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/*
2 * linux/net/sunrpc/xprtsock.c
3 *
4 * Client-side transport implementation for sockets.
5 *
6 * TCP callback races fixes (C) 1998 Red Hat
7 * TCP send fixes (C) 1998 Red Hat
8 * TCP NFS related read + write fixes
9 * (C) 1999 Dave Airlie, University of Limerick, Ireland <airlied@linux.ie>
10 *
11 * Rewrite of larges part of the code in order to stabilize TCP stuff.
12 * Fix behaviour when socket buffer is full.
13 * (C) 1999 Trond Myklebust <trond.myklebust@fys.uio.no>
14 *
15 * IP socket transport implementation, (C) 2005 Chuck Lever <cel@netapp.com>
16 *
17 * IPv6 support contributed by Gilles Quillard, Bull Open Source, 2005.
18 * <gilles.quillard@bull.net>
19 */
20
21#include <linux/types.h>
22#include <linux/string.h>
23#include <linux/slab.h>
24#include <linux/module.h>
25#include <linux/capability.h>
26#include <linux/pagemap.h>
27#include <linux/errno.h>
28#include <linux/socket.h>
29#include <linux/in.h>
30#include <linux/net.h>
31#include <linux/mm.h>
32#include <linux/un.h>
33#include <linux/udp.h>
34#include <linux/tcp.h>
35#include <linux/sunrpc/clnt.h>
36#include <linux/sunrpc/addr.h>
37#include <linux/sunrpc/sched.h>
38#include <linux/sunrpc/svcsock.h>
39#include <linux/sunrpc/xprtsock.h>
40#include <linux/file.h>
41#ifdef CONFIG_SUNRPC_BACKCHANNEL
42#include <linux/sunrpc/bc_xprt.h>
43#endif
44
45#include <net/sock.h>
46#include <net/checksum.h>
47#include <net/udp.h>
48#include <net/tcp.h>
49
50#include <trace/events/sunrpc.h>
51
52#include "sunrpc.h"
53
54static void xs_close(struct rpc_xprt *xprt);
55
56/*
57 * xprtsock tunables
58 */
59static unsigned int xprt_udp_slot_table_entries = RPC_DEF_SLOT_TABLE;
60static unsigned int xprt_tcp_slot_table_entries = RPC_MIN_SLOT_TABLE;
61static unsigned int xprt_max_tcp_slot_table_entries = RPC_MAX_SLOT_TABLE;
62
63static unsigned int xprt_min_resvport = RPC_DEF_MIN_RESVPORT;
64static unsigned int xprt_max_resvport = RPC_DEF_MAX_RESVPORT;
65
66#if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
67
68#define XS_TCP_LINGER_TO (15U * HZ)
69static unsigned int xs_tcp_fin_timeout __read_mostly = XS_TCP_LINGER_TO;
70
71/*
72 * We can register our own files under /proc/sys/sunrpc by
73 * calling register_sysctl_table() again. The files in that
74 * directory become the union of all files registered there.
75 *
76 * We simply need to make sure that we don't collide with
77 * someone else's file names!
78 */
79
80static unsigned int min_slot_table_size = RPC_MIN_SLOT_TABLE;
81static unsigned int max_slot_table_size = RPC_MAX_SLOT_TABLE;
82static unsigned int max_tcp_slot_table_limit = RPC_MAX_SLOT_TABLE_LIMIT;
83static unsigned int xprt_min_resvport_limit = RPC_MIN_RESVPORT;
84static unsigned int xprt_max_resvport_limit = RPC_MAX_RESVPORT;
85
86static struct ctl_table_header *sunrpc_table_header;
87
88/*
89 * FIXME: changing the UDP slot table size should also resize the UDP
90 * socket buffers for existing UDP transports
91 */
92static struct ctl_table xs_tunables_table[] = {
93 {
94 .procname = "udp_slot_table_entries",
95 .data = &xprt_udp_slot_table_entries,
96 .maxlen = sizeof(unsigned int),
97 .mode = 0644,
98 .proc_handler = proc_dointvec_minmax,
99 .extra1 = &min_slot_table_size,
100 .extra2 = &max_slot_table_size
101 },
102 {
103 .procname = "tcp_slot_table_entries",
104 .data = &xprt_tcp_slot_table_entries,
105 .maxlen = sizeof(unsigned int),
106 .mode = 0644,
107 .proc_handler = proc_dointvec_minmax,
108 .extra1 = &min_slot_table_size,
109 .extra2 = &max_slot_table_size
110 },
111 {
112 .procname = "tcp_max_slot_table_entries",
113 .data = &xprt_max_tcp_slot_table_entries,
114 .maxlen = sizeof(unsigned int),
115 .mode = 0644,
116 .proc_handler = proc_dointvec_minmax,
117 .extra1 = &min_slot_table_size,
118 .extra2 = &max_tcp_slot_table_limit
119 },
120 {
121 .procname = "min_resvport",
122 .data = &xprt_min_resvport,
123 .maxlen = sizeof(unsigned int),
124 .mode = 0644,
125 .proc_handler = proc_dointvec_minmax,
126 .extra1 = &xprt_min_resvport_limit,
127 .extra2 = &xprt_max_resvport_limit
128 },
129 {
130 .procname = "max_resvport",
131 .data = &xprt_max_resvport,
132 .maxlen = sizeof(unsigned int),
133 .mode = 0644,
134 .proc_handler = proc_dointvec_minmax,
135 .extra1 = &xprt_min_resvport_limit,
136 .extra2 = &xprt_max_resvport_limit
137 },
138 {
139 .procname = "tcp_fin_timeout",
140 .data = &xs_tcp_fin_timeout,
141 .maxlen = sizeof(xs_tcp_fin_timeout),
142 .mode = 0644,
143 .proc_handler = proc_dointvec_jiffies,
144 },
145 { },
146};
147
148static struct ctl_table sunrpc_table[] = {
149 {
150 .procname = "sunrpc",
151 .mode = 0555,
152 .child = xs_tunables_table
153 },
154 { },
155};
156
157#endif
158
159/*
160 * Wait duration for a reply from the RPC portmapper.
161 */
162#define XS_BIND_TO (60U * HZ)
163
164/*
165 * Delay if a UDP socket connect error occurs. This is most likely some
166 * kind of resource problem on the local host.
167 */
168#define XS_UDP_REEST_TO (2U * HZ)
169
170/*
171 * The reestablish timeout allows clients to delay for a bit before attempting
172 * to reconnect to a server that just dropped our connection.
173 *
174 * We implement an exponential backoff when trying to reestablish a TCP
175 * transport connection with the server. Some servers like to drop a TCP
176 * connection when they are overworked, so we start with a short timeout and
177 * increase over time if the server is down or not responding.
178 */
179#define XS_TCP_INIT_REEST_TO (3U * HZ)
180#define XS_TCP_MAX_REEST_TO (5U * 60 * HZ)
181
182/*
183 * TCP idle timeout; client drops the transport socket if it is idle
184 * for this long. Note that we also timeout UDP sockets to prevent
185 * holding port numbers when there is no RPC traffic.
186 */
187#define XS_IDLE_DISC_TO (5U * 60 * HZ)
188
189#if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
190# undef RPC_DEBUG_DATA
191# define RPCDBG_FACILITY RPCDBG_TRANS
192#endif
193
194#ifdef RPC_DEBUG_DATA
195static void xs_pktdump(char *msg, u32 *packet, unsigned int count)
196{
197 u8 *buf = (u8 *) packet;
198 int j;
199
200 dprintk("RPC: %s\n", msg);
201 for (j = 0; j < count && j < 128; j += 4) {
202 if (!(j & 31)) {
203 if (j)
204 dprintk("\n");
205 dprintk("0x%04x ", j);
206 }
207 dprintk("%02x%02x%02x%02x ",
208 buf[j], buf[j+1], buf[j+2], buf[j+3]);
209 }
210 dprintk("\n");
211}
212#else
213static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count)
214{
215 /* NOP */
216}
217#endif
218
219static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
220{
221 return (struct rpc_xprt *) sk->sk_user_data;
222}
223
224static inline struct sockaddr *xs_addr(struct rpc_xprt *xprt)
225{
226 return (struct sockaddr *) &xprt->addr;
227}
228
229static inline struct sockaddr_un *xs_addr_un(struct rpc_xprt *xprt)
230{
231 return (struct sockaddr_un *) &xprt->addr;
232}
233
234static inline struct sockaddr_in *xs_addr_in(struct rpc_xprt *xprt)
235{
236 return (struct sockaddr_in *) &xprt->addr;
237}
238
239static inline struct sockaddr_in6 *xs_addr_in6(struct rpc_xprt *xprt)
240{
241 return (struct sockaddr_in6 *) &xprt->addr;
242}
243
244static void xs_format_common_peer_addresses(struct rpc_xprt *xprt)
245{
246 struct sockaddr *sap = xs_addr(xprt);
247 struct sockaddr_in6 *sin6;
248 struct sockaddr_in *sin;
249 struct sockaddr_un *sun;
250 char buf[128];
251
252 switch (sap->sa_family) {
253 case AF_LOCAL:
254 sun = xs_addr_un(xprt);
255 strlcpy(buf, sun->sun_path, sizeof(buf));
256 xprt->address_strings[RPC_DISPLAY_ADDR] =
257 kstrdup(buf, GFP_KERNEL);
258 break;
259 case AF_INET:
260 (void)rpc_ntop(sap, buf, sizeof(buf));
261 xprt->address_strings[RPC_DISPLAY_ADDR] =
262 kstrdup(buf, GFP_KERNEL);
263 sin = xs_addr_in(xprt);
264 snprintf(buf, sizeof(buf), "%08x", ntohl(sin->sin_addr.s_addr));
265 break;
266 case AF_INET6:
267 (void)rpc_ntop(sap, buf, sizeof(buf));
268 xprt->address_strings[RPC_DISPLAY_ADDR] =
269 kstrdup(buf, GFP_KERNEL);
270 sin6 = xs_addr_in6(xprt);
271 snprintf(buf, sizeof(buf), "%pi6", &sin6->sin6_addr);
272 break;
273 default:
274 BUG();
275 }
276
277 xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = kstrdup(buf, GFP_KERNEL);
278}
279
280static void xs_format_common_peer_ports(struct rpc_xprt *xprt)
281{
282 struct sockaddr *sap = xs_addr(xprt);
283 char buf[128];
284
285 snprintf(buf, sizeof(buf), "%u", rpc_get_port(sap));
286 xprt->address_strings[RPC_DISPLAY_PORT] = kstrdup(buf, GFP_KERNEL);
287
288 snprintf(buf, sizeof(buf), "%4hx", rpc_get_port(sap));
289 xprt->address_strings[RPC_DISPLAY_HEX_PORT] = kstrdup(buf, GFP_KERNEL);
290}
291
292static void xs_format_peer_addresses(struct rpc_xprt *xprt,
293 const char *protocol,
294 const char *netid)
295{
296 xprt->address_strings[RPC_DISPLAY_PROTO] = protocol;
297 xprt->address_strings[RPC_DISPLAY_NETID] = netid;
298 xs_format_common_peer_addresses(xprt);
299 xs_format_common_peer_ports(xprt);
300}
301
302static void xs_update_peer_port(struct rpc_xprt *xprt)
303{
304 kfree(xprt->address_strings[RPC_DISPLAY_HEX_PORT]);
305 kfree(xprt->address_strings[RPC_DISPLAY_PORT]);
306
307 xs_format_common_peer_ports(xprt);
308}
309
310static void xs_free_peer_addresses(struct rpc_xprt *xprt)
311{
312 unsigned int i;
313
314 for (i = 0; i < RPC_DISPLAY_MAX; i++)
315 switch (i) {
316 case RPC_DISPLAY_PROTO:
317 case RPC_DISPLAY_NETID:
318 continue;
319 default:
320 kfree(xprt->address_strings[i]);
321 }
322}
323
324#define XS_SENDMSG_FLAGS (MSG_DONTWAIT | MSG_NOSIGNAL)
325
326static int xs_send_kvec(struct socket *sock, struct sockaddr *addr, int addrlen, struct kvec *vec, unsigned int base, int more)
327{
328 struct msghdr msg = {
329 .msg_name = addr,
330 .msg_namelen = addrlen,
331 .msg_flags = XS_SENDMSG_FLAGS | (more ? MSG_MORE : 0),
332 };
333 struct kvec iov = {
334 .iov_base = vec->iov_base + base,
335 .iov_len = vec->iov_len - base,
336 };
337
338 if (iov.iov_len != 0)
339 return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
340 return kernel_sendmsg(sock, &msg, NULL, 0, 0);
341}
342
343static int xs_send_pagedata(struct socket *sock, struct xdr_buf *xdr, unsigned int base, int more, bool zerocopy, int *sent_p)
344{
345 ssize_t (*do_sendpage)(struct socket *sock, struct page *page,
346 int offset, size_t size, int flags);
347 struct page **ppage;
348 unsigned int remainder;
349 int err;
350
351 remainder = xdr->page_len - base;
352 base += xdr->page_base;
353 ppage = xdr->pages + (base >> PAGE_SHIFT);
354 base &= ~PAGE_MASK;
355 do_sendpage = sock->ops->sendpage;
356 if (!zerocopy)
357 do_sendpage = sock_no_sendpage;
358 for(;;) {
359 unsigned int len = min_t(unsigned int, PAGE_SIZE - base, remainder);
360 int flags = XS_SENDMSG_FLAGS;
361
362 remainder -= len;
363 if (more)
364 flags |= MSG_MORE;
365 if (remainder != 0)
366 flags |= MSG_SENDPAGE_NOTLAST | MSG_MORE;
367 err = do_sendpage(sock, *ppage, base, len, flags);
368 if (remainder == 0 || err != len)
369 break;
370 *sent_p += err;
371 ppage++;
372 base = 0;
373 }
374 if (err > 0) {
375 *sent_p += err;
376 err = 0;
377 }
378 return err;
379}
380
381/**
382 * xs_sendpages - write pages directly to a socket
383 * @sock: socket to send on
384 * @addr: UDP only -- address of destination
385 * @addrlen: UDP only -- length of destination address
386 * @xdr: buffer containing this request
387 * @base: starting position in the buffer
388 * @zerocopy: true if it is safe to use sendpage()
389 * @sent_p: return the total number of bytes successfully queued for sending
390 *
391 */
392static int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base, bool zerocopy, int *sent_p)
393{
394 unsigned int remainder = xdr->len - base;
395 int err = 0;
396 int sent = 0;
397
398 if (unlikely(!sock))
399 return -ENOTSOCK;
400
401 if (base != 0) {
402 addr = NULL;
403 addrlen = 0;
404 }
405
406 if (base < xdr->head[0].iov_len || addr != NULL) {
407 unsigned int len = xdr->head[0].iov_len - base;
408 remainder -= len;
409 err = xs_send_kvec(sock, addr, addrlen, &xdr->head[0], base, remainder != 0);
410 if (remainder == 0 || err != len)
411 goto out;
412 *sent_p += err;
413 base = 0;
414 } else
415 base -= xdr->head[0].iov_len;
416
417 if (base < xdr->page_len) {
418 unsigned int len = xdr->page_len - base;
419 remainder -= len;
420 err = xs_send_pagedata(sock, xdr, base, remainder != 0, zerocopy, &sent);
421 *sent_p += sent;
422 if (remainder == 0 || sent != len)
423 goto out;
424 base = 0;
425 } else
426 base -= xdr->page_len;
427
428 if (base >= xdr->tail[0].iov_len)
429 return 0;
430 err = xs_send_kvec(sock, NULL, 0, &xdr->tail[0], base, 0);
431out:
432 if (err > 0) {
433 *sent_p += err;
434 err = 0;
435 }
436 return err;
437}
438
439static void xs_nospace_callback(struct rpc_task *task)
440{
441 struct sock_xprt *transport = container_of(task->tk_rqstp->rq_xprt, struct sock_xprt, xprt);
442
443 transport->inet->sk_write_pending--;
444}
445
446/**
447 * xs_nospace - place task on wait queue if transmit was incomplete
448 * @task: task to put to sleep
449 *
450 */
451static int xs_nospace(struct rpc_task *task)
452{
453 struct rpc_rqst *req = task->tk_rqstp;
454 struct rpc_xprt *xprt = req->rq_xprt;
455 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
456 struct sock *sk = transport->inet;
457 int ret = -EAGAIN;
458
459 dprintk("RPC: %5u xmit incomplete (%u left of %u)\n",
460 task->tk_pid, req->rq_slen - req->rq_bytes_sent,
461 req->rq_slen);
462
463 /* Protect against races with write_space */
464 spin_lock_bh(&xprt->transport_lock);
465
466 /* Don't race with disconnect */
467 if (xprt_connected(xprt)) {
468 /* wait for more buffer space */
469 sk->sk_write_pending++;
470 xprt_wait_for_buffer_space(task, xs_nospace_callback);
471 } else
472 ret = -ENOTCONN;
473
474 spin_unlock_bh(&xprt->transport_lock);
475
476 /* Race breaker in case memory is freed before above code is called */
477 sk->sk_write_space(sk);
478 return ret;
479}
480
481/*
482 * Construct a stream transport record marker in @buf.
