projects / linux-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
SUNRPC: Don't reset the stream record info when the receive worker is running
[linux-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 50#include <linux/bvec.h>
6a829eb8 51#include <linux/highmem.h>
277e4ab7 52#include <linux/uio.h>
a1231fda 53#include <linux/sched/mm.h>
a246b010 54
40b5ea0c
TM		 55#include <trace/events/sunrpc.h>
56
4cfc7e60 57#include "sunrpc.h"
176e21ee
CL		 58
59static void xs_close(struct rpc_xprt *xprt);
7196dbb0
TM		 60static void xs_tcp_set_socket_timeouts(struct rpc_xprt *xprt,
61 struct socket *sock);
176e21ee 62
c556b754
CL		 63/*
64 * xprtsock tunables
65 */
09acfea5
TM		 66static unsigned int xprt_udp_slot_table_entries = RPC_DEF_SLOT_TABLE;
67static unsigned int xprt_tcp_slot_table_entries = RPC_MIN_SLOT_TABLE;
68static unsigned int xprt_max_tcp_slot_table_entries = RPC_MAX_SLOT_TABLE;
c556b754 69
09acfea5
TM		 70static unsigned int xprt_min_resvport = RPC_DEF_MIN_RESVPORT;
71static unsigned int xprt_max_resvport = RPC_DEF_MAX_RESVPORT;
c556b754 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
03bf4b70
CL		 162/*
163 * Wait duration for a reply from the RPC portmapper.
164 */
165#define XS_BIND_TO (60U * HZ)
166
167/*
168 * Delay if a UDP socket connect error occurs. This is most likely some
169 * kind of resource problem on the local host.
170 */
171#define XS_UDP_REEST_TO (2U * HZ)
172
173/*
174 * The reestablish timeout allows clients to delay for a bit before attempting
175 * to reconnect to a server that just dropped our connection.
176 *
177 * We implement an exponential backoff when trying to reestablish a TCP
178 * transport connection with the server. Some servers like to drop a TCP
179 * connection when they are overworked, so we start with a short timeout and
180 * increase over time if the server is down or not responding.
181 */
182#define XS_TCP_INIT_REEST_TO (3U * HZ)
03bf4b70
CL		 183
184/*
185 * TCP idle timeout; client drops the transport socket if it is idle
186 * for this long. Note that we also timeout UDP sockets to prevent
187 * holding port numbers when there is no RPC traffic.
188 */
189#define XS_IDLE_DISC_TO (5U * 60 * HZ)
190
f895b252 191#if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
a246b010 192# undef RPC_DEBUG_DATA
9903cd1c 193# define RPCDBG_FACILITY RPCDBG_TRANS
a246b010
CL		 194#endif
195
a246b010 196#ifdef RPC_DEBUG_DATA
9903cd1c 197static void xs_pktdump(char *msg, u32 *packet, unsigned int count)
a246b010 198{
9903cd1c
CL		 199 u8 *buf = (u8 *) packet;
200 int j;
a246b010 201
46121cf7 202 dprintk("RPC: %s\n", msg);
a246b010
CL		 203 for (j = 0; j < count && j < 128; j += 4) {
204 if (!(j & 31)) {
205 if (j)
206 dprintk("\n");
207 dprintk("0x%04x ", j);
208 }
209 dprintk("%02x%02x%02x%02x ",
210 buf[j], buf[j+1], buf[j+2], buf[j+3]);
211 }
212 dprintk("\n");
213}
214#else
9903cd1c 215static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count)
a246b010
CL		 216{
217 /* NOP */
218}
219#endif
220
2118071d
TM		 221static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
222{
223 return (struct rpc_xprt *) sk->sk_user_data;
224}
225
95392c59
CL		 226static inline struct sockaddr *xs_addr(struct rpc_xprt *xprt)
227{
228 return (struct sockaddr *) &xprt->addr;
229}
230
176e21ee
CL		 231static inline struct sockaddr_un *xs_addr_un(struct rpc_xprt *xprt)
232{
233 return (struct sockaddr_un *) &xprt->addr;
234}
235
95392c59 236static inline struct sockaddr_in *xs_addr_in(struct rpc_xprt *xprt)
edb267a6 237{
95392c59
CL		 238 return (struct sockaddr_in *) &xprt->addr;
239}
240
241static inline struct sockaddr_in6 *xs_addr_in6(struct rpc_xprt *xprt)
242{
243 return (struct sockaddr_in6 *) &xprt->addr;
244}
245
c877b849 246static void xs_format_common_peer_addresses(struct rpc_xprt *xprt)
edb267a6 247{
c877b849 248 struct sockaddr *sap = xs_addr(xprt);
9dc3b095
CL		 249 struct sockaddr_in6 *sin6;
250 struct sockaddr_in *sin;
176e21ee 251 struct sockaddr_un *sun;
c877b849 252 char buf[128];
edb267a6 253
9dc3b095 254 switch (sap->sa_family) {
176e21ee
CL		 255 case AF_LOCAL:
256 sun = xs_addr_un(xprt);
257 strlcpy(buf, sun->sun_path, sizeof(buf));
258 xprt->address_strings[RPC_DISPLAY_ADDR] =
259 kstrdup(buf, GFP_KERNEL);
260 break;
9dc3b095 261 case AF_INET:
176e21ee
CL		 262 (void)rpc_ntop(sap, buf, sizeof(buf));
263 xprt->address_strings[RPC_DISPLAY_ADDR] =
264 kstrdup(buf, GFP_KERNEL);
9dc3b095 265 sin = xs_addr_in(xprt);
fc0b5791 266 snprintf(buf, sizeof(buf), "%08x", ntohl(sin->sin_addr.s_addr));
9dc3b095
CL		 267 break;
268 case AF_INET6:
176e21ee
CL		 269 (void)rpc_ntop(sap, buf, sizeof(buf));
270 xprt->address_strings[RPC_DISPLAY_ADDR] =
271 kstrdup(buf, GFP_KERNEL);
9dc3b095 272 sin6 = xs_addr_in6(xprt);
fc0b5791 273 snprintf(buf, sizeof(buf), "%pi6", &sin6->sin6_addr);
9dc3b095
CL		 274 break;
275 default:
276 BUG();
756805e7 277 }
176e21ee 278
9dc3b095 279 xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = kstrdup(buf, GFP_KERNEL);
edb267a6
CL		 280}
281
9dc3b095 282static void xs_format_common_peer_ports(struct rpc_xprt *xprt)
4b6473fb 283{
9dc3b095
CL		 284 struct sockaddr *sap = xs_addr(xprt);
285 char buf[128];
4b6473fb 286
81160e66 287 snprintf(buf, sizeof(buf), "%u", rpc_get_port(sap));
c877b849 288 xprt->address_strings[RPC_DISPLAY_PORT] = kstrdup(buf, GFP_KERNEL);
4b6473fb 289
81160e66 290 snprintf(buf, sizeof(buf), "%4hx", rpc_get_port(sap));
c877b849
CL		 291 xprt->address_strings[RPC_DISPLAY_HEX_PORT] = kstrdup(buf, GFP_KERNEL);
292}
4b6473fb 293
9dc3b095
CL		 294static void xs_format_peer_addresses(struct rpc_xprt *xprt,
295 const char *protocol,
296 const char *netid)
c877b849 297{
b454ae90 298 xprt->address_strings[RPC_DISPLAY_PROTO] = protocol;
b454ae90 299 xprt->address_strings[RPC_DISPLAY_NETID] = netid;
c877b849 300 xs_format_common_peer_addresses(xprt);
9dc3b095 301 xs_format_common_peer_ports(xprt);
edb267a6 302}
4b6473fb 303
9dc3b095 304static void xs_update_peer_port(struct rpc_xprt *xprt)
4b6473fb 305{
9dc3b095
CL		 306 kfree(xprt->address_strings[RPC_DISPLAY_HEX_PORT]);
307 kfree(xprt->address_strings[RPC_DISPLAY_PORT]);
4417c8c4 308
9dc3b095 309 xs_format_common_peer_ports(xprt);
edb267a6
CL		 310}
311
312static void xs_free_peer_addresses(struct rpc_xprt *xprt)
313{
33e01dc7
CL		 314 unsigned int i;
315
316 for (i = 0; i < RPC_DISPLAY_MAX; i++)
317 switch (i) {
318 case RPC_DISPLAY_PROTO:
319 case RPC_DISPLAY_NETID:
320 continue;
321 default:
322 kfree(xprt->address_strings[i]);
323 }
edb267a6
CL		 324}
325
277e4ab7
TM		 326static size_t
327xs_alloc_sparse_pages(struct xdr_buf *buf, size_t want, gfp_t gfp)
328{
329 size_t i,n;
330
16e5e90f 331 if (!want || !(buf->flags & XDRBUF_SPARSE_PAGES))
277e4ab7 332 return want;
277e4ab7
TM		 333 n = (buf->page_base + want + PAGE_SIZE - 1) >> PAGE_SHIFT;
334 for (i = 0; i < n; i++) {
335 if (buf->pages[i])
336 continue;
337 buf->bvec[i].bv_page = buf->pages[i] = alloc_page(gfp);
338 if (!buf->pages[i]) {
16e5e90f
TM		 339 i *= PAGE_SIZE;
340 return i > buf->page_base ? i - buf->page_base : 0;
277e4ab7
TM		 341 }
342 }
343 return want;
344}
345
346static ssize_t
347xs_sock_recvmsg(struct socket *sock, struct msghdr *msg, int flags, size_t seek)
348{
349 ssize_t ret;
350 if (seek != 0)
351 iov_iter_advance(&msg->msg_iter, seek);
352 ret = sock_recvmsg(sock, msg, flags);
353 return ret > 0 ? ret + seek : ret;
354}
355
356static ssize_t
357xs_read_kvec(struct socket *sock, struct msghdr *msg, int flags,
358 struct kvec *kvec, size_t count, size_t seek)
359{
0e9b4a82 360 iov_iter_kvec(&msg->msg_iter, READ, kvec, 1, count);
277e4ab7
TM		 361 return xs_sock_recvmsg(sock, msg, flags, seek);
362}
363
364static ssize_t
365xs_read_bvec(struct socket *sock, struct msghdr *msg, int flags,
366 struct bio_vec *bvec, unsigned long nr, size_t count,
367 size_t seek)
368{
0e9b4a82 369 iov_iter_bvec(&msg->msg_iter, READ, bvec, nr, count);
277e4ab7
TM		 370 return xs_sock_recvmsg(sock, msg, flags, seek);
371}
372
373static ssize_t
374xs_read_discard(struct socket *sock, struct msghdr *msg, int flags,
375 size_t count)
376{
b76a5afd
TM		 377 iov_iter_discard(&msg->msg_iter, READ, count);
378 return sock_recvmsg(sock, msg, flags);
277e4ab7
TM		 379}
380
6a829eb8
TM		 381#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
382static void
383xs_flush_bvec(const struct bio_vec *bvec, size_t count, size_t seek)
384{
385 struct bvec_iter bi = {
386 .bi_size = count,
387 };
388 struct bio_vec bv;
389
390 bvec_iter_advance(bvec, &bi, seek & PAGE_MASK);
391 for_each_bvec(bv, bvec, bi, bi)
392 flush_dcache_page(bv.bv_page);
393}
394#else
395static inline void
396xs_flush_bvec(const struct bio_vec *bvec, size_t count, size_t seek)
397{
398}
399#endif
400
277e4ab7
TM		 401static ssize_t
402xs_read_xdr_buf(struct socket *sock, struct msghdr *msg, int flags,
403 struct xdr_buf *buf, size_t count, size_t seek, size_t *read)
404{
405 size_t want, seek_init = seek, offset = 0;
406 ssize_t ret;
407
408 if (seek < buf->head[0].iov_len) {
409 want = min_t(size_t, count, buf->head[0].iov_len);
410 ret = xs_read_kvec(sock, msg, flags, &buf->head[0], want, seek);
411 if (ret <= 0)
412 goto sock_err;
413 offset += ret;
414 if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC))
415 goto out;
416 if (ret != want)
c4433055 417 goto out;
277e4ab7
TM		 418 seek = 0;
419 } else {
420 seek -= buf->head[0].iov_len;
421 offset += buf->head[0].iov_len;
422 }
16e5e90f
TM		 423
424 want = xs_alloc_sparse_pages(buf,
425 min_t(size_t, count - offset, buf->page_len),
426 GFP_NOWAIT);
427 if (seek < want) {
277e4ab7
TM		 428 ret = xs_read_bvec(sock, msg, flags, buf->bvec,
429 xdr_buf_pagecount(buf),
430 want + buf->page_base,
431 seek + buf->page_base);
432 if (ret <= 0)
433 goto sock_err;
6a829eb8 434 xs_flush_bvec(buf->bvec, ret, seek + buf->page_base);
277e4ab7
TM		 435 offset += ret - buf->page_base;
436 if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC))
437 goto out;
438 if (ret != want)
c4433055 439 goto out;
277e4ab7
TM		 440 seek = 0;
441 } else {
16e5e90f
TM		 442 seek -= want;
443 offset += want;
277e4ab7 444 }
16e5e90f 445
277e4ab7
TM		 446 if (seek < buf->tail[0].iov_len) {
447 want = min_t(size_t, count - offset, buf->tail[0].iov_len);
448 ret = xs_read_kvec(sock, msg, flags, &buf->tail[0], want, seek);
449 if (ret <= 0)
450 goto sock_err;
451 offset += ret;
452 if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC))
453 goto out;
454 if (ret != want)
c4433055 455 goto out;
277e4ab7
TM		 456 } else
457 offset += buf->tail[0].iov_len;
458 ret = -EMSGSIZE;
277e4ab7
TM		 459out:
460 *read = offset - seek_init;
461 return ret;
277e4ab7
TM		 462sock_err:
463 offset += seek;
464 goto out;
465}
466
467static void
468xs_read_header(struct sock_xprt *transport, struct xdr_buf *buf)
469{
470 if (!transport->recv.copied) {
471 if (buf->head[0].iov_len >= transport->recv.offset)
472 memcpy(buf->head[0].iov_base,
473 &transport->recv.xid,
474 transport->recv.offset);
475 transport->recv.copied = transport->recv.offset;
476 }
477}
478
479static bool
480xs_read_stream_request_done(struct sock_xprt *transport)
481{
482 return transport->recv.fraghdr & cpu_to_be32(RPC_LAST_STREAM_FRAGMENT);
483}
484
485static ssize_t
486xs_read_stream_request(struct sock_xprt *transport, struct msghdr *msg,
487 int flags, struct rpc_rqst *req)
488{
489 struct xdr_buf *buf = &req->rq_private_buf;
490 size_t want, read;
491 ssize_t ret;
492
493 xs_read_header(transport, buf);
494
495 want = transport->recv.len - transport->recv.offset;
496 ret = xs_read_xdr_buf(transport->sock, msg, flags, buf,
497 transport->recv.copied + want, transport->recv.copied,
498 &read);
499 transport->recv.offset += read;
500 transport->recv.copied += read;
501 if (transport->recv.offset == transport->recv.len) {
502 if (xs_read_stream_request_done(transport))
503 msg->msg_flags |= MSG_EOR;
c4433055 504 return read;
277e4ab7
TM		 505 }
506
507 switch (ret) {
c4433055
TM		 508 default:
509 break;
26781eab 510 case -EFAULT:
277e4ab7 511 case -EMSGSIZE:
26781eab 512 msg->msg_flags |= MSG_TRUNC;
c4433055 513 return read;
277e4ab7
TM		 514 case 0:
515 return -ESHUTDOWN;
277e4ab7 516 }
