projects / deliverable / linux.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
ubifs: make ubifs_[get|set]xattr atomic
[deliverable/linux.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
262965f5
CL		 680 /* If we've sent the entire packet, immediately
681 * reset the count of bytes sent. */
f279cd00
JB		 682 req->rq_bytes_sent += sent;
683 req->rq_xmit_bytes_sent += sent;
262965f5
CL		 684 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
685 req->rq_bytes_sent = 0;
686 return 0;
687 }
a246b010 688
f580dd04
TM		 689 if (status < 0)
690 break;
691 if (sent == 0) {
692 status = -EAGAIN;
693 break;
694 }
a246b010 695 }
743c69e7
N		 696 if (status == -EAGAIN && sk_stream_is_writeable(transport->inet))
697 status = -ENOBUFS;
a246b010 698
262965f5 699 switch (status) {
fba91afb
TM		 700 case -ENOTSOCK:
701 status = -ENOTCONN;
702 /* Should we call xs_close() here? */
703 break;
262965f5 704 case -EAGAIN:
5e3771ce 705 status = xs_nospace(task);
262965f5 706 break;
c8485e4d
TM		 707 default:
708 dprintk("RPC: sendmsg returned unrecognized error %d\n",
709 -status);
262965f5 710 case -ECONNRESET:
2a9e1cfa 711 case -ECONNREFUSED:
262965f5 712 case -ENOTCONN:
3913c78c 713 case -EADDRINUSE:
b5872f0c 714 case -ENOBUFS:
b9d2bb2e 715 case -EPIPE:
b6ddf64f 716 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
a246b010 717 }
5fe46e9d 718
a246b010
CL		 719 return status;
720}
721
e0ab53de
TM		 722/**
723 * xs_tcp_release_xprt - clean up after a tcp transmission
724 * @xprt: transport
725 * @task: rpc task
726 *
727 * This cleans up if an error causes us to abort the transmission of a request.
728 * In this case, the socket may need to be reset in order to avoid confusing
729 * the server.
730 */
731static void xs_tcp_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
732{
733 struct rpc_rqst *req;
734
735 if (task != xprt->snd_task)
736 return;
737 if (task == NULL)
738 goto out_release;
739 req = task->tk_rqstp;
43cedbf0
TM		 740 if (req == NULL)
741 goto out_release;
e0ab53de
TM		 742 if (req->rq_bytes_sent == 0)
743 goto out_release;
744 if (req->rq_bytes_sent == req->rq_snd_buf.len)
745 goto out_release;
a4f0835c 746 set_bit(XPRT_CLOSE_WAIT, &xprt->state);
e0ab53de
TM		 747out_release:
748 xprt_release_xprt(xprt, task);
749}
750
2a9e1cfa
TM		 751static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk)
752{
753 transport->old_data_ready = sk->sk_data_ready;
754 transport->old_state_change = sk->sk_state_change;
755 transport->old_write_space = sk->sk_write_space;
2118071d 756 transport->old_error_report = sk->sk_error_report;
2a9e1cfa
TM		 757}
758
759static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk)
760{
761 sk->sk_data_ready = transport->old_data_ready;
762 sk->sk_state_change = transport->old_state_change;
763 sk->sk_write_space = transport->old_write_space;
2118071d
TM		 764 sk->sk_error_report = transport->old_error_report;
765}
766
b70ae915
TM		 767static void xs_sock_reset_connection_flags(struct rpc_xprt *xprt)
768{
769 smp_mb__before_atomic();
770 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
771 clear_bit(XPRT_CLOSING, &xprt->state);
772 smp_mb__after_atomic();
773}
774
775static void xs_sock_mark_closed(struct rpc_xprt *xprt)
776{
777 xs_sock_reset_connection_flags(xprt);
778 /* Mark transport as closed and wake up all pending tasks */
779 xprt_disconnect_done(xprt);
780}
781
2118071d
TM		 782/**
783 * xs_error_report - callback to handle TCP socket state errors
784 * @sk: socket
785 *
786 * Note: we don't call sock_error() since there may be a rpc_task
787 * using the socket, and so we don't want to clear sk->sk_err.
788 */
789static void xs_error_report(struct sock *sk)
790{
791 struct rpc_xprt *xprt;
792 int err;
793
794 read_lock_bh(&sk->sk_callback_lock);
795 if (!(xprt = xprt_from_sock(sk)))
796 goto out;
797
798 err = -sk->sk_err;
799 if (err == 0)
800 goto out;
b70ae915
TM		 801 /* Is this a reset event? */
802 if (sk->sk_state == TCP_CLOSE)
803 xs_sock_mark_closed(xprt);
2118071d
TM		 804 dprintk("RPC: xs_error_report client %p, error=%d...\n",
805 xprt, -err);
e8353c76 806 trace_rpc_socket_error(xprt, sk->sk_socket, err);
2118071d
TM		 807 xprt_wake_pending_tasks(xprt, err);
808 out:
809 read_unlock_bh(&sk->sk_callback_lock);
2a9e1cfa
TM		 810}
811
fe315e76 812static void xs_reset_transport(struct sock_xprt *transport)
a246b010 813{
ee0ac0c2
CL		 814 struct socket *sock = transport->sock;
815 struct sock *sk = transport->inet;
6cc7e908 816 struct rpc_xprt *xprt = &transport->xprt;
a246b010 817
fe315e76
CL		 818 if (sk == NULL)
819 return;
9903cd1c 820
264d1df3
JL		 821 if (atomic_read(&transport->xprt.swapper))
822 sk_clear_memalloc(sk);
823
09939204
TM		 824 kernel_sock_shutdown(sock, SHUT_RDWR);
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);
0c78789e 833 xprt_clear_connected(xprt);
a246b010 834 write_unlock_bh(&sk->sk_callback_lock);
6cc7e908 835 xs_sock_reset_connection_flags(xprt);
a246b010 836
6cc7e908 837 trace_rpc_socket_close(xprt, sock);
a246b010 838 sock_release(sock);
fe315e76
CL		 839}
840
841/**
842 * xs_close - close a socket
843 * @xprt: transport
844 *
845 * This is used when all requests are complete; ie, no DRC state remains
846 * on the server we want to save.
f75e6745
TM		 847 *
848 * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with
849 * xs_reset_transport() zeroing the socket from underneath a writer.
fe315e76
CL		 850 */
851static void xs_close(struct rpc_xprt *xprt)
852{
853 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
854
855 dprintk("RPC: xs_close xprt %p\n", xprt);
856
857 xs_reset_transport(transport);
61d0a8e6 858 xprt->reestablish_timeout = 0;
fe315e76 859
62da3b24 860 xprt_disconnect_done(xprt);
a246b010
CL		 861}
862
4a068258
CL		 863static void xs_inject_disconnect(struct rpc_xprt *xprt)
864{
865 dprintk("RPC: injecting transport disconnect on xprt=%p\n",
866 xprt);
867 xprt_disconnect_done(xprt);
868}
869
315f3812
KM		 870static void xs_xprt_free(struct rpc_xprt *xprt)
871{
872 xs_free_peer_addresses(xprt);
873 xprt_free(xprt);
874}
875
9903cd1c
CL		 876/**
877 * xs_destroy - prepare to shutdown a transport
878 * @xprt: doomed transport
879 *
880 */
881static void xs_destroy(struct rpc_xprt *xprt)
a246b010 882{
03c78827
TM		 883 struct sock_xprt *transport = container_of(xprt,
884 struct sock_xprt, xprt);
46121cf7 885 dprintk("RPC: xs_destroy xprt %p\n", xprt);
9903cd1c 886
03c78827 887 cancel_delayed_work_sync(&transport->connect_worker);
a1311d87 888 xs_close(xprt);
315f3812 889 xs_xprt_free(xprt);
a1311d87 890 module_put(THIS_MODULE);
a246b010
CL		 891}
892
176e21ee
CL		 893static int xs_local_copy_to_xdr(struct xdr_buf *xdr, struct sk_buff *skb)
894{
895 struct xdr_skb_reader desc = {
896 .skb = skb,
897 .offset = sizeof(rpc_fraghdr),
898 .count = skb->len - sizeof(rpc_fraghdr),
899 };
900
901 if (xdr_partial_copy_from_skb(xdr, 0, &desc, xdr_skb_read_bits) < 0)
902 return -1;
903 if (desc.count)
904 return -1;
905 return 0;
906}
907
908/**
909 * xs_local_data_ready - "data ready" callback for AF_LOCAL sockets
910 * @sk: socket with data to read
176e21ee
CL		 911 *
912 * Currently this assumes we can read the whole reply in a single gulp.
913 */
676d2369 914static void xs_local_data_ready(struct sock *sk)
176e21ee
CL		 915{
916 struct rpc_task *task;
917 struct rpc_xprt *xprt;
918 struct rpc_rqst *rovr;
919 struct sk_buff *skb;
920 int err, repsize, copied;
921 u32 _xid;
922 __be32 *xp;
923
924 read_lock_bh(&sk->sk_callback_lock);
925 dprintk("RPC: %s...\n", __func__);
926 xprt = xprt_from_sock(sk);
927 if (xprt == NULL)
928 goto out;
929
930 skb = skb_recv_datagram(sk, 0, 1, &err);
931 if (skb == NULL)
932 goto out;
933
176e21ee
CL		 934 repsize = skb->len - sizeof(rpc_fraghdr);
935 if (repsize < 4) {
936 dprintk("RPC: impossible RPC reply size %d\n", repsize);
937 goto dropit;
938 }
939
940 /* Copy the XID from the skb... */
941 xp = skb_header_pointer(skb, sizeof(rpc_fraghdr), sizeof(_xid), &_xid);
942 if (xp == NULL)
943 goto dropit;
944
945 /* Look up and lock the request corresponding to the given XID */
946 spin_lock(&xprt->transport_lock);
947 rovr = xprt_lookup_rqst(xprt, *xp);
948 if (!rovr)
949 goto out_unlock;
950 task = rovr->rq_task;
951
952 copied = rovr->rq_private_buf.buflen;
953 if (copied > repsize)
954 copied = repsize;
955
956 if (xs_local_copy_to_xdr(&rovr->rq_private_buf, skb)) {
957 dprintk("RPC: sk_buff copy failed\n");
958 goto out_unlock;
959 }
960
961 xprt_complete_rqst(task, copied);
962
963 out_unlock:
964 spin_unlock(&xprt->transport_lock);
965 dropit:
966 skb_free_datagram(sk, skb);
967 out:
968 read_unlock_bh(&sk->sk_callback_lock);
969}
970
9903cd1c
CL		 971/**
972 * xs_udp_data_ready - "data ready" callback for UDP sockets
973 * @sk: socket with data to read
9903cd1c 974 *
a246b010 975 */
676d2369 976static void xs_udp_data_ready(struct sock *sk)
a246b010 977{
9903cd1c
CL		 978 struct rpc_task *task;
979 struct rpc_xprt *xprt;
a246b010 980 struct rpc_rqst *rovr;
9903cd1c 981 struct sk_buff *skb;
a246b010 982 int err, repsize, copied;
d8ed029d
AD		 983 u32 _xid;
984 __be32 *xp;
a246b010 985
f064af1e 986 read_lock_bh(&sk->sk_callback_lock);
46121cf7 987 dprintk("RPC: xs_udp_data_ready...\n");
9903cd1c 988 if (!(xprt = xprt_from_sock(sk)))
a246b010 989 goto out;
a246b010
CL		 990
991 if ((skb = skb_recv_datagram(sk, 0, 1, &err)) == NULL)
992 goto out;
993
a246b010
CL		 994 repsize = skb->len - sizeof(struct udphdr);
995 if (repsize < 4) {
46121cf7 996 dprintk("RPC: impossible RPC reply size %d!\n", repsize);
a246b010
CL		 997 goto dropit;
998 }
999
1000 /* Copy the XID from the skb... */
1001 xp = skb_header_pointer(skb, sizeof(struct udphdr),
1002 sizeof(_xid), &_xid);
1003 if (xp == NULL)
1004 goto dropit;
1005
1006 /* Look up and lock the request corresponding to the given XID */
4a0f8c04 1007 spin_lock(&xprt->transport_lock);
a246b010
CL		 1008 rovr = xprt_lookup_rqst(xprt, *xp);
1009 if (!rovr)
1010 goto out_unlock;
1011 task = rovr->rq_task;
1012
a246b010
CL		 1013 if ((copied = rovr->rq_private_buf.buflen) > repsize)
1014 copied = repsize;
1015
1016 /* Suck it into the iovec, verify checksum if not done by hw. */
1781f7f5
HX		 1017 if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
1018 UDPX_INC_STATS_BH(sk, UDP_MIB_INERRORS);
a246b010 1019 goto out_unlock;
1781f7f5
HX		 1020 }
1021
1022 UDPX_INC_STATS_BH(sk, UDP_MIB_INDATAGRAMS);
a246b010 1023
6a24dfb6 1024 xprt_adjust_cwnd(xprt, task, copied);
1570c1e4 1025 xprt_complete_rqst(task, copied);
a246b010
CL		 1026
1027 out_unlock:
4a0f8c04 1028 spin_unlock(&xprt->transport_lock);
a246b010
CL		 1029 dropit:
1030 skb_free_datagram(sk, skb);
1031 out:
f064af1e 1032 read_unlock_bh(&sk->sk_callback_lock);
a246b010
CL		 1033}
1034
a519fc7a
TM		 1035/*
1036 * Helper function to force a TCP close if the server is sending
1037 * junk and/or it has put us in CLOSE_WAIT
1038 */
1039static void xs_tcp_force_close(struct rpc_xprt *xprt)
1040{
a519fc7a
TM		 1041 xprt_force_disconnect(xprt);
1042}
1043
dd456471 1044static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
a246b010 1045{
51971139 1046 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
a246b010
CL		 1047 size_t len, used;
1048 char *p;
1049
51971139
CL		 1050 p = ((char *) &transport->tcp_fraghdr) + transport->tcp_offset;
1051 len = sizeof(transport->tcp_fraghdr) - transport->tcp_offset;
9d292316 1052 used = xdr_skb_read_bits(desc, p, len);
51971139 1053 transport->tcp_offset += used;
a246b010
CL		 1054 if (used != len)
1055 return;
808012fb 1056
51971139
CL		 1057 transport->tcp_reclen = ntohl(transport->tcp_fraghdr);
1058 if (transport->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
e136d092 1059 transport->tcp_flags |= TCP_RCV_LAST_FRAG;
a246b010 1060 else
e136d092 1061 transport->tcp_flags &= ~TCP_RCV_LAST_FRAG;
51971139 1062 transport->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
808012fb 1063
e136d092 1064 transport->tcp_flags &= ~TCP_RCV_COPY_FRAGHDR;
51971139 1065 transport->tcp_offset = 0;
808012fb 1066
a246b010 1067 /* Sanity check of the record length */
18dca02a 1068 if (unlikely(transport->tcp_reclen < 8)) {
46121cf7 1069 dprintk("RPC: invalid TCP record fragment length\n");
a519fc7a 1070 xs_tcp_force_close(xprt);
9903cd1c 1071 return;
a246b010 1072 }
46121cf7 1073 dprintk("RPC: reading TCP record fragment of length %d\n",
51971139 1074 transport->tcp_reclen);
a246b010
CL		 1075}
1076
51971139 1077static void xs_tcp_check_fraghdr(struct sock_xprt *transport)
a246b010 1078{
51971139 1079 if (transport->tcp_offset == transport->tcp_reclen) {
e136d092 1080 transport->tcp_flags |= TCP_RCV_COPY_FRAGHDR;
51971139 1081 transport->tcp_offset = 0;
e136d092
CL		 1082 if (transport->tcp_flags & TCP_RCV_LAST_FRAG) {
1083 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1084 transport->tcp_flags |= TCP_RCV_COPY_XID;
51971139 1085 transport->tcp_copied = 0;
a246b010
CL		 1086 }
1087 }
1088}
1089
dd456471 1090static inline void xs_tcp_read_xid(struct sock_xprt *transport, struct xdr_skb_reader *desc)
a246b010
CL		 1091{
1092 size_t len, used;
1093 char *p;
1094
51971139 1095 len = sizeof(transport->tcp_xid) - transport->tcp_offset;
46121cf7 1096 dprintk("RPC: reading XID (%Zu bytes)\n", len);
51971139 1097 p = ((char *) &transport->tcp_xid) + transport->tcp_offset;
9d292316 1098 used = xdr_skb_read_bits(desc, p, len);
51971139 1099 transport->tcp_offset += used;
a246b010
CL		 1100 if (used != len)
1101 return;
e136d092 1102 transport->tcp_flags &= ~TCP_RCV_COPY_XID;
f4a2e418 1103 transport->tcp_flags |= TCP_RCV_READ_CALLDIR;
51971139 1104 transport->tcp_copied = 4;
18dca02a
RL		 1105 dprintk("RPC: reading %s XID %08x\n",
1106 (transport->tcp_flags & TCP_RPC_REPLY) ? "reply for"
1107 : "request with",
51971139
CL		 1108 ntohl(transport->tcp_xid));
1109 xs_tcp_check_fraghdr(transport);
a246b010
CL		 1110}
1111
18dca02a
RL		 1112static inline void xs_tcp_read_calldir(struct sock_xprt *transport,
1113 struct xdr_skb_reader *desc)
a246b010 1114{
18dca02a
RL		 1115 size_t len, used;
1116 u32 offset;
b76ce561 1117 char *p;
18dca02a
RL		 1118
1119 /*
1120 * We want transport->tcp_offset to be 8 at the end of this routine
1121 * (4 bytes for the xid and 4 bytes for the call/reply flag).
1122 * When this function is called for the first time,
1123 * transport->tcp_offset is 4 (after having already read the xid).
1124 */
1125 offset = transport->tcp_offset - sizeof(transport->tcp_xid);
b76ce561 1126 len = sizeof(transport->tcp_calldir) - offset;
18dca02a 1127 dprintk("RPC: reading CALL/REPLY flag (%Zu bytes)\n", len);
b76ce561
TM		 1128 p = ((char *) &transport->tcp_calldir) + offset;
1129 used = xdr_skb_read_bits(desc, p, len);
18dca02a
RL		 1130 transport->tcp_offset += used;
1131 if (used != len)
1132 return;
f4a2e418 1133 transport->tcp_flags &= ~TCP_RCV_READ_CALLDIR;
f4a2e418
RL		 1134 /*
1135 * We don't yet have the XDR buffer, so we will write the calldir
1136 * out after we get the buffer from the 'struct rpc_rqst'
1137 */
b76ce561
TM		 1138 switch (ntohl(transport->tcp_calldir)) {
1139 case RPC_REPLY:
1140 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1141 transport->tcp_flags |= TCP_RCV_COPY_DATA;
18dca02a 1142 transport->tcp_flags |= TCP_RPC_REPLY;
b76ce561
TM		 1143 break;
1144 case RPC_CALL:
1145 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1146 transport->tcp_flags |= TCP_RCV_COPY_DATA;
18dca02a 1147 transport->tcp_flags &= ~TCP_RPC_REPLY;
b76ce561
TM		 1148 break;
1149 default:
1150 dprintk("RPC: invalid request message type\n");
a519fc7a 1151 xs_tcp_force_close(&transport->xprt);
b76ce561 1152 }
18dca02a
RL		 1153 xs_tcp_check_fraghdr(transport);
1154}
1155
44b98efd
RL		 1156static inline void xs_tcp_read_common(struct rpc_xprt *xprt,
1157 struct xdr_skb_reader *desc,
1158 struct rpc_rqst *req)
a246b010 1159{
44b98efd
RL		 1160 struct sock_xprt *transport =
1161 container_of(xprt, struct sock_xprt, xprt);
a246b010
CL		 1162 struct xdr_buf *rcvbuf;
1163 size_t len;
1164 ssize_t r;
1165
a246b010 1166 rcvbuf = &req->rq_private_buf;
f4a2e418
RL		 1167
1168 if (transport->tcp_flags & TCP_RCV_COPY_CALLDIR) {
1169 /*
1170 * Save the RPC direction in the XDR buffer
1171 */
f4a2e418 1172 memcpy(rcvbuf->head[0].iov_base + transport->tcp_copied,
b76ce561
TM		 1173 &transport->tcp_calldir,
1174 sizeof(transport->tcp_calldir));
1175 transport->tcp_copied += sizeof(transport->tcp_calldir);
f4a2e418 1176 transport->tcp_flags &= ~TCP_RCV_COPY_CALLDIR;
a246b010
CL		 1177 }
1178
a246b010 1179 len = desc->count;
51971139 1180 if (len > transport->tcp_reclen - transport->tcp_offset) {
dd456471 1181 struct xdr_skb_reader my_desc;
a246b010 1182
51971139 1183 len = transport->tcp_reclen - transport->tcp_offset;
a246b010
CL		 1184 memcpy(&my_desc, desc, sizeof(my_desc));
1185 my_desc.count = len;
51971139 1186 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
9d292316 1187 &my_desc, xdr_skb_read_bits);
a246b010
CL		 1188 desc->count -= r;
1189 desc->offset += r;
1190 } else
51971139 1191 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
9d292316 1192 desc, xdr_skb_read_bits);
a246b010
CL		 1193
1194 if (r > 0) {
51971139
CL		 1195 transport->tcp_copied += r;
1196 transport->tcp_offset += r;
a246b010
CL		 1197 }
1198 if (r != len) {
1199 /* Error when copying to the receive buffer,
1200 * usually because we weren't able to allocate
1201 * additional buffer pages. All we can do now
e136d092 1202 * is turn off TCP_RCV_COPY_DATA, so the request
a246b010
CL		 1203 * will not receive any additional updates,
1204 * and time out.
1205 * Any remaining data from this record will
1206 * be discarded.
1207 */
e136d092 1208 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
46121cf7 1209 dprintk("RPC: XID %08x truncated request\n",
51971139 1210 ntohl(transport->tcp_xid));
46121cf7
CL		 1211 dprintk("RPC: xprt = %p, tcp_copied = %lu, "
1212 "tcp_offset = %u, tcp_reclen = %u\n",
1213 xprt, transport->tcp_copied,
1214 transport->tcp_offset, transport->tcp_reclen);
44b98efd 1215 return;
a246b010
CL		 1216 }
1217
46121cf7 1218 dprintk("RPC: XID %08x read %Zd bytes\n",
51971139 1219 ntohl(transport->tcp_xid), r);
46121cf7
CL		 1220 dprintk("RPC: xprt = %p, tcp_copied = %lu, tcp_offset = %u, "
1221 "tcp_reclen = %u\n", xprt, transport->tcp_copied,
1222 transport->tcp_offset, transport->tcp_reclen);
51971139
CL		 1223
1224 if (transport->tcp_copied == req->rq_private_buf.buflen)
e136d092 1225 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
51971139 1226 else if (transport->tcp_offset == transport->tcp_reclen) {
e136d092
CL		 1227 if (transport->tcp_flags & TCP_RCV_LAST_FRAG)
1228 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
a246b010 1229 }
44b98efd
RL		 1230}
1231
1232/*
1233 * Finds the request corresponding to the RPC xid and invokes the common
1234 * tcp read code to read the data.
1235 */
1236static inline int xs_tcp_read_reply(struct rpc_xprt *xprt,
1237 struct xdr_skb_reader *desc)
1238{
1239 struct sock_xprt *transport =
1240 container_of(xprt, struct sock_xprt, xprt);
1241 struct rpc_rqst *req;
1242
1243 dprintk("RPC: read reply XID %08x\n", ntohl(transport->tcp_xid));
1244
1245 /* Find and lock the request corresponding to this xid */
1246 spin_lock(&xprt->transport_lock);
1247 req = xprt_lookup_rqst(xprt, transport->tcp_xid);
1248 if (!req) {
1249 dprintk("RPC: XID %08x request not found!\n",
1250 ntohl(transport->tcp_xid));
1251 spin_unlock(&xprt->transport_lock);
1252 return -1;
1253 }
1254
1255 xs_tcp_read_common(xprt, desc, req);
1256
e136d092 1257 if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
51971139 1258 xprt_complete_rqst(req->rq_task, transport->tcp_copied);
44b98efd 1259
4a0f8c04 1260 spin_unlock(&xprt->transport_lock);
44b98efd
RL		 1261 return 0;
1262}
1263
9e00abc3 1264#if defined(CONFIG_SUNRPC_BACKCHANNEL)
44b98efd
RL		 1265/*
1266 * Obtains an rpc_rqst previously allocated and invokes the common
1267 * tcp read code to read the data. The result is placed in the callback
1268 * queue.
1269 * If we're unable to obtain the rpc_rqst we schedule the closing of the
1270 * connection and return -1.
1271 */
2ea24497 1272static int xs_tcp_read_callback(struct rpc_xprt *xprt,
44b98efd
RL		 1273 struct xdr_skb_reader *desc)
1274{
1275 struct sock_xprt *transport =
1276 container_of(xprt, struct sock_xprt, xprt);
1277 struct rpc_rqst *req;
1278
2ea24497
TM		 1279 /* Look up and lock the request corresponding to the given XID */
1280 spin_lock(&xprt->transport_lock);
1281 req = xprt_lookup_bc_request(xprt, transport->tcp_xid);
44b98efd 1282 if (req == NULL) {
2ea24497 1283 spin_unlock(&xprt->transport_lock);
44b98efd
RL		 1284 printk(KERN_WARNING "Callback slot table overflowed\n");
1285 xprt_force_disconnect(xprt);
1286 return -1;
1287 }
1288
44b98efd
RL		 1289 dprintk("RPC: read callback XID %08x\n", ntohl(req->rq_xid));
1290 xs_tcp_read_common(xprt, desc, req);
1291
2ea24497
TM		 1292 if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1293 xprt_complete_bc_request(req, transport->tcp_copied);
1294 spin_unlock(&xprt->transport_lock);
44b98efd
RL		 1295
1296 return 0;
1297}
1298
1299static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1300 struct xdr_skb_reader *desc)
1301{
1302 struct sock_xprt *transport =
1303 container_of(xprt, struct sock_xprt, xprt);
1304
1305 return (transport->tcp_flags & TCP_RPC_REPLY) ?
