projects / GitHub / mt8127 / android_kernel_alcatel_ttab.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Merge tag 'renesas-soc-r8a7790-for-v3.10' of git://git.kernel.org/pub/scm/linux/kerne...
[GitHub/mt8127/android_kernel_alcatel_ttab.git] / net / sunrpc / xprtsock.c
CommitLineData
a246b010
CL		 1/*
2 * linux/net/sunrpc/xprtsock.c
3 *
4 * Client-side transport implementation for sockets.
5 *
113aa838
AC		 6 * TCP callback races fixes (C) 1998 Red Hat
7 * TCP send fixes (C) 1998 Red Hat
a246b010
CL		 8 * TCP NFS related read + write fixes
9 * (C) 1999 Dave Airlie, University of Limerick, Ireland <airlied@linux.ie>
10 *
11 * Rewrite of larges part of the code in order to stabilize TCP stuff.
12 * Fix behaviour when socket buffer is full.
13 * (C) 1999 Trond Myklebust <trond.myklebust@fys.uio.no>
55aa4f58
CL		 14 *
15 * IP socket transport implementation, (C) 2005 Chuck Lever <cel@netapp.com>
8f9d5b1a
CL		 16 *
17 * IPv6 support contributed by Gilles Quillard, Bull Open Source, 2005.
18 * <gilles.quillard@bull.net>
a246b010
CL		 19 */
20
21#include <linux/types.h>
176e21ee 22#include <linux/string.h>
a246b010 23#include <linux/slab.h>
bc25571e 24#include <linux/module.h>
a246b010 25#include <linux/capability.h>
a246b010
CL		 26#include <linux/pagemap.h>
27#include <linux/errno.h>
28#include <linux/socket.h>
29#include <linux/in.h>
30#include <linux/net.h>
31#include <linux/mm.h>
176e21ee 32#include <linux/un.h>
a246b010
CL		 33#include <linux/udp.h>
34#include <linux/tcp.h>
35#include <linux/sunrpc/clnt.h>
5976687a 36#include <linux/sunrpc/addr.h>
02107148 37#include <linux/sunrpc/sched.h>
4cfc7e60 38#include <linux/sunrpc/svcsock.h>
49c36fcc 39#include <linux/sunrpc/xprtsock.h>
a246b010 40#include <linux/file.h>
9e00abc3 41#ifdef CONFIG_SUNRPC_BACKCHANNEL
44b98efd
RL		 42#include <linux/sunrpc/bc_xprt.h>
43#endif
a246b010
CL		 44
45#include <net/sock.h>
46#include <net/checksum.h>
47#include <net/udp.h>
48#include <net/tcp.h>
49
4cfc7e60 50#include "sunrpc.h"
176e21ee
CL		 51
52static void xs_close(struct rpc_xprt *xprt);
53
c556b754
CL		 54/*
55 * xprtsock tunables
56 */
09acfea5
TM		 57static unsigned int xprt_udp_slot_table_entries = RPC_DEF_SLOT_TABLE;
58static unsigned int xprt_tcp_slot_table_entries = RPC_MIN_SLOT_TABLE;
59static unsigned int xprt_max_tcp_slot_table_entries = RPC_MAX_SLOT_TABLE;
c556b754 60
09acfea5
TM		 61static unsigned int xprt_min_resvport = RPC_DEF_MIN_RESVPORT;
62static unsigned int xprt_max_resvport = RPC_DEF_MAX_RESVPORT;
c556b754 63
7d1e8255 64#define XS_TCP_LINGER_TO (15U * HZ)
25fe6142 65static unsigned int xs_tcp_fin_timeout __read_mostly = XS_TCP_LINGER_TO;
7d1e8255 66
fbf76683
CL		 67/*
68 * We can register our own files under /proc/sys/sunrpc by
69 * calling register_sysctl_table() again. The files in that
70 * directory become the union of all files registered there.
71 *
72 * We simply need to make sure that we don't collide with
73 * someone else's file names!
74 */
75
76#ifdef RPC_DEBUG
77
78static unsigned int min_slot_table_size = RPC_MIN_SLOT_TABLE;
79static unsigned int max_slot_table_size = RPC_MAX_SLOT_TABLE;
d9ba131d 80static unsigned int max_tcp_slot_table_limit = RPC_MAX_SLOT_TABLE_LIMIT;
fbf76683
CL		 81static unsigned int xprt_min_resvport_limit = RPC_MIN_RESVPORT;
82static unsigned int xprt_max_resvport_limit = RPC_MAX_RESVPORT;
83
84static struct ctl_table_header *sunrpc_table_header;
85
86/*
87 * FIXME: changing the UDP slot table size should also resize the UDP
88 * socket buffers for existing UDP transports
89 */
90static ctl_table xs_tunables_table[] = {
91 {
fbf76683
CL		 92 .procname = "udp_slot_table_entries",
93 .data = &xprt_udp_slot_table_entries,
94 .maxlen = sizeof(unsigned int),
95 .mode = 0644,
6d456111 96 .proc_handler = proc_dointvec_minmax,
fbf76683
CL		 97 .extra1 = &min_slot_table_size,
98 .extra2 = &max_slot_table_size
99 },
100 {
fbf76683
CL		 101 .procname = "tcp_slot_table_entries",
102 .data = &xprt_tcp_slot_table_entries,
103 .maxlen = sizeof(unsigned int),
104 .mode = 0644,
6d456111 105 .proc_handler = proc_dointvec_minmax,
fbf76683
CL		 106 .extra1 = &min_slot_table_size,
107 .extra2 = &max_slot_table_size
108 },
d9ba131d
TM		 109 {
110 .procname = "tcp_max_slot_table_entries",
111 .data = &xprt_max_tcp_slot_table_entries,
112 .maxlen = sizeof(unsigned int),
113 .mode = 0644,
114 .proc_handler = proc_dointvec_minmax,
115 .extra1 = &min_slot_table_size,
116 .extra2 = &max_tcp_slot_table_limit
117 },
fbf76683 118 {
fbf76683
CL		 119 .procname = "min_resvport",
120 .data = &xprt_min_resvport,
121 .maxlen = sizeof(unsigned int),
122 .mode = 0644,
6d456111 123 .proc_handler = proc_dointvec_minmax,
fbf76683
CL		 124 .extra1 = &xprt_min_resvport_limit,
125 .extra2 = &xprt_max_resvport_limit
126 },
127 {
fbf76683
CL		 128 .procname = "max_resvport",
129 .data = &xprt_max_resvport,
130 .maxlen = sizeof(unsigned int),
131 .mode = 0644,
6d456111 132 .proc_handler = proc_dointvec_minmax,
fbf76683
CL		 133 .extra1 = &xprt_min_resvport_limit,
134 .extra2 = &xprt_max_resvport_limit
135 },
25fe6142
TM		 136 {
137 .procname = "tcp_fin_timeout",
138 .data = &xs_tcp_fin_timeout,
139 .maxlen = sizeof(xs_tcp_fin_timeout),
140 .mode = 0644,
6d456111 141 .proc_handler = proc_dointvec_jiffies,
fbf76683 142 },
f8572d8f 143 { },
fbf76683
CL		 144};
145
146static ctl_table sunrpc_table[] = {
147 {
fbf76683
CL		 148 .procname = "sunrpc",
149 .mode = 0555,
150 .child = xs_tunables_table
151 },
f8572d8f 152 { },
fbf76683
CL		 153};
154
155#endif
156
03bf4b70
CL		 157/*
158 * Wait duration for a reply from the RPC portmapper.
159 */
160#define XS_BIND_TO (60U * HZ)
161
162/*
163 * Delay if a UDP socket connect error occurs. This is most likely some
164 * kind of resource problem on the local host.
165 */
166#define XS_UDP_REEST_TO (2U * HZ)
167
168/*
169 * The reestablish timeout allows clients to delay for a bit before attempting
170 * to reconnect to a server that just dropped our connection.
171 *
172 * We implement an exponential backoff when trying to reestablish a TCP
173 * transport connection with the server. Some servers like to drop a TCP
174 * connection when they are overworked, so we start with a short timeout and
175 * increase over time if the server is down or not responding.
176 */
177#define XS_TCP_INIT_REEST_TO (3U * HZ)
178#define XS_TCP_MAX_REEST_TO (5U * 60 * HZ)
179
180/*
181 * TCP idle timeout; client drops the transport socket if it is idle
182 * for this long. Note that we also timeout UDP sockets to prevent
183 * holding port numbers when there is no RPC traffic.
184 */
185#define XS_IDLE_DISC_TO (5U * 60 * HZ)
186
a246b010
CL		 187#ifdef RPC_DEBUG
188# undef RPC_DEBUG_DATA
9903cd1c 189# define RPCDBG_FACILITY RPCDBG_TRANS
a246b010
CL		 190#endif
191
a246b010 192#ifdef RPC_DEBUG_DATA
9903cd1c 193static void xs_pktdump(char *msg, u32 *packet, unsigned int count)
a246b010 194{
9903cd1c
CL		 195 u8 *buf = (u8 *) packet;
196 int j;
a246b010 197
46121cf7 198 dprintk("RPC: %s\n", msg);
a246b010
CL		 199 for (j = 0; j < count && j < 128; j += 4) {
200 if (!(j & 31)) {
201 if (j)
202 dprintk("\n");
203 dprintk("0x%04x ", j);
204 }
205 dprintk("%02x%02x%02x%02x ",
206 buf[j], buf[j+1], buf[j+2], buf[j+3]);
207 }
208 dprintk("\n");
209}
210#else
9903cd1c 211static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count)
a246b010
CL		 212{
213 /* NOP */
214}
215#endif
216
ffc2e518
CL		 217struct sock_xprt {
218 struct rpc_xprt xprt;
ee0ac0c2
CL		 219
220 /*
221 * Network layer
222 */
223 struct socket * sock;
224 struct sock * inet;
51971139
CL		 225
226 /*
227 * State of TCP reply receive
228 */
229 __be32 tcp_fraghdr,
b76ce561
TM		 230 tcp_xid,
231 tcp_calldir;
51971139
CL		 232
233 u32 tcp_offset,
234 tcp_reclen;
235
236 unsigned long tcp_copied,
237 tcp_flags;
c8475461
CL		 238
239 /*
240 * Connection of transports
241 */
34161db6 242 struct delayed_work connect_worker;
fbfffbd5
CL		 243 struct sockaddr_storage srcaddr;
244 unsigned short srcport;
7c6e066e
CL		 245
246 /*
247 * UDP socket buffer size parameters
248 */
249 size_t rcvsize,
250 sndsize;
314dfd79
CL		 251
252 /*
253 * Saved socket callback addresses
254 */
255 void (*old_data_ready)(struct sock *, int);
256 void (*old_state_change)(struct sock *);
257 void (*old_write_space)(struct sock *);
ffc2e518
CL		 258};
259
e136d092
CL		 260/*
261 * TCP receive state flags
262 */
263#define TCP_RCV_LAST_FRAG (1UL << 0)
264#define TCP_RCV_COPY_FRAGHDR (1UL << 1)
265#define TCP_RCV_COPY_XID (1UL << 2)
266#define TCP_RCV_COPY_DATA (1UL << 3)
f4a2e418
RL		 267#define TCP_RCV_READ_CALLDIR (1UL << 4)
268#define TCP_RCV_COPY_CALLDIR (1UL << 5)
18dca02a
RL		 269
270/*
271 * TCP RPC flags
272 */
f4a2e418 273#define TCP_RPC_REPLY (1UL << 6)
e136d092 274
95392c59
CL		 275static inline struct sockaddr *xs_addr(struct rpc_xprt *xprt)
276{
277 return (struct sockaddr *) &xprt->addr;
278}
279
176e21ee
CL		 280static inline struct sockaddr_un *xs_addr_un(struct rpc_xprt *xprt)
281{
282 return (struct sockaddr_un *) &xprt->addr;
283}
284
95392c59 285static inline struct sockaddr_in *xs_addr_in(struct rpc_xprt *xprt)
edb267a6 286{
95392c59
CL		 287 return (struct sockaddr_in *) &xprt->addr;
288}
289
290static inline struct sockaddr_in6 *xs_addr_in6(struct rpc_xprt *xprt)
291{
292 return (struct sockaddr_in6 *) &xprt->addr;
293}
294
c877b849 295static void xs_format_common_peer_addresses(struct rpc_xprt *xprt)
edb267a6 296{
c877b849 297 struct sockaddr *sap = xs_addr(xprt);
9dc3b095
CL		 298 struct sockaddr_in6 *sin6;
299 struct sockaddr_in *sin;
176e21ee 300 struct sockaddr_un *sun;
c877b849 301 char buf[128];
edb267a6 302
9dc3b095 303 switch (sap->sa_family) {
176e21ee
CL		 304 case AF_LOCAL:
305 sun = xs_addr_un(xprt);
306 strlcpy(buf, sun->sun_path, sizeof(buf));
307 xprt->address_strings[RPC_DISPLAY_ADDR] =
308 kstrdup(buf, GFP_KERNEL);
309 break;
9dc3b095 310 case AF_INET:
176e21ee
CL		 311 (void)rpc_ntop(sap, buf, sizeof(buf));
312 xprt->address_strings[RPC_DISPLAY_ADDR] =
313 kstrdup(buf, GFP_KERNEL);
9dc3b095 314 sin = xs_addr_in(xprt);
fc0b5791 315 snprintf(buf, sizeof(buf), "%08x", ntohl(sin->sin_addr.s_addr));
9dc3b095
CL		 316 break;
317 case AF_INET6:
176e21ee
CL		 318 (void)rpc_ntop(sap, buf, sizeof(buf));
319 xprt->address_strings[RPC_DISPLAY_ADDR] =
320 kstrdup(buf, GFP_KERNEL);
9dc3b095 321 sin6 = xs_addr_in6(xprt);
fc0b5791 322 snprintf(buf, sizeof(buf), "%pi6", &sin6->sin6_addr);
9dc3b095
CL		 323 break;
324 default:
325 BUG();
756805e7 326 }
176e21ee 327
9dc3b095 328 xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = kstrdup(buf, GFP_KERNEL);
edb267a6
CL		 329}
330
9dc3b095 331static void xs_format_common_peer_ports(struct rpc_xprt *xprt)
4b6473fb 332{
9dc3b095
CL		 333 struct sockaddr *sap = xs_addr(xprt);
334 char buf[128];
4b6473fb 335
81160e66 336 snprintf(buf, sizeof(buf), "%u", rpc_get_port(sap));
c877b849 337 xprt->address_strings[RPC_DISPLAY_PORT] = kstrdup(buf, GFP_KERNEL);
4b6473fb 338
81160e66 339 snprintf(buf, sizeof(buf), "%4hx", rpc_get_port(sap));
c877b849
CL		 340 xprt->address_strings[RPC_DISPLAY_HEX_PORT] = kstrdup(buf, GFP_KERNEL);
341}
4b6473fb 342
9dc3b095
CL		 343static void xs_format_peer_addresses(struct rpc_xprt *xprt,
344 const char *protocol,
345 const char *netid)
c877b849 346{
b454ae90 347 xprt->address_strings[RPC_DISPLAY_PROTO] = protocol;
b454ae90 348 xprt->address_strings[RPC_DISPLAY_NETID] = netid;
c877b849 349 xs_format_common_peer_addresses(xprt);
9dc3b095 350 xs_format_common_peer_ports(xprt);
edb267a6 351}
4b6473fb 352
9dc3b095 353static void xs_update_peer_port(struct rpc_xprt *xprt)
4b6473fb 354{
9dc3b095
CL		 355 kfree(xprt->address_strings[RPC_DISPLAY_HEX_PORT]);
356 kfree(xprt->address_strings[RPC_DISPLAY_PORT]);
4417c8c4 357
9dc3b095 358 xs_format_common_peer_ports(xprt);
edb267a6
CL		 359}
360
361static void xs_free_peer_addresses(struct rpc_xprt *xprt)
362{
33e01dc7
CL		 363 unsigned int i;
364
365 for (i = 0; i < RPC_DISPLAY_MAX; i++)
366 switch (i) {
367 case RPC_DISPLAY_PROTO:
368 case RPC_DISPLAY_NETID:
369 continue;
370 default:
371 kfree(xprt->address_strings[i]);
372 }
edb267a6
CL		 373}
374
b4b5cc85
CL		 375#define XS_SENDMSG_FLAGS (MSG_DONTWAIT | MSG_NOSIGNAL)
376
24c5684b 377static int xs_send_kvec(struct socket *sock, struct sockaddr *addr, int addrlen, struct kvec *vec, unsigned int base, int more)
