SUNRPC: fix print format for tk_pid
[linux-2.6] / 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
18 #include <linux/types.h>
19 #include <linux/slab.h>
20 #include <linux/capability.h>
21 #include <linux/sched.h>
22 #include <linux/pagemap.h>
23 #include <linux/errno.h>
24 #include <linux/socket.h>
25 #include <linux/in.h>
26 #include <linux/net.h>
27 #include <linux/mm.h>
28 #include <linux/udp.h>
29 #include <linux/tcp.h>
30 #include <linux/sunrpc/clnt.h>
31 #include <linux/sunrpc/sched.h>
32 #include <linux/file.h>
33
34 #include <net/sock.h>
35 #include <net/checksum.h>
36 #include <net/udp.h>
37 #include <net/tcp.h>
38
39 /*
40  * xprtsock tunables
41  */
42 unsigned int xprt_udp_slot_table_entries = RPC_DEF_SLOT_TABLE;
43 unsigned int xprt_tcp_slot_table_entries = RPC_DEF_SLOT_TABLE;
44
45 unsigned int xprt_min_resvport = RPC_DEF_MIN_RESVPORT;
46 unsigned int xprt_max_resvport = RPC_DEF_MAX_RESVPORT;
47
48 /*
49  * We can register our own files under /proc/sys/sunrpc by
50  * calling register_sysctl_table() again.  The files in that
51  * directory become the union of all files registered there.
52  *
53  * We simply need to make sure that we don't collide with
54  * someone else's file names!
55  */
56
57 #ifdef RPC_DEBUG
58
59 static unsigned int min_slot_table_size = RPC_MIN_SLOT_TABLE;
60 static unsigned int max_slot_table_size = RPC_MAX_SLOT_TABLE;
61 static unsigned int xprt_min_resvport_limit = RPC_MIN_RESVPORT;
62 static unsigned int xprt_max_resvport_limit = RPC_MAX_RESVPORT;
63
64 static struct ctl_table_header *sunrpc_table_header;
65
66 /*
67  * FIXME: changing the UDP slot table size should also resize the UDP
68  *        socket buffers for existing UDP transports
69  */
70 static ctl_table xs_tunables_table[] = {
71         {
72                 .ctl_name       = CTL_SLOTTABLE_UDP,
73                 .procname       = "udp_slot_table_entries",
74                 .data           = &xprt_udp_slot_table_entries,
75                 .maxlen         = sizeof(unsigned int),
76                 .mode           = 0644,
77                 .proc_handler   = &proc_dointvec_minmax,
78                 .strategy       = &sysctl_intvec,
79                 .extra1         = &min_slot_table_size,
80                 .extra2         = &max_slot_table_size
81         },
82         {
83                 .ctl_name       = CTL_SLOTTABLE_TCP,
84                 .procname       = "tcp_slot_table_entries",
85                 .data           = &xprt_tcp_slot_table_entries,
86                 .maxlen         = sizeof(unsigned int),
87                 .mode           = 0644,
88                 .proc_handler   = &proc_dointvec_minmax,
89                 .strategy       = &sysctl_intvec,
90                 .extra1         = &min_slot_table_size,
91                 .extra2         = &max_slot_table_size
92         },
93         {
94                 .ctl_name       = CTL_MIN_RESVPORT,
95                 .procname       = "min_resvport",
96                 .data           = &xprt_min_resvport,
97                 .maxlen         = sizeof(unsigned int),
98                 .mode           = 0644,
99                 .proc_handler   = &proc_dointvec_minmax,
100                 .strategy       = &sysctl_intvec,
101                 .extra1         = &xprt_min_resvport_limit,
102                 .extra2         = &xprt_max_resvport_limit
103         },
104         {
105                 .ctl_name       = CTL_MAX_RESVPORT,
106                 .procname       = "max_resvport",
107                 .data           = &xprt_max_resvport,
108                 .maxlen         = sizeof(unsigned int),
109                 .mode           = 0644,
110                 .proc_handler   = &proc_dointvec_minmax,
111                 .strategy       = &sysctl_intvec,
112                 .extra1         = &xprt_min_resvport_limit,
113                 .extra2         = &xprt_max_resvport_limit
114         },
115         {
116                 .ctl_name = 0,
117         },
118 };
119
120 static ctl_table sunrpc_table[] = {
121         {
122                 .ctl_name       = CTL_SUNRPC,
123                 .procname       = "sunrpc",
124                 .mode           = 0555,
125                 .child          = xs_tunables_table
126         },
127         {
128                 .ctl_name = 0,
129         },
130 };
131
132 #endif
133
134 /*
135  * How many times to try sending a request on a socket before waiting
136  * for the socket buffer to clear.
137  */
138 #define XS_SENDMSG_RETRY        (10U)
139
140 /*
141  * Time out for an RPC UDP socket connect.  UDP socket connects are
142  * synchronous, but we set a timeout anyway in case of resource
143  * exhaustion on the local host.
144  */
145 #define XS_UDP_CONN_TO          (5U * HZ)
146
147 /*
148  * Wait duration for an RPC TCP connection to be established.  Solaris
149  * NFS over TCP uses 60 seconds, for example, which is in line with how
150  * long a server takes to reboot.
151  */
152 #define XS_TCP_CONN_TO          (60U * HZ)
153
154 /*
155  * Wait duration for a reply from the RPC portmapper.
156  */
157 #define XS_BIND_TO              (60U * HZ)
158
159 /*
160  * Delay if a UDP socket connect error occurs.  This is most likely some
161  * kind of resource problem on the local host.
162  */
163 #define XS_UDP_REEST_TO         (2U * HZ)
164
165 /*
166  * The reestablish timeout allows clients to delay for a bit before attempting
167  * to reconnect to a server that just dropped our connection.
168  *
169  * We implement an exponential backoff when trying to reestablish a TCP
170  * transport connection with the server.  Some servers like to drop a TCP
171  * connection when they are overworked, so we start with a short timeout and
172  * increase over time if the server is down or not responding.
173  */
174 #define XS_TCP_INIT_REEST_TO    (3U * HZ)
175 #define XS_TCP_MAX_REEST_TO     (5U * 60 * HZ)
176
177 /*
178  * TCP idle timeout; client drops the transport socket if it is idle
179  * for this long.  Note that we also timeout UDP sockets to prevent
180  * holding port numbers when there is no RPC traffic.
