]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blame - net/sunrpc/xprtsock.c
projects / mirror_ubuntu-bionic-kernel.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
SUNRPC: Consolidate xs_tcp_data_ready and xs_data_ready
[mirror_ubuntu-bionic-kernel.git] / net / sunrpc / xprtsock.c
CommitLineData
a246b010
CL		 1/*
2 * linux/net/sunrpc/xprtsock.c
3 *
4 * Client-side transport implementation for sockets.
5 *
113aa838
AC		 6 * TCP callback races fixes (C) 1998 Red Hat
7 * TCP send fixes (C) 1998 Red Hat
a246b010
CL		 8 * TCP NFS related read + write fixes
9 * (C) 1999 Dave Airlie, University of Limerick, Ireland <airlied@linux.ie>
10 *
11 * Rewrite of larges part of the code in order to stabilize TCP stuff.
12 * Fix behaviour when socket buffer is full.
13 * (C) 1999 Trond Myklebust <trond.myklebust@fys.uio.no>
55aa4f58
CL		 14 *
15 * IP socket transport implementation, (C) 2005 Chuck Lever <cel@netapp.com>
8f9d5b1a
CL		 16 *
17 * IPv6 support contributed by Gilles Quillard, Bull Open Source, 2005.
18 * <gilles.quillard@bull.net>
a246b010
CL		 19 */
20
21#include <linux/types.h>
176e21ee 22#include <linux/string.h>
a246b010 23#include <linux/slab.h>
bc25571e 24#include <linux/module.h>
a246b010 25#include <linux/capability.h>
a246b010
CL		 26#include <linux/pagemap.h>
27#include <linux/errno.h>
28#include <linux/socket.h>
29#include <linux/in.h>
30#include <linux/net.h>
31#include <linux/mm.h>
176e21ee 32#include <linux/un.h>
a246b010
CL		 33#include <linux/udp.h>
34#include <linux/tcp.h>
35#include <linux/sunrpc/clnt.h>
5976687a 36#include <linux/sunrpc/addr.h>
02107148 37#include <linux/sunrpc/sched.h>
4cfc7e60 38#include <linux/sunrpc/svcsock.h>
49c36fcc 39#include <linux/sunrpc/xprtsock.h>
a246b010 40#include <linux/file.h>
9e00abc3 41#ifdef CONFIG_SUNRPC_BACKCHANNEL
44b98efd
RL		 42#include <linux/sunrpc/bc_xprt.h>
43#endif
a246b010
CL		 44
45#include <net/sock.h>
46#include <net/checksum.h>
47#include <net/udp.h>
48#include <net/tcp.h>
49
40b5ea0c
TM		 50#include <trace/events/sunrpc.h>
51
4cfc7e60 52#include "sunrpc.h"
176e21ee
CL		 53
54static void xs_close(struct rpc_xprt *xprt);
55
c556b754
CL		 56/*
57 * xprtsock tunables
58 */
09acfea5
TM		 59static unsigned int xprt_udp_slot_table_entries = RPC_DEF_SLOT_TABLE;
60static unsigned int xprt_tcp_slot_table_entries = RPC_MIN_SLOT_TABLE;
61static unsigned int xprt_max_tcp_slot_table_entries = RPC_MAX_SLOT_TABLE;
c556b754 62
09acfea5
TM		 63static unsigned int xprt_min_resvport = RPC_DEF_MIN_RESVPORT;
64static unsigned int xprt_max_resvport = RPC_DEF_MAX_RESVPORT;
c556b754 65
54c09874
TM		 66#if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
67
7d1e8255 68#define XS_TCP_LINGER_TO (15U * HZ)
25fe6142 69static unsigned int xs_tcp_fin_timeout __read_mostly = XS_TCP_LINGER_TO;
7d1e8255 70
fbf76683
CL		 71/*
72 * We can register our own files under /proc/sys/sunrpc by
73 * calling register_sysctl_table() again. The files in that
74 * directory become the union of all files registered there.
75 *
76 * We simply need to make sure that we don't collide with
77 * someone else's file names!
78 */
79
fbf76683
CL		 80static unsigned int min_slot_table_size = RPC_MIN_SLOT_TABLE;
81static unsigned int max_slot_table_size = RPC_MAX_SLOT_TABLE;
d9ba131d 82static unsigned int max_tcp_slot_table_limit = RPC_MAX_SLOT_TABLE_LIMIT;
fbf76683
CL		 83static unsigned int xprt_min_resvport_limit = RPC_MIN_RESVPORT;
84static unsigned int xprt_max_resvport_limit = RPC_MAX_RESVPORT;
85
86static struct ctl_table_header *sunrpc_table_header;
87
88/*
89 * FIXME: changing the UDP slot table size should also resize the UDP
90 * socket buffers for existing UDP transports
91 */
fe2c6338 92static struct ctl_table xs_tunables_table[] = {
fbf76683 93 {
fbf76683
CL		 94 .procname = "udp_slot_table_entries",
95 .data = &xprt_udp_slot_table_entries,
96 .maxlen = sizeof(unsigned int),
97 .mode = 0644,
6d456111 98 .proc_handler = proc_dointvec_minmax,
fbf76683
CL		 99 .extra1 = &min_slot_table_size,
100 .extra2 = &max_slot_table_size
101 },
102 {
fbf76683
CL		 103 .procname = "tcp_slot_table_entries",
104 .data = &xprt_tcp_slot_table_entries,
105 .maxlen = sizeof(unsigned int),
106 .mode = 0644,
6d456111 107 .proc_handler = proc_dointvec_minmax,
fbf76683
CL		 108 .extra1 = &min_slot_table_size,
109 .extra2 = &max_slot_table_size
110 },
d9ba131d
TM		 111 {
112 .procname = "tcp_max_slot_table_entries",
113 .data = &xprt_max_tcp_slot_table_entries,
114 .maxlen = sizeof(unsigned int),
115 .mode = 0644,
116 .proc_handler = proc_dointvec_minmax,
117 .extra1 = &min_slot_table_size,
118 .extra2 = &max_tcp_slot_table_limit
119 },
fbf76683 120 {
fbf76683
CL		 121 .procname = "min_resvport",
122 .data = &xprt_min_resvport,
123 .maxlen = sizeof(unsigned int),
124 .mode = 0644,
6d456111 125 .proc_handler = proc_dointvec_minmax,
fbf76683
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);
5157b956
TM		 1091 transport->old_data_ready(sk);
1092 /* Any data means we had a useful conversation, so
1093 * then we don't need to delay the next reconnect
1094 */
1095 if (xprt->reestablish_timeout)
1096 xprt->reestablish_timeout = 0;
42d42a5b
TM		 1097 if (!test_and_set_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
1098 queue_work(rpciod_workqueue, &transport->recv_worker);
f9b2ee71 1099 }
f064af1e 1100 read_unlock_bh(&sk->sk_callback_lock);
a246b010
CL		 1101}
1102
a519fc7a
TM		 1103/*
1104 * Helper function to force a TCP close if the server is sending
1105 * junk and/or it has put us in CLOSE_WAIT
1106 */
1107static void xs_tcp_force_close(struct rpc_xprt *xprt)
1108{
a519fc7a
TM		 1109 xprt_force_disconnect(xprt);
1110}
1111
dd456471 1112static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
a246b010 1113{
51971139 1114 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
a246b010
CL		 1115 size_t len, used;
1116 char *p;
1117
51971139
CL		 1118 p = ((char *) &transport->tcp_fraghdr) + transport->tcp_offset;
1119 len = sizeof(transport->tcp_fraghdr) - transport->tcp_offset;
9d292316 1120 used = xdr_skb_read_bits(desc, p, len);
51971139 1121 transport->tcp_offset += used;
a246b010
CL		 1122 if (used != len)
1123 return;
808012fb 1124
51971139
CL		 1125 transport->tcp_reclen = ntohl(transport->tcp_fraghdr);
1126 if (transport->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
e136d092 1127 transport->tcp_flags |= TCP_RCV_LAST_FRAG;
a246b010 1128 else
e136d092 1129 transport->tcp_flags &= ~TCP_RCV_LAST_FRAG;
51971139 1130 transport->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
808012fb 1131
e136d092 1132 transport->tcp_flags &= ~TCP_RCV_COPY_FRAGHDR;
51971139 1133 transport->tcp_offset = 0;
808012fb 1134
a246b010 1135 /* Sanity check of the record length */
18dca02a 1136 if (unlikely(transport->tcp_reclen < 8)) {
46121cf7 1137 dprintk("RPC: invalid TCP record fragment length\n");
a519fc7a 1138 xs_tcp_force_close(xprt);
9903cd1c 1139 return;
a246b010 1140 }
46121cf7 1141 dprintk("RPC: reading TCP record fragment of length %d\n",
51971139 1142 transport->tcp_reclen);
a246b010
CL		 1143}
1144
51971139 1145static void xs_tcp_check_fraghdr(struct sock_xprt *transport)
a246b010 1146{
51971139 1147 if (transport->tcp_offset == transport->tcp_reclen) {
e136d092 1148 transport->tcp_flags |= TCP_RCV_COPY_FRAGHDR;
51971139 1149 transport->tcp_offset = 0;
e136d092
CL		 1150 if (transport->tcp_flags & TCP_RCV_LAST_FRAG) {
1151 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1152 transport->tcp_flags |= TCP_RCV_COPY_XID;
51971139 1153 transport->tcp_copied = 0;
a246b010
CL		 1154 }
1155 }
1156}
1157
dd456471 1158static inline void xs_tcp_read_xid(struct sock_xprt *transport, struct xdr_skb_reader *desc)
a246b010
CL		 1159{
1160 size_t len, used;
1161 char *p;
1162
51971139 1163 len = sizeof(transport->tcp_xid) - transport->tcp_offset;
46121cf7 1164 dprintk("RPC: reading XID (%Zu bytes)\n", len);
51971139 1165 p = ((char *) &transport->tcp_xid) + transport->tcp_offset;
9d292316 1166 used = xdr_skb_read_bits(desc, p, len);
51971139 1167 transport->tcp_offset += used;
a246b010
CL		 1168 if (used != len)
1169 return;
e136d092 1170 transport->tcp_flags &= ~TCP_RCV_COPY_XID;
f4a2e418 1171 transport->tcp_flags |= TCP_RCV_READ_CALLDIR;
51971139 1172 transport->tcp_copied = 4;
18dca02a
RL		 1173 dprintk("RPC: reading %s XID %08x\n",
1174 (transport->tcp_flags & TCP_RPC_REPLY) ? "reply for"
1175 : "request with",
51971139
CL		 1176 ntohl(transport->tcp_xid));
1177 xs_tcp_check_fraghdr(transport);
a246b010
CL		 1178}
1179
18dca02a
RL		 1180static inline void xs_tcp_read_calldir(struct sock_xprt *transport,
1181 struct xdr_skb_reader *desc)
a246b010 1182{
18dca02a
RL		 1183 size_t len, used;
1184 u32 offset;
b76ce561 1185 char *p;
18dca02a
RL		 1186
1187 /*
1188 * We want transport->tcp_offset to be 8 at the end of this routine
1189 * (4 bytes for the xid and 4 bytes for the call/reply flag).
1190 * When this function is called for the first time,
1191 * transport->tcp_offset is 4 (after having already read the xid).
