Merge branch 'for-3.10' of git://linux-nfs.org/~bfields/linux
[linux-3.10.git] / net / sunrpc / xprtsock.c
1 /*
2  * linux/net/sunrpc/xprtsock.c
3  *
4  * Client-side transport implementation for sockets.
5  *
6  * TCP callback races fixes (C) 1998 Red Hat
7  * TCP send fixes (C) 1998 Red Hat
8  * TCP NFS related read + write fixes
9  *  (C) 1999 Dave Airlie, University of Limerick, Ireland <airlied@linux.ie>
10  *
11  * Rewrite of larges part of the code in order to stabilize TCP stuff.
12  * Fix behaviour when socket buffer is full.
13  *  (C) 1999 Trond Myklebust <trond.myklebust@fys.uio.no>
14  *
15  * IP socket transport implementation, (C) 2005 Chuck Lever <cel@netapp.com>
16  *
17  * IPv6 support contributed by Gilles Quillard, Bull Open Source, 2005.
18  *   <gilles.quillard@bull.net>
19  */
20
21 #include <linux/types.h>
22 #include <linux/string.h>
23 #include <linux/slab.h>
24 #include <linux/module.h>
25 #include <linux/capability.h>
26 #include <linux/pagemap.h>
27 #include <linux/errno.h>
28 #include <linux/socket.h>
29 #include <linux/in.h>
30 #include <linux/net.h>
31 #include <linux/mm.h>
32 #include <linux/un.h>
33 #include <linux/udp.h>
34 #include <linux/tcp.h>
35 #include <linux/sunrpc/clnt.h>
36 #include <linux/sunrpc/addr.h>
37 #include <linux/sunrpc/sched.h>
38 #include <linux/sunrpc/svcsock.h>
39 #include <linux/sunrpc/xprtsock.h>
40 #include <linux/file.h>
41 #ifdef CONFIG_SUNRPC_BACKCHANNEL
42 #include <linux/sunrpc/bc_xprt.h>
43 #endif
44
45 #include <net/sock.h>
46 #include <net/checksum.h>
47 #include <net/udp.h>
48 #include <net/tcp.h>
49
50 #include "sunrpc.h"
51
52 static void xs_close(struct rpc_xprt *xprt);
53
54 /*
55  * xprtsock tunables
56  */
57 static unsigned int xprt_udp_slot_table_entries = RPC_DEF_SLOT_TABLE;
58 static unsigned int xprt_tcp_slot_table_entries = RPC_MIN_SLOT_TABLE;
59 static unsigned int xprt_max_tcp_slot_table_entries = RPC_MAX_SLOT_TABLE;
60
61 static unsigned int xprt_min_resvport = RPC_DEF_MIN_RESVPORT;
62 static unsigned int xprt_max_resvport = RPC_DEF_MAX_RESVPORT;
63
64 #define XS_TCP_LINGER_TO        (15U * HZ)
65 static unsigned int xs_tcp_fin_timeout __read_mostly = XS_TCP_LINGER_TO;
66
67 /*
68  * We can register our own files under /proc/sys/sunrpc by
69  * calling register_sysctl_table() again.  The files in that
70  * directory become the union of all files registered there.
71  *
72  * We simply need to make sure that we don't collide with
73  * someone else's file names!
74  */
75
76 #ifdef RPC_DEBUG
77
78 static unsigned int min_slot_table_size = RPC_MIN_SLOT_TABLE;
79 static unsigned int max_slot_table_size = RPC_MAX_SLOT_TABLE;
80 static unsigned int max_tcp_slot_table_limit = RPC_MAX_SLOT_TABLE_LIMIT;
81 static unsigned int xprt_min_resvport_limit = RPC_MIN_RESVPORT;
82 static unsigned int xprt_max_resvport_limit = RPC_MAX_RESVPORT;
83
84 static struct ctl_table_header *sunrpc_table_header;
85
86 /*
87  * FIXME: changing the UDP slot table size should also resize the UDP
88  *        socket buffers for existing UDP transports
89  */
90 static ctl_table xs_tunables_table[] = {
91         {
92                 .procname       = "udp_slot_table_entries",
93                 .data           = &xprt_udp_slot_table_entries,
94                 .maxlen         = sizeof(unsigned int),
95                 .mode           = 0644,
96                 .proc_handler   = proc_dointvec_minmax,
97                 .extra1         = &min_slot_table_size,
98                 .extra2         = &max_slot_table_size
99         },
100         {
101                 .procname       = "tcp_slot_table_entries",
102                 .data           = &xprt_tcp_slot_table_entries,
103                 .maxlen         = sizeof(unsigned int),
104                 .mode           = 0644,
105                 .proc_handler   = proc_dointvec_minmax,
106                 .extra1         = &min_slot_table_size,
107                 .extra2         = &max_slot_table_size
108         },
109         {
110                 .procname       = "tcp_max_slot_table_entries",
111                 .data           = &xprt_max_tcp_slot_table_entries,
112                 .maxlen         = sizeof(unsigned int),
113                 .mode           = 0644,
114                 .proc_handler   = proc_dointvec_minmax,
115                 .extra1         = &min_slot_table_size,
116                 .extra2         = &max_tcp_slot_table_limit
117         },
118         {
119                 .procname       = "min_resvport",
120                 .data           = &xprt_min_resvport,
121                 .maxlen         = sizeof(unsigned int),
122                 .mode           = 0644,
123                 .proc_handler   = proc_dointvec_minmax,
124                 .extra1         = &xprt_min_resvport_limit,
125                 .extra2         = &xprt_max_resvport_limit
126         },
127         {
128                 .procname       = "max_resvport",
129                 .data           = &xprt_max_resvport,
130                 .maxlen         = sizeof(unsigned int),
131                 .mode           = 0644,
132                 .proc_handler   = proc_dointvec_minmax,
133                 .extra1         = &xprt_min_resvport_limit,
134                 .extra2         = &xprt_max_resvport_limit
135         },
136         {
137                 .procname       = "tcp_fin_timeout",
138                 .data           = &xs_tcp_fin_timeout,
139                 .maxlen         = sizeof(xs_tcp_fin_timeout),
140                 .mode           = 0644,
141                 .proc_handler   = proc_dointvec_jiffies,
142         },
143         { },
144 };
145
146 static ctl_table sunrpc_table[] = {
147         {
148                 .procname       = "sunrpc",
149                 .mode           = 0555,
150                 .child          = xs_tunables_table
151         },
152         { },
153 };
154
155 #endif
156
157 /*
158  * Wait duration for a reply from the RPC portmapper.
159  */
160 #define XS_BIND_TO              (60U * HZ)
161
162 /*
163  * Delay if a UDP socket connect error occurs.  This is most likely some
164  * kind of resource problem on the local host.
165  */
166 #define XS_UDP_REEST_TO         (2U * HZ)
167
168 /*
169  * The reestablish timeout allows clients to delay for a bit before attempting
170  * to reconnect to a server that just dropped our connection.
171  *
172  * We implement an exponential backoff when trying to reestablish a TCP
173  * transport connection with the server.  Some servers like to drop a TCP
174  * connection when they are overworked, so we start with a short timeout and
175  * increase over time if the server is down or not responding.
176  */
177 #define XS_TCP_INIT_REEST_TO    (3U * HZ)
178 #define XS_TCP_MAX_REEST_TO     (5U * 60 * HZ)
179
180 /*
181  * TCP idle timeout; client drops the transport socket if it is idle
182  * for this long.  Note that we also timeout UDP sockets to prevent
183  * holding port numbers when there is no RPC traffic.
184  */
185 #define XS_IDLE_DISC_TO         (5U * 60 * HZ)
186
187 #ifdef RPC_DEBUG
188 # undef  RPC_DEBUG_DATA
189 # define RPCDBG_FACILITY        RPCDBG_TRANS
190 #endif
191
192 #ifdef RPC_DEBUG_DATA
193 static void xs_pktdump(char *msg, u32 *packet, unsigned int count)
194 {
195         u8 *buf = (u8 *) packet;
196         int j;
197
198         dprintk("RPC:       %s\n", msg);
199         for (j = 0; j < count && j < 128; j += 4) {
200                 if (!(j & 31)) {
201                         if (j)
202                                 dprintk("\n");
203                         dprintk("0x%04x ", j);
204                 }
205                 dprintk("%02x%02x%02x%02x ",
206                         buf[j], buf[j+1], buf[j+2], buf[j+3]);
207         }
208         dprintk("\n");
209 }
210 #else
211 static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count)
212 {
213         /* NOP */
214 }
215 #endif
216
217 struct sock_xprt {
218         struct rpc_xprt         xprt;
219
220         /*
221          * Network layer
222          */
223         struct socket *         sock;
224         struct sock *           inet;
225
226         /*
227          * State of TCP reply receive
228          */
229         __be32                  tcp_fraghdr,
230                                 tcp_xid,
231                                 tcp_calldir;
232
233         u32                     tcp_offset,
234                                 tcp_reclen;
235
236         unsigned long           tcp_copied,
237                                 tcp_flags;
238
239         /*
240          * Connection of transports
241          */
242         struct delayed_work     connect_worker;
243         struct sockaddr_storage srcaddr;
244         unsigned short          srcport;
245
246         /*
247          * UDP socket buffer size parameters
248          */
249         size_t                  rcvsize,
250                                 sndsize;
251
252         /*
253          * Saved socket callback addresses
254          */
255         void                    (*old_data_ready)(struct sock *, int);
256         void                    (*old_state_change)(struct sock *);
257         void                    (*old_write_space)(struct sock *);
258 };
259
260 /*
261  * TCP receive state flags
262  */
263 #define TCP_RCV_LAST_FRAG       (1UL << 0)
264 #define TCP_RCV_COPY_FRAGHDR    (1UL << 1)
265 #define TCP_RCV_COPY_XID        (1UL << 2)
266 #define TCP_RCV_COPY_DATA       (1UL << 3)
267 #define TCP_RCV_READ_CALLDIR    (1UL << 4)
268 #define TCP_RCV_COPY_CALLDIR    (1UL << 5)
269
270 /*
271  * TCP RPC flags
272  */
273 #define TCP_RPC_REPLY           (1UL << 6)
274
275 static inline struct sockaddr *xs_addr(struct rpc_xprt *xprt)
276 {
277         return (struct sockaddr *) &xprt->addr;
278 }
279
280 static inline struct sockaddr_un *xs_addr_un(struct rpc_xprt *xprt)
281 {
282         return (struct sockaddr_un *) &xprt->addr;
283 }
284
285 static inline struct sockaddr_in *xs_addr_in(struct rpc_xprt *xprt)
286 {
287         return (struct sockaddr_in *) &xprt->addr;
288 }
289
290 static inline struct sockaddr_in6 *xs_addr_in6(struct rpc_xprt *xprt)
291 {
292         return (struct sockaddr_in6 *) &xprt->addr;
293 }
294
295 static void xs_format_common_peer_addresses(struct rpc_xprt *xprt)
296 {
297         struct sockaddr *sap = xs_addr(xprt);
298         struct sockaddr_in6 *sin6;
299         struct sockaddr_in *sin;
300         struct sockaddr_un *sun;
301         char buf[128];
302
303         switch (sap->sa_family) {
304         case AF_LOCAL:
305                 sun = xs_addr_un(xprt);
306                 strlcpy(buf, sun->sun_path, sizeof(buf));
307                 xprt->address_strings[RPC_DISPLAY_ADDR] =
308                                                 kstrdup(buf, GFP_KERNEL);
309                 break;
310         case AF_INET:
311                 (void)rpc_ntop(sap, buf, sizeof(buf));
312                 xprt->address_strings[RPC_DISPLAY_ADDR] =
313                                                 kstrdup(buf, GFP_KERNEL);
314                 sin = xs_addr_in(xprt);
315                 snprintf(buf, sizeof(buf), "%08x", ntohl(sin->sin_addr.s_addr));
316                 break;
317         case AF_INET6:
318                 (void)rpc_ntop(sap, buf, sizeof(buf));
319                 xprt->address_strings[RPC_DISPLAY_ADDR] =
320                                                 kstrdup(buf, GFP_KERNEL);
321                 sin6 = xs_addr_in6(xprt);
322                 snprintf(buf, sizeof(buf), "%pi6", &sin6->sin6_addr);
323                 break;
324         default:
325                 BUG();
326         }
327
328         xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = kstrdup(buf, GFP_KERNEL);
329 }
330
331 static void xs_format_common_peer_ports(struct rpc_xprt *xprt)
332 {
333         struct sockaddr *sap = xs_addr(xprt);
334         char buf[128];
335
336         snprintf(buf, sizeof(buf), "%u", rpc_get_port(sap));
337         xprt->address_strings[RPC_DISPLAY_PORT] = kstrdup(buf, GFP_KERNEL);
338
339         snprintf(buf, sizeof(buf), "%4hx", rpc_get_port(sap));
340         xprt->address_strings[RPC_DISPLAY_HEX_PORT] = kstrdup(buf, GFP_KERNEL);
341 }
342
343 static void xs_format_peer_addresses(struct rpc_xprt *xprt,
344                                      const char *protocol,
345                                      const char *netid)
346 {
347         xprt->address_strings[RPC_DISPLAY_PROTO] = protocol;
348         xprt->address_strings[RPC_DISPLAY_NETID] = netid;
349         xs_format_common_peer_addresses(xprt);
350         xs_format_common_peer_ports(xprt);
351 }
352
353 static void xs_update_peer_port(struct rpc_xprt *xprt)
354 {
355         kfree(xprt->address_strings[RPC_DISPLAY_HEX_PORT]);
356         kfree(xprt->address_strings[RPC_DISPLAY_PORT]);
357
358         xs_format_common_peer_ports(xprt);
359 }
360
361 static void xs_free_peer_addresses(struct rpc_xprt *xprt)
362 {
363         unsigned int i;
364
365         for (i = 0; i < RPC_DISPLAY_MAX; i++)
366                 switch (i) {
367                 case RPC_DISPLAY_PROTO:
368                 case RPC_DISPLAY_NETID:
369                         continue;
370                 default:
371                         kfree(xprt->address_strings[i]);
372                 }
373 }
374
375 #define XS_SENDMSG_FLAGS        (MSG_DONTWAIT | MSG_NOSIGNAL)
376
377 static int xs_send_kvec(struct socket *sock, struct sockaddr *addr, int addrlen, struct kvec *vec, unsigned int base, int more)
378 {
379         struct msghdr msg = {
380                 .msg_name       = addr,
381                 .msg_namelen    = addrlen,
382                 .msg_flags      = XS_SENDMSG_FLAGS | (more ? MSG_MORE : 0),
383         };
384         struct kvec iov = {
385                 .iov_base       = vec->iov_base + base,
386                 .iov_len        = vec->iov_len - base,