181  */
182 #define XS_IDLE_DISC_TO         (5U * 60 * HZ)
183
184 #ifdef RPC_DEBUG
185 # undef  RPC_DEBUG_DATA
186 # define RPCDBG_FACILITY        RPCDBG_TRANS
187 #endif
188
189 #ifdef RPC_DEBUG_DATA
190 static void xs_pktdump(char *msg, u32 *packet, unsigned int count)
191 {
192         u8 *buf = (u8 *) packet;
193         int j;
194
195         dprintk("RPC:       %s\n", msg);
196         for (j = 0; j < count && j < 128; j += 4) {
197                 if (!(j & 31)) {
198                         if (j)
199                                 dprintk("\n");
200                         dprintk("0x%04x ", j);
201                 }
202                 dprintk("%02x%02x%02x%02x ",
203                         buf[j], buf[j+1], buf[j+2], buf[j+3]);
204         }
205         dprintk("\n");
206 }
207 #else
208 static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count)
209 {
210         /* NOP */
211 }
212 #endif
213
214 struct sock_xprt {
215         struct rpc_xprt         xprt;
216
217         /*
218          * Network layer
219          */
220         struct socket *         sock;
221         struct sock *           inet;
222
223         /*
224          * State of TCP reply receive
225          */
226         __be32                  tcp_fraghdr,
227                                 tcp_xid;
228
229         u32                     tcp_offset,
230                                 tcp_reclen;
231
232         unsigned long           tcp_copied,
233                                 tcp_flags;
234
235         /*
236          * Connection of transports
237          */
238         struct delayed_work     connect_worker;
239         unsigned short          port;
240
241         /*
242          * UDP socket buffer size parameters
243          */
244         size_t                  rcvsize,
245                                 sndsize;
246
247         /*
248          * Saved socket callback addresses
249          */
250         void                    (*old_data_ready)(struct sock *, int);
251         void                    (*old_state_change)(struct sock *);
252         void                    (*old_write_space)(struct sock *);
253 };
254
255 /*
256  * TCP receive state flags
257  */
258 #define TCP_RCV_LAST_FRAG       (1UL << 0)
259 #define TCP_RCV_COPY_FRAGHDR    (1UL << 1)
260 #define TCP_RCV_COPY_XID        (1UL << 2)
261 #define TCP_RCV_COPY_DATA       (1UL << 3)
262
263 static void xs_format_peer_addresses(struct rpc_xprt *xprt)
264 {
265         struct sockaddr_in *addr = (struct sockaddr_in *) &xprt->addr;
266         char *buf;
267
268         buf = kzalloc(20, GFP_KERNEL);
269         if (buf) {
270                 snprintf(buf, 20, "%u.%u.%u.%u",
271                                 NIPQUAD(addr->sin_addr.s_addr));
272         }
273         xprt->address_strings[RPC_DISPLAY_ADDR] = buf;
274
275         buf = kzalloc(8, GFP_KERNEL);
276         if (buf) {
277                 snprintf(buf, 8, "%u",
278                                 ntohs(addr->sin_port));
279         }
280         xprt->address_strings[RPC_DISPLAY_PORT] = buf;
281
282         if (xprt->prot == IPPROTO_UDP)
283                 xprt->address_strings[RPC_DISPLAY_PROTO] = "udp";
284         else
285                 xprt->address_strings[RPC_DISPLAY_PROTO] = "tcp";
286
287         buf = kzalloc(48, GFP_KERNEL);
288         if (buf) {
289                 snprintf(buf, 48, "addr=%u.%u.%u.%u port=%u proto=%s",
290                         NIPQUAD(addr->sin_addr.s_addr),
291                         ntohs(addr->sin_port),
292                         xprt->prot == IPPROTO_UDP ? "udp" : "tcp");
293         }
294         xprt->address_strings[RPC_DISPLAY_ALL] = buf;
295 }
296
297 static void xs_free_peer_addresses(struct rpc_xprt *xprt)
298 {
299         kfree(xprt->address_strings[RPC_DISPLAY_ADDR]);
300         kfree(xprt->address_strings[RPC_DISPLAY_PORT]);
301         kfree(xprt->address_strings[RPC_DISPLAY_ALL]);
302 }
303
304 #define XS_SENDMSG_FLAGS        (MSG_DONTWAIT | MSG_NOSIGNAL)
305
306 static int xs_send_kvec(struct socket *sock, struct sockaddr *addr, int addrlen, struct kvec *vec, unsigned int base, int more)
307 {
308         struct msghdr msg = {
309                 .msg_name       = addr,
310                 .msg_namelen    = addrlen,
311                 .msg_flags      = XS_SENDMSG_FLAGS | (more ? MSG_MORE : 0),
312         };
313         struct kvec iov = {
314                 .iov_base       = vec->iov_base + base,
315                 .iov_len        = vec->iov_len - base,
316         };
317
318         if (iov.iov_len != 0)
319                 return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
320         return kernel_sendmsg(sock, &msg, NULL, 0, 0);
321 }
322
323 static int xs_send_pagedata(struct socket *sock, struct xdr_buf *xdr, unsigned int base, int more)
324 {
325         struct page **ppage;
326         unsigned int remainder;
327         int err, sent = 0;
328
329         remainder = xdr->page_len - base;
330         base += xdr->page_base;
331         ppage = xdr->pages + (base >> PAGE_SHIFT);
332         base &= ~PAGE_MASK;
333         for(;;) {
334                 unsigned int len = min_t(unsigned int, PAGE_SIZE - base, remainder);
335                 int flags = XS_SENDMSG_FLAGS;
336
337                 remainder -= len;
338                 if (remainder != 0 || more)
339                         flags |= MSG_MORE;
340                 err = sock->ops->sendpage(sock, *ppage, base, len, flags);
341                 if (remainder == 0 || err != len)
342                         break;
343                 sent += err;
344                 ppage++;
345                 base = 0;
346         }
347         if (sent == 0)
348                 return err;
349         if (err > 0)
350                 sent += err;
351         return sent;
352 }
353
354 /**
355  * xs_sendpages - write pages directly to a socket
356  * @sock: socket to send on
357  * @addr: UDP only -- address of destination
358  * @addrlen: UDP only -- length of destination address
359  * @xdr: buffer containing this request
360  * @base: starting position in the buffer
361  *
362  */
363 static int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base)
364 {
365         unsigned int remainder = xdr->len - base;
366         int err, sent = 0;
367
368         if (unlikely(!sock))
369                 return -ENOTCONN;
370
371         clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags);
372         if (base != 0) {
373                 addr = NULL;
374                 addrlen = 0;
375         }
376
377         if (base < xdr->head[0].iov_len || addr != NULL) {
378                 unsigned int len = xdr->head[0].iov_len - base;
379                 remainder -= len;
380                 err = xs_send_kvec(sock, addr, addrlen, &xdr->head[0], base, remainder != 0);
381                 if (remainder == 0 || err != len)
382                         goto out;
383                 sent += err;
384                 base = 0;
385         } else
386                 base -= xdr->head[0].iov_len;
387
388         if (base < xdr->page_len) {
389                 unsigned int len = xdr->page_len - base;
390                 remainder -= len;
391                 err = xs_send_pagedata(sock, xdr, base, remainder != 0);
392                 if (remainder == 0 || err != len)
393                         goto out;
394                 sent += err;
395                 base = 0;
396         } else
397                 base -= xdr->page_len;
398
399         if (base >= xdr->tail[0].iov_len)
400                 return sent;
401         err = xs_send_kvec(sock, NULL, 0, &xdr->tail[0], base, 0);
402 out:
403         if (sent == 0)
404                 return err;
405         if (err > 0)
