]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blame - net/sunrpc/xprtsock.c
projects / mirror_ubuntu-bionic-kernel.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
udp: do fwd memory scheduling on dequeue
[mirror_ubuntu-bionic-kernel.git] / net / sunrpc / xprtsock.c
CommitLineData
a246b010
CL		 1/*
2 * linux/net/sunrpc/xprtsock.c
3 *
4 * Client-side transport implementation for sockets.
5 *
113aa838
AC		 6 * TCP callback races fixes (C) 1998 Red Hat
7 * TCP send fixes (C) 1998 Red Hat
a246b010
CL		 8 * TCP NFS related read + write fixes
9 * (C) 1999 Dave Airlie, University of Limerick, Ireland <airlied@linux.ie>
10 *
11 * Rewrite of larges part of the code in order to stabilize TCP stuff.
12 * Fix behaviour when socket buffer is full.
13 * (C) 1999 Trond Myklebust <trond.myklebust@fys.uio.no>
55aa4f58
CL		 14 *
15 * IP socket transport implementation, (C) 2005 Chuck Lever <cel@netapp.com>
8f9d5b1a
CL		 16 *
17 * IPv6 support contributed by Gilles Quillard, Bull Open Source, 2005.
18 * <gilles.quillard@bull.net>
a246b010
CL		 19 */
20
21#include <linux/types.h>
176e21ee 22#include <linux/string.h>
a246b010 23#include <linux/slab.h>
bc25571e 24#include <linux/module.h>
a246b010 25#include <linux/capability.h>
a246b010
CL		 26#include <linux/pagemap.h>
27#include <linux/errno.h>
28#include <linux/socket.h>
29#include <linux/in.h>
30#include <linux/net.h>
31#include <linux/mm.h>
176e21ee 32#include <linux/un.h>
a246b010
CL		 33#include <linux/udp.h>
34#include <linux/tcp.h>
35#include <linux/sunrpc/clnt.h>
5976687a 36#include <linux/sunrpc/addr.h>
02107148 37#include <linux/sunrpc/sched.h>
4cfc7e60 38#include <linux/sunrpc/svcsock.h>
49c36fcc 39#include <linux/sunrpc/xprtsock.h>
a246b010 40#include <linux/file.h>
9e00abc3 41#ifdef CONFIG_SUNRPC_BACKCHANNEL
44b98efd
RL		 42#include <linux/sunrpc/bc_xprt.h>
43#endif
a246b010
CL		 44
45#include <net/sock.h>
46#include <net/checksum.h>
47#include <net/udp.h>
48#include <net/tcp.h>
49
40b5ea0c
TM		 50#include <trace/events/sunrpc.h>
51
4cfc7e60 52#include "sunrpc.h"
176e21ee
CL		 53
54static void xs_close(struct rpc_xprt *xprt);
55
c556b754
CL		 56/*
57 * xprtsock tunables
58 */
09acfea5
TM		 59static unsigned int xprt_udp_slot_table_entries = RPC_DEF_SLOT_TABLE;
60static unsigned int xprt_tcp_slot_table_entries = RPC_MIN_SLOT_TABLE;
61static unsigned int xprt_max_tcp_slot_table_entries = RPC_MAX_SLOT_TABLE;
c556b754 62
09acfea5
TM		 63static unsigned int xprt_min_resvport = RPC_DEF_MIN_RESVPORT;
64static unsigned int xprt_max_resvport = RPC_DEF_MAX_RESVPORT;
c556b754 65
54c09874
TM		 66#if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
67
7d1e8255 68#define XS_TCP_LINGER_TO (15U * HZ)
25fe6142 69static unsigned int xs_tcp_fin_timeout __read_mostly = XS_TCP_LINGER_TO;
7d1e8255 70
fbf76683
CL		 71/*
72 * We can register our own files under /proc/sys/sunrpc by
73 * calling register_sysctl_table() again. The files in that
74 * directory become the union of all files registered there.
75 *
76 * We simply need to make sure that we don't collide with
77 * someone else's file names!
78 */
79
fbf76683
CL		 80static unsigned int min_slot_table_size = RPC_MIN_SLOT_TABLE;
81static unsigned int max_slot_table_size = RPC_MAX_SLOT_TABLE;
d9ba131d 82static unsigned int max_tcp_slot_table_limit = RPC_MAX_SLOT_TABLE_LIMIT;
fbf76683
CL		 83static unsigned int xprt_min_resvport_limit = RPC_MIN_RESVPORT;
84static unsigned int xprt_max_resvport_limit = RPC_MAX_RESVPORT;
85
86static struct ctl_table_header *sunrpc_table_header;
87
88/*
89 * FIXME: changing the UDP slot table size should also resize the UDP
90 * socket buffers for existing UDP transports
91 */
fe2c6338 92static struct ctl_table xs_tunables_table[] = {
fbf76683 93 {
fbf76683
CL		 94 .procname = "udp_slot_table_entries",
95 .data = &xprt_udp_slot_table_entries,
96 .maxlen = sizeof(unsigned int),
97 .mode = 0644,
6d456111 98 .proc_handler = proc_dointvec_minmax,
fbf76683
CL		 99 .extra1 = &min_slot_table_size,
100 .extra2 = &max_slot_table_size
101 },
102 {
fbf76683
CL		 103 .procname = "tcp_slot_table_entries",
104 .data = &xprt_tcp_slot_table_entries,
105 .maxlen = sizeof(unsigned int),
106 .mode = 0644,
6d456111 107 .proc_handler = proc_dointvec_minmax,
fbf76683
CL		 108 .extra1 = &min_slot_table_size,
109 .extra2 = &max_slot_table_size
110 },
d9ba131d
TM		 111 {
112 .procname = "tcp_max_slot_table_entries",
113 .data = &xprt_max_tcp_slot_table_entries,
114 .maxlen = sizeof(unsigned int),
115 .mode = 0644,
116 .proc_handler = proc_dointvec_minmax,
117 .extra1 = &min_slot_table_size,
118 .extra2 = &max_tcp_slot_table_limit
119 },
fbf76683 120 {
fbf76683
CL		 121 .procname = "min_resvport",
122 .data = &xprt_min_resvport,
123 .maxlen = sizeof(unsigned int),
124 .mode = 0644,
6d456111 125 .proc_handler = proc_dointvec_minmax,
fbf76683 126 .extra1 = &xprt_min_resvport_limit,
e08ea3a9 127 .extra2 = &xprt_max_resvport
fbf76683
CL		 128 },
129 {
fbf76683
CL		 130 .procname = "max_resvport",
131 .data = &xprt_max_resvport,
132 .maxlen = sizeof(unsigned int),
133 .mode = 0644,
6d456111 134 .proc_handler = proc_dointvec_minmax,
e08ea3a9 135 .extra1 = &xprt_min_resvport,
fbf76683
CL		 136 .extra2 = &xprt_max_resvport_limit
137 },
25fe6142
TM		 138 {
139 .procname = "tcp_fin_timeout",
140 .data = &xs_tcp_fin_timeout,
141 .maxlen = sizeof(xs_tcp_fin_timeout),
142 .mode = 0644,
6d456111 143 .proc_handler = proc_dointvec_jiffies,
fbf76683 144 },
f8572d8f 145 { },
fbf76683
CL		 146};
147
fe2c6338 148static struct ctl_table sunrpc_table[] = {
fbf76683 149 {
fbf76683
CL		 150 .procname = "sunrpc",
151 .mode = 0555,
152 .child = xs_tunables_table
153 },
f8572d8f 154 { },
fbf76683
CL		 155};
156
157#endif
158
03bf4b70
CL		 159/*
160 * Wait duration for a reply from the RPC portmapper.
161 */
162#define XS_BIND_TO (60U * HZ)
163
164/*
165 * Delay if a UDP socket connect error occurs. This is most likely some
166 * kind of resource problem on the local host.
167 */
168#define XS_UDP_REEST_TO (2U * HZ)
169
170/*
171 * The reestablish timeout allows clients to delay for a bit before attempting
172 * to reconnect to a server that just dropped our connection.
173 *
174 * We implement an exponential backoff when trying to reestablish a TCP
175 * transport connection with the server. Some servers like to drop a TCP
176 * connection when they are overworked, so we start with a short timeout and
177 * increase over time if the server is down or not responding.
178 */
179#define XS_TCP_INIT_REEST_TO (3U * HZ)
03bf4b70
CL		 180
181/*
182 * TCP idle timeout; client drops the transport socket if it is idle
183 * for this long. Note that we also timeout UDP sockets to prevent
184 * holding port numbers when there is no RPC traffic.
185 */
186#define XS_IDLE_DISC_TO (5U * 60 * HZ)
187
f895b252 188#if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
a246b010 189# undef RPC_DEBUG_DATA
9903cd1c 190# define RPCDBG_FACILITY RPCDBG_TRANS
a246b010
CL		 191#endif
192
a246b010 193#ifdef RPC_DEBUG_DATA
9903cd1c 194static void xs_pktdump(char *msg, u32 *packet, unsigned int count)
a246b010 195{
9903cd1c
CL		 196 u8 *buf = (u8 *) packet;
197 int j;
a246b010 198
46121cf7 199 dprintk("RPC: %s\n", msg);
a246b010
CL		 200 for (j = 0; j < count && j < 128; j += 4) {
201 if (!(j & 31)) {
202 if (j)
203 dprintk("\n");
204 dprintk("0x%04x ", j);
205 }
206 dprintk("%02x%02x%02x%02x ",
207 buf[j], buf[j+1], buf[j+2], buf[j+3]);
208 }
209 dprintk("\n");
210}
211#else
9903cd1c 212static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count)
a246b010
CL		 213{
214 /* NOP */
215}
216#endif
217
2118071d
TM		 218static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
219{
220 return (struct rpc_xprt *) sk->sk_user_data;
221}
222
95392c59
CL		 223static inline struct sockaddr *xs_addr(struct rpc_xprt *xprt)
224{
225 return (struct sockaddr *) &xprt->addr;
226}
227
176e21ee
CL		 228static inline struct sockaddr_un *xs_addr_un(struct rpc_xprt *xprt)
229{
230 return (struct sockaddr_un *) &xprt->addr;
231}
232
95392c59 233static inline struct sockaddr_in *xs_addr_in(struct rpc_xprt *xprt)
edb267a6 234{
95392c59
CL		 235 return (struct sockaddr_in *) &xprt->addr;
236}
237
238static inline struct sockaddr_in6 *xs_addr_in6(struct rpc_xprt *xprt)
239{
240 return (struct sockaddr_in6 *) &xprt->addr;
241}
242
c877b849 243static void xs_format_common_peer_addresses(struct rpc_xprt *xprt)
edb267a6 244{
c877b849 245 struct sockaddr *sap = xs_addr(xprt);
9dc3b095
CL		 246 struct sockaddr_in6 *sin6;
247 struct sockaddr_in *sin;
176e21ee 248 struct sockaddr_un *sun;
c877b849 249 char buf[128];
edb267a6 250
9dc3b095 251 switch (sap->sa_family) {
176e21ee
CL		 252 case AF_LOCAL:
253 sun = xs_addr_un(xprt);
254 strlcpy(buf, sun->sun_path, sizeof(buf));
255 xprt->address_strings[RPC_DISPLAY_ADDR] =
256 kstrdup(buf, GFP_KERNEL);
257 break;
9dc3b095 258 case AF_INET:
176e21ee
CL		 259 (void)rpc_ntop(sap, buf, sizeof(buf));
260 xprt->address_strings[RPC_DISPLAY_ADDR] =
261 kstrdup(buf, GFP_KERNEL);
9dc3b095 262 sin = xs_addr_in(xprt);
fc0b5791 263 snprintf(buf, sizeof(buf), "%08x", ntohl(sin->sin_addr.s_addr));
9dc3b095
CL		 264 break;
265 case AF_INET6:
176e21ee
CL		 266 (void)rpc_ntop(sap, buf, sizeof(buf));
267 xprt->address_strings[RPC_DISPLAY_ADDR] =
268 kstrdup(buf, GFP_KERNEL);
9dc3b095 269 sin6 = xs_addr_in6(xprt);
fc0b5791 270 snprintf(buf, sizeof(buf), "%pi6", &sin6->sin6_addr);
9dc3b095
CL		 271 break;
272 default:
273 BUG();
756805e7 274 }
176e21ee 275
9dc3b095 276 xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = kstrdup(buf, GFP_KERNEL);
edb267a6
CL		 277}
278
9dc3b095 279static void xs_format_common_peer_ports(struct rpc_xprt *xprt)
4b6473fb 280{
9dc3b095
CL		 281 struct sockaddr *sap = xs_addr(xprt);
282 char buf[128];
4b6473fb 283
81160e66 284 snprintf(buf, sizeof(buf), "%u", rpc_get_port(sap));
c877b849 285 xprt->address_strings[RPC_DISPLAY_PORT] = kstrdup(buf, GFP_KERNEL);
4b6473fb 286
81160e66 287 snprintf(buf, sizeof(buf), "%4hx", rpc_get_port(sap));
c877b849
CL		 288 xprt->address_strings[RPC_DISPLAY_HEX_PORT] = kstrdup(buf, GFP_KERNEL);
289}
4b6473fb 290
9dc3b095
CL		 291static void xs_format_peer_addresses(struct rpc_xprt *xprt,
292 const char *protocol,
293 const char *netid)
c877b849 294{
b454ae90 295 xprt->address_strings[RPC_DISPLAY_PROTO] = protocol;
b454ae90 296 xprt->address_strings[RPC_DISPLAY_NETID] = netid;
c877b849 297 xs_format_common_peer_addresses(xprt);
9dc3b095 298 xs_format_common_peer_ports(xprt);
edb267a6 299}
4b6473fb 300
9dc3b095 301static void xs_update_peer_port(struct rpc_xprt *xprt)
4b6473fb 302{
9dc3b095
CL		 303 kfree(xprt->address_strings[RPC_DISPLAY_HEX_PORT]);
304 kfree(xprt->address_strings[RPC_DISPLAY_PORT]);
4417c8c4 305
9dc3b095 306 xs_format_common_peer_ports(xprt);
edb267a6
CL		 307}
308
309static void xs_free_peer_addresses(struct rpc_xprt *xprt)
310{
33e01dc7
CL		 311 unsigned int i;
312
313 for (i = 0; i < RPC_DISPLAY_MAX; i++)
314 switch (i) {
315 case RPC_DISPLAY_PROTO:
316 case RPC_DISPLAY_NETID:
317 continue;
318 default:
319 kfree(xprt->address_strings[i]);
320 }
edb267a6
CL		 321}
322
b4b5cc85
CL		 323#define XS_SENDMSG_FLAGS (MSG_DONTWAIT | MSG_NOSIGNAL)
324
24c5684b 325static int xs_send_kvec(struct socket *sock, struct sockaddr *addr, int addrlen, struct kvec *vec, unsigned int base, int more)
b4b5cc85 326{
b4b5cc85
CL		 327 struct msghdr msg = {
328 .msg_name = addr,
329 .msg_namelen = addrlen,
24c5684b
TM		 330 .msg_flags = XS_SENDMSG_FLAGS | (more ? MSG_MORE : 0),
331 };
332 struct kvec iov = {
333 .iov_base = vec->iov_base + base,
334 .iov_len = vec->iov_len - base,
b4b5cc85
CL		 335 };
336
24c5684b 337 if (iov.iov_len != 0)
b4b5cc85
CL		 338 return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
339 return kernel_sendmsg(sock, &msg, NULL, 0, 0);
340}
341
f279cd00 342static int xs_send_pagedata(struct socket *sock, struct xdr_buf *xdr, unsigned int base, int more, bool zerocopy, int *sent_p)
b4b5cc85 343{
a6b31d18
TM		 344 ssize_t (*do_sendpage)(struct socket *sock, struct page *page,
345 int offset, size_t size, int flags);
24c5684b
TM		 346 struct page **ppage;
347 unsigned int remainder;
f279cd00 348 int err;
24c5684b
TM		 349
350 remainder = xdr->page_len - base;
351 base += xdr->page_base;
352 ppage = xdr->pages + (base >> PAGE_SHIFT);
353 base &= ~PAGE_MASK;
a6b31d18
TM		 354 do_sendpage = sock->ops->sendpage;
355 if (!zerocopy)
356 do_sendpage = sock_no_sendpage;
24c5684b
TM		 357 for(;;) {
358 unsigned int len = min_t(unsigned int, PAGE_SIZE - base, remainder);
359 int flags = XS_SENDMSG_FLAGS;
b4b5cc85 360
24c5684b 361 remainder -= len;
31303d6c 362 if (more)
24c5684b 363 flags |= MSG_MORE;
31303d6c
TM		 364 if (remainder != 0)
365 flags |= MSG_SENDPAGE_NOTLAST | MSG_MORE;
a6b31d18 366 err = do_sendpage(sock, *ppage, base, len, flags);
24c5684b
TM		 367 if (remainder == 0 || err != len)
368 break;
f279cd00 369 *sent_p += err;
24c5684b
TM		 370 ppage++;
371 base = 0;
372 }
f279cd00
JB		 373 if (err > 0) {
374 *sent_p += err;
375 err = 0;
376 }
377 return err;
b4b5cc85
CL		 378}
379
9903cd1c
CL		 380/**
381 * xs_sendpages - write pages directly to a socket
382 * @sock: socket to send on
383 * @addr: UDP only -- address of destination
384 * @addrlen: UDP only -- length of destination address
385 * @xdr: buffer containing this request
386 * @base: starting position in the buffer
a6b31d18 387 * @zerocopy: true if it is safe to use sendpage()
f279cd00 388 * @sent_p: return the total number of bytes successfully queued for sending
9903cd1c 389 *
a246b010 390 */
f279cd00 391static int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base, bool zerocopy, int *sent_p)
a246b010 392{
24c5684b 393 unsigned int remainder = xdr->len - base;
f279cd00
JB		 394 int err = 0;
395 int sent = 0;
a246b010 396
262965f5 397 if (unlikely(!sock))
fba91afb 398 return -ENOTSOCK;
262965f5 399
24c5684b
TM		 400 if (base != 0) {
401 addr = NULL;
402 addrlen = 0;
403 }
262965f5 404
24c5684b
TM		 405 if (base < xdr->head[0].iov_len || addr != NULL) {
406 unsigned int len = xdr->head[0].iov_len - base;
407 remainder -= len;
408 err = xs_send_kvec(sock, addr, addrlen, &xdr->head[0], base, remainder != 0);
409 if (remainder == 0 || err != len)
a246b010 410 goto out;
f279cd00 411 *sent_p += err;
a246b010
CL		 412 base = 0;
413 } else
24c5684b 414 base -= xdr->head[0].iov_len;
a246b010 415
24c5684b
TM		 416 if (base < xdr->page_len) {
417 unsigned int len = xdr->page_len - base;
418 remainder -= len;
f279cd00
JB		 419 err = xs_send_pagedata(sock, xdr, base, remainder != 0, zerocopy, &sent);
420 *sent_p += sent;
421 if (remainder == 0 || sent != len)
a246b010
CL		 422 goto out;
423 base = 0;
24c5684b
TM		 424 } else
425 base -= xdr->page_len;
426
427 if (base >= xdr->tail[0].iov_len)
f279cd00 428 return 0;
24c5684b 429 err = xs_send_kvec(sock, NULL, 0, &xdr->tail[0], base, 0);
a246b010 430out:
f279cd00
JB		 431 if (err > 0) {
432 *sent_p += err;
433 err = 0;
434 }
435 return err;
a246b010
CL		 436}
437
b6ddf64f
TM		 438static void xs_nospace_callback(struct rpc_task *task)
439{
440 struct sock_xprt *transport = container_of(task->tk_rqstp->rq_xprt, struct sock_xprt, xprt);
441
442 transport->inet->sk_write_pending--;
b6ddf64f
TM		 443}
444
9903cd1c 445/**
262965f5
CL		 446 * xs_nospace - place task on wait queue if transmit was incomplete
447 * @task: task to put to sleep
9903cd1c 448 *
a246b010 449 */
5e3771ce 450static int xs_nospace(struct rpc_task *task)
a246b010 451{
262965f5
CL		 452 struct rpc_rqst *req = task->tk_rqstp;
453 struct rpc_xprt *xprt = req->rq_xprt;
ee0ac0c2 454 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
06ea0bfe 455 struct sock *sk = transport->inet;
24ca9a84 456 int ret = -EAGAIN;
a246b010 457
46121cf7 458 dprintk("RPC: %5u xmit incomplete (%u left of %u)\n",
262965f5
CL		 459 task->tk_pid, req->rq_slen - req->rq_bytes_sent,
460 req->rq_slen);
461
b6ddf64f
TM		 462 /* Protect against races with write_space */
463 spin_lock_bh(&xprt->transport_lock);
464
465 /* Don't race with disconnect */
466 if (xprt_connected(xprt)) {
13331a55
TM		 467 /* wait for more buffer space */
468 sk->sk_write_pending++;
469 xprt_wait_for_buffer_space(task, xs_nospace_callback);
470 } else
5e3771ce 471 ret = -ENOTCONN;
262965f5 472
b6ddf64f 473 spin_unlock_bh(&xprt->transport_lock);
06ea0bfe
TM		 474
475 /* Race breaker in case memory is freed before above code is called */
d48f9ce7
DV		 476 if (ret == -EAGAIN) {
477 struct socket_wq *wq;
478
479 rcu_read_lock();
480 wq = rcu_dereference(sk->sk_wq);
481 set_bit(SOCKWQ_ASYNC_NOSPACE, &wq->flags);
482 rcu_read_unlock();
483
484 sk->sk_write_space(sk);
485 }
5e3771ce 486 return ret;
262965f5
CL		 487}
488
61677eee
CL		 489/*
490 * Construct a stream transport record marker in @buf.
