projects / deliverable / linux.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Merge remote-tracking branch 'iommu/next'
[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 126 .extra1 = &xprt_min_resvport_limit,
e08ea3a9 127 .extra2 = &xprt_max_resvport
fbf76683
CL		 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,
e08ea3a9 135 .extra1 = &xprt_min_resvport,
fbf76683
CL		 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)
03bf4b70
CL		 180
181/*
182 * TCP idle timeout; client drops the transport socket if it is idle
183 * for this long. Note that we also timeout UDP sockets to prevent
184 * holding port numbers when there is no RPC traffic.
185 */
186#define XS_IDLE_DISC_TO (5U * 60 * HZ)
187
f895b252 188#if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
a246b010 189# undef RPC_DEBUG_DATA
9903cd1c 190# define RPCDBG_FACILITY RPCDBG_TRANS
a246b010
CL		 191#endif
192
a246b010 193#ifdef RPC_DEBUG_DATA
9903cd1c 194static void xs_pktdump(char *msg, u32 *packet, unsigned int count)
a246b010 195{
9903cd1c
CL		 196 u8 *buf = (u8 *) packet;
197 int j;
a246b010 198
46121cf7 199 dprintk("RPC: %s\n", msg);
a246b010
CL		 200 for (j = 0; j < count && j < 128; j += 4) {
201 if (!(j & 31)) {
202 if (j)
203 dprintk("\n");
204 dprintk("0x%04x ", j);
205 }
206 dprintk("%02x%02x%02x%02x ",
207 buf[j], buf[j+1], buf[j+2], buf[j+3]);
208 }
209 dprintk("\n");
210}
211#else
9903cd1c 212static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count)
a246b010
CL		 213{
214 /* NOP */
215}
216#endif
217
2118071d
TM		 218static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
219{
220 return (struct rpc_xprt *) sk->sk_user_data;
221}
222
95392c59
CL		 223static inline struct sockaddr *xs_addr(struct rpc_xprt *xprt)
224{
225 return (struct sockaddr *) &xprt->addr;
226}
227
176e21ee
CL		 228static inline struct sockaddr_un *xs_addr_un(struct rpc_xprt *xprt)
229{
230 return (struct sockaddr_un *) &xprt->addr;
231}
232
95392c59 233static inline struct sockaddr_in *xs_addr_in(struct rpc_xprt *xprt)
edb267a6 234{
95392c59
CL		 235 return (struct sockaddr_in *) &xprt->addr;
236}
237
238static inline struct sockaddr_in6 *xs_addr_in6(struct rpc_xprt *xprt)
239{
240 return (struct sockaddr_in6 *) &xprt->addr;
241}
242
c877b849 243static void xs_format_common_peer_addresses(struct rpc_xprt *xprt)
edb267a6 244{
c877b849 245 struct sockaddr *sap = xs_addr(xprt);
9dc3b095
CL		 246 struct sockaddr_in6 *sin6;
247 struct sockaddr_in *sin;
176e21ee 248 struct sockaddr_un *sun;
c877b849 249 char buf[128];
edb267a6 250
9dc3b095 251 switch (sap->sa_family) {
176e21ee
CL		 252 case AF_LOCAL:
253 sun = xs_addr_un(xprt);
254 strlcpy(buf, sun->sun_path, sizeof(buf));
255 xprt->address_strings[RPC_DISPLAY_ADDR] =
256 kstrdup(buf, GFP_KERNEL);
257 break;
9dc3b095 258 case AF_INET:
176e21ee
CL		 259 (void)rpc_ntop(sap, buf, sizeof(buf));
260 xprt->address_strings[RPC_DISPLAY_ADDR] =
261 kstrdup(buf, GFP_KERNEL);
9dc3b095 262 sin = xs_addr_in(xprt);
fc0b5791 263 snprintf(buf, sizeof(buf), "%08x", ntohl(sin->sin_addr.s_addr));
9dc3b095
CL		 264 break;
265 case AF_INET6:
176e21ee
CL		 266 (void)rpc_ntop(sap, buf, sizeof(buf));
267 xprt->address_strings[RPC_DISPLAY_ADDR] =
268 kstrdup(buf, GFP_KERNEL);
9dc3b095 269 sin6 = xs_addr_in6(xprt);
fc0b5791 270 snprintf(buf, sizeof(buf), "%pi6", &sin6->sin6_addr);
9dc3b095
CL		 271 break;
272 default:
273 BUG();
756805e7 274 }
176e21ee 275
9dc3b095 276 xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = kstrdup(buf, GFP_KERNEL);
edb267a6
CL		 277}
278
9dc3b095 279static void xs_format_common_peer_ports(struct rpc_xprt *xprt)
4b6473fb 280{
9dc3b095
CL		 281 struct sockaddr *sap = xs_addr(xprt);
282 char buf[128];
4b6473fb 283
81160e66 284 snprintf(buf, sizeof(buf), "%u", rpc_get_port(sap));
c877b849 285 xprt->address_strings[RPC_DISPLAY_PORT] = kstrdup(buf, GFP_KERNEL);
4b6473fb 286
81160e66 287 snprintf(buf, sizeof(buf), "%4hx", rpc_get_port(sap));
c877b849
CL		 288 xprt->address_strings[RPC_DISPLAY_HEX_PORT] = kstrdup(buf, GFP_KERNEL);
289}
4b6473fb 290
9dc3b095
CL		 291static void xs_format_peer_addresses(struct rpc_xprt *xprt,
292 const char *protocol,
293 const char *netid)
c877b849 294{
b454ae90 295 xprt->address_strings[RPC_DISPLAY_PROTO] = protocol;
b454ae90 296 xprt->address_strings[RPC_DISPLAY_NETID] = netid;
c877b849 297 xs_format_common_peer_addresses(xprt);
9dc3b095 298 xs_format_common_peer_ports(xprt);
edb267a6 299}
4b6473fb 300
9dc3b095 301static void xs_update_peer_port(struct rpc_xprt *xprt)
4b6473fb 302{
9dc3b095
CL		 303 kfree(xprt->address_strings[RPC_DISPLAY_HEX_PORT]);
304 kfree(xprt->address_strings[RPC_DISPLAY_PORT]);
4417c8c4 305
9dc3b095 306 xs_format_common_peer_ports(xprt);
edb267a6
CL		 307}
308
309static void xs_free_peer_addresses(struct rpc_xprt *xprt)
310{
33e01dc7
CL		 311 unsigned int i;
312
313 for (i = 0; i < RPC_DISPLAY_MAX; i++)
314 switch (i) {
315 case RPC_DISPLAY_PROTO:
316 case RPC_DISPLAY_NETID:
317 continue;
318 default:
319 kfree(xprt->address_strings[i]);
320 }
edb267a6
CL		 321}
322
b4b5cc85
CL		 323#define XS_SENDMSG_FLAGS (MSG_DONTWAIT | MSG_NOSIGNAL)
324
24c5684b 325static int xs_send_kvec(struct socket *sock, struct sockaddr *addr, int addrlen, struct kvec *vec, unsigned int base, int more)
b4b5cc85 326{
b4b5cc85
CL		 327 struct msghdr msg = {
328 .msg_name = addr,
329 .msg_namelen = addrlen,
24c5684b
TM		 330 .msg_flags = XS_SENDMSG_FLAGS | (more ? MSG_MORE : 0),
331 };
332 struct kvec iov = {
333 .iov_base = vec->iov_base + base,
334 .iov_len = vec->iov_len - base,
b4b5cc85
CL		 335 };
336
24c5684b 337 if (iov.iov_len != 0)
b4b5cc85
CL		 338 return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
339 return kernel_sendmsg(sock, &msg, NULL, 0, 0);
340}
341
f279cd00 342static int xs_send_pagedata(struct socket *sock, struct xdr_buf *xdr, unsigned int base, int more, bool zerocopy, int *sent_p)
b4b5cc85 343{
a6b31d18
TM		 344 ssize_t (*do_sendpage)(struct socket *sock, struct page *page,
345 int offset, size_t size, int flags);
24c5684b
TM		 346 struct page **ppage;
347 unsigned int remainder;
f279cd00 348 int err;
24c5684b
TM		 349
350 remainder = xdr->page_len - base;
351 base += xdr->page_base;
352 ppage = xdr->pages + (base >> PAGE_SHIFT);
353 base &= ~PAGE_MASK;
a6b31d18
TM		 354 do_sendpage = sock->ops->sendpage;
355 if (!zerocopy)
356 do_sendpage = sock_no_sendpage;
24c5684b
TM		 357 for(;;) {
358 unsigned int len = min_t(unsigned int, PAGE_SIZE - base, remainder);
359 int flags = XS_SENDMSG_FLAGS;
b4b5cc85 360
24c5684b 361 remainder -= len;
31303d6c 362 if (more)
24c5684b 363 flags |= MSG_MORE;
31303d6c
TM		 364 if (remainder != 0)
365 flags |= MSG_SENDPAGE_NOTLAST | MSG_MORE;
a6b31d18 366 err = do_sendpage(sock, *ppage, base, len, flags);
24c5684b
TM		 367 if (remainder == 0 || err != len)
368 break;
f279cd00 369 *sent_p += err;
24c5684b
TM		 370 ppage++;
371 base = 0;
372 }
f279cd00
JB		 373 if (err > 0) {
374 *sent_p += err;
375 err = 0;
376 }
377 return err;
b4b5cc85
CL		 378}
379
9903cd1c
CL		 380/**
381 * xs_sendpages - write pages directly to a socket
382 * @sock: socket to send on
383 * @addr: UDP only -- address of destination
384 * @addrlen: UDP only -- length of destination address
385 * @xdr: buffer containing this request
386 * @base: starting position in the buffer
a6b31d18 387 * @zerocopy: true if it is safe to use sendpage()
f279cd00 388 * @sent_p: return the total number of bytes successfully queued for sending
9903cd1c 389 *
a246b010 390 */
f279cd00 391static int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base, bool zerocopy, int *sent_p)
a246b010 392{
24c5684b 393 unsigned int remainder = xdr->len - base;
f279cd00
JB		 394 int err = 0;
395 int sent = 0;
a246b010 396
262965f5 397 if (unlikely(!sock))
fba91afb 398 return -ENOTSOCK;
262965f5 399
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--;
b6ddf64f
TM		 443}
444
9903cd1c 445/**
262965f5
CL		 446 * xs_nospace - place task on wait queue if transmit was incomplete
447 * @task: task to put to sleep
9903cd1c 448 *
a246b010 449 */
5e3771ce 450static int xs_nospace(struct rpc_task *task)
a246b010 451{
262965f5
CL		 452 struct rpc_rqst *req = task->tk_rqstp;
453 struct rpc_xprt *xprt = req->rq_xprt;
ee0ac0c2 454 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
06ea0bfe 455 struct sock *sk = transport->inet;
24ca9a84 456 int ret = -EAGAIN;
a246b010 457
46121cf7 458 dprintk("RPC: %5u xmit incomplete (%u left of %u)\n",
262965f5
CL		 459 task->tk_pid, req->rq_slen - req->rq_bytes_sent,
460 req->rq_slen);
461
b6ddf64f
TM		 462 /* Protect against races with write_space */
463 spin_lock_bh(&xprt->transport_lock);
464
465 /* Don't race with disconnect */
466 if (xprt_connected(xprt)) {
13331a55
TM		 467 /* wait for more buffer space */
468 sk->sk_write_pending++;
469 xprt_wait_for_buffer_space(task, xs_nospace_callback);
470 } else
5e3771ce 471 ret = -ENOTCONN;
262965f5 472
b6ddf64f 473 spin_unlock_bh(&xprt->transport_lock);
06ea0bfe
TM		 474
475 /* Race breaker in case memory is freed before above code is called */
476 sk->sk_write_space(sk);
5e3771ce 477 return ret;
262965f5
CL		 478}
479
61677eee
CL		 480/*
481 * Construct a stream transport record marker in @buf.
482 */
483static inline void xs_encode_stream_record_marker(struct xdr_buf *buf)
484{
485 u32 reclen = buf->len - sizeof(rpc_fraghdr);
486 rpc_fraghdr *base = buf->head[0].iov_base;
487 *base = cpu_to_be32(RPC_LAST_STREAM_FRAGMENT | reclen);
488}
489
176e21ee
CL		 490/**
491 * xs_local_send_request - write an RPC request to an AF_LOCAL socket
492 * @task: RPC task that manages the state of an RPC request
493 *
494 * Return values:
495 * 0: The request has been sent
496 * EAGAIN: The socket was blocked, please call again later to
497 * complete the request
498 * ENOTCONN: Caller needs to invoke connect logic then call again
499 * other: Some other error occured, the request was not sent
500 */
501static int xs_local_send_request(struct rpc_task *task)
502{
503 struct rpc_rqst *req = task->tk_rqstp;
504 struct rpc_xprt *xprt = req->rq_xprt;
505 struct sock_xprt *transport =
506 container_of(xprt, struct sock_xprt, xprt);
507 struct xdr_buf *xdr = &req->rq_snd_buf;
508 int status;
f279cd00 509 int sent = 0;
176e21ee
CL		 510
511 xs_encode_stream_record_marker(&req->rq_snd_buf);
512
513 xs_pktdump("packet data:",
514 req->rq_svec->iov_base, req->rq_svec->iov_len);
515
f279cd00
JB		 516 status = xs_sendpages(transport->sock, NULL, 0, xdr, req->rq_bytes_sent,
517 true, &sent);
176e21ee
CL		 518 dprintk("RPC: %s(%u) = %d\n",
519 __func__, xdr->len - req->rq_bytes_sent, status);
743c69e7
N		 520
521 if (status == -EAGAIN && sock_writeable(transport->inet))
522 status = -ENOBUFS;
523
f279cd00
JB		 524 if (likely(sent > 0) || status == 0) {
525 req->rq_bytes_sent += sent;
526 req->rq_xmit_bytes_sent += sent;
176e21ee
CL		 527 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
528 req->rq_bytes_sent = 0;
529 return 0;
530 }
531 status = -EAGAIN;
532 }
533
534 switch (status) {
3601c4a9 535 case -ENOBUFS:
b5872f0c 536 break;
176e21ee
CL		 537 case -EAGAIN:
538 status = xs_nospace(task);
539 break;
540 default:
541 dprintk("RPC: sendmsg returned unrecognized error %d\n",
542 -status);
543 case -EPIPE:
544 xs_close(xprt);
545 status = -ENOTCONN;
546 }
547
548 return status;
549}
550
262965f5
CL		 551/**
552 * xs_udp_send_request - write an RPC request to a UDP socket
553 * @task: address of RPC task that manages the state of an RPC request
554 *
555 * Return values:
556 * 0: The request has been sent
557 * EAGAIN: The socket was blocked, please call again later to
558 * complete the request
559 * ENOTCONN: Caller needs to invoke connect logic then call again
25985edc 560 * other: Some other error occurred, the request was not sent
262965f5
CL		 561 */
562static int xs_udp_send_request(struct rpc_task *task)
563{
564 struct rpc_rqst *req = task->tk_rqstp;
565 struct rpc_xprt *xprt = req->rq_xprt;
ee0ac0c2 566 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
262965f5 567 struct xdr_buf *xdr = &req->rq_snd_buf;
f279cd00 568 int sent = 0;
262965f5 569 int status;
a246b010 570
9903cd1c 571 xs_pktdump("packet data:",
a246b010
CL		 572 req->rq_svec->iov_base,
573 req->rq_svec->iov_len);
574
01d37c42
TM		 575 if (!xprt_bound(xprt))
576 return -ENOTCONN;
f279cd00
JB		 577 status = xs_sendpages(transport->sock, xs_addr(xprt), xprt->addrlen,
578 xdr, req->rq_bytes_sent, true, &sent);
a246b010 579
46121cf7 580 dprintk("RPC: xs_udp_send_request(%u) = %d\n",
262965f5 581 xdr->len - req->rq_bytes_sent, status);
a246b010 582
3dedbb5c
JB		 583 /* firewall is blocking us, don't return -EAGAIN or we end up looping */
584 if (status == -EPERM)
585 goto process_status;
586
743c69e7
N		 587 if (status == -EAGAIN && sock_writeable(transport->inet))
588 status = -ENOBUFS;
589
f279cd00
JB		 590 if (sent > 0 || status == 0) {
591 req->rq_xmit_bytes_sent += sent;
592 if (sent >= req->rq_slen)
2199700f
TM		 593 return 0;
594 /* Still some bytes left; set up for a retry later. */
262965f5 595 status = -EAGAIN;
2199700f 596 }
a246b010 597
3dedbb5c 598process_status:
262965f5 599 switch (status) {
fba91afb
TM		 600 case -ENOTSOCK:
601 status = -ENOTCONN;
602 /* Should we call xs_close() here? */
603 break;
b6ddf64f 604 case -EAGAIN:
5e3771ce 605 status = xs_nospace(task);
b6ddf64f 606 break;
262965f5 607 case -ENETUNREACH:
3601c4a9 608 case -ENOBUFS:
262965f5 609 case -EPIPE:
a246b010 610 case -ECONNREFUSED:
3dedbb5c 611 case -EPERM:
a246b010 612 /* When the server has died, an ICMP port unreachable message
9903cd1c 613 * prompts ECONNREFUSED. */
13331a55
TM		 614 break;
615 default:
616 dprintk("RPC: sendmsg returned unrecognized error %d\n",
617 -status);
a246b010 618 }
5fe46e9d 619
262965f5 620 return status;
a246b010
CL		 621}
622
9903cd1c 623/**
262965f5 624 * xs_tcp_send_request - write an RPC request to a TCP socket
9903cd1c
CL		 625 * @task: address of RPC task that manages the state of an RPC request
626 *
627 * Return values:
262965f5
CL		 628 * 0: The request has been sent
629 * EAGAIN: The socket was blocked, please call again later to
630 * complete the request
631 * ENOTCONN: Caller needs to invoke connect logic then call again
25985edc 632 * other: Some other error occurred, the request was not sent
9903cd1c
CL		 633 *
634 * XXX: In the case of soft timeouts, should we eventually give up
262965f5 635 * if sendmsg is not able to make progress?
