SAFE public projects git trees. - safe/jmp/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         void                    (*old_error_report)(struct sock *);
 253 };
 254
 255 /*
 256  * TCP receive state flags
 257  */
 258 #define TCP_RCV_LAST_FRAG       (1UL << 0)
 259 #define TCP_RCV_COPY_FRAGHDR    (1UL << 1)
 260 #define TCP_RCV_COPY_XID        (1UL << 2)
 261 #define TCP_RCV_COPY_DATA       (1UL << 3)
 262
 263 static inline struct sockaddr *xs_addr(struct rpc_xprt *xprt)
 264 {
 265         return (struct sockaddr *) &xprt->addr;
 266 }
 267
 268 static inline struct sockaddr_in *xs_addr_in(struct rpc_xprt *xprt)
 269 {
 270         return (struct sockaddr_in *) &xprt->addr;
 271 }
 272
 273 static inline struct sockaddr_in6 *xs_addr_in6(struct rpc_xprt *xprt)
 274 {
 275         return (struct sockaddr_in6 *) &xprt->addr;
 276 }
 277
 278 static void xs_format_ipv4_peer_addresses(struct rpc_xprt *xprt,
 279                                           const char *protocol,
 280                                           const char *netid)
 281 {
 282         struct sockaddr_in *addr = xs_addr_in(xprt);
 283         char *buf;
 284
 285         buf = kzalloc(20, GFP_KERNEL);
 286         if (buf) {
 287                 snprintf(buf, 20, NIPQUAD_FMT,
 288                                 NIPQUAD(addr->sin_addr.s_addr));
 289         }
 290         xprt->address_strings[RPC_DISPLAY_ADDR] = buf;
 291
 292         buf = kzalloc(8, GFP_KERNEL);
 293         if (buf) {
 294                 snprintf(buf, 8, "%u",
 295                                 ntohs(addr->sin_port));
 296         }
 297         xprt->address_strings[RPC_DISPLAY_PORT] = buf;
 298
 299         xprt->address_strings[RPC_DISPLAY_PROTO] = protocol;
 300
 301         buf = kzalloc(48, GFP_KERNEL);
 302         if (buf) {
 303                 snprintf(buf, 48, "addr="NIPQUAD_FMT" port=%u proto=%s",
 304                         NIPQUAD(addr->sin_addr.s_addr),
 305                         ntohs(addr->sin_port),
 306                         protocol);
 307         }
 308         xprt->address_strings[RPC_DISPLAY_ALL] = buf;
 309
 310         buf = kzalloc(10, GFP_KERNEL);
 311         if (buf) {
 312                 snprintf(buf, 10, "%02x%02x%02x%02x",
 313                                 NIPQUAD(addr->sin_addr.s_addr));
 314         }
 315         xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = buf;
 316
 317         buf = kzalloc(8, GFP_KERNEL);
 318         if (buf) {
 319                 snprintf(buf, 8, "%4hx",
 320                                 ntohs(addr->sin_port));
 321         }
 322         xprt->address_strings[RPC_DISPLAY_HEX_PORT] = buf;
 323
 324         buf = kzalloc(30, GFP_KERNEL);
 325         if (buf) {
 326                 snprintf(buf, 30, NIPQUAD_FMT".%u.%u",
 327                                 NIPQUAD(addr->sin_addr.s_addr),
 328                                 ntohs(addr->sin_port) >> 8,
 329                                 ntohs(addr->sin_port) & 0xff);
 330         }
 331         xprt->address_strings[RPC_DISPLAY_UNIVERSAL_ADDR] = buf;
 332
 333         xprt->address_strings[RPC_DISPLAY_NETID] = netid;
 334 }
 335
 336 static void xs_format_ipv6_peer_addresses(struct rpc_xprt *xprt,
 337                                           const char *protocol,
 338                                           const char *netid)
 339 {
 340         struct sockaddr_in6 *addr = xs_addr_in6(xprt);
 341         char *buf;
 342
 343         buf = kzalloc(40, GFP_KERNEL);
 344         if (buf) {
 345                 snprintf(buf, 40, "%pI6",&addr->sin6_addr);
 346         }
 347         xprt->address_strings[RPC_DISPLAY_ADDR] = buf;
 348
 349         buf = kzalloc(8, GFP_KERNEL);
 350         if (buf) {
 351                 snprintf(buf, 8, "%u",
 352                                 ntohs(addr->sin6_port));