c4433055 517 return ret < 0 ? ret : read;
277e4ab7
TM		 518}
519
520static size_t
521xs_read_stream_headersize(bool isfrag)
522{
523 if (isfrag)
524 return sizeof(__be32);
525 return 3 * sizeof(__be32);
526}
527
528static ssize_t
529xs_read_stream_header(struct sock_xprt *transport, struct msghdr *msg,
530 int flags, size_t want, size_t seek)
531{
532 struct kvec kvec = {
533 .iov_base = &transport->recv.fraghdr,
534 .iov_len = want,
535 };
536 return xs_read_kvec(transport->sock, msg, flags, &kvec, want, seek);
537}
538
539#if defined(CONFIG_SUNRPC_BACKCHANNEL)
540static ssize_t
541xs_read_stream_call(struct sock_xprt *transport, struct msghdr *msg, int flags)
542{
543 struct rpc_xprt *xprt = &transport->xprt;
544 struct rpc_rqst *req;
545 ssize_t ret;
546
547 /* Look up and lock the request corresponding to the given XID */
548 req = xprt_lookup_bc_request(xprt, transport->recv.xid);
549 if (!req) {
550 printk(KERN_WARNING "Callback slot table overflowed\n");
551 return -ESHUTDOWN;
552 }
553
554 ret = xs_read_stream_request(transport, msg, flags, req);
555 if (msg->msg_flags & (MSG_EOR|MSG_TRUNC))
c4433055 556 xprt_complete_bc_request(req, transport->recv.copied);
277e4ab7
TM		 557
558 return ret;
559}
560#else /* CONFIG_SUNRPC_BACKCHANNEL */
561static ssize_t
562xs_read_stream_call(struct sock_xprt *transport, struct msghdr *msg, int flags)
563{
564 return -ESHUTDOWN;
565}
566#endif /* CONFIG_SUNRPC_BACKCHANNEL */
567
568static ssize_t
569xs_read_stream_reply(struct sock_xprt *transport, struct msghdr *msg, int flags)
570{
571 struct rpc_xprt *xprt = &transport->xprt;
572 struct rpc_rqst *req;
573 ssize_t ret = 0;
574
575 /* Look up and lock the request corresponding to the given XID */
576 spin_lock(&xprt->queue_lock);
577 req = xprt_lookup_rqst(xprt, transport->recv.xid);
578 if (!req) {
579 msg->msg_flags |= MSG_TRUNC;
580 goto out;
581 }
582 xprt_pin_rqst(req);
583 spin_unlock(&xprt->queue_lock);
584
585 ret = xs_read_stream_request(transport, msg, flags, req);
586
587 spin_lock(&xprt->queue_lock);
588 if (msg->msg_flags & (MSG_EOR|MSG_TRUNC))
c4433055 589 xprt_complete_rqst(req->rq_task, transport->recv.copied);
277e4ab7
TM		 590 xprt_unpin_rqst(req);
591out:
592 spin_unlock(&xprt->queue_lock);
593 return ret;
594}
595
596static ssize_t
597xs_read_stream(struct sock_xprt *transport, int flags)
598{
599 struct msghdr msg = { 0 };
600 size_t want, read = 0;
601 ssize_t ret = 0;
602
603 if (transport->recv.len == 0) {
604 want = xs_read_stream_headersize(transport->recv.copied != 0);
605 ret = xs_read_stream_header(transport, &msg, flags, want,
606 transport->recv.offset);
607 if (ret <= 0)
608 goto out_err;
609 transport->recv.offset = ret;
c4433055
TM		 610 if (transport->recv.offset != want)
611 return transport->recv.offset;
277e4ab7
TM		 612 transport->recv.len = be32_to_cpu(transport->recv.fraghdr) &
613 RPC_FRAGMENT_SIZE_MASK;
614 transport->recv.offset -= sizeof(transport->recv.fraghdr);
615 read = ret;
616 }
617
618 switch (be32_to_cpu(transport->recv.calldir)) {
c4433055
TM		 619 default:
620 msg.msg_flags |= MSG_TRUNC;
621 break;
277e4ab7
TM		 622 case RPC_CALL:
623 ret = xs_read_stream_call(transport, &msg, flags);
624 break;
625 case RPC_REPLY:
626 ret = xs_read_stream_reply(transport, &msg, flags);
627 }
628 if (msg.msg_flags & MSG_TRUNC) {
629 transport->recv.calldir = cpu_to_be32(-1);
630 transport->recv.copied = -1;
631 }
632 if (ret < 0)
633 goto out_err;
634 read += ret;
635 if (transport->recv.offset < transport->recv.len) {
c4433055
TM		 636 if (!(msg.msg_flags & MSG_TRUNC))
637 return read;
b76a5afd 638 msg.msg_flags = 0;
277e4ab7
TM		 639 ret = xs_read_discard(transport->sock, &msg, flags,
640 transport->recv.len - transport->recv.offset);
641 if (ret <= 0)
642 goto out_err;
643 transport->recv.offset += ret;
644 read += ret;
645 if (transport->recv.offset != transport->recv.len)
c4433055 646 return read;
277e4ab7
TM		 647 }
648 if (xs_read_stream_request_done(transport)) {
c50b8ee0 649 trace_xs_stream_read_request(transport);
277e4ab7
TM		 650 transport->recv.copied = 0;
651 }
652 transport->recv.offset = 0;
653 transport->recv.len = 0;
654 return read;
655out_err:
79462857 656 return ret != 0 ? ret : -ESHUTDOWN;
277e4ab7
TM		 657}
658
0ffe86f4
TM		 659static __poll_t xs_poll_socket(struct sock_xprt *transport)
660{
661 return transport->sock->ops->poll(NULL, transport->sock, NULL);
662}
663
664static bool xs_poll_socket_readable(struct sock_xprt *transport)
665{
666 __poll_t events = xs_poll_socket(transport);
667
668 return (events & (EPOLLIN | EPOLLRDNORM)) && !(events & EPOLLRDHUP);
669}
670
671static void xs_poll_check_readable(struct sock_xprt *transport)
672{
673
674 clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state);
675 if (!xs_poll_socket_readable(transport))
676 return;
677 if (!test_and_set_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
678 queue_work(xprtiod_workqueue, &transport->recv_worker);
679}
680
c50b8ee0
TM		 681static void xs_stream_data_receive(struct sock_xprt *transport)
682{
683 size_t read = 0;
684 ssize_t ret = 0;
685
686 mutex_lock(&transport->recv_mutex);
687 if (transport->sock == NULL)
688 goto out;
c50b8ee0
TM		 689 for (;;) {
690 ret = xs_read_stream(transport, MSG_DONTWAIT);
c4433055 691 if (ret < 0)
c50b8ee0
TM		 692 break;
693 read += ret;
694 cond_resched();
695 }
0ffe86f4 696 xs_poll_check_readable(transport);
c50b8ee0
TM		 697out:
698 mutex_unlock(&transport->recv_mutex);
699 trace_xs_stream_read_data(&transport->xprt, ret, read);
700}
701
702static void xs_stream_data_receive_workfn(struct work_struct *work)
703{
704 struct sock_xprt *transport =
705 container_of(work, struct sock_xprt, recv_worker);
a1231fda
TM		 706 unsigned int pflags = memalloc_nofs_save();
707
c50b8ee0 708 xs_stream_data_receive(transport);
a1231fda 709 memalloc_nofs_restore(pflags);
c50b8ee0
TM		 710}
711
550aebfe
TM		 712static void
713xs_stream_reset_connect(struct sock_xprt *transport)
714{
715 transport->recv.offset = 0;
716 transport->recv.len = 0;
717 transport->recv.copied = 0;
718 transport->xmit.offset = 0;
ae053551
TM		 719}
720
721static void
722xs_stream_start_connect(struct sock_xprt *transport)
723{
550aebfe
TM		 724 transport->xprt.stat.connect_count++;
725 transport->xprt.stat.connect_start = jiffies;
726}
727
b4b5cc85
CL		 728#define XS_SENDMSG_FLAGS (MSG_DONTWAIT | MSG_NOSIGNAL)
729
24c5684b 730static int xs_send_kvec(struct socket *sock, struct sockaddr *addr, int addrlen, struct kvec *vec, unsigned int base, int more)
b4b5cc85 731{
b4b5cc85
CL		 732 struct msghdr msg = {
733 .msg_name = addr,
734 .msg_namelen = addrlen,
24c5684b
TM		 735 .msg_flags = XS_SENDMSG_FLAGS | (more ? MSG_MORE : 0),
736 };
737 struct kvec iov = {
738 .iov_base = vec->iov_base + base,
739 .iov_len = vec->iov_len - base,
b4b5cc85
CL		 740 };
741
24c5684b 742 if (iov.iov_len != 0)
b4b5cc85
CL		 743 return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
744 return kernel_sendmsg(sock, &msg, NULL, 0, 0);
745}
746
f279cd00 747static int xs_send_pagedata(struct socket *sock, struct xdr_buf *xdr, unsigned int base, int more, bool zerocopy, int *sent_p)
b4b5cc85 748{
a6b31d18
TM		 749 ssize_t (*do_sendpage)(struct socket *sock, struct page *page,
750 int offset, size_t size, int flags);
24c5684b
TM		 751 struct page **ppage;
752 unsigned int remainder;
f279cd00 753 int err;
24c5684b
TM		 754
755 remainder = xdr->page_len - base;
756 base += xdr->page_base;
757 ppage = xdr->pages + (base >> PAGE_SHIFT);
758 base &= ~PAGE_MASK;
a6b31d18
TM		 759 do_sendpage = sock->ops->sendpage;
760 if (!zerocopy)
761 do_sendpage = sock_no_sendpage;
24c5684b
TM		 762 for(;;) {
763 unsigned int len = min_t(unsigned int, PAGE_SIZE - base, remainder);
764 int flags = XS_SENDMSG_FLAGS;
b4b5cc85 765
24c5684b 766 remainder -= len;
31303d6c 767 if (more)
24c5684b 768 flags |= MSG_MORE;
31303d6c
TM		 769 if (remainder != 0)
770 flags |= MSG_SENDPAGE_NOTLAST | MSG_MORE;
a6b31d18 771 err = do_sendpage(sock, *ppage, base, len, flags);
24c5684b
TM		 772 if (remainder == 0 || err != len)
773 break;
f279cd00 774 *sent_p += err;
24c5684b
TM		 775 ppage++;
776 base = 0;
777 }
f279cd00
JB		 778 if (err > 0) {
779 *sent_p += err;
780 err = 0;
781 }
782 return err;
b4b5cc85
CL		 783}
784
9903cd1c
CL		 785/**
786 * xs_sendpages - write pages directly to a socket
787 * @sock: socket to send on
788 * @addr: UDP only -- address of destination
789 * @addrlen: UDP only -- length of destination address
790 * @xdr: buffer containing this request
791 * @base: starting position in the buffer
a6b31d18 792 * @zerocopy: true if it is safe to use sendpage()
f279cd00 793 * @sent_p: return the total number of bytes successfully queued for sending
9903cd1c 794 *
a246b010 795 */
f279cd00 796static int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base, bool zerocopy, int *sent_p)
a246b010 797{
24c5684b 798 unsigned int remainder = xdr->len - base;
f279cd00
JB		 799 int err = 0;
800 int sent = 0;
a246b010 801
262965f5 802 if (unlikely(!sock))
fba91afb 803 return -ENOTSOCK;
262965f5 804
24c5684b
TM		 805 if (base != 0) {
806 addr = NULL;
807 addrlen = 0;
808 }
262965f5 809
24c5684b
TM		 810 if (base < xdr->head[0].iov_len || addr != NULL) {
811 unsigned int len = xdr->head[0].iov_len - base;
812 remainder -= len;
813 err = xs_send_kvec(sock, addr, addrlen, &xdr->head[0], base, remainder != 0);
814 if (remainder == 0 || err != len)
a246b010 815 goto out;
f279cd00 816 *sent_p += err;
a246b010
CL		 817 base = 0;
818 } else
24c5684b 819 base -= xdr->head[0].iov_len;
a246b010 820
24c5684b
TM		 821 if (base < xdr->page_len) {
822 unsigned int len = xdr->page_len - base;
823 remainder -= len;
f279cd00
JB		 824 err = xs_send_pagedata(sock, xdr, base, remainder != 0, zerocopy, &sent);
825 *sent_p += sent;
826 if (remainder == 0 || sent != len)
a246b010
CL		 827 goto out;
828 base = 0;
24c5684b
TM		 829 } else
830 base -= xdr->page_len;
831
832 if (base >= xdr->tail[0].iov_len)
f279cd00 833 return 0;
24c5684b 834 err = xs_send_kvec(sock, NULL, 0, &xdr->tail[0], base, 0);
a246b010 835out:
f279cd00
JB		 836 if (err > 0) {
837 *sent_p += err;
838 err = 0;
839 }
840 return err;
a246b010
CL		 841}
842
9903cd1c 843/**
c544577d 844 * xs_nospace - handle transmit was incomplete
50f484e2 845 * @req: pointer to RPC request
9903cd1c 846 *
a246b010 847 */
c544577d 848static int xs_nospace(struct rpc_rqst *req)
a246b010 849{
262965f5 850 struct rpc_xprt *xprt = req->rq_xprt;
ee0ac0c2 851 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
06ea0bfe 852 struct sock *sk = transport->inet;
24ca9a84 853 int ret = -EAGAIN;
a246b010 854
46121cf7 855 dprintk("RPC: %5u xmit incomplete (%u left of %u)\n",
c544577d
TM		 856 req->rq_task->tk_pid,
857 req->rq_slen - transport->xmit.offset,
262965f5
CL		 858 req->rq_slen);
859
b6ddf64f
TM		 860 /* Protect against races with write_space */
861 spin_lock_bh(&xprt->transport_lock);
862
863 /* Don't race with disconnect */
864 if (xprt_connected(xprt)) {
13331a55
TM		 865 /* wait for more buffer space */
866 sk->sk_write_pending++;
c544577d 867 xprt_wait_for_buffer_space(xprt);
13331a55 868 } else
5e3771ce 869 ret = -ENOTCONN;
262965f5 870
b6ddf64f 871 spin_unlock_bh(&xprt->transport_lock);
06ea0bfe
TM		 872
873 /* Race breaker in case memory is freed before above code is called */
d48f9ce7
DV		 874 if (ret == -EAGAIN) {
875 struct socket_wq *wq;
876
877 rcu_read_lock();
878 wq = rcu_dereference(sk->sk_wq);
879 set_bit(SOCKWQ_ASYNC_NOSPACE, &wq->flags);
880 rcu_read_unlock();
881
882 sk->sk_write_space(sk);
883 }
5e3771ce 884 return ret;
262965f5
CL		 885}
886
277e4ab7
TM		 887static void
888xs_stream_prepare_request(struct rpc_rqst *req)
889{
890 req->rq_task->tk_status = xdr_alloc_bvec(&req->rq_rcv_buf, GFP_NOIO);
891}
892
4cd34e7c
TM		 893/*
894 * Determine if the previous message in the stream was aborted before it
895 * could complete transmission.
896 */
897static bool
898xs_send_request_was_aborted(struct sock_xprt *transport, struct rpc_rqst *req)
899{
900 return transport->xmit.offset != 0 && req->rq_bytes_sent == 0;
901}
902
61677eee
CL		 903/*
904 * Construct a stream transport record marker in @buf.
