Merge branch 'for-2.6.38' of git://linux-nfs.org/~bfields/linux
[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  * Wait duration for a reply from the RPC portmapper.
142  */
143 #define XS_BIND_TO              (60U * HZ)
144
145 /*
146  * Delay if a UDP socket connect error occurs.  This is most likely some
147  * kind of resource problem on the local host.
148  */
149 #define XS_UDP_REEST_TO         (2U * HZ)
150
151 /*
152  * The reestablish timeout allows clients to delay for a bit before attempting
153  * to reconnect to a server that just dropped our connection.
154  *
155  * We implement an exponential backoff when trying to reestablish a TCP
156  * transport connection with the server.  Some servers like to drop a TCP
157  * connection when they are overworked, so we start with a short timeout and
158  * increase over time if the server is down or not responding.
159  */
160 #define XS_TCP_INIT_REEST_TO    (3U * HZ)
161 #define XS_TCP_MAX_REEST_TO     (5U * 60 * HZ)
162
163 /*
164  * TCP idle timeout; client drops the transport socket if it is idle
165  * for this long.  Note that we also timeout UDP sockets to prevent
166  * holding port numbers when there is no RPC traffic.
167  */
168 #define XS_IDLE_DISC_TO         (5U * 60 * HZ)
169
170 #ifdef RPC_DEBUG
171 # undef  RPC_DEBUG_DATA
172 # define RPCDBG_FACILITY        RPCDBG_TRANS
173 #endif
174
175 #ifdef RPC_DEBUG_DATA
176 static void xs_pktdump(char *msg, u32 *packet, unsigned int count)
177 {
178         u8 *buf = (u8 *) packet;
179         int j;
180
181         dprintk("RPC:       %s\n", msg);
182         for (j = 0; j < count && j < 128; j += 4) {
183                 if (!(j & 31)) {
184                         if (j)
185                                 dprintk("\n");
186                         dprintk("0x%04x ", j);
187                 }
188                 dprintk("%02x%02x%02x%02x ",
189                         buf[j], buf[j+1], buf[j+2], buf[j+3]);
190         }
191         dprintk("\n");
192 }
193 #else
194 static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count)
195 {
196         /* NOP */
197 }
198 #endif
199
200 struct sock_xprt {
201         struct rpc_xprt         xprt;
202
203         /*
204          * Network layer
205          */
206         struct socket *         sock;
207         struct sock *           inet;
208
209         /*
210          * State of TCP reply receive
211          */
212         __be32                  tcp_fraghdr,
213                                 tcp_xid,
214                                 tcp_calldir;
215
216         u32                     tcp_offset,
217                                 tcp_reclen;
218
219         unsigned long           tcp_copied,
220                                 tcp_flags;
221
222         /*
223          * Connection of transports
224          */
225         struct delayed_work     connect_worker;
226         struct sockaddr_storage srcaddr;
227         unsigned short          srcport;
228
229         /*
230          * UDP socket buffer size parameters
231          */
232         size_t                  rcvsize,
233                                 sndsize;
234
235         /*
236          * Saved socket callback addresses
237          */
238         void                    (*old_data_ready)(struct sock *, int);
239         void                    (*old_state_change)(struct sock *);
240         void                    (*old_write_space)(struct sock *);
241         void                    (*old_error_report)(struct sock *);
242 };
243
244 /*
245  * TCP receive state flags
246  */
247 #define TCP_RCV_LAST_FRAG       (1UL << 0)
248 #define TCP_RCV_COPY_FRAGHDR    (1UL << 1)
249 #define TCP_RCV_COPY_XID        (1UL << 2)
250 #define TCP_RCV_COPY_DATA       (1UL << 3)
251 #define TCP_RCV_READ_CALLDIR    (1UL << 4)
252 #define TCP_RCV_COPY_CALLDIR    (1UL << 5)
253
254 /*
255  * TCP RPC flags
256  */
257 #define TCP_RPC_REPLY           (1UL << 6)
258
259 static inline struct sockaddr *xs_addr(struct rpc_xprt *xprt)
260 {
261         return (struct sockaddr *) &xprt->addr;
262 }
263
264 static inline struct sockaddr_in *xs_addr_in(struct rpc_xprt *xprt)
265 {
266         return (struct sockaddr_in *) &xprt->addr;
267 }
268
269 static inline struct sockaddr_in6 *xs_addr_in6(struct rpc_xprt *xprt)
270 {
271         return (struct sockaddr_in6 *) &xprt->addr;
272 }
273
274 static void xs_format_common_peer_addresses(struct rpc_xprt *xprt)
275 {
276         struct sockaddr *sap = xs_addr(xprt);
277         struct sockaddr_in6 *sin6;
278         struct sockaddr_in *sin;
279         char buf[128];
280
281         (void)rpc_ntop(sap, buf, sizeof(buf));
282         xprt->address_strings[RPC_DISPLAY_ADDR] = kstrdup(buf, GFP_KERNEL);
283
284         switch (sap->sa_family) {
285         case AF_INET:
286                 sin = xs_addr_in(xprt);
287                 snprintf(buf, sizeof(buf), "%08x", ntohl(sin->sin_addr.s_addr));
288                 break;
289         case AF_INET6:
290                 sin6 = xs_addr_in6(xprt);
291                 snprintf(buf, sizeof(buf), "%pi6", &sin6->sin6_addr);
292                 break;
293         default:
294                 BUG();
295         }
296         xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = kstrdup(buf, GFP_KERNEL);
297 }
298
299 static void xs_format_common_peer_ports(struct rpc_xprt *xprt)
300 {
301         struct sockaddr *sap = xs_addr(xprt);
302         char buf[128];
303
304         snprintf(buf, sizeof(buf), "%u", rpc_get_port(sap));
305         xprt->address_strings[RPC_DISPLAY_PORT] = kstrdup(buf, GFP_KERNEL);
306
307         snprintf(buf, sizeof(buf), "%4hx", rpc_get_port(sap));
308         xprt->address_strings[RPC_DISPLAY_HEX_PORT] = kstrdup(buf, GFP_KERNEL);
309 }
310
311 static void xs_format_peer_addresses(struct rpc_xprt *xprt,
312                                      const char *protocol,
313                                      const char *netid)
314 {
315         xprt->address_strings[RPC_DISPLAY_PROTO] = protocol;
316         xprt->address_strings[RPC_DISPLAY_NETID] = netid;
317         xs_format_common_peer_addresses(xprt);
318         xs_format_common_peer_ports(xprt);
319 }
320
321 static void xs_update_peer_port(struct rpc_xprt *xprt)
322 {
323         kfree(xprt->address_strings[RPC_DISPLAY_HEX_PORT]);
324         kfree(xprt->address_strings[RPC_DISPLAY_PORT]);
325
326         xs_format_common_peer_ports(xprt);
327 }
328
329 static void xs_free_peer_addresses(struct rpc_xprt *xprt)
330 {
331         unsigned int i;
332
333         for (i = 0; i < RPC_DISPLAY_MAX; i++)
334                 switch (i) {
335                 case RPC_DISPLAY_PROTO:
336                 case RPC_DISPLAY_NETID:
337                         continue;
338                 default:
339                         kfree(xprt->address_strings[i]);
340                 }
341 }
342
343 #define XS_SENDMSG_FLAGS        (MSG_DONTWAIT | MSG_NOSIGNAL)
344
345 static int xs_send_kvec(struct socket *sock, struct sockaddr *addr, int addrlen, struct kvec *vec, unsigned int base, int more)
346 {
347         struct msghdr msg = {
348                 .msg_name       = addr,
349                 .msg_namelen    = addrlen,
350                 .msg_flags      = XS_SENDMSG_FLAGS | (more ? MSG_MORE : 0),
351         };
352         struct kvec iov = {
353                 .iov_base       = vec->iov_base + base,
354                 .iov_len        = vec->iov_len - base,
355         };
356
357         if (iov.iov_len != 0)
358                 return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
359         return kernel_sendmsg(sock, &msg, NULL, 0, 0);
360 }
361
362 static int xs_send_pagedata(struct socket *sock, struct xdr_buf *xdr, unsigned int base, int more)
363 {
364         struct page **ppage;
365         unsigned int remainder;
366         int err, sent = 0;
367
368         remainder = xdr->page_len - base;
369         base += xdr->page_base;
370         ppage = xdr->pages + (base >> PAGE_SHIFT);
371         base &= ~PAGE_MASK;
372         for(;;) {
373                 unsigned int len = min_t(unsigned int, PAGE_SIZE - base, remainder);
374                 int flags = XS_SENDMSG_FLAGS;
375
376                 remainder -= len;
377                 if (remainder != 0 || more)
378                         flags |= MSG_MORE;
379                 err = sock->ops->sendpage(sock, *ppage, base, len, flags);
380                 if (remainder == 0 || err != len)
381                         break;
382                 sent += err;
383                 ppage++;
384                 base = 0;
385         }
386         if (sent == 0)
387                 return err;
388         if (err > 0)
389                 sent += err;
390         return sent;
391 }
392
393 /**
394  * xs_sendpages - write pages directly to a socket
395  * @sock: socket to send on
396  * @addr: UDP only -- address of destination
397  * @addrlen: UDP only -- length of destination address
398  * @xdr: buffer containing this request
399  * @base: starting position in the buffer
400  *
401  */
402 static int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base)
403 {
404         unsigned int remainder = xdr->len - base;
405         int err, sent = 0;
406
407         if (unlikely(!sock))
408                 return -ENOTSOCK;
409
410         clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags);
411         if (base != 0) {
412                 addr = NULL;
413                 addrlen = 0;
414         }
415
416         if (base < xdr->head[0].iov_len || addr != NULL) {
417                 unsigned int len = xdr->head[0].iov_len - base;
418                 remainder -= len;
419                 err = xs_send_kvec(sock, addr, addrlen, &xdr->head[0], base, remainder != 0);
420                 if (remainder == 0 || err != len)
421                         goto out;
422                 sent += err;
423                 base = 0;
424         } else
425                 base -= xdr->head[0].iov_len;
426
427         if (base < xdr->page_len) {
428                 unsigned int len = xdr->page_len - base;
429                 remainder -= len;
430                 err = xs_send_pagedata(sock, xdr, base, remainder != 0);
431                 if (remainder == 0 || err != len)
432                         goto out;
433                 sent += err;
434                 base = 0;
435         } else
436                 base -= xdr->page_len;
437
438         if (base >= xdr->tail[0].iov_len)
439                 return sent;
440         err = xs_send_kvec(sock, NULL, 0, &xdr->tail[0], base, 0);
441 out:
442         if (sent == 0)
443                 return err;
444         if (err > 0)
445                 sent += err;