1192 */
1193 offset = transport->tcp_offset - sizeof(transport->tcp_xid);
b76ce561 1194 len = sizeof(transport->tcp_calldir) - offset;
18dca02a 1195 dprintk("RPC: reading CALL/REPLY flag (%Zu bytes)\n", len);
b76ce561
TM		 1196 p = ((char *) &transport->tcp_calldir) + offset;
1197 used = xdr_skb_read_bits(desc, p, len);
18dca02a
RL		 1198 transport->tcp_offset += used;
1199 if (used != len)
1200 return;
f4a2e418 1201 transport->tcp_flags &= ~TCP_RCV_READ_CALLDIR;
f4a2e418
RL		 1202 /*
1203 * We don't yet have the XDR buffer, so we will write the calldir
1204 * out after we get the buffer from the 'struct rpc_rqst'
1205 */
b76ce561
TM		 1206 switch (ntohl(transport->tcp_calldir)) {
1207 case RPC_REPLY:
1208 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1209 transport->tcp_flags |= TCP_RCV_COPY_DATA;
18dca02a 1210 transport->tcp_flags |= TCP_RPC_REPLY;
b76ce561
TM		 1211 break;
1212 case RPC_CALL:
1213 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1214 transport->tcp_flags |= TCP_RCV_COPY_DATA;
18dca02a 1215 transport->tcp_flags &= ~TCP_RPC_REPLY;
b76ce561
TM		 1216 break;
1217 default:
1218 dprintk("RPC: invalid request message type\n");
a519fc7a 1219 xs_tcp_force_close(&transport->xprt);
b76ce561 1220 }
18dca02a
RL		 1221 xs_tcp_check_fraghdr(transport);
1222}
1223
44b98efd
RL		 1224static inline void xs_tcp_read_common(struct rpc_xprt *xprt,
1225 struct xdr_skb_reader *desc,
1226 struct rpc_rqst *req)
a246b010 1227{
44b98efd
RL		 1228 struct sock_xprt *transport =
1229 container_of(xprt, struct sock_xprt, xprt);
a246b010
CL		 1230 struct xdr_buf *rcvbuf;
1231 size_t len;
1232 ssize_t r;
1233
a246b010 1234 rcvbuf = &req->rq_private_buf;
f4a2e418
RL		 1235
1236 if (transport->tcp_flags & TCP_RCV_COPY_CALLDIR) {
1237 /*
1238 * Save the RPC direction in the XDR buffer
1239 */
f4a2e418 1240 memcpy(rcvbuf->head[0].iov_base + transport->tcp_copied,
b76ce561
TM		 1241 &transport->tcp_calldir,
1242 sizeof(transport->tcp_calldir));
1243 transport->tcp_copied += sizeof(transport->tcp_calldir);
f4a2e418 1244 transport->tcp_flags &= ~TCP_RCV_COPY_CALLDIR;
a246b010
CL		 1245 }
1246
a246b010 1247 len = desc->count;
51971139 1248 if (len > transport->tcp_reclen - transport->tcp_offset) {
dd456471 1249 struct xdr_skb_reader my_desc;
a246b010 1250
51971139 1251 len = transport->tcp_reclen - transport->tcp_offset;
a246b010
CL		 1252 memcpy(&my_desc, desc, sizeof(my_desc));
1253 my_desc.count = len;
51971139 1254 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
9d292316 1255 &my_desc, xdr_skb_read_bits);
a246b010
CL		 1256 desc->count -= r;
1257 desc->offset += r;
1258 } else
51971139 1259 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
9d292316 1260 desc, xdr_skb_read_bits);
a246b010
CL		 1261
1262 if (r > 0) {
51971139
CL		 1263 transport->tcp_copied += r;
1264 transport->tcp_offset += r;
a246b010
CL		 1265 }
1266 if (r != len) {
1267 /* Error when copying to the receive buffer,
1268 * usually because we weren't able to allocate
1269 * additional buffer pages. All we can do now
e136d092 1270 * is turn off TCP_RCV_COPY_DATA, so the request
a246b010
CL		 1271 * will not receive any additional updates,
1272 * and time out.
1273 * Any remaining data from this record will
1274 * be discarded.
1275 */
e136d092 1276 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
46121cf7 1277 dprintk("RPC: XID %08x truncated request\n",
51971139 1278 ntohl(transport->tcp_xid));
46121cf7
CL		 1279 dprintk("RPC: xprt = %p, tcp_copied = %lu, "
1280 "tcp_offset = %u, tcp_reclen = %u\n",
1281 xprt, transport->tcp_copied,
1282 transport->tcp_offset, transport->tcp_reclen);
44b98efd 1283 return;
a246b010
CL		 1284 }
1285
46121cf7 1286 dprintk("RPC: XID %08x read %Zd bytes\n",
51971139 1287 ntohl(transport->tcp_xid), r);
46121cf7
CL		 1288 dprintk("RPC: xprt = %p, tcp_copied = %lu, tcp_offset = %u, "
1289 "tcp_reclen = %u\n", xprt, transport->tcp_copied,
1290 transport->tcp_offset, transport->tcp_reclen);
51971139
CL		 1291
1292 if (transport->tcp_copied == req->rq_private_buf.buflen)
e136d092 1293 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
51971139 1294 else if (transport->tcp_offset == transport->tcp_reclen) {
e136d092
CL		 1295 if (transport->tcp_flags & TCP_RCV_LAST_FRAG)
1296 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
a246b010 1297 }
44b98efd
RL		 1298}
1299
1300/*
1301 * Finds the request corresponding to the RPC xid and invokes the common
1302 * tcp read code to read the data.
1303 */
1304static inline int xs_tcp_read_reply(struct rpc_xprt *xprt,
1305 struct xdr_skb_reader *desc)
1306{
1307 struct sock_xprt *transport =
1308 container_of(xprt, struct sock_xprt, xprt);
1309 struct rpc_rqst *req;
1310
1311 dprintk("RPC: read reply XID %08x\n", ntohl(transport->tcp_xid));
1312
1313 /* Find and lock the request corresponding to this xid */
edc1b01c 1314 spin_lock_bh(&xprt->transport_lock);
44b98efd
RL		 1315 req = xprt_lookup_rqst(xprt, transport->tcp_xid);
1316 if (!req) {
1317 dprintk("RPC: XID %08x request not found!\n",
1318 ntohl(transport->tcp_xid));
edc1b01c 1319 spin_unlock_bh(&xprt->transport_lock);
44b98efd
RL		 1320 return -1;
1321 }
1322
1323 xs_tcp_read_common(xprt, desc, req);
1324
e136d092 1325 if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
51971139 1326 xprt_complete_rqst(req->rq_task, transport->tcp_copied);
44b98efd 1327
edc1b01c 1328 spin_unlock_bh(&xprt->transport_lock);
44b98efd
RL		 1329 return 0;
1330}
1331
9e00abc3 1332#if defined(CONFIG_SUNRPC_BACKCHANNEL)
44b98efd
RL		 1333/*
1334 * Obtains an rpc_rqst previously allocated and invokes the common
1335 * tcp read code to read the data. The result is placed in the callback
1336 * queue.
1337 * If we're unable to obtain the rpc_rqst we schedule the closing of the
1338 * connection and return -1.
1339 */
2ea24497 1340static int xs_tcp_read_callback(struct rpc_xprt *xprt,
44b98efd
RL		 1341 struct xdr_skb_reader *desc)
1342{
1343 struct sock_xprt *transport =
1344 container_of(xprt, struct sock_xprt, xprt);
1345 struct rpc_rqst *req;
1346
2ea24497 1347 /* Look up and lock the request corresponding to the given XID */
edc1b01c 1348 spin_lock_bh(&xprt->transport_lock);
2ea24497 1349 req = xprt_lookup_bc_request(xprt, transport->tcp_xid);
44b98efd 1350 if (req == NULL) {
edc1b01c 1351 spin_unlock_bh(&xprt->transport_lock);
44b98efd
RL		 1352 printk(KERN_WARNING "Callback slot table overflowed\n");
1353 xprt_force_disconnect(xprt);
1354 return -1;
1355 }
1356
44b98efd
RL		 1357 dprintk("RPC: read callback XID %08x\n", ntohl(req->rq_xid));
1358 xs_tcp_read_common(xprt, desc, req);
1359
2ea24497
TM		 1360 if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1361 xprt_complete_bc_request(req, transport->tcp_copied);
edc1b01c 1362 spin_unlock_bh(&xprt->transport_lock);
44b98efd
RL		 1363
1364 return 0;
1365}
1366
1367static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1368 struct xdr_skb_reader *desc)
1369{
1370 struct sock_xprt *transport =
1371 container_of(xprt, struct sock_xprt, xprt);
1372
1373 return (transport->tcp_flags & TCP_RPC_REPLY) ?
1374 xs_tcp_read_reply(xprt, desc) :
1375 xs_tcp_read_callback(xprt, desc);
1376}
76566773
CL		 1377
1378static int xs_tcp_bc_up(struct svc_serv *serv, struct net *net)
1379{
1380 int ret;
1381
1382 ret = svc_create_xprt(serv, "tcp-bc", net, PF_INET, 0,
1383 SVC_SOCK_ANONYMOUS);
1384 if (ret < 0)
1385 return ret;
1386 return 0;
1387}
6b26cc8c
CL		 1388
1389static size_t xs_tcp_bc_maxpayload(struct rpc_xprt *xprt)
1390{
1391 return PAGE_SIZE;
1392}
44b98efd
RL		 1393#else
1394static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1395 struct xdr_skb_reader *desc)
1396{
1397 return xs_tcp_read_reply(xprt, desc);
1398}
9e00abc3 1399#endif /* CONFIG_SUNRPC_BACKCHANNEL */
44b98efd
RL		 1400
1401/*
1402 * Read data off the transport. This can be either an RPC_CALL or an
1403 * RPC_REPLY. Relay the processing to helper functions.
1404 */
1405static void xs_tcp_read_data(struct rpc_xprt *xprt,
1406 struct xdr_skb_reader *desc)
1407{
1408 struct sock_xprt *transport =
1409 container_of(xprt, struct sock_xprt, xprt);
1410
1411 if (_xs_tcp_read_data(xprt, desc) == 0)
1412 xs_tcp_check_fraghdr(transport);
1413 else {
1414 /*
1415 * The transport_lock protects the request handling.
1416 * There's no need to hold it to update the tcp_flags.
