]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blame - net/sunrpc/xprtsock.c
projects / mirror_ubuntu-artful-kernel.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
SUNRPC: Add correct rcu_dereference annotation in rpc_clnt_set_transport
[mirror_ubuntu-artful-kernel.git] / net / sunrpc / xprtsock.c
CommitLineData
a246b010
CL		 1/*
2 * linux/net/sunrpc/xprtsock.c
3 *
4 * Client-side transport implementation for sockets.
5 *
113aa838
AC		 6 * TCP callback races fixes (C) 1998 Red Hat
7 * TCP send fixes (C) 1998 Red Hat
a246b010
CL		 8 * TCP NFS related read + write fixes
9 * (C) 1999 Dave Airlie, University of Limerick, Ireland <airlied@linux.ie>
10 *
11 * Rewrite of larges part of the code in order to stabilize TCP stuff.
12 * Fix behaviour when socket buffer is full.
13 * (C) 1999 Trond Myklebust <trond.myklebust@fys.uio.no>
55aa4f58
CL		 14 *
15 * IP socket transport implementation, (C) 2005 Chuck Lever <cel@netapp.com>
8f9d5b1a
CL		 16 *
17 * IPv6 support contributed by Gilles Quillard, Bull Open Source, 2005.
18 * <gilles.quillard@bull.net>
a246b010
CL		 19 */
20
21#include <linux/types.h>
176e21ee 22#include <linux/string.h>
a246b010 23#include <linux/slab.h>
bc25571e 24#include <linux/module.h>
a246b010 25#include <linux/capability.h>
a246b010
CL		 26#include <linux/pagemap.h>
27#include <linux/errno.h>
28#include <linux/socket.h>
29#include <linux/in.h>
30#include <linux/net.h>
31#include <linux/mm.h>
176e21ee 32#include <linux/un.h>
a246b010
CL		 33#include <linux/udp.h>
34#include <linux/tcp.h>
35#include <linux/sunrpc/clnt.h>
5976687a 36#include <linux/sunrpc/addr.h>
02107148 37#include <linux/sunrpc/sched.h>
4cfc7e60 38#include <linux/sunrpc/svcsock.h>
49c36fcc 39#include <linux/sunrpc/xprtsock.h>
a246b010 40#include <linux/file.h>
9e00abc3 41#ifdef CONFIG_SUNRPC_BACKCHANNEL
44b98efd
RL		 42#include <linux/sunrpc/bc_xprt.h>
43#endif
a246b010
CL		 44
45#include <net/sock.h>
46#include <net/checksum.h>
47#include <net/udp.h>
48#include <net/tcp.h>
49
40b5ea0c
TM		 50#include <trace/events/sunrpc.h>
51
4cfc7e60 52#include "sunrpc.h"
176e21ee
CL		 53
54static void xs_close(struct rpc_xprt *xprt);
55
c556b754
CL		 56/*
57 * xprtsock tunables
58 */
09acfea5
TM		 59static unsigned int xprt_udp_slot_table_entries = RPC_DEF_SLOT_TABLE;
60static unsigned int xprt_tcp_slot_table_entries = RPC_MIN_SLOT_TABLE;
61static unsigned int xprt_max_tcp_slot_table_entries = RPC_MAX_SLOT_TABLE;
c556b754 62
09acfea5
TM		 63static unsigned int xprt_min_resvport = RPC_DEF_MIN_RESVPORT;
64static unsigned int xprt_max_resvport = RPC_DEF_MAX_RESVPORT;
c556b754 65
7d1e8255 66#define XS_TCP_LINGER_TO (15U * HZ)
25fe6142 67static unsigned int xs_tcp_fin_timeout __read_mostly = XS_TCP_LINGER_TO;
7d1e8255 68
fbf76683
CL		 69/*
70 * We can register our own files under /proc/sys/sunrpc by
71 * calling register_sysctl_table() again. The files in that
72 * directory become the union of all files registered there.
73 *
74 * We simply need to make sure that we don't collide with
75 * someone else's file names!
76 */
77
78#ifdef RPC_DEBUG
79
80static unsigned int min_slot_table_size = RPC_MIN_SLOT_TABLE;
81static unsigned int max_slot_table_size = RPC_MAX_SLOT_TABLE;
d9ba131d 82static unsigned int max_tcp_slot_table_limit = RPC_MAX_SLOT_TABLE_LIMIT;
fbf76683
CL		 83static unsigned int xprt_min_resvport_limit = RPC_MIN_RESVPORT;
84static unsigned int xprt_max_resvport_limit = RPC_MAX_RESVPORT;
85
86static struct ctl_table_header *sunrpc_table_header;
87
88/*
89 * FIXME: changing the UDP slot table size should also resize the UDP
90 * socket buffers for existing UDP transports
91 */
fe2c6338 92static struct ctl_table xs_tunables_table[] = {
fbf76683 93 {
fbf76683
CL		 94 .procname = "udp_slot_table_entries",
95 .data = &xprt_udp_slot_table_entries,
96 .maxlen = sizeof(unsigned int),
97 .mode = 0644,
6d456111 98 .proc_handler = proc_dointvec_minmax,
fbf76683
CL		 99 .extra1 = &min_slot_table_size,
100 .extra2 = &max_slot_table_size
101 },
102 {
fbf76683
CL		 103 .procname = "tcp_slot_table_entries",
104 .data = &xprt_tcp_slot_table_entries,
105 .maxlen = sizeof(unsigned int),
106 .mode = 0644,
6d456111 107 .proc_handler = proc_dointvec_minmax,
fbf76683
CL		 108 .extra1 = &min_slot_table_size,
109 .extra2 = &max_slot_table_size
110 },
d9ba131d
TM		 111 {
112 .procname = "tcp_max_slot_table_entries",
113 .data = &xprt_max_tcp_slot_table_entries,
114 .maxlen = sizeof(unsigned int),
115 .mode = 0644,
116 .proc_handler = proc_dointvec_minmax,
117 .extra1 = &min_slot_table_size,
118 .extra2 = &max_tcp_slot_table_limit
119 },
fbf76683 120 {
fbf76683
CL		 121 .procname = "min_resvport",
122 .data = &xprt_min_resvport,
123 .maxlen = sizeof(unsigned int),
124 .mode = 0644,
6d456111 125 .proc_handler = proc_dointvec_minmax,
fbf76683
CL		 126 .extra1 = &xprt_min_resvport_limit,
127 .extra2 = &xprt_max_resvport_limit
128 },
129 {
fbf76683
CL		 130 .procname = "max_resvport",
131 .data = &xprt_max_resvport,
132 .maxlen = sizeof(unsigned int),
133 .mode = 0644,
6d456111 134 .proc_handler = proc_dointvec_minmax,
fbf76683
CL		 135 .extra1 = &xprt_min_resvport_limit,
136 .extra2 = &xprt_max_resvport_limit
137 },
25fe6142
TM		 138 {
139 .procname = "tcp_fin_timeout",
140 .data = &xs_tcp_fin_timeout,
141 .maxlen = sizeof(xs_tcp_fin_timeout),
142 .mode = 0644,
6d456111 143 .proc_handler = proc_dointvec_jiffies,
fbf76683 144 },
f8572d8f 145 { },
fbf76683
CL		 146};
147
fe2c6338 148static struct ctl_table sunrpc_table[] = {
fbf76683 149 {
fbf76683
CL		 150 .procname = "sunrpc",
151 .mode = 0555,
152 .child = xs_tunables_table
153 },
f8572d8f 154 { },
fbf76683
CL		 155};
156
157#endif
158
03bf4b70
CL		 159/*
160 * Wait duration for a reply from the RPC portmapper.
161 */
162#define XS_BIND_TO (60U * HZ)
163
164/*
165 * Delay if a UDP socket connect error occurs. This is most likely some
166 * kind of resource problem on the local host.
167 */
168#define XS_UDP_REEST_TO (2U * HZ)
169
170/*
171 * The reestablish timeout allows clients to delay for a bit before attempting
172 * to reconnect to a server that just dropped our connection.
173 *
174 * We implement an exponential backoff when trying to reestablish a TCP
175 * transport connection with the server. Some servers like to drop a TCP
176 * connection when they are overworked, so we start with a short timeout and
177 * increase over time if the server is down or not responding.
178 */
179#define XS_TCP_INIT_REEST_TO (3U * HZ)
180#define XS_TCP_MAX_REEST_TO (5U * 60 * HZ)
181
182/*
183 * TCP idle timeout; client drops the transport socket if it is idle
184 * for this long. Note that we also timeout UDP sockets to prevent
185 * holding port numbers when there is no RPC traffic.
186 */
187#define XS_IDLE_DISC_TO (5U * 60 * HZ)
188
a246b010
CL		 189#ifdef RPC_DEBUG
190# undef RPC_DEBUG_DATA
9903cd1c 191# define RPCDBG_FACILITY RPCDBG_TRANS
a246b010
CL		 192#endif
193
a246b010 194#ifdef RPC_DEBUG_DATA
9903cd1c 195static void xs_pktdump(char *msg, u32 *packet, unsigned int count)
a246b010 196{
9903cd1c
CL		 197 u8 *buf = (u8 *) packet;
198 int j;
a246b010 199
46121cf7 200 dprintk("RPC: %s\n", msg);
a246b010
CL		 201 for (j = 0; j < count && j < 128; j += 4) {
202 if (!(j & 31)) {
203 if (j)
204 dprintk("\n");
205 dprintk("0x%04x ", j);
206 }
207 dprintk("%02x%02x%02x%02x ",
208 buf[j], buf[j+1], buf[j+2], buf[j+3]);
209 }
210 dprintk("\n");
211}
212#else
9903cd1c 213static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count)
a246b010
CL		 214{
215 /* NOP */
216}
217#endif
218
ffc2e518
CL		 219struct sock_xprt {
220 struct rpc_xprt xprt;
ee0ac0c2
CL		 221
222 /*
223 * Network layer
224 */
225 struct socket * sock;
226 struct sock * inet;
51971139
CL		 227
228 /*
229 * State of TCP reply receive
230 */
231 __be32 tcp_fraghdr,
b76ce561
TM		 232 tcp_xid,
233 tcp_calldir;
51971139
CL		 234
235 u32 tcp_offset,
236 tcp_reclen;
237
238 unsigned long tcp_copied,
239 tcp_flags;
c8475461
CL		 240
241 /*
242 * Connection of transports
243 */
34161db6 244 struct delayed_work connect_worker;
fbfffbd5
CL		 245 struct sockaddr_storage srcaddr;
246 unsigned short srcport;
7c6e066e
CL		 247
248 /*
249 * UDP socket buffer size parameters
250 */
251 size_t rcvsize,
252 sndsize;
314dfd79
CL		 253
254 /*
255 * Saved socket callback addresses
256 */
257 void (*old_data_ready)(struct sock *, int);
258 void (*old_state_change)(struct sock *);
259 void (*old_write_space)(struct sock *);
ffc2e518
CL		 260};
261
e136d092
CL		 262/*
263 * TCP receive state flags
264 */
265#define TCP_RCV_LAST_FRAG (1UL << 0)
266#define TCP_RCV_COPY_FRAGHDR (1UL << 1)
267#define TCP_RCV_COPY_XID (1UL << 2)
268#define TCP_RCV_COPY_DATA (1UL << 3)
f4a2e418
RL		 269#define TCP_RCV_READ_CALLDIR (1UL << 4)
270#define TCP_RCV_COPY_CALLDIR (1UL << 5)
18dca02a
RL		 271
272/*
273 * TCP RPC flags
274 */
f4a2e418 275#define TCP_RPC_REPLY (1UL << 6)
e136d092 276
95392c59
CL		 277static inline struct sockaddr *xs_addr(struct rpc_xprt *xprt)
278{
279 return (struct sockaddr *) &xprt->addr;
280}
281
176e21ee
CL		 282static inline struct sockaddr_un *xs_addr_un(struct rpc_xprt *xprt)
283{
284 return (struct sockaddr_un *) &xprt->addr;
285}
286
95392c59 287static inline struct sockaddr_in *xs_addr_in(struct rpc_xprt *xprt)
edb267a6 288{
95392c59
CL		 289 return (struct sockaddr_in *) &xprt->addr;
290}
291
292static inline struct sockaddr_in6 *xs_addr_in6(struct rpc_xprt *xprt)
293{
294 return (struct sockaddr_in6 *) &xprt->addr;
295}
296
c877b849 297static void xs_format_common_peer_addresses(struct rpc_xprt *xprt)
edb267a6 298{
c877b849 299 struct sockaddr *sap = xs_addr(xprt);
9dc3b095
CL		 300 struct sockaddr_in6 *sin6;
301 struct sockaddr_in *sin;
176e21ee 302 struct sockaddr_un *sun;
c877b849 303 char buf[128];
edb267a6 304
9dc3b095 305 switch (sap->sa_family) {
176e21ee
CL		 306 case AF_LOCAL:
307 sun = xs_addr_un(xprt);
308 strlcpy(buf, sun->sun_path, sizeof(buf));
309 xprt->address_strings[RPC_DISPLAY_ADDR] =
310 kstrdup(buf, GFP_KERNEL);
311 break;
9dc3b095 312 case AF_INET:
176e21ee
CL		 313 (void)rpc_ntop(sap, buf, sizeof(buf));
314 xprt->address_strings[RPC_DISPLAY_ADDR] =
315 kstrdup(buf, GFP_KERNEL);
9dc3b095 316 sin = xs_addr_in(xprt);
fc0b5791 317 snprintf(buf, sizeof(buf), "%08x", ntohl(sin->sin_addr.s_addr));
9dc3b095
CL		 318 break;
319 case AF_INET6:
176e21ee
CL		 320 (void)rpc_ntop(sap, buf, sizeof(buf));
321 xprt->address_strings[RPC_DISPLAY_ADDR] =
322 kstrdup(buf, GFP_KERNEL);
9dc3b095 323 sin6 = xs_addr_in6(xprt);
fc0b5791 324 snprintf(buf, sizeof(buf), "%pi6", &sin6->sin6_addr);
9dc3b095
CL		 325 break;
326 default:
327 BUG();
756805e7 328 }
176e21ee 329
9dc3b095 330 xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = kstrdup(buf, GFP_KERNEL);
edb267a6
CL		 331}
332
9dc3b095 333static void xs_format_common_peer_ports(struct rpc_xprt *xprt)
4b6473fb 334{
9dc3b095
CL		 335 struct sockaddr *sap = xs_addr(xprt);
336 char buf[128];
4b6473fb 337
81160e66 338 snprintf(buf, sizeof(buf), "%u", rpc_get_port(sap));
c877b849 339 xprt->address_strings[RPC_DISPLAY_PORT] = kstrdup(buf, GFP_KERNEL);
4b6473fb 340
81160e66 341 snprintf(buf, sizeof(buf), "%4hx", rpc_get_port(sap));
c877b849
CL		 342 xprt->address_strings[RPC_DISPLAY_HEX_PORT] = kstrdup(buf, GFP_KERNEL);
343}
4b6473fb 344
9dc3b095
CL		 345static void xs_format_peer_addresses(struct rpc_xprt *xprt,
346 const char *protocol,
347 const char *netid)
c877b849 348{
b454ae90 349 xprt->address_strings[RPC_DISPLAY_PROTO] = protocol;
b454ae90 350 xprt->address_strings[RPC_DISPLAY_NETID] = netid;
c877b849 351 xs_format_common_peer_addresses(xprt);
9dc3b095 352 xs_format_common_peer_ports(xprt);
edb267a6 353}
4b6473fb 354
9dc3b095 355static void xs_update_peer_port(struct rpc_xprt *xprt)
4b6473fb 356{
9dc3b095
CL		 357 kfree(xprt->address_strings[RPC_DISPLAY_HEX_PORT]);
358 kfree(xprt->address_strings[RPC_DISPLAY_PORT]);
4417c8c4 359
9dc3b095 360 xs_format_common_peer_ports(xprt);
edb267a6
CL		 361}
362
363static void xs_free_peer_addresses(struct rpc_xprt *xprt)
364{
33e01dc7
CL		 365 unsigned int i;
366
367 for (i = 0; i < RPC_DISPLAY_MAX; i++)
368 switch (i) {
369 case RPC_DISPLAY_PROTO:
370 case RPC_DISPLAY_NETID:
371 continue;
372 default:
373 kfree(xprt->address_strings[i]);
374 }
edb267a6
CL		 375}
376
b4b5cc85
CL		 377#define XS_SENDMSG_FLAGS (MSG_DONTWAIT | MSG_NOSIGNAL)
378
24c5684b 379static int xs_send_kvec(struct socket *sock, struct sockaddr *addr, int addrlen, struct kvec *vec, unsigned int base, int more)
