projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
missing bits of "iov_iter: Separate type from direction and use accessor functions"
[linux-2.6-block.git] / net / sunrpc / xprtsock.c
CommitLineData
b2441318 1// SPDX-License-Identifier: GPL-2.0
a246b010
CL		 2/*
3 * linux/net/sunrpc/xprtsock.c
4 *
5 * Client-side transport implementation for sockets.
6 *
113aa838
AC		 7 * TCP callback races fixes (C) 1998 Red Hat
8 * TCP send fixes (C) 1998 Red Hat
a246b010
CL		 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>
55aa4f58
CL		 15 *
16 * IP socket transport implementation, (C) 2005 Chuck Lever <cel@netapp.com>
8f9d5b1a
CL		 17 *
18 * IPv6 support contributed by Gilles Quillard, Bull Open Source, 2005.
19 * <gilles.quillard@bull.net>
a246b010
CL		 20 */
21
22#include <linux/types.h>
176e21ee 23#include <linux/string.h>
a246b010 24#include <linux/slab.h>
bc25571e 25#include <linux/module.h>
a246b010 26#include <linux/capability.h>
a246b010
CL		 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>
176e21ee 33#include <linux/un.h>
a246b010
CL		 34#include <linux/udp.h>
35#include <linux/tcp.h>
36#include <linux/sunrpc/clnt.h>
5976687a 37#include <linux/sunrpc/addr.h>
02107148 38#include <linux/sunrpc/sched.h>
4cfc7e60 39#include <linux/sunrpc/svcsock.h>
49c36fcc 40#include <linux/sunrpc/xprtsock.h>
a246b010 41#include <linux/file.h>
9e00abc3 42#ifdef CONFIG_SUNRPC_BACKCHANNEL
44b98efd
RL		 43#include <linux/sunrpc/bc_xprt.h>
44#endif
a246b010
CL		 45
46#include <net/sock.h>
47#include <net/checksum.h>
48#include <net/udp.h>
49#include <net/tcp.h>
277e4ab7
TM		 50#include <linux/bvec.h>
51#include <linux/uio.h>
a246b010 52
40b5ea0c
TM		 53#include <trace/events/sunrpc.h>
54
4cfc7e60 55#include "sunrpc.h"
176e21ee
CL		 56
57static void xs_close(struct rpc_xprt *xprt);
7196dbb0
TM		 58static void xs_tcp_set_socket_timeouts(struct rpc_xprt *xprt,
59 struct socket *sock);
176e21ee 60
c556b754
CL		 61/*
62 * xprtsock tunables
63 */
09acfea5
TM		 64static unsigned int xprt_udp_slot_table_entries = RPC_DEF_SLOT_TABLE;
65static unsigned int xprt_tcp_slot_table_entries = RPC_MIN_SLOT_TABLE;
66static unsigned int xprt_max_tcp_slot_table_entries = RPC_MAX_SLOT_TABLE;
c556b754 67
09acfea5
TM		 68static unsigned int xprt_min_resvport = RPC_DEF_MIN_RESVPORT;
69static unsigned int xprt_max_resvport = RPC_DEF_MAX_RESVPORT;
c556b754 70
54c09874
TM		 71#if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
72
7d1e8255 73#define XS_TCP_LINGER_TO (15U * HZ)
25fe6142 74static unsigned int xs_tcp_fin_timeout __read_mostly = XS_TCP_LINGER_TO;
7d1e8255 75
fbf76683
CL		 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
fbf76683
CL		 85static unsigned int min_slot_table_size = RPC_MIN_SLOT_TABLE;
86static unsigned int max_slot_table_size = RPC_MAX_SLOT_TABLE;
d9ba131d 87static unsigned int max_tcp_slot_table_limit = RPC_MAX_SLOT_TABLE_LIMIT;
fbf76683
CL		 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 */
fe2c6338 97static struct ctl_table xs_tunables_table[] = {
fbf76683 98 {
fbf76683
CL		 99 .procname = "udp_slot_table_entries",
100 .data = &xprt_udp_slot_table_entries,
101 .maxlen = sizeof(unsigned int),
102 .mode = 0644,
6d456111 103 .proc_handler = proc_dointvec_minmax,
fbf76683
CL		 104 .extra1 = &min_slot_table_size,
105 .extra2 = &max_slot_table_size
106 },
107 {
fbf76683
CL		 108 .procname = "tcp_slot_table_entries",
109 .data = &xprt_tcp_slot_table_entries,
110 .maxlen = sizeof(unsigned int),
111 .mode = 0644,
6d456111 112 .proc_handler = proc_dointvec_minmax,
fbf76683
CL		 113 .extra1 = &min_slot_table_size,
114 .extra2 = &max_slot_table_size
115 },
d9ba131d
TM		 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 },
fbf76683 125 {
fbf76683
CL		 126 .procname = "min_resvport",
127 .data = &xprt_min_resvport,
128 .maxlen = sizeof(unsigned int),
129 .mode = 0644,
6d456111 130 .proc_handler = proc_dointvec_minmax,
fbf76683 131 .extra1 = &xprt_min_resvport_limit,
826799e6 132 .extra2 = &xprt_max_resvport_limit
fbf76683
CL		 133 },
134 {
fbf76683
CL		 135 .procname = "max_resvport",
136 .data = &xprt_max_resvport,
137 .maxlen = sizeof(unsigned int),
138 .mode = 0644,
6d456111 139 .proc_handler = proc_dointvec_minmax,
826799e6 140 .extra1 = &xprt_min_resvport_limit,
fbf76683
CL		 141 .extra2 = &xprt_max_resvport_limit
142 },
25fe6142
TM		 143 {
144 .procname = "tcp_fin_timeout",
145 .data = &xs_tcp_fin_timeout,
146 .maxlen = sizeof(xs_tcp_fin_timeout),
147 .mode = 0644,
6d456111 148 .proc_handler = proc_dointvec_jiffies,
fbf76683 149 },
f8572d8f 150 { },
fbf76683
CL		 151};
152
fe2c6338 153static struct ctl_table sunrpc_table[] = {
fbf76683 154 {
fbf76683
CL		 155 .procname = "sunrpc",
156 .mode = 0555,
157 .child = xs_tunables_table
158 },
f8572d8f 159 { },
fbf76683
CL		 160};
161
162#endif
163
03bf4b70
CL		 164/*
165 * Wait duration for a reply from the RPC portmapper.
166 */
167#define XS_BIND_TO (60U * HZ)
168
169/*
170 * Delay if a UDP socket connect error occurs. This is most likely some
171 * kind of resource problem on the local host.
172 */
173#define XS_UDP_REEST_TO (2U * HZ)
174
175/*
176 * The reestablish timeout allows clients to delay for a bit before attempting
177 * to reconnect to a server that just dropped our connection.
178 *
179 * We implement an exponential backoff when trying to reestablish a TCP
180 * transport connection with the server. Some servers like to drop a TCP
181 * connection when they are overworked, so we start with a short timeout and
182 * increase over time if the server is down or not responding.
183 */
184#define XS_TCP_INIT_REEST_TO (3U * HZ)
03bf4b70
CL		 185
186/*
187 * TCP idle timeout; client drops the transport socket if it is idle
188 * for this long. Note that we also timeout UDP sockets to prevent
189 * holding port numbers when there is no RPC traffic.
190 */
191#define XS_IDLE_DISC_TO (5U * 60 * HZ)
192
f895b252 193#if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
a246b010 194# undef RPC_DEBUG_DATA
9903cd1c 195# define RPCDBG_FACILITY RPCDBG_TRANS
a246b010
CL		 196#endif
197
a246b010 198#ifdef RPC_DEBUG_DATA
9903cd1c 199static void xs_pktdump(char *msg, u32 *packet, unsigned int count)
a246b010 200{
9903cd1c
CL		 201 u8 *buf = (u8 *) packet;
202 int j;
a246b010 203
46121cf7 204 dprintk("RPC: %s\n", msg);
a246b010
CL		 205 for (j = 0; j < count && j < 128; j += 4) {
206 if (!(j & 31)) {
207 if (j)
208 dprintk("\n");
209 dprintk("0x%04x ", j);
210 }
211 dprintk("%02x%02x%02x%02x ",
212 buf[j], buf[j+1], buf[j+2], buf[j+3]);
213 }
214 dprintk("\n");
215}
216#else
9903cd1c 217static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count)
a246b010
CL		 218{
219 /* NOP */
220}
221#endif
222
2118071d
TM		 223static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
224{
225 return (struct rpc_xprt *) sk->sk_user_data;
226}
227
95392c59
CL		 228static inline struct sockaddr *xs_addr(struct rpc_xprt *xprt)
229{
230 return (struct sockaddr *) &xprt->addr;
231}
232
176e21ee
CL		 233static inline struct sockaddr_un *xs_addr_un(struct rpc_xprt *xprt)
234{
235 return (struct sockaddr_un *) &xprt->addr;
236}
237
95392c59 238static inline struct sockaddr_in *xs_addr_in(struct rpc_xprt *xprt)
edb267a6 239{
95392c59
CL		 240 return (struct sockaddr_in *) &xprt->addr;
241}
242
243static inline struct sockaddr_in6 *xs_addr_in6(struct rpc_xprt *xprt)
244{
245 return (struct sockaddr_in6 *) &xprt->addr;
246}
247
c877b849 248static void xs_format_common_peer_addresses(struct rpc_xprt *xprt)
edb267a6 249{
c877b849 250 struct sockaddr *sap = xs_addr(xprt);
9dc3b095
CL		 251 struct sockaddr_in6 *sin6;
252 struct sockaddr_in *sin;
176e21ee 253 struct sockaddr_un *sun;
c877b849 254 char buf[128];
edb267a6 255
9dc3b095 256 switch (sap->sa_family) {
176e21ee
CL		 257 case AF_LOCAL:
258 sun = xs_addr_un(xprt);
259 strlcpy(buf, sun->sun_path, sizeof(buf));
260 xprt->address_strings[RPC_DISPLAY_ADDR] =
261 kstrdup(buf, GFP_KERNEL);
262 break;
9dc3b095 263 case AF_INET:
176e21ee
CL		 264 (void)rpc_ntop(sap, buf, sizeof(buf));
265 xprt->address_strings[RPC_DISPLAY_ADDR] =
266 kstrdup(buf, GFP_KERNEL);
9dc3b095 267 sin = xs_addr_in(xprt);
fc0b5791 268 snprintf(buf, sizeof(buf), "%08x", ntohl(sin->sin_addr.s_addr));
9dc3b095
CL		 269 break;
270 case AF_INET6:
176e21ee
CL		 271 (void)rpc_ntop(sap, buf, sizeof(buf));
272 xprt->address_strings[RPC_DISPLAY_ADDR] =
273 kstrdup(buf, GFP_KERNEL);
9dc3b095 274 sin6 = xs_addr_in6(xprt);
fc0b5791 275 snprintf(buf, sizeof(buf), "%pi6", &sin6->sin6_addr);
9dc3b095
CL		 276 break;
277 default:
278 BUG();
756805e7 279 }
176e21ee 280
9dc3b095 281 xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = kstrdup(buf, GFP_KERNEL);
edb267a6
CL		 282}
283
9dc3b095 284static void xs_format_common_peer_ports(struct rpc_xprt *xprt)
4b6473fb 285{
9dc3b095
CL		 286 struct sockaddr *sap = xs_addr(xprt);
287 char buf[128];
4b6473fb 288
81160e66 289 snprintf(buf, sizeof(buf), "%u", rpc_get_port(sap));
c877b849 290 xprt->address_strings[RPC_DISPLAY_PORT] = kstrdup(buf, GFP_KERNEL);
4b6473fb 291
81160e66 292 snprintf(buf, sizeof(buf), "%4hx", rpc_get_port(sap));
c877b849
CL		 293 xprt->address_strings[RPC_DISPLAY_HEX_PORT] = kstrdup(buf, GFP_KERNEL);
294}
4b6473fb 295
9dc3b095
CL		 296static void xs_format_peer_addresses(struct rpc_xprt *xprt,
297 const char *protocol,
298 const char *netid)
c877b849 299{
b454ae90 300 xprt->address_strings[RPC_DISPLAY_PROTO] = protocol;
b454ae90 301 xprt->address_strings[RPC_DISPLAY_NETID] = netid;
c877b849 302 xs_format_common_peer_addresses(xprt);
9dc3b095 303 xs_format_common_peer_ports(xprt);
edb267a6 304}
4b6473fb 305
9dc3b095 306static void xs_update_peer_port(struct rpc_xprt *xprt)
4b6473fb 307{
9dc3b095
CL		 308 kfree(xprt->address_strings[RPC_DISPLAY_HEX_PORT]);
309 kfree(xprt->address_strings[RPC_DISPLAY_PORT]);
4417c8c4 310
9dc3b095 311 xs_format_common_peer_ports(xprt);
edb267a6
CL		 312}
313
314static void xs_free_peer_addresses(struct rpc_xprt *xprt)
315{
33e01dc7
CL		 316 unsigned int i;
317
318 for (i = 0; i < RPC_DISPLAY_MAX; i++)
319 switch (i) {
320 case RPC_DISPLAY_PROTO:
321 case RPC_DISPLAY_NETID:
322 continue;
323 default:
324 kfree(xprt->address_strings[i]);
325 }
edb267a6
CL		 326}
327
277e4ab7
TM		 328static size_t
329xs_alloc_sparse_pages(struct xdr_buf *buf, size_t want, gfp_t gfp)
330{
331 size_t i,n;
332
333 if (!(buf->flags & XDRBUF_SPARSE_PAGES))
334 return want;
335 if (want > buf->page_len)
336 want = buf->page_len;
337 n = (buf->page_base + want + PAGE_SIZE - 1) >> PAGE_SHIFT;
338 for (i = 0; i < n; i++) {
339 if (buf->pages[i])
340 continue;
341 buf->bvec[i].bv_page = buf->pages[i] = alloc_page(gfp);
342 if (!buf->pages[i]) {
343 buf->page_len = (i * PAGE_SIZE) - buf->page_base;
344 return buf->page_len;
345 }
346 }
347 return want;
348}
349
350static ssize_t
351xs_sock_recvmsg(struct socket *sock, struct msghdr *msg, int flags, size_t seek)
352{
353 ssize_t ret;
354 if (seek != 0)
355 iov_iter_advance(&msg->msg_iter, seek);
356 ret = sock_recvmsg(sock, msg, flags);
357 return ret > 0 ? ret + seek : ret;
358}
359
360static ssize_t
361xs_read_kvec(struct socket *sock, struct msghdr *msg, int flags,
362 struct kvec *kvec, size_t count, size_t seek)
363{
0e9b4a82 364 iov_iter_kvec(&msg->msg_iter, READ, kvec, 1, count);
277e4ab7
TM		 365 return xs_sock_recvmsg(sock, msg, flags, seek);
366}
367
368static ssize_t
369xs_read_bvec(struct socket *sock, struct msghdr *msg, int flags,
370 struct bio_vec *bvec, unsigned long nr, size_t count,
371 size_t seek)
372{
0e9b4a82 373 iov_iter_bvec(&msg->msg_iter, READ, bvec, nr, count);
277e4ab7
TM		 374 return xs_sock_recvmsg(sock, msg, flags, seek);
375}
376
377static ssize_t
378xs_read_discard(struct socket *sock, struct msghdr *msg, int flags,
379 size_t count)
380{
381 struct kvec kvec = { 0 };
382 return xs_read_kvec(sock, msg, flags | MSG_TRUNC, &kvec, count, 0);
383}
384
385static ssize_t
386xs_read_xdr_buf(struct socket *sock, struct msghdr *msg, int flags,
387 struct xdr_buf *buf, size_t count, size_t seek, size_t *read)
388{
389 size_t want, seek_init = seek, offset = 0;
390 ssize_t ret;
391
392 if (seek < buf->head[0].iov_len) {
393 want = min_t(size_t, count, buf->head[0].iov_len);
394 ret = xs_read_kvec(sock, msg, flags, &buf->head[0], want, seek);
395 if (ret <= 0)
396 goto sock_err;
397 offset += ret;
398 if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC))
399 goto out;
400 if (ret != want)
401 goto eagain;
402 seek = 0;
403 } else {
404 seek -= buf->head[0].iov_len;
405 offset += buf->head[0].iov_len;
406 }
407 if (seek < buf->page_len) {
408 want = xs_alloc_sparse_pages(buf,
409 min_t(size_t, count - offset, buf->page_len),
410 GFP_NOWAIT);
411 ret = xs_read_bvec(sock, msg, flags, buf->bvec,
412 xdr_buf_pagecount(buf),
413 want + buf->page_base,
414 seek + buf->page_base);
415 if (ret <= 0)
416 goto sock_err;
417 offset += ret - buf->page_base;
418 if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC))
