projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Merge branch 'stable/for-linus-4.6' of git://git.kernel.org/pub/scm/linux/kernel...
[linux-2.6-block.git] / net / sunrpc / xprtsock.c
CommitLineData
a246b010
CL		 1/*
2 * linux/net/sunrpc/xprtsock.c
3 *
4 * Client-side transport implementation for sockets.
5 *
113aa838
AC		 6 * TCP callback races fixes (C) 1998 Red Hat
7 * TCP send fixes (C) 1998 Red Hat
a246b010
CL		 8 * TCP NFS related read + write fixes
9 * (C) 1999 Dave Airlie, University of Limerick, Ireland <airlied@linux.ie>
10 *
11 * Rewrite of larges part of the code in order to stabilize TCP stuff.
12 * Fix behaviour when socket buffer is full.
13 * (C) 1999 Trond Myklebust <trond.myklebust@fys.uio.no>
55aa4f58
CL		 14 *
15 * IP socket transport implementation, (C) 2005 Chuck Lever <cel@netapp.com>
8f9d5b1a
CL		 16 *
17 * IPv6 support contributed by Gilles Quillard, Bull Open Source, 2005.
18 * <gilles.quillard@bull.net>
a246b010
CL		 19 */
20
21#include <linux/types.h>
176e21ee 22#include <linux/string.h>
a246b010 23#include <linux/slab.h>
bc25571e 24#include <linux/module.h>
a246b010 25#include <linux/capability.h>
a246b010
CL		 26#include <linux/pagemap.h>
27#include <linux/errno.h>
28#include <linux/socket.h>
29#include <linux/in.h>
30#include <linux/net.h>
31#include <linux/mm.h>
176e21ee 32#include <linux/un.h>
a246b010
CL		 33#include <linux/udp.h>
34#include <linux/tcp.h>
35#include <linux/sunrpc/clnt.h>
5976687a 36#include <linux/sunrpc/addr.h>
02107148 37#include <linux/sunrpc/sched.h>
4cfc7e60 38#include <linux/sunrpc/svcsock.h>
49c36fcc 39#include <linux/sunrpc/xprtsock.h>
a246b010 40#include <linux/file.h>
9e00abc3 41#ifdef CONFIG_SUNRPC_BACKCHANNEL
44b98efd
RL		 42#include <linux/sunrpc/bc_xprt.h>
43#endif
a246b010
CL		 44
45#include <net/sock.h>
46#include <net/checksum.h>
47#include <net/udp.h>
48#include <net/tcp.h>
49
40b5ea0c
TM		 50#include <trace/events/sunrpc.h>
51
4cfc7e60 52#include "sunrpc.h"
176e21ee
CL		 53
54static void xs_close(struct rpc_xprt *xprt);
55
c556b754
CL		 56/*
57 * xprtsock tunables
58 */
09acfea5
TM		 59static unsigned int xprt_udp_slot_table_entries = RPC_DEF_SLOT_TABLE;
60static unsigned int xprt_tcp_slot_table_entries = RPC_MIN_SLOT_TABLE;
61static unsigned int xprt_max_tcp_slot_table_entries = RPC_MAX_SLOT_TABLE;
c556b754 62
09acfea5
TM		 63static unsigned int xprt_min_resvport = RPC_DEF_MIN_RESVPORT;
64static unsigned int xprt_max_resvport = RPC_DEF_MAX_RESVPORT;
c556b754 65
54c09874
TM		 66#if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
67
7d1e8255 68#define XS_TCP_LINGER_TO (15U * HZ)
25fe6142 69static unsigned int xs_tcp_fin_timeout __read_mostly = XS_TCP_LINGER_TO;
7d1e8255 70
fbf76683
CL		 71/*
72 * We can register our own files under /proc/sys/sunrpc by
73 * calling register_sysctl_table() again. The files in that
74 * directory become the union of all files registered there.
75 *
76 * We simply need to make sure that we don't collide with
77 * someone else's file names!
78 */
79
fbf76683
CL		 80static unsigned int min_slot_table_size = RPC_MIN_SLOT_TABLE;
81static unsigned int max_slot_table_size = RPC_MAX_SLOT_TABLE;
d9ba131d 82static unsigned int max_tcp_slot_table_limit = RPC_MAX_SLOT_TABLE_LIMIT;
fbf76683
CL		 83static unsigned int xprt_min_resvport_limit = RPC_MIN_RESVPORT;
84static unsigned int xprt_max_resvport_limit = RPC_MAX_RESVPORT;
85
86static struct ctl_table_header *sunrpc_table_header;
87
88/*
89 * FIXME: changing the UDP slot table size should also resize the UDP
90 * socket buffers for existing UDP transports
91 */
fe2c6338 92static struct ctl_table xs_tunables_table[] = {
fbf76683 93 {
fbf76683
CL		 94 .procname = "udp_slot_table_entries",
95 .data = &xprt_udp_slot_table_entries,
96 .maxlen = sizeof(unsigned int),
97 .mode = 0644,
6d456111 98 .proc_handler = proc_dointvec_minmax,
fbf76683
CL		 99 .extra1 = &min_slot_table_size,
100 .extra2 = &max_slot_table_size
101 },
102 {
fbf76683
CL		 103 .procname = "tcp_slot_table_entries",
104 .data = &xprt_tcp_slot_table_entries,
105 .maxlen = sizeof(unsigned int),
106 .mode = 0644,
6d456111 107 .proc_handler = proc_dointvec_minmax,
fbf76683
CL		 108 .extra1 = &min_slot_table_size,
109 .extra2 = &max_slot_table_size
110 },
d9ba131d
TM		 111 {
112 .procname = "tcp_max_slot_table_entries",
113 .data = &xprt_max_tcp_slot_table_entries,
114 .maxlen = sizeof(unsigned int),
115 .mode = 0644,
116 .proc_handler = proc_dointvec_minmax,
117 .extra1 = &min_slot_table_size,
118 .extra2 = &max_tcp_slot_table_limit
119 },
fbf76683 120 {
fbf76683
CL		 121 .procname = "min_resvport",
122 .data = &xprt_min_resvport,
123 .maxlen = sizeof(unsigned int),
124 .mode = 0644,
6d456111 125 .proc_handler = proc_dointvec_minmax,
fbf76683
CL		 126 .extra1 = &xprt_min_resvport_limit,
127 .extra2 = &xprt_max_resvport_limit
128 },
129 {
fbf76683
CL		 130 .procname = "max_resvport",
131 .data = &xprt_max_resvport,
132 .maxlen = sizeof(unsigned int),
133 .mode = 0644,
6d456111 134 .proc_handler = proc_dointvec_minmax,
fbf76683
CL		 135 .extra1 = &xprt_min_resvport_limit,
136 .extra2 = &xprt_max_resvport_limit
137 },
25fe6142
TM		 138 {
139 .procname = "tcp_fin_timeout",
140 .data = &xs_tcp_fin_timeout,
141 .maxlen = sizeof(xs_tcp_fin_timeout),
142 .mode = 0644,
6d456111 143 .proc_handler = proc_dointvec_jiffies,
fbf76683 144 },
f8572d8f 145 { },
fbf76683
CL		 146};
147
fe2c6338 148static struct ctl_table sunrpc_table[] = {
fbf76683 149 {
fbf76683
CL		 150 .procname = "sunrpc",
151 .mode = 0555,
152 .child = xs_tunables_table
153 },
f8572d8f 154 { },
fbf76683
CL		 155};
156
157#endif
158
03bf4b70
CL		 159/*
160 * Wait duration for a reply from the RPC portmapper.
161 */
162#define XS_BIND_TO (60U * HZ)
163
164/*
165 * Delay if a UDP socket connect error occurs. This is most likely some
166 * kind of resource problem on the local host.
167 */
168#define XS_UDP_REEST_TO (2U * HZ)
169
170/*
171 * The reestablish timeout allows clients to delay for a bit before attempting
172 * to reconnect to a server that just dropped our connection.
173 *
174 * We implement an exponential backoff when trying to reestablish a TCP
175 * transport connection with the server. Some servers like to drop a TCP
176 * connection when they are overworked, so we start with a short timeout and
177 * increase over time if the server is down or not responding.
178 */
179#define XS_TCP_INIT_REEST_TO (3U * HZ)
180#define XS_TCP_MAX_REEST_TO (5U * 60 * HZ)
181
182/*
183 * TCP idle timeout; client drops the transport socket if it is idle
184 * for this long. Note that we also timeout UDP sockets to prevent
185 * holding port numbers when there is no RPC traffic.
186 */
187#define XS_IDLE_DISC_TO (5U * 60 * HZ)
188
f895b252 189#if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
a246b010 190# undef RPC_DEBUG_DATA
9903cd1c 191# define RPCDBG_FACILITY RPCDBG_TRANS
a246b010
CL		 192#endif
193
a246b010 194#ifdef RPC_DEBUG_DATA
9903cd1c 195static void xs_pktdump(char *msg, u32 *packet, unsigned int count)
a246b010 196{
9903cd1c
CL		 197 u8 *buf = (u8 *) packet;
198 int j;
a246b010 199
46121cf7 200 dprintk("RPC: %s\n", msg);
a246b010
CL		 201 for (j = 0; j < count && j < 128; j += 4) {
202 if (!(j & 31)) {
203 if (j)
204 dprintk("\n");
205 dprintk("0x%04x ", j);
206 }
207 dprintk("%02x%02x%02x%02x ",
208 buf[j], buf[j+1], buf[j+2], buf[j+3]);
209 }
210 dprintk("\n");
211}
212#else
9903cd1c 213static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count)
a246b010
CL		 214{
215 /* NOP */
216}
217#endif
218
2118071d
TM		 219static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
220{
221 return (struct rpc_xprt *) sk->sk_user_data;
222}
223
95392c59
CL		 224static inline struct sockaddr *xs_addr(struct rpc_xprt *xprt)
225{
226 return (struct sockaddr *) &xprt->addr;
227}
228
176e21ee
CL		 229static inline struct sockaddr_un *xs_addr_un(struct rpc_xprt *xprt)
230{
231 return (struct sockaddr_un *) &xprt->addr;
232}
233
95392c59 234static inline struct sockaddr_in *xs_addr_in(struct rpc_xprt *xprt)
edb267a6 235{
95392c59
CL		 236 return (struct sockaddr_in *) &xprt->addr;
237}
238
239static inline struct sockaddr_in6 *xs_addr_in6(struct rpc_xprt *xprt)
240{
241 return (struct sockaddr_in6 *) &xprt->addr;
242}
243
c877b849 244static void xs_format_common_peer_addresses(struct rpc_xprt *xprt)
edb267a6 245{
c877b849 246 struct sockaddr *sap = xs_addr(xprt);
9dc3b095
CL		 247 struct sockaddr_in6 *sin6;
248 struct sockaddr_in *sin;
176e21ee 249 struct sockaddr_un *sun;
c877b849 250 char buf[128];
edb267a6 251
9dc3b095 252 switch (sap->sa_family) {
176e21ee
CL		 253 case AF_LOCAL:
254 sun = xs_addr_un(xprt);
255 strlcpy(buf, sun->sun_path, sizeof(buf));
256 xprt->address_strings[RPC_DISPLAY_ADDR] =
257 kstrdup(buf, GFP_KERNEL);
258 break;
9dc3b095 259 case AF_INET:
176e21ee
CL		 260 (void)rpc_ntop(sap, buf, sizeof(buf));
261 xprt->address_strings[RPC_DISPLAY_ADDR] =
262 kstrdup(buf, GFP_KERNEL);
9dc3b095 263 sin = xs_addr_in(xprt);
fc0b5791 264 snprintf(buf, sizeof(buf), "%08x", ntohl(sin->sin_addr.s_addr));
9dc3b095
CL		 265 break;
266 case AF_INET6:
176e21ee
CL		 267 (void)rpc_ntop(sap, buf, sizeof(buf));
268 xprt->address_strings[RPC_DISPLAY_ADDR] =
269 kstrdup(buf, GFP_KERNEL);
9dc3b095 270 sin6 = xs_addr_in6(xprt);
fc0b5791 271 snprintf(buf, sizeof(buf), "%pi6", &sin6->sin6_addr);
9dc3b095
CL		 272 break;
273 default:
274 BUG();
756805e7 275 }
176e21ee 276
9dc3b095 277 xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = kstrdup(buf, GFP_KERNEL);
edb267a6
CL		 278}
279
9dc3b095 280static void xs_format_common_peer_ports(struct rpc_xprt *xprt)
4b6473fb 281{
9dc3b095
CL		 282 struct sockaddr *sap = xs_addr(xprt);
283 char buf[128];
4b6473fb 284
81160e66 285 snprintf(buf, sizeof(buf), "%u", rpc_get_port(sap));
c877b849 286 xprt->address_strings[RPC_DISPLAY_PORT] = kstrdup(buf, GFP_KERNEL);
4b6473fb 287
81160e66 288 snprintf(buf, sizeof(buf), "%4hx", rpc_get_port(sap));
c877b849
CL		 289 xprt->address_strings[RPC_DISPLAY_HEX_PORT] = kstrdup(buf, GFP_KERNEL);
290}
4b6473fb 291
9dc3b095
CL		 292static void xs_format_peer_addresses(struct rpc_xprt *xprt,
293 const char *protocol,
294 const char *netid)
c877b849 295{
b454ae90 296 xprt->address_strings[RPC_DISPLAY_PROTO] = protocol;
b454ae90 297 xprt->address_strings[RPC_DISPLAY_NETID] = netid;
c877b849 298 xs_format_common_peer_addresses(xprt);
9dc3b095 299 xs_format_common_peer_ports(xprt);
edb267a6 300}
4b6473fb 301
9dc3b095 302static void xs_update_peer_port(struct rpc_xprt *xprt)
4b6473fb 303{
9dc3b095
CL		 304 kfree(xprt->address_strings[RPC_DISPLAY_HEX_PORT]);
305 kfree(xprt->address_strings[RPC_DISPLAY_PORT]);
4417c8c4 306
9dc3b095 307 xs_format_common_peer_ports(xprt);
edb267a6
CL		 308}
309
310static void xs_free_peer_addresses(struct rpc_xprt *xprt)
311{
33e01dc7
CL		 312 unsigned int i;
313
314 for (i = 0; i < RPC_DISPLAY_MAX; i++)
315 switch (i) {
316 case RPC_DISPLAY_PROTO:
317 case RPC_DISPLAY_NETID:
318 continue;
319 default:
320 kfree(xprt->address_strings[i]);
321 }
edb267a6
CL		 322}
323
b4b5cc85
CL		 324#define XS_SENDMSG_FLAGS (MSG_DONTWAIT | MSG_NOSIGNAL)
325
24c5684b 326static int xs_send_kvec(struct socket *sock, struct sockaddr *addr, int addrlen, struct kvec *vec, unsigned int base, int more)
b4b5cc85 327{
b4b5cc85
CL		 328 struct msghdr msg = {
329 .msg_name = addr,
330 .msg_namelen = addrlen,
24c5684b
TM		 331 .msg_flags = XS_SENDMSG_FLAGS | (more ? MSG_MORE : 0),
332 };
333 struct kvec iov = {
334 .iov_base = vec->iov_base + base,
335 .iov_len = vec->iov_len - base,
b4b5cc85
CL		 336 };
337
24c5684b 338 if (iov.iov_len != 0)
b4b5cc85
CL		 339 return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
340 return kernel_sendmsg(sock, &msg, NULL, 0, 0);
341}
342
f279cd00 343static int xs_send_pagedata(struct socket *sock, struct xdr_buf *xdr, unsigned int base, int more, bool zerocopy, int *sent_p)
b4b5cc85 344{
a6b31d18
TM		 345 ssize_t (*do_sendpage)(struct socket *sock, struct page *page,
346 int offset, size_t size, int flags);
24c5684b
TM		 347 struct page **ppage;
348 unsigned int remainder;
f279cd00 349 int err;
24c5684b
TM		 350
351 remainder = xdr->page_len - base;
352 base += xdr->page_base;
353 ppage = xdr->pages + (base >> PAGE_SHIFT);
354 base &= ~PAGE_MASK;
a6b31d18
TM		 355 do_sendpage = sock->ops->sendpage;
356 if (!zerocopy)
357 do_sendpage = sock_no_sendpage;
24c5684b
TM		 358 for(;;) {
359 unsigned int len = min_t(unsigned int, PAGE_SIZE - base, remainder);
360 int flags = XS_SENDMSG_FLAGS;
b4b5cc85 361
24c5684b 362 remainder -= len;
31303d6c 363 if (more)
24c5684b 364 flags |= MSG_MORE;
31303d6c
TM		 365 if (remainder != 0)
366 flags |= MSG_SENDPAGE_NOTLAST | MSG_MORE;
a6b31d18 367 err = do_sendpage(sock, *ppage, base, len, flags);
24c5684b
TM		 368 if (remainder == 0 || err != len)
369 break;
f279cd00 370 *sent_p += err;
24c5684b
TM		 371 ppage++;
372 base = 0;
373 }
f279cd00
JB		 374 if (err > 0) {
375 *sent_p += err;
376 err = 0;
377 }
378 return err;
b4b5cc85
CL		 379}
380
9903cd1c
CL		 381/**
382 * xs_sendpages - write pages directly to a socket
383 * @sock: socket to send on
384 * @addr: UDP only -- address of destination
385 * @addrlen: UDP only -- length of destination address
386 * @xdr: buffer containing this request
387 * @base: starting position in the buffer
a6b31d18 388 * @zerocopy: true if it is safe to use sendpage()
f279cd00 389 * @sent_p: return the total number of bytes successfully queued for sending
9903cd1c 390 *
a246b010 391 */
f279cd00 392static int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base, bool zerocopy, int *sent_p)
a246b010 393{
24c5684b 394 unsigned int remainder = xdr->len - base;
f279cd00
JB		 395 int err = 0;
396 int sent = 0;
a246b010 397
262965f5 398 if (unlikely(!sock))
fba91afb 399 return -ENOTSOCK;
262965f5 400
24c5684b
TM		 401 if (base != 0) {
402 addr = NULL;
403 addrlen = 0;
404 }
262965f5 405
24c5684b
TM		 406 if (base < xdr->head[0].iov_len || addr != NULL) {
407 unsigned int len = xdr->head[0].iov_len - base;
408 remainder -= len;
409 err = xs_send_kvec(sock, addr, addrlen, &xdr->head[0], base, remainder != 0);
410 if (remainder == 0 || err != len)
a246b010 411 goto out;
f279cd00 412 *sent_p += err;
a246b010
CL		 413 base = 0;
414 } else
24c5684b 415 base -= xdr->head[0].iov_len;
a246b010 416
24c5684b
TM		 417 if (base < xdr->page_len) {
418 unsigned int len = xdr->page_len - base;
419 remainder -= len;
f279cd00
JB		 420 err = xs_send_pagedata(sock, xdr, base, remainder != 0, zerocopy, &sent);
421 *sent_p += sent;
422 if (remainder == 0 || sent != len)
a246b010
CL		 423 goto out;
424 base = 0;
24c5684b
TM		 425 } else
426 base -= xdr->page_len;
427
428 if (base >= xdr->tail[0].iov_len)
f279cd00 429 return 0;
24c5684b 430 err = xs_send_kvec(sock, NULL, 0, &xdr->tail[0], base, 0);
a246b010 431out:
f279cd00
JB		 432 if (err > 0) {
433 *sent_p += err;
434 err = 0;
435 }
436 return err;
a246b010
CL		 437}
438
b6ddf64f
TM		 439static void xs_nospace_callback(struct rpc_task *task)
440{
441 struct sock_xprt *transport = container_of(task->tk_rqstp->rq_xprt, struct sock_xprt, xprt);
442
443 transport->inet->sk_write_pending--;
b6ddf64f
TM		 444}
445
9903cd1c 446/**
262965f5
CL		 447 * xs_nospace - place task on wait queue if transmit was incomplete
448 * @task: task to put to sleep
9903cd1c 449 *
a246b010 450 */
5e3771ce 451static int xs_nospace(struct rpc_task *task)
a246b010 452{
262965f5
CL		 453 struct rpc_rqst *req = task->tk_rqstp;
454 struct rpc_xprt *xprt = req->rq_xprt;
ee0ac0c2 455 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
06ea0bfe 456 struct sock *sk = transport->inet;
24ca9a84 457 int ret = -EAGAIN;
a246b010 458
46121cf7 459 dprintk("RPC: %5u xmit incomplete (%u left of %u)\n",
262965f5
CL		 460 task->tk_pid, req->rq_slen - req->rq_bytes_sent,
461 req->rq_slen);
462
b6ddf64f
TM		 463 /* Protect against races with write_space */
464 spin_lock_bh(&xprt->transport_lock);
465
466 /* Don't race with disconnect */
467 if (xprt_connected(xprt)) {
13331a55
TM		 468 /* wait for more buffer space */
469 sk->sk_write_pending++;
470 xprt_wait_for_buffer_space(task, xs_nospace_callback);
471 } else
5e3771ce 472 ret = -ENOTCONN;
262965f5 473
b6ddf64f 474 spin_unlock_bh(&xprt->transport_lock);
06ea0bfe
TM		 475
476 /* Race breaker in case memory is freed before above code is called */
477 sk->sk_write_space(sk);
5e3771ce 478 return ret;
262965f5
CL		 479}
480
61677eee
CL		 481/*
482 * Construct a stream transport record marker in @buf.
