projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Merge branch 'fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/evalenti/linux...
[linux-2.6-block.git] / net / sunrpc / xprtsock.c
CommitLineData
b2441318 1// SPDX-License-Identifier: GPL-2.0
a246b010
CL		 2/*
3 * linux/net/sunrpc/xprtsock.c
4 *
5 * Client-side transport implementation for sockets.
6 *
113aa838
AC		 7 * TCP callback races fixes (C) 1998 Red Hat
8 * TCP send fixes (C) 1998 Red Hat
a246b010
CL		 9 * TCP NFS related read + write fixes
10 * (C) 1999 Dave Airlie, University of Limerick, Ireland <airlied@linux.ie>
11 *
12 * Rewrite of larges part of the code in order to stabilize TCP stuff.
13 * Fix behaviour when socket buffer is full.
14 * (C) 1999 Trond Myklebust <trond.myklebust@fys.uio.no>
55aa4f58
CL		 15 *
16 * IP socket transport implementation, (C) 2005 Chuck Lever <cel@netapp.com>
8f9d5b1a
CL		 17 *
18 * IPv6 support contributed by Gilles Quillard, Bull Open Source, 2005.
19 * <gilles.quillard@bull.net>
a246b010
CL		 20 */
21
22#include <linux/types.h>
176e21ee 23#include <linux/string.h>
a246b010 24#include <linux/slab.h>
bc25571e 25#include <linux/module.h>
a246b010 26#include <linux/capability.h>
a246b010
CL		 27#include <linux/pagemap.h>
28#include <linux/errno.h>
29#include <linux/socket.h>
30#include <linux/in.h>
31#include <linux/net.h>
32#include <linux/mm.h>
176e21ee 33#include <linux/un.h>
a246b010
CL		 34#include <linux/udp.h>
35#include <linux/tcp.h>
36#include <linux/sunrpc/clnt.h>
5976687a 37#include <linux/sunrpc/addr.h>
02107148 38#include <linux/sunrpc/sched.h>
4cfc7e60 39#include <linux/sunrpc/svcsock.h>
49c36fcc 40#include <linux/sunrpc/xprtsock.h>
a246b010 41#include <linux/file.h>
9e00abc3 42#ifdef CONFIG_SUNRPC_BACKCHANNEL
44b98efd
RL		 43#include <linux/sunrpc/bc_xprt.h>
44#endif
a246b010
CL		 45
46#include <net/sock.h>
47#include <net/checksum.h>
48#include <net/udp.h>
49#include <net/tcp.h>
50
40b5ea0c
TM		 51#include <trace/events/sunrpc.h>
52
4cfc7e60 53#include "sunrpc.h"
176e21ee 54
3d188805
TM		 55#define RPC_TCP_READ_CHUNK_SZ (3*512*1024)
56
176e21ee 57static void xs_close(struct rpc_xprt *xprt);
7196dbb0
TM		 58static void xs_tcp_set_socket_timeouts(struct rpc_xprt *xprt,
59 struct socket *sock);
176e21ee 60
c556b754
CL		 61/*
62 * xprtsock tunables
63 */
09acfea5
TM		 64static unsigned int xprt_udp_slot_table_entries = RPC_DEF_SLOT_TABLE;
65static unsigned int xprt_tcp_slot_table_entries = RPC_MIN_SLOT_TABLE;
66static unsigned int xprt_max_tcp_slot_table_entries = RPC_MAX_SLOT_TABLE;
c556b754 67
09acfea5
TM		 68static unsigned int xprt_min_resvport = RPC_DEF_MIN_RESVPORT;
69static unsigned int xprt_max_resvport = RPC_DEF_MAX_RESVPORT;
c556b754 70
54c09874
TM		 71#if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
72
7d1e8255 73#define XS_TCP_LINGER_TO (15U * HZ)
25fe6142 74static unsigned int xs_tcp_fin_timeout __read_mostly = XS_TCP_LINGER_TO;
7d1e8255 75
fbf76683
CL		 76/*
77 * We can register our own files under /proc/sys/sunrpc by
78 * calling register_sysctl_table() again. The files in that
79 * directory become the union of all files registered there.
80 *
81 * We simply need to make sure that we don't collide with
82 * someone else's file names!
83 */
84
fbf76683
CL		 85static unsigned int min_slot_table_size = RPC_MIN_SLOT_TABLE;
86static unsigned int max_slot_table_size = RPC_MAX_SLOT_TABLE;
d9ba131d 87static unsigned int max_tcp_slot_table_limit = RPC_MAX_SLOT_TABLE_LIMIT;
fbf76683
CL		 88static unsigned int xprt_min_resvport_limit = RPC_MIN_RESVPORT;
89static unsigned int xprt_max_resvport_limit = RPC_MAX_RESVPORT;
90
91static struct ctl_table_header *sunrpc_table_header;
92
93/*
94 * FIXME: changing the UDP slot table size should also resize the UDP
95 * socket buffers for existing UDP transports
96 */
fe2c6338 97static struct ctl_table xs_tunables_table[] = {
fbf76683 98 {
fbf76683
CL		 99 .procname = "udp_slot_table_entries",
100 .data = &xprt_udp_slot_table_entries,
101 .maxlen = sizeof(unsigned int),
102 .mode = 0644,
6d456111 103 .proc_handler = proc_dointvec_minmax,
fbf76683
CL		 104 .extra1 = &min_slot_table_size,
105 .extra2 = &max_slot_table_size
106 },
107 {
fbf76683
CL		 108 .procname = "tcp_slot_table_entries",
109 .data = &xprt_tcp_slot_table_entries,
110 .maxlen = sizeof(unsigned int),
111 .mode = 0644,
6d456111 112 .proc_handler = proc_dointvec_minmax,
fbf76683
CL		 113 .extra1 = &min_slot_table_size,
114 .extra2 = &max_slot_table_size
115 },
d9ba131d
TM		 116 {
117 .procname = "tcp_max_slot_table_entries",
118 .data = &xprt_max_tcp_slot_table_entries,
119 .maxlen = sizeof(unsigned int),
120 .mode = 0644,
121 .proc_handler = proc_dointvec_minmax,
122 .extra1 = &min_slot_table_size,
123 .extra2 = &max_tcp_slot_table_limit
124 },
fbf76683 125 {
fbf76683
CL		 126 .procname = "min_resvport",
127 .data = &xprt_min_resvport,
128 .maxlen = sizeof(unsigned int),
129 .mode = 0644,
6d456111 130 .proc_handler = proc_dointvec_minmax,
fbf76683 131 .extra1 = &xprt_min_resvport_limit,
e08ea3a9 132 .extra2 = &xprt_max_resvport
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,
e08ea3a9 140 .extra1 = &xprt_min_resvport,
fbf76683
CL		 141 .extra2 = &xprt_max_resvport_limit
142 },
25fe6142
TM		 143 {
144 .procname = "tcp_fin_timeout",
145 .data = &xs_tcp_fin_timeout,
146 .maxlen = sizeof(xs_tcp_fin_timeout),
147 .mode = 0644,
6d456111 148 .proc_handler = proc_dointvec_jiffies,
fbf76683 149 },
f8572d8f 150 { },
fbf76683
CL		 151};
152
fe2c6338 153static struct ctl_table sunrpc_table[] = {
fbf76683 154 {
fbf76683
CL		 155 .procname = "sunrpc",
156 .mode = 0555,
157 .child = xs_tunables_table
158 },
f8572d8f 159 { },
fbf76683
CL		 160};
161
162#endif
163
03bf4b70
CL		 164/*
165 * Wait duration for a reply from the RPC portmapper.
166 */
167#define XS_BIND_TO (60U * HZ)
168
169/*
170 * Delay if a UDP socket connect error occurs. This is most likely some
171 * kind of resource problem on the local host.
172 */
173#define XS_UDP_REEST_TO (2U * HZ)
174
175/*
176 * The reestablish timeout allows clients to delay for a bit before attempting
177 * to reconnect to a server that just dropped our connection.
178 *
179 * We implement an exponential backoff when trying to reestablish a TCP
180 * transport connection with the server. Some servers like to drop a TCP
181 * connection when they are overworked, so we start with a short timeout and
182 * increase over time if the server is down or not responding.
183 */
184#define XS_TCP_INIT_REEST_TO (3U * HZ)
03bf4b70
CL		 185
186/*
187 * TCP idle timeout; client drops the transport socket if it is idle
188 * for this long. Note that we also timeout UDP sockets to prevent
189 * holding port numbers when there is no RPC traffic.
190 */
191#define XS_IDLE_DISC_TO (5U * 60 * HZ)
192
f895b252 193#if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
a246b010 194# undef RPC_DEBUG_DATA
9903cd1c 195# define RPCDBG_FACILITY RPCDBG_TRANS
a246b010
CL		 196#endif
197
a246b010 198#ifdef RPC_DEBUG_DATA
9903cd1c 199static void xs_pktdump(char *msg, u32 *packet, unsigned int count)
a246b010 200{
9903cd1c
CL		 201 u8 *buf = (u8 *) packet;
202 int j;
a246b010 203
46121cf7 204 dprintk("RPC: %s\n", msg);
a246b010
CL		 205 for (j = 0; j < count && j < 128; j += 4) {
206 if (!(j & 31)) {
207 if (j)
208 dprintk("\n");
209 dprintk("0x%04x ", j);
210 }
211 dprintk("%02x%02x%02x%02x ",
212 buf[j], buf[j+1], buf[j+2], buf[j+3]);
213 }
214 dprintk("\n");
215}
216#else
9903cd1c 217static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count)
a246b010
CL		 218{
219 /* NOP */
220}
221#endif
222
2118071d
TM		 223static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
224{
225 return (struct rpc_xprt *) sk->sk_user_data;
226}
227
95392c59
CL		 228static inline struct sockaddr *xs_addr(struct rpc_xprt *xprt)
229{
230 return (struct sockaddr *) &xprt->addr;
231}
232
176e21ee
CL		 233static inline struct sockaddr_un *xs_addr_un(struct rpc_xprt *xprt)
234{
235 return (struct sockaddr_un *) &xprt->addr;
236}
237
95392c59 238static inline struct sockaddr_in *xs_addr_in(struct rpc_xprt *xprt)
edb267a6 239{
95392c59
CL		 240 return (struct sockaddr_in *) &xprt->addr;
241}
242
243static inline struct sockaddr_in6 *xs_addr_in6(struct rpc_xprt *xprt)
244{
245 return (struct sockaddr_in6 *) &xprt->addr;
246}
247
c877b849 248static void xs_format_common_peer_addresses(struct rpc_xprt *xprt)
edb267a6 249{
c877b849 250 struct sockaddr *sap = xs_addr(xprt);
9dc3b095
CL		 251 struct sockaddr_in6 *sin6;
252 struct sockaddr_in *sin;
176e21ee 253 struct sockaddr_un *sun;
c877b849 254 char buf[128];
edb267a6 255
9dc3b095 256 switch (sap->sa_family) {
176e21ee
CL		 257 case AF_LOCAL:
258 sun = xs_addr_un(xprt);
259 strlcpy(buf, sun->sun_path, sizeof(buf));
260 xprt->address_strings[RPC_DISPLAY_ADDR] =
261 kstrdup(buf, GFP_KERNEL);
262 break;
9dc3b095 263 case AF_INET:
176e21ee
CL		 264 (void)rpc_ntop(sap, buf, sizeof(buf));
265 xprt->address_strings[RPC_DISPLAY_ADDR] =
266 kstrdup(buf, GFP_KERNEL);
9dc3b095 267 sin = xs_addr_in(xprt);
fc0b5791 268 snprintf(buf, sizeof(buf), "%08x", ntohl(sin->sin_addr.s_addr));
9dc3b095
CL		 269 break;
270 case AF_INET6:
176e21ee
CL		 271 (void)rpc_ntop(sap, buf, sizeof(buf));
272 xprt->address_strings[RPC_DISPLAY_ADDR] =
273 kstrdup(buf, GFP_KERNEL);
9dc3b095 274 sin6 = xs_addr_in6(xprt);
fc0b5791 275 snprintf(buf, sizeof(buf), "%pi6", &sin6->sin6_addr);
9dc3b095
CL		 276 break;
277 default:
278 BUG();
756805e7 279 }
176e21ee 280
9dc3b095 281 xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = kstrdup(buf, GFP_KERNEL);
edb267a6
CL		 282}
283
9dc3b095 284static void xs_format_common_peer_ports(struct rpc_xprt *xprt)
4b6473fb 285{
9dc3b095
CL		 286 struct sockaddr *sap = xs_addr(xprt);
287 char buf[128];
4b6473fb 288
81160e66 289 snprintf(buf, sizeof(buf), "%u", rpc_get_port(sap));
c877b849 290 xprt->address_strings[RPC_DISPLAY_PORT] = kstrdup(buf, GFP_KERNEL);
4b6473fb 291
81160e66 292 snprintf(buf, sizeof(buf), "%4hx", rpc_get_port(sap));
c877b849
CL		 293 xprt->address_strings[RPC_DISPLAY_HEX_PORT] = kstrdup(buf, GFP_KERNEL);
294}
4b6473fb 295
9dc3b095
CL		 296static void xs_format_peer_addresses(struct rpc_xprt *xprt,
297 const char *protocol,
298 const char *netid)
c877b849 299{
b454ae90 300 xprt->address_strings[RPC_DISPLAY_PROTO] = protocol;
b454ae90 301 xprt->address_strings[RPC_DISPLAY_NETID] = netid;
c877b849 302 xs_format_common_peer_addresses(xprt);
9dc3b095 303 xs_format_common_peer_ports(xprt);
edb267a6 304}
4b6473fb 305
9dc3b095 306static void xs_update_peer_port(struct rpc_xprt *xprt)
4b6473fb 307{
9dc3b095
CL		 308 kfree(xprt->address_strings[RPC_DISPLAY_HEX_PORT]);
309 kfree(xprt->address_strings[RPC_DISPLAY_PORT]);
4417c8c4 310
9dc3b095 311 xs_format_common_peer_ports(xprt);
edb267a6
CL		 312}
313
314static void xs_free_peer_addresses(struct rpc_xprt *xprt)
315{
33e01dc7
CL		 316 unsigned int i;
317
318 for (i = 0; i < RPC_DISPLAY_MAX; i++)
319 switch (i) {
320 case RPC_DISPLAY_PROTO:
321 case RPC_DISPLAY_NETID:
322 continue;
323 default:
324 kfree(xprt->address_strings[i]);
325 }
edb267a6
CL		 326}
327
b4b5cc85
CL		 328#define XS_SENDMSG_FLAGS (MSG_DONTWAIT | MSG_NOSIGNAL)
329
24c5684b 330static int xs_send_kvec(struct socket *sock, struct sockaddr *addr, int addrlen, struct kvec *vec, unsigned int base, int more)
b4b5cc85 331{
b4b5cc85
CL		 332 struct msghdr msg = {
333 .msg_name = addr,
334 .msg_namelen = addrlen,
24c5684b
TM		 335 .msg_flags = XS_SENDMSG_FLAGS | (more ? MSG_MORE : 0),
336 };
337 struct kvec iov = {
338 .iov_base = vec->iov_base + base,
339 .iov_len = vec->iov_len - base,
b4b5cc85
CL		 340 };
341
24c5684b 342 if (iov.iov_len != 0)
b4b5cc85
CL		 343 return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
344 return kernel_sendmsg(sock, &msg, NULL, 0, 0);
345}
346
f279cd00 347static int xs_send_pagedata(struct socket *sock, struct xdr_buf *xdr, unsigned int base, int more, bool zerocopy, int *sent_p)
b4b5cc85 348{
a6b31d18
TM		 349 ssize_t (*do_sendpage)(struct socket *sock, struct page *page,
350 int offset, size_t size, int flags);
24c5684b
TM		 351 struct page **ppage;
352 unsigned int remainder;
f279cd00 353 int err;
24c5684b
TM		 354
355 remainder = xdr->page_len - base;
356 base += xdr->page_base;
357 ppage = xdr->pages + (base >> PAGE_SHIFT);
358 base &= ~PAGE_MASK;
a6b31d18
TM		 359 do_sendpage = sock->ops->sendpage;
360 if (!zerocopy)
361 do_sendpage = sock_no_sendpage;
24c5684b
TM		 362 for(;;) {
363 unsigned int len = min_t(unsigned int, PAGE_SIZE - base, remainder);
364 int flags = XS_SENDMSG_FLAGS;
b4b5cc85 365
24c5684b 366 remainder -= len;
31303d6c 367 if (more)
24c5684b 368 flags |= MSG_MORE;
31303d6c
TM		 369 if (remainder != 0)
370 flags |= MSG_SENDPAGE_NOTLAST | MSG_MORE;
a6b31d18 371 err = do_sendpage(sock, *ppage, base, len, flags);
24c5684b
TM		 372 if (remainder == 0 || err != len)
373 break;
f279cd00 374 *sent_p += err;
24c5684b
TM		 375 ppage++;
376 base = 0;
377 }
f279cd00
JB		 378 if (err > 0) {
379 *sent_p += err;
380 err = 0;
381 }
382 return err;
b4b5cc85
CL		 383}
384
9903cd1c
CL		 385/**
386 * xs_sendpages - write pages directly to a socket
387 * @sock: socket to send on
388 * @addr: UDP only -- address of destination
389 * @addrlen: UDP only -- length of destination address
390 * @xdr: buffer containing this request
391 * @base: starting position in the buffer
a6b31d18 392 * @zerocopy: true if it is safe to use sendpage()
f279cd00 393 * @sent_p: return the total number of bytes successfully queued for sending
9903cd1c 394 *
a246b010 395 */
f279cd00 396static int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base, bool zerocopy, int *sent_p)
a246b010 397{
24c5684b 398 unsigned int remainder = xdr->len - base;
f279cd00
JB		 399 int err = 0;
400 int sent = 0;
a246b010 401
262965f5 402 if (unlikely(!sock))
fba91afb 403 return -ENOTSOCK;
262965f5 404
24c5684b
TM		 405 if (base != 0) {
406 addr = NULL;
407 addrlen = 0;
408 }
262965f5 409
24c5684b
TM		 410 if (base < xdr->head[0].iov_len || addr != NULL) {
411 unsigned int len = xdr->head[0].iov_len - base;
412 remainder -= len;
413 err = xs_send_kvec(sock, addr, addrlen, &xdr->head[0], base, remainder != 0);
414 if (remainder == 0 || err != len)
a246b010 415 goto out;
f279cd00 416 *sent_p += err;
a246b010
CL		 417 base = 0;
418 } else
24c5684b 419 base -= xdr->head[0].iov_len;
a246b010 420
24c5684b
TM		 421 if (base < xdr->page_len) {
422 unsigned int len = xdr->page_len - base;
423 remainder -= len;
f279cd00
JB		 424 err = xs_send_pagedata(sock, xdr, base, remainder != 0, zerocopy, &sent);
425 *sent_p += sent;
426 if (remainder == 0 || sent != len)
a246b010
CL		 427 goto out;
428 base = 0;
24c5684b
TM		 429 } else
430 base -= xdr->page_len;
431
432 if (base >= xdr->tail[0].iov_len)
f279cd00 433 return 0;
24c5684b 434 err = xs_send_kvec(sock, NULL, 0, &xdr->tail[0], base, 0);
a246b010 435out:
f279cd00
JB		 436 if (err > 0) {
437 *sent_p += err;
438 err = 0;
439 }
440 return err;
a246b010
CL		 441}
442
b6ddf64f
TM		 443static void xs_nospace_callback(struct rpc_task *task)
444{
445 struct sock_xprt *transport = container_of(task->tk_rqstp->rq_xprt, struct sock_xprt, xprt);
446
447 transport->inet->sk_write_pending--;
b6ddf64f
TM		 448}
449
9903cd1c 450/**
262965f5
CL		 451 * xs_nospace - place task on wait queue if transmit was incomplete
452 * @task: task to put to sleep
9903cd1c 453 *
a246b010 454 */
5e3771ce 455static int xs_nospace(struct rpc_task *task)
a246b010 456{
262965f5
CL		 457 struct rpc_rqst *req = task->tk_rqstp;
458 struct rpc_xprt *xprt = req->rq_xprt;
ee0ac0c2 459 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
06ea0bfe 460 struct sock *sk = transport->inet;
24ca9a84 461 int ret = -EAGAIN;
a246b010 462
46121cf7 463 dprintk("RPC: %5u xmit incomplete (%u left of %u)\n",
262965f5
CL		 464 task->tk_pid, req->rq_slen - req->rq_bytes_sent,
465 req->rq_slen);
466
b6ddf64f
TM		 467 /* Protect against races with write_space */
468 spin_lock_bh(&xprt->transport_lock);
469
470 /* Don't race with disconnect */
471 if (xprt_connected(xprt)) {
13331a55
TM		 472 /* wait for more buffer space */
473 sk->sk_write_pending++;
474 xprt_wait_for_buffer_space(task, xs_nospace_callback);
475 } else
5e3771ce 476 ret = -ENOTCONN;
262965f5 477
b6ddf64f 478 spin_unlock_bh(&xprt->transport_lock);
06ea0bfe
TM		 479
480 /* Race breaker in case memory is freed before above code is called */
d48f9ce7
DV		 481 if (ret == -EAGAIN) {
482 struct socket_wq *wq;
483
484 rcu_read_lock();
485 wq = rcu_dereference(sk->sk_wq);
486 set_bit(SOCKWQ_ASYNC_NOSPACE, &wq->flags);
487 rcu_read_unlock();
488
489 sk->sk_write_space(sk);
490 }
5e3771ce 491 return ret;
262965f5
CL		 492}
493
61677eee
CL		 494/*
495 * Construct a stream transport record marker in @buf.