419 goto out;
420 if (ret != want)
421 goto eagain;
422 seek = 0;
423 } else {
424 seek -= buf->page_len;
425 offset += buf->page_len;
426 }
427 if (seek < buf->tail[0].iov_len) {
428 want = min_t(size_t, count - offset, buf->tail[0].iov_len);
429 ret = xs_read_kvec(sock, msg, flags, &buf->tail[0], want, seek);
430 if (ret <= 0)
431 goto sock_err;
432 offset += ret;
433 if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC))
434 goto out;
435 if (ret != want)
436 goto eagain;
437 } else
438 offset += buf->tail[0].iov_len;
439 ret = -EMSGSIZE;
440 msg->msg_flags |= MSG_TRUNC;
441out:
442 *read = offset - seek_init;
443 return ret;
444eagain:
445 ret = -EAGAIN;
446 goto out;
447sock_err:
448 offset += seek;
449 goto out;
450}
451
452static void
453xs_read_header(struct sock_xprt *transport, struct xdr_buf *buf)
454{
455 if (!transport->recv.copied) {
456 if (buf->head[0].iov_len >= transport->recv.offset)
457 memcpy(buf->head[0].iov_base,
458 &transport->recv.xid,
459 transport->recv.offset);
460 transport->recv.copied = transport->recv.offset;
461 }
462}
463
464static bool
465xs_read_stream_request_done(struct sock_xprt *transport)
466{
467 return transport->recv.fraghdr & cpu_to_be32(RPC_LAST_STREAM_FRAGMENT);
468}
469
470static ssize_t
471xs_read_stream_request(struct sock_xprt *transport, struct msghdr *msg,
472 int flags, struct rpc_rqst *req)
473{
474 struct xdr_buf *buf = &req->rq_private_buf;
475 size_t want, read;
476 ssize_t ret;
477
478 xs_read_header(transport, buf);
479
480 want = transport->recv.len - transport->recv.offset;
481 ret = xs_read_xdr_buf(transport->sock, msg, flags, buf,
482 transport->recv.copied + want, transport->recv.copied,
483 &read);
484 transport->recv.offset += read;
485 transport->recv.copied += read;
486 if (transport->recv.offset == transport->recv.len) {
487 if (xs_read_stream_request_done(transport))
488 msg->msg_flags |= MSG_EOR;
489 return transport->recv.copied;
490 }
491
492 switch (ret) {
493 case -EMSGSIZE:
494 return transport->recv.copied;
495 case 0:
496 return -ESHUTDOWN;
497 default:
498 if (ret < 0)
499 return ret;
500 }
501 return -EAGAIN;
502}
503
504static size_t
505xs_read_stream_headersize(bool isfrag)
506{
507 if (isfrag)
508 return sizeof(__be32);
509 return 3 * sizeof(__be32);
510}
511
512static ssize_t
513xs_read_stream_header(struct sock_xprt *transport, struct msghdr *msg,
514 int flags, size_t want, size_t seek)
515{
516 struct kvec kvec = {
517 .iov_base = &transport->recv.fraghdr,
518 .iov_len = want,
519 };
520 return xs_read_kvec(transport->sock, msg, flags, &kvec, want, seek);
521}
522
523#if defined(CONFIG_SUNRPC_BACKCHANNEL)
524static ssize_t
525xs_read_stream_call(struct sock_xprt *transport, struct msghdr *msg, int flags)
526{
527 struct rpc_xprt *xprt = &transport->xprt;
528 struct rpc_rqst *req;
529 ssize_t ret;
530
531 /* Look up and lock the request corresponding to the given XID */
532 req = xprt_lookup_bc_request(xprt, transport->recv.xid);
533 if (!req) {
534 printk(KERN_WARNING "Callback slot table overflowed\n");
535 return -ESHUTDOWN;
536 }
537
538 ret = xs_read_stream_request(transport, msg, flags, req);
539 if (msg->msg_flags & (MSG_EOR|MSG_TRUNC))
540 xprt_complete_bc_request(req, ret);
541
542 return ret;
543}
544#else /* CONFIG_SUNRPC_BACKCHANNEL */
545static ssize_t
546xs_read_stream_call(struct sock_xprt *transport, struct msghdr *msg, int flags)
547{
548 return -ESHUTDOWN;
549}
550#endif /* CONFIG_SUNRPC_BACKCHANNEL */
551
552static ssize_t
553xs_read_stream_reply(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 = 0;
558
559 /* Look up and lock the request corresponding to the given XID */
560 spin_lock(&xprt->queue_lock);
561 req = xprt_lookup_rqst(xprt, transport->recv.xid);
562 if (!req) {
563 msg->msg_flags |= MSG_TRUNC;
564 goto out;
565 }
566 xprt_pin_rqst(req);
567 spin_unlock(&xprt->queue_lock);
568
569 ret = xs_read_stream_request(transport, msg, flags, req);
570
571 spin_lock(&xprt->queue_lock);
572 if (msg->msg_flags & (MSG_EOR|MSG_TRUNC))
573 xprt_complete_rqst(req->rq_task, ret);
574 xprt_unpin_rqst(req);
575out:
576 spin_unlock(&xprt->queue_lock);
577 return ret;
578}
579
580static ssize_t
581xs_read_stream(struct sock_xprt *transport, int flags)
582{
583 struct msghdr msg = { 0 };
584 size_t want, read = 0;
585 ssize_t ret = 0;
586
587 if (transport->recv.len == 0) {
588 want = xs_read_stream_headersize(transport->recv.copied != 0);
589 ret = xs_read_stream_header(transport, &msg, flags, want,
590 transport->recv.offset);
591 if (ret <= 0)
592 goto out_err;
593 transport->recv.offset = ret;
594 if (ret != want) {
595 ret = -EAGAIN;
596 goto out_err;
597 }
598 transport->recv.len = be32_to_cpu(transport->recv.fraghdr) &
599 RPC_FRAGMENT_SIZE_MASK;
600 transport->recv.offset -= sizeof(transport->recv.fraghdr);
601 read = ret;
602 }
603
604 switch (be32_to_cpu(transport->recv.calldir)) {
605 case RPC_CALL:
606 ret = xs_read_stream_call(transport, &msg, flags);
607 break;
608 case RPC_REPLY:
609 ret = xs_read_stream_reply(transport, &msg, flags);
610 }
611 if (msg.msg_flags & MSG_TRUNC) {
612 transport->recv.calldir = cpu_to_be32(-1);
613 transport->recv.copied = -1;
614 }
615 if (ret < 0)
616 goto out_err;
617 read += ret;
618 if (transport->recv.offset < transport->recv.len) {
619 ret = xs_read_discard(transport->sock, &msg, flags,
620 transport->recv.len - transport->recv.offset);
621 if (ret <= 0)
622 goto out_err;
623 transport->recv.offset += ret;
624 read += ret;
625 if (transport->recv.offset != transport->recv.len)
626 return -EAGAIN;
627 }
628 if (xs_read_stream_request_done(transport)) {
c50b8ee0 629 trace_xs_stream_read_request(transport);
277e4ab7
TM		 630 transport->recv.copied = 0;
631 }
632 transport->recv.offset = 0;
633 transport->recv.len = 0;
634 return read;
635out_err:
636 switch (ret) {
637 case 0:
638 case -ESHUTDOWN:
639 xprt_force_disconnect(&transport->xprt);
640 return -ESHUTDOWN;
641 }
642 return ret;
643}
644
c50b8ee0
TM		 645static void xs_stream_data_receive(struct sock_xprt *transport)
646{
647 size_t read = 0;
648 ssize_t ret = 0;
649
650 mutex_lock(&transport->recv_mutex);
651 if (transport->sock == NULL)
652 goto out;
653 clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state);
654 for (;;) {
655 ret = xs_read_stream(transport, MSG_DONTWAIT);
656 if (ret <= 0)
657 break;
658 read += ret;
659 cond_resched();
660 }
661out:
662 mutex_unlock(&transport->recv_mutex);
663 trace_xs_stream_read_data(&transport->xprt, ret, read);
664}
665
666static void xs_stream_data_receive_workfn(struct work_struct *work)
667{
668 struct sock_xprt *transport =
669 container_of(work, struct sock_xprt, recv_worker);
670 xs_stream_data_receive(transport);
671}
672
550aebfe
TM		 673static void
674xs_stream_reset_connect(struct sock_xprt *transport)
675{
676 transport->recv.offset = 0;
677 transport->recv.len = 0;
678 transport->recv.copied = 0;
679 transport->xmit.offset = 0;
680 transport->xprt.stat.connect_count++;
681 transport->xprt.stat.connect_start = jiffies;
682}
683
b4b5cc85
CL		 684#define XS_SENDMSG_FLAGS (MSG_DONTWAIT | MSG_NOSIGNAL)
685
24c5684b 686static int xs_send_kvec(struct socket *sock, struct sockaddr *addr, int addrlen, struct kvec *vec, unsigned int base, int more)
b4b5cc85 687{
b4b5cc85
CL		 688 struct msghdr msg = {
689 .msg_name = addr,
690 .msg_namelen = addrlen,
24c5684b
TM		 691 .msg_flags = XS_SENDMSG_FLAGS | (more ? MSG_MORE : 0),
692 };
693 struct kvec iov = {
694 .iov_base = vec->iov_base + base,
695 .iov_len = vec->iov_len - base,
b4b5cc85
CL		 696 };
697
24c5684b 698 if (iov.iov_len != 0)
b4b5cc85
CL		 699 return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
700 return kernel_sendmsg(sock, &msg, NULL, 0, 0);
701}
702
f279cd00 703static int xs_send_pagedata(struct socket *sock, struct xdr_buf *xdr, unsigned int base, int more, bool zerocopy, int *sent_p)
b4b5cc85 704{
a6b31d18
TM		 705 ssize_t (*do_sendpage)(struct socket *sock, struct page *page,
706 int offset, size_t size, int flags);
24c5684b
TM		 707 struct page **ppage;
708 unsigned int remainder;
f279cd00 709 int err;
24c5684b
TM		 710
711 remainder = xdr->page_len - base;
712 base += xdr->page_base;
713 ppage = xdr->pages + (base >> PAGE_SHIFT);
714 base &= ~PAGE_MASK;
a6b31d18
TM		 715 do_sendpage = sock->ops->sendpage;
716 if (!zerocopy)
717 do_sendpage = sock_no_sendpage;
24c5684b
TM		 718 for(;;) {
719 unsigned int len = min_t(unsigned int, PAGE_SIZE - base, remainder);
720 int flags = XS_SENDMSG_FLAGS;
b4b5cc85 721
24c5684b 722 remainder -= len;
31303d6c 723 if (more)
24c5684b 724 flags |= MSG_MORE;
31303d6c
TM		 725 if (remainder != 0)
726 flags |= MSG_SENDPAGE_NOTLAST | MSG_MORE;
a6b31d18 727 err = do_sendpage(sock, *ppage, base, len, flags);
24c5684b
TM		 728 if (remainder == 0 || err != len)
729 break;
f279cd00 730 *sent_p += err;
24c5684b
TM		 731 ppage++;
732 base = 0;
733 }
f279cd00
JB		 734 if (err > 0) {
735 *sent_p += err;
736 err = 0;
737 }
738 return err;
b4b5cc85
CL		 739}
740
9903cd1c
CL		 741/**
742 * xs_sendpages - write pages directly to a socket
743 * @sock: socket to send on
744 * @addr: UDP only -- address of destination
745 * @addrlen: UDP only -- length of destination address
746 * @xdr: buffer containing this request
747 * @base: starting position in the buffer
a6b31d18 748 * @zerocopy: true if it is safe to use sendpage()
f279cd00 749 * @sent_p: return the total number of bytes successfully queued for sending
9903cd1c 750 *
a246b010 751 */
f279cd00 752static int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base, bool zerocopy, int *sent_p)
a246b010 753{
24c5684b 754 unsigned int remainder = xdr->len - base;
f279cd00
JB		 755 int err = 0;
756 int sent = 0;
a246b010 757
262965f5 758 if (unlikely(!sock))
fba91afb 759 return -ENOTSOCK;
262965f5 760
24c5684b
TM		 761 if (base != 0) {
762 addr = NULL;
763 addrlen = 0;
764 }
262965f5 765
24c5684b
TM		 766 if (base < xdr->head[0].iov_len || addr != NULL) {
767 unsigned int len = xdr->head[0].iov_len - base;
768 remainder -= len;
769 err = xs_send_kvec(sock, addr, addrlen, &xdr->head[0], base, remainder != 0);
770 if (remainder == 0 || err != len)
a246b010 771 goto out;
f279cd00 772 *sent_p += err;
a246b010
CL		 773 base = 0;
774 } else
24c5684b 775 base -= xdr->head[0].iov_len;
a246b010 776
24c5684b
TM		 777 if (base < xdr->page_len) {
778 unsigned int len = xdr->page_len - base;
779 remainder -= len;
f279cd00
JB		 780 err = xs_send_pagedata(sock, xdr, base, remainder != 0, zerocopy, &sent);
781 *sent_p += sent;
782 if (remainder == 0 || sent != len)
a246b010
CL		 783 goto out;
784 base = 0;
24c5684b
TM		 785 } else
786 base -= xdr->page_len;
787
788 if (base >= xdr->tail[0].iov_len)
f279cd00 789 return 0;
24c5684b 790 err = xs_send_kvec(sock, NULL, 0, &xdr->tail[0], base, 0);
a246b010 791out:
f279cd00
JB		 792 if (err > 0) {
793 *sent_p += err;
794 err = 0;
795 }
796 return err;
a246b010
CL		 797}
798
9903cd1c 799/**
c544577d 800 * xs_nospace - handle transmit was incomplete
50f484e2 801 * @req: pointer to RPC request
9903cd1c 802 *
a246b010 803 */
c544577d 804static int xs_nospace(struct rpc_rqst *req)
a246b010 805{
262965f5 806 struct rpc_xprt *xprt = req->rq_xprt;
ee0ac0c2 807 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
06ea0bfe 808 struct sock *sk = transport->inet;
24ca9a84 809 int ret = -EAGAIN;
a246b010 810
46121cf7 811 dprintk("RPC: %5u xmit incomplete (%u left of %u)\n",
c544577d
TM		 812 req->rq_task->tk_pid,
813 req->rq_slen - transport->xmit.offset,
262965f5
CL		 814 req->rq_slen);
815
b6ddf64f
TM		 816 /* Protect against races with write_space */
817 spin_lock_bh(&xprt->transport_lock);
818
819 /* Don't race with disconnect */
820 if (xprt_connected(xprt)) {
13331a55
TM		 821 /* wait for more buffer space */
822 sk->sk_write_pending++;
c544577d 823 xprt_wait_for_buffer_space(xprt);
13331a55 824 } else
5e3771ce 825 ret = -ENOTCONN;
262965f5 826
b6ddf64f 827 spin_unlock_bh(&xprt->transport_lock);
06ea0bfe
TM		 828
829 /* Race breaker in case memory is freed before above code is called */
d48f9ce7
DV		 830 if (ret == -EAGAIN) {
831 struct socket_wq *wq;
832
833 rcu_read_lock();
834 wq = rcu_dereference(sk->sk_wq);
835 set_bit(SOCKWQ_ASYNC_NOSPACE, &wq->flags);
836 rcu_read_unlock();
837
838 sk->sk_write_space(sk);
839 }
5e3771ce 840 return ret;
262965f5
CL		 841}
842
277e4ab7
TM		 843static void
844xs_stream_prepare_request(struct rpc_rqst *req)
845{
846 req->rq_task->tk_status = xdr_alloc_bvec(&req->rq_rcv_buf, GFP_NOIO);
847}
848
4cd34e7c
TM		 849/*
850 * Determine if the previous message in the stream was aborted before it
851 * could complete transmission.
852 */
853static bool
854xs_send_request_was_aborted(struct sock_xprt *transport, struct rpc_rqst *req)
855{
856 return transport->xmit.offset != 0 && req->rq_bytes_sent == 0;
857}
858
61677eee
CL		 859/*
860 * Construct a stream transport record marker in @buf.