483 */
484static inline void xs_encode_stream_record_marker(struct xdr_buf *buf)
485{
486 u32 reclen = buf->len - sizeof(rpc_fraghdr);
487 rpc_fraghdr *base = buf->head[0].iov_base;
488 *base = cpu_to_be32(RPC_LAST_STREAM_FRAGMENT | reclen);
489}
490
491/**
492 * xs_local_send_request - write an RPC request to an AF_LOCAL socket
493 * @task: RPC task that manages the state of an RPC request
494 *
495 * Return values:
496 * 0: The request has been sent
497 * EAGAIN: The socket was blocked, please call again later to
498 * complete the request
499 * ENOTCONN: Caller needs to invoke connect logic then call again
500 * other: Some other error occured, the request was not sent
501 */
502static int xs_local_send_request(struct rpc_task *task)
503{
504 struct rpc_rqst *req = task->tk_rqstp;
505 struct rpc_xprt *xprt = req->rq_xprt;
506 struct sock_xprt *transport =
507 container_of(xprt, struct sock_xprt, xprt);
508 struct xdr_buf *xdr = &req->rq_snd_buf;
509 int status;
510 int sent = 0;
511
512 xs_encode_stream_record_marker(&req->rq_snd_buf);
513
514 xs_pktdump("packet data:",
515 req->rq_svec->iov_base, req->rq_svec->iov_len);
516
517 status = xs_sendpages(transport->sock, NULL, 0, xdr, req->rq_bytes_sent,
518 true, &sent);
519 dprintk("RPC: %s(%u) = %d\n",
520 __func__, xdr->len - req->rq_bytes_sent, status);
521
522 if (status == -EAGAIN && sock_writeable(transport->inet))
523 status = -ENOBUFS;
524
525 if (likely(sent > 0) || status == 0) {
526 req->rq_bytes_sent += sent;
527 req->rq_xmit_bytes_sent += sent;
528 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
529 req->rq_bytes_sent = 0;
530 return 0;
531 }
532 status = -EAGAIN;
533 }
534
535 switch (status) {
536 case -ENOBUFS:
537 break;
538 case -EAGAIN:
539 status = xs_nospace(task);
540 break;
541 default:
542 dprintk("RPC: sendmsg returned unrecognized error %d\n",
543 -status);
544 case -EPIPE:
545 xs_close(xprt);
546 status = -ENOTCONN;
547 }
548
549 return status;
550}
551
552/**
553 * xs_udp_send_request - write an RPC request to a UDP socket
554 * @task: address of RPC task that manages the state of an RPC request
555 *
556 * Return values:
557 * 0: The request has been sent
558 * EAGAIN: The socket was blocked, please call again later to
559 * complete the request
560 * ENOTCONN: Caller needs to invoke connect logic then call again
561 * other: Some other error occurred, the request was not sent
562 */
563static int xs_udp_send_request(struct rpc_task *task)
564{
565 struct rpc_rqst *req = task->tk_rqstp;
566 struct rpc_xprt *xprt = req->rq_xprt;
567 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
568 struct xdr_buf *xdr = &req->rq_snd_buf;
569 int sent = 0;
570 int status;
571
572 xs_pktdump("packet data:",
573 req->rq_svec->iov_base,
574 req->rq_svec->iov_len);
575
576 if (!xprt_bound(xprt))
577 return -ENOTCONN;
578 status = xs_sendpages(transport->sock, xs_addr(xprt), xprt->addrlen,
579 xdr, req->rq_bytes_sent, true, &sent);
580
581 dprintk("RPC: xs_udp_send_request(%u) = %d\n",
582 xdr->len - req->rq_bytes_sent, status);
583
584 /* firewall is blocking us, don't return -EAGAIN or we end up looping */
585 if (status == -EPERM)
586 goto process_status;
587
588 if (status == -EAGAIN && sock_writeable(transport->inet))
589 status = -ENOBUFS;
590
591 if (sent > 0 || status == 0) {
592 req->rq_xmit_bytes_sent += sent;
593 if (sent >= req->rq_slen)
594 return 0;
595 /* Still some bytes left; set up for a retry later. */
596 status = -EAGAIN;
597 }
598
599process_status:
600 switch (status) {
601 case -ENOTSOCK:
602 status = -ENOTCONN;
603 /* Should we call xs_close() here? */
604 break;
605 case -EAGAIN:
606 status = xs_nospace(task);
607 break;
608 case -ENETUNREACH:
609 case -ENOBUFS:
610 case -EPIPE:
611 case -ECONNREFUSED:
612 case -EPERM:
613 /* When the server has died, an ICMP port unreachable message
614 * prompts ECONNREFUSED. */
615 break;
616 default:
617 dprintk("RPC: sendmsg returned unrecognized error %d\n",
618 -status);
619 }
620
621 return status;
622}
623
624/**
625 * xs_tcp_send_request - write an RPC request to a TCP socket
626 * @task: address of RPC task that manages the state of an RPC request
627 *
628 * Return values:
629 * 0: The request has been sent
630 * EAGAIN: The socket was blocked, please call again later to
631 * complete the request
632 * ENOTCONN: Caller needs to invoke connect logic then call again
633 * other: Some other error occurred, the request was not sent
634 *
635 * XXX: In the case of soft timeouts, should we eventually give up
636 * if sendmsg is not able to make progress?
637 */
638static int xs_tcp_send_request(struct rpc_task *task)
639{
640 struct rpc_rqst *req = task->tk_rqstp;
641 struct rpc_xprt *xprt = req->rq_xprt;
642 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
643 struct xdr_buf *xdr = &req->rq_snd_buf;
644 bool zerocopy = true;
645 int status;
646 int sent;
647
648 xs_encode_stream_record_marker(&req->rq_snd_buf);
649
650 xs_pktdump("packet data:",
651 req->rq_svec->iov_base,
652 req->rq_svec->iov_len);
653 /* Don't use zero copy if this is a resend. If the RPC call
654 * completes while the socket holds a reference to the pages,
655 * then we may end up resending corrupted data.
656 */
657 if (task->tk_flags & RPC_TASK_SENT)
658 zerocopy = false;
659
660 /* Continue transmitting the packet/record. We must be careful
661 * to cope with writespace callbacks arriving _after_ we have
662 * called sendmsg(). */
663 while (1) {
664 sent = 0;
665 status = xs_sendpages(transport->sock, NULL, 0, xdr,
666 req->rq_bytes_sent, zerocopy, &sent);
667
668 dprintk("RPC: xs_tcp_send_request(%u) = %d\n",
669 xdr->len - req->rq_bytes_sent, status);
670
671 /* If we've sent the entire packet, immediately
672 * reset the count of bytes sent. */
673 req->rq_bytes_sent += sent;
674 req->rq_xmit_bytes_sent += sent;
675 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
676 req->rq_bytes_sent = 0;
677 return 0;
678 }
679
680 if (status < 0)
681 break;
682 if (sent == 0) {
683 status = -EAGAIN;
684 break;
685 }
686 }
687 if (status == -EAGAIN && sk_stream_is_writeable(transport->inet))
688 status = -ENOBUFS;
689
690 switch (status) {
691 case -ENOTSOCK:
692 status = -ENOTCONN;
693 /* Should we call xs_close() here? */
694 break;
695 case -EAGAIN:
696 status = xs_nospace(task);
697 break;
698 case -ECONNRESET:
699 case -ECONNREFUSED:
700 case -ENOTCONN:
701 case -EADDRINUSE:
702 case -ENOBUFS:
703 case -EPIPE:
704 break;
705 default:
706 dprintk("RPC: sendmsg returned unrecognized error %d\n",
707 -status);
708 }
709
710 return status;
711}
712
713/**
714 * xs_tcp_release_xprt - clean up after a tcp transmission
715 * @xprt: transport
716 * @task: rpc task
717 *
718 * This cleans up if an error causes us to abort the transmission of a request.
719 * In this case, the socket may need to be reset in order to avoid confusing
720 * the server.
721 */
722static void xs_tcp_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
723{
724 struct rpc_rqst *req;
725
726 if (task != xprt->snd_task)
727 return;
728 if (task == NULL)
729 goto out_release;
730 req = task->tk_rqstp;
731 if (req == NULL)
732 goto out_release;
733 if (req->rq_bytes_sent == 0)
734 goto out_release;
735 if (req->rq_bytes_sent == req->rq_snd_buf.len)
736 goto out_release;
737 set_bit(XPRT_CLOSE_WAIT, &xprt->state);
738out_release:
739 xprt_release_xprt(xprt, task);
740}
741
742static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk)
743{
744 transport->old_data_ready = sk->sk_data_ready;
745 transport->old_state_change = sk->sk_state_change;
746 transport->old_write_space = sk->sk_write_space;
747 transport->old_error_report = sk->sk_error_report;
748}
749
750static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk)
751{
752 sk->sk_data_ready = transport->old_data_ready;
753 sk->sk_state_change = transport->old_state_change;
754 sk->sk_write_space = transport->old_write_space;
755 sk->sk_error_report = transport->old_error_report;
756}
757
758static void xs_sock_reset_connection_flags(struct rpc_xprt *xprt)
759{
760 smp_mb__before_atomic();
761 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
762 clear_bit(XPRT_CLOSING, &xprt->state);
763 smp_mb__after_atomic();
764}
765
766static void xs_sock_mark_closed(struct rpc_xprt *xprt)
767{
768 xs_sock_reset_connection_flags(xprt);
769 /* Mark transport as closed and wake up all pending tasks */
770 xprt_disconnect_done(xprt);
771}
772
773/**
774 * xs_error_report - callback to handle TCP socket state errors
775 * @sk: socket
776 *
777 * Note: we don't call sock_error() since there may be a rpc_task
778 * using the socket, and so we don't want to clear sk->sk_err.
