projects / linux-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
pNFS: Actively set attributes as invalid if LAYOUTCOMMIT is outstanding
[linux-block.git] / net / sunrpc / xprtsock.c
CommitLineData
a246b010
CL		 1/*
2 * linux/net/sunrpc/xprtsock.c
3 *
4 * Client-side transport implementation for sockets.
5 *
113aa838
AC		 6 * TCP callback races fixes (C) 1998 Red Hat
7 * TCP send fixes (C) 1998 Red Hat
a246b010
CL		 8 * TCP NFS related read + write fixes
9 * (C) 1999 Dave Airlie, University of Limerick, Ireland <airlied@linux.ie>
10 *
11 * Rewrite of larges part of the code in order to stabilize TCP stuff.
12 * Fix behaviour when socket buffer is full.
13 * (C) 1999 Trond Myklebust <trond.myklebust@fys.uio.no>
55aa4f58
CL		 14 *
15 * IP socket transport implementation, (C) 2005 Chuck Lever <cel@netapp.com>
8f9d5b1a
CL		 16 *
17 * IPv6 support contributed by Gilles Quillard, Bull Open Source, 2005.
18 * <gilles.quillard@bull.net>
a246b010
CL		 19 */
20
21#include <linux/types.h>
176e21ee 22#include <linux/string.h>
a246b010 23#include <linux/slab.h>
bc25571e 24#include <linux/module.h>
a246b010 25#include <linux/capability.h>
a246b010
CL		 26#include <linux/pagemap.h>
27#include <linux/errno.h>
28#include <linux/socket.h>
29#include <linux/in.h>
30#include <linux/net.h>
31#include <linux/mm.h>
176e21ee 32#include <linux/un.h>
a246b010
CL		 33#include <linux/udp.h>
34#include <linux/tcp.h>
35#include <linux/sunrpc/clnt.h>
5976687a 36#include <linux/sunrpc/addr.h>
02107148 37#include <linux/sunrpc/sched.h>
4cfc7e60 38#include <linux/sunrpc/svcsock.h>
49c36fcc 39#include <linux/sunrpc/xprtsock.h>
a246b010 40#include <linux/file.h>
9e00abc3 41#ifdef CONFIG_SUNRPC_BACKCHANNEL
44b98efd
RL		 42#include <linux/sunrpc/bc_xprt.h>
43#endif
a246b010
CL		 44
45#include <net/sock.h>
46#include <net/checksum.h>
47#include <net/udp.h>
48#include <net/tcp.h>
49
40b5ea0c
TM		 50#include <trace/events/sunrpc.h>
51
4cfc7e60 52#include "sunrpc.h"
176e21ee
CL		 53
54static void xs_close(struct rpc_xprt *xprt);
55
c556b754
CL		 56/*
57 * xprtsock tunables
58 */
09acfea5
TM		 59static unsigned int xprt_udp_slot_table_entries = RPC_DEF_SLOT_TABLE;
60static unsigned int xprt_tcp_slot_table_entries = RPC_MIN_SLOT_TABLE;
61static unsigned int xprt_max_tcp_slot_table_entries = RPC_MAX_SLOT_TABLE;
c556b754 62
09acfea5
TM		 63static unsigned int xprt_min_resvport = RPC_DEF_MIN_RESVPORT;
64static unsigned int xprt_max_resvport = RPC_DEF_MAX_RESVPORT;
c556b754 65
54c09874
TM		 66#if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
67
7d1e8255 68#define XS_TCP_LINGER_TO (15U * HZ)
25fe6142 69static unsigned int xs_tcp_fin_timeout __read_mostly = XS_TCP_LINGER_TO;
7d1e8255 70
fbf76683
CL		 71/*
72 * We can register our own files under /proc/sys/sunrpc by
73 * calling register_sysctl_table() again. The files in that
74 * directory become the union of all files registered there.
75 *
76 * We simply need to make sure that we don't collide with
77 * someone else's file names!
78 */
79
fbf76683
CL		 80static unsigned int min_slot_table_size = RPC_MIN_SLOT_TABLE;
81static unsigned int max_slot_table_size = RPC_MAX_SLOT_TABLE;
d9ba131d 82static unsigned int max_tcp_slot_table_limit = RPC_MAX_SLOT_TABLE_LIMIT;
fbf76683
CL		 83static unsigned int xprt_min_resvport_limit = RPC_MIN_RESVPORT;
84static unsigned int xprt_max_resvport_limit = RPC_MAX_RESVPORT;
85
86static struct ctl_table_header *sunrpc_table_header;
87
88/*
89 * FIXME: changing the UDP slot table size should also resize the UDP
90 * socket buffers for existing UDP transports
91 */
fe2c6338 92static struct ctl_table xs_tunables_table[] = {
fbf76683 93 {
fbf76683
CL		 94 .procname = "udp_slot_table_entries",
95 .data = &xprt_udp_slot_table_entries,
96 .maxlen = sizeof(unsigned int),
97 .mode = 0644,
6d456111 98 .proc_handler = proc_dointvec_minmax,
fbf76683
CL		 99 .extra1 = &min_slot_table_size,
100 .extra2 = &max_slot_table_size
101 },
102 {
fbf76683
CL		 103 .procname = "tcp_slot_table_entries",
104 .data = &xprt_tcp_slot_table_entries,
105 .maxlen = sizeof(unsigned int),
106 .mode = 0644,
6d456111 107 .proc_handler = proc_dointvec_minmax,
fbf76683
CL		 108 .extra1 = &min_slot_table_size,
109 .extra2 = &max_slot_table_size
110 },
d9ba131d
TM		 111 {
112 .procname = "tcp_max_slot_table_entries",
113 .data = &xprt_max_tcp_slot_table_entries,
114 .maxlen = sizeof(unsigned int),
115 .mode = 0644,
116 .proc_handler = proc_dointvec_minmax,
117 .extra1 = &min_slot_table_size,
118 .extra2 = &max_tcp_slot_table_limit
119 },
fbf76683 120 {
fbf76683
CL		 121 .procname = "min_resvport",
122 .data = &xprt_min_resvport,
123 .maxlen = sizeof(unsigned int),
124 .mode = 0644,
6d456111 125 .proc_handler = proc_dointvec_minmax,
fbf76683 126 .extra1 = &xprt_min_resvport_limit,
e08ea3a9 127 .extra2 = &xprt_max_resvport
fbf76683
CL		 128 },
129 {
fbf76683
CL		 130 .procname = "max_resvport",
131 .data = &xprt_max_resvport,
132 .maxlen = sizeof(unsigned int),
133 .mode = 0644,
6d456111 134 .proc_handler = proc_dointvec_minmax,
e08ea3a9 135 .extra1 = &xprt_min_resvport,
fbf76683
CL		 136 .extra2 = &xprt_max_resvport_limit
137 },
25fe6142
TM		 138 {
139 .procname = "tcp_fin_timeout",
140 .data = &xs_tcp_fin_timeout,
141 .maxlen = sizeof(xs_tcp_fin_timeout),
142 .mode = 0644,
6d456111 143 .proc_handler = proc_dointvec_jiffies,
fbf76683 144 },
f8572d8f 145 { },
fbf76683
CL		 146};
147
fe2c6338 148static struct ctl_table sunrpc_table[] = {
fbf76683 149 {
fbf76683
CL		 150 .procname = "sunrpc",
151 .mode = 0555,
152 .child = xs_tunables_table
153 },
f8572d8f 154 { },
fbf76683
CL		 155};
156
157#endif
158
03bf4b70
CL		 159/*
160 * Wait duration for a reply from the RPC portmapper.
161 */
162#define XS_BIND_TO (60U * HZ)
163
164/*
165 * Delay if a UDP socket connect error occurs. This is most likely some
166 * kind of resource problem on the local host.
167 */
168#define XS_UDP_REEST_TO (2U * HZ)
169
170/*
171 * The reestablish timeout allows clients to delay for a bit before attempting
172 * to reconnect to a server that just dropped our connection.
173 *
174 * We implement an exponential backoff when trying to reestablish a TCP
175 * transport connection with the server. Some servers like to drop a TCP
176 * connection when they are overworked, so we start with a short timeout and
177 * increase over time if the server is down or not responding.
178 */
179#define XS_TCP_INIT_REEST_TO (3U * HZ)
180#define XS_TCP_MAX_REEST_TO (5U * 60 * HZ)
181
182/*
183 * TCP idle timeout; client drops the transport socket if it is idle
184 * for this long. Note that we also timeout UDP sockets to prevent
185 * holding port numbers when there is no RPC traffic.
186 */
187#define XS_IDLE_DISC_TO (5U * 60 * HZ)
188
f895b252 189#if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
a246b010 190# undef RPC_DEBUG_DATA
9903cd1c 191# define RPCDBG_FACILITY RPCDBG_TRANS
a246b010
CL		 192#endif
193
a246b010 194#ifdef RPC_DEBUG_DATA
9903cd1c 195static void xs_pktdump(char *msg, u32 *packet, unsigned int count)
a246b010 196{
9903cd1c
CL		 197 u8 *buf = (u8 *) packet;
198 int j;
a246b010 199
46121cf7 200 dprintk("RPC: %s\n", msg);
a246b010
CL		 201 for (j = 0; j < count && j < 128; j += 4) {
202 if (!(j & 31)) {
203 if (j)
204 dprintk("\n");
205 dprintk("0x%04x ", j);
206 }
207 dprintk("%02x%02x%02x%02x ",
208 buf[j], buf[j+1], buf[j+2], buf[j+3]);
209 }
210 dprintk("\n");
211}
212#else
9903cd1c 213static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count)
a246b010
CL		 214{
215 /* NOP */
216}
217#endif
218
2118071d
TM		 219static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
220{
221 return (struct rpc_xprt *) sk->sk_user_data;
222}
223
95392c59
CL		 224static inline struct sockaddr *xs_addr(struct rpc_xprt *xprt)
225{
226 return (struct sockaddr *) &xprt->addr;
227}
228
176e21ee
CL		 229static inline struct sockaddr_un *xs_addr_un(struct rpc_xprt *xprt)
230{
231 return (struct sockaddr_un *) &xprt->addr;
232}
233
95392c59 234static inline struct sockaddr_in *xs_addr_in(struct rpc_xprt *xprt)
edb267a6 235{
95392c59
CL		 236 return (struct sockaddr_in *) &xprt->addr;
237}
238
239static inline struct sockaddr_in6 *xs_addr_in6(struct rpc_xprt *xprt)
240{
241 return (struct sockaddr_in6 *) &xprt->addr;
242}
243
c877b849 244static void xs_format_common_peer_addresses(struct rpc_xprt *xprt)
edb267a6 245{
c877b849 246 struct sockaddr *sap = xs_addr(xprt);
9dc3b095
CL		 247 struct sockaddr_in6 *sin6;
248 struct sockaddr_in *sin;
176e21ee 249 struct sockaddr_un *sun;
c877b849 250 char buf[128];
edb267a6 251
9dc3b095 252 switch (sap->sa_family) {
176e21ee
CL		 253 case AF_LOCAL:
254 sun = xs_addr_un(xprt);
255 strlcpy(buf, sun->sun_path, sizeof(buf));
256 xprt->address_strings[RPC_DISPLAY_ADDR] =
257 kstrdup(buf, GFP_KERNEL);
258 break;
9dc3b095 259 case AF_INET:
176e21ee
CL		 260 (void)rpc_ntop(sap, buf, sizeof(buf));
261 xprt->address_strings[RPC_DISPLAY_ADDR] =
262 kstrdup(buf, GFP_KERNEL);
9dc3b095 263 sin = xs_addr_in(xprt);
fc0b5791 264 snprintf(buf, sizeof(buf), "%08x", ntohl(sin->sin_addr.s_addr));
9dc3b095
CL		 265 break;
266 case AF_INET6:
176e21ee
CL		 267 (void)rpc_ntop(sap, buf, sizeof(buf));
268 xprt->address_strings[RPC_DISPLAY_ADDR] =
269 kstrdup(buf, GFP_KERNEL);
9dc3b095 270 sin6 = xs_addr_in6(xprt);
fc0b5791 271 snprintf(buf, sizeof(buf), "%pi6", &sin6->sin6_addr);
9dc3b095
CL		 272 break;
273 default:
274 BUG();
756805e7 275 }
176e21ee 276
9dc3b095 277 xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = kstrdup(buf, GFP_KERNEL);
edb267a6
CL		 278}
279
9dc3b095 280static void xs_format_common_peer_ports(struct rpc_xprt *xprt)
4b6473fb 281{
9dc3b095
CL		 282 struct sockaddr *sap = xs_addr(xprt);
283 char buf[128];
4b6473fb 284
81160e66 285 snprintf(buf, sizeof(buf), "%u", rpc_get_port(sap));
c877b849 286 xprt->address_strings[RPC_DISPLAY_PORT] = kstrdup(buf, GFP_KERNEL);
4b6473fb 287
81160e66 288 snprintf(buf, sizeof(buf), "%4hx", rpc_get_port(sap));
c877b849
CL		 289 xprt->address_strings[RPC_DISPLAY_HEX_PORT] = kstrdup(buf, GFP_KERNEL);
290}
4b6473fb 291
9dc3b095
CL		 292static void xs_format_peer_addresses(struct rpc_xprt *xprt,
293 const char *protocol,
294 const char *netid)
c877b849 295{
b454ae90 296 xprt->address_strings[RPC_DISPLAY_PROTO] = protocol;
b454ae90 297 xprt->address_strings[RPC_DISPLAY_NETID] = netid;
c877b849 298 xs_format_common_peer_addresses(xprt);
9dc3b095 299 xs_format_common_peer_ports(xprt);
edb267a6 300}
4b6473fb 301
9dc3b095 302static void xs_update_peer_port(struct rpc_xprt *xprt)
4b6473fb 303{
9dc3b095
CL		 304 kfree(xprt->address_strings[RPC_DISPLAY_HEX_PORT]);
305 kfree(xprt->address_strings[RPC_DISPLAY_PORT]);
4417c8c4 306
9dc3b095 307 xs_format_common_peer_ports(xprt);
edb267a6
CL		 308}
309
310static void xs_free_peer_addresses(struct rpc_xprt *xprt)
311{
33e01dc7
CL		 312 unsigned int i;
313
314 for (i = 0; i < RPC_DISPLAY_MAX; i++)
315 switch (i) {
316 case RPC_DISPLAY_PROTO:
317 case RPC_DISPLAY_NETID:
318 continue;
319 default:
320 kfree(xprt->address_strings[i]);
321 }
edb267a6
CL		 322}
323
b4b5cc85
CL		 324#define XS_SENDMSG_FLAGS (MSG_DONTWAIT | MSG_NOSIGNAL)
325
24c5684b 326static int xs_send_kvec(struct socket *sock, struct sockaddr *addr, int addrlen, struct kvec *vec, unsigned int base, int more)
b4b5cc85 327{
b4b5cc85
CL		 328 struct msghdr msg = {
329 .msg_name = addr,
330 .msg_namelen = addrlen,
24c5684b
TM		 331 .msg_flags = XS_SENDMSG_FLAGS | (more ? MSG_MORE : 0),
332 };
333 struct kvec iov = {
334 .iov_base = vec->iov_base + base,
335 .iov_len = vec->iov_len - base,
b4b5cc85
CL		 336 };
337
24c5684b 338 if (iov.iov_len != 0)
b4b5cc85
CL		 339 return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
340 return kernel_sendmsg(sock, &msg, NULL, 0, 0);
341}
342
f279cd00 343static int xs_send_pagedata(struct socket *sock, struct xdr_buf *xdr, unsigned int base, int more, bool zerocopy, int *sent_p)
b4b5cc85 344{
a6b31d18
TM		 345 ssize_t (*do_sendpage)(struct socket *sock, struct page *page,
346 int offset, size_t size, int flags);
24c5684b
TM		 347 struct page **ppage;
348 unsigned int remainder;
f279cd00 349 int err;
24c5684b
TM		 350
351 remainder = xdr->page_len - base;
352 base += xdr->page_base;
353 ppage = xdr->pages + (base >> PAGE_SHIFT);
354 base &= ~PAGE_MASK;
a6b31d18
TM		 355 do_sendpage = sock->ops->sendpage;
356 if (!zerocopy)
357 do_sendpage = sock_no_sendpage;
24c5684b
TM		 358 for(;;) {
359 unsigned int len = min_t(unsigned int, PAGE_SIZE - base, remainder);
360 int flags = XS_SENDMSG_FLAGS;
b4b5cc85 361
24c5684b 362 remainder -= len;
31303d6c 363 if (more)
24c5684b 364 flags |= MSG_MORE;
31303d6c
TM		 365 if (remainder != 0)
366 flags |= MSG_SENDPAGE_NOTLAST | MSG_MORE;
a6b31d18 367 err = do_sendpage(sock, *ppage, base, len, flags);
24c5684b
TM		 368 if (remainder == 0 || err != len)
369 break;
f279cd00 370 *sent_p += err;
24c5684b
TM		 371 ppage++;
372 base = 0;
373 }
f279cd00
JB		 374 if (err > 0) {
375 *sent_p += err;
376 err = 0;
377 }
378 return err;
b4b5cc85
CL		 379}
380
9903cd1c
CL		 381/**
382 * xs_sendpages - write pages directly to a socket
383 * @sock: socket to send on
384 * @addr: UDP only -- address of destination
385 * @addrlen: UDP only -- length of destination address
386 * @xdr: buffer containing this request
387 * @base: starting position in the buffer
a6b31d18 388 * @zerocopy: true if it is safe to use sendpage()
f279cd00 389 * @sent_p: return the total number of bytes successfully queued for sending
9903cd1c 390 *
a246b010 391 */
f279cd00 392static int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base, bool zerocopy, int *sent_p)
a246b010 393{
24c5684b 394 unsigned int remainder = xdr->len - base;
f279cd00
JB		 395 int err = 0;
396 int sent = 0;
a246b010 397
262965f5 398 if (unlikely(!sock))
fba91afb 399 return -ENOTSOCK;
262965f5 400
24c5684b
TM		 401 if (base != 0) {
402 addr = NULL;
403 addrlen = 0;
404 }
262965f5 405
24c5684b
TM		 406 if (base < xdr->head[0].iov_len || addr != NULL) {
407 unsigned int len = xdr->head[0].iov_len - base;
408 remainder -= len;
409 err = xs_send_kvec(sock, addr, addrlen, &xdr->head[0], base, remainder != 0);
410 if (remainder == 0 || err != len)
a246b010 411 goto out;
f279cd00 412 *sent_p += err;
a246b010
CL		 413 base = 0;
414 } else
24c5684b 415 base -= xdr->head[0].iov_len;
a246b010 416
24c5684b
TM		 417 if (base < xdr->page_len) {
418 unsigned int len = xdr->page_len - base;
419 remainder -= len;
f279cd00
JB		 420 err = xs_send_pagedata(sock, xdr, base, remainder != 0, zerocopy, &sent);
421 *sent_p += sent;
422 if (remainder == 0 || sent != len)
a246b010
CL		 423 goto out;
424 base = 0;
24c5684b
TM		 425 } else
426 base -= xdr->page_len;
427
428 if (base >= xdr->tail[0].iov_len)
f279cd00 429 return 0;
24c5684b 430 err = xs_send_kvec(sock, NULL, 0, &xdr->tail[0], base, 0);
a246b010 431out:
f279cd00
JB		 432 if (err > 0) {
433 *sent_p += err;
434 err = 0;
435 }
436 return err;
a246b010
CL		 437}
438
b6ddf64f
TM		 439static void xs_nospace_callback(struct rpc_task *task)
440{
441 struct sock_xprt *transport = container_of(task->tk_rqstp->rq_xprt, struct sock_xprt, xprt);
442
443 transport->inet->sk_write_pending--;
b6ddf64f
TM		 444}
445
9903cd1c 446/**
262965f5
CL		 447 * xs_nospace - place task on wait queue if transmit was incomplete
448 * @task: task to put to sleep
9903cd1c 449 *
a246b010 450 */
5e3771ce 451static int xs_nospace(struct rpc_task *task)
a246b010 452{
262965f5
CL		 453 struct rpc_rqst *req = task->tk_rqstp;
454 struct rpc_xprt *xprt = req->rq_xprt;
ee0ac0c2 455 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
06ea0bfe 456 struct sock *sk = transport->inet;
24ca9a84 457 int ret = -EAGAIN;
a246b010 458
46121cf7 459 dprintk("RPC: %5u xmit incomplete (%u left of %u)\n",
262965f5
CL		 460 task->tk_pid, req->rq_slen - req->rq_bytes_sent,
461 req->rq_slen);
462
b6ddf64f
TM		 463 /* Protect against races with write_space */
464 spin_lock_bh(&xprt->transport_lock);
465
466 /* Don't race with disconnect */
467 if (xprt_connected(xprt)) {
13331a55
TM		 468 /* wait for more buffer space */
469 sk->sk_write_pending++;
470 xprt_wait_for_buffer_space(task, xs_nospace_callback);
471 } else
5e3771ce 472 ret = -ENOTCONN;
262965f5 473
b6ddf64f 474 spin_unlock_bh(&xprt->transport_lock);
06ea0bfe
TM		 475
476 /* Race breaker in case memory is freed before above code is called */
477 sk->sk_write_space(sk);
5e3771ce 478 return ret;
262965f5
CL		 479}
480
61677eee
CL		 481/*
482 * Construct a stream transport record marker in @buf.
