blob: 7ca65c7005ea5d2445ddddd617017ab3e6539f29 [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 *
Alan Cox113aa832008-10-13 19:01:08 -07006 * TCP callback races fixes (C) 1998 Red Hat
7 * TCP send fixes (C) 1998 Red Hat
Chuck Levera246b012005-08-11 16:25:23 -04008 * 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>
Rahul Iyer4cfc7e62009-09-10 17:32:28 +030035#include <linux/sunrpc/svcsock.h>
\"Talpey, Thomas\49c36fc2007-09-10 13:47:31 -040036#include <linux/sunrpc/xprtsock.h>
Chuck Levera246b012005-08-11 16:25:23 -040037#include <linux/file.h>
Ricardo Labiaga44b98ef2009-04-01 09:23:02 -040038#ifdef CONFIG_NFS_V4_1
39#include <linux/sunrpc/bc_xprt.h>
40#endif
Chuck Levera246b012005-08-11 16:25:23 -040041
42#include <net/sock.h>
43#include <net/checksum.h>
44#include <net/udp.h>
45#include <net/tcp.h>
46
Rahul Iyer4cfc7e62009-09-10 17:32:28 +030047#include "sunrpc.h"
Chuck Lever9903cd12005-08-11 16:25:26 -040048/*
Chuck Leverc556b752005-11-01 12:24:48 -050049 * xprtsock tunables
50 */
51unsigned int xprt_udp_slot_table_entries = RPC_DEF_SLOT_TABLE;
52unsigned int xprt_tcp_slot_table_entries = RPC_DEF_SLOT_TABLE;
53
54unsigned int xprt_min_resvport = RPC_DEF_MIN_RESVPORT;
55unsigned int xprt_max_resvport = RPC_DEF_MAX_RESVPORT;
56
Trond Myklebust7d1e8252009-03-11 14:38:03 -040057#define XS_TCP_LINGER_TO (15U * HZ)
Trond Myklebust25fe6142009-03-11 14:38:03 -040058static unsigned int xs_tcp_fin_timeout __read_mostly = XS_TCP_LINGER_TO;
Trond Myklebust7d1e8252009-03-11 14:38:03 -040059
Chuck Leverc556b752005-11-01 12:24:48 -050060/*
Chuck Leverfbf76682006-12-05 16:35:54 -050061 * We can register our own files under /proc/sys/sunrpc by
62 * calling register_sysctl_table() again. The files in that
63 * directory become the union of all files registered there.
64 *
65 * We simply need to make sure that we don't collide with
66 * someone else's file names!
67 */
68
69#ifdef RPC_DEBUG
70
71static unsigned int min_slot_table_size = RPC_MIN_SLOT_TABLE;
72static unsigned int max_slot_table_size = RPC_MAX_SLOT_TABLE;
73static unsigned int xprt_min_resvport_limit = RPC_MIN_RESVPORT;
74static unsigned int xprt_max_resvport_limit = RPC_MAX_RESVPORT;
75
76static struct ctl_table_header *sunrpc_table_header;
77
78/*
79 * FIXME: changing the UDP slot table size should also resize the UDP
80 * socket buffers for existing UDP transports
81 */
82static ctl_table xs_tunables_table[] = {
83 {
Chuck Leverfbf76682006-12-05 16:35:54 -050084 .procname = "udp_slot_table_entries",
85 .data = &xprt_udp_slot_table_entries,
86 .maxlen = sizeof(unsigned int),
87 .mode = 0644,
Eric W. Biederman6d456112009-11-16 03:11:48 -080088 .proc_handler = proc_dointvec_minmax,
Chuck Leverfbf76682006-12-05 16:35:54 -050089 .extra1 = &min_slot_table_size,
90 .extra2 = &max_slot_table_size
91 },
92 {
Chuck Leverfbf76682006-12-05 16:35:54 -050093 .procname = "tcp_slot_table_entries",
94 .data = &xprt_tcp_slot_table_entries,
95 .maxlen = sizeof(unsigned int),
96 .mode = 0644,
Eric W. Biederman6d456112009-11-16 03:11:48 -080097 .proc_handler = proc_dointvec_minmax,
Chuck Leverfbf76682006-12-05 16:35:54 -050098 .extra1 = &min_slot_table_size,
99 .extra2 = &max_slot_table_size
100 },
101 {
Chuck Leverfbf76682006-12-05 16:35:54 -0500102 .procname = "min_resvport",
103 .data = &xprt_min_resvport,
104 .maxlen = sizeof(unsigned int),
105 .mode = 0644,
Eric W. Biederman6d456112009-11-16 03:11:48 -0800106 .proc_handler = proc_dointvec_minmax,
Chuck Leverfbf76682006-12-05 16:35:54 -0500107 .extra1 = &xprt_min_resvport_limit,
108 .extra2 = &xprt_max_resvport_limit
109 },
110 {
Chuck Leverfbf76682006-12-05 16:35:54 -0500111 .procname = "max_resvport",
112 .data = &xprt_max_resvport,
113 .maxlen = sizeof(unsigned int),
114 .mode = 0644,
Eric W. Biederman6d456112009-11-16 03:11:48 -0800115 .proc_handler = proc_dointvec_minmax,
Chuck Leverfbf76682006-12-05 16:35:54 -0500116 .extra1 = &xprt_min_resvport_limit,
117 .extra2 = &xprt_max_resvport_limit
118 },
119 {
Trond Myklebust25fe6142009-03-11 14:38:03 -0400120 .procname = "tcp_fin_timeout",
121 .data = &xs_tcp_fin_timeout,
122 .maxlen = sizeof(xs_tcp_fin_timeout),
123 .mode = 0644,
Eric W. Biederman6d456112009-11-16 03:11:48 -0800124 .proc_handler = proc_dointvec_jiffies,
Trond Myklebust25fe6142009-03-11 14:38:03 -0400125 },
Eric W. Biedermanf8572d82009-11-05 13:32:03 -0800126 { },
Chuck Leverfbf76682006-12-05 16:35:54 -0500127};
128
129static ctl_table sunrpc_table[] = {
130 {
Chuck Leverfbf76682006-12-05 16:35:54 -0500131 .procname = "sunrpc",
132 .mode = 0555,
133 .child = xs_tunables_table
134 },
Eric W. Biedermanf8572d82009-11-05 13:32:03 -0800135 { },
Chuck Leverfbf76682006-12-05 16:35:54 -0500136};
137
138#endif
139
140/*
Chuck Lever03bf4b72005-08-25 16:25:55 -0700141 * Wait duration for a reply from the RPC portmapper.
142 */
143#define XS_BIND_TO (60U * HZ)
144
145/*
146 * Delay if a UDP socket connect error occurs. This is most likely some
147 * kind of resource problem on the local host.
148 */
149#define XS_UDP_REEST_TO (2U * HZ)
150
151/*
152 * The reestablish timeout allows clients to delay for a bit before attempting
153 * to reconnect to a server that just dropped our connection.
154 *
155 * We implement an exponential backoff when trying to reestablish a TCP
156 * transport connection with the server. Some servers like to drop a TCP
157 * connection when they are overworked, so we start with a short timeout and
158 * increase over time if the server is down or not responding.
159 */
160#define XS_TCP_INIT_REEST_TO (3U * HZ)
161#define XS_TCP_MAX_REEST_TO (5U * 60 * HZ)
162
163/*
164 * TCP idle timeout; client drops the transport socket if it is idle
165 * for this long. Note that we also timeout UDP sockets to prevent
166 * holding port numbers when there is no RPC traffic.
167 */
168#define XS_IDLE_DISC_TO (5U * 60 * HZ)
169
Chuck Levera246b012005-08-11 16:25:23 -0400170#ifdef RPC_DEBUG
171# undef RPC_DEBUG_DATA
Chuck Lever9903cd12005-08-11 16:25:26 -0400172# define RPCDBG_FACILITY RPCDBG_TRANS
Chuck Levera246b012005-08-11 16:25:23 -0400173#endif
174
Chuck Levera246b012005-08-11 16:25:23 -0400175#ifdef RPC_DEBUG_DATA
Chuck Lever9903cd12005-08-11 16:25:26 -0400176static void xs_pktdump(char *msg, u32 *packet, unsigned int count)
Chuck Levera246b012005-08-11 16:25:23 -0400177{
Chuck Lever9903cd12005-08-11 16:25:26 -0400178 u8 *buf = (u8 *) packet;
179 int j;
Chuck Levera246b012005-08-11 16:25:23 -0400180
Chuck Lever46121cf2007-01-31 12:14:08 -0500181 dprintk("RPC: %s\n", msg);
Chuck Levera246b012005-08-11 16:25:23 -0400182 for (j = 0; j < count && j < 128; j += 4) {
183 if (!(j & 31)) {
184 if (j)
185 dprintk("\n");
186 dprintk("0x%04x ", j);
187 }
188 dprintk("%02x%02x%02x%02x ",
189 buf[j], buf[j+1], buf[j+2], buf[j+3]);
190 }
191 dprintk("\n");
192}
193#else
Chuck Lever9903cd12005-08-11 16:25:26 -0400194static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count)
Chuck Levera246b012005-08-11 16:25:23 -0400195{
196 /* NOP */
197}
198#endif
199
Chuck Leverffc2e512006-12-05 16:35:11 -0500200struct sock_xprt {
201 struct rpc_xprt xprt;
Chuck Leveree0ac0c2006-12-05 16:35:15 -0500202
203 /*
204 * Network layer
205 */
206 struct socket * sock;
207 struct sock * inet;
Chuck Lever51971132006-12-05 16:35:19 -0500208
209 /*
210 * State of TCP reply receive
211 */
212 __be32 tcp_fraghdr,
Trond Myklebustb76ce562010-06-16 13:57:32 -0400213 tcp_xid,
214 tcp_calldir;
Chuck Lever51971132006-12-05 16:35:19 -0500215
216 u32 tcp_offset,
217 tcp_reclen;
218
219 unsigned long tcp_copied,
220 tcp_flags;
Chuck Leverc8475462006-12-05 16:35:26 -0500221
222 /*
223 * Connection of transports
224 */
Trond Myklebust34161db2006-12-07 15:48:15 -0500225 struct delayed_work connect_worker;
Chuck Leverfbfffbd2009-08-09 15:09:46 -0400226 struct sockaddr_storage srcaddr;
227 unsigned short srcport;
Chuck Lever7c6e0662006-12-05 16:35:30 -0500228
229 /*
230 * UDP socket buffer size parameters
231 */
232 size_t rcvsize,
233 sndsize;
Chuck Lever314dfd72006-12-05 16:35:34 -0500234
235 /*
236 * Saved socket callback addresses
237 */
238 void (*old_data_ready)(struct sock *, int);
239 void (*old_state_change)(struct sock *);
240 void (*old_write_space)(struct sock *);
Trond Myklebust2a9e1cf2008-10-28 15:21:39 -0400241 void (*old_error_report)(struct sock *);
Chuck Leverffc2e512006-12-05 16:35:11 -0500242};
243
Chuck Levere136d092006-12-05 16:35:23 -0500244/*
245 * TCP receive state flags
246 */
247#define TCP_RCV_LAST_FRAG (1UL << 0)
248#define TCP_RCV_COPY_FRAGHDR (1UL << 1)
249#define TCP_RCV_COPY_XID (1UL << 2)
250#define TCP_RCV_COPY_DATA (1UL << 3)
Ricardo Labiagaf4a2e412009-04-01 09:22:54 -0400251#define TCP_RCV_READ_CALLDIR (1UL << 4)
252#define TCP_RCV_COPY_CALLDIR (1UL << 5)
Ricardo Labiaga18dca022009-04-01 09:22:53 -0400253
254/*
255 * TCP RPC flags
256 */
Ricardo Labiagaf4a2e412009-04-01 09:22:54 -0400257#define TCP_RPC_REPLY (1UL << 6)
Chuck Levere136d092006-12-05 16:35:23 -0500258
Chuck Lever95392c52007-08-06 11:57:58 -0400259static inline struct sockaddr *xs_addr(struct rpc_xprt *xprt)
Chuck Leveredb267a2006-08-22 20:06:18 -0400260{
Chuck Lever95392c52007-08-06 11:57:58 -0400261 return (struct sockaddr *) &xprt->addr;
262}
263
264static inline struct sockaddr_in *xs_addr_in(struct rpc_xprt *xprt)
265{
266 return (struct sockaddr_in *) &xprt->addr;
267}
268
269static inline struct sockaddr_in6 *xs_addr_in6(struct rpc_xprt *xprt)
270{
271 return (struct sockaddr_in6 *) &xprt->addr;
272}
273
Chuck Leverc877b842009-08-09 15:09:36 -0400274static void xs_format_common_peer_addresses(struct rpc_xprt *xprt)
Chuck Leveredb267a2006-08-22 20:06:18 -0400275{
Chuck Leverc877b842009-08-09 15:09:36 -0400276 struct sockaddr *sap = xs_addr(xprt);
Chuck Lever9dc3b092009-08-09 15:09:46 -0400277 struct sockaddr_in6 *sin6;
278 struct sockaddr_in *sin;
Chuck Leverc877b842009-08-09 15:09:36 -0400279 char buf[128];
Chuck Leveredb267a2006-08-22 20:06:18 -0400280
Chuck Leverc877b842009-08-09 15:09:36 -0400281 (void)rpc_ntop(sap, buf, sizeof(buf));
282 xprt->address_strings[RPC_DISPLAY_ADDR] = kstrdup(buf, GFP_KERNEL);
283
Chuck Lever9dc3b092009-08-09 15:09:46 -0400284 switch (sap->sa_family) {
285 case AF_INET:
286 sin = xs_addr_in(xprt);
Joe Perchesfc0b57912010-03-08 12:15:28 -0800287 snprintf(buf, sizeof(buf), "%08x", ntohl(sin->sin_addr.s_addr));
Chuck Lever9dc3b092009-08-09 15:09:46 -0400288 break;
289 case AF_INET6:
290 sin6 = xs_addr_in6(xprt);
Joe Perchesfc0b57912010-03-08 12:15:28 -0800291 snprintf(buf, sizeof(buf), "%pi6", &sin6->sin6_addr);
Chuck Lever9dc3b092009-08-09 15:09:46 -0400292 break;
293 default:
294 BUG();
Chuck Leveredb267a2006-08-22 20:06:18 -0400295 }
Chuck Lever9dc3b092009-08-09 15:09:46 -0400296 xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = kstrdup(buf, GFP_KERNEL);
Chuck Leveredb267a2006-08-22 20:06:18 -0400297}
298
Chuck Lever9dc3b092009-08-09 15:09:46 -0400299static void xs_format_common_peer_ports(struct rpc_xprt *xprt)
Chuck Lever4b6473f2007-08-06 11:57:12 -0400300{
Chuck Lever9dc3b092009-08-09 15:09:46 -0400301 struct sockaddr *sap = xs_addr(xprt);
302 char buf[128];
Chuck Lever4b6473f2007-08-06 11:57:12 -0400303
Joe Perches81160e662010-03-08 12:15:59 -0800304 snprintf(buf, sizeof(buf), "%u", rpc_get_port(sap));
Chuck Leverc877b842009-08-09 15:09:36 -0400305 xprt->address_strings[RPC_DISPLAY_PORT] = kstrdup(buf, GFP_KERNEL);
Chuck Lever4b6473f2007-08-06 11:57:12 -0400306
Joe Perches81160e662010-03-08 12:15:59 -0800307 snprintf(buf, sizeof(buf), "%4hx", rpc_get_port(sap));
Chuck Leverc877b842009-08-09 15:09:36 -0400308 xprt->address_strings[RPC_DISPLAY_HEX_PORT] = kstrdup(buf, GFP_KERNEL);
309}
Chuck Lever4b6473f2007-08-06 11:57:12 -0400310
Chuck Lever9dc3b092009-08-09 15:09:46 -0400311static void xs_format_peer_addresses(struct rpc_xprt *xprt,
312 const char *protocol,
313 const char *netid)
Chuck Leveredb267a2006-08-22 20:06:18 -0400314{
Chuck Leverb454ae92008-01-07 18:34:48 -0500315 xprt->address_strings[RPC_DISPLAY_PROTO] = protocol;
Chuck Leverb454ae92008-01-07 18:34:48 -0500316 xprt->address_strings[RPC_DISPLAY_NETID] = netid;
Chuck Leverc877b842009-08-09 15:09:36 -0400317 xs_format_common_peer_addresses(xprt);
Chuck Lever9dc3b092009-08-09 15:09:46 -0400318 xs_format_common_peer_ports(xprt);
Chuck Lever4b6473f2007-08-06 11:57:12 -0400319}
320
Chuck Lever9dc3b092009-08-09 15:09:46 -0400321static void xs_update_peer_port(struct rpc_xprt *xprt)
Chuck Lever4b6473f2007-08-06 11:57:12 -0400322{
Chuck Lever9dc3b092009-08-09 15:09:46 -0400323 kfree(xprt->address_strings[RPC_DISPLAY_HEX_PORT]);
324 kfree(xprt->address_strings[RPC_DISPLAY_PORT]);
Chuck Lever4b6473f2007-08-06 11:57:12 -0400325
Chuck Lever9dc3b092009-08-09 15:09:46 -0400326 xs_format_common_peer_ports(xprt);
Chuck Leveredb267a2006-08-22 20:06:18 -0400327}
328
329static void xs_free_peer_addresses(struct rpc_xprt *xprt)
330{
Chuck Lever33e01dc2008-01-14 12:32:20 -0500331 unsigned int i;
332
333 for (i = 0; i < RPC_DISPLAY_MAX; i++)
334 switch (i) {
335 case RPC_DISPLAY_PROTO:
336 case RPC_DISPLAY_NETID:
337 continue;
338 default:
339 kfree(xprt->address_strings[i]);
340 }
Chuck Leveredb267a2006-08-22 20:06:18 -0400341}
342
Chuck Leverb4b5cc82005-08-11 16:25:29 -0400343#define XS_SENDMSG_FLAGS (MSG_DONTWAIT | MSG_NOSIGNAL)
344
Trond Myklebust24c56842006-10-17 15:06:22 -0400345static 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 -0400346{
Chuck Leverb4b5cc82005-08-11 16:25:29 -0400347 struct msghdr msg = {
348 .msg_name = addr,
349 .msg_namelen = addrlen,
Trond Myklebust24c56842006-10-17 15:06:22 -0400350 .msg_flags = XS_SENDMSG_FLAGS | (more ? MSG_MORE : 0),
351 };
352 struct kvec iov = {
353 .iov_base = vec->iov_base + base,
354 .iov_len = vec->iov_len - base,
Chuck Leverb4b5cc82005-08-11 16:25:29 -0400355 };
356
Trond Myklebust24c56842006-10-17 15:06:22 -0400357 if (iov.iov_len != 0)
Chuck Leverb4b5cc82005-08-11 16:25:29 -0400358 return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
359 return kernel_sendmsg(sock, &msg, NULL, 0, 0);
360}
361
Trond Myklebust24c56842006-10-17 15:06:22 -0400362static int xs_send_pagedata(struct socket *sock, struct xdr_buf *xdr, unsigned int base, int more)
Chuck Leverb4b5cc82005-08-11 16:25:29 -0400363{
Trond Myklebust24c56842006-10-17 15:06:22 -0400364 struct page **ppage;
365 unsigned int remainder;
366 int err, sent = 0;
Chuck Leverb4b5cc82005-08-11 16:25:29 -0400367
Trond Myklebust24c56842006-10-17 15:06:22 -0400368 remainder = xdr->page_len - base;
