projects / linux-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
SUNRPC: Handle connect errors ECONNABORTED and EHOSTUNREACH
[linux-block.git] / net / sunrpc / xprtsock.c
CommitLineData
a246b010
CL		 1/*
2 * linux/net/sunrpc/xprtsock.c
3 *
4 * Client-side transport implementation for sockets.
5 *
113aa838
AC		 6 * TCP callback races fixes (C) 1998 Red Hat
7 * TCP send fixes (C) 1998 Red Hat
a246b010
CL		 8 * TCP NFS related read + write fixes
9 * (C) 1999 Dave Airlie, University of Limerick, Ireland <airlied@linux.ie>
10 *
11 * Rewrite of larges part of the code in order to stabilize TCP stuff.
12 * Fix behaviour when socket buffer is full.
13 * (C) 1999 Trond Myklebust <trond.myklebust@fys.uio.no>
55aa4f58
CL		 14 *
15 * IP socket transport implementation, (C) 2005 Chuck Lever <cel@netapp.com>
8f9d5b1a
CL		 16 *
17 * IPv6 support contributed by Gilles Quillard, Bull Open Source, 2005.
18 * <gilles.quillard@bull.net>
a246b010
CL		 19 */
20
21#include <linux/types.h>
176e21ee 22#include <linux/string.h>
a246b010 23#include <linux/slab.h>
bc25571e 24#include <linux/module.h>
a246b010 25#include <linux/capability.h>
a246b010
CL		 26#include <linux/pagemap.h>
27#include <linux/errno.h>
28#include <linux/socket.h>
29#include <linux/in.h>
30#include <linux/net.h>
31#include <linux/mm.h>
176e21ee 32#include <linux/un.h>
a246b010
CL		 33#include <linux/udp.h>
34#include <linux/tcp.h>
35#include <linux/sunrpc/clnt.h>
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
a6b31d18 396static int xs_send_pagedata(struct socket *sock, struct xdr_buf *xdr, unsigned int base, int more, bool zerocopy)
b4b5cc85 397{
a6b31d18
TM		 398 ssize_t (*do_sendpage)(struct socket *sock, struct page *page,
399 int offset, size_t size, int flags);
24c5684b
TM		 400 struct page **ppage;
401 unsigned int remainder;
402 int err, sent = 0;
403
404 remainder = xdr->page_len - base;
405 base += xdr->page_base;
406 ppage = xdr->pages + (base >> PAGE_SHIFT);
407 base &= ~PAGE_MASK;
a6b31d18
TM		 408 do_sendpage = sock->ops->sendpage;
409 if (!zerocopy)
410 do_sendpage = sock_no_sendpage;
24c5684b
TM		 411 for(;;) {
412 unsigned int len = min_t(unsigned int, PAGE_SIZE - base, remainder);
413 int flags = XS_SENDMSG_FLAGS;
b4b5cc85 414
24c5684b
TM		 415 remainder -= len;
416 if (remainder != 0 || more)
417 flags |= MSG_MORE;
a6b31d18 418 err = do_sendpage(sock, *ppage, base, len, flags);
24c5684b
TM		 419 if (remainder == 0 || err != len)
420 break;
421 sent += err;
422 ppage++;
423 base = 0;
424 }
425 if (sent == 0)
426 return err;
427 if (err > 0)
428 sent += err;
429 return sent;
b4b5cc85
CL		 430}
431
9903cd1c
CL		 432/**
433 * xs_sendpages - write pages directly to a socket
434 * @sock: socket to send on
435 * @addr: UDP only -- address of destination
436 * @addrlen: UDP only -- length of destination address
437 * @xdr: buffer containing this request
438 * @base: starting position in the buffer
a6b31d18 439 * @zerocopy: true if it is safe to use sendpage()
9903cd1c 440 *
a246b010 441 */
a6b31d18 442static int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base, bool zerocopy)
a246b010 443{
24c5684b
TM		 444 unsigned int remainder = xdr->len - base;
445 int err, sent = 0;
a246b010 446
262965f5 447 if (unlikely(!sock))
fba91afb 448 return -ENOTSOCK;
262965f5
CL		 449
450 clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags);
24c5684b
TM		 451 if (base != 0) {
452 addr = NULL;
453 addrlen = 0;
454 }
262965f5 455
24c5684b
TM		 456 if (base < xdr->head[0].iov_len || addr != NULL) {
457 unsigned int len = xdr->head[0].iov_len - base;
458 remainder -= len;
459 err = xs_send_kvec(sock, addr, addrlen, &xdr->head[0], base, remainder != 0);
460 if (remainder == 0 || err != len)
a246b010 461 goto out;
24c5684b 462 sent += err;
a246b010
CL		 463 base = 0;
464 } else
24c5684b 465 base -= xdr->head[0].iov_len;
a246b010 466
24c5684b
TM		 467 if (base < xdr->page_len) {
468 unsigned int len = xdr->page_len - base;
469 remainder -= len;
a6b31d18 470 err = xs_send_pagedata(sock, xdr, base, remainder != 0, zerocopy);
24c5684b 471 if (remainder == 0 || err != len)
a246b010 472 goto out;
24c5684b 473 sent += err;
a246b010 474 base = 0;
24c5684b
TM		 475 } else
476 base -= xdr->page_len;
477
478 if (base >= xdr->tail[0].iov_len)
479 return sent;
480 err = xs_send_kvec(sock, NULL, 0, &xdr->tail[0], base, 0);
a246b010 481out:
24c5684b
TM		 482 if (sent == 0)
483 return err;
484 if (err > 0)
485 sent += err;
486 return sent;
a246b010
CL		 487}
488
b6ddf64f
TM		 489static void xs_nospace_callback(struct rpc_task *task)
490{
491 struct sock_xprt *transport = container_of(task->tk_rqstp->rq_xprt, struct sock_xprt, xprt);
492
493 transport->inet->sk_write_pending--;
494 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
495}
496
9903cd1c 497/**
262965f5
CL		 498 * xs_nospace - place task on wait queue if transmit was incomplete
499 * @task: task to put to sleep
9903cd1c 500 *
a246b010 501 */
5e3771ce 502static int xs_nospace(struct rpc_task *task)
a246b010 503{
262965f5
CL		 504 struct rpc_rqst *req = task->tk_rqstp;
505 struct rpc_xprt *xprt = req->rq_xprt;
ee0ac0c2 506 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
24ca9a84 507 int ret = -EAGAIN;
a246b010 508
46121cf7 509 dprintk("RPC: %5u xmit incomplete (%u left of %u)\n",
262965f5
CL		 510 task->tk_pid, req->rq_slen - req->rq_bytes_sent,
511 req->rq_slen);
512
b6ddf64f
TM		 513 /* Protect against races with write_space */
514 spin_lock_bh(&xprt->transport_lock);
515
516 /* Don't race with disconnect */
517 if (xprt_connected(xprt)) {
518 if (test_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags)) {
519 /*
520 * Notify TCP that we're limited by the application
521 * window size
522 */
523 set_bit(SOCK_NOSPACE, &transport->sock->flags);
524 transport->inet->sk_write_pending++;
525 /* ...and wait for more buffer space */
526 xprt_wait_for_buffer_space(task, xs_nospace_callback);
527 }
528 } else {
529 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
5e3771ce 530 ret = -ENOTCONN;
b6ddf64f 531 }
262965f5 532
b6ddf64f 533 spin_unlock_bh(&xprt->transport_lock);
5e3771ce 534 return ret;
262965f5
CL		 535}
536
61677eee
CL		 537/*
538 * Construct a stream transport record marker in @buf.
539 */
540static inline void xs_encode_stream_record_marker(struct xdr_buf *buf)
541{
542 u32 reclen = buf->len - sizeof(rpc_fraghdr);
543 rpc_fraghdr *base = buf->head[0].iov_base;
544 *base = cpu_to_be32(RPC_LAST_STREAM_FRAGMENT | reclen);
545}
546
176e21ee
CL		 547/**
548 * xs_local_send_request - write an RPC request to an AF_LOCAL socket
549 * @task: RPC task that manages the state of an RPC request
550 *
551 * Return values:
552 * 0: The request has been sent
553 * EAGAIN: The socket was blocked, please call again later to
554 * complete the request
555 * ENOTCONN: Caller needs to invoke connect logic then call again
556 * other: Some other error occured, the request was not sent
557 */
558static int xs_local_send_request(struct rpc_task *task)
559{
560 struct rpc_rqst *req = task->tk_rqstp;
561 struct rpc_xprt *xprt = req->rq_xprt;
562 struct sock_xprt *transport =
563 container_of(xprt, struct sock_xprt, xprt);
564 struct xdr_buf *xdr = &req->rq_snd_buf;
565 int status;
566
567 xs_encode_stream_record_marker(&req->rq_snd_buf);
568
569 xs_pktdump("packet data:",
570 req->rq_svec->iov_base, req->rq_svec->iov_len);
571
572 status = xs_sendpages(transport->sock, NULL, 0,
a6b31d18 573 xdr, req->rq_bytes_sent, true);
176e21ee
CL		 574 dprintk("RPC: %s(%u) = %d\n",
575 __func__, xdr->len - req->rq_bytes_sent, status);
576 if (likely(status >= 0)) {
577 req->rq_bytes_sent += status;
578 req->rq_xmit_bytes_sent += status;
579 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
580 req->rq_bytes_sent = 0;
581 return 0;
582 }
583 status = -EAGAIN;
584 }
585
586 switch (status) {
587 case -EAGAIN:
588 status = xs_nospace(task);
589 break;
590 default:
591 dprintk("RPC: sendmsg returned unrecognized error %d\n",
592 -status);
593 case -EPIPE:
594 xs_close(xprt);
595 status = -ENOTCONN;
596 }
597
598 return status;
599}
600
262965f5
CL		 601/**
602 * xs_udp_send_request - write an RPC request to a UDP socket
603 * @task: address of RPC task that manages the state of an RPC request
604 *
605 * Return values:
606 * 0: The request has been sent
607 * EAGAIN: The socket was blocked, please call again later to
608 * complete the request
609 * ENOTCONN: Caller needs to invoke connect logic then call again
25985edc 610 * other: Some other error occurred, the request was not sent
262965f5
CL		 611 */
612static int xs_udp_send_request(struct rpc_task *task)
613{
614 struct rpc_rqst *req = task->tk_rqstp;
615 struct rpc_xprt *xprt = req->rq_xprt;
ee0ac0c2 616 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
262965f5
CL		 617 struct xdr_buf *xdr = &req->rq_snd_buf;
618 int status;
a246b010 619
9903cd1c 620 xs_pktdump("packet data:",
a246b010
CL		 621 req->rq_svec->iov_base,
622 req->rq_svec->iov_len);
623
01d37c42
TM		 624 if (!xprt_bound(xprt))
625 return -ENOTCONN;
ee0ac0c2 626 status = xs_sendpages(transport->sock,
95392c59 627 xs_addr(xprt),
ee0ac0c2 628 xprt->addrlen, xdr,
a6b31d18 629 req->rq_bytes_sent, true);
a246b010 630
46121cf7 631 dprintk("RPC: xs_udp_send_request(%u) = %d\n",
262965f5 632 xdr->len - req->rq_bytes_sent, status);
a246b010 633
2199700f 634 if (status >= 0) {
d60dbb20 635 req->rq_xmit_bytes_sent += status;
2199700f
TM		 636 if (status >= req->rq_slen)
637 return 0;
638 /* Still some bytes left; set up for a retry later. */
262965f5 639 status = -EAGAIN;
2199700f 640 }
a246b010 641
262965f5 642 switch (status) {
fba91afb
TM		 643 case -ENOTSOCK:
644 status = -ENOTCONN;
645 /* Should we call xs_close() here? */
646 break;
b6ddf64f 647 case -EAGAIN:
5e3771ce 648 status = xs_nospace(task);
b6ddf64f 649 break;
c8485e4d
TM		 650 default:
651 dprintk("RPC: sendmsg returned unrecognized error %d\n",
652 -status);
262965f5
CL		 653 case -ENETUNREACH:
654 case -EPIPE:
a246b010
CL		 655 case -ECONNREFUSED:
656 /* When the server has died, an ICMP port unreachable message
9903cd1c 657 * prompts ECONNREFUSED. */
b6ddf64f 658 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
a246b010 659 }
5fe46e9d 660
262965f5 661 return status;
a246b010
CL		 662}
663
e06799f9
TM		 664/**
665 * xs_tcp_shutdown - gracefully shut down a TCP socket
666 * @xprt: transport
667 *
668 * Initiates a graceful shutdown of the TCP socket by calling the
669 * equivalent of shutdown(SHUT_WR);
670 */
671static void xs_tcp_shutdown(struct rpc_xprt *xprt)
672{
673 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
674 struct socket *sock = transport->sock;
675
40b5ea0c 676 if (sock != NULL) {
e06799f9 677 kernel_sock_shutdown(sock, SHUT_WR);
40b5ea0c
TM		 678 trace_rpc_socket_shutdown(xprt, sock);
679 }
e06799f9
TM		 680}
681
9903cd1c 682/**
262965f5 683 * xs_tcp_send_request - write an RPC request to a TCP socket
9903cd1c
CL		 684 * @task: address of RPC task that manages the state of an RPC request
685 *
686 * Return values:
262965f5
CL		 687 * 0: The request has been sent
688 * EAGAIN: The socket was blocked, please call again later to
689 * complete the request
690 * ENOTCONN: Caller needs to invoke connect logic then call again
25985edc 691 * other: Some other error occurred, the request was not sent
9903cd1c
CL		 692 *
693 * XXX: In the case of soft timeouts, should we eventually give up
262965f5 694 * if sendmsg is not able to make progress?
9903cd1c 695 */
262965f5 696static int xs_tcp_send_request(struct rpc_task *task)
a246b010
CL		 697{
698 struct rpc_rqst *req = task->tk_rqstp;
699 struct rpc_xprt *xprt = req->rq_xprt;
ee0ac0c2 700 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
262965f5 701 struct xdr_buf *xdr = &req->rq_snd_buf;
a6b31d18 702 bool zerocopy = true;
b595bb15 703 int status;
a246b010 704
61677eee 705 xs_encode_stream_record_marker(&req->rq_snd_buf);
a246b010 706
262965f5
CL		 707 xs_pktdump("packet data:",
708 req->rq_svec->iov_base,
709 req->rq_svec->iov_len);
a6b31d18
TM		 710 /* Don't use zero copy if this is a resend. If the RPC call
711 * completes while the socket holds a reference to the pages,
712 * then we may end up resending corrupted data.