b4b5cc85 378{
b4b5cc85
CL		 379 struct msghdr msg = {
380 .msg_name = addr,
381 .msg_namelen = addrlen,
24c5684b
TM		 382 .msg_flags = XS_SENDMSG_FLAGS | (more ? MSG_MORE : 0),
383 };
384 struct kvec iov = {
385 .iov_base = vec->iov_base + base,
386 .iov_len = vec->iov_len - base,
b4b5cc85
CL		 387 };
388
24c5684b 389 if (iov.iov_len != 0)
b4b5cc85
CL		 390 return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
391 return kernel_sendmsg(sock, &msg, NULL, 0, 0);
392}
393
24c5684b 394static int xs_send_pagedata(struct socket *sock, struct xdr_buf *xdr, unsigned int base, int more)
b4b5cc85 395{
24c5684b
TM		 396 struct page **ppage;
397 unsigned int remainder;
398 int err, sent = 0;
399
400 remainder = xdr->page_len - base;
401 base += xdr->page_base;
402 ppage = xdr->pages + (base >> PAGE_SHIFT);
403 base &= ~PAGE_MASK;
404 for(;;) {
405 unsigned int len = min_t(unsigned int, PAGE_SIZE - base, remainder);
406 int flags = XS_SENDMSG_FLAGS;
b4b5cc85 407
24c5684b
TM		 408 remainder -= len;
409 if (remainder != 0 || more)
410 flags |= MSG_MORE;
411 err = sock->ops->sendpage(sock, *ppage, base, len, flags);
412 if (remainder == 0 || err != len)
413 break;
414 sent += err;
415 ppage++;
416 base = 0;
417 }
418 if (sent == 0)
419 return err;
420 if (err > 0)
421 sent += err;
422 return sent;
b4b5cc85
CL		 423}
424
9903cd1c
CL		 425/**
426 * xs_sendpages - write pages directly to a socket
427 * @sock: socket to send on
428 * @addr: UDP only -- address of destination
429 * @addrlen: UDP only -- length of destination address
430 * @xdr: buffer containing this request
431 * @base: starting position in the buffer
432 *
a246b010 433 */
24c5684b 434static int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base)
a246b010 435{
24c5684b
TM		 436 unsigned int remainder = xdr->len - base;
437 int err, sent = 0;
a246b010 438
262965f5 439 if (unlikely(!sock))
fba91afb 440 return -ENOTSOCK;
262965f5
CL		 441
442 clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags);
24c5684b
TM		 443 if (base != 0) {
444 addr = NULL;
445 addrlen = 0;
446 }
262965f5 447
24c5684b
TM		 448 if (base < xdr->head[0].iov_len || addr != NULL) {
449 unsigned int len = xdr->head[0].iov_len - base;
450 remainder -= len;
451 err = xs_send_kvec(sock, addr, addrlen, &xdr->head[0], base, remainder != 0);
452 if (remainder == 0 || err != len)
a246b010 453 goto out;
24c5684b 454 sent += err;
a246b010
CL		 455 base = 0;
456 } else
24c5684b 457 base -= xdr->head[0].iov_len;
a246b010 458
24c5684b
TM		 459 if (base < xdr->page_len) {
460 unsigned int len = xdr->page_len - base;
461 remainder -= len;
462 err = xs_send_pagedata(sock, xdr, base, remainder != 0);
463 if (remainder == 0 || err != len)
a246b010 464 goto out;
24c5684b 465 sent += err;
a246b010 466 base = 0;
24c5684b
TM		 467 } else
468 base -= xdr->page_len;
469
470 if (base >= xdr->tail[0].iov_len)
471 return sent;
472 err = xs_send_kvec(sock, NULL, 0, &xdr->tail[0], base, 0);
a246b010 473out:
24c5684b
TM		 474 if (sent == 0)
475 return err;
476 if (err > 0)
477 sent += err;
478 return sent;
a246b010
CL		 479}
480
b6ddf64f
TM		 481static void xs_nospace_callback(struct rpc_task *task)
482{
483 struct sock_xprt *transport = container_of(task->tk_rqstp->rq_xprt, struct sock_xprt, xprt);
484
485 transport->inet->sk_write_pending--;
486 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
487}
488
9903cd1c 489/**
262965f5
CL		 490 * xs_nospace - place task on wait queue if transmit was incomplete
491 * @task: task to put to sleep
9903cd1c 492 *
a246b010 493 */
5e3771ce 494static int xs_nospace(struct rpc_task *task)
a246b010 495{
262965f5
CL		 496 struct rpc_rqst *req = task->tk_rqstp;
497 struct rpc_xprt *xprt = req->rq_xprt;
ee0ac0c2 498 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
24ca9a84 499 int ret = -EAGAIN;
a246b010 500
46121cf7 501 dprintk("RPC: %5u xmit incomplete (%u left of %u)\n",
262965f5
CL		 502 task->tk_pid, req->rq_slen - req->rq_bytes_sent,
503 req->rq_slen);
504
b6ddf64f
TM		 505 /* Protect against races with write_space */
506 spin_lock_bh(&xprt->transport_lock);
507
508 /* Don't race with disconnect */
509 if (xprt_connected(xprt)) {
510 if (test_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags)) {
511 /*
512 * Notify TCP that we're limited by the application
513 * window size
514 */
515 set_bit(SOCK_NOSPACE, &transport->sock->flags);
516 transport->inet->sk_write_pending++;
517 /* ...and wait for more buffer space */
518 xprt_wait_for_buffer_space(task, xs_nospace_callback);
519 }
520 } else {
521 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
5e3771ce 522 ret = -ENOTCONN;
b6ddf64f 523 }
262965f5 524
b6ddf64f 525 spin_unlock_bh(&xprt->transport_lock);
5e3771ce 526 return ret;
262965f5
CL		 527}
528
61677eee
CL		 529/*
530 * Construct a stream transport record marker in @buf.
531 */
532static inline void xs_encode_stream_record_marker(struct xdr_buf *buf)
533{
534 u32 reclen = buf->len - sizeof(rpc_fraghdr);
535 rpc_fraghdr *base = buf->head[0].iov_base;
536 *base = cpu_to_be32(RPC_LAST_STREAM_FRAGMENT | reclen);
537}
538
176e21ee
CL		 539/**
540 * xs_local_send_request - write an RPC request to an AF_LOCAL socket
541 * @task: RPC task that manages the state of an RPC request
542 *
543 * Return values:
544 * 0: The request has been sent
545 * EAGAIN: The socket was blocked, please call again later to
546 * complete the request
547 * ENOTCONN: Caller needs to invoke connect logic then call again
548 * other: Some other error occured, the request was not sent
549 */
550static int xs_local_send_request(struct rpc_task *task)
551{
552 struct rpc_rqst *req = task->tk_rqstp;
553 struct rpc_xprt *xprt = req->rq_xprt;
554 struct sock_xprt *transport =
555 container_of(xprt, struct sock_xprt, xprt);
556 struct xdr_buf *xdr = &req->rq_snd_buf;
557 int status;
558
559 xs_encode_stream_record_marker(&req->rq_snd_buf);
560
561 xs_pktdump("packet data:",
562 req->rq_svec->iov_base, req->rq_svec->iov_len);
563
564 status = xs_sendpages(transport->sock, NULL, 0,
565 xdr, req->rq_bytes_sent);
566 dprintk("RPC: %s(%u) = %d\n",
567 __func__, xdr->len - req->rq_bytes_sent, status);
568 if (likely(status >= 0)) {
569 req->rq_bytes_sent += status;
570 req->rq_xmit_bytes_sent += status;
571 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
572 req->rq_bytes_sent = 0;
573 return 0;
574 }
575 status = -EAGAIN;
576 }
577
578 switch (status) {
579 case -EAGAIN:
580 status = xs_nospace(task);
581 break;
582 default:
583 dprintk("RPC: sendmsg returned unrecognized error %d\n",
584 -status);
585 case -EPIPE:
586 xs_close(xprt);
587 status = -ENOTCONN;
588 }
589
590 return status;
591}
592
262965f5
CL		 593/**
594 * xs_udp_send_request - write an RPC request to a UDP socket
595 * @task: address of RPC task that manages the state of an RPC request
596 *
597 * Return values:
598 * 0: The request has been sent
599 * EAGAIN: The socket was blocked, please call again later to
600 * complete the request
601 * ENOTCONN: Caller needs to invoke connect logic then call again
25985edc 602 * other: Some other error occurred, the request was not sent
262965f5
CL		 603 */
604static int xs_udp_send_request(struct rpc_task *task)
605{
606 struct rpc_rqst *req = task->tk_rqstp;
607 struct rpc_xprt *xprt = req->rq_xprt;
ee0ac0c2 608 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
262965f5
CL		 609 struct xdr_buf *xdr = &req->rq_snd_buf;
610 int status;
a246b010 611
9903cd1c 612 xs_pktdump("packet data:",
a246b010
CL		 613 req->rq_svec->iov_base,
614 req->rq_svec->iov_len);
615
01d37c42
TM		 616 if (!xprt_bound(xprt))
617 return -ENOTCONN;
ee0ac0c2 618 status = xs_sendpages(transport->sock,
95392c59 619 xs_addr(xprt),
ee0ac0c2
CL		 620 xprt->addrlen, xdr,
621 req->rq_bytes_sent);
a246b010 622
46121cf7 623 dprintk("RPC: xs_udp_send_request(%u) = %d\n",
262965f5 624 xdr->len - req->rq_bytes_sent, status);
a246b010 625
2199700f 626 if (status >= 0) {
d60dbb20 627 req->rq_xmit_bytes_sent += status;
2199700f
TM		 628 if (status >= req->rq_slen)
629 return 0;
630 /* Still some bytes left; set up for a retry later. */
262965f5 631 status = -EAGAIN;
2199700f 632 }
a246b010 633
262965f5 634 switch (status) {
fba91afb
TM		 635 case -ENOTSOCK:
636 status = -ENOTCONN;
637 /* Should we call xs_close() here? */
638 break;
b6ddf64f 639 case -EAGAIN:
5e3771ce 640 status = xs_nospace(task);
b6ddf64f 641 break;
c8485e4d
TM		 642 default:
643 dprintk("RPC: sendmsg returned unrecognized error %d\n",
644 -status);
262965f5
CL		 645 case -ENETUNREACH:
646 case -EPIPE:
a246b010
CL		 647 case -ECONNREFUSED:
648 /* When the server has died, an ICMP port unreachable message
9903cd1c 649 * prompts ECONNREFUSED. */
b6ddf64f 650 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
a246b010 651 }
5fe46e9d 652
262965f5 653 return status;
a246b010
CL		 654}
655
e06799f9
TM		 656/**
657 * xs_tcp_shutdown - gracefully shut down a TCP socket
658 * @xprt: transport
659 *
660 * Initiates a graceful shutdown of the TCP socket by calling the
661 * equivalent of shutdown(SHUT_WR);
662 */
663static void xs_tcp_shutdown(struct rpc_xprt *xprt)
664{
665 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
666 struct socket *sock = transport->sock;
667
668 if (sock != NULL)
669 kernel_sock_shutdown(sock, SHUT_WR);
670}
671
9903cd1c 672/**
262965f5 673 * xs_tcp_send_request - write an RPC request to a TCP socket
9903cd1c
CL		 674 * @task: address of RPC task that manages the state of an RPC request
675 *
676 * Return values:
262965f5
CL		 677 * 0: The request has been sent
678 * EAGAIN: The socket was blocked, please call again later to
679 * complete the request
680 * ENOTCONN: Caller needs to invoke connect logic then call again
25985edc 681 * other: Some other error occurred, the request was not sent
9903cd1c
CL		 682 *
683 * XXX: In the case of soft timeouts, should we eventually give up
262965f5 684 * if sendmsg is not able to make progress?
9903cd1c 685 */
262965f5 686static int xs_tcp_send_request(struct rpc_task *task)
a246b010
CL		 687{
688 struct rpc_rqst *req = task->tk_rqstp;
689 struct rpc_xprt *xprt = req->rq_xprt;
ee0ac0c2 690 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
262965f5 691 struct xdr_buf *xdr = &req->rq_snd_buf;
b595bb15 692 int status;
a246b010 693
61677eee 694 xs_encode_stream_record_marker(&req->rq_snd_buf);
a246b010 695
262965f5
CL		 696 xs_pktdump("packet data:",
697 req->rq_svec->iov_base,
698 req->rq_svec->iov_len);
a246b010
CL		 699
700 /* Continue transmitting the packet/record. We must be careful
701 * to cope with writespace callbacks arriving _after_ we have
262965f5 702 * called sendmsg(). */
a246b010 703 while (1) {
ee0ac0c2
CL		 704 status = xs_sendpages(transport->sock,
705 NULL, 0, xdr, req->rq_bytes_sent);
a246b010 706
46121cf7 707 dprintk("RPC: xs_tcp_send_request(%u) = %d\n",
262965f5 708 xdr->len - req->rq_bytes_sent, status);
a246b010 709
262965f5 710 if (unlikely(status < 0))
a246b010 711 break;
a246b010 712
262965f5
CL		 713 /* If we've sent the entire packet, immediately
714 * reset the count of bytes sent. */
715 req->rq_bytes_sent += status;
d60dbb20 716 req->rq_xmit_bytes_sent += status;
262965f5
CL		 717 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
718 req->rq_bytes_sent = 0;
719 return 0;
720 }
a246b010 721
06b4b681
TM		 722 if (status != 0)
723 continue;
a246b010 724 status = -EAGAIN;
06b4b681 725 break;
a246b010
CL		 726 }
727
262965f5 728 switch (status) {
fba91afb
TM		 729 case -ENOTSOCK:
730 status = -ENOTCONN;
731 /* Should we call xs_close() here? */
732 break;
262965f5 733 case -EAGAIN:
5e3771ce 734 status = xs_nospace(task);
262965f5 735 break;
c8485e4d
TM		 736 default:
737 dprintk("RPC: sendmsg returned unrecognized error %d\n",
738 -status);
262965f5 739 case -ECONNRESET:
2a9e1cfa
TM		 740 xs_tcp_shutdown(xprt);
741 case -ECONNREFUSED:
262965f5 742 case -ENOTCONN:
b9d2bb2e 743 case -EPIPE:
b6ddf64f 744 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
a246b010 745 }
5fe46e9d 746
a246b010
CL		 747 return status;
748}
749
e0ab53de
TM		 750/**
751 * xs_tcp_release_xprt - clean up after a tcp transmission
752 * @xprt: transport
753 * @task: rpc task
754 *
755 * This cleans up if an error causes us to abort the transmission of a request.
756 * In this case, the socket may need to be reset in order to avoid confusing
757 * the server.
