projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
NFSD: Clear wcc data between compound ops
[linux-2.6-block.git] / net / sunrpc / xprtsock.c
CommitLineData
a246b010
CL		 1/*
2 * linux/net/sunrpc/xprtsock.c
3 *
4 * Client-side transport implementation for sockets.
5 *
113aa838
AC		 6 * TCP callback races fixes (C) 1998 Red Hat
7 * TCP send fixes (C) 1998 Red Hat
a246b010
CL		 8 * TCP NFS related read + write fixes
9 * (C) 1999 Dave Airlie, University of Limerick, Ireland <airlied@linux.ie>
10 *
11 * Rewrite of larges part of the code in order to stabilize TCP stuff.
12 * Fix behaviour when socket buffer is full.
13 * (C) 1999 Trond Myklebust <trond.myklebust@fys.uio.no>
55aa4f58
CL		 14 *
15 * IP socket transport implementation, (C) 2005 Chuck Lever <cel@netapp.com>
8f9d5b1a
CL		 16 *
17 * IPv6 support contributed by Gilles Quillard, Bull Open Source, 2005.
18 * <gilles.quillard@bull.net>
a246b010
CL		 19 */
20
21#include <linux/types.h>
176e21ee 22#include <linux/string.h>
a246b010 23#include <linux/slab.h>
bc25571e 24#include <linux/module.h>
a246b010 25#include <linux/capability.h>
a246b010
CL		 26#include <linux/pagemap.h>
27#include <linux/errno.h>
28#include <linux/socket.h>
29#include <linux/in.h>
30#include <linux/net.h>
31#include <linux/mm.h>
176e21ee 32#include <linux/un.h>
a246b010
CL		 33#include <linux/udp.h>
34#include <linux/tcp.h>
35#include <linux/sunrpc/clnt.h>
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 *);
2118071d 260 void (*old_error_report)(struct sock *);
ffc2e518
CL		 261};
262
e136d092
CL		 263/*
264 * TCP receive state flags
265 */
266#define TCP_RCV_LAST_FRAG (1UL << 0)
267#define TCP_RCV_COPY_FRAGHDR (1UL << 1)
268#define TCP_RCV_COPY_XID (1UL << 2)
269#define TCP_RCV_COPY_DATA (1UL << 3)
f4a2e418
RL		 270#define TCP_RCV_READ_CALLDIR (1UL << 4)
271#define TCP_RCV_COPY_CALLDIR (1UL << 5)
18dca02a
RL		 272
273/*
274 * TCP RPC flags
275 */
f4a2e418 276#define TCP_RPC_REPLY (1UL << 6)
e136d092 277
2118071d
TM		 278static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
279{
280 return (struct rpc_xprt *) sk->sk_user_data;
281}
282
95392c59
CL		 283static inline struct sockaddr *xs_addr(struct rpc_xprt *xprt)
284{
285 return (struct sockaddr *) &xprt->addr;
286}
287
176e21ee
CL		 288static inline struct sockaddr_un *xs_addr_un(struct rpc_xprt *xprt)
289{
290 return (struct sockaddr_un *) &xprt->addr;
291}
292
95392c59 293static inline struct sockaddr_in *xs_addr_in(struct rpc_xprt *xprt)
edb267a6 294{
95392c59
CL		 295 return (struct sockaddr_in *) &xprt->addr;
296}
297
298static inline struct sockaddr_in6 *xs_addr_in6(struct rpc_xprt *xprt)
299{
300 return (struct sockaddr_in6 *) &xprt->addr;
301}
302
c877b849 303static void xs_format_common_peer_addresses(struct rpc_xprt *xprt)
edb267a6 304{
c877b849 305 struct sockaddr *sap = xs_addr(xprt);
9dc3b095
CL		 306 struct sockaddr_in6 *sin6;
307 struct sockaddr_in *sin;
176e21ee 308 struct sockaddr_un *sun;
c877b849 309 char buf[128];
edb267a6 310
9dc3b095 311 switch (sap->sa_family) {
176e21ee
CL		 312 case AF_LOCAL:
313 sun = xs_addr_un(xprt);
314 strlcpy(buf, sun->sun_path, sizeof(buf));
315 xprt->address_strings[RPC_DISPLAY_ADDR] =
316 kstrdup(buf, GFP_KERNEL);
317 break;
9dc3b095 318 case AF_INET:
176e21ee
CL		 319 (void)rpc_ntop(sap, buf, sizeof(buf));
320 xprt->address_strings[RPC_DISPLAY_ADDR] =
321 kstrdup(buf, GFP_KERNEL);
9dc3b095 322 sin = xs_addr_in(xprt);
fc0b5791 323 snprintf(buf, sizeof(buf), "%08x", ntohl(sin->sin_addr.s_addr));
9dc3b095
CL		 324 break;
325 case AF_INET6:
176e21ee
CL		 326 (void)rpc_ntop(sap, buf, sizeof(buf));
327 xprt->address_strings[RPC_DISPLAY_ADDR] =
328 kstrdup(buf, GFP_KERNEL);
9dc3b095 329 sin6 = xs_addr_in6(xprt);
fc0b5791 330 snprintf(buf, sizeof(buf), "%pi6", &sin6->sin6_addr);
9dc3b095
CL		 331 break;
332 default:
333 BUG();
756805e7 334 }
176e21ee 335
9dc3b095 336 xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = kstrdup(buf, GFP_KERNEL);
edb267a6
CL		 337}
338
9dc3b095 339static void xs_format_common_peer_ports(struct rpc_xprt *xprt)
4b6473fb 340{
9dc3b095
CL		 341 struct sockaddr *sap = xs_addr(xprt);
342 char buf[128];
4b6473fb 343
81160e66 344 snprintf(buf, sizeof(buf), "%u", rpc_get_port(sap));
c877b849 345 xprt->address_strings[RPC_DISPLAY_PORT] = kstrdup(buf, GFP_KERNEL);
4b6473fb 346
81160e66 347 snprintf(buf, sizeof(buf), "%4hx", rpc_get_port(sap));
c877b849
CL		 348 xprt->address_strings[RPC_DISPLAY_HEX_PORT] = kstrdup(buf, GFP_KERNEL);
349}
4b6473fb 350
9dc3b095
CL		 351static void xs_format_peer_addresses(struct rpc_xprt *xprt,
352 const char *protocol,
353 const char *netid)
c877b849 354{
b454ae90 355 xprt->address_strings[RPC_DISPLAY_PROTO] = protocol;
b454ae90 356 xprt->address_strings[RPC_DISPLAY_NETID] = netid;
c877b849 357 xs_format_common_peer_addresses(xprt);
9dc3b095 358 xs_format_common_peer_ports(xprt);
edb267a6 359}
4b6473fb 360
9dc3b095 361static void xs_update_peer_port(struct rpc_xprt *xprt)
4b6473fb 362{
9dc3b095
CL		 363 kfree(xprt->address_strings[RPC_DISPLAY_HEX_PORT]);
364 kfree(xprt->address_strings[RPC_DISPLAY_PORT]);
4417c8c4 365
9dc3b095 366 xs_format_common_peer_ports(xprt);
edb267a6
CL		 367}
368
369static void xs_free_peer_addresses(struct rpc_xprt *xprt)
370{
33e01dc7
CL		 371 unsigned int i;
372
373 for (i = 0; i < RPC_DISPLAY_MAX; i++)
374 switch (i) {
375 case RPC_DISPLAY_PROTO:
376 case RPC_DISPLAY_NETID:
377 continue;
378 default:
379 kfree(xprt->address_strings[i]);
380 }
edb267a6
CL		 381}
382
b4b5cc85
CL		 383#define XS_SENDMSG_FLAGS (MSG_DONTWAIT | MSG_NOSIGNAL)
384
24c5684b 385static int xs_send_kvec(struct socket *sock, struct sockaddr *addr, int addrlen, struct kvec *vec, unsigned int base, int more)
b4b5cc85 386{
b4b5cc85
CL		 387 struct msghdr msg = {
388 .msg_name = addr,
389 .msg_namelen = addrlen,
24c5684b
TM		 390 .msg_flags = XS_SENDMSG_FLAGS | (more ? MSG_MORE : 0),
391 };
392 struct kvec iov = {
393 .iov_base = vec->iov_base + base,
394 .iov_len = vec->iov_len - base,
b4b5cc85
CL		 395 };
396
24c5684b 397 if (iov.iov_len != 0)
b4b5cc85
CL		 398 return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
399 return kernel_sendmsg(sock, &msg, NULL, 0, 0);
400}
401
a6b31d18 402static int xs_send_pagedata(struct socket *sock, struct xdr_buf *xdr, unsigned int base, int more, bool zerocopy)
b4b5cc85 403{
a6b31d18
TM		 404 ssize_t (*do_sendpage)(struct socket *sock, struct page *page,
405 int offset, size_t size, int flags);
24c5684b
TM		 406 struct page **ppage;
407 unsigned int remainder;
408 int err, sent = 0;
409
410 remainder = xdr->page_len - base;
411 base += xdr->page_base;
412 ppage = xdr->pages + (base >> PAGE_SHIFT);
413 base &= ~PAGE_MASK;
a6b31d18
TM		 414 do_sendpage = sock->ops->sendpage;
415 if (!zerocopy)
416 do_sendpage = sock_no_sendpage;
24c5684b
TM		 417 for(;;) {
418 unsigned int len = min_t(unsigned int, PAGE_SIZE - base, remainder);
419 int flags = XS_SENDMSG_FLAGS;
b4b5cc85 420
24c5684b
TM		 421 remainder -= len;
422 if (remainder != 0 || more)
423 flags |= MSG_MORE;
a6b31d18 424 err = do_sendpage(sock, *ppage, base, len, flags);
24c5684b
TM		 425 if (remainder == 0 || err != len)
426 break;
427 sent += err;
428 ppage++;
429 base = 0;
430 }
431 if (sent == 0)
432 return err;
433 if (err > 0)
434 sent += err;
435 return sent;
b4b5cc85
CL		 436}
437
9903cd1c
CL		 438/**
439 * xs_sendpages - write pages directly to a socket
440 * @sock: socket to send on
441 * @addr: UDP only -- address of destination
442 * @addrlen: UDP only -- length of destination address
443 * @xdr: buffer containing this request
444 * @base: starting position in the buffer
a6b31d18 445 * @zerocopy: true if it is safe to use sendpage()
9903cd1c 446 *
a246b010 447 */
a6b31d18 448static int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base, bool zerocopy)
a246b010 449{
24c5684b
TM		 450 unsigned int remainder = xdr->len - base;
451 int err, sent = 0;
a246b010 452
262965f5 453 if (unlikely(!sock))
fba91afb 454 return -ENOTSOCK;
262965f5
CL		 455
456 clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags);
24c5684b
TM		 457 if (base != 0) {
458 addr = NULL;
459 addrlen = 0;
460 }
262965f5 461
24c5684b
TM		 462 if (base < xdr->head[0].iov_len || addr != NULL) {
463 unsigned int len = xdr->head[0].iov_len - base;
464 remainder -= len;
465 err = xs_send_kvec(sock, addr, addrlen, &xdr->head[0], base, remainder != 0);
466 if (remainder == 0 || err != len)
a246b010 467 goto out;
24c5684b 468 sent += err;
a246b010
CL		 469 base = 0;
470 } else
24c5684b 471 base -= xdr->head[0].iov_len;
a246b010 472
24c5684b
TM		 473 if (base < xdr->page_len) {
474 unsigned int len = xdr->page_len - base;
475 remainder -= len;
a6b31d18 476 err = xs_send_pagedata(sock, xdr, base, remainder != 0, zerocopy);
24c5684b 477 if (remainder == 0 || err != len)
a246b010 478 goto out;
24c5684b 479 sent += err;
a246b010 480 base = 0;
24c5684b
TM		 481 } else
482 base -= xdr->page_len;
483
484 if (base >= xdr->tail[0].iov_len)
485 return sent;
486 err = xs_send_kvec(sock, NULL, 0, &xdr->tail[0], base, 0);
a246b010 487out:
24c5684b
TM		 488 if (sent == 0)
489 return err;
490 if (err > 0)
491 sent += err;
492 return sent;
a246b010
CL		 493}
494
b6ddf64f
TM		 495static void xs_nospace_callback(struct rpc_task *task)
496{
497 struct sock_xprt *transport = container_of(task->tk_rqstp->rq_xprt, struct sock_xprt, xprt);
498
499 transport->inet->sk_write_pending--;
500 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
501}
502
9903cd1c 503/**
262965f5
CL		 504 * xs_nospace - place task on wait queue if transmit was incomplete
505 * @task: task to put to sleep
9903cd1c 506 *
a246b010 507 */
5e3771ce 508static int xs_nospace(struct rpc_task *task)
a246b010 509{
262965f5
CL		 510 struct rpc_rqst *req = task->tk_rqstp;
511 struct rpc_xprt *xprt = req->rq_xprt;
ee0ac0c2 512 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
24ca9a84 513 int ret = -EAGAIN;
a246b010 514
46121cf7 515 dprintk("RPC: %5u xmit incomplete (%u left of %u)\n",
262965f5
CL		 516 task->tk_pid, req->rq_slen - req->rq_bytes_sent,
517 req->rq_slen);
518
b6ddf64f
TM		 519 /* Protect against races with write_space */
520 spin_lock_bh(&xprt->transport_lock);
521
522 /* Don't race with disconnect */
523 if (xprt_connected(xprt)) {
524 if (test_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags)) {
525 /*
526 * Notify TCP that we're limited by the application
527 * window size
528 */
529 set_bit(SOCK_NOSPACE, &transport->sock->flags);
530 transport->inet->sk_write_pending++;
531 /* ...and wait for more buffer space */
532 xprt_wait_for_buffer_space(task, xs_nospace_callback);
533 }
534 } else {
535 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
5e3771ce 536 ret = -ENOTCONN;
b6ddf64f 537 }
262965f5 538
b6ddf64f 539 spin_unlock_bh(&xprt->transport_lock);
5e3771ce 540 return ret;
262965f5
CL		 541}
542
61677eee
CL		 543/*
544 * Construct a stream transport record marker in @buf.
