projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
SUNRPC: Move UDP receive data path into a workqueue context
[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
TM		 362 remainder -= len;
363 if (remainder != 0 || more)
364 flags |= MSG_MORE;
a6b31d18 365 err = do_sendpage(sock, *ppage, base, len, flags);
24c5684b
TM		 366 if (remainder == 0 || err != len)
367 break;
f279cd00 368 *sent_p += err;
24c5684b
TM		 369 ppage++;
370 base = 0;
371 }
f279cd00
JB		 372 if (err > 0) {
373 *sent_p += err;
374 err = 0;
375 }
376 return err;
b4b5cc85
CL		 377}
378
9903cd1c
CL		 379/**
380 * xs_sendpages - write pages directly to a socket
381 * @sock: socket to send on
382 * @addr: UDP only -- address of destination
383 * @addrlen: UDP only -- length of destination address
384 * @xdr: buffer containing this request
385 * @base: starting position in the buffer
a6b31d18 386 * @zerocopy: true if it is safe to use sendpage()
f279cd00 387 * @sent_p: return the total number of bytes successfully queued for sending
9903cd1c 388 *
a246b010 389 */
f279cd00 390static int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base, bool zerocopy, int *sent_p)
a246b010 391{
24c5684b 392 unsigned int remainder = xdr->len - base;
f279cd00
JB		 393 int err = 0;
394 int sent = 0;
a246b010 395
262965f5 396 if (unlikely(!sock))
fba91afb 397 return -ENOTSOCK;
262965f5
CL		 398
399 clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags);
24c5684b
TM		 400 if (base != 0) {
401 addr = NULL;
402 addrlen = 0;
403 }
262965f5 404
24c5684b
TM		 405 if (base < xdr->head[0].iov_len || addr != NULL) {
406 unsigned int len = xdr->head[0].iov_len - base;
407 remainder -= len;
408 err = xs_send_kvec(sock, addr, addrlen, &xdr->head[0], base, remainder != 0);
409 if (remainder == 0 || err != len)
a246b010 410 goto out;
f279cd00 411 *sent_p += err;
a246b010
CL		 412 base = 0;
413 } else
24c5684b 414 base -= xdr->head[0].iov_len;
a246b010 415
24c5684b
TM		 416 if (base < xdr->page_len) {
417 unsigned int len = xdr->page_len - base;
418 remainder -= len;
f279cd00
JB		 419 err = xs_send_pagedata(sock, xdr, base, remainder != 0, zerocopy, &sent);
420 *sent_p += sent;
421 if (remainder == 0 || sent != len)
a246b010
CL		 422 goto out;
423 base = 0;
24c5684b
TM		 424 } else
425 base -= xdr->page_len;
426
427 if (base >= xdr->tail[0].iov_len)
f279cd00 428 return 0;
24c5684b 429 err = xs_send_kvec(sock, NULL, 0, &xdr->tail[0], base, 0);
a246b010 430out:
f279cd00
JB		 431 if (err > 0) {
432 *sent_p += err;
433 err = 0;
434 }
435 return err;
a246b010
CL		 436}
437
b6ddf64f
TM		 438static void xs_nospace_callback(struct rpc_task *task)
439{
440 struct sock_xprt *transport = container_of(task->tk_rqstp->rq_xprt, struct sock_xprt, xprt);
441
442 transport->inet->sk_write_pending--;
443 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
444}
445
9903cd1c 446/**
262965f5
CL		 447 * xs_nospace - place task on wait queue if transmit was incomplete
448 * @task: task to put to sleep
9903cd1c 449 *
a246b010 450 */
5e3771ce 451static int xs_nospace(struct rpc_task *task)
a246b010 452{
262965f5
CL		 453 struct rpc_rqst *req = task->tk_rqstp;
454 struct rpc_xprt *xprt = req->rq_xprt;
ee0ac0c2 455 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
06ea0bfe 456 struct sock *sk = transport->inet;
24ca9a84 457 int ret = -EAGAIN;
a246b010 458
46121cf7 459 dprintk("RPC: %5u xmit incomplete (%u left of %u)\n",
262965f5
CL		 460 task->tk_pid, req->rq_slen - req->rq_bytes_sent,
461 req->rq_slen);
462
b6ddf64f
TM		 463 /* Protect against races with write_space */
464 spin_lock_bh(&xprt->transport_lock);
465
466 /* Don't race with disconnect */
467 if (xprt_connected(xprt)) {
468 if (test_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags)) {
469 /*
470 * Notify TCP that we're limited by the application
471 * window size
472 */
473 set_bit(SOCK_NOSPACE, &transport->sock->flags);
06ea0bfe 474 sk->sk_write_pending++;
b6ddf64f
TM		 475 /* ...and wait for more buffer space */
476 xprt_wait_for_buffer_space(task, xs_nospace_callback);
477 }
478 } else {
479 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
5e3771ce 480 ret = -ENOTCONN;
b6ddf64f 481 }
262965f5 482
b6ddf64f 483 spin_unlock_bh(&xprt->transport_lock);
06ea0bfe
TM		 484
485 /* Race breaker in case memory is freed before above code is called */
486 sk->sk_write_space(sk);
5e3771ce 487 return ret;
262965f5
CL		 488}
489
61677eee
CL		 490/*
491 * Construct a stream transport record marker in @buf.
492 */
493static inline void xs_encode_stream_record_marker(struct xdr_buf *buf)
494{
495 u32 reclen = buf->len - sizeof(rpc_fraghdr);
496 rpc_fraghdr *base = buf->head[0].iov_base;
497 *base = cpu_to_be32(RPC_LAST_STREAM_FRAGMENT | reclen);
498}
499
176e21ee
CL		 500/**
501 * xs_local_send_request - write an RPC request to an AF_LOCAL socket
502 * @task: RPC task that manages the state of an RPC request
503 *
504 * Return values:
505 * 0: The request has been sent
506 * EAGAIN: The socket was blocked, please call again later to
507 * complete the request
508 * ENOTCONN: Caller needs to invoke connect logic then call again
509 * other: Some other error occured, the request was not sent
510 */
511static int xs_local_send_request(struct rpc_task *task)
512{
513 struct rpc_rqst *req = task->tk_rqstp;
514 struct rpc_xprt *xprt = req->rq_xprt;
515 struct sock_xprt *transport =
516 container_of(xprt, struct sock_xprt, xprt);
517 struct xdr_buf *xdr = &req->rq_snd_buf;
518 int status;
f279cd00 519 int sent = 0;
176e21ee
CL		 520
521 xs_encode_stream_record_marker(&req->rq_snd_buf);
522
523 xs_pktdump("packet data:",
524 req->rq_svec->iov_base, req->rq_svec->iov_len);
525
f279cd00
JB		 526 status = xs_sendpages(transport->sock, NULL, 0, xdr, req->rq_bytes_sent,
527 true, &sent);
176e21ee
CL		 528 dprintk("RPC: %s(%u) = %d\n",
529 __func__, xdr->len - req->rq_bytes_sent, status);
743c69e7
N		 530
531 if (status == -EAGAIN && sock_writeable(transport->inet))
532 status = -ENOBUFS;
533
f279cd00
JB		 534 if (likely(sent > 0) || status == 0) {
535 req->rq_bytes_sent += sent;
536 req->rq_xmit_bytes_sent += sent;
176e21ee
CL		 537 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
538 req->rq_bytes_sent = 0;
539 return 0;
540 }
541 status = -EAGAIN;
542 }
543
544 switch (status) {
3601c4a9 545 case -ENOBUFS:
b5872f0c 546 break;
176e21ee
CL		 547 case -EAGAIN:
548 status = xs_nospace(task);
549 break;
550 default:
551 dprintk("RPC: sendmsg returned unrecognized error %d\n",
552 -status);
553 case -EPIPE:
554 xs_close(xprt);
555 status = -ENOTCONN;
556 }
557
558 return status;
559}
560
262965f5
CL		 561/**
562 * xs_udp_send_request - write an RPC request to a UDP socket
563 * @task: address of RPC task that manages the state of an RPC request
564 *
565 * Return values:
566 * 0: The request has been sent
567 * EAGAIN: The socket was blocked, please call again later to
568 * complete the request
569 * ENOTCONN: Caller needs to invoke connect logic then call again
25985edc 570 * other: Some other error occurred, the request was not sent
262965f5
CL		 571 */
572static int xs_udp_send_request(struct rpc_task *task)
573{
574 struct rpc_rqst *req = task->tk_rqstp;
575 struct rpc_xprt *xprt = req->rq_xprt;
ee0ac0c2 576 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
262965f5 577 struct xdr_buf *xdr = &req->rq_snd_buf;
f279cd00 578 int sent = 0;
262965f5 579 int status;
a246b010 580
9903cd1c 581 xs_pktdump("packet data:",
a246b010
CL		 582 req->rq_svec->iov_base,
583 req->rq_svec->iov_len);
584
01d37c42
TM		 585 if (!xprt_bound(xprt))
586 return -ENOTCONN;
f279cd00
JB		 587 status = xs_sendpages(transport->sock, xs_addr(xprt), xprt->addrlen,
588 xdr, req->rq_bytes_sent, true, &sent);
a246b010 589
46121cf7 590 dprintk("RPC: xs_udp_send_request(%u) = %d\n",
262965f5 591 xdr->len - req->rq_bytes_sent, status);
a246b010 592
3dedbb5c
JB		 593 /* firewall is blocking us, don't return -EAGAIN or we end up looping */
594 if (status == -EPERM)
595 goto process_status;
596
743c69e7
N		 597 if (status == -EAGAIN && sock_writeable(transport->inet))
598 status = -ENOBUFS;
599
f279cd00
JB		 600 if (sent > 0 || status == 0) {
601 req->rq_xmit_bytes_sent += sent;
602 if (sent >= req->rq_slen)
2199700f
TM		 603 return 0;
604 /* Still some bytes left; set up for a retry later. */
262965f5 605 status = -EAGAIN;
2199700f 606 }
a246b010 607
3dedbb5c 608process_status:
262965f5 609 switch (status) {
fba91afb
TM		 610 case -ENOTSOCK:
611 status = -ENOTCONN;
612 /* Should we call xs_close() here? */
613 break;
b6ddf64f 614 case -EAGAIN:
5e3771ce 615 status = xs_nospace(task);
b6ddf64f 616 break;
c8485e4d
TM		 617 default:
618 dprintk("RPC: sendmsg returned unrecognized error %d\n",
619 -status);
262965f5 620 case -ENETUNREACH:
3601c4a9 621 case -ENOBUFS:
262965f5 622 case -EPIPE:
a246b010 623 case -ECONNREFUSED:
3dedbb5c 624 case -EPERM:
a246b010 625 /* When the server has died, an ICMP port unreachable message
9903cd1c 626 * prompts ECONNREFUSED. */
b6ddf64f 627 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
a246b010 628 }
5fe46e9d 629
262965f5 630 return status;
a246b010
CL		 631}
632
9903cd1c 633/**
262965f5 634 * xs_tcp_send_request - write an RPC request to a TCP socket
9903cd1c
CL		 635 * @task: address of RPC task that manages the state of an RPC request
636 *
637 * Return values:
262965f5
CL		 638 * 0: The request has been sent
639 * EAGAIN: The socket was blocked, please call again later to
640 * complete the request
641 * ENOTCONN: Caller needs to invoke connect logic then call again
25985edc 642 * other: Some other error occurred, the request was not sent
9903cd1c
CL		 643 *
644 * XXX: In the case of soft timeouts, should we eventually give up
262965f5 645 * if sendmsg is not able to make progress?
9903cd1c 646 */
262965f5 647static int xs_tcp_send_request(struct rpc_task *task)
a246b010
CL		 648{
649 struct rpc_rqst *req = task->tk_rqstp;
650 struct rpc_xprt *xprt = req->rq_xprt;
ee0ac0c2 651 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
262965f5 652 struct xdr_buf *xdr = &req->rq_snd_buf;
a6b31d18 653 bool zerocopy = true;
b595bb15 654 int status;
f279cd00 655 int sent;
a246b010 656
61677eee 657 xs_encode_stream_record_marker(&req->rq_snd_buf);
a246b010 658
262965f5
CL		 659 xs_pktdump("packet data:",
660 req->rq_svec->iov_base,
661 req->rq_svec->iov_len);
a6b31d18
TM		 662 /* Don't use zero copy if this is a resend. If the RPC call
663 * completes while the socket holds a reference to the pages,
664 * then we may end up resending corrupted data.
665 */
666 if (task->tk_flags & RPC_TASK_SENT)
667 zerocopy = false;
a246b010
CL		 668
669 /* Continue transmitting the packet/record. We must be careful
670 * to cope with writespace callbacks arriving _after_ we have
262965f5 671 * called sendmsg(). */
a246b010 672 while (1) {
f279cd00
JB		 673 sent = 0;
674 status = xs_sendpages(transport->sock, NULL, 0, xdr,
675 req->rq_bytes_sent, zerocopy, &sent);
a246b010 676
46121cf7 677 dprintk("RPC: xs_tcp_send_request(%u) = %d\n",
262965f5 678 xdr->len - req->rq_bytes_sent, status);
a246b010 679
262965f5
CL		 680 /* If we've sent the entire packet, immediately
681 * reset the count of bytes sent. */
f279cd00
JB		 682 req->rq_bytes_sent += sent;
683 req->rq_xmit_bytes_sent += sent;
262965f5
CL		 684 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
685 req->rq_bytes_sent = 0;
686 return 0;
687 }
a246b010 688
f580dd04
TM		 689 if (status < 0)
690 break;
691 if (sent == 0) {
692 status = -EAGAIN;
693 break;
694 }
a246b010 695 }
743c69e7
N		 696 if (status == -EAGAIN && sk_stream_is_writeable(transport->inet))
697 status = -ENOBUFS;
a246b010 698
262965f5 699 switch (status) {
fba91afb
TM		 700 case -ENOTSOCK:
701 status = -ENOTCONN;
702 /* Should we call xs_close() here? */
703 break;
262965f5 704 case -EAGAIN:
5e3771ce 705 status = xs_nospace(task);
262965f5 706 break;
c8485e4d
TM		 707 default:
708 dprintk("RPC: sendmsg returned unrecognized error %d\n",
709 -status);
262965f5 710 case -ECONNRESET:
2a9e1cfa 711 case -ECONNREFUSED:
262965f5 712 case -ENOTCONN:
3913c78c 713 case -EADDRINUSE:
b5872f0c 714 case -ENOBUFS:
b9d2bb2e 715 case -EPIPE:
b6ddf64f 716 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
a246b010 717 }
5fe46e9d 718
a246b010
CL		 719 return status;
720}
721
e0ab53de
TM		 722/**
723 * xs_tcp_release_xprt - clean up after a tcp transmission
724 * @xprt: transport
725 * @task: rpc task
726 *
727 * This cleans up if an error causes us to abort the transmission of a request.
728 * In this case, the socket may need to be reset in order to avoid confusing
729 * the server.
