projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
SUNRPC: remove BUG_ON in xdr_shrink_bufhead
[linux-2.6-block.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>
02107148 36#include <linux/sunrpc/sched.h>
4cfc7e60 37#include <linux/sunrpc/svcsock.h>
49c36fcc 38#include <linux/sunrpc/xprtsock.h>
a246b010 39#include <linux/file.h>
9e00abc3 40#ifdef CONFIG_SUNRPC_BACKCHANNEL
44b98efd
RL		 41#include <linux/sunrpc/bc_xprt.h>
42#endif
a246b010
CL		 43
44#include <net/sock.h>
45#include <net/checksum.h>
46#include <net/udp.h>
47#include <net/tcp.h>
48
4cfc7e60 49#include "sunrpc.h"
176e21ee
CL		 50
51static void xs_close(struct rpc_xprt *xprt);
52
c556b754
CL		 53/*
54 * xprtsock tunables
55 */
09acfea5
TM		 56static unsigned int xprt_udp_slot_table_entries = RPC_DEF_SLOT_TABLE;
57static unsigned int xprt_tcp_slot_table_entries = RPC_MIN_SLOT_TABLE;
58static unsigned int xprt_max_tcp_slot_table_entries = RPC_MAX_SLOT_TABLE;
c556b754 59
09acfea5
TM		 60static unsigned int xprt_min_resvport = RPC_DEF_MIN_RESVPORT;
61static unsigned int xprt_max_resvport = RPC_DEF_MAX_RESVPORT;
c556b754 62
7d1e8255 63#define XS_TCP_LINGER_TO (15U * HZ)
25fe6142 64static unsigned int xs_tcp_fin_timeout __read_mostly = XS_TCP_LINGER_TO;
7d1e8255 65
fbf76683
CL		 66/*
67 * We can register our own files under /proc/sys/sunrpc by
68 * calling register_sysctl_table() again. The files in that
69 * directory become the union of all files registered there.
70 *
71 * We simply need to make sure that we don't collide with
72 * someone else's file names!
73 */
74
75#ifdef RPC_DEBUG
76
77static unsigned int min_slot_table_size = RPC_MIN_SLOT_TABLE;
78static unsigned int max_slot_table_size = RPC_MAX_SLOT_TABLE;
d9ba131d 79static unsigned int max_tcp_slot_table_limit = RPC_MAX_SLOT_TABLE_LIMIT;
fbf76683
CL		 80static unsigned int xprt_min_resvport_limit = RPC_MIN_RESVPORT;
81static unsigned int xprt_max_resvport_limit = RPC_MAX_RESVPORT;
82
83static struct ctl_table_header *sunrpc_table_header;
84
85/*
86 * FIXME: changing the UDP slot table size should also resize the UDP
87 * socket buffers for existing UDP transports
88 */
89static ctl_table xs_tunables_table[] = {
90 {
fbf76683
CL		 91 .procname = "udp_slot_table_entries",
92 .data = &xprt_udp_slot_table_entries,
93 .maxlen = sizeof(unsigned int),
94 .mode = 0644,
6d456111 95 .proc_handler = proc_dointvec_minmax,
fbf76683
CL		 96 .extra1 = &min_slot_table_size,
97 .extra2 = &max_slot_table_size
98 },
99 {
fbf76683
CL		 100 .procname = "tcp_slot_table_entries",
101 .data = &xprt_tcp_slot_table_entries,
102 .maxlen = sizeof(unsigned int),
103 .mode = 0644,
6d456111 104 .proc_handler = proc_dointvec_minmax,
fbf76683
CL		 105 .extra1 = &min_slot_table_size,
106 .extra2 = &max_slot_table_size
107 },
d9ba131d
TM		 108 {
109 .procname = "tcp_max_slot_table_entries",
110 .data = &xprt_max_tcp_slot_table_entries,
111 .maxlen = sizeof(unsigned int),
112 .mode = 0644,
113 .proc_handler = proc_dointvec_minmax,
114 .extra1 = &min_slot_table_size,
115 .extra2 = &max_tcp_slot_table_limit
116 },
fbf76683 117 {
fbf76683
CL		 118 .procname = "min_resvport",
119 .data = &xprt_min_resvport,
120 .maxlen = sizeof(unsigned int),
121 .mode = 0644,
6d456111 122 .proc_handler = proc_dointvec_minmax,
fbf76683
CL		 123 .extra1 = &xprt_min_resvport_limit,
124 .extra2 = &xprt_max_resvport_limit
125 },
126 {
fbf76683
CL		 127 .procname = "max_resvport",
128 .data = &xprt_max_resvport,
129 .maxlen = sizeof(unsigned int),
130 .mode = 0644,
6d456111 131 .proc_handler = proc_dointvec_minmax,
fbf76683
CL		 132 .extra1 = &xprt_min_resvport_limit,
133 .extra2 = &xprt_max_resvport_limit
134 },
25fe6142
TM		 135 {
136 .procname = "tcp_fin_timeout",
137 .data = &xs_tcp_fin_timeout,
138 .maxlen = sizeof(xs_tcp_fin_timeout),
139 .mode = 0644,
6d456111 140 .proc_handler = proc_dointvec_jiffies,
fbf76683 141 },
f8572d8f 142 { },
fbf76683
CL		 143};
144
145static ctl_table sunrpc_table[] = {
146 {
fbf76683
CL		 147 .procname = "sunrpc",
148 .mode = 0555,
149 .child = xs_tunables_table
150 },
f8572d8f 151 { },
fbf76683
CL		 152};
153
154#endif
155
03bf4b70
CL		 156/*
157 * Wait duration for a reply from the RPC portmapper.
158 */
159#define XS_BIND_TO (60U * HZ)
160
161/*
162 * Delay if a UDP socket connect error occurs. This is most likely some
163 * kind of resource problem on the local host.
164 */
165#define XS_UDP_REEST_TO (2U * HZ)
166
167/*
168 * The reestablish timeout allows clients to delay for a bit before attempting
169 * to reconnect to a server that just dropped our connection.
170 *
171 * We implement an exponential backoff when trying to reestablish a TCP
172 * transport connection with the server. Some servers like to drop a TCP
173 * connection when they are overworked, so we start with a short timeout and
174 * increase over time if the server is down or not responding.
175 */
176#define XS_TCP_INIT_REEST_TO (3U * HZ)
177#define XS_TCP_MAX_REEST_TO (5U * 60 * HZ)
178
179/*
180 * TCP idle timeout; client drops the transport socket if it is idle
181 * for this long. Note that we also timeout UDP sockets to prevent
182 * holding port numbers when there is no RPC traffic.
183 */
184#define XS_IDLE_DISC_TO (5U * 60 * HZ)
185
a246b010
CL		 186#ifdef RPC_DEBUG
187# undef RPC_DEBUG_DATA
9903cd1c 188# define RPCDBG_FACILITY RPCDBG_TRANS
a246b010
CL		 189#endif
190
a246b010 191#ifdef RPC_DEBUG_DATA
9903cd1c 192static void xs_pktdump(char *msg, u32 *packet, unsigned int count)
a246b010 193{
9903cd1c
CL		 194 u8 *buf = (u8 *) packet;
195 int j;
a246b010 196
46121cf7 197 dprintk("RPC: %s\n", msg);
a246b010
CL		 198 for (j = 0; j < count && j < 128; j += 4) {
199 if (!(j & 31)) {
200 if (j)
201 dprintk("\n");
202 dprintk("0x%04x ", j);
203 }
204 dprintk("%02x%02x%02x%02x ",
205 buf[j], buf[j+1], buf[j+2], buf[j+3]);
206 }
207 dprintk("\n");
208}
209#else
9903cd1c 210static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count)
a246b010
CL		 211{
212 /* NOP */
213}
214#endif
215
ffc2e518
CL		 216struct sock_xprt {
217 struct rpc_xprt xprt;
ee0ac0c2
CL		 218
219 /*
220 * Network layer
221 */
222 struct socket * sock;
223 struct sock * inet;
51971139
CL		 224
225 /*
226 * State of TCP reply receive
227 */
228 __be32 tcp_fraghdr,
b76ce561
TM		 229 tcp_xid,
230 tcp_calldir;
51971139
CL		 231
232 u32 tcp_offset,
233 tcp_reclen;
234
235 unsigned long tcp_copied,
236 tcp_flags;
c8475461
CL		 237
238 /*
239 * Connection of transports
240 */
34161db6 241 struct delayed_work connect_worker;
fbfffbd5
CL		 242 struct sockaddr_storage srcaddr;
243 unsigned short srcport;
7c6e066e
CL		 244
245 /*
246 * UDP socket buffer size parameters
247 */
248 size_t rcvsize,
249 sndsize;
314dfd79
CL		 250
251 /*
252 * Saved socket callback addresses
253 */
254 void (*old_data_ready)(struct sock *, int);
255 void (*old_state_change)(struct sock *);
256 void (*old_write_space)(struct sock *);
ffc2e518
CL		 257};
258
e136d092
CL		 259/*
260 * TCP receive state flags
261 */
262#define TCP_RCV_LAST_FRAG (1UL << 0)
263#define TCP_RCV_COPY_FRAGHDR (1UL << 1)
264#define TCP_RCV_COPY_XID (1UL << 2)
265#define TCP_RCV_COPY_DATA (1UL << 3)
f4a2e418
RL		 266#define TCP_RCV_READ_CALLDIR (1UL << 4)
267#define TCP_RCV_COPY_CALLDIR (1UL << 5)
18dca02a
RL		 268
269/*
270 * TCP RPC flags
271 */
f4a2e418 272#define TCP_RPC_REPLY (1UL << 6)
e136d092 273
95392c59
CL		 274static inline struct sockaddr *xs_addr(struct rpc_xprt *xprt)
275{
276 return (struct sockaddr *) &xprt->addr;
277}
278
176e21ee
CL		 279static inline struct sockaddr_un *xs_addr_un(struct rpc_xprt *xprt)
280{
281 return (struct sockaddr_un *) &xprt->addr;
282}
283
95392c59 284static inline struct sockaddr_in *xs_addr_in(struct rpc_xprt *xprt)
edb267a6 285{
95392c59
CL		 286 return (struct sockaddr_in *) &xprt->addr;
287}
288
289static inline struct sockaddr_in6 *xs_addr_in6(struct rpc_xprt *xprt)
290{
291 return (struct sockaddr_in6 *) &xprt->addr;
292}
293
c877b849 294static void xs_format_common_peer_addresses(struct rpc_xprt *xprt)
edb267a6 295{
c877b849 296 struct sockaddr *sap = xs_addr(xprt);
9dc3b095
CL		 297 struct sockaddr_in6 *sin6;
298 struct sockaddr_in *sin;
176e21ee 299 struct sockaddr_un *sun;
c877b849 300 char buf[128];
edb267a6 301
9dc3b095 302 switch (sap->sa_family) {
176e21ee
CL		 303 case AF_LOCAL:
304 sun = xs_addr_un(xprt);
305 strlcpy(buf, sun->sun_path, sizeof(buf));
306 xprt->address_strings[RPC_DISPLAY_ADDR] =
307 kstrdup(buf, GFP_KERNEL);
308 break;
9dc3b095 309 case AF_INET:
176e21ee
CL		 310 (void)rpc_ntop(sap, buf, sizeof(buf));
311 xprt->address_strings[RPC_DISPLAY_ADDR] =
312 kstrdup(buf, GFP_KERNEL);
9dc3b095 313 sin = xs_addr_in(xprt);
fc0b5791 314 snprintf(buf, sizeof(buf), "%08x", ntohl(sin->sin_addr.s_addr));
9dc3b095
CL		 315 break;
316 case AF_INET6:
176e21ee
CL		 317 (void)rpc_ntop(sap, buf, sizeof(buf));
318 xprt->address_strings[RPC_DISPLAY_ADDR] =
319 kstrdup(buf, GFP_KERNEL);
9dc3b095 320 sin6 = xs_addr_in6(xprt);
fc0b5791 321 snprintf(buf, sizeof(buf), "%pi6", &sin6->sin6_addr);
9dc3b095
CL		 322 break;
323 default:
324 BUG();
756805e7 325 }
176e21ee 326
9dc3b095 327 xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = kstrdup(buf, GFP_KERNEL);
edb267a6
CL		 328}
329
9dc3b095 330static void xs_format_common_peer_ports(struct rpc_xprt *xprt)
4b6473fb 331{
9dc3b095
CL		 332 struct sockaddr *sap = xs_addr(xprt);
333 char buf[128];
4b6473fb 334
81160e66 335 snprintf(buf, sizeof(buf), "%u", rpc_get_port(sap));
c877b849 336 xprt->address_strings[RPC_DISPLAY_PORT] = kstrdup(buf, GFP_KERNEL);
4b6473fb 337
81160e66 338 snprintf(buf, sizeof(buf), "%4hx", rpc_get_port(sap));
c877b849
CL		 339 xprt->address_strings[RPC_DISPLAY_HEX_PORT] = kstrdup(buf, GFP_KERNEL);
340}
4b6473fb 341
9dc3b095
CL		 342static void xs_format_peer_addresses(struct rpc_xprt *xprt,
343 const char *protocol,
344 const char *netid)
c877b849 345{
b454ae90 346 xprt->address_strings[RPC_DISPLAY_PROTO] = protocol;
b454ae90 347 xprt->address_strings[RPC_DISPLAY_NETID] = netid;
c877b849 348 xs_format_common_peer_addresses(xprt);
9dc3b095 349 xs_format_common_peer_ports(xprt);
edb267a6 350}
4b6473fb 351
9dc3b095 352static void xs_update_peer_port(struct rpc_xprt *xprt)
4b6473fb 353{
9dc3b095
CL		 354 kfree(xprt->address_strings[RPC_DISPLAY_HEX_PORT]);
355 kfree(xprt->address_strings[RPC_DISPLAY_PORT]);
4417c8c4 356
9dc3b095 357 xs_format_common_peer_ports(xprt);
edb267a6
CL		 358}
359
360static void xs_free_peer_addresses(struct rpc_xprt *xprt)
361{
33e01dc7
CL		 362 unsigned int i;
363
364 for (i = 0; i < RPC_DISPLAY_MAX; i++)
365 switch (i) {
366 case RPC_DISPLAY_PROTO:
367 case RPC_DISPLAY_NETID:
368 continue;
369 default:
370 kfree(xprt->address_strings[i]);
371 }
edb267a6
CL		 372}
373
b4b5cc85
CL		 374#define XS_SENDMSG_FLAGS (MSG_DONTWAIT | MSG_NOSIGNAL)
375
24c5684b 376static int xs_send_kvec(struct socket *sock, struct sockaddr *addr, int addrlen, struct kvec *vec, unsigned int base, int more)
b4b5cc85 377{
b4b5cc85
CL		 378 struct msghdr msg = {
379 .msg_name = addr,
380 .msg_namelen = addrlen,
24c5684b
TM		 381 .msg_flags = XS_SENDMSG_FLAGS | (more ? MSG_MORE : 0),
382 };
383 struct kvec iov = {
384 .iov_base = vec->iov_base + base,
385 .iov_len = vec->iov_len - base,
b4b5cc85
CL		 386 };
387
24c5684b 388 if (iov.iov_len != 0)
b4b5cc85
CL		 389 return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
390 return kernel_sendmsg(sock, &msg, NULL, 0, 0);
391}
392
24c5684b 393static int xs_send_pagedata(struct socket *sock, struct xdr_buf *xdr, unsigned int base, int more)
b4b5cc85 394{
24c5684b
TM		 395 struct page **ppage;
396 unsigned int remainder;
397 int err, sent = 0;
398