483 */
484static inline void xs_encode_stream_record_marker(struct xdr_buf *buf)
485{
486 u32 reclen = buf->len - sizeof(rpc_fraghdr);
487 rpc_fraghdr *base = buf->head[0].iov_base;
488 *base = cpu_to_be32(RPC_LAST_STREAM_FRAGMENT | reclen);
489}
490
176e21ee
CL		 491/**
492 * xs_local_send_request - write an RPC request to an AF_LOCAL socket
493 * @task: RPC task that manages the state of an RPC request
494 *
495 * Return values:
496 * 0: The request has been sent
497 * EAGAIN: The socket was blocked, please call again later to
498 * complete the request
499 * ENOTCONN: Caller needs to invoke connect logic then call again
500 * other: Some other error occured, the request was not sent
501 */
502static int xs_local_send_request(struct rpc_task *task)
503{
504 struct rpc_rqst *req = task->tk_rqstp;
505 struct rpc_xprt *xprt = req->rq_xprt;
506 struct sock_xprt *transport =
507 container_of(xprt, struct sock_xprt, xprt);
508 struct xdr_buf *xdr = &req->rq_snd_buf;
509 int status;
f279cd00 510 int sent = 0;
176e21ee
CL		 511
512 xs_encode_stream_record_marker(&req->rq_snd_buf);
513
514 xs_pktdump("packet data:",
515 req->rq_svec->iov_base, req->rq_svec->iov_len);
516
f279cd00
JB		 517 status = xs_sendpages(transport->sock, NULL, 0, xdr, req->rq_bytes_sent,
518 true, &sent);
176e21ee
CL		 519 dprintk("RPC: %s(%u) = %d\n",
520 __func__, xdr->len - req->rq_bytes_sent, status);
743c69e7
N		 521
522 if (status == -EAGAIN && sock_writeable(transport->inet))
523 status = -ENOBUFS;
524
f279cd00
JB		 525 if (likely(sent > 0) || status == 0) {
526 req->rq_bytes_sent += sent;
527 req->rq_xmit_bytes_sent += sent;
176e21ee
CL		 528 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
529 req->rq_bytes_sent = 0;
530 return 0;
531 }
532 status = -EAGAIN;
533 }
534
535 switch (status) {
3601c4a9 536 case -ENOBUFS:
b5872f0c 537 break;
176e21ee
CL		 538 case -EAGAIN:
539 status = xs_nospace(task);
540 break;
541 default:
542 dprintk("RPC: sendmsg returned unrecognized error %d\n",
543 -status);
544 case -EPIPE:
545 xs_close(xprt);
546 status = -ENOTCONN;
547 }
548
549 return status;
550}
551
262965f5
CL		 552/**
553 * xs_udp_send_request - write an RPC request to a UDP socket
554 * @task: address of RPC task that manages the state of an RPC request
555 *
556 * Return values:
557 * 0: The request has been sent
558 * EAGAIN: The socket was blocked, please call again later to
559 * complete the request
560 * ENOTCONN: Caller needs to invoke connect logic then call again
25985edc 561 * other: Some other error occurred, the request was not sent
262965f5
CL		 562 */
563static int xs_udp_send_request(struct rpc_task *task)
564{
565 struct rpc_rqst *req = task->tk_rqstp;
566 struct rpc_xprt *xprt = req->rq_xprt;
ee0ac0c2 567 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
262965f5 568 struct xdr_buf *xdr = &req->rq_snd_buf;
f279cd00 569 int sent = 0;
262965f5 570 int status;
a246b010 571
9903cd1c 572 xs_pktdump("packet data:",
a246b010
CL		 573 req->rq_svec->iov_base,
574 req->rq_svec->iov_len);
575
01d37c42
TM		 576 if (!xprt_bound(xprt))
577 return -ENOTCONN;
f279cd00
JB		 578 status = xs_sendpages(transport->sock, xs_addr(xprt), xprt->addrlen,
579 xdr, req->rq_bytes_sent, true, &sent);
a246b010 580
46121cf7 581 dprintk("RPC: xs_udp_send_request(%u) = %d\n",
262965f5 582 xdr->len - req->rq_bytes_sent, status);
a246b010 583
3dedbb5c
JB		 584 /* firewall is blocking us, don't return -EAGAIN or we end up looping */
585 if (status == -EPERM)
586 goto process_status;
587
743c69e7
N		 588 if (status == -EAGAIN && sock_writeable(transport->inet))
589 status = -ENOBUFS;
590
f279cd00
JB		 591 if (sent > 0 || status == 0) {
592 req->rq_xmit_bytes_sent += sent;
593 if (sent >= req->rq_slen)
2199700f
TM		 594 return 0;
595 /* Still some bytes left; set up for a retry later. */
262965f5 596 status = -EAGAIN;
2199700f 597 }
a246b010 598
3dedbb5c 599process_status:
262965f5 600 switch (status) {
fba91afb
TM		 601 case -ENOTSOCK:
602 status = -ENOTCONN;
603 /* Should we call xs_close() here? */
604 break;
b6ddf64f 605 case -EAGAIN:
5e3771ce 606 status = xs_nospace(task);
b6ddf64f 607 break;
262965f5 608 case -ENETUNREACH:
3601c4a9 609 case -ENOBUFS:
262965f5 610 case -EPIPE:
a246b010 611 case -ECONNREFUSED:
3dedbb5c 612 case -EPERM:
a246b010 613 /* When the server has died, an ICMP port unreachable message
9903cd1c 614 * prompts ECONNREFUSED. */
13331a55
TM		 615 break;
616 default:
617 dprintk("RPC: sendmsg returned unrecognized error %d\n",
618 -status);
a246b010 619 }
5fe46e9d 620
262965f5 621 return status;
a246b010
CL		 622}
623
9903cd1c 624/**
262965f5 625 * xs_tcp_send_request - write an RPC request to a TCP socket
9903cd1c
CL		 626 * @task: address of RPC task that manages the state of an RPC request
627 *
628 * Return values:
262965f5
CL		 629 * 0: The request has been sent
630 * EAGAIN: The socket was blocked, please call again later to
631 * complete the request
632 * ENOTCONN: Caller needs to invoke connect logic then call again
25985edc 633 * other: Some other error occurred, the request was not sent
9903cd1c
CL		 634 *
635 * XXX: In the case of soft timeouts, should we eventually give up
262965f5 636 * if sendmsg is not able to make progress?
9903cd1c 637 */
262965f5 638static int xs_tcp_send_request(struct rpc_task *task)
a246b010
CL		 639{
640 struct rpc_rqst *req = task->tk_rqstp;
641 struct rpc_xprt *xprt = req->rq_xprt;
ee0ac0c2 642 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
262965f5 643 struct xdr_buf *xdr = &req->rq_snd_buf;
a6b31d18 644 bool zerocopy = true;
9ffadfbc 645 bool vm_wait = false;
b595bb15 646 int status;
f279cd00 647 int sent;
a246b010 648
61677eee 649 xs_encode_stream_record_marker(&req->rq_snd_buf);
a246b010 650
262965f5
CL		 651 xs_pktdump("packet data:",
652 req->rq_svec->iov_base,
653 req->rq_svec->iov_len);
a6b31d18
TM		 654 /* Don't use zero copy if this is a resend. If the RPC call
655 * completes while the socket holds a reference to the pages,
656 * then we may end up resending corrupted data.
657 */
658 if (task->tk_flags & RPC_TASK_SENT)
659 zerocopy = false;
a246b010
CL		 660
661 /* Continue transmitting the packet/record. We must be careful
662 * to cope with writespace callbacks arriving _after_ we have
262965f5 663 * called sendmsg(). */
a246b010 664 while (1) {
f279cd00
JB		 665 sent = 0;
666 status = xs_sendpages(transport->sock, NULL, 0, xdr,
667 req->rq_bytes_sent, zerocopy, &sent);
a246b010 668
46121cf7 669 dprintk("RPC: xs_tcp_send_request(%u) = %d\n",
262965f5 670 xdr->len - req->rq_bytes_sent, status);
a246b010 671
262965f5
CL		 672 /* If we've sent the entire packet, immediately
673 * reset the count of bytes sent. */
f279cd00
JB		 674 req->rq_bytes_sent += sent;
675 req->rq_xmit_bytes_sent += sent;
262965f5
CL		 676 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
677 req->rq_bytes_sent = 0;
678 return 0;
679 }
a246b010 680
9ffadfbc
TM		 681 WARN_ON_ONCE(sent == 0 && status == 0);
682
683 if (status == -EAGAIN ) {
684 /*
685 * Return EAGAIN if we're sure we're hitting the
686 * socket send buffer limits.
687 */
688 if (test_bit(SOCK_NOSPACE, &transport->sock->flags))
689 break;
690 /*
691 * Did we hit a memory allocation failure?
692 */
693 if (sent == 0) {
694 status = -ENOBUFS;
695 if (vm_wait)
696 break;
697 /* Retry, knowing now that we're below the
698 * socket send buffer limit
699 */
700 vm_wait = true;
701 }
702 continue;
703 }
f580dd04
TM		 704 if (status < 0)
705 break;
9ffadfbc 706 vm_wait = false;
a246b010
CL		 707 }
708
262965f5 709 switch (status) {
fba91afb
TM		 710 case -ENOTSOCK:
711 status = -ENOTCONN;
712 /* Should we call xs_close() here? */
713 break;
262965f5 714 case -EAGAIN:
5e3771ce 715 status = xs_nospace(task);
262965f5 716 break;
262965f5 717 case -ECONNRESET:
2a9e1cfa 718 case -ECONNREFUSED:
262965f5 719 case -ENOTCONN:
3913c78c 720 case -EADDRINUSE:
b5872f0c 721 case -ENOBUFS:
b9d2bb2e 722 case -EPIPE:
13331a55
TM		 723 break;
724 default:
725 dprintk("RPC: sendmsg returned unrecognized error %d\n",
726 -status);
a246b010 727 }
5fe46e9d 728
a246b010
CL		 729 return status;
730}
731
e0ab53de
TM		 732/**
733 * xs_tcp_release_xprt - clean up after a tcp transmission
734 * @xprt: transport
735 * @task: rpc task
736 *
737 * This cleans up if an error causes us to abort the transmission of a request.
738 * In this case, the socket may need to be reset in order to avoid confusing
739 * the server.
740 */
741static void xs_tcp_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
742{
743 struct rpc_rqst *req;
744
745 if (task != xprt->snd_task)
746 return;
747 if (task == NULL)
748 goto out_release;
749 req = task->tk_rqstp;
43cedbf0
TM		 750 if (req == NULL)
751 goto out_release;
e0ab53de
TM		 752 if (req->rq_bytes_sent == 0)
753 goto out_release;
754 if (req->rq_bytes_sent == req->rq_snd_buf.len)
755 goto out_release;
a4f0835c 756 set_bit(XPRT_CLOSE_WAIT, &xprt->state);
e0ab53de
TM		 757out_release:
758 xprt_release_xprt(xprt, task);
759}
760
2a9e1cfa
TM		 761static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk)
762{
763 transport->old_data_ready = sk->sk_data_ready;
764 transport->old_state_change = sk->sk_state_change;
765 transport->old_write_space = sk->sk_write_space;
2118071d 766 transport->old_error_report = sk->sk_error_report;
2a9e1cfa
TM		 767}
768
769static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk)
770{
771 sk->sk_data_ready = transport->old_data_ready;
772 sk->sk_state_change = transport->old_state_change;
773 sk->sk_write_space = transport->old_write_space;
2118071d
TM		 774 sk->sk_error_report = transport->old_error_report;
775}
776
42d42a5b
TM		 777static void xs_sock_reset_state_flags(struct rpc_xprt *xprt)
778{
779 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
780
781 clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state);
782}
783
b70ae915
TM		 784static void xs_sock_reset_connection_flags(struct rpc_xprt *xprt)
785{
786 smp_mb__before_atomic();
787 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
788 clear_bit(XPRT_CLOSING, &xprt->state);
42d42a5b 789 xs_sock_reset_state_flags(xprt);
b70ae915
TM		 790 smp_mb__after_atomic();
791}
792
793static void xs_sock_mark_closed(struct rpc_xprt *xprt)
794{
795 xs_sock_reset_connection_flags(xprt);
796 /* Mark transport as closed and wake up all pending tasks */
797 xprt_disconnect_done(xprt);
798}
799
2118071d
TM		 800/**
801 * xs_error_report - callback to handle TCP socket state errors
802 * @sk: socket
803 *
804 * Note: we don't call sock_error() since there may be a rpc_task
805 * using the socket, and so we don't want to clear sk->sk_err.