 353         }
 354         xprt->address_strings[RPC_DISPLAY_PORT] = buf;
 355
 356         xprt->address_strings[RPC_DISPLAY_PROTO] = protocol;
 357
 358         buf = kzalloc(64, GFP_KERNEL);
 359         if (buf) {
 360                 snprintf(buf, 64, "addr=%pI6 port=%u proto=%s",
 361                                 &addr->sin6_addr,
 362                                 ntohs(addr->sin6_port),
 363                                 protocol);
 364         }
 365         xprt->address_strings[RPC_DISPLAY_ALL] = buf;
 366
 367         buf = kzalloc(36, GFP_KERNEL);
 368         if (buf)
 369                 snprintf(buf, 36, "%pi6", &addr->sin6_addr);
 370
 371         xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = buf;
 372
 373         buf = kzalloc(8, GFP_KERNEL);
 374         if (buf) {
 375                 snprintf(buf, 8, "%4hx",
 376                                 ntohs(addr->sin6_port));
 377         }
 378         xprt->address_strings[RPC_DISPLAY_HEX_PORT] = buf;
 379
 380         buf = kzalloc(50, GFP_KERNEL);
 381         if (buf) {
 382                 snprintf(buf, 50, "%pI6.%u.%u",
 383                          &addr->sin6_addr,
 384                          ntohs(addr->sin6_port) >> 8,
 385                          ntohs(addr->sin6_port) & 0xff);
 386         }
 387         xprt->address_strings[RPC_DISPLAY_UNIVERSAL_ADDR] = buf;
 388
 389         xprt->address_strings[RPC_DISPLAY_NETID] = netid;
 390 }
 391
 392 static void xs_free_peer_addresses(struct rpc_xprt *xprt)
 393 {
 394         unsigned int i;
 395
 396         for (i = 0; i < RPC_DISPLAY_MAX; i++)
 397                 switch (i) {
 398                 case RPC_DISPLAY_PROTO:
 399                 case RPC_DISPLAY_NETID:
 400                         continue;
 401                 default:
 402                         kfree(xprt->address_strings[i]);
 403                 }
 404 }
 405
 406 #define XS_SENDMSG_FLAGS        (MSG_DONTWAIT | MSG_NOSIGNAL)
 407
 408 static int xs_send_kvec(struct socket *sock, struct sockaddr *addr, int addrlen, struct kvec *vec, unsigned int base, int more)
 409 {
 410         struct msghdr msg = {
 411                 .msg_name       = addr,
 412                 .msg_namelen    = addrlen,
 413                 .msg_flags      = XS_SENDMSG_FLAGS | (more ? MSG_MORE : 0),
 414         };
 415         struct kvec iov = {
 416                 .iov_base       = vec->iov_base + base,
 417                 .iov_len        = vec->iov_len - base,
 418         };
 419
 420         if (iov.iov_len != 0)
 421                 return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
 422         return kernel_sendmsg(sock, &msg, NULL, 0, 0);
 423 }
 424
 425 static int xs_send_pagedata(struct socket *sock, struct xdr_buf *xdr, unsigned int base, int more)
 426 {
 427         struct page **ppage;
 428         unsigned int remainder;
 429         int err, sent = 0;
 430
 431         remainder = xdr->page_len - base;
 432         base += xdr->page_base;
 433         ppage = xdr->pages + (base >> PAGE_SHIFT);
 434         base &= ~PAGE_MASK;
 435         for(;;) {
 436                 unsigned int len = min_t(unsigned int, PAGE_SIZE - base, remainder);
 437                 int flags = XS_SENDMSG_FLAGS;
 438
 439                 remainder -= len;
 440                 if (remainder != 0 || more)
 441                         flags |= MSG_MORE;
 442                 err = sock->ops->sendpage(sock, *ppage, base, len, flags);
 443                 if (remainder == 0 || err != len)
 444                         break;
 445                 sent += err;
 446                 ppage++;
 447                 base = 0;
 448         }
 449         if (sent == 0)
 450                 return err;
 451         if (err > 0)
 452                 sent += err;
 453         return sent;
 454 }
 455