483 */
484static inline void xs_encode_stream_record_marker(struct xdr_buf *buf)
485{
486 u32 reclen = buf->len - sizeof(rpc_fraghdr);
487 rpc_fraghdr *base = buf->head[0].iov_base;
488 *base = cpu_to_be32(RPC_LAST_STREAM_FRAGMENT | reclen);
489}
490
176e21ee
CL		 491/**
492 * xs_local_send_request - write an RPC request to an AF_LOCAL socket
493 * @task: RPC task that manages the state of an RPC request
494 *
495 * Return values:
496 * 0: The request has been sent
497 * EAGAIN: The socket was blocked, please call again later to
498 * complete the request
499 * ENOTCONN: Caller needs to invoke connect logic then call again
500 * other: Some other error occured, the request was not sent
501 */
502static int xs_local_send_request(struct rpc_task *task)
503{
504 struct rpc_rqst *req = task->tk_rqstp;
505 struct rpc_xprt *xprt = req->rq_xprt;
506 struct sock_xprt *transport =
507 container_of(xprt, struct sock_xprt, xprt);
508 struct xdr_buf *xdr = &req->rq_snd_buf;
509 int status;
f279cd00 510 int sent = 0;
176e21ee
CL		 511
512 xs_encode_stream_record_marker(&req->rq_snd_buf);
513
514 xs_pktdump("packet data:",
515 req->rq_svec->iov_base, req->rq_svec->iov_len);
516
f279cd00
JB		 517 status = xs_sendpages(transport->sock, NULL, 0, xdr, req->rq_bytes_sent,
518 true, &sent);
176e21ee
CL		 519 dprintk("RPC: %s(%u) = %d\n",
520 __func__, xdr->len - req->rq_bytes_sent, status);
743c69e7
N		 521
522 if (status == -EAGAIN && sock_writeable(transport->inet))
523 status = -ENOBUFS;
524
f279cd00
JB		 525 if (likely(sent > 0) || status == 0) {
526 req->rq_bytes_sent += sent;
527 req->rq_xmit_bytes_sent += sent;
176e21ee
CL		 528 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
529 req->rq_bytes_sent = 0;
530 return 0;
531 }
532 status = -EAGAIN;
533 }
534
535 switch (status) {
3601c4a9 536 case -ENOBUFS:
b5872f0c 537 break;
176e21ee
CL		 538 case -EAGAIN:
539 status = xs_nospace(task);
540 break;
541 default:
542 dprintk("RPC: sendmsg returned unrecognized error %d\n",
543 -status);
544 case -EPIPE:
545 xs_close(xprt);
546 status = -ENOTCONN;
547 }
548
549 return status;
550}
551
262965f5
CL		 552/**
553 * xs_udp_send_request - write an RPC request to a UDP socket
554 * @task: address of RPC task that manages the state of an RPC request
555 *
556 * Return values:
557 * 0: The request has been sent
558 * EAGAIN: The socket was blocked, please call again later to
559 * complete the request
560 * ENOTCONN: Caller needs to invoke connect logic then call again
25985edc 561 * other: Some other error occurred, the request was not sent
262965f5
CL		 562 */
563static int xs_udp_send_request(struct rpc_task *task)
564{
565 struct rpc_rqst *req = task->tk_rqstp;
566 struct rpc_xprt *xprt = req->rq_xprt;
ee0ac0c2 567 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
262965f5 568 struct xdr_buf *xdr = &req->rq_snd_buf;
f279cd00 569 int sent = 0;
262965f5 570 int status;
a246b010 571
9903cd1c 572 xs_pktdump("packet data:",
a246b010
CL		 573 req->rq_svec->iov_base,
574 req->rq_svec->iov_len);
575
01d37c42
TM		 576 if (!xprt_bound(xprt))
577 return -ENOTCONN;
f279cd00
JB		 578 status = xs_sendpages(transport->sock, xs_addr(xprt), xprt->addrlen,
579 xdr, req->rq_bytes_sent, true, &sent);
a246b010 580
46121cf7 581 dprintk("RPC: xs_udp_send_request(%u) = %d\n",
262965f5 582 xdr->len - req->rq_bytes_sent, status);
a246b010 583
3dedbb5c
JB		 584 /* firewall is blocking us, don't return -EAGAIN or we end up looping */
585 if (status == -EPERM)
586 goto process_status;
587
743c69e7
N		 588 if (status == -EAGAIN && sock_writeable(transport->inet))
589 status = -ENOBUFS;
590
f279cd00
JB		 591 if (sent > 0 || status == 0) {
592 req->rq_xmit_bytes_sent += sent;
593 if (sent >= req->rq_slen)
2199700f
TM		 594 return 0;
595 /* Still some bytes left; set up for a retry later. */
262965f5 596 status = -EAGAIN;
2199700f 597 }
a246b010 598
3dedbb5c 599process_status:
262965f5 600 switch (status) {
fba91afb
TM		 601 case -ENOTSOCK:
602 status = -ENOTCONN;
603 /* Should we call xs_close() here? */
604 break;
b6ddf64f 605 case -EAGAIN:
5e3771ce 606 status = xs_nospace(task);
b6ddf64f 607 break;
262965f5 608 case -ENETUNREACH:
3601c4a9 609 case -ENOBUFS:
262965f5 610 case -EPIPE:
a246b010 611 case -ECONNREFUSED:
3dedbb5c 612 case -EPERM:
a246b010 613 /* When the server has died, an ICMP port unreachable message
9903cd1c 614 * prompts ECONNREFUSED. */
13331a55
TM		 615 break;
616 default:
617 dprintk("RPC: sendmsg returned unrecognized error %d\n",
618 -status);
a246b010 619 }
5fe46e9d 620
262965f5 621 return status;
a246b010
CL		 622}
623
9903cd1c 624/**
262965f5 625 * xs_tcp_send_request - write an RPC request to a TCP socket
9903cd1c
CL		 626 * @task: address of RPC task that manages the state of an RPC request
627 *
628 * Return values:
262965f5
CL		 629 * 0: The request has been sent
630 * EAGAIN: The socket was blocked, please call again later to
631 * complete the request
632 * ENOTCONN: Caller needs to invoke connect logic then call again
25985edc 633 * other: Some other error occurred, the request was not sent
9903cd1c
CL		 634 *
635 * XXX: In the case of soft timeouts, should we eventually give up
262965f5 636 * if sendmsg is not able to make progress?
9903cd1c 637 */
262965f5 638static int xs_tcp_send_request(struct rpc_task *task)
a246b010
CL		 639{
640 struct rpc_rqst *req = task->tk_rqstp;
641 struct rpc_xprt *xprt = req->rq_xprt;
ee0ac0c2 642 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
262965f5 643 struct xdr_buf *xdr = &req->rq_snd_buf;
a6b31d18 644 bool zerocopy = true;
b595bb15 645 int status;
f279cd00 646 int sent;
a246b010 647
61677eee 648 xs_encode_stream_record_marker(&req->rq_snd_buf);
a246b010 649
262965f5
CL		 650 xs_pktdump("packet data:",
651 req->rq_svec->iov_base,
652 req->rq_svec->iov_len);
a6b31d18
TM		 653 /* Don't use zero copy if this is a resend. If the RPC call
654 * completes while the socket holds a reference to the pages,
655 * then we may end up resending corrupted data.
656 */
657 if (task->tk_flags & RPC_TASK_SENT)
658 zerocopy = false;
a246b010
CL		 659
660 /* Continue transmitting the packet/record. We must be careful
661 * to cope with writespace callbacks arriving _after_ we have
262965f5 662 * called sendmsg(). */
a246b010 663 while (1) {
f279cd00
JB		 664 sent = 0;
665 status = xs_sendpages(transport->sock, NULL, 0, xdr,
666 req->rq_bytes_sent, zerocopy, &sent);
a246b010 667
46121cf7 668 dprintk("RPC: xs_tcp_send_request(%u) = %d\n",
262965f5 669 xdr->len - req->rq_bytes_sent, status);
a246b010 670
262965f5
CL		 671 /* If we've sent the entire packet, immediately
672 * reset the count of bytes sent. */
f279cd00
JB		 673 req->rq_bytes_sent += sent;
674 req->rq_xmit_bytes_sent += sent;
262965f5
CL		 675 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
676 req->rq_bytes_sent = 0;
677 return 0;
678 }
a246b010 679
f580dd04
TM		 680 if (status < 0)
681 break;
682 if (sent == 0) {
683 status = -EAGAIN;
684 break;
685 }
a246b010 686 }
743c69e7
N		 687 if (status == -EAGAIN && sk_stream_is_writeable(transport->inet))
688 status = -ENOBUFS;
a246b010 689
262965f5 690 switch (status) {
fba91afb
TM		 691 case -ENOTSOCK:
692 status = -ENOTCONN;
693 /* Should we call xs_close() here? */
694 break;
262965f5 695 case -EAGAIN:
5e3771ce 696 status = xs_nospace(task);
262965f5 697 break;
262965f5 698 case -ECONNRESET:
2a9e1cfa 699 case -ECONNREFUSED:
262965f5 700 case -ENOTCONN:
3913c78c 701 case -EADDRINUSE:
b5872f0c 702 case -ENOBUFS:
b9d2bb2e 703 case -EPIPE:
13331a55
TM		 704 break;
705 default:
706 dprintk("RPC: sendmsg returned unrecognized error %d\n",
707 -status);
a246b010 708 }
5fe46e9d 709
a246b010
CL		 710 return status;
711}
712
e0ab53de
TM		 713/**
714 * xs_tcp_release_xprt - clean up after a tcp transmission
715 * @xprt: transport
716 * @task: rpc task
717 *
718 * This cleans up if an error causes us to abort the transmission of a request.
719 * In this case, the socket may need to be reset in order to avoid confusing
720 * the server.
721 */
722static void xs_tcp_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
723{
724 struct rpc_rqst *req;
725
726 if (task != xprt->snd_task)
727 return;
728 if (task == NULL)
729 goto out_release;
730 req = task->tk_rqstp;
43cedbf0
TM		 731 if (req == NULL)
732 goto out_release;
e0ab53de
TM		 733 if (req->rq_bytes_sent == 0)
734 goto out_release;
735 if (req->rq_bytes_sent == req->rq_snd_buf.len)
736 goto out_release;
a4f0835c 737 set_bit(XPRT_CLOSE_WAIT, &xprt->state);
e0ab53de
TM		 738out_release:
739 xprt_release_xprt(xprt, task);
740}
741
2a9e1cfa
TM		 742static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk)
743{
744 transport->old_data_ready = sk->sk_data_ready;
745 transport->old_state_change = sk->sk_state_change;
746 transport->old_write_space = sk->sk_write_space;
2118071d 747 transport->old_error_report = sk->sk_error_report;
2a9e1cfa
TM		 748}
749
750static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk)
751{
752 sk->sk_data_ready = transport->old_data_ready;
753 sk->sk_state_change = transport->old_state_change;
754 sk->sk_write_space = transport->old_write_space;
2118071d
TM		 755 sk->sk_error_report = transport->old_error_report;
756}
757
b70ae915
TM		 758static void xs_sock_reset_connection_flags(struct rpc_xprt *xprt)
759{
760 smp_mb__before_atomic();
761 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
762 clear_bit(XPRT_CLOSING, &xprt->state);
763 smp_mb__after_atomic();
764}
765
766static void xs_sock_mark_closed(struct rpc_xprt *xprt)
767{
768 xs_sock_reset_connection_flags(xprt);
769 /* Mark transport as closed and wake up all pending tasks */
770 xprt_disconnect_done(xprt);
771}
772
2118071d
TM		 773/**
774 * xs_error_report - callback to handle TCP socket state errors
775 * @sk: socket
776 *
777 * Note: we don't call sock_error() since there may be a rpc_task
778 * using the socket, and so we don't want to clear sk->sk_err.
