blob: 2f630a512ab753f7e2df18a42a2abb02f04893ba [file] [log] [blame]
Chuck Levera246b012005-08-11 16:25:23 -04001/*
2 * linux/net/sunrpc/xprtsock.c
3 *
4 * Client-side transport implementation for sockets.
5 *
6 * TCP callback races fixes (C) 1998 Red Hat Software <alan@redhat.com>
7 * TCP send fixes (C) 1998 Red Hat Software <alan@redhat.com>
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>
Chuck Lever55aa4f52005-08-11 16:25:47 -040014 *
15 * IP socket transport implementation, (C) 2005 Chuck Lever <cel@netapp.com>
Chuck Lever8f9d5b12007-08-06 11:57:53 -040016 *
17 * IPv6 support contributed by Gilles Quillard, Bull Open Source, 2005.
18 * <gilles.quillard@bull.net>
Chuck Levera246b012005-08-11 16:25:23 -040019 */
20
21#include <linux/types.h>
22#include <linux/slab.h>
\"Talpey, Thomas\bc255712007-09-10 13:46:39 -040023#include <linux/module.h>
Chuck Levera246b012005-08-11 16:25:23 -040024#include <linux/capability.h>
Chuck Levera246b012005-08-11 16:25:23 -040025#include <linux/pagemap.h>
26#include <linux/errno.h>
27#include <linux/socket.h>
28#include <linux/in.h>
29#include <linux/net.h>
30#include <linux/mm.h>
31#include <linux/udp.h>
32#include <linux/tcp.h>
33#include <linux/sunrpc/clnt.h>
Chuck Lever02107142006-01-03 09:55:49 +010034#include <linux/sunrpc/sched.h>
\"Talpey, Thomas\49c36fc2007-09-10 13:47:31 -040035#include <linux/sunrpc/xprtsock.h>
Chuck Levera246b012005-08-11 16:25:23 -040036#include <linux/file.h>
37
38#include <net/sock.h>
39#include <net/checksum.h>
40#include <net/udp.h>
41#include <net/tcp.h>
42
Chuck Lever9903cd12005-08-11 16:25:26 -040043/*
Chuck Leverc556b752005-11-01 12:24:48 -050044 * xprtsock tunables
45 */
46unsigned int xprt_udp_slot_table_entries = RPC_DEF_SLOT_TABLE;
47unsigned int xprt_tcp_slot_table_entries = RPC_DEF_SLOT_TABLE;
48
49unsigned int xprt_min_resvport = RPC_DEF_MIN_RESVPORT;
50unsigned int xprt_max_resvport = RPC_DEF_MAX_RESVPORT;
51
52/*
Chuck Leverfbf76682006-12-05 16:35:54 -050053 * We can register our own files under /proc/sys/sunrpc by
54 * calling register_sysctl_table() again. The files in that
55 * directory become the union of all files registered there.
56 *
57 * We simply need to make sure that we don't collide with
58 * someone else's file names!
59 */
60
61#ifdef RPC_DEBUG
62
63static unsigned int min_slot_table_size = RPC_MIN_SLOT_TABLE;
64static unsigned int max_slot_table_size = RPC_MAX_SLOT_TABLE;
65static unsigned int xprt_min_resvport_limit = RPC_MIN_RESVPORT;
66static unsigned int xprt_max_resvport_limit = RPC_MAX_RESVPORT;
67
68static struct ctl_table_header *sunrpc_table_header;
69
70/*
71 * FIXME: changing the UDP slot table size should also resize the UDP
72 * socket buffers for existing UDP transports
73 */
74static ctl_table xs_tunables_table[] = {
75 {
76 .ctl_name = CTL_SLOTTABLE_UDP,
77 .procname = "udp_slot_table_entries",
78 .data = &xprt_udp_slot_table_entries,
79 .maxlen = sizeof(unsigned int),
80 .mode = 0644,
81 .proc_handler = &proc_dointvec_minmax,
82 .strategy = &sysctl_intvec,
83 .extra1 = &min_slot_table_size,
84 .extra2 = &max_slot_table_size
85 },
86 {
87 .ctl_name = CTL_SLOTTABLE_TCP,
88 .procname = "tcp_slot_table_entries",
89 .data = &xprt_tcp_slot_table_entries,
90 .maxlen = sizeof(unsigned int),
91 .mode = 0644,
92 .proc_handler = &proc_dointvec_minmax,
93 .strategy = &sysctl_intvec,
94 .extra1 = &min_slot_table_size,
95 .extra2 = &max_slot_table_size
96 },
97 {
98 .ctl_name = CTL_MIN_RESVPORT,
99 .procname = "min_resvport",
100 .data = &xprt_min_resvport,
101 .maxlen = sizeof(unsigned int),
102 .mode = 0644,
103 .proc_handler = &proc_dointvec_minmax,
104 .strategy = &sysctl_intvec,
105 .extra1 = &xprt_min_resvport_limit,
106 .extra2 = &xprt_max_resvport_limit
107 },
108 {
109 .ctl_name = CTL_MAX_RESVPORT,
110 .procname = "max_resvport",
111 .data = &xprt_max_resvport,
112 .maxlen = sizeof(unsigned int),
113 .mode = 0644,
114 .proc_handler = &proc_dointvec_minmax,
115 .strategy = &sysctl_intvec,
116 .extra1 = &xprt_min_resvport_limit,
117 .extra2 = &xprt_max_resvport_limit
118 },
119 {
120 .ctl_name = 0,
121 },
122};
123
124static ctl_table sunrpc_table[] = {
125 {
126 .ctl_name = CTL_SUNRPC,
127 .procname = "sunrpc",
128 .mode = 0555,
129 .child = xs_tunables_table
130 },
131 {
132 .ctl_name = 0,
133 },
134};
135
136#endif
137
138/*
Chuck Lever262965f2005-08-11 16:25:56 -0400139 * How many times to try sending a request on a socket before waiting
140 * for the socket buffer to clear.
141 */
142#define XS_SENDMSG_RETRY (10U)
143
Chuck Lever03bf4b72005-08-25 16:25:55 -0700144/*
145 * Time out for an RPC UDP socket connect. UDP socket connects are
146 * synchronous, but we set a timeout anyway in case of resource
147 * exhaustion on the local host.
148 */
149#define XS_UDP_CONN_TO (5U * HZ)
150
151/*
152 * Wait duration for an RPC TCP connection to be established. Solaris
153 * NFS over TCP uses 60 seconds, for example, which is in line with how
154 * long a server takes to reboot.
155 */
156#define XS_TCP_CONN_TO (60U * HZ)
157
158/*
159 * Wait duration for a reply from the RPC portmapper.
160 */
161#define XS_BIND_TO (60U * HZ)
162
163/*
164 * Delay if a UDP socket connect error occurs. This is most likely some
165 * kind of resource problem on the local host.
166 */
167#define XS_UDP_REEST_TO (2U * HZ)
168
169/*
170 * The reestablish timeout allows clients to delay for a bit before attempting
171 * to reconnect to a server that just dropped our connection.
172 *
173 * We implement an exponential backoff when trying to reestablish a TCP
174 * transport connection with the server. Some servers like to drop a TCP
175 * connection when they are overworked, so we start with a short timeout and
176 * increase over time if the server is down or not responding.
177 */
178#define XS_TCP_INIT_REEST_TO (3U * HZ)
179#define XS_TCP_MAX_REEST_TO (5U * 60 * HZ)
180
181/*
182 * TCP idle timeout; client drops the transport socket if it is idle
183 * for this long. Note that we also timeout UDP sockets to prevent
184 * holding port numbers when there is no RPC traffic.
185 */
186#define XS_IDLE_DISC_TO (5U * 60 * HZ)
187
Chuck Levera246b012005-08-11 16:25:23 -0400188#ifdef RPC_DEBUG
189# undef RPC_DEBUG_DATA
Chuck Lever9903cd12005-08-11 16:25:26 -0400190# define RPCDBG_FACILITY RPCDBG_TRANS
Chuck Levera246b012005-08-11 16:25:23 -0400191#endif
192
Chuck Levera246b012005-08-11 16:25:23 -0400193#ifdef RPC_DEBUG_DATA
Chuck Lever9903cd12005-08-11 16:25:26 -0400194static void xs_pktdump(char *msg, u32 *packet, unsigned int count)
Chuck Levera246b012005-08-11 16:25:23 -0400195{
Chuck Lever9903cd12005-08-11 16:25:26 -0400196 u8 *buf = (u8 *) packet;
197 int j;
Chuck Levera246b012005-08-11 16:25:23 -0400198
Chuck Lever46121cf2007-01-31 12:14:08 -0500199 dprintk("RPC: %s\n", msg);
Chuck Levera246b012005-08-11 16:25:23 -0400200 for (j = 0; j < count && j < 128; j += 4) {
201 if (!(j & 31)) {
202 if (j)
203 dprintk("\n");
204 dprintk("0x%04x ", j);
205 }
206 dprintk("%02x%02x%02x%02x ",
207 buf[j], buf[j+1], buf[j+2], buf[j+3]);
208 }
209 dprintk("\n");
210}
211#else
Chuck Lever9903cd12005-08-11 16:25:26 -0400212static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count)
Chuck Levera246b012005-08-11 16:25:23 -0400213{
214 /* NOP */
215}
216#endif
217
Chuck Leverffc2e512006-12-05 16:35:11 -0500218struct sock_xprt {
219 struct rpc_xprt xprt;
Chuck Leveree0ac0c2006-12-05 16:35:15 -0500220
221 /*
222 * Network layer
223 */
224 struct socket * sock;
225 struct sock * inet;
Chuck Lever51971132006-12-05 16:35:19 -0500226
227 /*
228 * State of TCP reply receive
229 */
230 __be32 tcp_fraghdr,
231 tcp_xid;
232
233 u32 tcp_offset,
234 tcp_reclen;
235
236 unsigned long tcp_copied,
237 tcp_flags;
Chuck Leverc8475462006-12-05 16:35:26 -0500238
239 /*
240 * Connection of transports
241 */
Trond Myklebust34161db2006-12-07 15:48:15 -0500242 struct delayed_work connect_worker;
Frank van Maarseveend3bc9a12007-07-09 22:23:35 +0200243 struct sockaddr_storage addr;
Chuck Leverc8475462006-12-05 16:35:26 -0500244 unsigned short port;
Chuck Lever7c6e0662006-12-05 16:35:30 -0500245
246 /*
247 * UDP socket buffer size parameters
248 */
249 size_t rcvsize,
250 sndsize;
Chuck Lever314dfd72006-12-05 16:35:34 -0500251
252 /*
253 * Saved socket callback addresses
254 */
255 void (*old_data_ready)(struct sock *, int);
256 void (*old_state_change)(struct sock *);
257 void (*old_write_space)(struct sock *);
Chuck Leverffc2e512006-12-05 16:35:11 -0500258};
259
Chuck Levere136d092006-12-05 16:35:23 -0500260/*
261 * TCP receive state flags
262 */
263#define TCP_RCV_LAST_FRAG (1UL << 0)
264#define TCP_RCV_COPY_FRAGHDR (1UL << 1)
265#define TCP_RCV_COPY_XID (1UL << 2)
266#define TCP_RCV_COPY_DATA (1UL << 3)
267
Chuck Lever95392c52007-08-06 11:57:58 -0400268static inline struct sockaddr *xs_addr(struct rpc_xprt *xprt)
Chuck Leveredb267a2006-08-22 20:06:18 -0400269{
Chuck Lever95392c52007-08-06 11:57:58 -0400270 return (struct sockaddr *) &xprt->addr;
271}
272
273static inline struct sockaddr_in *xs_addr_in(struct rpc_xprt *xprt)
274{
275 return (struct sockaddr_in *) &xprt->addr;
276}
277
278static inline struct sockaddr_in6 *xs_addr_in6(struct rpc_xprt *xprt)
279{
280 return (struct sockaddr_in6 *) &xprt->addr;
281}
282
Chuck Leverba10f2c2007-08-06 11:57:07 -0400283static void xs_format_ipv4_peer_addresses(struct rpc_xprt *xprt)
Chuck Leveredb267a2006-08-22 20:06:18 -0400284{
Chuck Lever95392c52007-08-06 11:57:58 -0400285 struct sockaddr_in *addr = xs_addr_in(xprt);
Chuck Leveredb267a2006-08-22 20:06:18 -0400286 char *buf;
287
288 buf = kzalloc(20, GFP_KERNEL);
289 if (buf) {
Chuck Leverbda243d2007-08-06 11:56:52 -0400290 snprintf(buf, 20, NIPQUAD_FMT,
Chuck Leveredb267a2006-08-22 20:06:18 -0400291 NIPQUAD(addr->sin_addr.s_addr));
292 }
293 xprt->address_strings[RPC_DISPLAY_ADDR] = buf;
294
295 buf = kzalloc(8, GFP_KERNEL);
296 if (buf) {
297 snprintf(buf, 8, "%u",
298 ntohs(addr->sin_port));
299 }
300 xprt->address_strings[RPC_DISPLAY_PORT] = buf;
301
Chuck Lever0c43b3d2007-08-06 11:56:57 -0400302 buf = kzalloc(8, GFP_KERNEL);
303 if (buf) {
304 if (xprt->prot == IPPROTO_UDP)
305 snprintf(buf, 8, "udp");
306 else
307 snprintf(buf, 8, "tcp");
308 }
309 xprt->address_strings[RPC_DISPLAY_PROTO] = buf;
Chuck Leveredb267a2006-08-22 20:06:18 -0400310
311 buf = kzalloc(48, GFP_KERNEL);
312 if (buf) {
Chuck Leverbda243d2007-08-06 11:56:52 -0400313 snprintf(buf, 48, "addr="NIPQUAD_FMT" port=%u proto=%s",
Chuck Leveredb267a2006-08-22 20:06:18 -0400314 NIPQUAD(addr->sin_addr.s_addr),
315 ntohs(addr->sin_port),
316 xprt->prot == IPPROTO_UDP ? "udp" : "tcp");
317 }
318 xprt->address_strings[RPC_DISPLAY_ALL] = buf;
Chuck Leverfbfe3cc2007-08-06 11:57:02 -0400319
320 buf = kzalloc(10, GFP_KERNEL);
321 if (buf) {
322 snprintf(buf, 10, "%02x%02x%02x%02x",
323 NIPQUAD(addr->sin_addr.s_addr));
324 }
325 xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = buf;
326
327 buf = kzalloc(8, GFP_KERNEL);
328 if (buf) {
329 snprintf(buf, 8, "%4hx",
330 ntohs(addr->sin_port));
331 }
332 xprt->address_strings[RPC_DISPLAY_HEX_PORT] = buf;
Chuck Lever756805e2007-08-16 16:03:26 -0400333
334 buf = kzalloc(30, GFP_KERNEL);
335 if (buf) {
336 snprintf(buf, 30, NIPQUAD_FMT".%u.%u",
337 NIPQUAD(addr->sin_addr.s_addr),
338 ntohs(addr->sin_port) >> 8,
339 ntohs(addr->sin_port) & 0xff);
340 }
341 xprt->address_strings[RPC_DISPLAY_UNIVERSAL_ADDR] = buf;
\"Talpey, Thomas\4417c8c2007-09-10 13:43:05 -0400342
343 xprt->address_strings[RPC_DISPLAY_NETID] =
344 kstrdup(xprt->prot == IPPROTO_UDP ?