406                 sent += err;
407         return sent;
408 }
409
410 /**
411  * xs_nospace - place task on wait queue if transmit was incomplete
412  * @task: task to put to sleep
413  *
414  */
415 static void xs_nospace(struct rpc_task *task)
416 {
417         struct rpc_rqst *req = task->tk_rqstp;
418         struct rpc_xprt *xprt = req->rq_xprt;
419         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
420
421         dprintk("RPC: %5u xmit incomplete (%u left of %u)\n",
422                         task->tk_pid, req->rq_slen - req->rq_bytes_sent,
423                         req->rq_slen);
424
425         if (test_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags)) {
426                 /* Protect against races with write_space */
427                 spin_lock_bh(&xprt->transport_lock);
428
429                 /* Don't race with disconnect */
430                 if (!xprt_connected(xprt))
431                         task->tk_status = -ENOTCONN;
432                 else if (test_bit(SOCK_NOSPACE, &transport->sock->flags))
433                         xprt_wait_for_buffer_space(task);
434
435                 spin_unlock_bh(&xprt->transport_lock);
436         } else
437                 /* Keep holding the socket if it is blocked */
438                 rpc_delay(task, HZ>>4);
439 }
440
441 /**
442  * xs_udp_send_request - write an RPC request to a UDP socket
443  * @task: address of RPC task that manages the state of an RPC request
444  *
445  * Return values:
446  *        0:    The request has been sent
447  *   EAGAIN:    The socket was blocked, please call again later to
448  *              complete the request
449  * ENOTCONN:    Caller needs to invoke connect logic then call again
450  *    other:    Some other error occured, the request was not sent
451  */
452 static int xs_udp_send_request(struct rpc_task *task)
453 {
454         struct rpc_rqst *req = task->tk_rqstp;
455         struct rpc_xprt *xprt = req->rq_xprt;
456         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
457         struct xdr_buf *xdr = &req->rq_snd_buf;
458         int status;
459
460         xs_pktdump("packet data:",
461                                 req->rq_svec->iov_base,
462                                 req->rq_svec->iov_len);
463
464         req->rq_xtime = jiffies;
465         status = xs_sendpages(transport->sock,
466                               (struct sockaddr *) &xprt->addr,
467                               xprt->addrlen, xdr,
468                               req->rq_bytes_sent);
469
470         dprintk("RPC:       xs_udp_send_request(%u) = %d\n",
471                         xdr->len - req->rq_bytes_sent, status);
472
473         if (likely(status >= (int) req->rq_slen))
474                 return 0;
475
476         /* Still some bytes left; set up for a retry later. */
477         if (status > 0)
478                 status = -EAGAIN;
479
480         switch (status) {
481         case -ENETUNREACH:
482         case -EPIPE:
483         case -ECONNREFUSED:
484                 /* When the server has died, an ICMP port unreachable message
485                  * prompts ECONNREFUSED. */
486                 break;
487         case -EAGAIN:
488                 xs_nospace(task);
489                 break;
490         default:
491                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
492                         -status);
493                 break;
494         }
495
496         return status;
497 }
498
499 static inline void xs_encode_tcp_record_marker(struct xdr_buf *buf)
500 {
501         u32 reclen = buf->len - sizeof(rpc_fraghdr);
502         rpc_fraghdr *base = buf->head[0].iov_base;
503         *base = htonl(RPC_LAST_STREAM_FRAGMENT | reclen);
504 }
505
506 /**
507  * xs_tcp_send_request - write an RPC request to a TCP socket
508  * @task: address of RPC task that manages the state of an RPC request
509  *
510  * Return values:
511  *        0:    The request has been sent
512  *   EAGAIN:    The socket was blocked, please call again later to
513  *              complete the request
514  * ENOTCONN:    Caller needs to invoke connect logic then call again
515  *    other:    Some other error occured, the request was not sent
516  *
517  * XXX: In the case of soft timeouts, should we eventually give up
518  *      if sendmsg is not able to make progress?
519  */
520 static int xs_tcp_send_request(struct rpc_task *task)
521 {
522         struct rpc_rqst *req = task->tk_rqstp;
523         struct rpc_xprt *xprt = req->rq_xprt;
524         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
525         struct xdr_buf *xdr = &req->rq_snd_buf;
526         int status, retry = 0;
527
528         xs_encode_tcp_record_marker(&req->rq_snd_buf);
529
530         xs_pktdump("packet data:",
531                                 req->rq_svec->iov_base,
532                                 req->rq_svec->iov_len);
533
534         /* Continue transmitting the packet/record. We must be careful
535          * to cope with writespace callbacks arriving _after_ we have
536          * called sendmsg(). */
537         while (1) {
538                 req->rq_xtime = jiffies;
539                 status = xs_sendpages(transport->sock,
540                                         NULL, 0, xdr, req->rq_bytes_sent);
541
542                 dprintk("RPC:       xs_tcp_send_request(%u) = %d\n",
543                                 xdr->len - req->rq_bytes_sent, status);
544
545                 if (unlikely(status < 0))
546                         break;
547
548                 /* If we've sent the entire packet, immediately
549                  * reset the count of bytes sent. */
550                 req->rq_bytes_sent += status;
551                 task->tk_bytes_sent += status;
552                 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
553                         req->rq_bytes_sent = 0;
554                         return 0;
555                 }
556
557                 status = -EAGAIN;
558                 if (retry++ > XS_SENDMSG_RETRY)
559                         break;
560         }
561
562         switch (status) {
563         case -EAGAIN:
564                 xs_nospace(task);
565                 break;
566         case -ECONNREFUSED:
567         case -ECONNRESET:
568         case -ENOTCONN:
569         case -EPIPE:
570                 status = -ENOTCONN;
571                 break;
572         default:
573                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
574                         -status);
575                 xprt_disconnect(xprt);
576                 break;
577         }
578
579         return status;
580 }
581
582 /**
583  * xs_tcp_release_xprt - clean up after a tcp transmission
584  * @xprt: transport
585  * @task: rpc task
586  *
587  * This cleans up if an error causes us to abort the transmission of a request.
588  * In this case, the socket may need to be reset in order to avoid confusing
589  * the server.
590  */
591 static void xs_tcp_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
592 {
593         struct rpc_rqst *req;
594
595         if (task != xprt->snd_task)
596                 return;
597         if (task == NULL)
598                 goto out_release;
599         req = task->tk_rqstp;
600         if (req->rq_bytes_sent == 0)
601                 goto out_release;
602         if (req->rq_bytes_sent == req->rq_snd_buf.len)
603                 goto out_release;
604         set_bit(XPRT_CLOSE_WAIT, &task->tk_xprt->state);
605 out_release:
606         xprt_release_xprt(xprt, task);
607 }
608
609 /**
610  * xs_close - close a socket
611  * @xprt: transport
612  *
613  * This is used when all requests are complete; ie, no DRC state remains
614  * on the server we want to save.
