projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Revert "pinctrl: digicolor: add missing platform_set_drvdata() call"
[linux-2.6-block.git] / net / sunrpc / xprtsock.c
CommitLineData
a246b010
CL		 1/*
2 * linux/net/sunrpc/xprtsock.c
3 *
4 * Client-side transport implementation for sockets.
5 *
113aa838
AC		 6 * TCP callback races fixes (C) 1998 Red Hat
7 * TCP send fixes (C) 1998 Red Hat
a246b010
CL		 8 * TCP NFS related read + write fixes
9 * (C) 1999 Dave Airlie, University of Limerick, Ireland <airlied@linux.ie>
10 *
11 * Rewrite of larges part of the code in order to stabilize TCP stuff.
12 * Fix behaviour when socket buffer is full.
13 * (C) 1999 Trond Myklebust <trond.myklebust@fys.uio.no>
55aa4f58
CL		 14 *
15 * IP socket transport implementation, (C) 2005 Chuck Lever <cel@netapp.com>
8f9d5b1a
CL		 16 *
17 * IPv6 support contributed by Gilles Quillard, Bull Open Source, 2005.
18 * <gilles.quillard@bull.net>
a246b010
CL		 19 */
20
21#include <linux/types.h>
176e21ee 22#include <linux/string.h>
a246b010 23#include <linux/slab.h>
bc25571e 24#include <linux/module.h>
a246b010 25#include <linux/capability.h>
a246b010
CL		 26#include <linux/pagemap.h>
27#include <linux/errno.h>
28#include <linux/socket.h>
29#include <linux/in.h>
30#include <linux/net.h>
31#include <linux/mm.h>
176e21ee 32#include <linux/un.h>
a246b010
CL		 33#include <linux/udp.h>
34#include <linux/tcp.h>
35#include <linux/sunrpc/clnt.h>
5976687a 36#include <linux/sunrpc/addr.h>
02107148 37#include <linux/sunrpc/sched.h>
4cfc7e60 38#include <linux/sunrpc/svcsock.h>
49c36fcc 39#include <linux/sunrpc/xprtsock.h>
a246b010 40#include <linux/file.h>
9e00abc3 41#ifdef CONFIG_SUNRPC_BACKCHANNEL
44b98efd
RL		 42#include <linux/sunrpc/bc_xprt.h>
43#endif
a246b010
CL		 44
45#include <net/sock.h>
46#include <net/checksum.h>
47#include <net/udp.h>
48#include <net/tcp.h>
49
40b5ea0c
TM		 50#include <trace/events/sunrpc.h>
51
4cfc7e60 52#include "sunrpc.h"
176e21ee
CL		 53
54static void xs_close(struct rpc_xprt *xprt);
55
c556b754
CL		 56/*
57 * xprtsock tunables
58 */
09acfea5
TM		 59static unsigned int xprt_udp_slot_table_entries = RPC_DEF_SLOT_TABLE;
60static unsigned int xprt_tcp_slot_table_entries = RPC_MIN_SLOT_TABLE;
61static unsigned int xprt_max_tcp_slot_table_entries = RPC_MAX_SLOT_TABLE;
c556b754 62
09acfea5
TM		 63static unsigned int xprt_min_resvport = RPC_DEF_MIN_RESVPORT;
64static unsigned int xprt_max_resvport = RPC_DEF_MAX_RESVPORT;
c556b754 65
54c09874
TM		 66#if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
67
7d1e8255 68#define XS_TCP_LINGER_TO (15U * HZ)
25fe6142 69static unsigned int xs_tcp_fin_timeout __read_mostly = XS_TCP_LINGER_TO;
7d1e8255 70
fbf76683
CL		 71/*
72 * We can register our own files under /proc/sys/sunrpc by
73 * calling register_sysctl_table() again. The files in that
74 * directory become the union of all files registered there.
75 *
76 * We simply need to make sure that we don't collide with
77 * someone else's file names!
78 */
79
fbf76683
CL		 80static unsigned int min_slot_table_size = RPC_MIN_SLOT_TABLE;
81static unsigned int max_slot_table_size = RPC_MAX_SLOT_TABLE;
d9ba131d 82static unsigned int max_tcp_slot_table_limit = RPC_MAX_SLOT_TABLE_LIMIT;
fbf76683
CL		 83static unsigned int xprt_min_resvport_limit = RPC_MIN_RESVPORT;
84static unsigned int xprt_max_resvport_limit = RPC_MAX_RESVPORT;
85
86static struct ctl_table_header *sunrpc_table_header;
87
88/*
89 * FIXME: changing the UDP slot table size should also resize the UDP
90 * socket buffers for existing UDP transports
91 */
fe2c6338 92static struct ctl_table xs_tunables_table[] = {
fbf76683 93 {
fbf76683
CL		 94 .procname = "udp_slot_table_entries",
95 .data = &xprt_udp_slot_table_entries,
96 .maxlen = sizeof(unsigned int),
97 .mode = 0644,
6d456111 98 .proc_handler = proc_dointvec_minmax,
fbf76683
CL		 99 .extra1 = &min_slot_table_size,
100 .extra2 = &max_slot_table_size
101 },
102 {
fbf76683
CL		 103 .procname = "tcp_slot_table_entries",
104 .data = &xprt_tcp_slot_table_entries,
105 .maxlen = sizeof(unsigned int),
106 .mode = 0644,
6d456111 107 .proc_handler = proc_dointvec_minmax,
fbf76683
CL		 108 .extra1 = &min_slot_table_size,
109 .extra2 = &max_slot_table_size
110 },
d9ba131d
TM		 111 {
112 .procname = "tcp_max_slot_table_entries",
113 .data = &xprt_max_tcp_slot_table_entries,
114 .maxlen = sizeof(unsigned int),
115 .mode = 0644,
116 .proc_handler = proc_dointvec_minmax,
117 .extra1 = &min_slot_table_size,
118 .extra2 = &max_tcp_slot_table_limit
119 },
fbf76683 120 {
fbf76683
CL		 121 .procname = "min_resvport",
122 .data = &xprt_min_resvport,
123 .maxlen = sizeof(unsigned int),
124 .mode = 0644,
6d456111 125 .proc_handler = proc_dointvec_minmax,
fbf76683
CL		 126 .extra1 = &xprt_min_resvport_limit,
127 .extra2 = &xprt_max_resvport_limit
128 },
129 {
fbf76683
CL		 130 .procname = "max_resvport",
131 .data = &xprt_max_resvport,
132 .maxlen = sizeof(unsigned int),
133 .mode = 0644,
6d456111 134 .proc_handler = proc_dointvec_minmax,
fbf76683
CL		 135 .extra1 = &xprt_min_resvport_limit,
136 .extra2 = &xprt_max_resvport_limit
137 },
25fe6142
TM		 138 {
139 .procname = "tcp_fin_timeout",
140 .data = &xs_tcp_fin_timeout,
141 .maxlen = sizeof(xs_tcp_fin_timeout),
142 .mode = 0644,
6d456111 143 .proc_handler = proc_dointvec_jiffies,
fbf76683 144 },
f8572d8f 145 { },
fbf76683
CL		 146};
147
fe2c6338 148static struct ctl_table sunrpc_table[] = {
fbf76683 149 {
fbf76683
CL		 150 .procname = "sunrpc",
151 .mode = 0555,
152 .child = xs_tunables_table
153 },
f8572d8f 154 { },
fbf76683
CL		 155};
156
157#endif
158
03bf4b70
CL		 159/*
160 * Wait duration for a reply from the RPC portmapper.
161 */
162#define XS_BIND_TO (60U * HZ)
163
164/*
165 * Delay if a UDP socket connect error occurs. This is most likely some
166 * kind of resource problem on the local host.
167 */
168#define XS_UDP_REEST_TO (2U * HZ)
169
170/*
171 * The reestablish timeout allows clients to delay for a bit before attempting
172 * to reconnect to a server that just dropped our connection.
173 *
174 * We implement an exponential backoff when trying to reestablish a TCP
175 * transport connection with the server. Some servers like to drop a TCP
176 * connection when they are overworked, so we start with a short timeout and
177 * increase over time if the server is down or not responding.
178 */
179#define XS_TCP_INIT_REEST_TO (3U * HZ)
180#define XS_TCP_MAX_REEST_TO (5U * 60 * HZ)
181
182/*
183 * TCP idle timeout; client drops the transport socket if it is idle
184 * for this long. Note that we also timeout UDP sockets to prevent
185 * holding port numbers when there is no RPC traffic.
186 */
187#define XS_IDLE_DISC_TO (5U * 60 * HZ)
188
f895b252 189#if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
a246b010 190# undef RPC_DEBUG_DATA
9903cd1c 191# define RPCDBG_FACILITY RPCDBG_TRANS
a246b010
CL		 192#endif
193
a246b010 194#ifdef RPC_DEBUG_DATA
9903cd1c 195static void xs_pktdump(char *msg, u32 *packet, unsigned int count)
a246b010 196{
9903cd1c
CL		 197 u8 *buf = (u8 *) packet;
198 int j;
a246b010 199
46121cf7 200 dprintk("RPC: %s\n", msg);
a246b010
CL		 201 for (j = 0; j < count && j < 128; j += 4) {
202 if (!(j & 31)) {
203 if (j)
204 dprintk("\n");
205 dprintk("0x%04x ", j);
206 }
207 dprintk("%02x%02x%02x%02x ",
208 buf[j], buf[j+1], buf[j+2], buf[j+3]);
209 }
210 dprintk("\n");
211}
212#else
9903cd1c 213static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count)
a246b010
CL		 214{
215 /* NOP */
216}
217#endif
218
2118071d
TM		 219static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
220{
221 return (struct rpc_xprt *) sk->sk_user_data;
222}
223
95392c59
CL		 224static inline struct sockaddr *xs_addr(struct rpc_xprt *xprt)
225{
226 return (struct sockaddr *) &xprt->addr;
227}
228
176e21ee
CL		 229static inline struct sockaddr_un *xs_addr_un(struct rpc_xprt *xprt)
230{
231 return (struct sockaddr_un *) &xprt->addr;
232}
233
95392c59 234static inline struct sockaddr_in *xs_addr_in(struct rpc_xprt *xprt)
edb267a6 235{
95392c59
CL		 236 return (struct sockaddr_in *) &xprt->addr;
237}
238
239static inline struct sockaddr_in6 *xs_addr_in6(struct rpc_xprt *xprt)
240{
241 return (struct sockaddr_in6 *) &xprt->addr;
242}
243
c877b849 244static void xs_format_common_peer_addresses(struct rpc_xprt *xprt)
edb267a6 245{
c877b849 246 struct sockaddr *sap = xs_addr(xprt);
9dc3b095
CL		 247 struct sockaddr_in6 *sin6;
248 struct sockaddr_in *sin;
176e21ee 249 struct sockaddr_un *sun;
c877b849 250 char buf[128];
edb267a6 251
9dc3b095 252 switch (sap->sa_family) {
176e21ee
CL		 253 case AF_LOCAL:
254 sun = xs_addr_un(xprt);
255 strlcpy(buf, sun->sun_path, sizeof(buf));
256 xprt->address_strings[RPC_DISPLAY_ADDR] =
257 kstrdup(buf, GFP_KERNEL);
258 break;
9dc3b095 259 case AF_INET:
176e21ee
CL		 260 (void)rpc_ntop(sap, buf, sizeof(buf));
261 xprt->address_strings[RPC_DISPLAY_ADDR] =
262 kstrdup(buf, GFP_KERNEL);
9dc3b095 263 sin = xs_addr_in(xprt);
fc0b5791 264 snprintf(buf, sizeof(buf), "%08x", ntohl(sin->sin_addr.s_addr));
9dc3b095
CL		 265 break;
266 case AF_INET6:
176e21ee
CL		 267 (void)rpc_ntop(sap, buf, sizeof(buf));
268 xprt->address_strings[RPC_DISPLAY_ADDR] =
269 kstrdup(buf, GFP_KERNEL);
9dc3b095 270 sin6 = xs_addr_in6(xprt);
fc0b5791 271 snprintf(buf, sizeof(buf), "%pi6", &sin6->sin6_addr);
9dc3b095
CL		 272 break;
273 default:
274 BUG();
756805e7 275 }
176e21ee 276
9dc3b095 277 xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = kstrdup(buf, GFP_KERNEL);
edb267a6
CL		 278}
279
9dc3b095 280static void xs_format_common_peer_ports(struct rpc_xprt *xprt)
4b6473fb 281{
9dc3b095
CL		 282 struct sockaddr *sap = xs_addr(xprt);
283 char buf[128];
4b6473fb 284
81160e66 285 snprintf(buf, sizeof(buf), "%u", rpc_get_port(sap));
c877b849 286 xprt->address_strings[RPC_DISPLAY_PORT] = kstrdup(buf, GFP_KERNEL);
4b6473fb 287
81160e66 288 snprintf(buf, sizeof(buf), "%4hx", rpc_get_port(sap));
c877b849
CL		 289 xprt->address_strings[RPC_DISPLAY_HEX_PORT] = kstrdup(buf, GFP_KERNEL);
290}
4b6473fb 291
9dc3b095
CL		 292static void xs_format_peer_addresses(struct rpc_xprt *xprt,
293 const char *protocol,
294 const char *netid)
c877b849 295{
b454ae90 296 xprt->address_strings[RPC_DISPLAY_PROTO] = protocol;
b454ae90 297 xprt->address_strings[RPC_DISPLAY_NETID] = netid;
c877b849 298 xs_format_common_peer_addresses(xprt);
9dc3b095 299 xs_format_common_peer_ports(xprt);
edb267a6 300}
4b6473fb 301
9dc3b095 302static void xs_update_peer_port(struct rpc_xprt *xprt)
4b6473fb 303{
9dc3b095
CL		 304 kfree(xprt->address_strings[RPC_DISPLAY_HEX_PORT]);
305 kfree(xprt->address_strings[RPC_DISPLAY_PORT]);
4417c8c4 306
9dc3b095 307 xs_format_common_peer_ports(xprt);
edb267a6
CL		 308}
309
310static void xs_free_peer_addresses(struct rpc_xprt *xprt)
311{
33e01dc7
CL		 312 unsigned int i;
313
314 for (i = 0; i < RPC_DISPLAY_MAX; i++)
315 switch (i) {
316 case RPC_DISPLAY_PROTO:
317 case RPC_DISPLAY_NETID:
318 continue;
319 default:
320 kfree(xprt->address_strings[i]);
321 }
edb267a6
CL		 322}
323
b4b5cc85
CL		 324#define XS_SENDMSG_FLAGS (MSG_DONTWAIT | MSG_NOSIGNAL)
325
24c5684b 326static int xs_send_kvec(struct socket *sock, struct sockaddr *addr, int addrlen, struct kvec *vec, unsigned int base, int more)
b4b5cc85 327{
b4b5cc85
CL		 328 struct msghdr msg = {
329 .msg_name = addr,
330 .msg_namelen = addrlen,
24c5684b
TM		 331 .msg_flags = XS_SENDMSG_FLAGS | (more ? MSG_MORE : 0),
332 };
333 struct kvec iov = {
334 .iov_base = vec->iov_base + base,
335 .iov_len = vec->iov_len - base,
b4b5cc85
CL		 336 };
337
24c5684b 338 if (iov.iov_len != 0)
b4b5cc85
CL		 339 return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
340 return kernel_sendmsg(sock, &msg, NULL, 0, 0);
341}
342
f279cd00 343static int xs_send_pagedata(struct socket *sock, struct xdr_buf *xdr, unsigned int base, int more, bool zerocopy, int *sent_p)
b4b5cc85 344{
a6b31d18
TM		 345 ssize_t (*do_sendpage)(struct socket *sock, struct page *page,
346 int offset, size_t size, int flags);
24c5684b
TM		 347 struct page **ppage;
348 unsigned int remainder;
f279cd00 349 int err;
24c5684b
TM		 350
351 remainder = xdr->page_len - base;
352 base += xdr->page_base;
353 ppage = xdr->pages + (base >> PAGE_SHIFT);
354 base &= ~PAGE_MASK;
a6b31d18
TM		 355 do_sendpage = sock->ops->sendpage;
356 if (!zerocopy)
357 do_sendpage = sock_no_sendpage;
24c5684b
TM		 358 for(;;) {
359 unsigned int len = min_t(unsigned int, PAGE_SIZE - base, remainder);
360 int flags = XS_SENDMSG_FLAGS;
b4b5cc85 361
24c5684b 362 remainder -= len;
31303d6c 363 if (more)
24c5684b 364 flags |= MSG_MORE;
31303d6c
TM		 365 if (remainder != 0)
366 flags |= MSG_SENDPAGE_NOTLAST | MSG_MORE;
a6b31d18 367 err = do_sendpage(sock, *ppage, base, len, flags);
24c5684b
TM		 368 if (remainder == 0 || err != len)
369 break;
f279cd00 370 *sent_p += err;
24c5684b
TM		 371 ppage++;
372 base = 0;
373 }
f279cd00
JB		 374 if (err > 0) {
375 *sent_p += err;
376 err = 0;
377 }
378 return err;
b4b5cc85
CL		 379}
380
9903cd1c
CL		 381/**
382 * xs_sendpages - write pages directly to a socket
383 * @sock: socket to send on
384 * @addr: UDP only -- address of destination
385 * @addrlen: UDP only -- length of destination address
386 * @xdr: buffer containing this request
387 * @base: starting position in the buffer
a6b31d18 388 * @zerocopy: true if it is safe to use sendpage()
f279cd00 389 * @sent_p: return the total number of bytes successfully queued for sending
9903cd1c 390 *
a246b010 391 */
f279cd00 392static int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base, bool zerocopy, int *sent_p)
a246b010 393{
24c5684b 394 unsigned int remainder = xdr->len - base;
f279cd00
JB		 395 int err = 0;
396 int sent = 0;
a246b010 397
262965f5 398 if (unlikely(!sock))
fba91afb 399 return -ENOTSOCK;
262965f5 400
24c5684b
TM		 401 if (base != 0) {
402 addr = NULL;
403 addrlen = 0;
404 }
262965f5 405
24c5684b
TM		 406 if (base < xdr->head[0].iov_len || addr != NULL) {
407 unsigned int len = xdr->head[0].iov_len - base;
408 remainder -= len;
409 err = xs_send_kvec(sock, addr, addrlen, &xdr->head[0], base, remainder != 0);
410 if (remainder == 0 || err != len)
a246b010 411 goto out;
f279cd00 412 *sent_p += err;
a246b010
CL		 413 base = 0;
414 } else
24c5684b 415 base -= xdr->head[0].iov_len;
a246b010 416
24c5684b
TM		 417 if (base < xdr->page_len) {
418 unsigned int len = xdr->page_len - base;
419 remainder -= len;
f279cd00
JB		 420 err = xs_send_pagedata(sock, xdr, base, remainder != 0, zerocopy, &sent);
421 *sent_p += sent;
422 if (remainder == 0 || sent != len)
a246b010
CL		 423 goto out;
424 base = 0;
24c5684b
TM		 425 } else
426 base -= xdr->page_len;
427
428 if (base >= xdr->tail[0].iov_len)
f279cd00 429 return 0;
24c5684b 430 err = xs_send_kvec(sock, NULL, 0, &xdr->tail[0], base, 0);
a246b010 431out:
f279cd00
JB		 432 if (err > 0) {
433 *sent_p += err;
434 err = 0;
435 }
436 return err;
a246b010
CL		 437}
438
b6ddf64f
TM		 439static void xs_nospace_callback(struct rpc_task *task)
440{
441 struct sock_xprt *transport = container_of(task->tk_rqstp->rq_xprt, struct sock_xprt, xprt);
442
443 transport->inet->sk_write_pending--;
b6ddf64f
TM		 444}
445
9903cd1c 446/**
262965f5
CL		 447 * xs_nospace - place task on wait queue if transmit was incomplete
448 * @task: task to put to sleep
9903cd1c 449 *
a246b010 450 */
5e3771ce 451static int xs_nospace(struct rpc_task *task)
a246b010 452{
262965f5
CL		 453 struct rpc_rqst *req = task->tk_rqstp;
454 struct rpc_xprt *xprt = req->rq_xprt;
ee0ac0c2 455 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
06ea0bfe 456 struct sock *sk = transport->inet;
24ca9a84 457 int ret = -EAGAIN;
a246b010 458
46121cf7 459 dprintk("RPC: %5u xmit incomplete (%u left of %u)\n",
262965f5
CL		 460 task->tk_pid, req->rq_slen - req->rq_bytes_sent,
461 req->rq_slen);
462
b6ddf64f
TM		 463 /* Protect against races with write_space */
464 spin_lock_bh(&xprt->transport_lock);
465
466 /* Don't race with disconnect */
467 if (xprt_connected(xprt)) {
13331a55
TM		 468 /* wait for more buffer space */
469 sk->sk_write_pending++;
470 xprt_wait_for_buffer_space(task, xs_nospace_callback);
471 } else
5e3771ce 472 ret = -ENOTCONN;
262965f5 473
b6ddf64f 474 spin_unlock_bh(&xprt->transport_lock);
06ea0bfe
TM		 475
476 /* Race breaker in case memory is freed before above code is called */
477 sk->sk_write_space(sk);
5e3771ce 478 return ret;
262965f5
CL		 479}
480
61677eee
CL		 481/*
482 * Construct a stream transport record marker in @buf.
