SUNRPC: TCP clear XPRT_CLOSE_WAIT when the socket is closed for writes
[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 Software <alan@redhat.com>
7  * TCP send fixes (C) 1998 Red Hat Software <alan@redhat.com>
8  * TCP NFS related read + write fixes
9  *  (C) 1999 Dave Airlie, University of Limerick, Ireland <airlied@linux.ie>
10  *
11  * Rewrite of larges part of the code in order to stabilize TCP stuff.
12  * Fix behaviour when socket buffer is full.
13  *  (C) 1999 Trond Myklebust <trond.myklebust@fys.uio.no>
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/xprtsock.h>
36 #include <linux/file.h>
37
38 #include <net/sock.h>
39 #include <net/checksum.h>
40 #include <net/udp.h>
41 #include <net/tcp.h>
42
43 /*
44  * xprtsock tunables
45  */
46 unsigned int xprt_udp_slot_table_entries = RPC_DEF_SLOT_TABLE;
47 unsigned int xprt_tcp_slot_table_entries = RPC_DEF_SLOT_TABLE;
48
49 unsigned int xprt_min_resvport = RPC_DEF_MIN_RESVPORT;
50 unsigned int xprt_max_resvport = RPC_DEF_MAX_RESVPORT;
51
52 /*
53  * We can register our own files under /proc/sys/sunrpc by
54  * calling register_sysctl_table() again.  The files in that
55  * directory become the union of all files registered there.
56  *
57  * We simply need to make sure that we don't collide with
58  * someone else's file names!
59  */
60
61 #ifdef RPC_DEBUG
62
63 static unsigned int min_slot_table_size = RPC_MIN_SLOT_TABLE;
64 static unsigned int max_slot_table_size = RPC_MAX_SLOT_TABLE;
65 static unsigned int xprt_min_resvport_limit = RPC_MIN_RESVPORT;
66 static unsigned int xprt_max_resvport_limit = RPC_MAX_RESVPORT;
67
68 static struct ctl_table_header *sunrpc_table_header;
69
70 /*
71  * FIXME: changing the UDP slot table size should also resize the UDP
72  *        socket buffers for existing UDP transports
73  */
74 static ctl_table xs_tunables_table[] = {
75         {
76                 .ctl_name       = CTL_SLOTTABLE_UDP,
77                 .procname       = "udp_slot_table_entries",
78                 .data           = &xprt_udp_slot_table_entries,
79                 .maxlen         = sizeof(unsigned int),
80                 .mode           = 0644,
81                 .proc_handler   = &proc_dointvec_minmax,
82                 .strategy       = &sysctl_intvec,
83                 .extra1         = &min_slot_table_size,
84                 .extra2         = &max_slot_table_size
85         },
86         {
87                 .ctl_name       = CTL_SLOTTABLE_TCP,
88                 .procname       = "tcp_slot_table_entries",
89                 .data           = &xprt_tcp_slot_table_entries,
90                 .maxlen         = sizeof(unsigned int),
91                 .mode           = 0644,
92                 .proc_handler   = &proc_dointvec_minmax,
93                 .strategy       = &sysctl_intvec,
94                 .extra1         = &min_slot_table_size,
95                 .extra2         = &max_slot_table_size
96         },
97         {
98                 .ctl_name       = CTL_MIN_RESVPORT,
99                 .procname       = "min_resvport",
100                 .data           = &xprt_min_resvport,
101                 .maxlen         = sizeof(unsigned int),
102                 .mode           = 0644,
103                 .proc_handler   = &proc_dointvec_minmax,
104                 .strategy       = &sysctl_intvec,
105                 .extra1         = &xprt_min_resvport_limit,
106                 .extra2         = &xprt_max_resvport_limit
107         },
108         {
109                 .ctl_name       = CTL_MAX_RESVPORT,
110                 .procname       = "max_resvport",
111                 .data           = &xprt_max_resvport,
112                 .maxlen         = sizeof(unsigned int),
113                 .mode           = 0644,
114                 .proc_handler   = &proc_dointvec_minmax,
115                 .strategy       = &sysctl_intvec,
116                 .extra1         = &xprt_min_resvport_limit,
117                 .extra2         = &xprt_max_resvport_limit
118         },
119         {
120                 .ctl_name = 0,
121         },
122 };
123
124 static ctl_table sunrpc_table[] = {
125         {
126                 .ctl_name       = CTL_SUNRPC,
127                 .procname       = "sunrpc",
128                 .mode           = 0555,
129                 .child          = xs_tunables_table
130         },
131         {
132                 .ctl_name = 0,
133         },
134 };
135
136 #endif
137
138 /*
139  * How many times to try sending a request on a socket before waiting
140  * for the socket buffer to clear.
141  */
142 #define XS_SENDMSG_RETRY        (10U)
143
144 /*
145  * Time out for an RPC UDP socket connect.  UDP socket connects are
146  * synchronous, but we set a timeout anyway in case of resource
147  * exhaustion on the local host.
148  */
149 #define XS_UDP_CONN_TO          (5U * HZ)
150
151 /*
152  * Wait duration for an RPC TCP connection to be established.  Solaris
153  * NFS over TCP uses 60 seconds, for example, which is in line with how
154  * long a server takes to reboot.
155  */
156 #define XS_TCP_CONN_TO          (60U * HZ)
157
158 /*
159  * Wait duration for a reply from the RPC portmapper.
160  */
161 #define XS_BIND_TO              (60U * HZ)
162
163 /*
164  * Delay if a UDP socket connect error occurs.  This is most likely some
165  * kind of resource problem on the local host.
166  */
167 #define XS_UDP_REEST_TO         (2U * HZ)
168
169 /*
170  * The reestablish timeout allows clients to delay for a bit before attempting
171  * to reconnect to a server that just dropped our connection.
172  *
173  * We implement an exponential backoff when trying to reestablish a TCP
174  * transport connection with the server.  Some servers like to drop a TCP
175  * connection when they are overworked, so we start with a short timeout and
176  * increase over time if the server is down or not responding.
177  */
178 #define XS_TCP_INIT_REEST_TO    (3U * HZ)
179 #define XS_TCP_MAX_REEST_TO     (5U * 60 * HZ)
180
181 /*
182  * TCP idle timeout; client drops the transport socket if it is idle
183  * for this long.  Note that we also timeout UDP sockets to prevent
184  * holding port numbers when there is no RPC traffic.