1306 xs_tcp_read_reply(xprt, desc) :
1307 xs_tcp_read_callback(xprt, desc);
1308}
1309#else
1310static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1311 struct xdr_skb_reader *desc)
1312{
1313 return xs_tcp_read_reply(xprt, desc);
1314}
9e00abc3 1315#endif /* CONFIG_SUNRPC_BACKCHANNEL */
44b98efd
RL		 1316
1317/*
1318 * Read data off the transport. This can be either an RPC_CALL or an
1319 * RPC_REPLY. Relay the processing to helper functions.
1320 */
1321static void xs_tcp_read_data(struct rpc_xprt *xprt,
1322 struct xdr_skb_reader *desc)
1323{
1324 struct sock_xprt *transport =
1325 container_of(xprt, struct sock_xprt, xprt);
1326
1327 if (_xs_tcp_read_data(xprt, desc) == 0)
1328 xs_tcp_check_fraghdr(transport);
1329 else {
1330 /*
1331 * The transport_lock protects the request handling.
1332 * There's no need to hold it to update the tcp_flags.
1333 */
1334 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1335 }
a246b010
CL		 1336}
1337
dd456471 1338static inline void xs_tcp_read_discard(struct sock_xprt *transport, struct xdr_skb_reader *desc)
a246b010
CL		 1339{
1340 size_t len;
1341
51971139 1342 len = transport->tcp_reclen - transport->tcp_offset;
a246b010
CL		 1343 if (len > desc->count)
1344 len = desc->count;
1345 desc->count -= len;
1346 desc->offset += len;
51971139 1347 transport->tcp_offset += len;
46121cf7 1348 dprintk("RPC: discarded %Zu bytes\n", len);
51971139 1349 xs_tcp_check_fraghdr(transport);
a246b010
CL		 1350}
1351
9903cd1c 1352static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
a246b010
CL		 1353{
1354 struct rpc_xprt *xprt = rd_desc->arg.data;
51971139 1355 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
dd456471 1356 struct xdr_skb_reader desc = {
a246b010
CL		 1357 .skb = skb,
1358 .offset = offset,
1359 .count = len,
9903cd1c 1360 };
a246b010 1361
46121cf7 1362 dprintk("RPC: xs_tcp_data_recv started\n");
a246b010 1363 do {
1a867a08 1364 trace_xs_tcp_data_recv(transport);
a246b010
CL		 1365 /* Read in a new fragment marker if necessary */
1366 /* Can we ever really expect to get completely empty fragments? */
e136d092 1367 if (transport->tcp_flags & TCP_RCV_COPY_FRAGHDR) {
9903cd1c 1368 xs_tcp_read_fraghdr(xprt, &desc);
a246b010
CL		 1369 continue;
1370 }
1371 /* Read in the xid if necessary */
e136d092 1372 if (transport->tcp_flags & TCP_RCV_COPY_XID) {
51971139 1373 xs_tcp_read_xid(transport, &desc);
a246b010
CL		 1374 continue;
1375 }
18dca02a 1376 /* Read in the call/reply flag */
f4a2e418 1377 if (transport->tcp_flags & TCP_RCV_READ_CALLDIR) {
18dca02a
RL		 1378 xs_tcp_read_calldir(transport, &desc);
1379 continue;
1380 }
a246b010 1381 /* Read in the request data */
e136d092 1382 if (transport->tcp_flags & TCP_RCV_COPY_DATA) {
44b98efd 1383 xs_tcp_read_data(xprt, &desc);
a246b010
CL		 1384 continue;
1385 }
1386 /* Skip over any trailing bytes on short reads */
51971139 1387 xs_tcp_read_discard(transport, &desc);
a246b010 1388 } while (desc.count);
1a867a08 1389 trace_xs_tcp_data_recv(transport);
46121cf7 1390 dprintk("RPC: xs_tcp_data_recv done\n");
a246b010
CL		 1391 return len - desc.count;
1392}
1393
9903cd1c
CL		 1394/**
1395 * xs_tcp_data_ready - "data ready" callback for TCP sockets
1396 * @sk: socket with data to read
9903cd1c
CL		 1397 *
1398 */
676d2369 1399static void xs_tcp_data_ready(struct sock *sk)
a246b010
CL		 1400{
1401 struct rpc_xprt *xprt;
1402 read_descriptor_t rd_desc;
ff2d7db8 1403 int read;
3705ad64 1404 unsigned long total = 0;
a246b010 1405
46121cf7
CL		 1406 dprintk("RPC: xs_tcp_data_ready...\n");
1407
f064af1e 1408 read_lock_bh(&sk->sk_callback_lock);
3705ad64
JL		 1409 if (!(xprt = xprt_from_sock(sk))) {
1410 read = 0;
a246b010 1411 goto out;
3705ad64 1412 }
61d0a8e6
NB		 1413 /* Any data means we had a useful conversation, so
1414 * the we don't need to delay the next reconnect
1415 */
1416 if (xprt->reestablish_timeout)
1417 xprt->reestablish_timeout = 0;
1418
9903cd1c 1419 /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
a246b010 1420 rd_desc.arg.data = xprt;
ff2d7db8
TM		 1421 do {
1422 rd_desc.count = 65536;
1423 read = tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
3705ad64
JL		 1424 if (read > 0)
1425 total += read;
ff2d7db8 1426 } while (read > 0);
a246b010 1427out:
3705ad64 1428 trace_xs_tcp_data_ready(xprt, read, total);
f064af1e 1429 read_unlock_bh(&sk->sk_callback_lock);
a246b010
CL		 1430}
1431
9903cd1c
CL		 1432/**
1433 * xs_tcp_state_change - callback to handle TCP socket state changes
1434 * @sk: socket whose state has changed
1435 *
1436 */
1437static void xs_tcp_state_change(struct sock *sk)
a246b010 1438{
9903cd1c 1439 struct rpc_xprt *xprt;
0fdea1e8 1440 struct sock_xprt *transport;
a246b010 1441
f064af1e 1442 read_lock_bh(&sk->sk_callback_lock);
a246b010
CL		 1443 if (!(xprt = xprt_from_sock(sk)))
1444 goto out;
46121cf7 1445 dprintk("RPC: xs_tcp_state_change client %p...\n", xprt);
669502ff 1446 dprintk("RPC: state %x conn %d dead %d zapped %d sk_shutdown %d\n",
46121cf7
CL		 1447 sk->sk_state, xprt_connected(xprt),
1448 sock_flag(sk, SOCK_DEAD),
669502ff
AC		 1449 sock_flag(sk, SOCK_ZAPPED),
1450 sk->sk_shutdown);
a246b010 1451
0fdea1e8 1452 transport = container_of(xprt, struct sock_xprt, xprt);
40b5ea0c 1453 trace_rpc_socket_state_change(xprt, sk->sk_socket);
a246b010
CL		 1454 switch (sk->sk_state) {
1455 case TCP_ESTABLISHED:
f064af1e 1456 spin_lock(&xprt->transport_lock);
a246b010 1457 if (!xprt_test_and_set_connected(xprt)) {
51971139 1458
a246b010 1459 /* Reset TCP record info */
51971139
CL		 1460 transport->tcp_offset = 0;
1461 transport->tcp_reclen = 0;
1462 transport->tcp_copied = 0;
e136d092
CL		 1463 transport->tcp_flags =
1464 TCP_RCV_COPY_FRAGHDR | TCP_RCV_COPY_XID;
8b71798c 1465 xprt->connect_cookie++;
0fdea1e8
TM		 1466 clear_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
1467 xprt_clear_connecting(xprt);
51971139 1468
2a491991 1469 xprt_wake_pending_tasks(xprt, -EAGAIN);
a246b010 1470 }
f064af1e 1471 spin_unlock(&xprt->transport_lock);
a246b010 1472 break;
3b948ae5
TM		 1473 case TCP_FIN_WAIT1:
1474 /* The client initiated a shutdown of the socket */
7c1d71cf 1475 xprt->connect_cookie++;
663b8858 1476 xprt->reestablish_timeout = 0;
3b948ae5 1477 set_bit(XPRT_CLOSING, &xprt->state);
4e857c58 1478 smp_mb__before_atomic();
3b948ae5 1479 clear_bit(XPRT_CONNECTED, &xprt->state);
ef803670 1480 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
4e857c58 1481 smp_mb__after_atomic();
a246b010 1482 break;
632e3bdc 1483 case TCP_CLOSE_WAIT:
3b948ae5 1484 /* The server initiated a shutdown of the socket */
7c1d71cf 1485 xprt->connect_cookie++;
d0bea455 1486 clear_bit(XPRT_CONNECTED, &xprt->state);
a519fc7a 1487 xs_tcp_force_close(xprt);
663b8858
TM		 1488 case TCP_CLOSING:
1489 /*
1490 * If the server closed down the connection, make sure that
1491 * we back off before reconnecting
1492 */
1493 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1494 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
3b948ae5
TM		 1495 break;
1496 case TCP_LAST_ACK:
670f9457 1497 set_bit(XPRT_CLOSING, &xprt->state);
4e857c58 1498 smp_mb__before_atomic();
3b948ae5 1499 clear_bit(XPRT_CONNECTED, &xprt->state);
4e857c58 1500 smp_mb__after_atomic();
3b948ae5
TM		 1501 break;
1502 case TCP_CLOSE:
0fdea1e8
TM		 1503 if (test_and_clear_bit(XPRT_SOCK_CONNECTING,
1504 &transport->sock_state))
1505 xprt_clear_connecting(xprt);
7d1e8255 1506 xs_sock_mark_closed(xprt);
a246b010
CL		 1507 }
1508 out:
f064af1e 1509 read_unlock_bh(&sk->sk_callback_lock);
a246b010
CL		 1510}
1511
1f0fa154
IJ		 1512static void xs_write_space(struct sock *sk)
1513{
1514 struct socket *sock;
1515 struct rpc_xprt *xprt;
1516
1517 if (unlikely(!(sock = sk->sk_socket)))
1518 return;
1519 clear_bit(SOCK_NOSPACE, &sock->flags);
1520
1521 if (unlikely(!(xprt = xprt_from_sock(sk))))
1522 return;
1523 if (test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags) == 0)
1524 return;
1525
1526 xprt_write_space(xprt);
1527}
1528
9903cd1c 1529/**
c7b2cae8
CL		 1530 * xs_udp_write_space - callback invoked when socket buffer space
1531 * becomes available
9903cd1c
CL		 1532 * @sk: socket whose state has changed
1533 *
a246b010
CL		 1534 * Called when more output buffer space is available for this socket.
1535 * We try not to wake our writers until they can make "significant"
c7b2cae8 1536 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
a246b010
CL		 1537 * with a bunch of small requests.
1538 */
c7b2cae8 1539static void xs_udp_write_space(struct sock *sk)
a246b010 1540{
f064af1e 1541 read_lock_bh(&sk->sk_callback_lock);
a246b010 1542
c7b2cae8 1543 /* from net/core/sock.c:sock_def_write_space */
1f0fa154
IJ		 1544 if (sock_writeable(sk))
1545 xs_write_space(sk);
a246b010 1546
f064af1e 1547 read_unlock_bh(&sk->sk_callback_lock);
c7b2cae8 1548}
a246b010 1549
c7b2cae8
CL		 1550/**
1551 * xs_tcp_write_space - callback invoked when socket buffer space
1552 * becomes available
1553 * @sk: socket whose state has changed
1554 *
1555 * Called when more output buffer space is available for this socket.
1556 * We try not to wake our writers until they can make "significant"
1557 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1558 * with a bunch of small requests.
1559 */
1560static void xs_tcp_write_space(struct sock *sk)
1561{
f064af1e 1562 read_lock_bh(&sk->sk_callback_lock);
c7b2cae8
CL		 1563
1564 /* from net/core/stream.c:sk_stream_write_space */
64dc6130 1565 if (sk_stream_is_writeable(sk))
1f0fa154 1566 xs_write_space(sk);
b6ddf64f 1567
f064af1e 1568 read_unlock_bh(&sk->sk_callback_lock);
a246b010
CL		 1569}
1570
470056c2 1571static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
a246b010 1572{
ee0ac0c2
CL		 1573 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1574 struct sock *sk = transport->inet;
a246b010 1575
7c6e066e 1576 if (transport->rcvsize) {
a246b010 1577 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
7c6e066e 1578 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
a246b010 1579 }
7c6e066e 1580 if (transport->sndsize) {
a246b010 1581 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
7c6e066e 1582 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
a246b010
CL		 1583 sk->sk_write_space(sk);
1584 }
1585}
1586
43118c29 1587/**
470056c2 1588 * xs_udp_set_buffer_size - set send and receive limits
43118c29 1589 * @xprt: generic transport
470056c2
CL		 1590 * @sndsize: requested size of send buffer, in bytes
1591 * @rcvsize: requested size of receive buffer, in bytes
43118c29 1592 *
470056c2 1593 * Set socket send and receive buffer size limits.
