projects / linux-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
sunrpc: translate -EAGAIN to -ENOBUFS when socket is writable.
[linux-block.git] / net / sunrpc / xprtsock.c
CommitLineData
a246b010
CL		 1/*
2 * linux/net/sunrpc/xprtsock.c
3 *
4 * Client-side transport implementation for sockets.
5 *
113aa838
AC		 6 * TCP callback races fixes (C) 1998 Red Hat
7 * TCP send fixes (C) 1998 Red Hat
a246b010
CL		 8 * TCP NFS related read + write fixes
9 * (C) 1999 Dave Airlie, University of Limerick, Ireland <airlied@linux.ie>
10 *
11 * Rewrite of larges part of the code in order to stabilize TCP stuff.
12 * Fix behaviour when socket buffer is full.
13 * (C) 1999 Trond Myklebust <trond.myklebust@fys.uio.no>
55aa4f58
CL		 14 *
15 * IP socket transport implementation, (C) 2005 Chuck Lever <cel@netapp.com>
8f9d5b1a
CL		 16 *
17 * IPv6 support contributed by Gilles Quillard, Bull Open Source, 2005.
18 * <gilles.quillard@bull.net>
a246b010
CL		 19 */
20
21#include <linux/types.h>
176e21ee 22#include <linux/string.h>
a246b010 23#include <linux/slab.h>
bc25571e 24#include <linux/module.h>
a246b010 25#include <linux/capability.h>
a246b010
CL		 26#include <linux/pagemap.h>
27#include <linux/errno.h>
28#include <linux/socket.h>
29#include <linux/in.h>
30#include <linux/net.h>
31#include <linux/mm.h>
176e21ee 32#include <linux/un.h>
a246b010
CL		 33#include <linux/udp.h>
34#include <linux/tcp.h>
35#include <linux/sunrpc/clnt.h>
5976687a 36#include <linux/sunrpc/addr.h>
02107148 37#include <linux/sunrpc/sched.h>
4cfc7e60 38#include <linux/sunrpc/svcsock.h>
49c36fcc 39#include <linux/sunrpc/xprtsock.h>
a246b010 40#include <linux/file.h>
9e00abc3 41#ifdef CONFIG_SUNRPC_BACKCHANNEL
44b98efd
RL		 42#include <linux/sunrpc/bc_xprt.h>
43#endif
a246b010
CL		 44
45#include <net/sock.h>
46#include <net/checksum.h>
47#include <net/udp.h>
48#include <net/tcp.h>
49
40b5ea0c
TM		 50#include <trace/events/sunrpc.h>
51
4cfc7e60 52#include "sunrpc.h"
176e21ee
CL		 53
54static void xs_close(struct rpc_xprt *xprt);
55
c556b754
CL		 56/*
57 * xprtsock tunables
58 */
09acfea5
TM		 59static unsigned int xprt_udp_slot_table_entries = RPC_DEF_SLOT_TABLE;
60static unsigned int xprt_tcp_slot_table_entries = RPC_MIN_SLOT_TABLE;
61static unsigned int xprt_max_tcp_slot_table_entries = RPC_MAX_SLOT_TABLE;
c556b754 62
09acfea5
TM		 63static unsigned int xprt_min_resvport = RPC_DEF_MIN_RESVPORT;
64static unsigned int xprt_max_resvport = RPC_DEF_MAX_RESVPORT;
c556b754 65
54c09874
TM		 66#if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
67
7d1e8255 68#define XS_TCP_LINGER_TO (15U * HZ)
25fe6142 69static unsigned int xs_tcp_fin_timeout __read_mostly = XS_TCP_LINGER_TO;
7d1e8255 70
fbf76683
CL		 71/*
72 * We can register our own files under /proc/sys/sunrpc by
73 * calling register_sysctl_table() again. The files in that
74 * directory become the union of all files registered there.
75 *
76 * We simply need to make sure that we don't collide with
77 * someone else's file names!
78 */
79
fbf76683
CL		 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
f895b252 189#if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
a246b010 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
2118071d
TM		 219static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
220{
221 return (struct rpc_xprt *) sk->sk_user_data;
222}
223
95392c59
CL		 224static inline struct sockaddr *xs_addr(struct rpc_xprt *xprt)
225{
226 return (struct sockaddr *) &xprt->addr;
227}
228
176e21ee
CL		 229static inline struct sockaddr_un *xs_addr_un(struct rpc_xprt *xprt)
230{
231 return (struct sockaddr_un *) &xprt->addr;
232}
233
95392c59 234static inline struct sockaddr_in *xs_addr_in(struct rpc_xprt *xprt)
edb267a6 235{
95392c59
CL		 236 return (struct sockaddr_in *) &xprt->addr;
237}
238
239static inline struct sockaddr_in6 *xs_addr_in6(struct rpc_xprt *xprt)
240{
241 return (struct sockaddr_in6 *) &xprt->addr;
242}
243
c877b849 244static void xs_format_common_peer_addresses(struct rpc_xprt *xprt)
edb267a6 245{
c877b849 246 struct sockaddr *sap = xs_addr(xprt);
9dc3b095
CL		 247 struct sockaddr_in6 *sin6;
248 struct sockaddr_in *sin;
176e21ee 249 struct sockaddr_un *sun;
c877b849 250 char buf[128];
edb267a6 251
9dc3b095 252 switch (sap->sa_family) {
176e21ee
CL		 253 case AF_LOCAL:
254 sun = xs_addr_un(xprt);
255 strlcpy(buf, sun->sun_path, sizeof(buf));
256 xprt->address_strings[RPC_DISPLAY_ADDR] =
257 kstrdup(buf, GFP_KERNEL);
258 break;
9dc3b095 259 case AF_INET:
176e21ee
CL		 260 (void)rpc_ntop(sap, buf, sizeof(buf));
261 xprt->address_strings[RPC_DISPLAY_ADDR] =
262 kstrdup(buf, GFP_KERNEL);
9dc3b095 263 sin = xs_addr_in(xprt);
fc0b5791 264 snprintf(buf, sizeof(buf), "%08x", ntohl(sin->sin_addr.s_addr));
9dc3b095
CL		 265 break;
266 case AF_INET6:
176e21ee
CL		 267 (void)rpc_ntop(sap, buf, sizeof(buf));
268 xprt->address_strings[RPC_DISPLAY_ADDR] =
269 kstrdup(buf, GFP_KERNEL);
9dc3b095 270 sin6 = xs_addr_in6(xprt);
fc0b5791 271 snprintf(buf, sizeof(buf), "%pi6", &sin6->sin6_addr);
9dc3b095
CL		 272 break;
273 default:
274 BUG();
756805e7 275 }
176e21ee 276
9dc3b095 277 xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = kstrdup(buf, GFP_KERNEL);
edb267a6
CL		 278}
279
9dc3b095 280static void xs_format_common_peer_ports(struct rpc_xprt *xprt)
4b6473fb 281{
9dc3b095
CL		 282 struct sockaddr *sap = xs_addr(xprt);
283 char buf[128];
4b6473fb 284
81160e66 285 snprintf(buf, sizeof(buf), "%u", rpc_get_port(sap));
c877b849 286 xprt->address_strings[RPC_DISPLAY_PORT] = kstrdup(buf, GFP_KERNEL);
4b6473fb 287
81160e66 288 snprintf(buf, sizeof(buf), "%4hx", rpc_get_port(sap));
c877b849
CL		 289 xprt->address_strings[RPC_DISPLAY_HEX_PORT] = kstrdup(buf, GFP_KERNEL);
290}
4b6473fb 291
9dc3b095
CL		 292static void xs_format_peer_addresses(struct rpc_xprt *xprt,
293 const char *protocol,
294 const char *netid)
c877b849 295{
b454ae90 296 xprt->address_strings[RPC_DISPLAY_PROTO] = protocol;
b454ae90 297 xprt->address_strings[RPC_DISPLAY_NETID] = netid;
c877b849 298 xs_format_common_peer_addresses(xprt);
9dc3b095 299 xs_format_common_peer_ports(xprt);
edb267a6 300}
4b6473fb 301
9dc3b095 302static void xs_update_peer_port(struct rpc_xprt *xprt)
4b6473fb 303{
9dc3b095
CL		 304 kfree(xprt->address_strings[RPC_DISPLAY_HEX_PORT]);
305 kfree(xprt->address_strings[RPC_DISPLAY_PORT]);
4417c8c4 306
9dc3b095 307 xs_format_common_peer_ports(xprt);
edb267a6
CL		 308}
309
310static void xs_free_peer_addresses(struct rpc_xprt *xprt)
311{
33e01dc7
CL		 312 unsigned int i;
313
314 for (i = 0; i < RPC_DISPLAY_MAX; i++)
315 switch (i) {
316 case RPC_DISPLAY_PROTO:
317 case RPC_DISPLAY_NETID:
318 continue;
319 default:
320 kfree(xprt->address_strings[i]);
321 }
edb267a6
CL		 322}
323
b4b5cc85
CL		 324#define XS_SENDMSG_FLAGS (MSG_DONTWAIT | MSG_NOSIGNAL)
325
24c5684b 326static int xs_send_kvec(struct socket *sock, struct sockaddr *addr, int addrlen, struct kvec *vec, unsigned int base, int more)
b4b5cc85 327{
b4b5cc85
CL		 328 struct msghdr msg = {
329 .msg_name = addr,
330 .msg_namelen = addrlen,
24c5684b
TM		 331 .msg_flags = XS_SENDMSG_FLAGS | (more ? MSG_MORE : 0),
332 };
333 struct kvec iov = {
334 .iov_base = vec->iov_base + base,
335 .iov_len = vec->iov_len - base,
b4b5cc85
CL		 336 };
337
24c5684b 338 if (iov.iov_len != 0)
b4b5cc85
CL		 339 return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
340 return kernel_sendmsg(sock, &msg, NULL, 0, 0);
341}
342
f279cd00 343static int xs_send_pagedata(struct socket *sock, struct xdr_buf *xdr, unsigned int base, int more, bool zerocopy, int *sent_p)
b4b5cc85 344{
a6b31d18
TM		 345 ssize_t (*do_sendpage)(struct socket *sock, struct page *page,
346 int offset, size_t size, int flags);
24c5684b
TM		 347 struct page **ppage;
348 unsigned int remainder;
f279cd00 349 int err;
24c5684b
TM		 350
351 remainder = xdr->page_len - base;
352 base += xdr->page_base;
353 ppage = xdr->pages + (base >> PAGE_SHIFT);
354 base &= ~PAGE_MASK;
a6b31d18
TM		 355 do_sendpage = sock->ops->sendpage;
356 if (!zerocopy)
357 do_sendpage = sock_no_sendpage;
24c5684b
TM		 358 for(;;) {
359 unsigned int len = min_t(unsigned int, PAGE_SIZE - base, remainder);
360 int flags = XS_SENDMSG_FLAGS;
b4b5cc85 361
24c5684b
TM		 362 remainder -= len;
363 if (remainder != 0 || more)
364 flags |= MSG_MORE;
a6b31d18 365 err = do_sendpage(sock, *ppage, base, len, flags);
24c5684b
TM		 366 if (remainder == 0 || err != len)
367 break;
f279cd00 368 *sent_p += err;
24c5684b
TM		 369 ppage++;
370 base = 0;
371 }
f279cd00
JB		 372 if (err > 0) {
373 *sent_p += err;
374 err = 0;
375 }
376 return err;
b4b5cc85
CL		 377}
378
9903cd1c
CL		 379/**
380 * xs_sendpages - write pages directly to a socket
381 * @sock: socket to send on
382 * @addr: UDP only -- address of destination
383 * @addrlen: UDP only -- length of destination address
384 * @xdr: buffer containing this request
385 * @base: starting position in the buffer
a6b31d18 386 * @zerocopy: true if it is safe to use sendpage()
f279cd00 387 * @sent_p: return the total number of bytes successfully queued for sending
9903cd1c 388 *
a246b010 389 */
f279cd00 390static int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base, bool zerocopy, int *sent_p)
a246b010 391{
24c5684b 392 unsigned int remainder = xdr->len - base;
f279cd00
JB		 393 int err = 0;
394 int sent = 0;
a246b010 395
262965f5 396 if (unlikely(!sock))
fba91afb 397 return -ENOTSOCK;
262965f5
CL		 398
399 clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags);
24c5684b
TM		 400 if (base != 0) {
401 addr = NULL;
402 addrlen = 0;
403 }
262965f5 404
24c5684b
TM		 405 if (base < xdr->head[0].iov_len || addr != NULL) {
406 unsigned int len = xdr->head[0].iov_len - base;
407 remainder -= len;
408 err = xs_send_kvec(sock, addr, addrlen, &xdr->head[0], base, remainder != 0);
409 if (remainder == 0 || err != len)
a246b010 410 goto out;
f279cd00 411 *sent_p += err;
a246b010
CL		 412 base = 0;
413 } else
24c5684b 414 base -= xdr->head[0].iov_len;
a246b010 415
24c5684b
TM		 416 if (base < xdr->page_len) {
417 unsigned int len = xdr->page_len - base;
418 remainder -= len;
f279cd00
JB		 419 err = xs_send_pagedata(sock, xdr, base, remainder != 0, zerocopy, &sent);
420 *sent_p += sent;
421 if (remainder == 0 || sent != len)
a246b010
CL		 422 goto out;
423 base = 0;
24c5684b
TM		 424 } else
425 base -= xdr->page_len;
426
427 if (base >= xdr->tail[0].iov_len)
f279cd00 428 return 0;
24c5684b 429 err = xs_send_kvec(sock, NULL, 0, &xdr->tail[0], base, 0);
a246b010 430out:
f279cd00
JB		 431 if (err > 0) {
432 *sent_p += err;
433 err = 0;
434 }
435 return err;
a246b010
CL		 436}
437
b6ddf64f
TM		 438static void xs_nospace_callback(struct rpc_task *task)
439{
440 struct sock_xprt *transport = container_of(task->tk_rqstp->rq_xprt, struct sock_xprt, xprt);
441
442 transport->inet->sk_write_pending--;
443 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
444}
445
9903cd1c 446/**
262965f5
CL		 447 * xs_nospace - place task on wait queue if transmit was incomplete
448 * @task: task to put to sleep
9903cd1c 449 *
a246b010 450 */
5e3771ce 451static int xs_nospace(struct rpc_task *task)
a246b010 452{
262965f5
CL		 453 struct rpc_rqst *req = task->tk_rqstp;
454 struct rpc_xprt *xprt = req->rq_xprt;
ee0ac0c2 455 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
06ea0bfe 456 struct sock *sk = transport->inet;
24ca9a84 457 int ret = -EAGAIN;
a246b010 458
46121cf7 459 dprintk("RPC: %5u xmit incomplete (%u left of %u)\n",
262965f5
CL		 460 task->tk_pid, req->rq_slen - req->rq_bytes_sent,
461 req->rq_slen);
462
b6ddf64f
TM		 463 /* Protect against races with write_space */
464 spin_lock_bh(&xprt->transport_lock);
465
466 /* Don't race with disconnect */
467 if (xprt_connected(xprt)) {
468 if (test_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags)) {
469 /*
470 * Notify TCP that we're limited by the application
471 * window size
472 */
473 set_bit(SOCK_NOSPACE, &transport->sock->flags);
06ea0bfe 474 sk->sk_write_pending++;
b6ddf64f
TM		 475 /* ...and wait for more buffer space */
476 xprt_wait_for_buffer_space(task, xs_nospace_callback);
477 }
478 } else {
479 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
5e3771ce 480 ret = -ENOTCONN;
b6ddf64f 481 }
262965f5 482
b6ddf64f 483 spin_unlock_bh(&xprt->transport_lock);
06ea0bfe
TM		 484
485 /* Race breaker in case memory is freed before above code is called */
486 sk->sk_write_space(sk);
5e3771ce 487 return ret;
262965f5
CL		 488}
489
61677eee
CL		 490/*
491 * Construct a stream transport record marker in @buf.