185  */
186 #define XS_IDLE_DISC_TO         (5U * 60 * HZ)
187
188 #ifdef RPC_DEBUG
189 # undef  RPC_DEBUG_DATA
190 # define RPCDBG_FACILITY        RPCDBG_TRANS
191 #endif
192
193 #ifdef RPC_DEBUG_DATA
194 static void xs_pktdump(char *msg, u32 *packet, unsigned int count)
195 {
196         u8 *buf = (u8 *) packet;
197         int j;
198
199         dprintk("RPC:       %s\n", msg);
200         for (j = 0; j < count && j < 128; j += 4) {
201                 if (!(j & 31)) {
202                         if (j)
203                                 dprintk("\n");
204                         dprintk("0x%04x ", j);
205                 }
206                 dprintk("%02x%02x%02x%02x ",
207                         buf[j], buf[j+1], buf[j+2], buf[j+3]);
208         }
209         dprintk("\n");
210 }
211 #else
212 static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count)
213 {
214         /* NOP */
215 }
216 #endif
217
218 struct sock_xprt {
219         struct rpc_xprt         xprt;
220
221         /*
222          * Network layer
223          */
224         struct socket *         sock;
225         struct sock *           inet;
226
227         /*
228          * State of TCP reply receive
229          */
230         __be32                  tcp_fraghdr,
231                                 tcp_xid;
232
233         u32                     tcp_offset,
234                                 tcp_reclen;
235
236         unsigned long           tcp_copied,
237                                 tcp_flags;
238
239         /*
240          * Connection of transports
241          */
242         struct delayed_work     connect_worker;
243         struct sockaddr_storage addr;
244         unsigned short          port;
245
246         /*
247          * UDP socket buffer size parameters
248          */
249         size_t                  rcvsize,
250                                 sndsize;
251
252         /*
253          * Saved socket callback addresses
254          */
255         void                    (*old_data_ready)(struct sock *, int);
256         void                    (*old_state_change)(struct sock *);
257         void                    (*old_write_space)(struct sock *);
258 };
259
260 /*
261  * TCP receive state flags
262  */
263 #define TCP_RCV_LAST_FRAG       (1UL << 0)
264 #define TCP_RCV_COPY_FRAGHDR    (1UL << 1)
265 #define TCP_RCV_COPY_XID        (1UL << 2)
266 #define TCP_RCV_COPY_DATA       (1UL << 3)
267
268 static inline struct sockaddr *xs_addr(struct rpc_xprt *xprt)
269 {
270         return (struct sockaddr *) &xprt->addr;
271 }
272
273 static inline struct sockaddr_in *xs_addr_in(struct rpc_xprt *xprt)
274 {
275         return (struct sockaddr_in *) &xprt->addr;
276 }
277
278 static inline struct sockaddr_in6 *xs_addr_in6(struct rpc_xprt *xprt)
279 {
280         return (struct sockaddr_in6 *) &xprt->addr;
281 }
282
283 static void xs_format_ipv4_peer_addresses(struct rpc_xprt *xprt)
284 {
285         struct sockaddr_in *addr = xs_addr_in(xprt);
286         char *buf;
287
288         buf = kzalloc(20, GFP_KERNEL);
289         if (buf) {
290                 snprintf(buf, 20, NIPQUAD_FMT,
291                                 NIPQUAD(addr->sin_addr.s_addr));
292         }
293         xprt->address_strings[RPC_DISPLAY_ADDR] = buf;
294
295         buf = kzalloc(8, GFP_KERNEL);
296         if (buf) {
297                 snprintf(buf, 8, "%u",
298                                 ntohs(addr->sin_port));
299         }
300         xprt->address_strings[RPC_DISPLAY_PORT] = buf;
301
302         buf = kzalloc(8, GFP_KERNEL);
303         if (buf) {
304                 if (xprt->prot == IPPROTO_UDP)
305                         snprintf(buf, 8, "udp");
306                 else
307                         snprintf(buf, 8, "tcp");
308         }
309         xprt->address_strings[RPC_DISPLAY_PROTO] = buf;
310
311         buf = kzalloc(48, GFP_KERNEL);
312         if (buf) {
313                 snprintf(buf, 48, "addr="NIPQUAD_FMT" port=%u proto=%s",
314                         NIPQUAD(addr->sin_addr.s_addr),
315                         ntohs(addr->sin_port),
316                         xprt->prot == IPPROTO_UDP ? "udp" : "tcp");
317         }
318         xprt->address_strings[RPC_DISPLAY_ALL] = buf;
319
320         buf = kzalloc(10, GFP_KERNEL);
321         if (buf) {
322                 snprintf(buf, 10, "%02x%02x%02x%02x",
323                                 NIPQUAD(addr->sin_addr.s_addr));
324         }
325         xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = buf;
326
327         buf = kzalloc(8, GFP_KERNEL);
328         if (buf) {
329                 snprintf(buf, 8, "%4hx",
330                                 ntohs(addr->sin_port));
331         }
332         xprt->address_strings[RPC_DISPLAY_HEX_PORT] = buf;
333
334         buf = kzalloc(30, GFP_KERNEL);
335         if (buf) {
336                 snprintf(buf, 30, NIPQUAD_FMT".%u.%u",
337                                 NIPQUAD(addr->sin_addr.s_addr),
338                                 ntohs(addr->sin_port) >> 8,
339                                 ntohs(addr->sin_port) & 0xff);
340         }
341         xprt->address_strings[RPC_DISPLAY_UNIVERSAL_ADDR] = buf;
342
343         xprt->address_strings[RPC_DISPLAY_NETID] =
344                 kstrdup(xprt->prot == IPPROTO_UDP ?
345                         RPCBIND_NETID_UDP : RPCBIND_NETID_TCP, GFP_KERNEL);
346 }
347
348 static void xs_format_ipv6_peer_addresses(struct rpc_xprt *xprt)
349 {
350         struct sockaddr_in6 *addr = xs_addr_in6(xprt);
351         char *buf;
352
353         buf = kzalloc(40, GFP_KERNEL);
354         if (buf) {
355                 snprintf(buf, 40, NIP6_FMT,
356                                 NIP6(addr->sin6_addr));
357         }
358         xprt->address_strings[RPC_DISPLAY_ADDR] = buf;
359
360         buf = kzalloc(8, GFP_KERNEL);
361         if (buf) {
362                 snprintf(buf, 8, "%u",
363                                 ntohs(addr->sin6_port));
364         }
365         xprt->address_strings[RPC_DISPLAY_PORT] = buf;
366
367         buf = kzalloc(8, GFP_KERNEL);
368         if (buf) {
369                 if (xprt->prot == IPPROTO_UDP)
370                         snprintf(buf, 8, "udp");
371                 else
372                         snprintf(buf, 8, "tcp");
373         }
374         xprt->address_strings[RPC_DISPLAY_PROTO] = buf;
375
376         buf = kzalloc(64, GFP_KERNEL);
377         if (buf) {
378                 snprintf(buf, 64, "addr="NIP6_FMT" port=%u proto=%s",
379                                 NIP6(addr->sin6_addr),
380                                 ntohs(addr->sin6_port),
381                                 xprt->prot == IPPROTO_UDP ? "udp" : "tcp");
382         }
383         xprt->address_strings[RPC_DISPLAY_ALL] = buf;
384
385         buf = kzalloc(36, GFP_KERNEL);
386         if (buf) {
387                 snprintf(buf, 36, NIP6_SEQFMT,
388                                 NIP6(addr->sin6_addr));
389         }
390         xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = buf;
391
392         buf = kzalloc(8, GFP_KERNEL);
393         if (buf) {
394                 snprintf(buf, 8, "%4hx",
395                                 ntohs(addr->sin6_port));
396         }
397         xprt->address_strings[RPC_DISPLAY_HEX_PORT] = buf;
398
399         buf = kzalloc(50, GFP_KERNEL);
400         if (buf) {
401                 snprintf(buf, 50, NIP6_FMT".%u.%u",
402                                 NIP6(addr->sin6_addr),
403                                 ntohs(addr->sin6_port) >> 8,
404                                 ntohs(addr->sin6_port) & 0xff);
405         }
406         xprt->address_strings[RPC_DISPLAY_UNIVERSAL_ADDR] = buf;
407
408         xprt->address_strings[RPC_DISPLAY_NETID] =
409                 kstrdup(xprt->prot == IPPROTO_UDP ?
410                         RPCBIND_NETID_UDP6 : RPCBIND_NETID_TCP6, GFP_KERNEL);
411 }
412
413 static void xs_free_peer_addresses(struct rpc_xprt *xprt)
414 {
415         int i;
416
417         for (i = 0; i < RPC_DISPLAY_MAX; i++)
418                 kfree(xprt->address_strings[i]);
419 }
420
421 #define XS_SENDMSG_FLAGS        (MSG_DONTWAIT | MSG_NOSIGNAL)
422
423 static int xs_send_kvec(struct socket *sock, struct sockaddr *addr, int addrlen, struct kvec *vec, unsigned int base, int more)
424 {
425         struct msghdr msg = {
426                 .msg_name       = addr,
427                 .msg_namelen    = addrlen,
428                 .msg_flags      = XS_SENDMSG_FLAGS | (more ? MSG_MORE : 0),
429         };
430         struct kvec iov = {
431                 .iov_base       = vec->iov_base + base,
432                 .iov_len        = vec->iov_len - base,
433         };
434
435         if (iov.iov_len != 0)
436                 return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
437         return kernel_sendmsg(sock, &msg, NULL, 0, 0);
438 }
439
440 static int xs_send_pagedata(struct socket *sock, struct xdr_buf *xdr, unsigned int base, int more)
441 {
442         struct page **ppage;
443         unsigned int remainder;
444         int err, sent = 0;
445
446         remainder = xdr->page_len - base;
447         base += xdr->page_base;
448         ppage = xdr->pages + (base >> PAGE_SHIFT);
449         base &= ~PAGE_MASK;
450         for(;;) {
451                 unsigned int len = min_t(unsigned int, PAGE_SIZE - base, remainder);
452                 int flags = XS_SENDMSG_FLAGS;
453
454                 remainder -= len;
455                 if (remainder != 0 || more)
456                         flags |= MSG_MORE;
457                 err = sock->ops->sendpage(sock, *ppage, base, len, flags);
458                 if (remainder == 0 || err != len)
459                         break;
460                 sent += err;
461                 ppage++;
462                 base = 0;
463         }
464         if (sent == 0)