905 */
906static inline void xs_encode_stream_record_marker(struct xdr_buf *buf)
907{
908 u32 reclen = buf->len - sizeof(rpc_fraghdr);
909 rpc_fraghdr *base = buf->head[0].iov_base;
910 *base = cpu_to_be32(RPC_LAST_STREAM_FRAGMENT | reclen);
911}
912
176e21ee
CL		 913/**
914 * xs_local_send_request - write an RPC request to an AF_LOCAL socket
50f484e2 915 * @req: pointer to RPC request
176e21ee
CL		 916 *
917 * Return values:
918 * 0: The request has been sent
919 * EAGAIN: The socket was blocked, please call again later to
920 * complete the request
921 * ENOTCONN: Caller needs to invoke connect logic then call again
922 * other: Some other error occured, the request was not sent
923 */
adfa7144 924static int xs_local_send_request(struct rpc_rqst *req)
176e21ee 925{
176e21ee
CL		 926 struct rpc_xprt *xprt = req->rq_xprt;
927 struct sock_xprt *transport =
928 container_of(xprt, struct sock_xprt, xprt);
929 struct xdr_buf *xdr = &req->rq_snd_buf;
930 int status;
f279cd00 931 int sent = 0;
176e21ee 932
4cd34e7c
TM		 933 /* Close the stream if the previous transmission was incomplete */
934 if (xs_send_request_was_aborted(transport, req)) {
935 xs_close(xprt);
936 return -ENOTCONN;
937 }
938
176e21ee
CL		 939 xs_encode_stream_record_marker(&req->rq_snd_buf);
940
941 xs_pktdump("packet data:",
942 req->rq_svec->iov_base, req->rq_svec->iov_len);
943
78215759 944 req->rq_xtime = ktime_get();
6c7a64e5
TM		 945 status = xs_sendpages(transport->sock, NULL, 0, xdr,
946 transport->xmit.offset,
f279cd00 947 true, &sent);
176e21ee 948 dprintk("RPC: %s(%u) = %d\n",
6c7a64e5 949 __func__, xdr->len - transport->xmit.offset, status);
743c69e7
N		 950
951 if (status == -EAGAIN && sock_writeable(transport->inet))
952 status = -ENOBUFS;
953
f279cd00 954 if (likely(sent > 0) || status == 0) {
6c7a64e5
TM		 955 transport->xmit.offset += sent;
956 req->rq_bytes_sent = transport->xmit.offset;
176e21ee 957 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
6c7a64e5 958 req->rq_xmit_bytes_sent += transport->xmit.offset;
6c7a64e5 959 transport->xmit.offset = 0;
176e21ee
CL		 960 return 0;
961 }
962 status = -EAGAIN;
963 }
964
965 switch (status) {
3601c4a9 966 case -ENOBUFS:
b5872f0c 967 break;
176e21ee 968 case -EAGAIN:
c544577d 969 status = xs_nospace(req);
176e21ee
CL		 970 break;
971 default:
972 dprintk("RPC: sendmsg returned unrecognized error %d\n",
973 -status);
e9d47639 974 /* fall through */
176e21ee
CL		 975 case -EPIPE:
976 xs_close(xprt);
977 status = -ENOTCONN;
978 }
979
980 return status;
981}
982
262965f5
CL		 983/**
984 * xs_udp_send_request - write an RPC request to a UDP socket
50f484e2 985 * @req: pointer to RPC request
262965f5
CL		 986 *
987 * Return values:
988 * 0: The request has been sent
989 * EAGAIN: The socket was blocked, please call again later to
990 * complete the request
991 * ENOTCONN: Caller needs to invoke connect logic then call again
25985edc 992 * other: Some other error occurred, the request was not sent
262965f5 993 */
adfa7144 994static int xs_udp_send_request(struct rpc_rqst *req)
262965f5 995{
262965f5 996 struct rpc_xprt *xprt = req->rq_xprt;
ee0ac0c2 997 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
262965f5 998 struct xdr_buf *xdr = &req->rq_snd_buf;
f279cd00 999 int sent = 0;
262965f5 1000 int status;
a246b010 1001
9903cd1c 1002 xs_pktdump("packet data:",
a246b010
CL		 1003 req->rq_svec->iov_base,
1004 req->rq_svec->iov_len);
1005
01d37c42
TM		 1006 if (!xprt_bound(xprt))
1007 return -ENOTCONN;
75891f50
TM		 1008
1009 if (!xprt_request_get_cong(xprt, req))
1010 return -EBADSLT;
1011
78215759 1012 req->rq_xtime = ktime_get();
f279cd00 1013 status = xs_sendpages(transport->sock, xs_addr(xprt), xprt->addrlen,
6c7a64e5 1014 xdr, 0, true, &sent);
a246b010 1015
46121cf7 1016 dprintk("RPC: xs_udp_send_request(%u) = %d\n",
6c7a64e5 1017 xdr->len, status);
a246b010 1018
3dedbb5c
JB		 1019 /* firewall is blocking us, don't return -EAGAIN or we end up looping */
1020 if (status == -EPERM)
1021 goto process_status;
1022
743c69e7
N		 1023 if (status == -EAGAIN && sock_writeable(transport->inet))
1024 status = -ENOBUFS;
1025
f279cd00
JB		 1026 if (sent > 0 || status == 0) {
1027 req->rq_xmit_bytes_sent += sent;
1028 if (sent >= req->rq_slen)
2199700f
TM		 1029 return 0;
1030 /* Still some bytes left; set up for a retry later. */
262965f5 1031 status = -EAGAIN;
2199700f 1032 }
a246b010 1033
3dedbb5c 1034process_status:
262965f5 1035 switch (status) {
fba91afb
TM		 1036 case -ENOTSOCK:
1037 status = -ENOTCONN;
1038 /* Should we call xs_close() here? */
1039 break;
b6ddf64f 1040 case -EAGAIN:
c544577d 1041 status = xs_nospace(req);
b6ddf64f 1042 break;
262965f5 1043 case -ENETUNREACH:
3601c4a9 1044 case -ENOBUFS:
262965f5 1045 case -EPIPE:
a246b010 1046 case -ECONNREFUSED:
3dedbb5c 1047 case -EPERM:
a246b010 1048 /* When the server has died, an ICMP port unreachable message
9903cd1c 1049 * prompts ECONNREFUSED. */
13331a55
TM		 1050 break;
1051 default:
1052 dprintk("RPC: sendmsg returned unrecognized error %d\n",
1053 -status);
a246b010 1054 }
5fe46e9d 1055
262965f5 1056 return status;
a246b010
CL		 1057}
1058
9903cd1c 1059/**
262965f5 1060 * xs_tcp_send_request - write an RPC request to a TCP socket
50f484e2 1061 * @req: pointer to RPC request
9903cd1c
CL		 1062 *
1063 * Return values:
262965f5
CL		 1064 * 0: The request has been sent
1065 * EAGAIN: The socket was blocked, please call again later to
1066 * complete the request
1067 * ENOTCONN: Caller needs to invoke connect logic then call again
25985edc 1068 * other: Some other error occurred, the request was not sent
9903cd1c
CL		 1069 *
1070 * XXX: In the case of soft timeouts, should we eventually give up
262965f5 1071 * if sendmsg is not able to make progress?
9903cd1c 1072 */
adfa7144 1073static int xs_tcp_send_request(struct rpc_rqst *req)
a246b010 1074{
a246b010 1075 struct rpc_xprt *xprt = req->rq_xprt;
ee0ac0c2 1076 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
262965f5 1077 struct xdr_buf *xdr = &req->rq_snd_buf;
a6b31d18 1078 bool zerocopy = true;
9ffadfbc 1079 bool vm_wait = false;
b595bb15 1080 int status;
f279cd00 1081 int sent;
a246b010 1082
4cd34e7c
TM		 1083 /* Close the stream if the previous transmission was incomplete */
1084 if (xs_send_request_was_aborted(transport, req)) {
1085 if (transport->sock != NULL)
1086 kernel_sock_shutdown(transport->sock, SHUT_RDWR);
1087 return -ENOTCONN;
1088 }
1089
61677eee 1090 xs_encode_stream_record_marker(&req->rq_snd_buf);
a246b010 1091
262965f5
CL		 1092 xs_pktdump("packet data:",
1093 req->rq_svec->iov_base,
1094 req->rq_svec->iov_len);
a6b31d18
TM		 1095 /* Don't use zero copy if this is a resend. If the RPC call
1096 * completes while the socket holds a reference to the pages,
1097 * then we may end up resending corrupted data.
1098 */
50f484e2 1099 if (req->rq_task->tk_flags & RPC_TASK_SENT)
a6b31d18 1100 zerocopy = false;
a246b010 1101
7196dbb0
TM		 1102 if (test_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state))
1103 xs_tcp_set_socket_timeouts(xprt, transport->sock);
1104
a246b010
CL		 1105 /* Continue transmitting the packet/record. We must be careful
1106 * to cope with writespace callbacks arriving _after_ we have
262965f5 1107 * called sendmsg(). */
78215759 1108 req->rq_xtime = ktime_get();
a246b010 1109 while (1) {
f279cd00
JB		 1110 sent = 0;
1111 status = xs_sendpages(transport->sock, NULL, 0, xdr,
6c7a64e5
TM		 1112 transport->xmit.offset,
1113 zerocopy, &sent);
a246b010 1114
46121cf7 1115 dprintk("RPC: xs_tcp_send_request(%u) = %d\n",
6c7a64e5 1116 xdr->len - transport->xmit.offset, status);
a246b010 1117
262965f5
CL		 1118 /* If we've sent the entire packet, immediately
1119 * reset the count of bytes sent. */
6c7a64e5
TM		 1120 transport->xmit.offset += sent;
1121 req->rq_bytes_sent = transport->xmit.offset;
262965f5 1122 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
6c7a64e5 1123 req->rq_xmit_bytes_sent += transport->xmit.offset;
6c7a64e5 1124 transport->xmit.offset = 0;
262965f5
CL		 1125 return 0;
1126 }
a246b010 1127
9ffadfbc
TM		 1128 WARN_ON_ONCE(sent == 0 && status == 0);
1129
1130 if (status == -EAGAIN ) {
1131 /*
1132 * Return EAGAIN if we're sure we're hitting the
1133 * socket send buffer limits.
1134 */
1135 if (test_bit(SOCK_NOSPACE, &transport->sock->flags))
1136 break;
1137 /*
1138 * Did we hit a memory allocation failure?
1139 */
1140 if (sent == 0) {
1141 status = -ENOBUFS;
1142 if (vm_wait)
1143 break;
1144 /* Retry, knowing now that we're below the
1145 * socket send buffer limit
1146 */
1147 vm_wait = true;
1148 }
1149 continue;
1150 }
f580dd04
TM		 1151 if (status < 0)
1152 break;
9ffadfbc 1153 vm_wait = false;
a246b010
CL		 1154 }
1155
262965f5 1156 switch (status) {
fba91afb
TM		 1157 case -ENOTSOCK:
1158 status = -ENOTCONN;
1159 /* Should we call xs_close() here? */
1160 break;
262965f5 1161 case -EAGAIN:
c544577d 1162 status = xs_nospace(req);
262965f5 1163 break;
262965f5 1164 case -ECONNRESET:
2a9e1cfa 1165 case -ECONNREFUSED:
262965f5 1166 case -ENOTCONN:
3913c78c 1167 case -EADDRINUSE:
b5872f0c 1168 case -ENOBUFS:
b9d2bb2e 1169 case -EPIPE:
13331a55
TM		 1170 break;
1171 default:
1172 dprintk("RPC: sendmsg returned unrecognized error %d\n",
1173 -status);
a246b010 1174 }
5fe46e9d 1175
a246b010
CL		 1176 return status;
1177}
1178
2a9e1cfa
TM		 1179static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk)
1180{
1181 transport->old_data_ready = sk->sk_data_ready;
1182 transport->old_state_change = sk->sk_state_change;
1183 transport->old_write_space = sk->sk_write_space;
2118071d 1184 transport->old_error_report = sk->sk_error_report;
2a9e1cfa
TM		 1185}
1186
1187static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk)
1188{
1189 sk->sk_data_ready = transport->old_data_ready;
1190 sk->sk_state_change = transport->old_state_change;
1191 sk->sk_write_space = transport->old_write_space;
2118071d
TM		 1192 sk->sk_error_report = transport->old_error_report;
1193}
1194
42d42a5b
TM		 1195static void xs_sock_reset_state_flags(struct rpc_xprt *xprt)
1196{
1197 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1198
1199 clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state);
1200}
1201
b70ae915
TM		 1202static void xs_sock_reset_connection_flags(struct rpc_xprt *xprt)
1203{
1204 smp_mb__before_atomic();
1205 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1206 clear_bit(XPRT_CLOSING, &xprt->state);
42d42a5b 1207 xs_sock_reset_state_flags(xprt);
b70ae915
TM		 1208 smp_mb__after_atomic();
1209}
1210
2118071d
TM		 1211/**
1212 * xs_error_report - callback to handle TCP socket state errors
1213 * @sk: socket
1214 *
1215 * Note: we don't call sock_error() since there may be a rpc_task
1216 * using the socket, and so we don't want to clear sk->sk_err.
1217 */
1218static void xs_error_report(struct sock *sk)
1219{
1220 struct rpc_xprt *xprt;
1221 int err;
1222
1223 read_lock_bh(&sk->sk_callback_lock);
1224 if (!(xprt = xprt_from_sock(sk)))
1225 goto out;
1226
1227 err = -sk->sk_err;
1228 if (err == 0)
1229 goto out;
1230 dprintk("RPC: xs_error_report client %p, error=%d...\n",
1231 xprt, -err);
e8353c76 1232 trace_rpc_socket_error(xprt, sk->sk_socket, err);
2118071d
TM		 1233 xprt_wake_pending_tasks(xprt, err);
1234 out:
1235 read_unlock_bh(&sk->sk_callback_lock);
2a9e1cfa
TM		 1236}
1237
fe315e76 1238static void xs_reset_transport(struct sock_xprt *transport)
a246b010 1239{
ee0ac0c2
CL		 1240 struct socket *sock = transport->sock;
1241 struct sock *sk = transport->inet;
6cc7e908 1242 struct rpc_xprt *xprt = &transport->xprt;
a246b010 1243
fe315e76
CL		 1244 if (sk == NULL)
1245 return;
9903cd1c 1246
264d1df3
JL		 1247 if (atomic_read(&transport->xprt.swapper))
1248 sk_clear_memalloc(sk);
1249
09939204
TM		 1250 kernel_sock_shutdown(sock, SHUT_RDWR);
1251
edc1b01c 1252 mutex_lock(&transport->recv_mutex);
a246b010 1253 write_lock_bh(&sk->sk_callback_lock);
ee0ac0c2
CL		 1254 transport->inet = NULL;
1255 transport->sock = NULL;
a246b010 1256
9903cd1c 1257 sk->sk_user_data = NULL;
2a9e1cfa
TM		 1258
1259 xs_restore_old_callbacks(transport, sk);
0c78789e 1260 xprt_clear_connected(xprt);
a246b010 1261 write_unlock_bh(&sk->sk_callback_lock);
6cc7e908 1262 xs_sock_reset_connection_flags(xprt);
ae053551
TM		 1263 /* Reset stream record info */
1264 xs_stream_reset_connect(transport);
edc1b01c 1265 mutex_unlock(&transport->recv_mutex);
a246b010 1266
6cc7e908 1267 trace_rpc_socket_close(xprt, sock);
a246b010 1268 sock_release(sock);
0445f92c
TM		 1269
1270 xprt_disconnect_done(xprt);
fe315e76
CL		 1271}
1272
1273/**
1274 * xs_close - close a socket
1275 * @xprt: transport
1276 *
1277 * This is used when all requests are complete; ie, no DRC state remains
1278 * on the server we want to save.
f75e6745
TM		 1279 *
1280 * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with
1281 * xs_reset_transport() zeroing the socket from underneath a writer.