779 */
780static void xs_error_report(struct sock *sk)
781{
782 struct rpc_xprt *xprt;
783 int err;
784
785 read_lock_bh(&sk->sk_callback_lock);
786 if (!(xprt = xprt_from_sock(sk)))
787 goto out;
788
789 err = -sk->sk_err;
790 if (err == 0)
791 goto out;
792 /* Is this a reset event? */
793 if (sk->sk_state == TCP_CLOSE)
794 xs_sock_mark_closed(xprt);
795 dprintk("RPC: xs_error_report client %p, error=%d...\n",
796 xprt, -err);
797 trace_rpc_socket_error(xprt, sk->sk_socket, err);
798 xprt_wake_pending_tasks(xprt, err);
799 out:
800 read_unlock_bh(&sk->sk_callback_lock);
801}
802
803static void xs_reset_transport(struct sock_xprt *transport)
804{
805 struct socket *sock = transport->sock;
806 struct sock *sk = transport->inet;
807 struct rpc_xprt *xprt = &transport->xprt;
808
809 if (sk == NULL)
810 return;
811
812 if (atomic_read(&transport->xprt.swapper))
813 sk_clear_memalloc(sk);
814
815 kernel_sock_shutdown(sock, SHUT_RDWR);
816
817 mutex_lock(&transport->recv_mutex);
818 write_lock_bh(&sk->sk_callback_lock);
819 transport->inet = NULL;
820 transport->sock = NULL;
821
822 sk->sk_user_data = NULL;
823
824 xs_restore_old_callbacks(transport, sk);
825 xprt_clear_connected(xprt);
826 write_unlock_bh(&sk->sk_callback_lock);
827 xs_sock_reset_connection_flags(xprt);
828 mutex_unlock(&transport->recv_mutex);
829
830 trace_rpc_socket_close(xprt, sock);
831 sock_release(sock);
832}
833
834/**
835 * xs_close - close a socket
836 * @xprt: transport
837 *
838 * This is used when all requests are complete; ie, no DRC state remains
839 * on the server we want to save.
840 *
841 * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with
842 * xs_reset_transport() zeroing the socket from underneath a writer.
843 */
844static void xs_close(struct rpc_xprt *xprt)
845{
846 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
847
848 dprintk("RPC: xs_close xprt %p\n", xprt);
849
850 xs_reset_transport(transport);
851 xprt->reestablish_timeout = 0;
852
853 xprt_disconnect_done(xprt);
854}
855
856static void xs_inject_disconnect(struct rpc_xprt *xprt)
857{
858 dprintk("RPC: injecting transport disconnect on xprt=%p\n",
859 xprt);
860 xprt_disconnect_done(xprt);
861}
862
863static void xs_xprt_free(struct rpc_xprt *xprt)
864{
865 xs_free_peer_addresses(xprt);
866 xprt_free(xprt);
867}
868
869/**
870 * xs_destroy - prepare to shutdown a transport
871 * @xprt: doomed transport
872 *
873 */
874static void xs_destroy(struct rpc_xprt *xprt)
875{
876 struct sock_xprt *transport = container_of(xprt,
877 struct sock_xprt, xprt);
878 dprintk("RPC: xs_destroy xprt %p\n", xprt);
879
880 cancel_delayed_work_sync(&transport->connect_worker);
881 xs_close(xprt);
882 cancel_work_sync(&transport->recv_worker);
883 xs_xprt_free(xprt);
884 module_put(THIS_MODULE);
885}
886
887static int xs_local_copy_to_xdr(struct xdr_buf *xdr, struct sk_buff *skb)
888{
889 struct xdr_skb_reader desc = {
890 .skb = skb,
891 .offset = sizeof(rpc_fraghdr),
892 .count = skb->len - sizeof(rpc_fraghdr),
893 };
894
895 if (xdr_partial_copy_from_skb(xdr, 0, &desc, xdr_skb_read_bits) < 0)
896 return -1;
897 if (desc.count)
898 return -1;
899 return 0;
900}
901
902/**
903 * xs_local_data_read_skb
904 * @xprt: transport
905 * @sk: socket
906 * @skb: skbuff
907 *
908 * Currently this assumes we can read the whole reply in a single gulp.
909 */
910static void xs_local_data_read_skb(struct rpc_xprt *xprt,
911 struct sock *sk,
912 struct sk_buff *skb)
913{
914 struct rpc_task *task;
915 struct rpc_rqst *rovr;
916 int repsize, copied;
917 u32 _xid;
918 __be32 *xp;
919
920 repsize = skb->len - sizeof(rpc_fraghdr);
921 if (repsize < 4) {
922 dprintk("RPC: impossible RPC reply size %d\n", repsize);
923 return;
924 }
925
926 /* Copy the XID from the skb... */
927 xp = skb_header_pointer(skb, sizeof(rpc_fraghdr), sizeof(_xid), &_xid);
928 if (xp == NULL)
929 return;
930
931 /* Look up and lock the request corresponding to the given XID */
932 spin_lock_bh(&xprt->transport_lock);
933 rovr = xprt_lookup_rqst(xprt, *xp);
934 if (!rovr)
935 goto out_unlock;
936 task = rovr->rq_task;
937
938 copied = rovr->rq_private_buf.buflen;
939 if (copied > repsize)
940 copied = repsize;
941
942 if (xs_local_copy_to_xdr(&rovr->rq_private_buf, skb)) {
943 dprintk("RPC: sk_buff copy failed\n");
944 goto out_unlock;
945 }
946
947 xprt_complete_rqst(task, copied);
948
949 out_unlock:
950 spin_unlock_bh(&xprt->transport_lock);
951}
952
953static void xs_local_data_receive(struct sock_xprt *transport)
954{
955 struct sk_buff *skb;
956 struct sock *sk;
957 int err;
958
959 mutex_lock(&transport->recv_mutex);
960 sk = transport->inet;
961 if (sk == NULL)
962 goto out;
963 for (;;) {
964 skb = skb_recv_datagram(sk, 0, 1, &err);
965 if (skb == NULL)
966 break;
967 xs_local_data_read_skb(&transport->xprt, sk, skb);
968 skb_free_datagram(sk, skb);
969 }
970out:
971 mutex_unlock(&transport->recv_mutex);
972}
973
974static void xs_local_data_receive_workfn(struct work_struct *work)
975{
976 struct sock_xprt *transport =
977 container_of(work, struct sock_xprt, recv_worker);
978 xs_local_data_receive(transport);
979}
980
981/**
982 * xs_udp_data_read_skb - receive callback for UDP sockets
983 * @xprt: transport
984 * @sk: socket
985 * @skb: skbuff
986 *
987 */
988static void xs_udp_data_read_skb(struct rpc_xprt *xprt,
989 struct sock *sk,
990 struct sk_buff *skb)
991{
992 struct rpc_task *task;
993 struct rpc_rqst *rovr;
994 int repsize, copied;
995 u32 _xid;
996 __be32 *xp;
997
998 repsize = skb->len - sizeof(struct udphdr);
999 if (repsize < 4) {
1000 dprintk("RPC: impossible RPC reply size %d!\n", repsize);
1001 return;
1002 }
1003
1004 /* Copy the XID from the skb... */
1005 xp = skb_header_pointer(skb, sizeof(struct udphdr),
1006 sizeof(_xid), &_xid);
1007 if (xp == NULL)
1008 return;
1009
1010 /* Look up and lock the request corresponding to the given XID */
1011 spin_lock_bh(&xprt->transport_lock);
1012 rovr = xprt_lookup_rqst(xprt, *xp);
1013 if (!rovr)
1014 goto out_unlock;
1015 task = rovr->rq_task;
1016
1017 if ((copied = rovr->rq_private_buf.buflen) > repsize)
1018 copied = repsize;
1019
1020 /* Suck it into the iovec, verify checksum if not done by hw. */
1021 if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
1022 UDPX_INC_STATS_BH(sk, UDP_MIB_INERRORS);
1023 goto out_unlock;
1024 }
1025
1026 UDPX_INC_STATS_BH(sk, UDP_MIB_INDATAGRAMS);
1027
1028 xprt_adjust_cwnd(xprt, task, copied);
1029 xprt_complete_rqst(task, copied);
1030
1031 out_unlock:
1032 spin_unlock_bh(&xprt->transport_lock);
1033}
1034
1035static void xs_udp_data_receive(struct sock_xprt *transport)
1036{
1037 struct sk_buff *skb;
1038 struct sock *sk;
1039 int err;
1040
1041 mutex_lock(&transport->recv_mutex);
1042 sk = transport->inet;
1043 if (sk == NULL)
1044 goto out;
1045 for (;;) {
1046 skb = skb_recv_datagram(sk, 0, 1, &err);
1047 if (skb == NULL)
1048 break;
1049 xs_udp_data_read_skb(&transport->xprt, sk, skb);
1050 skb_free_datagram(sk, skb);
1051 }
1052out:
1053 mutex_unlock(&transport->recv_mutex);
1054}
1055
1056static void xs_udp_data_receive_workfn(struct work_struct *work)
1057{
1058 struct sock_xprt *transport =
1059 container_of(work, struct sock_xprt, recv_worker);
1060 xs_udp_data_receive(transport);
1061}
1062
1063/**
1064 * xs_data_ready - "data ready" callback for UDP sockets
1065 * @sk: socket with data to read
1066 *
1067 */
1068static void xs_data_ready(struct sock *sk)
1069{
1070 struct rpc_xprt *xprt;
1071
1072 read_lock_bh(&sk->sk_callback_lock);
1073 dprintk("RPC: xs_data_ready...\n");
1074 xprt = xprt_from_sock(sk);
1075 if (xprt != NULL) {
1076 struct sock_xprt *transport = container_of(xprt,
1077 struct sock_xprt, xprt);
1078 queue_work(rpciod_workqueue, &transport->recv_worker);
1079 }
1080 read_unlock_bh(&sk->sk_callback_lock);
1081}
1082
1083/*
1084 * Helper function to force a TCP close if the server is sending
1085 * junk and/or it has put us in CLOSE_WAIT
1086 */
1087static void xs_tcp_force_close(struct rpc_xprt *xprt)
1088{
1089 xprt_force_disconnect(xprt);
1090}
1091
1092static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
1093{
1094 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1095 size_t len, used;
1096 char *p;
1097
1098 p = ((char *) &transport->tcp_fraghdr) + transport->tcp_offset;
1099 len = sizeof(transport->tcp_fraghdr) - transport->tcp_offset;
1100 used = xdr_skb_read_bits(desc, p, len);
1101 transport->tcp_offset += used;
1102 if (used != len)
1103 return;
1104
1105 transport->tcp_reclen = ntohl(transport->tcp_fraghdr);
1106 if (transport->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
1107 transport->tcp_flags |= TCP_RCV_LAST_FRAG;
1108 else
1109 transport->tcp_flags &= ~TCP_RCV_LAST_FRAG;
1110 transport->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
1111
1112 transport->tcp_flags &= ~TCP_RCV_COPY_FRAGHDR;
1113 transport->tcp_offset = 0;
1114
1115 /* Sanity check of the record length */
1116 if (unlikely(transport->tcp_reclen < 8)) {
1117 dprintk("RPC: invalid TCP record fragment length\n");
1118 xs_tcp_force_close(xprt);
1119 return;
1120 }
1121 dprintk("RPC: reading TCP record fragment of length %d\n",
1122 transport->tcp_reclen);
1123}
1124
1125static void xs_tcp_check_fraghdr(struct sock_xprt *transport)
1126{
1127 if (transport->tcp_offset == transport->tcp_reclen) {
1128 transport->tcp_flags |= TCP_RCV_COPY_FRAGHDR;
1129 transport->tcp_offset = 0;
1130 if (transport->tcp_flags & TCP_RCV_LAST_FRAG) {
1131 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1132 transport->tcp_flags |= TCP_RCV_COPY_XID;
1133 transport->tcp_copied = 0;
1134 }
1135 }
1136}
1137
1138static inline void xs_tcp_read_xid(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1139{
1140 size_t len, used;
1141 char *p;
1142
1143 len = sizeof(transport->tcp_xid) - transport->tcp_offset;
1144 dprintk("RPC: reading XID (%Zu bytes)\n", len);
1145 p = ((char *) &transport->tcp_xid) + transport->tcp_offset;
1146 used = xdr_skb_read_bits(desc, p, len);
1147 transport->tcp_offset += used;
1148 if (used != len)
1149 return;
1150 transport->tcp_flags &= ~TCP_RCV_COPY_XID;
1151 transport->tcp_flags |= TCP_RCV_READ_CALLDIR;
1152 transport->tcp_copied = 4;
1153 dprintk("RPC: reading %s XID %08x\n",
1154 (transport->tcp_flags & TCP_RPC_REPLY) ? "reply for"
1155 : "request with",
1156 ntohl(transport->tcp_xid));
1157 xs_tcp_check_fraghdr(transport);
1158}
1159
1160static inline void xs_tcp_read_calldir(struct sock_xprt *transport,
1161 struct xdr_skb_reader *desc)
1162{
1163 size_t len, used;
1164 u32 offset;
1165 char *p;
1166
1167 /*
1168 * We want transport->tcp_offset to be 8 at the end of this routine
1169 * (4 bytes for the xid and 4 bytes for the call/reply flag).
1170 * When this function is called for the first time,
1171 * transport->tcp_offset is 4 (after having already read the xid).