399 remainder = xdr->page_len - base;
400 base += xdr->page_base;
401 ppage = xdr->pages + (base >> PAGE_SHIFT);
402 base &= ~PAGE_MASK;
403 for(;;) {
404 unsigned int len = min_t(unsigned int, PAGE_SIZE - base, remainder);
405 int flags = XS_SENDMSG_FLAGS;
b4b5cc85 406
24c5684b
TM		 407 remainder -= len;
408 if (remainder != 0 || more)
409 flags |= MSG_MORE;
410 err = sock->ops->sendpage(sock, *ppage, base, len, flags);
411 if (remainder == 0 || err != len)
412 break;
413 sent += err;
414 ppage++;
415 base = 0;
416 }
417 if (sent == 0)
418 return err;
419 if (err > 0)
420 sent += err;
421 return sent;
b4b5cc85
CL		 422}
423
9903cd1c
CL		 424/**
425 * xs_sendpages - write pages directly to a socket
426 * @sock: socket to send on
427 * @addr: UDP only -- address of destination
428 * @addrlen: UDP only -- length of destination address
429 * @xdr: buffer containing this request
430 * @base: starting position in the buffer
431 *
a246b010 432 */
24c5684b 433static int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base)
a246b010 434{
24c5684b
TM		 435 unsigned int remainder = xdr->len - base;
436 int err, sent = 0;
a246b010 437
262965f5 438 if (unlikely(!sock))
fba91afb 439 return -ENOTSOCK;
262965f5
CL		 440
441 clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags);
24c5684b
TM		 442 if (base != 0) {
443 addr = NULL;
444 addrlen = 0;
445 }
262965f5 446
24c5684b
TM		 447 if (base < xdr->head[0].iov_len || addr != NULL) {
448 unsigned int len = xdr->head[0].iov_len - base;
449 remainder -= len;
450 err = xs_send_kvec(sock, addr, addrlen, &xdr->head[0], base, remainder != 0);
451 if (remainder == 0 || err != len)
a246b010 452 goto out;
24c5684b 453 sent += err;
a246b010
CL		 454 base = 0;
455 } else
24c5684b 456 base -= xdr->head[0].iov_len;
a246b010 457
24c5684b
TM		 458 if (base < xdr->page_len) {
459 unsigned int len = xdr->page_len - base;
460 remainder -= len;
461 err = xs_send_pagedata(sock, xdr, base, remainder != 0);
462 if (remainder == 0 || err != len)
a246b010 463 goto out;
24c5684b 464 sent += err;
a246b010 465 base = 0;
24c5684b
TM		 466 } else
467 base -= xdr->page_len;
468
469 if (base >= xdr->tail[0].iov_len)
470 return sent;
471 err = xs_send_kvec(sock, NULL, 0, &xdr->tail[0], base, 0);
a246b010 472out:
24c5684b
TM		 473 if (sent == 0)
474 return err;
475 if (err > 0)
476 sent += err;
477 return sent;
a246b010
CL		 478}
479
b6ddf64f
TM		 480static void xs_nospace_callback(struct rpc_task *task)
481{
482 struct sock_xprt *transport = container_of(task->tk_rqstp->rq_xprt, struct sock_xprt, xprt);
483
484 transport->inet->sk_write_pending--;
485 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
486}
487
9903cd1c 488/**
262965f5
CL		 489 * xs_nospace - place task on wait queue if transmit was incomplete
490 * @task: task to put to sleep
9903cd1c 491 *
a246b010 492 */
5e3771ce 493static int xs_nospace(struct rpc_task *task)
a246b010 494{
262965f5
CL		 495 struct rpc_rqst *req = task->tk_rqstp;
496 struct rpc_xprt *xprt = req->rq_xprt;
ee0ac0c2 497 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
24ca9a84 498 int ret = -EAGAIN;
a246b010 499
46121cf7 500 dprintk("RPC: %5u xmit incomplete (%u left of %u)\n",
262965f5
CL		 501 task->tk_pid, req->rq_slen - req->rq_bytes_sent,
502 req->rq_slen);
503
b6ddf64f
TM		 504 /* Protect against races with write_space */
505 spin_lock_bh(&xprt->transport_lock);
506
507 /* Don't race with disconnect */
508 if (xprt_connected(xprt)) {
509 if (test_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags)) {
510 /*
511 * Notify TCP that we're limited by the application
512 * window size
513 */
514 set_bit(SOCK_NOSPACE, &transport->sock->flags);
515 transport->inet->sk_write_pending++;
516 /* ...and wait for more buffer space */
517 xprt_wait_for_buffer_space(task, xs_nospace_callback);
518 }
519 } else {
520 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
5e3771ce 521 ret = -ENOTCONN;
b6ddf64f 522 }
262965f5 523
b6ddf64f 524 spin_unlock_bh(&xprt->transport_lock);
5e3771ce 525 return ret;
262965f5
CL		 526}
527
61677eee
CL		 528/*
529 * Construct a stream transport record marker in @buf.
530 */
531static inline void xs_encode_stream_record_marker(struct xdr_buf *buf)
532{
533 u32 reclen = buf->len - sizeof(rpc_fraghdr);
534 rpc_fraghdr *base = buf->head[0].iov_base;
535 *base = cpu_to_be32(RPC_LAST_STREAM_FRAGMENT | reclen);
536}
537
176e21ee
CL		 538/**
539 * xs_local_send_request - write an RPC request to an AF_LOCAL socket
540 * @task: RPC task that manages the state of an RPC request
541 *
542 * Return values:
543 * 0: The request has been sent
544 * EAGAIN: The socket was blocked, please call again later to
545 * complete the request
546 * ENOTCONN: Caller needs to invoke connect logic then call again
547 * other: Some other error occured, the request was not sent
548 */
549static int xs_local_send_request(struct rpc_task *task)
550{
551 struct rpc_rqst *req = task->tk_rqstp;
552 struct rpc_xprt *xprt = req->rq_xprt;
553 struct sock_xprt *transport =
554 container_of(xprt, struct sock_xprt, xprt);
555 struct xdr_buf *xdr = &req->rq_snd_buf;
556 int status;
557
558 xs_encode_stream_record_marker(&req->rq_snd_buf);
559
560 xs_pktdump("packet data:",
561 req->rq_svec->iov_base, req->rq_svec->iov_len);
562
563 status = xs_sendpages(transport->sock, NULL, 0,
564 xdr, req->rq_bytes_sent);
565 dprintk("RPC: %s(%u) = %d\n",
566 __func__, xdr->len - req->rq_bytes_sent, status);
567 if (likely(status >= 0)) {
568 req->rq_bytes_sent += status;
569 req->rq_xmit_bytes_sent += status;
570 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
571 req->rq_bytes_sent = 0;
572 return 0;
573 }
574 status = -EAGAIN;
575 }
576
577 switch (status) {
578 case -EAGAIN:
579 status = xs_nospace(task);
580 break;
581 default:
582 dprintk("RPC: sendmsg returned unrecognized error %d\n",
583 -status);
584 case -EPIPE:
585 xs_close(xprt);
586 status = -ENOTCONN;
587 }
588
589 return status;
590}
591
262965f5
CL		 592/**
593 * xs_udp_send_request - write an RPC request to a UDP socket
594 * @task: address of RPC task that manages the state of an RPC request
595 *
596 * Return values:
597 * 0: The request has been sent
598 * EAGAIN: The socket was blocked, please call again later to
599 * complete the request
600 * ENOTCONN: Caller needs to invoke connect logic then call again
25985edc 601 * other: Some other error occurred, the request was not sent
262965f5
CL		 602 */
603static int xs_udp_send_request(struct rpc_task *task)
604{
605 struct rpc_rqst *req = task->tk_rqstp;
606 struct rpc_xprt *xprt = req->rq_xprt;
ee0ac0c2 607 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
262965f5
CL		 608 struct xdr_buf *xdr = &req->rq_snd_buf;
609 int status;
a246b010 610
9903cd1c 611 xs_pktdump("packet data:",
a246b010
CL		 612 req->rq_svec->iov_base,
613 req->rq_svec->iov_len);
614
01d37c42
TM		 615 if (!xprt_bound(xprt))
616 return -ENOTCONN;
ee0ac0c2 617 status = xs_sendpages(transport->sock,
95392c59 618 xs_addr(xprt),
ee0ac0c2
CL		 619 xprt->addrlen, xdr,
620 req->rq_bytes_sent);
a246b010 621
46121cf7 622 dprintk("RPC: xs_udp_send_request(%u) = %d\n",
262965f5 623 xdr->len - req->rq_bytes_sent, status);
a246b010 624
2199700f 625 if (status >= 0) {
d60dbb20 626 req->rq_xmit_bytes_sent += status;
2199700f
TM		 627 if (status >= req->rq_slen)
628 return 0;
629 /* Still some bytes left; set up for a retry later. */
262965f5 630 status = -EAGAIN;
2199700f 631 }
a246b010 632
262965f5 633 switch (status) {
fba91afb
TM		 634 case -ENOTSOCK:
635 status = -ENOTCONN;
636 /* Should we call xs_close() here? */
637 break;
b6ddf64f 638 case -EAGAIN:
5e3771ce 639 status = xs_nospace(task);
b6ddf64f 640 break;
c8485e4d
TM		 641 default:
642 dprintk("RPC: sendmsg returned unrecognized error %d\n",
643 -status);
262965f5
CL		 644 case -ENETUNREACH:
645 case -EPIPE:
a246b010
CL		 646 case -ECONNREFUSED:
647 /* When the server has died, an ICMP port unreachable message
9903cd1c 648 * prompts ECONNREFUSED. */
b6ddf64f 649 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
a246b010 650 }
5fe46e9d 651
262965f5 652 return status;
a246b010
CL		 653}
654
e06799f9
TM		 655/**
656 * xs_tcp_shutdown - gracefully shut down a TCP socket
657 * @xprt: transport
658 *
659 * Initiates a graceful shutdown of the TCP socket by calling the
660 * equivalent of shutdown(SHUT_WR);
661 */
662static void xs_tcp_shutdown(struct rpc_xprt *xprt)
663{
664 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
665 struct socket *sock = transport->sock;
666
667 if (sock != NULL)
668 kernel_sock_shutdown(sock, SHUT_WR);
669}
670
9903cd1c 671/**
262965f5 672 * xs_tcp_send_request - write an RPC request to a TCP socket
9903cd1c
CL		 673 * @task: address of RPC task that manages the state of an RPC request
674 *
675 * Return values:
262965f5
CL		 676 * 0: The request has been sent
677 * EAGAIN: The socket was blocked, please call again later to
678 * complete the request
679 * ENOTCONN: Caller needs to invoke connect logic then call again
25985edc 680 * other: Some other error occurred, the request was not sent
9903cd1c
CL		 681 *
682 * XXX: In the case of soft timeouts, should we eventually give up
262965f5 683 * if sendmsg is not able to make progress?
9903cd1c 684 */
262965f5 685static int xs_tcp_send_request(struct rpc_task *task)
a246b010
CL		 686{
687 struct rpc_rqst *req = task->tk_rqstp;
688 struct rpc_xprt *xprt = req->rq_xprt;
ee0ac0c2 689 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
262965f5 690 struct xdr_buf *xdr = &req->rq_snd_buf;
b595bb15 691 int status;
a246b010 692
61677eee 693 xs_encode_stream_record_marker(&req->rq_snd_buf);
a246b010 694
262965f5
CL		 695 xs_pktdump("packet data:",
696 req->rq_svec->iov_base,
697 req->rq_svec->iov_len);
a246b010
CL		 698
699 /* Continue transmitting the packet/record. We must be careful
700 * to cope with writespace callbacks arriving _after_ we have
262965f5 701 * called sendmsg(). */
a246b010 702 while (1) {
ee0ac0c2
CL		 703 status = xs_sendpages(transport->sock,
704 NULL, 0, xdr, req->rq_bytes_sent);
a246b010 705
46121cf7 706 dprintk("RPC: xs_tcp_send_request(%u) = %d\n",
262965f5 707 xdr->len - req->rq_bytes_sent, status);
a246b010 708
262965f5 709 if (unlikely(status < 0))
a246b010 710 break;
a246b010 711
262965f5
CL		 712 /* If we've sent the entire packet, immediately
713 * reset the count of bytes sent. */
714 req->rq_bytes_sent += status;
d60dbb20 715 req->rq_xmit_bytes_sent += status;
262965f5
CL		 716 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
717 req->rq_bytes_sent = 0;
718 return 0;
719 }
a246b010 720
06b4b681
TM		 721 if (status != 0)
722 continue;
a246b010 723 status = -EAGAIN;
06b4b681 724 break;
a246b010
CL		 725 }
726
262965f5 727 switch (status) {
fba91afb
TM		 728 case -ENOTSOCK:
729 status = -ENOTCONN;
730 /* Should we call xs_close() here? */
731 break;
262965f5 732 case -EAGAIN:
5e3771ce 733 status = xs_nospace(task);
262965f5 734 break;
c8485e4d
TM		 735 default:
736 dprintk("RPC: sendmsg returned unrecognized error %d\n",
737 -status);
262965f5 738 case -ECONNRESET:
2a9e1cfa
TM		 739 xs_tcp_shutdown(xprt);
740 case -ECONNREFUSED:
262965f5 741 case -ENOTCONN:
b9d2bb2e 742 case -EPIPE:
b6ddf64f 743 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
a246b010 744 }
5fe46e9d 745
a246b010
CL		 746 return status;
747}
748
e0ab53de
TM		 749/**
750 * xs_tcp_release_xprt - clean up after a tcp transmission
751 * @xprt: transport
752 * @task: rpc task
753 *
754 * This cleans up if an error causes us to abort the transmission of a request.
755 * In this case, the socket may need to be reset in order to avoid confusing
756 * the server.