345 RPCBIND_NETID_UDP : RPCBIND_NETID_TCP, GFP_KERNEL);
Chuck Leveredb267a2006-08-22 20:06:18 -0400346}
347
Chuck Lever4b6473f2007-08-06 11:57:12 -0400348static void xs_format_ipv6_peer_addresses(struct rpc_xprt *xprt)
349{
Chuck Lever95392c52007-08-06 11:57:58 -0400350 struct sockaddr_in6 *addr = xs_addr_in6(xprt);
Chuck Lever4b6473f2007-08-06 11:57:12 -0400351 char *buf;
352
353 buf = kzalloc(40, GFP_KERNEL);
354 if (buf) {
355 snprintf(buf, 40, NIP6_FMT,
356 NIP6(addr->sin6_addr));
357 }
358 xprt->address_strings[RPC_DISPLAY_ADDR] = buf;
359
360 buf = kzalloc(8, GFP_KERNEL);
361 if (buf) {
362 snprintf(buf, 8, "%u",
363 ntohs(addr->sin6_port));
364 }
365 xprt->address_strings[RPC_DISPLAY_PORT] = buf;
366
367 buf = kzalloc(8, GFP_KERNEL);
368 if (buf) {
369 if (xprt->prot == IPPROTO_UDP)
370 snprintf(buf, 8, "udp");
371 else
372 snprintf(buf, 8, "tcp");
373 }
374 xprt->address_strings[RPC_DISPLAY_PROTO] = buf;
375
376 buf = kzalloc(64, GFP_KERNEL);
377 if (buf) {
378 snprintf(buf, 64, "addr="NIP6_FMT" port=%u proto=%s",
379 NIP6(addr->sin6_addr),
380 ntohs(addr->sin6_port),
381 xprt->prot == IPPROTO_UDP ? "udp" : "tcp");
382 }
383 xprt->address_strings[RPC_DISPLAY_ALL] = buf;
384
385 buf = kzalloc(36, GFP_KERNEL);
386 if (buf) {
387 snprintf(buf, 36, NIP6_SEQFMT,
388 NIP6(addr->sin6_addr));
389 }
390 xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = buf;
391
392 buf = kzalloc(8, GFP_KERNEL);
393 if (buf) {
394 snprintf(buf, 8, "%4hx",
395 ntohs(addr->sin6_port));
396 }
397 xprt->address_strings[RPC_DISPLAY_HEX_PORT] = buf;
Chuck Lever756805e2007-08-16 16:03:26 -0400398
399 buf = kzalloc(50, GFP_KERNEL);
400 if (buf) {
401 snprintf(buf, 50, NIP6_FMT".%u.%u",
402 NIP6(addr->sin6_addr),
403 ntohs(addr->sin6_port) >> 8,
404 ntohs(addr->sin6_port) & 0xff);
405 }
406 xprt->address_strings[RPC_DISPLAY_UNIVERSAL_ADDR] = buf;
\"Talpey, Thomas\4417c8c2007-09-10 13:43:05 -0400407
408 xprt->address_strings[RPC_DISPLAY_NETID] =
409 kstrdup(xprt->prot == IPPROTO_UDP ?
410 RPCBIND_NETID_UDP6 : RPCBIND_NETID_TCP6, GFP_KERNEL);
Chuck Leveredb267a2006-08-22 20:06:18 -0400411}
412
413static void xs_free_peer_addresses(struct rpc_xprt *xprt)
414{
Chuck Lever0c43b3d2007-08-06 11:56:57 -0400415 int i;
416
417 for (i = 0; i < RPC_DISPLAY_MAX; i++)
418 kfree(xprt->address_strings[i]);
Chuck Leveredb267a2006-08-22 20:06:18 -0400419}
420
Chuck Leverb4b5cc82005-08-11 16:25:29 -0400421#define XS_SENDMSG_FLAGS (MSG_DONTWAIT | MSG_NOSIGNAL)
422
Trond Myklebust24c56842006-10-17 15:06:22 -0400423static int xs_send_kvec(struct socket *sock, struct sockaddr *addr, int addrlen, struct kvec *vec, unsigned int base, int more)
Chuck Leverb4b5cc82005-08-11 16:25:29 -0400424{
Chuck Leverb4b5cc82005-08-11 16:25:29 -0400425 struct msghdr msg = {
426 .msg_name = addr,
427 .msg_namelen = addrlen,
Trond Myklebust24c56842006-10-17 15:06:22 -0400428 .msg_flags = XS_SENDMSG_FLAGS | (more ? MSG_MORE : 0),
429 };
430 struct kvec iov = {
431 .iov_base = vec->iov_base + base,
432 .iov_len = vec->iov_len - base,
Chuck Leverb4b5cc82005-08-11 16:25:29 -0400433 };
434
Trond Myklebust24c56842006-10-17 15:06:22 -0400435 if (iov.iov_len != 0)
Chuck Leverb4b5cc82005-08-11 16:25:29 -0400436 return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
437 return kernel_sendmsg(sock, &msg, NULL, 0, 0);
438}
439
Trond Myklebust24c56842006-10-17 15:06:22 -0400440static int xs_send_pagedata(struct socket *sock, struct xdr_buf *xdr, unsigned int base, int more)
Chuck Leverb4b5cc82005-08-11 16:25:29 -0400441{
Trond Myklebust24c56842006-10-17 15:06:22 -0400442 struct page **ppage;
443 unsigned int remainder;
444 int err, sent = 0;
Chuck Leverb4b5cc82005-08-11 16:25:29 -0400445
Trond Myklebust24c56842006-10-17 15:06:22 -0400446 remainder = xdr->page_len - base;
447 base += xdr->page_base;
448 ppage = xdr->pages + (base >> PAGE_SHIFT);
449 base &= ~PAGE_MASK;
450 for(;;) {
451 unsigned int len = min_t(unsigned int, PAGE_SIZE - base, remainder);
452 int flags = XS_SENDMSG_FLAGS;
453
454 remainder -= len;
455 if (remainder != 0 || more)
456 flags |= MSG_MORE;
457 err = sock->ops->sendpage(sock, *ppage, base, len, flags);
458 if (remainder == 0 || err != len)
459 break;
460 sent += err;
461 ppage++;
462 base = 0;
463 }
464 if (sent == 0)
465 return err;
466 if (err > 0)
467 sent += err;
468 return sent;
Chuck Leverb4b5cc82005-08-11 16:25:29 -0400469}
470
Chuck Lever9903cd12005-08-11 16:25:26 -0400471/**
472 * xs_sendpages - write pages directly to a socket
473 * @sock: socket to send on
474 * @addr: UDP only -- address of destination
475 * @addrlen: UDP only -- length of destination address
476 * @xdr: buffer containing this request
477 * @base: starting position in the buffer
478 *
Chuck Levera246b012005-08-11 16:25:23 -0400479 */
Trond Myklebust24c56842006-10-17 15:06:22 -0400480static int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base)
Chuck Levera246b012005-08-11 16:25:23 -0400481{
Trond Myklebust24c56842006-10-17 15:06:22 -0400482 unsigned int remainder = xdr->len - base;
483 int err, sent = 0;
Chuck Levera246b012005-08-11 16:25:23 -0400484
Chuck Lever262965f2005-08-11 16:25:56 -0400485 if (unlikely(!sock))
486 return -ENOTCONN;
487
488 clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags);
Trond Myklebust24c56842006-10-17 15:06:22 -0400489 if (base != 0) {
490 addr = NULL;
491 addrlen = 0;
492 }
Chuck Lever262965f2005-08-11 16:25:56 -0400493
Trond Myklebust24c56842006-10-17 15:06:22 -0400494 if (base < xdr->head[0].iov_len || addr != NULL) {
495 unsigned int len = xdr->head[0].iov_len - base;
496 remainder -= len;
497 err = xs_send_kvec(sock, addr, addrlen, &xdr->head[0], base, remainder != 0);
498 if (remainder == 0 || err != len)
Chuck Levera246b012005-08-11 16:25:23 -0400499 goto out;
Trond Myklebust24c56842006-10-17 15:06:22 -0400500 sent += err;
Chuck Levera246b012005-08-11 16:25:23 -0400501 base = 0;
502 } else
Trond Myklebust24c56842006-10-17 15:06:22 -0400503 base -= xdr->head[0].iov_len;
Chuck Levera246b012005-08-11 16:25:23 -0400504
Trond Myklebust24c56842006-10-17 15:06:22 -0400505 if (base < xdr->page_len) {
506 unsigned int len = xdr->page_len - base;
507 remainder -= len;
508 err = xs_send_pagedata(sock, xdr, base, remainder != 0);
509 if (remainder == 0 || err != len)
Chuck Levera246b012005-08-11 16:25:23 -0400510 goto out;
Trond Myklebust24c56842006-10-17 15:06:22 -0400511 sent += err;
Chuck Levera246b012005-08-11 16:25:23 -0400512 base = 0;
Trond Myklebust24c56842006-10-17 15:06:22 -0400513 } else
514 base -= xdr->page_len;
515
516 if (base >= xdr->tail[0].iov_len)
517 return sent;
518 err = xs_send_kvec(sock, NULL, 0, &xdr->tail[0], base, 0);
Chuck Levera246b012005-08-11 16:25:23 -0400519out:
Trond Myklebust24c56842006-10-17 15:06:22 -0400520 if (sent == 0)
521 return err;
522 if (err > 0)
523 sent += err;
524 return sent;
Chuck Levera246b012005-08-11 16:25:23 -0400525}
526
Chuck Lever9903cd12005-08-11 16:25:26 -0400527/**
Chuck Lever262965f2005-08-11 16:25:56 -0400528 * xs_nospace - place task on wait queue if transmit was incomplete
529 * @task: task to put to sleep
Chuck Lever9903cd12005-08-11 16:25:26 -0400530 *
Chuck Levera246b012005-08-11 16:25:23 -0400531 */
Chuck Lever262965f2005-08-11 16:25:56 -0400532static void xs_nospace(struct rpc_task *task)
Chuck Levera246b012005-08-11 16:25:23 -0400533{
Chuck Lever262965f2005-08-11 16:25:56 -0400534 struct rpc_rqst *req = task->tk_rqstp;
535 struct rpc_xprt *xprt = req->rq_xprt;
Chuck Leveree0ac0c2006-12-05 16:35:15 -0500536 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
Chuck Levera246b012005-08-11 16:25:23 -0400537
Chuck Lever46121cf2007-01-31 12:14:08 -0500538 dprintk("RPC: %5u xmit incomplete (%u left of %u)\n",
Chuck Lever262965f2005-08-11 16:25:56 -0400539 task->tk_pid, req->rq_slen - req->rq_bytes_sent,
540 req->rq_slen);
541
Chuck Leveree0ac0c2006-12-05 16:35:15 -0500542 if (test_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags)) {
Chuck Lever262965f2005-08-11 16:25:56 -0400543 /* Protect against races with write_space */
544 spin_lock_bh(&xprt->transport_lock);
545
546 /* Don't race with disconnect */
547 if (!xprt_connected(xprt))
548 task->tk_status = -ENOTCONN;
Chuck Leveree0ac0c2006-12-05 16:35:15 -0500549 else if (test_bit(SOCK_NOSPACE, &transport->sock->flags))
Chuck Lever262965f2005-08-11 16:25:56 -0400550 xprt_wait_for_buffer_space(task);
551
552 spin_unlock_bh(&xprt->transport_lock);
553 } else
554 /* Keep holding the socket if it is blocked */
555 rpc_delay(task, HZ>>4);
556}
557
558/**
559 * xs_udp_send_request - write an RPC request to a UDP socket
560 * @task: address of RPC task that manages the state of an RPC request
561 *
562 * Return values:
563 * 0: The request has been sent
564 * EAGAIN: The socket was blocked, please call again later to
565 * complete the request
566 * ENOTCONN: Caller needs to invoke connect logic then call again
567 * other: Some other error occured, the request was not sent
568 */
569static int xs_udp_send_request(struct rpc_task *task)
570{
571 struct rpc_rqst *req = task->tk_rqstp;
572 struct rpc_xprt *xprt = req->rq_xprt;
Chuck Leveree0ac0c2006-12-05 16:35:15 -0500573 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
Chuck Lever262965f2005-08-11 16:25:56 -0400574 struct xdr_buf *xdr = &req->rq_snd_buf;
575 int status;
Chuck Levera246b012005-08-11 16:25:23 -0400576
Chuck Lever9903cd12005-08-11 16:25:26 -0400577 xs_pktdump("packet data:",
Chuck Levera246b012005-08-11 16:25:23 -0400578 req->rq_svec->iov_base,
579 req->rq_svec->iov_len);
580
Chuck Lever262965f2005-08-11 16:25:56 -0400581 req->rq_xtime = jiffies;
Chuck Leveree0ac0c2006-12-05 16:35:15 -0500582 status = xs_sendpages(transport->sock,
Chuck Lever95392c52007-08-06 11:57:58 -0400583 xs_addr(xprt),
Chuck Leveree0ac0c2006-12-05 16:35:15 -0500584 xprt->addrlen, xdr,
585 req->rq_bytes_sent);
Chuck Levera246b012005-08-11 16:25:23 -0400586
Chuck Lever46121cf2007-01-31 12:14:08 -0500587 dprintk("RPC: xs_udp_send_request(%u) = %d\n",
Chuck Lever262965f2005-08-11 16:25:56 -0400588 xdr->len - req->rq_bytes_sent, status);
Chuck Levera246b012005-08-11 16:25:23 -0400589
Trond Myklebust21997002007-10-01 11:43:37 -0400590 if (status >= 0) {
591 task->tk_bytes_sent += status;
592 if (status >= req->rq_slen)
593 return 0;
594 /* Still some bytes left; set up for a retry later. */
Chuck Lever262965f2005-08-11 16:25:56 -0400595 status = -EAGAIN;
Trond Myklebust21997002007-10-01 11:43:37 -0400596 }
Chuck Levera246b012005-08-11 16:25:23 -0400597
Chuck Lever262965f2005-08-11 16:25:56 -0400598 switch (status) {
599 case -ENETUNREACH:
600 case -EPIPE:
Chuck Levera246b012005-08-11 16:25:23 -0400601 case -ECONNREFUSED:
602 /* When the server has died, an ICMP port unreachable message
Chuck Lever9903cd12005-08-11 16:25:26 -0400603 * prompts ECONNREFUSED. */
Chuck Levera246b012005-08-11 16:25:23 -0400604 break;
Chuck Lever262965f2005-08-11 16:25:56 -0400605 case -EAGAIN:
606 xs_nospace(task);
Chuck Levera246b012005-08-11 16:25:23 -0400607 break;
608 default:
Chuck Lever46121cf2007-01-31 12:14:08 -0500609 dprintk("RPC: sendmsg returned unrecognized error %d\n",
Chuck Lever262965f2005-08-11 16:25:56 -0400610 -status);
Chuck Lever9903cd12005-08-11 16:25:26 -0400611 break;
Chuck Levera246b012005-08-11 16:25:23 -0400612 }
Chuck Lever262965f2005-08-11 16:25:56 -0400613
614 return status;
Chuck Levera246b012005-08-11 16:25:23 -0400615}
616
Chuck Lever808012f2005-08-25 16:25:49 -0700617static inline void xs_encode_tcp_record_marker(struct xdr_buf *buf)
618{
619 u32 reclen = buf->len - sizeof(rpc_fraghdr);
620 rpc_fraghdr *base = buf->head[0].iov_base;
621 *base = htonl(RPC_LAST_STREAM_FRAGMENT | reclen);
622}
623
Chuck Lever9903cd12005-08-11 16:25:26 -0400624/**
Chuck Lever262965f2005-08-11 16:25:56 -0400625 * xs_tcp_send_request - write an RPC request to a TCP socket
Chuck Lever9903cd12005-08-11 16:25:26 -0400626 * @task: address of RPC task that manages the state of an RPC request
627 *
628 * Return values:
Chuck Lever262965f2005-08-11 16:25:56 -0400629 * 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
633 * other: Some other error occured, the request was not sent
Chuck Lever9903cd12005-08-11 16:25:26 -0400634 *
635 * XXX: In the case of soft timeouts, should we eventually give up
Chuck Lever262965f2005-08-11 16:25:56 -0400636 * if sendmsg is not able to make progress?