fe315e76
CL		 1282 */
1283static void xs_close(struct rpc_xprt *xprt)
1284{
1285 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1286
1287 dprintk("RPC: xs_close xprt %p\n", xprt);
1288
1289 xs_reset_transport(transport);
61d0a8e6 1290 xprt->reestablish_timeout = 0;
a246b010
CL		 1291}
1292
4a068258
CL		 1293static void xs_inject_disconnect(struct rpc_xprt *xprt)
1294{
1295 dprintk("RPC: injecting transport disconnect on xprt=%p\n",
1296 xprt);
1297 xprt_disconnect_done(xprt);
1298}
1299
315f3812
KM		 1300static void xs_xprt_free(struct rpc_xprt *xprt)
1301{
1302 xs_free_peer_addresses(xprt);
1303 xprt_free(xprt);
1304}
1305
9903cd1c
CL		 1306/**
1307 * xs_destroy - prepare to shutdown a transport
1308 * @xprt: doomed transport
1309 *
1310 */
1311static void xs_destroy(struct rpc_xprt *xprt)
a246b010 1312{
03c78827
TM		 1313 struct sock_xprt *transport = container_of(xprt,
1314 struct sock_xprt, xprt);
46121cf7 1315 dprintk("RPC: xs_destroy xprt %p\n", xprt);
9903cd1c 1316
03c78827 1317 cancel_delayed_work_sync(&transport->connect_worker);
a1311d87 1318 xs_close(xprt);
edc1b01c 1319 cancel_work_sync(&transport->recv_worker);
315f3812 1320 xs_xprt_free(xprt);
a1311d87 1321 module_put(THIS_MODULE);
a246b010
CL		 1322}
1323
9903cd1c 1324/**
f9b2ee71
TM		 1325 * xs_udp_data_read_skb - receive callback for UDP sockets
1326 * @xprt: transport
1327 * @sk: socket
1328 * @skb: skbuff
9903cd1c 1329 *
a246b010 1330 */
f9b2ee71
TM		 1331static void xs_udp_data_read_skb(struct rpc_xprt *xprt,
1332 struct sock *sk,
1333 struct sk_buff *skb)
a246b010 1334{
9903cd1c 1335 struct rpc_task *task;
a246b010 1336 struct rpc_rqst *rovr;
f9b2ee71 1337 int repsize, copied;
d8ed029d
AD		 1338 u32 _xid;
1339 __be32 *xp;
a246b010 1340
1da8c681 1341 repsize = skb->len;
a246b010 1342 if (repsize < 4) {
46121cf7 1343 dprintk("RPC: impossible RPC reply size %d!\n", repsize);
f9b2ee71 1344 return;
a246b010
CL		 1345 }
1346
1347 /* Copy the XID from the skb... */
1da8c681 1348 xp = skb_header_pointer(skb, 0, sizeof(_xid), &_xid);
a246b010 1349 if (xp == NULL)
f9b2ee71 1350 return;
a246b010
CL		 1351
1352 /* Look up and lock the request corresponding to the given XID */
75c84151 1353 spin_lock(&xprt->queue_lock);
a246b010
CL		 1354 rovr = xprt_lookup_rqst(xprt, *xp);
1355 if (!rovr)
1356 goto out_unlock;
729749bb 1357 xprt_pin_rqst(rovr);
ecd465ee 1358 xprt_update_rtt(rovr->rq_task);
75c84151 1359 spin_unlock(&xprt->queue_lock);
a246b010
CL		 1360 task = rovr->rq_task;
1361
a246b010
CL		 1362 if ((copied = rovr->rq_private_buf.buflen) > repsize)
1363 copied = repsize;
1364
1365 /* Suck it into the iovec, verify checksum if not done by hw. */
1781f7f5 1366 if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
75c84151 1367 spin_lock(&xprt->queue_lock);
0afa6b44 1368 __UDPX_INC_STATS(sk, UDP_MIB_INERRORS);
729749bb 1369 goto out_unpin;
1781f7f5
HX		 1370 }
1371
a246b010 1372
729749bb 1373 spin_lock_bh(&xprt->transport_lock);
6a24dfb6 1374 xprt_adjust_cwnd(xprt, task, copied);
ce7c252a 1375 spin_unlock_bh(&xprt->transport_lock);
75c84151 1376 spin_lock(&xprt->queue_lock);
1570c1e4 1377 xprt_complete_rqst(task, copied);
0afa6b44 1378 __UDPX_INC_STATS(sk, UDP_MIB_INDATAGRAMS);
729749bb
TM		 1379out_unpin:
1380 xprt_unpin_rqst(rovr);
a246b010 1381 out_unlock:
75c84151 1382 spin_unlock(&xprt->queue_lock);
f9b2ee71
TM		 1383}
1384
1385static void xs_udp_data_receive(struct sock_xprt *transport)
1386{
1387 struct sk_buff *skb;
1388 struct sock *sk;
1389 int err;
1390
1391 mutex_lock(&transport->recv_mutex);
1392 sk = transport->inet;
1393 if (sk == NULL)
1394 goto out;
1395 for (;;) {
7c13f97f 1396 skb = skb_recv_udp(sk, 0, 1, &err);
4f546149 1397 if (skb == NULL)
f9b2ee71 1398 break;
4f546149
TM		 1399 xs_udp_data_read_skb(&transport->xprt, sk, skb);
1400 consume_skb(skb);
1401 cond_resched();
f9b2ee71 1402 }
0ffe86f4 1403 xs_poll_check_readable(transport);
f9b2ee71
TM		 1404out:
1405 mutex_unlock(&transport->recv_mutex);
1406}
1407
1408static void xs_udp_data_receive_workfn(struct work_struct *work)
1409{
1410 struct sock_xprt *transport =
1411 container_of(work, struct sock_xprt, recv_worker);
a1231fda
TM		 1412 unsigned int pflags = memalloc_nofs_save();
1413
f9b2ee71 1414 xs_udp_data_receive(transport);
a1231fda 1415 memalloc_nofs_restore(pflags);
f9b2ee71
TM		 1416}
1417
1418/**
1419 * xs_data_ready - "data ready" callback for UDP sockets
1420 * @sk: socket with data to read
1421 *
1422 */
1423static void xs_data_ready(struct sock *sk)
1424{
1425 struct rpc_xprt *xprt;
1426
1427 read_lock_bh(&sk->sk_callback_lock);
1428 dprintk("RPC: xs_data_ready...\n");
1429 xprt = xprt_from_sock(sk);
1430 if (xprt != NULL) {
1431 struct sock_xprt *transport = container_of(xprt,
1432 struct sock_xprt, xprt);
5157b956
TM		 1433 transport->old_data_ready(sk);
1434 /* Any data means we had a useful conversation, so
1435 * then we don't need to delay the next reconnect
1436 */
1437 if (xprt->reestablish_timeout)
1438 xprt->reestablish_timeout = 0;
42d42a5b 1439 if (!test_and_set_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
40a5f1b1 1440 queue_work(xprtiod_workqueue, &transport->recv_worker);
f9b2ee71 1441 }
f064af1e 1442 read_unlock_bh(&sk->sk_callback_lock);
a246b010
CL		 1443}
1444
a519fc7a
TM		 1445/*
1446 * Helper function to force a TCP close if the server is sending
1447 * junk and/or it has put us in CLOSE_WAIT
1448 */
1449static void xs_tcp_force_close(struct rpc_xprt *xprt)
1450{
a519fc7a
TM		 1451 xprt_force_disconnect(xprt);
1452}
1453
9e00abc3 1454#if defined(CONFIG_SUNRPC_BACKCHANNEL)
6b26cc8c
CL		 1455static size_t xs_tcp_bc_maxpayload(struct rpc_xprt *xprt)
1456{
1457 return PAGE_SIZE;
1458}
9e00abc3 1459#endif /* CONFIG_SUNRPC_BACKCHANNEL */
44b98efd 1460
9903cd1c
CL		 1461/**
1462 * xs_tcp_state_change - callback to handle TCP socket state changes
1463 * @sk: socket whose state has changed
1464 *
1465 */
1466static void xs_tcp_state_change(struct sock *sk)
a246b010 1467{
9903cd1c 1468 struct rpc_xprt *xprt;
0fdea1e8 1469 struct sock_xprt *transport;
a246b010 1470
f064af1e 1471 read_lock_bh(&sk->sk_callback_lock);
a246b010
CL		 1472 if (!(xprt = xprt_from_sock(sk)))
1473 goto out;
46121cf7 1474 dprintk("RPC: xs_tcp_state_change client %p...\n", xprt);
669502ff 1475 dprintk("RPC: state %x conn %d dead %d zapped %d sk_shutdown %d\n",
46121cf7
CL		 1476 sk->sk_state, xprt_connected(xprt),
1477 sock_flag(sk, SOCK_DEAD),
669502ff
AC		 1478 sock_flag(sk, SOCK_ZAPPED),
1479 sk->sk_shutdown);
a246b010 1480
0fdea1e8 1481 transport = container_of(xprt, struct sock_xprt, xprt);
40b5ea0c 1482 trace_rpc_socket_state_change(xprt, sk->sk_socket);
a246b010
CL		 1483 switch (sk->sk_state) {
1484 case TCP_ESTABLISHED:
f064af1e 1485 spin_lock(&xprt->transport_lock);
a246b010 1486 if (!xprt_test_and_set_connected(xprt)) {
8b71798c 1487 xprt->connect_cookie++;
0fdea1e8
TM		 1488 clear_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
1489 xprt_clear_connecting(xprt);
51971139 1490
3968a8a5
CL		 1491 xprt->stat.connect_count++;
1492 xprt->stat.connect_time += (long)jiffies -
1493 xprt->stat.connect_start;
2a491991 1494 xprt_wake_pending_tasks(xprt, -EAGAIN);
a246b010 1495 }
f064af1e 1496 spin_unlock(&xprt->transport_lock);
a246b010 1497 break;
3b948ae5
TM		 1498 case TCP_FIN_WAIT1:
1499 /* The client initiated a shutdown of the socket */
7c1d71cf 1500 xprt->connect_cookie++;
663b8858 1501 xprt->reestablish_timeout = 0;
3b948ae5 1502 set_bit(XPRT_CLOSING, &xprt->state);
4e857c58 1503 smp_mb__before_atomic();
3b948ae5 1504 clear_bit(XPRT_CONNECTED, &xprt->state);
ef803670 1505 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
4e857c58 1506 smp_mb__after_atomic();
a246b010 1507 break;
632e3bdc 1508 case TCP_CLOSE_WAIT:
3b948ae5 1509 /* The server initiated a shutdown of the socket */
7c1d71cf 1510 xprt->connect_cookie++;
d0bea455 1511 clear_bit(XPRT_CONNECTED, &xprt->state);
a519fc7a 1512 xs_tcp_force_close(xprt);
e9d47639 1513 /* fall through */
663b8858
TM		 1514 case TCP_CLOSING:
1515 /*
1516 * If the server closed down the connection, make sure that
1517 * we back off before reconnecting
1518 */
1519 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1520 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
3b948ae5
TM		 1521 break;
1522 case TCP_LAST_ACK:
670f9457 1523 set_bit(XPRT_CLOSING, &xprt->state);
4e857c58 1524 smp_mb__before_atomic();
3b948ae5 1525 clear_bit(XPRT_CONNECTED, &xprt->state);
4e857c58 1526 smp_mb__after_atomic();
3b948ae5
TM		 1527 break;
1528 case TCP_CLOSE:
0fdea1e8
TM		 1529 if (test_and_clear_bit(XPRT_SOCK_CONNECTING,
1530 &transport->sock_state))
1531 xprt_clear_connecting(xprt);
9b30889c 1532 clear_bit(XPRT_CLOSING, &xprt->state);
9b30889c
TM		 1533 /* Trigger the socket release */
1534 xs_tcp_force_close(xprt);
a246b010
CL		 1535 }
1536 out:
f064af1e 1537 read_unlock_bh(&sk->sk_callback_lock);
a246b010
CL		 1538}
1539
1f0fa154
IJ		 1540static void xs_write_space(struct sock *sk)
1541{
13331a55 1542 struct socket_wq *wq;
1f0fa154
IJ		 1543 struct rpc_xprt *xprt;
1544
13331a55 1545 if (!sk->sk_socket)
1f0fa154 1546 return;
13331a55 1547 clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1f0fa154
IJ		 1548
1549 if (unlikely(!(xprt = xprt_from_sock(sk))))
1550 return;
13331a55
TM		 1551 rcu_read_lock();
1552 wq = rcu_dereference(sk->sk_wq);
1553 if (!wq || test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &wq->flags) == 0)
1554 goto out;
1f0fa154 1555
c544577d
TM		 1556 if (xprt_write_space(xprt))
1557 sk->sk_write_pending--;
13331a55
TM		 1558out:
1559 rcu_read_unlock();
1f0fa154
IJ		 1560}
1561
9903cd1c 1562/**
c7b2cae8
CL		 1563 * xs_udp_write_space - callback invoked when socket buffer space
1564 * becomes available
9903cd1c
CL		 1565 * @sk: socket whose state has changed
1566 *
a246b010
CL		 1567 * Called when more output buffer space is available for this socket.
1568 * We try not to wake our writers until they can make "significant"
c7b2cae8 1569 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
a246b010
CL		 1570 * with a bunch of small requests.
1571 */
c7b2cae8 1572static void xs_udp_write_space(struct sock *sk)
a246b010 1573{
f064af1e 1574 read_lock_bh(&sk->sk_callback_lock);
a246b010 1575
c7b2cae8 1576 /* from net/core/sock.c:sock_def_write_space */
1f0fa154
IJ		 1577 if (sock_writeable(sk))
1578 xs_write_space(sk);
a246b010 1579
f064af1e 1580 read_unlock_bh(&sk->sk_callback_lock);
c7b2cae8 1581}
a246b010 1582
c7b2cae8
CL		 1583/**
1584 * xs_tcp_write_space - callback invoked when socket buffer space
1585 * becomes available
1586 * @sk: socket whose state has changed
1587 *
1588 * Called when more output buffer space is available for this socket.
1589 * We try not to wake our writers until they can make "significant"
1590 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1591 * with a bunch of small requests.
1592 */
1593static void xs_tcp_write_space(struct sock *sk)
1594{
f064af1e 1595 read_lock_bh(&sk->sk_callback_lock);
c7b2cae8
CL		 1596
1597 /* from net/core/stream.c:sk_stream_write_space */
64dc6130 1598 if (sk_stream_is_writeable(sk))
1f0fa154 1599 xs_write_space(sk);
b6ddf64f 1600
f064af1e 1601 read_unlock_bh(&sk->sk_callback_lock);
a246b010
CL		 1602}
1603
470056c2 1604static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
a246b010 1605{
ee0ac0c2
CL		 1606 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1607 struct sock *sk = transport->inet;
a246b010 1608
7c6e066e 1609 if (transport->rcvsize) {
a246b010 1610 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
7c6e066e 1611 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
a246b010 1612 }
7c6e066e 1613 if (transport->sndsize) {
a246b010 1614 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
7c6e066e 1615 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
a246b010
CL		 1616 sk->sk_write_space(sk);
1617 }
1618}
1619
43118c29 1620/**
470056c2 1621 * xs_udp_set_buffer_size - set send and receive limits
43118c29 1622 * @xprt: generic transport
470056c2
CL		 1623 * @sndsize: requested size of send buffer, in bytes
1624 * @rcvsize: requested size of receive buffer, in bytes
43118c29 1625 *
470056c2 1626 * Set socket send and receive buffer size limits.
43118c29 1627 */
470056c2 1628static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
43118c29 1629{
7c6e066e
CL		 1630 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1631
1632 transport->sndsize = 0;
470056c2 1633 if (sndsize)
7c6e066e
CL		 1634 transport->sndsize = sndsize + 1024;
1635 transport->rcvsize = 0;
470056c2 1636 if (rcvsize)
7c6e066e 1637 transport->rcvsize = rcvsize + 1024;
470056c2
CL		 1638
1639 xs_udp_do_set_buffer_size(xprt);
43118c29
CL		 1640}
1641
46c0ee8b
CL		 1642/**
1643 * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
acf0a39f 1644 * @xprt: controlling transport
46c0ee8b
CL		 1645 * @task: task that timed out
1646 *
1647 * Adjust the congestion window after a retransmit timeout has occurred.
1648 */
6a24dfb6 1649static void xs_udp_timer(struct rpc_xprt *xprt, struct rpc_task *task)
46c0ee8b 1650{
b977b644 1651 spin_lock_bh(&xprt->transport_lock);
6a24dfb6 1652 xprt_adjust_cwnd(xprt, task, -ETIMEDOUT);
b977b644 1653 spin_unlock_bh(&xprt->transport_lock);
46c0ee8b
CL		 1654}
1655
826799e6 1656static int xs_get_random_port(void)
b85d8806 1657{
826799e6
BF		 1658 unsigned short min = xprt_min_resvport, max = xprt_max_resvport;
1659 unsigned short range;
1660 unsigned short rand;
1661
1662 if (max < min)
1663 return -EADDRINUSE;
1664 range = max - min + 1;
1665 rand = (unsigned short) prandom_u32() % range;
1666 return rand + min;
b85d8806
CL		 1667}
1668
4dda9c8a
TM		 1669/**
1670 * xs_set_reuseaddr_port - set the socket's port and address reuse options
1671 * @sock: socket
1672 *
1673 * Note that this function has to be called on all sockets that share the
1674 * same port, and it must be called before binding.
1675 */
1676static void xs_sock_set_reuseport(struct socket *sock)
1677{
402e23b4 1678 int opt = 1;
4dda9c8a 1679
402e23b4
TM		 1680 kernel_setsockopt(sock, SOL_SOCKET, SO_REUSEPORT,
1681 (char *)&opt, sizeof(opt));
4dda9c8a
TM		 1682}
1683
1684static unsigned short xs_sock_getport(struct socket *sock)
1685{
1686 struct sockaddr_storage buf;
4dda9c8a
TM		 1687 unsigned short port = 0;
1688
9b2c45d4 1689 if (kernel_getsockname(sock, (struct sockaddr *)&buf) < 0)
4dda9c8a
TM		 1690 goto out;
1691 switch (buf.ss_family) {
1692 case AF_INET6:
1693 port = ntohs(((struct sockaddr_in6 *)&buf)->sin6_port);
1694 break;
1695 case AF_INET:
1696 port = ntohs(((struct sockaddr_in *)&buf)->sin_port);
1697 }
1698out:
1699 return port;
1700}
1701
92200412
CL		 1702/**
1703 * xs_set_port - reset the port number in the remote endpoint address
1704 * @xprt: generic transport
1705 * @port: new port number
1706 *
1707 */
1708static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1709{
46121cf7 1710 dprintk("RPC: setting port for xprt %p to %u\n", xprt, port);
c4efcb1d 1711
9dc3b095
CL		 1712 rpc_set_port(xs_addr(xprt), port);
1713 xs_update_peer_port(xprt);
92200412
CL		 1714}
1715
4dda9c8a
TM		 1716static void xs_set_srcport(struct sock_xprt *transport, struct socket *sock)
1717{
1718 if (transport->srcport == 0)
1719 transport->srcport = xs_sock_getport(sock);
1720}
1721
826799e6 1722static int xs_get_srcport(struct sock_xprt *transport)
67a391d7 1723{
826799e6 1724 int port = transport->srcport;
67a391d7
TM		 1725
1726 if (port == 0 && transport->xprt.resvport)
1727 port = xs_get_random_port();
1728 return port;
1729}
1730
baaf4e48 1731static unsigned short xs_next_srcport(struct sock_xprt *transport, unsigned short port)
67a391d7 1732{
fbfffbd5
CL		 1733 if (transport->srcport != 0)
1734 transport->srcport = 0;
67a391d7
TM		 1735 if (!transport->xprt.resvport)
1736 return 0;
1737 if (port <= xprt_min_resvport || port > xprt_max_resvport)
1738 return xprt_max_resvport;
1739 return --port;
1740}
beb59b68 1741static int xs_bind(struct sock_xprt *transport, struct socket *sock)
a246b010 1742{
beb59b68 1743 struct sockaddr_storage myaddr;
67a391d7 1744 int err, nloop = 0;
826799e6 1745 int port = xs_get_srcport(transport);
67a391d7 1746 unsigned short last;
a246b010 1747
0f7a622c
CP		 1748 /*
1749 * If we are asking for any ephemeral port (i.e. port == 0 &&
1750 * transport->xprt.resvport == 0), don't bind. Let the local
1751 * port selection happen implicitly when the socket is used
1752 * (for example at connect time).
