nfsd4: don't try to map gid's in generic rpc code
[linux-2.6.git] / net / sunrpc / xprtsock.c
1 /*
2  * linux/net/sunrpc/xprtsock.c
3  *
4  * Client-side transport implementation for sockets.
5  *
6  * TCP callback races fixes (C) 1998 Red Hat
7  * TCP send fixes (C) 1998 Red Hat
8  * TCP NFS related read + write fixes
9  *  (C) 1999 Dave Airlie, University of Limerick, Ireland <airlied@linux.ie>
10  *
11  * Rewrite of larges part of the code in order to stabilize TCP stuff.
12  * Fix behaviour when socket buffer is full.
13  *  (C) 1999 Trond Myklebust <trond.myklebust@fys.uio.no>
14  *
15  * IP socket transport implementation, (C) 2005 Chuck Lever <cel@netapp.com>
16  *
17  * IPv6 support contributed by Gilles Quillard, Bull Open Source, 2005.
18  *   <gilles.quillard@bull.net>
19  */
20
21 #include <linux/types.h>
22 #include <linux/slab.h>
23 #include <linux/module.h>
24 #include <linux/capability.h>
25 #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>
34 #include <linux/sunrpc/sched.h>
35 #include <linux/sunrpc/svcsock.h>
36 #include <linux/sunrpc/xprtsock.h>
37 #include <linux/file.h>
38 #ifdef CONFIG_NFS_V4_1
39 #include <linux/sunrpc/bc_xprt.h>
40 #endif
41
42 #include <net/sock.h>
43 #include <net/checksum.h>
44 #include <net/udp.h>
45 #include <net/tcp.h>
46
47 #include "sunrpc.h"
48 /*
49  * xprtsock tunables
50  */
51 unsigned int xprt_udp_slot_table_entries = RPC_DEF_SLOT_TABLE;
52 unsigned int xprt_tcp_slot_table_entries = RPC_DEF_SLOT_TABLE;
53
54 unsigned int xprt_min_resvport = RPC_DEF_MIN_RESVPORT;
55 unsigned int xprt_max_resvport = RPC_DEF_MAX_RESVPORT;
56
57 #define XS_TCP_LINGER_TO        (15U * HZ)
58 static unsigned int xs_tcp_fin_timeout __read_mostly = XS_TCP_LINGER_TO;
59
60 /*
61  * 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
71 static unsigned int min_slot_table_size = RPC_MIN_SLOT_TABLE;
72 static unsigned int max_slot_table_size = RPC_MAX_SLOT_TABLE;
73 static unsigned int xprt_min_resvport_limit = RPC_MIN_RESVPORT;
74 static unsigned int xprt_max_resvport_limit = RPC_MAX_RESVPORT;
75
76 static 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  */
82 static ctl_table xs_tunables_table[] = {
83         {
84                 .ctl_name       = CTL_SLOTTABLE_UDP,
85                 .procname       = "udp_slot_table_entries",
86                 .data           = &xprt_udp_slot_table_entries,
87                 .maxlen         = sizeof(unsigned int),
88                 .mode           = 0644,
89                 .proc_handler   = &proc_dointvec_minmax,
90                 .strategy       = &sysctl_intvec,
91                 .extra1         = &min_slot_table_size,
92                 .extra2         = &max_slot_table_size
93         },
94         {
95                 .ctl_name       = CTL_SLOTTABLE_TCP,
96                 .procname       = "tcp_slot_table_entries",
97                 .data           = &xprt_tcp_slot_table_entries,
98                 .maxlen         = sizeof(unsigned int),
99                 .mode           = 0644,
100                 .proc_handler   = &proc_dointvec_minmax,
101                 .strategy       = &sysctl_intvec,
102                 .extra1         = &min_slot_table_size,
103                 .extra2         = &max_slot_table_size
104         },
105         {
106                 .ctl_name       = CTL_MIN_RESVPORT,
107                 .procname       = "min_resvport",
108                 .data           = &xprt_min_resvport,
109                 .maxlen         = sizeof(unsigned int),
110                 .mode           = 0644,
111                 .proc_handler   = &proc_dointvec_minmax,
112                 .strategy       = &sysctl_intvec,
113                 .extra1         = &xprt_min_resvport_limit,
114                 .extra2         = &xprt_max_resvport_limit
115         },
116         {
117                 .ctl_name       = CTL_MAX_RESVPORT,
118                 .procname       = "max_resvport",
119                 .data           = &xprt_max_resvport,
120                 .maxlen         = sizeof(unsigned int),
121                 .mode           = 0644,
122                 .proc_handler   = &proc_dointvec_minmax,
123                 .strategy       = &sysctl_intvec,
124                 .extra1         = &xprt_min_resvport_limit,
125                 .extra2         = &xprt_max_resvport_limit
126         },
127         {
128                 .procname       = "tcp_fin_timeout",
129                 .data           = &xs_tcp_fin_timeout,
130                 .maxlen         = sizeof(xs_tcp_fin_timeout),
131                 .mode           = 0644,
132                 .proc_handler   = &proc_dointvec_jiffies,
133                 .strategy       = sysctl_jiffies
134         },
135         {
136                 .ctl_name = 0,
137         },
138 };
139
140 static ctl_table sunrpc_table[] = {
141         {
142                 .ctl_name       = CTL_SUNRPC,
143                 .procname       = "sunrpc",
144                 .mode           = 0555,
145                 .child          = xs_tunables_table
146         },
147         {
148                 .ctl_name = 0,
149         },
150 };
151
152 #endif
153
154 /*
155  * Time out for an RPC UDP socket connect.  UDP socket connects are
156  * synchronous, but we set a timeout anyway in case of resource
157  * exhaustion on the local host.
158  */
159 #define XS_UDP_CONN_TO          (5U * HZ)
160
161 /*
162  * Wait duration for an RPC TCP connection to be established.  Solaris
163  * NFS over TCP uses 60 seconds, for example, which is in line with how
164  * long a server takes to reboot.
165  */
166 #define XS_TCP_CONN_TO          (60U * HZ)
167
168 /*
169  * Wait duration for a reply from the RPC portmapper.
170  */
171 #define XS_BIND_TO              (60U * HZ)
172
173 /*
174  * Delay if a UDP socket connect error occurs.  This is most likely some
175  * kind of resource problem on the local host.
176  */
177 #define XS_UDP_REEST_TO         (2U * HZ)
178
179 /*
180  * The reestablish timeout allows clients to delay for a bit before attempting
181  * to reconnect to a server that just dropped our connection.
182  *
183  * We implement an exponential backoff when trying to reestablish a TCP
184  * transport connection with the server.  Some servers like to drop a TCP
185  * connection when they are overworked, so we start with a short timeout and
186  * increase over time if the server is down or not responding.
187  */
188 #define XS_TCP_INIT_REEST_TO    (3U * HZ)
189 #define XS_TCP_MAX_REEST_TO     (5U * 60 * HZ)
190
191 /*
192  * TCP idle timeout; client drops the transport socket if it is idle
193  * for this long.  Note that we also timeout UDP sockets to prevent
194  * holding port numbers when there is no RPC traffic.
195  */
196 #define XS_IDLE_DISC_TO         (5U * 60 * HZ)
197
198 #ifdef RPC_DEBUG
199 # undef  RPC_DEBUG_DATA
200 # define RPCDBG_FACILITY        RPCDBG_TRANS
201 #endif
202
203 #ifdef RPC_DEBUG_DATA
204 static void xs_pktdump(char *msg, u32 *packet, unsigned int count)
205 {
206         u8 *buf = (u8 *) packet;
207         int j;
208
209         dprintk("RPC:       %s\n", msg);
210         for (j = 0; j < count && j < 128; j += 4) {
211                 if (!(j & 31)) {
212                         if (j)
213                                 dprintk("\n");
214                         dprintk("0x%04x ", j);
215                 }
216                 dprintk("%02x%02x%02x%02x ",
217                         buf[j], buf[j+1], buf[j+2], buf[j+3]);
218         }
219         dprintk("\n");
220 }
221 #else
222 static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count)
223 {
224         /* NOP */
225 }
226 #endif
227
228 struct sock_xprt {
229         struct rpc_xprt         xprt;
230
231         /*
232          * Network layer
233          */
234         struct socket *         sock;
235         struct sock *           inet;
236
237         /*
238          * State of TCP reply receive
239          */
240         __be32                  tcp_fraghdr,
241                                 tcp_xid;
242
243         u32                     tcp_offset,
244                                 tcp_reclen;
245
246         unsigned long           tcp_copied,
247                                 tcp_flags;
248
249         /*
250          * Connection of transports
251          */
252         struct delayed_work     connect_worker;
253         struct sockaddr_storage srcaddr;
254         unsigned short          srcport;
255
256         /*
257          * UDP socket buffer size parameters
258          */
259         size_t                  rcvsize,
260                                 sndsize;
261
262         /*
263          * Saved socket callback addresses
264          */
265         void                    (*old_data_ready)(struct sock *, int);
266         void                    (*old_state_change)(struct sock *);
267         void                    (*old_write_space)(struct sock *);
268         void                    (*old_error_report)(struct sock *);
269 };
270
271 /*
272  * TCP receive state flags
273  */
274 #define TCP_RCV_LAST_FRAG       (1UL << 0)
275 #define TCP_RCV_COPY_FRAGHDR    (1UL << 1)
276 #define TCP_RCV_COPY_XID        (1UL << 2)
277 #define TCP_RCV_COPY_DATA       (1UL << 3)
278 #define TCP_RCV_READ_CALLDIR    (1UL << 4)
279 #define TCP_RCV_COPY_CALLDIR    (1UL << 5)
280
281 /*
282  * TCP RPC flags
283  */
284 #define TCP_RPC_REPLY           (1UL << 6)
285
286 static inline struct sockaddr *xs_addr(struct rpc_xprt *xprt)
287 {
288         return (struct sockaddr *) &xprt->addr;
289 }
290
291 static inline struct sockaddr_in *xs_addr_in(struct rpc_xprt *xprt)
292 {
293         return (struct sockaddr_in *) &xprt->addr;
294 }
295
296 static inline struct sockaddr_in6 *xs_addr_in6(struct rpc_xprt *xprt)
297 {
298         return (struct sockaddr_in6 *) &xprt->addr;
299 }
300
301 static void xs_format_common_peer_addresses(struct rpc_xprt *xprt)
302 {
303         struct sockaddr *sap = xs_addr(xprt);
304         struct sockaddr_in6 *sin6;
305         struct sockaddr_in *sin;
306         char buf[128];
307
308         (void)rpc_ntop(sap, buf, sizeof(buf));
309         xprt->address_strings[RPC_DISPLAY_ADDR] = kstrdup(buf, GFP_KERNEL);
310
311         switch (sap->sa_family) {
312         case AF_INET:
313                 sin = xs_addr_in(xprt);
314                 (void)snprintf(buf, sizeof(buf), "%02x%02x%02x%02x",
315                                         NIPQUAD(sin->sin_addr.s_addr));
316                 break;
317         case AF_INET6:
318                 sin6 = xs_addr_in6(xprt);
319                 (void)snprintf(buf, sizeof(buf), "%pi6", &sin6->sin6_addr);
320                 break;
321         default:
322                 BUG();
323         }
324         xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = kstrdup(buf, GFP_KERNEL);
325 }
326
327 static void xs_format_common_peer_ports(struct rpc_xprt *xprt)
328 {
329         struct sockaddr *sap = xs_addr(xprt);
330         char buf[128];
331
332         (void)snprintf(buf, sizeof(buf), "%u", rpc_get_port(sap));
333         xprt->address_strings[RPC_DISPLAY_PORT] = kstrdup(buf, GFP_KERNEL);
334
335         (void)snprintf(buf, sizeof(buf), "%4hx", rpc_get_port(sap));
336         xprt->address_strings[RPC_DISPLAY_HEX_PORT] = kstrdup(buf, GFP_KERNEL);
337 }
338
339 static void xs_format_peer_addresses(struct rpc_xprt *xprt,
340                                      const char *protocol,
341                                      const char *netid)
342 {
343         xprt->address_strings[RPC_DISPLAY_PROTO] = protocol;
344         xprt->address_strings[RPC_DISPLAY_NETID] = netid;
345         xs_format_common_peer_addresses(xprt);
346         xs_format_common_peer_ports(xprt);
347 }
348
349 static void xs_update_peer_port(struct rpc_xprt *xprt)
350 {
351         kfree(xprt->address_strings[RPC_DISPLAY_HEX_PORT]);
352         kfree(xprt->address_strings[RPC_DISPLAY_PORT]);
353
354         xs_format_common_peer_ports(xprt);
355 }
356
357 static void xs_free_peer_addresses(struct rpc_xprt *xprt)
358 {
359         unsigned int i;
360
361         for (i = 0; i < RPC_DISPLAY_MAX; i++)
362                 switch (i) {
363                 case RPC_DISPLAY_PROTO:
364                 case RPC_DISPLAY_NETID:
365                         continue;
366                 default:
367                         kfree(xprt->address_strings[i]);