545 */
546static inline void xs_encode_stream_record_marker(struct xdr_buf *buf)
547{
548 u32 reclen = buf->len - sizeof(rpc_fraghdr);
549 rpc_fraghdr *base = buf->head[0].iov_base;
550 *base = cpu_to_be32(RPC_LAST_STREAM_FRAGMENT | reclen);
551}
552
176e21ee
CL		 553/**
554 * xs_local_send_request - write an RPC request to an AF_LOCAL socket
555 * @task: RPC task that manages the state of an RPC request
556 *
557 * Return values:
558 * 0: The request has been sent
559 * EAGAIN: The socket was blocked, please call again later to
560 * complete the request
561 * ENOTCONN: Caller needs to invoke connect logic then call again
562 * other: Some other error occured, the request was not sent
563 */
564static int xs_local_send_request(struct rpc_task *task)
565{
566 struct rpc_rqst *req = task->tk_rqstp;
567 struct rpc_xprt *xprt = req->rq_xprt;
568 struct sock_xprt *transport =
569 container_of(xprt, struct sock_xprt, xprt);
570 struct xdr_buf *xdr = &req->rq_snd_buf;
571 int status;
572
573 xs_encode_stream_record_marker(&req->rq_snd_buf);
574
575 xs_pktdump("packet data:",
576 req->rq_svec->iov_base, req->rq_svec->iov_len);
577
578 status = xs_sendpages(transport->sock, NULL, 0,
a6b31d18 579 xdr, req->rq_bytes_sent, true);
176e21ee
CL		 580 dprintk("RPC: %s(%u) = %d\n",
581 __func__, xdr->len - req->rq_bytes_sent, status);
582 if (likely(status >= 0)) {
583 req->rq_bytes_sent += status;
584 req->rq_xmit_bytes_sent += status;
585 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
586 req->rq_bytes_sent = 0;
587 return 0;
588 }
589 status = -EAGAIN;
590 }
591
592 switch (status) {
593 case -EAGAIN:
594 status = xs_nospace(task);
595 break;
596 default:
597 dprintk("RPC: sendmsg returned unrecognized error %d\n",
598 -status);
599 case -EPIPE:
600 xs_close(xprt);
601 status = -ENOTCONN;
602 }
603
604 return status;
605}
606
262965f5
CL		 607/**
608 * xs_udp_send_request - write an RPC request to a UDP socket
609 * @task: address of RPC task that manages the state of an RPC request
610 *
611 * Return values:
612 * 0: The request has been sent
613 * EAGAIN: The socket was blocked, please call again later to
614 * complete the request
615 * ENOTCONN: Caller needs to invoke connect logic then call again
25985edc 616 * other: Some other error occurred, the request was not sent
262965f5
CL		 617 */
618static int xs_udp_send_request(struct rpc_task *task)
619{
620 struct rpc_rqst *req = task->tk_rqstp;
621 struct rpc_xprt *xprt = req->rq_xprt;
ee0ac0c2 622 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
262965f5
CL		 623 struct xdr_buf *xdr = &req->rq_snd_buf;
624 int status;
a246b010 625
9903cd1c 626 xs_pktdump("packet data:",
a246b010
CL		 627 req->rq_svec->iov_base,
628 req->rq_svec->iov_len);
629
01d37c42
TM		 630 if (!xprt_bound(xprt))
631 return -ENOTCONN;
ee0ac0c2 632 status = xs_sendpages(transport->sock,
95392c59 633 xs_addr(xprt),
ee0ac0c2 634 xprt->addrlen, xdr,
a6b31d18 635 req->rq_bytes_sent, true);
a246b010 636
46121cf7 637 dprintk("RPC: xs_udp_send_request(%u) = %d\n",
262965f5 638 xdr->len - req->rq_bytes_sent, status);
a246b010 639
2199700f 640 if (status >= 0) {
d60dbb20 641 req->rq_xmit_bytes_sent += status;
2199700f
TM		 642 if (status >= req->rq_slen)
643 return 0;
644 /* Still some bytes left; set up for a retry later. */
262965f5 645 status = -EAGAIN;
2199700f 646 }
a246b010 647
262965f5 648 switch (status) {
fba91afb
TM		 649 case -ENOTSOCK:
650 status = -ENOTCONN;
651 /* Should we call xs_close() here? */
652 break;
b6ddf64f 653 case -EAGAIN:
5e3771ce 654 status = xs_nospace(task);
b6ddf64f 655 break;
c8485e4d
TM		 656 default:
657 dprintk("RPC: sendmsg returned unrecognized error %d\n",
658 -status);
262965f5
CL		 659 case -ENETUNREACH:
660 case -EPIPE:
a246b010
CL		 661 case -ECONNREFUSED:
662 /* When the server has died, an ICMP port unreachable message
9903cd1c 663 * prompts ECONNREFUSED. */
b6ddf64f 664 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
a246b010 665 }
5fe46e9d 666
262965f5 667 return status;
a246b010
CL		 668}
669
e06799f9
TM		 670/**
671 * xs_tcp_shutdown - gracefully shut down a TCP socket
672 * @xprt: transport
673 *
674 * Initiates a graceful shutdown of the TCP socket by calling the
675 * equivalent of shutdown(SHUT_WR);
676 */
677static void xs_tcp_shutdown(struct rpc_xprt *xprt)
678{
679 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
680 struct socket *sock = transport->sock;
681
40b5ea0c 682 if (sock != NULL) {
e06799f9 683 kernel_sock_shutdown(sock, SHUT_WR);
40b5ea0c
TM		 684 trace_rpc_socket_shutdown(xprt, sock);
685 }
e06799f9
TM		 686}
687
9903cd1c 688/**
262965f5 689 * xs_tcp_send_request - write an RPC request to a TCP socket
9903cd1c
CL		 690 * @task: address of RPC task that manages the state of an RPC request
691 *
692 * Return values:
262965f5
CL		 693 * 0: The request has been sent
694 * EAGAIN: The socket was blocked, please call again later to
695 * complete the request
696 * ENOTCONN: Caller needs to invoke connect logic then call again
25985edc 697 * other: Some other error occurred, the request was not sent
9903cd1c
CL		 698 *
699 * XXX: In the case of soft timeouts, should we eventually give up
262965f5 700 * if sendmsg is not able to make progress?
9903cd1c 701 */
262965f5 702static int xs_tcp_send_request(struct rpc_task *task)
a246b010
CL		 703{
704 struct rpc_rqst *req = task->tk_rqstp;
705 struct rpc_xprt *xprt = req->rq_xprt;
ee0ac0c2 706 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
262965f5 707 struct xdr_buf *xdr = &req->rq_snd_buf;
a6b31d18 708 bool zerocopy = true;
b595bb15 709 int status;
a246b010 710
61677eee 711 xs_encode_stream_record_marker(&req->rq_snd_buf);
a246b010 712
262965f5
CL		 713 xs_pktdump("packet data:",
714 req->rq_svec->iov_base,
715 req->rq_svec->iov_len);
a6b31d18
TM		 716 /* Don't use zero copy if this is a resend. If the RPC call
717 * completes while the socket holds a reference to the pages,
718 * then we may end up resending corrupted data.
719 */
720 if (task->tk_flags & RPC_TASK_SENT)
721 zerocopy = false;
a246b010
CL		 722
723 /* Continue transmitting the packet/record. We must be careful
724 * to cope with writespace callbacks arriving _after_ we have
262965f5 725 * called sendmsg(). */
a246b010 726 while (1) {
ee0ac0c2 727 status = xs_sendpages(transport->sock,
a6b31d18
TM		 728 NULL, 0, xdr, req->rq_bytes_sent,
729 zerocopy);
a246b010 730
46121cf7 731 dprintk("RPC: xs_tcp_send_request(%u) = %d\n",
262965f5 732 xdr->len - req->rq_bytes_sent, status);
a246b010 733
262965f5 734 if (unlikely(status < 0))
a246b010 735 break;
a246b010 736
262965f5
CL		 737 /* If we've sent the entire packet, immediately
738 * reset the count of bytes sent. */
739 req->rq_bytes_sent += status;
d60dbb20 740 req->rq_xmit_bytes_sent += status;
262965f5
CL		 741 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
742 req->rq_bytes_sent = 0;
743 return 0;
744 }
a246b010 745
06b4b681
TM		 746 if (status != 0)
747 continue;
a246b010 748 status = -EAGAIN;
06b4b681 749 break;
a246b010
CL		 750 }
751
262965f5 752 switch (status) {
fba91afb
TM		 753 case -ENOTSOCK:
754 status = -ENOTCONN;
755 /* Should we call xs_close() here? */
756 break;
262965f5 757 case -EAGAIN:
5e3771ce 758 status = xs_nospace(task);
262965f5 759 break;
c8485e4d
TM		 760 default:
761 dprintk("RPC: sendmsg returned unrecognized error %d\n",
762 -status);
262965f5 763 case -ECONNRESET:
2a9e1cfa
TM		 764 xs_tcp_shutdown(xprt);
765 case -ECONNREFUSED:
262965f5 766 case -ENOTCONN:
b9d2bb2e 767 case -EPIPE:
b6ddf64f 768 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
a246b010 769 }
5fe46e9d 770
a246b010
CL		 771 return status;
772}
773
e0ab53de
TM		 774/**
775 * xs_tcp_release_xprt - clean up after a tcp transmission
776 * @xprt: transport
777 * @task: rpc task
778 *
779 * This cleans up if an error causes us to abort the transmission of a request.
780 * In this case, the socket may need to be reset in order to avoid confusing
781 * the server.
782 */
783static void xs_tcp_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
784{
785 struct rpc_rqst *req;
786
787 if (task != xprt->snd_task)
788 return;
789 if (task == NULL)
790 goto out_release;
791 req = task->tk_rqstp;
43cedbf0
TM		 792 if (req == NULL)
793 goto out_release;
e0ab53de
TM		 794 if (req->rq_bytes_sent == 0)
795 goto out_release;
796 if (req->rq_bytes_sent == req->rq_snd_buf.len)
797 goto out_release;
a4f0835c 798 set_bit(XPRT_CLOSE_WAIT, &xprt->state);
e0ab53de
TM		 799out_release:
800 xprt_release_xprt(xprt, task);
801}
802
2a9e1cfa
TM		 803static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk)
804{
805 transport->old_data_ready = sk->sk_data_ready;
806 transport->old_state_change = sk->sk_state_change;
807 transport->old_write_space = sk->sk_write_space;
2118071d 808 transport->old_error_report = sk->sk_error_report;
2a9e1cfa
TM		 809}
810
811static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk)
812{
813 sk->sk_data_ready = transport->old_data_ready;
814 sk->sk_state_change = transport->old_state_change;
815 sk->sk_write_space = transport->old_write_space;
2118071d
TM		 816 sk->sk_error_report = transport->old_error_report;
817}
818
819/**
820 * xs_error_report - callback to handle TCP socket state errors
821 * @sk: socket
822 *
823 * Note: we don't call sock_error() since there may be a rpc_task
824 * using the socket, and so we don't want to clear sk->sk_err.
825 */
826static void xs_error_report(struct sock *sk)
827{
828 struct rpc_xprt *xprt;
829 int err;
830
831 read_lock_bh(&sk->sk_callback_lock);
832 if (!(xprt = xprt_from_sock(sk)))
833 goto out;
834
835 err = -sk->sk_err;
836 if (err == 0)
837 goto out;
838 dprintk("RPC: xs_error_report client %p, error=%d...\n",
839 xprt, -err);
e8353c76 840 trace_rpc_socket_error(xprt, sk->sk_socket, err);
2118071d
TM		 841 xprt_wake_pending_tasks(xprt, err);
842 out:
843 read_unlock_bh(&sk->sk_callback_lock);
2a9e1cfa
TM		 844}
845
fe315e76 846static void xs_reset_transport(struct sock_xprt *transport)
a246b010 847{
ee0ac0c2
CL		 848 struct socket *sock = transport->sock;
849 struct sock *sk = transport->inet;
a246b010 850
fe315e76
CL		 851 if (sk == NULL)
852 return;
9903cd1c 853
246408dc
TM		 854 transport->srcport = 0;
855
a246b010 856 write_lock_bh(&sk->sk_callback_lock);
ee0ac0c2
CL		 857 transport->inet = NULL;
858 transport->sock = NULL;
a246b010 859
9903cd1c 860 sk->sk_user_data = NULL;
2a9e1cfa
TM		 861
862 xs_restore_old_callbacks(transport, sk);
a246b010
CL		 863 write_unlock_bh(&sk->sk_callback_lock);
864
9903cd1c 865 sk->sk_no_check = 0;
a246b010 866
40b5ea0c 867 trace_rpc_socket_close(&transport->xprt, sock);
a246b010 868 sock_release(sock);
fe315e76
CL		 869}
870
871/**
872 * xs_close - close a socket
873 * @xprt: transport
874 *
875 * This is used when all requests are complete; ie, no DRC state remains
876 * on the server we want to save.
f75e6745
TM		 877 *
878 * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with
879 * xs_reset_transport() zeroing the socket from underneath a writer.
fe315e76
CL		 880 */
881static void xs_close(struct rpc_xprt *xprt)
882{
883 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
884
885 dprintk("RPC: xs_close xprt %p\n", xprt);
886
93dc41bd
N		 887 cancel_delayed_work_sync(&transport->connect_worker);
888
fe315e76 889 xs_reset_transport(transport);
61d0a8e6 890 xprt->reestablish_timeout = 0;
fe315e76 891
632e3bdc 892 smp_mb__before_clear_bit();
7d1e8255 893 clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
632e3bdc 894 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
3b948ae5 895 clear_bit(XPRT_CLOSING, &xprt->state);
632e3bdc 896 smp_mb__after_clear_bit();
62da3b24 897 xprt_disconnect_done(xprt);
a246b010
CL		 898}
899
f75e6745
TM		 900static void xs_tcp_close(struct rpc_xprt *xprt)
901{
902 if (test_and_clear_bit(XPRT_CONNECTION_CLOSE, &xprt->state))
903 xs_close(xprt);
904 else
905 xs_tcp_shutdown(xprt);
906}
907
315f3812
KM		 908static void xs_xprt_free(struct rpc_xprt *xprt)
909{
910 xs_free_peer_addresses(xprt);
911 xprt_free(xprt);
912}
913
9903cd1c
CL		 914/**
915 * xs_destroy - prepare to shutdown a transport
916 * @xprt: doomed transport
917 *
918 */
919static void xs_destroy(struct rpc_xprt *xprt)
a246b010 920{
46121cf7 921 dprintk("RPC: xs_destroy xprt %p\n", xprt);
9903cd1c 922
a1311d87 923 xs_close(xprt);
315f3812 924 xs_xprt_free(xprt);
a1311d87 925 module_put(THIS_MODULE);
a246b010
CL		 926}
927
176e21ee
CL		 928static int xs_local_copy_to_xdr(struct xdr_buf *xdr, struct sk_buff *skb)
929{
930 struct xdr_skb_reader desc = {
931 .skb = skb,
932 .offset = sizeof(rpc_fraghdr),
933 .count = skb->len - sizeof(rpc_fraghdr),
934 };
935
936 if (xdr_partial_copy_from_skb(xdr, 0, &desc, xdr_skb_read_bits) < 0)
937 return -1;
938 if (desc.count)
939 return -1;
940 return 0;
941}
942
943/**
944 * xs_local_data_ready - "data ready" callback for AF_LOCAL sockets
945 * @sk: socket with data to read
946 * @len: how much data to read
947 *
948 * Currently this assumes we can read the whole reply in a single gulp.