43118c29 1594 */
470056c2 1595static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
43118c29 1596{
7c6e066e
CL		 1597 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1598
1599 transport->sndsize = 0;
470056c2 1600 if (sndsize)
7c6e066e
CL		 1601 transport->sndsize = sndsize + 1024;
1602 transport->rcvsize = 0;
470056c2 1603 if (rcvsize)
7c6e066e 1604 transport->rcvsize = rcvsize + 1024;
470056c2
CL		 1605
1606 xs_udp_do_set_buffer_size(xprt);
43118c29
CL		 1607}
1608
46c0ee8b
CL		 1609/**
1610 * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1611 * @task: task that timed out
1612 *
1613 * Adjust the congestion window after a retransmit timeout has occurred.
1614 */
6a24dfb6 1615static void xs_udp_timer(struct rpc_xprt *xprt, struct rpc_task *task)
46c0ee8b 1616{
6a24dfb6 1617 xprt_adjust_cwnd(xprt, task, -ETIMEDOUT);
46c0ee8b
CL		 1618}
1619
b85d8806
CL		 1620static unsigned short xs_get_random_port(void)
1621{
1622 unsigned short range = xprt_max_resvport - xprt_min_resvport;
63862b5b 1623 unsigned short rand = (unsigned short) prandom_u32() % range;
b85d8806
CL		 1624 return rand + xprt_min_resvport;
1625}
1626
4dda9c8a
TM		 1627/**
1628 * xs_set_reuseaddr_port - set the socket's port and address reuse options
1629 * @sock: socket
1630 *
1631 * Note that this function has to be called on all sockets that share the
1632 * same port, and it must be called before binding.
1633 */
1634static void xs_sock_set_reuseport(struct socket *sock)
1635{
402e23b4 1636 int opt = 1;
4dda9c8a 1637
402e23b4
TM		 1638 kernel_setsockopt(sock, SOL_SOCKET, SO_REUSEPORT,
1639 (char *)&opt, sizeof(opt));
4dda9c8a
TM		 1640}
1641
1642static unsigned short xs_sock_getport(struct socket *sock)
1643{
1644 struct sockaddr_storage buf;
1645 int buflen;
1646 unsigned short port = 0;
1647
1648 if (kernel_getsockname(sock, (struct sockaddr *)&buf, &buflen) < 0)
1649 goto out;
1650 switch (buf.ss_family) {
1651 case AF_INET6:
1652 port = ntohs(((struct sockaddr_in6 *)&buf)->sin6_port);
1653 break;
1654 case AF_INET:
1655 port = ntohs(((struct sockaddr_in *)&buf)->sin_port);
1656 }
1657out:
1658 return port;
1659}
1660
92200412
CL		 1661/**
1662 * xs_set_port - reset the port number in the remote endpoint address
1663 * @xprt: generic transport
1664 * @port: new port number
1665 *
1666 */
1667static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1668{
46121cf7 1669 dprintk("RPC: setting port for xprt %p to %u\n", xprt, port);
c4efcb1d 1670
9dc3b095
CL		 1671 rpc_set_port(xs_addr(xprt), port);
1672 xs_update_peer_port(xprt);
92200412
CL		 1673}
1674
4dda9c8a
TM		 1675static void xs_set_srcport(struct sock_xprt *transport, struct socket *sock)
1676{
1677 if (transport->srcport == 0)
1678 transport->srcport = xs_sock_getport(sock);
1679}
1680
5d4ec932 1681static unsigned short xs_get_srcport(struct sock_xprt *transport)
67a391d7 1682{
fbfffbd5 1683 unsigned short port = transport->srcport;
67a391d7
TM		 1684
1685 if (port == 0 && transport->xprt.resvport)
1686 port = xs_get_random_port();
1687 return port;
1688}
1689
baaf4e48 1690static unsigned short xs_next_srcport(struct sock_xprt *transport, unsigned short port)
67a391d7 1691{
fbfffbd5
CL		 1692 if (transport->srcport != 0)
1693 transport->srcport = 0;
67a391d7
TM		 1694 if (!transport->xprt.resvport)
1695 return 0;
1696 if (port <= xprt_min_resvport || port > xprt_max_resvport)
1697 return xprt_max_resvport;
1698 return --port;
1699}
beb59b68 1700static int xs_bind(struct sock_xprt *transport, struct socket *sock)
a246b010 1701{
beb59b68 1702 struct sockaddr_storage myaddr;
67a391d7 1703 int err, nloop = 0;
5d4ec932 1704 unsigned short port = xs_get_srcport(transport);
67a391d7 1705 unsigned short last;
a246b010 1706
0f7a622c
CP		 1707 /*
1708 * If we are asking for any ephemeral port (i.e. port == 0 &&
1709 * transport->xprt.resvport == 0), don't bind. Let the local
1710 * port selection happen implicitly when the socket is used
1711 * (for example at connect time).
1712 *
1713 * This ensures that we can continue to establish TCP
1714 * connections even when all local ephemeral ports are already
1715 * a part of some TCP connection. This makes no difference
1716 * for UDP sockets, but also doens't harm them.
1717 *
1718 * If we're asking for any reserved port (i.e. port == 0 &&
1719 * transport->xprt.resvport == 1) xs_get_srcport above will
1720 * ensure that port is non-zero and we will bind as needed.
1721 */
1722 if (port == 0)
1723 return 0;
1724
beb59b68 1725 memcpy(&myaddr, &transport->srcaddr, transport->xprt.addrlen);
a246b010 1726 do {
beb59b68
PE		 1727 rpc_set_port((struct sockaddr *)&myaddr, port);
1728 err = kernel_bind(sock, (struct sockaddr *)&myaddr,
1729 transport->xprt.addrlen);
a246b010 1730 if (err == 0) {
fbfffbd5 1731 transport->srcport = port;
d3bc9a1d 1732 break;
a246b010 1733 }
67a391d7 1734 last = port;
baaf4e48 1735 port = xs_next_srcport(transport, port);
67a391d7
TM		 1736 if (port > last)
1737 nloop++;
1738 } while (err == -EADDRINUSE && nloop != 2);
90058d37 1739
4232e863 1740 if (myaddr.ss_family == AF_INET)
beb59b68
PE		 1741 dprintk("RPC: %s %pI4:%u: %s (%d)\n", __func__,
1742 &((struct sockaddr_in *)&myaddr)->sin_addr,
1743 port, err ? "failed" : "ok", err);
1744 else
1745 dprintk("RPC: %s %pI6:%u: %s (%d)\n", __func__,
1746 &((struct sockaddr_in6 *)&myaddr)->sin6_addr,
1747 port, err ? "failed" : "ok", err);
a246b010
CL		 1748 return err;
1749}
1750
176e21ee
CL		 1751/*
1752 * We don't support autobind on AF_LOCAL sockets
1753 */
1754static void xs_local_rpcbind(struct rpc_task *task)
1755{
3dc0da27
TM		 1756 rcu_read_lock();
1757 xprt_set_bound(rcu_dereference(task->tk_client->cl_xprt));
1758 rcu_read_unlock();
176e21ee
CL		 1759}
1760
1761static void xs_local_set_port(struct rpc_xprt *xprt, unsigned short port)
1762{
1763}
beb59b68 1764
ed07536e
PZ		 1765#ifdef CONFIG_DEBUG_LOCK_ALLOC
1766static struct lock_class_key xs_key[2];
1767static struct lock_class_key xs_slock_key[2];
1768
176e21ee
CL		 1769static inline void xs_reclassify_socketu(struct socket *sock)
1770{
1771 struct sock *sk = sock->sk;
1772
176e21ee
CL		 1773 sock_lock_init_class_and_name(sk, "slock-AF_LOCAL-RPC",
1774 &xs_slock_key[1], "sk_lock-AF_LOCAL-RPC", &xs_key[1]);
1775}
1776
8945ee5e 1777static inline void xs_reclassify_socket4(struct socket *sock)
ed07536e
PZ		 1778{
1779 struct sock *sk = sock->sk;
8945ee5e 1780
8945ee5e
CL		 1781 sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1782 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1783}
ed07536e 1784
8945ee5e
CL		 1785static inline void xs_reclassify_socket6(struct socket *sock)
1786{
1787 struct sock *sk = sock->sk;
ed07536e 1788
8945ee5e
CL		 1789 sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1790 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
ed07536e 1791}
6bc9638a
PE		 1792
1793static inline void xs_reclassify_socket(int family, struct socket *sock)
1794{
1b7a1819
WAA		 1795 WARN_ON_ONCE(sock_owned_by_user(sock->sk));
1796 if (sock_owned_by_user(sock->sk))
1797 return;
1798
4232e863 1799 switch (family) {
176e21ee
CL		 1800 case AF_LOCAL:
1801 xs_reclassify_socketu(sock);
1802 break;
4232e863 1803 case AF_INET:
6bc9638a 1804 xs_reclassify_socket4(sock);
4232e863
CL		 1805 break;
1806 case AF_INET6:
6bc9638a 1807 xs_reclassify_socket6(sock);
4232e863
CL		 1808 break;
1809 }
6bc9638a 1810}
ed07536e 1811#else
176e21ee
CL		 1812static inline void xs_reclassify_socketu(struct socket *sock)
1813{
1814}
1815
8945ee5e
CL		 1816static inline void xs_reclassify_socket4(struct socket *sock)
1817{
1818}
1819
1820static inline void xs_reclassify_socket6(struct socket *sock)
ed07536e
PZ		 1821{
1822}
6bc9638a
PE		 1823
1824static inline void xs_reclassify_socket(int family, struct socket *sock)
1825{
1826}
ed07536e
PZ		 1827#endif
1828
93dc41bd
N		 1829static void xs_dummy_setup_socket(struct work_struct *work)
1830{
1831}
1832
6bc9638a 1833static struct socket *xs_create_sock(struct rpc_xprt *xprt,
4dda9c8a
TM		 1834 struct sock_xprt *transport, int family, int type,
1835 int protocol, bool reuseport)
22f79326
PE		 1836{
1837 struct socket *sock;
1838 int err;
1839
6bc9638a 1840 err = __sock_create(xprt->xprt_net, family, type, protocol, &sock, 1);
22f79326
PE		 1841 if (err < 0) {
1842 dprintk("RPC: can't create %d transport socket (%d).\n",
1843 protocol, -err);
1844 goto out;
1845 }
6bc9638a 1846 xs_reclassify_socket(family, sock);
22f79326 1847
4dda9c8a
TM		 1848 if (reuseport)
1849 xs_sock_set_reuseport(sock);
1850
4cea288a
BH		 1851 err = xs_bind(transport, sock);
1852 if (err) {
22f79326
PE		 1853 sock_release(sock);
1854 goto out;
1855 }
1856
1857 return sock;
1858out:
1859 return ERR_PTR(err);
1860}
1861
176e21ee
CL		 1862static int xs_local_finish_connecting(struct rpc_xprt *xprt,
1863 struct socket *sock)
1864{
1865 struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1866 xprt);
1867
1868 if (!transport->inet) {
1869 struct sock *sk = sock->sk;
1870
1871 write_lock_bh(&sk->sk_callback_lock);
1872
1873 xs_save_old_callbacks(transport, sk);
1874
1875 sk->sk_user_data = xprt;
1876 sk->sk_data_ready = xs_local_data_ready;
1877 sk->sk_write_space = xs_udp_write_space;
2118071d 1878 sk->sk_error_report = xs_error_report;
c2126157 1879 sk->sk_allocation = GFP_NOIO;
176e21ee
CL		 1880
1881 xprt_clear_connected(xprt);
1882
1883 /* Reset to new socket */
1884 transport->sock = sock;
1885 transport->inet = sk;
1886
1887 write_unlock_bh(&sk->sk_callback_lock);
1888 }
1889
1890 /* Tell the socket layer to start connecting... */
1891 xprt->stat.connect_count++;
1892 xprt->stat.connect_start = jiffies;
1893 return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, 0);
1894}
1895
1896/**
1897 * xs_local_setup_socket - create AF_LOCAL socket, connect to a local endpoint
176e21ee 1898 * @transport: socket transport to connect
176e21ee 1899 */
dc107402 1900static int xs_local_setup_socket(struct sock_xprt *transport)
176e21ee 1901{
176e21ee
CL		 1902 struct rpc_xprt *xprt = &transport->xprt;
1903 struct socket *sock;
1904 int status = -EIO;
1905
176e21ee
CL		 1906 status = __sock_create(xprt->xprt_net, AF_LOCAL,
1907 SOCK_STREAM, 0, &sock, 1);
1908 if (status < 0) {
1909 dprintk("RPC: can't create AF_LOCAL "
1910 "transport socket (%d).\n", -status);
1911 goto out;
1912 }
1913 xs_reclassify_socketu(sock);
1914
1915 dprintk("RPC: worker connecting xprt %p via AF_LOCAL to %s\n",
1916 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1917
1918 status = xs_local_finish_connecting(xprt, sock);
40b5ea0c 1919 trace_rpc_socket_connect(xprt, sock, status);
176e21ee
CL		 1920 switch (status) {
1921 case 0:
1922 dprintk("RPC: xprt %p connected to %s\n",
1923 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1924 xprt_set_connected(xprt);
3601c4a9 1925 case -ENOBUFS:
176e21ee
CL		 1926 break;
1927 case -ENOENT:
1928 dprintk("RPC: xprt %p: socket %s does not exist\n",
1929 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1930 break;
4a20a988
TM		 1931 case -ECONNREFUSED:
1932 dprintk("RPC: xprt %p: connection refused for %s\n",
1933 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1934 break;
176e21ee
CL		 1935 default:
1936 printk(KERN_ERR "%s: unhandled error (%d) connecting to %s\n",
1937 __func__, -status,
1938 xprt->address_strings[RPC_DISPLAY_ADDR]);
1939 }
1940
1941out:
1942 xprt_clear_connecting(xprt);
1943 xprt_wake_pending_tasks(xprt, status);
dc107402
BF		 1944 return status;
1945}
1946
b6669737 1947static void xs_local_connect(struct rpc_xprt *xprt, struct rpc_task *task)
dc107402 1948{
dc107402
BF		 1949 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1950 int ret;
1951
1952 if (RPC_IS_ASYNC(task)) {
1953 /*
1954 * We want the AF_LOCAL connect to be resolved in the
1955 * filesystem namespace of the process making the rpc
1956 * call. Thus we connect synchronously.