368                 }
369 }
370
371 #define XS_SENDMSG_FLAGS        (MSG_DONTWAIT | MSG_NOSIGNAL)
372
373 static int xs_send_kvec(struct socket *sock, struct sockaddr *addr, int addrlen, struct kvec *vec, unsigned int base, int more)
374 {
375         struct msghdr msg = {
376                 .msg_name       = addr,
377                 .msg_namelen    = addrlen,
378                 .msg_flags      = XS_SENDMSG_FLAGS | (more ? MSG_MORE : 0),
379         };
380         struct kvec iov = {
381                 .iov_base       = vec->iov_base + base,
382                 .iov_len        = vec->iov_len - base,
383         };
384
385         if (iov.iov_len != 0)
386                 return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
387         return kernel_sendmsg(sock, &msg, NULL, 0, 0);
388 }
389
390 static int xs_send_pagedata(struct socket *sock, struct xdr_buf *xdr, unsigned int base, int more)
391 {
392         struct page **ppage;
393         unsigned int remainder;
394         int err, sent = 0;
395
396         remainder = xdr->page_len - base;
397         base += xdr->page_base;
398         ppage = xdr->pages + (base >> PAGE_SHIFT);
399         base &= ~PAGE_MASK;
400         for(;;) {
401                 unsigned int len = min_t(unsigned int, PAGE_SIZE - base, remainder);
402                 int flags = XS_SENDMSG_FLAGS;
403
404                 remainder -= len;
405                 if (remainder != 0 || more)
406                         flags |= MSG_MORE;
407                 err = sock->ops->sendpage(sock, *ppage, base, len, flags);
408                 if (remainder == 0 || err != len)
409                         break;
410                 sent += err;
411                 ppage++;
412                 base = 0;
413         }
414         if (sent == 0)
415                 return err;
416         if (err > 0)
417                 sent += err;
418         return sent;
419 }
420
421 /**
422  * xs_sendpages - write pages directly to a socket
423  * @sock: socket to send on
424  * @addr: UDP only -- address of destination
425  * @addrlen: UDP only -- length of destination address
426  * @xdr: buffer containing this request
427  * @base: starting position in the buffer
428  *
429  */
430 static int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base)
431 {
432         unsigned int remainder = xdr->len - base;
433         int err, sent = 0;
434
435         if (unlikely(!sock))
436                 return -ENOTSOCK;
437
438         clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags);
439         if (base != 0) {
440                 addr = NULL;
441                 addrlen = 0;
442         }
443
444         if (base < xdr->head[0].iov_len || addr != NULL) {
445                 unsigned int len = xdr->head[0].iov_len - base;
446                 remainder -= len;
447                 err = xs_send_kvec(sock, addr, addrlen, &xdr->head[0], base, remainder != 0);
448                 if (remainder == 0 || err != len)
449                         goto out;
450                 sent += err;
451                 base = 0;
452         } else
453                 base -= xdr->head[0].iov_len;
454
455         if (base < xdr->page_len) {
456                 unsigned int len = xdr->page_len - base;
457                 remainder -= len;
458                 err = xs_send_pagedata(sock, xdr, base, remainder != 0);
459                 if (remainder == 0 || err != len)
460                         goto out;
461                 sent += err;
462                 base = 0;
463         } else
464                 base -= xdr->page_len;
465
466         if (base >= xdr->tail[0].iov_len)
467                 return sent;
468         err = xs_send_kvec(sock, NULL, 0, &xdr->tail[0], base, 0);
469 out:
470         if (sent == 0)
471                 return err;
472         if (err > 0)
473                 sent += err;
474         return sent;
475 }
476
477 static void xs_nospace_callback(struct rpc_task *task)
478 {
479         struct sock_xprt *transport = container_of(task->tk_rqstp->rq_xprt, struct sock_xprt, xprt);
480
481         transport->inet->sk_write_pending--;
482         clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
483 }
484
485 /**
486  * xs_nospace - place task on wait queue if transmit was incomplete
487  * @task: task to put to sleep
488  *
489  */
490 static int xs_nospace(struct rpc_task *task)
491 {
492         struct rpc_rqst *req = task->tk_rqstp;
493         struct rpc_xprt *xprt = req->rq_xprt;
494         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
495         int ret = 0;
496
497         dprintk("RPC: %5u xmit incomplete (%u left of %u)\n",
498                         task->tk_pid, req->rq_slen - req->rq_bytes_sent,
499                         req->rq_slen);
500
501         /* Protect against races with write_space */
502         spin_lock_bh(&xprt->transport_lock);
503
504         /* Don't race with disconnect */
505         if (xprt_connected(xprt)) {
506                 if (test_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags)) {
507                         ret = -EAGAIN;
508                         /*
509                          * Notify TCP that we're limited by the application
510                          * window size
511                          */
512                         set_bit(SOCK_NOSPACE, &transport->sock->flags);
513                         transport->inet->sk_write_pending++;
514                         /* ...and wait for more buffer space */
515                         xprt_wait_for_buffer_space(task, xs_nospace_callback);
516                 }
517         } else {
518                 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
519                 ret = -ENOTCONN;
520         }
521
522         spin_unlock_bh(&xprt->transport_lock);
523         return ret;
524 }
525
526 /**
527  * xs_udp_send_request - write an RPC request to a UDP socket
528  * @task: address of RPC task that manages the state of an RPC request
529  *
530  * Return values:
531  *        0:    The request has been sent
532  *   EAGAIN:    The socket was blocked, please call again later to
533  *              complete the request
534  * ENOTCONN:    Caller needs to invoke connect logic then call again
535  *    other:    Some other error occured, the request was not sent
536  */
537 static int xs_udp_send_request(struct rpc_task *task)
538 {
539         struct rpc_rqst *req = task->tk_rqstp;
540         struct rpc_xprt *xprt = req->rq_xprt;
541         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
542         struct xdr_buf *xdr = &req->rq_snd_buf;
543         int status;
544
545         xs_pktdump("packet data:",
546                                 req->rq_svec->iov_base,
547                                 req->rq_svec->iov_len);
548
549         if (!xprt_bound(xprt))
550                 return -ENOTCONN;
551         status = xs_sendpages(transport->sock,
552                               xs_addr(xprt),
553                               xprt->addrlen, xdr,
554                               req->rq_bytes_sent);
555
556         dprintk("RPC:       xs_udp_send_request(%u) = %d\n",
557                         xdr->len - req->rq_bytes_sent, status);
558
559         if (status >= 0) {
560                 task->tk_bytes_sent += status;
561                 if (status >= req->rq_slen)
562                         return 0;
563                 /* Still some bytes left; set up for a retry later. */
564                 status = -EAGAIN;
565         }
566         if (!transport->sock)
567                 goto out;
568
569         switch (status) {
570         case -ENOTSOCK:
571                 status = -ENOTCONN;
572                 /* Should we call xs_close() here? */
573                 break;
574         case -EAGAIN:
575                 status = xs_nospace(task);
576                 break;
577         default:
578                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
579                         -status);
580         case -ENETUNREACH:
581         case -EPIPE:
582         case -ECONNREFUSED:
583                 /* When the server has died, an ICMP port unreachable message
584                  * prompts ECONNREFUSED. */
585                 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
586         }
587 out:
588         return status;
589 }
590
591 /**
592  * xs_tcp_shutdown - gracefully shut down a TCP socket
593  * @xprt: transport
594  *
595  * Initiates a graceful shutdown of the TCP socket by calling the
596  * equivalent of shutdown(SHUT_WR);
597  */
598 static void xs_tcp_shutdown(struct rpc_xprt *xprt)
599 {
600         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
601         struct socket *sock = transport->sock;
602
603         if (sock != NULL)
604                 kernel_sock_shutdown(sock, SHUT_WR);
605 }
606
607 static inline void xs_encode_tcp_record_marker(struct xdr_buf *buf)
608 {
609         u32 reclen = buf->len - sizeof(rpc_fraghdr);
610         rpc_fraghdr *base = buf->head[0].iov_base;
611         *base = htonl(RPC_LAST_STREAM_FRAGMENT | reclen);
612 }
613
614 /**
615  * xs_tcp_send_request - write an RPC request to a TCP socket
616  * @task: address of RPC task that manages the state of an RPC request
617  *
618  * Return values:
619  *        0:    The request has been sent
620  *   EAGAIN:    The socket was blocked, please call again later to
621  *              complete the request
622  * ENOTCONN:    Caller needs to invoke connect logic then call again
623  *    other:    Some other error occured, the request was not sent
624  *
625  * XXX: In the case of soft timeouts, should we eventually give up
626  *      if sendmsg is not able to make progress?
627  */
628 static int xs_tcp_send_request(struct rpc_task *task)
629 {
630         struct rpc_rqst *req = task->tk_rqstp;
631         struct rpc_xprt *xprt = req->rq_xprt;
632         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
633         struct xdr_buf *xdr = &req->rq_snd_buf;
634         int status;
635
636         xs_encode_tcp_record_marker(&req->rq_snd_buf);
637
638         xs_pktdump("packet data:",
639                                 req->rq_svec->iov_base,
640                                 req->rq_svec->iov_len);
641
642         /* Continue transmitting the packet/record. We must be careful
643          * to cope with writespace callbacks arriving _after_ we have
644          * called sendmsg(). */
645         while (1) {
646                 status = xs_sendpages(transport->sock,
647                                         NULL, 0, xdr, req->rq_bytes_sent);
648
649                 dprintk("RPC:       xs_tcp_send_request(%u) = %d\n",
650                                 xdr->len - req->rq_bytes_sent, status);
651
652                 if (unlikely(status < 0))
653                         break;
654
655                 /* If we've sent the entire packet, immediately
656                  * reset the count of bytes sent. */
657                 req->rq_bytes_sent += status;
658                 task->tk_bytes_sent += status;
659                 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
660                         req->rq_bytes_sent = 0;
661                         return 0;
662                 }
663
664                 if (status != 0)
665                         continue;
666                 status = -EAGAIN;
667                 break;
668         }
669         if (!transport->sock)
670                 goto out;
671
672         switch (status) {
673         case -ENOTSOCK:
674                 status = -ENOTCONN;
675                 /* Should we call xs_close() here? */
676                 break;
677         case -EAGAIN:
678                 status = xs_nospace(task);
679                 break;
680         default:
681                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
682                         -status);
683         case -ECONNRESET:
684         case -EPIPE:
685                 xs_tcp_shutdown(xprt);
686         case -ECONNREFUSED:
687         case -ENOTCONN:
688                 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
689         }
690 out:
691         return status;
692 }
693
694 /**
695  * xs_tcp_release_xprt - clean up after a tcp transmission
696  * @xprt: transport
697  * @task: rpc task
698  *
699  * This cleans up if an error causes us to abort the transmission of a request.