496 */
497static inline void xs_encode_stream_record_marker(struct xdr_buf *buf)
498{
499 u32 reclen = buf->len - sizeof(rpc_fraghdr);
500 rpc_fraghdr *base = buf->head[0].iov_base;
501 *base = cpu_to_be32(RPC_LAST_STREAM_FRAGMENT | reclen);
502}
503
176e21ee
CL		 504/**
505 * xs_local_send_request - write an RPC request to an AF_LOCAL socket
506 * @task: RPC task that manages the state of an RPC request
507 *
508 * Return values:
509 * 0: The request has been sent
510 * EAGAIN: The socket was blocked, please call again later to
511 * complete the request
512 * ENOTCONN: Caller needs to invoke connect logic then call again
513 * other: Some other error occured, the request was not sent
514 */
515static int xs_local_send_request(struct rpc_task *task)
516{
517 struct rpc_rqst *req = task->tk_rqstp;
518 struct rpc_xprt *xprt = req->rq_xprt;
519 struct sock_xprt *transport =
520 container_of(xprt, struct sock_xprt, xprt);
521 struct xdr_buf *xdr = &req->rq_snd_buf;
522 int status;
f279cd00 523 int sent = 0;
176e21ee
CL		 524
525 xs_encode_stream_record_marker(&req->rq_snd_buf);
526
527 xs_pktdump("packet data:",
528 req->rq_svec->iov_base, req->rq_svec->iov_len);
529
78215759 530 req->rq_xtime = ktime_get();
f279cd00
JB		 531 status = xs_sendpages(transport->sock, NULL, 0, xdr, req->rq_bytes_sent,
532 true, &sent);
176e21ee
CL		 533 dprintk("RPC: %s(%u) = %d\n",
534 __func__, xdr->len - req->rq_bytes_sent, status);
743c69e7
N		 535
536 if (status == -EAGAIN && sock_writeable(transport->inet))
537 status = -ENOBUFS;
538
f279cd00
JB		 539 if (likely(sent > 0) || status == 0) {
540 req->rq_bytes_sent += sent;
541 req->rq_xmit_bytes_sent += sent;
176e21ee
CL		 542 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
543 req->rq_bytes_sent = 0;
544 return 0;
545 }
546 status = -EAGAIN;
547 }
548
549 switch (status) {
3601c4a9 550 case -ENOBUFS:
b5872f0c 551 break;
176e21ee
CL		 552 case -EAGAIN:
553 status = xs_nospace(task);
554 break;
555 default:
556 dprintk("RPC: sendmsg returned unrecognized error %d\n",
557 -status);
e9d47639 558 /* fall through */
176e21ee
CL		 559 case -EPIPE:
560 xs_close(xprt);
561 status = -ENOTCONN;
562 }
563
564 return status;
565}
566
262965f5
CL		 567/**
568 * xs_udp_send_request - write an RPC request to a UDP socket
569 * @task: address of RPC task that manages the state of an RPC request
570 *
571 * Return values:
572 * 0: The request has been sent
573 * EAGAIN: The socket was blocked, please call again later to
574 * complete the request
575 * ENOTCONN: Caller needs to invoke connect logic then call again
25985edc 576 * other: Some other error occurred, the request was not sent
262965f5
CL		 577 */
578static int xs_udp_send_request(struct rpc_task *task)
579{
580 struct rpc_rqst *req = task->tk_rqstp;
581 struct rpc_xprt *xprt = req->rq_xprt;
ee0ac0c2 582 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
262965f5 583 struct xdr_buf *xdr = &req->rq_snd_buf;
f279cd00 584 int sent = 0;
262965f5 585 int status;
a246b010 586
9903cd1c 587 xs_pktdump("packet data:",
a246b010
CL		 588 req->rq_svec->iov_base,
589 req->rq_svec->iov_len);
590
01d37c42
TM		 591 if (!xprt_bound(xprt))
592 return -ENOTCONN;
78215759 593 req->rq_xtime = ktime_get();
f279cd00
JB		 594 status = xs_sendpages(transport->sock, xs_addr(xprt), xprt->addrlen,
595 xdr, req->rq_bytes_sent, true, &sent);
a246b010 596
46121cf7 597 dprintk("RPC: xs_udp_send_request(%u) = %d\n",
262965f5 598 xdr->len - req->rq_bytes_sent, status);
a246b010 599
3dedbb5c
JB		 600 /* firewall is blocking us, don't return -EAGAIN or we end up looping */
601 if (status == -EPERM)
602 goto process_status;
603
743c69e7
N		 604 if (status == -EAGAIN && sock_writeable(transport->inet))
605 status = -ENOBUFS;
606
f279cd00
JB		 607 if (sent > 0 || status == 0) {
608 req->rq_xmit_bytes_sent += sent;
609 if (sent >= req->rq_slen)
2199700f
TM		 610 return 0;
611 /* Still some bytes left; set up for a retry later. */
262965f5 612 status = -EAGAIN;
2199700f 613 }
a246b010 614
3dedbb5c 615process_status:
262965f5 616 switch (status) {
fba91afb
TM		 617 case -ENOTSOCK:
618 status = -ENOTCONN;
619 /* Should we call xs_close() here? */
620 break;
b6ddf64f 621 case -EAGAIN:
5e3771ce 622 status = xs_nospace(task);
b6ddf64f 623 break;
262965f5 624 case -ENETUNREACH:
3601c4a9 625 case -ENOBUFS:
262965f5 626 case -EPIPE:
a246b010 627 case -ECONNREFUSED:
3dedbb5c 628 case -EPERM:
a246b010 629 /* When the server has died, an ICMP port unreachable message
9903cd1c 630 * prompts ECONNREFUSED. */
13331a55
TM		 631 break;
632 default:
633 dprintk("RPC: sendmsg returned unrecognized error %d\n",
634 -status);
a246b010 635 }
5fe46e9d 636
262965f5 637 return status;
a246b010
CL		 638}
639
9903cd1c 640/**
262965f5 641 * xs_tcp_send_request - write an RPC request to a TCP socket
9903cd1c
CL		 642 * @task: address of RPC task that manages the state of an RPC request
643 *
644 * Return values:
262965f5
CL		 645 * 0: The request has been sent
646 * EAGAIN: The socket was blocked, please call again later to
647 * complete the request
648 * ENOTCONN: Caller needs to invoke connect logic then call again
25985edc 649 * other: Some other error occurred, the request was not sent
9903cd1c
CL		 650 *
651 * XXX: In the case of soft timeouts, should we eventually give up
262965f5 652 * if sendmsg is not able to make progress?
9903cd1c 653 */
262965f5 654static int xs_tcp_send_request(struct rpc_task *task)
a246b010
CL		 655{
656 struct rpc_rqst *req = task->tk_rqstp;
657 struct rpc_xprt *xprt = req->rq_xprt;
ee0ac0c2 658 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
262965f5 659 struct xdr_buf *xdr = &req->rq_snd_buf;
a6b31d18 660 bool zerocopy = true;
9ffadfbc 661 bool vm_wait = false;
b595bb15 662 int status;
f279cd00 663 int sent;
a246b010 664
61677eee 665 xs_encode_stream_record_marker(&req->rq_snd_buf);
a246b010 666
262965f5
CL		 667 xs_pktdump("packet data:",
668 req->rq_svec->iov_base,
669 req->rq_svec->iov_len);
a6b31d18
TM		 670 /* Don't use zero copy if this is a resend. If the RPC call
671 * completes while the socket holds a reference to the pages,
672 * then we may end up resending corrupted data.
673 */
674 if (task->tk_flags & RPC_TASK_SENT)
675 zerocopy = false;
a246b010 676
7196dbb0
TM		 677 if (test_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state))
678 xs_tcp_set_socket_timeouts(xprt, transport->sock);
679
a246b010
CL		 680 /* Continue transmitting the packet/record. We must be careful
681 * to cope with writespace callbacks arriving _after_ we have
262965f5 682 * called sendmsg(). */
78215759 683 req->rq_xtime = ktime_get();
a246b010 684 while (1) {
f279cd00
JB		 685 sent = 0;
686 status = xs_sendpages(transport->sock, NULL, 0, xdr,
687 req->rq_bytes_sent, zerocopy, &sent);
a246b010 688
46121cf7 689 dprintk("RPC: xs_tcp_send_request(%u) = %d\n",
262965f5 690 xdr->len - req->rq_bytes_sent, status);
a246b010 691
262965f5
CL		 692 /* If we've sent the entire packet, immediately
693 * reset the count of bytes sent. */
f279cd00
JB		 694 req->rq_bytes_sent += sent;
695 req->rq_xmit_bytes_sent += sent;
262965f5
CL		 696 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
697 req->rq_bytes_sent = 0;
698 return 0;
699 }
a246b010 700
9ffadfbc
TM		 701 WARN_ON_ONCE(sent == 0 && status == 0);
702
703 if (status == -EAGAIN ) {
704 /*
705 * Return EAGAIN if we're sure we're hitting the
706 * socket send buffer limits.
707 */
708 if (test_bit(SOCK_NOSPACE, &transport->sock->flags))
709 break;
710 /*
711 * Did we hit a memory allocation failure?
712 */
713 if (sent == 0) {
714 status = -ENOBUFS;
715 if (vm_wait)
716 break;
717 /* Retry, knowing now that we're below the
718 * socket send buffer limit
719 */
720 vm_wait = true;
721 }
722 continue;
723 }
f580dd04
TM		 724 if (status < 0)
725 break;
9ffadfbc 726 vm_wait = false;
a246b010
CL		 727 }
728
262965f5 729 switch (status) {
fba91afb
TM		 730 case -ENOTSOCK:
731 status = -ENOTCONN;
732 /* Should we call xs_close() here? */
733 break;
262965f5 734 case -EAGAIN:
5e3771ce 735 status = xs_nospace(task);
262965f5 736 break;
262965f5 737 case -ECONNRESET:
2a9e1cfa 738 case -ECONNREFUSED:
262965f5 739 case -ENOTCONN:
3913c78c 740 case -EADDRINUSE:
b5872f0c 741 case -ENOBUFS:
b9d2bb2e 742 case -EPIPE:
13331a55
TM		 743 break;
744 default:
745 dprintk("RPC: sendmsg returned unrecognized error %d\n",
746 -status);
a246b010 747 }
5fe46e9d 748
a246b010
CL		 749 return status;
750}
751
e0ab53de
TM		 752/**
753 * xs_tcp_release_xprt - clean up after a tcp transmission
754 * @xprt: transport
755 * @task: rpc task
756 *
757 * This cleans up if an error causes us to abort the transmission of a request.
758 * In this case, the socket may need to be reset in order to avoid confusing
759 * the server.
760 */
761static void xs_tcp_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
762{
763 struct rpc_rqst *req;
764
765 if (task != xprt->snd_task)
766 return;
767 if (task == NULL)
768 goto out_release;
769 req = task->tk_rqstp;
43cedbf0
TM		 770 if (req == NULL)
771 goto out_release;
e0ab53de
TM		 772 if (req->rq_bytes_sent == 0)
773 goto out_release;
774 if (req->rq_bytes_sent == req->rq_snd_buf.len)
775 goto out_release;
a4f0835c 776 set_bit(XPRT_CLOSE_WAIT, &xprt->state);
e0ab53de
TM		 777out_release:
778 xprt_release_xprt(xprt, task);
779}
780
2a9e1cfa
TM		 781static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk)
782{
783 transport->old_data_ready = sk->sk_data_ready;
784 transport->old_state_change = sk->sk_state_change;
785 transport->old_write_space = sk->sk_write_space;
2118071d 786 transport->old_error_report = sk->sk_error_report;
2a9e1cfa
TM		 787}
788
789static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk)
790{
791 sk->sk_data_ready = transport->old_data_ready;
792 sk->sk_state_change = transport->old_state_change;
793 sk->sk_write_space = transport->old_write_space;
2118071d
TM		 794 sk->sk_error_report = transport->old_error_report;
795}
796
42d42a5b
TM		 797static void xs_sock_reset_state_flags(struct rpc_xprt *xprt)
798{
799 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
800
801 clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state);
802}
803
b70ae915
TM		 804static void xs_sock_reset_connection_flags(struct rpc_xprt *xprt)
805{
806 smp_mb__before_atomic();
807 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
808 clear_bit(XPRT_CLOSING, &xprt->state);
42d42a5b 809 xs_sock_reset_state_flags(xprt);
b70ae915
TM		 810 smp_mb__after_atomic();
811}
812
2118071d
TM		 813/**
814 * xs_error_report - callback to handle TCP socket state errors
815 * @sk: socket
816 *
817 * Note: we don't call sock_error() since there may be a rpc_task
818 * using the socket, and so we don't want to clear sk->sk_err.