615  */
616 static void xs_close(struct rpc_xprt *xprt)
617 {
618         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
619         struct socket *sock = transport->sock;
620         struct sock *sk = transport->inet;
621
622         if (!sk)
623                 goto clear_close_wait;
624
625         dprintk("RPC:       xs_close xprt %p\n", xprt);
626
627         write_lock_bh(&sk->sk_callback_lock);
628         transport->inet = NULL;
629         transport->sock = NULL;
630
631         sk->sk_user_data = NULL;
632         sk->sk_data_ready = transport->old_data_ready;
633         sk->sk_state_change = transport->old_state_change;
634         sk->sk_write_space = transport->old_write_space;
635         write_unlock_bh(&sk->sk_callback_lock);
636
637         sk->sk_no_check = 0;
638
639         sock_release(sock);
640 clear_close_wait:
641         smp_mb__before_clear_bit();
642         clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
643         smp_mb__after_clear_bit();
644 }
645
646 /**
647  * xs_destroy - prepare to shutdown a transport
648  * @xprt: doomed transport
649  *
650  */
651 static void xs_destroy(struct rpc_xprt *xprt)
652 {
653         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
654
655         dprintk("RPC:       xs_destroy xprt %p\n", xprt);
656
657         cancel_delayed_work(&transport->connect_worker);
658         flush_scheduled_work();
659
660         xprt_disconnect(xprt);
661         xs_close(xprt);
662         xs_free_peer_addresses(xprt);
663         kfree(xprt->slot);
664         kfree(xprt);
665 }
666
667 static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
668 {
669         return (struct rpc_xprt *) sk->sk_user_data;
670 }
671
672 /**
673  * xs_udp_data_ready - "data ready" callback for UDP sockets
674  * @sk: socket with data to read
675  * @len: how much data to read
676  *
677  */
678 static void xs_udp_data_ready(struct sock *sk, int len)
679 {
680         struct rpc_task *task;
681         struct rpc_xprt *xprt;
682         struct rpc_rqst *rovr;
683         struct sk_buff *skb;
684         int err, repsize, copied;
685         u32 _xid;
686         __be32 *xp;
687
688         read_lock(&sk->sk_callback_lock);
689         dprintk("RPC:       xs_udp_data_ready...\n");
690         if (!(xprt = xprt_from_sock(sk)))
691                 goto out;
692
693         if ((skb = skb_recv_datagram(sk, 0, 1, &err)) == NULL)
694                 goto out;
695
696         if (xprt->shutdown)
697                 goto dropit;
698
699         repsize = skb->len - sizeof(struct udphdr);
700         if (repsize < 4) {
701                 dprintk("RPC:       impossible RPC reply size %d!\n", repsize);
702                 goto dropit;
703         }
704
705         /* Copy the XID from the skb... */
706         xp = skb_header_pointer(skb, sizeof(struct udphdr),
707                                 sizeof(_xid), &_xid);
708         if (xp == NULL)
709                 goto dropit;
710
711         /* Look up and lock the request corresponding to the given XID */
712         spin_lock(&xprt->transport_lock);
713         rovr = xprt_lookup_rqst(xprt, *xp);
714         if (!rovr)
715                 goto out_unlock;
716         task = rovr->rq_task;
717
718         if ((copied = rovr->rq_private_buf.buflen) > repsize)
719                 copied = repsize;
720
721         /* Suck it into the iovec, verify checksum if not done by hw. */
722         if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb))
723                 goto out_unlock;
724
725         /* Something worked... */
726         dst_confirm(skb->dst);
727
728         xprt_adjust_cwnd(task, copied);
729         xprt_update_rtt(task);
730         xprt_complete_rqst(task, copied);
731
732  out_unlock:
733         spin_unlock(&xprt->transport_lock);
734  dropit:
735         skb_free_datagram(sk, skb);
736  out:
737         read_unlock(&sk->sk_callback_lock);
738 }
739
740 static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
741 {
742         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
743         size_t len, used;
744         char *p;
745
746         p = ((char *) &transport->tcp_fraghdr) + transport->tcp_offset;
747         len = sizeof(transport->tcp_fraghdr) - transport->tcp_offset;
748         used = xdr_skb_read_bits(desc, p, len);
749         transport->tcp_offset += used;
750         if (used != len)
751                 return;
752
753         transport->tcp_reclen = ntohl(transport->tcp_fraghdr);
754         if (transport->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
755                 transport->tcp_flags |= TCP_RCV_LAST_FRAG;
756         else
757                 transport->tcp_flags &= ~TCP_RCV_LAST_FRAG;
758         transport->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
759
760         transport->tcp_flags &= ~TCP_RCV_COPY_FRAGHDR;
761         transport->tcp_offset = 0;
762
763         /* Sanity check of the record length */
764         if (unlikely(transport->tcp_reclen < 4)) {
765                 dprintk("RPC:       invalid TCP record fragment length\n");
766                 xprt_disconnect(xprt);
767                 return;
768         }
769         dprintk("RPC:       reading TCP record fragment of length %d\n",
770                         transport->tcp_reclen);
771 }
772
773 static void xs_tcp_check_fraghdr(struct sock_xprt *transport)
774 {
775         if (transport->tcp_offset == transport->tcp_reclen) {
776                 transport->tcp_flags |= TCP_RCV_COPY_FRAGHDR;
777                 transport->tcp_offset = 0;
778                 if (transport->tcp_flags & TCP_RCV_LAST_FRAG) {
779                         transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
780                         transport->tcp_flags |= TCP_RCV_COPY_XID;
781                         transport->tcp_copied = 0;
782                 }
783         }
784 }
785
786 static inline void xs_tcp_read_xid(struct sock_xprt *transport, struct xdr_skb_reader *desc)
787 {
788         size_t len, used;
789         char *p;
790
791         len = sizeof(transport->tcp_xid) - transport->tcp_offset;
792         dprintk("RPC:       reading XID (%Zu bytes)\n", len);
793         p = ((char *) &transport->tcp_xid) + transport->tcp_offset;
794         used = xdr_skb_read_bits(desc, p, len);
795         transport->tcp_offset += used;
796         if (used != len)
797                 return;
798         transport->tcp_flags &= ~TCP_RCV_COPY_XID;
799         transport->tcp_flags |= TCP_RCV_COPY_DATA;
800         transport->tcp_copied = 4;
801         dprintk("RPC:       reading reply for XID %08x\n",
802                         ntohl(transport->tcp_xid));
803         xs_tcp_check_fraghdr(transport);
804 }
805
806 static inline void xs_tcp_read_request(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
807 {
808         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
809         struct rpc_rqst *req;
810         struct xdr_buf *rcvbuf;
811         size_t len;
812         ssize_t r;
813
814         /* Find and lock the request corresponding to this xid */
815         spin_lock(&xprt->transport_lock);
816         req = xprt_lookup_rqst(xprt, transport->tcp_xid);
817         if (!req) {
818                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
819                 dprintk("RPC:       XID %08x request not found!\n",
820                                 ntohl(transport->tcp_xid));
821                 spin_unlock(&xprt->transport_lock);
822                 return;
823         }
824
825         rcvbuf = &req->rq_private_buf;
826         len = desc->count;
827         if (len > transport->tcp_reclen - transport->tcp_offset) {
828                 struct xdr_skb_reader my_desc;
829
830                 len = transport->tcp_reclen - transport->tcp_offset;
831                 memcpy(&my_desc, desc, sizeof(my_desc));
832                 my_desc.count = len;
833                 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
834                                           &my_desc, xdr_skb_read_bits);
835                 desc->count -= r;
836                 desc->offset += r;
837         } else
838                 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
839                                           desc, xdr_skb_read_bits);
840
841         if (r > 0) {
842                 transport->tcp_copied += r;
843                 transport->tcp_offset += r;
844         }
845         if (r != len) {
846                 /* Error when copying to the receive buffer,
847                  * usually because we weren't able to allocate
848                  * additional buffer pages. All we can do now
849                  * is turn off TCP_RCV_COPY_DATA, so the request
850                  * will not receive any additional updates,
851                  * and time out.
852                  * Any remaining data from this record will
853                  * be discarded.