700  * In this case, the socket may need to be reset in order to avoid confusing
701  * the server.
702  */
703 static void xs_tcp_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
704 {
705         struct rpc_rqst *req;
706
707         if (task != xprt->snd_task)
708                 return;
709         if (task == NULL)
710                 goto out_release;
711         req = task->tk_rqstp;
712         if (req->rq_bytes_sent == 0)
713                 goto out_release;
714         if (req->rq_bytes_sent == req->rq_snd_buf.len)
715                 goto out_release;
716         set_bit(XPRT_CLOSE_WAIT, &task->tk_xprt->state);
717 out_release:
718         xprt_release_xprt(xprt, task);
719 }
720
721 static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk)
722 {
723         transport->old_data_ready = sk->sk_data_ready;
724         transport->old_state_change = sk->sk_state_change;
725         transport->old_write_space = sk->sk_write_space;
726         transport->old_error_report = sk->sk_error_report;
727 }
728
729 static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk)
730 {
731         sk->sk_data_ready = transport->old_data_ready;
732         sk->sk_state_change = transport->old_state_change;
733         sk->sk_write_space = transport->old_write_space;
734         sk->sk_error_report = transport->old_error_report;
735 }
736
737 static void xs_reset_transport(struct sock_xprt *transport)
738 {
739         struct socket *sock = transport->sock;
740         struct sock *sk = transport->inet;
741
742         if (sk == NULL)
743                 return;
744
745         write_lock_bh(&sk->sk_callback_lock);
746         transport->inet = NULL;
747         transport->sock = NULL;
748
749         sk->sk_user_data = NULL;
750
751         xs_restore_old_callbacks(transport, sk);
752         write_unlock_bh(&sk->sk_callback_lock);
753
754         sk->sk_no_check = 0;
755
756         sock_release(sock);
757 }
758
759 /**
760  * xs_close - close a socket
761  * @xprt: transport
762  *
763  * This is used when all requests are complete; ie, no DRC state remains
764  * on the server we want to save.
765  *
766  * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with
767  * xs_reset_transport() zeroing the socket from underneath a writer.
768  */
769 static void xs_close(struct rpc_xprt *xprt)
770 {
771         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
772
773         dprintk("RPC:       xs_close xprt %p\n", xprt);
774
775         xs_reset_transport(transport);
776         xprt->reestablish_timeout = 0;
777
778         smp_mb__before_clear_bit();
779         clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
780         clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
781         clear_bit(XPRT_CLOSING, &xprt->state);
782         smp_mb__after_clear_bit();
783         xprt_disconnect_done(xprt);
784 }
785
786 static void xs_tcp_close(struct rpc_xprt *xprt)
787 {
788         if (test_and_clear_bit(XPRT_CONNECTION_CLOSE, &xprt->state))
789                 xs_close(xprt);
790         else
791                 xs_tcp_shutdown(xprt);
792 }
793
794 /**
795  * xs_destroy - prepare to shutdown a transport
796  * @xprt: doomed transport
797  *
798  */
799 static void xs_destroy(struct rpc_xprt *xprt)
800 {
801         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
802
803         dprintk("RPC:       xs_destroy xprt %p\n", xprt);
804
805         cancel_rearming_delayed_work(&transport->connect_worker);
806
807         xs_close(xprt);
808         xs_free_peer_addresses(xprt);
809         kfree(xprt->slot);
810         kfree(xprt);
811         module_put(THIS_MODULE);
812 }
813
814 static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
815 {
816         return (struct rpc_xprt *) sk->sk_user_data;
817 }
818
819 /**
820  * xs_udp_data_ready - "data ready" callback for UDP sockets
821  * @sk: socket with data to read
822  * @len: how much data to read
823  *
824  */
825 static void xs_udp_data_ready(struct sock *sk, int len)
826 {
827         struct rpc_task *task;
828         struct rpc_xprt *xprt;
829         struct rpc_rqst *rovr;
830         struct sk_buff *skb;
831         int err, repsize, copied;
832         u32 _xid;
833         __be32 *xp;
834
835         read_lock(&sk->sk_callback_lock);
836         dprintk("RPC:       xs_udp_data_ready...\n");
837         if (!(xprt = xprt_from_sock(sk)))
838                 goto out;
839
840         if ((skb = skb_recv_datagram(sk, 0, 1, &err)) == NULL)
841                 goto out;
842
843         if (xprt->shutdown)
844                 goto dropit;
845
846         repsize = skb->len - sizeof(struct udphdr);
847         if (repsize < 4) {
848                 dprintk("RPC:       impossible RPC reply size %d!\n", repsize);
849                 goto dropit;
850         }
851
852         /* Copy the XID from the skb... */
853         xp = skb_header_pointer(skb, sizeof(struct udphdr),
854                                 sizeof(_xid), &_xid);
855         if (xp == NULL)
856                 goto dropit;
857
858         /* Look up and lock the request corresponding to the given XID */
859         spin_lock(&xprt->transport_lock);
860         rovr = xprt_lookup_rqst(xprt, *xp);
861         if (!rovr)
862                 goto out_unlock;
863         task = rovr->rq_task;
864
865         if ((copied = rovr->rq_private_buf.buflen) > repsize)
866                 copied = repsize;
867
868         /* Suck it into the iovec, verify checksum if not done by hw. */
869         if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
870                 UDPX_INC_STATS_BH(sk, UDP_MIB_INERRORS);
871                 goto out_unlock;
872         }
873
874         UDPX_INC_STATS_BH(sk, UDP_MIB_INDATAGRAMS);
875
876         /* Something worked... */
877         dst_confirm(skb_dst(skb));
878
879         xprt_adjust_cwnd(task, copied);
880         xprt_update_rtt(task);
881         xprt_complete_rqst(task, copied);
882
883  out_unlock:
884         spin_unlock(&xprt->transport_lock);
885  dropit:
886         skb_free_datagram(sk, skb);
887  out:
888         read_unlock(&sk->sk_callback_lock);
889 }
890
891 static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
892 {
893         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
894         size_t len, used;
895         char *p;
896
897         p = ((char *) &transport->tcp_fraghdr) + transport->tcp_offset;
898         len = sizeof(transport->tcp_fraghdr) - transport->tcp_offset;
899         used = xdr_skb_read_bits(desc, p, len);
900         transport->tcp_offset += used;
901         if (used != len)
902                 return;
903
904         transport->tcp_reclen = ntohl(transport->tcp_fraghdr);
905         if (transport->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
906                 transport->tcp_flags |= TCP_RCV_LAST_FRAG;
907         else
908                 transport->tcp_flags &= ~TCP_RCV_LAST_FRAG;
909         transport->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
910
911         transport->tcp_flags &= ~TCP_RCV_COPY_FRAGHDR;
912         transport->tcp_offset = 0;
913
914         /* Sanity check of the record length */
915         if (unlikely(transport->tcp_reclen < 8)) {
916                 dprintk("RPC:       invalid TCP record fragment length\n");
917                 xprt_force_disconnect(xprt);
918                 return;
919         }
920         dprintk("RPC:       reading TCP record fragment of length %d\n",
921                         transport->tcp_reclen);
922 }
923
924 static void xs_tcp_check_fraghdr(struct sock_xprt *transport)
925 {
926         if (transport->tcp_offset == transport->tcp_reclen) {
927                 transport->tcp_flags |= TCP_RCV_COPY_FRAGHDR;
928                 transport->tcp_offset = 0;
929                 if (transport->tcp_flags & TCP_RCV_LAST_FRAG) {
930                         transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
931                         transport->tcp_flags |= TCP_RCV_COPY_XID;
932                         transport->tcp_copied = 0;
933                 }
934         }
935 }
936
937 static inline void xs_tcp_read_xid(struct sock_xprt *transport, struct xdr_skb_reader *desc)
938 {
939         size_t len, used;
940         char *p;
941
942         len = sizeof(transport->tcp_xid) - transport->tcp_offset;
943         dprintk("RPC:       reading XID (%Zu bytes)\n", len);
944         p = ((char *) &transport->tcp_xid) + transport->tcp_offset;
945         used = xdr_skb_read_bits(desc, p, len);
946         transport->tcp_offset += used;
947         if (used != len)
948                 return;
949         transport->tcp_flags &= ~TCP_RCV_COPY_XID;
950         transport->tcp_flags |= TCP_RCV_READ_CALLDIR;
951         transport->tcp_copied = 4;
952         dprintk("RPC:       reading %s XID %08x\n",
953                         (transport->tcp_flags & TCP_RPC_REPLY) ? "reply for"
954                                                               : "request with",
955                         ntohl(transport->tcp_xid));
956         xs_tcp_check_fraghdr(transport);
957 }
958
959 static inline void xs_tcp_read_calldir(struct sock_xprt *transport,
960                                        struct xdr_skb_reader *desc)
961 {
962         size_t len, used;
963         u32 offset;
964         __be32  calldir;
965
966         /*
967          * We want transport->tcp_offset to be 8 at the end of this routine
968          * (4 bytes for the xid and 4 bytes for the call/reply flag).
969          * When this function is called for the first time,
970          * transport->tcp_offset is 4 (after having already read the xid).
971          */
972         offset = transport->tcp_offset - sizeof(transport->tcp_xid);
973         len = sizeof(calldir) - offset;
974         dprintk("RPC:       reading CALL/REPLY flag (%Zu bytes)\n", len);
975         used = xdr_skb_read_bits(desc, &calldir, len);
976         transport->tcp_offset += used;
977         if (used != len)
978                 return;
979         transport->tcp_flags &= ~TCP_RCV_READ_CALLDIR;
980         transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
981         transport->tcp_flags |= TCP_RCV_COPY_DATA;
982         /*
983          * We don't yet have the XDR buffer, so we will write the calldir
984          * out after we get the buffer from the 'struct rpc_rqst'
985          */
986         if (ntohl(calldir) == RPC_REPLY)
987                 transport->tcp_flags |= TCP_RPC_REPLY;
988         else
989                 transport->tcp_flags &= ~TCP_RPC_REPLY;
990         dprintk("RPC:       reading %s CALL/REPLY flag %08x\n",
991                         (transport->tcp_flags & TCP_RPC_REPLY) ?
992                                 "reply for" : "request with", calldir);
993         xs_tcp_check_fraghdr(transport);
994 }
995
996 static inline void xs_tcp_read_common(struct rpc_xprt *xprt,
997                                      struct xdr_skb_reader *desc,
998                                      struct rpc_rqst *req)
999 {
1000         struct sock_xprt *transport =
1001                                 container_of(xprt, struct sock_xprt, xprt);
1002         struct xdr_buf *rcvbuf;
1003         size_t len;
1004         ssize_t r;
1005
1006         rcvbuf = &req->rq_private_buf;
1007
1008         if (transport->tcp_flags & TCP_RCV_COPY_CALLDIR) {
1009                 /*
1010                  * Save the RPC direction in the XDR buffer
1011                  */
1012                 __be32  calldir = transport->tcp_flags & TCP_RPC_REPLY ?
1013                                         htonl(RPC_REPLY) : 0;
1014
1015                 memcpy(rcvbuf->head[0].iov_base + transport->tcp_copied,
1016                         &calldir, sizeof(calldir));
1017                 transport->tcp_copied += sizeof(calldir);
1018                 transport->tcp_flags &= ~TCP_RCV_COPY_CALLDIR;
1019         }
1020
1021         len = desc->count;
1022         if (len > transport->tcp_reclen - transport->tcp_offset) {
1023                 struct xdr_skb_reader my_desc;
1024
1025                 len = transport->tcp_reclen - transport->tcp_offset;
1026                 memcpy(&my_desc, desc, sizeof(my_desc));
1027                 my_desc.count = len;
1028                 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
1029                                           &my_desc, xdr_skb_read_bits);
1030                 desc->count -= r;
1031                 desc->offset += r;
1032         } else
1033                 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
1034                                           desc, xdr_skb_read_bits);
1035
1036         if (r > 0) {
1037                 transport->tcp_copied += r;
1038                 transport->tcp_offset += r;
1039         }
1040         if (r != len) {
1041                 /* Error when copying to the receive buffer,
1042                  * usually because we weren't able to allocate
1043                  * additional buffer pages. All we can do now
1044                  * is turn off TCP_RCV_COPY_DATA, so the request
1045                  * will not receive any additional updates,
1046                  * and time out.