779 */
780static void xs_error_report(struct sock *sk)
781{
782 struct rpc_xprt *xprt;
783 int err;
784
785 read_lock_bh(&sk->sk_callback_lock);
786 if (!(xprt = xprt_from_sock(sk)))
787 goto out;
788
789 err = -sk->sk_err;
790 if (err == 0)
791 goto out;
b70ae915
TM		 792 /* Is this a reset event? */
793 if (sk->sk_state == TCP_CLOSE)
794 xs_sock_mark_closed(xprt);
2118071d
TM		 795 dprintk("RPC: xs_error_report client %p, error=%d...\n",
796 xprt, -err);
e8353c76 797 trace_rpc_socket_error(xprt, sk->sk_socket, err);
2118071d
TM		 798 xprt_wake_pending_tasks(xprt, err);
799 out:
800 read_unlock_bh(&sk->sk_callback_lock);
2a9e1cfa
TM		 801}
802
fe315e76 803static void xs_reset_transport(struct sock_xprt *transport)
a246b010 804{
ee0ac0c2
CL		 805 struct socket *sock = transport->sock;
806 struct sock *sk = transport->inet;
6cc7e908 807 struct rpc_xprt *xprt = &transport->xprt;
a246b010 808
fe315e76
CL		 809 if (sk == NULL)
810 return;
9903cd1c 811
264d1df3
JL		 812 if (atomic_read(&transport->xprt.swapper))
813 sk_clear_memalloc(sk);
814
09939204
TM		 815 kernel_sock_shutdown(sock, SHUT_RDWR);
816
edc1b01c 817 mutex_lock(&transport->recv_mutex);
a246b010 818 write_lock_bh(&sk->sk_callback_lock);
ee0ac0c2
CL		 819 transport->inet = NULL;
820 transport->sock = NULL;
a246b010 821
9903cd1c 822 sk->sk_user_data = NULL;
2a9e1cfa
TM		 823
824 xs_restore_old_callbacks(transport, sk);
0c78789e 825 xprt_clear_connected(xprt);
a246b010 826 write_unlock_bh(&sk->sk_callback_lock);
6cc7e908 827 xs_sock_reset_connection_flags(xprt);
edc1b01c 828 mutex_unlock(&transport->recv_mutex);
a246b010 829
6cc7e908 830 trace_rpc_socket_close(xprt, sock);
a246b010 831 sock_release(sock);
fe315e76
CL		 832}
833
834/**
835 * xs_close - close a socket
836 * @xprt: transport
837 *
838 * This is used when all requests are complete; ie, no DRC state remains
839 * on the server we want to save.
f75e6745
TM		 840 *
841 * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with
842 * xs_reset_transport() zeroing the socket from underneath a writer.
fe315e76
CL		 843 */
844static void xs_close(struct rpc_xprt *xprt)
845{
846 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
847
848 dprintk("RPC: xs_close xprt %p\n", xprt);
849
850 xs_reset_transport(transport);
61d0a8e6 851 xprt->reestablish_timeout = 0;
fe315e76 852
62da3b24 853 xprt_disconnect_done(xprt);
a246b010
CL		 854}
855
4a068258
CL		 856static void xs_inject_disconnect(struct rpc_xprt *xprt)
857{
858 dprintk("RPC: injecting transport disconnect on xprt=%p\n",
859 xprt);
860 xprt_disconnect_done(xprt);
861}
862
315f3812
KM		 863static void xs_xprt_free(struct rpc_xprt *xprt)
864{
865 xs_free_peer_addresses(xprt);
866 xprt_free(xprt);
867}
868
9903cd1c
CL		 869/**
870 * xs_destroy - prepare to shutdown a transport
871 * @xprt: doomed transport
872 *
873 */
874static void xs_destroy(struct rpc_xprt *xprt)
a246b010 875{
03c78827
TM		 876 struct sock_xprt *transport = container_of(xprt,
877 struct sock_xprt, xprt);
46121cf7 878 dprintk("RPC: xs_destroy xprt %p\n", xprt);
9903cd1c 879
03c78827 880 cancel_delayed_work_sync(&transport->connect_worker);
a1311d87 881 xs_close(xprt);
edc1b01c 882 cancel_work_sync(&transport->recv_worker);
315f3812 883 xs_xprt_free(xprt);
a1311d87 884 module_put(THIS_MODULE);
a246b010
CL		 885}
886
176e21ee
CL		 887static int xs_local_copy_to_xdr(struct xdr_buf *xdr, struct sk_buff *skb)
888{
889 struct xdr_skb_reader desc = {
890 .skb = skb,
891 .offset = sizeof(rpc_fraghdr),
892 .count = skb->len - sizeof(rpc_fraghdr),
893 };
894
895 if (xdr_partial_copy_from_skb(xdr, 0, &desc, xdr_skb_read_bits) < 0)
896 return -1;
897 if (desc.count)
898 return -1;
899 return 0;
900}
901
902/**
a2648094
TM		 903 * xs_local_data_read_skb
904 * @xprt: transport
905 * @sk: socket
906 * @skb: skbuff
176e21ee
CL		 907 *
908 * Currently this assumes we can read the whole reply in a single gulp.
909 */
a2648094
TM		 910static void xs_local_data_read_skb(struct rpc_xprt *xprt,
911 struct sock *sk,
912 struct sk_buff *skb)
176e21ee
CL		 913{
914 struct rpc_task *task;
176e21ee 915 struct rpc_rqst *rovr;
a2648094 916 int repsize, copied;
176e21ee
CL		 917 u32 _xid;
918 __be32 *xp;
919
176e21ee
CL		 920 repsize = skb->len - sizeof(rpc_fraghdr);
921 if (repsize < 4) {
922 dprintk("RPC: impossible RPC reply size %d\n", repsize);
a2648094 923 return;
176e21ee
CL		 924 }
925
926 /* Copy the XID from the skb... */
927 xp = skb_header_pointer(skb, sizeof(rpc_fraghdr), sizeof(_xid), &_xid);
928 if (xp == NULL)
a2648094 929 return;
176e21ee
CL		 930
931 /* Look up and lock the request corresponding to the given XID */
a2648094 932 spin_lock_bh(&xprt->transport_lock);
176e21ee
CL		 933 rovr = xprt_lookup_rqst(xprt, *xp);
934 if (!rovr)
935 goto out_unlock;
936 task = rovr->rq_task;
937
938 copied = rovr->rq_private_buf.buflen;
939 if (copied > repsize)
940 copied = repsize;
941
942 if (xs_local_copy_to_xdr(&rovr->rq_private_buf, skb)) {
943 dprintk("RPC: sk_buff copy failed\n");
944 goto out_unlock;
945 }
946
947 xprt_complete_rqst(task, copied);
948
949 out_unlock:
a2648094
TM		 950 spin_unlock_bh(&xprt->transport_lock);
951}
952
953static void xs_local_data_receive(struct sock_xprt *transport)
954{
955 struct sk_buff *skb;
956 struct sock *sk;
957 int err;
958
959 mutex_lock(&transport->recv_mutex);
960 sk = transport->inet;
961 if (sk == NULL)
962 goto out;
963 for (;;) {
964 skb = skb_recv_datagram(sk, 0, 1, &err);
965 if (skb == NULL)
966 break;
967 xs_local_data_read_skb(&transport->xprt, sk, skb);
968 skb_free_datagram(sk, skb);
969 }
970out:
971 mutex_unlock(&transport->recv_mutex);
972}
973
974static void xs_local_data_receive_workfn(struct work_struct *work)
975{
976 struct sock_xprt *transport =
977 container_of(work, struct sock_xprt, recv_worker);
978 xs_local_data_receive(transport);
176e21ee
CL		 979}
980
9903cd1c 981/**
f9b2ee71
TM		 982 * xs_udp_data_read_skb - receive callback for UDP sockets
983 * @xprt: transport
984 * @sk: socket
985 * @skb: skbuff
9903cd1c 986 *
a246b010 987 */
f9b2ee71
TM		 988static void xs_udp_data_read_skb(struct rpc_xprt *xprt,
989 struct sock *sk,
990 struct sk_buff *skb)
a246b010 991{
9903cd1c 992 struct rpc_task *task;
a246b010 993 struct rpc_rqst *rovr;
f9b2ee71 994 int repsize, copied;
d8ed029d
AD		 995 u32 _xid;
996 __be32 *xp;
a246b010 997
a246b010
CL		 998 repsize = skb->len - sizeof(struct udphdr);
999 if (repsize < 4) {
46121cf7 1000 dprintk("RPC: impossible RPC reply size %d!\n", repsize);
f9b2ee71 1001 return;
a246b010
CL		 1002 }
1003
1004 /* Copy the XID from the skb... */
1005 xp = skb_header_pointer(skb, sizeof(struct udphdr),
1006 sizeof(_xid), &_xid);
1007 if (xp == NULL)
f9b2ee71 1008 return;
a246b010
CL		 1009
1010 /* Look up and lock the request corresponding to the given XID */
f9b2ee71 1011 spin_lock_bh(&xprt->transport_lock);
a246b010
CL		 1012 rovr = xprt_lookup_rqst(xprt, *xp);
1013 if (!rovr)
1014 goto out_unlock;
1015 task = rovr->rq_task;
1016
a246b010
CL		 1017 if ((copied = rovr->rq_private_buf.buflen) > repsize)
1018 copied = repsize;
1019
1020 /* Suck it into the iovec, verify checksum if not done by hw. */
1781f7f5
HX		 1021 if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
1022 UDPX_INC_STATS_BH(sk, UDP_MIB_INERRORS);
a246b010 1023 goto out_unlock;
1781f7f5
HX		 1024 }
1025
1026 UDPX_INC_STATS_BH(sk, UDP_MIB_INDATAGRAMS);
a246b010 1027
6a24dfb6 1028 xprt_adjust_cwnd(xprt, task, copied);
1570c1e4 1029 xprt_complete_rqst(task, copied);
a246b010
CL		 1030
1031 out_unlock:
f9b2ee71
TM		 1032 spin_unlock_bh(&xprt->transport_lock);
1033}
1034
1035static void xs_udp_data_receive(struct sock_xprt *transport)
1036{
1037 struct sk_buff *skb;
1038 struct sock *sk;
1039 int err;
1040
1041 mutex_lock(&transport->recv_mutex);
1042 sk = transport->inet;
1043 if (sk == NULL)
1044 goto out;
1045 for (;;) {
1046 skb = skb_recv_datagram(sk, 0, 1, &err);
1047 if (skb == NULL)
1048 break;
1049 xs_udp_data_read_skb(&transport->xprt, sk, skb);
1050 skb_free_datagram(sk, skb);
1051 }
1052out:
1053 mutex_unlock(&transport->recv_mutex);
1054}
1055
1056static void xs_udp_data_receive_workfn(struct work_struct *work)
1057{
1058 struct sock_xprt *transport =
1059 container_of(work, struct sock_xprt, recv_worker);
1060 xs_udp_data_receive(transport);
1061}
1062
1063/**
1064 * xs_data_ready - "data ready" callback for UDP sockets
1065 * @sk: socket with data to read
1066 *
1067 */
1068static void xs_data_ready(struct sock *sk)
1069{
1070 struct rpc_xprt *xprt;
1071
1072 read_lock_bh(&sk->sk_callback_lock);
1073 dprintk("RPC: xs_data_ready...\n");
1074 xprt = xprt_from_sock(sk);
1075 if (xprt != NULL) {
1076 struct sock_xprt *transport = container_of(xprt,
1077 struct sock_xprt, xprt);
1078 queue_work(rpciod_workqueue, &transport->recv_worker);
1079 }
f064af1e 1080 read_unlock_bh(&sk->sk_callback_lock);
a246b010
CL		 1081}
1082
a519fc7a
TM		 1083/*
1084 * Helper function to force a TCP close if the server is sending
1085 * junk and/or it has put us in CLOSE_WAIT
1086 */
1087static void xs_tcp_force_close(struct rpc_xprt *xprt)
1088{
a519fc7a
TM		 1089 xprt_force_disconnect(xprt);
1090}
1091
dd456471 1092static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
a246b010 1093{
51971139 1094 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
a246b010
CL		 1095 size_t len, used;
1096 char *p;
1097
51971139
CL		 1098 p = ((char *) &transport->tcp_fraghdr) + transport->tcp_offset;
1099 len = sizeof(transport->tcp_fraghdr) - transport->tcp_offset;
9d292316 1100 used = xdr_skb_read_bits(desc, p, len);
51971139 1101 transport->tcp_offset += used;
a246b010
CL		 1102 if (used != len)
1103 return;
808012fb 1104
51971139
CL		 1105 transport->tcp_reclen = ntohl(transport->tcp_fraghdr);
1106 if (transport->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
e136d092 1107 transport->tcp_flags |= TCP_RCV_LAST_FRAG;
a246b010 1108 else
e136d092 1109 transport->tcp_flags &= ~TCP_RCV_LAST_FRAG;
51971139 1110 transport->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
808012fb 1111
e136d092 1112 transport->tcp_flags &= ~TCP_RCV_COPY_FRAGHDR;
51971139 1113 transport->tcp_offset = 0;
808012fb 1114
a246b010 1115 /* Sanity check of the record length */
18dca02a 1116 if (unlikely(transport->tcp_reclen < 8)) {
46121cf7 1117 dprintk("RPC: invalid TCP record fragment length\n");
a519fc7a 1118 xs_tcp_force_close(xprt);
9903cd1c 1119 return;
a246b010 1120 }
46121cf7 1121 dprintk("RPC: reading TCP record fragment of length %d\n",
51971139 1122 transport->tcp_reclen);
a246b010
CL		 1123}
1124
51971139 1125static void xs_tcp_check_fraghdr(struct sock_xprt *transport)
a246b010 1126{
51971139 1127 if (transport->tcp_offset == transport->tcp_reclen) {
e136d092 1128 transport->tcp_flags |= TCP_RCV_COPY_FRAGHDR;
51971139 1129 transport->tcp_offset = 0;
e136d092
CL		 1130 if (transport->tcp_flags & TCP_RCV_LAST_FRAG) {
1131 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1132 transport->tcp_flags |= TCP_RCV_COPY_XID;
51971139 1133 transport->tcp_copied = 0;
a246b010
CL		 1134 }
1135 }
1136}
1137
dd456471 1138static inline void xs_tcp_read_xid(struct sock_xprt *transport, struct xdr_skb_reader *desc)
a246b010
CL		 1139{
1140 size_t len, used;
1141 char *p;
1142
51971139 1143 len = sizeof(transport->tcp_xid) - transport->tcp_offset;
46121cf7 1144 dprintk("RPC: reading XID (%Zu bytes)\n", len);
51971139 1145 p = ((char *) &transport->tcp_xid) + transport->tcp_offset;
9d292316 1146 used = xdr_skb_read_bits(desc, p, len);
51971139 1147 transport->tcp_offset += used;
a246b010
CL		 1148 if (used != len)
1149 return;
e136d092 1150 transport->tcp_flags &= ~TCP_RCV_COPY_XID;
f4a2e418 1151 transport->tcp_flags |= TCP_RCV_READ_CALLDIR;
51971139 1152 transport->tcp_copied = 4;
18dca02a
RL		 1153 dprintk("RPC: reading %s XID %08x\n",
1154 (transport->tcp_flags & TCP_RPC_REPLY) ? "reply for"
1155 : "request with",
51971139
CL		 1156 ntohl(transport->tcp_xid));
1157 xs_tcp_check_fraghdr(transport);
a246b010
CL		 1158}
1159
18dca02a
RL		 1160static inline void xs_tcp_read_calldir(struct sock_xprt *transport,
1161 struct xdr_skb_reader *desc)
a246b010 1162{
18dca02a
RL		 1163 size_t len, used;
1164 u32 offset;
b76ce561 1165 char *p;
18dca02a
RL		 1166
1167 /*
1168 * We want transport->tcp_offset to be 8 at the end of this routine
1169 * (4 bytes for the xid and 4 bytes for the call/reply flag).
1170 * When this function is called for the first time,
1171 * transport->tcp_offset is 4 (after having already read the xid).
1172 */
1173 offset = transport->tcp_offset - sizeof(transport->tcp_xid);
b76ce561 1174 len = sizeof(transport->tcp_calldir) - offset;
18dca02a 1175 dprintk("RPC: reading CALL/REPLY flag (%Zu bytes)\n", len);
b76ce561
TM		 1176 p = ((char *) &transport->tcp_calldir) + offset;
1177 used = xdr_skb_read_bits(desc, p, len);
18dca02a
RL		 1178 transport->tcp_offset += used;
1179 if (used != len)
1180 return;
f4a2e418 1181 transport->tcp_flags &= ~TCP_RCV_READ_CALLDIR;
f4a2e418
RL		 1182 /*
1183 * We don't yet have the XDR buffer, so we will write the calldir
1184 * out after we get the buffer from the 'struct rpc_rqst'
1185 */
b76ce561
TM		 1186 switch (ntohl(transport->tcp_calldir)) {
1187 case RPC_REPLY:
1188 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1189 transport->tcp_flags |= TCP_RCV_COPY_DATA;
18dca02a 1190 transport->tcp_flags |= TCP_RPC_REPLY;
b76ce561
TM		 1191 break;
1192 case RPC_CALL:
1193 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1194 transport->tcp_flags |= TCP_RCV_COPY_DATA;
18dca02a 1195 transport->tcp_flags &= ~TCP_RPC_REPLY;
b76ce561
TM		 1196 break;
1197 default:
1198 dprintk("RPC: invalid request message type\n");
a519fc7a 1199 xs_tcp_force_close(&transport->xprt);
b76ce561 1200 }
18dca02a
RL		 1201 xs_tcp_check_fraghdr(transport);
1202}
1203
44b98efd
RL		 1204static inline void xs_tcp_read_common(struct rpc_xprt *xprt,
1205 struct xdr_skb_reader *desc,
1206 struct rpc_rqst *req)
a246b010 1207{
44b98efd
RL		 1208 struct sock_xprt *transport =
1209 container_of(xprt, struct sock_xprt, xprt);
a246b010
CL		 1210 struct xdr_buf *rcvbuf;
1211 size_t len;
1212 ssize_t r;
1213
a246b010 1214 rcvbuf = &req->rq_private_buf;
f4a2e418
RL		 1215
1216 if (transport->tcp_flags & TCP_RCV_COPY_CALLDIR) {
1217 /*
1218 * Save the RPC direction in the XDR buffer
1219 */
f4a2e418 1220 memcpy(rcvbuf->head[0].iov_base + transport->tcp_copied,
b76ce561
TM		 1221 &transport->tcp_calldir,
1222 sizeof(transport->tcp_calldir));
1223 transport->tcp_copied += sizeof(transport->tcp_calldir);
f4a2e418 1224 transport->tcp_flags &= ~TCP_RCV_COPY_CALLDIR;
a246b010
CL		 1225 }
1226
a246b010 1227 len = desc->count;
51971139 1228 if (len > transport->tcp_reclen - transport->tcp_offset) {
dd456471 1229 struct xdr_skb_reader my_desc;
a246b010 1230
51971139 1231 len = transport->tcp_reclen - transport->tcp_offset;
a246b010
CL		 1232 memcpy(&my_desc, desc, sizeof(my_desc));
1233 my_desc.count = len;
51971139 1234 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
9d292316 1235 &my_desc, xdr_skb_read_bits);
a246b010
CL		 1236 desc->count -= r;
1237 desc->offset += r;
1238 } else
51971139 1239 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
9d292316 1240 desc, xdr_skb_read_bits);
a246b010
CL		 1241
1242 if (r > 0) {
51971139
CL		 1243 transport->tcp_copied += r;
1244 transport->tcp_offset += r;
a246b010
CL		 1245 }
1246 if (r != len) {
1247 /* Error when copying to the receive buffer,
1248 * usually because we weren't able to allocate
1249 * additional buffer pages. All we can do now
e136d092 1250 * is turn off TCP_RCV_COPY_DATA, so the request
a246b010
CL		 1251 * will not receive any additional updates,
1252 * and time out.