1172 */
1173 offset = transport->tcp_offset - sizeof(transport->tcp_xid);
1174 len = sizeof(transport->tcp_calldir) - offset;
1175 dprintk("RPC: reading CALL/REPLY flag (%Zu bytes)\n", len);
1176 p = ((char *) &transport->tcp_calldir) + offset;
1177 used = xdr_skb_read_bits(desc, p, len);
1178 transport->tcp_offset += used;
1179 if (used != len)
1180 return;
1181 transport->tcp_flags &= ~TCP_RCV_READ_CALLDIR;
1182 /*
1183 * We don't yet have the XDR buffer, so we will write the calldir
1184 * out after we get the buffer from the 'struct rpc_rqst'
1185 */
1186 switch (ntohl(transport->tcp_calldir)) {
1187 case RPC_REPLY:
1188 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1189 transport->tcp_flags |= TCP_RCV_COPY_DATA;
1190 transport->tcp_flags |= TCP_RPC_REPLY;
1191 break;
1192 case RPC_CALL:
1193 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1194 transport->tcp_flags |= TCP_RCV_COPY_DATA;
1195 transport->tcp_flags &= ~TCP_RPC_REPLY;
1196 break;
1197 default:
1198 dprintk("RPC: invalid request message type\n");
1199 xs_tcp_force_close(&transport->xprt);
1200 }
1201 xs_tcp_check_fraghdr(transport);
1202}
1203
1204static inline void xs_tcp_read_common(struct rpc_xprt *xprt,
1205 struct xdr_skb_reader *desc,
1206 struct rpc_rqst *req)
1207{
1208 struct sock_xprt *transport =
1209 container_of(xprt, struct sock_xprt, xprt);
1210 struct xdr_buf *rcvbuf;
1211 size_t len;
1212 ssize_t r;
1213
1214 rcvbuf = &req->rq_private_buf;
1215
1216 if (transport->tcp_flags & TCP_RCV_COPY_CALLDIR) {
1217 /*
1218 * Save the RPC direction in the XDR buffer
1219 */
1220 memcpy(rcvbuf->head[0].iov_base + transport->tcp_copied,
1221 &transport->tcp_calldir,
1222 sizeof(transport->tcp_calldir));
1223 transport->tcp_copied += sizeof(transport->tcp_calldir);
1224 transport->tcp_flags &= ~TCP_RCV_COPY_CALLDIR;
1225 }
1226
1227 len = desc->count;
1228 if (len > transport->tcp_reclen - transport->tcp_offset) {
1229 struct xdr_skb_reader my_desc;
1230
1231 len = transport->tcp_reclen - transport->tcp_offset;
1232 memcpy(&my_desc, desc, sizeof(my_desc));
1233 my_desc.count = len;
1234 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
1235 &my_desc, xdr_skb_read_bits);
1236 desc->count -= r;
1237 desc->offset += r;
1238 } else
1239 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
1240 desc, xdr_skb_read_bits);
1241
1242 if (r > 0) {
1243 transport->tcp_copied += r;
1244 transport->tcp_offset += r;
1245 }
1246 if (r != len) {
1247 /* Error when copying to the receive buffer,
1248 * usually because we weren't able to allocate
1249 * additional buffer pages. All we can do now
1250 * is turn off TCP_RCV_COPY_DATA, so the request
1251 * will not receive any additional updates,
1252 * and time out.
1253 * Any remaining data from this record will
1254 * be discarded.
1255 */
1256 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1257 dprintk("RPC: XID %08x truncated request\n",
1258 ntohl(transport->tcp_xid));
1259 dprintk("RPC: xprt = %p, tcp_copied = %lu, "
1260 "tcp_offset = %u, tcp_reclen = %u\n",
1261 xprt, transport->tcp_copied,
1262 transport->tcp_offset, transport->tcp_reclen);
1263 return;
1264 }
1265
1266 dprintk("RPC: XID %08x read %Zd bytes\n",
1267 ntohl(transport->tcp_xid), r);
1268 dprintk("RPC: xprt = %p, tcp_copied = %lu, tcp_offset = %u, "
1269 "tcp_reclen = %u\n", xprt, transport->tcp_copied,
1270 transport->tcp_offset, transport->tcp_reclen);
1271
1272 if (transport->tcp_copied == req->rq_private_buf.buflen)
1273 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1274 else if (transport->tcp_offset == transport->tcp_reclen) {
1275 if (transport->tcp_flags & TCP_RCV_LAST_FRAG)
1276 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1277 }
1278}
1279
1280/*
1281 * Finds the request corresponding to the RPC xid and invokes the common
1282 * tcp read code to read the data.
1283 */
1284static inline int xs_tcp_read_reply(struct rpc_xprt *xprt,
1285 struct xdr_skb_reader *desc)
1286{
1287 struct sock_xprt *transport =
1288 container_of(xprt, struct sock_xprt, xprt);
1289 struct rpc_rqst *req;
1290
1291 dprintk("RPC: read reply XID %08x\n", ntohl(transport->tcp_xid));
1292
1293 /* Find and lock the request corresponding to this xid */
1294 spin_lock_bh(&xprt->transport_lock);
1295 req = xprt_lookup_rqst(xprt, transport->tcp_xid);
1296 if (!req) {
1297 dprintk("RPC: XID %08x request not found!\n",
1298 ntohl(transport->tcp_xid));
1299 spin_unlock_bh(&xprt->transport_lock);
1300 return -1;
1301 }
1302
1303 xs_tcp_read_common(xprt, desc, req);
1304
1305 if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1306 xprt_complete_rqst(req->rq_task, transport->tcp_copied);
1307
1308 spin_unlock_bh(&xprt->transport_lock);
1309 return 0;
1310}
1311
1312#if defined(CONFIG_SUNRPC_BACKCHANNEL)
1313/*
1314 * Obtains an rpc_rqst previously allocated and invokes the common
1315 * tcp read code to read the data. The result is placed in the callback
1316 * queue.
1317 * If we're unable to obtain the rpc_rqst we schedule the closing of the
1318 * connection and return -1.
1319 */
1320static int xs_tcp_read_callback(struct rpc_xprt *xprt,
1321 struct xdr_skb_reader *desc)
1322{
1323 struct sock_xprt *transport =
1324 container_of(xprt, struct sock_xprt, xprt);
1325 struct rpc_rqst *req;
1326
1327 /* Look up and lock the request corresponding to the given XID */
1328 spin_lock_bh(&xprt->transport_lock);
1329 req = xprt_lookup_bc_request(xprt, transport->tcp_xid);
1330 if (req == NULL) {
1331 spin_unlock_bh(&xprt->transport_lock);
1332 printk(KERN_WARNING "Callback slot table overflowed\n");
1333 xprt_force_disconnect(xprt);
1334 return -1;
1335 }
1336
1337 dprintk("RPC: read callback XID %08x\n", ntohl(req->rq_xid));
1338 xs_tcp_read_common(xprt, desc, req);
1339
1340 if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1341 xprt_complete_bc_request(req, transport->tcp_copied);
1342 spin_unlock_bh(&xprt->transport_lock);
1343
1344 return 0;
1345}
1346
1347static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1348 struct xdr_skb_reader *desc)
1349{
1350 struct sock_xprt *transport =
1351 container_of(xprt, struct sock_xprt, xprt);
1352
1353 return (transport->tcp_flags & TCP_RPC_REPLY) ?
1354 xs_tcp_read_reply(xprt, desc) :
1355 xs_tcp_read_callback(xprt, desc);
1356}
1357
1358static int xs_tcp_bc_up(struct svc_serv *serv, struct net *net)
1359{
1360 int ret;
1361
1362 ret = svc_create_xprt(serv, "tcp-bc", net, PF_INET, 0,
1363 SVC_SOCK_ANONYMOUS);
1364 if (ret < 0)
1365 return ret;
1366 return 0;
1367}
1368#else
1369static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1370 struct xdr_skb_reader *desc)
1371{
1372 return xs_tcp_read_reply(xprt, desc);
1373}
1374#endif /* CONFIG_SUNRPC_BACKCHANNEL */
1375
1376/*
1377 * Read data off the transport. This can be either an RPC_CALL or an
1378 * RPC_REPLY. Relay the processing to helper functions.
1379 */
1380static void xs_tcp_read_data(struct rpc_xprt *xprt,
1381 struct xdr_skb_reader *desc)
1382{
1383 struct sock_xprt *transport =
1384 container_of(xprt, struct sock_xprt, xprt);
1385
1386 if (_xs_tcp_read_data(xprt, desc) == 0)
1387 xs_tcp_check_fraghdr(transport);
1388 else {
1389 /*
1390 * The transport_lock protects the request handling.
1391 * There's no need to hold it to update the tcp_flags.
1392 */
1393 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1394 }
1395}
1396
1397static inline void xs_tcp_read_discard(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1398{
1399 size_t len;
1400
1401 len = transport->tcp_reclen - transport->tcp_offset;
1402 if (len > desc->count)
1403 len = desc->count;
1404 desc->count -= len;
1405 desc->offset += len;
1406 transport->tcp_offset += len;
1407 dprintk("RPC: discarded %Zu bytes\n", len);
1408 xs_tcp_check_fraghdr(transport);
1409}
1410
1411static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
1412{
1413 struct rpc_xprt *xprt = rd_desc->arg.data;
1414 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1415 struct xdr_skb_reader desc = {
1416 .skb = skb,
1417 .offset = offset,
1418 .count = len,
1419 };
1420
1421 dprintk("RPC: xs_tcp_data_recv started\n");
1422 do {
1423 trace_xs_tcp_data_recv(transport);
1424 /* Read in a new fragment marker if necessary */
1425 /* Can we ever really expect to get completely empty fragments? */
1426 if (transport->tcp_flags & TCP_RCV_COPY_FRAGHDR) {
1427 xs_tcp_read_fraghdr(xprt, &desc);
1428 continue;
1429 }
1430 /* Read in the xid if necessary */
1431 if (transport->tcp_flags & TCP_RCV_COPY_XID) {
1432 xs_tcp_read_xid(transport, &desc);
1433 continue;
1434 }
1435 /* Read in the call/reply flag */
1436 if (transport->tcp_flags & TCP_RCV_READ_CALLDIR) {
1437 xs_tcp_read_calldir(transport, &desc);
1438 continue;
1439 }
1440 /* Read in the request data */
1441 if (transport->tcp_flags & TCP_RCV_COPY_DATA) {
1442 xs_tcp_read_data(xprt, &desc);
1443 continue;
1444 }
1445 /* Skip over any trailing bytes on short reads */
1446 xs_tcp_read_discard(transport, &desc);
1447 } while (desc.count);
1448 trace_xs_tcp_data_recv(transport);
1449 dprintk("RPC: xs_tcp_data_recv done\n");
1450 return len - desc.count;
1451}
1452
1453static void xs_tcp_data_receive(struct sock_xprt *transport)
1454{
1455 struct rpc_xprt *xprt = &transport->xprt;
1456 struct sock *sk;
1457 read_descriptor_t rd_desc = {
1458 .count = 2*1024*1024,
1459 .arg.data = xprt,
1460 };
1461 unsigned long total = 0;
1462 int read = 0;
1463
1464 mutex_lock(&transport->recv_mutex);
1465 sk = transport->inet;
1466 if (sk == NULL)
1467 goto out;
1468
1469 /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
1470 for (;;) {
1471 lock_sock(sk);
1472 read = tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
1473 release_sock(sk);
1474 if (read <= 0)
1475 break;
1476 total += read;
1477 rd_desc.count = 65536;
1478 }
1479out:
1480 mutex_unlock(&transport->recv_mutex);
1481 trace_xs_tcp_data_ready(xprt, read, total);
1482}
1483
1484static void xs_tcp_data_receive_workfn(struct work_struct *work)
1485{
1486 struct sock_xprt *transport =
1487 container_of(work, struct sock_xprt, recv_worker);
1488 xs_tcp_data_receive(transport);
1489}
1490
1491/**
1492 * xs_tcp_data_ready - "data ready" callback for TCP sockets
1493 * @sk: socket with data to read
1494 *
1495 */
1496static void xs_tcp_data_ready(struct sock *sk)
1497{
1498 struct sock_xprt *transport;
1499 struct rpc_xprt *xprt;
1500
1501 dprintk("RPC: xs_tcp_data_ready...\n");
1502
1503 read_lock_bh(&sk->sk_callback_lock);
1504 if (!(xprt = xprt_from_sock(sk)))
1505 goto out;
1506 transport = container_of(xprt, struct sock_xprt, xprt);
1507
1508 /* Any data means we had a useful conversation, so
1509 * the we don't need to delay the next reconnect
1510 */
1511 if (xprt->reestablish_timeout)
1512 xprt->reestablish_timeout = 0;
1513 queue_work(rpciod_workqueue, &transport->recv_worker);
1514
1515out:
1516 read_unlock_bh(&sk->sk_callback_lock);
1517}
1518
1519/**
1520 * xs_tcp_state_change - callback to handle TCP socket state changes
1521 * @sk: socket whose state has changed
1522 *
1523 */
1524static void xs_tcp_state_change(struct sock *sk)
1525{
1526 struct rpc_xprt *xprt;
1527 struct sock_xprt *transport;
1528
1529 read_lock_bh(&sk->sk_callback_lock);
1530 if (!(xprt = xprt_from_sock(sk)))
1531 goto out;
1532 dprintk("RPC: xs_tcp_state_change client %p...\n", xprt);
1533 dprintk("RPC: state %x conn %d dead %d zapped %d sk_shutdown %d\n",
1534 sk->sk_state, xprt_connected(xprt),
1535 sock_flag(sk, SOCK_DEAD),
1536 sock_flag(sk, SOCK_ZAPPED),
1537 sk->sk_shutdown);
1538
1539 transport = container_of(xprt, struct sock_xprt, xprt);
1540 trace_rpc_socket_state_change(xprt, sk->sk_socket);
1541 switch (sk->sk_state) {
1542 case TCP_ESTABLISHED:
1543 spin_lock(&xprt->transport_lock);
1544 if (!xprt_test_and_set_connected(xprt)) {
1545
1546 /* Reset TCP record info */
1547 transport->tcp_offset = 0;
1548 transport->tcp_reclen = 0;
1549 transport->tcp_copied = 0;
1550 transport->tcp_flags =
1551 TCP_RCV_COPY_FRAGHDR | TCP_RCV_COPY_XID;
1552 xprt->connect_cookie++;
1553 clear_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
1554 xprt_clear_connecting(xprt);
1555
1556 xprt_wake_pending_tasks(xprt, -EAGAIN);
1557 }
1558 spin_unlock(&xprt->transport_lock);
1559 break;
1560 case TCP_FIN_WAIT1:
1561 /* The client initiated a shutdown of the socket */
1562 xprt->connect_cookie++;
1563 xprt->reestablish_timeout = 0;
1564 set_bit(XPRT_CLOSING, &xprt->state);
1565 smp_mb__before_atomic();
1566 clear_bit(XPRT_CONNECTED, &xprt->state);
1567 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1568 smp_mb__after_atomic();
1569 break;
1570 case TCP_CLOSE_WAIT:
1571 /* The server initiated a shutdown of the socket */
1572 xprt->connect_cookie++;
1573 clear_bit(XPRT_CONNECTED, &xprt->state);
1574 xs_tcp_force_close(xprt);
1575 case TCP_CLOSING:
1576 /*
1577 * If the server closed down the connection, make sure that
1578 * we back off before reconnecting
1579 */
1580 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1581 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1582 break;
1583 case TCP_LAST_ACK:
1584 set_bit(XPRT_CLOSING, &xprt->state);
1585 smp_mb__before_atomic();
1586 clear_bit(XPRT_CONNECTED, &xprt->state);
1587 smp_mb__after_atomic();
1588 break;
1589 case TCP_CLOSE:
1590 if (test_and_clear_bit(XPRT_SOCK_CONNECTING,
1591 &transport->sock_state))
1592 xprt_clear_connecting(xprt);
1593 xs_sock_mark_closed(xprt);
1594 }
1595 out:
1596 read_unlock_bh(&sk->sk_callback_lock);
1597}
1598
1599static void xs_write_space(struct sock *sk)
1600{
1601 struct socket_wq *wq;
1602 struct rpc_xprt *xprt;
1603
1604 if (!sk->sk_socket)
1605 return;
1606 clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1607
1608 if (unlikely(!(xprt = xprt_from_sock(sk))))
1609 return;
1610 rcu_read_lock();
1611 wq = rcu_dereference(sk->sk_wq);
1612 if (!wq || test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &wq->flags) == 0)
1613 goto out;
1614
1615 xprt_write_space(xprt);
1616out:
1617 rcu_read_unlock();
1618}
1619
1620/**
1621 * xs_udp_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_udp_write_space(struct sock *sk)
1631{
1632 read_lock_bh(&sk->sk_callback_lock);
1633
1634 /* from net/core/sock.c:sock_def_write_space */
1635 if (sock_writeable(sk))
1636 xs_write_space(sk);
1637
1638 read_unlock_bh(&sk->sk_callback_lock);
1639}
1640
1641/**
1642 * xs_tcp_write_space - callback invoked when socket buffer space
1643 * becomes available
1644 * @sk: socket whose state has changed
1645 *
1646 * Called when more output buffer space is available for this socket.