861 */
862static inline void xs_encode_stream_record_marker(struct xdr_buf *buf)
863{
864 u32 reclen = buf->len - sizeof(rpc_fraghdr);
865 rpc_fraghdr *base = buf->head[0].iov_base;
866 *base = cpu_to_be32(RPC_LAST_STREAM_FRAGMENT | reclen);
867}
868
176e21ee
CL		 869/**
870 * xs_local_send_request - write an RPC request to an AF_LOCAL socket
50f484e2 871 * @req: pointer to RPC request
176e21ee
CL		 872 *
873 * Return values:
874 * 0: The request has been sent
875 * EAGAIN: The socket was blocked, please call again later to
876 * complete the request
877 * ENOTCONN: Caller needs to invoke connect logic then call again
878 * other: Some other error occured, the request was not sent
879 */
adfa7144 880static int xs_local_send_request(struct rpc_rqst *req)
176e21ee 881{
176e21ee
CL		 882 struct rpc_xprt *xprt = req->rq_xprt;
883 struct sock_xprt *transport =
884 container_of(xprt, struct sock_xprt, xprt);
885 struct xdr_buf *xdr = &req->rq_snd_buf;
886 int status;
f279cd00 887 int sent = 0;
176e21ee 888
4cd34e7c
TM		 889 /* Close the stream if the previous transmission was incomplete */
890 if (xs_send_request_was_aborted(transport, req)) {
891 xs_close(xprt);
892 return -ENOTCONN;
893 }
894
176e21ee
CL		 895 xs_encode_stream_record_marker(&req->rq_snd_buf);
896
897 xs_pktdump("packet data:",
898 req->rq_svec->iov_base, req->rq_svec->iov_len);
899
78215759 900 req->rq_xtime = ktime_get();
6c7a64e5
TM		 901 status = xs_sendpages(transport->sock, NULL, 0, xdr,
902 transport->xmit.offset,
f279cd00 903 true, &sent);
176e21ee 904 dprintk("RPC: %s(%u) = %d\n",
6c7a64e5 905 __func__, xdr->len - transport->xmit.offset, status);
743c69e7
N		 906
907 if (status == -EAGAIN && sock_writeable(transport->inet))
908 status = -ENOBUFS;
909
f279cd00 910 if (likely(sent > 0) || status == 0) {
6c7a64e5
TM		 911 transport->xmit.offset += sent;
912 req->rq_bytes_sent = transport->xmit.offset;
176e21ee 913 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
6c7a64e5 914 req->rq_xmit_bytes_sent += transport->xmit.offset;
176e21ee 915 req->rq_bytes_sent = 0;
6c7a64e5 916 transport->xmit.offset = 0;
176e21ee
CL		 917 return 0;
918 }
919 status = -EAGAIN;
920 }
921
922 switch (status) {
3601c4a9 923 case -ENOBUFS:
b5872f0c 924 break;
176e21ee 925 case -EAGAIN:
c544577d 926 status = xs_nospace(req);
176e21ee
CL		 927 break;
928 default:
929 dprintk("RPC: sendmsg returned unrecognized error %d\n",
930 -status);
e9d47639 931 /* fall through */
176e21ee
CL		 932 case -EPIPE:
933 xs_close(xprt);
934 status = -ENOTCONN;
935 }
936
937 return status;
938}
939
262965f5
CL		 940/**
941 * xs_udp_send_request - write an RPC request to a UDP socket
50f484e2 942 * @req: pointer to RPC request
262965f5
CL		 943 *
944 * Return values:
945 * 0: The request has been sent
946 * EAGAIN: The socket was blocked, please call again later to
947 * complete the request
948 * ENOTCONN: Caller needs to invoke connect logic then call again
25985edc 949 * other: Some other error occurred, the request was not sent
262965f5 950 */
adfa7144 951static int xs_udp_send_request(struct rpc_rqst *req)
262965f5 952{
262965f5 953 struct rpc_xprt *xprt = req->rq_xprt;
ee0ac0c2 954 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
262965f5 955 struct xdr_buf *xdr = &req->rq_snd_buf;
f279cd00 956 int sent = 0;
262965f5 957 int status;
a246b010 958
9903cd1c 959 xs_pktdump("packet data:",
a246b010
CL		 960 req->rq_svec->iov_base,
961 req->rq_svec->iov_len);
962
01d37c42
TM		 963 if (!xprt_bound(xprt))
964 return -ENOTCONN;
75891f50
TM		 965
966 if (!xprt_request_get_cong(xprt, req))
967 return -EBADSLT;
968
78215759 969 req->rq_xtime = ktime_get();
f279cd00 970 status = xs_sendpages(transport->sock, xs_addr(xprt), xprt->addrlen,
6c7a64e5 971 xdr, 0, true, &sent);
a246b010 972
46121cf7 973 dprintk("RPC: xs_udp_send_request(%u) = %d\n",
6c7a64e5 974 xdr->len, status);
a246b010 975
3dedbb5c
JB		 976 /* firewall is blocking us, don't return -EAGAIN or we end up looping */
977 if (status == -EPERM)
978 goto process_status;
979
743c69e7
N		 980 if (status == -EAGAIN && sock_writeable(transport->inet))
981 status = -ENOBUFS;
982
f279cd00
JB		 983 if (sent > 0 || status == 0) {
984 req->rq_xmit_bytes_sent += sent;
985 if (sent >= req->rq_slen)
2199700f
TM		 986 return 0;
987 /* Still some bytes left; set up for a retry later. */
262965f5 988 status = -EAGAIN;
2199700f 989 }
a246b010 990
3dedbb5c 991process_status:
262965f5 992 switch (status) {
fba91afb
TM		 993 case -ENOTSOCK:
994 status = -ENOTCONN;
995 /* Should we call xs_close() here? */
996 break;
b6ddf64f 997 case -EAGAIN:
c544577d 998 status = xs_nospace(req);
b6ddf64f 999 break;
262965f5 1000 case -ENETUNREACH:
3601c4a9 1001 case -ENOBUFS:
262965f5 1002 case -EPIPE:
a246b010 1003 case -ECONNREFUSED:
3dedbb5c 1004 case -EPERM:
a246b010 1005 /* When the server has died, an ICMP port unreachable message
9903cd1c 1006 * prompts ECONNREFUSED. */
13331a55
TM		 1007 break;
1008 default:
1009 dprintk("RPC: sendmsg returned unrecognized error %d\n",
1010 -status);
a246b010 1011 }
5fe46e9d 1012
262965f5 1013 return status;
a246b010
CL		 1014}
1015
9903cd1c 1016/**
262965f5 1017 * xs_tcp_send_request - write an RPC request to a TCP socket
50f484e2 1018 * @req: pointer to RPC request
9903cd1c
CL		 1019 *
1020 * Return values:
262965f5
CL		 1021 * 0: The request has been sent
1022 * EAGAIN: The socket was blocked, please call again later to
1023 * complete the request
1024 * ENOTCONN: Caller needs to invoke connect logic then call again
25985edc 1025 * other: Some other error occurred, the request was not sent
9903cd1c
CL		 1026 *
1027 * XXX: In the case of soft timeouts, should we eventually give up
262965f5 1028 * if sendmsg is not able to make progress?
9903cd1c 1029 */
adfa7144 1030static int xs_tcp_send_request(struct rpc_rqst *req)
a246b010 1031{
a246b010 1032 struct rpc_xprt *xprt = req->rq_xprt;
ee0ac0c2 1033 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
262965f5 1034 struct xdr_buf *xdr = &req->rq_snd_buf;
a6b31d18 1035 bool zerocopy = true;
9ffadfbc 1036 bool vm_wait = false;
b595bb15 1037 int status;
f279cd00 1038 int sent;
a246b010 1039
4cd34e7c
TM		 1040 /* Close the stream if the previous transmission was incomplete */
1041 if (xs_send_request_was_aborted(transport, req)) {
1042 if (transport->sock != NULL)
1043 kernel_sock_shutdown(transport->sock, SHUT_RDWR);
1044 return -ENOTCONN;
1045 }
1046
61677eee 1047 xs_encode_stream_record_marker(&req->rq_snd_buf);
a246b010 1048
262965f5
CL		 1049 xs_pktdump("packet data:",
1050 req->rq_svec->iov_base,
1051 req->rq_svec->iov_len);
a6b31d18
TM		 1052 /* Don't use zero copy if this is a resend. If the RPC call
1053 * completes while the socket holds a reference to the pages,
1054 * then we may end up resending corrupted data.
1055 */
50f484e2 1056 if (req->rq_task->tk_flags & RPC_TASK_SENT)
a6b31d18 1057 zerocopy = false;
a246b010 1058
7196dbb0
TM		 1059 if (test_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state))
1060 xs_tcp_set_socket_timeouts(xprt, transport->sock);
1061
a246b010
CL		 1062 /* Continue transmitting the packet/record. We must be careful
1063 * to cope with writespace callbacks arriving _after_ we have
262965f5 1064 * called sendmsg(). */
78215759 1065 req->rq_xtime = ktime_get();
a246b010 1066 while (1) {
f279cd00
JB		 1067 sent = 0;
1068 status = xs_sendpages(transport->sock, NULL, 0, xdr,
6c7a64e5
TM		 1069 transport->xmit.offset,
1070 zerocopy, &sent);
a246b010 1071
46121cf7 1072 dprintk("RPC: xs_tcp_send_request(%u) = %d\n",
6c7a64e5 1073 xdr->len - transport->xmit.offset, status);
a246b010 1074
262965f5
CL		 1075 /* If we've sent the entire packet, immediately
1076 * reset the count of bytes sent. */
6c7a64e5
TM		 1077 transport->xmit.offset += sent;
1078 req->rq_bytes_sent = transport->xmit.offset;
262965f5 1079 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
6c7a64e5 1080 req->rq_xmit_bytes_sent += transport->xmit.offset;
262965f5 1081 req->rq_bytes_sent = 0;
6c7a64e5 1082 transport->xmit.offset = 0;
262965f5
CL		 1083 return 0;
1084 }
a246b010 1085
9ffadfbc
TM		 1086 WARN_ON_ONCE(sent == 0 && status == 0);
1087
1088 if (status == -EAGAIN ) {
1089 /*
1090 * Return EAGAIN if we're sure we're hitting the
1091 * socket send buffer limits.
1092 */
1093 if (test_bit(SOCK_NOSPACE, &transport->sock->flags))
1094 break;
1095 /*
1096 * Did we hit a memory allocation failure?
1097 */
1098 if (sent == 0) {
1099 status = -ENOBUFS;
1100 if (vm_wait)
1101 break;
1102 /* Retry, knowing now that we're below the
1103 * socket send buffer limit
1104 */
1105 vm_wait = true;
1106 }
1107 continue;
1108 }
f580dd04
TM		 1109 if (status < 0)
1110 break;
9ffadfbc 1111 vm_wait = false;
a246b010
CL		 1112 }
1113
262965f5 1114 switch (status) {
fba91afb
TM		 1115 case -ENOTSOCK:
1116 status = -ENOTCONN;
1117 /* Should we call xs_close() here? */
1118 break;
262965f5 1119 case -EAGAIN:
c544577d 1120 status = xs_nospace(req);
262965f5 1121 break;
262965f5 1122 case -ECONNRESET:
2a9e1cfa 1123 case -ECONNREFUSED:
262965f5 1124 case -ENOTCONN:
3913c78c 1125 case -EADDRINUSE:
b5872f0c 1126 case -ENOBUFS:
b9d2bb2e 1127 case -EPIPE:
13331a55
TM		 1128 break;
1129 default:
1130 dprintk("RPC: sendmsg returned unrecognized error %d\n",
1131 -status);
a246b010 1132 }
5fe46e9d 1133
a246b010
CL		 1134 return status;
1135}
1136
2a9e1cfa
TM		 1137static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk)
1138{
1139 transport->old_data_ready = sk->sk_data_ready;
1140 transport->old_state_change = sk->sk_state_change;
1141 transport->old_write_space = sk->sk_write_space;
2118071d 1142 transport->old_error_report = sk->sk_error_report;
2a9e1cfa
TM		 1143}
1144
1145static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk)
1146{
1147 sk->sk_data_ready = transport->old_data_ready;
1148 sk->sk_state_change = transport->old_state_change;
1149 sk->sk_write_space = transport->old_write_space;
2118071d
TM		 1150 sk->sk_error_report = transport->old_error_report;
1151}
1152
42d42a5b
TM		 1153static void xs_sock_reset_state_flags(struct rpc_xprt *xprt)
1154{
1155 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1156
1157 clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state);
1158}
1159
b70ae915
TM		 1160static void xs_sock_reset_connection_flags(struct rpc_xprt *xprt)
1161{
1162 smp_mb__before_atomic();
1163 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1164 clear_bit(XPRT_CLOSING, &xprt->state);
42d42a5b 1165 xs_sock_reset_state_flags(xprt);
b70ae915
TM		 1166 smp_mb__after_atomic();
1167}
1168
2118071d
TM		 1169/**
1170 * xs_error_report - callback to handle TCP socket state errors
1171 * @sk: socket
1172 *
1173 * Note: we don't call sock_error() since there may be a rpc_task
1174 * using the socket, and so we don't want to clear sk->sk_err.
1175 */
1176static void xs_error_report(struct sock *sk)
1177{
1178 struct rpc_xprt *xprt;
1179 int err;
1180
1181 read_lock_bh(&sk->sk_callback_lock);
1182 if (!(xprt = xprt_from_sock(sk)))
1183 goto out;
1184
1185 err = -sk->sk_err;
1186 if (err == 0)
1187 goto out;
1188 dprintk("RPC: xs_error_report client %p, error=%d...\n",
1189 xprt, -err);
e8353c76 1190 trace_rpc_socket_error(xprt, sk->sk_socket, err);
2118071d
TM		 1191 xprt_wake_pending_tasks(xprt, err);
1192 out:
1193 read_unlock_bh(&sk->sk_callback_lock);
2a9e1cfa
TM		 1194}
1195
fe315e76 1196static void xs_reset_transport(struct sock_xprt *transport)
a246b010 1197{
ee0ac0c2
CL		 1198 struct socket *sock = transport->sock;
1199 struct sock *sk = transport->inet;
6cc7e908 1200 struct rpc_xprt *xprt = &transport->xprt;
a246b010 1201
fe315e76
CL		 1202 if (sk == NULL)
1203 return;
9903cd1c 1204
264d1df3
JL		 1205 if (atomic_read(&transport->xprt.swapper))
1206 sk_clear_memalloc(sk);
1207
09939204
TM		 1208 kernel_sock_shutdown(sock, SHUT_RDWR);
1209
edc1b01c 1210 mutex_lock(&transport->recv_mutex);
a246b010 1211 write_lock_bh(&sk->sk_callback_lock);
ee0ac0c2
CL		 1212 transport->inet = NULL;
1213 transport->sock = NULL;
a246b010 1214
9903cd1c 1215 sk->sk_user_data = NULL;
2a9e1cfa
TM		 1216
1217 xs_restore_old_callbacks(transport, sk);
0c78789e 1218 xprt_clear_connected(xprt);
a246b010 1219 write_unlock_bh(&sk->sk_callback_lock);
6cc7e908 1220 xs_sock_reset_connection_flags(xprt);
edc1b01c 1221 mutex_unlock(&transport->recv_mutex);
a246b010 1222
6cc7e908 1223 trace_rpc_socket_close(xprt, sock);
a246b010 1224 sock_release(sock);
fe315e76
CL		 1225}
1226
1227/**
1228 * xs_close - close a socket
1229 * @xprt: transport
1230 *
1231 * This is used when all requests are complete; ie, no DRC state remains
1232 * on the server we want to save.
f75e6745
TM		 1233 *
1234 * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with
1235 * xs_reset_transport() zeroing the socket from underneath a writer.
