]> git.proxmox.com Git - mirror_ubuntu-focal-kernel.git/blame - net/sunrpc/xprtsock.c
projects / mirror_ubuntu-focal-kernel.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
SUNRPC: Small optimisation of client receive
[mirror_ubuntu-focal-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
CL		 126 .extra1 = &xprt_min_resvport_limit,
127 .extra2 = &xprt_max_resvport_limit
128 },
129 {
fbf76683
CL		 130 .procname = "max_resvport",
131 .data = &xprt_max_resvport,
132 .maxlen = sizeof(unsigned int),
133 .mode = 0644,
6d456111 134 .proc_handler = proc_dointvec_minmax,
fbf76683
CL		 135 .extra1 = &xprt_min_resvport_limit,
136 .extra2 = &xprt_max_resvport_limit
137 },
25fe6142
TM		 138 {
139 .procname = "tcp_fin_timeout",
140 .data = &xs_tcp_fin_timeout,
141 .maxlen = sizeof(xs_tcp_fin_timeout),
142 .mode = 0644,
6d456111 143 .proc_handler = proc_dointvec_jiffies,
fbf76683 144 },
f8572d8f 145 { },
fbf76683
CL		 146};
147
fe2c6338 148static struct ctl_table sunrpc_table[] = {
fbf76683 149 {
fbf76683
CL		 150 .procname = "sunrpc",
151 .mode = 0555,
152 .child = xs_tunables_table
153 },
f8572d8f 154 { },
fbf76683
CL		 155};
156
157#endif
158
03bf4b70
CL		 159/*
160 * Wait duration for a reply from the RPC portmapper.
161 */
162#define XS_BIND_TO (60U * HZ)
163
164/*
165 * Delay if a UDP socket connect error occurs. This is most likely some
166 * kind of resource problem on the local host.
167 */
168#define XS_UDP_REEST_TO (2U * HZ)
169
170/*
171 * The reestablish timeout allows clients to delay for a bit before attempting
172 * to reconnect to a server that just dropped our connection.
173 *
174 * We implement an exponential backoff when trying to reestablish a TCP
175 * transport connection with the server. Some servers like to drop a TCP
176 * connection when they are overworked, so we start with a short timeout and
177 * increase over time if the server is down or not responding.
178 */
179#define XS_TCP_INIT_REEST_TO (3U * HZ)
180#define XS_TCP_MAX_REEST_TO (5U * 60 * HZ)
181
182/*
183 * TCP idle timeout; client drops the transport socket if it is idle
184 * for this long. Note that we also timeout UDP sockets to prevent
185 * holding port numbers when there is no RPC traffic.
186 */
187#define XS_IDLE_DISC_TO (5U * 60 * HZ)
188
f895b252 189#if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
a246b010 190# undef RPC_DEBUG_DATA
9903cd1c 191# define RPCDBG_FACILITY RPCDBG_TRANS
a246b010
CL		 192#endif
193
a246b010 194#ifdef RPC_DEBUG_DATA
9903cd1c 195static void xs_pktdump(char *msg, u32 *packet, unsigned int count)
a246b010 196{
9903cd1c
CL		 197 u8 *buf = (u8 *) packet;
198 int j;
a246b010 199
46121cf7 200 dprintk("RPC: %s\n", msg);
a246b010
CL		 201 for (j = 0; j < count && j < 128; j += 4) {
202 if (!(j & 31)) {
203 if (j)
204 dprintk("\n");
205 dprintk("0x%04x ", j);
206 }
207 dprintk("%02x%02x%02x%02x ",
208 buf[j], buf[j+1], buf[j+2], buf[j+3]);
209 }
210 dprintk("\n");
211}
212#else
9903cd1c 213static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count)
a246b010
CL		 214{
215 /* NOP */
216}
217#endif
218
2118071d
TM		 219static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
220{
221 return (struct rpc_xprt *) sk->sk_user_data;
222}
223
95392c59
CL		 224static inline struct sockaddr *xs_addr(struct rpc_xprt *xprt)
225{
226 return (struct sockaddr *) &xprt->addr;
227}
228
176e21ee
CL		 229static inline struct sockaddr_un *xs_addr_un(struct rpc_xprt *xprt)
230{
231 return (struct sockaddr_un *) &xprt->addr;
232}
233
95392c59 234static inline struct sockaddr_in *xs_addr_in(struct rpc_xprt *xprt)
edb267a6 235{
95392c59
CL		 236 return (struct sockaddr_in *) &xprt->addr;
237}
238
239static inline struct sockaddr_in6 *xs_addr_in6(struct rpc_xprt *xprt)
240{
241 return (struct sockaddr_in6 *) &xprt->addr;
242}
243
c877b849 244static void xs_format_common_peer_addresses(struct rpc_xprt *xprt)
edb267a6 245{
c877b849 246 struct sockaddr *sap = xs_addr(xprt);
9dc3b095
CL		 247 struct sockaddr_in6 *sin6;
248 struct sockaddr_in *sin;
176e21ee 249 struct sockaddr_un *sun;
c877b849 250 char buf[128];
edb267a6 251
9dc3b095 252 switch (sap->sa_family) {
176e21ee
CL		 253 case AF_LOCAL:
254 sun = xs_addr_un(xprt);
255 strlcpy(buf, sun->sun_path, sizeof(buf));
256 xprt->address_strings[RPC_DISPLAY_ADDR] =
257 kstrdup(buf, GFP_KERNEL);
258 break;
9dc3b095 259 case AF_INET:
176e21ee
CL		 260 (void)rpc_ntop(sap, buf, sizeof(buf));
261 xprt->address_strings[RPC_DISPLAY_ADDR] =
262 kstrdup(buf, GFP_KERNEL);
9dc3b095 263 sin = xs_addr_in(xprt);
fc0b5791 264 snprintf(buf, sizeof(buf), "%08x", ntohl(sin->sin_addr.s_addr));
9dc3b095
CL		 265 break;
266 case AF_INET6:
176e21ee
CL		 267 (void)rpc_ntop(sap, buf, sizeof(buf));
268 xprt->address_strings[RPC_DISPLAY_ADDR] =
269 kstrdup(buf, GFP_KERNEL);
9dc3b095 270 sin6 = xs_addr_in6(xprt);
fc0b5791 271 snprintf(buf, sizeof(buf), "%pi6", &sin6->sin6_addr);
9dc3b095
CL		 272 break;
273 default:
274 BUG();
756805e7 275 }
176e21ee 276
9dc3b095 277 xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = kstrdup(buf, GFP_KERNEL);
edb267a6
CL		 278}
279
9dc3b095 280static void xs_format_common_peer_ports(struct rpc_xprt *xprt)
4b6473fb 281{
9dc3b095
CL		 282 struct sockaddr *sap = xs_addr(xprt);
283 char buf[128];
4b6473fb 284
81160e66 285 snprintf(buf, sizeof(buf), "%u", rpc_get_port(sap));
c877b849 286 xprt->address_strings[RPC_DISPLAY_PORT] = kstrdup(buf, GFP_KERNEL);
4b6473fb 287
81160e66 288 snprintf(buf, sizeof(buf), "%4hx", rpc_get_port(sap));
c877b849
CL		 289 xprt->address_strings[RPC_DISPLAY_HEX_PORT] = kstrdup(buf, GFP_KERNEL);
290}
4b6473fb 291
9dc3b095
CL		 292static void xs_format_peer_addresses(struct rpc_xprt *xprt,
293 const char *protocol,
294 const char *netid)
c877b849 295{
b454ae90 296 xprt->address_strings[RPC_DISPLAY_PROTO] = protocol;
b454ae90 297 xprt->address_strings[RPC_DISPLAY_NETID] = netid;
c877b849 298 xs_format_common_peer_addresses(xprt);
9dc3b095 299 xs_format_common_peer_ports(xprt);
edb267a6 300}
4b6473fb 301
9dc3b095 302static void xs_update_peer_port(struct rpc_xprt *xprt)
4b6473fb 303{
9dc3b095
CL		 304 kfree(xprt->address_strings[RPC_DISPLAY_HEX_PORT]);
305 kfree(xprt->address_strings[RPC_DISPLAY_PORT]);
4417c8c4 306
9dc3b095 307 xs_format_common_peer_ports(xprt);
edb267a6
CL		 308}
309
310static void xs_free_peer_addresses(struct rpc_xprt *xprt)
311{
33e01dc7
CL		 312 unsigned int i;
313
314 for (i = 0; i < RPC_DISPLAY_MAX; i++)
315 switch (i) {
316 case RPC_DISPLAY_PROTO:
317 case RPC_DISPLAY_NETID:
318 continue;
319 default:
320 kfree(xprt->address_strings[i]);
321 }
edb267a6
CL		 322}
323
b4b5cc85
CL		 324#define XS_SENDMSG_FLAGS (MSG_DONTWAIT | MSG_NOSIGNAL)
325
24c5684b 326static int xs_send_kvec(struct socket *sock, struct sockaddr *addr, int addrlen, struct kvec *vec, unsigned int base, int more)
b4b5cc85 327{
b4b5cc85
CL		 328 struct msghdr msg = {
329 .msg_name = addr,
330 .msg_namelen = addrlen,
24c5684b
TM		 331 .msg_flags = XS_SENDMSG_FLAGS | (more ? MSG_MORE : 0),
332 };
333 struct kvec iov = {
334 .iov_base = vec->iov_base + base,
335 .iov_len = vec->iov_len - base,
b4b5cc85
CL		 336 };
337
24c5684b 338 if (iov.iov_len != 0)
b4b5cc85
CL		 339 return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
340 return kernel_sendmsg(sock, &msg, NULL, 0, 0);
341}
342
f279cd00 343static int xs_send_pagedata(struct socket *sock, struct xdr_buf *xdr, unsigned int base, int more, bool zerocopy, int *sent_p)
b4b5cc85 344{
a6b31d18
TM		 345 ssize_t (*do_sendpage)(struct socket *sock, struct page *page,
346 int offset, size_t size, int flags);
24c5684b
TM		 347 struct page **ppage;
348 unsigned int remainder;
f279cd00 349 int err;
24c5684b
TM		 350
351 remainder = xdr->page_len - base;
352 base += xdr->page_base;
353 ppage = xdr->pages + (base >> PAGE_SHIFT);
354 base &= ~PAGE_MASK;
a6b31d18
TM		 355 do_sendpage = sock->ops->sendpage;
356 if (!zerocopy)
357 do_sendpage = sock_no_sendpage;
24c5684b
TM		 358 for(;;) {
359 unsigned int len = min_t(unsigned int, PAGE_SIZE - base, remainder);
360 int flags = XS_SENDMSG_FLAGS;
b4b5cc85 361
24c5684b 362 remainder -= len;
31303d6c 363 if (more)
24c5684b 364 flags |= MSG_MORE;
31303d6c
TM		 365 if (remainder != 0)
366 flags |= MSG_SENDPAGE_NOTLAST | MSG_MORE;
a6b31d18 367 err = do_sendpage(sock, *ppage, base, len, flags);
24c5684b
TM		 368 if (remainder == 0 || err != len)
369 break;
f279cd00 370 *sent_p += err;
24c5684b
TM		 371 ppage++;
372 base = 0;
373 }
f279cd00
JB		 374 if (err > 0) {
375 *sent_p += err;
376 err = 0;
377 }
378 return err;
b4b5cc85
CL		 379}
380
9903cd1c
CL		 381/**
382 * xs_sendpages - write pages directly to a socket
383 * @sock: socket to send on
384 * @addr: UDP only -- address of destination
385 * @addrlen: UDP only -- length of destination address
386 * @xdr: buffer containing this request
387 * @base: starting position in the buffer
a6b31d18 388 * @zerocopy: true if it is safe to use sendpage()
f279cd00 389 * @sent_p: return the total number of bytes successfully queued for sending
9903cd1c 390 *
a246b010 391 */
f279cd00 392static int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base, bool zerocopy, int *sent_p)
a246b010 393{
24c5684b 394 unsigned int remainder = xdr->len - base;
f279cd00
JB		 395 int err = 0;
396 int sent = 0;
a246b010 397
262965f5 398 if (unlikely(!sock))
fba91afb 399 return -ENOTSOCK;
262965f5 400
24c5684b
TM		 401 if (base != 0) {
402 addr = NULL;
403 addrlen = 0;
404 }
262965f5 405
24c5684b
TM		 406 if (base < xdr->head[0].iov_len || addr != NULL) {
407 unsigned int len = xdr->head[0].iov_len - base;
408 remainder -= len;
409 err = xs_send_kvec(sock, addr, addrlen, &xdr->head[0], base, remainder != 0);
410 if (remainder == 0 || err != len)
a246b010 411 goto out;
f279cd00 412 *sent_p += err;
a246b010
CL		 413 base = 0;
414 } else
24c5684b 415 base -= xdr->head[0].iov_len;
a246b010 416
24c5684b
TM		 417 if (base < xdr->page_len) {
418 unsigned int len = xdr->page_len - base;
419 remainder -= len;
f279cd00
JB		 420 err = xs_send_pagedata(sock, xdr, base, remainder != 0, zerocopy, &sent);
421 *sent_p += sent;
422 if (remainder == 0 || sent != len)
a246b010
CL		 423 goto out;
424 base = 0;
24c5684b
TM		 425 } else
426 base -= xdr->page_len;
427
428 if (base >= xdr->tail[0].iov_len)
f279cd00 429 return 0;
24c5684b 430 err = xs_send_kvec(sock, NULL, 0, &xdr->tail[0], base, 0);
a246b010 431out:
f279cd00
JB		 432 if (err > 0) {
433 *sent_p += err;
434 err = 0;
435 }
436 return err;
a246b010
CL		 437}
438
b6ddf64f
TM		 439static void xs_nospace_callback(struct rpc_task *task)
440{
441 struct sock_xprt *transport = container_of(task->tk_rqstp->rq_xprt, struct sock_xprt, xprt);
442
443 transport->inet->sk_write_pending--;
b6ddf64f
TM		 444}
445
9903cd1c 446/**
262965f5
CL		 447 * xs_nospace - place task on wait queue if transmit was incomplete
448 * @task: task to put to sleep
9903cd1c 449 *
a246b010 450 */
5e3771ce 451static int xs_nospace(struct rpc_task *task)
a246b010 452{
262965f5
CL		 453 struct rpc_rqst *req = task->tk_rqstp;
454 struct rpc_xprt *xprt = req->rq_xprt;
ee0ac0c2 455 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
06ea0bfe 456 struct sock *sk = transport->inet;
24ca9a84 457 int ret = -EAGAIN;
a246b010 458
46121cf7 459 dprintk("RPC: %5u xmit incomplete (%u left of %u)\n",
262965f5
CL		 460 task->tk_pid, req->rq_slen - req->rq_bytes_sent,
461 req->rq_slen);
462
b6ddf64f
TM		 463 /* Protect against races with write_space */
464 spin_lock_bh(&xprt->transport_lock);
465
466 /* Don't race with disconnect */
467 if (xprt_connected(xprt)) {
13331a55
TM		 468 /* wait for more buffer space */
469 sk->sk_write_pending++;
470 xprt_wait_for_buffer_space(task, xs_nospace_callback);
471 } else
5e3771ce 472 ret = -ENOTCONN;
262965f5 473
b6ddf64f 474 spin_unlock_bh(&xprt->transport_lock);
06ea0bfe
TM		 475
476 /* Race breaker in case memory is freed before above code is called */
477 sk->sk_write_space(sk);
5e3771ce 478 return ret;
262965f5
CL		 479}
480
61677eee
CL		 481/*
482 * Construct a stream transport record marker in @buf.
