015f29edb619a8f381d0a8b1e9221bccb6cd21f3
[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
18 #include <linux/types.h>
19 #include <linux/slab.h>
20 #include <linux/capability.h>
21 #include <linux/pagemap.h>
22 #include <linux/errno.h>
23 #include <linux/socket.h>
24 #include <linux/in.h>
25 #include <linux/net.h>
26 #include <linux/mm.h>
27 #include <linux/udp.h>
28 #include <linux/tcp.h>
29 #include <linux/sunrpc/clnt.h>
30 #include <linux/sunrpc/sched.h>
31 #include <linux/file.h>
32
33 #include <net/sock.h>
34 #include <net/checksum.h>
35 #include <net/udp.h>
36 #include <net/tcp.h>
37
38 /*
39  * xprtsock tunables
40  */
41 unsigned int xprt_udp_slot_table_entries = RPC_DEF_SLOT_TABLE;
42 unsigned int xprt_tcp_slot_table_entries = RPC_DEF_SLOT_TABLE;
43
44 unsigned int xprt_min_resvport = RPC_DEF_MIN_RESVPORT;
45 unsigned int xprt_max_resvport = RPC_DEF_MAX_RESVPORT;
46
47 /*
48  * We can register our own files under /proc/sys/sunrpc by
49  * calling register_sysctl_table() again.  The files in that
50  * directory become the union of all files registered there.
51  *
52  * We simply need to make sure that we don't collide with
53  * someone else's file names!
54  */
55
56 #ifdef RPC_DEBUG
57
58 static unsigned int min_slot_table_size = RPC_MIN_SLOT_TABLE;
59 static unsigned int max_slot_table_size = RPC_MAX_SLOT_TABLE;
60 static unsigned int xprt_min_resvport_limit = RPC_MIN_RESVPORT;
61 static unsigned int xprt_max_resvport_limit = RPC_MAX_RESVPORT;
62
63 static struct ctl_table_header *sunrpc_table_header;
64
65 /*
66  * FIXME: changing the UDP slot table size should also resize the UDP
67  *        socket buffers for existing UDP transports
68  */
69 static ctl_table xs_tunables_table[] = {
70         {
71                 .ctl_name       = CTL_SLOTTABLE_UDP,
72                 .procname       = "udp_slot_table_entries",
73                 .data           = &xprt_udp_slot_table_entries,
74                 .maxlen         = sizeof(unsigned int),
75                 .mode           = 0644,
76                 .proc_handler   = &proc_dointvec_minmax,
77                 .strategy       = &sysctl_intvec,
78                 .extra1         = &min_slot_table_size,
79                 .extra2         = &max_slot_table_size
80         },
81         {
82                 .ctl_name       = CTL_SLOTTABLE_TCP,
83                 .procname       = "tcp_slot_table_entries",
84                 .data           = &xprt_tcp_slot_table_entries,
85                 .maxlen         = sizeof(unsigned int),
86                 .mode           = 0644,
87                 .proc_handler   = &proc_dointvec_minmax,
88                 .strategy       = &sysctl_intvec,
89                 .extra1         = &min_slot_table_size,
90                 .extra2         = &max_slot_table_size
91         },
92         {
93                 .ctl_name       = CTL_MIN_RESVPORT,
94                 .procname       = "min_resvport",
95                 .data           = &xprt_min_resvport,
96                 .maxlen         = sizeof(unsigned int),
97                 .mode           = 0644,
98                 .proc_handler   = &proc_dointvec_minmax,
99                 .strategy       = &sysctl_intvec,
100                 .extra1         = &xprt_min_resvport_limit,
101                 .extra2         = &xprt_max_resvport_limit
102         },
103         {
104                 .ctl_name       = CTL_MAX_RESVPORT,
105                 .procname       = "max_resvport",
106                 .data           = &xprt_max_resvport,
107                 .maxlen         = sizeof(unsigned int),
108                 .mode           = 0644,
109                 .proc_handler   = &proc_dointvec_minmax,
110                 .strategy       = &sysctl_intvec,
111                 .extra1         = &xprt_min_resvport_limit,
112                 .extra2         = &xprt_max_resvport_limit
113         },
114         {
115                 .ctl_name = 0,
116         },
117 };
118
119 static ctl_table sunrpc_table[] = {
120         {
121                 .ctl_name       = CTL_SUNRPC,
122                 .procname       = "sunrpc",
123                 .mode           = 0555,
124                 .child          = xs_tunables_table
125         },
126         {
127                 .ctl_name = 0,
128         },
129 };
130
131 #endif
132
133 /*
134  * How many times to try sending a request on a socket before waiting
135  * for the socket buffer to clear.