b4b5cc85 380{
b4b5cc85
CL		 381 struct msghdr msg = {
382 .msg_name = addr,
383 .msg_namelen = addrlen,
24c5684b
TM		 384 .msg_flags = XS_SENDMSG_FLAGS | (more ? MSG_MORE : 0),
385 };
386 struct kvec iov = {
387 .iov_base = vec->iov_base + base,
388 .iov_len = vec->iov_len - base,
b4b5cc85
CL		 389 };
390
24c5684b 391 if (iov.iov_len != 0)
b4b5cc85
CL		 392 return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
393 return kernel_sendmsg(sock, &msg, NULL, 0, 0);
394}
395
24c5684b 396static int xs_send_pagedata(struct socket *sock, struct xdr_buf *xdr, unsigned int base, int more)
b4b5cc85 397{
24c5684b
TM		 398 struct page **ppage;
399 unsigned int remainder;
400 int err, sent = 0;
401
402 remainder = xdr->page_len - base;
403 base += xdr->page_base;
404 ppage = xdr->pages + (base >> PAGE_SHIFT);
405 base &= ~PAGE_MASK;
406 for(;;) {
407 unsigned int len = min_t(unsigned int, PAGE_SIZE - base, remainder);
408 int flags = XS_SENDMSG_FLAGS;
b4b5cc85 409
24c5684b
TM		 410 remainder -= len;
411 if (remainder != 0 || more)
412 flags |= MSG_MORE;
413 err = sock->ops->sendpage(sock, *ppage, base, len, flags);
414 if (remainder == 0 || err != len)
415 break;
416 sent += err;
417 ppage++;
418 base = 0;
419 }
420 if (sent == 0)
421 return err;
422 if (err > 0)
423 sent += err;
424 return sent;
b4b5cc85
CL		 425}
426
9903cd1c
CL		 427/**
428 * xs_sendpages - write pages directly to a socket
429 * @sock: socket to send on
430 * @addr: UDP only -- address of destination
431 * @addrlen: UDP only -- length of destination address
432 * @xdr: buffer containing this request
433 * @base: starting position in the buffer
434 *
a246b010 435 */
24c5684b 436static int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base)
a246b010 437{
24c5684b
TM		 438 unsigned int remainder = xdr->len - base;
439 int err, sent = 0;
a246b010 440
262965f5 441 if (unlikely(!sock))
fba91afb 442 return -ENOTSOCK;
262965f5
CL		 443
444 clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags);
24c5684b
TM		 445 if (base != 0) {
446 addr = NULL;
447 addrlen = 0;
448 }
262965f5 449
24c5684b
TM		 450 if (base < xdr->head[0].iov_len || addr != NULL) {
451 unsigned int len = xdr->head[0].iov_len - base;
452 remainder -= len;
453 err = xs_send_kvec(sock, addr, addrlen, &xdr->head[0], base, remainder != 0);
454 if (remainder == 0 || err != len)
a246b010 455 goto out;
24c5684b 456 sent += err;
a246b010
CL		 457 base = 0;
458 } else
24c5684b 459 base -= xdr->head[0].iov_len;
a246b010 460
24c5684b
TM		 461 if (base < xdr->page_len) {
462 unsigned int len = xdr->page_len - base;
463 remainder -= len;
464 err = xs_send_pagedata(sock, xdr, base, remainder != 0);
465 if (remainder == 0 || err != len)
a246b010 466 goto out;
24c5684b 467 sent += err;
a246b010 468 base = 0;
24c5684b
TM		 469 } else
470 base -= xdr->page_len;
471
472 if (base >= xdr->tail[0].iov_len)
473 return sent;
474 err = xs_send_kvec(sock, NULL, 0, &xdr->tail[0], base, 0);
a246b010 475out:
24c5684b
TM		 476 if (sent == 0)
477 return err;
478 if (err > 0)
479 sent += err;
480 return sent;
a246b010
CL		 481}
482
b6ddf64f
TM		 483static void xs_nospace_callback(struct rpc_task *task)
484{
485 struct sock_xprt *transport = container_of(task->tk_rqstp->rq_xprt, struct sock_xprt, xprt);
486
487 transport->inet->sk_write_pending--;
488 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
489}
490
9903cd1c 491/**
262965f5
CL		 492 * xs_nospace - place task on wait queue if transmit was incomplete
493 * @task: task to put to sleep
9903cd1c 494 *
a246b010 495 */
5e3771ce 496static int xs_nospace(struct rpc_task *task)
a246b010 497{
262965f5
CL		 498 struct rpc_rqst *req = task->tk_rqstp;
499 struct rpc_xprt *xprt = req->rq_xprt;
ee0ac0c2 500 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
24ca9a84 501 int ret = -EAGAIN;
a246b010 502
46121cf7 503 dprintk("RPC: %5u xmit incomplete (%u left of %u)\n",
262965f5
CL		 504 task->tk_pid, req->rq_slen - req->rq_bytes_sent,
505 req->rq_slen);
506
b6ddf64f
TM		 507 /* Protect against races with write_space */
508 spin_lock_bh(&xprt->transport_lock);
509
510 /* Don't race with disconnect */
511 if (xprt_connected(xprt)) {
512 if (test_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags)) {
513 /*
514 * Notify TCP that we're limited by the application
515 * window size
516 */
517 set_bit(SOCK_NOSPACE, &transport->sock->flags);
518 transport->inet->sk_write_pending++;
519 /* ...and wait for more buffer space */
520 xprt_wait_for_buffer_space(task, xs_nospace_callback);
521 }
522 } else {
523 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
5e3771ce 524 ret = -ENOTCONN;
b6ddf64f 525 }
262965f5 526
b6ddf64f 527 spin_unlock_bh(&xprt->transport_lock);
5e3771ce 528 return ret;
262965f5
CL		 529}
530
61677eee
CL		 531/*
532 * Construct a stream transport record marker in @buf.
533 */
534static inline void xs_encode_stream_record_marker(struct xdr_buf *buf)
535{
536 u32 reclen = buf->len - sizeof(rpc_fraghdr);
537 rpc_fraghdr *base = buf->head[0].iov_base;
538 *base = cpu_to_be32(RPC_LAST_STREAM_FRAGMENT | reclen);
539}
540
176e21ee
CL		 541/**
542 * xs_local_send_request - write an RPC request to an AF_LOCAL socket
543 * @task: RPC task that manages the state of an RPC request
544 *
545 * Return values:
546 * 0: The request has been sent
547 * EAGAIN: The socket was blocked, please call again later to
548 * complete the request
549 * ENOTCONN: Caller needs to invoke connect logic then call again
550 * other: Some other error occured, the request was not sent
551 */
552static int xs_local_send_request(struct rpc_task *task)
553{
554 struct rpc_rqst *req = task->tk_rqstp;
555 struct rpc_xprt *xprt = req->rq_xprt;
556 struct sock_xprt *transport =
557 container_of(xprt, struct sock_xprt, xprt);
558 struct xdr_buf *xdr = &req->rq_snd_buf;
559 int status;
560
561 xs_encode_stream_record_marker(&req->rq_snd_buf);
562
563 xs_pktdump("packet data:",
564 req->rq_svec->iov_base, req->rq_svec->iov_len);
565
566 status = xs_sendpages(transport->sock, NULL, 0,
567 xdr, req->rq_bytes_sent);
568 dprintk("RPC: %s(%u) = %d\n",
569 __func__, xdr->len - req->rq_bytes_sent, status);
570 if (likely(status >= 0)) {
571 req->rq_bytes_sent += status;
572 req->rq_xmit_bytes_sent += status;
573 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
574 req->rq_bytes_sent = 0;
575 return 0;
576 }
577 status = -EAGAIN;
578 }
579
580 switch (status) {
581 case -EAGAIN:
582 status = xs_nospace(task);
583 break;
584 default:
585 dprintk("RPC: sendmsg returned unrecognized error %d\n",
586 -status);
587 case -EPIPE:
588 xs_close(xprt);
589 status = -ENOTCONN;
590 }
591
592 return status;
593}
594
262965f5
CL		 595/**
596 * xs_udp_send_request - write an RPC request to a UDP socket
597 * @task: address of RPC task that manages the state of an RPC request
598 *
599 * Return values:
600 * 0: The request has been sent
601 * EAGAIN: The socket was blocked, please call again later to
602 * complete the request
603 * ENOTCONN: Caller needs to invoke connect logic then call again
25985edc 604 * other: Some other error occurred, the request was not sent
262965f5
CL		 605 */
606static int xs_udp_send_request(struct rpc_task *task)
607{
608 struct rpc_rqst *req = task->tk_rqstp;
609 struct rpc_xprt *xprt = req->rq_xprt;
ee0ac0c2 610 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
262965f5
CL		 611 struct xdr_buf *xdr = &req->rq_snd_buf;
612 int status;
a246b010 613
9903cd1c 614 xs_pktdump("packet data:",
a246b010
CL		 615 req->rq_svec->iov_base,
616 req->rq_svec->iov_len);
617
01d37c42
TM		 618 if (!xprt_bound(xprt))
619 return -ENOTCONN;
ee0ac0c2 620 status = xs_sendpages(transport->sock,
95392c59 621 xs_addr(xprt),
ee0ac0c2
CL		 622 xprt->addrlen, xdr,
623 req->rq_bytes_sent);
a246b010 624
46121cf7 625 dprintk("RPC: xs_udp_send_request(%u) = %d\n",
262965f5 626 xdr->len - req->rq_bytes_sent, status);
a246b010 627
2199700f 628 if (status >= 0) {
d60dbb20 629 req->rq_xmit_bytes_sent += status;
2199700f
TM		 630 if (status >= req->rq_slen)
631 return 0;
632 /* Still some bytes left; set up for a retry later. */
262965f5 633 status = -EAGAIN;
2199700f 634 }
a246b010 635
262965f5 636 switch (status) {
fba91afb
TM		 637 case -ENOTSOCK:
638 status = -ENOTCONN;
639 /* Should we call xs_close() here? */
640 break;
b6ddf64f 641 case -EAGAIN:
5e3771ce 642 status = xs_nospace(task);
b6ddf64f 643 break;
c8485e4d
TM		 644 default:
645 dprintk("RPC: sendmsg returned unrecognized error %d\n",
646 -status);
262965f5
CL		 647 case -ENETUNREACH:
648 case -EPIPE:
a246b010
CL		 649 case -ECONNREFUSED:
650 /* When the server has died, an ICMP port unreachable message
9903cd1c 651 * prompts ECONNREFUSED. */
b6ddf64f 652 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
a246b010 653 }
5fe46e9d 654
262965f5 655 return status;
a246b010
CL		 656}
657
e06799f9
TM		 658/**
659 * xs_tcp_shutdown - gracefully shut down a TCP socket
660 * @xprt: transport
661 *
662 * Initiates a graceful shutdown of the TCP socket by calling the
663 * equivalent of shutdown(SHUT_WR);
664 */
665static void xs_tcp_shutdown(struct rpc_xprt *xprt)
666{
667 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
668 struct socket *sock = transport->sock;
669
40b5ea0c 670 if (sock != NULL) {
e06799f9 671 kernel_sock_shutdown(sock, SHUT_WR);
40b5ea0c
TM		 672 trace_rpc_socket_shutdown(xprt, sock);
673 }
e06799f9
TM		 674}
675
9903cd1c 676/**
262965f5 677 * xs_tcp_send_request - write an RPC request to a TCP socket
9903cd1c
CL		 678 * @task: address of RPC task that manages the state of an RPC request
679 *
680 * Return values:
262965f5
CL		 681 * 0: The request has been sent
682 * EAGAIN: The socket was blocked, please call again later to
683 * complete the request
684 * ENOTCONN: Caller needs to invoke connect logic then call again
25985edc 685 * other: Some other error occurred, the request was not sent
9903cd1c
CL		 686 *
687 * XXX: In the case of soft timeouts, should we eventually give up
262965f5 688 * if sendmsg is not able to make progress?
9903cd1c 689 */
262965f5 690static int xs_tcp_send_request(struct rpc_task *task)
a246b010
CL		 691{
692 struct rpc_rqst *req = task->tk_rqstp;
693 struct rpc_xprt *xprt = req->rq_xprt;
ee0ac0c2 694 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
262965f5 695 struct xdr_buf *xdr = &req->rq_snd_buf;
b595bb15 696 int status;
a246b010 697
61677eee 698 xs_encode_stream_record_marker(&req->rq_snd_buf);
a246b010 699
262965f5
CL		 700 xs_pktdump("packet data:",
701 req->rq_svec->iov_base,
702 req->rq_svec->iov_len);
a246b010
CL		 703
704 /* Continue transmitting the packet/record. We must be careful
705 * to cope with writespace callbacks arriving _after_ we have
262965f5 706 * called sendmsg(). */
a246b010 707 while (1) {
ee0ac0c2
CL		 708 status = xs_sendpages(transport->sock,
709 NULL, 0, xdr, req->rq_bytes_sent);
a246b010 710
46121cf7 711 dprintk("RPC: xs_tcp_send_request(%u) = %d\n",
262965f5 712 xdr->len - req->rq_bytes_sent, status);
a246b010 713
262965f5 714 if (unlikely(status < 0))
a246b010 715 break;
a246b010 716
262965f5
CL		 717 /* If we've sent the entire packet, immediately
718 * reset the count of bytes sent. */
719 req->rq_bytes_sent += status;
d60dbb20 720 req->rq_xmit_bytes_sent += status;
262965f5
CL		 721 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
722 req->rq_bytes_sent = 0;
723 return 0;
724 }
a246b010 725
06b4b681
TM		 726 if (status != 0)
727 continue;
a246b010 728 status = -EAGAIN;
06b4b681 729 break;
a246b010
CL		 730 }
731
262965f5 732 switch (status) {
fba91afb
TM		 733 case -ENOTSOCK:
734 status = -ENOTCONN;
735 /* Should we call xs_close() here? */
736 break;
262965f5 737 case -EAGAIN:
5e3771ce 738 status = xs_nospace(task);
262965f5 739 break;
c8485e4d
TM		 740 default:
741 dprintk("RPC: sendmsg returned unrecognized error %d\n",
742 -status);
262965f5 743 case -ECONNRESET:
2a9e1cfa
TM		 744 xs_tcp_shutdown(xprt);
745 case -ECONNREFUSED:
262965f5 746 case -ENOTCONN:
b9d2bb2e 747 case -EPIPE:
b6ddf64f 748 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
a246b010 749 }
5fe46e9d 750
a246b010
CL		 751 return status;
752}
753
e0ab53de
TM		 754/**
755 * xs_tcp_release_xprt - clean up after a tcp transmission
756 * @xprt: transport
757 * @task: rpc task
758 *
759 * This cleans up if an error causes us to abort the transmission of a request.
760 * In this case, the socket may need to be reset in order to avoid confusing
761 * the server.