483 */
484static inline void xs_encode_stream_record_marker(struct xdr_buf *buf)
485{
486 u32 reclen = buf->len - sizeof(rpc_fraghdr);
487 rpc_fraghdr *base = buf->head[0].iov_base;
488 *base = cpu_to_be32(RPC_LAST_STREAM_FRAGMENT | reclen);
489}
490
176e21ee
CL		 491/**
492 * xs_local_send_request - write an RPC request to an AF_LOCAL socket
493 * @task: RPC task that manages the state of an RPC request
494 *
495 * Return values:
496 * 0: The request has been sent
497 * EAGAIN: The socket was blocked, please call again later to
498 * complete the request
499 * ENOTCONN: Caller needs to invoke connect logic then call again
500 * other: Some other error occured, the request was not sent
501 */
502static int xs_local_send_request(struct rpc_task *task)
503{
504 struct rpc_rqst *req = task->tk_rqstp;
505 struct rpc_xprt *xprt = req->rq_xprt;
506 struct sock_xprt *transport =
507 container_of(xprt, struct sock_xprt, xprt);
508 struct xdr_buf *xdr = &req->rq_snd_buf;
509 int status;
f279cd00 510 int sent = 0;
176e21ee
CL		 511
512 xs_encode_stream_record_marker(&req->rq_snd_buf);
513
514 xs_pktdump("packet data:",
515 req->rq_svec->iov_base, req->rq_svec->iov_len);
516
f279cd00
JB		 517 status = xs_sendpages(transport->sock, NULL, 0, xdr, req->rq_bytes_sent,
518 true, &sent);
176e21ee
CL		 519 dprintk("RPC: %s(%u) = %d\n",
520 __func__, xdr->len - req->rq_bytes_sent, status);
743c69e7
N		 521
522 if (status == -EAGAIN && sock_writeable(transport->inet))
523 status = -ENOBUFS;
524
f279cd00
JB		 525 if (likely(sent > 0) || status == 0) {
526 req->rq_bytes_sent += sent;
527 req->rq_xmit_bytes_sent += sent;
176e21ee
CL		 528 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
529 req->rq_bytes_sent = 0;
530 return 0;
531 }
532 status = -EAGAIN;
533 }
534
535 switch (status) {
3601c4a9 536 case -ENOBUFS:
b5872f0c 537 break;
176e21ee
CL		 538 case -EAGAIN:
539 status = xs_nospace(task);
540 break;
541 default:
542 dprintk("RPC: sendmsg returned unrecognized error %d\n",
543 -status);
544 case -EPIPE:
545 xs_close(xprt);
546 status = -ENOTCONN;
547 }
548
549 return status;
550}
551
262965f5
CL		 552/**
553 * xs_udp_send_request - write an RPC request to a UDP socket
554 * @task: address of RPC task that manages the state of an RPC request
555 *
556 * Return values:
557 * 0: The request has been sent
558 * EAGAIN: The socket was blocked, please call again later to
559 * complete the request
560 * ENOTCONN: Caller needs to invoke connect logic then call again
25985edc 561 * other: Some other error occurred, the request was not sent
262965f5
CL		 562 */
563static int xs_udp_send_request(struct rpc_task *task)
564{
565 struct rpc_rqst *req = task->tk_rqstp;
566 struct rpc_xprt *xprt = req->rq_xprt;
ee0ac0c2 567 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
262965f5 568 struct xdr_buf *xdr = &req->rq_snd_buf;
f279cd00 569 int sent = 0;
262965f5 570 int status;
a246b010 571
9903cd1c 572 xs_pktdump("packet data:",
a246b010
CL		 573 req->rq_svec->iov_base,
574 req->rq_svec->iov_len);
575
01d37c42
TM		 576 if (!xprt_bound(xprt))
577 return -ENOTCONN;
f279cd00
JB		 578 status = xs_sendpages(transport->sock, xs_addr(xprt), xprt->addrlen,
579 xdr, req->rq_bytes_sent, true, &sent);
a246b010 580
46121cf7 581 dprintk("RPC: xs_udp_send_request(%u) = %d\n",
262965f5 582 xdr->len - req->rq_bytes_sent, status);
a246b010 583
3dedbb5c
JB		 584 /* firewall is blocking us, don't return -EAGAIN or we end up looping */
585 if (status == -EPERM)
586 goto process_status;
587
743c69e7
N		 588 if (status == -EAGAIN && sock_writeable(transport->inet))
589 status = -ENOBUFS;
590
f279cd00
JB		 591 if (sent > 0 || status == 0) {
592 req->rq_xmit_bytes_sent += sent;
593 if (sent >= req->rq_slen)
2199700f
TM		 594 return 0;
595 /* Still some bytes left; set up for a retry later. */
262965f5 596 status = -EAGAIN;
2199700f 597 }
a246b010 598
3dedbb5c 599process_status:
262965f5 600 switch (status) {
fba91afb
TM		 601 case -ENOTSOCK:
602 status = -ENOTCONN;
603 /* Should we call xs_close() here? */
604 break;
b6ddf64f 605 case -EAGAIN:
5e3771ce 606 status = xs_nospace(task);
b6ddf64f 607 break;
262965f5 608 case -ENETUNREACH:
3601c4a9 609 case -ENOBUFS:
262965f5 610 case -EPIPE:
a246b010 611 case -ECONNREFUSED:
3dedbb5c 612 case -EPERM:
a246b010 613 /* When the server has died, an ICMP port unreachable message
9903cd1c 614 * prompts ECONNREFUSED. */
13331a55
TM		 615 break;
616 default:
617 dprintk("RPC: sendmsg returned unrecognized error %d\n",
618 -status);
a246b010 619 }
5fe46e9d 620
262965f5 621 return status;
a246b010
CL		 622}
623
9903cd1c 624/**
262965f5 625 * xs_tcp_send_request - write an RPC request to a TCP socket
9903cd1c
CL		 626 * @task: address of RPC task that manages the state of an RPC request
627 *
628 * Return values:
262965f5
CL		 629 * 0: The request has been sent
630 * EAGAIN: The socket was blocked, please call again later to
631 * complete the request
632 * ENOTCONN: Caller needs to invoke connect logic then call again
25985edc 633 * other: Some other error occurred, the request was not sent
9903cd1c
CL		 634 *
635 * XXX: In the case of soft timeouts, should we eventually give up
262965f5 636 * if sendmsg is not able to make progress?
9903cd1c 637 */
262965f5 638static int xs_tcp_send_request(struct rpc_task *task)
a246b010
CL		 639{
640 struct rpc_rqst *req = task->tk_rqstp;
641 struct rpc_xprt *xprt = req->rq_xprt;
ee0ac0c2 642 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
262965f5 643 struct xdr_buf *xdr = &req->rq_snd_buf;
a6b31d18 644 bool zerocopy = true;
b595bb15 645 int status;
f279cd00 646 int sent;
a246b010 647
61677eee 648 xs_encode_stream_record_marker(&req->rq_snd_buf);
a246b010 649
262965f5
CL		 650 xs_pktdump("packet data:",
651 req->rq_svec->iov_base,
652 req->rq_svec->iov_len);
a6b31d18
TM		 653 /* Don't use zero copy if this is a resend. If the RPC call
654 * completes while the socket holds a reference to the pages,
655 * then we may end up resending corrupted data.
656 */
657 if (task->tk_flags & RPC_TASK_SENT)
658 zerocopy = false;
a246b010
CL		 659
660 /* Continue transmitting the packet/record. We must be careful
661 * to cope with writespace callbacks arriving _after_ we have
262965f5 662 * called sendmsg(). */
a246b010 663 while (1) {
f279cd00
JB		 664 sent = 0;
665 status = xs_sendpages(transport->sock, NULL, 0, xdr,
666 req->rq_bytes_sent, zerocopy, &sent);
a246b010 667
46121cf7 668 dprintk("RPC: xs_tcp_send_request(%u) = %d\n",
262965f5 669 xdr->len - req->rq_bytes_sent, status);
a246b010 670
262965f5
CL		 671 /* If we've sent the entire packet, immediately
672 * reset the count of bytes sent. */
f279cd00
JB		 673 req->rq_bytes_sent += sent;
674 req->rq_xmit_bytes_sent += sent;
262965f5
CL		 675 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
676 req->rq_bytes_sent = 0;
677 return 0;
678 }
a246b010 679
f580dd04
TM		 680 if (status < 0)
681 break;
682 if (sent == 0) {
683 status = -EAGAIN;
684 break;
685 }
a246b010 686 }
743c69e7
N		 687 if (status == -EAGAIN && sk_stream_is_writeable(transport->inet))
688 status = -ENOBUFS;
a246b010 689
262965f5 690 switch (status) {
fba91afb
TM		 691 case -ENOTSOCK:
692 status = -ENOTCONN;
693 /* Should we call xs_close() here? */
694 break;
262965f5 695 case -EAGAIN:
5e3771ce 696 status = xs_nospace(task);
262965f5 697 break;
262965f5 698 case -ECONNRESET:
2a9e1cfa 699 case -ECONNREFUSED:
262965f5 700 case -ENOTCONN:
3913c78c 701 case -EADDRINUSE:
b5872f0c 702 case -ENOBUFS:
b9d2bb2e 703 case -EPIPE:
13331a55
TM		 704 break;
705 default:
706 dprintk("RPC: sendmsg returned unrecognized error %d\n",
707 -status);
a246b010 708 }
5fe46e9d 709
a246b010
CL		 710 return status;
711}
712
e0ab53de
TM		 713/**
714 * xs_tcp_release_xprt - clean up after a tcp transmission
715 * @xprt: transport
716 * @task: rpc task
717 *
718 * This cleans up if an error causes us to abort the transmission of a request.
719 * In this case, the socket may need to be reset in order to avoid confusing
720 * the server.
721 */
722static void xs_tcp_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
723{
724 struct rpc_rqst *req;
725
726 if (task != xprt->snd_task)
727 return;
728 if (task == NULL)
729 goto out_release;
730 req = task->tk_rqstp;
43cedbf0
TM		 731 if (req == NULL)
732 goto out_release;
e0ab53de
TM		 733 if (req->rq_bytes_sent == 0)
734 goto out_release;
735 if (req->rq_bytes_sent == req->rq_snd_buf.len)
736 goto out_release;
a4f0835c 737 set_bit(XPRT_CLOSE_WAIT, &xprt->state);
e0ab53de
TM		 738out_release:
739 xprt_release_xprt(xprt, task);
740}
741
2a9e1cfa
TM		 742static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk)
743{
744 transport->old_data_ready = sk->sk_data_ready;
745 transport->old_state_change = sk->sk_state_change;
746 transport->old_write_space = sk->sk_write_space;
2118071d 747 transport->old_error_report = sk->sk_error_report;
2a9e1cfa
TM		 748}
749
750static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk)
751{
752 sk->sk_data_ready = transport->old_data_ready;
753 sk->sk_state_change = transport->old_state_change;
754 sk->sk_write_space = transport->old_write_space;
2118071d
TM		 755 sk->sk_error_report = transport->old_error_report;
756}
757
42d42a5b
TM		 758static void xs_sock_reset_state_flags(struct rpc_xprt *xprt)
759{
760 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
761
762 clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state);
763}
764
b70ae915
TM		 765static void xs_sock_reset_connection_flags(struct rpc_xprt *xprt)
766{
767 smp_mb__before_atomic();
768 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
769 clear_bit(XPRT_CLOSING, &xprt->state);
42d42a5b 770 xs_sock_reset_state_flags(xprt);
b70ae915
TM		 771 smp_mb__after_atomic();
772}
773
774static void xs_sock_mark_closed(struct rpc_xprt *xprt)
775{
776 xs_sock_reset_connection_flags(xprt);
777 /* Mark transport as closed and wake up all pending tasks */
778 xprt_disconnect_done(xprt);
779}
780
2118071d
TM		 781/**
782 * xs_error_report - callback to handle TCP socket state errors
783 * @sk: socket
784 *
785 * Note: we don't call sock_error() since there may be a rpc_task
786 * using the socket, and so we don't want to clear sk->sk_err.
787 */
788static void xs_error_report(struct sock *sk)
789{
790 struct rpc_xprt *xprt;
791 int err;
792
793 read_lock_bh(&sk->sk_callback_lock);
794 if (!(xprt = xprt_from_sock(sk)))
795 goto out;
796
797 err = -sk->sk_err;
798 if (err == 0)
799 goto out;
b70ae915
TM		 800 /* Is this a reset event? */
801 if (sk->sk_state == TCP_CLOSE)
802 xs_sock_mark_closed(xprt);
2118071d
TM		 803 dprintk("RPC: xs_error_report client %p, error=%d...\n",
804 xprt, -err);
e8353c76 805 trace_rpc_socket_error(xprt, sk->sk_socket, err);
2118071d
TM		 806 xprt_wake_pending_tasks(xprt, err);
807 out:
808 read_unlock_bh(&sk->sk_callback_lock);
2a9e1cfa
TM		 809}
810
fe315e76 811static void xs_reset_transport(struct sock_xprt *transport)
a246b010 812{
ee0ac0c2
CL		 813 struct socket *sock = transport->sock;
814 struct sock *sk = transport->inet;
6cc7e908 815 struct rpc_xprt *xprt = &transport->xprt;
a246b010 816
fe315e76
CL		 817 if (sk == NULL)
818 return;
9903cd1c 819
264d1df3
JL		 820 if (atomic_read(&transport->xprt.swapper))
821 sk_clear_memalloc(sk);
822
09939204
TM		 823 kernel_sock_shutdown(sock, SHUT_RDWR);
824
edc1b01c 825 mutex_lock(&transport->recv_mutex);
a246b010 826 write_lock_bh(&sk->sk_callback_lock);
ee0ac0c2
CL		 827 transport->inet = NULL;
828 transport->sock = NULL;
a246b010 829
9903cd1c 830 sk->sk_user_data = NULL;
2a9e1cfa
TM		 831
832 xs_restore_old_callbacks(transport, sk);
0c78789e 833 xprt_clear_connected(xprt);
a246b010 834 write_unlock_bh(&sk->sk_callback_lock);
6cc7e908 835 xs_sock_reset_connection_flags(xprt);
edc1b01c 836 mutex_unlock(&transport->recv_mutex);
a246b010 837
6cc7e908 838 trace_rpc_socket_close(xprt, sock);
a246b010 839 sock_release(sock);
fe315e76
CL		 840}
841
842/**
843 * xs_close - close a socket
844 * @xprt: transport
845 *
846 * This is used when all requests are complete; ie, no DRC state remains
847 * on the server we want to save.
f75e6745
TM		 848 *
849 * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with
850 * xs_reset_transport() zeroing the socket from underneath a writer.
fe315e76
CL		 851 */
852static void xs_close(struct rpc_xprt *xprt)
853{
854 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
855
856 dprintk("RPC: xs_close xprt %p\n", xprt);
857
858 xs_reset_transport(transport);
61d0a8e6 859 xprt->reestablish_timeout = 0;
fe315e76 860
62da3b24 861 xprt_disconnect_done(xprt);
a246b010
CL		 862}
863
4a068258
CL		 864static void xs_inject_disconnect(struct rpc_xprt *xprt)
865{
866 dprintk("RPC: injecting transport disconnect on xprt=%p\n",
867 xprt);
868 xprt_disconnect_done(xprt);
869}
870
315f3812
KM		 871static void xs_xprt_free(struct rpc_xprt *xprt)
872{
873 xs_free_peer_addresses(xprt);
874 xprt_free(xprt);
875}
876
9903cd1c
CL		 877/**
878 * xs_destroy - prepare to shutdown a transport
879 * @xprt: doomed transport
880 *
881 */
882static void xs_destroy(struct rpc_xprt *xprt)
a246b010 883{
03c78827
TM		 884 struct sock_xprt *transport = container_of(xprt,
885 struct sock_xprt, xprt);
46121cf7 886 dprintk("RPC: xs_destroy xprt %p\n", xprt);
9903cd1c 887
03c78827 888 cancel_delayed_work_sync(&transport->connect_worker);
a1311d87 889 xs_close(xprt);
edc1b01c 890 cancel_work_sync(&transport->recv_worker);
315f3812 891 xs_xprt_free(xprt);
a1311d87 892 module_put(THIS_MODULE);
a246b010
CL		 893}
894
176e21ee
CL		 895static int xs_local_copy_to_xdr(struct xdr_buf *xdr, struct sk_buff *skb)
896{
897 struct xdr_skb_reader desc = {
898 .skb = skb,
899 .offset = sizeof(rpc_fraghdr),
900 .count = skb->len - sizeof(rpc_fraghdr),
901 };
902
903 if (xdr_partial_copy_from_skb(xdr, 0, &desc, xdr_skb_read_bits) < 0)
904 return -1;
905 if (desc.count)
906 return -1;
907 return 0;
908}
909
910/**
a2648094
TM		 911 * xs_local_data_read_skb
912 * @xprt: transport
913 * @sk: socket
914 * @skb: skbuff
176e21ee
CL		 915 *
916 * Currently this assumes we can read the whole reply in a single gulp.