483 */
484static inline void xs_encode_stream_record_marker(struct xdr_buf *buf)
485{
486 u32 reclen = buf->len - sizeof(rpc_fraghdr);
487 rpc_fraghdr *base = buf->head[0].iov_base;
488 *base = cpu_to_be32(RPC_LAST_STREAM_FRAGMENT | reclen);
489}
490
176e21ee
CL		 491/**
492 * xs_local_send_request - write an RPC request to an AF_LOCAL socket
493 * @task: RPC task that manages the state of an RPC request
494 *
495 * Return values:
496 * 0: The request has been sent
497 * EAGAIN: The socket was blocked, please call again later to
498 * complete the request
499 * ENOTCONN: Caller needs to invoke connect logic then call again
500 * other: Some other error occured, the request was not sent
501 */
502static int xs_local_send_request(struct rpc_task *task)
503{
504 struct rpc_rqst *req = task->tk_rqstp;
505 struct rpc_xprt *xprt = req->rq_xprt;
506 struct sock_xprt *transport =
507 container_of(xprt, struct sock_xprt, xprt);
508 struct xdr_buf *xdr = &req->rq_snd_buf;
509 int status;
f279cd00 510 int sent = 0;
176e21ee
CL		 511
512 xs_encode_stream_record_marker(&req->rq_snd_buf);
513
514 xs_pktdump("packet data:",
515 req->rq_svec->iov_base, req->rq_svec->iov_len);
516
f279cd00
JB		 517 status = xs_sendpages(transport->sock, NULL, 0, xdr, req->rq_bytes_sent,
518 true, &sent);
176e21ee
CL		 519 dprintk("RPC: %s(%u) = %d\n",
520 __func__, xdr->len - req->rq_bytes_sent, status);
743c69e7
N		 521
522 if (status == -EAGAIN && sock_writeable(transport->inet))
523 status = -ENOBUFS;
524
f279cd00
JB		 525 if (likely(sent > 0) || status == 0) {
526 req->rq_bytes_sent += sent;
527 req->rq_xmit_bytes_sent += sent;
176e21ee
CL		 528 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
529 req->rq_bytes_sent = 0;
530 return 0;
531 }
532 status = -EAGAIN;
533 }
534
535 switch (status) {
3601c4a9 536 case -ENOBUFS:
b5872f0c 537 break;
176e21ee
CL		 538 case -EAGAIN:
539 status = xs_nospace(task);
540 break;
541 default:
542 dprintk("RPC: sendmsg returned unrecognized error %d\n",
543 -status);
544 case -EPIPE:
545 xs_close(xprt);
546 status = -ENOTCONN;
547 }
548
549 return status;
550}
551
262965f5
CL		 552/**
553 * xs_udp_send_request - write an RPC request to a UDP socket
554 * @task: address of RPC task that manages the state of an RPC request
555 *
556 * Return values:
557 * 0: The request has been sent
558 * EAGAIN: The socket was blocked, please call again later to
559 * complete the request
560 * ENOTCONN: Caller needs to invoke connect logic then call again
25985edc 561 * other: Some other error occurred, the request was not sent
262965f5
CL		 562 */
563static int xs_udp_send_request(struct rpc_task *task)
564{
565 struct rpc_rqst *req = task->tk_rqstp;
566 struct rpc_xprt *xprt = req->rq_xprt;
ee0ac0c2 567 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
262965f5 568 struct xdr_buf *xdr = &req->rq_snd_buf;
f279cd00 569 int sent = 0;
262965f5 570 int status;
a246b010 571
9903cd1c 572 xs_pktdump("packet data:",
a246b010
CL		 573 req->rq_svec->iov_base,
574 req->rq_svec->iov_len);
575
01d37c42
TM		 576 if (!xprt_bound(xprt))
577 return -ENOTCONN;
f279cd00
JB		 578 status = xs_sendpages(transport->sock, xs_addr(xprt), xprt->addrlen,
579 xdr, req->rq_bytes_sent, true, &sent);
a246b010 580
46121cf7 581 dprintk("RPC: xs_udp_send_request(%u) = %d\n",
262965f5 582 xdr->len - req->rq_bytes_sent, status);
a246b010 583
3dedbb5c
JB		 584 /* firewall is blocking us, don't return -EAGAIN or we end up looping */
585 if (status == -EPERM)
586 goto process_status;
587
743c69e7
N		 588 if (status == -EAGAIN && sock_writeable(transport->inet))
589 status = -ENOBUFS;
590
f279cd00
JB		 591 if (sent > 0 || status == 0) {
592 req->rq_xmit_bytes_sent += sent;
593 if (sent >= req->rq_slen)
2199700f
TM		 594 return 0;
595 /* Still some bytes left; set up for a retry later. */
262965f5 596 status = -EAGAIN;
2199700f 597 }
a246b010 598
3dedbb5c 599process_status:
262965f5 600 switch (status) {
fba91afb
TM		 601 case -ENOTSOCK:
602 status = -ENOTCONN;
603 /* Should we call xs_close() here? */
604 break;
b6ddf64f 605 case -EAGAIN:
5e3771ce 606 status = xs_nospace(task);
b6ddf64f 607 break;
262965f5 608 case -ENETUNREACH:
3601c4a9 609 case -ENOBUFS:
262965f5 610 case -EPIPE:
a246b010 611 case -ECONNREFUSED:
3dedbb5c 612 case -EPERM:
a246b010 613 /* When the server has died, an ICMP port unreachable message
9903cd1c 614 * prompts ECONNREFUSED. */
13331a55
TM		 615 break;
616 default:
617 dprintk("RPC: sendmsg returned unrecognized error %d\n",
618 -status);
a246b010 619 }
5fe46e9d 620
262965f5 621 return status;
a246b010
CL		 622}
623
9903cd1c 624/**
262965f5 625 * xs_tcp_send_request - write an RPC request to a TCP socket
9903cd1c
CL		 626 * @task: address of RPC task that manages the state of an RPC request
627 *
628 * Return values:
262965f5
CL		 629 * 0: The request has been sent
630 * EAGAIN: The socket was blocked, please call again later to
631 * complete the request
632 * ENOTCONN: Caller needs to invoke connect logic then call again
25985edc 633 * other: Some other error occurred, the request was not sent
9903cd1c
CL		 634 *
635 * XXX: In the case of soft timeouts, should we eventually give up
262965f5 636 * if sendmsg is not able to make progress?
9903cd1c 637 */
262965f5 638static int xs_tcp_send_request(struct rpc_task *task)
a246b010
CL		 639{
640 struct rpc_rqst *req = task->tk_rqstp;
641 struct rpc_xprt *xprt = req->rq_xprt;
ee0ac0c2 642 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
262965f5 643 struct xdr_buf *xdr = &req->rq_snd_buf;
a6b31d18 644 bool zerocopy = true;
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
f580dd04
TM		 680 if (status < 0)
681 break;
682 if (sent == 0) {
683 status = -EAGAIN;
684 break;
685 }
a246b010 686 }
743c69e7
N		 687 if (status == -EAGAIN && sk_stream_is_writeable(transport->inet))
688 status = -ENOBUFS;
a246b010 689
262965f5 690 switch (status) {
fba91afb
TM		 691 case -ENOTSOCK:
692 status = -ENOTCONN;
693 /* Should we call xs_close() here? */
694 break;
262965f5 695 case -EAGAIN:
5e3771ce 696 status = xs_nospace(task);
262965f5 697 break;
262965f5 698 case -ECONNRESET:
2a9e1cfa 699 case -ECONNREFUSED:
262965f5 700 case -ENOTCONN:
3913c78c 701 case -EADDRINUSE:
b5872f0c 702 case -ENOBUFS:
b9d2bb2e 703 case -EPIPE:
13331a55
TM		 704 break;
705 default:
706 dprintk("RPC: sendmsg returned unrecognized error %d\n",
707 -status);
a246b010 708 }
5fe46e9d 709
a246b010
CL		 710 return status;
711}
712
e0ab53de
TM		 713/**
714 * xs_tcp_release_xprt - clean up after a tcp transmission
715 * @xprt: transport
716 * @task: rpc task
717 *
718 * This cleans up if an error causes us to abort the transmission of a request.
719 * In this case, the socket may need to be reset in order to avoid confusing
720 * the server.
721 */
722static void xs_tcp_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
723{
724 struct rpc_rqst *req;
725
726 if (task != xprt->snd_task)
727 return;
728 if (task == NULL)
729 goto out_release;
730 req = task->tk_rqstp;
43cedbf0
TM		 731 if (req == NULL)
732 goto out_release;
e0ab53de
TM		 733 if (req->rq_bytes_sent == 0)
734 goto out_release;
735 if (req->rq_bytes_sent == req->rq_snd_buf.len)
736 goto out_release;
a4f0835c 737 set_bit(XPRT_CLOSE_WAIT, &xprt->state);
e0ab53de
TM		 738out_release:
739 xprt_release_xprt(xprt, task);
740}
741
2a9e1cfa
TM		 742static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk)
743{
744 transport->old_data_ready = sk->sk_data_ready;
745 transport->old_state_change = sk->sk_state_change;
746 transport->old_write_space = sk->sk_write_space;
2118071d 747 transport->old_error_report = sk->sk_error_report;
2a9e1cfa
TM		 748}
749
750static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk)
751{
752 sk->sk_data_ready = transport->old_data_ready;
753 sk->sk_state_change = transport->old_state_change;
754 sk->sk_write_space = transport->old_write_space;
2118071d
TM		 755 sk->sk_error_report = transport->old_error_report;
756}
757
b70ae915
TM		 758static void xs_sock_reset_connection_flags(struct rpc_xprt *xprt)
759{
760 smp_mb__before_atomic();
761 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
762 clear_bit(XPRT_CLOSING, &xprt->state);
763 smp_mb__after_atomic();
764}
765
766static void xs_sock_mark_closed(struct rpc_xprt *xprt)
767{
768 xs_sock_reset_connection_flags(xprt);
769 /* Mark transport as closed and wake up all pending tasks */
770 xprt_disconnect_done(xprt);
771}
772
2118071d
TM		 773/**
774 * xs_error_report - callback to handle TCP socket state errors
775 * @sk: socket
776 *
777 * Note: we don't call sock_error() since there may be a rpc_task
778 * using the socket, and so we don't want to clear sk->sk_err.