757 */
758static void xs_tcp_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
759{
760 struct rpc_rqst *req;
761
762 if (task != xprt->snd_task)
763 return;
764 if (task == NULL)
765 goto out_release;
766 req = task->tk_rqstp;
43cedbf0
TM		 767 if (req == NULL)
768 goto out_release;
e0ab53de
TM		 769 if (req->rq_bytes_sent == 0)
770 goto out_release;
771 if (req->rq_bytes_sent == req->rq_snd_buf.len)
772 goto out_release;
773 set_bit(XPRT_CLOSE_WAIT, &task->tk_xprt->state);
774out_release:
775 xprt_release_xprt(xprt, task);
776}
777
2a9e1cfa
TM		 778static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk)
779{
780 transport->old_data_ready = sk->sk_data_ready;
781 transport->old_state_change = sk->sk_state_change;
782 transport->old_write_space = sk->sk_write_space;
2a9e1cfa
TM		 783}
784
785static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk)
786{
787 sk->sk_data_ready = transport->old_data_ready;
788 sk->sk_state_change = transport->old_state_change;
789 sk->sk_write_space = transport->old_write_space;
2a9e1cfa
TM		 790}
791
fe315e76 792static void xs_reset_transport(struct sock_xprt *transport)
a246b010 793{
ee0ac0c2
CL		 794 struct socket *sock = transport->sock;
795 struct sock *sk = transport->inet;
a246b010 796
fe315e76
CL		 797 if (sk == NULL)
798 return;
9903cd1c 799
246408dc
TM		 800 transport->srcport = 0;
801
a246b010 802 write_lock_bh(&sk->sk_callback_lock);
ee0ac0c2
CL		 803 transport->inet = NULL;
804 transport->sock = NULL;
a246b010 805
9903cd1c 806 sk->sk_user_data = NULL;
2a9e1cfa
TM		 807
808 xs_restore_old_callbacks(transport, sk);
a246b010
CL		 809 write_unlock_bh(&sk->sk_callback_lock);
810
9903cd1c 811 sk->sk_no_check = 0;
a246b010
CL		 812
813 sock_release(sock);
fe315e76
CL		 814}
815
816/**
817 * xs_close - close a socket
818 * @xprt: transport
819 *
820 * This is used when all requests are complete; ie, no DRC state remains
821 * on the server we want to save.
f75e6745
TM		 822 *
823 * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with
824 * xs_reset_transport() zeroing the socket from underneath a writer.
fe315e76
CL		 825 */
826static void xs_close(struct rpc_xprt *xprt)
827{
828 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
829
830 dprintk("RPC: xs_close xprt %p\n", xprt);
831
832 xs_reset_transport(transport);
61d0a8e6 833 xprt->reestablish_timeout = 0;
fe315e76 834
632e3bdc 835 smp_mb__before_clear_bit();
7d1e8255 836 clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
632e3bdc 837 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
3b948ae5 838 clear_bit(XPRT_CLOSING, &xprt->state);
632e3bdc 839 smp_mb__after_clear_bit();
62da3b24 840 xprt_disconnect_done(xprt);
a246b010
CL		 841}
842
f75e6745
TM		 843static void xs_tcp_close(struct rpc_xprt *xprt)
844{
845 if (test_and_clear_bit(XPRT_CONNECTION_CLOSE, &xprt->state))
846 xs_close(xprt);
847 else
848 xs_tcp_shutdown(xprt);
849}
850
9903cd1c
CL		 851/**
852 * xs_destroy - prepare to shutdown a transport
853 * @xprt: doomed transport
854 *
855 */
856static void xs_destroy(struct rpc_xprt *xprt)
a246b010 857{
c8475461
CL		 858 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
859
46121cf7 860 dprintk("RPC: xs_destroy xprt %p\n", xprt);
9903cd1c 861
afe2c511 862 cancel_delayed_work_sync(&transport->connect_worker);
a246b010 863
9903cd1c 864 xs_close(xprt);
edb267a6 865 xs_free_peer_addresses(xprt);
e204e621 866 xprt_free(xprt);
bc25571e 867 module_put(THIS_MODULE);
a246b010
CL		 868}
869
9903cd1c
CL		 870static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
871{
872 return (struct rpc_xprt *) sk->sk_user_data;
873}
874
176e21ee
CL		 875static int xs_local_copy_to_xdr(struct xdr_buf *xdr, struct sk_buff *skb)
876{
877 struct xdr_skb_reader desc = {
878 .skb = skb,
879 .offset = sizeof(rpc_fraghdr),
880 .count = skb->len - sizeof(rpc_fraghdr),
881 };
882
883 if (xdr_partial_copy_from_skb(xdr, 0, &desc, xdr_skb_read_bits) < 0)
884 return -1;
885 if (desc.count)
886 return -1;
887 return 0;
888}
889
890/**
891 * xs_local_data_ready - "data ready" callback for AF_LOCAL sockets
892 * @sk: socket with data to read
893 * @len: how much data to read
894 *
895 * Currently this assumes we can read the whole reply in a single gulp.
896 */
897static void xs_local_data_ready(struct sock *sk, int len)
898{
899 struct rpc_task *task;
900 struct rpc_xprt *xprt;
901 struct rpc_rqst *rovr;
902 struct sk_buff *skb;
903 int err, repsize, copied;
904 u32 _xid;
905 __be32 *xp;
906
907 read_lock_bh(&sk->sk_callback_lock);
908 dprintk("RPC: %s...\n", __func__);
909 xprt = xprt_from_sock(sk);
910 if (xprt == NULL)
911 goto out;
912
913 skb = skb_recv_datagram(sk, 0, 1, &err);
914 if (skb == NULL)
915 goto out;
916
176e21ee
CL		 917 repsize = skb->len - sizeof(rpc_fraghdr);
918 if (repsize < 4) {
919 dprintk("RPC: impossible RPC reply size %d\n", repsize);
920 goto dropit;
921 }
922
923 /* Copy the XID from the skb... */
924 xp = skb_header_pointer(skb, sizeof(rpc_fraghdr), sizeof(_xid), &_xid);
925 if (xp == NULL)
926 goto dropit;
927
928 /* Look up and lock the request corresponding to the given XID */
929 spin_lock(&xprt->transport_lock);
930 rovr = xprt_lookup_rqst(xprt, *xp);
931 if (!rovr)
932 goto out_unlock;
933 task = rovr->rq_task;
934
935 copied = rovr->rq_private_buf.buflen;
936 if (copied > repsize)
937 copied = repsize;
938
939 if (xs_local_copy_to_xdr(&rovr->rq_private_buf, skb)) {
940 dprintk("RPC: sk_buff copy failed\n");
941 goto out_unlock;
942 }
943
944 xprt_complete_rqst(task, copied);
945
946 out_unlock:
947 spin_unlock(&xprt->transport_lock);
948 dropit:
949 skb_free_datagram(sk, skb);
950 out:
951 read_unlock_bh(&sk->sk_callback_lock);
952}
953
9903cd1c
CL		 954/**
955 * xs_udp_data_ready - "data ready" callback for UDP sockets
956 * @sk: socket with data to read
957 * @len: how much data to read
958 *
a246b010 959 */
9903cd1c 960static void xs_udp_data_ready(struct sock *sk, int len)
a246b010 961{
9903cd1c
CL		 962 struct rpc_task *task;
963 struct rpc_xprt *xprt;
a246b010 964 struct rpc_rqst *rovr;
9903cd1c 965 struct sk_buff *skb;
a246b010 966 int err, repsize, copied;
d8ed029d
AD		 967 u32 _xid;
968 __be32 *xp;
a246b010 969
f064af1e 970 read_lock_bh(&sk->sk_callback_lock);
46121cf7 971 dprintk("RPC: xs_udp_data_ready...\n");
9903cd1c 972 if (!(xprt = xprt_from_sock(sk)))
a246b010 973 goto out;
a246b010
CL		 974
975 if ((skb = skb_recv_datagram(sk, 0, 1, &err)) == NULL)
976 goto out;
977
a246b010
CL		 978 repsize = skb->len - sizeof(struct udphdr);
979 if (repsize < 4) {
46121cf7 980 dprintk("RPC: impossible RPC reply size %d!\n", repsize);
a246b010
CL		 981 goto dropit;
982 }
983
984 /* Copy the XID from the skb... */
985 xp = skb_header_pointer(skb, sizeof(struct udphdr),
986 sizeof(_xid), &_xid);
987 if (xp == NULL)
988 goto dropit;
989
990 /* Look up and lock the request corresponding to the given XID */
4a0f8c04 991 spin_lock(&xprt->transport_lock);
a246b010
CL		 992 rovr = xprt_lookup_rqst(xprt, *xp);
993 if (!rovr)
994 goto out_unlock;
995 task = rovr->rq_task;
996
a246b010
CL		 997 if ((copied = rovr->rq_private_buf.buflen) > repsize)
998 copied = repsize;
999
1000 /* Suck it into the iovec, verify checksum if not done by hw. */
1781f7f5
HX		 1001 if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
1002 UDPX_INC_STATS_BH(sk, UDP_MIB_INERRORS);
a246b010 1003 goto out_unlock;
1781f7f5
HX		 1004 }
1005
1006 UDPX_INC_STATS_BH(sk, UDP_MIB_INDATAGRAMS);
a246b010 1007
1570c1e4 1008 xprt_adjust_cwnd(task, copied);
1570c1e4 1009 xprt_complete_rqst(task, copied);
a246b010
CL		 1010
1011 out_unlock:
4a0f8c04 1012 spin_unlock(&xprt->transport_lock);
a246b010
CL		 1013 dropit:
1014 skb_free_datagram(sk, skb);
1015 out:
f064af1e 1016 read_unlock_bh(&sk->sk_callback_lock);
a246b010
CL		 1017}
1018
a519fc7a
TM		 1019/*
1020 * Helper function to force a TCP close if the server is sending
1021 * junk and/or it has put us in CLOSE_WAIT
1022 */
1023static void xs_tcp_force_close(struct rpc_xprt *xprt)
1024{
1025 set_bit(XPRT_CONNECTION_CLOSE, &xprt->state);
1026 xprt_force_disconnect(xprt);
1027}
1028
dd456471 1029static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
a246b010 1030{
51971139 1031 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
a246b010
CL		 1032 size_t len, used;
1033 char *p;
1034
51971139
CL		 1035 p = ((char *) &transport->tcp_fraghdr) + transport->tcp_offset;
1036 len = sizeof(transport->tcp_fraghdr) - transport->tcp_offset;
9d292316 1037 used = xdr_skb_read_bits(desc, p, len);
51971139 1038 transport->tcp_offset += used;
a246b010
CL		 1039 if (used != len)
1040 return;
808012fb 1041
51971139
CL		 1042 transport->tcp_reclen = ntohl(transport->tcp_fraghdr);
1043 if (transport->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
e136d092 1044 transport->tcp_flags |= TCP_RCV_LAST_FRAG;
a246b010 1045 else
e136d092 1046 transport->tcp_flags &= ~TCP_RCV_LAST_FRAG;
51971139 1047 transport->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
808012fb 1048
e136d092 1049 transport->tcp_flags &= ~TCP_RCV_COPY_FRAGHDR;
51971139 1050 transport->tcp_offset = 0;
808012fb 1051
a246b010 1052 /* Sanity check of the record length */
18dca02a 1053 if (unlikely(transport->tcp_reclen < 8)) {
46121cf7 1054 dprintk("RPC: invalid TCP record fragment length\n");
a519fc7a 1055 xs_tcp_force_close(xprt);
9903cd1c 1056 return;
a246b010 1057 }
46121cf7 1058 dprintk("RPC: reading TCP record fragment of length %d\n",
51971139 1059 transport->tcp_reclen);
a246b010
CL		 1060}
1061
51971139 1062static void xs_tcp_check_fraghdr(struct sock_xprt *transport)
a246b010 1063{
51971139 1064 if (transport->tcp_offset == transport->tcp_reclen) {
e136d092 1065 transport->tcp_flags |= TCP_RCV_COPY_FRAGHDR;
51971139 1066 transport->tcp_offset = 0;
e136d092
CL		 1067 if (transport->tcp_flags & TCP_RCV_LAST_FRAG) {
1068 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1069 transport->tcp_flags |= TCP_RCV_COPY_XID;
51971139 1070 transport->tcp_copied = 0;
a246b010
CL		 1071 }
1072 }
1073}
1074
dd456471 1075static inline void xs_tcp_read_xid(struct sock_xprt *transport, struct xdr_skb_reader *desc)
a246b010
CL		 1076{
1077 size_t len, used;
1078 char *p;
1079
51971139 1080 len = sizeof(transport->tcp_xid) - transport->tcp_offset;
46121cf7 1081 dprintk("RPC: reading XID (%Zu bytes)\n", len);
51971139 1082 p = ((char *) &transport->tcp_xid) + transport->tcp_offset;
9d292316 1083 used = xdr_skb_read_bits(desc, p, len);
51971139 1084 transport->tcp_offset += used;
a246b010
CL		 1085 if (used != len)
1086 return;
e136d092 1087 transport->tcp_flags &= ~TCP_RCV_COPY_XID;
f4a2e418 1088 transport->tcp_flags |= TCP_RCV_READ_CALLDIR;
51971139 1089 transport->tcp_copied = 4;
18dca02a
RL		 1090 dprintk("RPC: reading %s XID %08x\n",
1091 (transport->tcp_flags & TCP_RPC_REPLY) ? "reply for"
1092 : "request with",
51971139
CL		 1093 ntohl(transport->tcp_xid));
1094 xs_tcp_check_fraghdr(transport);
a246b010
CL		 1095}
1096
18dca02a
RL		 1097static inline void xs_tcp_read_calldir(struct sock_xprt *transport,
1098 struct xdr_skb_reader *desc)
a246b010 1099{
18dca02a
RL		 1100 size_t len, used;
1101 u32 offset;
b76ce561 1102 char *p;
18dca02a
RL		 1103
1104 /*
1105 * We want transport->tcp_offset to be 8 at the end of this routine
1106 * (4 bytes for the xid and 4 bytes for the call/reply flag).
1107 * When this function is called for the first time,
1108 * transport->tcp_offset is 4 (after having already read the xid).
1109 */
1110 offset = transport->tcp_offset - sizeof(transport->tcp_xid);
b76ce561 1111 len = sizeof(transport->tcp_calldir) - offset;
18dca02a 1112 dprintk("RPC: reading CALL/REPLY flag (%Zu bytes)\n", len);
b76ce561
TM		 1113 p = ((char *) &transport->tcp_calldir) + offset;
1114 used = xdr_skb_read_bits(desc, p, len);
18dca02a
RL		 1115 transport->tcp_offset += used;
1116 if (used != len)
1117 return;
f4a2e418 1118 transport->tcp_flags &= ~TCP_RCV_READ_CALLDIR;
f4a2e418
RL		 1119 /*
1120 * We don't yet have the XDR buffer, so we will write the calldir
1121 * out after we get the buffer from the 'struct rpc_rqst'
1122 */
b76ce561
TM		 1123 switch (ntohl(transport->tcp_calldir)) {
1124 case RPC_REPLY:
1125 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1126 transport->tcp_flags |= TCP_RCV_COPY_DATA;
18dca02a 1127 transport->tcp_flags |= TCP_RPC_REPLY;
b76ce561
TM		 1128 break;
1129 case RPC_CALL:
1130 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1131 transport->tcp_flags |= TCP_RCV_COPY_DATA;
18dca02a 1132 transport->tcp_flags &= ~TCP_RPC_REPLY;
b76ce561
TM		 1133 break;
1134 default:
1135 dprintk("RPC: invalid request message type\n");
a519fc7a 1136 xs_tcp_force_close(&transport->xprt);
b76ce561 1137 }
18dca02a
RL		 1138 xs_tcp_check_fraghdr(transport);
1139}
1140
44b98efd
RL		 1141static inline void xs_tcp_read_common(struct rpc_xprt *xprt,
1142 struct xdr_skb_reader *desc,
1143 struct rpc_rqst *req)
a246b010 1144{
44b98efd
RL		 1145 struct sock_xprt *transport =
1146 container_of(xprt, struct sock_xprt, xprt);
a246b010
CL		 1147 struct xdr_buf *rcvbuf;
1148 size_t len;
1149 ssize_t r;
1150
a246b010 1151 rcvbuf = &req->rq_private_buf;
f4a2e418
RL		 1152
1153 if (transport->tcp_flags & TCP_RCV_COPY_CALLDIR) {
1154 /*
1155 * Save the RPC direction in the XDR buffer
1156 */
f4a2e418 1157 memcpy(rcvbuf->head[0].iov_base + transport->tcp_copied,
b76ce561
TM		 1158 &transport->tcp_calldir,
1159 sizeof(transport->tcp_calldir));
1160 transport->tcp_copied += sizeof(transport->tcp_calldir);
f4a2e418 1161 transport->tcp_flags &= ~TCP_RCV_COPY_CALLDIR;
a246b010
CL		 1162 }
1163
a246b010 1164 len = desc->count;
51971139 1165 if (len > transport->tcp_reclen - transport->tcp_offset) {
dd456471 1166 struct xdr_skb_reader my_desc;
a246b010 1167
51971139 1168 len = transport->tcp_reclen - transport->tcp_offset;
a246b010
CL		 1169 memcpy(&my_desc, desc, sizeof(my_desc));
1170 my_desc.count = len;
51971139 1171 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
9d292316 1172 &my_desc, xdr_skb_read_bits);
a246b010
CL		 1173 desc->count -= r;
1174 desc->offset += r;
1175 } else
51971139 1176 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
9d292316 1177 desc, xdr_skb_read_bits);
a246b010
CL		 1178
1179 if (r > 0) {
51971139
CL		 1180 transport->tcp_copied += r;
1181 transport->tcp_offset += r;
a246b010
CL		 1182 }
1183 if (r != len) {
1184 /* Error when copying to the receive buffer,
1185 * usually because we weren't able to allocate
1186 * additional buffer pages. All we can do now
e136d092 1187 * is turn off TCP_RCV_COPY_DATA, so the request
a246b010
CL		 1188 * will not receive any additional updates,
1189 * and time out.