917 */
a2648094
TM		 918static void xs_local_data_read_skb(struct rpc_xprt *xprt,
919 struct sock *sk,
920 struct sk_buff *skb)
176e21ee
CL		 921{
922 struct rpc_task *task;
176e21ee 923 struct rpc_rqst *rovr;
a2648094 924 int repsize, copied;
176e21ee
CL		 925 u32 _xid;
926 __be32 *xp;
927
176e21ee
CL		 928 repsize = skb->len - sizeof(rpc_fraghdr);
929 if (repsize < 4) {
930 dprintk("RPC: impossible RPC reply size %d\n", repsize);
a2648094 931 return;
176e21ee
CL		 932 }
933
934 /* Copy the XID from the skb... */
935 xp = skb_header_pointer(skb, sizeof(rpc_fraghdr), sizeof(_xid), &_xid);
936 if (xp == NULL)
a2648094 937 return;
176e21ee
CL		 938
939 /* Look up and lock the request corresponding to the given XID */
a2648094 940 spin_lock_bh(&xprt->transport_lock);
176e21ee
CL		 941 rovr = xprt_lookup_rqst(xprt, *xp);
942 if (!rovr)
943 goto out_unlock;
944 task = rovr->rq_task;
945
946 copied = rovr->rq_private_buf.buflen;
947 if (copied > repsize)
948 copied = repsize;
949
950 if (xs_local_copy_to_xdr(&rovr->rq_private_buf, skb)) {
951 dprintk("RPC: sk_buff copy failed\n");
952 goto out_unlock;
953 }
954
955 xprt_complete_rqst(task, copied);
956
957 out_unlock:
a2648094
TM		 958 spin_unlock_bh(&xprt->transport_lock);
959}
960
961static void xs_local_data_receive(struct sock_xprt *transport)
962{
963 struct sk_buff *skb;
964 struct sock *sk;
965 int err;
966
967 mutex_lock(&transport->recv_mutex);
968 sk = transport->inet;
969 if (sk == NULL)
970 goto out;
971 for (;;) {
972 skb = skb_recv_datagram(sk, 0, 1, &err);
42d42a5b
TM		 973 if (skb != NULL) {
974 xs_local_data_read_skb(&transport->xprt, sk, skb);
975 skb_free_datagram(sk, skb);
976 continue;
977 }
978 if (!test_and_clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
a2648094 979 break;
a2648094
TM		 980 }
981out:
982 mutex_unlock(&transport->recv_mutex);
983}
984
985static void xs_local_data_receive_workfn(struct work_struct *work)
986{
987 struct sock_xprt *transport =
988 container_of(work, struct sock_xprt, recv_worker);
989 xs_local_data_receive(transport);
176e21ee
CL		 990}
991
9903cd1c 992/**
f9b2ee71
TM		 993 * xs_udp_data_read_skb - receive callback for UDP sockets
994 * @xprt: transport
995 * @sk: socket
996 * @skb: skbuff
9903cd1c 997 *
a246b010 998 */
f9b2ee71
TM		 999static void xs_udp_data_read_skb(struct rpc_xprt *xprt,
1000 struct sock *sk,
1001 struct sk_buff *skb)
a246b010 1002{
9903cd1c 1003 struct rpc_task *task;
a246b010 1004 struct rpc_rqst *rovr;
f9b2ee71 1005 int repsize, copied;
d8ed029d
AD		 1006 u32 _xid;
1007 __be32 *xp;
a246b010 1008
1da8c681 1009 repsize = skb->len;
a246b010 1010 if (repsize < 4) {
46121cf7 1011 dprintk("RPC: impossible RPC reply size %d!\n", repsize);
f9b2ee71 1012 return;
a246b010
CL		 1013 }
1014
1015 /* Copy the XID from the skb... */
1da8c681 1016 xp = skb_header_pointer(skb, 0, sizeof(_xid), &_xid);
a246b010 1017 if (xp == NULL)
f9b2ee71 1018 return;
a246b010
CL		 1019
1020 /* Look up and lock the request corresponding to the given XID */
f9b2ee71 1021 spin_lock_bh(&xprt->transport_lock);
a246b010
CL		 1022 rovr = xprt_lookup_rqst(xprt, *xp);
1023 if (!rovr)
1024 goto out_unlock;
1025 task = rovr->rq_task;
1026
a246b010
CL		 1027 if ((copied = rovr->rq_private_buf.buflen) > repsize)
1028 copied = repsize;
1029
1030 /* Suck it into the iovec, verify checksum if not done by hw. */
1781f7f5 1031 if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
02c22347 1032 __UDPX_INC_STATS(sk, UDP_MIB_INERRORS);
a246b010 1033 goto out_unlock;
1781f7f5
HX		 1034 }
1035
02c22347 1036 __UDPX_INC_STATS(sk, UDP_MIB_INDATAGRAMS);
a246b010 1037
6a24dfb6 1038 xprt_adjust_cwnd(xprt, task, copied);
1570c1e4 1039 xprt_complete_rqst(task, copied);
a246b010
CL		 1040
1041 out_unlock:
f9b2ee71
TM		 1042 spin_unlock_bh(&xprt->transport_lock);
1043}
1044
1045static void xs_udp_data_receive(struct sock_xprt *transport)
1046{
1047 struct sk_buff *skb;
1048 struct sock *sk;
1049 int err;
1050
1051 mutex_lock(&transport->recv_mutex);
1052 sk = transport->inet;
1053 if (sk == NULL)
1054 goto out;
1055 for (;;) {
1056 skb = skb_recv_datagram(sk, 0, 1, &err);
42d42a5b
TM		 1057 if (skb != NULL) {
1058 xs_udp_data_read_skb(&transport->xprt, sk, skb);
1059 skb_free_datagram(sk, skb);
1060 continue;
1061 }
1062 if (!test_and_clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
f9b2ee71 1063 break;
f9b2ee71
TM		 1064 }
1065out:
1066 mutex_unlock(&transport->recv_mutex);
1067}
1068
1069static void xs_udp_data_receive_workfn(struct work_struct *work)
1070{
1071 struct sock_xprt *transport =
1072 container_of(work, struct sock_xprt, recv_worker);
1073 xs_udp_data_receive(transport);
1074}
1075
1076/**
1077 * xs_data_ready - "data ready" callback for UDP sockets
1078 * @sk: socket with data to read
1079 *
1080 */
1081static void xs_data_ready(struct sock *sk)
1082{
1083 struct rpc_xprt *xprt;
1084
1085 read_lock_bh(&sk->sk_callback_lock);
1086 dprintk("RPC: xs_data_ready...\n");
1087 xprt = xprt_from_sock(sk);
1088 if (xprt != NULL) {
1089 struct sock_xprt *transport = container_of(xprt,
1090 struct sock_xprt, xprt);
42d42a5b
TM		 1091 if (!test_and_set_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
1092 queue_work(rpciod_workqueue, &transport->recv_worker);
f9b2ee71 1093 }
f064af1e 1094 read_unlock_bh(&sk->sk_callback_lock);
a246b010
CL		 1095}
1096
a519fc7a
TM		 1097/*
1098 * Helper function to force a TCP close if the server is sending
1099 * junk and/or it has put us in CLOSE_WAIT
1100 */
1101static void xs_tcp_force_close(struct rpc_xprt *xprt)
1102{
a519fc7a
TM		 1103 xprt_force_disconnect(xprt);
1104}
1105
dd456471 1106static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
a246b010 1107{
51971139 1108 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
a246b010
CL		 1109 size_t len, used;
1110 char *p;
1111
51971139
CL		 1112 p = ((char *) &transport->tcp_fraghdr) + transport->tcp_offset;
1113 len = sizeof(transport->tcp_fraghdr) - transport->tcp_offset;
9d292316 1114 used = xdr_skb_read_bits(desc, p, len);
51971139 1115 transport->tcp_offset += used;
a246b010
CL		 1116 if (used != len)
1117 return;
808012fb 1118
51971139
CL		 1119 transport->tcp_reclen = ntohl(transport->tcp_fraghdr);
1120 if (transport->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
e136d092 1121 transport->tcp_flags |= TCP_RCV_LAST_FRAG;
a246b010 1122 else
e136d092 1123 transport->tcp_flags &= ~TCP_RCV_LAST_FRAG;
51971139 1124 transport->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
808012fb 1125
e136d092 1126 transport->tcp_flags &= ~TCP_RCV_COPY_FRAGHDR;
51971139 1127 transport->tcp_offset = 0;
808012fb 1128
a246b010 1129 /* Sanity check of the record length */
18dca02a 1130 if (unlikely(transport->tcp_reclen < 8)) {
46121cf7 1131 dprintk("RPC: invalid TCP record fragment length\n");
a519fc7a 1132 xs_tcp_force_close(xprt);
9903cd1c 1133 return;
a246b010 1134 }
46121cf7 1135 dprintk("RPC: reading TCP record fragment of length %d\n",
51971139 1136 transport->tcp_reclen);
a246b010
CL		 1137}
1138
51971139 1139static void xs_tcp_check_fraghdr(struct sock_xprt *transport)
a246b010 1140{
51971139 1141 if (transport->tcp_offset == transport->tcp_reclen) {
e136d092 1142 transport->tcp_flags |= TCP_RCV_COPY_FRAGHDR;
51971139 1143 transport->tcp_offset = 0;
e136d092
CL		 1144 if (transport->tcp_flags & TCP_RCV_LAST_FRAG) {
1145 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1146 transport->tcp_flags |= TCP_RCV_COPY_XID;
51971139 1147 transport->tcp_copied = 0;
a246b010
CL		 1148 }
1149 }
1150}
1151
dd456471 1152static inline void xs_tcp_read_xid(struct sock_xprt *transport, struct xdr_skb_reader *desc)
a246b010
CL		 1153{
1154 size_t len, used;
1155 char *p;
1156
51971139 1157 len = sizeof(transport->tcp_xid) - transport->tcp_offset;
46121cf7 1158 dprintk("RPC: reading XID (%Zu bytes)\n", len);
51971139 1159 p = ((char *) &transport->tcp_xid) + transport->tcp_offset;
9d292316 1160 used = xdr_skb_read_bits(desc, p, len);
51971139 1161 transport->tcp_offset += used;
a246b010
CL		 1162 if (used != len)
1163 return;
e136d092 1164 transport->tcp_flags &= ~TCP_RCV_COPY_XID;
f4a2e418 1165 transport->tcp_flags |= TCP_RCV_READ_CALLDIR;
51971139 1166 transport->tcp_copied = 4;
18dca02a
RL		 1167 dprintk("RPC: reading %s XID %08x\n",
1168 (transport->tcp_flags & TCP_RPC_REPLY) ? "reply for"
1169 : "request with",
51971139
CL		 1170 ntohl(transport->tcp_xid));
1171 xs_tcp_check_fraghdr(transport);
a246b010
CL		 1172}
1173
18dca02a
RL		 1174static inline void xs_tcp_read_calldir(struct sock_xprt *transport,
1175 struct xdr_skb_reader *desc)
a246b010 1176{
18dca02a
RL		 1177 size_t len, used;
1178 u32 offset;
b76ce561 1179 char *p;
18dca02a
RL		 1180
1181 /*
1182 * We want transport->tcp_offset to be 8 at the end of this routine
1183 * (4 bytes for the xid and 4 bytes for the call/reply flag).
1184 * When this function is called for the first time,
1185 * transport->tcp_offset is 4 (after having already read the xid).
1186 */
1187 offset = transport->tcp_offset - sizeof(transport->tcp_xid);
b76ce561 1188 len = sizeof(transport->tcp_calldir) - offset;
18dca02a 1189 dprintk("RPC: reading CALL/REPLY flag (%Zu bytes)\n", len);
b76ce561
TM		 1190 p = ((char *) &transport->tcp_calldir) + offset;
1191 used = xdr_skb_read_bits(desc, p, len);
18dca02a
RL		 1192 transport->tcp_offset += used;
1193 if (used != len)
1194 return;
f4a2e418 1195 transport->tcp_flags &= ~TCP_RCV_READ_CALLDIR;
f4a2e418
RL		 1196 /*
1197 * We don't yet have the XDR buffer, so we will write the calldir
1198 * out after we get the buffer from the 'struct rpc_rqst'
1199 */
b76ce561
TM		 1200 switch (ntohl(transport->tcp_calldir)) {
1201 case RPC_REPLY:
1202 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1203 transport->tcp_flags |= TCP_RCV_COPY_DATA;
18dca02a 1204 transport->tcp_flags |= TCP_RPC_REPLY;
b76ce561
TM		 1205 break;
1206 case RPC_CALL:
1207 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1208 transport->tcp_flags |= TCP_RCV_COPY_DATA;
18dca02a 1209 transport->tcp_flags &= ~TCP_RPC_REPLY;
b76ce561
TM		 1210 break;
1211 default:
1212 dprintk("RPC: invalid request message type\n");
a519fc7a 1213 xs_tcp_force_close(&transport->xprt);
b76ce561 1214 }
18dca02a
RL		 1215 xs_tcp_check_fraghdr(transport);
1216}
1217
44b98efd
RL		 1218static inline void xs_tcp_read_common(struct rpc_xprt *xprt,
1219 struct xdr_skb_reader *desc,
1220 struct rpc_rqst *req)
a246b010 1221{
44b98efd
RL		 1222 struct sock_xprt *transport =
1223 container_of(xprt, struct sock_xprt, xprt);
a246b010
CL		 1224 struct xdr_buf *rcvbuf;
1225 size_t len;
1226 ssize_t r;
1227
a246b010 1228 rcvbuf = &req->rq_private_buf;
f4a2e418
RL		 1229
1230 if (transport->tcp_flags & TCP_RCV_COPY_CALLDIR) {
1231 /*
1232 * Save the RPC direction in the XDR buffer
1233 */
f4a2e418 1234 memcpy(rcvbuf->head[0].iov_base + transport->tcp_copied,
b76ce561
TM		 1235 &transport->tcp_calldir,
1236 sizeof(transport->tcp_calldir));
1237 transport->tcp_copied += sizeof(transport->tcp_calldir);
f4a2e418 1238 transport->tcp_flags &= ~TCP_RCV_COPY_CALLDIR;
a246b010
CL		 1239 }
1240
a246b010 1241 len = desc->count;
51971139 1242 if (len > transport->tcp_reclen - transport->tcp_offset) {
dd456471 1243 struct xdr_skb_reader my_desc;
a246b010 1244
51971139 1245 len = transport->tcp_reclen - transport->tcp_offset;
a246b010
CL		 1246 memcpy(&my_desc, desc, sizeof(my_desc));
1247 my_desc.count = len;
51971139 1248 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
9d292316 1249 &my_desc, xdr_skb_read_bits);
a246b010
CL		 1250 desc->count -= r;
1251 desc->offset += r;
1252 } else
51971139 1253 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
9d292316 1254 desc, xdr_skb_read_bits);
a246b010
CL		 1255
1256 if (r > 0) {
51971139
CL		 1257 transport->tcp_copied += r;
1258 transport->tcp_offset += r;
a246b010
CL		 1259 }
1260 if (r != len) {
1261 /* Error when copying to the receive buffer,
1262 * usually because we weren't able to allocate
1263 * additional buffer pages. All we can do now
e136d092 1264 * is turn off TCP_RCV_COPY_DATA, so the request
a246b010
CL		 1265 * will not receive any additional updates,
1266 * and time out.
1267 * Any remaining data from this record will
1268 * be discarded.
1269 */
e136d092 1270 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
46121cf7 1271 dprintk("RPC: XID %08x truncated request\n",
51971139 1272 ntohl(transport->tcp_xid));
46121cf7
CL		 1273 dprintk("RPC: xprt = %p, tcp_copied = %lu, "
1274 "tcp_offset = %u, tcp_reclen = %u\n",
1275 xprt, transport->tcp_copied,
1276 transport->tcp_offset, transport->tcp_reclen);
44b98efd 1277 return;
a246b010
CL		 1278 }
1279
46121cf7 1280 dprintk("RPC: XID %08x read %Zd bytes\n",
51971139 1281 ntohl(transport->tcp_xid), r);
46121cf7
CL		 1282 dprintk("RPC: xprt = %p, tcp_copied = %lu, tcp_offset = %u, "
1283 "tcp_reclen = %u\n", xprt, transport->tcp_copied,
1284 transport->tcp_offset, transport->tcp_reclen);
51971139
CL		 1285
1286 if (transport->tcp_copied == req->rq_private_buf.buflen)
e136d092 1287 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
51971139 1288 else if (transport->tcp_offset == transport->tcp_reclen) {
e136d092
CL		 1289 if (transport->tcp_flags & TCP_RCV_LAST_FRAG)
1290 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
a246b010 1291 }
44b98efd
RL		 1292}
1293
1294/*
1295 * Finds the request corresponding to the RPC xid and invokes the common
1296 * tcp read code to read the data.