387         };
388
389         if (iov.iov_len != 0)
390                 return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
391         return kernel_sendmsg(sock, &msg, NULL, 0, 0);
392 }
393
394 static int xs_send_pagedata(struct socket *sock, struct xdr_buf *xdr, unsigned int base, int more)
395 {
396         struct page **ppage;
397         unsigned int remainder;
398         int err, sent = 0;
399
400         remainder = xdr->page_len - base;
401         base += xdr->page_base;
402         ppage = xdr->pages + (base >> PAGE_SHIFT);
403         base &= ~PAGE_MASK;
404         for(;;) {
405                 unsigned int len = min_t(unsigned int, PAGE_SIZE - base, remainder);
406                 int flags = XS_SENDMSG_FLAGS;
407
408                 remainder -= len;
409                 if (remainder != 0 || more)
410                         flags |= MSG_MORE;
411                 err = sock->ops->sendpage(sock, *ppage, base, len, flags);
412                 if (remainder == 0 || err != len)
413                         break;
414                 sent += err;
415                 ppage++;
416                 base = 0;
417         }
418         if (sent == 0)
419                 return err;
420         if (err > 0)
421                 sent += err;
422         return sent;
423 }
424
425 /**
426  * xs_sendpages - write pages directly to a socket
427  * @sock: socket to send on
428  * @addr: UDP only -- address of destination
429  * @addrlen: UDP only -- length of destination address
430  * @xdr: buffer containing this request
431  * @base: starting position in the buffer
432  *
433  */
434 static int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base)
435 {
436         unsigned int remainder = xdr->len - base;
437         int err, sent = 0;
438
439         if (unlikely(!sock))
440                 return -ENOTSOCK;
441
442         clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags);
443         if (base != 0) {
444                 addr = NULL;
445                 addrlen = 0;
446         }
447
448         if (base < xdr->head[0].iov_len || addr != NULL) {
449                 unsigned int len = xdr->head[0].iov_len - base;
450                 remainder -= len;
451                 err = xs_send_kvec(sock, addr, addrlen, &xdr->head[0], base, remainder != 0);
452                 if (remainder == 0 || err != len)
453                         goto out;
454                 sent += err;
455                 base = 0;
456         } else
457                 base -= xdr->head[0].iov_len;
458
459         if (base < xdr->page_len) {
460                 unsigned int len = xdr->page_len - base;
461                 remainder -= len;
462                 err = xs_send_pagedata(sock, xdr, base, remainder != 0);
463                 if (remainder == 0 || err != len)
464                         goto out;
465                 sent += err;
466                 base = 0;
467         } else
468                 base -= xdr->page_len;
469
470         if (base >= xdr->tail[0].iov_len)
471                 return sent;
472         err = xs_send_kvec(sock, NULL, 0, &xdr->tail[0], base, 0);
473 out:
474         if (sent == 0)
475                 return err;
476         if (err > 0)
477                 sent += err;
478         return sent;
479 }
480
481 static void xs_nospace_callback(struct rpc_task *task)
482 {
483         struct sock_xprt *transport = container_of(task->tk_rqstp->rq_xprt, struct sock_xprt, xprt);
484
485         transport->inet->sk_write_pending--;
486         clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
487 }
488
489 /**
490  * xs_nospace - place task on wait queue if transmit was incomplete
491  * @task: task to put to sleep
492  *
493  */
494 static int xs_nospace(struct rpc_task *task)
495 {
496         struct rpc_rqst *req = task->tk_rqstp;
497         struct rpc_xprt *xprt = req->rq_xprt;
498         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
499         int ret = -EAGAIN;
500
501         dprintk("RPC: %5u xmit incomplete (%u left of %u)\n",
502                         task->tk_pid, req->rq_slen - req->rq_bytes_sent,
503                         req->rq_slen);
504
505         /* Protect against races with write_space */
506         spin_lock_bh(&xprt->transport_lock);
507
508         /* Don't race with disconnect */
509         if (xprt_connected(xprt)) {
510                 if (test_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags)) {
511                         /*
512                          * Notify TCP that we're limited by the application
513                          * window size
514                          */
515                         set_bit(SOCK_NOSPACE, &transport->sock->flags);
516                         transport->inet->sk_write_pending++;
517                         /* ...and wait for more buffer space */
518                         xprt_wait_for_buffer_space(task, xs_nospace_callback);
519                 }
520         } else {
521                 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
522                 ret = -ENOTCONN;
523         }
524
525         spin_unlock_bh(&xprt->transport_lock);
526         return ret;
527 }
528
529 /*
530  * Construct a stream transport record marker in @buf.
531  */
532 static inline void xs_encode_stream_record_marker(struct xdr_buf *buf)
533 {
534         u32 reclen = buf->len - sizeof(rpc_fraghdr);
535         rpc_fraghdr *base = buf->head[0].iov_base;
536         *base = cpu_to_be32(RPC_LAST_STREAM_FRAGMENT | reclen);
537 }
538
539 /**
540  * xs_local_send_request - write an RPC request to an AF_LOCAL socket
541  * @task: RPC task that manages the state of an RPC request
542  *
543  * Return values:
544  *        0:    The request has been sent
545  *   EAGAIN:    The socket was blocked, please call again later to
546  *              complete the request
547  * ENOTCONN:    Caller needs to invoke connect logic then call again
548  *    other:    Some other error occured, the request was not sent
549  */
550 static int xs_local_send_request(struct rpc_task *task)
551 {
552         struct rpc_rqst *req = task->tk_rqstp;
553         struct rpc_xprt *xprt = req->rq_xprt;
554         struct sock_xprt *transport =
555                                 container_of(xprt, struct sock_xprt, xprt);
556         struct xdr_buf *xdr = &req->rq_snd_buf;
557         int status;
558
559         xs_encode_stream_record_marker(&req->rq_snd_buf);
560
561         xs_pktdump("packet data:",
562                         req->rq_svec->iov_base, req->rq_svec->iov_len);
563
564         status = xs_sendpages(transport->sock, NULL, 0,
565                                                 xdr, req->rq_bytes_sent);
566         dprintk("RPC:       %s(%u) = %d\n",
567                         __func__, xdr->len - req->rq_bytes_sent, status);
568         if (likely(status >= 0)) {
569                 req->rq_bytes_sent += status;
570                 req->rq_xmit_bytes_sent += status;
571                 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
572                         req->rq_bytes_sent = 0;
573                         return 0;
574                 }
575                 status = -EAGAIN;
576         }
577
578         switch (status) {
579         case -EAGAIN:
580                 status = xs_nospace(task);
581                 break;
582         default:
583                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
584                         -status);
585         case -EPIPE:
586                 xs_close(xprt);
587                 status = -ENOTCONN;
588         }
589
590         return status;
591 }
592
593 /**
594  * xs_udp_send_request - write an RPC request to a UDP socket
595  * @task: address of RPC task that manages the state of an RPC request
596  *
597  * Return values:
598  *        0:    The request has been sent
599  *   EAGAIN:    The socket was blocked, please call again later to
600  *              complete the request
601  * ENOTCONN:    Caller needs to invoke connect logic then call again
602  *    other:    Some other error occurred, the request was not sent
603  */
604 static int xs_udp_send_request(struct rpc_task *task)
605 {
606         struct rpc_rqst *req = task->tk_rqstp;
607         struct rpc_xprt *xprt = req->rq_xprt;
608         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
609         struct xdr_buf *xdr = &req->rq_snd_buf;
610         int status;
611
612         xs_pktdump("packet data:",
613                                 req->rq_svec->iov_base,
614                                 req->rq_svec->iov_len);
615
616         if (!xprt_bound(xprt))
617                 return -ENOTCONN;
618         status = xs_sendpages(transport->sock,
619                               xs_addr(xprt),
620                               xprt->addrlen, xdr,
621                               req->rq_bytes_sent);
622
623         dprintk("RPC:       xs_udp_send_request(%u) = %d\n",
624                         xdr->len - req->rq_bytes_sent, status);
625
626         if (status >= 0) {
627                 req->rq_xmit_bytes_sent += status;
628                 if (status >= req->rq_slen)
629                         return 0;
630                 /* Still some bytes left; set up for a retry later. */
631                 status = -EAGAIN;
632         }
633
634         switch (status) {
635         case -ENOTSOCK:
636                 status = -ENOTCONN;
637                 /* Should we call xs_close() here? */
638                 break;
639         case -EAGAIN:
640                 status = xs_nospace(task);
641                 break;
642         default:
643                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
644                         -status);
645         case -ENETUNREACH:
646         case -EPIPE:
647         case -ECONNREFUSED:
648                 /* When the server has died, an ICMP port unreachable message
649                  * prompts ECONNREFUSED. */
650                 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
651         }
652
653         return status;
654 }
655
656 /**
657  * xs_tcp_shutdown - gracefully shut down a TCP socket
658  * @xprt: transport
659  *
660  * Initiates a graceful shutdown of the TCP socket by calling the
661  * equivalent of shutdown(SHUT_WR);
662  */
663 static void xs_tcp_shutdown(struct rpc_xprt *xprt)
664 {
665         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
666         struct socket *sock = transport->sock;
667
668         if (sock != NULL)
669                 kernel_sock_shutdown(sock, SHUT_WR);
670 }
671
672 /**
673  * xs_tcp_send_request - write an RPC request to a TCP socket
674  * @task: address of RPC task that manages the state of an RPC request
675  *
676  * Return values:
677  *        0:    The request has been sent
678  *   EAGAIN:    The socket was blocked, please call again later to
679  *              complete the request
680  * ENOTCONN:    Caller needs to invoke connect logic then call again
681  *    other:    Some other error occurred, the request was not sent
682  *
683  * XXX: In the case of soft timeouts, should we eventually give up
684  *      if sendmsg is not able to make progress?
685  */
686 static int xs_tcp_send_request(struct rpc_task *task)
687 {
688         struct rpc_rqst *req = task->tk_rqstp;
689         struct rpc_xprt *xprt = req->rq_xprt;
690         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
691         struct xdr_buf *xdr = &req->rq_snd_buf;
692         int status;
693
694         xs_encode_stream_record_marker(&req->rq_snd_buf);
695
696         xs_pktdump("packet data:",
697                                 req->rq_svec->iov_base,
698                                 req->rq_svec->iov_len);
699
700         /* Continue transmitting the packet/record. We must be careful
701          * to cope with writespace callbacks arriving _after_ we have
702          * called sendmsg(). */
703         while (1) {
704                 status = xs_sendpages(transport->sock,
705                                         NULL, 0, xdr, req->rq_bytes_sent);
706
707                 dprintk("RPC:       xs_tcp_send_request(%u) = %d\n",
708                                 xdr->len - req->rq_bytes_sent, status);
709
710                 if (unlikely(status < 0))
711                         break;
712
713                 /* If we've sent the entire packet, immediately
714                  * reset the count of bytes sent. */
715                 req->rq_bytes_sent += status;
716                 req->rq_xmit_bytes_sent += status;
717                 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
718                         req->rq_bytes_sent = 0;
719                         return 0;
720                 }
721
722                 if (status != 0)
723                         continue;
724                 status = -EAGAIN;
725                 break;
726         }
727
728         switch (status) {
729         case -ENOTSOCK:
730                 status = -ENOTCONN;
731                 /* Should we call xs_close() here? */
732                 break;
733         case -EAGAIN:
734                 status = xs_nospace(task);
735                 break;
736         default:
737                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
738                         -status);
739         case -ECONNRESET:
740                 xs_tcp_shutdown(xprt);
741         case -ECONNREFUSED:
742         case -ENOTCONN:
743         case -EPIPE:
744                 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
745         }
746
747         return status;
748 }
749
750 /**
751  * xs_tcp_release_xprt - clean up after a tcp transmission
752  * @xprt: transport
753  * @task: rpc task
754  *
755  * This cleans up if an error causes us to abort the transmission of a request.
756  * In this case, the socket may need to be reset in order to avoid confusing
757  * the server.