806 */
807static void xs_error_report(struct sock *sk)
808{
809 struct rpc_xprt *xprt;
810 int err;
811
812 read_lock_bh(&sk->sk_callback_lock);
813 if (!(xprt = xprt_from_sock(sk)))
814 goto out;
815
816 err = -sk->sk_err;
817 if (err == 0)
818 goto out;
b70ae915
TM		 819 /* Is this a reset event? */
820 if (sk->sk_state == TCP_CLOSE)
821 xs_sock_mark_closed(xprt);
2118071d
TM		 822 dprintk("RPC: xs_error_report client %p, error=%d...\n",
823 xprt, -err);
e8353c76 824 trace_rpc_socket_error(xprt, sk->sk_socket, err);
2118071d
TM		 825 xprt_wake_pending_tasks(xprt, err);
826 out:
827 read_unlock_bh(&sk->sk_callback_lock);
2a9e1cfa
TM		 828}
829
fe315e76 830static void xs_reset_transport(struct sock_xprt *transport)
a246b010 831{
ee0ac0c2
CL		 832 struct socket *sock = transport->sock;
833 struct sock *sk = transport->inet;
6cc7e908 834 struct rpc_xprt *xprt = &transport->xprt;
a246b010 835
fe315e76
CL		 836 if (sk == NULL)
837 return;
9903cd1c 838
264d1df3
JL		 839 if (atomic_read(&transport->xprt.swapper))
840 sk_clear_memalloc(sk);
841
09939204
TM		 842 kernel_sock_shutdown(sock, SHUT_RDWR);
843
edc1b01c 844 mutex_lock(&transport->recv_mutex);
a246b010 845 write_lock_bh(&sk->sk_callback_lock);
ee0ac0c2
CL		 846 transport->inet = NULL;
847 transport->sock = NULL;
a246b010 848
9903cd1c 849 sk->sk_user_data = NULL;
2a9e1cfa
TM		 850
851 xs_restore_old_callbacks(transport, sk);
0c78789e 852 xprt_clear_connected(xprt);
a246b010 853 write_unlock_bh(&sk->sk_callback_lock);
6cc7e908 854 xs_sock_reset_connection_flags(xprt);
edc1b01c 855 mutex_unlock(&transport->recv_mutex);
a246b010 856
6cc7e908 857 trace_rpc_socket_close(xprt, sock);
a246b010 858 sock_release(sock);
fe315e76
CL		 859}
860
861/**
862 * xs_close - close a socket
863 * @xprt: transport
864 *
865 * This is used when all requests are complete; ie, no DRC state remains
866 * on the server we want to save.
f75e6745
TM		 867 *
868 * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with
869 * xs_reset_transport() zeroing the socket from underneath a writer.
fe315e76
CL		 870 */
871static void xs_close(struct rpc_xprt *xprt)
872{
873 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
874
875 dprintk("RPC: xs_close xprt %p\n", xprt);
876
877 xs_reset_transport(transport);
61d0a8e6 878 xprt->reestablish_timeout = 0;
fe315e76 879
62da3b24 880 xprt_disconnect_done(xprt);
a246b010
CL		 881}
882
4a068258
CL		 883static void xs_inject_disconnect(struct rpc_xprt *xprt)
884{
885 dprintk("RPC: injecting transport disconnect on xprt=%p\n",
886 xprt);
887 xprt_disconnect_done(xprt);
888}
889
315f3812
KM		 890static void xs_xprt_free(struct rpc_xprt *xprt)
891{
892 xs_free_peer_addresses(xprt);
893 xprt_free(xprt);
894}
895
9903cd1c
CL		 896/**
897 * xs_destroy - prepare to shutdown a transport
898 * @xprt: doomed transport
899 *
900 */
901static void xs_destroy(struct rpc_xprt *xprt)
a246b010 902{
03c78827
TM		 903 struct sock_xprt *transport = container_of(xprt,
904 struct sock_xprt, xprt);
46121cf7 905 dprintk("RPC: xs_destroy xprt %p\n", xprt);
9903cd1c 906
03c78827 907 cancel_delayed_work_sync(&transport->connect_worker);
a1311d87 908 xs_close(xprt);
edc1b01c 909 cancel_work_sync(&transport->recv_worker);
315f3812 910 xs_xprt_free(xprt);
a1311d87 911 module_put(THIS_MODULE);
a246b010
CL		 912}
913
176e21ee
CL		 914static int xs_local_copy_to_xdr(struct xdr_buf *xdr, struct sk_buff *skb)
915{
916 struct xdr_skb_reader desc = {
917 .skb = skb,
918 .offset = sizeof(rpc_fraghdr),
919 .count = skb->len - sizeof(rpc_fraghdr),
920 };
921
922 if (xdr_partial_copy_from_skb(xdr, 0, &desc, xdr_skb_read_bits) < 0)
923 return -1;
924 if (desc.count)
925 return -1;
926 return 0;
927}
928
929/**
a2648094
TM		 930 * xs_local_data_read_skb
931 * @xprt: transport
932 * @sk: socket
933 * @skb: skbuff
176e21ee
CL		 934 *
935 * Currently this assumes we can read the whole reply in a single gulp.
936 */
a2648094
TM		 937static void xs_local_data_read_skb(struct rpc_xprt *xprt,
938 struct sock *sk,
939 struct sk_buff *skb)
176e21ee
CL		 940{
941 struct rpc_task *task;
176e21ee 942 struct rpc_rqst *rovr;
a2648094 943 int repsize, copied;
176e21ee
CL		 944 u32 _xid;
945 __be32 *xp;
946
176e21ee
CL		 947 repsize = skb->len - sizeof(rpc_fraghdr);
948 if (repsize < 4) {
949 dprintk("RPC: impossible RPC reply size %d\n", repsize);
a2648094 950 return;
176e21ee
CL		 951 }
952
953 /* Copy the XID from the skb... */
954 xp = skb_header_pointer(skb, sizeof(rpc_fraghdr), sizeof(_xid), &_xid);
955 if (xp == NULL)
a2648094 956 return;
176e21ee
CL		 957
958 /* Look up and lock the request corresponding to the given XID */
a2648094 959 spin_lock_bh(&xprt->transport_lock);
176e21ee
CL		 960 rovr = xprt_lookup_rqst(xprt, *xp);
961 if (!rovr)
962 goto out_unlock;
963 task = rovr->rq_task;
964
965 copied = rovr->rq_private_buf.buflen;
966 if (copied > repsize)
967 copied = repsize;
968
969 if (xs_local_copy_to_xdr(&rovr->rq_private_buf, skb)) {
970 dprintk("RPC: sk_buff copy failed\n");
971 goto out_unlock;
972 }
973
974 xprt_complete_rqst(task, copied);
975
976 out_unlock:
a2648094
TM		 977 spin_unlock_bh(&xprt->transport_lock);
978}
979
980static void xs_local_data_receive(struct sock_xprt *transport)
981{
982 struct sk_buff *skb;
983 struct sock *sk;
984 int err;
985
986 mutex_lock(&transport->recv_mutex);
987 sk = transport->inet;
988 if (sk == NULL)
989 goto out;
990 for (;;) {
991 skb = skb_recv_datagram(sk, 0, 1, &err);
42d42a5b
TM		 992 if (skb != NULL) {
993 xs_local_data_read_skb(&transport->xprt, sk, skb);
994 skb_free_datagram(sk, skb);
995 continue;
996 }
997 if (!test_and_clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
a2648094 998 break;
a2648094
TM		 999 }
1000out:
1001 mutex_unlock(&transport->recv_mutex);
1002}
1003
1004static void xs_local_data_receive_workfn(struct work_struct *work)
1005{
1006 struct sock_xprt *transport =
1007 container_of(work, struct sock_xprt, recv_worker);
1008 xs_local_data_receive(transport);
176e21ee
CL		 1009}
1010
9903cd1c 1011/**
f9b2ee71
TM		 1012 * xs_udp_data_read_skb - receive callback for UDP sockets
1013 * @xprt: transport
1014 * @sk: socket
1015 * @skb: skbuff
9903cd1c 1016 *
a246b010 1017 */
f9b2ee71
TM		 1018static void xs_udp_data_read_skb(struct rpc_xprt *xprt,
1019 struct sock *sk,
1020 struct sk_buff *skb)
a246b010 1021{
9903cd1c 1022 struct rpc_task *task;
a246b010 1023 struct rpc_rqst *rovr;
f9b2ee71 1024 int repsize, copied;
d8ed029d
AD		 1025 u32 _xid;
1026 __be32 *xp;
a246b010 1027
1da8c681 1028 repsize = skb->len;
a246b010 1029 if (repsize < 4) {
46121cf7 1030 dprintk("RPC: impossible RPC reply size %d!\n", repsize);
f9b2ee71 1031 return;
a246b010
CL		 1032 }
1033
1034 /* Copy the XID from the skb... */
1da8c681 1035 xp = skb_header_pointer(skb, 0, sizeof(_xid), &_xid);
a246b010 1036 if (xp == NULL)
f9b2ee71 1037 return;
a246b010
CL		 1038
1039 /* Look up and lock the request corresponding to the given XID */
f9b2ee71 1040 spin_lock_bh(&xprt->transport_lock);
a246b010
CL		 1041 rovr = xprt_lookup_rqst(xprt, *xp);
1042 if (!rovr)
1043 goto out_unlock;
1044 task = rovr->rq_task;
1045
a246b010
CL		 1046 if ((copied = rovr->rq_private_buf.buflen) > repsize)
1047 copied = repsize;
1048
1049 /* Suck it into the iovec, verify checksum if not done by hw. */
1781f7f5 1050 if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
02c22347 1051 __UDPX_INC_STATS(sk, UDP_MIB_INERRORS);
a246b010 1052 goto out_unlock;
1781f7f5
HX		 1053 }
1054
02c22347 1055 __UDPX_INC_STATS(sk, UDP_MIB_INDATAGRAMS);
a246b010 1056
6a24dfb6 1057 xprt_adjust_cwnd(xprt, task, copied);
1570c1e4 1058 xprt_complete_rqst(task, copied);
a246b010
CL		 1059
1060 out_unlock:
f9b2ee71
TM		 1061 spin_unlock_bh(&xprt->transport_lock);
1062}
1063
1064static void xs_udp_data_receive(struct sock_xprt *transport)
1065{
1066 struct sk_buff *skb;
1067 struct sock *sk;
1068 int err;
1069
1070 mutex_lock(&transport->recv_mutex);
1071 sk = transport->inet;
1072 if (sk == NULL)
1073 goto out;
1074 for (;;) {
1075 skb = skb_recv_datagram(sk, 0, 1, &err);
42d42a5b
TM		 1076 if (skb != NULL) {
1077 xs_udp_data_read_skb(&transport->xprt, sk, skb);
1078 skb_free_datagram(sk, skb);
1079 continue;
1080 }
1081 if (!test_and_clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
f9b2ee71 1082 break;
f9b2ee71
TM		 1083 }
1084out:
1085 mutex_unlock(&transport->recv_mutex);
1086}
1087
1088static void xs_udp_data_receive_workfn(struct work_struct *work)
1089{
1090 struct sock_xprt *transport =
1091 container_of(work, struct sock_xprt, recv_worker);
1092 xs_udp_data_receive(transport);
1093}
1094
1095/**
1096 * xs_data_ready - "data ready" callback for UDP sockets
1097 * @sk: socket with data to read
1098 *
1099 */
1100static void xs_data_ready(struct sock *sk)
1101{
1102 struct rpc_xprt *xprt;
1103
1104 read_lock_bh(&sk->sk_callback_lock);
1105 dprintk("RPC: xs_data_ready...\n");
1106 xprt = xprt_from_sock(sk);
1107 if (xprt != NULL) {
1108 struct sock_xprt *transport = container_of(xprt,
1109 struct sock_xprt, xprt);
5157b956
TM		 1110 transport->old_data_ready(sk);
1111 /* Any data means we had a useful conversation, so
1112 * then we don't need to delay the next reconnect
1113 */
1114 if (xprt->reestablish_timeout)
1115 xprt->reestablish_timeout = 0;
42d42a5b 1116 if (!test_and_set_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
40a5f1b1 1117 queue_work(xprtiod_workqueue, &transport->recv_worker);
f9b2ee71 1118 }
f064af1e 1119 read_unlock_bh(&sk->sk_callback_lock);
a246b010
CL		 1120}
1121
a519fc7a
TM		 1122/*
1123 * Helper function to force a TCP close if the server is sending
1124 * junk and/or it has put us in CLOSE_WAIT
1125 */
1126static void xs_tcp_force_close(struct rpc_xprt *xprt)
1127{
a519fc7a
TM		 1128 xprt_force_disconnect(xprt);
1129}
1130
dd456471 1131static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
a246b010 1132{
51971139 1133 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
a246b010
CL		 1134 size_t len, used;
1135 char *p;
1136
51971139
CL		 1137 p = ((char *) &transport->tcp_fraghdr) + transport->tcp_offset;
1138 len = sizeof(transport->tcp_fraghdr) - 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;
808012fb 1143
51971139
CL		 1144 transport->tcp_reclen = ntohl(transport->tcp_fraghdr);
1145 if (transport->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
e136d092 1146 transport->tcp_flags |= TCP_RCV_LAST_FRAG;
a246b010 1147 else
e136d092 1148 transport->tcp_flags &= ~TCP_RCV_LAST_FRAG;
51971139 1149 transport->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
808012fb 1150
e136d092 1151 transport->tcp_flags &= ~TCP_RCV_COPY_FRAGHDR;
51971139 1152 transport->tcp_offset = 0;
808012fb 1153
a246b010 1154 /* Sanity check of the record length */
18dca02a 1155 if (unlikely(transport->tcp_reclen < 8)) {
46121cf7 1156 dprintk("RPC: invalid TCP record fragment length\n");
a519fc7a 1157 xs_tcp_force_close(xprt);
9903cd1c 1158 return;
a246b010 1159 }
46121cf7 1160 dprintk("RPC: reading TCP record fragment of length %d\n",
51971139 1161 transport->tcp_reclen);
a246b010
CL		 1162}
1163
51971139 1164static void xs_tcp_check_fraghdr(struct sock_xprt *transport)
a246b010 1165{
51971139 1166 if (transport->tcp_offset == transport->tcp_reclen) {
e136d092 1167 transport->tcp_flags |= TCP_RCV_COPY_FRAGHDR;
51971139 1168 transport->tcp_offset = 0;
e136d092
CL		 1169 if (transport->tcp_flags & TCP_RCV_LAST_FRAG) {
1170 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1171 transport->tcp_flags |= TCP_RCV_COPY_XID;
51971139 1172 transport->tcp_copied = 0;
a246b010
CL		 1173 }
1174 }
1175}
1176
dd456471 1177static inline void xs_tcp_read_xid(struct sock_xprt *transport, struct xdr_skb_reader *desc)
a246b010
CL		 1178{
1179 size_t len, used;
1180 char *p;
1181
51971139 1182 len = sizeof(transport->tcp_xid) - transport->tcp_offset;
46121cf7 1183 dprintk("RPC: reading XID (%Zu bytes)\n", len);
51971139 1184 p = ((char *) &transport->tcp_xid) + transport->tcp_offset;
9d292316 1185 used = xdr_skb_read_bits(desc, p, len);
51971139 1186 transport->tcp_offset += used;
a246b010
CL		 1187 if (used != len)
1188 return;
e136d092 1189 transport->tcp_flags &= ~TCP_RCV_COPY_XID;
f4a2e418 1190 transport->tcp_flags |= TCP_RCV_READ_CALLDIR;
51971139 1191 transport->tcp_copied = 4;
18dca02a
RL		 1192 dprintk("RPC: reading %s XID %08x\n",
1193 (transport->tcp_flags & TCP_RPC_REPLY) ? "reply for"
1194 : "request with",
51971139
CL		 1195 ntohl(transport->tcp_xid));
1196 xs_tcp_check_fraghdr(transport);
a246b010
CL		 1197}
1198
18dca02a
RL		 1199static inline void xs_tcp_read_calldir(struct sock_xprt *transport,
1200 struct xdr_skb_reader *desc)
a246b010 1201{
18dca02a
RL		 1202 size_t len, used;
1203 u32 offset;
b76ce561 1204 char *p;
18dca02a
RL		 1205
1206 /*
1207 * We want transport->tcp_offset to be 8 at the end of this routine
1208 * (4 bytes for the xid and 4 bytes for the call/reply flag).
1209 * When this function is called for the first time,
1210 * transport->tcp_offset is 4 (after having already read the xid).