779 */
780static void xs_error_report(struct sock *sk)
781{
782 struct rpc_xprt *xprt;
783 int err;
784
785 read_lock_bh(&sk->sk_callback_lock);
786 if (!(xprt = xprt_from_sock(sk)))
787 goto out;
788
789 err = -sk->sk_err;
790 if (err == 0)
791 goto out;
b70ae915
TM		 792 /* Is this a reset event? */
793 if (sk->sk_state == TCP_CLOSE)
794 xs_sock_mark_closed(xprt);
2118071d
TM		 795 dprintk("RPC: xs_error_report client %p, error=%d...\n",
796 xprt, -err);
e8353c76 797 trace_rpc_socket_error(xprt, sk->sk_socket, err);
2118071d
TM		 798 xprt_wake_pending_tasks(xprt, err);
799 out:
800 read_unlock_bh(&sk->sk_callback_lock);
2a9e1cfa
TM		 801}
802
fe315e76 803static void xs_reset_transport(struct sock_xprt *transport)
a246b010 804{
ee0ac0c2
CL		 805 struct socket *sock = transport->sock;
806 struct sock *sk = transport->inet;
6cc7e908 807 struct rpc_xprt *xprt = &transport->xprt;
a246b010 808
fe315e76
CL		 809 if (sk == NULL)
810 return;
9903cd1c 811
264d1df3
JL		 812 if (atomic_read(&transport->xprt.swapper))
813 sk_clear_memalloc(sk);
814
09939204
TM		 815 kernel_sock_shutdown(sock, SHUT_RDWR);
816
edc1b01c 817 mutex_lock(&transport->recv_mutex);
a246b010 818 write_lock_bh(&sk->sk_callback_lock);
ee0ac0c2
CL		 819 transport->inet = NULL;
820 transport->sock = NULL;
a246b010 821
9903cd1c 822 sk->sk_user_data = NULL;
2a9e1cfa
TM		 823
824 xs_restore_old_callbacks(transport, sk);
0c78789e 825 xprt_clear_connected(xprt);
a246b010 826 write_unlock_bh(&sk->sk_callback_lock);
6cc7e908 827 xs_sock_reset_connection_flags(xprt);
edc1b01c 828 mutex_unlock(&transport->recv_mutex);
a246b010 829
6cc7e908 830 trace_rpc_socket_close(xprt, sock);
a246b010 831 sock_release(sock);
fe315e76
CL		 832}
833
834/**
835 * xs_close - close a socket
836 * @xprt: transport
837 *
838 * This is used when all requests are complete; ie, no DRC state remains
839 * on the server we want to save.
f75e6745
TM		 840 *
841 * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with
842 * xs_reset_transport() zeroing the socket from underneath a writer.
fe315e76
CL		 843 */
844static void xs_close(struct rpc_xprt *xprt)
845{
846 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
847
848 dprintk("RPC: xs_close xprt %p\n", xprt);
849
850 xs_reset_transport(transport);
61d0a8e6 851 xprt->reestablish_timeout = 0;
fe315e76 852
62da3b24 853 xprt_disconnect_done(xprt);
a246b010
CL		 854}
855
4a068258
CL		 856static void xs_inject_disconnect(struct rpc_xprt *xprt)
857{
858 dprintk("RPC: injecting transport disconnect on xprt=%p\n",
859 xprt);
860 xprt_disconnect_done(xprt);
861}
862
315f3812
KM		 863static void xs_xprt_free(struct rpc_xprt *xprt)
864{
865 xs_free_peer_addresses(xprt);
866 xprt_free(xprt);
867}
868
9903cd1c
CL		 869/**
870 * xs_destroy - prepare to shutdown a transport
871 * @xprt: doomed transport
872 *
873 */
874static void xs_destroy(struct rpc_xprt *xprt)
a246b010 875{
03c78827
TM		 876 struct sock_xprt *transport = container_of(xprt,
877 struct sock_xprt, xprt);
46121cf7 878 dprintk("RPC: xs_destroy xprt %p\n", xprt);
9903cd1c 879
03c78827 880 cancel_delayed_work_sync(&transport->connect_worker);
a1311d87 881 xs_close(xprt);
edc1b01c 882 cancel_work_sync(&transport->recv_worker);
315f3812 883 xs_xprt_free(xprt);
a1311d87 884 module_put(THIS_MODULE);
a246b010
CL		 885}
886
176e21ee
CL		 887static int xs_local_copy_to_xdr(struct xdr_buf *xdr, struct sk_buff *skb)
888{
889 struct xdr_skb_reader desc = {
890 .skb = skb,
891 .offset = sizeof(rpc_fraghdr),
892 .count = skb->len - sizeof(rpc_fraghdr),
893 };
894
895 if (xdr_partial_copy_from_skb(xdr, 0, &desc, xdr_skb_read_bits) < 0)
896 return -1;
897 if (desc.count)
898 return -1;
899 return 0;
900}
901
902/**
a2648094
TM		 903 * xs_local_data_read_skb
904 * @xprt: transport
905 * @sk: socket
906 * @skb: skbuff
176e21ee
CL		 907 *
908 * Currently this assumes we can read the whole reply in a single gulp.
909 */
a2648094
TM		 910static void xs_local_data_read_skb(struct rpc_xprt *xprt,
911 struct sock *sk,
912 struct sk_buff *skb)
176e21ee
CL		 913{
914 struct rpc_task *task;
176e21ee 915 struct rpc_rqst *rovr;
a2648094 916 int repsize, copied;
176e21ee
CL		 917 u32 _xid;
918 __be32 *xp;
919
176e21ee
CL		 920 repsize = skb->len - sizeof(rpc_fraghdr);
921 if (repsize < 4) {
922 dprintk("RPC: impossible RPC reply size %d\n", repsize);
a2648094 923 return;
176e21ee
CL		 924 }
925
926 /* Copy the XID from the skb... */
927 xp = skb_header_pointer(skb, sizeof(rpc_fraghdr), sizeof(_xid), &_xid);
928 if (xp == NULL)
a2648094 929 return;
176e21ee
CL		 930
931 /* Look up and lock the request corresponding to the given XID */
a2648094 932 spin_lock_bh(&xprt->transport_lock);
176e21ee
CL		 933 rovr = xprt_lookup_rqst(xprt, *xp);
934 if (!rovr)
935 goto out_unlock;
936 task = rovr->rq_task;
937
938 copied = rovr->rq_private_buf.buflen;
939 if (copied > repsize)
940 copied = repsize;
941
942 if (xs_local_copy_to_xdr(&rovr->rq_private_buf, skb)) {
943 dprintk("RPC: sk_buff copy failed\n");
944 goto out_unlock;
945 }
946
947 xprt_complete_rqst(task, copied);
948
949 out_unlock:
a2648094
TM		 950 spin_unlock_bh(&xprt->transport_lock);
951}
952
953static void xs_local_data_receive(struct sock_xprt *transport)
954{
955 struct sk_buff *skb;
956 struct sock *sk;
957 int err;
958
959 mutex_lock(&transport->recv_mutex);
960 sk = transport->inet;
961 if (sk == NULL)
962 goto out;
963 for (;;) {
964 skb = skb_recv_datagram(sk, 0, 1, &err);
965 if (skb == NULL)
966 break;
967 xs_local_data_read_skb(&transport->xprt, sk, skb);
968 skb_free_datagram(sk, skb);
969 }
970out:
971 mutex_unlock(&transport->recv_mutex);
972}
973
974static void xs_local_data_receive_workfn(struct work_struct *work)
975{
976 struct sock_xprt *transport =
977 container_of(work, struct sock_xprt, recv_worker);
978 xs_local_data_receive(transport);
176e21ee
CL		 979}
980
9903cd1c 981/**
f9b2ee71
TM		 982 * xs_udp_data_read_skb - receive callback for UDP sockets
983 * @xprt: transport
984 * @sk: socket
985 * @skb: skbuff
9903cd1c 986 *
a246b010 987 */
f9b2ee71
TM		 988static void xs_udp_data_read_skb(struct rpc_xprt *xprt,
989 struct sock *sk,
990 struct sk_buff *skb)
a246b010 991{
9903cd1c 992 struct rpc_task *task;
a246b010 993 struct rpc_rqst *rovr;
f9b2ee71 994 int repsize, copied;
d8ed029d
AD		 995 u32 _xid;
996 __be32 *xp;
a246b010 997
1da8c681 998 repsize = skb->len;
a246b010 999 if (repsize < 4) {
46121cf7 1000 dprintk("RPC: impossible RPC reply size %d!\n", repsize);
f9b2ee71 1001 return;
a246b010
CL		 1002 }
1003
1004 /* Copy the XID from the skb... */
1da8c681 1005 xp = skb_header_pointer(skb, 0, sizeof(_xid), &_xid);
a246b010 1006 if (xp == NULL)
f9b2ee71 1007 return;
a246b010
CL		 1008
1009 /* Look up and lock the request corresponding to the given XID */
f9b2ee71 1010 spin_lock_bh(&xprt->transport_lock);
a246b010
CL		 1011 rovr = xprt_lookup_rqst(xprt, *xp);
1012 if (!rovr)
1013 goto out_unlock;
1014 task = rovr->rq_task;
1015
a246b010
CL		 1016 if ((copied = rovr->rq_private_buf.buflen) > repsize)
1017 copied = repsize;
1018
1019 /* Suck it into the iovec, verify checksum if not done by hw. */
1781f7f5 1020 if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
02c22347 1021 __UDPX_INC_STATS(sk, UDP_MIB_INERRORS);
a246b010 1022 goto out_unlock;
1781f7f5
HX		 1023 }
1024
02c22347 1025 __UDPX_INC_STATS(sk, UDP_MIB_INDATAGRAMS);
a246b010 1026
6a24dfb6 1027 xprt_adjust_cwnd(xprt, task, copied);
1570c1e4 1028 xprt_complete_rqst(task, copied);
a246b010
CL		 1029
1030 out_unlock:
f9b2ee71
TM		 1031 spin_unlock_bh(&xprt->transport_lock);
1032}
1033
1034static void xs_udp_data_receive(struct sock_xprt *transport)
1035{
1036 struct sk_buff *skb;
1037 struct sock *sk;
1038 int err;
1039
1040 mutex_lock(&transport->recv_mutex);
1041 sk = transport->inet;
1042 if (sk == NULL)
1043 goto out;
1044 for (;;) {
1045 skb = skb_recv_datagram(sk, 0, 1, &err);
1046 if (skb == NULL)
1047 break;
1048 xs_udp_data_read_skb(&transport->xprt, sk, skb);
1049 skb_free_datagram(sk, skb);
1050 }
1051out:
1052 mutex_unlock(&transport->recv_mutex);
1053}
1054
1055static void xs_udp_data_receive_workfn(struct work_struct *work)
1056{
1057 struct sock_xprt *transport =
1058 container_of(work, struct sock_xprt, recv_worker);
1059 xs_udp_data_receive(transport);
1060}
1061
1062/**
1063 * xs_data_ready - "data ready" callback for UDP sockets
1064 * @sk: socket with data to read
1065 *
1066 */
1067static void xs_data_ready(struct sock *sk)
1068{
1069 struct rpc_xprt *xprt;
1070
1071 read_lock_bh(&sk->sk_callback_lock);
1072 dprintk("RPC: xs_data_ready...\n");
1073 xprt = xprt_from_sock(sk);
1074 if (xprt != NULL) {
1075 struct sock_xprt *transport = container_of(xprt,
1076 struct sock_xprt, xprt);
1077 queue_work(rpciod_workqueue, &transport->recv_worker);
1078 }
f064af1e 1079 read_unlock_bh(&sk->sk_callback_lock);
a246b010
CL		 1080}
1081
a519fc7a
TM		 1082/*
1083 * Helper function to force a TCP close if the server is sending
1084 * junk and/or it has put us in CLOSE_WAIT
1085 */
1086static void xs_tcp_force_close(struct rpc_xprt *xprt)
1087{
a519fc7a
TM		 1088 xprt_force_disconnect(xprt);
1089}
1090
dd456471 1091static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
a246b010 1092{
51971139 1093 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
a246b010
CL		 1094 size_t len, used;
1095 char *p;
1096
51971139
CL		 1097 p = ((char *) &transport->tcp_fraghdr) + transport->tcp_offset;
1098 len = sizeof(transport->tcp_fraghdr) - transport->tcp_offset;
9d292316 1099 used = xdr_skb_read_bits(desc, p, len);
51971139 1100 transport->tcp_offset += used;
a246b010
CL		 1101 if (used != len)
1102 return;
808012fb 1103
51971139
CL		 1104 transport->tcp_reclen = ntohl(transport->tcp_fraghdr);
1105 if (transport->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
e136d092 1106 transport->tcp_flags |= TCP_RCV_LAST_FRAG;
a246b010 1107 else
e136d092 1108 transport->tcp_flags &= ~TCP_RCV_LAST_FRAG;
51971139 1109 transport->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
808012fb 1110
e136d092 1111 transport->tcp_flags &= ~TCP_RCV_COPY_FRAGHDR;
51971139 1112 transport->tcp_offset = 0;
808012fb 1113
a246b010 1114 /* Sanity check of the record length */
18dca02a 1115 if (unlikely(transport->tcp_reclen < 8)) {
46121cf7 1116 dprintk("RPC: invalid TCP record fragment length\n");
a519fc7a 1117 xs_tcp_force_close(xprt);
9903cd1c 1118 return;
a246b010 1119 }
46121cf7 1120 dprintk("RPC: reading TCP record fragment of length %d\n",
51971139 1121 transport->tcp_reclen);
a246b010
CL		 1122}
1123
51971139 1124static void xs_tcp_check_fraghdr(struct sock_xprt *transport)
a246b010 1125{
51971139 1126 if (transport->tcp_offset == transport->tcp_reclen) {
e136d092 1127 transport->tcp_flags |= TCP_RCV_COPY_FRAGHDR;
51971139 1128 transport->tcp_offset = 0;
e136d092
CL		 1129 if (transport->tcp_flags & TCP_RCV_LAST_FRAG) {
1130 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1131 transport->tcp_flags |= TCP_RCV_COPY_XID;
51971139 1132 transport->tcp_copied = 0;
a246b010
CL		 1133 }
1134 }
1135}
1136
dd456471 1137static inline void xs_tcp_read_xid(struct sock_xprt *transport, struct xdr_skb_reader *desc)
a246b010
CL		 1138{
1139 size_t len, used;
1140 char *p;
1141
51971139 1142 len = sizeof(transport->tcp_xid) - transport->tcp_offset;
46121cf7 1143 dprintk("RPC: reading XID (%Zu bytes)\n", len);
51971139 1144 p = ((char *) &transport->tcp_xid) + transport->tcp_offset;
9d292316 1145 used = xdr_skb_read_bits(desc, p, len);
51971139 1146 transport->tcp_offset += used;
a246b010
CL		 1147 if (used != len)
1148 return;
e136d092 1149 transport->tcp_flags &= ~TCP_RCV_COPY_XID;
f4a2e418 1150 transport->tcp_flags |= TCP_RCV_READ_CALLDIR;
51971139 1151 transport->tcp_copied = 4;
18dca02a
RL		 1152 dprintk("RPC: reading %s XID %08x\n",
1153 (transport->tcp_flags & TCP_RPC_REPLY) ? "reply for"
1154 : "request with",
51971139
CL		 1155 ntohl(transport->tcp_xid));
1156 xs_tcp_check_fraghdr(transport);
a246b010
CL		 1157}
1158
18dca02a
RL		 1159static inline void xs_tcp_read_calldir(struct sock_xprt *transport,
1160 struct xdr_skb_reader *desc)
a246b010 1161{
18dca02a
RL		 1162 size_t len, used;
1163 u32 offset;
b76ce561 1164 char *p;
18dca02a
RL		 1165
1166 /*
1167 * We want transport->tcp_offset to be 8 at the end of this routine
1168 * (4 bytes for the xid and 4 bytes for the call/reply flag).
1169 * When this function is called for the first time,
1170 * transport->tcp_offset is 4 (after having already read the xid).
1171 */
1172 offset = transport->tcp_offset - sizeof(transport->tcp_xid);
b76ce561 1173 len = sizeof(transport->tcp_calldir) - offset;
18dca02a 1174 dprintk("RPC: reading CALL/REPLY flag (%Zu bytes)\n", len);
b76ce561
TM		 1175 p = ((char *) &transport->tcp_calldir) + offset;
1176 used = xdr_skb_read_bits(desc, p, len);
18dca02a
RL		 1177 transport->tcp_offset += used;
1178 if (used != len)
1179 return;
f4a2e418 1180 transport->tcp_flags &= ~TCP_RCV_READ_CALLDIR;
f4a2e418
RL		 1181 /*
1182 * We don't yet have the XDR buffer, so we will write the calldir
1183 * out after we get the buffer from the 'struct rpc_rqst'
1184 */
b76ce561
TM		 1185 switch (ntohl(transport->tcp_calldir)) {
1186 case RPC_REPLY:
1187 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1188 transport->tcp_flags |= TCP_RCV_COPY_DATA;
18dca02a 1189 transport->tcp_flags |= TCP_RPC_REPLY;
b76ce561
TM		 1190 break;
1191 case RPC_CALL:
1192 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1193 transport->tcp_flags |= TCP_RCV_COPY_DATA;
18dca02a 1194 transport->tcp_flags &= ~TCP_RPC_REPLY;
b76ce561
TM		 1195 break;
1196 default:
1197 dprintk("RPC: invalid request message type\n");
a519fc7a 1198 xs_tcp_force_close(&transport->xprt);
b76ce561 1199 }
18dca02a
RL		 1200 xs_tcp_check_fraghdr(transport);
1201}
1202
44b98efd
RL		 1203static inline void xs_tcp_read_common(struct rpc_xprt *xprt,
1204 struct xdr_skb_reader *desc,
1205 struct rpc_rqst *req)
a246b010 1206{
44b98efd
RL		 1207 struct sock_xprt *transport =
1208 container_of(xprt, struct sock_xprt, xprt);
a246b010
CL		 1209 struct xdr_buf *rcvbuf;
1210 size_t len;
1211 ssize_t r;
1212
a246b010 1213 rcvbuf = &req->rq_private_buf;
f4a2e418
RL		 1214
1215 if (transport->tcp_flags & TCP_RCV_COPY_CALLDIR) {
1216 /*
1217 * Save the RPC direction in the XDR buffer
1218 */
f4a2e418 1219 memcpy(rcvbuf->head[0].iov_base + transport->tcp_copied,
b76ce561
TM		 1220 &transport->tcp_calldir,
1221 sizeof(transport->tcp_calldir));
1222 transport->tcp_copied += sizeof(transport->tcp_calldir);
f4a2e418 1223 transport->tcp_flags &= ~TCP_RCV_COPY_CALLDIR;
a246b010
CL		 1224 }
1225
a246b010 1226 len = desc->count;
51971139 1227 if (len > transport->tcp_reclen - transport->tcp_offset) {
dd456471 1228 struct xdr_skb_reader my_desc;
a246b010 1229
51971139 1230 len = transport->tcp_reclen - transport->tcp_offset;
a246b010
CL		 1231 memcpy(&my_desc, desc, sizeof(my_desc));
1232 my_desc.count = len;
51971139 1233 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
9d292316 1234 &my_desc, xdr_skb_read_bits);
a246b010
CL		 1235 desc->count -= r;
1236 desc->offset += r;
1237 } else
51971139 1238 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
9d292316 1239 desc, xdr_skb_read_bits);
a246b010
CL		 1240
1241 if (r > 0) {
51971139
CL		 1242 transport->tcp_copied += r;
1243 transport->tcp_offset += r;
a246b010
CL		 1244 }
1245 if (r != len) {
1246 /* Error when copying to the receive buffer,
1247 * usually because we weren't able to allocate
1248 * additional buffer pages. All we can do now
e136d092 1249 * is turn off TCP_RCV_COPY_DATA, so the request
a246b010
CL		 1250 * will not receive any additional updates,
1251 * and time out.