819 */
820static void xs_error_report(struct sock *sk)
821{
822 struct rpc_xprt *xprt;
823 int err;
824
825 read_lock_bh(&sk->sk_callback_lock);
826 if (!(xprt = xprt_from_sock(sk)))
827 goto out;
828
829 err = -sk->sk_err;
830 if (err == 0)
831 goto out;
832 dprintk("RPC: xs_error_report client %p, error=%d...\n",
833 xprt, -err);
e8353c76 834 trace_rpc_socket_error(xprt, sk->sk_socket, err);
2118071d
TM		 835 xprt_wake_pending_tasks(xprt, err);
836 out:
837 read_unlock_bh(&sk->sk_callback_lock);
2a9e1cfa
TM		 838}
839
fe315e76 840static void xs_reset_transport(struct sock_xprt *transport)
a246b010 841{
ee0ac0c2
CL		 842 struct socket *sock = transport->sock;
843 struct sock *sk = transport->inet;
6cc7e908 844 struct rpc_xprt *xprt = &transport->xprt;
a246b010 845
fe315e76
CL		 846 if (sk == NULL)
847 return;
9903cd1c 848
264d1df3
JL		 849 if (atomic_read(&transport->xprt.swapper))
850 sk_clear_memalloc(sk);
851
09939204
TM		 852 kernel_sock_shutdown(sock, SHUT_RDWR);
853
edc1b01c 854 mutex_lock(&transport->recv_mutex);
a246b010 855 write_lock_bh(&sk->sk_callback_lock);
ee0ac0c2
CL		 856 transport->inet = NULL;
857 transport->sock = NULL;
a246b010 858
9903cd1c 859 sk->sk_user_data = NULL;
2a9e1cfa
TM		 860
861 xs_restore_old_callbacks(transport, sk);
0c78789e 862 xprt_clear_connected(xprt);
a246b010 863 write_unlock_bh(&sk->sk_callback_lock);
6cc7e908 864 xs_sock_reset_connection_flags(xprt);
edc1b01c 865 mutex_unlock(&transport->recv_mutex);
a246b010 866
6cc7e908 867 trace_rpc_socket_close(xprt, sock);
a246b010 868 sock_release(sock);
fe315e76
CL		 869}
870
871/**
872 * xs_close - close a socket
873 * @xprt: transport
874 *
875 * This is used when all requests are complete; ie, no DRC state remains
876 * on the server we want to save.
f75e6745
TM		 877 *
878 * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with
879 * xs_reset_transport() zeroing the socket from underneath a writer.
fe315e76
CL		 880 */
881static void xs_close(struct rpc_xprt *xprt)
882{
883 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
884
885 dprintk("RPC: xs_close xprt %p\n", xprt);
886
887 xs_reset_transport(transport);
61d0a8e6 888 xprt->reestablish_timeout = 0;
fe315e76 889
62da3b24 890 xprt_disconnect_done(xprt);
a246b010
CL		 891}
892
4a068258
CL		 893static void xs_inject_disconnect(struct rpc_xprt *xprt)
894{
895 dprintk("RPC: injecting transport disconnect on xprt=%p\n",
896 xprt);
897 xprt_disconnect_done(xprt);
898}
899
315f3812
KM		 900static void xs_xprt_free(struct rpc_xprt *xprt)
901{
902 xs_free_peer_addresses(xprt);
903 xprt_free(xprt);
904}
905
9903cd1c
CL		 906/**
907 * xs_destroy - prepare to shutdown a transport
908 * @xprt: doomed transport
909 *
910 */
911static void xs_destroy(struct rpc_xprt *xprt)
a246b010 912{
03c78827
TM		 913 struct sock_xprt *transport = container_of(xprt,
914 struct sock_xprt, xprt);
46121cf7 915 dprintk("RPC: xs_destroy xprt %p\n", xprt);
9903cd1c 916
03c78827 917 cancel_delayed_work_sync(&transport->connect_worker);
a1311d87 918 xs_close(xprt);
edc1b01c 919 cancel_work_sync(&transport->recv_worker);
315f3812 920 xs_xprt_free(xprt);
a1311d87 921 module_put(THIS_MODULE);
a246b010
CL		 922}
923
176e21ee
CL		 924static int xs_local_copy_to_xdr(struct xdr_buf *xdr, struct sk_buff *skb)
925{
926 struct xdr_skb_reader desc = {
927 .skb = skb,
928 .offset = sizeof(rpc_fraghdr),
929 .count = skb->len - sizeof(rpc_fraghdr),
930 };
931
932 if (xdr_partial_copy_from_skb(xdr, 0, &desc, xdr_skb_read_bits) < 0)
933 return -1;
934 if (desc.count)
935 return -1;
936 return 0;
937}
938
939/**
a2648094
TM		 940 * xs_local_data_read_skb
941 * @xprt: transport
942 * @sk: socket
943 * @skb: skbuff
176e21ee
CL		 944 *
945 * Currently this assumes we can read the whole reply in a single gulp.
946 */
a2648094
TM		 947static void xs_local_data_read_skb(struct rpc_xprt *xprt,
948 struct sock *sk,
949 struct sk_buff *skb)
176e21ee
CL		 950{
951 struct rpc_task *task;
176e21ee 952 struct rpc_rqst *rovr;
a2648094 953 int repsize, copied;
176e21ee
CL		 954 u32 _xid;
955 __be32 *xp;
956
176e21ee
CL		 957 repsize = skb->len - sizeof(rpc_fraghdr);
958 if (repsize < 4) {
959 dprintk("RPC: impossible RPC reply size %d\n", repsize);
a2648094 960 return;
176e21ee
CL		 961 }
962
963 /* Copy the XID from the skb... */
964 xp = skb_header_pointer(skb, sizeof(rpc_fraghdr), sizeof(_xid), &_xid);
965 if (xp == NULL)
a2648094 966 return;
176e21ee
CL		 967
968 /* Look up and lock the request corresponding to the given XID */
ce7c252a 969 spin_lock(&xprt->recv_lock);
176e21ee
CL		 970 rovr = xprt_lookup_rqst(xprt, *xp);
971 if (!rovr)
972 goto out_unlock;
729749bb 973 xprt_pin_rqst(rovr);
ce7c252a 974 spin_unlock(&xprt->recv_lock);
176e21ee
CL		 975 task = rovr->rq_task;
976
977 copied = rovr->rq_private_buf.buflen;
978 if (copied > repsize)
979 copied = repsize;
980
981 if (xs_local_copy_to_xdr(&rovr->rq_private_buf, skb)) {
982 dprintk("RPC: sk_buff copy failed\n");
ce7c252a 983 spin_lock(&xprt->recv_lock);
729749bb 984 goto out_unpin;
176e21ee
CL		 985 }
986
ce7c252a 987 spin_lock(&xprt->recv_lock);
176e21ee 988 xprt_complete_rqst(task, copied);
729749bb
TM		 989out_unpin:
990 xprt_unpin_rqst(rovr);
176e21ee 991 out_unlock:
ce7c252a 992 spin_unlock(&xprt->recv_lock);
a2648094
TM		 993}
994
995static void xs_local_data_receive(struct sock_xprt *transport)
996{
997 struct sk_buff *skb;
998 struct sock *sk;
999 int err;
1000
0af3442a 1001restart:
a2648094
TM		 1002 mutex_lock(&transport->recv_mutex);
1003 sk = transport->inet;
1004 if (sk == NULL)
1005 goto out;
1006 for (;;) {
1007 skb = skb_recv_datagram(sk, 0, 1, &err);
42d42a5b
TM		 1008 if (skb != NULL) {
1009 xs_local_data_read_skb(&transport->xprt, sk, skb);
1010 skb_free_datagram(sk, skb);
1011 continue;
1012 }
1013 if (!test_and_clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
a2648094 1014 break;
0af3442a
TM		 1015 if (need_resched()) {
1016 mutex_unlock(&transport->recv_mutex);
1017 cond_resched();
1018 goto restart;
1019 }
a2648094
TM		 1020 }
1021out:
1022 mutex_unlock(&transport->recv_mutex);
1023}
1024
1025static void xs_local_data_receive_workfn(struct work_struct *work)
1026{
1027 struct sock_xprt *transport =
1028 container_of(work, struct sock_xprt, recv_worker);
1029 xs_local_data_receive(transport);
176e21ee
CL		 1030}
1031
9903cd1c 1032/**
f9b2ee71
TM		 1033 * xs_udp_data_read_skb - receive callback for UDP sockets
1034 * @xprt: transport
1035 * @sk: socket
1036 * @skb: skbuff
9903cd1c 1037 *
a246b010 1038 */
f9b2ee71
TM		 1039static void xs_udp_data_read_skb(struct rpc_xprt *xprt,
1040 struct sock *sk,
1041 struct sk_buff *skb)
a246b010 1042{
9903cd1c 1043 struct rpc_task *task;
a246b010 1044 struct rpc_rqst *rovr;
f9b2ee71 1045 int repsize, copied;
d8ed029d
AD		 1046 u32 _xid;
1047 __be32 *xp;
a246b010 1048
1da8c681 1049 repsize = skb->len;
a246b010 1050 if (repsize < 4) {
46121cf7 1051 dprintk("RPC: impossible RPC reply size %d!\n", repsize);
f9b2ee71 1052 return;
a246b010
CL		 1053 }
1054
1055 /* Copy the XID from the skb... */
1da8c681 1056 xp = skb_header_pointer(skb, 0, sizeof(_xid), &_xid);
a246b010 1057 if (xp == NULL)
f9b2ee71 1058 return;
a246b010
CL		 1059
1060 /* Look up and lock the request corresponding to the given XID */
ce7c252a 1061 spin_lock(&xprt->recv_lock);
a246b010
CL		 1062 rovr = xprt_lookup_rqst(xprt, *xp);
1063 if (!rovr)
1064 goto out_unlock;
729749bb 1065 xprt_pin_rqst(rovr);
ecd465ee 1066 xprt_update_rtt(rovr->rq_task);
ce7c252a 1067 spin_unlock(&xprt->recv_lock);
a246b010
CL		 1068 task = rovr->rq_task;
1069
a246b010
CL		 1070 if ((copied = rovr->rq_private_buf.buflen) > repsize)
1071 copied = repsize;
1072
1073 /* Suck it into the iovec, verify checksum if not done by hw. */
1781f7f5 1074 if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
ce7c252a 1075 spin_lock(&xprt->recv_lock);
0afa6b44 1076 __UDPX_INC_STATS(sk, UDP_MIB_INERRORS);
729749bb 1077 goto out_unpin;
1781f7f5
HX		 1078 }
1079
a246b010 1080
729749bb 1081 spin_lock_bh(&xprt->transport_lock);
6a24dfb6 1082 xprt_adjust_cwnd(xprt, task, copied);
ce7c252a
TM		 1083 spin_unlock_bh(&xprt->transport_lock);
1084 spin_lock(&xprt->recv_lock);
1570c1e4 1085 xprt_complete_rqst(task, copied);
0afa6b44 1086 __UDPX_INC_STATS(sk, UDP_MIB_INDATAGRAMS);
729749bb
TM		 1087out_unpin:
1088 xprt_unpin_rqst(rovr);
a246b010 1089 out_unlock:
ce7c252a 1090 spin_unlock(&xprt->recv_lock);
f9b2ee71
TM		 1091}
1092
1093static void xs_udp_data_receive(struct sock_xprt *transport)
1094{
1095 struct sk_buff *skb;
1096 struct sock *sk;
1097 int err;
1098
0af3442a 1099restart:
f9b2ee71
TM		 1100 mutex_lock(&transport->recv_mutex);
1101 sk = transport->inet;
1102 if (sk == NULL)
1103 goto out;
1104 for (;;) {
7c13f97f 1105 skb = skb_recv_udp(sk, 0, 1, &err);
42d42a5b
TM		 1106 if (skb != NULL) {
1107 xs_udp_data_read_skb(&transport->xprt, sk, skb);
850cbadd 1108 consume_skb(skb);
42d42a5b
TM		 1109 continue;
1110 }
1111 if (!test_and_clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
f9b2ee71 1112 break;
0af3442a
TM		 1113 if (need_resched()) {
1114 mutex_unlock(&transport->recv_mutex);
1115 cond_resched();
1116 goto restart;
1117 }
f9b2ee71
TM		 1118 }
1119out:
1120 mutex_unlock(&transport->recv_mutex);
1121}
1122
1123static void xs_udp_data_receive_workfn(struct work_struct *work)
1124{
1125 struct sock_xprt *transport =
1126 container_of(work, struct sock_xprt, recv_worker);
1127 xs_udp_data_receive(transport);
1128}
1129
1130/**
1131 * xs_data_ready - "data ready" callback for UDP sockets
1132 * @sk: socket with data to read
1133 *
1134 */
1135static void xs_data_ready(struct sock *sk)
1136{
1137 struct rpc_xprt *xprt;
1138
1139 read_lock_bh(&sk->sk_callback_lock);
1140 dprintk("RPC: xs_data_ready...\n");
1141 xprt = xprt_from_sock(sk);
1142 if (xprt != NULL) {
1143 struct sock_xprt *transport = container_of(xprt,
1144 struct sock_xprt, xprt);
5157b956
TM		 1145 transport->old_data_ready(sk);
1146 /* Any data means we had a useful conversation, so
1147 * then we don't need to delay the next reconnect
1148 */
1149 if (xprt->reestablish_timeout)
1150 xprt->reestablish_timeout = 0;
42d42a5b 1151 if (!test_and_set_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
40a5f1b1 1152 queue_work(xprtiod_workqueue, &transport->recv_worker);
f9b2ee71 1153 }
f064af1e 1154 read_unlock_bh(&sk->sk_callback_lock);
a246b010
CL		 1155}
1156
a519fc7a
TM		 1157/*
1158 * Helper function to force a TCP close if the server is sending
1159 * junk and/or it has put us in CLOSE_WAIT
1160 */
1161static void xs_tcp_force_close(struct rpc_xprt *xprt)
1162{
a519fc7a
TM		 1163 xprt_force_disconnect(xprt);
1164}
1165
dd456471 1166static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
a246b010 1167{
51971139 1168 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
a246b010
CL		 1169 size_t len, used;
1170 char *p;
1171
51971139
CL		 1172 p = ((char *) &transport->tcp_fraghdr) + transport->tcp_offset;
1173 len = sizeof(transport->tcp_fraghdr) - transport->tcp_offset;
9d292316 1174 used = xdr_skb_read_bits(desc, p, len);
51971139 1175 transport->tcp_offset += used;
a246b010
CL		 1176 if (used != len)
1177 return;
808012fb 1178
51971139
CL		 1179 transport->tcp_reclen = ntohl(transport->tcp_fraghdr);
1180 if (transport->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
e136d092 1181 transport->tcp_flags |= TCP_RCV_LAST_FRAG;
a246b010 1182 else
e136d092 1183 transport->tcp_flags &= ~TCP_RCV_LAST_FRAG;
51971139 1184 transport->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
808012fb 1185
e136d092 1186 transport->tcp_flags &= ~TCP_RCV_COPY_FRAGHDR;
51971139 1187 transport->tcp_offset = 0;
808012fb 1188
a246b010 1189 /* Sanity check of the record length */
18dca02a 1190 if (unlikely(transport->tcp_reclen < 8)) {
46121cf7 1191 dprintk("RPC: invalid TCP record fragment length\n");
a519fc7a 1192 xs_tcp_force_close(xprt);
9903cd1c 1193 return;
a246b010 1194 }
46121cf7 1195 dprintk("RPC: reading TCP record fragment of length %d\n",
51971139 1196 transport->tcp_reclen);
a246b010
CL		 1197}
1198
51971139 1199static void xs_tcp_check_fraghdr(struct sock_xprt *transport)
a246b010 1200{
51971139 1201 if (transport->tcp_offset == transport->tcp_reclen) {
e136d092 1202 transport->tcp_flags |= TCP_RCV_COPY_FRAGHDR;
51971139 1203 transport->tcp_offset = 0;
e136d092
CL		 1204 if (transport->tcp_flags & TCP_RCV_LAST_FRAG) {
1205 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1206 transport->tcp_flags |= TCP_RCV_COPY_XID;
51971139 1207 transport->tcp_copied = 0;
a246b010
CL		 1208 }
1209 }
1210}
1211
dd456471 1212static inline void xs_tcp_read_xid(struct sock_xprt *transport, struct xdr_skb_reader *desc)
a246b010
CL		 1213{
1214 size_t len, used;
1215 char *p;
1216
51971139 1217 len = sizeof(transport->tcp_xid) - transport->tcp_offset;
5b5e0928 1218 dprintk("RPC: reading XID (%zu bytes)\n", len);
51971139 1219 p = ((char *) &transport->tcp_xid) + transport->tcp_offset;
9d292316 1220 used = xdr_skb_read_bits(desc, p, len);
51971139 1221 transport->tcp_offset += used;
a246b010
CL		 1222 if (used != len)
1223 return;
e136d092 1224 transport->tcp_flags &= ~TCP_RCV_COPY_XID;
f4a2e418 1225 transport->tcp_flags |= TCP_RCV_READ_CALLDIR;
51971139 1226 transport->tcp_copied = 4;
18dca02a
RL		 1227 dprintk("RPC: reading %s XID %08x\n",
1228 (transport->tcp_flags & TCP_RPC_REPLY) ? "reply for"
1229 : "request with",
51971139
CL		 1230 ntohl(transport->tcp_xid));
1231 xs_tcp_check_fraghdr(transport);
a246b010
CL		 1232}
1233
18dca02a
RL		 1234static inline void xs_tcp_read_calldir(struct sock_xprt *transport,
1235 struct xdr_skb_reader *desc)
a246b010 1236{
18dca02a
RL		 1237 size_t len, used;
1238 u32 offset;
b76ce561 1239 char *p;
18dca02a
RL		 1240
1241 /*
1242 * We want transport->tcp_offset to be 8 at the end of this routine
1243 * (4 bytes for the xid and 4 bytes for the call/reply flag).
1244 * When this function is called for the first time,
1245 * transport->tcp_offset is 4 (after having already read the xid).
1246 */
1247 offset = transport->tcp_offset - sizeof(transport->tcp_xid);
b76ce561 1248 len = sizeof(transport->tcp_calldir) - offset;
5b5e0928 1249 dprintk("RPC: reading CALL/REPLY flag (%zu bytes)\n", len);
b76ce561
TM		 1250 p = ((char *) &transport->tcp_calldir) + offset;
1251 used = xdr_skb_read_bits(desc, p, len);
18dca02a
RL		 1252 transport->tcp_offset += used;
1253 if (used != len)
1254 return;
f4a2e418 1255 transport->tcp_flags &= ~TCP_RCV_READ_CALLDIR;
f4a2e418
RL		 1256 /*
1257 * We don't yet have the XDR buffer, so we will write the calldir
1258 * out after we get the buffer from the 'struct rpc_rqst'
1259 */
b76ce561
TM		 1260 switch (ntohl(transport->tcp_calldir)) {
1261 case RPC_REPLY:
1262 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1263 transport->tcp_flags |= TCP_RCV_COPY_DATA;
18dca02a 1264 transport->tcp_flags |= TCP_RPC_REPLY;
b76ce561
TM		 1265 break;
1266 case RPC_CALL:
1267 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1268 transport->tcp_flags |= TCP_RCV_COPY_DATA;
18dca02a 1269 transport->tcp_flags &= ~TCP_RPC_REPLY;
b76ce561
TM		 1270 break;
1271 default:
1272 dprintk("RPC: invalid request message type\n");
a519fc7a 1273 xs_tcp_force_close(&transport->xprt);
b76ce561 1274 }
18dca02a
RL		 1275 xs_tcp_check_fraghdr(transport);
1276}
1277
44b98efd
RL		 1278static inline void xs_tcp_read_common(struct rpc_xprt *xprt,
1279 struct xdr_skb_reader *desc,
1280 struct rpc_rqst *req)
a246b010 1281{
44b98efd
RL		 1282 struct sock_xprt *transport =
1283 container_of(xprt, struct sock_xprt, xprt);
a246b010
CL		 1284 struct xdr_buf *rcvbuf;
1285 size_t len;
1286 ssize_t r;
1287
a246b010 1288 rcvbuf = &req->rq_private_buf;
f4a2e418
RL		 1289
1290 if (transport->tcp_flags & TCP_RCV_COPY_CALLDIR) {
1291 /*
1292 * Save the RPC direction in the XDR buffer
1293 */
f4a2e418 1294 memcpy(rcvbuf->head[0].iov_base + transport->tcp_copied,
b76ce561
TM		 1295 &transport->tcp_calldir,
1296 sizeof(transport->tcp_calldir));
1297 transport->tcp_copied += sizeof(transport->tcp_calldir);
f4a2e418 1298 transport->tcp_flags &= ~TCP_RCV_COPY_CALLDIR;
a246b010
CL		 1299 }
1300
a246b010 1301 len = desc->count;
040249df
TM		 1302 if (len > transport->tcp_reclen - transport->tcp_offset)
1303 desc->count = transport->tcp_reclen - transport->tcp_offset;
1304 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
9d292316 1305 desc, xdr_skb_read_bits);
a246b010 1306
040249df 1307 if (desc->count) {
a246b010
CL		 1308 /* Error when copying to the receive buffer,
1309 * usually because we weren't able to allocate
1310 * additional buffer pages. All we can do now
e136d092 1311 * is turn off TCP_RCV_COPY_DATA, so the request
a246b010
CL		 1312 * will not receive any additional updates,
1313 * and time out.