1190 * Any remaining data from this record will
1191 * be discarded.
1192 */
e136d092 1193 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
46121cf7 1194 dprintk("RPC: XID %08x truncated request\n",
51971139 1195 ntohl(transport->tcp_xid));
46121cf7
CL		 1196 dprintk("RPC: xprt = %p, tcp_copied = %lu, "
1197 "tcp_offset = %u, tcp_reclen = %u\n",
1198 xprt, transport->tcp_copied,
1199 transport->tcp_offset, transport->tcp_reclen);
44b98efd 1200 return;
a246b010
CL		 1201 }
1202
46121cf7 1203 dprintk("RPC: XID %08x read %Zd bytes\n",
51971139 1204 ntohl(transport->tcp_xid), r);
46121cf7
CL		 1205 dprintk("RPC: xprt = %p, tcp_copied = %lu, tcp_offset = %u, "
1206 "tcp_reclen = %u\n", xprt, transport->tcp_copied,
1207 transport->tcp_offset, transport->tcp_reclen);
51971139
CL		 1208
1209 if (transport->tcp_copied == req->rq_private_buf.buflen)
e136d092 1210 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
51971139 1211 else if (transport->tcp_offset == transport->tcp_reclen) {
e136d092
CL		 1212 if (transport->tcp_flags & TCP_RCV_LAST_FRAG)
1213 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
a246b010 1214 }
44b98efd
RL		 1215}
1216
1217/*
1218 * Finds the request corresponding to the RPC xid and invokes the common
1219 * tcp read code to read the data.
1220 */
1221static inline int xs_tcp_read_reply(struct rpc_xprt *xprt,
1222 struct xdr_skb_reader *desc)
1223{
1224 struct sock_xprt *transport =
1225 container_of(xprt, struct sock_xprt, xprt);
1226 struct rpc_rqst *req;
1227
1228 dprintk("RPC: read reply XID %08x\n", ntohl(transport->tcp_xid));
1229
1230 /* Find and lock the request corresponding to this xid */
1231 spin_lock(&xprt->transport_lock);
1232 req = xprt_lookup_rqst(xprt, transport->tcp_xid);
1233 if (!req) {
1234 dprintk("RPC: XID %08x request not found!\n",
1235 ntohl(transport->tcp_xid));
1236 spin_unlock(&xprt->transport_lock);
1237 return -1;
1238 }
1239
1240 xs_tcp_read_common(xprt, desc, req);
1241
e136d092 1242 if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
51971139 1243 xprt_complete_rqst(req->rq_task, transport->tcp_copied);
44b98efd 1244
4a0f8c04 1245 spin_unlock(&xprt->transport_lock);
44b98efd
RL		 1246 return 0;
1247}
1248
9e00abc3 1249#if defined(CONFIG_SUNRPC_BACKCHANNEL)
44b98efd
RL		 1250/*
1251 * Obtains an rpc_rqst previously allocated and invokes the common
1252 * tcp read code to read the data. The result is placed in the callback
1253 * queue.
1254 * If we're unable to obtain the rpc_rqst we schedule the closing of the
1255 * connection and return -1.
1256 */
1257static inline int xs_tcp_read_callback(struct rpc_xprt *xprt,
1258 struct xdr_skb_reader *desc)
1259{
1260 struct sock_xprt *transport =
1261 container_of(xprt, struct sock_xprt, xprt);
1262 struct rpc_rqst *req;
1263
1264 req = xprt_alloc_bc_request(xprt);
1265 if (req == NULL) {
1266 printk(KERN_WARNING "Callback slot table overflowed\n");
1267 xprt_force_disconnect(xprt);
1268 return -1;
1269 }
1270
1271 req->rq_xid = transport->tcp_xid;
1272 dprintk("RPC: read callback XID %08x\n", ntohl(req->rq_xid));
1273 xs_tcp_read_common(xprt, desc, req);
1274
1275 if (!(transport->tcp_flags & TCP_RCV_COPY_DATA)) {
1276 struct svc_serv *bc_serv = xprt->bc_serv;
1277
1278 /*
1279 * Add callback request to callback list. The callback
1280 * service sleeps on the sv_cb_waitq waiting for new
1281 * requests. Wake it up after adding enqueing the
1282 * request.
1283 */
1284 dprintk("RPC: add callback request to list\n");
1285 spin_lock(&bc_serv->sv_cb_lock);
1286 list_add(&req->rq_bc_list, &bc_serv->sv_cb_list);
1287 spin_unlock(&bc_serv->sv_cb_lock);
1288 wake_up(&bc_serv->sv_cb_waitq);
1289 }
1290
1291 req->rq_private_buf.len = transport->tcp_copied;
1292
1293 return 0;
1294}
1295
1296static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1297 struct xdr_skb_reader *desc)
1298{
1299 struct sock_xprt *transport =
1300 container_of(xprt, struct sock_xprt, xprt);
1301
1302 return (transport->tcp_flags & TCP_RPC_REPLY) ?
1303 xs_tcp_read_reply(xprt, desc) :
1304 xs_tcp_read_callback(xprt, desc);
1305}
1306#else
1307static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1308 struct xdr_skb_reader *desc)
1309{
1310 return xs_tcp_read_reply(xprt, desc);
1311}
9e00abc3 1312#endif /* CONFIG_SUNRPC_BACKCHANNEL */
44b98efd
RL		 1313
1314/*
1315 * Read data off the transport. This can be either an RPC_CALL or an
1316 * RPC_REPLY. Relay the processing to helper functions.
1317 */
1318static void xs_tcp_read_data(struct rpc_xprt *xprt,
1319 struct xdr_skb_reader *desc)
1320{
1321 struct sock_xprt *transport =
1322 container_of(xprt, struct sock_xprt, xprt);
1323
1324 if (_xs_tcp_read_data(xprt, desc) == 0)
1325 xs_tcp_check_fraghdr(transport);
1326 else {
1327 /*
1328 * The transport_lock protects the request handling.
1329 * There's no need to hold it to update the tcp_flags.
1330 */
1331 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1332 }
a246b010
CL		 1333}
1334
dd456471 1335static inline void xs_tcp_read_discard(struct sock_xprt *transport, struct xdr_skb_reader *desc)
a246b010
CL		 1336{
1337 size_t len;
1338
51971139 1339 len = transport->tcp_reclen - transport->tcp_offset;
a246b010
CL		 1340 if (len > desc->count)
1341 len = desc->count;
1342 desc->count -= len;
1343 desc->offset += len;
51971139 1344 transport->tcp_offset += len;
46121cf7 1345 dprintk("RPC: discarded %Zu bytes\n", len);
51971139 1346 xs_tcp_check_fraghdr(transport);
a246b010
CL		 1347}
1348
9903cd1c 1349static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
a246b010
CL		 1350{
1351 struct rpc_xprt *xprt = rd_desc->arg.data;
51971139 1352 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
dd456471 1353 struct xdr_skb_reader desc = {
a246b010
CL		 1354 .skb = skb,
1355 .offset = offset,
1356 .count = len,
9903cd1c 1357 };
a246b010 1358
46121cf7 1359 dprintk("RPC: xs_tcp_data_recv started\n");
a246b010
CL		 1360 do {
1361 /* Read in a new fragment marker if necessary */
1362 /* Can we ever really expect to get completely empty fragments? */
e136d092 1363 if (transport->tcp_flags & TCP_RCV_COPY_FRAGHDR) {
9903cd1c 1364 xs_tcp_read_fraghdr(xprt, &desc);
a246b010
CL		 1365 continue;
1366 }
1367 /* Read in the xid if necessary */
e136d092 1368 if (transport->tcp_flags & TCP_RCV_COPY_XID) {
51971139 1369 xs_tcp_read_xid(transport, &desc);
a246b010
CL		 1370 continue;
1371 }
18dca02a 1372 /* Read in the call/reply flag */
f4a2e418 1373 if (transport->tcp_flags & TCP_RCV_READ_CALLDIR) {
18dca02a
RL		 1374 xs_tcp_read_calldir(transport, &desc);
1375 continue;
1376 }
a246b010 1377 /* Read in the request data */
e136d092 1378 if (transport->tcp_flags & TCP_RCV_COPY_DATA) {
44b98efd 1379 xs_tcp_read_data(xprt, &desc);
a246b010
CL		 1380 continue;
1381 }
1382 /* Skip over any trailing bytes on short reads */
51971139 1383 xs_tcp_read_discard(transport, &desc);
a246b010 1384 } while (desc.count);
46121cf7 1385 dprintk("RPC: xs_tcp_data_recv done\n");
a246b010
CL		 1386 return len - desc.count;
1387}
1388
9903cd1c
CL		 1389/**
1390 * xs_tcp_data_ready - "data ready" callback for TCP sockets
1391 * @sk: socket with data to read
1392 * @bytes: how much data to read
1393 *
1394 */
1395static void xs_tcp_data_ready(struct sock *sk, int bytes)
a246b010
CL		 1396{
1397 struct rpc_xprt *xprt;
1398 read_descriptor_t rd_desc;
ff2d7db8 1399 int read;
a246b010 1400
46121cf7
CL		 1401 dprintk("RPC: xs_tcp_data_ready...\n");
1402
f064af1e 1403 read_lock_bh(&sk->sk_callback_lock);
9903cd1c 1404 if (!(xprt = xprt_from_sock(sk)))
a246b010 1405 goto out;
61d0a8e6
NB		 1406 /* Any data means we had a useful conversation, so
1407 * the we don't need to delay the next reconnect
1408 */
1409 if (xprt->reestablish_timeout)
1410 xprt->reestablish_timeout = 0;
1411
9903cd1c 1412 /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
a246b010 1413 rd_desc.arg.data = xprt;
ff2d7db8
TM		 1414 do {
1415 rd_desc.count = 65536;
1416 read = tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
1417 } while (read > 0);
a246b010 1418out:
f064af1e 1419 read_unlock_bh(&sk->sk_callback_lock);
a246b010
CL		 1420}
1421
7d1e8255
TM		 1422/*
1423 * Do the equivalent of linger/linger2 handling for dealing with
1424 * broken servers that don't close the socket in a timely
1425 * fashion
1426 */
1427static void xs_tcp_schedule_linger_timeout(struct rpc_xprt *xprt,
1428 unsigned long timeout)
1429{
1430 struct sock_xprt *transport;
1431
1432 if (xprt_test_and_set_connecting(xprt))
1433 return;
1434 set_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1435 transport = container_of(xprt, struct sock_xprt, xprt);
1436 queue_delayed_work(rpciod_workqueue, &transport->connect_worker,
1437 timeout);
1438}
1439
1440static void xs_tcp_cancel_linger_timeout(struct rpc_xprt *xprt)
1441{
1442 struct sock_xprt *transport;
1443
1444 transport = container_of(xprt, struct sock_xprt, xprt);
1445
1446 if (!test_bit(XPRT_CONNECTION_ABORT, &xprt->state) ||
1447 !cancel_delayed_work(&transport->connect_worker))
1448 return;
1449 clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1450 xprt_clear_connecting(xprt);
1451}
1452
4bc1e68e 1453static void xs_sock_reset_connection_flags(struct rpc_xprt *xprt)
7d1e8255
TM		 1454{
1455 smp_mb__before_clear_bit();
a519fc7a
TM		 1456 clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1457 clear_bit(XPRT_CONNECTION_CLOSE, &xprt->state);
7d1e8255
TM		 1458 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1459 clear_bit(XPRT_CLOSING, &xprt->state);
1460 smp_mb__after_clear_bit();
4bc1e68e
TM		 1461}
1462
1463static void xs_sock_mark_closed(struct rpc_xprt *xprt)
1464{
1465 xs_sock_reset_connection_flags(xprt);
7d1e8255
TM		 1466 /* Mark transport as closed and wake up all pending tasks */
1467 xprt_disconnect_done(xprt);
1468}
1469
9903cd1c
CL		 1470/**
1471 * xs_tcp_state_change - callback to handle TCP socket state changes
1472 * @sk: socket whose state has changed
1473 *
1474 */
1475static void xs_tcp_state_change(struct sock *sk)
a246b010 1476{
9903cd1c 1477 struct rpc_xprt *xprt;
a246b010 1478
f064af1e 1479 read_lock_bh(&sk->sk_callback_lock);
a246b010
CL		 1480 if (!(xprt = xprt_from_sock(sk)))
1481 goto out;
46121cf7 1482 dprintk("RPC: xs_tcp_state_change client %p...\n", xprt);
669502ff 1483 dprintk("RPC: state %x conn %d dead %d zapped %d sk_shutdown %d\n",
46121cf7
CL		 1484 sk->sk_state, xprt_connected(xprt),
1485 sock_flag(sk, SOCK_DEAD),
669502ff
AC		 1486 sock_flag(sk, SOCK_ZAPPED),
1487 sk->sk_shutdown);
a246b010
CL		 1488
1489 switch (sk->sk_state) {
1490 case TCP_ESTABLISHED:
f064af1e 1491 spin_lock(&xprt->transport_lock);
a246b010 1492 if (!xprt_test_and_set_connected(xprt)) {
51971139
CL		 1493 struct sock_xprt *transport = container_of(xprt,
1494 struct sock_xprt, xprt);
1495
a246b010 1496 /* Reset TCP record info */
51971139
CL		 1497 transport->tcp_offset = 0;
1498 transport->tcp_reclen = 0;
1499 transport->tcp_copied = 0;
e136d092
CL		 1500 transport->tcp_flags =
1501 TCP_RCV_COPY_FRAGHDR | TCP_RCV_COPY_XID;
51971139 1502
2a491991 1503 xprt_wake_pending_tasks(xprt, -EAGAIN);
a246b010 1504 }
f064af1e 1505 spin_unlock(&xprt->transport_lock);
a246b010 1506 break;
3b948ae5
TM		 1507 case TCP_FIN_WAIT1:
1508 /* The client initiated a shutdown of the socket */
7c1d71cf 1509 xprt->connect_cookie++;
663b8858 1510 xprt->reestablish_timeout = 0;
3b948ae5
TM		 1511 set_bit(XPRT_CLOSING, &xprt->state);
1512 smp_mb__before_clear_bit();
1513 clear_bit(XPRT_CONNECTED, &xprt->state);
ef803670 1514 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
3b948ae5 1515 smp_mb__after_clear_bit();
25fe6142 1516 xs_tcp_schedule_linger_timeout(xprt, xs_tcp_fin_timeout);
a246b010 1517 break;
632e3bdc 1518 case TCP_CLOSE_WAIT:
3b948ae5 1519 /* The server initiated a shutdown of the socket */
7c1d71cf 1520 xprt->connect_cookie++;
d0bea455 1521 clear_bit(XPRT_CONNECTED, &xprt->state);
a519fc7a 1522 xs_tcp_force_close(xprt);
663b8858
TM		 1523 case TCP_CLOSING:
1524 /*
1525 * If the server closed down the connection, make sure that
1526 * we back off before reconnecting
1527 */
1528 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1529 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
3b948ae5
TM		 1530 break;
1531 case TCP_LAST_ACK:
670f9457 1532 set_bit(XPRT_CLOSING, &xprt->state);
25fe6142 1533 xs_tcp_schedule_linger_timeout(xprt, xs_tcp_fin_timeout);
3b948ae5
TM		 1534 smp_mb__before_clear_bit();
1535 clear_bit(XPRT_CONNECTED, &xprt->state);
1536 smp_mb__after_clear_bit();
1537 break;
1538 case TCP_CLOSE:
7d1e8255
TM		 1539 xs_tcp_cancel_linger_timeout(xprt);
1540 xs_sock_mark_closed(xprt);
a246b010
CL		 1541 }
1542 out:
f064af1e 1543 read_unlock_bh(&sk->sk_callback_lock);
a246b010
CL		 1544}
1545
1f0fa154
IJ		 1546static void xs_write_space(struct sock *sk)
1547{
1548 struct socket *sock;
1549 struct rpc_xprt *xprt;
1550
1551 if (unlikely(!(sock = sk->sk_socket)))
1552 return;
1553 clear_bit(SOCK_NOSPACE, &sock->flags);
1554
1555 if (unlikely(!(xprt = xprt_from_sock(sk))))
1556 return;
1557 if (test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags) == 0)
1558 return;
1559
1560 xprt_write_space(xprt);
1561}
1562
9903cd1c 1563/**
c7b2cae8
CL		 1564 * xs_udp_write_space - callback invoked when socket buffer space
1565 * becomes available
9903cd1c
CL		 1566 * @sk: socket whose state has changed
1567 *
a246b010
CL		 1568 * Called when more output buffer space is available for this socket.