Chuck Lever9903cd12005-08-11 16:25:26 -0400637 */
Chuck Lever262965f2005-08-11 16:25:56 -0400638static int xs_tcp_send_request(struct rpc_task *task)
Chuck Levera246b012005-08-11 16:25:23 -0400639{
640 struct rpc_rqst *req = task->tk_rqstp;
641 struct rpc_xprt *xprt = req->rq_xprt;
Chuck Leveree0ac0c2006-12-05 16:35:15 -0500642 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
Chuck Lever262965f2005-08-11 16:25:56 -0400643 struct xdr_buf *xdr = &req->rq_snd_buf;
Chuck Leverb595bb12007-08-06 11:56:42 -0400644 int status;
645 unsigned int retry = 0;
Chuck Levera246b012005-08-11 16:25:23 -0400646
Chuck Lever808012f2005-08-25 16:25:49 -0700647 xs_encode_tcp_record_marker(&req->rq_snd_buf);
Chuck Levera246b012005-08-11 16:25:23 -0400648
Chuck Lever262965f2005-08-11 16:25:56 -0400649 xs_pktdump("packet data:",
650 req->rq_svec->iov_base,
651 req->rq_svec->iov_len);
Chuck Levera246b012005-08-11 16:25:23 -0400652
653 /* Continue transmitting the packet/record. We must be careful
654 * to cope with writespace callbacks arriving _after_ we have
Chuck Lever262965f2005-08-11 16:25:56 -0400655 * called sendmsg(). */
Chuck Levera246b012005-08-11 16:25:23 -0400656 while (1) {
657 req->rq_xtime = jiffies;
Chuck Leveree0ac0c2006-12-05 16:35:15 -0500658 status = xs_sendpages(transport->sock,
659 NULL, 0, xdr, req->rq_bytes_sent);
Chuck Levera246b012005-08-11 16:25:23 -0400660
Chuck Lever46121cf2007-01-31 12:14:08 -0500661 dprintk("RPC: xs_tcp_send_request(%u) = %d\n",
Chuck Lever262965f2005-08-11 16:25:56 -0400662 xdr->len - req->rq_bytes_sent, status);
663
664 if (unlikely(status < 0))
Chuck Levera246b012005-08-11 16:25:23 -0400665 break;
666
Chuck Lever262965f2005-08-11 16:25:56 -0400667 /* If we've sent the entire packet, immediately
668 * reset the count of bytes sent. */
669 req->rq_bytes_sent += status;
Chuck Leveref759a22006-03-20 13:44:17 -0500670 task->tk_bytes_sent += status;
Chuck Lever262965f2005-08-11 16:25:56 -0400671 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
672 req->rq_bytes_sent = 0;
673 return 0;
Chuck Levera246b012005-08-11 16:25:23 -0400674 }
675
Chuck Levera246b012005-08-11 16:25:23 -0400676 status = -EAGAIN;
Chuck Lever262965f2005-08-11 16:25:56 -0400677 if (retry++ > XS_SENDMSG_RETRY)
Chuck Levera246b012005-08-11 16:25:23 -0400678 break;
679 }
680
Chuck Lever262965f2005-08-11 16:25:56 -0400681 switch (status) {
682 case -EAGAIN:
683 xs_nospace(task);
684 break;
685 case -ECONNREFUSED:
686 case -ECONNRESET:
687 case -ENOTCONN:
688 case -EPIPE:
689 status = -ENOTCONN;
690 break;
691 default:
Chuck Lever46121cf2007-01-31 12:14:08 -0500692 dprintk("RPC: sendmsg returned unrecognized error %d\n",
Chuck Lever262965f2005-08-11 16:25:56 -0400693 -status);
Chuck Lever43118c22005-08-25 16:25:49 -0700694 xprt_disconnect(xprt);
Chuck Lever262965f2005-08-11 16:25:56 -0400695 break;
Chuck Levera246b012005-08-11 16:25:23 -0400696 }
Chuck Lever262965f2005-08-11 16:25:56 -0400697
Chuck Levera246b012005-08-11 16:25:23 -0400698 return status;
699}
700
Chuck Lever9903cd12005-08-11 16:25:26 -0400701/**
Trond Myklebuste0ab53d2006-07-27 17:22:50 -0400702 * xs_tcp_release_xprt - clean up after a tcp transmission
703 * @xprt: transport
704 * @task: rpc task
705 *
706 * This cleans up if an error causes us to abort the transmission of a request.
707 * In this case, the socket may need to be reset in order to avoid confusing
708 * the server.
709 */
710static void xs_tcp_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
711{
712 struct rpc_rqst *req;
713
714 if (task != xprt->snd_task)
715 return;
716 if (task == NULL)
717 goto out_release;
718 req = task->tk_rqstp;
719 if (req->rq_bytes_sent == 0)
720 goto out_release;
721 if (req->rq_bytes_sent == req->rq_snd_buf.len)
722 goto out_release;
723 set_bit(XPRT_CLOSE_WAIT, &task->tk_xprt->state);
724out_release:
725 xprt_release_xprt(xprt, task);
726}
727
728/**
Chuck Lever9903cd12005-08-11 16:25:26 -0400729 * xs_close - close a socket
730 * @xprt: transport
731 *
Chuck Lever3167e122005-08-25 16:25:55 -0700732 * This is used when all requests are complete; ie, no DRC state remains
733 * on the server we want to save.
Chuck Levera246b012005-08-11 16:25:23 -0400734 */
Chuck Lever9903cd12005-08-11 16:25:26 -0400735static void xs_close(struct rpc_xprt *xprt)
Chuck Levera246b012005-08-11 16:25:23 -0400736{
Chuck Leveree0ac0c2006-12-05 16:35:15 -0500737 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
738 struct socket *sock = transport->sock;
739 struct sock *sk = transport->inet;
Chuck Levera246b012005-08-11 16:25:23 -0400740
741 if (!sk)
Trond Myklebust632e3bd2006-01-03 09:55:55 +0100742 goto clear_close_wait;
Chuck Levera246b012005-08-11 16:25:23 -0400743
Chuck Lever46121cf2007-01-31 12:14:08 -0500744 dprintk("RPC: xs_close xprt %p\n", xprt);
Chuck Lever9903cd12005-08-11 16:25:26 -0400745
Chuck Levera246b012005-08-11 16:25:23 -0400746 write_lock_bh(&sk->sk_callback_lock);
Chuck Leveree0ac0c2006-12-05 16:35:15 -0500747 transport->inet = NULL;
748 transport->sock = NULL;
Chuck Levera246b012005-08-11 16:25:23 -0400749
Chuck Lever9903cd12005-08-11 16:25:26 -0400750 sk->sk_user_data = NULL;
Chuck Lever314dfd72006-12-05 16:35:34 -0500751 sk->sk_data_ready = transport->old_data_ready;
752 sk->sk_state_change = transport->old_state_change;
753 sk->sk_write_space = transport->old_write_space;
Chuck Levera246b012005-08-11 16:25:23 -0400754 write_unlock_bh(&sk->sk_callback_lock);
755
Chuck Lever9903cd12005-08-11 16:25:26 -0400756 sk->sk_no_check = 0;
Chuck Levera246b012005-08-11 16:25:23 -0400757
758 sock_release(sock);
Trond Myklebust632e3bd2006-01-03 09:55:55 +0100759clear_close_wait:
760 smp_mb__before_clear_bit();
761 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
762 smp_mb__after_clear_bit();
Chuck Levera246b012005-08-11 16:25:23 -0400763}
764
Chuck Lever9903cd12005-08-11 16:25:26 -0400765/**
766 * xs_destroy - prepare to shutdown a transport
767 * @xprt: doomed transport
768 *
769 */
770static void xs_destroy(struct rpc_xprt *xprt)
Chuck Levera246b012005-08-11 16:25:23 -0400771{
Chuck Leverc8475462006-12-05 16:35:26 -0500772 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
773
Chuck Lever46121cf2007-01-31 12:14:08 -0500774 dprintk("RPC: xs_destroy xprt %p\n", xprt);
Chuck Lever9903cd12005-08-11 16:25:26 -0400775
Trond Myklebustc1384c92007-06-14 18:00:42 -0400776 cancel_rearming_delayed_work(&transport->connect_worker);
Chuck Levera246b012005-08-11 16:25:23 -0400777
778 xprt_disconnect(xprt);
Chuck Lever9903cd12005-08-11 16:25:26 -0400779 xs_close(xprt);
Chuck Leveredb267a2006-08-22 20:06:18 -0400780 xs_free_peer_addresses(xprt);
Chuck Levera246b012005-08-11 16:25:23 -0400781 kfree(xprt->slot);
Chuck Leverc8541ec2006-10-17 14:44:27 -0400782 kfree(xprt);
\"Talpey, Thomas\bc255712007-09-10 13:46:39 -0400783 module_put(THIS_MODULE);
Chuck Levera246b012005-08-11 16:25:23 -0400784}
785
Chuck Lever9903cd12005-08-11 16:25:26 -0400786static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
Chuck Levera246b012005-08-11 16:25:23 -0400787{
Chuck Lever9903cd12005-08-11 16:25:26 -0400788 return (struct rpc_xprt *) sk->sk_user_data;
789}
790
791/**
792 * xs_udp_data_ready - "data ready" callback for UDP sockets
793 * @sk: socket with data to read
794 * @len: how much data to read
795 *
796 */
797static void xs_udp_data_ready(struct sock *sk, int len)
798{
799 struct rpc_task *task;
800 struct rpc_xprt *xprt;
Chuck Levera246b012005-08-11 16:25:23 -0400801 struct rpc_rqst *rovr;
Chuck Lever9903cd12005-08-11 16:25:26 -0400802 struct sk_buff *skb;
Chuck Levera246b012005-08-11 16:25:23 -0400803 int err, repsize, copied;
Alexey Dobriyand8ed0292006-09-26 22:29:38 -0700804 u32 _xid;
805 __be32 *xp;
Chuck Levera246b012005-08-11 16:25:23 -0400806
807 read_lock(&sk->sk_callback_lock);
Chuck Lever46121cf2007-01-31 12:14:08 -0500808 dprintk("RPC: xs_udp_data_ready...\n");
Chuck Lever9903cd12005-08-11 16:25:26 -0400809 if (!(xprt = xprt_from_sock(sk)))
Chuck Levera246b012005-08-11 16:25:23 -0400810 goto out;
Chuck Levera246b012005-08-11 16:25:23 -0400811
812 if ((skb = skb_recv_datagram(sk, 0, 1, &err)) == NULL)
813 goto out;
814
815 if (xprt->shutdown)
816 goto dropit;
817
818 repsize = skb->len - sizeof(struct udphdr);
819 if (repsize < 4) {
Chuck Lever46121cf2007-01-31 12:14:08 -0500820 dprintk("RPC: impossible RPC reply size %d!\n", repsize);