1417 */
1418 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1419 }
a246b010
CL		 1420}
1421
dd456471 1422static inline void xs_tcp_read_discard(struct sock_xprt *transport, struct xdr_skb_reader *desc)
a246b010
CL		 1423{
1424 size_t len;
1425
51971139 1426 len = transport->tcp_reclen - transport->tcp_offset;
a246b010
CL		 1427 if (len > desc->count)
1428 len = desc->count;
1429 desc->count -= len;
1430 desc->offset += len;
51971139 1431 transport->tcp_offset += len;
46121cf7 1432 dprintk("RPC: discarded %Zu bytes\n", len);
51971139 1433 xs_tcp_check_fraghdr(transport);
a246b010
CL		 1434}
1435
9903cd1c 1436static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
a246b010
CL		 1437{
1438 struct rpc_xprt *xprt = rd_desc->arg.data;
51971139 1439 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
dd456471 1440 struct xdr_skb_reader desc = {
a246b010
CL		 1441 .skb = skb,
1442 .offset = offset,
1443 .count = len,
9903cd1c 1444 };
a246b010 1445
46121cf7 1446 dprintk("RPC: xs_tcp_data_recv started\n");
a246b010 1447 do {
1a867a08 1448 trace_xs_tcp_data_recv(transport);
a246b010
CL		 1449 /* Read in a new fragment marker if necessary */
1450 /* Can we ever really expect to get completely empty fragments? */
e136d092 1451 if (transport->tcp_flags & TCP_RCV_COPY_FRAGHDR) {
9903cd1c 1452 xs_tcp_read_fraghdr(xprt, &desc);
a246b010
CL		 1453 continue;
1454 }
1455 /* Read in the xid if necessary */
e136d092 1456 if (transport->tcp_flags & TCP_RCV_COPY_XID) {
51971139 1457 xs_tcp_read_xid(transport, &desc);
a246b010
CL		 1458 continue;
1459 }
18dca02a 1460 /* Read in the call/reply flag */
f4a2e418 1461 if (transport->tcp_flags & TCP_RCV_READ_CALLDIR) {
18dca02a
RL		 1462 xs_tcp_read_calldir(transport, &desc);
1463 continue;
1464 }
a246b010 1465 /* Read in the request data */
e136d092 1466 if (transport->tcp_flags & TCP_RCV_COPY_DATA) {
44b98efd 1467 xs_tcp_read_data(xprt, &desc);
a246b010
CL		 1468 continue;
1469 }
1470 /* Skip over any trailing bytes on short reads */
51971139 1471 xs_tcp_read_discard(transport, &desc);
a246b010 1472 } while (desc.count);
1a867a08 1473 trace_xs_tcp_data_recv(transport);
46121cf7 1474 dprintk("RPC: xs_tcp_data_recv done\n");
a246b010
CL		 1475 return len - desc.count;
1476}
1477
66d7a56a
TM		 1478static void xs_tcp_data_receive(struct sock_xprt *transport)
1479{
1480 struct rpc_xprt *xprt = &transport->xprt;
1481 struct sock *sk;
1482 read_descriptor_t rd_desc = {
1483 .count = 2*1024*1024,
1484 .arg.data = xprt,
1485 };
1486 unsigned long total = 0;
1487 int read = 0;
1488
edc1b01c 1489 mutex_lock(&transport->recv_mutex);
66d7a56a 1490 sk = transport->inet;
edc1b01c
TM		 1491 if (sk == NULL)
1492 goto out;
66d7a56a
TM		 1493
1494 /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
1495 for (;;) {
edc1b01c 1496 lock_sock(sk);
66d7a56a 1497 read = tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
42d42a5b
TM		 1498 if (read <= 0) {
1499 clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state);
1500 release_sock(sk);
1501 if (!test_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
1502 break;
1503 } else {
1504 release_sock(sk);
1505 total += read;
1506 }
66d7a56a
TM		 1507 rd_desc.count = 65536;
1508 }
edc1b01c
TM		 1509out:
1510 mutex_unlock(&transport->recv_mutex);
66d7a56a
TM		 1511 trace_xs_tcp_data_ready(xprt, read, total);
1512}
1513
edc1b01c
TM		 1514static void xs_tcp_data_receive_workfn(struct work_struct *work)
1515{
1516 struct sock_xprt *transport =
1517 container_of(work, struct sock_xprt, recv_worker);
1518 xs_tcp_data_receive(transport);
1519}
1520
9903cd1c
CL		 1521/**
1522 * xs_tcp_state_change - callback to handle TCP socket state changes
1523 * @sk: socket whose state has changed
1524 *
1525 */
1526static void xs_tcp_state_change(struct sock *sk)
a246b010 1527{
9903cd1c 1528 struct rpc_xprt *xprt;
0fdea1e8 1529 struct sock_xprt *transport;
a246b010 1530
f064af1e 1531 read_lock_bh(&sk->sk_callback_lock);
a246b010
CL		 1532 if (!(xprt = xprt_from_sock(sk)))
1533 goto out;
46121cf7 1534 dprintk("RPC: xs_tcp_state_change client %p...\n", xprt);
669502ff 1535 dprintk("RPC: state %x conn %d dead %d zapped %d sk_shutdown %d\n",
46121cf7
CL		 1536 sk->sk_state, xprt_connected(xprt),
1537 sock_flag(sk, SOCK_DEAD),
669502ff
AC		 1538 sock_flag(sk, SOCK_ZAPPED),
1539 sk->sk_shutdown);
a246b010 1540
0fdea1e8 1541 transport = container_of(xprt, struct sock_xprt, xprt);
40b5ea0c 1542 trace_rpc_socket_state_change(xprt, sk->sk_socket);
a246b010
CL		 1543 switch (sk->sk_state) {
1544 case TCP_ESTABLISHED:
f064af1e 1545 spin_lock(&xprt->transport_lock);
a246b010 1546 if (!xprt_test_and_set_connected(xprt)) {
51971139 1547
a246b010 1548 /* Reset TCP record info */
51971139
CL		 1549 transport->tcp_offset = 0;
1550 transport->tcp_reclen = 0;
1551 transport->tcp_copied = 0;
e136d092
CL		 1552 transport->tcp_flags =
1553 TCP_RCV_COPY_FRAGHDR | TCP_RCV_COPY_XID;
8b71798c 1554 xprt->connect_cookie++;
0fdea1e8
TM		 1555 clear_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
1556 xprt_clear_connecting(xprt);
51971139 1557
2a491991 1558 xprt_wake_pending_tasks(xprt, -EAGAIN);
a246b010 1559 }
f064af1e 1560 spin_unlock(&xprt->transport_lock);
a246b010 1561 break;
3b948ae5
TM		 1562 case TCP_FIN_WAIT1:
1563 /* The client initiated a shutdown of the socket */
7c1d71cf 1564 xprt->connect_cookie++;
663b8858 1565 xprt->reestablish_timeout = 0;
3b948ae5 1566 set_bit(XPRT_CLOSING, &xprt->state);
4e857c58 1567 smp_mb__before_atomic();
3b948ae5 1568 clear_bit(XPRT_CONNECTED, &xprt->state);
ef803670 1569 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
4e857c58 1570 smp_mb__after_atomic();
a246b010 1571 break;
632e3bdc 1572 case TCP_CLOSE_WAIT:
3b948ae5 1573 /* The server initiated a shutdown of the socket */
7c1d71cf 1574 xprt->connect_cookie++;
d0bea455 1575 clear_bit(XPRT_CONNECTED, &xprt->state);
a519fc7a 1576 xs_tcp_force_close(xprt);
663b8858
TM		 1577 case TCP_CLOSING:
1578 /*
1579 * If the server closed down the connection, make sure that
1580 * we back off before reconnecting
1581 */
1582 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1583 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
3b948ae5
TM		 1584 break;
1585 case TCP_LAST_ACK:
670f9457 1586 set_bit(XPRT_CLOSING, &xprt->state);
4e857c58 1587 smp_mb__before_atomic();
3b948ae5 1588 clear_bit(XPRT_CONNECTED, &xprt->state);
4e857c58 1589 smp_mb__after_atomic();
3b948ae5
TM		 1590 break;
1591 case TCP_CLOSE:
0fdea1e8
TM		 1592 if (test_and_clear_bit(XPRT_SOCK_CONNECTING,
1593 &transport->sock_state))
1594 xprt_clear_connecting(xprt);
7d1e8255 1595 xs_sock_mark_closed(xprt);
a246b010
CL		 1596 }
1597 out:
f064af1e 1598 read_unlock_bh(&sk->sk_callback_lock);
a246b010
CL		 1599}
1600
1f0fa154
IJ		 1601static void xs_write_space(struct sock *sk)
1602{
13331a55 1603 struct socket_wq *wq;
1f0fa154
IJ		 1604 struct rpc_xprt *xprt;
1605
13331a55 1606 if (!sk->sk_socket)
1f0fa154 1607 return;
13331a55 1608 clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1f0fa154
IJ		 1609
1610 if (unlikely(!(xprt = xprt_from_sock(sk))))
1611 return;
13331a55
TM		 1612 rcu_read_lock();
1613 wq = rcu_dereference(sk->sk_wq);
1614 if (!wq || test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &wq->flags) == 0)
1615 goto out;
1f0fa154
IJ		 1616
1617 xprt_write_space(xprt);
13331a55
TM		 1618out:
1619 rcu_read_unlock();
1f0fa154
IJ		 1620}
1621
9903cd1c 1622/**
c7b2cae8
CL		 1623 * xs_udp_write_space - callback invoked when socket buffer space
1624 * becomes available
9903cd1c
CL		 1625 * @sk: socket whose state has changed
1626 *
a246b010
CL		 1627 * Called when more output buffer space is available for this socket.
1628 * We try not to wake our writers until they can make "significant"
c7b2cae8 1629 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
a246b010
CL		 1630 * with a bunch of small requests.
1631 */
c7b2cae8 1632static void xs_udp_write_space(struct sock *sk)
a246b010 1633{
f064af1e 1634 read_lock_bh(&sk->sk_callback_lock);
a246b010 1635
c7b2cae8 1636 /* from net/core/sock.c:sock_def_write_space */
1f0fa154
IJ		 1637 if (sock_writeable(sk))
1638 xs_write_space(sk);
a246b010 1639
f064af1e 1640 read_unlock_bh(&sk->sk_callback_lock);
c7b2cae8 1641}
a246b010 1642
c7b2cae8
CL		 1643/**
1644 * xs_tcp_write_space - callback invoked when socket buffer space
1645 * becomes available
1646 * @sk: socket whose state has changed
1647 *
1648 * Called when more output buffer space is available for this socket.
1649 * We try not to wake our writers until they can make "significant"
1650 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1651 * with a bunch of small requests.
1652 */
1653static void xs_tcp_write_space(struct sock *sk)
1654{
f064af1e 1655 read_lock_bh(&sk->sk_callback_lock);
c7b2cae8
CL		 1656
1657 /* from net/core/stream.c:sk_stream_write_space */
64dc6130 1658 if (sk_stream_is_writeable(sk))
1f0fa154 1659 xs_write_space(sk);
b6ddf64f 1660
f064af1e 1661 read_unlock_bh(&sk->sk_callback_lock);
a246b010
CL		 1662}
1663
470056c2 1664static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
a246b010 1665{
ee0ac0c2
CL		 1666 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1667 struct sock *sk = transport->inet;
a246b010 1668
7c6e066e 1669 if (transport->rcvsize) {
a246b010 1670 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
7c6e066e 1671 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
a246b010 1672 }
7c6e066e 1673 if (transport->sndsize) {
a246b010 1674 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
7c6e066e 1675 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
a246b010
CL		 1676 sk->sk_write_space(sk);
1677 }
1678}
1679
43118c29 1680/**
470056c2 1681 * xs_udp_set_buffer_size - set send and receive limits
43118c29 1682 * @xprt: generic transport
470056c2
CL		 1683 * @sndsize: requested size of send buffer, in bytes
1684 * @rcvsize: requested size of receive buffer, in bytes
43118c29 1685 *
470056c2 1686 * Set socket send and receive buffer size limits.
43118c29 1687 */
470056c2 1688static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
43118c29 1689{
7c6e066e
CL		 1690 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1691
1692 transport->sndsize = 0;
470056c2 1693 if (sndsize)
7c6e066e
CL		 1694 transport->sndsize = sndsize + 1024;
1695 transport->rcvsize = 0;
470056c2 1696 if (rcvsize)
7c6e066e 1697 transport->rcvsize = rcvsize + 1024;
470056c2
CL		 1698
1699 xs_udp_do_set_buffer_size(xprt);
43118c29
CL		 1700}
1701
46c0ee8b
CL		 1702/**
1703 * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1704 * @task: task that timed out
1705 *
1706 * Adjust the congestion window after a retransmit timeout has occurred.
1707 */
6a24dfb6 1708static void xs_udp_timer(struct rpc_xprt *xprt, struct rpc_task *task)
46c0ee8b 1709{
6a24dfb6 1710 xprt_adjust_cwnd(xprt, task, -ETIMEDOUT);
46c0ee8b
CL		 1711}
1712
b85d8806
CL		 1713static unsigned short xs_get_random_port(void)
1714{
1715 unsigned short range = xprt_max_resvport - xprt_min_resvport;
63862b5b 1716 unsigned short rand = (unsigned short) prandom_u32() % range;
b85d8806
CL		 1717 return rand + xprt_min_resvport;
1718}
1719
4dda9c8a
TM		 1720/**
1721 * xs_set_reuseaddr_port - set the socket's port and address reuse options
1722 * @sock: socket
1723 *
1724 * Note that this function has to be called on all sockets that share the
1725 * same port, and it must be called before binding.