9903cd1c 636 */
262965f5 637static int xs_tcp_send_request(struct rpc_task *task)
a246b010
CL		 638{
639 struct rpc_rqst *req = task->tk_rqstp;
640 struct rpc_xprt *xprt = req->rq_xprt;
ee0ac0c2 641 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
262965f5 642 struct xdr_buf *xdr = &req->rq_snd_buf;
a6b31d18 643 bool zerocopy = true;
9ffadfbc 644 bool vm_wait = false;
b595bb15 645 int status;
f279cd00 646 int sent;
a246b010 647
61677eee 648 xs_encode_stream_record_marker(&req->rq_snd_buf);
a246b010 649
262965f5
CL		 650 xs_pktdump("packet data:",
651 req->rq_svec->iov_base,
652 req->rq_svec->iov_len);
a6b31d18
TM		 653 /* Don't use zero copy if this is a resend. If the RPC call
654 * completes while the socket holds a reference to the pages,
655 * then we may end up resending corrupted data.
656 */
657 if (task->tk_flags & RPC_TASK_SENT)
658 zerocopy = false;
a246b010
CL		 659
660 /* Continue transmitting the packet/record. We must be careful
661 * to cope with writespace callbacks arriving _after_ we have
262965f5 662 * called sendmsg(). */
a246b010 663 while (1) {
f279cd00
JB		 664 sent = 0;
665 status = xs_sendpages(transport->sock, NULL, 0, xdr,
666 req->rq_bytes_sent, zerocopy, &sent);
a246b010 667
46121cf7 668 dprintk("RPC: xs_tcp_send_request(%u) = %d\n",
262965f5 669 xdr->len - req->rq_bytes_sent, status);
a246b010 670
262965f5
CL		 671 /* If we've sent the entire packet, immediately
672 * reset the count of bytes sent. */
f279cd00
JB		 673 req->rq_bytes_sent += sent;
674 req->rq_xmit_bytes_sent += sent;
262965f5
CL		 675 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
676 req->rq_bytes_sent = 0;
677 return 0;
678 }
a246b010 679
9ffadfbc
TM		 680 WARN_ON_ONCE(sent == 0 && status == 0);
681
682 if (status == -EAGAIN ) {
683 /*
684 * Return EAGAIN if we're sure we're hitting the
685 * socket send buffer limits.
686 */
687 if (test_bit(SOCK_NOSPACE, &transport->sock->flags))
688 break;
689 /*
690 * Did we hit a memory allocation failure?
691 */
692 if (sent == 0) {
693 status = -ENOBUFS;
694 if (vm_wait)
695 break;
696 /* Retry, knowing now that we're below the
697 * socket send buffer limit
698 */
699 vm_wait = true;
700 }
701 continue;
702 }
f580dd04
TM		 703 if (status < 0)
704 break;
9ffadfbc 705 vm_wait = false;
a246b010
CL		 706 }
707
262965f5 708 switch (status) {
fba91afb
TM		 709 case -ENOTSOCK:
710 status = -ENOTCONN;
711 /* Should we call xs_close() here? */
712 break;
262965f5 713 case -EAGAIN:
5e3771ce 714 status = xs_nospace(task);
262965f5 715 break;
262965f5 716 case -ECONNRESET:
2a9e1cfa 717 case -ECONNREFUSED:
262965f5 718 case -ENOTCONN:
3913c78c 719 case -EADDRINUSE:
b5872f0c 720 case -ENOBUFS:
b9d2bb2e 721 case -EPIPE:
13331a55
TM		 722 break;
723 default:
724 dprintk("RPC: sendmsg returned unrecognized error %d\n",
725 -status);
a246b010 726 }
5fe46e9d 727
a246b010
CL		 728 return status;
729}
730
e0ab53de
TM		 731/**
732 * xs_tcp_release_xprt - clean up after a tcp transmission
733 * @xprt: transport
734 * @task: rpc task
735 *
736 * This cleans up if an error causes us to abort the transmission of a request.
737 * In this case, the socket may need to be reset in order to avoid confusing
738 * the server.
739 */
740static void xs_tcp_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
741{
742 struct rpc_rqst *req;
743
744 if (task != xprt->snd_task)
745 return;
746 if (task == NULL)
747 goto out_release;
748 req = task->tk_rqstp;
43cedbf0
TM		 749 if (req == NULL)
750 goto out_release;
e0ab53de
TM		 751 if (req->rq_bytes_sent == 0)
752 goto out_release;
753 if (req->rq_bytes_sent == req->rq_snd_buf.len)
754 goto out_release;
a4f0835c 755 set_bit(XPRT_CLOSE_WAIT, &xprt->state);
e0ab53de
TM		 756out_release:
757 xprt_release_xprt(xprt, task);
758}
759
2a9e1cfa
TM		 760static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk)
761{
762 transport->old_data_ready = sk->sk_data_ready;
763 transport->old_state_change = sk->sk_state_change;
764 transport->old_write_space = sk->sk_write_space;
2118071d 765 transport->old_error_report = sk->sk_error_report;
2a9e1cfa
TM		 766}
767
768static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk)
769{
770 sk->sk_data_ready = transport->old_data_ready;
771 sk->sk_state_change = transport->old_state_change;
772 sk->sk_write_space = transport->old_write_space;
2118071d
TM		 773 sk->sk_error_report = transport->old_error_report;
774}
775
42d42a5b
TM		 776static void xs_sock_reset_state_flags(struct rpc_xprt *xprt)
777{
778 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
779
780 clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state);
781}
782
b70ae915
TM		 783static void xs_sock_reset_connection_flags(struct rpc_xprt *xprt)
784{
785 smp_mb__before_atomic();
786 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
787 clear_bit(XPRT_CLOSING, &xprt->state);
42d42a5b 788 xs_sock_reset_state_flags(xprt);
b70ae915
TM		 789 smp_mb__after_atomic();
790}
791
792static void xs_sock_mark_closed(struct rpc_xprt *xprt)
793{
794 xs_sock_reset_connection_flags(xprt);
795 /* Mark transport as closed and wake up all pending tasks */
796 xprt_disconnect_done(xprt);
797}
798
2118071d
TM		 799/**
800 * xs_error_report - callback to handle TCP socket state errors
801 * @sk: socket
802 *
803 * Note: we don't call sock_error() since there may be a rpc_task
804 * using the socket, and so we don't want to clear sk->sk_err.
805 */
806static void xs_error_report(struct sock *sk)
807{
808 struct rpc_xprt *xprt;
809 int err;
810
811 read_lock_bh(&sk->sk_callback_lock);
812 if (!(xprt = xprt_from_sock(sk)))
813 goto out;
814
815 err = -sk->sk_err;
816 if (err == 0)
817 goto out;
b70ae915
TM		 818 /* Is this a reset event? */
819 if (sk->sk_state == TCP_CLOSE)
820 xs_sock_mark_closed(xprt);
2118071d
TM		 821 dprintk("RPC: xs_error_report client %p, error=%d...\n",
822 xprt, -err);
e8353c76 823 trace_rpc_socket_error(xprt, sk->sk_socket, err);
2118071d
TM		 824 xprt_wake_pending_tasks(xprt, err);
825 out:
826 read_unlock_bh(&sk->sk_callback_lock);
2a9e1cfa
TM		 827}
828
fe315e76 829static void xs_reset_transport(struct sock_xprt *transport)
a246b010 830{
ee0ac0c2
CL		 831 struct socket *sock = transport->sock;
832 struct sock *sk = transport->inet;
6cc7e908 833 struct rpc_xprt *xprt = &transport->xprt;
a246b010 834
fe315e76
CL		 835 if (sk == NULL)
836 return;
9903cd1c 837
264d1df3
JL		 838 if (atomic_read(&transport->xprt.swapper))
839 sk_clear_memalloc(sk);
840
09939204
TM		 841 kernel_sock_shutdown(sock, SHUT_RDWR);
842
edc1b01c 843 mutex_lock(&transport->recv_mutex);
a246b010 844 write_lock_bh(&sk->sk_callback_lock);
ee0ac0c2
CL		 845 transport->inet = NULL;
846 transport->sock = NULL;
a246b010 847
9903cd1c 848 sk->sk_user_data = NULL;
2a9e1cfa
TM		 849
850 xs_restore_old_callbacks(transport, sk);
0c78789e 851 xprt_clear_connected(xprt);
a246b010 852 write_unlock_bh(&sk->sk_callback_lock);
6cc7e908 853 xs_sock_reset_connection_flags(xprt);
edc1b01c 854 mutex_unlock(&transport->recv_mutex);
a246b010 855
6cc7e908 856 trace_rpc_socket_close(xprt, sock);
a246b010 857 sock_release(sock);
fe315e76
CL		 858}
859
860/**
861 * xs_close - close a socket
862 * @xprt: transport
863 *
864 * This is used when all requests are complete; ie, no DRC state remains
865 * on the server we want to save.
f75e6745
TM		 866 *
867 * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with
868 * xs_reset_transport() zeroing the socket from underneath a writer.
fe315e76
CL		 869 */
870static void xs_close(struct rpc_xprt *xprt)
871{
872 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
873
874 dprintk("RPC: xs_close xprt %p\n", xprt);
875
876 xs_reset_transport(transport);
61d0a8e6 877 xprt->reestablish_timeout = 0;
fe315e76 878
62da3b24 879 xprt_disconnect_done(xprt);
a246b010
CL		 880}
881
4a068258
CL		 882static void xs_inject_disconnect(struct rpc_xprt *xprt)
883{
884 dprintk("RPC: injecting transport disconnect on xprt=%p\n",
885 xprt);
886 xprt_disconnect_done(xprt);
887}
888
315f3812
KM		 889static void xs_xprt_free(struct rpc_xprt *xprt)
890{
891 xs_free_peer_addresses(xprt);
892 xprt_free(xprt);
893}
894
9903cd1c
CL		 895/**
896 * xs_destroy - prepare to shutdown a transport
897 * @xprt: doomed transport
898 *
899 */
900static void xs_destroy(struct rpc_xprt *xprt)
a246b010 901{
03c78827
TM		 902 struct sock_xprt *transport = container_of(xprt,
903 struct sock_xprt, xprt);
46121cf7 904 dprintk("RPC: xs_destroy xprt %p\n", xprt);
9903cd1c 905
03c78827 906 cancel_delayed_work_sync(&transport->connect_worker);
a1311d87 907 xs_close(xprt);
edc1b01c 908 cancel_work_sync(&transport->recv_worker);
315f3812 909 xs_xprt_free(xprt);
a1311d87 910 module_put(THIS_MODULE);
a246b010
CL		 911}
912
176e21ee
CL		 913static int xs_local_copy_to_xdr(struct xdr_buf *xdr, struct sk_buff *skb)
914{
915 struct xdr_skb_reader desc = {
916 .skb = skb,
917 .offset = sizeof(rpc_fraghdr),
918 .count = skb->len - sizeof(rpc_fraghdr),
919 };
920
921 if (xdr_partial_copy_from_skb(xdr, 0, &desc, xdr_skb_read_bits) < 0)
922 return -1;
923 if (desc.count)
924 return -1;
925 return 0;
926}
927
928/**
a2648094
TM		 929 * xs_local_data_read_skb
930 * @xprt: transport
931 * @sk: socket
932 * @skb: skbuff
176e21ee
CL		 933 *
934 * Currently this assumes we can read the whole reply in a single gulp.