1047                  * Any remaining data from this record will
1048                  * be discarded.
1049                  */
1050                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1051                 dprintk("RPC:       XID %08x truncated request\n",
1052                                 ntohl(transport->tcp_xid));
1053                 dprintk("RPC:       xprt = %p, tcp_copied = %lu, "
1054                                 "tcp_offset = %u, tcp_reclen = %u\n",
1055                                 xprt, transport->tcp_copied,
1056                                 transport->tcp_offset, transport->tcp_reclen);
1057                 return;
1058         }
1059
1060         dprintk("RPC:       XID %08x read %Zd bytes\n",
1061                         ntohl(transport->tcp_xid), r);
1062         dprintk("RPC:       xprt = %p, tcp_copied = %lu, tcp_offset = %u, "
1063                         "tcp_reclen = %u\n", xprt, transport->tcp_copied,
1064                         transport->tcp_offset, transport->tcp_reclen);
1065
1066         if (transport->tcp_copied == req->rq_private_buf.buflen)
1067                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1068         else if (transport->tcp_offset == transport->tcp_reclen) {
1069                 if (transport->tcp_flags & TCP_RCV_LAST_FRAG)
1070                         transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1071         }
1072
1073         return;
1074 }
1075
1076 /*
1077  * Finds the request corresponding to the RPC xid and invokes the common
1078  * tcp read code to read the data.
1079  */
1080 static inline int xs_tcp_read_reply(struct rpc_xprt *xprt,
1081                                     struct xdr_skb_reader *desc)
1082 {
1083         struct sock_xprt *transport =
1084                                 container_of(xprt, struct sock_xprt, xprt);
1085         struct rpc_rqst *req;
1086
1087         dprintk("RPC:       read reply XID %08x\n", ntohl(transport->tcp_xid));
1088
1089         /* Find and lock the request corresponding to this xid */
1090         spin_lock(&xprt->transport_lock);
1091         req = xprt_lookup_rqst(xprt, transport->tcp_xid);
1092         if (!req) {
1093                 dprintk("RPC:       XID %08x request not found!\n",
1094                                 ntohl(transport->tcp_xid));
1095                 spin_unlock(&xprt->transport_lock);
1096                 return -1;
1097         }
1098
1099         xs_tcp_read_common(xprt, desc, req);
1100
1101         if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1102                 xprt_complete_rqst(req->rq_task, transport->tcp_copied);
1103
1104         spin_unlock(&xprt->transport_lock);
1105         return 0;
1106 }
1107
1108 #if defined(CONFIG_NFS_V4_1)
1109 /*
1110  * Obtains an rpc_rqst previously allocated and invokes the common
1111  * tcp read code to read the data.  The result is placed in the callback
1112  * queue.
1113  * If we're unable to obtain the rpc_rqst we schedule the closing of the
1114  * connection and return -1.
1115  */
1116 static inline int xs_tcp_read_callback(struct rpc_xprt *xprt,
1117                                        struct xdr_skb_reader *desc)
1118 {
1119         struct sock_xprt *transport =
1120                                 container_of(xprt, struct sock_xprt, xprt);
1121         struct rpc_rqst *req;
1122
1123         req = xprt_alloc_bc_request(xprt);
1124         if (req == NULL) {
1125                 printk(KERN_WARNING "Callback slot table overflowed\n");
1126                 xprt_force_disconnect(xprt);
1127                 return -1;
1128         }
1129
1130         req->rq_xid = transport->tcp_xid;
1131         dprintk("RPC:       read callback  XID %08x\n", ntohl(req->rq_xid));
1132         xs_tcp_read_common(xprt, desc, req);
1133
1134         if (!(transport->tcp_flags & TCP_RCV_COPY_DATA)) {
1135                 struct svc_serv *bc_serv = xprt->bc_serv;
1136
1137                 /*
1138                  * Add callback request to callback list.  The callback
1139                  * service sleeps on the sv_cb_waitq waiting for new
1140                  * requests.  Wake it up after adding enqueing the
1141                  * request.
1142                  */
1143                 dprintk("RPC:       add callback request to list\n");
1144                 spin_lock(&bc_serv->sv_cb_lock);
1145                 list_add(&req->rq_bc_list, &bc_serv->sv_cb_list);
1146                 spin_unlock(&bc_serv->sv_cb_lock);
1147                 wake_up(&bc_serv->sv_cb_waitq);
1148         }
1149
1150         req->rq_private_buf.len = transport->tcp_copied;
1151
1152         return 0;
1153 }
1154
1155 static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1156                                         struct xdr_skb_reader *desc)
1157 {
1158         struct sock_xprt *transport =
1159                                 container_of(xprt, struct sock_xprt, xprt);
1160
1161         return (transport->tcp_flags & TCP_RPC_REPLY) ?
1162                 xs_tcp_read_reply(xprt, desc) :
1163                 xs_tcp_read_callback(xprt, desc);
1164 }
1165 #else
1166 static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1167                                         struct xdr_skb_reader *desc)
1168 {
1169         return xs_tcp_read_reply(xprt, desc);
1170 }
1171 #endif /* CONFIG_NFS_V4_1 */
1172
1173 /*
1174  * Read data off the transport.  This can be either an RPC_CALL or an
1175  * RPC_REPLY.  Relay the processing to helper functions.
1176  */
1177 static void xs_tcp_read_data(struct rpc_xprt *xprt,
1178                                     struct xdr_skb_reader *desc)
1179 {
1180         struct sock_xprt *transport =
1181                                 container_of(xprt, struct sock_xprt, xprt);
1182
1183         if (_xs_tcp_read_data(xprt, desc) == 0)
1184                 xs_tcp_check_fraghdr(transport);
1185         else {
1186                 /*
1187                  * The transport_lock protects the request handling.
1188                  * There's no need to hold it to update the tcp_flags.
1189                  */
1190                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1191         }
1192 }
1193
1194 static inline void xs_tcp_read_discard(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1195 {
1196         size_t len;
1197
1198         len = transport->tcp_reclen - transport->tcp_offset;
1199         if (len > desc->count)
1200                 len = desc->count;
1201         desc->count -= len;
1202         desc->offset += len;
1203         transport->tcp_offset += len;
1204         dprintk("RPC:       discarded %Zu bytes\n", len);
1205         xs_tcp_check_fraghdr(transport);
1206 }
1207
1208 static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
1209 {
1210         struct rpc_xprt *xprt = rd_desc->arg.data;
1211         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1212         struct xdr_skb_reader desc = {
1213                 .skb    = skb,
1214                 .offset = offset,
1215                 .count  = len,
1216         };
1217
1218         dprintk("RPC:       xs_tcp_data_recv started\n");
1219         do {
1220                 /* Read in a new fragment marker if necessary */
1221                 /* Can we ever really expect to get completely empty fragments? */
1222                 if (transport->tcp_flags & TCP_RCV_COPY_FRAGHDR) {
1223                         xs_tcp_read_fraghdr(xprt, &desc);
1224                         continue;
1225                 }
1226                 /* Read in the xid if necessary */
1227                 if (transport->tcp_flags & TCP_RCV_COPY_XID) {
1228                         xs_tcp_read_xid(transport, &desc);
1229                         continue;
1230                 }
1231                 /* Read in the call/reply flag */
1232                 if (transport->tcp_flags & TCP_RCV_READ_CALLDIR) {
1233                         xs_tcp_read_calldir(transport, &desc);
1234                         continue;
1235                 }
1236                 /* Read in the request data */
1237                 if (transport->tcp_flags & TCP_RCV_COPY_DATA) {
1238                         xs_tcp_read_data(xprt, &desc);
1239                         continue;
1240                 }
1241                 /* Skip over any trailing bytes on short reads */
1242                 xs_tcp_read_discard(transport, &desc);
1243         } while (desc.count);
1244         dprintk("RPC:       xs_tcp_data_recv done\n");
1245         return len - desc.count;
1246 }
1247
1248 /**
1249  * xs_tcp_data_ready - "data ready" callback for TCP sockets
1250  * @sk: socket with data to read
1251  * @bytes: how much data to read
1252  *
1253  */
1254 static void xs_tcp_data_ready(struct sock *sk, int bytes)
1255 {
1256         struct rpc_xprt *xprt;
1257         read_descriptor_t rd_desc;
1258         int read;
1259
1260         dprintk("RPC:       xs_tcp_data_ready...\n");
1261
1262         read_lock(&sk->sk_callback_lock);
1263         if (!(xprt = xprt_from_sock(sk)))
1264                 goto out;
1265         if (xprt->shutdown)
1266                 goto out;
1267
1268         /* Any data means we had a useful conversation, so
1269          * the we don't need to delay the next reconnect
1270          */
1271         if (xprt->reestablish_timeout)
1272                 xprt->reestablish_timeout = 0;
1273
1274         /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
1275         rd_desc.arg.data = xprt;
1276         do {
1277                 rd_desc.count = 65536;
1278                 read = tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
1279         } while (read > 0);
1280 out:
1281         read_unlock(&sk->sk_callback_lock);
1282 }
1283
1284 /*
1285  * Do the equivalent of linger/linger2 handling for dealing with
1286  * broken servers that don't close the socket in a timely
1287  * fashion
1288  */
1289 static void xs_tcp_schedule_linger_timeout(struct rpc_xprt *xprt,
1290                 unsigned long timeout)
1291 {
1292         struct sock_xprt *transport;
1293
1294         if (xprt_test_and_set_connecting(xprt))
1295                 return;
1296         set_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1297         transport = container_of(xprt, struct sock_xprt, xprt);
1298         queue_delayed_work(rpciod_workqueue, &transport->connect_worker,
1299                            timeout);
1300 }
1301
1302 static void xs_tcp_cancel_linger_timeout(struct rpc_xprt *xprt)
1303 {
1304         struct sock_xprt *transport;
1305
1306         transport = container_of(xprt, struct sock_xprt, xprt);
1307
1308         if (!test_bit(XPRT_CONNECTION_ABORT, &xprt->state) ||
1309             !cancel_delayed_work(&transport->connect_worker))
1310                 return;
1311         clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1312         xprt_clear_connecting(xprt);
1313 }
1314
1315 static void xs_sock_mark_closed(struct rpc_xprt *xprt)
1316 {
1317         smp_mb__before_clear_bit();
1318         clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1319         clear_bit(XPRT_CLOSING, &xprt->state);
1320         smp_mb__after_clear_bit();
1321         /* Mark transport as closed and wake up all pending tasks */
1322         xprt_disconnect_done(xprt);
1323 }
1324
1325 /**
1326  * xs_tcp_state_change - callback to handle TCP socket state changes
1327  * @sk: socket whose state has changed
1328  *
1329  */
1330 static void xs_tcp_state_change(struct sock *sk)
1331 {
1332         struct rpc_xprt *xprt;
1333
1334         read_lock(&sk->sk_callback_lock);
1335         if (!(xprt = xprt_from_sock(sk)))
1336                 goto out;
1337         dprintk("RPC:       xs_tcp_state_change client %p...\n", xprt);
1338         dprintk("RPC:       state %x conn %d dead %d zapped %d\n",
1339                         sk->sk_state, xprt_connected(xprt),
1340                         sock_flag(sk, SOCK_DEAD),
1341                         sock_flag(sk, SOCK_ZAPPED));
1342
1343         switch (sk->sk_state) {
1344         case TCP_ESTABLISHED:
1345                 spin_lock_bh(&xprt->transport_lock);
1346                 if (!xprt_test_and_set_connected(xprt)) {
1347                         struct sock_xprt *transport = container_of(xprt,
1348                                         struct sock_xprt, xprt);
1349
1350                         /* Reset TCP record info */
1351                         transport->tcp_offset = 0;
1352                         transport->tcp_reclen = 0;
1353                         transport->tcp_copied = 0;
1354                         transport->tcp_flags =
