git.oblomov.eu Git - linux-2.6/blob - net/sunrpc/xprtsock.c

   1 /*
   2  * linux/net/sunrpc/xprtsock.c
   3  *
   4  * Client-side transport implementation for sockets.
   5  *
   6  * TCP callback races fixes (C) 1998 Red Hat
   7  * TCP send fixes (C) 1998 Red Hat
   8  * TCP NFS related read + write fixes
   9  *  (C) 1999 Dave Airlie, University of Limerick, Ireland <airlied@linux.ie>
  10  *
  11  * Rewrite of larges part of the code in order to stabilize TCP stuff.
  12  * Fix behaviour when socket buffer is full.
  13  *  (C) 1999 Trond Myklebust <trond.myklebust@fys.uio.no>
  14  *
  15  * IP socket transport implementation, (C) 2005 Chuck Lever <cel@netapp.com>
  16  *
  17  * IPv6 support contributed by Gilles Quillard, Bull Open Source, 2005.
  18  *   <gilles.quillard@bull.net>
  19  */
  20
  21 #include <linux/types.h>
  22 #include <linux/slab.h>
  23 #include <linux/module.h>
  24 #include <linux/capability.h>
  25 #include <linux/pagemap.h>
  26 #include <linux/errno.h>
  27 #include <linux/socket.h>
  28 #include <linux/in.h>
  29 #include <linux/net.h>
  30 #include <linux/mm.h>
  31 #include <linux/udp.h>
  32 #include <linux/tcp.h>
  33 #include <linux/sunrpc/clnt.h>
  34 #include <linux/sunrpc/sched.h>
  35 #include <linux/sunrpc/xprtsock.h>
  36 #include <linux/file.h>
  37
  38 #include <net/sock.h>
  39 #include <net/checksum.h>
  40 #include <net/udp.h>
  41 #include <net/tcp.h>
  42
  43 /*
  44  * xprtsock tunables
  45  */
  46 unsigned int xprt_udp_slot_table_entries = RPC_DEF_SLOT_TABLE;
  47 unsigned int xprt_tcp_slot_table_entries = RPC_DEF_SLOT_TABLE;
  48
  49 unsigned int xprt_min_resvport = RPC_DEF_MIN_RESVPORT;
  50 unsigned int xprt_max_resvport = RPC_DEF_MAX_RESVPORT;
  51
  52 /*
  53  * We can register our own files under /proc/sys/sunrpc by
  54  * calling register_sysctl_table() again.  The files in that
  55  * directory become the union of all files registered there.
  56  *
  57  * We simply need to make sure that we don't collide with
  58  * someone else's file names!
  59  */
  60
  61 #ifdef RPC_DEBUG
  62
  63 static unsigned int min_slot_table_size = RPC_MIN_SLOT_TABLE;
  64 static unsigned int max_slot_table_size = RPC_MAX_SLOT_TABLE;
  65 static unsigned int xprt_min_resvport_limit = RPC_MIN_RESVPORT;
  66 static unsigned int xprt_max_resvport_limit = RPC_MAX_RESVPORT;
  67
  68 static struct ctl_table_header *sunrpc_table_header;
  69
  70 /*
  71  * FIXME: changing the UDP slot table size should also resize the UDP
  72  *        socket buffers for existing UDP transports
  73  */
  74 static ctl_table xs_tunables_table[] = {
  75         {
  76                 .ctl_name       = CTL_SLOTTABLE_UDP,
  77                 .procname       = "udp_slot_table_entries",
  78                 .data           = &xprt_udp_slot_table_entries,
  79                 .maxlen         = sizeof(unsigned int),
  80                 .mode           = 0644,
  81                 .proc_handler   = &proc_dointvec_minmax,
  82                 .strategy       = &sysctl_intvec,
  83                 .extra1         = &min_slot_table_size,
  84                 .extra2         = &max_slot_table_size
  85         },
  86         {
  87                 .ctl_name       = CTL_SLOTTABLE_TCP,
  88                 .procname       = "tcp_slot_table_entries",
  89                 .data           = &xprt_tcp_slot_table_entries,
  90                 .maxlen         = sizeof(unsigned int),
  91                 .mode           = 0644,
  92                 .proc_handler   = &proc_dointvec_minmax,
  93                 .strategy       = &sysctl_intvec,
  94                 .extra1         = &min_slot_table_size,
  95                 .extra2         = &max_slot_table_size
  96         },
  97         {
  98                 .ctl_name       = CTL_MIN_RESVPORT,
  99                 .procname       = "min_resvport",
 100                 .data           = &xprt_min_resvport,
 101                 .maxlen         = sizeof(unsigned int),
 102                 .mode           = 0644,
 103                 .proc_handler   = &proc_dointvec_minmax,
 104                 .strategy       = &sysctl_intvec,
 105                 .extra1         = &xprt_min_resvport_limit,
 106                 .extra2         = &xprt_max_resvport_limit
 107         },
 108         {
 109                 .ctl_name       = CTL_MAX_RESVPORT,
 110                 .procname       = "max_resvport",
 111                 .data           = &xprt_max_resvport,
 112                 .maxlen         = sizeof(unsigned int),
 113                 .mode           = 0644,
 114                 .proc_handler   = &proc_dointvec_minmax,
 115                 .strategy       = &sysctl_intvec,
 116                 .extra1         = &xprt_min_resvport_limit,
 117                 .extra2         = &xprt_max_resvport_limit
 118         },
 119         {
 120                 .ctl_name = 0,
 121         },
 122 };
 123
 124 static ctl_table sunrpc_table[] = {
 125         {
 126                 .ctl_name       = CTL_SUNRPC,
 127                 .procname       = "sunrpc",
 128                 .mode           = 0555,
 129                 .child          = xs_tunables_table
 130         },
 131         {
 132                 .ctl_name = 0,
 133         },
 134 };
 135
 136 #endif
 137
 138 /*
 139  * Time out for an RPC UDP socket connect.  UDP socket connects are
 140  * synchronous, but we set a timeout anyway in case of resource
 141  * exhaustion on the local host.
 142  */
 143 #define XS_UDP_CONN_TO          (5U * HZ)
 144
 145 /*
 146  * Wait duration for an RPC TCP connection to be established.  Solaris
 147  * NFS over TCP uses 60 seconds, for example, which is in line with how
 148  * long a server takes to reboot.
 149  */
 150 #define XS_TCP_CONN_TO          (60U * HZ)
 151
 152 /*
 153  * Wait duration for a reply from the RPC portmapper.
 154  */
 155 #define XS_BIND_TO              (60U * HZ)
 156
 157 /*
 158  * Delay if a UDP socket connect error occurs.  This is most likely some
 159  * kind of resource problem on the local host.
 160  */
 161 #define XS_UDP_REEST_TO         (2U * HZ)
 162
 163 /*
 164  * The reestablish timeout allows clients to delay for a bit before attempting
 165  * to reconnect to a server that just dropped our connection.
 166  *
 167  * We implement an exponential backoff when trying to reestablish a TCP
 168  * transport connection with the server.  Some servers like to drop a TCP
 169  * connection when they are overworked, so we start with a short timeout and
 170  * increase over time if the server is down or not responding.
 171  */
 172 #define XS_TCP_INIT_REEST_TO    (3U * HZ)
 173 #define XS_TCP_MAX_REEST_TO     (5U * 60 * HZ)
 174
 175 /*
 176  * TCP idle timeout; client drops the transport socket if it is idle
 177  * for this long.  Note that we also timeout UDP sockets to prevent
 178  * holding port numbers when there is no RPC traffic.