1211 */
1212 offset = transport->tcp_offset - sizeof(transport->tcp_xid);
b76ce561 1213 len = sizeof(transport->tcp_calldir) - offset;
18dca02a 1214 dprintk("RPC: reading CALL/REPLY flag (%Zu bytes)\n", len);
b76ce561
TM		 1215 p = ((char *) &transport->tcp_calldir) + offset;
1216 used = xdr_skb_read_bits(desc, p, len);
18dca02a
RL		 1217 transport->tcp_offset += used;
1218 if (used != len)
1219 return;
f4a2e418 1220 transport->tcp_flags &= ~TCP_RCV_READ_CALLDIR;
f4a2e418
RL		 1221 /*
1222 * We don't yet have the XDR buffer, so we will write the calldir
1223 * out after we get the buffer from the 'struct rpc_rqst'
1224 */
b76ce561
TM		 1225 switch (ntohl(transport->tcp_calldir)) {
1226 case RPC_REPLY:
1227 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1228 transport->tcp_flags |= TCP_RCV_COPY_DATA;
18dca02a 1229 transport->tcp_flags |= TCP_RPC_REPLY;
b76ce561
TM		 1230 break;
1231 case RPC_CALL:
1232 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1233 transport->tcp_flags |= TCP_RCV_COPY_DATA;
18dca02a 1234 transport->tcp_flags &= ~TCP_RPC_REPLY;
b76ce561
TM		 1235 break;
1236 default:
1237 dprintk("RPC: invalid request message type\n");
a519fc7a 1238 xs_tcp_force_close(&transport->xprt);
b76ce561 1239 }
18dca02a
RL		 1240 xs_tcp_check_fraghdr(transport);
1241}
1242
44b98efd
RL		 1243static inline void xs_tcp_read_common(struct rpc_xprt *xprt,
1244 struct xdr_skb_reader *desc,
1245 struct rpc_rqst *req)
a246b010 1246{
44b98efd
RL		 1247 struct sock_xprt *transport =
1248 container_of(xprt, struct sock_xprt, xprt);
a246b010
CL		 1249 struct xdr_buf *rcvbuf;
1250 size_t len;
1251 ssize_t r;
1252
a246b010 1253 rcvbuf = &req->rq_private_buf;
f4a2e418
RL		 1254
1255 if (transport->tcp_flags & TCP_RCV_COPY_CALLDIR) {
1256 /*
1257 * Save the RPC direction in the XDR buffer
1258 */
f4a2e418 1259 memcpy(rcvbuf->head[0].iov_base + transport->tcp_copied,
b76ce561
TM		 1260 &transport->tcp_calldir,
1261 sizeof(transport->tcp_calldir));
1262 transport->tcp_copied += sizeof(transport->tcp_calldir);
f4a2e418 1263 transport->tcp_flags &= ~TCP_RCV_COPY_CALLDIR;
a246b010
CL		 1264 }
1265
a246b010 1266 len = desc->count;
51971139 1267 if (len > transport->tcp_reclen - transport->tcp_offset) {
dd456471 1268 struct xdr_skb_reader my_desc;
a246b010 1269
51971139 1270 len = transport->tcp_reclen - transport->tcp_offset;
a246b010
CL		 1271 memcpy(&my_desc, desc, sizeof(my_desc));
1272 my_desc.count = len;
51971139 1273 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
9d292316 1274 &my_desc, xdr_skb_read_bits);
a246b010
CL		 1275 desc->count -= r;
1276 desc->offset += r;
1277 } else
51971139 1278 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
9d292316 1279 desc, xdr_skb_read_bits);
a246b010
CL		 1280
1281 if (r > 0) {
51971139
CL		 1282 transport->tcp_copied += r;
1283 transport->tcp_offset += r;
a246b010
CL		 1284 }
1285 if (r != len) {
1286 /* Error when copying to the receive buffer,
1287 * usually because we weren't able to allocate
1288 * additional buffer pages. All we can do now
e136d092 1289 * is turn off TCP_RCV_COPY_DATA, so the request
a246b010
CL		 1290 * will not receive any additional updates,
1291 * and time out.
1292 * Any remaining data from this record will
1293 * be discarded.
1294 */
e136d092 1295 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
46121cf7 1296 dprintk("RPC: XID %08x truncated request\n",
51971139 1297 ntohl(transport->tcp_xid));
46121cf7
CL		 1298 dprintk("RPC: xprt = %p, tcp_copied = %lu, "
1299 "tcp_offset = %u, tcp_reclen = %u\n",
1300 xprt, transport->tcp_copied,
1301 transport->tcp_offset, transport->tcp_reclen);
44b98efd 1302 return;
a246b010
CL		 1303 }
1304
46121cf7 1305 dprintk("RPC: XID %08x read %Zd bytes\n",
51971139 1306 ntohl(transport->tcp_xid), r);
46121cf7
CL		 1307 dprintk("RPC: xprt = %p, tcp_copied = %lu, tcp_offset = %u, "
1308 "tcp_reclen = %u\n", xprt, transport->tcp_copied,
1309 transport->tcp_offset, transport->tcp_reclen);
51971139
CL		 1310
1311 if (transport->tcp_copied == req->rq_private_buf.buflen)
e136d092 1312 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
51971139 1313 else if (transport->tcp_offset == transport->tcp_reclen) {
e136d092
CL		 1314 if (transport->tcp_flags & TCP_RCV_LAST_FRAG)
1315 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
a246b010 1316 }
44b98efd
RL		 1317}
1318
1319/*
1320 * Finds the request corresponding to the RPC xid and invokes the common
1321 * tcp read code to read the data.
1322 */
1323static inline int xs_tcp_read_reply(struct rpc_xprt *xprt,
1324 struct xdr_skb_reader *desc)
1325{
1326 struct sock_xprt *transport =
1327 container_of(xprt, struct sock_xprt, xprt);
1328 struct rpc_rqst *req;
1329
1330 dprintk("RPC: read reply XID %08x\n", ntohl(transport->tcp_xid));
1331
1332 /* Find and lock the request corresponding to this xid */
edc1b01c 1333 spin_lock_bh(&xprt->transport_lock);
44b98efd
RL		 1334 req = xprt_lookup_rqst(xprt, transport->tcp_xid);
1335 if (!req) {
1336 dprintk("RPC: XID %08x request not found!\n",
1337 ntohl(transport->tcp_xid));
edc1b01c 1338 spin_unlock_bh(&xprt->transport_lock);
44b98efd
RL		 1339 return -1;
1340 }
1341
1342 xs_tcp_read_common(xprt, desc, req);
1343
e136d092 1344 if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
51971139 1345 xprt_complete_rqst(req->rq_task, transport->tcp_copied);
44b98efd 1346
edc1b01c 1347 spin_unlock_bh(&xprt->transport_lock);
44b98efd
RL		 1348 return 0;
1349}
1350
9e00abc3 1351#if defined(CONFIG_SUNRPC_BACKCHANNEL)
44b98efd
RL		 1352/*
1353 * Obtains an rpc_rqst previously allocated and invokes the common
1354 * tcp read code to read the data. The result is placed in the callback
1355 * queue.
1356 * If we're unable to obtain the rpc_rqst we schedule the closing of the
1357 * connection and return -1.
1358 */
2ea24497 1359static int xs_tcp_read_callback(struct rpc_xprt *xprt,
44b98efd
RL		 1360 struct xdr_skb_reader *desc)
1361{
1362 struct sock_xprt *transport =
1363 container_of(xprt, struct sock_xprt, xprt);
1364 struct rpc_rqst *req;
1365
2ea24497 1366 /* Look up and lock the request corresponding to the given XID */
edc1b01c 1367 spin_lock_bh(&xprt->transport_lock);
2ea24497 1368 req = xprt_lookup_bc_request(xprt, transport->tcp_xid);
44b98efd 1369 if (req == NULL) {
edc1b01c 1370 spin_unlock_bh(&xprt->transport_lock);
44b98efd
RL		 1371 printk(KERN_WARNING "Callback slot table overflowed\n");
1372 xprt_force_disconnect(xprt);
1373 return -1;
1374 }
1375
44b98efd
RL		 1376 dprintk("RPC: read callback XID %08x\n", ntohl(req->rq_xid));
1377 xs_tcp_read_common(xprt, desc, req);
1378
2ea24497
TM		 1379 if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1380 xprt_complete_bc_request(req, transport->tcp_copied);
edc1b01c 1381 spin_unlock_bh(&xprt->transport_lock);
44b98efd
RL		 1382
1383 return 0;
1384}
1385
1386static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1387 struct xdr_skb_reader *desc)
1388{
1389 struct sock_xprt *transport =
1390 container_of(xprt, struct sock_xprt, xprt);
1391
1392 return (transport->tcp_flags & TCP_RPC_REPLY) ?
1393 xs_tcp_read_reply(xprt, desc) :
1394 xs_tcp_read_callback(xprt, desc);
1395}
76566773
CL		 1396
1397static int xs_tcp_bc_up(struct svc_serv *serv, struct net *net)
1398{
1399 int ret;
1400
1401 ret = svc_create_xprt(serv, "tcp-bc", net, PF_INET, 0,
1402 SVC_SOCK_ANONYMOUS);
1403 if (ret < 0)
1404 return ret;
1405 return 0;
1406}
6b26cc8c
CL		 1407
1408static size_t xs_tcp_bc_maxpayload(struct rpc_xprt *xprt)
1409{
1410 return PAGE_SIZE;
1411}
44b98efd
RL		 1412#else
1413static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1414 struct xdr_skb_reader *desc)
1415{
1416 return xs_tcp_read_reply(xprt, desc);
1417}
9e00abc3 1418#endif /* CONFIG_SUNRPC_BACKCHANNEL */
44b98efd
RL		 1419
1420/*
1421 * Read data off the transport. This can be either an RPC_CALL or an
1422 * RPC_REPLY. Relay the processing to helper functions.
1423 */
1424static void xs_tcp_read_data(struct rpc_xprt *xprt,
1425 struct xdr_skb_reader *desc)
1426{
1427 struct sock_xprt *transport =
1428 container_of(xprt, struct sock_xprt, xprt);
1429
1430 if (_xs_tcp_read_data(xprt, desc) == 0)
1431 xs_tcp_check_fraghdr(transport);
1432 else {
1433 /*
1434 * The transport_lock protects the request handling.
1435 * There's no need to hold it to update the tcp_flags.
1436 */
1437 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1438 }
a246b010
CL		 1439}
1440
dd456471 1441static inline void xs_tcp_read_discard(struct sock_xprt *transport, struct xdr_skb_reader *desc)
a246b010
CL		 1442{
1443 size_t len;
1444
51971139 1445 len = transport->tcp_reclen - transport->tcp_offset;
a246b010
CL		 1446 if (len > desc->count)
1447 len = desc->count;
1448 desc->count -= len;
1449 desc->offset += len;
51971139 1450 transport->tcp_offset += len;
46121cf7 1451 dprintk("RPC: discarded %Zu bytes\n", len);
51971139 1452 xs_tcp_check_fraghdr(transport);
a246b010
CL		 1453}
1454
9903cd1c 1455static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
a246b010
CL		 1456{
1457 struct rpc_xprt *xprt = rd_desc->arg.data;
51971139 1458 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
dd456471 1459 struct xdr_skb_reader desc = {
a246b010
CL		 1460 .skb = skb,
1461 .offset = offset,
1462 .count = len,
9903cd1c 1463 };
a246b010 1464
46121cf7 1465 dprintk("RPC: xs_tcp_data_recv started\n");
a246b010 1466 do {
1a867a08 1467 trace_xs_tcp_data_recv(transport);
a246b010
CL		 1468 /* Read in a new fragment marker if necessary */
1469 /* Can we ever really expect to get completely empty fragments? */
e136d092 1470 if (transport->tcp_flags & TCP_RCV_COPY_FRAGHDR) {
9903cd1c 1471 xs_tcp_read_fraghdr(xprt, &desc);
a246b010
CL		 1472 continue;
1473 }
1474 /* Read in the xid if necessary */
e136d092 1475 if (transport->tcp_flags & TCP_RCV_COPY_XID) {
51971139 1476 xs_tcp_read_xid(transport, &desc);
a246b010
CL		 1477 continue;
1478 }
18dca02a 1479 /* Read in the call/reply flag */
f4a2e418 1480 if (transport->tcp_flags & TCP_RCV_READ_CALLDIR) {
18dca02a
RL		 1481 xs_tcp_read_calldir(transport, &desc);
1482 continue;
1483 }
a246b010 1484 /* Read in the request data */
e136d092 1485 if (transport->tcp_flags & TCP_RCV_COPY_DATA) {
44b98efd 1486 xs_tcp_read_data(xprt, &desc);
a246b010
CL		 1487 continue;
1488 }
1489 /* Skip over any trailing bytes on short reads */
51971139 1490 xs_tcp_read_discard(transport, &desc);
a246b010 1491 } while (desc.count);
1a867a08 1492 trace_xs_tcp_data_recv(transport);
46121cf7 1493 dprintk("RPC: xs_tcp_data_recv done\n");
a246b010
CL		 1494 return len - desc.count;
1495}
1496
66d7a56a
TM		 1497static void xs_tcp_data_receive(struct sock_xprt *transport)
1498{
1499 struct rpc_xprt *xprt = &transport->xprt;
1500 struct sock *sk;
1501 read_descriptor_t rd_desc = {
1502 .count = 2*1024*1024,
1503 .arg.data = xprt,
1504 };
1505 unsigned long total = 0;
1506 int read = 0;
1507
edc1b01c 1508 mutex_lock(&transport->recv_mutex);
66d7a56a 1509 sk = transport->inet;
edc1b01c
TM		 1510 if (sk == NULL)
1511 goto out;
66d7a56a
TM		 1512
1513 /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
1514 for (;;) {
edc1b01c 1515 lock_sock(sk);
66d7a56a 1516 read = tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
42d42a5b
TM		 1517 if (read <= 0) {
1518 clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state);
1519 release_sock(sk);
1520 if (!test_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
1521 break;
1522 } else {
1523 release_sock(sk);
1524 total += read;
1525 }
66d7a56a
TM		 1526 rd_desc.count = 65536;
1527 }
edc1b01c
TM		 1528out:
1529 mutex_unlock(&transport->recv_mutex);
66d7a56a
TM		 1530 trace_xs_tcp_data_ready(xprt, read, total);
1531}
1532
edc1b01c
TM		 1533static void xs_tcp_data_receive_workfn(struct work_struct *work)
1534{
1535 struct sock_xprt *transport =
1536 container_of(work, struct sock_xprt, recv_worker);
1537 xs_tcp_data_receive(transport);
1538}
1539
9903cd1c
CL		 1540/**
1541 * xs_tcp_state_change - callback to handle TCP socket state changes
1542 * @sk: socket whose state has changed
1543 *
1544 */
1545static void xs_tcp_state_change(struct sock *sk)
a246b010 1546{
9903cd1c 1547 struct rpc_xprt *xprt;
0fdea1e8 1548 struct sock_xprt *transport;
a246b010 1549
f064af1e 1550 read_lock_bh(&sk->sk_callback_lock);
a246b010
CL		 1551 if (!(xprt = xprt_from_sock(sk)))
1552 goto out;
46121cf7 1553 dprintk("RPC: xs_tcp_state_change client %p...\n", xprt);
669502ff 1554 dprintk("RPC: state %x conn %d dead %d zapped %d sk_shutdown %d\n",
46121cf7
CL		 1555 sk->sk_state, xprt_connected(xprt),
1556 sock_flag(sk, SOCK_DEAD),
669502ff
AC		 1557 sock_flag(sk, SOCK_ZAPPED),
1558 sk->sk_shutdown);
a246b010 1559
0fdea1e8 1560 transport = container_of(xprt, struct sock_xprt, xprt);
40b5ea0c 1561 trace_rpc_socket_state_change(xprt, sk->sk_socket);
a246b010
CL		 1562 switch (sk->sk_state) {
1563 case TCP_ESTABLISHED:
f064af1e 1564 spin_lock(&xprt->transport_lock);
a246b010 1565 if (!xprt_test_and_set_connected(xprt)) {
51971139 1566
a246b010 1567 /* Reset TCP record info */
51971139
CL		 1568 transport->tcp_offset = 0;
1569 transport->tcp_reclen = 0;
1570 transport->tcp_copied = 0;
e136d092
CL		 1571 transport->tcp_flags =
1572 TCP_RCV_COPY_FRAGHDR | TCP_RCV_COPY_XID;
8b71798c 1573 xprt->connect_cookie++;
0fdea1e8
TM		 1574 clear_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
1575 xprt_clear_connecting(xprt);
51971139 1576
2a491991 1577 xprt_wake_pending_tasks(xprt, -EAGAIN);
a246b010 1578 }
f064af1e 1579 spin_unlock(&xprt->transport_lock);
a246b010 1580 break;
3b948ae5
TM		 1581 case TCP_FIN_WAIT1:
1582 /* The client initiated a shutdown of the socket */
7c1d71cf 1583 xprt->connect_cookie++;
663b8858 1584 xprt->reestablish_timeout = 0;
3b948ae5 1585 set_bit(XPRT_CLOSING, &xprt->state);
4e857c58 1586 smp_mb__before_atomic();
3b948ae5 1587 clear_bit(XPRT_CONNECTED, &xprt->state);
ef803670 1588 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
4e857c58 1589 smp_mb__after_atomic();
a246b010 1590 break;
632e3bdc 1591 case TCP_CLOSE_WAIT:
3b948ae5 1592 /* The server initiated a shutdown of the socket */
7c1d71cf 1593 xprt->connect_cookie++;
d0bea455 1594 clear_bit(XPRT_CONNECTED, &xprt->state);
a519fc7a 1595 xs_tcp_force_close(xprt);
663b8858
TM		 1596 case TCP_CLOSING:
1597 /*
1598 * If the server closed down the connection, make sure that
1599 * we back off before reconnecting
1600 */
1601 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1602 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
3b948ae5
TM		 1603 break;
1604 case TCP_LAST_ACK:
670f9457 1605 set_bit(XPRT_CLOSING, &xprt->state);
4e857c58 1606 smp_mb__before_atomic();
3b948ae5 1607 clear_bit(XPRT_CONNECTED, &xprt->state);
4e857c58 1608 smp_mb__after_atomic();
3b948ae5
TM		 1609 break;
1610 case TCP_CLOSE:
0fdea1e8
TM		 1611 if (test_and_clear_bit(XPRT_SOCK_CONNECTING,
1612 &transport->sock_state))
1613 xprt_clear_connecting(xprt);
7d1e8255 1614 xs_sock_mark_closed(xprt);
a246b010
CL		 1615 }
1616 out:
f064af1e 1617 read_unlock_bh(&sk->sk_callback_lock);
a246b010
CL		 1618}
1619
1f0fa154
IJ		 1620static void xs_write_space(struct sock *sk)
1621{
13331a55 1622 struct socket_wq *wq;
1f0fa154
IJ		 1623 struct rpc_xprt *xprt;
1624
13331a55 1625 if (!sk->sk_socket)
1f0fa154 1626 return;
13331a55 1627 clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1f0fa154
IJ		 1628
1629 if (unlikely(!(xprt = xprt_from_sock(sk))))
1630 return;
13331a55
TM		 1631 rcu_read_lock();
1632 wq = rcu_dereference(sk->sk_wq);
1633 if (!wq || test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &wq->flags) == 0)
1634 goto out;
1f0fa154
IJ		 1635
1636 xprt_write_space(xprt);
13331a55
TM		 1637out:
1638 rcu_read_unlock();
1f0fa154
IJ		 1639}
1640
9903cd1c 1641/**
c7b2cae8
CL		 1642 * xs_udp_write_space - callback invoked when socket buffer space
1643 * becomes available
9903cd1c
CL		 1644 * @sk: socket whose state has changed
1645 *
a246b010
CL		 1646 * Called when more output buffer space is available for this socket.