1355                                 TCP_RCV_COPY_FRAGHDR | TCP_RCV_COPY_XID;
1356
1357                         xprt_wake_pending_tasks(xprt, -EAGAIN);
1358                 }
1359                 spin_unlock_bh(&xprt->transport_lock);
1360                 break;
1361         case TCP_FIN_WAIT1:
1362                 /* The client initiated a shutdown of the socket */
1363                 xprt->connect_cookie++;
1364                 xprt->reestablish_timeout = 0;
1365                 set_bit(XPRT_CLOSING, &xprt->state);
1366                 smp_mb__before_clear_bit();
1367                 clear_bit(XPRT_CONNECTED, &xprt->state);
1368                 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1369                 smp_mb__after_clear_bit();
1370                 xs_tcp_schedule_linger_timeout(xprt, xs_tcp_fin_timeout);
1371                 break;
1372         case TCP_CLOSE_WAIT:
1373                 /* The server initiated a shutdown of the socket */
1374                 xprt_force_disconnect(xprt);
1375         case TCP_SYN_SENT:
1376                 xprt->connect_cookie++;
1377         case TCP_CLOSING:
1378                 /*
1379                  * If the server closed down the connection, make sure that
1380                  * we back off before reconnecting
1381                  */
1382                 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1383                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1384                 break;
1385         case TCP_LAST_ACK:
1386                 set_bit(XPRT_CLOSING, &xprt->state);
1387                 xs_tcp_schedule_linger_timeout(xprt, xs_tcp_fin_timeout);
1388                 smp_mb__before_clear_bit();
1389                 clear_bit(XPRT_CONNECTED, &xprt->state);
1390                 smp_mb__after_clear_bit();
1391                 break;
1392         case TCP_CLOSE:
1393                 xs_tcp_cancel_linger_timeout(xprt);
1394                 xs_sock_mark_closed(xprt);
1395         }
1396  out:
1397         read_unlock(&sk->sk_callback_lock);
1398 }
1399
1400 /**
1401  * xs_error_report - callback mainly for catching socket errors
1402  * @sk: socket
1403  */
1404 static void xs_error_report(struct sock *sk)
1405 {
1406         struct rpc_xprt *xprt;
1407
1408         read_lock(&sk->sk_callback_lock);
1409         if (!(xprt = xprt_from_sock(sk)))
1410                 goto out;
1411         dprintk("RPC:       %s client %p...\n"
1412                         "RPC:       error %d\n",
1413                         __func__, xprt, sk->sk_err);
1414         xprt_wake_pending_tasks(xprt, -EAGAIN);
1415 out:
1416         read_unlock(&sk->sk_callback_lock);
1417 }
1418
1419 static void xs_write_space(struct sock *sk)
1420 {
1421         struct socket *sock;
1422         struct rpc_xprt *xprt;
1423
1424         if (unlikely(!(sock = sk->sk_socket)))
1425                 return;
1426         clear_bit(SOCK_NOSPACE, &sock->flags);
1427
1428         if (unlikely(!(xprt = xprt_from_sock(sk))))
1429                 return;
1430         if (test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags) == 0)
1431                 return;
1432
1433         xprt_write_space(xprt);
1434 }
1435
1436 /**
1437  * xs_udp_write_space - callback invoked when socket buffer space
1438  *                             becomes available
1439  * @sk: socket whose state has changed
1440  *
1441  * Called when more output buffer space is available for this socket.
1442  * We try not to wake our writers until they can make "significant"
1443  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1444  * with a bunch of small requests.
1445  */
1446 static void xs_udp_write_space(struct sock *sk)
1447 {
1448         read_lock(&sk->sk_callback_lock);
1449
1450         /* from net/core/sock.c:sock_def_write_space */
1451         if (sock_writeable(sk))
1452                 xs_write_space(sk);
1453
1454         read_unlock(&sk->sk_callback_lock);
1455 }
1456
1457 /**
1458  * xs_tcp_write_space - callback invoked when socket buffer space
1459  *                             becomes available
1460  * @sk: socket whose state has changed
1461  *
1462  * Called when more output buffer space is available for this socket.
1463  * We try not to wake our writers until they can make "significant"
1464  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1465  * with a bunch of small requests.
1466  */
1467 static void xs_tcp_write_space(struct sock *sk)
1468 {
1469         read_lock(&sk->sk_callback_lock);
1470
1471         /* from net/core/stream.c:sk_stream_write_space */
1472         if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk))
1473                 xs_write_space(sk);
1474
1475         read_unlock(&sk->sk_callback_lock);
1476 }
1477
1478 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
1479 {
1480         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1481         struct sock *sk = transport->inet;
1482
1483         if (transport->rcvsize) {
1484                 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1485                 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
1486         }
1487         if (transport->sndsize) {
1488                 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1489                 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1490                 sk->sk_write_space(sk);
1491         }
1492 }
1493
1494 /**
1495  * xs_udp_set_buffer_size - set send and receive limits
1496  * @xprt: generic transport
1497  * @sndsize: requested size of send buffer, in bytes
1498  * @rcvsize: requested size of receive buffer, in bytes
1499  *
1500  * Set socket send and receive buffer size limits.
1501  */
1502 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1503 {
1504         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1505
1506         transport->sndsize = 0;
1507         if (sndsize)
1508                 transport->sndsize = sndsize + 1024;
1509         transport->rcvsize = 0;
1510         if (rcvsize)
1511                 transport->rcvsize = rcvsize + 1024;
1512
1513         xs_udp_do_set_buffer_size(xprt);
1514 }
1515
1516 /**
1517  * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1518  * @task: task that timed out
1519  *
1520  * Adjust the congestion window after a retransmit timeout has occurred.
1521  */
1522 static void xs_udp_timer(struct rpc_task *task)
1523 {
1524         xprt_adjust_cwnd(task, -ETIMEDOUT);
1525 }
1526
1527 static unsigned short xs_get_random_port(void)
1528 {
1529         unsigned short range = xprt_max_resvport - xprt_min_resvport;
1530         unsigned short rand = (unsigned short) net_random() % range;
1531         return rand + xprt_min_resvport;
1532 }
1533
1534 /**
1535  * xs_set_port - reset the port number in the remote endpoint address
1536  * @xprt: generic transport
1537  * @port: new port number
1538  *
1539  */
1540 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1541 {
1542         dprintk("RPC:       setting port for xprt %p to %u\n", xprt, port);
1543
1544         rpc_set_port(xs_addr(xprt), port);
1545         xs_update_peer_port(xprt);
1546 }
1547
1548 static unsigned short xs_get_srcport(struct sock_xprt *transport, struct socket *sock)
1549 {
1550         unsigned short port = transport->srcport;
1551
1552         if (port == 0 && transport->xprt.resvport)
1553                 port = xs_get_random_port();
1554         return port;
1555 }
1556
1557 static unsigned short xs_next_srcport(struct sock_xprt *transport, struct socket *sock, unsigned short port)
1558 {
1559         if (transport->srcport != 0)
1560                 transport->srcport = 0;
1561         if (!transport->xprt.resvport)
1562                 return 0;
1563         if (port <= xprt_min_resvport || port > xprt_max_resvport)
1564                 return xprt_max_resvport;
1565         return --port;
1566 }
1567
1568 static int xs_bind4(struct sock_xprt *transport, struct socket *sock)
1569 {
1570         struct sockaddr_in myaddr = {
1571                 .sin_family = AF_INET,
1572         };
1573         struct sockaddr_in *sa;
1574         int err, nloop = 0;
1575         unsigned short port = xs_get_srcport(transport, sock);
1576         unsigned short last;
1577
1578         sa = (struct sockaddr_in *)&transport->srcaddr;
1579         myaddr.sin_addr = sa->sin_addr;
1580         do {
1581                 myaddr.sin_port = htons(port);
1582                 err = kernel_bind(sock, (struct sockaddr *) &myaddr,
1583                                                 sizeof(myaddr));
1584                 if (port == 0)
1585                         break;
1586                 if (err == 0) {
1587                         transport->srcport = port;
1588                         break;
1589                 }
1590                 last = port;
1591                 port = xs_next_srcport(transport, sock, port);
1592                 if (port > last)
1593                         nloop++;
1594         } while (err == -EADDRINUSE && nloop != 2);
1595         dprintk("RPC:       %s %pI4:%u: %s (%d)\n",
1596                         __func__, &myaddr.sin_addr,
1597                         port, err ? "failed" : "ok", err);
1598         return err;
1599 }
1600
1601 static int xs_bind6(struct sock_xprt *transport, struct socket *sock)
1602 {
1603         struct sockaddr_in6 myaddr = {
1604                 .sin6_family = AF_INET6,
1605         };
1606         struct sockaddr_in6 *sa;
1607         int err, nloop = 0;
1608         unsigned short port = xs_get_srcport(transport, sock);
1609         unsigned short last;
1610
1611         sa = (struct sockaddr_in6 *)&transport->srcaddr;
1612         myaddr.sin6_addr = sa->sin6_addr;
1613         do {
1614                 myaddr.sin6_port = htons(port);
1615                 err = kernel_bind(sock, (struct sockaddr *) &myaddr,
1616                                                 sizeof(myaddr));
1617                 if (port == 0)
1618                         break;
1619                 if (err == 0) {
1620                         transport->srcport = port;
1621                         break;
1622                 }
1623                 last = port;
1624                 port = xs_next_srcport(transport, sock, port);
1625                 if (port > last)
1626                         nloop++;
1627         } while (err == -EADDRINUSE && nloop != 2);
1628         dprintk("RPC:       xs_bind6 %pI6:%u: %s (%d)\n",
1629                 &myaddr.sin6_addr, port, err ? "failed" : "ok", err);
1630         return err;
1631 }
1632
1633 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1634 static struct lock_class_key xs_key[2];
1635 static struct lock_class_key xs_slock_key[2];
1636
1637 static inline void xs_reclassify_socket4(struct socket *sock)
1638 {
1639         struct sock *sk = sock->sk;
1640
1641         BUG_ON(sock_owned_by_user(sk));
1642         sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1643                 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1644 }
1645
1646 static inline void xs_reclassify_socket6(struct socket *sock)
1647 {
1648         struct sock *sk = sock->sk;
1649
1650         BUG_ON(sock_owned_by_user(sk));
1651         sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1652                 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
1653 }
1654 #else
1655 static inline void xs_reclassify_socket4(struct socket *sock)
1656 {
1657 }
1658
1659 static inline void xs_reclassify_socket6(struct socket *sock)
1660 {
1661 }
1662 #endif
1663
1664 static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
1665 {
1666         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1667
1668         if (!transport->inet) {
1669                 struct sock *sk = sock->sk;
1670
1671                 write_lock_bh(&sk->sk_callback_lock);
1672
1673                 xs_save_old_callbacks(transport, sk);
1674
1675                 sk->sk_user_data = xprt;
1676                 sk->sk_data_ready = xs_udp_data_ready;
1677                 sk->sk_write_space = xs_udp_write_space;
1678                 sk->sk_error_report = xs_error_report;
1679                 sk->sk_no_check = UDP_CSUM_NORCV;
1680                 sk->sk_allocation = GFP_ATOMIC;
1681
1682                 xprt_set_connected(xprt);
1683
1684                 /* Reset to new socket */
1685                 transport->sock = sock;
1686                 transport->inet = sk;
1687
1688                 write_unlock_bh(&sk->sk_callback_lock);
1689         }
1690         xs_udp_do_set_buffer_size(xprt);
1691 }
1692
1693 /**
1694  * xs_udp_connect_worker4 - set up a UDP socket
1695  * @work: RPC transport to connect
1696  *
1697  * Invoked by a work queue tasklet.