 179  */
 180 #define XS_IDLE_DISC_TO         (5U * 60 * HZ)
 181
 182 #ifdef RPC_DEBUG
 183 # undef  RPC_DEBUG_DATA
 184 # define RPCDBG_FACILITY        RPCDBG_TRANS
 185 #endif
 186
 187 #ifdef RPC_DEBUG_DATA
 188 static void xs_pktdump(char *msg, u32 *packet, unsigned int count)
 189 {
 190         u8 *buf = (u8 *) packet;
 191         int j;
 192
 193         dprintk("RPC:       %s\n", msg);
 194         for (j = 0; j < count && j < 128; j += 4) {
 195                 if (!(j & 31)) {
 196                         if (j)
 197                                 dprintk("\n");
 198                         dprintk("0x%04x ", j);
 199                 }
 200                 dprintk("%02x%02x%02x%02x ",
 201                         buf[j], buf[j+1], buf[j+2], buf[j+3]);
 202         }
 203         dprintk("\n");
 204 }
 205 #else
 206 static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count)
 207 {
 208         /* NOP */
 209 }
 210 #endif
 211
 212 struct sock_xprt {
 213         struct rpc_xprt         xprt;
 214
 215         /*
 216          * Network layer
 217          */
 218         struct socket *         sock;
 219         struct sock *           inet;
 220
 221         /*
 222          * State of TCP reply receive
 223          */
 224         __be32                  tcp_fraghdr,
 225                                 tcp_xid;
 226
 227         u32                     tcp_offset,
 228                                 tcp_reclen;
 229
 230         unsigned long           tcp_copied,
 231                                 tcp_flags;
 232
 233         /*
 234          * Connection of transports
 235          */
 236         struct delayed_work     connect_worker;
 237         struct sockaddr_storage addr;
 238         unsigned short          port;
 239
 240         /*
 241          * UDP socket buffer size parameters
 242          */
 243         size_t                  rcvsize,
 244                                 sndsize;
 245
 246         /*
 247          * Saved socket callback addresses
 248          */
 249         void                    (*old_data_ready)(struct sock *, int);
 250         void                    (*old_state_change)(struct sock *);
 251         void                    (*old_write_space)(struct sock *);
 252 };
 253
 254 /*
 255  * TCP receive state flags
 256  */
 257 #define TCP_RCV_LAST_FRAG       (1UL << 0)
 258 #define TCP_RCV_COPY_FRAGHDR    (1UL << 1)
 259 #define TCP_RCV_COPY_XID        (1UL << 2)
 260 #define TCP_RCV_COPY_DATA       (1UL << 3)
 261
 262 static inline struct sockaddr *xs_addr(struct rpc_xprt *xprt)
 263 {
 264         return (struct sockaddr *) &xprt->addr;
 265 }
 266
 267 static inline struct sockaddr_in *xs_addr_in(struct rpc_xprt *xprt)
 268 {
 269         return (struct sockaddr_in *) &xprt->addr;
 270 }
 271
 272 static inline struct sockaddr_in6 *xs_addr_in6(struct rpc_xprt *xprt)
 273 {
 274         return (struct sockaddr_in6 *) &xprt->addr;
 275 }
 276
 277 static void xs_format_ipv4_peer_addresses(struct rpc_xprt *xprt,
 278                                           const char *protocol,
 279                                           const char *netid)
 280 {
 281         struct sockaddr_in *addr = xs_addr_in(xprt);
 282         char *buf;
 283
 284         buf = kzalloc(20, GFP_KERNEL);
 285         if (buf) {
 286                 snprintf(buf, 20, NIPQUAD_FMT,
 287                                 NIPQUAD(addr->sin_addr.s_addr));
 288         }
 289         xprt->address_strings[RPC_DISPLAY_ADDR] = buf;
 290
 291         buf = kzalloc(8, GFP_KERNEL);
 292         if (buf) {
 293                 snprintf(buf, 8, "%u",
 294                                 ntohs(addr->sin_port));
 295         }
 296         xprt->address_strings[RPC_DISPLAY_PORT] = buf;
 297
 298         xprt->address_strings[RPC_DISPLAY_PROTO] = protocol;
 299
 300         buf = kzalloc(48, GFP_KERNEL);
 301         if (buf) {
 302                 snprintf(buf, 48, "addr="NIPQUAD_FMT" port=%u proto=%s",
 303                         NIPQUAD(addr->sin_addr.s_addr),
 304                         ntohs(addr->sin_port),
 305                         protocol);
 306         }
 307         xprt->address_strings[RPC_DISPLAY_ALL] = buf;
 308
 309         buf = kzalloc(10, GFP_KERNEL);
 310         if (buf) {
 311                 snprintf(buf, 10, "%02x%02x%02x%02x",
 312                                 NIPQUAD(addr->sin_addr.s_addr));
 313         }
 314         xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = buf;
 315
 316         buf = kzalloc(8, GFP_KERNEL);
 317         if (buf) {
 318                 snprintf(buf, 8, "%4hx",
 319                                 ntohs(addr->sin_port));
 320         }
 321         xprt->address_strings[RPC_DISPLAY_HEX_PORT] = buf;
 322
 323         buf = kzalloc(30, GFP_KERNEL);
 324         if (buf) {
 325                 snprintf(buf, 30, NIPQUAD_FMT".%u.%u",
 326                                 NIPQUAD(addr->sin_addr.s_addr),
 327                                 ntohs(addr->sin_port) >> 8,
 328                                 ntohs(addr->sin_port) & 0xff);
 329         }
 330         xprt->address_strings[RPC_DISPLAY_UNIVERSAL_ADDR] = buf;
 331
 332         xprt->address_strings[RPC_DISPLAY_NETID] = netid;
 333 }
 334
 335 static void xs_format_ipv6_peer_addresses(struct rpc_xprt *xprt,
 336                                           const char *protocol,
 337                                           const char *netid)
 338 {
 339         struct sockaddr_in6 *addr = xs_addr_in6(xprt);
 340         char *buf;
 341
 342         buf = kzalloc(40, GFP_KERNEL);
 343         if (buf) {
 344                 snprintf(buf, 40, "%p6",&addr->sin6_addr);
 345         }
 346         xprt->address_strings[RPC_DISPLAY_ADDR] = buf;
 347
 348         buf = kzalloc(8, GFP_KERNEL);
 349         if (buf) {
 350                 snprintf(buf, 8, "%u",
 351                                 ntohs(addr->sin6_port));
 352         }
 353         xprt->address_strings[RPC_DISPLAY_PORT] = buf;
 354
 355         xprt->address_strings[RPC_DISPLAY_PROTO] = protocol;
 356
 357         buf = kzalloc(64, GFP_KERNEL);
 358         if (buf) {
 359                 snprintf(buf, 64, "addr=%p6 port=%u proto=%s",
 360                                 &addr->sin6_addr,
 361                                 ntohs(addr->sin6_port),
 362                                 protocol);
 363         }
 364         xprt->address_strings[RPC_DISPLAY_ALL] = buf;
 365
 366         buf = kzalloc(36, GFP_KERNEL);
 367         if (buf)
 368                 snprintf(buf, 36, "%#p6", &addr->sin6_addr);
 369
 370         xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = buf;
 371
 372         buf = kzalloc(8, GFP_KERNEL);
 373         if (buf) {
 374                 snprintf(buf, 8, "%4hx",
 375                                 ntohs(addr->sin6_port));
 376         }
 377         xprt->address_strings[RPC_DISPLAY_HEX_PORT] = buf;
 378
 379         buf = kzalloc(50, GFP_KERNEL);
 380         if (buf) {
 381                 snprintf(buf, 50, "%p6.%u.%u",
 382                          &addr->sin6_addr,
 383                          ntohs(addr->sin6_port) >> 8,
 384                          ntohs(addr->sin6_port) & 0xff);
 385         }
 386         xprt->address_strings[RPC_DISPLAY_UNIVERSAL_ADDR] = buf;
 387
 388         xprt->address_strings[RPC_DISPLAY_NETID] = netid;
 389 }
 390
 391 static void xs_free_peer_addresses(struct rpc_xprt *xprt)
 392 {
 393         unsigned int i;
 394
 395         for (i = 0; i < RPC_DISPLAY_MAX; i++)
 396                 switch (i) {
 397                 case RPC_DISPLAY_PROTO:
 398                 case RPC_DISPLAY_NETID:
 399                         continue;
 400                 default:
 401                         kfree(xprt->address_strings[i]);
 402                 }
 403 }
 404
 405 #define XS_SENDMSG_FLAGS        (MSG_DONTWAIT | MSG_NOSIGNAL)
 406
 407 static int xs_send_kvec(struct socket *sock, struct sockaddr *addr, int addrlen, struct kvec *vec, unsigned int base, int more)
 408 {
 409         struct msghdr msg = {
 410                 .msg_name       = addr,
 411                 .msg_namelen    = addrlen,
 412                 .msg_flags      = XS_SENDMSG_FLAGS | (more ? MSG_MORE : 0),
 413         };
 414         struct kvec iov = {
 415                 .iov_base       = vec->iov_base + base,
 416                 .iov_len        = vec->iov_len - base,
 417         };
 418
 419         if (iov.iov_len != 0)
 420                 return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
 421         return kernel_sendmsg(sock, &msg, NULL, 0, 0);
 422 }
 423
 424 static int xs_send_pagedata(struct socket *sock, struct xdr_buf *xdr, unsigned int base, int more)
 425 {
 426         struct page **ppage;
 427         unsigned int remainder;
 428         int err, sent = 0;
 429
 430         remainder = xdr->page_len - base;
 431         base += xdr->page_base;
 432         ppage = xdr->pages + (base >> PAGE_SHIFT);
 433         base &= ~PAGE_MASK;
 434         for(;;) {
 435                 unsigned int len = min_t(unsigned int, PAGE_SIZE - base, remainder);
 436                 int flags = XS_SENDMSG_FLAGS;
 437
 438                 remainder -= len;
 439                 if (remainder != 0 || more)
 440                         flags |= MSG_MORE;
 441                 err = sock->ops->sendpage(sock, *ppage, base, len, flags);
 442                 if (remainder == 0 || err != len)
 443                         break;
 444                 sent += err;
 445                 ppage++;
 446                 base = 0;
 447         }
 448         if (sent == 0)
 449                 return err;
 450         if (err > 0)
 451                 sent += err;
 452         return sent;
 453 }
 454
 455 /**
 456  * xs_sendpages - write pages directly to a socket
 457  * @sock: socket to send on
 458  * @addr: UDP only -- address of destination
 459  * @addrlen: UDP only -- length of destination address
 460  * @xdr: buffer containing this request
 461  * @base: starting position in the buffer
 462  *
 463  */
 464 static int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base)
 465 {
 466         unsigned int remainder = xdr->len - base;
 467         int err, sent = 0;
 468
 469         if (unlikely(!sock))
 470                 return -ENOTCONN;
 471
 472         clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags);
 473         if (base != 0) {
 474                 addr = NULL;
 475                 addrlen = 0;