446         return sent;
447 }
448
449 static void xs_nospace_callback(struct rpc_task *task)
450 {
451         struct sock_xprt *transport = container_of(task->tk_rqstp->rq_xprt, struct sock_xprt, xprt);
452
453         transport->inet->sk_write_pending--;
454         clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
455 }
456
457 /**
458  * xs_nospace - place task on wait queue if transmit was incomplete
459  * @task: task to put to sleep
460  *
461  */
462 static int xs_nospace(struct rpc_task *task)
463 {
464         struct rpc_rqst *req = task->tk_rqstp;
465         struct rpc_xprt *xprt = req->rq_xprt;
466         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
467         int ret = 0;
468
469         dprintk("RPC: %5u xmit incomplete (%u left of %u)\n",
470                         task->tk_pid, req->rq_slen - req->rq_bytes_sent,
471                         req->rq_slen);
472
473         /* Protect against races with write_space */
474         spin_lock_bh(&xprt->transport_lock);
475
476         /* Don't race with disconnect */
477         if (xprt_connected(xprt)) {
478                 if (test_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags)) {
479                         ret = -EAGAIN;
480                         /*
481                          * Notify TCP that we're limited by the application
482                          * window size
483                          */
484                         set_bit(SOCK_NOSPACE, &transport->sock->flags);
485                         transport->inet->sk_write_pending++;
486                         /* ...and wait for more buffer space */
487                         xprt_wait_for_buffer_space(task, xs_nospace_callback);
488                 }
489         } else {
490                 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
491                 ret = -ENOTCONN;
492         }
493
494         spin_unlock_bh(&xprt->transport_lock);
495         return ret;
496 }
497
498 /**
499  * xs_udp_send_request - write an RPC request to a UDP socket
500  * @task: address of RPC task that manages the state of an RPC request
501  *
502  * Return values:
503  *        0:    The request has been sent
504  *   EAGAIN:    The socket was blocked, please call again later to
505  *              complete the request
506  * ENOTCONN:    Caller needs to invoke connect logic then call again
507  *    other:    Some other error occured, the request was not sent
508  */
509 static int xs_udp_send_request(struct rpc_task *task)
510 {
511         struct rpc_rqst *req = task->tk_rqstp;
512         struct rpc_xprt *xprt = req->rq_xprt;
513         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
514         struct xdr_buf *xdr = &req->rq_snd_buf;
515         int status;
516
517         xs_pktdump("packet data:",
518                                 req->rq_svec->iov_base,
519                                 req->rq_svec->iov_len);
520
521         if (!xprt_bound(xprt))
522                 return -ENOTCONN;
523         status = xs_sendpages(transport->sock,
524                               xs_addr(xprt),
525                               xprt->addrlen, xdr,
526                               req->rq_bytes_sent);
527
528         dprintk("RPC:       xs_udp_send_request(%u) = %d\n",
529                         xdr->len - req->rq_bytes_sent, status);
530
531         if (status >= 0) {
532                 req->rq_xmit_bytes_sent += status;
533                 if (status >= req->rq_slen)
534                         return 0;
535                 /* Still some bytes left; set up for a retry later. */
536                 status = -EAGAIN;
537         }
538
539         switch (status) {
540         case -ENOTSOCK:
541                 status = -ENOTCONN;
542                 /* Should we call xs_close() here? */
543                 break;
544         case -EAGAIN:
545                 status = xs_nospace(task);
546                 break;
547         default:
548                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
549                         -status);
550         case -ENETUNREACH:
551         case -EPIPE:
552         case -ECONNREFUSED:
553                 /* When the server has died, an ICMP port unreachable message
554                  * prompts ECONNREFUSED. */
555                 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
556         }
557
558         return status;
559 }
560
561 /**
562  * xs_tcp_shutdown - gracefully shut down a TCP socket
563  * @xprt: transport
564  *
565  * Initiates a graceful shutdown of the TCP socket by calling the
566  * equivalent of shutdown(SHUT_WR);
567  */
568 static void xs_tcp_shutdown(struct rpc_xprt *xprt)
569 {
570         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
571         struct socket *sock = transport->sock;
572
573         if (sock != NULL)
574                 kernel_sock_shutdown(sock, SHUT_WR);
575 }
576
577 static inline void xs_encode_tcp_record_marker(struct xdr_buf *buf)
578 {
579         u32 reclen = buf->len - sizeof(rpc_fraghdr);
580         rpc_fraghdr *base = buf->head[0].iov_base;
581         *base = htonl(RPC_LAST_STREAM_FRAGMENT | reclen);
582 }
583
584 /**
585  * xs_tcp_send_request - write an RPC request to a TCP socket
586  * @task: address of RPC task that manages the state of an RPC request
587  *
588  * Return values:
589  *        0:    The request has been sent
590  *   EAGAIN:    The socket was blocked, please call again later to
591  *              complete the request
592  * ENOTCONN:    Caller needs to invoke connect logic then call again
593  *    other:    Some other error occured, the request was not sent
594  *
595  * XXX: In the case of soft timeouts, should we eventually give up
596  *      if sendmsg is not able to make progress?
597  */
598 static int xs_tcp_send_request(struct rpc_task *task)
599 {
600         struct rpc_rqst *req = task->tk_rqstp;
601         struct rpc_xprt *xprt = req->rq_xprt;
602         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
603         struct xdr_buf *xdr = &req->rq_snd_buf;
604         int status;
605
606         xs_encode_tcp_record_marker(&req->rq_snd_buf);
607
608         xs_pktdump("packet data:",
609                                 req->rq_svec->iov_base,
610                                 req->rq_svec->iov_len);
611
612         /* Continue transmitting the packet/record. We must be careful
613          * to cope with writespace callbacks arriving _after_ we have
614          * called sendmsg(). */
615         while (1) {
616                 status = xs_sendpages(transport->sock,
617                                         NULL, 0, xdr, req->rq_bytes_sent);
618
619                 dprintk("RPC:       xs_tcp_send_request(%u) = %d\n",
620                                 xdr->len - req->rq_bytes_sent, status);
621
622                 if (unlikely(status < 0))
623                         break;
624
625                 /* If we've sent the entire packet, immediately
626                  * reset the count of bytes sent. */
627                 req->rq_bytes_sent += status;
628                 req->rq_xmit_bytes_sent += status;
629                 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
630                         req->rq_bytes_sent = 0;
631                         return 0;
632                 }
633
634                 if (status != 0)
635                         continue;
636                 status = -EAGAIN;
637                 break;
638         }
639
640         switch (status) {
641         case -ENOTSOCK:
642                 status = -ENOTCONN;
643                 /* Should we call xs_close() here? */
644                 break;
645         case -EAGAIN:
646                 status = xs_nospace(task);
647                 break;
648         default:
649                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
650                         -status);
651         case -ECONNRESET:
652         case -EPIPE:
653                 xs_tcp_shutdown(xprt);
654         case -ECONNREFUSED:
655         case -ENOTCONN:
656                 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
657         }
658
659         return status;
660 }
661
662 /**
663  * xs_tcp_release_xprt - clean up after a tcp transmission
664  * @xprt: transport
665  * @task: rpc task
666  *
667  * This cleans up if an error causes us to abort the transmission of a request.
668  * In this case, the socket may need to be reset in order to avoid confusing
669  * the server.
670  */
671 static void xs_tcp_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
672 {
673         struct rpc_rqst *req;
674
675         if (task != xprt->snd_task)
676                 return;
677         if (task == NULL)
678                 goto out_release;
679         req = task->tk_rqstp;
680         if (req->rq_bytes_sent == 0)
681                 goto out_release;
682         if (req->rq_bytes_sent == req->rq_snd_buf.len)
683                 goto out_release;
684         set_bit(XPRT_CLOSE_WAIT, &task->tk_xprt->state);
685 out_release:
686         xprt_release_xprt(xprt, task);
687 }
688
689 static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk)
690 {
691         transport->old_data_ready = sk->sk_data_ready;
692         transport->old_state_change = sk->sk_state_change;
693         transport->old_write_space = sk->sk_write_space;
694         transport->old_error_report = sk->sk_error_report;
695 }
696
697 static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk)
698 {
699         sk->sk_data_ready = transport->old_data_ready;
700         sk->sk_state_change = transport->old_state_change;
701         sk->sk_write_space = transport->old_write_space;
702         sk->sk_error_report = transport->old_error_report;
703 }
704
705 static void xs_reset_transport(struct sock_xprt *transport)
706 {
707         struct socket *sock = transport->sock;
708         struct sock *sk = transport->inet;
709
710         if (sk == NULL)
711                 return;
712
713         write_lock_bh(&sk->sk_callback_lock);
714         transport->inet = NULL;
715         transport->sock = NULL;
716
717         sk->sk_user_data = NULL;
718
719         xs_restore_old_callbacks(transport, sk);
720         write_unlock_bh(&sk->sk_callback_lock);
721
722         sk->sk_no_check = 0;
723
724         sock_release(sock);
725 }
726
727 /**
728  * xs_close - close a socket
729  * @xprt: transport
730  *
731  * This is used when all requests are complete; ie, no DRC state remains
732  * on the server we want to save.
733  *
734  * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with
735  * xs_reset_transport() zeroing the socket from underneath a writer.