1297 */
1298static inline int xs_tcp_read_reply(struct rpc_xprt *xprt,
1299 struct xdr_skb_reader *desc)
1300{
1301 struct sock_xprt *transport =
1302 container_of(xprt, struct sock_xprt, xprt);
1303 struct rpc_rqst *req;
1304
1305 dprintk("RPC: read reply XID %08x\n", ntohl(transport->tcp_xid));
1306
1307 /* Find and lock the request corresponding to this xid */
edc1b01c 1308 spin_lock_bh(&xprt->transport_lock);
44b98efd
RL		 1309 req = xprt_lookup_rqst(xprt, transport->tcp_xid);
1310 if (!req) {
1311 dprintk("RPC: XID %08x request not found!\n",
1312 ntohl(transport->tcp_xid));
edc1b01c 1313 spin_unlock_bh(&xprt->transport_lock);
44b98efd
RL		 1314 return -1;
1315 }
1316
1317 xs_tcp_read_common(xprt, desc, req);
1318
e136d092 1319 if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
51971139 1320 xprt_complete_rqst(req->rq_task, transport->tcp_copied);
44b98efd 1321
edc1b01c 1322 spin_unlock_bh(&xprt->transport_lock);
44b98efd
RL		 1323 return 0;
1324}
1325
9e00abc3 1326#if defined(CONFIG_SUNRPC_BACKCHANNEL)
44b98efd
RL		 1327/*
1328 * Obtains an rpc_rqst previously allocated and invokes the common
1329 * tcp read code to read the data. The result is placed in the callback
1330 * queue.
1331 * If we're unable to obtain the rpc_rqst we schedule the closing of the
1332 * connection and return -1.
1333 */
2ea24497 1334static int xs_tcp_read_callback(struct rpc_xprt *xprt,
44b98efd
RL		 1335 struct xdr_skb_reader *desc)
1336{
1337 struct sock_xprt *transport =
1338 container_of(xprt, struct sock_xprt, xprt);
1339 struct rpc_rqst *req;
1340
2ea24497 1341 /* Look up and lock the request corresponding to the given XID */
edc1b01c 1342 spin_lock_bh(&xprt->transport_lock);
2ea24497 1343 req = xprt_lookup_bc_request(xprt, transport->tcp_xid);
44b98efd 1344 if (req == NULL) {
edc1b01c 1345 spin_unlock_bh(&xprt->transport_lock);
44b98efd
RL		 1346 printk(KERN_WARNING "Callback slot table overflowed\n");
1347 xprt_force_disconnect(xprt);
1348 return -1;
1349 }
1350
44b98efd
RL		 1351 dprintk("RPC: read callback XID %08x\n", ntohl(req->rq_xid));
1352 xs_tcp_read_common(xprt, desc, req);
1353
2ea24497
TM		 1354 if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1355 xprt_complete_bc_request(req, transport->tcp_copied);
edc1b01c 1356 spin_unlock_bh(&xprt->transport_lock);
44b98efd
RL		 1357
1358 return 0;
1359}
1360
1361static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1362 struct xdr_skb_reader *desc)
1363{
1364 struct sock_xprt *transport =
1365 container_of(xprt, struct sock_xprt, xprt);
1366
1367 return (transport->tcp_flags & TCP_RPC_REPLY) ?
1368 xs_tcp_read_reply(xprt, desc) :
1369 xs_tcp_read_callback(xprt, desc);
1370}
76566773
CL		 1371
1372static int xs_tcp_bc_up(struct svc_serv *serv, struct net *net)
1373{
1374 int ret;
1375
1376 ret = svc_create_xprt(serv, "tcp-bc", net, PF_INET, 0,
1377 SVC_SOCK_ANONYMOUS);
1378 if (ret < 0)
1379 return ret;
1380 return 0;
1381}
6b26cc8c
CL		 1382
1383static size_t xs_tcp_bc_maxpayload(struct rpc_xprt *xprt)
1384{
1385 return PAGE_SIZE;
1386}
44b98efd
RL		 1387#else
1388static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1389 struct xdr_skb_reader *desc)
1390{
1391 return xs_tcp_read_reply(xprt, desc);
1392}
9e00abc3 1393#endif /* CONFIG_SUNRPC_BACKCHANNEL */
44b98efd
RL		 1394
1395/*
1396 * Read data off the transport. This can be either an RPC_CALL or an
1397 * RPC_REPLY. Relay the processing to helper functions.
1398 */
1399static void xs_tcp_read_data(struct rpc_xprt *xprt,
1400 struct xdr_skb_reader *desc)
1401{
1402 struct sock_xprt *transport =
1403 container_of(xprt, struct sock_xprt, xprt);
1404
1405 if (_xs_tcp_read_data(xprt, desc) == 0)
1406 xs_tcp_check_fraghdr(transport);
1407 else {
1408 /*
1409 * The transport_lock protects the request handling.
1410 * There's no need to hold it to update the tcp_flags.
1411 */
1412 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1413 }
a246b010
CL		 1414}
1415
dd456471 1416static inline void xs_tcp_read_discard(struct sock_xprt *transport, struct xdr_skb_reader *desc)
a246b010
CL		 1417{
1418 size_t len;
1419
51971139 1420 len = transport->tcp_reclen - transport->tcp_offset;
a246b010
CL		 1421 if (len > desc->count)
1422 len = desc->count;
1423 desc->count -= len;
1424 desc->offset += len;
51971139 1425 transport->tcp_offset += len;
46121cf7 1426 dprintk("RPC: discarded %Zu bytes\n", len);
51971139 1427 xs_tcp_check_fraghdr(transport);
a246b010
CL		 1428}
1429
9903cd1c 1430static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
a246b010
CL		 1431{
1432 struct rpc_xprt *xprt = rd_desc->arg.data;
51971139 1433 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
dd456471 1434 struct xdr_skb_reader desc = {
a246b010
CL		 1435 .skb = skb,
1436 .offset = offset,
1437 .count = len,
9903cd1c 1438 };
a246b010 1439
46121cf7 1440 dprintk("RPC: xs_tcp_data_recv started\n");
a246b010 1441 do {
1a867a08 1442 trace_xs_tcp_data_recv(transport);
a246b010
CL		 1443 /* Read in a new fragment marker if necessary */
1444 /* Can we ever really expect to get completely empty fragments? */
e136d092 1445 if (transport->tcp_flags & TCP_RCV_COPY_FRAGHDR) {
9903cd1c 1446 xs_tcp_read_fraghdr(xprt, &desc);
a246b010
CL		 1447 continue;
1448 }
1449 /* Read in the xid if necessary */
e136d092 1450 if (transport->tcp_flags & TCP_RCV_COPY_XID) {
51971139 1451 xs_tcp_read_xid(transport, &desc);
a246b010
CL		 1452 continue;
1453 }
18dca02a 1454 /* Read in the call/reply flag */
f4a2e418 1455 if (transport->tcp_flags & TCP_RCV_READ_CALLDIR) {
18dca02a
RL		 1456 xs_tcp_read_calldir(transport, &desc);
1457 continue;
1458 }
a246b010 1459 /* Read in the request data */
e136d092 1460 if (transport->tcp_flags & TCP_RCV_COPY_DATA) {
44b98efd 1461 xs_tcp_read_data(xprt, &desc);
a246b010
CL		 1462 continue;
1463 }
1464 /* Skip over any trailing bytes on short reads */
51971139 1465 xs_tcp_read_discard(transport, &desc);
a246b010 1466 } while (desc.count);
1a867a08 1467 trace_xs_tcp_data_recv(transport);
46121cf7 1468 dprintk("RPC: xs_tcp_data_recv done\n");
a246b010
CL		 1469 return len - desc.count;
1470}
1471
66d7a56a
TM		 1472static void xs_tcp_data_receive(struct sock_xprt *transport)
1473{
1474 struct rpc_xprt *xprt = &transport->xprt;
1475 struct sock *sk;
1476 read_descriptor_t rd_desc = {
1477 .count = 2*1024*1024,
1478 .arg.data = xprt,
1479 };
1480 unsigned long total = 0;
1481 int read = 0;
1482
edc1b01c 1483 mutex_lock(&transport->recv_mutex);
66d7a56a 1484 sk = transport->inet;
edc1b01c
TM		 1485 if (sk == NULL)
1486 goto out;
66d7a56a
TM		 1487
1488 /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
1489 for (;;) {
edc1b01c 1490 lock_sock(sk);
66d7a56a 1491 read = tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
42d42a5b
TM		 1492 if (read <= 0) {
1493 clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state);
1494 release_sock(sk);
1495 if (!test_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
1496 break;
1497 } else {
1498 release_sock(sk);
1499 total += read;
1500 }
66d7a56a
TM		 1501 rd_desc.count = 65536;
1502 }
edc1b01c
TM		 1503out:
1504 mutex_unlock(&transport->recv_mutex);
66d7a56a
TM		 1505 trace_xs_tcp_data_ready(xprt, read, total);
1506}
1507
edc1b01c
TM		 1508static void xs_tcp_data_receive_workfn(struct work_struct *work)
1509{
1510 struct sock_xprt *transport =
1511 container_of(work, struct sock_xprt, recv_worker);
1512 xs_tcp_data_receive(transport);
1513}
1514
9903cd1c
CL		 1515/**
1516 * xs_tcp_data_ready - "data ready" callback for TCP sockets
1517 * @sk: socket with data to read
9903cd1c
CL		 1518 *
1519 */
676d2369 1520static void xs_tcp_data_ready(struct sock *sk)
a246b010 1521{
66d7a56a 1522 struct sock_xprt *transport;
a246b010 1523 struct rpc_xprt *xprt;
a246b010 1524
46121cf7
CL		 1525 dprintk("RPC: xs_tcp_data_ready...\n");
1526
f064af1e 1527 read_lock_bh(&sk->sk_callback_lock);
66d7a56a 1528 if (!(xprt = xprt_from_sock(sk)))
a246b010 1529 goto out;
66d7a56a 1530 transport = container_of(xprt, struct sock_xprt, xprt);
42d42a5b
TM		 1531 if (test_and_set_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
1532 goto out;
66d7a56a 1533
61d0a8e6
NB		 1534 /* Any data means we had a useful conversation, so
1535 * the we don't need to delay the next reconnect
1536 */
1537 if (xprt->reestablish_timeout)
1538 xprt->reestablish_timeout = 0;
edc1b01c 1539 queue_work(rpciod_workqueue, &transport->recv_worker);
61d0a8e6 1540
a246b010 1541out:
f064af1e 1542 read_unlock_bh(&sk->sk_callback_lock);
a246b010
CL		 1543}
1544
9903cd1c
CL		 1545/**
1546 * xs_tcp_state_change - callback to handle TCP socket state changes
1547 * @sk: socket whose state has changed
1548 *
1549 */
1550static void xs_tcp_state_change(struct sock *sk)
a246b010 1551{
9903cd1c 1552 struct rpc_xprt *xprt;
0fdea1e8 1553 struct sock_xprt *transport;
a246b010 1554
f064af1e 1555 read_lock_bh(&sk->sk_callback_lock);
a246b010
CL		 1556 if (!(xprt = xprt_from_sock(sk)))
1557 goto out;
46121cf7 1558 dprintk("RPC: xs_tcp_state_change client %p...\n", xprt);
669502ff 1559 dprintk("RPC: state %x conn %d dead %d zapped %d sk_shutdown %d\n",
46121cf7
CL		 1560 sk->sk_state, xprt_connected(xprt),
1561 sock_flag(sk, SOCK_DEAD),
669502ff
AC		 1562 sock_flag(sk, SOCK_ZAPPED),
1563 sk->sk_shutdown);
a246b010 1564
0fdea1e8 1565 transport = container_of(xprt, struct sock_xprt, xprt);
40b5ea0c 1566 trace_rpc_socket_state_change(xprt, sk->sk_socket);
a246b010
CL		 1567 switch (sk->sk_state) {
1568 case TCP_ESTABLISHED:
f064af1e 1569 spin_lock(&xprt->transport_lock);
a246b010 1570 if (!xprt_test_and_set_connected(xprt)) {
51971139 1571
a246b010 1572 /* Reset TCP record info */
51971139
CL		 1573 transport->tcp_offset = 0;
1574 transport->tcp_reclen = 0;
1575 transport->tcp_copied = 0;
e136d092
CL		 1576 transport->tcp_flags =
1577 TCP_RCV_COPY_FRAGHDR | TCP_RCV_COPY_XID;
8b71798c 1578 xprt->connect_cookie++;
0fdea1e8
TM		 1579 clear_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
1580 xprt_clear_connecting(xprt);
51971139 1581
2a491991 1582 xprt_wake_pending_tasks(xprt, -EAGAIN);
a246b010 1583 }
f064af1e 1584 spin_unlock(&xprt->transport_lock);
a246b010 1585 break;
3b948ae5
TM		 1586 case TCP_FIN_WAIT1:
1587 /* The client initiated a shutdown of the socket */
7c1d71cf 1588 xprt->connect_cookie++;
663b8858 1589 xprt->reestablish_timeout = 0;
3b948ae5 1590 set_bit(XPRT_CLOSING, &xprt->state);
4e857c58 1591 smp_mb__before_atomic();
3b948ae5 1592 clear_bit(XPRT_CONNECTED, &xprt->state);
ef803670 1593 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
4e857c58 1594 smp_mb__after_atomic();
a246b010 1595 break;
632e3bdc 1596 case TCP_CLOSE_WAIT:
3b948ae5 1597 /* The server initiated a shutdown of the socket */
7c1d71cf 1598 xprt->connect_cookie++;
d0bea455 1599 clear_bit(XPRT_CONNECTED, &xprt->state);
a519fc7a 1600 xs_tcp_force_close(xprt);
663b8858
TM		 1601 case TCP_CLOSING:
1602 /*
1603 * If the server closed down the connection, make sure that
1604 * we back off before reconnecting
1605 */
1606 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1607 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
3b948ae5
TM		 1608 break;
1609 case TCP_LAST_ACK:
670f9457 1610 set_bit(XPRT_CLOSING, &xprt->state);
4e857c58 1611 smp_mb__before_atomic();
3b948ae5 1612 clear_bit(XPRT_CONNECTED, &xprt->state);
4e857c58 1613 smp_mb__after_atomic();
3b948ae5
TM		 1614 break;
1615 case TCP_CLOSE:
0fdea1e8
TM		 1616 if (test_and_clear_bit(XPRT_SOCK_CONNECTING,
1617 &transport->sock_state))
1618 xprt_clear_connecting(xprt);
7d1e8255 1619 xs_sock_mark_closed(xprt);
a246b010
CL		 1620 }
1621 out:
f064af1e 1622 read_unlock_bh(&sk->sk_callback_lock);
a246b010
CL		 1623}
1624
1f0fa154
IJ		 1625static void xs_write_space(struct sock *sk)
1626{
13331a55 1627 struct socket_wq *wq;
1f0fa154
IJ		 1628 struct rpc_xprt *xprt;
1629
13331a55 1630 if (!sk->sk_socket)
1f0fa154 1631 return;
13331a55 1632 clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1f0fa154
IJ		 1633
1634 if (unlikely(!(xprt = xprt_from_sock(sk))))
1635 return;
13331a55
TM		 1636 rcu_read_lock();
1637 wq = rcu_dereference(sk->sk_wq);
1638 if (!wq || test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &wq->flags) == 0)
1639 goto out;
1f0fa154
IJ		 1640
1641 xprt_write_space(xprt);
13331a55
TM		 1642out:
1643 rcu_read_unlock();
1f0fa154
IJ		 1644}
1645
9903cd1c 1646/**
c7b2cae8
CL		 1647 * xs_udp_write_space - callback invoked when socket buffer space
1648 * becomes available
9903cd1c
CL		 1649 * @sk: socket whose state has changed
1650 *
a246b010
CL		 1651 * Called when more output buffer space is available for this socket.
1652 * We try not to wake our writers until they can make "significant"
c7b2cae8 1653 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
a246b010
CL		 1654 * with a bunch of small requests.
1655 */
c7b2cae8 1656static void xs_udp_write_space(struct sock *sk)
a246b010 1657{
f064af1e 1658 read_lock_bh(&sk->sk_callback_lock);
a246b010 1659
c7b2cae8 1660 /* from net/core/sock.c:sock_def_write_space */
1f0fa154
IJ		 1661 if (sock_writeable(sk))
1662 xs_write_space(sk);
a246b010 1663
f064af1e 1664 read_unlock_bh(&sk->sk_callback_lock);
c7b2cae8 1665}
a246b010 1666
c7b2cae8
CL		 1667/**
1668 * xs_tcp_write_space - callback invoked when socket buffer space
1669 * becomes available
1670 * @sk: socket whose state has changed
1671 *
1672 * Called when more output buffer space is available for this socket.