762 */
763static void xs_tcp_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
764{
765 struct rpc_rqst *req;
766
767 if (task != xprt->snd_task)
768 return;
769 if (task == NULL)
770 goto out_release;
771 req = task->tk_rqstp;
43cedbf0
TM		 772 if (req == NULL)
773 goto out_release;
e0ab53de
TM		 774 if (req->rq_bytes_sent == 0)
775 goto out_release;
776 if (req->rq_bytes_sent == req->rq_snd_buf.len)
777 goto out_release;
a4f0835c 778 set_bit(XPRT_CLOSE_WAIT, &xprt->state);
e0ab53de
TM		 779out_release:
780 xprt_release_xprt(xprt, task);
781}
782
2a9e1cfa
TM		 783static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk)
784{
785 transport->old_data_ready = sk->sk_data_ready;
786 transport->old_state_change = sk->sk_state_change;
787 transport->old_write_space = sk->sk_write_space;
2a9e1cfa
TM		 788}
789
790static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk)
791{
792 sk->sk_data_ready = transport->old_data_ready;
793 sk->sk_state_change = transport->old_state_change;
794 sk->sk_write_space = transport->old_write_space;
2a9e1cfa
TM		 795}
796
fe315e76 797static void xs_reset_transport(struct sock_xprt *transport)
a246b010 798{
ee0ac0c2
CL		 799 struct socket *sock = transport->sock;
800 struct sock *sk = transport->inet;
a246b010 801
fe315e76
CL		 802 if (sk == NULL)
803 return;
9903cd1c 804
246408dc
TM		 805 transport->srcport = 0;
806
a246b010 807 write_lock_bh(&sk->sk_callback_lock);
ee0ac0c2
CL		 808 transport->inet = NULL;
809 transport->sock = NULL;
a246b010 810
9903cd1c 811 sk->sk_user_data = NULL;
2a9e1cfa
TM		 812
813 xs_restore_old_callbacks(transport, sk);
a246b010
CL		 814 write_unlock_bh(&sk->sk_callback_lock);
815
9903cd1c 816 sk->sk_no_check = 0;
a246b010 817
40b5ea0c 818 trace_rpc_socket_close(&transport->xprt, sock);
a246b010 819 sock_release(sock);
fe315e76
CL		 820}
821
822/**
823 * xs_close - close a socket
824 * @xprt: transport
825 *
826 * This is used when all requests are complete; ie, no DRC state remains
827 * on the server we want to save.
f75e6745
TM		 828 *
829 * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with
830 * xs_reset_transport() zeroing the socket from underneath a writer.
fe315e76
CL		 831 */
832static void xs_close(struct rpc_xprt *xprt)
833{
834 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
835
836 dprintk("RPC: xs_close xprt %p\n", xprt);
837
838 xs_reset_transport(transport);
61d0a8e6 839 xprt->reestablish_timeout = 0;
fe315e76 840
632e3bdc 841 smp_mb__before_clear_bit();
7d1e8255 842 clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
632e3bdc 843 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
3b948ae5 844 clear_bit(XPRT_CLOSING, &xprt->state);
632e3bdc 845 smp_mb__after_clear_bit();
62da3b24 846 xprt_disconnect_done(xprt);
a246b010
CL		 847}
848
f75e6745
TM		 849static void xs_tcp_close(struct rpc_xprt *xprt)
850{
851 if (test_and_clear_bit(XPRT_CONNECTION_CLOSE, &xprt->state))
852 xs_close(xprt);
853 else
854 xs_tcp_shutdown(xprt);
855}
856
190b1ecf
BF		 857static void xs_local_destroy(struct rpc_xprt *xprt)
858{
859 xs_close(xprt);
860 xs_free_peer_addresses(xprt);
861 xprt_free(xprt);
862 module_put(THIS_MODULE);
863}
864
9903cd1c
CL		 865/**
866 * xs_destroy - prepare to shutdown a transport
867 * @xprt: doomed transport
868 *
869 */
870static void xs_destroy(struct rpc_xprt *xprt)
a246b010 871{
c8475461
CL		 872 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
873
46121cf7 874 dprintk("RPC: xs_destroy xprt %p\n", xprt);
9903cd1c 875
afe2c511 876 cancel_delayed_work_sync(&transport->connect_worker);
a246b010 877
190b1ecf 878 xs_local_destroy(xprt);
a246b010
CL		 879}
880
9903cd1c
CL		 881static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
882{
883 return (struct rpc_xprt *) sk->sk_user_data;
884}
885
176e21ee
CL		 886static int xs_local_copy_to_xdr(struct xdr_buf *xdr, struct sk_buff *skb)
887{
888 struct xdr_skb_reader desc = {
889 .skb = skb,
890 .offset = sizeof(rpc_fraghdr),
891 .count = skb->len - sizeof(rpc_fraghdr),
892 };
893
894 if (xdr_partial_copy_from_skb(xdr, 0, &desc, xdr_skb_read_bits) < 0)
895 return -1;
896 if (desc.count)
897 return -1;
898 return 0;
899}
900
901/**
902 * xs_local_data_ready - "data ready" callback for AF_LOCAL sockets
903 * @sk: socket with data to read
904 * @len: how much data to read
905 *
906 * Currently this assumes we can read the whole reply in a single gulp.
907 */
908static void xs_local_data_ready(struct sock *sk, int len)
909{
910 struct rpc_task *task;
911 struct rpc_xprt *xprt;
912 struct rpc_rqst *rovr;
913 struct sk_buff *skb;
914 int err, repsize, copied;
915 u32 _xid;
916 __be32 *xp;
917
918 read_lock_bh(&sk->sk_callback_lock);
919 dprintk("RPC: %s...\n", __func__);
920 xprt = xprt_from_sock(sk);
921 if (xprt == NULL)
922 goto out;
923
924 skb = skb_recv_datagram(sk, 0, 1, &err);
925 if (skb == NULL)
926 goto out;
927
176e21ee
CL		 928 repsize = skb->len - sizeof(rpc_fraghdr);
929 if (repsize < 4) {
930 dprintk("RPC: impossible RPC reply size %d\n", repsize);
931 goto dropit;
932 }
933
934 /* Copy the XID from the skb... */
935 xp = skb_header_pointer(skb, sizeof(rpc_fraghdr), sizeof(_xid), &_xid);
936 if (xp == NULL)
937 goto dropit;
938
939 /* Look up and lock the request corresponding to the given XID */
940 spin_lock(&xprt->transport_lock);
941 rovr = xprt_lookup_rqst(xprt, *xp);
942 if (!rovr)
943 goto out_unlock;
944 task = rovr->rq_task;
945
946 copied = rovr->rq_private_buf.buflen;
947 if (copied > repsize)
948 copied = repsize;
949
950 if (xs_local_copy_to_xdr(&rovr->rq_private_buf, skb)) {
951 dprintk("RPC: sk_buff copy failed\n");
952 goto out_unlock;
953 }
954
955 xprt_complete_rqst(task, copied);
956
957 out_unlock:
958 spin_unlock(&xprt->transport_lock);
959 dropit:
960 skb_free_datagram(sk, skb);
961 out:
962 read_unlock_bh(&sk->sk_callback_lock);
963}
964
9903cd1c
CL		 965/**
966 * xs_udp_data_ready - "data ready" callback for UDP sockets
967 * @sk: socket with data to read
968 * @len: how much data to read
969 *
a246b010 970 */
9903cd1c 971static void xs_udp_data_ready(struct sock *sk, int len)
a246b010 972{
9903cd1c
CL		 973 struct rpc_task *task;
974 struct rpc_xprt *xprt;
a246b010 975 struct rpc_rqst *rovr;
9903cd1c 976 struct sk_buff *skb;
a246b010 977 int err, repsize, copied;
d8ed029d
AD		 978 u32 _xid;
979 __be32 *xp;
a246b010 980
f064af1e 981 read_lock_bh(&sk->sk_callback_lock);
46121cf7 982 dprintk("RPC: xs_udp_data_ready...\n");
9903cd1c 983 if (!(xprt = xprt_from_sock(sk)))
a246b010 984 goto out;
a246b010
CL		 985
986 if ((skb = skb_recv_datagram(sk, 0, 1, &err)) == NULL)
987 goto out;
988
a246b010
CL		 989 repsize = skb->len - sizeof(struct udphdr);
990 if (repsize < 4) {
46121cf7 991 dprintk("RPC: impossible RPC reply size %d!\n", repsize);
a246b010
CL		 992 goto dropit;
993 }
994
995 /* Copy the XID from the skb... */
996 xp = skb_header_pointer(skb, sizeof(struct udphdr),
997 sizeof(_xid), &_xid);
998 if (xp == NULL)
999 goto dropit;
1000
1001 /* Look up and lock the request corresponding to the given XID */
4a0f8c04 1002 spin_lock(&xprt->transport_lock);
a246b010
CL		 1003 rovr = xprt_lookup_rqst(xprt, *xp);
1004 if (!rovr)
1005 goto out_unlock;
1006 task = rovr->rq_task;
1007
a246b010
CL		 1008 if ((copied = rovr->rq_private_buf.buflen) > repsize)
1009 copied = repsize;
1010
1011 /* Suck it into the iovec, verify checksum if not done by hw. */
1781f7f5
HX		 1012 if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
1013 UDPX_INC_STATS_BH(sk, UDP_MIB_INERRORS);
a246b010 1014 goto out_unlock;
1781f7f5
HX		 1015 }
1016
1017 UDPX_INC_STATS_BH(sk, UDP_MIB_INDATAGRAMS);
a246b010 1018
6a24dfb6 1019 xprt_adjust_cwnd(xprt, task, copied);
1570c1e4 1020 xprt_complete_rqst(task, copied);
a246b010
CL		 1021
1022 out_unlock:
4a0f8c04 1023 spin_unlock(&xprt->transport_lock);
a246b010
CL		 1024 dropit:
1025 skb_free_datagram(sk, skb);
1026 out:
f064af1e 1027 read_unlock_bh(&sk->sk_callback_lock);
a246b010
CL		 1028}
1029
a519fc7a
TM		 1030/*
1031 * Helper function to force a TCP close if the server is sending
1032 * junk and/or it has put us in CLOSE_WAIT
1033 */
1034static void xs_tcp_force_close(struct rpc_xprt *xprt)
1035{
1036 set_bit(XPRT_CONNECTION_CLOSE, &xprt->state);
1037 xprt_force_disconnect(xprt);
1038}
1039
dd456471 1040static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
a246b010 1041{
51971139 1042 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
a246b010
CL		 1043 size_t len, used;
1044 char *p;
1045
51971139
CL		 1046 p = ((char *) &transport->tcp_fraghdr) + transport->tcp_offset;
1047 len = sizeof(transport->tcp_fraghdr) - transport->tcp_offset;
9d292316 1048 used = xdr_skb_read_bits(desc, p, len);
51971139 1049 transport->tcp_offset += used;
a246b010
CL		 1050 if (used != len)
1051 return;
808012fb 1052
51971139
CL		 1053 transport->tcp_reclen = ntohl(transport->tcp_fraghdr);
1054 if (transport->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
e136d092 1055 transport->tcp_flags |= TCP_RCV_LAST_FRAG;
a246b010 1056 else
e136d092 1057 transport->tcp_flags &= ~TCP_RCV_LAST_FRAG;
51971139 1058 transport->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
808012fb 1059
e136d092 1060 transport->tcp_flags &= ~TCP_RCV_COPY_FRAGHDR;
51971139 1061 transport->tcp_offset = 0;
808012fb 1062
a246b010 1063 /* Sanity check of the record length */
18dca02a 1064 if (unlikely(transport->tcp_reclen < 8)) {
46121cf7 1065 dprintk("RPC: invalid TCP record fragment length\n");
a519fc7a 1066 xs_tcp_force_close(xprt);
9903cd1c 1067 return;
a246b010 1068 }
46121cf7 1069 dprintk("RPC: reading TCP record fragment of length %d\n",
51971139 1070 transport->tcp_reclen);
a246b010
CL		 1071}
1072
51971139 1073static void xs_tcp_check_fraghdr(struct sock_xprt *transport)
a246b010 1074{
51971139 1075 if (transport->tcp_offset == transport->tcp_reclen) {
e136d092 1076 transport->tcp_flags |= TCP_RCV_COPY_FRAGHDR;
51971139 1077 transport->tcp_offset = 0;
e136d092
CL		 1078 if (transport->tcp_flags & TCP_RCV_LAST_FRAG) {
1079 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1080 transport->tcp_flags |= TCP_RCV_COPY_XID;
51971139 1081 transport->tcp_copied = 0;
a246b010
CL		 1082 }
1083 }
1084}
1085
dd456471 1086static inline void xs_tcp_read_xid(struct sock_xprt *transport, struct xdr_skb_reader *desc)
a246b010
CL		 1087{
1088 size_t len, used;
1089 char *p;
1090
51971139 1091 len = sizeof(transport->tcp_xid) - transport->tcp_offset;
46121cf7 1092 dprintk("RPC: reading XID (%Zu bytes)\n", len);
51971139 1093 p = ((char *) &transport->tcp_xid) + transport->tcp_offset;
9d292316 1094 used = xdr_skb_read_bits(desc, p, len);
51971139 1095 transport->tcp_offset += used;
a246b010
CL		 1096 if (used != len)
1097 return;
e136d092 1098 transport->tcp_flags &= ~TCP_RCV_COPY_XID;
f4a2e418 1099 transport->tcp_flags |= TCP_RCV_READ_CALLDIR;
51971139 1100 transport->tcp_copied = 4;
18dca02a
RL		 1101 dprintk("RPC: reading %s XID %08x\n",
1102 (transport->tcp_flags & TCP_RPC_REPLY) ? "reply for"
1103 : "request with",
51971139
CL		 1104 ntohl(transport->tcp_xid));
1105 xs_tcp_check_fraghdr(transport);
a246b010
CL		 1106}
1107
18dca02a
RL		 1108static inline void xs_tcp_read_calldir(struct sock_xprt *transport,
1109 struct xdr_skb_reader *desc)
a246b010 1110{
18dca02a
RL		 1111 size_t len, used;
1112 u32 offset;
b76ce561 1113 char *p;
18dca02a
RL		 1114
1115 /*
1116 * We want transport->tcp_offset to be 8 at the end of this routine
1117 * (4 bytes for the xid and 4 bytes for the call/reply flag).
1118 * When this function is called for the first time,
1119 * transport->tcp_offset is 4 (after having already read the xid).
1120 */
1121 offset = transport->tcp_offset - sizeof(transport->tcp_xid);
b76ce561 1122 len = sizeof(transport->tcp_calldir) - offset;
18dca02a 1123 dprintk("RPC: reading CALL/REPLY flag (%Zu bytes)\n", len);
b76ce561
TM		 1124 p = ((char *) &transport->tcp_calldir) + offset;
1125 used = xdr_skb_read_bits(desc, p, len);
18dca02a
RL		 1126 transport->tcp_offset += used;
1127 if (used != len)
1128 return;
f4a2e418 1129 transport->tcp_flags &= ~TCP_RCV_READ_CALLDIR;
f4a2e418
RL		 1130 /*
1131 * We don't yet have the XDR buffer, so we will write the calldir
1132 * out after we get the buffer from the 'struct rpc_rqst'
1133 */
b76ce561
TM		 1134 switch (ntohl(transport->tcp_calldir)) {
1135 case RPC_REPLY:
1136 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1137 transport->tcp_flags |= TCP_RCV_COPY_DATA;
18dca02a 1138 transport->tcp_flags |= TCP_RPC_REPLY;
b76ce561
TM		 1139 break;
1140 case RPC_CALL:
1141 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1142 transport->tcp_flags |= TCP_RCV_COPY_DATA;
18dca02a 1143 transport->tcp_flags &= ~TCP_RPC_REPLY;
b76ce561
TM		 1144 break;
1145 default:
1146 dprintk("RPC: invalid request message type\n");
a519fc7a 1147 xs_tcp_force_close(&transport->xprt);
b76ce561 1148 }
18dca02a
RL		 1149 xs_tcp_check_fraghdr(transport);
1150}
1151
44b98efd
RL		 1152static inline void xs_tcp_read_common(struct rpc_xprt *xprt,
1153 struct xdr_skb_reader *desc,
1154 struct rpc_rqst *req)
a246b010 1155{
44b98efd
RL		 1156 struct sock_xprt *transport =
1157 container_of(xprt, struct sock_xprt, xprt);
a246b010
CL		 1158 struct xdr_buf *rcvbuf;
1159 size_t len;
1160 ssize_t r;
1161
a246b010 1162 rcvbuf = &req->rq_private_buf;
f4a2e418
RL		 1163
1164 if (transport->tcp_flags & TCP_RCV_COPY_CALLDIR) {
1165 /*
1166 * Save the RPC direction in the XDR buffer
1167 */
f4a2e418 1168 memcpy(rcvbuf->head[0].iov_base + transport->tcp_copied,
b76ce561
TM		 1169 &transport->tcp_calldir,
1170 sizeof(transport->tcp_calldir));
1171 transport->tcp_copied += sizeof(transport->tcp_calldir);
f4a2e418 1172 transport->tcp_flags &= ~TCP_RCV_COPY_CALLDIR;
a246b010
CL		 1173 }
1174
a246b010 1175 len = desc->count;
51971139 1176 if (len > transport->tcp_reclen - transport->tcp_offset) {
dd456471 1177 struct xdr_skb_reader my_desc;
a246b010 1178
51971139 1179 len = transport->tcp_reclen - transport->tcp_offset;
a246b010
CL		 1180 memcpy(&my_desc, desc, sizeof(my_desc));
1181 my_desc.count = len;
51971139 1182 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
9d292316 1183 &my_desc, xdr_skb_read_bits);
a246b010
CL		 1184 desc->count -= r;
1185 desc->offset += r;
1186 } else
51971139 1187 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
9d292316 1188 desc, xdr_skb_read_bits);
a246b010
CL		 1189
1190 if (r > 0) {
51971139
CL		 1191 transport->tcp_copied += r;
1192 transport->tcp_offset += r;
a246b010
CL		 1193 }
1194 if (r != len) {
1195 /* Error when copying to the receive buffer,
1196 * usually because we weren't able to allocate
1197 * additional buffer pages. All we can do now
e136d092 1198 * is turn off TCP_RCV_COPY_DATA, so the request
a246b010
CL		 1199 * will not receive any additional updates,
1200 * and time out.