713 */
714 if (task->tk_flags & RPC_TASK_SENT)
715 zerocopy = false;
a246b010
CL		 716
717 /* Continue transmitting the packet/record. We must be careful
718 * to cope with writespace callbacks arriving _after_ we have
262965f5 719 * called sendmsg(). */
a246b010 720 while (1) {
ee0ac0c2 721 status = xs_sendpages(transport->sock,
a6b31d18
TM		 722 NULL, 0, xdr, req->rq_bytes_sent,
723 zerocopy);
a246b010 724
46121cf7 725 dprintk("RPC: xs_tcp_send_request(%u) = %d\n",
262965f5 726 xdr->len - req->rq_bytes_sent, status);
a246b010 727
262965f5 728 if (unlikely(status < 0))
a246b010 729 break;
a246b010 730
262965f5
CL		 731 /* If we've sent the entire packet, immediately
732 * reset the count of bytes sent. */
733 req->rq_bytes_sent += status;
d60dbb20 734 req->rq_xmit_bytes_sent += status;
262965f5
CL		 735 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
736 req->rq_bytes_sent = 0;
737 return 0;
738 }
a246b010 739
06b4b681
TM		 740 if (status != 0)
741 continue;
a246b010 742 status = -EAGAIN;
06b4b681 743 break;
a246b010
CL		 744 }
745
262965f5 746 switch (status) {
fba91afb
TM		 747 case -ENOTSOCK:
748 status = -ENOTCONN;
749 /* Should we call xs_close() here? */
750 break;
262965f5 751 case -EAGAIN:
5e3771ce 752 status = xs_nospace(task);
262965f5 753 break;
c8485e4d
TM		 754 default:
755 dprintk("RPC: sendmsg returned unrecognized error %d\n",
756 -status);
262965f5 757 case -ECONNRESET:
2a9e1cfa
TM		 758 xs_tcp_shutdown(xprt);
759 case -ECONNREFUSED:
262965f5 760 case -ENOTCONN:
b9d2bb2e 761 case -EPIPE:
b6ddf64f 762 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
a246b010 763 }
5fe46e9d 764
a246b010
CL		 765 return status;
766}
767
e0ab53de
TM		 768/**
769 * xs_tcp_release_xprt - clean up after a tcp transmission
770 * @xprt: transport
771 * @task: rpc task
772 *
773 * This cleans up if an error causes us to abort the transmission of a request.
774 * In this case, the socket may need to be reset in order to avoid confusing
775 * the server.
776 */
777static void xs_tcp_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
778{
779 struct rpc_rqst *req;
780
781 if (task != xprt->snd_task)
782 return;
783 if (task == NULL)
784 goto out_release;
785 req = task->tk_rqstp;
43cedbf0
TM		 786 if (req == NULL)
787 goto out_release;
e0ab53de
TM		 788 if (req->rq_bytes_sent == 0)
789 goto out_release;
790 if (req->rq_bytes_sent == req->rq_snd_buf.len)
791 goto out_release;
a4f0835c 792 set_bit(XPRT_CLOSE_WAIT, &xprt->state);
e0ab53de
TM		 793out_release:
794 xprt_release_xprt(xprt, task);
795}
796
2a9e1cfa
TM		 797static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk)
798{
799 transport->old_data_ready = sk->sk_data_ready;
800 transport->old_state_change = sk->sk_state_change;
801 transport->old_write_space = sk->sk_write_space;
2a9e1cfa
TM		 802}
803
804static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk)
805{
806 sk->sk_data_ready = transport->old_data_ready;
807 sk->sk_state_change = transport->old_state_change;
808 sk->sk_write_space = transport->old_write_space;
2a9e1cfa
TM		 809}
810
fe315e76 811static void xs_reset_transport(struct sock_xprt *transport)
a246b010 812{
ee0ac0c2
CL		 813 struct socket *sock = transport->sock;
814 struct sock *sk = transport->inet;
a246b010 815
fe315e76
CL		 816 if (sk == NULL)
817 return;
9903cd1c 818
246408dc
TM		 819 transport->srcport = 0;
820
a246b010 821 write_lock_bh(&sk->sk_callback_lock);
ee0ac0c2
CL		 822 transport->inet = NULL;
823 transport->sock = NULL;
a246b010 824
9903cd1c 825 sk->sk_user_data = NULL;
2a9e1cfa
TM		 826
827 xs_restore_old_callbacks(transport, sk);
a246b010
CL		 828 write_unlock_bh(&sk->sk_callback_lock);
829
9903cd1c 830 sk->sk_no_check = 0;
a246b010 831
40b5ea0c 832 trace_rpc_socket_close(&transport->xprt, sock);
a246b010 833 sock_release(sock);
fe315e76
CL		 834}
835
836/**
837 * xs_close - close a socket
838 * @xprt: transport
839 *
840 * This is used when all requests are complete; ie, no DRC state remains
841 * on the server we want to save.
f75e6745
TM		 842 *
843 * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with
844 * xs_reset_transport() zeroing the socket from underneath a writer.
fe315e76
CL		 845 */
846static void xs_close(struct rpc_xprt *xprt)
847{
848 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
849
850 dprintk("RPC: xs_close xprt %p\n", xprt);
851
93dc41bd
N		 852 cancel_delayed_work_sync(&transport->connect_worker);
853
fe315e76 854 xs_reset_transport(transport);
61d0a8e6 855 xprt->reestablish_timeout = 0;
fe315e76 856
632e3bdc 857 smp_mb__before_clear_bit();
7d1e8255 858 clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
632e3bdc 859 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
3b948ae5 860 clear_bit(XPRT_CLOSING, &xprt->state);
632e3bdc 861 smp_mb__after_clear_bit();
62da3b24 862 xprt_disconnect_done(xprt);
a246b010
CL		 863}
864
f75e6745
TM		 865static void xs_tcp_close(struct rpc_xprt *xprt)
866{
867 if (test_and_clear_bit(XPRT_CONNECTION_CLOSE, &xprt->state))
868 xs_close(xprt);
869 else
870 xs_tcp_shutdown(xprt);
871}
872
9903cd1c
CL		 873/**
874 * xs_destroy - prepare to shutdown a transport
875 * @xprt: doomed transport
876 *
877 */
878static void xs_destroy(struct rpc_xprt *xprt)
a246b010 879{
46121cf7 880 dprintk("RPC: xs_destroy xprt %p\n", xprt);
9903cd1c 881
a1311d87
TM		 882 xs_close(xprt);
883 xs_free_peer_addresses(xprt);
884 xprt_free(xprt);
885 module_put(THIS_MODULE);
a246b010
CL		 886}
887
9903cd1c
CL		 888static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
889{
890 return (struct rpc_xprt *) sk->sk_user_data;
891}
892
176e21ee
CL		 893static int xs_local_copy_to_xdr(struct xdr_buf *xdr, struct sk_buff *skb)
894{
895 struct xdr_skb_reader desc = {
896 .skb = skb,
897 .offset = sizeof(rpc_fraghdr),
898 .count = skb->len - sizeof(rpc_fraghdr),
899 };
900
901 if (xdr_partial_copy_from_skb(xdr, 0, &desc, xdr_skb_read_bits) < 0)
902 return -1;
903 if (desc.count)
904 return -1;
905 return 0;
906}
907
908/**
909 * xs_local_data_ready - "data ready" callback for AF_LOCAL sockets
910 * @sk: socket with data to read
911 * @len: how much data to read
912 *
913 * Currently this assumes we can read the whole reply in a single gulp.
914 */
915static void xs_local_data_ready(struct sock *sk, int len)
916{
917 struct rpc_task *task;
918 struct rpc_xprt *xprt;
919 struct rpc_rqst *rovr;
920 struct sk_buff *skb;
921 int err, repsize, copied;
922 u32 _xid;
923 __be32 *xp;
924
925 read_lock_bh(&sk->sk_callback_lock);
926 dprintk("RPC: %s...\n", __func__);
927 xprt = xprt_from_sock(sk);
928 if (xprt == NULL)
929 goto out;
930
931 skb = skb_recv_datagram(sk, 0, 1, &err);
932 if (skb == NULL)
933 goto out;
934
176e21ee
CL		 935 repsize = skb->len - sizeof(rpc_fraghdr);
936 if (repsize < 4) {
937 dprintk("RPC: impossible RPC reply size %d\n", repsize);
938 goto dropit;
939 }
940
941 /* Copy the XID from the skb... */
942 xp = skb_header_pointer(skb, sizeof(rpc_fraghdr), sizeof(_xid), &_xid);
943 if (xp == NULL)
944 goto dropit;
945
946 /* Look up and lock the request corresponding to the given XID */
947 spin_lock(&xprt->transport_lock);
948 rovr = xprt_lookup_rqst(xprt, *xp);
949 if (!rovr)
950 goto out_unlock;
951 task = rovr->rq_task;
952
953 copied = rovr->rq_private_buf.buflen;
954 if (copied > repsize)
955 copied = repsize;
956
957 if (xs_local_copy_to_xdr(&rovr->rq_private_buf, skb)) {
958 dprintk("RPC: sk_buff copy failed\n");
959 goto out_unlock;
960 }
961
962 xprt_complete_rqst(task, copied);
963
964 out_unlock:
965 spin_unlock(&xprt->transport_lock);
966 dropit:
967 skb_free_datagram(sk, skb);
968 out:
969 read_unlock_bh(&sk->sk_callback_lock);
970}
971
9903cd1c
CL		 972/**
973 * xs_udp_data_ready - "data ready" callback for UDP sockets
974 * @sk: socket with data to read
975 * @len: how much data to read
976 *
a246b010 977 */
9903cd1c 978static void xs_udp_data_ready(struct sock *sk, int len)
a246b010 979{
9903cd1c
CL		 980 struct rpc_task *task;
981 struct rpc_xprt *xprt;
a246b010 982 struct rpc_rqst *rovr;
9903cd1c 983 struct sk_buff *skb;
a246b010 984 int err, repsize, copied;
d8ed029d
AD		 985 u32 _xid;
986 __be32 *xp;
a246b010 987
f064af1e 988 read_lock_bh(&sk->sk_callback_lock);
46121cf7 989 dprintk("RPC: xs_udp_data_ready...\n");
9903cd1c 990 if (!(xprt = xprt_from_sock(sk)))
a246b010 991 goto out;
a246b010
CL		 992
993 if ((skb = skb_recv_datagram(sk, 0, 1, &err)) == NULL)
994 goto out;
995
a246b010
CL		 996 repsize = skb->len - sizeof(struct udphdr);
997 if (repsize < 4) {
46121cf7 998 dprintk("RPC: impossible RPC reply size %d!\n", repsize);
a246b010
CL		 999 goto dropit;
1000 }
1001
1002 /* Copy the XID from the skb... */
1003 xp = skb_header_pointer(skb, sizeof(struct udphdr),
1004 sizeof(_xid), &_xid);
1005 if (xp == NULL)
1006 goto dropit;
1007
1008 /* Look up and lock the request corresponding to the given XID */
4a0f8c04 1009 spin_lock(&xprt->transport_lock);
a246b010
CL		 1010 rovr = xprt_lookup_rqst(xprt, *xp);
1011 if (!rovr)
1012 goto out_unlock;
1013 task = rovr->rq_task;
1014
a246b010
CL		 1015 if ((copied = rovr->rq_private_buf.buflen) > repsize)
1016 copied = repsize;
1017
1018 /* Suck it into the iovec, verify checksum if not done by hw. */
1781f7f5
HX		 1019 if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
1020 UDPX_INC_STATS_BH(sk, UDP_MIB_INERRORS);
a246b010 1021 goto out_unlock;
1781f7f5
HX		 1022 }
1023
1024 UDPX_INC_STATS_BH(sk, UDP_MIB_INDATAGRAMS);
a246b010 1025
6a24dfb6 1026 xprt_adjust_cwnd(xprt, task, copied);
1570c1e4 1027 xprt_complete_rqst(task, copied);
a246b010
CL		 1028
1029 out_unlock:
4a0f8c04 1030 spin_unlock(&xprt->transport_lock);
a246b010
CL		 1031 dropit:
1032 skb_free_datagram(sk, skb);
1033 out:
f064af1e 1034 read_unlock_bh(&sk->sk_callback_lock);
a246b010
CL		 1035}
1036
a519fc7a
TM		 1037/*
1038 * Helper function to force a TCP close if the server is sending
1039 * junk and/or it has put us in CLOSE_WAIT
1040 */
1041static void xs_tcp_force_close(struct rpc_xprt *xprt)
1042{
1043 set_bit(XPRT_CONNECTION_CLOSE, &xprt->state);
1044 xprt_force_disconnect(xprt);
1045}
1046
dd456471 1047static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
a246b010 1048{
51971139 1049 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
a246b010
CL		 1050 size_t len, used;
1051 char *p;
1052
51971139
CL		 1053 p = ((char *) &transport->tcp_fraghdr) + transport->tcp_offset;
1054 len = sizeof(transport->tcp_fraghdr) - transport->tcp_offset;
9d292316 1055 used = xdr_skb_read_bits(desc, p, len);
51971139 1056 transport->tcp_offset += used;
a246b010
CL		 1057 if (used != len)
1058 return;
808012fb 1059
51971139
CL		 1060 transport->tcp_reclen = ntohl(transport->tcp_fraghdr);
1061 if (transport->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
e136d092 1062 transport->tcp_flags |= TCP_RCV_LAST_FRAG;
a246b010 1063 else
e136d092 1064 transport->tcp_flags &= ~TCP_RCV_LAST_FRAG;
51971139 1065 transport->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
808012fb 1066
e136d092 1067 transport->tcp_flags &= ~TCP_RCV_COPY_FRAGHDR;
51971139 1068 transport->tcp_offset = 0;
808012fb 1069
a246b010 1070 /* Sanity check of the record length */
18dca02a 1071 if (unlikely(transport->tcp_reclen < 8)) {
46121cf7 1072 dprintk("RPC: invalid TCP record fragment length\n");
a519fc7a 1073 xs_tcp_force_close(xprt);
9903cd1c 1074 return;
a246b010 1075 }
46121cf7 1076 dprintk("RPC: reading TCP record fragment of length %d\n",
51971139 1077 transport->tcp_reclen);
a246b010
CL		 1078}
1079
51971139 1080static void xs_tcp_check_fraghdr(struct sock_xprt *transport)
a246b010 1081{
51971139 1082 if (transport->tcp_offset == transport->tcp_reclen) {
e136d092 1083 transport->tcp_flags |= TCP_RCV_COPY_FRAGHDR;
51971139 1084 transport->tcp_offset = 0;
e136d092
CL		 1085 if (transport->tcp_flags & TCP_RCV_LAST_FRAG) {
1086 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1087 transport->tcp_flags |= TCP_RCV_COPY_XID;
51971139 1088 transport->tcp_copied = 0;
a246b010
CL		 1089 }
1090 }
1091}
1092
dd456471 1093static inline void xs_tcp_read_xid(struct sock_xprt *transport, struct xdr_skb_reader *desc)
a246b010
CL		 1094{
1095 size_t len, used;
1096 char *p;
1097
51971139 1098 len = sizeof(transport->tcp_xid) - transport->tcp_offset;
46121cf7 1099 dprintk("RPC: reading XID (%Zu bytes)\n", len);
51971139 1100 p = ((char *) &transport->tcp_xid) + transport->tcp_offset;
9d292316 1101 used = xdr_skb_read_bits(desc, p, len);
51971139 1102 transport->tcp_offset += used;
a246b010
CL		 1103 if (used != len)
1104 return;
e136d092 1105 transport->tcp_flags &= ~TCP_RCV_COPY_XID;
f4a2e418 1106 transport->tcp_flags |= TCP_RCV_READ_CALLDIR;
51971139 1107 transport->tcp_copied = 4;
18dca02a
RL		 1108 dprintk("RPC: reading %s XID %08x\n",
1109 (transport->tcp_flags & TCP_RPC_REPLY) ? "reply for"
1110 : "request with",
51971139
CL		 1111 ntohl(transport->tcp_xid));
1112 xs_tcp_check_fraghdr(transport);
a246b010
CL		 1113}
1114
18dca02a
RL		 1115static inline void xs_tcp_read_calldir(struct sock_xprt *transport,
1116 struct xdr_skb_reader *desc)
a246b010 1117{
18dca02a
RL		 1118 size_t len, used;
1119 u32 offset;
b76ce561 1120 char *p;
18dca02a
RL		 1121
1122 /*
1123 * We want transport->tcp_offset to be 8 at the end of this routine
1124 * (4 bytes for the xid and 4 bytes for the call/reply flag).