758 */
759static void xs_tcp_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
760{
761 struct rpc_rqst *req;
762
763 if (task != xprt->snd_task)
764 return;
765 if (task == NULL)
766 goto out_release;
767 req = task->tk_rqstp;
43cedbf0
TM		 768 if (req == NULL)
769 goto out_release;
e0ab53de
TM		 770 if (req->rq_bytes_sent == 0)
771 goto out_release;
772 if (req->rq_bytes_sent == req->rq_snd_buf.len)
773 goto out_release;
a4f0835c 774 set_bit(XPRT_CLOSE_WAIT, &xprt->state);
e0ab53de
TM		 775out_release:
776 xprt_release_xprt(xprt, task);
777}
778
2a9e1cfa
TM		 779static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk)
780{
781 transport->old_data_ready = sk->sk_data_ready;
782 transport->old_state_change = sk->sk_state_change;
783 transport->old_write_space = sk->sk_write_space;
2a9e1cfa
TM		 784}
785
786static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk)
787{
788 sk->sk_data_ready = transport->old_data_ready;
789 sk->sk_state_change = transport->old_state_change;
790 sk->sk_write_space = transport->old_write_space;
2a9e1cfa
TM		 791}
792
fe315e76 793static void xs_reset_transport(struct sock_xprt *transport)
a246b010 794{
ee0ac0c2
CL		 795 struct socket *sock = transport->sock;
796 struct sock *sk = transport->inet;
a246b010 797
fe315e76
CL		 798 if (sk == NULL)
799 return;
9903cd1c 800
246408dc
TM		 801 transport->srcport = 0;
802
a246b010 803 write_lock_bh(&sk->sk_callback_lock);
ee0ac0c2
CL		 804 transport->inet = NULL;
805 transport->sock = NULL;
a246b010 806
9903cd1c 807 sk->sk_user_data = NULL;
2a9e1cfa
TM		 808
809 xs_restore_old_callbacks(transport, sk);
a246b010
CL		 810 write_unlock_bh(&sk->sk_callback_lock);
811
9903cd1c 812 sk->sk_no_check = 0;
a246b010
CL		 813
814 sock_release(sock);
fe315e76
CL		 815}
816
817/**
818 * xs_close - close a socket
819 * @xprt: transport
820 *
821 * This is used when all requests are complete; ie, no DRC state remains
822 * on the server we want to save.
f75e6745
TM		 823 *
824 * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with
825 * xs_reset_transport() zeroing the socket from underneath a writer.
fe315e76
CL		 826 */
827static void xs_close(struct rpc_xprt *xprt)
828{
829 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
830
831 dprintk("RPC: xs_close xprt %p\n", xprt);
832
833 xs_reset_transport(transport);
61d0a8e6 834 xprt->reestablish_timeout = 0;
fe315e76 835
632e3bdc 836 smp_mb__before_clear_bit();
7d1e8255 837 clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
632e3bdc 838 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
3b948ae5 839 clear_bit(XPRT_CLOSING, &xprt->state);
632e3bdc 840 smp_mb__after_clear_bit();
62da3b24 841 xprt_disconnect_done(xprt);
a246b010
CL		 842}
843
f75e6745
TM		 844static void xs_tcp_close(struct rpc_xprt *xprt)
845{
846 if (test_and_clear_bit(XPRT_CONNECTION_CLOSE, &xprt->state))
847 xs_close(xprt);
848 else
849 xs_tcp_shutdown(xprt);
850}
851
190b1ecf
BF		 852static void xs_local_destroy(struct rpc_xprt *xprt)
853{
854 xs_close(xprt);
855 xs_free_peer_addresses(xprt);
856 xprt_free(xprt);
857 module_put(THIS_MODULE);
858}
859
9903cd1c
CL		 860/**
861 * xs_destroy - prepare to shutdown a transport
862 * @xprt: doomed transport
863 *
864 */
865static void xs_destroy(struct rpc_xprt *xprt)
a246b010 866{
c8475461
CL		 867 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
868
46121cf7 869 dprintk("RPC: xs_destroy xprt %p\n", xprt);
9903cd1c 870
afe2c511 871 cancel_delayed_work_sync(&transport->connect_worker);
a246b010 872
190b1ecf 873 xs_local_destroy(xprt);
a246b010
CL		 874}
875
9903cd1c
CL		 876static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
877{
878 return (struct rpc_xprt *) sk->sk_user_data;
879}
880
176e21ee
CL		 881static int xs_local_copy_to_xdr(struct xdr_buf *xdr, struct sk_buff *skb)
882{
883 struct xdr_skb_reader desc = {
884 .skb = skb,
885 .offset = sizeof(rpc_fraghdr),
886 .count = skb->len - sizeof(rpc_fraghdr),
887 };
888
889 if (xdr_partial_copy_from_skb(xdr, 0, &desc, xdr_skb_read_bits) < 0)
890 return -1;
891 if (desc.count)
892 return -1;
893 return 0;
894}
895
896/**
897 * xs_local_data_ready - "data ready" callback for AF_LOCAL sockets
898 * @sk: socket with data to read
899 * @len: how much data to read
900 *
901 * Currently this assumes we can read the whole reply in a single gulp.
902 */
903static void xs_local_data_ready(struct sock *sk, int len)
904{
905 struct rpc_task *task;
906 struct rpc_xprt *xprt;
907 struct rpc_rqst *rovr;
908 struct sk_buff *skb;
909 int err, repsize, copied;
910 u32 _xid;
911 __be32 *xp;
912
913 read_lock_bh(&sk->sk_callback_lock);
914 dprintk("RPC: %s...\n", __func__);
915 xprt = xprt_from_sock(sk);
916 if (xprt == NULL)
917 goto out;
918
919 skb = skb_recv_datagram(sk, 0, 1, &err);
920 if (skb == NULL)
921 goto out;
922
176e21ee
CL		 923 repsize = skb->len - sizeof(rpc_fraghdr);
924 if (repsize < 4) {
925 dprintk("RPC: impossible RPC reply size %d\n", repsize);
926 goto dropit;
927 }
928
929 /* Copy the XID from the skb... */
930 xp = skb_header_pointer(skb, sizeof(rpc_fraghdr), sizeof(_xid), &_xid);
931 if (xp == NULL)
932 goto dropit;
933
934 /* Look up and lock the request corresponding to the given XID */
935 spin_lock(&xprt->transport_lock);
936 rovr = xprt_lookup_rqst(xprt, *xp);
937 if (!rovr)
938 goto out_unlock;
939 task = rovr->rq_task;
940
941 copied = rovr->rq_private_buf.buflen;
942 if (copied > repsize)
943 copied = repsize;
944
945 if (xs_local_copy_to_xdr(&rovr->rq_private_buf, skb)) {
946 dprintk("RPC: sk_buff copy failed\n");
947 goto out_unlock;
948 }
949
950 xprt_complete_rqst(task, copied);
951
952 out_unlock:
953 spin_unlock(&xprt->transport_lock);
954 dropit:
955 skb_free_datagram(sk, skb);
956 out:
957 read_unlock_bh(&sk->sk_callback_lock);
958}
959
9903cd1c
CL		 960/**
961 * xs_udp_data_ready - "data ready" callback for UDP sockets
962 * @sk: socket with data to read
963 * @len: how much data to read
964 *
a246b010 965 */
9903cd1c 966static void xs_udp_data_ready(struct sock *sk, int len)
a246b010 967{
9903cd1c
CL		 968 struct rpc_task *task;
969 struct rpc_xprt *xprt;
a246b010 970 struct rpc_rqst *rovr;
9903cd1c 971 struct sk_buff *skb;
a246b010 972 int err, repsize, copied;
d8ed029d
AD		 973 u32 _xid;
974 __be32 *xp;
a246b010 975
f064af1e 976 read_lock_bh(&sk->sk_callback_lock);
46121cf7 977 dprintk("RPC: xs_udp_data_ready...\n");
9903cd1c 978 if (!(xprt = xprt_from_sock(sk)))
a246b010 979 goto out;
a246b010
CL		 980
981 if ((skb = skb_recv_datagram(sk, 0, 1, &err)) == NULL)
982 goto out;
983
a246b010
CL		 984 repsize = skb->len - sizeof(struct udphdr);
985 if (repsize < 4) {
46121cf7 986 dprintk("RPC: impossible RPC reply size %d!\n", repsize);
a246b010
CL		 987 goto dropit;
988 }
989
990 /* Copy the XID from the skb... */
991 xp = skb_header_pointer(skb, sizeof(struct udphdr),
992 sizeof(_xid), &_xid);
993 if (xp == NULL)
994 goto dropit;
995
996 /* Look up and lock the request corresponding to the given XID */
4a0f8c04 997 spin_lock(&xprt->transport_lock);
a246b010
CL		 998 rovr = xprt_lookup_rqst(xprt, *xp);
999 if (!rovr)
1000 goto out_unlock;
1001 task = rovr->rq_task;
1002
a246b010
CL		 1003 if ((copied = rovr->rq_private_buf.buflen) > repsize)
1004 copied = repsize;
1005
1006 /* Suck it into the iovec, verify checksum if not done by hw. */
1781f7f5
HX		 1007 if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
1008 UDPX_INC_STATS_BH(sk, UDP_MIB_INERRORS);
a246b010 1009 goto out_unlock;
1781f7f5
HX		 1010 }
1011
1012 UDPX_INC_STATS_BH(sk, UDP_MIB_INDATAGRAMS);
a246b010 1013
6a24dfb6 1014 xprt_adjust_cwnd(xprt, task, copied);
1570c1e4 1015 xprt_complete_rqst(task, copied);
a246b010
CL		 1016
1017 out_unlock:
4a0f8c04 1018 spin_unlock(&xprt->transport_lock);
a246b010
CL		 1019 dropit:
1020 skb_free_datagram(sk, skb);
1021 out:
f064af1e 1022 read_unlock_bh(&sk->sk_callback_lock);
a246b010
CL		 1023}
1024
a519fc7a
TM		 1025/*
1026 * Helper function to force a TCP close if the server is sending
1027 * junk and/or it has put us in CLOSE_WAIT
1028 */
1029static void xs_tcp_force_close(struct rpc_xprt *xprt)
1030{
1031 set_bit(XPRT_CONNECTION_CLOSE, &xprt->state);
1032 xprt_force_disconnect(xprt);
1033}
1034
dd456471 1035static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
a246b010 1036{
51971139 1037 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
a246b010
CL		 1038 size_t len, used;
1039 char *p;
1040
51971139
CL		 1041 p = ((char *) &transport->tcp_fraghdr) + transport->tcp_offset;
1042 len = sizeof(transport->tcp_fraghdr) - transport->tcp_offset;
9d292316 1043 used = xdr_skb_read_bits(desc, p, len);
51971139 1044 transport->tcp_offset += used;
a246b010
CL		 1045 if (used != len)
1046 return;
808012fb 1047
51971139
CL		 1048 transport->tcp_reclen = ntohl(transport->tcp_fraghdr);
1049 if (transport->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
e136d092 1050 transport->tcp_flags |= TCP_RCV_LAST_FRAG;
a246b010 1051 else
e136d092 1052 transport->tcp_flags &= ~TCP_RCV_LAST_FRAG;
51971139 1053 transport->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
808012fb 1054
e136d092 1055 transport->tcp_flags &= ~TCP_RCV_COPY_FRAGHDR;
51971139 1056 transport->tcp_offset = 0;
808012fb 1057
a246b010 1058 /* Sanity check of the record length */
18dca02a 1059 if (unlikely(transport->tcp_reclen < 8)) {
46121cf7 1060 dprintk("RPC: invalid TCP record fragment length\n");
a519fc7a 1061 xs_tcp_force_close(xprt);
9903cd1c 1062 return;
a246b010 1063 }
46121cf7 1064 dprintk("RPC: reading TCP record fragment of length %d\n",
51971139 1065 transport->tcp_reclen);
a246b010
CL		 1066}
1067
51971139 1068static void xs_tcp_check_fraghdr(struct sock_xprt *transport)
a246b010 1069{
51971139 1070 if (transport->tcp_offset == transport->tcp_reclen) {
e136d092 1071 transport->tcp_flags |= TCP_RCV_COPY_FRAGHDR;
51971139 1072 transport->tcp_offset = 0;
e136d092
CL		 1073 if (transport->tcp_flags & TCP_RCV_LAST_FRAG) {
1074 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1075 transport->tcp_flags |= TCP_RCV_COPY_XID;
51971139 1076 transport->tcp_copied = 0;
a246b010
CL		 1077 }
1078 }
1079}
1080
dd456471 1081static inline void xs_tcp_read_xid(struct sock_xprt *transport, struct xdr_skb_reader *desc)
a246b010
CL		 1082{
1083 size_t len, used;
1084 char *p;
1085
51971139 1086 len = sizeof(transport->tcp_xid) - transport->tcp_offset;
46121cf7 1087 dprintk("RPC: reading XID (%Zu bytes)\n", len);
51971139 1088 p = ((char *) &transport->tcp_xid) + transport->tcp_offset;
9d292316 1089 used = xdr_skb_read_bits(desc, p, len);
51971139 1090 transport->tcp_offset += used;
a246b010
CL		 1091 if (used != len)
1092 return;
e136d092 1093 transport->tcp_flags &= ~TCP_RCV_COPY_XID;
f4a2e418 1094 transport->tcp_flags |= TCP_RCV_READ_CALLDIR;
51971139 1095 transport->tcp_copied = 4;
18dca02a
RL		 1096 dprintk("RPC: reading %s XID %08x\n",
1097 (transport->tcp_flags & TCP_RPC_REPLY) ? "reply for"
1098 : "request with",
51971139
CL		 1099 ntohl(transport->tcp_xid));
1100 xs_tcp_check_fraghdr(transport);
a246b010
CL		 1101}
1102
18dca02a
RL		 1103static inline void xs_tcp_read_calldir(struct sock_xprt *transport,
1104 struct xdr_skb_reader *desc)
a246b010 1105{
18dca02a
RL		 1106 size_t len, used;
1107 u32 offset;
b76ce561 1108 char *p;
18dca02a
RL		 1109
1110 /*
1111 * We want transport->tcp_offset to be 8 at the end of this routine
1112 * (4 bytes for the xid and 4 bytes for the call/reply flag).
1113 * When this function is called for the first time,
1114 * transport->tcp_offset is 4 (after having already read the xid).
1115 */
1116 offset = transport->tcp_offset - sizeof(transport->tcp_xid);
b76ce561 1117 len = sizeof(transport->tcp_calldir) - offset;
18dca02a 1118 dprintk("RPC: reading CALL/REPLY flag (%Zu bytes)\n", len);
b76ce561
TM		 1119 p = ((char *) &transport->tcp_calldir) + offset;
1120 used = xdr_skb_read_bits(desc, p, len);
18dca02a
RL		 1121 transport->tcp_offset += used;
1122 if (used != len)
1123 return;
f4a2e418 1124 transport->tcp_flags &= ~TCP_RCV_READ_CALLDIR;
f4a2e418
RL		 1125 /*
1126 * We don't yet have the XDR buffer, so we will write the calldir
1127 * out after we get the buffer from the 'struct rpc_rqst'
1128 */
b76ce561
TM		 1129 switch (ntohl(transport->tcp_calldir)) {
1130 case RPC_REPLY:
1131 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1132 transport->tcp_flags |= TCP_RCV_COPY_DATA;
18dca02a 1133 transport->tcp_flags |= TCP_RPC_REPLY;
b76ce561
TM		 1134 break;
1135 case RPC_CALL:
1136 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1137 transport->tcp_flags |= TCP_RCV_COPY_DATA;
18dca02a 1138 transport->tcp_flags &= ~TCP_RPC_REPLY;
b76ce561
TM		 1139 break;
1140 default:
1141 dprintk("RPC: invalid request message type\n");
a519fc7a 1142 xs_tcp_force_close(&transport->xprt);
b76ce561 1143 }
18dca02a
RL		 1144 xs_tcp_check_fraghdr(transport);
1145}
1146
44b98efd
RL		 1147static inline void xs_tcp_read_common(struct rpc_xprt *xprt,
1148 struct xdr_skb_reader *desc,
1149 struct rpc_rqst *req)
a246b010 1150{
44b98efd
RL		 1151 struct sock_xprt *transport =
1152 container_of(xprt, struct sock_xprt, xprt);
a246b010
CL		 1153 struct xdr_buf *rcvbuf;
1154 size_t len;
1155 ssize_t r;
1156
a246b010 1157 rcvbuf = &req->rq_private_buf;
f4a2e418
RL		 1158
1159 if (transport->tcp_flags & TCP_RCV_COPY_CALLDIR) {
1160 /*
1161 * Save the RPC direction in the XDR buffer
1162 */
f4a2e418 1163 memcpy(rcvbuf->head[0].iov_base + transport->tcp_copied,
b76ce561
TM		 1164 &transport->tcp_calldir,
1165 sizeof(transport->tcp_calldir));
1166 transport->tcp_copied += sizeof(transport->tcp_calldir);
f4a2e418 1167 transport->tcp_flags &= ~TCP_RCV_COPY_CALLDIR;
a246b010
CL		 1168 }
1169
a246b010 1170 len = desc->count;
51971139 1171 if (len > transport->tcp_reclen - transport->tcp_offset) {
dd456471 1172 struct xdr_skb_reader my_desc;
a246b010 1173
51971139 1174 len = transport->tcp_reclen - transport->tcp_offset;
a246b010
CL		 1175 memcpy(&my_desc, desc, sizeof(my_desc));
1176 my_desc.count = len;
51971139 1177 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
9d292316 1178 &my_desc, xdr_skb_read_bits);
a246b010
CL		 1179 desc->count -= r;
1180 desc->offset += r;
1181 } else
51971139 1182 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
9d292316 1183 desc, xdr_skb_read_bits);
a246b010
CL		 1184
1185 if (r > 0) {
51971139
CL		 1186 transport->tcp_copied += r;
1187 transport->tcp_offset += r;
a246b010
CL		 1188 }
1189 if (r != len) {
1190 /* Error when copying to the receive buffer,
1191 * usually because we weren't able to allocate
1192 * additional buffer pages. All we can do now
e136d092 1193 * is turn off TCP_RCV_COPY_DATA, so the request
a246b010
CL		 1194 * will not receive any additional updates,
1195 * and time out.