949 */
950static void xs_local_data_ready(struct sock *sk, int len)
951{
952 struct rpc_task *task;
953 struct rpc_xprt *xprt;
954 struct rpc_rqst *rovr;
955 struct sk_buff *skb;
956 int err, repsize, copied;
957 u32 _xid;
958 __be32 *xp;
959
960 read_lock_bh(&sk->sk_callback_lock);
961 dprintk("RPC: %s...\n", __func__);
962 xprt = xprt_from_sock(sk);
963 if (xprt == NULL)
964 goto out;
965
966 skb = skb_recv_datagram(sk, 0, 1, &err);
967 if (skb == NULL)
968 goto out;
969
176e21ee
CL		 970 repsize = skb->len - sizeof(rpc_fraghdr);
971 if (repsize < 4) {
972 dprintk("RPC: impossible RPC reply size %d\n", repsize);
973 goto dropit;
974 }
975
976 /* Copy the XID from the skb... */
977 xp = skb_header_pointer(skb, sizeof(rpc_fraghdr), sizeof(_xid), &_xid);
978 if (xp == NULL)
979 goto dropit;
980
981 /* Look up and lock the request corresponding to the given XID */
982 spin_lock(&xprt->transport_lock);
983 rovr = xprt_lookup_rqst(xprt, *xp);
984 if (!rovr)
985 goto out_unlock;
986 task = rovr->rq_task;
987
988 copied = rovr->rq_private_buf.buflen;
989 if (copied > repsize)
990 copied = repsize;
991
992 if (xs_local_copy_to_xdr(&rovr->rq_private_buf, skb)) {
993 dprintk("RPC: sk_buff copy failed\n");
994 goto out_unlock;
995 }
996
997 xprt_complete_rqst(task, copied);
998
999 out_unlock:
1000 spin_unlock(&xprt->transport_lock);
1001 dropit:
1002 skb_free_datagram(sk, skb);
1003 out:
1004 read_unlock_bh(&sk->sk_callback_lock);
1005}
1006
9903cd1c
CL		 1007/**
1008 * xs_udp_data_ready - "data ready" callback for UDP sockets
1009 * @sk: socket with data to read
1010 * @len: how much data to read
1011 *
a246b010 1012 */
9903cd1c 1013static void xs_udp_data_ready(struct sock *sk, int len)
a246b010 1014{
9903cd1c
CL		 1015 struct rpc_task *task;
1016 struct rpc_xprt *xprt;
a246b010 1017 struct rpc_rqst *rovr;
9903cd1c 1018 struct sk_buff *skb;
a246b010 1019 int err, repsize, copied;
d8ed029d
AD		 1020 u32 _xid;
1021 __be32 *xp;
a246b010 1022
f064af1e 1023 read_lock_bh(&sk->sk_callback_lock);
46121cf7 1024 dprintk("RPC: xs_udp_data_ready...\n");
9903cd1c 1025 if (!(xprt = xprt_from_sock(sk)))
a246b010 1026 goto out;
a246b010
CL		 1027
1028 if ((skb = skb_recv_datagram(sk, 0, 1, &err)) == NULL)
1029 goto out;
1030
a246b010
CL		 1031 repsize = skb->len - sizeof(struct udphdr);
1032 if (repsize < 4) {
46121cf7 1033 dprintk("RPC: impossible RPC reply size %d!\n", repsize);
a246b010
CL		 1034 goto dropit;
1035 }
1036
1037 /* Copy the XID from the skb... */
1038 xp = skb_header_pointer(skb, sizeof(struct udphdr),
1039 sizeof(_xid), &_xid);
1040 if (xp == NULL)
1041 goto dropit;
1042
1043 /* Look up and lock the request corresponding to the given XID */
4a0f8c04 1044 spin_lock(&xprt->transport_lock);
a246b010
CL		 1045 rovr = xprt_lookup_rqst(xprt, *xp);
1046 if (!rovr)
1047 goto out_unlock;
1048 task = rovr->rq_task;
1049
a246b010
CL		 1050 if ((copied = rovr->rq_private_buf.buflen) > repsize)
1051 copied = repsize;
1052
1053 /* Suck it into the iovec, verify checksum if not done by hw. */
1781f7f5
HX		 1054 if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
1055 UDPX_INC_STATS_BH(sk, UDP_MIB_INERRORS);
a246b010 1056 goto out_unlock;
1781f7f5
HX		 1057 }
1058
1059 UDPX_INC_STATS_BH(sk, UDP_MIB_INDATAGRAMS);
a246b010 1060
6a24dfb6 1061 xprt_adjust_cwnd(xprt, task, copied);
1570c1e4 1062 xprt_complete_rqst(task, copied);
a246b010
CL		 1063
1064 out_unlock:
4a0f8c04 1065 spin_unlock(&xprt->transport_lock);
a246b010
CL		 1066 dropit:
1067 skb_free_datagram(sk, skb);
1068 out:
f064af1e 1069 read_unlock_bh(&sk->sk_callback_lock);
a246b010
CL		 1070}
1071
a519fc7a
TM		 1072/*
1073 * Helper function to force a TCP close if the server is sending
1074 * junk and/or it has put us in CLOSE_WAIT
1075 */
1076static void xs_tcp_force_close(struct rpc_xprt *xprt)
1077{
1078 set_bit(XPRT_CONNECTION_CLOSE, &xprt->state);
1079 xprt_force_disconnect(xprt);
1080}
1081
dd456471 1082static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
a246b010 1083{
51971139 1084 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
a246b010
CL		 1085 size_t len, used;
1086 char *p;
1087
51971139
CL		 1088 p = ((char *) &transport->tcp_fraghdr) + transport->tcp_offset;
1089 len = sizeof(transport->tcp_fraghdr) - transport->tcp_offset;
9d292316 1090 used = xdr_skb_read_bits(desc, p, len);
51971139 1091 transport->tcp_offset += used;
a246b010
CL		 1092 if (used != len)
1093 return;
808012fb 1094
51971139
CL		 1095 transport->tcp_reclen = ntohl(transport->tcp_fraghdr);
1096 if (transport->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
e136d092 1097 transport->tcp_flags |= TCP_RCV_LAST_FRAG;
a246b010 1098 else
e136d092 1099 transport->tcp_flags &= ~TCP_RCV_LAST_FRAG;
51971139 1100 transport->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
808012fb 1101
e136d092 1102 transport->tcp_flags &= ~TCP_RCV_COPY_FRAGHDR;
51971139 1103 transport->tcp_offset = 0;
808012fb 1104
a246b010 1105 /* Sanity check of the record length */
18dca02a 1106 if (unlikely(transport->tcp_reclen < 8)) {
46121cf7 1107 dprintk("RPC: invalid TCP record fragment length\n");
a519fc7a 1108 xs_tcp_force_close(xprt);
9903cd1c 1109 return;
a246b010 1110 }
46121cf7 1111 dprintk("RPC: reading TCP record fragment of length %d\n",
51971139 1112 transport->tcp_reclen);
a246b010
CL		 1113}
1114
51971139 1115static void xs_tcp_check_fraghdr(struct sock_xprt *transport)
a246b010 1116{
51971139 1117 if (transport->tcp_offset == transport->tcp_reclen) {
e136d092 1118 transport->tcp_flags |= TCP_RCV_COPY_FRAGHDR;
51971139 1119 transport->tcp_offset = 0;
e136d092
CL		 1120 if (transport->tcp_flags & TCP_RCV_LAST_FRAG) {
1121 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1122 transport->tcp_flags |= TCP_RCV_COPY_XID;
51971139 1123 transport->tcp_copied = 0;
a246b010
CL		 1124 }
1125 }
1126}
1127
dd456471 1128static inline void xs_tcp_read_xid(struct sock_xprt *transport, struct xdr_skb_reader *desc)
a246b010
CL		 1129{
1130 size_t len, used;
1131 char *p;
1132
51971139 1133 len = sizeof(transport->tcp_xid) - transport->tcp_offset;
46121cf7 1134 dprintk("RPC: reading XID (%Zu bytes)\n", len);
51971139 1135 p = ((char *) &transport->tcp_xid) + transport->tcp_offset;
9d292316 1136 used = xdr_skb_read_bits(desc, p, len);
51971139 1137 transport->tcp_offset += used;
a246b010
CL		 1138 if (used != len)
1139 return;
e136d092 1140 transport->tcp_flags &= ~TCP_RCV_COPY_XID;
f4a2e418 1141 transport->tcp_flags |= TCP_RCV_READ_CALLDIR;
51971139 1142 transport->tcp_copied = 4;
18dca02a
RL		 1143 dprintk("RPC: reading %s XID %08x\n",
1144 (transport->tcp_flags & TCP_RPC_REPLY) ? "reply for"
1145 : "request with",
51971139
CL		 1146 ntohl(transport->tcp_xid));
1147 xs_tcp_check_fraghdr(transport);
a246b010
CL		 1148}
1149
18dca02a
RL		 1150static inline void xs_tcp_read_calldir(struct sock_xprt *transport,
1151 struct xdr_skb_reader *desc)
a246b010 1152{
18dca02a
RL		 1153 size_t len, used;
1154 u32 offset;
b76ce561 1155 char *p;
18dca02a
RL		 1156
1157 /*
1158 * We want transport->tcp_offset to be 8 at the end of this routine
1159 * (4 bytes for the xid and 4 bytes for the call/reply flag).
1160 * When this function is called for the first time,
1161 * transport->tcp_offset is 4 (after having already read the xid).
1162 */
1163 offset = transport->tcp_offset - sizeof(transport->tcp_xid);
b76ce561 1164 len = sizeof(transport->tcp_calldir) - offset;
18dca02a 1165 dprintk("RPC: reading CALL/REPLY flag (%Zu bytes)\n", len);
b76ce561
TM		 1166 p = ((char *) &transport->tcp_calldir) + offset;
1167 used = xdr_skb_read_bits(desc, p, len);
18dca02a
RL		 1168 transport->tcp_offset += used;
1169 if (used != len)
1170 return;
f4a2e418 1171 transport->tcp_flags &= ~TCP_RCV_READ_CALLDIR;
f4a2e418
RL		 1172 /*
1173 * We don't yet have the XDR buffer, so we will write the calldir
1174 * out after we get the buffer from the 'struct rpc_rqst'
1175 */
b76ce561
TM		 1176 switch (ntohl(transport->tcp_calldir)) {
1177 case RPC_REPLY:
1178 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1179 transport->tcp_flags |= TCP_RCV_COPY_DATA;
18dca02a 1180 transport->tcp_flags |= TCP_RPC_REPLY;
b76ce561
TM		 1181 break;
1182 case RPC_CALL:
1183 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1184 transport->tcp_flags |= TCP_RCV_COPY_DATA;
18dca02a 1185 transport->tcp_flags &= ~TCP_RPC_REPLY;
b76ce561
TM		 1186 break;
1187 default:
1188 dprintk("RPC: invalid request message type\n");
a519fc7a 1189 xs_tcp_force_close(&transport->xprt);
b76ce561 1190 }
18dca02a
RL		 1191 xs_tcp_check_fraghdr(transport);
1192}
1193
44b98efd
RL		 1194static inline void xs_tcp_read_common(struct rpc_xprt *xprt,
1195 struct xdr_skb_reader *desc,
1196 struct rpc_rqst *req)
a246b010 1197{
44b98efd
RL		 1198 struct sock_xprt *transport =
1199 container_of(xprt, struct sock_xprt, xprt);
a246b010
CL		 1200 struct xdr_buf *rcvbuf;
1201 size_t len;
1202 ssize_t r;
1203
a246b010 1204 rcvbuf = &req->rq_private_buf;
f4a2e418
RL		 1205
1206 if (transport->tcp_flags & TCP_RCV_COPY_CALLDIR) {
1207 /*
1208 * Save the RPC direction in the XDR buffer
1209 */
f4a2e418 1210 memcpy(rcvbuf->head[0].iov_base + transport->tcp_copied,
b76ce561
TM		 1211 &transport->tcp_calldir,
1212 sizeof(transport->tcp_calldir));
1213 transport->tcp_copied += sizeof(transport->tcp_calldir);
f4a2e418 1214 transport->tcp_flags &= ~TCP_RCV_COPY_CALLDIR;
a246b010
CL		 1215 }
1216
a246b010 1217 len = desc->count;
51971139 1218 if (len > transport->tcp_reclen - transport->tcp_offset) {
dd456471 1219 struct xdr_skb_reader my_desc;
a246b010 1220
51971139 1221 len = transport->tcp_reclen - transport->tcp_offset;
a246b010
CL		 1222 memcpy(&my_desc, desc, sizeof(my_desc));
1223 my_desc.count = len;
51971139 1224 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
9d292316 1225 &my_desc, xdr_skb_read_bits);
a246b010
CL		 1226 desc->count -= r;
1227 desc->offset += r;
1228 } else
51971139 1229 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
9d292316 1230 desc, xdr_skb_read_bits);
a246b010
CL		 1231
1232 if (r > 0) {
51971139
CL		 1233 transport->tcp_copied += r;
1234 transport->tcp_offset += r;
a246b010
CL		 1235 }
1236 if (r != len) {
1237 /* Error when copying to the receive buffer,
1238 * usually because we weren't able to allocate
1239 * additional buffer pages. All we can do now
e136d092 1240 * is turn off TCP_RCV_COPY_DATA, so the request
a246b010
CL		 1241 * will not receive any additional updates,
1242 * and time out.
1243 * Any remaining data from this record will
1244 * be discarded.
1245 */
e136d092 1246 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
46121cf7 1247 dprintk("RPC: XID %08x truncated request\n",
51971139 1248 ntohl(transport->tcp_xid));
46121cf7
CL		 1249 dprintk("RPC: xprt = %p, tcp_copied = %lu, "
1250 "tcp_offset = %u, tcp_reclen = %u\n",
1251 xprt, transport->tcp_copied,
1252 transport->tcp_offset, transport->tcp_reclen);
44b98efd 1253 return;
a246b010
CL		 1254 }
1255
46121cf7 1256 dprintk("RPC: XID %08x read %Zd bytes\n",
51971139 1257 ntohl(transport->tcp_xid), r);
46121cf7
CL		 1258 dprintk("RPC: xprt = %p, tcp_copied = %lu, tcp_offset = %u, "
1259 "tcp_reclen = %u\n", xprt, transport->tcp_copied,
1260 transport->tcp_offset, transport->tcp_reclen);
51971139
CL		 1261
1262 if (transport->tcp_copied == req->rq_private_buf.buflen)
e136d092 1263 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
51971139 1264 else if (transport->tcp_offset == transport->tcp_reclen) {
e136d092
CL		 1265 if (transport->tcp_flags & TCP_RCV_LAST_FRAG)
1266 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
a246b010 1267 }
44b98efd
RL		 1268}
1269
1270/*
1271 * Finds the request corresponding to the RPC xid and invokes the common
1272 * tcp read code to read the data.
1273 */
1274static inline int xs_tcp_read_reply(struct rpc_xprt *xprt,
1275 struct xdr_skb_reader *desc)
1276{
1277 struct sock_xprt *transport =
1278 container_of(xprt, struct sock_xprt, xprt);
1279 struct rpc_rqst *req;
1280
1281 dprintk("RPC: read reply XID %08x\n", ntohl(transport->tcp_xid));
1282
1283 /* Find and lock the request corresponding to this xid */
1284 spin_lock(&xprt->transport_lock);
1285 req = xprt_lookup_rqst(xprt, transport->tcp_xid);
1286 if (!req) {
1287 dprintk("RPC: XID %08x request not found!\n",
1288 ntohl(transport->tcp_xid));
1289 spin_unlock(&xprt->transport_lock);
1290 return -1;
1291 }
1292
1293 xs_tcp_read_common(xprt, desc, req);
1294
e136d092 1295 if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
51971139 1296 xprt_complete_rqst(req->rq_task, transport->tcp_copied);
44b98efd 1297
4a0f8c04 1298 spin_unlock(&xprt->transport_lock);
44b98efd
RL		 1299 return 0;
1300}
1301
9e00abc3 1302#if defined(CONFIG_SUNRPC_BACKCHANNEL)
44b98efd
RL		 1303/*
1304 * Obtains an rpc_rqst previously allocated and invokes the common
1305 * tcp read code to read the data. The result is placed in the callback
1306 * queue.
1307 * If we're unable to obtain the rpc_rqst we schedule the closing of the
1308 * connection and return -1.