fe315e76
CL		 1236 */
1237static void xs_close(struct rpc_xprt *xprt)
1238{
1239 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1240
1241 dprintk("RPC: xs_close xprt %p\n", xprt);
1242
1243 xs_reset_transport(transport);
61d0a8e6 1244 xprt->reestablish_timeout = 0;
fe315e76 1245
62da3b24 1246 xprt_disconnect_done(xprt);
a246b010
CL		 1247}
1248
4a068258
CL		 1249static void xs_inject_disconnect(struct rpc_xprt *xprt)
1250{
1251 dprintk("RPC: injecting transport disconnect on xprt=%p\n",
1252 xprt);
1253 xprt_disconnect_done(xprt);
1254}
1255
315f3812
KM		 1256static void xs_xprt_free(struct rpc_xprt *xprt)
1257{
1258 xs_free_peer_addresses(xprt);
1259 xprt_free(xprt);
1260}
1261
9903cd1c
CL		 1262/**
1263 * xs_destroy - prepare to shutdown a transport
1264 * @xprt: doomed transport
1265 *
1266 */
1267static void xs_destroy(struct rpc_xprt *xprt)
a246b010 1268{
03c78827
TM		 1269 struct sock_xprt *transport = container_of(xprt,
1270 struct sock_xprt, xprt);
46121cf7 1271 dprintk("RPC: xs_destroy xprt %p\n", xprt);
9903cd1c 1272
03c78827 1273 cancel_delayed_work_sync(&transport->connect_worker);
a1311d87 1274 xs_close(xprt);
edc1b01c 1275 cancel_work_sync(&transport->recv_worker);
315f3812 1276 xs_xprt_free(xprt);
a1311d87 1277 module_put(THIS_MODULE);
a246b010
CL		 1278}
1279
9903cd1c 1280/**
f9b2ee71
TM		 1281 * xs_udp_data_read_skb - receive callback for UDP sockets
1282 * @xprt: transport
1283 * @sk: socket
1284 * @skb: skbuff
9903cd1c 1285 *
a246b010 1286 */
f9b2ee71
TM		 1287static void xs_udp_data_read_skb(struct rpc_xprt *xprt,
1288 struct sock *sk,
1289 struct sk_buff *skb)
a246b010 1290{
9903cd1c 1291 struct rpc_task *task;
a246b010 1292 struct rpc_rqst *rovr;
f9b2ee71 1293 int repsize, copied;
d8ed029d
AD		 1294 u32 _xid;
1295 __be32 *xp;
a246b010 1296
1da8c681 1297 repsize = skb->len;
a246b010 1298 if (repsize < 4) {
46121cf7 1299 dprintk("RPC: impossible RPC reply size %d!\n", repsize);
f9b2ee71 1300 return;
a246b010
CL		 1301 }
1302
1303 /* Copy the XID from the skb... */
1da8c681 1304 xp = skb_header_pointer(skb, 0, sizeof(_xid), &_xid);
a246b010 1305 if (xp == NULL)
f9b2ee71 1306 return;
a246b010
CL		 1307
1308 /* Look up and lock the request corresponding to the given XID */
75c84151 1309 spin_lock(&xprt->queue_lock);
a246b010
CL		 1310 rovr = xprt_lookup_rqst(xprt, *xp);
1311 if (!rovr)
1312 goto out_unlock;
729749bb 1313 xprt_pin_rqst(rovr);
ecd465ee 1314 xprt_update_rtt(rovr->rq_task);
75c84151 1315 spin_unlock(&xprt->queue_lock);
a246b010
CL		 1316 task = rovr->rq_task;
1317
a246b010
CL		 1318 if ((copied = rovr->rq_private_buf.buflen) > repsize)
1319 copied = repsize;
1320
1321 /* Suck it into the iovec, verify checksum if not done by hw. */
1781f7f5 1322 if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
75c84151 1323 spin_lock(&xprt->queue_lock);
0afa6b44 1324 __UDPX_INC_STATS(sk, UDP_MIB_INERRORS);
729749bb 1325 goto out_unpin;
1781f7f5
HX		 1326 }
1327
a246b010 1328
729749bb 1329 spin_lock_bh(&xprt->transport_lock);
6a24dfb6 1330 xprt_adjust_cwnd(xprt, task, copied);
ce7c252a 1331 spin_unlock_bh(&xprt->transport_lock);
75c84151 1332 spin_lock(&xprt->queue_lock);
1570c1e4 1333 xprt_complete_rqst(task, copied);
0afa6b44 1334 __UDPX_INC_STATS(sk, UDP_MIB_INDATAGRAMS);
729749bb
TM		 1335out_unpin:
1336 xprt_unpin_rqst(rovr);
a246b010 1337 out_unlock:
75c84151 1338 spin_unlock(&xprt->queue_lock);
f9b2ee71
TM		 1339}
1340
1341static void xs_udp_data_receive(struct sock_xprt *transport)
1342{
1343 struct sk_buff *skb;
1344 struct sock *sk;
1345 int err;
1346
1347 mutex_lock(&transport->recv_mutex);
1348 sk = transport->inet;
1349 if (sk == NULL)
1350 goto out;
4f546149 1351 clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state);
f9b2ee71 1352 for (;;) {
7c13f97f 1353 skb = skb_recv_udp(sk, 0, 1, &err);
4f546149 1354 if (skb == NULL)
f9b2ee71 1355 break;
4f546149
TM		 1356 xs_udp_data_read_skb(&transport->xprt, sk, skb);
1357 consume_skb(skb);
1358 cond_resched();
f9b2ee71
TM		 1359 }
1360out:
1361 mutex_unlock(&transport->recv_mutex);
1362}
1363
1364static void xs_udp_data_receive_workfn(struct work_struct *work)
1365{
1366 struct sock_xprt *transport =
1367 container_of(work, struct sock_xprt, recv_worker);
1368 xs_udp_data_receive(transport);
1369}
1370
1371/**
1372 * xs_data_ready - "data ready" callback for UDP sockets
1373 * @sk: socket with data to read
1374 *
1375 */
1376static void xs_data_ready(struct sock *sk)
1377{
1378 struct rpc_xprt *xprt;
1379
1380 read_lock_bh(&sk->sk_callback_lock);
1381 dprintk("RPC: xs_data_ready...\n");
1382 xprt = xprt_from_sock(sk);
1383 if (xprt != NULL) {
1384 struct sock_xprt *transport = container_of(xprt,
1385 struct sock_xprt, xprt);
5157b956
TM		 1386 transport->old_data_ready(sk);
1387 /* Any data means we had a useful conversation, so
1388 * then we don't need to delay the next reconnect
1389 */
1390 if (xprt->reestablish_timeout)
1391 xprt->reestablish_timeout = 0;
42d42a5b 1392 if (!test_and_set_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
40a5f1b1 1393 queue_work(xprtiod_workqueue, &transport->recv_worker);
f9b2ee71 1394 }
f064af1e 1395 read_unlock_bh(&sk->sk_callback_lock);
a246b010
CL		 1396}
1397
a519fc7a
TM		 1398/*
1399 * Helper function to force a TCP close if the server is sending
1400 * junk and/or it has put us in CLOSE_WAIT
1401 */
1402static void xs_tcp_force_close(struct rpc_xprt *xprt)
1403{
a519fc7a
TM		 1404 xprt_force_disconnect(xprt);
1405}
1406
9e00abc3 1407#if defined(CONFIG_SUNRPC_BACKCHANNEL)
76566773
CL		 1408static int xs_tcp_bc_up(struct svc_serv *serv, struct net *net)
1409{
1410 int ret;
1411
1412 ret = svc_create_xprt(serv, "tcp-bc", net, PF_INET, 0,
1413 SVC_SOCK_ANONYMOUS);
1414 if (ret < 0)
1415 return ret;
1416 return 0;
1417}
6b26cc8c
CL		 1418
1419static size_t xs_tcp_bc_maxpayload(struct rpc_xprt *xprt)
1420{
1421 return PAGE_SIZE;
1422}
9e00abc3 1423#endif /* CONFIG_SUNRPC_BACKCHANNEL */
44b98efd 1424
9903cd1c
CL		 1425/**
1426 * xs_tcp_state_change - callback to handle TCP socket state changes
1427 * @sk: socket whose state has changed
1428 *
1429 */
1430static void xs_tcp_state_change(struct sock *sk)
a246b010 1431{
9903cd1c 1432 struct rpc_xprt *xprt;
0fdea1e8 1433 struct sock_xprt *transport;
a246b010 1434
f064af1e 1435 read_lock_bh(&sk->sk_callback_lock);
a246b010
CL		 1436 if (!(xprt = xprt_from_sock(sk)))
1437 goto out;
46121cf7 1438 dprintk("RPC: xs_tcp_state_change client %p...\n", xprt);
669502ff 1439 dprintk("RPC: state %x conn %d dead %d zapped %d sk_shutdown %d\n",
46121cf7
CL		 1440 sk->sk_state, xprt_connected(xprt),
1441 sock_flag(sk, SOCK_DEAD),
669502ff
AC		 1442 sock_flag(sk, SOCK_ZAPPED),
1443 sk->sk_shutdown);
a246b010 1444
0fdea1e8 1445 transport = container_of(xprt, struct sock_xprt, xprt);
40b5ea0c 1446 trace_rpc_socket_state_change(xprt, sk->sk_socket);
a246b010
CL		 1447 switch (sk->sk_state) {
1448 case TCP_ESTABLISHED:
f064af1e 1449 spin_lock(&xprt->transport_lock);
a246b010 1450 if (!xprt_test_and_set_connected(xprt)) {
8b71798c 1451 xprt->connect_cookie++;
0fdea1e8
TM		 1452 clear_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
1453 xprt_clear_connecting(xprt);
51971139 1454
3968a8a5
CL		 1455 xprt->stat.connect_count++;
1456 xprt->stat.connect_time += (long)jiffies -
1457 xprt->stat.connect_start;
2a491991 1458 xprt_wake_pending_tasks(xprt, -EAGAIN);
a246b010 1459 }
f064af1e 1460 spin_unlock(&xprt->transport_lock);
a246b010 1461 break;
3b948ae5
TM		 1462 case TCP_FIN_WAIT1:
1463 /* The client initiated a shutdown of the socket */
7c1d71cf 1464 xprt->connect_cookie++;
663b8858 1465 xprt->reestablish_timeout = 0;
3b948ae5 1466 set_bit(XPRT_CLOSING, &xprt->state);
4e857c58 1467 smp_mb__before_atomic();
3b948ae5 1468 clear_bit(XPRT_CONNECTED, &xprt->state);
ef803670 1469 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
4e857c58 1470 smp_mb__after_atomic();
a246b010 1471 break;
632e3bdc 1472 case TCP_CLOSE_WAIT:
3b948ae5 1473 /* The server initiated a shutdown of the socket */
7c1d71cf 1474 xprt->connect_cookie++;
d0bea455 1475 clear_bit(XPRT_CONNECTED, &xprt->state);
a519fc7a 1476 xs_tcp_force_close(xprt);
e9d47639 1477 /* fall through */
663b8858
TM		 1478 case TCP_CLOSING:
1479 /*
1480 * If the server closed down the connection, make sure that
1481 * we back off before reconnecting
1482 */
1483 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1484 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
3b948ae5
TM		 1485 break;
1486 case TCP_LAST_ACK:
670f9457 1487 set_bit(XPRT_CLOSING, &xprt->state);
4e857c58 1488 smp_mb__before_atomic();
3b948ae5 1489 clear_bit(XPRT_CONNECTED, &xprt->state);
4e857c58 1490 smp_mb__after_atomic();
3b948ae5
TM		 1491 break;
1492 case TCP_CLOSE:
0fdea1e8
TM		 1493 if (test_and_clear_bit(XPRT_SOCK_CONNECTING,
1494 &transport->sock_state))
1495 xprt_clear_connecting(xprt);
9b30889c 1496 clear_bit(XPRT_CLOSING, &xprt->state);
3ffbc1d6
N		 1497 if (sk->sk_err)
1498 xprt_wake_pending_tasks(xprt, -sk->sk_err);
9b30889c
TM		 1499 /* Trigger the socket release */
1500 xs_tcp_force_close(xprt);
a246b010
CL		 1501 }
1502 out:
f064af1e 1503 read_unlock_bh(&sk->sk_callback_lock);
a246b010
CL		 1504}
1505
1f0fa154
IJ		 1506static void xs_write_space(struct sock *sk)
1507{
13331a55 1508 struct socket_wq *wq;
1f0fa154
IJ		 1509 struct rpc_xprt *xprt;
1510
13331a55 1511 if (!sk->sk_socket)
1f0fa154 1512 return;
13331a55 1513 clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1f0fa154
IJ		 1514
1515 if (unlikely(!(xprt = xprt_from_sock(sk))))
1516 return;
13331a55
TM		 1517 rcu_read_lock();
1518 wq = rcu_dereference(sk->sk_wq);
1519 if (!wq || test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &wq->flags) == 0)
1520 goto out;
1f0fa154 1521
c544577d
TM		 1522 if (xprt_write_space(xprt))
1523 sk->sk_write_pending--;
13331a55
TM		 1524out:
1525 rcu_read_unlock();
1f0fa154
IJ		 1526}
1527
9903cd1c 1528/**
c7b2cae8
CL		 1529 * xs_udp_write_space - callback invoked when socket buffer space
1530 * becomes available
9903cd1c
CL		 1531 * @sk: socket whose state has changed
1532 *
a246b010
CL		 1533 * Called when more output buffer space is available for this socket.
1534 * We try not to wake our writers until they can make "significant"
c7b2cae8 1535 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
a246b010
CL		 1536 * with a bunch of small requests.
1537 */
c7b2cae8 1538static void xs_udp_write_space(struct sock *sk)
a246b010 1539{
f064af1e 1540 read_lock_bh(&sk->sk_callback_lock);
a246b010 1541
c7b2cae8 1542 /* from net/core/sock.c:sock_def_write_space */
1f0fa154
IJ		 1543 if (sock_writeable(sk))
1544 xs_write_space(sk);
a246b010 1545
f064af1e 1546 read_unlock_bh(&sk->sk_callback_lock);
c7b2cae8 1547}
a246b010 1548
c7b2cae8
CL		 1549/**
1550 * xs_tcp_write_space - callback invoked when socket buffer space
1551 * becomes available
1552 * @sk: socket whose state has changed
1553 *
1554 * Called when more output buffer space is available for this socket.
1555 * We try not to wake our writers until they can make "significant"
1556 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1557 * with a bunch of small requests.
1558 */
1559static void xs_tcp_write_space(struct sock *sk)
1560{
f064af1e 1561 read_lock_bh(&sk->sk_callback_lock);
c7b2cae8
CL		 1562
1563 /* from net/core/stream.c:sk_stream_write_space */
64dc6130 1564 if (sk_stream_is_writeable(sk))
1f0fa154 1565 xs_write_space(sk);
b6ddf64f 1566
f064af1e 1567 read_unlock_bh(&sk->sk_callback_lock);
a246b010
CL		 1568}
1569
470056c2 1570static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
a246b010 1571{
ee0ac0c2
CL		 1572 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1573 struct sock *sk = transport->inet;
a246b010 1574
7c6e066e 1575 if (transport->rcvsize) {
a246b010 1576 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
7c6e066e 1577 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
a246b010 1578 }
7c6e066e 1579 if (transport->sndsize) {
a246b010 1580 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
7c6e066e 1581 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
a246b010
CL		 1582 sk->sk_write_space(sk);
1583 }
1584}
1585
43118c29 1586/**
470056c2 1587 * xs_udp_set_buffer_size - set send and receive limits
43118c29 1588 * @xprt: generic transport
470056c2
CL		 1589 * @sndsize: requested size of send buffer, in bytes
1590 * @rcvsize: requested size of receive buffer, in bytes
43118c29 1591 *
470056c2 1592 * Set socket send and receive buffer size limits.
43118c29 1593 */
470056c2 1594static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
43118c29 1595{
7c6e066e
CL		 1596 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1597
1598 transport->sndsize = 0;
470056c2 1599 if (sndsize)
7c6e066e
CL		 1600 transport->sndsize = sndsize + 1024;
1601 transport->rcvsize = 0;
470056c2 1602 if (rcvsize)
7c6e066e 1603 transport->rcvsize = rcvsize + 1024;
470056c2
CL		 1604
1605 xs_udp_do_set_buffer_size(xprt);
43118c29
CL		 1606}
1607
46c0ee8b
CL		 1608/**
1609 * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1610 * @task: task that timed out
1611 *
1612 * Adjust the congestion window after a retransmit timeout has occurred.
1613 */
6a24dfb6 1614static void xs_udp_timer(struct rpc_xprt *xprt, struct rpc_task *task)
46c0ee8b 1615{
b977b644 1616 spin_lock_bh(&xprt->transport_lock);
6a24dfb6 1617 xprt_adjust_cwnd(xprt, task, -ETIMEDOUT);
b977b644 1618 spin_unlock_bh(&xprt->transport_lock);
46c0ee8b
CL		 1619}
1620
826799e6 1621static int xs_get_random_port(void)
b85d8806 1622{
826799e6
BF		 1623 unsigned short min = xprt_min_resvport, max = xprt_max_resvport;
1624 unsigned short range;
1625 unsigned short rand;
1626
1627 if (max < min)
1628 return -EADDRINUSE;
1629 range = max - min + 1;
1630 rand = (unsigned short) prandom_u32() % range;
1631 return rand + min;
b85d8806
CL		 1632}
1633
4dda9c8a
TM		 1634/**
1635 * xs_set_reuseaddr_port - set the socket's port and address reuse options
1636 * @sock: socket
1637 *
1638 * Note that this function has to be called on all sockets that share the
1639 * same port, and it must be called before binding.