1196 * Any remaining data from this record will
1197 * be discarded.
1198 */
e136d092 1199 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
46121cf7 1200 dprintk("RPC: XID %08x truncated request\n",
51971139 1201 ntohl(transport->tcp_xid));
46121cf7
CL		 1202 dprintk("RPC: xprt = %p, tcp_copied = %lu, "
1203 "tcp_offset = %u, tcp_reclen = %u\n",
1204 xprt, transport->tcp_copied,
1205 transport->tcp_offset, transport->tcp_reclen);
44b98efd 1206 return;
a246b010
CL		 1207 }
1208
46121cf7 1209 dprintk("RPC: XID %08x read %Zd bytes\n",
51971139 1210 ntohl(transport->tcp_xid), r);
46121cf7
CL		 1211 dprintk("RPC: xprt = %p, tcp_copied = %lu, tcp_offset = %u, "
1212 "tcp_reclen = %u\n", xprt, transport->tcp_copied,
1213 transport->tcp_offset, transport->tcp_reclen);
51971139
CL		 1214
1215 if (transport->tcp_copied == req->rq_private_buf.buflen)
e136d092 1216 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
51971139 1217 else if (transport->tcp_offset == transport->tcp_reclen) {
e136d092
CL		 1218 if (transport->tcp_flags & TCP_RCV_LAST_FRAG)
1219 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
a246b010 1220 }
44b98efd
RL		 1221}
1222
1223/*
1224 * Finds the request corresponding to the RPC xid and invokes the common
1225 * tcp read code to read the data.
1226 */
1227static inline int xs_tcp_read_reply(struct rpc_xprt *xprt,
1228 struct xdr_skb_reader *desc)
1229{
1230 struct sock_xprt *transport =
1231 container_of(xprt, struct sock_xprt, xprt);
1232 struct rpc_rqst *req;
1233
1234 dprintk("RPC: read reply XID %08x\n", ntohl(transport->tcp_xid));
1235
1236 /* Find and lock the request corresponding to this xid */
1237 spin_lock(&xprt->transport_lock);
1238 req = xprt_lookup_rqst(xprt, transport->tcp_xid);
1239 if (!req) {
1240 dprintk("RPC: XID %08x request not found!\n",
1241 ntohl(transport->tcp_xid));
1242 spin_unlock(&xprt->transport_lock);
1243 return -1;
1244 }
1245
1246 xs_tcp_read_common(xprt, desc, req);
1247
e136d092 1248 if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
51971139 1249 xprt_complete_rqst(req->rq_task, transport->tcp_copied);
44b98efd 1250
4a0f8c04 1251 spin_unlock(&xprt->transport_lock);
44b98efd
RL		 1252 return 0;
1253}
1254
9e00abc3 1255#if defined(CONFIG_SUNRPC_BACKCHANNEL)
44b98efd
RL		 1256/*
1257 * Obtains an rpc_rqst previously allocated and invokes the common
1258 * tcp read code to read the data. The result is placed in the callback
1259 * queue.
1260 * If we're unable to obtain the rpc_rqst we schedule the closing of the
1261 * connection and return -1.
1262 */
1263static inline int xs_tcp_read_callback(struct rpc_xprt *xprt,
1264 struct xdr_skb_reader *desc)
1265{
1266 struct sock_xprt *transport =
1267 container_of(xprt, struct sock_xprt, xprt);
1268 struct rpc_rqst *req;
1269
1270 req = xprt_alloc_bc_request(xprt);
1271 if (req == NULL) {
1272 printk(KERN_WARNING "Callback slot table overflowed\n");
1273 xprt_force_disconnect(xprt);
1274 return -1;
1275 }
1276
1277 req->rq_xid = transport->tcp_xid;
1278 dprintk("RPC: read callback XID %08x\n", ntohl(req->rq_xid));
1279 xs_tcp_read_common(xprt, desc, req);
1280
1281 if (!(transport->tcp_flags & TCP_RCV_COPY_DATA)) {
1282 struct svc_serv *bc_serv = xprt->bc_serv;
1283
1284 /*
1285 * Add callback request to callback list. The callback
1286 * service sleeps on the sv_cb_waitq waiting for new
1287 * requests. Wake it up after adding enqueing the
1288 * request.
1289 */
1290 dprintk("RPC: add callback request to list\n");
1291 spin_lock(&bc_serv->sv_cb_lock);
1292 list_add(&req->rq_bc_list, &bc_serv->sv_cb_list);
1293 spin_unlock(&bc_serv->sv_cb_lock);
1294 wake_up(&bc_serv->sv_cb_waitq);
1295 }
1296
1297 req->rq_private_buf.len = transport->tcp_copied;
1298
1299 return 0;
1300}
1301
1302static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1303 struct xdr_skb_reader *desc)
1304{
1305 struct sock_xprt *transport =
1306 container_of(xprt, struct sock_xprt, xprt);
1307
1308 return (transport->tcp_flags & TCP_RPC_REPLY) ?
1309 xs_tcp_read_reply(xprt, desc) :
1310 xs_tcp_read_callback(xprt, desc);
1311}
1312#else
1313static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1314 struct xdr_skb_reader *desc)
1315{
1316 return xs_tcp_read_reply(xprt, desc);
1317}
9e00abc3 1318#endif /* CONFIG_SUNRPC_BACKCHANNEL */
44b98efd
RL		 1319
1320/*
1321 * Read data off the transport. This can be either an RPC_CALL or an
1322 * RPC_REPLY. Relay the processing to helper functions.
1323 */
1324static void xs_tcp_read_data(struct rpc_xprt *xprt,
1325 struct xdr_skb_reader *desc)
1326{
1327 struct sock_xprt *transport =
1328 container_of(xprt, struct sock_xprt, xprt);
1329
1330 if (_xs_tcp_read_data(xprt, desc) == 0)
1331 xs_tcp_check_fraghdr(transport);
1332 else {
1333 /*
1334 * The transport_lock protects the request handling.
1335 * There's no need to hold it to update the tcp_flags.
1336 */
1337 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1338 }
a246b010
CL		 1339}
1340
dd456471 1341static inline void xs_tcp_read_discard(struct sock_xprt *transport, struct xdr_skb_reader *desc)
a246b010
CL		 1342{
1343 size_t len;
1344
51971139 1345 len = transport->tcp_reclen - transport->tcp_offset;
a246b010
CL		 1346 if (len > desc->count)
1347 len = desc->count;
1348 desc->count -= len;
1349 desc->offset += len;
51971139 1350 transport->tcp_offset += len;
46121cf7 1351 dprintk("RPC: discarded %Zu bytes\n", len);
51971139 1352 xs_tcp_check_fraghdr(transport);
a246b010
CL		 1353}
1354
9903cd1c 1355static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
a246b010
CL		 1356{
1357 struct rpc_xprt *xprt = rd_desc->arg.data;
51971139 1358 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
dd456471 1359 struct xdr_skb_reader desc = {
a246b010
CL		 1360 .skb = skb,
1361 .offset = offset,
1362 .count = len,
9903cd1c 1363 };
a246b010 1364
46121cf7 1365 dprintk("RPC: xs_tcp_data_recv started\n");
a246b010
CL		 1366 do {
1367 /* Read in a new fragment marker if necessary */
1368 /* Can we ever really expect to get completely empty fragments? */
e136d092 1369 if (transport->tcp_flags & TCP_RCV_COPY_FRAGHDR) {
9903cd1c 1370 xs_tcp_read_fraghdr(xprt, &desc);
a246b010
CL		 1371 continue;
1372 }
1373 /* Read in the xid if necessary */
e136d092 1374 if (transport->tcp_flags & TCP_RCV_COPY_XID) {
51971139 1375 xs_tcp_read_xid(transport, &desc);
a246b010
CL		 1376 continue;
1377 }
18dca02a 1378 /* Read in the call/reply flag */
f4a2e418 1379 if (transport->tcp_flags & TCP_RCV_READ_CALLDIR) {
18dca02a
RL		 1380 xs_tcp_read_calldir(transport, &desc);
1381 continue;
1382 }
a246b010 1383 /* Read in the request data */
e136d092 1384 if (transport->tcp_flags & TCP_RCV_COPY_DATA) {
44b98efd 1385 xs_tcp_read_data(xprt, &desc);
a246b010
CL		 1386 continue;
1387 }
1388 /* Skip over any trailing bytes on short reads */
51971139 1389 xs_tcp_read_discard(transport, &desc);
a246b010 1390 } while (desc.count);
46121cf7 1391 dprintk("RPC: xs_tcp_data_recv done\n");
a246b010
CL		 1392 return len - desc.count;
1393}
1394
9903cd1c
CL		 1395/**
1396 * xs_tcp_data_ready - "data ready" callback for TCP sockets
1397 * @sk: socket with data to read
1398 * @bytes: how much data to read
1399 *
1400 */
1401static void xs_tcp_data_ready(struct sock *sk, int bytes)
a246b010
CL		 1402{
1403 struct rpc_xprt *xprt;
1404 read_descriptor_t rd_desc;
ff2d7db8 1405 int read;
a246b010 1406
46121cf7
CL		 1407 dprintk("RPC: xs_tcp_data_ready...\n");
1408
f064af1e 1409 read_lock_bh(&sk->sk_callback_lock);
9903cd1c 1410 if (!(xprt = xprt_from_sock(sk)))
a246b010 1411 goto out;
61d0a8e6
NB		 1412 /* Any data means we had a useful conversation, so
1413 * the we don't need to delay the next reconnect
1414 */
1415 if (xprt->reestablish_timeout)
1416 xprt->reestablish_timeout = 0;
1417
9903cd1c 1418 /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
a246b010 1419 rd_desc.arg.data = xprt;
ff2d7db8
TM		 1420 do {
1421 rd_desc.count = 65536;
1422 read = tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
1423 } while (read > 0);
a246b010 1424out:
f064af1e 1425 read_unlock_bh(&sk->sk_callback_lock);
a246b010
CL		 1426}
1427
7d1e8255
TM		 1428/*
1429 * Do the equivalent of linger/linger2 handling for dealing with
1430 * broken servers that don't close the socket in a timely
1431 * fashion
1432 */
1433static void xs_tcp_schedule_linger_timeout(struct rpc_xprt *xprt,
1434 unsigned long timeout)
1435{
1436 struct sock_xprt *transport;
1437
1438 if (xprt_test_and_set_connecting(xprt))
1439 return;
1440 set_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1441 transport = container_of(xprt, struct sock_xprt, xprt);
1442 queue_delayed_work(rpciod_workqueue, &transport->connect_worker,
1443 timeout);
1444}
1445
1446static void xs_tcp_cancel_linger_timeout(struct rpc_xprt *xprt)
1447{
1448 struct sock_xprt *transport;
1449
1450 transport = container_of(xprt, struct sock_xprt, xprt);
1451
1452 if (!test_bit(XPRT_CONNECTION_ABORT, &xprt->state) ||
1453 !cancel_delayed_work(&transport->connect_worker))
1454 return;
1455 clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1456 xprt_clear_connecting(xprt);
1457}
1458
4bc1e68e 1459static void xs_sock_reset_connection_flags(struct rpc_xprt *xprt)
7d1e8255
TM		 1460{
1461 smp_mb__before_clear_bit();
a519fc7a
TM		 1462 clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1463 clear_bit(XPRT_CONNECTION_CLOSE, &xprt->state);
7d1e8255
TM		 1464 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1465 clear_bit(XPRT_CLOSING, &xprt->state);
1466 smp_mb__after_clear_bit();
4bc1e68e
TM		 1467}
1468
1469static void xs_sock_mark_closed(struct rpc_xprt *xprt)
1470{
1471 xs_sock_reset_connection_flags(xprt);
7d1e8255
TM		 1472 /* Mark transport as closed and wake up all pending tasks */
1473 xprt_disconnect_done(xprt);
1474}
1475
9903cd1c
CL		 1476/**
1477 * xs_tcp_state_change - callback to handle TCP socket state changes
1478 * @sk: socket whose state has changed
1479 *
1480 */
1481static void xs_tcp_state_change(struct sock *sk)
a246b010 1482{
9903cd1c 1483 struct rpc_xprt *xprt;
a246b010 1484
f064af1e 1485 read_lock_bh(&sk->sk_callback_lock);
a246b010
CL		 1486 if (!(xprt = xprt_from_sock(sk)))
1487 goto out;
46121cf7 1488 dprintk("RPC: xs_tcp_state_change client %p...\n", xprt);
669502ff 1489 dprintk("RPC: state %x conn %d dead %d zapped %d sk_shutdown %d\n",
46121cf7
CL		 1490 sk->sk_state, xprt_connected(xprt),
1491 sock_flag(sk, SOCK_DEAD),
669502ff
AC		 1492 sock_flag(sk, SOCK_ZAPPED),
1493 sk->sk_shutdown);
a246b010
CL		 1494
1495 switch (sk->sk_state) {
1496 case TCP_ESTABLISHED:
f064af1e 1497 spin_lock(&xprt->transport_lock);
a246b010 1498 if (!xprt_test_and_set_connected(xprt)) {
51971139
CL		 1499 struct sock_xprt *transport = container_of(xprt,
1500 struct sock_xprt, xprt);
1501
a246b010 1502 /* Reset TCP record info */
51971139
CL		 1503 transport->tcp_offset = 0;
1504 transport->tcp_reclen = 0;
1505 transport->tcp_copied = 0;
e136d092
CL		 1506 transport->tcp_flags =
1507 TCP_RCV_COPY_FRAGHDR | TCP_RCV_COPY_XID;
51971139 1508
2a491991 1509 xprt_wake_pending_tasks(xprt, -EAGAIN);
a246b010 1510 }
f064af1e 1511 spin_unlock(&xprt->transport_lock);
a246b010 1512 break;
3b948ae5
TM		 1513 case TCP_FIN_WAIT1:
1514 /* The client initiated a shutdown of the socket */
7c1d71cf 1515 xprt->connect_cookie++;
663b8858 1516 xprt->reestablish_timeout = 0;
3b948ae5
TM		 1517 set_bit(XPRT_CLOSING, &xprt->state);
1518 smp_mb__before_clear_bit();
1519 clear_bit(XPRT_CONNECTED, &xprt->state);
ef803670 1520 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
3b948ae5 1521 smp_mb__after_clear_bit();
25fe6142 1522 xs_tcp_schedule_linger_timeout(xprt, xs_tcp_fin_timeout);
a246b010 1523 break;
632e3bdc 1524 case TCP_CLOSE_WAIT:
3b948ae5 1525 /* The server initiated a shutdown of the socket */
7c1d71cf 1526 xprt->connect_cookie++;
d0bea455 1527 clear_bit(XPRT_CONNECTED, &xprt->state);
a519fc7a 1528 xs_tcp_force_close(xprt);
663b8858
TM		 1529 case TCP_CLOSING:
1530 /*
1531 * If the server closed down the connection, make sure that
1532 * we back off before reconnecting
1533 */
1534 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1535 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
3b948ae5
TM		 1536 break;
1537 case TCP_LAST_ACK:
670f9457 1538 set_bit(XPRT_CLOSING, &xprt->state);
25fe6142 1539 xs_tcp_schedule_linger_timeout(xprt, xs_tcp_fin_timeout);
3b948ae5
TM		 1540 smp_mb__before_clear_bit();
1541 clear_bit(XPRT_CONNECTED, &xprt->state);
1542 smp_mb__after_clear_bit();
1543 break;
1544 case TCP_CLOSE:
7d1e8255
TM		 1545 xs_tcp_cancel_linger_timeout(xprt);
1546 xs_sock_mark_closed(xprt);
a246b010
CL		 1547 }
1548 out:
f064af1e 1549 read_unlock_bh(&sk->sk_callback_lock);
a246b010
CL		 1550}
1551
1f0fa154
IJ		 1552static void xs_write_space(struct sock *sk)
1553{
1554 struct socket *sock;
1555 struct rpc_xprt *xprt;
1556
1557 if (unlikely(!(sock = sk->sk_socket)))
1558 return;
1559 clear_bit(SOCK_NOSPACE, &sock->flags);
1560
1561 if (unlikely(!(xprt = xprt_from_sock(sk))))
1562 return;
1563 if (test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags) == 0)
1564 return;
1565
1566 xprt_write_space(xprt);
1567}
1568
9903cd1c 1569/**
c7b2cae8
CL		 1570 * xs_udp_write_space - callback invoked when socket buffer space
1571 * becomes available
9903cd1c
CL		 1572 * @sk: socket whose state has changed
1573 *
a246b010
CL		 1574 * Called when more output buffer space is available for this socket.