136  */
137 #define XS_SENDMSG_RETRY        (10U)
138
139 /*
140  * Time out for an RPC UDP socket connect.  UDP socket connects are
141  * synchronous, but we set a timeout anyway in case of resource
142  * exhaustion on the local host.
143  */
144 #define XS_UDP_CONN_TO          (5U * HZ)
145
146 /*
147  * Wait duration for an RPC TCP connection to be established.  Solaris
148  * NFS over TCP uses 60 seconds, for example, which is in line with how
149  * long a server takes to reboot.
150  */
151 #define XS_TCP_CONN_TO          (60U * HZ)
152
153 /*
154  * Wait duration for a reply from the RPC portmapper.
155  */
156 #define XS_BIND_TO              (60U * HZ)
157
158 /*
159  * Delay if a UDP socket connect error occurs.  This is most likely some
160  * kind of resource problem on the local host.
161  */
162 #define XS_UDP_REEST_TO         (2U * HZ)
163
164 /*
165  * The reestablish timeout allows clients to delay for a bit before attempting
166  * to reconnect to a server that just dropped our connection.
167  *
168  * We implement an exponential backoff when trying to reestablish a TCP
169  * transport connection with the server.  Some servers like to drop a TCP
170  * connection when they are overworked, so we start with a short timeout and
171  * increase over time if the server is down or not responding.
172  */
173 #define XS_TCP_INIT_REEST_TO    (3U * HZ)
174 #define XS_TCP_MAX_REEST_TO     (5U * 60 * HZ)
175
176 /*
177  * TCP idle timeout; client drops the transport socket if it is idle
178  * for this long.  Note that we also timeout UDP sockets to prevent
179  * holding port numbers when there is no RPC traffic.
180  */
181 #define XS_IDLE_DISC_TO         (5U * 60 * HZ)
182
183 #ifdef RPC_DEBUG
184 # undef  RPC_DEBUG_DATA
185 # define RPCDBG_FACILITY        RPCDBG_TRANS
186 #endif
187
188 #ifdef RPC_DEBUG_DATA
189 static void xs_pktdump(char *msg, u32 *packet, unsigned int count)
190 {
191         u8 *buf = (u8 *) packet;
192         int j;
193
194         dprintk("RPC:       %s\n", msg);
195         for (j = 0; j < count && j < 128; j += 4) {
196                 if (!(j & 31)) {
197                         if (j)
198                                 dprintk("\n");
199                         dprintk("0x%04x ", j);
200                 }
201                 dprintk("%02x%02x%02x%02x ",
202                         buf[j], buf[j+1], buf[j+2], buf[j+3]);
203         }
204         dprintk("\n");
205 }
206 #else
207 static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count)
208 {
209         /* NOP */
210 }
211 #endif
212
213 struct sock_xprt {
214         struct rpc_xprt         xprt;
215
216         /*
217          * Network layer
218          */
219         struct socket *         sock;
220         struct sock *           inet;
221
222         /*
223          * State of TCP reply receive
224          */
225         __be32                  tcp_fraghdr,
226                                 tcp_xid;
227
228         u32                     tcp_offset,
229                                 tcp_reclen;
230
231         unsigned long           tcp_copied,
232                                 tcp_flags;
233
234         /*
235          * Connection of transports
236          */
237         struct delayed_work     connect_worker;
238         struct sockaddr_storage addr;
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, NIPQUAD_FMT,
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="NIPQUAD_FMT" 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;
527         unsigned int retry = 0;
528
529         xs_encode_tcp_record_marker(&req->rq_snd_buf);
530
531         xs_pktdump("packet data:",
532                                 req->rq_svec->iov_base,
533                                 req->rq_svec->iov_len);
534
535         /* Continue transmitting the packet/record. We must be careful
536          * to cope with writespace callbacks arriving _after_ we have
537          * called sendmsg(). */
538         while (1) {
539                 req->rq_xtime = jiffies;
540                 status = xs_sendpages(transport->sock,
541                                         NULL, 0, xdr, req->rq_bytes_sent);
542
543                 dprintk("RPC:       xs_tcp_send_request(%u) = %d\n",
544                                 xdr->len - req->rq_bytes_sent, status);
545
546                 if (unlikely(status < 0))
547                         break;
548
549                 /* If we've sent the entire packet, immediately
550                  * reset the count of bytes sent. */
551                 req->rq_bytes_sent += status;
552                 task->tk_bytes_sent += status;
553                 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
554                         req->rq_bytes_sent = 0;
555                         return 0;
556                 }
557
558                 status = -EAGAIN;
559                 if (retry++ > XS_SENDMSG_RETRY)
560                         break;
561         }
562
563         switch (status) {
564         case -EAGAIN:
565                 xs_nospace(task);
566                 break;
567         case -ECONNREFUSED:
568         case -ECONNRESET:
569         case -ENOTCONN:
570         case -EPIPE:
571                 status = -ENOTCONN;
572                 break;
573         default:
574                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
575                         -status);
576                 xprt_disconnect(xprt);
577                 break;
578         }
579
580         return status;
581 }
582
583 /**
584  * xs_tcp_release_xprt - clean up after a tcp transmission
585  * @xprt: transport
586  * @task: rpc task
587  *
588  * This cleans up if an error causes us to abort the transmission of a request.
589  * In this case, the socket may need to be reset in order to avoid confusing
590  * the server.
591  */
592 static void xs_tcp_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
593 {
594         struct rpc_rqst *req;
595
596         if (task != xprt->snd_task)
597                 return;
598         if (task == NULL)
599                 goto out_release;
600         req = task->tk_rqstp;
601         if (req->rq_bytes_sent == 0)
602                 goto out_release;
603         if (req->rq_bytes_sent == req->rq_snd_buf.len)
604                 goto out_release;
605         set_bit(XPRT_CLOSE_WAIT, &task->tk_xprt->state);
606 out_release:
607         xprt_release_xprt(xprt, task);
608 }
609
610 /**
611  * xs_close - close a socket
612  * @xprt: transport
613  *
614  * This is used when all requests are complete; ie, no DRC state remains
615  * on the server we want to save.