1201 * Any remaining data from this record will
1202 * be discarded.
1203 */
e136d092 1204 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
46121cf7 1205 dprintk("RPC: XID %08x truncated request\n",
51971139 1206 ntohl(transport->tcp_xid));
46121cf7
CL		 1207 dprintk("RPC: xprt = %p, tcp_copied = %lu, "
1208 "tcp_offset = %u, tcp_reclen = %u\n",
1209 xprt, transport->tcp_copied,
1210 transport->tcp_offset, transport->tcp_reclen);
44b98efd 1211 return;
a246b010
CL		 1212 }
1213
46121cf7 1214 dprintk("RPC: XID %08x read %Zd bytes\n",
51971139 1215 ntohl(transport->tcp_xid), r);
46121cf7
CL		 1216 dprintk("RPC: xprt = %p, tcp_copied = %lu, tcp_offset = %u, "
1217 "tcp_reclen = %u\n", xprt, transport->tcp_copied,
1218 transport->tcp_offset, transport->tcp_reclen);
51971139
CL		 1219
1220 if (transport->tcp_copied == req->rq_private_buf.buflen)
e136d092 1221 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
51971139 1222 else if (transport->tcp_offset == transport->tcp_reclen) {
e136d092
CL		 1223 if (transport->tcp_flags & TCP_RCV_LAST_FRAG)
1224 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
a246b010 1225 }
44b98efd
RL		 1226}
1227
1228/*
1229 * Finds the request corresponding to the RPC xid and invokes the common
1230 * tcp read code to read the data.
1231 */
1232static inline int xs_tcp_read_reply(struct rpc_xprt *xprt,
1233 struct xdr_skb_reader *desc)
1234{
1235 struct sock_xprt *transport =
1236 container_of(xprt, struct sock_xprt, xprt);
1237 struct rpc_rqst *req;
1238
1239 dprintk("RPC: read reply XID %08x\n", ntohl(transport->tcp_xid));
1240
1241 /* Find and lock the request corresponding to this xid */
1242 spin_lock(&xprt->transport_lock);
1243 req = xprt_lookup_rqst(xprt, transport->tcp_xid);
1244 if (!req) {
1245 dprintk("RPC: XID %08x request not found!\n",
1246 ntohl(transport->tcp_xid));
1247 spin_unlock(&xprt->transport_lock);
1248 return -1;
1249 }
1250
1251 xs_tcp_read_common(xprt, desc, req);
1252
e136d092 1253 if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
51971139 1254 xprt_complete_rqst(req->rq_task, transport->tcp_copied);
44b98efd 1255
4a0f8c04 1256 spin_unlock(&xprt->transport_lock);
44b98efd
RL		 1257 return 0;
1258}
1259
9e00abc3 1260#if defined(CONFIG_SUNRPC_BACKCHANNEL)
44b98efd
RL		 1261/*
1262 * Obtains an rpc_rqst previously allocated and invokes the common
1263 * tcp read code to read the data. The result is placed in the callback
1264 * queue.
1265 * If we're unable to obtain the rpc_rqst we schedule the closing of the
1266 * connection and return -1.
1267 */
1268static inline int xs_tcp_read_callback(struct rpc_xprt *xprt,
1269 struct xdr_skb_reader *desc)
1270{
1271 struct sock_xprt *transport =
1272 container_of(xprt, struct sock_xprt, xprt);
1273 struct rpc_rqst *req;
1274
1275 req = xprt_alloc_bc_request(xprt);
1276 if (req == NULL) {
1277 printk(KERN_WARNING "Callback slot table overflowed\n");
1278 xprt_force_disconnect(xprt);
1279 return -1;
1280 }
1281
1282 req->rq_xid = transport->tcp_xid;
1283 dprintk("RPC: read callback XID %08x\n", ntohl(req->rq_xid));
1284 xs_tcp_read_common(xprt, desc, req);
1285
1286 if (!(transport->tcp_flags & TCP_RCV_COPY_DATA)) {
1287 struct svc_serv *bc_serv = xprt->bc_serv;
1288
1289 /*
1290 * Add callback request to callback list. The callback
1291 * service sleeps on the sv_cb_waitq waiting for new
1292 * requests. Wake it up after adding enqueing the
1293 * request.
1294 */
1295 dprintk("RPC: add callback request to list\n");
1296 spin_lock(&bc_serv->sv_cb_lock);
1297 list_add(&req->rq_bc_list, &bc_serv->sv_cb_list);
1298 spin_unlock(&bc_serv->sv_cb_lock);
1299 wake_up(&bc_serv->sv_cb_waitq);
1300 }
1301
1302 req->rq_private_buf.len = transport->tcp_copied;
1303
1304 return 0;
1305}
1306
1307static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1308 struct xdr_skb_reader *desc)
1309{
1310 struct sock_xprt *transport =
1311 container_of(xprt, struct sock_xprt, xprt);
1312
1313 return (transport->tcp_flags & TCP_RPC_REPLY) ?
1314 xs_tcp_read_reply(xprt, desc) :
1315 xs_tcp_read_callback(xprt, desc);
1316}
1317#else
1318static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1319 struct xdr_skb_reader *desc)
1320{
1321 return xs_tcp_read_reply(xprt, desc);
1322}
9e00abc3 1323#endif /* CONFIG_SUNRPC_BACKCHANNEL */
44b98efd
RL		 1324
1325/*
1326 * Read data off the transport. This can be either an RPC_CALL or an
1327 * RPC_REPLY. Relay the processing to helper functions.
1328 */
1329static void xs_tcp_read_data(struct rpc_xprt *xprt,
1330 struct xdr_skb_reader *desc)
1331{
1332 struct sock_xprt *transport =
1333 container_of(xprt, struct sock_xprt, xprt);
1334
1335 if (_xs_tcp_read_data(xprt, desc) == 0)
1336 xs_tcp_check_fraghdr(transport);
1337 else {
1338 /*
1339 * The transport_lock protects the request handling.
1340 * There's no need to hold it to update the tcp_flags.
1341 */
1342 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1343 }
a246b010
CL		 1344}
1345
dd456471 1346static inline void xs_tcp_read_discard(struct sock_xprt *transport, struct xdr_skb_reader *desc)
a246b010
CL		 1347{
1348 size_t len;
1349
51971139 1350 len = transport->tcp_reclen - transport->tcp_offset;
a246b010
CL		 1351 if (len > desc->count)
1352 len = desc->count;
1353 desc->count -= len;
1354 desc->offset += len;
51971139 1355 transport->tcp_offset += len;
46121cf7 1356 dprintk("RPC: discarded %Zu bytes\n", len);
51971139 1357 xs_tcp_check_fraghdr(transport);
a246b010
CL		 1358}
1359
9903cd1c 1360static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
a246b010
CL		 1361{
1362 struct rpc_xprt *xprt = rd_desc->arg.data;
51971139 1363 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
dd456471 1364 struct xdr_skb_reader desc = {
a246b010
CL		 1365 .skb = skb,
1366 .offset = offset,
1367 .count = len,
9903cd1c 1368 };
a246b010 1369
46121cf7 1370 dprintk("RPC: xs_tcp_data_recv started\n");
a246b010
CL		 1371 do {
1372 /* Read in a new fragment marker if necessary */
1373 /* Can we ever really expect to get completely empty fragments? */
e136d092 1374 if (transport->tcp_flags & TCP_RCV_COPY_FRAGHDR) {
9903cd1c 1375 xs_tcp_read_fraghdr(xprt, &desc);
a246b010
CL		 1376 continue;
1377 }
1378 /* Read in the xid if necessary */
e136d092 1379 if (transport->tcp_flags & TCP_RCV_COPY_XID) {
51971139 1380 xs_tcp_read_xid(transport, &desc);
a246b010
CL		 1381 continue;
1382 }
18dca02a 1383 /* Read in the call/reply flag */
f4a2e418 1384 if (transport->tcp_flags & TCP_RCV_READ_CALLDIR) {
18dca02a
RL		 1385 xs_tcp_read_calldir(transport, &desc);
1386 continue;
1387 }
a246b010 1388 /* Read in the request data */
e136d092 1389 if (transport->tcp_flags & TCP_RCV_COPY_DATA) {
44b98efd 1390 xs_tcp_read_data(xprt, &desc);
a246b010
CL		 1391 continue;
1392 }
1393 /* Skip over any trailing bytes on short reads */
51971139 1394 xs_tcp_read_discard(transport, &desc);
a246b010 1395 } while (desc.count);
46121cf7 1396 dprintk("RPC: xs_tcp_data_recv done\n");
a246b010
CL		 1397 return len - desc.count;
1398}
1399
9903cd1c
CL		 1400/**
1401 * xs_tcp_data_ready - "data ready" callback for TCP sockets
1402 * @sk: socket with data to read
1403 * @bytes: how much data to read
1404 *
1405 */
1406static void xs_tcp_data_ready(struct sock *sk, int bytes)
a246b010
CL		 1407{
1408 struct rpc_xprt *xprt;
1409 read_descriptor_t rd_desc;
ff2d7db8 1410 int read;
a246b010 1411
46121cf7
CL		 1412 dprintk("RPC: xs_tcp_data_ready...\n");
1413
f064af1e 1414 read_lock_bh(&sk->sk_callback_lock);
9903cd1c 1415 if (!(xprt = xprt_from_sock(sk)))
a246b010 1416 goto out;
61d0a8e6
NB		 1417 /* Any data means we had a useful conversation, so
1418 * the we don't need to delay the next reconnect
1419 */
1420 if (xprt->reestablish_timeout)
1421 xprt->reestablish_timeout = 0;
1422
9903cd1c 1423 /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
a246b010 1424 rd_desc.arg.data = xprt;
ff2d7db8
TM		 1425 do {
1426 rd_desc.count = 65536;
1427 read = tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
1428 } while (read > 0);
a246b010 1429out:
f064af1e 1430 read_unlock_bh(&sk->sk_callback_lock);
a246b010
CL		 1431}
1432
7d1e8255
TM		 1433/*
1434 * Do the equivalent of linger/linger2 handling for dealing with
1435 * broken servers that don't close the socket in a timely
1436 * fashion
1437 */
1438static void xs_tcp_schedule_linger_timeout(struct rpc_xprt *xprt,
1439 unsigned long timeout)
1440{
1441 struct sock_xprt *transport;
1442
1443 if (xprt_test_and_set_connecting(xprt))
1444 return;
1445 set_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1446 transport = container_of(xprt, struct sock_xprt, xprt);
1447 queue_delayed_work(rpciod_workqueue, &transport->connect_worker,
1448 timeout);
1449}
1450
1451static void xs_tcp_cancel_linger_timeout(struct rpc_xprt *xprt)
1452{
1453 struct sock_xprt *transport;
1454
1455 transport = container_of(xprt, struct sock_xprt, xprt);
1456
1457 if (!test_bit(XPRT_CONNECTION_ABORT, &xprt->state) ||
1458 !cancel_delayed_work(&transport->connect_worker))
1459 return;
1460 clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1461 xprt_clear_connecting(xprt);
1462}
1463
4bc1e68e 1464static void xs_sock_reset_connection_flags(struct rpc_xprt *xprt)
7d1e8255
TM		 1465{
1466 smp_mb__before_clear_bit();
a519fc7a
TM		 1467 clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1468 clear_bit(XPRT_CONNECTION_CLOSE, &xprt->state);
7d1e8255
TM		 1469 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1470 clear_bit(XPRT_CLOSING, &xprt->state);
1471 smp_mb__after_clear_bit();
4bc1e68e
TM		 1472}
1473
1474static void xs_sock_mark_closed(struct rpc_xprt *xprt)
1475{
1476 xs_sock_reset_connection_flags(xprt);
7d1e8255
TM		 1477 /* Mark transport as closed and wake up all pending tasks */
1478 xprt_disconnect_done(xprt);
1479}
1480
9903cd1c
CL		 1481/**
1482 * xs_tcp_state_change - callback to handle TCP socket state changes
1483 * @sk: socket whose state has changed
1484 *
1485 */
1486static void xs_tcp_state_change(struct sock *sk)
a246b010 1487{
9903cd1c 1488 struct rpc_xprt *xprt;
a246b010 1489
f064af1e 1490 read_lock_bh(&sk->sk_callback_lock);
a246b010
CL		 1491 if (!(xprt = xprt_from_sock(sk)))
1492 goto out;
46121cf7 1493 dprintk("RPC: xs_tcp_state_change client %p...\n", xprt);
669502ff 1494 dprintk("RPC: state %x conn %d dead %d zapped %d sk_shutdown %d\n",
46121cf7
CL		 1495 sk->sk_state, xprt_connected(xprt),
1496 sock_flag(sk, SOCK_DEAD),
669502ff
AC		 1497 sock_flag(sk, SOCK_ZAPPED),
1498 sk->sk_shutdown);
a246b010 1499
40b5ea0c 1500 trace_rpc_socket_state_change(xprt, sk->sk_socket);
a246b010
CL		 1501 switch (sk->sk_state) {
1502 case TCP_ESTABLISHED:
f064af1e 1503 spin_lock(&xprt->transport_lock);
a246b010 1504 if (!xprt_test_and_set_connected(xprt)) {
51971139
CL		 1505 struct sock_xprt *transport = container_of(xprt,
1506 struct sock_xprt, xprt);
1507
a246b010 1508 /* Reset TCP record info */
51971139
CL		 1509 transport->tcp_offset = 0;
1510 transport->tcp_reclen = 0;
1511 transport->tcp_copied = 0;
e136d092
CL		 1512 transport->tcp_flags =
1513 TCP_RCV_COPY_FRAGHDR | TCP_RCV_COPY_XID;
8b71798c 1514 xprt->connect_cookie++;
51971139 1515
2a491991 1516 xprt_wake_pending_tasks(xprt, -EAGAIN);
a246b010 1517 }
f064af1e 1518 spin_unlock(&xprt->transport_lock);
a246b010 1519 break;
3b948ae5
TM		 1520 case TCP_FIN_WAIT1:
1521 /* The client initiated a shutdown of the socket */
7c1d71cf 1522 xprt->connect_cookie++;
663b8858 1523 xprt->reestablish_timeout = 0;
3b948ae5
TM		 1524 set_bit(XPRT_CLOSING, &xprt->state);
1525 smp_mb__before_clear_bit();
1526 clear_bit(XPRT_CONNECTED, &xprt->state);
ef803670 1527 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
3b948ae5 1528 smp_mb__after_clear_bit();
25fe6142 1529 xs_tcp_schedule_linger_timeout(xprt, xs_tcp_fin_timeout);
a246b010 1530 break;
632e3bdc 1531 case TCP_CLOSE_WAIT:
3b948ae5 1532 /* The server initiated a shutdown of the socket */
7c1d71cf 1533 xprt->connect_cookie++;
d0bea455 1534 clear_bit(XPRT_CONNECTED, &xprt->state);
a519fc7a 1535 xs_tcp_force_close(xprt);
663b8858
TM		 1536 case TCP_CLOSING:
1537 /*
1538 * If the server closed down the connection, make sure that
1539 * we back off before reconnecting
1540 */
1541 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1542 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
3b948ae5
TM		 1543 break;
1544 case TCP_LAST_ACK:
670f9457 1545 set_bit(XPRT_CLOSING, &xprt->state);
25fe6142 1546 xs_tcp_schedule_linger_timeout(xprt, xs_tcp_fin_timeout);
3b948ae5
TM		 1547 smp_mb__before_clear_bit();
1548 clear_bit(XPRT_CONNECTED, &xprt->state);
1549 smp_mb__after_clear_bit();
1550 break;
1551 case TCP_CLOSE:
7d1e8255
TM		 1552 xs_tcp_cancel_linger_timeout(xprt);
1553 xs_sock_mark_closed(xprt);
a246b010
CL		 1554 }
1555 out:
f064af1e 1556 read_unlock_bh(&sk->sk_callback_lock);
a246b010
CL		 1557}
1558
1f0fa154
IJ		 1559static void xs_write_space(struct sock *sk)
1560{
1561 struct socket *sock;
1562 struct rpc_xprt *xprt;
1563
1564 if (unlikely(!(sock = sk->sk_socket)))
1565 return;
1566 clear_bit(SOCK_NOSPACE, &sock->flags);
1567
1568 if (unlikely(!(xprt = xprt_from_sock(sk))))
1569 return;
1570 if (test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags) == 0)
1571 return;
1572
1573 xprt_write_space(xprt);
1574}
1575
9903cd1c 1576/**
c7b2cae8
CL		 1577 * xs_udp_write_space - callback invoked when socket buffer space
1578 * becomes available
9903cd1c
CL		 1579 * @sk: socket whose state has changed
1580 *
a246b010
CL		 1581 * Called when more output buffer space is available for this socket.