 456 /**
 457  * xs_sendpages - write pages directly to a socket
 458  * @sock: socket to send on
 459  * @addr: UDP only -- address of destination
 460  * @addrlen: UDP only -- length of destination address
 461  * @xdr: buffer containing this request
 462  * @base: starting position in the buffer
 463  *
 464  */
 465 static int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base)
 466 {
 467         unsigned int remainder = xdr->len - base;
 468         int err, sent = 0;
 469
 470         if (unlikely(!sock))
 471                 return -ENOTCONN;
 472
 473         clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags);
 474         if (base != 0) {
 475                 addr = NULL;
 476                 addrlen = 0;
 477         }
 478
 479         if (base < xdr->head[0].iov_len || addr != NULL) {
 480                 unsigned int len = xdr->head[0].iov_len - base;
 481                 remainder -= len;
 482                 err = xs_send_kvec(sock, addr, addrlen, &xdr->head[0], base, remainder != 0);
 483                 if (remainder == 0 || err != len)
 484                         goto out;
 485                 sent += err;
 486                 base = 0;
 487         } else
 488                 base -= xdr->head[0].iov_len;
 489
 490         if (base < xdr->page_len) {
 491                 unsigned int len = xdr->page_len - base;
 492                 remainder -= len;
 493                 err = xs_send_pagedata(sock, xdr, base, remainder != 0);
 494                 if (remainder == 0 || err != len)
 495                         goto out;
 496                 sent += err;
 497                 base = 0;
 498         } else
 499                 base -= xdr->page_len;
 500
 501         if (base >= xdr->tail[0].iov_len)
 502                 return sent;
 503         err = xs_send_kvec(sock, NULL, 0, &xdr->tail[0], base, 0);
 504 out:
 505         if (sent == 0)
 506                 return err;
 507         if (err > 0)
 508                 sent += err;
 509         return sent;
 510 }
 511
 512 static void xs_nospace_callback(struct rpc_task *task)
 513 {
 514         struct sock_xprt *transport = container_of(task->tk_rqstp->rq_xprt, struct sock_xprt, xprt);
 515
 516         transport->inet->sk_write_pending--;
 517         clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 518 }
 519
 520 /**
 521  * xs_nospace - place task on wait queue if transmit was incomplete
 522  * @task: task to put to sleep
 523  *
 524  */
 525 static void xs_nospace(struct rpc_task *task)
 526 {
 527         struct rpc_rqst *req = task->tk_rqstp;
 528         struct rpc_xprt *xprt = req->rq_xprt;
 529         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 530
 531         dprintk("RPC: %5u xmit incomplete (%u left of %u)\n",
 532                         task->tk_pid, req->rq_slen - req->rq_bytes_sent,
 533                         req->rq_slen);
 534
 535         /* Protect against races with write_space */
 536         spin_lock_bh(&xprt->transport_lock);
 537
 538         /* Don't race with disconnect */
 539         if (xprt_connected(xprt)) {
 540                 if (test_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags)) {
 541                         /*
 542                          * Notify TCP that we're limited by the application
 543                          * window size
 544                          */
 545                         set_bit(SOCK_NOSPACE, &transport->sock->flags);
 546                         transport->inet->sk_write_pending++;
 547                         /* ...and wait for more buffer space */
 548                         xprt_wait_for_buffer_space(task, xs_nospace_callback);
 549                 }
 550         } else {
 551                 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 552                 task->tk_status = -ENOTCONN;