1252 * Any remaining data from this record will
1253 * be discarded.
1254 */
e136d092 1255 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
46121cf7 1256 dprintk("RPC: XID %08x truncated request\n",
51971139 1257 ntohl(transport->tcp_xid));
46121cf7
CL		 1258 dprintk("RPC: xprt = %p, tcp_copied = %lu, "
1259 "tcp_offset = %u, tcp_reclen = %u\n",
1260 xprt, transport->tcp_copied,
1261 transport->tcp_offset, transport->tcp_reclen);
44b98efd 1262 return;
a246b010
CL		 1263 }
1264
46121cf7 1265 dprintk("RPC: XID %08x read %Zd bytes\n",
51971139 1266 ntohl(transport->tcp_xid), r);
46121cf7
CL		 1267 dprintk("RPC: xprt = %p, tcp_copied = %lu, tcp_offset = %u, "
1268 "tcp_reclen = %u\n", xprt, transport->tcp_copied,
1269 transport->tcp_offset, transport->tcp_reclen);
51971139
CL		 1270
1271 if (transport->tcp_copied == req->rq_private_buf.buflen)
e136d092 1272 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
51971139 1273 else if (transport->tcp_offset == transport->tcp_reclen) {
e136d092
CL		 1274 if (transport->tcp_flags & TCP_RCV_LAST_FRAG)
1275 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
a246b010 1276 }
44b98efd
RL		 1277}
1278
1279/*
1280 * Finds the request corresponding to the RPC xid and invokes the common
1281 * tcp read code to read the data.
1282 */
1283static inline int xs_tcp_read_reply(struct rpc_xprt *xprt,
1284 struct xdr_skb_reader *desc)
1285{
1286 struct sock_xprt *transport =
1287 container_of(xprt, struct sock_xprt, xprt);
1288 struct rpc_rqst *req;
1289
1290 dprintk("RPC: read reply XID %08x\n", ntohl(transport->tcp_xid));
1291
1292 /* Find and lock the request corresponding to this xid */
edc1b01c 1293 spin_lock_bh(&xprt->transport_lock);
44b98efd
RL		 1294 req = xprt_lookup_rqst(xprt, transport->tcp_xid);
1295 if (!req) {
1296 dprintk("RPC: XID %08x request not found!\n",
1297 ntohl(transport->tcp_xid));
edc1b01c 1298 spin_unlock_bh(&xprt->transport_lock);
44b98efd
RL		 1299 return -1;
1300 }
1301
1302 xs_tcp_read_common(xprt, desc, req);
1303
e136d092 1304 if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
51971139 1305 xprt_complete_rqst(req->rq_task, transport->tcp_copied);
44b98efd 1306
edc1b01c 1307 spin_unlock_bh(&xprt->transport_lock);
44b98efd
RL		 1308 return 0;
1309}
1310
9e00abc3 1311#if defined(CONFIG_SUNRPC_BACKCHANNEL)
44b98efd
RL		 1312/*
1313 * Obtains an rpc_rqst previously allocated and invokes the common
1314 * tcp read code to read the data. The result is placed in the callback
1315 * queue.
1316 * If we're unable to obtain the rpc_rqst we schedule the closing of the
1317 * connection and return -1.
1318 */
2ea24497 1319static int xs_tcp_read_callback(struct rpc_xprt *xprt,
44b98efd
RL		 1320 struct xdr_skb_reader *desc)
1321{
1322 struct sock_xprt *transport =
1323 container_of(xprt, struct sock_xprt, xprt);
1324 struct rpc_rqst *req;
1325
2ea24497 1326 /* Look up and lock the request corresponding to the given XID */
edc1b01c 1327 spin_lock_bh(&xprt->transport_lock);
2ea24497 1328 req = xprt_lookup_bc_request(xprt, transport->tcp_xid);
44b98efd 1329 if (req == NULL) {
edc1b01c 1330 spin_unlock_bh(&xprt->transport_lock);
44b98efd
RL		 1331 printk(KERN_WARNING "Callback slot table overflowed\n");
1332 xprt_force_disconnect(xprt);
1333 return -1;
1334 }
1335
44b98efd
RL		 1336 dprintk("RPC: read callback XID %08x\n", ntohl(req->rq_xid));
1337 xs_tcp_read_common(xprt, desc, req);
1338
2ea24497
TM		 1339 if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1340 xprt_complete_bc_request(req, transport->tcp_copied);
edc1b01c 1341 spin_unlock_bh(&xprt->transport_lock);
44b98efd
RL		 1342
1343 return 0;
1344}
1345
1346static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1347 struct xdr_skb_reader *desc)
1348{
1349 struct sock_xprt *transport =
1350 container_of(xprt, struct sock_xprt, xprt);
1351
1352 return (transport->tcp_flags & TCP_RPC_REPLY) ?
1353 xs_tcp_read_reply(xprt, desc) :
1354 xs_tcp_read_callback(xprt, desc);
1355}
76566773
CL		 1356
1357static int xs_tcp_bc_up(struct svc_serv *serv, struct net *net)
1358{
1359 int ret;
1360
1361 ret = svc_create_xprt(serv, "tcp-bc", net, PF_INET, 0,
1362 SVC_SOCK_ANONYMOUS);
1363 if (ret < 0)
1364 return ret;
1365 return 0;
1366}
6b26cc8c
CL		 1367
1368static size_t xs_tcp_bc_maxpayload(struct rpc_xprt *xprt)
1369{
1370 return PAGE_SIZE;
1371}
44b98efd
RL		 1372#else
1373static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1374 struct xdr_skb_reader *desc)
1375{
1376 return xs_tcp_read_reply(xprt, desc);
1377}
9e00abc3 1378#endif /* CONFIG_SUNRPC_BACKCHANNEL */
44b98efd
RL		 1379
1380/*
1381 * Read data off the transport. This can be either an RPC_CALL or an
1382 * RPC_REPLY. Relay the processing to helper functions.
1383 */
1384static void xs_tcp_read_data(struct rpc_xprt *xprt,
1385 struct xdr_skb_reader *desc)
1386{
1387 struct sock_xprt *transport =
1388 container_of(xprt, struct sock_xprt, xprt);
1389
1390 if (_xs_tcp_read_data(xprt, desc) == 0)
1391 xs_tcp_check_fraghdr(transport);
1392 else {
1393 /*
1394 * The transport_lock protects the request handling.
1395 * There's no need to hold it to update the tcp_flags.
1396 */
1397 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1398 }
a246b010
CL		 1399}
1400
dd456471 1401static inline void xs_tcp_read_discard(struct sock_xprt *transport, struct xdr_skb_reader *desc)
a246b010
CL		 1402{
1403 size_t len;
1404
51971139 1405 len = transport->tcp_reclen - transport->tcp_offset;
a246b010
CL		 1406 if (len > desc->count)
1407 len = desc->count;
1408 desc->count -= len;
1409 desc->offset += len;
51971139 1410 transport->tcp_offset += len;
46121cf7 1411 dprintk("RPC: discarded %Zu bytes\n", len);
51971139 1412 xs_tcp_check_fraghdr(transport);
a246b010
CL		 1413}
1414
9903cd1c 1415static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
a246b010
CL		 1416{
1417 struct rpc_xprt *xprt = rd_desc->arg.data;
51971139 1418 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
dd456471 1419 struct xdr_skb_reader desc = {
a246b010
CL		 1420 .skb = skb,
1421 .offset = offset,
1422 .count = len,
9903cd1c 1423 };
a246b010 1424
46121cf7 1425 dprintk("RPC: xs_tcp_data_recv started\n");
a246b010 1426 do {
1a867a08 1427 trace_xs_tcp_data_recv(transport);
a246b010
CL		 1428 /* Read in a new fragment marker if necessary */
1429 /* Can we ever really expect to get completely empty fragments? */
e136d092 1430 if (transport->tcp_flags & TCP_RCV_COPY_FRAGHDR) {
9903cd1c 1431 xs_tcp_read_fraghdr(xprt, &desc);
a246b010
CL		 1432 continue;
1433 }
1434 /* Read in the xid if necessary */
e136d092 1435 if (transport->tcp_flags & TCP_RCV_COPY_XID) {
51971139 1436 xs_tcp_read_xid(transport, &desc);
a246b010
CL		 1437 continue;
1438 }
18dca02a 1439 /* Read in the call/reply flag */
f4a2e418 1440 if (transport->tcp_flags & TCP_RCV_READ_CALLDIR) {
18dca02a
RL		 1441 xs_tcp_read_calldir(transport, &desc);
1442 continue;
1443 }
a246b010 1444 /* Read in the request data */
e136d092 1445 if (transport->tcp_flags & TCP_RCV_COPY_DATA) {
44b98efd 1446 xs_tcp_read_data(xprt, &desc);
a246b010
CL		 1447 continue;
1448 }
1449 /* Skip over any trailing bytes on short reads */
51971139 1450 xs_tcp_read_discard(transport, &desc);
a246b010 1451 } while (desc.count);
1a867a08 1452 trace_xs_tcp_data_recv(transport);
46121cf7 1453 dprintk("RPC: xs_tcp_data_recv done\n");
a246b010
CL		 1454 return len - desc.count;
1455}
1456
66d7a56a
TM		 1457static void xs_tcp_data_receive(struct sock_xprt *transport)
1458{
1459 struct rpc_xprt *xprt = &transport->xprt;
1460 struct sock *sk;
1461 read_descriptor_t rd_desc = {
1462 .count = 2*1024*1024,
1463 .arg.data = xprt,
1464 };
1465 unsigned long total = 0;
1466 int read = 0;
1467
edc1b01c 1468 mutex_lock(&transport->recv_mutex);
66d7a56a 1469 sk = transport->inet;
edc1b01c
TM		 1470 if (sk == NULL)
1471 goto out;
66d7a56a
TM		 1472
1473 /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
1474 for (;;) {
edc1b01c 1475 lock_sock(sk);
66d7a56a 1476 read = tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
edc1b01c 1477 release_sock(sk);
66d7a56a
TM		 1478 if (read <= 0)
1479 break;
1480 total += read;
1481 rd_desc.count = 65536;
1482 }
edc1b01c
TM		 1483out:
1484 mutex_unlock(&transport->recv_mutex);
66d7a56a
TM		 1485 trace_xs_tcp_data_ready(xprt, read, total);
1486}
1487
edc1b01c
TM		 1488static void xs_tcp_data_receive_workfn(struct work_struct *work)
1489{
1490 struct sock_xprt *transport =
1491 container_of(work, struct sock_xprt, recv_worker);
1492 xs_tcp_data_receive(transport);
1493}
1494
9903cd1c
CL		 1495/**
1496 * xs_tcp_data_ready - "data ready" callback for TCP sockets
1497 * @sk: socket with data to read
9903cd1c
CL		 1498 *
1499 */
676d2369 1500static void xs_tcp_data_ready(struct sock *sk)
a246b010 1501{
66d7a56a 1502 struct sock_xprt *transport;
a246b010 1503 struct rpc_xprt *xprt;
a246b010 1504
46121cf7
CL		 1505 dprintk("RPC: xs_tcp_data_ready...\n");
1506
f064af1e 1507 read_lock_bh(&sk->sk_callback_lock);
66d7a56a 1508 if (!(xprt = xprt_from_sock(sk)))
a246b010 1509 goto out;
66d7a56a
TM		 1510 transport = container_of(xprt, struct sock_xprt, xprt);
1511
61d0a8e6
NB		 1512 /* Any data means we had a useful conversation, so
1513 * the we don't need to delay the next reconnect
1514 */
1515 if (xprt->reestablish_timeout)
1516 xprt->reestablish_timeout = 0;
edc1b01c 1517 queue_work(rpciod_workqueue, &transport->recv_worker);
61d0a8e6 1518
a246b010 1519out:
f064af1e 1520 read_unlock_bh(&sk->sk_callback_lock);
a246b010
CL		 1521}
1522
9903cd1c
CL		 1523/**
1524 * xs_tcp_state_change - callback to handle TCP socket state changes
1525 * @sk: socket whose state has changed
1526 *
1527 */
1528static void xs_tcp_state_change(struct sock *sk)
a246b010 1529{
9903cd1c 1530 struct rpc_xprt *xprt;
0fdea1e8 1531 struct sock_xprt *transport;
a246b010 1532
f064af1e 1533 read_lock_bh(&sk->sk_callback_lock);
a246b010
CL		 1534 if (!(xprt = xprt_from_sock(sk)))
1535 goto out;
46121cf7 1536 dprintk("RPC: xs_tcp_state_change client %p...\n", xprt);
669502ff 1537 dprintk("RPC: state %x conn %d dead %d zapped %d sk_shutdown %d\n",
46121cf7
CL		 1538 sk->sk_state, xprt_connected(xprt),
1539 sock_flag(sk, SOCK_DEAD),
669502ff
AC		 1540 sock_flag(sk, SOCK_ZAPPED),
1541 sk->sk_shutdown);
a246b010 1542
0fdea1e8 1543 transport = container_of(xprt, struct sock_xprt, xprt);
40b5ea0c 1544 trace_rpc_socket_state_change(xprt, sk->sk_socket);
a246b010
CL		 1545 switch (sk->sk_state) {
1546 case TCP_ESTABLISHED:
f064af1e 1547 spin_lock(&xprt->transport_lock);
a246b010 1548 if (!xprt_test_and_set_connected(xprt)) {
51971139 1549
a246b010 1550 /* Reset TCP record info */
51971139
CL		 1551 transport->tcp_offset = 0;
1552 transport->tcp_reclen = 0;
1553 transport->tcp_copied = 0;
e136d092
CL		 1554 transport->tcp_flags =
1555 TCP_RCV_COPY_FRAGHDR | TCP_RCV_COPY_XID;
8b71798c 1556 xprt->connect_cookie++;
0fdea1e8
TM		 1557 clear_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
1558 xprt_clear_connecting(xprt);
51971139 1559
2a491991 1560 xprt_wake_pending_tasks(xprt, -EAGAIN);
a246b010 1561 }
f064af1e 1562 spin_unlock(&xprt->transport_lock);
a246b010 1563 break;
3b948ae5
TM		 1564 case TCP_FIN_WAIT1:
1565 /* The client initiated a shutdown of the socket */
7c1d71cf 1566 xprt->connect_cookie++;
663b8858 1567 xprt->reestablish_timeout = 0;
3b948ae5 1568 set_bit(XPRT_CLOSING, &xprt->state);
4e857c58 1569 smp_mb__before_atomic();
3b948ae5 1570 clear_bit(XPRT_CONNECTED, &xprt->state);
ef803670 1571 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
4e857c58 1572 smp_mb__after_atomic();
a246b010 1573 break;
632e3bdc 1574 case TCP_CLOSE_WAIT:
3b948ae5 1575 /* The server initiated a shutdown of the socket */
7c1d71cf 1576 xprt->connect_cookie++;
d0bea455 1577 clear_bit(XPRT_CONNECTED, &xprt->state);
a519fc7a 1578 xs_tcp_force_close(xprt);
663b8858
TM		 1579 case TCP_CLOSING:
1580 /*
1581 * If the server closed down the connection, make sure that
1582 * we back off before reconnecting
1583 */
1584 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1585 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
3b948ae5
TM		 1586 break;
1587 case TCP_LAST_ACK:
670f9457 1588 set_bit(XPRT_CLOSING, &xprt->state);
4e857c58 1589 smp_mb__before_atomic();
3b948ae5 1590 clear_bit(XPRT_CONNECTED, &xprt->state);
4e857c58 1591 smp_mb__after_atomic();
3b948ae5
TM		 1592 break;
1593 case TCP_CLOSE:
0fdea1e8
TM		 1594 if (test_and_clear_bit(XPRT_SOCK_CONNECTING,
1595 &transport->sock_state))
1596 xprt_clear_connecting(xprt);
7d1e8255 1597 xs_sock_mark_closed(xprt);
a246b010
CL		 1598 }
1599 out:
f064af1e 1600 read_unlock_bh(&sk->sk_callback_lock);
a246b010
CL		 1601}
1602
1f0fa154
IJ		 1603static void xs_write_space(struct sock *sk)
1604{
13331a55 1605 struct socket_wq *wq;
1f0fa154
IJ		 1606 struct rpc_xprt *xprt;
1607
13331a55 1608 if (!sk->sk_socket)
1f0fa154 1609 return;
13331a55 1610 clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1f0fa154
IJ		 1611
1612 if (unlikely(!(xprt = xprt_from_sock(sk))))
1613 return;
13331a55
TM		 1614 rcu_read_lock();
1615 wq = rcu_dereference(sk->sk_wq);
1616 if (!wq || test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &wq->flags) == 0)
1617 goto out;
1f0fa154
IJ		 1618
1619 xprt_write_space(xprt);
13331a55
TM		 1620out:
1621 rcu_read_unlock();
1f0fa154
IJ		 1622}
1623
9903cd1c 1624/**
c7b2cae8
CL		 1625 * xs_udp_write_space - callback invoked when socket buffer space
1626 * becomes available
9903cd1c
CL		 1627 * @sk: socket whose state has changed
1628 *
a246b010
CL		 1629 * Called when more output buffer space is available for this socket.