1569 * We try not to wake our writers until they can make "significant"
c7b2cae8 1570 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
a246b010
CL		 1571 * with a bunch of small requests.
1572 */
c7b2cae8 1573static void xs_udp_write_space(struct sock *sk)
a246b010 1574{
f064af1e 1575 read_lock_bh(&sk->sk_callback_lock);
a246b010 1576
c7b2cae8 1577 /* from net/core/sock.c:sock_def_write_space */
1f0fa154
IJ		 1578 if (sock_writeable(sk))
1579 xs_write_space(sk);
a246b010 1580
f064af1e 1581 read_unlock_bh(&sk->sk_callback_lock);
c7b2cae8 1582}
a246b010 1583
c7b2cae8
CL		 1584/**
1585 * xs_tcp_write_space - callback invoked when socket buffer space
1586 * becomes available
1587 * @sk: socket whose state has changed
1588 *
1589 * Called when more output buffer space is available for this socket.
1590 * We try not to wake our writers until they can make "significant"
1591 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1592 * with a bunch of small requests.
1593 */
1594static void xs_tcp_write_space(struct sock *sk)
1595{
f064af1e 1596 read_lock_bh(&sk->sk_callback_lock);
c7b2cae8
CL		 1597
1598 /* from net/core/stream.c:sk_stream_write_space */
1f0fa154
IJ		 1599 if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk))
1600 xs_write_space(sk);
b6ddf64f 1601
f064af1e 1602 read_unlock_bh(&sk->sk_callback_lock);
a246b010
CL		 1603}
1604
470056c2 1605static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
a246b010 1606{
ee0ac0c2
CL		 1607 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1608 struct sock *sk = transport->inet;
a246b010 1609
7c6e066e 1610 if (transport->rcvsize) {
a246b010 1611 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
7c6e066e 1612 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
a246b010 1613 }
7c6e066e 1614 if (transport->sndsize) {
a246b010 1615 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
7c6e066e 1616 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
a246b010
CL		 1617 sk->sk_write_space(sk);
1618 }
1619}
1620
43118c29 1621/**
470056c2 1622 * xs_udp_set_buffer_size - set send and receive limits
43118c29 1623 * @xprt: generic transport
470056c2
CL		 1624 * @sndsize: requested size of send buffer, in bytes
1625 * @rcvsize: requested size of receive buffer, in bytes
43118c29 1626 *
470056c2 1627 * Set socket send and receive buffer size limits.
43118c29 1628 */
470056c2 1629static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
43118c29 1630{
7c6e066e
CL		 1631 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1632
1633 transport->sndsize = 0;
470056c2 1634 if (sndsize)
7c6e066e
CL		 1635 transport->sndsize = sndsize + 1024;
1636 transport->rcvsize = 0;
470056c2 1637 if (rcvsize)
7c6e066e 1638 transport->rcvsize = rcvsize + 1024;
470056c2
CL		 1639
1640 xs_udp_do_set_buffer_size(xprt);
43118c29
CL		 1641}
1642
46c0ee8b
CL		 1643/**
1644 * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1645 * @task: task that timed out
1646 *
1647 * Adjust the congestion window after a retransmit timeout has occurred.
1648 */
1649static void xs_udp_timer(struct rpc_task *task)
1650{
1651 xprt_adjust_cwnd(task, -ETIMEDOUT);
1652}
1653
b85d8806
CL		 1654static unsigned short xs_get_random_port(void)
1655{
1656 unsigned short range = xprt_max_resvport - xprt_min_resvport;
1657 unsigned short rand = (unsigned short) net_random() % range;
1658 return rand + xprt_min_resvport;
1659}
1660
92200412
CL		 1661/**
1662 * xs_set_port - reset the port number in the remote endpoint address
1663 * @xprt: generic transport
1664 * @port: new port number
1665 *
1666 */
1667static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1668{
46121cf7 1669 dprintk("RPC: setting port for xprt %p to %u\n", xprt, port);
c4efcb1d 1670
9dc3b095
CL		 1671 rpc_set_port(xs_addr(xprt), port);
1672 xs_update_peer_port(xprt);
92200412
CL		 1673}
1674
5d4ec932 1675static unsigned short xs_get_srcport(struct sock_xprt *transport)
67a391d7 1676{
fbfffbd5 1677 unsigned short port = transport->srcport;
67a391d7
TM		 1678
1679 if (port == 0 && transport->xprt.resvport)
1680 port = xs_get_random_port();
1681 return port;
1682}
1683
baaf4e48 1684static unsigned short xs_next_srcport(struct sock_xprt *transport, unsigned short port)
67a391d7 1685{
fbfffbd5
CL		 1686 if (transport->srcport != 0)
1687 transport->srcport = 0;
67a391d7
TM		 1688 if (!transport->xprt.resvport)
1689 return 0;
1690 if (port <= xprt_min_resvport || port > xprt_max_resvport)
1691 return xprt_max_resvport;
1692 return --port;
1693}
beb59b68 1694static int xs_bind(struct sock_xprt *transport, struct socket *sock)
a246b010 1695{
beb59b68 1696 struct sockaddr_storage myaddr;
67a391d7 1697 int err, nloop = 0;
5d4ec932 1698 unsigned short port = xs_get_srcport(transport);
67a391d7 1699 unsigned short last;
a246b010 1700
beb59b68 1701 memcpy(&myaddr, &transport->srcaddr, transport->xprt.addrlen);
a246b010 1702 do {
beb59b68
PE		 1703 rpc_set_port((struct sockaddr *)&myaddr, port);
1704 err = kernel_bind(sock, (struct sockaddr *)&myaddr,
1705 transport->xprt.addrlen);
67a391d7 1706 if (port == 0)
d3bc9a1d 1707 break;
a246b010 1708 if (err == 0) {
fbfffbd5 1709 transport->srcport = port;
d3bc9a1d 1710 break;
a246b010 1711 }
67a391d7 1712 last = port;
baaf4e48 1713 port = xs_next_srcport(transport, port);
67a391d7
TM		 1714 if (port > last)
1715 nloop++;
1716 } while (err == -EADDRINUSE && nloop != 2);
90058d37 1717
4232e863 1718 if (myaddr.ss_family == AF_INET)
beb59b68
PE		 1719 dprintk("RPC: %s %pI4:%u: %s (%d)\n", __func__,
1720 &((struct sockaddr_in *)&myaddr)->sin_addr,
1721 port, err ? "failed" : "ok", err);
1722 else
1723 dprintk("RPC: %s %pI6:%u: %s (%d)\n", __func__,
1724 &((struct sockaddr_in6 *)&myaddr)->sin6_addr,
1725 port, err ? "failed" : "ok", err);
a246b010
CL		 1726 return err;
1727}
1728
176e21ee
CL		 1729/*
1730 * We don't support autobind on AF_LOCAL sockets
1731 */
1732static void xs_local_rpcbind(struct rpc_task *task)
1733{
1734 xprt_set_bound(task->tk_xprt);
1735}
1736
1737static void xs_local_set_port(struct rpc_xprt *xprt, unsigned short port)
1738{
1739}
beb59b68 1740
ed07536e
PZ		 1741#ifdef CONFIG_DEBUG_LOCK_ALLOC
1742static struct lock_class_key xs_key[2];
1743static struct lock_class_key xs_slock_key[2];
1744
176e21ee
CL		 1745static inline void xs_reclassify_socketu(struct socket *sock)
1746{
1747 struct sock *sk = sock->sk;
1748
176e21ee
CL		 1749 sock_lock_init_class_and_name(sk, "slock-AF_LOCAL-RPC",
1750 &xs_slock_key[1], "sk_lock-AF_LOCAL-RPC", &xs_key[1]);
1751}
1752
8945ee5e 1753static inline void xs_reclassify_socket4(struct socket *sock)
ed07536e
PZ		 1754{
1755 struct sock *sk = sock->sk;
8945ee5e 1756
8945ee5e
CL		 1757 sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1758 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1759}
ed07536e 1760
8945ee5e
CL		 1761static inline void xs_reclassify_socket6(struct socket *sock)
1762{
1763 struct sock *sk = sock->sk;
ed07536e 1764
8945ee5e
CL		 1765 sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1766 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
ed07536e 1767}
6bc9638a
PE		 1768
1769static inline void xs_reclassify_socket(int family, struct socket *sock)
1770{
1b7a1819
WAA		 1771 WARN_ON_ONCE(sock_owned_by_user(sock->sk));
1772 if (sock_owned_by_user(sock->sk))
1773 return;
1774
4232e863 1775 switch (family) {
176e21ee
CL		 1776 case AF_LOCAL:
1777 xs_reclassify_socketu(sock);
1778 break;
4232e863 1779 case AF_INET:
6bc9638a 1780 xs_reclassify_socket4(sock);
4232e863
CL		 1781 break;
1782 case AF_INET6:
6bc9638a 1783 xs_reclassify_socket6(sock);
4232e863
CL		 1784 break;
1785 }
6bc9638a 1786}
ed07536e 1787#else
176e21ee
CL		 1788static inline void xs_reclassify_socketu(struct socket *sock)
1789{
1790}
1791
8945ee5e
CL		 1792static inline void xs_reclassify_socket4(struct socket *sock)
1793{
1794}
1795
1796static inline void xs_reclassify_socket6(struct socket *sock)
ed07536e
PZ		 1797{
1798}
6bc9638a
PE		 1799
1800static inline void xs_reclassify_socket(int family, struct socket *sock)
1801{
1802}
ed07536e
PZ		 1803#endif
1804
6bc9638a
PE		 1805static struct socket *xs_create_sock(struct rpc_xprt *xprt,
1806 struct sock_xprt *transport, int family, int type, int protocol)
22f79326
PE		 1807{
1808 struct socket *sock;
1809 int err;
1810
6bc9638a 1811 err = __sock_create(xprt->xprt_net, family, type, protocol, &sock, 1);
22f79326
PE		 1812 if (err < 0) {
1813 dprintk("RPC: can't create %d transport socket (%d).\n",
1814 protocol, -err);
1815 goto out;
1816 }
6bc9638a 1817 xs_reclassify_socket(family, sock);
22f79326 1818
4cea288a
BH		 1819 err = xs_bind(transport, sock);
1820 if (err) {
22f79326
PE		 1821 sock_release(sock);
1822 goto out;
1823 }
1824
1825 return sock;
1826out:
1827 return ERR_PTR(err);
1828}
1829
176e21ee
CL		 1830static int xs_local_finish_connecting(struct rpc_xprt *xprt,
1831 struct socket *sock)
1832{
1833 struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1834 xprt);
1835
1836 if (!transport->inet) {
1837 struct sock *sk = sock->sk;
1838
1839 write_lock_bh(&sk->sk_callback_lock);
1840
1841 xs_save_old_callbacks(transport, sk);
1842
1843 sk->sk_user_data = xprt;
1844 sk->sk_data_ready = xs_local_data_ready;
1845 sk->sk_write_space = xs_udp_write_space;
176e21ee
CL		 1846 sk->sk_allocation = GFP_ATOMIC;
1847
1848 xprt_clear_connected(xprt);
1849
1850 /* Reset to new socket */
1851 transport->sock = sock;
1852 transport->inet = sk;
1853
1854 write_unlock_bh(&sk->sk_callback_lock);
1855 }
1856
1857 /* Tell the socket layer to start connecting... */
1858 xprt->stat.connect_count++;
1859 xprt->stat.connect_start = jiffies;
1860 return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, 0);
1861}
1862
1863/**
1864 * xs_local_setup_socket - create AF_LOCAL socket, connect to a local endpoint
1865 * @xprt: RPC transport to connect
1866 * @transport: socket transport to connect
1867 * @create_sock: function to create a socket of the correct type
1868 *
1869 * Invoked by a work queue tasklet.
1870 */
1871static void xs_local_setup_socket(struct work_struct *work)
1872{
1873 struct sock_xprt *transport =
1874 container_of(work, struct sock_xprt, connect_worker.work);
1875 struct rpc_xprt *xprt = &transport->xprt;
1876 struct socket *sock;
1877 int status = -EIO;
1878
5cf02d09
JL		 1879 current->flags |= PF_FSTRANS;
1880
176e21ee
CL		 1881 clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1882 status = __sock_create(xprt->xprt_net, AF_LOCAL,
1883 SOCK_STREAM, 0, &sock, 1);
1884 if (status < 0) {
1885 dprintk("RPC: can't create AF_LOCAL "
1886 "transport socket (%d).\n", -status);
1887 goto out;
1888 }
1889 xs_reclassify_socketu(sock);
1890
1891 dprintk("RPC: worker connecting xprt %p via AF_LOCAL to %s\n",
1892 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1893
1894 status = xs_local_finish_connecting(xprt, sock);
1895 switch (status) {
1896 case 0:
1897 dprintk("RPC: xprt %p connected to %s\n",
1898 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1899 xprt_set_connected(xprt);
1900 break;
1901 case -ENOENT:
1902 dprintk("RPC: xprt %p: socket %s does not exist\n",
1903 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1904 break;
1905 default:
1906 printk(KERN_ERR "%s: unhandled error (%d) connecting to %s\n",
1907 __func__, -status,
1908 xprt->address_strings[RPC_DISPLAY_ADDR]);
1909 }
1910
1911out:
1912 xprt_clear_connecting(xprt);
1913 xprt_wake_pending_tasks(xprt, status);
5cf02d09 1914 current->flags &= ~PF_FSTRANS;
176e21ee
CL		 1915}
1916
a564b8f0
MG		 1917#ifdef CONFIG_SUNRPC_SWAP
1918static void xs_set_memalloc(struct rpc_xprt *xprt)
1919{
1920 struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1921 xprt);
1922
1923 if (xprt->swapper)
1924 sk_set_memalloc(transport->inet);
1925}
1926
1927/**
1928 * xs_swapper - Tag this transport as being used for swap.