935 */
a2648094
TM		 936static void xs_local_data_read_skb(struct rpc_xprt *xprt,
937 struct sock *sk,
938 struct sk_buff *skb)
176e21ee
CL		 939{
940 struct rpc_task *task;
176e21ee 941 struct rpc_rqst *rovr;
a2648094 942 int repsize, copied;
176e21ee
CL		 943 u32 _xid;
944 __be32 *xp;
945
176e21ee
CL		 946 repsize = skb->len - sizeof(rpc_fraghdr);
947 if (repsize < 4) {
948 dprintk("RPC: impossible RPC reply size %d\n", repsize);
a2648094 949 return;
176e21ee
CL		 950 }
951
952 /* Copy the XID from the skb... */
953 xp = skb_header_pointer(skb, sizeof(rpc_fraghdr), sizeof(_xid), &_xid);
954 if (xp == NULL)
a2648094 955 return;
176e21ee
CL		 956
957 /* Look up and lock the request corresponding to the given XID */
a2648094 958 spin_lock_bh(&xprt->transport_lock);
176e21ee
CL		 959 rovr = xprt_lookup_rqst(xprt, *xp);
960 if (!rovr)
961 goto out_unlock;
962 task = rovr->rq_task;
963
964 copied = rovr->rq_private_buf.buflen;
965 if (copied > repsize)
966 copied = repsize;
967
968 if (xs_local_copy_to_xdr(&rovr->rq_private_buf, skb)) {
969 dprintk("RPC: sk_buff copy failed\n");
970 goto out_unlock;
971 }
972
973 xprt_complete_rqst(task, copied);
974
975 out_unlock:
a2648094
TM		 976 spin_unlock_bh(&xprt->transport_lock);
977}
978
979static void xs_local_data_receive(struct sock_xprt *transport)
980{
981 struct sk_buff *skb;
982 struct sock *sk;
983 int err;
984
985 mutex_lock(&transport->recv_mutex);
986 sk = transport->inet;
987 if (sk == NULL)
988 goto out;
989 for (;;) {
990 skb = skb_recv_datagram(sk, 0, 1, &err);
42d42a5b
TM		 991 if (skb != NULL) {
992 xs_local_data_read_skb(&transport->xprt, sk, skb);
993 skb_free_datagram(sk, skb);
994 continue;
995 }
996 if (!test_and_clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
a2648094 997 break;
a2648094
TM		 998 }
999out:
1000 mutex_unlock(&transport->recv_mutex);
1001}
1002
1003static void xs_local_data_receive_workfn(struct work_struct *work)
1004{
1005 struct sock_xprt *transport =
1006 container_of(work, struct sock_xprt, recv_worker);
1007 xs_local_data_receive(transport);
176e21ee
CL		 1008}
1009
9903cd1c 1010/**
f9b2ee71
TM		 1011 * xs_udp_data_read_skb - receive callback for UDP sockets
1012 * @xprt: transport
1013 * @sk: socket
1014 * @skb: skbuff
9903cd1c 1015 *
a246b010 1016 */
f9b2ee71
TM		 1017static void xs_udp_data_read_skb(struct rpc_xprt *xprt,
1018 struct sock *sk,
1019 struct sk_buff *skb)
a246b010 1020{
9903cd1c 1021 struct rpc_task *task;
a246b010 1022 struct rpc_rqst *rovr;
f9b2ee71 1023 int repsize, copied;
d8ed029d
AD		 1024 u32 _xid;
1025 __be32 *xp;
a246b010 1026
1da8c681 1027 repsize = skb->len;
a246b010 1028 if (repsize < 4) {
46121cf7 1029 dprintk("RPC: impossible RPC reply size %d!\n", repsize);
f9b2ee71 1030 return;
a246b010
CL		 1031 }
1032
1033 /* Copy the XID from the skb... */
1da8c681 1034 xp = skb_header_pointer(skb, 0, sizeof(_xid), &_xid);
a246b010 1035 if (xp == NULL)
f9b2ee71 1036 return;
a246b010
CL		 1037
1038 /* Look up and lock the request corresponding to the given XID */
f9b2ee71 1039 spin_lock_bh(&xprt->transport_lock);
a246b010
CL		 1040 rovr = xprt_lookup_rqst(xprt, *xp);
1041 if (!rovr)
1042 goto out_unlock;
1043 task = rovr->rq_task;
1044
a246b010
CL		 1045 if ((copied = rovr->rq_private_buf.buflen) > repsize)
1046 copied = repsize;
1047
1048 /* Suck it into the iovec, verify checksum if not done by hw. */
1781f7f5 1049 if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
02c22347 1050 __UDPX_INC_STATS(sk, UDP_MIB_INERRORS);
a246b010 1051 goto out_unlock;
1781f7f5
HX		 1052 }
1053
02c22347 1054 __UDPX_INC_STATS(sk, UDP_MIB_INDATAGRAMS);
a246b010 1055
6a24dfb6 1056 xprt_adjust_cwnd(xprt, task, copied);
1570c1e4 1057 xprt_complete_rqst(task, copied);
a246b010
CL		 1058
1059 out_unlock:
f9b2ee71
TM		 1060 spin_unlock_bh(&xprt->transport_lock);
1061}
1062
1063static void xs_udp_data_receive(struct sock_xprt *transport)
1064{
1065 struct sk_buff *skb;
1066 struct sock *sk;
1067 int err;
1068
1069 mutex_lock(&transport->recv_mutex);
1070 sk = transport->inet;
1071 if (sk == NULL)
1072 goto out;
1073 for (;;) {
1074 skb = skb_recv_datagram(sk, 0, 1, &err);
42d42a5b
TM		 1075 if (skb != NULL) {
1076 xs_udp_data_read_skb(&transport->xprt, sk, skb);
a41bd25a 1077 skb_free_datagram_locked(sk, skb);
42d42a5b
TM		 1078 continue;
1079 }
1080 if (!test_and_clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
f9b2ee71 1081 break;
f9b2ee71
TM		 1082 }
1083out:
1084 mutex_unlock(&transport->recv_mutex);
1085}
1086
1087static void xs_udp_data_receive_workfn(struct work_struct *work)
1088{
1089 struct sock_xprt *transport =
1090 container_of(work, struct sock_xprt, recv_worker);
1091 xs_udp_data_receive(transport);
1092}
1093
1094/**
1095 * xs_data_ready - "data ready" callback for UDP sockets
1096 * @sk: socket with data to read
1097 *
1098 */
1099static void xs_data_ready(struct sock *sk)
1100{
1101 struct rpc_xprt *xprt;
1102
1103 read_lock_bh(&sk->sk_callback_lock);
1104 dprintk("RPC: xs_data_ready...\n");
1105 xprt = xprt_from_sock(sk);
1106 if (xprt != NULL) {
1107 struct sock_xprt *transport = container_of(xprt,
1108 struct sock_xprt, xprt);
5157b956
TM		 1109 transport->old_data_ready(sk);
1110 /* Any data means we had a useful conversation, so
1111 * then we don't need to delay the next reconnect
1112 */
1113 if (xprt->reestablish_timeout)
1114 xprt->reestablish_timeout = 0;
42d42a5b 1115 if (!test_and_set_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
40a5f1b1 1116 queue_work(xprtiod_workqueue, &transport->recv_worker);
f9b2ee71 1117 }
f064af1e 1118 read_unlock_bh(&sk->sk_callback_lock);
a246b010
CL		 1119}
1120
a519fc7a
TM		 1121/*
1122 * Helper function to force a TCP close if the server is sending
1123 * junk and/or it has put us in CLOSE_WAIT
1124 */
1125static void xs_tcp_force_close(struct rpc_xprt *xprt)
1126{
a519fc7a
TM		 1127 xprt_force_disconnect(xprt);
1128}
1129
dd456471 1130static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
a246b010 1131{
51971139 1132 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
a246b010
CL		 1133 size_t len, used;
1134 char *p;
1135
51971139
CL		 1136 p = ((char *) &transport->tcp_fraghdr) + transport->tcp_offset;
1137 len = sizeof(transport->tcp_fraghdr) - transport->tcp_offset;
9d292316 1138 used = xdr_skb_read_bits(desc, p, len);
51971139 1139 transport->tcp_offset += used;
a246b010
CL		 1140 if (used != len)
1141 return;
808012fb 1142
51971139
CL		 1143 transport->tcp_reclen = ntohl(transport->tcp_fraghdr);
1144 if (transport->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
e136d092 1145 transport->tcp_flags |= TCP_RCV_LAST_FRAG;
a246b010 1146 else
e136d092 1147 transport->tcp_flags &= ~TCP_RCV_LAST_FRAG;
51971139 1148 transport->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
808012fb 1149
e136d092 1150 transport->tcp_flags &= ~TCP_RCV_COPY_FRAGHDR;
51971139 1151 transport->tcp_offset = 0;
808012fb 1152
a246b010 1153 /* Sanity check of the record length */
18dca02a 1154 if (unlikely(transport->tcp_reclen < 8)) {
46121cf7 1155 dprintk("RPC: invalid TCP record fragment length\n");
a519fc7a 1156 xs_tcp_force_close(xprt);
9903cd1c 1157 return;
a246b010 1158 }
46121cf7 1159 dprintk("RPC: reading TCP record fragment of length %d\n",
51971139 1160 transport->tcp_reclen);
a246b010
CL		 1161}
1162
51971139 1163static void xs_tcp_check_fraghdr(struct sock_xprt *transport)
a246b010 1164{
51971139 1165 if (transport->tcp_offset == transport->tcp_reclen) {
e136d092 1166 transport->tcp_flags |= TCP_RCV_COPY_FRAGHDR;
51971139 1167 transport->tcp_offset = 0;
e136d092
CL		 1168 if (transport->tcp_flags & TCP_RCV_LAST_FRAG) {
1169 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1170 transport->tcp_flags |= TCP_RCV_COPY_XID;
51971139 1171 transport->tcp_copied = 0;
a246b010
CL		 1172 }
1173 }
1174}
1175
dd456471 1176static inline void xs_tcp_read_xid(struct sock_xprt *transport, struct xdr_skb_reader *desc)
a246b010
CL		 1177{
1178 size_t len, used;
1179 char *p;
1180
51971139 1181 len = sizeof(transport->tcp_xid) - transport->tcp_offset;
46121cf7 1182 dprintk("RPC: reading XID (%Zu bytes)\n", len);
51971139 1183 p = ((char *) &transport->tcp_xid) + transport->tcp_offset;
9d292316 1184 used = xdr_skb_read_bits(desc, p, len);
51971139 1185 transport->tcp_offset += used;
a246b010
CL		 1186 if (used != len)
1187 return;
e136d092 1188 transport->tcp_flags &= ~TCP_RCV_COPY_XID;
f4a2e418 1189 transport->tcp_flags |= TCP_RCV_READ_CALLDIR;
51971139 1190 transport->tcp_copied = 4;
18dca02a
RL		 1191 dprintk("RPC: reading %s XID %08x\n",
1192 (transport->tcp_flags & TCP_RPC_REPLY) ? "reply for"
1193 : "request with",
51971139
CL		 1194 ntohl(transport->tcp_xid));
1195 xs_tcp_check_fraghdr(transport);
a246b010
CL		 1196}
1197
18dca02a
RL		 1198static inline void xs_tcp_read_calldir(struct sock_xprt *transport,
1199 struct xdr_skb_reader *desc)
a246b010 1200{
18dca02a
RL		 1201 size_t len, used;
1202 u32 offset;
b76ce561 1203 char *p;
18dca02a
RL		 1204
1205 /*
1206 * We want transport->tcp_offset to be 8 at the end of this routine
1207 * (4 bytes for the xid and 4 bytes for the call/reply flag).
1208 * When this function is called for the first time,
1209 * transport->tcp_offset is 4 (after having already read the xid).
1210 */
1211 offset = transport->tcp_offset - sizeof(transport->tcp_xid);
b76ce561 1212 len = sizeof(transport->tcp_calldir) - offset;
18dca02a 1213 dprintk("RPC: reading CALL/REPLY flag (%Zu bytes)\n", len);
b76ce561
TM		 1214 p = ((char *) &transport->tcp_calldir) + offset;
1215 used = xdr_skb_read_bits(desc, p, len);
18dca02a
RL		 1216 transport->tcp_offset += used;
1217 if (used != len)
1218 return;
f4a2e418 1219 transport->tcp_flags &= ~TCP_RCV_READ_CALLDIR;
f4a2e418
RL		 1220 /*
1221 * We don't yet have the XDR buffer, so we will write the calldir
1222 * out after we get the buffer from the 'struct rpc_rqst'
1223 */
b76ce561
TM		 1224 switch (ntohl(transport->tcp_calldir)) {
1225 case RPC_REPLY:
1226 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1227 transport->tcp_flags |= TCP_RCV_COPY_DATA;
18dca02a 1228 transport->tcp_flags |= TCP_RPC_REPLY;
b76ce561
TM		 1229 break;
1230 case RPC_CALL:
1231 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1232 transport->tcp_flags |= TCP_RCV_COPY_DATA;
18dca02a 1233 transport->tcp_flags &= ~TCP_RPC_REPLY;
b76ce561
TM		 1234 break;
1235 default:
1236 dprintk("RPC: invalid request message type\n");
a519fc7a 1237 xs_tcp_force_close(&transport->xprt);
b76ce561 1238 }
18dca02a
RL		 1239 xs_tcp_check_fraghdr(transport);
1240}
1241
44b98efd
RL		 1242static inline void xs_tcp_read_common(struct rpc_xprt *xprt,
1243 struct xdr_skb_reader *desc,
1244 struct rpc_rqst *req)
a246b010 1245{
44b98efd
RL		 1246 struct sock_xprt *transport =
1247 container_of(xprt, struct sock_xprt, xprt);
a246b010
CL		 1248 struct xdr_buf *rcvbuf;
1249 size_t len;
1250 ssize_t r;
1251
a246b010 1252 rcvbuf = &req->rq_private_buf;
f4a2e418
RL		 1253
1254 if (transport->tcp_flags & TCP_RCV_COPY_CALLDIR) {
1255 /*
1256 * Save the RPC direction in the XDR buffer
1257 */
f4a2e418 1258 memcpy(rcvbuf->head[0].iov_base + transport->tcp_copied,
b76ce561
TM		 1259 &transport->tcp_calldir,
1260 sizeof(transport->tcp_calldir));
1261 transport->tcp_copied += sizeof(transport->tcp_calldir);
f4a2e418 1262 transport->tcp_flags &= ~TCP_RCV_COPY_CALLDIR;
a246b010
CL		 1263 }
1264
a246b010 1265 len = desc->count;
51971139 1266 if (len > transport->tcp_reclen - transport->tcp_offset) {
dd456471 1267 struct xdr_skb_reader my_desc;
a246b010 1268
51971139 1269 len = transport->tcp_reclen - transport->tcp_offset;
a246b010
CL		 1270 memcpy(&my_desc, desc, sizeof(my_desc));
1271 my_desc.count = len;
51971139 1272 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
9d292316 1273 &my_desc, xdr_skb_read_bits);
a246b010
CL		 1274 desc->count -= r;
1275 desc->offset += r;
1276 } else
51971139 1277 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
9d292316 1278 desc, xdr_skb_read_bits);
a246b010
CL		 1279
1280 if (r > 0) {
51971139
CL		 1281 transport->tcp_copied += r;
1282 transport->tcp_offset += r;
a246b010
CL		 1283 }
1284 if (r != len) {
1285 /* Error when copying to the receive buffer,
1286 * usually because we weren't able to allocate
1287 * additional buffer pages. All we can do now
e136d092 1288 * is turn off TCP_RCV_COPY_DATA, so the request
a246b010
CL		 1289 * will not receive any additional updates,
1290 * and time out.
1291 * Any remaining data from this record will
1292 * be discarded.
1293 */
e136d092 1294 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
46121cf7 1295 dprintk("RPC: XID %08x truncated request\n",
51971139 1296 ntohl(transport->tcp_xid));
46121cf7
CL		 1297 dprintk("RPC: xprt = %p, tcp_copied = %lu, "
1298 "tcp_offset = %u, tcp_reclen = %u\n",
1299 xprt, transport->tcp_copied,
1300 transport->tcp_offset, transport->tcp_reclen);
44b98efd 1301 return;
a246b010
CL		 1302 }
1303
46121cf7 1304 dprintk("RPC: XID %08x read %Zd bytes\n",
51971139 1305 ntohl(transport->tcp_xid), r);
46121cf7
CL		 1306 dprintk("RPC: xprt = %p, tcp_copied = %lu, tcp_offset = %u, "
1307 "tcp_reclen = %u\n", xprt, transport->tcp_copied,
1308 transport->tcp_offset, transport->tcp_reclen);
51971139
CL		 1309
1310 if (transport->tcp_copied == req->rq_private_buf.buflen)
e136d092 1311 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
51971139 1312 else if (transport->tcp_offset == transport->tcp_reclen) {
e136d092
CL		 1313 if (transport->tcp_flags & TCP_RCV_LAST_FRAG)
1314 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
a246b010 1315 }
44b98efd
RL		 1316}
1317
1318/*
1319 * Finds the request corresponding to the RPC xid and invokes the common
1320 * tcp read code to read the data.
1321 */
1322static inline int xs_tcp_read_reply(struct rpc_xprt *xprt,
1323 struct xdr_skb_reader *desc)
1324{
1325 struct sock_xprt *transport =
1326 container_of(xprt, struct sock_xprt, xprt);
1327 struct rpc_rqst *req;
1328
1329 dprintk("RPC: read reply XID %08x\n", ntohl(transport->tcp_xid));
1330
1331 /* Find and lock the request corresponding to this xid */
edc1b01c 1332 spin_lock_bh(&xprt->transport_lock);
44b98efd
RL		 1333 req = xprt_lookup_rqst(xprt, transport->tcp_xid);
1334 if (!req) {
1335 dprintk("RPC: XID %08x request not found!\n",
1336 ntohl(transport->tcp_xid));
edc1b01c 1337 spin_unlock_bh(&xprt->transport_lock);
44b98efd
RL		 1338 return -1;
1339 }
1340
1341 xs_tcp_read_common(xprt, desc, req);
1342
e136d092 1343 if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
51971139 1344 xprt_complete_rqst(req->rq_task, transport->tcp_copied);
44b98efd 1345
edc1b01c 1346 spin_unlock_bh(&xprt->transport_lock);
44b98efd
RL		 1347 return 0;
1348}
1349
9e00abc3 1350#if defined(CONFIG_SUNRPC_BACKCHANNEL)
44b98efd
RL		 1351/*
1352 * Obtains an rpc_rqst previously allocated and invokes the common
1353 * tcp read code to read the data. The result is placed in the callback
1354 * queue.
1355 * If we're unable to obtain the rpc_rqst we schedule the closing of the
1356 * connection and return -1.