1314 * Any remaining data from this record will
1315 * be discarded.
1316 */
e136d092 1317 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
46121cf7 1318 dprintk("RPC: XID %08x truncated request\n",
51971139 1319 ntohl(transport->tcp_xid));
46121cf7
CL		 1320 dprintk("RPC: xprt = %p, tcp_copied = %lu, "
1321 "tcp_offset = %u, tcp_reclen = %u\n",
1322 xprt, transport->tcp_copied,
1323 transport->tcp_offset, transport->tcp_reclen);
44b98efd 1324 return;
a246b010
CL		 1325 }
1326
040249df
TM		 1327 transport->tcp_copied += r;
1328 transport->tcp_offset += r;
1329 desc->count = len - r;
1330
5b5e0928 1331 dprintk("RPC: XID %08x read %zd bytes\n",
51971139 1332 ntohl(transport->tcp_xid), r);
46121cf7
CL		 1333 dprintk("RPC: xprt = %p, tcp_copied = %lu, tcp_offset = %u, "
1334 "tcp_reclen = %u\n", xprt, transport->tcp_copied,
1335 transport->tcp_offset, transport->tcp_reclen);
51971139
CL		 1336
1337 if (transport->tcp_copied == req->rq_private_buf.buflen)
e136d092 1338 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
51971139 1339 else if (transport->tcp_offset == transport->tcp_reclen) {
e136d092
CL		 1340 if (transport->tcp_flags & TCP_RCV_LAST_FRAG)
1341 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
a246b010 1342 }
44b98efd
RL		 1343}
1344
1345/*
1346 * Finds the request corresponding to the RPC xid and invokes the common
1347 * tcp read code to read the data.
1348 */
1349static inline int xs_tcp_read_reply(struct rpc_xprt *xprt,
1350 struct xdr_skb_reader *desc)
1351{
1352 struct sock_xprt *transport =
1353 container_of(xprt, struct sock_xprt, xprt);
1354 struct rpc_rqst *req;
1355
1356 dprintk("RPC: read reply XID %08x\n", ntohl(transport->tcp_xid));
1357
1358 /* Find and lock the request corresponding to this xid */
ce7c252a 1359 spin_lock(&xprt->recv_lock);
44b98efd
RL		 1360 req = xprt_lookup_rqst(xprt, transport->tcp_xid);
1361 if (!req) {
1362 dprintk("RPC: XID %08x request not found!\n",
1363 ntohl(transport->tcp_xid));
ce7c252a 1364 spin_unlock(&xprt->recv_lock);
44b98efd
RL		 1365 return -1;
1366 }
729749bb 1367 xprt_pin_rqst(req);
ce7c252a 1368 spin_unlock(&xprt->recv_lock);
44b98efd
RL		 1369
1370 xs_tcp_read_common(xprt, desc, req);
1371
ce7c252a 1372 spin_lock(&xprt->recv_lock);
e136d092 1373 if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
51971139 1374 xprt_complete_rqst(req->rq_task, transport->tcp_copied);
729749bb 1375 xprt_unpin_rqst(req);
ce7c252a 1376 spin_unlock(&xprt->recv_lock);
44b98efd
RL		 1377 return 0;
1378}
1379
9e00abc3 1380#if defined(CONFIG_SUNRPC_BACKCHANNEL)
44b98efd
RL		 1381/*
1382 * Obtains an rpc_rqst previously allocated and invokes the common
1383 * tcp read code to read the data. The result is placed in the callback
1384 * queue.
1385 * If we're unable to obtain the rpc_rqst we schedule the closing of the
1386 * connection and return -1.
1387 */
2ea24497 1388static int xs_tcp_read_callback(struct rpc_xprt *xprt,
44b98efd
RL		 1389 struct xdr_skb_reader *desc)
1390{
1391 struct sock_xprt *transport =
1392 container_of(xprt, struct sock_xprt, xprt);
1393 struct rpc_rqst *req;
1394
c89091c8 1395 /* Look up the request corresponding to the given XID */
2ea24497 1396 req = xprt_lookup_bc_request(xprt, transport->tcp_xid);
44b98efd
RL		 1397 if (req == NULL) {
1398 printk(KERN_WARNING "Callback slot table overflowed\n");
1399 xprt_force_disconnect(xprt);
1400 return -1;
1401 }
1402
44b98efd
RL		 1403 dprintk("RPC: read callback XID %08x\n", ntohl(req->rq_xid));
1404 xs_tcp_read_common(xprt, desc, req);
1405
2ea24497
TM		 1406 if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1407 xprt_complete_bc_request(req, transport->tcp_copied);
44b98efd
RL		 1408
1409 return 0;
1410}
1411
1412static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1413 struct xdr_skb_reader *desc)
1414{
1415 struct sock_xprt *transport =
1416 container_of(xprt, struct sock_xprt, xprt);
1417
1418 return (transport->tcp_flags & TCP_RPC_REPLY) ?
1419 xs_tcp_read_reply(xprt, desc) :
1420 xs_tcp_read_callback(xprt, desc);
1421}
76566773
CL		 1422
1423static int xs_tcp_bc_up(struct svc_serv *serv, struct net *net)
1424{
1425 int ret;
1426
1427 ret = svc_create_xprt(serv, "tcp-bc", net, PF_INET, 0,
1428 SVC_SOCK_ANONYMOUS);
1429 if (ret < 0)
1430 return ret;
1431 return 0;
1432}
6b26cc8c
CL		 1433
1434static size_t xs_tcp_bc_maxpayload(struct rpc_xprt *xprt)
1435{
1436 return PAGE_SIZE;
1437}
44b98efd
RL		 1438#else
1439static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1440 struct xdr_skb_reader *desc)
1441{
1442 return xs_tcp_read_reply(xprt, desc);
1443}
9e00abc3 1444#endif /* CONFIG_SUNRPC_BACKCHANNEL */
44b98efd
RL		 1445
1446/*
1447 * Read data off the transport. This can be either an RPC_CALL or an
1448 * RPC_REPLY. Relay the processing to helper functions.
1449 */
1450static void xs_tcp_read_data(struct rpc_xprt *xprt,
1451 struct xdr_skb_reader *desc)
1452{
1453 struct sock_xprt *transport =
1454 container_of(xprt, struct sock_xprt, xprt);
1455
1456 if (_xs_tcp_read_data(xprt, desc) == 0)
1457 xs_tcp_check_fraghdr(transport);
1458 else {
1459 /*
1460 * The transport_lock protects the request handling.
1461 * There's no need to hold it to update the tcp_flags.
1462 */
1463 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1464 }
a246b010
CL		 1465}
1466
dd456471 1467static inline void xs_tcp_read_discard(struct sock_xprt *transport, struct xdr_skb_reader *desc)
a246b010
CL		 1468{
1469 size_t len;
1470
51971139 1471 len = transport->tcp_reclen - transport->tcp_offset;
a246b010
CL		 1472 if (len > desc->count)
1473 len = desc->count;
1474 desc->count -= len;
1475 desc->offset += len;
51971139 1476 transport->tcp_offset += len;
5b5e0928 1477 dprintk("RPC: discarded %zu bytes\n", len);
51971139 1478 xs_tcp_check_fraghdr(transport);
a246b010
CL		 1479}
1480
9903cd1c 1481static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
a246b010
CL		 1482{
1483 struct rpc_xprt *xprt = rd_desc->arg.data;
51971139 1484 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
dd456471 1485 struct xdr_skb_reader desc = {
a246b010
CL		 1486 .skb = skb,
1487 .offset = offset,
1488 .count = len,
9903cd1c 1489 };
3d188805 1490 size_t ret;
a246b010 1491
46121cf7 1492 dprintk("RPC: xs_tcp_data_recv started\n");
a246b010 1493 do {
1a867a08 1494 trace_xs_tcp_data_recv(transport);
a246b010
CL		 1495 /* Read in a new fragment marker if necessary */
1496 /* Can we ever really expect to get completely empty fragments? */
e136d092 1497 if (transport->tcp_flags & TCP_RCV_COPY_FRAGHDR) {
9903cd1c 1498 xs_tcp_read_fraghdr(xprt, &desc);
a246b010
CL		 1499 continue;
1500 }
1501 /* Read in the xid if necessary */
e136d092 1502 if (transport->tcp_flags & TCP_RCV_COPY_XID) {
51971139 1503 xs_tcp_read_xid(transport, &desc);
a246b010
CL		 1504 continue;
1505 }
18dca02a 1506 /* Read in the call/reply flag */
f4a2e418 1507 if (transport->tcp_flags & TCP_RCV_READ_CALLDIR) {
18dca02a
RL		 1508 xs_tcp_read_calldir(transport, &desc);
1509 continue;
1510 }
a246b010 1511 /* Read in the request data */
e136d092 1512 if (transport->tcp_flags & TCP_RCV_COPY_DATA) {
44b98efd 1513 xs_tcp_read_data(xprt, &desc);
a246b010
CL		 1514 continue;
1515 }
1516 /* Skip over any trailing bytes on short reads */
51971139 1517 xs_tcp_read_discard(transport, &desc);
a246b010 1518 } while (desc.count);
3d188805
TM		 1519 ret = len - desc.count;
1520 if (ret < rd_desc->count)
1521 rd_desc->count -= ret;
1522 else
1523 rd_desc->count = 0;
1a867a08 1524 trace_xs_tcp_data_recv(transport);
46121cf7 1525 dprintk("RPC: xs_tcp_data_recv done\n");
3d188805 1526 return ret;
a246b010
CL		 1527}
1528
66d7a56a
TM		 1529static void xs_tcp_data_receive(struct sock_xprt *transport)
1530{
1531 struct rpc_xprt *xprt = &transport->xprt;
1532 struct sock *sk;
1533 read_descriptor_t rd_desc = {
66d7a56a
TM		 1534 .arg.data = xprt,
1535 };
1536 unsigned long total = 0;
1537 int read = 0;
1538
0af3442a 1539restart:
edc1b01c 1540 mutex_lock(&transport->recv_mutex);
66d7a56a 1541 sk = transport->inet;
edc1b01c
TM		 1542 if (sk == NULL)
1543 goto out;
66d7a56a
TM		 1544
1545 /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
0af3442a 1546 for (;;) {
3d188805 1547 rd_desc.count = RPC_TCP_READ_CHUNK_SZ;
edc1b01c 1548 lock_sock(sk);
66d7a56a 1549 read = tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
3d188805 1550 if (rd_desc.count != 0 || read < 0) {
42d42a5b
TM		 1551 clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state);
1552 release_sock(sk);
8d6f97d6 1553 break;
42d42a5b 1554 }
8d6f97d6
TM		 1555 release_sock(sk);
1556 total += read;
0af3442a
TM		 1557 if (need_resched()) {
1558 mutex_unlock(&transport->recv_mutex);
1559 cond_resched();
1560 goto restart;
1561 }
66d7a56a 1562 }
8d6f97d6
TM		 1563 if (test_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
1564 queue_work(xprtiod_workqueue, &transport->recv_worker);
edc1b01c
TM		 1565out:
1566 mutex_unlock(&transport->recv_mutex);
66d7a56a
TM		 1567 trace_xs_tcp_data_ready(xprt, read, total);
1568}
1569
edc1b01c
TM		 1570static void xs_tcp_data_receive_workfn(struct work_struct *work)
1571{
1572 struct sock_xprt *transport =
1573 container_of(work, struct sock_xprt, recv_worker);
1574 xs_tcp_data_receive(transport);
1575}
1576
9903cd1c
CL		 1577/**
1578 * xs_tcp_state_change - callback to handle TCP socket state changes
1579 * @sk: socket whose state has changed
1580 *
1581 */
1582static void xs_tcp_state_change(struct sock *sk)
a246b010 1583{
9903cd1c 1584 struct rpc_xprt *xprt;
0fdea1e8 1585 struct sock_xprt *transport;
a246b010 1586
f064af1e 1587 read_lock_bh(&sk->sk_callback_lock);
a246b010
CL		 1588 if (!(xprt = xprt_from_sock(sk)))
1589 goto out;
46121cf7 1590 dprintk("RPC: xs_tcp_state_change client %p...\n", xprt);
669502ff 1591 dprintk("RPC: state %x conn %d dead %d zapped %d sk_shutdown %d\n",
46121cf7
CL		 1592 sk->sk_state, xprt_connected(xprt),
1593 sock_flag(sk, SOCK_DEAD),
669502ff
AC		 1594 sock_flag(sk, SOCK_ZAPPED),
1595 sk->sk_shutdown);
a246b010 1596
0fdea1e8 1597 transport = container_of(xprt, struct sock_xprt, xprt);
40b5ea0c 1598 trace_rpc_socket_state_change(xprt, sk->sk_socket);
a246b010
CL		 1599 switch (sk->sk_state) {
1600 case TCP_ESTABLISHED:
f064af1e 1601 spin_lock(&xprt->transport_lock);
a246b010 1602 if (!xprt_test_and_set_connected(xprt)) {
51971139 1603
a246b010 1604 /* Reset TCP record info */
51971139
CL		 1605 transport->tcp_offset = 0;
1606 transport->tcp_reclen = 0;
1607 transport->tcp_copied = 0;
e136d092
CL		 1608 transport->tcp_flags =
1609 TCP_RCV_COPY_FRAGHDR | TCP_RCV_COPY_XID;
8b71798c 1610 xprt->connect_cookie++;
0fdea1e8
TM		 1611 clear_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
1612 xprt_clear_connecting(xprt);
51971139 1613
2a491991 1614 xprt_wake_pending_tasks(xprt, -EAGAIN);
a246b010 1615 }
f064af1e 1616 spin_unlock(&xprt->transport_lock);
a246b010 1617 break;
3b948ae5
TM		 1618 case TCP_FIN_WAIT1:
1619 /* The client initiated a shutdown of the socket */
7c1d71cf 1620 xprt->connect_cookie++;
663b8858 1621 xprt->reestablish_timeout = 0;
3b948ae5 1622 set_bit(XPRT_CLOSING, &xprt->state);
4e857c58 1623 smp_mb__before_atomic();
3b948ae5 1624 clear_bit(XPRT_CONNECTED, &xprt->state);
ef803670 1625 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
4e857c58 1626 smp_mb__after_atomic();
a246b010 1627 break;
632e3bdc 1628 case TCP_CLOSE_WAIT:
3b948ae5 1629 /* The server initiated a shutdown of the socket */
7c1d71cf 1630 xprt->connect_cookie++;
d0bea455 1631 clear_bit(XPRT_CONNECTED, &xprt->state);
a519fc7a 1632 xs_tcp_force_close(xprt);
e9d47639 1633 /* fall through */
663b8858
TM		 1634 case TCP_CLOSING:
1635 /*
1636 * If the server closed down the connection, make sure that
1637 * we back off before reconnecting
1638 */
1639 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1640 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
3b948ae5
TM		 1641 break;
1642 case TCP_LAST_ACK:
670f9457 1643 set_bit(XPRT_CLOSING, &xprt->state);
4e857c58 1644 smp_mb__before_atomic();
3b948ae5 1645 clear_bit(XPRT_CONNECTED, &xprt->state);
4e857c58 1646 smp_mb__after_atomic();
3b948ae5
TM		 1647 break;
1648 case TCP_CLOSE:
0fdea1e8
TM		 1649 if (test_and_clear_bit(XPRT_SOCK_CONNECTING,
1650 &transport->sock_state))
1651 xprt_clear_connecting(xprt);
9b30889c 1652 clear_bit(XPRT_CLOSING, &xprt->state);
3ffbc1d6
N		 1653 if (sk->sk_err)
1654 xprt_wake_pending_tasks(xprt, -sk->sk_err);
9b30889c
TM		 1655 /* Trigger the socket release */
1656 xs_tcp_force_close(xprt);
a246b010
CL		 1657 }
1658 out:
f064af1e 1659 read_unlock_bh(&sk->sk_callback_lock);
a246b010
CL		 1660}
1661
1f0fa154
IJ		 1662static void xs_write_space(struct sock *sk)
1663{
13331a55 1664 struct socket_wq *wq;
1f0fa154
IJ		 1665 struct rpc_xprt *xprt;
1666
13331a55 1667 if (!sk->sk_socket)
1f0fa154 1668 return;
13331a55 1669 clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1f0fa154
IJ		 1670
1671 if (unlikely(!(xprt = xprt_from_sock(sk))))
1672 return;
13331a55
TM		 1673 rcu_read_lock();
1674 wq = rcu_dereference(sk->sk_wq);
1675 if (!wq || test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &wq->flags) == 0)
1676 goto out;
1f0fa154
IJ		 1677
1678 xprt_write_space(xprt);
13331a55
TM		 1679out:
1680 rcu_read_unlock();
1f0fa154
IJ		 1681}
1682
9903cd1c 1683/**
c7b2cae8
CL		 1684 * xs_udp_write_space - callback invoked when socket buffer space
1685 * becomes available
9903cd1c
CL		 1686 * @sk: socket whose state has changed
1687 *
a246b010
CL		 1688 * Called when more output buffer space is available for this socket.