 553         }
 554
 555         spin_unlock_bh(&xprt->transport_lock);
 556 }
 557
 558 /**
 559  * xs_udp_send_request - write an RPC request to a UDP socket
 560  * @task: address of RPC task that manages the state of an RPC request
 561  *
 562  * Return values:
 563  *        0:    The request has been sent
 564  *   EAGAIN:    The socket was blocked, please call again later to
 565  *              complete the request
 566  * ENOTCONN:    Caller needs to invoke connect logic then call again
 567  *    other:    Some other error occured, the request was not sent
 568  */
 569 static int xs_udp_send_request(struct rpc_task *task)
 570 {
 571         struct rpc_rqst *req = task->tk_rqstp;
 572         struct rpc_xprt *xprt = req->rq_xprt;
 573         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 574         struct xdr_buf *xdr = &req->rq_snd_buf;
 575         int status;
 576
 577         xs_pktdump("packet data:",
 578                                 req->rq_svec->iov_base,
 579                                 req->rq_svec->iov_len);
 580
 581         status = xs_sendpages(transport->sock,
 582                               xs_addr(xprt),
 583                               xprt->addrlen, xdr,
 584                               req->rq_bytes_sent);
 585
 586         dprintk("RPC:       xs_udp_send_request(%u) = %d\n",
 587                         xdr->len - req->rq_bytes_sent, status);
 588
 589         if (status >= 0) {
 590                 task->tk_bytes_sent += status;
 591                 if (status >= req->rq_slen)
 592                         return 0;
 593                 /* Still some bytes left; set up for a retry later. */
 594                 status = -EAGAIN;
 595         }
 596
 597         switch (status) {
 598         case -EAGAIN:
 599                 xs_nospace(task);
 600                 break;
 601         case -ENETUNREACH:
 602         case -EPIPE:
 603         case -ECONNREFUSED:
 604                 /* When the server has died, an ICMP port unreachable message
 605                  * prompts ECONNREFUSED. */
 606                 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 607                 break;
 608         default:
 609                 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 610                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
 611                         -status);
 612         }
 613
 614         return status;
 615 }
 616
 617 /**
 618  * xs_tcp_shutdown - gracefully shut down a TCP socket
 619  * @xprt: transport
 620  *
 621  * Initiates a graceful shutdown of the TCP socket by calling the
 622  * equivalent of shutdown(SHUT_WR);
 623  */
 624 static void xs_tcp_shutdown(struct rpc_xprt *xprt)
 625 {
 626         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 627         struct socket *sock = transport->sock;
 628
 629         if (sock != NULL)
 630                 kernel_sock_shutdown(sock, SHUT_WR);
 631 }
 632
 633 static inline void xs_encode_tcp_record_marker(struct xdr_buf *buf)
 634 {
 635         u32 reclen = buf->len - sizeof(rpc_fraghdr);
 636         rpc_fraghdr *base = buf->head[0].iov_base;
 637         *base = htonl(RPC_LAST_STREAM_FRAGMENT | reclen);
 638 }
 639
 640 /**
 641  * xs_tcp_send_request - write an RPC request to a TCP socket
 642  * @task: address of RPC task that manages the state of an RPC request
 643  *
 644  * Return values:
 645  *        0:    The request has been sent
 646  *   EAGAIN:    The socket was blocked, please call again later to
 647  *              complete the request
 648  * ENOTCONN:    Caller needs to invoke connect logic then call again
 649  *    other:    Some other error occured, the request was not sent
 650  *
 651  * XXX: In the case of soft timeouts, should we eventually give up
 652  *      if sendmsg is not able to make progress?