1673 * We try not to wake our writers until they can make "significant"
1674 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1675 * with a bunch of small requests.
1676 */
1677static void xs_tcp_write_space(struct sock *sk)
1678{
f064af1e 1679 read_lock_bh(&sk->sk_callback_lock);
c7b2cae8
CL		 1680
1681 /* from net/core/stream.c:sk_stream_write_space */
64dc6130 1682 if (sk_stream_is_writeable(sk))
1f0fa154 1683 xs_write_space(sk);
b6ddf64f 1684
f064af1e 1685 read_unlock_bh(&sk->sk_callback_lock);
a246b010
CL		 1686}
1687
470056c2 1688static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
a246b010 1689{
ee0ac0c2
CL		 1690 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1691 struct sock *sk = transport->inet;
a246b010 1692
7c6e066e 1693 if (transport->rcvsize) {
a246b010 1694 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
7c6e066e 1695 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
a246b010 1696 }
7c6e066e 1697 if (transport->sndsize) {
a246b010 1698 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
7c6e066e 1699 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
a246b010
CL		 1700 sk->sk_write_space(sk);
1701 }
1702}
1703
43118c29 1704/**
470056c2 1705 * xs_udp_set_buffer_size - set send and receive limits
43118c29 1706 * @xprt: generic transport
470056c2
CL		 1707 * @sndsize: requested size of send buffer, in bytes
1708 * @rcvsize: requested size of receive buffer, in bytes
43118c29 1709 *
470056c2 1710 * Set socket send and receive buffer size limits.
43118c29 1711 */
470056c2 1712static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
43118c29 1713{
7c6e066e
CL		 1714 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1715
1716 transport->sndsize = 0;
470056c2 1717 if (sndsize)
7c6e066e
CL		 1718 transport->sndsize = sndsize + 1024;
1719 transport->rcvsize = 0;
470056c2 1720 if (rcvsize)
7c6e066e 1721 transport->rcvsize = rcvsize + 1024;
470056c2
CL		 1722
1723 xs_udp_do_set_buffer_size(xprt);
43118c29
CL		 1724}
1725
46c0ee8b
CL		 1726/**
1727 * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1728 * @task: task that timed out
1729 *
1730 * Adjust the congestion window after a retransmit timeout has occurred.
1731 */
6a24dfb6 1732static void xs_udp_timer(struct rpc_xprt *xprt, struct rpc_task *task)
46c0ee8b 1733{
6a24dfb6 1734 xprt_adjust_cwnd(xprt, task, -ETIMEDOUT);
46c0ee8b
CL		 1735}
1736
b85d8806
CL		 1737static unsigned short xs_get_random_port(void)
1738{
1739 unsigned short range = xprt_max_resvport - xprt_min_resvport;
63862b5b 1740 unsigned short rand = (unsigned short) prandom_u32() % range;
b85d8806
CL		 1741 return rand + xprt_min_resvport;
1742}
1743
4dda9c8a
TM		 1744/**
1745 * xs_set_reuseaddr_port - set the socket's port and address reuse options
1746 * @sock: socket
1747 *
1748 * Note that this function has to be called on all sockets that share the
1749 * same port, and it must be called before binding.
1750 */
1751static void xs_sock_set_reuseport(struct socket *sock)
1752{
402e23b4 1753 int opt = 1;
4dda9c8a 1754
402e23b4
TM		 1755 kernel_setsockopt(sock, SOL_SOCKET, SO_REUSEPORT,
1756 (char *)&opt, sizeof(opt));
4dda9c8a
TM		 1757}
1758
1759static unsigned short xs_sock_getport(struct socket *sock)
1760{
1761 struct sockaddr_storage buf;
1762 int buflen;
1763 unsigned short port = 0;
1764
1765 if (kernel_getsockname(sock, (struct sockaddr *)&buf, &buflen) < 0)
1766 goto out;
1767 switch (buf.ss_family) {
1768 case AF_INET6:
1769 port = ntohs(((struct sockaddr_in6 *)&buf)->sin6_port);
1770 break;
1771 case AF_INET:
1772 port = ntohs(((struct sockaddr_in *)&buf)->sin_port);
1773 }
1774out:
1775 return port;
1776}
1777
92200412
CL		 1778/**
1779 * xs_set_port - reset the port number in the remote endpoint address
1780 * @xprt: generic transport
1781 * @port: new port number
1782 *
1783 */
1784static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1785{
46121cf7 1786 dprintk("RPC: setting port for xprt %p to %u\n", xprt, port);
c4efcb1d 1787
9dc3b095
CL		 1788 rpc_set_port(xs_addr(xprt), port);
1789 xs_update_peer_port(xprt);
92200412
CL		 1790}
1791
4dda9c8a
TM		 1792static void xs_set_srcport(struct sock_xprt *transport, struct socket *sock)
1793{
1794 if (transport->srcport == 0)
1795 transport->srcport = xs_sock_getport(sock);
1796}
1797
5d4ec932 1798static unsigned short xs_get_srcport(struct sock_xprt *transport)
67a391d7 1799{
fbfffbd5 1800 unsigned short port = transport->srcport;
67a391d7
TM		 1801
1802 if (port == 0 && transport->xprt.resvport)
1803 port = xs_get_random_port();
1804 return port;
1805}
1806
baaf4e48 1807static unsigned short xs_next_srcport(struct sock_xprt *transport, unsigned short port)
67a391d7 1808{
fbfffbd5
CL		 1809 if (transport->srcport != 0)
1810 transport->srcport = 0;
67a391d7
TM		 1811 if (!transport->xprt.resvport)
1812 return 0;
1813 if (port <= xprt_min_resvport || port > xprt_max_resvport)
1814 return xprt_max_resvport;
1815 return --port;
1816}
beb59b68 1817static int xs_bind(struct sock_xprt *transport, struct socket *sock)
a246b010 1818{
beb59b68 1819 struct sockaddr_storage myaddr;
67a391d7 1820 int err, nloop = 0;
5d4ec932 1821 unsigned short port = xs_get_srcport(transport);
67a391d7 1822 unsigned short last;
a246b010 1823
0f7a622c
CP		 1824 /*
1825 * If we are asking for any ephemeral port (i.e. port == 0 &&
1826 * transport->xprt.resvport == 0), don't bind. Let the local
1827 * port selection happen implicitly when the socket is used
1828 * (for example at connect time).
1829 *
1830 * This ensures that we can continue to establish TCP
1831 * connections even when all local ephemeral ports are already
1832 * a part of some TCP connection. This makes no difference
1833 * for UDP sockets, but also doens't harm them.
1834 *
1835 * If we're asking for any reserved port (i.e. port == 0 &&
1836 * transport->xprt.resvport == 1) xs_get_srcport above will
1837 * ensure that port is non-zero and we will bind as needed.
1838 */
1839 if (port == 0)
1840 return 0;
1841
beb59b68 1842 memcpy(&myaddr, &transport->srcaddr, transport->xprt.addrlen);
a246b010 1843 do {
beb59b68
PE		 1844 rpc_set_port((struct sockaddr *)&myaddr, port);
1845 err = kernel_bind(sock, (struct sockaddr *)&myaddr,
1846 transport->xprt.addrlen);
a246b010 1847 if (err == 0) {
fbfffbd5 1848 transport->srcport = port;
d3bc9a1d 1849 break;
a246b010 1850 }
67a391d7 1851 last = port;
baaf4e48 1852 port = xs_next_srcport(transport, port);
67a391d7
TM		 1853 if (port > last)
1854 nloop++;
1855 } while (err == -EADDRINUSE && nloop != 2);
90058d37 1856
4232e863 1857 if (myaddr.ss_family == AF_INET)
beb59b68
PE		 1858 dprintk("RPC: %s %pI4:%u: %s (%d)\n", __func__,
1859 &((struct sockaddr_in *)&myaddr)->sin_addr,
1860 port, err ? "failed" : "ok", err);
1861 else
1862 dprintk("RPC: %s %pI6:%u: %s (%d)\n", __func__,
1863 &((struct sockaddr_in6 *)&myaddr)->sin6_addr,
1864 port, err ? "failed" : "ok", err);
a246b010
CL		 1865 return err;
1866}
1867
176e21ee
CL		 1868/*
1869 * We don't support autobind on AF_LOCAL sockets
1870 */
1871static void xs_local_rpcbind(struct rpc_task *task)
1872{
fb43d172 1873 xprt_set_bound(task->tk_xprt);
176e21ee
CL		 1874}
1875
1876static void xs_local_set_port(struct rpc_xprt *xprt, unsigned short port)
1877{
1878}
beb59b68 1879
ed07536e
PZ		 1880#ifdef CONFIG_DEBUG_LOCK_ALLOC
1881static struct lock_class_key xs_key[2];
1882static struct lock_class_key xs_slock_key[2];
1883
176e21ee
CL		 1884static inline void xs_reclassify_socketu(struct socket *sock)
1885{
1886 struct sock *sk = sock->sk;
1887
176e21ee
CL		 1888 sock_lock_init_class_and_name(sk, "slock-AF_LOCAL-RPC",
1889 &xs_slock_key[1], "sk_lock-AF_LOCAL-RPC", &xs_key[1]);
1890}
1891
8945ee5e 1892static inline void xs_reclassify_socket4(struct socket *sock)
ed07536e
PZ		 1893{
1894 struct sock *sk = sock->sk;
8945ee5e 1895
8945ee5e
CL		 1896 sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1897 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1898}
ed07536e 1899
8945ee5e
CL		 1900static inline void xs_reclassify_socket6(struct socket *sock)
1901{
1902 struct sock *sk = sock->sk;
ed07536e 1903
8945ee5e
CL		 1904 sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1905 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
ed07536e 1906}
6bc9638a
PE		 1907
1908static inline void xs_reclassify_socket(int family, struct socket *sock)
1909{
fafc4e1e 1910 if (WARN_ON_ONCE(!sock_allow_reclassification(sock->sk)))
1b7a1819
WAA		 1911 return;
1912
4232e863 1913 switch (family) {
176e21ee
CL		 1914 case AF_LOCAL:
1915 xs_reclassify_socketu(sock);
1916 break;
4232e863 1917 case AF_INET:
6bc9638a 1918 xs_reclassify_socket4(sock);
4232e863
CL		 1919 break;
1920 case AF_INET6:
6bc9638a 1921 xs_reclassify_socket6(sock);
4232e863
CL		 1922 break;
1923 }
6bc9638a 1924}
ed07536e 1925#else
6bc9638a
PE		 1926static inline void xs_reclassify_socket(int family, struct socket *sock)
1927{
1928}
ed07536e
PZ		 1929#endif
1930
93dc41bd
N		 1931static void xs_dummy_setup_socket(struct work_struct *work)
1932{
1933}
1934
6bc9638a 1935static struct socket *xs_create_sock(struct rpc_xprt *xprt,
4dda9c8a
TM		 1936 struct sock_xprt *transport, int family, int type,
1937 int protocol, bool reuseport)
22f79326
PE		 1938{
1939 struct socket *sock;
1940 int err;
1941
6bc9638a 1942 err = __sock_create(xprt->xprt_net, family, type, protocol, &sock, 1);
22f79326
PE		 1943 if (err < 0) {
1944 dprintk("RPC: can't create %d transport socket (%d).\n",
1945 protocol, -err);
1946 goto out;
1947 }
6bc9638a 1948 xs_reclassify_socket(family, sock);
22f79326 1949
4dda9c8a
TM		 1950 if (reuseport)
1951 xs_sock_set_reuseport(sock);
1952
4cea288a
BH		 1953 err = xs_bind(transport, sock);
1954 if (err) {
22f79326
PE		 1955 sock_release(sock);
1956 goto out;
1957 }
1958
1959 return sock;
1960out:
1961 return ERR_PTR(err);
1962}
1963
176e21ee
CL		 1964static int xs_local_finish_connecting(struct rpc_xprt *xprt,
1965 struct socket *sock)
1966{
1967 struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1968 xprt);
1969
1970 if (!transport->inet) {
1971 struct sock *sk = sock->sk;
1972
1973 write_lock_bh(&sk->sk_callback_lock);
1974
1975 xs_save_old_callbacks(transport, sk);
1976
1977 sk->sk_user_data = xprt;
a2648094 1978 sk->sk_data_ready = xs_data_ready;
176e21ee 1979 sk->sk_write_space = xs_udp_write_space;
b4411457 1980 sock_set_flag(sk, SOCK_FASYNC);
2118071d 1981 sk->sk_error_report = xs_error_report;
c2126157 1982 sk->sk_allocation = GFP_NOIO;
176e21ee
CL		 1983
1984 xprt_clear_connected(xprt);
1985
1986 /* Reset to new socket */
1987 transport->sock = sock;
1988 transport->inet = sk;
1989
1990 write_unlock_bh(&sk->sk_callback_lock);
1991 }
1992
1993 /* Tell the socket layer to start connecting... */
1994 xprt->stat.connect_count++;
1995 xprt->stat.connect_start = jiffies;
1996 return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, 0);
1997}
1998
1999/**
2000 * xs_local_setup_socket - create AF_LOCAL socket, connect to a local endpoint
176e21ee 2001 * @transport: socket transport to connect
176e21ee 2002 */
dc107402 2003static int xs_local_setup_socket(struct sock_xprt *transport)
176e21ee 2004{
176e21ee
CL		 2005 struct rpc_xprt *xprt = &transport->xprt;
2006 struct socket *sock;
2007 int status = -EIO;
2008
176e21ee
CL		 2009 status = __sock_create(xprt->xprt_net, AF_LOCAL,
2010 SOCK_STREAM, 0, &sock, 1);
2011 if (status < 0) {
2012 dprintk("RPC: can't create AF_LOCAL "
2013 "transport socket (%d).\n", -status);
2014 goto out;
2015 }
d1358917 2016 xs_reclassify_socket(AF_LOCAL, sock);
176e21ee
CL		 2017
2018 dprintk("RPC: worker connecting xprt %p via AF_LOCAL to %s\n",
2019 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2020
2021 status = xs_local_finish_connecting(xprt, sock);
40b5ea0c 2022 trace_rpc_socket_connect(xprt, sock, status);
176e21ee
CL		 2023 switch (status) {
2024 case 0:
2025 dprintk("RPC: xprt %p connected to %s\n",
2026 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2027 xprt_set_connected(xprt);
3601c4a9 2028 case -ENOBUFS:
176e21ee
CL		 2029 break;
2030 case -ENOENT:
2031 dprintk("RPC: xprt %p: socket %s does not exist\n",
2032 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2033 break;
4a20a988
TM		 2034 case -ECONNREFUSED:
2035 dprintk("RPC: xprt %p: connection refused for %s\n",
2036 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2037 break;
176e21ee
CL		 2038 default:
2039 printk(KERN_ERR "%s: unhandled error (%d) connecting to %s\n",
2040 __func__, -status,
2041 xprt->address_strings[RPC_DISPLAY_ADDR]);
2042 }
2043
2044out:
2045 xprt_clear_connecting(xprt);
2046 xprt_wake_pending_tasks(xprt, status);
dc107402
BF		 2047 return status;
2048}
2049
b6669737 2050static void xs_local_connect(struct rpc_xprt *xprt, struct rpc_task *task)
dc107402 2051{
dc107402
BF		 2052 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2053 int ret;
2054
2055 if (RPC_IS_ASYNC(task)) {
2056 /*
2057 * We want the AF_LOCAL connect to be resolved in the
2058 * filesystem namespace of the process making the rpc
2059 * call. Thus we connect synchronously.
2060 *
2061 * If we want to support asynchronous AF_LOCAL calls,
2062 * we'll need to figure out how to pass a namespace to
2063 * connect.
2064 */
2065 rpc_exit(task, -ENOTCONN);
2066 return;
2067 }
2068 ret = xs_local_setup_socket(transport);
2069 if (ret && !RPC_IS_SOFTCONN(task))
2070 msleep_interruptible(15000);
176e21ee
CL		 2071}
2072
3c87ef6e 2073#if IS_ENABLED(CONFIG_SUNRPC_SWAP)
d6e971d8
JL		 2074/*
2075 * Note that this should be called with XPRT_LOCKED held (or when we otherwise
2076 * know that we have exclusive access to the socket), to guard against
2077 * races with xs_reset_transport.
2078 */
a564b8f0
MG		 2079static void xs_set_memalloc(struct rpc_xprt *xprt)
2080{
2081 struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
2082 xprt);
2083
d6e971d8
JL		 2084 /*
2085 * If there's no sock, then we have nothing to set. The
2086 * reconnecting process will get it for us.