730 */
731static void xs_tcp_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
732{
733 struct rpc_rqst *req;
734
735 if (task != xprt->snd_task)
736 return;
737 if (task == NULL)
738 goto out_release;
739 req = task->tk_rqstp;
43cedbf0
TM		 740 if (req == NULL)
741 goto out_release;
e0ab53de
TM		 742 if (req->rq_bytes_sent == 0)
743 goto out_release;
744 if (req->rq_bytes_sent == req->rq_snd_buf.len)
745 goto out_release;
a4f0835c 746 set_bit(XPRT_CLOSE_WAIT, &xprt->state);
e0ab53de
TM		 747out_release:
748 xprt_release_xprt(xprt, task);
749}
750
2a9e1cfa
TM		 751static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk)
752{
753 transport->old_data_ready = sk->sk_data_ready;
754 transport->old_state_change = sk->sk_state_change;
755 transport->old_write_space = sk->sk_write_space;
2118071d 756 transport->old_error_report = sk->sk_error_report;
2a9e1cfa
TM		 757}
758
759static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk)
760{
761 sk->sk_data_ready = transport->old_data_ready;
762 sk->sk_state_change = transport->old_state_change;
763 sk->sk_write_space = transport->old_write_space;
2118071d
TM		 764 sk->sk_error_report = transport->old_error_report;
765}
766
b70ae915
TM		 767static void xs_sock_reset_connection_flags(struct rpc_xprt *xprt)
768{
769 smp_mb__before_atomic();
770 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
771 clear_bit(XPRT_CLOSING, &xprt->state);
772 smp_mb__after_atomic();
773}
774
775static void xs_sock_mark_closed(struct rpc_xprt *xprt)
776{
777 xs_sock_reset_connection_flags(xprt);
778 /* Mark transport as closed and wake up all pending tasks */
779 xprt_disconnect_done(xprt);
780}
781
2118071d
TM		 782/**
783 * xs_error_report - callback to handle TCP socket state errors
784 * @sk: socket
785 *
786 * Note: we don't call sock_error() since there may be a rpc_task
787 * using the socket, and so we don't want to clear sk->sk_err.
788 */
789static void xs_error_report(struct sock *sk)
790{
791 struct rpc_xprt *xprt;
792 int err;
793
794 read_lock_bh(&sk->sk_callback_lock);
795 if (!(xprt = xprt_from_sock(sk)))
796 goto out;
797
798 err = -sk->sk_err;
799 if (err == 0)
800 goto out;
b70ae915
TM		 801 /* Is this a reset event? */
802 if (sk->sk_state == TCP_CLOSE)
803 xs_sock_mark_closed(xprt);
2118071d
TM		 804 dprintk("RPC: xs_error_report client %p, error=%d...\n",
805 xprt, -err);
e8353c76 806 trace_rpc_socket_error(xprt, sk->sk_socket, err);
2118071d
TM		 807 xprt_wake_pending_tasks(xprt, err);
808 out:
809 read_unlock_bh(&sk->sk_callback_lock);
2a9e1cfa
TM		 810}
811
fe315e76 812static void xs_reset_transport(struct sock_xprt *transport)
a246b010 813{
ee0ac0c2
CL		 814 struct socket *sock = transport->sock;
815 struct sock *sk = transport->inet;
6cc7e908 816 struct rpc_xprt *xprt = &transport->xprt;
a246b010 817
fe315e76
CL		 818 if (sk == NULL)
819 return;
9903cd1c 820
264d1df3
JL		 821 if (atomic_read(&transport->xprt.swapper))
822 sk_clear_memalloc(sk);
823
09939204
TM		 824 kernel_sock_shutdown(sock, SHUT_RDWR);
825
edc1b01c 826 mutex_lock(&transport->recv_mutex);
a246b010 827 write_lock_bh(&sk->sk_callback_lock);
ee0ac0c2
CL		 828 transport->inet = NULL;
829 transport->sock = NULL;
a246b010 830
9903cd1c 831 sk->sk_user_data = NULL;
2a9e1cfa
TM		 832
833 xs_restore_old_callbacks(transport, sk);
0c78789e 834 xprt_clear_connected(xprt);
a246b010 835 write_unlock_bh(&sk->sk_callback_lock);
6cc7e908 836 xs_sock_reset_connection_flags(xprt);
edc1b01c 837 mutex_unlock(&transport->recv_mutex);
a246b010 838
6cc7e908 839 trace_rpc_socket_close(xprt, sock);
a246b010 840 sock_release(sock);
fe315e76
CL		 841}
842
843/**
844 * xs_close - close a socket
845 * @xprt: transport
846 *
847 * This is used when all requests are complete; ie, no DRC state remains
848 * on the server we want to save.
f75e6745
TM		 849 *
850 * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with
851 * xs_reset_transport() zeroing the socket from underneath a writer.
fe315e76
CL		 852 */
853static void xs_close(struct rpc_xprt *xprt)
854{
855 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
856
857 dprintk("RPC: xs_close xprt %p\n", xprt);
858
859 xs_reset_transport(transport);
61d0a8e6 860 xprt->reestablish_timeout = 0;
fe315e76 861
62da3b24 862 xprt_disconnect_done(xprt);
a246b010
CL		 863}
864
4a068258
CL		 865static void xs_inject_disconnect(struct rpc_xprt *xprt)
866{
867 dprintk("RPC: injecting transport disconnect on xprt=%p\n",
868 xprt);
869 xprt_disconnect_done(xprt);
870}
871
315f3812
KM		 872static void xs_xprt_free(struct rpc_xprt *xprt)
873{
874 xs_free_peer_addresses(xprt);
875 xprt_free(xprt);
876}
877
9903cd1c
CL		 878/**
879 * xs_destroy - prepare to shutdown a transport
880 * @xprt: doomed transport
881 *
882 */
883static void xs_destroy(struct rpc_xprt *xprt)
a246b010 884{
03c78827
TM		 885 struct sock_xprt *transport = container_of(xprt,
886 struct sock_xprt, xprt);
46121cf7 887 dprintk("RPC: xs_destroy xprt %p\n", xprt);
9903cd1c 888
03c78827 889 cancel_delayed_work_sync(&transport->connect_worker);
a1311d87 890 xs_close(xprt);
edc1b01c 891 cancel_work_sync(&transport->recv_worker);
315f3812 892 xs_xprt_free(xprt);
a1311d87 893 module_put(THIS_MODULE);
a246b010
CL		 894}
895
176e21ee
CL		 896static int xs_local_copy_to_xdr(struct xdr_buf *xdr, struct sk_buff *skb)
897{
898 struct xdr_skb_reader desc = {
899 .skb = skb,
900 .offset = sizeof(rpc_fraghdr),
901 .count = skb->len - sizeof(rpc_fraghdr),
902 };
903
904 if (xdr_partial_copy_from_skb(xdr, 0, &desc, xdr_skb_read_bits) < 0)
905 return -1;
906 if (desc.count)
907 return -1;
908 return 0;
909}
910
911/**
912 * xs_local_data_ready - "data ready" callback for AF_LOCAL sockets
913 * @sk: socket with data to read
176e21ee
CL		 914 *
915 * Currently this assumes we can read the whole reply in a single gulp.
916 */
676d2369 917static void xs_local_data_ready(struct sock *sk)
176e21ee
CL		 918{
919 struct rpc_task *task;
920 struct rpc_xprt *xprt;
921 struct rpc_rqst *rovr;
922 struct sk_buff *skb;
923 int err, repsize, copied;
924 u32 _xid;
925 __be32 *xp;
926
927 read_lock_bh(&sk->sk_callback_lock);
928 dprintk("RPC: %s...\n", __func__);
929 xprt = xprt_from_sock(sk);
930 if (xprt == NULL)
931 goto out;
932
933 skb = skb_recv_datagram(sk, 0, 1, &err);
934 if (skb == NULL)
935 goto out;
936
176e21ee
CL		 937 repsize = skb->len - sizeof(rpc_fraghdr);
938 if (repsize < 4) {
939 dprintk("RPC: impossible RPC reply size %d\n", repsize);
940 goto dropit;
941 }
942
943 /* Copy the XID from the skb... */
944 xp = skb_header_pointer(skb, sizeof(rpc_fraghdr), sizeof(_xid), &_xid);
945 if (xp == NULL)
946 goto dropit;
947
948 /* Look up and lock the request corresponding to the given XID */
949 spin_lock(&xprt->transport_lock);
950 rovr = xprt_lookup_rqst(xprt, *xp);
951 if (!rovr)
952 goto out_unlock;
953 task = rovr->rq_task;
954
955 copied = rovr->rq_private_buf.buflen;
956 if (copied > repsize)
957 copied = repsize;
958
959 if (xs_local_copy_to_xdr(&rovr->rq_private_buf, skb)) {
960 dprintk("RPC: sk_buff copy failed\n");
961 goto out_unlock;
962 }
963
964 xprt_complete_rqst(task, copied);
965
966 out_unlock:
967 spin_unlock(&xprt->transport_lock);
968 dropit:
969 skb_free_datagram(sk, skb);
970 out:
971 read_unlock_bh(&sk->sk_callback_lock);
972}
973
9903cd1c 974/**
f9b2ee71
TM		 975 * xs_udp_data_read_skb - receive callback for UDP sockets
976 * @xprt: transport
977 * @sk: socket
978 * @skb: skbuff
9903cd1c 979 *
a246b010 980 */
f9b2ee71
TM		 981static void xs_udp_data_read_skb(struct rpc_xprt *xprt,
982 struct sock *sk,
983 struct sk_buff *skb)
a246b010 984{
9903cd1c 985 struct rpc_task *task;
a246b010 986 struct rpc_rqst *rovr;
f9b2ee71 987 int repsize, copied;
d8ed029d
AD		 988 u32 _xid;
989 __be32 *xp;
a246b010 990
a246b010
CL		 991 repsize = skb->len - sizeof(struct udphdr);
992 if (repsize < 4) {
46121cf7 993 dprintk("RPC: impossible RPC reply size %d!\n", repsize);
f9b2ee71 994 return;
a246b010
CL		 995 }
996
997 /* Copy the XID from the skb... */
998 xp = skb_header_pointer(skb, sizeof(struct udphdr),
999 sizeof(_xid), &_xid);
1000 if (xp == NULL)
f9b2ee71 1001 return;
a246b010
CL		 1002
1003 /* Look up and lock the request corresponding to the given XID */
f9b2ee71 1004 spin_lock_bh(&xprt->transport_lock);
a246b010
CL		 1005 rovr = xprt_lookup_rqst(xprt, *xp);
1006 if (!rovr)
1007 goto out_unlock;
1008 task = rovr->rq_task;
1009
a246b010
CL		 1010 if ((copied = rovr->rq_private_buf.buflen) > repsize)
1011 copied = repsize;
1012
1013 /* Suck it into the iovec, verify checksum if not done by hw. */
1781f7f5
HX		 1014 if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
1015 UDPX_INC_STATS_BH(sk, UDP_MIB_INERRORS);
a246b010 1016 goto out_unlock;
1781f7f5
HX		 1017 }
1018
1019 UDPX_INC_STATS_BH(sk, UDP_MIB_INDATAGRAMS);
a246b010 1020
6a24dfb6 1021 xprt_adjust_cwnd(xprt, task, copied);
1570c1e4 1022 xprt_complete_rqst(task, copied);
a246b010
CL		 1023
1024 out_unlock:
f9b2ee71
TM		 1025 spin_unlock_bh(&xprt->transport_lock);
1026}
1027
1028static void xs_udp_data_receive(struct sock_xprt *transport)
1029{
1030 struct sk_buff *skb;
1031 struct sock *sk;
1032 int err;
1033
1034 mutex_lock(&transport->recv_mutex);
1035 sk = transport->inet;
1036 if (sk == NULL)
1037 goto out;
1038 for (;;) {
1039 skb = skb_recv_datagram(sk, 0, 1, &err);
1040 if (skb == NULL)
1041 break;
1042 xs_udp_data_read_skb(&transport->xprt, sk, skb);
1043 skb_free_datagram(sk, skb);
1044 }
1045out:
1046 mutex_unlock(&transport->recv_mutex);
1047}
1048
1049static void xs_udp_data_receive_workfn(struct work_struct *work)
1050{
1051 struct sock_xprt *transport =
1052 container_of(work, struct sock_xprt, recv_worker);
1053 xs_udp_data_receive(transport);
1054}
1055
1056/**
1057 * xs_data_ready - "data ready" callback for UDP sockets
1058 * @sk: socket with data to read
1059 *
1060 */
1061static void xs_data_ready(struct sock *sk)
1062{
1063 struct rpc_xprt *xprt;
1064
1065 read_lock_bh(&sk->sk_callback_lock);
1066 dprintk("RPC: xs_data_ready...\n");
1067 xprt = xprt_from_sock(sk);
1068 if (xprt != NULL) {
1069 struct sock_xprt *transport = container_of(xprt,
1070 struct sock_xprt, xprt);
1071 queue_work(rpciod_workqueue, &transport->recv_worker);
1072 }
f064af1e 1073 read_unlock_bh(&sk->sk_callback_lock);
a246b010
CL		 1074}
1075
a519fc7a
TM		 1076/*
1077 * Helper function to force a TCP close if the server is sending
1078 * junk and/or it has put us in CLOSE_WAIT
1079 */
1080static void xs_tcp_force_close(struct rpc_xprt *xprt)
1081{
a519fc7a
TM		 1082 xprt_force_disconnect(xprt);
1083}
1084
dd456471 1085static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
a246b010 1086{
51971139 1087 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
a246b010
CL		 1088 size_t len, used;
1089 char *p;
1090
51971139
CL		 1091 p = ((char *) &transport->tcp_fraghdr) + transport->tcp_offset;
1092 len = sizeof(transport->tcp_fraghdr) - transport->tcp_offset;
9d292316 1093 used = xdr_skb_read_bits(desc, p, len);
51971139 1094 transport->tcp_offset += used;
a246b010
CL		 1095 if (used != len)
1096 return;
808012fb 1097
51971139
CL		 1098 transport->tcp_reclen = ntohl(transport->tcp_fraghdr);
1099 if (transport->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
e136d092 1100 transport->tcp_flags |= TCP_RCV_LAST_FRAG;
a246b010 1101 else
e136d092 1102 transport->tcp_flags &= ~TCP_RCV_LAST_FRAG;
51971139 1103 transport->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
808012fb 1104
e136d092 1105 transport->tcp_flags &= ~TCP_RCV_COPY_FRAGHDR;
51971139 1106 transport->tcp_offset = 0;
808012fb 1107
a246b010 1108 /* Sanity check of the record length */
18dca02a 1109 if (unlikely(transport->tcp_reclen < 8)) {
46121cf7 1110 dprintk("RPC: invalid TCP record fragment length\n");
a519fc7a 1111 xs_tcp_force_close(xprt);
9903cd1c 1112 return;
a246b010 1113 }
46121cf7 1114 dprintk("RPC: reading TCP record fragment of length %d\n",
51971139 1115 transport->tcp_reclen);
a246b010
CL		 1116}
1117
51971139 1118static void xs_tcp_check_fraghdr(struct sock_xprt *transport)
a246b010 1119{
51971139 1120 if (transport->tcp_offset == transport->tcp_reclen) {
e136d092 1121 transport->tcp_flags |= TCP_RCV_COPY_FRAGHDR;
51971139 1122 transport->tcp_offset = 0;
e136d092
CL		 1123 if (transport->tcp_flags & TCP_RCV_LAST_FRAG) {
1124 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1125 transport->tcp_flags |= TCP_RCV_COPY_XID;
51971139 1126 transport->tcp_copied = 0;
a246b010
CL		 1127 }
1128 }
1129}
1130
dd456471 1131static inline void xs_tcp_read_xid(struct sock_xprt *transport, struct xdr_skb_reader *desc)
a246b010
CL		 1132{
1133 size_t len, used;
1134 char *p;
1135
51971139 1136 len = sizeof(transport->tcp_xid) - transport->tcp_offset;
46121cf7 1137 dprintk("RPC: reading XID (%Zu bytes)\n", len);
51971139 1138 p = ((char *) &transport->tcp_xid) + transport->tcp_offset;
9d292316 1139 used = xdr_skb_read_bits(desc, p, len);
51971139 1140 transport->tcp_offset += used;
a246b010
CL		 1141 if (used != len)
1142 return;
e136d092 1143 transport->tcp_flags &= ~TCP_RCV_COPY_XID;
f4a2e418 1144 transport->tcp_flags |= TCP_RCV_READ_CALLDIR;
51971139 1145 transport->tcp_copied = 4;
18dca02a
RL		 1146 dprintk("RPC: reading %s XID %08x\n",
1147 (transport->tcp_flags & TCP_RPC_REPLY) ? "reply for"
1148 : "request with",
51971139
CL		 1149 ntohl(transport->tcp_xid));
1150 xs_tcp_check_fraghdr(transport);
a246b010
CL		 1151}
1152
18dca02a
RL		 1153static inline void xs_tcp_read_calldir(struct sock_xprt *transport,
1154 struct xdr_skb_reader *desc)
a246b010 1155{
18dca02a
RL		 1156 size_t len, used;
1157 u32 offset;
b76ce561 1158 char *p;
18dca02a
RL		 1159
1160 /*
1161 * We want transport->tcp_offset to be 8 at the end of this routine
1162 * (4 bytes for the xid and 4 bytes for the call/reply flag).