491 */
492static inline void xs_encode_stream_record_marker(struct xdr_buf *buf)
493{
494 u32 reclen = buf->len - sizeof(rpc_fraghdr);
495 rpc_fraghdr *base = buf->head[0].iov_base;
496 *base = cpu_to_be32(RPC_LAST_STREAM_FRAGMENT | reclen);
497}
498
176e21ee
CL		 499/**
500 * xs_local_send_request - write an RPC request to an AF_LOCAL socket
501 * @task: RPC task that manages the state of an RPC request
502 *
503 * Return values:
504 * 0: The request has been sent
505 * EAGAIN: The socket was blocked, please call again later to
506 * complete the request
507 * ENOTCONN: Caller needs to invoke connect logic then call again
508 * other: Some other error occured, the request was not sent
509 */
510static int xs_local_send_request(struct rpc_task *task)
511{
512 struct rpc_rqst *req = task->tk_rqstp;
513 struct rpc_xprt *xprt = req->rq_xprt;
514 struct sock_xprt *transport =
515 container_of(xprt, struct sock_xprt, xprt);
516 struct xdr_buf *xdr = &req->rq_snd_buf;
517 int status;
f279cd00 518 int sent = 0;
176e21ee
CL		 519
520 xs_encode_stream_record_marker(&req->rq_snd_buf);
521
522 xs_pktdump("packet data:",
523 req->rq_svec->iov_base, req->rq_svec->iov_len);
524
f279cd00
JB		 525 status = xs_sendpages(transport->sock, NULL, 0, xdr, req->rq_bytes_sent,
526 true, &sent);
176e21ee
CL		 527 dprintk("RPC: %s(%u) = %d\n",
528 __func__, xdr->len - req->rq_bytes_sent, status);
743c69e7
N		 529
530 if (status == -EAGAIN && sock_writeable(transport->inet))
531 status = -ENOBUFS;
532
f279cd00
JB		 533 if (likely(sent > 0) || status == 0) {
534 req->rq_bytes_sent += sent;
535 req->rq_xmit_bytes_sent += sent;
176e21ee
CL		 536 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
537 req->rq_bytes_sent = 0;
538 return 0;
539 }
540 status = -EAGAIN;
541 }
542
543 switch (status) {
3601c4a9 544 case -ENOBUFS:
b5872f0c 545 break;
176e21ee
CL		 546 case -EAGAIN:
547 status = xs_nospace(task);
548 break;
549 default:
550 dprintk("RPC: sendmsg returned unrecognized error %d\n",
551 -status);
552 case -EPIPE:
553 xs_close(xprt);
554 status = -ENOTCONN;
555 }
556
557 return status;
558}
559
262965f5
CL		 560/**
561 * xs_udp_send_request - write an RPC request to a UDP socket
562 * @task: address of RPC task that manages the state of an RPC request
563 *
564 * Return values:
565 * 0: The request has been sent
566 * EAGAIN: The socket was blocked, please call again later to
567 * complete the request
568 * ENOTCONN: Caller needs to invoke connect logic then call again
25985edc 569 * other: Some other error occurred, the request was not sent
262965f5
CL		 570 */
571static int xs_udp_send_request(struct rpc_task *task)
572{
573 struct rpc_rqst *req = task->tk_rqstp;
574 struct rpc_xprt *xprt = req->rq_xprt;
ee0ac0c2 575 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
262965f5 576 struct xdr_buf *xdr = &req->rq_snd_buf;
f279cd00 577 int sent = 0;
262965f5 578 int status;
a246b010 579
9903cd1c 580 xs_pktdump("packet data:",
a246b010
CL		 581 req->rq_svec->iov_base,
582 req->rq_svec->iov_len);
583
01d37c42
TM		 584 if (!xprt_bound(xprt))
585 return -ENOTCONN;
f279cd00
JB		 586 status = xs_sendpages(transport->sock, xs_addr(xprt), xprt->addrlen,
587 xdr, req->rq_bytes_sent, true, &sent);
a246b010 588
46121cf7 589 dprintk("RPC: xs_udp_send_request(%u) = %d\n",
262965f5 590 xdr->len - req->rq_bytes_sent, status);
a246b010 591
3dedbb5c
JB		 592 /* firewall is blocking us, don't return -EAGAIN or we end up looping */
593 if (status == -EPERM)
594 goto process_status;
595
743c69e7
N		 596 if (status == -EAGAIN && sock_writeable(transport->inet))
597 status = -ENOBUFS;
598
f279cd00
JB		 599 if (sent > 0 || status == 0) {
600 req->rq_xmit_bytes_sent += sent;
601 if (sent >= req->rq_slen)
2199700f
TM		 602 return 0;
603 /* Still some bytes left; set up for a retry later. */
262965f5 604 status = -EAGAIN;
2199700f 605 }
a246b010 606
3dedbb5c 607process_status:
262965f5 608 switch (status) {
fba91afb
TM		 609 case -ENOTSOCK:
610 status = -ENOTCONN;
611 /* Should we call xs_close() here? */
612 break;
b6ddf64f 613 case -EAGAIN:
5e3771ce 614 status = xs_nospace(task);
b6ddf64f 615 break;
262965f5 616 case -ENETUNREACH:
3601c4a9 617 case -ENOBUFS:
262965f5 618 case -EPIPE:
a246b010 619 case -ECONNREFUSED:
3dedbb5c 620 case -EPERM:
a246b010 621 /* When the server has died, an ICMP port unreachable message
9903cd1c 622 * prompts ECONNREFUSED. */
13331a55
TM		 623 break;
624 default:
625 dprintk("RPC: sendmsg returned unrecognized error %d\n",
626 -status);
a246b010 627 }
5fe46e9d 628
262965f5 629 return status;
a246b010
CL		 630}
631
9903cd1c 632/**
262965f5 633 * xs_tcp_send_request - write an RPC request to a TCP socket
9903cd1c
CL		 634 * @task: address of RPC task that manages the state of an RPC request
635 *
636 * Return values:
262965f5
CL		 637 * 0: The request has been sent
638 * EAGAIN: The socket was blocked, please call again later to
639 * complete the request
640 * ENOTCONN: Caller needs to invoke connect logic then call again
25985edc 641 * other: Some other error occurred, the request was not sent
9903cd1c
CL		 642 *
643 * XXX: In the case of soft timeouts, should we eventually give up
262965f5 644 * if sendmsg is not able to make progress?
9903cd1c 645 */
262965f5 646static int xs_tcp_send_request(struct rpc_task *task)
a246b010
CL		 647{
648 struct rpc_rqst *req = task->tk_rqstp;
649 struct rpc_xprt *xprt = req->rq_xprt;
ee0ac0c2 650 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
262965f5 651 struct xdr_buf *xdr = &req->rq_snd_buf;
a6b31d18 652 bool zerocopy = true;
9ffadfbc 653 bool vm_wait = false;
b595bb15 654 int status;
f279cd00 655 int sent;
a246b010 656
61677eee 657 xs_encode_stream_record_marker(&req->rq_snd_buf);
a246b010 658
262965f5
CL		 659 xs_pktdump("packet data:",
660 req->rq_svec->iov_base,
661 req->rq_svec->iov_len);
a6b31d18
TM		 662 /* Don't use zero copy if this is a resend. If the RPC call
663 * completes while the socket holds a reference to the pages,
664 * then we may end up resending corrupted data.
665 */
666 if (task->tk_flags & RPC_TASK_SENT)
667 zerocopy = false;
a246b010
CL		 668
669 /* Continue transmitting the packet/record. We must be careful
670 * to cope with writespace callbacks arriving _after_ we have
262965f5 671 * called sendmsg(). */
a246b010 672 while (1) {
f279cd00
JB		 673 sent = 0;
674 status = xs_sendpages(transport->sock, NULL, 0, xdr,
675 req->rq_bytes_sent, zerocopy, &sent);
a246b010 676
46121cf7 677 dprintk("RPC: xs_tcp_send_request(%u) = %d\n",
262965f5 678 xdr->len - req->rq_bytes_sent, status);
a246b010 679
262965f5
CL		 680 /* If we've sent the entire packet, immediately
681 * reset the count of bytes sent. */
f279cd00
JB		 682 req->rq_bytes_sent += sent;
683 req->rq_xmit_bytes_sent += sent;
262965f5
CL		 684 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
685 req->rq_bytes_sent = 0;
686 return 0;
687 }
a246b010 688
9ffadfbc
TM		 689 WARN_ON_ONCE(sent == 0 && status == 0);
690
691 if (status == -EAGAIN ) {
692 /*
693 * Return EAGAIN if we're sure we're hitting the
694 * socket send buffer limits.
695 */
696 if (test_bit(SOCK_NOSPACE, &transport->sock->flags))
697 break;
698 /*
699 * Did we hit a memory allocation failure?
700 */
701 if (sent == 0) {
702 status = -ENOBUFS;
703 if (vm_wait)
704 break;
705 /* Retry, knowing now that we're below the
706 * socket send buffer limit
707 */
708 vm_wait = true;
709 }
710 continue;
711 }
f580dd04
TM		 712 if (status < 0)
713 break;
9ffadfbc 714 vm_wait = false;
a246b010
CL		 715 }
716
262965f5 717 switch (status) {
fba91afb
TM		 718 case -ENOTSOCK:
719 status = -ENOTCONN;
720 /* Should we call xs_close() here? */
721 break;
262965f5 722 case -EAGAIN:
5e3771ce 723 status = xs_nospace(task);
262965f5 724 break;
262965f5 725 case -ECONNRESET:
2a9e1cfa 726 case -ECONNREFUSED:
262965f5 727 case -ENOTCONN:
3913c78c 728 case -EADDRINUSE:
b5872f0c 729 case -ENOBUFS:
b9d2bb2e 730 case -EPIPE:
13331a55
TM		 731 break;
732 default:
733 dprintk("RPC: sendmsg returned unrecognized error %d\n",
734 -status);
a246b010 735 }
5fe46e9d 736
a246b010
CL		 737 return status;
738}
739
e0ab53de
TM		 740/**
741 * xs_tcp_release_xprt - clean up after a tcp transmission
742 * @xprt: transport
743 * @task: rpc task
744 *
745 * This cleans up if an error causes us to abort the transmission of a request.
746 * In this case, the socket may need to be reset in order to avoid confusing
747 * the server.
748 */
749static void xs_tcp_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
750{
751 struct rpc_rqst *req;
752
753 if (task != xprt->snd_task)
754 return;
755 if (task == NULL)
756 goto out_release;
757 req = task->tk_rqstp;
43cedbf0
TM		 758 if (req == NULL)
759 goto out_release;
e0ab53de
TM		 760 if (req->rq_bytes_sent == 0)
761 goto out_release;
762 if (req->rq_bytes_sent == req->rq_snd_buf.len)
763 goto out_release;
a4f0835c 764 set_bit(XPRT_CLOSE_WAIT, &xprt->state);
e0ab53de
TM		 765out_release:
766 xprt_release_xprt(xprt, task);
767}
768
2a9e1cfa
TM		 769static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk)
770{
771 transport->old_data_ready = sk->sk_data_ready;
772 transport->old_state_change = sk->sk_state_change;
773 transport->old_write_space = sk->sk_write_space;
2118071d 774 transport->old_error_report = sk->sk_error_report;
2a9e1cfa
TM		 775}
776
777static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk)
778{
779 sk->sk_data_ready = transport->old_data_ready;
780 sk->sk_state_change = transport->old_state_change;
781 sk->sk_write_space = transport->old_write_space;
2118071d
TM		 782 sk->sk_error_report = transport->old_error_report;
783}
784
42d42a5b
TM		 785static void xs_sock_reset_state_flags(struct rpc_xprt *xprt)
786{
787 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
788
789 clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state);
790}
791
b70ae915
TM		 792static void xs_sock_reset_connection_flags(struct rpc_xprt *xprt)
793{
794 smp_mb__before_atomic();
795 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
796 clear_bit(XPRT_CLOSING, &xprt->state);
42d42a5b 797 xs_sock_reset_state_flags(xprt);
b70ae915
TM		 798 smp_mb__after_atomic();
799}
800
801static void xs_sock_mark_closed(struct rpc_xprt *xprt)
802{
803 xs_sock_reset_connection_flags(xprt);
804 /* Mark transport as closed and wake up all pending tasks */
805 xprt_disconnect_done(xprt);
806}
807
2118071d
TM		 808/**
809 * xs_error_report - callback to handle TCP socket state errors
810 * @sk: socket
811 *
812 * Note: we don't call sock_error() since there may be a rpc_task
813 * using the socket, and so we don't want to clear sk->sk_err.