1929 * @xprt: transport to tag
1930 * @enable: enable/disable
1931 *
1932 */
1933int xs_swapper(struct rpc_xprt *xprt, int enable)
1934{
1935 struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1936 xprt);
1937 int err = 0;
1938
1939 if (enable) {
1940 xprt->swapper++;
1941 xs_set_memalloc(xprt);
1942 } else if (xprt->swapper) {
1943 xprt->swapper--;
1944 sk_clear_memalloc(transport->inet);
1945 }
1946
1947 return err;
1948}
1949EXPORT_SYMBOL_GPL(xs_swapper);
1950#else
1951static void xs_set_memalloc(struct rpc_xprt *xprt)
1952{
1953}
1954#endif
1955
16be2d20
CL		 1956static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
1957{
1958 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1959
1960 if (!transport->inet) {
1961 struct sock *sk = sock->sk;
1962
1963 write_lock_bh(&sk->sk_callback_lock);
1964
2a9e1cfa
TM		 1965 xs_save_old_callbacks(transport, sk);
1966
16be2d20 1967 sk->sk_user_data = xprt;
16be2d20
CL		 1968 sk->sk_data_ready = xs_udp_data_ready;
1969 sk->sk_write_space = xs_udp_write_space;
1970 sk->sk_no_check = UDP_CSUM_NORCV;
1971 sk->sk_allocation = GFP_ATOMIC;
1972
1973 xprt_set_connected(xprt);
1974
1975 /* Reset to new socket */
1976 transport->sock = sock;
1977 transport->inet = sk;
1978
a564b8f0
MG		 1979 xs_set_memalloc(xprt);
1980
16be2d20
CL		 1981 write_unlock_bh(&sk->sk_callback_lock);
1982 }
1983 xs_udp_do_set_buffer_size(xprt);
1984}
1985
8c14ff2a 1986static void xs_udp_setup_socket(struct work_struct *work)
a246b010 1987{
34161db6
TM		 1988 struct sock_xprt *transport =
1989 container_of(work, struct sock_xprt, connect_worker.work);
c8475461 1990 struct rpc_xprt *xprt = &transport->xprt;
ee0ac0c2 1991 struct socket *sock = transport->sock;
b65c0310 1992 int status = -EIO;
9903cd1c 1993
5cf02d09
JL		 1994 current->flags |= PF_FSTRANS;
1995
b0d93ad5 1996 /* Start by resetting any existing state */
fe315e76 1997 xs_reset_transport(transport);
8c14ff2a
PE		 1998 sock = xs_create_sock(xprt, transport,
1999 xs_addr(xprt)->sa_family, SOCK_DGRAM, IPPROTO_UDP);
b65c0310 2000 if (IS_ERR(sock))
b0d93ad5 2001 goto out;
68e220bd 2002
c740eff8
CL		 2003 dprintk("RPC: worker connecting xprt %p via %s to "
2004 "%s (port %s)\n", xprt,
2005 xprt->address_strings[RPC_DISPLAY_PROTO],
2006 xprt->address_strings[RPC_DISPLAY_ADDR],
2007 xprt->address_strings[RPC_DISPLAY_PORT]);
68e220bd
CL		 2008
2009 xs_udp_finish_connecting(xprt, sock);
b0d93ad5
CL		 2010 status = 0;
2011out:
b0d93ad5 2012 xprt_clear_connecting(xprt);
7d1e8255 2013 xprt_wake_pending_tasks(xprt, status);
5cf02d09 2014 current->flags &= ~PF_FSTRANS;
a246b010
CL		 2015}
2016
3167e12c
CL		 2017/*
2018 * We need to preserve the port number so the reply cache on the server can
2019 * find our cached RPC replies when we get around to reconnecting.
2020 */
58dddac9 2021static void xs_abort_connection(struct sock_xprt *transport)
3167e12c
CL		 2022{
2023 int result;
3167e12c
CL		 2024 struct sockaddr any;
2025
58dddac9 2026 dprintk("RPC: disconnecting xprt %p to reuse port\n", transport);
3167e12c
CL		 2027
2028 /*
2029 * Disconnect the transport socket by doing a connect operation
2030 * with AF_UNSPEC. This should return immediately...
2031 */
2032 memset(&any, 0, sizeof(any));
2033 any.sa_family = AF_UNSPEC;
ee0ac0c2 2034 result = kernel_connect(transport->sock, &any, sizeof(any), 0);
7d1e8255 2035 if (!result)
4bc1e68e
TM		 2036 xs_sock_reset_connection_flags(&transport->xprt);
2037 dprintk("RPC: AF_UNSPEC connect return code %d\n", result);
3167e12c
CL		 2038}
2039
58dddac9 2040static void xs_tcp_reuse_connection(struct sock_xprt *transport)
40d2549d
TM		 2041{
2042 unsigned int state = transport->inet->sk_state;
2043
669502ff
AC		 2044 if (state == TCP_CLOSE && transport->sock->state == SS_UNCONNECTED) {
2045 /* we don't need to abort the connection if the socket
2046 * hasn't undergone a shutdown
2047 */
2048 if (transport->inet->sk_shutdown == 0)
2049 return;
2050 dprintk("RPC: %s: TCP_CLOSEd and sk_shutdown set to %d\n",
2051 __func__, transport->inet->sk_shutdown);
2052 }
2053 if ((1 << state) & (TCPF_ESTABLISHED|TCPF_SYN_SENT)) {
2054 /* we don't need to abort the connection if the socket
2055 * hasn't undergone a shutdown
2056 */
2057 if (transport->inet->sk_shutdown == 0)
2058 return;
2059 dprintk("RPC: %s: ESTABLISHED/SYN_SENT "
2060 "sk_shutdown set to %d\n",
2061 __func__, transport->inet->sk_shutdown);
2062 }
58dddac9 2063 xs_abort_connection(transport);
40d2549d
TM		 2064}
2065
16be2d20 2066static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
a246b010 2067{
16be2d20 2068 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
fe19a96b 2069 int ret = -ENOTCONN;
edb267a6 2070
ee0ac0c2 2071 if (!transport->inet) {
b0d93ad5
CL		 2072 struct sock *sk = sock->sk;
2073
2074 write_lock_bh(&sk->sk_callback_lock);
2075
2a9e1cfa
TM		 2076 xs_save_old_callbacks(transport, sk);
2077
b0d93ad5 2078 sk->sk_user_data = xprt;
b0d93ad5
CL		 2079 sk->sk_data_ready = xs_tcp_data_ready;
2080 sk->sk_state_change = xs_tcp_state_change;
2081 sk->sk_write_space = xs_tcp_write_space;
b079fa7b 2082 sk->sk_allocation = GFP_ATOMIC;
3167e12c
CL		 2083
2084 /* socket options */
2085 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
2086 sock_reset_flag(sk, SOCK_LINGER);
2087 tcp_sk(sk)->linger2 = 0;
2088 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
b0d93ad5
CL		 2089
2090 xprt_clear_connected(xprt);
2091
2092 /* Reset to new socket */
ee0ac0c2
CL		 2093 transport->sock = sock;
2094 transport->inet = sk;
b0d93ad5
CL		 2095
2096 write_unlock_bh(&sk->sk_callback_lock);
2097 }
2098
01d37c42 2099 if (!xprt_bound(xprt))
fe19a96b 2100 goto out;
01d37c42 2101
a564b8f0
MG		 2102 xs_set_memalloc(xprt);
2103
b0d93ad5 2104 /* Tell the socket layer to start connecting... */
262ca07d
CL		 2105 xprt->stat.connect_count++;
2106 xprt->stat.connect_start = jiffies;
fe19a96b
TM		 2107 ret = kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
2108 switch (ret) {
2109 case 0:
2110 case -EINPROGRESS:
2111 /* SYN_SENT! */
2112 xprt->connect_cookie++;
2113 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2114 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2115 }
2116out:
2117 return ret;
16be2d20
CL		 2118}
2119
9903cd1c 2120/**
b61d59ff
TM		 2121 * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint
2122 * @xprt: RPC transport to connect
2123 * @transport: socket transport to connect
2124 * @create_sock: function to create a socket of the correct type
9903cd1c
CL		 2125 *
2126 * Invoked by a work queue tasklet.
a246b010 2127 */
cdd518d5 2128static void xs_tcp_setup_socket(struct work_struct *work)
a246b010 2129{
cdd518d5
PE		 2130 struct sock_xprt *transport =
2131 container_of(work, struct sock_xprt, connect_worker.work);
ee0ac0c2 2132 struct socket *sock = transport->sock;
a9f5f0f7 2133 struct rpc_xprt *xprt = &transport->xprt;
b61d59ff 2134 int status = -EIO;
a246b010 2135
5cf02d09
JL		 2136 current->flags |= PF_FSTRANS;
2137
ee0ac0c2 2138 if (!sock) {
7d1e8255 2139 clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
cdd518d5
PE		 2140 sock = xs_create_sock(xprt, transport,
2141 xs_addr(xprt)->sa_family, SOCK_STREAM, IPPROTO_TCP);
b61d59ff
TM		 2142 if (IS_ERR(sock)) {
2143 status = PTR_ERR(sock);
3167e12c
CL		 2144 goto out;
2145 }
7d1e8255
TM		 2146 } else {
2147 int abort_and_exit;
a246b010 2148
7d1e8255
TM		 2149 abort_and_exit = test_and_clear_bit(XPRT_CONNECTION_ABORT,
2150 &xprt->state);
3167e12c 2151 /* "close" the socket, preserving the local port */
58dddac9 2152 xs_tcp_reuse_connection(transport);
a246b010 2153
7d1e8255
TM		 2154 if (abort_and_exit)
2155 goto out_eagain;
2156 }
a246b010 2157
c740eff8
CL		 2158 dprintk("RPC: worker connecting xprt %p via %s to "
2159 "%s (port %s)\n", xprt,
2160 xprt->address_strings[RPC_DISPLAY_PROTO],
2161 xprt->address_strings[RPC_DISPLAY_ADDR],
2162 xprt->address_strings[RPC_DISPLAY_PORT]);
edb267a6 2163
16be2d20 2164 status = xs_tcp_finish_connecting(xprt, sock);
46121cf7
CL		 2165 dprintk("RPC: %p connect status %d connected %d sock state %d\n",
2166 xprt, -status, xprt_connected(xprt),
2167 sock->sk->sk_state);
2a491991 2168 switch (status) {
f75e6745
TM		 2169 default:
2170 printk("%s: connect returned unhandled error %d\n",
2171 __func__, status);
2172 case -EADDRNOTAVAIL:
2173 /* We're probably in TIME_WAIT. Get rid of existing socket,
2174 * and retry
2175 */
a519fc7a 2176 xs_tcp_force_close(xprt);
88b5ed73 2177 break;
8a2cec29
TM		 2178 case -ECONNREFUSED:
2179 case -ECONNRESET:
2180 case -ENETUNREACH:
2181 /* retry with existing socket, after a delay */
2a491991
TM		 2182 case 0:
2183 case -EINPROGRESS:
2184 case -EALREADY:
7d1e8255 2185 xprt_clear_connecting(xprt);
5cf02d09 2186 current->flags &= ~PF_FSTRANS;
7d1e8255 2187 return;
9fcfe0c8
TM		 2188 case -EINVAL:
2189 /* Happens, for instance, if the user specified a link
2190 * local IPv6 address without a scope-id.
2191 */
2192 goto out;
a246b010 2193 }
7d1e8255 2194out_eagain:
2a491991 2195 status = -EAGAIN;
a246b010 2196out:
68e220bd 2197 xprt_clear_connecting(xprt);
7d1e8255 2198 xprt_wake_pending_tasks(xprt, status);
5cf02d09 2199 current->flags &= ~PF_FSTRANS;
68e220bd 2200}
b0d93ad5 2201
9903cd1c
CL		 2202/**
2203 * xs_connect - connect a socket to a remote endpoint
2204 * @task: address of RPC task that manages state of connect request
2205 *
2206 * TCP: If the remote end dropped the connection, delay reconnecting.
03bf4b70
CL		 2207 *
2208 * UDP socket connects are synchronous, but we use a work queue anyway
2209 * to guarantee that even unprivileged user processes can set up a
2210 * socket on a privileged port.
2211 *
2212 * If a UDP socket connect fails, the delay behavior here prevents
2213 * retry floods (hard mounts).
9903cd1c
CL		 2214 */
2215static void xs_connect(struct rpc_task *task)
a246b010
CL		 2216{
2217 struct rpc_xprt *xprt = task->tk_xprt;
ee0ac0c2 2218 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
a246b010 2219
09a21c41 2220 if (transport->sock != NULL && !RPC_IS_SOFTCONN(task)) {
46121cf7
CL		 2221 dprintk("RPC: xs_connect delayed xprt %p for %lu "
2222 "seconds\n",
03bf4b70 2223 xprt, xprt->reestablish_timeout / HZ);
c1384c9c
TM		 2224 queue_delayed_work(rpciod_workqueue,
2225 &transport->connect_worker,
2226 xprt->reestablish_timeout);
03bf4b70 2227 xprt->reestablish_timeout <<= 1;
61d0a8e6
NB		 2228 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2229 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
03bf4b70
CL		 2230 if (xprt->reestablish_timeout > XS_TCP_MAX_REEST_TO)
2231 xprt->reestablish_timeout = XS_TCP_MAX_REEST_TO;
b0d93ad5 2232 } else {
46121cf7 2233 dprintk("RPC: xs_connect scheduled xprt %p\n", xprt);
c1384c9c
TM		 2234 queue_delayed_work(rpciod_workqueue,
2235 &transport->connect_worker, 0);
a246b010
CL		 2236 }
2237}
2238
176e21ee
CL		 2239/**
2240 * xs_local_print_stats - display AF_LOCAL socket-specifc stats
2241 * @xprt: rpc_xprt struct containing statistics
2242 * @seq: output file
2243 *
2244 */
2245static void xs_local_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2246{
2247 long idle_time = 0;
2248
2249 if (xprt_connected(xprt))
2250 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2251
2252 seq_printf(seq, "\txprt:\tlocal %lu %lu %lu %ld %lu %lu %lu "
15a45206 2253 "%llu %llu %lu %llu %llu\n",
176e21ee
CL		 2254 xprt->stat.bind_count,
2255 xprt->stat.connect_count,
2256 xprt->stat.connect_time,
2257 idle_time,
2258 xprt->stat.sends,
2259 xprt->stat.recvs,
2260 xprt->stat.bad_xids,
2261 xprt->stat.req_u,
15a45206
AA		 2262 xprt->stat.bklog_u,
2263 xprt->stat.max_slots,
2264 xprt->stat.sending_u,
2265 xprt->stat.pending_u);
176e21ee
CL		 2266}
2267
262ca07d
CL		 2268/**
2269 * xs_udp_print_stats - display UDP socket-specifc stats
2270 * @xprt: rpc_xprt struct containing statistics
2271 * @seq: output file
2272 *
2273 */
2274static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2275{
c8475461
CL		 2276 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2277
15a45206
AA		 2278 seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %llu %llu "
2279 "%lu %llu %llu\n",
fbfffbd5 2280 transport->srcport,
262ca07d
CL		 2281 xprt->stat.bind_count,
2282 xprt->stat.sends,
2283 xprt->stat.recvs,
2284 xprt->stat.bad_xids,
2285 xprt->stat.req_u,
15a45206
AA		 2286 xprt->stat.bklog_u,
2287 xprt->stat.max_slots,
2288 xprt->stat.sending_u,
2289 xprt->stat.pending_u);
262ca07d
CL		 2290}
2291
2292/**
2293 * xs_tcp_print_stats - display TCP socket-specifc stats
2294 * @xprt: rpc_xprt struct containing statistics
2295 * @seq: output file
2296 *
2297 */
2298static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2299{
c8475461 2300 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
262ca07d
CL		 2301 long idle_time = 0;
2302
2303 if (xprt_connected(xprt))
2304 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2305
15a45206
AA		 2306 seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu "
2307 "%llu %llu %lu %llu %llu\n",
fbfffbd5 2308 transport->srcport,
262ca07d
CL		 2309 xprt->stat.bind_count,
2310 xprt->stat.connect_count,
2311 xprt->stat.connect_time,
2312 idle_time,
2313 xprt->stat.sends,
2314 xprt->stat.recvs,
2315 xprt->stat.bad_xids,
2316 xprt->stat.req_u,
15a45206
AA		 2317 xprt->stat.bklog_u,
2318 xprt->stat.max_slots,
2319 xprt->stat.sending_u,
2320 xprt->stat.pending_u);
262ca07d
CL		 2321}
2322
4cfc7e60
RI		 2323/*
2324 * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason
2325 * we allocate pages instead doing a kmalloc like rpc_malloc is because we want
2326 * to use the server side send routines.