616  */
617 static void xs_close(struct rpc_xprt *xprt)
618 {
619         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
620         struct socket *sock = transport->sock;
621         struct sock *sk = transport->inet;
622
623         if (!sk)
624                 goto clear_close_wait;
625
626         dprintk("RPC:       xs_close xprt %p\n", xprt);
627
628         write_lock_bh(&sk->sk_callback_lock);
629         transport->inet = NULL;
630         transport->sock = NULL;
631
632         sk->sk_user_data = NULL;
633         sk->sk_data_ready = transport->old_data_ready;
634         sk->sk_state_change = transport->old_state_change;
635         sk->sk_write_space = transport->old_write_space;
636         write_unlock_bh(&sk->sk_callback_lock);
637
638         sk->sk_no_check = 0;
639
640         sock_release(sock);
641 clear_close_wait:
642         smp_mb__before_clear_bit();
643         clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
644         smp_mb__after_clear_bit();
645 }
646
647 /**
648  * xs_destroy - prepare to shutdown a transport
649  * @xprt: doomed transport
650  *
651  */
652 static void xs_destroy(struct rpc_xprt *xprt)
653 {
654         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
655
656         dprintk("RPC:       xs_destroy xprt %p\n", xprt);
657
658         cancel_rearming_delayed_work(&transport->connect_worker);
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                         queue_work(rpciod_workqueue, &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_bind(struct sock_xprt *transport, struct socket *sock)
1151 {
1152         struct sockaddr_in myaddr = {
1153                 .sin_family = AF_INET,
1154         };
1155         struct sockaddr_in *sa;
1156         int err;
1157         unsigned short port = transport->port;
1158
1159         if (!transport->xprt.resvport)
1160                 port = 0;
1161         sa = (struct sockaddr_in *)&transport->addr;
1162         myaddr.sin_addr = sa->sin_addr;
1163         do {
1164                 myaddr.sin_port = htons(port);
1165                 err = kernel_bind(sock, (struct sockaddr *) &myaddr,
1166                                                 sizeof(myaddr));
1167                 if (!transport->xprt.resvport)
1168                         break;
1169                 if (err == 0) {
1170                         transport->port = port;
1171                         break;
1172                 }
1173                 if (port <= xprt_min_resvport)
1174                         port = xprt_max_resvport;
1175                 else
1176                         port--;
1177         } while (err == -EADDRINUSE && port != transport->port);
1178         dprintk("RPC:       xs_bind "NIPQUAD_FMT":%u: %s (%d)\n",
1179                 NIPQUAD(myaddr.sin_addr), port, err ? "failed" : "ok", err);
1180         return err;
1181 }
1182
1183 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1184 static struct lock_class_key xs_key[2];
1185 static struct lock_class_key xs_slock_key[2];
1186
1187 static inline void xs_reclassify_socket(struct socket *sock)
1188 {
1189         struct sock *sk = sock->sk;
1190         BUG_ON(sk->sk_lock.owner != NULL);