1689 * We try not to wake our writers until they can make "significant"
c7b2cae8 1690 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
a246b010
CL		 1691 * with a bunch of small requests.
1692 */
c7b2cae8 1693static void xs_udp_write_space(struct sock *sk)
a246b010 1694{
f064af1e 1695 read_lock_bh(&sk->sk_callback_lock);
a246b010 1696
c7b2cae8 1697 /* from net/core/sock.c:sock_def_write_space */
1f0fa154
IJ		 1698 if (sock_writeable(sk))
1699 xs_write_space(sk);
a246b010 1700
f064af1e 1701 read_unlock_bh(&sk->sk_callback_lock);
c7b2cae8 1702}
a246b010 1703
c7b2cae8
CL		 1704/**
1705 * xs_tcp_write_space - callback invoked when socket buffer space
1706 * becomes available
1707 * @sk: socket whose state has changed
1708 *
1709 * Called when more output buffer space is available for this socket.
1710 * We try not to wake our writers until they can make "significant"
1711 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1712 * with a bunch of small requests.
1713 */
1714static void xs_tcp_write_space(struct sock *sk)
1715{
f064af1e 1716 read_lock_bh(&sk->sk_callback_lock);
c7b2cae8
CL		 1717
1718 /* from net/core/stream.c:sk_stream_write_space */
64dc6130 1719 if (sk_stream_is_writeable(sk))
1f0fa154 1720 xs_write_space(sk);
b6ddf64f 1721
f064af1e 1722 read_unlock_bh(&sk->sk_callback_lock);
a246b010
CL		 1723}
1724
470056c2 1725static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
a246b010 1726{
ee0ac0c2
CL		 1727 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1728 struct sock *sk = transport->inet;
a246b010 1729
7c6e066e 1730 if (transport->rcvsize) {
a246b010 1731 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
7c6e066e 1732 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
a246b010 1733 }
7c6e066e 1734 if (transport->sndsize) {
a246b010 1735 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
7c6e066e 1736 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
a246b010
CL		 1737 sk->sk_write_space(sk);
1738 }
1739}
1740
43118c29 1741/**
470056c2 1742 * xs_udp_set_buffer_size - set send and receive limits
43118c29 1743 * @xprt: generic transport
470056c2
CL		 1744 * @sndsize: requested size of send buffer, in bytes
1745 * @rcvsize: requested size of receive buffer, in bytes
43118c29 1746 *
470056c2 1747 * Set socket send and receive buffer size limits.
43118c29 1748 */
470056c2 1749static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
43118c29 1750{
7c6e066e
CL		 1751 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1752
1753 transport->sndsize = 0;
470056c2 1754 if (sndsize)
7c6e066e
CL		 1755 transport->sndsize = sndsize + 1024;
1756 transport->rcvsize = 0;
470056c2 1757 if (rcvsize)
7c6e066e 1758 transport->rcvsize = rcvsize + 1024;
470056c2
CL		 1759
1760 xs_udp_do_set_buffer_size(xprt);
43118c29
CL		 1761}
1762
46c0ee8b
CL		 1763/**
1764 * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1765 * @task: task that timed out
1766 *
1767 * Adjust the congestion window after a retransmit timeout has occurred.
1768 */
6a24dfb6 1769static void xs_udp_timer(struct rpc_xprt *xprt, struct rpc_task *task)
46c0ee8b 1770{
b977b644 1771 spin_lock_bh(&xprt->transport_lock);
6a24dfb6 1772 xprt_adjust_cwnd(xprt, task, -ETIMEDOUT);
b977b644 1773 spin_unlock_bh(&xprt->transport_lock);
46c0ee8b
CL		 1774}
1775
b85d8806
CL		 1776static unsigned short xs_get_random_port(void)
1777{
5d71899a 1778 unsigned short range = xprt_max_resvport - xprt_min_resvport + 1;
63862b5b 1779 unsigned short rand = (unsigned short) prandom_u32() % range;
b85d8806
CL		 1780 return rand + xprt_min_resvport;
1781}
1782
4dda9c8a
TM		 1783/**
1784 * xs_set_reuseaddr_port - set the socket's port and address reuse options
1785 * @sock: socket
1786 *
1787 * Note that this function has to be called on all sockets that share the
1788 * same port, and it must be called before binding.
1789 */
1790static void xs_sock_set_reuseport(struct socket *sock)
1791{
402e23b4 1792 int opt = 1;
4dda9c8a 1793
402e23b4
TM		 1794 kernel_setsockopt(sock, SOL_SOCKET, SO_REUSEPORT,
1795 (char *)&opt, sizeof(opt));
4dda9c8a
TM		 1796}
1797
1798static unsigned short xs_sock_getport(struct socket *sock)
1799{
1800 struct sockaddr_storage buf;
4dda9c8a
TM		 1801 unsigned short port = 0;
1802
9b2c45d4 1803 if (kernel_getsockname(sock, (struct sockaddr *)&buf) < 0)
4dda9c8a
TM		 1804 goto out;
1805 switch (buf.ss_family) {
1806 case AF_INET6:
1807 port = ntohs(((struct sockaddr_in6 *)&buf)->sin6_port);
1808 break;
1809 case AF_INET:
1810 port = ntohs(((struct sockaddr_in *)&buf)->sin_port);
1811 }
1812out:
1813 return port;
1814}
1815
92200412
CL		 1816/**
1817 * xs_set_port - reset the port number in the remote endpoint address
1818 * @xprt: generic transport
1819 * @port: new port number
1820 *
1821 */
1822static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1823{
46121cf7 1824 dprintk("RPC: setting port for xprt %p to %u\n", xprt, port);
c4efcb1d 1825
9dc3b095
CL		 1826 rpc_set_port(xs_addr(xprt), port);
1827 xs_update_peer_port(xprt);
92200412
CL		 1828}
1829
4dda9c8a
TM		 1830static void xs_set_srcport(struct sock_xprt *transport, struct socket *sock)
1831{
1832 if (transport->srcport == 0)
1833 transport->srcport = xs_sock_getport(sock);
1834}
1835
5d4ec932 1836static unsigned short xs_get_srcport(struct sock_xprt *transport)
67a391d7 1837{
fbfffbd5 1838 unsigned short port = transport->srcport;
67a391d7
TM		 1839
1840 if (port == 0 && transport->xprt.resvport)
1841 port = xs_get_random_port();
1842 return port;
1843}
1844
baaf4e48 1845static unsigned short xs_next_srcport(struct sock_xprt *transport, unsigned short port)
67a391d7 1846{
fbfffbd5
CL		 1847 if (transport->srcport != 0)
1848 transport->srcport = 0;
67a391d7
TM		 1849 if (!transport->xprt.resvport)
1850 return 0;
1851 if (port <= xprt_min_resvport || port > xprt_max_resvport)
1852 return xprt_max_resvport;
1853 return --port;
1854}
beb59b68 1855static int xs_bind(struct sock_xprt *transport, struct socket *sock)
a246b010 1856{
beb59b68 1857 struct sockaddr_storage myaddr;
67a391d7 1858 int err, nloop = 0;
5d4ec932 1859 unsigned short port = xs_get_srcport(transport);
67a391d7 1860 unsigned short last;
a246b010 1861
0f7a622c
CP		 1862 /*
1863 * If we are asking for any ephemeral port (i.e. port == 0 &&
1864 * transport->xprt.resvport == 0), don't bind. Let the local
1865 * port selection happen implicitly when the socket is used
1866 * (for example at connect time).
1867 *
1868 * This ensures that we can continue to establish TCP
1869 * connections even when all local ephemeral ports are already
1870 * a part of some TCP connection. This makes no difference
1871 * for UDP sockets, but also doens't harm them.
1872 *
1873 * If we're asking for any reserved port (i.e. port == 0 &&
1874 * transport->xprt.resvport == 1) xs_get_srcport above will
1875 * ensure that port is non-zero and we will bind as needed.
1876 */
1877 if (port == 0)
1878 return 0;
1879
beb59b68 1880 memcpy(&myaddr, &transport->srcaddr, transport->xprt.addrlen);
a246b010 1881 do {
beb59b68
PE		 1882 rpc_set_port((struct sockaddr *)&myaddr, port);
1883 err = kernel_bind(sock, (struct sockaddr *)&myaddr,
1884 transport->xprt.addrlen);
a246b010 1885 if (err == 0) {
fbfffbd5 1886 transport->srcport = port;
d3bc9a1d 1887 break;
a246b010 1888 }
67a391d7 1889 last = port;
baaf4e48 1890 port = xs_next_srcport(transport, port);
67a391d7
TM		 1891 if (port > last)
1892 nloop++;
1893 } while (err == -EADDRINUSE && nloop != 2);
90058d37 1894
4232e863 1895 if (myaddr.ss_family == AF_INET)
beb59b68
PE		 1896 dprintk("RPC: %s %pI4:%u: %s (%d)\n", __func__,
1897 &((struct sockaddr_in *)&myaddr)->sin_addr,
1898 port, err ? "failed" : "ok", err);
1899 else
1900 dprintk("RPC: %s %pI6:%u: %s (%d)\n", __func__,
1901 &((struct sockaddr_in6 *)&myaddr)->sin6_addr,
1902 port, err ? "failed" : "ok", err);
a246b010
CL		 1903 return err;
1904}
1905
176e21ee
CL		 1906/*
1907 * We don't support autobind on AF_LOCAL sockets
1908 */
1909static void xs_local_rpcbind(struct rpc_task *task)
1910{
fb43d172 1911 xprt_set_bound(task->tk_xprt);
176e21ee
CL		 1912}
1913
1914static void xs_local_set_port(struct rpc_xprt *xprt, unsigned short port)
1915{
1916}
beb59b68 1917
ed07536e
PZ		 1918#ifdef CONFIG_DEBUG_LOCK_ALLOC
1919static struct lock_class_key xs_key[2];
1920static struct lock_class_key xs_slock_key[2];
1921
176e21ee
CL		 1922static inline void xs_reclassify_socketu(struct socket *sock)
1923{
1924 struct sock *sk = sock->sk;
1925
176e21ee
CL		 1926 sock_lock_init_class_and_name(sk, "slock-AF_LOCAL-RPC",
1927 &xs_slock_key[1], "sk_lock-AF_LOCAL-RPC", &xs_key[1]);
1928}
1929
8945ee5e 1930static inline void xs_reclassify_socket4(struct socket *sock)
ed07536e
PZ		 1931{
1932 struct sock *sk = sock->sk;
8945ee5e 1933
8945ee5e
CL		 1934 sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1935 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1936}
ed07536e 1937
8945ee5e
CL		 1938static inline void xs_reclassify_socket6(struct socket *sock)
1939{
1940 struct sock *sk = sock->sk;
ed07536e 1941
8945ee5e
CL		 1942 sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1943 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
ed07536e 1944}
6bc9638a
PE		 1945
1946static inline void xs_reclassify_socket(int family, struct socket *sock)
1947{
fafc4e1e 1948 if (WARN_ON_ONCE(!sock_allow_reclassification(sock->sk)))
1b7a1819
WAA		 1949 return;
1950
4232e863 1951 switch (family) {
176e21ee
CL		 1952 case AF_LOCAL:
1953 xs_reclassify_socketu(sock);
1954 break;
4232e863 1955 case AF_INET:
6bc9638a 1956 xs_reclassify_socket4(sock);
4232e863
CL		 1957 break;
1958 case AF_INET6:
6bc9638a 1959 xs_reclassify_socket6(sock);
4232e863
CL		 1960 break;
1961 }
6bc9638a 1962}
ed07536e 1963#else
6bc9638a
PE		 1964static inline void xs_reclassify_socket(int family, struct socket *sock)
1965{
1966}
ed07536e
PZ		 1967#endif
1968
93dc41bd
N		 1969static void xs_dummy_setup_socket(struct work_struct *work)
1970{
1971}
1972
6bc9638a 1973static struct socket *xs_create_sock(struct rpc_xprt *xprt,
4dda9c8a
TM		 1974 struct sock_xprt *transport, int family, int type,
1975 int protocol, bool reuseport)
22f79326
PE		 1976{
1977 struct socket *sock;
1978 int err;
1979
6bc9638a 1980 err = __sock_create(xprt->xprt_net, family, type, protocol, &sock, 1);
22f79326
PE		 1981 if (err < 0) {
1982 dprintk("RPC: can't create %d transport socket (%d).\n",
1983 protocol, -err);
1984 goto out;
1985 }
6bc9638a 1986 xs_reclassify_socket(family, sock);
22f79326 1987
4dda9c8a
TM		 1988 if (reuseport)
1989 xs_sock_set_reuseport(sock);
1990
4cea288a
BH		 1991 err = xs_bind(transport, sock);
1992 if (err) {
22f79326
PE		 1993 sock_release(sock);
1994 goto out;
1995 }
1996
1997 return sock;
1998out:
1999 return ERR_PTR(err);
2000}
2001
176e21ee
CL		 2002static int xs_local_finish_connecting(struct rpc_xprt *xprt,
2003 struct socket *sock)
2004{
2005 struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
2006 xprt);
2007
2008 if (!transport->inet) {
2009 struct sock *sk = sock->sk;
2010
2011 write_lock_bh(&sk->sk_callback_lock);
2012
2013 xs_save_old_callbacks(transport, sk);
2014
2015 sk->sk_user_data = xprt;
a2648094 2016 sk->sk_data_ready = xs_data_ready;
176e21ee 2017 sk->sk_write_space = xs_udp_write_space;
b4411457 2018 sock_set_flag(sk, SOCK_FASYNC);
2118071d 2019 sk->sk_error_report = xs_error_report;
c2126157 2020 sk->sk_allocation = GFP_NOIO;
176e21ee
CL		 2021
2022 xprt_clear_connected(xprt);
2023
2024 /* Reset to new socket */
2025 transport->sock = sock;
2026 transport->inet = sk;
2027
2028 write_unlock_bh(&sk->sk_callback_lock);
2029 }
2030
2031 /* Tell the socket layer to start connecting... */
2032 xprt->stat.connect_count++;
2033 xprt->stat.connect_start = jiffies;
2034 return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, 0);
2035}
2036
2037/**
2038 * xs_local_setup_socket - create AF_LOCAL socket, connect to a local endpoint
176e21ee 2039 * @transport: socket transport to connect
176e21ee 2040 */
dc107402 2041static int xs_local_setup_socket(struct sock_xprt *transport)
176e21ee 2042{
176e21ee
CL		 2043 struct rpc_xprt *xprt = &transport->xprt;
2044 struct socket *sock;
2045 int status = -EIO;
2046
176e21ee
CL		 2047 status = __sock_create(xprt->xprt_net, AF_LOCAL,
2048 SOCK_STREAM, 0, &sock, 1);
2049 if (status < 0) {
2050 dprintk("RPC: can't create AF_LOCAL "
2051 "transport socket (%d).\n", -status);
2052 goto out;
2053 }
d1358917 2054 xs_reclassify_socket(AF_LOCAL, sock);
176e21ee
CL		 2055
2056 dprintk("RPC: worker connecting xprt %p via AF_LOCAL to %s\n",
2057 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2058
2059 status = xs_local_finish_connecting(xprt, sock);
40b5ea0c 2060 trace_rpc_socket_connect(xprt, sock, status);
176e21ee
CL		 2061 switch (status) {
2062 case 0:
2063 dprintk("RPC: xprt %p connected to %s\n",
2064 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2065 xprt_set_connected(xprt);
3601c4a9 2066 case -ENOBUFS:
176e21ee
CL		 2067 break;
2068 case -ENOENT:
2069 dprintk("RPC: xprt %p: socket %s does not exist\n",
2070 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2071 break;
4a20a988
TM		 2072 case -ECONNREFUSED:
2073 dprintk("RPC: xprt %p: connection refused for %s\n",
2074 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2075 break;
176e21ee
CL		 2076 default:
2077 printk(KERN_ERR "%s: unhandled error (%d) connecting to %s\n",
2078 __func__, -status,
2079 xprt->address_strings[RPC_DISPLAY_ADDR]);
2080 }
2081
2082out:
2083 xprt_clear_connecting(xprt);
2084 xprt_wake_pending_tasks(xprt, status);
dc107402
BF		 2085 return status;
2086}
2087
b6669737 2088static void xs_local_connect(struct rpc_xprt *xprt, struct rpc_task *task)
dc107402 2089{
dc107402
BF		 2090 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2091 int ret;
2092
2093 if (RPC_IS_ASYNC(task)) {
2094 /*
2095 * We want the AF_LOCAL connect to be resolved in the
2096 * filesystem namespace of the process making the rpc
2097 * call. Thus we connect synchronously.
2098 *
2099 * If we want to support asynchronous AF_LOCAL calls,
2100 * we'll need to figure out how to pass a namespace to
2101 * connect.
2102 */
2103 rpc_exit(task, -ENOTCONN);
2104 return;
2105 }
2106 ret = xs_local_setup_socket(transport);
2107 if (ret && !RPC_IS_SOFTCONN(task))
2108 msleep_interruptible(15000);
176e21ee
CL		 2109}
2110
3c87ef6e 2111#if IS_ENABLED(CONFIG_SUNRPC_SWAP)
d6e971d8
JL		 2112/*
2113 * Note that this should be called with XPRT_LOCKED held (or when we otherwise
2114 * know that we have exclusive access to the socket), to guard against
2115 * races with xs_reset_transport.
2116 */
a564b8f0
MG		 2117static void xs_set_memalloc(struct rpc_xprt *xprt)
2118{
2119 struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
2120 xprt);
2121
d6e971d8
JL		 2122 /*
2123 * If there's no sock, then we have nothing to set. The
2124 * reconnecting process will get it for us.