2327 */
5a51f13a 2328static void *bc_malloc(struct rpc_task *task, size_t size)
4cfc7e60
RI		 2329{
2330 struct page *page;
2331 struct rpc_buffer *buf;
2332
2333 BUG_ON(size > PAGE_SIZE - sizeof(struct rpc_buffer));
2334 page = alloc_page(GFP_KERNEL);
2335
2336 if (!page)
2337 return NULL;
2338
2339 buf = page_address(page);
2340 buf->len = PAGE_SIZE;
2341
2342 return buf->data;
2343}
2344
2345/*
2346 * Free the space allocated in the bc_alloc routine
2347 */
5a51f13a 2348static void bc_free(void *buffer)
4cfc7e60
RI		 2349{
2350 struct rpc_buffer *buf;
2351
2352 if (!buffer)
2353 return;
2354
2355 buf = container_of(buffer, struct rpc_buffer, data);
2356 free_page((unsigned long)buf);
2357}
2358
2359/*
2360 * Use the svc_sock to send the callback. Must be called with svsk->sk_mutex
2361 * held. Borrows heavily from svc_tcp_sendto and xs_tcp_send_request.
2362 */
2363static int bc_sendto(struct rpc_rqst *req)
2364{
2365 int len;
2366 struct xdr_buf *xbufp = &req->rq_snd_buf;
2367 struct rpc_xprt *xprt = req->rq_xprt;
2368 struct sock_xprt *transport =
2369 container_of(xprt, struct sock_xprt, xprt);
2370 struct socket *sock = transport->sock;
2371 unsigned long headoff;
2372 unsigned long tailoff;
2373
61677eee 2374 xs_encode_stream_record_marker(xbufp);
4cfc7e60
RI		 2375
2376 tailoff = (unsigned long)xbufp->tail[0].iov_base & ~PAGE_MASK;
2377 headoff = (unsigned long)xbufp->head[0].iov_base & ~PAGE_MASK;
2378 len = svc_send_common(sock, xbufp,
2379 virt_to_page(xbufp->head[0].iov_base), headoff,
2380 xbufp->tail[0].iov_base, tailoff);
2381
2382 if (len != xbufp->len) {
2383 printk(KERN_NOTICE "Error sending entire callback!\n");
2384 len = -EAGAIN;
2385 }
2386
2387 return len;
2388}
2389
2390/*
2391 * The send routine. Borrows from svc_send
2392 */
2393static int bc_send_request(struct rpc_task *task)
2394{
2395 struct rpc_rqst *req = task->tk_rqstp;
2396 struct svc_xprt *xprt;
2397 struct svc_sock *svsk;
2398 u32 len;
2399
2400 dprintk("sending request with xid: %08x\n", ntohl(req->rq_xid));
2401 /*
2402 * Get the server socket associated with this callback xprt
2403 */
2404 xprt = req->rq_xprt->bc_xprt;
2405 svsk = container_of(xprt, struct svc_sock, sk_xprt);
2406
2407 /*
2408 * Grab the mutex to serialize data as the connection is shared
2409 * with the fore channel
2410 */
2411 if (!mutex_trylock(&xprt->xpt_mutex)) {
2412 rpc_sleep_on(&xprt->xpt_bc_pending, task, NULL);
2413 if (!mutex_trylock(&xprt->xpt_mutex))
2414 return -EAGAIN;
2415 rpc_wake_up_queued_task(&xprt->xpt_bc_pending, task);
2416 }
2417 if (test_bit(XPT_DEAD, &xprt->xpt_flags))
2418 len = -ENOTCONN;
2419 else
2420 len = bc_sendto(req);
2421 mutex_unlock(&xprt->xpt_mutex);
2422
2423 if (len > 0)
2424 len = 0;
2425
2426 return len;
2427}
2428
2429/*
2430 * The close routine. Since this is client initiated, we do nothing
2431 */
2432
2433static void bc_close(struct rpc_xprt *xprt)
2434{
4cfc7e60
RI		 2435}
2436
2437/*
2438 * The xprt destroy routine. Again, because this connection is client
2439 * initiated, we do nothing
2440 */
2441
2442static void bc_destroy(struct rpc_xprt *xprt)
2443{
4cfc7e60
RI		 2444}
2445
176e21ee
CL		 2446static struct rpc_xprt_ops xs_local_ops = {
2447 .reserve_xprt = xprt_reserve_xprt,
2448 .release_xprt = xs_tcp_release_xprt,
f39c1bfb 2449 .alloc_slot = xprt_alloc_slot,
176e21ee
CL		 2450 .rpcbind = xs_local_rpcbind,
2451 .set_port = xs_local_set_port,
2452 .connect = xs_connect,
2453 .buf_alloc = rpc_malloc,
2454 .buf_free = rpc_free,
2455 .send_request = xs_local_send_request,
2456 .set_retrans_timeout = xprt_set_retrans_timeout_def,
2457 .close = xs_close,
2458 .destroy = xs_destroy,
2459 .print_stats = xs_local_print_stats,
2460};
2461
262965f5 2462static struct rpc_xprt_ops xs_udp_ops = {
43118c29 2463 .set_buffer_size = xs_udp_set_buffer_size,
12a80469 2464 .reserve_xprt = xprt_reserve_xprt_cong,
49e9a890 2465 .release_xprt = xprt_release_xprt_cong,
f39c1bfb 2466 .alloc_slot = xprt_alloc_slot,
45160d62 2467 .rpcbind = rpcb_getport_async,
92200412 2468 .set_port = xs_set_port,
262965f5 2469 .connect = xs_connect,
02107148
CL		 2470 .buf_alloc = rpc_malloc,
2471 .buf_free = rpc_free,
262965f5 2472 .send_request = xs_udp_send_request,
fe3aca29 2473 .set_retrans_timeout = xprt_set_retrans_timeout_rtt,
46c0ee8b 2474 .timer = xs_udp_timer,
a58dd398 2475 .release_request = xprt_release_rqst_cong,
262965f5
CL		 2476 .close = xs_close,
2477 .destroy = xs_destroy,
262ca07d 2478 .print_stats = xs_udp_print_stats,
262965f5
CL		 2479};
2480
2481static struct rpc_xprt_ops xs_tcp_ops = {
12a80469 2482 .reserve_xprt = xprt_reserve_xprt,
e0ab53de 2483 .release_xprt = xs_tcp_release_xprt,
f39c1bfb 2484 .alloc_slot = xprt_lock_and_alloc_slot,
45160d62 2485 .rpcbind = rpcb_getport_async,
92200412 2486 .set_port = xs_set_port,
0b9e7943 2487 .connect = xs_connect,
02107148
CL		 2488 .buf_alloc = rpc_malloc,
2489 .buf_free = rpc_free,
262965f5 2490 .send_request = xs_tcp_send_request,
fe3aca29 2491 .set_retrans_timeout = xprt_set_retrans_timeout_def,
f75e6745 2492 .close = xs_tcp_close,
9903cd1c 2493 .destroy = xs_destroy,
262ca07d 2494 .print_stats = xs_tcp_print_stats,
a246b010
CL		 2495};
2496
4cfc7e60
RI		 2497/*
2498 * The rpc_xprt_ops for the server backchannel
2499 */
2500
2501static struct rpc_xprt_ops bc_tcp_ops = {
2502 .reserve_xprt = xprt_reserve_xprt,
2503 .release_xprt = xprt_release_xprt,
84e28a30 2504 .alloc_slot = xprt_alloc_slot,
92769108 2505 .rpcbind = xs_local_rpcbind,
4cfc7e60
RI		 2506 .buf_alloc = bc_malloc,
2507 .buf_free = bc_free,
2508 .send_request = bc_send_request,
2509 .set_retrans_timeout = xprt_set_retrans_timeout_def,
2510 .close = bc_close,
2511 .destroy = bc_destroy,
2512 .print_stats = xs_tcp_print_stats,
2513};
2514
92476850
CL		 2515static int xs_init_anyaddr(const int family, struct sockaddr *sap)
2516{
2517 static const struct sockaddr_in sin = {
2518 .sin_family = AF_INET,
2519 .sin_addr.s_addr = htonl(INADDR_ANY),
2520 };
2521 static const struct sockaddr_in6 sin6 = {
2522 .sin6_family = AF_INET6,
2523 .sin6_addr = IN6ADDR_ANY_INIT,
2524 };
2525
2526 switch (family) {
176e21ee
CL		 2527 case AF_LOCAL:
2528 break;
92476850
CL		 2529 case AF_INET:
2530 memcpy(sap, &sin, sizeof(sin));
2531 break;
2532 case AF_INET6:
2533 memcpy(sap, &sin6, sizeof(sin6));
2534 break;
2535 default:
2536 dprintk("RPC: %s: Bad address family\n", __func__);
2537 return -EAFNOSUPPORT;
2538 }
2539 return 0;
2540}
2541
3c341b0b 2542static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
d9ba131d
TM		 2543 unsigned int slot_table_size,
2544 unsigned int max_slot_table_size)
c8541ecd
CL		 2545{
2546 struct rpc_xprt *xprt;
ffc2e518 2547 struct sock_xprt *new;
c8541ecd 2548
96802a09 2549 if (args->addrlen > sizeof(xprt->addr)) {
46121cf7 2550 dprintk("RPC: xs_setup_xprt: address too large\n");
c8541ecd
CL		 2551 return ERR_PTR(-EBADF);
2552 }
2553
d9ba131d
TM		 2554 xprt = xprt_alloc(args->net, sizeof(*new), slot_table_size,
2555 max_slot_table_size);
bd1722d4 2556 if (xprt == NULL) {
46121cf7
CL		 2557 dprintk("RPC: xs_setup_xprt: couldn't allocate "
2558 "rpc_xprt\n");
c8541ecd
CL		 2559 return ERR_PTR(-ENOMEM);
2560 }
c8541ecd 2561
bd1722d4 2562 new = container_of(xprt, struct sock_xprt, xprt);
96802a09
FM		 2563 memcpy(&xprt->addr, args->dstaddr, args->addrlen);
2564 xprt->addrlen = args->addrlen;
d3bc9a1d 2565 if (args->srcaddr)
fbfffbd5 2566 memcpy(&new->srcaddr, args->srcaddr, args->addrlen);
92476850
CL		 2567 else {
2568 int err;
2569 err = xs_init_anyaddr(args->dstaddr->sa_family,
2570 (struct sockaddr *)&new->srcaddr);
2aa13531
SK		 2571 if (err != 0) {
2572 xprt_free(xprt);
92476850 2573 return ERR_PTR(err);
2aa13531 2574 }
92476850 2575 }
c8541ecd
CL		 2576
2577 return xprt;
2578}
2579
176e21ee
CL		 2580static const struct rpc_timeout xs_local_default_timeout = {
2581 .to_initval = 10 * HZ,
2582 .to_maxval = 10 * HZ,
2583 .to_retries = 2,
2584};
2585
2586/**
2587 * xs_setup_local - Set up transport to use an AF_LOCAL socket
2588 * @args: rpc transport creation arguments
2589 *
2590 * AF_LOCAL is a "tpi_cots_ord" transport, just like TCP
2591 */
2592static struct rpc_xprt *xs_setup_local(struct xprt_create *args)
2593{
2594 struct sockaddr_un *sun = (struct sockaddr_un *)args->dstaddr;
2595 struct sock_xprt *transport;
2596 struct rpc_xprt *xprt;
2597 struct rpc_xprt *ret;
2598
d9ba131d
TM		 2599 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2600 xprt_max_tcp_slot_table_entries);
176e21ee
CL		 2601 if (IS_ERR(xprt))
2602 return xprt;
2603 transport = container_of(xprt, struct sock_xprt, xprt);
2604
2605 xprt->prot = 0;
2606 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2607 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2608
2609 xprt->bind_timeout = XS_BIND_TO;
2610 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2611 xprt->idle_timeout = XS_IDLE_DISC_TO;
2612
2613 xprt->ops = &xs_local_ops;
2614 xprt->timeout = &xs_local_default_timeout;
2615
2616 switch (sun->sun_family) {
2617 case AF_LOCAL:
2618 if (sun->sun_path[0] != '/') {
2619 dprintk("RPC: bad AF_LOCAL address: %s\n",
2620 sun->sun_path);
2621 ret = ERR_PTR(-EINVAL);
2622 goto out_err;
2623 }
2624 xprt_set_bound(xprt);
2625 INIT_DELAYED_WORK(&transport->connect_worker,
2626 xs_local_setup_socket);
2627 xs_format_peer_addresses(xprt, "local", RPCBIND_NETID_LOCAL);
2628 break;
2629 default:
2630 ret = ERR_PTR(-EAFNOSUPPORT);
2631 goto out_err;
2632 }
2633
2634 dprintk("RPC: set up xprt to %s via AF_LOCAL\n",
2635 xprt->address_strings[RPC_DISPLAY_ADDR]);
2636
2637 if (try_module_get(THIS_MODULE))
2638 return xprt;
2639 ret = ERR_PTR(-EINVAL);
2640out_err:
2641 xprt_free(xprt);
2642 return ret;
2643}
2644
2881ae74
TM		 2645static const struct rpc_timeout xs_udp_default_timeout = {
2646 .to_initval = 5 * HZ,
2647 .to_maxval = 30 * HZ,
2648 .to_increment = 5 * HZ,
2649 .to_retries = 5,
2650};
2651
9903cd1c
CL		 2652/**
2653 * xs_setup_udp - Set up transport to use a UDP socket
96802a09 2654 * @args: rpc transport creation arguments
9903cd1c
CL		 2655 *
2656 */
483066d6 2657static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
a246b010 2658{
8f9d5b1a 2659 struct sockaddr *addr = args->dstaddr;
c8541ecd 2660 struct rpc_xprt *xprt;
c8475461 2661 struct sock_xprt *transport;
0a68b0be 2662 struct rpc_xprt *ret;
a246b010 2663
d9ba131d
TM		 2664 xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries,
2665 xprt_udp_slot_table_entries);
c8541ecd
CL		 2666 if (IS_ERR(xprt))
2667 return xprt;
c8475461 2668 transport = container_of(xprt, struct sock_xprt, xprt);
a246b010 2669
ec739ef0 2670 xprt->prot = IPPROTO_UDP;
808012fb 2671 xprt->tsh_size = 0;
a246b010
CL		 2672 /* XXX: header size can vary due to auth type, IPv6, etc. */
2673 xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
2674
03bf4b70 2675 xprt->bind_timeout = XS_BIND_TO;
03bf4b70
CL		 2676 xprt->reestablish_timeout = XS_UDP_REEST_TO;
2677 xprt->idle_timeout = XS_IDLE_DISC_TO;
a246b010 2678
262965f5 2679 xprt->ops = &xs_udp_ops;
a246b010 2680
ba7392bb 2681 xprt->timeout = &xs_udp_default_timeout;
a246b010 2682
8f9d5b1a
CL		 2683 switch (addr->sa_family) {
2684 case AF_INET:
2685 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2686 xprt_set_bound(xprt);
2687
2688 INIT_DELAYED_WORK(&transport->connect_worker,
8c14ff2a 2689 xs_udp_setup_socket);
9dc3b095 2690 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