 476         }
 477
 478         if (base < xdr->head[0].iov_len || addr != NULL) {
 479                 unsigned int len = xdr->head[0].iov_len - base;
 480                 remainder -= len;
 481                 err = xs_send_kvec(sock, addr, addrlen, &xdr->head[0], base, remainder != 0);
 482                 if (remainder == 0 || err != len)
 483                         goto out;
 484                 sent += err;
 485                 base = 0;
 486         } else
 487                 base -= xdr->head[0].iov_len;
 488
 489         if (base < xdr->page_len) {
 490                 unsigned int len = xdr->page_len - base;
 491                 remainder -= len;
 492                 err = xs_send_pagedata(sock, xdr, base, remainder != 0);
 493                 if (remainder == 0 || err != len)
 494                         goto out;
 495                 sent += err;
 496                 base = 0;
 497         } else
 498                 base -= xdr->page_len;
 499
 500         if (base >= xdr->tail[0].iov_len)
 501                 return sent;
 502         err = xs_send_kvec(sock, NULL, 0, &xdr->tail[0], base, 0);
 503 out:
 504         if (sent == 0)
 505                 return err;
 506         if (err > 0)
 507                 sent += err;
 508         return sent;
 509 }
 510
 511 static void xs_nospace_callback(struct rpc_task *task)
 512 {
 513         struct sock_xprt *transport = container_of(task->tk_rqstp->rq_xprt, struct sock_xprt, xprt);
 514
 515         transport->inet->sk_write_pending--;
 516         clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 517 }
 518
 519 /**
 520  * xs_nospace - place task on wait queue if transmit was incomplete
 521  * @task: task to put to sleep
 522  *
 523  */
 524 static void xs_nospace(struct rpc_task *task)
 525 {
 526         struct rpc_rqst *req = task->tk_rqstp;
 527         struct rpc_xprt *xprt = req->rq_xprt;
 528         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 529
 530         dprintk("RPC: %5u xmit incomplete (%u left of %u)\n",
 531                         task->tk_pid, req->rq_slen - req->rq_bytes_sent,
 532                         req->rq_slen);
 533
 534         /* Protect against races with write_space */
 535         spin_lock_bh(&xprt->transport_lock);
 536
 537         /* Don't race with disconnect */
 538         if (xprt_connected(xprt)) {
 539                 if (test_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags)) {
 540                         /*
 541                          * Notify TCP that we're limited by the application
 542                          * window size
 543                          */
 544                         set_bit(SOCK_NOSPACE, &transport->sock->flags);
 545                         transport->inet->sk_write_pending++;
 546                         /* ...and wait for more buffer space */
 547                         xprt_wait_for_buffer_space(task, xs_nospace_callback);
 548                 }
 549         } else {
 550                 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 551                 task->tk_status = -ENOTCONN;
 552         }
 553
 554         spin_unlock_bh(&xprt->transport_lock);
 555 }
 556
 557 /**
 558  * xs_udp_send_request - write an RPC request to a UDP socket
 559  * @task: address of RPC task that manages the state of an RPC request
 560  *
 561  * Return values:
 562  *        0:    The request has been sent
 563  *   EAGAIN:    The socket was blocked, please call again later to
 564  *              complete the request
 565  * ENOTCONN:    Caller needs to invoke connect logic then call again
 566  *    other:    Some other error occured, the request was not sent
 567  */
 568 static int xs_udp_send_request(struct rpc_task *task)
 569 {
 570         struct rpc_rqst *req = task->tk_rqstp;
 571         struct rpc_xprt *xprt = req->rq_xprt;
 572         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 573         struct xdr_buf *xdr = &req->rq_snd_buf;
 574         int status;
 575
 576         xs_pktdump("packet data:",
 577                                 req->rq_svec->iov_base,
 578                                 req->rq_svec->iov_len);
 579
 580         status = xs_sendpages(transport->sock,
 581                               xs_addr(xprt),
 582                               xprt->addrlen, xdr,
 583                               req->rq_bytes_sent);
 584
 585         dprintk("RPC:       xs_udp_send_request(%u) = %d\n",
 586                         xdr->len - req->rq_bytes_sent, status);
 587
 588         if (status >= 0) {
 589                 task->tk_bytes_sent += status;
 590                 if (status >= req->rq_slen)
 591                         return 0;
 592                 /* Still some bytes left; set up for a retry later. */
 593                 status = -EAGAIN;
 594         }
 595
 596         switch (status) {
 597         case -EAGAIN:
 598                 xs_nospace(task);
 599                 break;
 600         case -ENETUNREACH:
 601         case -EPIPE:
 602         case -ECONNREFUSED:
 603                 /* When the server has died, an ICMP port unreachable message
 604                  * prompts ECONNREFUSED. */
 605                 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 606                 break;
 607         default:
 608                 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 609                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
 610                         -status);
 611         }
 612
 613         return status;
 614 }
 615
 616 /**
 617  * xs_tcp_shutdown - gracefully shut down a TCP socket
 618  * @xprt: transport
 619  *
 620  * Initiates a graceful shutdown of the TCP socket by calling the
 621  * equivalent of shutdown(SHUT_WR);
 622  */
 623 static void xs_tcp_shutdown(struct rpc_xprt *xprt)
 624 {
 625         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 626         struct socket *sock = transport->sock;
 627
 628         if (sock != NULL)
 629                 kernel_sock_shutdown(sock, SHUT_WR);
 630 }
 631
 632 static inline void xs_encode_tcp_record_marker(struct xdr_buf *buf)
 633 {
 634         u32 reclen = buf->len - sizeof(rpc_fraghdr);
 635         rpc_fraghdr *base = buf->head[0].iov_base;
 636         *base = htonl(RPC_LAST_STREAM_FRAGMENT | reclen);
 637 }
 638
 639 /**
 640  * xs_tcp_send_request - write an RPC request to a TCP socket
 641  * @task: address of RPC task that manages the state of an RPC request
 642  *
 643  * Return values:
 644  *        0:    The request has been sent
 645  *   EAGAIN:    The socket was blocked, please call again later to
 646  *              complete the request
 647  * ENOTCONN:    Caller needs to invoke connect logic then call again
 648  *    other:    Some other error occured, the request was not sent
 649  *
 650  * XXX: In the case of soft timeouts, should we eventually give up
 651  *      if sendmsg is not able to make progress?
 652  */
 653 static int xs_tcp_send_request(struct rpc_task *task)
 654 {
 655         struct rpc_rqst *req = task->tk_rqstp;
 656         struct rpc_xprt *xprt = req->rq_xprt;
 657         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 658         struct xdr_buf *xdr = &req->rq_snd_buf;
 659         int status;
 660
 661         xs_encode_tcp_record_marker(&req->rq_snd_buf);
 662
 663         xs_pktdump("packet data:",
 664                                 req->rq_svec->iov_base,
 665                                 req->rq_svec->iov_len);
 666
 667         /* Continue transmitting the packet/record. We must be careful
 668          * to cope with writespace callbacks arriving _after_ we have
 669          * called sendmsg(). */
 670         while (1) {
 671                 status = xs_sendpages(transport->sock,
 672                                         NULL, 0, xdr, req->rq_bytes_sent);
 673
 674                 dprintk("RPC:       xs_tcp_send_request(%u) = %d\n",
 675                                 xdr->len - req->rq_bytes_sent, status);
 676
 677                 if (unlikely(status < 0))
 678                         break;
 679
 680                 /* If we've sent the entire packet, immediately
 681                  * reset the count of bytes sent. */
 682                 req->rq_bytes_sent += status;
 683                 task->tk_bytes_sent += status;
 684                 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
 685                         req->rq_bytes_sent = 0;
 686                         return 0;
 687                 }
 688
 689                 if (status != 0)
 690                         continue;
 691                 status = -EAGAIN;
 692                 break;
 693         }
 694
 695         switch (status) {
 696         case -EAGAIN:
 697                 xs_nospace(task);
 698                 break;
 699         case -ECONNREFUSED:
 700         case -ECONNRESET:
 701         case -ENOTCONN:
 702         case -EPIPE:
 703                 status = -ENOTCONN;
 704                 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 705                 break;
 706         default:
 707                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
 708                         -status);
 709                 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 710                 xs_tcp_shutdown(xprt);
 711         }
 712
 713         return status;
 714 }
 715
 716 /**
 717  * xs_tcp_release_xprt - clean up after a tcp transmission
 718  * @xprt: transport
 719  * @task: rpc task
 720  *
 721  * This cleans up if an error causes us to abort the transmission of a request.
 722  * In this case, the socket may need to be reset in order to avoid confusing
 723  * the server.
 724  */
 725 static void xs_tcp_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
 726 {
 727         struct rpc_rqst *req;
 728
 729         if (task != xprt->snd_task)
 730                 return;
 731         if (task == NULL)
 732                 goto out_release;
 733         req = task->tk_rqstp;
 734         if (req->rq_bytes_sent == 0)
 735                 goto out_release;
 736         if (req->rq_bytes_sent == req->rq_snd_buf.len)
 737                 goto out_release;
 738         set_bit(XPRT_CLOSE_WAIT, &task->tk_xprt->state);
 739 out_release:
 740         xprt_release_xprt(xprt, task);
 741 }
 742
 743 /**
 744  * xs_close - close a socket
 745  * @xprt: transport
 746  *
 747  * This is used when all requests are complete; ie, no DRC state remains
 748  * on the server we want to save.