465                 return err;
466         if (err > 0)
467                 sent += err;
468         return sent;
469 }
470
471 /**
472  * xs_sendpages - write pages directly to a socket
473  * @sock: socket to send on
474  * @addr: UDP only -- address of destination
475  * @addrlen: UDP only -- length of destination address
476  * @xdr: buffer containing this request
477  * @base: starting position in the buffer
478  *
479  */
480 static int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base)
481 {
482         unsigned int remainder = xdr->len - base;
483         int err, sent = 0;
484
485         if (unlikely(!sock))
486                 return -ENOTCONN;
487
488         clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags);
489         if (base != 0) {
490                 addr = NULL;
491                 addrlen = 0;
492         }
493
494         if (base < xdr->head[0].iov_len || addr != NULL) {
495                 unsigned int len = xdr->head[0].iov_len - base;
496                 remainder -= len;
497                 err = xs_send_kvec(sock, addr, addrlen, &xdr->head[0], base, remainder != 0);
498                 if (remainder == 0 || err != len)
499                         goto out;
500                 sent += err;
501                 base = 0;
502         } else
503                 base -= xdr->head[0].iov_len;
504
505         if (base < xdr->page_len) {
506                 unsigned int len = xdr->page_len - base;
507                 remainder -= len;
508                 err = xs_send_pagedata(sock, xdr, base, remainder != 0);
509                 if (remainder == 0 || err != len)
510                         goto out;
511                 sent += err;
512                 base = 0;
513         } else
514                 base -= xdr->page_len;
515
516         if (base >= xdr->tail[0].iov_len)
517                 return sent;
518         err = xs_send_kvec(sock, NULL, 0, &xdr->tail[0], base, 0);
519 out:
520         if (sent == 0)
521                 return err;
522         if (err > 0)
523                 sent += err;
524         return sent;
525 }
526
527 /**
528  * xs_nospace - place task on wait queue if transmit was incomplete
529  * @task: task to put to sleep
530  *
531  */
532 static void xs_nospace(struct rpc_task *task)
533 {
534         struct rpc_rqst *req = task->tk_rqstp;
535         struct rpc_xprt *xprt = req->rq_xprt;
536         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
537
538         dprintk("RPC: %5u xmit incomplete (%u left of %u)\n",
539                         task->tk_pid, req->rq_slen - req->rq_bytes_sent,
540                         req->rq_slen);
541
542         if (test_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags)) {
543                 /* Protect against races with write_space */
544                 spin_lock_bh(&xprt->transport_lock);
545
546                 /* Don't race with disconnect */
547                 if (!xprt_connected(xprt))
548                         task->tk_status = -ENOTCONN;
549                 else if (test_bit(SOCK_NOSPACE, &transport->sock->flags))
550                         xprt_wait_for_buffer_space(task);
551
552                 spin_unlock_bh(&xprt->transport_lock);
553         } else
554                 /* Keep holding the socket if it is blocked */
555                 rpc_delay(task, HZ>>4);
556 }
557
558 /**
559  * xs_udp_send_request - write an RPC request to a UDP socket
560  * @task: address of RPC task that manages the state of an RPC request
561  *
562  * Return values:
563  *        0:    The request has been sent
564  *   EAGAIN:    The socket was blocked, please call again later to
565  *              complete the request
566  * ENOTCONN:    Caller needs to invoke connect logic then call again
567  *    other:    Some other error occured, the request was not sent
568  */
569 static int xs_udp_send_request(struct rpc_task *task)
570 {
571         struct rpc_rqst *req = task->tk_rqstp;
572         struct rpc_xprt *xprt = req->rq_xprt;
573         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
574         struct xdr_buf *xdr = &req->rq_snd_buf;
575         int status;
576
577         xs_pktdump("packet data:",
578                                 req->rq_svec->iov_base,
579                                 req->rq_svec->iov_len);
580
581         req->rq_xtime = jiffies;
582         status = xs_sendpages(transport->sock,
583                               xs_addr(xprt),
584                               xprt->addrlen, xdr,
585                               req->rq_bytes_sent);
586
587         dprintk("RPC:       xs_udp_send_request(%u) = %d\n",
588                         xdr->len - req->rq_bytes_sent, status);
589
590         if (status >= 0) {
591                 task->tk_bytes_sent += status;
592                 if (status >= req->rq_slen)
593                         return 0;
594                 /* Still some bytes left; set up for a retry later. */
595                 status = -EAGAIN;
596         }
597
598         switch (status) {
599         case -ENETUNREACH:
600         case -EPIPE:
601         case -ECONNREFUSED:
602                 /* When the server has died, an ICMP port unreachable message
603                  * prompts ECONNREFUSED. */
604                 break;
605         case -EAGAIN:
606                 xs_nospace(task);
607                 break;
608         default:
609                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
610                         -status);
611                 break;
612         }
613
614         return status;
615 }
616
617 static inline void xs_encode_tcp_record_marker(struct xdr_buf *buf)
618 {
619         u32 reclen = buf->len - sizeof(rpc_fraghdr);
620         rpc_fraghdr *base = buf->head[0].iov_base;
621         *base = htonl(RPC_LAST_STREAM_FRAGMENT | reclen);
622 }
623
624 /**
625  * xs_tcp_send_request - write an RPC request to a TCP socket
626  * @task: address of RPC task that manages the state of an RPC request
627  *
628  * Return values:
629  *        0:    The request has been sent
630  *   EAGAIN:    The socket was blocked, please call again later to
631  *              complete the request
632  * ENOTCONN:    Caller needs to invoke connect logic then call again
633  *    other:    Some other error occured, the request was not sent
634  *
635  * XXX: In the case of soft timeouts, should we eventually give up
636  *      if sendmsg is not able to make progress?
637  */
638 static int xs_tcp_send_request(struct rpc_task *task)
639 {
640         struct rpc_rqst *req = task->tk_rqstp;
641         struct rpc_xprt *xprt = req->rq_xprt;
642         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
643         struct xdr_buf *xdr = &req->rq_snd_buf;
644         int status;
645         unsigned int retry = 0;
646
647         xs_encode_tcp_record_marker(&req->rq_snd_buf);
648
649         xs_pktdump("packet data:",
650                                 req->rq_svec->iov_base,
651                                 req->rq_svec->iov_len);
652
653         /* Continue transmitting the packet/record. We must be careful
654          * to cope with writespace callbacks arriving _after_ we have
655          * called sendmsg(). */
656         while (1) {
657                 req->rq_xtime = jiffies;
658                 status = xs_sendpages(transport->sock,
659                                         NULL, 0, xdr, req->rq_bytes_sent);
660
661                 dprintk("RPC:       xs_tcp_send_request(%u) = %d\n",
662                                 xdr->len - req->rq_bytes_sent, status);
663
664                 if (unlikely(status < 0))
665                         break;
666
667                 /* If we've sent the entire packet, immediately
668                  * reset the count of bytes sent. */
669                 req->rq_bytes_sent += status;
670                 task->tk_bytes_sent += status;
671                 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
672                         req->rq_bytes_sent = 0;
673                         return 0;
674                 }
675
676                 status = -EAGAIN;
677                 if (retry++ > XS_SENDMSG_RETRY)
678                         break;
679         }
680
681         switch (status) {
682         case -EAGAIN:
683                 xs_nospace(task);
684                 break;
685         case -ECONNREFUSED:
686         case -ECONNRESET:
687         case -ENOTCONN:
688         case -EPIPE:
689                 status = -ENOTCONN;
690                 break;
691         default:
692                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
693                         -status);
694                 xprt_disconnect(xprt);
695                 break;
696         }
697
698         return status;
699 }
700
701 /**
702  * xs_tcp_release_xprt - clean up after a tcp transmission
703  * @xprt: transport
704  * @task: rpc task
705  *
706  * This cleans up if an error causes us to abort the transmission of a request.
707  * In this case, the socket may need to be reset in order to avoid confusing
708  * the server.
709  */
710 static void xs_tcp_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
711 {
712         struct rpc_rqst *req;
713
714         if (task != xprt->snd_task)
715                 return;
716         if (task == NULL)
717                 goto out_release;
718         req = task->tk_rqstp;
719         if (req->rq_bytes_sent == 0)
720                 goto out_release;
721         if (req->rq_bytes_sent == req->rq_snd_buf.len)
722                 goto out_release;
723         set_bit(XPRT_CLOSE_WAIT, &task->tk_xprt->state);
724 out_release:
725         xprt_release_xprt(xprt, task);
726 }
727
728 /**
729  * xs_close - close a socket
730  * @xprt: transport
731  *
732  * This is used when all requests are complete; ie, no DRC state remains
733  * on the server we want to save.