1163 * When this function is called for the first time,
1164 * transport->tcp_offset is 4 (after having already read the xid).
1165 */
1166 offset = transport->tcp_offset - sizeof(transport->tcp_xid);
b76ce561 1167 len = sizeof(transport->tcp_calldir) - offset;
18dca02a 1168 dprintk("RPC: reading CALL/REPLY flag (%Zu bytes)\n", len);
b76ce561
TM		 1169 p = ((char *) &transport->tcp_calldir) + offset;
1170 used = xdr_skb_read_bits(desc, p, len);
18dca02a
RL		 1171 transport->tcp_offset += used;
1172 if (used != len)
1173 return;
f4a2e418 1174 transport->tcp_flags &= ~TCP_RCV_READ_CALLDIR;
f4a2e418
RL		 1175 /*
1176 * We don't yet have the XDR buffer, so we will write the calldir
1177 * out after we get the buffer from the 'struct rpc_rqst'
1178 */
b76ce561
TM		 1179 switch (ntohl(transport->tcp_calldir)) {
1180 case RPC_REPLY:
1181 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1182 transport->tcp_flags |= TCP_RCV_COPY_DATA;
18dca02a 1183 transport->tcp_flags |= TCP_RPC_REPLY;
b76ce561
TM		 1184 break;
1185 case RPC_CALL:
1186 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1187 transport->tcp_flags |= TCP_RCV_COPY_DATA;
18dca02a 1188 transport->tcp_flags &= ~TCP_RPC_REPLY;
b76ce561
TM		 1189 break;
1190 default:
1191 dprintk("RPC: invalid request message type\n");
a519fc7a 1192 xs_tcp_force_close(&transport->xprt);
b76ce561 1193 }
18dca02a
RL		 1194 xs_tcp_check_fraghdr(transport);
1195}
1196
44b98efd
RL		 1197static inline void xs_tcp_read_common(struct rpc_xprt *xprt,
1198 struct xdr_skb_reader *desc,
1199 struct rpc_rqst *req)
a246b010 1200{
44b98efd
RL		 1201 struct sock_xprt *transport =
1202 container_of(xprt, struct sock_xprt, xprt);
a246b010
CL		 1203 struct xdr_buf *rcvbuf;
1204 size_t len;
1205 ssize_t r;
1206
a246b010 1207 rcvbuf = &req->rq_private_buf;
f4a2e418
RL		 1208
1209 if (transport->tcp_flags & TCP_RCV_COPY_CALLDIR) {
1210 /*
1211 * Save the RPC direction in the XDR buffer
1212 */
f4a2e418 1213 memcpy(rcvbuf->head[0].iov_base + transport->tcp_copied,
b76ce561
TM		 1214 &transport->tcp_calldir,
1215 sizeof(transport->tcp_calldir));
1216 transport->tcp_copied += sizeof(transport->tcp_calldir);
f4a2e418 1217 transport->tcp_flags &= ~TCP_RCV_COPY_CALLDIR;
a246b010
CL		 1218 }
1219
a246b010 1220 len = desc->count;
51971139 1221 if (len > transport->tcp_reclen - transport->tcp_offset) {
dd456471 1222 struct xdr_skb_reader my_desc;
a246b010 1223
51971139 1224 len = transport->tcp_reclen - transport->tcp_offset;
a246b010
CL		 1225 memcpy(&my_desc, desc, sizeof(my_desc));
1226 my_desc.count = len;
51971139 1227 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
9d292316 1228 &my_desc, xdr_skb_read_bits);
a246b010
CL		 1229 desc->count -= r;
1230 desc->offset += r;
1231 } else
51971139 1232 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
9d292316 1233 desc, xdr_skb_read_bits);
a246b010
CL		 1234
1235 if (r > 0) {
51971139
CL		 1236 transport->tcp_copied += r;
1237 transport->tcp_offset += r;
a246b010
CL		 1238 }
1239 if (r != len) {
1240 /* Error when copying to the receive buffer,
1241 * usually because we weren't able to allocate
1242 * additional buffer pages. All we can do now
e136d092 1243 * is turn off TCP_RCV_COPY_DATA, so the request
a246b010
CL		 1244 * will not receive any additional updates,
1245 * and time out.
1246 * Any remaining data from this record will
1247 * be discarded.
1248 */
e136d092 1249 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
46121cf7 1250 dprintk("RPC: XID %08x truncated request\n",
51971139 1251 ntohl(transport->tcp_xid));
46121cf7
CL		 1252 dprintk("RPC: xprt = %p, tcp_copied = %lu, "
1253 "tcp_offset = %u, tcp_reclen = %u\n",
1254 xprt, transport->tcp_copied,
1255 transport->tcp_offset, transport->tcp_reclen);
44b98efd 1256 return;
a246b010
CL		 1257 }
1258
46121cf7 1259 dprintk("RPC: XID %08x read %Zd bytes\n",
51971139 1260 ntohl(transport->tcp_xid), r);
46121cf7
CL		 1261 dprintk("RPC: xprt = %p, tcp_copied = %lu, tcp_offset = %u, "
1262 "tcp_reclen = %u\n", xprt, transport->tcp_copied,
1263 transport->tcp_offset, transport->tcp_reclen);
51971139
CL		 1264
1265 if (transport->tcp_copied == req->rq_private_buf.buflen)
e136d092 1266 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
51971139 1267 else if (transport->tcp_offset == transport->tcp_reclen) {
e136d092
CL		 1268 if (transport->tcp_flags & TCP_RCV_LAST_FRAG)
1269 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
a246b010 1270 }
44b98efd
RL		 1271}
1272
1273/*
1274 * Finds the request corresponding to the RPC xid and invokes the common
1275 * tcp read code to read the data.
1276 */
1277static inline int xs_tcp_read_reply(struct rpc_xprt *xprt,
1278 struct xdr_skb_reader *desc)
1279{
1280 struct sock_xprt *transport =
1281 container_of(xprt, struct sock_xprt, xprt);
1282 struct rpc_rqst *req;
1283
1284 dprintk("RPC: read reply XID %08x\n", ntohl(transport->tcp_xid));
1285
1286 /* Find and lock the request corresponding to this xid */
edc1b01c 1287 spin_lock_bh(&xprt->transport_lock);
44b98efd
RL		 1288 req = xprt_lookup_rqst(xprt, transport->tcp_xid);
1289 if (!req) {
1290 dprintk("RPC: XID %08x request not found!\n",
1291 ntohl(transport->tcp_xid));
edc1b01c 1292 spin_unlock_bh(&xprt->transport_lock);
44b98efd
RL		 1293 return -1;
1294 }
1295
1296 xs_tcp_read_common(xprt, desc, req);
1297
e136d092 1298 if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
51971139 1299 xprt_complete_rqst(req->rq_task, transport->tcp_copied);
44b98efd 1300
edc1b01c 1301 spin_unlock_bh(&xprt->transport_lock);
44b98efd
RL		 1302 return 0;
1303}
1304
9e00abc3 1305#if defined(CONFIG_SUNRPC_BACKCHANNEL)
44b98efd
RL		 1306/*
1307 * Obtains an rpc_rqst previously allocated and invokes the common
1308 * tcp read code to read the data. The result is placed in the callback
1309 * queue.
1310 * If we're unable to obtain the rpc_rqst we schedule the closing of the
1311 * connection and return -1.
1312 */
2ea24497 1313static int xs_tcp_read_callback(struct rpc_xprt *xprt,
44b98efd
RL		 1314 struct xdr_skb_reader *desc)
1315{
1316 struct sock_xprt *transport =
1317 container_of(xprt, struct sock_xprt, xprt);
1318 struct rpc_rqst *req;
1319
2ea24497 1320 /* Look up and lock the request corresponding to the given XID */
edc1b01c 1321 spin_lock_bh(&xprt->transport_lock);
2ea24497 1322 req = xprt_lookup_bc_request(xprt, transport->tcp_xid);
44b98efd 1323 if (req == NULL) {
edc1b01c 1324 spin_unlock_bh(&xprt->transport_lock);
44b98efd
RL		 1325 printk(KERN_WARNING "Callback slot table overflowed\n");
1326 xprt_force_disconnect(xprt);
1327 return -1;
1328 }
1329
44b98efd
RL		 1330 dprintk("RPC: read callback XID %08x\n", ntohl(req->rq_xid));
1331 xs_tcp_read_common(xprt, desc, req);
1332
2ea24497
TM		 1333 if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1334 xprt_complete_bc_request(req, transport->tcp_copied);
edc1b01c 1335 spin_unlock_bh(&xprt->transport_lock);
44b98efd
RL		 1336
1337 return 0;
1338}
1339
1340static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1341 struct xdr_skb_reader *desc)
1342{
1343 struct sock_xprt *transport =
1344 container_of(xprt, struct sock_xprt, xprt);
1345
1346 return (transport->tcp_flags & TCP_RPC_REPLY) ?
1347 xs_tcp_read_reply(xprt, desc) :
1348 xs_tcp_read_callback(xprt, desc);
1349}
1350#else
1351static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1352 struct xdr_skb_reader *desc)
1353{
1354 return xs_tcp_read_reply(xprt, desc);
1355}
9e00abc3 1356#endif /* CONFIG_SUNRPC_BACKCHANNEL */
44b98efd
RL		 1357
1358/*
1359 * Read data off the transport. This can be either an RPC_CALL or an
1360 * RPC_REPLY. Relay the processing to helper functions.
1361 */
1362static void xs_tcp_read_data(struct rpc_xprt *xprt,
1363 struct xdr_skb_reader *desc)
1364{
1365 struct sock_xprt *transport =
1366 container_of(xprt, struct sock_xprt, xprt);
1367
1368 if (_xs_tcp_read_data(xprt, desc) == 0)
1369 xs_tcp_check_fraghdr(transport);
1370 else {
1371 /*
1372 * The transport_lock protects the request handling.
1373 * There's no need to hold it to update the tcp_flags.