1630 * We try not to wake our writers until they can make "significant"
c7b2cae8 1631 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
a246b010
CL		 1632 * with a bunch of small requests.
1633 */
c7b2cae8 1634static void xs_udp_write_space(struct sock *sk)
a246b010 1635{
f064af1e 1636 read_lock_bh(&sk->sk_callback_lock);
a246b010 1637
c7b2cae8 1638 /* from net/core/sock.c:sock_def_write_space */
1f0fa154
IJ		 1639 if (sock_writeable(sk))
1640 xs_write_space(sk);
a246b010 1641
f064af1e 1642 read_unlock_bh(&sk->sk_callback_lock);
c7b2cae8 1643}
a246b010 1644
c7b2cae8
CL		 1645/**
1646 * xs_tcp_write_space - callback invoked when socket buffer space
1647 * becomes available
1648 * @sk: socket whose state has changed
1649 *
1650 * Called when more output buffer space is available for this socket.
1651 * We try not to wake our writers until they can make "significant"
1652 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1653 * with a bunch of small requests.
1654 */
1655static void xs_tcp_write_space(struct sock *sk)
1656{
f064af1e 1657 read_lock_bh(&sk->sk_callback_lock);
c7b2cae8
CL		 1658
1659 /* from net/core/stream.c:sk_stream_write_space */
64dc6130 1660 if (sk_stream_is_writeable(sk))
1f0fa154 1661 xs_write_space(sk);
b6ddf64f 1662
f064af1e 1663 read_unlock_bh(&sk->sk_callback_lock);
a246b010
CL		 1664}
1665
470056c2 1666static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
a246b010 1667{
ee0ac0c2
CL		 1668 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1669 struct sock *sk = transport->inet;
a246b010 1670
7c6e066e 1671 if (transport->rcvsize) {
a246b010 1672 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
7c6e066e 1673 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
a246b010 1674 }
7c6e066e 1675 if (transport->sndsize) {
a246b010 1676 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
7c6e066e 1677 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
a246b010
CL		 1678 sk->sk_write_space(sk);
1679 }
1680}
1681
43118c29 1682/**
470056c2 1683 * xs_udp_set_buffer_size - set send and receive limits
43118c29 1684 * @xprt: generic transport
470056c2
CL		 1685 * @sndsize: requested size of send buffer, in bytes
1686 * @rcvsize: requested size of receive buffer, in bytes
43118c29 1687 *
470056c2 1688 * Set socket send and receive buffer size limits.
43118c29 1689 */
470056c2 1690static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
43118c29 1691{
7c6e066e
CL		 1692 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1693
1694 transport->sndsize = 0;
470056c2 1695 if (sndsize)
7c6e066e
CL		 1696 transport->sndsize = sndsize + 1024;
1697 transport->rcvsize = 0;
470056c2 1698 if (rcvsize)
7c6e066e 1699 transport->rcvsize = rcvsize + 1024;
470056c2
CL		 1700
1701 xs_udp_do_set_buffer_size(xprt);
43118c29
CL		 1702}
1703
46c0ee8b
CL		 1704/**
1705 * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1706 * @task: task that timed out
1707 *
1708 * Adjust the congestion window after a retransmit timeout has occurred.
1709 */
6a24dfb6 1710static void xs_udp_timer(struct rpc_xprt *xprt, struct rpc_task *task)
46c0ee8b 1711{
6a24dfb6 1712 xprt_adjust_cwnd(xprt, task, -ETIMEDOUT);
46c0ee8b
CL		 1713}
1714
b85d8806
CL		 1715static unsigned short xs_get_random_port(void)
1716{
1717 unsigned short range = xprt_max_resvport - xprt_min_resvport;
63862b5b 1718 unsigned short rand = (unsigned short) prandom_u32() % range;
b85d8806
CL		 1719 return rand + xprt_min_resvport;
1720}
1721
4dda9c8a
TM		 1722/**
1723 * xs_set_reuseaddr_port - set the socket's port and address reuse options
1724 * @sock: socket
1725 *
1726 * Note that this function has to be called on all sockets that share the
1727 * same port, and it must be called before binding.
1728 */
1729static void xs_sock_set_reuseport(struct socket *sock)
1730{
402e23b4 1731 int opt = 1;
4dda9c8a 1732
402e23b4
TM		 1733 kernel_setsockopt(sock, SOL_SOCKET, SO_REUSEPORT,
1734 (char *)&opt, sizeof(opt));
4dda9c8a
TM		 1735}
1736
1737static unsigned short xs_sock_getport(struct socket *sock)
1738{
1739 struct sockaddr_storage buf;
1740 int buflen;
1741 unsigned short port = 0;
1742
1743 if (kernel_getsockname(sock, (struct sockaddr *)&buf, &buflen) < 0)
1744 goto out;
1745 switch (buf.ss_family) {
1746 case AF_INET6:
1747 port = ntohs(((struct sockaddr_in6 *)&buf)->sin6_port);
1748 break;
1749 case AF_INET:
1750 port = ntohs(((struct sockaddr_in *)&buf)->sin_port);
1751 }
1752out:
1753 return port;
1754}
1755
92200412
CL		 1756/**
1757 * xs_set_port - reset the port number in the remote endpoint address
1758 * @xprt: generic transport
1759 * @port: new port number
1760 *
1761 */
1762static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1763{
46121cf7 1764 dprintk("RPC: setting port for xprt %p to %u\n", xprt, port);
c4efcb1d 1765
9dc3b095
CL		 1766 rpc_set_port(xs_addr(xprt), port);
1767 xs_update_peer_port(xprt);
92200412
CL		 1768}
1769
4dda9c8a
TM		 1770static void xs_set_srcport(struct sock_xprt *transport, struct socket *sock)
1771{
1772 if (transport->srcport == 0)
1773 transport->srcport = xs_sock_getport(sock);
1774}
1775
5d4ec932 1776static unsigned short xs_get_srcport(struct sock_xprt *transport)
67a391d7 1777{
fbfffbd5 1778 unsigned short port = transport->srcport;
67a391d7
TM		 1779
1780 if (port == 0 && transport->xprt.resvport)
1781 port = xs_get_random_port();
1782 return port;
1783}
1784
baaf4e48 1785static unsigned short xs_next_srcport(struct sock_xprt *transport, unsigned short port)
67a391d7 1786{
fbfffbd5
CL		 1787 if (transport->srcport != 0)
1788 transport->srcport = 0;
67a391d7
TM		 1789 if (!transport->xprt.resvport)
1790 return 0;
1791 if (port <= xprt_min_resvport || port > xprt_max_resvport)
1792 return xprt_max_resvport;
1793 return --port;
1794}
beb59b68 1795static int xs_bind(struct sock_xprt *transport, struct socket *sock)
a246b010 1796{
beb59b68 1797 struct sockaddr_storage myaddr;
67a391d7 1798 int err, nloop = 0;
5d4ec932 1799 unsigned short port = xs_get_srcport(transport);
67a391d7 1800 unsigned short last;
a246b010 1801
0f7a622c
CP		 1802 /*
1803 * If we are asking for any ephemeral port (i.e. port == 0 &&
1804 * transport->xprt.resvport == 0), don't bind. Let the local
1805 * port selection happen implicitly when the socket is used
1806 * (for example at connect time).
1807 *
1808 * This ensures that we can continue to establish TCP
1809 * connections even when all local ephemeral ports are already
1810 * a part of some TCP connection. This makes no difference
1811 * for UDP sockets, but also doens't harm them.
1812 *
1813 * If we're asking for any reserved port (i.e. port == 0 &&
1814 * transport->xprt.resvport == 1) xs_get_srcport above will
1815 * ensure that port is non-zero and we will bind as needed.
1816 */
1817 if (port == 0)
1818 return 0;
1819
beb59b68 1820 memcpy(&myaddr, &transport->srcaddr, transport->xprt.addrlen);
a246b010 1821 do {
beb59b68
PE		 1822 rpc_set_port((struct sockaddr *)&myaddr, port);
1823 err = kernel_bind(sock, (struct sockaddr *)&myaddr,
1824 transport->xprt.addrlen);
a246b010 1825 if (err == 0) {
fbfffbd5 1826 transport->srcport = port;
d3bc9a1d 1827 break;
a246b010 1828 }
67a391d7 1829 last = port;
baaf4e48 1830 port = xs_next_srcport(transport, port);
67a391d7
TM		 1831 if (port > last)
1832 nloop++;
1833 } while (err == -EADDRINUSE && nloop != 2);
90058d37 1834
4232e863 1835 if (myaddr.ss_family == AF_INET)
beb59b68
PE		 1836 dprintk("RPC: %s %pI4:%u: %s (%d)\n", __func__,
1837 &((struct sockaddr_in *)&myaddr)->sin_addr,
1838 port, err ? "failed" : "ok", err);
1839 else
1840 dprintk("RPC: %s %pI6:%u: %s (%d)\n", __func__,
1841 &((struct sockaddr_in6 *)&myaddr)->sin6_addr,
1842 port, err ? "failed" : "ok", err);
a246b010
CL		 1843 return err;
1844}
1845
176e21ee
CL		 1846/*
1847 * We don't support autobind on AF_LOCAL sockets
1848 */
1849static void xs_local_rpcbind(struct rpc_task *task)
1850{
fb43d172 1851 xprt_set_bound(task->tk_xprt);
176e21ee
CL		 1852}
1853
1854static void xs_local_set_port(struct rpc_xprt *xprt, unsigned short port)
1855{
1856}
beb59b68 1857
ed07536e
PZ		 1858#ifdef CONFIG_DEBUG_LOCK_ALLOC
1859static struct lock_class_key xs_key[2];
1860static struct lock_class_key xs_slock_key[2];
1861
176e21ee
CL		 1862static inline void xs_reclassify_socketu(struct socket *sock)
1863{
1864 struct sock *sk = sock->sk;
1865
176e21ee
CL		 1866 sock_lock_init_class_and_name(sk, "slock-AF_LOCAL-RPC",
1867 &xs_slock_key[1], "sk_lock-AF_LOCAL-RPC", &xs_key[1]);
1868}
1869
8945ee5e 1870static inline void xs_reclassify_socket4(struct socket *sock)
ed07536e
PZ		 1871{
1872 struct sock *sk = sock->sk;
8945ee5e 1873
8945ee5e
CL		 1874 sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1875 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1876}
ed07536e 1877
8945ee5e
CL		 1878static inline void xs_reclassify_socket6(struct socket *sock)
1879{
1880 struct sock *sk = sock->sk;
ed07536e 1881
8945ee5e
CL		 1882 sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1883 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
ed07536e 1884}
6bc9638a
PE		 1885
1886static inline void xs_reclassify_socket(int family, struct socket *sock)
1887{
fafc4e1e 1888 if (WARN_ON_ONCE(!sock_allow_reclassification(sock->sk)))
1b7a1819
WAA		 1889 return;
1890
4232e863 1891 switch (family) {
176e21ee
CL		 1892 case AF_LOCAL:
1893 xs_reclassify_socketu(sock);
1894 break;
4232e863 1895 case AF_INET:
6bc9638a 1896 xs_reclassify_socket4(sock);
4232e863
CL		 1897 break;
1898 case AF_INET6:
6bc9638a 1899 xs_reclassify_socket6(sock);
4232e863
CL		 1900 break;
1901 }
6bc9638a 1902}
ed07536e 1903#else
6bc9638a
PE		 1904static inline void xs_reclassify_socket(int family, struct socket *sock)
1905{
1906}
ed07536e
PZ		 1907#endif
1908
93dc41bd
N		 1909static void xs_dummy_setup_socket(struct work_struct *work)
1910{
1911}
1912
6bc9638a 1913static struct socket *xs_create_sock(struct rpc_xprt *xprt,
4dda9c8a
TM		 1914 struct sock_xprt *transport, int family, int type,
1915 int protocol, bool reuseport)
22f79326
PE		 1916{
1917 struct socket *sock;
1918 int err;
1919
6bc9638a 1920 err = __sock_create(xprt->xprt_net, family, type, protocol, &sock, 1);
22f79326
PE		 1921 if (err < 0) {
1922 dprintk("RPC: can't create %d transport socket (%d).\n",
1923 protocol, -err);
1924 goto out;
1925 }
6bc9638a 1926 xs_reclassify_socket(family, sock);
22f79326 1927
4dda9c8a
TM		 1928 if (reuseport)
1929 xs_sock_set_reuseport(sock);
1930
4cea288a
BH		 1931 err = xs_bind(transport, sock);
1932 if (err) {
22f79326
PE		 1933 sock_release(sock);
1934 goto out;
1935 }
1936
1937 return sock;
1938out:
1939 return ERR_PTR(err);
1940}
1941
176e21ee
CL		 1942static int xs_local_finish_connecting(struct rpc_xprt *xprt,
1943 struct socket *sock)
1944{
1945 struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1946 xprt);
1947
1948 if (!transport->inet) {
1949 struct sock *sk = sock->sk;
1950
1951 write_lock_bh(&sk->sk_callback_lock);
1952
1953 xs_save_old_callbacks(transport, sk);
1954
1955 sk->sk_user_data = xprt;
a2648094 1956 sk->sk_data_ready = xs_data_ready;
176e21ee 1957 sk->sk_write_space = xs_udp_write_space;
b4411457 1958 sock_set_flag(sk, SOCK_FASYNC);
2118071d 1959 sk->sk_error_report = xs_error_report;
c2126157 1960 sk->sk_allocation = GFP_NOIO;
176e21ee
CL		 1961
1962 xprt_clear_connected(xprt);
1963
1964 /* Reset to new socket */
1965 transport->sock = sock;
1966 transport->inet = sk;
1967
1968 write_unlock_bh(&sk->sk_callback_lock);
1969 }
1970
1971 /* Tell the socket layer to start connecting... */
1972 xprt->stat.connect_count++;
1973 xprt->stat.connect_start = jiffies;
1974 return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, 0);
1975}
1976
1977/**
1978 * xs_local_setup_socket - create AF_LOCAL socket, connect to a local endpoint
176e21ee 1979 * @transport: socket transport to connect
176e21ee 1980 */
dc107402 1981static int xs_local_setup_socket(struct sock_xprt *transport)
176e21ee 1982{
176e21ee
CL		 1983 struct rpc_xprt *xprt = &transport->xprt;
1984 struct socket *sock;
1985 int status = -EIO;
1986
176e21ee
CL		 1987 status = __sock_create(xprt->xprt_net, AF_LOCAL,
1988 SOCK_STREAM, 0, &sock, 1);
1989 if (status < 0) {
1990 dprintk("RPC: can't create AF_LOCAL "
1991 "transport socket (%d).\n", -status);
1992 goto out;
1993 }
d1358917 1994 xs_reclassify_socket(AF_LOCAL, sock);
176e21ee
CL		 1995
1996 dprintk("RPC: worker connecting xprt %p via AF_LOCAL to %s\n",
1997 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1998
1999 status = xs_local_finish_connecting(xprt, sock);
40b5ea0c 2000 trace_rpc_socket_connect(xprt, sock, status);
176e21ee
CL		 2001 switch (status) {
2002 case 0:
2003 dprintk("RPC: xprt %p connected to %s\n",
2004 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2005 xprt_set_connected(xprt);
3601c4a9 2006 case -ENOBUFS:
176e21ee
CL		 2007 break;
2008 case -ENOENT:
2009 dprintk("RPC: xprt %p: socket %s does not exist\n",
2010 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2011 break;
4a20a988
TM		 2012 case -ECONNREFUSED:
2013 dprintk("RPC: xprt %p: connection refused for %s\n",
2014 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2015 break;
176e21ee
CL		 2016 default:
2017 printk(KERN_ERR "%s: unhandled error (%d) connecting to %s\n",
2018 __func__, -status,
2019 xprt->address_strings[RPC_DISPLAY_ADDR]);
2020 }
2021
2022out:
2023 xprt_clear_connecting(xprt);
2024 xprt_wake_pending_tasks(xprt, status);
dc107402
BF		 2025 return status;
2026}
2027
b6669737 2028static void xs_local_connect(struct rpc_xprt *xprt, struct rpc_task *task)
dc107402 2029{
dc107402
BF		 2030 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2031 int ret;
2032
2033 if (RPC_IS_ASYNC(task)) {
2034 /*
2035 * We want the AF_LOCAL connect to be resolved in the
2036 * filesystem namespace of the process making the rpc
2037 * call. Thus we connect synchronously.