758  */
759 static void xs_tcp_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
760 {
761         struct rpc_rqst *req;
762
763         if (task != xprt->snd_task)
764                 return;
765         if (task == NULL)
766                 goto out_release;
767         req = task->tk_rqstp;
768         if (req == NULL)
769                 goto out_release;
770         if (req->rq_bytes_sent == 0)
771                 goto out_release;
772         if (req->rq_bytes_sent == req->rq_snd_buf.len)
773                 goto out_release;
774         set_bit(XPRT_CLOSE_WAIT, &xprt->state);
775 out_release:
776         xprt_release_xprt(xprt, task);
777 }
778
779 static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk)
780 {
781         transport->old_data_ready = sk->sk_data_ready;
782         transport->old_state_change = sk->sk_state_change;
783         transport->old_write_space = sk->sk_write_space;
784 }
785
786 static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk)
787 {
788         sk->sk_data_ready = transport->old_data_ready;
789         sk->sk_state_change = transport->old_state_change;
790         sk->sk_write_space = transport->old_write_space;
791 }
792
793 static void xs_reset_transport(struct sock_xprt *transport)
794 {
795         struct socket *sock = transport->sock;
796         struct sock *sk = transport->inet;
797
798         if (sk == NULL)
799                 return;
800
801         transport->srcport = 0;
802
803         write_lock_bh(&sk->sk_callback_lock);
804         transport->inet = NULL;
805         transport->sock = NULL;
806
807         sk->sk_user_data = NULL;
808
809         xs_restore_old_callbacks(transport, sk);
810         write_unlock_bh(&sk->sk_callback_lock);
811
812         sk->sk_no_check = 0;
813
814         sock_release(sock);
815 }
816
817 /**
818  * xs_close - close a socket
819  * @xprt: transport
820  *
821  * This is used when all requests are complete; ie, no DRC state remains
822  * on the server we want to save.
823  *
824  * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with
825  * xs_reset_transport() zeroing the socket from underneath a writer.
826  */
827 static void xs_close(struct rpc_xprt *xprt)
828 {
829         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
830
831         dprintk("RPC:       xs_close xprt %p\n", xprt);
832
833         xs_reset_transport(transport);
834         xprt->reestablish_timeout = 0;
835
836         smp_mb__before_clear_bit();
837         clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
838         clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
839         clear_bit(XPRT_CLOSING, &xprt->state);
840         smp_mb__after_clear_bit();
841         xprt_disconnect_done(xprt);
842 }
843
844 static void xs_tcp_close(struct rpc_xprt *xprt)
845 {
846         if (test_and_clear_bit(XPRT_CONNECTION_CLOSE, &xprt->state))
847                 xs_close(xprt);
848         else
849                 xs_tcp_shutdown(xprt);
850 }
851
852 static void xs_local_destroy(struct rpc_xprt *xprt)
853 {
854         xs_close(xprt);
855         xs_free_peer_addresses(xprt);
856         xprt_free(xprt);
857         module_put(THIS_MODULE);
858 }
859
860 /**
861  * xs_destroy - prepare to shutdown a transport
862  * @xprt: doomed transport
863  *
864  */
865 static void xs_destroy(struct rpc_xprt *xprt)
866 {
867         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
868
869         dprintk("RPC:       xs_destroy xprt %p\n", xprt);
870
871         cancel_delayed_work_sync(&transport->connect_worker);
872
873         xs_local_destroy(xprt);
874 }
875
876 static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
877 {
878         return (struct rpc_xprt *) sk->sk_user_data;
879 }
880
881 static int xs_local_copy_to_xdr(struct xdr_buf *xdr, struct sk_buff *skb)
882 {
883         struct xdr_skb_reader desc = {
884                 .skb            = skb,
885                 .offset         = sizeof(rpc_fraghdr),
886                 .count          = skb->len - sizeof(rpc_fraghdr),
887         };
888
889         if (xdr_partial_copy_from_skb(xdr, 0, &desc, xdr_skb_read_bits) < 0)
890                 return -1;
891         if (desc.count)
892                 return -1;
893         return 0;
894 }
895
896 /**
897  * xs_local_data_ready - "data ready" callback for AF_LOCAL sockets
898  * @sk: socket with data to read
899  * @len: how much data to read
900  *
901  * Currently this assumes we can read the whole reply in a single gulp.
902  */
903 static void xs_local_data_ready(struct sock *sk, int len)
904 {
905         struct rpc_task *task;
906         struct rpc_xprt *xprt;
907         struct rpc_rqst *rovr;
908         struct sk_buff *skb;
909         int err, repsize, copied;
910         u32 _xid;
911         __be32 *xp;
912
913         read_lock_bh(&sk->sk_callback_lock);
914         dprintk("RPC:       %s...\n", __func__);
915         xprt = xprt_from_sock(sk);
916         if (xprt == NULL)
917                 goto out;
918
919         skb = skb_recv_datagram(sk, 0, 1, &err);
920         if (skb == NULL)
921                 goto out;
922
923         repsize = skb->len - sizeof(rpc_fraghdr);
924         if (repsize < 4) {
925                 dprintk("RPC:       impossible RPC reply size %d\n", repsize);
926                 goto dropit;
927         }
928
929         /* Copy the XID from the skb... */
930         xp = skb_header_pointer(skb, sizeof(rpc_fraghdr), sizeof(_xid), &_xid);
931         if (xp == NULL)
932                 goto dropit;
933
934         /* Look up and lock the request corresponding to the given XID */
935         spin_lock(&xprt->transport_lock);
936         rovr = xprt_lookup_rqst(xprt, *xp);
937         if (!rovr)
938                 goto out_unlock;
939         task = rovr->rq_task;
940
941         copied = rovr->rq_private_buf.buflen;
942         if (copied > repsize)
943                 copied = repsize;
944
945         if (xs_local_copy_to_xdr(&rovr->rq_private_buf, skb)) {
946                 dprintk("RPC:       sk_buff copy failed\n");
947                 goto out_unlock;
948         }
949
950         xprt_complete_rqst(task, copied);
951
952  out_unlock:
953         spin_unlock(&xprt->transport_lock);
954  dropit:
955         skb_free_datagram(sk, skb);
956  out:
957         read_unlock_bh(&sk->sk_callback_lock);
958 }
959
960 /**
961  * xs_udp_data_ready - "data ready" callback for UDP sockets
962  * @sk: socket with data to read
963  * @len: how much data to read
964  *
965  */
966 static void xs_udp_data_ready(struct sock *sk, int len)
967 {
968         struct rpc_task *task;
969         struct rpc_xprt *xprt;
970         struct rpc_rqst *rovr;
971         struct sk_buff *skb;
972         int err, repsize, copied;
973         u32 _xid;
974         __be32 *xp;
975
976         read_lock_bh(&sk->sk_callback_lock);
977         dprintk("RPC:       xs_udp_data_ready...\n");
978         if (!(xprt = xprt_from_sock(sk)))
979                 goto out;
980
981         if ((skb = skb_recv_datagram(sk, 0, 1, &err)) == NULL)
982                 goto out;
983
984         repsize = skb->len - sizeof(struct udphdr);
985         if (repsize < 4) {
986                 dprintk("RPC:       impossible RPC reply size %d!\n", repsize);
987                 goto dropit;
988         }
989
990         /* Copy the XID from the skb... */
991         xp = skb_header_pointer(skb, sizeof(struct udphdr),
992                                 sizeof(_xid), &_xid);
993         if (xp == NULL)
994                 goto dropit;
995
996         /* Look up and lock the request corresponding to the given XID */
997         spin_lock(&xprt->transport_lock);
998         rovr = xprt_lookup_rqst(xprt, *xp);
999         if (!rovr)
1000                 goto out_unlock;
1001         task = rovr->rq_task;
1002
1003         if ((copied = rovr->rq_private_buf.buflen) > repsize)
1004                 copied = repsize;
1005
1006         /* Suck it into the iovec, verify checksum if not done by hw. */
1007         if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
1008                 UDPX_INC_STATS_BH(sk, UDP_MIB_INERRORS);
1009                 goto out_unlock;
1010         }
1011
1012         UDPX_INC_STATS_BH(sk, UDP_MIB_INDATAGRAMS);
1013
1014         xprt_adjust_cwnd(xprt, task, copied);
1015         xprt_complete_rqst(task, copied);
1016
1017  out_unlock:
1018         spin_unlock(&xprt->transport_lock);
1019  dropit:
1020         skb_free_datagram(sk, skb);
1021  out:
1022         read_unlock_bh(&sk->sk_callback_lock);
1023 }
1024
1025 /*
1026  * Helper function to force a TCP close if the server is sending
1027  * junk and/or it has put us in CLOSE_WAIT
1028  */
1029 static void xs_tcp_force_close(struct rpc_xprt *xprt)
1030 {
1031         set_bit(XPRT_CONNECTION_CLOSE, &xprt->state);
1032         xprt_force_disconnect(xprt);
1033 }
1034
1035 static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
1036 {
1037         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1038         size_t len, used;
1039         char *p;
1040
1041         p = ((char *) &transport->tcp_fraghdr) + transport->tcp_offset;
1042         len = sizeof(transport->tcp_fraghdr) - transport->tcp_offset;
1043         used = xdr_skb_read_bits(desc, p, len);
1044         transport->tcp_offset += used;
1045         if (used != len)
1046                 return;
1047
1048         transport->tcp_reclen = ntohl(transport->tcp_fraghdr);
1049         if (transport->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
1050                 transport->tcp_flags |= TCP_RCV_LAST_FRAG;
1051         else
1052                 transport->tcp_flags &= ~TCP_RCV_LAST_FRAG;
1053         transport->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
1054
1055         transport->tcp_flags &= ~TCP_RCV_COPY_FRAGHDR;
1056         transport->tcp_offset = 0;
1057
1058         /* Sanity check of the record length */
1059         if (unlikely(transport->tcp_reclen < 8)) {
1060                 dprintk("RPC:       invalid TCP record fragment length\n");
1061                 xs_tcp_force_close(xprt);
1062                 return;
1063         }
1064         dprintk("RPC:       reading TCP record fragment of length %d\n",
1065                         transport->tcp_reclen);
1066 }
1067
1068 static void xs_tcp_check_fraghdr(struct sock_xprt *transport)
1069 {
1070         if (transport->tcp_offset == transport->tcp_reclen) {
1071                 transport->tcp_flags |= TCP_RCV_COPY_FRAGHDR;
1072                 transport->tcp_offset = 0;
1073                 if (transport->tcp_flags & TCP_RCV_LAST_FRAG) {
1074                         transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1075                         transport->tcp_flags |= TCP_RCV_COPY_XID;
1076                         transport->tcp_copied = 0;
1077                 }
1078         }
1079 }
1080
1081 static inline void xs_tcp_read_xid(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1082 {
1083         size_t len, used;
1084         char *p;
1085
1086         len = sizeof(transport->tcp_xid) - transport->tcp_offset;
1087         dprintk("RPC:       reading XID (%Zu bytes)\n", len);
1088         p = ((char *) &transport->tcp_xid) + transport->tcp_offset;
1089         used = xdr_skb_read_bits(desc, p, len);
1090         transport->tcp_offset += used;
1091         if (used != len)
1092                 return;
1093         transport->tcp_flags &= ~TCP_RCV_COPY_XID;
1094         transport->tcp_flags |= TCP_RCV_READ_CALLDIR;
1095         transport->tcp_copied = 4;
1096         dprintk("RPC:       reading %s XID %08x\n",
1097                         (transport->tcp_flags & TCP_RPC_REPLY) ? "reply for"
1098                                                               : "request with",
1099                         ntohl(transport->tcp_xid));
1100         xs_tcp_check_fraghdr(transport);
1101 }
1102
1103 static inline void xs_tcp_read_calldir(struct sock_xprt *transport,
1104                                        struct xdr_skb_reader *desc)
1105 {
1106         size_t len, used;
1107         u32 offset;
1108         char *p;
1109
1110         /*
1111          * We want transport->tcp_offset to be 8 at the end of this routine
1112          * (4 bytes for the xid and 4 bytes for the call/reply flag).
1113          * When this function is called for the first time,
1114          * transport->tcp_offset is 4 (after having already read the xid).
1115          */
1116         offset = transport->tcp_offset - sizeof(transport->tcp_xid);
1117         len = sizeof(transport->tcp_calldir) - offset;
1118         dprintk("RPC:       reading CALL/REPLY flag (%Zu bytes)\n", len);
1119         p = ((char *) &transport->tcp_calldir) + offset;
1120         used = xdr_skb_read_bits(desc, p, len);
1121         transport->tcp_offset += used;
1122         if (used != len)
1123                 return;
1124         transport->tcp_flags &= ~TCP_RCV_READ_CALLDIR;
1125         /*
1126          * We don't yet have the XDR buffer, so we will write the calldir
1127          * out after we get the buffer from the 'struct rpc_rqst'
1128          */
1129         switch (ntohl(transport->tcp_calldir)) {
1130         case RPC_REPLY:
1131                 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1132                 transport->tcp_flags |= TCP_RCV_COPY_DATA;
1133                 transport->tcp_flags |= TCP_RPC_REPLY;
1134                 break;
1135         case RPC_CALL:
1136                 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1137                 transport->tcp_flags |= TCP_RCV_COPY_DATA;
1138                 transport->tcp_flags &= ~TCP_RPC_REPLY;
1139                 break;
1140         default:
1141                 dprintk("RPC:       invalid request message type\n");
1142                 xs_tcp_force_close(&transport->xprt);
1143         }
1144         xs_tcp_check_fraghdr(transport);
1145 }
1146
1147 static inline void xs_tcp_read_common(struct rpc_xprt *xprt,
1148                                      struct xdr_skb_reader *desc,
1149                                      struct rpc_rqst *req)
1150 {
1151         struct sock_xprt *transport =
1152                                 container_of(xprt, struct sock_xprt, xprt);
1153         struct xdr_buf *rcvbuf;
1154         size_t len;
1155         ssize_t r;
1156
1157         rcvbuf = &req->rq_private_buf;
1158
1159         if (transport->tcp_flags & TCP_RCV_COPY_CALLDIR) {
1160                 /*
1161                  * Save the RPC direction in the XDR buffer
1162                  */
1163                 memcpy(rcvbuf->head[0].iov_base + transport->tcp_copied,
1164                         &transport->tcp_calldir,
1165                         sizeof(transport->tcp_calldir));
1166                 transport->tcp_copied += sizeof(transport->tcp_calldir);
1167                 transport->tcp_flags &= ~TCP_RCV_COPY_CALLDIR;
1168         }
1169
1170         len = desc->count;
1171         if (len > transport->tcp_reclen - transport->tcp_offset) {
1172                 struct xdr_skb_reader my_desc;
1173
1174                 len = transport->tcp_reclen - transport->tcp_offset;
1175                 memcpy(&my_desc, desc, sizeof(my_desc));
1176                 my_desc.count = len;
1177                 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
1178                                           &my_desc, xdr_skb_read_bits);
1179                 desc->count -= r;
1180                 desc->offset += r;
1181         } else
1182                 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
1183                                           desc, xdr_skb_read_bits);
1184
1185         if (r > 0) {
1186                 transport->tcp_copied += r;
1187                 transport->tcp_offset += r;
1188         }
1189         if (r != len) {
1190                 /* Error when copying to the receive buffer,
1191                  * usually because we weren't able to allocate
1192                  * additional buffer pages. All we can do now
1193                  * is turn off TCP_RCV_COPY_DATA, so the request
1194                  * will not receive any additional updates,
1195                  * and time out.