1309 */
1310static inline int xs_tcp_read_callback(struct rpc_xprt *xprt,
1311 struct xdr_skb_reader *desc)
1312{
1313 struct sock_xprt *transport =
1314 container_of(xprt, struct sock_xprt, xprt);
1315 struct rpc_rqst *req;
1316
1317 req = xprt_alloc_bc_request(xprt);
1318 if (req == NULL) {
1319 printk(KERN_WARNING "Callback slot table overflowed\n");
1320 xprt_force_disconnect(xprt);
1321 return -1;
1322 }
1323
1324 req->rq_xid = transport->tcp_xid;
1325 dprintk("RPC: read callback XID %08x\n", ntohl(req->rq_xid));
1326 xs_tcp_read_common(xprt, desc, req);
1327
1328 if (!(transport->tcp_flags & TCP_RCV_COPY_DATA)) {
1329 struct svc_serv *bc_serv = xprt->bc_serv;
1330
1331 /*
1332 * Add callback request to callback list. The callback
1333 * service sleeps on the sv_cb_waitq waiting for new
1334 * requests. Wake it up after adding enqueing the
1335 * request.
1336 */
1337 dprintk("RPC: add callback request to list\n");
1338 spin_lock(&bc_serv->sv_cb_lock);
1339 list_add(&req->rq_bc_list, &bc_serv->sv_cb_list);
1340 spin_unlock(&bc_serv->sv_cb_lock);
1341 wake_up(&bc_serv->sv_cb_waitq);
1342 }
1343
1344 req->rq_private_buf.len = transport->tcp_copied;
1345
1346 return 0;
1347}
1348
1349static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1350 struct xdr_skb_reader *desc)
1351{
1352 struct sock_xprt *transport =
1353 container_of(xprt, struct sock_xprt, xprt);
1354
1355 return (transport->tcp_flags & TCP_RPC_REPLY) ?
1356 xs_tcp_read_reply(xprt, desc) :
1357 xs_tcp_read_callback(xprt, desc);
1358}
1359#else
1360static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1361 struct xdr_skb_reader *desc)
1362{
1363 return xs_tcp_read_reply(xprt, desc);
1364}
9e00abc3 1365#endif /* CONFIG_SUNRPC_BACKCHANNEL */
44b98efd
RL		 1366
1367/*
1368 * Read data off the transport. This can be either an RPC_CALL or an
1369 * RPC_REPLY. Relay the processing to helper functions.
1370 */
1371static void xs_tcp_read_data(struct rpc_xprt *xprt,
1372 struct xdr_skb_reader *desc)
1373{
1374 struct sock_xprt *transport =
1375 container_of(xprt, struct sock_xprt, xprt);
1376
1377 if (_xs_tcp_read_data(xprt, desc) == 0)
1378 xs_tcp_check_fraghdr(transport);
1379 else {
1380 /*
1381 * The transport_lock protects the request handling.
1382 * There's no need to hold it to update the tcp_flags.
1383 */
1384 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1385 }
a246b010
CL		 1386}
1387
dd456471 1388static inline void xs_tcp_read_discard(struct sock_xprt *transport, struct xdr_skb_reader *desc)
a246b010
CL		 1389{
1390 size_t len;
1391
51971139 1392 len = transport->tcp_reclen - transport->tcp_offset;
a246b010
CL		 1393 if (len > desc->count)
1394 len = desc->count;
1395 desc->count -= len;
1396 desc->offset += len;
51971139 1397 transport->tcp_offset += len;
46121cf7 1398 dprintk("RPC: discarded %Zu bytes\n", len);
51971139 1399 xs_tcp_check_fraghdr(transport);
a246b010
CL		 1400}
1401
9903cd1c 1402static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
a246b010
CL		 1403{
1404 struct rpc_xprt *xprt = rd_desc->arg.data;
51971139 1405 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
dd456471 1406 struct xdr_skb_reader desc = {
a246b010
CL		 1407 .skb = skb,
1408 .offset = offset,
1409 .count = len,
9903cd1c 1410 };
a246b010 1411
46121cf7 1412 dprintk("RPC: xs_tcp_data_recv started\n");
a246b010
CL		 1413 do {
1414 /* Read in a new fragment marker if necessary */
1415 /* Can we ever really expect to get completely empty fragments? */
e136d092 1416 if (transport->tcp_flags & TCP_RCV_COPY_FRAGHDR) {
9903cd1c 1417 xs_tcp_read_fraghdr(xprt, &desc);
a246b010
CL		 1418 continue;
1419 }
1420 /* Read in the xid if necessary */
e136d092 1421 if (transport->tcp_flags & TCP_RCV_COPY_XID) {
51971139 1422 xs_tcp_read_xid(transport, &desc);
a246b010
CL		 1423 continue;
1424 }
18dca02a 1425 /* Read in the call/reply flag */
f4a2e418 1426 if (transport->tcp_flags & TCP_RCV_READ_CALLDIR) {
18dca02a
RL		 1427 xs_tcp_read_calldir(transport, &desc);
1428 continue;
1429 }
a246b010 1430 /* Read in the request data */
e136d092 1431 if (transport->tcp_flags & TCP_RCV_COPY_DATA) {
44b98efd 1432 xs_tcp_read_data(xprt, &desc);
a246b010
CL		 1433 continue;
1434 }
1435 /* Skip over any trailing bytes on short reads */
51971139 1436 xs_tcp_read_discard(transport, &desc);
a246b010 1437 } while (desc.count);
46121cf7 1438 dprintk("RPC: xs_tcp_data_recv done\n");
a246b010
CL		 1439 return len - desc.count;
1440}
1441
9903cd1c
CL		 1442/**
1443 * xs_tcp_data_ready - "data ready" callback for TCP sockets
1444 * @sk: socket with data to read
1445 * @bytes: how much data to read
1446 *
1447 */
1448static void xs_tcp_data_ready(struct sock *sk, int bytes)
a246b010
CL		 1449{
1450 struct rpc_xprt *xprt;
1451 read_descriptor_t rd_desc;
ff2d7db8 1452 int read;
a246b010 1453
46121cf7
CL		 1454 dprintk("RPC: xs_tcp_data_ready...\n");
1455
f064af1e 1456 read_lock_bh(&sk->sk_callback_lock);
9903cd1c 1457 if (!(xprt = xprt_from_sock(sk)))
a246b010 1458 goto out;
61d0a8e6
NB		 1459 /* Any data means we had a useful conversation, so
1460 * the we don't need to delay the next reconnect
1461 */
1462 if (xprt->reestablish_timeout)
1463 xprt->reestablish_timeout = 0;
1464
9903cd1c 1465 /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
a246b010 1466 rd_desc.arg.data = xprt;
ff2d7db8
TM		 1467 do {
1468 rd_desc.count = 65536;
1469 read = tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
1470 } while (read > 0);
a246b010 1471out:
f064af1e 1472 read_unlock_bh(&sk->sk_callback_lock);
a246b010
CL		 1473}
1474
7d1e8255
TM		 1475/*
1476 * Do the equivalent of linger/linger2 handling for dealing with
1477 * broken servers that don't close the socket in a timely
1478 * fashion
1479 */
1480static void xs_tcp_schedule_linger_timeout(struct rpc_xprt *xprt,
1481 unsigned long timeout)
1482{
1483 struct sock_xprt *transport;
1484
1485 if (xprt_test_and_set_connecting(xprt))
1486 return;
1487 set_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1488 transport = container_of(xprt, struct sock_xprt, xprt);
1489 queue_delayed_work(rpciod_workqueue, &transport->connect_worker,
1490 timeout);
1491}
1492
1493static void xs_tcp_cancel_linger_timeout(struct rpc_xprt *xprt)
1494{
1495 struct sock_xprt *transport;
1496
1497 transport = container_of(xprt, struct sock_xprt, xprt);
1498
1499 if (!test_bit(XPRT_CONNECTION_ABORT, &xprt->state) ||
1500 !cancel_delayed_work(&transport->connect_worker))
1501 return;
1502 clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1503 xprt_clear_connecting(xprt);
1504}
1505
4bc1e68e 1506static void xs_sock_reset_connection_flags(struct rpc_xprt *xprt)
7d1e8255
TM		 1507{
1508 smp_mb__before_clear_bit();
a519fc7a
TM		 1509 clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1510 clear_bit(XPRT_CONNECTION_CLOSE, &xprt->state);
7d1e8255
TM		 1511 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1512 clear_bit(XPRT_CLOSING, &xprt->state);
1513 smp_mb__after_clear_bit();
4bc1e68e
TM		 1514}
1515
1516static void xs_sock_mark_closed(struct rpc_xprt *xprt)
1517{
1518 xs_sock_reset_connection_flags(xprt);
7d1e8255
TM		 1519 /* Mark transport as closed and wake up all pending tasks */
1520 xprt_disconnect_done(xprt);
1521}
1522
9903cd1c
CL		 1523/**
1524 * xs_tcp_state_change - callback to handle TCP socket state changes
1525 * @sk: socket whose state has changed
1526 *
1527 */
1528static void xs_tcp_state_change(struct sock *sk)
a246b010 1529{
9903cd1c 1530 struct rpc_xprt *xprt;
a246b010 1531
f064af1e 1532 read_lock_bh(&sk->sk_callback_lock);
a246b010
CL		 1533 if (!(xprt = xprt_from_sock(sk)))
1534 goto out;
46121cf7 1535 dprintk("RPC: xs_tcp_state_change client %p...\n", xprt);
669502ff 1536 dprintk("RPC: state %x conn %d dead %d zapped %d sk_shutdown %d\n",
46121cf7
CL		 1537 sk->sk_state, xprt_connected(xprt),
1538 sock_flag(sk, SOCK_DEAD),
669502ff
AC		 1539 sock_flag(sk, SOCK_ZAPPED),
1540 sk->sk_shutdown);
a246b010 1541
40b5ea0c 1542 trace_rpc_socket_state_change(xprt, sk->sk_socket);
a246b010
CL		 1543 switch (sk->sk_state) {
1544 case TCP_ESTABLISHED:
f064af1e 1545 spin_lock(&xprt->transport_lock);
a246b010 1546 if (!xprt_test_and_set_connected(xprt)) {
51971139
CL		 1547 struct sock_xprt *transport = container_of(xprt,
1548 struct sock_xprt, xprt);
1549
a246b010 1550 /* Reset TCP record info */
51971139
CL		 1551 transport->tcp_offset = 0;
1552 transport->tcp_reclen = 0;
1553 transport->tcp_copied = 0;
e136d092
CL		 1554 transport->tcp_flags =
1555 TCP_RCV_COPY_FRAGHDR | TCP_RCV_COPY_XID;
8b71798c 1556 xprt->connect_cookie++;
51971139 1557
2a491991 1558 xprt_wake_pending_tasks(xprt, -EAGAIN);
a246b010 1559 }
f064af1e 1560 spin_unlock(&xprt->transport_lock);
a246b010 1561 break;
3b948ae5
TM		 1562 case TCP_FIN_WAIT1:
1563 /* The client initiated a shutdown of the socket */
7c1d71cf 1564 xprt->connect_cookie++;
663b8858 1565 xprt->reestablish_timeout = 0;
3b948ae5
TM		 1566 set_bit(XPRT_CLOSING, &xprt->state);
1567 smp_mb__before_clear_bit();
1568 clear_bit(XPRT_CONNECTED, &xprt->state);
ef803670 1569 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
3b948ae5 1570 smp_mb__after_clear_bit();
25fe6142 1571 xs_tcp_schedule_linger_timeout(xprt, xs_tcp_fin_timeout);
a246b010 1572 break;
632e3bdc 1573 case TCP_CLOSE_WAIT:
3b948ae5 1574 /* The server initiated a shutdown of the socket */
7c1d71cf 1575 xprt->connect_cookie++;
d0bea455 1576 clear_bit(XPRT_CONNECTED, &xprt->state);
a519fc7a 1577 xs_tcp_force_close(xprt);
663b8858
TM		 1578 case TCP_CLOSING:
1579 /*
1580 * If the server closed down the connection, make sure that
1581 * we back off before reconnecting
1582 */
1583 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1584 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
3b948ae5
TM		 1585 break;
1586 case TCP_LAST_ACK:
670f9457 1587 set_bit(XPRT_CLOSING, &xprt->state);
25fe6142 1588 xs_tcp_schedule_linger_timeout(xprt, xs_tcp_fin_timeout);
3b948ae5
TM		 1589 smp_mb__before_clear_bit();
1590 clear_bit(XPRT_CONNECTED, &xprt->state);
1591 smp_mb__after_clear_bit();
1592 break;
1593 case TCP_CLOSE:
7d1e8255
TM		 1594 xs_tcp_cancel_linger_timeout(xprt);
1595 xs_sock_mark_closed(xprt);
a246b010
CL		 1596 }
1597 out:
f064af1e 1598 read_unlock_bh(&sk->sk_callback_lock);
a246b010
CL		 1599}
1600
1f0fa154
IJ		 1601static void xs_write_space(struct sock *sk)
1602{
1603 struct socket *sock;
1604 struct rpc_xprt *xprt;
1605
1606 if (unlikely(!(sock = sk->sk_socket)))
1607 return;
1608 clear_bit(SOCK_NOSPACE, &sock->flags);
1609
1610 if (unlikely(!(xprt = xprt_from_sock(sk))))
1611 return;
1612 if (test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags) == 0)
1613 return;
1614
1615 xprt_write_space(xprt);
1616}
1617
9903cd1c 1618/**
c7b2cae8
CL		 1619 * xs_udp_write_space - callback invoked when socket buffer space
1620 * becomes available
9903cd1c
CL		 1621 * @sk: socket whose state has changed
1622 *
a246b010
CL		 1623 * Called when more output buffer space is available for this socket.
1624 * We try not to wake our writers until they can make "significant"
c7b2cae8 1625 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
a246b010
CL		 1626 * with a bunch of small requests.
1627 */
c7b2cae8 1628static void xs_udp_write_space(struct sock *sk)
a246b010 1629{
f064af1e 1630 read_lock_bh(&sk->sk_callback_lock);
a246b010 1631
c7b2cae8 1632 /* from net/core/sock.c:sock_def_write_space */
1f0fa154
IJ		 1633 if (sock_writeable(sk))
1634 xs_write_space(sk);
a246b010 1635
f064af1e 1636 read_unlock_bh(&sk->sk_callback_lock);
c7b2cae8 1637}
a246b010 1638
c7b2cae8
CL		 1639/**
1640 * xs_tcp_write_space - callback invoked when socket buffer space
1641 * becomes available
1642 * @sk: socket whose state has changed
1643 *
1644 * Called when more output buffer space is available for this socket.
1645 * We try not to wake our writers until they can make "significant"
1646 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1647 * with a bunch of small requests.
1648 */
1649static void xs_tcp_write_space(struct sock *sk)
1650{
f064af1e 1651 read_lock_bh(&sk->sk_callback_lock);
c7b2cae8
CL		 1652
1653 /* from net/core/stream.c:sk_stream_write_space */
64dc6130 1654 if (sk_stream_is_writeable(sk))
1f0fa154 1655 xs_write_space(sk);
b6ddf64f 1656
f064af1e 1657 read_unlock_bh(&sk->sk_callback_lock);
a246b010
CL		 1658}
1659
470056c2 1660static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
a246b010 1661{
ee0ac0c2
CL		 1662 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1663 struct sock *sk = transport->inet;
a246b010 1664
7c6e066e 1665 if (transport->rcvsize) {
a246b010 1666 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
7c6e066e 1667 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
a246b010 1668 }
7c6e066e 1669 if (transport->sndsize) {
a246b010 1670 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
7c6e066e 1671 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
a246b010
CL		 1672 sk->sk_write_space(sk);
1673 }
1674}
1675
43118c29 1676/**
470056c2 1677 * xs_udp_set_buffer_size - set send and receive limits
43118c29 1678 * @xprt: generic transport
470056c2
CL		 1679 * @sndsize: requested size of send buffer, in bytes
1680 * @rcvsize: requested size of receive buffer, in bytes
43118c29 1681 *
470056c2 1682 * Set socket send and receive buffer size limits.