1357 */
2ea24497 1358static int xs_tcp_read_callback(struct rpc_xprt *xprt,
44b98efd
RL		 1359 struct xdr_skb_reader *desc)
1360{
1361 struct sock_xprt *transport =
1362 container_of(xprt, struct sock_xprt, xprt);
1363 struct rpc_rqst *req;
1364
2ea24497 1365 /* Look up and lock the request corresponding to the given XID */
edc1b01c 1366 spin_lock_bh(&xprt->transport_lock);
2ea24497 1367 req = xprt_lookup_bc_request(xprt, transport->tcp_xid);
44b98efd 1368 if (req == NULL) {
edc1b01c 1369 spin_unlock_bh(&xprt->transport_lock);
44b98efd
RL		 1370 printk(KERN_WARNING "Callback slot table overflowed\n");
1371 xprt_force_disconnect(xprt);
1372 return -1;
1373 }
1374
44b98efd
RL		 1375 dprintk("RPC: read callback XID %08x\n", ntohl(req->rq_xid));
1376 xs_tcp_read_common(xprt, desc, req);
1377
2ea24497
TM		 1378 if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1379 xprt_complete_bc_request(req, transport->tcp_copied);
edc1b01c 1380 spin_unlock_bh(&xprt->transport_lock);
44b98efd
RL		 1381
1382 return 0;
1383}
1384
1385static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1386 struct xdr_skb_reader *desc)
1387{
1388 struct sock_xprt *transport =
1389 container_of(xprt, struct sock_xprt, xprt);
1390
1391 return (transport->tcp_flags & TCP_RPC_REPLY) ?
1392 xs_tcp_read_reply(xprt, desc) :
1393 xs_tcp_read_callback(xprt, desc);
1394}
76566773
CL		 1395
1396static int xs_tcp_bc_up(struct svc_serv *serv, struct net *net)
1397{
1398 int ret;
1399
1400 ret = svc_create_xprt(serv, "tcp-bc", net, PF_INET, 0,
1401 SVC_SOCK_ANONYMOUS);
1402 if (ret < 0)
1403 return ret;
1404 return 0;
1405}
6b26cc8c
CL		 1406
1407static size_t xs_tcp_bc_maxpayload(struct rpc_xprt *xprt)
1408{
1409 return PAGE_SIZE;
1410}
44b98efd
RL		 1411#else
1412static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1413 struct xdr_skb_reader *desc)
1414{
1415 return xs_tcp_read_reply(xprt, desc);
1416}
9e00abc3 1417#endif /* CONFIG_SUNRPC_BACKCHANNEL */
44b98efd
RL		 1418
1419/*
1420 * Read data off the transport. This can be either an RPC_CALL or an
1421 * RPC_REPLY. Relay the processing to helper functions.
1422 */
1423static void xs_tcp_read_data(struct rpc_xprt *xprt,
1424 struct xdr_skb_reader *desc)
1425{
1426 struct sock_xprt *transport =
1427 container_of(xprt, struct sock_xprt, xprt);
1428
1429 if (_xs_tcp_read_data(xprt, desc) == 0)
1430 xs_tcp_check_fraghdr(transport);
1431 else {
1432 /*
1433 * The transport_lock protects the request handling.
1434 * There's no need to hold it to update the tcp_flags.
1435 */
1436 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1437 }
a246b010
CL		 1438}
1439
dd456471 1440static inline void xs_tcp_read_discard(struct sock_xprt *transport, struct xdr_skb_reader *desc)
a246b010
CL		 1441{
1442 size_t len;
1443
51971139 1444 len = transport->tcp_reclen - transport->tcp_offset;
a246b010
CL		 1445 if (len > desc->count)
1446 len = desc->count;
1447 desc->count -= len;
1448 desc->offset += len;
51971139 1449 transport->tcp_offset += len;
46121cf7 1450 dprintk("RPC: discarded %Zu bytes\n", len);
51971139 1451 xs_tcp_check_fraghdr(transport);
a246b010
CL		 1452}
1453
9903cd1c 1454static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
a246b010
CL		 1455{
1456 struct rpc_xprt *xprt = rd_desc->arg.data;
51971139 1457 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
dd456471 1458 struct xdr_skb_reader desc = {
a246b010
CL		 1459 .skb = skb,
1460 .offset = offset,
1461 .count = len,
9903cd1c 1462 };
a246b010 1463
46121cf7 1464 dprintk("RPC: xs_tcp_data_recv started\n");
a246b010 1465 do {
1a867a08 1466 trace_xs_tcp_data_recv(transport);
a246b010
CL		 1467 /* Read in a new fragment marker if necessary */
1468 /* Can we ever really expect to get completely empty fragments? */
e136d092 1469 if (transport->tcp_flags & TCP_RCV_COPY_FRAGHDR) {
9903cd1c 1470 xs_tcp_read_fraghdr(xprt, &desc);
a246b010
CL		 1471 continue;
1472 }
1473 /* Read in the xid if necessary */
e136d092 1474 if (transport->tcp_flags & TCP_RCV_COPY_XID) {
51971139 1475 xs_tcp_read_xid(transport, &desc);
a246b010
CL		 1476 continue;
1477 }
18dca02a 1478 /* Read in the call/reply flag */
f4a2e418 1479 if (transport->tcp_flags & TCP_RCV_READ_CALLDIR) {
18dca02a
RL		 1480 xs_tcp_read_calldir(transport, &desc);
1481 continue;
1482 }
a246b010 1483 /* Read in the request data */
e136d092 1484 if (transport->tcp_flags & TCP_RCV_COPY_DATA) {
44b98efd 1485 xs_tcp_read_data(xprt, &desc);
a246b010
CL		 1486 continue;
1487 }
1488 /* Skip over any trailing bytes on short reads */
51971139 1489 xs_tcp_read_discard(transport, &desc);
a246b010 1490 } while (desc.count);
1a867a08 1491 trace_xs_tcp_data_recv(transport);
46121cf7 1492 dprintk("RPC: xs_tcp_data_recv done\n");
a246b010
CL		 1493 return len - desc.count;
1494}
1495
66d7a56a
TM		 1496static void xs_tcp_data_receive(struct sock_xprt *transport)
1497{
1498 struct rpc_xprt *xprt = &transport->xprt;
1499 struct sock *sk;
1500 read_descriptor_t rd_desc = {
1501 .count = 2*1024*1024,
1502 .arg.data = xprt,
1503 };
1504 unsigned long total = 0;
1505 int read = 0;
1506
edc1b01c 1507 mutex_lock(&transport->recv_mutex);
66d7a56a 1508 sk = transport->inet;
edc1b01c
TM		 1509 if (sk == NULL)
1510 goto out;
66d7a56a
TM		 1511
1512 /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
1513 for (;;) {
edc1b01c 1514 lock_sock(sk);
66d7a56a 1515 read = tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
42d42a5b
TM		 1516 if (read <= 0) {
1517 clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state);
1518 release_sock(sk);
1519 if (!test_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
1520 break;
1521 } else {
1522 release_sock(sk);
1523 total += read;
1524 }
66d7a56a
TM		 1525 rd_desc.count = 65536;
1526 }
edc1b01c
TM		 1527out:
1528 mutex_unlock(&transport->recv_mutex);
66d7a56a
TM		 1529 trace_xs_tcp_data_ready(xprt, read, total);
1530}
1531
edc1b01c
TM		 1532static void xs_tcp_data_receive_workfn(struct work_struct *work)
1533{
1534 struct sock_xprt *transport =
1535 container_of(work, struct sock_xprt, recv_worker);
1536 xs_tcp_data_receive(transport);
1537}
1538
9903cd1c
CL		 1539/**
1540 * xs_tcp_state_change - callback to handle TCP socket state changes
1541 * @sk: socket whose state has changed
1542 *
1543 */
1544static void xs_tcp_state_change(struct sock *sk)
a246b010 1545{
9903cd1c 1546 struct rpc_xprt *xprt;
0fdea1e8 1547 struct sock_xprt *transport;
a246b010 1548
f064af1e 1549 read_lock_bh(&sk->sk_callback_lock);
a246b010
CL		 1550 if (!(xprt = xprt_from_sock(sk)))
1551 goto out;
46121cf7 1552 dprintk("RPC: xs_tcp_state_change client %p...\n", xprt);
669502ff 1553 dprintk("RPC: state %x conn %d dead %d zapped %d sk_shutdown %d\n",
46121cf7
CL		 1554 sk->sk_state, xprt_connected(xprt),
1555 sock_flag(sk, SOCK_DEAD),
669502ff
AC		 1556 sock_flag(sk, SOCK_ZAPPED),
1557 sk->sk_shutdown);
a246b010 1558
0fdea1e8 1559 transport = container_of(xprt, struct sock_xprt, xprt);
40b5ea0c 1560 trace_rpc_socket_state_change(xprt, sk->sk_socket);
a246b010
CL		 1561 switch (sk->sk_state) {
1562 case TCP_ESTABLISHED:
f064af1e 1563 spin_lock(&xprt->transport_lock);
a246b010 1564 if (!xprt_test_and_set_connected(xprt)) {
51971139 1565
a246b010 1566 /* Reset TCP record info */
51971139
CL		 1567 transport->tcp_offset = 0;
1568 transport->tcp_reclen = 0;
1569 transport->tcp_copied = 0;
e136d092
CL		 1570 transport->tcp_flags =
1571 TCP_RCV_COPY_FRAGHDR | TCP_RCV_COPY_XID;
8b71798c 1572 xprt->connect_cookie++;
0fdea1e8
TM		 1573 clear_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
1574 xprt_clear_connecting(xprt);
51971139 1575
2a491991 1576 xprt_wake_pending_tasks(xprt, -EAGAIN);
a246b010 1577 }
f064af1e 1578 spin_unlock(&xprt->transport_lock);
a246b010 1579 break;
3b948ae5
TM		 1580 case TCP_FIN_WAIT1:
1581 /* The client initiated a shutdown of the socket */
7c1d71cf 1582 xprt->connect_cookie++;
663b8858 1583 xprt->reestablish_timeout = 0;
3b948ae5 1584 set_bit(XPRT_CLOSING, &xprt->state);
4e857c58 1585 smp_mb__before_atomic();
3b948ae5 1586 clear_bit(XPRT_CONNECTED, &xprt->state);
ef803670 1587 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
4e857c58 1588 smp_mb__after_atomic();
a246b010 1589 break;
632e3bdc 1590 case TCP_CLOSE_WAIT:
3b948ae5 1591 /* The server initiated a shutdown of the socket */
7c1d71cf 1592 xprt->connect_cookie++;
d0bea455 1593 clear_bit(XPRT_CONNECTED, &xprt->state);
a519fc7a 1594 xs_tcp_force_close(xprt);
663b8858
TM		 1595 case TCP_CLOSING:
1596 /*
1597 * If the server closed down the connection, make sure that
1598 * we back off before reconnecting
1599 */
1600 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1601 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
3b948ae5
TM		 1602 break;
1603 case TCP_LAST_ACK:
670f9457 1604 set_bit(XPRT_CLOSING, &xprt->state);
4e857c58 1605 smp_mb__before_atomic();
3b948ae5 1606 clear_bit(XPRT_CONNECTED, &xprt->state);
4e857c58 1607 smp_mb__after_atomic();
3b948ae5
TM		 1608 break;
1609 case TCP_CLOSE:
0fdea1e8
TM		 1610 if (test_and_clear_bit(XPRT_SOCK_CONNECTING,
1611 &transport->sock_state))
1612 xprt_clear_connecting(xprt);
7d1e8255 1613 xs_sock_mark_closed(xprt);
a246b010
CL		 1614 }
1615 out:
f064af1e 1616 read_unlock_bh(&sk->sk_callback_lock);
a246b010
CL		 1617}
1618
1f0fa154
IJ		 1619static void xs_write_space(struct sock *sk)
1620{
13331a55 1621 struct socket_wq *wq;
1f0fa154
IJ		 1622 struct rpc_xprt *xprt;
1623
13331a55 1624 if (!sk->sk_socket)
1f0fa154 1625 return;
13331a55 1626 clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1f0fa154
IJ		 1627
1628 if (unlikely(!(xprt = xprt_from_sock(sk))))
1629 return;
13331a55
TM		 1630 rcu_read_lock();
1631 wq = rcu_dereference(sk->sk_wq);
1632 if (!wq || test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &wq->flags) == 0)
1633 goto out;
1f0fa154
IJ		 1634
1635 xprt_write_space(xprt);
13331a55
TM		 1636out:
1637 rcu_read_unlock();
1f0fa154
IJ		 1638}
1639
9903cd1c 1640/**
c7b2cae8
CL		 1641 * xs_udp_write_space - callback invoked when socket buffer space
1642 * becomes available
9903cd1c
CL		 1643 * @sk: socket whose state has changed
1644 *
a246b010
CL		 1645 * Called when more output buffer space is available for this socket.
1646 * We try not to wake our writers until they can make "significant"
c7b2cae8 1647 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
a246b010
CL		 1648 * with a bunch of small requests.
1649 */
c7b2cae8 1650static void xs_udp_write_space(struct sock *sk)
a246b010 1651{
f064af1e 1652 read_lock_bh(&sk->sk_callback_lock);
a246b010 1653
c7b2cae8 1654 /* from net/core/sock.c:sock_def_write_space */
1f0fa154
IJ		 1655 if (sock_writeable(sk))
1656 xs_write_space(sk);
a246b010 1657
f064af1e 1658 read_unlock_bh(&sk->sk_callback_lock);
c7b2cae8 1659}
a246b010 1660
c7b2cae8
CL		 1661/**
1662 * xs_tcp_write_space - callback invoked when socket buffer space
1663 * becomes available
1664 * @sk: socket whose state has changed
1665 *
1666 * Called when more output buffer space is available for this socket.
1667 * We try not to wake our writers until they can make "significant"
1668 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1669 * with a bunch of small requests.
1670 */
1671static void xs_tcp_write_space(struct sock *sk)
1672{
f064af1e 1673 read_lock_bh(&sk->sk_callback_lock);
c7b2cae8
CL		 1674
1675 /* from net/core/stream.c:sk_stream_write_space */
64dc6130 1676 if (sk_stream_is_writeable(sk))
1f0fa154 1677 xs_write_space(sk);
b6ddf64f 1678
f064af1e 1679 read_unlock_bh(&sk->sk_callback_lock);
a246b010
CL		 1680}
1681
470056c2 1682static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
a246b010 1683{
ee0ac0c2
CL		 1684 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1685 struct sock *sk = transport->inet;
a246b010 1686
7c6e066e 1687 if (transport->rcvsize) {
a246b010 1688 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
7c6e066e 1689 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
a246b010 1690 }
7c6e066e 1691 if (transport->sndsize) {
a246b010 1692 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
7c6e066e 1693 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
a246b010
CL		 1694 sk->sk_write_space(sk);
1695 }
1696}
1697
43118c29 1698/**
470056c2 1699 * xs_udp_set_buffer_size - set send and receive limits
43118c29 1700 * @xprt: generic transport
470056c2
CL		 1701 * @sndsize: requested size of send buffer, in bytes
1702 * @rcvsize: requested size of receive buffer, in bytes
43118c29 1703 *
470056c2 1704 * Set socket send and receive buffer size limits.
43118c29 1705 */
470056c2 1706static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
43118c29 1707{
7c6e066e
CL		 1708 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1709
1710 transport->sndsize = 0;
470056c2 1711 if (sndsize)
7c6e066e
CL		 1712 transport->sndsize = sndsize + 1024;
1713 transport->rcvsize = 0;
470056c2 1714 if (rcvsize)
7c6e066e 1715 transport->rcvsize = rcvsize + 1024;
470056c2
CL		 1716
1717 xs_udp_do_set_buffer_size(xprt);
43118c29
CL		 1718}
1719
46c0ee8b
CL		 1720/**
1721 * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1722 * @task: task that timed out
1723 *
1724 * Adjust the congestion window after a retransmit timeout has occurred.
1725 */
6a24dfb6 1726static void xs_udp_timer(struct rpc_xprt *xprt, struct rpc_task *task)
46c0ee8b 1727{
6a24dfb6 1728 xprt_adjust_cwnd(xprt, task, -ETIMEDOUT);
46c0ee8b
CL		 1729}
1730
b85d8806
CL		 1731static unsigned short xs_get_random_port(void)
1732{
5d71899a 1733 unsigned short range = xprt_max_resvport - xprt_min_resvport + 1;
63862b5b 1734 unsigned short rand = (unsigned short) prandom_u32() % range;
b85d8806
CL		 1735 return rand + xprt_min_resvport;
1736}
1737
4dda9c8a
TM		 1738/**
1739 * xs_set_reuseaddr_port - set the socket's port and address reuse options
1740 * @sock: socket
1741 *
1742 * Note that this function has to be called on all sockets that share the
1743 * same port, and it must be called before binding.