369 base += xdr->page_base;
370 ppage = xdr->pages + (base >> PAGE_SHIFT);
371 base &= ~PAGE_MASK;
372 for(;;) {
373 unsigned int len = min_t(unsigned int, PAGE_SIZE - base, remainder);
374 int flags = XS_SENDMSG_FLAGS;
375
376 remainder -= len;
377 if (remainder != 0 || more)
378 flags |= MSG_MORE;
379 err = sock->ops->sendpage(sock, *ppage, base, len, flags);
380 if (remainder == 0 || err != len)
381 break;
382 sent += err;
383 ppage++;
384 base = 0;
385 }
386 if (sent == 0)
387 return err;
388 if (err > 0)
389 sent += err;
390 return sent;
Chuck Leverb4b5cc82005-08-11 16:25:29 -0400391}
392
Chuck Lever9903cd12005-08-11 16:25:26 -0400393/**
394 * xs_sendpages - write pages directly to a socket
395 * @sock: socket to send on
396 * @addr: UDP only -- address of destination
397 * @addrlen: UDP only -- length of destination address
398 * @xdr: buffer containing this request
399 * @base: starting position in the buffer
400 *
Chuck Levera246b012005-08-11 16:25:23 -0400401 */
Trond Myklebust24c56842006-10-17 15:06:22 -0400402static 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 -0400403{
Trond Myklebust24c56842006-10-17 15:06:22 -0400404 unsigned int remainder = xdr->len - base;
405 int err, sent = 0;
Chuck Levera246b012005-08-11 16:25:23 -0400406
Chuck Lever262965f2005-08-11 16:25:56 -0400407 if (unlikely(!sock))
Trond Myklebustfba91af2009-03-11 14:06:41 -0400408 return -ENOTSOCK;
Chuck Lever262965f2005-08-11 16:25:56 -0400409
410 clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags);
Trond Myklebust24c56842006-10-17 15:06:22 -0400411 if (base != 0) {
412 addr = NULL;
413 addrlen = 0;
414 }
Chuck Lever262965f2005-08-11 16:25:56 -0400415
Trond Myklebust24c56842006-10-17 15:06:22 -0400416 if (base < xdr->head[0].iov_len || addr != NULL) {
417 unsigned int len = xdr->head[0].iov_len - base;
418 remainder -= len;
419 err = xs_send_kvec(sock, addr, addrlen, &xdr->head[0], base, remainder != 0);
420 if (remainder == 0 || err != len)
Chuck Levera246b012005-08-11 16:25:23 -0400421 goto out;
Trond Myklebust24c56842006-10-17 15:06:22 -0400422 sent += err;
Chuck Levera246b012005-08-11 16:25:23 -0400423 base = 0;
424 } else
Trond Myklebust24c56842006-10-17 15:06:22 -0400425 base -= xdr->head[0].iov_len;
Chuck Levera246b012005-08-11 16:25:23 -0400426
Trond Myklebust24c56842006-10-17 15:06:22 -0400427 if (base < xdr->page_len) {
428 unsigned int len = xdr->page_len - base;
429 remainder -= len;
430 err = xs_send_pagedata(sock, xdr, base, remainder != 0);
431 if (remainder == 0 || err != len)
Chuck Levera246b012005-08-11 16:25:23 -0400432 goto out;
Trond Myklebust24c56842006-10-17 15:06:22 -0400433 sent += err;
Chuck Levera246b012005-08-11 16:25:23 -0400434 base = 0;
Trond Myklebust24c56842006-10-17 15:06:22 -0400435 } else
436 base -= xdr->page_len;
437
438 if (base >= xdr->tail[0].iov_len)
439 return sent;
440 err = xs_send_kvec(sock, NULL, 0, &xdr->tail[0], base, 0);
Chuck Levera246b012005-08-11 16:25:23 -0400441out:
Trond Myklebust24c56842006-10-17 15:06:22 -0400442 if (sent == 0)
443 return err;
444 if (err > 0)
445 sent += err;
446 return sent;
Chuck Levera246b012005-08-11 16:25:23 -0400447}
448
Trond Myklebustb6ddf642008-04-17 18:52:19 -0400449static void xs_nospace_callback(struct rpc_task *task)
450{
451 struct sock_xprt *transport = container_of(task->tk_rqstp->rq_xprt, struct sock_xprt, xprt);
452
453 transport->inet->sk_write_pending--;
454 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
455}
456
Chuck Lever9903cd12005-08-11 16:25:26 -0400457/**
Chuck Lever262965f2005-08-11 16:25:56 -0400458 * xs_nospace - place task on wait queue if transmit was incomplete
459 * @task: task to put to sleep
Chuck Lever9903cd12005-08-11 16:25:26 -0400460 *
Chuck Levera246b012005-08-11 16:25:23 -0400461 */
Trond Myklebust5e3771c2009-03-11 14:38:01 -0400462static int xs_nospace(struct rpc_task *task)
Chuck Levera246b012005-08-11 16:25:23 -0400463{
Chuck Lever262965f2005-08-11 16:25:56 -0400464 struct rpc_rqst *req = task->tk_rqstp;
465 struct rpc_xprt *xprt = req->rq_xprt;
Chuck Leveree0ac0c2006-12-05 16:35:15 -0500466 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
Trond Myklebust5e3771c2009-03-11 14:38:01 -0400467 int ret = 0;
Chuck Levera246b012005-08-11 16:25:23 -0400468
Chuck Lever46121cf2007-01-31 12:14:08 -0500469 dprintk("RPC: %5u xmit incomplete (%u left of %u)\n",
Chuck Lever262965f2005-08-11 16:25:56 -0400470 task->tk_pid, req->rq_slen - req->rq_bytes_sent,
471 req->rq_slen);
472
Trond Myklebustb6ddf642008-04-17 18:52:19 -0400473 /* Protect against races with write_space */
474 spin_lock_bh(&xprt->transport_lock);
Chuck Lever262965f2005-08-11 16:25:56 -0400475
Trond Myklebustb6ddf642008-04-17 18:52:19 -0400476 /* Don't race with disconnect */
477 if (xprt_connected(xprt)) {
478 if (test_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags)) {
Trond Myklebust5e3771c2009-03-11 14:38:01 -0400479 ret = -EAGAIN;
Trond Myklebustb6ddf642008-04-17 18:52:19 -0400480 /*
481 * Notify TCP that we're limited by the application
482 * window size
483 */
484 set_bit(SOCK_NOSPACE, &transport->sock->flags);
485 transport->inet->sk_write_pending++;
486 /* ...and wait for more buffer space */
487 xprt_wait_for_buffer_space(task, xs_nospace_callback);
488 }
489 } else {
490 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
Trond Myklebust5e3771c2009-03-11 14:38:01 -0400491 ret = -ENOTCONN;
Trond Myklebustb6ddf642008-04-17 18:52:19 -0400492 }
Chuck Lever262965f2005-08-11 16:25:56 -0400493
Trond Myklebustb6ddf642008-04-17 18:52:19 -0400494 spin_unlock_bh(&xprt->transport_lock);
Trond Myklebust5e3771c2009-03-11 14:38:01 -0400495 return ret;
Chuck Lever262965f2005-08-11 16:25:56 -0400496}
497
498/**
499 * xs_udp_send_request - write an RPC request to a UDP socket
500 * @task: address of RPC task that manages the state of an RPC request
501 *
502 * Return values:
503 * 0: The request has been sent
504 * EAGAIN: The socket was blocked, please call again later to
505 * complete the request
506 * ENOTCONN: Caller needs to invoke connect logic then call again
507 * other: Some other error occured, the request was not sent
508 */
509static int xs_udp_send_request(struct rpc_task *task)
510{
511 struct rpc_rqst *req = task->tk_rqstp;
512 struct rpc_xprt *xprt = req->rq_xprt;
Chuck Leveree0ac0c2006-12-05 16:35:15 -0500513 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
Chuck Lever262965f2005-08-11 16:25:56 -0400514 struct xdr_buf *xdr = &req->rq_snd_buf;
515 int status;
Chuck Levera246b012005-08-11 16:25:23 -0400516
Chuck Lever9903cd12005-08-11 16:25:26 -0400517 xs_pktdump("packet data:",
Chuck Levera246b012005-08-11 16:25:23 -0400518 req->rq_svec->iov_base,
519 req->rq_svec->iov_len);
520
Trond Myklebust01d37c42009-03-11 14:09:39 -0400521 if (!xprt_bound(xprt))
522 return -ENOTCONN;
Chuck Leveree0ac0c2006-12-05 16:35:15 -0500523 status = xs_sendpages(transport->sock,
Chuck Lever95392c52007-08-06 11:57:58 -0400524 xs_addr(xprt),
Chuck Leveree0ac0c2006-12-05 16:35:15 -0500525 xprt->addrlen, xdr,
526 req->rq_bytes_sent);
Chuck Levera246b012005-08-11 16:25:23 -0400527
Chuck Lever46121cf2007-01-31 12:14:08 -0500528 dprintk("RPC: xs_udp_send_request(%u) = %d\n",
Chuck Lever262965f2005-08-11 16:25:56 -0400529 xdr->len - req->rq_bytes_sent, status);
Chuck Levera246b012005-08-11 16:25:23 -0400530
Trond Myklebust21997002007-10-01 11:43:37 -0400531 if (status >= 0) {
Trond Myklebustd60dbb22010-05-13 12:51:49 -0400532 req->rq_xmit_bytes_sent += status;
Trond Myklebust21997002007-10-01 11:43:37 -0400533 if (status >= req->rq_slen)
534 return 0;
535 /* Still some bytes left; set up for a retry later. */
Chuck Lever262965f2005-08-11 16:25:56 -0400536 status = -EAGAIN;
Trond Myklebust21997002007-10-01 11:43:37 -0400537 }
Chuck Levera246b012005-08-11 16:25:23 -0400538
Chuck Lever262965f2005-08-11 16:25:56 -0400539 switch (status) {
Trond Myklebustfba91af2009-03-11 14:06:41 -0400540 case -ENOTSOCK:
541 status = -ENOTCONN;
542 /* Should we call xs_close() here? */
543 break;
Trond Myklebustb6ddf642008-04-17 18:52:19 -0400544 case -EAGAIN:
Trond Myklebust5e3771c2009-03-11 14:38:01 -0400545 status = xs_nospace(task);
Trond Myklebustb6ddf642008-04-17 18:52:19 -0400546 break;
Trond Myklebustc8485e42009-03-11 14:37:59 -0400547 default:
548 dprintk("RPC: sendmsg returned unrecognized error %d\n",
549 -status);
Chuck Lever262965f2005-08-11 16:25:56 -0400550 case -ENETUNREACH:
551 case -EPIPE:
Chuck Levera246b012005-08-11 16:25:23 -0400552 case -ECONNREFUSED:
553 /* When the server has died, an ICMP port unreachable message
Chuck Lever9903cd12005-08-11 16:25:26 -0400554 * prompts ECONNREFUSED. */
Trond Myklebustb6ddf642008-04-17 18:52:19 -0400555 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
Chuck Levera246b012005-08-11 16:25:23 -0400556 }
Bian Naimeng5fe46e92010-03-08 14:49:01 +0800557
Chuck Lever262965f2005-08-11 16:25:56 -0400558 return status;
Chuck Levera246b012005-08-11 16:25:23 -0400559}
560
Trond Myklebuste06799f2007-11-05 15:44:12 -0500561/**
562 * xs_tcp_shutdown - gracefully shut down a TCP socket
563 * @xprt: transport
564 *
565 * Initiates a graceful shutdown of the TCP socket by calling the
566 * equivalent of shutdown(SHUT_WR);
567 */
568static void xs_tcp_shutdown(struct rpc_xprt *xprt)
569{
570 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
571 struct socket *sock = transport->sock;
572
573 if (sock != NULL)
574 kernel_sock_shutdown(sock, SHUT_WR);
575}
576
Chuck Lever808012f2005-08-25 16:25:49 -0700577static inline void xs_encode_tcp_record_marker(struct xdr_buf *buf)
578{
579 u32 reclen = buf->len - sizeof(rpc_fraghdr);
580 rpc_fraghdr *base = buf->head[0].iov_base;
581 *base = htonl(RPC_LAST_STREAM_FRAGMENT | reclen);
582}
583
Chuck Lever9903cd12005-08-11 16:25:26 -0400584/**
Chuck Lever262965f2005-08-11 16:25:56 -0400585 * xs_tcp_send_request - write an RPC request to a TCP socket
Chuck Lever9903cd12005-08-11 16:25:26 -0400586 * @task: address of RPC task that manages the state of an RPC request
587 *
588 * Return values:
Chuck Lever262965f2005-08-11 16:25:56 -0400589 * 0: The request has been sent
590 * EAGAIN: The socket was blocked, please call again later to
591 * complete the request
592 * ENOTCONN: Caller needs to invoke connect logic then call again
593 * other: Some other error occured, the request was not sent
Chuck Lever9903cd12005-08-11 16:25:26 -0400594 *
595 * XXX: In the case of soft timeouts, should we eventually give up
Chuck Lever262965f2005-08-11 16:25:56 -0400596 * if sendmsg is not able to make progress?
Chuck Lever9903cd12005-08-11 16:25:26 -0400597 */
Chuck Lever262965f2005-08-11 16:25:56 -0400598static int xs_tcp_send_request(struct rpc_task *task)
Chuck Levera246b012005-08-11 16:25:23 -0400599{
600 struct rpc_rqst *req = task->tk_rqstp;
601 struct rpc_xprt *xprt = req->rq_xprt;
Chuck Leveree0ac0c2006-12-05 16:35:15 -0500602 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
Chuck Lever262965f2005-08-11 16:25:56 -0400603 struct xdr_buf *xdr = &req->rq_snd_buf;
Chuck Leverb595bb12007-08-06 11:56:42 -0400604 int status;
Chuck Levera246b012005-08-11 16:25:23 -0400605
Chuck Lever808012f2005-08-25 16:25:49 -0700606 xs_encode_tcp_record_marker(&req->rq_snd_buf);
Chuck Levera246b012005-08-11 16:25:23 -0400607
Chuck Lever262965f2005-08-11 16:25:56 -0400608 xs_pktdump("packet data:",
609 req->rq_svec->iov_base,
610 req->rq_svec->iov_len);
Chuck Levera246b012005-08-11 16:25:23 -0400611
612 /* Continue transmitting the packet/record. We must be careful
613 * to cope with writespace callbacks arriving _after_ we have
Chuck Lever262965f2005-08-11 16:25:56 -0400614 * called sendmsg(). */
Chuck Levera246b012005-08-11 16:25:23 -0400615 while (1) {
Chuck Leveree0ac0c2006-12-05 16:35:15 -0500616 status = xs_sendpages(transport->sock,
617 NULL, 0, xdr, req->rq_bytes_sent);
Chuck Levera246b012005-08-11 16:25:23 -0400618
Chuck Lever46121cf2007-01-31 12:14:08 -0500619 dprintk("RPC: xs_tcp_send_request(%u) = %d\n",
Chuck Lever262965f2005-08-11 16:25:56 -0400620 xdr->len - req->rq_bytes_sent, status);
621
622 if (unlikely(status < 0))
Chuck Levera246b012005-08-11 16:25:23 -0400623 break;
624
Chuck Lever262965f2005-08-11 16:25:56 -0400625 /* If we've sent the entire packet, immediately
626 * reset the count of bytes sent. */
627 req->rq_bytes_sent += status;
Trond Myklebustd60dbb22010-05-13 12:51:49 -0400628 req->rq_xmit_bytes_sent += status;
Chuck Lever262965f2005-08-11 16:25:56 -0400629 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
630 req->rq_bytes_sent = 0;
631 return 0;
Chuck Levera246b012005-08-11 16:25:23 -0400632 }
633
Trond Myklebust06b4b682008-04-16 16:51:38 -0400634 if (status != 0)
635 continue;
Chuck Levera246b012005-08-11 16:25:23 -0400636 status = -EAGAIN;
Trond Myklebust06b4b682008-04-16 16:51:38 -0400637 break;
Chuck Levera246b012005-08-11 16:25:23 -0400638 }
639
Chuck Lever262965f2005-08-11 16:25:56 -0400640 switch (status) {
Trond Myklebustfba91af2009-03-11 14:06:41 -0400641 case -ENOTSOCK:
642 status = -ENOTCONN;
643 /* Should we call xs_close() here? */
644 break;
Chuck Lever262965f2005-08-11 16:25:56 -0400645 case -EAGAIN:
Trond Myklebust5e3771c2009-03-11 14:38:01 -0400646 status = xs_nospace(task);
Chuck Lever262965f2005-08-11 16:25:56 -0400647 break;
648 default:
Chuck Lever46121cf2007-01-31 12:14:08 -0500649 dprintk("RPC: sendmsg returned unrecognized error %d\n",
Chuck Lever262965f2005-08-11 16:25:56 -0400650 -status);
Chuck Lever262965f2005-08-11 16:25:56 -0400651 case -ECONNRESET:
Trond Myklebust55420c22009-03-11 15:29:24 -0400652 case -EPIPE:
Trond Myklebuste06799f2007-11-05 15:44:12 -0500653 xs_tcp_shutdown(xprt);
Chuck Lever262965f2005-08-11 16:25:56 -0400654 case -ECONNREFUSED:
655 case -ENOTCONN:
Chuck Lever262965f2005-08-11 16:25:56 -0400656 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
Chuck Levera246b012005-08-11 16:25:23 -0400657 }
Bian Naimeng5fe46e92010-03-08 14:49:01 +0800658
Chuck Levera246b012005-08-11 16:25:23 -0400659 return status;
660}
661
Chuck Lever9903cd12005-08-11 16:25:26 -0400662/**
Trond Myklebuste0ab53d2006-07-27 17:22:50 -0400663 * xs_tcp_release_xprt - clean up after a tcp transmission
664 * @xprt: transport
665 * @task: rpc task
666 *
667 * This cleans up if an error causes us to abort the transmission of a request.
668 * In this case, the socket may need to be reset in order to avoid confusing
669 * the server.