1191         switch (sk->sk_family) {
1192         case AF_INET:
1193                 sock_lock_init_class_and_name(sk, "slock-AF_INET-NFS",
1194                         &xs_slock_key[0], "sk_lock-AF_INET-NFS", &xs_key[0]);
1195                 break;
1196
1197         case AF_INET6:
1198                 sock_lock_init_class_and_name(sk, "slock-AF_INET6-NFS",
1199                         &xs_slock_key[1], "sk_lock-AF_INET6-NFS", &xs_key[1]);
1200                 break;
1201
1202         default:
1203                 BUG();
1204         }
1205 }
1206 #else
1207 static inline void xs_reclassify_socket(struct socket *sock)
1208 {
1209 }
1210 #endif
1211
1212 /**
1213  * xs_udp_connect_worker - set up a UDP socket
1214  * @work: RPC transport to connect
1215  *
1216  * Invoked by a work queue tasklet.
1217  */
1218 static void xs_udp_connect_worker(struct work_struct *work)
1219 {
1220         struct sock_xprt *transport =
1221                 container_of(work, struct sock_xprt, connect_worker.work);
1222         struct rpc_xprt *xprt = &transport->xprt;
1223         struct socket *sock = transport->sock;
1224         int err, status = -EIO;
1225
1226         if (xprt->shutdown || !xprt_bound(xprt))
1227                 goto out;
1228
1229         /* Start by resetting any existing state */
1230         xs_close(xprt);
1231
1232         if ((err = sock_create_kern(PF_INET, SOCK_DGRAM, IPPROTO_UDP, &sock)) < 0) {
1233                 dprintk("RPC:       can't create UDP transport socket (%d).\n", -err);
1234                 goto out;
1235         }
1236         xs_reclassify_socket(sock);
1237
1238         if (xs_bind(transport, sock)) {
1239                 sock_release(sock);
1240                 goto out;
1241         }
1242
1243         dprintk("RPC:       worker connecting xprt %p to address: %s\n",
1244                         xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1245
1246         if (!transport->inet) {
1247                 struct sock *sk = sock->sk;
1248
1249                 write_lock_bh(&sk->sk_callback_lock);
1250
1251                 sk->sk_user_data = xprt;
1252                 transport->old_data_ready = sk->sk_data_ready;
1253                 transport->old_state_change = sk->sk_state_change;
1254                 transport->old_write_space = sk->sk_write_space;
1255                 sk->sk_data_ready = xs_udp_data_ready;
1256                 sk->sk_write_space = xs_udp_write_space;
1257                 sk->sk_no_check = UDP_CSUM_NORCV;
1258                 sk->sk_allocation = GFP_ATOMIC;
1259
1260                 xprt_set_connected(xprt);
1261
1262                 /* Reset to new socket */
1263                 transport->sock = sock;
1264                 transport->inet = sk;
1265
1266                 write_unlock_bh(&sk->sk_callback_lock);
1267         }
1268         xs_udp_do_set_buffer_size(xprt);
1269         status = 0;
1270 out:
1271         xprt_wake_pending_tasks(xprt, status);
1272         xprt_clear_connecting(xprt);
1273 }
1274
1275 /*
1276  * We need to preserve the port number so the reply cache on the server can
1277  * find our cached RPC replies when we get around to reconnecting.