1582 * We try not to wake our writers until they can make "significant"
c7b2cae8 1583 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
a246b010
CL		 1584 * with a bunch of small requests.
1585 */
c7b2cae8 1586static void xs_udp_write_space(struct sock *sk)
a246b010 1587{
f064af1e 1588 read_lock_bh(&sk->sk_callback_lock);
a246b010 1589
c7b2cae8 1590 /* from net/core/sock.c:sock_def_write_space */
1f0fa154
IJ		 1591 if (sock_writeable(sk))
1592 xs_write_space(sk);
a246b010 1593
f064af1e 1594 read_unlock_bh(&sk->sk_callback_lock);
c7b2cae8 1595}
a246b010 1596
c7b2cae8
CL		 1597/**
1598 * xs_tcp_write_space - callback invoked when socket buffer space
1599 * becomes available
1600 * @sk: socket whose state has changed
1601 *
1602 * Called when more output buffer space is available for this socket.
1603 * We try not to wake our writers until they can make "significant"
1604 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1605 * with a bunch of small requests.
1606 */
1607static void xs_tcp_write_space(struct sock *sk)
1608{
f064af1e 1609 read_lock_bh(&sk->sk_callback_lock);
c7b2cae8
CL		 1610
1611 /* from net/core/stream.c:sk_stream_write_space */
64dc6130 1612 if (sk_stream_is_writeable(sk))
1f0fa154 1613 xs_write_space(sk);
b6ddf64f 1614
f064af1e 1615 read_unlock_bh(&sk->sk_callback_lock);
a246b010
CL		 1616}
1617
470056c2 1618static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
a246b010 1619{
ee0ac0c2
CL		 1620 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1621 struct sock *sk = transport->inet;
a246b010 1622
7c6e066e 1623 if (transport->rcvsize) {
a246b010 1624 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
7c6e066e 1625 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
a246b010 1626 }
7c6e066e 1627 if (transport->sndsize) {
a246b010 1628 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
7c6e066e 1629 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
a246b010
CL		 1630 sk->sk_write_space(sk);
1631 }
1632}
1633
43118c29 1634/**
470056c2 1635 * xs_udp_set_buffer_size - set send and receive limits
43118c29 1636 * @xprt: generic transport
470056c2
CL		 1637 * @sndsize: requested size of send buffer, in bytes
1638 * @rcvsize: requested size of receive buffer, in bytes
43118c29 1639 *
470056c2 1640 * Set socket send and receive buffer size limits.
43118c29 1641 */
470056c2 1642static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
43118c29 1643{
7c6e066e
CL		 1644 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1645
1646 transport->sndsize = 0;
470056c2 1647 if (sndsize)
7c6e066e
CL		 1648 transport->sndsize = sndsize + 1024;
1649 transport->rcvsize = 0;
470056c2 1650 if (rcvsize)
7c6e066e 1651 transport->rcvsize = rcvsize + 1024;
470056c2
CL		 1652
1653 xs_udp_do_set_buffer_size(xprt);
43118c29
CL		 1654}
1655
46c0ee8b
CL		 1656/**
1657 * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1658 * @task: task that timed out
1659 *
1660 * Adjust the congestion window after a retransmit timeout has occurred.
1661 */
6a24dfb6 1662static void xs_udp_timer(struct rpc_xprt *xprt, struct rpc_task *task)
46c0ee8b 1663{
6a24dfb6 1664 xprt_adjust_cwnd(xprt, task, -ETIMEDOUT);
46c0ee8b
CL		 1665}
1666
b85d8806
CL		 1667static unsigned short xs_get_random_port(void)
1668{
1669 unsigned short range = xprt_max_resvport - xprt_min_resvport;
1670 unsigned short rand = (unsigned short) net_random() % range;
1671 return rand + xprt_min_resvport;
1672}
1673
92200412
CL		 1674/**
1675 * xs_set_port - reset the port number in the remote endpoint address
1676 * @xprt: generic transport
1677 * @port: new port number
1678 *
1679 */
1680static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1681{
46121cf7 1682 dprintk("RPC: setting port for xprt %p to %u\n", xprt, port);
c4efcb1d 1683
9dc3b095
CL		 1684 rpc_set_port(xs_addr(xprt), port);
1685 xs_update_peer_port(xprt);
92200412
CL		 1686}
1687
5d4ec932 1688static unsigned short xs_get_srcport(struct sock_xprt *transport)
67a391d7 1689{
fbfffbd5 1690 unsigned short port = transport->srcport;
67a391d7
TM		 1691
1692 if (port == 0 && transport->xprt.resvport)
1693 port = xs_get_random_port();
1694 return port;
1695}
1696
baaf4e48 1697static unsigned short xs_next_srcport(struct sock_xprt *transport, unsigned short port)
67a391d7 1698{
fbfffbd5
CL		 1699 if (transport->srcport != 0)
1700 transport->srcport = 0;
67a391d7
TM		 1701 if (!transport->xprt.resvport)
1702 return 0;
1703 if (port <= xprt_min_resvport || port > xprt_max_resvport)
1704 return xprt_max_resvport;
1705 return --port;
1706}
beb59b68 1707static int xs_bind(struct sock_xprt *transport, struct socket *sock)
a246b010 1708{
beb59b68 1709 struct sockaddr_storage myaddr;
67a391d7 1710 int err, nloop = 0;
5d4ec932 1711 unsigned short port = xs_get_srcport(transport);
67a391d7 1712 unsigned short last;
a246b010 1713
beb59b68 1714 memcpy(&myaddr, &transport->srcaddr, transport->xprt.addrlen);
a246b010 1715 do {
beb59b68
PE		 1716 rpc_set_port((struct sockaddr *)&myaddr, port);
1717 err = kernel_bind(sock, (struct sockaddr *)&myaddr,
1718 transport->xprt.addrlen);
67a391d7 1719 if (port == 0)
d3bc9a1d 1720 break;
a246b010 1721 if (err == 0) {
fbfffbd5 1722 transport->srcport = port;
d3bc9a1d 1723 break;
a246b010 1724 }
67a391d7 1725 last = port;
baaf4e48 1726 port = xs_next_srcport(transport, port);
67a391d7
TM		 1727 if (port > last)
1728 nloop++;
1729 } while (err == -EADDRINUSE && nloop != 2);
90058d37 1730
4232e863 1731 if (myaddr.ss_family == AF_INET)
beb59b68
PE		 1732 dprintk("RPC: %s %pI4:%u: %s (%d)\n", __func__,
1733 &((struct sockaddr_in *)&myaddr)->sin_addr,
1734 port, err ? "failed" : "ok", err);
1735 else
1736 dprintk("RPC: %s %pI6:%u: %s (%d)\n", __func__,
1737 &((struct sockaddr_in6 *)&myaddr)->sin6_addr,
1738 port, err ? "failed" : "ok", err);
a246b010
CL		 1739 return err;
1740}
1741
176e21ee
CL		 1742/*
1743 * We don't support autobind on AF_LOCAL sockets
1744 */
1745static void xs_local_rpcbind(struct rpc_task *task)
1746{
3dc0da27
TM		 1747 rcu_read_lock();
1748 xprt_set_bound(rcu_dereference(task->tk_client->cl_xprt));
1749 rcu_read_unlock();
176e21ee
CL		 1750}
1751
1752static void xs_local_set_port(struct rpc_xprt *xprt, unsigned short port)
1753{
1754}
beb59b68 1755
ed07536e
PZ		 1756#ifdef CONFIG_DEBUG_LOCK_ALLOC
1757static struct lock_class_key xs_key[2];
1758static struct lock_class_key xs_slock_key[2];
1759
176e21ee
CL		 1760static inline void xs_reclassify_socketu(struct socket *sock)
1761{
1762 struct sock *sk = sock->sk;
1763
176e21ee
CL		 1764 sock_lock_init_class_and_name(sk, "slock-AF_LOCAL-RPC",
1765 &xs_slock_key[1], "sk_lock-AF_LOCAL-RPC", &xs_key[1]);
1766}
1767
8945ee5e 1768static inline void xs_reclassify_socket4(struct socket *sock)
ed07536e
PZ		 1769{
1770 struct sock *sk = sock->sk;
8945ee5e 1771
8945ee5e
CL		 1772 sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1773 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1774}
ed07536e 1775
8945ee5e
CL		 1776static inline void xs_reclassify_socket6(struct socket *sock)
1777{
1778 struct sock *sk = sock->sk;
ed07536e 1779
8945ee5e
CL		 1780 sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1781 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
ed07536e 1782}
6bc9638a
PE		 1783
1784static inline void xs_reclassify_socket(int family, struct socket *sock)
1785{
1b7a1819
WAA		 1786 WARN_ON_ONCE(sock_owned_by_user(sock->sk));
1787 if (sock_owned_by_user(sock->sk))
1788 return;
1789
4232e863 1790 switch (family) {
176e21ee
CL		 1791 case AF_LOCAL:
1792 xs_reclassify_socketu(sock);
1793 break;
4232e863 1794 case AF_INET:
6bc9638a 1795 xs_reclassify_socket4(sock);
4232e863
CL		 1796 break;
1797 case AF_INET6:
6bc9638a 1798 xs_reclassify_socket6(sock);
4232e863
CL		 1799 break;
1800 }
6bc9638a 1801}
ed07536e 1802#else
176e21ee
CL		 1803static inline void xs_reclassify_socketu(struct socket *sock)
1804{
1805}
1806
8945ee5e
CL		 1807static inline void xs_reclassify_socket4(struct socket *sock)
1808{
1809}
1810
1811static inline void xs_reclassify_socket6(struct socket *sock)
ed07536e
PZ		 1812{
1813}
6bc9638a
PE		 1814
1815static inline void xs_reclassify_socket(int family, struct socket *sock)
1816{
1817}
ed07536e
PZ		 1818#endif
1819
6bc9638a
PE		 1820static struct socket *xs_create_sock(struct rpc_xprt *xprt,
1821 struct sock_xprt *transport, int family, int type, int protocol)
22f79326
PE		 1822{
1823 struct socket *sock;
1824 int err;
1825
6bc9638a 1826 err = __sock_create(xprt->xprt_net, family, type, protocol, &sock, 1);
22f79326
PE		 1827 if (err < 0) {
1828 dprintk("RPC: can't create %d transport socket (%d).\n",
1829 protocol, -err);
1830 goto out;
1831 }
6bc9638a 1832 xs_reclassify_socket(family, sock);
22f79326 1833
4cea288a
BH		 1834 err = xs_bind(transport, sock);
1835 if (err) {
22f79326
PE		 1836 sock_release(sock);
1837 goto out;
1838 }
1839
1840 return sock;
1841out:
1842 return ERR_PTR(err);
1843}
1844
176e21ee
CL		 1845static int xs_local_finish_connecting(struct rpc_xprt *xprt,
1846 struct socket *sock)
1847{
1848 struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1849 xprt);
1850
1851 if (!transport->inet) {
1852 struct sock *sk = sock->sk;
1853
1854 write_lock_bh(&sk->sk_callback_lock);
1855
1856 xs_save_old_callbacks(transport, sk);
1857
1858 sk->sk_user_data = xprt;
1859 sk->sk_data_ready = xs_local_data_ready;
1860 sk->sk_write_space = xs_udp_write_space;
176e21ee
CL		 1861 sk->sk_allocation = GFP_ATOMIC;
1862
1863 xprt_clear_connected(xprt);
1864
1865 /* Reset to new socket */
1866 transport->sock = sock;
1867 transport->inet = sk;
1868
1869 write_unlock_bh(&sk->sk_callback_lock);
1870 }
1871
1872 /* Tell the socket layer to start connecting... */
1873 xprt->stat.connect_count++;
1874 xprt->stat.connect_start = jiffies;
1875 return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, 0);
1876}
1877
1878/**
1879 * xs_local_setup_socket - create AF_LOCAL socket, connect to a local endpoint
1880 * @xprt: RPC transport to connect
1881 * @transport: socket transport to connect
1882 * @create_sock: function to create a socket of the correct type
176e21ee 1883 */
dc107402 1884static int xs_local_setup_socket(struct sock_xprt *transport)
176e21ee 1885{
176e21ee
CL		 1886 struct rpc_xprt *xprt = &transport->xprt;
1887 struct socket *sock;
1888 int status = -EIO;
1889
5cf02d09
JL		 1890 current->flags |= PF_FSTRANS;
1891
176e21ee
CL		 1892 clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1893 status = __sock_create(xprt->xprt_net, AF_LOCAL,
1894 SOCK_STREAM, 0, &sock, 1);
1895 if (status < 0) {
1896 dprintk("RPC: can't create AF_LOCAL "
1897 "transport socket (%d).\n", -status);
1898 goto out;
1899 }
1900 xs_reclassify_socketu(sock);
1901
1902 dprintk("RPC: worker connecting xprt %p via AF_LOCAL to %s\n",
1903 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1904
1905 status = xs_local_finish_connecting(xprt, sock);
40b5ea0c 1906 trace_rpc_socket_connect(xprt, sock, status);
176e21ee
CL		 1907 switch (status) {
1908 case 0:
1909 dprintk("RPC: xprt %p connected to %s\n",
1910 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1911 xprt_set_connected(xprt);
1912 break;
1913 case -ENOENT:
1914 dprintk("RPC: xprt %p: socket %s does not exist\n",
1915 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1916 break;
4a20a988
TM		 1917 case -ECONNREFUSED:
1918 dprintk("RPC: xprt %p: connection refused for %s\n",
1919 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1920 break;
176e21ee
CL		 1921 default:
1922 printk(KERN_ERR "%s: unhandled error (%d) connecting to %s\n",
1923 __func__, -status,
1924 xprt->address_strings[RPC_DISPLAY_ADDR]);
1925 }
1926
1927out:
1928 xprt_clear_connecting(xprt);
1929 xprt_wake_pending_tasks(xprt, status);
5cf02d09 1930 current->flags &= ~PF_FSTRANS;
dc107402
BF		 1931 return status;
1932}
1933
b6669737 1934static void xs_local_connect(struct rpc_xprt *xprt, struct rpc_task *task)
dc107402 1935{
dc107402
BF		 1936 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1937 int ret;
1938
1939 if (RPC_IS_ASYNC(task)) {
1940 /*
1941 * We want the AF_LOCAL connect to be resolved in the
1942 * filesystem namespace of the process making the rpc
1943 * call. Thus we connect synchronously.