2125 */
2126 if (!transport->inet)
2127 return;
8e228133 2128 if (atomic_read(&xprt->swapper))
a564b8f0
MG		 2129 sk_set_memalloc(transport->inet);
2130}
2131
2132/**
d67fa4d8 2133 * xs_enable_swap - Tag this transport as being used for swap.
a564b8f0 2134 * @xprt: transport to tag
a564b8f0 2135 *
8e228133
JL		 2136 * Take a reference to this transport on behalf of the rpc_clnt, and
2137 * optionally mark it for swapping if it wasn't already.
a564b8f0 2138 */
d67fa4d8
JL		 2139static int
2140xs_enable_swap(struct rpc_xprt *xprt)
a564b8f0 2141{
d6e971d8 2142 struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
a564b8f0 2143
d6e971d8
JL		 2144 if (atomic_inc_return(&xprt->swapper) != 1)
2145 return 0;
2146 if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE))
2147 return -ERESTARTSYS;
2148 if (xs->inet)
2149 sk_set_memalloc(xs->inet);
2150 xprt_release_xprt(xprt, NULL);
8e228133
JL		 2151 return 0;
2152}
a564b8f0 2153
8e228133 2154/**
d67fa4d8 2155 * xs_disable_swap - Untag this transport as being used for swap.
8e228133
JL		 2156 * @xprt: transport to tag
2157 *
2158 * Drop a "swapper" reference to this xprt on behalf of the rpc_clnt. If the
2159 * swapper refcount goes to 0, untag the socket as a memalloc socket.
2160 */
d67fa4d8
JL		 2161static void
2162xs_disable_swap(struct rpc_xprt *xprt)
8e228133 2163{
d6e971d8 2164 struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
8e228133 2165
d6e971d8
JL		 2166 if (!atomic_dec_and_test(&xprt->swapper))
2167 return;
2168 if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE))
2169 return;
2170 if (xs->inet)
2171 sk_clear_memalloc(xs->inet);
2172 xprt_release_xprt(xprt, NULL);
a564b8f0 2173}
a564b8f0
MG		 2174#else
2175static void xs_set_memalloc(struct rpc_xprt *xprt)
2176{
2177}
d67fa4d8
JL		 2178
2179static int
2180xs_enable_swap(struct rpc_xprt *xprt)
2181{
2182 return -EINVAL;
2183}
2184
2185static void
2186xs_disable_swap(struct rpc_xprt *xprt)
2187{
2188}
a564b8f0
MG		 2189#endif
2190
16be2d20
CL		 2191static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2192{
2193 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2194
2195 if (!transport->inet) {
2196 struct sock *sk = sock->sk;
2197
2198 write_lock_bh(&sk->sk_callback_lock);
2199
2a9e1cfa
TM		 2200 xs_save_old_callbacks(transport, sk);
2201
16be2d20 2202 sk->sk_user_data = xprt;
f9b2ee71 2203 sk->sk_data_ready = xs_data_ready;
16be2d20 2204 sk->sk_write_space = xs_udp_write_space;
b4411457 2205 sock_set_flag(sk, SOCK_FASYNC);
c2126157 2206 sk->sk_allocation = GFP_NOIO;
16be2d20
CL		 2207
2208 xprt_set_connected(xprt);
2209
2210 /* Reset to new socket */
2211 transport->sock = sock;
2212 transport->inet = sk;
2213
a564b8f0
MG		 2214 xs_set_memalloc(xprt);
2215
16be2d20
CL		 2216 write_unlock_bh(&sk->sk_callback_lock);
2217 }
2218 xs_udp_do_set_buffer_size(xprt);
02910177
TM		 2219
2220 xprt->stat.connect_start = jiffies;
16be2d20
CL		 2221}
2222
8c14ff2a 2223static void xs_udp_setup_socket(struct work_struct *work)
a246b010 2224{
34161db6
TM		 2225 struct sock_xprt *transport =
2226 container_of(work, struct sock_xprt, connect_worker.work);
c8475461 2227 struct rpc_xprt *xprt = &transport->xprt;
d099b8af 2228 struct socket *sock;
b65c0310 2229 int status = -EIO;
9903cd1c 2230
8c14ff2a 2231 sock = xs_create_sock(xprt, transport,
4dda9c8a
TM		 2232 xs_addr(xprt)->sa_family, SOCK_DGRAM,
2233 IPPROTO_UDP, false);
b65c0310 2234 if (IS_ERR(sock))
b0d93ad5 2235 goto out;
68e220bd 2236
c740eff8
CL		 2237 dprintk("RPC: worker connecting xprt %p via %s to "
2238 "%s (port %s)\n", xprt,
2239 xprt->address_strings[RPC_DISPLAY_PROTO],
2240 xprt->address_strings[RPC_DISPLAY_ADDR],
2241 xprt->address_strings[RPC_DISPLAY_PORT]);
68e220bd
CL		 2242
2243 xs_udp_finish_connecting(xprt, sock);
40b5ea0c 2244 trace_rpc_socket_connect(xprt, sock, 0);
b0d93ad5
CL		 2245 status = 0;
2246out:
718ba5b8 2247 xprt_unlock_connect(xprt, transport);
b0d93ad5 2248 xprt_clear_connecting(xprt);
7d1e8255 2249 xprt_wake_pending_tasks(xprt, status);
a246b010
CL		 2250}
2251
4876cc77
TM		 2252/**
2253 * xs_tcp_shutdown - gracefully shut down a TCP socket
2254 * @xprt: transport
2255 *
2256 * Initiates a graceful shutdown of the TCP socket by calling the
2257 * equivalent of shutdown(SHUT_RDWR);
2258 */
2259static void xs_tcp_shutdown(struct rpc_xprt *xprt)
2260{
2261 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2262 struct socket *sock = transport->sock;
9b30889c 2263 int skst = transport->inet ? transport->inet->sk_state : TCP_CLOSE;
4876cc77
TM		 2264
2265 if (sock == NULL)
2266 return;
9b30889c
TM		 2267 switch (skst) {
2268 default:
4876cc77
TM		 2269 kernel_sock_shutdown(sock, SHUT_RDWR);
2270 trace_rpc_socket_shutdown(xprt, sock);
9b30889c
TM		 2271 break;
2272 case TCP_CLOSE:
2273 case TCP_TIME_WAIT:
4876cc77 2274 xs_reset_transport(transport);
9b30889c 2275 }
4876cc77
TM		 2276}
2277
8d1b8c62
TM		 2278static void xs_tcp_set_socket_timeouts(struct rpc_xprt *xprt,
2279 struct socket *sock)
2280{
7196dbb0
TM		 2281 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2282 unsigned int keepidle;
2283 unsigned int keepcnt;
8d1b8c62
TM		 2284 unsigned int opt_on = 1;
2285 unsigned int timeo;
2286
7196dbb0
TM		 2287 spin_lock_bh(&xprt->transport_lock);
2288 keepidle = DIV_ROUND_UP(xprt->timeout->to_initval, HZ);
2289 keepcnt = xprt->timeout->to_retries + 1;
2290 timeo = jiffies_to_msecs(xprt->timeout->to_initval) *
2291 (xprt->timeout->to_retries + 1);
2292 clear_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state);
2293 spin_unlock_bh(&xprt->transport_lock);
2294
8d1b8c62
TM		 2295 /* TCP Keepalive options */
2296 kernel_setsockopt(sock, SOL_SOCKET, SO_KEEPALIVE,
2297 (char *)&opt_on, sizeof(opt_on));
2298 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPIDLE,
2299 (char *)&keepidle, sizeof(keepidle));
2300 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPINTVL,
2301 (char *)&keepidle, sizeof(keepidle));
2302 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPCNT,
2303 (char *)&keepcnt, sizeof(keepcnt));
2304
2305 /* TCP user timeout (see RFC5482) */
8d1b8c62
TM		 2306 kernel_setsockopt(sock, SOL_TCP, TCP_USER_TIMEOUT,
2307 (char *)&timeo, sizeof(timeo));
2308}
2309
7196dbb0
TM		 2310static void xs_tcp_set_connect_timeout(struct rpc_xprt *xprt,
2311 unsigned long connect_timeout,
2312 unsigned long reconnect_timeout)
2313{
2314 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2315 struct rpc_timeout to;
2316 unsigned long initval;
2317
2318 spin_lock_bh(&xprt->transport_lock);
2319 if (reconnect_timeout < xprt->max_reconnect_timeout)
2320 xprt->max_reconnect_timeout = reconnect_timeout;
2321 if (connect_timeout < xprt->connect_timeout) {
2322 memcpy(&to, xprt->timeout, sizeof(to));
2323 initval = DIV_ROUND_UP(connect_timeout, to.to_retries + 1);
2324 /* Arbitrary lower limit */
2325 if (initval < XS_TCP_INIT_REEST_TO << 1)
2326 initval = XS_TCP_INIT_REEST_TO << 1;
2327 to.to_initval = initval;
2328 to.to_maxval = initval;
2329 memcpy(&transport->tcp_timeout, &to,
2330 sizeof(transport->tcp_timeout));
2331 xprt->timeout = &transport->tcp_timeout;
2332 xprt->connect_timeout = connect_timeout;
2333 }
2334 set_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state);
2335 spin_unlock_bh(&xprt->transport_lock);
2336}
2337
16be2d20 2338static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
a246b010 2339{
16be2d20 2340 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
fe19a96b 2341 int ret = -ENOTCONN;
edb267a6 2342
ee0ac0c2 2343 if (!transport->inet) {
b0d93ad5 2344 struct sock *sk = sock->sk;
d88e4d82 2345 unsigned int addr_pref = IPV6_PREFER_SRC_PUBLIC;
7f260e85 2346
d88e4d82
N		 2347 /* Avoid temporary address, they are bad for long-lived
2348 * connections such as NFS mounts.
2349 * RFC4941, section 3.6 suggests that:
2350 * Individual applications, which have specific
2351 * knowledge about the normal duration of connections,
2352 * MAY override this as appropriate.
2353 */
2354 kernel_setsockopt(sock, SOL_IPV6, IPV6_ADDR_PREFERENCES,
2355 (char *)&addr_pref, sizeof(addr_pref));
2356
8d1b8c62 2357 xs_tcp_set_socket_timeouts(xprt, sock);
775f06ab 2358
b0d93ad5
CL		 2359 write_lock_bh(&sk->sk_callback_lock);
2360
2a9e1cfa
TM		 2361 xs_save_old_callbacks(transport, sk);
2362
b0d93ad5 2363 sk->sk_user_data = xprt;
5157b956 2364 sk->sk_data_ready = xs_data_ready;
b0d93ad5
CL		 2365 sk->sk_state_change = xs_tcp_state_change;
2366 sk->sk_write_space = xs_tcp_write_space;
b4411457 2367 sock_set_flag(sk, SOCK_FASYNC);
2118071d 2368 sk->sk_error_report = xs_error_report;
c2126157 2369 sk->sk_allocation = GFP_NOIO;
3167e12c
CL		 2370
2371 /* socket options */
3167e12c 2372 sock_reset_flag(sk, SOCK_LINGER);
3167e12c 2373 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
b0d93ad5
CL		 2374
2375 xprt_clear_connected(xprt);
2376
2377 /* Reset to new socket */
ee0ac0c2
CL		 2378 transport->sock = sock;
2379 transport->inet = sk;
b0d93ad5
CL		 2380
2381 write_unlock_bh(&sk->sk_callback_lock);
2382 }
2383
01d37c42 2384 if (!xprt_bound(xprt))
fe19a96b 2385 goto out;
01d37c42 2386
a564b8f0
MG		 2387 xs_set_memalloc(xprt);
2388
b0d93ad5 2389 /* Tell the socket layer to start connecting... */
262ca07d
CL		 2390 xprt->stat.connect_count++;
2391 xprt->stat.connect_start = jiffies;
0fdea1e8 2392 set_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
fe19a96b
TM		 2393 ret = kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
2394 switch (ret) {
2395 case 0:
4dda9c8a 2396 xs_set_srcport(transport, sock);
e9d47639 2397 /* fall through */
fe19a96b
TM		 2398 case -EINPROGRESS:
2399 /* SYN_SENT! */
fe19a96b
TM		 2400 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2401 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1f4c17a0
TM		 2402 break;
2403 case -EADDRNOTAVAIL:
2404 /* Source port number is unavailable. Try a new one! */
2405 transport->srcport = 0;
fe19a96b
TM		 2406 }
2407out:
2408 return ret;
16be2d20
CL		 2409}
2410
9903cd1c 2411/**
b61d59ff 2412 * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint
9903cd1c
CL		 2413 *
2414 * Invoked by a work queue tasklet.
a246b010 2415 */
cdd518d5 2416static void xs_tcp_setup_socket(struct work_struct *work)
a246b010 2417{
cdd518d5
PE		 2418 struct sock_xprt *transport =
2419 container_of(work, struct sock_xprt, connect_worker.work);
ee0ac0c2 2420 struct socket *sock = transport->sock;
a9f5f0f7 2421 struct rpc_xprt *xprt = &transport->xprt;
b61d59ff 2422 int status = -EIO;
a246b010 2423
ee0ac0c2 2424 if (!sock) {
cdd518d5 2425 sock = xs_create_sock(xprt, transport,
4dda9c8a
TM		 2426 xs_addr(xprt)->sa_family, SOCK_STREAM,
2427 IPPROTO_TCP, true);
b61d59ff
TM		 2428 if (IS_ERR(sock)) {
2429 status = PTR_ERR(sock);
3167e12c
CL		 2430 goto out;
2431 }
7d1e8255 2432 }
a246b010 2433
c740eff8
CL		 2434 dprintk("RPC: worker connecting xprt %p via %s to "
2435 "%s (port %s)\n", xprt,
2436 xprt->address_strings[RPC_DISPLAY_PROTO],
2437 xprt->address_strings[RPC_DISPLAY_ADDR],
2438 xprt->address_strings[RPC_DISPLAY_PORT]);
edb267a6 2439
16be2d20 2440 status = xs_tcp_finish_connecting(xprt, sock);
40b5ea0c 2441 trace_rpc_socket_connect(xprt, sock, status);
46121cf7
CL		 2442 dprintk("RPC: %p connect status %d connected %d sock state %d\n",
2443 xprt, -status, xprt_connected(xprt),
2444 sock->sk->sk_state);
2a491991 2445 switch (status) {
f75e6745
TM		 2446 default:
2447 printk("%s: connect returned unhandled error %d\n",
2448 __func__, status);
e9d47639 2449 /* fall through */
f75e6745
TM		 2450 case -EADDRNOTAVAIL:
2451 /* We're probably in TIME_WAIT. Get rid of existing socket,
2452 * and retry
2453 */
a519fc7a 2454 xs_tcp_force_close(xprt);
88b5ed73 2455 break;
2a491991
TM		 2456 case 0:
2457 case -EINPROGRESS:
2458 case -EALREADY:
718ba5b8 2459 xprt_unlock_connect(xprt, transport);
7d1e8255 2460 return;
9fcfe0c8
TM		 2461 case -EINVAL:
2462 /* Happens, for instance, if the user specified a link
2463 * local IPv6 address without a scope-id.
2464 */
3ed5e2a2
TM		 2465 case -ECONNREFUSED:
2466 case -ECONNRESET:
eb5b46fa 2467 case -ENETDOWN:
3ed5e2a2 2468 case -ENETUNREACH:
4ba161a7 2469 case -EHOSTUNREACH:
3913c78c 2470 case -EADDRINUSE:
3601c4a9 2471 case -ENOBUFS:
6ea44adc
N		 2472 /*
2473 * xs_tcp_force_close() wakes tasks with -EIO.
2474 * We need to wake them first to ensure the
2475 * correct error code.
2476 */
2477 xprt_wake_pending_tasks(xprt, status);
4efdd92c 2478 xs_tcp_force_close(xprt);
9fcfe0c8 2479 goto out;
a246b010 2480 }
2a491991 2481 status = -EAGAIN;
a246b010 2482out:
718ba5b8 2483 xprt_unlock_connect(xprt, transport);
68e220bd 2484 xprt_clear_connecting(xprt);
7d1e8255 2485 xprt_wake_pending_tasks(xprt, status);
68e220bd 2486}
b0d93ad5 2487
02910177
TM		 2488static unsigned long xs_reconnect_delay(const struct rpc_xprt *xprt)
2489{
2490 unsigned long start, now = jiffies;
2491
2492 start = xprt->stat.connect_start + xprt->reestablish_timeout;
2493 if (time_after(start, now))
2494 return start - now;
2495 return 0;
2496}
2497
3851f1cd
TM		 2498static void xs_reconnect_backoff(struct rpc_xprt *xprt)
2499{
2500 xprt->reestablish_timeout <<= 1;
2501 if (xprt->reestablish_timeout > xprt->max_reconnect_timeout)
2502 xprt->reestablish_timeout = xprt->max_reconnect_timeout;
2503 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2504 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2505}
2506
9903cd1c
CL		 2507/**
2508 * xs_connect - connect a socket to a remote endpoint
1b092092 2509 * @xprt: pointer to transport structure
9903cd1c
CL		 2510 * @task: address of RPC task that manages state of connect request
2511 *
2512 * TCP: If the remote end dropped the connection, delay reconnecting.
03bf4b70
CL		 2513 *
2514 * UDP socket connects are synchronous, but we use a work queue anyway
2515 * to guarantee that even unprivileged user processes can set up a
2516 * socket on a privileged port.
2517 *
2518 * If a UDP socket connect fails, the delay behavior here prevents
2519 * retry floods (hard mounts).
