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