1753 *
1754 * This ensures that we can continue to establish TCP
1755 * connections even when all local ephemeral ports are already
1756 * a part of some TCP connection. This makes no difference
1757 * for UDP sockets, but also doens't harm them.
1758 *
1759 * If we're asking for any reserved port (i.e. port == 0 &&
1760 * transport->xprt.resvport == 1) xs_get_srcport above will
1761 * ensure that port is non-zero and we will bind as needed.
1762 */
826799e6
BF		 1763 if (port <= 0)
1764 return port;
0f7a622c 1765
beb59b68 1766 memcpy(&myaddr, &transport->srcaddr, transport->xprt.addrlen);
a246b010 1767 do {
beb59b68
PE		 1768 rpc_set_port((struct sockaddr *)&myaddr, port);
1769 err = kernel_bind(sock, (struct sockaddr *)&myaddr,
1770 transport->xprt.addrlen);
a246b010 1771 if (err == 0) {
fbfffbd5 1772 transport->srcport = port;
d3bc9a1d 1773 break;
a246b010 1774 }
67a391d7 1775 last = port;
baaf4e48 1776 port = xs_next_srcport(transport, port);
67a391d7
TM		 1777 if (port > last)
1778 nloop++;
1779 } while (err == -EADDRINUSE && nloop != 2);
90058d37 1780
4232e863 1781 if (myaddr.ss_family == AF_INET)
beb59b68
PE		 1782 dprintk("RPC: %s %pI4:%u: %s (%d)\n", __func__,
1783 &((struct sockaddr_in *)&myaddr)->sin_addr,
1784 port, err ? "failed" : "ok", err);
1785 else
1786 dprintk("RPC: %s %pI6:%u: %s (%d)\n", __func__,
1787 &((struct sockaddr_in6 *)&myaddr)->sin6_addr,
1788 port, err ? "failed" : "ok", err);
a246b010
CL		 1789 return err;
1790}
1791
176e21ee
CL		 1792/*
1793 * We don't support autobind on AF_LOCAL sockets
1794 */
1795static void xs_local_rpcbind(struct rpc_task *task)
1796{
fb43d172 1797 xprt_set_bound(task->tk_xprt);
176e21ee
CL		 1798}
1799
1800static void xs_local_set_port(struct rpc_xprt *xprt, unsigned short port)
1801{
1802}
beb59b68 1803
ed07536e
PZ		 1804#ifdef CONFIG_DEBUG_LOCK_ALLOC
1805static struct lock_class_key xs_key[2];
1806static struct lock_class_key xs_slock_key[2];
1807
176e21ee
CL		 1808static inline void xs_reclassify_socketu(struct socket *sock)
1809{
1810 struct sock *sk = sock->sk;
1811
176e21ee
CL		 1812 sock_lock_init_class_and_name(sk, "slock-AF_LOCAL-RPC",
1813 &xs_slock_key[1], "sk_lock-AF_LOCAL-RPC", &xs_key[1]);
1814}
1815
8945ee5e 1816static inline void xs_reclassify_socket4(struct socket *sock)
ed07536e
PZ		 1817{
1818 struct sock *sk = sock->sk;
8945ee5e 1819
8945ee5e
CL		 1820 sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1821 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1822}
ed07536e 1823
8945ee5e
CL		 1824static inline void xs_reclassify_socket6(struct socket *sock)
1825{
1826 struct sock *sk = sock->sk;
ed07536e 1827
8945ee5e
CL		 1828 sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1829 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
ed07536e 1830}
6bc9638a
PE		 1831
1832static inline void xs_reclassify_socket(int family, struct socket *sock)
1833{
fafc4e1e 1834 if (WARN_ON_ONCE(!sock_allow_reclassification(sock->sk)))
1b7a1819
WAA		 1835 return;
1836
4232e863 1837 switch (family) {
176e21ee
CL		 1838 case AF_LOCAL:
1839 xs_reclassify_socketu(sock);
1840 break;
4232e863 1841 case AF_INET:
6bc9638a 1842 xs_reclassify_socket4(sock);
4232e863
CL		 1843 break;
1844 case AF_INET6:
6bc9638a 1845 xs_reclassify_socket6(sock);
4232e863
CL		 1846 break;
1847 }
6bc9638a 1848}
ed07536e 1849#else
6bc9638a
PE		 1850static inline void xs_reclassify_socket(int family, struct socket *sock)
1851{
1852}
ed07536e
PZ		 1853#endif
1854
93dc41bd
N		 1855static void xs_dummy_setup_socket(struct work_struct *work)
1856{
1857}
1858
6bc9638a 1859static struct socket *xs_create_sock(struct rpc_xprt *xprt,
4dda9c8a
TM		 1860 struct sock_xprt *transport, int family, int type,
1861 int protocol, bool reuseport)
22f79326
PE		 1862{
1863 struct socket *sock;
1864 int err;
1865
6bc9638a 1866 err = __sock_create(xprt->xprt_net, family, type, protocol, &sock, 1);
22f79326
PE		 1867 if (err < 0) {
1868 dprintk("RPC: can't create %d transport socket (%d).\n",
1869 protocol, -err);
1870 goto out;
1871 }
6bc9638a 1872 xs_reclassify_socket(family, sock);
22f79326 1873
4dda9c8a
TM		 1874 if (reuseport)
1875 xs_sock_set_reuseport(sock);
1876
4cea288a
BH		 1877 err = xs_bind(transport, sock);
1878 if (err) {
22f79326
PE		 1879 sock_release(sock);
1880 goto out;
1881 }
1882
1883 return sock;
1884out:
1885 return ERR_PTR(err);
1886}
1887
176e21ee
CL		 1888static int xs_local_finish_connecting(struct rpc_xprt *xprt,
1889 struct socket *sock)
1890{
1891 struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1892 xprt);
1893
1894 if (!transport->inet) {
1895 struct sock *sk = sock->sk;
1896
1897 write_lock_bh(&sk->sk_callback_lock);
1898
1899 xs_save_old_callbacks(transport, sk);
1900
1901 sk->sk_user_data = xprt;
a2648094 1902 sk->sk_data_ready = xs_data_ready;
176e21ee 1903 sk->sk_write_space = xs_udp_write_space;
b4411457 1904 sock_set_flag(sk, SOCK_FASYNC);
2118071d 1905 sk->sk_error_report = xs_error_report;
176e21ee
CL		 1906
1907 xprt_clear_connected(xprt);
1908
1909 /* Reset to new socket */
1910 transport->sock = sock;
1911 transport->inet = sk;
1912
1913 write_unlock_bh(&sk->sk_callback_lock);
1914 }
1915
ae053551 1916 xs_stream_start_connect(transport);
6c7a64e5 1917
176e21ee
CL		 1918 return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, 0);
1919}
1920
1921/**
1922 * xs_local_setup_socket - create AF_LOCAL socket, connect to a local endpoint
176e21ee 1923 * @transport: socket transport to connect
176e21ee 1924 */
dc107402 1925static int xs_local_setup_socket(struct sock_xprt *transport)
176e21ee 1926{
176e21ee
CL		 1927 struct rpc_xprt *xprt = &transport->xprt;
1928 struct socket *sock;
1929 int status = -EIO;
1930
176e21ee
CL		 1931 status = __sock_create(xprt->xprt_net, AF_LOCAL,
1932 SOCK_STREAM, 0, &sock, 1);
1933 if (status < 0) {
1934 dprintk("RPC: can't create AF_LOCAL "
1935 "transport socket (%d).\n", -status);
1936 goto out;
1937 }
d1358917 1938 xs_reclassify_socket(AF_LOCAL, sock);
176e21ee
CL		 1939
1940 dprintk("RPC: worker connecting xprt %p via AF_LOCAL to %s\n",
1941 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1942
1943 status = xs_local_finish_connecting(xprt, sock);
40b5ea0c 1944 trace_rpc_socket_connect(xprt, sock, status);
176e21ee
CL		 1945 switch (status) {
1946 case 0:
1947 dprintk("RPC: xprt %p connected to %s\n",
1948 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
3968a8a5
CL		 1949 xprt->stat.connect_count++;
1950 xprt->stat.connect_time += (long)jiffies -
1951 xprt->stat.connect_start;
176e21ee 1952 xprt_set_connected(xprt);
3601c4a9 1953 case -ENOBUFS:
176e21ee
CL		 1954 break;
1955 case -ENOENT:
1956 dprintk("RPC: xprt %p: socket %s does not exist\n",
1957 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1958 break;
4a20a988
TM		 1959 case -ECONNREFUSED:
1960 dprintk("RPC: xprt %p: connection refused for %s\n",
1961 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1962 break;
176e21ee
CL		 1963 default:
1964 printk(KERN_ERR "%s: unhandled error (%d) connecting to %s\n",
1965 __func__, -status,
1966 xprt->address_strings[RPC_DISPLAY_ADDR]);
1967 }
1968
1969out:
1970 xprt_clear_connecting(xprt);
1971 xprt_wake_pending_tasks(xprt, status);
dc107402
BF		 1972 return status;
1973}
1974
b6669737 1975static void xs_local_connect(struct rpc_xprt *xprt, struct rpc_task *task)
dc107402 1976{
dc107402
BF		 1977 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1978 int ret;
1979
1980 if (RPC_IS_ASYNC(task)) {
1981 /*
1982 * We want the AF_LOCAL connect to be resolved in the
1983 * filesystem namespace of the process making the rpc
1984 * call. Thus we connect synchronously.
1985 *
1986 * If we want to support asynchronous AF_LOCAL calls,
1987 * we'll need to figure out how to pass a namespace to
1988 * connect.
1989 */
1990 rpc_exit(task, -ENOTCONN);
1991 return;
1992 }
1993 ret = xs_local_setup_socket(transport);
1994 if (ret && !RPC_IS_SOFTCONN(task))
1995 msleep_interruptible(15000);
176e21ee
CL		 1996}
1997
3c87ef6e 1998#if IS_ENABLED(CONFIG_SUNRPC_SWAP)
d6e971d8
JL		 1999/*
2000 * Note that this should be called with XPRT_LOCKED held (or when we otherwise
2001 * know that we have exclusive access to the socket), to guard against
2002 * races with xs_reset_transport.
2003 */
a564b8f0
MG		 2004static void xs_set_memalloc(struct rpc_xprt *xprt)
2005{
2006 struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
2007 xprt);
2008
d6e971d8
JL		 2009 /*
2010 * If there's no sock, then we have nothing to set. The
2011 * reconnecting process will get it for us.
2012 */
2013 if (!transport->inet)
2014 return;
8e228133 2015 if (atomic_read(&xprt->swapper))
a564b8f0
MG		 2016 sk_set_memalloc(transport->inet);
2017}
2018
2019/**
d67fa4d8 2020 * xs_enable_swap - Tag this transport as being used for swap.
a564b8f0 2021 * @xprt: transport to tag
a564b8f0 2022 *
8e228133
JL		 2023 * Take a reference to this transport on behalf of the rpc_clnt, and
2024 * optionally mark it for swapping if it wasn't already.
a564b8f0 2025 */
d67fa4d8
JL		 2026static int
2027xs_enable_swap(struct rpc_xprt *xprt)
a564b8f0 2028{
d6e971d8 2029 struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
a564b8f0 2030
d6e971d8
JL		 2031 if (atomic_inc_return(&xprt->swapper) != 1)
2032 return 0;
2033 if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE))
2034 return -ERESTARTSYS;
2035 if (xs->inet)
2036 sk_set_memalloc(xs->inet);
2037 xprt_release_xprt(xprt, NULL);
8e228133
JL		 2038 return 0;
2039}
a564b8f0 2040
8e228133 2041/**
d67fa4d8 2042 * xs_disable_swap - Untag this transport as being used for swap.
8e228133
JL		 2043 * @xprt: transport to tag
2044 *
2045 * Drop a "swapper" reference to this xprt on behalf of the rpc_clnt. If the
2046 * swapper refcount goes to 0, untag the socket as a memalloc socket.