1726 */
1727static void xs_sock_set_reuseport(struct socket *sock)
1728{
402e23b4 1729 int opt = 1;
4dda9c8a 1730
402e23b4
TM		 1731 kernel_setsockopt(sock, SOL_SOCKET, SO_REUSEPORT,
1732 (char *)&opt, sizeof(opt));
4dda9c8a
TM		 1733}
1734
1735static unsigned short xs_sock_getport(struct socket *sock)
1736{
1737 struct sockaddr_storage buf;
1738 int buflen;
1739 unsigned short port = 0;
1740
1741 if (kernel_getsockname(sock, (struct sockaddr *)&buf, &buflen) < 0)
1742 goto out;
1743 switch (buf.ss_family) {
1744 case AF_INET6:
1745 port = ntohs(((struct sockaddr_in6 *)&buf)->sin6_port);
1746 break;
1747 case AF_INET:
1748 port = ntohs(((struct sockaddr_in *)&buf)->sin_port);
1749 }
1750out:
1751 return port;
1752}
1753
92200412
CL		 1754/**
1755 * xs_set_port - reset the port number in the remote endpoint address
1756 * @xprt: generic transport
1757 * @port: new port number
1758 *
1759 */
1760static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1761{
46121cf7 1762 dprintk("RPC: setting port for xprt %p to %u\n", xprt, port);
c4efcb1d 1763
9dc3b095
CL		 1764 rpc_set_port(xs_addr(xprt), port);
1765 xs_update_peer_port(xprt);
92200412
CL		 1766}
1767
4dda9c8a
TM		 1768static void xs_set_srcport(struct sock_xprt *transport, struct socket *sock)
1769{
1770 if (transport->srcport == 0)
1771 transport->srcport = xs_sock_getport(sock);
1772}
1773
5d4ec932 1774static unsigned short xs_get_srcport(struct sock_xprt *transport)
67a391d7 1775{
fbfffbd5 1776 unsigned short port = transport->srcport;
67a391d7
TM		 1777
1778 if (port == 0 && transport->xprt.resvport)
1779 port = xs_get_random_port();
1780 return port;
1781}
1782
baaf4e48 1783static unsigned short xs_next_srcport(struct sock_xprt *transport, unsigned short port)
67a391d7 1784{
fbfffbd5
CL		 1785 if (transport->srcport != 0)
1786 transport->srcport = 0;
67a391d7
TM		 1787 if (!transport->xprt.resvport)
1788 return 0;
1789 if (port <= xprt_min_resvport || port > xprt_max_resvport)
1790 return xprt_max_resvport;
1791 return --port;
1792}
beb59b68 1793static int xs_bind(struct sock_xprt *transport, struct socket *sock)
a246b010 1794{
beb59b68 1795 struct sockaddr_storage myaddr;
67a391d7 1796 int err, nloop = 0;
5d4ec932 1797 unsigned short port = xs_get_srcport(transport);
67a391d7 1798 unsigned short last;
a246b010 1799
0f7a622c
CP		 1800 /*
1801 * If we are asking for any ephemeral port (i.e. port == 0 &&
1802 * transport->xprt.resvport == 0), don't bind. Let the local
1803 * port selection happen implicitly when the socket is used
1804 * (for example at connect time).
1805 *
1806 * This ensures that we can continue to establish TCP
1807 * connections even when all local ephemeral ports are already
1808 * a part of some TCP connection. This makes no difference
1809 * for UDP sockets, but also doens't harm them.
1810 *
1811 * If we're asking for any reserved port (i.e. port == 0 &&
1812 * transport->xprt.resvport == 1) xs_get_srcport above will
1813 * ensure that port is non-zero and we will bind as needed.
1814 */
1815 if (port == 0)
1816 return 0;
1817
beb59b68 1818 memcpy(&myaddr, &transport->srcaddr, transport->xprt.addrlen);
a246b010 1819 do {
beb59b68
PE		 1820 rpc_set_port((struct sockaddr *)&myaddr, port);
1821 err = kernel_bind(sock, (struct sockaddr *)&myaddr,
1822 transport->xprt.addrlen);
a246b010 1823 if (err == 0) {
fbfffbd5 1824 transport->srcport = port;
d3bc9a1d 1825 break;
a246b010 1826 }
67a391d7 1827 last = port;
baaf4e48 1828 port = xs_next_srcport(transport, port);
67a391d7
TM		 1829 if (port > last)
1830 nloop++;
1831 } while (err == -EADDRINUSE && nloop != 2);
90058d37 1832
4232e863 1833 if (myaddr.ss_family == AF_INET)
beb59b68
PE		 1834 dprintk("RPC: %s %pI4:%u: %s (%d)\n", __func__,
1835 &((struct sockaddr_in *)&myaddr)->sin_addr,
1836 port, err ? "failed" : "ok", err);
1837 else
1838 dprintk("RPC: %s %pI6:%u: %s (%d)\n", __func__,
1839 &((struct sockaddr_in6 *)&myaddr)->sin6_addr,
1840 port, err ? "failed" : "ok", err);
a246b010
CL		 1841 return err;
1842}
1843
176e21ee
CL		 1844/*
1845 * We don't support autobind on AF_LOCAL sockets
1846 */
1847static void xs_local_rpcbind(struct rpc_task *task)
1848{
fb43d172 1849 xprt_set_bound(task->tk_xprt);
176e21ee
CL		 1850}
1851
1852static void xs_local_set_port(struct rpc_xprt *xprt, unsigned short port)
1853{
1854}
beb59b68 1855
ed07536e
PZ		 1856#ifdef CONFIG_DEBUG_LOCK_ALLOC
1857static struct lock_class_key xs_key[2];
1858static struct lock_class_key xs_slock_key[2];
1859
176e21ee
CL		 1860static inline void xs_reclassify_socketu(struct socket *sock)
1861{
1862 struct sock *sk = sock->sk;
1863
176e21ee
CL		 1864 sock_lock_init_class_and_name(sk, "slock-AF_LOCAL-RPC",
1865 &xs_slock_key[1], "sk_lock-AF_LOCAL-RPC", &xs_key[1]);
1866}
1867
8945ee5e 1868static inline void xs_reclassify_socket4(struct socket *sock)
ed07536e
PZ		 1869{
1870 struct sock *sk = sock->sk;
8945ee5e 1871
8945ee5e
CL		 1872 sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1873 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1874}
ed07536e 1875
8945ee5e
CL		 1876static inline void xs_reclassify_socket6(struct socket *sock)
1877{
1878 struct sock *sk = sock->sk;
ed07536e 1879
8945ee5e
CL		 1880 sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1881 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
ed07536e 1882}
6bc9638a
PE		 1883
1884static inline void xs_reclassify_socket(int family, struct socket *sock)
1885{
fafc4e1e 1886 if (WARN_ON_ONCE(!sock_allow_reclassification(sock->sk)))
1b7a1819
WAA		 1887 return;
1888
4232e863 1889 switch (family) {
176e21ee
CL		 1890 case AF_LOCAL:
1891 xs_reclassify_socketu(sock);
1892 break;
4232e863 1893 case AF_INET:
6bc9638a 1894 xs_reclassify_socket4(sock);
4232e863
CL		 1895 break;
1896 case AF_INET6:
6bc9638a 1897 xs_reclassify_socket6(sock);
4232e863
CL		 1898 break;
1899 }
6bc9638a 1900}
ed07536e 1901#else
6bc9638a
PE		 1902static inline void xs_reclassify_socket(int family, struct socket *sock)
1903{
1904}
ed07536e
PZ		 1905#endif
1906
93dc41bd
N		 1907static void xs_dummy_setup_socket(struct work_struct *work)
1908{
1909}
1910
6bc9638a 1911static struct socket *xs_create_sock(struct rpc_xprt *xprt,
4dda9c8a
TM		 1912 struct sock_xprt *transport, int family, int type,
1913 int protocol, bool reuseport)
22f79326
PE		 1914{
1915 struct socket *sock;
1916 int err;
1917
6bc9638a 1918 err = __sock_create(xprt->xprt_net, family, type, protocol, &sock, 1);
22f79326
PE		 1919 if (err < 0) {
1920 dprintk("RPC: can't create %d transport socket (%d).\n",
1921 protocol, -err);
1922 goto out;
1923 }
6bc9638a 1924 xs_reclassify_socket(family, sock);
22f79326 1925
4dda9c8a
TM		 1926 if (reuseport)
1927 xs_sock_set_reuseport(sock);
1928
4cea288a
BH		 1929 err = xs_bind(transport, sock);
1930 if (err) {
22f79326
PE		 1931 sock_release(sock);
1932 goto out;
1933 }
1934
1935 return sock;
1936out:
1937 return ERR_PTR(err);
1938}
1939
176e21ee
CL		 1940static int xs_local_finish_connecting(struct rpc_xprt *xprt,
1941 struct socket *sock)
1942{
1943 struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1944 xprt);
1945
1946 if (!transport->inet) {
1947 struct sock *sk = sock->sk;
1948
1949 write_lock_bh(&sk->sk_callback_lock);
1950
1951 xs_save_old_callbacks(transport, sk);
1952
1953 sk->sk_user_data = xprt;
a2648094 1954 sk->sk_data_ready = xs_data_ready;
176e21ee 1955 sk->sk_write_space = xs_udp_write_space;
b4411457 1956 sock_set_flag(sk, SOCK_FASYNC);
2118071d 1957 sk->sk_error_report = xs_error_report;
c2126157 1958 sk->sk_allocation = GFP_NOIO;
176e21ee
CL		 1959
1960 xprt_clear_connected(xprt);
1961
1962 /* Reset to new socket */
1963 transport->sock = sock;
1964 transport->inet = sk;
1965
1966 write_unlock_bh(&sk->sk_callback_lock);
1967 }
1968
1969 /* Tell the socket layer to start connecting... */
1970 xprt->stat.connect_count++;
1971 xprt->stat.connect_start = jiffies;
1972 return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, 0);
1973}
1974
1975/**
1976 * xs_local_setup_socket - create AF_LOCAL socket, connect to a local endpoint
176e21ee 1977 * @transport: socket transport to connect
176e21ee 1978 */
dc107402 1979static int xs_local_setup_socket(struct sock_xprt *transport)
176e21ee 1980{
176e21ee
CL		 1981 struct rpc_xprt *xprt = &transport->xprt;
1982 struct socket *sock;
1983 int status = -EIO;
1984
176e21ee
CL		 1985 status = __sock_create(xprt->xprt_net, AF_LOCAL,
1986 SOCK_STREAM, 0, &sock, 1);
1987 if (status < 0) {
1988 dprintk("RPC: can't create AF_LOCAL "
1989 "transport socket (%d).\n", -status);
1990 goto out;
1991 }
d1358917 1992 xs_reclassify_socket(AF_LOCAL, sock);
176e21ee
CL		 1993
1994 dprintk("RPC: worker connecting xprt %p via AF_LOCAL to %s\n",
1995 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1996
1997 status = xs_local_finish_connecting(xprt, sock);
40b5ea0c 1998 trace_rpc_socket_connect(xprt, sock, status);
176e21ee
CL		 1999 switch (status) {
2000 case 0:
2001 dprintk("RPC: xprt %p connected to %s\n",
2002 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2003 xprt_set_connected(xprt);
3601c4a9 2004 case -ENOBUFS:
176e21ee
CL		 2005 break;
2006 case -ENOENT:
2007 dprintk("RPC: xprt %p: socket %s does not exist\n",
2008 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2009 break;
4a20a988
TM		 2010 case -ECONNREFUSED:
2011 dprintk("RPC: xprt %p: connection refused for %s\n",
2012 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2013 break;
176e21ee
CL		 2014 default:
2015 printk(KERN_ERR "%s: unhandled error (%d) connecting to %s\n",
2016 __func__, -status,
2017 xprt->address_strings[RPC_DISPLAY_ADDR]);
2018 }
2019
2020out:
2021 xprt_clear_connecting(xprt);
2022 xprt_wake_pending_tasks(xprt, status);
dc107402
BF		 2023 return status;
2024}
2025
b6669737 2026static void xs_local_connect(struct rpc_xprt *xprt, struct rpc_task *task)
dc107402 2027{
dc107402
BF		 2028 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2029 int ret;
2030
2031 if (RPC_IS_ASYNC(task)) {
2032 /*
2033 * We want the AF_LOCAL connect to be resolved in the
2034 * filesystem namespace of the process making the rpc
2035 * call. Thus we connect synchronously.
2036 *
2037 * If we want to support asynchronous AF_LOCAL calls,
2038 * we'll need to figure out how to pass a namespace to
2039 * connect.
2040 */
2041 rpc_exit(task, -ENOTCONN);
2042 return;
2043 }
2044 ret = xs_local_setup_socket(transport);
2045 if (ret && !RPC_IS_SOFTCONN(task))
2046 msleep_interruptible(15000);
176e21ee
CL		 2047}
2048
3c87ef6e 2049#if IS_ENABLED(CONFIG_SUNRPC_SWAP)
d6e971d8
JL		 2050/*
2051 * Note that this should be called with XPRT_LOCKED held (or when we otherwise
2052 * know that we have exclusive access to the socket), to guard against
2053 * races with xs_reset_transport.
2054 */
a564b8f0
MG		 2055static void xs_set_memalloc(struct rpc_xprt *xprt)
2056{
2057 struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
2058 xprt);
2059
d6e971d8
JL		 2060 /*
2061 * If there's no sock, then we have nothing to set. The
2062 * reconnecting process will get it for us.