1125 * When this function is called for the first time,
1126 * transport->tcp_offset is 4 (after having already read the xid).
1127 */
1128 offset = transport->tcp_offset - sizeof(transport->tcp_xid);
b76ce561 1129 len = sizeof(transport->tcp_calldir) - offset;
18dca02a 1130 dprintk("RPC: reading CALL/REPLY flag (%Zu bytes)\n", len);
b76ce561
TM		 1131 p = ((char *) &transport->tcp_calldir) + offset;
1132 used = xdr_skb_read_bits(desc, p, len);
18dca02a
RL		 1133 transport->tcp_offset += used;
1134 if (used != len)
1135 return;
f4a2e418 1136 transport->tcp_flags &= ~TCP_RCV_READ_CALLDIR;
f4a2e418
RL		 1137 /*
1138 * We don't yet have the XDR buffer, so we will write the calldir
1139 * out after we get the buffer from the 'struct rpc_rqst'
1140 */
b76ce561
TM		 1141 switch (ntohl(transport->tcp_calldir)) {
1142 case RPC_REPLY:
1143 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1144 transport->tcp_flags |= TCP_RCV_COPY_DATA;
18dca02a 1145 transport->tcp_flags |= TCP_RPC_REPLY;
b76ce561
TM		 1146 break;
1147 case RPC_CALL:
1148 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1149 transport->tcp_flags |= TCP_RCV_COPY_DATA;
18dca02a 1150 transport->tcp_flags &= ~TCP_RPC_REPLY;
b76ce561
TM		 1151 break;
1152 default:
1153 dprintk("RPC: invalid request message type\n");
a519fc7a 1154 xs_tcp_force_close(&transport->xprt);
b76ce561 1155 }
18dca02a
RL		 1156 xs_tcp_check_fraghdr(transport);
1157}
1158
44b98efd
RL		 1159static inline void xs_tcp_read_common(struct rpc_xprt *xprt,
1160 struct xdr_skb_reader *desc,
1161 struct rpc_rqst *req)
a246b010 1162{
44b98efd
RL		 1163 struct sock_xprt *transport =
1164 container_of(xprt, struct sock_xprt, xprt);
a246b010
CL		 1165 struct xdr_buf *rcvbuf;
1166 size_t len;
1167 ssize_t r;
1168
a246b010 1169 rcvbuf = &req->rq_private_buf;
f4a2e418
RL		 1170
1171 if (transport->tcp_flags & TCP_RCV_COPY_CALLDIR) {
1172 /*
1173 * Save the RPC direction in the XDR buffer
1174 */
f4a2e418 1175 memcpy(rcvbuf->head[0].iov_base + transport->tcp_copied,
b76ce561
TM		 1176 &transport->tcp_calldir,
1177 sizeof(transport->tcp_calldir));
1178 transport->tcp_copied += sizeof(transport->tcp_calldir);
f4a2e418 1179 transport->tcp_flags &= ~TCP_RCV_COPY_CALLDIR;
a246b010
CL		 1180 }
1181
a246b010 1182 len = desc->count;
51971139 1183 if (len > transport->tcp_reclen - transport->tcp_offset) {
dd456471 1184 struct xdr_skb_reader my_desc;
a246b010 1185
51971139 1186 len = transport->tcp_reclen - transport->tcp_offset;
a246b010
CL		 1187 memcpy(&my_desc, desc, sizeof(my_desc));
1188 my_desc.count = len;
51971139 1189 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
9d292316 1190 &my_desc, xdr_skb_read_bits);
a246b010
CL		 1191 desc->count -= r;
1192 desc->offset += r;
1193 } else
51971139 1194 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
9d292316 1195 desc, xdr_skb_read_bits);
a246b010
CL		 1196
1197 if (r > 0) {
51971139
CL		 1198 transport->tcp_copied += r;
1199 transport->tcp_offset += r;
a246b010
CL		 1200 }
1201 if (r != len) {
1202 /* Error when copying to the receive buffer,
1203 * usually because we weren't able to allocate
1204 * additional buffer pages. All we can do now
e136d092 1205 * is turn off TCP_RCV_COPY_DATA, so the request
a246b010
CL		 1206 * will not receive any additional updates,
1207 * and time out.
1208 * Any remaining data from this record will
1209 * be discarded.
1210 */
e136d092 1211 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
46121cf7 1212 dprintk("RPC: XID %08x truncated request\n",
51971139 1213 ntohl(transport->tcp_xid));
46121cf7
CL		 1214 dprintk("RPC: xprt = %p, tcp_copied = %lu, "
1215 "tcp_offset = %u, tcp_reclen = %u\n",
1216 xprt, transport->tcp_copied,
1217 transport->tcp_offset, transport->tcp_reclen);
44b98efd 1218 return;
a246b010
CL		 1219 }
1220
46121cf7 1221 dprintk("RPC: XID %08x read %Zd bytes\n",
51971139 1222 ntohl(transport->tcp_xid), r);
46121cf7
CL		 1223 dprintk("RPC: xprt = %p, tcp_copied = %lu, tcp_offset = %u, "
1224 "tcp_reclen = %u\n", xprt, transport->tcp_copied,
1225 transport->tcp_offset, transport->tcp_reclen);
51971139
CL		 1226
1227 if (transport->tcp_copied == req->rq_private_buf.buflen)
e136d092 1228 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
51971139 1229 else if (transport->tcp_offset == transport->tcp_reclen) {
e136d092
CL		 1230 if (transport->tcp_flags & TCP_RCV_LAST_FRAG)
1231 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
a246b010 1232 }
44b98efd
RL		 1233}
1234
1235/*
1236 * Finds the request corresponding to the RPC xid and invokes the common
1237 * tcp read code to read the data.
1238 */
1239static inline int xs_tcp_read_reply(struct rpc_xprt *xprt,
1240 struct xdr_skb_reader *desc)
1241{
1242 struct sock_xprt *transport =
1243 container_of(xprt, struct sock_xprt, xprt);
1244 struct rpc_rqst *req;
1245
1246 dprintk("RPC: read reply XID %08x\n", ntohl(transport->tcp_xid));
1247
1248 /* Find and lock the request corresponding to this xid */
1249 spin_lock(&xprt->transport_lock);
1250 req = xprt_lookup_rqst(xprt, transport->tcp_xid);
1251 if (!req) {
1252 dprintk("RPC: XID %08x request not found!\n",
1253 ntohl(transport->tcp_xid));
1254 spin_unlock(&xprt->transport_lock);
1255 return -1;
1256 }
1257
1258 xs_tcp_read_common(xprt, desc, req);
1259
e136d092 1260 if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
51971139 1261 xprt_complete_rqst(req->rq_task, transport->tcp_copied);
44b98efd 1262
4a0f8c04 1263 spin_unlock(&xprt->transport_lock);
44b98efd
RL		 1264 return 0;
1265}
1266
9e00abc3 1267#if defined(CONFIG_SUNRPC_BACKCHANNEL)
44b98efd
RL		 1268/*
1269 * Obtains an rpc_rqst previously allocated and invokes the common
1270 * tcp read code to read the data. The result is placed in the callback
1271 * queue.
1272 * If we're unable to obtain the rpc_rqst we schedule the closing of the
1273 * connection and return -1.
1274 */
1275static inline int xs_tcp_read_callback(struct rpc_xprt *xprt,
1276 struct xdr_skb_reader *desc)
1277{
1278 struct sock_xprt *transport =
1279 container_of(xprt, struct sock_xprt, xprt);
1280 struct rpc_rqst *req;
1281
1282 req = xprt_alloc_bc_request(xprt);
1283 if (req == NULL) {
1284 printk(KERN_WARNING "Callback slot table overflowed\n");
1285 xprt_force_disconnect(xprt);
1286 return -1;
1287 }
1288
1289 req->rq_xid = transport->tcp_xid;
1290 dprintk("RPC: read callback XID %08x\n", ntohl(req->rq_xid));
1291 xs_tcp_read_common(xprt, desc, req);
1292
1293 if (!(transport->tcp_flags & TCP_RCV_COPY_DATA)) {
1294 struct svc_serv *bc_serv = xprt->bc_serv;
1295
1296 /*
1297 * Add callback request to callback list. The callback
1298 * service sleeps on the sv_cb_waitq waiting for new
1299 * requests. Wake it up after adding enqueing the
1300 * request.
1301 */
1302 dprintk("RPC: add callback request to list\n");
1303 spin_lock(&bc_serv->sv_cb_lock);
1304 list_add(&req->rq_bc_list, &bc_serv->sv_cb_list);
1305 spin_unlock(&bc_serv->sv_cb_lock);
1306 wake_up(&bc_serv->sv_cb_waitq);
1307 }
1308
1309 req->rq_private_buf.len = transport->tcp_copied;
1310
1311 return 0;
1312}
1313
1314static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1315 struct xdr_skb_reader *desc)
1316{
1317 struct sock_xprt *transport =
1318 container_of(xprt, struct sock_xprt, xprt);
1319
1320 return (transport->tcp_flags & TCP_RPC_REPLY) ?
1321 xs_tcp_read_reply(xprt, desc) :
1322 xs_tcp_read_callback(xprt, desc);
1323}
1324#else
1325static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1326 struct xdr_skb_reader *desc)
1327{
1328 return xs_tcp_read_reply(xprt, desc);
1329}
9e00abc3 1330#endif /* CONFIG_SUNRPC_BACKCHANNEL */
44b98efd
RL		 1331
1332/*
1333 * Read data off the transport. This can be either an RPC_CALL or an
1334 * RPC_REPLY. Relay the processing to helper functions.
1335 */
1336static void xs_tcp_read_data(struct rpc_xprt *xprt,
1337 struct xdr_skb_reader *desc)
1338{
1339 struct sock_xprt *transport =
1340 container_of(xprt, struct sock_xprt, xprt);
1341
1342 if (_xs_tcp_read_data(xprt, desc) == 0)
1343 xs_tcp_check_fraghdr(transport);
1344 else {
1345 /*
1346 * The transport_lock protects the request handling.
1347 * There's no need to hold it to update the tcp_flags.