1253 * Any remaining data from this record will
1254 * be discarded.
1255 */
e136d092 1256 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
46121cf7 1257 dprintk("RPC: XID %08x truncated request\n",
51971139 1258 ntohl(transport->tcp_xid));
46121cf7
CL		 1259 dprintk("RPC: xprt = %p, tcp_copied = %lu, "
1260 "tcp_offset = %u, tcp_reclen = %u\n",
1261 xprt, transport->tcp_copied,
1262 transport->tcp_offset, transport->tcp_reclen);
44b98efd 1263 return;
a246b010
CL		 1264 }
1265
46121cf7 1266 dprintk("RPC: XID %08x read %Zd bytes\n",
51971139 1267 ntohl(transport->tcp_xid), r);
46121cf7
CL		 1268 dprintk("RPC: xprt = %p, tcp_copied = %lu, tcp_offset = %u, "
1269 "tcp_reclen = %u\n", xprt, transport->tcp_copied,
1270 transport->tcp_offset, transport->tcp_reclen);
51971139
CL		 1271
1272 if (transport->tcp_copied == req->rq_private_buf.buflen)
e136d092 1273 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
51971139 1274 else if (transport->tcp_offset == transport->tcp_reclen) {
e136d092
CL		 1275 if (transport->tcp_flags & TCP_RCV_LAST_FRAG)
1276 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
a246b010 1277 }
44b98efd
RL		 1278}
1279
1280/*
1281 * Finds the request corresponding to the RPC xid and invokes the common
1282 * tcp read code to read the data.
1283 */
1284static inline int xs_tcp_read_reply(struct rpc_xprt *xprt,
1285 struct xdr_skb_reader *desc)
1286{
1287 struct sock_xprt *transport =
1288 container_of(xprt, struct sock_xprt, xprt);
1289 struct rpc_rqst *req;
1290
1291 dprintk("RPC: read reply XID %08x\n", ntohl(transport->tcp_xid));
1292
1293 /* Find and lock the request corresponding to this xid */
edc1b01c 1294 spin_lock_bh(&xprt->transport_lock);
44b98efd
RL		 1295 req = xprt_lookup_rqst(xprt, transport->tcp_xid);
1296 if (!req) {
1297 dprintk("RPC: XID %08x request not found!\n",
1298 ntohl(transport->tcp_xid));
edc1b01c 1299 spin_unlock_bh(&xprt->transport_lock);
44b98efd
RL		 1300 return -1;
1301 }
1302
1303 xs_tcp_read_common(xprt, desc, req);
1304
e136d092 1305 if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
51971139 1306 xprt_complete_rqst(req->rq_task, transport->tcp_copied);
44b98efd 1307
edc1b01c 1308 spin_unlock_bh(&xprt->transport_lock);
44b98efd
RL		 1309 return 0;
1310}
1311
9e00abc3 1312#if defined(CONFIG_SUNRPC_BACKCHANNEL)
44b98efd
RL		 1313/*
1314 * Obtains an rpc_rqst previously allocated and invokes the common
1315 * tcp read code to read the data. The result is placed in the callback
1316 * queue.
1317 * If we're unable to obtain the rpc_rqst we schedule the closing of the
1318 * connection and return -1.
1319 */
2ea24497 1320static int xs_tcp_read_callback(struct rpc_xprt *xprt,
44b98efd
RL		 1321 struct xdr_skb_reader *desc)
1322{
1323 struct sock_xprt *transport =
1324 container_of(xprt, struct sock_xprt, xprt);
1325 struct rpc_rqst *req;
1326
2ea24497 1327 /* Look up and lock the request corresponding to the given XID */
edc1b01c 1328 spin_lock_bh(&xprt->transport_lock);
2ea24497 1329 req = xprt_lookup_bc_request(xprt, transport->tcp_xid);
44b98efd 1330 if (req == NULL) {
edc1b01c 1331 spin_unlock_bh(&xprt->transport_lock);
44b98efd
RL		 1332 printk(KERN_WARNING "Callback slot table overflowed\n");
1333 xprt_force_disconnect(xprt);
1334 return -1;
1335 }
1336
44b98efd
RL		 1337 dprintk("RPC: read callback XID %08x\n", ntohl(req->rq_xid));
1338 xs_tcp_read_common(xprt, desc, req);
1339
2ea24497
TM		 1340 if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1341 xprt_complete_bc_request(req, transport->tcp_copied);
edc1b01c 1342 spin_unlock_bh(&xprt->transport_lock);
44b98efd
RL		 1343
1344 return 0;
1345}
1346
1347static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1348 struct xdr_skb_reader *desc)
1349{
1350 struct sock_xprt *transport =
1351 container_of(xprt, struct sock_xprt, xprt);
1352
1353 return (transport->tcp_flags & TCP_RPC_REPLY) ?
1354 xs_tcp_read_reply(xprt, desc) :
1355 xs_tcp_read_callback(xprt, desc);
1356}
76566773
CL		 1357
1358static int xs_tcp_bc_up(struct svc_serv *serv, struct net *net)
1359{
1360 int ret;
1361
1362 ret = svc_create_xprt(serv, "tcp-bc", net, PF_INET, 0,
1363 SVC_SOCK_ANONYMOUS);
1364 if (ret < 0)
1365 return ret;
1366 return 0;
1367}
44b98efd
RL		 1368#else
1369static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1370 struct xdr_skb_reader *desc)
1371{
1372 return xs_tcp_read_reply(xprt, desc);
1373}
9e00abc3 1374#endif /* CONFIG_SUNRPC_BACKCHANNEL */
44b98efd
RL		 1375
1376/*
1377 * Read data off the transport. This can be either an RPC_CALL or an
1378 * RPC_REPLY. Relay the processing to helper functions.
1379 */
1380static void xs_tcp_read_data(struct rpc_xprt *xprt,
1381 struct xdr_skb_reader *desc)
1382{
1383 struct sock_xprt *transport =
1384 container_of(xprt, struct sock_xprt, xprt);
1385
1386 if (_xs_tcp_read_data(xprt, desc) == 0)
1387 xs_tcp_check_fraghdr(transport);
1388 else {
1389 /*
1390 * The transport_lock protects the request handling.
1391 * There's no need to hold it to update the tcp_flags.
1392 */
1393 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1394 }
a246b010
CL		 1395}
1396
dd456471 1397static inline void xs_tcp_read_discard(struct sock_xprt *transport, struct xdr_skb_reader *desc)
a246b010
CL		 1398{
1399 size_t len;
1400
51971139 1401 len = transport->tcp_reclen - transport->tcp_offset;
a246b010
CL		 1402 if (len > desc->count)
1403 len = desc->count;
1404 desc->count -= len;
1405 desc->offset += len;
51971139 1406 transport->tcp_offset += len;
46121cf7 1407 dprintk("RPC: discarded %Zu bytes\n", len);
51971139 1408 xs_tcp_check_fraghdr(transport);
a246b010
CL		 1409}
1410
9903cd1c 1411static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
a246b010
CL		 1412{
1413 struct rpc_xprt *xprt = rd_desc->arg.data;
51971139 1414 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
dd456471 1415 struct xdr_skb_reader desc = {
a246b010
CL		 1416 .skb = skb,
1417 .offset = offset,
1418 .count = len,
9903cd1c 1419 };
a246b010 1420
46121cf7 1421 dprintk("RPC: xs_tcp_data_recv started\n");
a246b010 1422 do {
1a867a08 1423 trace_xs_tcp_data_recv(transport);
a246b010
CL		 1424 /* Read in a new fragment marker if necessary */
1425 /* Can we ever really expect to get completely empty fragments? */
e136d092 1426 if (transport->tcp_flags & TCP_RCV_COPY_FRAGHDR) {
9903cd1c 1427 xs_tcp_read_fraghdr(xprt, &desc);
a246b010
CL		 1428 continue;
1429 }
1430 /* Read in the xid if necessary */
e136d092 1431 if (transport->tcp_flags & TCP_RCV_COPY_XID) {
51971139 1432 xs_tcp_read_xid(transport, &desc);
a246b010
CL		 1433 continue;
1434 }
18dca02a 1435 /* Read in the call/reply flag */
f4a2e418 1436 if (transport->tcp_flags & TCP_RCV_READ_CALLDIR) {
18dca02a
RL		 1437 xs_tcp_read_calldir(transport, &desc);
1438 continue;
1439 }
a246b010 1440 /* Read in the request data */
e136d092 1441 if (transport->tcp_flags & TCP_RCV_COPY_DATA) {
44b98efd 1442 xs_tcp_read_data(xprt, &desc);
a246b010
CL		 1443 continue;
1444 }
1445 /* Skip over any trailing bytes on short reads */
51971139 1446 xs_tcp_read_discard(transport, &desc);
a246b010 1447 } while (desc.count);
1a867a08 1448 trace_xs_tcp_data_recv(transport);
46121cf7 1449 dprintk("RPC: xs_tcp_data_recv done\n");
a246b010
CL		 1450 return len - desc.count;
1451}
1452
66d7a56a
TM		 1453static void xs_tcp_data_receive(struct sock_xprt *transport)
1454{
1455 struct rpc_xprt *xprt = &transport->xprt;
1456 struct sock *sk;
1457 read_descriptor_t rd_desc = {
1458 .count = 2*1024*1024,
1459 .arg.data = xprt,
1460 };
1461 unsigned long total = 0;
1462 int read = 0;
1463
edc1b01c 1464 mutex_lock(&transport->recv_mutex);
66d7a56a 1465 sk = transport->inet;
edc1b01c
TM		 1466 if (sk == NULL)
1467 goto out;
66d7a56a
TM		 1468
1469 /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
1470 for (;;) {
edc1b01c 1471 lock_sock(sk);
66d7a56a 1472 read = tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
edc1b01c 1473 release_sock(sk);
66d7a56a
TM		 1474 if (read <= 0)
1475 break;
1476 total += read;
1477 rd_desc.count = 65536;
1478 }
edc1b01c
TM		 1479out:
1480 mutex_unlock(&transport->recv_mutex);
66d7a56a
TM		 1481 trace_xs_tcp_data_ready(xprt, read, total);
1482}
1483
edc1b01c
TM		 1484static void xs_tcp_data_receive_workfn(struct work_struct *work)
1485{
1486 struct sock_xprt *transport =
1487 container_of(work, struct sock_xprt, recv_worker);
1488 xs_tcp_data_receive(transport);
1489}
1490
9903cd1c
CL		 1491/**
1492 * xs_tcp_data_ready - "data ready" callback for TCP sockets
1493 * @sk: socket with data to read
9903cd1c
CL		 1494 *
1495 */
676d2369 1496static void xs_tcp_data_ready(struct sock *sk)
a246b010 1497{
66d7a56a 1498 struct sock_xprt *transport;
a246b010 1499 struct rpc_xprt *xprt;
a246b010 1500
46121cf7
CL		 1501 dprintk("RPC: xs_tcp_data_ready...\n");
1502
f064af1e 1503 read_lock_bh(&sk->sk_callback_lock);
66d7a56a 1504 if (!(xprt = xprt_from_sock(sk)))
a246b010 1505 goto out;
66d7a56a
TM		 1506 transport = container_of(xprt, struct sock_xprt, xprt);
1507
61d0a8e6
NB		 1508 /* Any data means we had a useful conversation, so
1509 * the we don't need to delay the next reconnect
1510 */
1511 if (xprt->reestablish_timeout)
1512 xprt->reestablish_timeout = 0;
edc1b01c 1513 queue_work(rpciod_workqueue, &transport->recv_worker);
61d0a8e6 1514
a246b010 1515out:
f064af1e 1516 read_unlock_bh(&sk->sk_callback_lock);
a246b010
CL		 1517}
1518
9903cd1c
CL		 1519/**
1520 * xs_tcp_state_change - callback to handle TCP socket state changes
1521 * @sk: socket whose state has changed
1522 *
1523 */
1524static void xs_tcp_state_change(struct sock *sk)
a246b010 1525{
9903cd1c 1526 struct rpc_xprt *xprt;
0fdea1e8 1527 struct sock_xprt *transport;
a246b010 1528
f064af1e 1529 read_lock_bh(&sk->sk_callback_lock);
a246b010
CL		 1530 if (!(xprt = xprt_from_sock(sk)))
1531 goto out;
46121cf7 1532 dprintk("RPC: xs_tcp_state_change client %p...\n", xprt);
669502ff 1533 dprintk("RPC: state %x conn %d dead %d zapped %d sk_shutdown %d\n",
46121cf7
CL		 1534 sk->sk_state, xprt_connected(xprt),
1535 sock_flag(sk, SOCK_DEAD),
669502ff
AC		 1536 sock_flag(sk, SOCK_ZAPPED),
1537 sk->sk_shutdown);
a246b010 1538
0fdea1e8 1539 transport = container_of(xprt, struct sock_xprt, xprt);
40b5ea0c 1540 trace_rpc_socket_state_change(xprt, sk->sk_socket);
a246b010
CL		 1541 switch (sk->sk_state) {
1542 case TCP_ESTABLISHED:
f064af1e 1543 spin_lock(&xprt->transport_lock);
a246b010 1544 if (!xprt_test_and_set_connected(xprt)) {
51971139 1545
a246b010 1546 /* Reset TCP record info */
51971139
CL		 1547 transport->tcp_offset = 0;
1548 transport->tcp_reclen = 0;
1549 transport->tcp_copied = 0;
e136d092
CL		 1550 transport->tcp_flags =
1551 TCP_RCV_COPY_FRAGHDR | TCP_RCV_COPY_XID;
8b71798c 1552 xprt->connect_cookie++;
0fdea1e8
TM		 1553 clear_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
1554 xprt_clear_connecting(xprt);
51971139 1555
2a491991 1556 xprt_wake_pending_tasks(xprt, -EAGAIN);
a246b010 1557 }
f064af1e 1558 spin_unlock(&xprt->transport_lock);
a246b010 1559 break;
3b948ae5
TM		 1560 case TCP_FIN_WAIT1:
1561 /* The client initiated a shutdown of the socket */
7c1d71cf 1562 xprt->connect_cookie++;
663b8858 1563 xprt->reestablish_timeout = 0;
3b948ae5 1564 set_bit(XPRT_CLOSING, &xprt->state);
4e857c58 1565 smp_mb__before_atomic();
3b948ae5 1566 clear_bit(XPRT_CONNECTED, &xprt->state);
ef803670 1567 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
4e857c58 1568 smp_mb__after_atomic();
a246b010 1569 break;
632e3bdc 1570 case TCP_CLOSE_WAIT:
3b948ae5 1571 /* The server initiated a shutdown of the socket */
7c1d71cf 1572 xprt->connect_cookie++;
d0bea455 1573 clear_bit(XPRT_CONNECTED, &xprt->state);
a519fc7a 1574 xs_tcp_force_close(xprt);
663b8858
TM		 1575 case TCP_CLOSING:
1576 /*
1577 * If the server closed down the connection, make sure that
1578 * we back off before reconnecting
1579 */
1580 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1581 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
3b948ae5
TM		 1582 break;
1583 case TCP_LAST_ACK:
670f9457 1584 set_bit(XPRT_CLOSING, &xprt->state);
4e857c58 1585 smp_mb__before_atomic();
3b948ae5 1586 clear_bit(XPRT_CONNECTED, &xprt->state);
4e857c58 1587 smp_mb__after_atomic();
3b948ae5
TM		 1588 break;
1589 case TCP_CLOSE:
0fdea1e8
TM		 1590 if (test_and_clear_bit(XPRT_SOCK_CONNECTING,
1591 &transport->sock_state))
1592 xprt_clear_connecting(xprt);
7d1e8255 1593 xs_sock_mark_closed(xprt);
a246b010
CL		 1594 }
1595 out:
f064af1e 1596 read_unlock_bh(&sk->sk_callback_lock);
a246b010
CL		 1597}
1598
1f0fa154
IJ		 1599static void xs_write_space(struct sock *sk)
1600{
13331a55 1601 struct socket_wq *wq;
1f0fa154
IJ		 1602 struct rpc_xprt *xprt;
1603
13331a55 1604 if (!sk->sk_socket)
1f0fa154 1605 return;
13331a55 1606 clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1f0fa154
IJ		 1607
1608 if (unlikely(!(xprt = xprt_from_sock(sk))))
1609 return;
13331a55
TM		 1610 rcu_read_lock();
1611 wq = rcu_dereference(sk->sk_wq);
1612 if (!wq || test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &wq->flags) == 0)
1613 goto out;
1f0fa154
IJ		 1614
1615 xprt_write_space(xprt);
13331a55
TM		 1616out:
1617 rcu_read_unlock();
1f0fa154
IJ		 1618}
1619
9903cd1c 1620/**
c7b2cae8
CL		 1621 * xs_udp_write_space - callback invoked when socket buffer space
1622 * becomes available
9903cd1c
CL		 1623 * @sk: socket whose state has changed
1624 *
a246b010
CL		 1625 * Called when more output buffer space is available for this socket.