2063 */
2064 if (!transport->inet)
2065 return;
8e228133 2066 if (atomic_read(&xprt->swapper))
a564b8f0
MG		 2067 sk_set_memalloc(transport->inet);
2068}
2069
2070/**
d67fa4d8 2071 * xs_enable_swap - Tag this transport as being used for swap.
a564b8f0 2072 * @xprt: transport to tag
a564b8f0 2073 *
8e228133
JL		 2074 * Take a reference to this transport on behalf of the rpc_clnt, and
2075 * optionally mark it for swapping if it wasn't already.
a564b8f0 2076 */
d67fa4d8
JL		 2077static int
2078xs_enable_swap(struct rpc_xprt *xprt)
a564b8f0 2079{
d6e971d8 2080 struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
a564b8f0 2081
d6e971d8
JL		 2082 if (atomic_inc_return(&xprt->swapper) != 1)
2083 return 0;
2084 if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE))
2085 return -ERESTARTSYS;
2086 if (xs->inet)
2087 sk_set_memalloc(xs->inet);
2088 xprt_release_xprt(xprt, NULL);
8e228133
JL		 2089 return 0;
2090}
a564b8f0 2091
8e228133 2092/**
d67fa4d8 2093 * xs_disable_swap - Untag this transport as being used for swap.
8e228133
JL		 2094 * @xprt: transport to tag
2095 *
2096 * Drop a "swapper" reference to this xprt on behalf of the rpc_clnt. If the
2097 * swapper refcount goes to 0, untag the socket as a memalloc socket.
2098 */
d67fa4d8
JL		 2099static void
2100xs_disable_swap(struct rpc_xprt *xprt)
8e228133 2101{
d6e971d8 2102 struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
8e228133 2103
d6e971d8
JL		 2104 if (!atomic_dec_and_test(&xprt->swapper))
2105 return;
2106 if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE))
2107 return;
2108 if (xs->inet)
2109 sk_clear_memalloc(xs->inet);
2110 xprt_release_xprt(xprt, NULL);
a564b8f0 2111}
a564b8f0
MG		 2112#else
2113static void xs_set_memalloc(struct rpc_xprt *xprt)
2114{
2115}
d67fa4d8
JL		 2116
2117static int
2118xs_enable_swap(struct rpc_xprt *xprt)
2119{
2120 return -EINVAL;
2121}
2122
2123static void
2124xs_disable_swap(struct rpc_xprt *xprt)
2125{
2126}
a564b8f0
MG		 2127#endif
2128
16be2d20
CL		 2129static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2130{
2131 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2132
2133 if (!transport->inet) {
2134 struct sock *sk = sock->sk;
2135
2136 write_lock_bh(&sk->sk_callback_lock);
2137
2a9e1cfa
TM		 2138 xs_save_old_callbacks(transport, sk);
2139
16be2d20 2140 sk->sk_user_data = xprt;
f9b2ee71 2141 sk->sk_data_ready = xs_data_ready;
16be2d20 2142 sk->sk_write_space = xs_udp_write_space;
b4411457 2143 sock_set_flag(sk, SOCK_FASYNC);
c2126157 2144 sk->sk_allocation = GFP_NOIO;
16be2d20
CL		 2145
2146 xprt_set_connected(xprt);
2147
2148 /* Reset to new socket */
2149 transport->sock = sock;
2150 transport->inet = sk;
2151
a564b8f0
MG		 2152 xs_set_memalloc(xprt);
2153
16be2d20
CL		 2154 write_unlock_bh(&sk->sk_callback_lock);
2155 }
2156 xs_udp_do_set_buffer_size(xprt);
2157}
2158
8c14ff2a 2159static void xs_udp_setup_socket(struct work_struct *work)
a246b010 2160{
34161db6
TM		 2161 struct sock_xprt *transport =
2162 container_of(work, struct sock_xprt, connect_worker.work);
c8475461 2163 struct rpc_xprt *xprt = &transport->xprt;
ee0ac0c2 2164 struct socket *sock = transport->sock;
b65c0310 2165 int status = -EIO;
9903cd1c 2166
8c14ff2a 2167 sock = xs_create_sock(xprt, transport,
4dda9c8a
TM		 2168 xs_addr(xprt)->sa_family, SOCK_DGRAM,
2169 IPPROTO_UDP, false);
b65c0310 2170 if (IS_ERR(sock))
b0d93ad5 2171 goto out;
68e220bd 2172
c740eff8
CL		 2173 dprintk("RPC: worker connecting xprt %p via %s to "
2174 "%s (port %s)\n", xprt,
2175 xprt->address_strings[RPC_DISPLAY_PROTO],
2176 xprt->address_strings[RPC_DISPLAY_ADDR],
2177 xprt->address_strings[RPC_DISPLAY_PORT]);
68e220bd
CL		 2178
2179 xs_udp_finish_connecting(xprt, sock);
40b5ea0c 2180 trace_rpc_socket_connect(xprt, sock, 0);
b0d93ad5
CL		 2181 status = 0;
2182out:
718ba5b8 2183 xprt_unlock_connect(xprt, transport);
b0d93ad5 2184 xprt_clear_connecting(xprt);
7d1e8255 2185 xprt_wake_pending_tasks(xprt, status);
a246b010
CL		 2186}
2187
4876cc77
TM		 2188/**
2189 * xs_tcp_shutdown - gracefully shut down a TCP socket
2190 * @xprt: transport
2191 *
2192 * Initiates a graceful shutdown of the TCP socket by calling the
2193 * equivalent of shutdown(SHUT_RDWR);
2194 */
2195static void xs_tcp_shutdown(struct rpc_xprt *xprt)
2196{
2197 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2198 struct socket *sock = transport->sock;
2199
2200 if (sock == NULL)
2201 return;
2202 if (xprt_connected(xprt)) {
2203 kernel_sock_shutdown(sock, SHUT_RDWR);
2204 trace_rpc_socket_shutdown(xprt, sock);
2205 } else
2206 xs_reset_transport(transport);
2207}
2208
16be2d20 2209static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
a246b010 2210{
16be2d20 2211 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
fe19a96b 2212 int ret = -ENOTCONN;
edb267a6 2213
ee0ac0c2 2214 if (!transport->inet) {
b0d93ad5 2215 struct sock *sk = sock->sk;
7f260e85
TM		 2216 unsigned int keepidle = xprt->timeout->to_initval / HZ;
2217 unsigned int keepcnt = xprt->timeout->to_retries + 1;
2218 unsigned int opt_on = 1;
775f06ab 2219 unsigned int timeo;
7f260e85
TM		 2220
2221 /* TCP Keepalive options */
2222 kernel_setsockopt(sock, SOL_SOCKET, SO_KEEPALIVE,
2223 (char *)&opt_on, sizeof(opt_on));
2224 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPIDLE,
2225 (char *)&keepidle, sizeof(keepidle));
2226 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPINTVL,
2227 (char *)&keepidle, sizeof(keepidle));
2228 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPCNT,
2229 (char *)&keepcnt, sizeof(keepcnt));
b0d93ad5 2230
775f06ab
TM		 2231 /* TCP user timeout (see RFC5482) */
2232 timeo = jiffies_to_msecs(xprt->timeout->to_initval) *
2233 (xprt->timeout->to_retries + 1);
2234 kernel_setsockopt(sock, SOL_TCP, TCP_USER_TIMEOUT,
2235 (char *)&timeo, sizeof(timeo));
2236
b0d93ad5
CL		 2237 write_lock_bh(&sk->sk_callback_lock);
2238
2a9e1cfa
TM		 2239 xs_save_old_callbacks(transport, sk);
2240
b0d93ad5 2241 sk->sk_user_data = xprt;
5157b956 2242 sk->sk_data_ready = xs_data_ready;
b0d93ad5
CL		 2243 sk->sk_state_change = xs_tcp_state_change;
2244 sk->sk_write_space = xs_tcp_write_space;
b4411457 2245 sock_set_flag(sk, SOCK_FASYNC);
2118071d 2246 sk->sk_error_report = xs_error_report;
c2126157 2247 sk->sk_allocation = GFP_NOIO;
3167e12c
CL		 2248
2249 /* socket options */
3167e12c 2250 sock_reset_flag(sk, SOCK_LINGER);
3167e12c 2251 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
b0d93ad5
CL		 2252
2253 xprt_clear_connected(xprt);
2254
2255 /* Reset to new socket */
ee0ac0c2
CL		 2256 transport->sock = sock;
2257 transport->inet = sk;
b0d93ad5
CL		 2258
2259 write_unlock_bh(&sk->sk_callback_lock);
2260 }
2261
01d37c42 2262 if (!xprt_bound(xprt))
fe19a96b 2263 goto out;
01d37c42 2264
a564b8f0
MG		 2265 xs_set_memalloc(xprt);
2266
b0d93ad5 2267 /* Tell the socket layer to start connecting... */
262ca07d
CL		 2268 xprt->stat.connect_count++;
2269 xprt->stat.connect_start = jiffies;
0fdea1e8 2270 set_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
fe19a96b
TM		 2271 ret = kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
2272 switch (ret) {
2273 case 0:
4dda9c8a 2274 xs_set_srcport(transport, sock);
fe19a96b
TM		 2275 case -EINPROGRESS:
2276 /* SYN_SENT! */
fe19a96b
TM		 2277 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2278 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2279 }
2280out:
2281 return ret;
16be2d20
CL		 2282}
2283
9903cd1c 2284/**
b61d59ff 2285 * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint
9903cd1c
CL		 2286 *
2287 * Invoked by a work queue tasklet.
a246b010 2288 */
cdd518d5 2289static void xs_tcp_setup_socket(struct work_struct *work)
a246b010 2290{
cdd518d5
PE		 2291 struct sock_xprt *transport =
2292 container_of(work, struct sock_xprt, connect_worker.work);
ee0ac0c2 2293 struct socket *sock = transport->sock;
a9f5f0f7 2294 struct rpc_xprt *xprt = &transport->xprt;
b61d59ff 2295 int status = -EIO;
a246b010 2296
ee0ac0c2 2297 if (!sock) {
cdd518d5 2298 sock = xs_create_sock(xprt, transport,
4dda9c8a
TM		 2299 xs_addr(xprt)->sa_family, SOCK_STREAM,
2300 IPPROTO_TCP, true);
b61d59ff
TM		 2301 if (IS_ERR(sock)) {
2302 status = PTR_ERR(sock);
3167e12c
CL		 2303 goto out;
2304 }
7d1e8255 2305 }
a246b010 2306
c740eff8
CL		 2307 dprintk("RPC: worker connecting xprt %p via %s to "
2308 "%s (port %s)\n", xprt,
2309 xprt->address_strings[RPC_DISPLAY_PROTO],
2310 xprt->address_strings[RPC_DISPLAY_ADDR],
2311 xprt->address_strings[RPC_DISPLAY_PORT]);
edb267a6 2312
16be2d20 2313 status = xs_tcp_finish_connecting(xprt, sock);
40b5ea0c 2314 trace_rpc_socket_connect(xprt, sock, status);
46121cf7
CL		 2315 dprintk("RPC: %p connect status %d connected %d sock state %d\n",
2316 xprt, -status, xprt_connected(xprt),
2317 sock->sk->sk_state);
2a491991 2318 switch (status) {
f75e6745
TM		 2319 default:
2320 printk("%s: connect returned unhandled error %d\n",
2321 __func__, status);
2322 case -EADDRNOTAVAIL:
2323 /* We're probably in TIME_WAIT. Get rid of existing socket,
2324 * and retry
2325 */
a519fc7a 2326 xs_tcp_force_close(xprt);
88b5ed73 2327 break;
2a491991
TM		 2328 case 0:
2329 case -EINPROGRESS:
2330 case -EALREADY:
718ba5b8 2331 xprt_unlock_connect(xprt, transport);
7d1e8255 2332 return;
9fcfe0c8
TM		 2333 case -EINVAL:
2334 /* Happens, for instance, if the user specified a link
2335 * local IPv6 address without a scope-id.
2336 */
3ed5e2a2
TM		 2337 case -ECONNREFUSED:
2338 case -ECONNRESET:
2339 case -ENETUNREACH:
3913c78c 2340 case -EADDRINUSE:
3601c4a9 2341 case -ENOBUFS:
3ed5e2a2 2342 /* retry with existing socket, after a delay */
4efdd92c 2343 xs_tcp_force_close(xprt);
9fcfe0c8 2344 goto out;
a246b010 2345 }
2a491991 2346 status = -EAGAIN;
a246b010 2347out:
718ba5b8 2348 xprt_unlock_connect(xprt, transport);
68e220bd 2349 xprt_clear_connecting(xprt);
7d1e8255 2350 xprt_wake_pending_tasks(xprt, status);
68e220bd 2351}
b0d93ad5 2352
9903cd1c
CL		 2353/**
2354 * xs_connect - connect a socket to a remote endpoint
1b092092 2355 * @xprt: pointer to transport structure
9903cd1c
CL		 2356 * @task: address of RPC task that manages state of connect request
2357 *
2358 * TCP: If the remote end dropped the connection, delay reconnecting.