734  */
735 static void xs_close(struct rpc_xprt *xprt)
736 {
737         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
738         struct socket *sock = transport->sock;
739         struct sock *sk = transport->inet;
740
741         if (!sk)
742                 goto clear_close_wait;
743
744         dprintk("RPC:       xs_close xprt %p\n", xprt);
745
746         write_lock_bh(&sk->sk_callback_lock);
747         transport->inet = NULL;
748         transport->sock = NULL;
749
750         sk->sk_user_data = NULL;
751         sk->sk_data_ready = transport->old_data_ready;
752         sk->sk_state_change = transport->old_state_change;
753         sk->sk_write_space = transport->old_write_space;
754         write_unlock_bh(&sk->sk_callback_lock);
755
756         sk->sk_no_check = 0;
757
758         sock_release(sock);
759 clear_close_wait:
760         smp_mb__before_clear_bit();
761         clear_bit(XPRT_CONNECTED, &xprt->state);
762         clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
763         clear_bit(XPRT_CLOSING, &xprt->state);
764         smp_mb__after_clear_bit();
765 }
766
767 /**
768  * xs_destroy - prepare to shutdown a transport
769  * @xprt: doomed transport
770  *
771  */
772 static void xs_destroy(struct rpc_xprt *xprt)
773 {
774         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
775
776         dprintk("RPC:       xs_destroy xprt %p\n", xprt);
777
778         cancel_rearming_delayed_work(&transport->connect_worker);
779
780         xprt_disconnect(xprt);
781         xs_close(xprt);
782         xs_free_peer_addresses(xprt);
783         kfree(xprt->slot);
784         kfree(xprt);
785         module_put(THIS_MODULE);
786 }
787
788 static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
789 {
790         return (struct rpc_xprt *) sk->sk_user_data;
791 }
792
793 /**
794  * xs_udp_data_ready - "data ready" callback for UDP sockets
795  * @sk: socket with data to read
796  * @len: how much data to read
797  *
798  */
799 static void xs_udp_data_ready(struct sock *sk, int len)
800 {
801         struct rpc_task *task;
802         struct rpc_xprt *xprt;
803         struct rpc_rqst *rovr;
804         struct sk_buff *skb;
805         int err, repsize, copied;
806         u32 _xid;
807         __be32 *xp;
808
809         read_lock(&sk->sk_callback_lock);
810         dprintk("RPC:       xs_udp_data_ready...\n");
811         if (!(xprt = xprt_from_sock(sk)))
812                 goto out;
813
814         if ((skb = skb_recv_datagram(sk, 0, 1, &err)) == NULL)
815                 goto out;
816
817         if (xprt->shutdown)
818                 goto dropit;
819
820         repsize = skb->len - sizeof(struct udphdr);
821         if (repsize < 4) {
822                 dprintk("RPC:       impossible RPC reply size %d!\n", repsize);
823                 goto dropit;
824         }
825
826         /* Copy the XID from the skb... */
827         xp = skb_header_pointer(skb, sizeof(struct udphdr),
828                                 sizeof(_xid), &_xid);
829         if (xp == NULL)
830                 goto dropit;
831
832         /* Look up and lock the request corresponding to the given XID */
833         spin_lock(&xprt->transport_lock);
834         rovr = xprt_lookup_rqst(xprt, *xp);
835         if (!rovr)
836                 goto out_unlock;
837         task = rovr->rq_task;
838
839         if ((copied = rovr->rq_private_buf.buflen) > repsize)
840                 copied = repsize;
841
842         /* Suck it into the iovec, verify checksum if not done by hw. */
843         if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
844                 UDPX_INC_STATS_BH(sk, UDP_MIB_INERRORS);
845                 goto out_unlock;
846         }
847
848         UDPX_INC_STATS_BH(sk, UDP_MIB_INDATAGRAMS);
849
850         /* Something worked... */
851         dst_confirm(skb->dst);
852
853         xprt_adjust_cwnd(task, copied);
854         xprt_update_rtt(task);
855         xprt_complete_rqst(task, copied);
856
857  out_unlock:
858         spin_unlock(&xprt->transport_lock);
859  dropit:
860         skb_free_datagram(sk, skb);
861  out:
862         read_unlock(&sk->sk_callback_lock);
863 }
864
865 static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
866 {
867         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
868         size_t len, used;
869         char *p;
870
871         p = ((char *) &transport->tcp_fraghdr) + transport->tcp_offset;
872         len = sizeof(transport->tcp_fraghdr) - transport->tcp_offset;
873         used = xdr_skb_read_bits(desc, p, len);
874         transport->tcp_offset += used;
875         if (used != len)
876                 return;
877
878         transport->tcp_reclen = ntohl(transport->tcp_fraghdr);
879         if (transport->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
880                 transport->tcp_flags |= TCP_RCV_LAST_FRAG;
881         else
882                 transport->tcp_flags &= ~TCP_RCV_LAST_FRAG;
883         transport->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
884
885         transport->tcp_flags &= ~TCP_RCV_COPY_FRAGHDR;
886         transport->tcp_offset = 0;
887
888         /* Sanity check of the record length */
889         if (unlikely(transport->tcp_reclen < 4)) {
890                 dprintk("RPC:       invalid TCP record fragment length\n");
891                 xprt_disconnect(xprt);
892                 return;
893         }
894         dprintk("RPC:       reading TCP record fragment of length %d\n",
895                         transport->tcp_reclen);
896 }
897
898 static void xs_tcp_check_fraghdr(struct sock_xprt *transport)
899 {
900         if (transport->tcp_offset == transport->tcp_reclen) {
901                 transport->tcp_flags |= TCP_RCV_COPY_FRAGHDR;
902                 transport->tcp_offset = 0;
903                 if (transport->tcp_flags & TCP_RCV_LAST_FRAG) {
904                         transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
905                         transport->tcp_flags |= TCP_RCV_COPY_XID;
906                         transport->tcp_copied = 0;
907                 }
908         }
909 }
910
911 static inline void xs_tcp_read_xid(struct sock_xprt *transport, struct xdr_skb_reader *desc)
912 {
913         size_t len, used;
914         char *p;
915
916         len = sizeof(transport->tcp_xid) - transport->tcp_offset;
917         dprintk("RPC:       reading XID (%Zu bytes)\n", len);
918         p = ((char *) &transport->tcp_xid) + transport->tcp_offset;
919         used = xdr_skb_read_bits(desc, p, len);
920         transport->tcp_offset += used;
921         if (used != len)
922                 return;
923         transport->tcp_flags &= ~TCP_RCV_COPY_XID;
924         transport->tcp_flags |= TCP_RCV_COPY_DATA;
925         transport->tcp_copied = 4;
926         dprintk("RPC:       reading reply for XID %08x\n",
927                         ntohl(transport->tcp_xid));
928         xs_tcp_check_fraghdr(transport);
929 }
930
931 static inline void xs_tcp_read_request(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
932 {
933         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
934         struct rpc_rqst *req;
935         struct xdr_buf *rcvbuf;
936         size_t len;
937         ssize_t r;
938
939         /* Find and lock the request corresponding to this xid */
940         spin_lock(&xprt->transport_lock);
941         req = xprt_lookup_rqst(xprt, transport->tcp_xid);
942         if (!req) {
943                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
944                 dprintk("RPC:       XID %08x request not found!\n",
945                                 ntohl(transport->tcp_xid));
946                 spin_unlock(&xprt->transport_lock);
947                 return;
948         }
949
950         rcvbuf = &req->rq_private_buf;
951         len = desc->count;
952         if (len > transport->tcp_reclen - transport->tcp_offset) {
953                 struct xdr_skb_reader my_desc;
954
955                 len = transport->tcp_reclen - transport->tcp_offset;
956                 memcpy(&my_desc, desc, sizeof(my_desc));
957                 my_desc.count = len;
958                 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
959                                           &my_desc, xdr_skb_read_bits);
960                 desc->count -= r;
961                 desc->offset += r;
962         } else
963                 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
964                                           desc, xdr_skb_read_bits);
965
966         if (r > 0) {
967                 transport->tcp_copied += r;
968                 transport->tcp_offset += r;
969         }
970         if (r != len) {
971                 /* Error when copying to the receive buffer,
972                  * usually because we weren't able to allocate
973                  * additional buffer pages. All we can do now
974                  * is turn off TCP_RCV_COPY_DATA, so the request
975                  * will not receive any additional updates,
976                  * and time out.
977                  * Any remaining data from this record will
978                  * be discarded.