1374 */
1375 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1376 }
a246b010
CL		 1377}
1378
dd456471 1379static inline void xs_tcp_read_discard(struct sock_xprt *transport, struct xdr_skb_reader *desc)
a246b010
CL		 1380{
1381 size_t len;
1382
51971139 1383 len = transport->tcp_reclen - transport->tcp_offset;
a246b010
CL		 1384 if (len > desc->count)
1385 len = desc->count;
1386 desc->count -= len;
1387 desc->offset += len;
51971139 1388 transport->tcp_offset += len;
46121cf7 1389 dprintk("RPC: discarded %Zu bytes\n", len);
51971139 1390 xs_tcp_check_fraghdr(transport);
a246b010
CL		 1391}
1392
9903cd1c 1393static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
a246b010
CL		 1394{
1395 struct rpc_xprt *xprt = rd_desc->arg.data;
51971139 1396 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
dd456471 1397 struct xdr_skb_reader desc = {
a246b010
CL		 1398 .skb = skb,
1399 .offset = offset,
1400 .count = len,
9903cd1c 1401 };
a246b010 1402
46121cf7 1403 dprintk("RPC: xs_tcp_data_recv started\n");
a246b010 1404 do {
1a867a08 1405 trace_xs_tcp_data_recv(transport);
a246b010
CL		 1406 /* Read in a new fragment marker if necessary */
1407 /* Can we ever really expect to get completely empty fragments? */
e136d092 1408 if (transport->tcp_flags & TCP_RCV_COPY_FRAGHDR) {
9903cd1c 1409 xs_tcp_read_fraghdr(xprt, &desc);
a246b010
CL		 1410 continue;
1411 }
1412 /* Read in the xid if necessary */
e136d092 1413 if (transport->tcp_flags & TCP_RCV_COPY_XID) {
51971139 1414 xs_tcp_read_xid(transport, &desc);
a246b010
CL		 1415 continue;
1416 }
18dca02a 1417 /* Read in the call/reply flag */
f4a2e418 1418 if (transport->tcp_flags & TCP_RCV_READ_CALLDIR) {
18dca02a
RL		 1419 xs_tcp_read_calldir(transport, &desc);
1420 continue;
1421 }
a246b010 1422 /* Read in the request data */
e136d092 1423 if (transport->tcp_flags & TCP_RCV_COPY_DATA) {
44b98efd 1424 xs_tcp_read_data(xprt, &desc);
a246b010
CL		 1425 continue;
1426 }
1427 /* Skip over any trailing bytes on short reads */
51971139 1428 xs_tcp_read_discard(transport, &desc);
a246b010 1429 } while (desc.count);
1a867a08 1430 trace_xs_tcp_data_recv(transport);
46121cf7 1431 dprintk("RPC: xs_tcp_data_recv done\n");
a246b010
CL		 1432 return len - desc.count;
1433}
1434
66d7a56a
TM		 1435static void xs_tcp_data_receive(struct sock_xprt *transport)
1436{
1437 struct rpc_xprt *xprt = &transport->xprt;
1438 struct sock *sk;
1439 read_descriptor_t rd_desc = {
1440 .count = 2*1024*1024,
1441 .arg.data = xprt,
1442 };
1443 unsigned long total = 0;
1444 int read = 0;
1445
edc1b01c 1446 mutex_lock(&transport->recv_mutex);
66d7a56a 1447 sk = transport->inet;
edc1b01c
TM		 1448 if (sk == NULL)
1449 goto out;
66d7a56a
TM		 1450
1451 /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
1452 for (;;) {
edc1b01c 1453 lock_sock(sk);
66d7a56a 1454 read = tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
edc1b01c 1455 release_sock(sk);
66d7a56a
TM		 1456 if (read <= 0)
1457 break;
1458 total += read;
1459 rd_desc.count = 65536;
1460 }
edc1b01c
TM		 1461out:
1462 mutex_unlock(&transport->recv_mutex);
66d7a56a
TM		 1463 trace_xs_tcp_data_ready(xprt, read, total);
1464}
1465
edc1b01c
TM		 1466static void xs_tcp_data_receive_workfn(struct work_struct *work)
1467{
1468 struct sock_xprt *transport =
1469 container_of(work, struct sock_xprt, recv_worker);
1470 xs_tcp_data_receive(transport);
1471}
1472
9903cd1c
CL		 1473/**
1474 * xs_tcp_data_ready - "data ready" callback for TCP sockets
1475 * @sk: socket with data to read
9903cd1c
CL		 1476 *
1477 */
676d2369 1478static void xs_tcp_data_ready(struct sock *sk)
a246b010 1479{
66d7a56a 1480 struct sock_xprt *transport;
a246b010 1481 struct rpc_xprt *xprt;
a246b010 1482
46121cf7
CL		 1483 dprintk("RPC: xs_tcp_data_ready...\n");
1484
f064af1e 1485 read_lock_bh(&sk->sk_callback_lock);
66d7a56a 1486 if (!(xprt = xprt_from_sock(sk)))
a246b010 1487 goto out;
66d7a56a
TM		 1488 transport = container_of(xprt, struct sock_xprt, xprt);
1489
61d0a8e6
NB		 1490 /* Any data means we had a useful conversation, so
1491 * the we don't need to delay the next reconnect
1492 */
1493 if (xprt->reestablish_timeout)
1494 xprt->reestablish_timeout = 0;
edc1b01c 1495 queue_work(rpciod_workqueue, &transport->recv_worker);
61d0a8e6 1496
a246b010 1497out:
f064af1e 1498 read_unlock_bh(&sk->sk_callback_lock);
a246b010
CL		 1499}
1500
9903cd1c
CL		 1501/**
1502 * xs_tcp_state_change - callback to handle TCP socket state changes
1503 * @sk: socket whose state has changed
1504 *
1505 */
1506static void xs_tcp_state_change(struct sock *sk)
a246b010 1507{
9903cd1c 1508 struct rpc_xprt *xprt;
0fdea1e8 1509 struct sock_xprt *transport;
a246b010 1510
f064af1e 1511 read_lock_bh(&sk->sk_callback_lock);
a246b010
CL		 1512 if (!(xprt = xprt_from_sock(sk)))
1513 goto out;
46121cf7 1514 dprintk("RPC: xs_tcp_state_change client %p...\n", xprt);
669502ff 1515 dprintk("RPC: state %x conn %d dead %d zapped %d sk_shutdown %d\n",
46121cf7
CL		 1516 sk->sk_state, xprt_connected(xprt),
1517 sock_flag(sk, SOCK_DEAD),
669502ff
AC		 1518 sock_flag(sk, SOCK_ZAPPED),
1519 sk->sk_shutdown);
a246b010 1520
0fdea1e8 1521 transport = container_of(xprt, struct sock_xprt, xprt);
40b5ea0c 1522 trace_rpc_socket_state_change(xprt, sk->sk_socket);
a246b010
CL		 1523 switch (sk->sk_state) {
1524 case TCP_ESTABLISHED:
f064af1e 1525 spin_lock(&xprt->transport_lock);
a246b010 1526 if (!xprt_test_and_set_connected(xprt)) {
51971139 1527
a246b010 1528 /* Reset TCP record info */
51971139
CL		 1529 transport->tcp_offset = 0;
1530 transport->tcp_reclen = 0;
1531 transport->tcp_copied = 0;
e136d092
CL		 1532 transport->tcp_flags =
1533 TCP_RCV_COPY_FRAGHDR | TCP_RCV_COPY_XID;
8b71798c 1534 xprt->connect_cookie++;
0fdea1e8
TM		 1535 clear_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
1536 xprt_clear_connecting(xprt);
51971139 1537
2a491991 1538 xprt_wake_pending_tasks(xprt, -EAGAIN);
a246b010 1539 }
f064af1e 1540 spin_unlock(&xprt->transport_lock);
a246b010 1541 break;
3b948ae5
TM		 1542 case TCP_FIN_WAIT1:
1543 /* The client initiated a shutdown of the socket */
7c1d71cf 1544 xprt->connect_cookie++;
663b8858 1545 xprt->reestablish_timeout = 0;
3b948ae5 1546 set_bit(XPRT_CLOSING, &xprt->state);
4e857c58 1547 smp_mb__before_atomic();
3b948ae5 1548 clear_bit(XPRT_CONNECTED, &xprt->state);
ef803670 1549 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
4e857c58 1550 smp_mb__after_atomic();
a246b010 1551 break;
632e3bdc 1552 case TCP_CLOSE_WAIT:
3b948ae5 1553 /* The server initiated a shutdown of the socket */
7c1d71cf 1554 xprt->connect_cookie++;
d0bea455 1555 clear_bit(XPRT_CONNECTED, &xprt->state);
a519fc7a 1556 xs_tcp_force_close(xprt);
663b8858
TM		 1557 case TCP_CLOSING:
1558 /*
1559 * If the server closed down the connection, make sure that
1560 * we back off before reconnecting
1561 */
1562 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1563 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
3b948ae5
TM		 1564 break;
1565 case TCP_LAST_ACK:
670f9457 1566 set_bit(XPRT_CLOSING, &xprt->state);
4e857c58 1567 smp_mb__before_atomic();
3b948ae5 1568 clear_bit(XPRT_CONNECTED, &xprt->state);
4e857c58 1569 smp_mb__after_atomic();
3b948ae5
TM		 1570 break;
1571 case TCP_CLOSE:
0fdea1e8
TM		 1572 if (test_and_clear_bit(XPRT_SOCK_CONNECTING,
1573 &transport->sock_state))
1574 xprt_clear_connecting(xprt);
7d1e8255 1575 xs_sock_mark_closed(xprt);
a246b010
CL		 1576 }
1577 out:
f064af1e 1578 read_unlock_bh(&sk->sk_callback_lock);
a246b010
CL		 1579}
1580
1f0fa154
IJ		 1581static void xs_write_space(struct sock *sk)
1582{
1583 struct socket *sock;
1584 struct rpc_xprt *xprt;
1585
1586 if (unlikely(!(sock = sk->sk_socket)))
1587 return;
1588 clear_bit(SOCK_NOSPACE, &sock->flags);
1589
1590 if (unlikely(!(xprt = xprt_from_sock(sk))))
1591 return;
1592 if (test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags) == 0)
1593 return;
1594
1595 xprt_write_space(xprt);
1596}
1597
9903cd1c 1598/**
c7b2cae8
CL		 1599 * xs_udp_write_space - callback invoked when socket buffer space
1600 * becomes available
9903cd1c
CL		 1601 * @sk: socket whose state has changed
1602 *
a246b010
CL		 1603 * Called when more output buffer space is available for this socket.
1604 * We try not to wake our writers until they can make "significant"
c7b2cae8 1605 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
a246b010
CL		 1606 * with a bunch of small requests.
1607 */
c7b2cae8 1608static void xs_udp_write_space(struct sock *sk)
a246b010 1609{
f064af1e 1610 read_lock_bh(&sk->sk_callback_lock);
a246b010 1611
c7b2cae8 1612 /* from net/core/sock.c:sock_def_write_space */
1f0fa154
IJ		 1613 if (sock_writeable(sk))
1614 xs_write_space(sk);
a246b010 1615
f064af1e 1616 read_unlock_bh(&sk->sk_callback_lock);
c7b2cae8 1617}
a246b010 1618
c7b2cae8
CL		 1619/**
1620 * xs_tcp_write_space - callback invoked when socket buffer space
1621 * becomes available
1622 * @sk: socket whose state has changed
1623 *
1624 * Called when more output buffer space is available for this socket.
1625 * We try not to wake our writers until they can make "significant"
1626 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1627 * with a bunch of small requests.
1628 */
1629static void xs_tcp_write_space(struct sock *sk)
1630{
f064af1e 1631 read_lock_bh(&sk->sk_callback_lock);
c7b2cae8
CL		 1632
1633 /* from net/core/stream.c:sk_stream_write_space */
64dc6130 1634 if (sk_stream_is_writeable(sk))
1f0fa154 1635 xs_write_space(sk);
b6ddf64f 1636
f064af1e 1637 read_unlock_bh(&sk->sk_callback_lock);
a246b010
CL		 1638}
1639
470056c2 1640static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
a246b010 1641{
ee0ac0c2
CL		 1642 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1643 struct sock *sk = transport->inet;
a246b010 1644
7c6e066e 1645 if (transport->rcvsize) {
a246b010 1646 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
7c6e066e 1647 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
a246b010 1648 }
7c6e066e 1649 if (transport->sndsize) {
a246b010 1650 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
7c6e066e 1651 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
a246b010
CL		 1652 sk->sk_write_space(sk);
1653 }
1654}
1655
43118c29 1656/**
470056c2 1657 * xs_udp_set_buffer_size - set send and receive limits
43118c29 1658 * @xprt: generic transport
470056c2
CL		 1659 * @sndsize: requested size of send buffer, in bytes
1660 * @rcvsize: requested size of receive buffer, in bytes
43118c29 1661 *
470056c2 1662 * Set socket send and receive buffer size limits.
43118c29 1663 */
470056c2 1664static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
43118c29 1665{
7c6e066e
CL		 1666 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1667
1668 transport->sndsize = 0;
470056c2 1669 if (sndsize)
7c6e066e
CL		 1670 transport->sndsize = sndsize + 1024;
1671 transport->rcvsize = 0;
470056c2 1672 if (rcvsize)
7c6e066e 1673 transport->rcvsize = rcvsize + 1024;
470056c2
CL		 1674
1675 xs_udp_do_set_buffer_size(xprt);
43118c29
CL		 1676}
1677
46c0ee8b
CL		 1678/**
1679 * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1680 * @task: task that timed out
1681 *
1682 * Adjust the congestion window after a retransmit timeout has occurred.
1683 */
6a24dfb6 1684static void xs_udp_timer(struct rpc_xprt *xprt, struct rpc_task *task)
46c0ee8b 1685{
6a24dfb6 1686 xprt_adjust_cwnd(xprt, task, -ETIMEDOUT);
46c0ee8b
CL		 1687}
1688
b85d8806
CL		 1689static unsigned short xs_get_random_port(void)
1690{
1691 unsigned short range = xprt_max_resvport - xprt_min_resvport;
63862b5b 1692 unsigned short rand = (unsigned short) prandom_u32() % range;
b85d8806
CL		 1693 return rand + xprt_min_resvport;
1694}
1695
4dda9c8a
TM		 1696/**
1697 * xs_set_reuseaddr_port - set the socket's port and address reuse options
1698 * @sock: socket
1699 *
1700 * Note that this function has to be called on all sockets that share the
1701 * same port, and it must be called before binding.
1702 */
1703static void xs_sock_set_reuseport(struct socket *sock)
1704{
402e23b4 1705 int opt = 1;
4dda9c8a 1706
402e23b4
TM		 1707 kernel_setsockopt(sock, SOL_SOCKET, SO_REUSEPORT,
1708 (char *)&opt, sizeof(opt));
4dda9c8a
TM		 1709}
1710
1711static unsigned short xs_sock_getport(struct socket *sock)
1712{
1713 struct sockaddr_storage buf;
1714 int buflen;
1715 unsigned short port = 0;
1716
1717 if (kernel_getsockname(sock, (struct sockaddr *)&buf, &buflen) < 0)
1718 goto out;
1719 switch (buf.ss_family) {
1720 case AF_INET6:
1721 port = ntohs(((struct sockaddr_in6 *)&buf)->sin6_port);
1722 break;
1723 case AF_INET:
1724 port = ntohs(((struct sockaddr_in *)&buf)->sin_port);
1725 }
1726out:
1727 return port;
1728}
1729
92200412
CL		 1730/**
1731 * xs_set_port - reset the port number in the remote endpoint address
1732 * @xprt: generic transport
1733 * @port: new port number
1734 *
1735 */
1736static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1737{
46121cf7 1738 dprintk("RPC: setting port for xprt %p to %u\n", xprt, port);
c4efcb1d 1739
9dc3b095
CL		 1740 rpc_set_port(xs_addr(xprt), port);
1741 xs_update_peer_port(xprt);
92200412
CL		 1742}
1743
4dda9c8a
TM		 1744static void xs_set_srcport(struct sock_xprt *transport, struct socket *sock)
1745{
1746 if (transport->srcport == 0)
1747 transport->srcport = xs_sock_getport(sock);
1748}
1749
5d4ec932 1750static unsigned short xs_get_srcport(struct sock_xprt *transport)
67a391d7 1751{
fbfffbd5 1752 unsigned short port = transport->srcport;
67a391d7
TM		 1753
1754 if (port == 0 && transport->xprt.resvport)
1755 port = xs_get_random_port();
1756 return port;
1757}
1758
baaf4e48 1759static unsigned short xs_next_srcport(struct sock_xprt *transport, unsigned short port)
67a391d7 1760{
fbfffbd5
CL		 1761 if (transport->srcport != 0)
1762 transport->srcport = 0;
67a391d7
TM		 1763 if (!transport->xprt.resvport)
1764 return 0;
1765 if (port <= xprt_min_resvport || port > xprt_max_resvport)
1766 return xprt_max_resvport;
1767 return --port;
1768}
beb59b68 1769static int xs_bind(struct sock_xprt *transport, struct socket *sock)
a246b010 1770{
beb59b68 1771 struct sockaddr_storage myaddr;
67a391d7 1772 int err, nloop = 0;
5d4ec932 1773 unsigned short port = xs_get_srcport(transport);
67a391d7 1774 unsigned short last;
a246b010 1775
0f7a622c
CP		 1776 /*
1777 * If we are asking for any ephemeral port (i.e. port == 0 &&
1778 * transport->xprt.resvport == 0), don't bind. Let the local
1779 * port selection happen implicitly when the socket is used
1780 * (for example at connect time).