2047 */
d67fa4d8
JL		 2048static void
2049xs_disable_swap(struct rpc_xprt *xprt)
8e228133 2050{
d6e971d8 2051 struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
8e228133 2052
d6e971d8
JL		 2053 if (!atomic_dec_and_test(&xprt->swapper))
2054 return;
2055 if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE))
2056 return;
2057 if (xs->inet)
2058 sk_clear_memalloc(xs->inet);
2059 xprt_release_xprt(xprt, NULL);
a564b8f0 2060}
a564b8f0
MG		 2061#else
2062static void xs_set_memalloc(struct rpc_xprt *xprt)
2063{
2064}
d67fa4d8
JL		 2065
2066static int
2067xs_enable_swap(struct rpc_xprt *xprt)
2068{
2069 return -EINVAL;
2070}
2071
2072static void
2073xs_disable_swap(struct rpc_xprt *xprt)
2074{
2075}
a564b8f0
MG		 2076#endif
2077
16be2d20
CL		 2078static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2079{
2080 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2081
2082 if (!transport->inet) {
2083 struct sock *sk = sock->sk;
2084
2085 write_lock_bh(&sk->sk_callback_lock);
2086
2a9e1cfa
TM		 2087 xs_save_old_callbacks(transport, sk);
2088
16be2d20 2089 sk->sk_user_data = xprt;
f9b2ee71 2090 sk->sk_data_ready = xs_data_ready;
16be2d20 2091 sk->sk_write_space = xs_udp_write_space;
b4411457 2092 sock_set_flag(sk, SOCK_FASYNC);
16be2d20
CL		 2093
2094 xprt_set_connected(xprt);
2095
2096 /* Reset to new socket */
2097 transport->sock = sock;
2098 transport->inet = sk;
2099
a564b8f0
MG		 2100 xs_set_memalloc(xprt);
2101
16be2d20
CL		 2102 write_unlock_bh(&sk->sk_callback_lock);
2103 }
2104 xs_udp_do_set_buffer_size(xprt);
02910177
TM		 2105
2106 xprt->stat.connect_start = jiffies;
16be2d20
CL		 2107}
2108
8c14ff2a 2109static void xs_udp_setup_socket(struct work_struct *work)
a246b010 2110{
34161db6
TM		 2111 struct sock_xprt *transport =
2112 container_of(work, struct sock_xprt, connect_worker.work);
c8475461 2113 struct rpc_xprt *xprt = &transport->xprt;
d099b8af 2114 struct socket *sock;
b65c0310 2115 int status = -EIO;
9903cd1c 2116
8c14ff2a 2117 sock = xs_create_sock(xprt, transport,
4dda9c8a
TM		 2118 xs_addr(xprt)->sa_family, SOCK_DGRAM,
2119 IPPROTO_UDP, false);
b65c0310 2120 if (IS_ERR(sock))
b0d93ad5 2121 goto out;
68e220bd 2122
c740eff8
CL		 2123 dprintk("RPC: worker connecting xprt %p via %s to "
2124 "%s (port %s)\n", xprt,
2125 xprt->address_strings[RPC_DISPLAY_PROTO],
2126 xprt->address_strings[RPC_DISPLAY_ADDR],
2127 xprt->address_strings[RPC_DISPLAY_PORT]);
68e220bd
CL		 2128
2129 xs_udp_finish_connecting(xprt, sock);
40b5ea0c 2130 trace_rpc_socket_connect(xprt, sock, 0);
b0d93ad5
CL		 2131 status = 0;
2132out:
b0d93ad5 2133 xprt_clear_connecting(xprt);
cf76785d 2134 xprt_unlock_connect(xprt, transport);
7d1e8255 2135 xprt_wake_pending_tasks(xprt, status);
a246b010
CL		 2136}
2137
4876cc77
TM		 2138/**
2139 * xs_tcp_shutdown - gracefully shut down a TCP socket
2140 * @xprt: transport
2141 *
2142 * Initiates a graceful shutdown of the TCP socket by calling the
2143 * equivalent of shutdown(SHUT_RDWR);
2144 */
2145static void xs_tcp_shutdown(struct rpc_xprt *xprt)
2146{
2147 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2148 struct socket *sock = transport->sock;
9b30889c 2149 int skst = transport->inet ? transport->inet->sk_state : TCP_CLOSE;
4876cc77
TM		 2150
2151 if (sock == NULL)
2152 return;
9b30889c
TM		 2153 switch (skst) {
2154 default:
4876cc77
TM		 2155 kernel_sock_shutdown(sock, SHUT_RDWR);
2156 trace_rpc_socket_shutdown(xprt, sock);
9b30889c
TM		 2157 break;
2158 case TCP_CLOSE:
2159 case TCP_TIME_WAIT:
4876cc77 2160 xs_reset_transport(transport);
9b30889c 2161 }
4876cc77
TM		 2162}
2163
8d1b8c62
TM		 2164static void xs_tcp_set_socket_timeouts(struct rpc_xprt *xprt,
2165 struct socket *sock)
2166{
7196dbb0
TM		 2167 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2168 unsigned int keepidle;
2169 unsigned int keepcnt;
8d1b8c62
TM		 2170 unsigned int opt_on = 1;
2171 unsigned int timeo;
2172
7196dbb0
TM		 2173 spin_lock_bh(&xprt->transport_lock);
2174 keepidle = DIV_ROUND_UP(xprt->timeout->to_initval, HZ);
2175 keepcnt = xprt->timeout->to_retries + 1;
2176 timeo = jiffies_to_msecs(xprt->timeout->to_initval) *
2177 (xprt->timeout->to_retries + 1);
2178 clear_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state);
2179 spin_unlock_bh(&xprt->transport_lock);
2180
8d1b8c62
TM		 2181 /* TCP Keepalive options */
2182 kernel_setsockopt(sock, SOL_SOCKET, SO_KEEPALIVE,
2183 (char *)&opt_on, sizeof(opt_on));
2184 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPIDLE,
2185 (char *)&keepidle, sizeof(keepidle));
2186 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPINTVL,
2187 (char *)&keepidle, sizeof(keepidle));
2188 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPCNT,
2189 (char *)&keepcnt, sizeof(keepcnt));
2190
2191 /* TCP user timeout (see RFC5482) */
8d1b8c62
TM		 2192 kernel_setsockopt(sock, SOL_TCP, TCP_USER_TIMEOUT,
2193 (char *)&timeo, sizeof(timeo));
2194}
2195
7196dbb0
TM		 2196static void xs_tcp_set_connect_timeout(struct rpc_xprt *xprt,
2197 unsigned long connect_timeout,
2198 unsigned long reconnect_timeout)
2199{
2200 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2201 struct rpc_timeout to;
2202 unsigned long initval;
2203
2204 spin_lock_bh(&xprt->transport_lock);
2205 if (reconnect_timeout < xprt->max_reconnect_timeout)
2206 xprt->max_reconnect_timeout = reconnect_timeout;
2207 if (connect_timeout < xprt->connect_timeout) {
2208 memcpy(&to, xprt->timeout, sizeof(to));
2209 initval = DIV_ROUND_UP(connect_timeout, to.to_retries + 1);
2210 /* Arbitrary lower limit */
2211 if (initval < XS_TCP_INIT_REEST_TO << 1)
2212 initval = XS_TCP_INIT_REEST_TO << 1;
2213 to.to_initval = initval;
2214 to.to_maxval = initval;
2215 memcpy(&transport->tcp_timeout, &to,
2216 sizeof(transport->tcp_timeout));
2217 xprt->timeout = &transport->tcp_timeout;
2218 xprt->connect_timeout = connect_timeout;
2219 }
2220 set_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state);
2221 spin_unlock_bh(&xprt->transport_lock);
2222}
2223
16be2d20 2224static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
a246b010 2225{
16be2d20 2226 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
fe19a96b 2227 int ret = -ENOTCONN;
edb267a6 2228
ee0ac0c2 2229 if (!transport->inet) {
b0d93ad5 2230 struct sock *sk = sock->sk;
d88e4d82 2231 unsigned int addr_pref = IPV6_PREFER_SRC_PUBLIC;
7f260e85 2232
d88e4d82
N		 2233 /* Avoid temporary address, they are bad for long-lived
2234 * connections such as NFS mounts.
2235 * RFC4941, section 3.6 suggests that:
2236 * Individual applications, which have specific
2237 * knowledge about the normal duration of connections,
2238 * MAY override this as appropriate.
2239 */
2240 kernel_setsockopt(sock, SOL_IPV6, IPV6_ADDR_PREFERENCES,
2241 (char *)&addr_pref, sizeof(addr_pref));
2242
8d1b8c62 2243 xs_tcp_set_socket_timeouts(xprt, sock);
775f06ab 2244
b0d93ad5
CL		 2245 write_lock_bh(&sk->sk_callback_lock);
2246
2a9e1cfa
TM		 2247 xs_save_old_callbacks(transport, sk);
2248
b0d93ad5 2249 sk->sk_user_data = xprt;
5157b956 2250 sk->sk_data_ready = xs_data_ready;
b0d93ad5
CL		 2251 sk->sk_state_change = xs_tcp_state_change;
2252 sk->sk_write_space = xs_tcp_write_space;
b4411457 2253 sock_set_flag(sk, SOCK_FASYNC);
2118071d 2254 sk->sk_error_report = xs_error_report;
3167e12c
CL		 2255
2256 /* socket options */
3167e12c 2257 sock_reset_flag(sk, SOCK_LINGER);
3167e12c 2258 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
b0d93ad5
CL		 2259
2260 xprt_clear_connected(xprt);
2261
2262 /* Reset to new socket */
ee0ac0c2
CL		 2263 transport->sock = sock;
2264 transport->inet = sk;
b0d93ad5
CL		 2265
2266 write_unlock_bh(&sk->sk_callback_lock);
2267 }
2268
01d37c42 2269 if (!xprt_bound(xprt))
fe19a96b 2270 goto out;
01d37c42 2271
a564b8f0
MG		 2272 xs_set_memalloc(xprt);
2273
ae053551 2274 xs_stream_start_connect(transport);
e1806c7b 2275
b0d93ad5 2276 /* Tell the socket layer to start connecting... */
0fdea1e8 2277 set_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
fe19a96b
TM		 2278 ret = kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
2279 switch (ret) {
2280 case 0:
4dda9c8a 2281 xs_set_srcport(transport, sock);
e9d47639 2282 /* fall through */
fe19a96b
TM		 2283 case -EINPROGRESS:
2284 /* SYN_SENT! */
fe19a96b
TM		 2285 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2286 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1f4c17a0
TM		 2287 break;
2288 case -EADDRNOTAVAIL:
2289 /* Source port number is unavailable. Try a new one! */
2290 transport->srcport = 0;
fe19a96b
TM		 2291 }
2292out:
2293 return ret;
16be2d20
CL		 2294}
2295
9903cd1c 2296/**
b61d59ff 2297 * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint
acf0a39f 2298 * @work: queued work item
9903cd1c
CL		 2299 *
2300 * Invoked by a work queue tasklet.
a246b010 2301 */
cdd518d5 2302static void xs_tcp_setup_socket(struct work_struct *work)
a246b010 2303{
cdd518d5
PE		 2304 struct sock_xprt *transport =
2305 container_of(work, struct sock_xprt, connect_worker.work);
ee0ac0c2 2306 struct socket *sock = transport->sock;
a9f5f0f7 2307 struct rpc_xprt *xprt = &transport->xprt;
b61d59ff 2308 int status = -EIO;
a246b010 2309
ee0ac0c2 2310 if (!sock) {
cdd518d5 2311 sock = xs_create_sock(xprt, transport,
4dda9c8a
TM		 2312 xs_addr(xprt)->sa_family, SOCK_STREAM,
2313 IPPROTO_TCP, true);
b61d59ff
TM		 2314 if (IS_ERR(sock)) {
2315 status = PTR_ERR(sock);
3167e12c
CL		 2316 goto out;
2317 }
7d1e8255 2318 }
a246b010 2319
c740eff8
CL		 2320 dprintk("RPC: worker connecting xprt %p via %s to "
2321 "%s (port %s)\n", xprt,
2322 xprt->address_strings[RPC_DISPLAY_PROTO],
2323 xprt->address_strings[RPC_DISPLAY_ADDR],
2324 xprt->address_strings[RPC_DISPLAY_PORT]);
edb267a6 2325
16be2d20 2326 status = xs_tcp_finish_connecting(xprt, sock);
40b5ea0c 2327 trace_rpc_socket_connect(xprt, sock, status);
46121cf7
CL		 2328 dprintk("RPC: %p connect status %d connected %d sock state %d\n",
2329 xprt, -status, xprt_connected(xprt),
2330 sock->sk->sk_state);
2a491991 2331 switch (status) {
f75e6745
TM		 2332 default:
2333 printk("%s: connect returned unhandled error %d\n",
2334 __func__, status);
e9d47639 2335 /* fall through */
f75e6745
TM		 2336 case -EADDRNOTAVAIL:
2337 /* We're probably in TIME_WAIT. Get rid of existing socket,
2338 * and retry
2339 */
a519fc7a 2340 xs_tcp_force_close(xprt);
88b5ed73 2341 break;
2a491991
TM		 2342 case 0:
2343 case -EINPROGRESS:
2344 case -EALREADY:
718ba5b8 2345 xprt_unlock_connect(xprt, transport);
7d1e8255 2346 return;
9fcfe0c8
TM		 2347 case -EINVAL:
2348 /* Happens, for instance, if the user specified a link
2349 * local IPv6 address without a scope-id.
2350 */
3ed5e2a2
TM		 2351 case -ECONNREFUSED:
2352 case -ECONNRESET:
eb5b46fa 2353 case -ENETDOWN:
3ed5e2a2 2354 case -ENETUNREACH:
4ba161a7 2355 case -EHOSTUNREACH:
3913c78c 2356 case -EADDRINUSE:
3601c4a9 2357 case -ENOBUFS:
6ea44adc
N		 2358 /*
2359 * xs_tcp_force_close() wakes tasks with -EIO.
2360 * We need to wake them first to ensure the
2361 * correct error code.
2362 */
2363 xprt_wake_pending_tasks(xprt, status);
4efdd92c 2364 xs_tcp_force_close(xprt);
9fcfe0c8 2365 goto out;
a246b010 2366 }
2a491991 2367 status = -EAGAIN;
a246b010 2368out:
68e220bd 2369 xprt_clear_connecting(xprt);
cf76785d 2370 xprt_unlock_connect(xprt, transport);
7d1e8255 2371 xprt_wake_pending_tasks(xprt, status);
68e220bd 2372}
b0d93ad5 2373
02910177
TM		 2374static unsigned long xs_reconnect_delay(const struct rpc_xprt *xprt)
2375{
2376 unsigned long start, now = jiffies;
2377
2378 start = xprt->stat.connect_start + xprt->reestablish_timeout;
2379 if (time_after(start, now))
2380 return start - now;
2381 return 0;
2382}
2383
3851f1cd
TM		 2384static void xs_reconnect_backoff(struct rpc_xprt *xprt)
2385{
2386 xprt->reestablish_timeout <<= 1;
2387 if (xprt->reestablish_timeout > xprt->max_reconnect_timeout)
2388 xprt->reestablish_timeout = xprt->max_reconnect_timeout;
2389 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2390 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2391}
2392
9903cd1c
CL		 2393/**
2394 * xs_connect - connect a socket to a remote endpoint
1b092092 2395 * @xprt: pointer to transport structure
9903cd1c
CL		 2396 * @task: address of RPC task that manages state of connect request
2397 *
2398 * TCP: If the remote end dropped the connection, delay reconnecting.
03bf4b70
CL		 2399 *
2400 * UDP socket connects are synchronous, but we use a work queue anyway
2401 * to guarantee that even unprivileged user processes can set up a
2402 * socket on a privileged port.
2403 *
2404 * If a UDP socket connect fails, the delay behavior here prevents
2405 * retry floods (hard mounts).
9903cd1c 2406 */
1b092092 2407static void xs_connect(struct rpc_xprt *xprt, struct rpc_task *task)
a246b010 2408{
ee0ac0c2 2409 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
02910177 2410 unsigned long delay = 0;
a246b010 2411
718ba5b8
TM		 2412 WARN_ON_ONCE(!xprt_lock_connect(xprt, task, transport));
2413
99b1a4c3 2414 if (transport->sock != NULL) {
46121cf7
CL		 2415 dprintk("RPC: xs_connect delayed xprt %p for %lu "
2416 "seconds\n",
03bf4b70 2417 xprt, xprt->reestablish_timeout / HZ);
99b1a4c3
TM		 2418
2419 /* Start by resetting any existing state */
2420 xs_reset_transport(transport);
2421
02910177 2422 delay = xs_reconnect_delay(xprt);
3851f1cd 2423 xs_reconnect_backoff(xprt);
02910177 2424
02910177 2425 } else
46121cf7 2426 dprintk("RPC: xs_connect scheduled xprt %p\n", xprt);
02910177
TM		 2427
2428 queue_delayed_work(xprtiod_workqueue,
2429 &transport->connect_worker,
2430 delay);
a246b010
CL		 2431}
2432
176e21ee
CL		 2433/**
2434 * xs_local_print_stats - display AF_LOCAL socket-specifc stats
2435 * @xprt: rpc_xprt struct containing statistics
2436 * @seq: output file
2437 *
2438 */
2439static void xs_local_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2440{
2441 long idle_time = 0;
2442
2443 if (xprt_connected(xprt))
2444 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2445
2446 seq_printf(seq, "\txprt:\tlocal %lu %lu %lu %ld %lu %lu %lu "
15a45206 2447 "%llu %llu %lu %llu %llu\n",
176e21ee
CL		 2448 xprt->stat.bind_count,
2449 xprt->stat.connect_count,
8440a886 2450 xprt->stat.connect_time / HZ,
176e21ee
CL		 2451 idle_time,
2452 xprt->stat.sends,
2453 xprt->stat.recvs,
2454 xprt->stat.bad_xids,
2455 xprt->stat.req_u,
15a45206
AA		 2456 xprt->stat.bklog_u,
2457 xprt->stat.max_slots,
2458 xprt->stat.sending_u,
2459 xprt->stat.pending_u);
176e21ee
CL		 2460}
2461
262ca07d
CL		 2462/**
2463 * xs_udp_print_stats - display UDP socket-specifc stats
2464 * @xprt: rpc_xprt struct containing statistics
2465 * @seq: output file
2466 *
2467 */
2468static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2469{
c8475461
CL		 2470 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2471
15a45206
AA		 2472 seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %llu %llu "
2473 "%lu %llu %llu\n",
fbfffbd5 2474 transport->srcport,
262ca07d
CL		 2475 xprt->stat.bind_count,
2476 xprt->stat.sends,
2477 xprt->stat.recvs,
2478 xprt->stat.bad_xids,
2479 xprt->stat.req_u,
15a45206
AA		 2480 xprt->stat.bklog_u,
2481 xprt->stat.max_slots,
2482 xprt->stat.sending_u,
2483 xprt->stat.pending_u);
262ca07d
CL		 2484}
2485
2486/**
2487 * xs_tcp_print_stats - display TCP socket-specifc stats
2488 * @xprt: rpc_xprt struct containing statistics
2489 * @seq: output file
2490 *
2491 */
2492static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2493{
c8475461 2494 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
262ca07d
CL		 2495 long idle_time = 0;
2496
2497 if (xprt_connected(xprt))
2498 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2499
15a45206
AA		 2500 seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu "
2501 "%llu %llu %lu %llu %llu\n",
fbfffbd5 2502 transport->srcport,
262ca07d
CL		 2503 xprt->stat.bind_count,
2504 xprt->stat.connect_count,
8440a886 2505 xprt->stat.connect_time / HZ,
262ca07d
CL		 2506 idle_time,
2507 xprt->stat.sends,
2508 xprt->stat.recvs,
2509 xprt->stat.bad_xids,
2510 xprt->stat.req_u,
15a45206
AA		 2511 xprt->stat.bklog_u,
2512 xprt->stat.max_slots,
2513 xprt->stat.sending_u,
2514 xprt->stat.pending_u);
262ca07d
CL		 2515}
2516
4cfc7e60
RI		 2517/*
2518 * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason
2519 * we allocate pages instead doing a kmalloc like rpc_malloc is because we want
2520 * to use the server side send routines.