1698  */
1699 static void xs_udp_connect_worker4(struct work_struct *work)
1700 {
1701         struct sock_xprt *transport =
1702                 container_of(work, struct sock_xprt, connect_worker.work);
1703         struct rpc_xprt *xprt = &transport->xprt;
1704         struct socket *sock = transport->sock;
1705         int err, status = -EIO;
1706
1707         if (xprt->shutdown)
1708                 goto out;
1709
1710         /* Start by resetting any existing state */
1711         xs_reset_transport(transport);
1712
1713         err = sock_create_kern(PF_INET, SOCK_DGRAM, IPPROTO_UDP, &sock);
1714         if (err < 0) {
1715                 dprintk("RPC:       can't create UDP transport socket (%d).\n", -err);
1716                 goto out;
1717         }
1718         xs_reclassify_socket4(sock);
1719
1720         if (xs_bind4(transport, sock)) {
1721                 sock_release(sock);
1722                 goto out;
1723         }
1724
1725         dprintk("RPC:       worker connecting xprt %p via %s to "
1726                                 "%s (port %s)\n", xprt,
1727                         xprt->address_strings[RPC_DISPLAY_PROTO],
1728                         xprt->address_strings[RPC_DISPLAY_ADDR],
1729                         xprt->address_strings[RPC_DISPLAY_PORT]);
1730
1731         xs_udp_finish_connecting(xprt, sock);
1732         status = 0;
1733 out:
1734         xprt_clear_connecting(xprt);
1735         xprt_wake_pending_tasks(xprt, status);
1736 }
1737
1738 /**
1739  * xs_udp_connect_worker6 - set up a UDP socket
1740  * @work: RPC transport to connect
1741  *
1742  * Invoked by a work queue tasklet.
1743  */
1744 static void xs_udp_connect_worker6(struct work_struct *work)
1745 {
1746         struct sock_xprt *transport =
1747                 container_of(work, struct sock_xprt, connect_worker.work);
1748         struct rpc_xprt *xprt = &transport->xprt;
1749         struct socket *sock = transport->sock;
1750         int err, status = -EIO;
1751
1752         if (xprt->shutdown)
1753                 goto out;
1754
1755         /* Start by resetting any existing state */
1756         xs_reset_transport(transport);
1757
1758         err = sock_create_kern(PF_INET6, SOCK_DGRAM, IPPROTO_UDP, &sock);
1759         if (err < 0) {
1760                 dprintk("RPC:       can't create UDP transport socket (%d).\n", -err);
1761                 goto out;
1762         }
1763         xs_reclassify_socket6(sock);
1764
1765         if (xs_bind6(transport, sock) < 0) {
1766                 sock_release(sock);
1767                 goto out;
1768         }
1769
1770         dprintk("RPC:       worker connecting xprt %p via %s to "
1771                                 "%s (port %s)\n", xprt,
1772                         xprt->address_strings[RPC_DISPLAY_PROTO],
1773                         xprt->address_strings[RPC_DISPLAY_ADDR],
1774                         xprt->address_strings[RPC_DISPLAY_PORT]);
1775
1776         xs_udp_finish_connecting(xprt, sock);
1777         status = 0;
1778 out:
1779         xprt_clear_connecting(xprt);
1780         xprt_wake_pending_tasks(xprt, status);
1781 }
1782
1783 /*
1784  * We need to preserve the port number so the reply cache on the server can
1785  * find our cached RPC replies when we get around to reconnecting.
1786  */
1787 static void xs_abort_connection(struct rpc_xprt *xprt, struct sock_xprt *transport)
1788 {
1789         int result;
1790         struct sockaddr any;
1791
1792         dprintk("RPC:       disconnecting xprt %p to reuse port\n", xprt);
1793
1794         /*
1795          * Disconnect the transport socket by doing a connect operation
1796          * with AF_UNSPEC.  This should return immediately...
1797          */
1798         memset(&any, 0, sizeof(any));
1799         any.sa_family = AF_UNSPEC;
1800         result = kernel_connect(transport->sock, &any, sizeof(any), 0);
1801         if (!result)
1802                 xs_sock_mark_closed(xprt);
1803         else
1804                 dprintk("RPC:       AF_UNSPEC connect return code %d\n",
1805                                 result);
1806 }
1807
1808 static void xs_tcp_reuse_connection(struct rpc_xprt *xprt, struct sock_xprt *transport)
1809 {
1810         unsigned int state = transport->inet->sk_state;
1811
1812         if (state == TCP_CLOSE && transport->sock->state == SS_UNCONNECTED)
1813                 return;
1814         if ((1 << state) & (TCPF_ESTABLISHED|TCPF_SYN_SENT))
1815                 return;
1816         xs_abort_connection(xprt, transport);
1817 }
1818
1819 static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
1820 {
1821         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1822
1823         if (!transport->inet) {
1824                 struct sock *sk = sock->sk;
1825
1826                 write_lock_bh(&sk->sk_callback_lock);
1827
1828                 xs_save_old_callbacks(transport, sk);
1829
1830                 sk->sk_user_data = xprt;
1831                 sk->sk_data_ready = xs_tcp_data_ready;
1832                 sk->sk_state_change = xs_tcp_state_change;
1833                 sk->sk_write_space = xs_tcp_write_space;
1834                 sk->sk_error_report = xs_error_report;
1835                 sk->sk_allocation = GFP_ATOMIC;
1836
1837                 /* socket options */
1838                 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
1839                 sock_reset_flag(sk, SOCK_LINGER);
1840                 tcp_sk(sk)->linger2 = 0;
1841                 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
1842
1843                 xprt_clear_connected(xprt);
1844
1845                 /* Reset to new socket */
1846                 transport->sock = sock;
1847                 transport->inet = sk;
1848
1849                 write_unlock_bh(&sk->sk_callback_lock);
1850         }
1851
1852         if (!xprt_bound(xprt))
1853                 return -ENOTCONN;
1854
1855         /* Tell the socket layer to start connecting... */
1856         xprt->stat.connect_count++;
1857         xprt->stat.connect_start = jiffies;
1858         return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
1859 }
1860
1861 /**
1862  * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint
1863  * @xprt: RPC transport to connect
1864  * @transport: socket transport to connect
1865  * @create_sock: function to create a socket of the correct type
1866  *
1867  * Invoked by a work queue tasklet.
1868  */
1869 static void xs_tcp_setup_socket(struct rpc_xprt *xprt,
1870                 struct sock_xprt *transport,
1871                 struct socket *(*create_sock)(struct rpc_xprt *,
1872                         struct sock_xprt *))
1873 {
1874         struct socket *sock = transport->sock;
1875         int status = -EIO;
1876
1877         if (xprt->shutdown)
1878                 goto out;
1879
1880         if (!sock) {
1881                 clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1882                 sock = create_sock(xprt, transport);
1883                 if (IS_ERR(sock)) {
1884                         status = PTR_ERR(sock);
1885                         goto out;
1886                 }
1887         } else {
1888                 int abort_and_exit;
1889
1890                 abort_and_exit = test_and_clear_bit(XPRT_CONNECTION_ABORT,
1891                                 &xprt->state);
1892                 /* "close" the socket, preserving the local port */
1893                 xs_tcp_reuse_connection(xprt, transport);
1894
1895                 if (abort_and_exit)
1896                         goto out_eagain;
1897         }
1898
1899         dprintk("RPC:       worker connecting xprt %p via %s to "
1900                                 "%s (port %s)\n", xprt,
1901                         xprt->address_strings[RPC_DISPLAY_PROTO],
1902                         xprt->address_strings[RPC_DISPLAY_ADDR],
1903                         xprt->address_strings[RPC_DISPLAY_PORT]);
1904
1905         status = xs_tcp_finish_connecting(xprt, sock);
1906         dprintk("RPC:       %p connect status %d connected %d sock state %d\n",
1907                         xprt, -status, xprt_connected(xprt),
1908                         sock->sk->sk_state);
1909         switch (status) {
1910         default:
1911                 printk("%s: connect returned unhandled error %d\n",
1912                         __func__, status);
1913         case -EADDRNOTAVAIL:
1914                 /* We're probably in TIME_WAIT. Get rid of existing socket,
1915                  * and retry
1916                  */
1917                 set_bit(XPRT_CONNECTION_CLOSE, &xprt->state);
1918                 xprt_force_disconnect(xprt);
1919                 break;
1920         case -ECONNREFUSED:
1921         case -ECONNRESET:
1922         case -ENETUNREACH:
1923                 /* retry with existing socket, after a delay */
1924         case 0:
1925         case -EINPROGRESS:
1926         case -EALREADY:
1927                 xprt_clear_connecting(xprt);
1928                 return;
1929         }
1930 out_eagain:
1931         status = -EAGAIN;
1932 out:
1933         xprt_clear_connecting(xprt);
1934         xprt_wake_pending_tasks(xprt, status);
1935 }
1936
1937 static struct socket *xs_create_tcp_sock4(struct rpc_xprt *xprt,
1938                 struct sock_xprt *transport)
1939 {
1940         struct socket *sock;
1941         int err;
1942
1943         /* start from scratch */
1944         err = sock_create_kern(PF_INET, SOCK_STREAM, IPPROTO_TCP, &sock);
1945         if (err < 0) {
1946                 dprintk("RPC:       can't create TCP transport socket (%d).\n",
1947                                 -err);
1948                 goto out_err;
1949         }
1950         xs_reclassify_socket4(sock);
1951
1952         if (xs_bind4(transport, sock) < 0) {
1953                 sock_release(sock);
1954                 goto out_err;
1955         }
1956         return sock;
1957 out_err:
1958         return ERR_PTR(-EIO);
1959 }
1960
1961 /**
1962  * xs_tcp_connect_worker4 - connect a TCP socket to a remote endpoint
1963  * @work: RPC transport to connect
1964  *
1965  * Invoked by a work queue tasklet.
1966  */
1967 static void xs_tcp_connect_worker4(struct work_struct *work)
1968 {
1969         struct sock_xprt *transport =
1970                 container_of(work, struct sock_xprt, connect_worker.work);
1971         struct rpc_xprt *xprt = &transport->xprt;
1972
1973         xs_tcp_setup_socket(xprt, transport, xs_create_tcp_sock4);
1974 }
1975
1976 static struct socket *xs_create_tcp_sock6(struct rpc_xprt *xprt,
1977                 struct sock_xprt *transport)
1978 {
1979         struct socket *sock;
1980         int err;
1981
1982         /* start from scratch */
1983         err = sock_create_kern(PF_INET6, SOCK_STREAM, IPPROTO_TCP, &sock);
1984         if (err < 0) {
1985                 dprintk("RPC:       can't create TCP transport socket (%d).\n",
1986                                 -err);
1987                 goto out_err;
1988         }
1989         xs_reclassify_socket6(sock);
1990
1991         if (xs_bind6(transport, sock) < 0) {
1992                 sock_release(sock);
1993                 goto out_err;
1994         }
1995         return sock;
1996 out_err:
1997         return ERR_PTR(-EIO);
1998 }
1999
2000 /**
2001  * xs_tcp_connect_worker6 - connect a TCP socket to a remote endpoint
2002  * @work: RPC transport to connect
2003  *
2004  * Invoked by a work queue tasklet.