Chuck Levera246b012005-08-11 16:25:23 -0400821 goto dropit;
822 }
823
824 /* Copy the XID from the skb... */
825 xp = skb_header_pointer(skb, sizeof(struct udphdr),
826 sizeof(_xid), &_xid);
827 if (xp == NULL)
828 goto dropit;
829
830 /* Look up and lock the request corresponding to the given XID */
Chuck Lever4a0f8c02005-08-11 16:25:32 -0400831 spin_lock(&xprt->transport_lock);
Chuck Levera246b012005-08-11 16:25:23 -0400832 rovr = xprt_lookup_rqst(xprt, *xp);
833 if (!rovr)
834 goto out_unlock;
835 task = rovr->rq_task;
836
Chuck Levera246b012005-08-11 16:25:23 -0400837 if ((copied = rovr->rq_private_buf.buflen) > repsize)
838 copied = repsize;
839
840 /* Suck it into the iovec, verify checksum if not done by hw. */
841 if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb))
842 goto out_unlock;
843
844 /* Something worked... */
845 dst_confirm(skb->dst);
846
Chuck Lever1570c1e2005-08-25 16:25:52 -0700847 xprt_adjust_cwnd(task, copied);
848 xprt_update_rtt(task);
849 xprt_complete_rqst(task, copied);
Chuck Levera246b012005-08-11 16:25:23 -0400850
851 out_unlock:
Chuck Lever4a0f8c02005-08-11 16:25:32 -0400852 spin_unlock(&xprt->transport_lock);
Chuck Levera246b012005-08-11 16:25:23 -0400853 dropit:
854 skb_free_datagram(sk, skb);
855 out:
856 read_unlock(&sk->sk_callback_lock);
857}
858
Chuck Leverdd456472006-12-05 16:35:44 -0500859static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
Chuck Levera246b012005-08-11 16:25:23 -0400860{
Chuck Lever51971132006-12-05 16:35:19 -0500861 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
Chuck Levera246b012005-08-11 16:25:23 -0400862 size_t len, used;
863 char *p;
864
Chuck Lever51971132006-12-05 16:35:19 -0500865 p = ((char *) &transport->tcp_fraghdr) + transport->tcp_offset;
866 len = sizeof(transport->tcp_fraghdr) - transport->tcp_offset;
Chuck Lever9d292312006-12-05 16:35:41 -0500867 used = xdr_skb_read_bits(desc, p, len);
Chuck Lever51971132006-12-05 16:35:19 -0500868 transport->tcp_offset += used;
Chuck Levera246b012005-08-11 16:25:23 -0400869 if (used != len)
870 return;
Chuck Lever808012f2005-08-25 16:25:49 -0700871
Chuck Lever51971132006-12-05 16:35:19 -0500872 transport->tcp_reclen = ntohl(transport->tcp_fraghdr);
873 if (transport->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
Chuck Levere136d092006-12-05 16:35:23 -0500874 transport->tcp_flags |= TCP_RCV_LAST_FRAG;
Chuck Levera246b012005-08-11 16:25:23 -0400875 else
Chuck Levere136d092006-12-05 16:35:23 -0500876 transport->tcp_flags &= ~TCP_RCV_LAST_FRAG;
Chuck Lever51971132006-12-05 16:35:19 -0500877 transport->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
Chuck Lever808012f2005-08-25 16:25:49 -0700878
Chuck Levere136d092006-12-05 16:35:23 -0500879 transport->tcp_flags &= ~TCP_RCV_COPY_FRAGHDR;
Chuck Lever51971132006-12-05 16:35:19 -0500880 transport->tcp_offset = 0;
Chuck Lever808012f2005-08-25 16:25:49 -0700881
Chuck Levera246b012005-08-11 16:25:23 -0400882 /* Sanity check of the record length */
Chuck Lever51971132006-12-05 16:35:19 -0500883 if (unlikely(transport->tcp_reclen < 4)) {
Chuck Lever46121cf2007-01-31 12:14:08 -0500884 dprintk("RPC: invalid TCP record fragment length\n");
Chuck Levera246b012005-08-11 16:25:23 -0400885 xprt_disconnect(xprt);
Chuck Lever9903cd12005-08-11 16:25:26 -0400886 return;
Chuck Levera246b012005-08-11 16:25:23 -0400887 }
Chuck Lever46121cf2007-01-31 12:14:08 -0500888 dprintk("RPC: reading TCP record fragment of length %d\n",
Chuck Lever51971132006-12-05 16:35:19 -0500889 transport->tcp_reclen);
Chuck Levera246b012005-08-11 16:25:23 -0400890}
891
Chuck Lever51971132006-12-05 16:35:19 -0500892static void xs_tcp_check_fraghdr(struct sock_xprt *transport)
Chuck Levera246b012005-08-11 16:25:23 -0400893{
Chuck Lever51971132006-12-05 16:35:19 -0500894 if (transport->tcp_offset == transport->tcp_reclen) {
Chuck Levere136d092006-12-05 16:35:23 -0500895 transport->tcp_flags |= TCP_RCV_COPY_FRAGHDR;
Chuck Lever51971132006-12-05 16:35:19 -0500896 transport->tcp_offset = 0;
Chuck Levere136d092006-12-05 16:35:23 -0500897 if (transport->tcp_flags & TCP_RCV_LAST_FRAG) {
898 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
899 transport->tcp_flags |= TCP_RCV_COPY_XID;
Chuck Lever51971132006-12-05 16:35:19 -0500900 transport->tcp_copied = 0;
Chuck Levera246b012005-08-11 16:25:23 -0400901 }
902 }
903}
904
Chuck Leverdd456472006-12-05 16:35:44 -0500905static inline void xs_tcp_read_xid(struct sock_xprt *transport, struct xdr_skb_reader *desc)
Chuck Levera246b012005-08-11 16:25:23 -0400906{
907 size_t len, used;
908 char *p;
909
Chuck Lever51971132006-12-05 16:35:19 -0500910 len = sizeof(transport->tcp_xid) - transport->tcp_offset;
Chuck Lever46121cf2007-01-31 12:14:08 -0500911 dprintk("RPC: reading XID (%Zu bytes)\n", len);
Chuck Lever51971132006-12-05 16:35:19 -0500912 p = ((char *) &transport->tcp_xid) + transport->tcp_offset;
Chuck Lever9d292312006-12-05 16:35:41 -0500913 used = xdr_skb_read_bits(desc, p, len);
Chuck Lever51971132006-12-05 16:35:19 -0500914 transport->tcp_offset += used;
Chuck Levera246b012005-08-11 16:25:23 -0400915 if (used != len)
916 return;
Chuck Levere136d092006-12-05 16:35:23 -0500917 transport->tcp_flags &= ~TCP_RCV_COPY_XID;
918 transport->tcp_flags |= TCP_RCV_COPY_DATA;
Chuck Lever51971132006-12-05 16:35:19 -0500919 transport->tcp_copied = 4;
Chuck Lever46121cf2007-01-31 12:14:08 -0500920 dprintk("RPC: reading reply for XID %08x\n",
Chuck Lever51971132006-12-05 16:35:19 -0500921 ntohl(transport->tcp_xid));
922 xs_tcp_check_fraghdr(transport);
Chuck Levera246b012005-08-11 16:25:23 -0400923}
924
Chuck Leverdd456472006-12-05 16:35:44 -0500925static inline void xs_tcp_read_request(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
Chuck Levera246b012005-08-11 16:25:23 -0400926{
Chuck Lever51971132006-12-05 16:35:19 -0500927 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
Chuck Levera246b012005-08-11 16:25:23 -0400928 struct rpc_rqst *req;
929 struct xdr_buf *rcvbuf;
930 size_t len;
931 ssize_t r;
932
933 /* Find and lock the request corresponding to this xid */
Chuck Lever4a0f8c02005-08-11 16:25:32 -0400934 spin_lock(&xprt->transport_lock);
Chuck Lever51971132006-12-05 16:35:19 -0500935 req = xprt_lookup_rqst(xprt, transport->tcp_xid);
Chuck Levera246b012005-08-11 16:25:23 -0400936 if (!req) {
Chuck Levere136d092006-12-05 16:35:23 -0500937 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
Chuck Lever46121cf2007-01-31 12:14:08 -0500938 dprintk("RPC: XID %08x request not found!\n",
Chuck Lever51971132006-12-05 16:35:19 -0500939 ntohl(transport->tcp_xid));
Chuck Lever4a0f8c02005-08-11 16:25:32 -0400940 spin_unlock(&xprt->transport_lock);
Chuck Levera246b012005-08-11 16:25:23 -0400941 return;
942 }
943
944 rcvbuf = &req->rq_private_buf;
945 len = desc->count;
Chuck Lever51971132006-12-05 16:35:19 -0500946 if (len > transport->tcp_reclen - transport->tcp_offset) {
Chuck Leverdd456472006-12-05 16:35:44 -0500947 struct xdr_skb_reader my_desc;
Chuck Levera246b012005-08-11 16:25:23 -0400948
Chuck Lever51971132006-12-05 16:35:19 -0500949 len = transport->tcp_reclen - transport->tcp_offset;
Chuck Levera246b012005-08-11 16:25:23 -0400950 memcpy(&my_desc, desc, sizeof(my_desc));
951 my_desc.count = len;
Chuck Lever51971132006-12-05 16:35:19 -0500952 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
Chuck Lever9d292312006-12-05 16:35:41 -0500953 &my_desc, xdr_skb_read_bits);
Chuck Levera246b012005-08-11 16:25:23 -0400954 desc->count -= r;
955 desc->offset += r;
956 } else
Chuck Lever51971132006-12-05 16:35:19 -0500957 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
Chuck Lever9d292312006-12-05 16:35:41 -0500958 desc, xdr_skb_read_bits);
Chuck Levera246b012005-08-11 16:25:23 -0400959
960 if (r > 0) {
Chuck Lever51971132006-12-05 16:35:19 -0500961 transport->tcp_copied += r;
962 transport->tcp_offset += r;
Chuck Levera246b012005-08-11 16:25:23 -0400963 }
964 if (r != len) {
965 /* Error when copying to the receive buffer,
966 * usually because we weren't able to allocate
967 * additional buffer pages. All we can do now
Chuck Levere136d092006-12-05 16:35:23 -0500968 * is turn off TCP_RCV_COPY_DATA, so the request
Chuck Levera246b012005-08-11 16:25:23 -0400969 * will not receive any additional updates,
970 * and time out.
971 * Any remaining data from this record will
972 * be discarded.