854                  */
855                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
856                 dprintk("RPC:       XID %08x truncated request\n",
857                                 ntohl(transport->tcp_xid));
858                 dprintk("RPC:       xprt = %p, tcp_copied = %lu, "
859                                 "tcp_offset = %u, tcp_reclen = %u\n",
860                                 xprt, transport->tcp_copied,
861                                 transport->tcp_offset, transport->tcp_reclen);
862                 goto out;
863         }
864
865         dprintk("RPC:       XID %08x read %Zd bytes\n",
866                         ntohl(transport->tcp_xid), r);
867         dprintk("RPC:       xprt = %p, tcp_copied = %lu, tcp_offset = %u, "
868                         "tcp_reclen = %u\n", xprt, transport->tcp_copied,
869                         transport->tcp_offset, transport->tcp_reclen);
870
871         if (transport->tcp_copied == req->rq_private_buf.buflen)
872                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
873         else if (transport->tcp_offset == transport->tcp_reclen) {
874                 if (transport->tcp_flags & TCP_RCV_LAST_FRAG)
875                         transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
876         }
877
878 out:
879         if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
880                 xprt_complete_rqst(req->rq_task, transport->tcp_copied);
881         spin_unlock(&xprt->transport_lock);
882         xs_tcp_check_fraghdr(transport);
883 }
884
885 static inline void xs_tcp_read_discard(struct sock_xprt *transport, struct xdr_skb_reader *desc)
886 {
887         size_t len;
888
889         len = transport->tcp_reclen - transport->tcp_offset;
890         if (len > desc->count)
891                 len = desc->count;
892         desc->count -= len;
893         desc->offset += len;
894         transport->tcp_offset += len;
895         dprintk("RPC:       discarded %Zu bytes\n", len);
896         xs_tcp_check_fraghdr(transport);
897 }
898
899 static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
900 {
901         struct rpc_xprt *xprt = rd_desc->arg.data;
902         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
903         struct xdr_skb_reader desc = {
904                 .skb    = skb,
905                 .offset = offset,
906                 .count  = len,
907         };
908
909         dprintk("RPC:       xs_tcp_data_recv started\n");
910         do {
911                 /* Read in a new fragment marker if necessary */
912                 /* Can we ever really expect to get completely empty fragments? */
913                 if (transport->tcp_flags & TCP_RCV_COPY_FRAGHDR) {
914                         xs_tcp_read_fraghdr(xprt, &desc);
915                         continue;
916                 }
917                 /* Read in the xid if necessary */
918                 if (transport->tcp_flags & TCP_RCV_COPY_XID) {
919                         xs_tcp_read_xid(transport, &desc);
920                         continue;
921                 }
922                 /* Read in the request data */
923                 if (transport->tcp_flags & TCP_RCV_COPY_DATA) {
924                         xs_tcp_read_request(xprt, &desc);
925                         continue;
926                 }
927                 /* Skip over any trailing bytes on short reads */
928                 xs_tcp_read_discard(transport, &desc);
929         } while (desc.count);
930         dprintk("RPC:       xs_tcp_data_recv done\n");
931         return len - desc.count;
932 }
933
934 /**
935  * xs_tcp_data_ready - "data ready" callback for TCP sockets
936  * @sk: socket with data to read
937  * @bytes: how much data to read
938  *
939  */
940 static void xs_tcp_data_ready(struct sock *sk, int bytes)
941 {
942         struct rpc_xprt *xprt;
943         read_descriptor_t rd_desc;
944
945         dprintk("RPC:       xs_tcp_data_ready...\n");
946
947         read_lock(&sk->sk_callback_lock);
948         if (!(xprt = xprt_from_sock(sk)))
949                 goto out;
950         if (xprt->shutdown)
951                 goto out;
952
953         /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
954         rd_desc.arg.data = xprt;
955         rd_desc.count = 65536;
956         tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
957 out:
958         read_unlock(&sk->sk_callback_lock);
959 }
960
961 /**
962  * xs_tcp_state_change - callback to handle TCP socket state changes
963  * @sk: socket whose state has changed
964  *
965  */
966 static void xs_tcp_state_change(struct sock *sk)
967 {
968         struct rpc_xprt *xprt;
969
970         read_lock(&sk->sk_callback_lock);
971         if (!(xprt = xprt_from_sock(sk)))
972                 goto out;
973         dprintk("RPC:       xs_tcp_state_change client %p...\n", xprt);
974         dprintk("RPC:       state %x conn %d dead %d zapped %d\n",
975                         sk->sk_state, xprt_connected(xprt),
976                         sock_flag(sk, SOCK_DEAD),
977                         sock_flag(sk, SOCK_ZAPPED));
978
979         switch (sk->sk_state) {
980         case TCP_ESTABLISHED:
981                 spin_lock_bh(&xprt->transport_lock);
982                 if (!xprt_test_and_set_connected(xprt)) {
983                         struct sock_xprt *transport = container_of(xprt,
984                                         struct sock_xprt, xprt);
985
986                         /* Reset TCP record info */
987                         transport->tcp_offset = 0;
988                         transport->tcp_reclen = 0;
989                         transport->tcp_copied = 0;
990                         transport->tcp_flags =
991                                 TCP_RCV_COPY_FRAGHDR | TCP_RCV_COPY_XID;
992
993                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
994                         xprt_wake_pending_tasks(xprt, 0);
995                 }
996                 spin_unlock_bh(&xprt->transport_lock);
997                 break;
998         case TCP_SYN_SENT:
999         case TCP_SYN_RECV:
1000                 break;
1001         case TCP_CLOSE_WAIT:
1002                 /* Try to schedule an autoclose RPC calls */
1003                 set_bit(XPRT_CLOSE_WAIT, &xprt->state);
1004                 if (test_and_set_bit(XPRT_LOCKED, &xprt->state) == 0)
1005                         schedule_work(&xprt->task_cleanup);
1006         default:
1007                 xprt_disconnect(xprt);
1008         }
1009  out:
1010         read_unlock(&sk->sk_callback_lock);
1011 }
1012
1013 /**
1014  * xs_udp_write_space - callback invoked when socket buffer space
1015  *                             becomes available
1016  * @sk: socket whose state has changed
1017  *
1018  * Called when more output buffer space is available for this socket.
1019  * We try not to wake our writers until they can make "significant"
1020  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1021  * with a bunch of small requests.
1022  */
1023 static void xs_udp_write_space(struct sock *sk)
1024 {
1025         read_lock(&sk->sk_callback_lock);
1026
1027         /* from net/core/sock.c:sock_def_write_space */
1028         if (sock_writeable(sk)) {
1029                 struct socket *sock;
1030                 struct rpc_xprt *xprt;
1031
1032                 if (unlikely(!(sock = sk->sk_socket)))
1033                         goto out;
1034                 if (unlikely(!(xprt = xprt_from_sock(sk))))
1035                         goto out;
1036                 if (unlikely(!test_and_clear_bit(SOCK_NOSPACE, &sock->flags)))
1037                         goto out;
1038
1039                 xprt_write_space(xprt);
1040         }
1041
1042  out:
1043         read_unlock(&sk->sk_callback_lock);
1044 }
1045
1046 /**
1047  * xs_tcp_write_space - callback invoked when socket buffer space
1048  *                             becomes available
1049  * @sk: socket whose state has changed
1050  *
1051  * Called when more output buffer space is available for this socket.
1052  * We try not to wake our writers until they can make "significant"
1053  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1054  * with a bunch of small requests.