1640 */
1641static void xs_sock_set_reuseport(struct socket *sock)
1642{
402e23b4 1643 int opt = 1;
4dda9c8a 1644
402e23b4
TM		 1645 kernel_setsockopt(sock, SOL_SOCKET, SO_REUSEPORT,
1646 (char *)&opt, sizeof(opt));
4dda9c8a
TM		 1647}
1648
1649static unsigned short xs_sock_getport(struct socket *sock)
1650{
1651 struct sockaddr_storage buf;
4dda9c8a
TM		 1652 unsigned short port = 0;
1653
9b2c45d4 1654 if (kernel_getsockname(sock, (struct sockaddr *)&buf) < 0)
4dda9c8a
TM		 1655 goto out;
1656 switch (buf.ss_family) {
1657 case AF_INET6:
1658 port = ntohs(((struct sockaddr_in6 *)&buf)->sin6_port);
1659 break;
1660 case AF_INET:
1661 port = ntohs(((struct sockaddr_in *)&buf)->sin_port);
1662 }
1663out:
1664 return port;
1665}
1666
92200412
CL		 1667/**
1668 * xs_set_port - reset the port number in the remote endpoint address
1669 * @xprt: generic transport
1670 * @port: new port number
1671 *
1672 */
1673static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1674{
46121cf7 1675 dprintk("RPC: setting port for xprt %p to %u\n", xprt, port);
c4efcb1d 1676
9dc3b095
CL		 1677 rpc_set_port(xs_addr(xprt), port);
1678 xs_update_peer_port(xprt);
92200412
CL		 1679}
1680
4dda9c8a
TM		 1681static void xs_set_srcport(struct sock_xprt *transport, struct socket *sock)
1682{
1683 if (transport->srcport == 0)
1684 transport->srcport = xs_sock_getport(sock);
1685}
1686
826799e6 1687static int xs_get_srcport(struct sock_xprt *transport)
67a391d7 1688{
826799e6 1689 int port = transport->srcport;
67a391d7
TM		 1690
1691 if (port == 0 && transport->xprt.resvport)
1692 port = xs_get_random_port();
1693 return port;
1694}
1695
baaf4e48 1696static unsigned short xs_next_srcport(struct sock_xprt *transport, unsigned short port)
67a391d7 1697{
fbfffbd5
CL		 1698 if (transport->srcport != 0)
1699 transport->srcport = 0;
67a391d7
TM		 1700 if (!transport->xprt.resvport)
1701 return 0;
1702 if (port <= xprt_min_resvport || port > xprt_max_resvport)
1703 return xprt_max_resvport;
1704 return --port;
1705}
beb59b68 1706static int xs_bind(struct sock_xprt *transport, struct socket *sock)
a246b010 1707{
beb59b68 1708 struct sockaddr_storage myaddr;
67a391d7 1709 int err, nloop = 0;
826799e6 1710 int port = xs_get_srcport(transport);
67a391d7 1711 unsigned short last;
a246b010 1712
0f7a622c
CP		 1713 /*
1714 * If we are asking for any ephemeral port (i.e. port == 0 &&
1715 * transport->xprt.resvport == 0), don't bind. Let the local
1716 * port selection happen implicitly when the socket is used
1717 * (for example at connect time).
1718 *
1719 * This ensures that we can continue to establish TCP
1720 * connections even when all local ephemeral ports are already
1721 * a part of some TCP connection. This makes no difference
1722 * for UDP sockets, but also doens't harm them.
1723 *
1724 * If we're asking for any reserved port (i.e. port == 0 &&
1725 * transport->xprt.resvport == 1) xs_get_srcport above will
1726 * ensure that port is non-zero and we will bind as needed.
1727 */
826799e6
BF		 1728 if (port <= 0)
1729 return port;
0f7a622c 1730
beb59b68 1731 memcpy(&myaddr, &transport->srcaddr, transport->xprt.addrlen);
a246b010 1732 do {
beb59b68
PE		 1733 rpc_set_port((struct sockaddr *)&myaddr, port);
1734 err = kernel_bind(sock, (struct sockaddr *)&myaddr,
1735 transport->xprt.addrlen);
a246b010 1736 if (err == 0) {
fbfffbd5 1737 transport->srcport = port;
d3bc9a1d 1738 break;
a246b010 1739 }
67a391d7 1740 last = port;
baaf4e48 1741 port = xs_next_srcport(transport, port);
67a391d7
TM		 1742 if (port > last)
1743 nloop++;
1744 } while (err == -EADDRINUSE && nloop != 2);
90058d37 1745
4232e863 1746 if (myaddr.ss_family == AF_INET)
beb59b68
PE		 1747 dprintk("RPC: %s %pI4:%u: %s (%d)\n", __func__,
1748 &((struct sockaddr_in *)&myaddr)->sin_addr,
1749 port, err ? "failed" : "ok", err);
1750 else
1751 dprintk("RPC: %s %pI6:%u: %s (%d)\n", __func__,
1752 &((struct sockaddr_in6 *)&myaddr)->sin6_addr,
1753 port, err ? "failed" : "ok", err);
a246b010
CL		 1754 return err;
1755}
1756
176e21ee
CL		 1757/*
1758 * We don't support autobind on AF_LOCAL sockets
1759 */
1760static void xs_local_rpcbind(struct rpc_task *task)
1761{
fb43d172 1762 xprt_set_bound(task->tk_xprt);
176e21ee
CL		 1763}
1764
1765static void xs_local_set_port(struct rpc_xprt *xprt, unsigned short port)
1766{
1767}
beb59b68 1768
ed07536e
PZ		 1769#ifdef CONFIG_DEBUG_LOCK_ALLOC
1770static struct lock_class_key xs_key[2];
1771static struct lock_class_key xs_slock_key[2];
1772
176e21ee
CL		 1773static inline void xs_reclassify_socketu(struct socket *sock)
1774{
1775 struct sock *sk = sock->sk;
1776
176e21ee
CL		 1777 sock_lock_init_class_and_name(sk, "slock-AF_LOCAL-RPC",
1778 &xs_slock_key[1], "sk_lock-AF_LOCAL-RPC", &xs_key[1]);
1779}
1780
8945ee5e 1781static inline void xs_reclassify_socket4(struct socket *sock)
ed07536e
PZ		 1782{
1783 struct sock *sk = sock->sk;
8945ee5e 1784
8945ee5e
CL		 1785 sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1786 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1787}
ed07536e 1788
8945ee5e
CL		 1789static inline void xs_reclassify_socket6(struct socket *sock)
1790{
1791 struct sock *sk = sock->sk;
ed07536e 1792
8945ee5e
CL		 1793 sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1794 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
ed07536e 1795}
6bc9638a
PE		 1796
1797static inline void xs_reclassify_socket(int family, struct socket *sock)
1798{
fafc4e1e 1799 if (WARN_ON_ONCE(!sock_allow_reclassification(sock->sk)))
1b7a1819
WAA		 1800 return;
1801
4232e863 1802 switch (family) {
176e21ee
CL		 1803 case AF_LOCAL:
1804 xs_reclassify_socketu(sock);
1805 break;
4232e863 1806 case AF_INET:
6bc9638a 1807 xs_reclassify_socket4(sock);
4232e863
CL		 1808 break;
1809 case AF_INET6:
6bc9638a 1810 xs_reclassify_socket6(sock);
4232e863
CL		 1811 break;
1812 }
6bc9638a 1813}
ed07536e 1814#else
6bc9638a
PE		 1815static inline void xs_reclassify_socket(int family, struct socket *sock)
1816{
1817}
ed07536e
PZ		 1818#endif
1819
93dc41bd
N		 1820static void xs_dummy_setup_socket(struct work_struct *work)
1821{
1822}
1823
6bc9638a 1824static struct socket *xs_create_sock(struct rpc_xprt *xprt,
4dda9c8a
TM		 1825 struct sock_xprt *transport, int family, int type,
1826 int protocol, bool reuseport)
22f79326
PE		 1827{
1828 struct socket *sock;
1829 int err;
1830
6bc9638a 1831 err = __sock_create(xprt->xprt_net, family, type, protocol, &sock, 1);
22f79326
PE		 1832 if (err < 0) {
1833 dprintk("RPC: can't create %d transport socket (%d).\n",
1834 protocol, -err);
1835 goto out;
1836 }
6bc9638a 1837 xs_reclassify_socket(family, sock);
22f79326 1838
4dda9c8a
TM		 1839 if (reuseport)
1840 xs_sock_set_reuseport(sock);
1841
4cea288a
BH		 1842 err = xs_bind(transport, sock);
1843 if (err) {
22f79326
PE		 1844 sock_release(sock);
1845 goto out;
1846 }
1847
1848 return sock;
1849out:
1850 return ERR_PTR(err);
1851}
1852
176e21ee
CL		 1853static int xs_local_finish_connecting(struct rpc_xprt *xprt,
1854 struct socket *sock)
1855{
1856 struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1857 xprt);
1858
1859 if (!transport->inet) {
1860 struct sock *sk = sock->sk;
1861
1862 write_lock_bh(&sk->sk_callback_lock);
1863
1864 xs_save_old_callbacks(transport, sk);
1865
1866 sk->sk_user_data = xprt;
a2648094 1867 sk->sk_data_ready = xs_data_ready;
176e21ee 1868 sk->sk_write_space = xs_udp_write_space;
b4411457 1869 sock_set_flag(sk, SOCK_FASYNC);
2118071d 1870 sk->sk_error_report = xs_error_report;
c2126157 1871 sk->sk_allocation = GFP_NOIO;
176e21ee
CL		 1872
1873 xprt_clear_connected(xprt);
1874
1875 /* Reset to new socket */
1876 transport->sock = sock;
1877 transport->inet = sk;
1878
1879 write_unlock_bh(&sk->sk_callback_lock);
1880 }
1881
550aebfe 1882 xs_stream_reset_connect(transport);
6c7a64e5 1883
176e21ee
CL		 1884 return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, 0);
1885}
1886
1887/**
1888 * xs_local_setup_socket - create AF_LOCAL socket, connect to a local endpoint
176e21ee 1889 * @transport: socket transport to connect
176e21ee 1890 */
dc107402 1891static int xs_local_setup_socket(struct sock_xprt *transport)
176e21ee 1892{
176e21ee
CL		 1893 struct rpc_xprt *xprt = &transport->xprt;
1894 struct socket *sock;
1895 int status = -EIO;
1896
176e21ee
CL		 1897 status = __sock_create(xprt->xprt_net, AF_LOCAL,
1898 SOCK_STREAM, 0, &sock, 1);
1899 if (status < 0) {
1900 dprintk("RPC: can't create AF_LOCAL "
1901 "transport socket (%d).\n", -status);
1902 goto out;
1903 }
d1358917 1904 xs_reclassify_socket(AF_LOCAL, sock);
176e21ee
CL		 1905
1906 dprintk("RPC: worker connecting xprt %p via AF_LOCAL to %s\n",
1907 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1908
1909 status = xs_local_finish_connecting(xprt, sock);
40b5ea0c 1910 trace_rpc_socket_connect(xprt, sock, status);
176e21ee
CL		 1911 switch (status) {
1912 case 0:
1913 dprintk("RPC: xprt %p connected to %s\n",
1914 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
3968a8a5
CL		 1915 xprt->stat.connect_count++;
1916 xprt->stat.connect_time += (long)jiffies -
1917 xprt->stat.connect_start;
176e21ee 1918 xprt_set_connected(xprt);
3601c4a9 1919 case -ENOBUFS:
176e21ee
CL		 1920 break;
1921 case -ENOENT:
1922 dprintk("RPC: xprt %p: socket %s does not exist\n",
1923 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1924 break;
4a20a988
TM		 1925 case -ECONNREFUSED:
1926 dprintk("RPC: xprt %p: connection refused for %s\n",
1927 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1928 break;
176e21ee
CL		 1929 default:
1930 printk(KERN_ERR "%s: unhandled error (%d) connecting to %s\n",
1931 __func__, -status,
1932 xprt->address_strings[RPC_DISPLAY_ADDR]);
1933 }
1934
1935out:
1936 xprt_clear_connecting(xprt);
1937 xprt_wake_pending_tasks(xprt, status);
dc107402
BF		 1938 return status;
1939}
1940
b6669737 1941static void xs_local_connect(struct rpc_xprt *xprt, struct rpc_task *task)
dc107402 1942{
dc107402
BF		 1943 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1944 int ret;
1945
1946 if (RPC_IS_ASYNC(task)) {
1947 /*
1948 * We want the AF_LOCAL connect to be resolved in the
1949 * filesystem namespace of the process making the rpc
1950 * call. Thus we connect synchronously.
1951 *
1952 * If we want to support asynchronous AF_LOCAL calls,
1953 * we'll need to figure out how to pass a namespace to
1954 * connect.
1955 */
1956 rpc_exit(task, -ENOTCONN);
1957 return;
1958 }
1959 ret = xs_local_setup_socket(transport);
1960 if (ret && !RPC_IS_SOFTCONN(task))
1961 msleep_interruptible(15000);
176e21ee
CL		 1962}
1963
3c87ef6e 1964#if IS_ENABLED(CONFIG_SUNRPC_SWAP)
d6e971d8
JL		 1965/*
1966 * Note that this should be called with XPRT_LOCKED held (or when we otherwise
1967 * know that we have exclusive access to the socket), to guard against
1968 * races with xs_reset_transport.
1969 */
a564b8f0
MG		 1970static void xs_set_memalloc(struct rpc_xprt *xprt)
1971{
1972 struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1973 xprt);
1974
d6e971d8
JL		 1975 /*
1976 * If there's no sock, then we have nothing to set. The
1977 * reconnecting process will get it for us.
1978 */
1979 if (!transport->inet)
1980 return;
8e228133 1981 if (atomic_read(&xprt->swapper))
a564b8f0
MG		 1982 sk_set_memalloc(transport->inet);
1983}
1984
1985/**
d67fa4d8 1986 * xs_enable_swap - Tag this transport as being used for swap.
a564b8f0 1987 * @xprt: transport to tag
a564b8f0 1988 *
8e228133
JL		 1989 * Take a reference to this transport on behalf of the rpc_clnt, and
1990 * optionally mark it for swapping if it wasn't already.
a564b8f0 1991 */
d67fa4d8
JL		 1992static int
1993xs_enable_swap(struct rpc_xprt *xprt)
a564b8f0 1994{
d6e971d8 1995 struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
a564b8f0 1996
d6e971d8
JL		 1997 if (atomic_inc_return(&xprt->swapper) != 1)
1998 return 0;
1999 if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE))
2000 return -ERESTARTSYS;
2001 if (xs->inet)
2002 sk_set_memalloc(xs->inet);
2003 xprt_release_xprt(xprt, NULL);
8e228133
JL		 2004 return 0;
2005}
a564b8f0 2006
8e228133 2007/**
d67fa4d8 2008 * xs_disable_swap - Untag this transport as being used for swap.
8e228133
JL		 2009 * @xprt: transport to tag
2010 *
2011 * Drop a "swapper" reference to this xprt on behalf of the rpc_clnt. If the
2012 * swapper refcount goes to 0, untag the socket as a memalloc socket.