736  */
737 static void xs_close(struct rpc_xprt *xprt)
738 {
739         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
740
741         dprintk("RPC:       xs_close xprt %p\n", xprt);
742
743         xs_reset_transport(transport);
744         xprt->reestablish_timeout = 0;
745
746         smp_mb__before_clear_bit();
747         clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
748         clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
749         clear_bit(XPRT_CLOSING, &xprt->state);
750         smp_mb__after_clear_bit();
751         xprt_disconnect_done(xprt);
752 }
753
754 static void xs_tcp_close(struct rpc_xprt *xprt)
755 {
756         if (test_and_clear_bit(XPRT_CONNECTION_CLOSE, &xprt->state))
757                 xs_close(xprt);
758         else
759                 xs_tcp_shutdown(xprt);
760 }
761
762 /**
763  * xs_destroy - prepare to shutdown a transport
764  * @xprt: doomed transport
765  *
766  */
767 static void xs_destroy(struct rpc_xprt *xprt)
768 {
769         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
770
771         dprintk("RPC:       xs_destroy xprt %p\n", xprt);
772
773         cancel_delayed_work_sync(&transport->connect_worker);
774
775         xs_close(xprt);
776         xs_free_peer_addresses(xprt);
777         xprt_free(xprt);
778         module_put(THIS_MODULE);
779 }
780
781 static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
782 {
783         return (struct rpc_xprt *) sk->sk_user_data;
784 }
785
786 /**
787  * xs_udp_data_ready - "data ready" callback for UDP sockets
788  * @sk: socket with data to read
789  * @len: how much data to read
790  *
791  */
792 static void xs_udp_data_ready(struct sock *sk, int len)
793 {
794         struct rpc_task *task;
795         struct rpc_xprt *xprt;
796         struct rpc_rqst *rovr;
797         struct sk_buff *skb;
798         int err, repsize, copied;
799         u32 _xid;
800         __be32 *xp;
801
802         read_lock_bh(&sk->sk_callback_lock);
803         dprintk("RPC:       xs_udp_data_ready...\n");
804         if (!(xprt = xprt_from_sock(sk)))
805                 goto out;
806
807         if ((skb = skb_recv_datagram(sk, 0, 1, &err)) == NULL)
808                 goto out;
809
810         if (xprt->shutdown)
811                 goto dropit;
812
813         repsize = skb->len - sizeof(struct udphdr);
814         if (repsize < 4) {
815                 dprintk("RPC:       impossible RPC reply size %d!\n", repsize);
816                 goto dropit;
817         }
818
819         /* Copy the XID from the skb... */
820         xp = skb_header_pointer(skb, sizeof(struct udphdr),
821                                 sizeof(_xid), &_xid);
822         if (xp == NULL)
823                 goto dropit;
824
825         /* Look up and lock the request corresponding to the given XID */
826         spin_lock(&xprt->transport_lock);
827         rovr = xprt_lookup_rqst(xprt, *xp);
828         if (!rovr)
829                 goto out_unlock;
830         task = rovr->rq_task;
831
832         if ((copied = rovr->rq_private_buf.buflen) > repsize)
833                 copied = repsize;
834
835         /* Suck it into the iovec, verify checksum if not done by hw. */
836         if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
837                 UDPX_INC_STATS_BH(sk, UDP_MIB_INERRORS);
838                 goto out_unlock;
839         }
840
841         UDPX_INC_STATS_BH(sk, UDP_MIB_INDATAGRAMS);
842
843         /* Something worked... */
844         dst_confirm(skb_dst(skb));
845
846         xprt_adjust_cwnd(task, copied);
847         xprt_complete_rqst(task, copied);
848
849  out_unlock:
850         spin_unlock(&xprt->transport_lock);
851  dropit:
852         skb_free_datagram(sk, skb);
853  out:
854         read_unlock_bh(&sk->sk_callback_lock);
855 }
856
857 static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
858 {
859         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
860         size_t len, used;
861         char *p;
862
863         p = ((char *) &transport->tcp_fraghdr) + transport->tcp_offset;
864         len = sizeof(transport->tcp_fraghdr) - transport->tcp_offset;
865         used = xdr_skb_read_bits(desc, p, len);
866         transport->tcp_offset += used;
867         if (used != len)
868                 return;
869
870         transport->tcp_reclen = ntohl(transport->tcp_fraghdr);
871         if (transport->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
872                 transport->tcp_flags |= TCP_RCV_LAST_FRAG;
873         else
874                 transport->tcp_flags &= ~TCP_RCV_LAST_FRAG;
875         transport->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
876
877         transport->tcp_flags &= ~TCP_RCV_COPY_FRAGHDR;
878         transport->tcp_offset = 0;
879
880         /* Sanity check of the record length */
881         if (unlikely(transport->tcp_reclen < 8)) {
882                 dprintk("RPC:       invalid TCP record fragment length\n");
883                 xprt_force_disconnect(xprt);
884                 return;
885         }
886         dprintk("RPC:       reading TCP record fragment of length %d\n",
887                         transport->tcp_reclen);
888 }
889
890 static void xs_tcp_check_fraghdr(struct sock_xprt *transport)
891 {
892         if (transport->tcp_offset == transport->tcp_reclen) {
893                 transport->tcp_flags |= TCP_RCV_COPY_FRAGHDR;
894                 transport->tcp_offset = 0;
895                 if (transport->tcp_flags & TCP_RCV_LAST_FRAG) {
896                         transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
897                         transport->tcp_flags |= TCP_RCV_COPY_XID;
898                         transport->tcp_copied = 0;
899                 }
900         }
901 }
902
903 static inline void xs_tcp_read_xid(struct sock_xprt *transport, struct xdr_skb_reader *desc)
904 {
905         size_t len, used;
906         char *p;
907
908         len = sizeof(transport->tcp_xid) - transport->tcp_offset;
909         dprintk("RPC:       reading XID (%Zu bytes)\n", len);
910         p = ((char *) &transport->tcp_xid) + transport->tcp_offset;
911         used = xdr_skb_read_bits(desc, p, len);
912         transport->tcp_offset += used;
913         if (used != len)
914                 return;
915         transport->tcp_flags &= ~TCP_RCV_COPY_XID;
916         transport->tcp_flags |= TCP_RCV_READ_CALLDIR;
917         transport->tcp_copied = 4;
918         dprintk("RPC:       reading %s XID %08x\n",
919                         (transport->tcp_flags & TCP_RPC_REPLY) ? "reply for"
920                                                               : "request with",
921                         ntohl(transport->tcp_xid));
922         xs_tcp_check_fraghdr(transport);
923 }
924
925 static inline void xs_tcp_read_calldir(struct sock_xprt *transport,
926                                        struct xdr_skb_reader *desc)
927 {
928         size_t len, used;
929         u32 offset;
930         char *p;
931
932         /*
933          * We want transport->tcp_offset to be 8 at the end of this routine
934          * (4 bytes for the xid and 4 bytes for the call/reply flag).
935          * When this function is called for the first time,
936          * transport->tcp_offset is 4 (after having already read the xid).
937          */
938         offset = transport->tcp_offset - sizeof(transport->tcp_xid);
939         len = sizeof(transport->tcp_calldir) - offset;
940         dprintk("RPC:       reading CALL/REPLY flag (%Zu bytes)\n", len);
941         p = ((char *) &transport->tcp_calldir) + offset;
942         used = xdr_skb_read_bits(desc, p, len);
943         transport->tcp_offset += used;
944         if (used != len)
945                 return;
946         transport->tcp_flags &= ~TCP_RCV_READ_CALLDIR;
947         /*
948          * We don't yet have the XDR buffer, so we will write the calldir
949          * out after we get the buffer from the 'struct rpc_rqst'
950          */
951         switch (ntohl(transport->tcp_calldir)) {
952         case RPC_REPLY:
953                 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
954                 transport->tcp_flags |= TCP_RCV_COPY_DATA;
955                 transport->tcp_flags |= TCP_RPC_REPLY;
956                 break;
957         case RPC_CALL:
958                 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
959                 transport->tcp_flags |= TCP_RCV_COPY_DATA;
960                 transport->tcp_flags &= ~TCP_RPC_REPLY;
961                 break;
962         default:
963                 dprintk("RPC:       invalid request message type\n");
964                 xprt_force_disconnect(&transport->xprt);
965         }
966         xs_tcp_check_fraghdr(transport);
967 }
968
969 static inline void xs_tcp_read_common(struct rpc_xprt *xprt,
970                                      struct xdr_skb_reader *desc,
971                                      struct rpc_rqst *req)
972 {
973         struct sock_xprt *transport =
974                                 container_of(xprt, struct sock_xprt, xprt);
975         struct xdr_buf *rcvbuf;
976         size_t len;
977         ssize_t r;
978
979         rcvbuf = &req->rq_private_buf;
980
981         if (transport->tcp_flags & TCP_RCV_COPY_CALLDIR) {
982                 /*
983                  * Save the RPC direction in the XDR buffer
984                  */
985                 memcpy(rcvbuf->head[0].iov_base + transport->tcp_copied,
986                         &transport->tcp_calldir,
987                         sizeof(transport->tcp_calldir));
988                 transport->tcp_copied += sizeof(transport->tcp_calldir);
989                 transport->tcp_flags &= ~TCP_RCV_COPY_CALLDIR;
990         }
991
992         len = desc->count;
993         if (len > transport->tcp_reclen - transport->tcp_offset) {
994                 struct xdr_skb_reader my_desc;
995
996                 len = transport->tcp_reclen - transport->tcp_offset;
997                 memcpy(&my_desc, desc, sizeof(my_desc));
998                 my_desc.count = len;
999                 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
1000                                           &my_desc, xdr_skb_read_bits);
1001                 desc->count -= r;
1002                 desc->offset += r;
1003         } else
1004                 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
1005                                           desc, xdr_skb_read_bits);
1006
1007         if (r > 0) {
1008                 transport->tcp_copied += r;
1009                 transport->tcp_offset += r;
1010         }
1011         if (r != len) {
1012                 /* Error when copying to the receive buffer,
1013                  * usually because we weren't able to allocate
1014                  * additional buffer pages. All we can do now
1015                  * is turn off TCP_RCV_COPY_DATA, so the request
1016                  * will not receive any additional updates,
1017                  * and time out.
1018                  * Any remaining data from this record will
1019                  * be discarded.
1020                  */
1021                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1022                 dprintk("RPC:       XID %08x truncated request\n",
1023                                 ntohl(transport->tcp_xid));
1024                 dprintk("RPC:       xprt = %p, tcp_copied = %lu, "
1025                                 "tcp_offset = %u, tcp_reclen = %u\n",
1026                                 xprt, transport->tcp_copied,
1027                                 transport->tcp_offset, transport->tcp_reclen);
1028                 return;
1029         }
1030
1031         dprintk("RPC:       XID %08x read %Zd bytes\n",
1032                         ntohl(transport->tcp_xid), r);
1033         dprintk("RPC:       xprt = %p, tcp_copied = %lu, tcp_offset = %u, "
1034                         "tcp_reclen = %u\n", xprt, transport->tcp_copied,
1035                         transport->tcp_offset, transport->tcp_reclen);
1036
1037         if (transport->tcp_copied == req->rq_private_buf.buflen)
1038                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1039         else if (transport->tcp_offset == transport->tcp_reclen) {
1040                 if (transport->tcp_flags & TCP_RCV_LAST_FRAG)
1041                         transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1042         }
1043 }
1044
1045 /*
1046  * Finds the request corresponding to the RPC xid and invokes the common
1047  * tcp read code to read the data.
1048  */
1049 static inline int xs_tcp_read_reply(struct rpc_xprt *xprt,
1050                                     struct xdr_skb_reader *desc)
1051 {
1052         struct sock_xprt *transport =
1053                                 container_of(xprt, struct sock_xprt, xprt);
1054         struct rpc_rqst *req;
1055
1056         dprintk("RPC:       read reply XID %08x\n", ntohl(transport->tcp_xid));
1057
1058         /* Find and lock the request corresponding to this xid */
1059         spin_lock(&xprt->transport_lock);
1060         req = xprt_lookup_rqst(xprt, transport->tcp_xid);
1061         if (!req) {
1062                 dprintk("RPC:       XID %08x request not found!\n",
1063                                 ntohl(transport->tcp_xid));
1064                 spin_unlock(&xprt->transport_lock);
1065                 return -1;
1066         }
1067
1068         xs_tcp_read_common(xprt, desc, req);
1069
1070         if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1071                 xprt_complete_rqst(req->rq_task, transport->tcp_copied);
1072
1073         spin_unlock(&xprt->transport_lock);
1074         return 0;
1075 }
1076
1077 #if defined(CONFIG_NFS_V4_1)
1078 /*
1079  * Obtains an rpc_rqst previously allocated and invokes the common
1080  * tcp read code to read the data.  The result is placed in the callback
1081  * queue.