1626 * We try not to wake our writers until they can make "significant"
c7b2cae8 1627 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
a246b010
CL		 1628 * with a bunch of small requests.
1629 */
c7b2cae8 1630static void xs_udp_write_space(struct sock *sk)
a246b010 1631{
f064af1e 1632 read_lock_bh(&sk->sk_callback_lock);
a246b010 1633
c7b2cae8 1634 /* from net/core/sock.c:sock_def_write_space */
1f0fa154
IJ		 1635 if (sock_writeable(sk))
1636 xs_write_space(sk);
a246b010 1637
f064af1e 1638 read_unlock_bh(&sk->sk_callback_lock);
c7b2cae8 1639}
a246b010 1640
c7b2cae8
CL		 1641/**
1642 * xs_tcp_write_space - callback invoked when socket buffer space
1643 * becomes available
1644 * @sk: socket whose state has changed
1645 *
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"
1648 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1649 * with a bunch of small requests.
1650 */
1651static void xs_tcp_write_space(struct sock *sk)
1652{
f064af1e 1653 read_lock_bh(&sk->sk_callback_lock);
c7b2cae8
CL		 1654
1655 /* from net/core/stream.c:sk_stream_write_space */
64dc6130 1656 if (sk_stream_is_writeable(sk))
1f0fa154 1657 xs_write_space(sk);
b6ddf64f 1658
f064af1e 1659 read_unlock_bh(&sk->sk_callback_lock);
a246b010
CL		 1660}
1661
470056c2 1662static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
a246b010 1663{
ee0ac0c2
CL		 1664 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1665 struct sock *sk = transport->inet;
a246b010 1666
7c6e066e 1667 if (transport->rcvsize) {
a246b010 1668 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
7c6e066e 1669 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
a246b010 1670 }
7c6e066e 1671 if (transport->sndsize) {
a246b010 1672 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
7c6e066e 1673 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
a246b010
CL		 1674 sk->sk_write_space(sk);
1675 }
1676}
1677
43118c29 1678/**
470056c2 1679 * xs_udp_set_buffer_size - set send and receive limits
43118c29 1680 * @xprt: generic transport
470056c2
CL		 1681 * @sndsize: requested size of send buffer, in bytes
1682 * @rcvsize: requested size of receive buffer, in bytes
43118c29 1683 *
470056c2 1684 * Set socket send and receive buffer size limits.
43118c29 1685 */
470056c2 1686static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
43118c29 1687{
7c6e066e
CL		 1688 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1689
1690 transport->sndsize = 0;
470056c2 1691 if (sndsize)
7c6e066e
CL		 1692 transport->sndsize = sndsize + 1024;
1693 transport->rcvsize = 0;
470056c2 1694 if (rcvsize)
7c6e066e 1695 transport->rcvsize = rcvsize + 1024;
470056c2
CL		 1696
1697 xs_udp_do_set_buffer_size(xprt);
43118c29
CL		 1698}
1699
46c0ee8b
CL		 1700/**
1701 * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1702 * @task: task that timed out
1703 *
1704 * Adjust the congestion window after a retransmit timeout has occurred.
1705 */
6a24dfb6 1706static void xs_udp_timer(struct rpc_xprt *xprt, struct rpc_task *task)
46c0ee8b 1707{
6a24dfb6 1708 xprt_adjust_cwnd(xprt, task, -ETIMEDOUT);
46c0ee8b
CL		 1709}
1710
b85d8806
CL		 1711static unsigned short xs_get_random_port(void)
1712{
1713 unsigned short range = xprt_max_resvport - xprt_min_resvport;
63862b5b 1714 unsigned short rand = (unsigned short) prandom_u32() % range;
b85d8806
CL		 1715 return rand + xprt_min_resvport;
1716}
1717
4dda9c8a
TM		 1718/**
1719 * xs_set_reuseaddr_port - set the socket's port and address reuse options
1720 * @sock: socket
1721 *
1722 * Note that this function has to be called on all sockets that share the
1723 * same port, and it must be called before binding.
1724 */
1725static void xs_sock_set_reuseport(struct socket *sock)
1726{
402e23b4 1727 int opt = 1;
4dda9c8a 1728
402e23b4
TM		 1729 kernel_setsockopt(sock, SOL_SOCKET, SO_REUSEPORT,
1730 (char *)&opt, sizeof(opt));
4dda9c8a
TM		 1731}
1732
1733static unsigned short xs_sock_getport(struct socket *sock)
1734{
1735 struct sockaddr_storage buf;
1736 int buflen;
1737 unsigned short port = 0;
1738
1739 if (kernel_getsockname(sock, (struct sockaddr *)&buf, &buflen) < 0)
1740 goto out;
1741 switch (buf.ss_family) {
1742 case AF_INET6:
1743 port = ntohs(((struct sockaddr_in6 *)&buf)->sin6_port);
1744 break;
1745 case AF_INET:
1746 port = ntohs(((struct sockaddr_in *)&buf)->sin_port);
1747 }
1748out:
1749 return port;
1750}
1751
92200412
CL		 1752/**
1753 * xs_set_port - reset the port number in the remote endpoint address
1754 * @xprt: generic transport
1755 * @port: new port number
1756 *
1757 */
1758static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1759{
46121cf7 1760 dprintk("RPC: setting port for xprt %p to %u\n", xprt, port);
c4efcb1d 1761
9dc3b095
CL		 1762 rpc_set_port(xs_addr(xprt), port);
1763 xs_update_peer_port(xprt);
92200412
CL		 1764}
1765
4dda9c8a
TM		 1766static void xs_set_srcport(struct sock_xprt *transport, struct socket *sock)
1767{
1768 if (transport->srcport == 0)
1769 transport->srcport = xs_sock_getport(sock);
1770}
1771
5d4ec932 1772static unsigned short xs_get_srcport(struct sock_xprt *transport)
67a391d7 1773{
fbfffbd5 1774 unsigned short port = transport->srcport;
67a391d7
TM		 1775
1776 if (port == 0 && transport->xprt.resvport)
1777 port = xs_get_random_port();
1778 return port;
1779}
1780
baaf4e48 1781static unsigned short xs_next_srcport(struct sock_xprt *transport, unsigned short port)
67a391d7 1782{
fbfffbd5
CL		 1783 if (transport->srcport != 0)
1784 transport->srcport = 0;
67a391d7
TM		 1785 if (!transport->xprt.resvport)
1786 return 0;
1787 if (port <= xprt_min_resvport || port > xprt_max_resvport)
1788 return xprt_max_resvport;
1789 return --port;
1790}
beb59b68 1791static int xs_bind(struct sock_xprt *transport, struct socket *sock)
a246b010 1792{
beb59b68 1793 struct sockaddr_storage myaddr;
67a391d7 1794 int err, nloop = 0;
5d4ec932 1795 unsigned short port = xs_get_srcport(transport);
67a391d7 1796 unsigned short last;
a246b010 1797
0f7a622c
CP		 1798 /*
1799 * If we are asking for any ephemeral port (i.e. port == 0 &&
1800 * transport->xprt.resvport == 0), don't bind. Let the local
1801 * port selection happen implicitly when the socket is used
1802 * (for example at connect time).
1803 *
1804 * This ensures that we can continue to establish TCP
1805 * connections even when all local ephemeral ports are already
1806 * a part of some TCP connection. This makes no difference
1807 * for UDP sockets, but also doens't harm them.
1808 *
1809 * If we're asking for any reserved port (i.e. port == 0 &&
1810 * transport->xprt.resvport == 1) xs_get_srcport above will
1811 * ensure that port is non-zero and we will bind as needed.
1812 */
1813 if (port == 0)
1814 return 0;
1815
beb59b68 1816 memcpy(&myaddr, &transport->srcaddr, transport->xprt.addrlen);
a246b010 1817 do {
beb59b68
PE		 1818 rpc_set_port((struct sockaddr *)&myaddr, port);
1819 err = kernel_bind(sock, (struct sockaddr *)&myaddr,
1820 transport->xprt.addrlen);
a246b010 1821 if (err == 0) {
fbfffbd5 1822 transport->srcport = port;
d3bc9a1d 1823 break;
a246b010 1824 }
67a391d7 1825 last = port;
baaf4e48 1826 port = xs_next_srcport(transport, port);
67a391d7
TM		 1827 if (port > last)
1828 nloop++;
1829 } while (err == -EADDRINUSE && nloop != 2);
90058d37 1830
4232e863 1831 if (myaddr.ss_family == AF_INET)
beb59b68
PE		 1832 dprintk("RPC: %s %pI4:%u: %s (%d)\n", __func__,
1833 &((struct sockaddr_in *)&myaddr)->sin_addr,
1834 port, err ? "failed" : "ok", err);
1835 else
1836 dprintk("RPC: %s %pI6:%u: %s (%d)\n", __func__,
1837 &((struct sockaddr_in6 *)&myaddr)->sin6_addr,
1838 port, err ? "failed" : "ok", err);
a246b010
CL		 1839 return err;
1840}
1841
176e21ee
CL		 1842/*
1843 * We don't support autobind on AF_LOCAL sockets
1844 */
1845static void xs_local_rpcbind(struct rpc_task *task)
1846{
3dc0da27
TM		 1847 rcu_read_lock();
1848 xprt_set_bound(rcu_dereference(task->tk_client->cl_xprt));
1849 rcu_read_unlock();
176e21ee
CL		 1850}
1851
1852static void xs_local_set_port(struct rpc_xprt *xprt, unsigned short port)
1853{
1854}
beb59b68 1855
ed07536e
PZ		 1856#ifdef CONFIG_DEBUG_LOCK_ALLOC
1857static struct lock_class_key xs_key[2];
1858static struct lock_class_key xs_slock_key[2];
1859
176e21ee
CL		 1860static inline void xs_reclassify_socketu(struct socket *sock)
1861{
1862 struct sock *sk = sock->sk;
1863
176e21ee
CL		 1864 sock_lock_init_class_and_name(sk, "slock-AF_LOCAL-RPC",
1865 &xs_slock_key[1], "sk_lock-AF_LOCAL-RPC", &xs_key[1]);
1866}
1867
8945ee5e 1868static inline void xs_reclassify_socket4(struct socket *sock)
ed07536e
PZ		 1869{
1870 struct sock *sk = sock->sk;
8945ee5e 1871
8945ee5e
CL		 1872 sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1873 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1874}
ed07536e 1875
8945ee5e
CL		 1876static inline void xs_reclassify_socket6(struct socket *sock)
1877{
1878 struct sock *sk = sock->sk;
ed07536e 1879
8945ee5e
CL		 1880 sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1881 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
ed07536e 1882}
6bc9638a
PE		 1883
1884static inline void xs_reclassify_socket(int family, struct socket *sock)
1885{
1b7a1819
WAA		 1886 WARN_ON_ONCE(sock_owned_by_user(sock->sk));
1887 if (sock_owned_by_user(sock->sk))
1888 return;
1889
4232e863 1890 switch (family) {
176e21ee
CL		 1891 case AF_LOCAL:
1892 xs_reclassify_socketu(sock);
1893 break;
4232e863 1894 case AF_INET:
6bc9638a 1895 xs_reclassify_socket4(sock);
4232e863
CL		 1896 break;
1897 case AF_INET6:
6bc9638a 1898 xs_reclassify_socket6(sock);
4232e863
CL		 1899 break;
1900 }
6bc9638a 1901}
ed07536e 1902#else
6bc9638a
PE		 1903static inline void xs_reclassify_socket(int family, struct socket *sock)
1904{
1905}
ed07536e
PZ		 1906#endif
1907
93dc41bd
N		 1908static void xs_dummy_setup_socket(struct work_struct *work)
1909{
1910}
1911
6bc9638a 1912static struct socket *xs_create_sock(struct rpc_xprt *xprt,
4dda9c8a
TM		 1913 struct sock_xprt *transport, int family, int type,
1914 int protocol, bool reuseport)
22f79326
PE		 1915{
1916 struct socket *sock;
1917 int err;
1918
6bc9638a 1919 err = __sock_create(xprt->xprt_net, family, type, protocol, &sock, 1);
22f79326
PE		 1920 if (err < 0) {
1921 dprintk("RPC: can't create %d transport socket (%d).\n",
1922 protocol, -err);
1923 goto out;
1924 }
6bc9638a 1925 xs_reclassify_socket(family, sock);
22f79326 1926
4dda9c8a
TM		 1927 if (reuseport)
1928 xs_sock_set_reuseport(sock);
1929
4cea288a
BH		 1930 err = xs_bind(transport, sock);
1931 if (err) {
22f79326
PE		 1932 sock_release(sock);
1933 goto out;
1934 }
1935
1936 return sock;
1937out:
1938 return ERR_PTR(err);
1939}
1940
176e21ee
CL		 1941static int xs_local_finish_connecting(struct rpc_xprt *xprt,
1942 struct socket *sock)
1943{
1944 struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1945 xprt);
1946
1947 if (!transport->inet) {
1948 struct sock *sk = sock->sk;
1949
1950 write_lock_bh(&sk->sk_callback_lock);
1951
1952 xs_save_old_callbacks(transport, sk);
1953
1954 sk->sk_user_data = xprt;
a2648094 1955 sk->sk_data_ready = xs_data_ready;
176e21ee 1956 sk->sk_write_space = xs_udp_write_space;
2118071d 1957 sk->sk_error_report = xs_error_report;
c2126157 1958 sk->sk_allocation = GFP_NOIO;
176e21ee
CL		 1959
1960 xprt_clear_connected(xprt);
1961
1962 /* Reset to new socket */
1963 transport->sock = sock;
1964 transport->inet = sk;
1965
1966 write_unlock_bh(&sk->sk_callback_lock);
1967 }
1968
1969 /* Tell the socket layer to start connecting... */
1970 xprt->stat.connect_count++;
1971 xprt->stat.connect_start = jiffies;
1972 return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, 0);
1973}
1974
1975/**
1976 * xs_local_setup_socket - create AF_LOCAL socket, connect to a local endpoint
176e21ee 1977 * @transport: socket transport to connect
176e21ee 1978 */
dc107402 1979static int xs_local_setup_socket(struct sock_xprt *transport)
176e21ee 1980{
176e21ee
CL		 1981 struct rpc_xprt *xprt = &transport->xprt;
1982 struct socket *sock;
1983 int status = -EIO;
1984
176e21ee
CL		 1985 status = __sock_create(xprt->xprt_net, AF_LOCAL,
1986 SOCK_STREAM, 0, &sock, 1);
1987 if (status < 0) {
1988 dprintk("RPC: can't create AF_LOCAL "
1989 "transport socket (%d).\n", -status);
1990 goto out;
1991 }
d1358917 1992 xs_reclassify_socket(AF_LOCAL, sock);
176e21ee
CL		 1993
1994 dprintk("RPC: worker connecting xprt %p via AF_LOCAL to %s\n",
1995 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1996
1997 status = xs_local_finish_connecting(xprt, sock);
40b5ea0c 1998 trace_rpc_socket_connect(xprt, sock, status);
176e21ee
CL		 1999 switch (status) {
2000 case 0:
2001 dprintk("RPC: xprt %p connected to %s\n",
2002 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2003 xprt_set_connected(xprt);
3601c4a9 2004 case -ENOBUFS:
176e21ee
CL		 2005 break;
2006 case -ENOENT:
2007 dprintk("RPC: xprt %p: socket %s does not exist\n",
2008 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2009 break;
4a20a988
TM		 2010 case -ECONNREFUSED:
2011 dprintk("RPC: xprt %p: connection refused for %s\n",
2012 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2013 break;
176e21ee
CL		 2014 default:
2015 printk(KERN_ERR "%s: unhandled error (%d) connecting to %s\n",
2016 __func__, -status,
2017 xprt->address_strings[RPC_DISPLAY_ADDR]);
2018 }
2019
2020out:
2021 xprt_clear_connecting(xprt);
2022 xprt_wake_pending_tasks(xprt, status);
dc107402
BF		 2023 return status;
2024}
2025
b6669737 2026static void xs_local_connect(struct rpc_xprt *xprt, struct rpc_task *task)
dc107402 2027{
dc107402
BF		 2028 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2029 int ret;
2030
2031 if (RPC_IS_ASYNC(task)) {
2032 /*
2033 * We want the AF_LOCAL connect to be resolved in the
2034 * filesystem namespace of the process making the rpc
2035 * call. Thus we connect synchronously.