1575 * We try not to wake our writers until they can make "significant"
c7b2cae8 1576 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
a246b010
CL		 1577 * with a bunch of small requests.
1578 */
c7b2cae8 1579static void xs_udp_write_space(struct sock *sk)
a246b010 1580{
f064af1e 1581 read_lock_bh(&sk->sk_callback_lock);
a246b010 1582
c7b2cae8 1583 /* from net/core/sock.c:sock_def_write_space */
1f0fa154
IJ		 1584 if (sock_writeable(sk))
1585 xs_write_space(sk);
a246b010 1586
f064af1e 1587 read_unlock_bh(&sk->sk_callback_lock);
c7b2cae8 1588}
a246b010 1589
c7b2cae8
CL		 1590/**
1591 * xs_tcp_write_space - callback invoked when socket buffer space
1592 * becomes available
1593 * @sk: socket whose state has changed
1594 *
1595 * Called when more output buffer space is available for this socket.
1596 * We try not to wake our writers until they can make "significant"
1597 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1598 * with a bunch of small requests.
1599 */
1600static void xs_tcp_write_space(struct sock *sk)
1601{
f064af1e 1602 read_lock_bh(&sk->sk_callback_lock);
c7b2cae8
CL		 1603
1604 /* from net/core/stream.c:sk_stream_write_space */
1f0fa154
IJ		 1605 if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk))
1606 xs_write_space(sk);
b6ddf64f 1607
f064af1e 1608 read_unlock_bh(&sk->sk_callback_lock);
a246b010
CL		 1609}
1610
470056c2 1611static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
a246b010 1612{
ee0ac0c2
CL		 1613 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1614 struct sock *sk = transport->inet;
a246b010 1615
7c6e066e 1616 if (transport->rcvsize) {
a246b010 1617 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
7c6e066e 1618 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
a246b010 1619 }
7c6e066e 1620 if (transport->sndsize) {
a246b010 1621 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
7c6e066e 1622 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
a246b010
CL		 1623 sk->sk_write_space(sk);
1624 }
1625}
1626
43118c29 1627/**
470056c2 1628 * xs_udp_set_buffer_size - set send and receive limits
43118c29 1629 * @xprt: generic transport
470056c2
CL		 1630 * @sndsize: requested size of send buffer, in bytes
1631 * @rcvsize: requested size of receive buffer, in bytes
43118c29 1632 *
470056c2 1633 * Set socket send and receive buffer size limits.
43118c29 1634 */
470056c2 1635static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
43118c29 1636{
7c6e066e
CL		 1637 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1638
1639 transport->sndsize = 0;
470056c2 1640 if (sndsize)
7c6e066e
CL		 1641 transport->sndsize = sndsize + 1024;
1642 transport->rcvsize = 0;
470056c2 1643 if (rcvsize)
7c6e066e 1644 transport->rcvsize = rcvsize + 1024;
470056c2
CL		 1645
1646 xs_udp_do_set_buffer_size(xprt);
43118c29
CL		 1647}
1648
46c0ee8b
CL		 1649/**
1650 * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1651 * @task: task that timed out
1652 *
1653 * Adjust the congestion window after a retransmit timeout has occurred.
1654 */
6a24dfb6 1655static void xs_udp_timer(struct rpc_xprt *xprt, struct rpc_task *task)
46c0ee8b 1656{
6a24dfb6 1657 xprt_adjust_cwnd(xprt, task, -ETIMEDOUT);
46c0ee8b
CL		 1658}
1659
b85d8806
CL		 1660static unsigned short xs_get_random_port(void)
1661{
1662 unsigned short range = xprt_max_resvport - xprt_min_resvport;
1663 unsigned short rand = (unsigned short) net_random() % range;
1664 return rand + xprt_min_resvport;
1665}
1666
92200412
CL		 1667/**
1668 * xs_set_port - reset the port number in the remote endpoint address
1669 * @xprt: generic transport
1670 * @port: new port number
1671 *
1672 */
1673static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1674{
46121cf7 1675 dprintk("RPC: setting port for xprt %p to %u\n", xprt, port);
c4efcb1d 1676
9dc3b095
CL		 1677 rpc_set_port(xs_addr(xprt), port);
1678 xs_update_peer_port(xprt);
92200412
CL		 1679}
1680
5d4ec932 1681static unsigned short xs_get_srcport(struct sock_xprt *transport)
67a391d7 1682{
fbfffbd5 1683 unsigned short port = transport->srcport;
67a391d7
TM		 1684
1685 if (port == 0 && transport->xprt.resvport)
1686 port = xs_get_random_port();
1687 return port;
1688}
1689
baaf4e48 1690static unsigned short xs_next_srcport(struct sock_xprt *transport, unsigned short port)
67a391d7 1691{
fbfffbd5
CL		 1692 if (transport->srcport != 0)
1693 transport->srcport = 0;
67a391d7
TM		 1694 if (!transport->xprt.resvport)
1695 return 0;
1696 if (port <= xprt_min_resvport || port > xprt_max_resvport)
1697 return xprt_max_resvport;
1698 return --port;
1699}
beb59b68 1700static int xs_bind(struct sock_xprt *transport, struct socket *sock)
a246b010 1701{
beb59b68 1702 struct sockaddr_storage myaddr;
67a391d7 1703 int err, nloop = 0;
5d4ec932 1704 unsigned short port = xs_get_srcport(transport);
67a391d7 1705 unsigned short last;
a246b010 1706
beb59b68 1707 memcpy(&myaddr, &transport->srcaddr, transport->xprt.addrlen);
a246b010 1708 do {
beb59b68
PE		 1709 rpc_set_port((struct sockaddr *)&myaddr, port);
1710 err = kernel_bind(sock, (struct sockaddr *)&myaddr,
1711 transport->xprt.addrlen);
67a391d7 1712 if (port == 0)
d3bc9a1d 1713 break;
a246b010 1714 if (err == 0) {
fbfffbd5 1715 transport->srcport = port;
d3bc9a1d 1716 break;
a246b010 1717 }
67a391d7 1718 last = port;
baaf4e48 1719 port = xs_next_srcport(transport, port);
67a391d7
TM		 1720 if (port > last)
1721 nloop++;
1722 } while (err == -EADDRINUSE && nloop != 2);
90058d37 1723
4232e863 1724 if (myaddr.ss_family == AF_INET)
beb59b68
PE		 1725 dprintk("RPC: %s %pI4:%u: %s (%d)\n", __func__,
1726 &((struct sockaddr_in *)&myaddr)->sin_addr,
1727 port, err ? "failed" : "ok", err);
1728 else
1729 dprintk("RPC: %s %pI6:%u: %s (%d)\n", __func__,
1730 &((struct sockaddr_in6 *)&myaddr)->sin6_addr,
1731 port, err ? "failed" : "ok", err);
a246b010
CL		 1732 return err;
1733}
1734
176e21ee
CL		 1735/*
1736 * We don't support autobind on AF_LOCAL sockets
1737 */
1738static void xs_local_rpcbind(struct rpc_task *task)
1739{
3dc0da27
TM		 1740 rcu_read_lock();
1741 xprt_set_bound(rcu_dereference(task->tk_client->cl_xprt));
1742 rcu_read_unlock();
176e21ee
CL		 1743}
1744
1745static void xs_local_set_port(struct rpc_xprt *xprt, unsigned short port)
1746{
1747}
beb59b68 1748
ed07536e
PZ		 1749#ifdef CONFIG_DEBUG_LOCK_ALLOC
1750static struct lock_class_key xs_key[2];
1751static struct lock_class_key xs_slock_key[2];
1752
176e21ee
CL		 1753static inline void xs_reclassify_socketu(struct socket *sock)
1754{
1755 struct sock *sk = sock->sk;
1756
176e21ee
CL		 1757 sock_lock_init_class_and_name(sk, "slock-AF_LOCAL-RPC",
1758 &xs_slock_key[1], "sk_lock-AF_LOCAL-RPC", &xs_key[1]);
1759}
1760
8945ee5e 1761static inline void xs_reclassify_socket4(struct socket *sock)
ed07536e
PZ		 1762{
1763 struct sock *sk = sock->sk;
8945ee5e 1764
8945ee5e
CL		 1765 sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1766 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1767}
ed07536e 1768
8945ee5e
CL		 1769static inline void xs_reclassify_socket6(struct socket *sock)
1770{
1771 struct sock *sk = sock->sk;
ed07536e 1772
8945ee5e
CL		 1773 sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1774 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
ed07536e 1775}
6bc9638a
PE		 1776
1777static inline void xs_reclassify_socket(int family, struct socket *sock)
1778{
1b7a1819
WAA		 1779 WARN_ON_ONCE(sock_owned_by_user(sock->sk));
1780 if (sock_owned_by_user(sock->sk))
1781 return;
1782
4232e863 1783 switch (family) {
176e21ee
CL		 1784 case AF_LOCAL:
1785 xs_reclassify_socketu(sock);
1786 break;
4232e863 1787 case AF_INET:
6bc9638a 1788 xs_reclassify_socket4(sock);
4232e863
CL		 1789 break;
1790 case AF_INET6:
6bc9638a 1791 xs_reclassify_socket6(sock);
4232e863
CL		 1792 break;
1793 }
6bc9638a 1794}
ed07536e 1795#else
176e21ee
CL		 1796static inline void xs_reclassify_socketu(struct socket *sock)
1797{
1798}
1799
8945ee5e
CL		 1800static inline void xs_reclassify_socket4(struct socket *sock)
1801{
1802}
1803
1804static inline void xs_reclassify_socket6(struct socket *sock)
ed07536e
PZ		 1805{
1806}
6bc9638a
PE		 1807
1808static inline void xs_reclassify_socket(int family, struct socket *sock)
1809{
1810}
ed07536e
PZ		 1811#endif
1812
6bc9638a
PE		 1813static struct socket *xs_create_sock(struct rpc_xprt *xprt,
1814 struct sock_xprt *transport, int family, int type, int protocol)
22f79326
PE		 1815{
1816 struct socket *sock;
1817 int err;
1818
6bc9638a 1819 err = __sock_create(xprt->xprt_net, family, type, protocol, &sock, 1);
22f79326
PE		 1820 if (err < 0) {
1821 dprintk("RPC: can't create %d transport socket (%d).\n",
1822 protocol, -err);
1823 goto out;
1824 }
6bc9638a 1825 xs_reclassify_socket(family, sock);
22f79326 1826
4cea288a
BH		 1827 err = xs_bind(transport, sock);
1828 if (err) {
22f79326
PE		 1829 sock_release(sock);
1830 goto out;
1831 }
1832
1833 return sock;
1834out:
1835 return ERR_PTR(err);
1836}
1837
176e21ee
CL		 1838static int xs_local_finish_connecting(struct rpc_xprt *xprt,
1839 struct socket *sock)
1840{
1841 struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1842 xprt);
1843
1844 if (!transport->inet) {
1845 struct sock *sk = sock->sk;
1846
1847 write_lock_bh(&sk->sk_callback_lock);
1848
1849 xs_save_old_callbacks(transport, sk);
1850
1851 sk->sk_user_data = xprt;
1852 sk->sk_data_ready = xs_local_data_ready;
1853 sk->sk_write_space = xs_udp_write_space;
176e21ee
CL		 1854 sk->sk_allocation = GFP_ATOMIC;
1855
1856 xprt_clear_connected(xprt);
1857
1858 /* Reset to new socket */
1859 transport->sock = sock;
1860 transport->inet = sk;
1861
1862 write_unlock_bh(&sk->sk_callback_lock);
1863 }
1864
1865 /* Tell the socket layer to start connecting... */
1866 xprt->stat.connect_count++;
1867 xprt->stat.connect_start = jiffies;
1868 return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, 0);
1869}
1870
1871/**
1872 * xs_local_setup_socket - create AF_LOCAL socket, connect to a local endpoint
1873 * @xprt: RPC transport to connect
1874 * @transport: socket transport to connect
1875 * @create_sock: function to create a socket of the correct type
176e21ee 1876 */
dc107402 1877static int xs_local_setup_socket(struct sock_xprt *transport)
176e21ee 1878{
176e21ee
CL		 1879 struct rpc_xprt *xprt = &transport->xprt;
1880 struct socket *sock;
1881 int status = -EIO;
1882
5cf02d09
JL		 1883 current->flags |= PF_FSTRANS;
1884
176e21ee
CL		 1885 clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1886 status = __sock_create(xprt->xprt_net, AF_LOCAL,
1887 SOCK_STREAM, 0, &sock, 1);
1888 if (status < 0) {
1889 dprintk("RPC: can't create AF_LOCAL "
1890 "transport socket (%d).\n", -status);
1891 goto out;
1892 }
1893 xs_reclassify_socketu(sock);
1894
1895 dprintk("RPC: worker connecting xprt %p via AF_LOCAL to %s\n",
1896 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1897
1898 status = xs_local_finish_connecting(xprt, sock);
1899 switch (status) {
1900 case 0:
1901 dprintk("RPC: xprt %p connected to %s\n",
1902 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1903 xprt_set_connected(xprt);
1904 break;
1905 case -ENOENT:
1906 dprintk("RPC: xprt %p: socket %s does not exist\n",
1907 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1908 break;
4a20a988
TM		 1909 case -ECONNREFUSED:
1910 dprintk("RPC: xprt %p: connection refused for %s\n",
1911 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1912 break;
176e21ee
CL		 1913 default:
1914 printk(KERN_ERR "%s: unhandled error (%d) connecting to %s\n",
1915 __func__, -status,
1916 xprt->address_strings[RPC_DISPLAY_ADDR]);
1917 }
1918
1919out:
1920 xprt_clear_connecting(xprt);
1921 xprt_wake_pending_tasks(xprt, status);
5cf02d09 1922 current->flags &= ~PF_FSTRANS;
dc107402
BF		 1923 return status;
1924}
1925
b6669737 1926static void xs_local_connect(struct rpc_xprt *xprt, struct rpc_task *task)
dc107402 1927{
dc107402
BF		 1928 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1929 int ret;
1930
1931 if (RPC_IS_ASYNC(task)) {
1932 /*
1933 * We want the AF_LOCAL connect to be resolved in the
1934 * filesystem namespace of the process making the rpc
1935 * call. Thus we connect synchronously.