973 */
Chuck Levere136d092006-12-05 16:35:23 -0500974 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
Chuck Lever46121cf2007-01-31 12:14:08 -0500975 dprintk("RPC: XID %08x truncated request\n",
Chuck Lever51971132006-12-05 16:35:19 -0500976 ntohl(transport->tcp_xid));
Chuck Lever46121cf2007-01-31 12:14:08 -0500977 dprintk("RPC: xprt = %p, tcp_copied = %lu, "
978 "tcp_offset = %u, tcp_reclen = %u\n",
979 xprt, transport->tcp_copied,
980 transport->tcp_offset, transport->tcp_reclen);
Chuck Levera246b012005-08-11 16:25:23 -0400981 goto out;
982 }
983
Chuck Lever46121cf2007-01-31 12:14:08 -0500984 dprintk("RPC: XID %08x read %Zd bytes\n",
Chuck Lever51971132006-12-05 16:35:19 -0500985 ntohl(transport->tcp_xid), r);
Chuck Lever46121cf2007-01-31 12:14:08 -0500986 dprintk("RPC: xprt = %p, tcp_copied = %lu, tcp_offset = %u, "
987 "tcp_reclen = %u\n", xprt, transport->tcp_copied,
988 transport->tcp_offset, transport->tcp_reclen);
Chuck Levera246b012005-08-11 16:25:23 -0400989
Chuck Lever51971132006-12-05 16:35:19 -0500990 if (transport->tcp_copied == req->rq_private_buf.buflen)
Chuck Levere136d092006-12-05 16:35:23 -0500991 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
Chuck Lever51971132006-12-05 16:35:19 -0500992 else if (transport->tcp_offset == transport->tcp_reclen) {
Chuck Levere136d092006-12-05 16:35:23 -0500993 if (transport->tcp_flags & TCP_RCV_LAST_FRAG)
994 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
Chuck Levera246b012005-08-11 16:25:23 -0400995 }
996
997out:
Chuck Levere136d092006-12-05 16:35:23 -0500998 if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
Chuck Lever51971132006-12-05 16:35:19 -0500999 xprt_complete_rqst(req->rq_task, transport->tcp_copied);
Chuck Lever4a0f8c02005-08-11 16:25:32 -04001000 spin_unlock(&xprt->transport_lock);
Chuck Lever51971132006-12-05 16:35:19 -05001001 xs_tcp_check_fraghdr(transport);
Chuck Levera246b012005-08-11 16:25:23 -04001002}
1003
Chuck Leverdd456472006-12-05 16:35:44 -05001004static inline void xs_tcp_read_discard(struct sock_xprt *transport, struct xdr_skb_reader *desc)
Chuck Levera246b012005-08-11 16:25:23 -04001005{
1006 size_t len;
1007
Chuck Lever51971132006-12-05 16:35:19 -05001008 len = transport->tcp_reclen - transport->tcp_offset;
Chuck Levera246b012005-08-11 16:25:23 -04001009 if (len > desc->count)
1010 len = desc->count;
1011 desc->count -= len;
1012 desc->offset += len;
Chuck Lever51971132006-12-05 16:35:19 -05001013 transport->tcp_offset += len;
Chuck Lever46121cf2007-01-31 12:14:08 -05001014 dprintk("RPC: discarded %Zu bytes\n", len);
Chuck Lever51971132006-12-05 16:35:19 -05001015 xs_tcp_check_fraghdr(transport);
Chuck Levera246b012005-08-11 16:25:23 -04001016}
1017
Chuck Lever9903cd12005-08-11 16:25:26 -04001018static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
Chuck Levera246b012005-08-11 16:25:23 -04001019{
1020 struct rpc_xprt *xprt = rd_desc->arg.data;
Chuck Lever51971132006-12-05 16:35:19 -05001021 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
Chuck Leverdd456472006-12-05 16:35:44 -05001022 struct xdr_skb_reader desc = {
Chuck Levera246b012005-08-11 16:25:23 -04001023 .skb = skb,
1024 .offset = offset,
1025 .count = len,
Chuck Lever9903cd12005-08-11 16:25:26 -04001026 };
Chuck Levera246b012005-08-11 16:25:23 -04001027
Chuck Lever46121cf2007-01-31 12:14:08 -05001028 dprintk("RPC: xs_tcp_data_recv started\n");
Chuck Levera246b012005-08-11 16:25:23 -04001029 do {
1030 /* Read in a new fragment marker if necessary */
1031 /* Can we ever really expect to get completely empty fragments? */
Chuck Levere136d092006-12-05 16:35:23 -05001032 if (transport->tcp_flags & TCP_RCV_COPY_FRAGHDR) {
Chuck Lever9903cd12005-08-11 16:25:26 -04001033 xs_tcp_read_fraghdr(xprt, &desc);
Chuck Levera246b012005-08-11 16:25:23 -04001034 continue;
1035 }
1036 /* Read in the xid if necessary */
Chuck Levere136d092006-12-05 16:35:23 -05001037 if (transport->tcp_flags & TCP_RCV_COPY_XID) {
Chuck Lever51971132006-12-05 16:35:19 -05001038 xs_tcp_read_xid(transport, &desc);
Chuck Levera246b012005-08-11 16:25:23 -04001039 continue;
1040 }
1041 /* Read in the request data */
Chuck Levere136d092006-12-05 16:35:23 -05001042 if (transport->tcp_flags & TCP_RCV_COPY_DATA) {
Chuck Lever9903cd12005-08-11 16:25:26 -04001043 xs_tcp_read_request(xprt, &desc);
Chuck Levera246b012005-08-11 16:25:23 -04001044 continue;
1045 }
1046 /* Skip over any trailing bytes on short reads */
Chuck Lever51971132006-12-05 16:35:19 -05001047 xs_tcp_read_discard(transport, &desc);
Chuck Levera246b012005-08-11 16:25:23 -04001048 } while (desc.count);
Chuck Lever46121cf2007-01-31 12:14:08 -05001049 dprintk("RPC: xs_tcp_data_recv done\n");
Chuck Levera246b012005-08-11 16:25:23 -04001050 return len - desc.count;
1051}
1052
Chuck Lever9903cd12005-08-11 16:25:26 -04001053/**
1054 * xs_tcp_data_ready - "data ready" callback for TCP sockets
1055 * @sk: socket with data to read
1056 * @bytes: how much data to read
1057 *
1058 */
1059static void xs_tcp_data_ready(struct sock *sk, int bytes)
Chuck Levera246b012005-08-11 16:25:23 -04001060{
1061 struct rpc_xprt *xprt;
1062 read_descriptor_t rd_desc;
1063
Chuck Lever46121cf2007-01-31 12:14:08 -05001064 dprintk("RPC: xs_tcp_data_ready...\n");
1065
Chuck Levera246b012005-08-11 16:25:23 -04001066 read_lock(&sk->sk_callback_lock);
Chuck Lever9903cd12005-08-11 16:25:26 -04001067 if (!(xprt = xprt_from_sock(sk)))
Chuck Levera246b012005-08-11 16:25:23 -04001068 goto out;
Chuck Levera246b012005-08-11 16:25:23 -04001069 if (xprt->shutdown)
1070 goto out;
1071
Chuck Lever9903cd12005-08-11 16:25:26 -04001072 /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
Chuck Levera246b012005-08-11 16:25:23 -04001073 rd_desc.arg.data = xprt;
1074 rd_desc.count = 65536;
Chuck Lever9903cd12005-08-11 16:25:26 -04001075 tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
Chuck Levera246b012005-08-11 16:25:23 -04001076out:
1077 read_unlock(&sk->sk_callback_lock);
1078}
1079
Chuck Lever9903cd12005-08-11 16:25:26 -04001080/**
1081 * xs_tcp_state_change - callback to handle TCP socket state changes
1082 * @sk: socket whose state has changed
1083 *
1084 */
1085static void xs_tcp_state_change(struct sock *sk)
Chuck Levera246b012005-08-11 16:25:23 -04001086{
Chuck Lever9903cd12005-08-11 16:25:26 -04001087 struct rpc_xprt *xprt;
Chuck Levera246b012005-08-11 16:25:23 -04001088
1089 read_lock(&sk->sk_callback_lock);
1090 if (!(xprt = xprt_from_sock(sk)))
1091 goto out;
Chuck Lever46121cf2007-01-31 12:14:08 -05001092 dprintk("RPC: xs_tcp_state_change client %p...\n", xprt);
1093 dprintk("RPC: state %x conn %d dead %d zapped %d\n",
1094 sk->sk_state, xprt_connected(xprt),
1095 sock_flag(sk, SOCK_DEAD),
1096 sock_flag(sk, SOCK_ZAPPED));
Chuck Levera246b012005-08-11 16:25:23 -04001097
1098 switch (sk->sk_state) {
1099 case TCP_ESTABLISHED:
Chuck Lever4a0f8c02005-08-11 16:25:32 -04001100 spin_lock_bh(&xprt->transport_lock);
Chuck Levera246b012005-08-11 16:25:23 -04001101 if (!xprt_test_and_set_connected(xprt)) {
Chuck Lever51971132006-12-05 16:35:19 -05001102 struct sock_xprt *transport = container_of(xprt,
1103 struct sock_xprt, xprt);
1104
Chuck Levera246b012005-08-11 16:25:23 -04001105 /* Reset TCP record info */
Chuck Lever51971132006-12-05 16:35:19 -05001106 transport->tcp_offset = 0;
1107 transport->tcp_reclen = 0;
1108 transport->tcp_copied = 0;
Chuck Levere136d092006-12-05 16:35:23 -05001109 transport->tcp_flags =
1110 TCP_RCV_COPY_FRAGHDR | TCP_RCV_COPY_XID;
Chuck Lever51971132006-12-05 16:35:19 -05001111
Chuck Lever03bf4b72005-08-25 16:25:55 -07001112 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
Chuck Lever44fbac22005-08-11 16:25:44 -04001113 xprt_wake_pending_tasks(xprt, 0);
Chuck Levera246b012005-08-11 16:25:23 -04001114 }
Chuck Lever4a0f8c02005-08-11 16:25:32 -04001115 spin_unlock_bh(&xprt->transport_lock);
Chuck Levera246b012005-08-11 16:25:23 -04001116 break;
1117 case TCP_SYN_SENT:
1118 case TCP_SYN_RECV:
1119 break;
Trond Myklebust632e3bd2006-01-03 09:55:55 +01001120 case TCP_CLOSE_WAIT:
1121 /* Try to schedule an autoclose RPC calls */
1122 set_bit(XPRT_CLOSE_WAIT, &xprt->state);
1123 if (test_and_set_bit(XPRT_LOCKED, &xprt->state) == 0)
Trond Myklebustc1384c92007-06-14 18:00:42 -04001124 queue_work(rpciod_workqueue, &xprt->task_cleanup);
Chuck Levera246b012005-08-11 16:25:23 -04001125 default:
1126 xprt_disconnect(xprt);
Chuck Levera246b012005-08-11 16:25:23 -04001127 }
1128 out:
1129 read_unlock(&sk->sk_callback_lock);
1130}
1131
Chuck Lever9903cd12005-08-11 16:25:26 -04001132/**
Chuck Leverc7b2cae2005-08-11 16:25:50 -04001133 * xs_udp_write_space - callback invoked when socket buffer space
1134 * becomes available
Chuck Lever9903cd12005-08-11 16:25:26 -04001135 * @sk: socket whose state has changed
1136 *
Chuck Levera246b012005-08-11 16:25:23 -04001137 * Called when more output buffer space is available for this socket.
1138 * We try not to wake our writers until they can make "significant"
Chuck Leverc7b2cae2005-08-11 16:25:50 -04001139 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
Chuck Levera246b012005-08-11 16:25:23 -04001140 * with a bunch of small requests.
1141 */
Chuck Leverc7b2cae2005-08-11 16:25:50 -04001142static void xs_udp_write_space(struct sock *sk)
Chuck Levera246b012005-08-11 16:25:23 -04001143{
Chuck Levera246b012005-08-11 16:25:23 -04001144 read_lock(&sk->sk_callback_lock);
Chuck Levera246b012005-08-11 16:25:23 -04001145
Chuck Leverc7b2cae2005-08-11 16:25:50 -04001146 /* from net/core/sock.c:sock_def_write_space */
1147 if (sock_writeable(sk)) {
1148 struct socket *sock;
1149 struct rpc_xprt *xprt;
1150
1151 if (unlikely(!(sock = sk->sk_socket)))
Chuck Levera246b012005-08-11 16:25:23 -04001152 goto out;
Chuck Leverc7b2cae2005-08-11 16:25:50 -04001153 if (unlikely(!(xprt = xprt_from_sock(sk))))
Chuck Levera246b012005-08-11 16:25:23 -04001154 goto out;
Chuck Leverc7b2cae2005-08-11 16:25:50 -04001155 if (unlikely(!test_and_clear_bit(SOCK_NOSPACE, &sock->flags)))
1156 goto out;
1157
1158 xprt_write_space(xprt);
Chuck Levera246b012005-08-11 16:25:23 -04001159 }
1160
Chuck Leverc7b2cae2005-08-11 16:25:50 -04001161 out:
1162 read_unlock(&sk->sk_callback_lock);
1163}
Chuck Levera246b012005-08-11 16:25:23 -04001164
Chuck Leverc7b2cae2005-08-11 16:25:50 -04001165/**
1166 * xs_tcp_write_space - callback invoked when socket buffer space
1167 * becomes available
1168 * @sk: socket whose state has changed
1169 *
1170 * Called when more output buffer space is available for this socket.
1171 * We try not to wake our writers until they can make "significant"
1172 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1173 * with a bunch of small requests.
1174 */
1175static void xs_tcp_write_space(struct sock *sk)
1176{
1177 read_lock(&sk->sk_callback_lock);
1178
1179 /* from net/core/stream.c:sk_stream_write_space */
1180 if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk)) {
1181 struct socket *sock;
1182 struct rpc_xprt *xprt;
1183
1184 if (unlikely(!(sock = sk->sk_socket)))
1185 goto out;
1186 if (unlikely(!(xprt = xprt_from_sock(sk))))
1187 goto out;
1188 if (unlikely(!test_and_clear_bit(SOCK_NOSPACE, &sock->flags)))
1189 goto out;
1190
1191 xprt_write_space(xprt);
1192 }
1193
1194 out:
Chuck Levera246b012005-08-11 16:25:23 -04001195 read_unlock(&sk->sk_callback_lock);
1196}
1197
Chuck Lever470056c2005-08-25 16:25:56 -07001198static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
Chuck Levera246b012005-08-11 16:25:23 -04001199{
Chuck Leveree0ac0c2006-12-05 16:35:15 -05001200 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1201 struct sock *sk = transport->inet;
Chuck Levera246b012005-08-11 16:25:23 -04001202
Chuck Lever7c6e0662006-12-05 16:35:30 -05001203 if (transport->rcvsize) {
Chuck Levera246b012005-08-11 16:25:23 -04001204 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
Chuck Lever7c6e0662006-12-05 16:35:30 -05001205 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
Chuck Levera246b012005-08-11 16:25:23 -04001206 }
Chuck Lever7c6e0662006-12-05 16:35:30 -05001207 if (transport->sndsize) {
Chuck Levera246b012005-08-11 16:25:23 -04001208 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
Chuck Lever7c6e0662006-12-05 16:35:30 -05001209 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
Chuck Levera246b012005-08-11 16:25:23 -04001210 sk->sk_write_space(sk);
1211 }
1212}
1213
Chuck Lever43118c22005-08-25 16:25:49 -07001214/**
Chuck Lever470056c2005-08-25 16:25:56 -07001215 * xs_udp_set_buffer_size - set send and receive limits
Chuck Lever43118c22005-08-25 16:25:49 -07001216 * @xprt: generic transport
Chuck Lever470056c2005-08-25 16:25:56 -07001217 * @sndsize: requested size of send buffer, in bytes
1218 * @rcvsize: requested size of receive buffer, in bytes
Chuck Lever43118c22005-08-25 16:25:49 -07001219 *
Chuck Lever470056c2005-08-25 16:25:56 -07001220 * Set socket send and receive buffer size limits.
Chuck Lever43118c22005-08-25 16:25:49 -07001221 */
Chuck Lever470056c2005-08-25 16:25:56 -07001222static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
Chuck Lever43118c22005-08-25 16:25:49 -07001223{
Chuck Lever7c6e0662006-12-05 16:35:30 -05001224 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1225
1226 transport->sndsize = 0;
Chuck Lever470056c2005-08-25 16:25:56 -07001227 if (sndsize)
Chuck Lever7c6e0662006-12-05 16:35:30 -05001228 transport->sndsize = sndsize + 1024;
1229 transport->rcvsize = 0;
Chuck Lever470056c2005-08-25 16:25:56 -07001230 if (rcvsize)
Chuck Lever7c6e0662006-12-05 16:35:30 -05001231 transport->rcvsize = rcvsize + 1024;
Chuck Lever470056c2005-08-25 16:25:56 -07001232
1233 xs_udp_do_set_buffer_size(xprt);
Chuck Lever43118c22005-08-25 16:25:49 -07001234}
1235
Chuck Lever46c0ee82005-08-25 16:25:52 -07001236/**
1237 * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1238 * @task: task that timed out
1239 *
1240 * Adjust the congestion window after a retransmit timeout has occurred.