1944 *
1945 * If we want to support asynchronous AF_LOCAL calls,
1946 * we'll need to figure out how to pass a namespace to
1947 * connect.
1948 */
1949 rpc_exit(task, -ENOTCONN);
1950 return;
1951 }
1952 ret = xs_local_setup_socket(transport);
1953 if (ret && !RPC_IS_SOFTCONN(task))
1954 msleep_interruptible(15000);
176e21ee
CL		 1955}
1956
a564b8f0
MG		 1957#ifdef CONFIG_SUNRPC_SWAP
1958static void xs_set_memalloc(struct rpc_xprt *xprt)
1959{
1960 struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1961 xprt);
1962
1963 if (xprt->swapper)
1964 sk_set_memalloc(transport->inet);
1965}
1966
1967/**
1968 * xs_swapper - Tag this transport as being used for swap.
1969 * @xprt: transport to tag
1970 * @enable: enable/disable
1971 *
1972 */
1973int xs_swapper(struct rpc_xprt *xprt, int enable)
1974{
1975 struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1976 xprt);
1977 int err = 0;
1978
1979 if (enable) {
1980 xprt->swapper++;
1981 xs_set_memalloc(xprt);
1982 } else if (xprt->swapper) {
1983 xprt->swapper--;
1984 sk_clear_memalloc(transport->inet);
1985 }
1986
1987 return err;
1988}
1989EXPORT_SYMBOL_GPL(xs_swapper);
1990#else
1991static void xs_set_memalloc(struct rpc_xprt *xprt)
1992{
1993}
1994#endif
1995
16be2d20
CL		 1996static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
1997{
1998 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1999
2000 if (!transport->inet) {
2001 struct sock *sk = sock->sk;
2002
2003 write_lock_bh(&sk->sk_callback_lock);
2004
2a9e1cfa
TM		 2005 xs_save_old_callbacks(transport, sk);
2006
16be2d20 2007 sk->sk_user_data = xprt;
16be2d20
CL		 2008 sk->sk_data_ready = xs_udp_data_ready;
2009 sk->sk_write_space = xs_udp_write_space;
2010 sk->sk_no_check = UDP_CSUM_NORCV;
2011 sk->sk_allocation = GFP_ATOMIC;
2012
2013 xprt_set_connected(xprt);
2014
2015 /* Reset to new socket */
2016 transport->sock = sock;
2017 transport->inet = sk;
2018
a564b8f0
MG		 2019 xs_set_memalloc(xprt);
2020
16be2d20
CL		 2021 write_unlock_bh(&sk->sk_callback_lock);
2022 }
2023 xs_udp_do_set_buffer_size(xprt);
2024}
2025
8c14ff2a 2026static void xs_udp_setup_socket(struct work_struct *work)
a246b010 2027{
34161db6
TM		 2028 struct sock_xprt *transport =
2029 container_of(work, struct sock_xprt, connect_worker.work);
c8475461 2030 struct rpc_xprt *xprt = &transport->xprt;
ee0ac0c2 2031 struct socket *sock = transport->sock;
b65c0310 2032 int status = -EIO;
9903cd1c 2033
5cf02d09
JL		 2034 current->flags |= PF_FSTRANS;
2035
b0d93ad5 2036 /* Start by resetting any existing state */
fe315e76 2037 xs_reset_transport(transport);
8c14ff2a
PE		 2038 sock = xs_create_sock(xprt, transport,
2039 xs_addr(xprt)->sa_family, SOCK_DGRAM, IPPROTO_UDP);
b65c0310 2040 if (IS_ERR(sock))
b0d93ad5 2041 goto out;
68e220bd 2042
c740eff8
CL		 2043 dprintk("RPC: worker connecting xprt %p via %s to "
2044 "%s (port %s)\n", xprt,
2045 xprt->address_strings[RPC_DISPLAY_PROTO],
2046 xprt->address_strings[RPC_DISPLAY_ADDR],
2047 xprt->address_strings[RPC_DISPLAY_PORT]);
68e220bd
CL		 2048
2049 xs_udp_finish_connecting(xprt, sock);
40b5ea0c 2050 trace_rpc_socket_connect(xprt, sock, 0);
b0d93ad5
CL		 2051 status = 0;
2052out:
b0d93ad5 2053 xprt_clear_connecting(xprt);
7d1e8255 2054 xprt_wake_pending_tasks(xprt, status);
5cf02d09 2055 current->flags &= ~PF_FSTRANS;
a246b010
CL		 2056}
2057
3167e12c
CL		 2058/*
2059 * We need to preserve the port number so the reply cache on the server can
2060 * find our cached RPC replies when we get around to reconnecting.
2061 */
58dddac9 2062static void xs_abort_connection(struct sock_xprt *transport)
3167e12c
CL		 2063{
2064 int result;
3167e12c
CL		 2065 struct sockaddr any;
2066
58dddac9 2067 dprintk("RPC: disconnecting xprt %p to reuse port\n", transport);
3167e12c
CL		 2068
2069 /*
2070 * Disconnect the transport socket by doing a connect operation
2071 * with AF_UNSPEC. This should return immediately...
2072 */
2073 memset(&any, 0, sizeof(any));
2074 any.sa_family = AF_UNSPEC;
ee0ac0c2 2075 result = kernel_connect(transport->sock, &any, sizeof(any), 0);
40b5ea0c
TM		 2076 trace_rpc_socket_reset_connection(&transport->xprt,
2077 transport->sock, result);
7d1e8255 2078 if (!result)
4bc1e68e
TM		 2079 xs_sock_reset_connection_flags(&transport->xprt);
2080 dprintk("RPC: AF_UNSPEC connect return code %d\n", result);
3167e12c
CL		 2081}
2082
58dddac9 2083static void xs_tcp_reuse_connection(struct sock_xprt *transport)
40d2549d
TM		 2084{
2085 unsigned int state = transport->inet->sk_state;
2086
669502ff
AC		 2087 if (state == TCP_CLOSE && transport->sock->state == SS_UNCONNECTED) {
2088 /* we don't need to abort the connection if the socket
2089 * hasn't undergone a shutdown
2090 */
2091 if (transport->inet->sk_shutdown == 0)
2092 return;
2093 dprintk("RPC: %s: TCP_CLOSEd and sk_shutdown set to %d\n",
2094 __func__, transport->inet->sk_shutdown);
2095 }
2096 if ((1 << state) & (TCPF_ESTABLISHED|TCPF_SYN_SENT)) {
2097 /* we don't need to abort the connection if the socket
2098 * hasn't undergone a shutdown
2099 */
2100 if (transport->inet->sk_shutdown == 0)
2101 return;
2102 dprintk("RPC: %s: ESTABLISHED/SYN_SENT "
2103 "sk_shutdown set to %d\n",
2104 __func__, transport->inet->sk_shutdown);
2105 }
58dddac9 2106 xs_abort_connection(transport);
40d2549d
TM		 2107}
2108
16be2d20 2109static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
a246b010 2110{
16be2d20 2111 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
fe19a96b 2112 int ret = -ENOTCONN;
edb267a6 2113
ee0ac0c2 2114 if (!transport->inet) {
b0d93ad5 2115 struct sock *sk = sock->sk;
7f260e85
TM		 2116 unsigned int keepidle = xprt->timeout->to_initval / HZ;
2117 unsigned int keepcnt = xprt->timeout->to_retries + 1;
2118 unsigned int opt_on = 1;
2119
2120 /* TCP Keepalive options */
2121 kernel_setsockopt(sock, SOL_SOCKET, SO_KEEPALIVE,
2122 (char *)&opt_on, sizeof(opt_on));
2123 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPIDLE,
2124 (char *)&keepidle, sizeof(keepidle));
2125 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPINTVL,
2126 (char *)&keepidle, sizeof(keepidle));
2127 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPCNT,
2128 (char *)&keepcnt, sizeof(keepcnt));
b0d93ad5
CL		 2129
2130 write_lock_bh(&sk->sk_callback_lock);
2131
2a9e1cfa
TM		 2132 xs_save_old_callbacks(transport, sk);
2133
b0d93ad5 2134 sk->sk_user_data = xprt;
b0d93ad5
CL		 2135 sk->sk_data_ready = xs_tcp_data_ready;
2136 sk->sk_state_change = xs_tcp_state_change;
2137 sk->sk_write_space = xs_tcp_write_space;
b079fa7b 2138 sk->sk_allocation = GFP_ATOMIC;
3167e12c
CL		 2139
2140 /* socket options */
2141 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
2142 sock_reset_flag(sk, SOCK_LINGER);
2143 tcp_sk(sk)->linger2 = 0;
2144 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
b0d93ad5
CL		 2145
2146 xprt_clear_connected(xprt);
2147
2148 /* Reset to new socket */
ee0ac0c2
CL		 2149 transport->sock = sock;
2150 transport->inet = sk;
b0d93ad5
CL		 2151
2152 write_unlock_bh(&sk->sk_callback_lock);
2153 }
2154
01d37c42 2155 if (!xprt_bound(xprt))
fe19a96b 2156 goto out;
01d37c42 2157
a564b8f0
MG		 2158 xs_set_memalloc(xprt);
2159
b0d93ad5 2160 /* Tell the socket layer to start connecting... */
262ca07d
CL		 2161 xprt->stat.connect_count++;
2162 xprt->stat.connect_start = jiffies;
fe19a96b
TM		 2163 ret = kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
2164 switch (ret) {
2165 case 0:
2166 case -EINPROGRESS:
2167 /* SYN_SENT! */
fe19a96b
TM		 2168 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2169 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2170 }
2171out:
2172 return ret;
16be2d20
CL		 2173}
2174
9903cd1c 2175/**
b61d59ff
TM		 2176 * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint
2177 * @xprt: RPC transport to connect
2178 * @transport: socket transport to connect
2179 * @create_sock: function to create a socket of the correct type
9903cd1c
CL		 2180 *
2181 * Invoked by a work queue tasklet.
a246b010 2182 */
cdd518d5 2183static void xs_tcp_setup_socket(struct work_struct *work)
a246b010 2184{
cdd518d5
PE		 2185 struct sock_xprt *transport =
2186 container_of(work, struct sock_xprt, connect_worker.work);
ee0ac0c2 2187 struct socket *sock = transport->sock;
a9f5f0f7 2188 struct rpc_xprt *xprt = &transport->xprt;
b61d59ff 2189 int status = -EIO;
a246b010 2190
5cf02d09
JL		 2191 current->flags |= PF_FSTRANS;
2192
ee0ac0c2 2193 if (!sock) {
7d1e8255 2194 clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
cdd518d5
PE		 2195 sock = xs_create_sock(xprt, transport,
2196 xs_addr(xprt)->sa_family, SOCK_STREAM, IPPROTO_TCP);
b61d59ff
TM		 2197 if (IS_ERR(sock)) {
2198 status = PTR_ERR(sock);
3167e12c
CL		 2199 goto out;
2200 }
7d1e8255
TM		 2201 } else {
2202 int abort_and_exit;
a246b010 2203
7d1e8255
TM		 2204 abort_and_exit = test_and_clear_bit(XPRT_CONNECTION_ABORT,
2205 &xprt->state);
3167e12c 2206 /* "close" the socket, preserving the local port */
58dddac9 2207 xs_tcp_reuse_connection(transport);
a246b010 2208
7d1e8255
TM		 2209 if (abort_and_exit)
2210 goto out_eagain;
2211 }
a246b010 2212
c740eff8
CL		 2213 dprintk("RPC: worker connecting xprt %p via %s to "
2214 "%s (port %s)\n", xprt,
2215 xprt->address_strings[RPC_DISPLAY_PROTO],
2216 xprt->address_strings[RPC_DISPLAY_ADDR],
2217 xprt->address_strings[RPC_DISPLAY_PORT]);
edb267a6 2218
16be2d20 2219 status = xs_tcp_finish_connecting(xprt, sock);
40b5ea0c 2220 trace_rpc_socket_connect(xprt, sock, status);
46121cf7
CL		 2221 dprintk("RPC: %p connect status %d connected %d sock state %d\n",
2222 xprt, -status, xprt_connected(xprt),
2223 sock->sk->sk_state);
2a491991 2224 switch (status) {
f75e6745
TM		 2225 default:
2226 printk("%s: connect returned unhandled error %d\n",
2227 __func__, status);
2228 case -EADDRNOTAVAIL:
2229 /* We're probably in TIME_WAIT. Get rid of existing socket,
2230 * and retry
2231 */
a519fc7a 2232 xs_tcp_force_close(xprt);
88b5ed73 2233 break;
2a491991
TM		 2234 case 0:
2235 case -EINPROGRESS:
2236 case -EALREADY:
7d1e8255 2237 xprt_clear_connecting(xprt);
5cf02d09 2238 current->flags &= ~PF_FSTRANS;
7d1e8255 2239 return;
9fcfe0c8
TM		 2240 case -EINVAL:
2241 /* Happens, for instance, if the user specified a link
2242 * local IPv6 address without a scope-id.
2243 */
3ed5e2a2
TM		 2244 case -ECONNREFUSED:
2245 case -ECONNRESET:
2246 case -ENETUNREACH:
2247 /* retry with existing socket, after a delay */
9fcfe0c8 2248 goto out;
a246b010 2249 }
7d1e8255 2250out_eagain:
2a491991 2251 status = -EAGAIN;
a246b010 2252out:
68e220bd 2253 xprt_clear_connecting(xprt);
7d1e8255 2254 xprt_wake_pending_tasks(xprt, status);
5cf02d09 2255 current->flags &= ~PF_FSTRANS;
68e220bd 2256}
b0d93ad5 2257
9903cd1c
CL		 2258/**
2259 * xs_connect - connect a socket to a remote endpoint
1b092092 2260 * @xprt: pointer to transport structure
9903cd1c
CL		 2261 * @task: address of RPC task that manages state of connect request
2262 *
2263 * TCP: If the remote end dropped the connection, delay reconnecting.
03bf4b70
CL		 2264 *
2265 * UDP socket connects are synchronous, but we use a work queue anyway
2266 * to guarantee that even unprivileged user processes can set up a
2267 * socket on a privileged port.
2268 *
2269 * If a UDP socket connect fails, the delay behavior here prevents
2270 * retry floods (hard mounts).