1055  */
1056 static void xs_tcp_write_space(struct sock *sk)
1057 {
1058         read_lock(&sk->sk_callback_lock);
1059
1060         /* from net/core/stream.c:sk_stream_write_space */
1061         if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk)) {
1062                 struct socket *sock;
1063                 struct rpc_xprt *xprt;
1064
1065                 if (unlikely(!(sock = sk->sk_socket)))
1066                         goto out;
1067                 if (unlikely(!(xprt = xprt_from_sock(sk))))
1068                         goto out;
1069                 if (unlikely(!test_and_clear_bit(SOCK_NOSPACE, &sock->flags)))
1070                         goto out;
1071
1072                 xprt_write_space(xprt);
1073         }
1074
1075  out:
1076         read_unlock(&sk->sk_callback_lock);
1077 }
1078
1079 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
1080 {
1081         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1082         struct sock *sk = transport->inet;
1083
1084         if (transport->rcvsize) {
1085                 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1086                 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
1087         }
1088         if (transport->sndsize) {
1089                 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1090                 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1091                 sk->sk_write_space(sk);
1092         }
1093 }
1094
1095 /**
1096  * xs_udp_set_buffer_size - set send and receive limits
1097  * @xprt: generic transport
1098  * @sndsize: requested size of send buffer, in bytes
1099  * @rcvsize: requested size of receive buffer, in bytes
1100  *
1101  * Set socket send and receive buffer size limits.
1102  */
1103 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1104 {
1105         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1106
1107         transport->sndsize = 0;
1108         if (sndsize)
1109                 transport->sndsize = sndsize + 1024;
1110         transport->rcvsize = 0;
1111         if (rcvsize)
1112                 transport->rcvsize = rcvsize + 1024;
1113
1114         xs_udp_do_set_buffer_size(xprt);
1115 }
1116
1117 /**
1118  * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1119  * @task: task that timed out
1120  *
1121  * Adjust the congestion window after a retransmit timeout has occurred.
1122  */
1123 static void xs_udp_timer(struct rpc_task *task)
1124 {
1125         xprt_adjust_cwnd(task, -ETIMEDOUT);
1126 }
1127
1128 static unsigned short xs_get_random_port(void)
1129 {
1130         unsigned short range = xprt_max_resvport - xprt_min_resvport;
1131         unsigned short rand = (unsigned short) net_random() % range;
1132         return rand + xprt_min_resvport;
1133 }
1134
1135 /**
1136  * xs_set_port - reset the port number in the remote endpoint address
1137  * @xprt: generic transport
1138  * @port: new port number
1139  *
1140  */
1141 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1142 {
1143         struct sockaddr_in *sap = (struct sockaddr_in *) &xprt->addr;
1144
1145         dprintk("RPC:       setting port for xprt %p to %u\n", xprt, port);
1146
1147         sap->sin_port = htons(port);
1148 }
1149
1150 static int xs_bindresvport(struct sock_xprt *transport, struct socket *sock)
1151 {
1152         struct sockaddr_in myaddr = {
1153                 .sin_family = AF_INET,
1154         };
1155         int err;
1156         unsigned short port = transport->port;
1157
1158         do {
1159                 myaddr.sin_port = htons(port);
1160                 err = kernel_bind(sock, (struct sockaddr *) &myaddr,
1161                                                 sizeof(myaddr));
1162                 if (err == 0) {
1163                         transport->port = port;
1164                         dprintk("RPC:       xs_bindresvport bound to port %u\n",
1165                                         port);
1166                         return 0;
1167                 }
1168                 if (port <= xprt_min_resvport)
1169                         port = xprt_max_resvport;
1170                 else
1171                         port--;
1172         } while (err == -EADDRINUSE && port != transport->port);
1173
1174         dprintk("RPC:       can't bind to reserved port (%d).\n", -err);
1175         return err;
1176 }
1177
1178 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1179 static struct lock_class_key xs_key[2];
1180 static struct lock_class_key xs_slock_key[2];
1181
1182 static inline void xs_reclassify_socket(struct socket *sock)
1183 {
1184         struct sock *sk = sock->sk;
1185         BUG_ON(sk->sk_lock.owner != NULL);
1186         switch (sk->sk_family) {
1187         case AF_INET:
1188                 sock_lock_init_class_and_name(sk, "slock-AF_INET-NFS",
1189                         &xs_slock_key[0], "sk_lock-AF_INET-NFS", &xs_key[0]);
1190                 break;
1191
1192         case AF_INET6:
1193                 sock_lock_init_class_and_name(sk, "slock-AF_INET6-NFS",
1194                         &xs_slock_key[1], "sk_lock-AF_INET6-NFS", &xs_key[1]);
1195                 break;
1196
1197         default:
1198                 BUG();
1199         }
1200 }
1201 #else
1202 static inline void xs_reclassify_socket(struct socket *sock)
1203 {
1204 }
1205 #endif
1206
1207 /**
1208  * xs_udp_connect_worker - set up a UDP socket
1209  * @work: RPC transport to connect
1210  *
1211  * Invoked by a work queue tasklet.
1212  */
1213 static void xs_udp_connect_worker(struct work_struct *work)
1214 {
1215         struct sock_xprt *transport =
1216                 container_of(work, struct sock_xprt, connect_worker.work);
1217         struct rpc_xprt *xprt = &transport->xprt;
1218         struct socket *sock = transport->sock;
1219         int err, status = -EIO;
1220
1221         if (xprt->shutdown || !xprt_bound(xprt))
1222                 goto out;
1223
1224         /* Start by resetting any existing state */
1225         xs_close(xprt);
1226
1227         if ((err = sock_create_kern(PF_INET, SOCK_DGRAM, IPPROTO_UDP, &sock)) < 0) {
1228                 dprintk("RPC:       can't create UDP transport socket (%d).\n", -err);
1229                 goto out;
1230         }
1231         xs_reclassify_socket(sock);
1232
1233         if (xprt->resvport && xs_bindresvport(transport, sock) < 0) {
1234                 sock_release(sock);
1235                 goto out;
1236         }
1237
1238         dprintk("RPC:       worker connecting xprt %p to address: %s\n",
1239                         xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1240
1241         if (!transport->inet) {
1242                 struct sock *sk = sock->sk;
1243
1244                 write_lock_bh(&sk->sk_callback_lock);
1245
1246                 sk->sk_user_data = xprt;
1247                 transport->old_data_ready = sk->sk_data_ready;
1248                 transport->old_state_change = sk->sk_state_change;
1249                 transport->old_write_space = sk->sk_write_space;
1250                 sk->sk_data_ready = xs_udp_data_ready;
1251                 sk->sk_write_space = xs_udp_write_space;
1252                 sk->sk_no_check = UDP_CSUM_NORCV;
1253                 sk->sk_allocation = GFP_ATOMIC;
1254
1255                 xprt_set_connected(xprt);
1256
1257                 /* Reset to new socket */
1258                 transport->sock = sock;
1259                 transport->inet = sk;
1260
1261                 write_unlock_bh(&sk->sk_callback_lock);
1262         }
1263         xs_udp_do_set_buffer_size(xprt);
1264         status = 0;
1265 out:
1266         xprt_wake_pending_tasks(xprt, status);
1267         xprt_clear_connecting(xprt);
1268 }
1269
1270 /*
1271  * We need to preserve the port number so the reply cache on the server can
1272  * find our cached RPC replies when we get around to reconnecting.
1273  */
1274 static void xs_tcp_reuse_connection(struct rpc_xprt *xprt)
1275 {
1276         int result;
1277         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1278         struct sockaddr any;
1279
1280         dprintk("RPC:       disconnecting xprt %p to reuse port\n", xprt);
1281
1282         /*
1283          * Disconnect the transport socket by doing a connect operation
1284          * with AF_UNSPEC.  This should return immediately...