1781 *
1782 * This ensures that we can continue to establish TCP
1783 * connections even when all local ephemeral ports are already
1784 * a part of some TCP connection. This makes no difference
1785 * for UDP sockets, but also doens't harm them.
1786 *
1787 * If we're asking for any reserved port (i.e. port == 0 &&
1788 * transport->xprt.resvport == 1) xs_get_srcport above will
1789 * ensure that port is non-zero and we will bind as needed.
1790 */
1791 if (port == 0)
1792 return 0;
1793
beb59b68 1794 memcpy(&myaddr, &transport->srcaddr, transport->xprt.addrlen);
a246b010 1795 do {
beb59b68
PE		 1796 rpc_set_port((struct sockaddr *)&myaddr, port);
1797 err = kernel_bind(sock, (struct sockaddr *)&myaddr,
1798 transport->xprt.addrlen);
a246b010 1799 if (err == 0) {
fbfffbd5 1800 transport->srcport = port;
d3bc9a1d 1801 break;
a246b010 1802 }
67a391d7 1803 last = port;
baaf4e48 1804 port = xs_next_srcport(transport, port);
67a391d7
TM		 1805 if (port > last)
1806 nloop++;
1807 } while (err == -EADDRINUSE && nloop != 2);
90058d37 1808
4232e863 1809 if (myaddr.ss_family == AF_INET)
beb59b68
PE		 1810 dprintk("RPC: %s %pI4:%u: %s (%d)\n", __func__,
1811 &((struct sockaddr_in *)&myaddr)->sin_addr,
1812 port, err ? "failed" : "ok", err);
1813 else
1814 dprintk("RPC: %s %pI6:%u: %s (%d)\n", __func__,
1815 &((struct sockaddr_in6 *)&myaddr)->sin6_addr,
1816 port, err ? "failed" : "ok", err);
a246b010
CL		 1817 return err;
1818}
1819
176e21ee
CL		 1820/*
1821 * We don't support autobind on AF_LOCAL sockets
1822 */
1823static void xs_local_rpcbind(struct rpc_task *task)
1824{
3dc0da27
TM		 1825 rcu_read_lock();
1826 xprt_set_bound(rcu_dereference(task->tk_client->cl_xprt));
1827 rcu_read_unlock();
176e21ee
CL		 1828}
1829
1830static void xs_local_set_port(struct rpc_xprt *xprt, unsigned short port)
1831{
1832}
beb59b68 1833
ed07536e
PZ		 1834#ifdef CONFIG_DEBUG_LOCK_ALLOC
1835static struct lock_class_key xs_key[2];
1836static struct lock_class_key xs_slock_key[2];
1837
176e21ee
CL		 1838static inline void xs_reclassify_socketu(struct socket *sock)
1839{
1840 struct sock *sk = sock->sk;
1841
176e21ee
CL		 1842 sock_lock_init_class_and_name(sk, "slock-AF_LOCAL-RPC",
1843 &xs_slock_key[1], "sk_lock-AF_LOCAL-RPC", &xs_key[1]);
1844}
1845
8945ee5e 1846static inline void xs_reclassify_socket4(struct socket *sock)
ed07536e
PZ		 1847{
1848 struct sock *sk = sock->sk;
8945ee5e 1849
8945ee5e
CL		 1850 sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1851 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1852}
ed07536e 1853
8945ee5e
CL		 1854static inline void xs_reclassify_socket6(struct socket *sock)
1855{
1856 struct sock *sk = sock->sk;
ed07536e 1857
8945ee5e
CL		 1858 sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1859 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
ed07536e 1860}
6bc9638a
PE		 1861
1862static inline void xs_reclassify_socket(int family, struct socket *sock)
1863{
1b7a1819
WAA		 1864 WARN_ON_ONCE(sock_owned_by_user(sock->sk));
1865 if (sock_owned_by_user(sock->sk))
1866 return;
1867
4232e863 1868 switch (family) {
176e21ee
CL		 1869 case AF_LOCAL:
1870 xs_reclassify_socketu(sock);
1871 break;
4232e863 1872 case AF_INET:
6bc9638a 1873 xs_reclassify_socket4(sock);
4232e863
CL		 1874 break;
1875 case AF_INET6:
6bc9638a 1876 xs_reclassify_socket6(sock);
4232e863
CL		 1877 break;
1878 }
6bc9638a 1879}
ed07536e 1880#else
176e21ee
CL		 1881static inline void xs_reclassify_socketu(struct socket *sock)
1882{
1883}
1884
8945ee5e
CL		 1885static inline void xs_reclassify_socket4(struct socket *sock)
1886{
1887}
1888
1889static inline void xs_reclassify_socket6(struct socket *sock)
ed07536e
PZ		 1890{
1891}
6bc9638a
PE		 1892
1893static inline void xs_reclassify_socket(int family, struct socket *sock)
1894{
1895}
ed07536e
PZ		 1896#endif
1897
edc1b01c
TM		 1898static void xs_dummy_data_receive_workfn(struct work_struct *work)
1899{
1900}
1901
93dc41bd
N		 1902static void xs_dummy_setup_socket(struct work_struct *work)
1903{
1904}
1905
6bc9638a 1906static struct socket *xs_create_sock(struct rpc_xprt *xprt,
4dda9c8a
TM		 1907 struct sock_xprt *transport, int family, int type,
1908 int protocol, bool reuseport)
22f79326
PE		 1909{
1910 struct socket *sock;
1911 int err;
1912
6bc9638a 1913 err = __sock_create(xprt->xprt_net, family, type, protocol, &sock, 1);
22f79326
PE		 1914 if (err < 0) {
1915 dprintk("RPC: can't create %d transport socket (%d).\n",
1916 protocol, -err);
1917 goto out;
1918 }
6bc9638a 1919 xs_reclassify_socket(family, sock);
22f79326 1920
4dda9c8a
TM		 1921 if (reuseport)
1922 xs_sock_set_reuseport(sock);
1923
4cea288a
BH		 1924 err = xs_bind(transport, sock);
1925 if (err) {
22f79326
PE		 1926 sock_release(sock);
1927 goto out;
1928 }
1929
1930 return sock;
1931out:
1932 return ERR_PTR(err);
1933}
1934
176e21ee
CL		 1935static int xs_local_finish_connecting(struct rpc_xprt *xprt,
1936 struct socket *sock)
1937{
1938 struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1939 xprt);
1940
1941 if (!transport->inet) {
1942 struct sock *sk = sock->sk;
1943
1944 write_lock_bh(&sk->sk_callback_lock);
1945
1946 xs_save_old_callbacks(transport, sk);
1947
1948 sk->sk_user_data = xprt;
1949 sk->sk_data_ready = xs_local_data_ready;
1950 sk->sk_write_space = xs_udp_write_space;
2118071d 1951 sk->sk_error_report = xs_error_report;
c2126157 1952 sk->sk_allocation = GFP_NOIO;
176e21ee
CL		 1953
1954 xprt_clear_connected(xprt);
1955
1956 /* Reset to new socket */
1957 transport->sock = sock;
1958 transport->inet = sk;
1959
1960 write_unlock_bh(&sk->sk_callback_lock);
1961 }
1962
1963 /* Tell the socket layer to start connecting... */
1964 xprt->stat.connect_count++;
1965 xprt->stat.connect_start = jiffies;
1966 return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, 0);
1967}
1968
1969/**
1970 * xs_local_setup_socket - create AF_LOCAL socket, connect to a local endpoint
176e21ee 1971 * @transport: socket transport to connect
176e21ee 1972 */
dc107402 1973static int xs_local_setup_socket(struct sock_xprt *transport)
176e21ee 1974{
176e21ee
CL		 1975 struct rpc_xprt *xprt = &transport->xprt;
1976 struct socket *sock;
1977 int status = -EIO;
1978
176e21ee
CL		 1979 status = __sock_create(xprt->xprt_net, AF_LOCAL,
1980 SOCK_STREAM, 0, &sock, 1);
1981 if (status < 0) {
1982 dprintk("RPC: can't create AF_LOCAL "
1983 "transport socket (%d).\n", -status);
1984 goto out;
1985 }
1986 xs_reclassify_socketu(sock);
1987
1988 dprintk("RPC: worker connecting xprt %p via AF_LOCAL to %s\n",
1989 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1990
1991 status = xs_local_finish_connecting(xprt, sock);
40b5ea0c 1992 trace_rpc_socket_connect(xprt, sock, status);
176e21ee
CL		 1993 switch (status) {
1994 case 0:
1995 dprintk("RPC: xprt %p connected to %s\n",
1996 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1997 xprt_set_connected(xprt);
3601c4a9 1998 case -ENOBUFS:
176e21ee
CL		 1999 break;
2000 case -ENOENT:
2001 dprintk("RPC: xprt %p: socket %s does not exist\n",
2002 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2003 break;
4a20a988
TM		 2004 case -ECONNREFUSED:
2005 dprintk("RPC: xprt %p: connection refused for %s\n",
2006 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2007 break;
176e21ee
CL		 2008 default:
2009 printk(KERN_ERR "%s: unhandled error (%d) connecting to %s\n",
2010 __func__, -status,
2011 xprt->address_strings[RPC_DISPLAY_ADDR]);
2012 }
2013
2014out:
2015 xprt_clear_connecting(xprt);
2016 xprt_wake_pending_tasks(xprt, status);
dc107402
BF		 2017 return status;
2018}
2019
b6669737 2020static void xs_local_connect(struct rpc_xprt *xprt, struct rpc_task *task)
dc107402 2021{
dc107402
BF		 2022 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2023 int ret;
2024
2025 if (RPC_IS_ASYNC(task)) {
2026 /*
2027 * We want the AF_LOCAL connect to be resolved in the
2028 * filesystem namespace of the process making the rpc
2029 * call. Thus we connect synchronously.
2030 *
2031 * If we want to support asynchronous AF_LOCAL calls,
2032 * we'll need to figure out how to pass a namespace to
2033 * connect.
2034 */
2035 rpc_exit(task, -ENOTCONN);
2036 return;
2037 }
2038 ret = xs_local_setup_socket(transport);
2039 if (ret && !RPC_IS_SOFTCONN(task))
2040 msleep_interruptible(15000);
176e21ee
CL		 2041}
2042
3c87ef6e 2043#if IS_ENABLED(CONFIG_SUNRPC_SWAP)
d6e971d8
JL		 2044/*
2045 * Note that this should be called with XPRT_LOCKED held (or when we otherwise
2046 * know that we have exclusive access to the socket), to guard against
2047 * races with xs_reset_transport.
2048 */
a564b8f0
MG		 2049static void xs_set_memalloc(struct rpc_xprt *xprt)
2050{
2051 struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
2052 xprt);
2053
d6e971d8
JL		 2054 /*
2055 * If there's no sock, then we have nothing to set. The
2056 * reconnecting process will get it for us.
2057 */
2058 if (!transport->inet)
2059 return;
8e228133 2060 if (atomic_read(&xprt->swapper))
a564b8f0
MG		 2061 sk_set_memalloc(transport->inet);
2062}
2063
2064/**
d67fa4d8 2065 * xs_enable_swap - Tag this transport as being used for swap.
a564b8f0 2066 * @xprt: transport to tag
a564b8f0 2067 *
8e228133
JL		 2068 * Take a reference to this transport on behalf of the rpc_clnt, and
2069 * optionally mark it for swapping if it wasn't already.
a564b8f0 2070 */
d67fa4d8
JL		 2071static int
2072xs_enable_swap(struct rpc_xprt *xprt)
a564b8f0 2073{
d6e971d8 2074 struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
a564b8f0 2075
d6e971d8
JL		 2076 if (atomic_inc_return(&xprt->swapper) != 1)
2077 return 0;
2078 if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE))
2079 return -ERESTARTSYS;
2080 if (xs->inet)
2081 sk_set_memalloc(xs->inet);
2082 xprt_release_xprt(xprt, NULL);
8e228133
JL		 2083 return 0;
2084}
a564b8f0 2085
8e228133 2086/**
d67fa4d8 2087 * xs_disable_swap - Untag this transport as being used for swap.
8e228133
JL		 2088 * @xprt: transport to tag
2089 *
2090 * Drop a "swapper" reference to this xprt on behalf of the rpc_clnt. If the
2091 * swapper refcount goes to 0, untag the socket as a memalloc socket.