1241 */
1242static void xs_udp_timer(struct rpc_task *task)
1243{
1244 xprt_adjust_cwnd(task, -ETIMEDOUT);
1245}
1246
Chuck Leverb85d8802006-05-25 01:40:49 -04001247static unsigned short xs_get_random_port(void)
1248{
1249 unsigned short range = xprt_max_resvport - xprt_min_resvport;
1250 unsigned short rand = (unsigned short) net_random() % range;
1251 return rand + xprt_min_resvport;
1252}
1253
Chuck Lever92200412006-01-03 09:55:51 +01001254/**
1255 * xs_set_port - reset the port number in the remote endpoint address
1256 * @xprt: generic transport
1257 * @port: new port number
1258 *
1259 */
1260static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1261{
Chuck Lever95392c52007-08-06 11:57:58 -04001262 struct sockaddr *addr = xs_addr(xprt);
Chuck Leverc4efcb12006-08-22 20:06:19 -04001263
Chuck Lever46121cf2007-01-31 12:14:08 -05001264 dprintk("RPC: setting port for xprt %p to %u\n", xprt, port);
Chuck Leverc4efcb12006-08-22 20:06:19 -04001265
Chuck Lever20612002007-08-06 11:57:23 -04001266 switch (addr->sa_family) {
1267 case AF_INET:
1268 ((struct sockaddr_in *)addr)->sin_port = htons(port);
1269 break;
1270 case AF_INET6:
1271 ((struct sockaddr_in6 *)addr)->sin6_port = htons(port);
1272 break;
1273 default:
1274 BUG();
1275 }
Chuck Lever92200412006-01-03 09:55:51 +01001276}
1277
Chuck Lever7dc753f2007-08-06 11:57:28 -04001278static int xs_bind4(struct sock_xprt *transport, struct socket *sock)
Chuck Levera246b012005-08-11 16:25:23 -04001279{
1280 struct sockaddr_in myaddr = {
1281 .sin_family = AF_INET,
1282 };
Frank van Maarseveend3bc9a12007-07-09 22:23:35 +02001283 struct sockaddr_in *sa;
Chuck Lever529b33c2005-08-25 16:25:54 -07001284 int err;
Chuck Leverc8475462006-12-05 16:35:26 -05001285 unsigned short port = transport->port;
Chuck Levera246b012005-08-11 16:25:23 -04001286
Frank van Maarseveend3bc9a12007-07-09 22:23:35 +02001287 if (!transport->xprt.resvport)
1288 port = 0;
1289 sa = (struct sockaddr_in *)&transport->addr;
1290 myaddr.sin_addr = sa->sin_addr;
Chuck Levera246b012005-08-11 16:25:23 -04001291 do {
1292 myaddr.sin_port = htons(port);
Sridhar Samudralae6242e92006-08-07 20:58:01 -07001293 err = kernel_bind(sock, (struct sockaddr *) &myaddr,
Chuck Levera246b012005-08-11 16:25:23 -04001294 sizeof(myaddr));
Frank van Maarseveend3bc9a12007-07-09 22:23:35 +02001295 if (!transport->xprt.resvport)
1296 break;
Chuck Levera246b012005-08-11 16:25:23 -04001297 if (err == 0) {
Chuck Leverc8475462006-12-05 16:35:26 -05001298 transport->port = port;
Frank van Maarseveend3bc9a12007-07-09 22:23:35 +02001299 break;
Chuck Levera246b012005-08-11 16:25:23 -04001300 }
Chuck Lever529b33c2005-08-25 16:25:54 -07001301 if (port <= xprt_min_resvport)
1302 port = xprt_max_resvport;
1303 else
1304 port--;
Chuck Leverc8475462006-12-05 16:35:26 -05001305 } while (err == -EADDRINUSE && port != transport->port);
Chuck Lever7dc753f2007-08-06 11:57:28 -04001306 dprintk("RPC: %s "NIPQUAD_FMT":%u: %s (%d)\n",
1307 __FUNCTION__, NIPQUAD(myaddr.sin_addr),
1308 port, err ? "failed" : "ok", err);
Chuck Levera246b012005-08-11 16:25:23 -04001309 return err;
1310}
1311
Chuck Lever90058d32007-08-06 11:57:33 -04001312static int xs_bind6(struct sock_xprt *transport, struct socket *sock)
1313{
1314 struct sockaddr_in6 myaddr = {
1315 .sin6_family = AF_INET6,
1316 };
1317 struct sockaddr_in6 *sa;
1318 int err;
1319 unsigned short port = transport->port;
1320
1321 if (!transport->xprt.resvport)
1322 port = 0;
1323 sa = (struct sockaddr_in6 *)&transport->addr;
1324 myaddr.sin6_addr = sa->sin6_addr;
1325 do {
1326 myaddr.sin6_port = htons(port);
1327 err = kernel_bind(sock, (struct sockaddr *) &myaddr,
1328 sizeof(myaddr));
1329 if (!transport->xprt.resvport)
1330 break;
1331 if (err == 0) {
1332 transport->port = port;
1333 break;
1334 }
1335 if (port <= xprt_min_resvport)
1336 port = xprt_max_resvport;
1337 else
1338 port--;
1339 } while (err == -EADDRINUSE && port != transport->port);
1340 dprintk("RPC: xs_bind6 "NIP6_FMT":%u: %s (%d)\n",
1341 NIP6(myaddr.sin6_addr), port, err ? "failed" : "ok", err);
Chuck Levera246b012005-08-11 16:25:23 -04001342 return err;
1343}
1344
Peter Zijlstraed075362006-12-06 20:35:24 -08001345#ifdef CONFIG_DEBUG_LOCK_ALLOC
1346static struct lock_class_key xs_key[2];
1347static struct lock_class_key xs_slock_key[2];
1348
Chuck Lever8945ee52007-08-06 11:58:04 -04001349static inline void xs_reclassify_socket4(struct socket *sock)
Peter Zijlstraed075362006-12-06 20:35:24 -08001350{
1351 struct sock *sk = sock->sk;
Chuck Lever8945ee52007-08-06 11:58:04 -04001352
John Heffner02b3d342007-09-12 10:42:12 +02001353 BUG_ON(sock_owned_by_user(sk));
Chuck Lever8945ee52007-08-06 11:58:04 -04001354 sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1355 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1356}
Peter Zijlstraed075362006-12-06 20:35:24 -08001357
Chuck Lever8945ee52007-08-06 11:58:04 -04001358static inline void xs_reclassify_socket6(struct socket *sock)
1359{
1360 struct sock *sk = sock->sk;
Peter Zijlstraed075362006-12-06 20:35:24 -08001361
Linus Torvaldsf4921af2007-10-15 10:46:05 -07001362 BUG_ON(sock_owned_by_user(sk));
Chuck Lever8945ee52007-08-06 11:58:04 -04001363 sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1364 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
Peter Zijlstraed075362006-12-06 20:35:24 -08001365}
1366#else
Chuck Lever8945ee52007-08-06 11:58:04 -04001367static inline void xs_reclassify_socket4(struct socket *sock)
1368{
1369}
1370
1371static inline void xs_reclassify_socket6(struct socket *sock)
Peter Zijlstraed075362006-12-06 20:35:24 -08001372{
1373}
1374#endif
1375
Chuck Lever16be2d22007-08-06 11:57:38 -04001376static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
Chuck Levera246b012005-08-11 16:25:23 -04001377{
Chuck Lever16be2d22007-08-06 11:57:38 -04001378 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
Chuck Leveredb267a2006-08-22 20:06:18 -04001379
Chuck Leveree0ac0c2006-12-05 16:35:15 -05001380 if (!transport->inet) {
Chuck Leverb0d93ad2005-08-11 16:25:53 -04001381 struct sock *sk = sock->sk;
1382
1383 write_lock_bh(&sk->sk_callback_lock);
1384
1385 sk->sk_user_data = xprt;
Chuck Lever314dfd72006-12-05 16:35:34 -05001386 transport->old_data_ready = sk->sk_data_ready;
1387 transport->old_state_change = sk->sk_state_change;
1388 transport->old_write_space = sk->sk_write_space;
Chuck Leverb0d93ad2005-08-11 16:25:53 -04001389 sk->sk_data_ready = xs_udp_data_ready;
1390 sk->sk_write_space = xs_udp_write_space;
1391 sk->sk_no_check = UDP_CSUM_NORCV;
Trond Myklebustb079fa7b2005-12-13 16:13:52 -05001392 sk->sk_allocation = GFP_ATOMIC;
Chuck Leverb0d93ad2005-08-11 16:25:53 -04001393
1394 xprt_set_connected(xprt);
1395
1396 /* Reset to new socket */
Chuck Leveree0ac0c2006-12-05 16:35:15 -05001397 transport->sock = sock;
1398 transport->inet = sk;
Chuck Leverb0d93ad2005-08-11 16:25:53 -04001399
1400 write_unlock_bh(&sk->sk_callback_lock);
1401 }
Chuck Lever470056c2005-08-25 16:25:56 -07001402 xs_udp_do_set_buffer_size(xprt);
Chuck Lever16be2d22007-08-06 11:57:38 -04001403}
1404
Chuck Levera246b012005-08-11 16:25:23 -04001405/**
Chuck Lever9c3d72d2007-08-06 11:57:43 -04001406 * xs_udp_connect_worker4 - set up a UDP socket
Chuck Levera246b012005-08-11 16:25:23 -04001407 * @work: RPC transport to connect
1408 *
1409 * Invoked by a work queue tasklet.
1410 */
Chuck Lever9c3d72d2007-08-06 11:57:43 -04001411static void xs_udp_connect_worker4(struct work_struct *work)
Chuck Levera246b012005-08-11 16:25:23 -04001412{
1413 struct sock_xprt *transport =
1414 container_of(work, struct sock_xprt, connect_worker.work);
1415 struct rpc_xprt *xprt = &transport->xprt;
1416 struct socket *sock = transport->sock;
1417 int err, status = -EIO;
1418
1419 if (xprt->shutdown || !xprt_bound(xprt))
Chuck Lever9903cd12005-08-11 16:25:26 -04001420 goto out;
1421
Chuck Levera246b012005-08-11 16:25:23 -04001422 /* Start by resetting any existing state */
1423 xs_close(xprt);
1424
Chuck Lever9903cd12005-08-11 16:25:26 -04001425 if ((err = sock_create_kern(PF_INET, SOCK_DGRAM, IPPROTO_UDP, &sock)) < 0) {
1426 dprintk("RPC: can't create UDP transport socket (%d).\n", -err);
Chuck Levera246b012005-08-11 16:25:23 -04001427 goto out;
1428 }
Chuck Lever8945ee52007-08-06 11:58:04 -04001429 xs_reclassify_socket4(sock);
Chuck Levera246b012005-08-11 16:25:23 -04001430
Chuck Lever7dc753f2007-08-06 11:57:28 -04001431 if (xs_bind4(transport, sock)) {
Chuck Lever9903cd12005-08-11 16:25:26 -04001432 sock_release(sock);
1433 goto out;
Chuck Levera246b012005-08-11 16:25:23 -04001434 }
Chuck Leverb0d93ad2005-08-11 16:25:53 -04001435
1436 dprintk("RPC: worker connecting xprt %p to address: %s\n",
1437 xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1438
Chuck Lever16be2d22007-08-06 11:57:38 -04001439 xs_udp_finish_connecting(xprt, sock);
Chuck Levera246b012005-08-11 16:25:23 -04001440 status = 0;
Chuck Leverb0d93ad2005-08-11 16:25:53 -04001441out:
1442 xprt_wake_pending_tasks(xprt, status);
1443 xprt_clear_connecting(xprt);
1444}
1445
Chuck Lever68e220b2007-08-06 11:57:48 -04001446/**
1447 * xs_udp_connect_worker6 - set up a UDP socket
1448 * @work: RPC transport to connect
1449 *
1450 * Invoked by a work queue tasklet.
1451 */
1452static void xs_udp_connect_worker6(struct work_struct *work)
1453{
1454 struct sock_xprt *transport =
1455 container_of(work, struct sock_xprt, connect_worker.work);
1456 struct rpc_xprt *xprt = &transport->xprt;
1457 struct socket *sock = transport->sock;
1458 int err, status = -EIO;
1459
1460 if (xprt->shutdown || !xprt_bound(xprt))
1461 goto out;
1462
1463 /* Start by resetting any existing state */
1464 xs_close(xprt);
1465
1466 if ((err = sock_create_kern(PF_INET6, SOCK_DGRAM, IPPROTO_UDP, &sock)) < 0) {
1467 dprintk("RPC: can't create UDP transport socket (%d).\n", -err);
1468 goto out;
1469 }
Chuck Lever8945ee52007-08-06 11:58:04 -04001470 xs_reclassify_socket6(sock);
Chuck Lever68e220b2007-08-06 11:57:48 -04001471
1472 if (xs_bind6(transport, sock) < 0) {
1473 sock_release(sock);
1474 goto out;
1475 }
1476
1477 dprintk("RPC: worker connecting xprt %p to address: %s\n",
1478 xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1479
1480 xs_udp_finish_connecting(xprt, sock);
Chuck Levera246b012005-08-11 16:25:23 -04001481 status = 0;
Chuck Leverb0d93ad2005-08-11 16:25:53 -04001482out:
1483 xprt_wake_pending_tasks(xprt, status);
1484 xprt_clear_connecting(xprt);
1485}
1486
Chuck Lever3167e122005-08-25 16:25:55 -07001487/*
1488 * We need to preserve the port number so the reply cache on the server can
1489 * find our cached RPC replies when we get around to reconnecting.
1490 */
1491static void xs_tcp_reuse_connection(struct rpc_xprt *xprt)
1492{
1493 int result;
Chuck Leveree0ac0c2006-12-05 16:35:15 -05001494 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
Chuck Lever3167e122005-08-25 16:25:55 -07001495 struct sockaddr any;
1496
Chuck Lever46121cf2007-01-31 12:14:08 -05001497 dprintk("RPC: disconnecting xprt %p to reuse port\n", xprt);
Chuck Lever3167e122005-08-25 16:25:55 -07001498
1499 /*
1500 * Disconnect the transport socket by doing a connect operation
1501 * with AF_UNSPEC. This should return immediately...