1957 *
1958 * If we want to support asynchronous AF_LOCAL calls,
1959 * we'll need to figure out how to pass a namespace to
1960 * connect.
1961 */
1962 rpc_exit(task, -ENOTCONN);
1963 return;
1964 }
1965 ret = xs_local_setup_socket(transport);
1966 if (ret && !RPC_IS_SOFTCONN(task))
1967 msleep_interruptible(15000);
176e21ee
CL		 1968}
1969
3c87ef6e 1970#if IS_ENABLED(CONFIG_SUNRPC_SWAP)
d6e971d8
JL		 1971/*
1972 * Note that this should be called with XPRT_LOCKED held (or when we otherwise
1973 * know that we have exclusive access to the socket), to guard against
1974 * races with xs_reset_transport.
1975 */
a564b8f0
MG		 1976static void xs_set_memalloc(struct rpc_xprt *xprt)
1977{
1978 struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1979 xprt);
1980
d6e971d8
JL		 1981 /*
1982 * If there's no sock, then we have nothing to set. The
1983 * reconnecting process will get it for us.
1984 */
1985 if (!transport->inet)
1986 return;
8e228133 1987 if (atomic_read(&xprt->swapper))
a564b8f0
MG		 1988 sk_set_memalloc(transport->inet);
1989}
1990
1991/**
d67fa4d8 1992 * xs_enable_swap - Tag this transport as being used for swap.
a564b8f0 1993 * @xprt: transport to tag
a564b8f0 1994 *
8e228133
JL		 1995 * Take a reference to this transport on behalf of the rpc_clnt, and
1996 * optionally mark it for swapping if it wasn't already.
a564b8f0 1997 */
d67fa4d8
JL		 1998static int
1999xs_enable_swap(struct rpc_xprt *xprt)
a564b8f0 2000{
d6e971d8 2001 struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
a564b8f0 2002
d6e971d8
JL		 2003 if (atomic_inc_return(&xprt->swapper) != 1)
2004 return 0;
2005 if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE))
2006 return -ERESTARTSYS;
2007 if (xs->inet)
2008 sk_set_memalloc(xs->inet);
2009 xprt_release_xprt(xprt, NULL);
8e228133
JL		 2010 return 0;
2011}
a564b8f0 2012
8e228133 2013/**
d67fa4d8 2014 * xs_disable_swap - Untag this transport as being used for swap.
8e228133
JL		 2015 * @xprt: transport to tag
2016 *
2017 * Drop a "swapper" reference to this xprt on behalf of the rpc_clnt. If the
2018 * swapper refcount goes to 0, untag the socket as a memalloc socket.
2019 */
d67fa4d8
JL		 2020static void
2021xs_disable_swap(struct rpc_xprt *xprt)
8e228133 2022{
d6e971d8 2023 struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
8e228133 2024
d6e971d8
JL		 2025 if (!atomic_dec_and_test(&xprt->swapper))
2026 return;
2027 if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE))
2028 return;
2029 if (xs->inet)
2030 sk_clear_memalloc(xs->inet);
2031 xprt_release_xprt(xprt, NULL);
a564b8f0 2032}
a564b8f0
MG		 2033#else
2034static void xs_set_memalloc(struct rpc_xprt *xprt)
2035{
2036}
d67fa4d8
JL		 2037
2038static int
2039xs_enable_swap(struct rpc_xprt *xprt)
2040{
2041 return -EINVAL;
2042}
2043
2044static void
2045xs_disable_swap(struct rpc_xprt *xprt)
2046{
2047}
a564b8f0
MG		 2048#endif
2049
16be2d20
CL		 2050static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2051{
2052 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2053
2054 if (!transport->inet) {
2055 struct sock *sk = sock->sk;
2056
2057 write_lock_bh(&sk->sk_callback_lock);
2058
2a9e1cfa
TM		 2059 xs_save_old_callbacks(transport, sk);
2060
16be2d20 2061 sk->sk_user_data = xprt;
16be2d20
CL		 2062 sk->sk_data_ready = xs_udp_data_ready;
2063 sk->sk_write_space = xs_udp_write_space;
c2126157 2064 sk->sk_allocation = GFP_NOIO;
16be2d20
CL		 2065
2066 xprt_set_connected(xprt);
2067
2068 /* Reset to new socket */
2069 transport->sock = sock;
2070 transport->inet = sk;
2071
a564b8f0
MG		 2072 xs_set_memalloc(xprt);
2073
16be2d20
CL		 2074 write_unlock_bh(&sk->sk_callback_lock);
2075 }
2076 xs_udp_do_set_buffer_size(xprt);
2077}
2078
8c14ff2a 2079static void xs_udp_setup_socket(struct work_struct *work)
a246b010 2080{
34161db6
TM		 2081 struct sock_xprt *transport =
2082 container_of(work, struct sock_xprt, connect_worker.work);
c8475461 2083 struct rpc_xprt *xprt = &transport->xprt;
ee0ac0c2 2084 struct socket *sock = transport->sock;
b65c0310 2085 int status = -EIO;
9903cd1c 2086
8c14ff2a 2087 sock = xs_create_sock(xprt, transport,
4dda9c8a
TM		 2088 xs_addr(xprt)->sa_family, SOCK_DGRAM,
2089 IPPROTO_UDP, false);
b65c0310 2090 if (IS_ERR(sock))
b0d93ad5 2091 goto out;
68e220bd 2092
c740eff8
CL		 2093 dprintk("RPC: worker connecting xprt %p via %s to "
2094 "%s (port %s)\n", xprt,
2095 xprt->address_strings[RPC_DISPLAY_PROTO],
2096 xprt->address_strings[RPC_DISPLAY_ADDR],
2097 xprt->address_strings[RPC_DISPLAY_PORT]);
68e220bd
CL		 2098
2099 xs_udp_finish_connecting(xprt, sock);
40b5ea0c 2100 trace_rpc_socket_connect(xprt, sock, 0);
b0d93ad5
CL		 2101 status = 0;
2102out:
718ba5b8 2103 xprt_unlock_connect(xprt, transport);
b0d93ad5 2104 xprt_clear_connecting(xprt);
7d1e8255 2105 xprt_wake_pending_tasks(xprt, status);
a246b010
CL		 2106}
2107
4876cc77
TM		 2108/**
2109 * xs_tcp_shutdown - gracefully shut down a TCP socket
2110 * @xprt: transport
2111 *
2112 * Initiates a graceful shutdown of the TCP socket by calling the
2113 * equivalent of shutdown(SHUT_RDWR);
2114 */
2115static void xs_tcp_shutdown(struct rpc_xprt *xprt)
2116{
2117 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2118 struct socket *sock = transport->sock;
2119
2120 if (sock == NULL)
2121 return;
2122 if (xprt_connected(xprt)) {
2123 kernel_sock_shutdown(sock, SHUT_RDWR);
2124 trace_rpc_socket_shutdown(xprt, sock);
2125 } else
2126 xs_reset_transport(transport);
2127}
2128
16be2d20 2129static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
a246b010 2130{
16be2d20 2131 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
fe19a96b 2132 int ret = -ENOTCONN;
edb267a6 2133
ee0ac0c2 2134 if (!transport->inet) {
b0d93ad5 2135 struct sock *sk = sock->sk;
7f260e85
TM		 2136 unsigned int keepidle = xprt->timeout->to_initval / HZ;
2137 unsigned int keepcnt = xprt->timeout->to_retries + 1;
2138 unsigned int opt_on = 1;
775f06ab 2139 unsigned int timeo;
7f260e85
TM		 2140
2141 /* TCP Keepalive options */
2142 kernel_setsockopt(sock, SOL_SOCKET, SO_KEEPALIVE,
2143 (char *)&opt_on, sizeof(opt_on));
2144 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPIDLE,
2145 (char *)&keepidle, sizeof(keepidle));
2146 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPINTVL,
2147 (char *)&keepidle, sizeof(keepidle));
2148 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPCNT,
2149 (char *)&keepcnt, sizeof(keepcnt));
b0d93ad5 2150
775f06ab
TM		 2151 /* TCP user timeout (see RFC5482) */
2152 timeo = jiffies_to_msecs(xprt->timeout->to_initval) *
2153 (xprt->timeout->to_retries + 1);
2154 kernel_setsockopt(sock, SOL_TCP, TCP_USER_TIMEOUT,
2155 (char *)&timeo, sizeof(timeo));
2156
b0d93ad5
CL		 2157 write_lock_bh(&sk->sk_callback_lock);
2158
2a9e1cfa
TM		 2159 xs_save_old_callbacks(transport, sk);
2160
b0d93ad5 2161 sk->sk_user_data = xprt;
b0d93ad5
CL		 2162 sk->sk_data_ready = xs_tcp_data_ready;
2163 sk->sk_state_change = xs_tcp_state_change;
2164 sk->sk_write_space = xs_tcp_write_space;
2118071d 2165 sk->sk_error_report = xs_error_report;
c2126157 2166 sk->sk_allocation = GFP_NOIO;
3167e12c
CL		 2167
2168 /* socket options */
3167e12c 2169 sock_reset_flag(sk, SOCK_LINGER);
3167e12c 2170 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
b0d93ad5
CL		 2171
2172 xprt_clear_connected(xprt);
2173
2174 /* Reset to new socket */
ee0ac0c2
CL		 2175 transport->sock = sock;
2176 transport->inet = sk;
b0d93ad5
CL		 2177
2178 write_unlock_bh(&sk->sk_callback_lock);
2179 }
2180
01d37c42 2181 if (!xprt_bound(xprt))
fe19a96b 2182 goto out;
01d37c42 2183
a564b8f0
MG		 2184 xs_set_memalloc(xprt);
2185
b0d93ad5 2186 /* Tell the socket layer to start connecting... */
262ca07d
CL		 2187 xprt->stat.connect_count++;
2188 xprt->stat.connect_start = jiffies;
0fdea1e8 2189 set_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
fe19a96b
TM		 2190 ret = kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
2191 switch (ret) {
2192 case 0:
4dda9c8a 2193 xs_set_srcport(transport, sock);
fe19a96b
TM		 2194 case -EINPROGRESS:
2195 /* SYN_SENT! */
fe19a96b
TM		 2196 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2197 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2198 }
2199out:
2200 return ret;
16be2d20
CL		 2201}
2202
9903cd1c 2203/**
b61d59ff 2204 * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint
9903cd1c
CL		 2205 *
2206 * Invoked by a work queue tasklet.
a246b010 2207 */
cdd518d5 2208static void xs_tcp_setup_socket(struct work_struct *work)
a246b010 2209{
cdd518d5
PE		 2210 struct sock_xprt *transport =
2211 container_of(work, struct sock_xprt, connect_worker.work);
ee0ac0c2 2212 struct socket *sock = transport->sock;
a9f5f0f7 2213 struct rpc_xprt *xprt = &transport->xprt;
b61d59ff 2214 int status = -EIO;
a246b010 2215
ee0ac0c2 2216 if (!sock) {
cdd518d5 2217 sock = xs_create_sock(xprt, transport,
4dda9c8a
TM		 2218 xs_addr(xprt)->sa_family, SOCK_STREAM,
2219 IPPROTO_TCP, true);
b61d59ff
TM		 2220 if (IS_ERR(sock)) {
2221 status = PTR_ERR(sock);
3167e12c
CL		 2222 goto out;
2223 }
7d1e8255 2224 }
a246b010 2225
c740eff8
CL		 2226 dprintk("RPC: worker connecting xprt %p via %s to "
2227 "%s (port %s)\n", xprt,
2228 xprt->address_strings[RPC_DISPLAY_PROTO],
2229 xprt->address_strings[RPC_DISPLAY_ADDR],
2230 xprt->address_strings[RPC_DISPLAY_PORT]);
edb267a6 2231
16be2d20 2232 status = xs_tcp_finish_connecting(xprt, sock);
40b5ea0c 2233 trace_rpc_socket_connect(xprt, sock, status);
46121cf7
CL		 2234 dprintk("RPC: %p connect status %d connected %d sock state %d\n",
2235 xprt, -status, xprt_connected(xprt),
2236 sock->sk->sk_state);
2a491991 2237 switch (status) {
f75e6745
TM		 2238 default:
2239 printk("%s: connect returned unhandled error %d\n",
2240 __func__, status);
2241 case -EADDRNOTAVAIL:
2242 /* We're probably in TIME_WAIT. Get rid of existing socket,
2243 * and retry
2244 */
a519fc7a 2245 xs_tcp_force_close(xprt);
88b5ed73 2246 break;
2a491991
TM		 2247 case 0:
2248 case -EINPROGRESS:
2249 case -EALREADY:
718ba5b8 2250 xprt_unlock_connect(xprt, transport);
7d1e8255 2251 return;
9fcfe0c8
TM		 2252 case -EINVAL:
2253 /* Happens, for instance, if the user specified a link
2254 * local IPv6 address without a scope-id.