492 */
493static inline void xs_encode_stream_record_marker(struct xdr_buf *buf)
494{
495 u32 reclen = buf->len - sizeof(rpc_fraghdr);
496 rpc_fraghdr *base = buf->head[0].iov_base;
497 *base = cpu_to_be32(RPC_LAST_STREAM_FRAGMENT | reclen);
498}
499
176e21ee
CL		 500/**
501 * xs_local_send_request - write an RPC request to an AF_LOCAL socket
502 * @task: RPC task that manages the state of an RPC request
503 *
504 * Return values:
505 * 0: The request has been sent
506 * EAGAIN: The socket was blocked, please call again later to
507 * complete the request
508 * ENOTCONN: Caller needs to invoke connect logic then call again
509 * other: Some other error occured, the request was not sent
510 */
511static int xs_local_send_request(struct rpc_task *task)
512{
513 struct rpc_rqst *req = task->tk_rqstp;
514 struct rpc_xprt *xprt = req->rq_xprt;
515 struct sock_xprt *transport =
516 container_of(xprt, struct sock_xprt, xprt);
517 struct xdr_buf *xdr = &req->rq_snd_buf;
518 int status;
f279cd00 519 int sent = 0;
176e21ee
CL		 520
521 xs_encode_stream_record_marker(&req->rq_snd_buf);
522
523 xs_pktdump("packet data:",
524 req->rq_svec->iov_base, req->rq_svec->iov_len);
525
f279cd00
JB		 526 status = xs_sendpages(transport->sock, NULL, 0, xdr, req->rq_bytes_sent,
527 true, &sent);
176e21ee
CL		 528 dprintk("RPC: %s(%u) = %d\n",
529 __func__, xdr->len - req->rq_bytes_sent, status);
743c69e7
N		 530
531 if (status == -EAGAIN && sock_writeable(transport->inet))
532 status = -ENOBUFS;
533
f279cd00
JB		 534 if (likely(sent > 0) || status == 0) {
535 req->rq_bytes_sent += sent;
536 req->rq_xmit_bytes_sent += sent;
176e21ee
CL		 537 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
538 req->rq_bytes_sent = 0;
539 return 0;
540 }
541 status = -EAGAIN;
542 }
543
544 switch (status) {
3601c4a9 545 case -ENOBUFS:
b5872f0c 546 break;
176e21ee
CL		 547 case -EAGAIN:
548 status = xs_nospace(task);
549 break;
550 default:
551 dprintk("RPC: sendmsg returned unrecognized error %d\n",
552 -status);
553 case -EPIPE:
554 xs_close(xprt);
555 status = -ENOTCONN;
556 }
557
558 return status;
559}
560
262965f5
CL		 561/**
562 * xs_udp_send_request - write an RPC request to a UDP socket
563 * @task: address of RPC task that manages the state of an RPC request
564 *
565 * Return values:
566 * 0: The request has been sent
567 * EAGAIN: The socket was blocked, please call again later to
568 * complete the request
569 * ENOTCONN: Caller needs to invoke connect logic then call again
25985edc 570 * other: Some other error occurred, the request was not sent
262965f5
CL		 571 */
572static int xs_udp_send_request(struct rpc_task *task)
573{
574 struct rpc_rqst *req = task->tk_rqstp;
575 struct rpc_xprt *xprt = req->rq_xprt;
ee0ac0c2 576 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
262965f5 577 struct xdr_buf *xdr = &req->rq_snd_buf;
f279cd00 578 int sent = 0;
262965f5 579 int status;
a246b010 580
9903cd1c 581 xs_pktdump("packet data:",
a246b010
CL		 582 req->rq_svec->iov_base,
583 req->rq_svec->iov_len);
584
01d37c42
TM		 585 if (!xprt_bound(xprt))
586 return -ENOTCONN;
f279cd00
JB		 587 status = xs_sendpages(transport->sock, xs_addr(xprt), xprt->addrlen,
588 xdr, req->rq_bytes_sent, true, &sent);
a246b010 589
46121cf7 590 dprintk("RPC: xs_udp_send_request(%u) = %d\n",
262965f5 591 xdr->len - req->rq_bytes_sent, status);
a246b010 592
3dedbb5c
JB		 593 /* firewall is blocking us, don't return -EAGAIN or we end up looping */
594 if (status == -EPERM)
595 goto process_status;
596
743c69e7
N		 597 if (status == -EAGAIN && sock_writeable(transport->inet))
598 status = -ENOBUFS;
599
f279cd00
JB		 600 if (sent > 0 || status == 0) {
601 req->rq_xmit_bytes_sent += sent;
602 if (sent >= req->rq_slen)
2199700f
TM		 603 return 0;
604 /* Still some bytes left; set up for a retry later. */
262965f5 605 status = -EAGAIN;
2199700f 606 }
a246b010 607
3dedbb5c 608process_status:
262965f5 609 switch (status) {
fba91afb
TM		 610 case -ENOTSOCK:
611 status = -ENOTCONN;
612 /* Should we call xs_close() here? */
613 break;
b6ddf64f 614 case -EAGAIN:
5e3771ce 615 status = xs_nospace(task);
b6ddf64f 616 break;
c8485e4d
TM		 617 default:
618 dprintk("RPC: sendmsg returned unrecognized error %d\n",
619 -status);
262965f5 620 case -ENETUNREACH:
3601c4a9 621 case -ENOBUFS:
262965f5 622 case -EPIPE:
a246b010 623 case -ECONNREFUSED:
3dedbb5c 624 case -EPERM:
a246b010 625 /* When the server has died, an ICMP port unreachable message
9903cd1c 626 * prompts ECONNREFUSED. */
b6ddf64f 627 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
a246b010 628 }
5fe46e9d 629
262965f5 630 return status;
a246b010
CL		 631}
632
9903cd1c 633/**
262965f5 634 * xs_tcp_send_request - write an RPC request to a TCP socket
9903cd1c
CL		 635 * @task: address of RPC task that manages the state of an RPC request
636 *
637 * Return values:
262965f5
CL		 638 * 0: The request has been sent
639 * EAGAIN: The socket was blocked, please call again later to
640 * complete the request
641 * ENOTCONN: Caller needs to invoke connect logic then call again
25985edc 642 * other: Some other error occurred, the request was not sent
9903cd1c
CL		 643 *
644 * XXX: In the case of soft timeouts, should we eventually give up
262965f5 645 * if sendmsg is not able to make progress?
9903cd1c 646 */
262965f5 647static int xs_tcp_send_request(struct rpc_task *task)
a246b010
CL		 648{
649 struct rpc_rqst *req = task->tk_rqstp;
650 struct rpc_xprt *xprt = req->rq_xprt;
ee0ac0c2 651 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
262965f5 652 struct xdr_buf *xdr = &req->rq_snd_buf;
a6b31d18 653 bool zerocopy = true;
b595bb15 654 int status;
f279cd00 655 int sent;
a246b010 656
61677eee 657 xs_encode_stream_record_marker(&req->rq_snd_buf);
a246b010 658
262965f5
CL		 659 xs_pktdump("packet data:",
660 req->rq_svec->iov_base,
661 req->rq_svec->iov_len);
a6b31d18
TM		 662 /* Don't use zero copy if this is a resend. If the RPC call
663 * completes while the socket holds a reference to the pages,
664 * then we may end up resending corrupted data.
665 */
666 if (task->tk_flags & RPC_TASK_SENT)
667 zerocopy = false;
a246b010
CL		 668
669 /* Continue transmitting the packet/record. We must be careful
670 * to cope with writespace callbacks arriving _after_ we have
262965f5 671 * called sendmsg(). */
a246b010 672 while (1) {
f279cd00
JB		 673 sent = 0;
674 status = xs_sendpages(transport->sock, NULL, 0, xdr,
675 req->rq_bytes_sent, zerocopy, &sent);
a246b010 676
46121cf7 677 dprintk("RPC: xs_tcp_send_request(%u) = %d\n",
262965f5 678 xdr->len - req->rq_bytes_sent, status);
a246b010 679
f279cd00 680 if (unlikely(sent == 0 && status < 0))
a246b010 681 break;
a246b010 682
262965f5
CL		 683 /* If we've sent the entire packet, immediately
684 * reset the count of bytes sent. */
f279cd00
JB		 685 req->rq_bytes_sent += sent;
686 req->rq_xmit_bytes_sent += sent;
262965f5
CL		 687 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
688 req->rq_bytes_sent = 0;
689 return 0;
690 }
a246b010 691
f279cd00 692 if (sent != 0)
06b4b681 693 continue;
a246b010 694 status = -EAGAIN;
06b4b681 695 break;
a246b010 696 }
743c69e7
N		 697 if (status == -EAGAIN && sk_stream_is_writeable(transport->inet))
698 status = -ENOBUFS;
a246b010 699
262965f5 700 switch (status) {
fba91afb
TM		 701 case -ENOTSOCK:
702 status = -ENOTCONN;
703 /* Should we call xs_close() here? */
704 break;
262965f5 705 case -EAGAIN:
5e3771ce 706 status = xs_nospace(task);
262965f5 707 break;
c8485e4d
TM		 708 default:
709 dprintk("RPC: sendmsg returned unrecognized error %d\n",
710 -status);
262965f5 711 case -ECONNRESET:
2a9e1cfa 712 case -ECONNREFUSED:
262965f5 713 case -ENOTCONN:
3913c78c 714 case -EADDRINUSE:
b5872f0c 715 case -ENOBUFS:
b9d2bb2e 716 case -EPIPE:
b6ddf64f 717 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
a246b010 718 }
5fe46e9d 719
a246b010
CL		 720 return status;
721}
722
e0ab53de
TM		 723/**
724 * xs_tcp_release_xprt - clean up after a tcp transmission
725 * @xprt: transport
726 * @task: rpc task
727 *
728 * This cleans up if an error causes us to abort the transmission of a request.
729 * In this case, the socket may need to be reset in order to avoid confusing
730 * the server.
731 */
732static void xs_tcp_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
733{
734 struct rpc_rqst *req;
735
736 if (task != xprt->snd_task)
737 return;
738 if (task == NULL)
739 goto out_release;
740 req = task->tk_rqstp;
43cedbf0
TM		 741 if (req == NULL)
742 goto out_release;
e0ab53de
TM		 743 if (req->rq_bytes_sent == 0)
744 goto out_release;
745 if (req->rq_bytes_sent == req->rq_snd_buf.len)
746 goto out_release;
a4f0835c 747 set_bit(XPRT_CLOSE_WAIT, &xprt->state);
e0ab53de
TM		 748out_release:
749 xprt_release_xprt(xprt, task);
750}
751
2a9e1cfa
TM		 752static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk)
753{
754 transport->old_data_ready = sk->sk_data_ready;
755 transport->old_state_change = sk->sk_state_change;
756 transport->old_write_space = sk->sk_write_space;
2118071d 757 transport->old_error_report = sk->sk_error_report;
2a9e1cfa
TM		 758}
759
760static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk)
761{
762 sk->sk_data_ready = transport->old_data_ready;
763 sk->sk_state_change = transport->old_state_change;
764 sk->sk_write_space = transport->old_write_space;
2118071d
TM		 765 sk->sk_error_report = transport->old_error_report;
766}
767
b70ae915
TM		 768static void xs_sock_reset_connection_flags(struct rpc_xprt *xprt)
769{
770 smp_mb__before_atomic();
771 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
772 clear_bit(XPRT_CLOSING, &xprt->state);
773 smp_mb__after_atomic();
774}
775
776static void xs_sock_mark_closed(struct rpc_xprt *xprt)
777{
778 xs_sock_reset_connection_flags(xprt);
779 /* Mark transport as closed and wake up all pending tasks */
780 xprt_disconnect_done(xprt);
4876cc77 781 xprt_force_disconnect(xprt);
b70ae915
TM		 782}
783
2118071d
TM		 784/**
785 * xs_error_report - callback to handle TCP socket state errors
786 * @sk: socket
787 *
788 * Note: we don't call sock_error() since there may be a rpc_task
789 * using the socket, and so we don't want to clear sk->sk_err.
790 */
791static void xs_error_report(struct sock *sk)
792{
793 struct rpc_xprt *xprt;
794 int err;
795
796 read_lock_bh(&sk->sk_callback_lock);
797 if (!(xprt = xprt_from_sock(sk)))
798 goto out;
799
800 err = -sk->sk_err;
801 if (err == 0)
802 goto out;
b70ae915
TM		 803 /* Is this a reset event? */
804 if (sk->sk_state == TCP_CLOSE)
805 xs_sock_mark_closed(xprt);
2118071d
TM		 806 dprintk("RPC: xs_error_report client %p, error=%d...\n",
807 xprt, -err);
e8353c76 808 trace_rpc_socket_error(xprt, sk->sk_socket, err);
2118071d
TM		 809 xprt_wake_pending_tasks(xprt, err);
810 out:
811 read_unlock_bh(&sk->sk_callback_lock);
2a9e1cfa
TM		 812}
813
fe315e76 814static void xs_reset_transport(struct sock_xprt *transport)
a246b010 815{
ee0ac0c2
CL		 816 struct socket *sock = transport->sock;
817 struct sock *sk = transport->inet;
6cc7e908 818 struct rpc_xprt *xprt = &transport->xprt;
a246b010 819
fe315e76
CL		 820 if (sk == NULL)
821 return;
9903cd1c 822
264d1df3
JL		 823 if (atomic_read(&transport->xprt.swapper))
824 sk_clear_memalloc(sk);
825
a246b010 826 write_lock_bh(&sk->sk_callback_lock);
ee0ac0c2
CL		 827 transport->inet = NULL;
828 transport->sock = NULL;
a246b010 829
9903cd1c 830 sk->sk_user_data = NULL;
2a9e1cfa
TM		 831
832 xs_restore_old_callbacks(transport, sk);
a246b010 833 write_unlock_bh(&sk->sk_callback_lock);
6cc7e908 834 xs_sock_reset_connection_flags(xprt);
a246b010 835
6cc7e908 836 trace_rpc_socket_close(xprt, sock);
a246b010 837 sock_release(sock);
fe315e76
CL		 838}
839
840/**
841 * xs_close - close a socket
842 * @xprt: transport
843 *
844 * This is used when all requests are complete; ie, no DRC state remains
845 * on the server we want to save.
f75e6745
TM		 846 *
847 * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with
848 * xs_reset_transport() zeroing the socket from underneath a writer.
fe315e76
CL		 849 */
850static void xs_close(struct rpc_xprt *xprt)
851{
852 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
853
854 dprintk("RPC: xs_close xprt %p\n", xprt);
855
856 xs_reset_transport(transport);
61d0a8e6 857 xprt->reestablish_timeout = 0;
fe315e76 858
62da3b24 859 xprt_disconnect_done(xprt);
a246b010
CL		 860}
861
4a068258
CL		 862static void xs_inject_disconnect(struct rpc_xprt *xprt)
863{
864 dprintk("RPC: injecting transport disconnect on xprt=%p\n",
865 xprt);
866 xprt_disconnect_done(xprt);
867}
868
315f3812
KM		 869static void xs_xprt_free(struct rpc_xprt *xprt)
870{
871 xs_free_peer_addresses(xprt);
872 xprt_free(xprt);
873}
874
9903cd1c
CL		 875/**
876 * xs_destroy - prepare to shutdown a transport
877 * @xprt: doomed transport
878 *
879 */
880static void xs_destroy(struct rpc_xprt *xprt)
a246b010 881{
46121cf7 882 dprintk("RPC: xs_destroy xprt %p\n", xprt);
9903cd1c 883
a1311d87 884 xs_close(xprt);
315f3812 885 xs_xprt_free(xprt);
a1311d87 886 module_put(THIS_MODULE);
a246b010
CL		 887}
888
176e21ee
CL		 889static int xs_local_copy_to_xdr(struct xdr_buf *xdr, struct sk_buff *skb)
890{
891 struct xdr_skb_reader desc = {
892 .skb = skb,
893 .offset = sizeof(rpc_fraghdr),
894 .count = skb->len - sizeof(rpc_fraghdr),
895 };
896
897 if (xdr_partial_copy_from_skb(xdr, 0, &desc, xdr_skb_read_bits) < 0)
898 return -1;
899 if (desc.count)
900 return -1;
901 return 0;
902}
903
904/**
905 * xs_local_data_ready - "data ready" callback for AF_LOCAL sockets
906 * @sk: socket with data to read
176e21ee
CL		 907 *
908 * Currently this assumes we can read the whole reply in a single gulp.