2521 */
5fe6eaa1 2522static int bc_malloc(struct rpc_task *task)
4cfc7e60 2523{
5fe6eaa1
CL		 2524 struct rpc_rqst *rqst = task->tk_rqstp;
2525 size_t size = rqst->rq_callsize;
4cfc7e60
RI		 2526 struct page *page;
2527 struct rpc_buffer *buf;
2528
5fe6eaa1
CL		 2529 if (size > PAGE_SIZE - sizeof(struct rpc_buffer)) {
2530 WARN_ONCE(1, "xprtsock: large bc buffer request (size %zu)\n",
2531 size);
2532 return -EINVAL;
2533 }
4cfc7e60 2534
b8a13d03 2535 page = alloc_page(GFP_KERNEL);
4cfc7e60 2536 if (!page)
5fe6eaa1 2537 return -ENOMEM;
4cfc7e60
RI		 2538
2539 buf = page_address(page);
2540 buf->len = PAGE_SIZE;
2541
5fe6eaa1 2542 rqst->rq_buffer = buf->data;
18e601d6 2543 rqst->rq_rbuffer = (char *)rqst->rq_buffer + rqst->rq_callsize;
5fe6eaa1 2544 return 0;
4cfc7e60
RI		 2545}
2546
2547/*
2548 * Free the space allocated in the bc_alloc routine
2549 */
3435c74a 2550static void bc_free(struct rpc_task *task)
4cfc7e60 2551{
3435c74a 2552 void *buffer = task->tk_rqstp->rq_buffer;
4cfc7e60
RI		 2553 struct rpc_buffer *buf;
2554
4cfc7e60
RI		 2555 buf = container_of(buffer, struct rpc_buffer, data);
2556 free_page((unsigned long)buf);
2557}
2558
2559/*
2560 * Use the svc_sock to send the callback. Must be called with svsk->sk_mutex
2561 * held. Borrows heavily from svc_tcp_sendto and xs_tcp_send_request.
2562 */
2563static int bc_sendto(struct rpc_rqst *req)
2564{
2565 int len;
2566 struct xdr_buf *xbufp = &req->rq_snd_buf;
2567 struct rpc_xprt *xprt = req->rq_xprt;
2568 struct sock_xprt *transport =
2569 container_of(xprt, struct sock_xprt, xprt);
2570 struct socket *sock = transport->sock;
2571 unsigned long headoff;
2572 unsigned long tailoff;
2573
61677eee 2574 xs_encode_stream_record_marker(xbufp);
4cfc7e60
RI		 2575
2576 tailoff = (unsigned long)xbufp->tail[0].iov_base & ~PAGE_MASK;
2577 headoff = (unsigned long)xbufp->head[0].iov_base & ~PAGE_MASK;
2578 len = svc_send_common(sock, xbufp,
2579 virt_to_page(xbufp->head[0].iov_base), headoff,
2580 xbufp->tail[0].iov_base, tailoff);
2581
2582 if (len != xbufp->len) {
2583 printk(KERN_NOTICE "Error sending entire callback!\n");
2584 len = -EAGAIN;
2585 }
2586
2587 return len;
2588}
2589
2590/*
2591 * The send routine. Borrows from svc_send
2592 */
adfa7144 2593static int bc_send_request(struct rpc_rqst *req)
4cfc7e60 2594{
4cfc7e60 2595 struct svc_xprt *xprt;
7fc56136 2596 int len;
4cfc7e60
RI		 2597
2598 dprintk("sending request with xid: %08x\n", ntohl(req->rq_xid));
2599 /*
2600 * Get the server socket associated with this callback xprt
2601 */
2602 xprt = req->rq_xprt->bc_xprt;
4cfc7e60
RI		 2603
2604 /*
2605 * Grab the mutex to serialize data as the connection is shared
2606 * with the fore channel
2607 */
c544577d 2608 mutex_lock(&xprt->xpt_mutex);
4cfc7e60
RI		 2609 if (test_bit(XPT_DEAD, &xprt->xpt_flags))
2610 len = -ENOTCONN;
2611 else
2612 len = bc_sendto(req);
2613 mutex_unlock(&xprt->xpt_mutex);
2614
2615 if (len > 0)
2616 len = 0;
2617
2618 return len;
2619}
2620
2621/*
2622 * The close routine. Since this is client initiated, we do nothing
2623 */
2624
2625static void bc_close(struct rpc_xprt *xprt)
2626{
4cfc7e60
RI		 2627}
2628
2629/*
2630 * The xprt destroy routine. Again, because this connection is client
2631 * initiated, we do nothing
2632 */
2633
2634static void bc_destroy(struct rpc_xprt *xprt)
2635{
47f72efa
KM		 2636 dprintk("RPC: bc_destroy xprt %p\n", xprt);
2637
2638 xs_xprt_free(xprt);
2639 module_put(THIS_MODULE);
4cfc7e60
RI		 2640}
2641
d31ae254 2642static const struct rpc_xprt_ops xs_local_ops = {
176e21ee 2643 .reserve_xprt = xprt_reserve_xprt,
4cd34e7c 2644 .release_xprt = xprt_release_xprt,
f39c1bfb 2645 .alloc_slot = xprt_alloc_slot,
a9cde23a 2646 .free_slot = xprt_free_slot,
176e21ee
CL		 2647 .rpcbind = xs_local_rpcbind,
2648 .set_port = xs_local_set_port,
dc107402 2649 .connect = xs_local_connect,
176e21ee
CL		 2650 .buf_alloc = rpc_malloc,
2651 .buf_free = rpc_free,
550aebfe 2652 .prepare_request = xs_stream_prepare_request,
176e21ee
CL		 2653 .send_request = xs_local_send_request,
2654 .set_retrans_timeout = xprt_set_retrans_timeout_def,
2655 .close = xs_close,
a1311d87 2656 .destroy = xs_destroy,
176e21ee 2657 .print_stats = xs_local_print_stats,
d67fa4d8
JL		 2658 .enable_swap = xs_enable_swap,
2659 .disable_swap = xs_disable_swap,
176e21ee
CL		 2660};
2661
d31ae254 2662static const struct rpc_xprt_ops xs_udp_ops = {
43118c29 2663 .set_buffer_size = xs_udp_set_buffer_size,
12a80469 2664 .reserve_xprt = xprt_reserve_xprt_cong,
49e9a890 2665 .release_xprt = xprt_release_xprt_cong,
f39c1bfb 2666 .alloc_slot = xprt_alloc_slot,
a9cde23a 2667 .free_slot = xprt_free_slot,
45160d62 2668 .rpcbind = rpcb_getport_async,
92200412 2669 .set_port = xs_set_port,
262965f5 2670 .connect = xs_connect,
02107148
CL		 2671 .buf_alloc = rpc_malloc,
2672 .buf_free = rpc_free,
262965f5 2673 .send_request = xs_udp_send_request,
fe3aca29 2674 .set_retrans_timeout = xprt_set_retrans_timeout_rtt,
46c0ee8b 2675 .timer = xs_udp_timer,
a58dd398 2676 .release_request = xprt_release_rqst_cong,
262965f5
CL		 2677 .close = xs_close,
2678 .destroy = xs_destroy,
262ca07d 2679 .print_stats = xs_udp_print_stats,
d67fa4d8
JL		 2680 .enable_swap = xs_enable_swap,
2681 .disable_swap = xs_disable_swap,
4a068258 2682 .inject_disconnect = xs_inject_disconnect,
262965f5
CL		 2683};
2684
d31ae254 2685static const struct rpc_xprt_ops xs_tcp_ops = {
12a80469 2686 .reserve_xprt = xprt_reserve_xprt,
4cd34e7c 2687 .release_xprt = xprt_release_xprt,
36bd7de9 2688 .alloc_slot = xprt_alloc_slot,
a9cde23a 2689 .free_slot = xprt_free_slot,
45160d62 2690 .rpcbind = rpcb_getport_async,
92200412 2691 .set_port = xs_set_port,
0b9e7943 2692 .connect = xs_connect,
02107148
CL		 2693 .buf_alloc = rpc_malloc,
2694 .buf_free = rpc_free,
277e4ab7 2695 .prepare_request = xs_stream_prepare_request,
262965f5 2696 .send_request = xs_tcp_send_request,
fe3aca29 2697 .set_retrans_timeout = xprt_set_retrans_timeout_def,
c627d31b 2698 .close = xs_tcp_shutdown,
9903cd1c 2699 .destroy = xs_destroy,
7196dbb0 2700 .set_connect_timeout = xs_tcp_set_connect_timeout,
262ca07d 2701 .print_stats = xs_tcp_print_stats,
d67fa4d8
JL		 2702 .enable_swap = xs_enable_swap,
2703 .disable_swap = xs_disable_swap,
4a068258 2704 .inject_disconnect = xs_inject_disconnect,
42e5c3e2
CL		 2705#ifdef CONFIG_SUNRPC_BACKCHANNEL
2706 .bc_setup = xprt_setup_bc,
6b26cc8c 2707 .bc_maxpayload = xs_tcp_bc_maxpayload,
42e5c3e2
CL		 2708 .bc_free_rqst = xprt_free_bc_rqst,
2709 .bc_destroy = xprt_destroy_bc,
2710#endif
a246b010
CL		 2711};
2712
4cfc7e60
RI		 2713/*
2714 * The rpc_xprt_ops for the server backchannel
2715 */
2716
d31ae254 2717static const struct rpc_xprt_ops bc_tcp_ops = {
4cfc7e60
RI		 2718 .reserve_xprt = xprt_reserve_xprt,
2719 .release_xprt = xprt_release_xprt,
84e28a30 2720 .alloc_slot = xprt_alloc_slot,
a9cde23a 2721 .free_slot = xprt_free_slot,
4cfc7e60
RI		 2722 .buf_alloc = bc_malloc,
2723 .buf_free = bc_free,
2724 .send_request = bc_send_request,
2725 .set_retrans_timeout = xprt_set_retrans_timeout_def,
2726 .close = bc_close,
2727 .destroy = bc_destroy,
2728 .print_stats = xs_tcp_print_stats,
d67fa4d8
JL		 2729 .enable_swap = xs_enable_swap,
2730 .disable_swap = xs_disable_swap,
4a068258 2731 .inject_disconnect = xs_inject_disconnect,
4cfc7e60
RI		 2732};
2733
92476850
CL		 2734static int xs_init_anyaddr(const int family, struct sockaddr *sap)
2735{
2736 static const struct sockaddr_in sin = {
2737 .sin_family = AF_INET,
2738 .sin_addr.s_addr = htonl(INADDR_ANY),
2739 };
2740 static const struct sockaddr_in6 sin6 = {
2741 .sin6_family = AF_INET6,
2742 .sin6_addr = IN6ADDR_ANY_INIT,
2743 };
2744
2745 switch (family) {
176e21ee
CL		 2746 case AF_LOCAL:
2747 break;
92476850
CL		 2748 case AF_INET:
2749 memcpy(sap, &sin, sizeof(sin));
2750 break;
2751 case AF_INET6:
2752 memcpy(sap, &sin6, sizeof(sin6));
2753 break;
2754 default:
2755 dprintk("RPC: %s: Bad address family\n", __func__);
2756 return -EAFNOSUPPORT;
2757 }
2758 return 0;
2759}
2760
3c341b0b 2761static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
d9ba131d
TM		 2762 unsigned int slot_table_size,
2763 unsigned int max_slot_table_size)
c8541ecd
CL		 2764{
2765 struct rpc_xprt *xprt;
ffc2e518 2766 struct sock_xprt *new;
c8541ecd 2767
96802a09 2768 if (args->addrlen > sizeof(xprt->addr)) {
46121cf7 2769 dprintk("RPC: xs_setup_xprt: address too large\n");
c8541ecd
CL		 2770 return ERR_PTR(-EBADF);
2771 }
2772
d9ba131d
TM		 2773 xprt = xprt_alloc(args->net, sizeof(*new), slot_table_size,
2774 max_slot_table_size);
bd1722d4 2775 if (xprt == NULL) {
46121cf7
CL		 2776 dprintk("RPC: xs_setup_xprt: couldn't allocate "
2777 "rpc_xprt\n");
c8541ecd
CL		 2778 return ERR_PTR(-ENOMEM);
2779 }
c8541ecd 2780
bd1722d4 2781 new = container_of(xprt, struct sock_xprt, xprt);
edc1b01c 2782 mutex_init(&new->recv_mutex);
96802a09
FM		 2783 memcpy(&xprt->addr, args->dstaddr, args->addrlen);
2784 xprt->addrlen = args->addrlen;
d3bc9a1d 2785 if (args->srcaddr)
fbfffbd5 2786 memcpy(&new->srcaddr, args->srcaddr, args->addrlen);
92476850
CL		 2787 else {
2788 int err;
2789 err = xs_init_anyaddr(args->dstaddr->sa_family,
2790 (struct sockaddr *)&new->srcaddr);
2aa13531
SK		 2791 if (err != 0) {
2792 xprt_free(xprt);
92476850 2793 return ERR_PTR(err);
2aa13531 2794 }
92476850 2795 }
c8541ecd
CL		 2796
2797 return xprt;
2798}
2799
176e21ee
CL		 2800static const struct rpc_timeout xs_local_default_timeout = {
2801 .to_initval = 10 * HZ,
2802 .to_maxval = 10 * HZ,
2803 .to_retries = 2,
2804};
2805
2806/**
2807 * xs_setup_local - Set up transport to use an AF_LOCAL socket
2808 * @args: rpc transport creation arguments
2809 *
2810 * AF_LOCAL is a "tpi_cots_ord" transport, just like TCP
2811 */
2812static struct rpc_xprt *xs_setup_local(struct xprt_create *args)
2813{
2814 struct sockaddr_un *sun = (struct sockaddr_un *)args->dstaddr;
2815 struct sock_xprt *transport;
2816 struct rpc_xprt *xprt;
2817 struct rpc_xprt *ret;
2818
d9ba131d
TM		 2819 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2820 xprt_max_tcp_slot_table_entries);
176e21ee
CL		 2821 if (IS_ERR(xprt))
2822 return xprt;
2823 transport = container_of(xprt, struct sock_xprt, xprt);
2824
2825 xprt->prot = 0;
2826 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2827 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2828
2829 xprt->bind_timeout = XS_BIND_TO;
2830 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2831 xprt->idle_timeout = XS_IDLE_DISC_TO;
2832
2833 xprt->ops = &xs_local_ops;
2834 xprt->timeout = &xs_local_default_timeout;
2835
550aebfe
TM		 2836 INIT_WORK(&transport->recv_worker, xs_stream_data_receive_workfn);
2837 INIT_DELAYED_WORK(&transport->connect_worker, xs_dummy_setup_socket);
93dc41bd 2838
176e21ee
CL		 2839 switch (sun->sun_family) {
2840 case AF_LOCAL:
2841 if (sun->sun_path[0] != '/') {
2842 dprintk("RPC: bad AF_LOCAL address: %s\n",
2843 sun->sun_path);
2844 ret = ERR_PTR(-EINVAL);
2845 goto out_err;
2846 }
2847 xprt_set_bound(xprt);
176e21ee 2848 xs_format_peer_addresses(xprt, "local", RPCBIND_NETID_LOCAL);
7073ea87
BF		 2849 ret = ERR_PTR(xs_local_setup_socket(transport));
2850 if (ret)
2851 goto out_err;
176e21ee
CL		 2852 break;
2853 default:
2854 ret = ERR_PTR(-EAFNOSUPPORT);
2855 goto out_err;
2856 }
2857
2858 dprintk("RPC: set up xprt to %s via AF_LOCAL\n",
2859 xprt->address_strings[RPC_DISPLAY_ADDR]);
2860
2861 if (try_module_get(THIS_MODULE))
2862 return xprt;
2863 ret = ERR_PTR(-EINVAL);
2864out_err:
315f3812 2865 xs_xprt_free(xprt);
176e21ee
CL		 2866 return ret;
2867}
2868
2881ae74
TM		 2869static const struct rpc_timeout xs_udp_default_timeout = {
2870 .to_initval = 5 * HZ,
2871 .to_maxval = 30 * HZ,
2872 .to_increment = 5 * HZ,
2873 .to_retries = 5,
2874};
2875
9903cd1c
CL		 2876/**
2877 * xs_setup_udp - Set up transport to use a UDP socket
96802a09 2878 * @args: rpc transport creation arguments
9903cd1c
CL		 2879 *
2880 */
483066d6 2881static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
a246b010 2882{
8f9d5b1a 2883 struct sockaddr *addr = args->dstaddr;
c8541ecd 2884 struct rpc_xprt *xprt;
c8475461 2885 struct sock_xprt *transport;
0a68b0be 2886 struct rpc_xprt *ret;
a246b010 2887
d9ba131d
TM		 2888 xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries,
2889 xprt_udp_slot_table_entries);
c8541ecd
CL		 2890 if (IS_ERR(xprt))
2891 return xprt;
c8475461 2892 transport = container_of(xprt, struct sock_xprt, xprt);