1647 * We try not to wake our writers until they can make "significant"
c7b2cae8 1648 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
a246b010
CL		 1649 * with a bunch of small requests.
1650 */
c7b2cae8 1651static void xs_udp_write_space(struct sock *sk)
a246b010 1652{
f064af1e 1653 read_lock_bh(&sk->sk_callback_lock);
a246b010 1654
c7b2cae8 1655 /* from net/core/sock.c:sock_def_write_space */
1f0fa154
IJ		 1656 if (sock_writeable(sk))
1657 xs_write_space(sk);
a246b010 1658
f064af1e 1659 read_unlock_bh(&sk->sk_callback_lock);
c7b2cae8 1660}
a246b010 1661
c7b2cae8
CL		 1662/**
1663 * xs_tcp_write_space - callback invoked when socket buffer space
1664 * becomes available
1665 * @sk: socket whose state has changed
1666 *
1667 * Called when more output buffer space is available for this socket.
1668 * We try not to wake our writers until they can make "significant"
1669 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1670 * with a bunch of small requests.
1671 */
1672static void xs_tcp_write_space(struct sock *sk)
1673{
f064af1e 1674 read_lock_bh(&sk->sk_callback_lock);
c7b2cae8
CL		 1675
1676 /* from net/core/stream.c:sk_stream_write_space */
64dc6130 1677 if (sk_stream_is_writeable(sk))
1f0fa154 1678 xs_write_space(sk);
b6ddf64f 1679
f064af1e 1680 read_unlock_bh(&sk->sk_callback_lock);
a246b010
CL		 1681}
1682
470056c2 1683static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
a246b010 1684{
ee0ac0c2
CL		 1685 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1686 struct sock *sk = transport->inet;
a246b010 1687
7c6e066e 1688 if (transport->rcvsize) {
a246b010 1689 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
7c6e066e 1690 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
a246b010 1691 }
7c6e066e 1692 if (transport->sndsize) {
a246b010 1693 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
7c6e066e 1694 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
a246b010
CL		 1695 sk->sk_write_space(sk);
1696 }
1697}
1698
43118c29 1699/**
470056c2 1700 * xs_udp_set_buffer_size - set send and receive limits
43118c29 1701 * @xprt: generic transport
470056c2
CL		 1702 * @sndsize: requested size of send buffer, in bytes
1703 * @rcvsize: requested size of receive buffer, in bytes
43118c29 1704 *
470056c2 1705 * Set socket send and receive buffer size limits.
43118c29 1706 */
470056c2 1707static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
43118c29 1708{
7c6e066e
CL		 1709 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1710
1711 transport->sndsize = 0;
470056c2 1712 if (sndsize)
7c6e066e
CL		 1713 transport->sndsize = sndsize + 1024;
1714 transport->rcvsize = 0;
470056c2 1715 if (rcvsize)
7c6e066e 1716 transport->rcvsize = rcvsize + 1024;
470056c2
CL		 1717
1718 xs_udp_do_set_buffer_size(xprt);
43118c29
CL		 1719}
1720
46c0ee8b
CL		 1721/**
1722 * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1723 * @task: task that timed out
1724 *
1725 * Adjust the congestion window after a retransmit timeout has occurred.
1726 */
6a24dfb6 1727static void xs_udp_timer(struct rpc_xprt *xprt, struct rpc_task *task)
46c0ee8b 1728{
6a24dfb6 1729 xprt_adjust_cwnd(xprt, task, -ETIMEDOUT);
46c0ee8b
CL		 1730}
1731
b85d8806
CL		 1732static unsigned short xs_get_random_port(void)
1733{
5d71899a 1734 unsigned short range = xprt_max_resvport - xprt_min_resvport + 1;
63862b5b 1735 unsigned short rand = (unsigned short) prandom_u32() % range;
b85d8806
CL		 1736 return rand + xprt_min_resvport;
1737}
1738
4dda9c8a
TM		 1739/**
1740 * xs_set_reuseaddr_port - set the socket's port and address reuse options
1741 * @sock: socket
1742 *
1743 * Note that this function has to be called on all sockets that share the
1744 * same port, and it must be called before binding.
1745 */
1746static void xs_sock_set_reuseport(struct socket *sock)
1747{
402e23b4 1748 int opt = 1;
4dda9c8a 1749
402e23b4
TM		 1750 kernel_setsockopt(sock, SOL_SOCKET, SO_REUSEPORT,
1751 (char *)&opt, sizeof(opt));
4dda9c8a
TM		 1752}
1753
1754static unsigned short xs_sock_getport(struct socket *sock)
1755{
1756 struct sockaddr_storage buf;
1757 int buflen;
1758 unsigned short port = 0;
1759
1760 if (kernel_getsockname(sock, (struct sockaddr *)&buf, &buflen) < 0)
1761 goto out;
1762 switch (buf.ss_family) {
1763 case AF_INET6:
1764 port = ntohs(((struct sockaddr_in6 *)&buf)->sin6_port);
1765 break;
1766 case AF_INET:
1767 port = ntohs(((struct sockaddr_in *)&buf)->sin_port);
1768 }
1769out:
1770 return port;
1771}
1772
92200412
CL		 1773/**
1774 * xs_set_port - reset the port number in the remote endpoint address
1775 * @xprt: generic transport
1776 * @port: new port number
1777 *
1778 */
1779static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1780{
46121cf7 1781 dprintk("RPC: setting port for xprt %p to %u\n", xprt, port);
c4efcb1d 1782
9dc3b095
CL		 1783 rpc_set_port(xs_addr(xprt), port);
1784 xs_update_peer_port(xprt);
92200412
CL		 1785}
1786
4dda9c8a
TM		 1787static void xs_set_srcport(struct sock_xprt *transport, struct socket *sock)
1788{
1789 if (transport->srcport == 0)
1790 transport->srcport = xs_sock_getport(sock);
1791}
1792
5d4ec932 1793static unsigned short xs_get_srcport(struct sock_xprt *transport)
67a391d7 1794{
fbfffbd5 1795 unsigned short port = transport->srcport;
67a391d7
TM		 1796
1797 if (port == 0 && transport->xprt.resvport)
1798 port = xs_get_random_port();
1799 return port;
1800}
1801
baaf4e48 1802static unsigned short xs_next_srcport(struct sock_xprt *transport, unsigned short port)
67a391d7 1803{
fbfffbd5
CL		 1804 if (transport->srcport != 0)
1805 transport->srcport = 0;
67a391d7
TM		 1806 if (!transport->xprt.resvport)
1807 return 0;
1808 if (port <= xprt_min_resvport || port > xprt_max_resvport)
1809 return xprt_max_resvport;
1810 return --port;
1811}
beb59b68 1812static int xs_bind(struct sock_xprt *transport, struct socket *sock)
a246b010 1813{
beb59b68 1814 struct sockaddr_storage myaddr;
67a391d7 1815 int err, nloop = 0;
5d4ec932 1816 unsigned short port = xs_get_srcport(transport);
67a391d7 1817 unsigned short last;
a246b010 1818
0f7a622c
CP		 1819 /*
1820 * If we are asking for any ephemeral port (i.e. port == 0 &&
1821 * transport->xprt.resvport == 0), don't bind. Let the local
1822 * port selection happen implicitly when the socket is used
1823 * (for example at connect time).
1824 *
1825 * This ensures that we can continue to establish TCP
1826 * connections even when all local ephemeral ports are already
1827 * a part of some TCP connection. This makes no difference
1828 * for UDP sockets, but also doens't harm them.
1829 *
1830 * If we're asking for any reserved port (i.e. port == 0 &&
1831 * transport->xprt.resvport == 1) xs_get_srcport above will
1832 * ensure that port is non-zero and we will bind as needed.
1833 */
1834 if (port == 0)
1835 return 0;
1836
beb59b68 1837 memcpy(&myaddr, &transport->srcaddr, transport->xprt.addrlen);
a246b010 1838 do {
beb59b68
PE		 1839 rpc_set_port((struct sockaddr *)&myaddr, port);
1840 err = kernel_bind(sock, (struct sockaddr *)&myaddr,
1841 transport->xprt.addrlen);
a246b010 1842 if (err == 0) {
fbfffbd5 1843 transport->srcport = port;
d3bc9a1d 1844 break;
a246b010 1845 }
67a391d7 1846 last = port;
baaf4e48 1847 port = xs_next_srcport(transport, port);
67a391d7
TM		 1848 if (port > last)
1849 nloop++;
1850 } while (err == -EADDRINUSE && nloop != 2);
90058d37 1851
4232e863 1852 if (myaddr.ss_family == AF_INET)
beb59b68
PE		 1853 dprintk("RPC: %s %pI4:%u: %s (%d)\n", __func__,
1854 &((struct sockaddr_in *)&myaddr)->sin_addr,
1855 port, err ? "failed" : "ok", err);
1856 else
1857 dprintk("RPC: %s %pI6:%u: %s (%d)\n", __func__,
1858 &((struct sockaddr_in6 *)&myaddr)->sin6_addr,
1859 port, err ? "failed" : "ok", err);
a246b010
CL		 1860 return err;
1861}
1862
176e21ee
CL		 1863/*
1864 * We don't support autobind on AF_LOCAL sockets
1865 */
1866static void xs_local_rpcbind(struct rpc_task *task)
1867{
fb43d172 1868 xprt_set_bound(task->tk_xprt);
176e21ee
CL		 1869}
1870
1871static void xs_local_set_port(struct rpc_xprt *xprt, unsigned short port)
1872{
1873}
beb59b68 1874
ed07536e
PZ		 1875#ifdef CONFIG_DEBUG_LOCK_ALLOC
1876static struct lock_class_key xs_key[2];
1877static struct lock_class_key xs_slock_key[2];
1878
176e21ee
CL		 1879static inline void xs_reclassify_socketu(struct socket *sock)
1880{
1881 struct sock *sk = sock->sk;
1882
176e21ee
CL		 1883 sock_lock_init_class_and_name(sk, "slock-AF_LOCAL-RPC",
1884 &xs_slock_key[1], "sk_lock-AF_LOCAL-RPC", &xs_key[1]);
1885}
1886
8945ee5e 1887static inline void xs_reclassify_socket4(struct socket *sock)
ed07536e
PZ		 1888{
1889 struct sock *sk = sock->sk;
8945ee5e 1890
8945ee5e
CL		 1891 sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1892 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1893}
ed07536e 1894
8945ee5e
CL		 1895static inline void xs_reclassify_socket6(struct socket *sock)
1896{
1897 struct sock *sk = sock->sk;
ed07536e 1898
8945ee5e
CL		 1899 sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1900 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
ed07536e 1901}
6bc9638a
PE		 1902
1903static inline void xs_reclassify_socket(int family, struct socket *sock)
1904{
fafc4e1e 1905 if (WARN_ON_ONCE(!sock_allow_reclassification(sock->sk)))
1b7a1819
WAA		 1906 return;
1907
4232e863 1908 switch (family) {
176e21ee
CL		 1909 case AF_LOCAL:
1910 xs_reclassify_socketu(sock);
1911 break;
4232e863 1912 case AF_INET:
6bc9638a 1913 xs_reclassify_socket4(sock);
4232e863
CL		 1914 break;
1915 case AF_INET6:
6bc9638a 1916 xs_reclassify_socket6(sock);
4232e863
CL		 1917 break;
1918 }
6bc9638a 1919}
ed07536e 1920#else
6bc9638a
PE		 1921static inline void xs_reclassify_socket(int family, struct socket *sock)
1922{
1923}
ed07536e
PZ		 1924#endif
1925
93dc41bd
N		 1926static void xs_dummy_setup_socket(struct work_struct *work)
1927{
1928}
1929
6bc9638a 1930static struct socket *xs_create_sock(struct rpc_xprt *xprt,
4dda9c8a
TM		 1931 struct sock_xprt *transport, int family, int type,
1932 int protocol, bool reuseport)
22f79326
PE		 1933{
1934 struct socket *sock;
1935 int err;
1936
6bc9638a 1937 err = __sock_create(xprt->xprt_net, family, type, protocol, &sock, 1);
22f79326
PE		 1938 if (err < 0) {
1939 dprintk("RPC: can't create %d transport socket (%d).\n",
1940 protocol, -err);
1941 goto out;
1942 }
6bc9638a 1943 xs_reclassify_socket(family, sock);
22f79326 1944
4dda9c8a
TM		 1945 if (reuseport)
1946 xs_sock_set_reuseport(sock);
1947
4cea288a
BH		 1948 err = xs_bind(transport, sock);
1949 if (err) {
22f79326
PE		 1950 sock_release(sock);
1951 goto out;
1952 }
1953
1954 return sock;
1955out:
1956 return ERR_PTR(err);
1957}
1958
176e21ee
CL		 1959static int xs_local_finish_connecting(struct rpc_xprt *xprt,
1960 struct socket *sock)
1961{
1962 struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1963 xprt);
1964
1965 if (!transport->inet) {
1966 struct sock *sk = sock->sk;
1967
1968 write_lock_bh(&sk->sk_callback_lock);
1969
1970 xs_save_old_callbacks(transport, sk);
1971
1972 sk->sk_user_data = xprt;
a2648094 1973 sk->sk_data_ready = xs_data_ready;
176e21ee 1974 sk->sk_write_space = xs_udp_write_space;
b4411457 1975 sock_set_flag(sk, SOCK_FASYNC);
2118071d 1976 sk->sk_error_report = xs_error_report;
c2126157 1977 sk->sk_allocation = GFP_NOIO;
176e21ee
CL		 1978
1979 xprt_clear_connected(xprt);
1980
1981 /* Reset to new socket */
1982 transport->sock = sock;
1983 transport->inet = sk;
1984
1985 write_unlock_bh(&sk->sk_callback_lock);
1986 }
1987
1988 /* Tell the socket layer to start connecting... */
1989 xprt->stat.connect_count++;
1990 xprt->stat.connect_start = jiffies;
1991 return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, 0);
1992}
1993
1994/**
1995 * xs_local_setup_socket - create AF_LOCAL socket, connect to a local endpoint
176e21ee 1996 * @transport: socket transport to connect
176e21ee 1997 */
dc107402 1998static int xs_local_setup_socket(struct sock_xprt *transport)
176e21ee 1999{
176e21ee
CL		 2000 struct rpc_xprt *xprt = &transport->xprt;
2001 struct socket *sock;
2002 int status = -EIO;
2003
176e21ee
CL		 2004 status = __sock_create(xprt->xprt_net, AF_LOCAL,
2005 SOCK_STREAM, 0, &sock, 1);
2006 if (status < 0) {
2007 dprintk("RPC: can't create AF_LOCAL "
2008 "transport socket (%d).\n", -status);
2009 goto out;
2010 }
d1358917 2011 xs_reclassify_socket(AF_LOCAL, sock);
176e21ee
CL		 2012
2013 dprintk("RPC: worker connecting xprt %p via AF_LOCAL to %s\n",
2014 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2015
2016 status = xs_local_finish_connecting(xprt, sock);
40b5ea0c 2017 trace_rpc_socket_connect(xprt, sock, status);
176e21ee
CL		 2018 switch (status) {
2019 case 0:
2020 dprintk("RPC: xprt %p connected to %s\n",
2021 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2022 xprt_set_connected(xprt);
3601c4a9 2023 case -ENOBUFS:
176e21ee
CL		 2024 break;
2025 case -ENOENT:
2026 dprintk("RPC: xprt %p: socket %s does not exist\n",
2027 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2028 break;
4a20a988
TM		 2029 case -ECONNREFUSED:
2030 dprintk("RPC: xprt %p: connection refused for %s\n",
2031 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2032 break;
176e21ee
CL		 2033 default:
2034 printk(KERN_ERR "%s: unhandled error (%d) connecting to %s\n",
2035 __func__, -status,
2036 xprt->address_strings[RPC_DISPLAY_ADDR]);
2037 }
2038
2039out:
2040 xprt_clear_connecting(xprt);
2041 xprt_wake_pending_tasks(xprt, status);
dc107402
BF		 2042 return status;
2043}
2044
b6669737 2045static void xs_local_connect(struct rpc_xprt *xprt, struct rpc_task *task)
dc107402 2046{
dc107402
BF		 2047 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2048 int ret;
2049
2050 if (RPC_IS_ASYNC(task)) {
2051 /*
2052 * We want the AF_LOCAL connect to be resolved in the
2053 * filesystem namespace of the process making the rpc
2054 * call. Thus we connect synchronously.