1278  */
1279 static void xs_tcp_reuse_connection(struct rpc_xprt *xprt)
1280 {
1281         int result;
1282         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1283         struct sockaddr any;
1284
1285         dprintk("RPC:       disconnecting xprt %p to reuse port\n", xprt);
1286
1287         /*
1288          * Disconnect the transport socket by doing a connect operation
1289          * with AF_UNSPEC.  This should return immediately...
1290          */
1291         memset(&any, 0, sizeof(any));
1292         any.sa_family = AF_UNSPEC;
1293         result = kernel_connect(transport->sock, &any, sizeof(any), 0);
1294         if (result)
1295                 dprintk("RPC:       AF_UNSPEC connect return code %d\n",
1296                                 result);
1297 }
1298
1299 /**
1300  * xs_tcp_connect_worker - connect a TCP socket to a remote endpoint
1301  * @work: RPC transport to connect
1302  *
1303  * Invoked by a work queue tasklet.
1304  */
1305 static void xs_tcp_connect_worker(struct work_struct *work)
1306 {
1307         struct sock_xprt *transport =
1308                 container_of(work, struct sock_xprt, connect_worker.work);
1309         struct rpc_xprt *xprt = &transport->xprt;
1310         struct socket *sock = transport->sock;
1311         int err, status = -EIO;
1312
1313         if (xprt->shutdown || !xprt_bound(xprt))
1314                 goto out;
1315
1316         if (!sock) {
1317                 /* start from scratch */
1318                 if ((err = sock_create_kern(PF_INET, SOCK_STREAM, IPPROTO_TCP, &sock)) < 0) {
1319                         dprintk("RPC:       can't create TCP transport "
1320                                         "socket (%d).\n", -err);
1321                         goto out;
1322                 }
1323                 xs_reclassify_socket(sock);
1324
1325                 if (xs_bind(transport, sock)) {
1326                         sock_release(sock);
1327                         goto out;
1328                 }
1329         } else
1330                 /* "close" the socket, preserving the local port */
1331                 xs_tcp_reuse_connection(xprt);
1332
1333         dprintk("RPC:       worker connecting xprt %p to address: %s\n",
1334                         xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1335
1336         if (!transport->inet) {
1337                 struct sock *sk = sock->sk;
1338
1339                 write_lock_bh(&sk->sk_callback_lock);
1340
1341                 sk->sk_user_data = xprt;
1342                 transport->old_data_ready = sk->sk_data_ready;
1343                 transport->old_state_change = sk->sk_state_change;
1344                 transport->old_write_space = sk->sk_write_space;
1345                 sk->sk_data_ready = xs_tcp_data_ready;
1346                 sk->sk_state_change = xs_tcp_state_change;
1347                 sk->sk_write_space = xs_tcp_write_space;
1348                 sk->sk_allocation = GFP_ATOMIC;
1349
1350                 /* socket options */
1351                 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
1352                 sock_reset_flag(sk, SOCK_LINGER);
1353                 tcp_sk(sk)->linger2 = 0;
1354                 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
1355
1356                 xprt_clear_connected(xprt);
1357
1358                 /* Reset to new socket */
1359                 transport->sock = sock;
1360                 transport->inet = sk;
1361
1362                 write_unlock_bh(&sk->sk_callback_lock);
1363         }
1364
1365         /* Tell the socket layer to start connecting... */
1366         xprt->stat.connect_count++;
1367         xprt->stat.connect_start = jiffies;
1368         status = kernel_connect(sock, (struct sockaddr *) &xprt->addr,