814 */
815static void xs_error_report(struct sock *sk)
816{
817 struct rpc_xprt *xprt;
818 int err;
819
820 read_lock_bh(&sk->sk_callback_lock);
821 if (!(xprt = xprt_from_sock(sk)))
822 goto out;
823
824 err = -sk->sk_err;
825 if (err == 0)
826 goto out;
b70ae915
TM		 827 /* Is this a reset event? */
828 if (sk->sk_state == TCP_CLOSE)
829 xs_sock_mark_closed(xprt);
2118071d
TM		 830 dprintk("RPC: xs_error_report client %p, error=%d...\n",
831 xprt, -err);
e8353c76 832 trace_rpc_socket_error(xprt, sk->sk_socket, err);
2118071d
TM		 833 xprt_wake_pending_tasks(xprt, err);
834 out:
835 read_unlock_bh(&sk->sk_callback_lock);
2a9e1cfa
TM		 836}
837
fe315e76 838static void xs_reset_transport(struct sock_xprt *transport)
a246b010 839{
ee0ac0c2
CL		 840 struct socket *sock = transport->sock;
841 struct sock *sk = transport->inet;
6cc7e908 842 struct rpc_xprt *xprt = &transport->xprt;
a246b010 843
fe315e76
CL		 844 if (sk == NULL)
845 return;
9903cd1c 846
264d1df3
JL		 847 if (atomic_read(&transport->xprt.swapper))
848 sk_clear_memalloc(sk);
849
09939204
TM		 850 kernel_sock_shutdown(sock, SHUT_RDWR);
851
edc1b01c 852 mutex_lock(&transport->recv_mutex);
a246b010 853 write_lock_bh(&sk->sk_callback_lock);
ee0ac0c2
CL		 854 transport->inet = NULL;
855 transport->sock = NULL;
a246b010 856
9903cd1c 857 sk->sk_user_data = NULL;
2a9e1cfa
TM		 858
859 xs_restore_old_callbacks(transport, sk);
0c78789e 860 xprt_clear_connected(xprt);
a246b010 861 write_unlock_bh(&sk->sk_callback_lock);
6cc7e908 862 xs_sock_reset_connection_flags(xprt);
edc1b01c 863 mutex_unlock(&transport->recv_mutex);
a246b010 864
6cc7e908 865 trace_rpc_socket_close(xprt, sock);
a246b010 866 sock_release(sock);
fe315e76
CL		 867}
868
869/**
870 * xs_close - close a socket
871 * @xprt: transport
872 *
873 * This is used when all requests are complete; ie, no DRC state remains
874 * on the server we want to save.
f75e6745
TM		 875 *
876 * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with
877 * xs_reset_transport() zeroing the socket from underneath a writer.
fe315e76
CL		 878 */
879static void xs_close(struct rpc_xprt *xprt)
880{
881 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
882
883 dprintk("RPC: xs_close xprt %p\n", xprt);
884
885 xs_reset_transport(transport);
61d0a8e6 886 xprt->reestablish_timeout = 0;
fe315e76 887
62da3b24 888 xprt_disconnect_done(xprt);
a246b010
CL		 889}
890
4a068258
CL		 891static void xs_inject_disconnect(struct rpc_xprt *xprt)
892{
893 dprintk("RPC: injecting transport disconnect on xprt=%p\n",
894 xprt);
895 xprt_disconnect_done(xprt);
896}
897
315f3812
KM		 898static void xs_xprt_free(struct rpc_xprt *xprt)
899{
900 xs_free_peer_addresses(xprt);
901 xprt_free(xprt);
902}
903
9903cd1c
CL		 904/**
905 * xs_destroy - prepare to shutdown a transport
906 * @xprt: doomed transport
907 *
908 */
909static void xs_destroy(struct rpc_xprt *xprt)
a246b010 910{
03c78827
TM		 911 struct sock_xprt *transport = container_of(xprt,
912 struct sock_xprt, xprt);
46121cf7 913 dprintk("RPC: xs_destroy xprt %p\n", xprt);
9903cd1c 914
03c78827 915 cancel_delayed_work_sync(&transport->connect_worker);
a1311d87 916 xs_close(xprt);
edc1b01c 917 cancel_work_sync(&transport->recv_worker);
315f3812 918 xs_xprt_free(xprt);
a1311d87 919 module_put(THIS_MODULE);
a246b010
CL		 920}
921
176e21ee
CL		 922static int xs_local_copy_to_xdr(struct xdr_buf *xdr, struct sk_buff *skb)
923{
924 struct xdr_skb_reader desc = {
925 .skb = skb,
926 .offset = sizeof(rpc_fraghdr),
927 .count = skb->len - sizeof(rpc_fraghdr),
928 };
929
930 if (xdr_partial_copy_from_skb(xdr, 0, &desc, xdr_skb_read_bits) < 0)
931 return -1;
932 if (desc.count)
933 return -1;
934 return 0;
935}
936
937/**
a2648094
TM		 938 * xs_local_data_read_skb
939 * @xprt: transport
940 * @sk: socket
941 * @skb: skbuff
176e21ee
CL		 942 *
943 * Currently this assumes we can read the whole reply in a single gulp.
944 */
a2648094
TM		 945static void xs_local_data_read_skb(struct rpc_xprt *xprt,
946 struct sock *sk,
947 struct sk_buff *skb)
176e21ee
CL		 948{
949 struct rpc_task *task;
176e21ee 950 struct rpc_rqst *rovr;
a2648094 951 int repsize, copied;
176e21ee
CL		 952 u32 _xid;
953 __be32 *xp;
954
176e21ee
CL		 955 repsize = skb->len - sizeof(rpc_fraghdr);
956 if (repsize < 4) {
957 dprintk("RPC: impossible RPC reply size %d\n", repsize);
a2648094 958 return;
176e21ee
CL		 959 }
960
961 /* Copy the XID from the skb... */
962 xp = skb_header_pointer(skb, sizeof(rpc_fraghdr), sizeof(_xid), &_xid);
963 if (xp == NULL)
a2648094 964 return;
176e21ee
CL		 965
966 /* Look up and lock the request corresponding to the given XID */
a2648094 967 spin_lock_bh(&xprt->transport_lock);
176e21ee
CL		 968 rovr = xprt_lookup_rqst(xprt, *xp);
969 if (!rovr)
970 goto out_unlock;
971 task = rovr->rq_task;
972
973 copied = rovr->rq_private_buf.buflen;
974 if (copied > repsize)
975 copied = repsize;
976
977 if (xs_local_copy_to_xdr(&rovr->rq_private_buf, skb)) {
978 dprintk("RPC: sk_buff copy failed\n");
979 goto out_unlock;
980 }
981
982 xprt_complete_rqst(task, copied);
983
984 out_unlock:
a2648094
TM		 985 spin_unlock_bh(&xprt->transport_lock);
986}
987
988static void xs_local_data_receive(struct sock_xprt *transport)
989{
990 struct sk_buff *skb;
991 struct sock *sk;
992 int err;
993
994 mutex_lock(&transport->recv_mutex);
995 sk = transport->inet;
996 if (sk == NULL)
997 goto out;
998 for (;;) {
999 skb = skb_recv_datagram(sk, 0, 1, &err);
42d42a5b
TM		 1000 if (skb != NULL) {
1001 xs_local_data_read_skb(&transport->xprt, sk, skb);
1002 skb_free_datagram(sk, skb);
1003 continue;
1004 }
1005 if (!test_and_clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
a2648094 1006 break;
a2648094
TM		 1007 }
1008out:
1009 mutex_unlock(&transport->recv_mutex);
1010}
1011
1012static void xs_local_data_receive_workfn(struct work_struct *work)
1013{
1014 struct sock_xprt *transport =
1015 container_of(work, struct sock_xprt, recv_worker);
1016 xs_local_data_receive(transport);
176e21ee
CL		 1017}
1018
9903cd1c 1019/**
f9b2ee71
TM		 1020 * xs_udp_data_read_skb - receive callback for UDP sockets
1021 * @xprt: transport
1022 * @sk: socket
1023 * @skb: skbuff
9903cd1c 1024 *
a246b010 1025 */
f9b2ee71
TM		 1026static void xs_udp_data_read_skb(struct rpc_xprt *xprt,
1027 struct sock *sk,
1028 struct sk_buff *skb)
a246b010 1029{
9903cd1c 1030 struct rpc_task *task;
a246b010 1031 struct rpc_rqst *rovr;
f9b2ee71 1032 int repsize, copied;
d8ed029d
AD		 1033 u32 _xid;
1034 __be32 *xp;
a246b010 1035
1da8c681 1036 repsize = skb->len;
a246b010 1037 if (repsize < 4) {
46121cf7 1038 dprintk("RPC: impossible RPC reply size %d!\n", repsize);
f9b2ee71 1039 return;
a246b010
CL		 1040 }
1041
1042 /* Copy the XID from the skb... */
1da8c681 1043 xp = skb_header_pointer(skb, 0, sizeof(_xid), &_xid);
a246b010 1044 if (xp == NULL)
f9b2ee71 1045 return;
a246b010
CL		 1046
1047 /* Look up and lock the request corresponding to the given XID */
f9b2ee71 1048 spin_lock_bh(&xprt->transport_lock);
a246b010
CL		 1049 rovr = xprt_lookup_rqst(xprt, *xp);
1050 if (!rovr)
1051 goto out_unlock;
1052 task = rovr->rq_task;
1053
a246b010
CL		 1054 if ((copied = rovr->rq_private_buf.buflen) > repsize)
1055 copied = repsize;
1056
1057 /* Suck it into the iovec, verify checksum if not done by hw. */
1781f7f5 1058 if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
02c22347 1059 __UDPX_INC_STATS(sk, UDP_MIB_INERRORS);
a246b010 1060 goto out_unlock;
1781f7f5
HX		 1061 }
1062
02c22347 1063 __UDPX_INC_STATS(sk, UDP_MIB_INDATAGRAMS);
a246b010 1064
6a24dfb6 1065 xprt_adjust_cwnd(xprt, task, copied);
1570c1e4 1066 xprt_complete_rqst(task, copied);
a246b010
CL		 1067
1068 out_unlock:
f9b2ee71
TM		 1069 spin_unlock_bh(&xprt->transport_lock);
1070}
1071
1072static void xs_udp_data_receive(struct sock_xprt *transport)
1073{
1074 struct sk_buff *skb;
1075 struct sock *sk;
1076 int err;
1077
1078 mutex_lock(&transport->recv_mutex);
1079 sk = transport->inet;
1080 if (sk == NULL)
1081 goto out;
1082 for (;;) {
7c13f97f 1083 skb = skb_recv_udp(sk, 0, 1, &err);
42d42a5b
TM		 1084 if (skb != NULL) {
1085 xs_udp_data_read_skb(&transport->xprt, sk, skb);
850cbadd 1086 consume_skb(skb);
42d42a5b
TM		 1087 continue;
1088 }
1089 if (!test_and_clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
f9b2ee71 1090 break;
f9b2ee71
TM		 1091 }
1092out:
1093 mutex_unlock(&transport->recv_mutex);
1094}
1095
1096static void xs_udp_data_receive_workfn(struct work_struct *work)
1097{
1098 struct sock_xprt *transport =
1099 container_of(work, struct sock_xprt, recv_worker);
1100 xs_udp_data_receive(transport);
1101}
1102
1103/**
1104 * xs_data_ready - "data ready" callback for UDP sockets
1105 * @sk: socket with data to read
1106 *
1107 */
1108static void xs_data_ready(struct sock *sk)
1109{
1110 struct rpc_xprt *xprt;
1111
1112 read_lock_bh(&sk->sk_callback_lock);
1113 dprintk("RPC: xs_data_ready...\n");
1114 xprt = xprt_from_sock(sk);
1115 if (xprt != NULL) {
1116 struct sock_xprt *transport = container_of(xprt,
1117 struct sock_xprt, xprt);
5157b956
TM		 1118 transport->old_data_ready(sk);
1119 /* Any data means we had a useful conversation, so
1120 * then we don't need to delay the next reconnect
1121 */
1122 if (xprt->reestablish_timeout)
1123 xprt->reestablish_timeout = 0;
42d42a5b 1124 if (!test_and_set_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
40a5f1b1 1125 queue_work(xprtiod_workqueue, &transport->recv_worker);
f9b2ee71 1126 }
f064af1e 1127 read_unlock_bh(&sk->sk_callback_lock);
a246b010
CL		 1128}
1129
a519fc7a
TM		 1130/*
1131 * Helper function to force a TCP close if the server is sending
1132 * junk and/or it has put us in CLOSE_WAIT
1133 */
1134static void xs_tcp_force_close(struct rpc_xprt *xprt)
1135{
a519fc7a
TM		 1136 xprt_force_disconnect(xprt);
1137}
1138
dd456471 1139static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
a246b010 1140{
51971139 1141 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
a246b010
CL		 1142 size_t len, used;
1143 char *p;
1144
51971139
CL		 1145 p = ((char *) &transport->tcp_fraghdr) + transport->tcp_offset;
1146 len = sizeof(transport->tcp_fraghdr) - transport->tcp_offset;
9d292316 1147 used = xdr_skb_read_bits(desc, p, len);
51971139 1148 transport->tcp_offset += used;
a246b010
CL		 1149 if (used != len)
1150 return;
808012fb 1151
51971139
CL		 1152 transport->tcp_reclen = ntohl(transport->tcp_fraghdr);
1153 if (transport->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
e136d092 1154 transport->tcp_flags |= TCP_RCV_LAST_FRAG;
a246b010 1155 else
e136d092 1156 transport->tcp_flags &= ~TCP_RCV_LAST_FRAG;
51971139 1157 transport->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
808012fb 1158
e136d092 1159 transport->tcp_flags &= ~TCP_RCV_COPY_FRAGHDR;
51971139 1160 transport->tcp_offset = 0;
808012fb 1161
a246b010 1162 /* Sanity check of the record length */
18dca02a 1163 if (unlikely(transport->tcp_reclen < 8)) {
46121cf7 1164 dprintk("RPC: invalid TCP record fragment length\n");
a519fc7a 1165 xs_tcp_force_close(xprt);
9903cd1c 1166 return;
a246b010 1167 }
46121cf7 1168 dprintk("RPC: reading TCP record fragment of length %d\n",
51971139 1169 transport->tcp_reclen);
a246b010
CL		 1170}
1171
51971139 1172static void xs_tcp_check_fraghdr(struct sock_xprt *transport)
a246b010 1173{
51971139 1174 if (transport->tcp_offset == transport->tcp_reclen) {
e136d092 1175 transport->tcp_flags |= TCP_RCV_COPY_FRAGHDR;
51971139 1176 transport->tcp_offset = 0;
e136d092
CL		 1177 if (transport->tcp_flags & TCP_RCV_LAST_FRAG) {
1178 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1179 transport->tcp_flags |= TCP_RCV_COPY_XID;
51971139 1180 transport->tcp_copied = 0;
a246b010
CL		 1181 }
1182 }
1183}
1184
dd456471 1185static inline void xs_tcp_read_xid(struct sock_xprt *transport, struct xdr_skb_reader *desc)
a246b010
CL		 1186{
1187 size_t len, used;
1188 char *p;
1189
51971139 1190 len = sizeof(transport->tcp_xid) - transport->tcp_offset;
46121cf7 1191 dprintk("RPC: reading XID (%Zu bytes)\n", len);
51971139 1192 p = ((char *) &transport->tcp_xid) + transport->tcp_offset;
9d292316 1193 used = xdr_skb_read_bits(desc, p, len);
51971139 1194 transport->tcp_offset += used;
a246b010
CL		 1195 if (used != len)
1196 return;
e136d092 1197 transport->tcp_flags &= ~TCP_RCV_COPY_XID;
f4a2e418 1198 transport->tcp_flags |= TCP_RCV_READ_CALLDIR;
51971139 1199 transport->tcp_copied = 4;
18dca02a
RL		 1200 dprintk("RPC: reading %s XID %08x\n",
1201 (transport->tcp_flags & TCP_RPC_REPLY) ? "reply for"
1202 : "request with",
51971139
CL		 1203 ntohl(transport->tcp_xid));
1204 xs_tcp_check_fraghdr(transport);
a246b010
CL		 1205}
1206
18dca02a
RL		 1207static inline void xs_tcp_read_calldir(struct sock_xprt *transport,
1208 struct xdr_skb_reader *desc)
a246b010 1209{
18dca02a
RL		 1210 size_t len, used;
1211 u32 offset;
b76ce561 1212 char *p;
18dca02a
RL		 1213
1214 /*
1215 * We want transport->tcp_offset to be 8 at the end of this routine
1216 * (4 bytes for the xid and 4 bytes for the call/reply flag).
1217 * When this function is called for the first time,
1218 * transport->tcp_offset is 4 (after having already read the xid).