909 */
676d2369 910static void xs_local_data_ready(struct sock *sk)
176e21ee
CL		 911{
912 struct rpc_task *task;
913 struct rpc_xprt *xprt;
914 struct rpc_rqst *rovr;
915 struct sk_buff *skb;
916 int err, repsize, copied;
917 u32 _xid;
918 __be32 *xp;
919
920 read_lock_bh(&sk->sk_callback_lock);
921 dprintk("RPC: %s...\n", __func__);
922 xprt = xprt_from_sock(sk);
923 if (xprt == NULL)
924 goto out;
925
926 skb = skb_recv_datagram(sk, 0, 1, &err);
927 if (skb == NULL)
928 goto out;
929
176e21ee
CL		 930 repsize = skb->len - sizeof(rpc_fraghdr);
931 if (repsize < 4) {
932 dprintk("RPC: impossible RPC reply size %d\n", repsize);
933 goto dropit;
934 }
935
936 /* Copy the XID from the skb... */
937 xp = skb_header_pointer(skb, sizeof(rpc_fraghdr), sizeof(_xid), &_xid);
938 if (xp == NULL)
939 goto dropit;
940
941 /* Look up and lock the request corresponding to the given XID */
942 spin_lock(&xprt->transport_lock);
943 rovr = xprt_lookup_rqst(xprt, *xp);
944 if (!rovr)
945 goto out_unlock;
946 task = rovr->rq_task;
947
948 copied = rovr->rq_private_buf.buflen;
949 if (copied > repsize)
950 copied = repsize;
951
952 if (xs_local_copy_to_xdr(&rovr->rq_private_buf, skb)) {
953 dprintk("RPC: sk_buff copy failed\n");
954 goto out_unlock;
955 }
956
957 xprt_complete_rqst(task, copied);
958
959 out_unlock:
960 spin_unlock(&xprt->transport_lock);
961 dropit:
962 skb_free_datagram(sk, skb);
963 out:
964 read_unlock_bh(&sk->sk_callback_lock);
965}
966
9903cd1c
CL		 967/**
968 * xs_udp_data_ready - "data ready" callback for UDP sockets
969 * @sk: socket with data to read
9903cd1c 970 *
a246b010 971 */
676d2369 972static void xs_udp_data_ready(struct sock *sk)
a246b010 973{
9903cd1c
CL		 974 struct rpc_task *task;
975 struct rpc_xprt *xprt;
a246b010 976 struct rpc_rqst *rovr;
9903cd1c 977 struct sk_buff *skb;
a246b010 978 int err, repsize, copied;
d8ed029d
AD		 979 u32 _xid;
980 __be32 *xp;
a246b010 981
f064af1e 982 read_lock_bh(&sk->sk_callback_lock);
46121cf7 983 dprintk("RPC: xs_udp_data_ready...\n");
9903cd1c 984 if (!(xprt = xprt_from_sock(sk)))
a246b010 985 goto out;
a246b010
CL		 986
987 if ((skb = skb_recv_datagram(sk, 0, 1, &err)) == NULL)
988 goto out;
989
a246b010
CL		 990 repsize = skb->len - sizeof(struct udphdr);
991 if (repsize < 4) {
46121cf7 992 dprintk("RPC: impossible RPC reply size %d!\n", repsize);
a246b010
CL		 993 goto dropit;
994 }
995
996 /* Copy the XID from the skb... */
997 xp = skb_header_pointer(skb, sizeof(struct udphdr),
998 sizeof(_xid), &_xid);
999 if (xp == NULL)
1000 goto dropit;
1001
1002 /* Look up and lock the request corresponding to the given XID */
4a0f8c04 1003 spin_lock(&xprt->transport_lock);
a246b010
CL		 1004 rovr = xprt_lookup_rqst(xprt, *xp);
1005 if (!rovr)
1006 goto out_unlock;
1007 task = rovr->rq_task;
1008
a246b010
CL		 1009 if ((copied = rovr->rq_private_buf.buflen) > repsize)
1010 copied = repsize;
1011
1012 /* Suck it into the iovec, verify checksum if not done by hw. */
1781f7f5
HX		 1013 if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
1014 UDPX_INC_STATS_BH(sk, UDP_MIB_INERRORS);
a246b010 1015 goto out_unlock;
1781f7f5
HX		 1016 }
1017
1018 UDPX_INC_STATS_BH(sk, UDP_MIB_INDATAGRAMS);
a246b010 1019
6a24dfb6 1020 xprt_adjust_cwnd(xprt, task, copied);
1570c1e4 1021 xprt_complete_rqst(task, copied);
a246b010
CL		 1022
1023 out_unlock:
4a0f8c04 1024 spin_unlock(&xprt->transport_lock);
a246b010
CL		 1025 dropit:
1026 skb_free_datagram(sk, skb);
1027 out:
f064af1e 1028 read_unlock_bh(&sk->sk_callback_lock);
a246b010
CL		 1029}
1030
a519fc7a
TM		 1031/*
1032 * Helper function to force a TCP close if the server is sending
1033 * junk and/or it has put us in CLOSE_WAIT
1034 */
1035static void xs_tcp_force_close(struct rpc_xprt *xprt)
1036{
a519fc7a
TM		 1037 xprt_force_disconnect(xprt);
1038}
1039
dd456471 1040static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
a246b010 1041{
51971139 1042 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
a246b010
CL		 1043 size_t len, used;
1044 char *p;
1045
51971139
CL		 1046 p = ((char *) &transport->tcp_fraghdr) + transport->tcp_offset;
1047 len = sizeof(transport->tcp_fraghdr) - transport->tcp_offset;
9d292316 1048 used = xdr_skb_read_bits(desc, p, len);
51971139 1049 transport->tcp_offset += used;
a246b010
CL		 1050 if (used != len)
1051 return;
808012fb 1052
51971139
CL		 1053 transport->tcp_reclen = ntohl(transport->tcp_fraghdr);
1054 if (transport->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
e136d092 1055 transport->tcp_flags |= TCP_RCV_LAST_FRAG;
a246b010 1056 else
e136d092 1057 transport->tcp_flags &= ~TCP_RCV_LAST_FRAG;
51971139 1058 transport->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
808012fb 1059
e136d092 1060 transport->tcp_flags &= ~TCP_RCV_COPY_FRAGHDR;
51971139 1061 transport->tcp_offset = 0;
808012fb 1062
a246b010 1063 /* Sanity check of the record length */
18dca02a 1064 if (unlikely(transport->tcp_reclen < 8)) {
46121cf7 1065 dprintk("RPC: invalid TCP record fragment length\n");
a519fc7a 1066 xs_tcp_force_close(xprt);
9903cd1c 1067 return;
a246b010 1068 }
46121cf7 1069 dprintk("RPC: reading TCP record fragment of length %d\n",
51971139 1070 transport->tcp_reclen);
a246b010
CL		 1071}
1072
51971139 1073static void xs_tcp_check_fraghdr(struct sock_xprt *transport)
a246b010 1074{
51971139 1075 if (transport->tcp_offset == transport->tcp_reclen) {
e136d092 1076 transport->tcp_flags |= TCP_RCV_COPY_FRAGHDR;
51971139 1077 transport->tcp_offset = 0;
e136d092
CL		 1078 if (transport->tcp_flags & TCP_RCV_LAST_FRAG) {
1079 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1080 transport->tcp_flags |= TCP_RCV_COPY_XID;
51971139 1081 transport->tcp_copied = 0;
a246b010
CL		 1082 }
1083 }
1084}
1085
dd456471 1086static inline void xs_tcp_read_xid(struct sock_xprt *transport, struct xdr_skb_reader *desc)
a246b010
CL		 1087{
1088 size_t len, used;
1089 char *p;
1090
51971139 1091 len = sizeof(transport->tcp_xid) - transport->tcp_offset;
46121cf7 1092 dprintk("RPC: reading XID (%Zu bytes)\n", len);
51971139 1093 p = ((char *) &transport->tcp_xid) + transport->tcp_offset;
9d292316 1094 used = xdr_skb_read_bits(desc, p, len);
51971139 1095 transport->tcp_offset += used;
a246b010
CL		 1096 if (used != len)
1097 return;
e136d092 1098 transport->tcp_flags &= ~TCP_RCV_COPY_XID;
f4a2e418 1099 transport->tcp_flags |= TCP_RCV_READ_CALLDIR;
51971139 1100 transport->tcp_copied = 4;
18dca02a
RL		 1101 dprintk("RPC: reading %s XID %08x\n",
1102 (transport->tcp_flags & TCP_RPC_REPLY) ? "reply for"
1103 : "request with",
51971139
CL		 1104 ntohl(transport->tcp_xid));
1105 xs_tcp_check_fraghdr(transport);
a246b010
CL		 1106}
1107
18dca02a
RL		 1108static inline void xs_tcp_read_calldir(struct sock_xprt *transport,
1109 struct xdr_skb_reader *desc)
a246b010 1110{
18dca02a
RL		 1111 size_t len, used;
1112 u32 offset;
b76ce561 1113 char *p;
18dca02a
RL		 1114
1115 /*
1116 * We want transport->tcp_offset to be 8 at the end of this routine
1117 * (4 bytes for the xid and 4 bytes for the call/reply flag).
1118 * When this function is called for the first time,
1119 * transport->tcp_offset is 4 (after having already read the xid).
1120 */
1121 offset = transport->tcp_offset - sizeof(transport->tcp_xid);
b76ce561 1122 len = sizeof(transport->tcp_calldir) - offset;
18dca02a 1123 dprintk("RPC: reading CALL/REPLY flag (%Zu bytes)\n", len);
b76ce561
TM		 1124 p = ((char *) &transport->tcp_calldir) + offset;
1125 used = xdr_skb_read_bits(desc, p, len);
18dca02a
RL		 1126 transport->tcp_offset += used;
1127 if (used != len)
1128 return;
f4a2e418 1129 transport->tcp_flags &= ~TCP_RCV_READ_CALLDIR;
f4a2e418
RL		 1130 /*
1131 * We don't yet have the XDR buffer, so we will write the calldir
1132 * out after we get the buffer from the 'struct rpc_rqst'
1133 */
b76ce561
TM		 1134 switch (ntohl(transport->tcp_calldir)) {
1135 case RPC_REPLY:
1136 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1137 transport->tcp_flags |= TCP_RCV_COPY_DATA;
18dca02a 1138 transport->tcp_flags |= TCP_RPC_REPLY;
b76ce561
TM		 1139 break;
1140 case RPC_CALL:
1141 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1142 transport->tcp_flags |= TCP_RCV_COPY_DATA;
18dca02a 1143 transport->tcp_flags &= ~TCP_RPC_REPLY;
b76ce561
TM		 1144 break;
1145 default:
1146 dprintk("RPC: invalid request message type\n");
a519fc7a 1147 xs_tcp_force_close(&transport->xprt);
b76ce561 1148 }
18dca02a
RL		 1149 xs_tcp_check_fraghdr(transport);
1150}
1151
44b98efd
RL		 1152static inline void xs_tcp_read_common(struct rpc_xprt *xprt,
1153 struct xdr_skb_reader *desc,
1154 struct rpc_rqst *req)
a246b010 1155{
44b98efd
RL		 1156 struct sock_xprt *transport =
1157 container_of(xprt, struct sock_xprt, xprt);
a246b010
CL		 1158 struct xdr_buf *rcvbuf;
1159 size_t len;
1160 ssize_t r;
1161
a246b010 1162 rcvbuf = &req->rq_private_buf;
f4a2e418
RL		 1163
1164 if (transport->tcp_flags & TCP_RCV_COPY_CALLDIR) {
1165 /*
1166 * Save the RPC direction in the XDR buffer
1167 */
f4a2e418 1168 memcpy(rcvbuf->head[0].iov_base + transport->tcp_copied,
b76ce561
TM		 1169 &transport->tcp_calldir,
1170 sizeof(transport->tcp_calldir));
1171 transport->tcp_copied += sizeof(transport->tcp_calldir);
f4a2e418 1172 transport->tcp_flags &= ~TCP_RCV_COPY_CALLDIR;
a246b010
CL		 1173 }
1174
a246b010 1175 len = desc->count;
51971139 1176 if (len > transport->tcp_reclen - transport->tcp_offset) {
dd456471 1177 struct xdr_skb_reader my_desc;
a246b010 1178
51971139 1179 len = transport->tcp_reclen - transport->tcp_offset;
a246b010
CL		 1180 memcpy(&my_desc, desc, sizeof(my_desc));
1181 my_desc.count = len;
51971139 1182 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
9d292316 1183 &my_desc, xdr_skb_read_bits);
a246b010
CL		 1184 desc->count -= r;
1185 desc->offset += r;
1186 } else
51971139 1187 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
9d292316 1188 desc, xdr_skb_read_bits);
a246b010
CL		 1189
1190 if (r > 0) {
51971139
CL		 1191 transport->tcp_copied += r;
1192 transport->tcp_offset += r;
a246b010
CL		 1193 }
1194 if (r != len) {
1195 /* Error when copying to the receive buffer,
1196 * usually because we weren't able to allocate
1197 * additional buffer pages. All we can do now
e136d092 1198 * is turn off TCP_RCV_COPY_DATA, so the request
a246b010
CL		 1199 * will not receive any additional updates,
1200 * and time out.
1201 * Any remaining data from this record will
1202 * be discarded.
1203 */
e136d092 1204 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
46121cf7 1205 dprintk("RPC: XID %08x truncated request\n",
51971139 1206 ntohl(transport->tcp_xid));
46121cf7
CL		 1207 dprintk("RPC: xprt = %p, tcp_copied = %lu, "
1208 "tcp_offset = %u, tcp_reclen = %u\n",
1209 xprt, transport->tcp_copied,
1210 transport->tcp_offset, transport->tcp_reclen);
44b98efd 1211 return;
a246b010
CL		 1212 }
1213
46121cf7 1214 dprintk("RPC: XID %08x read %Zd bytes\n",
51971139 1215 ntohl(transport->tcp_xid), r);
46121cf7
CL		 1216 dprintk("RPC: xprt = %p, tcp_copied = %lu, tcp_offset = %u, "
1217 "tcp_reclen = %u\n", xprt, transport->tcp_copied,
1218 transport->tcp_offset, transport->tcp_reclen);
51971139
CL		 1219
1220 if (transport->tcp_copied == req->rq_private_buf.buflen)
e136d092 1221 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
51971139 1222 else if (transport->tcp_offset == transport->tcp_reclen) {
e136d092
CL		 1223 if (transport->tcp_flags & TCP_RCV_LAST_FRAG)
1224 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
a246b010 1225 }
44b98efd
RL		 1226}
1227
1228/*
1229 * Finds the request corresponding to the RPC xid and invokes the common
1230 * tcp read code to read the data.