2087 */
2088 if (!transport->inet)
2089 return;
8e228133 2090 if (atomic_read(&xprt->swapper))
a564b8f0
MG		 2091 sk_set_memalloc(transport->inet);
2092}
2093
2094/**
d67fa4d8 2095 * xs_enable_swap - Tag this transport as being used for swap.
a564b8f0 2096 * @xprt: transport to tag
a564b8f0 2097 *
8e228133
JL		 2098 * Take a reference to this transport on behalf of the rpc_clnt, and
2099 * optionally mark it for swapping if it wasn't already.
a564b8f0 2100 */
d67fa4d8
JL		 2101static int
2102xs_enable_swap(struct rpc_xprt *xprt)
a564b8f0 2103{
d6e971d8 2104 struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
a564b8f0 2105
d6e971d8
JL		 2106 if (atomic_inc_return(&xprt->swapper) != 1)
2107 return 0;
2108 if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE))
2109 return -ERESTARTSYS;
2110 if (xs->inet)
2111 sk_set_memalloc(xs->inet);
2112 xprt_release_xprt(xprt, NULL);
8e228133
JL		 2113 return 0;
2114}
a564b8f0 2115
8e228133 2116/**
d67fa4d8 2117 * xs_disable_swap - Untag this transport as being used for swap.
8e228133
JL		 2118 * @xprt: transport to tag
2119 *
2120 * Drop a "swapper" reference to this xprt on behalf of the rpc_clnt. If the
2121 * swapper refcount goes to 0, untag the socket as a memalloc socket.
2122 */
d67fa4d8
JL		 2123static void
2124xs_disable_swap(struct rpc_xprt *xprt)
8e228133 2125{
d6e971d8 2126 struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
8e228133 2127
d6e971d8
JL		 2128 if (!atomic_dec_and_test(&xprt->swapper))
2129 return;
2130 if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE))
2131 return;
2132 if (xs->inet)
2133 sk_clear_memalloc(xs->inet);
2134 xprt_release_xprt(xprt, NULL);
a564b8f0 2135}
a564b8f0
MG		 2136#else
2137static void xs_set_memalloc(struct rpc_xprt *xprt)
2138{
2139}
d67fa4d8
JL		 2140
2141static int
2142xs_enable_swap(struct rpc_xprt *xprt)
2143{
2144 return -EINVAL;
2145}
2146
2147static void
2148xs_disable_swap(struct rpc_xprt *xprt)
2149{
2150}
a564b8f0
MG		 2151#endif
2152
16be2d20
CL		 2153static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2154{
2155 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2156
2157 if (!transport->inet) {
2158 struct sock *sk = sock->sk;
2159
2160 write_lock_bh(&sk->sk_callback_lock);
2161
2a9e1cfa
TM		 2162 xs_save_old_callbacks(transport, sk);
2163
16be2d20 2164 sk->sk_user_data = xprt;
f9b2ee71 2165 sk->sk_data_ready = xs_data_ready;
16be2d20 2166 sk->sk_write_space = xs_udp_write_space;
b4411457 2167 sock_set_flag(sk, SOCK_FASYNC);
c2126157 2168 sk->sk_allocation = GFP_NOIO;
16be2d20
CL		 2169
2170 xprt_set_connected(xprt);
2171
2172 /* Reset to new socket */
2173 transport->sock = sock;
2174 transport->inet = sk;
2175
a564b8f0
MG		 2176 xs_set_memalloc(xprt);
2177
16be2d20
CL		 2178 write_unlock_bh(&sk->sk_callback_lock);
2179 }
2180 xs_udp_do_set_buffer_size(xprt);
2181}
2182
8c14ff2a 2183static void xs_udp_setup_socket(struct work_struct *work)
a246b010 2184{
34161db6
TM		 2185 struct sock_xprt *transport =
2186 container_of(work, struct sock_xprt, connect_worker.work);
c8475461 2187 struct rpc_xprt *xprt = &transport->xprt;
ee0ac0c2 2188 struct socket *sock = transport->sock;
b65c0310 2189 int status = -EIO;
9903cd1c 2190
8c14ff2a 2191 sock = xs_create_sock(xprt, transport,
4dda9c8a
TM		 2192 xs_addr(xprt)->sa_family, SOCK_DGRAM,
2193 IPPROTO_UDP, false);
b65c0310 2194 if (IS_ERR(sock))
b0d93ad5 2195 goto out;
68e220bd 2196
c740eff8
CL		 2197 dprintk("RPC: worker connecting xprt %p via %s to "
2198 "%s (port %s)\n", xprt,
2199 xprt->address_strings[RPC_DISPLAY_PROTO],
2200 xprt->address_strings[RPC_DISPLAY_ADDR],
2201 xprt->address_strings[RPC_DISPLAY_PORT]);
68e220bd
CL		 2202
2203 xs_udp_finish_connecting(xprt, sock);
40b5ea0c 2204 trace_rpc_socket_connect(xprt, sock, 0);
b0d93ad5
CL		 2205 status = 0;
2206out:
718ba5b8 2207 xprt_unlock_connect(xprt, transport);
b0d93ad5 2208 xprt_clear_connecting(xprt);
7d1e8255 2209 xprt_wake_pending_tasks(xprt, status);
a246b010
CL		 2210}
2211
4876cc77
TM		 2212/**
2213 * xs_tcp_shutdown - gracefully shut down a TCP socket
2214 * @xprt: transport
2215 *
2216 * Initiates a graceful shutdown of the TCP socket by calling the
2217 * equivalent of shutdown(SHUT_RDWR);
2218 */
2219static void xs_tcp_shutdown(struct rpc_xprt *xprt)
2220{
2221 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2222 struct socket *sock = transport->sock;
2223
2224 if (sock == NULL)
2225 return;
2226 if (xprt_connected(xprt)) {
2227 kernel_sock_shutdown(sock, SHUT_RDWR);
2228 trace_rpc_socket_shutdown(xprt, sock);
2229 } else
2230 xs_reset_transport(transport);
2231}
2232
16be2d20 2233static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
a246b010 2234{
16be2d20 2235 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
fe19a96b 2236 int ret = -ENOTCONN;
edb267a6 2237
ee0ac0c2 2238 if (!transport->inet) {
b0d93ad5 2239 struct sock *sk = sock->sk;
7f260e85
TM		 2240 unsigned int keepidle = xprt->timeout->to_initval / HZ;
2241 unsigned int keepcnt = xprt->timeout->to_retries + 1;
2242 unsigned int opt_on = 1;
775f06ab 2243 unsigned int timeo;
7f260e85
TM		 2244
2245 /* TCP Keepalive options */
2246 kernel_setsockopt(sock, SOL_SOCKET, SO_KEEPALIVE,
2247 (char *)&opt_on, sizeof(opt_on));
2248 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPIDLE,
2249 (char *)&keepidle, sizeof(keepidle));
2250 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPINTVL,
2251 (char *)&keepidle, sizeof(keepidle));
2252 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPCNT,
2253 (char *)&keepcnt, sizeof(keepcnt));
b0d93ad5 2254
775f06ab
TM		 2255 /* TCP user timeout (see RFC5482) */
2256 timeo = jiffies_to_msecs(xprt->timeout->to_initval) *
2257 (xprt->timeout->to_retries + 1);
2258 kernel_setsockopt(sock, SOL_TCP, TCP_USER_TIMEOUT,
2259 (char *)&timeo, sizeof(timeo));
2260
b0d93ad5
CL		 2261 write_lock_bh(&sk->sk_callback_lock);
2262
2a9e1cfa
TM		 2263 xs_save_old_callbacks(transport, sk);
2264
b0d93ad5 2265 sk->sk_user_data = xprt;
b0d93ad5
CL		 2266 sk->sk_data_ready = xs_tcp_data_ready;
2267 sk->sk_state_change = xs_tcp_state_change;
2268 sk->sk_write_space = xs_tcp_write_space;
b4411457 2269 sock_set_flag(sk, SOCK_FASYNC);
2118071d 2270 sk->sk_error_report = xs_error_report;
c2126157 2271 sk->sk_allocation = GFP_NOIO;
3167e12c
CL		 2272
2273 /* socket options */
3167e12c 2274 sock_reset_flag(sk, SOCK_LINGER);
3167e12c 2275 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
b0d93ad5
CL		 2276
2277 xprt_clear_connected(xprt);
2278
2279 /* Reset to new socket */
ee0ac0c2
CL		 2280 transport->sock = sock;
2281 transport->inet = sk;
b0d93ad5
CL		 2282
2283 write_unlock_bh(&sk->sk_callback_lock);
2284 }
2285
01d37c42 2286 if (!xprt_bound(xprt))
fe19a96b 2287 goto out;
01d37c42 2288
a564b8f0
MG		 2289 xs_set_memalloc(xprt);
2290
b0d93ad5 2291 /* Tell the socket layer to start connecting... */
262ca07d
CL		 2292 xprt->stat.connect_count++;
2293 xprt->stat.connect_start = jiffies;
0fdea1e8 2294 set_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
fe19a96b
TM		 2295 ret = kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
2296 switch (ret) {
2297 case 0:
4dda9c8a 2298 xs_set_srcport(transport, sock);
fe19a96b
TM		 2299 case -EINPROGRESS:
2300 /* SYN_SENT! */
fe19a96b
TM		 2301 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2302 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2303 }
2304out:
2305 return ret;
16be2d20
CL		 2306}
2307
9903cd1c 2308/**
b61d59ff 2309 * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint
9903cd1c
CL		 2310 *
2311 * Invoked by a work queue tasklet.
a246b010 2312 */
cdd518d5 2313static void xs_tcp_setup_socket(struct work_struct *work)
a246b010 2314{
cdd518d5
PE		 2315 struct sock_xprt *transport =
2316 container_of(work, struct sock_xprt, connect_worker.work);
ee0ac0c2 2317 struct socket *sock = transport->sock;
a9f5f0f7 2318 struct rpc_xprt *xprt = &transport->xprt;
b61d59ff 2319 int status = -EIO;
a246b010 2320
ee0ac0c2 2321 if (!sock) {
cdd518d5 2322 sock = xs_create_sock(xprt, transport,
4dda9c8a
TM		 2323 xs_addr(xprt)->sa_family, SOCK_STREAM,
2324 IPPROTO_TCP, true);
b61d59ff
TM		 2325 if (IS_ERR(sock)) {
2326 status = PTR_ERR(sock);
3167e12c
CL		 2327 goto out;
2328 }
7d1e8255 2329 }
a246b010 2330
c740eff8
CL		 2331 dprintk("RPC: worker connecting xprt %p via %s to "
2332 "%s (port %s)\n", xprt,
2333 xprt->address_strings[RPC_DISPLAY_PROTO],
2334 xprt->address_strings[RPC_DISPLAY_ADDR],
2335 xprt->address_strings[RPC_DISPLAY_PORT]);
edb267a6 2336
16be2d20 2337 status = xs_tcp_finish_connecting(xprt, sock);
40b5ea0c 2338 trace_rpc_socket_connect(xprt, sock, status);
46121cf7
CL		 2339 dprintk("RPC: %p connect status %d connected %d sock state %d\n",
2340 xprt, -status, xprt_connected(xprt),
2341 sock->sk->sk_state);
2a491991 2342 switch (status) {
f75e6745
TM		 2343 default:
2344 printk("%s: connect returned unhandled error %d\n",
2345 __func__, status);
2346 case -EADDRNOTAVAIL:
2347 /* We're probably in TIME_WAIT. Get rid of existing socket,
2348 * and retry
2349 */
a519fc7a 2350 xs_tcp_force_close(xprt);
88b5ed73 2351 break;
2a491991
TM		 2352 case 0:
2353 case -EINPROGRESS:
2354 case -EALREADY:
718ba5b8 2355 xprt_unlock_connect(xprt, transport);
7d1e8255 2356 return;
9fcfe0c8
TM		 2357 case -EINVAL:
2358 /* Happens, for instance, if the user specified a link
2359 * local IPv6 address without a scope-id.
2360 */
3ed5e2a2
TM		 2361 case -ECONNREFUSED:
2362 case -ECONNRESET:
2363 case -ENETUNREACH:
3913c78c 2364 case -EADDRINUSE:
3601c4a9 2365 case -ENOBUFS:
3ed5e2a2 2366 /* retry with existing socket, after a delay */
4efdd92c 2367 xs_tcp_force_close(xprt);
9fcfe0c8 2368 goto out;
a246b010 2369 }
2a491991 2370 status = -EAGAIN;
a246b010 2371out:
718ba5b8 2372 xprt_unlock_connect(xprt, transport);
68e220bd 2373 xprt_clear_connecting(xprt);
7d1e8255 2374 xprt_wake_pending_tasks(xprt, status);
68e220bd 2375}
b0d93ad5 2376
9903cd1c
CL		 2377/**
2378 * xs_connect - connect a socket to a remote endpoint
1b092092 2379 * @xprt: pointer to transport structure
9903cd1c
CL		 2380 * @task: address of RPC task that manages state of connect request
2381 *
2382 * TCP: If the remote end dropped the connection, delay reconnecting.
03bf4b70
CL		 2383 *
2384 * UDP socket connects are synchronous, but we use a work queue anyway
2385 * to guarantee that even unprivileged user processes can set up a
2386 * socket on a privileged port.
2387 *
2388 * If a UDP socket connect fails, the delay behavior here prevents
2389 * retry floods (hard mounts).
9903cd1c 2390 */
1b092092 2391static void xs_connect(struct rpc_xprt *xprt, struct rpc_task *task)
a246b010 2392{
ee0ac0c2 2393 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
a246b010 2394
718ba5b8
TM		 2395 WARN_ON_ONCE(!xprt_lock_connect(xprt, task, transport));
2396
99b1a4c3 2397 if (transport->sock != NULL) {
46121cf7
CL		 2398 dprintk("RPC: xs_connect delayed xprt %p for %lu "
2399 "seconds\n",
03bf4b70 2400 xprt, xprt->reestablish_timeout / HZ);
99b1a4c3
TM		 2401
2402 /* Start by resetting any existing state */
2403 xs_reset_transport(transport);
2404
c1384c9c
TM		 2405 queue_delayed_work(rpciod_workqueue,
2406 &transport->connect_worker,
2407 xprt->reestablish_timeout);
03bf4b70 2408 xprt->reestablish_timeout <<= 1;
61d0a8e6
NB		 2409 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2410 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
03bf4b70
CL		 2411 if (xprt->reestablish_timeout > XS_TCP_MAX_REEST_TO)
2412 xprt->reestablish_timeout = XS_TCP_MAX_REEST_TO;
b0d93ad5 2413 } else {
46121cf7 2414 dprintk("RPC: xs_connect scheduled xprt %p\n", xprt);
c1384c9c
TM		 2415 queue_delayed_work(rpciod_workqueue,
2416 &transport->connect_worker, 0);
a246b010
CL		 2417 }
2418}
2419
176e21ee
CL		 2420/**
2421 * xs_local_print_stats - display AF_LOCAL socket-specifc stats
2422 * @xprt: rpc_xprt struct containing statistics
2423 * @seq: output file
2424 *
2425 */
2426static void xs_local_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2427{
2428 long idle_time = 0;
2429
2430 if (xprt_connected(xprt))
2431 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2432
2433 seq_printf(seq, "\txprt:\tlocal %lu %lu %lu %ld %lu %lu %lu "
15a45206 2434 "%llu %llu %lu %llu %llu\n",
176e21ee
CL		 2435 xprt->stat.bind_count,
2436 xprt->stat.connect_count,
2437 xprt->stat.connect_time,
2438 idle_time,
2439 xprt->stat.sends,
2440 xprt->stat.recvs,
2441 xprt->stat.bad_xids,
2442 xprt->stat.req_u,
15a45206
AA		 2443 xprt->stat.bklog_u,
2444 xprt->stat.max_slots,
2445 xprt->stat.sending_u,
2446 xprt->stat.pending_u);
176e21ee
CL		 2447}
2448
262ca07d
CL		 2449/**
2450 * xs_udp_print_stats - display UDP socket-specifc stats
2451 * @xprt: rpc_xprt struct containing statistics
2452 * @seq: output file
2453 *
2454 */
2455static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2456{
c8475461
CL		 2457 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2458
15a45206
AA		 2459 seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %llu %llu "
2460 "%lu %llu %llu\n",
fbfffbd5 2461 transport->srcport,
262ca07d
CL		 2462 xprt->stat.bind_count,
2463 xprt->stat.sends,
2464 xprt->stat.recvs,
2465 xprt->stat.bad_xids,
2466 xprt->stat.req_u,
15a45206
AA		 2467 xprt->stat.bklog_u,
2468 xprt->stat.max_slots,
2469 xprt->stat.sending_u,
2470 xprt->stat.pending_u);
262ca07d
CL		 2471}
2472
2473/**
2474 * xs_tcp_print_stats - display TCP socket-specifc stats
2475 * @xprt: rpc_xprt struct containing statistics
2476 * @seq: output file
2477 *
2478 */
2479static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2480{
c8475461 2481 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
262ca07d
CL		 2482 long idle_time = 0;
2483
2484 if (xprt_connected(xprt))
2485 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2486
15a45206
AA		 2487 seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu "
2488 "%llu %llu %lu %llu %llu\n",
fbfffbd5 2489 transport->srcport,
262ca07d
CL		 2490 xprt->stat.bind_count,
2491 xprt->stat.connect_count,
2492 xprt->stat.connect_time,
2493 idle_time,
2494 xprt->stat.sends,
2495 xprt->stat.recvs,
2496 xprt->stat.bad_xids,
2497 xprt->stat.req_u,
15a45206
AA		 2498 xprt->stat.bklog_u,
2499 xprt->stat.max_slots,
2500 xprt->stat.sending_u,
2501 xprt->stat.pending_u);
262ca07d
CL		 2502}
2503
4cfc7e60
RI		 2504/*
2505 * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason
2506 * we allocate pages instead doing a kmalloc like rpc_malloc is because we want
2507 * to use the server side send routines.