1369                         xprt->addrlen, O_NONBLOCK);
1370         dprintk("RPC:       %p connect status %d connected %d sock state %d\n",
1371                         xprt, -status, xprt_connected(xprt),
1372                         sock->sk->sk_state);
1373         if (status < 0) {
1374                 switch (status) {
1375                         case -EINPROGRESS:
1376                         case -EALREADY:
1377                                 goto out_clear;
1378                         case -ECONNREFUSED:
1379                         case -ECONNRESET:
1380                                 /* retry with existing socket, after a delay */
1381                                 break;
1382                         default:
1383                                 /* get rid of existing socket, and retry */
1384                                 xs_close(xprt);
1385                                 break;
1386                 }
1387         }
1388 out:
1389         xprt_wake_pending_tasks(xprt, status);
1390 out_clear:
1391         xprt_clear_connecting(xprt);
1392 }
1393
1394 /**
1395  * xs_connect - connect a socket to a remote endpoint
1396  * @task: address of RPC task that manages state of connect request
1397  *
1398  * TCP: If the remote end dropped the connection, delay reconnecting.
1399  *
1400  * UDP socket connects are synchronous, but we use a work queue anyway
1401  * to guarantee that even unprivileged user processes can set up a
1402  * socket on a privileged port.
1403  *
1404  * If a UDP socket connect fails, the delay behavior here prevents
1405  * retry floods (hard mounts).
1406  */
1407 static void xs_connect(struct rpc_task *task)
1408 {
1409         struct rpc_xprt *xprt = task->tk_xprt;
1410         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1411
1412         if (xprt_test_and_set_connecting(xprt))
1413                 return;
1414
1415         if (transport->sock != NULL) {
1416                 dprintk("RPC:       xs_connect delayed xprt %p for %lu "
1417                                 "seconds\n",
1418                                 xprt, xprt->reestablish_timeout / HZ);
1419                 queue_delayed_work(rpciod_workqueue,
1420                                    &transport->connect_worker,
1421                                    xprt->reestablish_timeout);
1422                 xprt->reestablish_timeout <<= 1;
1423                 if (xprt->reestablish_timeout > XS_TCP_MAX_REEST_TO)
1424                         xprt->reestablish_timeout = XS_TCP_MAX_REEST_TO;
1425         } else {
1426                 dprintk("RPC:       xs_connect scheduled xprt %p\n", xprt);
1427                 queue_delayed_work(rpciod_workqueue,
1428                                    &transport->connect_worker, 0);
1429         }
1430 }
1431
1432 /**
1433  * xs_udp_print_stats - display UDP socket-specifc stats
1434  * @xprt: rpc_xprt struct containing statistics
1435  * @seq: output file
1436  *
1437  */
1438 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
1439 {
1440         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1441
1442         seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %Lu %Lu\n",
1443                         transport->port,
1444                         xprt->stat.bind_count,
1445                         xprt->stat.sends,
1446                         xprt->stat.recvs,
1447                         xprt->stat.bad_xids,
1448                         xprt->stat.req_u,
1449                         xprt->stat.bklog_u);
1450 }
1451
1452 /**
1453  * xs_tcp_print_stats - display TCP socket-specifc stats
1454  * @xprt: rpc_xprt struct containing statistics
1455  * @seq: output file
1456  *
1457  */
1458 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
1459 {