1744 */
1745static void xs_sock_set_reuseport(struct socket *sock)
1746{
402e23b4 1747 int opt = 1;
4dda9c8a 1748
402e23b4
TM		 1749 kernel_setsockopt(sock, SOL_SOCKET, SO_REUSEPORT,
1750 (char *)&opt, sizeof(opt));
4dda9c8a
TM		 1751}
1752
1753static unsigned short xs_sock_getport(struct socket *sock)
1754{
1755 struct sockaddr_storage buf;
1756 int buflen;
1757 unsigned short port = 0;
1758
1759 if (kernel_getsockname(sock, (struct sockaddr *)&buf, &buflen) < 0)
1760 goto out;
1761 switch (buf.ss_family) {
1762 case AF_INET6:
1763 port = ntohs(((struct sockaddr_in6 *)&buf)->sin6_port);
1764 break;
1765 case AF_INET:
1766 port = ntohs(((struct sockaddr_in *)&buf)->sin_port);
1767 }
1768out:
1769 return port;
1770}
1771
92200412
CL		 1772/**
1773 * xs_set_port - reset the port number in the remote endpoint address
1774 * @xprt: generic transport
1775 * @port: new port number
1776 *
1777 */
1778static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1779{
46121cf7 1780 dprintk("RPC: setting port for xprt %p to %u\n", xprt, port);
c4efcb1d 1781
9dc3b095
CL		 1782 rpc_set_port(xs_addr(xprt), port);
1783 xs_update_peer_port(xprt);
92200412
CL		 1784}
1785
4dda9c8a
TM		 1786static void xs_set_srcport(struct sock_xprt *transport, struct socket *sock)
1787{
1788 if (transport->srcport == 0)
1789 transport->srcport = xs_sock_getport(sock);
1790}
1791
5d4ec932 1792static unsigned short xs_get_srcport(struct sock_xprt *transport)
67a391d7 1793{
fbfffbd5 1794 unsigned short port = transport->srcport;
67a391d7
TM		 1795
1796 if (port == 0 && transport->xprt.resvport)
1797 port = xs_get_random_port();
1798 return port;
1799}
1800
baaf4e48 1801static unsigned short xs_next_srcport(struct sock_xprt *transport, unsigned short port)
67a391d7 1802{
fbfffbd5
CL		 1803 if (transport->srcport != 0)
1804 transport->srcport = 0;
67a391d7
TM		 1805 if (!transport->xprt.resvport)
1806 return 0;
1807 if (port <= xprt_min_resvport || port > xprt_max_resvport)
1808 return xprt_max_resvport;
1809 return --port;
1810}
beb59b68 1811static int xs_bind(struct sock_xprt *transport, struct socket *sock)
a246b010 1812{
beb59b68 1813 struct sockaddr_storage myaddr;
67a391d7 1814 int err, nloop = 0;
5d4ec932 1815 unsigned short port = xs_get_srcport(transport);
67a391d7 1816 unsigned short last;
a246b010 1817
0f7a622c
CP		 1818 /*
1819 * If we are asking for any ephemeral port (i.e. port == 0 &&
1820 * transport->xprt.resvport == 0), don't bind. Let the local
1821 * port selection happen implicitly when the socket is used
1822 * (for example at connect time).
1823 *
1824 * This ensures that we can continue to establish TCP
1825 * connections even when all local ephemeral ports are already
1826 * a part of some TCP connection. This makes no difference
1827 * for UDP sockets, but also doens't harm them.
1828 *
1829 * If we're asking for any reserved port (i.e. port == 0 &&
1830 * transport->xprt.resvport == 1) xs_get_srcport above will
1831 * ensure that port is non-zero and we will bind as needed.
1832 */
1833 if (port == 0)
1834 return 0;
1835
beb59b68 1836 memcpy(&myaddr, &transport->srcaddr, transport->xprt.addrlen);
a246b010 1837 do {
beb59b68
PE		 1838 rpc_set_port((struct sockaddr *)&myaddr, port);
1839 err = kernel_bind(sock, (struct sockaddr *)&myaddr,
1840 transport->xprt.addrlen);
a246b010 1841 if (err == 0) {
fbfffbd5 1842 transport->srcport = port;
d3bc9a1d 1843 break;
a246b010 1844 }
67a391d7 1845 last = port;
baaf4e48 1846 port = xs_next_srcport(transport, port);
67a391d7
TM		 1847 if (port > last)
1848 nloop++;
1849 } while (err == -EADDRINUSE && nloop != 2);
90058d37 1850
4232e863 1851 if (myaddr.ss_family == AF_INET)
beb59b68
PE		 1852 dprintk("RPC: %s %pI4:%u: %s (%d)\n", __func__,
1853 &((struct sockaddr_in *)&myaddr)->sin_addr,
1854 port, err ? "failed" : "ok", err);
1855 else
1856 dprintk("RPC: %s %pI6:%u: %s (%d)\n", __func__,
1857 &((struct sockaddr_in6 *)&myaddr)->sin6_addr,
1858 port, err ? "failed" : "ok", err);
a246b010
CL		 1859 return err;
1860}
1861
176e21ee
CL		 1862/*
1863 * We don't support autobind on AF_LOCAL sockets
1864 */
1865static void xs_local_rpcbind(struct rpc_task *task)
1866{
fb43d172 1867 xprt_set_bound(task->tk_xprt);
176e21ee
CL		 1868}
1869
1870static void xs_local_set_port(struct rpc_xprt *xprt, unsigned short port)
1871{
1872}
beb59b68 1873
ed07536e
PZ		 1874#ifdef CONFIG_DEBUG_LOCK_ALLOC
1875static struct lock_class_key xs_key[2];
1876static struct lock_class_key xs_slock_key[2];
1877
176e21ee
CL		 1878static inline void xs_reclassify_socketu(struct socket *sock)
1879{
1880 struct sock *sk = sock->sk;
1881
176e21ee
CL		 1882 sock_lock_init_class_and_name(sk, "slock-AF_LOCAL-RPC",
1883 &xs_slock_key[1], "sk_lock-AF_LOCAL-RPC", &xs_key[1]);
1884}
1885
8945ee5e 1886static inline void xs_reclassify_socket4(struct socket *sock)
ed07536e
PZ		 1887{
1888 struct sock *sk = sock->sk;
8945ee5e 1889
8945ee5e
CL		 1890 sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1891 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1892}
ed07536e 1893
8945ee5e
CL		 1894static inline void xs_reclassify_socket6(struct socket *sock)
1895{
1896 struct sock *sk = sock->sk;
ed07536e 1897
8945ee5e
CL		 1898 sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1899 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
ed07536e 1900}
6bc9638a
PE		 1901
1902static inline void xs_reclassify_socket(int family, struct socket *sock)
1903{
fafc4e1e 1904 if (WARN_ON_ONCE(!sock_allow_reclassification(sock->sk)))
1b7a1819
WAA		 1905 return;
1906
4232e863 1907 switch (family) {
176e21ee
CL		 1908 case AF_LOCAL:
1909 xs_reclassify_socketu(sock);
1910 break;
4232e863 1911 case AF_INET:
6bc9638a 1912 xs_reclassify_socket4(sock);
4232e863
CL		 1913 break;
1914 case AF_INET6:
6bc9638a 1915 xs_reclassify_socket6(sock);
4232e863
CL		 1916 break;
1917 }
6bc9638a 1918}
ed07536e 1919#else
6bc9638a
PE		 1920static inline void xs_reclassify_socket(int family, struct socket *sock)
1921{
1922}
ed07536e
PZ		 1923#endif
1924
93dc41bd
N		 1925static void xs_dummy_setup_socket(struct work_struct *work)
1926{
1927}
1928
6bc9638a 1929static struct socket *xs_create_sock(struct rpc_xprt *xprt,
4dda9c8a
TM		 1930 struct sock_xprt *transport, int family, int type,
1931 int protocol, bool reuseport)
22f79326
PE		 1932{
1933 struct socket *sock;
1934 int err;
1935
6bc9638a 1936 err = __sock_create(xprt->xprt_net, family, type, protocol, &sock, 1);
22f79326
PE		 1937 if (err < 0) {
1938 dprintk("RPC: can't create %d transport socket (%d).\n",
1939 protocol, -err);
1940 goto out;
1941 }
6bc9638a 1942 xs_reclassify_socket(family, sock);
22f79326 1943
4dda9c8a
TM		 1944 if (reuseport)
1945 xs_sock_set_reuseport(sock);
1946
4cea288a
BH		 1947 err = xs_bind(transport, sock);
1948 if (err) {
22f79326
PE		 1949 sock_release(sock);
1950 goto out;
1951 }
1952
1953 return sock;
1954out:
1955 return ERR_PTR(err);
1956}
1957
176e21ee
CL		 1958static int xs_local_finish_connecting(struct rpc_xprt *xprt,
1959 struct socket *sock)
1960{
1961 struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1962 xprt);
1963
1964 if (!transport->inet) {
1965 struct sock *sk = sock->sk;
1966
1967 write_lock_bh(&sk->sk_callback_lock);
1968
1969 xs_save_old_callbacks(transport, sk);
1970
1971 sk->sk_user_data = xprt;
a2648094 1972 sk->sk_data_ready = xs_data_ready;
176e21ee 1973 sk->sk_write_space = xs_udp_write_space;
b4411457 1974 sock_set_flag(sk, SOCK_FASYNC);
2118071d 1975 sk->sk_error_report = xs_error_report;
c2126157 1976 sk->sk_allocation = GFP_NOIO;
176e21ee
CL		 1977
1978 xprt_clear_connected(xprt);
1979
1980 /* Reset to new socket */
1981 transport->sock = sock;
1982 transport->inet = sk;
1983
1984 write_unlock_bh(&sk->sk_callback_lock);
1985 }
1986
1987 /* Tell the socket layer to start connecting... */
1988 xprt->stat.connect_count++;
1989 xprt->stat.connect_start = jiffies;
1990 return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, 0);
1991}
1992
1993/**
1994 * xs_local_setup_socket - create AF_LOCAL socket, connect to a local endpoint
176e21ee 1995 * @transport: socket transport to connect
176e21ee 1996 */
dc107402 1997static int xs_local_setup_socket(struct sock_xprt *transport)
176e21ee 1998{
176e21ee
CL		 1999 struct rpc_xprt *xprt = &transport->xprt;
2000 struct socket *sock;
2001 int status = -EIO;
2002
176e21ee
CL		 2003 status = __sock_create(xprt->xprt_net, AF_LOCAL,
2004 SOCK_STREAM, 0, &sock, 1);
2005 if (status < 0) {
2006 dprintk("RPC: can't create AF_LOCAL "
2007 "transport socket (%d).\n", -status);
2008 goto out;
2009 }
d1358917 2010 xs_reclassify_socket(AF_LOCAL, sock);
176e21ee
CL		 2011
2012 dprintk("RPC: worker connecting xprt %p via AF_LOCAL to %s\n",
2013 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2014
2015 status = xs_local_finish_connecting(xprt, sock);
40b5ea0c 2016 trace_rpc_socket_connect(xprt, sock, status);
176e21ee
CL		 2017 switch (status) {
2018 case 0:
2019 dprintk("RPC: xprt %p connected to %s\n",
2020 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2021 xprt_set_connected(xprt);
3601c4a9 2022 case -ENOBUFS:
176e21ee
CL		 2023 break;
2024 case -ENOENT:
2025 dprintk("RPC: xprt %p: socket %s does not exist\n",
2026 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2027 break;
4a20a988
TM		 2028 case -ECONNREFUSED:
2029 dprintk("RPC: xprt %p: connection refused for %s\n",
2030 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2031 break;
176e21ee
CL		 2032 default:
2033 printk(KERN_ERR "%s: unhandled error (%d) connecting to %s\n",
2034 __func__, -status,
2035 xprt->address_strings[RPC_DISPLAY_ADDR]);
2036 }
2037
2038out:
2039 xprt_clear_connecting(xprt);
2040 xprt_wake_pending_tasks(xprt, status);
dc107402
BF		 2041 return status;
2042}
2043
b6669737 2044static void xs_local_connect(struct rpc_xprt *xprt, struct rpc_task *task)
dc107402 2045{
dc107402
BF		 2046 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2047 int ret;
2048
2049 if (RPC_IS_ASYNC(task)) {
2050 /*
2051 * We want the AF_LOCAL connect to be resolved in the
2052 * filesystem namespace of the process making the rpc
2053 * call. Thus we connect synchronously.
2054 *
2055 * If we want to support asynchronous AF_LOCAL calls,
2056 * we'll need to figure out how to pass a namespace to
2057 * connect.
2058 */
2059 rpc_exit(task, -ENOTCONN);
2060 return;
2061 }
2062 ret = xs_local_setup_socket(transport);
2063 if (ret && !RPC_IS_SOFTCONN(task))
2064 msleep_interruptible(15000);
176e21ee
CL		 2065}
2066
3c87ef6e 2067#if IS_ENABLED(CONFIG_SUNRPC_SWAP)
d6e971d8
JL		 2068/*
2069 * Note that this should be called with XPRT_LOCKED held (or when we otherwise
2070 * know that we have exclusive access to the socket), to guard against
2071 * races with xs_reset_transport.
2072 */
a564b8f0
MG		 2073static void xs_set_memalloc(struct rpc_xprt *xprt)
2074{
2075 struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
2076 xprt);
2077
d6e971d8
JL		 2078 /*
2079 * If there's no sock, then we have nothing to set. The
2080 * reconnecting process will get it for us.
2081 */
2082 if (!transport->inet)
2083 return;
8e228133 2084 if (atomic_read(&xprt->swapper))
a564b8f0
MG		 2085 sk_set_memalloc(transport->inet);
2086}
2087
2088/**
d67fa4d8 2089 * xs_enable_swap - Tag this transport as being used for swap.
a564b8f0 2090 * @xprt: transport to tag
a564b8f0 2091 *
8e228133
JL		 2092 * Take a reference to this transport on behalf of the rpc_clnt, and
2093 * optionally mark it for swapping if it wasn't already.
a564b8f0 2094 */
d67fa4d8
JL		 2095static int
2096xs_enable_swap(struct rpc_xprt *xprt)
a564b8f0 2097{
d6e971d8 2098 struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
a564b8f0 2099
d6e971d8
JL		 2100 if (atomic_inc_return(&xprt->swapper) != 1)
2101 return 0;
2102 if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE))
2103 return -ERESTARTSYS;
2104 if (xs->inet)
2105 sk_set_memalloc(xs->inet);
2106 xprt_release_xprt(xprt, NULL);
8e228133
JL		 2107 return 0;
2108}
a564b8f0 2109
8e228133 2110/**
d67fa4d8 2111 * xs_disable_swap - Untag this transport as being used for swap.
8e228133
JL		 2112 * @xprt: transport to tag
2113 *
2114 * Drop a "swapper" reference to this xprt on behalf of the rpc_clnt. If the
2115 * swapper refcount goes to 0, untag the socket as a memalloc socket.