 653  */
 654 static int xs_tcp_send_request(struct rpc_task *task)
 655 {
 656         struct rpc_rqst *req = task->tk_rqstp;
 657         struct rpc_xprt *xprt = req->rq_xprt;
 658         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 659         struct xdr_buf *xdr = &req->rq_snd_buf;
 660         int status;
 661
 662         xs_encode_tcp_record_marker(&req->rq_snd_buf);
 663
 664         xs_pktdump("packet data:",
 665                                 req->rq_svec->iov_base,
 666                                 req->rq_svec->iov_len);
 667
 668         /* Continue transmitting the packet/record. We must be careful
 669          * to cope with writespace callbacks arriving _after_ we have
 670          * called sendmsg(). */
 671         while (1) {
 672                 status = xs_sendpages(transport->sock,
 673                                         NULL, 0, xdr, req->rq_bytes_sent);
 674
 675                 dprintk("RPC:       xs_tcp_send_request(%u) = %d\n",
 676                                 xdr->len - req->rq_bytes_sent, status);
 677
 678                 if (unlikely(status < 0))
 679                         break;
 680
 681                 /* If we've sent the entire packet, immediately
 682                  * reset the count of bytes sent. */
 683                 req->rq_bytes_sent += status;
 684                 task->tk_bytes_sent += status;
 685                 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
 686                         req->rq_bytes_sent = 0;
 687                         return 0;
 688                 }
 689
 690                 if (status != 0)
 691                         continue;
 692                 status = -EAGAIN;
 693                 break;
 694         }
 695
 696         switch (status) {
 697         case -EAGAIN:
 698                 xs_nospace(task);
 699                 break;
 700         case -ECONNRESET:
 701                 xs_tcp_shutdown(xprt);
 702         case -ECONNREFUSED:
 703         case -ENOTCONN:
 704         case -EPIPE:
 705                 status = -ENOTCONN;
 706                 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 707                 break;
 708         default:
 709                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
 710                         -status);
 711                 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 712                 xs_tcp_shutdown(xprt);
 713         }
 714
 715         return status;
 716 }
 717
 718 /**
 719  * xs_tcp_release_xprt - clean up after a tcp transmission
 720  * @xprt: transport
 721  * @task: rpc task
 722  *
 723  * This cleans up if an error causes us to abort the transmission of a request.
 724  * In this case, the socket may need to be reset in order to avoid confusing
 725  * the server.
 726  */
 727 static void xs_tcp_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
 728 {
 729         struct rpc_rqst *req;
 730
 731         if (task != xprt->snd_task)
 732                 return;
 733         if (task == NULL)
 734                 goto out_release;
 735         req = task->tk_rqstp;
 736         if (req->rq_bytes_sent == 0)
 737                 goto out_release;
 738         if (req->rq_bytes_sent == req->rq_snd_buf.len)
 739                 goto out_release;
 740         set_bit(XPRT_CLOSE_WAIT, &task->tk_xprt->state);
 741 out_release:
 742         xprt_release_xprt(xprt, task);
 743 }
 744
 745 static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk)
 746 {
 747         transport->old_data_ready = sk->sk_data_ready;
 748         transport->old_state_change = sk->sk_state_change;
 749         transport->old_write_space = sk->sk_write_space;
 750         transport->old_error_report = sk->sk_error_report;
 751 }
 752
 753 static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk)
 754 {
 755         sk->sk_data_ready = transport->old_data_ready;
 756         sk->sk_state_change = transport->old_state_change;
 757         sk->sk_write_space = transport->old_write_space;
 758         sk->sk_error_report = transport->old_error_report;
 759 }
 760
 761 /**
 762  * xs_close - close a socket
 763  * @xprt: transport
 764  *
 765  * This is used when all requests are complete; ie, no DRC state remains
 766  * on the server we want to save.