9903cd1c 2520 */
1b092092 2521static void xs_connect(struct rpc_xprt *xprt, struct rpc_task *task)
a246b010 2522{
ee0ac0c2 2523 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
02910177 2524 unsigned long delay = 0;
a246b010 2525
718ba5b8
TM		 2526 WARN_ON_ONCE(!xprt_lock_connect(xprt, task, transport));
2527
99b1a4c3 2528 if (transport->sock != NULL) {
46121cf7
CL		 2529 dprintk("RPC: xs_connect delayed xprt %p for %lu "
2530 "seconds\n",
03bf4b70 2531 xprt, xprt->reestablish_timeout / HZ);
99b1a4c3
TM		 2532
2533 /* Start by resetting any existing state */
2534 xs_reset_transport(transport);
2535
02910177 2536 delay = xs_reconnect_delay(xprt);
3851f1cd 2537 xs_reconnect_backoff(xprt);
02910177 2538
02910177 2539 } else
46121cf7 2540 dprintk("RPC: xs_connect scheduled xprt %p\n", xprt);
02910177
TM		 2541
2542 queue_delayed_work(xprtiod_workqueue,
2543 &transport->connect_worker,
2544 delay);
a246b010
CL		 2545}
2546
176e21ee
CL		 2547/**
2548 * xs_local_print_stats - display AF_LOCAL socket-specifc stats
2549 * @xprt: rpc_xprt struct containing statistics
2550 * @seq: output file
2551 *
2552 */
2553static void xs_local_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2554{
2555 long idle_time = 0;
2556
2557 if (xprt_connected(xprt))
2558 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2559
2560 seq_printf(seq, "\txprt:\tlocal %lu %lu %lu %ld %lu %lu %lu "
15a45206 2561 "%llu %llu %lu %llu %llu\n",
176e21ee
CL		 2562 xprt->stat.bind_count,
2563 xprt->stat.connect_count,
2564 xprt->stat.connect_time,
2565 idle_time,
2566 xprt->stat.sends,
2567 xprt->stat.recvs,
2568 xprt->stat.bad_xids,
2569 xprt->stat.req_u,
15a45206
AA		 2570 xprt->stat.bklog_u,
2571 xprt->stat.max_slots,
2572 xprt->stat.sending_u,
2573 xprt->stat.pending_u);
176e21ee
CL		 2574}
2575
262ca07d
CL		 2576/**
2577 * xs_udp_print_stats - display UDP socket-specifc stats
2578 * @xprt: rpc_xprt struct containing statistics
2579 * @seq: output file
2580 *
2581 */
2582static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2583{
c8475461
CL		 2584 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2585
15a45206
AA		 2586 seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %llu %llu "
2587 "%lu %llu %llu\n",
fbfffbd5 2588 transport->srcport,
262ca07d
CL		 2589 xprt->stat.bind_count,
2590 xprt->stat.sends,
2591 xprt->stat.recvs,
2592 xprt->stat.bad_xids,
2593 xprt->stat.req_u,
15a45206
AA		 2594 xprt->stat.bklog_u,
2595 xprt->stat.max_slots,
2596 xprt->stat.sending_u,
2597 xprt->stat.pending_u);
262ca07d
CL		 2598}
2599
2600/**
2601 * xs_tcp_print_stats - display TCP socket-specifc stats
2602 * @xprt: rpc_xprt struct containing statistics
2603 * @seq: output file
2604 *
2605 */
2606static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2607{
c8475461 2608 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
262ca07d
CL		 2609 long idle_time = 0;
2610
2611 if (xprt_connected(xprt))
2612 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2613
15a45206
AA		 2614 seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu "
2615 "%llu %llu %lu %llu %llu\n",
fbfffbd5 2616 transport->srcport,
262ca07d
CL		 2617 xprt->stat.bind_count,
2618 xprt->stat.connect_count,
2619 xprt->stat.connect_time,
2620 idle_time,
2621 xprt->stat.sends,
2622 xprt->stat.recvs,
2623 xprt->stat.bad_xids,
2624 xprt->stat.req_u,
15a45206
AA		 2625 xprt->stat.bklog_u,
2626 xprt->stat.max_slots,
2627 xprt->stat.sending_u,
2628 xprt->stat.pending_u);
262ca07d
CL		 2629}
2630
4cfc7e60
RI		 2631/*
2632 * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason
2633 * we allocate pages instead doing a kmalloc like rpc_malloc is because we want
2634 * to use the server side send routines.
2635 */
5fe6eaa1 2636static int bc_malloc(struct rpc_task *task)
4cfc7e60 2637{
5fe6eaa1
CL		 2638 struct rpc_rqst *rqst = task->tk_rqstp;
2639 size_t size = rqst->rq_callsize;
4cfc7e60
RI		 2640 struct page *page;
2641 struct rpc_buffer *buf;
2642
5fe6eaa1
CL		 2643 if (size > PAGE_SIZE - sizeof(struct rpc_buffer)) {
2644 WARN_ONCE(1, "xprtsock: large bc buffer request (size %zu)\n",
2645 size);
2646 return -EINVAL;
2647 }
4cfc7e60 2648
b8a13d03 2649 page = alloc_page(GFP_KERNEL);
4cfc7e60 2650 if (!page)
5fe6eaa1 2651 return -ENOMEM;
4cfc7e60
RI		 2652
2653 buf = page_address(page);
2654 buf->len = PAGE_SIZE;
2655
5fe6eaa1 2656 rqst->rq_buffer = buf->data;
18e601d6 2657 rqst->rq_rbuffer = (char *)rqst->rq_buffer + rqst->rq_callsize;
5fe6eaa1 2658 return 0;
4cfc7e60
RI		 2659}
2660
2661/*
2662 * Free the space allocated in the bc_alloc routine
2663 */
3435c74a 2664static void bc_free(struct rpc_task *task)
4cfc7e60 2665{
3435c74a 2666 void *buffer = task->tk_rqstp->rq_buffer;
4cfc7e60
RI		 2667 struct rpc_buffer *buf;
2668
4cfc7e60
RI		 2669 buf = container_of(buffer, struct rpc_buffer, data);
2670 free_page((unsigned long)buf);
2671}
2672
2673/*
2674 * Use the svc_sock to send the callback. Must be called with svsk->sk_mutex
2675 * held. Borrows heavily from svc_tcp_sendto and xs_tcp_send_request.
2676 */
2677static int bc_sendto(struct rpc_rqst *req)
2678{
2679 int len;
2680 struct xdr_buf *xbufp = &req->rq_snd_buf;
2681 struct rpc_xprt *xprt = req->rq_xprt;
2682 struct sock_xprt *transport =
2683 container_of(xprt, struct sock_xprt, xprt);
2684 struct socket *sock = transport->sock;
2685 unsigned long headoff;
2686 unsigned long tailoff;
2687
61677eee 2688 xs_encode_stream_record_marker(xbufp);
4cfc7e60
RI		 2689
2690 tailoff = (unsigned long)xbufp->tail[0].iov_base & ~PAGE_MASK;
2691 headoff = (unsigned long)xbufp->head[0].iov_base & ~PAGE_MASK;
2692 len = svc_send_common(sock, xbufp,
2693 virt_to_page(xbufp->head[0].iov_base), headoff,
2694 xbufp->tail[0].iov_base, tailoff);
2695
2696 if (len != xbufp->len) {
2697 printk(KERN_NOTICE "Error sending entire callback!\n");
2698 len = -EAGAIN;
2699 }
2700
2701 return len;
2702}
2703
2704/*
2705 * The send routine. Borrows from svc_send
2706 */
2707static int bc_send_request(struct rpc_task *task)
2708{
2709 struct rpc_rqst *req = task->tk_rqstp;
2710 struct svc_xprt *xprt;
7fc56136 2711 int len;
4cfc7e60
RI		 2712
2713 dprintk("sending request with xid: %08x\n", ntohl(req->rq_xid));
2714 /*
2715 * Get the server socket associated with this callback xprt
2716 */
2717 xprt = req->rq_xprt->bc_xprt;
4cfc7e60
RI		 2718
2719 /*
2720 * Grab the mutex to serialize data as the connection is shared
2721 * with the fore channel
2722 */
2723 if (!mutex_trylock(&xprt->xpt_mutex)) {
2724 rpc_sleep_on(&xprt->xpt_bc_pending, task, NULL);
2725 if (!mutex_trylock(&xprt->xpt_mutex))
2726 return -EAGAIN;
2727 rpc_wake_up_queued_task(&xprt->xpt_bc_pending, task);
2728 }
2729 if (test_bit(XPT_DEAD, &xprt->xpt_flags))
2730 len = -ENOTCONN;
2731 else
2732 len = bc_sendto(req);
2733 mutex_unlock(&xprt->xpt_mutex);
2734
2735 if (len > 0)
2736 len = 0;
2737
2738 return len;
2739}
2740
2741/*
2742 * The close routine. Since this is client initiated, we do nothing
2743 */
2744
2745static void bc_close(struct rpc_xprt *xprt)
2746{
4cfc7e60
RI		 2747}
2748
2749/*
2750 * The xprt destroy routine. Again, because this connection is client
2751 * initiated, we do nothing
2752 */
2753
2754static void bc_destroy(struct rpc_xprt *xprt)
2755{
47f72efa
KM		 2756 dprintk("RPC: bc_destroy xprt %p\n", xprt);
2757
2758 xs_xprt_free(xprt);
2759 module_put(THIS_MODULE);
4cfc7e60
RI		 2760}
2761
d31ae254 2762static const struct rpc_xprt_ops xs_local_ops = {
176e21ee
CL		 2763 .reserve_xprt = xprt_reserve_xprt,
2764 .release_xprt = xs_tcp_release_xprt,
f39c1bfb 2765 .alloc_slot = xprt_alloc_slot,
176e21ee
CL		 2766 .rpcbind = xs_local_rpcbind,
2767 .set_port = xs_local_set_port,
dc107402 2768 .connect = xs_local_connect,
176e21ee
CL		 2769 .buf_alloc = rpc_malloc,
2770 .buf_free = rpc_free,
2771 .send_request = xs_local_send_request,
2772 .set_retrans_timeout = xprt_set_retrans_timeout_def,
2773 .close = xs_close,
a1311d87 2774 .destroy = xs_destroy,
176e21ee 2775 .print_stats = xs_local_print_stats,
d67fa4d8
JL		 2776 .enable_swap = xs_enable_swap,
2777 .disable_swap = xs_disable_swap,
176e21ee
CL		 2778};
2779
d31ae254 2780static const struct rpc_xprt_ops xs_udp_ops = {
43118c29 2781 .set_buffer_size = xs_udp_set_buffer_size,
12a80469 2782 .reserve_xprt = xprt_reserve_xprt_cong,
49e9a890 2783 .release_xprt = xprt_release_xprt_cong,
f39c1bfb 2784 .alloc_slot = xprt_alloc_slot,
45160d62 2785 .rpcbind = rpcb_getport_async,
92200412 2786 .set_port = xs_set_port,
262965f5 2787 .connect = xs_connect,
02107148
CL		 2788 .buf_alloc = rpc_malloc,
2789 .buf_free = rpc_free,
262965f5 2790 .send_request = xs_udp_send_request,
fe3aca29 2791 .set_retrans_timeout = xprt_set_retrans_timeout_rtt,
46c0ee8b 2792 .timer = xs_udp_timer,
a58dd398 2793 .release_request = xprt_release_rqst_cong,
262965f5
CL		 2794 .close = xs_close,
2795 .destroy = xs_destroy,
262ca07d 2796 .print_stats = xs_udp_print_stats,
d67fa4d8
JL		 2797 .enable_swap = xs_enable_swap,
2798 .disable_swap = xs_disable_swap,
4a068258 2799 .inject_disconnect = xs_inject_disconnect,
262965f5
CL		 2800};
2801
d31ae254 2802static const struct rpc_xprt_ops xs_tcp_ops = {
12a80469 2803 .reserve_xprt = xprt_reserve_xprt,
e0ab53de 2804 .release_xprt = xs_tcp_release_xprt,
f39c1bfb 2805 .alloc_slot = xprt_lock_and_alloc_slot,
45160d62 2806 .rpcbind = rpcb_getport_async,
92200412 2807 .set_port = xs_set_port,
0b9e7943 2808 .connect = xs_connect,
02107148
CL		 2809 .buf_alloc = rpc_malloc,
2810 .buf_free = rpc_free,
262965f5 2811 .send_request = xs_tcp_send_request,
fe3aca29 2812 .set_retrans_timeout = xprt_set_retrans_timeout_def,
c627d31b 2813 .close = xs_tcp_shutdown,
9903cd1c 2814 .destroy = xs_destroy,
7196dbb0 2815 .set_connect_timeout = xs_tcp_set_connect_timeout,
262ca07d 2816 .print_stats = xs_tcp_print_stats,
d67fa4d8
JL		 2817 .enable_swap = xs_enable_swap,
2818 .disable_swap = xs_disable_swap,
4a068258 2819 .inject_disconnect = xs_inject_disconnect,
42e5c3e2
CL		 2820#ifdef CONFIG_SUNRPC_BACKCHANNEL
2821 .bc_setup = xprt_setup_bc,
76566773 2822 .bc_up = xs_tcp_bc_up,
6b26cc8c 2823 .bc_maxpayload = xs_tcp_bc_maxpayload,
42e5c3e2
CL		 2824 .bc_free_rqst = xprt_free_bc_rqst,
2825 .bc_destroy = xprt_destroy_bc,
2826#endif
a246b010
CL		 2827};
2828
4cfc7e60
RI		 2829/*
2830 * The rpc_xprt_ops for the server backchannel
2831 */
2832
d31ae254 2833static const struct rpc_xprt_ops bc_tcp_ops = {
4cfc7e60
RI		 2834 .reserve_xprt = xprt_reserve_xprt,
2835 .release_xprt = xprt_release_xprt,
84e28a30 2836 .alloc_slot = xprt_alloc_slot,
4cfc7e60
RI		 2837 .buf_alloc = bc_malloc,
2838 .buf_free = bc_free,
2839 .send_request = bc_send_request,
2840 .set_retrans_timeout = xprt_set_retrans_timeout_def,
2841 .close = bc_close,
2842 .destroy = bc_destroy,
2843 .print_stats = xs_tcp_print_stats,
d67fa4d8
JL		 2844 .enable_swap = xs_enable_swap,
2845 .disable_swap = xs_disable_swap,
4a068258 2846 .inject_disconnect = xs_inject_disconnect,
4cfc7e60
RI		 2847};
2848
92476850
CL		 2849static int xs_init_anyaddr(const int family, struct sockaddr *sap)
2850{
2851 static const struct sockaddr_in sin = {
2852 .sin_family = AF_INET,
2853 .sin_addr.s_addr = htonl(INADDR_ANY),
2854 };
2855 static const struct sockaddr_in6 sin6 = {
2856 .sin6_family = AF_INET6,
2857 .sin6_addr = IN6ADDR_ANY_INIT,
2858 };
2859
2860 switch (family) {
176e21ee
CL		 2861 case AF_LOCAL:
2862 break;
92476850
CL		 2863 case AF_INET:
2864 memcpy(sap, &sin, sizeof(sin));
2865 break;
2866 case AF_INET6:
2867 memcpy(sap, &sin6, sizeof(sin6));
2868 break;
2869 default:
2870 dprintk("RPC: %s: Bad address family\n", __func__);
2871 return -EAFNOSUPPORT;
2872 }
2873 return 0;
2874}
2875
3c341b0b 2876static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
d9ba131d
TM		 2877 unsigned int slot_table_size,
2878 unsigned int max_slot_table_size)
c8541ecd
CL		 2879{
2880 struct rpc_xprt *xprt;
ffc2e518 2881 struct sock_xprt *new;
c8541ecd 2882
96802a09 2883 if (args->addrlen > sizeof(xprt->addr)) {
46121cf7 2884 dprintk("RPC: xs_setup_xprt: address too large\n");
c8541ecd
CL		 2885 return ERR_PTR(-EBADF);
2886 }
2887
d9ba131d
TM		 2888 xprt = xprt_alloc(args->net, sizeof(*new), slot_table_size,
2889 max_slot_table_size);
bd1722d4 2890 if (xprt == NULL) {
46121cf7
CL		 2891 dprintk("RPC: xs_setup_xprt: couldn't allocate "
2892 "rpc_xprt\n");
c8541ecd
CL		 2893 return ERR_PTR(-ENOMEM);
2894 }
c8541ecd 2895
bd1722d4 2896 new = container_of(xprt, struct sock_xprt, xprt);
edc1b01c 2897 mutex_init(&new->recv_mutex);
96802a09
FM		 2898 memcpy(&xprt->addr, args->dstaddr, args->addrlen);
2899 xprt->addrlen = args->addrlen;
d3bc9a1d 2900 if (args->srcaddr)
fbfffbd5 2901 memcpy(&new->srcaddr, args->srcaddr, args->addrlen);
92476850
CL		 2902 else {
2903 int err;
2904 err = xs_init_anyaddr(args->dstaddr->sa_family,
2905 (struct sockaddr *)&new->srcaddr);
2aa13531
SK		 2906 if (err != 0) {
2907 xprt_free(xprt);
92476850 2908 return ERR_PTR(err);
2aa13531 2909 }
92476850 2910 }
c8541ecd
CL		 2911
2912 return xprt;
2913}
2914
176e21ee
CL		 2915static const struct rpc_timeout xs_local_default_timeout = {
2916 .to_initval = 10 * HZ,
2917 .to_maxval = 10 * HZ,
2918 .to_retries = 2,
2919};
2920
2921/**
2922 * xs_setup_local - Set up transport to use an AF_LOCAL socket
2923 * @args: rpc transport creation arguments
2924 *
2925 * AF_LOCAL is a "tpi_cots_ord" transport, just like TCP
2926 */
2927static struct rpc_xprt *xs_setup_local(struct xprt_create *args)
2928{
2929 struct sockaddr_un *sun = (struct sockaddr_un *)args->dstaddr;
2930 struct sock_xprt *transport;
2931 struct rpc_xprt *xprt;
2932 struct rpc_xprt *ret;
2933
d9ba131d
TM		 2934 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2935 xprt_max_tcp_slot_table_entries);
176e21ee
CL		 2936 if (IS_ERR(xprt))
2937 return xprt;
2938 transport = container_of(xprt, struct sock_xprt, xprt);
2939
2940 xprt->prot = 0;
2941 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2942 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2943
2944 xprt->bind_timeout = XS_BIND_TO;
2945 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2946 xprt->idle_timeout = XS_IDLE_DISC_TO;
2947
2948 xprt->ops = &xs_local_ops;
2949 xprt->timeout = &xs_local_default_timeout;
2950
a2648094 2951 INIT_WORK(&transport->recv_worker, xs_local_data_receive_workfn);
93dc41bd
N		 2952 INIT_DELAYED_WORK(&transport->connect_worker,
2953 xs_dummy_setup_socket);
2954
176e21ee