2255 */
3ed5e2a2
TM		 2256 case -ECONNREFUSED:
2257 case -ECONNRESET:
2258 case -ENETUNREACH:
3913c78c 2259 case -EADDRINUSE:
3601c4a9 2260 case -ENOBUFS:
3ed5e2a2 2261 /* retry with existing socket, after a delay */
4efdd92c 2262 xs_tcp_force_close(xprt);
9fcfe0c8 2263 goto out;
a246b010 2264 }
2a491991 2265 status = -EAGAIN;
a246b010 2266out:
718ba5b8 2267 xprt_unlock_connect(xprt, transport);
68e220bd 2268 xprt_clear_connecting(xprt);
7d1e8255 2269 xprt_wake_pending_tasks(xprt, status);
68e220bd 2270}
b0d93ad5 2271
9903cd1c
CL		 2272/**
2273 * xs_connect - connect a socket to a remote endpoint
1b092092 2274 * @xprt: pointer to transport structure
9903cd1c
CL		 2275 * @task: address of RPC task that manages state of connect request
2276 *
2277 * TCP: If the remote end dropped the connection, delay reconnecting.
03bf4b70
CL		 2278 *
2279 * UDP socket connects are synchronous, but we use a work queue anyway
2280 * to guarantee that even unprivileged user processes can set up a
2281 * socket on a privileged port.
2282 *
2283 * If a UDP socket connect fails, the delay behavior here prevents
2284 * retry floods (hard mounts).
9903cd1c 2285 */
1b092092 2286static void xs_connect(struct rpc_xprt *xprt, struct rpc_task *task)
a246b010 2287{
ee0ac0c2 2288 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
a246b010 2289
718ba5b8
TM		 2290 WARN_ON_ONCE(!xprt_lock_connect(xprt, task, transport));
2291
99b1a4c3 2292 if (transport->sock != NULL) {
46121cf7
CL		 2293 dprintk("RPC: xs_connect delayed xprt %p for %lu "
2294 "seconds\n",
03bf4b70 2295 xprt, xprt->reestablish_timeout / HZ);
99b1a4c3
TM		 2296
2297 /* Start by resetting any existing state */
2298 xs_reset_transport(transport);
2299
c1384c9c
TM		 2300 queue_delayed_work(rpciod_workqueue,
2301 &transport->connect_worker,
2302 xprt->reestablish_timeout);
03bf4b70 2303 xprt->reestablish_timeout <<= 1;
61d0a8e6
NB		 2304 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2305 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
03bf4b70
CL		 2306 if (xprt->reestablish_timeout > XS_TCP_MAX_REEST_TO)
2307 xprt->reestablish_timeout = XS_TCP_MAX_REEST_TO;
b0d93ad5 2308 } else {
46121cf7 2309 dprintk("RPC: xs_connect scheduled xprt %p\n", xprt);
c1384c9c
TM		 2310 queue_delayed_work(rpciod_workqueue,
2311 &transport->connect_worker, 0);
a246b010
CL		 2312 }
2313}
2314
176e21ee
CL		 2315/**
2316 * xs_local_print_stats - display AF_LOCAL socket-specifc stats
2317 * @xprt: rpc_xprt struct containing statistics
2318 * @seq: output file
2319 *
2320 */
2321static void xs_local_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2322{
2323 long idle_time = 0;
2324
2325 if (xprt_connected(xprt))
2326 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2327
2328 seq_printf(seq, "\txprt:\tlocal %lu %lu %lu %ld %lu %lu %lu "
15a45206 2329 "%llu %llu %lu %llu %llu\n",
176e21ee
CL		 2330 xprt->stat.bind_count,
2331 xprt->stat.connect_count,
2332 xprt->stat.connect_time,
2333 idle_time,
2334 xprt->stat.sends,
2335 xprt->stat.recvs,
2336 xprt->stat.bad_xids,
2337 xprt->stat.req_u,
15a45206
AA		 2338 xprt->stat.bklog_u,
2339 xprt->stat.max_slots,
2340 xprt->stat.sending_u,
2341 xprt->stat.pending_u);
176e21ee
CL		 2342}
2343
262ca07d
CL		 2344/**
2345 * xs_udp_print_stats - display UDP socket-specifc stats
2346 * @xprt: rpc_xprt struct containing statistics
2347 * @seq: output file
2348 *
2349 */
2350static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2351{
c8475461
CL		 2352 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2353
15a45206
AA		 2354 seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %llu %llu "
2355 "%lu %llu %llu\n",
fbfffbd5 2356 transport->srcport,
262ca07d
CL		 2357 xprt->stat.bind_count,
2358 xprt->stat.sends,
2359 xprt->stat.recvs,
2360 xprt->stat.bad_xids,
2361 xprt->stat.req_u,
15a45206
AA		 2362 xprt->stat.bklog_u,
2363 xprt->stat.max_slots,
2364 xprt->stat.sending_u,
2365 xprt->stat.pending_u);
262ca07d
CL		 2366}
2367
2368/**
2369 * xs_tcp_print_stats - display TCP socket-specifc stats
2370 * @xprt: rpc_xprt struct containing statistics
2371 * @seq: output file
2372 *
2373 */
2374static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2375{
c8475461 2376 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
262ca07d
CL		 2377 long idle_time = 0;
2378
2379 if (xprt_connected(xprt))
2380 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2381
15a45206
AA		 2382 seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu "
2383 "%llu %llu %lu %llu %llu\n",
fbfffbd5 2384 transport->srcport,
262ca07d
CL		 2385 xprt->stat.bind_count,
2386 xprt->stat.connect_count,
2387 xprt->stat.connect_time,
2388 idle_time,
2389 xprt->stat.sends,
2390 xprt->stat.recvs,
2391 xprt->stat.bad_xids,
2392 xprt->stat.req_u,
15a45206
AA		 2393 xprt->stat.bklog_u,
2394 xprt->stat.max_slots,
2395 xprt->stat.sending_u,
2396 xprt->stat.pending_u);
262ca07d
CL		 2397}
2398
4cfc7e60
RI		 2399/*
2400 * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason
2401 * we allocate pages instead doing a kmalloc like rpc_malloc is because we want
2402 * to use the server side send routines.
2403 */
5a51f13a 2404static void *bc_malloc(struct rpc_task *task, size_t size)
4cfc7e60
RI		 2405{
2406 struct page *page;
2407 struct rpc_buffer *buf;
2408
b8a13d03
WAA		 2409 WARN_ON_ONCE(size > PAGE_SIZE - sizeof(struct rpc_buffer));
2410 if (size > PAGE_SIZE - sizeof(struct rpc_buffer))
2411 return NULL;
4cfc7e60 2412
b8a13d03 2413 page = alloc_page(GFP_KERNEL);
4cfc7e60
RI		 2414 if (!page)
2415 return NULL;
2416
2417 buf = page_address(page);
2418 buf->len = PAGE_SIZE;
2419
2420 return buf->data;
2421}
2422
2423/*
2424 * Free the space allocated in the bc_alloc routine
2425 */
5a51f13a 2426static void bc_free(void *buffer)
4cfc7e60
RI		 2427{
2428 struct rpc_buffer *buf;
2429
2430 if (!buffer)
2431 return;
2432
2433 buf = container_of(buffer, struct rpc_buffer, data);
2434 free_page((unsigned long)buf);
2435}
2436
2437/*
2438 * Use the svc_sock to send the callback. Must be called with svsk->sk_mutex
2439 * held. Borrows heavily from svc_tcp_sendto and xs_tcp_send_request.