2013 */
d67fa4d8
JL		 2014static void
2015xs_disable_swap(struct rpc_xprt *xprt)
8e228133 2016{
d6e971d8 2017 struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
8e228133 2018
d6e971d8
JL		 2019 if (!atomic_dec_and_test(&xprt->swapper))
2020 return;
2021 if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE))
2022 return;
2023 if (xs->inet)
2024 sk_clear_memalloc(xs->inet);
2025 xprt_release_xprt(xprt, NULL);
a564b8f0 2026}
a564b8f0
MG		 2027#else
2028static void xs_set_memalloc(struct rpc_xprt *xprt)
2029{
2030}
d67fa4d8
JL		 2031
2032static int
2033xs_enable_swap(struct rpc_xprt *xprt)
2034{
2035 return -EINVAL;
2036}
2037
2038static void
2039xs_disable_swap(struct rpc_xprt *xprt)
2040{
2041}
a564b8f0
MG		 2042#endif
2043
16be2d20
CL		 2044static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2045{
2046 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2047
2048 if (!transport->inet) {
2049 struct sock *sk = sock->sk;
2050
2051 write_lock_bh(&sk->sk_callback_lock);
2052
2a9e1cfa
TM		 2053 xs_save_old_callbacks(transport, sk);
2054
16be2d20 2055 sk->sk_user_data = xprt;
f9b2ee71 2056 sk->sk_data_ready = xs_data_ready;
16be2d20 2057 sk->sk_write_space = xs_udp_write_space;
b4411457 2058 sock_set_flag(sk, SOCK_FASYNC);
c2126157 2059 sk->sk_allocation = GFP_NOIO;
16be2d20
CL		 2060
2061 xprt_set_connected(xprt);
2062
2063 /* Reset to new socket */
2064 transport->sock = sock;
2065 transport->inet = sk;
2066
a564b8f0
MG		 2067 xs_set_memalloc(xprt);
2068
16be2d20
CL		 2069 write_unlock_bh(&sk->sk_callback_lock);
2070 }
2071 xs_udp_do_set_buffer_size(xprt);
02910177
TM		 2072
2073 xprt->stat.connect_start = jiffies;
16be2d20
CL		 2074}
2075
8c14ff2a 2076static void xs_udp_setup_socket(struct work_struct *work)
a246b010 2077{
34161db6
TM		 2078 struct sock_xprt *transport =
2079 container_of(work, struct sock_xprt, connect_worker.work);
c8475461 2080 struct rpc_xprt *xprt = &transport->xprt;
d099b8af 2081 struct socket *sock;
b65c0310 2082 int status = -EIO;
9903cd1c 2083
8c14ff2a 2084 sock = xs_create_sock(xprt, transport,
4dda9c8a
TM		 2085 xs_addr(xprt)->sa_family, SOCK_DGRAM,
2086 IPPROTO_UDP, false);
b65c0310 2087 if (IS_ERR(sock))
b0d93ad5 2088 goto out;
68e220bd 2089
c740eff8
CL		 2090 dprintk("RPC: worker connecting xprt %p via %s to "
2091 "%s (port %s)\n", xprt,
2092 xprt->address_strings[RPC_DISPLAY_PROTO],
2093 xprt->address_strings[RPC_DISPLAY_ADDR],
2094 xprt->address_strings[RPC_DISPLAY_PORT]);
68e220bd
CL		 2095
2096 xs_udp_finish_connecting(xprt, sock);
40b5ea0c 2097 trace_rpc_socket_connect(xprt, sock, 0);
b0d93ad5
CL		 2098 status = 0;
2099out:
718ba5b8 2100 xprt_unlock_connect(xprt, transport);
b0d93ad5 2101 xprt_clear_connecting(xprt);
7d1e8255 2102 xprt_wake_pending_tasks(xprt, status);
a246b010
CL		 2103}
2104
4876cc77
TM		 2105/**
2106 * xs_tcp_shutdown - gracefully shut down a TCP socket
2107 * @xprt: transport
2108 *
2109 * Initiates a graceful shutdown of the TCP socket by calling the
2110 * equivalent of shutdown(SHUT_RDWR);
2111 */
2112static void xs_tcp_shutdown(struct rpc_xprt *xprt)
2113{
2114 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2115 struct socket *sock = transport->sock;
9b30889c 2116 int skst = transport->inet ? transport->inet->sk_state : TCP_CLOSE;
4876cc77
TM		 2117
2118 if (sock == NULL)
2119 return;
9b30889c
TM		 2120 switch (skst) {
2121 default:
4876cc77
TM		 2122 kernel_sock_shutdown(sock, SHUT_RDWR);
2123 trace_rpc_socket_shutdown(xprt, sock);
9b30889c
TM		 2124 break;
2125 case TCP_CLOSE:
2126 case TCP_TIME_WAIT:
4876cc77 2127 xs_reset_transport(transport);
9b30889c 2128 }
4876cc77
TM		 2129}
2130
8d1b8c62
TM		 2131static void xs_tcp_set_socket_timeouts(struct rpc_xprt *xprt,
2132 struct socket *sock)
2133{
7196dbb0
TM		 2134 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2135 unsigned int keepidle;
2136 unsigned int keepcnt;
8d1b8c62
TM		 2137 unsigned int opt_on = 1;
2138 unsigned int timeo;
2139
7196dbb0
TM		 2140 spin_lock_bh(&xprt->transport_lock);
2141 keepidle = DIV_ROUND_UP(xprt->timeout->to_initval, HZ);
2142 keepcnt = xprt->timeout->to_retries + 1;
2143 timeo = jiffies_to_msecs(xprt->timeout->to_initval) *
2144 (xprt->timeout->to_retries + 1);
2145 clear_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state);
2146 spin_unlock_bh(&xprt->transport_lock);
2147
8d1b8c62
TM		 2148 /* TCP Keepalive options */
2149 kernel_setsockopt(sock, SOL_SOCKET, SO_KEEPALIVE,
2150 (char *)&opt_on, sizeof(opt_on));
2151 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPIDLE,
2152 (char *)&keepidle, sizeof(keepidle));
2153 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPINTVL,
2154 (char *)&keepidle, sizeof(keepidle));
2155 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPCNT,
2156 (char *)&keepcnt, sizeof(keepcnt));
2157
2158 /* TCP user timeout (see RFC5482) */
8d1b8c62
TM		 2159 kernel_setsockopt(sock, SOL_TCP, TCP_USER_TIMEOUT,
2160 (char *)&timeo, sizeof(timeo));
2161}
2162
7196dbb0
TM		 2163static void xs_tcp_set_connect_timeout(struct rpc_xprt *xprt,
2164 unsigned long connect_timeout,
2165 unsigned long reconnect_timeout)
2166{
2167 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2168 struct rpc_timeout to;
2169 unsigned long initval;
2170
2171 spin_lock_bh(&xprt->transport_lock);
2172 if (reconnect_timeout < xprt->max_reconnect_timeout)
2173 xprt->max_reconnect_timeout = reconnect_timeout;
2174 if (connect_timeout < xprt->connect_timeout) {
2175 memcpy(&to, xprt->timeout, sizeof(to));
2176 initval = DIV_ROUND_UP(connect_timeout, to.to_retries + 1);
2177 /* Arbitrary lower limit */
2178 if (initval < XS_TCP_INIT_REEST_TO << 1)
2179 initval = XS_TCP_INIT_REEST_TO << 1;
2180 to.to_initval = initval;
2181 to.to_maxval = initval;
2182 memcpy(&transport->tcp_timeout, &to,
2183 sizeof(transport->tcp_timeout));
2184 xprt->timeout = &transport->tcp_timeout;
2185 xprt->connect_timeout = connect_timeout;
2186 }
2187 set_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state);
2188 spin_unlock_bh(&xprt->transport_lock);
2189}
2190
16be2d20 2191static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
a246b010 2192{
16be2d20 2193 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
fe19a96b 2194 int ret = -ENOTCONN;
edb267a6 2195
ee0ac0c2 2196 if (!transport->inet) {
b0d93ad5 2197 struct sock *sk = sock->sk;
d88e4d82 2198 unsigned int addr_pref = IPV6_PREFER_SRC_PUBLIC;
7f260e85 2199
d88e4d82
N		 2200 /* Avoid temporary address, they are bad for long-lived
2201 * connections such as NFS mounts.
2202 * RFC4941, section 3.6 suggests that:
2203 * Individual applications, which have specific
2204 * knowledge about the normal duration of connections,
2205 * MAY override this as appropriate.
2206 */
2207 kernel_setsockopt(sock, SOL_IPV6, IPV6_ADDR_PREFERENCES,
2208 (char *)&addr_pref, sizeof(addr_pref));
2209
8d1b8c62 2210 xs_tcp_set_socket_timeouts(xprt, sock);
775f06ab 2211
b0d93ad5
CL		 2212 write_lock_bh(&sk->sk_callback_lock);
2213
2a9e1cfa
TM		 2214 xs_save_old_callbacks(transport, sk);
2215
b0d93ad5 2216 sk->sk_user_data = xprt;
5157b956 2217 sk->sk_data_ready = xs_data_ready;
b0d93ad5
CL		 2218 sk->sk_state_change = xs_tcp_state_change;
2219 sk->sk_write_space = xs_tcp_write_space;
b4411457 2220 sock_set_flag(sk, SOCK_FASYNC);
2118071d 2221 sk->sk_error_report = xs_error_report;
c2126157 2222 sk->sk_allocation = GFP_NOIO;
3167e12c
CL		 2223
2224 /* socket options */
3167e12c 2225 sock_reset_flag(sk, SOCK_LINGER);
3167e12c 2226 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
b0d93ad5
CL		 2227
2228 xprt_clear_connected(xprt);
2229
2230 /* Reset to new socket */
ee0ac0c2
CL		 2231 transport->sock = sock;
2232 transport->inet = sk;
b0d93ad5
CL		 2233
2234 write_unlock_bh(&sk->sk_callback_lock);
2235 }
2236
01d37c42 2237 if (!xprt_bound(xprt))
fe19a96b 2238 goto out;
01d37c42 2239
a564b8f0
MG		 2240 xs_set_memalloc(xprt);
2241
e1806c7b 2242 /* Reset TCP record info */
550aebfe 2243 xs_stream_reset_connect(transport);
e1806c7b 2244
b0d93ad5 2245 /* Tell the socket layer to start connecting... */
0fdea1e8 2246 set_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
fe19a96b
TM		 2247 ret = kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
2248 switch (ret) {
2249 case 0:
4dda9c8a 2250 xs_set_srcport(transport, sock);
e9d47639 2251 /* fall through */
fe19a96b
TM		 2252 case -EINPROGRESS:
2253 /* SYN_SENT! */
fe19a96b
TM		 2254 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2255 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1f4c17a0
TM		 2256 break;
2257 case -EADDRNOTAVAIL:
2258 /* Source port number is unavailable. Try a new one! */
2259 transport->srcport = 0;
fe19a96b
TM		 2260 }
2261out:
2262 return ret;
16be2d20
CL		 2263}
2264
9903cd1c 2265/**
b61d59ff 2266 * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint
9903cd1c
CL		 2267 *
2268 * Invoked by a work queue tasklet.
a246b010 2269 */
cdd518d5 2270static void xs_tcp_setup_socket(struct work_struct *work)
a246b010 2271{
cdd518d5
PE		 2272 struct sock_xprt *transport =
2273 container_of(work, struct sock_xprt, connect_worker.work);
ee0ac0c2 2274 struct socket *sock = transport->sock;
a9f5f0f7 2275 struct rpc_xprt *xprt = &transport->xprt;
b61d59ff 2276 int status = -EIO;
a246b010 2277
ee0ac0c2 2278 if (!sock) {
cdd518d5 2279 sock = xs_create_sock(xprt, transport,
4dda9c8a
TM		 2280 xs_addr(xprt)->sa_family, SOCK_STREAM,
2281 IPPROTO_TCP, true);
b61d59ff
TM		 2282 if (IS_ERR(sock)) {
2283 status = PTR_ERR(sock);
3167e12c
CL		 2284 goto out;
2285 }
7d1e8255 2286 }
a246b010 2287
c740eff8
CL		 2288 dprintk("RPC: worker connecting xprt %p via %s to "
2289 "%s (port %s)\n", xprt,
2290 xprt->address_strings[RPC_DISPLAY_PROTO],
2291 xprt->address_strings[RPC_DISPLAY_ADDR],
2292 xprt->address_strings[RPC_DISPLAY_PORT]);
edb267a6 2293
16be2d20 2294 status = xs_tcp_finish_connecting(xprt, sock);
40b5ea0c 2295 trace_rpc_socket_connect(xprt, sock, status);
46121cf7
CL		 2296 dprintk("RPC: %p connect status %d connected %d sock state %d\n",
2297 xprt, -status, xprt_connected(xprt),
2298 sock->sk->sk_state);
2a491991 2299 switch (status) {
f75e6745
TM		 2300 default:
2301 printk("%s: connect returned unhandled error %d\n",
2302 __func__, status);
e9d47639 2303 /* fall through */
f75e6745
TM		 2304 case -EADDRNOTAVAIL:
2305 /* We're probably in TIME_WAIT. Get rid of existing socket,
2306 * and retry
2307 */
a519fc7a 2308 xs_tcp_force_close(xprt);
88b5ed73 2309 break;
2a491991
TM		 2310 case 0:
2311 case -EINPROGRESS:
2312 case -EALREADY:
718ba5b8 2313 xprt_unlock_connect(xprt, transport);
7d1e8255 2314 return;
9fcfe0c8
TM		 2315 case -EINVAL:
2316 /* Happens, for instance, if the user specified a link
2317 * local IPv6 address without a scope-id.
2318 */
3ed5e2a2
TM		 2319 case -ECONNREFUSED:
2320 case -ECONNRESET:
eb5b46fa 2321 case -ENETDOWN:
3ed5e2a2 2322 case -ENETUNREACH:
4ba161a7 2323 case -EHOSTUNREACH:
3913c78c 2324 case -EADDRINUSE:
3601c4a9 2325 case -ENOBUFS:
6ea44adc
N		 2326 /*
2327 * xs_tcp_force_close() wakes tasks with -EIO.
2328 * We need to wake them first to ensure the
2329 * correct error code.
2330 */
2331 xprt_wake_pending_tasks(xprt, status);
4efdd92c 2332 xs_tcp_force_close(xprt);
9fcfe0c8 2333 goto out;
a246b010 2334 }
2a491991 2335 status = -EAGAIN;
a246b010 2336out:
718ba5b8 2337 xprt_unlock_connect(xprt, transport);
68e220bd 2338 xprt_clear_connecting(xprt);
7d1e8255 2339 xprt_wake_pending_tasks(xprt, status);
68e220bd 2340}
b0d93ad5 2341
02910177
TM		 2342static unsigned long xs_reconnect_delay(const struct rpc_xprt *xprt)
2343{
2344 unsigned long start, now = jiffies;
2345
2346 start = xprt->stat.connect_start + xprt->reestablish_timeout;
2347 if (time_after(start, now))
2348 return start - now;
2349 return 0;
2350}
2351
3851f1cd
TM		 2352static void xs_reconnect_backoff(struct rpc_xprt *xprt)
2353{
2354 xprt->reestablish_timeout <<= 1;
2355 if (xprt->reestablish_timeout > xprt->max_reconnect_timeout)
2356 xprt->reestablish_timeout = xprt->max_reconnect_timeout;
2357 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2358 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2359}
2360
9903cd1c
CL		 2361/**
2362 * xs_connect - connect a socket to a remote endpoint
1b092092 2363 * @xprt: pointer to transport structure
9903cd1c
CL		 2364 * @task: address of RPC task that manages state of connect request
2365 *
2366 * TCP: If the remote end dropped the connection, delay reconnecting.
03bf4b70
CL		 2367 *
2368 * UDP socket connects are synchronous, but we use a work queue anyway
2369 * to guarantee that even unprivileged user processes can set up a
2370 * socket on a privileged port.
2371 *
2372 * If a UDP socket connect fails, the delay behavior here prevents
2373 * retry floods (hard mounts).
9903cd1c 2374 */
1b092092 2375static void xs_connect(struct rpc_xprt *xprt, struct rpc_task *task)
a246b010 2376{
ee0ac0c2 2377 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
02910177 2378 unsigned long delay = 0;
a246b010 2379
718ba5b8
TM		 2380 WARN_ON_ONCE(!xprt_lock_connect(xprt, task, transport));
2381
99b1a4c3 2382 if (transport->sock != NULL) {
46121cf7
CL		 2383 dprintk("RPC: xs_connect delayed xprt %p for %lu "
2384 "seconds\n",
03bf4b70 2385 xprt, xprt->reestablish_timeout / HZ);
99b1a4c3
TM		 2386
2387 /* Start by resetting any existing state */
2388 xs_reset_transport(transport);
2389
02910177 2390 delay = xs_reconnect_delay(xprt);
3851f1cd 2391 xs_reconnect_backoff(xprt);
02910177 2392
02910177 2393 } else
46121cf7 2394 dprintk("RPC: xs_connect scheduled xprt %p\n", xprt);
02910177
TM		 2395
2396 queue_delayed_work(xprtiod_workqueue,
2397 &transport->connect_worker,
2398 delay);
a246b010
CL		 2399}
2400
176e21ee
CL		 2401/**
2402 * xs_local_print_stats - display AF_LOCAL socket-specifc stats
2403 * @xprt: rpc_xprt struct containing statistics
2404 * @seq: output file
2405 *
2406 */
2407static void xs_local_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2408{
2409 long idle_time = 0;
2410
2411 if (xprt_connected(xprt))
2412 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2413
2414 seq_printf(seq, "\txprt:\tlocal %lu %lu %lu %ld %lu %lu %lu "
15a45206 2415 "%llu %llu %lu %llu %llu\n",
176e21ee
CL		 2416 xprt->stat.bind_count,
2417 xprt->stat.connect_count,
8440a886 2418 xprt->stat.connect_time / HZ,
176e21ee
CL		 2419 idle_time,
2420 xprt->stat.sends,
2421 xprt->stat.recvs,
2422 xprt->stat.bad_xids,
2423 xprt->stat.req_u,
15a45206
AA		 2424 xprt->stat.bklog_u,
2425 xprt->stat.max_slots,
2426 xprt->stat.sending_u,
2427 xprt->stat.pending_u);
176e21ee
CL		 2428}
2429
262ca07d
CL		 2430/**
2431 * xs_udp_print_stats - display UDP socket-specifc stats
2432 * @xprt: rpc_xprt struct containing statistics
2433 * @seq: output file
2434 *
2435 */
2436static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2437{
c8475461
CL		 2438 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2439
15a45206
AA		 2440 seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %llu %llu "
2441 "%lu %llu %llu\n",
fbfffbd5 2442 transport->srcport,
262ca07d
CL		 2443 xprt->stat.bind_count,
2444 xprt->stat.sends,
2445 xprt->stat.recvs,
2446 xprt->stat.bad_xids,
2447 xprt->stat.req_u,
15a45206
AA		 2448 xprt->stat.bklog_u,
2449 xprt->stat.max_slots,
2450 xprt->stat.sending_u,
2451 xprt->stat.pending_u);
262ca07d
CL		 2452}
2453
2454/**
2455 * xs_tcp_print_stats - display TCP socket-specifc stats
2456 * @xprt: rpc_xprt struct containing statistics
2457 * @seq: output file
2458 *
2459 */
2460static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2461{
c8475461 2462 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
262ca07d
CL		 2463 long idle_time = 0;
2464
2465 if (xprt_connected(xprt))
2466 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2467
15a45206
AA		 2468 seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu "
2469 "%llu %llu %lu %llu %llu\n",
fbfffbd5 2470 transport->srcport,
262ca07d
CL		 2471 xprt->stat.bind_count,
2472 xprt->stat.connect_count,
8440a886 2473 xprt->stat.connect_time / HZ,
262ca07d
CL		 2474 idle_time,
2475 xprt->stat.sends,
2476 xprt->stat.recvs,
2477 xprt->stat.bad_xids,
2478 xprt->stat.req_u,
15a45206
AA		 2479 xprt->stat.bklog_u,
2480 xprt->stat.max_slots,
2481 xprt->stat.sending_u,
2482 xprt->stat.pending_u);
262ca07d
CL		 2483}
2484
4cfc7e60
RI		 2485/*
2486 * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason
2487 * we allocate pages instead doing a kmalloc like rpc_malloc is because we want
2488 * to use the server side send routines.