979                  */
980                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
981                 dprintk("RPC:       XID %08x truncated request\n",
982                                 ntohl(transport->tcp_xid));
983                 dprintk("RPC:       xprt = %p, tcp_copied = %lu, "
984                                 "tcp_offset = %u, tcp_reclen = %u\n",
985                                 xprt, transport->tcp_copied,
986                                 transport->tcp_offset, transport->tcp_reclen);
987                 goto out;
988         }
989
990         dprintk("RPC:       XID %08x read %Zd bytes\n",
991                         ntohl(transport->tcp_xid), r);
992         dprintk("RPC:       xprt = %p, tcp_copied = %lu, tcp_offset = %u, "
993                         "tcp_reclen = %u\n", xprt, transport->tcp_copied,
994                         transport->tcp_offset, transport->tcp_reclen);
995
996         if (transport->tcp_copied == req->rq_private_buf.buflen)
997                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
998         else if (transport->tcp_offset == transport->tcp_reclen) {
999                 if (transport->tcp_flags & TCP_RCV_LAST_FRAG)
1000                         transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1001         }
1002
1003 out:
1004         if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1005                 xprt_complete_rqst(req->rq_task, transport->tcp_copied);
1006         spin_unlock(&xprt->transport_lock);
1007         xs_tcp_check_fraghdr(transport);
1008 }
1009
1010 static inline void xs_tcp_read_discard(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1011 {
1012         size_t len;
1013
1014         len = transport->tcp_reclen - transport->tcp_offset;
1015         if (len > desc->count)
1016                 len = desc->count;
1017         desc->count -= len;
1018         desc->offset += len;
1019         transport->tcp_offset += len;
1020         dprintk("RPC:       discarded %Zu bytes\n", len);
1021         xs_tcp_check_fraghdr(transport);
1022 }
1023
1024 static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
1025 {
1026         struct rpc_xprt *xprt = rd_desc->arg.data;
1027         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1028         struct xdr_skb_reader desc = {
1029                 .skb    = skb,
1030                 .offset = offset,
1031                 .count  = len,
1032         };
1033
1034         dprintk("RPC:       xs_tcp_data_recv started\n");
1035         do {
1036                 /* Read in a new fragment marker if necessary */
1037                 /* Can we ever really expect to get completely empty fragments? */
1038                 if (transport->tcp_flags & TCP_RCV_COPY_FRAGHDR) {
1039                         xs_tcp_read_fraghdr(xprt, &desc);
1040                         continue;
1041                 }
1042                 /* Read in the xid if necessary */
1043                 if (transport->tcp_flags & TCP_RCV_COPY_XID) {
1044                         xs_tcp_read_xid(transport, &desc);
1045                         continue;
1046                 }
1047                 /* Read in the request data */
1048                 if (transport->tcp_flags & TCP_RCV_COPY_DATA) {
1049                         xs_tcp_read_request(xprt, &desc);
1050                         continue;
1051                 }
1052                 /* Skip over any trailing bytes on short reads */
1053                 xs_tcp_read_discard(transport, &desc);
1054         } while (desc.count);
1055         dprintk("RPC:       xs_tcp_data_recv done\n");
1056         return len - desc.count;
1057 }
1058
1059 /**
1060  * xs_tcp_data_ready - "data ready" callback for TCP sockets
1061  * @sk: socket with data to read
1062  * @bytes: how much data to read
1063  *
1064  */
1065 static void xs_tcp_data_ready(struct sock *sk, int bytes)
1066 {
1067         struct rpc_xprt *xprt;
1068         read_descriptor_t rd_desc;
1069
1070         dprintk("RPC:       xs_tcp_data_ready...\n");
1071
1072         read_lock(&sk->sk_callback_lock);
1073         if (!(xprt = xprt_from_sock(sk)))
1074                 goto out;
1075         if (xprt->shutdown)
1076                 goto out;
1077
1078         /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
1079         rd_desc.arg.data = xprt;
1080         rd_desc.count = 65536;
1081         tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
1082 out:
1083         read_unlock(&sk->sk_callback_lock);
1084 }
1085
1086 /**
1087  * xs_tcp_state_change - callback to handle TCP socket state changes
1088  * @sk: socket whose state has changed
1089  *
1090  */
1091 static void xs_tcp_state_change(struct sock *sk)
1092 {
1093         struct rpc_xprt *xprt;
1094
1095         read_lock(&sk->sk_callback_lock);
1096         if (!(xprt = xprt_from_sock(sk)))
1097                 goto out;
1098         dprintk("RPC:       xs_tcp_state_change client %p...\n", xprt);
1099         dprintk("RPC:       state %x conn %d dead %d zapped %d\n",
1100                         sk->sk_state, xprt_connected(xprt),
1101                         sock_flag(sk, SOCK_DEAD),
1102                         sock_flag(sk, SOCK_ZAPPED));
1103
1104         switch (sk->sk_state) {
1105         case TCP_ESTABLISHED:
1106                 spin_lock_bh(&xprt->transport_lock);
1107                 if (!xprt_test_and_set_connected(xprt)) {
1108                         struct sock_xprt *transport = container_of(xprt,
1109                                         struct sock_xprt, xprt);
1110
1111                         /* Reset TCP record info */
1112                         transport->tcp_offset = 0;
1113                         transport->tcp_reclen = 0;
1114                         transport->tcp_copied = 0;
1115                         transport->tcp_flags =
1116                                 TCP_RCV_COPY_FRAGHDR | TCP_RCV_COPY_XID;
1117
1118                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1119                         xprt_wake_pending_tasks(xprt, 0);
1120                 }
1121                 spin_unlock_bh(&xprt->transport_lock);
1122                 break;
1123         case TCP_FIN_WAIT1:
1124                 /* The client initiated a shutdown of the socket */
1125                 set_bit(XPRT_CLOSING, &xprt->state);
1126                 smp_mb__before_clear_bit();
1127                 clear_bit(XPRT_CONNECTED, &xprt->state);
1128                 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1129                 smp_mb__after_clear_bit();
1130                 break;
1131         case TCP_CLOSE_WAIT:
1132                 /* The server initiated a shutdown of the socket */
1133                 set_bit(XPRT_CLOSING, &xprt->state);
1134                 xprt_force_disconnect(xprt);
1135                 break;
1136         case TCP_LAST_ACK:
1137                 smp_mb__before_clear_bit();
1138                 clear_bit(XPRT_CONNECTED, &xprt->state);
1139                 smp_mb__after_clear_bit();
1140                 break;
1141         case TCP_CLOSE:
1142                 smp_mb__before_clear_bit();
1143                 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1144                 clear_bit(XPRT_CLOSING, &xprt->state);
1145                 smp_mb__after_clear_bit();
1146                 /* Mark transport as closed and wake up all pending tasks */
1147                 xprt_disconnect(xprt);
1148         }
1149  out:
1150         read_unlock(&sk->sk_callback_lock);
1151 }
1152
1153 /**
1154  * xs_udp_write_space - callback invoked when socket buffer space
1155  *                             becomes available
1156  * @sk: socket whose state has changed
1157  *
1158  * Called when more output buffer space is available for this socket.
1159  * We try not to wake our writers until they can make "significant"
1160  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1161  * with a bunch of small requests.
1162  */
1163 static void xs_udp_write_space(struct sock *sk)
1164 {
1165         read_lock(&sk->sk_callback_lock);
1166
1167         /* from net/core/sock.c:sock_def_write_space */
1168         if (sock_writeable(sk)) {
1169                 struct socket *sock;
1170                 struct rpc_xprt *xprt;
1171
1172                 if (unlikely(!(sock = sk->sk_socket)))
1173                         goto out;
1174                 if (unlikely(!(xprt = xprt_from_sock(sk))))
1175                         goto out;
1176                 if (unlikely(!test_and_clear_bit(SOCK_NOSPACE, &sock->flags)))
1177                         goto out;
1178
1179                 xprt_write_space(xprt);
1180         }
1181
1182  out:
1183         read_unlock(&sk->sk_callback_lock);
1184 }
1185
1186 /**
1187  * xs_tcp_write_space - callback invoked when socket buffer space
1188  *                             becomes available
1189  * @sk: socket whose state has changed
1190  *
1191  * Called when more output buffer space is available for this socket.
1192  * We try not to wake our writers until they can make "significant"
1193  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1194  * with a bunch of small requests.
1195  */
1196 static void xs_tcp_write_space(struct sock *sk)
1197 {
1198         read_lock(&sk->sk_callback_lock);
1199
1200         /* from net/core/stream.c:sk_stream_write_space */
1201         if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk)) {
1202                 struct socket *sock;
1203                 struct rpc_xprt *xprt;
1204
1205                 if (unlikely(!(sock = sk->sk_socket)))
1206                         goto out;
1207                 if (unlikely(!(xprt = xprt_from_sock(sk))))
1208                         goto out;
1209                 if (unlikely(!test_and_clear_bit(SOCK_NOSPACE, &sock->flags)))
1210                         goto out;
1211
1212                 xprt_write_space(xprt);
1213         }
1214
1215  out:
1216         read_unlock(&sk->sk_callback_lock);
1217 }
1218
1219 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
1220 {
1221         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1222         struct sock *sk = transport->inet;
1223
1224         if (transport->rcvsize) {
1225                 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1226                 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
1227         }
1228         if (transport->sndsize) {
1229                 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1230                 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1231                 sk->sk_write_space(sk);
1232         }
1233 }
1234
1235 /**
1236  * xs_udp_set_buffer_size - set send and receive limits
1237  * @xprt: generic transport
1238  * @sndsize: requested size of send buffer, in bytes
1239  * @rcvsize: requested size of receive buffer, in bytes
1240  *
1241  * Set socket send and receive buffer size limits.
1242  */
1243 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1244 {
1245         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1246
1247         transport->sndsize = 0;
1248         if (sndsize)
1249                 transport->sndsize = sndsize + 1024;
1250         transport->rcvsize = 0;
1251         if (rcvsize)
1252                 transport->rcvsize = rcvsize + 1024;
1253
1254         xs_udp_do_set_buffer_size(xprt);
1255 }
1256
1257 /**
1258  * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1259  * @task: task that timed out
1260  *
1261  * Adjust the congestion window after a retransmit timeout has occurred.