1219 */
1220 offset = transport->tcp_offset - sizeof(transport->tcp_xid);
b76ce561 1221 len = sizeof(transport->tcp_calldir) - offset;
18dca02a 1222 dprintk("RPC: reading CALL/REPLY flag (%Zu bytes)\n", len);
b76ce561
TM		 1223 p = ((char *) &transport->tcp_calldir) + offset;
1224 used = xdr_skb_read_bits(desc, p, len);
18dca02a
RL		 1225 transport->tcp_offset += used;
1226 if (used != len)
1227 return;
f4a2e418 1228 transport->tcp_flags &= ~TCP_RCV_READ_CALLDIR;
f4a2e418
RL		 1229 /*
1230 * We don't yet have the XDR buffer, so we will write the calldir
1231 * out after we get the buffer from the 'struct rpc_rqst'
1232 */
b76ce561
TM		 1233 switch (ntohl(transport->tcp_calldir)) {
1234 case RPC_REPLY:
1235 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1236 transport->tcp_flags |= TCP_RCV_COPY_DATA;
18dca02a 1237 transport->tcp_flags |= TCP_RPC_REPLY;
b76ce561
TM		 1238 break;
1239 case RPC_CALL:
1240 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1241 transport->tcp_flags |= TCP_RCV_COPY_DATA;
18dca02a 1242 transport->tcp_flags &= ~TCP_RPC_REPLY;
b76ce561
TM		 1243 break;
1244 default:
1245 dprintk("RPC: invalid request message type\n");
a519fc7a 1246 xs_tcp_force_close(&transport->xprt);
b76ce561 1247 }
18dca02a
RL		 1248 xs_tcp_check_fraghdr(transport);
1249}
1250
44b98efd
RL		 1251static inline void xs_tcp_read_common(struct rpc_xprt *xprt,
1252 struct xdr_skb_reader *desc,
1253 struct rpc_rqst *req)
a246b010 1254{
44b98efd
RL		 1255 struct sock_xprt *transport =
1256 container_of(xprt, struct sock_xprt, xprt);
a246b010
CL		 1257 struct xdr_buf *rcvbuf;
1258 size_t len;
1259 ssize_t r;
1260
a246b010 1261 rcvbuf = &req->rq_private_buf;
f4a2e418
RL		 1262
1263 if (transport->tcp_flags & TCP_RCV_COPY_CALLDIR) {
1264 /*
1265 * Save the RPC direction in the XDR buffer
1266 */
f4a2e418 1267 memcpy(rcvbuf->head[0].iov_base + transport->tcp_copied,
b76ce561
TM		 1268 &transport->tcp_calldir,
1269 sizeof(transport->tcp_calldir));
1270 transport->tcp_copied += sizeof(transport->tcp_calldir);
f4a2e418 1271 transport->tcp_flags &= ~TCP_RCV_COPY_CALLDIR;
a246b010
CL		 1272 }
1273
a246b010 1274 len = desc->count;
51971139 1275 if (len > transport->tcp_reclen - transport->tcp_offset) {
dd456471 1276 struct xdr_skb_reader my_desc;
a246b010 1277
51971139 1278 len = transport->tcp_reclen - transport->tcp_offset;
a246b010
CL		 1279 memcpy(&my_desc, desc, sizeof(my_desc));
1280 my_desc.count = len;
51971139 1281 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
9d292316 1282 &my_desc, xdr_skb_read_bits);
a246b010
CL		 1283 desc->count -= r;
1284 desc->offset += r;
1285 } else
51971139 1286 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
9d292316 1287 desc, xdr_skb_read_bits);
a246b010
CL		 1288
1289 if (r > 0) {
51971139
CL		 1290 transport->tcp_copied += r;
1291 transport->tcp_offset += r;
a246b010
CL		 1292 }
1293 if (r != len) {
1294 /* Error when copying to the receive buffer,
1295 * usually because we weren't able to allocate
1296 * additional buffer pages. All we can do now
e136d092 1297 * is turn off TCP_RCV_COPY_DATA, so the request
a246b010
CL		 1298 * will not receive any additional updates,
1299 * and time out.
1300 * Any remaining data from this record will
1301 * be discarded.
1302 */
e136d092 1303 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
46121cf7 1304 dprintk("RPC: XID %08x truncated request\n",
51971139 1305 ntohl(transport->tcp_xid));
46121cf7
CL		 1306 dprintk("RPC: xprt = %p, tcp_copied = %lu, "
1307 "tcp_offset = %u, tcp_reclen = %u\n",
1308 xprt, transport->tcp_copied,
1309 transport->tcp_offset, transport->tcp_reclen);
44b98efd 1310 return;
a246b010
CL		 1311 }
1312
46121cf7 1313 dprintk("RPC: XID %08x read %Zd bytes\n",
51971139 1314 ntohl(transport->tcp_xid), r);
46121cf7
CL		 1315 dprintk("RPC: xprt = %p, tcp_copied = %lu, tcp_offset = %u, "
1316 "tcp_reclen = %u\n", xprt, transport->tcp_copied,
1317 transport->tcp_offset, transport->tcp_reclen);
51971139
CL		 1318
1319 if (transport->tcp_copied == req->rq_private_buf.buflen)
e136d092 1320 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
51971139 1321 else if (transport->tcp_offset == transport->tcp_reclen) {
e136d092
CL		 1322 if (transport->tcp_flags & TCP_RCV_LAST_FRAG)
1323 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
a246b010 1324 }
44b98efd
RL		 1325}
1326
1327/*
1328 * Finds the request corresponding to the RPC xid and invokes the common
1329 * tcp read code to read the data.
1330 */
1331static inline int xs_tcp_read_reply(struct rpc_xprt *xprt,
1332 struct xdr_skb_reader *desc)
1333{
1334 struct sock_xprt *transport =
1335 container_of(xprt, struct sock_xprt, xprt);
1336 struct rpc_rqst *req;
1337
1338 dprintk("RPC: read reply XID %08x\n", ntohl(transport->tcp_xid));
1339
1340 /* Find and lock the request corresponding to this xid */
edc1b01c 1341 spin_lock_bh(&xprt->transport_lock);
44b98efd
RL		 1342 req = xprt_lookup_rqst(xprt, transport->tcp_xid);
1343 if (!req) {
1344 dprintk("RPC: XID %08x request not found!\n",
1345 ntohl(transport->tcp_xid));
edc1b01c 1346 spin_unlock_bh(&xprt->transport_lock);
44b98efd
RL		 1347 return -1;
1348 }
1349
1350 xs_tcp_read_common(xprt, desc, req);
1351
e136d092 1352 if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
51971139 1353 xprt_complete_rqst(req->rq_task, transport->tcp_copied);
44b98efd 1354
edc1b01c 1355 spin_unlock_bh(&xprt->transport_lock);
44b98efd
RL		 1356 return 0;
1357}
1358
9e00abc3 1359#if defined(CONFIG_SUNRPC_BACKCHANNEL)
44b98efd
RL		 1360/*
1361 * Obtains an rpc_rqst previously allocated and invokes the common
1362 * tcp read code to read the data. The result is placed in the callback
1363 * queue.
1364 * If we're unable to obtain the rpc_rqst we schedule the closing of the
1365 * connection and return -1.
1366 */
2ea24497 1367static int xs_tcp_read_callback(struct rpc_xprt *xprt,
44b98efd
RL		 1368 struct xdr_skb_reader *desc)
1369{
1370 struct sock_xprt *transport =
1371 container_of(xprt, struct sock_xprt, xprt);
1372 struct rpc_rqst *req;
1373
2ea24497 1374 /* Look up and lock the request corresponding to the given XID */
edc1b01c 1375 spin_lock_bh(&xprt->transport_lock);
2ea24497 1376 req = xprt_lookup_bc_request(xprt, transport->tcp_xid);
44b98efd 1377 if (req == NULL) {
edc1b01c 1378 spin_unlock_bh(&xprt->transport_lock);
44b98efd
RL		 1379 printk(KERN_WARNING "Callback slot table overflowed\n");
1380 xprt_force_disconnect(xprt);
1381 return -1;
1382 }
1383
44b98efd
RL		 1384 dprintk("RPC: read callback XID %08x\n", ntohl(req->rq_xid));
1385 xs_tcp_read_common(xprt, desc, req);
1386
2ea24497
TM		 1387 if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1388 xprt_complete_bc_request(req, transport->tcp_copied);
edc1b01c 1389 spin_unlock_bh(&xprt->transport_lock);
44b98efd
RL		 1390
1391 return 0;
1392}
1393
1394static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1395 struct xdr_skb_reader *desc)
1396{
1397 struct sock_xprt *transport =
1398 container_of(xprt, struct sock_xprt, xprt);
1399
1400 return (transport->tcp_flags & TCP_RPC_REPLY) ?
1401 xs_tcp_read_reply(xprt, desc) :
1402 xs_tcp_read_callback(xprt, desc);
1403}
76566773
CL		 1404
1405static int xs_tcp_bc_up(struct svc_serv *serv, struct net *net)
1406{
1407 int ret;
1408
1409 ret = svc_create_xprt(serv, "tcp-bc", net, PF_INET, 0,
1410 SVC_SOCK_ANONYMOUS);
1411 if (ret < 0)
1412 return ret;
1413 return 0;
1414}
6b26cc8c
CL		 1415
1416static size_t xs_tcp_bc_maxpayload(struct rpc_xprt *xprt)
1417{
1418 return PAGE_SIZE;
1419}
44b98efd
RL		 1420#else
1421static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1422 struct xdr_skb_reader *desc)
1423{
1424 return xs_tcp_read_reply(xprt, desc);
1425}
9e00abc3 1426#endif /* CONFIG_SUNRPC_BACKCHANNEL */
44b98efd
RL		 1427
1428/*
1429 * Read data off the transport. This can be either an RPC_CALL or an
1430 * RPC_REPLY. Relay the processing to helper functions.
1431 */
1432static void xs_tcp_read_data(struct rpc_xprt *xprt,
1433 struct xdr_skb_reader *desc)
1434{
1435 struct sock_xprt *transport =
1436 container_of(xprt, struct sock_xprt, xprt);
1437
1438 if (_xs_tcp_read_data(xprt, desc) == 0)
1439 xs_tcp_check_fraghdr(transport);
1440 else {
1441 /*
1442 * The transport_lock protects the request handling.
1443 * There's no need to hold it to update the tcp_flags.
1444 */
1445 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1446 }
a246b010
CL		 1447}
1448
dd456471 1449static inline void xs_tcp_read_discard(struct sock_xprt *transport, struct xdr_skb_reader *desc)
a246b010
CL		 1450{
1451 size_t len;
1452
51971139 1453 len = transport->tcp_reclen - transport->tcp_offset;
a246b010
CL		 1454 if (len > desc->count)
1455 len = desc->count;
1456 desc->count -= len;
1457 desc->offset += len;
51971139 1458 transport->tcp_offset += len;
46121cf7 1459 dprintk("RPC: discarded %Zu bytes\n", len);
51971139 1460 xs_tcp_check_fraghdr(transport);
a246b010
CL		 1461}
1462
9903cd1c 1463static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
a246b010
CL		 1464{
1465 struct rpc_xprt *xprt = rd_desc->arg.data;
51971139 1466 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
dd456471 1467 struct xdr_skb_reader desc = {
a246b010
CL		 1468 .skb = skb,
1469 .offset = offset,
1470 .count = len,
9903cd1c 1471 };
a246b010 1472
46121cf7 1473 dprintk("RPC: xs_tcp_data_recv started\n");
a246b010 1474 do {
1a867a08 1475 trace_xs_tcp_data_recv(transport);
a246b010
CL		 1476 /* Read in a new fragment marker if necessary */
1477 /* Can we ever really expect to get completely empty fragments? */
e136d092 1478 if (transport->tcp_flags & TCP_RCV_COPY_FRAGHDR) {
9903cd1c 1479 xs_tcp_read_fraghdr(xprt, &desc);
a246b010
CL		 1480 continue;
1481 }
1482 /* Read in the xid if necessary */
e136d092 1483 if (transport->tcp_flags & TCP_RCV_COPY_XID) {
51971139 1484 xs_tcp_read_xid(transport, &desc);
a246b010
CL		 1485 continue;
1486 }
18dca02a 1487 /* Read in the call/reply flag */
f4a2e418 1488 if (transport->tcp_flags & TCP_RCV_READ_CALLDIR) {
18dca02a
RL		 1489 xs_tcp_read_calldir(transport, &desc);
1490 continue;
1491 }
a246b010 1492 /* Read in the request data */
e136d092 1493 if (transport->tcp_flags & TCP_RCV_COPY_DATA) {
44b98efd 1494 xs_tcp_read_data(xprt, &desc);
a246b010
CL		 1495 continue;
1496 }
1497 /* Skip over any trailing bytes on short reads */
51971139 1498 xs_tcp_read_discard(transport, &desc);
a246b010 1499 } while (desc.count);
1a867a08 1500 trace_xs_tcp_data_recv(transport);
46121cf7 1501 dprintk("RPC: xs_tcp_data_recv done\n");
a246b010
CL		 1502 return len - desc.count;
1503}
1504
66d7a56a
TM		 1505static void xs_tcp_data_receive(struct sock_xprt *transport)
1506{
1507 struct rpc_xprt *xprt = &transport->xprt;
1508 struct sock *sk;
1509 read_descriptor_t rd_desc = {
1510 .count = 2*1024*1024,
1511 .arg.data = xprt,
1512 };
1513 unsigned long total = 0;
1514 int read = 0;
1515
edc1b01c 1516 mutex_lock(&transport->recv_mutex);
66d7a56a 1517 sk = transport->inet;
edc1b01c
TM		 1518 if (sk == NULL)
1519 goto out;
66d7a56a
TM		 1520
1521 /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
1522 for (;;) {
edc1b01c 1523 lock_sock(sk);
66d7a56a 1524 read = tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
42d42a5b
TM		 1525 if (read <= 0) {
1526 clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state);
1527 release_sock(sk);
1528 if (!test_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
1529 break;
1530 } else {
1531 release_sock(sk);
1532 total += read;
1533 }
66d7a56a
TM		 1534 rd_desc.count = 65536;
1535 }
edc1b01c
TM		 1536out:
1537 mutex_unlock(&transport->recv_mutex);
66d7a56a
TM		 1538 trace_xs_tcp_data_ready(xprt, read, total);
1539}
1540
edc1b01c
TM		 1541static void xs_tcp_data_receive_workfn(struct work_struct *work)
1542{
1543 struct sock_xprt *transport =
1544 container_of(work, struct sock_xprt, recv_worker);
1545 xs_tcp_data_receive(transport);
1546}
1547
9903cd1c
CL		 1548/**
1549 * xs_tcp_state_change - callback to handle TCP socket state changes
1550 * @sk: socket whose state has changed
1551 *
1552 */
1553static void xs_tcp_state_change(struct sock *sk)
a246b010 1554{
9903cd1c 1555 struct rpc_xprt *xprt;
0fdea1e8 1556 struct sock_xprt *transport;
a246b010 1557
f064af1e 1558 read_lock_bh(&sk->sk_callback_lock);
a246b010
CL		 1559 if (!(xprt = xprt_from_sock(sk)))
1560 goto out;
46121cf7 1561 dprintk("RPC: xs_tcp_state_change client %p...\n", xprt);
669502ff 1562 dprintk("RPC: state %x conn %d dead %d zapped %d sk_shutdown %d\n",
46121cf7
CL		 1563 sk->sk_state, xprt_connected(xprt),
1564 sock_flag(sk, SOCK_DEAD),
669502ff
AC		 1565 sock_flag(sk, SOCK_ZAPPED),
1566 sk->sk_shutdown);
a246b010 1567
0fdea1e8 1568 transport = container_of(xprt, struct sock_xprt, xprt);
40b5ea0c 1569 trace_rpc_socket_state_change(xprt, sk->sk_socket);
a246b010
CL		 1570 switch (sk->sk_state) {
1571 case TCP_ESTABLISHED:
f064af1e 1572 spin_lock(&xprt->transport_lock);
a246b010 1573 if (!xprt_test_and_set_connected(xprt)) {
51971139 1574
a246b010 1575 /* Reset TCP record info */
51971139
CL		 1576 transport->tcp_offset = 0;
1577 transport->tcp_reclen = 0;
1578 transport->tcp_copied = 0;
e136d092
CL		 1579 transport->tcp_flags =
1580 TCP_RCV_COPY_FRAGHDR | TCP_RCV_COPY_XID;
8b71798c 1581 xprt->connect_cookie++;
0fdea1e8
TM		 1582 clear_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
1583 xprt_clear_connecting(xprt);
51971139 1584
2a491991 1585 xprt_wake_pending_tasks(xprt, -EAGAIN);
a246b010 1586 }
f064af1e 1587 spin_unlock(&xprt->transport_lock);
a246b010 1588 break;
3b948ae5
TM		 1589 case TCP_FIN_WAIT1:
1590 /* The client initiated a shutdown of the socket */
7c1d71cf 1591 xprt->connect_cookie++;
663b8858 1592 xprt->reestablish_timeout = 0;
3b948ae5 1593 set_bit(XPRT_CLOSING, &xprt->state);
4e857c58 1594 smp_mb__before_atomic();
3b948ae5 1595 clear_bit(XPRT_CONNECTED, &xprt->state);
ef803670 1596 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
4e857c58 1597 smp_mb__after_atomic();
a246b010 1598 break;
632e3bdc 1599 case TCP_CLOSE_WAIT:
3b948ae5 1600 /* The server initiated a shutdown of the socket */
7c1d71cf 1601 xprt->connect_cookie++;
d0bea455 1602 clear_bit(XPRT_CONNECTED, &xprt->state);
a519fc7a 1603 xs_tcp_force_close(xprt);
663b8858
TM		 1604 case TCP_CLOSING:
1605 /*
1606 * If the server closed down the connection, make sure that
1607 * we back off before reconnecting
1608 */
1609 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1610 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
3b948ae5
TM		 1611 break;
1612 case TCP_LAST_ACK:
670f9457 1613 set_bit(XPRT_CLOSING, &xprt->state);
4e857c58 1614 smp_mb__before_atomic();
3b948ae5 1615 clear_bit(XPRT_CONNECTED, &xprt->state);
4e857c58 1616 smp_mb__after_atomic();
3b948ae5
TM		 1617 break;
1618 case TCP_CLOSE:
0fdea1e8
TM		 1619 if (test_and_clear_bit(XPRT_SOCK_CONNECTING,
1620 &transport->sock_state))
1621 xprt_clear_connecting(xprt);
7d1e8255 1622 xs_sock_mark_closed(xprt);
a246b010
CL		 1623 }
1624 out:
f064af1e 1625 read_unlock_bh(&sk->sk_callback_lock);
a246b010
CL		 1626}
1627
1f0fa154
IJ		 1628static void xs_write_space(struct sock *sk)
1629{
13331a55 1630 struct socket_wq *wq;
1f0fa154
IJ		 1631 struct rpc_xprt *xprt;
1632
13331a55 1633 if (!sk->sk_socket)
1f0fa154 1634 return;
13331a55 1635 clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1f0fa154
IJ		 1636
1637 if (unlikely(!(xprt = xprt_from_sock(sk))))
1638 return;
13331a55
TM		 1639 rcu_read_lock();
1640 wq = rcu_dereference(sk->sk_wq);
1641 if (!wq || test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &wq->flags) == 0)
1642 goto out;
1f0fa154
IJ		 1643
1644 xprt_write_space(xprt);
13331a55
TM		 1645out:
1646 rcu_read_unlock();
1f0fa154
IJ		 1647}
1648
9903cd1c 1649/**
c7b2cae8
CL		 1650 * xs_udp_write_space - callback invoked when socket buffer space
1651 * becomes available
9903cd1c
CL		 1652 * @sk: socket whose state has changed
1653 *
a246b010
CL		 1654 * Called when more output buffer space is available for this socket.