2036 *
2037 * If we want to support asynchronous AF_LOCAL calls,
2038 * we'll need to figure out how to pass a namespace to
2039 * connect.
2040 */
2041 rpc_exit(task, -ENOTCONN);
2042 return;
2043 }
2044 ret = xs_local_setup_socket(transport);
2045 if (ret && !RPC_IS_SOFTCONN(task))
2046 msleep_interruptible(15000);
176e21ee
CL		 2047}
2048
3c87ef6e 2049#if IS_ENABLED(CONFIG_SUNRPC_SWAP)
d6e971d8
JL		 2050/*
2051 * Note that this should be called with XPRT_LOCKED held (or when we otherwise
2052 * know that we have exclusive access to the socket), to guard against
2053 * races with xs_reset_transport.
2054 */
a564b8f0
MG		 2055static void xs_set_memalloc(struct rpc_xprt *xprt)
2056{
2057 struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
2058 xprt);
2059
d6e971d8
JL		 2060 /*
2061 * If there's no sock, then we have nothing to set. The
2062 * reconnecting process will get it for us.
2063 */
2064 if (!transport->inet)
2065 return;
8e228133 2066 if (atomic_read(&xprt->swapper))
a564b8f0
MG		 2067 sk_set_memalloc(transport->inet);
2068}
2069
2070/**
d67fa4d8 2071 * xs_enable_swap - Tag this transport as being used for swap.
a564b8f0 2072 * @xprt: transport to tag
a564b8f0 2073 *
8e228133
JL		 2074 * Take a reference to this transport on behalf of the rpc_clnt, and
2075 * optionally mark it for swapping if it wasn't already.
a564b8f0 2076 */
d67fa4d8
JL		 2077static int
2078xs_enable_swap(struct rpc_xprt *xprt)
a564b8f0 2079{
d6e971d8 2080 struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
a564b8f0 2081
d6e971d8
JL		 2082 if (atomic_inc_return(&xprt->swapper) != 1)
2083 return 0;
2084 if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE))
2085 return -ERESTARTSYS;
2086 if (xs->inet)
2087 sk_set_memalloc(xs->inet);
2088 xprt_release_xprt(xprt, NULL);
8e228133
JL		 2089 return 0;
2090}
a564b8f0 2091
8e228133 2092/**
d67fa4d8 2093 * xs_disable_swap - Untag this transport as being used for swap.
8e228133
JL		 2094 * @xprt: transport to tag
2095 *
2096 * Drop a "swapper" reference to this xprt on behalf of the rpc_clnt. If the
2097 * swapper refcount goes to 0, untag the socket as a memalloc socket.
2098 */
d67fa4d8
JL		 2099static void
2100xs_disable_swap(struct rpc_xprt *xprt)
8e228133 2101{
d6e971d8 2102 struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
8e228133 2103
d6e971d8
JL		 2104 if (!atomic_dec_and_test(&xprt->swapper))
2105 return;
2106 if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE))
2107 return;
2108 if (xs->inet)
2109 sk_clear_memalloc(xs->inet);
2110 xprt_release_xprt(xprt, NULL);
a564b8f0 2111}
a564b8f0
MG		 2112#else
2113static void xs_set_memalloc(struct rpc_xprt *xprt)
2114{
2115}
d67fa4d8
JL		 2116
2117static int
2118xs_enable_swap(struct rpc_xprt *xprt)
2119{
2120 return -EINVAL;
2121}
2122
2123static void
2124xs_disable_swap(struct rpc_xprt *xprt)
2125{
2126}
a564b8f0
MG		 2127#endif
2128
16be2d20
CL		 2129static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2130{
2131 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2132
2133 if (!transport->inet) {
2134 struct sock *sk = sock->sk;
2135
2136 write_lock_bh(&sk->sk_callback_lock);
2137
2a9e1cfa
TM		 2138 xs_save_old_callbacks(transport, sk);
2139
16be2d20 2140 sk->sk_user_data = xprt;
f9b2ee71 2141 sk->sk_data_ready = xs_data_ready;
16be2d20 2142 sk->sk_write_space = xs_udp_write_space;
c2126157 2143 sk->sk_allocation = GFP_NOIO;
16be2d20
CL		 2144
2145 xprt_set_connected(xprt);
2146
2147 /* Reset to new socket */
2148 transport->sock = sock;
2149 transport->inet = sk;
2150
a564b8f0
MG		 2151 xs_set_memalloc(xprt);
2152
16be2d20
CL		 2153 write_unlock_bh(&sk->sk_callback_lock);
2154 }
2155 xs_udp_do_set_buffer_size(xprt);
2156}
2157
8c14ff2a 2158static void xs_udp_setup_socket(struct work_struct *work)
a246b010 2159{
34161db6
TM		 2160 struct sock_xprt *transport =
2161 container_of(work, struct sock_xprt, connect_worker.work);
c8475461 2162 struct rpc_xprt *xprt = &transport->xprt;
ee0ac0c2 2163 struct socket *sock = transport->sock;
b65c0310 2164 int status = -EIO;
9903cd1c 2165
8c14ff2a 2166 sock = xs_create_sock(xprt, transport,
4dda9c8a
TM		 2167 xs_addr(xprt)->sa_family, SOCK_DGRAM,
2168 IPPROTO_UDP, false);
b65c0310 2169 if (IS_ERR(sock))
b0d93ad5 2170 goto out;
68e220bd 2171
c740eff8
CL		 2172 dprintk("RPC: worker connecting xprt %p via %s to "
2173 "%s (port %s)\n", xprt,
2174 xprt->address_strings[RPC_DISPLAY_PROTO],
2175 xprt->address_strings[RPC_DISPLAY_ADDR],
2176 xprt->address_strings[RPC_DISPLAY_PORT]);
68e220bd
CL		 2177
2178 xs_udp_finish_connecting(xprt, sock);
40b5ea0c 2179 trace_rpc_socket_connect(xprt, sock, 0);
b0d93ad5
CL		 2180 status = 0;
2181out:
718ba5b8 2182 xprt_unlock_connect(xprt, transport);
b0d93ad5 2183 xprt_clear_connecting(xprt);
7d1e8255 2184 xprt_wake_pending_tasks(xprt, status);
a246b010
CL		 2185}
2186
4876cc77
TM		 2187/**
2188 * xs_tcp_shutdown - gracefully shut down a TCP socket
2189 * @xprt: transport
2190 *
2191 * Initiates a graceful shutdown of the TCP socket by calling the
2192 * equivalent of shutdown(SHUT_RDWR);
2193 */
2194static void xs_tcp_shutdown(struct rpc_xprt *xprt)
2195{
2196 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2197 struct socket *sock = transport->sock;
2198
2199 if (sock == NULL)
2200 return;
2201 if (xprt_connected(xprt)) {
2202 kernel_sock_shutdown(sock, SHUT_RDWR);
2203 trace_rpc_socket_shutdown(xprt, sock);
2204 } else
2205 xs_reset_transport(transport);
2206}
2207
16be2d20 2208static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
a246b010 2209{
16be2d20 2210 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
fe19a96b 2211 int ret = -ENOTCONN;
edb267a6 2212
ee0ac0c2 2213 if (!transport->inet) {
b0d93ad5 2214 struct sock *sk = sock->sk;
7f260e85
TM		 2215 unsigned int keepidle = xprt->timeout->to_initval / HZ;
2216 unsigned int keepcnt = xprt->timeout->to_retries + 1;
2217 unsigned int opt_on = 1;
775f06ab 2218 unsigned int timeo;
7f260e85
TM		 2219
2220 /* TCP Keepalive options */
2221 kernel_setsockopt(sock, SOL_SOCKET, SO_KEEPALIVE,
2222 (char *)&opt_on, sizeof(opt_on));
2223 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPIDLE,
2224 (char *)&keepidle, sizeof(keepidle));
2225 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPINTVL,
2226 (char *)&keepidle, sizeof(keepidle));
2227 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPCNT,
2228 (char *)&keepcnt, sizeof(keepcnt));
b0d93ad5 2229
775f06ab
TM		 2230 /* TCP user timeout (see RFC5482) */
2231 timeo = jiffies_to_msecs(xprt->timeout->to_initval) *
2232 (xprt->timeout->to_retries + 1);
2233 kernel_setsockopt(sock, SOL_TCP, TCP_USER_TIMEOUT,
2234 (char *)&timeo, sizeof(timeo));
2235
b0d93ad5
CL		 2236 write_lock_bh(&sk->sk_callback_lock);
2237
2a9e1cfa
TM		 2238 xs_save_old_callbacks(transport, sk);
2239
b0d93ad5 2240 sk->sk_user_data = xprt;
b0d93ad5
CL		 2241 sk->sk_data_ready = xs_tcp_data_ready;
2242 sk->sk_state_change = xs_tcp_state_change;
2243 sk->sk_write_space = xs_tcp_write_space;
2118071d 2244 sk->sk_error_report = xs_error_report;
c2126157 2245 sk->sk_allocation = GFP_NOIO;
3167e12c
CL		 2246
2247 /* socket options */
3167e12c 2248 sock_reset_flag(sk, SOCK_LINGER);
3167e12c 2249 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
b0d93ad5
CL		 2250
2251 xprt_clear_connected(xprt);
2252
2253 /* Reset to new socket */
ee0ac0c2
CL		 2254 transport->sock = sock;
2255 transport->inet = sk;
b0d93ad5
CL		 2256
2257 write_unlock_bh(&sk->sk_callback_lock);
2258 }
2259
01d37c42 2260 if (!xprt_bound(xprt))
fe19a96b 2261 goto out;
01d37c42 2262
a564b8f0
MG		 2263 xs_set_memalloc(xprt);
2264
b0d93ad5 2265 /* Tell the socket layer to start connecting... */
262ca07d
CL		 2266 xprt->stat.connect_count++;
2267 xprt->stat.connect_start = jiffies;
0fdea1e8 2268 set_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
fe19a96b
TM		 2269 ret = kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
2270 switch (ret) {
2271 case 0:
4dda9c8a 2272 xs_set_srcport(transport, sock);
fe19a96b
TM		 2273 case -EINPROGRESS:
2274 /* SYN_SENT! */
fe19a96b
TM		 2275 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2276 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2277 }
2278out:
2279 return ret;
16be2d20
CL		 2280}
2281
9903cd1c 2282/**
b61d59ff 2283 * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint
9903cd1c
CL		 2284 *
2285 * Invoked by a work queue tasklet.
a246b010 2286 */
cdd518d5 2287static void xs_tcp_setup_socket(struct work_struct *work)
a246b010 2288{
cdd518d5
PE		 2289 struct sock_xprt *transport =
2290 container_of(work, struct sock_xprt, connect_worker.work);
ee0ac0c2 2291 struct socket *sock = transport->sock;
a9f5f0f7 2292 struct rpc_xprt *xprt = &transport->xprt;
b61d59ff 2293 int status = -EIO;
a246b010 2294
ee0ac0c2 2295 if (!sock) {
cdd518d5 2296 sock = xs_create_sock(xprt, transport,
4dda9c8a
TM		 2297 xs_addr(xprt)->sa_family, SOCK_STREAM,
2298 IPPROTO_TCP, true);
b61d59ff
TM		 2299 if (IS_ERR(sock)) {
2300 status = PTR_ERR(sock);
3167e12c
CL		 2301 goto out;
2302 }
7d1e8255 2303 }
a246b010 2304
c740eff8
CL		 2305 dprintk("RPC: worker connecting xprt %p via %s to "
2306 "%s (port %s)\n", xprt,
2307 xprt->address_strings[RPC_DISPLAY_PROTO],
2308 xprt->address_strings[RPC_DISPLAY_ADDR],
2309 xprt->address_strings[RPC_DISPLAY_PORT]);
edb267a6 2310
16be2d20 2311 status = xs_tcp_finish_connecting(xprt, sock);
40b5ea0c 2312 trace_rpc_socket_connect(xprt, sock, status);
46121cf7
CL		 2313 dprintk("RPC: %p connect status %d connected %d sock state %d\n",
2314 xprt, -status, xprt_connected(xprt),
2315 sock->sk->sk_state);
2a491991 2316 switch (status) {
f75e6745
TM		 2317 default:
2318 printk("%s: connect returned unhandled error %d\n",
2319 __func__, status);
2320 case -EADDRNOTAVAIL:
2321 /* We're probably in TIME_WAIT. Get rid of existing socket,
2322 * and retry
2323 */
a519fc7a 2324 xs_tcp_force_close(xprt);
88b5ed73 2325 break;
2a491991
TM		 2326 case 0:
2327 case -EINPROGRESS:
2328 case -EALREADY:
718ba5b8 2329 xprt_unlock_connect(xprt, transport);
7d1e8255 2330 return;
9fcfe0c8
TM		 2331 case -EINVAL:
2332 /* Happens, for instance, if the user specified a link
2333 * local IPv6 address without a scope-id.
2334 */
3ed5e2a2
TM		 2335 case -ECONNREFUSED:
2336 case -ECONNRESET:
2337 case -ENETUNREACH:
3913c78c 2338 case -EADDRINUSE:
3601c4a9 2339 case -ENOBUFS:
3ed5e2a2 2340 /* retry with existing socket, after a delay */
4efdd92c 2341 xs_tcp_force_close(xprt);
9fcfe0c8 2342 goto out;
a246b010 2343 }
2a491991 2344 status = -EAGAIN;
a246b010 2345out:
718ba5b8 2346 xprt_unlock_connect(xprt, transport);
68e220bd 2347 xprt_clear_connecting(xprt);
7d1e8255 2348 xprt_wake_pending_tasks(xprt, status);
68e220bd 2349}
b0d93ad5 2350
9903cd1c
CL		 2351/**
2352 * xs_connect - connect a socket to a remote endpoint
1b092092 2353 * @xprt: pointer to transport structure
9903cd1c
CL		 2354 * @task: address of RPC task that manages state of connect request
2355 *
2356 * TCP: If the remote end dropped the connection, delay reconnecting.
03bf4b70
CL		 2357 *
2358 * UDP socket connects are synchronous, but we use a work queue anyway
2359 * to guarantee that even unprivileged user processes can set up a
2360 * socket on a privileged port.
2361 *
2362 * If a UDP socket connect fails, the delay behavior here prevents
2363 * retry floods (hard mounts).