1936 *
1937 * If we want to support asynchronous AF_LOCAL calls,
1938 * we'll need to figure out how to pass a namespace to
1939 * connect.
1940 */
1941 rpc_exit(task, -ENOTCONN);
1942 return;
1943 }
1944 ret = xs_local_setup_socket(transport);
1945 if (ret && !RPC_IS_SOFTCONN(task))
1946 msleep_interruptible(15000);
176e21ee
CL		 1947}
1948
a564b8f0
MG		 1949#ifdef CONFIG_SUNRPC_SWAP
1950static void xs_set_memalloc(struct rpc_xprt *xprt)
1951{
1952 struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1953 xprt);
1954
1955 if (xprt->swapper)
1956 sk_set_memalloc(transport->inet);
1957}
1958
1959/**
1960 * xs_swapper - Tag this transport as being used for swap.
1961 * @xprt: transport to tag
1962 * @enable: enable/disable
1963 *
1964 */
1965int xs_swapper(struct rpc_xprt *xprt, int enable)
1966{
1967 struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1968 xprt);
1969 int err = 0;
1970
1971 if (enable) {
1972 xprt->swapper++;
1973 xs_set_memalloc(xprt);
1974 } else if (xprt->swapper) {
1975 xprt->swapper--;
1976 sk_clear_memalloc(transport->inet);
1977 }
1978
1979 return err;
1980}
1981EXPORT_SYMBOL_GPL(xs_swapper);
1982#else
1983static void xs_set_memalloc(struct rpc_xprt *xprt)
1984{
1985}
1986#endif
1987
16be2d20
CL		 1988static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
1989{
1990 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1991
1992 if (!transport->inet) {
1993 struct sock *sk = sock->sk;
1994
1995 write_lock_bh(&sk->sk_callback_lock);
1996
2a9e1cfa
TM		 1997 xs_save_old_callbacks(transport, sk);
1998
16be2d20 1999 sk->sk_user_data = xprt;
16be2d20
CL		 2000 sk->sk_data_ready = xs_udp_data_ready;
2001 sk->sk_write_space = xs_udp_write_space;
2002 sk->sk_no_check = UDP_CSUM_NORCV;
2003 sk->sk_allocation = GFP_ATOMIC;
2004
2005 xprt_set_connected(xprt);
2006
2007 /* Reset to new socket */
2008 transport->sock = sock;
2009 transport->inet = sk;
2010
a564b8f0
MG		 2011 xs_set_memalloc(xprt);
2012
16be2d20
CL		 2013 write_unlock_bh(&sk->sk_callback_lock);
2014 }
2015 xs_udp_do_set_buffer_size(xprt);
2016}
2017
8c14ff2a 2018static void xs_udp_setup_socket(struct work_struct *work)
a246b010 2019{
34161db6
TM		 2020 struct sock_xprt *transport =
2021 container_of(work, struct sock_xprt, connect_worker.work);
c8475461 2022 struct rpc_xprt *xprt = &transport->xprt;
ee0ac0c2 2023 struct socket *sock = transport->sock;
b65c0310 2024 int status = -EIO;
9903cd1c 2025
5cf02d09
JL		 2026 current->flags |= PF_FSTRANS;
2027
b0d93ad5 2028 /* Start by resetting any existing state */
fe315e76 2029 xs_reset_transport(transport);
8c14ff2a
PE		 2030 sock = xs_create_sock(xprt, transport,
2031 xs_addr(xprt)->sa_family, SOCK_DGRAM, IPPROTO_UDP);
b65c0310 2032 if (IS_ERR(sock))
b0d93ad5 2033 goto out;
68e220bd 2034
c740eff8
CL		 2035 dprintk("RPC: worker connecting xprt %p via %s to "
2036 "%s (port %s)\n", xprt,
2037 xprt->address_strings[RPC_DISPLAY_PROTO],
2038 xprt->address_strings[RPC_DISPLAY_ADDR],
2039 xprt->address_strings[RPC_DISPLAY_PORT]);
68e220bd
CL		 2040
2041 xs_udp_finish_connecting(xprt, sock);
b0d93ad5
CL		 2042 status = 0;
2043out:
b0d93ad5 2044 xprt_clear_connecting(xprt);
7d1e8255 2045 xprt_wake_pending_tasks(xprt, status);
5cf02d09 2046 current->flags &= ~PF_FSTRANS;
a246b010
CL		 2047}
2048
3167e12c
CL		 2049/*
2050 * We need to preserve the port number so the reply cache on the server can
2051 * find our cached RPC replies when we get around to reconnecting.
2052 */
58dddac9 2053static void xs_abort_connection(struct sock_xprt *transport)
3167e12c
CL		 2054{
2055 int result;
3167e12c
CL		 2056 struct sockaddr any;
2057
58dddac9 2058 dprintk("RPC: disconnecting xprt %p to reuse port\n", transport);
3167e12c
CL		 2059
2060 /*
2061 * Disconnect the transport socket by doing a connect operation
2062 * with AF_UNSPEC. This should return immediately...
2063 */
2064 memset(&any, 0, sizeof(any));
2065 any.sa_family = AF_UNSPEC;
ee0ac0c2 2066 result = kernel_connect(transport->sock, &any, sizeof(any), 0);
7d1e8255 2067 if (!result)
4bc1e68e
TM		 2068 xs_sock_reset_connection_flags(&transport->xprt);
2069 dprintk("RPC: AF_UNSPEC connect return code %d\n", result);
3167e12c
CL		 2070}
2071
58dddac9 2072static void xs_tcp_reuse_connection(struct sock_xprt *transport)
40d2549d
TM		 2073{
2074 unsigned int state = transport->inet->sk_state;
2075
669502ff
AC		 2076 if (state == TCP_CLOSE && transport->sock->state == SS_UNCONNECTED) {
2077 /* we don't need to abort the connection if the socket
2078 * hasn't undergone a shutdown
2079 */
2080 if (transport->inet->sk_shutdown == 0)
2081 return;
2082 dprintk("RPC: %s: TCP_CLOSEd and sk_shutdown set to %d\n",
2083 __func__, transport->inet->sk_shutdown);
2084 }
2085 if ((1 << state) & (TCPF_ESTABLISHED|TCPF_SYN_SENT)) {
2086 /* we don't need to abort the connection if the socket
2087 * hasn't undergone a shutdown
2088 */
2089 if (transport->inet->sk_shutdown == 0)
2090 return;
2091 dprintk("RPC: %s: ESTABLISHED/SYN_SENT "
2092 "sk_shutdown set to %d\n",
2093 __func__, transport->inet->sk_shutdown);
2094 }
58dddac9 2095 xs_abort_connection(transport);
40d2549d
TM		 2096}
2097
16be2d20 2098static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
a246b010 2099{
16be2d20 2100 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
fe19a96b 2101 int ret = -ENOTCONN;
edb267a6 2102
ee0ac0c2 2103 if (!transport->inet) {
b0d93ad5
CL		 2104 struct sock *sk = sock->sk;
2105
2106 write_lock_bh(&sk->sk_callback_lock);
2107
2a9e1cfa
TM		 2108 xs_save_old_callbacks(transport, sk);
2109
b0d93ad5 2110 sk->sk_user_data = xprt;
b0d93ad5
CL		 2111 sk->sk_data_ready = xs_tcp_data_ready;
2112 sk->sk_state_change = xs_tcp_state_change;
2113 sk->sk_write_space = xs_tcp_write_space;
b079fa7b 2114 sk->sk_allocation = GFP_ATOMIC;
3167e12c
CL		 2115
2116 /* socket options */
2117 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
2118 sock_reset_flag(sk, SOCK_LINGER);
2119 tcp_sk(sk)->linger2 = 0;
2120 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
b0d93ad5
CL		 2121
2122 xprt_clear_connected(xprt);
2123
2124 /* Reset to new socket */
ee0ac0c2
CL		 2125 transport->sock = sock;
2126 transport->inet = sk;
b0d93ad5
CL		 2127
2128 write_unlock_bh(&sk->sk_callback_lock);
2129 }
2130
01d37c42 2131 if (!xprt_bound(xprt))
fe19a96b 2132 goto out;
01d37c42 2133
a564b8f0
MG		 2134 xs_set_memalloc(xprt);
2135
b0d93ad5 2136 /* Tell the socket layer to start connecting... */
262ca07d
CL		 2137 xprt->stat.connect_count++;
2138 xprt->stat.connect_start = jiffies;
fe19a96b
TM		 2139 ret = kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
2140 switch (ret) {
2141 case 0:
2142 case -EINPROGRESS:
2143 /* SYN_SENT! */
2144 xprt->connect_cookie++;
2145 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2146 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2147 }
2148out:
2149 return ret;
16be2d20
CL		 2150}
2151
9903cd1c 2152/**
b61d59ff
TM		 2153 * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint
2154 * @xprt: RPC transport to connect
2155 * @transport: socket transport to connect
2156 * @create_sock: function to create a socket of the correct type
9903cd1c
CL		 2157 *
2158 * Invoked by a work queue tasklet.
a246b010 2159 */
cdd518d5 2160static void xs_tcp_setup_socket(struct work_struct *work)
a246b010 2161{
cdd518d5
PE		 2162 struct sock_xprt *transport =
2163 container_of(work, struct sock_xprt, connect_worker.work);
ee0ac0c2 2164 struct socket *sock = transport->sock;
a9f5f0f7 2165 struct rpc_xprt *xprt = &transport->xprt;
b61d59ff 2166 int status = -EIO;
a246b010 2167
5cf02d09
JL		 2168 current->flags |= PF_FSTRANS;
2169
ee0ac0c2 2170 if (!sock) {
7d1e8255 2171 clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
cdd518d5
PE		 2172 sock = xs_create_sock(xprt, transport,
2173 xs_addr(xprt)->sa_family, SOCK_STREAM, IPPROTO_TCP);
b61d59ff
TM		 2174 if (IS_ERR(sock)) {
2175 status = PTR_ERR(sock);
3167e12c
CL		 2176 goto out;
2177 }
7d1e8255
TM		 2178 } else {
2179 int abort_and_exit;
a246b010 2180
7d1e8255
TM		 2181 abort_and_exit = test_and_clear_bit(XPRT_CONNECTION_ABORT,
2182 &xprt->state);
3167e12c 2183 /* "close" the socket, preserving the local port */
58dddac9 2184 xs_tcp_reuse_connection(transport);
a246b010 2185
7d1e8255
TM		 2186 if (abort_and_exit)
2187 goto out_eagain;
2188 }
a246b010 2189
c740eff8
CL		 2190 dprintk("RPC: worker connecting xprt %p via %s to "
2191 "%s (port %s)\n", xprt,
2192 xprt->address_strings[RPC_DISPLAY_PROTO],
2193 xprt->address_strings[RPC_DISPLAY_ADDR],
2194 xprt->address_strings[RPC_DISPLAY_PORT]);
edb267a6 2195
16be2d20 2196 status = xs_tcp_finish_connecting(xprt, sock);
46121cf7
CL		 2197 dprintk("RPC: %p connect status %d connected %d sock state %d\n",
2198 xprt, -status, xprt_connected(xprt),
2199 sock->sk->sk_state);
2a491991 2200 switch (status) {
f75e6745
TM		 2201 default:
2202 printk("%s: connect returned unhandled error %d\n",
2203 __func__, status);
2204 case -EADDRNOTAVAIL:
2205 /* We're probably in TIME_WAIT. Get rid of existing socket,
2206 * and retry
2207 */
a519fc7a 2208 xs_tcp_force_close(xprt);
88b5ed73 2209 break;
8a2cec29
TM		 2210 case -ECONNREFUSED:
2211 case -ECONNRESET:
2212 case -ENETUNREACH:
2213 /* retry with existing socket, after a delay */
2a491991
TM		 2214 case 0:
2215 case -EINPROGRESS:
2216 case -EALREADY:
7d1e8255 2217 xprt_clear_connecting(xprt);
5cf02d09 2218 current->flags &= ~PF_FSTRANS;
7d1e8255 2219 return;
9fcfe0c8
TM		 2220 case -EINVAL:
2221 /* Happens, for instance, if the user specified a link
2222 * local IPv6 address without a scope-id.
2223 */
2224 goto out;
a246b010 2225 }
7d1e8255 2226out_eagain:
2a491991 2227 status = -EAGAIN;
a246b010 2228out:
68e220bd 2229 xprt_clear_connecting(xprt);
7d1e8255 2230 xprt_wake_pending_tasks(xprt, status);
5cf02d09 2231 current->flags &= ~PF_FSTRANS;
68e220bd 2232}
b0d93ad5 2233
9903cd1c
CL		 2234/**
2235 * xs_connect - connect a socket to a remote endpoint
1b092092 2236 * @xprt: pointer to transport structure
9903cd1c
CL		 2237 * @task: address of RPC task that manages state of connect request
2238 *
2239 * TCP: If the remote end dropped the connection, delay reconnecting.
03bf4b70
CL		 2240 *
2241 * UDP socket connects are synchronous, but we use a work queue anyway
2242 * to guarantee that even unprivileged user processes can set up a
2243 * socket on a privileged port.
2244 *
2245 * If a UDP socket connect fails, the delay behavior here prevents
2246 * retry floods (hard mounts).