2116 */
d67fa4d8
JL		 2117static void
2118xs_disable_swap(struct rpc_xprt *xprt)
8e228133 2119{
d6e971d8 2120 struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
8e228133 2121
d6e971d8
JL		 2122 if (!atomic_dec_and_test(&xprt->swapper))
2123 return;
2124 if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE))
2125 return;
2126 if (xs->inet)
2127 sk_clear_memalloc(xs->inet);
2128 xprt_release_xprt(xprt, NULL);
a564b8f0 2129}
a564b8f0
MG		 2130#else
2131static void xs_set_memalloc(struct rpc_xprt *xprt)
2132{
2133}
d67fa4d8
JL		 2134
2135static int
2136xs_enable_swap(struct rpc_xprt *xprt)
2137{
2138 return -EINVAL;
2139}
2140
2141static void
2142xs_disable_swap(struct rpc_xprt *xprt)
2143{
2144}
a564b8f0
MG		 2145#endif
2146
16be2d20
CL		 2147static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2148{
2149 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2150
2151 if (!transport->inet) {
2152 struct sock *sk = sock->sk;
2153
2154 write_lock_bh(&sk->sk_callback_lock);
2155
2a9e1cfa
TM		 2156 xs_save_old_callbacks(transport, sk);
2157
16be2d20 2158 sk->sk_user_data = xprt;
f9b2ee71 2159 sk->sk_data_ready = xs_data_ready;
16be2d20 2160 sk->sk_write_space = xs_udp_write_space;
b4411457 2161 sock_set_flag(sk, SOCK_FASYNC);
c2126157 2162 sk->sk_allocation = GFP_NOIO;
16be2d20
CL		 2163
2164 xprt_set_connected(xprt);
2165
2166 /* Reset to new socket */
2167 transport->sock = sock;
2168 transport->inet = sk;
2169
a564b8f0
MG		 2170 xs_set_memalloc(xprt);
2171
16be2d20
CL		 2172 write_unlock_bh(&sk->sk_callback_lock);
2173 }
2174 xs_udp_do_set_buffer_size(xprt);
02910177
TM		 2175
2176 xprt->stat.connect_start = jiffies;
16be2d20
CL		 2177}
2178
8c14ff2a 2179static void xs_udp_setup_socket(struct work_struct *work)
a246b010 2180{
34161db6
TM		 2181 struct sock_xprt *transport =
2182 container_of(work, struct sock_xprt, connect_worker.work);
c8475461 2183 struct rpc_xprt *xprt = &transport->xprt;
ee0ac0c2 2184 struct socket *sock = transport->sock;
b65c0310 2185 int status = -EIO;
9903cd1c 2186
8c14ff2a 2187 sock = xs_create_sock(xprt, transport,
4dda9c8a
TM		 2188 xs_addr(xprt)->sa_family, SOCK_DGRAM,
2189 IPPROTO_UDP, false);
b65c0310 2190 if (IS_ERR(sock))
b0d93ad5 2191 goto out;
68e220bd 2192
c740eff8
CL		 2193 dprintk("RPC: worker connecting xprt %p via %s to "
2194 "%s (port %s)\n", xprt,
2195 xprt->address_strings[RPC_DISPLAY_PROTO],
2196 xprt->address_strings[RPC_DISPLAY_ADDR],
2197 xprt->address_strings[RPC_DISPLAY_PORT]);
68e220bd
CL		 2198
2199 xs_udp_finish_connecting(xprt, sock);
40b5ea0c 2200 trace_rpc_socket_connect(xprt, sock, 0);
b0d93ad5
CL		 2201 status = 0;
2202out:
718ba5b8 2203 xprt_unlock_connect(xprt, transport);
b0d93ad5 2204 xprt_clear_connecting(xprt);
7d1e8255 2205 xprt_wake_pending_tasks(xprt, status);
a246b010
CL		 2206}
2207
4876cc77
TM		 2208/**
2209 * xs_tcp_shutdown - gracefully shut down a TCP socket
2210 * @xprt: transport
2211 *
2212 * Initiates a graceful shutdown of the TCP socket by calling the
2213 * equivalent of shutdown(SHUT_RDWR);
2214 */
2215static void xs_tcp_shutdown(struct rpc_xprt *xprt)
2216{
2217 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2218 struct socket *sock = transport->sock;
2219
2220 if (sock == NULL)
2221 return;
2222 if (xprt_connected(xprt)) {
2223 kernel_sock_shutdown(sock, SHUT_RDWR);
2224 trace_rpc_socket_shutdown(xprt, sock);
2225 } else
2226 xs_reset_transport(transport);
2227}
2228
16be2d20 2229static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
a246b010 2230{
16be2d20 2231 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
fe19a96b 2232 int ret = -ENOTCONN;
edb267a6 2233
ee0ac0c2 2234 if (!transport->inet) {
b0d93ad5 2235 struct sock *sk = sock->sk;
7f260e85
TM		 2236 unsigned int keepidle = xprt->timeout->to_initval / HZ;
2237 unsigned int keepcnt = xprt->timeout->to_retries + 1;
2238 unsigned int opt_on = 1;
775f06ab 2239 unsigned int timeo;
d88e4d82 2240 unsigned int addr_pref = IPV6_PREFER_SRC_PUBLIC;
7f260e85
TM		 2241
2242 /* TCP Keepalive options */
2243 kernel_setsockopt(sock, SOL_SOCKET, SO_KEEPALIVE,
2244 (char *)&opt_on, sizeof(opt_on));
2245 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPIDLE,
2246 (char *)&keepidle, sizeof(keepidle));
2247 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPINTVL,
2248 (char *)&keepidle, sizeof(keepidle));
2249 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPCNT,
2250 (char *)&keepcnt, sizeof(keepcnt));
b0d93ad5 2251
d88e4d82
N		 2252 /* Avoid temporary address, they are bad for long-lived
2253 * connections such as NFS mounts.
2254 * RFC4941, section 3.6 suggests that:
2255 * Individual applications, which have specific
2256 * knowledge about the normal duration of connections,
2257 * MAY override this as appropriate.
2258 */
2259 kernel_setsockopt(sock, SOL_IPV6, IPV6_ADDR_PREFERENCES,
2260 (char *)&addr_pref, sizeof(addr_pref));
2261
775f06ab
TM		 2262 /* TCP user timeout (see RFC5482) */
2263 timeo = jiffies_to_msecs(xprt->timeout->to_initval) *
2264 (xprt->timeout->to_retries + 1);
2265 kernel_setsockopt(sock, SOL_TCP, TCP_USER_TIMEOUT,
2266 (char *)&timeo, sizeof(timeo));
2267
b0d93ad5
CL		 2268 write_lock_bh(&sk->sk_callback_lock);
2269
2a9e1cfa
TM		 2270 xs_save_old_callbacks(transport, sk);
2271
b0d93ad5 2272 sk->sk_user_data = xprt;
5157b956 2273 sk->sk_data_ready = xs_data_ready;
b0d93ad5
CL		 2274 sk->sk_state_change = xs_tcp_state_change;
2275 sk->sk_write_space = xs_tcp_write_space;
b4411457 2276 sock_set_flag(sk, SOCK_FASYNC);
2118071d 2277 sk->sk_error_report = xs_error_report;
c2126157 2278 sk->sk_allocation = GFP_NOIO;
3167e12c
CL		 2279
2280 /* socket options */
3167e12c 2281 sock_reset_flag(sk, SOCK_LINGER);
3167e12c 2282 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
b0d93ad5
CL		 2283
2284 xprt_clear_connected(xprt);
2285
2286 /* Reset to new socket */
ee0ac0c2
CL		 2287 transport->sock = sock;
2288 transport->inet = sk;
b0d93ad5
CL		 2289
2290 write_unlock_bh(&sk->sk_callback_lock);
2291 }
2292
01d37c42 2293 if (!xprt_bound(xprt))
fe19a96b 2294 goto out;
01d37c42 2295
a564b8f0
MG		 2296 xs_set_memalloc(xprt);
2297
b0d93ad5 2298 /* Tell the socket layer to start connecting... */
262ca07d
CL		 2299 xprt->stat.connect_count++;
2300 xprt->stat.connect_start = jiffies;
0fdea1e8 2301 set_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
fe19a96b
TM		 2302 ret = kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
2303 switch (ret) {
2304 case 0:
4dda9c8a 2305 xs_set_srcport(transport, sock);
fe19a96b
TM		 2306 case -EINPROGRESS:
2307 /* SYN_SENT! */
fe19a96b
TM		 2308 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2309 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1f4c17a0
TM		 2310 break;
2311 case -EADDRNOTAVAIL:
2312 /* Source port number is unavailable. Try a new one! */
2313 transport->srcport = 0;
fe19a96b
TM		 2314 }
2315out:
2316 return ret;
16be2d20
CL		 2317}
2318
9903cd1c 2319/**
b61d59ff 2320 * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint
9903cd1c
CL		 2321 *
2322 * Invoked by a work queue tasklet.
a246b010 2323 */
cdd518d5 2324static void xs_tcp_setup_socket(struct work_struct *work)
a246b010 2325{
cdd518d5
PE		 2326 struct sock_xprt *transport =
2327 container_of(work, struct sock_xprt, connect_worker.work);
ee0ac0c2 2328 struct socket *sock = transport->sock;
a9f5f0f7 2329 struct rpc_xprt *xprt = &transport->xprt;
b61d59ff 2330 int status = -EIO;
a246b010 2331
ee0ac0c2 2332 if (!sock) {
cdd518d5 2333 sock = xs_create_sock(xprt, transport,
4dda9c8a
TM		 2334 xs_addr(xprt)->sa_family, SOCK_STREAM,
2335 IPPROTO_TCP, true);
b61d59ff
TM		 2336 if (IS_ERR(sock)) {
2337 status = PTR_ERR(sock);
3167e12c
CL		 2338 goto out;
2339 }
7d1e8255 2340 }
a246b010 2341
c740eff8
CL		 2342 dprintk("RPC: worker connecting xprt %p via %s to "
2343 "%s (port %s)\n", xprt,
2344 xprt->address_strings[RPC_DISPLAY_PROTO],
2345 xprt->address_strings[RPC_DISPLAY_ADDR],
2346 xprt->address_strings[RPC_DISPLAY_PORT]);
edb267a6 2347
16be2d20 2348 status = xs_tcp_finish_connecting(xprt, sock);
40b5ea0c 2349 trace_rpc_socket_connect(xprt, sock, status);
46121cf7
CL		 2350 dprintk("RPC: %p connect status %d connected %d sock state %d\n",
2351 xprt, -status, xprt_connected(xprt),
2352 sock->sk->sk_state);
2a491991 2353 switch (status) {
f75e6745
TM		 2354 default:
2355 printk("%s: connect returned unhandled error %d\n",
2356 __func__, status);
2357 case -EADDRNOTAVAIL:
2358 /* We're probably in TIME_WAIT. Get rid of existing socket,
2359 * and retry
2360 */
a519fc7a 2361 xs_tcp_force_close(xprt);
88b5ed73 2362 break;
2a491991
TM		 2363 case 0:
2364 case -EINPROGRESS:
2365 case -EALREADY:
718ba5b8 2366 xprt_unlock_connect(xprt, transport);
7d1e8255 2367 return;
9fcfe0c8
TM		 2368 case -EINVAL:
2369 /* Happens, for instance, if the user specified a link
2370 * local IPv6 address without a scope-id.
2371 */
3ed5e2a2
TM		 2372 case -ECONNREFUSED:
2373 case -ECONNRESET:
2374 case -ENETUNREACH:
3913c78c 2375 case -EADDRINUSE:
3601c4a9 2376 case -ENOBUFS:
3ed5e2a2 2377 /* retry with existing socket, after a delay */
4efdd92c 2378 xs_tcp_force_close(xprt);
9fcfe0c8 2379 goto out;
a246b010 2380 }
2a491991 2381 status = -EAGAIN;
a246b010 2382out:
718ba5b8 2383 xprt_unlock_connect(xprt, transport);
68e220bd 2384 xprt_clear_connecting(xprt);
7d1e8255 2385 xprt_wake_pending_tasks(xprt, status);
68e220bd 2386}
b0d93ad5 2387
02910177
TM		 2388static unsigned long xs_reconnect_delay(const struct rpc_xprt *xprt)
2389{
2390 unsigned long start, now = jiffies;
2391
2392 start = xprt->stat.connect_start + xprt->reestablish_timeout;
2393 if (time_after(start, now))
2394 return start - now;
2395 return 0;
2396}
2397
3851f1cd
TM		 2398static void xs_reconnect_backoff(struct rpc_xprt *xprt)
2399{
2400 xprt->reestablish_timeout <<= 1;
2401 if (xprt->reestablish_timeout > xprt->max_reconnect_timeout)
2402 xprt->reestablish_timeout = xprt->max_reconnect_timeout;
2403 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2404 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2405}
2406
9903cd1c
CL		 2407/**
2408 * xs_connect - connect a socket to a remote endpoint
1b092092 2409 * @xprt: pointer to transport structure
9903cd1c
CL		 2410 * @task: address of RPC task that manages state of connect request
2411 *
2412 * TCP: If the remote end dropped the connection, delay reconnecting.
03bf4b70
CL		 2413 *
2414 * UDP socket connects are synchronous, but we use a work queue anyway
2415 * to guarantee that even unprivileged user processes can set up a
2416 * socket on a privileged port.
2417 *
2418 * If a UDP socket connect fails, the delay behavior here prevents
2419 * retry floods (hard mounts).
9903cd1c 2420 */
1b092092 2421static void xs_connect(struct rpc_xprt *xprt, struct rpc_task *task)
a246b010 2422{
ee0ac0c2 2423 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
02910177 2424 unsigned long delay = 0;
a246b010 2425
718ba5b8
TM		 2426 WARN_ON_ONCE(!xprt_lock_connect(xprt, task, transport));
2427
99b1a4c3 2428 if (transport->sock != NULL) {
46121cf7
CL		 2429 dprintk("RPC: xs_connect delayed xprt %p for %lu "
2430 "seconds\n",
03bf4b70 2431 xprt, xprt->reestablish_timeout / HZ);
99b1a4c3
TM		 2432
2433 /* Start by resetting any existing state */
2434 xs_reset_transport(transport);
2435
02910177 2436 delay = xs_reconnect_delay(xprt);
3851f1cd 2437 xs_reconnect_backoff(xprt);
02910177 2438
02910177 2439 } else
46121cf7 2440 dprintk("RPC: xs_connect scheduled xprt %p\n", xprt);
02910177
TM		 2441
2442 queue_delayed_work(xprtiod_workqueue,
2443 &transport->connect_worker,
2444 delay);
a246b010
CL		 2445}
2446
176e21ee
CL		 2447/**
2448 * xs_local_print_stats - display AF_LOCAL socket-specifc stats
2449 * @xprt: rpc_xprt struct containing statistics
2450 * @seq: output file
2451 *
2452 */
2453static void xs_local_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2454{
2455 long idle_time = 0;
2456
2457 if (xprt_connected(xprt))
2458 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2459
2460 seq_printf(seq, "\txprt:\tlocal %lu %lu %lu %ld %lu %lu %lu "
15a45206 2461 "%llu %llu %lu %llu %llu\n",
176e21ee
CL		 2462 xprt->stat.bind_count,
2463 xprt->stat.connect_count,
2464 xprt->stat.connect_time,
2465 idle_time,
2466 xprt->stat.sends,
2467 xprt->stat.recvs,
2468 xprt->stat.bad_xids,
2469 xprt->stat.req_u,
15a45206
AA		 2470 xprt->stat.bklog_u,
2471 xprt->stat.max_slots,
2472 xprt->stat.sending_u,
2473 xprt->stat.pending_u);
176e21ee
CL		 2474}
2475
262ca07d
CL		 2476/**
2477 * xs_udp_print_stats - display UDP socket-specifc stats
2478 * @xprt: rpc_xprt struct containing statistics
2479 * @seq: output file
2480 *
2481 */
2482static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2483{
c8475461
CL		 2484 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2485
15a45206
AA		 2486 seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %llu %llu "
2487 "%lu %llu %llu\n",
fbfffbd5 2488 transport->srcport,
262ca07d
CL		 2489 xprt->stat.bind_count,
2490 xprt->stat.sends,
2491 xprt->stat.recvs,
2492 xprt->stat.bad_xids,
2493 xprt->stat.req_u,
15a45206
AA		 2494 xprt->stat.bklog_u,
2495 xprt->stat.max_slots,
2496 xprt->stat.sending_u,
2497 xprt->stat.pending_u);
262ca07d
CL		 2498}
2499
2500/**
2501 * xs_tcp_print_stats - display TCP socket-specifc stats
2502 * @xprt: rpc_xprt struct containing statistics
2503 * @seq: output file
2504 *
2505 */
2506static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2507{
c8475461 2508 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
262ca07d
CL		 2509 long idle_time = 0;
2510
2511 if (xprt_connected(xprt))
2512 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2513
15a45206
AA		 2514 seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu "
2515 "%llu %llu %lu %llu %llu\n",
fbfffbd5 2516 transport->srcport,
262ca07d
CL		 2517 xprt->stat.bind_count,
2518 xprt->stat.connect_count,
2519 xprt->stat.connect_time,
2520 idle_time,
2521 xprt->stat.sends,
2522 xprt->stat.recvs,
2523 xprt->stat.bad_xids,
2524 xprt->stat.req_u,
15a45206
AA		 2525 xprt->stat.bklog_u,
2526 xprt->stat.max_slots,
2527 xprt->stat.sending_u,
2528 xprt->stat.pending_u);
262ca07d
CL		 2529}
2530
4cfc7e60
RI		 2531/*
2532 * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason
2533 * we allocate pages instead doing a kmalloc like rpc_malloc is because we want
2534 * to use the server side send routines.