 749  */
 750 static void xs_close(struct rpc_xprt *xprt)
 751 {
 752         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 753         struct socket *sock = transport->sock;
 754         struct sock *sk = transport->inet;
 755
 756         if (!sk)
 757                 goto clear_close_wait;
 758
 759         dprintk("RPC:       xs_close xprt %p\n", xprt);
 760
 761         write_lock_bh(&sk->sk_callback_lock);
 762         transport->inet = NULL;
 763         transport->sock = NULL;
 764
 765         sk->sk_user_data = NULL;
 766         sk->sk_data_ready = transport->old_data_ready;
 767         sk->sk_state_change = transport->old_state_change;
 768         sk->sk_write_space = transport->old_write_space;
 769         write_unlock_bh(&sk->sk_callback_lock);
 770
 771         sk->sk_no_check = 0;
 772
 773         sock_release(sock);
 774 clear_close_wait:
 775         smp_mb__before_clear_bit();
 776         clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
 777         clear_bit(XPRT_CLOSING, &xprt->state);
 778         smp_mb__after_clear_bit();
 779         xprt_disconnect_done(xprt);
 780 }
 781
 782 /**
 783  * xs_destroy - prepare to shutdown a transport
 784  * @xprt: doomed transport
 785  *
 786  */
 787 static void xs_destroy(struct rpc_xprt *xprt)
 788 {
 789         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 790
 791         dprintk("RPC:       xs_destroy xprt %p\n", xprt);
 792
 793         cancel_rearming_delayed_work(&transport->connect_worker);
 794
 795         xs_close(xprt);
 796         xs_free_peer_addresses(xprt);
 797         kfree(xprt->slot);
 798         kfree(xprt);
 799         module_put(THIS_MODULE);
 800 }
 801
 802 static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
 803 {
 804         return (struct rpc_xprt *) sk->sk_user_data;
 805 }
 806
 807 /**
 808  * xs_udp_data_ready - "data ready" callback for UDP sockets
 809  * @sk: socket with data to read
 810  * @len: how much data to read
 811  *
 812  */
 813 static void xs_udp_data_ready(struct sock *sk, int len)
 814 {
 815         struct rpc_task *task;
 816         struct rpc_xprt *xprt;
 817         struct rpc_rqst *rovr;
 818         struct sk_buff *skb;
 819         int err, repsize, copied;
 820         u32 _xid;
 821         __be32 *xp;
 822
 823         read_lock(&sk->sk_callback_lock);
 824         dprintk("RPC:       xs_udp_data_ready...\n");
 825         if (!(xprt = xprt_from_sock(sk)))
 826                 goto out;
 827
 828         if ((skb = skb_recv_datagram(sk, 0, 1, &err)) == NULL)
 829                 goto out;
 830
 831         if (xprt->shutdown)
 832                 goto dropit;
 833
 834         repsize = skb->len - sizeof(struct udphdr);
 835         if (repsize < 4) {
 836                 dprintk("RPC:       impossible RPC reply size %d!\n", repsize);
 837                 goto dropit;
 838         }
 839
 840         /* Copy the XID from the skb... */
 841         xp = skb_header_pointer(skb, sizeof(struct udphdr),
 842                                 sizeof(_xid), &_xid);
 843         if (xp == NULL)
 844                 goto dropit;
 845
 846         /* Look up and lock the request corresponding to the given XID */
 847         spin_lock(&xprt->transport_lock);
 848         rovr = xprt_lookup_rqst(xprt, *xp);
 849         if (!rovr)
 850                 goto out_unlock;
 851         task = rovr->rq_task;
 852
 853         if ((copied = rovr->rq_private_buf.buflen) > repsize)
 854                 copied = repsize;
 855
 856         /* Suck it into the iovec, verify checksum if not done by hw. */
 857         if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
 858                 UDPX_INC_STATS_BH(sk, UDP_MIB_INERRORS);
 859                 goto out_unlock;
 860         }
 861
 862         UDPX_INC_STATS_BH(sk, UDP_MIB_INDATAGRAMS);
 863
 864         /* Something worked... */
 865         dst_confirm(skb->dst);
 866
 867         xprt_adjust_cwnd(task, copied);
 868         xprt_update_rtt(task);
 869         xprt_complete_rqst(task, copied);
 870
 871  out_unlock:
 872         spin_unlock(&xprt->transport_lock);
 873  dropit:
 874         skb_free_datagram(sk, skb);
 875  out:
 876         read_unlock(&sk->sk_callback_lock);
 877 }
 878
 879 static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
 880 {
 881         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 882         size_t len, used;
 883         char *p;
 884
 885         p = ((char *) &transport->tcp_fraghdr) + transport->tcp_offset;
 886         len = sizeof(transport->tcp_fraghdr) - transport->tcp_offset;
 887         used = xdr_skb_read_bits(desc, p, len);
 888         transport->tcp_offset += used;
 889         if (used != len)
 890                 return;
 891
 892         transport->tcp_reclen = ntohl(transport->tcp_fraghdr);
 893         if (transport->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
 894                 transport->tcp_flags |= TCP_RCV_LAST_FRAG;
 895         else
 896                 transport->tcp_flags &= ~TCP_RCV_LAST_FRAG;
 897         transport->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
 898
 899         transport->tcp_flags &= ~TCP_RCV_COPY_FRAGHDR;
 900         transport->tcp_offset = 0;
 901
 902         /* Sanity check of the record length */
 903         if (unlikely(transport->tcp_reclen < 4)) {
 904                 dprintk("RPC:       invalid TCP record fragment length\n");
 905                 xprt_force_disconnect(xprt);
 906                 return;
 907         }
 908         dprintk("RPC:       reading TCP record fragment of length %d\n",
 909                         transport->tcp_reclen);
 910 }
 911
 912 static void xs_tcp_check_fraghdr(struct sock_xprt *transport)
 913 {
 914         if (transport->tcp_offset == transport->tcp_reclen) {
 915                 transport->tcp_flags |= TCP_RCV_COPY_FRAGHDR;
 916                 transport->tcp_offset = 0;
 917                 if (transport->tcp_flags & TCP_RCV_LAST_FRAG) {
 918                         transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
 919                         transport->tcp_flags |= TCP_RCV_COPY_XID;
 920                         transport->tcp_copied = 0;
 921                 }
 922         }
 923 }
 924
 925 static inline void xs_tcp_read_xid(struct sock_xprt *transport, struct xdr_skb_reader *desc)
 926 {
 927         size_t len, used;
 928         char *p;
 929
 930         len = sizeof(transport->tcp_xid) - transport->tcp_offset;
 931         dprintk("RPC:       reading XID (%Zu bytes)\n", len);
 932         p = ((char *) &transport->tcp_xid) + transport->tcp_offset;
 933         used = xdr_skb_read_bits(desc, p, len);
 934         transport->tcp_offset += used;
 935         if (used != len)
 936                 return;
 937         transport->tcp_flags &= ~TCP_RCV_COPY_XID;
 938         transport->tcp_flags |= TCP_RCV_COPY_DATA;
 939         transport->tcp_copied = 4;
 940         dprintk("RPC:       reading reply for XID %08x\n",
 941                         ntohl(transport->tcp_xid));
 942         xs_tcp_check_fraghdr(transport);
 943 }
 944
 945 static inline void xs_tcp_read_request(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
 946 {
 947         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 948         struct rpc_rqst *req;
 949         struct xdr_buf *rcvbuf;
 950         size_t len;
 951         ssize_t r;
 952
 953         /* Find and lock the request corresponding to this xid */
 954         spin_lock(&xprt->transport_lock);
 955         req = xprt_lookup_rqst(xprt, transport->tcp_xid);
 956         if (!req) {
 957                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
 958                 dprintk("RPC:       XID %08x request not found!\n",
 959                                 ntohl(transport->tcp_xid));
 960                 spin_unlock(&xprt->transport_lock);
 961                 return;
 962         }
 963
 964         rcvbuf = &req->rq_private_buf;
 965         len = desc->count;
 966         if (len > transport->tcp_reclen - transport->tcp_offset) {
 967                 struct xdr_skb_reader my_desc;
 968
 969                 len = transport->tcp_reclen - transport->tcp_offset;
 970                 memcpy(&my_desc, desc, sizeof(my_desc));
 971                 my_desc.count = len;
 972                 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
 973                                           &my_desc, xdr_skb_read_bits);
 974                 desc->count -= r;
 975                 desc->offset += r;
 976         } else
 977                 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
 978                                           desc, xdr_skb_read_bits);
 979
 980         if (r > 0) {
 981                 transport->tcp_copied += r;
 982                 transport->tcp_offset += r;
 983         }
 984         if (r != len) {
 985                 /* Error when copying to the receive buffer,
 986                  * usually because we weren't able to allocate
 987                  * additional buffer pages. All we can do now
 988                  * is turn off TCP_RCV_COPY_DATA, so the request
 989                  * will not receive any additional updates,
 990                  * and time out.
 991                  * Any remaining data from this record will
 992                  * be discarded.