670 */
671static void xs_tcp_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
672{
673 struct rpc_rqst *req;
674
675 if (task != xprt->snd_task)
676 return;
677 if (task == NULL)
678 goto out_release;
679 req = task->tk_rqstp;
680 if (req->rq_bytes_sent == 0)
681 goto out_release;
682 if (req->rq_bytes_sent == req->rq_snd_buf.len)
683 goto out_release;
684 set_bit(XPRT_CLOSE_WAIT, &task->tk_xprt->state);
685out_release:
686 xprt_release_xprt(xprt, task);
687}
688
Trond Myklebust2a9e1cf2008-10-28 15:21:39 -0400689static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk)
690{
691 transport->old_data_ready = sk->sk_data_ready;
692 transport->old_state_change = sk->sk_state_change;
693 transport->old_write_space = sk->sk_write_space;
694 transport->old_error_report = sk->sk_error_report;
695}
696
697static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk)
698{
699 sk->sk_data_ready = transport->old_data_ready;
700 sk->sk_state_change = transport->old_state_change;
701 sk->sk_write_space = transport->old_write_space;
702 sk->sk_error_report = transport->old_error_report;
703}
704
Chuck Leverfe315e72009-03-11 14:10:21 -0400705static void xs_reset_transport(struct sock_xprt *transport)
Chuck Levera246b012005-08-11 16:25:23 -0400706{
Chuck Leveree0ac0c2006-12-05 16:35:15 -0500707 struct socket *sock = transport->sock;
708 struct sock *sk = transport->inet;
Chuck Levera246b012005-08-11 16:25:23 -0400709
Chuck Leverfe315e72009-03-11 14:10:21 -0400710 if (sk == NULL)
711 return;
Chuck Lever9903cd12005-08-11 16:25:26 -0400712
Chuck Levera246b012005-08-11 16:25:23 -0400713 write_lock_bh(&sk->sk_callback_lock);
Chuck Leveree0ac0c2006-12-05 16:35:15 -0500714 transport->inet = NULL;
715 transport->sock = NULL;
Chuck Levera246b012005-08-11 16:25:23 -0400716
Chuck Lever9903cd12005-08-11 16:25:26 -0400717 sk->sk_user_data = NULL;
Trond Myklebust2a9e1cf2008-10-28 15:21:39 -0400718
719 xs_restore_old_callbacks(transport, sk);
Chuck Levera246b012005-08-11 16:25:23 -0400720 write_unlock_bh(&sk->sk_callback_lock);
721
Chuck Lever9903cd12005-08-11 16:25:26 -0400722 sk->sk_no_check = 0;
Chuck Levera246b012005-08-11 16:25:23 -0400723
724 sock_release(sock);
Chuck Leverfe315e72009-03-11 14:10:21 -0400725}
726
727/**
728 * xs_close - close a socket
729 * @xprt: transport
730 *
731 * This is used when all requests are complete; ie, no DRC state remains
732 * on the server we want to save.
Trond Myklebustf75e6742009-04-21 17:18:20 -0400733 *
734 * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with
735 * xs_reset_transport() zeroing the socket from underneath a writer.
Chuck Leverfe315e72009-03-11 14:10:21 -0400736 */
737static void xs_close(struct rpc_xprt *xprt)
738{
739 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
740
741 dprintk("RPC: xs_close xprt %p\n", xprt);
742
743 xs_reset_transport(transport);
Neil Brown61d0a8e2009-09-23 14:36:37 -0400744 xprt->reestablish_timeout = 0;
Chuck Leverfe315e72009-03-11 14:10:21 -0400745
Trond Myklebust632e3bd2006-01-03 09:55:55 +0100746 smp_mb__before_clear_bit();
Trond Myklebust7d1e8252009-03-11 14:38:03 -0400747 clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
Trond Myklebust632e3bd2006-01-03 09:55:55 +0100748 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
Trond Myklebust3b948ae2007-11-05 17:42:39 -0500749 clear_bit(XPRT_CLOSING, &xprt->state);
Trond Myklebust632e3bd2006-01-03 09:55:55 +0100750 smp_mb__after_clear_bit();
Trond Myklebust62da3b22007-11-06 18:44:20 -0500751 xprt_disconnect_done(xprt);
Chuck Levera246b012005-08-11 16:25:23 -0400752}
753
Trond Myklebustf75e6742009-04-21 17:18:20 -0400754static void xs_tcp_close(struct rpc_xprt *xprt)
755{
756 if (test_and_clear_bit(XPRT_CONNECTION_CLOSE, &xprt->state))
757 xs_close(xprt);
758 else
759 xs_tcp_shutdown(xprt);
760}
761
Chuck Lever9903cd12005-08-11 16:25:26 -0400762/**
763 * xs_destroy - prepare to shutdown a transport
764 * @xprt: doomed transport
765 *
766 */
767static void xs_destroy(struct rpc_xprt *xprt)
Chuck Levera246b012005-08-11 16:25:23 -0400768{
Chuck Leverc8475462006-12-05 16:35:26 -0500769 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
770
Chuck Lever46121cf2007-01-31 12:14:08 -0500771 dprintk("RPC: xs_destroy xprt %p\n", xprt);
Chuck Lever9903cd12005-08-11 16:25:26 -0400772
Trond Myklebustc1384c92007-06-14 18:00:42 -0400773 cancel_rearming_delayed_work(&transport->connect_worker);
Chuck Levera246b012005-08-11 16:25:23 -0400774
Chuck Lever9903cd12005-08-11 16:25:26 -0400775 xs_close(xprt);
Chuck Leveredb267a2006-08-22 20:06:18 -0400776 xs_free_peer_addresses(xprt);
Chuck Levera246b012005-08-11 16:25:23 -0400777 kfree(xprt->slot);
Chuck Leverc8541ec2006-10-17 14:44:27 -0400778 kfree(xprt);
\"Talpey, Thomas\bc255712007-09-10 13:46:39 -0400779 module_put(THIS_MODULE);
Chuck Levera246b012005-08-11 16:25:23 -0400780}
781
Chuck Lever9903cd12005-08-11 16:25:26 -0400782static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
Chuck Levera246b012005-08-11 16:25:23 -0400783{
Chuck Lever9903cd12005-08-11 16:25:26 -0400784 return (struct rpc_xprt *) sk->sk_user_data;
785}
786
787/**
788 * xs_udp_data_ready - "data ready" callback for UDP sockets
789 * @sk: socket with data to read
790 * @len: how much data to read
791 *
792 */
793static void xs_udp_data_ready(struct sock *sk, int len)
794{
795 struct rpc_task *task;
796 struct rpc_xprt *xprt;
Chuck Levera246b012005-08-11 16:25:23 -0400797 struct rpc_rqst *rovr;
Chuck Lever9903cd12005-08-11 16:25:26 -0400798 struct sk_buff *skb;
Chuck Levera246b012005-08-11 16:25:23 -0400799 int err, repsize, copied;
Alexey Dobriyand8ed0292006-09-26 22:29:38 -0700800 u32 _xid;
801 __be32 *xp;
Chuck Levera246b012005-08-11 16:25:23 -0400802
803 read_lock(&sk->sk_callback_lock);
Chuck Lever46121cf2007-01-31 12:14:08 -0500804 dprintk("RPC: xs_udp_data_ready...\n");
Chuck Lever9903cd12005-08-11 16:25:26 -0400805 if (!(xprt = xprt_from_sock(sk)))
Chuck Levera246b012005-08-11 16:25:23 -0400806 goto out;
Chuck Levera246b012005-08-11 16:25:23 -0400807
808 if ((skb = skb_recv_datagram(sk, 0, 1, &err)) == NULL)
809 goto out;
810
811 if (xprt->shutdown)
812 goto dropit;
813
814 repsize = skb->len - sizeof(struct udphdr);
815 if (repsize < 4) {
Chuck Lever46121cf2007-01-31 12:14:08 -0500816 dprintk("RPC: impossible RPC reply size %d!\n", repsize);
Chuck Levera246b012005-08-11 16:25:23 -0400817 goto dropit;
818 }
819
820 /* Copy the XID from the skb... */
821 xp = skb_header_pointer(skb, sizeof(struct udphdr),
822 sizeof(_xid), &_xid);
823 if (xp == NULL)
824 goto dropit;
825
826 /* Look up and lock the request corresponding to the given XID */
Chuck Lever4a0f8c02005-08-11 16:25:32 -0400827 spin_lock(&xprt->transport_lock);
Chuck Levera246b012005-08-11 16:25:23 -0400828 rovr = xprt_lookup_rqst(xprt, *xp);
829 if (!rovr)
830 goto out_unlock;
831 task = rovr->rq_task;
832
Chuck Levera246b012005-08-11 16:25:23 -0400833 if ((copied = rovr->rq_private_buf.buflen) > repsize)
834 copied = repsize;
835
836 /* Suck it into the iovec, verify checksum if not done by hw. */
Herbert Xu1781f7f2007-12-11 11:30:32 -0800837 if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
838 UDPX_INC_STATS_BH(sk, UDP_MIB_INERRORS);
Chuck Levera246b012005-08-11 16:25:23 -0400839 goto out_unlock;
Herbert Xu1781f7f2007-12-11 11:30:32 -0800840 }
841
842 UDPX_INC_STATS_BH(sk, UDP_MIB_INDATAGRAMS);
Chuck Levera246b012005-08-11 16:25:23 -0400843
844 /* Something worked... */
Eric Dumazetadf30902009-06-02 05:19:30 +0000845 dst_confirm(skb_dst(skb));
Chuck Levera246b012005-08-11 16:25:23 -0400846
Chuck Lever1570c1e2005-08-25 16:25:52 -0700847 xprt_adjust_cwnd(task, copied);
Chuck Lever1570c1e2005-08-25 16:25:52 -0700848 xprt_complete_rqst(task, copied);
Chuck Levera246b012005-08-11 16:25:23 -0400849
850 out_unlock:
Chuck Lever4a0f8c02005-08-11 16:25:32 -0400851 spin_unlock(&xprt->transport_lock);
Chuck Levera246b012005-08-11 16:25:23 -0400852 dropit:
853 skb_free_datagram(sk, skb);
854 out:
855 read_unlock(&sk->sk_callback_lock);
856}
857
Chuck Leverdd456472006-12-05 16:35:44 -0500858static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
Chuck Levera246b012005-08-11 16:25:23 -0400859{
Chuck Lever51971132006-12-05 16:35:19 -0500860 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
Chuck Levera246b012005-08-11 16:25:23 -0400861 size_t len, used;
862 char *p;
863
Chuck Lever51971132006-12-05 16:35:19 -0500864 p = ((char *) &transport->tcp_fraghdr) + transport->tcp_offset;
865 len = sizeof(transport->tcp_fraghdr) - transport->tcp_offset;
Chuck Lever9d292312006-12-05 16:35:41 -0500866 used = xdr_skb_read_bits(desc, p, len);
Chuck Lever51971132006-12-05 16:35:19 -0500867 transport->tcp_offset += used;
Chuck Levera246b012005-08-11 16:25:23 -0400868 if (used != len)
869 return;
Chuck Lever808012f2005-08-25 16:25:49 -0700870
Chuck Lever51971132006-12-05 16:35:19 -0500871 transport->tcp_reclen = ntohl(transport->tcp_fraghdr);
872 if (transport->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
Chuck Levere136d092006-12-05 16:35:23 -0500873 transport->tcp_flags |= TCP_RCV_LAST_FRAG;
Chuck Levera246b012005-08-11 16:25:23 -0400874 else
Chuck Levere136d092006-12-05 16:35:23 -0500875 transport->tcp_flags &= ~TCP_RCV_LAST_FRAG;
Chuck Lever51971132006-12-05 16:35:19 -0500876 transport->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
Chuck Lever808012f2005-08-25 16:25:49 -0700877
Chuck Levere136d092006-12-05 16:35:23 -0500878 transport->tcp_flags &= ~TCP_RCV_COPY_FRAGHDR;
Chuck Lever51971132006-12-05 16:35:19 -0500879 transport->tcp_offset = 0;
Chuck Lever808012f2005-08-25 16:25:49 -0700880
Chuck Levera246b012005-08-11 16:25:23 -0400881 /* Sanity check of the record length */
Ricardo Labiaga18dca022009-04-01 09:22:53 -0400882 if (unlikely(transport->tcp_reclen < 8)) {
Chuck Lever46121cf2007-01-31 12:14:08 -0500883 dprintk("RPC: invalid TCP record fragment length\n");
Trond Myklebust3ebb0672007-11-06 18:40:12 -0500884 xprt_force_disconnect(xprt);
Chuck Lever9903cd12005-08-11 16:25:26 -0400885 return;
Chuck Levera246b012005-08-11 16:25:23 -0400886 }
Chuck Lever46121cf2007-01-31 12:14:08 -0500887 dprintk("RPC: reading TCP record fragment of length %d\n",
Chuck Lever51971132006-12-05 16:35:19 -0500888 transport->tcp_reclen);
Chuck Levera246b012005-08-11 16:25:23 -0400889}
890
Chuck Lever51971132006-12-05 16:35:19 -0500891static void xs_tcp_check_fraghdr(struct sock_xprt *transport)
Chuck Levera246b012005-08-11 16:25:23 -0400892{
Chuck Lever51971132006-12-05 16:35:19 -0500893 if (transport->tcp_offset == transport->tcp_reclen) {
Chuck Levere136d092006-12-05 16:35:23 -0500894 transport->tcp_flags |= TCP_RCV_COPY_FRAGHDR;
Chuck Lever51971132006-12-05 16:35:19 -0500895 transport->tcp_offset = 0;
Chuck Levere136d092006-12-05 16:35:23 -0500896 if (transport->tcp_flags & TCP_RCV_LAST_FRAG) {
897 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
898 transport->tcp_flags |= TCP_RCV_COPY_XID;
Chuck Lever51971132006-12-05 16:35:19 -0500899 transport->tcp_copied = 0;
Chuck Levera246b012005-08-11 16:25:23 -0400900 }
901 }
902}
903
Chuck Leverdd456472006-12-05 16:35:44 -0500904static inline void xs_tcp_read_xid(struct sock_xprt *transport, struct xdr_skb_reader *desc)
Chuck Levera246b012005-08-11 16:25:23 -0400905{
906 size_t len, used;
907 char *p;
908
Chuck Lever51971132006-12-05 16:35:19 -0500909 len = sizeof(transport->tcp_xid) - transport->tcp_offset;
Chuck Lever46121cf2007-01-31 12:14:08 -0500910 dprintk("RPC: reading XID (%Zu bytes)\n", len);
Chuck Lever51971132006-12-05 16:35:19 -0500911 p = ((char *) &transport->tcp_xid) + transport->tcp_offset;
Chuck Lever9d292312006-12-05 16:35:41 -0500912 used = xdr_skb_read_bits(desc, p, len);
Chuck Lever51971132006-12-05 16:35:19 -0500913 transport->tcp_offset += used;
Chuck Levera246b012005-08-11 16:25:23 -0400914 if (used != len)
915 return;
Chuck Levere136d092006-12-05 16:35:23 -0500916 transport->tcp_flags &= ~TCP_RCV_COPY_XID;
Ricardo Labiagaf4a2e412009-04-01 09:22:54 -0400917 transport->tcp_flags |= TCP_RCV_READ_CALLDIR;
Chuck Lever51971132006-12-05 16:35:19 -0500918 transport->tcp_copied = 4;
Ricardo Labiaga18dca022009-04-01 09:22:53 -0400919 dprintk("RPC: reading %s XID %08x\n",
920 (transport->tcp_flags & TCP_RPC_REPLY) ? "reply for"
921 : "request with",
Chuck Lever51971132006-12-05 16:35:19 -0500922 ntohl(transport->tcp_xid));
923 xs_tcp_check_fraghdr(transport);
Chuck Levera246b012005-08-11 16:25:23 -0400924}
925
Ricardo Labiaga18dca022009-04-01 09:22:53 -0400926static inline void xs_tcp_read_calldir(struct sock_xprt *transport,
927 struct xdr_skb_reader *desc)
Chuck Levera246b012005-08-11 16:25:23 -0400928{
Ricardo Labiaga18dca022009-04-01 09:22:53 -0400929 size_t len, used;
930 u32 offset;
Trond Myklebustb76ce562010-06-16 13:57:32 -0400931 char *p;
Ricardo Labiaga18dca022009-04-01 09:22:53 -0400932
933 /*
934 * We want transport->tcp_offset to be 8 at the end of this routine
935 * (4 bytes for the xid and 4 bytes for the call/reply flag).
936 * When this function is called for the first time,
937 * transport->tcp_offset is 4 (after having already read the xid).