2038 *
2039 * If we want to support asynchronous AF_LOCAL calls,
2040 * we'll need to figure out how to pass a namespace to
2041 * connect.
2042 */
2043 rpc_exit(task, -ENOTCONN);
2044 return;
2045 }
2046 ret = xs_local_setup_socket(transport);
2047 if (ret && !RPC_IS_SOFTCONN(task))
2048 msleep_interruptible(15000);
176e21ee
CL		 2049}
2050
3c87ef6e 2051#if IS_ENABLED(CONFIG_SUNRPC_SWAP)
d6e971d8
JL		 2052/*
2053 * Note that this should be called with XPRT_LOCKED held (or when we otherwise
2054 * know that we have exclusive access to the socket), to guard against
2055 * races with xs_reset_transport.
2056 */
a564b8f0
MG		 2057static void xs_set_memalloc(struct rpc_xprt *xprt)
2058{
2059 struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
2060 xprt);
2061
d6e971d8
JL		 2062 /*
2063 * If there's no sock, then we have nothing to set. The
2064 * reconnecting process will get it for us.
2065 */
2066 if (!transport->inet)
2067 return;
8e228133 2068 if (atomic_read(&xprt->swapper))
a564b8f0
MG		 2069 sk_set_memalloc(transport->inet);
2070}
2071
2072/**
d67fa4d8 2073 * xs_enable_swap - Tag this transport as being used for swap.
a564b8f0 2074 * @xprt: transport to tag
a564b8f0 2075 *
8e228133
JL		 2076 * Take a reference to this transport on behalf of the rpc_clnt, and
2077 * optionally mark it for swapping if it wasn't already.
a564b8f0 2078 */
d67fa4d8
JL		 2079static int
2080xs_enable_swap(struct rpc_xprt *xprt)
a564b8f0 2081{
d6e971d8 2082 struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
a564b8f0 2083
d6e971d8
JL		 2084 if (atomic_inc_return(&xprt->swapper) != 1)
2085 return 0;
2086 if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE))
2087 return -ERESTARTSYS;
2088 if (xs->inet)
2089 sk_set_memalloc(xs->inet);
2090 xprt_release_xprt(xprt, NULL);
8e228133
JL		 2091 return 0;
2092}
a564b8f0 2093
8e228133 2094/**
d67fa4d8 2095 * xs_disable_swap - Untag this transport as being used for swap.
8e228133
JL		 2096 * @xprt: transport to tag
2097 *
2098 * Drop a "swapper" reference to this xprt on behalf of the rpc_clnt. If the
2099 * swapper refcount goes to 0, untag the socket as a memalloc socket.
2100 */
d67fa4d8
JL		 2101static void
2102xs_disable_swap(struct rpc_xprt *xprt)
8e228133 2103{
d6e971d8 2104 struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
8e228133 2105
d6e971d8
JL		 2106 if (!atomic_dec_and_test(&xprt->swapper))
2107 return;
2108 if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE))
2109 return;
2110 if (xs->inet)
2111 sk_clear_memalloc(xs->inet);
2112 xprt_release_xprt(xprt, NULL);
a564b8f0 2113}
a564b8f0
MG		 2114#else
2115static void xs_set_memalloc(struct rpc_xprt *xprt)
2116{
2117}
d67fa4d8
JL		 2118
2119static int
2120xs_enable_swap(struct rpc_xprt *xprt)
2121{
2122 return -EINVAL;
2123}
2124
2125static void
2126xs_disable_swap(struct rpc_xprt *xprt)
2127{
2128}
a564b8f0
MG		 2129#endif
2130
16be2d20
CL		 2131static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2132{
2133 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2134
2135 if (!transport->inet) {
2136 struct sock *sk = sock->sk;
2137
2138 write_lock_bh(&sk->sk_callback_lock);
2139
2a9e1cfa
TM		 2140 xs_save_old_callbacks(transport, sk);
2141
16be2d20 2142 sk->sk_user_data = xprt;
f9b2ee71 2143 sk->sk_data_ready = xs_data_ready;
16be2d20 2144 sk->sk_write_space = xs_udp_write_space;
b4411457 2145 sock_set_flag(sk, SOCK_FASYNC);
c2126157 2146 sk->sk_allocation = GFP_NOIO;
16be2d20
CL		 2147
2148 xprt_set_connected(xprt);
2149
2150 /* Reset to new socket */
2151 transport->sock = sock;
2152 transport->inet = sk;
2153
a564b8f0
MG		 2154 xs_set_memalloc(xprt);
2155
16be2d20
CL		 2156 write_unlock_bh(&sk->sk_callback_lock);
2157 }
2158 xs_udp_do_set_buffer_size(xprt);
2159}
2160
8c14ff2a 2161static void xs_udp_setup_socket(struct work_struct *work)
a246b010 2162{
34161db6
TM		 2163 struct sock_xprt *transport =
2164 container_of(work, struct sock_xprt, connect_worker.work);
c8475461 2165 struct rpc_xprt *xprt = &transport->xprt;
ee0ac0c2 2166 struct socket *sock = transport->sock;
b65c0310 2167 int status = -EIO;
9903cd1c 2168
8c14ff2a 2169 sock = xs_create_sock(xprt, transport,
4dda9c8a
TM		 2170 xs_addr(xprt)->sa_family, SOCK_DGRAM,
2171 IPPROTO_UDP, false);
b65c0310 2172 if (IS_ERR(sock))
b0d93ad5 2173 goto out;
68e220bd 2174
c740eff8
CL		 2175 dprintk("RPC: worker connecting xprt %p via %s to "
2176 "%s (port %s)\n", xprt,
2177 xprt->address_strings[RPC_DISPLAY_PROTO],
2178 xprt->address_strings[RPC_DISPLAY_ADDR],
2179 xprt->address_strings[RPC_DISPLAY_PORT]);
68e220bd
CL		 2180
2181 xs_udp_finish_connecting(xprt, sock);
40b5ea0c 2182 trace_rpc_socket_connect(xprt, sock, 0);
b0d93ad5
CL		 2183 status = 0;
2184out:
718ba5b8 2185 xprt_unlock_connect(xprt, transport);
b0d93ad5 2186 xprt_clear_connecting(xprt);
7d1e8255 2187 xprt_wake_pending_tasks(xprt, status);
a246b010
CL		 2188}
2189
4876cc77
TM		 2190/**
2191 * xs_tcp_shutdown - gracefully shut down a TCP socket
2192 * @xprt: transport
2193 *
2194 * Initiates a graceful shutdown of the TCP socket by calling the
2195 * equivalent of shutdown(SHUT_RDWR);
2196 */
2197static void xs_tcp_shutdown(struct rpc_xprt *xprt)
2198{
2199 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2200 struct socket *sock = transport->sock;
2201
2202 if (sock == NULL)
2203 return;
2204 if (xprt_connected(xprt)) {
2205 kernel_sock_shutdown(sock, SHUT_RDWR);
2206 trace_rpc_socket_shutdown(xprt, sock);
2207 } else
2208 xs_reset_transport(transport);
2209}
2210
16be2d20 2211static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
a246b010 2212{
16be2d20 2213 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
fe19a96b 2214 int ret = -ENOTCONN;
edb267a6 2215
ee0ac0c2 2216 if (!transport->inet) {
b0d93ad5 2217 struct sock *sk = sock->sk;
7f260e85
TM		 2218 unsigned int keepidle = xprt->timeout->to_initval / HZ;
2219 unsigned int keepcnt = xprt->timeout->to_retries + 1;
2220 unsigned int opt_on = 1;
775f06ab 2221 unsigned int timeo;
7f260e85
TM		 2222
2223 /* TCP Keepalive options */
2224 kernel_setsockopt(sock, SOL_SOCKET, SO_KEEPALIVE,
2225 (char *)&opt_on, sizeof(opt_on));
2226 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPIDLE,
2227 (char *)&keepidle, sizeof(keepidle));
2228 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPINTVL,
2229 (char *)&keepidle, sizeof(keepidle));
2230 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPCNT,
2231 (char *)&keepcnt, sizeof(keepcnt));
b0d93ad5 2232
775f06ab
TM		 2233 /* TCP user timeout (see RFC5482) */
2234 timeo = jiffies_to_msecs(xprt->timeout->to_initval) *
2235 (xprt->timeout->to_retries + 1);
2236 kernel_setsockopt(sock, SOL_TCP, TCP_USER_TIMEOUT,
2237 (char *)&timeo, sizeof(timeo));
2238
b0d93ad5
CL		 2239 write_lock_bh(&sk->sk_callback_lock);
2240
2a9e1cfa
TM		 2241 xs_save_old_callbacks(transport, sk);
2242
b0d93ad5 2243 sk->sk_user_data = xprt;
b0d93ad5
CL		 2244 sk->sk_data_ready = xs_tcp_data_ready;
2245 sk->sk_state_change = xs_tcp_state_change;
2246 sk->sk_write_space = xs_tcp_write_space;
b4411457 2247 sock_set_flag(sk, SOCK_FASYNC);
2118071d 2248 sk->sk_error_report = xs_error_report;
c2126157 2249 sk->sk_allocation = GFP_NOIO;
3167e12c
CL		 2250
2251 /* socket options */
3167e12c 2252 sock_reset_flag(sk, SOCK_LINGER);
3167e12c 2253 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
b0d93ad5
CL		 2254
2255 xprt_clear_connected(xprt);
2256
2257 /* Reset to new socket */
ee0ac0c2
CL		 2258 transport->sock = sock;
2259 transport->inet = sk;
b0d93ad5
CL		 2260
2261 write_unlock_bh(&sk->sk_callback_lock);
2262 }
2263
01d37c42 2264 if (!xprt_bound(xprt))
fe19a96b 2265 goto out;
01d37c42 2266
a564b8f0
MG		 2267 xs_set_memalloc(xprt);
2268
b0d93ad5 2269 /* Tell the socket layer to start connecting... */
262ca07d
CL		 2270 xprt->stat.connect_count++;
2271 xprt->stat.connect_start = jiffies;
0fdea1e8 2272 set_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
fe19a96b
TM		 2273 ret = kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
2274 switch (ret) {
2275 case 0:
4dda9c8a 2276 xs_set_srcport(transport, sock);
fe19a96b
TM		 2277 case -EINPROGRESS:
2278 /* SYN_SENT! */
fe19a96b
TM		 2279 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2280 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2281 }
2282out:
2283 return ret;
16be2d20
CL		 2284}
2285
9903cd1c 2286/**
b61d59ff 2287 * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint
9903cd1c
CL		 2288 *
2289 * Invoked by a work queue tasklet.
a246b010 2290 */
cdd518d5 2291static void xs_tcp_setup_socket(struct work_struct *work)
a246b010 2292{
cdd518d5
PE		 2293 struct sock_xprt *transport =
2294 container_of(work, struct sock_xprt, connect_worker.work);
ee0ac0c2 2295 struct socket *sock = transport->sock;
a9f5f0f7 2296 struct rpc_xprt *xprt = &transport->xprt;
b61d59ff 2297 int status = -EIO;
a246b010 2298
ee0ac0c2 2299 if (!sock) {
cdd518d5 2300 sock = xs_create_sock(xprt, transport,
4dda9c8a
TM		 2301 xs_addr(xprt)->sa_family, SOCK_STREAM,
2302 IPPROTO_TCP, true);
b61d59ff
TM		 2303 if (IS_ERR(sock)) {
2304 status = PTR_ERR(sock);
3167e12c
CL		 2305 goto out;
2306 }
7d1e8255 2307 }
a246b010 2308
c740eff8
CL		 2309 dprintk("RPC: worker connecting xprt %p via %s to "
2310 "%s (port %s)\n", xprt,
2311 xprt->address_strings[RPC_DISPLAY_PROTO],
2312 xprt->address_strings[RPC_DISPLAY_ADDR],
2313 xprt->address_strings[RPC_DISPLAY_PORT]);
edb267a6 2314
16be2d20 2315 status = xs_tcp_finish_connecting(xprt, sock);
40b5ea0c 2316 trace_rpc_socket_connect(xprt, sock, status);
46121cf7
CL		 2317 dprintk("RPC: %p connect status %d connected %d sock state %d\n",
2318 xprt, -status, xprt_connected(xprt),
2319 sock->sk->sk_state);
2a491991 2320 switch (status) {
f75e6745
TM		 2321 default:
2322 printk("%s: connect returned unhandled error %d\n",
2323 __func__, status);
2324 case -EADDRNOTAVAIL:
2325 /* We're probably in TIME_WAIT. Get rid of existing socket,
2326 * and retry
2327 */
a519fc7a 2328 xs_tcp_force_close(xprt);
88b5ed73 2329 break;
2a491991
TM		 2330 case 0:
2331 case -EINPROGRESS:
2332 case -EALREADY:
718ba5b8 2333 xprt_unlock_connect(xprt, transport);
7d1e8255 2334 return;
9fcfe0c8
TM		 2335 case -EINVAL:
2336 /* Happens, for instance, if the user specified a link
2337 * local IPv6 address without a scope-id.
2338 */
3ed5e2a2
TM		 2339 case -ECONNREFUSED:
2340 case -ECONNRESET:
2341 case -ENETUNREACH:
3913c78c 2342 case -EADDRINUSE:
3601c4a9 2343 case -ENOBUFS:
3ed5e2a2 2344 /* retry with existing socket, after a delay */
4efdd92c 2345 xs_tcp_force_close(xprt);
9fcfe0c8 2346 goto out;
a246b010 2347 }
2a491991 2348 status = -EAGAIN;
a246b010 2349out:
718ba5b8 2350 xprt_unlock_connect(xprt, transport);
68e220bd 2351 xprt_clear_connecting(xprt);
7d1e8255 2352 xprt_wake_pending_tasks(xprt, status);
68e220bd 2353}
b0d93ad5 2354
9903cd1c
CL		 2355/**
2356 * xs_connect - connect a socket to a remote endpoint
1b092092 2357 * @xprt: pointer to transport structure
9903cd1c
CL		 2358 * @task: address of RPC task that manages state of connect request
2359 *
2360 * TCP: If the remote end dropped the connection, delay reconnecting.
03bf4b70
CL		 2361 *
2362 * UDP socket connects are synchronous, but we use a work queue anyway
2363 * to guarantee that even unprivileged user processes can set up a
2364 * socket on a privileged port.
2365 *
2366 * If a UDP socket connect fails, the delay behavior here prevents
2367 * retry floods (hard mounts).