9903cd1c 2364 */
1b092092 2365static void xs_connect(struct rpc_xprt *xprt, struct rpc_task *task)
a246b010 2366{
ee0ac0c2 2367 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
a246b010 2368
718ba5b8
TM		 2369 WARN_ON_ONCE(!xprt_lock_connect(xprt, task, transport));
2370
99b1a4c3 2371 if (transport->sock != NULL) {
46121cf7
CL		 2372 dprintk("RPC: xs_connect delayed xprt %p for %lu "
2373 "seconds\n",
03bf4b70 2374 xprt, xprt->reestablish_timeout / HZ);
99b1a4c3
TM		 2375
2376 /* Start by resetting any existing state */
2377 xs_reset_transport(transport);
2378
c1384c9c
TM		 2379 queue_delayed_work(rpciod_workqueue,
2380 &transport->connect_worker,
2381 xprt->reestablish_timeout);
03bf4b70 2382 xprt->reestablish_timeout <<= 1;
61d0a8e6
NB		 2383 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2384 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
03bf4b70
CL		 2385 if (xprt->reestablish_timeout > XS_TCP_MAX_REEST_TO)
2386 xprt->reestablish_timeout = XS_TCP_MAX_REEST_TO;
b0d93ad5 2387 } else {
46121cf7 2388 dprintk("RPC: xs_connect scheduled xprt %p\n", xprt);
c1384c9c
TM		 2389 queue_delayed_work(rpciod_workqueue,
2390 &transport->connect_worker, 0);
a246b010
CL		 2391 }
2392}
2393
176e21ee
CL		 2394/**
2395 * xs_local_print_stats - display AF_LOCAL socket-specifc stats
2396 * @xprt: rpc_xprt struct containing statistics
2397 * @seq: output file
2398 *
2399 */
2400static void xs_local_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2401{
2402 long idle_time = 0;
2403
2404 if (xprt_connected(xprt))
2405 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2406
2407 seq_printf(seq, "\txprt:\tlocal %lu %lu %lu %ld %lu %lu %lu "
15a45206 2408 "%llu %llu %lu %llu %llu\n",
176e21ee
CL		 2409 xprt->stat.bind_count,
2410 xprt->stat.connect_count,
2411 xprt->stat.connect_time,
2412 idle_time,
2413 xprt->stat.sends,
2414 xprt->stat.recvs,
2415 xprt->stat.bad_xids,
2416 xprt->stat.req_u,
15a45206
AA		 2417 xprt->stat.bklog_u,
2418 xprt->stat.max_slots,
2419 xprt->stat.sending_u,
2420 xprt->stat.pending_u);
176e21ee
CL		 2421}
2422
262ca07d
CL		 2423/**
2424 * xs_udp_print_stats - display UDP socket-specifc stats
2425 * @xprt: rpc_xprt struct containing statistics
2426 * @seq: output file
2427 *
2428 */
2429static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2430{
c8475461
CL		 2431 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2432
15a45206
AA		 2433 seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %llu %llu "
2434 "%lu %llu %llu\n",
fbfffbd5 2435 transport->srcport,
262ca07d
CL		 2436 xprt->stat.bind_count,
2437 xprt->stat.sends,
2438 xprt->stat.recvs,
2439 xprt->stat.bad_xids,
2440 xprt->stat.req_u,
15a45206
AA		 2441 xprt->stat.bklog_u,
2442 xprt->stat.max_slots,
2443 xprt->stat.sending_u,
2444 xprt->stat.pending_u);
262ca07d
CL		 2445}
2446
2447/**
2448 * xs_tcp_print_stats - display TCP socket-specifc stats
2449 * @xprt: rpc_xprt struct containing statistics
2450 * @seq: output file
2451 *
2452 */
2453static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2454{
c8475461 2455 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
262ca07d
CL		 2456 long idle_time = 0;
2457
2458 if (xprt_connected(xprt))
2459 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2460
15a45206
AA		 2461 seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu "
2462 "%llu %llu %lu %llu %llu\n",
fbfffbd5 2463 transport->srcport,
262ca07d
CL		 2464 xprt->stat.bind_count,
2465 xprt->stat.connect_count,
2466 xprt->stat.connect_time,
2467 idle_time,
2468 xprt->stat.sends,
2469 xprt->stat.recvs,
2470 xprt->stat.bad_xids,
2471 xprt->stat.req_u,
15a45206
AA		 2472 xprt->stat.bklog_u,
2473 xprt->stat.max_slots,
2474 xprt->stat.sending_u,
2475 xprt->stat.pending_u);
262ca07d
CL		 2476}
2477
4cfc7e60
RI		 2478/*
2479 * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason
2480 * we allocate pages instead doing a kmalloc like rpc_malloc is because we want
2481 * to use the server side send routines.
2482 */
5a51f13a 2483static void *bc_malloc(struct rpc_task *task, size_t size)
4cfc7e60
RI		 2484{
2485 struct page *page;
2486 struct rpc_buffer *buf;
2487
b8a13d03
WAA		 2488 WARN_ON_ONCE(size > PAGE_SIZE - sizeof(struct rpc_buffer));
2489 if (size > PAGE_SIZE - sizeof(struct rpc_buffer))
2490 return NULL;
4cfc7e60 2491
b8a13d03 2492 page = alloc_page(GFP_KERNEL);
4cfc7e60
RI		 2493 if (!page)
2494 return NULL;
2495
2496 buf = page_address(page);
2497 buf->len = PAGE_SIZE;
2498
2499 return buf->data;
2500}
2501
2502/*
2503 * Free the space allocated in the bc_alloc routine
2504 */
5a51f13a 2505static void bc_free(void *buffer)
4cfc7e60
RI		 2506{
2507 struct rpc_buffer *buf;
2508
2509 if (!buffer)
2510 return;
2511
2512 buf = container_of(buffer, struct rpc_buffer, data);
2513 free_page((unsigned long)buf);
2514}
2515
2516/*
2517 * Use the svc_sock to send the callback. Must be called with svsk->sk_mutex
2518 * held. Borrows heavily from svc_tcp_sendto and xs_tcp_send_request.
2519 */
2520static int bc_sendto(struct rpc_rqst *req)
2521{
2522 int len;
2523 struct xdr_buf *xbufp = &req->rq_snd_buf;
2524 struct rpc_xprt *xprt = req->rq_xprt;
2525 struct sock_xprt *transport =
2526 container_of(xprt, struct sock_xprt, xprt);
2527 struct socket *sock = transport->sock;
2528 unsigned long headoff;
2529 unsigned long tailoff;
2530
61677eee 2531 xs_encode_stream_record_marker(xbufp);
4cfc7e60
RI		 2532
2533 tailoff = (unsigned long)xbufp->tail[0].iov_base & ~PAGE_MASK;
2534 headoff = (unsigned long)xbufp->head[0].iov_base & ~PAGE_MASK;
2535 len = svc_send_common(sock, xbufp,
2536 virt_to_page(xbufp->head[0].iov_base), headoff,
2537 xbufp->tail[0].iov_base, tailoff);
2538
2539 if (len != xbufp->len) {
2540 printk(KERN_NOTICE "Error sending entire callback!\n");
2541 len = -EAGAIN;
2542 }
2543
2544 return len;
2545}
2546
2547/*
2548 * The send routine. Borrows from svc_send
2549 */
2550static int bc_send_request(struct rpc_task *task)
2551{
2552 struct rpc_rqst *req = task->tk_rqstp;
2553 struct svc_xprt *xprt;
7fc56136 2554 int len;
4cfc7e60
RI		 2555
2556 dprintk("sending request with xid: %08x\n", ntohl(req->rq_xid));
2557 /*
2558 * Get the server socket associated with this callback xprt
2559 */
2560 xprt = req->rq_xprt->bc_xprt;
4cfc7e60
RI		 2561
2562 /*
2563 * Grab the mutex to serialize data as the connection is shared
2564 * with the fore channel
2565 */
2566 if (!mutex_trylock(&xprt->xpt_mutex)) {
2567 rpc_sleep_on(&xprt->xpt_bc_pending, task, NULL);
2568 if (!mutex_trylock(&xprt->xpt_mutex))
2569 return -EAGAIN;
2570 rpc_wake_up_queued_task(&xprt->xpt_bc_pending, task);
2571 }
2572 if (test_bit(XPT_DEAD, &xprt->xpt_flags))
2573 len = -ENOTCONN;
2574 else
2575 len = bc_sendto(req);
2576 mutex_unlock(&xprt->xpt_mutex);
2577
2578 if (len > 0)
2579 len = 0;
2580
2581 return len;
2582}
2583
2584/*
2585 * The close routine. Since this is client initiated, we do nothing
2586 */
2587
2588static void bc_close(struct rpc_xprt *xprt)
2589{
4cfc7e60
RI		 2590}
2591
2592/*
2593 * The xprt destroy routine. Again, because this connection is client
2594 * initiated, we do nothing
2595 */
2596
2597static void bc_destroy(struct rpc_xprt *xprt)
2598{
47f72efa
KM		 2599 dprintk("RPC: bc_destroy xprt %p\n", xprt);
2600
2601 xs_xprt_free(xprt);
2602 module_put(THIS_MODULE);
4cfc7e60
RI		 2603}
2604
176e21ee
CL		 2605static struct rpc_xprt_ops xs_local_ops = {
2606 .reserve_xprt = xprt_reserve_xprt,
2607 .release_xprt = xs_tcp_release_xprt,
f39c1bfb 2608 .alloc_slot = xprt_alloc_slot,
176e21ee
CL		 2609 .rpcbind = xs_local_rpcbind,
2610 .set_port = xs_local_set_port,
dc107402 2611 .connect = xs_local_connect,
176e21ee
CL		 2612 .buf_alloc = rpc_malloc,
2613 .buf_free = rpc_free,
2614 .send_request = xs_local_send_request,
2615 .set_retrans_timeout = xprt_set_retrans_timeout_def,
2616 .close = xs_close,
a1311d87 2617 .destroy = xs_destroy,
176e21ee 2618 .print_stats = xs_local_print_stats,
d67fa4d8
JL		 2619 .enable_swap = xs_enable_swap,
2620 .disable_swap = xs_disable_swap,
176e21ee
CL		 2621};
2622
262965f5 2623static struct rpc_xprt_ops xs_udp_ops = {
43118c29 2624 .set_buffer_size = xs_udp_set_buffer_size,
12a80469 2625 .reserve_xprt = xprt_reserve_xprt_cong,
49e9a890 2626 .release_xprt = xprt_release_xprt_cong,
f39c1bfb 2627 .alloc_slot = xprt_alloc_slot,
45160d62 2628 .rpcbind = rpcb_getport_async,
92200412 2629 .set_port = xs_set_port,
262965f5 2630 .connect = xs_connect,
02107148
CL		 2631 .buf_alloc = rpc_malloc,
2632 .buf_free = rpc_free,
262965f5 2633 .send_request = xs_udp_send_request,
fe3aca29 2634 .set_retrans_timeout = xprt_set_retrans_timeout_rtt,
46c0ee8b 2635 .timer = xs_udp_timer,
a58dd398 2636 .release_request = xprt_release_rqst_cong,
262965f5
CL		 2637 .close = xs_close,
2638 .destroy = xs_destroy,
262ca07d 2639 .print_stats = xs_udp_print_stats,
d67fa4d8
JL		 2640 .enable_swap = xs_enable_swap,
2641 .disable_swap = xs_disable_swap,
4a068258 2642 .inject_disconnect = xs_inject_disconnect,
262965f5
CL		 2643};
2644
2645static struct rpc_xprt_ops xs_tcp_ops = {
12a80469 2646 .reserve_xprt = xprt_reserve_xprt,
e0ab53de 2647 .release_xprt = xs_tcp_release_xprt,
f39c1bfb 2648 .alloc_slot = xprt_lock_and_alloc_slot,
45160d62 2649 .rpcbind = rpcb_getport_async,
92200412 2650 .set_port = xs_set_port,
0b9e7943 2651 .connect = xs_connect,
02107148
CL		 2652 .buf_alloc = rpc_malloc,
2653 .buf_free = rpc_free,
262965f5 2654 .send_request = xs_tcp_send_request,
fe3aca29 2655 .set_retrans_timeout = xprt_set_retrans_timeout_def,
c627d31b 2656 .close = xs_tcp_shutdown,
9903cd1c 2657 .destroy = xs_destroy,
262ca07d 2658 .print_stats = xs_tcp_print_stats,
d67fa4d8
JL		 2659 .enable_swap = xs_enable_swap,
2660 .disable_swap = xs_disable_swap,
4a068258 2661 .inject_disconnect = xs_inject_disconnect,
42e5c3e2
CL		 2662#ifdef CONFIG_SUNRPC_BACKCHANNEL
2663 .bc_setup = xprt_setup_bc,
76566773 2664 .bc_up = xs_tcp_bc_up,
42e5c3e2
CL		 2665 .bc_free_rqst = xprt_free_bc_rqst,
2666 .bc_destroy = xprt_destroy_bc,
2667#endif
a246b010
CL		 2668};
2669
4cfc7e60
RI		 2670/*
2671 * The rpc_xprt_ops for the server backchannel
2672 */
2673
2674static struct rpc_xprt_ops bc_tcp_ops = {
2675 .reserve_xprt = xprt_reserve_xprt,
2676 .release_xprt = xprt_release_xprt,
84e28a30 2677 .alloc_slot = xprt_alloc_slot,
4cfc7e60
RI		 2678 .buf_alloc = bc_malloc,
2679 .buf_free = bc_free,
2680 .send_request = bc_send_request,
2681 .set_retrans_timeout = xprt_set_retrans_timeout_def,
2682 .close = bc_close,
2683 .destroy = bc_destroy,
2684 .print_stats = xs_tcp_print_stats,
d67fa4d8
JL		 2685 .enable_swap = xs_enable_swap,
2686 .disable_swap = xs_disable_swap,
4a068258 2687 .inject_disconnect = xs_inject_disconnect,
4cfc7e60
RI		 2688};
2689
92476850
CL		 2690static int xs_init_anyaddr(const int family, struct sockaddr *sap)
2691{
2692 static const struct sockaddr_in sin = {
2693 .sin_family = AF_INET,
2694 .sin_addr.s_addr = htonl(INADDR_ANY),
2695 };
2696 static const struct sockaddr_in6 sin6 = {
2697 .sin6_family = AF_INET6,
2698 .sin6_addr = IN6ADDR_ANY_INIT,
2699 };
2700
2701 switch (family) {
176e21ee
CL		 2702 case AF_LOCAL:
2703 break;
92476850
CL		 2704 case AF_INET:
2705 memcpy(sap, &sin, sizeof(sin));
2706 break;
2707 case AF_INET6:
2708 memcpy(sap, &sin6, sizeof(sin6));
2709 break;
2710 default:
2711 dprintk("RPC: %s: Bad address family\n", __func__);
2712 return -EAFNOSUPPORT;
2713 }
2714 return 0;
2715}
2716
3c341b0b 2717static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
d9ba131d
TM		 2718 unsigned int slot_table_size,
2719 unsigned int max_slot_table_size)
c8541ecd
CL		 2720{
2721 struct rpc_xprt *xprt;
ffc2e518 2722 struct sock_xprt *new;
c8541ecd 2723
96802a09 2724 if (args->addrlen > sizeof(xprt->addr)) {
46121cf7 2725 dprintk("RPC: xs_setup_xprt: address too large\n");
c8541ecd
CL		 2726 return ERR_PTR(-EBADF);
2727 }
2728
d9ba131d
TM		 2729 xprt = xprt_alloc(args->net, sizeof(*new), slot_table_size,
2730 max_slot_table_size);
bd1722d4 2731 if (xprt == NULL) {
46121cf7
CL		 2732 dprintk("RPC: xs_setup_xprt: couldn't allocate "
2733 "rpc_xprt\n");
c8541ecd
CL		 2734 return ERR_PTR(-ENOMEM);
2735 }
c8541ecd 2736
bd1722d4 2737 new = container_of(xprt, struct sock_xprt, xprt);
edc1b01c 2738 mutex_init(&new->recv_mutex);
96802a09
FM		 2739 memcpy(&xprt->addr, args->dstaddr, args->addrlen);
2740 xprt->addrlen = args->addrlen;
d3bc9a1d 2741 if (args->srcaddr)
fbfffbd5 2742 memcpy(&new->srcaddr, args->srcaddr, args->addrlen);
92476850
CL		 2743 else {
2744 int err;
2745 err = xs_init_anyaddr(args->dstaddr->sa_family,
2746 (struct sockaddr *)&new->srcaddr);
2aa13531
SK		 2747 if (err != 0) {
2748 xprt_free(xprt);
92476850 2749 return ERR_PTR(err);
2aa13531 2750 }
92476850 2751 }
c8541ecd
CL		 2752
2753 return xprt;
2754}
2755
176e21ee
CL		 2756static const struct rpc_timeout xs_local_default_timeout = {
2757 .to_initval = 10 * HZ,
2758 .to_maxval = 10 * HZ,
2759 .to_retries = 2,
2760};
2761
2762/**
2763 * xs_setup_local - Set up transport to use an AF_LOCAL socket
2764 * @args: rpc transport creation arguments
2765 *
2766 * AF_LOCAL is a "tpi_cots_ord" transport, just like TCP
2767 */
2768static struct rpc_xprt *xs_setup_local(struct xprt_create *args)
2769{
2770 struct sockaddr_un *sun = (struct sockaddr_un *)args->dstaddr;
2771 struct sock_xprt *transport;
2772 struct rpc_xprt *xprt;
2773 struct rpc_xprt *ret;
2774
d9ba131d
TM		 2775 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2776 xprt_max_tcp_slot_table_entries);
176e21ee
CL		 2777 if (IS_ERR(xprt))
2778 return xprt;
2779 transport = container_of(xprt, struct sock_xprt, xprt);
2780
2781 xprt->prot = 0;
2782 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2783 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2784
2785 xprt->bind_timeout = XS_BIND_TO;
2786 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2787 xprt->idle_timeout = XS_IDLE_DISC_TO;
2788
2789 xprt->ops = &xs_local_ops;
2790 xprt->timeout = &xs_local_default_timeout;
2791
a2648094 2792 INIT_WORK(&transport->recv_worker, xs_local_data_receive_workfn);
93dc41bd