1082  * If we're unable to obtain the rpc_rqst we schedule the closing of the
1083  * connection and return -1.
1084  */
1085 static inline int xs_tcp_read_callback(struct rpc_xprt *xprt,
1086                                        struct xdr_skb_reader *desc)
1087 {
1088         struct sock_xprt *transport =
1089                                 container_of(xprt, struct sock_xprt, xprt);
1090         struct rpc_rqst *req;
1091
1092         req = xprt_alloc_bc_request(xprt);
1093         if (req == NULL) {
1094                 printk(KERN_WARNING "Callback slot table overflowed\n");
1095                 xprt_force_disconnect(xprt);
1096                 return -1;
1097         }
1098
1099         req->rq_xid = transport->tcp_xid;
1100         dprintk("RPC:       read callback  XID %08x\n", ntohl(req->rq_xid));
1101         xs_tcp_read_common(xprt, desc, req);
1102
1103         if (!(transport->tcp_flags & TCP_RCV_COPY_DATA)) {
1104                 struct svc_serv *bc_serv = xprt->bc_serv;
1105
1106                 /*
1107                  * Add callback request to callback list.  The callback
1108                  * service sleeps on the sv_cb_waitq waiting for new
1109                  * requests.  Wake it up after adding enqueing the
1110                  * request.
1111                  */
1112                 dprintk("RPC:       add callback request to list\n");
1113                 spin_lock(&bc_serv->sv_cb_lock);
1114                 list_add(&req->rq_bc_list, &bc_serv->sv_cb_list);
1115                 spin_unlock(&bc_serv->sv_cb_lock);
1116                 wake_up(&bc_serv->sv_cb_waitq);
1117         }
1118
1119         req->rq_private_buf.len = transport->tcp_copied;
1120
1121         return 0;
1122 }
1123
1124 static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1125                                         struct xdr_skb_reader *desc)
1126 {
1127         struct sock_xprt *transport =
1128                                 container_of(xprt, struct sock_xprt, xprt);
1129
1130         return (transport->tcp_flags & TCP_RPC_REPLY) ?
1131                 xs_tcp_read_reply(xprt, desc) :
1132                 xs_tcp_read_callback(xprt, desc);
1133 }
1134 #else
1135 static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1136                                         struct xdr_skb_reader *desc)
1137 {
1138         return xs_tcp_read_reply(xprt, desc);
1139 }
1140 #endif /* CONFIG_NFS_V4_1 */
1141
1142 /*
1143  * Read data off the transport.  This can be either an RPC_CALL or an
1144  * RPC_REPLY.  Relay the processing to helper functions.
1145  */
1146 static void xs_tcp_read_data(struct rpc_xprt *xprt,
1147                                     struct xdr_skb_reader *desc)
1148 {
1149         struct sock_xprt *transport =
1150                                 container_of(xprt, struct sock_xprt, xprt);
1151
1152         if (_xs_tcp_read_data(xprt, desc) == 0)
1153                 xs_tcp_check_fraghdr(transport);
1154         else {
1155                 /*
1156                  * The transport_lock protects the request handling.
1157                  * There's no need to hold it to update the tcp_flags.
1158                  */
1159                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1160         }
1161 }
1162
1163 static inline void xs_tcp_read_discard(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1164 {
1165         size_t len;
1166
1167         len = transport->tcp_reclen - transport->tcp_offset;
1168         if (len > desc->count)
1169                 len = desc->count;
1170         desc->count -= len;
1171         desc->offset += len;
1172         transport->tcp_offset += len;
1173         dprintk("RPC:       discarded %Zu bytes\n", len);
1174         xs_tcp_check_fraghdr(transport);
1175 }
1176
1177 static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
1178 {
1179         struct rpc_xprt *xprt = rd_desc->arg.data;
1180         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1181         struct xdr_skb_reader desc = {
1182                 .skb    = skb,
1183                 .offset = offset,
1184                 .count  = len,
1185         };
1186
1187         dprintk("RPC:       xs_tcp_data_recv started\n");
1188         do {
1189                 /* Read in a new fragment marker if necessary */
1190                 /* Can we ever really expect to get completely empty fragments? */
1191                 if (transport->tcp_flags & TCP_RCV_COPY_FRAGHDR) {
1192                         xs_tcp_read_fraghdr(xprt, &desc);
1193                         continue;
1194                 }
1195                 /* Read in the xid if necessary */
1196                 if (transport->tcp_flags & TCP_RCV_COPY_XID) {
1197                         xs_tcp_read_xid(transport, &desc);
1198                         continue;
1199                 }
1200                 /* Read in the call/reply flag */
1201                 if (transport->tcp_flags & TCP_RCV_READ_CALLDIR) {
1202                         xs_tcp_read_calldir(transport, &desc);
1203                         continue;
1204                 }
1205                 /* Read in the request data */
1206                 if (transport->tcp_flags & TCP_RCV_COPY_DATA) {
1207                         xs_tcp_read_data(xprt, &desc);
1208                         continue;
1209                 }
1210                 /* Skip over any trailing bytes on short reads */
1211                 xs_tcp_read_discard(transport, &desc);
1212         } while (desc.count);
1213         dprintk("RPC:       xs_tcp_data_recv done\n");
1214         return len - desc.count;
1215 }
1216
1217 /**
1218  * xs_tcp_data_ready - "data ready" callback for TCP sockets
1219  * @sk: socket with data to read
1220  * @bytes: how much data to read
1221  *
1222  */
1223 static void xs_tcp_data_ready(struct sock *sk, int bytes)
1224 {
1225         struct rpc_xprt *xprt;
1226         read_descriptor_t rd_desc;
1227         int read;
1228
1229         dprintk("RPC:       xs_tcp_data_ready...\n");
1230
1231         read_lock_bh(&sk->sk_callback_lock);
1232         if (!(xprt = xprt_from_sock(sk)))
1233                 goto out;
1234         if (xprt->shutdown)
1235                 goto out;
1236
1237         /* Any data means we had a useful conversation, so
1238          * the we don't need to delay the next reconnect
1239          */
1240         if (xprt->reestablish_timeout)
1241                 xprt->reestablish_timeout = 0;
1242
1243         /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
1244         rd_desc.arg.data = xprt;
1245         do {
1246                 rd_desc.count = 65536;
1247                 read = tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
1248         } while (read > 0);
1249 out:
1250         read_unlock_bh(&sk->sk_callback_lock);
1251 }
1252
1253 /*
1254  * Do the equivalent of linger/linger2 handling for dealing with
1255  * broken servers that don't close the socket in a timely
1256  * fashion
1257  */
1258 static void xs_tcp_schedule_linger_timeout(struct rpc_xprt *xprt,
1259                 unsigned long timeout)
1260 {
1261         struct sock_xprt *transport;
1262
1263         if (xprt_test_and_set_connecting(xprt))
1264                 return;
1265         set_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1266         transport = container_of(xprt, struct sock_xprt, xprt);
1267         queue_delayed_work(rpciod_workqueue, &transport->connect_worker,
1268                            timeout);
1269 }
1270
1271 static void xs_tcp_cancel_linger_timeout(struct rpc_xprt *xprt)
1272 {
1273         struct sock_xprt *transport;
1274
1275         transport = container_of(xprt, struct sock_xprt, xprt);
1276
1277         if (!test_bit(XPRT_CONNECTION_ABORT, &xprt->state) ||
1278             !cancel_delayed_work(&transport->connect_worker))
1279                 return;
1280         clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1281         xprt_clear_connecting(xprt);
1282 }
1283
1284 static void xs_sock_mark_closed(struct rpc_xprt *xprt)
1285 {
1286         smp_mb__before_clear_bit();
1287         clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1288         clear_bit(XPRT_CLOSING, &xprt->state);
1289         smp_mb__after_clear_bit();
1290         /* Mark transport as closed and wake up all pending tasks */
1291         xprt_disconnect_done(xprt);
1292 }
1293
1294 /**
1295  * xs_tcp_state_change - callback to handle TCP socket state changes
1296  * @sk: socket whose state has changed
1297  *
1298  */
1299 static void xs_tcp_state_change(struct sock *sk)
1300 {
1301         struct rpc_xprt *xprt;
1302
1303         read_lock_bh(&sk->sk_callback_lock);
1304         if (!(xprt = xprt_from_sock(sk)))
1305                 goto out;
1306         dprintk("RPC:       xs_tcp_state_change client %p...\n", xprt);
1307         dprintk("RPC:       state %x conn %d dead %d zapped %d sk_shutdown %d\n",
1308                         sk->sk_state, xprt_connected(xprt),
1309                         sock_flag(sk, SOCK_DEAD),
1310                         sock_flag(sk, SOCK_ZAPPED),
1311                         sk->sk_shutdown);
1312
1313         switch (sk->sk_state) {
1314         case TCP_ESTABLISHED:
1315                 spin_lock(&xprt->transport_lock);
1316                 if (!xprt_test_and_set_connected(xprt)) {
1317                         struct sock_xprt *transport = container_of(xprt,
1318                                         struct sock_xprt, xprt);
1319
1320                         /* Reset TCP record info */
1321                         transport->tcp_offset = 0;
1322                         transport->tcp_reclen = 0;
1323                         transport->tcp_copied = 0;
1324                         transport->tcp_flags =
1325                                 TCP_RCV_COPY_FRAGHDR | TCP_RCV_COPY_XID;
1326
1327                         xprt_wake_pending_tasks(xprt, -EAGAIN);
1328                 }
1329                 spin_unlock(&xprt->transport_lock);
1330                 break;
1331         case TCP_FIN_WAIT1:
1332                 /* The client initiated a shutdown of the socket */
1333                 xprt->connect_cookie++;
1334                 xprt->reestablish_timeout = 0;
1335                 set_bit(XPRT_CLOSING, &xprt->state);
1336                 smp_mb__before_clear_bit();
1337                 clear_bit(XPRT_CONNECTED, &xprt->state);
1338                 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1339                 smp_mb__after_clear_bit();
1340                 xs_tcp_schedule_linger_timeout(xprt, xs_tcp_fin_timeout);
1341                 break;
1342         case TCP_CLOSE_WAIT:
1343                 /* The server initiated a shutdown of the socket */
1344                 xprt_force_disconnect(xprt);
1345         case TCP_SYN_SENT:
1346                 xprt->connect_cookie++;
1347         case TCP_CLOSING:
1348                 /*
1349                  * If the server closed down the connection, make sure that
1350                  * we back off before reconnecting
1351                  */
1352                 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1353                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1354                 break;
1355         case TCP_LAST_ACK:
1356                 set_bit(XPRT_CLOSING, &xprt->state);
1357                 xs_tcp_schedule_linger_timeout(xprt, xs_tcp_fin_timeout);
1358                 smp_mb__before_clear_bit();
1359                 clear_bit(XPRT_CONNECTED, &xprt->state);
1360                 smp_mb__after_clear_bit();
1361                 break;
1362         case TCP_CLOSE:
1363                 xs_tcp_cancel_linger_timeout(xprt);
1364                 xs_sock_mark_closed(xprt);
1365         }
1366  out:
1367         read_unlock_bh(&sk->sk_callback_lock);
1368 }
1369
1370 /**
1371  * xs_error_report - callback mainly for catching socket errors
1372  * @sk: socket
1373  */
1374 static void xs_error_report(struct sock *sk)
1375 {
1376         struct rpc_xprt *xprt;
1377
1378         read_lock_bh(&sk->sk_callback_lock);
1379         if (!(xprt = xprt_from_sock(sk)))
1380                 goto out;
1381         dprintk("RPC:       %s client %p...\n"
1382                         "RPC:       error %d\n",
1383                         __func__, xprt, sk->sk_err);
1384         xprt_wake_pending_tasks(xprt, -EAGAIN);
1385 out:
1386         read_unlock_bh(&sk->sk_callback_lock);
1387 }
1388
1389 static void xs_write_space(struct sock *sk)
1390 {
1391         struct socket *sock;
1392         struct rpc_xprt *xprt;
1393
1394         if (unlikely(!(sock = sk->sk_socket)))
1395                 return;
1396         clear_bit(SOCK_NOSPACE, &sock->flags);
1397
1398         if (unlikely(!(xprt = xprt_from_sock(sk))))
1399                 return;
1400         if (test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags) == 0)
1401                 return;
1402
1403         xprt_write_space(xprt);
1404 }
1405
1406 /**
1407  * xs_udp_write_space - callback invoked when socket buffer space
1408  *                             becomes available
1409  * @sk: socket whose state has changed
1410  *
1411  * Called when more output buffer space is available for this socket.