1196                  * Any remaining data from this record will
1197                  * be discarded.
1198                  */
1199                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1200                 dprintk("RPC:       XID %08x truncated request\n",
1201                                 ntohl(transport->tcp_xid));
1202                 dprintk("RPC:       xprt = %p, tcp_copied = %lu, "
1203                                 "tcp_offset = %u, tcp_reclen = %u\n",
1204                                 xprt, transport->tcp_copied,
1205                                 transport->tcp_offset, transport->tcp_reclen);
1206                 return;
1207         }
1208
1209         dprintk("RPC:       XID %08x read %Zd bytes\n",
1210                         ntohl(transport->tcp_xid), r);
1211         dprintk("RPC:       xprt = %p, tcp_copied = %lu, tcp_offset = %u, "
1212                         "tcp_reclen = %u\n", xprt, transport->tcp_copied,
1213                         transport->tcp_offset, transport->tcp_reclen);
1214
1215         if (transport->tcp_copied == req->rq_private_buf.buflen)
1216                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1217         else if (transport->tcp_offset == transport->tcp_reclen) {
1218                 if (transport->tcp_flags & TCP_RCV_LAST_FRAG)
1219                         transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1220         }
1221 }
1222
1223 /*
1224  * Finds the request corresponding to the RPC xid and invokes the common
1225  * tcp read code to read the data.
1226  */
1227 static inline int xs_tcp_read_reply(struct rpc_xprt *xprt,
1228                                     struct xdr_skb_reader *desc)
1229 {
1230         struct sock_xprt *transport =
1231                                 container_of(xprt, struct sock_xprt, xprt);
1232         struct rpc_rqst *req;
1233
1234         dprintk("RPC:       read reply XID %08x\n", ntohl(transport->tcp_xid));
1235
1236         /* Find and lock the request corresponding to this xid */
1237         spin_lock(&xprt->transport_lock);
1238         req = xprt_lookup_rqst(xprt, transport->tcp_xid);
1239         if (!req) {
1240                 dprintk("RPC:       XID %08x request not found!\n",
1241                                 ntohl(transport->tcp_xid));
1242                 spin_unlock(&xprt->transport_lock);
1243                 return -1;
1244         }
1245
1246         xs_tcp_read_common(xprt, desc, req);
1247
1248         if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1249                 xprt_complete_rqst(req->rq_task, transport->tcp_copied);
1250
1251         spin_unlock(&xprt->transport_lock);
1252         return 0;
1253 }
1254
1255 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1256 /*
1257  * Obtains an rpc_rqst previously allocated and invokes the common
1258  * tcp read code to read the data.  The result is placed in the callback
1259  * queue.
1260  * If we're unable to obtain the rpc_rqst we schedule the closing of the
1261  * connection and return -1.
1262  */
1263 static inline int xs_tcp_read_callback(struct rpc_xprt *xprt,
1264                                        struct xdr_skb_reader *desc)
1265 {
1266         struct sock_xprt *transport =
1267                                 container_of(xprt, struct sock_xprt, xprt);
1268         struct rpc_rqst *req;
1269
1270         req = xprt_alloc_bc_request(xprt);
1271         if (req == NULL) {
1272                 printk(KERN_WARNING "Callback slot table overflowed\n");
1273                 xprt_force_disconnect(xprt);
1274                 return -1;
1275         }
1276
1277         req->rq_xid = transport->tcp_xid;
1278         dprintk("RPC:       read callback  XID %08x\n", ntohl(req->rq_xid));
1279         xs_tcp_read_common(xprt, desc, req);
1280
1281         if (!(transport->tcp_flags & TCP_RCV_COPY_DATA)) {
1282                 struct svc_serv *bc_serv = xprt->bc_serv;
1283
1284                 /*
1285                  * Add callback request to callback list.  The callback
1286                  * service sleeps on the sv_cb_waitq waiting for new
1287                  * requests.  Wake it up after adding enqueing the
1288                  * request.
1289                  */
1290                 dprintk("RPC:       add callback request to list\n");
1291                 spin_lock(&bc_serv->sv_cb_lock);
1292                 list_add(&req->rq_bc_list, &bc_serv->sv_cb_list);
1293                 spin_unlock(&bc_serv->sv_cb_lock);
1294                 wake_up(&bc_serv->sv_cb_waitq);
1295         }
1296
1297         req->rq_private_buf.len = transport->tcp_copied;
1298
1299         return 0;
1300 }
1301
1302 static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1303                                         struct xdr_skb_reader *desc)
1304 {
1305         struct sock_xprt *transport =
1306                                 container_of(xprt, struct sock_xprt, xprt);
1307
1308         return (transport->tcp_flags & TCP_RPC_REPLY) ?
1309                 xs_tcp_read_reply(xprt, desc) :
1310                 xs_tcp_read_callback(xprt, desc);
1311 }
1312 #else
1313 static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1314                                         struct xdr_skb_reader *desc)
1315 {
1316         return xs_tcp_read_reply(xprt, desc);
1317 }
1318 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1319
1320 /*
1321  * Read data off the transport.  This can be either an RPC_CALL or an
1322  * RPC_REPLY.  Relay the processing to helper functions.
1323  */
1324 static void xs_tcp_read_data(struct rpc_xprt *xprt,
1325                                     struct xdr_skb_reader *desc)
1326 {
1327         struct sock_xprt *transport =
1328                                 container_of(xprt, struct sock_xprt, xprt);
1329
1330         if (_xs_tcp_read_data(xprt, desc) == 0)
1331                 xs_tcp_check_fraghdr(transport);
1332         else {
1333                 /*
1334                  * The transport_lock protects the request handling.
1335                  * There's no need to hold it to update the tcp_flags.
1336                  */
1337                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1338         }
1339 }
1340
1341 static inline void xs_tcp_read_discard(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1342 {
1343         size_t len;
1344
1345         len = transport->tcp_reclen - transport->tcp_offset;
1346         if (len > desc->count)
1347                 len = desc->count;
1348         desc->count -= len;
1349         desc->offset += len;
1350         transport->tcp_offset += len;
1351         dprintk("RPC:       discarded %Zu bytes\n", len);
1352         xs_tcp_check_fraghdr(transport);
1353 }
1354
1355 static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
1356 {
1357         struct rpc_xprt *xprt = rd_desc->arg.data;
1358         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1359         struct xdr_skb_reader desc = {
1360                 .skb    = skb,
1361                 .offset = offset,
1362                 .count  = len,
1363         };
1364
1365         dprintk("RPC:       xs_tcp_data_recv started\n");
1366         do {
1367                 /* Read in a new fragment marker if necessary */
1368                 /* Can we ever really expect to get completely empty fragments? */
1369                 if (transport->tcp_flags & TCP_RCV_COPY_FRAGHDR) {
1370                         xs_tcp_read_fraghdr(xprt, &desc);
1371                         continue;
1372                 }
1373                 /* Read in the xid if necessary */
1374                 if (transport->tcp_flags & TCP_RCV_COPY_XID) {
1375                         xs_tcp_read_xid(transport, &desc);
1376                         continue;
1377                 }
1378                 /* Read in the call/reply flag */
1379                 if (transport->tcp_flags & TCP_RCV_READ_CALLDIR) {
1380                         xs_tcp_read_calldir(transport, &desc);
1381                         continue;
1382                 }
1383                 /* Read in the request data */
1384                 if (transport->tcp_flags & TCP_RCV_COPY_DATA) {
1385                         xs_tcp_read_data(xprt, &desc);
1386                         continue;
1387                 }
1388                 /* Skip over any trailing bytes on short reads */
1389                 xs_tcp_read_discard(transport, &desc);
1390         } while (desc.count);
1391         dprintk("RPC:       xs_tcp_data_recv done\n");
1392         return len - desc.count;
1393 }
1394
1395 /**
1396  * xs_tcp_data_ready - "data ready" callback for TCP sockets
1397  * @sk: socket with data to read
1398  * @bytes: how much data to read
1399  *
1400  */
1401 static void xs_tcp_data_ready(struct sock *sk, int bytes)
1402 {
1403         struct rpc_xprt *xprt;
1404         read_descriptor_t rd_desc;
1405         int read;
1406
1407         dprintk("RPC:       xs_tcp_data_ready...\n");
1408
1409         read_lock_bh(&sk->sk_callback_lock);
1410         if (!(xprt = xprt_from_sock(sk)))
1411                 goto out;
1412         /* Any data means we had a useful conversation, so
1413          * the we don't need to delay the next reconnect
1414          */
1415         if (xprt->reestablish_timeout)
1416                 xprt->reestablish_timeout = 0;
1417
1418         /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
1419         rd_desc.arg.data = xprt;
1420         do {
1421                 rd_desc.count = 65536;
1422                 read = tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
1423         } while (read > 0);
1424 out:
1425         read_unlock_bh(&sk->sk_callback_lock);
1426 }
1427
1428 /*
1429  * Do the equivalent of linger/linger2 handling for dealing with
1430  * broken servers that don't close the socket in a timely
1431  * fashion
1432  */
1433 static void xs_tcp_schedule_linger_timeout(struct rpc_xprt *xprt,
1434                 unsigned long timeout)
1435 {
1436         struct sock_xprt *transport;
1437
1438         if (xprt_test_and_set_connecting(xprt))
1439                 return;
1440         set_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1441         transport = container_of(xprt, struct sock_xprt, xprt);
1442         queue_delayed_work(rpciod_workqueue, &transport->connect_worker,
1443                            timeout);
1444 }
1445
1446 static void xs_tcp_cancel_linger_timeout(struct rpc_xprt *xprt)
1447 {
1448         struct sock_xprt *transport;
1449
1450         transport = container_of(xprt, struct sock_xprt, xprt);
1451
1452         if (!test_bit(XPRT_CONNECTION_ABORT, &xprt->state) ||
1453             !cancel_delayed_work(&transport->connect_worker))
1454                 return;
1455         clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1456         xprt_clear_connecting(xprt);
1457 }
1458
1459 static void xs_sock_reset_connection_flags(struct rpc_xprt *xprt)
1460 {
1461         smp_mb__before_clear_bit();
1462         clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1463         clear_bit(XPRT_CONNECTION_CLOSE, &xprt->state);
1464         clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1465         clear_bit(XPRT_CLOSING, &xprt->state);
1466         smp_mb__after_clear_bit();
1467 }
1468
1469 static void xs_sock_mark_closed(struct rpc_xprt *xprt)
1470 {
1471         xs_sock_reset_connection_flags(xprt);
1472         /* Mark transport as closed and wake up all pending tasks */
1473         xprt_disconnect_done(xprt);
1474 }
1475
1476 /**
1477  * xs_tcp_state_change - callback to handle TCP socket state changes
1478  * @sk: socket whose state has changed
1479  *
1480  */
1481 static void xs_tcp_state_change(struct sock *sk)
1482 {
1483         struct rpc_xprt *xprt;
1484
1485         read_lock_bh(&sk->sk_callback_lock);
1486         if (!(xprt = xprt_from_sock(sk)))
1487                 goto out;
1488         dprintk("RPC:       xs_tcp_state_change client %p...\n", xprt);
1489         dprintk("RPC:       state %x conn %d dead %d zapped %d sk_shutdown %d\n",
1490                         sk->sk_state, xprt_connected(xprt),
1491                         sock_flag(sk, SOCK_DEAD),
1492                         sock_flag(sk, SOCK_ZAPPED),
1493                         sk->sk_shutdown);
1494
1495         switch (sk->sk_state) {
1496         case TCP_ESTABLISHED:
1497                 spin_lock(&xprt->transport_lock);
1498                 if (!xprt_test_and_set_connected(xprt)) {
1499                         struct sock_xprt *transport = container_of(xprt,
1500                                         struct sock_xprt, xprt);
1501
1502                         /* Reset TCP record info */
1503                         transport->tcp_offset = 0;
1504                         transport->tcp_reclen = 0;
1505                         transport->tcp_copied = 0;
1506                         transport->tcp_flags =
1507                                 TCP_RCV_COPY_FRAGHDR | TCP_RCV_COPY_XID;
1508
1509                         xprt_wake_pending_tasks(xprt, -EAGAIN);
1510                 }
1511                 spin_unlock(&xprt->transport_lock);
1512                 break;
1513         case TCP_FIN_WAIT1:
1514                 /* The client initiated a shutdown of the socket */
1515                 xprt->connect_cookie++;
1516                 xprt->reestablish_timeout = 0;
1517                 set_bit(XPRT_CLOSING, &xprt->state);
1518                 smp_mb__before_clear_bit();
1519                 clear_bit(XPRT_CONNECTED, &xprt->state);
1520                 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1521                 smp_mb__after_clear_bit();
1522                 xs_tcp_schedule_linger_timeout(xprt, xs_tcp_fin_timeout);
1523                 break;
1524         case TCP_CLOSE_WAIT:
1525                 /* The server initiated a shutdown of the socket */
1526                 xprt->connect_cookie++;
1527                 clear_bit(XPRT_CONNECTED, &xprt->state);
1528                 xs_tcp_force_close(xprt);
1529         case TCP_CLOSING:
1530                 /*
1531                  * If the server closed down the connection, make sure that
1532                  * we back off before reconnecting
1533                  */
1534                 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1535                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1536                 break;
1537         case TCP_LAST_ACK:
1538                 set_bit(XPRT_CLOSING, &xprt->state);
1539                 xs_tcp_schedule_linger_timeout(xprt, xs_tcp_fin_timeout);
1540                 smp_mb__before_clear_bit();
1541                 clear_bit(XPRT_CONNECTED, &xprt->state);
1542                 smp_mb__after_clear_bit();
1543                 break;
1544         case TCP_CLOSE:
1545                 xs_tcp_cancel_linger_timeout(xprt);
1546                 xs_sock_mark_closed(xprt);
1547         }
1548  out:
1549         read_unlock_bh(&sk->sk_callback_lock);
1550 }
1551
1552 static void xs_write_space(struct sock *sk)
1553 {
1554         struct socket *sock;
1555         struct rpc_xprt *xprt;
1556
1557         if (unlikely(!(sock = sk->sk_socket)))
1558                 return;
1559         clear_bit(SOCK_NOSPACE, &sock->flags);
1560
1561         if (unlikely(!(xprt = xprt_from_sock(sk))))
1562                 return;
1563         if (test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags) == 0)
1564                 return;
1565
1566         xprt_write_space(xprt);
1567 }
1568
1569 /**
1570  * xs_udp_write_space - callback invoked when socket buffer space
1571  *                             becomes available
1572  * @sk: socket whose state has changed
1573  *
1574  * Called when more output buffer space is available for this socket.