43118c29 1683 */
470056c2 1684static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
43118c29 1685{
7c6e066e
CL		 1686 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1687
1688 transport->sndsize = 0;
470056c2 1689 if (sndsize)
7c6e066e
CL		 1690 transport->sndsize = sndsize + 1024;
1691 transport->rcvsize = 0;
470056c2 1692 if (rcvsize)
7c6e066e 1693 transport->rcvsize = rcvsize + 1024;
470056c2
CL		 1694
1695 xs_udp_do_set_buffer_size(xprt);
43118c29
CL		 1696}
1697
46c0ee8b
CL		 1698/**
1699 * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1700 * @task: task that timed out
1701 *
1702 * Adjust the congestion window after a retransmit timeout has occurred.
1703 */
6a24dfb6 1704static void xs_udp_timer(struct rpc_xprt *xprt, struct rpc_task *task)
46c0ee8b 1705{
6a24dfb6 1706 xprt_adjust_cwnd(xprt, task, -ETIMEDOUT);
46c0ee8b
CL		 1707}
1708
b85d8806
CL		 1709static unsigned short xs_get_random_port(void)
1710{
1711 unsigned short range = xprt_max_resvport - xprt_min_resvport;
63862b5b 1712 unsigned short rand = (unsigned short) prandom_u32() % range;
b85d8806
CL		 1713 return rand + xprt_min_resvport;
1714}
1715
92200412
CL		 1716/**
1717 * xs_set_port - reset the port number in the remote endpoint address
1718 * @xprt: generic transport
1719 * @port: new port number
1720 *
1721 */
1722static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1723{
46121cf7 1724 dprintk("RPC: setting port for xprt %p to %u\n", xprt, port);
c4efcb1d 1725
9dc3b095
CL		 1726 rpc_set_port(xs_addr(xprt), port);
1727 xs_update_peer_port(xprt);
92200412
CL		 1728}
1729
5d4ec932 1730static unsigned short xs_get_srcport(struct sock_xprt *transport)
67a391d7 1731{
fbfffbd5 1732 unsigned short port = transport->srcport;
67a391d7
TM		 1733
1734 if (port == 0 && transport->xprt.resvport)
1735 port = xs_get_random_port();
1736 return port;
1737}
1738
baaf4e48 1739static unsigned short xs_next_srcport(struct sock_xprt *transport, unsigned short port)
67a391d7 1740{
fbfffbd5
CL		 1741 if (transport->srcport != 0)
1742 transport->srcport = 0;
67a391d7
TM		 1743 if (!transport->xprt.resvport)
1744 return 0;
1745 if (port <= xprt_min_resvport || port > xprt_max_resvport)
1746 return xprt_max_resvport;
1747 return --port;
1748}
beb59b68 1749static int xs_bind(struct sock_xprt *transport, struct socket *sock)
a246b010 1750{
beb59b68 1751 struct sockaddr_storage myaddr;
67a391d7 1752 int err, nloop = 0;
5d4ec932 1753 unsigned short port = xs_get_srcport(transport);
67a391d7 1754 unsigned short last;
a246b010 1755
beb59b68 1756 memcpy(&myaddr, &transport->srcaddr, transport->xprt.addrlen);
a246b010 1757 do {
beb59b68
PE		 1758 rpc_set_port((struct sockaddr *)&myaddr, port);
1759 err = kernel_bind(sock, (struct sockaddr *)&myaddr,
1760 transport->xprt.addrlen);
67a391d7 1761 if (port == 0)
d3bc9a1d 1762 break;
a246b010 1763 if (err == 0) {
fbfffbd5 1764 transport->srcport = port;
d3bc9a1d 1765 break;
a246b010 1766 }
67a391d7 1767 last = port;
baaf4e48 1768 port = xs_next_srcport(transport, port);
67a391d7
TM		 1769 if (port > last)
1770 nloop++;
1771 } while (err == -EADDRINUSE && nloop != 2);
90058d37 1772
4232e863 1773 if (myaddr.ss_family == AF_INET)
beb59b68
PE		 1774 dprintk("RPC: %s %pI4:%u: %s (%d)\n", __func__,
1775 &((struct sockaddr_in *)&myaddr)->sin_addr,
1776 port, err ? "failed" : "ok", err);
1777 else
1778 dprintk("RPC: %s %pI6:%u: %s (%d)\n", __func__,
1779 &((struct sockaddr_in6 *)&myaddr)->sin6_addr,
1780 port, err ? "failed" : "ok", err);
a246b010
CL		 1781 return err;
1782}
1783
176e21ee
CL		 1784/*
1785 * We don't support autobind on AF_LOCAL sockets
1786 */
1787static void xs_local_rpcbind(struct rpc_task *task)
1788{
3dc0da27
TM		 1789 rcu_read_lock();
1790 xprt_set_bound(rcu_dereference(task->tk_client->cl_xprt));
1791 rcu_read_unlock();
176e21ee
CL		 1792}
1793
1794static void xs_local_set_port(struct rpc_xprt *xprt, unsigned short port)
1795{
1796}
beb59b68 1797
ed07536e
PZ		 1798#ifdef CONFIG_DEBUG_LOCK_ALLOC
1799static struct lock_class_key xs_key[2];
1800static struct lock_class_key xs_slock_key[2];
1801
176e21ee
CL		 1802static inline void xs_reclassify_socketu(struct socket *sock)
1803{
1804 struct sock *sk = sock->sk;
1805
176e21ee
CL		 1806 sock_lock_init_class_and_name(sk, "slock-AF_LOCAL-RPC",
1807 &xs_slock_key[1], "sk_lock-AF_LOCAL-RPC", &xs_key[1]);
1808}
1809
8945ee5e 1810static inline void xs_reclassify_socket4(struct socket *sock)
ed07536e
PZ		 1811{
1812 struct sock *sk = sock->sk;
8945ee5e 1813
8945ee5e
CL		 1814 sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1815 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1816}
ed07536e 1817
8945ee5e
CL		 1818static inline void xs_reclassify_socket6(struct socket *sock)
1819{
1820 struct sock *sk = sock->sk;
ed07536e 1821
8945ee5e
CL		 1822 sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1823 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
ed07536e 1824}
6bc9638a
PE		 1825
1826static inline void xs_reclassify_socket(int family, struct socket *sock)
1827{
1b7a1819
WAA		 1828 WARN_ON_ONCE(sock_owned_by_user(sock->sk));
1829 if (sock_owned_by_user(sock->sk))
1830 return;
1831
4232e863 1832 switch (family) {
176e21ee
CL		 1833 case AF_LOCAL:
1834 xs_reclassify_socketu(sock);
1835 break;
4232e863 1836 case AF_INET:
6bc9638a 1837 xs_reclassify_socket4(sock);
4232e863
CL		 1838 break;
1839 case AF_INET6:
6bc9638a 1840 xs_reclassify_socket6(sock);
4232e863
CL		 1841 break;
1842 }
6bc9638a 1843}
ed07536e 1844#else
176e21ee
CL		 1845static inline void xs_reclassify_socketu(struct socket *sock)
1846{
1847}
1848
8945ee5e
CL		 1849static inline void xs_reclassify_socket4(struct socket *sock)
1850{
1851}
1852
1853static inline void xs_reclassify_socket6(struct socket *sock)
ed07536e
PZ		 1854{
1855}
6bc9638a
PE		 1856
1857static inline void xs_reclassify_socket(int family, struct socket *sock)
1858{
1859}
ed07536e
PZ		 1860#endif
1861
93dc41bd
N		 1862static void xs_dummy_setup_socket(struct work_struct *work)
1863{
1864}
1865
6bc9638a
PE		 1866static struct socket *xs_create_sock(struct rpc_xprt *xprt,
1867 struct sock_xprt *transport, int family, int type, int protocol)
22f79326
PE		 1868{
1869 struct socket *sock;
1870 int err;
1871
6bc9638a 1872 err = __sock_create(xprt->xprt_net, family, type, protocol, &sock, 1);
22f79326
PE		 1873 if (err < 0) {
1874 dprintk("RPC: can't create %d transport socket (%d).\n",
1875 protocol, -err);
1876 goto out;
1877 }
6bc9638a 1878 xs_reclassify_socket(family, sock);
22f79326 1879
4cea288a
BH		 1880 err = xs_bind(transport, sock);
1881 if (err) {
22f79326
PE		 1882 sock_release(sock);
1883 goto out;
1884 }
1885
1886 return sock;
1887out:
1888 return ERR_PTR(err);
1889}
1890
176e21ee
CL		 1891static int xs_local_finish_connecting(struct rpc_xprt *xprt,
1892 struct socket *sock)
1893{
1894 struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1895 xprt);
1896
1897 if (!transport->inet) {
1898 struct sock *sk = sock->sk;
1899
1900 write_lock_bh(&sk->sk_callback_lock);
1901
1902 xs_save_old_callbacks(transport, sk);
1903
1904 sk->sk_user_data = xprt;
1905 sk->sk_data_ready = xs_local_data_ready;
1906 sk->sk_write_space = xs_udp_write_space;
2118071d 1907 sk->sk_error_report = xs_error_report;
176e21ee
CL		 1908 sk->sk_allocation = GFP_ATOMIC;
1909
1910 xprt_clear_connected(xprt);
1911
1912 /* Reset to new socket */
1913 transport->sock = sock;
1914 transport->inet = sk;
1915
1916 write_unlock_bh(&sk->sk_callback_lock);
1917 }
1918
1919 /* Tell the socket layer to start connecting... */
1920 xprt->stat.connect_count++;
1921 xprt->stat.connect_start = jiffies;
1922 return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, 0);
1923}
1924
1925/**
1926 * xs_local_setup_socket - create AF_LOCAL socket, connect to a local endpoint
1927 * @xprt: RPC transport to connect
1928 * @transport: socket transport to connect
1929 * @create_sock: function to create a socket of the correct type
176e21ee 1930 */
dc107402 1931static int xs_local_setup_socket(struct sock_xprt *transport)
176e21ee 1932{
176e21ee
CL		 1933 struct rpc_xprt *xprt = &transport->xprt;
1934 struct socket *sock;
1935 int status = -EIO;
1936
5cf02d09
JL		 1937 current->flags |= PF_FSTRANS;
1938
176e21ee
CL		 1939 clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1940 status = __sock_create(xprt->xprt_net, AF_LOCAL,
1941 SOCK_STREAM, 0, &sock, 1);
1942 if (status < 0) {
1943 dprintk("RPC: can't create AF_LOCAL "
1944 "transport socket (%d).\n", -status);
1945 goto out;
1946 }
1947 xs_reclassify_socketu(sock);
1948
1949 dprintk("RPC: worker connecting xprt %p via AF_LOCAL to %s\n",
1950 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1951
1952 status = xs_local_finish_connecting(xprt, sock);
40b5ea0c 1953 trace_rpc_socket_connect(xprt, sock, status);
176e21ee
CL		 1954 switch (status) {
1955 case 0:
1956 dprintk("RPC: xprt %p connected to %s\n",
1957 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1958 xprt_set_connected(xprt);
1959 break;
1960 case -ENOENT:
1961 dprintk("RPC: xprt %p: socket %s does not exist\n",
1962 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1963 break;
4a20a988
TM		 1964 case -ECONNREFUSED:
1965 dprintk("RPC: xprt %p: connection refused for %s\n",
1966 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1967 break;
176e21ee
CL		 1968 default:
1969 printk(KERN_ERR "%s: unhandled error (%d) connecting to %s\n",
1970 __func__, -status,
1971 xprt->address_strings[RPC_DISPLAY_ADDR]);
1972 }
1973
1974out:
1975 xprt_clear_connecting(xprt);
1976 xprt_wake_pending_tasks(xprt, status);
5cf02d09 1977 current->flags &= ~PF_FSTRANS;
dc107402
BF		 1978 return status;
1979}
1980
b6669737 1981static void xs_local_connect(struct rpc_xprt *xprt, struct rpc_task *task)
dc107402 1982{
dc107402
BF		 1983 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1984 int ret;
1985
1986 if (RPC_IS_ASYNC(task)) {
1987 /*
1988 * We want the AF_LOCAL connect to be resolved in the
1989 * filesystem namespace of the process making the rpc
1990 * call. Thus we connect synchronously.
1991 *
1992 * If we want to support asynchronous AF_LOCAL calls,
1993 * we'll need to figure out how to pass a namespace to
1994 * connect.
1995 */
1996 rpc_exit(task, -ENOTCONN);
1997 return;
1998 }
1999 ret = xs_local_setup_socket(transport);
2000 if (ret && !RPC_IS_SOFTCONN(task))
2001 msleep_interruptible(15000);
176e21ee
CL		 2002}
2003
a564b8f0
MG		 2004#ifdef CONFIG_SUNRPC_SWAP
2005static void xs_set_memalloc(struct rpc_xprt *xprt)
2006{
2007 struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
2008 xprt);
2009
2010 if (xprt->swapper)
2011 sk_set_memalloc(transport->inet);
2012}
2013
2014/**
2015 * xs_swapper - Tag this transport as being used for swap.
2016 * @xprt: transport to tag
2017 * @enable: enable/disable
2018 *
2019 */
2020int xs_swapper(struct rpc_xprt *xprt, int enable)
2021{
2022 struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
2023 xprt);
2024 int err = 0;
2025
2026 if (enable) {
2027 xprt->swapper++;
2028 xs_set_memalloc(xprt);
2029 } else if (xprt->swapper) {
2030 xprt->swapper--;
2031 sk_clear_memalloc(transport->inet);
2032 }
2033
2034 return err;
2035}
2036EXPORT_SYMBOL_GPL(xs_swapper);
2037#else
2038static void xs_set_memalloc(struct rpc_xprt *xprt)
2039{
2040}
2041#endif
2042
16be2d20
CL		 2043static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2044{
2045 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2046
2047 if (!transport->inet) {
2048 struct sock *sk = sock->sk;
2049
2050 write_lock_bh(&sk->sk_callback_lock);
2051
2a9e1cfa
TM		 2052 xs_save_old_callbacks(transport, sk);
2053
16be2d20 2054 sk->sk_user_data = xprt;
16be2d20
CL		 2055 sk->sk_data_ready = xs_udp_data_ready;
2056 sk->sk_write_space = xs_udp_write_space;
2057 sk->sk_no_check = UDP_CSUM_NORCV;
2058 sk->sk_allocation = GFP_ATOMIC;
2059
2060 xprt_set_connected(xprt);
2061
2062 /* Reset to new socket */
2063 transport->sock = sock;
2064 transport->inet = sk;
2065
a564b8f0
MG		 2066 xs_set_memalloc(xprt);
2067
16be2d20
CL		 2068 write_unlock_bh(&sk->sk_callback_lock);
2069 }
2070 xs_udp_do_set_buffer_size(xprt);
2071}
2072
8c14ff2a 2073static void xs_udp_setup_socket(struct work_struct *work)
a246b010 2074{
34161db6
TM		 2075 struct sock_xprt *transport =
2076 container_of(work, struct sock_xprt, connect_worker.work);
c8475461 2077 struct rpc_xprt *xprt = &transport->xprt;
ee0ac0c2 2078 struct socket *sock = transport->sock;
b65c0310 2079 int status = -EIO;
9903cd1c 2080
5cf02d09
JL		 2081 current->flags |= PF_FSTRANS;
2082
b0d93ad5 2083 /* Start by resetting any existing state */
fe315e76 2084 xs_reset_transport(transport);
8c14ff2a
PE		 2085 sock = xs_create_sock(xprt, transport,
2086 xs_addr(xprt)->sa_family, SOCK_DGRAM, IPPROTO_UDP);
b65c0310 2087 if (IS_ERR(sock))
b0d93ad5 2088 goto out;
68e220bd 2089
c740eff8
CL		 2090 dprintk("RPC: worker connecting xprt %p via %s to "
2091 "%s (port %s)\n", xprt,
2092 xprt->address_strings[RPC_DISPLAY_PROTO],
2093 xprt->address_strings[RPC_DISPLAY_ADDR],
2094 xprt->address_strings[RPC_DISPLAY_PORT]);
68e220bd
CL		 2095
2096 xs_udp_finish_connecting(xprt, sock);
40b5ea0c 2097 trace_rpc_socket_connect(xprt, sock, 0);
b0d93ad5
CL		 2098 status = 0;
2099out:
b0d93ad5 2100 xprt_clear_connecting(xprt);
7d1e8255 2101 xprt_wake_pending_tasks(xprt, status);
5cf02d09 2102 current->flags &= ~PF_FSTRANS;
a246b010
CL		 2103}
2104
3167e12c
CL		 2105/*
2106 * We need to preserve the port number so the reply cache on the server can
2107 * find our cached RPC replies when we get around to reconnecting.