1412  * We try not to wake our writers until they can make "significant"
1413  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1414  * with a bunch of small requests.
1415  */
1416 static void xs_udp_write_space(struct sock *sk)
1417 {
1418         read_lock_bh(&sk->sk_callback_lock);
1419
1420         /* from net/core/sock.c:sock_def_write_space */
1421         if (sock_writeable(sk))
1422                 xs_write_space(sk);
1423
1424         read_unlock_bh(&sk->sk_callback_lock);
1425 }
1426
1427 /**
1428  * xs_tcp_write_space - callback invoked when socket buffer space
1429  *                             becomes available
1430  * @sk: socket whose state has changed
1431  *
1432  * Called when more output buffer space is available for this socket.
1433  * We try not to wake our writers until they can make "significant"
1434  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1435  * with a bunch of small requests.
1436  */
1437 static void xs_tcp_write_space(struct sock *sk)
1438 {
1439         read_lock_bh(&sk->sk_callback_lock);
1440
1441         /* from net/core/stream.c:sk_stream_write_space */
1442         if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk))
1443                 xs_write_space(sk);
1444
1445         read_unlock_bh(&sk->sk_callback_lock);
1446 }
1447
1448 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
1449 {
1450         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1451         struct sock *sk = transport->inet;
1452
1453         if (transport->rcvsize) {
1454                 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1455                 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
1456         }
1457         if (transport->sndsize) {
1458                 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1459                 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1460                 sk->sk_write_space(sk);
1461         }
1462 }
1463
1464 /**
1465  * xs_udp_set_buffer_size - set send and receive limits
1466  * @xprt: generic transport
1467  * @sndsize: requested size of send buffer, in bytes
1468  * @rcvsize: requested size of receive buffer, in bytes
1469  *
1470  * Set socket send and receive buffer size limits.
1471  */
1472 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1473 {
1474         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1475
1476         transport->sndsize = 0;
1477         if (sndsize)
1478                 transport->sndsize = sndsize + 1024;
1479         transport->rcvsize = 0;
1480         if (rcvsize)
1481                 transport->rcvsize = rcvsize + 1024;
1482
1483         xs_udp_do_set_buffer_size(xprt);
1484 }
1485
1486 /**
1487  * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1488  * @task: task that timed out
1489  *
1490  * Adjust the congestion window after a retransmit timeout has occurred.
1491  */
1492 static void xs_udp_timer(struct rpc_task *task)
1493 {
1494         xprt_adjust_cwnd(task, -ETIMEDOUT);
1495 }
1496
1497 static unsigned short xs_get_random_port(void)
1498 {
1499         unsigned short range = xprt_max_resvport - xprt_min_resvport;
1500         unsigned short rand = (unsigned short) net_random() % range;
1501         return rand + xprt_min_resvport;
1502 }
1503
1504 /**
1505  * xs_set_port - reset the port number in the remote endpoint address
1506  * @xprt: generic transport
1507  * @port: new port number
1508  *
1509  */
1510 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1511 {
1512         dprintk("RPC:       setting port for xprt %p to %u\n", xprt, port);
1513
1514         rpc_set_port(xs_addr(xprt), port);
1515         xs_update_peer_port(xprt);
1516 }
1517
1518 static unsigned short xs_get_srcport(struct sock_xprt *transport)
1519 {
1520         unsigned short port = transport->srcport;
1521
1522         if (port == 0 && transport->xprt.resvport)
1523                 port = xs_get_random_port();
1524         return port;
1525 }
1526
1527 static unsigned short xs_next_srcport(struct sock_xprt *transport, unsigned short port)
1528 {
1529         if (transport->srcport != 0)
1530                 transport->srcport = 0;
1531         if (!transport->xprt.resvport)
1532                 return 0;
1533         if (port <= xprt_min_resvport || port > xprt_max_resvport)
1534                 return xprt_max_resvport;
1535         return --port;
1536 }
1537 static int xs_bind(struct sock_xprt *transport, struct socket *sock)
1538 {
1539         struct sockaddr_storage myaddr;
1540         int err, nloop = 0;
1541         unsigned short port = xs_get_srcport(transport);
1542         unsigned short last;
1543
1544         memcpy(&myaddr, &transport->srcaddr, transport->xprt.addrlen);
1545         do {
1546                 rpc_set_port((struct sockaddr *)&myaddr, port);
1547                 err = kernel_bind(sock, (struct sockaddr *)&myaddr,
1548                                 transport->xprt.addrlen);
1549                 if (port == 0)
1550                         break;
1551                 if (err == 0) {
1552                         transport->srcport = port;
1553                         break;
1554                 }
1555                 last = port;
1556                 port = xs_next_srcport(transport, port);
1557                 if (port > last)
1558                         nloop++;
1559         } while (err == -EADDRINUSE && nloop != 2);
1560
1561         if (myaddr.ss_family == AF_INET)
1562                 dprintk("RPC:       %s %pI4:%u: %s (%d)\n", __func__,
1563                                 &((struct sockaddr_in *)&myaddr)->sin_addr,
1564                                 port, err ? "failed" : "ok", err);
1565         else
1566                 dprintk("RPC:       %s %pI6:%u: %s (%d)\n", __func__,
1567                                 &((struct sockaddr_in6 *)&myaddr)->sin6_addr,
1568                                 port, err ? "failed" : "ok", err);
1569         return err;
1570 }
1571
1572
1573 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1574 static struct lock_class_key xs_key[2];
1575 static struct lock_class_key xs_slock_key[2];
1576
1577 static inline void xs_reclassify_socket4(struct socket *sock)
1578 {
1579         struct sock *sk = sock->sk;
1580
1581         BUG_ON(sock_owned_by_user(sk));
1582         sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1583                 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1584 }
1585
1586 static inline void xs_reclassify_socket6(struct socket *sock)
1587 {
1588         struct sock *sk = sock->sk;
1589
1590         BUG_ON(sock_owned_by_user(sk));
1591         sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1592                 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
1593 }
1594
1595 static inline void xs_reclassify_socket(int family, struct socket *sock)
1596 {
1597         switch (family) {
1598         case AF_INET:
1599                 xs_reclassify_socket4(sock);
1600                 break;
1601         case AF_INET6:
1602                 xs_reclassify_socket6(sock);
1603                 break;
1604         }
1605 }
1606 #else
1607 static inline void xs_reclassify_socket4(struct socket *sock)
1608 {
1609 }
1610
1611 static inline void xs_reclassify_socket6(struct socket *sock)
1612 {
1613 }
1614
1615 static inline void xs_reclassify_socket(int family, struct socket *sock)
1616 {
1617 }
1618 #endif
1619
1620 static struct socket *xs_create_sock(struct rpc_xprt *xprt,
1621                 struct sock_xprt *transport, int family, int type, int protocol)
1622 {
1623         struct socket *sock;
1624         int err;
1625
1626         err = __sock_create(xprt->xprt_net, family, type, protocol, &sock, 1);
1627         if (err < 0) {
1628                 dprintk("RPC:       can't create %d transport socket (%d).\n",
1629                                 protocol, -err);
1630                 goto out;
1631         }
1632         xs_reclassify_socket(family, sock);
1633
1634         if (xs_bind(transport, sock)) {
1635                 sock_release(sock);
1636                 goto out;
1637         }
1638
1639         return sock;
1640 out:
1641         return ERR_PTR(err);
1642 }
1643
1644 static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
1645 {
1646         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1647
1648         if (!transport->inet) {
1649                 struct sock *sk = sock->sk;
1650
1651                 write_lock_bh(&sk->sk_callback_lock);
1652
1653                 xs_save_old_callbacks(transport, sk);
1654
1655                 sk->sk_user_data = xprt;
1656                 sk->sk_data_ready = xs_udp_data_ready;
1657                 sk->sk_write_space = xs_udp_write_space;
1658                 sk->sk_error_report = xs_error_report;
1659                 sk->sk_no_check = UDP_CSUM_NORCV;
1660                 sk->sk_allocation = GFP_ATOMIC;
1661
1662                 xprt_set_connected(xprt);
1663
1664                 /* Reset to new socket */
1665                 transport->sock = sock;
1666                 transport->inet = sk;
1667
1668                 write_unlock_bh(&sk->sk_callback_lock);
1669         }
1670         xs_udp_do_set_buffer_size(xprt);
1671 }
1672
1673 static void xs_udp_setup_socket(struct work_struct *work)
1674 {
1675         struct sock_xprt *transport =
1676                 container_of(work, struct sock_xprt, connect_worker.work);
1677         struct rpc_xprt *xprt = &transport->xprt;
1678         struct socket *sock = transport->sock;
1679         int status = -EIO;
1680
1681         if (xprt->shutdown)
1682                 goto out;
1683
1684         /* Start by resetting any existing state */
1685         xs_reset_transport(transport);
1686         sock = xs_create_sock(xprt, transport,
1687                         xs_addr(xprt)->sa_family, SOCK_DGRAM, IPPROTO_UDP);
1688         if (IS_ERR(sock))
1689                 goto out;
1690
1691         dprintk("RPC:       worker connecting xprt %p via %s to "
1692                                 "%s (port %s)\n", xprt,
1693                         xprt->address_strings[RPC_DISPLAY_PROTO],
1694                         xprt->address_strings[RPC_DISPLAY_ADDR],
1695                         xprt->address_strings[RPC_DISPLAY_PORT]);
1696
1697         xs_udp_finish_connecting(xprt, sock);
1698         status = 0;
1699 out:
1700         xprt_clear_connecting(xprt);
1701         xprt_wake_pending_tasks(xprt, status);
1702 }
1703
1704 /*
1705  * We need to preserve the port number so the reply cache on the server can
1706  * find our cached RPC replies when we get around to reconnecting.