1655 * We try not to wake our writers until they can make "significant"
c7b2cae8 1656 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
a246b010
CL		 1657 * with a bunch of small requests.
1658 */
c7b2cae8 1659static void xs_udp_write_space(struct sock *sk)
a246b010 1660{
f064af1e 1661 read_lock_bh(&sk->sk_callback_lock);
a246b010 1662
c7b2cae8 1663 /* from net/core/sock.c:sock_def_write_space */
1f0fa154
IJ		 1664 if (sock_writeable(sk))
1665 xs_write_space(sk);
a246b010 1666
f064af1e 1667 read_unlock_bh(&sk->sk_callback_lock);
c7b2cae8 1668}
a246b010 1669
c7b2cae8
CL		 1670/**
1671 * xs_tcp_write_space - callback invoked when socket buffer space
1672 * becomes available
1673 * @sk: socket whose state has changed
1674 *
1675 * Called when more output buffer space is available for this socket.
1676 * We try not to wake our writers until they can make "significant"
1677 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1678 * with a bunch of small requests.
1679 */
1680static void xs_tcp_write_space(struct sock *sk)
1681{
f064af1e 1682 read_lock_bh(&sk->sk_callback_lock);
c7b2cae8
CL		 1683
1684 /* from net/core/stream.c:sk_stream_write_space */
64dc6130 1685 if (sk_stream_is_writeable(sk))
1f0fa154 1686 xs_write_space(sk);
b6ddf64f 1687
f064af1e 1688 read_unlock_bh(&sk->sk_callback_lock);
a246b010
CL		 1689}
1690
470056c2 1691static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
a246b010 1692{
ee0ac0c2
CL		 1693 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1694 struct sock *sk = transport->inet;
a246b010 1695
7c6e066e 1696 if (transport->rcvsize) {
a246b010 1697 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
7c6e066e 1698 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
a246b010 1699 }
7c6e066e 1700 if (transport->sndsize) {
a246b010 1701 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
7c6e066e 1702 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
a246b010
CL		 1703 sk->sk_write_space(sk);
1704 }
1705}
1706
43118c29 1707/**
470056c2 1708 * xs_udp_set_buffer_size - set send and receive limits
43118c29 1709 * @xprt: generic transport
470056c2
CL		 1710 * @sndsize: requested size of send buffer, in bytes
1711 * @rcvsize: requested size of receive buffer, in bytes
43118c29 1712 *
470056c2 1713 * Set socket send and receive buffer size limits.
43118c29 1714 */
470056c2 1715static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
43118c29 1716{
7c6e066e
CL		 1717 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1718
1719 transport->sndsize = 0;
470056c2 1720 if (sndsize)
7c6e066e
CL		 1721 transport->sndsize = sndsize + 1024;
1722 transport->rcvsize = 0;
470056c2 1723 if (rcvsize)
7c6e066e 1724 transport->rcvsize = rcvsize + 1024;
470056c2
CL		 1725
1726 xs_udp_do_set_buffer_size(xprt);
43118c29
CL		 1727}
1728
46c0ee8b
CL		 1729/**
1730 * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1731 * @task: task that timed out
1732 *
1733 * Adjust the congestion window after a retransmit timeout has occurred.
1734 */
6a24dfb6 1735static void xs_udp_timer(struct rpc_xprt *xprt, struct rpc_task *task)
46c0ee8b 1736{
6a24dfb6 1737 xprt_adjust_cwnd(xprt, task, -ETIMEDOUT);
46c0ee8b
CL		 1738}
1739
b85d8806
CL		 1740static unsigned short xs_get_random_port(void)
1741{
5d71899a 1742 unsigned short range = xprt_max_resvport - xprt_min_resvport + 1;
63862b5b 1743 unsigned short rand = (unsigned short) prandom_u32() % range;
b85d8806
CL		 1744 return rand + xprt_min_resvport;
1745}
1746
4dda9c8a
TM		 1747/**
1748 * xs_set_reuseaddr_port - set the socket's port and address reuse options
1749 * @sock: socket
1750 *
1751 * Note that this function has to be called on all sockets that share the
1752 * same port, and it must be called before binding.
1753 */
1754static void xs_sock_set_reuseport(struct socket *sock)
1755{
402e23b4 1756 int opt = 1;
4dda9c8a 1757
402e23b4
TM		 1758 kernel_setsockopt(sock, SOL_SOCKET, SO_REUSEPORT,
1759 (char *)&opt, sizeof(opt));
4dda9c8a
TM		 1760}
1761
1762static unsigned short xs_sock_getport(struct socket *sock)
1763{
1764 struct sockaddr_storage buf;
1765 int buflen;
1766 unsigned short port = 0;
1767
1768 if (kernel_getsockname(sock, (struct sockaddr *)&buf, &buflen) < 0)
1769 goto out;
1770 switch (buf.ss_family) {
1771 case AF_INET6:
1772 port = ntohs(((struct sockaddr_in6 *)&buf)->sin6_port);
1773 break;
1774 case AF_INET:
1775 port = ntohs(((struct sockaddr_in *)&buf)->sin_port);
1776 }
1777out:
1778 return port;
1779}
1780
92200412
CL		 1781/**
1782 * xs_set_port - reset the port number in the remote endpoint address
1783 * @xprt: generic transport
1784 * @port: new port number
1785 *
1786 */
1787static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1788{
46121cf7 1789 dprintk("RPC: setting port for xprt %p to %u\n", xprt, port);
c4efcb1d 1790
9dc3b095
CL		 1791 rpc_set_port(xs_addr(xprt), port);
1792 xs_update_peer_port(xprt);
92200412
CL		 1793}
1794
4dda9c8a
TM		 1795static void xs_set_srcport(struct sock_xprt *transport, struct socket *sock)
1796{
1797 if (transport->srcport == 0)
1798 transport->srcport = xs_sock_getport(sock);
1799}
1800
5d4ec932 1801static unsigned short xs_get_srcport(struct sock_xprt *transport)
67a391d7 1802{
fbfffbd5 1803 unsigned short port = transport->srcport;
67a391d7
TM		 1804
1805 if (port == 0 && transport->xprt.resvport)
1806 port = xs_get_random_port();
1807 return port;
1808}
1809
baaf4e48 1810static unsigned short xs_next_srcport(struct sock_xprt *transport, unsigned short port)
67a391d7 1811{
fbfffbd5
CL		 1812 if (transport->srcport != 0)
1813 transport->srcport = 0;
67a391d7
TM		 1814 if (!transport->xprt.resvport)
1815 return 0;
1816 if (port <= xprt_min_resvport || port > xprt_max_resvport)
1817 return xprt_max_resvport;
1818 return --port;
1819}
beb59b68 1820static int xs_bind(struct sock_xprt *transport, struct socket *sock)
a246b010 1821{
beb59b68 1822 struct sockaddr_storage myaddr;
67a391d7 1823 int err, nloop = 0;
5d4ec932 1824 unsigned short port = xs_get_srcport(transport);
67a391d7 1825 unsigned short last;
a246b010 1826
0f7a622c
CP		 1827 /*
1828 * If we are asking for any ephemeral port (i.e. port == 0 &&
1829 * transport->xprt.resvport == 0), don't bind. Let the local
1830 * port selection happen implicitly when the socket is used
1831 * (for example at connect time).
1832 *
1833 * This ensures that we can continue to establish TCP
1834 * connections even when all local ephemeral ports are already
1835 * a part of some TCP connection. This makes no difference
1836 * for UDP sockets, but also doens't harm them.
1837 *
1838 * If we're asking for any reserved port (i.e. port == 0 &&
1839 * transport->xprt.resvport == 1) xs_get_srcport above will
1840 * ensure that port is non-zero and we will bind as needed.
1841 */
1842 if (port == 0)
1843 return 0;
1844
beb59b68 1845 memcpy(&myaddr, &transport->srcaddr, transport->xprt.addrlen);
a246b010 1846 do {
beb59b68
PE		 1847 rpc_set_port((struct sockaddr *)&myaddr, port);
1848 err = kernel_bind(sock, (struct sockaddr *)&myaddr,
1849 transport->xprt.addrlen);
a246b010 1850 if (err == 0) {
fbfffbd5 1851 transport->srcport = port;
d3bc9a1d 1852 break;
a246b010 1853 }
67a391d7 1854 last = port;
baaf4e48 1855 port = xs_next_srcport(transport, port);
67a391d7
TM		 1856 if (port > last)
1857 nloop++;
1858 } while (err == -EADDRINUSE && nloop != 2);
90058d37 1859
4232e863 1860 if (myaddr.ss_family == AF_INET)
beb59b68
PE		 1861 dprintk("RPC: %s %pI4:%u: %s (%d)\n", __func__,
1862 &((struct sockaddr_in *)&myaddr)->sin_addr,
1863 port, err ? "failed" : "ok", err);
1864 else
1865 dprintk("RPC: %s %pI6:%u: %s (%d)\n", __func__,
1866 &((struct sockaddr_in6 *)&myaddr)->sin6_addr,
1867 port, err ? "failed" : "ok", err);
a246b010
CL		 1868 return err;
1869}
1870
176e21ee
CL		 1871/*
1872 * We don't support autobind on AF_LOCAL sockets
1873 */
1874static void xs_local_rpcbind(struct rpc_task *task)
1875{
fb43d172 1876 xprt_set_bound(task->tk_xprt);
176e21ee
CL		 1877}
1878
1879static void xs_local_set_port(struct rpc_xprt *xprt, unsigned short port)
1880{
1881}
beb59b68 1882
ed07536e
PZ		 1883#ifdef CONFIG_DEBUG_LOCK_ALLOC
1884static struct lock_class_key xs_key[2];
1885static struct lock_class_key xs_slock_key[2];
1886
176e21ee
CL		 1887static inline void xs_reclassify_socketu(struct socket *sock)
1888{
1889 struct sock *sk = sock->sk;
1890
176e21ee
CL		 1891 sock_lock_init_class_and_name(sk, "slock-AF_LOCAL-RPC",
1892 &xs_slock_key[1], "sk_lock-AF_LOCAL-RPC", &xs_key[1]);
1893}
1894
8945ee5e 1895static inline void xs_reclassify_socket4(struct socket *sock)
ed07536e
PZ		 1896{
1897 struct sock *sk = sock->sk;
8945ee5e 1898
8945ee5e
CL		 1899 sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1900 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1901}
ed07536e 1902
8945ee5e
CL		 1903static inline void xs_reclassify_socket6(struct socket *sock)
1904{
1905 struct sock *sk = sock->sk;
ed07536e 1906
8945ee5e
CL		 1907 sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1908 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
ed07536e 1909}
6bc9638a
PE		 1910
1911static inline void xs_reclassify_socket(int family, struct socket *sock)
1912{
fafc4e1e 1913 if (WARN_ON_ONCE(!sock_allow_reclassification(sock->sk)))
1b7a1819
WAA		 1914 return;
1915
4232e863 1916 switch (family) {
176e21ee
CL		 1917 case AF_LOCAL:
1918 xs_reclassify_socketu(sock);
1919 break;
4232e863 1920 case AF_INET:
6bc9638a 1921 xs_reclassify_socket4(sock);
4232e863
CL		 1922 break;
1923 case AF_INET6:
6bc9638a 1924 xs_reclassify_socket6(sock);
4232e863
CL		 1925 break;
1926 }
6bc9638a 1927}
ed07536e 1928#else
6bc9638a
PE		 1929static inline void xs_reclassify_socket(int family, struct socket *sock)
1930{
1931}
ed07536e
PZ		 1932#endif
1933
93dc41bd
N		 1934static void xs_dummy_setup_socket(struct work_struct *work)
1935{
1936}
1937
6bc9638a 1938static struct socket *xs_create_sock(struct rpc_xprt *xprt,
4dda9c8a
TM		 1939 struct sock_xprt *transport, int family, int type,
1940 int protocol, bool reuseport)
22f79326
PE		 1941{
1942 struct socket *sock;
1943 int err;
1944
6bc9638a 1945 err = __sock_create(xprt->xprt_net, family, type, protocol, &sock, 1);
22f79326
PE		 1946 if (err < 0) {
1947 dprintk("RPC: can't create %d transport socket (%d).\n",
1948 protocol, -err);
1949 goto out;
1950 }
6bc9638a 1951 xs_reclassify_socket(family, sock);
22f79326 1952
4dda9c8a
TM		 1953 if (reuseport)
1954 xs_sock_set_reuseport(sock);
1955
4cea288a
BH		 1956 err = xs_bind(transport, sock);
1957 if (err) {
22f79326
PE		 1958 sock_release(sock);
1959 goto out;
1960 }
1961
1962 return sock;
1963out:
1964 return ERR_PTR(err);
1965}
1966
176e21ee
CL		 1967static int xs_local_finish_connecting(struct rpc_xprt *xprt,
1968 struct socket *sock)
1969{
1970 struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1971 xprt);
1972
1973 if (!transport->inet) {
1974 struct sock *sk = sock->sk;
1975
1976 write_lock_bh(&sk->sk_callback_lock);
1977
1978 xs_save_old_callbacks(transport, sk);
1979
1980 sk->sk_user_data = xprt;
a2648094 1981 sk->sk_data_ready = xs_data_ready;
176e21ee 1982 sk->sk_write_space = xs_udp_write_space;
b4411457 1983 sock_set_flag(sk, SOCK_FASYNC);
2118071d 1984 sk->sk_error_report = xs_error_report;
c2126157 1985 sk->sk_allocation = GFP_NOIO;
176e21ee
CL		 1986
1987 xprt_clear_connected(xprt);
1988
1989 /* Reset to new socket */
1990 transport->sock = sock;
1991 transport->inet = sk;
1992
1993 write_unlock_bh(&sk->sk_callback_lock);
1994 }
1995
1996 /* Tell the socket layer to start connecting... */
1997 xprt->stat.connect_count++;
1998 xprt->stat.connect_start = jiffies;
1999 return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, 0);
2000}
2001
2002/**
2003 * xs_local_setup_socket - create AF_LOCAL socket, connect to a local endpoint
176e21ee 2004 * @transport: socket transport to connect
176e21ee 2005 */
dc107402 2006static int xs_local_setup_socket(struct sock_xprt *transport)
176e21ee 2007{
176e21ee
CL		 2008 struct rpc_xprt *xprt = &transport->xprt;
2009 struct socket *sock;
2010 int status = -EIO;
2011
176e21ee
CL		 2012 status = __sock_create(xprt->xprt_net, AF_LOCAL,
2013 SOCK_STREAM, 0, &sock, 1);
2014 if (status < 0) {
2015 dprintk("RPC: can't create AF_LOCAL "
2016 "transport socket (%d).\n", -status);
2017 goto out;
2018 }
d1358917 2019 xs_reclassify_socket(AF_LOCAL, sock);
176e21ee
CL		 2020
2021 dprintk("RPC: worker connecting xprt %p via AF_LOCAL to %s\n",
2022 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2023
2024 status = xs_local_finish_connecting(xprt, sock);
40b5ea0c 2025 trace_rpc_socket_connect(xprt, sock, status);
176e21ee
CL		 2026 switch (status) {
2027 case 0:
2028 dprintk("RPC: xprt %p connected to %s\n",
2029 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2030 xprt_set_connected(xprt);
3601c4a9 2031 case -ENOBUFS:
176e21ee
CL		 2032 break;
2033 case -ENOENT:
2034 dprintk("RPC: xprt %p: socket %s does not exist\n",
2035 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2036 break;
4a20a988
TM		 2037 case -ECONNREFUSED:
2038 dprintk("RPC: xprt %p: connection refused for %s\n",
2039 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2040 break;
176e21ee
CL		 2041 default:
2042 printk(KERN_ERR "%s: unhandled error (%d) connecting to %s\n",
2043 __func__, -status,
2044 xprt->address_strings[RPC_DISPLAY_ADDR]);
2045 }
2046
2047out:
2048 xprt_clear_connecting(xprt);
2049 xprt_wake_pending_tasks(xprt, status);
dc107402
BF		 2050 return status;
2051}
2052
b6669737 2053static void xs_local_connect(struct rpc_xprt *xprt, struct rpc_task *task)
dc107402 2054{
dc107402
BF		 2055 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2056 int ret;
2057
2058 if (RPC_IS_ASYNC(task)) {
2059 /*
2060 * We want the AF_LOCAL connect to be resolved in the
2061 * filesystem namespace of the process making the rpc
2062 * call. Thus we connect synchronously.
2063 *
2064 * If we want to support asynchronous AF_LOCAL calls,
2065 * we'll need to figure out how to pass a namespace to
2066 * connect.