2535 */
5a51f13a 2536static void *bc_malloc(struct rpc_task *task, size_t size)
4cfc7e60
RI		 2537{
2538 struct page *page;
2539 struct rpc_buffer *buf;
2540
b8a13d03
WAA		 2541 WARN_ON_ONCE(size > PAGE_SIZE - sizeof(struct rpc_buffer));
2542 if (size > PAGE_SIZE - sizeof(struct rpc_buffer))
2543 return NULL;
4cfc7e60 2544
b8a13d03 2545 page = alloc_page(GFP_KERNEL);
4cfc7e60
RI		 2546 if (!page)
2547 return NULL;
2548
2549 buf = page_address(page);
2550 buf->len = PAGE_SIZE;
2551
2552 return buf->data;
2553}
2554
2555/*
2556 * Free the space allocated in the bc_alloc routine
2557 */
5a51f13a 2558static void bc_free(void *buffer)
4cfc7e60
RI		 2559{
2560 struct rpc_buffer *buf;
2561
2562 if (!buffer)
2563 return;
2564
2565 buf = container_of(buffer, struct rpc_buffer, data);
2566 free_page((unsigned long)buf);
2567}
2568
2569/*
2570 * Use the svc_sock to send the callback. Must be called with svsk->sk_mutex
2571 * held. Borrows heavily from svc_tcp_sendto and xs_tcp_send_request.
2572 */
2573static int bc_sendto(struct rpc_rqst *req)
2574{
2575 int len;
2576 struct xdr_buf *xbufp = &req->rq_snd_buf;
2577 struct rpc_xprt *xprt = req->rq_xprt;
2578 struct sock_xprt *transport =
2579 container_of(xprt, struct sock_xprt, xprt);
2580 struct socket *sock = transport->sock;
2581 unsigned long headoff;
2582 unsigned long tailoff;
2583
61677eee 2584 xs_encode_stream_record_marker(xbufp);
4cfc7e60
RI		 2585
2586 tailoff = (unsigned long)xbufp->tail[0].iov_base & ~PAGE_MASK;
2587 headoff = (unsigned long)xbufp->head[0].iov_base & ~PAGE_MASK;
2588 len = svc_send_common(sock, xbufp,
2589 virt_to_page(xbufp->head[0].iov_base), headoff,
2590 xbufp->tail[0].iov_base, tailoff);
2591
2592 if (len != xbufp->len) {
2593 printk(KERN_NOTICE "Error sending entire callback!\n");
2594 len = -EAGAIN;
2595 }
2596
2597 return len;
2598}
2599
2600/*
2601 * The send routine. Borrows from svc_send
2602 */
2603static int bc_send_request(struct rpc_task *task)
2604{
2605 struct rpc_rqst *req = task->tk_rqstp;
2606 struct svc_xprt *xprt;
7fc56136 2607 int len;
4cfc7e60
RI		 2608
2609 dprintk("sending request with xid: %08x\n", ntohl(req->rq_xid));
2610 /*
2611 * Get the server socket associated with this callback xprt
2612 */
2613 xprt = req->rq_xprt->bc_xprt;
4cfc7e60
RI		 2614
2615 /*
2616 * Grab the mutex to serialize data as the connection is shared
2617 * with the fore channel
2618 */
2619 if (!mutex_trylock(&xprt->xpt_mutex)) {
2620 rpc_sleep_on(&xprt->xpt_bc_pending, task, NULL);
2621 if (!mutex_trylock(&xprt->xpt_mutex))
2622 return -EAGAIN;
2623 rpc_wake_up_queued_task(&xprt->xpt_bc_pending, task);
2624 }
2625 if (test_bit(XPT_DEAD, &xprt->xpt_flags))
2626 len = -ENOTCONN;
2627 else
2628 len = bc_sendto(req);
2629 mutex_unlock(&xprt->xpt_mutex);
2630
2631 if (len > 0)
2632 len = 0;
2633
2634 return len;
2635}
2636
2637/*
2638 * The close routine. Since this is client initiated, we do nothing
2639 */
2640
2641static void bc_close(struct rpc_xprt *xprt)
2642{
4cfc7e60
RI		 2643}
2644
2645/*
2646 * The xprt destroy routine. Again, because this connection is client
2647 * initiated, we do nothing
2648 */
2649
2650static void bc_destroy(struct rpc_xprt *xprt)
2651{
47f72efa
KM		 2652 dprintk("RPC: bc_destroy xprt %p\n", xprt);
2653
2654 xs_xprt_free(xprt);
2655 module_put(THIS_MODULE);
4cfc7e60
RI		 2656}
2657
176e21ee
CL		 2658static struct rpc_xprt_ops xs_local_ops = {
2659 .reserve_xprt = xprt_reserve_xprt,
2660 .release_xprt = xs_tcp_release_xprt,
f39c1bfb 2661 .alloc_slot = xprt_alloc_slot,
176e21ee
CL		 2662 .rpcbind = xs_local_rpcbind,
2663 .set_port = xs_local_set_port,
dc107402 2664 .connect = xs_local_connect,
176e21ee
CL		 2665 .buf_alloc = rpc_malloc,
2666 .buf_free = rpc_free,
2667 .send_request = xs_local_send_request,
2668 .set_retrans_timeout = xprt_set_retrans_timeout_def,
2669 .close = xs_close,
a1311d87 2670 .destroy = xs_destroy,
176e21ee 2671 .print_stats = xs_local_print_stats,
d67fa4d8
JL		 2672 .enable_swap = xs_enable_swap,
2673 .disable_swap = xs_disable_swap,
176e21ee
CL		 2674};
2675
262965f5 2676static struct rpc_xprt_ops xs_udp_ops = {
43118c29 2677 .set_buffer_size = xs_udp_set_buffer_size,
12a80469 2678 .reserve_xprt = xprt_reserve_xprt_cong,
49e9a890 2679 .release_xprt = xprt_release_xprt_cong,
f39c1bfb 2680 .alloc_slot = xprt_alloc_slot,
45160d62 2681 .rpcbind = rpcb_getport_async,
92200412 2682 .set_port = xs_set_port,
262965f5 2683 .connect = xs_connect,
02107148
CL		 2684 .buf_alloc = rpc_malloc,
2685 .buf_free = rpc_free,
262965f5 2686 .send_request = xs_udp_send_request,
fe3aca29 2687 .set_retrans_timeout = xprt_set_retrans_timeout_rtt,
46c0ee8b 2688 .timer = xs_udp_timer,
a58dd398 2689 .release_request = xprt_release_rqst_cong,
262965f5
CL		 2690 .close = xs_close,
2691 .destroy = xs_destroy,
262ca07d 2692 .print_stats = xs_udp_print_stats,
d67fa4d8
JL		 2693 .enable_swap = xs_enable_swap,
2694 .disable_swap = xs_disable_swap,
4a068258 2695 .inject_disconnect = xs_inject_disconnect,
262965f5
CL		 2696};
2697
2698static struct rpc_xprt_ops xs_tcp_ops = {
12a80469 2699 .reserve_xprt = xprt_reserve_xprt,
e0ab53de 2700 .release_xprt = xs_tcp_release_xprt,
f39c1bfb 2701 .alloc_slot = xprt_lock_and_alloc_slot,
45160d62 2702 .rpcbind = rpcb_getport_async,
92200412 2703 .set_port = xs_set_port,
0b9e7943 2704 .connect = xs_connect,
02107148
CL		 2705 .buf_alloc = rpc_malloc,
2706 .buf_free = rpc_free,
262965f5 2707 .send_request = xs_tcp_send_request,
fe3aca29 2708 .set_retrans_timeout = xprt_set_retrans_timeout_def,
c627d31b 2709 .close = xs_tcp_shutdown,
9903cd1c 2710 .destroy = xs_destroy,
262ca07d 2711 .print_stats = xs_tcp_print_stats,
d67fa4d8
JL		 2712 .enable_swap = xs_enable_swap,
2713 .disable_swap = xs_disable_swap,
4a068258 2714 .inject_disconnect = xs_inject_disconnect,
42e5c3e2
CL		 2715#ifdef CONFIG_SUNRPC_BACKCHANNEL
2716 .bc_setup = xprt_setup_bc,
76566773 2717 .bc_up = xs_tcp_bc_up,
6b26cc8c 2718 .bc_maxpayload = xs_tcp_bc_maxpayload,
42e5c3e2
CL		 2719 .bc_free_rqst = xprt_free_bc_rqst,
2720 .bc_destroy = xprt_destroy_bc,
2721#endif
a246b010
CL		 2722};
2723
4cfc7e60
RI		 2724/*
2725 * The rpc_xprt_ops for the server backchannel
2726 */
2727
2728static struct rpc_xprt_ops bc_tcp_ops = {
2729 .reserve_xprt = xprt_reserve_xprt,
2730 .release_xprt = xprt_release_xprt,
84e28a30 2731 .alloc_slot = xprt_alloc_slot,
4cfc7e60
RI		 2732 .buf_alloc = bc_malloc,
2733 .buf_free = bc_free,
2734 .send_request = bc_send_request,
2735 .set_retrans_timeout = xprt_set_retrans_timeout_def,
2736 .close = bc_close,
2737 .destroy = bc_destroy,
2738 .print_stats = xs_tcp_print_stats,
d67fa4d8
JL		 2739 .enable_swap = xs_enable_swap,
2740 .disable_swap = xs_disable_swap,
4a068258 2741 .inject_disconnect = xs_inject_disconnect,
4cfc7e60
RI		 2742};
2743
92476850
CL		 2744static int xs_init_anyaddr(const int family, struct sockaddr *sap)
2745{
2746 static const struct sockaddr_in sin = {
2747 .sin_family = AF_INET,
2748 .sin_addr.s_addr = htonl(INADDR_ANY),
2749 };
2750 static const struct sockaddr_in6 sin6 = {
2751 .sin6_family = AF_INET6,
2752 .sin6_addr = IN6ADDR_ANY_INIT,
2753 };
2754
2755 switch (family) {
176e21ee
CL		 2756 case AF_LOCAL:
2757 break;
92476850
CL		 2758 case AF_INET:
2759 memcpy(sap, &sin, sizeof(sin));
2760 break;
2761 case AF_INET6:
2762 memcpy(sap, &sin6, sizeof(sin6));
2763 break;
2764 default:
2765 dprintk("RPC: %s: Bad address family\n", __func__);
2766 return -EAFNOSUPPORT;
2767 }
2768 return 0;
2769}
2770
3c341b0b 2771static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
d9ba131d
TM		 2772 unsigned int slot_table_size,
2773 unsigned int max_slot_table_size)
c8541ecd
CL		 2774{
2775 struct rpc_xprt *xprt;
ffc2e518 2776 struct sock_xprt *new;
c8541ecd 2777
96802a09 2778 if (args->addrlen > sizeof(xprt->addr)) {
46121cf7 2779 dprintk("RPC: xs_setup_xprt: address too large\n");
c8541ecd
CL		 2780 return ERR_PTR(-EBADF);
2781 }
2782
d9ba131d
TM		 2783 xprt = xprt_alloc(args->net, sizeof(*new), slot_table_size,
2784 max_slot_table_size);
bd1722d4 2785 if (xprt == NULL) {
46121cf7
CL		 2786 dprintk("RPC: xs_setup_xprt: couldn't allocate "
2787 "rpc_xprt\n");
c8541ecd
CL		 2788 return ERR_PTR(-ENOMEM);
2789 }
c8541ecd 2790
bd1722d4 2791 new = container_of(xprt, struct sock_xprt, xprt);
edc1b01c 2792 mutex_init(&new->recv_mutex);
96802a09
FM		 2793 memcpy(&xprt->addr, args->dstaddr, args->addrlen);
2794 xprt->addrlen = args->addrlen;
d3bc9a1d 2795 if (args->srcaddr)
fbfffbd5 2796 memcpy(&new->srcaddr, args->srcaddr, args->addrlen);
92476850
CL		 2797 else {
2798 int err;
2799 err = xs_init_anyaddr(args->dstaddr->sa_family,
2800 (struct sockaddr *)&new->srcaddr);
2aa13531
SK		 2801 if (err != 0) {
2802 xprt_free(xprt);
92476850 2803 return ERR_PTR(err);
2aa13531 2804 }
92476850 2805 }
c8541ecd
CL		 2806
2807 return xprt;
2808}
2809
176e21ee
CL		 2810static const struct rpc_timeout xs_local_default_timeout = {
2811 .to_initval = 10 * HZ,
2812 .to_maxval = 10 * HZ,
2813 .to_retries = 2,
2814};
2815
2816/**
2817 * xs_setup_local - Set up transport to use an AF_LOCAL socket
2818 * @args: rpc transport creation arguments
2819 *
2820 * AF_LOCAL is a "tpi_cots_ord" transport, just like TCP
2821 */
2822static struct rpc_xprt *xs_setup_local(struct xprt_create *args)
2823{
2824 struct sockaddr_un *sun = (struct sockaddr_un *)args->dstaddr;
2825 struct sock_xprt *transport;
2826 struct rpc_xprt *xprt;
2827 struct rpc_xprt *ret;
2828
d9ba131d
TM		 2829 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2830 xprt_max_tcp_slot_table_entries);
176e21ee
CL		 2831 if (IS_ERR(xprt))
2832 return xprt;
2833 transport = container_of(xprt, struct sock_xprt, xprt);
2834
2835 xprt->prot = 0;
2836 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2837 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2838
2839 xprt->bind_timeout = XS_BIND_TO;
2840 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2841 xprt->idle_timeout = XS_IDLE_DISC_TO;
2842
2843 xprt->ops = &xs_local_ops;
2844 xprt->timeout = &xs_local_default_timeout;
2845
a2648094 2846 INIT_WORK(&transport->recv_worker, xs_local_data_receive_workfn);
93dc41bd
N		 2847 INIT_DELAYED_WORK(&transport->connect_worker,
2848 xs_dummy_setup_socket);
2849
176e21ee
CL		 2850 switch (sun->sun_family) {
2851 case AF_LOCAL:
2852 if (sun->sun_path[0] != '/') {
2853 dprintk("RPC: bad AF_LOCAL address: %s\n",
2854 sun->sun_path);
2855 ret = ERR_PTR(-EINVAL);
2856 goto out_err;
2857 }
2858 xprt_set_bound(xprt);
176e21ee 2859 xs_format_peer_addresses(xprt, "local", RPCBIND_NETID_LOCAL);
7073ea87
BF		 2860 ret = ERR_PTR(xs_local_setup_socket(transport));
2861 if (ret)
2862 goto out_err;
176e21ee
CL		 2863 break;
2864 default:
2865 ret = ERR_PTR(-EAFNOSUPPORT);
2866 goto out_err;
2867 }
2868
2869 dprintk("RPC: set up xprt to %s via AF_LOCAL\n",
2870 xprt->address_strings[RPC_DISPLAY_ADDR]);
2871
2872 if (try_module_get(THIS_MODULE))
2873 return xprt;
2874 ret = ERR_PTR(-EINVAL);
2875out_err:
315f3812 2876 xs_xprt_free(xprt);
176e21ee
CL		 2877 return ret;
2878}
2879
2881ae74
TM		 2880static const struct rpc_timeout xs_udp_default_timeout = {
2881 .to_initval = 5 * HZ,
2882 .to_maxval = 30 * HZ,
2883 .to_increment = 5 * HZ,
2884 .to_retries = 5,
2885};
2886
9903cd1c
CL		 2887/**
2888 * xs_setup_udp - Set up transport to use a UDP socket
96802a09 2889 * @args: rpc transport creation arguments
9903cd1c
CL		 2890 *
2891 */
483066d6 2892static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
a246b010 2893{
8f9d5b1a 2894 struct sockaddr *addr = args->dstaddr;
c8541ecd 2895 struct rpc_xprt *xprt;
c8475461 2896 struct sock_xprt *transport;
0a68b0be 2897 struct rpc_xprt *ret;
a246b010 2898
d9ba131d
TM		 2899 xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries,
2900 xprt_udp_slot_table_entries);
c8541ecd
CL		 2901 if (IS_ERR(xprt))
2902 return xprt;
c8475461 2903 transport = container_of(xprt, struct sock_xprt, xprt);
a246b010 2904
ec739ef0 2905 xprt->prot = IPPROTO_UDP;
808012fb 2906 xprt->tsh_size = 0;
a246b010
CL		 2907 /* XXX: header size can vary due to auth type, IPv6, etc. */
2908 xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
2909
03bf4b70 2910 xprt->bind_timeout = XS_BIND_TO;
03bf4b70
CL		 2911 xprt->reestablish_timeout = XS_UDP_REEST_TO;
2912 xprt->idle_timeout = XS_IDLE_DISC_TO;