1707  */
1708 static void xs_abort_connection(struct sock_xprt *transport)
1709 {
1710         int result;
1711         struct sockaddr any;
1712
1713         dprintk("RPC:       disconnecting xprt %p to reuse port\n", transport);
1714
1715         /*
1716          * Disconnect the transport socket by doing a connect operation
1717          * with AF_UNSPEC.  This should return immediately...
1718          */
1719         memset(&any, 0, sizeof(any));
1720         any.sa_family = AF_UNSPEC;
1721         result = kernel_connect(transport->sock, &any, sizeof(any), 0);
1722         if (!result)
1723                 xs_sock_mark_closed(&transport->xprt);
1724         else
1725                 dprintk("RPC:       AF_UNSPEC connect return code %d\n",
1726                                 result);
1727 }
1728
1729 static void xs_tcp_reuse_connection(struct sock_xprt *transport)
1730 {
1731         unsigned int state = transport->inet->sk_state;
1732
1733         if (state == TCP_CLOSE && transport->sock->state == SS_UNCONNECTED) {
1734                 /* we don't need to abort the connection if the socket
1735                  * hasn't undergone a shutdown
1736                  */
1737                 if (transport->inet->sk_shutdown == 0)
1738                         return;
1739                 dprintk("RPC:       %s: TCP_CLOSEd and sk_shutdown set to %d\n",
1740                                 __func__, transport->inet->sk_shutdown);
1741         }
1742         if ((1 << state) & (TCPF_ESTABLISHED|TCPF_SYN_SENT)) {
1743                 /* we don't need to abort the connection if the socket
1744                  * hasn't undergone a shutdown
1745                  */
1746                 if (transport->inet->sk_shutdown == 0)
1747                         return;
1748                 dprintk("RPC:       %s: ESTABLISHED/SYN_SENT "
1749                                 "sk_shutdown set to %d\n",
1750                                 __func__, transport->inet->sk_shutdown);
1751         }
1752         xs_abort_connection(transport);
1753 }
1754
1755 static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
1756 {
1757         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1758
1759         if (!transport->inet) {
1760                 struct sock *sk = sock->sk;
1761
1762                 write_lock_bh(&sk->sk_callback_lock);
1763
1764                 xs_save_old_callbacks(transport, sk);
1765
1766                 sk->sk_user_data = xprt;
1767                 sk->sk_data_ready = xs_tcp_data_ready;
1768                 sk->sk_state_change = xs_tcp_state_change;
1769                 sk->sk_write_space = xs_tcp_write_space;
1770                 sk->sk_error_report = xs_error_report;
1771                 sk->sk_allocation = GFP_ATOMIC;
1772
1773                 /* socket options */
1774                 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
1775                 sock_reset_flag(sk, SOCK_LINGER);
1776                 tcp_sk(sk)->linger2 = 0;
1777                 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
1778
1779                 xprt_clear_connected(xprt);
1780
1781                 /* Reset to new socket */
1782                 transport->sock = sock;
1783                 transport->inet = sk;
1784
1785                 write_unlock_bh(&sk->sk_callback_lock);
1786         }
1787
1788         if (!xprt_bound(xprt))
1789                 return -ENOTCONN;
1790
1791         /* Tell the socket layer to start connecting... */
1792         xprt->stat.connect_count++;
1793         xprt->stat.connect_start = jiffies;
1794         return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
1795 }
1796
1797 /**
1798  * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint
1799  * @xprt: RPC transport to connect
1800  * @transport: socket transport to connect
1801  * @create_sock: function to create a socket of the correct type
1802  *
1803  * Invoked by a work queue tasklet.
1804  */
1805 static void xs_tcp_setup_socket(struct work_struct *work)
1806 {
1807         struct sock_xprt *transport =
1808                 container_of(work, struct sock_xprt, connect_worker.work);
1809         struct socket *sock = transport->sock;
1810         struct rpc_xprt *xprt = &transport->xprt;
1811         int status = -EIO;
1812
1813         if (xprt->shutdown)
1814                 goto out;
1815
1816         if (!sock) {
1817                 clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1818                 sock = xs_create_sock(xprt, transport,
1819                                 xs_addr(xprt)->sa_family, SOCK_STREAM, IPPROTO_TCP);
1820                 if (IS_ERR(sock)) {
1821                         status = PTR_ERR(sock);
1822                         goto out;
1823                 }
1824         } else {
1825                 int abort_and_exit;
1826
1827                 abort_and_exit = test_and_clear_bit(XPRT_CONNECTION_ABORT,
1828                                 &xprt->state);
1829                 /* "close" the socket, preserving the local port */
1830                 xs_tcp_reuse_connection(transport);
1831
1832                 if (abort_and_exit)
1833                         goto out_eagain;
1834         }
1835
1836         dprintk("RPC:       worker connecting xprt %p via %s to "
1837                                 "%s (port %s)\n", xprt,
1838                         xprt->address_strings[RPC_DISPLAY_PROTO],
1839                         xprt->address_strings[RPC_DISPLAY_ADDR],
1840                         xprt->address_strings[RPC_DISPLAY_PORT]);
1841
1842         status = xs_tcp_finish_connecting(xprt, sock);
1843         dprintk("RPC:       %p connect status %d connected %d sock state %d\n",
1844                         xprt, -status, xprt_connected(xprt),
1845                         sock->sk->sk_state);
1846         switch (status) {
1847         default:
1848                 printk("%s: connect returned unhandled error %d\n",
1849                         __func__, status);
1850         case -EADDRNOTAVAIL:
1851                 /* We're probably in TIME_WAIT. Get rid of existing socket,
1852                  * and retry
1853                  */
1854                 set_bit(XPRT_CONNECTION_CLOSE, &xprt->state);
1855                 xprt_force_disconnect(xprt);
1856                 break;
1857         case -ECONNREFUSED:
1858         case -ECONNRESET:
1859         case -ENETUNREACH:
1860                 /* retry with existing socket, after a delay */
1861         case 0:
1862         case -EINPROGRESS:
1863         case -EALREADY:
1864                 xprt_clear_connecting(xprt);
1865                 return;
1866         case -EINVAL:
1867                 /* Happens, for instance, if the user specified a link
1868                  * local IPv6 address without a scope-id.
1869                  */
1870                 goto out;
1871         }
1872 out_eagain:
1873         status = -EAGAIN;
1874 out:
1875         xprt_clear_connecting(xprt);
1876         xprt_wake_pending_tasks(xprt, status);
1877 }
1878
1879 /**
1880  * xs_connect - connect a socket to a remote endpoint
1881  * @task: address of RPC task that manages state of connect request
1882  *
1883  * TCP: If the remote end dropped the connection, delay reconnecting.
1884  *
1885  * UDP socket connects are synchronous, but we use a work queue anyway
1886  * to guarantee that even unprivileged user processes can set up a
1887  * socket on a privileged port.
1888  *
1889  * If a UDP socket connect fails, the delay behavior here prevents
1890  * retry floods (hard mounts).
1891  */
1892 static void xs_connect(struct rpc_task *task)
1893 {
1894         struct rpc_xprt *xprt = task->tk_xprt;
1895         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1896
1897         if (transport->sock != NULL && !RPC_IS_SOFTCONN(task)) {
1898                 dprintk("RPC:       xs_connect delayed xprt %p for %lu "
1899                                 "seconds\n",
1900                                 xprt, xprt->reestablish_timeout / HZ);
1901                 queue_delayed_work(rpciod_workqueue,
1902                                    &transport->connect_worker,
1903                                    xprt->reestablish_timeout);
1904                 xprt->reestablish_timeout <<= 1;
1905                 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1906                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1907                 if (xprt->reestablish_timeout > XS_TCP_MAX_REEST_TO)
1908                         xprt->reestablish_timeout = XS_TCP_MAX_REEST_TO;
1909         } else {
1910                 dprintk("RPC:       xs_connect scheduled xprt %p\n", xprt);
1911                 queue_delayed_work(rpciod_workqueue,
1912                                    &transport->connect_worker, 0);
1913         }
1914 }
1915
1916 /**
1917  * xs_udp_print_stats - display UDP socket-specifc stats
1918  * @xprt: rpc_xprt struct containing statistics
1919  * @seq: output file
1920  *
1921  */
1922 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
1923 {
1924         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1925
1926         seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %Lu %Lu\n",
1927                         transport->srcport,
1928                         xprt->stat.bind_count,
1929                         xprt->stat.sends,
1930                         xprt->stat.recvs,
1931                         xprt->stat.bad_xids,
1932                         xprt->stat.req_u,
1933                         xprt->stat.bklog_u);
1934 }
1935
1936 /**
1937  * xs_tcp_print_stats - display TCP socket-specifc stats
1938  * @xprt: rpc_xprt struct containing statistics
1939  * @seq: output file
1940  *
1941  */
1942 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
1943 {
1944         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1945         long idle_time = 0;
1946
1947         if (xprt_connected(xprt))
1948                 idle_time = (long)(jiffies - xprt->last_used) / HZ;
1949
1950         seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu %Lu %Lu\n",
1951                         transport->srcport,
1952                         xprt->stat.bind_count,
1953                         xprt->stat.connect_count,
1954                         xprt->stat.connect_time,
1955                         idle_time,
1956                         xprt->stat.sends,
1957                         xprt->stat.recvs,
1958                         xprt->stat.bad_xids,
1959                         xprt->stat.req_u,
1960                         xprt->stat.bklog_u);
1961 }
1962
1963 /*
1964  * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason
1965  * we allocate pages instead doing a kmalloc like rpc_malloc is because we want
1966  * to use the server side send routines.