2067 */
2068 rpc_exit(task, -ENOTCONN);
2069 return;
2070 }
2071 ret = xs_local_setup_socket(transport);
2072 if (ret && !RPC_IS_SOFTCONN(task))
2073 msleep_interruptible(15000);
176e21ee
CL		 2074}
2075
3c87ef6e 2076#if IS_ENABLED(CONFIG_SUNRPC_SWAP)
d6e971d8
JL		 2077/*
2078 * Note that this should be called with XPRT_LOCKED held (or when we otherwise
2079 * know that we have exclusive access to the socket), to guard against
2080 * races with xs_reset_transport.
2081 */
a564b8f0
MG		 2082static void xs_set_memalloc(struct rpc_xprt *xprt)
2083{
2084 struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
2085 xprt);
2086
d6e971d8
JL		 2087 /*
2088 * If there's no sock, then we have nothing to set. The
2089 * reconnecting process will get it for us.
2090 */
2091 if (!transport->inet)
2092 return;
8e228133 2093 if (atomic_read(&xprt->swapper))
a564b8f0
MG		 2094 sk_set_memalloc(transport->inet);
2095}
2096
2097/**
d67fa4d8 2098 * xs_enable_swap - Tag this transport as being used for swap.
a564b8f0 2099 * @xprt: transport to tag
a564b8f0 2100 *
8e228133
JL		 2101 * Take a reference to this transport on behalf of the rpc_clnt, and
2102 * optionally mark it for swapping if it wasn't already.
a564b8f0 2103 */
d67fa4d8
JL		 2104static int
2105xs_enable_swap(struct rpc_xprt *xprt)
a564b8f0 2106{
d6e971d8 2107 struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
a564b8f0 2108
d6e971d8
JL		 2109 if (atomic_inc_return(&xprt->swapper) != 1)
2110 return 0;
2111 if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE))
2112 return -ERESTARTSYS;
2113 if (xs->inet)
2114 sk_set_memalloc(xs->inet);
2115 xprt_release_xprt(xprt, NULL);
8e228133
JL		 2116 return 0;
2117}
a564b8f0 2118
8e228133 2119/**
d67fa4d8 2120 * xs_disable_swap - Untag this transport as being used for swap.
8e228133
JL		 2121 * @xprt: transport to tag
2122 *
2123 * Drop a "swapper" reference to this xprt on behalf of the rpc_clnt. If the
2124 * swapper refcount goes to 0, untag the socket as a memalloc socket.
2125 */
d67fa4d8
JL		 2126static void
2127xs_disable_swap(struct rpc_xprt *xprt)
8e228133 2128{
d6e971d8 2129 struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
8e228133 2130
d6e971d8
JL		 2131 if (!atomic_dec_and_test(&xprt->swapper))
2132 return;
2133 if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE))
2134 return;
2135 if (xs->inet)
2136 sk_clear_memalloc(xs->inet);
2137 xprt_release_xprt(xprt, NULL);
a564b8f0 2138}
a564b8f0
MG		 2139#else
2140static void xs_set_memalloc(struct rpc_xprt *xprt)
2141{
2142}
d67fa4d8
JL		 2143
2144static int
2145xs_enable_swap(struct rpc_xprt *xprt)
2146{
2147 return -EINVAL;
2148}
2149
2150static void
2151xs_disable_swap(struct rpc_xprt *xprt)
2152{
2153}
a564b8f0
MG		 2154#endif
2155
16be2d20
CL		 2156static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2157{
2158 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2159
2160 if (!transport->inet) {
2161 struct sock *sk = sock->sk;
2162
2163 write_lock_bh(&sk->sk_callback_lock);
2164
2a9e1cfa
TM		 2165 xs_save_old_callbacks(transport, sk);
2166
16be2d20 2167 sk->sk_user_data = xprt;
f9b2ee71 2168 sk->sk_data_ready = xs_data_ready;
16be2d20 2169 sk->sk_write_space = xs_udp_write_space;
b4411457 2170 sock_set_flag(sk, SOCK_FASYNC);
c2126157 2171 sk->sk_allocation = GFP_NOIO;
16be2d20
CL		 2172
2173 xprt_set_connected(xprt);
2174
2175 /* Reset to new socket */
2176 transport->sock = sock;
2177 transport->inet = sk;
2178
a564b8f0
MG		 2179 xs_set_memalloc(xprt);
2180
16be2d20
CL		 2181 write_unlock_bh(&sk->sk_callback_lock);
2182 }
2183 xs_udp_do_set_buffer_size(xprt);
02910177
TM		 2184
2185 xprt->stat.connect_start = jiffies;
16be2d20
CL		 2186}
2187
8c14ff2a 2188static void xs_udp_setup_socket(struct work_struct *work)
a246b010 2189{
34161db6
TM		 2190 struct sock_xprt *transport =
2191 container_of(work, struct sock_xprt, connect_worker.work);
c8475461 2192 struct rpc_xprt *xprt = &transport->xprt;
ee0ac0c2 2193 struct socket *sock = transport->sock;
b65c0310 2194 int status = -EIO;
9903cd1c 2195
8c14ff2a 2196 sock = xs_create_sock(xprt, transport,
4dda9c8a
TM		 2197 xs_addr(xprt)->sa_family, SOCK_DGRAM,
2198 IPPROTO_UDP, false);
b65c0310 2199 if (IS_ERR(sock))
b0d93ad5 2200 goto out;
68e220bd 2201
c740eff8
CL		 2202 dprintk("RPC: worker connecting xprt %p via %s to "
2203 "%s (port %s)\n", xprt,
2204 xprt->address_strings[RPC_DISPLAY_PROTO],
2205 xprt->address_strings[RPC_DISPLAY_ADDR],
2206 xprt->address_strings[RPC_DISPLAY_PORT]);
68e220bd
CL		 2207
2208 xs_udp_finish_connecting(xprt, sock);
40b5ea0c 2209 trace_rpc_socket_connect(xprt, sock, 0);
b0d93ad5
CL		 2210 status = 0;
2211out:
718ba5b8 2212 xprt_unlock_connect(xprt, transport);
b0d93ad5 2213 xprt_clear_connecting(xprt);
7d1e8255 2214 xprt_wake_pending_tasks(xprt, status);
a246b010
CL		 2215}
2216
4876cc77
TM		 2217/**
2218 * xs_tcp_shutdown - gracefully shut down a TCP socket
2219 * @xprt: transport
2220 *
2221 * Initiates a graceful shutdown of the TCP socket by calling the
2222 * equivalent of shutdown(SHUT_RDWR);
2223 */
2224static void xs_tcp_shutdown(struct rpc_xprt *xprt)
2225{
2226 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2227 struct socket *sock = transport->sock;
2228
2229 if (sock == NULL)
2230 return;
2231 if (xprt_connected(xprt)) {
2232 kernel_sock_shutdown(sock, SHUT_RDWR);
2233 trace_rpc_socket_shutdown(xprt, sock);
2234 } else
2235 xs_reset_transport(transport);
2236}
2237
16be2d20 2238static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
a246b010 2239{
16be2d20 2240 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
fe19a96b 2241 int ret = -ENOTCONN;
edb267a6 2242
ee0ac0c2 2243 if (!transport->inet) {
b0d93ad5 2244 struct sock *sk = sock->sk;
7f260e85
TM		 2245 unsigned int keepidle = xprt->timeout->to_initval / HZ;
2246 unsigned int keepcnt = xprt->timeout->to_retries + 1;
2247 unsigned int opt_on = 1;
775f06ab 2248 unsigned int timeo;
d88e4d82 2249 unsigned int addr_pref = IPV6_PREFER_SRC_PUBLIC;
7f260e85
TM		 2250
2251 /* TCP Keepalive options */
2252 kernel_setsockopt(sock, SOL_SOCKET, SO_KEEPALIVE,
2253 (char *)&opt_on, sizeof(opt_on));
2254 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPIDLE,
2255 (char *)&keepidle, sizeof(keepidle));
2256 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPINTVL,
2257 (char *)&keepidle, sizeof(keepidle));
2258 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPCNT,
2259 (char *)&keepcnt, sizeof(keepcnt));
b0d93ad5 2260
d88e4d82
N		 2261 /* Avoid temporary address, they are bad for long-lived
2262 * connections such as NFS mounts.
2263 * RFC4941, section 3.6 suggests that:
2264 * Individual applications, which have specific
2265 * knowledge about the normal duration of connections,
2266 * MAY override this as appropriate.
2267 */
2268 kernel_setsockopt(sock, SOL_IPV6, IPV6_ADDR_PREFERENCES,
2269 (char *)&addr_pref, sizeof(addr_pref));
2270
775f06ab
TM		 2271 /* TCP user timeout (see RFC5482) */
2272 timeo = jiffies_to_msecs(xprt->timeout->to_initval) *
2273 (xprt->timeout->to_retries + 1);
2274 kernel_setsockopt(sock, SOL_TCP, TCP_USER_TIMEOUT,
2275 (char *)&timeo, sizeof(timeo));
2276
b0d93ad5
CL		 2277 write_lock_bh(&sk->sk_callback_lock);
2278
2a9e1cfa
TM		 2279 xs_save_old_callbacks(transport, sk);
2280
b0d93ad5 2281 sk->sk_user_data = xprt;
5157b956 2282 sk->sk_data_ready = xs_data_ready;
b0d93ad5
CL		 2283 sk->sk_state_change = xs_tcp_state_change;
2284 sk->sk_write_space = xs_tcp_write_space;
b4411457 2285 sock_set_flag(sk, SOCK_FASYNC);
2118071d 2286 sk->sk_error_report = xs_error_report;
c2126157 2287 sk->sk_allocation = GFP_NOIO;
3167e12c
CL		 2288
2289 /* socket options */
3167e12c 2290 sock_reset_flag(sk, SOCK_LINGER);
3167e12c 2291 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
b0d93ad5
CL		 2292
2293 xprt_clear_connected(xprt);
2294
2295 /* Reset to new socket */
ee0ac0c2
CL		 2296 transport->sock = sock;
2297 transport->inet = sk;
b0d93ad5
CL		 2298
2299 write_unlock_bh(&sk->sk_callback_lock);
2300 }
2301
01d37c42 2302 if (!xprt_bound(xprt))
fe19a96b 2303 goto out;
01d37c42 2304
a564b8f0
MG		 2305 xs_set_memalloc(xprt);
2306
b0d93ad5 2307 /* Tell the socket layer to start connecting... */
262ca07d
CL		 2308 xprt->stat.connect_count++;
2309 xprt->stat.connect_start = jiffies;
0fdea1e8 2310 set_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
fe19a96b
TM		 2311 ret = kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
2312 switch (ret) {
2313 case 0:
4dda9c8a 2314 xs_set_srcport(transport, sock);
fe19a96b
TM		 2315 case -EINPROGRESS:
2316 /* SYN_SENT! */
fe19a96b
TM		 2317 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2318 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1f4c17a0
TM		 2319 break;
2320 case -EADDRNOTAVAIL:
2321 /* Source port number is unavailable. Try a new one! */
2322 transport->srcport = 0;
fe19a96b
TM		 2323 }
2324out:
2325 return ret;
16be2d20
CL		 2326}
2327
9903cd1c 2328/**
b61d59ff 2329 * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint
9903cd1c
CL		 2330 *
2331 * Invoked by a work queue tasklet.
a246b010 2332 */
cdd518d5 2333static void xs_tcp_setup_socket(struct work_struct *work)
a246b010 2334{
cdd518d5
PE		 2335 struct sock_xprt *transport =
2336 container_of(work, struct sock_xprt, connect_worker.work);
ee0ac0c2 2337 struct socket *sock = transport->sock;
a9f5f0f7 2338 struct rpc_xprt *xprt = &transport->xprt;
b61d59ff 2339 int status = -EIO;
a246b010 2340
ee0ac0c2 2341 if (!sock) {
cdd518d5 2342 sock = xs_create_sock(xprt, transport,
4dda9c8a
TM		 2343 xs_addr(xprt)->sa_family, SOCK_STREAM,
2344 IPPROTO_TCP, true);
b61d59ff
TM		 2345 if (IS_ERR(sock)) {
2346 status = PTR_ERR(sock);
3167e12c
CL		 2347 goto out;
2348 }
7d1e8255 2349 }
a246b010 2350
c740eff8
CL		 2351 dprintk("RPC: worker connecting xprt %p via %s to "
2352 "%s (port %s)\n", xprt,
2353 xprt->address_strings[RPC_DISPLAY_PROTO],
2354 xprt->address_strings[RPC_DISPLAY_ADDR],
2355 xprt->address_strings[RPC_DISPLAY_PORT]);
edb267a6 2356
16be2d20 2357 status = xs_tcp_finish_connecting(xprt, sock);
40b5ea0c 2358 trace_rpc_socket_connect(xprt, sock, status);
46121cf7
CL		 2359 dprintk("RPC: %p connect status %d connected %d sock state %d\n",
2360 xprt, -status, xprt_connected(xprt),
2361 sock->sk->sk_state);
2a491991 2362 switch (status) {
f75e6745
TM		 2363 default:
2364 printk("%s: connect returned unhandled error %d\n",
2365 __func__, status);
2366 case -EADDRNOTAVAIL:
2367 /* We're probably in TIME_WAIT. Get rid of existing socket,
2368 * and retry
2369 */
a519fc7a 2370 xs_tcp_force_close(xprt);
88b5ed73 2371 break;
2a491991
TM		 2372 case 0:
2373 case -EINPROGRESS:
2374 case -EALREADY:
718ba5b8 2375 xprt_unlock_connect(xprt, transport);
7d1e8255 2376 return;
9fcfe0c8
TM		 2377 case -EINVAL:
2378 /* Happens, for instance, if the user specified a link
2379 * local IPv6 address without a scope-id.
2380 */
3ed5e2a2
TM		 2381 case -ECONNREFUSED:
2382 case -ECONNRESET:
2383 case -ENETUNREACH:
3913c78c 2384 case -EADDRINUSE:
3601c4a9 2385 case -ENOBUFS:
3ed5e2a2 2386 /* retry with existing socket, after a delay */
4efdd92c 2387 xs_tcp_force_close(xprt);
9fcfe0c8 2388 goto out;
a246b010 2389 }
2a491991 2390 status = -EAGAIN;
a246b010 2391out:
718ba5b8 2392 xprt_unlock_connect(xprt, transport);
68e220bd 2393 xprt_clear_connecting(xprt);
7d1e8255 2394 xprt_wake_pending_tasks(xprt, status);
68e220bd 2395}
b0d93ad5 2396
02910177
TM		 2397static unsigned long xs_reconnect_delay(const struct rpc_xprt *xprt)
2398{
2399 unsigned long start, now = jiffies;
2400
2401 start = xprt->stat.connect_start + xprt->reestablish_timeout;
2402 if (time_after(start, now))
2403 return start - now;
2404 return 0;
2405}
2406
3851f1cd
TM		 2407static void xs_reconnect_backoff(struct rpc_xprt *xprt)
2408{
2409 xprt->reestablish_timeout <<= 1;
2410 if (xprt->reestablish_timeout > xprt->max_reconnect_timeout)
2411 xprt->reestablish_timeout = xprt->max_reconnect_timeout;
2412 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2413 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2414}
2415
9903cd1c
CL		 2416/**
2417 * xs_connect - connect a socket to a remote endpoint
1b092092 2418 * @xprt: pointer to transport structure
9903cd1c
CL		 2419 * @task: address of RPC task that manages state of connect request
2420 *
2421 * TCP: If the remote end dropped the connection, delay reconnecting.
03bf4b70
CL		 2422 *
2423 * UDP socket connects are synchronous, but we use a work queue anyway
2424 * to guarantee that even unprivileged user processes can set up a
2425 * socket on a privileged port.
2426 *
2427 * If a UDP socket connect fails, the delay behavior here prevents
2428 * retry floods (hard mounts).