2108 */
58dddac9 2109static void xs_abort_connection(struct sock_xprt *transport)
3167e12c
CL		 2110{
2111 int result;
3167e12c
CL		 2112 struct sockaddr any;
2113
58dddac9 2114 dprintk("RPC: disconnecting xprt %p to reuse port\n", transport);
3167e12c
CL		 2115
2116 /*
2117 * Disconnect the transport socket by doing a connect operation
2118 * with AF_UNSPEC. This should return immediately...
2119 */
2120 memset(&any, 0, sizeof(any));
2121 any.sa_family = AF_UNSPEC;
ee0ac0c2 2122 result = kernel_connect(transport->sock, &any, sizeof(any), 0);
40b5ea0c
TM		 2123 trace_rpc_socket_reset_connection(&transport->xprt,
2124 transport->sock, result);
7d1e8255 2125 if (!result)
4bc1e68e
TM		 2126 xs_sock_reset_connection_flags(&transport->xprt);
2127 dprintk("RPC: AF_UNSPEC connect return code %d\n", result);
3167e12c
CL		 2128}
2129
58dddac9 2130static void xs_tcp_reuse_connection(struct sock_xprt *transport)
40d2549d
TM		 2131{
2132 unsigned int state = transport->inet->sk_state;
2133
669502ff
AC		 2134 if (state == TCP_CLOSE && transport->sock->state == SS_UNCONNECTED) {
2135 /* we don't need to abort the connection if the socket
2136 * hasn't undergone a shutdown
2137 */
2138 if (transport->inet->sk_shutdown == 0)
2139 return;
2140 dprintk("RPC: %s: TCP_CLOSEd and sk_shutdown set to %d\n",
2141 __func__, transport->inet->sk_shutdown);
2142 }
2143 if ((1 << state) & (TCPF_ESTABLISHED|TCPF_SYN_SENT)) {
2144 /* we don't need to abort the connection if the socket
2145 * hasn't undergone a shutdown
2146 */
2147 if (transport->inet->sk_shutdown == 0)
2148 return;
2149 dprintk("RPC: %s: ESTABLISHED/SYN_SENT "
2150 "sk_shutdown set to %d\n",
2151 __func__, transport->inet->sk_shutdown);
2152 }
58dddac9 2153 xs_abort_connection(transport);
40d2549d
TM		 2154}
2155
16be2d20 2156static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
a246b010 2157{
16be2d20 2158 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
fe19a96b 2159 int ret = -ENOTCONN;
edb267a6 2160
ee0ac0c2 2161 if (!transport->inet) {
b0d93ad5 2162 struct sock *sk = sock->sk;
7f260e85
TM		 2163 unsigned int keepidle = xprt->timeout->to_initval / HZ;
2164 unsigned int keepcnt = xprt->timeout->to_retries + 1;
2165 unsigned int opt_on = 1;
2166
2167 /* TCP Keepalive options */
2168 kernel_setsockopt(sock, SOL_SOCKET, SO_KEEPALIVE,
2169 (char *)&opt_on, sizeof(opt_on));
2170 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPIDLE,
2171 (char *)&keepidle, sizeof(keepidle));
2172 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPINTVL,
2173 (char *)&keepidle, sizeof(keepidle));
2174 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPCNT,
2175 (char *)&keepcnt, sizeof(keepcnt));
b0d93ad5
CL		 2176
2177 write_lock_bh(&sk->sk_callback_lock);
2178
2a9e1cfa
TM		 2179 xs_save_old_callbacks(transport, sk);
2180
b0d93ad5 2181 sk->sk_user_data = xprt;
b0d93ad5
CL		 2182 sk->sk_data_ready = xs_tcp_data_ready;
2183 sk->sk_state_change = xs_tcp_state_change;
2184 sk->sk_write_space = xs_tcp_write_space;
2118071d 2185 sk->sk_error_report = xs_error_report;
b079fa7b 2186 sk->sk_allocation = GFP_ATOMIC;
3167e12c
CL		 2187
2188 /* socket options */
2189 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
2190 sock_reset_flag(sk, SOCK_LINGER);
2191 tcp_sk(sk)->linger2 = 0;
2192 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
b0d93ad5
CL		 2193
2194 xprt_clear_connected(xprt);
2195
2196 /* Reset to new socket */
ee0ac0c2
CL		 2197 transport->sock = sock;
2198 transport->inet = sk;
b0d93ad5
CL		 2199
2200 write_unlock_bh(&sk->sk_callback_lock);
2201 }
2202
01d37c42 2203 if (!xprt_bound(xprt))
fe19a96b 2204 goto out;
01d37c42 2205
a564b8f0
MG		 2206 xs_set_memalloc(xprt);
2207
b0d93ad5 2208 /* Tell the socket layer to start connecting... */
262ca07d
CL		 2209 xprt->stat.connect_count++;
2210 xprt->stat.connect_start = jiffies;
fe19a96b
TM		 2211 ret = kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
2212 switch (ret) {
2213 case 0:
2214 case -EINPROGRESS:
2215 /* SYN_SENT! */
fe19a96b
TM		 2216 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2217 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2218 }
2219out:
2220 return ret;
16be2d20
CL		 2221}
2222
9903cd1c 2223/**
b61d59ff
TM		 2224 * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint
2225 * @xprt: RPC transport to connect
2226 * @transport: socket transport to connect
2227 * @create_sock: function to create a socket of the correct type
9903cd1c
CL		 2228 *
2229 * Invoked by a work queue tasklet.
a246b010 2230 */
cdd518d5 2231static void xs_tcp_setup_socket(struct work_struct *work)
a246b010 2232{
cdd518d5
PE		 2233 struct sock_xprt *transport =
2234 container_of(work, struct sock_xprt, connect_worker.work);
ee0ac0c2 2235 struct socket *sock = transport->sock;
a9f5f0f7 2236 struct rpc_xprt *xprt = &transport->xprt;
b61d59ff 2237 int status = -EIO;
a246b010 2238
5cf02d09
JL		 2239 current->flags |= PF_FSTRANS;
2240
ee0ac0c2 2241 if (!sock) {
7d1e8255 2242 clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
cdd518d5
PE		 2243 sock = xs_create_sock(xprt, transport,
2244 xs_addr(xprt)->sa_family, SOCK_STREAM, IPPROTO_TCP);
b61d59ff
TM		 2245 if (IS_ERR(sock)) {
2246 status = PTR_ERR(sock);
3167e12c
CL		 2247 goto out;
2248 }
7d1e8255
TM		 2249 } else {
2250 int abort_and_exit;
a246b010 2251
7d1e8255
TM		 2252 abort_and_exit = test_and_clear_bit(XPRT_CONNECTION_ABORT,
2253 &xprt->state);
3167e12c 2254 /* "close" the socket, preserving the local port */
58dddac9 2255 xs_tcp_reuse_connection(transport);
a246b010 2256
7d1e8255
TM		 2257 if (abort_and_exit)
2258 goto out_eagain;
2259 }
a246b010 2260
c740eff8
CL		 2261 dprintk("RPC: worker connecting xprt %p via %s to "
2262 "%s (port %s)\n", xprt,
2263 xprt->address_strings[RPC_DISPLAY_PROTO],
2264 xprt->address_strings[RPC_DISPLAY_ADDR],
2265 xprt->address_strings[RPC_DISPLAY_PORT]);
edb267a6 2266
16be2d20 2267 status = xs_tcp_finish_connecting(xprt, sock);
40b5ea0c 2268 trace_rpc_socket_connect(xprt, sock, status);
46121cf7
CL		 2269 dprintk("RPC: %p connect status %d connected %d sock state %d\n",
2270 xprt, -status, xprt_connected(xprt),
2271 sock->sk->sk_state);
2a491991 2272 switch (status) {
f75e6745
TM		 2273 default:
2274 printk("%s: connect returned unhandled error %d\n",
2275 __func__, status);
2276 case -EADDRNOTAVAIL:
2277 /* We're probably in TIME_WAIT. Get rid of existing socket,
2278 * and retry
2279 */
a519fc7a 2280 xs_tcp_force_close(xprt);
88b5ed73 2281 break;
2a491991
TM		 2282 case 0:
2283 case -EINPROGRESS:
2284 case -EALREADY:
7d1e8255 2285 xprt_clear_connecting(xprt);
5cf02d09 2286 current->flags &= ~PF_FSTRANS;
7d1e8255 2287 return;
9fcfe0c8
TM		 2288 case -EINVAL:
2289 /* Happens, for instance, if the user specified a link
2290 * local IPv6 address without a scope-id.
2291 */
3ed5e2a2
TM		 2292 case -ECONNREFUSED:
2293 case -ECONNRESET:
2294 case -ENETUNREACH:
2295 /* retry with existing socket, after a delay */
9fcfe0c8 2296 goto out;
a246b010 2297 }
7d1e8255 2298out_eagain:
2a491991 2299 status = -EAGAIN;
a246b010 2300out:
68e220bd 2301 xprt_clear_connecting(xprt);
7d1e8255 2302 xprt_wake_pending_tasks(xprt, status);
5cf02d09 2303 current->flags &= ~PF_FSTRANS;
68e220bd 2304}
b0d93ad5 2305
9903cd1c
CL		 2306/**
2307 * xs_connect - connect a socket to a remote endpoint
1b092092 2308 * @xprt: pointer to transport structure
9903cd1c
CL		 2309 * @task: address of RPC task that manages state of connect request
2310 *
2311 * TCP: If the remote end dropped the connection, delay reconnecting.
03bf4b70
CL		 2312 *
2313 * UDP socket connects are synchronous, but we use a work queue anyway
2314 * to guarantee that even unprivileged user processes can set up a
2315 * socket on a privileged port.
2316 *
2317 * If a UDP socket connect fails, the delay behavior here prevents
2318 * retry floods (hard mounts).
9903cd1c 2319 */
1b092092 2320static void xs_connect(struct rpc_xprt *xprt, struct rpc_task *task)
a246b010 2321{
ee0ac0c2 2322 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
a246b010 2323
09a21c41 2324 if (transport->sock != NULL && !RPC_IS_SOFTCONN(task)) {
46121cf7
CL		 2325 dprintk("RPC: xs_connect delayed xprt %p for %lu "
2326 "seconds\n",
03bf4b70 2327 xprt, xprt->reestablish_timeout / HZ);
c1384c9c
TM		 2328 queue_delayed_work(rpciod_workqueue,
2329 &transport->connect_worker,
2330 xprt->reestablish_timeout);
03bf4b70 2331 xprt->reestablish_timeout <<= 1;
61d0a8e6
NB		 2332 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2333 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
03bf4b70
CL		 2334 if (xprt->reestablish_timeout > XS_TCP_MAX_REEST_TO)
2335 xprt->reestablish_timeout = XS_TCP_MAX_REEST_TO;
b0d93ad5 2336 } else {
46121cf7 2337 dprintk("RPC: xs_connect scheduled xprt %p\n", xprt);
c1384c9c
TM		 2338 queue_delayed_work(rpciod_workqueue,
2339 &transport->connect_worker, 0);
a246b010
CL		 2340 }
2341}
2342
176e21ee
CL		 2343/**
2344 * xs_local_print_stats - display AF_LOCAL socket-specifc stats
2345 * @xprt: rpc_xprt struct containing statistics
2346 * @seq: output file
2347 *
2348 */
2349static void xs_local_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2350{
2351 long idle_time = 0;
2352
2353 if (xprt_connected(xprt))
2354 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2355
2356 seq_printf(seq, "\txprt:\tlocal %lu %lu %lu %ld %lu %lu %lu "
15a45206 2357 "%llu %llu %lu %llu %llu\n",
176e21ee
CL		 2358 xprt->stat.bind_count,
2359 xprt->stat.connect_count,
2360 xprt->stat.connect_time,
2361 idle_time,
2362 xprt->stat.sends,
2363 xprt->stat.recvs,
2364 xprt->stat.bad_xids,
2365 xprt->stat.req_u,
15a45206
AA		 2366 xprt->stat.bklog_u,
2367 xprt->stat.max_slots,
2368 xprt->stat.sending_u,
2369 xprt->stat.pending_u);
176e21ee
CL		 2370}
2371
262ca07d
CL		 2372/**
2373 * xs_udp_print_stats - display UDP socket-specifc stats
2374 * @xprt: rpc_xprt struct containing statistics
2375 * @seq: output file
2376 *
2377 */
2378static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2379{
c8475461
CL		 2380 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2381
15a45206
AA		 2382 seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %llu %llu "
2383 "%lu %llu %llu\n",
fbfffbd5 2384 transport->srcport,
262ca07d
CL		 2385 xprt->stat.bind_count,
2386 xprt->stat.sends,
2387 xprt->stat.recvs,
2388 xprt->stat.bad_xids,
2389 xprt->stat.req_u,
15a45206
AA		 2390 xprt->stat.bklog_u,
2391 xprt->stat.max_slots,
2392 xprt->stat.sending_u,
2393 xprt->stat.pending_u);
262ca07d
CL		 2394}
2395
2396/**
2397 * xs_tcp_print_stats - display TCP socket-specifc stats
2398 * @xprt: rpc_xprt struct containing statistics
2399 * @seq: output file
2400 *
2401 */
2402static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2403{
c8475461 2404 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
262ca07d
CL		 2405 long idle_time = 0;
2406
2407 if (xprt_connected(xprt))
2408 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2409
15a45206
AA		 2410 seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu "
2411 "%llu %llu %lu %llu %llu\n",
fbfffbd5 2412 transport->srcport,
262ca07d
CL		 2413 xprt->stat.bind_count,
2414 xprt->stat.connect_count,
2415 xprt->stat.connect_time,
2416 idle_time,
2417 xprt->stat.sends,
2418 xprt->stat.recvs,
2419 xprt->stat.bad_xids,
2420 xprt->stat.req_u,
15a45206
AA		 2421 xprt->stat.bklog_u,
2422 xprt->stat.max_slots,
2423 xprt->stat.sending_u,
2424 xprt->stat.pending_u);
262ca07d
CL		 2425}
2426
4cfc7e60
RI		 2427/*
2428 * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason
2429 * we allocate pages instead doing a kmalloc like rpc_malloc is because we want
2430 * to use the server side send routines.