N		 2793 INIT_DELAYED_WORK(&transport->connect_worker,
2794 xs_dummy_setup_socket);
2795
176e21ee
CL		 2796 switch (sun->sun_family) {
2797 case AF_LOCAL:
2798 if (sun->sun_path[0] != '/') {
2799 dprintk("RPC: bad AF_LOCAL address: %s\n",
2800 sun->sun_path);
2801 ret = ERR_PTR(-EINVAL);
2802 goto out_err;
2803 }
2804 xprt_set_bound(xprt);
176e21ee 2805 xs_format_peer_addresses(xprt, "local", RPCBIND_NETID_LOCAL);
7073ea87
BF		 2806 ret = ERR_PTR(xs_local_setup_socket(transport));
2807 if (ret)
2808 goto out_err;
176e21ee
CL		 2809 break;
2810 default:
2811 ret = ERR_PTR(-EAFNOSUPPORT);
2812 goto out_err;
2813 }
2814
2815 dprintk("RPC: set up xprt to %s via AF_LOCAL\n",
2816 xprt->address_strings[RPC_DISPLAY_ADDR]);
2817
2818 if (try_module_get(THIS_MODULE))
2819 return xprt;
2820 ret = ERR_PTR(-EINVAL);
2821out_err:
315f3812 2822 xs_xprt_free(xprt);
176e21ee
CL		 2823 return ret;
2824}
2825
2881ae74
TM		 2826static const struct rpc_timeout xs_udp_default_timeout = {
2827 .to_initval = 5 * HZ,
2828 .to_maxval = 30 * HZ,
2829 .to_increment = 5 * HZ,
2830 .to_retries = 5,
2831};
2832
9903cd1c
CL		 2833/**
2834 * xs_setup_udp - Set up transport to use a UDP socket
96802a09 2835 * @args: rpc transport creation arguments
9903cd1c
CL		 2836 *
2837 */
483066d6 2838static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
a246b010 2839{
8f9d5b1a 2840 struct sockaddr *addr = args->dstaddr;
c8541ecd 2841 struct rpc_xprt *xprt;
c8475461 2842 struct sock_xprt *transport;
0a68b0be 2843 struct rpc_xprt *ret;
a246b010 2844
d9ba131d
TM		 2845 xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries,
2846 xprt_udp_slot_table_entries);
c8541ecd
CL		 2847 if (IS_ERR(xprt))
2848 return xprt;
c8475461 2849 transport = container_of(xprt, struct sock_xprt, xprt);
a246b010 2850
ec739ef0 2851 xprt->prot = IPPROTO_UDP;
808012fb 2852 xprt->tsh_size = 0;
a246b010
CL		 2853 /* XXX: header size can vary due to auth type, IPv6, etc. */
2854 xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
2855
03bf4b70 2856 xprt->bind_timeout = XS_BIND_TO;
03bf4b70
CL		 2857 xprt->reestablish_timeout = XS_UDP_REEST_TO;
2858 xprt->idle_timeout = XS_IDLE_DISC_TO;
a246b010 2859
262965f5 2860 xprt->ops = &xs_udp_ops;
a246b010 2861
ba7392bb 2862 xprt->timeout = &xs_udp_default_timeout;
a246b010 2863
f9b2ee71 2864 INIT_WORK(&transport->recv_worker, xs_udp_data_receive_workfn);
edc1b01c
TM		 2865 INIT_DELAYED_WORK(&transport->connect_worker, xs_udp_setup_socket);
2866
8f9d5b1a
CL		 2867 switch (addr->sa_family) {
2868 case AF_INET:
2869 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2870 xprt_set_bound(xprt);
2871
9dc3b095 2872 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
8f9d5b1a
CL		 2873 break;
2874 case AF_INET6:
2875 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2876 xprt_set_bound(xprt);
2877
9dc3b095 2878 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
8f9d5b1a
CL		 2879 break;
2880 default:
0a68b0be
BF		 2881 ret = ERR_PTR(-EAFNOSUPPORT);
2882 goto out_err;
8f9d5b1a
CL		 2883 }
2884
c740eff8
CL		 2885 if (xprt_bound(xprt))
2886 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2887 xprt->address_strings[RPC_DISPLAY_ADDR],
2888 xprt->address_strings[RPC_DISPLAY_PORT],
2889 xprt->address_strings[RPC_DISPLAY_PROTO]);
2890 else
2891 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
2892 xprt->address_strings[RPC_DISPLAY_ADDR],
2893 xprt->address_strings[RPC_DISPLAY_PROTO]);
edb267a6 2894
bc25571e
TT		 2895 if (try_module_get(THIS_MODULE))
2896 return xprt;
0a68b0be
BF		 2897 ret = ERR_PTR(-EINVAL);
2898out_err:
315f3812 2899 xs_xprt_free(xprt);
0a68b0be 2900 return ret;
a246b010
CL		 2901}
2902
2881ae74
TM		 2903static const struct rpc_timeout xs_tcp_default_timeout = {
2904 .to_initval = 60 * HZ,
2905 .to_maxval = 60 * HZ,
2906 .to_retries = 2,
2907};
2908
9903cd1c
CL		 2909/**
2910 * xs_setup_tcp - Set up transport to use a TCP socket
96802a09 2911 * @args: rpc transport creation arguments
9903cd1c
CL		 2912 *
2913 */
483066d6 2914static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
a246b010 2915{
8f9d5b1a 2916 struct sockaddr *addr = args->dstaddr;
c8541ecd 2917 struct rpc_xprt *xprt;
c8475461 2918 struct sock_xprt *transport;
0a68b0be 2919 struct rpc_xprt *ret;
b7993ceb
TM		 2920 unsigned int max_slot_table_size = xprt_max_tcp_slot_table_entries;
2921
2922 if (args->flags & XPRT_CREATE_INFINITE_SLOTS)
2923 max_slot_table_size = RPC_MAX_SLOT_TABLE_LIMIT;
a246b010 2924
d9ba131d 2925 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
b7993ceb 2926 max_slot_table_size);
c8541ecd
CL		 2927 if (IS_ERR(xprt))
2928 return xprt;
c8475461 2929 transport = container_of(xprt, struct sock_xprt, xprt);
a246b010 2930
ec739ef0 2931 xprt->prot = IPPROTO_TCP;
808012fb 2932 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
808012fb 2933 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
a246b010 2934
03bf4b70 2935 xprt->bind_timeout = XS_BIND_TO;
03bf4b70
CL		 2936 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2937 xprt->idle_timeout = XS_IDLE_DISC_TO;
a246b010 2938
262965f5 2939 xprt->ops = &xs_tcp_ops;
ba7392bb 2940 xprt->timeout = &xs_tcp_default_timeout;
a246b010 2941
edc1b01c
TM		 2942 INIT_WORK(&transport->recv_worker, xs_tcp_data_receive_workfn);
2943 INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_setup_socket);
2944
8f9d5b1a
CL		 2945 switch (addr->sa_family) {
2946 case AF_INET:
2947 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2948 xprt_set_bound(xprt);
2949
9dc3b095 2950 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
8f9d5b1a
CL		 2951 break;
2952 case AF_INET6:
2953 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2954 xprt_set_bound(xprt);
2955
9dc3b095 2956 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
8f9d5b1a
CL		 2957 break;
2958 default:
0a68b0be
BF		 2959 ret = ERR_PTR(-EAFNOSUPPORT);
2960 goto out_err;
8f9d5b1a
CL		 2961 }
2962
c740eff8
CL		 2963 if (xprt_bound(xprt))
2964 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2965 xprt->address_strings[RPC_DISPLAY_ADDR],
2966 xprt->address_strings[RPC_DISPLAY_PORT],
2967 xprt->address_strings[RPC_DISPLAY_PROTO]);
2968 else
2969 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
2970 xprt->address_strings[RPC_DISPLAY_ADDR],
2971 xprt->address_strings[RPC_DISPLAY_PROTO]);
2972
bc25571e
TT		 2973 if (try_module_get(THIS_MODULE))
2974 return xprt;
0a68b0be
BF		 2975 ret = ERR_PTR(-EINVAL);
2976out_err:
315f3812 2977 xs_xprt_free(xprt);
0a68b0be 2978 return ret;
a246b010 2979}
282b32e1 2980
f300baba
AB		 2981/**
2982 * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket
2983 * @args: rpc transport creation arguments
2984 *
2985 */
2986static struct rpc_xprt *xs_setup_bc_tcp(struct xprt_create *args)
2987{
2988 struct sockaddr *addr = args->dstaddr;
2989 struct rpc_xprt *xprt;
2990 struct sock_xprt *transport;
2991 struct svc_sock *bc_sock;
0a68b0be 2992 struct rpc_xprt *ret;
f300baba 2993
d9ba131d
TM		 2994 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2995 xprt_tcp_slot_table_entries);
f300baba
AB		 2996 if (IS_ERR(xprt))
2997 return xprt;
2998 transport = container_of(xprt, struct sock_xprt, xprt);
2999
3000 xprt->prot = IPPROTO_TCP;
3001 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
3002 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
3003 xprt->timeout = &xs_tcp_default_timeout;
3004
3005 /* backchannel */
3006 xprt_set_bound(xprt);
3007 xprt->bind_timeout = 0;
f300baba
AB		 3008 xprt->reestablish_timeout = 0;
3009 xprt->idle_timeout = 0;
3010
f300baba
AB		 3011 xprt->ops = &bc_tcp_ops;
3012
3013 switch (addr->sa_family) {
3014 case AF_INET:
3015 xs_format_peer_addresses(xprt, "tcp",
3016 RPCBIND_NETID_TCP);
3017 break;
3018 case AF_INET6:
3019 xs_format_peer_addresses(xprt, "tcp",
3020 RPCBIND_NETID_TCP6);
3021 break;
3022 default:
0a68b0be
BF		 3023 ret = ERR_PTR(-EAFNOSUPPORT);
3024 goto out_err;
f300baba
AB		 3025 }
3026
50fa0d40
PE		 3027 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
3028 xprt->address_strings[RPC_DISPLAY_ADDR],
3029 xprt->address_strings[RPC_DISPLAY_PORT],
3030 xprt->address_strings[RPC_DISPLAY_PROTO]);
f300baba 3031
99de8ea9
BF		 3032 /*
3033 * Once we've associated a backchannel xprt with a connection,
28303ca3
WM		 3034 * we want to keep it around as long as the connection lasts,
3035 * in case we need to start using it for a backchannel again;
3036 * this reference won't be dropped until bc_xprt is destroyed.
99de8ea9
BF		 3037 */
3038 xprt_get(xprt);
3039 args->bc_xprt->xpt_bc_xprt = xprt;
3040 xprt->bc_xprt = args->bc_xprt;
3041 bc_sock = container_of(args->bc_xprt, struct svc_sock, sk_xprt);
3042 transport->sock = bc_sock->sk_sock;
3043 transport->inet = bc_sock->sk_sk;
3044
f300baba
AB		 3045 /*
3046 * Since we don't want connections for the backchannel, we set
3047 * the xprt status to connected
3048 */
3049 xprt_set_connected(xprt);
3050
f300baba
AB		 3051 if (try_module_get(THIS_MODULE))
3052 return xprt;
642aab58
KM		 3053
3054 args->bc_xprt->xpt_bc_xprt = NULL;
99de8ea9 3055 xprt_put(xprt);
0a68b0be
BF		 3056 ret = ERR_PTR(-EINVAL);
3057out_err:
315f3812 3058 xs_xprt_free(xprt);
0a68b0be 3059 return ret;
f300baba
AB		 3060}
3061
176e21ee
CL		 3062static struct xprt_class xs_local_transport = {
3063 .list = LIST_HEAD_INIT(xs_local_transport.list),
3064 .name = "named UNIX socket",
3065 .owner = THIS_MODULE,
3066 .ident = XPRT_TRANSPORT_LOCAL,
3067 .setup = xs_setup_local,
3068};
3069
bc25571e
TT		 3070static struct xprt_class xs_udp_transport = {
3071 .list = LIST_HEAD_INIT(xs_udp_transport.list),
3072 .name = "udp",
3073 .owner = THIS_MODULE,
f300baba 3074 .ident = XPRT_TRANSPORT_UDP,
bc25571e
TT		 3075 .setup = xs_setup_udp,
3076};
3077
3078static struct xprt_class xs_tcp_transport = {
3079 .list = LIST_HEAD_INIT(xs_tcp_transport.list),
3080 .name = "tcp",
3081 .owner = THIS_MODULE,
f300baba 3082 .ident = XPRT_TRANSPORT_TCP,
bc25571e
TT		 3083 .setup = xs_setup_tcp,
3084};
3085
f300baba
AB		 3086static struct xprt_class xs_bc_tcp_transport = {
3087 .list = LIST_HEAD_INIT(xs_bc_tcp_transport.list),
3088 .name = "tcp NFSv4.1 backchannel",
3089 .owner = THIS_MODULE,
3090 .ident = XPRT_TRANSPORT_BC_TCP,
3091 .setup = xs_setup_bc_tcp,
3092};
3093
282b32e1 3094/**
bc25571e 3095 * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
282b32e1
CL		 3096 *
3097 */
3098int init_socket_xprt(void)
3099{
f895b252 3100#if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
2b1bec5f 3101 if (!sunrpc_table_header)
0b4d4147 3102 sunrpc_table_header = register_sysctl_table(sunrpc_table);
fbf76683
CL		 3103#endif
3104
176e21ee 3105 xprt_register_transport(&xs_local_transport);
bc25571e
TT		 3106 xprt_register_transport(&xs_udp_transport);
3107 xprt_register_transport(&xs_tcp_transport);
f300baba 3108 xprt_register_transport(&xs_bc_tcp_transport);
bc25571e 3109
282b32e1
CL		 3110 return 0;
3111}
3112
3113/**
bc25571e 3114 * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
282b32e1
CL		 3115 *
3116 */
3117void cleanup_socket_xprt(void)
3118{
f895b252 3119#if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
fbf76683
CL		 3120 if (sunrpc_table_header) {
3121 unregister_sysctl_table(sunrpc_table_header);
3122 sunrpc_table_header = NULL;
3123 }
3124#endif
bc25571e 3125
176e21ee 3126 xprt_unregister_transport(&xs_local_transport);
bc25571e
TT		 3127 xprt_unregister_transport(&xs_udp_transport);
3128 xprt_unregister_transport(&xs_tcp_transport);
f300baba 3129 xprt_unregister_transport(&xs_bc_tcp_transport);
282b32e1 3130}
cbf11071 3131
9bbb9e5a
RR		 3132static int param_set_uint_minmax(const char *val,
3133 const struct kernel_param *kp,
cbf11071
TM		 3134 unsigned int min, unsigned int max)
3135{
00cfaa94 3136 unsigned int num;
cbf11071
TM		 3137 int ret;
3138
3139 if (!val)
3140 return -EINVAL;
00cfaa94 3141 ret = kstrtouint(val, 0, &num);
cbf11071
TM		 3142 if (ret == -EINVAL || num < min || num > max)
3143 return -EINVAL;
3144 *((unsigned int *)kp->arg) = num;
3145 return 0;
3146}
3147
9bbb9e5a 3148static int param_set_portnr(const char *val, const struct kernel_param *kp)
cbf11071
TM		 3149{
3150 return param_set_uint_minmax(val, kp,
3151 RPC_MIN_RESVPORT,
3152 RPC_MAX_RESVPORT);
3153}
3154
9c27847d 3155static const struct kernel_param_ops param_ops_portnr = {
9bbb9e5a
RR		 3156 .set = param_set_portnr,
3157 .get = param_get_uint,
3158};
3159
cbf11071
TM		 3160#define param_check_portnr(name, p) \
3161 __param_check(name, p, unsigned int);
3162
3163module_param_named(min_resvport, xprt_min_resvport, portnr, 0644);
3164module_param_named(max_resvport, xprt_max_resvport, portnr, 0644);
3165
9bbb9e5a
RR		 3166static int param_set_slot_table_size(const char *val,
3167 const struct kernel_param *kp)
cbf11071
TM		 3168{
3169 return param_set_uint_minmax(val, kp,
3170 RPC_MIN_SLOT_TABLE,
3171 RPC_MAX_SLOT_TABLE);
3172}
3173
9c27847d 3174static const struct kernel_param_ops param_ops_slot_table_size = {
9bbb9e5a
RR		 3175 .set = param_set_slot_table_size,
3176 .get = param_get_uint,
3177};
3178
cbf11071
TM		 3179#define param_check_slot_table_size(name, p) \
3180 __param_check(name, p, unsigned int);
3181
d9ba131d
TM		 3182static int param_set_max_slot_table_size(const char *val,
3183 const struct kernel_param *kp)
3184{
3185 return param_set_uint_minmax(val, kp,
3186 RPC_MIN_SLOT_TABLE,
3187 RPC_MAX_SLOT_TABLE_LIMIT);
3188}
3189
9c27847d 3190static const struct kernel_param_ops param_ops_max_slot_table_size = {
d9ba131d
TM		 3191 .set = param_set_max_slot_table_size,
3192 .get = param_get_uint,
3193};
3194
3195#define param_check_max_slot_table_size(name, p) \
3196 __param_check(name, p, unsigned int);
3197
cbf11071
TM		 3198module_param_named(tcp_slot_table_entries, xprt_tcp_slot_table_entries,
3199 slot_table_size, 0644);
d9ba131d
TM		 3200module_param_named(tcp_max_slot_table_entries, xprt_max_tcp_slot_table_entries,
3201 max_slot_table_size, 0644);
cbf11071
TM		 3202module_param_named(udp_slot_table_entries, xprt_udp_slot_table_entries,
3203 slot_table_size, 0644);
3204
Linux 6.x block layer and io_uring tree(s)