03bf4b70
CL		 2359 *
2360 * UDP socket connects are synchronous, but we use a work queue anyway
2361 * to guarantee that even unprivileged user processes can set up a
2362 * socket on a privileged port.
2363 *
2364 * If a UDP socket connect fails, the delay behavior here prevents
2365 * retry floods (hard mounts).
9903cd1c 2366 */
1b092092 2367static void xs_connect(struct rpc_xprt *xprt, struct rpc_task *task)
a246b010 2368{
ee0ac0c2 2369 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
a246b010 2370
718ba5b8
TM		 2371 WARN_ON_ONCE(!xprt_lock_connect(xprt, task, transport));
2372
99b1a4c3 2373 if (transport->sock != NULL) {
46121cf7
CL		 2374 dprintk("RPC: xs_connect delayed xprt %p for %lu "
2375 "seconds\n",
03bf4b70 2376 xprt, xprt->reestablish_timeout / HZ);
99b1a4c3
TM		 2377
2378 /* Start by resetting any existing state */
2379 xs_reset_transport(transport);
2380
c1384c9c
TM		 2381 queue_delayed_work(rpciod_workqueue,
2382 &transport->connect_worker,
2383 xprt->reestablish_timeout);
03bf4b70 2384 xprt->reestablish_timeout <<= 1;
61d0a8e6
NB		 2385 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2386 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
03bf4b70
CL		 2387 if (xprt->reestablish_timeout > XS_TCP_MAX_REEST_TO)
2388 xprt->reestablish_timeout = XS_TCP_MAX_REEST_TO;
b0d93ad5 2389 } else {
46121cf7 2390 dprintk("RPC: xs_connect scheduled xprt %p\n", xprt);
c1384c9c
TM		 2391 queue_delayed_work(rpciod_workqueue,
2392 &transport->connect_worker, 0);
a246b010
CL		 2393 }
2394}
2395
176e21ee
CL		 2396/**
2397 * xs_local_print_stats - display AF_LOCAL socket-specifc stats
2398 * @xprt: rpc_xprt struct containing statistics
2399 * @seq: output file
2400 *
2401 */
2402static void xs_local_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2403{
2404 long idle_time = 0;
2405
2406 if (xprt_connected(xprt))
2407 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2408
2409 seq_printf(seq, "\txprt:\tlocal %lu %lu %lu %ld %lu %lu %lu "
15a45206 2410 "%llu %llu %lu %llu %llu\n",
176e21ee
CL		 2411 xprt->stat.bind_count,
2412 xprt->stat.connect_count,
2413 xprt->stat.connect_time,
2414 idle_time,
2415 xprt->stat.sends,
2416 xprt->stat.recvs,
2417 xprt->stat.bad_xids,
2418 xprt->stat.req_u,
15a45206
AA		 2419 xprt->stat.bklog_u,
2420 xprt->stat.max_slots,
2421 xprt->stat.sending_u,
2422 xprt->stat.pending_u);
176e21ee
CL		 2423}
2424
262ca07d
CL		 2425/**
2426 * xs_udp_print_stats - display UDP socket-specifc stats
2427 * @xprt: rpc_xprt struct containing statistics
2428 * @seq: output file
2429 *
2430 */
2431static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2432{
c8475461
CL		 2433 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2434
15a45206
AA		 2435 seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %llu %llu "
2436 "%lu %llu %llu\n",
fbfffbd5 2437 transport->srcport,
262ca07d
CL		 2438 xprt->stat.bind_count,
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);
262ca07d
CL		 2447}
2448
2449/**
2450 * xs_tcp_print_stats - display TCP socket-specifc stats
2451 * @xprt: rpc_xprt struct containing statistics
2452 * @seq: output file
2453 *
2454 */
2455static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2456{
c8475461 2457 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
262ca07d
CL		 2458 long idle_time = 0;
2459
2460 if (xprt_connected(xprt))
2461 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2462
15a45206
AA		 2463 seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu "
2464 "%llu %llu %lu %llu %llu\n",
fbfffbd5 2465 transport->srcport,
262ca07d
CL		 2466 xprt->stat.bind_count,
2467 xprt->stat.connect_count,
2468 xprt->stat.connect_time,
2469 idle_time,
2470 xprt->stat.sends,
2471 xprt->stat.recvs,
2472 xprt->stat.bad_xids,
2473 xprt->stat.req_u,
15a45206
AA		 2474 xprt->stat.bklog_u,
2475 xprt->stat.max_slots,
2476 xprt->stat.sending_u,
2477 xprt->stat.pending_u);
262ca07d
CL		 2478}
2479
4cfc7e60
RI		 2480/*
2481 * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason
2482 * we allocate pages instead doing a kmalloc like rpc_malloc is because we want
2483 * to use the server side send routines.
2484 */
5a51f13a 2485static void *bc_malloc(struct rpc_task *task, size_t size)
4cfc7e60
RI		 2486{
2487 struct page *page;
2488 struct rpc_buffer *buf;
2489
b8a13d03
WAA		 2490 WARN_ON_ONCE(size > PAGE_SIZE - sizeof(struct rpc_buffer));
2491 if (size > PAGE_SIZE - sizeof(struct rpc_buffer))
2492 return NULL;
4cfc7e60 2493
b8a13d03 2494 page = alloc_page(GFP_KERNEL);
4cfc7e60
RI		 2495 if (!page)
2496 return NULL;
2497
2498 buf = page_address(page);
2499 buf->len = PAGE_SIZE;
2500
2501 return buf->data;
2502}
2503
2504/*
2505 * Free the space allocated in the bc_alloc routine
2506 */
5a51f13a 2507static void bc_free(void *buffer)
4cfc7e60
RI		 2508{
2509 struct rpc_buffer *buf;
2510
2511 if (!buffer)
2512 return;
2513
2514 buf = container_of(buffer, struct rpc_buffer, data);
2515 free_page((unsigned long)buf);
2516}
2517
2518/*
2519 * Use the svc_sock to send the callback. Must be called with svsk->sk_mutex
2520 * held. Borrows heavily from svc_tcp_sendto and xs_tcp_send_request.
2521 */
2522static int bc_sendto(struct rpc_rqst *req)
2523{
2524 int len;
2525 struct xdr_buf *xbufp = &req->rq_snd_buf;
2526 struct rpc_xprt *xprt = req->rq_xprt;
2527 struct sock_xprt *transport =
2528 container_of(xprt, struct sock_xprt, xprt);
2529 struct socket *sock = transport->sock;
2530 unsigned long headoff;
2531 unsigned long tailoff;
2532
61677eee 2533 xs_encode_stream_record_marker(xbufp);
4cfc7e60
RI		 2534
2535 tailoff = (unsigned long)xbufp->tail[0].iov_base & ~PAGE_MASK;
2536 headoff = (unsigned long)xbufp->head[0].iov_base & ~PAGE_MASK;
2537 len = svc_send_common(sock, xbufp,
2538 virt_to_page(xbufp->head[0].iov_base), headoff,
2539 xbufp->tail[0].iov_base, tailoff);
2540
2541 if (len != xbufp->len) {
2542 printk(KERN_NOTICE "Error sending entire callback!\n");
2543 len = -EAGAIN;
2544 }
2545
2546 return len;
2547}
2548
2549/*
2550 * The send routine. Borrows from svc_send
2551 */
2552static int bc_send_request(struct rpc_task *task)
2553{
2554 struct rpc_rqst *req = task->tk_rqstp;
2555 struct svc_xprt *xprt;
7fc56136 2556 int len;
4cfc7e60
RI		 2557
2558 dprintk("sending request with xid: %08x\n", ntohl(req->rq_xid));
2559 /*
2560 * Get the server socket associated with this callback xprt
2561 */
2562 xprt = req->rq_xprt->bc_xprt;
4cfc7e60
RI		 2563
2564 /*
2565 * Grab the mutex to serialize data as the connection is shared
2566 * with the fore channel
2567 */
2568 if (!mutex_trylock(&xprt->xpt_mutex)) {
2569 rpc_sleep_on(&xprt->xpt_bc_pending, task, NULL);
2570 if (!mutex_trylock(&xprt->xpt_mutex))
2571 return -EAGAIN;
2572 rpc_wake_up_queued_task(&xprt->xpt_bc_pending, task);
2573 }
2574 if (test_bit(XPT_DEAD, &xprt->xpt_flags))
2575 len = -ENOTCONN;
2576 else
2577 len = bc_sendto(req);
2578 mutex_unlock(&xprt->xpt_mutex);
2579
2580 if (len > 0)
2581 len = 0;
2582
2583 return len;
2584}
2585
2586/*
2587 * The close routine. Since this is client initiated, we do nothing
2588 */
2589
2590static void bc_close(struct rpc_xprt *xprt)
2591{
4cfc7e60
RI		 2592}
2593
2594/*
2595 * The xprt destroy routine. Again, because this connection is client
2596 * initiated, we do nothing
2597 */
2598
2599static void bc_destroy(struct rpc_xprt *xprt)
2600{
47f72efa
KM		 2601 dprintk("RPC: bc_destroy xprt %p\n", xprt);
2602
2603 xs_xprt_free(xprt);
2604 module_put(THIS_MODULE);
4cfc7e60
RI		 2605}
2606
176e21ee
CL		 2607static struct rpc_xprt_ops xs_local_ops = {
2608 .reserve_xprt = xprt_reserve_xprt,
2609 .release_xprt = xs_tcp_release_xprt,
f39c1bfb 2610 .alloc_slot = xprt_alloc_slot,
176e21ee
CL		 2611 .rpcbind = xs_local_rpcbind,
2612 .set_port = xs_local_set_port,
dc107402 2613 .connect = xs_local_connect,
176e21ee
CL		 2614 .buf_alloc = rpc_malloc,
2615 .buf_free = rpc_free,
2616 .send_request = xs_local_send_request,
2617 .set_retrans_timeout = xprt_set_retrans_timeout_def,
2618 .close = xs_close,
a1311d87 2619 .destroy = xs_destroy,
176e21ee 2620 .print_stats = xs_local_print_stats,
d67fa4d8
JL		 2621 .enable_swap = xs_enable_swap,
2622 .disable_swap = xs_disable_swap,
176e21ee
CL		 2623};
2624
262965f5 2625static struct rpc_xprt_ops xs_udp_ops = {
43118c29 2626 .set_buffer_size = xs_udp_set_buffer_size,
12a80469 2627 .reserve_xprt = xprt_reserve_xprt_cong,
49e9a890 2628 .release_xprt = xprt_release_xprt_cong,
f39c1bfb 2629 .alloc_slot = xprt_alloc_slot,
45160d62 2630 .rpcbind = rpcb_getport_async,
92200412 2631 .set_port = xs_set_port,
262965f5 2632 .connect = xs_connect,
02107148
CL		 2633 .buf_alloc = rpc_malloc,
2634 .buf_free = rpc_free,
262965f5 2635 .send_request = xs_udp_send_request,
fe3aca29 2636 .set_retrans_timeout = xprt_set_retrans_timeout_rtt,
46c0ee8b 2637 .timer = xs_udp_timer,
a58dd398 2638 .release_request = xprt_release_rqst_cong,
262965f5
CL		 2639 .close = xs_close,
2640 .destroy = xs_destroy,
262ca07d 2641 .print_stats = xs_udp_print_stats,
d67fa4d8
JL		 2642 .enable_swap = xs_enable_swap,
2643 .disable_swap = xs_disable_swap,
4a068258 2644 .inject_disconnect = xs_inject_disconnect,
262965f5
CL		 2645};
2646
2647static struct rpc_xprt_ops xs_tcp_ops = {
12a80469 2648 .reserve_xprt = xprt_reserve_xprt,
e0ab53de 2649 .release_xprt = xs_tcp_release_xprt,
f39c1bfb 2650 .alloc_slot = xprt_lock_and_alloc_slot,
45160d62 2651 .rpcbind = rpcb_getport_async,
92200412 2652 .set_port = xs_set_port,