9903cd1c 2429 */
1b092092 2430static void xs_connect(struct rpc_xprt *xprt, struct rpc_task *task)
a246b010 2431{
ee0ac0c2 2432 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
02910177 2433 unsigned long delay = 0;
a246b010 2434
718ba5b8
TM		 2435 WARN_ON_ONCE(!xprt_lock_connect(xprt, task, transport));
2436
99b1a4c3 2437 if (transport->sock != NULL) {
46121cf7
CL		 2438 dprintk("RPC: xs_connect delayed xprt %p for %lu "
2439 "seconds\n",
03bf4b70 2440 xprt, xprt->reestablish_timeout / HZ);
99b1a4c3
TM		 2441
2442 /* Start by resetting any existing state */
2443 xs_reset_transport(transport);
2444
02910177 2445 delay = xs_reconnect_delay(xprt);
3851f1cd 2446 xs_reconnect_backoff(xprt);
02910177 2447
02910177 2448 } else
46121cf7 2449 dprintk("RPC: xs_connect scheduled xprt %p\n", xprt);
02910177
TM		 2450
2451 queue_delayed_work(xprtiod_workqueue,
2452 &transport->connect_worker,
2453 delay);
a246b010
CL		 2454}
2455
176e21ee
CL		 2456/**
2457 * xs_local_print_stats - display AF_LOCAL socket-specifc stats
2458 * @xprt: rpc_xprt struct containing statistics
2459 * @seq: output file
2460 *
2461 */
2462static void xs_local_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2463{
2464 long idle_time = 0;
2465
2466 if (xprt_connected(xprt))
2467 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2468
2469 seq_printf(seq, "\txprt:\tlocal %lu %lu %lu %ld %lu %lu %lu "
15a45206 2470 "%llu %llu %lu %llu %llu\n",
176e21ee
CL		 2471 xprt->stat.bind_count,
2472 xprt->stat.connect_count,
2473 xprt->stat.connect_time,
2474 idle_time,
2475 xprt->stat.sends,
2476 xprt->stat.recvs,
2477 xprt->stat.bad_xids,
2478 xprt->stat.req_u,
15a45206
AA		 2479 xprt->stat.bklog_u,
2480 xprt->stat.max_slots,
2481 xprt->stat.sending_u,
2482 xprt->stat.pending_u);
176e21ee
CL		 2483}
2484
262ca07d
CL		 2485/**
2486 * xs_udp_print_stats - display UDP socket-specifc stats
2487 * @xprt: rpc_xprt struct containing statistics
2488 * @seq: output file
2489 *
2490 */
2491static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2492{
c8475461
CL		 2493 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2494
15a45206
AA		 2495 seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %llu %llu "
2496 "%lu %llu %llu\n",
fbfffbd5 2497 transport->srcport,
262ca07d
CL		 2498 xprt->stat.bind_count,
2499 xprt->stat.sends,
2500 xprt->stat.recvs,
2501 xprt->stat.bad_xids,
2502 xprt->stat.req_u,
15a45206
AA		 2503 xprt->stat.bklog_u,
2504 xprt->stat.max_slots,
2505 xprt->stat.sending_u,
2506 xprt->stat.pending_u);
262ca07d
CL		 2507}
2508
2509/**
2510 * xs_tcp_print_stats - display TCP socket-specifc stats
2511 * @xprt: rpc_xprt struct containing statistics
2512 * @seq: output file
2513 *
2514 */
2515static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2516{
c8475461 2517 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
262ca07d
CL		 2518 long idle_time = 0;
2519
2520 if (xprt_connected(xprt))
2521 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2522
15a45206
AA		 2523 seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu "
2524 "%llu %llu %lu %llu %llu\n",
fbfffbd5 2525 transport->srcport,
262ca07d
CL		 2526 xprt->stat.bind_count,
2527 xprt->stat.connect_count,
2528 xprt->stat.connect_time,
2529 idle_time,
2530 xprt->stat.sends,
2531 xprt->stat.recvs,
2532 xprt->stat.bad_xids,
2533 xprt->stat.req_u,
15a45206
AA		 2534 xprt->stat.bklog_u,
2535 xprt->stat.max_slots,
2536 xprt->stat.sending_u,
2537 xprt->stat.pending_u);
262ca07d
CL		 2538}
2539
4cfc7e60
RI		 2540/*
2541 * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason
2542 * we allocate pages instead doing a kmalloc like rpc_malloc is because we want
2543 * to use the server side send routines.
2544 */
5fe6eaa1 2545static int bc_malloc(struct rpc_task *task)
4cfc7e60 2546{
5fe6eaa1
CL		 2547 struct rpc_rqst *rqst = task->tk_rqstp;
2548 size_t size = rqst->rq_callsize;
4cfc7e60
RI		 2549 struct page *page;
2550 struct rpc_buffer *buf;
2551
5fe6eaa1
CL		 2552 if (size > PAGE_SIZE - sizeof(struct rpc_buffer)) {
2553 WARN_ONCE(1, "xprtsock: large bc buffer request (size %zu)\n",
2554 size);
2555 return -EINVAL;
2556 }
4cfc7e60 2557
b8a13d03 2558 page = alloc_page(GFP_KERNEL);
4cfc7e60 2559 if (!page)
5fe6eaa1 2560 return -ENOMEM;
4cfc7e60
RI		 2561
2562 buf = page_address(page);
2563 buf->len = PAGE_SIZE;
2564
5fe6eaa1
CL		 2565 rqst->rq_buffer = buf->data;
2566 return 0;
4cfc7e60
RI		 2567}
2568
2569/*
2570 * Free the space allocated in the bc_alloc routine
2571 */
3435c74a 2572static void bc_free(struct rpc_task *task)
4cfc7e60 2573{
3435c74a 2574 void *buffer = task->tk_rqstp->rq_buffer;
4cfc7e60
RI		 2575 struct rpc_buffer *buf;
2576
4cfc7e60
RI		 2577 buf = container_of(buffer, struct rpc_buffer, data);
2578 free_page((unsigned long)buf);
2579}
2580
2581/*
2582 * Use the svc_sock to send the callback. Must be called with svsk->sk_mutex
2583 * held. Borrows heavily from svc_tcp_sendto and xs_tcp_send_request.
2584 */
2585static int bc_sendto(struct rpc_rqst *req)
2586{
2587 int len;
2588 struct xdr_buf *xbufp = &req->rq_snd_buf;
2589 struct rpc_xprt *xprt = req->rq_xprt;
2590 struct sock_xprt *transport =
2591 container_of(xprt, struct sock_xprt, xprt);
2592 struct socket *sock = transport->sock;
2593 unsigned long headoff;
2594 unsigned long tailoff;
2595
61677eee 2596 xs_encode_stream_record_marker(xbufp);
4cfc7e60
RI		 2597
2598 tailoff = (unsigned long)xbufp->tail[0].iov_base & ~PAGE_MASK;
2599 headoff = (unsigned long)xbufp->head[0].iov_base & ~PAGE_MASK;
2600 len = svc_send_common(sock, xbufp,
2601 virt_to_page(xbufp->head[0].iov_base), headoff,
2602 xbufp->tail[0].iov_base, tailoff);
2603
2604 if (len != xbufp->len) {
2605 printk(KERN_NOTICE "Error sending entire callback!\n");
2606 len = -EAGAIN;
2607 }
2608
2609 return len;
2610}
2611
2612/*
2613 * The send routine. Borrows from svc_send
2614 */
2615static int bc_send_request(struct rpc_task *task)
2616{
2617 struct rpc_rqst *req = task->tk_rqstp;
2618 struct svc_xprt *xprt;
7fc56136 2619 int len;
4cfc7e60
RI		 2620
2621 dprintk("sending request with xid: %08x\n", ntohl(req->rq_xid));
2622 /*
2623 * Get the server socket associated with this callback xprt
2624 */
2625 xprt = req->rq_xprt->bc_xprt;
4cfc7e60
RI		 2626
2627 /*
2628 * Grab the mutex to serialize data as the connection is shared
2629 * with the fore channel
2630 */
2631 if (!mutex_trylock(&xprt->xpt_mutex)) {
2632 rpc_sleep_on(&xprt->xpt_bc_pending, task, NULL);
2633 if (!mutex_trylock(&xprt->xpt_mutex))
2634 return -EAGAIN;
2635 rpc_wake_up_queued_task(&xprt->xpt_bc_pending, task);
2636 }
2637 if (test_bit(XPT_DEAD, &xprt->xpt_flags))
2638 len = -ENOTCONN;
2639 else
2640 len = bc_sendto(req);
2641 mutex_unlock(&xprt->xpt_mutex);
2642
2643 if (len > 0)
2644 len = 0;
2645
2646 return len;
2647}
2648
2649/*
2650 * The close routine. Since this is client initiated, we do nothing
2651 */
2652
2653static void bc_close(struct rpc_xprt *xprt)
2654{
4cfc7e60
RI		 2655}
2656
2657/*
2658 * The xprt destroy routine. Again, because this connection is client
2659 * initiated, we do nothing
2660 */
2661
2662static void bc_destroy(struct rpc_xprt *xprt)
2663{
47f72efa
KM		 2664 dprintk("RPC: bc_destroy xprt %p\n", xprt);
2665
2666 xs_xprt_free(xprt);
2667 module_put(THIS_MODULE);
4cfc7e60
RI		 2668}
2669
176e21ee
CL		 2670static struct rpc_xprt_ops xs_local_ops = {
2671 .reserve_xprt = xprt_reserve_xprt,
2672 .release_xprt = xs_tcp_release_xprt,
f39c1bfb 2673 .alloc_slot = xprt_alloc_slot,
176e21ee
CL		 2674 .rpcbind = xs_local_rpcbind,
2675 .set_port = xs_local_set_port,
dc107402 2676 .connect = xs_local_connect,
176e21ee
CL		 2677 .buf_alloc = rpc_malloc,
2678 .buf_free = rpc_free,
2679 .send_request = xs_local_send_request,
2680 .set_retrans_timeout = xprt_set_retrans_timeout_def,
2681 .close = xs_close,
a1311d87 2682 .destroy = xs_destroy,
176e21ee 2683 .print_stats = xs_local_print_stats,
d67fa4d8
JL		 2684 .enable_swap = xs_enable_swap,
2685 .disable_swap = xs_disable_swap,
176e21ee
CL		 2686};
2687
262965f5 2688static struct rpc_xprt_ops xs_udp_ops = {
43118c29 2689 .set_buffer_size = xs_udp_set_buffer_size,
12a80469 2690 .reserve_xprt = xprt_reserve_xprt_cong,
49e9a890 2691 .release_xprt = xprt_release_xprt_cong,
f39c1bfb 2692 .alloc_slot = xprt_alloc_slot,
45160d62 2693 .rpcbind = rpcb_getport_async,
92200412 2694 .set_port = xs_set_port,
262965f5 2695 .connect = xs_connect,
02107148
CL		 2696 .buf_alloc = rpc_malloc,
2697 .buf_free = rpc_free,
262965f5 2698 .send_request = xs_udp_send_request,
fe3aca29 2699 .set_retrans_timeout = xprt_set_retrans_timeout_rtt,
46c0ee8b 2700 .timer = xs_udp_timer,
a58dd398 2701 .release_request = xprt_release_rqst_cong,
262965f5
CL		 2702 .close = xs_close,
2703 .destroy = xs_destroy,
262ca07d 2704 .print_stats = xs_udp_print_stats,
d67fa4d8
JL		 2705 .enable_swap = xs_enable_swap,
2706 .disable_swap = xs_disable_swap,
4a068258 2707 .inject_disconnect = xs_inject_disconnect,
262965f5
CL		 2708};
2709
2710static struct rpc_xprt_ops xs_tcp_ops = {
12a80469 2711 .reserve_xprt = xprt_reserve_xprt,
e0ab53de 2712 .release_xprt = xs_tcp_release_xprt,
f39c1bfb 2713 .alloc_slot = xprt_lock_and_alloc_slot,
45160d62 2714 .rpcbind = rpcb_getport_async,
92200412 2715 .set_port = xs_set_port,
0b9e7943 2716 .connect = xs_connect,
02107148
CL		 2717 .buf_alloc = rpc_malloc,
2718 .buf_free = rpc_free,
262965f5 2719 .send_request = xs_tcp_send_request,
fe3aca29 2720 .set_retrans_timeout = xprt_set_retrans_timeout_def,
c627d31b 2721 .close = xs_tcp_shutdown,
9903cd1c 2722 .destroy = xs_destroy,
262ca07d 2723 .print_stats = xs_tcp_print_stats,
d67fa4d8
JL		 2724 .enable_swap = xs_enable_swap,
2725 .disable_swap = xs_disable_swap,
4a068258 2726 .inject_disconnect = xs_inject_disconnect,
42e5c3e2
CL		 2727#ifdef CONFIG_SUNRPC_BACKCHANNEL
2728 .bc_setup = xprt_setup_bc,
76566773 2729 .bc_up = xs_tcp_bc_up,
6b26cc8c 2730 .bc_maxpayload = xs_tcp_bc_maxpayload,
42e5c3e2
CL		 2731 .bc_free_rqst = xprt_free_bc_rqst,
2732 .bc_destroy = xprt_destroy_bc,
2733#endif
a246b010
CL		 2734};
2735
4cfc7e60
RI		 2736/*
2737 * The rpc_xprt_ops for the server backchannel
2738 */
2739
2740static struct rpc_xprt_ops bc_tcp_ops = {
2741 .reserve_xprt = xprt_reserve_xprt,
2742 .release_xprt = xprt_release_xprt,
84e28a30 2743 .alloc_slot = xprt_alloc_slot,
4cfc7e60
RI		 2744 .buf_alloc = bc_malloc,
2745 .buf_free = bc_free,
2746 .send_request = bc_send_request,
2747 .set_retrans_timeout = xprt_set_retrans_timeout_def,
2748 .close = bc_close,
2749 .destroy = bc_destroy,
2750 .print_stats = xs_tcp_print_stats,
d67fa4d8
JL		 2751 .enable_swap = xs_enable_swap,
2752 .disable_swap = xs_disable_swap,
4a068258 2753 .inject_disconnect = xs_inject_disconnect,
4cfc7e60
RI		 2754};
2755
92476850
CL		 2756static int xs_init_anyaddr(const int family, struct sockaddr *sap)
2757{
2758 static const struct sockaddr_in sin = {
2759 .sin_family = AF_INET,
2760 .sin_addr.s_addr = htonl(INADDR_ANY),
2761 };
2762 static const struct sockaddr_in6 sin6 = {
2763 .sin6_family = AF_INET6,
2764 .sin6_addr = IN6ADDR_ANY_INIT,
2765 };
2766
2767 switch (family) {
176e21ee
CL		 2768 case AF_LOCAL:
2769 break;
92476850
CL		 2770 case AF_INET:
2771 memcpy(sap, &sin, sizeof(sin));
2772 break;
2773 case AF_INET6:
2774 memcpy(sap, &sin6, sizeof(sin6));
2775 break;
2776 default:
2777 dprintk("RPC: %s: Bad address family\n", __func__);
2778 return -EAFNOSUPPORT;
2779 }
2780 return 0;
2781}
2782
3c341b0b 2783static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
d9ba131d
TM		 2784 unsigned int slot_table_size,
2785 unsigned int max_slot_table_size)
c8541ecd
CL		 2786{
2787 struct rpc_xprt *xprt;
ffc2e518 2788 struct sock_xprt *new;
c8541ecd 2789
96802a09 2790 if (args->addrlen > sizeof(xprt->addr)) {
46121cf7 2791 dprintk("RPC: xs_setup_xprt: address too large\n");
c8541ecd
CL		 2792 return ERR_PTR(-EBADF);
2793 }
2794
d9ba131d
TM		 2795 xprt = xprt_alloc(args->net, sizeof(*new), slot_table_size,
2796 max_slot_table_size);
bd1722d4 2797 if (xprt == NULL) {
46121cf7
CL		 2798 dprintk("RPC: xs_setup_xprt: couldn't allocate "
2799 "rpc_xprt\n");
c8541ecd
CL		 2800 return ERR_PTR(-ENOMEM);
2801 }
c8541ecd 2802
bd1722d4 2803 new = container_of(xprt, struct sock_xprt, xprt);
edc1b01c 2804 mutex_init(&new->recv_mutex);
96802a09
FM		 2805 memcpy(&xprt->addr, args->dstaddr, args->addrlen);
2806 xprt->addrlen = args->addrlen;
d3bc9a1d 2807 if (args->srcaddr)
fbfffbd5 2808 memcpy(&new->srcaddr, args->srcaddr, args->addrlen);
92476850
CL		 2809 else {
2810 int err;
2811 err = xs_init_anyaddr(args->dstaddr->sa_family,
2812 (struct sockaddr *)&new->srcaddr);
2aa13531
SK		 2813 if (err != 0) {
2814 xprt_free(xprt);
92476850 2815 return ERR_PTR(err);
2aa13531 2816 }
92476850 2817 }
c8541ecd
CL		 2818
2819 return xprt;
2820}
2821
176e21ee
CL		 2822static const struct rpc_timeout xs_local_default_timeout = {
2823 .to_initval = 10 * HZ,
2824 .to_maxval = 10 * HZ,
2825 .to_retries = 2,
2826};
2827
2828/**
2829 * xs_setup_local - Set up transport to use an AF_LOCAL socket
2830 * @args: rpc transport creation arguments
2831 *
2832 * AF_LOCAL is a "tpi_cots_ord" transport, just like TCP
2833 */
2834static struct rpc_xprt *xs_setup_local(struct xprt_create *args)
2835{
2836 struct sockaddr_un *sun = (struct sockaddr_un *)args->dstaddr;
2837 struct sock_xprt *transport;
2838 struct rpc_xprt *xprt;
2839 struct rpc_xprt *ret;
2840
d9ba131d
TM		 2841 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2842 xprt_max_tcp_slot_table_entries);
176e21ee
CL		 2843 if (IS_ERR(xprt))
2844 return xprt;
2845 transport = container_of(xprt, struct sock_xprt, xprt);
2846
2847 xprt->prot = 0;
2848 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2849 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2850
2851 xprt->bind_timeout = XS_BIND_TO;
2852 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2853 xprt->idle_timeout = XS_IDLE_DISC_TO;
2854
2855 xprt->ops = &xs_local_ops;
2856 xprt->timeout = &xs_local_default_timeout;
2857
a2648094 2858 INIT_WORK(&transport->recv_worker, xs_local_data_receive_workfn);
93dc41bd
N		 2859 INIT_DELAYED_WORK(&transport->connect_worker,
2860 xs_dummy_setup_socket);
2861
176e21ee
CL		 2862 switch (sun->sun_family) {