2055 *
2056 * If we want to support asynchronous AF_LOCAL calls,
2057 * we'll need to figure out how to pass a namespace to
2058 * connect.
2059 */
2060 rpc_exit(task, -ENOTCONN);
2061 return;
2062 }
2063 ret = xs_local_setup_socket(transport);
2064 if (ret && !RPC_IS_SOFTCONN(task))
2065 msleep_interruptible(15000);
176e21ee
CL		 2066}
2067
3c87ef6e 2068#if IS_ENABLED(CONFIG_SUNRPC_SWAP)
d6e971d8
JL		 2069/*
2070 * Note that this should be called with XPRT_LOCKED held (or when we otherwise
2071 * know that we have exclusive access to the socket), to guard against
2072 * races with xs_reset_transport.
2073 */
a564b8f0
MG		 2074static void xs_set_memalloc(struct rpc_xprt *xprt)
2075{
2076 struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
2077 xprt);
2078
d6e971d8
JL		 2079 /*
2080 * If there's no sock, then we have nothing to set. The
2081 * reconnecting process will get it for us.
2082 */
2083 if (!transport->inet)
2084 return;
8e228133 2085 if (atomic_read(&xprt->swapper))
a564b8f0
MG		 2086 sk_set_memalloc(transport->inet);
2087}
2088
2089/**
d67fa4d8 2090 * xs_enable_swap - Tag this transport as being used for swap.
a564b8f0 2091 * @xprt: transport to tag
a564b8f0 2092 *
8e228133
JL		 2093 * Take a reference to this transport on behalf of the rpc_clnt, and
2094 * optionally mark it for swapping if it wasn't already.
a564b8f0 2095 */
d67fa4d8
JL		 2096static int
2097xs_enable_swap(struct rpc_xprt *xprt)
a564b8f0 2098{
d6e971d8 2099 struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
a564b8f0 2100
d6e971d8
JL		 2101 if (atomic_inc_return(&xprt->swapper) != 1)
2102 return 0;
2103 if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE))
2104 return -ERESTARTSYS;
2105 if (xs->inet)
2106 sk_set_memalloc(xs->inet);
2107 xprt_release_xprt(xprt, NULL);
8e228133
JL		 2108 return 0;
2109}
a564b8f0 2110
8e228133 2111/**
d67fa4d8 2112 * xs_disable_swap - Untag this transport as being used for swap.
8e228133
JL		 2113 * @xprt: transport to tag
2114 *
2115 * Drop a "swapper" reference to this xprt on behalf of the rpc_clnt. If the
2116 * swapper refcount goes to 0, untag the socket as a memalloc socket.
2117 */
d67fa4d8
JL		 2118static void
2119xs_disable_swap(struct rpc_xprt *xprt)
8e228133 2120{
d6e971d8 2121 struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
8e228133 2122
d6e971d8
JL		 2123 if (!atomic_dec_and_test(&xprt->swapper))
2124 return;
2125 if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE))
2126 return;
2127 if (xs->inet)
2128 sk_clear_memalloc(xs->inet);
2129 xprt_release_xprt(xprt, NULL);
a564b8f0 2130}
a564b8f0
MG		 2131#else
2132static void xs_set_memalloc(struct rpc_xprt *xprt)
2133{
2134}
d67fa4d8
JL		 2135
2136static int
2137xs_enable_swap(struct rpc_xprt *xprt)
2138{
2139 return -EINVAL;
2140}
2141
2142static void
2143xs_disable_swap(struct rpc_xprt *xprt)
2144{
2145}
a564b8f0
MG		 2146#endif
2147
16be2d20
CL		 2148static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2149{
2150 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2151
2152 if (!transport->inet) {
2153 struct sock *sk = sock->sk;
2154
2155 write_lock_bh(&sk->sk_callback_lock);
2156
2a9e1cfa
TM		 2157 xs_save_old_callbacks(transport, sk);
2158
16be2d20 2159 sk->sk_user_data = xprt;
f9b2ee71 2160 sk->sk_data_ready = xs_data_ready;
16be2d20 2161 sk->sk_write_space = xs_udp_write_space;
b4411457 2162 sock_set_flag(sk, SOCK_FASYNC);
c2126157 2163 sk->sk_allocation = GFP_NOIO;
16be2d20
CL		 2164
2165 xprt_set_connected(xprt);
2166
2167 /* Reset to new socket */
2168 transport->sock = sock;
2169 transport->inet = sk;
2170
a564b8f0
MG		 2171 xs_set_memalloc(xprt);
2172
16be2d20
CL		 2173 write_unlock_bh(&sk->sk_callback_lock);
2174 }
2175 xs_udp_do_set_buffer_size(xprt);
2176}
2177
8c14ff2a 2178static void xs_udp_setup_socket(struct work_struct *work)
a246b010 2179{
34161db6
TM		 2180 struct sock_xprt *transport =
2181 container_of(work, struct sock_xprt, connect_worker.work);
c8475461 2182 struct rpc_xprt *xprt = &transport->xprt;
ee0ac0c2 2183 struct socket *sock = transport->sock;
b65c0310 2184 int status = -EIO;
9903cd1c 2185
8c14ff2a 2186 sock = xs_create_sock(xprt, transport,
4dda9c8a
TM		 2187 xs_addr(xprt)->sa_family, SOCK_DGRAM,
2188 IPPROTO_UDP, false);
b65c0310 2189 if (IS_ERR(sock))
b0d93ad5 2190 goto out;
68e220bd 2191
c740eff8
CL		 2192 dprintk("RPC: worker connecting xprt %p via %s to "
2193 "%s (port %s)\n", xprt,
2194 xprt->address_strings[RPC_DISPLAY_PROTO],
2195 xprt->address_strings[RPC_DISPLAY_ADDR],
2196 xprt->address_strings[RPC_DISPLAY_PORT]);
68e220bd
CL		 2197
2198 xs_udp_finish_connecting(xprt, sock);
40b5ea0c 2199 trace_rpc_socket_connect(xprt, sock, 0);
b0d93ad5
CL		 2200 status = 0;
2201out:
718ba5b8 2202 xprt_unlock_connect(xprt, transport);
b0d93ad5 2203 xprt_clear_connecting(xprt);
7d1e8255 2204 xprt_wake_pending_tasks(xprt, status);
a246b010
CL		 2205}
2206
4876cc77
TM		 2207/**
2208 * xs_tcp_shutdown - gracefully shut down a TCP socket
2209 * @xprt: transport
2210 *
2211 * Initiates a graceful shutdown of the TCP socket by calling the
2212 * equivalent of shutdown(SHUT_RDWR);
2213 */
2214static void xs_tcp_shutdown(struct rpc_xprt *xprt)
2215{
2216 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2217 struct socket *sock = transport->sock;
2218
2219 if (sock == NULL)
2220 return;
2221 if (xprt_connected(xprt)) {
2222 kernel_sock_shutdown(sock, SHUT_RDWR);
2223 trace_rpc_socket_shutdown(xprt, sock);
2224 } else
2225 xs_reset_transport(transport);
2226}
2227
16be2d20 2228static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
a246b010 2229{
16be2d20 2230 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
fe19a96b 2231 int ret = -ENOTCONN;
edb267a6 2232
ee0ac0c2 2233 if (!transport->inet) {
b0d93ad5 2234 struct sock *sk = sock->sk;
7f260e85
TM		 2235 unsigned int keepidle = xprt->timeout->to_initval / HZ;
2236 unsigned int keepcnt = xprt->timeout->to_retries + 1;
2237 unsigned int opt_on = 1;
775f06ab 2238 unsigned int timeo;
7f260e85
TM		 2239
2240 /* TCP Keepalive options */
2241 kernel_setsockopt(sock, SOL_SOCKET, SO_KEEPALIVE,
2242 (char *)&opt_on, sizeof(opt_on));
2243 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPIDLE,
2244 (char *)&keepidle, sizeof(keepidle));
2245 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPINTVL,
2246 (char *)&keepidle, sizeof(keepidle));
2247 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPCNT,
2248 (char *)&keepcnt, sizeof(keepcnt));
b0d93ad5 2249
775f06ab
TM		 2250 /* TCP user timeout (see RFC5482) */
2251 timeo = jiffies_to_msecs(xprt->timeout->to_initval) *
2252 (xprt->timeout->to_retries + 1);
2253 kernel_setsockopt(sock, SOL_TCP, TCP_USER_TIMEOUT,
2254 (char *)&timeo, sizeof(timeo));
2255
b0d93ad5
CL		 2256 write_lock_bh(&sk->sk_callback_lock);
2257
2a9e1cfa
TM		 2258 xs_save_old_callbacks(transport, sk);
2259
b0d93ad5 2260 sk->sk_user_data = xprt;
5157b956 2261 sk->sk_data_ready = xs_data_ready;
b0d93ad5
CL		 2262 sk->sk_state_change = xs_tcp_state_change;
2263 sk->sk_write_space = xs_tcp_write_space;
b4411457 2264 sock_set_flag(sk, SOCK_FASYNC);
2118071d 2265 sk->sk_error_report = xs_error_report;
c2126157 2266 sk->sk_allocation = GFP_NOIO;
3167e12c
CL		 2267
2268 /* socket options */
3167e12c 2269 sock_reset_flag(sk, SOCK_LINGER);
3167e12c 2270 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
b0d93ad5
CL		 2271
2272 xprt_clear_connected(xprt);
2273
2274 /* Reset to new socket */
ee0ac0c2
CL		 2275 transport->sock = sock;
2276 transport->inet = sk;
b0d93ad5
CL		 2277
2278 write_unlock_bh(&sk->sk_callback_lock);
2279 }
2280
01d37c42 2281 if (!xprt_bound(xprt))
fe19a96b 2282 goto out;
01d37c42 2283
a564b8f0
MG		 2284 xs_set_memalloc(xprt);
2285
b0d93ad5 2286 /* Tell the socket layer to start connecting... */
262ca07d
CL		 2287 xprt->stat.connect_count++;
2288 xprt->stat.connect_start = jiffies;
0fdea1e8 2289 set_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
fe19a96b
TM		 2290 ret = kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
2291 switch (ret) {
2292 case 0:
4dda9c8a 2293 xs_set_srcport(transport, sock);
fe19a96b
TM		 2294 case -EINPROGRESS:
2295 /* SYN_SENT! */
fe19a96b
TM		 2296 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2297 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2298 }
2299out:
2300 return ret;
16be2d20
CL		 2301}
2302
9903cd1c 2303/**
b61d59ff 2304 * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint
9903cd1c
CL		 2305 *
2306 * Invoked by a work queue tasklet.
a246b010 2307 */
cdd518d5 2308static void xs_tcp_setup_socket(struct work_struct *work)
a246b010 2309{
cdd518d5
PE		 2310 struct sock_xprt *transport =
2311 container_of(work, struct sock_xprt, connect_worker.work);
ee0ac0c2 2312 struct socket *sock = transport->sock;
a9f5f0f7 2313 struct rpc_xprt *xprt = &transport->xprt;
b61d59ff 2314 int status = -EIO;
a246b010 2315
ee0ac0c2 2316 if (!sock) {
cdd518d5 2317 sock = xs_create_sock(xprt, transport,
4dda9c8a
TM		 2318 xs_addr(xprt)->sa_family, SOCK_STREAM,
2319 IPPROTO_TCP, true);
b61d59ff
TM		 2320 if (IS_ERR(sock)) {
2321 status = PTR_ERR(sock);
3167e12c
CL		 2322 goto out;
2323 }
7d1e8255 2324 }
a246b010 2325
c740eff8
CL		 2326 dprintk("RPC: worker connecting xprt %p via %s to "
2327 "%s (port %s)\n", xprt,
2328 xprt->address_strings[RPC_DISPLAY_PROTO],
2329 xprt->address_strings[RPC_DISPLAY_ADDR],
2330 xprt->address_strings[RPC_DISPLAY_PORT]);
edb267a6 2331
16be2d20 2332 status = xs_tcp_finish_connecting(xprt, sock);
40b5ea0c 2333 trace_rpc_socket_connect(xprt, sock, status);
46121cf7
CL		 2334 dprintk("RPC: %p connect status %d connected %d sock state %d\n",
2335 xprt, -status, xprt_connected(xprt),
2336 sock->sk->sk_state);
2a491991 2337 switch (status) {
f75e6745
TM		 2338 default:
2339 printk("%s: connect returned unhandled error %d\n",
2340 __func__, status);
2341 case -EADDRNOTAVAIL:
2342 /* We're probably in TIME_WAIT. Get rid of existing socket,
2343 * and retry
2344 */
a519fc7a 2345 xs_tcp_force_close(xprt);
88b5ed73 2346 break;
2a491991
TM		 2347 case 0:
2348 case -EINPROGRESS:
2349 case -EALREADY:
718ba5b8 2350 xprt_unlock_connect(xprt, transport);
7d1e8255 2351 return;
9fcfe0c8
TM		 2352 case -EINVAL:
2353 /* Happens, for instance, if the user specified a link
2354 * local IPv6 address without a scope-id.
2355 */
3ed5e2a2
TM		 2356 case -ECONNREFUSED:
2357 case -ECONNRESET:
2358 case -ENETUNREACH:
3913c78c 2359 case -EADDRINUSE:
3601c4a9 2360 case -ENOBUFS:
3ed5e2a2 2361 /* retry with existing socket, after a delay */
4efdd92c 2362 xs_tcp_force_close(xprt);
9fcfe0c8 2363 goto out;
a246b010 2364 }
2a491991 2365 status = -EAGAIN;
a246b010 2366out:
718ba5b8 2367 xprt_unlock_connect(xprt, transport);
68e220bd 2368 xprt_clear_connecting(xprt);
7d1e8255 2369 xprt_wake_pending_tasks(xprt, status);
68e220bd 2370}
b0d93ad5 2371
9903cd1c
CL		 2372/**
2373 * xs_connect - connect a socket to a remote endpoint
1b092092 2374 * @xprt: pointer to transport structure
9903cd1c
CL		 2375 * @task: address of RPC task that manages state of connect request
2376 *
2377 * TCP: If the remote end dropped the connection, delay reconnecting.
03bf4b70
CL		 2378 *
2379 * UDP socket connects are synchronous, but we use a work queue anyway
2380 * to guarantee that even unprivileged user processes can set up a
2381 * socket on a privileged port.
2382 *
2383 * If a UDP socket connect fails, the delay behavior here prevents
2384 * retry floods (hard mounts).