2092 */
d67fa4d8
JL		 2093static void
2094xs_disable_swap(struct rpc_xprt *xprt)
8e228133 2095{
d6e971d8 2096 struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
8e228133 2097
d6e971d8
JL		 2098 if (!atomic_dec_and_test(&xprt->swapper))
2099 return;
2100 if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE))
2101 return;
2102 if (xs->inet)
2103 sk_clear_memalloc(xs->inet);
2104 xprt_release_xprt(xprt, NULL);
a564b8f0 2105}
a564b8f0
MG		 2106#else
2107static void xs_set_memalloc(struct rpc_xprt *xprt)
2108{
2109}
d67fa4d8
JL		 2110
2111static int
2112xs_enable_swap(struct rpc_xprt *xprt)
2113{
2114 return -EINVAL;
2115}
2116
2117static void
2118xs_disable_swap(struct rpc_xprt *xprt)
2119{
2120}
a564b8f0
MG		 2121#endif
2122
16be2d20
CL		 2123static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2124{
2125 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2126
2127 if (!transport->inet) {
2128 struct sock *sk = sock->sk;
2129
2130 write_lock_bh(&sk->sk_callback_lock);
2131
2a9e1cfa
TM		 2132 xs_save_old_callbacks(transport, sk);
2133
16be2d20 2134 sk->sk_user_data = xprt;
f9b2ee71 2135 sk->sk_data_ready = xs_data_ready;
16be2d20 2136 sk->sk_write_space = xs_udp_write_space;
c2126157 2137 sk->sk_allocation = GFP_NOIO;
16be2d20
CL		 2138
2139 xprt_set_connected(xprt);
2140
2141 /* Reset to new socket */
2142 transport->sock = sock;
2143 transport->inet = sk;
2144
a564b8f0
MG		 2145 xs_set_memalloc(xprt);
2146
16be2d20
CL		 2147 write_unlock_bh(&sk->sk_callback_lock);
2148 }
2149 xs_udp_do_set_buffer_size(xprt);
2150}
2151
8c14ff2a 2152static void xs_udp_setup_socket(struct work_struct *work)
a246b010 2153{
34161db6
TM		 2154 struct sock_xprt *transport =
2155 container_of(work, struct sock_xprt, connect_worker.work);
c8475461 2156 struct rpc_xprt *xprt = &transport->xprt;
ee0ac0c2 2157 struct socket *sock = transport->sock;
b65c0310 2158 int status = -EIO;
9903cd1c 2159
8c14ff2a 2160 sock = xs_create_sock(xprt, transport,
4dda9c8a
TM		 2161 xs_addr(xprt)->sa_family, SOCK_DGRAM,
2162 IPPROTO_UDP, false);
b65c0310 2163 if (IS_ERR(sock))
b0d93ad5 2164 goto out;
68e220bd 2165
c740eff8
CL		 2166 dprintk("RPC: worker connecting xprt %p via %s to "
2167 "%s (port %s)\n", xprt,
2168 xprt->address_strings[RPC_DISPLAY_PROTO],
2169 xprt->address_strings[RPC_DISPLAY_ADDR],
2170 xprt->address_strings[RPC_DISPLAY_PORT]);
68e220bd
CL		 2171
2172 xs_udp_finish_connecting(xprt, sock);
40b5ea0c 2173 trace_rpc_socket_connect(xprt, sock, 0);
b0d93ad5
CL		 2174 status = 0;
2175out:
718ba5b8 2176 xprt_unlock_connect(xprt, transport);
b0d93ad5 2177 xprt_clear_connecting(xprt);
7d1e8255 2178 xprt_wake_pending_tasks(xprt, status);
a246b010
CL		 2179}
2180
4876cc77
TM		 2181/**
2182 * xs_tcp_shutdown - gracefully shut down a TCP socket
2183 * @xprt: transport
2184 *
2185 * Initiates a graceful shutdown of the TCP socket by calling the
2186 * equivalent of shutdown(SHUT_RDWR);
2187 */
2188static void xs_tcp_shutdown(struct rpc_xprt *xprt)
2189{
2190 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2191 struct socket *sock = transport->sock;
2192
2193 if (sock == NULL)
2194 return;
2195 if (xprt_connected(xprt)) {
2196 kernel_sock_shutdown(sock, SHUT_RDWR);
2197 trace_rpc_socket_shutdown(xprt, sock);
2198 } else
2199 xs_reset_transport(transport);
2200}
2201
16be2d20 2202static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
a246b010 2203{
16be2d20 2204 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
fe19a96b 2205 int ret = -ENOTCONN;
edb267a6 2206
ee0ac0c2 2207 if (!transport->inet) {
b0d93ad5 2208 struct sock *sk = sock->sk;
7f260e85
TM		 2209 unsigned int keepidle = xprt->timeout->to_initval / HZ;
2210 unsigned int keepcnt = xprt->timeout->to_retries + 1;
2211 unsigned int opt_on = 1;
775f06ab 2212 unsigned int timeo;
7f260e85
TM		 2213
2214 /* TCP Keepalive options */
2215 kernel_setsockopt(sock, SOL_SOCKET, SO_KEEPALIVE,
2216 (char *)&opt_on, sizeof(opt_on));
2217 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPIDLE,
2218 (char *)&keepidle, sizeof(keepidle));
2219 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPINTVL,
2220 (char *)&keepidle, sizeof(keepidle));
2221 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPCNT,
2222 (char *)&keepcnt, sizeof(keepcnt));
b0d93ad5 2223
775f06ab
TM		 2224 /* TCP user timeout (see RFC5482) */
2225 timeo = jiffies_to_msecs(xprt->timeout->to_initval) *
2226 (xprt->timeout->to_retries + 1);
2227 kernel_setsockopt(sock, SOL_TCP, TCP_USER_TIMEOUT,
2228 (char *)&timeo, sizeof(timeo));
2229
b0d93ad5
CL		 2230 write_lock_bh(&sk->sk_callback_lock);
2231
2a9e1cfa
TM		 2232 xs_save_old_callbacks(transport, sk);
2233
b0d93ad5 2234 sk->sk_user_data = xprt;
b0d93ad5
CL		 2235 sk->sk_data_ready = xs_tcp_data_ready;
2236 sk->sk_state_change = xs_tcp_state_change;
2237 sk->sk_write_space = xs_tcp_write_space;
2118071d 2238 sk->sk_error_report = xs_error_report;
c2126157 2239 sk->sk_allocation = GFP_NOIO;
3167e12c
CL		 2240
2241 /* socket options */
3167e12c 2242 sock_reset_flag(sk, SOCK_LINGER);
3167e12c 2243 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
b0d93ad5
CL		 2244
2245 xprt_clear_connected(xprt);
2246
2247 /* Reset to new socket */
ee0ac0c2
CL		 2248 transport->sock = sock;
2249 transport->inet = sk;
b0d93ad5
CL		 2250
2251 write_unlock_bh(&sk->sk_callback_lock);
2252 }
2253
01d37c42 2254 if (!xprt_bound(xprt))
fe19a96b 2255 goto out;
01d37c42 2256
a564b8f0
MG		 2257 xs_set_memalloc(xprt);
2258
b0d93ad5 2259 /* Tell the socket layer to start connecting... */
262ca07d
CL		 2260 xprt->stat.connect_count++;
2261 xprt->stat.connect_start = jiffies;
0fdea1e8 2262 set_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
fe19a96b
TM		 2263 ret = kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
2264 switch (ret) {
2265 case 0:
4dda9c8a 2266 xs_set_srcport(transport, sock);
fe19a96b
TM		 2267 case -EINPROGRESS:
2268 /* SYN_SENT! */
fe19a96b
TM		 2269 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2270 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2271 }
2272out:
2273 return ret;
16be2d20
CL		 2274}
2275
9903cd1c 2276/**
b61d59ff 2277 * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint
9903cd1c
CL		 2278 *
2279 * Invoked by a work queue tasklet.
a246b010 2280 */
cdd518d5 2281static void xs_tcp_setup_socket(struct work_struct *work)
a246b010 2282{
cdd518d5
PE		 2283 struct sock_xprt *transport =
2284 container_of(work, struct sock_xprt, connect_worker.work);
ee0ac0c2 2285 struct socket *sock = transport->sock;
a9f5f0f7 2286 struct rpc_xprt *xprt = &transport->xprt;
b61d59ff 2287 int status = -EIO;
a246b010 2288
ee0ac0c2 2289 if (!sock) {
cdd518d5 2290 sock = xs_create_sock(xprt, transport,
4dda9c8a
TM		 2291 xs_addr(xprt)->sa_family, SOCK_STREAM,
2292 IPPROTO_TCP, true);
b61d59ff
TM		 2293 if (IS_ERR(sock)) {
2294 status = PTR_ERR(sock);
3167e12c
CL		 2295 goto out;
2296 }
7d1e8255 2297 }
a246b010 2298
c740eff8
CL		 2299 dprintk("RPC: worker connecting xprt %p via %s to "
2300 "%s (port %s)\n", xprt,
2301 xprt->address_strings[RPC_DISPLAY_PROTO],
2302 xprt->address_strings[RPC_DISPLAY_ADDR],
2303 xprt->address_strings[RPC_DISPLAY_PORT]);
edb267a6 2304
16be2d20 2305 status = xs_tcp_finish_connecting(xprt, sock);
40b5ea0c 2306 trace_rpc_socket_connect(xprt, sock, status);
46121cf7
CL		 2307 dprintk("RPC: %p connect status %d connected %d sock state %d\n",
2308 xprt, -status, xprt_connected(xprt),
2309 sock->sk->sk_state);
2a491991 2310 switch (status) {
f75e6745
TM		 2311 default:
2312 printk("%s: connect returned unhandled error %d\n",
2313 __func__, status);
2314 case -EADDRNOTAVAIL:
2315 /* We're probably in TIME_WAIT. Get rid of existing socket,
2316 * and retry
2317 */
a519fc7a 2318 xs_tcp_force_close(xprt);
88b5ed73 2319 break;
2a491991
TM		 2320 case 0:
2321 case -EINPROGRESS:
2322 case -EALREADY:
718ba5b8 2323 xprt_unlock_connect(xprt, transport);
7d1e8255 2324 return;
9fcfe0c8
TM		 2325 case -EINVAL:
2326 /* Happens, for instance, if the user specified a link
2327 * local IPv6 address without a scope-id.
2328 */
3ed5e2a2
TM		 2329 case -ECONNREFUSED:
2330 case -ECONNRESET:
2331 case -ENETUNREACH:
3913c78c 2332 case -EADDRINUSE:
3601c4a9 2333 case -ENOBUFS:
3ed5e2a2 2334 /* retry with existing socket, after a delay */
4efdd92c 2335 xs_tcp_force_close(xprt);
9fcfe0c8 2336 goto out;
a246b010 2337 }
2a491991 2338 status = -EAGAIN;
a246b010 2339out:
718ba5b8 2340 xprt_unlock_connect(xprt, transport);
68e220bd 2341 xprt_clear_connecting(xprt);
7d1e8255 2342 xprt_wake_pending_tasks(xprt, status);
68e220bd 2343}
b0d93ad5 2344
9903cd1c
CL		 2345/**
2346 * xs_connect - connect a socket to a remote endpoint
1b092092 2347 * @xprt: pointer to transport structure
9903cd1c
CL		 2348 * @task: address of RPC task that manages state of connect request
2349 *
2350 * TCP: If the remote end dropped the connection, delay reconnecting.
03bf4b70
CL		 2351 *
2352 * UDP socket connects are synchronous, but we use a work queue anyway
2353 * to guarantee that even unprivileged user processes can set up a
2354 * socket on a privileged port.
2355 *
2356 * If a UDP socket connect fails, the delay behavior here prevents
2357 * retry floods (hard mounts).
9903cd1c 2358 */
1b092092 2359static void xs_connect(struct rpc_xprt *xprt, struct rpc_task *task)
a246b010 2360{
ee0ac0c2 2361 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
a246b010 2362
718ba5b8
TM		 2363 WARN_ON_ONCE(!xprt_lock_connect(xprt, task, transport));
2364
99b1a4c3 2365 if (transport->sock != NULL) {
46121cf7
CL		 2366 dprintk("RPC: xs_connect delayed xprt %p for %lu "
2367 "seconds\n",
03bf4b70 2368 xprt, xprt->reestablish_timeout / HZ);
99b1a4c3
TM		 2369
2370 /* Start by resetting any existing state */
2371 xs_reset_transport(transport);
2372
c1384c9c
TM		 2373 queue_delayed_work(rpciod_workqueue,
2374 &transport->connect_worker,
2375 xprt->reestablish_timeout);
03bf4b70 2376 xprt->reestablish_timeout <<= 1;
61d0a8e6
NB		 2377 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2378 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
03bf4b70
CL		 2379 if (xprt->reestablish_timeout > XS_TCP_MAX_REEST_TO)
2380 xprt->reestablish_timeout = XS_TCP_MAX_REEST_TO;
b0d93ad5 2381 } else {
46121cf7 2382 dprintk("RPC: xs_connect scheduled xprt %p\n", xprt);
c1384c9c
TM		 2383 queue_delayed_work(rpciod_workqueue,
2384 &transport->connect_worker, 0);
a246b010
CL		 2385 }
2386}
2387
176e21ee
CL		 2388/**
2389 * xs_local_print_stats - display AF_LOCAL socket-specifc stats
2390 * @xprt: rpc_xprt struct containing statistics
2391 * @seq: output file
2392 *
2393 */
2394static void xs_local_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2395{
2396 long idle_time = 0;
2397
2398 if (xprt_connected(xprt))
2399 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2400
2401 seq_printf(seq, "\txprt:\tlocal %lu %lu %lu %ld %lu %lu %lu "
15a45206 2402 "%llu %llu %lu %llu %llu\n",
176e21ee
CL		 2403 xprt->stat.bind_count,
2404 xprt->stat.connect_count,
2405 xprt->stat.connect_time,
2406 idle_time,
2407 xprt->stat.sends,
2408 xprt->stat.recvs,
2409 xprt->stat.bad_xids,
2410 xprt->stat.req_u,
15a45206
AA		 2411 xprt->stat.bklog_u,
2412 xprt->stat.max_slots,
2413 xprt->stat.sending_u,
2414 xprt->stat.pending_u);
176e21ee
CL		 2415}
2416
262ca07d
CL		 2417/**
2418 * xs_udp_print_stats - display UDP socket-specifc stats
2419 * @xprt: rpc_xprt struct containing statistics
2420 * @seq: output file
2421 *
2422 */
2423static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2424{
c8475461
CL		 2425 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2426
15a45206
AA		 2427 seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %llu %llu "
2428 "%lu %llu %llu\n",
fbfffbd5 2429 transport->srcport,
262ca07d
CL		 2430 xprt->stat.bind_count,
2431 xprt->stat.sends,
2432 xprt->stat.recvs,
2433 xprt->stat.bad_xids,
2434 xprt->stat.req_u,
15a45206
AA		 2435 xprt->stat.bklog_u,
2436 xprt->stat.max_slots,
2437 xprt->stat.sending_u,
2438 xprt->stat.pending_u);
262ca07d
CL		 2439}
2440
2441/**
2442 * xs_tcp_print_stats - display TCP socket-specifc stats
2443 * @xprt: rpc_xprt struct containing statistics
2444 * @seq: output file
2445 *
2446 */
2447static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2448{
c8475461 2449 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
262ca07d
CL		 2450 long idle_time = 0;
2451
2452 if (xprt_connected(xprt))
2453 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2454
15a45206
AA		 2455 seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu "
2456 "%llu %llu %lu %llu %llu\n",
fbfffbd5 2457 transport->srcport,
262ca07d
CL		 2458 xprt->stat.bind_count,
2459 xprt->stat.connect_count,
2460 xprt->stat.connect_time,
2461 idle_time,
2462 xprt->stat.sends,
2463 xprt->stat.recvs,
2464 xprt->stat.bad_xids,
2465 xprt->stat.req_u,
15a45206
AA		 2466 xprt->stat.bklog_u,
2467 xprt->stat.max_slots,
2468 xprt->stat.sending_u,
2469 xprt->stat.pending_u);
262ca07d
CL		 2470}
2471
4cfc7e60
RI		 2472/*
2473 * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason
2474 * we allocate pages instead doing a kmalloc like rpc_malloc is because we want
2475 * to use the server side send routines.
2476 */
5a51f13a 2477static void *bc_malloc(struct rpc_task *task, size_t size)
4cfc7e60
RI		 2478{
2479 struct page *page;
2480 struct rpc_buffer *buf;
2481
b8a13d03
WAA		 2482 WARN_ON_ONCE(size > PAGE_SIZE - sizeof(struct rpc_buffer));
2483 if (size > PAGE_SIZE - sizeof(struct rpc_buffer))
2484 return NULL;
4cfc7e60 2485
b8a13d03 2486 page = alloc_page(GFP_KERNEL);
4cfc7e60
RI		 2487 if (!page)
2488 return NULL;
2489
2490 buf = page_address(page);
2491 buf->len = PAGE_SIZE;
2492
2493 return buf->data;
2494}
2495
2496/*
2497 * Free the space allocated in the bc_alloc routine
2498 */
5a51f13a 2499static void bc_free(void *buffer)
4cfc7e60
RI		 2500{
2501 struct rpc_buffer *buf;
2502
2503 if (!buffer)
2504 return;
2505
2506 buf = container_of(buffer, struct rpc_buffer, data);
2507 free_page((unsigned long)buf);
2508}
2509
2510/*
2511 * Use the svc_sock to send the callback. Must be called with svsk->sk_mutex
2512 * held. Borrows heavily from svc_tcp_sendto and xs_tcp_send_request.