 993                  */
 994                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
 995                 dprintk("RPC:       XID %08x truncated request\n",
 996                                 ntohl(transport->tcp_xid));
 997                 dprintk("RPC:       xprt = %p, tcp_copied = %lu, "
 998                                 "tcp_offset = %u, tcp_reclen = %u\n",
 999                                 xprt, transport->tcp_copied,
1000                                 transport->tcp_offset, transport->tcp_reclen);
1001                 goto out;
1002         }
1003
1004         dprintk("RPC:       XID %08x read %Zd bytes\n",
1005                         ntohl(transport->tcp_xid), r);
1006         dprintk("RPC:       xprt = %p, tcp_copied = %lu, tcp_offset = %u, "
1007                         "tcp_reclen = %u\n", xprt, transport->tcp_copied,
1008                         transport->tcp_offset, transport->tcp_reclen);
1009
1010         if (transport->tcp_copied == req->rq_private_buf.buflen)
1011                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1012         else if (transport->tcp_offset == transport->tcp_reclen) {
1013                 if (transport->tcp_flags & TCP_RCV_LAST_FRAG)
1014                         transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1015         }
1016
1017 out:
1018         if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1019                 xprt_complete_rqst(req->rq_task, transport->tcp_copied);
1020         spin_unlock(&xprt->transport_lock);
1021         xs_tcp_check_fraghdr(transport);
1022 }
1023
1024 static inline void xs_tcp_read_discard(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1025 {
1026         size_t len;
1027
1028         len = transport->tcp_reclen - transport->tcp_offset;
1029         if (len > desc->count)
1030                 len = desc->count;
1031         desc->count -= len;
1032         desc->offset += len;
1033         transport->tcp_offset += len;
1034         dprintk("RPC:       discarded %Zu bytes\n", len);
1035         xs_tcp_check_fraghdr(transport);
1036 }
1037
1038 static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
1039 {
1040         struct rpc_xprt *xprt = rd_desc->arg.data;
1041         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1042         struct xdr_skb_reader desc = {
1043                 .skb    = skb,
1044                 .offset = offset,
1045                 .count  = len,
1046         };
1047
1048         dprintk("RPC:       xs_tcp_data_recv started\n");
1049         do {
1050                 /* Read in a new fragment marker if necessary */
1051                 /* Can we ever really expect to get completely empty fragments? */
1052                 if (transport->tcp_flags & TCP_RCV_COPY_FRAGHDR) {
1053                         xs_tcp_read_fraghdr(xprt, &desc);
1054                         continue;
1055                 }
1056                 /* Read in the xid if necessary */
1057                 if (transport->tcp_flags & TCP_RCV_COPY_XID) {
1058                         xs_tcp_read_xid(transport, &desc);
1059                         continue;
1060                 }
1061                 /* Read in the request data */
1062                 if (transport->tcp_flags & TCP_RCV_COPY_DATA) {
1063                         xs_tcp_read_request(xprt, &desc);
1064                         continue;
1065                 }
1066                 /* Skip over any trailing bytes on short reads */
1067                 xs_tcp_read_discard(transport, &desc);
1068         } while (desc.count);
1069         dprintk("RPC:       xs_tcp_data_recv done\n");
1070         return len - desc.count;
1071 }
1072
1073 /**
1074  * xs_tcp_data_ready - "data ready" callback for TCP sockets
1075  * @sk: socket with data to read
1076  * @bytes: how much data to read
1077  *
1078  */
1079 static void xs_tcp_data_ready(struct sock *sk, int bytes)
1080 {
1081         struct rpc_xprt *xprt;
1082         read_descriptor_t rd_desc;
1083         int read;
1084
1085         dprintk("RPC:       xs_tcp_data_ready...\n");
1086
1087         read_lock(&sk->sk_callback_lock);
1088         if (!(xprt = xprt_from_sock(sk)))
1089                 goto out;
1090         if (xprt->shutdown)
1091                 goto out;
1092
1093         /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
1094         rd_desc.arg.data = xprt;
1095         do {
1096                 rd_desc.count = 65536;
1097                 read = tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
1098         } while (read > 0);
1099 out:
1100         read_unlock(&sk->sk_callback_lock);
1101 }
1102
1103 /**
1104  * xs_tcp_state_change - callback to handle TCP socket state changes
1105  * @sk: socket whose state has changed
1106  *
1107  */
1108 static void xs_tcp_state_change(struct sock *sk)
1109 {
1110         struct rpc_xprt *xprt;
1111
1112         read_lock(&sk->sk_callback_lock);
1113         if (!(xprt = xprt_from_sock(sk)))
1114                 goto out;
1115         dprintk("RPC:       xs_tcp_state_change client %p...\n", xprt);
1116         dprintk("RPC:       state %x conn %d dead %d zapped %d\n",
1117                         sk->sk_state, xprt_connected(xprt),
1118                         sock_flag(sk, SOCK_DEAD),
1119                         sock_flag(sk, SOCK_ZAPPED));
1120
1121         switch (sk->sk_state) {
1122         case TCP_ESTABLISHED:
1123                 spin_lock_bh(&xprt->transport_lock);
1124                 if (!xprt_test_and_set_connected(xprt)) {
1125                         struct sock_xprt *transport = container_of(xprt,
1126                                         struct sock_xprt, xprt);
1127
1128                         /* Reset TCP record info */
1129                         transport->tcp_offset = 0;
1130                         transport->tcp_reclen = 0;
1131                         transport->tcp_copied = 0;
1132                         transport->tcp_flags =
1133                                 TCP_RCV_COPY_FRAGHDR | TCP_RCV_COPY_XID;
1134
1135                         xprt_wake_pending_tasks(xprt, 0);
1136                 }
1137                 spin_unlock_bh(&xprt->transport_lock);
1138                 break;
1139         case TCP_FIN_WAIT1:
1140                 /* The client initiated a shutdown of the socket */
1141                 xprt->connect_cookie++;
1142                 xprt->reestablish_timeout = 0;
1143                 set_bit(XPRT_CLOSING, &xprt->state);
1144                 smp_mb__before_clear_bit();
1145                 clear_bit(XPRT_CONNECTED, &xprt->state);
1146                 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1147                 smp_mb__after_clear_bit();
1148                 break;
1149         case TCP_CLOSE_WAIT:
1150                 /* The server initiated a shutdown of the socket */
1151                 set_bit(XPRT_CLOSING, &xprt->state);
1152                 xprt_force_disconnect(xprt);
1153         case TCP_SYN_SENT:
1154                 xprt->connect_cookie++;
1155         case TCP_CLOSING:
1156                 /*
1157                  * If the server closed down the connection, make sure that
1158                  * we back off before reconnecting
1159                  */
1160                 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1161                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1162                 break;
1163         case TCP_LAST_ACK:
1164                 smp_mb__before_clear_bit();
1165                 clear_bit(XPRT_CONNECTED, &xprt->state);
1166                 smp_mb__after_clear_bit();
1167                 break;
1168         case TCP_CLOSE:
1169                 smp_mb__before_clear_bit();
1170                 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1171                 clear_bit(XPRT_CLOSING, &xprt->state);
1172                 smp_mb__after_clear_bit();
1173                 /* Mark transport as closed and wake up all pending tasks */
1174                 xprt_disconnect_done(xprt);
1175         }
1176  out:
1177         read_unlock(&sk->sk_callback_lock);
1178 }
1179
1180 /**
1181  * xs_udp_write_space - callback invoked when socket buffer space
1182  *                             becomes available
1183  * @sk: socket whose state has changed
1184  *
1185  * Called when more output buffer space is available for this socket.
1186  * We try not to wake our writers until they can make "significant"
1187  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1188  * with a bunch of small requests.
1189  */
1190 static void xs_udp_write_space(struct sock *sk)
1191 {
1192         read_lock(&sk->sk_callback_lock);
1193
1194         /* from net/core/sock.c:sock_def_write_space */
1195         if (sock_writeable(sk)) {
1196                 struct socket *sock;
1197                 struct rpc_xprt *xprt;
1198
1199                 if (unlikely(!(sock = sk->sk_socket)))
1200                         goto out;
1201                 clear_bit(SOCK_NOSPACE, &sock->flags);
1202
1203                 if (unlikely(!(xprt = xprt_from_sock(sk))))
1204                         goto out;
1205                 if (test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags) == 0)
1206                         goto out;
1207
1208                 xprt_write_space(xprt);
1209         }
1210
1211  out:
1212         read_unlock(&sk->sk_callback_lock);
1213 }
1214
1215 /**
1216  * xs_tcp_write_space - callback invoked when socket buffer space
1217  *                             becomes available
1218  * @sk: socket whose state has changed
1219  *
1220  * Called when more output buffer space is available for this socket.
1221  * We try not to wake our writers until they can make "significant"
1222  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1223  * with a bunch of small requests.
1224  */
1225 static void xs_tcp_write_space(struct sock *sk)
1226 {
1227         read_lock(&sk->sk_callback_lock);
1228
1229         /* from net/core/stream.c:sk_stream_write_space */
1230         if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk)) {
1231                 struct socket *sock;
1232                 struct rpc_xprt *xprt;
1233
1234                 if (unlikely(!(sock = sk->sk_socket)))
1235                         goto out;
1236                 clear_bit(SOCK_NOSPACE, &sock->flags);
1237
1238                 if (unlikely(!(xprt = xprt_from_sock(sk))))
1239                         goto out;
1240                 if (test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags) == 0)
1241                         goto out;
1242
1243                 xprt_write_space(xprt);
1244         }
1245
1246  out:
1247         read_unlock(&sk->sk_callback_lock);
1248 }
1249
1250 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
1251 {
1252         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1253         struct sock *sk = transport->inet;
1254
1255         if (transport->rcvsize) {
1256                 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1257                 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
1258         }
1259         if (transport->sndsize) {
1260                 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1261                 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1262                 sk->sk_write_space(sk);
1263         }
1264 }
1265
1266 /**
1267  * xs_udp_set_buffer_size - set send and receive limits
1268  * @xprt: generic transport
1269  * @sndsize: requested size of send buffer, in bytes
1270  * @rcvsize: requested size of receive buffer, in bytes
1271  *
1272  * Set socket send and receive buffer size limits.
1273  */
1274 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1275 {
1276         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1277
1278         transport->sndsize = 0;
1279         if (sndsize)
1280                 transport->sndsize = sndsize + 1024;
1281         transport->rcvsize = 0;
1282         if (rcvsize)
1283                 transport->rcvsize = rcvsize + 1024;
1284
1285         xs_udp_do_set_buffer_size(xprt);
1286 }
1287
1288 /**
1289  * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1290  * @task: task that timed out
1291  *
1292  * Adjust the congestion window after a retransmit timeout has occurred.