0b9e7943 2653 .connect = xs_connect,
02107148
CL		 2654 .buf_alloc = rpc_malloc,
2655 .buf_free = rpc_free,
262965f5 2656 .send_request = xs_tcp_send_request,
fe3aca29 2657 .set_retrans_timeout = xprt_set_retrans_timeout_def,
c627d31b 2658 .close = xs_tcp_shutdown,
9903cd1c 2659 .destroy = xs_destroy,
262ca07d 2660 .print_stats = xs_tcp_print_stats,
d67fa4d8
JL		 2661 .enable_swap = xs_enable_swap,
2662 .disable_swap = xs_disable_swap,
4a068258 2663 .inject_disconnect = xs_inject_disconnect,
42e5c3e2
CL		 2664#ifdef CONFIG_SUNRPC_BACKCHANNEL
2665 .bc_setup = xprt_setup_bc,
76566773 2666 .bc_up = xs_tcp_bc_up,
6b26cc8c 2667 .bc_maxpayload = xs_tcp_bc_maxpayload,
42e5c3e2
CL		 2668 .bc_free_rqst = xprt_free_bc_rqst,
2669 .bc_destroy = xprt_destroy_bc,
2670#endif
a246b010
CL		 2671};
2672
4cfc7e60
RI		 2673/*
2674 * The rpc_xprt_ops for the server backchannel
2675 */
2676
2677static struct rpc_xprt_ops bc_tcp_ops = {
2678 .reserve_xprt = xprt_reserve_xprt,
2679 .release_xprt = xprt_release_xprt,
84e28a30 2680 .alloc_slot = xprt_alloc_slot,
4cfc7e60
RI		 2681 .buf_alloc = bc_malloc,
2682 .buf_free = bc_free,
2683 .send_request = bc_send_request,
2684 .set_retrans_timeout = xprt_set_retrans_timeout_def,
2685 .close = bc_close,
2686 .destroy = bc_destroy,
2687 .print_stats = xs_tcp_print_stats,
d67fa4d8
JL		 2688 .enable_swap = xs_enable_swap,
2689 .disable_swap = xs_disable_swap,
4a068258 2690 .inject_disconnect = xs_inject_disconnect,
4cfc7e60
RI		 2691};
2692
92476850
CL		 2693static int xs_init_anyaddr(const int family, struct sockaddr *sap)
2694{
2695 static const struct sockaddr_in sin = {
2696 .sin_family = AF_INET,
2697 .sin_addr.s_addr = htonl(INADDR_ANY),
2698 };
2699 static const struct sockaddr_in6 sin6 = {
2700 .sin6_family = AF_INET6,
2701 .sin6_addr = IN6ADDR_ANY_INIT,
2702 };
2703
2704 switch (family) {
176e21ee
CL		 2705 case AF_LOCAL:
2706 break;
92476850
CL		 2707 case AF_INET:
2708 memcpy(sap, &sin, sizeof(sin));
2709 break;
2710 case AF_INET6:
2711 memcpy(sap, &sin6, sizeof(sin6));
2712 break;
2713 default:
2714 dprintk("RPC: %s: Bad address family\n", __func__);
2715 return -EAFNOSUPPORT;
2716 }
2717 return 0;
2718}
2719
3c341b0b 2720static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
d9ba131d
TM		 2721 unsigned int slot_table_size,
2722 unsigned int max_slot_table_size)
c8541ecd
CL		 2723{
2724 struct rpc_xprt *xprt;
ffc2e518 2725 struct sock_xprt *new;
c8541ecd 2726
96802a09 2727 if (args->addrlen > sizeof(xprt->addr)) {
46121cf7 2728 dprintk("RPC: xs_setup_xprt: address too large\n");
c8541ecd
CL		 2729 return ERR_PTR(-EBADF);
2730 }
2731
d9ba131d
TM		 2732 xprt = xprt_alloc(args->net, sizeof(*new), slot_table_size,
2733 max_slot_table_size);
bd1722d4 2734 if (xprt == NULL) {
46121cf7
CL		 2735 dprintk("RPC: xs_setup_xprt: couldn't allocate "
2736 "rpc_xprt\n");
c8541ecd
CL		 2737 return ERR_PTR(-ENOMEM);
2738 }
c8541ecd 2739
bd1722d4 2740 new = container_of(xprt, struct sock_xprt, xprt);
edc1b01c 2741 mutex_init(&new->recv_mutex);
96802a09
FM		 2742 memcpy(&xprt->addr, args->dstaddr, args->addrlen);
2743 xprt->addrlen = args->addrlen;
d3bc9a1d 2744 if (args->srcaddr)
fbfffbd5 2745 memcpy(&new->srcaddr, args->srcaddr, args->addrlen);
92476850
CL		 2746 else {
2747 int err;
2748 err = xs_init_anyaddr(args->dstaddr->sa_family,
2749 (struct sockaddr *)&new->srcaddr);
2aa13531
SK		 2750 if (err != 0) {
2751 xprt_free(xprt);
92476850 2752 return ERR_PTR(err);
2aa13531 2753 }
92476850 2754 }
c8541ecd
CL		 2755
2756 return xprt;
2757}
2758
176e21ee
CL		 2759static const struct rpc_timeout xs_local_default_timeout = {
2760 .to_initval = 10 * HZ,
2761 .to_maxval = 10 * HZ,
2762 .to_retries = 2,
2763};
2764
2765/**
2766 * xs_setup_local - Set up transport to use an AF_LOCAL socket
2767 * @args: rpc transport creation arguments
2768 *
2769 * AF_LOCAL is a "tpi_cots_ord" transport, just like TCP
2770 */
2771static struct rpc_xprt *xs_setup_local(struct xprt_create *args)
2772{
2773 struct sockaddr_un *sun = (struct sockaddr_un *)args->dstaddr;
2774 struct sock_xprt *transport;
2775 struct rpc_xprt *xprt;
2776 struct rpc_xprt *ret;
2777
d9ba131d
TM		 2778 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2779 xprt_max_tcp_slot_table_entries);
176e21ee
CL		 2780 if (IS_ERR(xprt))
2781 return xprt;
2782 transport = container_of(xprt, struct sock_xprt, xprt);
2783
2784 xprt->prot = 0;
2785 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2786 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2787
2788 xprt->bind_timeout = XS_BIND_TO;
2789 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2790 xprt->idle_timeout = XS_IDLE_DISC_TO;
2791
2792 xprt->ops = &xs_local_ops;
2793 xprt->timeout = &xs_local_default_timeout;
2794
a2648094 2795 INIT_WORK(&transport->recv_worker, xs_local_data_receive_workfn);
93dc41bd
N		 2796 INIT_DELAYED_WORK(&transport->connect_worker,
2797 xs_dummy_setup_socket);
2798
176e21ee
CL		 2799 switch (sun->sun_family) {
2800 case AF_LOCAL:
2801 if (sun->sun_path[0] != '/') {
2802 dprintk("RPC: bad AF_LOCAL address: %s\n",
2803 sun->sun_path);
2804 ret = ERR_PTR(-EINVAL);
2805 goto out_err;
2806 }
2807 xprt_set_bound(xprt);
176e21ee 2808 xs_format_peer_addresses(xprt, "local", RPCBIND_NETID_LOCAL);
7073ea87
BF		 2809 ret = ERR_PTR(xs_local_setup_socket(transport));
2810 if (ret)
2811 goto out_err;
176e21ee
CL		 2812 break;
2813 default:
2814 ret = ERR_PTR(-EAFNOSUPPORT);
2815 goto out_err;
2816 }
2817
2818 dprintk("RPC: set up xprt to %s via AF_LOCAL\n",
2819 xprt->address_strings[RPC_DISPLAY_ADDR]);
2820
2821 if (try_module_get(THIS_MODULE))
2822 return xprt;
2823 ret = ERR_PTR(-EINVAL);
2824out_err:
315f3812 2825 xs_xprt_free(xprt);
176e21ee
CL		 2826 return ret;
2827}
2828
2881ae74
TM		 2829static const struct rpc_timeout xs_udp_default_timeout = {
2830 .to_initval = 5 * HZ,
2831 .to_maxval = 30 * HZ,
2832 .to_increment = 5 * HZ,
2833 .to_retries = 5,
2834};
2835
9903cd1c
CL		 2836/**
2837 * xs_setup_udp - Set up transport to use a UDP socket
96802a09 2838 * @args: rpc transport creation arguments
9903cd1c
CL		 2839 *
2840 */
483066d6 2841static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
a246b010 2842{
8f9d5b1a 2843 struct sockaddr *addr = args->dstaddr;
c8541ecd 2844 struct rpc_xprt *xprt;
c8475461 2845 struct sock_xprt *transport;
0a68b0be 2846 struct rpc_xprt *ret;
a246b010 2847
d9ba131d
TM		 2848 xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries,
2849 xprt_udp_slot_table_entries);
c8541ecd
CL		 2850 if (IS_ERR(xprt))
2851 return xprt;
c8475461 2852 transport = container_of(xprt, struct sock_xprt, xprt);
a246b010 2853
ec739ef0 2854 xprt->prot = IPPROTO_UDP;
808012fb 2855 xprt->tsh_size = 0;
a246b010
CL		 2856 /* XXX: header size can vary due to auth type, IPv6, etc. */
2857 xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
2858
03bf4b70 2859 xprt->bind_timeout = XS_BIND_TO;
03bf4b70
CL		 2860 xprt->reestablish_timeout = XS_UDP_REEST_TO;
2861 xprt->idle_timeout = XS_IDLE_DISC_TO;
a246b010 2862
262965f5 2863 xprt->ops = &xs_udp_ops;
a246b010 2864
ba7392bb 2865 xprt->timeout = &xs_udp_default_timeout;
a246b010 2866
f9b2ee71 2867 INIT_WORK(&transport->recv_worker, xs_udp_data_receive_workfn);
edc1b01c
TM		 2868 INIT_DELAYED_WORK(&transport->connect_worker, xs_udp_setup_socket);
2869
8f9d5b1a
CL		 2870 switch (addr->sa_family) {
2871 case AF_INET:
2872 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2873 xprt_set_bound(xprt);
2874
9dc3b095 2875 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
8f9d5b1a
CL		 2876 break;
2877 case AF_INET6:
2878 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2879 xprt_set_bound(xprt);
2880
9dc3b095 2881 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
8f9d5b1a
CL		 2882 break;
2883 default:
0a68b0be
BF		 2884 ret = ERR_PTR(-EAFNOSUPPORT);
2885 goto out_err;
8f9d5b1a
CL		 2886 }
2887
c740eff8
CL		 2888 if (xprt_bound(xprt))
2889 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2890 xprt->address_strings[RPC_DISPLAY_ADDR],
2891 xprt->address_strings[RPC_DISPLAY_PORT],
2892 xprt->address_strings[RPC_DISPLAY_PROTO]);
2893 else
2894 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
2895 xprt->address_strings[RPC_DISPLAY_ADDR],
2896 xprt->address_strings[RPC_DISPLAY_PROTO]);
edb267a6 2897
bc25571e
TT		 2898 if (try_module_get(THIS_MODULE))
2899 return xprt;
0a68b0be
BF		 2900 ret = ERR_PTR(-EINVAL);
2901out_err:
315f3812 2902 xs_xprt_free(xprt);
0a68b0be 2903 return ret;
a246b010
CL		 2904}
2905
2881ae74
TM		 2906static const struct rpc_timeout xs_tcp_default_timeout = {
2907 .to_initval = 60 * HZ,
2908 .to_maxval = 60 * HZ,
2909 .to_retries = 2,
2910};
2911
9903cd1c
CL		 2912/**
2913 * xs_setup_tcp - Set up transport to use a TCP socket
96802a09 2914 * @args: rpc transport creation arguments
9903cd1c
CL		 2915 *
2916 */
483066d6 2917static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
a246b010 2918{
8f9d5b1a 2919 struct sockaddr *addr = args->dstaddr;
c8541ecd 2920 struct rpc_xprt *xprt;
c8475461 2921 struct sock_xprt *transport;
0a68b0be 2922 struct rpc_xprt *ret;
b7993ceb
TM		 2923 unsigned int max_slot_table_size = xprt_max_tcp_slot_table_entries;
2924
2925 if (args->flags & XPRT_CREATE_INFINITE_SLOTS)
2926 max_slot_table_size = RPC_MAX_SLOT_TABLE_LIMIT;
a246b010 2927
d9ba131d 2928 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
b7993ceb 2929 max_slot_table_size);
c8541ecd
CL		 2930 if (IS_ERR(xprt))