9903cd1c 2368 */
1b092092 2369static void xs_connect(struct rpc_xprt *xprt, struct rpc_task *task)
a246b010 2370{
ee0ac0c2 2371 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
a246b010 2372
718ba5b8
TM		 2373 WARN_ON_ONCE(!xprt_lock_connect(xprt, task, transport));
2374
99b1a4c3 2375 if (transport->sock != NULL) {
46121cf7
CL		 2376 dprintk("RPC: xs_connect delayed xprt %p for %lu "
2377 "seconds\n",
03bf4b70 2378 xprt, xprt->reestablish_timeout / HZ);
99b1a4c3
TM		 2379
2380 /* Start by resetting any existing state */
2381 xs_reset_transport(transport);
2382
c1384c9c
TM		 2383 queue_delayed_work(rpciod_workqueue,
2384 &transport->connect_worker,
2385 xprt->reestablish_timeout);
03bf4b70 2386 xprt->reestablish_timeout <<= 1;
61d0a8e6
NB		 2387 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2388 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
03bf4b70
CL		 2389 if (xprt->reestablish_timeout > XS_TCP_MAX_REEST_TO)
2390 xprt->reestablish_timeout = XS_TCP_MAX_REEST_TO;
b0d93ad5 2391 } else {
46121cf7 2392 dprintk("RPC: xs_connect scheduled xprt %p\n", xprt);
c1384c9c
TM		 2393 queue_delayed_work(rpciod_workqueue,
2394 &transport->connect_worker, 0);
a246b010
CL		 2395 }
2396}
2397
176e21ee
CL		 2398/**
2399 * xs_local_print_stats - display AF_LOCAL socket-specifc stats
2400 * @xprt: rpc_xprt struct containing statistics
2401 * @seq: output file
2402 *
2403 */
2404static void xs_local_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2405{
2406 long idle_time = 0;
2407
2408 if (xprt_connected(xprt))
2409 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2410
2411 seq_printf(seq, "\txprt:\tlocal %lu %lu %lu %ld %lu %lu %lu "
15a45206 2412 "%llu %llu %lu %llu %llu\n",
176e21ee
CL		 2413 xprt->stat.bind_count,
2414 xprt->stat.connect_count,
2415 xprt->stat.connect_time,
2416 idle_time,
2417 xprt->stat.sends,
2418 xprt->stat.recvs,
2419 xprt->stat.bad_xids,
2420 xprt->stat.req_u,
15a45206
AA		 2421 xprt->stat.bklog_u,
2422 xprt->stat.max_slots,
2423 xprt->stat.sending_u,
2424 xprt->stat.pending_u);
176e21ee
CL		 2425}
2426
262ca07d
CL		 2427/**
2428 * xs_udp_print_stats - display UDP socket-specifc stats
2429 * @xprt: rpc_xprt struct containing statistics
2430 * @seq: output file
2431 *
2432 */
2433static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2434{
c8475461
CL		 2435 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2436
15a45206
AA		 2437 seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %llu %llu "
2438 "%lu %llu %llu\n",
fbfffbd5 2439 transport->srcport,
262ca07d
CL		 2440 xprt->stat.bind_count,
2441 xprt->stat.sends,
2442 xprt->stat.recvs,
2443 xprt->stat.bad_xids,
2444 xprt->stat.req_u,
15a45206
AA		 2445 xprt->stat.bklog_u,
2446 xprt->stat.max_slots,
2447 xprt->stat.sending_u,
2448 xprt->stat.pending_u);
262ca07d
CL		 2449}
2450
2451/**
2452 * xs_tcp_print_stats - display TCP socket-specifc stats
2453 * @xprt: rpc_xprt struct containing statistics
2454 * @seq: output file
2455 *
2456 */
2457static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2458{
c8475461 2459 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
262ca07d
CL		 2460 long idle_time = 0;
2461
2462 if (xprt_connected(xprt))
2463 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2464
15a45206
AA		 2465 seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu "
2466 "%llu %llu %lu %llu %llu\n",
fbfffbd5 2467 transport->srcport,
262ca07d
CL		 2468 xprt->stat.bind_count,
2469 xprt->stat.connect_count,
2470 xprt->stat.connect_time,
2471 idle_time,
2472 xprt->stat.sends,
2473 xprt->stat.recvs,
2474 xprt->stat.bad_xids,
2475 xprt->stat.req_u,
15a45206
AA		 2476 xprt->stat.bklog_u,
2477 xprt->stat.max_slots,
2478 xprt->stat.sending_u,
2479 xprt->stat.pending_u);
262ca07d
CL		 2480}
2481
4cfc7e60
RI		 2482/*
2483 * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason
2484 * we allocate pages instead doing a kmalloc like rpc_malloc is because we want
2485 * to use the server side send routines.
2486 */
5a51f13a 2487static void *bc_malloc(struct rpc_task *task, size_t size)
4cfc7e60
RI		 2488{
2489 struct page *page;
2490 struct rpc_buffer *buf;
2491
b8a13d03
WAA		 2492 WARN_ON_ONCE(size > PAGE_SIZE - sizeof(struct rpc_buffer));
2493 if (size > PAGE_SIZE - sizeof(struct rpc_buffer))
2494 return NULL;
4cfc7e60 2495
b8a13d03 2496 page = alloc_page(GFP_KERNEL);
4cfc7e60
RI		 2497 if (!page)
2498 return NULL;
2499
2500 buf = page_address(page);
2501 buf->len = PAGE_SIZE;
2502
2503 return buf->data;
2504}
2505
2506/*
2507 * Free the space allocated in the bc_alloc routine
2508 */
5a51f13a 2509static void bc_free(void *buffer)
4cfc7e60
RI		 2510{
2511 struct rpc_buffer *buf;
2512
2513 if (!buffer)
2514 return;
2515
2516 buf = container_of(buffer, struct rpc_buffer, data);
2517 free_page((unsigned long)buf);
2518}
2519
2520/*
2521 * Use the svc_sock to send the callback. Must be called with svsk->sk_mutex
2522 * held. Borrows heavily from svc_tcp_sendto and xs_tcp_send_request.
2523 */
2524static int bc_sendto(struct rpc_rqst *req)
2525{
2526 int len;
2527 struct xdr_buf *xbufp = &req->rq_snd_buf;
2528 struct rpc_xprt *xprt = req->rq_xprt;
2529 struct sock_xprt *transport =
2530 container_of(xprt, struct sock_xprt, xprt);
2531 struct socket *sock = transport->sock;
2532 unsigned long headoff;
2533 unsigned long tailoff;
2534
61677eee 2535 xs_encode_stream_record_marker(xbufp);
4cfc7e60
RI		 2536
2537 tailoff = (unsigned long)xbufp->tail[0].iov_base & ~PAGE_MASK;
2538 headoff = (unsigned long)xbufp->head[0].iov_base & ~PAGE_MASK;
2539 len = svc_send_common(sock, xbufp,
2540 virt_to_page(xbufp->head[0].iov_base), headoff,
2541 xbufp->tail[0].iov_base, tailoff);
2542
2543 if (len != xbufp->len) {
2544 printk(KERN_NOTICE "Error sending entire callback!\n");
2545 len = -EAGAIN;
2546 }
2547
2548 return len;
2549}
2550
2551/*
2552 * The send routine. Borrows from svc_send
2553 */
2554static int bc_send_request(struct rpc_task *task)
2555{
2556 struct rpc_rqst *req = task->tk_rqstp;
2557 struct svc_xprt *xprt;
7fc56136 2558 int len;
4cfc7e60
RI		 2559
2560 dprintk("sending request with xid: %08x\n", ntohl(req->rq_xid));
2561 /*
2562 * Get the server socket associated with this callback xprt
2563 */
2564 xprt = req->rq_xprt->bc_xprt;
4cfc7e60
RI		 2565
2566 /*
2567 * Grab the mutex to serialize data as the connection is shared
2568 * with the fore channel
2569 */
2570 if (!mutex_trylock(&xprt->xpt_mutex)) {
2571 rpc_sleep_on(&xprt->xpt_bc_pending, task, NULL);
2572 if (!mutex_trylock(&xprt->xpt_mutex))
2573 return -EAGAIN;
2574 rpc_wake_up_queued_task(&xprt->xpt_bc_pending, task);
2575 }
2576 if (test_bit(XPT_DEAD, &xprt->xpt_flags))
2577 len = -ENOTCONN;
2578 else
2579 len = bc_sendto(req);
2580 mutex_unlock(&xprt->xpt_mutex);
2581
2582 if (len > 0)
2583 len = 0;
2584
2585 return len;
2586}
2587
2588/*
2589 * The close routine. Since this is client initiated, we do nothing
2590 */
2591
2592static void bc_close(struct rpc_xprt *xprt)
2593{
4cfc7e60
RI		 2594}
2595
2596/*
2597 * The xprt destroy routine. Again, because this connection is client
2598 * initiated, we do nothing
2599 */
2600
2601static void bc_destroy(struct rpc_xprt *xprt)
2602{
47f72efa
KM		 2603 dprintk("RPC: bc_destroy xprt %p\n", xprt);
2604
2605 xs_xprt_free(xprt);
2606 module_put(THIS_MODULE);
4cfc7e60
RI		 2607}
2608
176e21ee
CL		 2609static struct rpc_xprt_ops xs_local_ops = {
2610 .reserve_xprt = xprt_reserve_xprt,
2611 .release_xprt = xs_tcp_release_xprt,
f39c1bfb 2612 .alloc_slot = xprt_alloc_slot,
176e21ee
CL		 2613 .rpcbind = xs_local_rpcbind,
2614 .set_port = xs_local_set_port,
dc107402 2615 .connect = xs_local_connect,
176e21ee
CL		 2616 .buf_alloc = rpc_malloc,
2617 .buf_free = rpc_free,
2618 .send_request = xs_local_send_request,
2619 .set_retrans_timeout = xprt_set_retrans_timeout_def,
2620 .close = xs_close,
a1311d87 2621 .destroy = xs_destroy,
176e21ee 2622 .print_stats = xs_local_print_stats,
d67fa4d8
JL		 2623 .enable_swap = xs_enable_swap,
2624 .disable_swap = xs_disable_swap,
176e21ee
CL		 2625};
2626
262965f5 2627static struct rpc_xprt_ops xs_udp_ops = {
43118c29 2628 .set_buffer_size = xs_udp_set_buffer_size,
12a80469 2629 .reserve_xprt = xprt_reserve_xprt_cong,
49e9a890 2630 .release_xprt = xprt_release_xprt_cong,
f39c1bfb 2631 .alloc_slot = xprt_alloc_slot,
45160d62 2632 .rpcbind = rpcb_getport_async,
92200412 2633 .set_port = xs_set_port,
262965f5 2634 .connect = xs_connect,
02107148
CL		 2635 .buf_alloc = rpc_malloc,
2636 .buf_free = rpc_free,
262965f5 2637 .send_request = xs_udp_send_request,
fe3aca29 2638 .set_retrans_timeout = xprt_set_retrans_timeout_rtt,
46c0ee8b 2639 .timer = xs_udp_timer,
a58dd398 2640 .release_request = xprt_release_rqst_cong,
262965f5
CL		 2641 .close = xs_close,
2642 .destroy = xs_destroy,
262ca07d 2643 .print_stats = xs_udp_print_stats,
d67fa4d8
JL		 2644 .enable_swap = xs_enable_swap,
2645 .disable_swap = xs_disable_swap,
4a068258 2646 .inject_disconnect = xs_inject_disconnect,
262965f5
CL		 2647};
2648
2649static struct rpc_xprt_ops xs_tcp_ops = {
12a80469 2650 .reserve_xprt = xprt_reserve_xprt,
e0ab53de 2651 .release_xprt = xs_tcp_release_xprt,
f39c1bfb 2652 .alloc_slot = xprt_lock_and_alloc_slot,
45160d62 2653 .rpcbind = rpcb_getport_async,
92200412 2654 .set_port = xs_set_port,
0b9e7943 2655 .connect = xs_connect,
02107148
CL		 2656 .buf_alloc = rpc_malloc,
2657 .buf_free = rpc_free,
262965f5 2658 .send_request = xs_tcp_send_request,
fe3aca29 2659 .set_retrans_timeout = xprt_set_retrans_timeout_def,
c627d31b 2660 .close = xs_tcp_shutdown,
9903cd1c 2661 .destroy = xs_destroy,
262ca07d 2662 .print_stats = xs_tcp_print_stats,
d67fa4d8
JL		 2663 .enable_swap = xs_enable_swap,
2664 .disable_swap = xs_disable_swap,
4a068258 2665 .inject_disconnect = xs_inject_disconnect,
42e5c3e2
CL		 2666#ifdef CONFIG_SUNRPC_BACKCHANNEL
2667 .bc_setup = xprt_setup_bc,
76566773 2668 .bc_up = xs_tcp_bc_up,
6b26cc8c 2669 .bc_maxpayload = xs_tcp_bc_maxpayload,
42e5c3e2
CL		 2670 .bc_free_rqst = xprt_free_bc_rqst,
2671 .bc_destroy = xprt_destroy_bc,
2672#endif
a246b010
CL		 2673};
2674
4cfc7e60
RI		 2675/*
2676 * The rpc_xprt_ops for the server backchannel
2677 */
2678
2679static struct rpc_xprt_ops bc_tcp_ops = {
2680 .reserve_xprt = xprt_reserve_xprt,
2681 .release_xprt = xprt_release_xprt,
84e28a30 2682 .alloc_slot = xprt_alloc_slot,
4cfc7e60
RI		 2683 .buf_alloc = bc_malloc,
2684 .buf_free = bc_free,
2685 .send_request = bc_send_request,
2686 .set_retrans_timeout = xprt_set_retrans_timeout_def,
2687 .close = bc_close,
2688 .destroy = bc_destroy,
2689 .print_stats = xs_tcp_print_stats,
d67fa4d8
JL		 2690 .enable_swap = xs_enable_swap,
2691 .disable_swap = xs_disable_swap,
4a068258 2692 .inject_disconnect = xs_inject_disconnect,
4cfc7e60
RI		 2693};
2694
92476850
CL		 2695static int xs_init_anyaddr(const int family, struct sockaddr *sap)
2696{
2697 static const struct sockaddr_in sin = {
2698 .sin_family = AF_INET,
2699 .sin_addr.s_addr = htonl(INADDR_ANY),
2700 };
2701 static const struct sockaddr_in6 sin6 = {
2702 .sin6_family = AF_INET6,
2703 .sin6_addr = IN6ADDR_ANY_INIT,
2704 };
2705
2706 switch (family) {
176e21ee
CL		 2707 case AF_LOCAL:
2708 break;
92476850
CL		 2709 case AF_INET:
2710 memcpy(sap, &sin, sizeof(sin));
2711 break;
2712 case AF_INET6:
2713 memcpy(sap, &sin6, sizeof(sin6));
2714 break;
2715 default:
2716 dprintk("RPC: %s: Bad address family\n", __func__);
2717 return -EAFNOSUPPORT;
2718 }
2719 return 0;
2720}
2721
3c341b0b 2722static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
d9ba131d
TM		 2723 unsigned int slot_table_size,
2724 unsigned int max_slot_table_size)
c8541ecd
CL		 2725{
2726 struct rpc_xprt *xprt;
ffc2e518 2727 struct sock_xprt *new;
c8541ecd 2728
96802a09 2729 if (args->addrlen > sizeof(xprt->addr)) {
46121cf7 2730 dprintk("RPC: xs_setup_xprt: address too large\n");
c8541ecd
CL		 2731 return ERR_PTR(-EBADF);
2732 }
2733
d9ba131d
TM		 2734 xprt = xprt_alloc(args->net, sizeof(*new), slot_table_size,
2735 max_slot_table_size);
bd1722d4 2736 if (xprt == NULL) {
46121cf7
CL		 2737 dprintk("RPC: xs_setup_xprt: couldn't allocate "
2738 "rpc_xprt\n");
c8541ecd
CL		 2739 return ERR_PTR(-ENOMEM);
2740 }
c8541ecd 2741
bd1722d4 2742 new = container_of(xprt, struct sock_xprt, xprt);
edc1b01c 2743 mutex_init(&new->recv_mutex);
96802a09
FM		 2744 memcpy(&xprt->addr, args->dstaddr, args->addrlen);
2745 xprt->addrlen = args->addrlen;
d3bc9a1d 2746 if (args->srcaddr)
fbfffbd5 2747 memcpy(&new->srcaddr, args->srcaddr, args->addrlen);
92476850
CL		 2748 else {
2749 int err;
2750 err = xs_init_anyaddr(args->dstaddr->sa_family,
2751 (struct sockaddr *)&new->srcaddr);
2aa13531
SK		 2752 if (err != 0) {
2753 xprt_free(xprt);
92476850 2754 return ERR_PTR(err);
2aa13531 2755 }
92476850 2756 }
c8541ecd
CL		 2757
2758 return xprt;
2759}
2760
176e21ee
CL		 2761static const struct rpc_timeout xs_local_default_timeout = {
2762 .to_initval = 10 * HZ,
2763 .to_maxval = 10 * HZ,
2764 .to_retries = 2,
2765};
2766
2767/**
2768 * xs_setup_local - Set up transport to use an AF_LOCAL socket
2769 * @args: rpc transport creation arguments
2770 *
2771 * AF_LOCAL is a "tpi_cots_ord" transport, just like TCP
2772 */
2773static struct rpc_xprt *xs_setup_local(struct xprt_create *args)
2774{
2775 struct sockaddr_un *sun = (struct sockaddr_un *)args->dstaddr;
2776 struct sock_xprt *transport;
2777 struct rpc_xprt *xprt;
2778 struct rpc_xprt *ret;
2779
d9ba131d
TM		 2780 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2781 xprt_max_tcp_slot_table_entries);
176e21ee
CL		 2782 if (IS_ERR(xprt))
2783 return xprt;
2784 transport = container_of(xprt, struct sock_xprt, xprt);
2785
2786 xprt->prot = 0;
2787 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2788 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2789
2790 xprt->bind_timeout = XS_BIND_TO;
2791 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2792 xprt->idle_timeout = XS_IDLE_DISC_TO;
2793
2794 xprt->ops = &xs_local_ops;
2795 xprt->timeout = &xs_local_default_timeout;
2796
a2648094 2797 INIT_WORK(&transport->recv_worker, xs_local_data_receive_workfn);