9903cd1c 2247 */
1b092092 2248static void xs_connect(struct rpc_xprt *xprt, struct rpc_task *task)
a246b010 2249{
ee0ac0c2 2250 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
a246b010 2251
09a21c41 2252 if (transport->sock != NULL && !RPC_IS_SOFTCONN(task)) {
46121cf7
CL		 2253 dprintk("RPC: xs_connect delayed xprt %p for %lu "
2254 "seconds\n",
03bf4b70 2255 xprt, xprt->reestablish_timeout / HZ);
c1384c9c
TM		 2256 queue_delayed_work(rpciod_workqueue,
2257 &transport->connect_worker,
2258 xprt->reestablish_timeout);
03bf4b70 2259 xprt->reestablish_timeout <<= 1;
61d0a8e6
NB		 2260 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2261 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
03bf4b70
CL		 2262 if (xprt->reestablish_timeout > XS_TCP_MAX_REEST_TO)
2263 xprt->reestablish_timeout = XS_TCP_MAX_REEST_TO;
b0d93ad5 2264 } else {
46121cf7 2265 dprintk("RPC: xs_connect scheduled xprt %p\n", xprt);
c1384c9c
TM		 2266 queue_delayed_work(rpciod_workqueue,
2267 &transport->connect_worker, 0);
a246b010
CL		 2268 }
2269}
2270
176e21ee
CL		 2271/**
2272 * xs_local_print_stats - display AF_LOCAL socket-specifc stats
2273 * @xprt: rpc_xprt struct containing statistics
2274 * @seq: output file
2275 *
2276 */
2277static void xs_local_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2278{
2279 long idle_time = 0;
2280
2281 if (xprt_connected(xprt))
2282 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2283
2284 seq_printf(seq, "\txprt:\tlocal %lu %lu %lu %ld %lu %lu %lu "
15a45206 2285 "%llu %llu %lu %llu %llu\n",
176e21ee
CL		 2286 xprt->stat.bind_count,
2287 xprt->stat.connect_count,
2288 xprt->stat.connect_time,
2289 idle_time,
2290 xprt->stat.sends,
2291 xprt->stat.recvs,
2292 xprt->stat.bad_xids,
2293 xprt->stat.req_u,
15a45206
AA		 2294 xprt->stat.bklog_u,
2295 xprt->stat.max_slots,
2296 xprt->stat.sending_u,
2297 xprt->stat.pending_u);
176e21ee
CL		 2298}
2299
262ca07d
CL		 2300/**
2301 * xs_udp_print_stats - display UDP socket-specifc stats
2302 * @xprt: rpc_xprt struct containing statistics
2303 * @seq: output file
2304 *
2305 */
2306static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2307{
c8475461
CL		 2308 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2309
15a45206
AA		 2310 seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %llu %llu "
2311 "%lu %llu %llu\n",
fbfffbd5 2312 transport->srcport,
262ca07d
CL		 2313 xprt->stat.bind_count,
2314 xprt->stat.sends,
2315 xprt->stat.recvs,
2316 xprt->stat.bad_xids,
2317 xprt->stat.req_u,
15a45206
AA		 2318 xprt->stat.bklog_u,
2319 xprt->stat.max_slots,
2320 xprt->stat.sending_u,
2321 xprt->stat.pending_u);
262ca07d
CL		 2322}
2323
2324/**
2325 * xs_tcp_print_stats - display TCP socket-specifc stats
2326 * @xprt: rpc_xprt struct containing statistics
2327 * @seq: output file
2328 *
2329 */
2330static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2331{
c8475461 2332 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
262ca07d
CL		 2333 long idle_time = 0;
2334
2335 if (xprt_connected(xprt))
2336 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2337
15a45206
AA		 2338 seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu "
2339 "%llu %llu %lu %llu %llu\n",
fbfffbd5 2340 transport->srcport,
262ca07d
CL		 2341 xprt->stat.bind_count,
2342 xprt->stat.connect_count,
2343 xprt->stat.connect_time,
2344 idle_time,
2345 xprt->stat.sends,
2346 xprt->stat.recvs,
2347 xprt->stat.bad_xids,
2348 xprt->stat.req_u,
15a45206
AA		 2349 xprt->stat.bklog_u,
2350 xprt->stat.max_slots,
2351 xprt->stat.sending_u,
2352 xprt->stat.pending_u);
262ca07d
CL		 2353}
2354
4cfc7e60
RI		 2355/*
2356 * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason
2357 * we allocate pages instead doing a kmalloc like rpc_malloc is because we want
2358 * to use the server side send routines.
2359 */
5a51f13a 2360static void *bc_malloc(struct rpc_task *task, size_t size)
4cfc7e60
RI		 2361{
2362 struct page *page;
2363 struct rpc_buffer *buf;
2364
b8a13d03
WAA		 2365 WARN_ON_ONCE(size > PAGE_SIZE - sizeof(struct rpc_buffer));
2366 if (size > PAGE_SIZE - sizeof(struct rpc_buffer))
2367 return NULL;
4cfc7e60 2368
b8a13d03 2369 page = alloc_page(GFP_KERNEL);
4cfc7e60
RI		 2370 if (!page)
2371 return NULL;
2372
2373 buf = page_address(page);
2374 buf->len = PAGE_SIZE;
2375
2376 return buf->data;
2377}
2378
2379/*
2380 * Free the space allocated in the bc_alloc routine
2381 */
5a51f13a 2382static void bc_free(void *buffer)
4cfc7e60
RI		 2383{
2384 struct rpc_buffer *buf;
2385
2386 if (!buffer)
2387 return;
2388
2389 buf = container_of(buffer, struct rpc_buffer, data);
2390 free_page((unsigned long)buf);
2391}
2392
2393/*
2394 * Use the svc_sock to send the callback. Must be called with svsk->sk_mutex
2395 * held. Borrows heavily from svc_tcp_sendto and xs_tcp_send_request.
2396 */
2397static int bc_sendto(struct rpc_rqst *req)
2398{
2399 int len;
2400 struct xdr_buf *xbufp = &req->rq_snd_buf;
2401 struct rpc_xprt *xprt = req->rq_xprt;
2402 struct sock_xprt *transport =
2403 container_of(xprt, struct sock_xprt, xprt);
2404 struct socket *sock = transport->sock;
2405 unsigned long headoff;
2406 unsigned long tailoff;
2407
61677eee 2408 xs_encode_stream_record_marker(xbufp);
4cfc7e60
RI		 2409
2410 tailoff = (unsigned long)xbufp->tail[0].iov_base & ~PAGE_MASK;
2411 headoff = (unsigned long)xbufp->head[0].iov_base & ~PAGE_MASK;
2412 len = svc_send_common(sock, xbufp,
2413 virt_to_page(xbufp->head[0].iov_base), headoff,
2414 xbufp->tail[0].iov_base, tailoff);
2415
2416 if (len != xbufp->len) {
2417 printk(KERN_NOTICE "Error sending entire callback!\n");
2418 len = -EAGAIN;
2419 }
2420
2421 return len;
2422}
2423
2424/*
2425 * The send routine. Borrows from svc_send
2426 */
2427static int bc_send_request(struct rpc_task *task)
2428{
2429 struct rpc_rqst *req = task->tk_rqstp;
2430 struct svc_xprt *xprt;
4cfc7e60
RI		 2431 u32 len;
2432
2433 dprintk("sending request with xid: %08x\n", ntohl(req->rq_xid));
2434 /*
2435 * Get the server socket associated with this callback xprt
2436 */
2437 xprt = req->rq_xprt->bc_xprt;
4cfc7e60
RI		 2438
2439 /*
2440 * Grab the mutex to serialize data as the connection is shared
2441 * with the fore channel
2442 */
2443 if (!mutex_trylock(&xprt->xpt_mutex)) {
2444 rpc_sleep_on(&xprt->xpt_bc_pending, task, NULL);
2445 if (!mutex_trylock(&xprt->xpt_mutex))
2446 return -EAGAIN;
2447 rpc_wake_up_queued_task(&xprt->xpt_bc_pending, task);
2448 }
2449 if (test_bit(XPT_DEAD, &xprt->xpt_flags))
2450 len = -ENOTCONN;
2451 else
2452 len = bc_sendto(req);
2453 mutex_unlock(&xprt->xpt_mutex);
2454
2455 if (len > 0)
2456 len = 0;
2457
2458 return len;
2459}
2460
2461/*
2462 * The close routine. Since this is client initiated, we do nothing
2463 */
2464
2465static void bc_close(struct rpc_xprt *xprt)
2466{
4cfc7e60
RI		 2467}
2468
2469/*
2470 * The xprt destroy routine. Again, because this connection is client
2471 * initiated, we do nothing
2472 */
2473
2474static void bc_destroy(struct rpc_xprt *xprt)
2475{
4cfc7e60
RI		 2476}
2477
176e21ee
CL		 2478static struct rpc_xprt_ops xs_local_ops = {
2479 .reserve_xprt = xprt_reserve_xprt,
2480 .release_xprt = xs_tcp_release_xprt,
f39c1bfb 2481 .alloc_slot = xprt_alloc_slot,
176e21ee
CL		 2482 .rpcbind = xs_local_rpcbind,
2483 .set_port = xs_local_set_port,
dc107402 2484 .connect = xs_local_connect,
176e21ee
CL		 2485 .buf_alloc = rpc_malloc,
2486 .buf_free = rpc_free,
2487 .send_request = xs_local_send_request,
2488 .set_retrans_timeout = xprt_set_retrans_timeout_def,
2489 .close = xs_close,
190b1ecf 2490 .destroy = xs_local_destroy,
176e21ee
CL		 2491 .print_stats = xs_local_print_stats,
2492};
2493
262965f5 2494static struct rpc_xprt_ops xs_udp_ops = {
43118c29 2495 .set_buffer_size = xs_udp_set_buffer_size,
12a80469 2496 .reserve_xprt = xprt_reserve_xprt_cong,
49e9a890 2497 .release_xprt = xprt_release_xprt_cong,
f39c1bfb 2498 .alloc_slot = xprt_alloc_slot,
45160d62 2499 .rpcbind = rpcb_getport_async,
92200412 2500 .set_port = xs_set_port,
262965f5 2501 .connect = xs_connect,
02107148
CL		 2502 .buf_alloc = rpc_malloc,
2503 .buf_free = rpc_free,
262965f5 2504 .send_request = xs_udp_send_request,
fe3aca29 2505 .set_retrans_timeout = xprt_set_retrans_timeout_rtt,
46c0ee8b 2506 .timer = xs_udp_timer,
a58dd398 2507 .release_request = xprt_release_rqst_cong,
262965f5
CL		 2508 .close = xs_close,
2509 .destroy = xs_destroy,
262ca07d 2510 .print_stats = xs_udp_print_stats,
262965f5
CL		 2511};
2512
2513static struct rpc_xprt_ops xs_tcp_ops = {
12a80469 2514 .reserve_xprt = xprt_reserve_xprt,
e0ab53de 2515 .release_xprt = xs_tcp_release_xprt,
f39c1bfb 2516 .alloc_slot = xprt_lock_and_alloc_slot,
45160d62 2517 .rpcbind = rpcb_getport_async,
92200412 2518 .set_port = xs_set_port,
0b9e7943 2519 .connect = xs_connect,
02107148
CL		 2520 .buf_alloc = rpc_malloc,
2521 .buf_free = rpc_free,
262965f5 2522 .send_request = xs_tcp_send_request,
fe3aca29 2523 .set_retrans_timeout = xprt_set_retrans_timeout_def,
f75e6745 2524 .close = xs_tcp_close,
9903cd1c 2525 .destroy = xs_destroy,
262ca07d 2526 .print_stats = xs_tcp_print_stats,
a246b010
CL		 2527};
2528
4cfc7e60
RI		 2529/*
2530 * The rpc_xprt_ops for the server backchannel
2531 */
2532
2533static struct rpc_xprt_ops bc_tcp_ops = {
2534 .reserve_xprt = xprt_reserve_xprt,
2535 .release_xprt = xprt_release_xprt,
84e28a30 2536 .alloc_slot = xprt_alloc_slot,
92769108 2537 .rpcbind = xs_local_rpcbind,
4cfc7e60
RI		 2538 .buf_alloc = bc_malloc,
2539 .buf_free = bc_free,
2540 .send_request = bc_send_request,
2541 .set_retrans_timeout = xprt_set_retrans_timeout_def,
2542 .close = bc_close,
2543 .destroy = bc_destroy,
2544 .print_stats = xs_tcp_print_stats,
2545};
2546
92476850
CL		 2547static int xs_init_anyaddr(const int family, struct sockaddr *sap)
2548{
2549 static const struct sockaddr_in sin = {
2550 .sin_family = AF_INET,
2551 .sin_addr.s_addr = htonl(INADDR_ANY),
2552 };
2553 static const struct sockaddr_in6 sin6 = {
2554 .sin6_family = AF_INET6,
2555 .sin6_addr = IN6ADDR_ANY_INIT,
2556 };
2557
2558 switch (family) {
176e21ee
CL		 2559 case AF_LOCAL:
2560 break;
92476850
CL		 2561 case AF_INET:
2562 memcpy(sap, &sin, sizeof(sin));
2563 break;
2564 case AF_INET6:
2565 memcpy(sap, &sin6, sizeof(sin6));
2566 break;
2567 default:
2568 dprintk("RPC: %s: Bad address family\n", __func__);
2569 return -EAFNOSUPPORT;
2570 }
2571 return 0;
2572}
2573
3c341b0b 2574static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
d9ba131d
TM		 2575 unsigned int slot_table_size,
2576 unsigned int max_slot_table_size)
c8541ecd
CL		 2577{
2578 struct rpc_xprt *xprt;
ffc2e518 2579 struct sock_xprt *new;
c8541ecd 2580
96802a09 2581 if (args->addrlen > sizeof(xprt->addr)) {
46121cf7 2582 dprintk("RPC: xs_setup_xprt: address too large\n");
c8541ecd
CL		 2583 return ERR_PTR(-EBADF);
2584 }
2585
d9ba131d
TM		 2586 xprt = xprt_alloc(args->net, sizeof(*new), slot_table_size,
2587 max_slot_table_size);
bd1722d4 2588 if (xprt == NULL) {
46121cf7
CL		 2589 dprintk("RPC: xs_setup_xprt: couldn't allocate "
2590 "rpc_xprt\n");
c8541ecd
CL		 2591 return ERR_PTR(-ENOMEM);
2592 }
c8541ecd 2593
bd1722d4 2594 new = container_of(xprt, struct sock_xprt, xprt);
96802a09
FM		 2595 memcpy(&xprt->addr, args->dstaddr, args->addrlen);
2596 xprt->addrlen = args->addrlen;
d3bc9a1d 2597 if (args->srcaddr)
fbfffbd5 2598 memcpy(&new->srcaddr, args->srcaddr, args->addrlen);
92476850
CL		 2599 else {
2600 int err;
2601 err = xs_init_anyaddr(args->dstaddr->sa_family,
2602 (struct sockaddr *)&new->srcaddr);
2aa13531
SK		 2603 if (err != 0) {
2604 xprt_free(xprt);
92476850 2605 return ERR_PTR(err);
2aa13531 2606 }
92476850 2607 }
c8541ecd
CL		 2608
2609 return xprt;
2610}
2611
176e21ee
CL		 2612static const struct rpc_timeout xs_local_default_timeout = {
2613 .to_initval = 10 * HZ,
2614 .to_maxval = 10 * HZ,
2615 .to_retries = 2,
2616};
2617
2618/**
2619 * xs_setup_local - Set up transport to use an AF_LOCAL socket
2620 * @args: rpc transport creation arguments
2621 *
2622 * AF_LOCAL is a "tpi_cots_ord" transport, just like TCP
2623 */
2624static struct rpc_xprt *xs_setup_local(struct xprt_create *args)
2625{
2626 struct sockaddr_un *sun = (struct sockaddr_un *)args->dstaddr;
2627 struct sock_xprt *transport;
2628 struct rpc_xprt *xprt;
2629 struct rpc_xprt *ret;
2630
d9ba131d
TM		 2631 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2632 xprt_max_tcp_slot_table_entries);
176e21ee
CL		 2633 if (IS_ERR(xprt))
2634 return xprt;
2635 transport = container_of(xprt, struct sock_xprt, xprt);
2636
2637 xprt->prot = 0;
2638 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2639 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2640
2641 xprt->bind_timeout = XS_BIND_TO;
2642 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2643 xprt->idle_timeout = XS_IDLE_DISC_TO;
2644
2645 xprt->ops = &xs_local_ops;
2646 xprt->timeout = &xs_local_default_timeout;
2647
2648 switch (sun->sun_family) {
2649 case AF_LOCAL:
2650 if (sun->sun_path[0] != '/') {
2651 dprintk("RPC: bad AF_LOCAL address: %s\n",
2652 sun->sun_path);
2653 ret = ERR_PTR(-EINVAL);
2654 goto out_err;
2655 }
2656 xprt_set_bound(xprt);
176e21ee
CL		 2657 xs_format_peer_addresses(xprt, "local", RPCBIND_NETID_LOCAL);
2658 break;
2659 default:
2660 ret = ERR_PTR(-EAFNOSUPPORT);
2661 goto out_err;
2662 }
2663