938 */
939 offset = transport->tcp_offset - sizeof(transport->tcp_xid);
Trond Myklebustb76ce562010-06-16 13:57:32 -0400940 len = sizeof(transport->tcp_calldir) - offset;
Ricardo Labiaga18dca022009-04-01 09:22:53 -0400941 dprintk("RPC: reading CALL/REPLY flag (%Zu bytes)\n", len);
Trond Myklebustb76ce562010-06-16 13:57:32 -0400942 p = ((char *) &transport->tcp_calldir) + offset;
943 used = xdr_skb_read_bits(desc, p, len);
Ricardo Labiaga18dca022009-04-01 09:22:53 -0400944 transport->tcp_offset += used;
945 if (used != len)
946 return;
Ricardo Labiagaf4a2e412009-04-01 09:22:54 -0400947 transport->tcp_flags &= ~TCP_RCV_READ_CALLDIR;
Ricardo Labiagaf4a2e412009-04-01 09:22:54 -0400948 /*
949 * We don't yet have the XDR buffer, so we will write the calldir
950 * out after we get the buffer from the 'struct rpc_rqst'
951 */
Trond Myklebustb76ce562010-06-16 13:57:32 -0400952 switch (ntohl(transport->tcp_calldir)) {
953 case RPC_REPLY:
954 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
955 transport->tcp_flags |= TCP_RCV_COPY_DATA;
Ricardo Labiaga18dca022009-04-01 09:22:53 -0400956 transport->tcp_flags |= TCP_RPC_REPLY;
Trond Myklebustb76ce562010-06-16 13:57:32 -0400957 break;
958 case RPC_CALL:
959 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
960 transport->tcp_flags |= TCP_RCV_COPY_DATA;
Ricardo Labiaga18dca022009-04-01 09:22:53 -0400961 transport->tcp_flags &= ~TCP_RPC_REPLY;
Trond Myklebustb76ce562010-06-16 13:57:32 -0400962 break;
963 default:
964 dprintk("RPC: invalid request message type\n");
965 xprt_force_disconnect(&transport->xprt);
966 }
Ricardo Labiaga18dca022009-04-01 09:22:53 -0400967 xs_tcp_check_fraghdr(transport);
968}
969
Ricardo Labiaga44b98ef2009-04-01 09:23:02 -0400970static inline void xs_tcp_read_common(struct rpc_xprt *xprt,
971 struct xdr_skb_reader *desc,
972 struct rpc_rqst *req)
Chuck Levera246b012005-08-11 16:25:23 -0400973{
Ricardo Labiaga44b98ef2009-04-01 09:23:02 -0400974 struct sock_xprt *transport =
975 container_of(xprt, struct sock_xprt, xprt);
Chuck Levera246b012005-08-11 16:25:23 -0400976 struct xdr_buf *rcvbuf;
977 size_t len;
978 ssize_t r;
979
Chuck Levera246b012005-08-11 16:25:23 -0400980 rcvbuf = &req->rq_private_buf;
Ricardo Labiagaf4a2e412009-04-01 09:22:54 -0400981
982 if (transport->tcp_flags & TCP_RCV_COPY_CALLDIR) {
983 /*
984 * Save the RPC direction in the XDR buffer
985 */
Ricardo Labiagaf4a2e412009-04-01 09:22:54 -0400986 memcpy(rcvbuf->head[0].iov_base + transport->tcp_copied,
Trond Myklebustb76ce562010-06-16 13:57:32 -0400987 &transport->tcp_calldir,
988 sizeof(transport->tcp_calldir));
989 transport->tcp_copied += sizeof(transport->tcp_calldir);
Ricardo Labiagaf4a2e412009-04-01 09:22:54 -0400990 transport->tcp_flags &= ~TCP_RCV_COPY_CALLDIR;
Chuck Levera246b012005-08-11 16:25:23 -0400991 }
992
Chuck Levera246b012005-08-11 16:25:23 -0400993 len = desc->count;
Chuck Lever51971132006-12-05 16:35:19 -0500994 if (len > transport->tcp_reclen - transport->tcp_offset) {
Chuck Leverdd456472006-12-05 16:35:44 -0500995 struct xdr_skb_reader my_desc;
Chuck Levera246b012005-08-11 16:25:23 -0400996
Chuck Lever51971132006-12-05 16:35:19 -0500997 len = transport->tcp_reclen - transport->tcp_offset;
Chuck Levera246b012005-08-11 16:25:23 -0400998 memcpy(&my_desc, desc, sizeof(my_desc));
999 my_desc.count = len;
Chuck Lever51971132006-12-05 16:35:19 -05001000 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
Chuck Lever9d292312006-12-05 16:35:41 -05001001 &my_desc, xdr_skb_read_bits);
Chuck Levera246b012005-08-11 16:25:23 -04001002 desc->count -= r;
1003 desc->offset += r;
1004 } else
Chuck Lever51971132006-12-05 16:35:19 -05001005 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
Chuck Lever9d292312006-12-05 16:35:41 -05001006 desc, xdr_skb_read_bits);
Chuck Levera246b012005-08-11 16:25:23 -04001007
1008 if (r > 0) {
Chuck Lever51971132006-12-05 16:35:19 -05001009 transport->tcp_copied += r;
1010 transport->tcp_offset += r;
Chuck Levera246b012005-08-11 16:25:23 -04001011 }
1012 if (r != len) {
1013 /* Error when copying to the receive buffer,
1014 * usually because we weren't able to allocate
1015 * additional buffer pages. All we can do now
Chuck Levere136d092006-12-05 16:35:23 -05001016 * is turn off TCP_RCV_COPY_DATA, so the request
Chuck Levera246b012005-08-11 16:25:23 -04001017 * will not receive any additional updates,
1018 * and time out.
1019 * Any remaining data from this record will
1020 * be discarded.
1021 */
Chuck Levere136d092006-12-05 16:35:23 -05001022 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
Chuck Lever46121cf2007-01-31 12:14:08 -05001023 dprintk("RPC: XID %08x truncated request\n",
Chuck Lever51971132006-12-05 16:35:19 -05001024 ntohl(transport->tcp_xid));
Chuck Lever46121cf2007-01-31 12:14:08 -05001025 dprintk("RPC: xprt = %p, tcp_copied = %lu, "
1026 "tcp_offset = %u, tcp_reclen = %u\n",
1027 xprt, transport->tcp_copied,
1028 transport->tcp_offset, transport->tcp_reclen);
Ricardo Labiaga44b98ef2009-04-01 09:23:02 -04001029 return;
Chuck Levera246b012005-08-11 16:25:23 -04001030 }
1031
Chuck Lever46121cf2007-01-31 12:14:08 -05001032 dprintk("RPC: XID %08x read %Zd bytes\n",
Chuck Lever51971132006-12-05 16:35:19 -05001033 ntohl(transport->tcp_xid), r);
Chuck Lever46121cf2007-01-31 12:14:08 -05001034 dprintk("RPC: xprt = %p, tcp_copied = %lu, tcp_offset = %u, "
1035 "tcp_reclen = %u\n", xprt, transport->tcp_copied,
1036 transport->tcp_offset, transport->tcp_reclen);
Chuck Levera246b012005-08-11 16:25:23 -04001037
Chuck Lever51971132006-12-05 16:35:19 -05001038 if (transport->tcp_copied == req->rq_private_buf.buflen)
Chuck Levere136d092006-12-05 16:35:23 -05001039 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
Chuck Lever51971132006-12-05 16:35:19 -05001040 else if (transport->tcp_offset == transport->tcp_reclen) {
Chuck Levere136d092006-12-05 16:35:23 -05001041 if (transport->tcp_flags & TCP_RCV_LAST_FRAG)
1042 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
Chuck Levera246b012005-08-11 16:25:23 -04001043 }
Ricardo Labiaga44b98ef2009-04-01 09:23:02 -04001044}
1045
1046/*
1047 * Finds the request corresponding to the RPC xid and invokes the common
1048 * tcp read code to read the data.
1049 */
1050static inline int xs_tcp_read_reply(struct rpc_xprt *xprt,
1051 struct xdr_skb_reader *desc)
1052{
1053 struct sock_xprt *transport =
1054 container_of(xprt, struct sock_xprt, xprt);
1055 struct rpc_rqst *req;
1056
1057 dprintk("RPC: read reply XID %08x\n", ntohl(transport->tcp_xid));
1058
1059 /* Find and lock the request corresponding to this xid */
1060 spin_lock(&xprt->transport_lock);
1061 req = xprt_lookup_rqst(xprt, transport->tcp_xid);
1062 if (!req) {
1063 dprintk("RPC: XID %08x request not found!\n",
1064 ntohl(transport->tcp_xid));
1065 spin_unlock(&xprt->transport_lock);
1066 return -1;
1067 }
1068
1069 xs_tcp_read_common(xprt, desc, req);
1070
Chuck Levere136d092006-12-05 16:35:23 -05001071 if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
Chuck Lever51971132006-12-05 16:35:19 -05001072 xprt_complete_rqst(req->rq_task, transport->tcp_copied);
Ricardo Labiaga44b98ef2009-04-01 09:23:02 -04001073
Chuck Lever4a0f8c02005-08-11 16:25:32 -04001074 spin_unlock(&xprt->transport_lock);
Ricardo Labiaga44b98ef2009-04-01 09:23:02 -04001075 return 0;
1076}
1077
1078#if defined(CONFIG_NFS_V4_1)
1079/*
1080 * Obtains an rpc_rqst previously allocated and invokes the common
1081 * tcp read code to read the data. The result is placed in the callback
1082 * queue.
1083 * If we're unable to obtain the rpc_rqst we schedule the closing of the
1084 * connection and return -1.
1085 */
1086static inline int xs_tcp_read_callback(struct rpc_xprt *xprt,
1087 struct xdr_skb_reader *desc)
1088{
1089 struct sock_xprt *transport =
1090 container_of(xprt, struct sock_xprt, xprt);
1091 struct rpc_rqst *req;
1092
1093 req = xprt_alloc_bc_request(xprt);
1094 if (req == NULL) {
1095 printk(KERN_WARNING "Callback slot table overflowed\n");
1096 xprt_force_disconnect(xprt);
1097 return -1;
1098 }
1099
1100 req->rq_xid = transport->tcp_xid;
1101 dprintk("RPC: read callback XID %08x\n", ntohl(req->rq_xid));
1102 xs_tcp_read_common(xprt, desc, req);
1103
1104 if (!(transport->tcp_flags & TCP_RCV_COPY_DATA)) {
1105 struct svc_serv *bc_serv = xprt->bc_serv;
1106
1107 /*
1108 * Add callback request to callback list. The callback
1109 * service sleeps on the sv_cb_waitq waiting for new
1110 * requests. Wake it up after adding enqueing the
1111 * request.
1112 */
1113 dprintk("RPC: add callback request to list\n");
1114 spin_lock(&bc_serv->sv_cb_lock);
1115 list_add(&req->rq_bc_list, &bc_serv->sv_cb_list);
1116 spin_unlock(&bc_serv->sv_cb_lock);
1117 wake_up(&bc_serv->sv_cb_waitq);
1118 }
1119
1120 req->rq_private_buf.len = transport->tcp_copied;
1121
1122 return 0;
1123}
1124
1125static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1126 struct xdr_skb_reader *desc)
1127{
1128 struct sock_xprt *transport =
1129 container_of(xprt, struct sock_xprt, xprt);
1130
1131 return (transport->tcp_flags & TCP_RPC_REPLY) ?
1132 xs_tcp_read_reply(xprt, desc) :
1133 xs_tcp_read_callback(xprt, desc);
1134}
1135#else
1136static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1137 struct xdr_skb_reader *desc)
1138{
1139 return xs_tcp_read_reply(xprt, desc);
1140}
1141#endif /* CONFIG_NFS_V4_1 */
1142
1143/*
1144 * Read data off the transport. This can be either an RPC_CALL or an
1145 * RPC_REPLY. Relay the processing to helper functions.
1146 */
1147static void xs_tcp_read_data(struct rpc_xprt *xprt,
1148 struct xdr_skb_reader *desc)
1149{
1150 struct sock_xprt *transport =
1151 container_of(xprt, struct sock_xprt, xprt);
1152
1153 if (_xs_tcp_read_data(xprt, desc) == 0)
1154 xs_tcp_check_fraghdr(transport);
1155 else {
1156 /*
1157 * The transport_lock protects the request handling.
1158 * There's no need to hold it to update the tcp_flags.
1159 */
1160 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1161 }
Chuck Levera246b012005-08-11 16:25:23 -04001162}
1163
Chuck Leverdd456472006-12-05 16:35:44 -05001164static inline void xs_tcp_read_discard(struct sock_xprt *transport, struct xdr_skb_reader *desc)
Chuck Levera246b012005-08-11 16:25:23 -04001165{
1166 size_t len;
1167
Chuck Lever51971132006-12-05 16:35:19 -05001168 len = transport->tcp_reclen - transport->tcp_offset;
Chuck Levera246b012005-08-11 16:25:23 -04001169 if (len > desc->count)
1170 len = desc->count;
1171 desc->count -= len;
1172 desc->offset += len;
Chuck Lever51971132006-12-05 16:35:19 -05001173 transport->tcp_offset += len;
Chuck Lever46121cf2007-01-31 12:14:08 -05001174 dprintk("RPC: discarded %Zu bytes\n", len);
Chuck Lever51971132006-12-05 16:35:19 -05001175 xs_tcp_check_fraghdr(transport);
Chuck Levera246b012005-08-11 16:25:23 -04001176}
1177
Chuck Lever9903cd12005-08-11 16:25:26 -04001178static 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 -04001179{
1180 struct rpc_xprt *xprt = rd_desc->arg.data;
Chuck Lever51971132006-12-05 16:35:19 -05001181 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
Chuck Leverdd456472006-12-05 16:35:44 -05001182 struct xdr_skb_reader desc = {
Chuck Levera246b012005-08-11 16:25:23 -04001183 .skb = skb,
1184 .offset = offset,
1185 .count = len,
Chuck Lever9903cd12005-08-11 16:25:26 -04001186 };
Chuck Levera246b012005-08-11 16:25:23 -04001187
Chuck Lever46121cf2007-01-31 12:14:08 -05001188 dprintk("RPC: xs_tcp_data_recv started\n");
Chuck Levera246b012005-08-11 16:25:23 -04001189 do {
1190 /* Read in a new fragment marker if necessary */
1191 /* Can we ever really expect to get completely empty fragments? */
Chuck Levere136d092006-12-05 16:35:23 -05001192 if (transport->tcp_flags & TCP_RCV_COPY_FRAGHDR) {
Chuck Lever9903cd12005-08-11 16:25:26 -04001193 xs_tcp_read_fraghdr(xprt, &desc);
Chuck Levera246b012005-08-11 16:25:23 -04001194 continue;
1195 }
1196 /* Read in the xid if necessary */
Chuck Levere136d092006-12-05 16:35:23 -05001197 if (transport->tcp_flags & TCP_RCV_COPY_XID) {
Chuck Lever51971132006-12-05 16:35:19 -05001198 xs_tcp_read_xid(transport, &desc);
Chuck Levera246b012005-08-11 16:25:23 -04001199 continue;
1200 }
Ricardo Labiaga18dca022009-04-01 09:22:53 -04001201 /* Read in the call/reply flag */
Ricardo Labiagaf4a2e412009-04-01 09:22:54 -04001202 if (transport->tcp_flags & TCP_RCV_READ_CALLDIR) {
Ricardo Labiaga18dca022009-04-01 09:22:53 -04001203 xs_tcp_read_calldir(transport, &desc);
1204 continue;
1205 }
Chuck Levera246b012005-08-11 16:25:23 -04001206 /* Read in the request data */
Chuck Levere136d092006-12-05 16:35:23 -05001207 if (transport->tcp_flags & TCP_RCV_COPY_DATA) {
Ricardo Labiaga44b98ef2009-04-01 09:23:02 -04001208 xs_tcp_read_data(xprt, &desc);
Chuck Levera246b012005-08-11 16:25:23 -04001209 continue;
1210 }
1211 /* Skip over any trailing bytes on short reads */
Chuck Lever51971132006-12-05 16:35:19 -05001212 xs_tcp_read_discard(transport, &desc);
Chuck Levera246b012005-08-11 16:25:23 -04001213 } while (desc.count);
Chuck Lever46121cf2007-01-31 12:14:08 -05001214 dprintk("RPC: xs_tcp_data_recv done\n");
Chuck Levera246b012005-08-11 16:25:23 -04001215 return len - desc.count;
1216}
1217
Chuck Lever9903cd12005-08-11 16:25:26 -04001218/**
1219 * xs_tcp_data_ready - "data ready" callback for TCP sockets
1220 * @sk: socket with data to read
1221 * @bytes: how much data to read
1222 *
1223 */
1224static void xs_tcp_data_ready(struct sock *sk, int bytes)
Chuck Levera246b012005-08-11 16:25:23 -04001225{
1226 struct rpc_xprt *xprt;
1227 read_descriptor_t rd_desc;
Trond Myklebustff2d7db2008-02-25 21:40:51 -08001228 int read;
Chuck Levera246b012005-08-11 16:25:23 -04001229
Chuck Lever46121cf2007-01-31 12:14:08 -05001230 dprintk("RPC: xs_tcp_data_ready...\n");
1231
Chuck Levera246b012005-08-11 16:25:23 -04001232 read_lock(&sk->sk_callback_lock);
Chuck Lever9903cd12005-08-11 16:25:26 -04001233 if (!(xprt = xprt_from_sock(sk)))
Chuck Levera246b012005-08-11 16:25:23 -04001234 goto out;
Chuck Levera246b012005-08-11 16:25:23 -04001235 if (xprt->shutdown)
1236 goto out;
1237
Neil Brown61d0a8e2009-09-23 14:36:37 -04001238 /* Any data means we had a useful conversation, so
1239 * the we don't need to delay the next reconnect
1240 */
1241 if (xprt->reestablish_timeout)
1242 xprt->reestablish_timeout = 0;
1243
Chuck Lever9903cd12005-08-11 16:25:26 -04001244 /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
Chuck Levera246b012005-08-11 16:25:23 -04001245 rd_desc.arg.data = xprt;
Trond Myklebustff2d7db2008-02-25 21:40:51 -08001246 do {
1247 rd_desc.count = 65536;
1248 read = tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
1249 } while (read > 0);
Chuck Levera246b012005-08-11 16:25:23 -04001250out:
1251 read_unlock(&sk->sk_callback_lock);
1252}
1253
Trond Myklebust7d1e8252009-03-11 14:38:03 -04001254/*
1255 * Do the equivalent of linger/linger2 handling for dealing with
1256 * broken servers that don't close the socket in a timely
1257 * fashion
1258 */
1259static void xs_tcp_schedule_linger_timeout(struct rpc_xprt *xprt,
1260 unsigned long timeout)
1261{
1262 struct sock_xprt *transport;
1263
1264 if (xprt_test_and_set_connecting(xprt))
1265 return;
1266 set_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1267 transport = container_of(xprt, struct sock_xprt, xprt);
1268 queue_delayed_work(rpciod_workqueue, &transport->connect_worker,
1269 timeout);
1270}
1271
1272static void xs_tcp_cancel_linger_timeout(struct rpc_xprt *xprt)
1273{
1274 struct sock_xprt *transport;
1275
1276 transport = container_of(xprt, struct sock_xprt, xprt);
1277
1278 if (!test_bit(XPRT_CONNECTION_ABORT, &xprt->state) ||
1279 !cancel_delayed_work(&transport->connect_worker))
1280 return;
1281 clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1282 xprt_clear_connecting(xprt);
1283}
1284
1285static void xs_sock_mark_closed(struct rpc_xprt *xprt)
1286{
1287 smp_mb__before_clear_bit();
1288 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1289 clear_bit(XPRT_CLOSING, &xprt->state);
1290 smp_mb__after_clear_bit();
1291 /* Mark transport as closed and wake up all pending tasks */
1292 xprt_disconnect_done(xprt);
1293}
1294
Chuck Lever9903cd12005-08-11 16:25:26 -04001295/**
1296 * xs_tcp_state_change - callback to handle TCP socket state changes
1297 * @sk: socket whose state has changed
1298 *
1299 */
1300static void xs_tcp_state_change(struct sock *sk)
Chuck Levera246b012005-08-11 16:25:23 -04001301{
Chuck Lever9903cd12005-08-11 16:25:26 -04001302 struct rpc_xprt *xprt;
Chuck Levera246b012005-08-11 16:25:23 -04001303
1304 read_lock(&sk->sk_callback_lock);
1305 if (!(xprt = xprt_from_sock(sk)))
1306 goto out;
Chuck Lever46121cf2007-01-31 12:14:08 -05001307 dprintk("RPC: xs_tcp_state_change client %p...\n", xprt);
1308 dprintk("RPC: state %x conn %d dead %d zapped %d\n",
1309 sk->sk_state, xprt_connected(xprt),
1310 sock_flag(sk, SOCK_DEAD),
1311 sock_flag(sk, SOCK_ZAPPED));
Chuck Levera246b012005-08-11 16:25:23 -04001312
1313 switch (sk->sk_state) {
1314 case TCP_ESTABLISHED:
Chuck Lever4a0f8c02005-08-11 16:25:32 -04001315 spin_lock_bh(&xprt->transport_lock);
Chuck Levera246b012005-08-11 16:25:23 -04001316 if (!xprt_test_and_set_connected(xprt)) {
Chuck Lever51971132006-12-05 16:35:19 -05001317 struct sock_xprt *transport = container_of(xprt,
1318 struct sock_xprt, xprt);
1319
Chuck Levera246b012005-08-11 16:25:23 -04001320 /* Reset TCP record info */