1967  */
1968 static void *bc_malloc(struct rpc_task *task, size_t size)
1969 {
1970         struct page *page;
1971         struct rpc_buffer *buf;
1972
1973         BUG_ON(size > PAGE_SIZE - sizeof(struct rpc_buffer));
1974         page = alloc_page(GFP_KERNEL);
1975
1976         if (!page)
1977                 return NULL;
1978
1979         buf = page_address(page);
1980         buf->len = PAGE_SIZE;
1981
1982         return buf->data;
1983 }
1984
1985 /*
1986  * Free the space allocated in the bc_alloc routine
1987  */
1988 static void bc_free(void *buffer)
1989 {
1990         struct rpc_buffer *buf;
1991
1992         if (!buffer)
1993                 return;
1994
1995         buf = container_of(buffer, struct rpc_buffer, data);
1996         free_page((unsigned long)buf);
1997 }
1998
1999 /*
2000  * Use the svc_sock to send the callback. Must be called with svsk->sk_mutex
2001  * held. Borrows heavily from svc_tcp_sendto and xs_tcp_send_request.
2002  */
2003 static int bc_sendto(struct rpc_rqst *req)
2004 {
2005         int len;
2006         struct xdr_buf *xbufp = &req->rq_snd_buf;
2007         struct rpc_xprt *xprt = req->rq_xprt;
2008         struct sock_xprt *transport =
2009                                 container_of(xprt, struct sock_xprt, xprt);
2010         struct socket *sock = transport->sock;
2011         unsigned long headoff;
2012         unsigned long tailoff;
2013
2014         /*
2015          * Set up the rpc header and record marker stuff
2016          */
2017         xs_encode_tcp_record_marker(xbufp);
2018
2019         tailoff = (unsigned long)xbufp->tail[0].iov_base & ~PAGE_MASK;
2020         headoff = (unsigned long)xbufp->head[0].iov_base & ~PAGE_MASK;
2021         len = svc_send_common(sock, xbufp,
2022                               virt_to_page(xbufp->head[0].iov_base), headoff,
2023                               xbufp->tail[0].iov_base, tailoff);
2024
2025         if (len != xbufp->len) {
2026                 printk(KERN_NOTICE "Error sending entire callback!\n");
2027                 len = -EAGAIN;
2028         }
2029
2030         return len;
2031 }
2032
2033 /*
2034  * The send routine. Borrows from svc_send
2035  */
2036 static int bc_send_request(struct rpc_task *task)
2037 {
2038         struct rpc_rqst *req = task->tk_rqstp;
2039         struct svc_xprt *xprt;
2040         struct svc_sock         *svsk;
2041         u32                     len;
2042
2043         dprintk("sending request with xid: %08x\n", ntohl(req->rq_xid));
2044         /*
2045          * Get the server socket associated with this callback xprt
2046          */
2047         xprt = req->rq_xprt->bc_xprt;
2048         svsk = container_of(xprt, struct svc_sock, sk_xprt);
2049
2050         /*
2051          * Grab the mutex to serialize data as the connection is shared
2052          * with the fore channel
2053          */
2054         if (!mutex_trylock(&xprt->xpt_mutex)) {
2055                 rpc_sleep_on(&xprt->xpt_bc_pending, task, NULL);
2056                 if (!mutex_trylock(&xprt->xpt_mutex))
2057                         return -EAGAIN;
2058                 rpc_wake_up_queued_task(&xprt->xpt_bc_pending, task);
2059         }
2060         if (test_bit(XPT_DEAD, &xprt->xpt_flags))
2061                 len = -ENOTCONN;
2062         else
2063                 len = bc_sendto(req);
2064         mutex_unlock(&xprt->xpt_mutex);
2065
2066         if (len > 0)
2067                 len = 0;
2068
2069         return len;
2070 }
2071
2072 /*
2073  * The close routine. Since this is client initiated, we do nothing
2074  */
2075
2076 static void bc_close(struct rpc_xprt *xprt)
2077 {
2078 }
2079
2080 /*
2081  * The xprt destroy routine. Again, because this connection is client
2082  * initiated, we do nothing
2083  */
2084
2085 static void bc_destroy(struct rpc_xprt *xprt)
2086 {
2087 }
2088
2089 static struct rpc_xprt_ops xs_udp_ops = {
2090         .set_buffer_size        = xs_udp_set_buffer_size,
2091         .reserve_xprt           = xprt_reserve_xprt_cong,
2092         .release_xprt           = xprt_release_xprt_cong,
2093         .rpcbind                = rpcb_getport_async,
2094         .set_port               = xs_set_port,
2095         .connect                = xs_connect,
2096         .buf_alloc              = rpc_malloc,
2097         .buf_free               = rpc_free,
2098         .send_request           = xs_udp_send_request,
2099         .set_retrans_timeout    = xprt_set_retrans_timeout_rtt,
2100         .timer                  = xs_udp_timer,
2101         .release_request        = xprt_release_rqst_cong,
2102         .close                  = xs_close,
2103         .destroy                = xs_destroy,
2104         .print_stats            = xs_udp_print_stats,
2105 };
2106
2107 static struct rpc_xprt_ops xs_tcp_ops = {
2108         .reserve_xprt           = xprt_reserve_xprt,
2109         .release_xprt           = xs_tcp_release_xprt,
2110         .rpcbind                = rpcb_getport_async,
2111         .set_port               = xs_set_port,
2112         .connect                = xs_connect,
2113         .buf_alloc              = rpc_malloc,
2114         .buf_free               = rpc_free,
2115         .send_request           = xs_tcp_send_request,
2116         .set_retrans_timeout    = xprt_set_retrans_timeout_def,
2117         .close                  = xs_tcp_close,
2118         .destroy                = xs_destroy,
2119         .print_stats            = xs_tcp_print_stats,
2120 };
2121
2122 /*
2123  * The rpc_xprt_ops for the server backchannel
2124  */
2125
2126 static struct rpc_xprt_ops bc_tcp_ops = {
2127         .reserve_xprt           = xprt_reserve_xprt,
2128         .release_xprt           = xprt_release_xprt,
2129         .buf_alloc              = bc_malloc,
2130         .buf_free               = bc_free,
2131         .send_request           = bc_send_request,
2132         .set_retrans_timeout    = xprt_set_retrans_timeout_def,
2133         .close                  = bc_close,
2134         .destroy                = bc_destroy,
2135         .print_stats            = xs_tcp_print_stats,
2136 };
2137
2138 static int xs_init_anyaddr(const int family, struct sockaddr *sap)
2139 {
2140         static const struct sockaddr_in sin = {
2141                 .sin_family             = AF_INET,
2142                 .sin_addr.s_addr        = htonl(INADDR_ANY),
2143         };
2144         static const struct sockaddr_in6 sin6 = {
2145                 .sin6_family            = AF_INET6,
2146                 .sin6_addr              = IN6ADDR_ANY_INIT,
2147         };
2148
2149         switch (family) {
2150         case AF_INET:
2151                 memcpy(sap, &sin, sizeof(sin));
2152                 break;
2153         case AF_INET6:
2154                 memcpy(sap, &sin6, sizeof(sin6));
2155                 break;
2156         default:
2157                 dprintk("RPC:       %s: Bad address family\n", __func__);
2158                 return -EAFNOSUPPORT;
2159         }
2160         return 0;
2161 }
2162
2163 static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
2164                                       unsigned int slot_table_size)
2165 {
2166         struct rpc_xprt *xprt;
2167         struct sock_xprt *new;
2168
2169         if (args->addrlen > sizeof(xprt->addr)) {
2170                 dprintk("RPC:       xs_setup_xprt: address too large\n");
2171                 return ERR_PTR(-EBADF);
2172         }
2173
2174         xprt = xprt_alloc(args->net, sizeof(*new), slot_table_size);
2175         if (xprt == NULL) {
2176                 dprintk("RPC:       xs_setup_xprt: couldn't allocate "
2177                                 "rpc_xprt\n");
2178                 return ERR_PTR(-ENOMEM);
2179         }
2180
2181         new = container_of(xprt, struct sock_xprt, xprt);
2182         memcpy(&xprt->addr, args->dstaddr, args->addrlen);
2183         xprt->addrlen = args->addrlen;
2184         if (args->srcaddr)
2185                 memcpy(&new->srcaddr, args->srcaddr, args->addrlen);
2186         else {
2187                 int err;
2188                 err = xs_init_anyaddr(args->dstaddr->sa_family,
2189                                         (struct sockaddr *)&new->srcaddr);
2190                 if (err != 0)
2191                         return ERR_PTR(err);
2192         }
2193
2194         return xprt;
2195 }
2196
2197 static const struct rpc_timeout xs_udp_default_timeout = {
2198         .to_initval = 5 * HZ,
2199         .to_maxval = 30 * HZ,
2200         .to_increment = 5 * HZ,
2201         .to_retries = 5,
2202 };
2203
2204 /**
2205  * xs_setup_udp - Set up transport to use a UDP socket
2206  * @args: rpc transport creation arguments
2207  *
2208  */
2209 static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
2210 {
2211         struct sockaddr *addr = args->dstaddr;
2212         struct rpc_xprt *xprt;
2213         struct sock_xprt *transport;
2214         struct rpc_xprt *ret;
2215
2216         xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries);
2217         if (IS_ERR(xprt))
2218                 return xprt;
2219         transport = container_of(xprt, struct sock_xprt, xprt);
2220
2221         xprt->prot = IPPROTO_UDP;
2222         xprt->tsh_size = 0;
2223         /* XXX: header size can vary due to auth type, IPv6, etc. */
2224         xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
2225
2226         xprt->bind_timeout = XS_BIND_TO;
2227         xprt->reestablish_timeout = XS_UDP_REEST_TO;
2228         xprt->idle_timeout = XS_IDLE_DISC_TO;
2229
2230         xprt->ops = &xs_udp_ops;
2231
2232         xprt->timeout = &xs_udp_default_timeout;
2233
2234         switch (addr->sa_family) {
2235         case AF_INET:
2236                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2237                         xprt_set_bound(xprt);
2238
2239                 INIT_DELAYED_WORK(&transport->connect_worker,
2240                                         xs_udp_setup_socket);
2241                 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
2242                 break;
2243         case AF_INET6:
2244                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2245                         xprt_set_bound(xprt);
2246
2247                 INIT_DELAYED_WORK(&transport->connect_worker,
2248                                         xs_udp_setup_socket);
2249                 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
2250                 break;
2251         default:
2252                 ret = ERR_PTR(-EAFNOSUPPORT);
2253                 goto out_err;
2254         }
2255
2256         if (xprt_bound(xprt))
2257                 dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
2258                                 xprt->address_strings[RPC_DISPLAY_ADDR],
2259                                 xprt->address_strings[RPC_DISPLAY_PORT],