1348 */
1349 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1350 }
a246b010
CL		 1351}
1352
dd456471 1353static inline void xs_tcp_read_discard(struct sock_xprt *transport, struct xdr_skb_reader *desc)
a246b010
CL		 1354{
1355 size_t len;
1356
51971139 1357 len = transport->tcp_reclen - transport->tcp_offset;
a246b010
CL		 1358 if (len > desc->count)
1359 len = desc->count;
1360 desc->count -= len;
1361 desc->offset += len;
51971139 1362 transport->tcp_offset += len;
46121cf7 1363 dprintk("RPC: discarded %Zu bytes\n", len);
51971139 1364 xs_tcp_check_fraghdr(transport);
a246b010
CL		 1365}
1366
9903cd1c 1367static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
a246b010
CL		 1368{
1369 struct rpc_xprt *xprt = rd_desc->arg.data;
51971139 1370 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
dd456471 1371 struct xdr_skb_reader desc = {
a246b010
CL		 1372 .skb = skb,
1373 .offset = offset,
1374 .count = len,
9903cd1c 1375 };
a246b010 1376
46121cf7 1377 dprintk("RPC: xs_tcp_data_recv started\n");
a246b010
CL		 1378 do {
1379 /* Read in a new fragment marker if necessary */
1380 /* Can we ever really expect to get completely empty fragments? */
e136d092 1381 if (transport->tcp_flags & TCP_RCV_COPY_FRAGHDR) {
9903cd1c 1382 xs_tcp_read_fraghdr(xprt, &desc);
a246b010
CL		 1383 continue;
1384 }
1385 /* Read in the xid if necessary */
e136d092 1386 if (transport->tcp_flags & TCP_RCV_COPY_XID) {
51971139 1387 xs_tcp_read_xid(transport, &desc);
a246b010
CL		 1388 continue;
1389 }
18dca02a 1390 /* Read in the call/reply flag */
f4a2e418 1391 if (transport->tcp_flags & TCP_RCV_READ_CALLDIR) {
18dca02a
RL		 1392 xs_tcp_read_calldir(transport, &desc);
1393 continue;
1394 }
a246b010 1395 /* Read in the request data */
e136d092 1396 if (transport->tcp_flags & TCP_RCV_COPY_DATA) {
44b98efd 1397 xs_tcp_read_data(xprt, &desc);
a246b010
CL		 1398 continue;
1399 }
1400 /* Skip over any trailing bytes on short reads */
51971139 1401 xs_tcp_read_discard(transport, &desc);
a246b010 1402 } while (desc.count);
46121cf7 1403 dprintk("RPC: xs_tcp_data_recv done\n");
a246b010
CL		 1404 return len - desc.count;
1405}
1406
9903cd1c
CL		 1407/**
1408 * xs_tcp_data_ready - "data ready" callback for TCP sockets
1409 * @sk: socket with data to read
1410 * @bytes: how much data to read
1411 *
1412 */
1413static void xs_tcp_data_ready(struct sock *sk, int bytes)
a246b010
CL		 1414{
1415 struct rpc_xprt *xprt;
1416 read_descriptor_t rd_desc;
ff2d7db8 1417 int read;
a246b010 1418
46121cf7
CL		 1419 dprintk("RPC: xs_tcp_data_ready...\n");
1420
f064af1e 1421 read_lock_bh(&sk->sk_callback_lock);
9903cd1c 1422 if (!(xprt = xprt_from_sock(sk)))
a246b010 1423 goto out;
61d0a8e6
NB		 1424 /* Any data means we had a useful conversation, so
1425 * the we don't need to delay the next reconnect
1426 */
1427 if (xprt->reestablish_timeout)
1428 xprt->reestablish_timeout = 0;
1429
9903cd1c 1430 /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
a246b010 1431 rd_desc.arg.data = xprt;
ff2d7db8
TM		 1432 do {
1433 rd_desc.count = 65536;
1434 read = tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
1435 } while (read > 0);
a246b010 1436out:
f064af1e 1437 read_unlock_bh(&sk->sk_callback_lock);
a246b010
CL		 1438}
1439
7d1e8255
TM		 1440/*
1441 * Do the equivalent of linger/linger2 handling for dealing with
1442 * broken servers that don't close the socket in a timely
1443 * fashion
1444 */
1445static void xs_tcp_schedule_linger_timeout(struct rpc_xprt *xprt,
1446 unsigned long timeout)
1447{
1448 struct sock_xprt *transport;
1449
1450 if (xprt_test_and_set_connecting(xprt))
1451 return;
1452 set_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1453 transport = container_of(xprt, struct sock_xprt, xprt);
1454 queue_delayed_work(rpciod_workqueue, &transport->connect_worker,
1455 timeout);
1456}
1457
1458static void xs_tcp_cancel_linger_timeout(struct rpc_xprt *xprt)
1459{
1460 struct sock_xprt *transport;
1461
1462 transport = container_of(xprt, struct sock_xprt, xprt);
1463
1464 if (!test_bit(XPRT_CONNECTION_ABORT, &xprt->state) ||
1465 !cancel_delayed_work(&transport->connect_worker))
1466 return;
1467 clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1468 xprt_clear_connecting(xprt);
1469}
1470
4bc1e68e 1471static void xs_sock_reset_connection_flags(struct rpc_xprt *xprt)
7d1e8255
TM		 1472{
1473 smp_mb__before_clear_bit();
a519fc7a
TM		 1474 clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1475 clear_bit(XPRT_CONNECTION_CLOSE, &xprt->state);
7d1e8255
TM		 1476 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1477 clear_bit(XPRT_CLOSING, &xprt->state);
1478 smp_mb__after_clear_bit();
4bc1e68e
TM		 1479}
1480
1481static void xs_sock_mark_closed(struct rpc_xprt *xprt)
1482{
1483 xs_sock_reset_connection_flags(xprt);
7d1e8255
TM		 1484 /* Mark transport as closed and wake up all pending tasks */
1485 xprt_disconnect_done(xprt);
1486}
1487
9903cd1c
CL		 1488/**
1489 * xs_tcp_state_change - callback to handle TCP socket state changes
1490 * @sk: socket whose state has changed
1491 *
1492 */
1493static void xs_tcp_state_change(struct sock *sk)
a246b010 1494{
9903cd1c 1495 struct rpc_xprt *xprt;
a246b010 1496
f064af1e 1497 read_lock_bh(&sk->sk_callback_lock);
a246b010
CL		 1498 if (!(xprt = xprt_from_sock(sk)))
1499 goto out;
46121cf7 1500 dprintk("RPC: xs_tcp_state_change client %p...\n", xprt);
669502ff 1501 dprintk("RPC: state %x conn %d dead %d zapped %d sk_shutdown %d\n",
46121cf7
CL		 1502 sk->sk_state, xprt_connected(xprt),
1503 sock_flag(sk, SOCK_DEAD),
669502ff
AC		 1504 sock_flag(sk, SOCK_ZAPPED),
1505 sk->sk_shutdown);
a246b010 1506
40b5ea0c 1507 trace_rpc_socket_state_change(xprt, sk->sk_socket);
a246b010
CL		 1508 switch (sk->sk_state) {
1509 case TCP_ESTABLISHED:
f064af1e 1510 spin_lock(&xprt->transport_lock);
a246b010 1511 if (!xprt_test_and_set_connected(xprt)) {
51971139
CL		 1512 struct sock_xprt *transport = container_of(xprt,
1513 struct sock_xprt, xprt);
1514
a246b010 1515 /* Reset TCP record info */
51971139
CL		 1516 transport->tcp_offset = 0;
1517 transport->tcp_reclen = 0;
1518 transport->tcp_copied = 0;
e136d092
CL		 1519 transport->tcp_flags =
1520 TCP_RCV_COPY_FRAGHDR | TCP_RCV_COPY_XID;
8b71798c 1521 xprt->connect_cookie++;
51971139 1522
2a491991 1523 xprt_wake_pending_tasks(xprt, -EAGAIN);
a246b010 1524 }
f064af1e 1525 spin_unlock(&xprt->transport_lock);
a246b010 1526 break;
3b948ae5
TM		 1527 case TCP_FIN_WAIT1:
1528 /* The client initiated a shutdown of the socket */
7c1d71cf 1529 xprt->connect_cookie++;
663b8858 1530 xprt->reestablish_timeout = 0;
3b948ae5
TM		 1531 set_bit(XPRT_CLOSING, &xprt->state);
1532 smp_mb__before_clear_bit();
1533 clear_bit(XPRT_CONNECTED, &xprt->state);
ef803670 1534 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
3b948ae5 1535 smp_mb__after_clear_bit();
25fe6142 1536 xs_tcp_schedule_linger_timeout(xprt, xs_tcp_fin_timeout);
a246b010 1537 break;
632e3bdc 1538 case TCP_CLOSE_WAIT:
3b948ae5 1539 /* The server initiated a shutdown of the socket */
7c1d71cf 1540 xprt->connect_cookie++;
d0bea455 1541 clear_bit(XPRT_CONNECTED, &xprt->state);
a519fc7a 1542 xs_tcp_force_close(xprt);
663b8858
TM		 1543 case TCP_CLOSING:
1544 /*
1545 * If the server closed down the connection, make sure that
1546 * we back off before reconnecting
1547 */
1548 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1549 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
3b948ae5
TM		 1550 break;
1551 case TCP_LAST_ACK:
670f9457 1552 set_bit(XPRT_CLOSING, &xprt->state);
25fe6142 1553 xs_tcp_schedule_linger_timeout(xprt, xs_tcp_fin_timeout);
3b948ae5
TM		 1554 smp_mb__before_clear_bit();
1555 clear_bit(XPRT_CONNECTED, &xprt->state);
1556 smp_mb__after_clear_bit();
1557 break;
1558 case TCP_CLOSE:
7d1e8255
TM		 1559 xs_tcp_cancel_linger_timeout(xprt);
1560 xs_sock_mark_closed(xprt);
a246b010
CL		 1561 }
1562 out:
f064af1e 1563 read_unlock_bh(&sk->sk_callback_lock);
a246b010
CL		 1564}
1565
1f0fa154
IJ		 1566static void xs_write_space(struct sock *sk)
1567{
1568 struct socket *sock;
1569 struct rpc_xprt *xprt;
1570
1571 if (unlikely(!(sock = sk->sk_socket)))
1572 return;
1573 clear_bit(SOCK_NOSPACE, &sock->flags);
1574
1575 if (unlikely(!(xprt = xprt_from_sock(sk))))
1576 return;
1577 if (test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags) == 0)
1578 return;
1579
1580 xprt_write_space(xprt);
1581}
1582
9903cd1c 1583/**
c7b2cae8
CL		 1584 * xs_udp_write_space - callback invoked when socket buffer space
1585 * becomes available
9903cd1c
CL		 1586 * @sk: socket whose state has changed
1587 *
a246b010
CL		 1588 * Called when more output buffer space is available for this socket.
1589 * We try not to wake our writers until they can make "significant"
c7b2cae8 1590 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
a246b010
CL		 1591 * with a bunch of small requests.
1592 */
c7b2cae8 1593static void xs_udp_write_space(struct sock *sk)
a246b010 1594{
f064af1e 1595 read_lock_bh(&sk->sk_callback_lock);
a246b010 1596
c7b2cae8 1597 /* from net/core/sock.c:sock_def_write_space */
1f0fa154
IJ		 1598 if (sock_writeable(sk))
1599 xs_write_space(sk);
a246b010 1600
f064af1e 1601 read_unlock_bh(&sk->sk_callback_lock);
c7b2cae8 1602}
a246b010 1603
c7b2cae8
CL		 1604/**
1605 * xs_tcp_write_space - callback invoked when socket buffer space
1606 * becomes available
1607 * @sk: socket whose state has changed
1608 *
1609 * Called when more output buffer space is available for this socket.
1610 * We try not to wake our writers until they can make "significant"
1611 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1612 * with a bunch of small requests.
1613 */
1614static void xs_tcp_write_space(struct sock *sk)
1615{
f064af1e 1616 read_lock_bh(&sk->sk_callback_lock);
c7b2cae8
CL		 1617
1618 /* from net/core/stream.c:sk_stream_write_space */
64dc6130 1619 if (sk_stream_is_writeable(sk))
1f0fa154 1620 xs_write_space(sk);
b6ddf64f 1621
f064af1e 1622 read_unlock_bh(&sk->sk_callback_lock);
a246b010
CL		 1623}
1624
470056c2 1625static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
a246b010 1626{
ee0ac0c2
CL		 1627 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1628 struct sock *sk = transport->inet;
a246b010 1629
7c6e066e 1630 if (transport->rcvsize) {
a246b010 1631 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
7c6e066e 1632 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
a246b010 1633 }
7c6e066e 1634 if (transport->sndsize) {
a246b010 1635 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
7c6e066e 1636 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
a246b010
CL		 1637 sk->sk_write_space(sk);
1638 }
1639}
1640
43118c29 1641/**
470056c2 1642 * xs_udp_set_buffer_size - set send and receive limits
43118c29 1643 * @xprt: generic transport
470056c2
CL		 1644 * @sndsize: requested size of send buffer, in bytes
1645 * @rcvsize: requested size of receive buffer, in bytes
43118c29 1646 *
470056c2 1647 * Set socket send and receive buffer size limits.
43118c29 1648 */
470056c2 1649static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
43118c29 1650{
7c6e066e
CL		 1651 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1652
1653 transport->sndsize = 0;
470056c2 1654 if (sndsize)
7c6e066e
CL		 1655 transport->sndsize = sndsize + 1024;
1656 transport->rcvsize = 0;
470056c2 1657 if (rcvsize)
7c6e066e 1658 transport->rcvsize = rcvsize + 1024;
470056c2
CL		 1659
1660 xs_udp_do_set_buffer_size(xprt);
43118c29
CL		 1661}
1662
46c0ee8b
CL		 1663/**
1664 * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1665 * @task: task that timed out
1666 *
1667 * Adjust the congestion window after a retransmit timeout has occurred.