2440 */
2441static int bc_sendto(struct rpc_rqst *req)
2442{
2443 int len;
2444 struct xdr_buf *xbufp = &req->rq_snd_buf;
2445 struct rpc_xprt *xprt = req->rq_xprt;
2446 struct sock_xprt *transport =
2447 container_of(xprt, struct sock_xprt, xprt);
2448 struct socket *sock = transport->sock;
2449 unsigned long headoff;
2450 unsigned long tailoff;
2451
61677eee 2452 xs_encode_stream_record_marker(xbufp);
4cfc7e60
RI		 2453
2454 tailoff = (unsigned long)xbufp->tail[0].iov_base & ~PAGE_MASK;
2455 headoff = (unsigned long)xbufp->head[0].iov_base & ~PAGE_MASK;
2456 len = svc_send_common(sock, xbufp,
2457 virt_to_page(xbufp->head[0].iov_base), headoff,
2458 xbufp->tail[0].iov_base, tailoff);
2459
2460 if (len != xbufp->len) {
2461 printk(KERN_NOTICE "Error sending entire callback!\n");
2462 len = -EAGAIN;
2463 }
2464
2465 return len;
2466}
2467
2468/*
2469 * The send routine. Borrows from svc_send
2470 */
2471static int bc_send_request(struct rpc_task *task)
2472{
2473 struct rpc_rqst *req = task->tk_rqstp;
2474 struct svc_xprt *xprt;
4cfc7e60
RI		 2475 u32 len;
2476
2477 dprintk("sending request with xid: %08x\n", ntohl(req->rq_xid));
2478 /*
2479 * Get the server socket associated with this callback xprt
2480 */
2481 xprt = req->rq_xprt->bc_xprt;
4cfc7e60
RI		 2482
2483 /*
2484 * Grab the mutex to serialize data as the connection is shared
2485 * with the fore channel
2486 */
2487 if (!mutex_trylock(&xprt->xpt_mutex)) {
2488 rpc_sleep_on(&xprt->xpt_bc_pending, task, NULL);
2489 if (!mutex_trylock(&xprt->xpt_mutex))
2490 return -EAGAIN;
2491 rpc_wake_up_queued_task(&xprt->xpt_bc_pending, task);
2492 }
2493 if (test_bit(XPT_DEAD, &xprt->xpt_flags))
2494 len = -ENOTCONN;
2495 else
2496 len = bc_sendto(req);
2497 mutex_unlock(&xprt->xpt_mutex);
2498
2499 if (len > 0)
2500 len = 0;
2501
2502 return len;
2503}
2504
2505/*
2506 * The close routine. Since this is client initiated, we do nothing
2507 */
2508
2509static void bc_close(struct rpc_xprt *xprt)
2510{
4cfc7e60
RI		 2511}
2512
2513/*
2514 * The xprt destroy routine. Again, because this connection is client
2515 * initiated, we do nothing
2516 */
2517
2518static void bc_destroy(struct rpc_xprt *xprt)
2519{
47f72efa
KM		 2520 dprintk("RPC: bc_destroy xprt %p\n", xprt);
2521
2522 xs_xprt_free(xprt);
2523 module_put(THIS_MODULE);
4cfc7e60
RI		 2524}
2525
176e21ee
CL		 2526static struct rpc_xprt_ops xs_local_ops = {
2527 .reserve_xprt = xprt_reserve_xprt,
2528 .release_xprt = xs_tcp_release_xprt,
f39c1bfb 2529 .alloc_slot = xprt_alloc_slot,
176e21ee
CL		 2530 .rpcbind = xs_local_rpcbind,
2531 .set_port = xs_local_set_port,
dc107402 2532 .connect = xs_local_connect,
176e21ee
CL		 2533 .buf_alloc = rpc_malloc,
2534 .buf_free = rpc_free,
2535 .send_request = xs_local_send_request,
2536 .set_retrans_timeout = xprt_set_retrans_timeout_def,
2537 .close = xs_close,
a1311d87 2538 .destroy = xs_destroy,
176e21ee 2539 .print_stats = xs_local_print_stats,
d67fa4d8
JL		 2540 .enable_swap = xs_enable_swap,
2541 .disable_swap = xs_disable_swap,
176e21ee
CL		 2542};
2543
262965f5 2544static struct rpc_xprt_ops xs_udp_ops = {
43118c29 2545 .set_buffer_size = xs_udp_set_buffer_size,
12a80469 2546 .reserve_xprt = xprt_reserve_xprt_cong,
49e9a890 2547 .release_xprt = xprt_release_xprt_cong,
f39c1bfb 2548 .alloc_slot = xprt_alloc_slot,
45160d62 2549 .rpcbind = rpcb_getport_async,
92200412 2550 .set_port = xs_set_port,
262965f5 2551 .connect = xs_connect,
02107148
CL		 2552 .buf_alloc = rpc_malloc,
2553 .buf_free = rpc_free,
262965f5 2554 .send_request = xs_udp_send_request,
fe3aca29 2555 .set_retrans_timeout = xprt_set_retrans_timeout_rtt,
46c0ee8b 2556 .timer = xs_udp_timer,
a58dd398 2557 .release_request = xprt_release_rqst_cong,
262965f5
CL		 2558 .close = xs_close,
2559 .destroy = xs_destroy,
262ca07d 2560 .print_stats = xs_udp_print_stats,
d67fa4d8
JL		 2561 .enable_swap = xs_enable_swap,
2562 .disable_swap = xs_disable_swap,
4a068258 2563 .inject_disconnect = xs_inject_disconnect,
262965f5
CL		 2564};
2565
2566static struct rpc_xprt_ops xs_tcp_ops = {
12a80469 2567 .reserve_xprt = xprt_reserve_xprt,
e0ab53de 2568 .release_xprt = xs_tcp_release_xprt,
f39c1bfb 2569 .alloc_slot = xprt_lock_and_alloc_slot,
45160d62 2570 .rpcbind = rpcb_getport_async,
92200412 2571 .set_port = xs_set_port,
0b9e7943 2572 .connect = xs_connect,
02107148
CL		 2573 .buf_alloc = rpc_malloc,
2574 .buf_free = rpc_free,
262965f5 2575 .send_request = xs_tcp_send_request,
fe3aca29 2576 .set_retrans_timeout = xprt_set_retrans_timeout_def,
c627d31b 2577 .close = xs_tcp_shutdown,
9903cd1c 2578 .destroy = xs_destroy,
262ca07d 2579 .print_stats = xs_tcp_print_stats,
d67fa4d8
JL		 2580 .enable_swap = xs_enable_swap,
2581 .disable_swap = xs_disable_swap,
4a068258 2582 .inject_disconnect = xs_inject_disconnect,
a246b010
CL		 2583};
2584
4cfc7e60
RI		 2585/*
2586 * The rpc_xprt_ops for the server backchannel
2587 */
2588
2589static struct rpc_xprt_ops bc_tcp_ops = {
2590 .reserve_xprt = xprt_reserve_xprt,
2591 .release_xprt = xprt_release_xprt,
84e28a30 2592 .alloc_slot = xprt_alloc_slot,
4cfc7e60
RI		 2593 .buf_alloc = bc_malloc,
2594 .buf_free = bc_free,
2595 .send_request = bc_send_request,
2596 .set_retrans_timeout = xprt_set_retrans_timeout_def,
2597 .close = bc_close,
2598 .destroy = bc_destroy,
2599 .print_stats = xs_tcp_print_stats,
d67fa4d8
JL		 2600 .enable_swap = xs_enable_swap,
2601 .disable_swap = xs_disable_swap,
4a068258 2602 .inject_disconnect = xs_inject_disconnect,
4cfc7e60
RI		 2603};
2604
92476850
CL		 2605static int xs_init_anyaddr(const int family, struct sockaddr *sap)
2606{
2607 static const struct sockaddr_in sin = {
2608 .sin_family = AF_INET,
2609 .sin_addr.s_addr = htonl(INADDR_ANY),
2610 };
2611 static const struct sockaddr_in6 sin6 = {
2612 .sin6_family = AF_INET6,
2613 .sin6_addr = IN6ADDR_ANY_INIT,
2614 };
2615
2616 switch (family) {
176e21ee
CL		 2617 case AF_LOCAL:
2618 break;
92476850
CL		 2619 case AF_INET:
2620 memcpy(sap, &sin, sizeof(sin));
2621 break;
2622 case AF_INET6:
2623 memcpy(sap, &sin6, sizeof(sin6));
2624 break;
2625 default:
2626 dprintk("RPC: %s: Bad address family\n", __func__);
2627 return -EAFNOSUPPORT;
2628 }
2629 return 0;
2630}
2631
3c341b0b 2632static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
d9ba131d
TM		 2633 unsigned int slot_table_size,
2634 unsigned int max_slot_table_size)
c8541ecd
CL		 2635{
2636 struct rpc_xprt *xprt;
ffc2e518 2637 struct sock_xprt *new;
c8541ecd 2638
96802a09 2639 if (args->addrlen > sizeof(xprt->addr)) {
46121cf7 2640 dprintk("RPC: xs_setup_xprt: address too large\n");
c8541ecd
CL		 2641 return ERR_PTR(-EBADF);
2642 }
2643
d9ba131d
TM		 2644 xprt = xprt_alloc(args->net, sizeof(*new), slot_table_size,
2645 max_slot_table_size);
bd1722d4 2646 if (xprt == NULL) {
46121cf7
CL		 2647 dprintk("RPC: xs_setup_xprt: couldn't allocate "
2648 "rpc_xprt\n");
c8541ecd
CL		 2649 return ERR_PTR(-ENOMEM);
2650 }
c8541ecd 2651
bd1722d4 2652 new = container_of(xprt, struct sock_xprt, xprt);
96802a09
FM		 2653 memcpy(&xprt->addr, args->dstaddr, args->addrlen);
2654 xprt->addrlen = args->addrlen;
d3bc9a1d 2655 if (args->srcaddr)
fbfffbd5 2656 memcpy(&new->srcaddr, args->srcaddr, args->addrlen);
92476850
CL		 2657 else {
2658 int err;
2659 err = xs_init_anyaddr(args->dstaddr->sa_family,
2660 (struct sockaddr *)&new->srcaddr);
2aa13531
SK		 2661 if (err != 0) {
2662 xprt_free(xprt);
92476850 2663 return ERR_PTR(err);
2aa13531 2664 }
92476850 2665 }
c8541ecd
CL		 2666
2667 return xprt;
2668}
2669
176e21ee
CL		 2670static const struct rpc_timeout xs_local_default_timeout = {
2671 .to_initval = 10 * HZ,
2672 .to_maxval = 10 * HZ,
2673 .to_retries = 2,
2674};
2675
2676/**
2677 * xs_setup_local - Set up transport to use an AF_LOCAL socket
2678 * @args: rpc transport creation arguments
2679 *
2680 * AF_LOCAL is a "tpi_cots_ord" transport, just like TCP
2681 */
2682static struct rpc_xprt *xs_setup_local(struct xprt_create *args)
2683{
2684 struct sockaddr_un *sun = (struct sockaddr_un *)args->dstaddr;
2685 struct sock_xprt *transport;
2686 struct rpc_xprt *xprt;
2687 struct rpc_xprt *ret;
2688
d9ba131d
TM		 2689 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2690 xprt_max_tcp_slot_table_entries);
176e21ee
CL		 2691 if (IS_ERR(xprt))
2692 return xprt;
2693 transport = container_of(xprt, struct sock_xprt, xprt);
2694
2695 xprt->prot = 0;
2696 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2697 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2698
2699 xprt->bind_timeout = XS_BIND_TO;
2700 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2701 xprt->idle_timeout = XS_IDLE_DISC_TO;
2702
2703 xprt->ops = &xs_local_ops;
2704 xprt->timeout = &xs_local_default_timeout;
2705
93dc41bd
N		 2706 INIT_DELAYED_WORK(&transport->connect_worker,
2707 xs_dummy_setup_socket);
2708
176e21ee
CL		 2709 switch (sun->sun_family) {
2710 case AF_LOCAL:
2711 if (sun->sun_path[0] != '/') {
2712 dprintk("RPC: bad AF_LOCAL address: %s\n",
2713 sun->sun_path);
2714 ret = ERR_PTR(-EINVAL);
2715 goto out_err;
2716 }
2717 xprt_set_bound(xprt);
176e21ee 2718 xs_format_peer_addresses(xprt, "local", RPCBIND_NETID_LOCAL);
7073ea87
BF		 2719 ret = ERR_PTR(xs_local_setup_socket(transport));
2720 if (ret)
2721 goto out_err;
176e21ee
CL		 2722 break;
2723 default:
2724 ret = ERR_PTR(-EAFNOSUPPORT);
2725 goto out_err;
2726 }
2727
2728 dprintk("RPC: set up xprt to %s via AF_LOCAL\n",
2729 xprt->address_strings[RPC_DISPLAY_ADDR]);
2730
2731 if (try_module_get(THIS_MODULE))
2732 return xprt;
2733 ret = ERR_PTR(-EINVAL);
2734out_err:
315f3812 2735 xs_xprt_free(xprt);
176e21ee
CL		 2736 return ret;
2737}
2738
2881ae74
TM		 2739static const struct rpc_timeout xs_udp_default_timeout = {
2740 .to_initval = 5 * HZ,
2741 .to_maxval = 30 * HZ,
2742 .to_increment = 5 * HZ,
2743 .to_retries = 5,
2744};
2745
9903cd1c
CL		 2746/**
2747 * xs_setup_udp - Set up transport to use a UDP socket
96802a09 2748 * @args: rpc transport creation arguments
9903cd1c
CL		 2749 *
2750 */
483066d6 2751static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
a246b010 2752{
8f9d5b1a 2753 struct sockaddr *addr = args->dstaddr;
c8541ecd 2754 struct rpc_xprt *xprt;
c8475461 2755 struct sock_xprt *transport;
0a68b0be 2756 struct rpc_xprt *ret;
a246b010 2757
d9ba131d
TM		 2758 xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries,
2759 xprt_udp_slot_table_entries);
c8541ecd
CL		 2760 if (IS_ERR(xprt))
2761 return xprt;
c8475461 2762 transport = container_of(xprt, struct sock_xprt, xprt);
a246b010 2763
ec739ef0 2764 xprt->prot = IPPROTO_UDP;
808012fb 2765 xprt->tsh_size = 0;
a246b010
CL		 2766 /* XXX: header size can vary due to auth type, IPv6, etc. */
2767 xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
2768
03bf4b70 2769 xprt->bind_timeout = XS_BIND_TO;
03bf4b70
CL		 2770 xprt->reestablish_timeout = XS_UDP_REEST_TO;
2771 xprt->idle_timeout = XS_IDLE_DISC_TO;
a246b010 2772
262965f5 2773 xprt->ops = &xs_udp_ops;
a246b010 2774
ba7392bb 2775 xprt->timeout = &xs_udp_default_timeout;
a246b010 2776
8f9d5b1a
CL		 2777 switch (addr->sa_family) {
2778 case AF_INET:
2779 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2780 xprt_set_bound(xprt);
2781
2782 INIT_DELAYED_WORK(&transport->connect_worker,
8c14ff2a 2783 xs_udp_setup_socket);
9dc3b095 2784 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
8f9d5b1a
CL		 2785 break;
2786 case AF_INET6:
2787 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2788 xprt_set_bound(xprt);