1231 */
1232static inline int xs_tcp_read_reply(struct rpc_xprt *xprt,
1233 struct xdr_skb_reader *desc)
1234{
1235 struct sock_xprt *transport =
1236 container_of(xprt, struct sock_xprt, xprt);
1237 struct rpc_rqst *req;
1238
1239 dprintk("RPC: read reply XID %08x\n", ntohl(transport->tcp_xid));
1240
1241 /* Find and lock the request corresponding to this xid */
1242 spin_lock(&xprt->transport_lock);
1243 req = xprt_lookup_rqst(xprt, transport->tcp_xid);
1244 if (!req) {
1245 dprintk("RPC: XID %08x request not found!\n",
1246 ntohl(transport->tcp_xid));
1247 spin_unlock(&xprt->transport_lock);
1248 return -1;
1249 }
1250
1251 xs_tcp_read_common(xprt, desc, req);
1252
e136d092 1253 if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
51971139 1254 xprt_complete_rqst(req->rq_task, transport->tcp_copied);
44b98efd 1255
4a0f8c04 1256 spin_unlock(&xprt->transport_lock);
44b98efd
RL		 1257 return 0;
1258}
1259
9e00abc3 1260#if defined(CONFIG_SUNRPC_BACKCHANNEL)
44b98efd
RL		 1261/*
1262 * Obtains an rpc_rqst previously allocated and invokes the common
1263 * tcp read code to read the data. The result is placed in the callback
1264 * queue.
1265 * If we're unable to obtain the rpc_rqst we schedule the closing of the
1266 * connection and return -1.
1267 */
2ea24497 1268static int xs_tcp_read_callback(struct rpc_xprt *xprt,
44b98efd
RL		 1269 struct xdr_skb_reader *desc)
1270{
1271 struct sock_xprt *transport =
1272 container_of(xprt, struct sock_xprt, xprt);
1273 struct rpc_rqst *req;
1274
2ea24497
TM		 1275 /* Look up and lock the request corresponding to the given XID */
1276 spin_lock(&xprt->transport_lock);
1277 req = xprt_lookup_bc_request(xprt, transport->tcp_xid);
44b98efd 1278 if (req == NULL) {
2ea24497 1279 spin_unlock(&xprt->transport_lock);
44b98efd
RL		 1280 printk(KERN_WARNING "Callback slot table overflowed\n");
1281 xprt_force_disconnect(xprt);
1282 return -1;
1283 }
1284
44b98efd
RL		 1285 dprintk("RPC: read callback XID %08x\n", ntohl(req->rq_xid));
1286 xs_tcp_read_common(xprt, desc, req);
1287
2ea24497
TM		 1288 if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1289 xprt_complete_bc_request(req, transport->tcp_copied);
1290 spin_unlock(&xprt->transport_lock);
44b98efd
RL		 1291
1292 return 0;
1293}
1294
1295static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1296 struct xdr_skb_reader *desc)
1297{
1298 struct sock_xprt *transport =
1299 container_of(xprt, struct sock_xprt, xprt);
1300
1301 return (transport->tcp_flags & TCP_RPC_REPLY) ?
1302 xs_tcp_read_reply(xprt, desc) :
1303 xs_tcp_read_callback(xprt, desc);
1304}
1305#else
1306static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1307 struct xdr_skb_reader *desc)
1308{
1309 return xs_tcp_read_reply(xprt, desc);
1310}
9e00abc3 1311#endif /* CONFIG_SUNRPC_BACKCHANNEL */
44b98efd
RL		 1312
1313/*
1314 * Read data off the transport. This can be either an RPC_CALL or an
1315 * RPC_REPLY. Relay the processing to helper functions.
1316 */
1317static void xs_tcp_read_data(struct rpc_xprt *xprt,
1318 struct xdr_skb_reader *desc)
1319{
1320 struct sock_xprt *transport =
1321 container_of(xprt, struct sock_xprt, xprt);
1322
1323 if (_xs_tcp_read_data(xprt, desc) == 0)
1324 xs_tcp_check_fraghdr(transport);
1325 else {
1326 /*
1327 * The transport_lock protects the request handling.
1328 * There's no need to hold it to update the tcp_flags.
1329 */
1330 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1331 }
a246b010
CL		 1332}
1333
dd456471 1334static inline void xs_tcp_read_discard(struct sock_xprt *transport, struct xdr_skb_reader *desc)
a246b010
CL		 1335{
1336 size_t len;
1337
51971139 1338 len = transport->tcp_reclen - transport->tcp_offset;
a246b010
CL		 1339 if (len > desc->count)
1340 len = desc->count;
1341 desc->count -= len;
1342 desc->offset += len;
51971139 1343 transport->tcp_offset += len;
46121cf7 1344 dprintk("RPC: discarded %Zu bytes\n", len);
51971139 1345 xs_tcp_check_fraghdr(transport);
a246b010
CL		 1346}
1347
9903cd1c 1348static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
a246b010
CL		 1349{
1350 struct rpc_xprt *xprt = rd_desc->arg.data;
51971139 1351 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
dd456471 1352 struct xdr_skb_reader desc = {
a246b010
CL		 1353 .skb = skb,
1354 .offset = offset,
1355 .count = len,
9903cd1c 1356 };
a246b010 1357
46121cf7 1358 dprintk("RPC: xs_tcp_data_recv started\n");
a246b010 1359 do {
1a867a08 1360 trace_xs_tcp_data_recv(transport);
a246b010
CL		 1361 /* Read in a new fragment marker if necessary */
1362 /* Can we ever really expect to get completely empty fragments? */
e136d092 1363 if (transport->tcp_flags & TCP_RCV_COPY_FRAGHDR) {
9903cd1c 1364 xs_tcp_read_fraghdr(xprt, &desc);
a246b010
CL		 1365 continue;
1366 }
1367 /* Read in the xid if necessary */
e136d092 1368 if (transport->tcp_flags & TCP_RCV_COPY_XID) {
51971139 1369 xs_tcp_read_xid(transport, &desc);
a246b010
CL		 1370 continue;
1371 }
18dca02a 1372 /* Read in the call/reply flag */
f4a2e418 1373 if (transport->tcp_flags & TCP_RCV_READ_CALLDIR) {
18dca02a
RL		 1374 xs_tcp_read_calldir(transport, &desc);
1375 continue;
1376 }
a246b010 1377 /* Read in the request data */
e136d092 1378 if (transport->tcp_flags & TCP_RCV_COPY_DATA) {
44b98efd 1379 xs_tcp_read_data(xprt, &desc);
a246b010
CL		 1380 continue;
1381 }
1382 /* Skip over any trailing bytes on short reads */
51971139 1383 xs_tcp_read_discard(transport, &desc);
a246b010 1384 } while (desc.count);
1a867a08 1385 trace_xs_tcp_data_recv(transport);
46121cf7 1386 dprintk("RPC: xs_tcp_data_recv done\n");
a246b010
CL		 1387 return len - desc.count;
1388}
1389
9903cd1c
CL		 1390/**
1391 * xs_tcp_data_ready - "data ready" callback for TCP sockets
1392 * @sk: socket with data to read
9903cd1c
CL		 1393 *
1394 */
676d2369 1395static void xs_tcp_data_ready(struct sock *sk)
a246b010
CL		 1396{
1397 struct rpc_xprt *xprt;
1398 read_descriptor_t rd_desc;
ff2d7db8 1399 int read;
3705ad64 1400 unsigned long total = 0;
a246b010 1401
46121cf7
CL		 1402 dprintk("RPC: xs_tcp_data_ready...\n");
1403
f064af1e 1404 read_lock_bh(&sk->sk_callback_lock);
3705ad64
JL		 1405 if (!(xprt = xprt_from_sock(sk))) {
1406 read = 0;
a246b010 1407 goto out;
3705ad64 1408 }
61d0a8e6
NB		 1409 /* Any data means we had a useful conversation, so
1410 * the we don't need to delay the next reconnect
1411 */
1412 if (xprt->reestablish_timeout)
1413 xprt->reestablish_timeout = 0;
1414
9903cd1c 1415 /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
a246b010 1416 rd_desc.arg.data = xprt;
ff2d7db8
TM		 1417 do {
1418 rd_desc.count = 65536;
1419 read = tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
3705ad64
JL		 1420 if (read > 0)
1421 total += read;
ff2d7db8 1422 } while (read > 0);
a246b010 1423out:
3705ad64 1424 trace_xs_tcp_data_ready(xprt, read, total);
f064af1e 1425 read_unlock_bh(&sk->sk_callback_lock);
a246b010
CL		 1426}
1427
9903cd1c
CL		 1428/**
1429 * xs_tcp_state_change - callback to handle TCP socket state changes
1430 * @sk: socket whose state has changed
1431 *
1432 */
1433static void xs_tcp_state_change(struct sock *sk)
a246b010 1434{
9903cd1c 1435 struct rpc_xprt *xprt;
a246b010 1436
f064af1e 1437 read_lock_bh(&sk->sk_callback_lock);
a246b010
CL		 1438 if (!(xprt = xprt_from_sock(sk)))
1439 goto out;
46121cf7 1440 dprintk("RPC: xs_tcp_state_change client %p...\n", xprt);
669502ff 1441 dprintk("RPC: state %x conn %d dead %d zapped %d sk_shutdown %d\n",
46121cf7
CL		 1442 sk->sk_state, xprt_connected(xprt),
1443 sock_flag(sk, SOCK_DEAD),
669502ff
AC		 1444 sock_flag(sk, SOCK_ZAPPED),
1445 sk->sk_shutdown);
a246b010 1446
40b5ea0c 1447 trace_rpc_socket_state_change(xprt, sk->sk_socket);
a246b010
CL		 1448 switch (sk->sk_state) {
1449 case TCP_ESTABLISHED:
f064af1e 1450 spin_lock(&xprt->transport_lock);
a246b010 1451 if (!xprt_test_and_set_connected(xprt)) {
51971139
CL		 1452 struct sock_xprt *transport = container_of(xprt,
1453 struct sock_xprt, xprt);
1454
a246b010 1455 /* Reset TCP record info */
51971139
CL		 1456 transport->tcp_offset = 0;
1457 transport->tcp_reclen = 0;
1458 transport->tcp_copied = 0;
e136d092
CL		 1459 transport->tcp_flags =
1460 TCP_RCV_COPY_FRAGHDR | TCP_RCV_COPY_XID;
8b71798c 1461 xprt->connect_cookie++;
51971139 1462
2a491991 1463 xprt_wake_pending_tasks(xprt, -EAGAIN);
a246b010 1464 }
f064af1e 1465 spin_unlock(&xprt->transport_lock);
a246b010 1466 break;
3b948ae5
TM		 1467 case TCP_FIN_WAIT1:
1468 /* The client initiated a shutdown of the socket */
7c1d71cf 1469 xprt->connect_cookie++;
663b8858 1470 xprt->reestablish_timeout = 0;
3b948ae5 1471 set_bit(XPRT_CLOSING, &xprt->state);
4e857c58 1472 smp_mb__before_atomic();
3b948ae5 1473 clear_bit(XPRT_CONNECTED, &xprt->state);
ef803670 1474 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
4e857c58 1475 smp_mb__after_atomic();
a246b010 1476 break;
632e3bdc 1477 case TCP_CLOSE_WAIT:
3b948ae5 1478 /* The server initiated a shutdown of the socket */
7c1d71cf 1479 xprt->connect_cookie++;
d0bea455 1480 clear_bit(XPRT_CONNECTED, &xprt->state);
a519fc7a 1481 xs_tcp_force_close(xprt);
663b8858
TM		 1482 case TCP_CLOSING:
1483 /*
1484 * If the server closed down the connection, make sure that
1485 * we back off before reconnecting
1486 */
1487 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1488 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
3b948ae5
TM		 1489 break;
1490 case TCP_LAST_ACK:
670f9457 1491 set_bit(XPRT_CLOSING, &xprt->state);
4e857c58 1492 smp_mb__before_atomic();
3b948ae5 1493 clear_bit(XPRT_CONNECTED, &xprt->state);
4e857c58 1494 smp_mb__after_atomic();
3b948ae5
TM		 1495 break;
1496 case TCP_CLOSE:
7d1e8255 1497 xs_sock_mark_closed(xprt);
a246b010
CL		 1498 }
1499 out:
f064af1e 1500 read_unlock_bh(&sk->sk_callback_lock);
a246b010
CL		 1501}
1502
1f0fa154
IJ		 1503static void xs_write_space(struct sock *sk)
1504{
1505 struct socket *sock;
1506 struct rpc_xprt *xprt;
1507
1508 if (unlikely(!(sock = sk->sk_socket)))
1509 return;
1510 clear_bit(SOCK_NOSPACE, &sock->flags);
1511
1512 if (unlikely(!(xprt = xprt_from_sock(sk))))
1513 return;
1514 if (test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags) == 0)
1515 return;
1516
1517 xprt_write_space(xprt);
1518}
1519
9903cd1c 1520/**
c7b2cae8
CL		 1521 * xs_udp_write_space - callback invoked when socket buffer space
1522 * becomes available
9903cd1c
CL		 1523 * @sk: socket whose state has changed
1524 *
a246b010
CL		 1525 * Called when more output buffer space is available for this socket.
1526 * We try not to wake our writers until they can make "significant"
c7b2cae8 1527 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
a246b010
CL		 1528 * with a bunch of small requests.
1529 */
c7b2cae8 1530static void xs_udp_write_space(struct sock *sk)
a246b010 1531{
f064af1e 1532 read_lock_bh(&sk->sk_callback_lock);
a246b010 1533
c7b2cae8 1534 /* from net/core/sock.c:sock_def_write_space */
1f0fa154
IJ		 1535 if (sock_writeable(sk))
1536 xs_write_space(sk);
a246b010 1537
f064af1e 1538 read_unlock_bh(&sk->sk_callback_lock);
c7b2cae8 1539}
a246b010 1540
c7b2cae8
CL		 1541/**
1542 * xs_tcp_write_space - callback invoked when socket buffer space
1543 * becomes available
1544 * @sk: socket whose state has changed
1545 *
1546 * Called when more output buffer space is available for this socket.
1547 * We try not to wake our writers until they can make "significant"
1548 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1549 * with a bunch of small requests.
1550 */
1551static void xs_tcp_write_space(struct sock *sk)
1552{
f064af1e 1553 read_lock_bh(&sk->sk_callback_lock);
c7b2cae8
CL		 1554
1555 /* from net/core/stream.c:sk_stream_write_space */
64dc6130 1556 if (sk_stream_is_writeable(sk))
1f0fa154 1557 xs_write_space(sk);
b6ddf64f 1558
f064af1e 1559 read_unlock_bh(&sk->sk_callback_lock);
a246b010
CL		 1560}
1561
470056c2 1562static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
a246b010 1563{
ee0ac0c2
CL		 1564 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1565 struct sock *sk = transport->inet;
a246b010 1566
7c6e066e 1567 if (transport->rcvsize) {
a246b010 1568 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
7c6e066e 1569 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
a246b010 1570 }
7c6e066e 1571 if (transport->sndsize) {
a246b010 1572 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
7c6e066e 1573 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
a246b010
CL		 1574 sk->sk_write_space(sk);
1575 }
1576}
1577
43118c29 1578/**
470056c2 1579 * xs_udp_set_buffer_size - set send and receive limits
43118c29 1580 * @xprt: generic transport
470056c2
CL		 1581 * @sndsize: requested size of send buffer, in bytes
1582 * @rcvsize: requested size of receive buffer, in bytes
43118c29 1583 *
470056c2 1584 * Set socket send and receive buffer size limits.
43118c29 1585 */
470056c2 1586static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
43118c29 1587{
7c6e066e
CL		 1588 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1589
1590 transport->sndsize = 0;
470056c2 1591 if (sndsize)
7c6e066e
CL		 1592 transport->sndsize = sndsize + 1024;
1593 transport->rcvsize = 0;
470056c2 1594 if (rcvsize)
7c6e066e 1595 transport->rcvsize = rcvsize + 1024;
470056c2
CL		 1596
1597 xs_udp_do_set_buffer_size(xprt);
43118c29
CL		 1598}
1599
46c0ee8b
CL		 1600/**
1601 * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1602 * @task: task that timed out
1603 *
1604 * Adjust the congestion window after a retransmit timeout has occurred.