1262  */
1263 static void xs_udp_timer(struct rpc_task *task)
1264 {
1265         xprt_adjust_cwnd(task, -ETIMEDOUT);
1266 }
1267
1268 static unsigned short xs_get_random_port(void)
1269 {
1270         unsigned short range = xprt_max_resvport - xprt_min_resvport;
1271         unsigned short rand = (unsigned short) net_random() % range;
1272         return rand + xprt_min_resvport;
1273 }
1274
1275 /**
1276  * xs_set_port - reset the port number in the remote endpoint address
1277  * @xprt: generic transport
1278  * @port: new port number
1279  *
1280  */
1281 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1282 {
1283         struct sockaddr *addr = xs_addr(xprt);
1284
1285         dprintk("RPC:       setting port for xprt %p to %u\n", xprt, port);
1286
1287         switch (addr->sa_family) {
1288         case AF_INET:
1289                 ((struct sockaddr_in *)addr)->sin_port = htons(port);
1290                 break;
1291         case AF_INET6:
1292                 ((struct sockaddr_in6 *)addr)->sin6_port = htons(port);
1293                 break;
1294         default:
1295                 BUG();
1296         }
1297 }
1298
1299 static unsigned short xs_get_srcport(struct sock_xprt *transport, struct socket *sock)
1300 {
1301         unsigned short port = transport->port;
1302
1303         if (port == 0 && transport->xprt.resvport)
1304                 port = xs_get_random_port();
1305         return port;
1306 }
1307
1308 static unsigned short xs_next_srcport(struct sock_xprt *transport, struct socket *sock, unsigned short port)
1309 {
1310         if (transport->port != 0)
1311                 transport->port = 0;
1312         if (!transport->xprt.resvport)
1313                 return 0;
1314         if (port <= xprt_min_resvport || port > xprt_max_resvport)
1315                 return xprt_max_resvport;
1316         return --port;
1317 }
1318
1319 static int xs_bind4(struct sock_xprt *transport, struct socket *sock)
1320 {
1321         struct sockaddr_in myaddr = {
1322                 .sin_family = AF_INET,
1323         };
1324         struct sockaddr_in *sa;
1325         int err, nloop = 0;
1326         unsigned short port = xs_get_srcport(transport, sock);
1327         unsigned short last;
1328
1329         sa = (struct sockaddr_in *)&transport->addr;
1330         myaddr.sin_addr = sa->sin_addr;
1331         do {
1332                 myaddr.sin_port = htons(port);
1333                 err = kernel_bind(sock, (struct sockaddr *) &myaddr,
1334                                                 sizeof(myaddr));
1335                 if (port == 0)
1336                         break;
1337                 if (err == 0) {
1338                         transport->port = port;
1339                         break;
1340                 }
1341                 last = port;
1342                 port = xs_next_srcport(transport, sock, port);
1343                 if (port > last)
1344                         nloop++;
1345         } while (err == -EADDRINUSE && nloop != 2);
1346         dprintk("RPC:       %s "NIPQUAD_FMT":%u: %s (%d)\n",
1347                         __FUNCTION__, NIPQUAD(myaddr.sin_addr),
1348                         port, err ? "failed" : "ok", err);
1349         return err;
1350 }
1351
1352 static int xs_bind6(struct sock_xprt *transport, struct socket *sock)
1353 {
1354         struct sockaddr_in6 myaddr = {
1355                 .sin6_family = AF_INET6,
1356         };
1357         struct sockaddr_in6 *sa;
1358         int err, nloop = 0;
1359         unsigned short port = xs_get_srcport(transport, sock);
1360         unsigned short last;
1361
1362         sa = (struct sockaddr_in6 *)&transport->addr;
1363         myaddr.sin6_addr = sa->sin6_addr;
1364         do {
1365                 myaddr.sin6_port = htons(port);
1366                 err = kernel_bind(sock, (struct sockaddr *) &myaddr,
1367                                                 sizeof(myaddr));
1368                 if (port == 0)
1369                         break;
1370                 if (err == 0) {
1371                         transport->port = port;
1372                         break;
1373                 }
1374                 last = port;
1375                 port = xs_next_srcport(transport, sock, port);
1376                 if (port > last)
1377                         nloop++;
1378         } while (err == -EADDRINUSE && nloop != 2);
1379         dprintk("RPC:       xs_bind6 "NIP6_FMT":%u: %s (%d)\n",
1380                 NIP6(myaddr.sin6_addr), port, err ? "failed" : "ok", err);
1381         return err;
1382 }
1383
1384 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1385 static struct lock_class_key xs_key[2];
1386 static struct lock_class_key xs_slock_key[2];
1387
1388 static inline void xs_reclassify_socket4(struct socket *sock)
1389 {
1390         struct sock *sk = sock->sk;
1391
1392         BUG_ON(sock_owned_by_user(sk));
1393         sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1394                 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1395 }
1396
1397 static inline void xs_reclassify_socket6(struct socket *sock)
1398 {
1399         struct sock *sk = sock->sk;
1400
1401         BUG_ON(sock_owned_by_user(sk));
1402         sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1403                 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
1404 }
1405 #else
1406 static inline void xs_reclassify_socket4(struct socket *sock)
1407 {
1408 }
1409
1410 static inline void xs_reclassify_socket6(struct socket *sock)
1411 {
1412 }
1413 #endif
1414
1415 static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
1416 {
1417         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1418
1419         if (!transport->inet) {
1420                 struct sock *sk = sock->sk;
1421
1422                 write_lock_bh(&sk->sk_callback_lock);
1423
1424                 sk->sk_user_data = xprt;
1425                 transport->old_data_ready = sk->sk_data_ready;
1426                 transport->old_state_change = sk->sk_state_change;
1427                 transport->old_write_space = sk->sk_write_space;
1428                 sk->sk_data_ready = xs_udp_data_ready;
1429                 sk->sk_write_space = xs_udp_write_space;
1430                 sk->sk_no_check = UDP_CSUM_NORCV;
1431                 sk->sk_allocation = GFP_ATOMIC;
1432
1433                 xprt_set_connected(xprt);
1434
1435                 /* Reset to new socket */
1436                 transport->sock = sock;
1437                 transport->inet = sk;
1438
1439                 write_unlock_bh(&sk->sk_callback_lock);
1440         }
1441         xs_udp_do_set_buffer_size(xprt);
1442 }
1443
1444 /**
1445  * xs_udp_connect_worker4 - set up a UDP socket
1446  * @work: RPC transport to connect
1447  *
1448  * Invoked by a work queue tasklet.
1449  */
1450 static void xs_udp_connect_worker4(struct work_struct *work)
1451 {
1452         struct sock_xprt *transport =
1453                 container_of(work, struct sock_xprt, connect_worker.work);
1454         struct rpc_xprt *xprt = &transport->xprt;
1455         struct socket *sock = transport->sock;
1456         int err, status = -EIO;
1457
1458         if (xprt->shutdown || !xprt_bound(xprt))
1459                 goto out;
1460
1461         /* Start by resetting any existing state */
1462         xs_close(xprt);
1463
1464         if ((err = sock_create_kern(PF_INET, SOCK_DGRAM, IPPROTO_UDP, &sock)) < 0) {
1465                 dprintk("RPC:       can't create UDP transport socket (%d).\n", -err);
1466                 goto out;
1467         }
1468         xs_reclassify_socket4(sock);
1469
1470         if (xs_bind4(transport, sock)) {
1471                 sock_release(sock);
1472                 goto out;
1473         }
1474
1475         dprintk("RPC:       worker connecting xprt %p to address: %s\n",
1476                         xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1477
1478         xs_udp_finish_connecting(xprt, sock);
1479         status = 0;
1480 out:
1481         xprt_wake_pending_tasks(xprt, status);
1482         xprt_clear_connecting(xprt);
1483 }
1484
1485 /**
1486  * xs_udp_connect_worker6 - set up a UDP socket
1487  * @work: RPC transport to connect
1488  *
1489  * Invoked by a work queue tasklet.
1490  */
1491 static void xs_udp_connect_worker6(struct work_struct *work)
1492 {
1493         struct sock_xprt *transport =
1494                 container_of(work, struct sock_xprt, connect_worker.work);
1495         struct rpc_xprt *xprt = &transport->xprt;
1496         struct socket *sock = transport->sock;
1497         int err, status = -EIO;
1498
1499         if (xprt->shutdown || !xprt_bound(xprt))
1500                 goto out;
1501
1502         /* Start by resetting any existing state */
1503         xs_close(xprt);
1504
1505         if ((err = sock_create_kern(PF_INET6, SOCK_DGRAM, IPPROTO_UDP, &sock)) < 0) {
1506                 dprintk("RPC:       can't create UDP transport socket (%d).\n", -err);
1507                 goto out;
1508         }
1509         xs_reclassify_socket6(sock);
1510
1511         if (xs_bind6(transport, sock) < 0) {
1512                 sock_release(sock);
1513                 goto out;
1514         }
1515
1516         dprintk("RPC:       worker connecting xprt %p to address: %s\n",
1517                         xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1518
1519         xs_udp_finish_connecting(xprt, sock);
1520         status = 0;
1521 out:
1522         xprt_wake_pending_tasks(xprt, status);
1523         xprt_clear_connecting(xprt);
1524 }
1525
1526 /*
1527  * We need to preserve the port number so the reply cache on the server can
1528  * find our cached RPC replies when we get around to reconnecting.
1529  */
1530 static void xs_tcp_reuse_connection(struct rpc_xprt *xprt)
1531 {
1532         int result;
1533         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1534         struct sockaddr any;
1535
1536         dprintk("RPC:       disconnecting xprt %p to reuse port\n", xprt);
1537
1538         /*
1539          * Disconnect the transport socket by doing a connect operation
1540          * with AF_UNSPEC.  This should return immediately...