2513 */
2514static int bc_sendto(struct rpc_rqst *req)
2515{
2516 int len;
2517 struct xdr_buf *xbufp = &req->rq_snd_buf;
2518 struct rpc_xprt *xprt = req->rq_xprt;
2519 struct sock_xprt *transport =
2520 container_of(xprt, struct sock_xprt, xprt);
2521 struct socket *sock = transport->sock;
2522 unsigned long headoff;
2523 unsigned long tailoff;
2524
61677eee 2525 xs_encode_stream_record_marker(xbufp);
4cfc7e60
RI		 2526
2527 tailoff = (unsigned long)xbufp->tail[0].iov_base & ~PAGE_MASK;
2528 headoff = (unsigned long)xbufp->head[0].iov_base & ~PAGE_MASK;
2529 len = svc_send_common(sock, xbufp,
2530 virt_to_page(xbufp->head[0].iov_base), headoff,
2531 xbufp->tail[0].iov_base, tailoff);
2532
2533 if (len != xbufp->len) {
2534 printk(KERN_NOTICE "Error sending entire callback!\n");
2535 len = -EAGAIN;
2536 }
2537
2538 return len;
2539}
2540
2541/*
2542 * The send routine. Borrows from svc_send
2543 */
2544static int bc_send_request(struct rpc_task *task)
2545{
2546 struct rpc_rqst *req = task->tk_rqstp;
2547 struct svc_xprt *xprt;
4cfc7e60
RI		 2548 u32 len;
2549
2550 dprintk("sending request with xid: %08x\n", ntohl(req->rq_xid));
2551 /*
2552 * Get the server socket associated with this callback xprt
2553 */
2554 xprt = req->rq_xprt->bc_xprt;
4cfc7e60
RI		 2555
2556 /*
2557 * Grab the mutex to serialize data as the connection is shared
2558 * with the fore channel
2559 */
2560 if (!mutex_trylock(&xprt->xpt_mutex)) {
2561 rpc_sleep_on(&xprt->xpt_bc_pending, task, NULL);
2562 if (!mutex_trylock(&xprt->xpt_mutex))
2563 return -EAGAIN;
2564 rpc_wake_up_queued_task(&xprt->xpt_bc_pending, task);
2565 }
2566 if (test_bit(XPT_DEAD, &xprt->xpt_flags))
2567 len = -ENOTCONN;
2568 else
2569 len = bc_sendto(req);
2570 mutex_unlock(&xprt->xpt_mutex);
2571
2572 if (len > 0)
2573 len = 0;
2574
2575 return len;
2576}
2577
2578/*
2579 * The close routine. Since this is client initiated, we do nothing
2580 */
2581
2582static void bc_close(struct rpc_xprt *xprt)
2583{
4cfc7e60
RI		 2584}
2585
2586/*
2587 * The xprt destroy routine. Again, because this connection is client
2588 * initiated, we do nothing
2589 */
2590
2591static void bc_destroy(struct rpc_xprt *xprt)
2592{
47f72efa
KM		 2593 dprintk("RPC: bc_destroy xprt %p\n", xprt);
2594
2595 xs_xprt_free(xprt);
2596 module_put(THIS_MODULE);
4cfc7e60
RI		 2597}
2598
176e21ee
CL		 2599static struct rpc_xprt_ops xs_local_ops = {
2600 .reserve_xprt = xprt_reserve_xprt,
2601 .release_xprt = xs_tcp_release_xprt,
f39c1bfb 2602 .alloc_slot = xprt_alloc_slot,
176e21ee
CL		 2603 .rpcbind = xs_local_rpcbind,
2604 .set_port = xs_local_set_port,
dc107402 2605 .connect = xs_local_connect,
176e21ee
CL		 2606 .buf_alloc = rpc_malloc,
2607 .buf_free = rpc_free,
2608 .send_request = xs_local_send_request,
2609 .set_retrans_timeout = xprt_set_retrans_timeout_def,
2610 .close = xs_close,
a1311d87 2611 .destroy = xs_destroy,
176e21ee 2612 .print_stats = xs_local_print_stats,
d67fa4d8
JL		 2613 .enable_swap = xs_enable_swap,
2614 .disable_swap = xs_disable_swap,
176e21ee
CL		 2615};
2616
262965f5 2617static struct rpc_xprt_ops xs_udp_ops = {
43118c29 2618 .set_buffer_size = xs_udp_set_buffer_size,
12a80469 2619 .reserve_xprt = xprt_reserve_xprt_cong,
49e9a890 2620 .release_xprt = xprt_release_xprt_cong,
f39c1bfb 2621 .alloc_slot = xprt_alloc_slot,
45160d62 2622 .rpcbind = rpcb_getport_async,
92200412 2623 .set_port = xs_set_port,
262965f5 2624 .connect = xs_connect,
02107148
CL		 2625 .buf_alloc = rpc_malloc,
2626 .buf_free = rpc_free,
262965f5 2627 .send_request = xs_udp_send_request,
fe3aca29 2628 .set_retrans_timeout = xprt_set_retrans_timeout_rtt,
46c0ee8b 2629 .timer = xs_udp_timer,
a58dd398 2630 .release_request = xprt_release_rqst_cong,
262965f5
CL		 2631 .close = xs_close,
2632 .destroy = xs_destroy,
262ca07d 2633 .print_stats = xs_udp_print_stats,
d67fa4d8
JL		 2634 .enable_swap = xs_enable_swap,
2635 .disable_swap = xs_disable_swap,
4a068258 2636 .inject_disconnect = xs_inject_disconnect,
262965f5
CL		 2637};
2638
2639static struct rpc_xprt_ops xs_tcp_ops = {
12a80469 2640 .reserve_xprt = xprt_reserve_xprt,
e0ab53de 2641 .release_xprt = xs_tcp_release_xprt,
f39c1bfb 2642 .alloc_slot = xprt_lock_and_alloc_slot,
45160d62 2643 .rpcbind = rpcb_getport_async,
92200412 2644 .set_port = xs_set_port,
0b9e7943 2645 .connect = xs_connect,
02107148
CL		 2646 .buf_alloc = rpc_malloc,
2647 .buf_free = rpc_free,
262965f5 2648 .send_request = xs_tcp_send_request,
fe3aca29 2649 .set_retrans_timeout = xprt_set_retrans_timeout_def,
c627d31b 2650 .close = xs_tcp_shutdown,
9903cd1c 2651 .destroy = xs_destroy,
262ca07d 2652 .print_stats = xs_tcp_print_stats,
d67fa4d8
JL		 2653 .enable_swap = xs_enable_swap,
2654 .disable_swap = xs_disable_swap,
4a068258 2655 .inject_disconnect = xs_inject_disconnect,
a246b010
CL		 2656};
2657
4cfc7e60
RI		 2658/*
2659 * The rpc_xprt_ops for the server backchannel
2660 */
2661
2662static struct rpc_xprt_ops bc_tcp_ops = {
2663 .reserve_xprt = xprt_reserve_xprt,
2664 .release_xprt = xprt_release_xprt,
84e28a30 2665 .alloc_slot = xprt_alloc_slot,
4cfc7e60
RI		 2666 .buf_alloc = bc_malloc,
2667 .buf_free = bc_free,
2668 .send_request = bc_send_request,
2669 .set_retrans_timeout = xprt_set_retrans_timeout_def,
2670 .close = bc_close,
2671 .destroy = bc_destroy,
2672 .print_stats = xs_tcp_print_stats,
d67fa4d8
JL		 2673 .enable_swap = xs_enable_swap,
2674 .disable_swap = xs_disable_swap,
4a068258 2675 .inject_disconnect = xs_inject_disconnect,
4cfc7e60
RI		 2676};
2677
92476850
CL		 2678static int xs_init_anyaddr(const int family, struct sockaddr *sap)
2679{
2680 static const struct sockaddr_in sin = {
2681 .sin_family = AF_INET,
2682 .sin_addr.s_addr = htonl(INADDR_ANY),
2683 };
2684 static const struct sockaddr_in6 sin6 = {
2685 .sin6_family = AF_INET6,
2686 .sin6_addr = IN6ADDR_ANY_INIT,
2687 };
2688
2689 switch (family) {
176e21ee
CL		 2690 case AF_LOCAL:
2691 break;
92476850
CL		 2692 case AF_INET:
2693 memcpy(sap, &sin, sizeof(sin));
2694 break;
2695 case AF_INET6:
2696 memcpy(sap, &sin6, sizeof(sin6));
2697 break;
2698 default:
2699 dprintk("RPC: %s: Bad address family\n", __func__);
2700 return -EAFNOSUPPORT;
2701 }
2702 return 0;
2703}
2704
3c341b0b 2705static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
d9ba131d
TM		 2706 unsigned int slot_table_size,
2707 unsigned int max_slot_table_size)
c8541ecd
CL		 2708{
2709 struct rpc_xprt *xprt;
ffc2e518 2710 struct sock_xprt *new;
c8541ecd 2711
96802a09 2712 if (args->addrlen > sizeof(xprt->addr)) {
46121cf7 2713 dprintk("RPC: xs_setup_xprt: address too large\n");
c8541ecd
CL		 2714 return ERR_PTR(-EBADF);
2715 }
2716
d9ba131d
TM		 2717 xprt = xprt_alloc(args->net, sizeof(*new), slot_table_size,
2718 max_slot_table_size);
bd1722d4 2719 if (xprt == NULL) {
46121cf7
CL		 2720 dprintk("RPC: xs_setup_xprt: couldn't allocate "
2721 "rpc_xprt\n");
c8541ecd
CL		 2722 return ERR_PTR(-ENOMEM);
2723 }
c8541ecd 2724
bd1722d4 2725 new = container_of(xprt, struct sock_xprt, xprt);
edc1b01c 2726 mutex_init(&new->recv_mutex);
96802a09
FM		 2727 memcpy(&xprt->addr, args->dstaddr, args->addrlen);
2728 xprt->addrlen = args->addrlen;
d3bc9a1d 2729 if (args->srcaddr)
fbfffbd5 2730 memcpy(&new->srcaddr, args->srcaddr, args->addrlen);
92476850
CL		 2731 else {
2732 int err;
2733 err = xs_init_anyaddr(args->dstaddr->sa_family,
2734 (struct sockaddr *)&new->srcaddr);
2aa13531
SK		 2735 if (err != 0) {
2736 xprt_free(xprt);
92476850 2737 return ERR_PTR(err);
2aa13531 2738 }
92476850 2739 }
c8541ecd
CL		 2740
2741 return xprt;
2742}
2743
176e21ee
CL		 2744static const struct rpc_timeout xs_local_default_timeout = {
2745 .to_initval = 10 * HZ,
2746 .to_maxval = 10 * HZ,
2747 .to_retries = 2,
2748};
2749
2750/**
2751 * xs_setup_local - Set up transport to use an AF_LOCAL socket
2752 * @args: rpc transport creation arguments
2753 *
2754 * AF_LOCAL is a "tpi_cots_ord" transport, just like TCP
2755 */
2756static struct rpc_xprt *xs_setup_local(struct xprt_create *args)
2757{
2758 struct sockaddr_un *sun = (struct sockaddr_un *)args->dstaddr;
2759 struct sock_xprt *transport;
2760 struct rpc_xprt *xprt;
2761 struct rpc_xprt *ret;
2762
d9ba131d
TM		 2763 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2764 xprt_max_tcp_slot_table_entries);
176e21ee
CL		 2765 if (IS_ERR(xprt))
2766 return xprt;
2767 transport = container_of(xprt, struct sock_xprt, xprt);
2768
2769 xprt->prot = 0;
2770 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2771 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2772
2773 xprt->bind_timeout = XS_BIND_TO;
2774 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2775 xprt->idle_timeout = XS_IDLE_DISC_TO;
2776
2777 xprt->ops = &xs_local_ops;
2778 xprt->timeout = &xs_local_default_timeout;
2779
edc1b01c 2780 INIT_WORK(&transport->recv_worker, xs_dummy_data_receive_workfn);
93dc41bd
N		 2781 INIT_DELAYED_WORK(&transport->connect_worker,
2782 xs_dummy_setup_socket);
2783
176e21ee
CL		 2784 switch (sun->sun_family) {
2785 case AF_LOCAL:
2786 if (sun->sun_path[0] != '/') {
2787 dprintk("RPC: bad AF_LOCAL address: %s\n",
2788 sun->sun_path);
2789 ret = ERR_PTR(-EINVAL);
2790 goto out_err;
2791 }
2792 xprt_set_bound(xprt);
176e21ee 2793 xs_format_peer_addresses(xprt, "local", RPCBIND_NETID_LOCAL);
7073ea87
BF		 2794 ret = ERR_PTR(xs_local_setup_socket(transport));
2795 if (ret)
2796 goto out_err;
176e21ee
CL		 2797 break;
2798 default:
2799 ret = ERR_PTR(-EAFNOSUPPORT);
2800 goto out_err;
2801 }
2802
2803 dprintk("RPC: set up xprt to %s via AF_LOCAL\n",
2804 xprt->address_strings[RPC_DISPLAY_ADDR]);
2805
2806 if (try_module_get(THIS_MODULE))
2807 return xprt;
2808 ret = ERR_PTR(-EINVAL);
2809out_err:
315f3812 2810 xs_xprt_free(xprt);
176e21ee
CL		 2811 return ret;
2812}
2813
2881ae74
TM		 2814static const struct rpc_timeout xs_udp_default_timeout = {
2815 .to_initval = 5 * HZ,
2816 .to_maxval = 30 * HZ,
2817 .to_increment = 5 * HZ,
2818 .to_retries = 5,
2819};
2820
9903cd1c
CL		 2821/**
2822 * xs_setup_udp - Set up transport to use a UDP socket
96802a09 2823 * @args: rpc transport creation arguments
9903cd1c
CL		 2824 *
2825 */
483066d6 2826static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
a246b010 2827{
8f9d5b1a 2828 struct sockaddr *addr = args->dstaddr;
c8541ecd 2829 struct rpc_xprt *xprt;
c8475461 2830 struct sock_xprt *transport;
0a68b0be 2831 struct rpc_xprt *ret;
a246b010 2832
d9ba131d
TM		 2833 xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries,
2834 xprt_udp_slot_table_entries);
c8541ecd
CL		 2835 if (IS_ERR(xprt))
2836 return xprt;
c8475461 2837 transport = container_of(xprt, struct sock_xprt, xprt);
a246b010 2838
ec739ef0 2839 xprt->prot = IPPROTO_UDP;
808012fb 2840 xprt->tsh_size = 0;
a246b010
CL		 2841 /* XXX: header size can vary due to auth type, IPv6, etc. */
2842 xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
2843
03bf4b70 2844 xprt->bind_timeout = XS_BIND_TO;
03bf4b70
CL		 2845 xprt->reestablish_timeout = XS_UDP_REEST_TO;
2846 xprt->idle_timeout = XS_IDLE_DISC_TO;
a246b010 2847
262965f5 2848 xprt->ops = &xs_udp_ops;
a246b010 2849
ba7392bb 2850 xprt->timeout = &xs_udp_default_timeout;
a246b010 2851
f9b2ee71 2852 INIT_WORK(&transport->recv_worker, xs_udp_data_receive_workfn);
edc1b01c
TM		 2853 INIT_DELAYED_WORK(&transport->connect_worker, xs_udp_setup_socket);
2854
8f9d5b1a
CL		 2855 switch (addr->sa_family) {
2856 case AF_INET:
2857 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2858 xprt_set_bound(xprt);