a246b010 2913
262965f5 2914 xprt->ops = &xs_udp_ops;
a246b010 2915
ba7392bb 2916 xprt->timeout = &xs_udp_default_timeout;
a246b010 2917
f9b2ee71 2918 INIT_WORK(&transport->recv_worker, xs_udp_data_receive_workfn);
edc1b01c
TM		 2919 INIT_DELAYED_WORK(&transport->connect_worker, xs_udp_setup_socket);
2920
8f9d5b1a
CL		 2921 switch (addr->sa_family) {
2922 case AF_INET:
2923 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2924 xprt_set_bound(xprt);
2925
9dc3b095 2926 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
8f9d5b1a
CL		 2927 break;
2928 case AF_INET6:
2929 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2930 xprt_set_bound(xprt);
2931
9dc3b095 2932 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
8f9d5b1a
CL		 2933 break;
2934 default:
0a68b0be
BF		 2935 ret = ERR_PTR(-EAFNOSUPPORT);
2936 goto out_err;
8f9d5b1a
CL		 2937 }
2938
c740eff8
CL		 2939 if (xprt_bound(xprt))
2940 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2941 xprt->address_strings[RPC_DISPLAY_ADDR],
2942 xprt->address_strings[RPC_DISPLAY_PORT],
2943 xprt->address_strings[RPC_DISPLAY_PROTO]);
2944 else
2945 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
2946 xprt->address_strings[RPC_DISPLAY_ADDR],
2947 xprt->address_strings[RPC_DISPLAY_PROTO]);
edb267a6 2948
bc25571e
TT		 2949 if (try_module_get(THIS_MODULE))
2950 return xprt;
0a68b0be
BF		 2951 ret = ERR_PTR(-EINVAL);
2952out_err:
315f3812 2953 xs_xprt_free(xprt);
0a68b0be 2954 return ret;
a246b010
CL		 2955}
2956
2881ae74
TM		 2957static const struct rpc_timeout xs_tcp_default_timeout = {
2958 .to_initval = 60 * HZ,
2959 .to_maxval = 60 * HZ,
2960 .to_retries = 2,
2961};
2962
9903cd1c
CL		 2963/**
2964 * xs_setup_tcp - Set up transport to use a TCP socket
96802a09 2965 * @args: rpc transport creation arguments
9903cd1c
CL		 2966 *
2967 */
483066d6 2968static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
a246b010 2969{
8f9d5b1a 2970 struct sockaddr *addr = args->dstaddr;
c8541ecd 2971 struct rpc_xprt *xprt;
c8475461 2972 struct sock_xprt *transport;
0a68b0be 2973 struct rpc_xprt *ret;
b7993ceb
TM		 2974 unsigned int max_slot_table_size = xprt_max_tcp_slot_table_entries;
2975
2976 if (args->flags & XPRT_CREATE_INFINITE_SLOTS)
2977 max_slot_table_size = RPC_MAX_SLOT_TABLE_LIMIT;
a246b010 2978
d9ba131d 2979 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
b7993ceb 2980 max_slot_table_size);
c8541ecd
CL		 2981 if (IS_ERR(xprt))
2982 return xprt;
c8475461 2983 transport = container_of(xprt, struct sock_xprt, xprt);
a246b010 2984
ec739ef0 2985 xprt->prot = IPPROTO_TCP;
808012fb 2986 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
808012fb 2987 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
a246b010 2988
03bf4b70 2989 xprt->bind_timeout = XS_BIND_TO;
03bf4b70
CL		 2990 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2991 xprt->idle_timeout = XS_IDLE_DISC_TO;
a246b010 2992
262965f5 2993 xprt->ops = &xs_tcp_ops;
ba7392bb 2994 xprt->timeout = &xs_tcp_default_timeout;
a246b010 2995
3851f1cd
TM		 2996 xprt->max_reconnect_timeout = xprt->timeout->to_maxval;
2997
edc1b01c
TM		 2998 INIT_WORK(&transport->recv_worker, xs_tcp_data_receive_workfn);
2999 INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_setup_socket);
3000
8f9d5b1a
CL		 3001 switch (addr->sa_family) {
3002 case AF_INET:
3003 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
3004 xprt_set_bound(xprt);
3005
9dc3b095 3006 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
8f9d5b1a
CL		 3007 break;
3008 case AF_INET6:
3009 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
3010 xprt_set_bound(xprt);
3011
9dc3b095 3012 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
8f9d5b1a
CL		 3013 break;
3014 default:
0a68b0be
BF		 3015 ret = ERR_PTR(-EAFNOSUPPORT);
3016 goto out_err;
8f9d5b1a
CL		 3017 }
3018
c740eff8
CL		 3019 if (xprt_bound(xprt))
3020 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
3021 xprt->address_strings[RPC_DISPLAY_ADDR],
3022 xprt->address_strings[RPC_DISPLAY_PORT],
3023 xprt->address_strings[RPC_DISPLAY_PROTO]);
3024 else
3025 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
3026 xprt->address_strings[RPC_DISPLAY_ADDR],
3027 xprt->address_strings[RPC_DISPLAY_PROTO]);
3028
bc25571e
TT		 3029 if (try_module_get(THIS_MODULE))
3030 return xprt;
0a68b0be
BF		 3031 ret = ERR_PTR(-EINVAL);
3032out_err:
315f3812 3033 xs_xprt_free(xprt);
0a68b0be 3034 return ret;
a246b010 3035}
282b32e1 3036
f300baba
AB		 3037/**
3038 * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket
3039 * @args: rpc transport creation arguments
3040 *
3041 */
3042static struct rpc_xprt *xs_setup_bc_tcp(struct xprt_create *args)
3043{
3044 struct sockaddr *addr = args->dstaddr;
3045 struct rpc_xprt *xprt;
3046 struct sock_xprt *transport;
3047 struct svc_sock *bc_sock;
0a68b0be 3048 struct rpc_xprt *ret;
f300baba 3049
d9ba131d
TM		 3050 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
3051 xprt_tcp_slot_table_entries);
f300baba
AB		 3052 if (IS_ERR(xprt))
3053 return xprt;
3054 transport = container_of(xprt, struct sock_xprt, xprt);
3055
3056 xprt->prot = IPPROTO_TCP;
3057 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
3058 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
3059 xprt->timeout = &xs_tcp_default_timeout;
3060
3061 /* backchannel */
3062 xprt_set_bound(xprt);
3063 xprt->bind_timeout = 0;
f300baba
AB		 3064 xprt->reestablish_timeout = 0;
3065 xprt->idle_timeout = 0;
3066
f300baba
AB		 3067 xprt->ops = &bc_tcp_ops;
3068
3069 switch (addr->sa_family) {
3070 case AF_INET:
3071 xs_format_peer_addresses(xprt, "tcp",
3072 RPCBIND_NETID_TCP);
3073 break;
3074 case AF_INET6:
3075 xs_format_peer_addresses(xprt, "tcp",
3076 RPCBIND_NETID_TCP6);
3077 break;
3078 default:
0a68b0be
BF		 3079 ret = ERR_PTR(-EAFNOSUPPORT);
3080 goto out_err;
f300baba
AB		 3081 }
3082
50fa0d40
PE		 3083 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
3084 xprt->address_strings[RPC_DISPLAY_ADDR],
3085 xprt->address_strings[RPC_DISPLAY_PORT],
3086 xprt->address_strings[RPC_DISPLAY_PROTO]);
f300baba 3087
99de8ea9
BF		 3088 /*
3089 * Once we've associated a backchannel xprt with a connection,
28303ca3
WM		 3090 * we want to keep it around as long as the connection lasts,
3091 * in case we need to start using it for a backchannel again;
3092 * this reference won't be dropped until bc_xprt is destroyed.
99de8ea9
BF		 3093 */
3094 xprt_get(xprt);
3095 args->bc_xprt->xpt_bc_xprt = xprt;
3096 xprt->bc_xprt = args->bc_xprt;
3097 bc_sock = container_of(args->bc_xprt, struct svc_sock, sk_xprt);
3098 transport->sock = bc_sock->sk_sock;
3099 transport->inet = bc_sock->sk_sk;
3100
f300baba
AB		 3101 /*
3102 * Since we don't want connections for the backchannel, we set
3103 * the xprt status to connected
3104 */
3105 xprt_set_connected(xprt);
3106
f300baba
AB		 3107 if (try_module_get(THIS_MODULE))
3108 return xprt;
642aab58
KM		 3109
3110 args->bc_xprt->xpt_bc_xprt = NULL;
39a9beab 3111 args->bc_xprt->xpt_bc_xps = NULL;
99de8ea9 3112 xprt_put(xprt);
0a68b0be
BF		 3113 ret = ERR_PTR(-EINVAL);
3114out_err:
315f3812 3115 xs_xprt_free(xprt);
0a68b0be 3116 return ret;
f300baba
AB		 3117}
3118
176e21ee
CL		 3119static struct xprt_class xs_local_transport = {
3120 .list = LIST_HEAD_INIT(xs_local_transport.list),
3121 .name = "named UNIX socket",
3122 .owner = THIS_MODULE,
3123 .ident = XPRT_TRANSPORT_LOCAL,
3124 .setup = xs_setup_local,
3125};
3126
bc25571e
TT		 3127static struct xprt_class xs_udp_transport = {
3128 .list = LIST_HEAD_INIT(xs_udp_transport.list),
3129 .name = "udp",
3130 .owner = THIS_MODULE,
f300baba 3131 .ident = XPRT_TRANSPORT_UDP,
bc25571e
TT		 3132 .setup = xs_setup_udp,
3133};
3134
3135static struct xprt_class xs_tcp_transport = {
3136 .list = LIST_HEAD_INIT(xs_tcp_transport.list),
3137 .name = "tcp",
3138 .owner = THIS_MODULE,
f300baba 3139 .ident = XPRT_TRANSPORT_TCP,
bc25571e
TT		 3140 .setup = xs_setup_tcp,
3141};
3142
f300baba
AB		 3143static struct xprt_class xs_bc_tcp_transport = {
3144 .list = LIST_HEAD_INIT(xs_bc_tcp_transport.list),
3145 .name = "tcp NFSv4.1 backchannel",
3146 .owner = THIS_MODULE,
3147 .ident = XPRT_TRANSPORT_BC_TCP,
3148 .setup = xs_setup_bc_tcp,
3149};
3150
282b32e1 3151/**
bc25571e 3152 * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
282b32e1
CL		 3153 *
3154 */
3155int init_socket_xprt(void)
3156{
f895b252 3157#if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
2b1bec5f 3158 if (!sunrpc_table_header)
0b4d4147 3159 sunrpc_table_header = register_sysctl_table(sunrpc_table);
fbf76683
CL		 3160#endif
3161
176e21ee 3162 xprt_register_transport(&xs_local_transport);
bc25571e
TT		 3163 xprt_register_transport(&xs_udp_transport);
3164 xprt_register_transport(&xs_tcp_transport);
f300baba 3165 xprt_register_transport(&xs_bc_tcp_transport);
bc25571e 3166
282b32e1
CL		 3167 return 0;
3168}
3169
3170/**
bc25571e 3171 * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
282b32e1
CL		 3172 *
3173 */
3174void cleanup_socket_xprt(void)
3175{
f895b252 3176#if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
fbf76683
CL		 3177 if (sunrpc_table_header) {
3178 unregister_sysctl_table(sunrpc_table_header);
3179 sunrpc_table_header = NULL;
3180 }
3181#endif
bc25571e 3182
176e21ee 3183 xprt_unregister_transport(&xs_local_transport);
bc25571e
TT		 3184 xprt_unregister_transport(&xs_udp_transport);
3185 xprt_unregister_transport(&xs_tcp_transport);
f300baba 3186 xprt_unregister_transport(&xs_bc_tcp_transport);
282b32e1 3187}
cbf11071 3188
9bbb9e5a
RR		 3189static int param_set_uint_minmax(const char *val,
3190 const struct kernel_param *kp,
cbf11071
TM		 3191 unsigned int min, unsigned int max)
3192{
00cfaa94 3193 unsigned int num;
cbf11071
TM		 3194 int ret;
3195
3196 if (!val)
3197 return -EINVAL;
00cfaa94 3198 ret = kstrtouint(val, 0, &num);
cbf11071
TM		 3199 if (ret == -EINVAL || num < min || num > max)
3200 return -EINVAL;
3201 *((unsigned int *)kp->arg) = num;
3202 return 0;
3203}
3204
9bbb9e5a 3205static int param_set_portnr(const char *val, const struct kernel_param *kp)
cbf11071 3206{
ffb6ca33
FS		 3207 if (kp->arg == &xprt_min_resvport)
3208 return param_set_uint_minmax(val, kp,
cbf11071 3209 RPC_MIN_RESVPORT,
ffb6ca33
FS		 3210 xprt_max_resvport);
3211 return param_set_uint_minmax(val, kp,
3212 xprt_min_resvport,
cbf11071
TM		 3213 RPC_MAX_RESVPORT);
3214}
3215
9c27847d 3216static const struct kernel_param_ops param_ops_portnr = {
9bbb9e5a
RR		 3217 .set = param_set_portnr,
3218 .get = param_get_uint,
3219};
3220
cbf11071
TM		 3221#define param_check_portnr(name, p) \
3222 __param_check(name, p, unsigned int);
3223
3224module_param_named(min_resvport, xprt_min_resvport, portnr, 0644);
3225module_param_named(max_resvport, xprt_max_resvport, portnr, 0644);
3226
9bbb9e5a
RR		 3227static int param_set_slot_table_size(const char *val,
3228 const struct kernel_param *kp)
cbf11071
TM		 3229{
3230 return param_set_uint_minmax(val, kp,
3231 RPC_MIN_SLOT_TABLE,
3232 RPC_MAX_SLOT_TABLE);
3233}
3234
9c27847d 3235static const struct kernel_param_ops param_ops_slot_table_size = {
9bbb9e5a
RR		 3236 .set = param_set_slot_table_size,
3237 .get = param_get_uint,
3238};
3239
cbf11071
TM		 3240#define param_check_slot_table_size(name, p) \
3241 __param_check(name, p, unsigned int);
3242
d9ba131d
TM		 3243static int param_set_max_slot_table_size(const char *val,
3244 const struct kernel_param *kp)
3245{
3246 return param_set_uint_minmax(val, kp,
3247 RPC_MIN_SLOT_TABLE,
3248 RPC_MAX_SLOT_TABLE_LIMIT);
3249}
3250
9c27847d 3251static const struct kernel_param_ops param_ops_max_slot_table_size = {
d9ba131d
TM		 3252 .set = param_set_max_slot_table_size,
3253 .get = param_get_uint,
3254};
3255
3256#define param_check_max_slot_table_size(name, p) \
3257 __param_check(name, p, unsigned int);
3258
cbf11071
TM		 3259module_param_named(tcp_slot_table_entries, xprt_tcp_slot_table_entries,
3260 slot_table_size, 0644);
d9ba131d
TM		 3261module_param_named(tcp_max_slot_table_entries, xprt_max_tcp_slot_table_entries,
3262 max_slot_table_size, 0644);
cbf11071
TM		 3263module_param_named(udp_slot_table_entries, xprt_udp_slot_table_entries,
3264 slot_table_size, 0644);
3265
Linux kernel - Deliverables
This page took 1.012937 seconds and 5 git commands to generate.