1647 * We try not to wake our writers until they can make "significant"
1648 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1649 * with a bunch of small requests.
1650 */
1651static void xs_tcp_write_space(struct sock *sk)
1652{
1653 read_lock_bh(&sk->sk_callback_lock);
1654
1655 /* from net/core/stream.c:sk_stream_write_space */
1656 if (sk_stream_is_writeable(sk))
1657 xs_write_space(sk);
1658
1659 read_unlock_bh(&sk->sk_callback_lock);
1660}
1661
1662static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
1663{
1664 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1665 struct sock *sk = transport->inet;
1666
1667 if (transport->rcvsize) {
1668 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1669 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
1670 }
1671 if (transport->sndsize) {
1672 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1673 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1674 sk->sk_write_space(sk);
1675 }
1676}
1677
1678/**
1679 * xs_udp_set_buffer_size - set send and receive limits
1680 * @xprt: generic transport
1681 * @sndsize: requested size of send buffer, in bytes
1682 * @rcvsize: requested size of receive buffer, in bytes
1683 *
1684 * Set socket send and receive buffer size limits.
1685 */
1686static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1687{
1688 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1689
1690 transport->sndsize = 0;
1691 if (sndsize)
1692 transport->sndsize = sndsize + 1024;
1693 transport->rcvsize = 0;
1694 if (rcvsize)
1695 transport->rcvsize = rcvsize + 1024;
1696
1697 xs_udp_do_set_buffer_size(xprt);
1698}
1699
1700/**
1701 * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1702 * @task: task that timed out
1703 *
1704 * Adjust the congestion window after a retransmit timeout has occurred.
1705 */
1706static void xs_udp_timer(struct rpc_xprt *xprt, struct rpc_task *task)
1707{
1708 xprt_adjust_cwnd(xprt, task, -ETIMEDOUT);
1709}
1710
1711static unsigned short xs_get_random_port(void)
1712{
1713 unsigned short range = xprt_max_resvport - xprt_min_resvport;
1714 unsigned short rand = (unsigned short) prandom_u32() % range;
1715 return rand + xprt_min_resvport;
1716}
1717
1718/**
1719 * xs_set_reuseaddr_port - set the socket's port and address reuse options
1720 * @sock: socket
1721 *
1722 * Note that this function has to be called on all sockets that share the
1723 * same port, and it must be called before binding.
1724 */
1725static void xs_sock_set_reuseport(struct socket *sock)
1726{
1727 int opt = 1;
1728
1729 kernel_setsockopt(sock, SOL_SOCKET, SO_REUSEPORT,
1730 (char *)&opt, sizeof(opt));
1731}
1732
1733static unsigned short xs_sock_getport(struct socket *sock)
1734{
1735 struct sockaddr_storage buf;
1736 int buflen;
1737 unsigned short port = 0;
1738
1739 if (kernel_getsockname(sock, (struct sockaddr *)&buf, &buflen) < 0)
1740 goto out;
1741 switch (buf.ss_family) {
1742 case AF_INET6:
1743 port = ntohs(((struct sockaddr_in6 *)&buf)->sin6_port);
1744 break;
1745 case AF_INET:
1746 port = ntohs(((struct sockaddr_in *)&buf)->sin_port);
1747 }
1748out:
1749 return port;
1750}
1751
1752/**
1753 * xs_set_port - reset the port number in the remote endpoint address
1754 * @xprt: generic transport
1755 * @port: new port number
1756 *
1757 */
1758static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1759{
1760 dprintk("RPC: setting port for xprt %p to %u\n", xprt, port);
1761
1762 rpc_set_port(xs_addr(xprt), port);
1763 xs_update_peer_port(xprt);
1764}
1765
1766static void xs_set_srcport(struct sock_xprt *transport, struct socket *sock)
1767{
1768 if (transport->srcport == 0)
1769 transport->srcport = xs_sock_getport(sock);
1770}
1771
1772static unsigned short xs_get_srcport(struct sock_xprt *transport)
1773{
1774 unsigned short port = transport->srcport;
1775
1776 if (port == 0 && transport->xprt.resvport)
1777 port = xs_get_random_port();
1778 return port;
1779}
1780
1781static unsigned short xs_next_srcport(struct sock_xprt *transport, unsigned short port)
1782{
1783 if (transport->srcport != 0)
1784 transport->srcport = 0;
1785 if (!transport->xprt.resvport)
1786 return 0;
1787 if (port <= xprt_min_resvport || port > xprt_max_resvport)
1788 return xprt_max_resvport;
1789 return --port;
1790}
1791static int xs_bind(struct sock_xprt *transport, struct socket *sock)
1792{
1793 struct sockaddr_storage myaddr;
1794 int err, nloop = 0;
1795 unsigned short port = xs_get_srcport(transport);
1796 unsigned short last;
1797
1798 /*
1799 * If we are asking for any ephemeral port (i.e. port == 0 &&
1800 * transport->xprt.resvport == 0), don't bind. Let the local
1801 * port selection happen implicitly when the socket is used
1802 * (for example at connect time).
1803 *
1804 * This ensures that we can continue to establish TCP
1805 * connections even when all local ephemeral ports are already
1806 * a part of some TCP connection. This makes no difference
1807 * for UDP sockets, but also doens't harm them.
1808 *
1809 * If we're asking for any reserved port (i.e. port == 0 &&
1810 * transport->xprt.resvport == 1) xs_get_srcport above will
1811 * ensure that port is non-zero and we will bind as needed.
1812 */
1813 if (port == 0)
1814 return 0;
1815
1816 memcpy(&myaddr, &transport->srcaddr, transport->xprt.addrlen);
1817 do {
1818 rpc_set_port((struct sockaddr *)&myaddr, port);
1819 err = kernel_bind(sock, (struct sockaddr *)&myaddr,
1820 transport->xprt.addrlen);
1821 if (err == 0) {
1822 transport->srcport = port;
1823 break;
1824 }
1825 last = port;
1826 port = xs_next_srcport(transport, port);
1827 if (port > last)
1828 nloop++;
1829 } while (err == -EADDRINUSE && nloop != 2);
1830
1831 if (myaddr.ss_family == AF_INET)
1832 dprintk("RPC: %s %pI4:%u: %s (%d)\n", __func__,
1833 &((struct sockaddr_in *)&myaddr)->sin_addr,
1834 port, err ? "failed" : "ok", err);
1835 else
1836 dprintk("RPC: %s %pI6:%u: %s (%d)\n", __func__,
1837 &((struct sockaddr_in6 *)&myaddr)->sin6_addr,
1838 port, err ? "failed" : "ok", err);
1839 return err;
1840}
1841
1842/*
1843 * We don't support autobind on AF_LOCAL sockets
1844 */
1845static void xs_local_rpcbind(struct rpc_task *task)
1846{
1847 xprt_set_bound(task->tk_xprt);
1848}
1849
1850static void xs_local_set_port(struct rpc_xprt *xprt, unsigned short port)
1851{
1852}
1853
1854#ifdef CONFIG_DEBUG_LOCK_ALLOC
1855static struct lock_class_key xs_key[2];
1856static struct lock_class_key xs_slock_key[2];
1857
1858static inline void xs_reclassify_socketu(struct socket *sock)
1859{
1860 struct sock *sk = sock->sk;
1861
1862 sock_lock_init_class_and_name(sk, "slock-AF_LOCAL-RPC",
1863 &xs_slock_key[1], "sk_lock-AF_LOCAL-RPC", &xs_key[1]);
1864}
1865
1866static inline void xs_reclassify_socket4(struct socket *sock)
1867{
1868 struct sock *sk = sock->sk;
1869
1870 sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1871 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1872}
1873
1874static inline void xs_reclassify_socket6(struct socket *sock)
1875{
1876 struct sock *sk = sock->sk;
1877
1878 sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1879 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
1880}
1881
1882static inline void xs_reclassify_socket(int family, struct socket *sock)
1883{
1884 WARN_ON_ONCE(sock_owned_by_user(sock->sk));
1885 if (sock_owned_by_user(sock->sk))
1886 return;
1887
1888 switch (family) {
1889 case AF_LOCAL:
1890 xs_reclassify_socketu(sock);
1891 break;
1892 case AF_INET:
1893 xs_reclassify_socket4(sock);
1894 break;
1895 case AF_INET6:
1896 xs_reclassify_socket6(sock);
1897 break;
1898 }
1899}
1900#else
1901static inline void xs_reclassify_socket(int family, struct socket *sock)
1902{
1903}
1904#endif
1905
1906static void xs_dummy_setup_socket(struct work_struct *work)
1907{
1908}
1909
1910static struct socket *xs_create_sock(struct rpc_xprt *xprt,
1911 struct sock_xprt *transport, int family, int type,
1912 int protocol, bool reuseport)
1913{
1914 struct socket *sock;
1915 int err;
1916
1917 err = __sock_create(xprt->xprt_net, family, type, protocol, &sock, 1);
1918 if (err < 0) {
1919 dprintk("RPC: can't create %d transport socket (%d).\n",
1920 protocol, -err);
1921 goto out;
1922 }
1923 xs_reclassify_socket(family, sock);
1924
1925 if (reuseport)
1926 xs_sock_set_reuseport(sock);
1927
1928 err = xs_bind(transport, sock);
1929 if (err) {
1930 sock_release(sock);
1931 goto out;
1932 }
1933
1934 return sock;
1935out:
1936 return ERR_PTR(err);
1937}
1938
1939static int xs_local_finish_connecting(struct rpc_xprt *xprt,
1940 struct socket *sock)
1941{
1942 struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1943 xprt);
1944
1945 if (!transport->inet) {
1946 struct sock *sk = sock->sk;
1947
1948 write_lock_bh(&sk->sk_callback_lock);
1949
1950 xs_save_old_callbacks(transport, sk);
1951
1952 sk->sk_user_data = xprt;
1953 sk->sk_data_ready = xs_data_ready;
1954 sk->sk_write_space = xs_udp_write_space;
1955 sk->sk_error_report = xs_error_report;
1956 sk->sk_allocation = GFP_NOIO;
1957
1958 xprt_clear_connected(xprt);
1959
1960 /* Reset to new socket */
1961 transport->sock = sock;
1962 transport->inet = sk;
1963
1964 write_unlock_bh(&sk->sk_callback_lock);
1965 }
1966
1967 /* Tell the socket layer to start connecting... */
1968 xprt->stat.connect_count++;
1969 xprt->stat.connect_start = jiffies;
1970 return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, 0);
1971}
1972
1973/**
1974 * xs_local_setup_socket - create AF_LOCAL socket, connect to a local endpoint
1975 * @transport: socket transport to connect
1976 */
1977static int xs_local_setup_socket(struct sock_xprt *transport)
1978{
1979 struct rpc_xprt *xprt = &transport->xprt;
1980 struct socket *sock;
1981 int status = -EIO;
1982
1983 status = __sock_create(xprt->xprt_net, AF_LOCAL,
1984 SOCK_STREAM, 0, &sock, 1);
1985 if (status < 0) {
1986 dprintk("RPC: can't create AF_LOCAL "
1987 "transport socket (%d).\n", -status);
1988 goto out;
1989 }
1990 xs_reclassify_socket(AF_LOCAL, sock);
1991
1992 dprintk("RPC: worker connecting xprt %p via AF_LOCAL to %s\n",
1993 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1994
1995 status = xs_local_finish_connecting(xprt, sock);
1996 trace_rpc_socket_connect(xprt, sock, status);
1997 switch (status) {
1998 case 0:
1999 dprintk("RPC: xprt %p connected to %s\n",
2000 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2001 xprt_set_connected(xprt);
2002 case -ENOBUFS:
2003 break;
2004 case -ENOENT:
2005 dprintk("RPC: xprt %p: socket %s does not exist\n",
2006 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2007 break;
2008 case -ECONNREFUSED:
2009 dprintk("RPC: xprt %p: connection refused for %s\n",
2010 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2011 break;
2012 default:
2013 printk(KERN_ERR "%s: unhandled error (%d) connecting to %s\n",
2014 __func__, -status,
2015 xprt->address_strings[RPC_DISPLAY_ADDR]);
2016 }
2017
2018out:
2019 xprt_clear_connecting(xprt);
2020 xprt_wake_pending_tasks(xprt, status);
2021 return status;
2022}
2023
2024static void xs_local_connect(struct rpc_xprt *xprt, struct rpc_task *task)
2025{
2026 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2027 int ret;
2028
2029 if (RPC_IS_ASYNC(task)) {
2030 /*
2031 * We want the AF_LOCAL connect to be resolved in the
2032 * filesystem namespace of the process making the rpc
2033 * call. Thus we connect synchronously.