2431 */
5a51f13a 2432static void *bc_malloc(struct rpc_task *task, size_t size)
4cfc7e60
RI		 2433{
2434 struct page *page;
2435 struct rpc_buffer *buf;
2436
b8a13d03
WAA		 2437 WARN_ON_ONCE(size > PAGE_SIZE - sizeof(struct rpc_buffer));
2438 if (size > PAGE_SIZE - sizeof(struct rpc_buffer))
2439 return NULL;
4cfc7e60 2440
b8a13d03 2441 page = alloc_page(GFP_KERNEL);
4cfc7e60
RI		 2442 if (!page)
2443 return NULL;
2444
2445 buf = page_address(page);
2446 buf->len = PAGE_SIZE;
2447
2448 return buf->data;
2449}
2450
2451/*
2452 * Free the space allocated in the bc_alloc routine
2453 */
5a51f13a 2454static void bc_free(void *buffer)
4cfc7e60
RI		 2455{
2456 struct rpc_buffer *buf;
2457
2458 if (!buffer)
2459 return;
2460
2461 buf = container_of(buffer, struct rpc_buffer, data);
2462 free_page((unsigned long)buf);
2463}
2464
2465/*
2466 * Use the svc_sock to send the callback. Must be called with svsk->sk_mutex
2467 * held. Borrows heavily from svc_tcp_sendto and xs_tcp_send_request.
2468 */
2469static int bc_sendto(struct rpc_rqst *req)
2470{
2471 int len;
2472 struct xdr_buf *xbufp = &req->rq_snd_buf;
2473 struct rpc_xprt *xprt = req->rq_xprt;
2474 struct sock_xprt *transport =
2475 container_of(xprt, struct sock_xprt, xprt);
2476 struct socket *sock = transport->sock;
2477 unsigned long headoff;
2478 unsigned long tailoff;
2479
61677eee 2480 xs_encode_stream_record_marker(xbufp);
4cfc7e60
RI		 2481
2482 tailoff = (unsigned long)xbufp->tail[0].iov_base & ~PAGE_MASK;
2483 headoff = (unsigned long)xbufp->head[0].iov_base & ~PAGE_MASK;
2484 len = svc_send_common(sock, xbufp,
2485 virt_to_page(xbufp->head[0].iov_base), headoff,
2486 xbufp->tail[0].iov_base, tailoff);
2487
2488 if (len != xbufp->len) {
2489 printk(KERN_NOTICE "Error sending entire callback!\n");
2490 len = -EAGAIN;
2491 }
2492
2493 return len;
2494}
2495
2496/*
2497 * The send routine. Borrows from svc_send
2498 */
2499static int bc_send_request(struct rpc_task *task)
2500{
2501 struct rpc_rqst *req = task->tk_rqstp;
2502 struct svc_xprt *xprt;
4cfc7e60
RI		 2503 u32 len;
2504
2505 dprintk("sending request with xid: %08x\n", ntohl(req->rq_xid));
2506 /*
2507 * Get the server socket associated with this callback xprt
2508 */
2509 xprt = req->rq_xprt->bc_xprt;
4cfc7e60
RI		 2510
2511 /*
2512 * Grab the mutex to serialize data as the connection is shared
2513 * with the fore channel
2514 */
2515 if (!mutex_trylock(&xprt->xpt_mutex)) {
2516 rpc_sleep_on(&xprt->xpt_bc_pending, task, NULL);
2517 if (!mutex_trylock(&xprt->xpt_mutex))
2518 return -EAGAIN;
2519 rpc_wake_up_queued_task(&xprt->xpt_bc_pending, task);
2520 }
2521 if (test_bit(XPT_DEAD, &xprt->xpt_flags))
2522 len = -ENOTCONN;
2523 else
2524 len = bc_sendto(req);
2525 mutex_unlock(&xprt->xpt_mutex);
2526
2527 if (len > 0)
2528 len = 0;
2529
2530 return len;
2531}
2532
2533/*
2534 * The close routine. Since this is client initiated, we do nothing
2535 */
2536
2537static void bc_close(struct rpc_xprt *xprt)
2538{
4cfc7e60
RI		 2539}
2540
2541/*
2542 * The xprt destroy routine. Again, because this connection is client
2543 * initiated, we do nothing
2544 */
2545
2546static void bc_destroy(struct rpc_xprt *xprt)
2547{
4cfc7e60
RI		 2548}
2549
176e21ee
CL		 2550static struct rpc_xprt_ops xs_local_ops = {
2551 .reserve_xprt = xprt_reserve_xprt,
2552 .release_xprt = xs_tcp_release_xprt,
f39c1bfb 2553 .alloc_slot = xprt_alloc_slot,
176e21ee
CL		 2554 .rpcbind = xs_local_rpcbind,
2555 .set_port = xs_local_set_port,
dc107402 2556 .connect = xs_local_connect,
176e21ee
CL		 2557 .buf_alloc = rpc_malloc,
2558 .buf_free = rpc_free,
2559 .send_request = xs_local_send_request,
2560 .set_retrans_timeout = xprt_set_retrans_timeout_def,
2561 .close = xs_close,
a1311d87 2562 .destroy = xs_destroy,
176e21ee
CL		 2563 .print_stats = xs_local_print_stats,
2564};
2565
262965f5 2566static struct rpc_xprt_ops xs_udp_ops = {
43118c29 2567 .set_buffer_size = xs_udp_set_buffer_size,
12a80469 2568 .reserve_xprt = xprt_reserve_xprt_cong,
49e9a890 2569 .release_xprt = xprt_release_xprt_cong,
f39c1bfb 2570 .alloc_slot = xprt_alloc_slot,
45160d62 2571 .rpcbind = rpcb_getport_async,
92200412 2572 .set_port = xs_set_port,
262965f5 2573 .connect = xs_connect,
02107148
CL		 2574 .buf_alloc = rpc_malloc,
2575 .buf_free = rpc_free,
262965f5 2576 .send_request = xs_udp_send_request,
fe3aca29 2577 .set_retrans_timeout = xprt_set_retrans_timeout_rtt,
46c0ee8b 2578 .timer = xs_udp_timer,
a58dd398 2579 .release_request = xprt_release_rqst_cong,
262965f5
CL		 2580 .close = xs_close,
2581 .destroy = xs_destroy,
262ca07d 2582 .print_stats = xs_udp_print_stats,
262965f5
CL		 2583};
2584
2585static struct rpc_xprt_ops xs_tcp_ops = {
12a80469 2586 .reserve_xprt = xprt_reserve_xprt,
e0ab53de 2587 .release_xprt = xs_tcp_release_xprt,
f39c1bfb 2588 .alloc_slot = xprt_lock_and_alloc_slot,
45160d62 2589 .rpcbind = rpcb_getport_async,
92200412 2590 .set_port = xs_set_port,
0b9e7943 2591 .connect = xs_connect,
02107148
CL		 2592 .buf_alloc = rpc_malloc,
2593 .buf_free = rpc_free,
262965f5 2594 .send_request = xs_tcp_send_request,
fe3aca29 2595 .set_retrans_timeout = xprt_set_retrans_timeout_def,
f75e6745 2596 .close = xs_tcp_close,
9903cd1c 2597 .destroy = xs_destroy,
262ca07d 2598 .print_stats = xs_tcp_print_stats,
a246b010
CL		 2599};
2600
4cfc7e60
RI		 2601/*
2602 * The rpc_xprt_ops for the server backchannel
2603 */
2604
2605static struct rpc_xprt_ops bc_tcp_ops = {
2606 .reserve_xprt = xprt_reserve_xprt,
2607 .release_xprt = xprt_release_xprt,
84e28a30 2608 .alloc_slot = xprt_alloc_slot,
4cfc7e60
RI		 2609 .buf_alloc = bc_malloc,
2610 .buf_free = bc_free,
2611 .send_request = bc_send_request,
2612 .set_retrans_timeout = xprt_set_retrans_timeout_def,
2613 .close = bc_close,
2614 .destroy = bc_destroy,
2615 .print_stats = xs_tcp_print_stats,
2616};
2617
92476850
CL		 2618static int xs_init_anyaddr(const int family, struct sockaddr *sap)
2619{
2620 static const struct sockaddr_in sin = {
2621 .sin_family = AF_INET,
2622 .sin_addr.s_addr = htonl(INADDR_ANY),
2623 };
2624 static const struct sockaddr_in6 sin6 = {
2625 .sin6_family = AF_INET6,
2626 .sin6_addr = IN6ADDR_ANY_INIT,
2627 };
2628
2629 switch (family) {
176e21ee
CL		 2630 case AF_LOCAL:
2631 break;
92476850
CL		 2632 case AF_INET:
2633 memcpy(sap, &sin, sizeof(sin));
2634 break;
2635 case AF_INET6:
2636 memcpy(sap, &sin6, sizeof(sin6));
2637 break;
2638 default:
2639 dprintk("RPC: %s: Bad address family\n", __func__);
2640 return -EAFNOSUPPORT;
2641 }
2642 return 0;
2643}
2644
3c341b0b 2645static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
d9ba131d
TM		 2646 unsigned int slot_table_size,
2647 unsigned int max_slot_table_size)
c8541ecd
CL		 2648{
2649 struct rpc_xprt *xprt;
ffc2e518 2650 struct sock_xprt *new;
c8541ecd 2651
96802a09 2652 if (args->addrlen > sizeof(xprt->addr)) {
46121cf7 2653 dprintk("RPC: xs_setup_xprt: address too large\n");
c8541ecd
CL		 2654 return ERR_PTR(-EBADF);
2655 }
2656
d9ba131d
TM		 2657 xprt = xprt_alloc(args->net, sizeof(*new), slot_table_size,
2658 max_slot_table_size);
bd1722d4 2659 if (xprt == NULL) {
46121cf7
CL		 2660 dprintk("RPC: xs_setup_xprt: couldn't allocate "
2661 "rpc_xprt\n");
c8541ecd
CL		 2662 return ERR_PTR(-ENOMEM);
2663 }
c8541ecd 2664
bd1722d4 2665 new = container_of(xprt, struct sock_xprt, xprt);
96802a09
FM		 2666 memcpy(&xprt->addr, args->dstaddr, args->addrlen);
2667 xprt->addrlen = args->addrlen;
d3bc9a1d 2668 if (args->srcaddr)
fbfffbd5 2669 memcpy(&new->srcaddr, args->srcaddr, args->addrlen);
92476850
CL		 2670 else {
2671 int err;
2672 err = xs_init_anyaddr(args->dstaddr->sa_family,
2673 (struct sockaddr *)&new->srcaddr);
2aa13531
SK		 2674 if (err != 0) {
2675 xprt_free(xprt);
92476850 2676 return ERR_PTR(err);
2aa13531 2677 }
92476850 2678 }
c8541ecd
CL		 2679
2680 return xprt;
2681}
2682
176e21ee
CL		 2683static const struct rpc_timeout xs_local_default_timeout = {
2684 .to_initval = 10 * HZ,
2685 .to_maxval = 10 * HZ,
2686 .to_retries = 2,
2687};
2688
2689/**
2690 * xs_setup_local - Set up transport to use an AF_LOCAL socket
2691 * @args: rpc transport creation arguments
2692 *
2693 * AF_LOCAL is a "tpi_cots_ord" transport, just like TCP
2694 */
2695static struct rpc_xprt *xs_setup_local(struct xprt_create *args)
2696{
2697 struct sockaddr_un *sun = (struct sockaddr_un *)args->dstaddr;
2698 struct sock_xprt *transport;
2699 struct rpc_xprt *xprt;
2700 struct rpc_xprt *ret;
2701
d9ba131d
TM		 2702 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2703 xprt_max_tcp_slot_table_entries);
176e21ee
CL		 2704 if (IS_ERR(xprt))
2705 return xprt;
2706 transport = container_of(xprt, struct sock_xprt, xprt);
2707
2708 xprt->prot = 0;
2709 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2710 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2711
2712 xprt->bind_timeout = XS_BIND_TO;
2713 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2714 xprt->idle_timeout = XS_IDLE_DISC_TO;
2715
2716 xprt->ops = &xs_local_ops;
2717 xprt->timeout = &xs_local_default_timeout;
2718
93dc41bd
N		 2719 INIT_DELAYED_WORK(&transport->connect_worker,
2720 xs_dummy_setup_socket);
2721
176e21ee
CL		 2722 switch (sun->sun_family) {
2723 case AF_LOCAL:
2724 if (sun->sun_path[0] != '/') {
2725 dprintk("RPC: bad AF_LOCAL address: %s\n",
2726 sun->sun_path);
2727 ret = ERR_PTR(-EINVAL);
2728 goto out_err;
2729 }
2730 xprt_set_bound(xprt);
176e21ee 2731 xs_format_peer_addresses(xprt, "local", RPCBIND_NETID_LOCAL);
7073ea87
BF		 2732 ret = ERR_PTR(xs_local_setup_socket(transport));
2733 if (ret)
2734 goto out_err;
176e21ee
CL		 2735 break;
2736 default:
2737 ret = ERR_PTR(-EAFNOSUPPORT);
2738 goto out_err;
2739 }
2740
2741 dprintk("RPC: set up xprt to %s via AF_LOCAL\n",
2742 xprt->address_strings[RPC_DISPLAY_ADDR]);
2743
2744 if (try_module_get(THIS_MODULE))
2745 return xprt;
2746 ret = ERR_PTR(-EINVAL);
2747out_err:
315f3812 2748 xs_xprt_free(xprt);
176e21ee
CL		 2749 return ret;
2750}
2751
2881ae74
TM		 2752static const struct rpc_timeout xs_udp_default_timeout = {
2753 .to_initval = 5 * HZ,
2754 .to_maxval = 30 * HZ,
2755 .to_increment = 5 * HZ,
2756 .to_retries = 5,
2757};
2758
9903cd1c
CL		 2759/**
2760 * xs_setup_udp - Set up transport to use a UDP socket
96802a09 2761 * @args: rpc transport creation arguments
9903cd1c
CL		 2762 *
2763 */
483066d6 2764static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
a246b010 2765{
8f9d5b1a 2766 struct sockaddr *addr = args->dstaddr;
c8541ecd 2767 struct rpc_xprt *xprt;
c8475461 2768 struct sock_xprt *transport;
0a68b0be 2769 struct rpc_xprt *ret;
a246b010 2770
d9ba131d
TM		 2771 xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries,
2772 xprt_udp_slot_table_entries);
c8541ecd
CL		 2773 if (IS_ERR(xprt))
2774 return xprt;
c8475461 2775 transport = container_of(xprt, struct sock_xprt, xprt);
a246b010 2776
ec739ef0 2777 xprt->prot = IPPROTO_UDP;
808012fb 2778 xprt->tsh_size = 0;
a246b010
CL		 2779 /* XXX: header size can vary due to auth type, IPv6, etc. */
2780 xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
2781
03bf4b70 2782 xprt->bind_timeout = XS_BIND_TO;
03bf4b70
CL		 2783 xprt->reestablish_timeout = XS_UDP_REEST_TO;
2784 xprt->idle_timeout = XS_IDLE_DISC_TO;
a246b010 2785
262965f5 2786 xprt->ops = &xs_udp_ops;
a246b010 2787
ba7392bb 2788 xprt->timeout = &xs_udp_default_timeout;
a246b010 2789
8f9d5b1a
CL		 2790 switch (addr->sa_family) {
2791 case AF_INET:
2792 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2793 xprt_set_bound(xprt);
2794
2795 INIT_DELAYED_WORK(&transport->connect_worker,
8c14ff2a 2796 xs_udp_setup_socket);