Chuck Lever51971132006-12-05 16:35:19 -05001321 transport->tcp_offset = 0;
1322 transport->tcp_reclen = 0;
1323 transport->tcp_copied = 0;
Chuck Levere136d092006-12-05 16:35:23 -05001324 transport->tcp_flags =
1325 TCP_RCV_COPY_FRAGHDR | TCP_RCV_COPY_XID;
Chuck Lever51971132006-12-05 16:35:19 -05001326
Trond Myklebust2a491992009-03-11 14:38:00 -04001327 xprt_wake_pending_tasks(xprt, -EAGAIN);
Chuck Levera246b012005-08-11 16:25:23 -04001328 }
Chuck Lever4a0f8c02005-08-11 16:25:32 -04001329 spin_unlock_bh(&xprt->transport_lock);
Chuck Levera246b012005-08-11 16:25:23 -04001330 break;
Trond Myklebust3b948ae2007-11-05 17:42:39 -05001331 case TCP_FIN_WAIT1:
1332 /* The client initiated a shutdown of the socket */
Trond Myklebust7c1d71c2008-04-17 16:52:57 -04001333 xprt->connect_cookie++;
Trond Myklebust663b8852008-01-01 18:42:12 -05001334 xprt->reestablish_timeout = 0;
Trond Myklebust3b948ae2007-11-05 17:42:39 -05001335 set_bit(XPRT_CLOSING, &xprt->state);
1336 smp_mb__before_clear_bit();
1337 clear_bit(XPRT_CONNECTED, &xprt->state);
Trond Myklebustef803672007-12-31 16:19:17 -05001338 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
Trond Myklebust3b948ae2007-11-05 17:42:39 -05001339 smp_mb__after_clear_bit();
Trond Myklebust25fe6142009-03-11 14:38:03 -04001340 xs_tcp_schedule_linger_timeout(xprt, xs_tcp_fin_timeout);
Chuck Levera246b012005-08-11 16:25:23 -04001341 break;
Trond Myklebust632e3bd2006-01-03 09:55:55 +01001342 case TCP_CLOSE_WAIT:
Trond Myklebust3b948ae2007-11-05 17:42:39 -05001343 /* The server initiated a shutdown of the socket */
Trond Myklebust66af1e52007-11-06 10:18:36 -05001344 xprt_force_disconnect(xprt);
Trond Myklebust663b8852008-01-01 18:42:12 -05001345 case TCP_SYN_SENT:
Trond Myklebust7c1d71c2008-04-17 16:52:57 -04001346 xprt->connect_cookie++;
Trond Myklebust663b8852008-01-01 18:42:12 -05001347 case TCP_CLOSING:
1348 /*
1349 * If the server closed down the connection, make sure that
1350 * we back off before reconnecting
1351 */
1352 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1353 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
Trond Myklebust3b948ae2007-11-05 17:42:39 -05001354 break;
1355 case TCP_LAST_ACK:
Trond Myklebust670f9452009-03-11 14:37:58 -04001356 set_bit(XPRT_CLOSING, &xprt->state);
Trond Myklebust25fe6142009-03-11 14:38:03 -04001357 xs_tcp_schedule_linger_timeout(xprt, xs_tcp_fin_timeout);
Trond Myklebust3b948ae2007-11-05 17:42:39 -05001358 smp_mb__before_clear_bit();
1359 clear_bit(XPRT_CONNECTED, &xprt->state);
1360 smp_mb__after_clear_bit();
1361 break;
1362 case TCP_CLOSE:
Trond Myklebust7d1e8252009-03-11 14:38:03 -04001363 xs_tcp_cancel_linger_timeout(xprt);
1364 xs_sock_mark_closed(xprt);
Chuck Levera246b012005-08-11 16:25:23 -04001365 }
1366 out:
1367 read_unlock(&sk->sk_callback_lock);
1368}
1369
Chuck Lever9903cd12005-08-11 16:25:26 -04001370/**
Trond Myklebust482f32e2009-03-11 14:38:00 -04001371 * xs_error_report - callback mainly for catching socket errors
Trond Myklebust2a9e1cf2008-10-28 15:21:39 -04001372 * @sk: socket
1373 */
Trond Myklebust482f32e2009-03-11 14:38:00 -04001374static void xs_error_report(struct sock *sk)
Trond Myklebust2a9e1cf2008-10-28 15:21:39 -04001375{
1376 struct rpc_xprt *xprt;
1377
1378 read_lock(&sk->sk_callback_lock);
Trond Myklebust2a9e1cf2008-10-28 15:21:39 -04001379 if (!(xprt = xprt_from_sock(sk)))
1380 goto out;
1381 dprintk("RPC: %s client %p...\n"
1382 "RPC: error %d\n",
1383 __func__, xprt, sk->sk_err);
Trond Myklebust482f32e2009-03-11 14:38:00 -04001384 xprt_wake_pending_tasks(xprt, -EAGAIN);
Trond Myklebust2a9e1cf2008-10-28 15:21:39 -04001385out:
1386 read_unlock(&sk->sk_callback_lock);
1387}
1388
Ilpo Järvinen1f0fa152009-02-06 23:48:33 -08001389static void xs_write_space(struct sock *sk)
1390{
1391 struct socket *sock;
1392 struct rpc_xprt *xprt;
1393
1394 if (unlikely(!(sock = sk->sk_socket)))
1395 return;
1396 clear_bit(SOCK_NOSPACE, &sock->flags);
1397
1398 if (unlikely(!(xprt = xprt_from_sock(sk))))
1399 return;
1400 if (test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags) == 0)
1401 return;
1402
1403 xprt_write_space(xprt);
1404}
1405
Trond Myklebust2a9e1cf2008-10-28 15:21:39 -04001406/**
Chuck Leverc7b2cae2005-08-11 16:25:50 -04001407 * xs_udp_write_space - callback invoked when socket buffer space
1408 * becomes available
Chuck Lever9903cd12005-08-11 16:25:26 -04001409 * @sk: socket whose state has changed
1410 *
Chuck Levera246b012005-08-11 16:25:23 -04001411 * Called when more output buffer space is available for this socket.
1412 * We try not to wake our writers until they can make "significant"
Chuck Leverc7b2cae2005-08-11 16:25:50 -04001413 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
Chuck Levera246b012005-08-11 16:25:23 -04001414 * with a bunch of small requests.
1415 */
Chuck Leverc7b2cae2005-08-11 16:25:50 -04001416static void xs_udp_write_space(struct sock *sk)
Chuck Levera246b012005-08-11 16:25:23 -04001417{
Chuck Levera246b012005-08-11 16:25:23 -04001418 read_lock(&sk->sk_callback_lock);
Chuck Levera246b012005-08-11 16:25:23 -04001419
Chuck Leverc7b2cae2005-08-11 16:25:50 -04001420 /* from net/core/sock.c:sock_def_write_space */
Ilpo Järvinen1f0fa152009-02-06 23:48:33 -08001421 if (sock_writeable(sk))
1422 xs_write_space(sk);
Chuck Leverc7b2cae2005-08-11 16:25:50 -04001423
Chuck Leverc7b2cae2005-08-11 16:25:50 -04001424 read_unlock(&sk->sk_callback_lock);
1425}
Chuck Levera246b012005-08-11 16:25:23 -04001426
Chuck Leverc7b2cae2005-08-11 16:25:50 -04001427/**
1428 * xs_tcp_write_space - callback invoked when socket buffer space
1429 * becomes available
1430 * @sk: socket whose state has changed
1431 *
1432 * Called when more output buffer space is available for this socket.
1433 * We try not to wake our writers until they can make "significant"
1434 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1435 * with a bunch of small requests.
1436 */
1437static void xs_tcp_write_space(struct sock *sk)
1438{
1439 read_lock(&sk->sk_callback_lock);
1440
1441 /* from net/core/stream.c:sk_stream_write_space */
Ilpo Järvinen1f0fa152009-02-06 23:48:33 -08001442 if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk))
1443 xs_write_space(sk);
Chuck Leverc7b2cae2005-08-11 16:25:50 -04001444
Chuck Levera246b012005-08-11 16:25:23 -04001445 read_unlock(&sk->sk_callback_lock);
1446}
1447
Chuck Lever470056c2005-08-25 16:25:56 -07001448static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
Chuck Levera246b012005-08-11 16:25:23 -04001449{
Chuck Leveree0ac0c2006-12-05 16:35:15 -05001450 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1451 struct sock *sk = transport->inet;
Chuck Levera246b012005-08-11 16:25:23 -04001452
Chuck Lever7c6e0662006-12-05 16:35:30 -05001453 if (transport->rcvsize) {
Chuck Levera246b012005-08-11 16:25:23 -04001454 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
Chuck Lever7c6e0662006-12-05 16:35:30 -05001455 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
Chuck Levera246b012005-08-11 16:25:23 -04001456 }
Chuck Lever7c6e0662006-12-05 16:35:30 -05001457 if (transport->sndsize) {
Chuck Levera246b012005-08-11 16:25:23 -04001458 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
Chuck Lever7c6e0662006-12-05 16:35:30 -05001459 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
Chuck Levera246b012005-08-11 16:25:23 -04001460 sk->sk_write_space(sk);
1461 }
1462}
1463
Chuck Lever43118c22005-08-25 16:25:49 -07001464/**
Chuck Lever470056c2005-08-25 16:25:56 -07001465 * xs_udp_set_buffer_size - set send and receive limits
Chuck Lever43118c22005-08-25 16:25:49 -07001466 * @xprt: generic transport
Chuck Lever470056c2005-08-25 16:25:56 -07001467 * @sndsize: requested size of send buffer, in bytes
1468 * @rcvsize: requested size of receive buffer, in bytes
Chuck Lever43118c22005-08-25 16:25:49 -07001469 *
Chuck Lever470056c2005-08-25 16:25:56 -07001470 * Set socket send and receive buffer size limits.
Chuck Lever43118c22005-08-25 16:25:49 -07001471 */
Chuck Lever470056c2005-08-25 16:25:56 -07001472static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
Chuck Lever43118c22005-08-25 16:25:49 -07001473{
Chuck Lever7c6e0662006-12-05 16:35:30 -05001474 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1475
1476 transport->sndsize = 0;
Chuck Lever470056c2005-08-25 16:25:56 -07001477 if (sndsize)
Chuck Lever7c6e0662006-12-05 16:35:30 -05001478 transport->sndsize = sndsize + 1024;
1479 transport->rcvsize = 0;
Chuck Lever470056c2005-08-25 16:25:56 -07001480 if (rcvsize)
Chuck Lever7c6e0662006-12-05 16:35:30 -05001481 transport->rcvsize = rcvsize + 1024;
Chuck Lever470056c2005-08-25 16:25:56 -07001482
1483 xs_udp_do_set_buffer_size(xprt);
Chuck Lever43118c22005-08-25 16:25:49 -07001484}
1485
Chuck Lever46c0ee82005-08-25 16:25:52 -07001486/**
1487 * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1488 * @task: task that timed out
1489 *
1490 * Adjust the congestion window after a retransmit timeout has occurred.
1491 */
1492static void xs_udp_timer(struct rpc_task *task)
1493{
1494 xprt_adjust_cwnd(task, -ETIMEDOUT);
1495}
1496
Chuck Leverb85d8802006-05-25 01:40:49 -04001497static unsigned short xs_get_random_port(void)
1498{
1499 unsigned short range = xprt_max_resvport - xprt_min_resvport;
1500 unsigned short rand = (unsigned short) net_random() % range;
1501 return rand + xprt_min_resvport;
1502}
1503
Chuck Lever92200412006-01-03 09:55:51 +01001504/**
1505 * xs_set_port - reset the port number in the remote endpoint address
1506 * @xprt: generic transport
1507 * @port: new port number
1508 *
1509 */
1510static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1511{
Chuck Lever46121cf2007-01-31 12:14:08 -05001512 dprintk("RPC: setting port for xprt %p to %u\n", xprt, port);
Chuck Leverc4efcb12006-08-22 20:06:19 -04001513
Chuck Lever9dc3b092009-08-09 15:09:46 -04001514 rpc_set_port(xs_addr(xprt), port);
1515 xs_update_peer_port(xprt);
Chuck Lever92200412006-01-03 09:55:51 +01001516}
1517
Trond Myklebust67a391d2007-11-05 17:40:58 -05001518static unsigned short xs_get_srcport(struct sock_xprt *transport, struct socket *sock)
1519{
Chuck Leverfbfffbd2009-08-09 15:09:46 -04001520 unsigned short port = transport->srcport;
Trond Myklebust67a391d2007-11-05 17:40:58 -05001521
1522 if (port == 0 && transport->xprt.resvport)
1523 port = xs_get_random_port();
1524 return port;
1525}
1526
1527static unsigned short xs_next_srcport(struct sock_xprt *transport, struct socket *sock, unsigned short port)
1528{
Chuck Leverfbfffbd2009-08-09 15:09:46 -04001529 if (transport->srcport != 0)
1530 transport->srcport = 0;
Trond Myklebust67a391d2007-11-05 17:40:58 -05001531 if (!transport->xprt.resvport)
1532 return 0;
1533 if (port <= xprt_min_resvport || port > xprt_max_resvport)
1534 return xprt_max_resvport;
1535 return --port;
1536}
1537
Chuck Lever7dc753f2007-08-06 11:57:28 -04001538static int xs_bind4(struct sock_xprt *transport, struct socket *sock)
Chuck Levera246b012005-08-11 16:25:23 -04001539{
1540 struct sockaddr_in myaddr = {
1541 .sin_family = AF_INET,
1542 };
Frank van Maarseveend3bc9a12007-07-09 22:23:35 +02001543 struct sockaddr_in *sa;
Trond Myklebust67a391d2007-11-05 17:40:58 -05001544 int err, nloop = 0;
1545 unsigned short port = xs_get_srcport(transport, sock);
1546 unsigned short last;
Chuck Levera246b012005-08-11 16:25:23 -04001547
Chuck Leverfbfffbd2009-08-09 15:09:46 -04001548 sa = (struct sockaddr_in *)&transport->srcaddr;
Frank van Maarseveend3bc9a12007-07-09 22:23:35 +02001549 myaddr.sin_addr = sa->sin_addr;
Chuck Levera246b012005-08-11 16:25:23 -04001550 do {
1551 myaddr.sin_port = htons(port);
Sridhar Samudralae6242e92006-08-07 20:58:01 -07001552 err = kernel_bind(sock, (struct sockaddr *) &myaddr,
Chuck Levera246b012005-08-11 16:25:23 -04001553 sizeof(myaddr));
Trond Myklebust67a391d2007-11-05 17:40:58 -05001554 if (port == 0)
Frank van Maarseveend3bc9a12007-07-09 22:23:35 +02001555 break;
Chuck Levera246b012005-08-11 16:25:23 -04001556 if (err == 0) {
Chuck Leverfbfffbd2009-08-09 15:09:46 -04001557 transport->srcport = port;
Frank van Maarseveend3bc9a12007-07-09 22:23:35 +02001558 break;
Chuck Levera246b012005-08-11 16:25:23 -04001559 }
Trond Myklebust67a391d2007-11-05 17:40:58 -05001560 last = port;
1561 port = xs_next_srcport(transport, sock, port);
1562 if (port > last)
1563 nloop++;
1564 } while (err == -EADDRINUSE && nloop != 2);
Harvey Harrison21454aa2008-10-31 00:54:56 -07001565 dprintk("RPC: %s %pI4:%u: %s (%d)\n",
1566 __func__, &myaddr.sin_addr,
Chuck Lever7dc753f2007-08-06 11:57:28 -04001567 port, err ? "failed" : "ok", err);
Chuck Levera246b012005-08-11 16:25:23 -04001568 return err;
1569}
1570
Chuck Lever90058d32007-08-06 11:57:33 -04001571static int xs_bind6(struct sock_xprt *transport, struct socket *sock)
1572{
1573 struct sockaddr_in6 myaddr = {
1574 .sin6_family = AF_INET6,
1575 };
1576 struct sockaddr_in6 *sa;
Trond Myklebust67a391d2007-11-05 17:40:58 -05001577 int err, nloop = 0;
1578 unsigned short port = xs_get_srcport(transport, sock);
1579 unsigned short last;
Chuck Lever90058d32007-08-06 11:57:33 -04001580
Chuck Leverfbfffbd2009-08-09 15:09:46 -04001581 sa = (struct sockaddr_in6 *)&transport->srcaddr;
Chuck Lever90058d32007-08-06 11:57:33 -04001582 myaddr.sin6_addr = sa->sin6_addr;
1583 do {
1584 myaddr.sin6_port = htons(port);
1585 err = kernel_bind(sock, (struct sockaddr *) &myaddr,
1586 sizeof(myaddr));
Trond Myklebust67a391d2007-11-05 17:40:58 -05001587 if (port == 0)
Chuck Lever90058d32007-08-06 11:57:33 -04001588 break;
1589 if (err == 0) {
Chuck Leverfbfffbd2009-08-09 15:09:46 -04001590 transport->srcport = port;
Chuck Lever90058d32007-08-06 11:57:33 -04001591 break;
1592 }
Trond Myklebust67a391d2007-11-05 17:40:58 -05001593 last = port;
1594 port = xs_next_srcport(transport, sock, port);
1595 if (port > last)
1596 nloop++;
1597 } while (err == -EADDRINUSE && nloop != 2);
Harvey Harrison5b095d9892008-10-29 12:52:50 -07001598 dprintk("RPC: xs_bind6 %pI6:%u: %s (%d)\n",
Harvey Harrisonfdb46ee2008-10-28 16:10:17 -07001599 &myaddr.sin6_addr, port, err ? "failed" : "ok", err);
Chuck Levera246b012005-08-11 16:25:23 -04001600 return err;
1601}
1602
Peter Zijlstraed075362006-12-06 20:35:24 -08001603#ifdef CONFIG_DEBUG_LOCK_ALLOC
1604static struct lock_class_key xs_key[2];
1605static struct lock_class_key xs_slock_key[2];
1606
Chuck Lever8945ee52007-08-06 11:58:04 -04001607static inline void xs_reclassify_socket4(struct socket *sock)
Peter Zijlstraed075362006-12-06 20:35:24 -08001608{
1609 struct sock *sk = sock->sk;
Chuck Lever8945ee52007-08-06 11:58:04 -04001610
John Heffner02b3d342007-09-12 10:42:12 +02001611 BUG_ON(sock_owned_by_user(sk));
Chuck Lever8945ee52007-08-06 11:58:04 -04001612 sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1613 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1614}
Peter Zijlstraed075362006-12-06 20:35:24 -08001615
Chuck Lever8945ee52007-08-06 11:58:04 -04001616static inline void xs_reclassify_socket6(struct socket *sock)
1617{
1618 struct sock *sk = sock->sk;
Peter Zijlstraed075362006-12-06 20:35:24 -08001619
Linus Torvaldsf4921af2007-10-15 10:46:05 -07001620 BUG_ON(sock_owned_by_user(sk));
Chuck Lever8945ee52007-08-06 11:58:04 -04001621 sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1622 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
Peter Zijlstraed075362006-12-06 20:35:24 -08001623}
1624#else
Chuck Lever8945ee52007-08-06 11:58:04 -04001625static inline void xs_reclassify_socket4(struct socket *sock)
1626{
1627}
1628
1629static inline void xs_reclassify_socket6(struct socket *sock)
Peter Zijlstraed075362006-12-06 20:35:24 -08001630{
1631}
1632#endif
1633
Chuck Lever16be2d22007-08-06 11:57:38 -04001634static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
Chuck Levera246b012005-08-11 16:25:23 -04001635{
Chuck Lever16be2d22007-08-06 11:57:38 -04001636 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
Chuck Leveredb267a2006-08-22 20:06:18 -04001637
Chuck Leveree0ac0c2006-12-05 16:35:15 -05001638 if (!transport->inet) {
Chuck Leverb0d93ad2005-08-11 16:25:53 -04001639 struct sock *sk = sock->sk;
1640
1641 write_lock_bh(&sk->sk_callback_lock);
1642
Trond Myklebust2a9e1cf2008-10-28 15:21:39 -04001643 xs_save_old_callbacks(transport, sk);
1644
Chuck Leverb0d93ad2005-08-11 16:25:53 -04001645 sk->sk_user_data = xprt;
Chuck Leverb0d93ad2005-08-11 16:25:53 -04001646 sk->sk_data_ready = xs_udp_data_ready;
1647 sk->sk_write_space = xs_udp_write_space;
Trond Myklebust482f32e2009-03-11 14:38:00 -04001648 sk->sk_error_report = xs_error_report;
Chuck Leverb0d93ad2005-08-11 16:25:53 -04001649 sk->sk_no_check = UDP_CSUM_NORCV;
Trond Myklebustb079fa7b2005-12-13 16:13:52 -05001650 sk->sk_allocation = GFP_ATOMIC;
Chuck Leverb0d93ad2005-08-11 16:25:53 -04001651
1652 xprt_set_connected(xprt);
1653
1654 /* Reset to new socket */
Chuck Leveree0ac0c2006-12-05 16:35:15 -05001655 transport->sock = sock;
1656 transport->inet = sk;
Chuck Leverb0d93ad2005-08-11 16:25:53 -04001657
1658 write_unlock_bh(&sk->sk_callback_lock);
1659 }
Chuck Lever470056c2005-08-25 16:25:56 -07001660 xs_udp_do_set_buffer_size(xprt);
Chuck Lever16be2d22007-08-06 11:57:38 -04001661}
1662
Chuck Levera246b012005-08-11 16:25:23 -04001663/**
Chuck Lever9c3d72d2007-08-06 11:57:43 -04001664 * xs_udp_connect_worker4 - set up a UDP socket
Chuck Levera246b012005-08-11 16:25:23 -04001665 * @work: RPC transport to connect
1666 *
1667 * Invoked by a work queue tasklet.