1460         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1461         long idle_time = 0;
1462
1463         if (xprt_connected(xprt))
1464                 idle_time = (long)(jiffies - xprt->last_used) / HZ;
1465
1466         seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu %Lu %Lu\n",
1467                         transport->port,
1468                         xprt->stat.bind_count,
1469                         xprt->stat.connect_count,
1470                         xprt->stat.connect_time,
1471                         idle_time,
1472                         xprt->stat.sends,
1473                         xprt->stat.recvs,
1474                         xprt->stat.bad_xids,
1475                         xprt->stat.req_u,
1476                         xprt->stat.bklog_u);
1477 }
1478
1479 static struct rpc_xprt_ops xs_udp_ops = {
1480         .set_buffer_size        = xs_udp_set_buffer_size,
1481         .reserve_xprt           = xprt_reserve_xprt_cong,
1482         .release_xprt           = xprt_release_xprt_cong,
1483         .rpcbind                = rpcb_getport_async,
1484         .set_port               = xs_set_port,
1485         .connect                = xs_connect,
1486         .buf_alloc              = rpc_malloc,
1487         .buf_free               = rpc_free,
1488         .send_request           = xs_udp_send_request,
1489         .set_retrans_timeout    = xprt_set_retrans_timeout_rtt,
1490         .timer                  = xs_udp_timer,
1491         .release_request        = xprt_release_rqst_cong,
1492         .close                  = xs_close,
1493         .destroy                = xs_destroy,
1494         .print_stats            = xs_udp_print_stats,
1495 };
1496
1497 static struct rpc_xprt_ops xs_tcp_ops = {
1498         .reserve_xprt           = xprt_reserve_xprt,
1499         .release_xprt           = xs_tcp_release_xprt,
1500         .rpcbind                = rpcb_getport_async,
1501         .set_port               = xs_set_port,
1502         .connect                = xs_connect,
1503         .buf_alloc              = rpc_malloc,
1504         .buf_free               = rpc_free,
1505         .send_request           = xs_tcp_send_request,
1506         .set_retrans_timeout    = xprt_set_retrans_timeout_def,
1507         .close                  = xs_close,
1508         .destroy                = xs_destroy,
1509         .print_stats            = xs_tcp_print_stats,
1510 };
1511
1512 static struct rpc_xprt *xs_setup_xprt(struct rpc_xprtsock_create *args, unsigned int slot_table_size)
1513 {
1514         struct rpc_xprt *xprt;
1515         struct sock_xprt *new;
1516
1517         if (args->addrlen > sizeof(xprt->addr)) {
1518                 dprintk("RPC:       xs_setup_xprt: address too large\n");
1519                 return ERR_PTR(-EBADF);
1520         }
1521
1522         new = kzalloc(sizeof(*new), GFP_KERNEL);
1523         if (new == NULL) {
1524                 dprintk("RPC:       xs_setup_xprt: couldn't allocate "
1525                                 "rpc_xprt\n");
1526                 return ERR_PTR(-ENOMEM);
1527         }
1528         xprt = &new->xprt;
1529
1530         xprt->max_reqs = slot_table_size;
1531         xprt->slot = kcalloc(xprt->max_reqs, sizeof(struct rpc_rqst), GFP_KERNEL);
1532         if (xprt->slot == NULL) {
1533                 kfree(xprt);
1534                 dprintk("RPC:       xs_setup_xprt: couldn't allocate slot "
1535                                 "table\n");
1536                 return ERR_PTR(-ENOMEM);
1537         }
1538
1539         memcpy(&xprt->addr, args->dstaddr, args->addrlen);
1540         xprt->addrlen = args->addrlen;
1541         if (args->srcaddr)
1542                 memcpy(&new->addr, args->srcaddr, args->addrlen);
1543         new->port = xs_get_random_port();
1544
1545         return xprt;
1546 }