1293  */
1294 static void xs_udp_timer(struct rpc_task *task)
1295 {
1296         xprt_adjust_cwnd(task, -ETIMEDOUT);
1297 }
1298
1299 static unsigned short xs_get_random_port(void)
1300 {
1301         unsigned short range = xprt_max_resvport - xprt_min_resvport;
1302         unsigned short rand = (unsigned short) net_random() % range;
1303         return rand + xprt_min_resvport;
1304 }
1305
1306 /**
1307  * xs_set_port - reset the port number in the remote endpoint address
1308  * @xprt: generic transport
1309  * @port: new port number
1310  *
1311  */
1312 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1313 {
1314         struct sockaddr *addr = xs_addr(xprt);
1315
1316         dprintk("RPC:       setting port for xprt %p to %u\n", xprt, port);
1317
1318         switch (addr->sa_family) {
1319         case AF_INET:
1320                 ((struct sockaddr_in *)addr)->sin_port = htons(port);
1321                 break;
1322         case AF_INET6:
1323                 ((struct sockaddr_in6 *)addr)->sin6_port = htons(port);
1324                 break;
1325         default:
1326                 BUG();
1327         }
1328 }
1329
1330 static unsigned short xs_get_srcport(struct sock_xprt *transport, struct socket *sock)
1331 {
1332         unsigned short port = transport->port;
1333
1334         if (port == 0 && transport->xprt.resvport)
1335                 port = xs_get_random_port();
1336         return port;
1337 }
1338
1339 static unsigned short xs_next_srcport(struct sock_xprt *transport, struct socket *sock, unsigned short port)
1340 {
1341         if (transport->port != 0)
1342                 transport->port = 0;
1343         if (!transport->xprt.resvport)
1344                 return 0;
1345         if (port <= xprt_min_resvport || port > xprt_max_resvport)
1346                 return xprt_max_resvport;
1347         return --port;
1348 }
1349
1350 static int xs_bind4(struct sock_xprt *transport, struct socket *sock)
1351 {
1352         struct sockaddr_in myaddr = {
1353                 .sin_family = AF_INET,
1354         };
1355         struct sockaddr_in *sa;
1356         int err, nloop = 0;
1357         unsigned short port = xs_get_srcport(transport, sock);
1358         unsigned short last;
1359
1360         sa = (struct sockaddr_in *)&transport->addr;
1361         myaddr.sin_addr = sa->sin_addr;
1362         do {
1363                 myaddr.sin_port = htons(port);
1364                 err = kernel_bind(sock, (struct sockaddr *) &myaddr,
1365                                                 sizeof(myaddr));
1366                 if (port == 0)
1367                         break;
1368                 if (err == 0) {
1369                         transport->port = port;
1370                         break;
1371                 }
1372                 last = port;
1373                 port = xs_next_srcport(transport, sock, port);
1374                 if (port > last)
1375                         nloop++;
1376         } while (err == -EADDRINUSE && nloop != 2);
1377         dprintk("RPC:       %s "NIPQUAD_FMT":%u: %s (%d)\n",
1378                         __func__, NIPQUAD(myaddr.sin_addr),
1379                         port, err ? "failed" : "ok", err);
1380         return err;
1381 }
1382
1383 static int xs_bind6(struct sock_xprt *transport, struct socket *sock)
1384 {
1385         struct sockaddr_in6 myaddr = {
1386                 .sin6_family = AF_INET6,
1387         };
1388         struct sockaddr_in6 *sa;
1389         int err, nloop = 0;
1390         unsigned short port = xs_get_srcport(transport, sock);
1391         unsigned short last;
1392
1393         sa = (struct sockaddr_in6 *)&transport->addr;
1394         myaddr.sin6_addr = sa->sin6_addr;
1395         do {
1396                 myaddr.sin6_port = htons(port);
1397                 err = kernel_bind(sock, (struct sockaddr *) &myaddr,
1398                                                 sizeof(myaddr));
1399                 if (port == 0)
1400                         break;
1401                 if (err == 0) {
1402                         transport->port = port;
1403                         break;
1404                 }
1405                 last = port;
1406                 port = xs_next_srcport(transport, sock, port);
1407                 if (port > last)
1408                         nloop++;
1409         } while (err == -EADDRINUSE && nloop != 2);
1410         dprintk("RPC:       xs_bind6 %p6:%u: %s (%d)\n",
1411                 &myaddr.sin6_addr, port, err ? "failed" : "ok", err);
1412         return err;
1413 }
1414
1415 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1416 static struct lock_class_key xs_key[2];
1417 static struct lock_class_key xs_slock_key[2];
1418
1419 static inline void xs_reclassify_socket4(struct socket *sock)
1420 {
1421         struct sock *sk = sock->sk;
1422
1423         BUG_ON(sock_owned_by_user(sk));
1424         sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1425                 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1426 }
1427
1428 static inline void xs_reclassify_socket6(struct socket *sock)
1429 {
1430         struct sock *sk = sock->sk;
1431
1432         BUG_ON(sock_owned_by_user(sk));
1433         sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1434                 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
1435 }
1436 #else
1437 static inline void xs_reclassify_socket4(struct socket *sock)
1438 {
1439 }
1440
1441 static inline void xs_reclassify_socket6(struct socket *sock)
1442 {
1443 }
1444 #endif
1445
1446 static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
1447 {
1448         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1449
1450         if (!transport->inet) {
1451                 struct sock *sk = sock->sk;
1452
1453                 write_lock_bh(&sk->sk_callback_lock);
1454
1455                 sk->sk_user_data = xprt;
1456                 transport->old_data_ready = sk->sk_data_ready;
1457                 transport->old_state_change = sk->sk_state_change;
1458                 transport->old_write_space = sk->sk_write_space;
1459                 sk->sk_data_ready = xs_udp_data_ready;
1460                 sk->sk_write_space = xs_udp_write_space;
1461                 sk->sk_no_check = UDP_CSUM_NORCV;
1462                 sk->sk_allocation = GFP_ATOMIC;
1463
1464                 xprt_set_connected(xprt);
1465
1466                 /* Reset to new socket */
1467                 transport->sock = sock;
1468                 transport->inet = sk;
1469
1470                 write_unlock_bh(&sk->sk_callback_lock);
1471         }
1472         xs_udp_do_set_buffer_size(xprt);
1473 }
1474
1475 /**
1476  * xs_udp_connect_worker4 - set up a UDP socket
1477  * @work: RPC transport to connect
1478  *
1479  * Invoked by a work queue tasklet.
1480  */
1481 static void xs_udp_connect_worker4(struct work_struct *work)
1482 {
1483         struct sock_xprt *transport =
1484                 container_of(work, struct sock_xprt, connect_worker.work);
1485         struct rpc_xprt *xprt = &transport->xprt;
1486         struct socket *sock = transport->sock;
1487         int err, status = -EIO;
1488
1489         if (xprt->shutdown || !xprt_bound(xprt))
1490                 goto out;
1491
1492         /* Start by resetting any existing state */
1493         xs_close(xprt);
1494
1495         if ((err = sock_create_kern(PF_INET, SOCK_DGRAM, IPPROTO_UDP, &sock)) < 0) {
1496                 dprintk("RPC:       can't create UDP transport socket (%d).\n", -err);
1497                 goto out;
1498         }
1499         xs_reclassify_socket4(sock);
1500
1501         if (xs_bind4(transport, sock)) {
1502                 sock_release(sock);
1503                 goto out;
1504         }
1505
1506         dprintk("RPC:       worker connecting xprt %p to address: %s\n",
1507                         xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1508
1509         xs_udp_finish_connecting(xprt, sock);
1510         status = 0;
1511 out:
1512         xprt_wake_pending_tasks(xprt, status);
1513         xprt_clear_connecting(xprt);
1514 }
1515
1516 /**
1517  * xs_udp_connect_worker6 - set up a UDP socket
1518  * @work: RPC transport to connect
1519  *
1520  * Invoked by a work queue tasklet.
1521  */
1522 static void xs_udp_connect_worker6(struct work_struct *work)
1523 {
1524         struct sock_xprt *transport =
1525                 container_of(work, struct sock_xprt, connect_worker.work);
1526         struct rpc_xprt *xprt = &transport->xprt;
1527         struct socket *sock = transport->sock;
1528         int err, status = -EIO;
1529
1530         if (xprt->shutdown || !xprt_bound(xprt))
1531                 goto out;
1532
1533         /* Start by resetting any existing state */
1534         xs_close(xprt);
1535
1536         if ((err = sock_create_kern(PF_INET6, SOCK_DGRAM, IPPROTO_UDP, &sock)) < 0) {
1537                 dprintk("RPC:       can't create UDP transport socket (%d).\n", -err);
1538                 goto out;
1539         }
1540         xs_reclassify_socket6(sock);
1541
1542         if (xs_bind6(transport, sock) < 0) {
1543                 sock_release(sock);
1544                 goto out;
1545         }
1546
1547         dprintk("RPC:       worker connecting xprt %p to address: %s\n",
1548                         xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1549
1550         xs_udp_finish_connecting(xprt, sock);
1551         status = 0;
1552 out:
1553         xprt_wake_pending_tasks(xprt, status);
1554         xprt_clear_connecting(xprt);
1555 }
1556
1557 /*
1558  * We need to preserve the port number so the reply cache on the server can
1559  * find our cached RPC replies when we get around to reconnecting.
1560  */
1561 static void xs_tcp_reuse_connection(struct rpc_xprt *xprt)
1562 {
1563         int result;
1564         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1565         struct sockaddr any;
1566
1567         dprintk("RPC:       disconnecting xprt %p to reuse port\n", xprt);
1568
1569         /*
1570          * Disconnect the transport socket by doing a connect operation
1571          * with AF_UNSPEC.  This should return immediately...