1668 */
Chuck Lever9c3d72d2007-08-06 11:57:43 -04001669static void xs_udp_connect_worker4(struct work_struct *work)
Chuck Levera246b012005-08-11 16:25:23 -04001670{
1671 struct sock_xprt *transport =
1672 container_of(work, struct sock_xprt, connect_worker.work);
1673 struct rpc_xprt *xprt = &transport->xprt;
1674 struct socket *sock = transport->sock;
1675 int err, status = -EIO;
1676
Trond Myklebust01d37c42009-03-11 14:09:39 -04001677 if (xprt->shutdown)
Chuck Lever9903cd12005-08-11 16:25:26 -04001678 goto out;
1679
Chuck Levera246b012005-08-11 16:25:23 -04001680 /* Start by resetting any existing state */
Chuck Leverfe315e72009-03-11 14:10:21 -04001681 xs_reset_transport(transport);
Chuck Levera246b012005-08-11 16:25:23 -04001682
Chuck Leverfe315e72009-03-11 14:10:21 -04001683 err = sock_create_kern(PF_INET, SOCK_DGRAM, IPPROTO_UDP, &sock);
1684 if (err < 0) {
Chuck Lever9903cd12005-08-11 16:25:26 -04001685 dprintk("RPC: can't create UDP transport socket (%d).\n", -err);
Chuck Levera246b012005-08-11 16:25:23 -04001686 goto out;
1687 }
Chuck Lever8945ee52007-08-06 11:58:04 -04001688 xs_reclassify_socket4(sock);
Chuck Levera246b012005-08-11 16:25:23 -04001689
Chuck Lever7dc753f2007-08-06 11:57:28 -04001690 if (xs_bind4(transport, sock)) {
Chuck Lever9903cd12005-08-11 16:25:26 -04001691 sock_release(sock);
1692 goto out;
Chuck Levera246b012005-08-11 16:25:23 -04001693 }
Chuck Leverb0d93ad2005-08-11 16:25:53 -04001694
Chuck Leverc740eff2009-08-09 15:09:46 -04001695 dprintk("RPC: worker connecting xprt %p via %s to "
1696 "%s (port %s)\n", xprt,
1697 xprt->address_strings[RPC_DISPLAY_PROTO],
1698 xprt->address_strings[RPC_DISPLAY_ADDR],
1699 xprt->address_strings[RPC_DISPLAY_PORT]);
Chuck Leverb0d93ad2005-08-11 16:25:53 -04001700
Chuck Lever16be2d22007-08-06 11:57:38 -04001701 xs_udp_finish_connecting(xprt, sock);
Chuck Levera246b012005-08-11 16:25:23 -04001702 status = 0;
Chuck Leverb0d93ad2005-08-11 16:25:53 -04001703out:
Chuck Leverb0d93ad2005-08-11 16:25:53 -04001704 xprt_clear_connecting(xprt);
Trond Myklebust7d1e8252009-03-11 14:38:03 -04001705 xprt_wake_pending_tasks(xprt, status);
Chuck Leverb0d93ad2005-08-11 16:25:53 -04001706}
1707
Chuck Lever68e220b2007-08-06 11:57:48 -04001708/**
1709 * xs_udp_connect_worker6 - set up a UDP socket
1710 * @work: RPC transport to connect
1711 *
1712 * Invoked by a work queue tasklet.
1713 */
1714static void xs_udp_connect_worker6(struct work_struct *work)
1715{
1716 struct sock_xprt *transport =
1717 container_of(work, struct sock_xprt, connect_worker.work);
1718 struct rpc_xprt *xprt = &transport->xprt;
1719 struct socket *sock = transport->sock;
1720 int err, status = -EIO;
1721
Trond Myklebust01d37c42009-03-11 14:09:39 -04001722 if (xprt->shutdown)
Chuck Lever68e220b2007-08-06 11:57:48 -04001723 goto out;
1724
1725 /* Start by resetting any existing state */
Chuck Leverfe315e72009-03-11 14:10:21 -04001726 xs_reset_transport(transport);
Chuck Lever68e220b2007-08-06 11:57:48 -04001727
Chuck Leverfe315e72009-03-11 14:10:21 -04001728 err = sock_create_kern(PF_INET6, SOCK_DGRAM, IPPROTO_UDP, &sock);
1729 if (err < 0) {
Chuck Lever68e220b2007-08-06 11:57:48 -04001730 dprintk("RPC: can't create UDP transport socket (%d).\n", -err);
1731 goto out;
1732 }
Chuck Lever8945ee52007-08-06 11:58:04 -04001733 xs_reclassify_socket6(sock);
Chuck Lever68e220b2007-08-06 11:57:48 -04001734
1735 if (xs_bind6(transport, sock) < 0) {
1736 sock_release(sock);
1737 goto out;
1738 }
1739
Chuck Leverc740eff2009-08-09 15:09:46 -04001740 dprintk("RPC: worker connecting xprt %p via %s to "
1741 "%s (port %s)\n", xprt,
1742 xprt->address_strings[RPC_DISPLAY_PROTO],
1743 xprt->address_strings[RPC_DISPLAY_ADDR],
1744 xprt->address_strings[RPC_DISPLAY_PORT]);
Chuck Lever68e220b2007-08-06 11:57:48 -04001745
1746 xs_udp_finish_connecting(xprt, sock);
Chuck Levera246b012005-08-11 16:25:23 -04001747 status = 0;
Chuck Leverb0d93ad2005-08-11 16:25:53 -04001748out:
Chuck Leverb0d93ad2005-08-11 16:25:53 -04001749 xprt_clear_connecting(xprt);
Trond Myklebust7d1e8252009-03-11 14:38:03 -04001750 xprt_wake_pending_tasks(xprt, status);
Chuck Leverb0d93ad2005-08-11 16:25:53 -04001751}
1752
Chuck Lever3167e122005-08-25 16:25:55 -07001753/*
1754 * We need to preserve the port number so the reply cache on the server can
1755 * find our cached RPC replies when we get around to reconnecting.
1756 */
Trond Myklebust40d25492009-03-11 14:37:58 -04001757static void xs_abort_connection(struct rpc_xprt *xprt, struct sock_xprt *transport)
Chuck Lever3167e122005-08-25 16:25:55 -07001758{
1759 int result;
Chuck Lever3167e122005-08-25 16:25:55 -07001760 struct sockaddr any;
1761
Chuck Lever46121cf2007-01-31 12:14:08 -05001762 dprintk("RPC: disconnecting xprt %p to reuse port\n", xprt);
Chuck Lever3167e122005-08-25 16:25:55 -07001763
1764 /*
1765 * Disconnect the transport socket by doing a connect operation
1766 * with AF_UNSPEC. This should return immediately...
1767 */
1768 memset(&any, 0, sizeof(any));
1769 any.sa_family = AF_UNSPEC;
Chuck Leveree0ac0c2006-12-05 16:35:15 -05001770 result = kernel_connect(transport->sock, &any, sizeof(any), 0);
Trond Myklebust7d1e8252009-03-11 14:38:03 -04001771 if (!result)
1772 xs_sock_mark_closed(xprt);
1773 else
Chuck Lever46121cf2007-01-31 12:14:08 -05001774 dprintk("RPC: AF_UNSPEC connect return code %d\n",
Chuck Lever3167e122005-08-25 16:25:55 -07001775 result);
1776}
1777
Trond Myklebust40d25492009-03-11 14:37:58 -04001778static void xs_tcp_reuse_connection(struct rpc_xprt *xprt, struct sock_xprt *transport)
1779{
1780 unsigned int state = transport->inet->sk_state;
1781
1782 if (state == TCP_CLOSE && transport->sock->state == SS_UNCONNECTED)
1783 return;
1784 if ((1 << state) & (TCPF_ESTABLISHED|TCPF_SYN_SENT))
1785 return;
1786 xs_abort_connection(xprt, transport);
1787}
1788
Chuck Lever16be2d22007-08-06 11:57:38 -04001789static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
Chuck Leverb0d93ad2005-08-11 16:25:53 -04001790{
Chuck Lever16be2d22007-08-06 11:57:38 -04001791 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
Chuck Leveredb267a2006-08-22 20:06:18 -04001792
Chuck Leveree0ac0c2006-12-05 16:35:15 -05001793 if (!transport->inet) {
Chuck Leverb0d93ad2005-08-11 16:25:53 -04001794 struct sock *sk = sock->sk;
1795
1796 write_lock_bh(&sk->sk_callback_lock);
1797
Trond Myklebust2a9e1cf2008-10-28 15:21:39 -04001798 xs_save_old_callbacks(transport, sk);
1799
Chuck Leverb0d93ad2005-08-11 16:25:53 -04001800 sk->sk_user_data = xprt;
Chuck Leverb0d93ad2005-08-11 16:25:53 -04001801 sk->sk_data_ready = xs_tcp_data_ready;
1802 sk->sk_state_change = xs_tcp_state_change;
1803 sk->sk_write_space = xs_tcp_write_space;
Trond Myklebust482f32e2009-03-11 14:38:00 -04001804 sk->sk_error_report = xs_error_report;
Trond Myklebustb079fa7b2005-12-13 16:13:52 -05001805 sk->sk_allocation = GFP_ATOMIC;
Chuck Lever3167e122005-08-25 16:25:55 -07001806
1807 /* socket options */
1808 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
1809 sock_reset_flag(sk, SOCK_LINGER);
1810 tcp_sk(sk)->linger2 = 0;
1811 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
Chuck Leverb0d93ad2005-08-11 16:25:53 -04001812
1813 xprt_clear_connected(xprt);
1814
1815 /* Reset to new socket */
Chuck Leveree0ac0c2006-12-05 16:35:15 -05001816 transport->sock = sock;
1817 transport->inet = sk;
Chuck Leverb0d93ad2005-08-11 16:25:53 -04001818
1819 write_unlock_bh(&sk->sk_callback_lock);
1820 }
1821
Trond Myklebust01d37c42009-03-11 14:09:39 -04001822 if (!xprt_bound(xprt))
1823 return -ENOTCONN;
1824
Chuck Leverb0d93ad2005-08-11 16:25:53 -04001825 /* Tell the socket layer to start connecting... */
Chuck Lever262ca072006-03-20 13:44:16 -05001826 xprt->stat.connect_count++;
1827 xprt->stat.connect_start = jiffies;
Chuck Lever95392c52007-08-06 11:57:58 -04001828 return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
Chuck Lever16be2d22007-08-06 11:57:38 -04001829}
1830
1831/**
Trond Myklebustb61d59f2009-03-11 14:38:04 -04001832 * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint
1833 * @xprt: RPC transport to connect
1834 * @transport: socket transport to connect
1835 * @create_sock: function to create a socket of the correct type
Chuck Lever16be2d22007-08-06 11:57:38 -04001836 *
1837 * Invoked by a work queue tasklet.
1838 */
Trond Myklebustb61d59f2009-03-11 14:38:04 -04001839static void xs_tcp_setup_socket(struct rpc_xprt *xprt,
1840 struct sock_xprt *transport,
1841 struct socket *(*create_sock)(struct rpc_xprt *,
1842 struct sock_xprt *))
Chuck Lever16be2d22007-08-06 11:57:38 -04001843{
Chuck Lever16be2d22007-08-06 11:57:38 -04001844 struct socket *sock = transport->sock;
Trond Myklebustb61d59f2009-03-11 14:38:04 -04001845 int status = -EIO;
Chuck Lever16be2d22007-08-06 11:57:38 -04001846
Trond Myklebust01d37c42009-03-11 14:09:39 -04001847 if (xprt->shutdown)
Chuck Lever16be2d22007-08-06 11:57:38 -04001848 goto out;
1849
1850 if (!sock) {
Trond Myklebust7d1e8252009-03-11 14:38:03 -04001851 clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
Trond Myklebustb61d59f2009-03-11 14:38:04 -04001852 sock = create_sock(xprt, transport);
1853 if (IS_ERR(sock)) {
1854 status = PTR_ERR(sock);
Chuck Lever16be2d22007-08-06 11:57:38 -04001855 goto out;
1856 }
Trond Myklebust7d1e8252009-03-11 14:38:03 -04001857 } else {
1858 int abort_and_exit;
1859
1860 abort_and_exit = test_and_clear_bit(XPRT_CONNECTION_ABORT,
1861 &xprt->state);
Chuck Lever16be2d22007-08-06 11:57:38 -04001862 /* "close" the socket, preserving the local port */
Trond Myklebust40d25492009-03-11 14:37:58 -04001863 xs_tcp_reuse_connection(xprt, transport);
Chuck Lever16be2d22007-08-06 11:57:38 -04001864
Trond Myklebust7d1e8252009-03-11 14:38:03 -04001865 if (abort_and_exit)
1866 goto out_eagain;
1867 }
1868
Chuck Leverc740eff2009-08-09 15:09:46 -04001869 dprintk("RPC: worker connecting xprt %p via %s to "
1870 "%s (port %s)\n", xprt,
1871 xprt->address_strings[RPC_DISPLAY_PROTO],
1872 xprt->address_strings[RPC_DISPLAY_ADDR],
1873 xprt->address_strings[RPC_DISPLAY_PORT]);
Chuck Lever16be2d22007-08-06 11:57:38 -04001874
1875 status = xs_tcp_finish_connecting(xprt, sock);
Chuck Lever46121cf2007-01-31 12:14:08 -05001876 dprintk("RPC: %p connect status %d connected %d sock state %d\n",
1877 xprt, -status, xprt_connected(xprt),
1878 sock->sk->sk_state);
Trond Myklebust2a491992009-03-11 14:38:00 -04001879 switch (status) {
Trond Myklebustf75e6742009-04-21 17:18:20 -04001880 default:
1881 printk("%s: connect returned unhandled error %d\n",
1882 __func__, status);
1883 case -EADDRNOTAVAIL:
1884 /* We're probably in TIME_WAIT. Get rid of existing socket,
1885 * and retry
1886 */
1887 set_bit(XPRT_CONNECTION_CLOSE, &xprt->state);
1888 xprt_force_disconnect(xprt);
Trond Myklebust88b5ed72009-06-17 13:22:57 -07001889 break;
Trond Myklebust8a2cec22009-03-11 14:38:01 -04001890 case -ECONNREFUSED:
1891 case -ECONNRESET:
1892 case -ENETUNREACH:
1893 /* retry with existing socket, after a delay */
Trond Myklebust2a491992009-03-11 14:38:00 -04001894 case 0:
1895 case -EINPROGRESS:
1896 case -EALREADY:
Trond Myklebust7d1e8252009-03-11 14:38:03 -04001897 xprt_clear_connecting(xprt);
1898 return;
Trond Myklebust9fcfe0c2010-03-02 13:06:21 -05001899 case -EINVAL:
1900 /* Happens, for instance, if the user specified a link
1901 * local IPv6 address without a scope-id.
1902 */
1903 goto out;
Chuck Levera246b012005-08-11 16:25:23 -04001904 }
Trond Myklebust7d1e8252009-03-11 14:38:03 -04001905out_eagain:
Trond Myklebust2a491992009-03-11 14:38:00 -04001906 status = -EAGAIN;
Chuck Levera246b012005-08-11 16:25:23 -04001907out:
Chuck Lever2226feb2005-08-11 16:25:38 -04001908 xprt_clear_connecting(xprt);
Trond Myklebust7d1e8252009-03-11 14:38:03 -04001909 xprt_wake_pending_tasks(xprt, status);
Chuck Levera246b012005-08-11 16:25:23 -04001910}
1911
Trond Myklebustb61d59f2009-03-11 14:38:04 -04001912static struct socket *xs_create_tcp_sock4(struct rpc_xprt *xprt,
1913 struct sock_xprt *transport)
1914{
1915 struct socket *sock;
1916 int err;
1917
1918 /* start from scratch */
1919 err = sock_create_kern(PF_INET, SOCK_STREAM, IPPROTO_TCP, &sock);
1920 if (err < 0) {
1921 dprintk("RPC: can't create TCP transport socket (%d).\n",
1922 -err);
1923 goto out_err;
1924 }
1925 xs_reclassify_socket4(sock);
1926
1927 if (xs_bind4(transport, sock) < 0) {
1928 sock_release(sock);
1929 goto out_err;
1930 }
1931 return sock;
1932out_err:
1933 return ERR_PTR(-EIO);
1934}
1935
1936/**
Chuck Levera246b012005-08-11 16:25:23 -04001937 * xs_tcp_connect_worker4 - connect a TCP socket to a remote endpoint
1938 * @work: RPC transport to connect
1939 *
1940 * Invoked by a work queue tasklet.