1541          */
1542         memset(&any, 0, sizeof(any));
1543         any.sa_family = AF_UNSPEC;
1544         result = kernel_connect(transport->sock, &any, sizeof(any), 0);
1545         if (result)
1546                 dprintk("RPC:       AF_UNSPEC connect return code %d\n",
1547                                 result);
1548 }
1549
1550 static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
1551 {
1552         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1553
1554         if (!transport->inet) {
1555                 struct sock *sk = sock->sk;
1556
1557                 write_lock_bh(&sk->sk_callback_lock);
1558
1559                 sk->sk_user_data = xprt;
1560                 transport->old_data_ready = sk->sk_data_ready;
1561                 transport->old_state_change = sk->sk_state_change;
1562                 transport->old_write_space = sk->sk_write_space;
1563                 sk->sk_data_ready = xs_tcp_data_ready;
1564                 sk->sk_state_change = xs_tcp_state_change;
1565                 sk->sk_write_space = xs_tcp_write_space;
1566                 sk->sk_allocation = GFP_ATOMIC;
1567
1568                 /* socket options */
1569                 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
1570                 sock_reset_flag(sk, SOCK_LINGER);
1571                 tcp_sk(sk)->linger2 = 0;
1572                 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
1573
1574                 xprt_clear_connected(xprt);
1575
1576                 /* Reset to new socket */
1577                 transport->sock = sock;
1578                 transport->inet = sk;
1579
1580                 write_unlock_bh(&sk->sk_callback_lock);
1581         }
1582
1583         /* Tell the socket layer to start connecting... */
1584         xprt->stat.connect_count++;
1585         xprt->stat.connect_start = jiffies;
1586         return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
1587 }
1588
1589 /**
1590  * xs_tcp_connect_worker4 - connect a TCP socket to a remote endpoint
1591  * @work: RPC transport to connect
1592  *
1593  * Invoked by a work queue tasklet.
1594  */
1595 static void xs_tcp_connect_worker4(struct work_struct *work)
1596 {
1597         struct sock_xprt *transport =
1598                 container_of(work, struct sock_xprt, connect_worker.work);
1599         struct rpc_xprt *xprt = &transport->xprt;
1600         struct socket *sock = transport->sock;
1601         int err, status = -EIO;
1602
1603         if (xprt->shutdown || !xprt_bound(xprt))
1604                 goto out;
1605
1606         if (!sock) {
1607                 /* start from scratch */
1608                 if ((err = sock_create_kern(PF_INET, SOCK_STREAM, IPPROTO_TCP, &sock)) < 0) {
1609                         dprintk("RPC:       can't create TCP transport socket (%d).\n", -err);
1610                         goto out;
1611                 }
1612                 xs_reclassify_socket4(sock);
1613
1614                 if (xs_bind4(transport, sock) < 0) {
1615                         sock_release(sock);
1616                         goto out;
1617                 }
1618         } else
1619                 /* "close" the socket, preserving the local port */
1620                 xs_tcp_reuse_connection(xprt);
1621
1622         dprintk("RPC:       worker connecting xprt %p to address: %s\n",
1623                         xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1624
1625         status = xs_tcp_finish_connecting(xprt, sock);
1626         dprintk("RPC:       %p connect status %d connected %d sock state %d\n",
1627                         xprt, -status, xprt_connected(xprt),
1628                         sock->sk->sk_state);
1629         if (status < 0) {
1630                 switch (status) {
1631                         case -EINPROGRESS:
1632                         case -EALREADY:
1633                                 goto out_clear;
1634                         case -ECONNREFUSED:
1635                         case -ECONNRESET:
1636                                 /* retry with existing socket, after a delay */
1637                                 break;
1638                         default:
1639                                 /* get rid of existing socket, and retry */
1640                                 xs_close(xprt);
1641                                 break;
1642                 }
1643         }
1644 out:
1645         xprt_wake_pending_tasks(xprt, status);
1646 out_clear:
1647         xprt_clear_connecting(xprt);
1648 }
1649
1650 /**
1651  * xs_tcp_connect_worker6 - connect a TCP socket to a remote endpoint
1652  * @work: RPC transport to connect
1653  *
1654  * Invoked by a work queue tasklet.
1655  */
1656 static void xs_tcp_connect_worker6(struct work_struct *work)
1657 {
1658         struct sock_xprt *transport =
1659                 container_of(work, struct sock_xprt, connect_worker.work);
1660         struct rpc_xprt *xprt = &transport->xprt;
1661         struct socket *sock = transport->sock;
1662         int err, status = -EIO;
1663
1664         if (xprt->shutdown || !xprt_bound(xprt))
1665                 goto out;
1666
1667         if (!sock) {
1668                 /* start from scratch */
1669                 if ((err = sock_create_kern(PF_INET6, SOCK_STREAM, IPPROTO_TCP, &sock)) < 0) {
1670                         dprintk("RPC:       can't create TCP transport socket (%d).\n", -err);
1671                         goto out;
1672                 }
1673                 xs_reclassify_socket6(sock);
1674
1675                 if (xs_bind6(transport, sock) < 0) {
1676                         sock_release(sock);
1677                         goto out;
1678                 }
1679         } else
1680                 /* "close" the socket, preserving the local port */
1681                 xs_tcp_reuse_connection(xprt);
1682
1683         dprintk("RPC:       worker connecting xprt %p to address: %s\n",
1684                         xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1685
1686         status = xs_tcp_finish_connecting(xprt, sock);
1687         dprintk("RPC:       %p connect status %d connected %d sock state %d\n",
1688                         xprt, -status, xprt_connected(xprt), sock->sk->sk_state);
1689         if (status < 0) {
1690                 switch (status) {
1691                         case -EINPROGRESS:
1692                         case -EALREADY:
1693                                 goto out_clear;
1694                         case -ECONNREFUSED:
1695                         case -ECONNRESET:
1696                                 /* retry with existing socket, after a delay */
1697                                 break;
1698                         default:
1699                                 /* get rid of existing socket, and retry */
1700                                 xs_close(xprt);
1701                                 break;
1702                 }
1703         }
1704 out:
1705         xprt_wake_pending_tasks(xprt, status);
1706 out_clear:
1707         xprt_clear_connecting(xprt);
1708 }
1709
1710 /**
1711  * xs_connect - connect a socket to a remote endpoint
1712  * @task: address of RPC task that manages state of connect request
1713  *
1714  * TCP: If the remote end dropped the connection, delay reconnecting.
1715  *
1716  * UDP socket connects are synchronous, but we use a work queue anyway
1717  * to guarantee that even unprivileged user processes can set up a
1718  * socket on a privileged port.
1719  *
1720  * If a UDP socket connect fails, the delay behavior here prevents
1721  * retry floods (hard mounts).