1572          */
1573         memset(&any, 0, sizeof(any));
1574         any.sa_family = AF_UNSPEC;
1575         result = kernel_connect(transport->sock, &any, sizeof(any), 0);
1576         if (result)
1577                 dprintk("RPC:       AF_UNSPEC connect return code %d\n",
1578                                 result);
1579 }
1580
1581 static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
1582 {
1583         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1584
1585         if (!transport->inet) {
1586                 struct sock *sk = sock->sk;
1587
1588                 write_lock_bh(&sk->sk_callback_lock);
1589
1590                 sk->sk_user_data = xprt;
1591                 transport->old_data_ready = sk->sk_data_ready;
1592                 transport->old_state_change = sk->sk_state_change;
1593                 transport->old_write_space = sk->sk_write_space;
1594                 sk->sk_data_ready = xs_tcp_data_ready;
1595                 sk->sk_state_change = xs_tcp_state_change;
1596                 sk->sk_write_space = xs_tcp_write_space;
1597                 sk->sk_allocation = GFP_ATOMIC;
1598
1599                 /* socket options */
1600                 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
1601                 sock_reset_flag(sk, SOCK_LINGER);
1602                 tcp_sk(sk)->linger2 = 0;
1603                 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
1604
1605                 xprt_clear_connected(xprt);
1606
1607                 /* Reset to new socket */
1608                 transport->sock = sock;
1609                 transport->inet = sk;
1610
1611                 write_unlock_bh(&sk->sk_callback_lock);
1612         }
1613
1614         /* Tell the socket layer to start connecting... */
1615         xprt->stat.connect_count++;
1616         xprt->stat.connect_start = jiffies;
1617         return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
1618 }
1619
1620 /**
1621  * xs_tcp_connect_worker4 - connect a TCP socket to a remote endpoint
1622  * @work: RPC transport to connect
1623  *
1624  * Invoked by a work queue tasklet.
1625  */
1626 static void xs_tcp_connect_worker4(struct work_struct *work)
1627 {
1628         struct sock_xprt *transport =
1629                 container_of(work, struct sock_xprt, connect_worker.work);
1630         struct rpc_xprt *xprt = &transport->xprt;
1631         struct socket *sock = transport->sock;
1632         int err, status = -EIO;
1633
1634         if (xprt->shutdown || !xprt_bound(xprt))
1635                 goto out;
1636
1637         if (!sock) {
1638                 /* start from scratch */
1639                 if ((err = sock_create_kern(PF_INET, SOCK_STREAM, IPPROTO_TCP, &sock)) < 0) {
1640                         dprintk("RPC:       can't create TCP transport socket (%d).\n", -err);
1641                         goto out;
1642                 }
1643                 xs_reclassify_socket4(sock);
1644
1645                 if (xs_bind4(transport, sock) < 0) {
1646                         sock_release(sock);
1647                         goto out;
1648                 }
1649         } else
1650                 /* "close" the socket, preserving the local port */
1651                 xs_tcp_reuse_connection(xprt);
1652
1653         dprintk("RPC:       worker connecting xprt %p to address: %s\n",
1654                         xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1655
1656         status = xs_tcp_finish_connecting(xprt, sock);
1657         dprintk("RPC:       %p connect status %d connected %d sock state %d\n",
1658                         xprt, -status, xprt_connected(xprt),
1659                         sock->sk->sk_state);
1660         if (status < 0) {
1661                 switch (status) {
1662                         case -EINPROGRESS:
1663                         case -EALREADY:
1664                                 goto out_clear;
1665                         case -ECONNREFUSED:
1666                         case -ECONNRESET:
1667                                 /* retry with existing socket, after a delay */
1668                                 break;
1669                         default:
1670                                 /* get rid of existing socket, and retry */
1671                                 xs_tcp_shutdown(xprt);
1672                 }
1673         }
1674 out:
1675         xprt_wake_pending_tasks(xprt, status);
1676 out_clear:
1677         xprt_clear_connecting(xprt);
1678 }
1679
1680 /**
1681  * xs_tcp_connect_worker6 - connect a TCP socket to a remote endpoint
1682  * @work: RPC transport to connect
1683  *
1684  * Invoked by a work queue tasklet.
1685  */
1686 static void xs_tcp_connect_worker6(struct work_struct *work)
1687 {
1688         struct sock_xprt *transport =
1689                 container_of(work, struct sock_xprt, connect_worker.work);
1690         struct rpc_xprt *xprt = &transport->xprt;
1691         struct socket *sock = transport->sock;
1692         int err, status = -EIO;
1693
1694         if (xprt->shutdown || !xprt_bound(xprt))
1695                 goto out;
1696
1697         if (!sock) {
1698                 /* start from scratch */
1699                 if ((err = sock_create_kern(PF_INET6, SOCK_STREAM, IPPROTO_TCP, &sock)) < 0) {
1700                         dprintk("RPC:       can't create TCP transport socket (%d).\n", -err);
1701                         goto out;
1702                 }
1703                 xs_reclassify_socket6(sock);
1704
1705                 if (xs_bind6(transport, sock) < 0) {
1706                         sock_release(sock);
1707                         goto out;
1708                 }
1709         } else
1710                 /* "close" the socket, preserving the local port */
1711                 xs_tcp_reuse_connection(xprt);
1712
1713         dprintk("RPC:       worker connecting xprt %p to address: %s\n",
1714                         xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1715
1716         status = xs_tcp_finish_connecting(xprt, sock);
1717         dprintk("RPC:       %p connect status %d connected %d sock state %d\n",
1718                         xprt, -status, xprt_connected(xprt), sock->sk->sk_state);
1719         if (status < 0) {
1720                 switch (status) {
1721                         case -EINPROGRESS:
1722                         case -EALREADY:
1723                                 goto out_clear;
1724                         case -ECONNREFUSED:
1725                         case -ECONNRESET:
1726                                 /* retry with existing socket, after a delay */
1727                                 break;
1728                         default:
1729                                 /* get rid of existing socket, and retry */
1730                                 xs_tcp_shutdown(xprt);
1731                 }
1732         }
1733 out:
1734         xprt_wake_pending_tasks(xprt, status);
1735 out_clear:
1736         xprt_clear_connecting(xprt);
1737 }
1738
1739 /**
1740  * xs_connect - connect a socket to a remote endpoint
1741  * @task: address of RPC task that manages state of connect request
1742  *
1743  * TCP: If the remote end dropped the connection, delay reconnecting.
1744  *
1745  * UDP socket connects are synchronous, but we use a work queue anyway
1746  * to guarantee that even unprivileged user processes can set up a
1747  * socket on a privileged port.
1748  *
1749  * If a UDP socket connect fails, the delay behavior here prevents
1750  * retry floods (hard mounts).