9903cd1c 2271 */
1b092092 2272static void xs_connect(struct rpc_xprt *xprt, struct rpc_task *task)
a246b010 2273{
ee0ac0c2 2274 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
a246b010 2275
09a21c41 2276 if (transport->sock != NULL && !RPC_IS_SOFTCONN(task)) {
46121cf7
CL		 2277 dprintk("RPC: xs_connect delayed xprt %p for %lu "
2278 "seconds\n",
03bf4b70 2279 xprt, xprt->reestablish_timeout / HZ);
c1384c9c
TM		 2280 queue_delayed_work(rpciod_workqueue,
2281 &transport->connect_worker,
2282 xprt->reestablish_timeout);
03bf4b70 2283 xprt->reestablish_timeout <<= 1;
61d0a8e6
NB		 2284 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2285 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
03bf4b70
CL		 2286 if (xprt->reestablish_timeout > XS_TCP_MAX_REEST_TO)
2287 xprt->reestablish_timeout = XS_TCP_MAX_REEST_TO;
b0d93ad5 2288 } else {
46121cf7 2289 dprintk("RPC: xs_connect scheduled xprt %p\n", xprt);
c1384c9c
TM		 2290 queue_delayed_work(rpciod_workqueue,
2291 &transport->connect_worker, 0);
a246b010
CL		 2292 }
2293}
2294
176e21ee
CL		 2295/**
2296 * xs_local_print_stats - display AF_LOCAL socket-specifc stats
2297 * @xprt: rpc_xprt struct containing statistics
2298 * @seq: output file
2299 *
2300 */
2301static void xs_local_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2302{
2303 long idle_time = 0;
2304
2305 if (xprt_connected(xprt))
2306 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2307
2308 seq_printf(seq, "\txprt:\tlocal %lu %lu %lu %ld %lu %lu %lu "
15a45206 2309 "%llu %llu %lu %llu %llu\n",
176e21ee
CL		 2310 xprt->stat.bind_count,
2311 xprt->stat.connect_count,
2312 xprt->stat.connect_time,
2313 idle_time,
2314 xprt->stat.sends,
2315 xprt->stat.recvs,
2316 xprt->stat.bad_xids,
2317 xprt->stat.req_u,
15a45206
AA		 2318 xprt->stat.bklog_u,
2319 xprt->stat.max_slots,
2320 xprt->stat.sending_u,
2321 xprt->stat.pending_u);
176e21ee
CL		 2322}
2323
262ca07d
CL		 2324/**
2325 * xs_udp_print_stats - display UDP socket-specifc stats
2326 * @xprt: rpc_xprt struct containing statistics
2327 * @seq: output file
2328 *
2329 */
2330static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2331{
c8475461
CL		 2332 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2333
15a45206
AA		 2334 seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %llu %llu "
2335 "%lu %llu %llu\n",
fbfffbd5 2336 transport->srcport,
262ca07d
CL		 2337 xprt->stat.bind_count,
2338 xprt->stat.sends,
2339 xprt->stat.recvs,
2340 xprt->stat.bad_xids,
2341 xprt->stat.req_u,
15a45206
AA		 2342 xprt->stat.bklog_u,
2343 xprt->stat.max_slots,
2344 xprt->stat.sending_u,
2345 xprt->stat.pending_u);
262ca07d
CL		 2346}
2347
2348/**
2349 * xs_tcp_print_stats - display TCP socket-specifc stats
2350 * @xprt: rpc_xprt struct containing statistics
2351 * @seq: output file
2352 *
2353 */
2354static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2355{
c8475461 2356 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
262ca07d
CL		 2357 long idle_time = 0;
2358
2359 if (xprt_connected(xprt))
2360 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2361
15a45206
AA		 2362 seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu "
2363 "%llu %llu %lu %llu %llu\n",
fbfffbd5 2364 transport->srcport,
262ca07d
CL		 2365 xprt->stat.bind_count,
2366 xprt->stat.connect_count,
2367 xprt->stat.connect_time,
2368 idle_time,
2369 xprt->stat.sends,
2370 xprt->stat.recvs,
2371 xprt->stat.bad_xids,
2372 xprt->stat.req_u,
15a45206
AA		 2373 xprt->stat.bklog_u,
2374 xprt->stat.max_slots,
2375 xprt->stat.sending_u,
2376 xprt->stat.pending_u);
262ca07d
CL		 2377}
2378
4cfc7e60
RI		 2379/*
2380 * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason
2381 * we allocate pages instead doing a kmalloc like rpc_malloc is because we want
2382 * to use the server side send routines.
2383 */
5a51f13a 2384static void *bc_malloc(struct rpc_task *task, size_t size)
4cfc7e60
RI		 2385{
2386 struct page *page;
2387 struct rpc_buffer *buf;
2388
b8a13d03
WAA		 2389 WARN_ON_ONCE(size > PAGE_SIZE - sizeof(struct rpc_buffer));
2390 if (size > PAGE_SIZE - sizeof(struct rpc_buffer))
2391 return NULL;
4cfc7e60 2392
b8a13d03 2393 page = alloc_page(GFP_KERNEL);
4cfc7e60
RI		 2394 if (!page)
2395 return NULL;
2396
2397 buf = page_address(page);
2398 buf->len = PAGE_SIZE;
2399
2400 return buf->data;
2401}
2402
2403/*
2404 * Free the space allocated in the bc_alloc routine
2405 */
5a51f13a 2406static void bc_free(void *buffer)
4cfc7e60
RI		 2407{
2408 struct rpc_buffer *buf;
2409
2410 if (!buffer)
2411 return;
2412
2413 buf = container_of(buffer, struct rpc_buffer, data);
2414 free_page((unsigned long)buf);
2415}
2416
2417/*
2418 * Use the svc_sock to send the callback. Must be called with svsk->sk_mutex
2419 * held. Borrows heavily from svc_tcp_sendto and xs_tcp_send_request.
2420 */
2421static int bc_sendto(struct rpc_rqst *req)
2422{
2423 int len;
2424 struct xdr_buf *xbufp = &req->rq_snd_buf;
2425 struct rpc_xprt *xprt = req->rq_xprt;
2426 struct sock_xprt *transport =
2427 container_of(xprt, struct sock_xprt, xprt);
2428 struct socket *sock = transport->sock;
2429 unsigned long headoff;
2430 unsigned long tailoff;
2431
61677eee 2432 xs_encode_stream_record_marker(xbufp);
4cfc7e60
RI		 2433
2434 tailoff = (unsigned long)xbufp->tail[0].iov_base & ~PAGE_MASK;
2435 headoff = (unsigned long)xbufp->head[0].iov_base & ~PAGE_MASK;
2436 len = svc_send_common(sock, xbufp,
2437 virt_to_page(xbufp->head[0].iov_base), headoff,
2438 xbufp->tail[0].iov_base, tailoff);
2439
2440 if (len != xbufp->len) {
2441 printk(KERN_NOTICE "Error sending entire callback!\n");
2442 len = -EAGAIN;
2443 }
2444
2445 return len;
2446}
2447
2448/*
2449 * The send routine. Borrows from svc_send
2450 */
2451static int bc_send_request(struct rpc_task *task)
2452{
2453 struct rpc_rqst *req = task->tk_rqstp;
2454 struct svc_xprt *xprt;
4cfc7e60
RI		 2455 u32 len;
2456
2457 dprintk("sending request with xid: %08x\n", ntohl(req->rq_xid));
2458 /*
2459 * Get the server socket associated with this callback xprt
2460 */
2461 xprt = req->rq_xprt->bc_xprt;
4cfc7e60
RI		 2462
2463 /*
2464 * Grab the mutex to serialize data as the connection is shared
2465 * with the fore channel
2466 */
2467 if (!mutex_trylock(&xprt->xpt_mutex)) {
2468 rpc_sleep_on(&xprt->xpt_bc_pending, task, NULL);
2469 if (!mutex_trylock(&xprt->xpt_mutex))
2470 return -EAGAIN;
2471 rpc_wake_up_queued_task(&xprt->xpt_bc_pending, task);
2472 }
2473 if (test_bit(XPT_DEAD, &xprt->xpt_flags))
2474 len = -ENOTCONN;
2475 else
2476 len = bc_sendto(req);
2477 mutex_unlock(&xprt->xpt_mutex);
2478
2479 if (len > 0)
2480 len = 0;
2481
2482 return len;
2483}
2484
2485/*
2486 * The close routine. Since this is client initiated, we do nothing
2487 */
2488
2489static void bc_close(struct rpc_xprt *xprt)
2490{
4cfc7e60
RI		 2491}
2492
2493/*
2494 * The xprt destroy routine. Again, because this connection is client
2495 * initiated, we do nothing
2496 */
2497
2498static void bc_destroy(struct rpc_xprt *xprt)
2499{
4cfc7e60
RI		 2500}
2501
176e21ee
CL		 2502static struct rpc_xprt_ops xs_local_ops = {
2503 .reserve_xprt = xprt_reserve_xprt,
2504 .release_xprt = xs_tcp_release_xprt,
f39c1bfb 2505 .alloc_slot = xprt_alloc_slot,
176e21ee
CL		 2506 .rpcbind = xs_local_rpcbind,
2507 .set_port = xs_local_set_port,
dc107402 2508 .connect = xs_local_connect,
176e21ee
CL		 2509 .buf_alloc = rpc_malloc,
2510 .buf_free = rpc_free,
2511 .send_request = xs_local_send_request,
2512 .set_retrans_timeout = xprt_set_retrans_timeout_def,
2513 .close = xs_close,
190b1ecf 2514 .destroy = xs_local_destroy,
176e21ee
CL		 2515 .print_stats = xs_local_print_stats,
2516};
2517
262965f5 2518static struct rpc_xprt_ops xs_udp_ops = {
43118c29 2519 .set_buffer_size = xs_udp_set_buffer_size,
12a80469 2520 .reserve_xprt = xprt_reserve_xprt_cong,
49e9a890 2521 .release_xprt = xprt_release_xprt_cong,
f39c1bfb 2522 .alloc_slot = xprt_alloc_slot,
45160d62 2523 .rpcbind = rpcb_getport_async,
92200412 2524 .set_port = xs_set_port,
262965f5 2525 .connect = xs_connect,
02107148
CL		 2526 .buf_alloc = rpc_malloc,
2527 .buf_free = rpc_free,
262965f5 2528 .send_request = xs_udp_send_request,
fe3aca29 2529 .set_retrans_timeout = xprt_set_retrans_timeout_rtt,
46c0ee8b 2530 .timer = xs_udp_timer,
a58dd398 2531 .release_request = xprt_release_rqst_cong,
262965f5
CL		 2532 .close = xs_close,
2533 .destroy = xs_destroy,
262ca07d 2534 .print_stats = xs_udp_print_stats,
262965f5
CL		 2535};
2536
2537static struct rpc_xprt_ops xs_tcp_ops = {
12a80469 2538 .reserve_xprt = xprt_reserve_xprt,
e0ab53de 2539 .release_xprt = xs_tcp_release_xprt,
f39c1bfb 2540 .alloc_slot = xprt_lock_and_alloc_slot,
45160d62 2541 .rpcbind = rpcb_getport_async,
92200412 2542 .set_port = xs_set_port,
0b9e7943 2543 .connect = xs_connect,
02107148
CL		 2544 .buf_alloc = rpc_malloc,
2545 .buf_free = rpc_free,
262965f5 2546 .send_request = xs_tcp_send_request,
fe3aca29 2547 .set_retrans_timeout = xprt_set_retrans_timeout_def,
f75e6745 2548 .close = xs_tcp_close,
9903cd1c 2549 .destroy = xs_destroy,
262ca07d 2550 .print_stats = xs_tcp_print_stats,
a246b010
CL		 2551};
2552
4cfc7e60
RI		 2553/*
2554 * The rpc_xprt_ops for the server backchannel
2555 */
2556
2557static struct rpc_xprt_ops bc_tcp_ops = {
2558 .reserve_xprt = xprt_reserve_xprt,
2559 .release_xprt = xprt_release_xprt,
84e28a30 2560 .alloc_slot = xprt_alloc_slot,
4cfc7e60
RI		 2561 .buf_alloc = bc_malloc,
2562 .buf_free = bc_free,
2563 .send_request = bc_send_request,
2564 .set_retrans_timeout = xprt_set_retrans_timeout_def,
2565 .close = bc_close,
2566 .destroy = bc_destroy,
2567 .print_stats = xs_tcp_print_stats,
2568};
2569
92476850
CL		 2570static int xs_init_anyaddr(const int family, struct sockaddr *sap)
2571{
2572 static const struct sockaddr_in sin = {
2573 .sin_family = AF_INET,
2574 .sin_addr.s_addr = htonl(INADDR_ANY),
2575 };
2576 static const struct sockaddr_in6 sin6 = {
2577 .sin6_family = AF_INET6,
2578 .sin6_addr = IN6ADDR_ANY_INIT,
2579 };
2580
2581 switch (family) {
176e21ee
CL		 2582 case AF_LOCAL:
2583 break;
92476850
CL		 2584 case AF_INET:
2585 memcpy(sap, &sin, sizeof(sin));
2586 break;
2587 case AF_INET6:
2588 memcpy(sap, &sin6, sizeof(sin6));
2589 break;
2590 default:
2591 dprintk("RPC: %s: Bad address family\n", __func__);
2592 return -EAFNOSUPPORT;
2593 }
2594 return 0;
2595}
2596
3c341b0b 2597static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
d9ba131d
TM		 2598 unsigned int slot_table_size,
2599 unsigned int max_slot_table_size)
c8541ecd
CL		 2600{
2601 struct rpc_xprt *xprt;
ffc2e518 2602 struct sock_xprt *new;
c8541ecd 2603
96802a09 2604 if (args->addrlen > sizeof(xprt->addr)) {
46121cf7 2605 dprintk("RPC: xs_setup_xprt: address too large\n");
c8541ecd
CL		 2606 return ERR_PTR(-EBADF);
2607 }
2608
d9ba131d
TM		 2609 xprt = xprt_alloc(args->net, sizeof(*new), slot_table_size,
2610 max_slot_table_size);
bd1722d4 2611 if (xprt == NULL) {
46121cf7
CL		 2612 dprintk("RPC: xs_setup_xprt: couldn't allocate "
2613 "rpc_xprt\n");
c8541ecd
CL		 2614 return ERR_PTR(-ENOMEM);
2615 }
c8541ecd 2616
bd1722d4 2617 new = container_of(xprt, struct sock_xprt, xprt);
96802a09
FM		 2618 memcpy(&xprt->addr, args->dstaddr, args->addrlen);
2619 xprt->addrlen = args->addrlen;
d3bc9a1d 2620 if (args->srcaddr)
fbfffbd5 2621 memcpy(&new->srcaddr, args->srcaddr, args->addrlen);
92476850
CL		 2622 else {
2623 int err;
2624 err = xs_init_anyaddr(args->dstaddr->sa_family,
2625 (struct sockaddr *)&new->srcaddr);
2aa13531
SK		 2626 if (err != 0) {
2627 xprt_free(xprt);
92476850 2628 return ERR_PTR(err);
2aa13531 2629 }
92476850 2630 }
c8541ecd
CL		 2631
2632 return xprt;
2633}
2634
176e21ee
CL		 2635static const struct rpc_timeout xs_local_default_timeout = {
2636 .to_initval = 10 * HZ,
2637 .to_maxval = 10 * HZ,
2638 .to_retries = 2,
2639};
2640
2641/**
2642 * xs_setup_local - Set up transport to use an AF_LOCAL socket
2643 * @args: rpc transport creation arguments
2644 *
2645 * AF_LOCAL is a "tpi_cots_ord" transport, just like TCP
2646 */
2647static struct rpc_xprt *xs_setup_local(struct xprt_create *args)
2648{
2649 struct sockaddr_un *sun = (struct sockaddr_un *)args->dstaddr;
2650 struct sock_xprt *transport;
2651 struct rpc_xprt *xprt;
2652 struct rpc_xprt *ret;
2653
d9ba131d
TM		 2654 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2655 xprt_max_tcp_slot_table_entries);
176e21ee
CL		 2656 if (IS_ERR(xprt))
2657 return xprt;
2658 transport = container_of(xprt, struct sock_xprt, xprt);
2659
2660 xprt->prot = 0;
2661 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2662 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2663
2664 xprt->bind_timeout = XS_BIND_TO;
2665 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2666 xprt->idle_timeout = XS_IDLE_DISC_TO;
2667
2668 xprt->ops = &xs_local_ops;
2669 xprt->timeout = &xs_local_default_timeout;
2670
2671 switch (sun->sun_family) {
2672 case AF_LOCAL:
2673 if (sun->sun_path[0] != '/') {
2674 dprintk("RPC: bad AF_LOCAL address: %s\n",
2675 sun->sun_path);
2676 ret = ERR_PTR(-EINVAL);
2677 goto out_err;
2678 }
2679 xprt_set_bound(xprt);
176e21ee 2680 xs_format_peer_addresses(xprt, "local", RPCBIND_NETID_LOCAL);
7073ea87
BF		 2681 ret = ERR_PTR(xs_local_setup_socket(transport));
2682 if (ret)
2683 goto out_err;
176e21ee
CL		 2684 break;
2685 default:
2686 ret = ERR_PTR(-EAFNOSUPPORT);
2687 goto out_err;
2688 }
2689
2690 dprintk("RPC: set up xprt to %s via AF_LOCAL\n",