1722  */
1723 static void xs_connect(struct rpc_task *task)
1724 {
1725         struct rpc_xprt *xprt = task->tk_xprt;
1726         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1727
1728         if (xprt_test_and_set_connecting(xprt))
1729                 return;
1730
1731         if (transport->sock != NULL) {
1732                 dprintk("RPC:       xs_connect delayed xprt %p for %lu "
1733                                 "seconds\n",
1734                                 xprt, xprt->reestablish_timeout / HZ);
1735                 queue_delayed_work(rpciod_workqueue,
1736                                    &transport->connect_worker,
1737                                    xprt->reestablish_timeout);
1738                 xprt->reestablish_timeout <<= 1;
1739                 if (xprt->reestablish_timeout > XS_TCP_MAX_REEST_TO)
1740                         xprt->reestablish_timeout = XS_TCP_MAX_REEST_TO;
1741         } else {
1742                 dprintk("RPC:       xs_connect scheduled xprt %p\n", xprt);
1743                 queue_delayed_work(rpciod_workqueue,
1744                                    &transport->connect_worker, 0);
1745         }
1746 }
1747
1748 /**
1749  * xs_udp_print_stats - display UDP socket-specifc stats
1750  * @xprt: rpc_xprt struct containing statistics
1751  * @seq: output file
1752  *
1753  */
1754 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
1755 {
1756         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1757
1758         seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %Lu %Lu\n",
1759                         transport->port,
1760                         xprt->stat.bind_count,
1761                         xprt->stat.sends,
1762                         xprt->stat.recvs,
1763                         xprt->stat.bad_xids,
1764                         xprt->stat.req_u,
1765                         xprt->stat.bklog_u);
1766 }
1767
1768 /**
1769  * xs_tcp_print_stats - display TCP socket-specifc stats
1770  * @xprt: rpc_xprt struct containing statistics
1771  * @seq: output file
1772  *
1773  */
1774 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
1775 {
1776         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1777         long idle_time = 0;
1778
1779         if (xprt_connected(xprt))
1780                 idle_time = (long)(jiffies - xprt->last_used) / HZ;
1781
1782         seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu %Lu %Lu\n",
1783                         transport->port,
1784                         xprt->stat.bind_count,
1785                         xprt->stat.connect_count,
1786                         xprt->stat.connect_time,
1787                         idle_time,
1788                         xprt->stat.sends,
1789                         xprt->stat.recvs,
1790                         xprt->stat.bad_xids,
1791                         xprt->stat.req_u,
1792                         xprt->stat.bklog_u);
1793 }
1794
1795 static struct rpc_xprt_ops xs_udp_ops = {
1796         .set_buffer_size        = xs_udp_set_buffer_size,
1797         .reserve_xprt           = xprt_reserve_xprt_cong,
1798         .release_xprt           = xprt_release_xprt_cong,
1799         .rpcbind                = rpcb_getport_async,
1800         .set_port               = xs_set_port,
1801         .connect                = xs_connect,
1802         .buf_alloc              = rpc_malloc,
1803         .buf_free               = rpc_free,
1804         .send_request           = xs_udp_send_request,
1805         .set_retrans_timeout    = xprt_set_retrans_timeout_rtt,
1806         .timer                  = xs_udp_timer,
1807         .release_request        = xprt_release_rqst_cong,
1808         .close                  = xs_close,
1809         .destroy                = xs_destroy,
1810         .print_stats            = xs_udp_print_stats,
1811 };
1812
1813 static struct rpc_xprt_ops xs_tcp_ops = {
1814         .reserve_xprt           = xprt_reserve_xprt,
1815         .release_xprt           = xs_tcp_release_xprt,
1816         .rpcbind                = rpcb_getport_async,
1817         .set_port               = xs_set_port,
1818         .connect                = xs_connect,
1819         .buf_alloc              = rpc_malloc,
1820         .buf_free               = rpc_free,
1821         .send_request           = xs_tcp_send_request,
1822         .set_retrans_timeout    = xprt_set_retrans_timeout_def,
1823         .close                  = xs_close,
1824         .destroy                = xs_destroy,
1825         .print_stats            = xs_tcp_print_stats,
1826 };
1827
1828 static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
1829                                       unsigned int slot_table_size)
1830 {
1831         struct rpc_xprt *xprt;
1832         struct sock_xprt *new;
1833
1834         if (args->addrlen > sizeof(xprt->addr)) {
1835                 dprintk("RPC:       xs_setup_xprt: address too large\n");
1836                 return ERR_PTR(-EBADF);
1837         }
1838
1839         new = kzalloc(sizeof(*new), GFP_KERNEL);
1840         if (new == NULL) {
1841                 dprintk("RPC:       xs_setup_xprt: couldn't allocate "
1842                                 "rpc_xprt\n");
1843                 return ERR_PTR(-ENOMEM);
1844         }
1845         xprt = &new->xprt;
1846
1847         xprt->max_reqs = slot_table_size;
1848         xprt->slot = kcalloc(xprt->max_reqs, sizeof(struct rpc_rqst), GFP_KERNEL);
1849         if (xprt->slot == NULL) {
1850                 kfree(xprt);
1851                 dprintk("RPC:       xs_setup_xprt: couldn't allocate slot "
1852                                 "table\n");
1853                 return ERR_PTR(-ENOMEM);
1854         }
1855
1856         memcpy(&xprt->addr, args->dstaddr, args->addrlen);
1857         xprt->addrlen = args->addrlen;
1858         if (args->srcaddr)
1859                 memcpy(&new->addr, args->srcaddr, args->addrlen);
1860
1861         return xprt;
1862 }
1863
1864 /**
1865  * xs_setup_udp - Set up transport to use a UDP socket
1866  * @args: rpc transport creation arguments
1867  *
1868  */
1869 static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
1870 {
1871         struct sockaddr *addr = args->dstaddr;
1872         struct rpc_xprt *xprt;
1873         struct sock_xprt *transport;
1874
1875         xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries);
1876         if (IS_ERR(xprt))
1877                 return xprt;
1878         transport = container_of(xprt, struct sock_xprt, xprt);
1879
1880         xprt->prot = IPPROTO_UDP;
1881         xprt->tsh_size = 0;
1882         /* XXX: header size can vary due to auth type, IPv6, etc. */
1883         xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
1884
1885         xprt->bind_timeout = XS_BIND_TO;
1886         xprt->connect_timeout = XS_UDP_CONN_TO;
1887         xprt->reestablish_timeout = XS_UDP_REEST_TO;
1888         xprt->idle_timeout = XS_IDLE_DISC_TO;
1889
1890         xprt->ops = &xs_udp_ops;
1891
1892         if (args->timeout)
1893                 xprt->timeout = *args->timeout;
1894         else
1895                 xprt_set_timeout(&xprt->timeout, 5, 5 * HZ);
1896
1897         switch (addr->sa_family) {
1898         case AF_INET:
1899                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
1900                         xprt_set_bound(xprt);
1901
1902                 INIT_DELAYED_WORK(&transport->connect_worker,
1903                                         xs_udp_connect_worker4);
1904                 xs_format_ipv4_peer_addresses(xprt);
1905                 break;
1906         case AF_INET6:
1907                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
1908                         xprt_set_bound(xprt);
1909
1910                 INIT_DELAYED_WORK(&transport->connect_worker,
1911                                         xs_udp_connect_worker6);
1912                 xs_format_ipv6_peer_addresses(xprt);
1913                 break;
1914         default:
1915                 kfree(xprt);
1916                 return ERR_PTR(-EAFNOSUPPORT);
1917         }
1918
1919         dprintk("RPC:       set up transport to address %s\n",
1920                         xprt->address_strings[RPC_DISPLAY_ALL]);
1921
1922         if (try_module_get(THIS_MODULE))
1923                 return xprt;
1924
1925         kfree(xprt->slot);
1926         kfree(xprt);
1927         return ERR_PTR(-EINVAL);
1928 }
1929
1930 /**
1931  * xs_setup_tcp - Set up transport to use a TCP socket
1932  * @args: rpc transport creation arguments
1933  *
1934  */
1935 static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
1936 {
1937         struct sockaddr *addr = args->dstaddr;
1938         struct rpc_xprt *xprt;
1939         struct sock_xprt *transport;
1940
1941         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries);
1942         if (IS_ERR(xprt))
1943                 return xprt;
1944         transport = container_of(xprt, struct sock_xprt, xprt);
1945
1946         xprt->prot = IPPROTO_TCP;
1947         xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
1948         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
1949
1950         xprt->bind_timeout = XS_BIND_TO;
1951         xprt->connect_timeout = XS_TCP_CONN_TO;
1952         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1953         xprt->idle_timeout = XS_IDLE_DISC_TO;
1954
1955         xprt->ops = &xs_tcp_ops;
1956
1957         if (args->timeout)
1958                 xprt->timeout = *args->timeout;
1959         else
1960                 xprt_set_timeout(&xprt->timeout, 2, 60 * HZ);
1961
1962         switch (addr->sa_family) {
1963         case AF_INET:
1964                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
1965                         xprt_set_bound(xprt);
1966
1967                 INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_connect_worker4);
1968                 xs_format_ipv4_peer_addresses(xprt);
1969                 break;
1970         case AF_INET6:
1971                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
1972                         xprt_set_bound(xprt);
1973
1974                 INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_connect_worker6);
1975                 xs_format_ipv6_peer_addresses(xprt);
1976                 break;
1977         default:
1978                 kfree(xprt);
1979                 return ERR_PTR(-EAFNOSUPPORT);
1980         }
1981
1982         dprintk("RPC:       set up transport to address %s\n",
1983                         xprt->address_strings[RPC_DISPLAY_ALL]);
1984
1985         if (try_module_get(THIS_MODULE))
1986                 return xprt;
1987
1988         kfree(xprt->slot);
1989         kfree(xprt);
1990         return ERR_PTR(-EINVAL);
1991 }
1992
1993 static struct xprt_class        xs_udp_transport = {
1994         .list           = LIST_HEAD_INIT(xs_udp_transport.list),
1995         .name           = "udp",
1996         .owner          = THIS_MODULE,
1997         .ident          = IPPROTO_UDP,
1998         .setup          = xs_setup_udp,
1999 };
2000
2001 static struct xprt_class        xs_tcp_transport = {
2002         .list           = LIST_HEAD_INIT(xs_tcp_transport.list),
2003         .name           = "tcp",
2004         .owner          = THIS_MODULE,
2005         .ident          = IPPROTO_TCP,
2006         .setup          = xs_setup_tcp,
2007 };
2008
2009 /**
2010  * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
2011  *
2012  */
2013 int init_socket_xprt(void)
2014 {
2015 #ifdef RPC_DEBUG
2016         if (!sunrpc_table_header)
2017                 sunrpc_table_header = register_sysctl_table(sunrpc_table);
2018 #endif
2019
2020         xprt_register_transport(&xs_udp_transport);
2021         xprt_register_transport(&xs_tcp_transport);
2022
2023         return 0;
2024 }
2025
2026 /**
2027  * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
2028  *
2029  */
2030 void cleanup_socket_xprt(void)
2031 {
2032 #ifdef RPC_DEBUG
2033         if (sunrpc_table_header) {
2034                 unregister_sysctl_table(sunrpc_table_header);
2035                 sunrpc_table_header = NULL;
2036         }
2037 #endif
2038
2039         xprt_unregister_transport(&xs_udp_transport);
2040         xprt_unregister_transport(&xs_tcp_transport);
2041 }