 767  */
 768 static void xs_close(struct rpc_xprt *xprt)
 769 {
 770         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 771         struct socket *sock = transport->sock;
 772         struct sock *sk = transport->inet;
 773
 774         if (!sk)
 775                 goto clear_close_wait;
 776
 777         dprintk("RPC:       xs_close xprt %p\n", xprt);
 778
 779         write_lock_bh(&sk->sk_callback_lock);
 780         transport->inet = NULL;
 781         transport->sock = NULL;
 782
 783         sk->sk_user_data = NULL;
 784
 785         xs_restore_old_callbacks(transport, sk);
 786         write_unlock_bh(&sk->sk_callback_lock);
 787
 788         sk->sk_no_check = 0;
 789
 790         sock_release(sock);
 791 clear_close_wait:
 792         smp_mb__before_clear_bit();
 793         clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
 794         clear_bit(XPRT_CLOSING, &xprt->state);
 795         smp_mb__after_clear_bit();
 796         xprt_disconnect_done(xprt);
 797 }
 798
 799 /**
 800  * xs_destroy - prepare to shutdown a transport
 801  * @xprt: doomed transport
 802  *
 803  */
 804 static void xs_destroy(struct rpc_xprt *xprt)
 805 {
 806         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 807
 808         dprintk("RPC:       xs_destroy xprt %p\n", xprt);
 809
 810         cancel_rearming_delayed_work(&transport->connect_worker);
 811
 812         xs_close(xprt);
 813         xs_free_peer_addresses(xprt);
 814         kfree(xprt->slot);
 815         kfree(xprt);
 816         module_put(THIS_MODULE);
 817 }
 818
 819 static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
 820 {
 821         return (struct rpc_xprt *) sk->sk_user_data;
 822 }
 823
 824 /**
 825  * xs_udp_data_ready - "data ready" callback for UDP sockets
 826  * @sk: socket with data to read
 827  * @len: how much data to read
 828  *
 829  */
 830 static void xs_udp_data_ready(struct sock *sk, int len)
 831 {
 832         struct rpc_task *task;
 833         struct rpc_xprt *xprt;
 834         struct rpc_rqst *rovr;
 835         struct sk_buff *skb;
 836         int err, repsize, copied;
 837         u32 _xid;
 838         __be32 *xp;
 839
 840         read_lock(&sk->sk_callback_lock);
 841         dprintk("RPC:       xs_udp_data_ready...\n");
 842         if (!(xprt = xprt_from_sock(sk)))
 843                 goto out;
 844
 845         if ((skb = skb_recv_datagram(sk, 0, 1, &err)) == NULL)
 846                 goto out;
 847
 848         if (xprt->shutdown)
 849                 goto dropit;
 850
 851         repsize = skb->len - sizeof(struct udphdr);
 852         if (repsize < 4) {
 853                 dprintk("RPC:       impossible RPC reply size %d!\n", repsize);
 854                 goto dropit;
 855         }
 856
 857         /* Copy the XID from the skb... */
 858         xp = skb_header_pointer(skb, sizeof(struct udphdr),
 859                                 sizeof(_xid), &_xid);
 860         if (xp == NULL)
 861                 goto dropit;
 862
 863         /* Look up and lock the request corresponding to the given XID */
 864         spin_lock(&xprt->transport_lock);
 865         rovr = xprt_lookup_rqst(xprt, *xp);
 866         if (!rovr)
 867                 goto out_unlock;
 868         task = rovr->rq_task;
 869
 870         if ((copied = rovr->rq_private_buf.buflen) > repsize)
 871                 copied = repsize;
 872
 873         /* Suck it into the iovec, verify checksum if not done by hw. */
 874         if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
 875                 UDPX_INC_STATS_BH(sk, UDP_MIB_INERRORS);
 876                 goto out_unlock;
 877         }
 878
 879         UDPX_INC_STATS_BH(sk, UDP_MIB_INDATAGRAMS);
 880
 881         /* Something worked... */
 882         dst_confirm(skb->dst);
 883
 884         xprt_adjust_cwnd(task, copied);
 885         xprt_update_rtt(task);
 886         xprt_complete_rqst(task, copied);
 887
 888  out_unlock:
 889         spin_unlock(&xprt->transport_lock);
 890  dropit:
 891         skb_free_datagram(sk, skb);
 892  out:
 893         read_unlock(&sk->sk_callback_lock);
 894 }
 895
 896 static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
 897 {
 898         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 899         size_t len, used;
 900         char *p;
 901
 902         p = ((char *) &transport->tcp_fraghdr) + transport->tcp_offset;
 903         len = sizeof(transport->tcp_fraghdr) - transport->tcp_offset;
 904         used = xdr_skb_read_bits(desc, p, len);
 905         transport->tcp_offset += used;
 906         if (used != len)
 907                 return;