1285          */
1286         memset(&any, 0, sizeof(any));
1287         any.sa_family = AF_UNSPEC;
1288         result = kernel_connect(transport->sock, &any, sizeof(any), 0);
1289         if (result)
1290                 dprintk("RPC:       AF_UNSPEC connect return code %d\n",
1291                                 result);
1292 }
1293
1294 /**
1295  * xs_tcp_connect_worker - connect a TCP socket to a remote endpoint
1296  * @work: RPC transport to connect
1297  *
1298  * Invoked by a work queue tasklet.
1299  */
1300 static void xs_tcp_connect_worker(struct work_struct *work)
1301 {
1302         struct sock_xprt *transport =
1303                 container_of(work, struct sock_xprt, connect_worker.work);
1304         struct rpc_xprt *xprt = &transport->xprt;
1305         struct socket *sock = transport->sock;
1306         int err, status = -EIO;
1307
1308         if (xprt->shutdown || !xprt_bound(xprt))
1309                 goto out;
1310
1311         if (!sock) {
1312                 /* start from scratch */
1313                 if ((err = sock_create_kern(PF_INET, SOCK_STREAM, IPPROTO_TCP, &sock)) < 0) {
1314                         dprintk("RPC:       can't create TCP transport "
1315                                         "socket (%d).\n", -err);
1316                         goto out;
1317                 }
1318                 xs_reclassify_socket(sock);
1319
1320                 if (xprt->resvport && xs_bindresvport(transport, sock) < 0) {
1321                         sock_release(sock);
1322                         goto out;
1323                 }
1324         } else
1325                 /* "close" the socket, preserving the local port */
1326                 xs_tcp_reuse_connection(xprt);
1327
1328         dprintk("RPC:       worker connecting xprt %p to address: %s\n",
1329                         xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1330
1331         if (!transport->inet) {
1332                 struct sock *sk = sock->sk;
1333
1334                 write_lock_bh(&sk->sk_callback_lock);
1335
1336                 sk->sk_user_data = xprt;
1337                 transport->old_data_ready = sk->sk_data_ready;
1338                 transport->old_state_change = sk->sk_state_change;
1339                 transport->old_write_space = sk->sk_write_space;
1340                 sk->sk_data_ready = xs_tcp_data_ready;
1341                 sk->sk_state_change = xs_tcp_state_change;
1342                 sk->sk_write_space = xs_tcp_write_space;
1343                 sk->sk_allocation = GFP_ATOMIC;
1344
1345                 /* socket options */
1346                 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
1347                 sock_reset_flag(sk, SOCK_LINGER);
1348                 tcp_sk(sk)->linger2 = 0;
1349                 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
1350
1351                 xprt_clear_connected(xprt);
1352
1353                 /* Reset to new socket */
1354                 transport->sock = sock;
1355                 transport->inet = sk;
1356
1357                 write_unlock_bh(&sk->sk_callback_lock);
1358         }
1359
1360         /* Tell the socket layer to start connecting... */
1361         xprt->stat.connect_count++;
1362         xprt->stat.connect_start = jiffies;
1363         status = kernel_connect(sock, (struct sockaddr *) &xprt->addr,
1364                         xprt->addrlen, O_NONBLOCK);
1365         dprintk("RPC:       %p connect status %d connected %d sock state %d\n",
1366                         xprt, -status, xprt_connected(xprt),
1367                         sock->sk->sk_state);
1368         if (status < 0) {
1369                 switch (status) {
1370                         case -EINPROGRESS:
1371                         case -EALREADY:
1372                                 goto out_clear;
1373                         case -ECONNREFUSED:
1374                         case -ECONNRESET:
1375                                 /* retry with existing socket, after a delay */
1376                                 break;
1377                         default:
1378                                 /* get rid of existing socket, and retry */
1379                                 xs_close(xprt);
1380                                 break;
1381                 }
1382         }
1383 out:
1384         xprt_wake_pending_tasks(xprt, status);
1385 out_clear:
1386         xprt_clear_connecting(xprt);
1387 }
1388
1389 /**
1390  * xs_connect - connect a socket to a remote endpoint
1391  * @task: address of RPC task that manages state of connect request
1392  *
1393  * TCP: If the remote end dropped the connection, delay reconnecting.
1394  *
1395  * UDP socket connects are synchronous, but we use a work queue anyway
1396  * to guarantee that even unprivileged user processes can set up a
1397  * socket on a privileged port.
1398  *
1399  * If a UDP socket connect fails, the delay behavior here prevents
1400  * retry floods (hard mounts).
1401  */
1402 static void xs_connect(struct rpc_task *task)
1403 {
1404         struct rpc_xprt *xprt = task->tk_xprt;
1405         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1406
1407         if (xprt_test_and_set_connecting(xprt))
1408                 return;
1409
1410         if (transport->sock != NULL) {
1411                 dprintk("RPC:       xs_connect delayed xprt %p for %lu "
1412                                 "seconds\n",
1413                                 xprt, xprt->reestablish_timeout / HZ);
1414                 schedule_delayed_work(&transport->connect_worker,
1415                                         xprt->reestablish_timeout);
1416                 xprt->reestablish_timeout <<= 1;
1417                 if (xprt->reestablish_timeout > XS_TCP_MAX_REEST_TO)
1418                         xprt->reestablish_timeout = XS_TCP_MAX_REEST_TO;
1419         } else {
1420                 dprintk("RPC:       xs_connect scheduled xprt %p\n", xprt);
1421                 schedule_delayed_work(&transport->connect_worker, 0);
1422
1423                 /* flush_scheduled_work can sleep... */
1424                 if (!RPC_IS_ASYNC(task))
1425                         flush_scheduled_work();
1426         }
1427 }
1428
1429 /**
1430  * xs_udp_print_stats - display UDP socket-specifc stats
1431  * @xprt: rpc_xprt struct containing statistics
1432  * @seq: output file
1433  *
1434  */
1435 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
1436 {
1437         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1438
1439         seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %Lu %Lu\n",
1440                         transport->port,
1441                         xprt->stat.bind_count,
1442                         xprt->stat.sends,
1443                         xprt->stat.recvs,
1444                         xprt->stat.bad_xids,
1445                         xprt->stat.req_u,
1446                         xprt->stat.bklog_u);
1447 }
1448
1449 /**
1450  * xs_tcp_print_stats - display TCP socket-specifc stats
1451  * @xprt: rpc_xprt struct containing statistics
1452  * @seq: output file
1453  *
1454  */
1455 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
1456 {
1457         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1458         long idle_time = 0;
1459
1460         if (xprt_connected(xprt))
1461                 idle_time = (long)(jiffies - xprt->last_used) / HZ;
1462
1463         seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu %Lu %Lu\n",