1605 */
6a24dfb6 1606static void xs_udp_timer(struct rpc_xprt *xprt, struct rpc_task *task)
46c0ee8b 1607{
6a24dfb6 1608 xprt_adjust_cwnd(xprt, task, -ETIMEDOUT);
46c0ee8b
CL		 1609}
1610
b85d8806
CL		 1611static unsigned short xs_get_random_port(void)
1612{
1613 unsigned short range = xprt_max_resvport - xprt_min_resvport;
63862b5b 1614 unsigned short rand = (unsigned short) prandom_u32() % range;
b85d8806
CL		 1615 return rand + xprt_min_resvport;
1616}
1617
4dda9c8a
TM		 1618/**
1619 * xs_set_reuseaddr_port - set the socket's port and address reuse options
1620 * @sock: socket
1621 *
1622 * Note that this function has to be called on all sockets that share the
1623 * same port, and it must be called before binding.
1624 */
1625static void xs_sock_set_reuseport(struct socket *sock)
1626{
402e23b4 1627 int opt = 1;
4dda9c8a 1628
402e23b4
TM		 1629 kernel_setsockopt(sock, SOL_SOCKET, SO_REUSEPORT,
1630 (char *)&opt, sizeof(opt));
4dda9c8a
TM		 1631}
1632
1633static unsigned short xs_sock_getport(struct socket *sock)
1634{
1635 struct sockaddr_storage buf;
1636 int buflen;
1637 unsigned short port = 0;
1638
1639 if (kernel_getsockname(sock, (struct sockaddr *)&buf, &buflen) < 0)
1640 goto out;
1641 switch (buf.ss_family) {
1642 case AF_INET6:
1643 port = ntohs(((struct sockaddr_in6 *)&buf)->sin6_port);
1644 break;
1645 case AF_INET:
1646 port = ntohs(((struct sockaddr_in *)&buf)->sin_port);
1647 }
1648out:
1649 return port;
1650}
1651
92200412
CL		 1652/**
1653 * xs_set_port - reset the port number in the remote endpoint address
1654 * @xprt: generic transport
1655 * @port: new port number
1656 *
1657 */
1658static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1659{
46121cf7 1660 dprintk("RPC: setting port for xprt %p to %u\n", xprt, port);
c4efcb1d 1661
9dc3b095
CL		 1662 rpc_set_port(xs_addr(xprt), port);
1663 xs_update_peer_port(xprt);
92200412
CL		 1664}
1665
4dda9c8a
TM		 1666static void xs_set_srcport(struct sock_xprt *transport, struct socket *sock)
1667{
1668 if (transport->srcport == 0)
1669 transport->srcport = xs_sock_getport(sock);
1670}
1671
5d4ec932 1672static unsigned short xs_get_srcport(struct sock_xprt *transport)
67a391d7 1673{
fbfffbd5 1674 unsigned short port = transport->srcport;
67a391d7
TM		 1675
1676 if (port == 0 && transport->xprt.resvport)
1677 port = xs_get_random_port();
1678 return port;
1679}
1680
baaf4e48 1681static unsigned short xs_next_srcport(struct sock_xprt *transport, unsigned short port)
67a391d7 1682{
fbfffbd5
CL		 1683 if (transport->srcport != 0)
1684 transport->srcport = 0;
67a391d7
TM		 1685 if (!transport->xprt.resvport)
1686 return 0;
1687 if (port <= xprt_min_resvport || port > xprt_max_resvport)
1688 return xprt_max_resvport;
1689 return --port;
1690}
beb59b68 1691static int xs_bind(struct sock_xprt *transport, struct socket *sock)
a246b010 1692{
beb59b68 1693 struct sockaddr_storage myaddr;
67a391d7 1694 int err, nloop = 0;
5d4ec932 1695 unsigned short port = xs_get_srcport(transport);
67a391d7 1696 unsigned short last;
a246b010 1697
0f7a622c
CP		 1698 /*
1699 * If we are asking for any ephemeral port (i.e. port == 0 &&
1700 * transport->xprt.resvport == 0), don't bind. Let the local
1701 * port selection happen implicitly when the socket is used
1702 * (for example at connect time).
1703 *
1704 * This ensures that we can continue to establish TCP
1705 * connections even when all local ephemeral ports are already
1706 * a part of some TCP connection. This makes no difference
1707 * for UDP sockets, but also doens't harm them.
1708 *
1709 * If we're asking for any reserved port (i.e. port == 0 &&
1710 * transport->xprt.resvport == 1) xs_get_srcport above will
1711 * ensure that port is non-zero and we will bind as needed.
1712 */
1713 if (port == 0)
1714 return 0;
1715
beb59b68 1716 memcpy(&myaddr, &transport->srcaddr, transport->xprt.addrlen);
a246b010 1717 do {
beb59b68
PE		 1718 rpc_set_port((struct sockaddr *)&myaddr, port);
1719 err = kernel_bind(sock, (struct sockaddr *)&myaddr,
1720 transport->xprt.addrlen);
a246b010 1721 if (err == 0) {
fbfffbd5 1722 transport->srcport = port;
d3bc9a1d 1723 break;
a246b010 1724 }
67a391d7 1725 last = port;
baaf4e48 1726 port = xs_next_srcport(transport, port);
67a391d7
TM		 1727 if (port > last)
1728 nloop++;
1729 } while (err == -EADDRINUSE && nloop != 2);
90058d37 1730
4232e863 1731 if (myaddr.ss_family == AF_INET)
beb59b68
PE		 1732 dprintk("RPC: %s %pI4:%u: %s (%d)\n", __func__,
1733 &((struct sockaddr_in *)&myaddr)->sin_addr,
1734 port, err ? "failed" : "ok", err);
1735 else
1736 dprintk("RPC: %s %pI6:%u: %s (%d)\n", __func__,
1737 &((struct sockaddr_in6 *)&myaddr)->sin6_addr,
1738 port, err ? "failed" : "ok", err);
a246b010
CL		 1739 return err;
1740}
1741
176e21ee
CL		 1742/*
1743 * We don't support autobind on AF_LOCAL sockets
1744 */
1745static void xs_local_rpcbind(struct rpc_task *task)
1746{
3dc0da27
TM		 1747 rcu_read_lock();
1748 xprt_set_bound(rcu_dereference(task->tk_client->cl_xprt));
1749 rcu_read_unlock();
176e21ee
CL		 1750}
1751
1752static void xs_local_set_port(struct rpc_xprt *xprt, unsigned short port)
1753{
1754}
beb59b68 1755
ed07536e
PZ		 1756#ifdef CONFIG_DEBUG_LOCK_ALLOC
1757static struct lock_class_key xs_key[2];
1758static struct lock_class_key xs_slock_key[2];
1759
176e21ee
CL		 1760static inline void xs_reclassify_socketu(struct socket *sock)
1761{
1762 struct sock *sk = sock->sk;
1763
176e21ee
CL		 1764 sock_lock_init_class_and_name(sk, "slock-AF_LOCAL-RPC",
1765 &xs_slock_key[1], "sk_lock-AF_LOCAL-RPC", &xs_key[1]);
1766}
1767
8945ee5e 1768static inline void xs_reclassify_socket4(struct socket *sock)
ed07536e
PZ		 1769{
1770 struct sock *sk = sock->sk;
8945ee5e 1771
8945ee5e
CL		 1772 sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1773 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1774}
ed07536e 1775
8945ee5e
CL		 1776static inline void xs_reclassify_socket6(struct socket *sock)
1777{
1778 struct sock *sk = sock->sk;
ed07536e 1779
8945ee5e
CL		 1780 sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1781 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
ed07536e 1782}
6bc9638a
PE		 1783
1784static inline void xs_reclassify_socket(int family, struct socket *sock)
1785{
1b7a1819
WAA		 1786 WARN_ON_ONCE(sock_owned_by_user(sock->sk));
1787 if (sock_owned_by_user(sock->sk))
1788 return;
1789
4232e863 1790 switch (family) {
176e21ee
CL		 1791 case AF_LOCAL:
1792 xs_reclassify_socketu(sock);
1793 break;
4232e863 1794 case AF_INET:
6bc9638a 1795 xs_reclassify_socket4(sock);
4232e863
CL		 1796 break;
1797 case AF_INET6:
6bc9638a 1798 xs_reclassify_socket6(sock);
4232e863
CL		 1799 break;
1800 }
6bc9638a 1801}
ed07536e 1802#else
176e21ee
CL		 1803static inline void xs_reclassify_socketu(struct socket *sock)
1804{
1805}
1806
8945ee5e
CL		 1807static inline void xs_reclassify_socket4(struct socket *sock)
1808{
1809}
1810
1811static inline void xs_reclassify_socket6(struct socket *sock)
ed07536e
PZ		 1812{
1813}
6bc9638a
PE		 1814
1815static inline void xs_reclassify_socket(int family, struct socket *sock)
1816{
1817}
ed07536e
PZ		 1818#endif
1819
93dc41bd
N		 1820static void xs_dummy_setup_socket(struct work_struct *work)
1821{
1822}
1823
6bc9638a 1824static struct socket *xs_create_sock(struct rpc_xprt *xprt,
4dda9c8a
TM		 1825 struct sock_xprt *transport, int family, int type,
1826 int protocol, bool reuseport)
22f79326
PE		 1827{
1828 struct socket *sock;
1829 int err;
1830
6bc9638a 1831 err = __sock_create(xprt->xprt_net, family, type, protocol, &sock, 1);
22f79326
PE		 1832 if (err < 0) {
1833 dprintk("RPC: can't create %d transport socket (%d).\n",
1834 protocol, -err);
1835 goto out;
1836 }
6bc9638a 1837 xs_reclassify_socket(family, sock);
22f79326 1838
4dda9c8a
TM		 1839 if (reuseport)
1840 xs_sock_set_reuseport(sock);
1841
4cea288a
BH		 1842 err = xs_bind(transport, sock);
1843 if (err) {
22f79326
PE		 1844 sock_release(sock);
1845 goto out;
1846 }
1847
1848 return sock;
1849out:
1850 return ERR_PTR(err);
1851}
1852
176e21ee
CL		 1853static int xs_local_finish_connecting(struct rpc_xprt *xprt,
1854 struct socket *sock)
1855{
1856 struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1857 xprt);
1858
1859 if (!transport->inet) {
1860 struct sock *sk = sock->sk;
1861
1862 write_lock_bh(&sk->sk_callback_lock);
1863
1864 xs_save_old_callbacks(transport, sk);
1865
1866 sk->sk_user_data = xprt;
1867 sk->sk_data_ready = xs_local_data_ready;
1868 sk->sk_write_space = xs_udp_write_space;
2118071d 1869 sk->sk_error_report = xs_error_report;
176e21ee
CL		 1870 sk->sk_allocation = GFP_ATOMIC;
1871
1872 xprt_clear_connected(xprt);
1873
1874 /* Reset to new socket */
1875 transport->sock = sock;
1876 transport->inet = sk;
1877
1878 write_unlock_bh(&sk->sk_callback_lock);
1879 }
1880
1881 /* Tell the socket layer to start connecting... */
1882 xprt->stat.connect_count++;
1883 xprt->stat.connect_start = jiffies;
1884 return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, 0);
1885}
1886
1887/**
1888 * xs_local_setup_socket - create AF_LOCAL socket, connect to a local endpoint
176e21ee 1889 * @transport: socket transport to connect
176e21ee 1890 */
dc107402 1891static int xs_local_setup_socket(struct sock_xprt *transport)
176e21ee 1892{
176e21ee
CL		 1893 struct rpc_xprt *xprt = &transport->xprt;
1894 struct socket *sock;
1895 int status = -EIO;
1896
176e21ee
CL		 1897 status = __sock_create(xprt->xprt_net, AF_LOCAL,
1898 SOCK_STREAM, 0, &sock, 1);
1899 if (status < 0) {
1900 dprintk("RPC: can't create AF_LOCAL "
1901 "transport socket (%d).\n", -status);
1902 goto out;
1903 }
1904 xs_reclassify_socketu(sock);
1905
1906 dprintk("RPC: worker connecting xprt %p via AF_LOCAL to %s\n",
1907 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1908
1909 status = xs_local_finish_connecting(xprt, sock);
40b5ea0c 1910 trace_rpc_socket_connect(xprt, sock, status);
176e21ee
CL		 1911 switch (status) {
1912 case 0:
1913 dprintk("RPC: xprt %p connected to %s\n",
1914 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1915 xprt_set_connected(xprt);
3601c4a9 1916 case -ENOBUFS:
176e21ee
CL		 1917 break;
1918 case -ENOENT:
1919 dprintk("RPC: xprt %p: socket %s does not exist\n",
1920 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1921 break;
4a20a988
TM		 1922 case -ECONNREFUSED:
1923 dprintk("RPC: xprt %p: connection refused for %s\n",
1924 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1925 break;
176e21ee
CL		 1926 default:
1927 printk(KERN_ERR "%s: unhandled error (%d) connecting to %s\n",
1928 __func__, -status,
1929 xprt->address_strings[RPC_DISPLAY_ADDR]);
1930 }
1931
1932out:
1933 xprt_clear_connecting(xprt);
1934 xprt_wake_pending_tasks(xprt, status);
dc107402
BF		 1935 return status;
1936}
1937
b6669737 1938static void xs_local_connect(struct rpc_xprt *xprt, struct rpc_task *task)
dc107402 1939{
dc107402
BF		 1940 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1941 int ret;
1942
1943 if (RPC_IS_ASYNC(task)) {
1944 /*
1945 * We want the AF_LOCAL connect to be resolved in the
1946 * filesystem namespace of the process making the rpc
1947 * call. Thus we connect synchronously.
1948 *
1949 * If we want to support asynchronous AF_LOCAL calls,
1950 * we'll need to figure out how to pass a namespace to
1951 * connect.
1952 */
1953 rpc_exit(task, -ENOTCONN);
1954 return;
1955 }
1956 ret = xs_local_setup_socket(transport);
1957 if (ret && !RPC_IS_SOFTCONN(task))
1958 msleep_interruptible(15000);
176e21ee
CL		 1959}
1960
3c87ef6e 1961#if IS_ENABLED(CONFIG_SUNRPC_SWAP)
d6e971d8
JL		 1962/*
1963 * Note that this should be called with XPRT_LOCKED held (or when we otherwise
1964 * know that we have exclusive access to the socket), to guard against
1965 * races with xs_reset_transport.
1966 */
a564b8f0
MG		 1967static void xs_set_memalloc(struct rpc_xprt *xprt)
1968{
1969 struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1970 xprt);
1971
d6e971d8
JL		 1972 /*
1973 * If there's no sock, then we have nothing to set. The
1974 * reconnecting process will get it for us.
1975 */
1976 if (!transport->inet)
1977 return;
8e228133 1978 if (atomic_read(&xprt->swapper))
a564b8f0
MG		 1979 sk_set_memalloc(transport->inet);
1980}
1981
1982/**
d67fa4d8 1983 * xs_enable_swap - Tag this transport as being used for swap.
a564b8f0 1984 * @xprt: transport to tag
a564b8f0 1985 *
8e228133
JL		 1986 * Take a reference to this transport on behalf of the rpc_clnt, and
1987 * optionally mark it for swapping if it wasn't already.
a564b8f0 1988 */
d67fa4d8
JL		 1989static int
1990xs_enable_swap(struct rpc_xprt *xprt)
a564b8f0 1991{
d6e971d8 1992 struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
a564b8f0 1993
d6e971d8
JL		 1994 if (atomic_inc_return(&xprt->swapper) != 1)
1995 return 0;
1996 if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE))
1997 return -ERESTARTSYS;
1998 if (xs->inet)
1999 sk_set_memalloc(xs->inet);
2000 xprt_release_xprt(xprt, NULL);
8e228133
JL		 2001 return 0;
2002}
a564b8f0 2003
8e228133 2004/**
d67fa4d8 2005 * xs_disable_swap - Untag this transport as being used for swap.