2489 */
5fe6eaa1 2490static int bc_malloc(struct rpc_task *task)
4cfc7e60 2491{
5fe6eaa1
CL		 2492 struct rpc_rqst *rqst = task->tk_rqstp;
2493 size_t size = rqst->rq_callsize;
4cfc7e60
RI		 2494 struct page *page;
2495 struct rpc_buffer *buf;
2496
5fe6eaa1
CL		 2497 if (size > PAGE_SIZE - sizeof(struct rpc_buffer)) {
2498 WARN_ONCE(1, "xprtsock: large bc buffer request (size %zu)\n",
2499 size);
2500 return -EINVAL;
2501 }
4cfc7e60 2502
b8a13d03 2503 page = alloc_page(GFP_KERNEL);
4cfc7e60 2504 if (!page)
5fe6eaa1 2505 return -ENOMEM;
4cfc7e60
RI		 2506
2507 buf = page_address(page);
2508 buf->len = PAGE_SIZE;
2509
5fe6eaa1 2510 rqst->rq_buffer = buf->data;
18e601d6 2511 rqst->rq_rbuffer = (char *)rqst->rq_buffer + rqst->rq_callsize;
5fe6eaa1 2512 return 0;
4cfc7e60
RI		 2513}
2514
2515/*
2516 * Free the space allocated in the bc_alloc routine
2517 */
3435c74a 2518static void bc_free(struct rpc_task *task)
4cfc7e60 2519{
3435c74a 2520 void *buffer = task->tk_rqstp->rq_buffer;
4cfc7e60
RI		 2521 struct rpc_buffer *buf;
2522
4cfc7e60
RI		 2523 buf = container_of(buffer, struct rpc_buffer, data);
2524 free_page((unsigned long)buf);
2525}
2526
2527/*
2528 * Use the svc_sock to send the callback. Must be called with svsk->sk_mutex
2529 * held. Borrows heavily from svc_tcp_sendto and xs_tcp_send_request.
2530 */
2531static int bc_sendto(struct rpc_rqst *req)
2532{
2533 int len;
2534 struct xdr_buf *xbufp = &req->rq_snd_buf;
2535 struct rpc_xprt *xprt = req->rq_xprt;
2536 struct sock_xprt *transport =
2537 container_of(xprt, struct sock_xprt, xprt);
2538 struct socket *sock = transport->sock;
2539 unsigned long headoff;
2540 unsigned long tailoff;
2541
61677eee 2542 xs_encode_stream_record_marker(xbufp);
4cfc7e60
RI		 2543
2544 tailoff = (unsigned long)xbufp->tail[0].iov_base & ~PAGE_MASK;
2545 headoff = (unsigned long)xbufp->head[0].iov_base & ~PAGE_MASK;
2546 len = svc_send_common(sock, xbufp,
2547 virt_to_page(xbufp->head[0].iov_base), headoff,
2548 xbufp->tail[0].iov_base, tailoff);
2549
2550 if (len != xbufp->len) {
2551 printk(KERN_NOTICE "Error sending entire callback!\n");
2552 len = -EAGAIN;
2553 }
2554
2555 return len;
2556}
2557
2558/*
2559 * The send routine. Borrows from svc_send
2560 */
adfa7144 2561static int bc_send_request(struct rpc_rqst *req)
4cfc7e60 2562{
4cfc7e60 2563 struct svc_xprt *xprt;
7fc56136 2564 int len;
4cfc7e60
RI		 2565
2566 dprintk("sending request with xid: %08x\n", ntohl(req->rq_xid));
2567 /*
2568 * Get the server socket associated with this callback xprt
2569 */
2570 xprt = req->rq_xprt->bc_xprt;
4cfc7e60
RI		 2571
2572 /*
2573 * Grab the mutex to serialize data as the connection is shared
2574 * with the fore channel
2575 */
c544577d 2576 mutex_lock(&xprt->xpt_mutex);
4cfc7e60
RI		 2577 if (test_bit(XPT_DEAD, &xprt->xpt_flags))
2578 len = -ENOTCONN;
2579 else
2580 len = bc_sendto(req);
2581 mutex_unlock(&xprt->xpt_mutex);
2582
2583 if (len > 0)
2584 len = 0;
2585
2586 return len;
2587}
2588
2589/*
2590 * The close routine. Since this is client initiated, we do nothing
2591 */
2592
2593static void bc_close(struct rpc_xprt *xprt)
2594{
4cfc7e60
RI		 2595}
2596
2597/*
2598 * The xprt destroy routine. Again, because this connection is client
2599 * initiated, we do nothing
2600 */
2601
2602static void bc_destroy(struct rpc_xprt *xprt)
2603{
47f72efa
KM		 2604 dprintk("RPC: bc_destroy xprt %p\n", xprt);
2605
2606 xs_xprt_free(xprt);
2607 module_put(THIS_MODULE);
4cfc7e60
RI		 2608}
2609
d31ae254 2610static const struct rpc_xprt_ops xs_local_ops = {
176e21ee 2611 .reserve_xprt = xprt_reserve_xprt,
4cd34e7c 2612 .release_xprt = xprt_release_xprt,
f39c1bfb 2613 .alloc_slot = xprt_alloc_slot,
a9cde23a 2614 .free_slot = xprt_free_slot,
176e21ee
CL		 2615 .rpcbind = xs_local_rpcbind,
2616 .set_port = xs_local_set_port,
dc107402 2617 .connect = xs_local_connect,
176e21ee
CL		 2618 .buf_alloc = rpc_malloc,
2619 .buf_free = rpc_free,
550aebfe 2620 .prepare_request = xs_stream_prepare_request,
176e21ee
CL		 2621 .send_request = xs_local_send_request,
2622 .set_retrans_timeout = xprt_set_retrans_timeout_def,
2623 .close = xs_close,
a1311d87 2624 .destroy = xs_destroy,
176e21ee 2625 .print_stats = xs_local_print_stats,
d67fa4d8
JL		 2626 .enable_swap = xs_enable_swap,
2627 .disable_swap = xs_disable_swap,
176e21ee
CL		 2628};
2629
d31ae254 2630static const struct rpc_xprt_ops xs_udp_ops = {
43118c29 2631 .set_buffer_size = xs_udp_set_buffer_size,
12a80469 2632 .reserve_xprt = xprt_reserve_xprt_cong,
49e9a890 2633 .release_xprt = xprt_release_xprt_cong,
f39c1bfb 2634 .alloc_slot = xprt_alloc_slot,
a9cde23a 2635 .free_slot = xprt_free_slot,
45160d62 2636 .rpcbind = rpcb_getport_async,
92200412 2637 .set_port = xs_set_port,
262965f5 2638 .connect = xs_connect,
02107148
CL		 2639 .buf_alloc = rpc_malloc,
2640 .buf_free = rpc_free,
262965f5 2641 .send_request = xs_udp_send_request,
fe3aca29 2642 .set_retrans_timeout = xprt_set_retrans_timeout_rtt,
46c0ee8b 2643 .timer = xs_udp_timer,
a58dd398 2644 .release_request = xprt_release_rqst_cong,
262965f5
CL		 2645 .close = xs_close,
2646 .destroy = xs_destroy,
262ca07d 2647 .print_stats = xs_udp_print_stats,
d67fa4d8
JL		 2648 .enable_swap = xs_enable_swap,
2649 .disable_swap = xs_disable_swap,
4a068258 2650 .inject_disconnect = xs_inject_disconnect,
262965f5
CL		 2651};
2652
d31ae254 2653static const struct rpc_xprt_ops xs_tcp_ops = {
12a80469 2654 .reserve_xprt = xprt_reserve_xprt,
4cd34e7c 2655 .release_xprt = xprt_release_xprt,
36bd7de9 2656 .alloc_slot = xprt_alloc_slot,
a9cde23a 2657 .free_slot = xprt_free_slot,
45160d62 2658 .rpcbind = rpcb_getport_async,
92200412 2659 .set_port = xs_set_port,
0b9e7943 2660 .connect = xs_connect,
02107148
CL		 2661 .buf_alloc = rpc_malloc,
2662 .buf_free = rpc_free,
277e4ab7 2663 .prepare_request = xs_stream_prepare_request,
262965f5 2664 .send_request = xs_tcp_send_request,
fe3aca29 2665 .set_retrans_timeout = xprt_set_retrans_timeout_def,
c627d31b 2666 .close = xs_tcp_shutdown,
9903cd1c 2667 .destroy = xs_destroy,
7196dbb0 2668 .set_connect_timeout = xs_tcp_set_connect_timeout,
262ca07d 2669 .print_stats = xs_tcp_print_stats,
d67fa4d8
JL		 2670 .enable_swap = xs_enable_swap,
2671 .disable_swap = xs_disable_swap,
4a068258 2672 .inject_disconnect = xs_inject_disconnect,
42e5c3e2
CL		 2673#ifdef CONFIG_SUNRPC_BACKCHANNEL
2674 .bc_setup = xprt_setup_bc,
76566773 2675 .bc_up = xs_tcp_bc_up,
6b26cc8c 2676 .bc_maxpayload = xs_tcp_bc_maxpayload,
42e5c3e2
CL		 2677 .bc_free_rqst = xprt_free_bc_rqst,
2678 .bc_destroy = xprt_destroy_bc,
2679#endif
a246b010
CL		 2680};
2681
4cfc7e60
RI		 2682/*
2683 * The rpc_xprt_ops for the server backchannel
2684 */
2685
d31ae254 2686static const struct rpc_xprt_ops bc_tcp_ops = {
4cfc7e60
RI		 2687 .reserve_xprt = xprt_reserve_xprt,
2688 .release_xprt = xprt_release_xprt,
84e28a30 2689 .alloc_slot = xprt_alloc_slot,
a9cde23a 2690 .free_slot = xprt_free_slot,
4cfc7e60
RI		 2691 .buf_alloc = bc_malloc,
2692 .buf_free = bc_free,
2693 .send_request = bc_send_request,
2694 .set_retrans_timeout = xprt_set_retrans_timeout_def,
2695 .close = bc_close,
2696 .destroy = bc_destroy,
2697 .print_stats = xs_tcp_print_stats,
d67fa4d8
JL		 2698 .enable_swap = xs_enable_swap,
2699 .disable_swap = xs_disable_swap,
4a068258 2700 .inject_disconnect = xs_inject_disconnect,
4cfc7e60
RI		 2701};
2702
92476850
CL		 2703static int xs_init_anyaddr(const int family, struct sockaddr *sap)
2704{
2705 static const struct sockaddr_in sin = {
2706 .sin_family = AF_INET,
2707 .sin_addr.s_addr = htonl(INADDR_ANY),
2708 };
2709 static const struct sockaddr_in6 sin6 = {
2710 .sin6_family = AF_INET6,
2711 .sin6_addr = IN6ADDR_ANY_INIT,
2712 };
2713
2714 switch (family) {
176e21ee
CL		 2715 case AF_LOCAL:
2716 break;
92476850
CL		 2717 case AF_INET:
2718 memcpy(sap, &sin, sizeof(sin));
2719 break;
2720 case AF_INET6:
2721 memcpy(sap, &sin6, sizeof(sin6));
2722 break;
2723 default:
2724 dprintk("RPC: %s: Bad address family\n", __func__);
2725 return -EAFNOSUPPORT;
2726 }
2727 return 0;
2728}
2729
3c341b0b 2730static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
d9ba131d
TM		 2731 unsigned int slot_table_size,
2732 unsigned int max_slot_table_size)
c8541ecd
CL		 2733{
2734 struct rpc_xprt *xprt;
ffc2e518 2735 struct sock_xprt *new;
c8541ecd 2736
96802a09 2737 if (args->addrlen > sizeof(xprt->addr)) {
46121cf7 2738 dprintk("RPC: xs_setup_xprt: address too large\n");
c8541ecd
CL		 2739 return ERR_PTR(-EBADF);
2740 }
2741
d9ba131d
TM		 2742 xprt = xprt_alloc(args->net, sizeof(*new), slot_table_size,
2743 max_slot_table_size);
bd1722d4 2744 if (xprt == NULL) {
46121cf7
CL		 2745 dprintk("RPC: xs_setup_xprt: couldn't allocate "
2746 "rpc_xprt\n");
c8541ecd
CL		 2747 return ERR_PTR(-ENOMEM);
2748 }
c8541ecd 2749
bd1722d4 2750 new = container_of(xprt, struct sock_xprt, xprt);
edc1b01c 2751 mutex_init(&new->recv_mutex);
96802a09
FM		 2752 memcpy(&xprt->addr, args->dstaddr, args->addrlen);
2753 xprt->addrlen = args->addrlen;
d3bc9a1d 2754 if (args->srcaddr)
fbfffbd5 2755 memcpy(&new->srcaddr, args->srcaddr, args->addrlen);
92476850
CL		 2756 else {
2757 int err;
2758 err = xs_init_anyaddr(args->dstaddr->sa_family,
2759 (struct sockaddr *)&new->srcaddr);
2aa13531
SK		 2760 if (err != 0) {
2761 xprt_free(xprt);
92476850 2762 return ERR_PTR(err);
2aa13531 2763 }
92476850 2764 }
c8541ecd
CL		 2765
2766 return xprt;
2767}
2768
176e21ee
CL		 2769static const struct rpc_timeout xs_local_default_timeout = {
2770 .to_initval = 10 * HZ,
2771 .to_maxval = 10 * HZ,
2772 .to_retries = 2,
2773};
2774
2775/**
2776 * xs_setup_local - Set up transport to use an AF_LOCAL socket
2777 * @args: rpc transport creation arguments
2778 *
2779 * AF_LOCAL is a "tpi_cots_ord" transport, just like TCP
2780 */
2781static struct rpc_xprt *xs_setup_local(struct xprt_create *args)
2782{
2783 struct sockaddr_un *sun = (struct sockaddr_un *)args->dstaddr;
2784 struct sock_xprt *transport;
2785 struct rpc_xprt *xprt;
2786 struct rpc_xprt *ret;
2787
d9ba131d
TM		 2788 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2789 xprt_max_tcp_slot_table_entries);
176e21ee
CL		 2790 if (IS_ERR(xprt))
2791 return xprt;
2792 transport = container_of(xprt, struct sock_xprt, xprt);
2793
2794 xprt->prot = 0;
2795 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2796 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2797
2798 xprt->bind_timeout = XS_BIND_TO;
2799 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2800 xprt->idle_timeout = XS_IDLE_DISC_TO;
2801
2802 xprt->ops = &xs_local_ops;
2803 xprt->timeout = &xs_local_default_timeout;
2804
550aebfe
TM		 2805 INIT_WORK(&transport->recv_worker, xs_stream_data_receive_workfn);
2806 INIT_DELAYED_WORK(&transport->connect_worker, xs_dummy_setup_socket);
93dc41bd 2807
176e21ee
CL		 2808 switch (sun->sun_family) {
2809 case AF_LOCAL:
2810 if (sun->sun_path[0] != '/') {
2811 dprintk("RPC: bad AF_LOCAL address: %s\n",
2812 sun->sun_path);
2813 ret = ERR_PTR(-EINVAL);
2814 goto out_err;
2815 }
2816 xprt_set_bound(xprt);
176e21ee 2817 xs_format_peer_addresses(xprt, "local", RPCBIND_NETID_LOCAL);
7073ea87
BF		 2818 ret = ERR_PTR(xs_local_setup_socket(transport));
2819 if (ret)
2820 goto out_err;
176e21ee
CL		 2821 break;
2822 default:
2823 ret = ERR_PTR(-EAFNOSUPPORT);
2824 goto out_err;
2825 }
2826
2827 dprintk("RPC: set up xprt to %s via AF_LOCAL\n",
2828 xprt->address_strings[RPC_DISPLAY_ADDR]);
2829
2830 if (try_module_get(THIS_MODULE))
2831 return xprt;
2832 ret = ERR_PTR(-EINVAL);
2833out_err:
315f3812 2834 xs_xprt_free(xprt);
176e21ee
CL		 2835 return ret;
2836}
2837
2881ae74
TM		 2838static const struct rpc_timeout xs_udp_default_timeout = {
2839 .to_initval = 5 * HZ,
2840 .to_maxval = 30 * HZ,
2841 .to_increment = 5 * HZ,
2842 .to_retries = 5,
2843};
2844
9903cd1c
CL		 2845/**
2846 * xs_setup_udp - Set up transport to use a UDP socket
96802a09 2847 * @args: rpc transport creation arguments
9903cd1c
CL		 2848 *
2849 */
483066d6 2850static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
a246b010 2851{
8f9d5b1a 2852 struct sockaddr *addr = args->dstaddr;
c8541ecd 2853 struct rpc_xprt *xprt;
c8475461 2854 struct sock_xprt *transport;
0a68b0be 2855 struct rpc_xprt *ret;
a246b010 2856
d9ba131d
TM		 2857 xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries,
2858 xprt_udp_slot_table_entries);
c8541ecd
CL		 2859 if (IS_ERR(xprt))
2860 return xprt;
c8475461 2861 transport = container_of(xprt, struct sock_xprt, xprt);
a246b010 2862
ec739ef0 2863 xprt->prot = IPPROTO_UDP;