1668 */
6a24dfb6 1669static void xs_udp_timer(struct rpc_xprt *xprt, struct rpc_task *task)
46c0ee8b 1670{
6a24dfb6 1671 xprt_adjust_cwnd(xprt, task, -ETIMEDOUT);
46c0ee8b
CL		 1672}
1673
b85d8806
CL		 1674static unsigned short xs_get_random_port(void)
1675{
1676 unsigned short range = xprt_max_resvport - xprt_min_resvport;
1677 unsigned short rand = (unsigned short) net_random() % range;
1678 return rand + xprt_min_resvport;
1679}
1680
92200412
CL		 1681/**
1682 * xs_set_port - reset the port number in the remote endpoint address
1683 * @xprt: generic transport
1684 * @port: new port number
1685 *
1686 */
1687static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1688{
46121cf7 1689 dprintk("RPC: setting port for xprt %p to %u\n", xprt, port);
c4efcb1d 1690
9dc3b095
CL		 1691 rpc_set_port(xs_addr(xprt), port);
1692 xs_update_peer_port(xprt);
92200412
CL		 1693}
1694
5d4ec932 1695static unsigned short xs_get_srcport(struct sock_xprt *transport)
67a391d7 1696{
fbfffbd5 1697 unsigned short port = transport->srcport;
67a391d7
TM		 1698
1699 if (port == 0 && transport->xprt.resvport)
1700 port = xs_get_random_port();
1701 return port;
1702}
1703
baaf4e48 1704static unsigned short xs_next_srcport(struct sock_xprt *transport, unsigned short port)
67a391d7 1705{
fbfffbd5
CL		 1706 if (transport->srcport != 0)
1707 transport->srcport = 0;
67a391d7
TM		 1708 if (!transport->xprt.resvport)
1709 return 0;
1710 if (port <= xprt_min_resvport || port > xprt_max_resvport)
1711 return xprt_max_resvport;
1712 return --port;
1713}
beb59b68 1714static int xs_bind(struct sock_xprt *transport, struct socket *sock)
a246b010 1715{
beb59b68 1716 struct sockaddr_storage myaddr;
67a391d7 1717 int err, nloop = 0;
5d4ec932 1718 unsigned short port = xs_get_srcport(transport);
67a391d7 1719 unsigned short last;
a246b010 1720
beb59b68 1721 memcpy(&myaddr, &transport->srcaddr, transport->xprt.addrlen);
a246b010 1722 do {
beb59b68
PE		 1723 rpc_set_port((struct sockaddr *)&myaddr, port);
1724 err = kernel_bind(sock, (struct sockaddr *)&myaddr,
1725 transport->xprt.addrlen);
67a391d7 1726 if (port == 0)
d3bc9a1d 1727 break;
a246b010 1728 if (err == 0) {
fbfffbd5 1729 transport->srcport = port;
d3bc9a1d 1730 break;
a246b010 1731 }
67a391d7 1732 last = port;
baaf4e48 1733 port = xs_next_srcport(transport, port);
67a391d7
TM		 1734 if (port > last)
1735 nloop++;
1736 } while (err == -EADDRINUSE && nloop != 2);
90058d37 1737
4232e863 1738 if (myaddr.ss_family == AF_INET)
beb59b68
PE		 1739 dprintk("RPC: %s %pI4:%u: %s (%d)\n", __func__,
1740 &((struct sockaddr_in *)&myaddr)->sin_addr,
1741 port, err ? "failed" : "ok", err);
1742 else
1743 dprintk("RPC: %s %pI6:%u: %s (%d)\n", __func__,
1744 &((struct sockaddr_in6 *)&myaddr)->sin6_addr,
1745 port, err ? "failed" : "ok", err);
a246b010
CL		 1746 return err;
1747}
1748
176e21ee
CL		 1749/*
1750 * We don't support autobind on AF_LOCAL sockets
1751 */
1752static void xs_local_rpcbind(struct rpc_task *task)
1753{
3dc0da27
TM		 1754 rcu_read_lock();
1755 xprt_set_bound(rcu_dereference(task->tk_client->cl_xprt));
1756 rcu_read_unlock();
176e21ee
CL		 1757}
1758
1759static void xs_local_set_port(struct rpc_xprt *xprt, unsigned short port)
1760{
1761}
beb59b68 1762
ed07536e
PZ		 1763#ifdef CONFIG_DEBUG_LOCK_ALLOC
1764static struct lock_class_key xs_key[2];
1765static struct lock_class_key xs_slock_key[2];
1766
176e21ee
CL		 1767static inline void xs_reclassify_socketu(struct socket *sock)
1768{
1769 struct sock *sk = sock->sk;
1770
176e21ee
CL		 1771 sock_lock_init_class_and_name(sk, "slock-AF_LOCAL-RPC",
1772 &xs_slock_key[1], "sk_lock-AF_LOCAL-RPC", &xs_key[1]);
1773}
1774
8945ee5e 1775static inline void xs_reclassify_socket4(struct socket *sock)
ed07536e
PZ		 1776{
1777 struct sock *sk = sock->sk;
8945ee5e 1778
8945ee5e
CL		 1779 sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1780 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1781}
ed07536e 1782
8945ee5e
CL		 1783static inline void xs_reclassify_socket6(struct socket *sock)
1784{
1785 struct sock *sk = sock->sk;
ed07536e 1786
8945ee5e
CL		 1787 sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1788 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
ed07536e 1789}
6bc9638a
PE		 1790
1791static inline void xs_reclassify_socket(int family, struct socket *sock)
1792{
1b7a1819
WAA		 1793 WARN_ON_ONCE(sock_owned_by_user(sock->sk));
1794 if (sock_owned_by_user(sock->sk))
1795 return;
1796
4232e863 1797 switch (family) {
176e21ee
CL		 1798 case AF_LOCAL:
1799 xs_reclassify_socketu(sock);
1800 break;
4232e863 1801 case AF_INET:
6bc9638a 1802 xs_reclassify_socket4(sock);
4232e863
CL		 1803 break;
1804 case AF_INET6:
6bc9638a 1805 xs_reclassify_socket6(sock);
4232e863
CL		 1806 break;
1807 }
6bc9638a 1808}
ed07536e 1809#else
176e21ee
CL		 1810static inline void xs_reclassify_socketu(struct socket *sock)
1811{
1812}
1813
8945ee5e
CL		 1814static inline void xs_reclassify_socket4(struct socket *sock)
1815{
1816}
1817
1818static inline void xs_reclassify_socket6(struct socket *sock)
ed07536e
PZ		 1819{
1820}
6bc9638a
PE		 1821
1822static inline void xs_reclassify_socket(int family, struct socket *sock)
1823{
1824}
ed07536e
PZ		 1825#endif
1826
93dc41bd
N		 1827static void xs_dummy_setup_socket(struct work_struct *work)
1828{
1829}
1830
6bc9638a
PE		 1831static struct socket *xs_create_sock(struct rpc_xprt *xprt,
1832 struct sock_xprt *transport, int family, int type, int protocol)
22f79326
PE		 1833{
1834 struct socket *sock;
1835 int err;
1836
6bc9638a 1837 err = __sock_create(xprt->xprt_net, family, type, protocol, &sock, 1);
22f79326
PE		 1838 if (err < 0) {
1839 dprintk("RPC: can't create %d transport socket (%d).\n",
1840 protocol, -err);
1841 goto out;
1842 }
6bc9638a 1843 xs_reclassify_socket(family, sock);
22f79326 1844
4cea288a
BH		 1845 err = xs_bind(transport, sock);
1846 if (err) {
22f79326
PE		 1847 sock_release(sock);
1848 goto out;
1849 }
1850
1851 return sock;
1852out:
1853 return ERR_PTR(err);
1854}
1855
176e21ee
CL		 1856static int xs_local_finish_connecting(struct rpc_xprt *xprt,
1857 struct socket *sock)
1858{
1859 struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1860 xprt);
1861
1862 if (!transport->inet) {
1863 struct sock *sk = sock->sk;
1864
1865 write_lock_bh(&sk->sk_callback_lock);
1866
1867 xs_save_old_callbacks(transport, sk);
1868
1869 sk->sk_user_data = xprt;
1870 sk->sk_data_ready = xs_local_data_ready;
1871 sk->sk_write_space = xs_udp_write_space;
176e21ee
CL		 1872 sk->sk_allocation = GFP_ATOMIC;
1873
1874 xprt_clear_connected(xprt);
1875
1876 /* Reset to new socket */
1877 transport->sock = sock;
1878 transport->inet = sk;
1879
1880 write_unlock_bh(&sk->sk_callback_lock);
1881 }
1882
1883 /* Tell the socket layer to start connecting... */
1884 xprt->stat.connect_count++;
1885 xprt->stat.connect_start = jiffies;
1886 return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, 0);
1887}
1888
1889/**
1890 * xs_local_setup_socket - create AF_LOCAL socket, connect to a local endpoint
1891 * @xprt: RPC transport to connect
1892 * @transport: socket transport to connect
1893 * @create_sock: function to create a socket of the correct type
176e21ee 1894 */
dc107402 1895static int xs_local_setup_socket(struct sock_xprt *transport)
176e21ee 1896{
176e21ee
CL		 1897 struct rpc_xprt *xprt = &transport->xprt;
1898 struct socket *sock;
1899 int status = -EIO;
1900
5cf02d09
JL		 1901 current->flags |= PF_FSTRANS;
1902
176e21ee
CL		 1903 clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1904 status = __sock_create(xprt->xprt_net, AF_LOCAL,
1905 SOCK_STREAM, 0, &sock, 1);
1906 if (status < 0) {
1907 dprintk("RPC: can't create AF_LOCAL "
1908 "transport socket (%d).\n", -status);
1909 goto out;
1910 }
1911 xs_reclassify_socketu(sock);
1912
1913 dprintk("RPC: worker connecting xprt %p via AF_LOCAL to %s\n",
1914 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1915
1916 status = xs_local_finish_connecting(xprt, sock);
40b5ea0c 1917 trace_rpc_socket_connect(xprt, sock, status);
176e21ee
CL		 1918 switch (status) {
1919 case 0:
1920 dprintk("RPC: xprt %p connected to %s\n",
1921 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1922 xprt_set_connected(xprt);
1923 break;
1924 case -ENOENT:
1925 dprintk("RPC: xprt %p: socket %s does not exist\n",
1926 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1927 break;
4a20a988
TM		 1928 case -ECONNREFUSED:
1929 dprintk("RPC: xprt %p: connection refused for %s\n",
1930 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1931 break;
176e21ee
CL		 1932 default:
1933 printk(KERN_ERR "%s: unhandled error (%d) connecting to %s\n",
1934 __func__, -status,
1935 xprt->address_strings[RPC_DISPLAY_ADDR]);
1936 }
1937
1938out:
1939 xprt_clear_connecting(xprt);
1940 xprt_wake_pending_tasks(xprt, status);
5cf02d09 1941 current->flags &= ~PF_FSTRANS;
dc107402
BF		 1942 return status;
1943}
1944
b6669737 1945static void xs_local_connect(struct rpc_xprt *xprt, struct rpc_task *task)
dc107402 1946{
dc107402
BF		 1947 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1948 int ret;
1949
1950 if (RPC_IS_ASYNC(task)) {
1951 /*
1952 * We want the AF_LOCAL connect to be resolved in the
1953 * filesystem namespace of the process making the rpc
1954 * call. Thus we connect synchronously.
1955 *
1956 * If we want to support asynchronous AF_LOCAL calls,
1957 * we'll need to figure out how to pass a namespace to
1958 * connect.
1959 */
1960 rpc_exit(task, -ENOTCONN);
1961 return;
1962 }
1963 ret = xs_local_setup_socket(transport);
1964 if (ret && !RPC_IS_SOFTCONN(task))
1965 msleep_interruptible(15000);
176e21ee
CL		 1966}
1967
a564b8f0
MG		 1968#ifdef CONFIG_SUNRPC_SWAP
1969static void xs_set_memalloc(struct rpc_xprt *xprt)
1970{
1971 struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1972 xprt);
1973
1974 if (xprt->swapper)
1975 sk_set_memalloc(transport->inet);
1976}
1977
1978/**
1979 * xs_swapper - Tag this transport as being used for swap.
1980 * @xprt: transport to tag
1981 * @enable: enable/disable
1982 *
1983 */
1984int xs_swapper(struct rpc_xprt *xprt, int enable)
1985{
1986 struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1987 xprt);
1988 int err = 0;
1989
1990 if (enable) {
1991 xprt->swapper++;
1992 xs_set_memalloc(xprt);
1993 } else if (xprt->swapper) {
1994 xprt->swapper--;
1995 sk_clear_memalloc(transport->inet);
1996 }
1997
1998 return err;
1999}
2000EXPORT_SYMBOL_GPL(xs_swapper);
2001#else
2002static void xs_set_memalloc(struct rpc_xprt *xprt)
2003{
2004}
2005#endif
2006
16be2d20
CL		 2007static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2008{
2009 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2010
2011 if (!transport->inet) {
2012 struct sock *sk = sock->sk;
2013
2014 write_lock_bh(&sk->sk_callback_lock);
2015
2a9e1cfa
TM		 2016 xs_save_old_callbacks(transport, sk);
2017
16be2d20 2018 sk->sk_user_data = xprt;
16be2d20
CL		 2019 sk->sk_data_ready = xs_udp_data_ready;
2020 sk->sk_write_space = xs_udp_write_space;
2021 sk->sk_no_check = UDP_CSUM_NORCV;
2022 sk->sk_allocation = GFP_ATOMIC;
2023
2024 xprt_set_connected(xprt);
2025
2026 /* Reset to new socket */
2027 transport->sock = sock;
2028 transport->inet = sk;
2029
a564b8f0
MG		 2030 xs_set_memalloc(xprt);
2031
16be2d20
CL		 2032 write_unlock_bh(&sk->sk_callback_lock);
2033 }
2034 xs_udp_do_set_buffer_size(xprt);
2035}
2036
8c14ff2a 2037static void xs_udp_setup_socket(struct work_struct *work)
a246b010 2038{
34161db6
TM		 2039 struct sock_xprt *transport =
2040 container_of(work, struct sock_xprt, connect_worker.work);
c8475461 2041 struct rpc_xprt *xprt = &transport->xprt;
ee0ac0c2 2042 struct socket *sock = transport->sock;
b65c0310 2043 int status = -EIO;
9903cd1c 2044
5cf02d09
JL		 2045 current->flags |= PF_FSTRANS;
2046
b0d93ad5 2047 /* Start by resetting any existing state */
fe315e76 2048 xs_reset_transport(transport);
8c14ff2a
PE		 2049 sock = xs_create_sock(xprt, transport,
2050 xs_addr(xprt)->sa_family, SOCK_DGRAM, IPPROTO_UDP);
b65c0310 2051 if (IS_ERR(sock))
b0d93ad5 2052 goto out;
68e220bd 2053
c740eff8
CL		 2054 dprintk("RPC: worker connecting xprt %p via %s to "
2055 "%s (port %s)\n", xprt,
2056 xprt->address_strings[RPC_DISPLAY_PROTO],
2057 xprt->address_strings[RPC_DISPLAY_ADDR],
2058 xprt->address_strings[RPC_DISPLAY_PORT]);
68e220bd
CL		 2059
2060 xs_udp_finish_connecting(xprt, sock);
40b5ea0c 2061 trace_rpc_socket_connect(xprt, sock, 0);
b0d93ad5
CL		 2062 status = 0;
2063out:
b0d93ad5 2064 xprt_clear_connecting(xprt);
7d1e8255 2065 xprt_wake_pending_tasks(xprt, status);
5cf02d09 2066 current->flags &= ~PF_FSTRANS;
a246b010
CL		 2067}
2068
3167e12c
CL		 2069/*
2070 * We need to preserve the port number so the reply cache on the server can
2071 * find our cached RPC replies when we get around to reconnecting.
2072 */
58dddac9 2073static void xs_abort_connection(struct sock_xprt *transport)
3167e12c
CL		 2074{
2075 int result;
3167e12c
CL		 2076 struct sockaddr any;
2077
58dddac9 2078 dprintk("RPC: disconnecting xprt %p to reuse port\n", transport);
3167e12c
CL		 2079
2080 /*
2081 * Disconnect the transport socket by doing a connect operation
2082 * with AF_UNSPEC. This should return immediately...