1464                         transport->port,
1465                         xprt->stat.bind_count,
1466                         xprt->stat.connect_count,
1467                         xprt->stat.connect_time,
1468                         idle_time,
1469                         xprt->stat.sends,
1470                         xprt->stat.recvs,
1471                         xprt->stat.bad_xids,
1472                         xprt->stat.req_u,
1473                         xprt->stat.bklog_u);
1474 }
1475
1476 static struct rpc_xprt_ops xs_udp_ops = {
1477         .set_buffer_size        = xs_udp_set_buffer_size,
1478         .reserve_xprt           = xprt_reserve_xprt_cong,
1479         .release_xprt           = xprt_release_xprt_cong,
1480         .rpcbind                = rpc_getport,
1481         .set_port               = xs_set_port,
1482         .connect                = xs_connect,
1483         .buf_alloc              = rpc_malloc,
1484         .buf_free               = rpc_free,
1485         .send_request           = xs_udp_send_request,
1486         .set_retrans_timeout    = xprt_set_retrans_timeout_rtt,
1487         .timer                  = xs_udp_timer,
1488         .release_request        = xprt_release_rqst_cong,
1489         .close                  = xs_close,
1490         .destroy                = xs_destroy,
1491         .print_stats            = xs_udp_print_stats,
1492 };
1493
1494 static struct rpc_xprt_ops xs_tcp_ops = {
1495         .reserve_xprt           = xprt_reserve_xprt,
1496         .release_xprt           = xs_tcp_release_xprt,
1497         .rpcbind                = rpc_getport,
1498         .set_port               = xs_set_port,
1499         .connect                = xs_connect,
1500         .buf_alloc              = rpc_malloc,
1501         .buf_free               = rpc_free,
1502         .send_request           = xs_tcp_send_request,
1503         .set_retrans_timeout    = xprt_set_retrans_timeout_def,
1504         .close                  = xs_close,
1505         .destroy                = xs_destroy,
1506         .print_stats            = xs_tcp_print_stats,
1507 };
1508
1509 static struct rpc_xprt *xs_setup_xprt(struct sockaddr *addr, size_t addrlen, unsigned int slot_table_size)
1510 {
1511         struct rpc_xprt *xprt;
1512         struct sock_xprt *new;
1513
1514         if (addrlen > sizeof(xprt->addr)) {
1515                 dprintk("RPC:       xs_setup_xprt: address too large\n");
1516                 return ERR_PTR(-EBADF);
1517         }
1518
1519         new = kzalloc(sizeof(*new), GFP_KERNEL);
1520         if (new == NULL) {
1521                 dprintk("RPC:       xs_setup_xprt: couldn't allocate "
1522                                 "rpc_xprt\n");
1523                 return ERR_PTR(-ENOMEM);
1524         }
1525         xprt = &new->xprt;
1526
1527         xprt->max_reqs = slot_table_size;
1528         xprt->slot = kcalloc(xprt->max_reqs, sizeof(struct rpc_rqst), GFP_KERNEL);
1529         if (xprt->slot == NULL) {
1530                 kfree(xprt);
1531                 dprintk("RPC:       xs_setup_xprt: couldn't allocate slot "
1532                                 "table\n");
1533                 return ERR_PTR(-ENOMEM);
1534         }
1535
1536         memcpy(&xprt->addr, addr, addrlen);
1537         xprt->addrlen = addrlen;
1538         new->port = xs_get_random_port();
1539
1540         return xprt;
1541 }
1542
1543 /**
1544  * xs_setup_udp - Set up transport to use a UDP socket
1545  * @addr: address of remote server
1546  * @addrlen: length of address in bytes
1547  * @to:   timeout parameters
1548  *
1549  */
1550 struct rpc_xprt *xs_setup_udp(struct sockaddr *addr, size_t addrlen, struct rpc_timeout *to)
1551 {
1552         struct rpc_xprt *xprt;
1553         struct sock_xprt *transport;
1554
1555         xprt = xs_setup_xprt(addr, addrlen, xprt_udp_slot_table_entries);
1556         if (IS_ERR(xprt))
1557                 return xprt;
1558         transport = container_of(xprt, struct sock_xprt, xprt);
1559
1560         if (ntohs(((struct sockaddr_in *)addr)->sin_port) != 0)
1561                 xprt_set_bound(xprt);
1562
1563         xprt->prot = IPPROTO_UDP;
1564         xprt->tsh_size = 0;
1565         /* XXX: header size can vary due to auth type, IPv6, etc. */
1566         xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
1567
1568         INIT_DELAYED_WORK(&transport->connect_worker, xs_udp_connect_worker);
1569         xprt->bind_timeout = XS_BIND_TO;
1570         xprt->connect_timeout = XS_UDP_CONN_TO;
1571         xprt->reestablish_timeout = XS_UDP_REEST_TO;
1572         xprt->idle_timeout = XS_IDLE_DISC_TO;
1573
1574         xprt->ops = &xs_udp_ops;
1575
1576         if (to)
1577                 xprt->timeout = *to;
1578         else
1579                 xprt_set_timeout(&xprt->timeout, 5, 5 * HZ);
1580
1581         xs_format_peer_addresses(xprt);
1582         dprintk("RPC:       set up transport to address %s\n",
1583                         xprt->address_strings[RPC_DISPLAY_ALL]);
1584
1585         return xprt;
1586 }
1587
1588 /**
1589  * xs_setup_tcp - Set up transport to use a TCP socket
1590  * @addr: address of remote server
1591  * @addrlen: length of address in bytes
1592  * @to: timeout parameters
1593  *
1594  */
1595 struct rpc_xprt *xs_setup_tcp(struct sockaddr *addr, size_t addrlen, struct rpc_timeout *to)
1596 {
1597         struct rpc_xprt *xprt;
1598         struct sock_xprt *transport;
1599
1600         xprt = xs_setup_xprt(addr, addrlen, xprt_tcp_slot_table_entries);
1601         if (IS_ERR(xprt))
1602                 return xprt;
1603         transport = container_of(xprt, struct sock_xprt, xprt);
1604
1605         if (ntohs(((struct sockaddr_in *)addr)->sin_port) != 0)
1606                 xprt_set_bound(xprt);
1607
1608         xprt->prot = IPPROTO_TCP;
1609         xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
1610         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
1611
1612         INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_connect_worker);
1613         xprt->bind_timeout = XS_BIND_TO;
1614         xprt->connect_timeout = XS_TCP_CONN_TO;
1615         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1616         xprt->idle_timeout = XS_IDLE_DISC_TO;
1617
1618         xprt->ops = &xs_tcp_ops;
1619
1620         if (to)
1621                 xprt->timeout = *to;
1622         else
1623                 xprt_set_timeout(&xprt->timeout, 2, 60 * HZ);
1624
1625         xs_format_peer_addresses(xprt);
1626         dprintk("RPC:       set up transport to address %s\n",
1627                         xprt->address_strings[RPC_DISPLAY_ALL]);
1628
1629         return xprt;
1630 }
1631
1632 /**
1633  * init_socket_xprt - set up xprtsock's sysctls
1634  *
1635  */
1636 int init_socket_xprt(void)
1637 {
1638 #ifdef RPC_DEBUG
1639         if (!sunrpc_table_header) {
1640                 sunrpc_table_header = register_sysctl_table(sunrpc_table, 1);
1641 #ifdef CONFIG_PROC_FS
1642                 if (sunrpc_table[0].de)
1643                         sunrpc_table[0].de->owner = THIS_MODULE;
1644 #endif
1645         }
1646 #endif
1647
1648         return 0;
1649 }
1650
1651 /**
1652  * cleanup_socket_xprt - remove xprtsock's sysctls
1653  *
1654  */
1655 void cleanup_socket_xprt(void)
1656 {
1657 #ifdef RPC_DEBUG
1658         if (sunrpc_table_header) {
1659                 unregister_sysctl_table(sunrpc_table_header);
1660                 sunrpc_table_header = NULL;
1661         }
1662 #endif
1663 }