CL		 2955 switch (sun->sun_family) {
2956 case AF_LOCAL:
2957 if (sun->sun_path[0] != '/') {
2958 dprintk("RPC: bad AF_LOCAL address: %s\n",
2959 sun->sun_path);
2960 ret = ERR_PTR(-EINVAL);
2961 goto out_err;
2962 }
2963 xprt_set_bound(xprt);
176e21ee 2964 xs_format_peer_addresses(xprt, "local", RPCBIND_NETID_LOCAL);
7073ea87
BF		 2965 ret = ERR_PTR(xs_local_setup_socket(transport));
2966 if (ret)
2967 goto out_err;
176e21ee
CL		 2968 break;
2969 default:
2970 ret = ERR_PTR(-EAFNOSUPPORT);
2971 goto out_err;
2972 }
2973
2974 dprintk("RPC: set up xprt to %s via AF_LOCAL\n",
2975 xprt->address_strings[RPC_DISPLAY_ADDR]);
2976
2977 if (try_module_get(THIS_MODULE))
2978 return xprt;
2979 ret = ERR_PTR(-EINVAL);
2980out_err:
315f3812 2981 xs_xprt_free(xprt);
176e21ee
CL		 2982 return ret;
2983}
2984
2881ae74
TM		 2985static const struct rpc_timeout xs_udp_default_timeout = {
2986 .to_initval = 5 * HZ,
2987 .to_maxval = 30 * HZ,
2988 .to_increment = 5 * HZ,
2989 .to_retries = 5,
2990};
2991
9903cd1c
CL		 2992/**
2993 * xs_setup_udp - Set up transport to use a UDP socket
96802a09 2994 * @args: rpc transport creation arguments
9903cd1c
CL		 2995 *
2996 */
483066d6 2997static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
a246b010 2998{
8f9d5b1a 2999 struct sockaddr *addr = args->dstaddr;
c8541ecd 3000 struct rpc_xprt *xprt;
c8475461 3001 struct sock_xprt *transport;
0a68b0be 3002 struct rpc_xprt *ret;
a246b010 3003
d9ba131d
TM		 3004 xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries,
3005 xprt_udp_slot_table_entries);
c8541ecd
CL		 3006 if (IS_ERR(xprt))
3007 return xprt;
c8475461 3008 transport = container_of(xprt, struct sock_xprt, xprt);
a246b010 3009
ec739ef0 3010 xprt->prot = IPPROTO_UDP;
808012fb 3011 xprt->tsh_size = 0;
a246b010
CL		 3012 /* XXX: header size can vary due to auth type, IPv6, etc. */
3013 xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
3014
03bf4b70 3015 xprt->bind_timeout = XS_BIND_TO;
03bf4b70
CL		 3016 xprt->reestablish_timeout = XS_UDP_REEST_TO;
3017 xprt->idle_timeout = XS_IDLE_DISC_TO;
a246b010 3018
262965f5 3019 xprt->ops = &xs_udp_ops;
a246b010 3020
ba7392bb 3021 xprt->timeout = &xs_udp_default_timeout;
a246b010 3022
f9b2ee71 3023 INIT_WORK(&transport->recv_worker, xs_udp_data_receive_workfn);
edc1b01c
TM		 3024 INIT_DELAYED_WORK(&transport->connect_worker, xs_udp_setup_socket);
3025
8f9d5b1a
CL		 3026 switch (addr->sa_family) {
3027 case AF_INET:
3028 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
3029 xprt_set_bound(xprt);
3030
9dc3b095 3031 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
8f9d5b1a
CL		 3032 break;
3033 case AF_INET6:
3034 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
3035 xprt_set_bound(xprt);
3036
9dc3b095 3037 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
8f9d5b1a
CL		 3038 break;
3039 default:
0a68b0be
BF		 3040 ret = ERR_PTR(-EAFNOSUPPORT);
3041 goto out_err;
8f9d5b1a
CL		 3042 }
3043
c740eff8
CL		 3044 if (xprt_bound(xprt))
3045 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
3046 xprt->address_strings[RPC_DISPLAY_ADDR],
3047 xprt->address_strings[RPC_DISPLAY_PORT],
3048 xprt->address_strings[RPC_DISPLAY_PROTO]);
3049 else
3050 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
3051 xprt->address_strings[RPC_DISPLAY_ADDR],
3052 xprt->address_strings[RPC_DISPLAY_PROTO]);
edb267a6 3053
bc25571e
TT		 3054 if (try_module_get(THIS_MODULE))
3055 return xprt;
0a68b0be
BF		 3056 ret = ERR_PTR(-EINVAL);
3057out_err:
315f3812 3058 xs_xprt_free(xprt);
0a68b0be 3059 return ret;
a246b010
CL		 3060}
3061
2881ae74
TM		 3062static const struct rpc_timeout xs_tcp_default_timeout = {
3063 .to_initval = 60 * HZ,
3064 .to_maxval = 60 * HZ,
3065 .to_retries = 2,
3066};
3067
9903cd1c
CL		 3068/**
3069 * xs_setup_tcp - Set up transport to use a TCP socket
96802a09 3070 * @args: rpc transport creation arguments
9903cd1c
CL		 3071 *
3072 */
483066d6 3073static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
a246b010 3074{
8f9d5b1a 3075 struct sockaddr *addr = args->dstaddr;
c8541ecd 3076 struct rpc_xprt *xprt;
c8475461 3077 struct sock_xprt *transport;
0a68b0be 3078 struct rpc_xprt *ret;
b7993ceb
TM		 3079 unsigned int max_slot_table_size = xprt_max_tcp_slot_table_entries;
3080
3081 if (args->flags & XPRT_CREATE_INFINITE_SLOTS)
3082 max_slot_table_size = RPC_MAX_SLOT_TABLE_LIMIT;
a246b010 3083
d9ba131d 3084 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
b7993ceb 3085 max_slot_table_size);
c8541ecd
CL		 3086 if (IS_ERR(xprt))
3087 return xprt;
c8475461 3088 transport = container_of(xprt, struct sock_xprt, xprt);
a246b010 3089
ec739ef0 3090 xprt->prot = IPPROTO_TCP;
808012fb 3091 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
808012fb 3092 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
a246b010 3093
03bf4b70 3094 xprt->bind_timeout = XS_BIND_TO;
03bf4b70
CL		 3095 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
3096 xprt->idle_timeout = XS_IDLE_DISC_TO;
a246b010 3097
262965f5 3098 xprt->ops = &xs_tcp_ops;
ba7392bb 3099 xprt->timeout = &xs_tcp_default_timeout;
a246b010 3100
3851f1cd 3101 xprt->max_reconnect_timeout = xprt->timeout->to_maxval;
7196dbb0
TM		 3102 xprt->connect_timeout = xprt->timeout->to_initval *
3103 (xprt->timeout->to_retries + 1);
3851f1cd 3104
edc1b01c
TM		 3105 INIT_WORK(&transport->recv_worker, xs_tcp_data_receive_workfn);
3106 INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_setup_socket);
3107
8f9d5b1a
CL		 3108 switch (addr->sa_family) {
3109 case AF_INET:
3110 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
3111 xprt_set_bound(xprt);
3112
9dc3b095 3113 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
8f9d5b1a
CL		 3114 break;
3115 case AF_INET6:
3116 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
3117 xprt_set_bound(xprt);
3118
9dc3b095 3119 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
8f9d5b1a
CL		 3120 break;
3121 default:
0a68b0be
BF		 3122 ret = ERR_PTR(-EAFNOSUPPORT);
3123 goto out_err;
8f9d5b1a
CL		 3124 }
3125
c740eff8
CL		 3126 if (xprt_bound(xprt))
3127 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
3128 xprt->address_strings[RPC_DISPLAY_ADDR],
3129 xprt->address_strings[RPC_DISPLAY_PORT],
3130 xprt->address_strings[RPC_DISPLAY_PROTO]);
3131 else
3132 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
3133 xprt->address_strings[RPC_DISPLAY_ADDR],
3134 xprt->address_strings[RPC_DISPLAY_PROTO]);
3135
bc25571e
TT		 3136 if (try_module_get(THIS_MODULE))
3137 return xprt;
0a68b0be
BF		 3138 ret = ERR_PTR(-EINVAL);
3139out_err:
315f3812 3140 xs_xprt_free(xprt);
0a68b0be 3141 return ret;
a246b010 3142}
282b32e1 3143
f300baba
AB		 3144/**
3145 * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket
3146 * @args: rpc transport creation arguments
3147 *
3148 */
3149static struct rpc_xprt *xs_setup_bc_tcp(struct xprt_create *args)
3150{
3151 struct sockaddr *addr = args->dstaddr;
3152 struct rpc_xprt *xprt;
3153 struct sock_xprt *transport;
3154 struct svc_sock *bc_sock;
0a68b0be 3155 struct rpc_xprt *ret;
f300baba 3156
d9ba131d
TM		 3157 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
3158 xprt_tcp_slot_table_entries);
f300baba
AB		 3159 if (IS_ERR(xprt))
3160 return xprt;
3161 transport = container_of(xprt, struct sock_xprt, xprt);
3162
3163 xprt->prot = IPPROTO_TCP;
3164 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
3165 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
3166 xprt->timeout = &xs_tcp_default_timeout;
3167
3168 /* backchannel */
3169 xprt_set_bound(xprt);
3170 xprt->bind_timeout = 0;
f300baba
AB		 3171 xprt->reestablish_timeout = 0;
3172 xprt->idle_timeout = 0;
3173
f300baba
AB		 3174 xprt->ops = &bc_tcp_ops;
3175
3176 switch (addr->sa_family) {
3177 case AF_INET:
3178 xs_format_peer_addresses(xprt, "tcp",
3179 RPCBIND_NETID_TCP);
3180 break;
3181 case AF_INET6:
3182 xs_format_peer_addresses(xprt, "tcp",
3183 RPCBIND_NETID_TCP6);
3184 break;
3185 default:
0a68b0be
BF		 3186 ret = ERR_PTR(-EAFNOSUPPORT);
3187 goto out_err;
f300baba
AB		 3188 }
3189
50fa0d40
PE		 3190 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
3191 xprt->address_strings[RPC_DISPLAY_ADDR],
3192 xprt->address_strings[RPC_DISPLAY_PORT],
3193 xprt->address_strings[RPC_DISPLAY_PROTO]);
f300baba 3194
99de8ea9
BF		 3195 /*
3196 * Once we've associated a backchannel xprt with a connection,
28303ca3
WM		 3197 * we want to keep it around as long as the connection lasts,
3198 * in case we need to start using it for a backchannel again;
3199 * this reference won't be dropped until bc_xprt is destroyed.
99de8ea9
BF		 3200 */
3201 xprt_get(xprt);
3202 args->bc_xprt->xpt_bc_xprt = xprt;
3203 xprt->bc_xprt = args->bc_xprt;
3204 bc_sock = container_of(args->bc_xprt, struct svc_sock, sk_xprt);
3205 transport->sock = bc_sock->sk_sock;
3206 transport->inet = bc_sock->sk_sk;
3207
f300baba
AB		 3208 /*
3209 * Since we don't want connections for the backchannel, we set
3210 * the xprt status to connected
3211 */
3212 xprt_set_connected(xprt);
3213
f300baba
AB		 3214 if (try_module_get(THIS_MODULE))
3215 return xprt;
642aab58
KM		 3216
3217 args->bc_xprt->xpt_bc_xprt = NULL;
39a9beab 3218 args->bc_xprt->xpt_bc_xps = NULL;
99de8ea9 3219 xprt_put(xprt);
0a68b0be
BF		 3220 ret = ERR_PTR(-EINVAL);
3221out_err:
315f3812 3222 xs_xprt_free(xprt);
0a68b0be 3223 return ret;
f300baba
AB		 3224}
3225
176e21ee
CL		 3226static struct xprt_class xs_local_transport = {
3227 .list = LIST_HEAD_INIT(xs_local_transport.list),
3228 .name = "named UNIX socket",
3229 .owner = THIS_MODULE,
3230 .ident = XPRT_TRANSPORT_LOCAL,
3231 .setup = xs_setup_local,
3232};
3233
bc25571e
TT		 3234static struct xprt_class xs_udp_transport = {
3235 .list = LIST_HEAD_INIT(xs_udp_transport.list),
3236 .name = "udp",
3237 .owner = THIS_MODULE,
f300baba 3238 .ident = XPRT_TRANSPORT_UDP,
bc25571e
TT		 3239 .setup = xs_setup_udp,
3240};
3241
3242static struct xprt_class xs_tcp_transport = {
3243 .list = LIST_HEAD_INIT(xs_tcp_transport.list),
3244 .name = "tcp",
3245 .owner = THIS_MODULE,
f300baba 3246 .ident = XPRT_TRANSPORT_TCP,
bc25571e
TT		 3247 .setup = xs_setup_tcp,
3248};
3249
f300baba
AB		 3250static struct xprt_class xs_bc_tcp_transport = {
3251 .list = LIST_HEAD_INIT(xs_bc_tcp_transport.list),
3252 .name = "tcp NFSv4.1 backchannel",
3253 .owner = THIS_MODULE,
3254 .ident = XPRT_TRANSPORT_BC_TCP,
3255 .setup = xs_setup_bc_tcp,
3256};
3257
282b32e1 3258/**
bc25571e 3259 * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
282b32e1
CL		 3260 *
3261 */
3262int init_socket_xprt(void)
3263{
f895b252 3264#if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
2b1bec5f 3265 if (!sunrpc_table_header)
0b4d4147 3266 sunrpc_table_header = register_sysctl_table(sunrpc_table);
fbf76683
CL		 3267#endif
3268
176e21ee 3269 xprt_register_transport(&xs_local_transport);
bc25571e
TT		 3270 xprt_register_transport(&xs_udp_transport);
3271 xprt_register_transport(&xs_tcp_transport);
f300baba 3272 xprt_register_transport(&xs_bc_tcp_transport);
bc25571e 3273
282b32e1
CL		 3274 return 0;
3275}
3276
3277/**
bc25571e 3278 * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
282b32e1
CL		 3279 *
3280 */
3281void cleanup_socket_xprt(void)
3282{
f895b252 3283#if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
fbf76683
CL		 3284 if (sunrpc_table_header) {
3285 unregister_sysctl_table(sunrpc_table_header);
3286 sunrpc_table_header = NULL;
3287 }
3288#endif
bc25571e 3289
176e21ee 3290 xprt_unregister_transport(&xs_local_transport);
bc25571e
TT		 3291 xprt_unregister_transport(&xs_udp_transport);
3292 xprt_unregister_transport(&xs_tcp_transport);
f300baba 3293 xprt_unregister_transport(&xs_bc_tcp_transport);
282b32e1 3294}
cbf11071 3295
9bbb9e5a
RR		 3296static int param_set_uint_minmax(const char *val,
3297 const struct kernel_param *kp,
cbf11071
TM		 3298 unsigned int min, unsigned int max)
3299{
00cfaa94 3300 unsigned int num;
cbf11071
TM		 3301 int ret;
3302
3303 if (!val)
3304 return -EINVAL;
00cfaa94 3305 ret = kstrtouint(val, 0, &num);
9761a246
DC		 3306 if (ret)
3307 return ret;
3308 if (num < min || num > max)
cbf11071
TM		 3309 return -EINVAL;
3310 *((unsigned int *)kp->arg) = num;
3311 return 0;
3312}
3313
9bbb9e5a 3314static int param_set_portnr(const char *val, const struct kernel_param *kp)
cbf11071 3315{
ffb6ca33
FS		 3316 if (kp->arg == &xprt_min_resvport)
3317 return param_set_uint_minmax(val, kp,
cbf11071 3318 RPC_MIN_RESVPORT,
ffb6ca33
FS		 3319 xprt_max_resvport);
3320 return param_set_uint_minmax(val, kp,
3321 xprt_min_resvport,
cbf11071
TM		 3322 RPC_MAX_RESVPORT);
3323}
3324
9c27847d 3325static const struct kernel_param_ops param_ops_portnr = {
9bbb9e5a
RR		 3326 .set = param_set_portnr,
3327 .get = param_get_uint,
3328};
3329
cbf11071
TM		 3330#define param_check_portnr(name, p) \
3331 __param_check(name, p, unsigned int);
3332
3333module_param_named(min_resvport, xprt_min_resvport, portnr, 0644);
3334module_param_named(max_resvport, xprt_max_resvport, portnr, 0644);
3335
9bbb9e5a
RR		 3336static int param_set_slot_table_size(const char *val,
3337 const struct kernel_param *kp)
cbf11071
TM		 3338{
3339 return param_set_uint_minmax(val, kp,
3340 RPC_MIN_SLOT_TABLE,
3341 RPC_MAX_SLOT_TABLE);
3342}
3343
9c27847d 3344static const struct kernel_param_ops param_ops_slot_table_size = {
9bbb9e5a
RR		 3345 .set = param_set_slot_table_size,
3346 .get = param_get_uint,
3347};
3348
cbf11071
TM		 3349#define param_check_slot_table_size(name, p) \
3350 __param_check(name, p, unsigned int);
3351
d9ba131d
TM		 3352static int param_set_max_slot_table_size(const char *val,
3353 const struct kernel_param *kp)
3354{
3355 return param_set_uint_minmax(val, kp,
3356 RPC_MIN_SLOT_TABLE,
3357 RPC_MAX_SLOT_TABLE_LIMIT);
3358}
3359
9c27847d 3360static const struct kernel_param_ops param_ops_max_slot_table_size = {
d9ba131d
TM		 3361 .set = param_set_max_slot_table_size,
3362 .get = param_get_uint,
3363};
3364
3365#define param_check_max_slot_table_size(name, p) \
3366 __param_check(name, p, unsigned int);
3367
cbf11071
TM		 3368module_param_named(tcp_slot_table_entries, xprt_tcp_slot_table_entries,
3369 slot_table_size, 0644);
d9ba131d
TM		 3370module_param_named(tcp_max_slot_table_entries, xprt_max_tcp_slot_table_entries,
3371 max_slot_table_size, 0644);
cbf11071
TM		 3372module_param_named(udp_slot_table_entries, xprt_udp_slot_table_entries,
3373 slot_table_size, 0644);
3374
Linux 6.x block layer and io_uring tree(s)