2863 case AF_LOCAL:
2864 if (sun->sun_path[0] != '/') {
2865 dprintk("RPC: bad AF_LOCAL address: %s\n",
2866 sun->sun_path);
2867 ret = ERR_PTR(-EINVAL);
2868 goto out_err;
2869 }
2870 xprt_set_bound(xprt);
176e21ee 2871 xs_format_peer_addresses(xprt, "local", RPCBIND_NETID_LOCAL);
7073ea87
BF		 2872 ret = ERR_PTR(xs_local_setup_socket(transport));
2873 if (ret)
2874 goto out_err;
176e21ee
CL		 2875 break;
2876 default:
2877 ret = ERR_PTR(-EAFNOSUPPORT);
2878 goto out_err;
2879 }
2880
2881 dprintk("RPC: set up xprt to %s via AF_LOCAL\n",
2882 xprt->address_strings[RPC_DISPLAY_ADDR]);
2883
2884 if (try_module_get(THIS_MODULE))
2885 return xprt;
2886 ret = ERR_PTR(-EINVAL);
2887out_err:
315f3812 2888 xs_xprt_free(xprt);
176e21ee
CL		 2889 return ret;
2890}
2891
2881ae74
TM		 2892static const struct rpc_timeout xs_udp_default_timeout = {
2893 .to_initval = 5 * HZ,
2894 .to_maxval = 30 * HZ,
2895 .to_increment = 5 * HZ,
2896 .to_retries = 5,
2897};
2898
9903cd1c
CL		 2899/**
2900 * xs_setup_udp - Set up transport to use a UDP socket
96802a09 2901 * @args: rpc transport creation arguments
9903cd1c
CL		 2902 *
2903 */
483066d6 2904static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
a246b010 2905{
8f9d5b1a 2906 struct sockaddr *addr = args->dstaddr;
c8541ecd 2907 struct rpc_xprt *xprt;
c8475461 2908 struct sock_xprt *transport;
0a68b0be 2909 struct rpc_xprt *ret;
a246b010 2910
d9ba131d
TM		 2911 xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries,
2912 xprt_udp_slot_table_entries);
c8541ecd
CL		 2913 if (IS_ERR(xprt))
2914 return xprt;
c8475461 2915 transport = container_of(xprt, struct sock_xprt, xprt);
a246b010 2916
ec739ef0 2917 xprt->prot = IPPROTO_UDP;
808012fb 2918 xprt->tsh_size = 0;
a246b010
CL		 2919 /* XXX: header size can vary due to auth type, IPv6, etc. */
2920 xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
2921
03bf4b70 2922 xprt->bind_timeout = XS_BIND_TO;
03bf4b70
CL		 2923 xprt->reestablish_timeout = XS_UDP_REEST_TO;
2924 xprt->idle_timeout = XS_IDLE_DISC_TO;
a246b010 2925
262965f5 2926 xprt->ops = &xs_udp_ops;
a246b010 2927
ba7392bb 2928 xprt->timeout = &xs_udp_default_timeout;
a246b010 2929
f9b2ee71 2930 INIT_WORK(&transport->recv_worker, xs_udp_data_receive_workfn);
edc1b01c
TM		 2931 INIT_DELAYED_WORK(&transport->connect_worker, xs_udp_setup_socket);
2932
8f9d5b1a
CL		 2933 switch (addr->sa_family) {
2934 case AF_INET:
2935 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2936 xprt_set_bound(xprt);
2937
9dc3b095 2938 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
8f9d5b1a
CL		 2939 break;
2940 case AF_INET6:
2941 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2942 xprt_set_bound(xprt);
2943
9dc3b095 2944 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
8f9d5b1a
CL		 2945 break;
2946 default:
0a68b0be
BF		 2947 ret = ERR_PTR(-EAFNOSUPPORT);
2948 goto out_err;
8f9d5b1a
CL		 2949 }
2950
c740eff8
CL		 2951 if (xprt_bound(xprt))
2952 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2953 xprt->address_strings[RPC_DISPLAY_ADDR],
2954 xprt->address_strings[RPC_DISPLAY_PORT],
2955 xprt->address_strings[RPC_DISPLAY_PROTO]);
2956 else
2957 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
2958 xprt->address_strings[RPC_DISPLAY_ADDR],
2959 xprt->address_strings[RPC_DISPLAY_PROTO]);
edb267a6 2960
bc25571e
TT		 2961 if (try_module_get(THIS_MODULE))
2962 return xprt;
0a68b0be
BF		 2963 ret = ERR_PTR(-EINVAL);
2964out_err:
315f3812 2965 xs_xprt_free(xprt);
0a68b0be 2966 return ret;
a246b010
CL		 2967}
2968
2881ae74
TM		 2969static const struct rpc_timeout xs_tcp_default_timeout = {
2970 .to_initval = 60 * HZ,
2971 .to_maxval = 60 * HZ,
2972 .to_retries = 2,
2973};
2974
9903cd1c
CL		 2975/**
2976 * xs_setup_tcp - Set up transport to use a TCP socket
96802a09 2977 * @args: rpc transport creation arguments
9903cd1c
CL		 2978 *
2979 */
483066d6 2980static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
a246b010 2981{
8f9d5b1a 2982 struct sockaddr *addr = args->dstaddr;
c8541ecd 2983 struct rpc_xprt *xprt;
c8475461 2984 struct sock_xprt *transport;
0a68b0be 2985 struct rpc_xprt *ret;
b7993ceb
TM		 2986 unsigned int max_slot_table_size = xprt_max_tcp_slot_table_entries;
2987
2988 if (args->flags & XPRT_CREATE_INFINITE_SLOTS)
2989 max_slot_table_size = RPC_MAX_SLOT_TABLE_LIMIT;
a246b010 2990
d9ba131d 2991 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
b7993ceb 2992 max_slot_table_size);
c8541ecd
CL		 2993 if (IS_ERR(xprt))
2994 return xprt;
c8475461 2995 transport = container_of(xprt, struct sock_xprt, xprt);
a246b010 2996
ec739ef0 2997 xprt->prot = IPPROTO_TCP;
808012fb 2998 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
808012fb 2999 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
a246b010 3000
03bf4b70 3001 xprt->bind_timeout = XS_BIND_TO;
03bf4b70
CL		 3002 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
3003 xprt->idle_timeout = XS_IDLE_DISC_TO;
a246b010 3004
262965f5 3005 xprt->ops = &xs_tcp_ops;
ba7392bb 3006 xprt->timeout = &xs_tcp_default_timeout;
a246b010 3007
3851f1cd
TM		 3008 xprt->max_reconnect_timeout = xprt->timeout->to_maxval;
3009
edc1b01c
TM		 3010 INIT_WORK(&transport->recv_worker, xs_tcp_data_receive_workfn);
3011 INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_setup_socket);
3012
8f9d5b1a
CL		 3013 switch (addr->sa_family) {
3014 case AF_INET:
3015 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
3016 xprt_set_bound(xprt);
3017
9dc3b095 3018 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
8f9d5b1a
CL		 3019 break;
3020 case AF_INET6:
3021 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
3022 xprt_set_bound(xprt);
3023
9dc3b095 3024 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
8f9d5b1a
CL		 3025 break;
3026 default:
0a68b0be
BF		 3027 ret = ERR_PTR(-EAFNOSUPPORT);
3028 goto out_err;
8f9d5b1a
CL		 3029 }
3030
c740eff8
CL		 3031 if (xprt_bound(xprt))
3032 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
3033 xprt->address_strings[RPC_DISPLAY_ADDR],
3034 xprt->address_strings[RPC_DISPLAY_PORT],
3035 xprt->address_strings[RPC_DISPLAY_PROTO]);
3036 else
3037 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
3038 xprt->address_strings[RPC_DISPLAY_ADDR],
3039 xprt->address_strings[RPC_DISPLAY_PROTO]);
3040
bc25571e
TT		 3041 if (try_module_get(THIS_MODULE))
3042 return xprt;
0a68b0be
BF		 3043 ret = ERR_PTR(-EINVAL);
3044out_err:
315f3812 3045 xs_xprt_free(xprt);
0a68b0be 3046 return ret;
a246b010 3047}
282b32e1 3048
f300baba
AB		 3049/**
3050 * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket
3051 * @args: rpc transport creation arguments
3052 *
3053 */
3054static struct rpc_xprt *xs_setup_bc_tcp(struct xprt_create *args)
3055{
3056 struct sockaddr *addr = args->dstaddr;
3057 struct rpc_xprt *xprt;
3058 struct sock_xprt *transport;
3059 struct svc_sock *bc_sock;
0a68b0be 3060 struct rpc_xprt *ret;
f300baba 3061
d9ba131d
TM		 3062 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
3063 xprt_tcp_slot_table_entries);
f300baba
AB		 3064 if (IS_ERR(xprt))
3065 return xprt;
3066 transport = container_of(xprt, struct sock_xprt, xprt);
3067
3068 xprt->prot = IPPROTO_TCP;
3069 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
3070 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
3071 xprt->timeout = &xs_tcp_default_timeout;
3072
3073 /* backchannel */
3074 xprt_set_bound(xprt);
3075 xprt->bind_timeout = 0;
f300baba
AB		 3076 xprt->reestablish_timeout = 0;
3077 xprt->idle_timeout = 0;
3078
f300baba
AB		 3079 xprt->ops = &bc_tcp_ops;
3080
3081 switch (addr->sa_family) {
3082 case AF_INET:
3083 xs_format_peer_addresses(xprt, "tcp",
3084 RPCBIND_NETID_TCP);
3085 break;
3086 case AF_INET6:
3087 xs_format_peer_addresses(xprt, "tcp",
3088 RPCBIND_NETID_TCP6);
3089 break;
3090 default:
0a68b0be
BF		 3091 ret = ERR_PTR(-EAFNOSUPPORT);
3092 goto out_err;
f300baba
AB		 3093 }
3094
50fa0d40
PE		 3095 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
3096 xprt->address_strings[RPC_DISPLAY_ADDR],
3097 xprt->address_strings[RPC_DISPLAY_PORT],
3098 xprt->address_strings[RPC_DISPLAY_PROTO]);
f300baba 3099
99de8ea9
BF		 3100 /*
3101 * Once we've associated a backchannel xprt with a connection,
28303ca3
WM		 3102 * we want to keep it around as long as the connection lasts,
3103 * in case we need to start using it for a backchannel again;
3104 * this reference won't be dropped until bc_xprt is destroyed.
99de8ea9
BF		 3105 */
3106 xprt_get(xprt);
3107 args->bc_xprt->xpt_bc_xprt = xprt;
3108 xprt->bc_xprt = args->bc_xprt;
3109 bc_sock = container_of(args->bc_xprt, struct svc_sock, sk_xprt);
3110 transport->sock = bc_sock->sk_sock;
3111 transport->inet = bc_sock->sk_sk;
3112
f300baba
AB		 3113 /*
3114 * Since we don't want connections for the backchannel, we set
3115 * the xprt status to connected
3116 */
3117 xprt_set_connected(xprt);
3118
f300baba
AB		 3119 if (try_module_get(THIS_MODULE))
3120 return xprt;
642aab58
KM		 3121
3122 args->bc_xprt->xpt_bc_xprt = NULL;
39a9beab 3123 args->bc_xprt->xpt_bc_xps = NULL;
99de8ea9 3124 xprt_put(xprt);
0a68b0be
BF		 3125 ret = ERR_PTR(-EINVAL);
3126out_err:
315f3812 3127 xs_xprt_free(xprt);
0a68b0be 3128 return ret;
f300baba
AB		 3129}
3130
176e21ee
CL		 3131static struct xprt_class xs_local_transport = {
3132 .list = LIST_HEAD_INIT(xs_local_transport.list),
3133 .name = "named UNIX socket",
3134 .owner = THIS_MODULE,
3135 .ident = XPRT_TRANSPORT_LOCAL,
3136 .setup = xs_setup_local,
3137};
3138
bc25571e
TT		 3139static struct xprt_class xs_udp_transport = {
3140 .list = LIST_HEAD_INIT(xs_udp_transport.list),
3141 .name = "udp",
3142 .owner = THIS_MODULE,
f300baba 3143 .ident = XPRT_TRANSPORT_UDP,
bc25571e
TT		 3144 .setup = xs_setup_udp,
3145};
3146
3147static struct xprt_class xs_tcp_transport = {
3148 .list = LIST_HEAD_INIT(xs_tcp_transport.list),
3149 .name = "tcp",
3150 .owner = THIS_MODULE,
f300baba 3151 .ident = XPRT_TRANSPORT_TCP,
bc25571e
TT		 3152 .setup = xs_setup_tcp,
3153};
3154
f300baba
AB		 3155static struct xprt_class xs_bc_tcp_transport = {
3156 .list = LIST_HEAD_INIT(xs_bc_tcp_transport.list),
3157 .name = "tcp NFSv4.1 backchannel",
3158 .owner = THIS_MODULE,
3159 .ident = XPRT_TRANSPORT_BC_TCP,
3160 .setup = xs_setup_bc_tcp,
3161};
3162
282b32e1 3163/**
bc25571e 3164 * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
282b32e1
CL		 3165 *
3166 */
3167int init_socket_xprt(void)
3168{
f895b252 3169#if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
2b1bec5f 3170 if (!sunrpc_table_header)
0b4d4147 3171 sunrpc_table_header = register_sysctl_table(sunrpc_table);
fbf76683
CL		 3172#endif
3173
176e21ee 3174 xprt_register_transport(&xs_local_transport);
bc25571e
TT		 3175 xprt_register_transport(&xs_udp_transport);
3176 xprt_register_transport(&xs_tcp_transport);
f300baba 3177 xprt_register_transport(&xs_bc_tcp_transport);
bc25571e 3178
282b32e1
CL		 3179 return 0;
3180}
3181
3182/**
bc25571e 3183 * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
282b32e1
CL		 3184 *
3185 */
3186void cleanup_socket_xprt(void)
3187{
f895b252 3188#if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
fbf76683
CL		 3189 if (sunrpc_table_header) {
3190 unregister_sysctl_table(sunrpc_table_header);
3191 sunrpc_table_header = NULL;
3192 }
3193#endif
bc25571e 3194
176e21ee 3195 xprt_unregister_transport(&xs_local_transport);
bc25571e
TT		 3196 xprt_unregister_transport(&xs_udp_transport);
3197 xprt_unregister_transport(&xs_tcp_transport);
f300baba 3198 xprt_unregister_transport(&xs_bc_tcp_transport);
282b32e1 3199}
cbf11071 3200
9bbb9e5a
RR		 3201static int param_set_uint_minmax(const char *val,
3202 const struct kernel_param *kp,
cbf11071
TM		 3203 unsigned int min, unsigned int max)
3204{
00cfaa94 3205 unsigned int num;
cbf11071
TM		 3206 int ret;
3207
3208 if (!val)
3209 return -EINVAL;
00cfaa94 3210 ret = kstrtouint(val, 0, &num);
cbf11071
TM		 3211 if (ret == -EINVAL || num < min || num > max)
3212 return -EINVAL;
3213 *((unsigned int *)kp->arg) = num;
3214 return 0;
3215}
3216
9bbb9e5a 3217static int param_set_portnr(const char *val, const struct kernel_param *kp)
cbf11071 3218{
ffb6ca33
FS		 3219 if (kp->arg == &xprt_min_resvport)
3220 return param_set_uint_minmax(val, kp,
cbf11071 3221 RPC_MIN_RESVPORT,
ffb6ca33
FS		 3222 xprt_max_resvport);
3223 return param_set_uint_minmax(val, kp,
3224 xprt_min_resvport,
cbf11071
TM		 3225 RPC_MAX_RESVPORT);
3226}
3227
9c27847d 3228static const struct kernel_param_ops param_ops_portnr = {
9bbb9e5a
RR		 3229 .set = param_set_portnr,
3230 .get = param_get_uint,
3231};
3232
cbf11071
TM		 3233#define param_check_portnr(name, p) \
3234 __param_check(name, p, unsigned int);
3235
3236module_param_named(min_resvport, xprt_min_resvport, portnr, 0644);
3237module_param_named(max_resvport, xprt_max_resvport, portnr, 0644);
3238
9bbb9e5a
RR		 3239static int param_set_slot_table_size(const char *val,
3240 const struct kernel_param *kp)
cbf11071
TM		 3241{
3242 return param_set_uint_minmax(val, kp,
3243 RPC_MIN_SLOT_TABLE,
3244 RPC_MAX_SLOT_TABLE);
3245}
3246
9c27847d 3247static const struct kernel_param_ops param_ops_slot_table_size = {
9bbb9e5a
RR		 3248 .set = param_set_slot_table_size,
3249 .get = param_get_uint,
3250};
3251
cbf11071
TM		 3252#define param_check_slot_table_size(name, p) \
3253 __param_check(name, p, unsigned int);
3254
d9ba131d
TM		 3255static int param_set_max_slot_table_size(const char *val,
3256 const struct kernel_param *kp)
3257{
3258 return param_set_uint_minmax(val, kp,
3259 RPC_MIN_SLOT_TABLE,
3260 RPC_MAX_SLOT_TABLE_LIMIT);
3261}
3262
9c27847d 3263static const struct kernel_param_ops param_ops_max_slot_table_size = {
d9ba131d
TM		 3264 .set = param_set_max_slot_table_size,
3265 .get = param_get_uint,
3266};
3267
3268#define param_check_max_slot_table_size(name, p) \
3269 __param_check(name, p, unsigned int);
3270
cbf11071
TM		 3271module_param_named(tcp_slot_table_entries, xprt_tcp_slot_table_entries,
3272 slot_table_size, 0644);
d9ba131d
TM		 3273module_param_named(tcp_max_slot_table_entries, xprt_max_tcp_slot_table_entries,
3274 max_slot_table_size, 0644);
cbf11071
TM		 3275module_param_named(udp_slot_table_entries, xprt_udp_slot_table_entries,
3276 slot_table_size, 0644);
3277
ubuntu-bionic-kernel mirror