2005  */
2006 static void xs_tcp_connect_worker6(struct work_struct *work)
2007 {
2008         struct sock_xprt *transport =
2009                 container_of(work, struct sock_xprt, connect_worker.work);
2010         struct rpc_xprt *xprt = &transport->xprt;
2011
2012         xs_tcp_setup_socket(xprt, transport, xs_create_tcp_sock6);
2013 }
2014
2015 /**
2016  * xs_connect - connect a socket to a remote endpoint
2017  * @task: address of RPC task that manages state of connect request
2018  *
2019  * TCP: If the remote end dropped the connection, delay reconnecting.
2020  *
2021  * UDP socket connects are synchronous, but we use a work queue anyway
2022  * to guarantee that even unprivileged user processes can set up a
2023  * socket on a privileged port.
2024  *
2025  * If a UDP socket connect fails, the delay behavior here prevents
2026  * retry floods (hard mounts).
2027  */
2028 static void xs_connect(struct rpc_task *task)
2029 {
2030         struct rpc_xprt *xprt = task->tk_xprt;
2031         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2032
2033         if (xprt_test_and_set_connecting(xprt))
2034                 return;
2035
2036         if (transport->sock != NULL) {
2037                 dprintk("RPC:       xs_connect delayed xprt %p for %lu "
2038                                 "seconds\n",
2039                                 xprt, xprt->reestablish_timeout / HZ);
2040                 queue_delayed_work(rpciod_workqueue,
2041                                    &transport->connect_worker,
2042                                    xprt->reestablish_timeout);
2043                 xprt->reestablish_timeout <<= 1;
2044                 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2045                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2046                 if (xprt->reestablish_timeout > XS_TCP_MAX_REEST_TO)
2047                         xprt->reestablish_timeout = XS_TCP_MAX_REEST_TO;
2048         } else {
2049                 dprintk("RPC:       xs_connect scheduled xprt %p\n", xprt);
2050                 queue_delayed_work(rpciod_workqueue,
2051                                    &transport->connect_worker, 0);
2052         }
2053 }
2054
2055 static void xs_tcp_connect(struct rpc_task *task)
2056 {
2057         struct rpc_xprt *xprt = task->tk_xprt;
2058
2059         /* Exit if we need to wait for socket shutdown to complete */
2060         if (test_bit(XPRT_CLOSING, &xprt->state))
2061                 return;
2062         xs_connect(task);
2063 }
2064
2065 /**
2066  * xs_udp_print_stats - display UDP socket-specifc stats
2067  * @xprt: rpc_xprt struct containing statistics
2068  * @seq: output file
2069  *
2070  */
2071 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2072 {
2073         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2074
2075         seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %Lu %Lu\n",
2076                         transport->srcport,
2077                         xprt->stat.bind_count,
2078                         xprt->stat.sends,
2079                         xprt->stat.recvs,
2080                         xprt->stat.bad_xids,
2081                         xprt->stat.req_u,
2082                         xprt->stat.bklog_u);
2083 }
2084
2085 /**
2086  * xs_tcp_print_stats - display TCP socket-specifc stats
2087  * @xprt: rpc_xprt struct containing statistics
2088  * @seq: output file
2089  *
2090  */
2091 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2092 {
2093         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2094         long idle_time = 0;
2095
2096         if (xprt_connected(xprt))
2097                 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2098
2099         seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu %Lu %Lu\n",
2100                         transport->srcport,
2101                         xprt->stat.bind_count,
2102                         xprt->stat.connect_count,
2103                         xprt->stat.connect_time,
2104                         idle_time,
2105                         xprt->stat.sends,
2106                         xprt->stat.recvs,
2107                         xprt->stat.bad_xids,
2108                         xprt->stat.req_u,
2109                         xprt->stat.bklog_u);
2110 }
2111
2112 /*
2113  * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason
2114  * we allocate pages instead doing a kmalloc like rpc_malloc is because we want
2115  * to use the server side send routines.
2116  */
2117 void *bc_malloc(struct rpc_task *task, size_t size)
2118 {
2119         struct page *page;
2120         struct rpc_buffer *buf;
2121
2122         BUG_ON(size > PAGE_SIZE - sizeof(struct rpc_buffer));
2123         page = alloc_page(GFP_KERNEL);
2124
2125         if (!page)
2126                 return NULL;
2127
2128         buf = page_address(page);
2129         buf->len = PAGE_SIZE;
2130
2131         return buf->data;
2132 }
2133
2134 /*
2135  * Free the space allocated in the bc_alloc routine
2136  */
2137 void bc_free(void *buffer)
2138 {
2139         struct rpc_buffer *buf;
2140
2141         if (!buffer)
2142                 return;
2143
2144         buf = container_of(buffer, struct rpc_buffer, data);
2145         free_page((unsigned long)buf);
2146 }
2147
2148 /*
2149  * Use the svc_sock to send the callback. Must be called with svsk->sk_mutex
2150  * held. Borrows heavily from svc_tcp_sendto and xs_tcp_send_request.
2151  */
2152 static int bc_sendto(struct rpc_rqst *req)
2153 {
2154         int len;
2155         struct xdr_buf *xbufp = &req->rq_snd_buf;
2156         struct rpc_xprt *xprt = req->rq_xprt;
2157         struct sock_xprt *transport =
2158                                 container_of(xprt, struct sock_xprt, xprt);
2159         struct socket *sock = transport->sock;
2160         unsigned long headoff;
2161         unsigned long tailoff;
2162
2163         /*
2164          * Set up the rpc header and record marker stuff
2165          */
2166         xs_encode_tcp_record_marker(xbufp);
2167
2168         tailoff = (unsigned long)xbufp->tail[0].iov_base & ~PAGE_MASK;
2169         headoff = (unsigned long)xbufp->head[0].iov_base & ~PAGE_MASK;
2170         len = svc_send_common(sock, xbufp,
2171                               virt_to_page(xbufp->head[0].iov_base), headoff,
2172                               xbufp->tail[0].iov_base, tailoff);
2173
2174         if (len != xbufp->len) {
2175                 printk(KERN_NOTICE "Error sending entire callback!\n");
2176                 len = -EAGAIN;
2177         }
2178
2179         return len;
2180 }
2181
2182 /*
2183  * The send routine. Borrows from svc_send
2184  */
2185 static int bc_send_request(struct rpc_task *task)
2186 {
2187         struct rpc_rqst *req = task->tk_rqstp;
2188         struct svc_xprt *xprt;
2189         struct svc_sock         *svsk;
2190         u32                     len;
2191
2192         dprintk("sending request with xid: %08x\n", ntohl(req->rq_xid));
2193         /*
2194          * Get the server socket associated with this callback xprt
2195          */
2196         xprt = req->rq_xprt->bc_xprt;
2197         svsk = container_of(xprt, struct svc_sock, sk_xprt);
2198
2199         /*
2200          * Grab the mutex to serialize data as the connection is shared
2201          * with the fore channel
2202          */
2203         if (!mutex_trylock(&xprt->xpt_mutex)) {
2204                 rpc_sleep_on(&xprt->xpt_bc_pending, task, NULL);
2205                 if (!mutex_trylock(&xprt->xpt_mutex))
2206                         return -EAGAIN;
2207                 rpc_wake_up_queued_task(&xprt->xpt_bc_pending, task);
2208         }
2209         if (test_bit(XPT_DEAD, &xprt->xpt_flags))
2210                 len = -ENOTCONN;
2211         else
2212                 len = bc_sendto(req);
2213         mutex_unlock(&xprt->xpt_mutex);
2214
2215         if (len > 0)
2216                 len = 0;
2217
2218         return len;
2219 }
2220
2221 /*
2222  * The close routine. Since this is client initiated, we do nothing
2223  */
2224
2225 static void bc_close(struct rpc_xprt *xprt)
2226 {
2227         return;
2228 }
2229
2230 /*
2231  * The xprt destroy routine. Again, because this connection is client
2232  * initiated, we do nothing
2233  */
2234
2235 static void bc_destroy(struct rpc_xprt *xprt)
2236 {
2237         return;
2238 }
2239
2240 static struct rpc_xprt_ops xs_udp_ops = {
2241         .set_buffer_size        = xs_udp_set_buffer_size,
2242         .reserve_xprt           = xprt_reserve_xprt_cong,
2243         .release_xprt           = xprt_release_xprt_cong,
2244         .rpcbind                = rpcb_getport_async,
2245         .set_port               = xs_set_port,
2246         .connect                = xs_connect,
2247         .buf_alloc              = rpc_malloc,
2248         .buf_free               = rpc_free,
2249         .send_request           = xs_udp_send_request,
2250         .set_retrans_timeout    = xprt_set_retrans_timeout_rtt,
2251         .timer                  = xs_udp_timer,
2252         .release_request        = xprt_release_rqst_cong,
2253         .close                  = xs_close,
2254         .destroy                = xs_destroy,
2255         .print_stats            = xs_udp_print_stats,
2256 };
2257
2258 static struct rpc_xprt_ops xs_tcp_ops = {
2259         .reserve_xprt           = xprt_reserve_xprt,
2260         .release_xprt           = xs_tcp_release_xprt,
2261         .rpcbind                = rpcb_getport_async,
2262         .set_port               = xs_set_port,
2263         .connect                = xs_tcp_connect,
2264         .buf_alloc              = rpc_malloc,
2265         .buf_free               = rpc_free,
2266         .send_request           = xs_tcp_send_request,
2267         .set_retrans_timeout    = xprt_set_retrans_timeout_def,
2268 #if defined(CONFIG_NFS_V4_1)
2269         .release_request        = bc_release_request,
2270 #endif /* CONFIG_NFS_V4_1 */
2271         .close                  = xs_tcp_close,
2272         .destroy                = xs_destroy,
2273         .print_stats            = xs_tcp_print_stats,
2274 };
2275
2276 /*
2277  * The rpc_xprt_ops for the server backchannel
2278  */
2279
2280 static struct rpc_xprt_ops bc_tcp_ops = {
2281         .reserve_xprt           = xprt_reserve_xprt,
2282         .release_xprt           = xprt_release_xprt,
2283         .buf_alloc              = bc_malloc,
2284         .buf_free               = bc_free,
2285         .send_request           = bc_send_request,
2286         .set_retrans_timeout    = xprt_set_retrans_timeout_def,
2287         .close                  = bc_close,
2288         .destroy                = bc_destroy,
2289         .print_stats            = xs_tcp_print_stats,
2290 };
2291
2292 static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
2293                                       unsigned int slot_table_size)
2294 {
2295         struct rpc_xprt *xprt;
2296         struct sock_xprt *new;
2297
2298         if (args->addrlen > sizeof(xprt->addr)) {
2299                 dprintk("RPC:       xs_setup_xprt: address too large\n");
2300                 return ERR_PTR(-EBADF);
2301         }
2302
2303         new = kzalloc(sizeof(*new), GFP_KERNEL);
2304         if (new == NULL) {
2305                 dprintk("RPC:       xs_setup_xprt: couldn't allocate "
2306                                 "rpc_xprt\n");
2307                 return ERR_PTR(-ENOMEM);
2308         }
2309         xprt = &new->xprt;
2310
2311         xprt->max_reqs = slot_table_size;
2312         xprt->slot = kcalloc(xprt->max_reqs, sizeof(struct rpc_rqst), GFP_KERNEL);
2313         if (xprt->slot == NULL) {
2314                 kfree(xprt);
2315                 dprintk("RPC:       xs_setup_xprt: couldn't allocate slot "
2316                                 "table\n");
2317                 return ERR_PTR(-ENOMEM);
2318         }
2319
2320         memcpy(&xprt->addr, args->dstaddr, args->addrlen);
2321         xprt->addrlen = args->addrlen;
2322         if (args->srcaddr)
2323                 memcpy(&new->srcaddr, args->srcaddr, args->addrlen);
2324
2325         return xprt;
2326 }
2327
2328 static const struct rpc_timeout xs_udp_default_timeout = {
2329         .to_initval = 5 * HZ,
2330         .to_maxval = 30 * HZ,
2331         .to_increment = 5 * HZ,
2332         .to_retries = 5,
2333 };
2334
2335 /**
2336  * xs_setup_udp - Set up transport to use a UDP socket
2337  * @args: rpc transport creation arguments
2338  *
2339  */
2340 static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