8f9d5b1a
CL		 2691 break;
2692 case AF_INET6:
2693 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2694 xprt_set_bound(xprt);
2695
2696 INIT_DELAYED_WORK(&transport->connect_worker,
8c14ff2a 2697 xs_udp_setup_socket);
9dc3b095 2698 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
8f9d5b1a
CL		 2699 break;
2700 default:
0a68b0be
BF		 2701 ret = ERR_PTR(-EAFNOSUPPORT);
2702 goto out_err;
8f9d5b1a
CL		 2703 }
2704
c740eff8
CL		 2705 if (xprt_bound(xprt))
2706 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2707 xprt->address_strings[RPC_DISPLAY_ADDR],
2708 xprt->address_strings[RPC_DISPLAY_PORT],
2709 xprt->address_strings[RPC_DISPLAY_PROTO]);
2710 else
2711 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
2712 xprt->address_strings[RPC_DISPLAY_ADDR],
2713 xprt->address_strings[RPC_DISPLAY_PROTO]);
edb267a6 2714
bc25571e
TT		 2715 if (try_module_get(THIS_MODULE))
2716 return xprt;
0a68b0be
BF		 2717 ret = ERR_PTR(-EINVAL);
2718out_err:
e204e621 2719 xprt_free(xprt);
0a68b0be 2720 return ret;
a246b010
CL		 2721}
2722
2881ae74
TM		 2723static const struct rpc_timeout xs_tcp_default_timeout = {
2724 .to_initval = 60 * HZ,
2725 .to_maxval = 60 * HZ,
2726 .to_retries = 2,
2727};
2728
9903cd1c
CL		 2729/**
2730 * xs_setup_tcp - Set up transport to use a TCP socket
96802a09 2731 * @args: rpc transport creation arguments
9903cd1c
CL		 2732 *
2733 */
483066d6 2734static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
a246b010 2735{
8f9d5b1a 2736 struct sockaddr *addr = args->dstaddr;
c8541ecd 2737 struct rpc_xprt *xprt;
c8475461 2738 struct sock_xprt *transport;
0a68b0be 2739 struct rpc_xprt *ret;
a246b010 2740
d9ba131d
TM		 2741 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2742 xprt_max_tcp_slot_table_entries);
c8541ecd
CL		 2743 if (IS_ERR(xprt))
2744 return xprt;
c8475461 2745 transport = container_of(xprt, struct sock_xprt, xprt);
a246b010 2746
ec739ef0 2747 xprt->prot = IPPROTO_TCP;
808012fb 2748 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
808012fb 2749 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
a246b010 2750
03bf4b70 2751 xprt->bind_timeout = XS_BIND_TO;
03bf4b70
CL		 2752 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2753 xprt->idle_timeout = XS_IDLE_DISC_TO;
a246b010 2754
262965f5 2755 xprt->ops = &xs_tcp_ops;
ba7392bb 2756 xprt->timeout = &xs_tcp_default_timeout;
a246b010 2757
8f9d5b1a
CL		 2758 switch (addr->sa_family) {
2759 case AF_INET:
2760 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2761 xprt_set_bound(xprt);
2762
9dc3b095 2763 INIT_DELAYED_WORK(&transport->connect_worker,
cdd518d5 2764 xs_tcp_setup_socket);
9dc3b095 2765 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
8f9d5b1a
CL		 2766 break;
2767 case AF_INET6:
2768 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2769 xprt_set_bound(xprt);
2770
9dc3b095 2771 INIT_DELAYED_WORK(&transport->connect_worker,
cdd518d5 2772 xs_tcp_setup_socket);
9dc3b095 2773 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
8f9d5b1a
CL		 2774 break;
2775 default:
0a68b0be
BF		 2776 ret = ERR_PTR(-EAFNOSUPPORT);
2777 goto out_err;
8f9d5b1a
CL		 2778 }
2779
c740eff8
CL		 2780 if (xprt_bound(xprt))
2781 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2782 xprt->address_strings[RPC_DISPLAY_ADDR],
2783 xprt->address_strings[RPC_DISPLAY_PORT],
2784 xprt->address_strings[RPC_DISPLAY_PROTO]);
2785 else
2786 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
2787 xprt->address_strings[RPC_DISPLAY_ADDR],
2788 xprt->address_strings[RPC_DISPLAY_PROTO]);
2789
edb267a6 2790
bc25571e
TT		 2791 if (try_module_get(THIS_MODULE))
2792 return xprt;
0a68b0be
BF		 2793 ret = ERR_PTR(-EINVAL);
2794out_err:
e204e621 2795 xprt_free(xprt);
0a68b0be 2796 return ret;
a246b010 2797}
282b32e1 2798
f300baba
AB		 2799/**
2800 * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket
2801 * @args: rpc transport creation arguments
2802 *
2803 */
2804static struct rpc_xprt *xs_setup_bc_tcp(struct xprt_create *args)
2805{
2806 struct sockaddr *addr = args->dstaddr;
2807 struct rpc_xprt *xprt;
2808 struct sock_xprt *transport;
2809 struct svc_sock *bc_sock;
0a68b0be 2810 struct rpc_xprt *ret;
f300baba 2811
f0418aa4
BF		 2812 if (args->bc_xprt->xpt_bc_xprt) {
2813 /*
2814 * This server connection already has a backchannel
2815 * export; we can't create a new one, as we wouldn't be
2816 * able to match replies based on xid any more. So,
2817 * reuse the already-existing one:
2818 */
2819 return args->bc_xprt->xpt_bc_xprt;
2820 }
d9ba131d
TM		 2821 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2822 xprt_tcp_slot_table_entries);
f300baba
AB		 2823 if (IS_ERR(xprt))
2824 return xprt;
2825 transport = container_of(xprt, struct sock_xprt, xprt);
2826
2827 xprt->prot = IPPROTO_TCP;
2828 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2829 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2830 xprt->timeout = &xs_tcp_default_timeout;
2831
2832 /* backchannel */
2833 xprt_set_bound(xprt);
2834 xprt->bind_timeout = 0;
f300baba
AB		 2835 xprt->reestablish_timeout = 0;
2836 xprt->idle_timeout = 0;
2837
f300baba
AB		 2838 xprt->ops = &bc_tcp_ops;
2839
2840 switch (addr->sa_family) {
2841 case AF_INET:
2842 xs_format_peer_addresses(xprt, "tcp",
2843 RPCBIND_NETID_TCP);
2844 break;
2845 case AF_INET6:
2846 xs_format_peer_addresses(xprt, "tcp",
2847 RPCBIND_NETID_TCP6);
2848 break;
2849 default:
0a68b0be
BF		 2850 ret = ERR_PTR(-EAFNOSUPPORT);
2851 goto out_err;
f300baba
AB		 2852 }
2853
50fa0d40
PE		 2854 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2855 xprt->address_strings[RPC_DISPLAY_ADDR],
2856 xprt->address_strings[RPC_DISPLAY_PORT],
2857 xprt->address_strings[RPC_DISPLAY_PROTO]);
f300baba 2858
99de8ea9
BF		 2859 /*
2860 * Once we've associated a backchannel xprt with a connection,
2861 * we want to keep it around as long as long as the connection
2862 * lasts, in case we need to start using it for a backchannel
2863 * again; this reference won't be dropped until bc_xprt is
2864 * destroyed.
2865 */
2866 xprt_get(xprt);
2867 args->bc_xprt->xpt_bc_xprt = xprt;
2868 xprt->bc_xprt = args->bc_xprt;
2869 bc_sock = container_of(args->bc_xprt, struct svc_sock, sk_xprt);
2870 transport->sock = bc_sock->sk_sock;
2871 transport->inet = bc_sock->sk_sk;
2872
f300baba
AB		 2873 /*
2874 * Since we don't want connections for the backchannel, we set
2875 * the xprt status to connected
2876 */
2877 xprt_set_connected(xprt);
2878
2879
2880 if (try_module_get(THIS_MODULE))
2881 return xprt;
99de8ea9 2882 xprt_put(xprt);
0a68b0be
BF		 2883 ret = ERR_PTR(-EINVAL);
2884out_err:
e204e621 2885 xprt_free(xprt);
0a68b0be 2886 return ret;
f300baba
AB		 2887}
2888
176e21ee
CL		 2889static struct xprt_class xs_local_transport = {
2890 .list = LIST_HEAD_INIT(xs_local_transport.list),
2891 .name = "named UNIX socket",
2892 .owner = THIS_MODULE,
2893 .ident = XPRT_TRANSPORT_LOCAL,
2894 .setup = xs_setup_local,
2895};
2896
bc25571e
TT		 2897static struct xprt_class xs_udp_transport = {
2898 .list = LIST_HEAD_INIT(xs_udp_transport.list),
2899 .name = "udp",
2900 .owner = THIS_MODULE,
f300baba 2901 .ident = XPRT_TRANSPORT_UDP,
bc25571e
TT		 2902 .setup = xs_setup_udp,
2903};
2904
2905static struct xprt_class xs_tcp_transport = {
2906 .list = LIST_HEAD_INIT(xs_tcp_transport.list),
2907 .name = "tcp",
2908 .owner = THIS_MODULE,
f300baba 2909 .ident = XPRT_TRANSPORT_TCP,
bc25571e
TT		 2910 .setup = xs_setup_tcp,
2911};
2912
f300baba
AB		 2913static struct xprt_class xs_bc_tcp_transport = {
2914 .list = LIST_HEAD_INIT(xs_bc_tcp_transport.list),
2915 .name = "tcp NFSv4.1 backchannel",
2916 .owner = THIS_MODULE,
2917 .ident = XPRT_TRANSPORT_BC_TCP,
2918 .setup = xs_setup_bc_tcp,
2919};
2920
282b32e1 2921/**
bc25571e 2922 * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
282b32e1
CL		 2923 *
2924 */
2925int init_socket_xprt(void)
2926{
fbf76683 2927#ifdef RPC_DEBUG
2b1bec5f 2928 if (!sunrpc_table_header)
0b4d4147 2929 sunrpc_table_header = register_sysctl_table(sunrpc_table);
fbf76683
CL		 2930#endif
2931
176e21ee 2932 xprt_register_transport(&xs_local_transport);
bc25571e
TT		 2933 xprt_register_transport(&xs_udp_transport);
2934 xprt_register_transport(&xs_tcp_transport);
f300baba 2935 xprt_register_transport(&xs_bc_tcp_transport);
bc25571e 2936
282b32e1
CL		 2937 return 0;
2938}
2939
2940/**
bc25571e 2941 * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
282b32e1
CL		 2942 *
2943 */
2944void cleanup_socket_xprt(void)
2945{
fbf76683
CL		 2946#ifdef RPC_DEBUG
2947 if (sunrpc_table_header) {
2948 unregister_sysctl_table(sunrpc_table_header);
2949 sunrpc_table_header = NULL;
2950 }
2951#endif
bc25571e 2952
176e21ee 2953 xprt_unregister_transport(&xs_local_transport);
bc25571e
TT		 2954 xprt_unregister_transport(&xs_udp_transport);
2955 xprt_unregister_transport(&xs_tcp_transport);
f300baba 2956 xprt_unregister_transport(&xs_bc_tcp_transport);
282b32e1 2957}
cbf11071 2958
9bbb9e5a
RR		 2959static int param_set_uint_minmax(const char *val,
2960 const struct kernel_param *kp,
cbf11071
TM		 2961 unsigned int min, unsigned int max)
2962{
2963 unsigned long num;
2964 int ret;
2965
2966 if (!val)
2967 return -EINVAL;
2968 ret = strict_strtoul(val, 0, &num);
2969 if (ret == -EINVAL || num < min || num > max)
2970 return -EINVAL;
2971 *((unsigned int *)kp->arg) = num;
2972 return 0;
2973}
2974
9bbb9e5a 2975static int param_set_portnr(const char *val, const struct kernel_param *kp)
cbf11071
TM		 2976{
2977 return param_set_uint_minmax(val, kp,
2978 RPC_MIN_RESVPORT,
2979 RPC_MAX_RESVPORT);
2980}
2981
9bbb9e5a
RR		 2982static struct kernel_param_ops param_ops_portnr = {
2983 .set = param_set_portnr,
2984 .get = param_get_uint,
2985};
2986
cbf11071
TM		 2987#define param_check_portnr(name, p) \
2988 __param_check(name, p, unsigned int);
2989
2990module_param_named(min_resvport, xprt_min_resvport, portnr, 0644);
2991module_param_named(max_resvport, xprt_max_resvport, portnr, 0644);
2992
9bbb9e5a
RR		 2993static int param_set_slot_table_size(const char *val,
2994 const struct kernel_param *kp)
cbf11071
TM		 2995{
2996 return param_set_uint_minmax(val, kp,
2997 RPC_MIN_SLOT_TABLE,
2998 RPC_MAX_SLOT_TABLE);
2999}
3000
9bbb9e5a
RR		 3001static struct kernel_param_ops param_ops_slot_table_size = {
3002 .set = param_set_slot_table_size,
3003 .get = param_get_uint,
3004};
3005
cbf11071
TM		 3006#define param_check_slot_table_size(name, p) \
3007 __param_check(name, p, unsigned int);
3008
d9ba131d
TM		 3009static int param_set_max_slot_table_size(const char *val,
3010 const struct kernel_param *kp)
3011{
3012 return param_set_uint_minmax(val, kp,
3013 RPC_MIN_SLOT_TABLE,
3014 RPC_MAX_SLOT_TABLE_LIMIT);
3015}
3016
3017static struct kernel_param_ops param_ops_max_slot_table_size = {
3018 .set = param_set_max_slot_table_size,
3019 .get = param_get_uint,
3020};
3021
3022#define param_check_max_slot_table_size(name, p) \
3023 __param_check(name, p, unsigned int);
3024
cbf11071
TM		 3025module_param_named(tcp_slot_table_entries, xprt_tcp_slot_table_entries,
3026 slot_table_size, 0644);
d9ba131d
TM		 3027module_param_named(tcp_max_slot_table_entries, xprt_max_tcp_slot_table_entries,
3028 max_slot_table_size, 0644);
cbf11071
TM		 3029module_param_named(udp_slot_table_entries, xprt_udp_slot_table_entries,
3030 slot_table_size, 0644);
3031
Linux 6.x block layer and io_uring tree(s)