2083 */
2084 memset(&any, 0, sizeof(any));
2085 any.sa_family = AF_UNSPEC;
ee0ac0c2 2086 result = kernel_connect(transport->sock, &any, sizeof(any), 0);
40b5ea0c
TM		 2087 trace_rpc_socket_reset_connection(&transport->xprt,
2088 transport->sock, result);
7d1e8255 2089 if (!result)
4bc1e68e
TM		 2090 xs_sock_reset_connection_flags(&transport->xprt);
2091 dprintk("RPC: AF_UNSPEC connect return code %d\n", result);
3167e12c
CL		 2092}
2093
58dddac9 2094static void xs_tcp_reuse_connection(struct sock_xprt *transport)
40d2549d
TM		 2095{
2096 unsigned int state = transport->inet->sk_state;
2097
669502ff
AC		 2098 if (state == TCP_CLOSE && transport->sock->state == SS_UNCONNECTED) {
2099 /* we don't need to abort the connection if the socket
2100 * hasn't undergone a shutdown
2101 */
2102 if (transport->inet->sk_shutdown == 0)
2103 return;
2104 dprintk("RPC: %s: TCP_CLOSEd and sk_shutdown set to %d\n",
2105 __func__, transport->inet->sk_shutdown);
2106 }
2107 if ((1 << state) & (TCPF_ESTABLISHED|TCPF_SYN_SENT)) {
2108 /* we don't need to abort the connection if the socket
2109 * hasn't undergone a shutdown
2110 */
2111 if (transport->inet->sk_shutdown == 0)
2112 return;
2113 dprintk("RPC: %s: ESTABLISHED/SYN_SENT "
2114 "sk_shutdown set to %d\n",
2115 __func__, transport->inet->sk_shutdown);
2116 }
58dddac9 2117 xs_abort_connection(transport);
40d2549d
TM		 2118}
2119
16be2d20 2120static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
a246b010 2121{
16be2d20 2122 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
fe19a96b 2123 int ret = -ENOTCONN;
edb267a6 2124
ee0ac0c2 2125 if (!transport->inet) {
b0d93ad5 2126 struct sock *sk = sock->sk;
7f260e85
TM		 2127 unsigned int keepidle = xprt->timeout->to_initval / HZ;
2128 unsigned int keepcnt = xprt->timeout->to_retries + 1;
2129 unsigned int opt_on = 1;
2130
2131 /* TCP Keepalive options */
2132 kernel_setsockopt(sock, SOL_SOCKET, SO_KEEPALIVE,
2133 (char *)&opt_on, sizeof(opt_on));
2134 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPIDLE,
2135 (char *)&keepidle, sizeof(keepidle));
2136 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPINTVL,
2137 (char *)&keepidle, sizeof(keepidle));
2138 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPCNT,
2139 (char *)&keepcnt, sizeof(keepcnt));
b0d93ad5
CL		 2140
2141 write_lock_bh(&sk->sk_callback_lock);
2142
2a9e1cfa
TM		 2143 xs_save_old_callbacks(transport, sk);
2144
b0d93ad5 2145 sk->sk_user_data = xprt;
b0d93ad5
CL		 2146 sk->sk_data_ready = xs_tcp_data_ready;
2147 sk->sk_state_change = xs_tcp_state_change;
2148 sk->sk_write_space = xs_tcp_write_space;
b079fa7b 2149 sk->sk_allocation = GFP_ATOMIC;
3167e12c
CL		 2150
2151 /* socket options */
2152 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
2153 sock_reset_flag(sk, SOCK_LINGER);
2154 tcp_sk(sk)->linger2 = 0;
2155 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
b0d93ad5
CL		 2156
2157 xprt_clear_connected(xprt);
2158
2159 /* Reset to new socket */
ee0ac0c2
CL		 2160 transport->sock = sock;
2161 transport->inet = sk;
b0d93ad5
CL		 2162
2163 write_unlock_bh(&sk->sk_callback_lock);
2164 }
2165
01d37c42 2166 if (!xprt_bound(xprt))
fe19a96b 2167 goto out;
01d37c42 2168
a564b8f0
MG		 2169 xs_set_memalloc(xprt);
2170
b0d93ad5 2171 /* Tell the socket layer to start connecting... */
262ca07d
CL		 2172 xprt->stat.connect_count++;
2173 xprt->stat.connect_start = jiffies;
fe19a96b
TM		 2174 ret = kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
2175 switch (ret) {
2176 case 0:
2177 case -EINPROGRESS:
2178 /* SYN_SENT! */
fe19a96b
TM		 2179 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2180 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2181 }
2182out:
2183 return ret;
16be2d20
CL		 2184}
2185
9903cd1c 2186/**
b61d59ff
TM		 2187 * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint
2188 * @xprt: RPC transport to connect
2189 * @transport: socket transport to connect
2190 * @create_sock: function to create a socket of the correct type
9903cd1c
CL		 2191 *
2192 * Invoked by a work queue tasklet.
a246b010 2193 */
cdd518d5 2194static void xs_tcp_setup_socket(struct work_struct *work)
a246b010 2195{
cdd518d5
PE		 2196 struct sock_xprt *transport =
2197 container_of(work, struct sock_xprt, connect_worker.work);
ee0ac0c2 2198 struct socket *sock = transport->sock;
a9f5f0f7 2199 struct rpc_xprt *xprt = &transport->xprt;
b61d59ff 2200 int status = -EIO;
a246b010 2201
5cf02d09
JL		 2202 current->flags |= PF_FSTRANS;
2203
ee0ac0c2 2204 if (!sock) {
7d1e8255 2205 clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
cdd518d5
PE		 2206 sock = xs_create_sock(xprt, transport,
2207 xs_addr(xprt)->sa_family, SOCK_STREAM, IPPROTO_TCP);
b61d59ff
TM		 2208 if (IS_ERR(sock)) {
2209 status = PTR_ERR(sock);
3167e12c
CL		 2210 goto out;
2211 }
7d1e8255
TM		 2212 } else {
2213 int abort_and_exit;
a246b010 2214
7d1e8255
TM		 2215 abort_and_exit = test_and_clear_bit(XPRT_CONNECTION_ABORT,
2216 &xprt->state);
3167e12c 2217 /* "close" the socket, preserving the local port */
58dddac9 2218 xs_tcp_reuse_connection(transport);
a246b010 2219
7d1e8255
TM		 2220 if (abort_and_exit)
2221 goto out_eagain;
2222 }
a246b010 2223
c740eff8
CL		 2224 dprintk("RPC: worker connecting xprt %p via %s to "
2225 "%s (port %s)\n", xprt,
2226 xprt->address_strings[RPC_DISPLAY_PROTO],
2227 xprt->address_strings[RPC_DISPLAY_ADDR],
2228 xprt->address_strings[RPC_DISPLAY_PORT]);
edb267a6 2229
16be2d20 2230 status = xs_tcp_finish_connecting(xprt, sock);
40b5ea0c 2231 trace_rpc_socket_connect(xprt, sock, status);
46121cf7
CL		 2232 dprintk("RPC: %p connect status %d connected %d sock state %d\n",
2233 xprt, -status, xprt_connected(xprt),
2234 sock->sk->sk_state);
2a491991 2235 switch (status) {
f75e6745
TM		 2236 default:
2237 printk("%s: connect returned unhandled error %d\n",
2238 __func__, status);
2239 case -EADDRNOTAVAIL:
2240 /* We're probably in TIME_WAIT. Get rid of existing socket,
2241 * and retry
2242 */
a519fc7a 2243 xs_tcp_force_close(xprt);
88b5ed73 2244 break;
2a491991
TM		 2245 case 0:
2246 case -EINPROGRESS:
2247 case -EALREADY:
7d1e8255 2248 xprt_clear_connecting(xprt);
5cf02d09 2249 current->flags &= ~PF_FSTRANS;
7d1e8255 2250 return;
9fcfe0c8
TM		 2251 case -EINVAL:
2252 /* Happens, for instance, if the user specified a link
2253 * local IPv6 address without a scope-id.
2254 */
3ed5e2a2
TM		 2255 case -ECONNREFUSED:
2256 case -ECONNRESET:
2257 case -ENETUNREACH:
2258 /* retry with existing socket, after a delay */
9fcfe0c8 2259 goto out;
a246b010 2260 }
7d1e8255 2261out_eagain:
2a491991 2262 status = -EAGAIN;
a246b010 2263out:
68e220bd 2264 xprt_clear_connecting(xprt);
7d1e8255 2265 xprt_wake_pending_tasks(xprt, status);
5cf02d09 2266 current->flags &= ~PF_FSTRANS;
68e220bd 2267}
b0d93ad5 2268
9903cd1c
CL		 2269/**
2270 * xs_connect - connect a socket to a remote endpoint
1b092092 2271 * @xprt: pointer to transport structure
9903cd1c
CL		 2272 * @task: address of RPC task that manages state of connect request
2273 *
2274 * TCP: If the remote end dropped the connection, delay reconnecting.
03bf4b70
CL		 2275 *
2276 * UDP socket connects are synchronous, but we use a work queue anyway
2277 * to guarantee that even unprivileged user processes can set up a
2278 * socket on a privileged port.
2279 *
2280 * If a UDP socket connect fails, the delay behavior here prevents
2281 * retry floods (hard mounts).
9903cd1c 2282 */
1b092092 2283static void xs_connect(struct rpc_xprt *xprt, struct rpc_task *task)
a246b010 2284{
ee0ac0c2 2285 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
a246b010 2286
09a21c41 2287 if (transport->sock != NULL && !RPC_IS_SOFTCONN(task)) {
46121cf7
CL		 2288 dprintk("RPC: xs_connect delayed xprt %p for %lu "
2289 "seconds\n",
03bf4b70 2290 xprt, xprt->reestablish_timeout / HZ);
c1384c9c
TM		 2291 queue_delayed_work(rpciod_workqueue,
2292 &transport->connect_worker,
2293 xprt->reestablish_timeout);
03bf4b70 2294 xprt->reestablish_timeout <<= 1;
61d0a8e6
NB		 2295 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2296 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
03bf4b70
CL		 2297 if (xprt->reestablish_timeout > XS_TCP_MAX_REEST_TO)
2298 xprt->reestablish_timeout = XS_TCP_MAX_REEST_TO;
b0d93ad5 2299 } else {
46121cf7 2300 dprintk("RPC: xs_connect scheduled xprt %p\n", xprt);
c1384c9c
TM		 2301 queue_delayed_work(rpciod_workqueue,
2302 &transport->connect_worker, 0);
a246b010
CL		 2303 }
2304}
2305
176e21ee
CL		 2306/**
2307 * xs_local_print_stats - display AF_LOCAL socket-specifc stats
2308 * @xprt: rpc_xprt struct containing statistics
2309 * @seq: output file
2310 *
2311 */
2312static void xs_local_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2313{
2314 long idle_time = 0;
2315
2316 if (xprt_connected(xprt))
2317 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2318
2319 seq_printf(seq, "\txprt:\tlocal %lu %lu %lu %ld %lu %lu %lu "
15a45206 2320 "%llu %llu %lu %llu %llu\n",
176e21ee
CL		 2321 xprt->stat.bind_count,
2322 xprt->stat.connect_count,
2323 xprt->stat.connect_time,
2324 idle_time,
2325 xprt->stat.sends,
2326 xprt->stat.recvs,
2327 xprt->stat.bad_xids,
2328 xprt->stat.req_u,
15a45206
AA		 2329 xprt->stat.bklog_u,
2330 xprt->stat.max_slots,
2331 xprt->stat.sending_u,
2332 xprt->stat.pending_u);
176e21ee
CL		 2333}
2334
262ca07d
CL		 2335/**
2336 * xs_udp_print_stats - display UDP socket-specifc stats
2337 * @xprt: rpc_xprt struct containing statistics
2338 * @seq: output file
2339 *
2340 */
2341static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2342{
c8475461
CL		 2343 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2344
15a45206
AA		 2345 seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %llu %llu "
2346 "%lu %llu %llu\n",
fbfffbd5 2347 transport->srcport,
262ca07d
CL		 2348 xprt->stat.bind_count,
2349 xprt->stat.sends,
2350 xprt->stat.recvs,
2351 xprt->stat.bad_xids,
2352 xprt->stat.req_u,
15a45206
AA		 2353 xprt->stat.bklog_u,
2354 xprt->stat.max_slots,
2355 xprt->stat.sending_u,
2356 xprt->stat.pending_u);
262ca07d
CL		 2357}
2358
2359/**
2360 * xs_tcp_print_stats - display TCP socket-specifc stats
2361 * @xprt: rpc_xprt struct containing statistics
2362 * @seq: output file
2363 *
2364 */
2365static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2366{
c8475461 2367 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
262ca07d
CL		 2368 long idle_time = 0;
2369
2370 if (xprt_connected(xprt))
2371 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2372
15a45206
AA		 2373 seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu "
2374 "%llu %llu %lu %llu %llu\n",
fbfffbd5 2375 transport->srcport,
262ca07d
CL		 2376 xprt->stat.bind_count,
2377 xprt->stat.connect_count,
2378 xprt->stat.connect_time,
2379 idle_time,
2380 xprt->stat.sends,
2381 xprt->stat.recvs,
2382 xprt->stat.bad_xids,
2383 xprt->stat.req_u,
15a45206
AA		 2384 xprt->stat.bklog_u,
2385 xprt->stat.max_slots,
2386 xprt->stat.sending_u,
2387 xprt->stat.pending_u);
262ca07d
CL		 2388}
2389
4cfc7e60
RI		 2390/*
2391 * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason
2392 * we allocate pages instead doing a kmalloc like rpc_malloc is because we want
2393 * to use the server side send routines.
2394 */
5a51f13a 2395static void *bc_malloc(struct rpc_task *task, size_t size)
4cfc7e60
RI		 2396{
2397 struct page *page;
2398 struct rpc_buffer *buf;
2399
b8a13d03
WAA		 2400 WARN_ON_ONCE(size > PAGE_SIZE - sizeof(struct rpc_buffer));
2401 if (size > PAGE_SIZE - sizeof(struct rpc_buffer))
2402 return NULL;
4cfc7e60 2403
b8a13d03 2404 page = alloc_page(GFP_KERNEL);
4cfc7e60
RI		 2405 if (!page)
2406 return NULL;
2407
2408 buf = page_address(page);
2409 buf->len = PAGE_SIZE;
2410
2411 return buf->data;
2412}
2413
2414/*
2415 * Free the space allocated in the bc_alloc routine
2416 */
5a51f13a 2417static void bc_free(void *buffer)
4cfc7e60
RI		 2418{
2419 struct rpc_buffer *buf;
2420
2421 if (!buffer)
2422 return;
2423
2424 buf = container_of(buffer, struct rpc_buffer, data);
2425 free_page((unsigned long)buf);
2426}
2427
2428/*
2429 * Use the svc_sock to send the callback. Must be called with svsk->sk_mutex
2430 * held. Borrows heavily from svc_tcp_sendto and xs_tcp_send_request.