2034 *
2035 * If we want to support asynchronous AF_LOCAL calls,
2036 * we'll need to figure out how to pass a namespace to
2037 * connect.
2038 */
2039 rpc_exit(task, -ENOTCONN);
2040 return;
2041 }
2042 ret = xs_local_setup_socket(transport);
2043 if (ret && !RPC_IS_SOFTCONN(task))
2044 msleep_interruptible(15000);
2045}
2046
2047#if IS_ENABLED(CONFIG_SUNRPC_SWAP)
2048/*
2049 * Note that this should be called with XPRT_LOCKED held (or when we otherwise
2050 * know that we have exclusive access to the socket), to guard against
2051 * races with xs_reset_transport.
2052 */
2053static void xs_set_memalloc(struct rpc_xprt *xprt)
2054{
2055 struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
2056 xprt);
2057
2058 /*
2059 * If there's no sock, then we have nothing to set. The
2060 * reconnecting process will get it for us.
2061 */
2062 if (!transport->inet)
2063 return;
2064 if (atomic_read(&xprt->swapper))
2065 sk_set_memalloc(transport->inet);
2066}
2067
2068/**
2069 * xs_enable_swap - Tag this transport as being used for swap.
2070 * @xprt: transport to tag
2071 *
2072 * Take a reference to this transport on behalf of the rpc_clnt, and
2073 * optionally mark it for swapping if it wasn't already.
2074 */
2075static int
2076xs_enable_swap(struct rpc_xprt *xprt)
2077{
2078 struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
2079
2080 if (atomic_inc_return(&xprt->swapper) != 1)
2081 return 0;
2082 if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE))
2083 return -ERESTARTSYS;
2084 if (xs->inet)
2085 sk_set_memalloc(xs->inet);
2086 xprt_release_xprt(xprt, NULL);
2087 return 0;
2088}
2089
2090/**
2091 * xs_disable_swap - Untag this transport as being used for swap.
2092 * @xprt: transport to tag
2093 *
2094 * Drop a "swapper" reference to this xprt on behalf of the rpc_clnt. If the
2095 * swapper refcount goes to 0, untag the socket as a memalloc socket.
2096 */
2097static void
2098xs_disable_swap(struct rpc_xprt *xprt)
2099{
2100 struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
2101
2102 if (!atomic_dec_and_test(&xprt->swapper))
2103 return;
2104 if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE))
2105 return;
2106 if (xs->inet)
2107 sk_clear_memalloc(xs->inet);
2108 xprt_release_xprt(xprt, NULL);
2109}
2110#else
2111static void xs_set_memalloc(struct rpc_xprt *xprt)
2112{
2113}
2114
2115static int
2116xs_enable_swap(struct rpc_xprt *xprt)
2117{
2118 return -EINVAL;
2119}
2120
2121static void
2122xs_disable_swap(struct rpc_xprt *xprt)
2123{
2124}
2125#endif
2126
2127static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2128{
2129 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2130
2131 if (!transport->inet) {
2132 struct sock *sk = sock->sk;
2133
2134 write_lock_bh(&sk->sk_callback_lock);
2135
2136 xs_save_old_callbacks(transport, sk);
2137
2138 sk->sk_user_data = xprt;
2139 sk->sk_data_ready = xs_data_ready;
2140 sk->sk_write_space = xs_udp_write_space;
2141 sk->sk_allocation = GFP_NOIO;
2142
2143 xprt_set_connected(xprt);
2144
2145 /* Reset to new socket */
2146 transport->sock = sock;
2147 transport->inet = sk;
2148
2149 xs_set_memalloc(xprt);
2150
2151 write_unlock_bh(&sk->sk_callback_lock);
2152 }
2153 xs_udp_do_set_buffer_size(xprt);
2154}
2155
2156static void xs_udp_setup_socket(struct work_struct *work)
2157{
2158 struct sock_xprt *transport =
2159 container_of(work, struct sock_xprt, connect_worker.work);
2160 struct rpc_xprt *xprt = &transport->xprt;
2161 struct socket *sock = transport->sock;
2162 int status = -EIO;
2163
2164 sock = xs_create_sock(xprt, transport,
2165 xs_addr(xprt)->sa_family, SOCK_DGRAM,
2166 IPPROTO_UDP, false);
2167 if (IS_ERR(sock))
2168 goto out;
2169
2170 dprintk("RPC: worker connecting xprt %p via %s to "
2171 "%s (port %s)\n", xprt,
2172 xprt->address_strings[RPC_DISPLAY_PROTO],
2173 xprt->address_strings[RPC_DISPLAY_ADDR],
2174 xprt->address_strings[RPC_DISPLAY_PORT]);
2175
2176 xs_udp_finish_connecting(xprt, sock);
2177 trace_rpc_socket_connect(xprt, sock, 0);
2178 status = 0;
2179out:
2180 xprt_unlock_connect(xprt, transport);
2181 xprt_clear_connecting(xprt);
2182 xprt_wake_pending_tasks(xprt, status);
2183}
2184
2185/**
2186 * xs_tcp_shutdown - gracefully shut down a TCP socket
2187 * @xprt: transport
2188 *
2189 * Initiates a graceful shutdown of the TCP socket by calling the
2190 * equivalent of shutdown(SHUT_RDWR);
2191 */
2192static void xs_tcp_shutdown(struct rpc_xprt *xprt)
2193{
2194 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2195 struct socket *sock = transport->sock;
2196
2197 if (sock == NULL)
2198 return;
2199 if (xprt_connected(xprt)) {
2200 kernel_sock_shutdown(sock, SHUT_RDWR);
2201 trace_rpc_socket_shutdown(xprt, sock);
2202 } else
2203 xs_reset_transport(transport);
2204}
2205
2206static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2207{
2208 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2209 int ret = -ENOTCONN;
2210
2211 if (!transport->inet) {
2212 struct sock *sk = sock->sk;
2213 unsigned int keepidle = xprt->timeout->to_initval / HZ;
2214 unsigned int keepcnt = xprt->timeout->to_retries + 1;
2215 unsigned int opt_on = 1;
2216 unsigned int timeo;
2217
2218 /* TCP Keepalive options */
2219 kernel_setsockopt(sock, SOL_SOCKET, SO_KEEPALIVE,
2220 (char *)&opt_on, sizeof(opt_on));
2221 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPIDLE,
2222 (char *)&keepidle, sizeof(keepidle));
2223 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPINTVL,
2224 (char *)&keepidle, sizeof(keepidle));
2225 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPCNT,
2226 (char *)&keepcnt, sizeof(keepcnt));
2227
2228 /* TCP user timeout (see RFC5482) */
2229 timeo = jiffies_to_msecs(xprt->timeout->to_initval) *
2230 (xprt->timeout->to_retries + 1);
2231 kernel_setsockopt(sock, SOL_TCP, TCP_USER_TIMEOUT,
2232 (char *)&timeo, sizeof(timeo));
2233
2234 write_lock_bh(&sk->sk_callback_lock);
2235
2236 xs_save_old_callbacks(transport, sk);
2237
2238 sk->sk_user_data = xprt;
2239 sk->sk_data_ready = xs_tcp_data_ready;
2240 sk->sk_state_change = xs_tcp_state_change;
2241 sk->sk_write_space = xs_tcp_write_space;
2242 sk->sk_error_report = xs_error_report;
2243 sk->sk_allocation = GFP_NOIO;
2244
2245 /* socket options */
2246 sock_reset_flag(sk, SOCK_LINGER);
2247 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
2248
2249 xprt_clear_connected(xprt);
2250
2251 /* Reset to new socket */
2252 transport->sock = sock;
2253 transport->inet = sk;
2254
2255 write_unlock_bh(&sk->sk_callback_lock);
2256 }
2257
2258 if (!xprt_bound(xprt))
2259 goto out;
2260
2261 xs_set_memalloc(xprt);
2262
2263 /* Tell the socket layer to start connecting... */
2264 xprt->stat.connect_count++;
2265 xprt->stat.connect_start = jiffies;
2266 set_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
2267 ret = kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
2268 switch (ret) {
2269 case 0:
2270 xs_set_srcport(transport, sock);
2271 case -EINPROGRESS:
2272 /* SYN_SENT! */
2273 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2274 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2275 }
2276out:
2277 return ret;
2278}
2279
2280/**
2281 * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint
2282 *
2283 * Invoked by a work queue tasklet.
2284 */
2285static void xs_tcp_setup_socket(struct work_struct *work)
2286{
2287 struct sock_xprt *transport =
2288 container_of(work, struct sock_xprt, connect_worker.work);
2289 struct socket *sock = transport->sock;
2290 struct rpc_xprt *xprt = &transport->xprt;
2291 int status = -EIO;
2292
2293 if (!sock) {
2294 sock = xs_create_sock(xprt, transport,
2295 xs_addr(xprt)->sa_family, SOCK_STREAM,
2296 IPPROTO_TCP, true);
2297 if (IS_ERR(sock)) {
2298 status = PTR_ERR(sock);
2299 goto out;
2300 }
2301 }
2302
2303 dprintk("RPC: worker connecting xprt %p via %s to "
2304 "%s (port %s)\n", xprt,
2305 xprt->address_strings[RPC_DISPLAY_PROTO],
2306 xprt->address_strings[RPC_DISPLAY_ADDR],
2307 xprt->address_strings[RPC_DISPLAY_PORT]);
2308
2309 status = xs_tcp_finish_connecting(xprt, sock);
2310 trace_rpc_socket_connect(xprt, sock, status);
2311 dprintk("RPC: %p connect status %d connected %d sock state %d\n",
2312 xprt, -status, xprt_connected(xprt),
2313 sock->sk->sk_state);
2314 switch (status) {
2315 default:
2316 printk("%s: connect returned unhandled error %d\n",
2317 __func__, status);
2318 case -EADDRNOTAVAIL:
2319 /* We're probably in TIME_WAIT. Get rid of existing socket,
2320 * and retry
2321 */
2322 xs_tcp_force_close(xprt);
2323 break;
2324 case 0:
2325 case -EINPROGRESS:
2326 case -EALREADY:
2327 xprt_unlock_connect(xprt, transport);
2328 return;
2329 case -EINVAL:
2330 /* Happens, for instance, if the user specified a link
2331 * local IPv6 address without a scope-id.
2332 */
2333 case -ECONNREFUSED:
2334 case -ECONNRESET:
2335 case -ENETUNREACH:
2336 case -EADDRINUSE:
2337 case -ENOBUFS:
2338 /* retry with existing socket, after a delay */
2339 xs_tcp_force_close(xprt);
2340 goto out;
2341 }
2342 status = -EAGAIN;
2343out:
2344 xprt_unlock_connect(xprt, transport);
2345 xprt_clear_connecting(xprt);
2346 xprt_wake_pending_tasks(xprt, status);
2347}
2348
2349/**
2350 * xs_connect - connect a socket to a remote endpoint
2351 * @xprt: pointer to transport structure
2352 * @task: address of RPC task that manages state of connect request
2353 *
2354 * TCP: If the remote end dropped the connection, delay reconnecting.
2355 *
2356 * UDP socket connects are synchronous, but we use a work queue anyway
2357 * to guarantee that even unprivileged user processes can set up a
2358 * socket on a privileged port.
2359 *
2360 * If a UDP socket connect fails, the delay behavior here prevents
2361 * retry floods (hard mounts).
2362 */
2363static void xs_connect(struct rpc_xprt *xprt, struct rpc_task *task)
2364{
2365 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2366
2367 WARN_ON_ONCE(!xprt_lock_connect(xprt, task, transport));
2368
2369 if (transport->sock != NULL) {
2370 dprintk("RPC: xs_connect delayed xprt %p for %lu "
2371 "seconds\n",
2372 xprt, xprt->reestablish_timeout / HZ);
2373
2374 /* Start by resetting any existing state */
2375 xs_reset_transport(transport);
2376
2377 queue_delayed_work(rpciod_workqueue,
2378 &transport->connect_worker,
2379 xprt->reestablish_timeout);
2380 xprt->reestablish_timeout <<= 1;
2381 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2382 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2383 if (xprt->reestablish_timeout > XS_TCP_MAX_REEST_TO)
2384 xprt->reestablish_timeout = XS_TCP_MAX_REEST_TO;
2385 } else {
2386 dprintk("RPC: xs_connect scheduled xprt %p\n", xprt);
2387 queue_delayed_work(rpciod_workqueue,
2388 &transport->connect_worker, 0);
2389 }
2390}
2391
2392/**
2393 * xs_local_print_stats - display AF_LOCAL socket-specifc stats
2394 * @xprt: rpc_xprt struct containing statistics
2395 * @seq: output file
2396 *
2397 */
2398static void xs_local_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2399{
2400 long idle_time = 0;
2401
2402 if (xprt_connected(xprt))
2403 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2404
2405 seq_printf(seq, "\txprt:\tlocal %lu %lu %lu %ld %lu %lu %lu "
2406 "%llu %llu %lu %llu %llu\n",
2407 xprt->stat.bind_count,
2408 xprt->stat.connect_count,
2409 xprt->stat.connect_time,
2410 idle_time,
2411 xprt->stat.sends,
2412 xprt->stat.recvs,
2413 xprt->stat.bad_xids,
2414 xprt->stat.req_u,
2415 xprt->stat.bklog_u,
2416 xprt->stat.max_slots,
2417 xprt->stat.sending_u,
2418 xprt->stat.pending_u);
2419}
2420
2421/**
2422 * xs_udp_print_stats - display UDP socket-specifc stats
2423 * @xprt: rpc_xprt struct containing statistics
2424 * @seq: output file
2425 *
2426 */
2427static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2428{
2429 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2430
2431 seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %llu %llu "
2432 "%lu %llu %llu\n",
2433 transport->srcport,
2434 xprt->stat.bind_count,
2435 xprt->stat.sends,
2436 xprt->stat.recvs,
2437 xprt->stat.bad_xids,
2438 xprt->stat.req_u,
2439 xprt->stat.bklog_u,
2440 xprt->stat.max_slots,
2441 xprt->stat.sending_u,
2442 xprt->stat.pending_u);
2443}
2444
2445/**
2446 * xs_tcp_print_stats - display TCP socket-specifc stats
2447 * @xprt: rpc_xprt struct containing statistics
2448 * @seq: output file
2449 *
2450 */
2451static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2452{
2453 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2454 long idle_time = 0;
2455
2456 if (xprt_connected(xprt))
2457 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2458
2459 seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu "
2460 "%llu %llu %lu %llu %llu\n",
2461 transport->srcport,
2462 xprt->stat.bind_count,
2463 xprt->stat.connect_count,
2464 xprt->stat.connect_time,
2465 idle_time,
2466 xprt->stat.sends,
2467 xprt->stat.recvs,
2468 xprt->stat.bad_xids,
2469 xprt->stat.req_u,
2470 xprt->stat.bklog_u,
2471 xprt->stat.max_slots,
2472 xprt->stat.sending_u,
2473 xprt->stat.pending_u);
2474}
2475
2476/*
2477 * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason
2478 * we allocate pages instead doing a kmalloc like rpc_malloc is because we want
2479 * to use the server side send routines.