2260                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
2261         else
2262                 dprintk("RPC:       set up xprt to %s (autobind) via %s\n",
2263                                 xprt->address_strings[RPC_DISPLAY_ADDR],
2264                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
2265
2266         if (try_module_get(THIS_MODULE))
2267                 return xprt;
2268         ret = ERR_PTR(-EINVAL);
2269 out_err:
2270         xprt_free(xprt);
2271         return ret;
2272 }
2273
2274 static const struct rpc_timeout xs_tcp_default_timeout = {
2275         .to_initval = 60 * HZ,
2276         .to_maxval = 60 * HZ,
2277         .to_retries = 2,
2278 };
2279
2280 /**
2281  * xs_setup_tcp - Set up transport to use a TCP socket
2282  * @args: rpc transport creation arguments
2283  *
2284  */
2285 static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
2286 {
2287         struct sockaddr *addr = args->dstaddr;
2288         struct rpc_xprt *xprt;
2289         struct sock_xprt *transport;
2290         struct rpc_xprt *ret;
2291
2292         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries);
2293         if (IS_ERR(xprt))
2294                 return xprt;
2295         transport = container_of(xprt, struct sock_xprt, xprt);
2296
2297         xprt->prot = IPPROTO_TCP;
2298         xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2299         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2300
2301         xprt->bind_timeout = XS_BIND_TO;
2302         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2303         xprt->idle_timeout = XS_IDLE_DISC_TO;
2304
2305         xprt->ops = &xs_tcp_ops;
2306         xprt->timeout = &xs_tcp_default_timeout;
2307
2308         switch (addr->sa_family) {
2309         case AF_INET:
2310                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2311                         xprt_set_bound(xprt);
2312
2313                 INIT_DELAYED_WORK(&transport->connect_worker,
2314                                         xs_tcp_setup_socket);
2315                 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
2316                 break;
2317         case AF_INET6:
2318                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2319                         xprt_set_bound(xprt);
2320
2321                 INIT_DELAYED_WORK(&transport->connect_worker,
2322                                         xs_tcp_setup_socket);
2323                 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
2324                 break;
2325         default:
2326                 ret = ERR_PTR(-EAFNOSUPPORT);
2327                 goto out_err;
2328         }
2329
2330         if (xprt_bound(xprt))
2331                 dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
2332                                 xprt->address_strings[RPC_DISPLAY_ADDR],
2333                                 xprt->address_strings[RPC_DISPLAY_PORT],
2334                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
2335         else
2336                 dprintk("RPC:       set up xprt to %s (autobind) via %s\n",
2337                                 xprt->address_strings[RPC_DISPLAY_ADDR],
2338                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
2339
2340
2341         if (try_module_get(THIS_MODULE))
2342                 return xprt;
2343         ret = ERR_PTR(-EINVAL);
2344 out_err:
2345         xprt_free(xprt);
2346         return ret;
2347 }
2348
2349 /**
2350  * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket
2351  * @args: rpc transport creation arguments
2352  *
2353  */
2354 static struct rpc_xprt *xs_setup_bc_tcp(struct xprt_create *args)
2355 {
2356         struct sockaddr *addr = args->dstaddr;
2357         struct rpc_xprt *xprt;
2358         struct sock_xprt *transport;
2359         struct svc_sock *bc_sock;
2360         struct rpc_xprt *ret;
2361
2362         if (args->bc_xprt->xpt_bc_xprt) {
2363                 /*
2364                  * This server connection already has a backchannel
2365                  * export; we can't create a new one, as we wouldn't be
2366                  * able to match replies based on xid any more.  So,
2367                  * reuse the already-existing one:
2368                  */
2369                  return args->bc_xprt->xpt_bc_xprt;
2370         }
2371         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries);
2372         if (IS_ERR(xprt))
2373                 return xprt;
2374         transport = container_of(xprt, struct sock_xprt, xprt);
2375
2376         xprt->prot = IPPROTO_TCP;
2377         xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2378         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2379         xprt->timeout = &xs_tcp_default_timeout;
2380
2381         /* backchannel */
2382         xprt_set_bound(xprt);
2383         xprt->bind_timeout = 0;
2384         xprt->reestablish_timeout = 0;
2385         xprt->idle_timeout = 0;
2386
2387         xprt->ops = &bc_tcp_ops;
2388
2389         switch (addr->sa_family) {
2390         case AF_INET:
2391                 xs_format_peer_addresses(xprt, "tcp",
2392                                          RPCBIND_NETID_TCP);
2393                 break;
2394         case AF_INET6:
2395                 xs_format_peer_addresses(xprt, "tcp",
2396                                    RPCBIND_NETID_TCP6);
2397                 break;
2398         default:
2399                 ret = ERR_PTR(-EAFNOSUPPORT);
2400                 goto out_err;
2401         }
2402
2403         dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
2404                         xprt->address_strings[RPC_DISPLAY_ADDR],
2405                         xprt->address_strings[RPC_DISPLAY_PORT],
2406                         xprt->address_strings[RPC_DISPLAY_PROTO]);
2407
2408         /*
2409          * Once we've associated a backchannel xprt with a connection,
2410          * we want to keep it around as long as long as the connection
2411          * lasts, in case we need to start using it for a backchannel
2412          * again; this reference won't be dropped until bc_xprt is
2413          * destroyed.
2414          */
2415         xprt_get(xprt);
2416         args->bc_xprt->xpt_bc_xprt = xprt;
2417         xprt->bc_xprt = args->bc_xprt;
2418         bc_sock = container_of(args->bc_xprt, struct svc_sock, sk_xprt);
2419         transport->sock = bc_sock->sk_sock;
2420         transport->inet = bc_sock->sk_sk;
2421
2422         /*
2423          * Since we don't want connections for the backchannel, we set
2424          * the xprt status to connected
2425          */
2426         xprt_set_connected(xprt);
2427
2428
2429         if (try_module_get(THIS_MODULE))
2430                 return xprt;
2431         xprt_put(xprt);
2432         ret = ERR_PTR(-EINVAL);
2433 out_err:
2434         xprt_free(xprt);
2435         return ret;
2436 }
2437
2438 static struct xprt_class        xs_udp_transport = {
2439         .list           = LIST_HEAD_INIT(xs_udp_transport.list),
2440         .name           = "udp",
2441         .owner          = THIS_MODULE,
2442         .ident          = XPRT_TRANSPORT_UDP,
2443         .setup          = xs_setup_udp,
2444 };
2445
2446 static struct xprt_class        xs_tcp_transport = {
2447         .list           = LIST_HEAD_INIT(xs_tcp_transport.list),
2448         .name           = "tcp",
2449         .owner          = THIS_MODULE,
2450         .ident          = XPRT_TRANSPORT_TCP,
2451         .setup          = xs_setup_tcp,
2452 };
2453
2454 static struct xprt_class        xs_bc_tcp_transport = {
2455         .list           = LIST_HEAD_INIT(xs_bc_tcp_transport.list),
2456         .name           = "tcp NFSv4.1 backchannel",
2457         .owner          = THIS_MODULE,
2458         .ident          = XPRT_TRANSPORT_BC_TCP,
2459         .setup          = xs_setup_bc_tcp,
2460 };
2461
2462 /**
2463  * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
2464  *
2465  */
2466 int init_socket_xprt(void)
2467 {
2468 #ifdef RPC_DEBUG
2469         if (!sunrpc_table_header)
2470                 sunrpc_table_header = register_sysctl_table(sunrpc_table);
2471 #endif
2472
2473         xprt_register_transport(&xs_udp_transport);
2474         xprt_register_transport(&xs_tcp_transport);
2475         xprt_register_transport(&xs_bc_tcp_transport);
2476
2477         return 0;
2478 }
2479
2480 /**
2481  * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
2482  *
2483  */
2484 void cleanup_socket_xprt(void)
2485 {
2486 #ifdef RPC_DEBUG
2487         if (sunrpc_table_header) {
2488                 unregister_sysctl_table(sunrpc_table_header);
2489                 sunrpc_table_header = NULL;
2490         }
2491 #endif
2492
2493         xprt_unregister_transport(&xs_udp_transport);
2494         xprt_unregister_transport(&xs_tcp_transport);
2495         xprt_unregister_transport(&xs_bc_tcp_transport);
2496 }
2497
2498 static int param_set_uint_minmax(const char *val,
2499                 const struct kernel_param *kp,
2500                 unsigned int min, unsigned int max)
2501 {
2502         unsigned long num;
2503         int ret;
2504
2505         if (!val)
2506                 return -EINVAL;
2507         ret = strict_strtoul(val, 0, &num);
2508         if (ret == -EINVAL || num < min || num > max)
2509                 return -EINVAL;
2510         *((unsigned int *)kp->arg) = num;
2511         return 0;
2512 }
2513
2514 static int param_set_portnr(const char *val, const struct kernel_param *kp)
2515 {
2516         return param_set_uint_minmax(val, kp,
2517                         RPC_MIN_RESVPORT,
2518                         RPC_MAX_RESVPORT);
2519 }
2520
2521 static struct kernel_param_ops param_ops_portnr = {
2522         .set = param_set_portnr,
2523         .get = param_get_uint,
2524 };
2525
2526 #define param_check_portnr(name, p) \
2527         __param_check(name, p, unsigned int);
2528
2529 module_param_named(min_resvport, xprt_min_resvport, portnr, 0644);
2530 module_param_named(max_resvport, xprt_max_resvport, portnr, 0644);
2531
2532 static int param_set_slot_table_size(const char *val,
2533                                      const struct kernel_param *kp)
2534 {
2535         return param_set_uint_minmax(val, kp,
2536                         RPC_MIN_SLOT_TABLE,
2537                         RPC_MAX_SLOT_TABLE);
2538 }
2539
2540 static struct kernel_param_ops param_ops_slot_table_size = {
2541         .set = param_set_slot_table_size,
2542         .get = param_get_uint,
2543 };
2544
2545 #define param_check_slot_table_size(name, p) \
2546         __param_check(name, p, unsigned int);
2547
2548 module_param_named(tcp_slot_table_entries, xprt_tcp_slot_table_entries,
2549                    slot_table_size, 0644);
2550 module_param_named(udp_slot_table_entries, xprt_udp_slot_table_entries,
2551                    slot_table_size, 0644);
2552