2789
2790 INIT_DELAYED_WORK(&transport->connect_worker,
8c14ff2a 2791 xs_udp_setup_socket);
9dc3b095 2792 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
8f9d5b1a
CL		 2793 break;
2794 default:
0a68b0be
BF		 2795 ret = ERR_PTR(-EAFNOSUPPORT);
2796 goto out_err;
8f9d5b1a
CL		 2797 }
2798
c740eff8
CL		 2799 if (xprt_bound(xprt))
2800 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2801 xprt->address_strings[RPC_DISPLAY_ADDR],
2802 xprt->address_strings[RPC_DISPLAY_PORT],
2803 xprt->address_strings[RPC_DISPLAY_PROTO]);
2804 else
2805 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
2806 xprt->address_strings[RPC_DISPLAY_ADDR],
2807 xprt->address_strings[RPC_DISPLAY_PROTO]);
edb267a6 2808
bc25571e
TT		 2809 if (try_module_get(THIS_MODULE))
2810 return xprt;
0a68b0be
BF		 2811 ret = ERR_PTR(-EINVAL);
2812out_err:
315f3812 2813 xs_xprt_free(xprt);
0a68b0be 2814 return ret;
a246b010
CL		 2815}
2816
2881ae74
TM		 2817static const struct rpc_timeout xs_tcp_default_timeout = {
2818 .to_initval = 60 * HZ,
2819 .to_maxval = 60 * HZ,
2820 .to_retries = 2,
2821};
2822
9903cd1c
CL		 2823/**
2824 * xs_setup_tcp - Set up transport to use a TCP socket
96802a09 2825 * @args: rpc transport creation arguments
9903cd1c
CL		 2826 *
2827 */
483066d6 2828static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
a246b010 2829{
8f9d5b1a 2830 struct sockaddr *addr = args->dstaddr;
c8541ecd 2831 struct rpc_xprt *xprt;
c8475461 2832 struct sock_xprt *transport;
0a68b0be 2833 struct rpc_xprt *ret;
b7993ceb
TM		 2834 unsigned int max_slot_table_size = xprt_max_tcp_slot_table_entries;
2835
2836 if (args->flags & XPRT_CREATE_INFINITE_SLOTS)
2837 max_slot_table_size = RPC_MAX_SLOT_TABLE_LIMIT;
a246b010 2838
d9ba131d 2839 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
b7993ceb 2840 max_slot_table_size);
c8541ecd
CL		 2841 if (IS_ERR(xprt))
2842 return xprt;
c8475461 2843 transport = container_of(xprt, struct sock_xprt, xprt);
a246b010 2844
ec739ef0 2845 xprt->prot = IPPROTO_TCP;
808012fb 2846 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
808012fb 2847 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
a246b010 2848
03bf4b70 2849 xprt->bind_timeout = XS_BIND_TO;
03bf4b70
CL		 2850 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2851 xprt->idle_timeout = XS_IDLE_DISC_TO;
a246b010 2852
262965f5 2853 xprt->ops = &xs_tcp_ops;
ba7392bb 2854 xprt->timeout = &xs_tcp_default_timeout;
a246b010 2855
8f9d5b1a
CL		 2856 switch (addr->sa_family) {
2857 case AF_INET:
2858 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2859 xprt_set_bound(xprt);
2860
9dc3b095 2861 INIT_DELAYED_WORK(&transport->connect_worker,
cdd518d5 2862 xs_tcp_setup_socket);
9dc3b095 2863 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
8f9d5b1a
CL		 2864 break;
2865 case AF_INET6:
2866 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2867 xprt_set_bound(xprt);
2868
9dc3b095 2869 INIT_DELAYED_WORK(&transport->connect_worker,
cdd518d5 2870 xs_tcp_setup_socket);
9dc3b095 2871 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
8f9d5b1a
CL		 2872 break;
2873 default:
0a68b0be
BF		 2874 ret = ERR_PTR(-EAFNOSUPPORT);
2875 goto out_err;
8f9d5b1a
CL		 2876 }
2877
c740eff8
CL		 2878 if (xprt_bound(xprt))
2879 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2880 xprt->address_strings[RPC_DISPLAY_ADDR],
2881 xprt->address_strings[RPC_DISPLAY_PORT],
2882 xprt->address_strings[RPC_DISPLAY_PROTO]);
2883 else
2884 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
2885 xprt->address_strings[RPC_DISPLAY_ADDR],
2886 xprt->address_strings[RPC_DISPLAY_PROTO]);
2887
bc25571e
TT		 2888 if (try_module_get(THIS_MODULE))
2889 return xprt;
0a68b0be
BF		 2890 ret = ERR_PTR(-EINVAL);
2891out_err:
315f3812 2892 xs_xprt_free(xprt);
0a68b0be 2893 return ret;
a246b010 2894}
282b32e1 2895
f300baba
AB		 2896/**
2897 * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket
2898 * @args: rpc transport creation arguments
2899 *
2900 */
2901static struct rpc_xprt *xs_setup_bc_tcp(struct xprt_create *args)
2902{
2903 struct sockaddr *addr = args->dstaddr;
2904 struct rpc_xprt *xprt;
2905 struct sock_xprt *transport;
2906 struct svc_sock *bc_sock;
0a68b0be 2907 struct rpc_xprt *ret;
f300baba 2908
d9ba131d
TM		 2909 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2910 xprt_tcp_slot_table_entries);
f300baba
AB		 2911 if (IS_ERR(xprt))
2912 return xprt;
2913 transport = container_of(xprt, struct sock_xprt, xprt);
2914
2915 xprt->prot = IPPROTO_TCP;
2916 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2917 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2918 xprt->timeout = &xs_tcp_default_timeout;
2919
2920 /* backchannel */
2921 xprt_set_bound(xprt);
2922 xprt->bind_timeout = 0;
f300baba
AB		 2923 xprt->reestablish_timeout = 0;
2924 xprt->idle_timeout = 0;
2925
f300baba
AB		 2926 xprt->ops = &bc_tcp_ops;
2927
2928 switch (addr->sa_family) {
2929 case AF_INET:
2930 xs_format_peer_addresses(xprt, "tcp",
2931 RPCBIND_NETID_TCP);
2932 break;
2933 case AF_INET6:
2934 xs_format_peer_addresses(xprt, "tcp",
2935 RPCBIND_NETID_TCP6);
2936 break;
2937 default:
0a68b0be
BF		 2938 ret = ERR_PTR(-EAFNOSUPPORT);
2939 goto out_err;
f300baba
AB		 2940 }
2941
50fa0d40
PE		 2942 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2943 xprt->address_strings[RPC_DISPLAY_ADDR],
2944 xprt->address_strings[RPC_DISPLAY_PORT],
2945 xprt->address_strings[RPC_DISPLAY_PROTO]);
f300baba 2946
99de8ea9
BF		 2947 /*
2948 * Once we've associated a backchannel xprt with a connection,
28303ca3
WM		 2949 * we want to keep it around as long as the connection lasts,
2950 * in case we need to start using it for a backchannel again;
2951 * this reference won't be dropped until bc_xprt is destroyed.
99de8ea9
BF		 2952 */
2953 xprt_get(xprt);
2954 args->bc_xprt->xpt_bc_xprt = xprt;
2955 xprt->bc_xprt = args->bc_xprt;
2956 bc_sock = container_of(args->bc_xprt, struct svc_sock, sk_xprt);
2957 transport->sock = bc_sock->sk_sock;
2958 transport->inet = bc_sock->sk_sk;
2959
f300baba
AB		 2960 /*
2961 * Since we don't want connections for the backchannel, we set
2962 * the xprt status to connected
2963 */
2964 xprt_set_connected(xprt);
2965
f300baba
AB		 2966 if (try_module_get(THIS_MODULE))
2967 return xprt;
642aab58
KM		 2968
2969 args->bc_xprt->xpt_bc_xprt = NULL;
99de8ea9 2970 xprt_put(xprt);
0a68b0be
BF		 2971 ret = ERR_PTR(-EINVAL);
2972out_err:
315f3812 2973 xs_xprt_free(xprt);
0a68b0be 2974 return ret;
f300baba
AB		 2975}
2976
176e21ee
CL		 2977static struct xprt_class xs_local_transport = {
2978 .list = LIST_HEAD_INIT(xs_local_transport.list),
2979 .name = "named UNIX socket",
2980 .owner = THIS_MODULE,
2981 .ident = XPRT_TRANSPORT_LOCAL,
2982 .setup = xs_setup_local,
2983};
2984
bc25571e
TT		 2985static struct xprt_class xs_udp_transport = {
2986 .list = LIST_HEAD_INIT(xs_udp_transport.list),
2987 .name = "udp",
2988 .owner = THIS_MODULE,
f300baba 2989 .ident = XPRT_TRANSPORT_UDP,
bc25571e
TT		 2990 .setup = xs_setup_udp,
2991};
2992
2993static struct xprt_class xs_tcp_transport = {
2994 .list = LIST_HEAD_INIT(xs_tcp_transport.list),
2995 .name = "tcp",
2996 .owner = THIS_MODULE,
f300baba 2997 .ident = XPRT_TRANSPORT_TCP,
bc25571e
TT		 2998 .setup = xs_setup_tcp,
2999};
3000
f300baba
AB		 3001static struct xprt_class xs_bc_tcp_transport = {
3002 .list = LIST_HEAD_INIT(xs_bc_tcp_transport.list),
3003 .name = "tcp NFSv4.1 backchannel",
3004 .owner = THIS_MODULE,
3005 .ident = XPRT_TRANSPORT_BC_TCP,
3006 .setup = xs_setup_bc_tcp,
3007};
3008
282b32e1 3009/**
bc25571e 3010 * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
282b32e1
CL		 3011 *
3012 */
3013int init_socket_xprt(void)
3014{
f895b252 3015#if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
2b1bec5f 3016 if (!sunrpc_table_header)
0b4d4147 3017 sunrpc_table_header = register_sysctl_table(sunrpc_table);
fbf76683
CL		 3018#endif
3019
176e21ee 3020 xprt_register_transport(&xs_local_transport);
bc25571e
TT		 3021 xprt_register_transport(&xs_udp_transport);
3022 xprt_register_transport(&xs_tcp_transport);
f300baba 3023 xprt_register_transport(&xs_bc_tcp_transport);
bc25571e 3024
282b32e1
CL		 3025 return 0;
3026}
3027
3028/**
bc25571e 3029 * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
282b32e1
CL		 3030 *
3031 */
3032void cleanup_socket_xprt(void)
3033{
f895b252 3034#if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
fbf76683
CL		 3035 if (sunrpc_table_header) {
3036 unregister_sysctl_table(sunrpc_table_header);
3037 sunrpc_table_header = NULL;
3038 }
3039#endif
bc25571e 3040
176e21ee 3041 xprt_unregister_transport(&xs_local_transport);
bc25571e
TT		 3042 xprt_unregister_transport(&xs_udp_transport);
3043 xprt_unregister_transport(&xs_tcp_transport);
f300baba 3044 xprt_unregister_transport(&xs_bc_tcp_transport);
282b32e1 3045}
cbf11071 3046
9bbb9e5a
RR		 3047static int param_set_uint_minmax(const char *val,
3048 const struct kernel_param *kp,
cbf11071
TM		 3049 unsigned int min, unsigned int max)
3050{
00cfaa94 3051 unsigned int num;
cbf11071
TM		 3052 int ret;
3053
3054 if (!val)
3055 return -EINVAL;
00cfaa94 3056 ret = kstrtouint(val, 0, &num);
cbf11071
TM		 3057 if (ret == -EINVAL || num < min || num > max)
3058 return -EINVAL;
3059 *((unsigned int *)kp->arg) = num;
3060 return 0;
3061}
3062
9bbb9e5a 3063static int param_set_portnr(const char *val, const struct kernel_param *kp)
cbf11071
TM		 3064{
3065 return param_set_uint_minmax(val, kp,
3066 RPC_MIN_RESVPORT,
3067 RPC_MAX_RESVPORT);
3068}
3069
9c27847d 3070static const struct kernel_param_ops param_ops_portnr = {
9bbb9e5a
RR		 3071 .set = param_set_portnr,
3072 .get = param_get_uint,
3073};
3074
cbf11071
TM		 3075#define param_check_portnr(name, p) \
3076 __param_check(name, p, unsigned int);
3077
3078module_param_named(min_resvport, xprt_min_resvport, portnr, 0644);
3079module_param_named(max_resvport, xprt_max_resvport, portnr, 0644);
3080
9bbb9e5a
RR		 3081static int param_set_slot_table_size(const char *val,
3082 const struct kernel_param *kp)
cbf11071
TM		 3083{
3084 return param_set_uint_minmax(val, kp,
3085 RPC_MIN_SLOT_TABLE,
3086 RPC_MAX_SLOT_TABLE);
3087}
3088
9c27847d 3089static const struct kernel_param_ops param_ops_slot_table_size = {
9bbb9e5a
RR		 3090 .set = param_set_slot_table_size,
3091 .get = param_get_uint,
3092};
3093
cbf11071
TM		 3094#define param_check_slot_table_size(name, p) \
3095 __param_check(name, p, unsigned int);
3096
d9ba131d
TM		 3097static int param_set_max_slot_table_size(const char *val,
3098 const struct kernel_param *kp)
3099{
3100 return param_set_uint_minmax(val, kp,
3101 RPC_MIN_SLOT_TABLE,
3102 RPC_MAX_SLOT_TABLE_LIMIT);
3103}
3104
9c27847d 3105static const struct kernel_param_ops param_ops_max_slot_table_size = {
d9ba131d
TM		 3106 .set = param_set_max_slot_table_size,
3107 .get = param_get_uint,
3108};
3109
3110#define param_check_max_slot_table_size(name, p) \
3111 __param_check(name, p, unsigned int);
3112
cbf11071
TM		 3113module_param_named(tcp_slot_table_entries, xprt_tcp_slot_table_entries,
3114 slot_table_size, 0644);
d9ba131d
TM		 3115module_param_named(tcp_max_slot_table_entries, xprt_max_tcp_slot_table_entries,
3116 max_slot_table_size, 0644);
cbf11071
TM		 3117module_param_named(udp_slot_table_entries, xprt_udp_slot_table_entries,
3118 slot_table_size, 0644);
3119
Linux kernel - Deliverables
This page took 1.557625 seconds and 5 git commands to generate.