8e228133
JL		 2006 * @xprt: transport to tag
2007 *
2008 * Drop a "swapper" reference to this xprt on behalf of the rpc_clnt. If the
2009 * swapper refcount goes to 0, untag the socket as a memalloc socket.
2010 */
d67fa4d8
JL		 2011static void
2012xs_disable_swap(struct rpc_xprt *xprt)
8e228133 2013{
d6e971d8 2014 struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
8e228133 2015
d6e971d8
JL		 2016 if (!atomic_dec_and_test(&xprt->swapper))
2017 return;
2018 if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE))
2019 return;
2020 if (xs->inet)
2021 sk_clear_memalloc(xs->inet);
2022 xprt_release_xprt(xprt, NULL);
a564b8f0 2023}
a564b8f0
MG		 2024#else
2025static void xs_set_memalloc(struct rpc_xprt *xprt)
2026{
2027}
d67fa4d8
JL		 2028
2029static int
2030xs_enable_swap(struct rpc_xprt *xprt)
2031{
2032 return -EINVAL;
2033}
2034
2035static void
2036xs_disable_swap(struct rpc_xprt *xprt)
2037{
2038}
a564b8f0
MG		 2039#endif
2040
16be2d20
CL		 2041static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2042{
2043 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2044
2045 if (!transport->inet) {
2046 struct sock *sk = sock->sk;
2047
2048 write_lock_bh(&sk->sk_callback_lock);
2049
2a9e1cfa
TM		 2050 xs_save_old_callbacks(transport, sk);
2051
16be2d20 2052 sk->sk_user_data = xprt;
16be2d20
CL		 2053 sk->sk_data_ready = xs_udp_data_ready;
2054 sk->sk_write_space = xs_udp_write_space;
16be2d20
CL		 2055 sk->sk_allocation = GFP_ATOMIC;
2056
2057 xprt_set_connected(xprt);
2058
2059 /* Reset to new socket */
2060 transport->sock = sock;
2061 transport->inet = sk;
2062
a564b8f0
MG		 2063 xs_set_memalloc(xprt);
2064
16be2d20
CL		 2065 write_unlock_bh(&sk->sk_callback_lock);
2066 }
2067 xs_udp_do_set_buffer_size(xprt);
2068}
2069
8c14ff2a 2070static void xs_udp_setup_socket(struct work_struct *work)
a246b010 2071{
34161db6
TM		 2072 struct sock_xprt *transport =
2073 container_of(work, struct sock_xprt, connect_worker.work);
c8475461 2074 struct rpc_xprt *xprt = &transport->xprt;
ee0ac0c2 2075 struct socket *sock = transport->sock;
b65c0310 2076 int status = -EIO;
9903cd1c 2077
8c14ff2a 2078 sock = xs_create_sock(xprt, transport,
4dda9c8a
TM		 2079 xs_addr(xprt)->sa_family, SOCK_DGRAM,
2080 IPPROTO_UDP, false);
b65c0310 2081 if (IS_ERR(sock))
b0d93ad5 2082 goto out;
68e220bd 2083
c740eff8
CL		 2084 dprintk("RPC: worker connecting xprt %p via %s to "
2085 "%s (port %s)\n", xprt,
2086 xprt->address_strings[RPC_DISPLAY_PROTO],
2087 xprt->address_strings[RPC_DISPLAY_ADDR],
2088 xprt->address_strings[RPC_DISPLAY_PORT]);
68e220bd
CL		 2089
2090 xs_udp_finish_connecting(xprt, sock);
40b5ea0c 2091 trace_rpc_socket_connect(xprt, sock, 0);
b0d93ad5
CL		 2092 status = 0;
2093out:
718ba5b8 2094 xprt_unlock_connect(xprt, transport);
b0d93ad5 2095 xprt_clear_connecting(xprt);
7d1e8255 2096 xprt_wake_pending_tasks(xprt, status);
a246b010
CL		 2097}
2098
4876cc77
TM		 2099/**
2100 * xs_tcp_shutdown - gracefully shut down a TCP socket
2101 * @xprt: transport
2102 *
2103 * Initiates a graceful shutdown of the TCP socket by calling the
2104 * equivalent of shutdown(SHUT_RDWR);
2105 */
2106static void xs_tcp_shutdown(struct rpc_xprt *xprt)
2107{
2108 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2109 struct socket *sock = transport->sock;
2110
2111 if (sock == NULL)
2112 return;
2113 if (xprt_connected(xprt)) {
2114 kernel_sock_shutdown(sock, SHUT_RDWR);
2115 trace_rpc_socket_shutdown(xprt, sock);
2116 } else
2117 xs_reset_transport(transport);
2118}
2119
16be2d20 2120static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
a246b010 2121{
16be2d20 2122 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
fe19a96b 2123 int ret = -ENOTCONN;
edb267a6 2124
ee0ac0c2 2125 if (!transport->inet) {
b0d93ad5 2126 struct sock *sk = sock->sk;
7f260e85
TM		 2127 unsigned int keepidle = xprt->timeout->to_initval / HZ;
2128 unsigned int keepcnt = xprt->timeout->to_retries + 1;
2129 unsigned int opt_on = 1;
775f06ab 2130 unsigned int timeo;
7f260e85
TM		 2131
2132 /* TCP Keepalive options */
2133 kernel_setsockopt(sock, SOL_SOCKET, SO_KEEPALIVE,
2134 (char *)&opt_on, sizeof(opt_on));
2135 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPIDLE,
2136 (char *)&keepidle, sizeof(keepidle));
2137 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPINTVL,
2138 (char *)&keepidle, sizeof(keepidle));
2139 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPCNT,
2140 (char *)&keepcnt, sizeof(keepcnt));
b0d93ad5 2141
775f06ab
TM		 2142 /* TCP user timeout (see RFC5482) */
2143 timeo = jiffies_to_msecs(xprt->timeout->to_initval) *
2144 (xprt->timeout->to_retries + 1);
2145 kernel_setsockopt(sock, SOL_TCP, TCP_USER_TIMEOUT,
2146 (char *)&timeo, sizeof(timeo));
2147
b0d93ad5
CL		 2148 write_lock_bh(&sk->sk_callback_lock);
2149
2a9e1cfa
TM		 2150 xs_save_old_callbacks(transport, sk);
2151
b0d93ad5 2152 sk->sk_user_data = xprt;
b0d93ad5
CL		 2153 sk->sk_data_ready = xs_tcp_data_ready;
2154 sk->sk_state_change = xs_tcp_state_change;
2155 sk->sk_write_space = xs_tcp_write_space;
2118071d 2156 sk->sk_error_report = xs_error_report;
b079fa7b 2157 sk->sk_allocation = GFP_ATOMIC;
3167e12c
CL		 2158
2159 /* socket options */
3167e12c 2160 sock_reset_flag(sk, SOCK_LINGER);
3167e12c 2161 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
b0d93ad5
CL		 2162
2163 xprt_clear_connected(xprt);
2164
2165 /* Reset to new socket */
ee0ac0c2
CL		 2166 transport->sock = sock;
2167 transport->inet = sk;
b0d93ad5
CL		 2168
2169 write_unlock_bh(&sk->sk_callback_lock);
2170 }
2171
01d37c42 2172 if (!xprt_bound(xprt))
fe19a96b 2173 goto out;
01d37c42 2174
a564b8f0
MG		 2175 xs_set_memalloc(xprt);
2176
b0d93ad5 2177 /* Tell the socket layer to start connecting... */
262ca07d
CL		 2178 xprt->stat.connect_count++;
2179 xprt->stat.connect_start = jiffies;
fe19a96b
TM		 2180 ret = kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
2181 switch (ret) {
2182 case 0:
4dda9c8a 2183 xs_set_srcport(transport, sock);
fe19a96b
TM		 2184 case -EINPROGRESS:
2185 /* SYN_SENT! */
fe19a96b
TM		 2186 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2187 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2188 }
2189out:
2190 return ret;
16be2d20
CL		 2191}
2192
9903cd1c 2193/**
b61d59ff 2194 * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint
9903cd1c
CL		 2195 *
2196 * Invoked by a work queue tasklet.
a246b010 2197 */
cdd518d5 2198static void xs_tcp_setup_socket(struct work_struct *work)
a246b010 2199{
cdd518d5
PE		 2200 struct sock_xprt *transport =
2201 container_of(work, struct sock_xprt, connect_worker.work);
ee0ac0c2 2202 struct socket *sock = transport->sock;
a9f5f0f7 2203 struct rpc_xprt *xprt = &transport->xprt;
b61d59ff 2204 int status = -EIO;
a246b010 2205
ee0ac0c2 2206 if (!sock) {
cdd518d5 2207 sock = xs_create_sock(xprt, transport,
4dda9c8a
TM		 2208 xs_addr(xprt)->sa_family, SOCK_STREAM,
2209 IPPROTO_TCP, true);
b61d59ff
TM		 2210 if (IS_ERR(sock)) {
2211 status = PTR_ERR(sock);
3167e12c
CL		 2212 goto out;
2213 }
7d1e8255 2214 }
a246b010 2215
c740eff8
CL		 2216 dprintk("RPC: worker connecting xprt %p via %s to "
2217 "%s (port %s)\n", xprt,
2218 xprt->address_strings[RPC_DISPLAY_PROTO],
2219 xprt->address_strings[RPC_DISPLAY_ADDR],
2220 xprt->address_strings[RPC_DISPLAY_PORT]);
edb267a6 2221
16be2d20 2222 status = xs_tcp_finish_connecting(xprt, sock);
40b5ea0c 2223 trace_rpc_socket_connect(xprt, sock, status);
46121cf7
CL		 2224 dprintk("RPC: %p connect status %d connected %d sock state %d\n",
2225 xprt, -status, xprt_connected(xprt),
2226 sock->sk->sk_state);
2a491991 2227 switch (status) {
f75e6745
TM		 2228 default:
2229 printk("%s: connect returned unhandled error %d\n",
2230 __func__, status);
2231 case -EADDRNOTAVAIL:
2232 /* We're probably in TIME_WAIT. Get rid of existing socket,
2233 * and retry
2234 */
a519fc7a 2235 xs_tcp_force_close(xprt);
88b5ed73 2236 break;
2a491991
TM		 2237 case 0:
2238 case -EINPROGRESS:
2239 case -EALREADY:
718ba5b8 2240 xprt_unlock_connect(xprt, transport);
7d1e8255
TM		 2241 xprt_clear_connecting(xprt);
2242 return;
9fcfe0c8
TM		 2243 case -EINVAL:
2244 /* Happens, for instance, if the user specified a link
2245 * local IPv6 address without a scope-id.
2246 */
3ed5e2a2
TM		 2247 case -ECONNREFUSED:
2248 case -ECONNRESET:
2249 case -ENETUNREACH:
3913c78c 2250 case -EADDRINUSE:
3601c4a9 2251 case -ENOBUFS:
3ed5e2a2 2252 /* retry with existing socket, after a delay */
4efdd92c 2253 xs_tcp_force_close(xprt);
9fcfe0c8 2254 goto out;
a246b010 2255 }
2a491991 2256 status = -EAGAIN;
a246b010 2257out:
718ba5b8 2258 xprt_unlock_connect(xprt, transport);
68e220bd 2259 xprt_clear_connecting(xprt);
7d1e8255 2260 xprt_wake_pending_tasks(xprt, status);
68e220bd 2261}
b0d93ad5 2262
9903cd1c
CL		 2263/**
2264 * xs_connect - connect a socket to a remote endpoint
1b092092 2265 * @xprt: pointer to transport structure
9903cd1c
CL		 2266 * @task: address of RPC task that manages state of connect request
2267 *
2268 * TCP: If the remote end dropped the connection, delay reconnecting.
03bf4b70
CL		 2269 *
2270 * UDP socket connects are synchronous, but we use a work queue anyway
2271 * to guarantee that even unprivileged user processes can set up a
2272 * socket on a privileged port.
2273 *
2274 * If a UDP socket connect fails, the delay behavior here prevents
2275 * retry floods (hard mounts).
9903cd1c 2276 */
1b092092 2277static void xs_connect(struct rpc_xprt *xprt, struct rpc_task *task)
a246b010 2278{
ee0ac0c2 2279 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
a246b010 2280
718ba5b8
TM		 2281 WARN_ON_ONCE(!xprt_lock_connect(xprt, task, transport));
2282
de84d890
TM		 2283 /* Start by resetting any existing state */
2284 xs_reset_transport(transport);
2285
09a21c41 2286 if (transport->sock != NULL && !RPC_IS_SOFTCONN(task)) {
46121cf7
CL		 2287 dprintk("RPC: xs_connect delayed xprt %p for %lu "
2288 "seconds\n",
03bf4b70 2289 xprt, xprt->reestablish_timeout / HZ);
c1384c9c
TM		 2290 queue_delayed_work(rpciod_workqueue,
2291 &transport->connect_worker,
2292 xprt->reestablish_timeout);
03bf4b70 2293 xprt->reestablish_timeout <<= 1;
61d0a8e6
NB		 2294 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2295 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
03bf4b70
CL		 2296 if (xprt->reestablish_timeout > XS_TCP_MAX_REEST_TO)
2297 xprt->reestablish_timeout = XS_TCP_MAX_REEST_TO;
b0d93ad5 2298 } else {
46121cf7 2299 dprintk("RPC: xs_connect scheduled xprt %p\n", xprt);
c1384c9c
TM		 2300 queue_delayed_work(rpciod_workqueue,
2301 &transport->connect_worker, 0);
a246b010
CL		 2302 }
2303}
2304
176e21ee
CL		 2305/**
2306 * xs_local_print_stats - display AF_LOCAL socket-specifc stats
2307 * @xprt: rpc_xprt struct containing statistics
2308 * @seq: output file
2309 *
2310 */
2311static void xs_local_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2312{
2313 long idle_time = 0;
2314
2315 if (xprt_connected(xprt))
2316 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2317
2318 seq_printf(seq, "\txprt:\tlocal %lu %lu %lu %ld %lu %lu %lu "
15a45206 2319 "%llu %llu %lu %llu %llu\n",
176e21ee
CL		 2320 xprt->stat.bind_count,
2321 xprt->stat.connect_count,
2322 xprt->stat.connect_time,
2323 idle_time,
2324 xprt->stat.sends,
2325 xprt->stat.recvs,
2326 xprt->stat.bad_xids,
2327 xprt->stat.req_u,
15a45206
AA		 2328 xprt->stat.bklog_u,
2329 xprt->stat.max_slots,
2330 xprt->stat.sending_u,
2331 xprt->stat.pending_u);
176e21ee
CL		 2332}
2333
262ca07d
CL		 2334/**
2335 * xs_udp_print_stats - display UDP socket-specifc stats
2336 * @xprt: rpc_xprt struct containing statistics
2337 * @seq: output file
2338 *
2339 */
2340static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2341{
c8475461
CL		 2342 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2343
15a45206
AA		 2344 seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %llu %llu "
2345 "%lu %llu %llu\n",
fbfffbd5 2346 transport->srcport,
262ca07d
CL		 2347 xprt->stat.bind_count,
2348 xprt->stat.sends,
2349 xprt->stat.recvs,
2350 xprt->stat.bad_xids,
2351 xprt->stat.req_u,
15a45206
AA		 2352 xprt->stat.bklog_u,
2353 xprt->stat.max_slots,
2354 xprt->stat.sending_u,
2355 xprt->stat.pending_u);
262ca07d
CL		 2356}
2357
2358/**
2359 * xs_tcp_print_stats - display TCP socket-specifc stats
2360 * @xprt: rpc_xprt struct containing statistics
2361 * @seq: output file
2362 *
2363 */
2364static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2365{
c8475461 2366 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
262ca07d
CL		 2367 long idle_time = 0;
2368
2369 if (xprt_connected(xprt))
2370 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2371
15a45206
AA		 2372 seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu "
2373 "%llu %llu %lu %llu %llu\n",
fbfffbd5 2374 transport->srcport,
262ca07d
CL		 2375 xprt->stat.bind_count,
2376 xprt->stat.connect_count,
2377 xprt->stat.connect_time,
2378 idle_time,
2379 xprt->stat.sends,
2380 xprt->stat.recvs,
2381 xprt->stat.bad_xids,
2382 xprt->stat.req_u,
15a45206
AA		 2383 xprt->stat.bklog_u,
2384 xprt->stat.max_slots,
2385 xprt->stat.sending_u,
2386 xprt->stat.pending_u);
262ca07d
CL		 2387}
2388
4cfc7e60
RI		 2389/*
2390 * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason
2391 * we allocate pages instead doing a kmalloc like rpc_malloc is because we want
2392 * to use the server side send routines.