9903cd1c 2385 */
1b092092 2386static void xs_connect(struct rpc_xprt *xprt, struct rpc_task *task)
a246b010 2387{
ee0ac0c2 2388 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
a246b010 2389
718ba5b8
TM		 2390 WARN_ON_ONCE(!xprt_lock_connect(xprt, task, transport));
2391
99b1a4c3 2392 if (transport->sock != NULL) {
46121cf7
CL		 2393 dprintk("RPC: xs_connect delayed xprt %p for %lu "
2394 "seconds\n",
03bf4b70 2395 xprt, xprt->reestablish_timeout / HZ);
99b1a4c3
TM		 2396
2397 /* Start by resetting any existing state */
2398 xs_reset_transport(transport);
2399
40a5f1b1 2400 queue_delayed_work(xprtiod_workqueue,
c1384c9c
TM		 2401 &transport->connect_worker,
2402 xprt->reestablish_timeout);
03bf4b70 2403 xprt->reestablish_timeout <<= 1;
61d0a8e6
NB		 2404 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2405 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
03bf4b70
CL		 2406 if (xprt->reestablish_timeout > XS_TCP_MAX_REEST_TO)
2407 xprt->reestablish_timeout = XS_TCP_MAX_REEST_TO;
b0d93ad5 2408 } else {
46121cf7 2409 dprintk("RPC: xs_connect scheduled xprt %p\n", xprt);
40a5f1b1 2410 queue_delayed_work(xprtiod_workqueue,
c1384c9c 2411 &transport->connect_worker, 0);
a246b010
CL		 2412 }
2413}
2414
176e21ee
CL		 2415/**
2416 * xs_local_print_stats - display AF_LOCAL socket-specifc stats
2417 * @xprt: rpc_xprt struct containing statistics
2418 * @seq: output file
2419 *
2420 */
2421static void xs_local_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2422{
2423 long idle_time = 0;
2424
2425 if (xprt_connected(xprt))
2426 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2427
2428 seq_printf(seq, "\txprt:\tlocal %lu %lu %lu %ld %lu %lu %lu "
15a45206 2429 "%llu %llu %lu %llu %llu\n",
176e21ee
CL		 2430 xprt->stat.bind_count,
2431 xprt->stat.connect_count,
2432 xprt->stat.connect_time,
2433 idle_time,
2434 xprt->stat.sends,
2435 xprt->stat.recvs,
2436 xprt->stat.bad_xids,
2437 xprt->stat.req_u,
15a45206
AA		 2438 xprt->stat.bklog_u,
2439 xprt->stat.max_slots,
2440 xprt->stat.sending_u,
2441 xprt->stat.pending_u);
176e21ee
CL		 2442}
2443
262ca07d
CL		 2444/**
2445 * xs_udp_print_stats - display UDP socket-specifc stats
2446 * @xprt: rpc_xprt struct containing statistics
2447 * @seq: output file
2448 *
2449 */
2450static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2451{
c8475461
CL		 2452 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2453
15a45206
AA		 2454 seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %llu %llu "
2455 "%lu %llu %llu\n",
fbfffbd5 2456 transport->srcport,
262ca07d
CL		 2457 xprt->stat.bind_count,
2458 xprt->stat.sends,
2459 xprt->stat.recvs,
2460 xprt->stat.bad_xids,
2461 xprt->stat.req_u,
15a45206
AA		 2462 xprt->stat.bklog_u,
2463 xprt->stat.max_slots,
2464 xprt->stat.sending_u,
2465 xprt->stat.pending_u);
262ca07d
CL		 2466}
2467
2468/**
2469 * xs_tcp_print_stats - display TCP socket-specifc stats
2470 * @xprt: rpc_xprt struct containing statistics
2471 * @seq: output file
2472 *
2473 */
2474static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2475{
c8475461 2476 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
262ca07d
CL		 2477 long idle_time = 0;
2478
2479 if (xprt_connected(xprt))
2480 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2481
15a45206
AA		 2482 seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu "
2483 "%llu %llu %lu %llu %llu\n",
fbfffbd5 2484 transport->srcport,
262ca07d
CL		 2485 xprt->stat.bind_count,
2486 xprt->stat.connect_count,
2487 xprt->stat.connect_time,
2488 idle_time,
2489 xprt->stat.sends,
2490 xprt->stat.recvs,
2491 xprt->stat.bad_xids,
2492 xprt->stat.req_u,
15a45206
AA		 2493 xprt->stat.bklog_u,
2494 xprt->stat.max_slots,
2495 xprt->stat.sending_u,
2496 xprt->stat.pending_u);
262ca07d
CL		 2497}
2498
4cfc7e60
RI		 2499/*
2500 * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason
2501 * we allocate pages instead doing a kmalloc like rpc_malloc is because we want
2502 * to use the server side send routines.
2503 */
5a51f13a 2504static void *bc_malloc(struct rpc_task *task, size_t size)
4cfc7e60
RI		 2505{
2506 struct page *page;
2507 struct rpc_buffer *buf;
2508
b8a13d03
WAA		 2509 WARN_ON_ONCE(size > PAGE_SIZE - sizeof(struct rpc_buffer));
2510 if (size > PAGE_SIZE - sizeof(struct rpc_buffer))
2511 return NULL;
4cfc7e60 2512
b8a13d03 2513 page = alloc_page(GFP_KERNEL);
4cfc7e60
RI		 2514 if (!page)
2515 return NULL;
2516
2517 buf = page_address(page);
2518 buf->len = PAGE_SIZE;
2519
2520 return buf->data;
2521}
2522
2523/*
2524 * Free the space allocated in the bc_alloc routine
2525 */
5a51f13a 2526static void bc_free(void *buffer)
4cfc7e60
RI		 2527{
2528 struct rpc_buffer *buf;
2529
2530 if (!buffer)
2531 return;
2532
2533 buf = container_of(buffer, struct rpc_buffer, data);
2534 free_page((unsigned long)buf);
2535}
2536
2537/*
2538 * Use the svc_sock to send the callback. Must be called with svsk->sk_mutex
2539 * held. Borrows heavily from svc_tcp_sendto and xs_tcp_send_request.
2540 */
2541static int bc_sendto(struct rpc_rqst *req)
2542{
2543 int len;
2544 struct xdr_buf *xbufp = &req->rq_snd_buf;
2545 struct rpc_xprt *xprt = req->rq_xprt;
2546 struct sock_xprt *transport =
2547 container_of(xprt, struct sock_xprt, xprt);
2548 struct socket *sock = transport->sock;
2549 unsigned long headoff;
2550 unsigned long tailoff;
2551
61677eee 2552 xs_encode_stream_record_marker(xbufp);
4cfc7e60
RI		 2553
2554 tailoff = (unsigned long)xbufp->tail[0].iov_base & ~PAGE_MASK;
2555 headoff = (unsigned long)xbufp->head[0].iov_base & ~PAGE_MASK;
2556 len = svc_send_common(sock, xbufp,
2557 virt_to_page(xbufp->head[0].iov_base), headoff,
2558 xbufp->tail[0].iov_base, tailoff);
2559
2560 if (len != xbufp->len) {
2561 printk(KERN_NOTICE "Error sending entire callback!\n");
2562 len = -EAGAIN;
2563 }
2564
2565 return len;
2566}
2567
2568/*
2569 * The send routine. Borrows from svc_send
2570 */
2571static int bc_send_request(struct rpc_task *task)
2572{
2573 struct rpc_rqst *req = task->tk_rqstp;
2574 struct svc_xprt *xprt;
7fc56136 2575 int len;
4cfc7e60
RI		 2576
2577 dprintk("sending request with xid: %08x\n", ntohl(req->rq_xid));
2578 /*
2579 * Get the server socket associated with this callback xprt
2580 */
2581 xprt = req->rq_xprt->bc_xprt;
4cfc7e60
RI		 2582
2583 /*
2584 * Grab the mutex to serialize data as the connection is shared
2585 * with the fore channel
2586 */
2587 if (!mutex_trylock(&xprt->xpt_mutex)) {
2588 rpc_sleep_on(&xprt->xpt_bc_pending, task, NULL);
2589 if (!mutex_trylock(&xprt->xpt_mutex))
2590 return -EAGAIN;
2591 rpc_wake_up_queued_task(&xprt->xpt_bc_pending, task);
2592 }
2593 if (test_bit(XPT_DEAD, &xprt->xpt_flags))
2594 len = -ENOTCONN;
2595 else
2596 len = bc_sendto(req);
2597 mutex_unlock(&xprt->xpt_mutex);
2598
2599 if (len > 0)
2600 len = 0;
2601
2602 return len;
2603}
2604
2605/*
2606 * The close routine. Since this is client initiated, we do nothing
2607 */
2608
2609static void bc_close(struct rpc_xprt *xprt)
2610{
4cfc7e60
RI		 2611}
2612
2613/*
2614 * The xprt destroy routine. Again, because this connection is client
2615 * initiated, we do nothing
2616 */
2617
2618static void bc_destroy(struct rpc_xprt *xprt)
2619{
47f72efa
KM		 2620 dprintk("RPC: bc_destroy xprt %p\n", xprt);
2621
2622 xs_xprt_free(xprt);
2623 module_put(THIS_MODULE);
4cfc7e60
RI		 2624}
2625
176e21ee
CL		 2626static struct rpc_xprt_ops xs_local_ops = {
2627 .reserve_xprt = xprt_reserve_xprt,
2628 .release_xprt = xs_tcp_release_xprt,
f39c1bfb 2629 .alloc_slot = xprt_alloc_slot,
176e21ee
CL		 2630 .rpcbind = xs_local_rpcbind,
2631 .set_port = xs_local_set_port,
dc107402 2632 .connect = xs_local_connect,
176e21ee
CL		 2633 .buf_alloc = rpc_malloc,
2634 .buf_free = rpc_free,
2635 .send_request = xs_local_send_request,
2636 .set_retrans_timeout = xprt_set_retrans_timeout_def,
2637 .close = xs_close,
a1311d87 2638 .destroy = xs_destroy,
176e21ee 2639 .print_stats = xs_local_print_stats,
d67fa4d8
JL		 2640 .enable_swap = xs_enable_swap,
2641 .disable_swap = xs_disable_swap,
176e21ee
CL		 2642};
2643
262965f5 2644static struct rpc_xprt_ops xs_udp_ops = {
43118c29 2645 .set_buffer_size = xs_udp_set_buffer_size,
12a80469 2646 .reserve_xprt = xprt_reserve_xprt_cong,
49e9a890 2647 .release_xprt = xprt_release_xprt_cong,
f39c1bfb 2648 .alloc_slot = xprt_alloc_slot,
45160d62 2649 .rpcbind = rpcb_getport_async,
92200412 2650 .set_port = xs_set_port,
262965f5 2651 .connect = xs_connect,
02107148
CL		 2652 .buf_alloc = rpc_malloc,
2653 .buf_free = rpc_free,
262965f5 2654 .send_request = xs_udp_send_request,
fe3aca29 2655 .set_retrans_timeout = xprt_set_retrans_timeout_rtt,
46c0ee8b 2656 .timer = xs_udp_timer,
a58dd398 2657 .release_request = xprt_release_rqst_cong,
262965f5
CL		 2658 .close = xs_close,
2659 .destroy = xs_destroy,
262ca07d 2660 .print_stats = xs_udp_print_stats,
d67fa4d8
JL		 2661 .enable_swap = xs_enable_swap,
2662 .disable_swap = xs_disable_swap,
4a068258 2663 .inject_disconnect = xs_inject_disconnect,
262965f5
CL		 2664};
2665
2666static struct rpc_xprt_ops xs_tcp_ops = {
12a80469 2667 .reserve_xprt = xprt_reserve_xprt,
e0ab53de 2668 .release_xprt = xs_tcp_release_xprt,
f39c1bfb 2669 .alloc_slot = xprt_lock_and_alloc_slot,
45160d62 2670 .rpcbind = rpcb_getport_async,
92200412 2671 .set_port = xs_set_port,
0b9e7943 2672 .connect = xs_connect,
02107148
CL		 2673 .buf_alloc = rpc_malloc,
2674 .buf_free = rpc_free,
262965f5 2675 .send_request = xs_tcp_send_request,
fe3aca29 2676 .set_retrans_timeout = xprt_set_retrans_timeout_def,
c627d31b 2677 .close = xs_tcp_shutdown,
9903cd1c 2678 .destroy = xs_destroy,
262ca07d 2679 .print_stats = xs_tcp_print_stats,
d67fa4d8
JL		 2680 .enable_swap = xs_enable_swap,
2681 .disable_swap = xs_disable_swap,
4a068258 2682 .inject_disconnect = xs_inject_disconnect,
42e5c3e2
CL		 2683#ifdef CONFIG_SUNRPC_BACKCHANNEL
2684 .bc_setup = xprt_setup_bc,
76566773 2685 .bc_up = xs_tcp_bc_up,
6b26cc8c 2686 .bc_maxpayload = xs_tcp_bc_maxpayload,
42e5c3e2
CL		 2687 .bc_free_rqst = xprt_free_bc_rqst,
2688 .bc_destroy = xprt_destroy_bc,
2689#endif
a246b010
CL		 2690};
2691
4cfc7e60
RI		 2692/*
2693 * The rpc_xprt_ops for the server backchannel
2694 */
2695
2696static struct rpc_xprt_ops bc_tcp_ops = {
2697 .reserve_xprt = xprt_reserve_xprt,
2698 .release_xprt = xprt_release_xprt,
84e28a30 2699 .alloc_slot = xprt_alloc_slot,
4cfc7e60
RI		 2700 .buf_alloc = bc_malloc,
2701 .buf_free = bc_free,
2702 .send_request = bc_send_request,
2703 .set_retrans_timeout = xprt_set_retrans_timeout_def,
2704 .close = bc_close,
2705 .destroy = bc_destroy,
2706 .print_stats = xs_tcp_print_stats,
d67fa4d8
JL		 2707 .enable_swap = xs_enable_swap,
2708 .disable_swap = xs_disable_swap,
4a068258 2709 .inject_disconnect = xs_inject_disconnect,
4cfc7e60
RI		 2710};
2711
92476850
CL		 2712static int xs_init_anyaddr(const int family, struct sockaddr *sap)
2713{
2714 static const struct sockaddr_in sin = {
2715 .sin_family = AF_INET,
2716 .sin_addr.s_addr = htonl(INADDR_ANY),
2717 };
2718 static const struct sockaddr_in6 sin6 = {
2719 .sin6_family = AF_INET6,
2720 .sin6_addr = IN6ADDR_ANY_INIT,
2721 };
2722
2723 switch (family) {
176e21ee
CL		 2724 case AF_LOCAL:
2725 break;
92476850
CL		 2726 case AF_INET:
2727 memcpy(sap, &sin, sizeof(sin));
2728 break;
2729 case AF_INET6:
2730 memcpy(sap, &sin6, sizeof(sin6));
2731 break;
2732 default:
2733 dprintk("RPC: %s: Bad address family\n", __func__);
2734 return -EAFNOSUPPORT;
2735 }
2736 return 0;
2737}
2738
3c341b0b 2739static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
d9ba131d
TM		 2740 unsigned int slot_table_size,
2741 unsigned int max_slot_table_size)
c8541ecd
CL		 2742{
2743 struct rpc_xprt *xprt;
ffc2e518 2744 struct sock_xprt *new;
c8541ecd 2745
96802a09 2746 if (args->addrlen > sizeof(xprt->addr)) {
46121cf7 2747 dprintk("RPC: xs_setup_xprt: address too large\n");
c8541ecd
CL		 2748 return ERR_PTR(-EBADF);
2749 }
2750
d9ba131d
TM		 2751 xprt = xprt_alloc(args->net, sizeof(*new), slot_table_size,
2752 max_slot_table_size);
bd1722d4 2753 if (xprt == NULL) {
46121cf7
CL		 2754 dprintk("RPC: xs_setup_xprt: couldn't allocate "
2755 "rpc_xprt\n");
c8541ecd
CL		 2756 return ERR_PTR(-ENOMEM);
2757 }
c8541ecd 2758
bd1722d4 2759 new = container_of(xprt, struct sock_xprt, xprt);
edc1b01c 2760 mutex_init(&new->recv_mutex);
96802a09
FM		 2761 memcpy(&xprt->addr, args->dstaddr, args->addrlen);
2762 xprt->addrlen = args->addrlen;
d3bc9a1d 2763 if (args->srcaddr)
fbfffbd5 2764 memcpy(&new->srcaddr, args->srcaddr, args->addrlen);
92476850
CL		 2765 else {
2766 int err;
2767 err = xs_init_anyaddr(args->dstaddr->sa_family,
2768 (struct sockaddr *)&new->srcaddr);
2aa13531
SK		 2769 if (err != 0) {
2770 xprt_free(xprt);
92476850 2771 return ERR_PTR(err);
2aa13531 2772 }
92476850 2773 }
c8541ecd
CL		 2774
2775 return xprt;
2776}
2777
176e21ee
CL		 2778static const struct rpc_timeout xs_local_default_timeout = {
2779 .to_initval = 10 * HZ,
2780 .to_maxval = 10 * HZ,
2781 .to_retries = 2,
2782};
2783
2784/**
2785 * xs_setup_local - Set up transport to use an AF_LOCAL socket
2786 * @args: rpc transport creation arguments
2787 *
2788 * AF_LOCAL is a "tpi_cots_ord" transport, just like TCP
2789 */
2790static struct rpc_xprt *xs_setup_local(struct xprt_create *args)
2791{
2792 struct sockaddr_un *sun = (struct sockaddr_un *)args->dstaddr;
2793 struct sock_xprt *transport;
2794 struct rpc_xprt *xprt;
2795 struct rpc_xprt *ret;
2796
d9ba131d
TM		 2797 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2798 xprt_max_tcp_slot_table_entries);
176e21ee
CL		 2799 if (IS_ERR(xprt))
2800 return xprt;
2801 transport = container_of(xprt, struct sock_xprt, xprt);
2802
2803 xprt->prot = 0;
2804 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2805 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2806
2807 xprt->bind_timeout = XS_BIND_TO;
2808 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2809 xprt->idle_timeout = XS_IDLE_DISC_TO;
2810
2811 xprt->ops = &xs_local_ops;
2812 xprt->timeout = &xs_local_default_timeout;
2813
a2648094 2814 INIT_WORK(&transport->recv_worker, xs_local_data_receive_workfn);
93dc41bd
N		 2815 INIT_DELAYED_WORK(&transport->connect_worker,
2816 xs_dummy_setup_socket);