1941 */
1942static void xs_tcp_connect_worker4(struct work_struct *work)
1943{
1944 struct sock_xprt *transport =
1945 container_of(work, struct sock_xprt, connect_worker.work);
1946 struct rpc_xprt *xprt = &transport->xprt;
Chuck Levera246b012005-08-11 16:25:23 -04001947
Trond Myklebustb61d59f2009-03-11 14:38:04 -04001948 xs_tcp_setup_socket(xprt, transport, xs_create_tcp_sock4);
1949}
Chuck Levera246b012005-08-11 16:25:23 -04001950
Trond Myklebustb61d59f2009-03-11 14:38:04 -04001951static struct socket *xs_create_tcp_sock6(struct rpc_xprt *xprt,
1952 struct sock_xprt *transport)
1953{
1954 struct socket *sock;
1955 int err;
Chuck Lever9903cd12005-08-11 16:25:26 -04001956
Trond Myklebustb61d59f2009-03-11 14:38:04 -04001957 /* start from scratch */
1958 err = sock_create_kern(PF_INET6, SOCK_STREAM, IPPROTO_TCP, &sock);
1959 if (err < 0) {
1960 dprintk("RPC: can't create TCP transport socket (%d).\n",
1961 -err);
1962 goto out_err;
Chuck Levera246b012005-08-11 16:25:23 -04001963 }
Trond Myklebustb61d59f2009-03-11 14:38:04 -04001964 xs_reclassify_socket6(sock);
1965
1966 if (xs_bind6(transport, sock) < 0) {
1967 sock_release(sock);
1968 goto out_err;
1969 }
1970 return sock;
1971out_err:
1972 return ERR_PTR(-EIO);
Chuck Levera246b012005-08-11 16:25:23 -04001973}
1974
Chuck Lever9903cd12005-08-11 16:25:26 -04001975/**
Chuck Lever68e220b2007-08-06 11:57:48 -04001976 * xs_tcp_connect_worker6 - connect a TCP socket to a remote endpoint
1977 * @work: RPC transport to connect
1978 *
1979 * Invoked by a work queue tasklet.
1980 */
1981static void xs_tcp_connect_worker6(struct work_struct *work)
1982{
1983 struct sock_xprt *transport =
1984 container_of(work, struct sock_xprt, connect_worker.work);
1985 struct rpc_xprt *xprt = &transport->xprt;
Chuck Lever68e220b2007-08-06 11:57:48 -04001986
Trond Myklebustb61d59f2009-03-11 14:38:04 -04001987 xs_tcp_setup_socket(xprt, transport, xs_create_tcp_sock6);
Chuck Lever68e220b2007-08-06 11:57:48 -04001988}
1989
1990/**
Chuck Lever9903cd12005-08-11 16:25:26 -04001991 * xs_connect - connect a socket to a remote endpoint
1992 * @task: address of RPC task that manages state of connect request
1993 *
1994 * TCP: If the remote end dropped the connection, delay reconnecting.
Chuck Lever03bf4b72005-08-25 16:25:55 -07001995 *
1996 * UDP socket connects are synchronous, but we use a work queue anyway
1997 * to guarantee that even unprivileged user processes can set up a
1998 * socket on a privileged port.
1999 *
2000 * If a UDP socket connect fails, the delay behavior here prevents
2001 * retry floods (hard mounts).
Chuck Lever9903cd12005-08-11 16:25:26 -04002002 */
2003static void xs_connect(struct rpc_task *task)
Chuck Levera246b012005-08-11 16:25:23 -04002004{
2005 struct rpc_xprt *xprt = task->tk_xprt;
Chuck Leveree0ac0c2006-12-05 16:35:15 -05002006 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
Chuck Levera246b012005-08-11 16:25:23 -04002007
Chuck Lever09a21c42009-12-03 15:58:56 -05002008 if (transport->sock != NULL && !RPC_IS_SOFTCONN(task)) {
Chuck Lever46121cf2007-01-31 12:14:08 -05002009 dprintk("RPC: xs_connect delayed xprt %p for %lu "
2010 "seconds\n",
Chuck Lever03bf4b72005-08-25 16:25:55 -07002011 xprt, xprt->reestablish_timeout / HZ);
Trond Myklebustc1384c92007-06-14 18:00:42 -04002012 queue_delayed_work(rpciod_workqueue,
2013 &transport->connect_worker,
2014 xprt->reestablish_timeout);
Chuck Lever03bf4b72005-08-25 16:25:55 -07002015 xprt->reestablish_timeout <<= 1;
Neil Brown61d0a8e2009-09-23 14:36:37 -04002016 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2017 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
Chuck Lever03bf4b72005-08-25 16:25:55 -07002018 if (xprt->reestablish_timeout > XS_TCP_MAX_REEST_TO)
2019 xprt->reestablish_timeout = XS_TCP_MAX_REEST_TO;
Chuck Leverb0d93ad2005-08-11 16:25:53 -04002020 } else {
Chuck Lever46121cf2007-01-31 12:14:08 -05002021 dprintk("RPC: xs_connect scheduled xprt %p\n", xprt);
Trond Myklebustc1384c92007-06-14 18:00:42 -04002022 queue_delayed_work(rpciod_workqueue,
2023 &transport->connect_worker, 0);
Chuck Levera246b012005-08-11 16:25:23 -04002024 }
2025}
2026
Chuck Lever262ca072006-03-20 13:44:16 -05002027/**
2028 * xs_udp_print_stats - display UDP socket-specifc stats
2029 * @xprt: rpc_xprt struct containing statistics
2030 * @seq: output file
2031 *
2032 */
2033static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2034{
Chuck Leverc8475462006-12-05 16:35:26 -05002035 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2036
Chuck Lever262ca072006-03-20 13:44:16 -05002037 seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %Lu %Lu\n",
Chuck Leverfbfffbd2009-08-09 15:09:46 -04002038 transport->srcport,
Chuck Lever262ca072006-03-20 13:44:16 -05002039 xprt->stat.bind_count,
2040 xprt->stat.sends,
2041 xprt->stat.recvs,
2042 xprt->stat.bad_xids,
2043 xprt->stat.req_u,
2044 xprt->stat.bklog_u);
2045}
2046
2047/**
2048 * xs_tcp_print_stats - display TCP socket-specifc stats
2049 * @xprt: rpc_xprt struct containing statistics
2050 * @seq: output file
2051 *
2052 */
2053static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2054{
Chuck Leverc8475462006-12-05 16:35:26 -05002055 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
Chuck Lever262ca072006-03-20 13:44:16 -05002056 long idle_time = 0;
2057
2058 if (xprt_connected(xprt))
2059 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2060
2061 seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu %Lu %Lu\n",
Chuck Leverfbfffbd2009-08-09 15:09:46 -04002062 transport->srcport,
Chuck Lever262ca072006-03-20 13:44:16 -05002063 xprt->stat.bind_count,
2064 xprt->stat.connect_count,
2065 xprt->stat.connect_time,
2066 idle_time,
2067 xprt->stat.sends,
2068 xprt->stat.recvs,
2069 xprt->stat.bad_xids,
2070 xprt->stat.req_u,
2071 xprt->stat.bklog_u);
2072}
2073
Rahul Iyer4cfc7e62009-09-10 17:32:28 +03002074/*
2075 * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason
2076 * we allocate pages instead doing a kmalloc like rpc_malloc is because we want
2077 * to use the server side send routines.
2078 */
H Hartley Sweeten5a51f132010-01-14 15:38:31 -07002079static void *bc_malloc(struct rpc_task *task, size_t size)
Rahul Iyer4cfc7e62009-09-10 17:32:28 +03002080{
2081 struct page *page;
2082 struct rpc_buffer *buf;
2083
2084 BUG_ON(size > PAGE_SIZE - sizeof(struct rpc_buffer));
2085 page = alloc_page(GFP_KERNEL);
2086
2087 if (!page)
2088 return NULL;
2089
2090 buf = page_address(page);
2091 buf->len = PAGE_SIZE;
2092
2093 return buf->data;
2094}
2095
2096/*
2097 * Free the space allocated in the bc_alloc routine
2098 */
H Hartley Sweeten5a51f132010-01-14 15:38:31 -07002099static void bc_free(void *buffer)
Rahul Iyer4cfc7e62009-09-10 17:32:28 +03002100{
2101 struct rpc_buffer *buf;
2102
2103 if (!buffer)
2104 return;
2105
2106 buf = container_of(buffer, struct rpc_buffer, data);
2107 free_page((unsigned long)buf);
2108}
2109
2110/*
2111 * Use the svc_sock to send the callback. Must be called with svsk->sk_mutex
2112 * held. Borrows heavily from svc_tcp_sendto and xs_tcp_send_request.
2113 */
2114static int bc_sendto(struct rpc_rqst *req)
2115{
2116 int len;
2117 struct xdr_buf *xbufp = &req->rq_snd_buf;
2118 struct rpc_xprt *xprt = req->rq_xprt;
2119 struct sock_xprt *transport =
2120 container_of(xprt, struct sock_xprt, xprt);
2121 struct socket *sock = transport->sock;
2122 unsigned long headoff;
2123 unsigned long tailoff;
2124
2125 /*
2126 * Set up the rpc header and record marker stuff
2127 */
2128 xs_encode_tcp_record_marker(xbufp);
2129
2130 tailoff = (unsigned long)xbufp->tail[0].iov_base & ~PAGE_MASK;
2131 headoff = (unsigned long)xbufp->head[0].iov_base & ~PAGE_MASK;
2132 len = svc_send_common(sock, xbufp,
2133 virt_to_page(xbufp->head[0].iov_base), headoff,
2134 xbufp->tail[0].iov_base, tailoff);
2135
2136 if (len != xbufp->len) {
2137 printk(KERN_NOTICE "Error sending entire callback!\n");
2138 len = -EAGAIN;
2139 }
2140
2141 return len;
2142}
2143
2144/*
2145 * The send routine. Borrows from svc_send
2146 */
2147static int bc_send_request(struct rpc_task *task)
2148{
2149 struct rpc_rqst *req = task->tk_rqstp;
2150 struct svc_xprt *xprt;
2151 struct svc_sock *svsk;
2152 u32 len;
2153
2154 dprintk("sending request with xid: %08x\n", ntohl(req->rq_xid));
2155 /*
2156 * Get the server socket associated with this callback xprt
2157 */
2158 xprt = req->rq_xprt->bc_xprt;
2159 svsk = container_of(xprt, struct svc_sock, sk_xprt);
2160
2161 /*
2162 * Grab the mutex to serialize data as the connection is shared
2163 * with the fore channel
2164 */
2165 if (!mutex_trylock(&xprt->xpt_mutex)) {
2166 rpc_sleep_on(&xprt->xpt_bc_pending, task, NULL);
2167 if (!mutex_trylock(&xprt->xpt_mutex))
2168 return -EAGAIN;
2169 rpc_wake_up_queued_task(&xprt->xpt_bc_pending, task);
2170 }
2171 if (test_bit(XPT_DEAD, &xprt->xpt_flags))
2172 len = -ENOTCONN;
2173 else
2174 len = bc_sendto(req);
2175 mutex_unlock(&xprt->xpt_mutex);
2176
2177 if (len > 0)
2178 len = 0;
2179
2180 return len;
2181}
2182
2183/*
2184 * The close routine. Since this is client initiated, we do nothing
2185 */
2186
2187static void bc_close(struct rpc_xprt *xprt)
2188{
Rahul Iyer4cfc7e62009-09-10 17:32:28 +03002189}
2190
2191/*
2192 * The xprt destroy routine. Again, because this connection is client
2193 * initiated, we do nothing
2194 */
2195
2196static void bc_destroy(struct rpc_xprt *xprt)
2197{
Rahul Iyer4cfc7e62009-09-10 17:32:28 +03002198}
2199
Chuck Lever262965f2005-08-11 16:25:56 -04002200static struct rpc_xprt_ops xs_udp_ops = {
Chuck Lever43118c22005-08-25 16:25:49 -07002201 .set_buffer_size = xs_udp_set_buffer_size,
Chuck Lever12a80462005-08-25 16:25:51 -07002202 .reserve_xprt = xprt_reserve_xprt_cong,
Chuck Lever49e9a892005-08-25 16:25:51 -07002203 .release_xprt = xprt_release_xprt_cong,
Chuck Lever45160d62007-07-01 12:13:17 -04002204 .rpcbind = rpcb_getport_async,
Chuck Lever92200412006-01-03 09:55:51 +01002205 .set_port = xs_set_port,
Chuck Lever9903cd12005-08-11 16:25:26 -04002206 .connect = xs_connect,
Chuck Lever02107142006-01-03 09:55:49 +01002207 .buf_alloc = rpc_malloc,
2208 .buf_free = rpc_free,
Chuck Lever262965f2005-08-11 16:25:56 -04002209 .send_request = xs_udp_send_request,
Chuck Leverfe3aca22005-08-25 16:25:50 -07002210 .set_retrans_timeout = xprt_set_retrans_timeout_rtt,
Chuck Lever46c0ee82005-08-25 16:25:52 -07002211 .timer = xs_udp_timer,
Chuck Levera58dd392005-08-25 16:25:53 -07002212 .release_request = xprt_release_rqst_cong,
Chuck Lever262965f2005-08-11 16:25:56 -04002213 .close = xs_close,
2214 .destroy = xs_destroy,
Chuck Lever262ca072006-03-20 13:44:16 -05002215 .print_stats = xs_udp_print_stats,
Chuck Lever262965f2005-08-11 16:25:56 -04002216};
2217
2218static struct rpc_xprt_ops xs_tcp_ops = {
Chuck Lever12a80462005-08-25 16:25:51 -07002219 .reserve_xprt = xprt_reserve_xprt,
Trond Myklebuste0ab53d2006-07-27 17:22:50 -04002220 .release_xprt = xs_tcp_release_xprt,
Chuck Lever45160d62007-07-01 12:13:17 -04002221 .rpcbind = rpcb_getport_async,
Chuck Lever92200412006-01-03 09:55:51 +01002222 .set_port = xs_set_port,
Trond Myklebust0b9e7942010-04-16 16:41:57 -04002223 .connect = xs_connect,
Chuck Lever02107142006-01-03 09:55:49 +01002224 .buf_alloc = rpc_malloc,
2225 .buf_free = rpc_free,
Chuck Lever262965f2005-08-11 16:25:56 -04002226 .send_request = xs_tcp_send_request,
Chuck Leverfe3aca22005-08-25 16:25:50 -07002227 .set_retrans_timeout = xprt_set_retrans_timeout_def,
Trond Myklebustf75e6742009-04-21 17:18:20 -04002228 .close = xs_tcp_close,
Chuck Lever9903cd12005-08-11 16:25:26 -04002229 .destroy = xs_destroy,
Chuck Lever262ca072006-03-20 13:44:16 -05002230 .print_stats = xs_tcp_print_stats,
Chuck Levera246b012005-08-11 16:25:23 -04002231};
2232
Rahul Iyer4cfc7e62009-09-10 17:32:28 +03002233/*
2234 * The rpc_xprt_ops for the server backchannel
2235 */
2236
2237static struct rpc_xprt_ops bc_tcp_ops = {
2238 .reserve_xprt = xprt_reserve_xprt,
2239 .release_xprt = xprt_release_xprt,
2240 .buf_alloc = bc_malloc,
2241 .buf_free = bc_free,
2242 .send_request = bc_send_request,
2243 .set_retrans_timeout = xprt_set_retrans_timeout_def,
2244 .close = bc_close,
2245 .destroy = bc_destroy,
2246 .print_stats = xs_tcp_print_stats,
2247};
2248
\"Talpey, Thomas\3c341b0b2007-09-10 13:47:07 -04002249static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
\"Talpey, Thomas\bc255712007-09-10 13:46:39 -04002250 unsigned int slot_table_size)
Chuck Leverc8541ec2006-10-17 14:44:27 -04002251{
2252 struct rpc_xprt *xprt;
Chuck Leverffc2e512006-12-05 16:35:11 -05002253 struct sock_xprt *new;
Chuck Leverc8541ec2006-10-17 14:44:27 -04002254
Frank van Maarseveen96802a02007-07-08 13:08:54 +02002255 if (args->addrlen > sizeof(xprt->addr)) {
Chuck Lever46121cf2007-01-31 12:14:08 -05002256 dprintk("RPC: xs_setup_xprt: address too large\n");
Chuck Leverc8541ec2006-10-17 14:44:27 -04002257 return ERR_PTR(-EBADF);
2258 }
2259
Chuck Leverffc2e512006-12-05 16:35:11 -05002260 new = kzalloc(sizeof(*new), GFP_KERNEL);
2261 if (new == NULL) {
Chuck Lever46121cf2007-01-31 12:14:08 -05002262 dprintk("RPC: xs_setup_xprt: couldn't allocate "
2263 "rpc_xprt\n");
Chuck Leverc8541ec2006-10-17 14:44:27 -04002264 return ERR_PTR(-ENOMEM);
2265 }
Chuck Leverffc2e512006-12-05 16:35:11 -05002266 xprt = &new->xprt;
Chuck Leverc8541ec2006-10-17 14:44:27 -04002267
2268 xprt->max_reqs = slot_table_size;
2269 xprt->slot = kcalloc(xprt->max_reqs, sizeof(struct rpc_rqst), GFP_KERNEL);
2270 if (xprt->slot == NULL) {
2271 kfree(xprt);
Chuck Lever46121cf2007-01-31 12:14:08 -05002272 dprintk("RPC: xs_setup_xprt: couldn't allocate slot "
2273 "table\n");
Chuck Leverc8541ec2006-10-17 14:44:27 -04002274 return ERR_PTR(-ENOMEM);
2275 }
2276
Frank van Maarseveen96802a02007-07-08 13:08:54 +02002277 memcpy(&xprt->addr, args->dstaddr, args->addrlen);
2278 xprt->addrlen = args->addrlen;
Frank van Maarseveend3bc9a12007-07-09 22:23:35 +02002279 if (args->srcaddr)
Chuck Leverfbfffbd2009-08-09 15:09:46 -04002280 memcpy(&new->srcaddr, args->srcaddr, args->addrlen);
Chuck Leverc8541ec2006-10-17 14:44:27 -04002281
2282 return xprt;
2283}
2284
Trond Myklebust2881ae72007-12-20 16:03:54 -05002285static const struct rpc_timeout xs_udp_default_timeout = {
2286 .to_initval = 5 * HZ,
2287 .to_maxval = 30 * HZ,
2288 .to_increment = 5 * HZ,
2289 .to_retries = 5,
2290};
2291
Chuck Lever9903cd12005-08-11 16:25:26 -04002292/**
2293 * xs_setup_udp - Set up transport to use a UDP socket
Frank van Maarseveen96802a02007-07-08 13:08:54 +02002294 * @args: rpc transport creation arguments
Chuck Lever9903cd12005-08-11 16:25:26 -04002295 *
2296 */