2393 */
5a51f13a 2394static void *bc_malloc(struct rpc_task *task, size_t size)
4cfc7e60
RI		 2395{
2396 struct page *page;
2397 struct rpc_buffer *buf;
2398
b8a13d03
WAA		 2399 WARN_ON_ONCE(size > PAGE_SIZE - sizeof(struct rpc_buffer));
2400 if (size > PAGE_SIZE - sizeof(struct rpc_buffer))
2401 return NULL;
4cfc7e60 2402
b8a13d03 2403 page = alloc_page(GFP_KERNEL);
4cfc7e60
RI		 2404 if (!page)
2405 return NULL;
2406
2407 buf = page_address(page);
2408 buf->len = PAGE_SIZE;
2409
2410 return buf->data;
2411}
2412
2413/*
2414 * Free the space allocated in the bc_alloc routine
2415 */
5a51f13a 2416static void bc_free(void *buffer)
4cfc7e60
RI		 2417{
2418 struct rpc_buffer *buf;
2419
2420 if (!buffer)
2421 return;
2422
2423 buf = container_of(buffer, struct rpc_buffer, data);
2424 free_page((unsigned long)buf);
2425}
2426
2427/*
2428 * Use the svc_sock to send the callback. Must be called with svsk->sk_mutex
2429 * held. Borrows heavily from svc_tcp_sendto and xs_tcp_send_request.
2430 */
2431static int bc_sendto(struct rpc_rqst *req)
2432{
2433 int len;
2434 struct xdr_buf *xbufp = &req->rq_snd_buf;
2435 struct rpc_xprt *xprt = req->rq_xprt;
2436 struct sock_xprt *transport =
2437 container_of(xprt, struct sock_xprt, xprt);
2438 struct socket *sock = transport->sock;
2439 unsigned long headoff;
2440 unsigned long tailoff;
2441
61677eee 2442 xs_encode_stream_record_marker(xbufp);
4cfc7e60
RI		 2443
2444 tailoff = (unsigned long)xbufp->tail[0].iov_base & ~PAGE_MASK;
2445 headoff = (unsigned long)xbufp->head[0].iov_base & ~PAGE_MASK;
2446 len = svc_send_common(sock, xbufp,
2447 virt_to_page(xbufp->head[0].iov_base), headoff,
2448 xbufp->tail[0].iov_base, tailoff);
2449
2450 if (len != xbufp->len) {
2451 printk(KERN_NOTICE "Error sending entire callback!\n");
2452 len = -EAGAIN;
2453 }
2454
2455 return len;
2456}
2457
2458/*
2459 * The send routine. Borrows from svc_send
2460 */
2461static int bc_send_request(struct rpc_task *task)
2462{
2463 struct rpc_rqst *req = task->tk_rqstp;
2464 struct svc_xprt *xprt;
4cfc7e60
RI		 2465 u32 len;
2466
2467 dprintk("sending request with xid: %08x\n", ntohl(req->rq_xid));
2468 /*
2469 * Get the server socket associated with this callback xprt
2470 */
2471 xprt = req->rq_xprt->bc_xprt;
4cfc7e60
RI		 2472
2473 /*
2474 * Grab the mutex to serialize data as the connection is shared
2475 * with the fore channel
2476 */
2477 if (!mutex_trylock(&xprt->xpt_mutex)) {
2478 rpc_sleep_on(&xprt->xpt_bc_pending, task, NULL);
2479 if (!mutex_trylock(&xprt->xpt_mutex))
2480 return -EAGAIN;
2481 rpc_wake_up_queued_task(&xprt->xpt_bc_pending, task);
2482 }
2483 if (test_bit(XPT_DEAD, &xprt->xpt_flags))
2484 len = -ENOTCONN;
2485 else
2486 len = bc_sendto(req);
2487 mutex_unlock(&xprt->xpt_mutex);
2488
2489 if (len > 0)
2490 len = 0;
2491
2492 return len;
2493}
2494
2495/*
2496 * The close routine. Since this is client initiated, we do nothing
2497 */
2498
2499static void bc_close(struct rpc_xprt *xprt)
2500{
4cfc7e60
RI		 2501}
2502
2503/*
2504 * The xprt destroy routine. Again, because this connection is client
2505 * initiated, we do nothing
2506 */
2507
2508static void bc_destroy(struct rpc_xprt *xprt)
2509{
47f72efa
KM		 2510 dprintk("RPC: bc_destroy xprt %p\n", xprt);
2511
2512 xs_xprt_free(xprt);
2513 module_put(THIS_MODULE);
4cfc7e60
RI		 2514}
2515
176e21ee
CL		 2516static struct rpc_xprt_ops xs_local_ops = {
2517 .reserve_xprt = xprt_reserve_xprt,
2518 .release_xprt = xs_tcp_release_xprt,
f39c1bfb 2519 .alloc_slot = xprt_alloc_slot,
176e21ee
CL		 2520 .rpcbind = xs_local_rpcbind,
2521 .set_port = xs_local_set_port,
dc107402 2522 .connect = xs_local_connect,
176e21ee
CL		 2523 .buf_alloc = rpc_malloc,
2524 .buf_free = rpc_free,
2525 .send_request = xs_local_send_request,
2526 .set_retrans_timeout = xprt_set_retrans_timeout_def,
2527 .close = xs_close,
a1311d87 2528 .destroy = xs_destroy,
176e21ee 2529 .print_stats = xs_local_print_stats,
d67fa4d8
JL		 2530 .enable_swap = xs_enable_swap,
2531 .disable_swap = xs_disable_swap,
176e21ee
CL		 2532};
2533
262965f5 2534static struct rpc_xprt_ops xs_udp_ops = {
43118c29 2535 .set_buffer_size = xs_udp_set_buffer_size,
12a80469 2536 .reserve_xprt = xprt_reserve_xprt_cong,
49e9a890 2537 .release_xprt = xprt_release_xprt_cong,
f39c1bfb 2538 .alloc_slot = xprt_alloc_slot,
45160d62 2539 .rpcbind = rpcb_getport_async,
92200412 2540 .set_port = xs_set_port,
262965f5 2541 .connect = xs_connect,
02107148
CL		 2542 .buf_alloc = rpc_malloc,
2543 .buf_free = rpc_free,
262965f5 2544 .send_request = xs_udp_send_request,
fe3aca29 2545 .set_retrans_timeout = xprt_set_retrans_timeout_rtt,
46c0ee8b 2546 .timer = xs_udp_timer,
a58dd398 2547 .release_request = xprt_release_rqst_cong,
262965f5
CL		 2548 .close = xs_close,
2549 .destroy = xs_destroy,
262ca07d 2550 .print_stats = xs_udp_print_stats,
d67fa4d8
JL		 2551 .enable_swap = xs_enable_swap,
2552 .disable_swap = xs_disable_swap,
4a068258 2553 .inject_disconnect = xs_inject_disconnect,
262965f5
CL		 2554};
2555
2556static struct rpc_xprt_ops xs_tcp_ops = {
12a80469 2557 .reserve_xprt = xprt_reserve_xprt,
e0ab53de 2558 .release_xprt = xs_tcp_release_xprt,
f39c1bfb 2559 .alloc_slot = xprt_lock_and_alloc_slot,
45160d62 2560 .rpcbind = rpcb_getport_async,
92200412 2561 .set_port = xs_set_port,
0b9e7943 2562 .connect = xs_connect,
02107148
CL		 2563 .buf_alloc = rpc_malloc,
2564 .buf_free = rpc_free,
262965f5 2565 .send_request = xs_tcp_send_request,
fe3aca29 2566 .set_retrans_timeout = xprt_set_retrans_timeout_def,
c627d31b 2567 .close = xs_tcp_shutdown,
9903cd1c 2568 .destroy = xs_destroy,
262ca07d 2569 .print_stats = xs_tcp_print_stats,
d67fa4d8
JL		 2570 .enable_swap = xs_enable_swap,
2571 .disable_swap = xs_disable_swap,
4a068258 2572 .inject_disconnect = xs_inject_disconnect,
a246b010
CL		 2573};
2574
4cfc7e60
RI		 2575/*
2576 * The rpc_xprt_ops for the server backchannel
2577 */
2578
2579static struct rpc_xprt_ops bc_tcp_ops = {
2580 .reserve_xprt = xprt_reserve_xprt,
2581 .release_xprt = xprt_release_xprt,
84e28a30 2582 .alloc_slot = xprt_alloc_slot,
4cfc7e60
RI		 2583 .buf_alloc = bc_malloc,
2584 .buf_free = bc_free,
2585 .send_request = bc_send_request,
2586 .set_retrans_timeout = xprt_set_retrans_timeout_def,
2587 .close = bc_close,
2588 .destroy = bc_destroy,
2589 .print_stats = xs_tcp_print_stats,
d67fa4d8
JL		 2590 .enable_swap = xs_enable_swap,
2591 .disable_swap = xs_disable_swap,
4a068258 2592 .inject_disconnect = xs_inject_disconnect,
4cfc7e60
RI		 2593};
2594
92476850
CL		 2595static int xs_init_anyaddr(const int family, struct sockaddr *sap)
2596{
2597 static const struct sockaddr_in sin = {
2598 .sin_family = AF_INET,
2599 .sin_addr.s_addr = htonl(INADDR_ANY),
2600 };
2601 static const struct sockaddr_in6 sin6 = {
2602 .sin6_family = AF_INET6,
2603 .sin6_addr = IN6ADDR_ANY_INIT,
2604 };
2605
2606 switch (family) {
176e21ee
CL		 2607 case AF_LOCAL:
2608 break;
92476850
CL		 2609 case AF_INET:
2610 memcpy(sap, &sin, sizeof(sin));
2611 break;
2612 case AF_INET6:
2613 memcpy(sap, &sin6, sizeof(sin6));
2614 break;
2615 default:
2616 dprintk("RPC: %s: Bad address family\n", __func__);
2617 return -EAFNOSUPPORT;
2618 }
2619 return 0;
2620}
2621
3c341b0b 2622static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
d9ba131d
TM		 2623 unsigned int slot_table_size,
2624 unsigned int max_slot_table_size)
c8541ecd
CL		 2625{
2626 struct rpc_xprt *xprt;
ffc2e518 2627 struct sock_xprt *new;
c8541ecd 2628
96802a09 2629 if (args->addrlen > sizeof(xprt->addr)) {
46121cf7 2630 dprintk("RPC: xs_setup_xprt: address too large\n");
c8541ecd
CL		 2631 return ERR_PTR(-EBADF);
2632 }
2633
d9ba131d
TM		 2634 xprt = xprt_alloc(args->net, sizeof(*new), slot_table_size,
2635 max_slot_table_size);
bd1722d4 2636 if (xprt == NULL) {
46121cf7
CL		 2637 dprintk("RPC: xs_setup_xprt: couldn't allocate "
2638 "rpc_xprt\n");
c8541ecd
CL		 2639 return ERR_PTR(-ENOMEM);
2640 }
c8541ecd 2641
bd1722d4 2642 new = container_of(xprt, struct sock_xprt, xprt);
96802a09
FM		 2643 memcpy(&xprt->addr, args->dstaddr, args->addrlen);
2644 xprt->addrlen = args->addrlen;
d3bc9a1d 2645 if (args->srcaddr)
fbfffbd5 2646 memcpy(&new->srcaddr, args->srcaddr, args->addrlen);
92476850
CL		 2647 else {
2648 int err;
2649 err = xs_init_anyaddr(args->dstaddr->sa_family,
2650 (struct sockaddr *)&new->srcaddr);
2aa13531
SK		 2651 if (err != 0) {
2652 xprt_free(xprt);
92476850 2653 return ERR_PTR(err);
2aa13531 2654 }
92476850 2655 }
c8541ecd
CL		 2656
2657 return xprt;
2658}
2659
176e21ee
CL		 2660static const struct rpc_timeout xs_local_default_timeout = {
2661 .to_initval = 10 * HZ,
2662 .to_maxval = 10 * HZ,
2663 .to_retries = 2,
2664};
2665
2666/**
2667 * xs_setup_local - Set up transport to use an AF_LOCAL socket
2668 * @args: rpc transport creation arguments
2669 *
2670 * AF_LOCAL is a "tpi_cots_ord" transport, just like TCP
2671 */
2672static struct rpc_xprt *xs_setup_local(struct xprt_create *args)
2673{
2674 struct sockaddr_un *sun = (struct sockaddr_un *)args->dstaddr;
2675 struct sock_xprt *transport;
2676 struct rpc_xprt *xprt;
2677 struct rpc_xprt *ret;
2678
d9ba131d
TM		 2679 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2680 xprt_max_tcp_slot_table_entries);
176e21ee
CL		 2681 if (IS_ERR(xprt))
2682 return xprt;
2683 transport = container_of(xprt, struct sock_xprt, xprt);
2684
2685 xprt->prot = 0;
2686 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2687 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2688
2689 xprt->bind_timeout = XS_BIND_TO;
2690 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2691 xprt->idle_timeout = XS_IDLE_DISC_TO;
2692
2693 xprt->ops = &xs_local_ops;
2694 xprt->timeout = &xs_local_default_timeout;
2695
93dc41bd
N		 2696 INIT_DELAYED_WORK(&transport->connect_worker,
2697 xs_dummy_setup_socket);
2698
176e21ee
CL		 2699 switch (sun->sun_family) {
2700 case AF_LOCAL:
2701 if (sun->sun_path[0] != '/') {
2702 dprintk("RPC: bad AF_LOCAL address: %s\n",
2703 sun->sun_path);
2704 ret = ERR_PTR(-EINVAL);
2705 goto out_err;
2706 }
2707 xprt_set_bound(xprt);
176e21ee 2708 xs_format_peer_addresses(xprt, "local", RPCBIND_NETID_LOCAL);
7073ea87
BF		 2709 ret = ERR_PTR(xs_local_setup_socket(transport));
2710 if (ret)
2711 goto out_err;
176e21ee
CL		 2712 break;
2713 default:
2714 ret = ERR_PTR(-EAFNOSUPPORT);
2715 goto out_err;
2716 }
2717
2718 dprintk("RPC: set up xprt to %s via AF_LOCAL\n",
2719 xprt->address_strings[RPC_DISPLAY_ADDR]);
2720
2721 if (try_module_get(THIS_MODULE))
2722 return xprt;
2723 ret = ERR_PTR(-EINVAL);
2724out_err:
315f3812 2725 xs_xprt_free(xprt);
176e21ee
CL		 2726 return ret;
2727}
2728
2881ae74
TM		 2729static const struct rpc_timeout xs_udp_default_timeout = {
2730 .to_initval = 5 * HZ,
2731 .to_maxval = 30 * HZ,
2732 .to_increment = 5 * HZ,
2733 .to_retries = 5,
2734};
2735
9903cd1c
CL		 2736/**
2737 * xs_setup_udp - Set up transport to use a UDP socket
96802a09 2738 * @args: rpc transport creation arguments
9903cd1c
CL		 2739 *
2740 */
483066d6 2741static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
a246b010 2742{
8f9d5b1a 2743 struct sockaddr *addr = args->dstaddr;
c8541ecd 2744 struct rpc_xprt *xprt;
c8475461 2745 struct sock_xprt *transport;
0a68b0be 2746 struct rpc_xprt *ret;
a246b010 2747
d9ba131d
TM		 2748 xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries,
2749 xprt_udp_slot_table_entries);
c8541ecd
CL		 2750 if (IS_ERR(xprt))
2751 return xprt;
c8475461 2752 transport = container_of(xprt, struct sock_xprt, xprt);
a246b010 2753
ec739ef0 2754 xprt->prot = IPPROTO_UDP;
808012fb 2755 xprt->tsh_size = 0;
a246b010
CL		 2756 /* XXX: header size can vary due to auth type, IPv6, etc. */
2757 xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
2758
03bf4b70 2759 xprt->bind_timeout = XS_BIND_TO;
03bf4b70
CL		 2760 xprt->reestablish_timeout = XS_UDP_REEST_TO;
2761 xprt->idle_timeout = XS_IDLE_DISC_TO;