1502 */
1503 memset(&any, 0, sizeof(any));
1504 any.sa_family = AF_UNSPEC;
Chuck Leveree0ac0c2006-12-05 16:35:15 -05001505 result = kernel_connect(transport->sock, &any, sizeof(any), 0);
Chuck Lever3167e122005-08-25 16:25:55 -07001506 if (result)
Chuck Lever46121cf2007-01-31 12:14:08 -05001507 dprintk("RPC: AF_UNSPEC connect return code %d\n",
Chuck Lever3167e122005-08-25 16:25:55 -07001508 result);
1509}
1510
Chuck Lever16be2d22007-08-06 11:57:38 -04001511static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
Chuck Leverb0d93ad2005-08-11 16:25:53 -04001512{
Chuck Lever16be2d22007-08-06 11:57:38 -04001513 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
Chuck Leveredb267a2006-08-22 20:06:18 -04001514
Chuck Leveree0ac0c2006-12-05 16:35:15 -05001515 if (!transport->inet) {
Chuck Leverb0d93ad2005-08-11 16:25:53 -04001516 struct sock *sk = sock->sk;
1517
1518 write_lock_bh(&sk->sk_callback_lock);
1519
1520 sk->sk_user_data = xprt;
Chuck Lever314dfd72006-12-05 16:35:34 -05001521 transport->old_data_ready = sk->sk_data_ready;
1522 transport->old_state_change = sk->sk_state_change;
1523 transport->old_write_space = sk->sk_write_space;
Chuck Leverb0d93ad2005-08-11 16:25:53 -04001524 sk->sk_data_ready = xs_tcp_data_ready;
1525 sk->sk_state_change = xs_tcp_state_change;
1526 sk->sk_write_space = xs_tcp_write_space;
Trond Myklebustb079fa7b2005-12-13 16:13:52 -05001527 sk->sk_allocation = GFP_ATOMIC;
Chuck Lever3167e122005-08-25 16:25:55 -07001528
1529 /* socket options */
1530 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
1531 sock_reset_flag(sk, SOCK_LINGER);
1532 tcp_sk(sk)->linger2 = 0;
1533 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
Chuck Leverb0d93ad2005-08-11 16:25:53 -04001534
1535 xprt_clear_connected(xprt);
1536
1537 /* Reset to new socket */
Chuck Leveree0ac0c2006-12-05 16:35:15 -05001538 transport->sock = sock;
1539 transport->inet = sk;
Chuck Leverb0d93ad2005-08-11 16:25:53 -04001540
1541 write_unlock_bh(&sk->sk_callback_lock);
1542 }
1543
1544 /* Tell the socket layer to start connecting... */
Chuck Lever262ca072006-03-20 13:44:16 -05001545 xprt->stat.connect_count++;
1546 xprt->stat.connect_start = jiffies;
Chuck Lever95392c52007-08-06 11:57:58 -04001547 return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
Chuck Lever16be2d22007-08-06 11:57:38 -04001548}
1549
1550/**
Chuck Lever9c3d72d2007-08-06 11:57:43 -04001551 * xs_tcp_connect_worker4 - connect a TCP socket to a remote endpoint
Chuck Lever16be2d22007-08-06 11:57:38 -04001552 * @work: RPC transport to connect
1553 *
1554 * Invoked by a work queue tasklet.
1555 */
Chuck Lever9c3d72d2007-08-06 11:57:43 -04001556static void xs_tcp_connect_worker4(struct work_struct *work)
Chuck Lever16be2d22007-08-06 11:57:38 -04001557{
1558 struct sock_xprt *transport =
1559 container_of(work, struct sock_xprt, connect_worker.work);
1560 struct rpc_xprt *xprt = &transport->xprt;
1561 struct socket *sock = transport->sock;
1562 int err, status = -EIO;
1563
1564 if (xprt->shutdown || !xprt_bound(xprt))
1565 goto out;
1566
1567 if (!sock) {
1568 /* start from scratch */
1569 if ((err = sock_create_kern(PF_INET, SOCK_STREAM, IPPROTO_TCP, &sock)) < 0) {
1570 dprintk("RPC: can't create TCP transport socket (%d).\n", -err);
1571 goto out;
1572 }
Chuck Lever8945ee52007-08-06 11:58:04 -04001573 xs_reclassify_socket4(sock);
Chuck Lever16be2d22007-08-06 11:57:38 -04001574
1575 if (xs_bind4(transport, sock) < 0) {
1576 sock_release(sock);
1577 goto out;
1578 }
1579 } else
1580 /* "close" the socket, preserving the local port */
1581 xs_tcp_reuse_connection(xprt);
1582
1583 dprintk("RPC: worker connecting xprt %p to address: %s\n",
1584 xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1585
1586 status = xs_tcp_finish_connecting(xprt, sock);
Chuck Lever46121cf2007-01-31 12:14:08 -05001587 dprintk("RPC: %p connect status %d connected %d sock state %d\n",
1588 xprt, -status, xprt_connected(xprt),
1589 sock->sk->sk_state);
Chuck Levera246b012005-08-11 16:25:23 -04001590 if (status < 0) {
1591 switch (status) {
1592 case -EINPROGRESS:
1593 case -EALREADY:
1594 goto out_clear;
Chuck Lever3167e122005-08-25 16:25:55 -07001595 case -ECONNREFUSED:
1596 case -ECONNRESET:
1597 /* retry with existing socket, after a delay */
1598 break;
1599 default:
1600 /* get rid of existing socket, and retry */
1601 xs_close(xprt);
1602 break;
Chuck Levera246b012005-08-11 16:25:23 -04001603 }
1604 }
1605out:
Chuck Lever44fbac22005-08-11 16:25:44 -04001606 xprt_wake_pending_tasks(xprt, status);
Chuck Levera246b012005-08-11 16:25:23 -04001607out_clear:
Chuck Lever2226feb2005-08-11 16:25:38 -04001608 xprt_clear_connecting(xprt);
Chuck Levera246b012005-08-11 16:25:23 -04001609}
1610
Chuck Lever9903cd12005-08-11 16:25:26 -04001611/**
Chuck Lever68e220b2007-08-06 11:57:48 -04001612 * xs_tcp_connect_worker6 - connect a TCP socket to a remote endpoint
1613 * @work: RPC transport to connect
1614 *
1615 * Invoked by a work queue tasklet.
1616 */
1617static void xs_tcp_connect_worker6(struct work_struct *work)
1618{
1619 struct sock_xprt *transport =
1620 container_of(work, struct sock_xprt, connect_worker.work);
1621 struct rpc_xprt *xprt = &transport->xprt;
1622 struct socket *sock = transport->sock;
1623 int err, status = -EIO;
1624
1625 if (xprt->shutdown || !xprt_bound(xprt))
1626 goto out;
1627
1628 if (!sock) {
1629 /* start from scratch */
1630 if ((err = sock_create_kern(PF_INET6, SOCK_STREAM, IPPROTO_TCP, &sock)) < 0) {
1631 dprintk("RPC: can't create TCP transport socket (%d).\n", -err);
1632 goto out;
1633 }
Chuck Lever8945ee52007-08-06 11:58:04 -04001634 xs_reclassify_socket6(sock);
Chuck Lever68e220b2007-08-06 11:57:48 -04001635
1636 if (xs_bind6(transport, sock) < 0) {
1637 sock_release(sock);
1638 goto out;
1639 }
1640 } else
1641 /* "close" the socket, preserving the local port */
1642 xs_tcp_reuse_connection(xprt);
1643
1644 dprintk("RPC: worker connecting xprt %p to address: %s\n",
1645 xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1646
1647 status = xs_tcp_finish_connecting(xprt, sock);
1648 dprintk("RPC: %p connect status %d connected %d sock state %d\n",
1649 xprt, -status, xprt_connected(xprt), sock->sk->sk_state);
1650 if (status < 0) {
1651 switch (status) {
1652 case -EINPROGRESS:
1653 case -EALREADY:
1654 goto out_clear;
1655 case -ECONNREFUSED:
1656 case -ECONNRESET:
1657 /* retry with existing socket, after a delay */
1658 break;
1659 default:
1660 /* get rid of existing socket, and retry */
1661 xs_close(xprt);
1662 break;
1663 }
1664 }
1665out:
1666 xprt_wake_pending_tasks(xprt, status);
1667out_clear:
1668 xprt_clear_connecting(xprt);
1669}
1670
1671/**
Chuck Lever9903cd12005-08-11 16:25:26 -04001672 * xs_connect - connect a socket to a remote endpoint
1673 * @task: address of RPC task that manages state of connect request
1674 *
1675 * TCP: If the remote end dropped the connection, delay reconnecting.
Chuck Lever03bf4b72005-08-25 16:25:55 -07001676 *
1677 * UDP socket connects are synchronous, but we use a work queue anyway
1678 * to guarantee that even unprivileged user processes can set up a
1679 * socket on a privileged port.
1680 *
1681 * If a UDP socket connect fails, the delay behavior here prevents
1682 * retry floods (hard mounts).
Chuck Lever9903cd12005-08-11 16:25:26 -04001683 */
1684static void xs_connect(struct rpc_task *task)
Chuck Levera246b012005-08-11 16:25:23 -04001685{
1686 struct rpc_xprt *xprt = task->tk_xprt;
Chuck Leveree0ac0c2006-12-05 16:35:15 -05001687 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
Chuck Levera246b012005-08-11 16:25:23 -04001688
Chuck Leverb0d93ad2005-08-11 16:25:53 -04001689 if (xprt_test_and_set_connecting(xprt))
1690 return;
1691
Chuck Leveree0ac0c2006-12-05 16:35:15 -05001692 if (transport->sock != NULL) {
Chuck Lever46121cf2007-01-31 12:14:08 -05001693 dprintk("RPC: xs_connect delayed xprt %p for %lu "
1694 "seconds\n",
Chuck Lever03bf4b72005-08-25 16:25:55 -07001695 xprt, xprt->reestablish_timeout / HZ);
Trond Myklebustc1384c92007-06-14 18:00:42 -04001696 queue_delayed_work(rpciod_workqueue,
1697 &transport->connect_worker,
1698 xprt->reestablish_timeout);
Chuck Lever03bf4b72005-08-25 16:25:55 -07001699 xprt->reestablish_timeout <<= 1;
1700 if (xprt->reestablish_timeout > XS_TCP_MAX_REEST_TO)
1701 xprt->reestablish_timeout = XS_TCP_MAX_REEST_TO;
Chuck Leverb0d93ad2005-08-11 16:25:53 -04001702 } else {
Chuck Lever46121cf2007-01-31 12:14:08 -05001703 dprintk("RPC: xs_connect scheduled xprt %p\n", xprt);
Trond Myklebustc1384c92007-06-14 18:00:42 -04001704 queue_delayed_work(rpciod_workqueue,
1705 &transport->connect_worker, 0);
Chuck Levera246b012005-08-11 16:25:23 -04001706 }
1707}
1708
Chuck Lever262ca072006-03-20 13:44:16 -05001709/**
1710 * xs_udp_print_stats - display UDP socket-specifc stats
1711 * @xprt: rpc_xprt struct containing statistics
1712 * @seq: output file
1713 *
1714 */
1715static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
1716{
Chuck Leverc8475462006-12-05 16:35:26 -05001717 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1718
Chuck Lever262ca072006-03-20 13:44:16 -05001719 seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %Lu %Lu\n",
Chuck Leverc8475462006-12-05 16:35:26 -05001720 transport->port,
Chuck Lever262ca072006-03-20 13:44:16 -05001721 xprt->stat.bind_count,
1722 xprt->stat.sends,
1723 xprt->stat.recvs,
1724 xprt->stat.bad_xids,
1725 xprt->stat.req_u,
1726 xprt->stat.bklog_u);
1727}
1728
1729/**
1730 * xs_tcp_print_stats - display TCP socket-specifc stats
1731 * @xprt: rpc_xprt struct containing statistics
1732 * @seq: output file
1733 *
1734 */
1735static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
1736{
Chuck Leverc8475462006-12-05 16:35:26 -05001737 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
Chuck Lever262ca072006-03-20 13:44:16 -05001738 long idle_time = 0;
1739
1740 if (xprt_connected(xprt))
1741 idle_time = (long)(jiffies - xprt->last_used) / HZ;
1742
1743 seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu %Lu %Lu\n",
Chuck Leverc8475462006-12-05 16:35:26 -05001744 transport->port,
Chuck Lever262ca072006-03-20 13:44:16 -05001745 xprt->stat.bind_count,
1746 xprt->stat.connect_count,
1747 xprt->stat.connect_time,
1748 idle_time,
1749 xprt->stat.sends,
1750 xprt->stat.recvs,
1751 xprt->stat.bad_xids,
1752 xprt->stat.req_u,
1753 xprt->stat.bklog_u);
1754}
1755
Chuck Lever262965f2005-08-11 16:25:56 -04001756static struct rpc_xprt_ops xs_udp_ops = {
Chuck Lever43118c22005-08-25 16:25:49 -07001757 .set_buffer_size = xs_udp_set_buffer_size,
Chuck Lever12a80462005-08-25 16:25:51 -07001758 .reserve_xprt = xprt_reserve_xprt_cong,
Chuck Lever49e9a892005-08-25 16:25:51 -07001759 .release_xprt = xprt_release_xprt_cong,
Chuck Lever45160d62007-07-01 12:13:17 -04001760 .rpcbind = rpcb_getport_async,
Chuck Lever92200412006-01-03 09:55:51 +01001761 .set_port = xs_set_port,
Chuck Lever9903cd12005-08-11 16:25:26 -04001762 .connect = xs_connect,
Chuck Lever02107142006-01-03 09:55:49 +01001763 .buf_alloc = rpc_malloc,
1764 .buf_free = rpc_free,
Chuck Lever262965f2005-08-11 16:25:56 -04001765 .send_request = xs_udp_send_request,
Chuck Leverfe3aca22005-08-25 16:25:50 -07001766 .set_retrans_timeout = xprt_set_retrans_timeout_rtt,
Chuck Lever46c0ee82005-08-25 16:25:52 -07001767 .timer = xs_udp_timer,
Chuck Levera58dd392005-08-25 16:25:53 -07001768 .release_request = xprt_release_rqst_cong,
Chuck Lever262965f2005-08-11 16:25:56 -04001769 .close = xs_close,
1770 .destroy = xs_destroy,