Adrian Bunk483066d2007-10-24 18:24:02 +02002297static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
Chuck Levera246b012005-08-11 16:25:23 -04002298{
Chuck Lever8f9d5b12007-08-06 11:57:53 -04002299 struct sockaddr *addr = args->dstaddr;
Chuck Leverc8541ec2006-10-17 14:44:27 -04002300 struct rpc_xprt *xprt;
Chuck Leverc8475462006-12-05 16:35:26 -05002301 struct sock_xprt *transport;
J. Bruce Fields0a68b0b2010-05-26 08:42:24 -04002302 struct rpc_xprt *ret;
Chuck Levera246b012005-08-11 16:25:23 -04002303
Frank van Maarseveen96802a02007-07-08 13:08:54 +02002304 xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries);
Chuck Leverc8541ec2006-10-17 14:44:27 -04002305 if (IS_ERR(xprt))
2306 return xprt;
Chuck Leverc8475462006-12-05 16:35:26 -05002307 transport = container_of(xprt, struct sock_xprt, xprt);
Chuck Levera246b012005-08-11 16:25:23 -04002308
Chuck Leverec739ef2006-08-22 20:06:15 -04002309 xprt->prot = IPPROTO_UDP;
Chuck Lever808012f2005-08-25 16:25:49 -07002310 xprt->tsh_size = 0;
Chuck Levera246b012005-08-11 16:25:23 -04002311 /* XXX: header size can vary due to auth type, IPv6, etc. */
2312 xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
2313
Chuck Lever03bf4b72005-08-25 16:25:55 -07002314 xprt->bind_timeout = XS_BIND_TO;
Chuck Lever03bf4b72005-08-25 16:25:55 -07002315 xprt->reestablish_timeout = XS_UDP_REEST_TO;
2316 xprt->idle_timeout = XS_IDLE_DISC_TO;
Chuck Levera246b012005-08-11 16:25:23 -04002317
Chuck Lever262965f2005-08-11 16:25:56 -04002318 xprt->ops = &xs_udp_ops;
Chuck Levera246b012005-08-11 16:25:23 -04002319
Trond Myklebustba7392b2007-12-20 16:03:55 -05002320 xprt->timeout = &xs_udp_default_timeout;
Chuck Levera246b012005-08-11 16:25:23 -04002321
Chuck Lever8f9d5b12007-08-06 11:57:53 -04002322 switch (addr->sa_family) {
2323 case AF_INET:
2324 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2325 xprt_set_bound(xprt);
2326
2327 INIT_DELAYED_WORK(&transport->connect_worker,
2328 xs_udp_connect_worker4);
Chuck Lever9dc3b092009-08-09 15:09:46 -04002329 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
Chuck Lever8f9d5b12007-08-06 11:57:53 -04002330 break;
2331 case AF_INET6:
2332 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2333 xprt_set_bound(xprt);
2334
2335 INIT_DELAYED_WORK(&transport->connect_worker,
2336 xs_udp_connect_worker6);
Chuck Lever9dc3b092009-08-09 15:09:46 -04002337 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
Chuck Lever8f9d5b12007-08-06 11:57:53 -04002338 break;
2339 default:
J. Bruce Fields0a68b0b2010-05-26 08:42:24 -04002340 ret = ERR_PTR(-EAFNOSUPPORT);
2341 goto out_err;
Chuck Lever8f9d5b12007-08-06 11:57:53 -04002342 }
2343
Chuck Leverc740eff2009-08-09 15:09:46 -04002344 if (xprt_bound(xprt))
2345 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2346 xprt->address_strings[RPC_DISPLAY_ADDR],
2347 xprt->address_strings[RPC_DISPLAY_PORT],
2348 xprt->address_strings[RPC_DISPLAY_PROTO]);
2349 else
2350 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
2351 xprt->address_strings[RPC_DISPLAY_ADDR],
2352 xprt->address_strings[RPC_DISPLAY_PROTO]);
Chuck Leveredb267a2006-08-22 20:06:18 -04002353
\"Talpey, Thomas\bc255712007-09-10 13:46:39 -04002354 if (try_module_get(THIS_MODULE))
2355 return xprt;
J. Bruce Fields0a68b0b2010-05-26 08:42:24 -04002356 ret = ERR_PTR(-EINVAL);
2357out_err:
\"Talpey, Thomas\bc255712007-09-10 13:46:39 -04002358 kfree(xprt->slot);
2359 kfree(xprt);
J. Bruce Fields0a68b0b2010-05-26 08:42:24 -04002360 return ret;
Chuck Levera246b012005-08-11 16:25:23 -04002361}
2362
Trond Myklebust2881ae72007-12-20 16:03:54 -05002363static const struct rpc_timeout xs_tcp_default_timeout = {
2364 .to_initval = 60 * HZ,
2365 .to_maxval = 60 * HZ,
2366 .to_retries = 2,
2367};
2368
Chuck Lever9903cd12005-08-11 16:25:26 -04002369/**
2370 * xs_setup_tcp - Set up transport to use a TCP socket
Frank van Maarseveen96802a02007-07-08 13:08:54 +02002371 * @args: rpc transport creation arguments
Chuck Lever9903cd12005-08-11 16:25:26 -04002372 *
2373 */
Adrian Bunk483066d2007-10-24 18:24:02 +02002374static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
Chuck Levera246b012005-08-11 16:25:23 -04002375{
Chuck Lever8f9d5b12007-08-06 11:57:53 -04002376 struct sockaddr *addr = args->dstaddr;
Chuck Leverc8541ec2006-10-17 14:44:27 -04002377 struct rpc_xprt *xprt;
Chuck Leverc8475462006-12-05 16:35:26 -05002378 struct sock_xprt *transport;
J. Bruce Fields0a68b0b2010-05-26 08:42:24 -04002379 struct rpc_xprt *ret;
Chuck Levera246b012005-08-11 16:25:23 -04002380
Frank van Maarseveen96802a02007-07-08 13:08:54 +02002381 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries);
Chuck Leverc8541ec2006-10-17 14:44:27 -04002382 if (IS_ERR(xprt))
2383 return xprt;
Chuck Leverc8475462006-12-05 16:35:26 -05002384 transport = container_of(xprt, struct sock_xprt, xprt);
Chuck Levera246b012005-08-11 16:25:23 -04002385
Chuck Leverec739ef2006-08-22 20:06:15 -04002386 xprt->prot = IPPROTO_TCP;
Chuck Lever808012f2005-08-25 16:25:49 -07002387 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
Chuck Lever808012f2005-08-25 16:25:49 -07002388 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
Chuck Levera246b012005-08-11 16:25:23 -04002389
Chuck Lever03bf4b72005-08-25 16:25:55 -07002390 xprt->bind_timeout = XS_BIND_TO;
Chuck Lever03bf4b72005-08-25 16:25:55 -07002391 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2392 xprt->idle_timeout = XS_IDLE_DISC_TO;
Chuck Levera246b012005-08-11 16:25:23 -04002393
Chuck Lever262965f2005-08-11 16:25:56 -04002394 xprt->ops = &xs_tcp_ops;
Trond Myklebustba7392b2007-12-20 16:03:55 -05002395 xprt->timeout = &xs_tcp_default_timeout;
Chuck Levera246b012005-08-11 16:25:23 -04002396
Chuck Lever8f9d5b12007-08-06 11:57:53 -04002397 switch (addr->sa_family) {
2398 case AF_INET:
2399 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2400 xprt_set_bound(xprt);
2401
Chuck Lever9dc3b092009-08-09 15:09:46 -04002402 INIT_DELAYED_WORK(&transport->connect_worker,
2403 xs_tcp_connect_worker4);
2404 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
Chuck Lever8f9d5b12007-08-06 11:57:53 -04002405 break;
2406 case AF_INET6:
2407 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2408 xprt_set_bound(xprt);
2409
Chuck Lever9dc3b092009-08-09 15:09:46 -04002410 INIT_DELAYED_WORK(&transport->connect_worker,
2411 xs_tcp_connect_worker6);
2412 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
Chuck Lever8f9d5b12007-08-06 11:57:53 -04002413 break;
2414 default:
J. Bruce Fields0a68b0b2010-05-26 08:42:24 -04002415 ret = ERR_PTR(-EAFNOSUPPORT);
2416 goto out_err;
Chuck Lever8f9d5b12007-08-06 11:57:53 -04002417 }
2418
Chuck Leverc740eff2009-08-09 15:09:46 -04002419 if (xprt_bound(xprt))
2420 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2421 xprt->address_strings[RPC_DISPLAY_ADDR],
2422 xprt->address_strings[RPC_DISPLAY_PORT],
2423 xprt->address_strings[RPC_DISPLAY_PROTO]);
2424 else
2425 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
2426 xprt->address_strings[RPC_DISPLAY_ADDR],
2427 xprt->address_strings[RPC_DISPLAY_PROTO]);
2428
Chuck Leveredb267a2006-08-22 20:06:18 -04002429
\"Talpey, Thomas\bc255712007-09-10 13:46:39 -04002430 if (try_module_get(THIS_MODULE))
2431 return xprt;
J. Bruce Fields0a68b0b2010-05-26 08:42:24 -04002432 ret = ERR_PTR(-EINVAL);
2433out_err:
\"Talpey, Thomas\bc255712007-09-10 13:46:39 -04002434 kfree(xprt->slot);
2435 kfree(xprt);
J. Bruce Fields0a68b0b2010-05-26 08:42:24 -04002436 return ret;
Chuck Levera246b012005-08-11 16:25:23 -04002437}
Chuck Lever282b32e2006-12-05 16:35:51 -05002438
Alexandros Batsakisf300bab2009-09-10 17:33:30 +03002439/**
2440 * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket
2441 * @args: rpc transport creation arguments
2442 *
2443 */
2444static struct rpc_xprt *xs_setup_bc_tcp(struct xprt_create *args)
2445{
2446 struct sockaddr *addr = args->dstaddr;
2447 struct rpc_xprt *xprt;
2448 struct sock_xprt *transport;
2449 struct svc_sock *bc_sock;
J. Bruce Fields0a68b0b2010-05-26 08:42:24 -04002450 struct rpc_xprt *ret;
Alexandros Batsakisf300bab2009-09-10 17:33:30 +03002451
Alexandros Batsakisf300bab2009-09-10 17:33:30 +03002452 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries);
2453 if (IS_ERR(xprt))
2454 return xprt;
2455 transport = container_of(xprt, struct sock_xprt, xprt);
2456
2457 xprt->prot = IPPROTO_TCP;
2458 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2459 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2460 xprt->timeout = &xs_tcp_default_timeout;
2461
2462 /* backchannel */
2463 xprt_set_bound(xprt);
2464 xprt->bind_timeout = 0;
Alexandros Batsakisf300bab2009-09-10 17:33:30 +03002465 xprt->reestablish_timeout = 0;
2466 xprt->idle_timeout = 0;
2467
2468 /*
2469 * The backchannel uses the same socket connection as the
2470 * forechannel
2471 */
2472 xprt->bc_xprt = args->bc_xprt;
2473 bc_sock = container_of(args->bc_xprt, struct svc_sock, sk_xprt);
2474 bc_sock->sk_bc_xprt = xprt;
2475 transport->sock = bc_sock->sk_sock;
2476 transport->inet = bc_sock->sk_sk;
2477
2478 xprt->ops = &bc_tcp_ops;
2479
2480 switch (addr->sa_family) {
2481 case AF_INET:
2482 xs_format_peer_addresses(xprt, "tcp",
2483 RPCBIND_NETID_TCP);
2484 break;
2485 case AF_INET6:
2486 xs_format_peer_addresses(xprt, "tcp",
2487 RPCBIND_NETID_TCP6);
2488 break;
2489 default:
J. Bruce Fields0a68b0b2010-05-26 08:42:24 -04002490 ret = ERR_PTR(-EAFNOSUPPORT);
2491 goto out_err;
Alexandros Batsakisf300bab2009-09-10 17:33:30 +03002492 }
2493
2494 if (xprt_bound(xprt))
2495 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2496 xprt->address_strings[RPC_DISPLAY_ADDR],
2497 xprt->address_strings[RPC_DISPLAY_PORT],
2498 xprt->address_strings[RPC_DISPLAY_PROTO]);
2499 else
2500 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
2501 xprt->address_strings[RPC_DISPLAY_ADDR],
2502 xprt->address_strings[RPC_DISPLAY_PROTO]);
2503
2504 /*
2505 * Since we don't want connections for the backchannel, we set
2506 * the xprt status to connected
2507 */
2508 xprt_set_connected(xprt);
2509
2510
2511 if (try_module_get(THIS_MODULE))
2512 return xprt;
J. Bruce Fields0a68b0b2010-05-26 08:42:24 -04002513 ret = ERR_PTR(-EINVAL);
2514out_err:
Alexandros Batsakisf300bab2009-09-10 17:33:30 +03002515 kfree(xprt->slot);
2516 kfree(xprt);
J. Bruce Fields0a68b0b2010-05-26 08:42:24 -04002517 return ret;
Alexandros Batsakisf300bab2009-09-10 17:33:30 +03002518}
2519
\"Talpey, Thomas\bc255712007-09-10 13:46:39 -04002520static struct xprt_class xs_udp_transport = {
2521 .list = LIST_HEAD_INIT(xs_udp_transport.list),
2522 .name = "udp",
2523 .owner = THIS_MODULE,
Alexandros Batsakisf300bab2009-09-10 17:33:30 +03002524 .ident = XPRT_TRANSPORT_UDP,
\"Talpey, Thomas\bc255712007-09-10 13:46:39 -04002525 .setup = xs_setup_udp,
2526};
2527
2528static struct xprt_class xs_tcp_transport = {
2529 .list = LIST_HEAD_INIT(xs_tcp_transport.list),
2530 .name = "tcp",
2531 .owner = THIS_MODULE,
Alexandros Batsakisf300bab2009-09-10 17:33:30 +03002532 .ident = XPRT_TRANSPORT_TCP,
\"Talpey, Thomas\bc255712007-09-10 13:46:39 -04002533 .setup = xs_setup_tcp,
2534};
2535
Alexandros Batsakisf300bab2009-09-10 17:33:30 +03002536static struct xprt_class xs_bc_tcp_transport = {
2537 .list = LIST_HEAD_INIT(xs_bc_tcp_transport.list),
2538 .name = "tcp NFSv4.1 backchannel",
2539 .owner = THIS_MODULE,
2540 .ident = XPRT_TRANSPORT_BC_TCP,
2541 .setup = xs_setup_bc_tcp,
2542};
2543
Chuck Lever282b32e2006-12-05 16:35:51 -05002544/**
\"Talpey, Thomas\bc255712007-09-10 13:46:39 -04002545 * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
Chuck Lever282b32e2006-12-05 16:35:51 -05002546 *
2547 */
2548int init_socket_xprt(void)
2549{
Chuck Leverfbf76682006-12-05 16:35:54 -05002550#ifdef RPC_DEBUG
Eric W. Biederman2b1bec52007-02-14 00:33:24 -08002551 if (!sunrpc_table_header)
Eric W. Biederman0b4d4142007-02-14 00:34:09 -08002552 sunrpc_table_header = register_sysctl_table(sunrpc_table);
Chuck Leverfbf76682006-12-05 16:35:54 -05002553#endif
2554
\"Talpey, Thomas\bc255712007-09-10 13:46:39 -04002555 xprt_register_transport(&xs_udp_transport);
2556 xprt_register_transport(&xs_tcp_transport);
Alexandros Batsakisf300bab2009-09-10 17:33:30 +03002557 xprt_register_transport(&xs_bc_tcp_transport);
\"Talpey, Thomas\bc255712007-09-10 13:46:39 -04002558
Chuck Lever282b32e2006-12-05 16:35:51 -05002559 return 0;
2560}
2561
2562/**
\"Talpey, Thomas\bc255712007-09-10 13:46:39 -04002563 * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
Chuck Lever282b32e2006-12-05 16:35:51 -05002564 *
2565 */
2566void cleanup_socket_xprt(void)
2567{
Chuck Leverfbf76682006-12-05 16:35:54 -05002568#ifdef RPC_DEBUG
2569 if (sunrpc_table_header) {
2570 unregister_sysctl_table(sunrpc_table_header);
2571 sunrpc_table_header = NULL;
2572 }
2573#endif
\"Talpey, Thomas\bc255712007-09-10 13:46:39 -04002574
2575 xprt_unregister_transport(&xs_udp_transport);
2576 xprt_unregister_transport(&xs_tcp_transport);
Alexandros Batsakisf300bab2009-09-10 17:33:30 +03002577 xprt_unregister_transport(&xs_bc_tcp_transport);
Chuck Lever282b32e2006-12-05 16:35:51 -05002578}
Trond Myklebustcbf11072009-08-09 15:06:19 -04002579
2580static int param_set_uint_minmax(const char *val, struct kernel_param *kp,
2581 unsigned int min, unsigned int max)
2582{
2583 unsigned long num;
2584 int ret;
2585
2586 if (!val)
2587 return -EINVAL;
2588 ret = strict_strtoul(val, 0, &num);
2589 if (ret == -EINVAL || num < min || num > max)
2590 return -EINVAL;
2591 *((unsigned int *)kp->arg) = num;
2592 return 0;
2593}
2594
2595static int param_set_portnr(const char *val, struct kernel_param *kp)
2596{
2597 return param_set_uint_minmax(val, kp,
2598 RPC_MIN_RESVPORT,
2599 RPC_MAX_RESVPORT);
2600}
2601
2602static int param_get_portnr(char *buffer, struct kernel_param *kp)
2603{
2604 return param_get_uint(buffer, kp);
2605}
2606#define param_check_portnr(name, p) \
2607 __param_check(name, p, unsigned int);
2608
2609module_param_named(min_resvport, xprt_min_resvport, portnr, 0644);
2610module_param_named(max_resvport, xprt_max_resvport, portnr, 0644);
2611
2612static int param_set_slot_table_size(const char *val, struct kernel_param *kp)
2613{
2614 return param_set_uint_minmax(val, kp,
2615 RPC_MIN_SLOT_TABLE,
2616 RPC_MAX_SLOT_TABLE);
2617}
2618
2619static int param_get_slot_table_size(char *buffer, struct kernel_param *kp)
2620{
2621 return param_get_uint(buffer, kp);
2622}
2623#define param_check_slot_table_size(name, p) \
2624 __param_check(name, p, unsigned int);
2625
2626module_param_named(tcp_slot_table_entries, xprt_tcp_slot_table_entries,
2627 slot_table_size, 0644);
2628module_param_named(udp_slot_table_entries, xprt_udp_slot_table_entries,
2629 slot_table_size, 0644);
2630