2431 */
2432static int bc_sendto(struct rpc_rqst *req)
2433{
2434 int len;
2435 struct xdr_buf *xbufp = &req->rq_snd_buf;
2436 struct rpc_xprt *xprt = req->rq_xprt;
2437 struct sock_xprt *transport =
2438 container_of(xprt, struct sock_xprt, xprt);
2439 struct socket *sock = transport->sock;
2440 unsigned long headoff;
2441 unsigned long tailoff;
2442
61677eee 2443 xs_encode_stream_record_marker(xbufp);
4cfc7e60
RI		 2444
2445 tailoff = (unsigned long)xbufp->tail[0].iov_base & ~PAGE_MASK;
2446 headoff = (unsigned long)xbufp->head[0].iov_base & ~PAGE_MASK;
2447 len = svc_send_common(sock, xbufp,
2448 virt_to_page(xbufp->head[0].iov_base), headoff,
2449 xbufp->tail[0].iov_base, tailoff);
2450
2451 if (len != xbufp->len) {
2452 printk(KERN_NOTICE "Error sending entire callback!\n");
2453 len = -EAGAIN;
2454 }
2455
2456 return len;
2457}
2458
2459/*
2460 * The send routine. Borrows from svc_send
2461 */
2462static int bc_send_request(struct rpc_task *task)
2463{
2464 struct rpc_rqst *req = task->tk_rqstp;
2465 struct svc_xprt *xprt;
4cfc7e60
RI		 2466 u32 len;
2467
2468 dprintk("sending request with xid: %08x\n", ntohl(req->rq_xid));
2469 /*
2470 * Get the server socket associated with this callback xprt
2471 */
2472 xprt = req->rq_xprt->bc_xprt;
4cfc7e60
RI		 2473
2474 /*
2475 * Grab the mutex to serialize data as the connection is shared
2476 * with the fore channel
2477 */
2478 if (!mutex_trylock(&xprt->xpt_mutex)) {
2479 rpc_sleep_on(&xprt->xpt_bc_pending, task, NULL);
2480 if (!mutex_trylock(&xprt->xpt_mutex))
2481 return -EAGAIN;
2482 rpc_wake_up_queued_task(&xprt->xpt_bc_pending, task);
2483 }
2484 if (test_bit(XPT_DEAD, &xprt->xpt_flags))
2485 len = -ENOTCONN;
2486 else
2487 len = bc_sendto(req);
2488 mutex_unlock(&xprt->xpt_mutex);
2489
2490 if (len > 0)
2491 len = 0;
2492
2493 return len;
2494}
2495
2496/*
2497 * The close routine. Since this is client initiated, we do nothing
2498 */
2499
2500static void bc_close(struct rpc_xprt *xprt)
2501{
4cfc7e60
RI		 2502}
2503
2504/*
2505 * The xprt destroy routine. Again, because this connection is client
2506 * initiated, we do nothing
2507 */
2508
2509static void bc_destroy(struct rpc_xprt *xprt)
2510{
4cfc7e60
RI		 2511}
2512
176e21ee
CL		 2513static struct rpc_xprt_ops xs_local_ops = {
2514 .reserve_xprt = xprt_reserve_xprt,
2515 .release_xprt = xs_tcp_release_xprt,
f39c1bfb 2516 .alloc_slot = xprt_alloc_slot,
176e21ee
CL		 2517 .rpcbind = xs_local_rpcbind,
2518 .set_port = xs_local_set_port,
dc107402 2519 .connect = xs_local_connect,
176e21ee
CL		 2520 .buf_alloc = rpc_malloc,
2521 .buf_free = rpc_free,
2522 .send_request = xs_local_send_request,
2523 .set_retrans_timeout = xprt_set_retrans_timeout_def,
2524 .close = xs_close,
a1311d87 2525 .destroy = xs_destroy,
176e21ee
CL		 2526 .print_stats = xs_local_print_stats,
2527};
2528
262965f5 2529static struct rpc_xprt_ops xs_udp_ops = {
43118c29 2530 .set_buffer_size = xs_udp_set_buffer_size,
12a80469 2531 .reserve_xprt = xprt_reserve_xprt_cong,
49e9a890 2532 .release_xprt = xprt_release_xprt_cong,
f39c1bfb 2533 .alloc_slot = xprt_alloc_slot,
45160d62 2534 .rpcbind = rpcb_getport_async,
92200412 2535 .set_port = xs_set_port,
262965f5 2536 .connect = xs_connect,
02107148
CL		 2537 .buf_alloc = rpc_malloc,
2538 .buf_free = rpc_free,
262965f5 2539 .send_request = xs_udp_send_request,
fe3aca29 2540 .set_retrans_timeout = xprt_set_retrans_timeout_rtt,
46c0ee8b 2541 .timer = xs_udp_timer,
a58dd398 2542 .release_request = xprt_release_rqst_cong,
262965f5
CL		 2543 .close = xs_close,
2544 .destroy = xs_destroy,
262ca07d 2545 .print_stats = xs_udp_print_stats,
262965f5
CL		 2546};
2547
2548static struct rpc_xprt_ops xs_tcp_ops = {
12a80469 2549 .reserve_xprt = xprt_reserve_xprt,
e0ab53de 2550 .release_xprt = xs_tcp_release_xprt,
f39c1bfb 2551 .alloc_slot = xprt_lock_and_alloc_slot,
45160d62 2552 .rpcbind = rpcb_getport_async,
92200412 2553 .set_port = xs_set_port,
0b9e7943 2554 .connect = xs_connect,
02107148
CL		 2555 .buf_alloc = rpc_malloc,
2556 .buf_free = rpc_free,
262965f5 2557 .send_request = xs_tcp_send_request,
fe3aca29 2558 .set_retrans_timeout = xprt_set_retrans_timeout_def,
f75e6745 2559 .close = xs_tcp_close,
9903cd1c 2560 .destroy = xs_destroy,
262ca07d 2561 .print_stats = xs_tcp_print_stats,
a246b010
CL		 2562};
2563
4cfc7e60
RI		 2564/*
2565 * The rpc_xprt_ops for the server backchannel
2566 */
2567
2568static struct rpc_xprt_ops bc_tcp_ops = {
2569 .reserve_xprt = xprt_reserve_xprt,
2570 .release_xprt = xprt_release_xprt,
84e28a30 2571 .alloc_slot = xprt_alloc_slot,
4cfc7e60
RI		 2572 .buf_alloc = bc_malloc,
2573 .buf_free = bc_free,
2574 .send_request = bc_send_request,
2575 .set_retrans_timeout = xprt_set_retrans_timeout_def,
2576 .close = bc_close,
2577 .destroy = bc_destroy,
2578 .print_stats = xs_tcp_print_stats,
2579};
2580
92476850
CL		 2581static int xs_init_anyaddr(const int family, struct sockaddr *sap)
2582{
2583 static const struct sockaddr_in sin = {
2584 .sin_family = AF_INET,
2585 .sin_addr.s_addr = htonl(INADDR_ANY),
2586 };
2587 static const struct sockaddr_in6 sin6 = {
2588 .sin6_family = AF_INET6,
2589 .sin6_addr = IN6ADDR_ANY_INIT,
2590 };
2591
2592 switch (family) {
176e21ee
CL		 2593 case AF_LOCAL:
2594 break;
92476850
CL		 2595 case AF_INET:
2596 memcpy(sap, &sin, sizeof(sin));
2597 break;
2598 case AF_INET6:
2599 memcpy(sap, &sin6, sizeof(sin6));
2600 break;
2601 default:
2602 dprintk("RPC: %s: Bad address family\n", __func__);
2603 return -EAFNOSUPPORT;
2604 }
2605 return 0;
2606}
2607
3c341b0b 2608static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
d9ba131d
TM		 2609 unsigned int slot_table_size,
2610 unsigned int max_slot_table_size)
c8541ecd
CL		 2611{
2612 struct rpc_xprt *xprt;
ffc2e518 2613 struct sock_xprt *new;
c8541ecd 2614
96802a09 2615 if (args->addrlen > sizeof(xprt->addr)) {
46121cf7 2616 dprintk("RPC: xs_setup_xprt: address too large\n");
c8541ecd
CL		 2617 return ERR_PTR(-EBADF);
2618 }
2619
d9ba131d
TM		 2620 xprt = xprt_alloc(args->net, sizeof(*new), slot_table_size,
2621 max_slot_table_size);
bd1722d4 2622 if (xprt == NULL) {
46121cf7
CL		 2623 dprintk("RPC: xs_setup_xprt: couldn't allocate "
2624 "rpc_xprt\n");
c8541ecd
CL		 2625 return ERR_PTR(-ENOMEM);
2626 }
c8541ecd 2627
bd1722d4 2628 new = container_of(xprt, struct sock_xprt, xprt);
96802a09
FM		 2629 memcpy(&xprt->addr, args->dstaddr, args->addrlen);
2630 xprt->addrlen = args->addrlen;
d3bc9a1d 2631 if (args->srcaddr)
fbfffbd5 2632 memcpy(&new->srcaddr, args->srcaddr, args->addrlen);
92476850
CL		 2633 else {
2634 int err;
2635 err = xs_init_anyaddr(args->dstaddr->sa_family,
2636 (struct sockaddr *)&new->srcaddr);
2aa13531
SK		 2637 if (err != 0) {
2638 xprt_free(xprt);
92476850 2639 return ERR_PTR(err);
2aa13531 2640 }
92476850 2641 }
c8541ecd
CL		 2642
2643 return xprt;
2644}
2645
176e21ee
CL		 2646static const struct rpc_timeout xs_local_default_timeout = {
2647 .to_initval = 10 * HZ,
2648 .to_maxval = 10 * HZ,
2649 .to_retries = 2,
2650};
2651
2652/**
2653 * xs_setup_local - Set up transport to use an AF_LOCAL socket
2654 * @args: rpc transport creation arguments
2655 *
2656 * AF_LOCAL is a "tpi_cots_ord" transport, just like TCP
2657 */
2658static struct rpc_xprt *xs_setup_local(struct xprt_create *args)
2659{
2660 struct sockaddr_un *sun = (struct sockaddr_un *)args->dstaddr;
2661 struct sock_xprt *transport;
2662 struct rpc_xprt *xprt;
2663 struct rpc_xprt *ret;
2664
d9ba131d
TM		 2665 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2666 xprt_max_tcp_slot_table_entries);
176e21ee
CL		 2667 if (IS_ERR(xprt))
2668 return xprt;
2669 transport = container_of(xprt, struct sock_xprt, xprt);
2670
2671 xprt->prot = 0;
2672 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2673 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2674
2675 xprt->bind_timeout = XS_BIND_TO;
2676 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2677 xprt->idle_timeout = XS_IDLE_DISC_TO;
2678
2679 xprt->ops = &xs_local_ops;
2680 xprt->timeout = &xs_local_default_timeout;
2681
93dc41bd
N		 2682 INIT_DELAYED_WORK(&transport->connect_worker,
2683 xs_dummy_setup_socket);
2684
176e21ee
CL		 2685 switch (sun->sun_family) {
2686 case AF_LOCAL:
2687 if (sun->sun_path[0] != '/') {
2688 dprintk("RPC: bad AF_LOCAL address: %s\n",
2689 sun->sun_path);
2690 ret = ERR_PTR(-EINVAL);
2691 goto out_err;
2692 }
2693 xprt_set_bound(xprt);
176e21ee 2694 xs_format_peer_addresses(xprt, "local", RPCBIND_NETID_LOCAL);
7073ea87
BF		 2695 ret = ERR_PTR(xs_local_setup_socket(transport));
2696 if (ret)
2697 goto out_err;
176e21ee
CL		 2698 break;
2699 default:
2700 ret = ERR_PTR(-EAFNOSUPPORT);
2701 goto out_err;
2702 }
2703
2704 dprintk("RPC: set up xprt to %s via AF_LOCAL\n",
2705 xprt->address_strings[RPC_DISPLAY_ADDR]);
2706
2707 if (try_module_get(THIS_MODULE))
2708 return xprt;
2709 ret = ERR_PTR(-EINVAL);
2710out_err:
2711 xprt_free(xprt);
2712 return ret;
2713}
2714
2881ae74
TM		 2715static const struct rpc_timeout xs_udp_default_timeout = {
2716 .to_initval = 5 * HZ,
2717 .to_maxval = 30 * HZ,
2718 .to_increment = 5 * HZ,
2719 .to_retries = 5,
2720};
2721
9903cd1c
CL		 2722/**
2723 * xs_setup_udp - Set up transport to use a UDP socket
96802a09 2724 * @args: rpc transport creation arguments
9903cd1c
CL		 2725 *
2726 */
483066d6 2727static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
a246b010 2728{
8f9d5b1a 2729 struct sockaddr *addr = args->dstaddr;
c8541ecd 2730 struct rpc_xprt *xprt;
c8475461 2731 struct sock_xprt *transport;
0a68b0be 2732 struct rpc_xprt *ret;
a246b010 2733
d9ba131d
TM		 2734 xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries,
2735 xprt_udp_slot_table_entries);
c8541ecd
CL		 2736 if (IS_ERR(xprt))
2737 return xprt;
c8475461 2738 transport = container_of(xprt, struct sock_xprt, xprt);
a246b010 2739
ec739ef0 2740 xprt->prot = IPPROTO_UDP;
808012fb 2741 xprt->tsh_size = 0;
a246b010
CL		 2742 /* XXX: header size can vary due to auth type, IPv6, etc. */
2743 xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
2744
03bf4b70 2745 xprt->bind_timeout = XS_BIND_TO;
03bf4b70
CL		 2746 xprt->reestablish_timeout = XS_UDP_REEST_TO;
2747 xprt->idle_timeout = XS_IDLE_DISC_TO;
a246b010 2748
262965f5 2749 xprt->ops = &xs_udp_ops;
a246b010 2750
ba7392bb 2751 xprt->timeout = &xs_udp_default_timeout;
a246b010 2752
8f9d5b1a
CL		 2753 switch (addr->sa_family) {
2754 case AF_INET:
2755 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2756 xprt_set_bound(xprt);
2757
2758 INIT_DELAYED_WORK(&transport->connect_worker,
8c14ff2a 2759 xs_udp_setup_socket);
9dc3b095 2760 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
8f9d5b1a
CL		 2761 break;
2762 case AF_INET6:
2763 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2764 xprt_set_bound(xprt);
2765
2766 INIT_DELAYED_WORK(&transport->connect_worker,
8c14ff2a 2767 xs_udp_setup_socket);
9dc3b095 2768 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
8f9d5b1a
CL		 2769 break;
2770 default:
0a68b0be
BF		 2771 ret = ERR_PTR(-EAFNOSUPPORT);
2772 goto out_err;
8f9d5b1a
CL		 2773 }
2774
c740eff8
CL		 2775 if (xprt_bound(xprt))