808012fb 2864 xprt->tsh_size = 0;
a246b010
CL		 2865 /* XXX: header size can vary due to auth type, IPv6, etc. */
2866 xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
2867
03bf4b70 2868 xprt->bind_timeout = XS_BIND_TO;
03bf4b70
CL		 2869 xprt->reestablish_timeout = XS_UDP_REEST_TO;
2870 xprt->idle_timeout = XS_IDLE_DISC_TO;
a246b010 2871
262965f5 2872 xprt->ops = &xs_udp_ops;
a246b010 2873
ba7392bb 2874 xprt->timeout = &xs_udp_default_timeout;
a246b010 2875
f9b2ee71 2876 INIT_WORK(&transport->recv_worker, xs_udp_data_receive_workfn);
edc1b01c
TM		 2877 INIT_DELAYED_WORK(&transport->connect_worker, xs_udp_setup_socket);
2878
8f9d5b1a
CL		 2879 switch (addr->sa_family) {
2880 case AF_INET:
2881 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2882 xprt_set_bound(xprt);
2883
9dc3b095 2884 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
8f9d5b1a
CL		 2885 break;
2886 case AF_INET6:
2887 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2888 xprt_set_bound(xprt);
2889
9dc3b095 2890 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
8f9d5b1a
CL		 2891 break;
2892 default:
0a68b0be
BF		 2893 ret = ERR_PTR(-EAFNOSUPPORT);
2894 goto out_err;
8f9d5b1a
CL		 2895 }
2896
c740eff8
CL		 2897 if (xprt_bound(xprt))
2898 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2899 xprt->address_strings[RPC_DISPLAY_ADDR],
2900 xprt->address_strings[RPC_DISPLAY_PORT],
2901 xprt->address_strings[RPC_DISPLAY_PROTO]);
2902 else
2903 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
2904 xprt->address_strings[RPC_DISPLAY_ADDR],
2905 xprt->address_strings[RPC_DISPLAY_PROTO]);
edb267a6 2906
bc25571e
TT		 2907 if (try_module_get(THIS_MODULE))
2908 return xprt;
0a68b0be
BF		 2909 ret = ERR_PTR(-EINVAL);
2910out_err:
315f3812 2911 xs_xprt_free(xprt);
0a68b0be 2912 return ret;
a246b010
CL		 2913}
2914
2881ae74
TM		 2915static const struct rpc_timeout xs_tcp_default_timeout = {
2916 .to_initval = 60 * HZ,
2917 .to_maxval = 60 * HZ,
2918 .to_retries = 2,
2919};
2920
9903cd1c
CL		 2921/**
2922 * xs_setup_tcp - Set up transport to use a TCP socket
96802a09 2923 * @args: rpc transport creation arguments
9903cd1c
CL		 2924 *
2925 */
483066d6 2926static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
a246b010 2927{
8f9d5b1a 2928 struct sockaddr *addr = args->dstaddr;
c8541ecd 2929 struct rpc_xprt *xprt;
c8475461 2930 struct sock_xprt *transport;
0a68b0be 2931 struct rpc_xprt *ret;
b7993ceb
TM		 2932 unsigned int max_slot_table_size = xprt_max_tcp_slot_table_entries;
2933
2934 if (args->flags & XPRT_CREATE_INFINITE_SLOTS)
2935 max_slot_table_size = RPC_MAX_SLOT_TABLE_LIMIT;
a246b010 2936
d9ba131d 2937 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
b7993ceb 2938 max_slot_table_size);
c8541ecd
CL		 2939 if (IS_ERR(xprt))
2940 return xprt;
c8475461 2941 transport = container_of(xprt, struct sock_xprt, xprt);
a246b010 2942
ec739ef0 2943 xprt->prot = IPPROTO_TCP;
808012fb 2944 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
808012fb 2945 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
a246b010 2946
03bf4b70 2947 xprt->bind_timeout = XS_BIND_TO;
03bf4b70
CL		 2948 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2949 xprt->idle_timeout = XS_IDLE_DISC_TO;
a246b010 2950
262965f5 2951 xprt->ops = &xs_tcp_ops;
ba7392bb 2952 xprt->timeout = &xs_tcp_default_timeout;
a246b010 2953
3851f1cd 2954 xprt->max_reconnect_timeout = xprt->timeout->to_maxval;
7196dbb0
TM		 2955 xprt->connect_timeout = xprt->timeout->to_initval *
2956 (xprt->timeout->to_retries + 1);
3851f1cd 2957
c50b8ee0 2958 INIT_WORK(&transport->recv_worker, xs_stream_data_receive_workfn);
edc1b01c
TM		 2959 INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_setup_socket);
2960
8f9d5b1a
CL		 2961 switch (addr->sa_family) {
2962 case AF_INET:
2963 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2964 xprt_set_bound(xprt);
2965
9dc3b095 2966 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
8f9d5b1a
CL		 2967 break;
2968 case AF_INET6:
2969 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2970 xprt_set_bound(xprt);
2971
9dc3b095 2972 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
8f9d5b1a
CL		 2973 break;
2974 default:
0a68b0be
BF		 2975 ret = ERR_PTR(-EAFNOSUPPORT);
2976 goto out_err;
8f9d5b1a
CL		 2977 }
2978
c740eff8
CL		 2979 if (xprt_bound(xprt))
2980 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2981 xprt->address_strings[RPC_DISPLAY_ADDR],
2982 xprt->address_strings[RPC_DISPLAY_PORT],
2983 xprt->address_strings[RPC_DISPLAY_PROTO]);
2984 else
2985 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
2986 xprt->address_strings[RPC_DISPLAY_ADDR],
2987 xprt->address_strings[RPC_DISPLAY_PROTO]);
2988
bc25571e
TT		 2989 if (try_module_get(THIS_MODULE))
2990 return xprt;
0a68b0be
BF		 2991 ret = ERR_PTR(-EINVAL);
2992out_err:
315f3812 2993 xs_xprt_free(xprt);
0a68b0be 2994 return ret;
a246b010 2995}
282b32e1 2996
f300baba
AB		 2997/**
2998 * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket
2999 * @args: rpc transport creation arguments
3000 *
3001 */
3002static struct rpc_xprt *xs_setup_bc_tcp(struct xprt_create *args)
3003{
3004 struct sockaddr *addr = args->dstaddr;
3005 struct rpc_xprt *xprt;
3006 struct sock_xprt *transport;
3007 struct svc_sock *bc_sock;
0a68b0be 3008 struct rpc_xprt *ret;
f300baba 3009
d9ba131d
TM		 3010 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
3011 xprt_tcp_slot_table_entries);
f300baba
AB		 3012 if (IS_ERR(xprt))
3013 return xprt;
3014 transport = container_of(xprt, struct sock_xprt, xprt);
3015
3016 xprt->prot = IPPROTO_TCP;
3017 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
3018 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
3019 xprt->timeout = &xs_tcp_default_timeout;
3020
3021 /* backchannel */
3022 xprt_set_bound(xprt);
3023 xprt->bind_timeout = 0;
f300baba
AB		 3024 xprt->reestablish_timeout = 0;
3025 xprt->idle_timeout = 0;
3026
f300baba
AB		 3027 xprt->ops = &bc_tcp_ops;
3028
3029 switch (addr->sa_family) {
3030 case AF_INET:
3031 xs_format_peer_addresses(xprt, "tcp",
3032 RPCBIND_NETID_TCP);
3033 break;
3034 case AF_INET6:
3035 xs_format_peer_addresses(xprt, "tcp",
3036 RPCBIND_NETID_TCP6);
3037 break;
3038 default:
0a68b0be
BF		 3039 ret = ERR_PTR(-EAFNOSUPPORT);
3040 goto out_err;
f300baba
AB		 3041 }
3042
50fa0d40
PE		 3043 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
3044 xprt->address_strings[RPC_DISPLAY_ADDR],
3045 xprt->address_strings[RPC_DISPLAY_PORT],
3046 xprt->address_strings[RPC_DISPLAY_PROTO]);
f300baba 3047
99de8ea9
BF		 3048 /*
3049 * Once we've associated a backchannel xprt with a connection,
28303ca3
WM		 3050 * we want to keep it around as long as the connection lasts,
3051 * in case we need to start using it for a backchannel again;
3052 * this reference won't be dropped until bc_xprt is destroyed.
99de8ea9
BF		 3053 */
3054 xprt_get(xprt);
3055 args->bc_xprt->xpt_bc_xprt = xprt;
3056 xprt->bc_xprt = args->bc_xprt;
3057 bc_sock = container_of(args->bc_xprt, struct svc_sock, sk_xprt);
3058 transport->sock = bc_sock->sk_sock;
3059 transport->inet = bc_sock->sk_sk;
3060
f300baba
AB		 3061 /*
3062 * Since we don't want connections for the backchannel, we set
3063 * the xprt status to connected
3064 */
3065 xprt_set_connected(xprt);
3066
f300baba
AB		 3067 if (try_module_get(THIS_MODULE))
3068 return xprt;
642aab58
KM		 3069
3070 args->bc_xprt->xpt_bc_xprt = NULL;
39a9beab 3071 args->bc_xprt->xpt_bc_xps = NULL;
99de8ea9 3072 xprt_put(xprt);
0a68b0be
BF		 3073 ret = ERR_PTR(-EINVAL);
3074out_err:
315f3812 3075 xs_xprt_free(xprt);
0a68b0be 3076 return ret;
f300baba
AB		 3077}
3078
176e21ee
CL		 3079static struct xprt_class xs_local_transport = {
3080 .list = LIST_HEAD_INIT(xs_local_transport.list),
3081 .name = "named UNIX socket",
3082 .owner = THIS_MODULE,
3083 .ident = XPRT_TRANSPORT_LOCAL,
3084 .setup = xs_setup_local,
3085};
3086
bc25571e
TT		 3087static struct xprt_class xs_udp_transport = {
3088 .list = LIST_HEAD_INIT(xs_udp_transport.list),
3089 .name = "udp",
3090 .owner = THIS_MODULE,
f300baba 3091 .ident = XPRT_TRANSPORT_UDP,
bc25571e
TT		 3092 .setup = xs_setup_udp,
3093};
3094
3095static struct xprt_class xs_tcp_transport = {
3096 .list = LIST_HEAD_INIT(xs_tcp_transport.list),
3097 .name = "tcp",
3098 .owner = THIS_MODULE,
f300baba 3099 .ident = XPRT_TRANSPORT_TCP,
bc25571e
TT		 3100 .setup = xs_setup_tcp,
3101};
3102
f300baba
AB		 3103static struct xprt_class xs_bc_tcp_transport = {
3104 .list = LIST_HEAD_INIT(xs_bc_tcp_transport.list),
3105 .name = "tcp NFSv4.1 backchannel",
3106 .owner = THIS_MODULE,
3107 .ident = XPRT_TRANSPORT_BC_TCP,
3108 .setup = xs_setup_bc_tcp,
3109};
3110
282b32e1 3111/**
bc25571e 3112 * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
282b32e1
CL		 3113 *
3114 */
3115int init_socket_xprt(void)
3116{
f895b252 3117#if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
2b1bec5f 3118 if (!sunrpc_table_header)
0b4d4147 3119 sunrpc_table_header = register_sysctl_table(sunrpc_table);
fbf76683
CL		 3120#endif
3121
176e21ee 3122 xprt_register_transport(&xs_local_transport);
bc25571e
TT		 3123 xprt_register_transport(&xs_udp_transport);
3124 xprt_register_transport(&xs_tcp_transport);
f300baba 3125 xprt_register_transport(&xs_bc_tcp_transport);
bc25571e 3126
282b32e1
CL		 3127 return 0;
3128}
3129
3130/**
bc25571e 3131 * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
282b32e1
CL		 3132 *
3133 */
3134void cleanup_socket_xprt(void)
3135{
f895b252 3136#if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
fbf76683
CL		 3137 if (sunrpc_table_header) {
3138 unregister_sysctl_table(sunrpc_table_header);
3139 sunrpc_table_header = NULL;
3140 }
3141#endif
bc25571e 3142
176e21ee 3143 xprt_unregister_transport(&xs_local_transport);
bc25571e
TT		 3144 xprt_unregister_transport(&xs_udp_transport);
3145 xprt_unregister_transport(&xs_tcp_transport);
f300baba 3146 xprt_unregister_transport(&xs_bc_tcp_transport);
282b32e1 3147}
cbf11071 3148
9bbb9e5a
RR		 3149static int param_set_uint_minmax(const char *val,
3150 const struct kernel_param *kp,
cbf11071
TM		 3151 unsigned int min, unsigned int max)
3152{
00cfaa94 3153 unsigned int num;
cbf11071
TM		 3154 int ret;
3155
3156 if (!val)
3157 return -EINVAL;
00cfaa94 3158 ret = kstrtouint(val, 0, &num);
9761a246
DC		 3159 if (ret)
3160 return ret;
3161 if (num < min || num > max)
cbf11071
TM		 3162 return -EINVAL;
3163 *((unsigned int *)kp->arg) = num;
3164 return 0;
3165}
3166
9bbb9e5a 3167static int param_set_portnr(const char *val, const struct kernel_param *kp)
cbf11071 3168{
ffb6ca33 3169 return param_set_uint_minmax(val, kp,
826799e6 3170 RPC_MIN_RESVPORT,
cbf11071
TM		 3171 RPC_MAX_RESVPORT);
3172}
3173
9c27847d 3174static const struct kernel_param_ops param_ops_portnr = {
9bbb9e5a
RR		 3175 .set = param_set_portnr,
3176 .get = param_get_uint,
3177};
3178
cbf11071
TM		 3179#define param_check_portnr(name, p) \
3180 __param_check(name, p, unsigned int);
3181
3182module_param_named(min_resvport, xprt_min_resvport, portnr, 0644);
3183module_param_named(max_resvport, xprt_max_resvport, portnr, 0644);
3184
9bbb9e5a
RR		 3185static int param_set_slot_table_size(const char *val,
3186 const struct kernel_param *kp)
cbf11071
TM		 3187{
3188 return param_set_uint_minmax(val, kp,
3189 RPC_MIN_SLOT_TABLE,
3190 RPC_MAX_SLOT_TABLE);
3191}
3192
9c27847d 3193static const struct kernel_param_ops param_ops_slot_table_size = {
9bbb9e5a
RR		 3194 .set = param_set_slot_table_size,
3195 .get = param_get_uint,
3196};
3197
cbf11071
TM		 3198#define param_check_slot_table_size(name, p) \
3199 __param_check(name, p, unsigned int);
3200
d9ba131d
TM		 3201static int param_set_max_slot_table_size(const char *val,
3202 const struct kernel_param *kp)
3203{
3204 return param_set_uint_minmax(val, kp,
3205 RPC_MIN_SLOT_TABLE,
3206 RPC_MAX_SLOT_TABLE_LIMIT);
3207}
3208
9c27847d 3209static const struct kernel_param_ops param_ops_max_slot_table_size = {
d9ba131d
TM		 3210 .set = param_set_max_slot_table_size,
3211 .get = param_get_uint,
3212};
3213
3214#define param_check_max_slot_table_size(name, p) \
3215 __param_check(name, p, unsigned int);
3216
cbf11071
TM		 3217module_param_named(tcp_slot_table_entries, xprt_tcp_slot_table_entries,
3218 slot_table_size, 0644);
d9ba131d
TM		 3219module_param_named(tcp_max_slot_table_entries, xprt_max_tcp_slot_table_entries,
3220 max_slot_table_size, 0644);
cbf11071
TM		 3221module_param_named(udp_slot_table_entries, xprt_udp_slot_table_entries,
3222 slot_table_size, 0644);
Linux 6.x block layer and io_uring tree(s)