2480 */
2481static void *bc_malloc(struct rpc_task *task, size_t size)
2482{
2483 struct page *page;
2484 struct rpc_buffer *buf;
2485
2486 WARN_ON_ONCE(size > PAGE_SIZE - sizeof(struct rpc_buffer));
2487 if (size > PAGE_SIZE - sizeof(struct rpc_buffer))
2488 return NULL;
2489
2490 page = alloc_page(GFP_KERNEL);
2491 if (!page)
2492 return NULL;
2493
2494 buf = page_address(page);
2495 buf->len = PAGE_SIZE;
2496
2497 return buf->data;
2498}
2499
2500/*
2501 * Free the space allocated in the bc_alloc routine
2502 */
2503static void bc_free(void *buffer)
2504{
2505 struct rpc_buffer *buf;
2506
2507 if (!buffer)
2508 return;
2509
2510 buf = container_of(buffer, struct rpc_buffer, data);
2511 free_page((unsigned long)buf);
2512}
2513
2514/*
2515 * Use the svc_sock to send the callback. Must be called with svsk->sk_mutex
2516 * held. Borrows heavily from svc_tcp_sendto and xs_tcp_send_request.
2517 */
2518static int bc_sendto(struct rpc_rqst *req)
2519{
2520 int len;
2521 struct xdr_buf *xbufp = &req->rq_snd_buf;
2522 struct rpc_xprt *xprt = req->rq_xprt;
2523 struct sock_xprt *transport =
2524 container_of(xprt, struct sock_xprt, xprt);
2525 struct socket *sock = transport->sock;
2526 unsigned long headoff;
2527 unsigned long tailoff;
2528
2529 xs_encode_stream_record_marker(xbufp);
2530
2531 tailoff = (unsigned long)xbufp->tail[0].iov_base & ~PAGE_MASK;
2532 headoff = (unsigned long)xbufp->head[0].iov_base & ~PAGE_MASK;
2533 len = svc_send_common(sock, xbufp,
2534 virt_to_page(xbufp->head[0].iov_base), headoff,
2535 xbufp->tail[0].iov_base, tailoff);
2536
2537 if (len != xbufp->len) {
2538 printk(KERN_NOTICE "Error sending entire callback!\n");
2539 len = -EAGAIN;
2540 }
2541
2542 return len;
2543}
2544
2545/*
2546 * The send routine. Borrows from svc_send
2547 */
2548static int bc_send_request(struct rpc_task *task)
2549{
2550 struct rpc_rqst *req = task->tk_rqstp;
2551 struct svc_xprt *xprt;
2552 int len;
2553
2554 dprintk("sending request with xid: %08x\n", ntohl(req->rq_xid));
2555 /*
2556 * Get the server socket associated with this callback xprt
2557 */
2558 xprt = req->rq_xprt->bc_xprt;
2559
2560 /*
2561 * Grab the mutex to serialize data as the connection is shared
2562 * with the fore channel
2563 */
2564 if (!mutex_trylock(&xprt->xpt_mutex)) {
2565 rpc_sleep_on(&xprt->xpt_bc_pending, task, NULL);
2566 if (!mutex_trylock(&xprt->xpt_mutex))
2567 return -EAGAIN;
2568 rpc_wake_up_queued_task(&xprt->xpt_bc_pending, task);
2569 }
2570 if (test_bit(XPT_DEAD, &xprt->xpt_flags))
2571 len = -ENOTCONN;
2572 else
2573 len = bc_sendto(req);
2574 mutex_unlock(&xprt->xpt_mutex);
2575
2576 if (len > 0)
2577 len = 0;
2578
2579 return len;
2580}
2581
2582/*
2583 * The close routine. Since this is client initiated, we do nothing
2584 */
2585
2586static void bc_close(struct rpc_xprt *xprt)
2587{
2588}
2589
2590/*
2591 * The xprt destroy routine. Again, because this connection is client
2592 * initiated, we do nothing
2593 */
2594
2595static void bc_destroy(struct rpc_xprt *xprt)
2596{
2597 dprintk("RPC: bc_destroy xprt %p\n", xprt);
2598
2599 xs_xprt_free(xprt);
2600 module_put(THIS_MODULE);
2601}
2602
2603static struct rpc_xprt_ops xs_local_ops = {
2604 .reserve_xprt = xprt_reserve_xprt,
2605 .release_xprt = xs_tcp_release_xprt,
2606 .alloc_slot = xprt_alloc_slot,
2607 .rpcbind = xs_local_rpcbind,
2608 .set_port = xs_local_set_port,
2609 .connect = xs_local_connect,
2610 .buf_alloc = rpc_malloc,
2611 .buf_free = rpc_free,
2612 .send_request = xs_local_send_request,
2613 .set_retrans_timeout = xprt_set_retrans_timeout_def,
2614 .close = xs_close,
2615 .destroy = xs_destroy,
2616 .print_stats = xs_local_print_stats,
2617 .enable_swap = xs_enable_swap,
2618 .disable_swap = xs_disable_swap,
2619};
2620
2621static struct rpc_xprt_ops xs_udp_ops = {
2622 .set_buffer_size = xs_udp_set_buffer_size,
2623 .reserve_xprt = xprt_reserve_xprt_cong,
2624 .release_xprt = xprt_release_xprt_cong,
2625 .alloc_slot = xprt_alloc_slot,
2626 .rpcbind = rpcb_getport_async,
2627 .set_port = xs_set_port,
2628 .connect = xs_connect,
2629 .buf_alloc = rpc_malloc,
2630 .buf_free = rpc_free,
2631 .send_request = xs_udp_send_request,
2632 .set_retrans_timeout = xprt_set_retrans_timeout_rtt,
2633 .timer = xs_udp_timer,
2634 .release_request = xprt_release_rqst_cong,
2635 .close = xs_close,
2636 .destroy = xs_destroy,
2637 .print_stats = xs_udp_print_stats,
2638 .enable_swap = xs_enable_swap,
2639 .disable_swap = xs_disable_swap,
2640 .inject_disconnect = xs_inject_disconnect,
2641};
2642
2643static struct rpc_xprt_ops xs_tcp_ops = {
2644 .reserve_xprt = xprt_reserve_xprt,
2645 .release_xprt = xs_tcp_release_xprt,
2646 .alloc_slot = xprt_lock_and_alloc_slot,
2647 .rpcbind = rpcb_getport_async,
2648 .set_port = xs_set_port,
2649 .connect = xs_connect,
2650 .buf_alloc = rpc_malloc,
2651 .buf_free = rpc_free,
2652 .send_request = xs_tcp_send_request,
2653 .set_retrans_timeout = xprt_set_retrans_timeout_def,
2654 .close = xs_tcp_shutdown,
2655 .destroy = xs_destroy,
2656 .print_stats = xs_tcp_print_stats,
2657 .enable_swap = xs_enable_swap,
2658 .disable_swap = xs_disable_swap,
2659 .inject_disconnect = xs_inject_disconnect,
2660#ifdef CONFIG_SUNRPC_BACKCHANNEL
2661 .bc_setup = xprt_setup_bc,
2662 .bc_up = xs_tcp_bc_up,
2663 .bc_free_rqst = xprt_free_bc_rqst,
2664 .bc_destroy = xprt_destroy_bc,
2665#endif
2666};
2667
2668/*
2669 * The rpc_xprt_ops for the server backchannel
2670 */
2671
2672static struct rpc_xprt_ops bc_tcp_ops = {
2673 .reserve_xprt = xprt_reserve_xprt,
2674 .release_xprt = xprt_release_xprt,
2675 .alloc_slot = xprt_alloc_slot,
2676 .buf_alloc = bc_malloc,
2677 .buf_free = bc_free,
2678 .send_request = bc_send_request,
2679 .set_retrans_timeout = xprt_set_retrans_timeout_def,
2680 .close = bc_close,
2681 .destroy = bc_destroy,
2682 .print_stats = xs_tcp_print_stats,
2683 .enable_swap = xs_enable_swap,
2684 .disable_swap = xs_disable_swap,
2685 .inject_disconnect = xs_inject_disconnect,
2686};
2687
2688static int xs_init_anyaddr(const int family, struct sockaddr *sap)
2689{
2690 static const struct sockaddr_in sin = {
2691 .sin_family = AF_INET,
2692 .sin_addr.s_addr = htonl(INADDR_ANY),
2693 };
2694 static const struct sockaddr_in6 sin6 = {
2695 .sin6_family = AF_INET6,
2696 .sin6_addr = IN6ADDR_ANY_INIT,
2697 };
2698
2699 switch (family) {
2700 case AF_LOCAL:
2701 break;
2702 case AF_INET:
2703 memcpy(sap, &sin, sizeof(sin));
2704 break;
2705 case AF_INET6:
2706 memcpy(sap, &sin6, sizeof(sin6));
2707 break;
2708 default:
2709 dprintk("RPC: %s: Bad address family\n", __func__);
2710 return -EAFNOSUPPORT;
2711 }
2712 return 0;
2713}
2714
2715static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
2716 unsigned int slot_table_size,
2717 unsigned int max_slot_table_size)
2718{
2719 struct rpc_xprt *xprt;
2720 struct sock_xprt *new;
2721
2722 if (args->addrlen > sizeof(xprt->addr)) {
2723 dprintk("RPC: xs_setup_xprt: address too large\n");
2724 return ERR_PTR(-EBADF);
2725 }
2726
2727 xprt = xprt_alloc(args->net, sizeof(*new), slot_table_size,
2728 max_slot_table_size);
2729 if (xprt == NULL) {
2730 dprintk("RPC: xs_setup_xprt: couldn't allocate "
2731 "rpc_xprt\n");
2732 return ERR_PTR(-ENOMEM);
2733 }
2734
2735 new = container_of(xprt, struct sock_xprt, xprt);
2736 mutex_init(&new->recv_mutex);
2737 memcpy(&xprt->addr, args->dstaddr, args->addrlen);
2738 xprt->addrlen = args->addrlen;
2739 if (args->srcaddr)
2740 memcpy(&new->srcaddr, args->srcaddr, args->addrlen);
2741 else {
2742 int err;
2743 err = xs_init_anyaddr(args->dstaddr->sa_family,
2744 (struct sockaddr *)&new->srcaddr);
2745 if (err != 0) {
2746 xprt_free(xprt);
2747 return ERR_PTR(err);
2748 }
2749 }
2750
2751 return xprt;
2752}
2753
2754static const struct rpc_timeout xs_local_default_timeout = {
2755 .to_initval = 10 * HZ,
2756 .to_maxval = 10 * HZ,
2757 .to_retries = 2,
2758};
2759
2760/**
2761 * xs_setup_local - Set up transport to use an AF_LOCAL socket
2762 * @args: rpc transport creation arguments
2763 *
2764 * AF_LOCAL is a "tpi_cots_ord" transport, just like TCP
2765 */
2766static struct rpc_xprt *xs_setup_local(struct xprt_create *args)
2767{
2768 struct sockaddr_un *sun = (struct sockaddr_un *)args->dstaddr;
2769 struct sock_xprt *transport;
2770 struct rpc_xprt *xprt;
2771 struct rpc_xprt *ret;
2772
2773 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2774 xprt_max_tcp_slot_table_entries);
2775 if (IS_ERR(xprt))
2776 return xprt;
2777 transport = container_of(xprt, struct sock_xprt, xprt);
2778
2779 xprt->prot = 0;
2780 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2781 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2782
2783 xprt->bind_timeout = XS_BIND_TO;
2784 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2785 xprt->idle_timeout = XS_IDLE_DISC_TO;
2786
2787 xprt->ops = &xs_local_ops;
2788 xprt->timeout = &xs_local_default_timeout;
2789
2790 INIT_WORK(&transport->recv_worker, xs_local_data_receive_workfn);
2791 INIT_DELAYED_WORK(&transport->connect_worker,
2792 xs_dummy_setup_socket);
2793
2794 switch (sun->sun_family) {
2795 case AF_LOCAL:
2796 if (sun->sun_path[0] != '/') {
2797 dprintk("RPC: bad AF_LOCAL address: %s\n",
2798 sun->sun_path);
2799 ret = ERR_PTR(-EINVAL);
2800 goto out_err;
2801 }
2802 xprt_set_bound(xprt);
2803 xs_format_peer_addresses(xprt, "local", RPCBIND_NETID_LOCAL);
2804 ret = ERR_PTR(xs_local_setup_socket(transport));
2805 if (ret)
2806 goto out_err;
2807 break;
2808 default:
2809 ret = ERR_PTR(-EAFNOSUPPORT);
2810 goto out_err;
2811 }
2812
2813 dprintk("RPC: set up xprt to %s via AF_LOCAL\n",
2814 xprt->address_strings[RPC_DISPLAY_ADDR]);
2815
2816 if (try_module_get(THIS_MODULE))
2817 return xprt;
2818 ret = ERR_PTR(-EINVAL);
2819out_err:
2820 xs_xprt_free(xprt);
2821 return ret;
2822}
2823
2824static const struct rpc_timeout xs_udp_default_timeout = {
2825 .to_initval = 5 * HZ,
2826 .to_maxval = 30 * HZ,
2827 .to_increment = 5 * HZ,
2828 .to_retries = 5,
2829};
2830
2831/**
2832 * xs_setup_udp - Set up transport to use a UDP socket
2833 * @args: rpc transport creation arguments
2834 *
2835 */
2836static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
2837{
2838 struct sockaddr *addr = args->dstaddr;
2839 struct rpc_xprt *xprt;
2840 struct sock_xprt *transport;
2841 struct rpc_xprt *ret;
2842
2843 xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries,
2844 xprt_udp_slot_table_entries);
2845 if (IS_ERR(xprt))
2846 return xprt;
2847 transport = container_of(xprt, struct sock_xprt, xprt);
2848
2849 xprt->prot = IPPROTO_UDP;