1575  * We try not to wake our writers until they can make "significant"
1576  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1577  * with a bunch of small requests.
1578  */
1579 static void xs_udp_write_space(struct sock *sk)
1580 {
1581         read_lock_bh(&sk->sk_callback_lock);
1582
1583         /* from net/core/sock.c:sock_def_write_space */
1584         if (sock_writeable(sk))
1585                 xs_write_space(sk);
1586
1587         read_unlock_bh(&sk->sk_callback_lock);
1588 }
1589
1590 /**
1591  * xs_tcp_write_space - callback invoked when socket buffer space
1592  *                             becomes available
1593  * @sk: socket whose state has changed
1594  *
1595  * Called when more output buffer space is available for this socket.
1596  * We try not to wake our writers until they can make "significant"
1597  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1598  * with a bunch of small requests.
1599  */
1600 static void xs_tcp_write_space(struct sock *sk)
1601 {
1602         read_lock_bh(&sk->sk_callback_lock);
1603
1604         /* from net/core/stream.c:sk_stream_write_space */
1605         if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk))
1606                 xs_write_space(sk);
1607
1608         read_unlock_bh(&sk->sk_callback_lock);
1609 }
1610
1611 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
1612 {
1613         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1614         struct sock *sk = transport->inet;
1615
1616         if (transport->rcvsize) {
1617                 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1618                 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
1619         }
1620         if (transport->sndsize) {
1621                 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1622                 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1623                 sk->sk_write_space(sk);
1624         }
1625 }
1626
1627 /**
1628  * xs_udp_set_buffer_size - set send and receive limits
1629  * @xprt: generic transport
1630  * @sndsize: requested size of send buffer, in bytes
1631  * @rcvsize: requested size of receive buffer, in bytes
1632  *
1633  * Set socket send and receive buffer size limits.
1634  */
1635 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1636 {
1637         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1638
1639         transport->sndsize = 0;
1640         if (sndsize)
1641                 transport->sndsize = sndsize + 1024;
1642         transport->rcvsize = 0;
1643         if (rcvsize)
1644                 transport->rcvsize = rcvsize + 1024;
1645
1646         xs_udp_do_set_buffer_size(xprt);
1647 }
1648
1649 /**
1650  * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1651  * @task: task that timed out
1652  *
1653  * Adjust the congestion window after a retransmit timeout has occurred.
1654  */
1655 static void xs_udp_timer(struct rpc_xprt *xprt, struct rpc_task *task)
1656 {
1657         xprt_adjust_cwnd(xprt, task, -ETIMEDOUT);
1658 }
1659
1660 static unsigned short xs_get_random_port(void)
1661 {
1662         unsigned short range = xprt_max_resvport - xprt_min_resvport;
1663         unsigned short rand = (unsigned short) net_random() % range;
1664         return rand + xprt_min_resvport;
1665 }
1666
1667 /**
1668  * xs_set_port - reset the port number in the remote endpoint address
1669  * @xprt: generic transport
1670  * @port: new port number
1671  *
1672  */
1673 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1674 {
1675         dprintk("RPC:       setting port for xprt %p to %u\n", xprt, port);
1676
1677         rpc_set_port(xs_addr(xprt), port);
1678         xs_update_peer_port(xprt);
1679 }
1680
1681 static unsigned short xs_get_srcport(struct sock_xprt *transport)
1682 {
1683         unsigned short port = transport->srcport;
1684
1685         if (port == 0 && transport->xprt.resvport)
1686                 port = xs_get_random_port();
1687         return port;
1688 }
1689
1690 static unsigned short xs_next_srcport(struct sock_xprt *transport, unsigned short port)
1691 {
1692         if (transport->srcport != 0)
1693                 transport->srcport = 0;
1694         if (!transport->xprt.resvport)
1695                 return 0;
1696         if (port <= xprt_min_resvport || port > xprt_max_resvport)
1697                 return xprt_max_resvport;
1698         return --port;
1699 }
1700 static int xs_bind(struct sock_xprt *transport, struct socket *sock)
1701 {
1702         struct sockaddr_storage myaddr;
1703         int err, nloop = 0;
1704         unsigned short port = xs_get_srcport(transport);
1705         unsigned short last;
1706
1707         memcpy(&myaddr, &transport->srcaddr, transport->xprt.addrlen);
1708         do {
1709                 rpc_set_port((struct sockaddr *)&myaddr, port);
1710                 err = kernel_bind(sock, (struct sockaddr *)&myaddr,
1711                                 transport->xprt.addrlen);
1712                 if (port == 0)
1713                         break;
1714                 if (err == 0) {
1715                         transport->srcport = port;
1716                         break;
1717                 }
1718                 last = port;
1719                 port = xs_next_srcport(transport, port);
1720                 if (port > last)
1721                         nloop++;
1722         } while (err == -EADDRINUSE && nloop != 2);
1723
1724         if (myaddr.ss_family == AF_INET)
1725                 dprintk("RPC:       %s %pI4:%u: %s (%d)\n", __func__,
1726                                 &((struct sockaddr_in *)&myaddr)->sin_addr,
1727                                 port, err ? "failed" : "ok", err);
1728         else
1729                 dprintk("RPC:       %s %pI6:%u: %s (%d)\n", __func__,
1730                                 &((struct sockaddr_in6 *)&myaddr)->sin6_addr,
1731                                 port, err ? "failed" : "ok", err);
1732         return err;
1733 }
1734
1735 /*
1736  * We don't support autobind on AF_LOCAL sockets
1737  */
1738 static void xs_local_rpcbind(struct rpc_task *task)
1739 {
1740         rcu_read_lock();
1741         xprt_set_bound(rcu_dereference(task->tk_client->cl_xprt));
1742         rcu_read_unlock();
1743 }
1744
1745 static void xs_local_set_port(struct rpc_xprt *xprt, unsigned short port)
1746 {
1747 }
1748
1749 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1750 static struct lock_class_key xs_key[2];
1751 static struct lock_class_key xs_slock_key[2];
1752
1753 static inline void xs_reclassify_socketu(struct socket *sock)
1754 {
1755         struct sock *sk = sock->sk;
1756
1757         sock_lock_init_class_and_name(sk, "slock-AF_LOCAL-RPC",
1758                 &xs_slock_key[1], "sk_lock-AF_LOCAL-RPC", &xs_key[1]);
1759 }
1760
1761 static inline void xs_reclassify_socket4(struct socket *sock)
1762 {
1763         struct sock *sk = sock->sk;
1764
1765         sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1766                 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1767 }
1768
1769 static inline void xs_reclassify_socket6(struct socket *sock)
1770 {
1771         struct sock *sk = sock->sk;
1772
1773         sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1774                 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
1775 }
1776
1777 static inline void xs_reclassify_socket(int family, struct socket *sock)
1778 {
1779         WARN_ON_ONCE(sock_owned_by_user(sock->sk));
1780         if (sock_owned_by_user(sock->sk))
1781                 return;
1782
1783         switch (family) {
1784         case AF_LOCAL:
1785                 xs_reclassify_socketu(sock);
1786                 break;
1787         case AF_INET:
1788                 xs_reclassify_socket4(sock);
1789                 break;
1790         case AF_INET6:
1791                 xs_reclassify_socket6(sock);
1792                 break;
1793         }
1794 }
1795 #else
1796 static inline void xs_reclassify_socketu(struct socket *sock)
1797 {
1798 }
1799
1800 static inline void xs_reclassify_socket4(struct socket *sock)
1801 {
1802 }
1803
1804 static inline void xs_reclassify_socket6(struct socket *sock)
1805 {
1806 }
1807
1808 static inline void xs_reclassify_socket(int family, struct socket *sock)
1809 {
1810 }
1811 #endif
1812
1813 static struct socket *xs_create_sock(struct rpc_xprt *xprt,
1814                 struct sock_xprt *transport, int family, int type, int protocol)
1815 {
1816         struct socket *sock;
1817         int err;
1818
1819         err = __sock_create(xprt->xprt_net, family, type, protocol, &sock, 1);
1820         if (err < 0) {
1821                 dprintk("RPC:       can't create %d transport socket (%d).\n",
1822                                 protocol, -err);
1823                 goto out;
1824         }
1825         xs_reclassify_socket(family, sock);
1826
1827         err = xs_bind(transport, sock);
1828         if (err) {
1829                 sock_release(sock);
1830                 goto out;
1831         }
1832
1833         return sock;
1834 out:
1835         return ERR_PTR(err);
1836 }
1837
1838 static int xs_local_finish_connecting(struct rpc_xprt *xprt,
1839                                       struct socket *sock)
1840 {
1841         struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1842                                                                         xprt);
1843
1844         if (!transport->inet) {
1845                 struct sock *sk = sock->sk;
1846
1847                 write_lock_bh(&sk->sk_callback_lock);
1848
1849                 xs_save_old_callbacks(transport, sk);
1850
1851                 sk->sk_user_data = xprt;
1852                 sk->sk_data_ready = xs_local_data_ready;
1853                 sk->sk_write_space = xs_udp_write_space;
1854                 sk->sk_allocation = GFP_ATOMIC;
1855
1856                 xprt_clear_connected(xprt);
1857
1858                 /* Reset to new socket */
1859                 transport->sock = sock;
1860                 transport->inet = sk;
1861
1862                 write_unlock_bh(&sk->sk_callback_lock);
1863         }
1864
1865         /* Tell the socket layer to start connecting... */
1866         xprt->stat.connect_count++;
1867         xprt->stat.connect_start = jiffies;
1868         return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, 0);
1869 }
1870
1871 /**
1872  * xs_local_setup_socket - create AF_LOCAL socket, connect to a local endpoint
1873  * @xprt: RPC transport to connect
1874  * @transport: socket transport to connect
1875  * @create_sock: function to create a socket of the correct type
1876  */
1877 static int xs_local_setup_socket(struct sock_xprt *transport)
1878 {
1879         struct rpc_xprt *xprt = &transport->xprt;
1880         struct socket *sock;
1881         int status = -EIO;
1882
1883         current->flags |= PF_FSTRANS;
1884
1885         clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1886         status = __sock_create(xprt->xprt_net, AF_LOCAL,
1887                                         SOCK_STREAM, 0, &sock, 1);
1888         if (status < 0) {
1889                 dprintk("RPC:       can't create AF_LOCAL "
1890                         "transport socket (%d).\n", -status);
1891                 goto out;
1892         }
1893         xs_reclassify_socketu(sock);
1894
1895         dprintk("RPC:       worker connecting xprt %p via AF_LOCAL to %s\n",
1896                         xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1897
1898         status = xs_local_finish_connecting(xprt, sock);
1899         switch (status) {
1900         case 0:
1901                 dprintk("RPC:       xprt %p connected to %s\n",
1902                                 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1903                 xprt_set_connected(xprt);
1904                 break;
1905         case -ENOENT:
1906                 dprintk("RPC:       xprt %p: socket %s does not exist\n",
1907                                 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1908                 break;
1909         case -ECONNREFUSED:
1910                 dprintk("RPC:       xprt %p: connection refused for %s\n",
1911                                 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1912                 break;
1913         default:
1914                 printk(KERN_ERR "%s: unhandled error (%d) connecting to %s\n",
1915                                 __func__, -status,
1916                                 xprt->address_strings[RPC_DISPLAY_ADDR]);
1917         }
1918
1919 out:
1920         xprt_clear_connecting(xprt);
1921         xprt_wake_pending_tasks(xprt, status);
1922         current->flags &= ~PF_FSTRANS;
1923         return status;
1924 }
1925
1926 static void xs_local_connect(struct rpc_xprt *xprt, struct rpc_task *task)
1927 {
1928         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1929         int ret;
1930
1931          if (RPC_IS_ASYNC(task)) {
1932                 /*
1933                  * We want the AF_LOCAL connect to be resolved in the
1934                  * filesystem namespace of the process making the rpc
1935                  * call.  Thus we connect synchronously.