a246b010 2762
262965f5 2763 xprt->ops = &xs_udp_ops;
a246b010 2764
ba7392bb 2765 xprt->timeout = &xs_udp_default_timeout;
a246b010 2766
8f9d5b1a
CL		 2767 switch (addr->sa_family) {
2768 case AF_INET:
2769 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2770 xprt_set_bound(xprt);
2771
2772 INIT_DELAYED_WORK(&transport->connect_worker,
8c14ff2a 2773 xs_udp_setup_socket);
9dc3b095 2774 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
8f9d5b1a
CL		 2775 break;
2776 case AF_INET6:
2777 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2778 xprt_set_bound(xprt);
2779
2780 INIT_DELAYED_WORK(&transport->connect_worker,
8c14ff2a 2781 xs_udp_setup_socket);
9dc3b095 2782 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
8f9d5b1a
CL		 2783 break;
2784 default:
0a68b0be
BF		 2785 ret = ERR_PTR(-EAFNOSUPPORT);
2786 goto out_err;
8f9d5b1a
CL		 2787 }
2788
c740eff8
CL		 2789 if (xprt_bound(xprt))
2790 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2791 xprt->address_strings[RPC_DISPLAY_ADDR],
2792 xprt->address_strings[RPC_DISPLAY_PORT],
2793 xprt->address_strings[RPC_DISPLAY_PROTO]);
2794 else
2795 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
2796 xprt->address_strings[RPC_DISPLAY_ADDR],
2797 xprt->address_strings[RPC_DISPLAY_PROTO]);
edb267a6 2798
bc25571e
TT		 2799 if (try_module_get(THIS_MODULE))
2800 return xprt;
0a68b0be
BF		 2801 ret = ERR_PTR(-EINVAL);
2802out_err:
315f3812 2803 xs_xprt_free(xprt);
0a68b0be 2804 return ret;
a246b010
CL		 2805}
2806
2881ae74
TM		 2807static const struct rpc_timeout xs_tcp_default_timeout = {
2808 .to_initval = 60 * HZ,
2809 .to_maxval = 60 * HZ,
2810 .to_retries = 2,
2811};
2812
9903cd1c
CL		 2813/**
2814 * xs_setup_tcp - Set up transport to use a TCP socket
96802a09 2815 * @args: rpc transport creation arguments
9903cd1c
CL		 2816 *
2817 */
483066d6 2818static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
a246b010 2819{
8f9d5b1a 2820 struct sockaddr *addr = args->dstaddr;
c8541ecd 2821 struct rpc_xprt *xprt;
c8475461 2822 struct sock_xprt *transport;
0a68b0be 2823 struct rpc_xprt *ret;
b7993ceb
TM		 2824 unsigned int max_slot_table_size = xprt_max_tcp_slot_table_entries;
2825
2826 if (args->flags & XPRT_CREATE_INFINITE_SLOTS)
2827 max_slot_table_size = RPC_MAX_SLOT_TABLE_LIMIT;
a246b010 2828
d9ba131d 2829 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
b7993ceb 2830 max_slot_table_size);
c8541ecd
CL		 2831 if (IS_ERR(xprt))
2832 return xprt;
c8475461 2833 transport = container_of(xprt, struct sock_xprt, xprt);
a246b010 2834
ec739ef0 2835 xprt->prot = IPPROTO_TCP;
808012fb 2836 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
808012fb 2837 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
a246b010 2838
03bf4b70 2839 xprt->bind_timeout = XS_BIND_TO;
03bf4b70
CL		 2840 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2841 xprt->idle_timeout = XS_IDLE_DISC_TO;
a246b010 2842
262965f5 2843 xprt->ops = &xs_tcp_ops;
ba7392bb 2844 xprt->timeout = &xs_tcp_default_timeout;
a246b010 2845
8f9d5b1a
CL		 2846 switch (addr->sa_family) {
2847 case AF_INET:
2848 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2849 xprt_set_bound(xprt);
2850
9dc3b095 2851 INIT_DELAYED_WORK(&transport->connect_worker,
cdd518d5 2852 xs_tcp_setup_socket);
9dc3b095 2853 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
8f9d5b1a
CL		 2854 break;
2855 case AF_INET6:
2856 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2857 xprt_set_bound(xprt);
2858
9dc3b095 2859 INIT_DELAYED_WORK(&transport->connect_worker,
cdd518d5 2860 xs_tcp_setup_socket);
9dc3b095 2861 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
8f9d5b1a
CL		 2862 break;
2863 default:
0a68b0be
BF		 2864 ret = ERR_PTR(-EAFNOSUPPORT);
2865 goto out_err;
8f9d5b1a
CL		 2866 }
2867
c740eff8
CL		 2868 if (xprt_bound(xprt))
2869 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2870 xprt->address_strings[RPC_DISPLAY_ADDR],
2871 xprt->address_strings[RPC_DISPLAY_PORT],
2872 xprt->address_strings[RPC_DISPLAY_PROTO]);
2873 else
2874 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
2875 xprt->address_strings[RPC_DISPLAY_ADDR],
2876 xprt->address_strings[RPC_DISPLAY_PROTO]);
2877
bc25571e
TT		 2878 if (try_module_get(THIS_MODULE))
2879 return xprt;
0a68b0be
BF		 2880 ret = ERR_PTR(-EINVAL);
2881out_err:
315f3812 2882 xs_xprt_free(xprt);
0a68b0be 2883 return ret;
a246b010 2884}
282b32e1 2885
f300baba
AB		 2886/**
2887 * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket
2888 * @args: rpc transport creation arguments
2889 *
2890 */
2891static struct rpc_xprt *xs_setup_bc_tcp(struct xprt_create *args)
2892{
2893 struct sockaddr *addr = args->dstaddr;
2894 struct rpc_xprt *xprt;
2895 struct sock_xprt *transport;
2896 struct svc_sock *bc_sock;
0a68b0be 2897 struct rpc_xprt *ret;
f300baba 2898
d9ba131d
TM		 2899 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2900 xprt_tcp_slot_table_entries);
f300baba
AB		 2901 if (IS_ERR(xprt))
2902 return xprt;
2903 transport = container_of(xprt, struct sock_xprt, xprt);
2904
2905 xprt->prot = IPPROTO_TCP;
2906 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2907 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2908 xprt->timeout = &xs_tcp_default_timeout;
2909
2910 /* backchannel */
2911 xprt_set_bound(xprt);
2912 xprt->bind_timeout = 0;
f300baba
AB		 2913 xprt->reestablish_timeout = 0;
2914 xprt->idle_timeout = 0;
2915
f300baba
AB		 2916 xprt->ops = &bc_tcp_ops;
2917
2918 switch (addr->sa_family) {
2919 case AF_INET:
2920 xs_format_peer_addresses(xprt, "tcp",
2921 RPCBIND_NETID_TCP);
2922 break;
2923 case AF_INET6:
2924 xs_format_peer_addresses(xprt, "tcp",
2925 RPCBIND_NETID_TCP6);
2926 break;
2927 default:
0a68b0be
BF		 2928 ret = ERR_PTR(-EAFNOSUPPORT);
2929 goto out_err;
f300baba
AB		 2930 }
2931
50fa0d40
PE		 2932 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2933 xprt->address_strings[RPC_DISPLAY_ADDR],
2934 xprt->address_strings[RPC_DISPLAY_PORT],
2935 xprt->address_strings[RPC_DISPLAY_PROTO]);
f300baba 2936
99de8ea9
BF		 2937 /*
2938 * Once we've associated a backchannel xprt with a connection,
28303ca3
WM		 2939 * we want to keep it around as long as the connection lasts,
2940 * in case we need to start using it for a backchannel again;
2941 * this reference won't be dropped until bc_xprt is destroyed.
99de8ea9
BF		 2942 */
2943 xprt_get(xprt);
2944 args->bc_xprt->xpt_bc_xprt = xprt;
2945 xprt->bc_xprt = args->bc_xprt;
2946 bc_sock = container_of(args->bc_xprt, struct svc_sock, sk_xprt);
2947 transport->sock = bc_sock->sk_sock;
2948 transport->inet = bc_sock->sk_sk;
2949
f300baba
AB		 2950 /*
2951 * Since we don't want connections for the backchannel, we set
2952 * the xprt status to connected
2953 */
2954 xprt_set_connected(xprt);
2955
f300baba
AB		 2956 if (try_module_get(THIS_MODULE))
2957 return xprt;
642aab58
KM		 2958
2959 args->bc_xprt->xpt_bc_xprt = NULL;
99de8ea9 2960 xprt_put(xprt);
0a68b0be
BF		 2961 ret = ERR_PTR(-EINVAL);
2962out_err:
315f3812 2963 xs_xprt_free(xprt);
0a68b0be 2964 return ret;
f300baba
AB		 2965}
2966
176e21ee
CL		 2967static struct xprt_class xs_local_transport = {
2968 .list = LIST_HEAD_INIT(xs_local_transport.list),
2969 .name = "named UNIX socket",
2970 .owner = THIS_MODULE,
2971 .ident = XPRT_TRANSPORT_LOCAL,
2972 .setup = xs_setup_local,
2973};
2974
bc25571e
TT		 2975static struct xprt_class xs_udp_transport = {
2976 .list = LIST_HEAD_INIT(xs_udp_transport.list),
2977 .name = "udp",
2978 .owner = THIS_MODULE,
f300baba 2979 .ident = XPRT_TRANSPORT_UDP,
bc25571e
TT		 2980 .setup = xs_setup_udp,
2981};
2982
2983static struct xprt_class xs_tcp_transport = {
2984 .list = LIST_HEAD_INIT(xs_tcp_transport.list),
2985 .name = "tcp",
2986 .owner = THIS_MODULE,
f300baba 2987 .ident = XPRT_TRANSPORT_TCP,
bc25571e
TT		 2988 .setup = xs_setup_tcp,
2989};
2990
f300baba
AB		 2991static struct xprt_class xs_bc_tcp_transport = {
2992 .list = LIST_HEAD_INIT(xs_bc_tcp_transport.list),
2993 .name = "tcp NFSv4.1 backchannel",
2994 .owner = THIS_MODULE,
2995 .ident = XPRT_TRANSPORT_BC_TCP,
2996 .setup = xs_setup_bc_tcp,
2997};
2998
282b32e1 2999/**
bc25571e 3000 * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
282b32e1
CL		 3001 *
3002 */
3003int init_socket_xprt(void)
3004{
f895b252 3005#if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
2b1bec5f 3006 if (!sunrpc_table_header)
0b4d4147 3007 sunrpc_table_header = register_sysctl_table(sunrpc_table);
fbf76683
CL		 3008#endif
3009
176e21ee 3010 xprt_register_transport(&xs_local_transport);
bc25571e
TT		 3011 xprt_register_transport(&xs_udp_transport);
3012 xprt_register_transport(&xs_tcp_transport);
f300baba 3013 xprt_register_transport(&xs_bc_tcp_transport);
bc25571e 3014
282b32e1
CL		 3015 return 0;
3016}
3017
3018/**
bc25571e 3019 * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
282b32e1
CL		 3020 *
3021 */
3022void cleanup_socket_xprt(void)
3023{
f895b252 3024#if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
fbf76683
CL		 3025 if (sunrpc_table_header) {
3026 unregister_sysctl_table(sunrpc_table_header);
3027 sunrpc_table_header = NULL;
3028 }
3029#endif
bc25571e 3030
176e21ee 3031 xprt_unregister_transport(&xs_local_transport);
bc25571e
TT		 3032 xprt_unregister_transport(&xs_udp_transport);
3033 xprt_unregister_transport(&xs_tcp_transport);
f300baba 3034 xprt_unregister_transport(&xs_bc_tcp_transport);
282b32e1 3035}
cbf11071 3036
9bbb9e5a
RR		 3037static int param_set_uint_minmax(const char *val,
3038 const struct kernel_param *kp,
cbf11071
TM		 3039 unsigned int min, unsigned int max)
3040{
00cfaa94 3041 unsigned int num;
cbf11071
TM		 3042 int ret;
3043
3044 if (!val)
3045 return -EINVAL;
00cfaa94 3046 ret = kstrtouint(val, 0, &num);
cbf11071
TM		 3047 if (ret == -EINVAL || num < min || num > max)
3048 return -EINVAL;
3049 *((unsigned int *)kp->arg) = num;
3050 return 0;
3051}
3052
9bbb9e5a 3053static int param_set_portnr(const char *val, const struct kernel_param *kp)
cbf11071
TM		 3054{
3055 return param_set_uint_minmax(val, kp,
3056 RPC_MIN_RESVPORT,
3057 RPC_MAX_RESVPORT);
3058}
3059
9bbb9e5a
RR		 3060static struct kernel_param_ops param_ops_portnr = {
3061 .set = param_set_portnr,
3062 .get = param_get_uint,
3063};
3064
cbf11071
TM		 3065#define param_check_portnr(name, p) \
3066 __param_check(name, p, unsigned int);
3067
3068module_param_named(min_resvport, xprt_min_resvport, portnr, 0644);
3069module_param_named(max_resvport, xprt_max_resvport, portnr, 0644);
3070
9bbb9e5a
RR		 3071static int param_set_slot_table_size(const char *val,
3072 const struct kernel_param *kp)
cbf11071
TM		 3073{
3074 return param_set_uint_minmax(val, kp,
3075 RPC_MIN_SLOT_TABLE,
3076 RPC_MAX_SLOT_TABLE);
3077}
3078
9bbb9e5a
RR		 3079static struct kernel_param_ops param_ops_slot_table_size = {
3080 .set = param_set_slot_table_size,
3081 .get = param_get_uint,
3082};
3083
cbf11071
TM		 3084#define param_check_slot_table_size(name, p) \
3085 __param_check(name, p, unsigned int);
3086
d9ba131d
TM		 3087static int param_set_max_slot_table_size(const char *val,
3088 const struct kernel_param *kp)
3089{
3090 return param_set_uint_minmax(val, kp,
3091 RPC_MIN_SLOT_TABLE,
3092 RPC_MAX_SLOT_TABLE_LIMIT);
3093}
3094
3095static struct kernel_param_ops param_ops_max_slot_table_size = {
3096 .set = param_set_max_slot_table_size,
3097 .get = param_get_uint,
3098};
3099
3100#define param_check_max_slot_table_size(name, p) \
3101 __param_check(name, p, unsigned int);
3102
cbf11071
TM		 3103module_param_named(tcp_slot_table_entries, xprt_tcp_slot_table_entries,
3104 slot_table_size, 0644);
d9ba131d
TM		 3105module_param_named(tcp_max_slot_table_entries, xprt_max_tcp_slot_table_entries,
3106 max_slot_table_size, 0644);
cbf11071
TM		 3107module_param_named(udp_slot_table_entries, xprt_udp_slot_table_entries,
3108 slot_table_size, 0644);
3109
Linux 6.x block layer and io_uring tree(s)