2341 {
2342         struct sockaddr *addr = args->dstaddr;
2343         struct rpc_xprt *xprt;
2344         struct sock_xprt *transport;
2345
2346         xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries);
2347         if (IS_ERR(xprt))
2348                 return xprt;
2349         transport = container_of(xprt, struct sock_xprt, xprt);
2350
2351         xprt->prot = IPPROTO_UDP;
2352         xprt->tsh_size = 0;
2353         /* XXX: header size can vary due to auth type, IPv6, etc. */
2354         xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
2355
2356         xprt->bind_timeout = XS_BIND_TO;
2357         xprt->connect_timeout = XS_UDP_CONN_TO;
2358         xprt->reestablish_timeout = XS_UDP_REEST_TO;
2359         xprt->idle_timeout = XS_IDLE_DISC_TO;
2360
2361         xprt->ops = &xs_udp_ops;
2362
2363         xprt->timeout = &xs_udp_default_timeout;
2364
2365         switch (addr->sa_family) {
2366         case AF_INET:
2367                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2368                         xprt_set_bound(xprt);
2369
2370                 INIT_DELAYED_WORK(&transport->connect_worker,
2371                                         xs_udp_connect_worker4);
2372                 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
2373                 break;
2374         case AF_INET6:
2375                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2376                         xprt_set_bound(xprt);
2377
2378                 INIT_DELAYED_WORK(&transport->connect_worker,
2379                                         xs_udp_connect_worker6);
2380                 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
2381                 break;
2382         default:
2383                 kfree(xprt);
2384                 return ERR_PTR(-EAFNOSUPPORT);
2385         }
2386
2387         if (xprt_bound(xprt))
2388                 dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
2389                                 xprt->address_strings[RPC_DISPLAY_ADDR],
2390                                 xprt->address_strings[RPC_DISPLAY_PORT],
2391                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
2392         else
2393                 dprintk("RPC:       set up xprt to %s (autobind) via %s\n",
2394                                 xprt->address_strings[RPC_DISPLAY_ADDR],
2395                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
2396
2397         if (try_module_get(THIS_MODULE))
2398                 return xprt;
2399
2400         kfree(xprt->slot);
2401         kfree(xprt);
2402         return ERR_PTR(-EINVAL);
2403 }
2404
2405 static const struct rpc_timeout xs_tcp_default_timeout = {
2406         .to_initval = 60 * HZ,
2407         .to_maxval = 60 * HZ,
2408         .to_retries = 2,
2409 };
2410
2411 /**
2412  * xs_setup_tcp - Set up transport to use a TCP socket
2413  * @args: rpc transport creation arguments
2414  *
2415  */
2416 static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
2417 {
2418         struct sockaddr *addr = args->dstaddr;
2419         struct rpc_xprt *xprt;
2420         struct sock_xprt *transport;
2421
2422         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries);
2423         if (IS_ERR(xprt))
2424                 return xprt;
2425         transport = container_of(xprt, struct sock_xprt, xprt);
2426
2427         xprt->prot = IPPROTO_TCP;
2428         xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2429         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2430
2431         xprt->bind_timeout = XS_BIND_TO;
2432         xprt->connect_timeout = XS_TCP_CONN_TO;
2433         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2434         xprt->idle_timeout = XS_IDLE_DISC_TO;
2435
2436         xprt->ops = &xs_tcp_ops;
2437         xprt->timeout = &xs_tcp_default_timeout;
2438
2439         switch (addr->sa_family) {
2440         case AF_INET:
2441                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2442                         xprt_set_bound(xprt);
2443
2444                 INIT_DELAYED_WORK(&transport->connect_worker,
2445                                         xs_tcp_connect_worker4);
2446                 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
2447                 break;
2448         case AF_INET6:
2449                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2450                         xprt_set_bound(xprt);
2451
2452                 INIT_DELAYED_WORK(&transport->connect_worker,
2453                                         xs_tcp_connect_worker6);
2454                 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
2455                 break;
2456         default:
2457                 kfree(xprt);
2458                 return ERR_PTR(-EAFNOSUPPORT);
2459         }
2460
2461         if (xprt_bound(xprt))
2462                 dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
2463                                 xprt->address_strings[RPC_DISPLAY_ADDR],
2464                                 xprt->address_strings[RPC_DISPLAY_PORT],
2465                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
2466         else
2467                 dprintk("RPC:       set up xprt to %s (autobind) via %s\n",
2468                                 xprt->address_strings[RPC_DISPLAY_ADDR],
2469                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
2470
2471
2472         if (try_module_get(THIS_MODULE))
2473                 return xprt;
2474
2475         kfree(xprt->slot);
2476         kfree(xprt);
2477         return ERR_PTR(-EINVAL);
2478 }
2479
2480 /**
2481  * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket
2482  * @args: rpc transport creation arguments
2483  *
2484  */
2485 static struct rpc_xprt *xs_setup_bc_tcp(struct xprt_create *args)
2486 {
2487         struct sockaddr *addr = args->dstaddr;
2488         struct rpc_xprt *xprt;
2489         struct sock_xprt *transport;
2490         struct svc_sock *bc_sock;
2491
2492         if (!args->bc_xprt)
2493                 ERR_PTR(-EINVAL);
2494
2495         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries);
2496         if (IS_ERR(xprt))
2497                 return xprt;
2498         transport = container_of(xprt, struct sock_xprt, xprt);
2499
2500         xprt->prot = IPPROTO_TCP;
2501         xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2502         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2503         xprt->timeout = &xs_tcp_default_timeout;
2504
2505         /* backchannel */
2506         xprt_set_bound(xprt);
2507         xprt->bind_timeout = 0;
2508         xprt->connect_timeout = 0;
2509         xprt->reestablish_timeout = 0;
2510         xprt->idle_timeout = 0;
2511
2512         /*
2513          * The backchannel uses the same socket connection as the
2514          * forechannel
2515          */
2516         xprt->bc_xprt = args->bc_xprt;
2517         bc_sock = container_of(args->bc_xprt, struct svc_sock, sk_xprt);
2518         bc_sock->sk_bc_xprt = xprt;
2519         transport->sock = bc_sock->sk_sock;
2520         transport->inet = bc_sock->sk_sk;
2521
2522         xprt->ops = &bc_tcp_ops;
2523
2524         switch (addr->sa_family) {
2525         case AF_INET:
2526                 xs_format_peer_addresses(xprt, "tcp",
2527                                          RPCBIND_NETID_TCP);
2528                 break;
2529         case AF_INET6:
2530                 xs_format_peer_addresses(xprt, "tcp",
2531                                    RPCBIND_NETID_TCP6);
2532                 break;
2533         default:
2534                 kfree(xprt);
2535                 return ERR_PTR(-EAFNOSUPPORT);
2536         }
2537
2538         if (xprt_bound(xprt))
2539                 dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
2540                                 xprt->address_strings[RPC_DISPLAY_ADDR],
2541                                 xprt->address_strings[RPC_DISPLAY_PORT],
2542                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
2543         else
2544                 dprintk("RPC:       set up xprt to %s (autobind) via %s\n",
2545                                 xprt->address_strings[RPC_DISPLAY_ADDR],
2546                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
2547
2548         /*
2549          * Since we don't want connections for the backchannel, we set
2550          * the xprt status to connected
2551          */
2552         xprt_set_connected(xprt);
2553
2554
2555         if (try_module_get(THIS_MODULE))
2556                 return xprt;
2557         kfree(xprt->slot);
2558         kfree(xprt);
2559         return ERR_PTR(-EINVAL);
2560 }
2561
2562 static struct xprt_class        xs_udp_transport = {
2563         .list           = LIST_HEAD_INIT(xs_udp_transport.list),
2564         .name           = "udp",
2565         .owner          = THIS_MODULE,
2566         .ident          = XPRT_TRANSPORT_UDP,
2567         .setup          = xs_setup_udp,
2568 };
2569
2570 static struct xprt_class        xs_tcp_transport = {
2571         .list           = LIST_HEAD_INIT(xs_tcp_transport.list),
2572         .name           = "tcp",
2573         .owner          = THIS_MODULE,
2574         .ident          = XPRT_TRANSPORT_TCP,
2575         .setup          = xs_setup_tcp,
2576 };
2577
2578 static struct xprt_class        xs_bc_tcp_transport = {
2579         .list           = LIST_HEAD_INIT(xs_bc_tcp_transport.list),
2580         .name           = "tcp NFSv4.1 backchannel",
2581         .owner          = THIS_MODULE,
2582         .ident          = XPRT_TRANSPORT_BC_TCP,
2583         .setup          = xs_setup_bc_tcp,
2584 };
2585
2586 /**
2587  * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
2588  *
2589  */
2590 int init_socket_xprt(void)
2591 {
2592 #ifdef RPC_DEBUG
2593         if (!sunrpc_table_header)
2594                 sunrpc_table_header = register_sysctl_table(sunrpc_table);
2595 #endif
2596
2597         xprt_register_transport(&xs_udp_transport);
2598         xprt_register_transport(&xs_tcp_transport);
2599         xprt_register_transport(&xs_bc_tcp_transport);
2600
2601         return 0;
2602 }
2603
2604 /**
2605  * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
2606  *
2607  */
2608 void cleanup_socket_xprt(void)
2609 {
2610 #ifdef RPC_DEBUG
2611         if (sunrpc_table_header) {
2612                 unregister_sysctl_table(sunrpc_table_header);
2613                 sunrpc_table_header = NULL;
2614         }
2615 #endif
2616
2617         xprt_unregister_transport(&xs_udp_transport);
2618         xprt_unregister_transport(&xs_tcp_transport);
2619         xprt_unregister_transport(&xs_bc_tcp_transport);
2620 }
2621
2622 static int param_set_uint_minmax(const char *val, struct kernel_param *kp,
2623                 unsigned int min, unsigned int max)
2624 {
2625         unsigned long num;
2626         int ret;
2627
2628         if (!val)
2629                 return -EINVAL;
2630         ret = strict_strtoul(val, 0, &num);
2631         if (ret == -EINVAL || num < min || num > max)
2632                 return -EINVAL;
2633         *((unsigned int *)kp->arg) = num;
2634         return 0;
2635 }
2636
2637 static int param_set_portnr(const char *val, struct kernel_param *kp)
2638 {
2639         return param_set_uint_minmax(val, kp,
2640                         RPC_MIN_RESVPORT,
2641                         RPC_MAX_RESVPORT);
2642 }
2643
2644 static int param_get_portnr(char *buffer, struct kernel_param *kp)
2645 {
2646         return param_get_uint(buffer, kp);
2647 }
2648 #define param_check_portnr(name, p) \
2649         __param_check(name, p, unsigned int);
2650
2651 module_param_named(min_resvport, xprt_min_resvport, portnr, 0644);
2652 module_param_named(max_resvport, xprt_max_resvport, portnr, 0644);
2653
2654 static int param_set_slot_table_size(const char *val, struct kernel_param *kp)
2655 {
2656         return param_set_uint_minmax(val, kp,
2657                         RPC_MIN_SLOT_TABLE,
2658                         RPC_MAX_SLOT_TABLE);
2659 }
2660
2661 static int param_get_slot_table_size(char *buffer, struct kernel_param *kp)
2662 {
2663         return param_get_uint(buffer, kp);
2664 }
2665 #define param_check_slot_table_size(name, p) \
2666         __param_check(name, p, unsigned int);
2667
2668 module_param_named(tcp_slot_table_entries, xprt_tcp_slot_table_entries,
2669                    slot_table_size, 0644);
2670 module_param_named(udp_slot_table_entries, xprt_udp_slot_table_entries,
2671                    slot_table_size, 0644);
2672