1547
1548 /**
1549  * xs_setup_udp - Set up transport to use a UDP socket
1550  * @args: rpc transport creation arguments
1551  *
1552  */
1553 struct rpc_xprt *xs_setup_udp(struct rpc_xprtsock_create *args)
1554 {
1555         struct rpc_xprt *xprt;
1556         struct sock_xprt *transport;
1557
1558         xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries);
1559         if (IS_ERR(xprt))
1560                 return xprt;
1561         transport = container_of(xprt, struct sock_xprt, xprt);
1562
1563         if (ntohs(((struct sockaddr_in *)args->dstaddr)->sin_port) != 0)
1564                 xprt_set_bound(xprt);
1565
1566         xprt->prot = IPPROTO_UDP;
1567         xprt->tsh_size = 0;
1568         /* XXX: header size can vary due to auth type, IPv6, etc. */
1569         xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
1570
1571         INIT_DELAYED_WORK(&transport->connect_worker, xs_udp_connect_worker);
1572         xprt->bind_timeout = XS_BIND_TO;
1573         xprt->connect_timeout = XS_UDP_CONN_TO;
1574         xprt->reestablish_timeout = XS_UDP_REEST_TO;
1575         xprt->idle_timeout = XS_IDLE_DISC_TO;
1576
1577         xprt->ops = &xs_udp_ops;
1578
1579         if (args->timeout)
1580                 xprt->timeout = *args->timeout;
1581         else
1582                 xprt_set_timeout(&xprt->timeout, 5, 5 * HZ);
1583
1584         xs_format_peer_addresses(xprt);
1585         dprintk("RPC:       set up transport to address %s\n",
1586                         xprt->address_strings[RPC_DISPLAY_ALL]);
1587
1588         return xprt;
1589 }
1590
1591 /**
1592  * xs_setup_tcp - Set up transport to use a TCP socket
1593  * @args: rpc transport creation arguments
1594  *
1595  */
1596 struct rpc_xprt *xs_setup_tcp(struct rpc_xprtsock_create *args)
1597 {
1598         struct rpc_xprt *xprt;
1599         struct sock_xprt *transport;
1600
1601         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries);
1602         if (IS_ERR(xprt))
1603                 return xprt;
1604         transport = container_of(xprt, struct sock_xprt, xprt);
1605
1606         if (ntohs(((struct sockaddr_in *)args->dstaddr)->sin_port) != 0)
1607                 xprt_set_bound(xprt);
1608
1609         xprt->prot = IPPROTO_TCP;
1610         xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
1611         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
1612
1613         INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_connect_worker);
1614         xprt->bind_timeout = XS_BIND_TO;
1615         xprt->connect_timeout = XS_TCP_CONN_TO;
1616         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1617         xprt->idle_timeout = XS_IDLE_DISC_TO;
1618
1619         xprt->ops = &xs_tcp_ops;
1620
1621         if (args->timeout)
1622                 xprt->timeout = *args->timeout;
1623         else
1624                 xprt_set_timeout(&xprt->timeout, 2, 60 * HZ);
1625
1626         xs_format_peer_addresses(xprt);
1627         dprintk("RPC:       set up transport to address %s\n",
1628                         xprt->address_strings[RPC_DISPLAY_ALL]);
1629
1630         return xprt;
1631 }
1632
1633 /**
1634  * init_socket_xprt - set up xprtsock's sysctls
1635  *
1636  */
1637 int init_socket_xprt(void)
1638 {
1639 #ifdef RPC_DEBUG
1640         if (!sunrpc_table_header)
1641                 sunrpc_table_header = register_sysctl_table(sunrpc_table);
1642 #endif
1643
1644         return 0;
1645 }
1646
1647 /**
1648  * cleanup_socket_xprt - remove xprtsock's sysctls
1649  *
1650  */
1651 void cleanup_socket_xprt(void)
1652 {
1653 #ifdef RPC_DEBUG
1654         if (sunrpc_table_header) {
1655                 unregister_sysctl_table(sunrpc_table_header);
1656                 sunrpc_table_header = NULL;
1657         }
1658 #endif
1659 }