2508 */
5a51f13a 2509static void *bc_malloc(struct rpc_task *task, size_t size)
4cfc7e60
RI		 2510{
2511 struct page *page;
2512 struct rpc_buffer *buf;
2513
b8a13d03
WAA		 2514 WARN_ON_ONCE(size > PAGE_SIZE - sizeof(struct rpc_buffer));
2515 if (size > PAGE_SIZE - sizeof(struct rpc_buffer))
2516 return NULL;
4cfc7e60 2517
b8a13d03 2518 page = alloc_page(GFP_KERNEL);
4cfc7e60
RI		 2519 if (!page)
2520 return NULL;
2521
2522 buf = page_address(page);
2523 buf->len = PAGE_SIZE;
2524
2525 return buf->data;
2526}
2527
2528/*
2529 * Free the space allocated in the bc_alloc routine
2530 */
5a51f13a 2531static void bc_free(void *buffer)
4cfc7e60
RI		 2532{
2533 struct rpc_buffer *buf;
2534
2535 if (!buffer)
2536 return;
2537
2538 buf = container_of(buffer, struct rpc_buffer, data);
2539 free_page((unsigned long)buf);
2540}
2541
2542/*
2543 * Use the svc_sock to send the callback. Must be called with svsk->sk_mutex
2544 * held. Borrows heavily from svc_tcp_sendto and xs_tcp_send_request.
2545 */
2546static int bc_sendto(struct rpc_rqst *req)
2547{
2548 int len;
2549 struct xdr_buf *xbufp = &req->rq_snd_buf;
2550 struct rpc_xprt *xprt = req->rq_xprt;
2551 struct sock_xprt *transport =
2552 container_of(xprt, struct sock_xprt, xprt);
2553 struct socket *sock = transport->sock;
2554 unsigned long headoff;
2555 unsigned long tailoff;
2556
61677eee 2557 xs_encode_stream_record_marker(xbufp);
4cfc7e60
RI		 2558
2559 tailoff = (unsigned long)xbufp->tail[0].iov_base & ~PAGE_MASK;
2560 headoff = (unsigned long)xbufp->head[0].iov_base & ~PAGE_MASK;
2561 len = svc_send_common(sock, xbufp,
2562 virt_to_page(xbufp->head[0].iov_base), headoff,
2563 xbufp->tail[0].iov_base, tailoff);
2564
2565 if (len != xbufp->len) {
2566 printk(KERN_NOTICE "Error sending entire callback!\n");
2567 len = -EAGAIN;
2568 }
2569
2570 return len;
2571}
2572
2573/*
2574 * The send routine. Borrows from svc_send
2575 */
2576static int bc_send_request(struct rpc_task *task)
2577{
2578 struct rpc_rqst *req = task->tk_rqstp;
2579 struct svc_xprt *xprt;
7fc56136 2580 int len;
4cfc7e60
RI		 2581
2582 dprintk("sending request with xid: %08x\n", ntohl(req->rq_xid));
2583 /*
2584 * Get the server socket associated with this callback xprt
2585 */
2586 xprt = req->rq_xprt->bc_xprt;
4cfc7e60
RI		 2587
2588 /*
2589 * Grab the mutex to serialize data as the connection is shared
2590 * with the fore channel
2591 */
2592 if (!mutex_trylock(&xprt->xpt_mutex)) {
2593 rpc_sleep_on(&xprt->xpt_bc_pending, task, NULL);
2594 if (!mutex_trylock(&xprt->xpt_mutex))
2595 return -EAGAIN;
2596 rpc_wake_up_queued_task(&xprt->xpt_bc_pending, task);
2597 }
2598 if (test_bit(XPT_DEAD, &xprt->xpt_flags))
2599 len = -ENOTCONN;
2600 else
2601 len = bc_sendto(req);
2602 mutex_unlock(&xprt->xpt_mutex);
2603
2604 if (len > 0)
2605 len = 0;
2606
2607 return len;
2608}
2609
2610/*
2611 * The close routine. Since this is client initiated, we do nothing
2612 */
2613
2614static void bc_close(struct rpc_xprt *xprt)
2615{
4cfc7e60
RI		 2616}
2617
2618/*
2619 * The xprt destroy routine. Again, because this connection is client
2620 * initiated, we do nothing
2621 */
2622
2623static void bc_destroy(struct rpc_xprt *xprt)
2624{
47f72efa
KM		 2625 dprintk("RPC: bc_destroy xprt %p\n", xprt);
2626
2627 xs_xprt_free(xprt);
2628 module_put(THIS_MODULE);
4cfc7e60
RI		 2629}
2630
176e21ee
CL		 2631static struct rpc_xprt_ops xs_local_ops = {
2632 .reserve_xprt = xprt_reserve_xprt,
2633 .release_xprt = xs_tcp_release_xprt,
f39c1bfb 2634 .alloc_slot = xprt_alloc_slot,
176e21ee
CL		 2635 .rpcbind = xs_local_rpcbind,
2636 .set_port = xs_local_set_port,
dc107402 2637 .connect = xs_local_connect,
176e21ee
CL		 2638 .buf_alloc = rpc_malloc,
2639 .buf_free = rpc_free,
2640 .send_request = xs_local_send_request,
2641 .set_retrans_timeout = xprt_set_retrans_timeout_def,
2642 .close = xs_close,
a1311d87 2643 .destroy = xs_destroy,
176e21ee 2644 .print_stats = xs_local_print_stats,
d67fa4d8
JL		 2645 .enable_swap = xs_enable_swap,
2646 .disable_swap = xs_disable_swap,
176e21ee
CL		 2647};
2648
262965f5 2649static struct rpc_xprt_ops xs_udp_ops = {
43118c29 2650 .set_buffer_size = xs_udp_set_buffer_size,
12a80469 2651 .reserve_xprt = xprt_reserve_xprt_cong,
49e9a890 2652 .release_xprt = xprt_release_xprt_cong,
f39c1bfb 2653 .alloc_slot = xprt_alloc_slot,
45160d62 2654 .rpcbind = rpcb_getport_async,
92200412 2655 .set_port = xs_set_port,
262965f5 2656 .connect = xs_connect,
02107148
CL		 2657 .buf_alloc = rpc_malloc,
2658 .buf_free = rpc_free,
262965f5 2659 .send_request = xs_udp_send_request,
fe3aca29 2660 .set_retrans_timeout = xprt_set_retrans_timeout_rtt,
46c0ee8b 2661 .timer = xs_udp_timer,
a58dd398 2662 .release_request = xprt_release_rqst_cong,
262965f5
CL		 2663 .close = xs_close,
2664 .destroy = xs_destroy,
262ca07d 2665 .print_stats = xs_udp_print_stats,
d67fa4d8
JL		 2666 .enable_swap = xs_enable_swap,
2667 .disable_swap = xs_disable_swap,
4a068258 2668 .inject_disconnect = xs_inject_disconnect,
262965f5
CL		 2669};
2670
2671static struct rpc_xprt_ops xs_tcp_ops = {
12a80469 2672 .reserve_xprt = xprt_reserve_xprt,
e0ab53de 2673 .release_xprt = xs_tcp_release_xprt,
f39c1bfb 2674 .alloc_slot = xprt_lock_and_alloc_slot,
45160d62 2675 .rpcbind = rpcb_getport_async,
92200412 2676 .set_port = xs_set_port,
0b9e7943 2677 .connect = xs_connect,
02107148
CL		 2678 .buf_alloc = rpc_malloc,
2679 .buf_free = rpc_free,
262965f5 2680 .send_request = xs_tcp_send_request,
fe3aca29 2681 .set_retrans_timeout = xprt_set_retrans_timeout_def,
c627d31b 2682 .close = xs_tcp_shutdown,
9903cd1c 2683 .destroy = xs_destroy,
262ca07d 2684 .print_stats = xs_tcp_print_stats,
d67fa4d8
JL		 2685 .enable_swap = xs_enable_swap,
2686 .disable_swap = xs_disable_swap,
4a068258 2687 .inject_disconnect = xs_inject_disconnect,
42e5c3e2
CL		 2688#ifdef CONFIG_SUNRPC_BACKCHANNEL
2689 .bc_setup = xprt_setup_bc,
76566773 2690 .bc_up = xs_tcp_bc_up,
6b26cc8c 2691 .bc_maxpayload = xs_tcp_bc_maxpayload,
42e5c3e2
CL		 2692 .bc_free_rqst = xprt_free_bc_rqst,
2693 .bc_destroy = xprt_destroy_bc,
2694#endif
a246b010
CL		 2695};
2696
4cfc7e60
RI		 2697/*
2698 * The rpc_xprt_ops for the server backchannel
2699 */
2700
2701static struct rpc_xprt_ops bc_tcp_ops = {
2702 .reserve_xprt = xprt_reserve_xprt,
2703 .release_xprt = xprt_release_xprt,
84e28a30 2704 .alloc_slot = xprt_alloc_slot,
4cfc7e60
RI		 2705 .buf_alloc = bc_malloc,
2706 .buf_free = bc_free,
2707 .send_request = bc_send_request,
2708 .set_retrans_timeout = xprt_set_retrans_timeout_def,
2709 .close = bc_close,
2710 .destroy = bc_destroy,
2711 .print_stats = xs_tcp_print_stats,
d67fa4d8
JL		 2712 .enable_swap = xs_enable_swap,
2713 .disable_swap = xs_disable_swap,
4a068258 2714 .inject_disconnect = xs_inject_disconnect,
4cfc7e60
RI		 2715};
2716
92476850
CL		 2717static int xs_init_anyaddr(const int family, struct sockaddr *sap)
2718{
2719 static const struct sockaddr_in sin = {
2720 .sin_family = AF_INET,
2721 .sin_addr.s_addr = htonl(INADDR_ANY),
2722 };
2723 static const struct sockaddr_in6 sin6 = {
2724 .sin6_family = AF_INET6,
2725 .sin6_addr = IN6ADDR_ANY_INIT,
2726 };
2727
2728 switch (family) {
176e21ee
CL		 2729 case AF_LOCAL:
2730 break;
92476850
CL		 2731 case AF_INET:
2732 memcpy(sap, &sin, sizeof(sin));
2733 break;
2734 case AF_INET6:
2735 memcpy(sap, &sin6, sizeof(sin6));
2736 break;
2737 default:
2738 dprintk("RPC: %s: Bad address family\n", __func__);
2739 return -EAFNOSUPPORT;
2740 }
2741 return 0;
2742}
2743
3c341b0b 2744static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
d9ba131d
TM		 2745 unsigned int slot_table_size,
2746 unsigned int max_slot_table_size)
c8541ecd
CL		 2747{
2748 struct rpc_xprt *xprt;
ffc2e518 2749 struct sock_xprt *new;
c8541ecd 2750
96802a09 2751 if (args->addrlen > sizeof(xprt->addr)) {
46121cf7 2752 dprintk("RPC: xs_setup_xprt: address too large\n");
c8541ecd
CL		 2753 return ERR_PTR(-EBADF);
2754 }
2755
d9ba131d
TM		 2756 xprt = xprt_alloc(args->net, sizeof(*new), slot_table_size,
2757 max_slot_table_size);
bd1722d4 2758 if (xprt == NULL) {
46121cf7
CL		 2759 dprintk("RPC: xs_setup_xprt: couldn't allocate "
2760 "rpc_xprt\n");
c8541ecd
CL		 2761 return ERR_PTR(-ENOMEM);
2762 }
c8541ecd 2763
bd1722d4 2764 new = container_of(xprt, struct sock_xprt, xprt);
edc1b01c 2765 mutex_init(&new->recv_mutex);
96802a09
FM		 2766 memcpy(&xprt->addr, args->dstaddr, args->addrlen);
2767 xprt->addrlen = args->addrlen;
d3bc9a1d 2768 if (args->srcaddr)
fbfffbd5 2769 memcpy(&new->srcaddr, args->srcaddr, args->addrlen);
92476850
CL		 2770 else {
2771 int err;
2772 err = xs_init_anyaddr(args->dstaddr->sa_family,
2773 (struct sockaddr *)&new->srcaddr);
2aa13531
SK		 2774 if (err != 0) {
2775 xprt_free(xprt);
92476850 2776 return ERR_PTR(err);
2aa13531 2777 }
92476850 2778 }
c8541ecd
CL		 2779
2780 return xprt;
2781}
2782
176e21ee
CL		 2783static const struct rpc_timeout xs_local_default_timeout = {
2784 .to_initval = 10 * HZ,
2785 .to_maxval = 10 * HZ,
2786 .to_retries = 2,
2787};
2788
2789/**
2790 * xs_setup_local - Set up transport to use an AF_LOCAL socket
2791 * @args: rpc transport creation arguments
2792 *
2793 * AF_LOCAL is a "tpi_cots_ord" transport, just like TCP
2794 */
2795static struct rpc_xprt *xs_setup_local(struct xprt_create *args)
2796{
2797 struct sockaddr_un *sun = (struct sockaddr_un *)args->dstaddr;
2798 struct sock_xprt *transport;
2799 struct rpc_xprt *xprt;
2800 struct rpc_xprt *ret;
2801
d9ba131d
TM		 2802 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2803 xprt_max_tcp_slot_table_entries);
176e21ee
CL		 2804 if (IS_ERR(xprt))
2805 return xprt;
2806 transport = container_of(xprt, struct sock_xprt, xprt);
2807
2808 xprt->prot = 0;
2809 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2810 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2811
2812 xprt->bind_timeout = XS_BIND_TO;
2813 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2814 xprt->idle_timeout = XS_IDLE_DISC_TO;
2815
2816 xprt->ops = &xs_local_ops;
2817 xprt->timeout = &xs_local_default_timeout;
2818
a2648094 2819 INIT_WORK(&transport->recv_worker, xs_local_data_receive_workfn);
93dc41bd
N		 2820 INIT_DELAYED_WORK(&transport->connect_worker,
2821 xs_dummy_setup_socket);
2822
176e21ee
CL		 2823 switch (sun->sun_family) {
2824 case AF_LOCAL:
2825 if (sun->sun_path[0] != '/') {
2826 dprintk("RPC: bad AF_LOCAL address: %s\n",
2827 sun->sun_path);
2828 ret = ERR_PTR(-EINVAL);
2829 goto out_err;
2830 }
2831 xprt_set_bound(xprt);
176e21ee 2832 xs_format_peer_addresses(xprt, "local", RPCBIND_NETID_LOCAL);
7073ea87
BF		 2833 ret = ERR_PTR(xs_local_setup_socket(transport));
2834 if (ret)
2835 goto out_err;
176e21ee
CL		 2836 break;
2837 default:
2838 ret = ERR_PTR(-EAFNOSUPPORT);
2839 goto out_err;
2840 }
2841
2842 dprintk("RPC: set up xprt to %s via AF_LOCAL\n",
2843 xprt->address_strings[RPC_DISPLAY_ADDR]);
2844
2845 if (try_module_get(THIS_MODULE))
2846 return xprt;
2847 ret = ERR_PTR(-EINVAL);
2848out_err:
315f3812 2849 xs_xprt_free(xprt);
176e21ee
CL		 2850 return ret;
2851}
2852
2881ae74
TM		 2853static const struct rpc_timeout xs_udp_default_timeout = {
2854 .to_initval = 5 * HZ,
2855 .to_maxval = 30 * HZ,
2856 .to_increment = 5 * HZ,
2857 .to_retries = 5,
2858};
2859
9903cd1c
CL		 2860/**
2861 * xs_setup_udp - Set up transport to use a UDP socket
96802a09 2862 * @args: rpc transport creation arguments
9903cd1c
CL		 2863 *
2864 */
483066d6 2865static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
a246b010 2866{
8f9d5b1a 2867 struct sockaddr *addr = args->dstaddr;
c8541ecd 2868 struct rpc_xprt *xprt;
c8475461 2869 struct sock_xprt *transport;
0a68b0be 2870 struct rpc_xprt *ret;
a246b010 2871
d9ba131d
TM		 2872 xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries,
2873 xprt_udp_slot_table_entries);
c8541ecd
CL		 2874 if (IS_ERR(xprt))
2875 return xprt;
c8475461 2876 transport = container_of(xprt, struct sock_xprt, xprt);
a246b010 2877
ec739ef0 2878 xprt->prot = IPPROTO_UDP;
808012fb 2879 xprt->tsh_size = 0;
a246b010
CL		 2880 /* XXX: header size can vary due to auth type, IPv6, etc. */