2664 dprintk("RPC: set up xprt to %s via AF_LOCAL\n",
2665 xprt->address_strings[RPC_DISPLAY_ADDR]);
2666
2667 if (try_module_get(THIS_MODULE))
2668 return xprt;
2669 ret = ERR_PTR(-EINVAL);
2670out_err:
2671 xprt_free(xprt);
2672 return ret;
2673}
2674
2881ae74
TM		 2675static const struct rpc_timeout xs_udp_default_timeout = {
2676 .to_initval = 5 * HZ,
2677 .to_maxval = 30 * HZ,
2678 .to_increment = 5 * HZ,
2679 .to_retries = 5,
2680};
2681
9903cd1c
CL		 2682/**
2683 * xs_setup_udp - Set up transport to use a UDP socket
96802a09 2684 * @args: rpc transport creation arguments
9903cd1c
CL		 2685 *
2686 */
483066d6 2687static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
a246b010 2688{
8f9d5b1a 2689 struct sockaddr *addr = args->dstaddr;
c8541ecd 2690 struct rpc_xprt *xprt;
c8475461 2691 struct sock_xprt *transport;
0a68b0be 2692 struct rpc_xprt *ret;
a246b010 2693
d9ba131d
TM		 2694 xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries,
2695 xprt_udp_slot_table_entries);
c8541ecd
CL		 2696 if (IS_ERR(xprt))
2697 return xprt;
c8475461 2698 transport = container_of(xprt, struct sock_xprt, xprt);
a246b010 2699
ec739ef0 2700 xprt->prot = IPPROTO_UDP;
808012fb 2701 xprt->tsh_size = 0;
a246b010
CL		 2702 /* XXX: header size can vary due to auth type, IPv6, etc. */
2703 xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
2704
03bf4b70 2705 xprt->bind_timeout = XS_BIND_TO;
03bf4b70
CL		 2706 xprt->reestablish_timeout = XS_UDP_REEST_TO;
2707 xprt->idle_timeout = XS_IDLE_DISC_TO;
a246b010 2708
262965f5 2709 xprt->ops = &xs_udp_ops;
a246b010 2710
ba7392bb 2711 xprt->timeout = &xs_udp_default_timeout;
a246b010 2712
8f9d5b1a
CL		 2713 switch (addr->sa_family) {
2714 case AF_INET:
2715 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2716 xprt_set_bound(xprt);
2717
2718 INIT_DELAYED_WORK(&transport->connect_worker,
8c14ff2a 2719 xs_udp_setup_socket);
9dc3b095 2720 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
8f9d5b1a
CL		 2721 break;
2722 case AF_INET6:
2723 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2724 xprt_set_bound(xprt);
2725
2726 INIT_DELAYED_WORK(&transport->connect_worker,
8c14ff2a 2727 xs_udp_setup_socket);
9dc3b095 2728 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
8f9d5b1a
CL		 2729 break;
2730 default:
0a68b0be
BF		 2731 ret = ERR_PTR(-EAFNOSUPPORT);
2732 goto out_err;
8f9d5b1a
CL		 2733 }
2734
c740eff8
CL		 2735 if (xprt_bound(xprt))
2736 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2737 xprt->address_strings[RPC_DISPLAY_ADDR],
2738 xprt->address_strings[RPC_DISPLAY_PORT],
2739 xprt->address_strings[RPC_DISPLAY_PROTO]);
2740 else
2741 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
2742 xprt->address_strings[RPC_DISPLAY_ADDR],
2743 xprt->address_strings[RPC_DISPLAY_PROTO]);
edb267a6 2744
bc25571e
TT		 2745 if (try_module_get(THIS_MODULE))
2746 return xprt;
0a68b0be
BF		 2747 ret = ERR_PTR(-EINVAL);
2748out_err:
e204e621 2749 xprt_free(xprt);
0a68b0be 2750 return ret;
a246b010
CL		 2751}
2752
2881ae74
TM		 2753static const struct rpc_timeout xs_tcp_default_timeout = {
2754 .to_initval = 60 * HZ,
2755 .to_maxval = 60 * HZ,
2756 .to_retries = 2,
2757};
2758
9903cd1c
CL		 2759/**
2760 * xs_setup_tcp - Set up transport to use a TCP socket
96802a09 2761 * @args: rpc transport creation arguments
9903cd1c
CL		 2762 *
2763 */
483066d6 2764static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
a246b010 2765{
8f9d5b1a 2766 struct sockaddr *addr = args->dstaddr;
c8541ecd 2767 struct rpc_xprt *xprt;
c8475461 2768 struct sock_xprt *transport;
0a68b0be 2769 struct rpc_xprt *ret;
a246b010 2770
d9ba131d
TM		 2771 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2772 xprt_max_tcp_slot_table_entries);
c8541ecd
CL		 2773 if (IS_ERR(xprt))
2774 return xprt;
c8475461 2775 transport = container_of(xprt, struct sock_xprt, xprt);
a246b010 2776
ec739ef0 2777 xprt->prot = IPPROTO_TCP;
808012fb 2778 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
808012fb 2779 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
a246b010 2780
03bf4b70 2781 xprt->bind_timeout = XS_BIND_TO;
03bf4b70
CL		 2782 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2783 xprt->idle_timeout = XS_IDLE_DISC_TO;
a246b010 2784
262965f5 2785 xprt->ops = &xs_tcp_ops;
ba7392bb 2786 xprt->timeout = &xs_tcp_default_timeout;
a246b010 2787
8f9d5b1a
CL		 2788 switch (addr->sa_family) {
2789 case AF_INET:
2790 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2791 xprt_set_bound(xprt);
2792
9dc3b095 2793 INIT_DELAYED_WORK(&transport->connect_worker,
cdd518d5 2794 xs_tcp_setup_socket);
9dc3b095 2795 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
8f9d5b1a
CL		 2796 break;
2797 case AF_INET6:
2798 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2799 xprt_set_bound(xprt);
2800
9dc3b095 2801 INIT_DELAYED_WORK(&transport->connect_worker,
cdd518d5 2802 xs_tcp_setup_socket);
9dc3b095 2803 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
8f9d5b1a
CL		 2804 break;
2805 default:
0a68b0be
BF		 2806 ret = ERR_PTR(-EAFNOSUPPORT);
2807 goto out_err;
8f9d5b1a
CL		 2808 }
2809
c740eff8
CL		 2810 if (xprt_bound(xprt))
2811 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2812 xprt->address_strings[RPC_DISPLAY_ADDR],
2813 xprt->address_strings[RPC_DISPLAY_PORT],
2814 xprt->address_strings[RPC_DISPLAY_PROTO]);
2815 else
2816 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
2817 xprt->address_strings[RPC_DISPLAY_ADDR],
2818 xprt->address_strings[RPC_DISPLAY_PROTO]);
2819
edb267a6 2820
bc25571e
TT		 2821 if (try_module_get(THIS_MODULE))
2822 return xprt;
0a68b0be
BF		 2823 ret = ERR_PTR(-EINVAL);
2824out_err:
e204e621 2825 xprt_free(xprt);
0a68b0be 2826 return ret;
a246b010 2827}
282b32e1 2828
f300baba
AB		 2829/**
2830 * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket
2831 * @args: rpc transport creation arguments
2832 *
2833 */
2834static struct rpc_xprt *xs_setup_bc_tcp(struct xprt_create *args)
2835{
2836 struct sockaddr *addr = args->dstaddr;
2837 struct rpc_xprt *xprt;
2838 struct sock_xprt *transport;
2839 struct svc_sock *bc_sock;
0a68b0be 2840 struct rpc_xprt *ret;
f300baba 2841
f0418aa4
BF		 2842 if (args->bc_xprt->xpt_bc_xprt) {
2843 /*
2844 * This server connection already has a backchannel
2845 * export; we can't create a new one, as we wouldn't be
2846 * able to match replies based on xid any more. So,
2847 * reuse the already-existing one:
2848 */
2849 return args->bc_xprt->xpt_bc_xprt;
2850 }
d9ba131d
TM		 2851 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2852 xprt_tcp_slot_table_entries);
f300baba
AB		 2853 if (IS_ERR(xprt))
2854 return xprt;
2855 transport = container_of(xprt, struct sock_xprt, xprt);
2856
2857 xprt->prot = IPPROTO_TCP;
2858 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2859 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2860 xprt->timeout = &xs_tcp_default_timeout;
2861
2862 /* backchannel */
2863 xprt_set_bound(xprt);
2864 xprt->bind_timeout = 0;
f300baba
AB		 2865 xprt->reestablish_timeout = 0;
2866 xprt->idle_timeout = 0;
2867
f300baba
AB		 2868 xprt->ops = &bc_tcp_ops;
2869
2870 switch (addr->sa_family) {
2871 case AF_INET:
2872 xs_format_peer_addresses(xprt, "tcp",
2873 RPCBIND_NETID_TCP);
2874 break;
2875 case AF_INET6:
2876 xs_format_peer_addresses(xprt, "tcp",
2877 RPCBIND_NETID_TCP6);
2878 break;
2879 default:
0a68b0be
BF		 2880 ret = ERR_PTR(-EAFNOSUPPORT);
2881 goto out_err;
f300baba
AB		 2882 }
2883
50fa0d40
PE		 2884 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2885 xprt->address_strings[RPC_DISPLAY_ADDR],
2886 xprt->address_strings[RPC_DISPLAY_PORT],
2887 xprt->address_strings[RPC_DISPLAY_PROTO]);
f300baba 2888
99de8ea9
BF		 2889 /*
2890 * Once we've associated a backchannel xprt with a connection,
2891 * we want to keep it around as long as long as the connection
2892 * lasts, in case we need to start using it for a backchannel
2893 * again; this reference won't be dropped until bc_xprt is
2894 * destroyed.
2895 */
2896 xprt_get(xprt);
2897 args->bc_xprt->xpt_bc_xprt = xprt;
2898 xprt->bc_xprt = args->bc_xprt;
2899 bc_sock = container_of(args->bc_xprt, struct svc_sock, sk_xprt);
2900 transport->sock = bc_sock->sk_sock;
2901 transport->inet = bc_sock->sk_sk;
2902
f300baba
AB		 2903 /*
2904 * Since we don't want connections for the backchannel, we set
2905 * the xprt status to connected
2906 */
2907 xprt_set_connected(xprt);
2908
2909
2910 if (try_module_get(THIS_MODULE))
2911 return xprt;
99de8ea9 2912 xprt_put(xprt);
0a68b0be
BF		 2913 ret = ERR_PTR(-EINVAL);
2914out_err:
e204e621 2915 xprt_free(xprt);
0a68b0be 2916 return ret;
f300baba
AB		 2917}
2918
176e21ee
CL		 2919static struct xprt_class xs_local_transport = {
2920 .list = LIST_HEAD_INIT(xs_local_transport.list),
2921 .name = "named UNIX socket",
2922 .owner = THIS_MODULE,
2923 .ident = XPRT_TRANSPORT_LOCAL,
2924 .setup = xs_setup_local,
2925};
2926
bc25571e
TT		 2927static struct xprt_class xs_udp_transport = {
2928 .list = LIST_HEAD_INIT(xs_udp_transport.list),
2929 .name = "udp",
2930 .owner = THIS_MODULE,
f300baba 2931 .ident = XPRT_TRANSPORT_UDP,
bc25571e
TT		 2932 .setup = xs_setup_udp,
2933};
2934
2935static struct xprt_class xs_tcp_transport = {
2936 .list = LIST_HEAD_INIT(xs_tcp_transport.list),
2937 .name = "tcp",
2938 .owner = THIS_MODULE,
f300baba 2939 .ident = XPRT_TRANSPORT_TCP,
bc25571e
TT		 2940 .setup = xs_setup_tcp,
2941};
2942
f300baba
AB		 2943static struct xprt_class xs_bc_tcp_transport = {
2944 .list = LIST_HEAD_INIT(xs_bc_tcp_transport.list),
2945 .name = "tcp NFSv4.1 backchannel",
2946 .owner = THIS_MODULE,
2947 .ident = XPRT_TRANSPORT_BC_TCP,
2948 .setup = xs_setup_bc_tcp,
2949};
2950
282b32e1 2951/**
bc25571e 2952 * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
282b32e1
CL		 2953 *
2954 */
2955int init_socket_xprt(void)
2956{
fbf76683 2957#ifdef RPC_DEBUG
2b1bec5f 2958 if (!sunrpc_table_header)
0b4d4147 2959 sunrpc_table_header = register_sysctl_table(sunrpc_table);
fbf76683
CL		 2960#endif
2961
176e21ee 2962 xprt_register_transport(&xs_local_transport);
bc25571e
TT		 2963 xprt_register_transport(&xs_udp_transport);
2964 xprt_register_transport(&xs_tcp_transport);
f300baba 2965 xprt_register_transport(&xs_bc_tcp_transport);
bc25571e 2966
282b32e1
CL		 2967 return 0;
2968}
2969
2970/**
bc25571e 2971 * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
282b32e1
CL		 2972 *
2973 */
2974void cleanup_socket_xprt(void)
2975{
fbf76683
CL		 2976#ifdef RPC_DEBUG
2977 if (sunrpc_table_header) {
2978 unregister_sysctl_table(sunrpc_table_header);
2979 sunrpc_table_header = NULL;
2980 }
2981#endif
bc25571e 2982
176e21ee 2983 xprt_unregister_transport(&xs_local_transport);
bc25571e
TT		 2984 xprt_unregister_transport(&xs_udp_transport);
2985 xprt_unregister_transport(&xs_tcp_transport);
f300baba 2986 xprt_unregister_transport(&xs_bc_tcp_transport);
282b32e1 2987}
cbf11071 2988
9bbb9e5a
RR		 2989static int param_set_uint_minmax(const char *val,
2990 const struct kernel_param *kp,
cbf11071
TM		 2991 unsigned int min, unsigned int max)
2992{
2993 unsigned long num;
2994 int ret;
2995
2996 if (!val)
2997 return -EINVAL;
2998 ret = strict_strtoul(val, 0, &num);
2999 if (ret == -EINVAL || num < min || num > max)
3000 return -EINVAL;
3001 *((unsigned int *)kp->arg) = num;
3002 return 0;
3003}
3004
9bbb9e5a 3005static int param_set_portnr(const char *val, const struct kernel_param *kp)
cbf11071
TM		 3006{
3007 return param_set_uint_minmax(val, kp,
3008 RPC_MIN_RESVPORT,
3009 RPC_MAX_RESVPORT);
3010}
3011
9bbb9e5a
RR		 3012static struct kernel_param_ops param_ops_portnr = {
3013 .set = param_set_portnr,
3014 .get = param_get_uint,
3015};
3016
cbf11071
TM		 3017#define param_check_portnr(name, p) \
3018 __param_check(name, p, unsigned int);
3019
3020module_param_named(min_resvport, xprt_min_resvport, portnr, 0644);
3021module_param_named(max_resvport, xprt_max_resvport, portnr, 0644);
3022
9bbb9e5a
RR		 3023static int param_set_slot_table_size(const char *val,
3024 const struct kernel_param *kp)
cbf11071
TM		 3025{
3026 return param_set_uint_minmax(val, kp,
3027 RPC_MIN_SLOT_TABLE,
3028 RPC_MAX_SLOT_TABLE);
3029}
3030
9bbb9e5a
RR		 3031static struct kernel_param_ops param_ops_slot_table_size = {
3032 .set = param_set_slot_table_size,
3033 .get = param_get_uint,
3034};
3035
cbf11071
TM		 3036#define param_check_slot_table_size(name, p) \
3037 __param_check(name, p, unsigned int);
3038
d9ba131d
TM		 3039static int param_set_max_slot_table_size(const char *val,
3040 const struct kernel_param *kp)
3041{
3042 return param_set_uint_minmax(val, kp,
3043 RPC_MIN_SLOT_TABLE,
3044 RPC_MAX_SLOT_TABLE_LIMIT);
3045}
3046
3047static struct kernel_param_ops param_ops_max_slot_table_size = {
3048 .set = param_set_max_slot_table_size,
3049 .get = param_get_uint,
3050};
3051
3052#define param_check_max_slot_table_size(name, p) \
3053 __param_check(name, p, unsigned int);
3054
cbf11071
TM		 3055module_param_named(tcp_slot_table_entries, xprt_tcp_slot_table_entries,
3056 slot_table_size, 0644);
d9ba131d
TM		 3057module_param_named(tcp_max_slot_table_entries, xprt_max_tcp_slot_table_entries,
3058 max_slot_table_size, 0644);
cbf11071
TM		 3059module_param_named(udp_slot_table_entries, xprt_udp_slot_table_entries,
3060 slot_table_size, 0644);
3061
kernel source for telekom puls (alcatel/mediatek)