1936                  *
1937                  * If we want to support asynchronous AF_LOCAL calls,
1938                  * we'll need to figure out how to pass a namespace to
1939                  * connect.
1940                  */
1941                 rpc_exit(task, -ENOTCONN);
1942                 return;
1943         }
1944         ret = xs_local_setup_socket(transport);
1945         if (ret && !RPC_IS_SOFTCONN(task))
1946                 msleep_interruptible(15000);
1947 }
1948
1949 #ifdef CONFIG_SUNRPC_SWAP
1950 static void xs_set_memalloc(struct rpc_xprt *xprt)
1951 {
1952         struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1953                         xprt);
1954
1955         if (xprt->swapper)
1956                 sk_set_memalloc(transport->inet);
1957 }
1958
1959 /**
1960  * xs_swapper - Tag this transport as being used for swap.
1961  * @xprt: transport to tag
1962  * @enable: enable/disable
1963  *
1964  */
1965 int xs_swapper(struct rpc_xprt *xprt, int enable)
1966 {
1967         struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1968                         xprt);
1969         int err = 0;
1970
1971         if (enable) {
1972                 xprt->swapper++;
1973                 xs_set_memalloc(xprt);
1974         } else if (xprt->swapper) {
1975                 xprt->swapper--;
1976                 sk_clear_memalloc(transport->inet);
1977         }
1978
1979         return err;
1980 }
1981 EXPORT_SYMBOL_GPL(xs_swapper);
1982 #else
1983 static void xs_set_memalloc(struct rpc_xprt *xprt)
1984 {
1985 }
1986 #endif
1987
1988 static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
1989 {
1990         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1991
1992         if (!transport->inet) {
1993                 struct sock *sk = sock->sk;
1994
1995                 write_lock_bh(&sk->sk_callback_lock);
1996
1997                 xs_save_old_callbacks(transport, sk);
1998
1999                 sk->sk_user_data = xprt;
2000                 sk->sk_data_ready = xs_udp_data_ready;
2001                 sk->sk_write_space = xs_udp_write_space;
2002                 sk->sk_no_check = UDP_CSUM_NORCV;
2003                 sk->sk_allocation = GFP_ATOMIC;
2004
2005                 xprt_set_connected(xprt);
2006
2007                 /* Reset to new socket */
2008                 transport->sock = sock;
2009                 transport->inet = sk;
2010
2011                 xs_set_memalloc(xprt);
2012
2013                 write_unlock_bh(&sk->sk_callback_lock);
2014         }
2015         xs_udp_do_set_buffer_size(xprt);
2016 }
2017
2018 static void xs_udp_setup_socket(struct work_struct *work)
2019 {
2020         struct sock_xprt *transport =
2021                 container_of(work, struct sock_xprt, connect_worker.work);
2022         struct rpc_xprt *xprt = &transport->xprt;
2023         struct socket *sock = transport->sock;
2024         int status = -EIO;
2025
2026         current->flags |= PF_FSTRANS;
2027
2028         /* Start by resetting any existing state */
2029         xs_reset_transport(transport);
2030         sock = xs_create_sock(xprt, transport,
2031                         xs_addr(xprt)->sa_family, SOCK_DGRAM, IPPROTO_UDP);
2032         if (IS_ERR(sock))
2033                 goto out;
2034
2035         dprintk("RPC:       worker connecting xprt %p via %s to "
2036                                 "%s (port %s)\n", xprt,
2037                         xprt->address_strings[RPC_DISPLAY_PROTO],
2038                         xprt->address_strings[RPC_DISPLAY_ADDR],
2039                         xprt->address_strings[RPC_DISPLAY_PORT]);
2040
2041         xs_udp_finish_connecting(xprt, sock);
2042         status = 0;
2043 out:
2044         xprt_clear_connecting(xprt);
2045         xprt_wake_pending_tasks(xprt, status);
2046         current->flags &= ~PF_FSTRANS;
2047 }
2048
2049 /*
2050  * We need to preserve the port number so the reply cache on the server can
2051  * find our cached RPC replies when we get around to reconnecting.
2052  */
2053 static void xs_abort_connection(struct sock_xprt *transport)
2054 {
2055         int result;
2056         struct sockaddr any;
2057
2058         dprintk("RPC:       disconnecting xprt %p to reuse port\n", transport);
2059
2060         /*
2061          * Disconnect the transport socket by doing a connect operation
2062          * with AF_UNSPEC.  This should return immediately...
2063          */
2064         memset(&any, 0, sizeof(any));
2065         any.sa_family = AF_UNSPEC;
2066         result = kernel_connect(transport->sock, &any, sizeof(any), 0);
2067         if (!result)
2068                 xs_sock_reset_connection_flags(&transport->xprt);
2069         dprintk("RPC:       AF_UNSPEC connect return code %d\n", result);
2070 }
2071
2072 static void xs_tcp_reuse_connection(struct sock_xprt *transport)
2073 {
2074         unsigned int state = transport->inet->sk_state;
2075
2076         if (state == TCP_CLOSE && transport->sock->state == SS_UNCONNECTED) {
2077                 /* we don't need to abort the connection if the socket
2078                  * hasn't undergone a shutdown
2079                  */
2080                 if (transport->inet->sk_shutdown == 0)
2081                         return;
2082                 dprintk("RPC:       %s: TCP_CLOSEd and sk_shutdown set to %d\n",
2083                                 __func__, transport->inet->sk_shutdown);
2084         }
2085         if ((1 << state) & (TCPF_ESTABLISHED|TCPF_SYN_SENT)) {
2086                 /* we don't need to abort the connection if the socket
2087                  * hasn't undergone a shutdown
2088                  */
2089                 if (transport->inet->sk_shutdown == 0)
2090                         return;
2091                 dprintk("RPC:       %s: ESTABLISHED/SYN_SENT "
2092                                 "sk_shutdown set to %d\n",
2093                                 __func__, transport->inet->sk_shutdown);
2094         }
2095         xs_abort_connection(transport);
2096 }
2097
2098 static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2099 {
2100         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2101         int ret = -ENOTCONN;
2102
2103         if (!transport->inet) {
2104                 struct sock *sk = sock->sk;
2105
2106                 write_lock_bh(&sk->sk_callback_lock);
2107
2108                 xs_save_old_callbacks(transport, sk);
2109
2110                 sk->sk_user_data = xprt;
2111                 sk->sk_data_ready = xs_tcp_data_ready;
2112                 sk->sk_state_change = xs_tcp_state_change;
2113                 sk->sk_write_space = xs_tcp_write_space;
2114                 sk->sk_allocation = GFP_ATOMIC;
2115
2116                 /* socket options */
2117                 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
2118                 sock_reset_flag(sk, SOCK_LINGER);
2119                 tcp_sk(sk)->linger2 = 0;
2120                 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
2121
2122                 xprt_clear_connected(xprt);
2123
2124                 /* Reset to new socket */
2125                 transport->sock = sock;
2126                 transport->inet = sk;
2127
2128                 write_unlock_bh(&sk->sk_callback_lock);
2129         }
2130
2131         if (!xprt_bound(xprt))
2132                 goto out;
2133
2134         xs_set_memalloc(xprt);
2135
2136         /* Tell the socket layer to start connecting... */
2137         xprt->stat.connect_count++;
2138         xprt->stat.connect_start = jiffies;
2139         ret = kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
2140         switch (ret) {
2141         case 0:
2142         case -EINPROGRESS:
2143                 /* SYN_SENT! */
2144                 xprt->connect_cookie++;
2145                 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2146                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2147         }
2148 out:
2149         return ret;
2150 }
2151
2152 /**
2153  * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint
2154  * @xprt: RPC transport to connect
2155  * @transport: socket transport to connect
2156  * @create_sock: function to create a socket of the correct type
2157  *
2158  * Invoked by a work queue tasklet.
2159  */
2160 static void xs_tcp_setup_socket(struct work_struct *work)
2161 {
2162         struct sock_xprt *transport =
2163                 container_of(work, struct sock_xprt, connect_worker.work);
2164         struct socket *sock = transport->sock;
2165         struct rpc_xprt *xprt = &transport->xprt;
2166         int status = -EIO;
2167
2168         current->flags |= PF_FSTRANS;
2169
2170         if (!sock) {
2171                 clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
2172                 sock = xs_create_sock(xprt, transport,
2173                                 xs_addr(xprt)->sa_family, SOCK_STREAM, IPPROTO_TCP);
2174                 if (IS_ERR(sock)) {
2175                         status = PTR_ERR(sock);
2176                         goto out;
2177                 }
2178         } else {
2179                 int abort_and_exit;
2180
2181                 abort_and_exit = test_and_clear_bit(XPRT_CONNECTION_ABORT,
2182                                 &xprt->state);
2183                 /* "close" the socket, preserving the local port */
2184                 xs_tcp_reuse_connection(transport);
2185
2186                 if (abort_and_exit)
2187                         goto out_eagain;
2188         }
2189
2190         dprintk("RPC:       worker connecting xprt %p via %s to "
2191                                 "%s (port %s)\n", xprt,
2192                         xprt->address_strings[RPC_DISPLAY_PROTO],
2193                         xprt->address_strings[RPC_DISPLAY_ADDR],
2194                         xprt->address_strings[RPC_DISPLAY_PORT]);
2195
2196         status = xs_tcp_finish_connecting(xprt, sock);
2197         dprintk("RPC:       %p connect status %d connected %d sock state %d\n",
2198                         xprt, -status, xprt_connected(xprt),
2199                         sock->sk->sk_state);
2200         switch (status) {
2201         default:
2202                 printk("%s: connect returned unhandled error %d\n",
2203                         __func__, status);
2204         case -EADDRNOTAVAIL:
2205                 /* We're probably in TIME_WAIT. Get rid of existing socket,
2206                  * and retry
2207                  */
2208                 xs_tcp_force_close(xprt);
2209                 break;
2210         case 0:
2211         case -EINPROGRESS:
2212         case -EALREADY:
2213                 xprt_clear_connecting(xprt);
2214                 current->flags &= ~PF_FSTRANS;
2215                 return;
2216         case -EINVAL:
2217                 /* Happens, for instance, if the user specified a link
2218                  * local IPv6 address without a scope-id.
2219                  */
2220         case -ECONNREFUSED:
2221         case -ECONNRESET:
2222         case -ENETUNREACH:
2223                 /* retry with existing socket, after a delay */
2224                 goto out;
2225         }
2226 out_eagain:
2227         status = -EAGAIN;
2228 out:
2229         xprt_clear_connecting(xprt);
2230         xprt_wake_pending_tasks(xprt, status);
2231         current->flags &= ~PF_FSTRANS;
2232 }
2233
2234 /**
2235  * xs_connect - connect a socket to a remote endpoint
2236  * @xprt: pointer to transport structure
2237  * @task: address of RPC task that manages state of connect request
2238  *
2239  * TCP: If the remote end dropped the connection, delay reconnecting.
2240  *
2241  * UDP socket connects are synchronous, but we use a work queue anyway
2242  * to guarantee that even unprivileged user processes can set up a
2243  * socket on a privileged port.
2244  *
2245  * If a UDP socket connect fails, the delay behavior here prevents
2246  * retry floods (hard mounts).
2247  */
2248 static void xs_connect(struct rpc_xprt *xprt, struct rpc_task *task)
2249 {
2250         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2251
2252         if (transport->sock != NULL && !RPC_IS_SOFTCONN(task)) {
2253                 dprintk("RPC:       xs_connect delayed xprt %p for %lu "
2254                                 "seconds\n",
2255                                 xprt, xprt->reestablish_timeout / HZ);
2256                 queue_delayed_work(rpciod_workqueue,
2257                                    &transport->connect_worker,
2258                                    xprt->reestablish_timeout);
2259                 xprt->reestablish_timeout <<= 1;
2260                 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2261                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2262                 if (xprt->reestablish_timeout > XS_TCP_MAX_REEST_TO)
2263                         xprt->reestablish_timeout = XS_TCP_MAX_REEST_TO;
2264         } else {
2265                 dprintk("RPC:       xs_connect scheduled xprt %p\n", xprt);
2266                 queue_delayed_work(rpciod_workqueue,
2267                                    &transport->connect_worker, 0);
2268         }
2269 }
2270
2271 /**
2272  * xs_local_print_stats - display AF_LOCAL socket-specifc stats
2273  * @xprt: rpc_xprt struct containing statistics
2274  * @seq: output file
2275  *
2276  */
2277 static void xs_local_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2278 {
2279         long idle_time = 0;
2280
2281         if (xprt_connected(xprt))
2282                 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2283
2284         seq_printf(seq, "\txprt:\tlocal %lu %lu %lu %ld %lu %lu %lu "
2285                         "%llu %llu %lu %llu %llu\n",
2286                         xprt->stat.bind_count,
2287                         xprt->stat.connect_count,
2288                         xprt->stat.connect_time,
2289                         idle_time,
2290                         xprt->stat.sends,
2291                         xprt->stat.recvs,
2292                         xprt->stat.bad_xids,
2293                         xprt->stat.req_u,
2294                         xprt->stat.bklog_u,
2295                         xprt->stat.max_slots,
2296                         xprt->stat.sending_u,
2297                         xprt->stat.pending_u);
2298 }
2299
2300 /**
2301  * xs_udp_print_stats - display UDP socket-specifc stats
2302  * @xprt: rpc_xprt struct containing statistics
2303  * @seq: output file
2304  *
2305  */
2306 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2307 {
2308         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2309
2310         seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %llu %llu "
2311                         "%lu %llu %llu\n",
2312                         transport->srcport,
2313                         xprt->stat.bind_count,
2314                         xprt->stat.sends,
2315                         xprt->stat.recvs,
2316                         xprt->stat.bad_xids,
2317                         xprt->stat.req_u,
2318                         xprt->stat.bklog_u,
2319                         xprt->stat.max_slots,
2320                         xprt->stat.sending_u,
2321                         xprt->stat.pending_u);
2322 }
2323
2324 /**
2325  * xs_tcp_print_stats - display TCP socket-specifc stats
2326  * @xprt: rpc_xprt struct containing statistics
2327  * @seq: output file
2328  *
2329  */
2330 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2331 {
2332         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2333         long idle_time = 0;
2334
2335         if (xprt_connected(xprt))
2336                 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2337
2338         seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu "
2339                         "%llu %llu %lu %llu %llu\n",
2340                         transport->srcport,
2341                         xprt->stat.bind_count,
2342                         xprt->stat.connect_count,
2343                         xprt->stat.connect_time,
2344                         idle_time,
2345                         xprt->stat.sends,
2346                         xprt->stat.recvs,
2347                         xprt->stat.bad_xids,
2348                         xprt->stat.req_u,
2349                         xprt->stat.bklog_u,
2350                         xprt->stat.max_slots,
2351                         xprt->stat.sending_u,
2352                         xprt->stat.pending_u);
2353 }
2354
2355 /*
2356  * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason
2357  * we allocate pages instead doing a kmalloc like rpc_malloc is because we want
2358  * to use the server side send routines.