2850 xprt->tsh_size = 0;
2851 /* XXX: header size can vary due to auth type, IPv6, etc. */
2852 xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
2853
2854 xprt->bind_timeout = XS_BIND_TO;
2855 xprt->reestablish_timeout = XS_UDP_REEST_TO;
2856 xprt->idle_timeout = XS_IDLE_DISC_TO;
2857
2858 xprt->ops = &xs_udp_ops;
2859
2860 xprt->timeout = &xs_udp_default_timeout;
2861
2862 INIT_WORK(&transport->recv_worker, xs_udp_data_receive_workfn);
2863 INIT_DELAYED_WORK(&transport->connect_worker, xs_udp_setup_socket);
2864
2865 switch (addr->sa_family) {
2866 case AF_INET:
2867 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2868 xprt_set_bound(xprt);
2869
2870 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
2871 break;
2872 case AF_INET6:
2873 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2874 xprt_set_bound(xprt);
2875
2876 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
2877 break;
2878 default:
2879 ret = ERR_PTR(-EAFNOSUPPORT);
2880 goto out_err;
2881 }
2882
2883 if (xprt_bound(xprt))
2884 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2885 xprt->address_strings[RPC_DISPLAY_ADDR],
2886 xprt->address_strings[RPC_DISPLAY_PORT],
2887 xprt->address_strings[RPC_DISPLAY_PROTO]);
2888 else
2889 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
2890 xprt->address_strings[RPC_DISPLAY_ADDR],
2891 xprt->address_strings[RPC_DISPLAY_PROTO]);
2892
2893 if (try_module_get(THIS_MODULE))
2894 return xprt;
2895 ret = ERR_PTR(-EINVAL);
2896out_err:
2897 xs_xprt_free(xprt);
2898 return ret;
2899}
2900
2901static const struct rpc_timeout xs_tcp_default_timeout = {
2902 .to_initval = 60 * HZ,
2903 .to_maxval = 60 * HZ,
2904 .to_retries = 2,
2905};
2906
2907/**
2908 * xs_setup_tcp - Set up transport to use a TCP socket
2909 * @args: rpc transport creation arguments
2910 *
2911 */
2912static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
2913{
2914 struct sockaddr *addr = args->dstaddr;
2915 struct rpc_xprt *xprt;
2916 struct sock_xprt *transport;
2917 struct rpc_xprt *ret;
2918 unsigned int max_slot_table_size = xprt_max_tcp_slot_table_entries;
2919
2920 if (args->flags & XPRT_CREATE_INFINITE_SLOTS)
2921 max_slot_table_size = RPC_MAX_SLOT_TABLE_LIMIT;
2922
2923 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2924 max_slot_table_size);
2925 if (IS_ERR(xprt))
2926 return xprt;
2927 transport = container_of(xprt, struct sock_xprt, xprt);
2928
2929 xprt->prot = IPPROTO_TCP;
2930 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2931 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2932
2933 xprt->bind_timeout = XS_BIND_TO;
2934 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2935 xprt->idle_timeout = XS_IDLE_DISC_TO;
2936
2937 xprt->ops = &xs_tcp_ops;
2938 xprt->timeout = &xs_tcp_default_timeout;
2939
2940 INIT_WORK(&transport->recv_worker, xs_tcp_data_receive_workfn);
2941 INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_setup_socket);
2942
2943 switch (addr->sa_family) {
2944 case AF_INET:
2945 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2946 xprt_set_bound(xprt);
2947
2948 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
2949 break;
2950 case AF_INET6:
2951 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2952 xprt_set_bound(xprt);
2953
2954 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
2955 break;
2956 default:
2957 ret = ERR_PTR(-EAFNOSUPPORT);
2958 goto out_err;
2959 }
2960
2961 if (xprt_bound(xprt))
2962 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2963 xprt->address_strings[RPC_DISPLAY_ADDR],
2964 xprt->address_strings[RPC_DISPLAY_PORT],
2965 xprt->address_strings[RPC_DISPLAY_PROTO]);
2966 else
2967 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
2968 xprt->address_strings[RPC_DISPLAY_ADDR],
2969 xprt->address_strings[RPC_DISPLAY_PROTO]);
2970
2971 if (try_module_get(THIS_MODULE))
2972 return xprt;
2973 ret = ERR_PTR(-EINVAL);
2974out_err:
2975 xs_xprt_free(xprt);
2976 return ret;
2977}
2978
2979/**
2980 * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket
2981 * @args: rpc transport creation arguments
2982 *
2983 */
2984static struct rpc_xprt *xs_setup_bc_tcp(struct xprt_create *args)
2985{
2986 struct sockaddr *addr = args->dstaddr;
2987 struct rpc_xprt *xprt;
2988 struct sock_xprt *transport;
2989 struct svc_sock *bc_sock;
2990 struct rpc_xprt *ret;
2991
2992 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2993 xprt_tcp_slot_table_entries);
2994 if (IS_ERR(xprt))
2995 return xprt;
2996 transport = container_of(xprt, struct sock_xprt, xprt);
2997
2998 xprt->prot = IPPROTO_TCP;
2999 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
3000 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
3001 xprt->timeout = &xs_tcp_default_timeout;
3002
3003 /* backchannel */
3004 xprt_set_bound(xprt);
3005 xprt->bind_timeout = 0;
3006 xprt->reestablish_timeout = 0;
3007 xprt->idle_timeout = 0;
3008
3009 xprt->ops = &bc_tcp_ops;
3010
3011 switch (addr->sa_family) {
3012 case AF_INET:
3013 xs_format_peer_addresses(xprt, "tcp",
3014 RPCBIND_NETID_TCP);
3015 break;
3016 case AF_INET6:
3017 xs_format_peer_addresses(xprt, "tcp",
3018 RPCBIND_NETID_TCP6);
3019 break;
3020 default:
3021 ret = ERR_PTR(-EAFNOSUPPORT);
3022 goto out_err;
3023 }
3024
3025 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
3026 xprt->address_strings[RPC_DISPLAY_ADDR],
3027 xprt->address_strings[RPC_DISPLAY_PORT],
3028 xprt->address_strings[RPC_DISPLAY_PROTO]);
3029
3030 /*
3031 * Once we've associated a backchannel xprt with a connection,
3032 * we want to keep it around as long as the connection lasts,
3033 * in case we need to start using it for a backchannel again;
3034 * this reference won't be dropped until bc_xprt is destroyed.
3035 */
3036 xprt_get(xprt);
3037 args->bc_xprt->xpt_bc_xprt = xprt;
3038 xprt->bc_xprt = args->bc_xprt;
3039 bc_sock = container_of(args->bc_xprt, struct svc_sock, sk_xprt);
3040 transport->sock = bc_sock->sk_sock;
3041 transport->inet = bc_sock->sk_sk;
3042
3043 /*
3044 * Since we don't want connections for the backchannel, we set
3045 * the xprt status to connected
3046 */
3047 xprt_set_connected(xprt);
3048
3049 if (try_module_get(THIS_MODULE))
3050 return xprt;
3051
3052 args->bc_xprt->xpt_bc_xprt = NULL;
3053 xprt_put(xprt);
3054 ret = ERR_PTR(-EINVAL);
3055out_err:
3056 xs_xprt_free(xprt);
3057 return ret;
3058}
3059
3060static struct xprt_class xs_local_transport = {
3061 .list = LIST_HEAD_INIT(xs_local_transport.list),
3062 .name = "named UNIX socket",
3063 .owner = THIS_MODULE,
3064 .ident = XPRT_TRANSPORT_LOCAL,
3065 .setup = xs_setup_local,
3066};
3067
3068static struct xprt_class xs_udp_transport = {
3069 .list = LIST_HEAD_INIT(xs_udp_transport.list),
3070 .name = "udp",
3071 .owner = THIS_MODULE,
3072 .ident = XPRT_TRANSPORT_UDP,
3073 .setup = xs_setup_udp,
3074};
3075
3076static struct xprt_class xs_tcp_transport = {
3077 .list = LIST_HEAD_INIT(xs_tcp_transport.list),
3078 .name = "tcp",
3079 .owner = THIS_MODULE,
3080 .ident = XPRT_TRANSPORT_TCP,
3081 .setup = xs_setup_tcp,
3082};
3083
3084static struct xprt_class xs_bc_tcp_transport = {
3085 .list = LIST_HEAD_INIT(xs_bc_tcp_transport.list),
3086 .name = "tcp NFSv4.1 backchannel",
3087 .owner = THIS_MODULE,
3088 .ident = XPRT_TRANSPORT_BC_TCP,
3089 .setup = xs_setup_bc_tcp,
3090};
3091
3092/**
3093 * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
3094 *
3095 */
3096int init_socket_xprt(void)
3097{
3098#if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
3099 if (!sunrpc_table_header)
3100 sunrpc_table_header = register_sysctl_table(sunrpc_table);
3101#endif
3102
3103 xprt_register_transport(&xs_local_transport);
3104 xprt_register_transport(&xs_udp_transport);
3105 xprt_register_transport(&xs_tcp_transport);
3106 xprt_register_transport(&xs_bc_tcp_transport);
3107
3108 return 0;
3109}
3110
3111/**
3112 * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
3113 *
3114 */
3115void cleanup_socket_xprt(void)
3116{
3117#if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
3118 if (sunrpc_table_header) {
3119 unregister_sysctl_table(sunrpc_table_header);
3120 sunrpc_table_header = NULL;
3121 }
3122#endif
3123
3124 xprt_unregister_transport(&xs_local_transport);
3125 xprt_unregister_transport(&xs_udp_transport);
3126 xprt_unregister_transport(&xs_tcp_transport);
3127 xprt_unregister_transport(&xs_bc_tcp_transport);
3128}
3129
3130static int param_set_uint_minmax(const char *val,
3131 const struct kernel_param *kp,
3132 unsigned int min, unsigned int max)
3133{
3134 unsigned int num;
3135 int ret;
3136
3137 if (!val)
3138 return -EINVAL;
3139 ret = kstrtouint(val, 0, &num);
3140 if (ret == -EINVAL || num < min || num > max)
3141 return -EINVAL;
3142 *((unsigned int *)kp->arg) = num;
3143 return 0;
3144}
3145
3146static int param_set_portnr(const char *val, const struct kernel_param *kp)
3147{
3148 return param_set_uint_minmax(val, kp,
3149 RPC_MIN_RESVPORT,
3150 RPC_MAX_RESVPORT);
3151}
3152
3153static const struct kernel_param_ops param_ops_portnr = {
3154 .set = param_set_portnr,
3155 .get = param_get_uint,
3156};
3157
3158#define param_check_portnr(name, p) \
3159 __param_check(name, p, unsigned int);
3160
3161module_param_named(min_resvport, xprt_min_resvport, portnr, 0644);
3162module_param_named(max_resvport, xprt_max_resvport, portnr, 0644);
3163
3164static int param_set_slot_table_size(const char *val,
3165 const struct kernel_param *kp)
3166{
3167 return param_set_uint_minmax(val, kp,
3168 RPC_MIN_SLOT_TABLE,
3169 RPC_MAX_SLOT_TABLE);
3170}
3171
3172static const struct kernel_param_ops param_ops_slot_table_size = {
3173 .set = param_set_slot_table_size,
3174 .get = param_get_uint,
3175};
3176
3177#define param_check_slot_table_size(name, p) \
3178 __param_check(name, p, unsigned int);
3179
3180static int param_set_max_slot_table_size(const char *val,
3181 const struct kernel_param *kp)
3182{
3183 return param_set_uint_minmax(val, kp,
3184 RPC_MIN_SLOT_TABLE,
3185 RPC_MAX_SLOT_TABLE_LIMIT);
3186}
3187
3188static const struct kernel_param_ops param_ops_max_slot_table_size = {
3189 .set = param_set_max_slot_table_size,
3190 .get = param_get_uint,
3191};
3192
3193#define param_check_max_slot_table_size(name, p) \
3194 __param_check(name, p, unsigned int);
3195
3196module_param_named(tcp_slot_table_entries, xprt_tcp_slot_table_entries,
3197 slot_table_size, 0644);
3198module_param_named(tcp_max_slot_table_entries, xprt_max_tcp_slot_table_entries,
3199 max_slot_table_size, 0644);
3200module_param_named(udp_slot_table_entries, xprt_udp_slot_table_entries,
3201 slot_table_size, 0644);
3202