2859
9dc3b095 2860 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
8f9d5b1a
CL		 2861 break;
2862 case AF_INET6:
2863 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2864 xprt_set_bound(xprt);
2865
9dc3b095 2866 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
8f9d5b1a
CL		 2867 break;
2868 default:
0a68b0be
BF		 2869 ret = ERR_PTR(-EAFNOSUPPORT);
2870 goto out_err;
8f9d5b1a
CL		 2871 }
2872
c740eff8
CL		 2873 if (xprt_bound(xprt))
2874 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2875 xprt->address_strings[RPC_DISPLAY_ADDR],
2876 xprt->address_strings[RPC_DISPLAY_PORT],
2877 xprt->address_strings[RPC_DISPLAY_PROTO]);
2878 else
2879 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
2880 xprt->address_strings[RPC_DISPLAY_ADDR],
2881 xprt->address_strings[RPC_DISPLAY_PROTO]);
edb267a6 2882
bc25571e
TT		 2883 if (try_module_get(THIS_MODULE))
2884 return xprt;
0a68b0be
BF		 2885 ret = ERR_PTR(-EINVAL);
2886out_err:
315f3812 2887 xs_xprt_free(xprt);
0a68b0be 2888 return ret;
a246b010
CL		 2889}
2890
2881ae74
TM		 2891static const struct rpc_timeout xs_tcp_default_timeout = {
2892 .to_initval = 60 * HZ,
2893 .to_maxval = 60 * HZ,
2894 .to_retries = 2,
2895};
2896
9903cd1c
CL		 2897/**
2898 * xs_setup_tcp - Set up transport to use a TCP socket
96802a09 2899 * @args: rpc transport creation arguments
9903cd1c
CL		 2900 *
2901 */
483066d6 2902static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
a246b010 2903{
8f9d5b1a 2904 struct sockaddr *addr = args->dstaddr;
c8541ecd 2905 struct rpc_xprt *xprt;
c8475461 2906 struct sock_xprt *transport;
0a68b0be 2907 struct rpc_xprt *ret;
b7993ceb
TM		 2908 unsigned int max_slot_table_size = xprt_max_tcp_slot_table_entries;
2909
2910 if (args->flags & XPRT_CREATE_INFINITE_SLOTS)
2911 max_slot_table_size = RPC_MAX_SLOT_TABLE_LIMIT;
a246b010 2912
d9ba131d 2913 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
b7993ceb 2914 max_slot_table_size);
c8541ecd
CL		 2915 if (IS_ERR(xprt))
2916 return xprt;
c8475461 2917 transport = container_of(xprt, struct sock_xprt, xprt);
a246b010 2918
ec739ef0 2919 xprt->prot = IPPROTO_TCP;
808012fb 2920 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
808012fb 2921 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
a246b010 2922
03bf4b70 2923 xprt->bind_timeout = XS_BIND_TO;
03bf4b70
CL		 2924 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2925 xprt->idle_timeout = XS_IDLE_DISC_TO;
a246b010 2926
262965f5 2927 xprt->ops = &xs_tcp_ops;
ba7392bb 2928 xprt->timeout = &xs_tcp_default_timeout;
a246b010 2929
edc1b01c
TM		 2930 INIT_WORK(&transport->recv_worker, xs_tcp_data_receive_workfn);
2931 INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_setup_socket);
2932
8f9d5b1a
CL		 2933 switch (addr->sa_family) {
2934 case AF_INET:
2935 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2936 xprt_set_bound(xprt);
2937
9dc3b095 2938 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
8f9d5b1a
CL		 2939 break;
2940 case AF_INET6:
2941 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2942 xprt_set_bound(xprt);
2943
9dc3b095 2944 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
8f9d5b1a
CL		 2945 break;
2946 default:
0a68b0be
BF		 2947 ret = ERR_PTR(-EAFNOSUPPORT);
2948 goto out_err;
8f9d5b1a
CL		 2949 }
2950
c740eff8
CL		 2951 if (xprt_bound(xprt))
2952 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2953 xprt->address_strings[RPC_DISPLAY_ADDR],
2954 xprt->address_strings[RPC_DISPLAY_PORT],
2955 xprt->address_strings[RPC_DISPLAY_PROTO]);
2956 else
2957 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
2958 xprt->address_strings[RPC_DISPLAY_ADDR],
2959 xprt->address_strings[RPC_DISPLAY_PROTO]);
2960
bc25571e
TT		 2961 if (try_module_get(THIS_MODULE))
2962 return xprt;
0a68b0be
BF		 2963 ret = ERR_PTR(-EINVAL);
2964out_err:
315f3812 2965 xs_xprt_free(xprt);
0a68b0be 2966 return ret;
a246b010 2967}
282b32e1 2968
f300baba
AB		 2969/**
2970 * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket
2971 * @args: rpc transport creation arguments
2972 *
2973 */
2974static struct rpc_xprt *xs_setup_bc_tcp(struct xprt_create *args)
2975{
2976 struct sockaddr *addr = args->dstaddr;
2977 struct rpc_xprt *xprt;
2978 struct sock_xprt *transport;
2979 struct svc_sock *bc_sock;
0a68b0be 2980 struct rpc_xprt *ret;
f300baba 2981
d9ba131d
TM		 2982 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2983 xprt_tcp_slot_table_entries);
f300baba
AB		 2984 if (IS_ERR(xprt))
2985 return xprt;
2986 transport = container_of(xprt, struct sock_xprt, xprt);
2987
2988 xprt->prot = IPPROTO_TCP;
2989 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2990 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2991 xprt->timeout = &xs_tcp_default_timeout;
2992
2993 /* backchannel */
2994 xprt_set_bound(xprt);
2995 xprt->bind_timeout = 0;
f300baba
AB		 2996 xprt->reestablish_timeout = 0;
2997 xprt->idle_timeout = 0;
2998
f300baba
AB		 2999 xprt->ops = &bc_tcp_ops;
3000
3001 switch (addr->sa_family) {
3002 case AF_INET:
3003 xs_format_peer_addresses(xprt, "tcp",
3004 RPCBIND_NETID_TCP);
3005 break;
3006 case AF_INET6:
3007 xs_format_peer_addresses(xprt, "tcp",
3008 RPCBIND_NETID_TCP6);
3009 break;
3010 default:
0a68b0be
BF		 3011 ret = ERR_PTR(-EAFNOSUPPORT);
3012 goto out_err;
f300baba
AB		 3013 }
3014
50fa0d40
PE		 3015 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
3016 xprt->address_strings[RPC_DISPLAY_ADDR],
3017 xprt->address_strings[RPC_DISPLAY_PORT],
3018 xprt->address_strings[RPC_DISPLAY_PROTO]);
f300baba 3019
99de8ea9
BF		 3020 /*
3021 * Once we've associated a backchannel xprt with a connection,
28303ca3
WM		 3022 * we want to keep it around as long as the connection lasts,
3023 * in case we need to start using it for a backchannel again;
3024 * this reference won't be dropped until bc_xprt is destroyed.
99de8ea9
BF		 3025 */
3026 xprt_get(xprt);
3027 args->bc_xprt->xpt_bc_xprt = xprt;
3028 xprt->bc_xprt = args->bc_xprt;
3029 bc_sock = container_of(args->bc_xprt, struct svc_sock, sk_xprt);
3030 transport->sock = bc_sock->sk_sock;
3031 transport->inet = bc_sock->sk_sk;
3032
f300baba
AB		 3033 /*
3034 * Since we don't want connections for the backchannel, we set
3035 * the xprt status to connected
3036 */
3037 xprt_set_connected(xprt);
3038
f300baba
AB		 3039 if (try_module_get(THIS_MODULE))
3040 return xprt;
642aab58
KM		 3041
3042 args->bc_xprt->xpt_bc_xprt = NULL;
99de8ea9 3043 xprt_put(xprt);
0a68b0be
BF		 3044 ret = ERR_PTR(-EINVAL);
3045out_err:
315f3812 3046 xs_xprt_free(xprt);
0a68b0be 3047 return ret;
f300baba
AB		 3048}
3049
176e21ee
CL		 3050static struct xprt_class xs_local_transport = {
3051 .list = LIST_HEAD_INIT(xs_local_transport.list),
3052 .name = "named UNIX socket",
3053 .owner = THIS_MODULE,
3054 .ident = XPRT_TRANSPORT_LOCAL,
3055 .setup = xs_setup_local,
3056};
3057
bc25571e
TT		 3058static struct xprt_class xs_udp_transport = {
3059 .list = LIST_HEAD_INIT(xs_udp_transport.list),
3060 .name = "udp",
3061 .owner = THIS_MODULE,
f300baba 3062 .ident = XPRT_TRANSPORT_UDP,
bc25571e
TT		 3063 .setup = xs_setup_udp,
3064};
3065
3066static struct xprt_class xs_tcp_transport = {
3067 .list = LIST_HEAD_INIT(xs_tcp_transport.list),
3068 .name = "tcp",
3069 .owner = THIS_MODULE,
f300baba 3070 .ident = XPRT_TRANSPORT_TCP,
bc25571e
TT		 3071 .setup = xs_setup_tcp,
3072};
3073
f300baba
AB		 3074static struct xprt_class xs_bc_tcp_transport = {
3075 .list = LIST_HEAD_INIT(xs_bc_tcp_transport.list),
3076 .name = "tcp NFSv4.1 backchannel",
3077 .owner = THIS_MODULE,
3078 .ident = XPRT_TRANSPORT_BC_TCP,
3079 .setup = xs_setup_bc_tcp,
3080};
3081
282b32e1 3082/**
bc25571e 3083 * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
282b32e1
CL		 3084 *
3085 */
3086int init_socket_xprt(void)
3087{
f895b252 3088#if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
2b1bec5f 3089 if (!sunrpc_table_header)
0b4d4147 3090 sunrpc_table_header = register_sysctl_table(sunrpc_table);
fbf76683
CL		 3091#endif
3092
176e21ee 3093 xprt_register_transport(&xs_local_transport);
bc25571e
TT		 3094 xprt_register_transport(&xs_udp_transport);
3095 xprt_register_transport(&xs_tcp_transport);
f300baba 3096 xprt_register_transport(&xs_bc_tcp_transport);
bc25571e 3097
282b32e1
CL		 3098 return 0;
3099}
3100
3101/**
bc25571e 3102 * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
282b32e1
CL		 3103 *
3104 */
3105void cleanup_socket_xprt(void)
3106{
f895b252 3107#if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
fbf76683
CL		 3108 if (sunrpc_table_header) {
3109 unregister_sysctl_table(sunrpc_table_header);
3110 sunrpc_table_header = NULL;
3111 }
3112#endif
bc25571e 3113
176e21ee 3114 xprt_unregister_transport(&xs_local_transport);
bc25571e
TT		 3115 xprt_unregister_transport(&xs_udp_transport);
3116 xprt_unregister_transport(&xs_tcp_transport);
f300baba 3117 xprt_unregister_transport(&xs_bc_tcp_transport);
282b32e1 3118}
cbf11071 3119
9bbb9e5a
RR		 3120static int param_set_uint_minmax(const char *val,
3121 const struct kernel_param *kp,
cbf11071
TM		 3122 unsigned int min, unsigned int max)
3123{
00cfaa94 3124 unsigned int num;
cbf11071
TM		 3125 int ret;
3126
3127 if (!val)
3128 return -EINVAL;
00cfaa94 3129 ret = kstrtouint(val, 0, &num);
cbf11071
TM		 3130 if (ret == -EINVAL || num < min || num > max)
3131 return -EINVAL;
3132 *((unsigned int *)kp->arg) = num;
3133 return 0;
3134}
3135
9bbb9e5a 3136static int param_set_portnr(const char *val, const struct kernel_param *kp)
cbf11071
TM		 3137{
3138 return param_set_uint_minmax(val, kp,
3139 RPC_MIN_RESVPORT,
3140 RPC_MAX_RESVPORT);
3141}
3142
9c27847d 3143static const struct kernel_param_ops param_ops_portnr = {
9bbb9e5a
RR		 3144 .set = param_set_portnr,
3145 .get = param_get_uint,
3146};
3147
cbf11071
TM		 3148#define param_check_portnr(name, p) \
3149 __param_check(name, p, unsigned int);
3150
3151module_param_named(min_resvport, xprt_min_resvport, portnr, 0644);
3152module_param_named(max_resvport, xprt_max_resvport, portnr, 0644);
3153
9bbb9e5a
RR		 3154static int param_set_slot_table_size(const char *val,
3155 const struct kernel_param *kp)
cbf11071
TM		 3156{
3157 return param_set_uint_minmax(val, kp,
3158 RPC_MIN_SLOT_TABLE,
3159 RPC_MAX_SLOT_TABLE);
3160}
3161
9c27847d 3162static const struct kernel_param_ops param_ops_slot_table_size = {
9bbb9e5a
RR		 3163 .set = param_set_slot_table_size,
3164 .get = param_get_uint,
3165};
3166
cbf11071
TM		 3167#define param_check_slot_table_size(name, p) \
3168 __param_check(name, p, unsigned int);
3169
d9ba131d
TM		 3170static int param_set_max_slot_table_size(const char *val,
3171 const struct kernel_param *kp)
3172{
3173 return param_set_uint_minmax(val, kp,
3174 RPC_MIN_SLOT_TABLE,
3175 RPC_MAX_SLOT_TABLE_LIMIT);
3176}
3177
9c27847d 3178static const struct kernel_param_ops param_ops_max_slot_table_size = {
d9ba131d
TM		 3179 .set = param_set_max_slot_table_size,
3180 .get = param_get_uint,
3181};
3182
3183#define param_check_max_slot_table_size(name, p) \
3184 __param_check(name, p, unsigned int);
3185
cbf11071
TM		 3186module_param_named(tcp_slot_table_entries, xprt_tcp_slot_table_entries,
3187 slot_table_size, 0644);
d9ba131d
TM		 3188module_param_named(tcp_max_slot_table_entries, xprt_max_tcp_slot_table_entries,
3189 max_slot_table_size, 0644);
cbf11071
TM		 3190module_param_named(udp_slot_table_entries, xprt_udp_slot_table_entries,
3191 slot_table_size, 0644);
3192
Linux 6.x block layer and io_uring tree(s)