 908
 909         transport->tcp_reclen = ntohl(transport->tcp_fraghdr);
 910         if (transport->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
 911                 transport->tcp_flags |= TCP_RCV_LAST_FRAG;
 912         else
 913                 transport->tcp_flags &= ~TCP_RCV_LAST_FRAG;
 914         transport->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
 915
 916         transport->tcp_flags &= ~TCP_RCV_COPY_FRAGHDR;
 917         transport->tcp_offset = 0;
 918
 919         /* Sanity check of the record length */
 920         if (unlikely(transport->tcp_reclen < 4)) {
 921                 dprintk("RPC:       invalid TCP record fragment length\n");
 922                 xprt_force_disconnect(xprt);
 923                 return;
 924         }
 925         dprintk("RPC:       reading TCP record fragment of length %d\n",
 926                         transport->tcp_reclen);
 927 }
 928
 929 static void xs_tcp_check_fraghdr(struct sock_xprt *transport)
 930 {
 931         if (transport->tcp_offset == transport->tcp_reclen) {
 932                 transport->tcp_flags |= TCP_RCV_COPY_FRAGHDR;
 933                 transport->tcp_offset = 0;
 934                 if (transport->tcp_flags & TCP_RCV_LAST_FRAG) {
 935                         transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
 936                         transport->tcp_flags |= TCP_RCV_COPY_XID;
 937                         transport->tcp_copied = 0;
 938                 }
 939         }
 940 }
 941
 942 static inline void xs_tcp_read_xid(struct sock_xprt *transport, struct xdr_skb_reader *desc)
 943 {
 944         size_t len, used;
 945         char *p;
 946
 947         len = sizeof(transport->tcp_xid) - transport->tcp_offset;
 948         dprintk("RPC:       reading XID (%Zu bytes)\n", len);
 949         p = ((char *) &transport->tcp_xid) + transport->tcp_offset;
 950         used = xdr_skb_read_bits(desc, p, len);
 951         transport->tcp_offset += used;
 952         if (used != len)
 953                 return;
 954         transport->tcp_flags &= ~TCP_RCV_COPY_XID;
 955         transport->tcp_flags |= TCP_RCV_COPY_DATA;
 956         transport->tcp_copied = 4;
 957         dprintk("RPC:       reading reply for XID %08x\n",
 958                         ntohl(transport->tcp_xid));
 959         xs_tcp_check_fraghdr(transport);
 960 }
 961
 962 static inline void xs_tcp_read_request(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
 963 {
 964         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 965         struct rpc_rqst *req;
 966         struct xdr_buf *rcvbuf;
 967         size_t len;
 968         ssize_t r;
 969
 970         /* Find and lock the request corresponding to this xid */
 971         spin_lock(&xprt->transport_lock);
 972         req = xprt_lookup_rqst(xprt, transport->tcp_xid);
 973         if (!req) {
 974                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
 975                 dprintk("RPC:       XID %08x request not found!\n",
 976                                 ntohl(transport->tcp_xid));
 977                 spin_unlock(&xprt->transport_lock);
 978                 return;
 979         }
 980
 981         rcvbuf = &req->rq_private_buf;
 982         len = desc->count;
 983         if (len > transport->tcp_reclen - transport->tcp_offset) {
 984                 struct xdr_skb_reader my_desc;
 985
 986                 len = transport->tcp_reclen - transport->tcp_offset;
 987                 memcpy(&my_desc, desc, sizeof(my_desc));
 988                 my_desc.count = len;
 989                 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
 990                                           &my_desc, xdr_skb_read_bits);
 991                 desc->count -= r;
 992                 desc->offset += r;
 993         } else
 994                 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
 995                                           desc, xdr_skb_read_bits);
 996
 997         if (r > 0) {
 998                 transport->tcp_copied += r;
 999                 transport->tcp_offset += r;
1000         }
1001         if (r != len) {
1002                 /* Error when copying to the receive buffer,
1003                  * usually because we weren't able to allocate
1004                  * additional buffer pages. All we can do now
1005                  * is turn off TCP_RCV_COPY_DATA, so the request
1006                  * will not receive any additional updates,
1007                  * and time out.
1008                  * Any remaining data from this record will
1009                  * be discarded.