93dc41bd
N		 2798 INIT_DELAYED_WORK(&transport->connect_worker,
2799 xs_dummy_setup_socket);
2800
176e21ee
CL		 2801 switch (sun->sun_family) {
2802 case AF_LOCAL:
2803 if (sun->sun_path[0] != '/') {
2804 dprintk("RPC: bad AF_LOCAL address: %s\n",
2805 sun->sun_path);
2806 ret = ERR_PTR(-EINVAL);
2807 goto out_err;
2808 }
2809 xprt_set_bound(xprt);
176e21ee 2810 xs_format_peer_addresses(xprt, "local", RPCBIND_NETID_LOCAL);
7073ea87
BF		 2811 ret = ERR_PTR(xs_local_setup_socket(transport));
2812 if (ret)
2813 goto out_err;
176e21ee
CL		 2814 break;
2815 default:
2816 ret = ERR_PTR(-EAFNOSUPPORT);
2817 goto out_err;
2818 }
2819
2820 dprintk("RPC: set up xprt to %s via AF_LOCAL\n",
2821 xprt->address_strings[RPC_DISPLAY_ADDR]);
2822
2823 if (try_module_get(THIS_MODULE))
2824 return xprt;
2825 ret = ERR_PTR(-EINVAL);
2826out_err:
315f3812 2827 xs_xprt_free(xprt);
176e21ee
CL		 2828 return ret;
2829}
2830
2881ae74
TM		 2831static const struct rpc_timeout xs_udp_default_timeout = {
2832 .to_initval = 5 * HZ,
2833 .to_maxval = 30 * HZ,
2834 .to_increment = 5 * HZ,
2835 .to_retries = 5,
2836};
2837
9903cd1c
CL		 2838/**
2839 * xs_setup_udp - Set up transport to use a UDP socket
96802a09 2840 * @args: rpc transport creation arguments
9903cd1c
CL		 2841 *
2842 */
483066d6 2843static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
a246b010 2844{
8f9d5b1a 2845 struct sockaddr *addr = args->dstaddr;
c8541ecd 2846 struct rpc_xprt *xprt;
c8475461 2847 struct sock_xprt *transport;
0a68b0be 2848 struct rpc_xprt *ret;
a246b010 2849
d9ba131d
TM		 2850 xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries,
2851 xprt_udp_slot_table_entries);
c8541ecd
CL		 2852 if (IS_ERR(xprt))
2853 return xprt;
c8475461 2854 transport = container_of(xprt, struct sock_xprt, xprt);
a246b010 2855
ec739ef0 2856 xprt->prot = IPPROTO_UDP;
808012fb 2857 xprt->tsh_size = 0;
a246b010
CL		 2858 /* XXX: header size can vary due to auth type, IPv6, etc. */
2859 xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
2860
03bf4b70 2861 xprt->bind_timeout = XS_BIND_TO;
03bf4b70
CL		 2862 xprt->reestablish_timeout = XS_UDP_REEST_TO;
2863 xprt->idle_timeout = XS_IDLE_DISC_TO;
a246b010 2864
262965f5 2865 xprt->ops = &xs_udp_ops;
a246b010 2866
ba7392bb 2867 xprt->timeout = &xs_udp_default_timeout;
a246b010 2868
f9b2ee71 2869 INIT_WORK(&transport->recv_worker, xs_udp_data_receive_workfn);
edc1b01c
TM		 2870 INIT_DELAYED_WORK(&transport->connect_worker, xs_udp_setup_socket);
2871
8f9d5b1a
CL		 2872 switch (addr->sa_family) {
2873 case AF_INET:
2874 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2875 xprt_set_bound(xprt);
2876
9dc3b095 2877 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
8f9d5b1a
CL		 2878 break;
2879 case AF_INET6:
2880 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2881 xprt_set_bound(xprt);
2882
9dc3b095 2883 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
8f9d5b1a
CL		 2884 break;
2885 default:
0a68b0be
BF		 2886 ret = ERR_PTR(-EAFNOSUPPORT);
2887 goto out_err;
8f9d5b1a
CL		 2888 }
2889
c740eff8
CL		 2890 if (xprt_bound(xprt))
2891 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2892 xprt->address_strings[RPC_DISPLAY_ADDR],
2893 xprt->address_strings[RPC_DISPLAY_PORT],
2894 xprt->address_strings[RPC_DISPLAY_PROTO]);
2895 else
2896 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
2897 xprt->address_strings[RPC_DISPLAY_ADDR],
2898 xprt->address_strings[RPC_DISPLAY_PROTO]);
edb267a6 2899
bc25571e
TT		 2900 if (try_module_get(THIS_MODULE))
2901 return xprt;
0a68b0be
BF		 2902 ret = ERR_PTR(-EINVAL);
2903out_err:
315f3812 2904 xs_xprt_free(xprt);
0a68b0be 2905 return ret;
a246b010
CL		 2906}
2907
2881ae74
TM		 2908static const struct rpc_timeout xs_tcp_default_timeout = {
2909 .to_initval = 60 * HZ,
2910 .to_maxval = 60 * HZ,
2911 .to_retries = 2,
2912};
2913
9903cd1c
CL		 2914/**
2915 * xs_setup_tcp - Set up transport to use a TCP socket
96802a09 2916 * @args: rpc transport creation arguments
9903cd1c
CL		 2917 *
2918 */
483066d6 2919static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
a246b010 2920{
8f9d5b1a 2921 struct sockaddr *addr = args->dstaddr;
c8541ecd 2922 struct rpc_xprt *xprt;
c8475461 2923 struct sock_xprt *transport;
0a68b0be 2924 struct rpc_xprt *ret;
b7993ceb
TM		 2925 unsigned int max_slot_table_size = xprt_max_tcp_slot_table_entries;
2926
2927 if (args->flags & XPRT_CREATE_INFINITE_SLOTS)
2928 max_slot_table_size = RPC_MAX_SLOT_TABLE_LIMIT;
a246b010 2929
d9ba131d 2930 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
b7993ceb 2931 max_slot_table_size);
c8541ecd
CL		 2932 if (IS_ERR(xprt))
2933 return xprt;
c8475461 2934 transport = container_of(xprt, struct sock_xprt, xprt);
a246b010 2935
ec739ef0 2936 xprt->prot = IPPROTO_TCP;
808012fb 2937 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
808012fb 2938 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
a246b010 2939
03bf4b70 2940 xprt->bind_timeout = XS_BIND_TO;
03bf4b70
CL		 2941 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2942 xprt->idle_timeout = XS_IDLE_DISC_TO;
a246b010 2943
262965f5 2944 xprt->ops = &xs_tcp_ops;
ba7392bb 2945 xprt->timeout = &xs_tcp_default_timeout;
a246b010 2946
edc1b01c
TM		 2947 INIT_WORK(&transport->recv_worker, xs_tcp_data_receive_workfn);
2948 INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_setup_socket);
2949
8f9d5b1a
CL		 2950 switch (addr->sa_family) {
2951 case AF_INET:
2952 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2953 xprt_set_bound(xprt);
2954
9dc3b095 2955 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
8f9d5b1a
CL		 2956 break;
2957 case AF_INET6:
2958 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2959 xprt_set_bound(xprt);
2960
9dc3b095 2961 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
8f9d5b1a
CL		 2962 break;
2963 default:
0a68b0be
BF		 2964 ret = ERR_PTR(-EAFNOSUPPORT);
2965 goto out_err;
8f9d5b1a
CL		 2966 }
2967
c740eff8
CL		 2968 if (xprt_bound(xprt))
2969 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2970 xprt->address_strings[RPC_DISPLAY_ADDR],
2971 xprt->address_strings[RPC_DISPLAY_PORT],
2972 xprt->address_strings[RPC_DISPLAY_PROTO]);
2973 else
2974 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
2975 xprt->address_strings[RPC_DISPLAY_ADDR],
2976 xprt->address_strings[RPC_DISPLAY_PROTO]);
2977
bc25571e
TT		 2978 if (try_module_get(THIS_MODULE))
2979 return xprt;
0a68b0be
BF		 2980 ret = ERR_PTR(-EINVAL);
2981out_err:
315f3812 2982 xs_xprt_free(xprt);
0a68b0be 2983 return ret;
a246b010 2984}
282b32e1 2985
f300baba
AB		 2986/**
2987 * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket
2988 * @args: rpc transport creation arguments
2989 *
2990 */
2991static struct rpc_xprt *xs_setup_bc_tcp(struct xprt_create *args)
2992{
2993 struct sockaddr *addr = args->dstaddr;
2994 struct rpc_xprt *xprt;
2995 struct sock_xprt *transport;
2996 struct svc_sock *bc_sock;
0a68b0be 2997 struct rpc_xprt *ret;
f300baba 2998
d9ba131d
TM		 2999 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
3000 xprt_tcp_slot_table_entries);
f300baba
AB		 3001 if (IS_ERR(xprt))
3002 return xprt;
3003 transport = container_of(xprt, struct sock_xprt, xprt);
3004
3005 xprt->prot = IPPROTO_TCP;
3006 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
3007 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
3008 xprt->timeout = &xs_tcp_default_timeout;
3009
3010 /* backchannel */
3011 xprt_set_bound(xprt);
3012 xprt->bind_timeout = 0;
f300baba
AB		 3013 xprt->reestablish_timeout = 0;
3014 xprt->idle_timeout = 0;
3015
f300baba
AB		 3016 xprt->ops = &bc_tcp_ops;
3017
3018 switch (addr->sa_family) {
3019 case AF_INET:
3020 xs_format_peer_addresses(xprt, "tcp",
3021 RPCBIND_NETID_TCP);
3022 break;
3023 case AF_INET6:
3024 xs_format_peer_addresses(xprt, "tcp",
3025 RPCBIND_NETID_TCP6);
3026 break;
3027 default:
0a68b0be
BF		 3028 ret = ERR_PTR(-EAFNOSUPPORT);
3029 goto out_err;
f300baba
AB		 3030 }
3031
50fa0d40
PE		 3032 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
3033 xprt->address_strings[RPC_DISPLAY_ADDR],
3034 xprt->address_strings[RPC_DISPLAY_PORT],
3035 xprt->address_strings[RPC_DISPLAY_PROTO]);
f300baba 3036
99de8ea9
BF		 3037 /*
3038 * Once we've associated a backchannel xprt with a connection,
28303ca3
WM		 3039 * we want to keep it around as long as the connection lasts,
3040 * in case we need to start using it for a backchannel again;
3041 * this reference won't be dropped until bc_xprt is destroyed.
99de8ea9
BF		 3042 */
3043 xprt_get(xprt);
3044 args->bc_xprt->xpt_bc_xprt = xprt;
3045 xprt->bc_xprt = args->bc_xprt;
3046 bc_sock = container_of(args->bc_xprt, struct svc_sock, sk_xprt);
3047 transport->sock = bc_sock->sk_sock;
3048 transport->inet = bc_sock->sk_sk;
3049
f300baba
AB		 3050 /*
3051 * Since we don't want connections for the backchannel, we set
3052 * the xprt status to connected
3053 */
3054 xprt_set_connected(xprt);
3055
f300baba
AB		 3056 if (try_module_get(THIS_MODULE))
3057 return xprt;
642aab58
KM		 3058
3059 args->bc_xprt->xpt_bc_xprt = NULL;
99de8ea9 3060 xprt_put(xprt);
0a68b0be
BF		 3061 ret = ERR_PTR(-EINVAL);
3062out_err:
315f3812 3063 xs_xprt_free(xprt);
0a68b0be 3064 return ret;
f300baba
AB		 3065}
3066
176e21ee
CL		 3067static struct xprt_class xs_local_transport = {
3068 .list = LIST_HEAD_INIT(xs_local_transport.list),
3069 .name = "named UNIX socket",
3070 .owner = THIS_MODULE,
3071 .ident = XPRT_TRANSPORT_LOCAL,
3072 .setup = xs_setup_local,
3073};
3074
bc25571e
TT		 3075static struct xprt_class xs_udp_transport = {
3076 .list = LIST_HEAD_INIT(xs_udp_transport.list),
3077 .name = "udp",
3078 .owner = THIS_MODULE,
f300baba 3079 .ident = XPRT_TRANSPORT_UDP,
bc25571e
TT		 3080 .setup = xs_setup_udp,
3081};
3082
3083static struct xprt_class xs_tcp_transport = {
3084 .list = LIST_HEAD_INIT(xs_tcp_transport.list),
3085 .name = "tcp",
3086 .owner = THIS_MODULE,
f300baba 3087 .ident = XPRT_TRANSPORT_TCP,
bc25571e
TT		 3088 .setup = xs_setup_tcp,
3089};
3090
f300baba
AB		 3091static struct xprt_class xs_bc_tcp_transport = {
3092 .list = LIST_HEAD_INIT(xs_bc_tcp_transport.list),
3093 .name = "tcp NFSv4.1 backchannel",
3094 .owner = THIS_MODULE,
3095 .ident = XPRT_TRANSPORT_BC_TCP,
3096 .setup = xs_setup_bc_tcp,
3097};
3098
282b32e1 3099/**
bc25571e 3100 * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
282b32e1
CL		 3101 *
3102 */
3103int init_socket_xprt(void)
3104{
f895b252 3105#if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
2b1bec5f 3106 if (!sunrpc_table_header)
0b4d4147 3107 sunrpc_table_header = register_sysctl_table(sunrpc_table);
fbf76683
CL		 3108#endif
3109
176e21ee 3110 xprt_register_transport(&xs_local_transport);
bc25571e
TT		 3111 xprt_register_transport(&xs_udp_transport);
3112 xprt_register_transport(&xs_tcp_transport);
f300baba 3113 xprt_register_transport(&xs_bc_tcp_transport);
bc25571e 3114
282b32e1
CL		 3115 return 0;
3116}
3117
3118/**
bc25571e 3119 * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
282b32e1
CL		 3120 *
3121 */
3122void cleanup_socket_xprt(void)
3123{
f895b252 3124#if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
fbf76683
CL		 3125 if (sunrpc_table_header) {
3126 unregister_sysctl_table(sunrpc_table_header);
3127 sunrpc_table_header = NULL;
3128 }
3129#endif
bc25571e 3130
176e21ee 3131 xprt_unregister_transport(&xs_local_transport);
bc25571e
TT		 3132 xprt_unregister_transport(&xs_udp_transport);
3133 xprt_unregister_transport(&xs_tcp_transport);
f300baba 3134 xprt_unregister_transport(&xs_bc_tcp_transport);
282b32e1 3135}
cbf11071 3136
9bbb9e5a
RR		 3137static int param_set_uint_minmax(const char *val,
3138 const struct kernel_param *kp,
cbf11071
TM		 3139 unsigned int min, unsigned int max)
3140{
00cfaa94 3141 unsigned int num;
cbf11071
TM		 3142 int ret;
3143
3144 if (!val)
3145 return -EINVAL;
00cfaa94 3146 ret = kstrtouint(val, 0, &num);
cbf11071
TM		 3147 if (ret == -EINVAL || num < min || num > max)
3148 return -EINVAL;
3149 *((unsigned int *)kp->arg) = num;
3150 return 0;
3151}
3152
9bbb9e5a 3153static int param_set_portnr(const char *val, const struct kernel_param *kp)
cbf11071
TM		 3154{
3155 return param_set_uint_minmax(val, kp,
3156 RPC_MIN_RESVPORT,
3157 RPC_MAX_RESVPORT);
3158}
3159
9c27847d 3160static const struct kernel_param_ops param_ops_portnr = {
9bbb9e5a
RR		 3161 .set = param_set_portnr,
3162 .get = param_get_uint,
3163};
3164
cbf11071
TM		 3165#define param_check_portnr(name, p) \
3166 __param_check(name, p, unsigned int);
3167
3168module_param_named(min_resvport, xprt_min_resvport, portnr, 0644);
3169module_param_named(max_resvport, xprt_max_resvport, portnr, 0644);
3170
9bbb9e5a
RR		 3171static int param_set_slot_table_size(const char *val,
3172 const struct kernel_param *kp)
cbf11071
TM		 3173{
3174 return param_set_uint_minmax(val, kp,
3175 RPC_MIN_SLOT_TABLE,
3176 RPC_MAX_SLOT_TABLE);
3177}
3178
9c27847d 3179static const struct kernel_param_ops param_ops_slot_table_size = {
9bbb9e5a
RR		 3180 .set = param_set_slot_table_size,
3181 .get = param_get_uint,
3182};
3183
cbf11071
TM		 3184#define param_check_slot_table_size(name, p) \
3185 __param_check(name, p, unsigned int);
3186
d9ba131d
TM		 3187static int param_set_max_slot_table_size(const char *val,
3188 const struct kernel_param *kp)
3189{
3190 return param_set_uint_minmax(val, kp,
3191 RPC_MIN_SLOT_TABLE,
3192 RPC_MAX_SLOT_TABLE_LIMIT);
3193}
3194
9c27847d 3195static const struct kernel_param_ops param_ops_max_slot_table_size = {
d9ba131d
TM		 3196 .set = param_set_max_slot_table_size,
3197 .get = param_get_uint,
3198};
3199
3200#define param_check_max_slot_table_size(name, p) \
3201 __param_check(name, p, unsigned int);
3202
cbf11071
TM		 3203module_param_named(tcp_slot_table_entries, xprt_tcp_slot_table_entries,
3204 slot_table_size, 0644);
d9ba131d
TM		 3205module_param_named(tcp_max_slot_table_entries, xprt_max_tcp_slot_table_entries,
3206 max_slot_table_size, 0644);
cbf11071
TM		 3207module_param_named(udp_slot_table_entries, xprt_udp_slot_table_entries,
3208 slot_table_size, 0644);
3209
Linux 6.x block layer and io_uring tree(s)