2931 return xprt;
c8475461 2932 transport = container_of(xprt, struct sock_xprt, xprt);
a246b010 2933
ec739ef0 2934 xprt->prot = IPPROTO_TCP;
808012fb 2935 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
808012fb 2936 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
a246b010 2937
03bf4b70 2938 xprt->bind_timeout = XS_BIND_TO;
03bf4b70
CL		 2939 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2940 xprt->idle_timeout = XS_IDLE_DISC_TO;
a246b010 2941
262965f5 2942 xprt->ops = &xs_tcp_ops;
ba7392bb 2943 xprt->timeout = &xs_tcp_default_timeout;
a246b010 2944
edc1b01c
TM		 2945 INIT_WORK(&transport->recv_worker, xs_tcp_data_receive_workfn);
2946 INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_setup_socket);
2947
8f9d5b1a
CL		 2948 switch (addr->sa_family) {
2949 case AF_INET:
2950 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2951 xprt_set_bound(xprt);
2952
9dc3b095 2953 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
8f9d5b1a
CL		 2954 break;
2955 case AF_INET6:
2956 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2957 xprt_set_bound(xprt);
2958
9dc3b095 2959 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
8f9d5b1a
CL		 2960 break;
2961 default:
0a68b0be
BF		 2962 ret = ERR_PTR(-EAFNOSUPPORT);
2963 goto out_err;
8f9d5b1a
CL		 2964 }
2965
c740eff8
CL		 2966 if (xprt_bound(xprt))
2967 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2968 xprt->address_strings[RPC_DISPLAY_ADDR],
2969 xprt->address_strings[RPC_DISPLAY_PORT],
2970 xprt->address_strings[RPC_DISPLAY_PROTO]);
2971 else
2972 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
2973 xprt->address_strings[RPC_DISPLAY_ADDR],
2974 xprt->address_strings[RPC_DISPLAY_PROTO]);
2975
bc25571e
TT		 2976 if (try_module_get(THIS_MODULE))
2977 return xprt;
0a68b0be
BF		 2978 ret = ERR_PTR(-EINVAL);
2979out_err:
315f3812 2980 xs_xprt_free(xprt);
0a68b0be 2981 return ret;
a246b010 2982}
282b32e1 2983
f300baba
AB		 2984/**
2985 * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket
2986 * @args: rpc transport creation arguments
2987 *
2988 */
2989static struct rpc_xprt *xs_setup_bc_tcp(struct xprt_create *args)
2990{
2991 struct sockaddr *addr = args->dstaddr;
2992 struct rpc_xprt *xprt;
2993 struct sock_xprt *transport;
2994 struct svc_sock *bc_sock;
0a68b0be 2995 struct rpc_xprt *ret;
f300baba 2996
d9ba131d
TM		 2997 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2998 xprt_tcp_slot_table_entries);
f300baba
AB		 2999 if (IS_ERR(xprt))
3000 return xprt;
3001 transport = container_of(xprt, struct sock_xprt, xprt);
3002
3003 xprt->prot = IPPROTO_TCP;
3004 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
3005 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
3006 xprt->timeout = &xs_tcp_default_timeout;
3007
3008 /* backchannel */
3009 xprt_set_bound(xprt);
3010 xprt->bind_timeout = 0;
f300baba
AB		 3011 xprt->reestablish_timeout = 0;
3012 xprt->idle_timeout = 0;
3013
f300baba
AB		 3014 xprt->ops = &bc_tcp_ops;
3015
3016 switch (addr->sa_family) {
3017 case AF_INET:
3018 xs_format_peer_addresses(xprt, "tcp",
3019 RPCBIND_NETID_TCP);
3020 break;
3021 case AF_INET6:
3022 xs_format_peer_addresses(xprt, "tcp",
3023 RPCBIND_NETID_TCP6);
3024 break;
3025 default:
0a68b0be
BF		 3026 ret = ERR_PTR(-EAFNOSUPPORT);
3027 goto out_err;
f300baba
AB		 3028 }
3029
50fa0d40
PE		 3030 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
3031 xprt->address_strings[RPC_DISPLAY_ADDR],
3032 xprt->address_strings[RPC_DISPLAY_PORT],
3033 xprt->address_strings[RPC_DISPLAY_PROTO]);
f300baba 3034
99de8ea9
BF		 3035 /*
3036 * Once we've associated a backchannel xprt with a connection,
28303ca3
WM		 3037 * we want to keep it around as long as the connection lasts,
3038 * in case we need to start using it for a backchannel again;
3039 * this reference won't be dropped until bc_xprt is destroyed.
99de8ea9
BF		 3040 */
3041 xprt_get(xprt);
3042 args->bc_xprt->xpt_bc_xprt = xprt;
3043 xprt->bc_xprt = args->bc_xprt;
3044 bc_sock = container_of(args->bc_xprt, struct svc_sock, sk_xprt);
3045 transport->sock = bc_sock->sk_sock;
3046 transport->inet = bc_sock->sk_sk;
3047
f300baba
AB		 3048 /*
3049 * Since we don't want connections for the backchannel, we set
3050 * the xprt status to connected
3051 */
3052 xprt_set_connected(xprt);
3053
f300baba
AB		 3054 if (try_module_get(THIS_MODULE))
3055 return xprt;
642aab58
KM		 3056
3057 args->bc_xprt->xpt_bc_xprt = NULL;
99de8ea9 3058 xprt_put(xprt);
0a68b0be
BF		 3059 ret = ERR_PTR(-EINVAL);
3060out_err:
315f3812 3061 xs_xprt_free(xprt);
0a68b0be 3062 return ret;
f300baba
AB		 3063}
3064
176e21ee
CL		 3065static struct xprt_class xs_local_transport = {
3066 .list = LIST_HEAD_INIT(xs_local_transport.list),
3067 .name = "named UNIX socket",
3068 .owner = THIS_MODULE,
3069 .ident = XPRT_TRANSPORT_LOCAL,
3070 .setup = xs_setup_local,
3071};
3072
bc25571e
TT		 3073static struct xprt_class xs_udp_transport = {
3074 .list = LIST_HEAD_INIT(xs_udp_transport.list),
3075 .name = "udp",
3076 .owner = THIS_MODULE,
f300baba 3077 .ident = XPRT_TRANSPORT_UDP,
bc25571e
TT		 3078 .setup = xs_setup_udp,
3079};
3080
3081static struct xprt_class xs_tcp_transport = {
3082 .list = LIST_HEAD_INIT(xs_tcp_transport.list),
3083 .name = "tcp",
3084 .owner = THIS_MODULE,
f300baba 3085 .ident = XPRT_TRANSPORT_TCP,
bc25571e
TT		 3086 .setup = xs_setup_tcp,
3087};
3088
f300baba
AB		 3089static struct xprt_class xs_bc_tcp_transport = {
3090 .list = LIST_HEAD_INIT(xs_bc_tcp_transport.list),
3091 .name = "tcp NFSv4.1 backchannel",
3092 .owner = THIS_MODULE,
3093 .ident = XPRT_TRANSPORT_BC_TCP,
3094 .setup = xs_setup_bc_tcp,
3095};
3096
282b32e1 3097/**
bc25571e 3098 * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
282b32e1
CL		 3099 *
3100 */
3101int init_socket_xprt(void)
3102{
f895b252 3103#if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
2b1bec5f 3104 if (!sunrpc_table_header)
0b4d4147 3105 sunrpc_table_header = register_sysctl_table(sunrpc_table);
fbf76683
CL		 3106#endif
3107
176e21ee 3108 xprt_register_transport(&xs_local_transport);
bc25571e
TT		 3109 xprt_register_transport(&xs_udp_transport);
3110 xprt_register_transport(&xs_tcp_transport);
f300baba 3111 xprt_register_transport(&xs_bc_tcp_transport);
bc25571e 3112
282b32e1
CL		 3113 return 0;
3114}
3115
3116/**
bc25571e 3117 * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
282b32e1
CL		 3118 *
3119 */
3120void cleanup_socket_xprt(void)
3121{
f895b252 3122#if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
fbf76683
CL		 3123 if (sunrpc_table_header) {
3124 unregister_sysctl_table(sunrpc_table_header);
3125 sunrpc_table_header = NULL;
3126 }
3127#endif
bc25571e 3128
176e21ee 3129 xprt_unregister_transport(&xs_local_transport);
bc25571e
TT		 3130 xprt_unregister_transport(&xs_udp_transport);
3131 xprt_unregister_transport(&xs_tcp_transport);
f300baba 3132 xprt_unregister_transport(&xs_bc_tcp_transport);
282b32e1 3133}
cbf11071 3134
9bbb9e5a
RR		 3135static int param_set_uint_minmax(const char *val,
3136 const struct kernel_param *kp,
cbf11071
TM		 3137 unsigned int min, unsigned int max)
3138{
00cfaa94 3139 unsigned int num;
cbf11071
TM		 3140 int ret;
3141
3142 if (!val)
3143 return -EINVAL;
00cfaa94 3144 ret = kstrtouint(val, 0, &num);
cbf11071
TM		 3145 if (ret == -EINVAL || num < min || num > max)
3146 return -EINVAL;
3147 *((unsigned int *)kp->arg) = num;
3148 return 0;
3149}
3150
9bbb9e5a 3151static int param_set_portnr(const char *val, const struct kernel_param *kp)
cbf11071
TM		 3152{
3153 return param_set_uint_minmax(val, kp,
3154 RPC_MIN_RESVPORT,
3155 RPC_MAX_RESVPORT);
3156}
3157
9c27847d 3158static const struct kernel_param_ops param_ops_portnr = {
9bbb9e5a
RR		 3159 .set = param_set_portnr,
3160 .get = param_get_uint,
3161};
3162
cbf11071
TM		 3163#define param_check_portnr(name, p) \
3164 __param_check(name, p, unsigned int);
3165
3166module_param_named(min_resvport, xprt_min_resvport, portnr, 0644);
3167module_param_named(max_resvport, xprt_max_resvport, portnr, 0644);
3168
9bbb9e5a
RR		 3169static int param_set_slot_table_size(const char *val,
3170 const struct kernel_param *kp)
cbf11071
TM		 3171{
3172 return param_set_uint_minmax(val, kp,
3173 RPC_MIN_SLOT_TABLE,
3174 RPC_MAX_SLOT_TABLE);
3175}
3176
9c27847d 3177static const struct kernel_param_ops param_ops_slot_table_size = {
9bbb9e5a
RR		 3178 .set = param_set_slot_table_size,
3179 .get = param_get_uint,
3180};
3181
cbf11071
TM		 3182#define param_check_slot_table_size(name, p) \
3183 __param_check(name, p, unsigned int);
3184
d9ba131d
TM		 3185static int param_set_max_slot_table_size(const char *val,
3186 const struct kernel_param *kp)
3187{
3188 return param_set_uint_minmax(val, kp,
3189 RPC_MIN_SLOT_TABLE,
3190 RPC_MAX_SLOT_TABLE_LIMIT);
3191}
3192
9c27847d 3193static const struct kernel_param_ops param_ops_max_slot_table_size = {
d9ba131d
TM		 3194 .set = param_set_max_slot_table_size,
3195 .get = param_get_uint,
3196};
3197
3198#define param_check_max_slot_table_size(name, p) \
3199 __param_check(name, p, unsigned int);
3200
cbf11071
TM		 3201module_param_named(tcp_slot_table_entries, xprt_tcp_slot_table_entries,
3202 slot_table_size, 0644);
d9ba131d
TM		 3203module_param_named(tcp_max_slot_table_entries, xprt_max_tcp_slot_table_entries,
3204 max_slot_table_size, 0644);
cbf11071
TM		 3205module_param_named(udp_slot_table_entries, xprt_udp_slot_table_entries,
3206 slot_table_size, 0644);
3207
ubuntu-bionic-kernel mirror