2776 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2777 xprt->address_strings[RPC_DISPLAY_ADDR],
2778 xprt->address_strings[RPC_DISPLAY_PORT],
2779 xprt->address_strings[RPC_DISPLAY_PROTO]);
2780 else
2781 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
2782 xprt->address_strings[RPC_DISPLAY_ADDR],
2783 xprt->address_strings[RPC_DISPLAY_PROTO]);
edb267a6 2784
bc25571e
TT		 2785 if (try_module_get(THIS_MODULE))
2786 return xprt;
0a68b0be
BF		 2787 ret = ERR_PTR(-EINVAL);
2788out_err:
e204e621 2789 xprt_free(xprt);
0a68b0be 2790 return ret;
a246b010
CL		 2791}
2792
2881ae74
TM		 2793static const struct rpc_timeout xs_tcp_default_timeout = {
2794 .to_initval = 60 * HZ,
2795 .to_maxval = 60 * HZ,
2796 .to_retries = 2,
2797};
2798
9903cd1c
CL		 2799/**
2800 * xs_setup_tcp - Set up transport to use a TCP socket
96802a09 2801 * @args: rpc transport creation arguments
9903cd1c
CL		 2802 *
2803 */
483066d6 2804static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
a246b010 2805{
8f9d5b1a 2806 struct sockaddr *addr = args->dstaddr;
c8541ecd 2807 struct rpc_xprt *xprt;
c8475461 2808 struct sock_xprt *transport;
0a68b0be 2809 struct rpc_xprt *ret;
b7993ceb
TM		 2810 unsigned int max_slot_table_size = xprt_max_tcp_slot_table_entries;
2811
2812 if (args->flags & XPRT_CREATE_INFINITE_SLOTS)
2813 max_slot_table_size = RPC_MAX_SLOT_TABLE_LIMIT;
a246b010 2814
d9ba131d 2815 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
b7993ceb 2816 max_slot_table_size);
c8541ecd
CL		 2817 if (IS_ERR(xprt))
2818 return xprt;
c8475461 2819 transport = container_of(xprt, struct sock_xprt, xprt);
a246b010 2820
ec739ef0 2821 xprt->prot = IPPROTO_TCP;
808012fb 2822 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
808012fb 2823 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
a246b010 2824
03bf4b70 2825 xprt->bind_timeout = XS_BIND_TO;
03bf4b70
CL		 2826 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2827 xprt->idle_timeout = XS_IDLE_DISC_TO;
a246b010 2828
262965f5 2829 xprt->ops = &xs_tcp_ops;
ba7392bb 2830 xprt->timeout = &xs_tcp_default_timeout;
a246b010 2831
8f9d5b1a
CL		 2832 switch (addr->sa_family) {
2833 case AF_INET:
2834 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2835 xprt_set_bound(xprt);
2836
9dc3b095 2837 INIT_DELAYED_WORK(&transport->connect_worker,
cdd518d5 2838 xs_tcp_setup_socket);
9dc3b095 2839 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
8f9d5b1a
CL		 2840 break;
2841 case AF_INET6:
2842 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2843 xprt_set_bound(xprt);
2844
9dc3b095 2845 INIT_DELAYED_WORK(&transport->connect_worker,
cdd518d5 2846 xs_tcp_setup_socket);
9dc3b095 2847 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
8f9d5b1a
CL		 2848 break;
2849 default:
0a68b0be
BF		 2850 ret = ERR_PTR(-EAFNOSUPPORT);
2851 goto out_err;
8f9d5b1a
CL		 2852 }
2853
c740eff8
CL		 2854 if (xprt_bound(xprt))
2855 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2856 xprt->address_strings[RPC_DISPLAY_ADDR],
2857 xprt->address_strings[RPC_DISPLAY_PORT],
2858 xprt->address_strings[RPC_DISPLAY_PROTO]);
2859 else
2860 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
2861 xprt->address_strings[RPC_DISPLAY_ADDR],
2862 xprt->address_strings[RPC_DISPLAY_PROTO]);
2863
edb267a6 2864
bc25571e
TT		 2865 if (try_module_get(THIS_MODULE))
2866 return xprt;
0a68b0be
BF		 2867 ret = ERR_PTR(-EINVAL);
2868out_err:
e204e621 2869 xprt_free(xprt);
0a68b0be 2870 return ret;
a246b010 2871}
282b32e1 2872
f300baba
AB		 2873/**
2874 * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket
2875 * @args: rpc transport creation arguments
2876 *
2877 */
2878static struct rpc_xprt *xs_setup_bc_tcp(struct xprt_create *args)
2879{
2880 struct sockaddr *addr = args->dstaddr;
2881 struct rpc_xprt *xprt;
2882 struct sock_xprt *transport;
2883 struct svc_sock *bc_sock;
0a68b0be 2884 struct rpc_xprt *ret;
f300baba 2885
f0418aa4
BF		 2886 if (args->bc_xprt->xpt_bc_xprt) {
2887 /*
2888 * This server connection already has a backchannel
e3bfab18
BF		 2889 * transport; we can't create a new one, as we wouldn't
2890 * be able to match replies based on xid any more. So,
f0418aa4
BF		 2891 * reuse the already-existing one:
2892 */
2893 return args->bc_xprt->xpt_bc_xprt;
2894 }
d9ba131d
TM		 2895 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2896 xprt_tcp_slot_table_entries);
f300baba
AB		 2897 if (IS_ERR(xprt))
2898 return xprt;
2899 transport = container_of(xprt, struct sock_xprt, xprt);
2900
2901 xprt->prot = IPPROTO_TCP;
2902 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2903 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2904 xprt->timeout = &xs_tcp_default_timeout;
2905
2906 /* backchannel */
2907 xprt_set_bound(xprt);
2908 xprt->bind_timeout = 0;
f300baba
AB		 2909 xprt->reestablish_timeout = 0;
2910 xprt->idle_timeout = 0;
2911
f300baba
AB		 2912 xprt->ops = &bc_tcp_ops;
2913
2914 switch (addr->sa_family) {
2915 case AF_INET:
2916 xs_format_peer_addresses(xprt, "tcp",
2917 RPCBIND_NETID_TCP);
2918 break;
2919 case AF_INET6:
2920 xs_format_peer_addresses(xprt, "tcp",
2921 RPCBIND_NETID_TCP6);
2922 break;
2923 default:
0a68b0be
BF		 2924 ret = ERR_PTR(-EAFNOSUPPORT);
2925 goto out_err;
f300baba
AB		 2926 }
2927
50fa0d40
PE		 2928 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2929 xprt->address_strings[RPC_DISPLAY_ADDR],
2930 xprt->address_strings[RPC_DISPLAY_PORT],
2931 xprt->address_strings[RPC_DISPLAY_PROTO]);
f300baba 2932
99de8ea9
BF		 2933 /*
2934 * Once we've associated a backchannel xprt with a connection,
2935 * we want to keep it around as long as long as the connection
2936 * lasts, in case we need to start using it for a backchannel
2937 * again; this reference won't be dropped until bc_xprt is
2938 * destroyed.
2939 */
2940 xprt_get(xprt);
2941 args->bc_xprt->xpt_bc_xprt = xprt;
2942 xprt->bc_xprt = args->bc_xprt;
2943 bc_sock = container_of(args->bc_xprt, struct svc_sock, sk_xprt);
2944 transport->sock = bc_sock->sk_sock;
2945 transport->inet = bc_sock->sk_sk;
2946
f300baba
AB		 2947 /*
2948 * Since we don't want connections for the backchannel, we set
2949 * the xprt status to connected
2950 */
2951 xprt_set_connected(xprt);
2952
2953
2954 if (try_module_get(THIS_MODULE))
2955 return xprt;
99de8ea9 2956 xprt_put(xprt);
0a68b0be
BF		 2957 ret = ERR_PTR(-EINVAL);
2958out_err:
e204e621 2959 xprt_free(xprt);
0a68b0be 2960 return ret;
f300baba
AB		 2961}
2962
176e21ee
CL		 2963static struct xprt_class xs_local_transport = {
2964 .list = LIST_HEAD_INIT(xs_local_transport.list),
2965 .name = "named UNIX socket",
2966 .owner = THIS_MODULE,
2967 .ident = XPRT_TRANSPORT_LOCAL,
2968 .setup = xs_setup_local,
2969};
2970
bc25571e
TT		 2971static struct xprt_class xs_udp_transport = {
2972 .list = LIST_HEAD_INIT(xs_udp_transport.list),
2973 .name = "udp",
2974 .owner = THIS_MODULE,
f300baba 2975 .ident = XPRT_TRANSPORT_UDP,
bc25571e
TT		 2976 .setup = xs_setup_udp,
2977};
2978
2979static struct xprt_class xs_tcp_transport = {
2980 .list = LIST_HEAD_INIT(xs_tcp_transport.list),
2981 .name = "tcp",
2982 .owner = THIS_MODULE,
f300baba 2983 .ident = XPRT_TRANSPORT_TCP,
bc25571e
TT		 2984 .setup = xs_setup_tcp,
2985};
2986
f300baba
AB		 2987static struct xprt_class xs_bc_tcp_transport = {
2988 .list = LIST_HEAD_INIT(xs_bc_tcp_transport.list),
2989 .name = "tcp NFSv4.1 backchannel",
2990 .owner = THIS_MODULE,
2991 .ident = XPRT_TRANSPORT_BC_TCP,
2992 .setup = xs_setup_bc_tcp,
2993};
2994
282b32e1 2995/**
bc25571e 2996 * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
282b32e1
CL		 2997 *
2998 */
2999int init_socket_xprt(void)
3000{
fbf76683 3001#ifdef RPC_DEBUG
2b1bec5f 3002 if (!sunrpc_table_header)
0b4d4147 3003 sunrpc_table_header = register_sysctl_table(sunrpc_table);
fbf76683
CL		 3004#endif
3005
176e21ee 3006 xprt_register_transport(&xs_local_transport);
bc25571e
TT		 3007 xprt_register_transport(&xs_udp_transport);
3008 xprt_register_transport(&xs_tcp_transport);
f300baba 3009 xprt_register_transport(&xs_bc_tcp_transport);
bc25571e 3010
282b32e1
CL		 3011 return 0;
3012}
3013
3014/**
bc25571e 3015 * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
282b32e1
CL		 3016 *
3017 */
3018void cleanup_socket_xprt(void)
3019{
fbf76683
CL		 3020#ifdef RPC_DEBUG
3021 if (sunrpc_table_header) {
3022 unregister_sysctl_table(sunrpc_table_header);
3023 sunrpc_table_header = NULL;
3024 }
3025#endif
bc25571e 3026
176e21ee 3027 xprt_unregister_transport(&xs_local_transport);
bc25571e
TT		 3028 xprt_unregister_transport(&xs_udp_transport);
3029 xprt_unregister_transport(&xs_tcp_transport);
f300baba 3030 xprt_unregister_transport(&xs_bc_tcp_transport);
282b32e1 3031}
cbf11071 3032
9bbb9e5a
RR		 3033static int param_set_uint_minmax(const char *val,
3034 const struct kernel_param *kp,
cbf11071
TM		 3035 unsigned int min, unsigned int max)
3036{
3037 unsigned long num;
3038 int ret;
3039
3040 if (!val)
3041 return -EINVAL;
3042 ret = strict_strtoul(val, 0, &num);
3043 if (ret == -EINVAL || num < min || num > max)
3044 return -EINVAL;
3045 *((unsigned int *)kp->arg) = num;
3046 return 0;
3047}
3048
9bbb9e5a 3049static int param_set_portnr(const char *val, const struct kernel_param *kp)
cbf11071
TM		 3050{
3051 return param_set_uint_minmax(val, kp,
3052 RPC_MIN_RESVPORT,
3053 RPC_MAX_RESVPORT);
3054}
3055
9bbb9e5a
RR		 3056static struct kernel_param_ops param_ops_portnr = {
3057 .set = param_set_portnr,
3058 .get = param_get_uint,
3059};
3060
cbf11071
TM		 3061#define param_check_portnr(name, p) \
3062 __param_check(name, p, unsigned int);
3063
3064module_param_named(min_resvport, xprt_min_resvport, portnr, 0644);
3065module_param_named(max_resvport, xprt_max_resvport, portnr, 0644);
3066
9bbb9e5a
RR		 3067static int param_set_slot_table_size(const char *val,
3068 const struct kernel_param *kp)
cbf11071
TM		 3069{
3070 return param_set_uint_minmax(val, kp,
3071 RPC_MIN_SLOT_TABLE,
3072 RPC_MAX_SLOT_TABLE);
3073}
3074
9bbb9e5a
RR		 3075static struct kernel_param_ops param_ops_slot_table_size = {
3076 .set = param_set_slot_table_size,
3077 .get = param_get_uint,
3078};
3079
cbf11071
TM		 3080#define param_check_slot_table_size(name, p) \
3081 __param_check(name, p, unsigned int);
3082
d9ba131d
TM		 3083static int param_set_max_slot_table_size(const char *val,
3084 const struct kernel_param *kp)
3085{
3086 return param_set_uint_minmax(val, kp,
3087 RPC_MIN_SLOT_TABLE,
3088 RPC_MAX_SLOT_TABLE_LIMIT);
3089}
3090
3091static struct kernel_param_ops param_ops_max_slot_table_size = {
3092 .set = param_set_max_slot_table_size,
3093 .get = param_get_uint,
3094};
3095
3096#define param_check_max_slot_table_size(name, p) \
3097 __param_check(name, p, unsigned int);
3098
cbf11071
TM		 3099module_param_named(tcp_slot_table_entries, xprt_tcp_slot_table_entries,
3100 slot_table_size, 0644);
d9ba131d
TM		 3101module_param_named(tcp_max_slot_table_entries, xprt_max_tcp_slot_table_entries,
3102 max_slot_table_size, 0644);
cbf11071
TM		 3103module_param_named(udp_slot_table_entries, xprt_udp_slot_table_entries,
3104 slot_table_size, 0644);
3105
Linux 6.x block layer and io_uring tree(s)