a246b010 2893
ec739ef0 2894 xprt->prot = IPPROTO_UDP;
808012fb 2895 xprt->tsh_size = 0;
a246b010
CL		 2896 /* XXX: header size can vary due to auth type, IPv6, etc. */
2897 xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
2898
03bf4b70 2899 xprt->bind_timeout = XS_BIND_TO;
03bf4b70
CL		 2900 xprt->reestablish_timeout = XS_UDP_REEST_TO;
2901 xprt->idle_timeout = XS_IDLE_DISC_TO;
a246b010 2902
262965f5 2903 xprt->ops = &xs_udp_ops;
a246b010 2904
ba7392bb 2905 xprt->timeout = &xs_udp_default_timeout;
a246b010 2906
f9b2ee71 2907 INIT_WORK(&transport->recv_worker, xs_udp_data_receive_workfn);
edc1b01c
TM		 2908 INIT_DELAYED_WORK(&transport->connect_worker, xs_udp_setup_socket);
2909
8f9d5b1a
CL		 2910 switch (addr->sa_family) {
2911 case AF_INET:
2912 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2913 xprt_set_bound(xprt);
2914
9dc3b095 2915 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
8f9d5b1a
CL		 2916 break;
2917 case AF_INET6:
2918 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2919 xprt_set_bound(xprt);
2920
9dc3b095 2921 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
8f9d5b1a
CL		 2922 break;
2923 default:
0a68b0be
BF		 2924 ret = ERR_PTR(-EAFNOSUPPORT);
2925 goto out_err;
8f9d5b1a
CL		 2926 }
2927
c740eff8
CL		 2928 if (xprt_bound(xprt))
2929 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2930 xprt->address_strings[RPC_DISPLAY_ADDR],
2931 xprt->address_strings[RPC_DISPLAY_PORT],
2932 xprt->address_strings[RPC_DISPLAY_PROTO]);
2933 else
2934 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
2935 xprt->address_strings[RPC_DISPLAY_ADDR],
2936 xprt->address_strings[RPC_DISPLAY_PROTO]);
edb267a6 2937
bc25571e
TT		 2938 if (try_module_get(THIS_MODULE))
2939 return xprt;
0a68b0be
BF		 2940 ret = ERR_PTR(-EINVAL);
2941out_err:
315f3812 2942 xs_xprt_free(xprt);
0a68b0be 2943 return ret;
a246b010
CL		 2944}
2945
2881ae74
TM		 2946static const struct rpc_timeout xs_tcp_default_timeout = {
2947 .to_initval = 60 * HZ,
2948 .to_maxval = 60 * HZ,
2949 .to_retries = 2,
2950};
2951
9903cd1c
CL		 2952/**
2953 * xs_setup_tcp - Set up transport to use a TCP socket
96802a09 2954 * @args: rpc transport creation arguments
9903cd1c
CL		 2955 *
2956 */
483066d6 2957static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
a246b010 2958{
8f9d5b1a 2959 struct sockaddr *addr = args->dstaddr;
c8541ecd 2960 struct rpc_xprt *xprt;
c8475461 2961 struct sock_xprt *transport;
0a68b0be 2962 struct rpc_xprt *ret;
b7993ceb
TM		 2963 unsigned int max_slot_table_size = xprt_max_tcp_slot_table_entries;
2964
2965 if (args->flags & XPRT_CREATE_INFINITE_SLOTS)
2966 max_slot_table_size = RPC_MAX_SLOT_TABLE_LIMIT;
a246b010 2967
d9ba131d 2968 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
b7993ceb 2969 max_slot_table_size);
c8541ecd
CL		 2970 if (IS_ERR(xprt))
2971 return xprt;
c8475461 2972 transport = container_of(xprt, struct sock_xprt, xprt);
a246b010 2973
ec739ef0 2974 xprt->prot = IPPROTO_TCP;
808012fb 2975 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
808012fb 2976 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
a246b010 2977
03bf4b70 2978 xprt->bind_timeout = XS_BIND_TO;
03bf4b70
CL		 2979 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2980 xprt->idle_timeout = XS_IDLE_DISC_TO;
a246b010 2981
262965f5 2982 xprt->ops = &xs_tcp_ops;
ba7392bb 2983 xprt->timeout = &xs_tcp_default_timeout;
a246b010 2984
3851f1cd 2985 xprt->max_reconnect_timeout = xprt->timeout->to_maxval;
7196dbb0
TM		 2986 xprt->connect_timeout = xprt->timeout->to_initval *
2987 (xprt->timeout->to_retries + 1);
3851f1cd 2988
c50b8ee0 2989 INIT_WORK(&transport->recv_worker, xs_stream_data_receive_workfn);
edc1b01c
TM		 2990 INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_setup_socket);
2991
8f9d5b1a
CL		 2992 switch (addr->sa_family) {
2993 case AF_INET:
2994 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2995 xprt_set_bound(xprt);
2996
9dc3b095 2997 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
8f9d5b1a
CL		 2998 break;
2999 case AF_INET6:
3000 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
3001 xprt_set_bound(xprt);
3002
9dc3b095 3003 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
8f9d5b1a
CL		 3004 break;
3005 default:
0a68b0be
BF		 3006 ret = ERR_PTR(-EAFNOSUPPORT);
3007 goto out_err;
8f9d5b1a
CL		 3008 }
3009
c740eff8
CL		 3010 if (xprt_bound(xprt))
3011 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
3012 xprt->address_strings[RPC_DISPLAY_ADDR],
3013 xprt->address_strings[RPC_DISPLAY_PORT],
3014 xprt->address_strings[RPC_DISPLAY_PROTO]);
3015 else
3016 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
3017 xprt->address_strings[RPC_DISPLAY_ADDR],
3018 xprt->address_strings[RPC_DISPLAY_PROTO]);
3019
bc25571e
TT		 3020 if (try_module_get(THIS_MODULE))
3021 return xprt;
0a68b0be
BF		 3022 ret = ERR_PTR(-EINVAL);
3023out_err:
315f3812 3024 xs_xprt_free(xprt);
0a68b0be 3025 return ret;
a246b010 3026}
282b32e1 3027
f300baba
AB		 3028/**
3029 * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket
3030 * @args: rpc transport creation arguments
3031 *
3032 */
3033static struct rpc_xprt *xs_setup_bc_tcp(struct xprt_create *args)
3034{
3035 struct sockaddr *addr = args->dstaddr;
3036 struct rpc_xprt *xprt;
3037 struct sock_xprt *transport;
3038 struct svc_sock *bc_sock;
0a68b0be 3039 struct rpc_xprt *ret;
f300baba 3040
d9ba131d
TM		 3041 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
3042 xprt_tcp_slot_table_entries);
f300baba
AB		 3043 if (IS_ERR(xprt))
3044 return xprt;
3045 transport = container_of(xprt, struct sock_xprt, xprt);
3046
3047 xprt->prot = IPPROTO_TCP;
3048 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
3049 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
3050 xprt->timeout = &xs_tcp_default_timeout;
3051
3052 /* backchannel */
3053 xprt_set_bound(xprt);
3054 xprt->bind_timeout = 0;
f300baba
AB		 3055 xprt->reestablish_timeout = 0;
3056 xprt->idle_timeout = 0;
3057
f300baba
AB		 3058 xprt->ops = &bc_tcp_ops;
3059
3060 switch (addr->sa_family) {
3061 case AF_INET:
3062 xs_format_peer_addresses(xprt, "tcp",
3063 RPCBIND_NETID_TCP);
3064 break;
3065 case AF_INET6:
3066 xs_format_peer_addresses(xprt, "tcp",
3067 RPCBIND_NETID_TCP6);
3068 break;
3069 default:
0a68b0be
BF		 3070 ret = ERR_PTR(-EAFNOSUPPORT);
3071 goto out_err;
f300baba
AB		 3072 }
3073
50fa0d40
PE		 3074 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
3075 xprt->address_strings[RPC_DISPLAY_ADDR],
3076 xprt->address_strings[RPC_DISPLAY_PORT],
3077 xprt->address_strings[RPC_DISPLAY_PROTO]);
f300baba 3078
99de8ea9
BF		 3079 /*
3080 * Once we've associated a backchannel xprt with a connection,
28303ca3
WM		 3081 * we want to keep it around as long as the connection lasts,
3082 * in case we need to start using it for a backchannel again;
3083 * this reference won't be dropped until bc_xprt is destroyed.
99de8ea9
BF		 3084 */
3085 xprt_get(xprt);
3086 args->bc_xprt->xpt_bc_xprt = xprt;
3087 xprt->bc_xprt = args->bc_xprt;
3088 bc_sock = container_of(args->bc_xprt, struct svc_sock, sk_xprt);
3089 transport->sock = bc_sock->sk_sock;
3090 transport->inet = bc_sock->sk_sk;
3091
f300baba
AB		 3092 /*
3093 * Since we don't want connections for the backchannel, we set
3094 * the xprt status to connected
3095 */
3096 xprt_set_connected(xprt);
3097
f300baba
AB		 3098 if (try_module_get(THIS_MODULE))
3099 return xprt;
642aab58
KM		 3100
3101 args->bc_xprt->xpt_bc_xprt = NULL;
39a9beab 3102 args->bc_xprt->xpt_bc_xps = NULL;
99de8ea9 3103 xprt_put(xprt);
0a68b0be
BF		 3104 ret = ERR_PTR(-EINVAL);
3105out_err:
315f3812 3106 xs_xprt_free(xprt);
0a68b0be 3107 return ret;
f300baba
AB		 3108}
3109
176e21ee
CL		 3110static struct xprt_class xs_local_transport = {
3111 .list = LIST_HEAD_INIT(xs_local_transport.list),
3112 .name = "named UNIX socket",
3113 .owner = THIS_MODULE,
3114 .ident = XPRT_TRANSPORT_LOCAL,
3115 .setup = xs_setup_local,
3116};
3117
bc25571e
TT		 3118static struct xprt_class xs_udp_transport = {
3119 .list = LIST_HEAD_INIT(xs_udp_transport.list),
3120 .name = "udp",
3121 .owner = THIS_MODULE,
f300baba 3122 .ident = XPRT_TRANSPORT_UDP,
bc25571e
TT		 3123 .setup = xs_setup_udp,
3124};
3125
3126static struct xprt_class xs_tcp_transport = {
3127 .list = LIST_HEAD_INIT(xs_tcp_transport.list),
3128 .name = "tcp",
3129 .owner = THIS_MODULE,
f300baba 3130 .ident = XPRT_TRANSPORT_TCP,
bc25571e
TT		 3131 .setup = xs_setup_tcp,
3132};
3133
f300baba
AB		 3134static struct xprt_class xs_bc_tcp_transport = {
3135 .list = LIST_HEAD_INIT(xs_bc_tcp_transport.list),
3136 .name = "tcp NFSv4.1 backchannel",
3137 .owner = THIS_MODULE,
3138 .ident = XPRT_TRANSPORT_BC_TCP,
3139 .setup = xs_setup_bc_tcp,
3140};
3141
282b32e1 3142/**
bc25571e 3143 * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
282b32e1
CL		 3144 *
3145 */
3146int init_socket_xprt(void)
3147{
2b1bec5f 3148 if (!sunrpc_table_header)
0b4d4147 3149 sunrpc_table_header = register_sysctl_table(sunrpc_table);
fbf76683 3150
176e21ee 3151 xprt_register_transport(&xs_local_transport);
bc25571e
TT		 3152 xprt_register_transport(&xs_udp_transport);
3153 xprt_register_transport(&xs_tcp_transport);
f300baba 3154 xprt_register_transport(&xs_bc_tcp_transport);
bc25571e 3155
282b32e1
CL		 3156 return 0;
3157}
3158
3159/**
bc25571e 3160 * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
282b32e1
CL		 3161 *
3162 */
3163void cleanup_socket_xprt(void)
3164{
fbf76683
CL		 3165 if (sunrpc_table_header) {
3166 unregister_sysctl_table(sunrpc_table_header);
3167 sunrpc_table_header = NULL;
3168 }
bc25571e 3169
176e21ee 3170 xprt_unregister_transport(&xs_local_transport);
bc25571e
TT		 3171 xprt_unregister_transport(&xs_udp_transport);
3172 xprt_unregister_transport(&xs_tcp_transport);
f300baba 3173 xprt_unregister_transport(&xs_bc_tcp_transport);
282b32e1 3174}
cbf11071 3175
9bbb9e5a
RR		 3176static int param_set_uint_minmax(const char *val,
3177 const struct kernel_param *kp,
cbf11071
TM		 3178 unsigned int min, unsigned int max)
3179{
00cfaa94 3180 unsigned int num;
cbf11071
TM		 3181 int ret;
3182
3183 if (!val)
3184 return -EINVAL;
00cfaa94 3185 ret = kstrtouint(val, 0, &num);
9761a246
DC		 3186 if (ret)
3187 return ret;
3188 if (num < min || num > max)
cbf11071
TM		 3189 return -EINVAL;
3190 *((unsigned int *)kp->arg) = num;
3191 return 0;
3192}
3193
9bbb9e5a 3194static int param_set_portnr(const char *val, const struct kernel_param *kp)
cbf11071 3195{
ffb6ca33 3196 return param_set_uint_minmax(val, kp,
826799e6 3197 RPC_MIN_RESVPORT,
cbf11071
TM		 3198 RPC_MAX_RESVPORT);
3199}
3200
9c27847d 3201static const struct kernel_param_ops param_ops_portnr = {
9bbb9e5a
RR		 3202 .set = param_set_portnr,
3203 .get = param_get_uint,
3204};
3205
cbf11071
TM		 3206#define param_check_portnr(name, p) \
3207 __param_check(name, p, unsigned int);
3208
3209module_param_named(min_resvport, xprt_min_resvport, portnr, 0644);
3210module_param_named(max_resvport, xprt_max_resvport, portnr, 0644);
3211
9bbb9e5a
RR		 3212static int param_set_slot_table_size(const char *val,
3213 const struct kernel_param *kp)
cbf11071
TM		 3214{
3215 return param_set_uint_minmax(val, kp,
3216 RPC_MIN_SLOT_TABLE,
3217 RPC_MAX_SLOT_TABLE);
3218}
3219
9c27847d 3220static const struct kernel_param_ops param_ops_slot_table_size = {
9bbb9e5a
RR		 3221 .set = param_set_slot_table_size,
3222 .get = param_get_uint,
3223};
3224
cbf11071
TM		 3225#define param_check_slot_table_size(name, p) \
3226 __param_check(name, p, unsigned int);
3227
d9ba131d
TM		 3228static int param_set_max_slot_table_size(const char *val,
3229 const struct kernel_param *kp)
3230{
3231 return param_set_uint_minmax(val, kp,
3232 RPC_MIN_SLOT_TABLE,
3233 RPC_MAX_SLOT_TABLE_LIMIT);
3234}
3235
9c27847d 3236static const struct kernel_param_ops param_ops_max_slot_table_size = {
d9ba131d
TM		 3237 .set = param_set_max_slot_table_size,
3238 .get = param_get_uint,
3239};
3240
3241#define param_check_max_slot_table_size(name, p) \
3242 __param_check(name, p, unsigned int);
3243
cbf11071
TM		 3244module_param_named(tcp_slot_table_entries, xprt_tcp_slot_table_entries,
3245 slot_table_size, 0644);
d9ba131d
TM		 3246module_param_named(tcp_max_slot_table_entries, xprt_max_tcp_slot_table_entries,
3247 max_slot_table_size, 0644);
cbf11071
TM		 3248module_param_named(udp_slot_table_entries, xprt_udp_slot_table_entries,
3249 slot_table_size, 0644);
Linux 6.x block layer and io_uring tree(s)