Chuck Lever262ca072006-03-20 13:44:16 -05001771 .print_stats = xs_udp_print_stats,
Chuck Lever262965f2005-08-11 16:25:56 -04001772};
1773
1774static struct rpc_xprt_ops xs_tcp_ops = {
Chuck Lever12a80462005-08-25 16:25:51 -07001775 .reserve_xprt = xprt_reserve_xprt,
Trond Myklebuste0ab53d2006-07-27 17:22:50 -04001776 .release_xprt = xs_tcp_release_xprt,
Chuck Lever45160d62007-07-01 12:13:17 -04001777 .rpcbind = rpcb_getport_async,
Chuck Lever92200412006-01-03 09:55:51 +01001778 .set_port = xs_set_port,
Chuck Lever262965f2005-08-11 16:25:56 -04001779 .connect = xs_connect,
Chuck Lever02107142006-01-03 09:55:49 +01001780 .buf_alloc = rpc_malloc,
1781 .buf_free = rpc_free,
Chuck Lever262965f2005-08-11 16:25:56 -04001782 .send_request = xs_tcp_send_request,
Chuck Leverfe3aca22005-08-25 16:25:50 -07001783 .set_retrans_timeout = xprt_set_retrans_timeout_def,
Chuck Lever9903cd12005-08-11 16:25:26 -04001784 .close = xs_close,
1785 .destroy = xs_destroy,
Chuck Lever262ca072006-03-20 13:44:16 -05001786 .print_stats = xs_tcp_print_stats,
Chuck Levera246b012005-08-11 16:25:23 -04001787};
1788
\"Talpey, Thomas\3c341b0b2007-09-10 13:47:07 -04001789static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
\"Talpey, Thomas\bc255712007-09-10 13:46:39 -04001790 unsigned int slot_table_size)
Chuck Leverc8541ec2006-10-17 14:44:27 -04001791{
1792 struct rpc_xprt *xprt;
Chuck Leverffc2e512006-12-05 16:35:11 -05001793 struct sock_xprt *new;
Chuck Leverc8541ec2006-10-17 14:44:27 -04001794
Frank van Maarseveen96802a02007-07-08 13:08:54 +02001795 if (args->addrlen > sizeof(xprt->addr)) {
Chuck Lever46121cf2007-01-31 12:14:08 -05001796 dprintk("RPC: xs_setup_xprt: address too large\n");
Chuck Leverc8541ec2006-10-17 14:44:27 -04001797 return ERR_PTR(-EBADF);
1798 }
1799
Chuck Leverffc2e512006-12-05 16:35:11 -05001800 new = kzalloc(sizeof(*new), GFP_KERNEL);
1801 if (new == NULL) {
Chuck Lever46121cf2007-01-31 12:14:08 -05001802 dprintk("RPC: xs_setup_xprt: couldn't allocate "
1803 "rpc_xprt\n");
Chuck Leverc8541ec2006-10-17 14:44:27 -04001804 return ERR_PTR(-ENOMEM);
1805 }
Chuck Leverffc2e512006-12-05 16:35:11 -05001806 xprt = &new->xprt;
Chuck Leverc8541ec2006-10-17 14:44:27 -04001807
1808 xprt->max_reqs = slot_table_size;
1809 xprt->slot = kcalloc(xprt->max_reqs, sizeof(struct rpc_rqst), GFP_KERNEL);
1810 if (xprt->slot == NULL) {
1811 kfree(xprt);
Chuck Lever46121cf2007-01-31 12:14:08 -05001812 dprintk("RPC: xs_setup_xprt: couldn't allocate slot "
1813 "table\n");
Chuck Leverc8541ec2006-10-17 14:44:27 -04001814 return ERR_PTR(-ENOMEM);
1815 }
1816
Frank van Maarseveen96802a02007-07-08 13:08:54 +02001817 memcpy(&xprt->addr, args->dstaddr, args->addrlen);
1818 xprt->addrlen = args->addrlen;
Frank van Maarseveend3bc9a12007-07-09 22:23:35 +02001819 if (args->srcaddr)
1820 memcpy(&new->addr, args->srcaddr, args->addrlen);
Chuck Leverc8475462006-12-05 16:35:26 -05001821 new->port = xs_get_random_port();
Chuck Leverc8541ec2006-10-17 14:44:27 -04001822
1823 return xprt;
1824}
1825
Chuck Lever9903cd12005-08-11 16:25:26 -04001826/**
1827 * xs_setup_udp - Set up transport to use a UDP socket
Frank van Maarseveen96802a02007-07-08 13:08:54 +02001828 * @args: rpc transport creation arguments
Chuck Lever9903cd12005-08-11 16:25:26 -04001829 *
1830 */
Adrian Bunk483066d2007-10-24 18:24:02 +02001831static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
Chuck Levera246b012005-08-11 16:25:23 -04001832{
Chuck Lever8f9d5b12007-08-06 11:57:53 -04001833 struct sockaddr *addr = args->dstaddr;
Chuck Leverc8541ec2006-10-17 14:44:27 -04001834 struct rpc_xprt *xprt;
Chuck Leverc8475462006-12-05 16:35:26 -05001835 struct sock_xprt *transport;
Chuck Levera246b012005-08-11 16:25:23 -04001836
Frank van Maarseveen96802a02007-07-08 13:08:54 +02001837 xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries);
Chuck Leverc8541ec2006-10-17 14:44:27 -04001838 if (IS_ERR(xprt))
1839 return xprt;
Chuck Leverc8475462006-12-05 16:35:26 -05001840 transport = container_of(xprt, struct sock_xprt, xprt);
Chuck Levera246b012005-08-11 16:25:23 -04001841
Chuck Leverec739ef2006-08-22 20:06:15 -04001842 xprt->prot = IPPROTO_UDP;
Chuck Lever808012f2005-08-25 16:25:49 -07001843 xprt->tsh_size = 0;
Chuck Levera246b012005-08-11 16:25:23 -04001844 /* XXX: header size can vary due to auth type, IPv6, etc. */
1845 xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
1846
Chuck Lever03bf4b72005-08-25 16:25:55 -07001847 xprt->bind_timeout = XS_BIND_TO;
1848 xprt->connect_timeout = XS_UDP_CONN_TO;
1849 xprt->reestablish_timeout = XS_UDP_REEST_TO;
1850 xprt->idle_timeout = XS_IDLE_DISC_TO;
Chuck Levera246b012005-08-11 16:25:23 -04001851
Chuck Lever262965f2005-08-11 16:25:56 -04001852 xprt->ops = &xs_udp_ops;
Chuck Levera246b012005-08-11 16:25:23 -04001853
Frank van Maarseveen96802a02007-07-08 13:08:54 +02001854 if (args->timeout)
1855 xprt->timeout = *args->timeout;
Chuck Levera246b012005-08-11 16:25:23 -04001856 else
Chuck Lever9903cd12005-08-11 16:25:26 -04001857 xprt_set_timeout(&xprt->timeout, 5, 5 * HZ);
Chuck Levera246b012005-08-11 16:25:23 -04001858
Chuck Lever8f9d5b12007-08-06 11:57:53 -04001859 switch (addr->sa_family) {
1860 case AF_INET:
1861 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
1862 xprt_set_bound(xprt);
1863
1864 INIT_DELAYED_WORK(&transport->connect_worker,
1865 xs_udp_connect_worker4);
1866 xs_format_ipv4_peer_addresses(xprt);
1867 break;
1868 case AF_INET6:
1869 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
1870 xprt_set_bound(xprt);
1871
1872 INIT_DELAYED_WORK(&transport->connect_worker,
1873 xs_udp_connect_worker6);
1874 xs_format_ipv6_peer_addresses(xprt);
1875 break;
1876 default:
1877 kfree(xprt);
1878 return ERR_PTR(-EAFNOSUPPORT);
1879 }
1880
Chuck Lever46121cf2007-01-31 12:14:08 -05001881 dprintk("RPC: set up transport to address %s\n",
Chuck Lever7559c7a2006-12-05 16:35:37 -05001882 xprt->address_strings[RPC_DISPLAY_ALL]);
Chuck Leveredb267a2006-08-22 20:06:18 -04001883
\"Talpey, Thomas\bc255712007-09-10 13:46:39 -04001884 if (try_module_get(THIS_MODULE))
1885 return xprt;
1886
1887 kfree(xprt->slot);
1888 kfree(xprt);
1889 return ERR_PTR(-EINVAL);
Chuck Levera246b012005-08-11 16:25:23 -04001890}
1891
Chuck Lever9903cd12005-08-11 16:25:26 -04001892/**
1893 * xs_setup_tcp - Set up transport to use a TCP socket
Frank van Maarseveen96802a02007-07-08 13:08:54 +02001894 * @args: rpc transport creation arguments
Chuck Lever9903cd12005-08-11 16:25:26 -04001895 *
1896 */
Adrian Bunk483066d2007-10-24 18:24:02 +02001897static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
Chuck Levera246b012005-08-11 16:25:23 -04001898{
Chuck Lever8f9d5b12007-08-06 11:57:53 -04001899 struct sockaddr *addr = args->dstaddr;
Chuck Leverc8541ec2006-10-17 14:44:27 -04001900 struct rpc_xprt *xprt;
Chuck Leverc8475462006-12-05 16:35:26 -05001901 struct sock_xprt *transport;
Chuck Levera246b012005-08-11 16:25:23 -04001902
Frank van Maarseveen96802a02007-07-08 13:08:54 +02001903 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries);
Chuck Leverc8541ec2006-10-17 14:44:27 -04001904 if (IS_ERR(xprt))
1905 return xprt;
Chuck Leverc8475462006-12-05 16:35:26 -05001906 transport = container_of(xprt, struct sock_xprt, xprt);
Chuck Levera246b012005-08-11 16:25:23 -04001907
Chuck Leverec739ef2006-08-22 20:06:15 -04001908 xprt->prot = IPPROTO_TCP;
Chuck Lever808012f2005-08-25 16:25:49 -07001909 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
Chuck Lever808012f2005-08-25 16:25:49 -07001910 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
Chuck Levera246b012005-08-11 16:25:23 -04001911
Chuck Lever03bf4b72005-08-25 16:25:55 -07001912 xprt->bind_timeout = XS_BIND_TO;
1913 xprt->connect_timeout = XS_TCP_CONN_TO;
1914 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1915 xprt->idle_timeout = XS_IDLE_DISC_TO;
Chuck Levera246b012005-08-11 16:25:23 -04001916
Chuck Lever262965f2005-08-11 16:25:56 -04001917 xprt->ops = &xs_tcp_ops;
Chuck Levera246b012005-08-11 16:25:23 -04001918
Frank van Maarseveen96802a02007-07-08 13:08:54 +02001919 if (args->timeout)
1920 xprt->timeout = *args->timeout;
Chuck Levera246b012005-08-11 16:25:23 -04001921 else
Chuck Lever9903cd12005-08-11 16:25:26 -04001922 xprt_set_timeout(&xprt->timeout, 2, 60 * HZ);
Chuck Levera246b012005-08-11 16:25:23 -04001923
Chuck Lever8f9d5b12007-08-06 11:57:53 -04001924 switch (addr->sa_family) {
1925 case AF_INET:
1926 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
1927 xprt_set_bound(xprt);
1928
1929 INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_connect_worker4);
1930 xs_format_ipv4_peer_addresses(xprt);
1931 break;
1932 case AF_INET6:
1933 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
1934 xprt_set_bound(xprt);
1935
1936 INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_connect_worker6);
1937 xs_format_ipv6_peer_addresses(xprt);
1938 break;
1939 default:
1940 kfree(xprt);
1941 return ERR_PTR(-EAFNOSUPPORT);
1942 }
1943
Chuck Lever46121cf2007-01-31 12:14:08 -05001944 dprintk("RPC: set up transport to address %s\n",
Chuck Lever7559c7a2006-12-05 16:35:37 -05001945 xprt->address_strings[RPC_DISPLAY_ALL]);
Chuck Leveredb267a2006-08-22 20:06:18 -04001946
\"Talpey, Thomas\bc255712007-09-10 13:46:39 -04001947 if (try_module_get(THIS_MODULE))
1948 return xprt;
1949
1950 kfree(xprt->slot);
1951 kfree(xprt);
1952 return ERR_PTR(-EINVAL);
Chuck Levera246b012005-08-11 16:25:23 -04001953}
Chuck Lever282b32e2006-12-05 16:35:51 -05001954
\"Talpey, Thomas\bc255712007-09-10 13:46:39 -04001955static struct xprt_class xs_udp_transport = {
1956 .list = LIST_HEAD_INIT(xs_udp_transport.list),
1957 .name = "udp",
1958 .owner = THIS_MODULE,
\"Talpey, Thomas\4fa016e2007-09-10 13:47:57 -04001959 .ident = IPPROTO_UDP,
\"Talpey, Thomas\bc255712007-09-10 13:46:39 -04001960 .setup = xs_setup_udp,
1961};
1962
1963static struct xprt_class xs_tcp_transport = {
1964 .list = LIST_HEAD_INIT(xs_tcp_transport.list),
1965 .name = "tcp",
1966 .owner = THIS_MODULE,
\"Talpey, Thomas\4fa016e2007-09-10 13:47:57 -04001967 .ident = IPPROTO_TCP,
\"Talpey, Thomas\bc255712007-09-10 13:46:39 -04001968 .setup = xs_setup_tcp,
1969};
1970
Chuck Lever282b32e2006-12-05 16:35:51 -05001971/**
\"Talpey, Thomas\bc255712007-09-10 13:46:39 -04001972 * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
Chuck Lever282b32e2006-12-05 16:35:51 -05001973 *
1974 */
1975int init_socket_xprt(void)
1976{
Chuck Leverfbf76682006-12-05 16:35:54 -05001977#ifdef RPC_DEBUG
Eric W. Biederman2b1bec52007-02-14 00:33:24 -08001978 if (!sunrpc_table_header)
Eric W. Biederman0b4d4142007-02-14 00:34:09 -08001979 sunrpc_table_header = register_sysctl_table(sunrpc_table);
Chuck Leverfbf76682006-12-05 16:35:54 -05001980#endif
1981
\"Talpey, Thomas\bc255712007-09-10 13:46:39 -04001982 xprt_register_transport(&xs_udp_transport);
1983 xprt_register_transport(&xs_tcp_transport);
1984
Chuck Lever282b32e2006-12-05 16:35:51 -05001985 return 0;
1986}
1987
1988/**
\"Talpey, Thomas\bc255712007-09-10 13:46:39 -04001989 * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
Chuck Lever282b32e2006-12-05 16:35:51 -05001990 *
1991 */
1992void cleanup_socket_xprt(void)
1993{
Chuck Leverfbf76682006-12-05 16:35:54 -05001994#ifdef RPC_DEBUG
1995 if (sunrpc_table_header) {
1996 unregister_sysctl_table(sunrpc_table_header);
1997 sunrpc_table_header = NULL;
1998 }
1999#endif
\"Talpey, Thomas\bc255712007-09-10 13:46:39 -04002000
2001 xprt_unregister_transport(&xs_udp_transport);
2002 xprt_unregister_transport(&xs_tcp_transport);
Chuck Lever282b32e2006-12-05 16:35:51 -05002003}