2817
176e21ee
CL		 2818 switch (sun->sun_family) {
2819 case AF_LOCAL:
2820 if (sun->sun_path[0] != '/') {
2821 dprintk("RPC: bad AF_LOCAL address: %s\n",
2822 sun->sun_path);
2823 ret = ERR_PTR(-EINVAL);
2824 goto out_err;
2825 }
2826 xprt_set_bound(xprt);
176e21ee 2827 xs_format_peer_addresses(xprt, "local", RPCBIND_NETID_LOCAL);
7073ea87
BF		 2828 ret = ERR_PTR(xs_local_setup_socket(transport));
2829 if (ret)
2830 goto out_err;
176e21ee
CL		 2831 break;
2832 default:
2833 ret = ERR_PTR(-EAFNOSUPPORT);
2834 goto out_err;
2835 }
2836
2837 dprintk("RPC: set up xprt to %s via AF_LOCAL\n",
2838 xprt->address_strings[RPC_DISPLAY_ADDR]);
2839
2840 if (try_module_get(THIS_MODULE))
2841 return xprt;
2842 ret = ERR_PTR(-EINVAL);
2843out_err:
315f3812 2844 xs_xprt_free(xprt);
176e21ee
CL		 2845 return ret;
2846}
2847
2881ae74
TM		 2848static const struct rpc_timeout xs_udp_default_timeout = {
2849 .to_initval = 5 * HZ,
2850 .to_maxval = 30 * HZ,
2851 .to_increment = 5 * HZ,
2852 .to_retries = 5,
2853};
2854
9903cd1c
CL		 2855/**
2856 * xs_setup_udp - Set up transport to use a UDP socket
96802a09 2857 * @args: rpc transport creation arguments
9903cd1c
CL		 2858 *
2859 */
483066d6 2860static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
a246b010 2861{
8f9d5b1a 2862 struct sockaddr *addr = args->dstaddr;
c8541ecd 2863 struct rpc_xprt *xprt;
c8475461 2864 struct sock_xprt *transport;
0a68b0be 2865 struct rpc_xprt *ret;
a246b010 2866
d9ba131d
TM		 2867 xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries,
2868 xprt_udp_slot_table_entries);
c8541ecd
CL		 2869 if (IS_ERR(xprt))
2870 return xprt;
c8475461 2871 transport = container_of(xprt, struct sock_xprt, xprt);
a246b010 2872
ec739ef0 2873 xprt->prot = IPPROTO_UDP;
808012fb 2874 xprt->tsh_size = 0;
a246b010
CL		 2875 /* XXX: header size can vary due to auth type, IPv6, etc. */
2876 xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
2877
03bf4b70 2878 xprt->bind_timeout = XS_BIND_TO;
03bf4b70
CL		 2879 xprt->reestablish_timeout = XS_UDP_REEST_TO;
2880 xprt->idle_timeout = XS_IDLE_DISC_TO;
a246b010 2881
262965f5 2882 xprt->ops = &xs_udp_ops;
a246b010 2883
ba7392bb 2884 xprt->timeout = &xs_udp_default_timeout;
a246b010 2885
f9b2ee71 2886 INIT_WORK(&transport->recv_worker, xs_udp_data_receive_workfn);
edc1b01c
TM		 2887 INIT_DELAYED_WORK(&transport->connect_worker, xs_udp_setup_socket);
2888
8f9d5b1a
CL		 2889 switch (addr->sa_family) {
2890 case AF_INET:
2891 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2892 xprt_set_bound(xprt);
2893
9dc3b095 2894 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
8f9d5b1a
CL		 2895 break;
2896 case AF_INET6:
2897 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2898 xprt_set_bound(xprt);
2899
9dc3b095 2900 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
8f9d5b1a
CL		 2901 break;
2902 default:
0a68b0be
BF		 2903 ret = ERR_PTR(-EAFNOSUPPORT);
2904 goto out_err;
8f9d5b1a
CL		 2905 }
2906
c740eff8
CL		 2907 if (xprt_bound(xprt))
2908 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2909 xprt->address_strings[RPC_DISPLAY_ADDR],
2910 xprt->address_strings[RPC_DISPLAY_PORT],
2911 xprt->address_strings[RPC_DISPLAY_PROTO]);
2912 else
2913 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
2914 xprt->address_strings[RPC_DISPLAY_ADDR],
2915 xprt->address_strings[RPC_DISPLAY_PROTO]);
edb267a6 2916
bc25571e
TT		 2917 if (try_module_get(THIS_MODULE))
2918 return xprt;
0a68b0be
BF		 2919 ret = ERR_PTR(-EINVAL);
2920out_err:
315f3812 2921 xs_xprt_free(xprt);
0a68b0be 2922 return ret;
a246b010
CL		 2923}
2924
2881ae74
TM		 2925static const struct rpc_timeout xs_tcp_default_timeout = {
2926 .to_initval = 60 * HZ,
2927 .to_maxval = 60 * HZ,
2928 .to_retries = 2,
2929};
2930
9903cd1c
CL		 2931/**
2932 * xs_setup_tcp - Set up transport to use a TCP socket
96802a09 2933 * @args: rpc transport creation arguments
9903cd1c
CL		 2934 *
2935 */
483066d6 2936static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
a246b010 2937{
8f9d5b1a 2938 struct sockaddr *addr = args->dstaddr;
c8541ecd 2939 struct rpc_xprt *xprt;
c8475461 2940 struct sock_xprt *transport;
0a68b0be 2941 struct rpc_xprt *ret;
b7993ceb
TM		 2942 unsigned int max_slot_table_size = xprt_max_tcp_slot_table_entries;
2943
2944 if (args->flags & XPRT_CREATE_INFINITE_SLOTS)
2945 max_slot_table_size = RPC_MAX_SLOT_TABLE_LIMIT;
a246b010 2946
d9ba131d 2947 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
b7993ceb 2948 max_slot_table_size);
c8541ecd
CL		 2949 if (IS_ERR(xprt))
2950 return xprt;
c8475461 2951 transport = container_of(xprt, struct sock_xprt, xprt);
a246b010 2952
ec739ef0 2953 xprt->prot = IPPROTO_TCP;
808012fb 2954 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
808012fb 2955 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
a246b010 2956
03bf4b70 2957 xprt->bind_timeout = XS_BIND_TO;
03bf4b70
CL		 2958 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2959 xprt->idle_timeout = XS_IDLE_DISC_TO;
a246b010 2960
262965f5 2961 xprt->ops = &xs_tcp_ops;
ba7392bb 2962 xprt->timeout = &xs_tcp_default_timeout;
a246b010 2963
edc1b01c
TM		 2964 INIT_WORK(&transport->recv_worker, xs_tcp_data_receive_workfn);
2965 INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_setup_socket);
2966
8f9d5b1a
CL		 2967 switch (addr->sa_family) {
2968 case AF_INET:
2969 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2970 xprt_set_bound(xprt);
2971
9dc3b095 2972 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
8f9d5b1a
CL		 2973 break;
2974 case AF_INET6:
2975 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2976 xprt_set_bound(xprt);
2977
9dc3b095 2978 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
8f9d5b1a
CL		 2979 break;
2980 default:
0a68b0be
BF		 2981 ret = ERR_PTR(-EAFNOSUPPORT);
2982 goto out_err;
8f9d5b1a
CL		 2983 }
2984
c740eff8
CL		 2985 if (xprt_bound(xprt))
2986 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2987 xprt->address_strings[RPC_DISPLAY_ADDR],
2988 xprt->address_strings[RPC_DISPLAY_PORT],
2989 xprt->address_strings[RPC_DISPLAY_PROTO]);
2990 else
2991 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
2992 xprt->address_strings[RPC_DISPLAY_ADDR],
2993 xprt->address_strings[RPC_DISPLAY_PROTO]);
2994
bc25571e
TT		 2995 if (try_module_get(THIS_MODULE))
2996 return xprt;
0a68b0be
BF		 2997 ret = ERR_PTR(-EINVAL);
2998out_err:
315f3812 2999 xs_xprt_free(xprt);
0a68b0be 3000 return ret;
a246b010 3001}
282b32e1 3002
f300baba
AB		 3003/**
3004 * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket
3005 * @args: rpc transport creation arguments
3006 *
3007 */
3008static struct rpc_xprt *xs_setup_bc_tcp(struct xprt_create *args)
3009{
3010 struct sockaddr *addr = args->dstaddr;
3011 struct rpc_xprt *xprt;
3012 struct sock_xprt *transport;
3013 struct svc_sock *bc_sock;
0a68b0be 3014 struct rpc_xprt *ret;
f300baba 3015
d9ba131d
TM		 3016 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
3017 xprt_tcp_slot_table_entries);
f300baba
AB		 3018 if (IS_ERR(xprt))
3019 return xprt;
3020 transport = container_of(xprt, struct sock_xprt, xprt);
3021
3022 xprt->prot = IPPROTO_TCP;
3023 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
3024 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
3025 xprt->timeout = &xs_tcp_default_timeout;
3026
3027 /* backchannel */
3028 xprt_set_bound(xprt);
3029 xprt->bind_timeout = 0;
f300baba
AB		 3030 xprt->reestablish_timeout = 0;
3031 xprt->idle_timeout = 0;
3032
f300baba
AB		 3033 xprt->ops = &bc_tcp_ops;
3034
3035 switch (addr->sa_family) {
3036 case AF_INET:
3037 xs_format_peer_addresses(xprt, "tcp",
3038 RPCBIND_NETID_TCP);
3039 break;
3040 case AF_INET6:
3041 xs_format_peer_addresses(xprt, "tcp",
3042 RPCBIND_NETID_TCP6);
3043 break;
3044 default:
0a68b0be
BF		 3045 ret = ERR_PTR(-EAFNOSUPPORT);
3046 goto out_err;
f300baba
AB		 3047 }
3048
50fa0d40
PE		 3049 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
3050 xprt->address_strings[RPC_DISPLAY_ADDR],
3051 xprt->address_strings[RPC_DISPLAY_PORT],
3052 xprt->address_strings[RPC_DISPLAY_PROTO]);
f300baba 3053
99de8ea9
BF		 3054 /*
3055 * Once we've associated a backchannel xprt with a connection,
28303ca3
WM		 3056 * we want to keep it around as long as the connection lasts,
3057 * in case we need to start using it for a backchannel again;
3058 * this reference won't be dropped until bc_xprt is destroyed.
99de8ea9
BF		 3059 */
3060 xprt_get(xprt);
3061 args->bc_xprt->xpt_bc_xprt = xprt;
3062 xprt->bc_xprt = args->bc_xprt;
3063 bc_sock = container_of(args->bc_xprt, struct svc_sock, sk_xprt);
3064 transport->sock = bc_sock->sk_sock;
3065 transport->inet = bc_sock->sk_sk;
3066
f300baba
AB		 3067 /*
3068 * Since we don't want connections for the backchannel, we set
3069 * the xprt status to connected
3070 */
3071 xprt_set_connected(xprt);
3072
f300baba
AB		 3073 if (try_module_get(THIS_MODULE))
3074 return xprt;
642aab58
KM		 3075
3076 args->bc_xprt->xpt_bc_xprt = NULL;
39a9beab 3077 args->bc_xprt->xpt_bc_xps = NULL;
99de8ea9 3078 xprt_put(xprt);
0a68b0be
BF		 3079 ret = ERR_PTR(-EINVAL);
3080out_err:
315f3812 3081 xs_xprt_free(xprt);
0a68b0be 3082 return ret;
f300baba
AB		 3083}
3084
176e21ee
CL		 3085static struct xprt_class xs_local_transport = {
3086 .list = LIST_HEAD_INIT(xs_local_transport.list),
3087 .name = "named UNIX socket",
3088 .owner = THIS_MODULE,
3089 .ident = XPRT_TRANSPORT_LOCAL,
3090 .setup = xs_setup_local,
3091};
3092
bc25571e
TT		 3093static struct xprt_class xs_udp_transport = {
3094 .list = LIST_HEAD_INIT(xs_udp_transport.list),
3095 .name = "udp",
3096 .owner = THIS_MODULE,
f300baba 3097 .ident = XPRT_TRANSPORT_UDP,
bc25571e
TT		 3098 .setup = xs_setup_udp,
3099};
3100
3101static struct xprt_class xs_tcp_transport = {
3102 .list = LIST_HEAD_INIT(xs_tcp_transport.list),
3103 .name = "tcp",
3104 .owner = THIS_MODULE,
f300baba 3105 .ident = XPRT_TRANSPORT_TCP,
bc25571e
TT		 3106 .setup = xs_setup_tcp,
3107};
3108
f300baba
AB		 3109static struct xprt_class xs_bc_tcp_transport = {
3110 .list = LIST_HEAD_INIT(xs_bc_tcp_transport.list),
3111 .name = "tcp NFSv4.1 backchannel",
3112 .owner = THIS_MODULE,
3113 .ident = XPRT_TRANSPORT_BC_TCP,
3114 .setup = xs_setup_bc_tcp,
3115};
3116
282b32e1 3117/**
bc25571e 3118 * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
282b32e1
CL		 3119 *
3120 */
3121int init_socket_xprt(void)
3122{
f895b252 3123#if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
2b1bec5f 3124 if (!sunrpc_table_header)
0b4d4147 3125 sunrpc_table_header = register_sysctl_table(sunrpc_table);
fbf76683
CL		 3126#endif
3127
176e21ee 3128 xprt_register_transport(&xs_local_transport);
bc25571e
TT		 3129 xprt_register_transport(&xs_udp_transport);
3130 xprt_register_transport(&xs_tcp_transport);
f300baba 3131 xprt_register_transport(&xs_bc_tcp_transport);
bc25571e 3132
282b32e1
CL		 3133 return 0;
3134}
3135
3136/**
bc25571e 3137 * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
282b32e1
CL		 3138 *
3139 */
3140void cleanup_socket_xprt(void)
3141{
f895b252 3142#if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
fbf76683
CL		 3143 if (sunrpc_table_header) {
3144 unregister_sysctl_table(sunrpc_table_header);
3145 sunrpc_table_header = NULL;
3146 }
3147#endif
bc25571e 3148
176e21ee 3149 xprt_unregister_transport(&xs_local_transport);
bc25571e
TT		 3150 xprt_unregister_transport(&xs_udp_transport);
3151 xprt_unregister_transport(&xs_tcp_transport);
f300baba 3152 xprt_unregister_transport(&xs_bc_tcp_transport);
282b32e1 3153}
cbf11071 3154
9bbb9e5a
RR		 3155static int param_set_uint_minmax(const char *val,
3156 const struct kernel_param *kp,
cbf11071
TM		 3157 unsigned int min, unsigned int max)
3158{
00cfaa94 3159 unsigned int num;
cbf11071
TM		 3160 int ret;
3161
3162 if (!val)
3163 return -EINVAL;
00cfaa94 3164 ret = kstrtouint(val, 0, &num);
cbf11071
TM		 3165 if (ret == -EINVAL || num < min || num > max)
3166 return -EINVAL;
3167 *((unsigned int *)kp->arg) = num;
3168 return 0;
3169}
3170
9bbb9e5a 3171static int param_set_portnr(const char *val, const struct kernel_param *kp)
cbf11071 3172{
ffb6ca33
FS		 3173 if (kp->arg == &xprt_min_resvport)
3174 return param_set_uint_minmax(val, kp,
cbf11071 3175 RPC_MIN_RESVPORT,
ffb6ca33
FS		 3176 xprt_max_resvport);
3177 return param_set_uint_minmax(val, kp,
3178 xprt_min_resvport,
cbf11071
TM		 3179 RPC_MAX_RESVPORT);
3180}
3181
9c27847d 3182static const struct kernel_param_ops param_ops_portnr = {
9bbb9e5a
RR		 3183 .set = param_set_portnr,
3184 .get = param_get_uint,
3185};
3186
cbf11071
TM		 3187#define param_check_portnr(name, p) \
3188 __param_check(name, p, unsigned int);
3189
3190module_param_named(min_resvport, xprt_min_resvport, portnr, 0644);
3191module_param_named(max_resvport, xprt_max_resvport, portnr, 0644);
3192
9bbb9e5a
RR		 3193static int param_set_slot_table_size(const char *val,
3194 const struct kernel_param *kp)
cbf11071
TM		 3195{
3196 return param_set_uint_minmax(val, kp,
3197 RPC_MIN_SLOT_TABLE,
3198 RPC_MAX_SLOT_TABLE);
3199}
3200
9c27847d 3201static const struct kernel_param_ops param_ops_slot_table_size = {
9bbb9e5a
RR		 3202 .set = param_set_slot_table_size,
3203 .get = param_get_uint,
3204};
3205
cbf11071
TM		 3206#define param_check_slot_table_size(name, p) \
3207 __param_check(name, p, unsigned int);
3208
d9ba131d
TM		 3209static int param_set_max_slot_table_size(const char *val,
3210 const struct kernel_param *kp)
3211{
3212 return param_set_uint_minmax(val, kp,
3213 RPC_MIN_SLOT_TABLE,
3214 RPC_MAX_SLOT_TABLE_LIMIT);
3215}
3216
9c27847d 3217static const struct kernel_param_ops param_ops_max_slot_table_size = {
d9ba131d
TM		 3218 .set = param_set_max_slot_table_size,
3219 .get = param_get_uint,
3220};
3221
3222#define param_check_max_slot_table_size(name, p) \
3223 __param_check(name, p, unsigned int);
3224
cbf11071
TM		 3225module_param_named(tcp_slot_table_entries, xprt_tcp_slot_table_entries,
3226 slot_table_size, 0644);
d9ba131d
TM		 3227module_param_named(tcp_max_slot_table_entries, xprt_max_tcp_slot_table_entries,
3228 max_slot_table_size, 0644);
cbf11071
TM		 3229module_param_named(udp_slot_table_entries, xprt_udp_slot_table_entries,
3230 slot_table_size, 0644);
3231
Linux 6.x block layer and io_uring tree(s)