1751  */
1752 static void xs_connect(struct rpc_task *task)
1753 {
1754         struct rpc_xprt *xprt = task->tk_xprt;
1755         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1756
1757         if (xprt_test_and_set_connecting(xprt))
1758                 return;
1759
1760         if (transport->sock != NULL) {
1761                 dprintk("RPC:       xs_connect delayed xprt %p for %lu "
1762                                 "seconds\n",
1763                                 xprt, xprt->reestablish_timeout / HZ);
1764                 queue_delayed_work(rpciod_workqueue,
1765                                    &transport->connect_worker,
1766                                    xprt->reestablish_timeout);
1767                 xprt->reestablish_timeout <<= 1;
1768                 if (xprt->reestablish_timeout > XS_TCP_MAX_REEST_TO)
1769                         xprt->reestablish_timeout = XS_TCP_MAX_REEST_TO;
1770         } else {
1771                 dprintk("RPC:       xs_connect scheduled xprt %p\n", xprt);
1772                 queue_delayed_work(rpciod_workqueue,
1773                                    &transport->connect_worker, 0);
1774         }
1775 }
1776
1777 static void xs_tcp_connect(struct rpc_task *task)
1778 {
1779         struct rpc_xprt *xprt = task->tk_xprt;
1780
1781         /* Initiate graceful shutdown of the socket if not already done */
1782         if (test_bit(XPRT_CONNECTED, &xprt->state))
1783                 xs_tcp_shutdown(xprt);
1784         /* Exit if we need to wait for socket shutdown to complete */
1785         if (test_bit(XPRT_CLOSING, &xprt->state))
1786                 return;
1787         xs_connect(task);
1788 }
1789
1790 /**
1791  * xs_udp_print_stats - display UDP socket-specifc stats
1792  * @xprt: rpc_xprt struct containing statistics
1793  * @seq: output file
1794  *
1795  */
1796 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
1797 {
1798         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1799
1800         seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %Lu %Lu\n",
1801                         transport->port,
1802                         xprt->stat.bind_count,
1803                         xprt->stat.sends,
1804                         xprt->stat.recvs,
1805                         xprt->stat.bad_xids,
1806                         xprt->stat.req_u,
1807                         xprt->stat.bklog_u);
1808 }
1809
1810 /**
1811  * xs_tcp_print_stats - display TCP socket-specifc stats
1812  * @xprt: rpc_xprt struct containing statistics
1813  * @seq: output file
1814  *
1815  */
1816 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
1817 {
1818         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1819         long idle_time = 0;
1820
1821         if (xprt_connected(xprt))
1822                 idle_time = (long)(jiffies - xprt->last_used) / HZ;
1823
1824         seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu %Lu %Lu\n",
1825                         transport->port,
1826                         xprt->stat.bind_count,
1827                         xprt->stat.connect_count,
1828                         xprt->stat.connect_time,
1829                         idle_time,
1830                         xprt->stat.sends,
1831                         xprt->stat.recvs,
1832                         xprt->stat.bad_xids,
1833                         xprt->stat.req_u,
1834                         xprt->stat.bklog_u);
1835 }
1836
1837 static struct rpc_xprt_ops xs_udp_ops = {
1838         .set_buffer_size        = xs_udp_set_buffer_size,
1839         .reserve_xprt           = xprt_reserve_xprt_cong,
1840         .release_xprt           = xprt_release_xprt_cong,
1841         .rpcbind                = rpcb_getport_async,
1842         .set_port               = xs_set_port,
1843         .connect                = xs_connect,
1844         .buf_alloc              = rpc_malloc,
1845         .buf_free               = rpc_free,
1846         .send_request           = xs_udp_send_request,
1847         .set_retrans_timeout    = xprt_set_retrans_timeout_rtt,
1848         .timer                  = xs_udp_timer,
1849         .release_request        = xprt_release_rqst_cong,
1850         .close                  = xs_close,
1851         .destroy                = xs_destroy,
1852         .print_stats            = xs_udp_print_stats,
1853 };
1854
1855 static struct rpc_xprt_ops xs_tcp_ops = {
1856         .reserve_xprt           = xprt_reserve_xprt,
1857         .release_xprt           = xs_tcp_release_xprt,
1858         .rpcbind                = rpcb_getport_async,
1859         .set_port               = xs_set_port,
1860         .connect                = xs_tcp_connect,
1861         .buf_alloc              = rpc_malloc,
1862         .buf_free               = rpc_free,
1863         .send_request           = xs_tcp_send_request,
1864         .set_retrans_timeout    = xprt_set_retrans_timeout_def,
1865         .close                  = xs_tcp_shutdown,
1866         .destroy                = xs_destroy,
1867         .print_stats            = xs_tcp_print_stats,
1868 };
1869
1870 static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
1871                                       unsigned int slot_table_size)
1872 {
1873         struct rpc_xprt *xprt;
1874         struct sock_xprt *new;
1875
1876         if (args->addrlen > sizeof(xprt->addr)) {
1877                 dprintk("RPC:       xs_setup_xprt: address too large\n");
1878                 return ERR_PTR(-EBADF);
1879         }
1880
1881         new = kzalloc(sizeof(*new), GFP_KERNEL);
1882         if (new == NULL) {
1883                 dprintk("RPC:       xs_setup_xprt: couldn't allocate "
1884                                 "rpc_xprt\n");
1885                 return ERR_PTR(-ENOMEM);
1886         }
1887         xprt = &new->xprt;
1888
1889         xprt->max_reqs = slot_table_size;
1890         xprt->slot = kcalloc(xprt->max_reqs, sizeof(struct rpc_rqst), GFP_KERNEL);
1891         if (xprt->slot == NULL) {
1892                 kfree(xprt);
1893                 dprintk("RPC:       xs_setup_xprt: couldn't allocate slot "
1894                                 "table\n");
1895                 return ERR_PTR(-ENOMEM);
1896         }
1897
1898         memcpy(&xprt->addr, args->dstaddr, args->addrlen);
1899         xprt->addrlen = args->addrlen;
1900         if (args->srcaddr)
1901                 memcpy(&new->addr, args->srcaddr, args->addrlen);
1902
1903         return xprt;
1904 }
1905
1906 static const struct rpc_timeout xs_udp_default_timeout = {
1907         .to_initval = 5 * HZ,
1908         .to_maxval = 30 * HZ,
1909         .to_increment = 5 * HZ,
1910         .to_retries = 5,
1911 };
1912
1913 /**
1914  * xs_setup_udp - Set up transport to use a UDP socket
1915  * @args: rpc transport creation arguments
1916  *
1917  */
1918 static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
1919 {
1920         struct sockaddr *addr = args->dstaddr;
1921         struct rpc_xprt *xprt;
1922         struct sock_xprt *transport;
1923
1924         xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries);
1925         if (IS_ERR(xprt))
1926                 return xprt;
1927         transport = container_of(xprt, struct sock_xprt, xprt);
1928
1929         xprt->prot = IPPROTO_UDP;
1930         xprt->tsh_size = 0;
1931         /* XXX: header size can vary due to auth type, IPv6, etc. */
1932         xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
1933
1934         xprt->bind_timeout = XS_BIND_TO;
1935         xprt->connect_timeout = XS_UDP_CONN_TO;
1936         xprt->reestablish_timeout = XS_UDP_REEST_TO;
1937         xprt->idle_timeout = XS_IDLE_DISC_TO;
1938
1939         xprt->ops = &xs_udp_ops;
1940
1941         xprt->timeout = &xs_udp_default_timeout;
1942
1943         switch (addr->sa_family) {
1944         case AF_INET:
1945                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
1946                         xprt_set_bound(xprt);
1947
1948                 INIT_DELAYED_WORK(&transport->connect_worker,
1949                                         xs_udp_connect_worker4);
1950                 xs_format_ipv4_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
1951                 break;
1952         case AF_INET6:
1953                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
1954                         xprt_set_bound(xprt);
1955
1956                 INIT_DELAYED_WORK(&transport->connect_worker,
1957                                         xs_udp_connect_worker6);
1958                 xs_format_ipv6_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
1959                 break;
1960         default:
1961                 kfree(xprt);
1962                 return ERR_PTR(-EAFNOSUPPORT);
1963         }
1964
1965         dprintk("RPC:       set up transport to address %s\n",
1966                         xprt->address_strings[RPC_DISPLAY_ALL]);
1967
1968         if (try_module_get(THIS_MODULE))
1969                 return xprt;
1970
1971         kfree(xprt->slot);
1972         kfree(xprt);
1973         return ERR_PTR(-EINVAL);
1974 }
1975
1976 static const struct rpc_timeout xs_tcp_default_timeout = {
1977         .to_initval = 60 * HZ,
1978         .to_maxval = 60 * HZ,
1979         .to_retries = 2,
1980 };
1981
1982 /**
1983  * xs_setup_tcp - Set up transport to use a TCP socket
1984  * @args: rpc transport creation arguments
1985  *
1986  */
1987 static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
1988 {
1989         struct sockaddr *addr = args->dstaddr;
1990         struct rpc_xprt *xprt;
1991         struct sock_xprt *transport;
1992
1993         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries);
1994         if (IS_ERR(xprt))
1995                 return xprt;
1996         transport = container_of(xprt, struct sock_xprt, xprt);
1997
1998         xprt->prot = IPPROTO_TCP;
1999         xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2000         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2001
2002         xprt->bind_timeout = XS_BIND_TO;
2003         xprt->connect_timeout = XS_TCP_CONN_TO;
2004         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2005         xprt->idle_timeout = XS_IDLE_DISC_TO;
2006
2007         xprt->ops = &xs_tcp_ops;
2008         xprt->timeout = &xs_tcp_default_timeout;
2009
2010         switch (addr->sa_family) {
2011         case AF_INET:
2012                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2013                         xprt_set_bound(xprt);
2014
2015                 INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_connect_worker4);
2016                 xs_format_ipv4_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
2017                 break;
2018         case AF_INET6:
2019                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2020                         xprt_set_bound(xprt);
2021
2022                 INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_connect_worker6);
2023                 xs_format_ipv6_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
2024                 break;
2025         default:
2026                 kfree(xprt);
2027                 return ERR_PTR(-EAFNOSUPPORT);
2028         }
2029
2030         dprintk("RPC:       set up transport to address %s\n",
2031                         xprt->address_strings[RPC_DISPLAY_ALL]);
2032
2033         if (try_module_get(THIS_MODULE))
2034                 return xprt;
2035
2036         kfree(xprt->slot);
2037         kfree(xprt);
2038         return ERR_PTR(-EINVAL);
2039 }
2040
2041 static struct xprt_class        xs_udp_transport = {
2042         .list           = LIST_HEAD_INIT(xs_udp_transport.list),
2043         .name           = "udp",
2044         .owner          = THIS_MODULE,
2045         .ident          = IPPROTO_UDP,
2046         .setup          = xs_setup_udp,
2047 };
2048
2049 static struct xprt_class        xs_tcp_transport = {
2050         .list           = LIST_HEAD_INIT(xs_tcp_transport.list),
2051         .name           = "tcp",
2052         .owner          = THIS_MODULE,
2053         .ident          = IPPROTO_TCP,
2054         .setup          = xs_setup_tcp,
2055 };
2056
2057 /**
2058  * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
2059  *
2060  */
2061 int init_socket_xprt(void)
2062 {
2063 #ifdef RPC_DEBUG
2064         if (!sunrpc_table_header)
2065                 sunrpc_table_header = register_sysctl_table(sunrpc_table);
2066 #endif
2067
2068         xprt_register_transport(&xs_udp_transport);
2069         xprt_register_transport(&xs_tcp_transport);
2070
2071         return 0;
2072 }
2073
2074 /**
2075  * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
2076  *
2077  */
2078 void cleanup_socket_xprt(void)
2079 {
2080 #ifdef RPC_DEBUG
2081         if (sunrpc_table_header) {
2082                 unregister_sysctl_table(sunrpc_table_header);
2083                 sunrpc_table_header = NULL;
2084         }
2085 #endif
2086
2087         xprt_unregister_transport(&xs_udp_transport);
2088         xprt_unregister_transport(&xs_tcp_transport);
2089 }