2881 xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
2882
03bf4b70 2883 xprt->bind_timeout = XS_BIND_TO;
03bf4b70
CL		 2884 xprt->reestablish_timeout = XS_UDP_REEST_TO;
2885 xprt->idle_timeout = XS_IDLE_DISC_TO;
a246b010 2886
262965f5 2887 xprt->ops = &xs_udp_ops;
a246b010 2888
ba7392bb 2889 xprt->timeout = &xs_udp_default_timeout;
a246b010 2890
f9b2ee71 2891 INIT_WORK(&transport->recv_worker, xs_udp_data_receive_workfn);
edc1b01c
TM		 2892 INIT_DELAYED_WORK(&transport->connect_worker, xs_udp_setup_socket);
2893
8f9d5b1a
CL		 2894 switch (addr->sa_family) {
2895 case AF_INET:
2896 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2897 xprt_set_bound(xprt);
2898
9dc3b095 2899 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
8f9d5b1a
CL		 2900 break;
2901 case AF_INET6:
2902 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2903 xprt_set_bound(xprt);
2904
9dc3b095 2905 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
8f9d5b1a
CL		 2906 break;
2907 default:
0a68b0be
BF		 2908 ret = ERR_PTR(-EAFNOSUPPORT);
2909 goto out_err;
8f9d5b1a
CL		 2910 }
2911
c740eff8
CL		 2912 if (xprt_bound(xprt))
2913 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2914 xprt->address_strings[RPC_DISPLAY_ADDR],
2915 xprt->address_strings[RPC_DISPLAY_PORT],
2916 xprt->address_strings[RPC_DISPLAY_PROTO]);
2917 else
2918 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
2919 xprt->address_strings[RPC_DISPLAY_ADDR],
2920 xprt->address_strings[RPC_DISPLAY_PROTO]);
edb267a6 2921
bc25571e
TT		 2922 if (try_module_get(THIS_MODULE))
2923 return xprt;
0a68b0be
BF		 2924 ret = ERR_PTR(-EINVAL);
2925out_err:
315f3812 2926 xs_xprt_free(xprt);
0a68b0be 2927 return ret;
a246b010
CL		 2928}
2929
2881ae74
TM		 2930static const struct rpc_timeout xs_tcp_default_timeout = {
2931 .to_initval = 60 * HZ,
2932 .to_maxval = 60 * HZ,
2933 .to_retries = 2,
2934};
2935
9903cd1c
CL		 2936/**
2937 * xs_setup_tcp - Set up transport to use a TCP socket
96802a09 2938 * @args: rpc transport creation arguments
9903cd1c
CL		 2939 *
2940 */
483066d6 2941static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
a246b010 2942{
8f9d5b1a 2943 struct sockaddr *addr = args->dstaddr;
c8541ecd 2944 struct rpc_xprt *xprt;
c8475461 2945 struct sock_xprt *transport;
0a68b0be 2946 struct rpc_xprt *ret;
b7993ceb
TM		 2947 unsigned int max_slot_table_size = xprt_max_tcp_slot_table_entries;
2948
2949 if (args->flags & XPRT_CREATE_INFINITE_SLOTS)
2950 max_slot_table_size = RPC_MAX_SLOT_TABLE_LIMIT;
a246b010 2951
d9ba131d 2952 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
b7993ceb 2953 max_slot_table_size);
c8541ecd
CL		 2954 if (IS_ERR(xprt))
2955 return xprt;
c8475461 2956 transport = container_of(xprt, struct sock_xprt, xprt);
a246b010 2957
ec739ef0 2958 xprt->prot = IPPROTO_TCP;
808012fb 2959 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
808012fb 2960 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
a246b010 2961
03bf4b70 2962 xprt->bind_timeout = XS_BIND_TO;
03bf4b70
CL		 2963 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2964 xprt->idle_timeout = XS_IDLE_DISC_TO;
a246b010 2965
262965f5 2966 xprt->ops = &xs_tcp_ops;
ba7392bb 2967 xprt->timeout = &xs_tcp_default_timeout;
a246b010 2968
edc1b01c
TM		 2969 INIT_WORK(&transport->recv_worker, xs_tcp_data_receive_workfn);
2970 INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_setup_socket);
2971
8f9d5b1a
CL		 2972 switch (addr->sa_family) {
2973 case AF_INET:
2974 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2975 xprt_set_bound(xprt);
2976
9dc3b095 2977 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
8f9d5b1a
CL		 2978 break;
2979 case AF_INET6:
2980 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2981 xprt_set_bound(xprt);
2982
9dc3b095 2983 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
8f9d5b1a
CL		 2984 break;
2985 default:
0a68b0be
BF		 2986 ret = ERR_PTR(-EAFNOSUPPORT);
2987 goto out_err;
8f9d5b1a
CL		 2988 }
2989
c740eff8
CL		 2990 if (xprt_bound(xprt))
2991 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2992 xprt->address_strings[RPC_DISPLAY_ADDR],
2993 xprt->address_strings[RPC_DISPLAY_PORT],
2994 xprt->address_strings[RPC_DISPLAY_PROTO]);
2995 else
2996 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
2997 xprt->address_strings[RPC_DISPLAY_ADDR],
2998 xprt->address_strings[RPC_DISPLAY_PROTO]);
2999
bc25571e
TT		 3000 if (try_module_get(THIS_MODULE))
3001 return xprt;
0a68b0be
BF		 3002 ret = ERR_PTR(-EINVAL);
3003out_err:
315f3812 3004 xs_xprt_free(xprt);
0a68b0be 3005 return ret;
a246b010 3006}
282b32e1 3007
f300baba
AB		 3008/**
3009 * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket
3010 * @args: rpc transport creation arguments
3011 *
3012 */
3013static struct rpc_xprt *xs_setup_bc_tcp(struct xprt_create *args)
3014{
3015 struct sockaddr *addr = args->dstaddr;
3016 struct rpc_xprt *xprt;
3017 struct sock_xprt *transport;
3018 struct svc_sock *bc_sock;
0a68b0be 3019 struct rpc_xprt *ret;
f300baba 3020
d9ba131d
TM		 3021 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
3022 xprt_tcp_slot_table_entries);
f300baba
AB		 3023 if (IS_ERR(xprt))
3024 return xprt;
3025 transport = container_of(xprt, struct sock_xprt, xprt);
3026
3027 xprt->prot = IPPROTO_TCP;
3028 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
3029 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
3030 xprt->timeout = &xs_tcp_default_timeout;
3031
3032 /* backchannel */
3033 xprt_set_bound(xprt);
3034 xprt->bind_timeout = 0;
f300baba
AB		 3035 xprt->reestablish_timeout = 0;
3036 xprt->idle_timeout = 0;
3037
f300baba
AB		 3038 xprt->ops = &bc_tcp_ops;
3039
3040 switch (addr->sa_family) {
3041 case AF_INET:
3042 xs_format_peer_addresses(xprt, "tcp",
3043 RPCBIND_NETID_TCP);
3044 break;
3045 case AF_INET6:
3046 xs_format_peer_addresses(xprt, "tcp",
3047 RPCBIND_NETID_TCP6);
3048 break;
3049 default:
0a68b0be
BF		 3050 ret = ERR_PTR(-EAFNOSUPPORT);
3051 goto out_err;
f300baba
AB		 3052 }
3053
50fa0d40
PE		 3054 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
3055 xprt->address_strings[RPC_DISPLAY_ADDR],
3056 xprt->address_strings[RPC_DISPLAY_PORT],
3057 xprt->address_strings[RPC_DISPLAY_PROTO]);
f300baba 3058
99de8ea9
BF		 3059 /*
3060 * Once we've associated a backchannel xprt with a connection,
28303ca3
WM		 3061 * we want to keep it around as long as the connection lasts,
3062 * in case we need to start using it for a backchannel again;
3063 * this reference won't be dropped until bc_xprt is destroyed.
99de8ea9
BF		 3064 */
3065 xprt_get(xprt);
3066 args->bc_xprt->xpt_bc_xprt = xprt;
3067 xprt->bc_xprt = args->bc_xprt;
3068 bc_sock = container_of(args->bc_xprt, struct svc_sock, sk_xprt);
3069 transport->sock = bc_sock->sk_sock;
3070 transport->inet = bc_sock->sk_sk;
3071
f300baba
AB		 3072 /*
3073 * Since we don't want connections for the backchannel, we set
3074 * the xprt status to connected
3075 */
3076 xprt_set_connected(xprt);
3077
f300baba
AB		 3078 if (try_module_get(THIS_MODULE))
3079 return xprt;
642aab58
KM		 3080
3081 args->bc_xprt->xpt_bc_xprt = NULL;
99de8ea9 3082 xprt_put(xprt);
0a68b0be
BF		 3083 ret = ERR_PTR(-EINVAL);
3084out_err:
315f3812 3085 xs_xprt_free(xprt);
0a68b0be 3086 return ret;
f300baba
AB		 3087}
3088
176e21ee
CL		 3089static struct xprt_class xs_local_transport = {
3090 .list = LIST_HEAD_INIT(xs_local_transport.list),
3091 .name = "named UNIX socket",
3092 .owner = THIS_MODULE,
3093 .ident = XPRT_TRANSPORT_LOCAL,
3094 .setup = xs_setup_local,
3095};
3096
bc25571e
TT		 3097static struct xprt_class xs_udp_transport = {
3098 .list = LIST_HEAD_INIT(xs_udp_transport.list),
3099 .name = "udp",
3100 .owner = THIS_MODULE,
f300baba 3101 .ident = XPRT_TRANSPORT_UDP,
bc25571e
TT		 3102 .setup = xs_setup_udp,
3103};
3104
3105static struct xprt_class xs_tcp_transport = {
3106 .list = LIST_HEAD_INIT(xs_tcp_transport.list),
3107 .name = "tcp",
3108 .owner = THIS_MODULE,
f300baba 3109 .ident = XPRT_TRANSPORT_TCP,
bc25571e
TT		 3110 .setup = xs_setup_tcp,
3111};
3112
f300baba
AB		 3113static struct xprt_class xs_bc_tcp_transport = {
3114 .list = LIST_HEAD_INIT(xs_bc_tcp_transport.list),
3115 .name = "tcp NFSv4.1 backchannel",
3116 .owner = THIS_MODULE,
3117 .ident = XPRT_TRANSPORT_BC_TCP,
3118 .setup = xs_setup_bc_tcp,
3119};
3120
282b32e1 3121/**
bc25571e 3122 * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
282b32e1
CL		 3123 *
3124 */
3125int init_socket_xprt(void)
3126{
f895b252 3127#if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
2b1bec5f 3128 if (!sunrpc_table_header)
0b4d4147 3129 sunrpc_table_header = register_sysctl_table(sunrpc_table);
fbf76683
CL		 3130#endif
3131
176e21ee 3132 xprt_register_transport(&xs_local_transport);
bc25571e
TT		 3133 xprt_register_transport(&xs_udp_transport);
3134 xprt_register_transport(&xs_tcp_transport);
f300baba 3135 xprt_register_transport(&xs_bc_tcp_transport);
bc25571e 3136
282b32e1
CL		 3137 return 0;
3138}
3139
3140/**
bc25571e 3141 * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
282b32e1
CL		 3142 *
3143 */
3144void cleanup_socket_xprt(void)
3145{
f895b252 3146#if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
fbf76683
CL		 3147 if (sunrpc_table_header) {
3148 unregister_sysctl_table(sunrpc_table_header);
3149 sunrpc_table_header = NULL;
3150 }
3151#endif
bc25571e 3152
176e21ee 3153 xprt_unregister_transport(&xs_local_transport);
bc25571e
TT		 3154 xprt_unregister_transport(&xs_udp_transport);
3155 xprt_unregister_transport(&xs_tcp_transport);
f300baba 3156 xprt_unregister_transport(&xs_bc_tcp_transport);
282b32e1 3157}
cbf11071 3158
9bbb9e5a
RR		 3159static int param_set_uint_minmax(const char *val,
3160 const struct kernel_param *kp,
cbf11071
TM		 3161 unsigned int min, unsigned int max)
3162{
00cfaa94 3163 unsigned int num;
cbf11071
TM		 3164 int ret;
3165
3166 if (!val)
3167 return -EINVAL;
00cfaa94 3168 ret = kstrtouint(val, 0, &num);
cbf11071
TM		 3169 if (ret == -EINVAL || num < min || num > max)
3170 return -EINVAL;
3171 *((unsigned int *)kp->arg) = num;
3172 return 0;
3173}
3174
9bbb9e5a 3175static int param_set_portnr(const char *val, const struct kernel_param *kp)
cbf11071
TM		 3176{
3177 return param_set_uint_minmax(val, kp,
3178 RPC_MIN_RESVPORT,
3179 RPC_MAX_RESVPORT);
3180}
3181
9c27847d 3182static const struct kernel_param_ops param_ops_portnr = {
9bbb9e5a
RR		 3183 .set = param_set_portnr,
3184 .get = param_get_uint,
3185};
3186
cbf11071
TM		 3187#define param_check_portnr(name, p) \
3188 __param_check(name, p, unsigned int);
3189
3190module_param_named(min_resvport, xprt_min_resvport, portnr, 0644);
3191module_param_named(max_resvport, xprt_max_resvport, portnr, 0644);
3192
9bbb9e5a
RR		 3193static int param_set_slot_table_size(const char *val,
3194 const struct kernel_param *kp)
cbf11071
TM		 3195{
3196 return param_set_uint_minmax(val, kp,
3197 RPC_MIN_SLOT_TABLE,
3198 RPC_MAX_SLOT_TABLE);
3199}
3200
9c27847d 3201static const struct kernel_param_ops param_ops_slot_table_size = {
9bbb9e5a
RR		 3202 .set = param_set_slot_table_size,
3203 .get = param_get_uint,
3204};
3205
cbf11071
TM		 3206#define param_check_slot_table_size(name, p) \
3207 __param_check(name, p, unsigned int);
3208
d9ba131d
TM		 3209static int param_set_max_slot_table_size(const char *val,
3210 const struct kernel_param *kp)
3211{
3212 return param_set_uint_minmax(val, kp,
3213 RPC_MIN_SLOT_TABLE,
3214 RPC_MAX_SLOT_TABLE_LIMIT);
3215}
3216
9c27847d 3217static const struct kernel_param_ops param_ops_max_slot_table_size = {
d9ba131d
TM		 3218 .set = param_set_max_slot_table_size,
3219 .get = param_get_uint,
3220};
3221
3222#define param_check_max_slot_table_size(name, p) \
3223 __param_check(name, p, unsigned int);
3224
cbf11071
TM		 3225module_param_named(tcp_slot_table_entries, xprt_tcp_slot_table_entries,
3226 slot_table_size, 0644);
d9ba131d
TM		 3227module_param_named(tcp_max_slot_table_entries, xprt_max_tcp_slot_table_entries,
3228 max_slot_table_size, 0644);
cbf11071
TM		 3229module_param_named(udp_slot_table_entries, xprt_udp_slot_table_entries,
3230 slot_table_size, 0644);
3231
Ubuntu Focal Linux kernel mirror