9dc3b095 2797 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
8f9d5b1a
CL		 2798 break;
2799 case AF_INET6:
2800 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2801 xprt_set_bound(xprt);
2802
2803 INIT_DELAYED_WORK(&transport->connect_worker,
8c14ff2a 2804 xs_udp_setup_socket);
9dc3b095 2805 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
8f9d5b1a
CL		 2806 break;
2807 default:
0a68b0be
BF		 2808 ret = ERR_PTR(-EAFNOSUPPORT);
2809 goto out_err;
8f9d5b1a
CL		 2810 }
2811
c740eff8
CL		 2812 if (xprt_bound(xprt))
2813 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2814 xprt->address_strings[RPC_DISPLAY_ADDR],
2815 xprt->address_strings[RPC_DISPLAY_PORT],
2816 xprt->address_strings[RPC_DISPLAY_PROTO]);
2817 else
2818 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
2819 xprt->address_strings[RPC_DISPLAY_ADDR],
2820 xprt->address_strings[RPC_DISPLAY_PROTO]);
edb267a6 2821
bc25571e
TT		 2822 if (try_module_get(THIS_MODULE))
2823 return xprt;
0a68b0be
BF		 2824 ret = ERR_PTR(-EINVAL);
2825out_err:
315f3812 2826 xs_xprt_free(xprt);
0a68b0be 2827 return ret;
a246b010
CL		 2828}
2829
2881ae74
TM		 2830static const struct rpc_timeout xs_tcp_default_timeout = {
2831 .to_initval = 60 * HZ,
2832 .to_maxval = 60 * HZ,
2833 .to_retries = 2,
2834};
2835
9903cd1c
CL		 2836/**
2837 * xs_setup_tcp - Set up transport to use a TCP socket
96802a09 2838 * @args: rpc transport creation arguments
9903cd1c
CL		 2839 *
2840 */
483066d6 2841static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
a246b010 2842{
8f9d5b1a 2843 struct sockaddr *addr = args->dstaddr;
c8541ecd 2844 struct rpc_xprt *xprt;
c8475461 2845 struct sock_xprt *transport;
0a68b0be 2846 struct rpc_xprt *ret;
b7993ceb
TM		 2847 unsigned int max_slot_table_size = xprt_max_tcp_slot_table_entries;
2848
2849 if (args->flags & XPRT_CREATE_INFINITE_SLOTS)
2850 max_slot_table_size = RPC_MAX_SLOT_TABLE_LIMIT;
a246b010 2851
d9ba131d 2852 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
b7993ceb 2853 max_slot_table_size);
c8541ecd
CL		 2854 if (IS_ERR(xprt))
2855 return xprt;
c8475461 2856 transport = container_of(xprt, struct sock_xprt, xprt);
a246b010 2857
ec739ef0 2858 xprt->prot = IPPROTO_TCP;
808012fb 2859 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
808012fb 2860 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
a246b010 2861
03bf4b70 2862 xprt->bind_timeout = XS_BIND_TO;
03bf4b70
CL		 2863 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2864 xprt->idle_timeout = XS_IDLE_DISC_TO;
a246b010 2865
262965f5 2866 xprt->ops = &xs_tcp_ops;
ba7392bb 2867 xprt->timeout = &xs_tcp_default_timeout;
a246b010 2868
8f9d5b1a
CL		 2869 switch (addr->sa_family) {
2870 case AF_INET:
2871 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2872 xprt_set_bound(xprt);
2873
9dc3b095 2874 INIT_DELAYED_WORK(&transport->connect_worker,
cdd518d5 2875 xs_tcp_setup_socket);
9dc3b095 2876 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
8f9d5b1a
CL		 2877 break;
2878 case AF_INET6:
2879 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2880 xprt_set_bound(xprt);
2881
9dc3b095 2882 INIT_DELAYED_WORK(&transport->connect_worker,
cdd518d5 2883 xs_tcp_setup_socket);
9dc3b095 2884 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
8f9d5b1a
CL		 2885 break;
2886 default:
0a68b0be
BF		 2887 ret = ERR_PTR(-EAFNOSUPPORT);
2888 goto out_err;
8f9d5b1a
CL		 2889 }
2890
c740eff8
CL		 2891 if (xprt_bound(xprt))
2892 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2893 xprt->address_strings[RPC_DISPLAY_ADDR],
2894 xprt->address_strings[RPC_DISPLAY_PORT],
2895 xprt->address_strings[RPC_DISPLAY_PROTO]);
2896 else
2897 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
2898 xprt->address_strings[RPC_DISPLAY_ADDR],
2899 xprt->address_strings[RPC_DISPLAY_PROTO]);
2900
bc25571e
TT		 2901 if (try_module_get(THIS_MODULE))
2902 return xprt;
0a68b0be
BF		 2903 ret = ERR_PTR(-EINVAL);
2904out_err:
315f3812 2905 xs_xprt_free(xprt);
0a68b0be 2906 return ret;
a246b010 2907}
282b32e1 2908
f300baba
AB		 2909/**
2910 * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket
2911 * @args: rpc transport creation arguments
2912 *
2913 */
2914static struct rpc_xprt *xs_setup_bc_tcp(struct xprt_create *args)
2915{
2916 struct sockaddr *addr = args->dstaddr;
2917 struct rpc_xprt *xprt;
2918 struct sock_xprt *transport;
2919 struct svc_sock *bc_sock;
0a68b0be 2920 struct rpc_xprt *ret;
f300baba 2921
f0418aa4
BF		 2922 if (args->bc_xprt->xpt_bc_xprt) {
2923 /*
2924 * This server connection already has a backchannel
e3bfab18
BF		 2925 * transport; we can't create a new one, as we wouldn't
2926 * be able to match replies based on xid any more. So,
f0418aa4
BF		 2927 * reuse the already-existing one:
2928 */
2929 return args->bc_xprt->xpt_bc_xprt;
2930 }
d9ba131d
TM		 2931 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2932 xprt_tcp_slot_table_entries);
f300baba
AB		 2933 if (IS_ERR(xprt))
2934 return xprt;
2935 transport = container_of(xprt, struct sock_xprt, xprt);
2936
2937 xprt->prot = IPPROTO_TCP;
2938 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2939 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2940 xprt->timeout = &xs_tcp_default_timeout;
2941
2942 /* backchannel */
2943 xprt_set_bound(xprt);
2944 xprt->bind_timeout = 0;
f300baba
AB		 2945 xprt->reestablish_timeout = 0;
2946 xprt->idle_timeout = 0;
2947
f300baba
AB		 2948 xprt->ops = &bc_tcp_ops;
2949
2950 switch (addr->sa_family) {
2951 case AF_INET:
2952 xs_format_peer_addresses(xprt, "tcp",
2953 RPCBIND_NETID_TCP);
2954 break;
2955 case AF_INET6:
2956 xs_format_peer_addresses(xprt, "tcp",
2957 RPCBIND_NETID_TCP6);
2958 break;
2959 default:
0a68b0be
BF		 2960 ret = ERR_PTR(-EAFNOSUPPORT);
2961 goto out_err;
f300baba
AB		 2962 }
2963
50fa0d40
PE		 2964 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2965 xprt->address_strings[RPC_DISPLAY_ADDR],
2966 xprt->address_strings[RPC_DISPLAY_PORT],
2967 xprt->address_strings[RPC_DISPLAY_PROTO]);
f300baba 2968
99de8ea9
BF		 2969 /*
2970 * Once we've associated a backchannel xprt with a connection,
28303ca3
WM		 2971 * we want to keep it around as long as the connection lasts,
2972 * in case we need to start using it for a backchannel again;
2973 * this reference won't be dropped until bc_xprt is destroyed.
99de8ea9
BF		 2974 */
2975 xprt_get(xprt);
2976 args->bc_xprt->xpt_bc_xprt = xprt;
2977 xprt->bc_xprt = args->bc_xprt;
2978 bc_sock = container_of(args->bc_xprt, struct svc_sock, sk_xprt);
2979 transport->sock = bc_sock->sk_sock;
2980 transport->inet = bc_sock->sk_sk;
2981
f300baba
AB		 2982 /*
2983 * Since we don't want connections for the backchannel, we set
2984 * the xprt status to connected
2985 */
2986 xprt_set_connected(xprt);
2987
f300baba
AB		 2988 if (try_module_get(THIS_MODULE))
2989 return xprt;
99de8ea9 2990 xprt_put(xprt);
0a68b0be
BF		 2991 ret = ERR_PTR(-EINVAL);
2992out_err:
315f3812 2993 xs_xprt_free(xprt);
0a68b0be 2994 return ret;
f300baba
AB		 2995}
2996
176e21ee
CL		 2997static struct xprt_class xs_local_transport = {
2998 .list = LIST_HEAD_INIT(xs_local_transport.list),
2999 .name = "named UNIX socket",
3000 .owner = THIS_MODULE,
3001 .ident = XPRT_TRANSPORT_LOCAL,
3002 .setup = xs_setup_local,
3003};
3004
bc25571e
TT		 3005static struct xprt_class xs_udp_transport = {
3006 .list = LIST_HEAD_INIT(xs_udp_transport.list),
3007 .name = "udp",
3008 .owner = THIS_MODULE,
f300baba 3009 .ident = XPRT_TRANSPORT_UDP,
bc25571e
TT		 3010 .setup = xs_setup_udp,
3011};
3012
3013static struct xprt_class xs_tcp_transport = {
3014 .list = LIST_HEAD_INIT(xs_tcp_transport.list),
3015 .name = "tcp",
3016 .owner = THIS_MODULE,
f300baba 3017 .ident = XPRT_TRANSPORT_TCP,
bc25571e
TT		 3018 .setup = xs_setup_tcp,
3019};
3020
f300baba
AB		 3021static struct xprt_class xs_bc_tcp_transport = {
3022 .list = LIST_HEAD_INIT(xs_bc_tcp_transport.list),
3023 .name = "tcp NFSv4.1 backchannel",
3024 .owner = THIS_MODULE,
3025 .ident = XPRT_TRANSPORT_BC_TCP,
3026 .setup = xs_setup_bc_tcp,
3027};
3028
282b32e1 3029/**
bc25571e 3030 * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
282b32e1
CL		 3031 *
3032 */
3033int init_socket_xprt(void)
3034{
fbf76683 3035#ifdef RPC_DEBUG
2b1bec5f 3036 if (!sunrpc_table_header)
0b4d4147 3037 sunrpc_table_header = register_sysctl_table(sunrpc_table);
fbf76683
CL		 3038#endif
3039
176e21ee 3040 xprt_register_transport(&xs_local_transport);
bc25571e
TT		 3041 xprt_register_transport(&xs_udp_transport);
3042 xprt_register_transport(&xs_tcp_transport);
f300baba 3043 xprt_register_transport(&xs_bc_tcp_transport);
bc25571e 3044
282b32e1
CL		 3045 return 0;
3046}
3047
3048/**
bc25571e 3049 * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
282b32e1
CL		 3050 *
3051 */
3052void cleanup_socket_xprt(void)
3053{
fbf76683
CL		 3054#ifdef RPC_DEBUG
3055 if (sunrpc_table_header) {
3056 unregister_sysctl_table(sunrpc_table_header);
3057 sunrpc_table_header = NULL;
3058 }
3059#endif
bc25571e 3060
176e21ee 3061 xprt_unregister_transport(&xs_local_transport);
bc25571e
TT		 3062 xprt_unregister_transport(&xs_udp_transport);
3063 xprt_unregister_transport(&xs_tcp_transport);
f300baba 3064 xprt_unregister_transport(&xs_bc_tcp_transport);
282b32e1 3065}
cbf11071 3066
9bbb9e5a
RR		 3067static int param_set_uint_minmax(const char *val,
3068 const struct kernel_param *kp,
cbf11071
TM		 3069 unsigned int min, unsigned int max)
3070{
3071 unsigned long num;
3072 int ret;
3073
3074 if (!val)
3075 return -EINVAL;
3076 ret = strict_strtoul(val, 0, &num);
3077 if (ret == -EINVAL || num < min || num > max)
3078 return -EINVAL;
3079 *((unsigned int *)kp->arg) = num;
3080 return 0;
3081}
3082
9bbb9e5a 3083static int param_set_portnr(const char *val, const struct kernel_param *kp)
cbf11071
TM		 3084{
3085 return param_set_uint_minmax(val, kp,
3086 RPC_MIN_RESVPORT,
3087 RPC_MAX_RESVPORT);
3088}
3089
9bbb9e5a
RR		 3090static struct kernel_param_ops param_ops_portnr = {
3091 .set = param_set_portnr,
3092 .get = param_get_uint,
3093};
3094
cbf11071
TM		 3095#define param_check_portnr(name, p) \
3096 __param_check(name, p, unsigned int);
3097
3098module_param_named(min_resvport, xprt_min_resvport, portnr, 0644);
3099module_param_named(max_resvport, xprt_max_resvport, portnr, 0644);
3100
9bbb9e5a
RR		 3101static int param_set_slot_table_size(const char *val,
3102 const struct kernel_param *kp)
cbf11071
TM		 3103{
3104 return param_set_uint_minmax(val, kp,
3105 RPC_MIN_SLOT_TABLE,
3106 RPC_MAX_SLOT_TABLE);
3107}
3108
9bbb9e5a
RR		 3109static struct kernel_param_ops param_ops_slot_table_size = {
3110 .set = param_set_slot_table_size,
3111 .get = param_get_uint,
3112};
3113
cbf11071
TM		 3114#define param_check_slot_table_size(name, p) \
3115 __param_check(name, p, unsigned int);
3116
d9ba131d
TM		 3117static int param_set_max_slot_table_size(const char *val,
3118 const struct kernel_param *kp)
3119{
3120 return param_set_uint_minmax(val, kp,
3121 RPC_MIN_SLOT_TABLE,
3122 RPC_MAX_SLOT_TABLE_LIMIT);
3123}
3124
3125static struct kernel_param_ops param_ops_max_slot_table_size = {
3126 .set = param_set_max_slot_table_size,
3127 .get = param_get_uint,
3128};
3129
3130#define param_check_max_slot_table_size(name, p) \
3131 __param_check(name, p, unsigned int);
3132
cbf11071
TM		 3133module_param_named(tcp_slot_table_entries, xprt_tcp_slot_table_entries,
3134 slot_table_size, 0644);
d9ba131d
TM		 3135module_param_named(tcp_max_slot_table_entries, xprt_max_tcp_slot_table_entries,
3136 max_slot_table_size, 0644);
cbf11071
TM		 3137module_param_named(udp_slot_table_entries, xprt_udp_slot_table_entries,
3138 slot_table_size, 0644);
3139
Linux 6.x block layer and io_uring tree(s)