2359  */
2360 static void *bc_malloc(struct rpc_task *task, size_t size)
2361 {
2362         struct page *page;
2363         struct rpc_buffer *buf;
2364
2365         WARN_ON_ONCE(size > PAGE_SIZE - sizeof(struct rpc_buffer));
2366         if (size > PAGE_SIZE - sizeof(struct rpc_buffer))
2367                 return NULL;
2368
2369         page = alloc_page(GFP_KERNEL);
2370         if (!page)
2371                 return NULL;
2372
2373         buf = page_address(page);
2374         buf->len = PAGE_SIZE;
2375
2376         return buf->data;
2377 }
2378
2379 /*
2380  * Free the space allocated in the bc_alloc routine
2381  */
2382 static void bc_free(void *buffer)
2383 {
2384         struct rpc_buffer *buf;
2385
2386         if (!buffer)
2387                 return;
2388
2389         buf = container_of(buffer, struct rpc_buffer, data);
2390         free_page((unsigned long)buf);
2391 }
2392
2393 /*
2394  * Use the svc_sock to send the callback. Must be called with svsk->sk_mutex
2395  * held. Borrows heavily from svc_tcp_sendto and xs_tcp_send_request.
2396  */
2397 static int bc_sendto(struct rpc_rqst *req)
2398 {
2399         int len;
2400         struct xdr_buf *xbufp = &req->rq_snd_buf;
2401         struct rpc_xprt *xprt = req->rq_xprt;
2402         struct sock_xprt *transport =
2403                                 container_of(xprt, struct sock_xprt, xprt);
2404         struct socket *sock = transport->sock;
2405         unsigned long headoff;
2406         unsigned long tailoff;
2407
2408         xs_encode_stream_record_marker(xbufp);
2409
2410         tailoff = (unsigned long)xbufp->tail[0].iov_base & ~PAGE_MASK;
2411         headoff = (unsigned long)xbufp->head[0].iov_base & ~PAGE_MASK;
2412         len = svc_send_common(sock, xbufp,
2413                               virt_to_page(xbufp->head[0].iov_base), headoff,
2414                               xbufp->tail[0].iov_base, tailoff);
2415
2416         if (len != xbufp->len) {
2417                 printk(KERN_NOTICE "Error sending entire callback!\n");
2418                 len = -EAGAIN;
2419         }
2420
2421         return len;
2422 }
2423
2424 /*
2425  * The send routine. Borrows from svc_send
2426  */
2427 static int bc_send_request(struct rpc_task *task)
2428 {
2429         struct rpc_rqst *req = task->tk_rqstp;
2430         struct svc_xprt *xprt;
2431         u32                     len;
2432
2433         dprintk("sending request with xid: %08x\n", ntohl(req->rq_xid));
2434         /*
2435          * Get the server socket associated with this callback xprt
2436          */
2437         xprt = req->rq_xprt->bc_xprt;
2438
2439         /*
2440          * Grab the mutex to serialize data as the connection is shared
2441          * with the fore channel
2442          */
2443         if (!mutex_trylock(&xprt->xpt_mutex)) {
2444                 rpc_sleep_on(&xprt->xpt_bc_pending, task, NULL);
2445                 if (!mutex_trylock(&xprt->xpt_mutex))
2446                         return -EAGAIN;
2447                 rpc_wake_up_queued_task(&xprt->xpt_bc_pending, task);
2448         }
2449         if (test_bit(XPT_DEAD, &xprt->xpt_flags))
2450                 len = -ENOTCONN;
2451         else
2452                 len = bc_sendto(req);
2453         mutex_unlock(&xprt->xpt_mutex);
2454
2455         if (len > 0)
2456                 len = 0;
2457
2458         return len;
2459 }
2460
2461 /*
2462  * The close routine. Since this is client initiated, we do nothing
2463  */
2464
2465 static void bc_close(struct rpc_xprt *xprt)
2466 {
2467 }
2468
2469 /*
2470  * The xprt destroy routine. Again, because this connection is client
2471  * initiated, we do nothing
2472  */
2473
2474 static void bc_destroy(struct rpc_xprt *xprt)
2475 {
2476 }
2477
2478 static struct rpc_xprt_ops xs_local_ops = {
2479         .reserve_xprt           = xprt_reserve_xprt,
2480         .release_xprt           = xs_tcp_release_xprt,
2481         .alloc_slot             = xprt_alloc_slot,
2482         .rpcbind                = xs_local_rpcbind,
2483         .set_port               = xs_local_set_port,
2484         .connect                = xs_local_connect,
2485         .buf_alloc              = rpc_malloc,
2486         .buf_free               = rpc_free,
2487         .send_request           = xs_local_send_request,
2488         .set_retrans_timeout    = xprt_set_retrans_timeout_def,
2489         .close                  = xs_close,
2490         .destroy                = xs_local_destroy,
2491         .print_stats            = xs_local_print_stats,
2492 };
2493
2494 static struct rpc_xprt_ops xs_udp_ops = {
2495         .set_buffer_size        = xs_udp_set_buffer_size,
2496         .reserve_xprt           = xprt_reserve_xprt_cong,
2497         .release_xprt           = xprt_release_xprt_cong,
2498         .alloc_slot             = xprt_alloc_slot,
2499         .rpcbind                = rpcb_getport_async,
2500         .set_port               = xs_set_port,
2501         .connect                = xs_connect,
2502         .buf_alloc              = rpc_malloc,
2503         .buf_free               = rpc_free,
2504         .send_request           = xs_udp_send_request,
2505         .set_retrans_timeout    = xprt_set_retrans_timeout_rtt,
2506         .timer                  = xs_udp_timer,
2507         .release_request        = xprt_release_rqst_cong,
2508         .close                  = xs_close,
2509         .destroy                = xs_destroy,
2510         .print_stats            = xs_udp_print_stats,
2511 };
2512
2513 static struct rpc_xprt_ops xs_tcp_ops = {
2514         .reserve_xprt           = xprt_reserve_xprt,
2515         .release_xprt           = xs_tcp_release_xprt,
2516         .alloc_slot             = xprt_lock_and_alloc_slot,
2517         .rpcbind                = rpcb_getport_async,
2518         .set_port               = xs_set_port,
2519         .connect                = xs_connect,
2520         .buf_alloc              = rpc_malloc,
2521         .buf_free               = rpc_free,
2522         .send_request           = xs_tcp_send_request,
2523         .set_retrans_timeout    = xprt_set_retrans_timeout_def,
2524         .close                  = xs_tcp_close,
2525         .destroy                = xs_destroy,
2526         .print_stats            = xs_tcp_print_stats,
2527 };
2528
2529 /*
2530  * The rpc_xprt_ops for the server backchannel
2531  */
2532
2533 static struct rpc_xprt_ops bc_tcp_ops = {
2534         .reserve_xprt           = xprt_reserve_xprt,
2535         .release_xprt           = xprt_release_xprt,
2536         .alloc_slot             = xprt_alloc_slot,
2537         .rpcbind                = xs_local_rpcbind,
2538         .buf_alloc              = bc_malloc,
2539         .buf_free               = bc_free,
2540         .send_request           = bc_send_request,
2541         .set_retrans_timeout    = xprt_set_retrans_timeout_def,
2542         .close                  = bc_close,
2543         .destroy                = bc_destroy,
2544         .print_stats            = xs_tcp_print_stats,
2545 };
2546
2547 static int xs_init_anyaddr(const int family, struct sockaddr *sap)
2548 {
2549         static const struct sockaddr_in sin = {
2550                 .sin_family             = AF_INET,
2551                 .sin_addr.s_addr        = htonl(INADDR_ANY),
2552         };
2553         static const struct sockaddr_in6 sin6 = {
2554                 .sin6_family            = AF_INET6,
2555                 .sin6_addr              = IN6ADDR_ANY_INIT,
2556         };
2557
2558         switch (family) {
2559         case AF_LOCAL:
2560                 break;
2561         case AF_INET:
2562                 memcpy(sap, &sin, sizeof(sin));
2563                 break;
2564         case AF_INET6:
2565                 memcpy(sap, &sin6, sizeof(sin6));
2566                 break;
2567         default:
2568                 dprintk("RPC:       %s: Bad address family\n", __func__);
2569                 return -EAFNOSUPPORT;
2570         }
2571         return 0;
2572 }
2573
2574 static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
2575                                       unsigned int slot_table_size,
2576                                       unsigned int max_slot_table_size)
2577 {
2578         struct rpc_xprt *xprt;
2579         struct sock_xprt *new;
2580
2581         if (args->addrlen > sizeof(xprt->addr)) {
2582                 dprintk("RPC:       xs_setup_xprt: address too large\n");
2583                 return ERR_PTR(-EBADF);
2584         }
2585
2586         xprt = xprt_alloc(args->net, sizeof(*new), slot_table_size,
2587                         max_slot_table_size);
2588         if (xprt == NULL) {
2589                 dprintk("RPC:       xs_setup_xprt: couldn't allocate "
2590                                 "rpc_xprt\n");
2591                 return ERR_PTR(-ENOMEM);
2592         }
2593
2594         new = container_of(xprt, struct sock_xprt, xprt);
2595         memcpy(&xprt->addr, args->dstaddr, args->addrlen);
2596         xprt->addrlen = args->addrlen;
2597         if (args->srcaddr)
2598                 memcpy(&new->srcaddr, args->srcaddr, args->addrlen);
2599         else {
2600                 int err;
2601                 err = xs_init_anyaddr(args->dstaddr->sa_family,
2602                                         (struct sockaddr *)&new->srcaddr);
2603                 if (err != 0) {
2604                         xprt_free(xprt);
2605                         return ERR_PTR(err);
2606                 }
2607         }
2608
2609         return xprt;
2610 }
2611
2612 static const struct rpc_timeout xs_local_default_timeout = {
2613         .to_initval = 10 * HZ,
2614         .to_maxval = 10 * HZ,
2615         .to_retries = 2,
2616 };
2617
2618 /**
2619  * xs_setup_local - Set up transport to use an AF_LOCAL socket
2620  * @args: rpc transport creation arguments
2621  *
2622  * AF_LOCAL is a "tpi_cots_ord" transport, just like TCP
2623  */
2624 static struct rpc_xprt *xs_setup_local(struct xprt_create *args)
2625 {
2626         struct sockaddr_un *sun = (struct sockaddr_un *)args->dstaddr;
2627         struct sock_xprt *transport;
2628         struct rpc_xprt *xprt;
2629         struct rpc_xprt *ret;
2630
2631         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2632                         xprt_max_tcp_slot_table_entries);
2633         if (IS_ERR(xprt))
2634                 return xprt;
2635         transport = container_of(xprt, struct sock_xprt, xprt);
2636
2637         xprt->prot = 0;
2638         xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2639         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2640
2641         xprt->bind_timeout = XS_BIND_TO;
2642         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2643         xprt->idle_timeout = XS_IDLE_DISC_TO;
2644
2645         xprt->ops = &xs_local_ops;
2646         xprt->timeout = &xs_local_default_timeout;
2647
2648         switch (sun->sun_family) {
2649         case AF_LOCAL:
2650                 if (sun->sun_path[0] != '/') {
2651                         dprintk("RPC:       bad AF_LOCAL address: %s\n",
2652                                         sun->sun_path);
2653                         ret = ERR_PTR(-EINVAL);
2654                         goto out_err;
2655                 }
2656                 xprt_set_bound(xprt);
2657                 xs_format_peer_addresses(xprt, "local", RPCBIND_NETID_LOCAL);
2658                 ret = ERR_PTR(xs_local_setup_socket(transport));
2659                 if (ret)
2660                         goto out_err;
2661                 break;
2662         default:
2663                 ret = ERR_PTR(-EAFNOSUPPORT);
2664                 goto out_err;
2665         }
2666
2667         dprintk("RPC:       set up xprt to %s via AF_LOCAL\n",
2668                         xprt->address_strings[RPC_DISPLAY_ADDR]);
2669
2670         if (try_module_get(THIS_MODULE))
2671                 return xprt;
2672         ret = ERR_PTR(-EINVAL);
2673 out_err:
2674         xprt_free(xprt);
2675         return ret;
2676 }
2677
2678 static const struct rpc_timeout xs_udp_default_timeout = {
2679         .to_initval = 5 * HZ,
2680         .to_maxval = 30 * HZ,
2681         .to_increment = 5 * HZ,
2682         .to_retries = 5,
2683 };
2684
2685 /**
2686  * xs_setup_udp - Set up transport to use a UDP socket
2687  * @args: rpc transport creation arguments
2688  *
2689  */
2690 static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
2691 {
2692         struct sockaddr *addr = args->dstaddr;
2693         struct rpc_xprt *xprt;
2694         struct sock_xprt *transport;
2695         struct rpc_xprt *ret;
2696
2697         xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries,
2698                         xprt_udp_slot_table_entries);
2699         if (IS_ERR(xprt))
2700                 return xprt;
2701         transport = container_of(xprt, struct sock_xprt, xprt);
2702
2703         xprt->prot = IPPROTO_UDP;
2704         xprt->tsh_size = 0;
2705         /* XXX: header size can vary due to auth type, IPv6, etc. */
2706         xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
2707
2708         xprt->bind_timeout = XS_BIND_TO;
2709         xprt->reestablish_timeout = XS_UDP_REEST_TO;
2710         xprt->idle_timeout = XS_IDLE_DISC_TO;
2711
2712         xprt->ops = &xs_udp_ops;
2713
2714         xprt->timeout = &xs_udp_default_timeout;
2715
2716         switch (addr->sa_family) {
2717         case AF_INET:
2718                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2719                         xprt_set_bound(xprt);
2720
2721                 INIT_DELAYED_WORK(&transport->connect_worker,
2722                                         xs_udp_setup_socket);
2723                 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
2724                 break;
2725         case AF_INET6:
2726                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2727                         xprt_set_bound(xprt);
2728
2729                 INIT_DELAYED_WORK(&transport->connect_worker,
2730                                         xs_udp_setup_socket);
2731                 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
2732                 break;
2733         default:
2734                 ret = ERR_PTR(-EAFNOSUPPORT);
2735                 goto out_err;
2736         }
2737
2738         if (xprt_bound(xprt))
2739                 dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
2740                                 xprt->address_strings[RPC_DISPLAY_ADDR],
2741                                 xprt->address_strings[RPC_DISPLAY_PORT],
2742                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
2743         else
2744                 dprintk("RPC:       set up xprt to %s (autobind) via %s\n",
2745                                 xprt->address_strings[RPC_DISPLAY_ADDR],
2746                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
2747
2748         if (try_module_get(THIS_MODULE))
2749                 return xprt;
2750         ret = ERR_PTR(-EINVAL);
2751 out_err:
2752         xprt_free(xprt);
2753         return ret;
2754 }
2755
2756 static const struct rpc_timeout xs_tcp_default_timeout = {
2757         .to_initval = 60 * HZ,
2758         .to_maxval = 60 * HZ,
2759         .to_retries = 2,
2760 };
2761
2762 /**
2763  * xs_setup_tcp - Set up transport to use a TCP socket
2764  * @args: rpc transport creation arguments
2765  *
2766  */
2767 static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
2768 {
2769         struct sockaddr *addr = args->dstaddr;
2770         struct rpc_xprt *xprt;
2771         struct sock_xprt *transport;
2772         struct rpc_xprt *ret;
2773         unsigned int max_slot_table_size = xprt_max_tcp_slot_table_entries;
2774
2775         if (args->flags & XPRT_CREATE_INFINITE_SLOTS)
2776                 max_slot_table_size = RPC_MAX_SLOT_TABLE_LIMIT;
2777
2778         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2779                         max_slot_table_size);
2780         if (IS_ERR(xprt))
2781                 return xprt;
2782         transport = container_of(xprt, struct sock_xprt, xprt);
2783
2784         xprt->prot = IPPROTO_TCP;
2785         xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2786         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2787
2788         xprt->bind_timeout = XS_BIND_TO;
2789         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2790         xprt->idle_timeout = XS_IDLE_DISC_TO;
2791
2792         xprt->ops = &xs_tcp_ops;
2793         xprt->timeout = &xs_tcp_default_timeout;
2794
2795         switch (addr->sa_family) {
2796         case AF_INET:
2797                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2798                         xprt_set_bound(xprt);
2799
2800                 INIT_DELAYED_WORK(&transport->connect_worker,
2801                                         xs_tcp_setup_socket);
2802                 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
2803                 break;
2804         case AF_INET6:
2805                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2806                         xprt_set_bound(xprt);
2807
2808                 INIT_DELAYED_WORK(&transport->connect_worker,
2809                                         xs_tcp_setup_socket);
2810                 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
2811                 break;
2812         default:
2813                 ret = ERR_PTR(-EAFNOSUPPORT);
2814                 goto out_err;
2815         }
2816
2817         if (xprt_bound(xprt))
2818                 dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
2819                                 xprt->address_strings[RPC_DISPLAY_ADDR],
2820                                 xprt->address_strings[RPC_DISPLAY_PORT],
2821                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
2822         else
2823                 dprintk("RPC:       set up xprt to %s (autobind) via %s\n",
2824                                 xprt->address_strings[RPC_DISPLAY_ADDR],
2825                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
2826
2827
2828         if (try_module_get(THIS_MODULE))
2829                 return xprt;
2830         ret = ERR_PTR(-EINVAL);
2831 out_err:
2832         xprt_free(xprt);
2833         return ret;
2834 }
2835
2836 /**
2837  * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket
2838  * @args: rpc transport creation arguments
2839  *
2840  */
2841 static struct rpc_xprt *xs_setup_bc_tcp(struct xprt_create *args)
2842 {
2843         struct sockaddr *addr = args->dstaddr;
2844         struct rpc_xprt *xprt;
2845         struct sock_xprt *transport;
2846         struct svc_sock *bc_sock;
2847         struct rpc_xprt *ret;
2848
2849         if (args->bc_xprt->xpt_bc_xprt) {
2850                 /*
2851                  * This server connection already has a backchannel
2852                  * export; we can't create a new one, as we wouldn't be
2853                  * able to match replies based on xid any more.  So,
2854                  * reuse the already-existing one:
2855                  */
2856                  return args->bc_xprt->xpt_bc_xprt;
2857         }
2858         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2859                         xprt_tcp_slot_table_entries);
2860         if (IS_ERR(xprt))
2861                 return xprt;
2862         transport = container_of(xprt, struct sock_xprt, xprt);
2863
2864         xprt->prot = IPPROTO_TCP;
2865         xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2866         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2867         xprt->timeout = &xs_tcp_default_timeout;
2868
2869         /* backchannel */
2870         xprt_set_bound(xprt);
2871         xprt->bind_timeout = 0;
2872         xprt->reestablish_timeout = 0;
2873         xprt->idle_timeout = 0;
2874
2875         xprt->ops = &bc_tcp_ops;
2876
2877         switch (addr->sa_family) {
2878         case AF_INET:
2879                 xs_format_peer_addresses(xprt, "tcp",
2880                                          RPCBIND_NETID_TCP);
2881                 break;
2882         case AF_INET6:
2883                 xs_format_peer_addresses(xprt, "tcp",
2884                                    RPCBIND_NETID_TCP6);
2885                 break;
2886         default:
2887                 ret = ERR_PTR(-EAFNOSUPPORT);
2888                 goto out_err;
2889         }
2890
2891         dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
2892                         xprt->address_strings[RPC_DISPLAY_ADDR],
2893                         xprt->address_strings[RPC_DISPLAY_PORT],
2894                         xprt->address_strings[RPC_DISPLAY_PROTO]);
2895
2896         /*
2897          * Once we've associated a backchannel xprt with a connection,
2898          * we want to keep it around as long as long as the connection
2899          * lasts, in case we need to start using it for a backchannel
2900          * again; this reference won't be dropped until bc_xprt is
2901          * destroyed.
2902          */
2903         xprt_get(xprt);
2904         args->bc_xprt->xpt_bc_xprt = xprt;
2905         xprt->bc_xprt = args->bc_xprt;
2906         bc_sock = container_of(args->bc_xprt, struct svc_sock, sk_xprt);
2907         transport->sock = bc_sock->sk_sock;
2908         transport->inet = bc_sock->sk_sk;
2909
2910         /*
2911          * Since we don't want connections for the backchannel, we set
2912          * the xprt status to connected
2913          */
2914         xprt_set_connected(xprt);
2915
2916
2917         if (try_module_get(THIS_MODULE))
2918                 return xprt;
2919         xprt_put(xprt);
2920         ret = ERR_PTR(-EINVAL);
2921 out_err:
2922         xprt_free(xprt);
2923         return ret;
2924 }
2925
2926 static struct xprt_class        xs_local_transport = {
2927         .list           = LIST_HEAD_INIT(xs_local_transport.list),
2928         .name           = "named UNIX socket",
2929         .owner          = THIS_MODULE,
2930         .ident          = XPRT_TRANSPORT_LOCAL,
2931         .setup          = xs_setup_local,
2932 };
2933
2934 static struct xprt_class        xs_udp_transport = {
2935         .list           = LIST_HEAD_INIT(xs_udp_transport.list),
2936         .name           = "udp",
2937         .owner          = THIS_MODULE,
2938         .ident          = XPRT_TRANSPORT_UDP,
2939         .setup          = xs_setup_udp,
2940 };
2941
2942 static struct xprt_class        xs_tcp_transport = {
2943         .list           = LIST_HEAD_INIT(xs_tcp_transport.list),
2944         .name           = "tcp",
2945         .owner          = THIS_MODULE,
2946         .ident          = XPRT_TRANSPORT_TCP,
2947         .setup          = xs_setup_tcp,
2948 };
2949
2950 static struct xprt_class        xs_bc_tcp_transport = {
2951         .list           = LIST_HEAD_INIT(xs_bc_tcp_transport.list),
2952         .name           = "tcp NFSv4.1 backchannel",
2953         .owner          = THIS_MODULE,
2954         .ident          = XPRT_TRANSPORT_BC_TCP,
2955         .setup          = xs_setup_bc_tcp,
2956 };
2957
2958 /**
2959  * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
2960  *
2961  */
2962 int init_socket_xprt(void)
2963 {
2964 #ifdef RPC_DEBUG
2965         if (!sunrpc_table_header)
2966                 sunrpc_table_header = register_sysctl_table(sunrpc_table);
2967 #endif
2968
2969         xprt_register_transport(&xs_local_transport);
2970         xprt_register_transport(&xs_udp_transport);
2971         xprt_register_transport(&xs_tcp_transport);
2972         xprt_register_transport(&xs_bc_tcp_transport);
2973
2974         return 0;
2975 }
2976
2977 /**
2978  * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
2979  *
2980  */
2981 void cleanup_socket_xprt(void)
2982 {
2983 #ifdef RPC_DEBUG
2984         if (sunrpc_table_header) {
2985                 unregister_sysctl_table(sunrpc_table_header);
2986                 sunrpc_table_header = NULL;
2987         }
2988 #endif
2989
2990         xprt_unregister_transport(&xs_local_transport);
2991         xprt_unregister_transport(&xs_udp_transport);
2992         xprt_unregister_transport(&xs_tcp_transport);
2993         xprt_unregister_transport(&xs_bc_tcp_transport);
2994 }
2995
2996 static int param_set_uint_minmax(const char *val,
2997                 const struct kernel_param *kp,
2998                 unsigned int min, unsigned int max)
2999 {
3000         unsigned long num;
3001         int ret;
3002
3003         if (!val)
3004                 return -EINVAL;
3005         ret = strict_strtoul(val, 0, &num);
3006         if (ret == -EINVAL || num < min || num > max)
3007                 return -EINVAL;
3008         *((unsigned int *)kp->arg) = num;
3009         return 0;
3010 }
3011
3012 static int param_set_portnr(const char *val, const struct kernel_param *kp)
3013 {
3014         return param_set_uint_minmax(val, kp,
3015                         RPC_MIN_RESVPORT,
3016                         RPC_MAX_RESVPORT);
3017 }
3018
3019 static struct kernel_param_ops param_ops_portnr = {
3020         .set = param_set_portnr,
3021         .get = param_get_uint,
3022 };
3023
3024 #define param_check_portnr(name, p) \
3025         __param_check(name, p, unsigned int);
3026
3027 module_param_named(min_resvport, xprt_min_resvport, portnr, 0644);
3028 module_param_named(max_resvport, xprt_max_resvport, portnr, 0644);
3029
3030 static int param_set_slot_table_size(const char *val,
3031                                      const struct kernel_param *kp)
3032 {
3033         return param_set_uint_minmax(val, kp,
3034                         RPC_MIN_SLOT_TABLE,
3035                         RPC_MAX_SLOT_TABLE);
3036 }
3037
3038 static struct kernel_param_ops param_ops_slot_table_size = {
3039         .set = param_set_slot_table_size,
3040         .get = param_get_uint,
3041 };
3042
3043 #define param_check_slot_table_size(name, p) \
3044         __param_check(name, p, unsigned int);
3045
3046 static int param_set_max_slot_table_size(const char *val,
3047                                      const struct kernel_param *kp)
3048 {
3049         return param_set_uint_minmax(val, kp,
3050                         RPC_MIN_SLOT_TABLE,
3051                         RPC_MAX_SLOT_TABLE_LIMIT);
3052 }
3053
3054 static struct kernel_param_ops param_ops_max_slot_table_size = {
3055         .set = param_set_max_slot_table_size,
3056         .get = param_get_uint,
3057 };
3058
3059 #define param_check_max_slot_table_size(name, p) \
3060         __param_check(name, p, unsigned int);
3061
3062 module_param_named(tcp_slot_table_entries, xprt_tcp_slot_table_entries,
3063                    slot_table_size, 0644);
3064 module_param_named(tcp_max_slot_table_entries, xprt_max_tcp_slot_table_entries,
3065                    max_slot_table_size, 0644);
3066 module_param_named(udp_slot_table_entries, xprt_udp_slot_table_entries,
3067                    slot_table_size, 0644);
3068