2691 xprt->address_strings[RPC_DISPLAY_ADDR]);
2692
2693 if (try_module_get(THIS_MODULE))
2694 return xprt;
2695 ret = ERR_PTR(-EINVAL);
2696out_err:
2697 xprt_free(xprt);
2698 return ret;
2699}
2700
2881ae74
TM		 2701static const struct rpc_timeout xs_udp_default_timeout = {
2702 .to_initval = 5 * HZ,
2703 .to_maxval = 30 * HZ,
2704 .to_increment = 5 * HZ,
2705 .to_retries = 5,
2706};
2707
9903cd1c
CL		 2708/**
2709 * xs_setup_udp - Set up transport to use a UDP socket
96802a09 2710 * @args: rpc transport creation arguments
9903cd1c
CL		 2711 *
2712 */
483066d6 2713static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
a246b010 2714{
8f9d5b1a 2715 struct sockaddr *addr = args->dstaddr;
c8541ecd 2716 struct rpc_xprt *xprt;
c8475461 2717 struct sock_xprt *transport;
0a68b0be 2718 struct rpc_xprt *ret;
a246b010 2719
d9ba131d
TM		 2720 xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries,
2721 xprt_udp_slot_table_entries);
c8541ecd
CL		 2722 if (IS_ERR(xprt))
2723 return xprt;
c8475461 2724 transport = container_of(xprt, struct sock_xprt, xprt);
a246b010 2725
ec739ef0 2726 xprt->prot = IPPROTO_UDP;
808012fb 2727 xprt->tsh_size = 0;
a246b010
CL		 2728 /* XXX: header size can vary due to auth type, IPv6, etc. */
2729 xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
2730
03bf4b70 2731 xprt->bind_timeout = XS_BIND_TO;
03bf4b70
CL		 2732 xprt->reestablish_timeout = XS_UDP_REEST_TO;
2733 xprt->idle_timeout = XS_IDLE_DISC_TO;
a246b010 2734
262965f5 2735 xprt->ops = &xs_udp_ops;
a246b010 2736
ba7392bb 2737 xprt->timeout = &xs_udp_default_timeout;
a246b010 2738
8f9d5b1a
CL		 2739 switch (addr->sa_family) {
2740 case AF_INET:
2741 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2742 xprt_set_bound(xprt);
2743
2744 INIT_DELAYED_WORK(&transport->connect_worker,
8c14ff2a 2745 xs_udp_setup_socket);
9dc3b095 2746 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
8f9d5b1a
CL		 2747 break;
2748 case AF_INET6:
2749 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2750 xprt_set_bound(xprt);
2751
2752 INIT_DELAYED_WORK(&transport->connect_worker,
8c14ff2a 2753 xs_udp_setup_socket);
9dc3b095 2754 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
8f9d5b1a
CL		 2755 break;
2756 default:
0a68b0be
BF		 2757 ret = ERR_PTR(-EAFNOSUPPORT);
2758 goto out_err;
8f9d5b1a
CL		 2759 }
2760
c740eff8
CL		 2761 if (xprt_bound(xprt))
2762 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2763 xprt->address_strings[RPC_DISPLAY_ADDR],
2764 xprt->address_strings[RPC_DISPLAY_PORT],
2765 xprt->address_strings[RPC_DISPLAY_PROTO]);
2766 else
2767 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
2768 xprt->address_strings[RPC_DISPLAY_ADDR],
2769 xprt->address_strings[RPC_DISPLAY_PROTO]);
edb267a6 2770
bc25571e
TT		 2771 if (try_module_get(THIS_MODULE))
2772 return xprt;
0a68b0be
BF		 2773 ret = ERR_PTR(-EINVAL);
2774out_err:
e204e621 2775 xprt_free(xprt);
0a68b0be 2776 return ret;
a246b010
CL		 2777}
2778
2881ae74
TM		 2779static const struct rpc_timeout xs_tcp_default_timeout = {
2780 .to_initval = 60 * HZ,
2781 .to_maxval = 60 * HZ,
2782 .to_retries = 2,
2783};
2784
9903cd1c
CL		 2785/**
2786 * xs_setup_tcp - Set up transport to use a TCP socket
96802a09 2787 * @args: rpc transport creation arguments
9903cd1c
CL		 2788 *
2789 */
483066d6 2790static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
a246b010 2791{
8f9d5b1a 2792 struct sockaddr *addr = args->dstaddr;
c8541ecd 2793 struct rpc_xprt *xprt;
c8475461 2794 struct sock_xprt *transport;
0a68b0be 2795 struct rpc_xprt *ret;
b7993ceb
TM		 2796 unsigned int max_slot_table_size = xprt_max_tcp_slot_table_entries;
2797
2798 if (args->flags & XPRT_CREATE_INFINITE_SLOTS)
2799 max_slot_table_size = RPC_MAX_SLOT_TABLE_LIMIT;
a246b010 2800
d9ba131d 2801 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
b7993ceb 2802 max_slot_table_size);
c8541ecd
CL		 2803 if (IS_ERR(xprt))
2804 return xprt;
c8475461 2805 transport = container_of(xprt, struct sock_xprt, xprt);
a246b010 2806
ec739ef0 2807 xprt->prot = IPPROTO_TCP;
808012fb 2808 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
808012fb 2809 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
a246b010 2810
03bf4b70 2811 xprt->bind_timeout = XS_BIND_TO;
03bf4b70
CL		 2812 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2813 xprt->idle_timeout = XS_IDLE_DISC_TO;
a246b010 2814
262965f5 2815 xprt->ops = &xs_tcp_ops;
ba7392bb 2816 xprt->timeout = &xs_tcp_default_timeout;
a246b010 2817
8f9d5b1a
CL		 2818 switch (addr->sa_family) {
2819 case AF_INET:
2820 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2821 xprt_set_bound(xprt);
2822
9dc3b095 2823 INIT_DELAYED_WORK(&transport->connect_worker,
cdd518d5 2824 xs_tcp_setup_socket);
9dc3b095 2825 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
8f9d5b1a
CL		 2826 break;
2827 case AF_INET6:
2828 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2829 xprt_set_bound(xprt);
2830
9dc3b095 2831 INIT_DELAYED_WORK(&transport->connect_worker,
cdd518d5 2832 xs_tcp_setup_socket);
9dc3b095 2833 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
8f9d5b1a
CL		 2834 break;
2835 default:
0a68b0be
BF		 2836 ret = ERR_PTR(-EAFNOSUPPORT);
2837 goto out_err;
8f9d5b1a
CL		 2838 }
2839
c740eff8
CL		 2840 if (xprt_bound(xprt))
2841 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2842 xprt->address_strings[RPC_DISPLAY_ADDR],
2843 xprt->address_strings[RPC_DISPLAY_PORT],
2844 xprt->address_strings[RPC_DISPLAY_PROTO]);
2845 else
2846 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
2847 xprt->address_strings[RPC_DISPLAY_ADDR],
2848 xprt->address_strings[RPC_DISPLAY_PROTO]);
2849
edb267a6 2850
bc25571e
TT		 2851 if (try_module_get(THIS_MODULE))
2852 return xprt;
0a68b0be
BF		 2853 ret = ERR_PTR(-EINVAL);
2854out_err:
e204e621 2855 xprt_free(xprt);
0a68b0be 2856 return ret;
a246b010 2857}
282b32e1 2858
f300baba
AB		 2859/**
2860 * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket
2861 * @args: rpc transport creation arguments
2862 *
2863 */
2864static struct rpc_xprt *xs_setup_bc_tcp(struct xprt_create *args)
2865{
2866 struct sockaddr *addr = args->dstaddr;
2867 struct rpc_xprt *xprt;
2868 struct sock_xprt *transport;
2869 struct svc_sock *bc_sock;
0a68b0be 2870 struct rpc_xprt *ret;
f300baba 2871
f0418aa4
BF		 2872 if (args->bc_xprt->xpt_bc_xprt) {
2873 /*
2874 * This server connection already has a backchannel
2875 * export; we can't create a new one, as we wouldn't be
2876 * able to match replies based on xid any more. So,
2877 * reuse the already-existing one:
2878 */
2879 return args->bc_xprt->xpt_bc_xprt;
2880 }
d9ba131d
TM		 2881 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2882 xprt_tcp_slot_table_entries);
f300baba
AB		 2883 if (IS_ERR(xprt))
2884 return xprt;
2885 transport = container_of(xprt, struct sock_xprt, xprt);
2886
2887 xprt->prot = IPPROTO_TCP;
2888 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2889 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2890 xprt->timeout = &xs_tcp_default_timeout;
2891
2892 /* backchannel */
2893 xprt_set_bound(xprt);
2894 xprt->bind_timeout = 0;
f300baba
AB		 2895 xprt->reestablish_timeout = 0;
2896 xprt->idle_timeout = 0;
2897
f300baba
AB		 2898 xprt->ops = &bc_tcp_ops;
2899
2900 switch (addr->sa_family) {
2901 case AF_INET:
2902 xs_format_peer_addresses(xprt, "tcp",
2903 RPCBIND_NETID_TCP);
2904 break;
2905 case AF_INET6:
2906 xs_format_peer_addresses(xprt, "tcp",
2907 RPCBIND_NETID_TCP6);
2908 break;
2909 default:
0a68b0be
BF		 2910 ret = ERR_PTR(-EAFNOSUPPORT);
2911 goto out_err;
f300baba
AB		 2912 }
2913
50fa0d40
PE		 2914 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2915 xprt->address_strings[RPC_DISPLAY_ADDR],
2916 xprt->address_strings[RPC_DISPLAY_PORT],
2917 xprt->address_strings[RPC_DISPLAY_PROTO]);
f300baba 2918
99de8ea9
BF		 2919 /*
2920 * Once we've associated a backchannel xprt with a connection,
2921 * we want to keep it around as long as long as the connection
2922 * lasts, in case we need to start using it for a backchannel
2923 * again; this reference won't be dropped until bc_xprt is
2924 * destroyed.
2925 */
2926 xprt_get(xprt);
2927 args->bc_xprt->xpt_bc_xprt = xprt;
2928 xprt->bc_xprt = args->bc_xprt;
2929 bc_sock = container_of(args->bc_xprt, struct svc_sock, sk_xprt);
2930 transport->sock = bc_sock->sk_sock;
2931 transport->inet = bc_sock->sk_sk;
2932
f300baba
AB		 2933 /*
2934 * Since we don't want connections for the backchannel, we set
2935 * the xprt status to connected
2936 */
2937 xprt_set_connected(xprt);
2938
2939
2940 if (try_module_get(THIS_MODULE))
2941 return xprt;
99de8ea9 2942 xprt_put(xprt);
0a68b0be
BF		 2943 ret = ERR_PTR(-EINVAL);
2944out_err:
e204e621 2945 xprt_free(xprt);
0a68b0be 2946 return ret;
f300baba
AB		 2947}
2948
176e21ee
CL		 2949static struct xprt_class xs_local_transport = {
2950 .list = LIST_HEAD_INIT(xs_local_transport.list),
2951 .name = "named UNIX socket",
2952 .owner = THIS_MODULE,
2953 .ident = XPRT_TRANSPORT_LOCAL,
2954 .setup = xs_setup_local,
2955};
2956
bc25571e
TT		 2957static struct xprt_class xs_udp_transport = {
2958 .list = LIST_HEAD_INIT(xs_udp_transport.list),
2959 .name = "udp",
2960 .owner = THIS_MODULE,
f300baba 2961 .ident = XPRT_TRANSPORT_UDP,
bc25571e
TT		 2962 .setup = xs_setup_udp,
2963};
2964
2965static struct xprt_class xs_tcp_transport = {
2966 .list = LIST_HEAD_INIT(xs_tcp_transport.list),
2967 .name = "tcp",
2968 .owner = THIS_MODULE,
f300baba 2969 .ident = XPRT_TRANSPORT_TCP,
bc25571e
TT		 2970 .setup = xs_setup_tcp,
2971};
2972
f300baba
AB		 2973static struct xprt_class xs_bc_tcp_transport = {
2974 .list = LIST_HEAD_INIT(xs_bc_tcp_transport.list),
2975 .name = "tcp NFSv4.1 backchannel",
2976 .owner = THIS_MODULE,
2977 .ident = XPRT_TRANSPORT_BC_TCP,
2978 .setup = xs_setup_bc_tcp,
2979};
2980
282b32e1 2981/**
bc25571e 2982 * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
282b32e1
CL		 2983 *
2984 */
2985int init_socket_xprt(void)
2986{
fbf76683 2987#ifdef RPC_DEBUG
2b1bec5f 2988 if (!sunrpc_table_header)
0b4d4147 2989 sunrpc_table_header = register_sysctl_table(sunrpc_table);
fbf76683
CL		 2990#endif
2991
176e21ee 2992 xprt_register_transport(&xs_local_transport);
bc25571e
TT		 2993 xprt_register_transport(&xs_udp_transport);
2994 xprt_register_transport(&xs_tcp_transport);
f300baba 2995 xprt_register_transport(&xs_bc_tcp_transport);
bc25571e 2996
282b32e1
CL		 2997 return 0;
2998}
2999
3000/**
bc25571e 3001 * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
282b32e1
CL		 3002 *
3003 */
3004void cleanup_socket_xprt(void)
3005{
fbf76683
CL		 3006#ifdef RPC_DEBUG
3007 if (sunrpc_table_header) {
3008 unregister_sysctl_table(sunrpc_table_header);
3009 sunrpc_table_header = NULL;
3010 }
3011#endif
bc25571e 3012
176e21ee 3013 xprt_unregister_transport(&xs_local_transport);
bc25571e
TT		 3014 xprt_unregister_transport(&xs_udp_transport);
3015 xprt_unregister_transport(&xs_tcp_transport);
f300baba 3016 xprt_unregister_transport(&xs_bc_tcp_transport);
282b32e1 3017}
cbf11071 3018
9bbb9e5a
RR		 3019static int param_set_uint_minmax(const char *val,
3020 const struct kernel_param *kp,
cbf11071
TM		 3021 unsigned int min, unsigned int max)
3022{
3023 unsigned long num;
3024 int ret;
3025
3026 if (!val)
3027 return -EINVAL;
3028 ret = strict_strtoul(val, 0, &num);
3029 if (ret == -EINVAL || num < min || num > max)
3030 return -EINVAL;
3031 *((unsigned int *)kp->arg) = num;
3032 return 0;
3033}
3034
9bbb9e5a 3035static int param_set_portnr(const char *val, const struct kernel_param *kp)
cbf11071
TM		 3036{
3037 return param_set_uint_minmax(val, kp,
3038 RPC_MIN_RESVPORT,
3039 RPC_MAX_RESVPORT);
3040}
3041
9bbb9e5a
RR		 3042static struct kernel_param_ops param_ops_portnr = {
3043 .set = param_set_portnr,
3044 .get = param_get_uint,
3045};
3046
cbf11071
TM		 3047#define param_check_portnr(name, p) \
3048 __param_check(name, p, unsigned int);
3049
3050module_param_named(min_resvport, xprt_min_resvport, portnr, 0644);
3051module_param_named(max_resvport, xprt_max_resvport, portnr, 0644);
3052
9bbb9e5a
RR		 3053static int param_set_slot_table_size(const char *val,
3054 const struct kernel_param *kp)
cbf11071
TM		 3055{
3056 return param_set_uint_minmax(val, kp,
3057 RPC_MIN_SLOT_TABLE,
3058 RPC_MAX_SLOT_TABLE);
3059}
3060
9bbb9e5a
RR		 3061static struct kernel_param_ops param_ops_slot_table_size = {
3062 .set = param_set_slot_table_size,
3063 .get = param_get_uint,
3064};
3065
cbf11071
TM		 3066#define param_check_slot_table_size(name, p) \
3067 __param_check(name, p, unsigned int);
3068
d9ba131d
TM		 3069static int param_set_max_slot_table_size(const char *val,
3070 const struct kernel_param *kp)
3071{
3072 return param_set_uint_minmax(val, kp,
3073 RPC_MIN_SLOT_TABLE,
3074 RPC_MAX_SLOT_TABLE_LIMIT);
3075}
3076
3077static struct kernel_param_ops param_ops_max_slot_table_size = {
3078 .set = param_set_max_slot_table_size,
3079 .get = param_get_uint,
3080};
3081
3082#define param_check_max_slot_table_size(name, p) \
3083 __param_check(name, p, unsigned int);
3084
cbf11071
TM		 3085module_param_named(tcp_slot_table_entries, xprt_tcp_slot_table_entries,
3086 slot_table_size, 0644);
d9ba131d
TM		 3087module_param_named(tcp_max_slot_table_entries, xprt_max_tcp_slot_table_entries,
3088 max_slot_table_size, 0644);
cbf11071
TM		 3089module_param_named(udp_slot_table_entries, xprt_udp_slot_table_entries,
3090 slot_table_size, 0644);
3091
Ubuntu Artful kernel mirror