1010                  */
1011                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1012                 dprintk("RPC:       XID %08x truncated request\n",
1013                                 ntohl(transport->tcp_xid));
1014                 dprintk("RPC:       xprt = %p, tcp_copied = %lu, "
1015                                 "tcp_offset = %u, tcp_reclen = %u\n",
1016                                 xprt, transport->tcp_copied,
1017                                 transport->tcp_offset, transport->tcp_reclen);
1018                 goto out;
1019         }
1020
1021         dprintk("RPC:       XID %08x read %Zd bytes\n",
1022                         ntohl(transport->tcp_xid), r);
1023         dprintk("RPC:       xprt = %p, tcp_copied = %lu, tcp_offset = %u, "
1024                         "tcp_reclen = %u\n", xprt, transport->tcp_copied,
1025                         transport->tcp_offset, transport->tcp_reclen);
1026
1027         if (transport->tcp_copied == req->rq_private_buf.buflen)
1028                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1029         else if (transport->tcp_offset == transport->tcp_reclen) {
1030                 if (transport->tcp_flags & TCP_RCV_LAST_FRAG)
1031                         transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1032         }
1033
1034 out:
1035         if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1036                 xprt_complete_rqst(req->rq_task, transport->tcp_copied);
1037         spin_unlock(&xprt->transport_lock);
1038         xs_tcp_check_fraghdr(transport);
1039 }
1040
1041 static inline void xs_tcp_read_discard(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1042 {
1043         size_t len;
1044
1045         len = transport->tcp_reclen - transport->tcp_offset;
1046         if (len > desc->count)
1047                 len = desc->count;
1048         desc->count -= len;
1049         desc->offset += len;
1050         transport->tcp_offset += len;
1051         dprintk("RPC:       discarded %Zu bytes\n", len);
1052         xs_tcp_check_fraghdr(transport);
1053 }
1054
1055 static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
1056 {
1057         struct rpc_xprt *xprt = rd_desc->arg.data;
1058         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1059         struct xdr_skb_reader desc = {
1060                 .skb    = skb,
1061                 .offset = offset,
1062                 .count  = len,
1063         };
1064
1065         dprintk("RPC:       xs_tcp_data_recv started\n");
1066         do {
1067                 /* Read in a new fragment marker if necessary */
1068                 /* Can we ever really expect to get completely empty fragments? */
1069                 if (transport->tcp_flags & TCP_RCV_COPY_FRAGHDR) {
1070                         xs_tcp_read_fraghdr(xprt, &desc);
1071                         continue;
1072                 }
1073                 /* Read in the xid if necessary */
1074                 if (transport->tcp_flags & TCP_RCV_COPY_XID) {
1075                         xs_tcp_read_xid(transport, &desc);
1076                         continue;
1077                 }
1078                 /* Read in the request data */
1079                 if (transport->tcp_flags & TCP_RCV_COPY_DATA) {
1080                         xs_tcp_read_request(xprt, &desc);
1081                         continue;
1082                 }
1083                 /* Skip over any trailing bytes on short reads */
1084                 xs_tcp_read_discard(transport, &desc);
1085         } while (desc.count);
1086         dprintk("RPC:       xs_tcp_data_recv done\n");
1087         return len - desc.count;
1088 }
1089
1090 /**
1091  * xs_tcp_data_ready - "data ready" callback for TCP sockets
1092  * @sk: socket with data to read
1093  * @bytes: how much data to read
1094  *
1095  */
1096 static void xs_tcp_data_ready(struct sock *sk, int bytes)
1097 {
1098         struct rpc_xprt *xprt;
1099         read_descriptor_t rd_desc;
1100         int read;
1101
1102         dprintk("RPC:       xs_tcp_data_ready...\n");
1103
1104         read_lock(&sk->sk_callback_lock);
1105         if (!(xprt = xprt_from_sock(sk)))
1106                 goto out;
1107         if (xprt->shutdown)
1108                 goto out;
1109
1110         /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
1111         rd_desc.arg.data = xprt;
1112         do {
1113                 rd_desc.count = 65536;
1114                 read = tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
1115         } while (read > 0);
1116 out:
1117         read_unlock(&sk->sk_callback_lock);
1118 }
1119
1120 /**
1121  * xs_tcp_state_change - callback to handle TCP socket state changes
1122  * @sk: socket whose state has changed
1123  *
1124  */
1125 static void xs_tcp_state_change(struct sock *sk)
1126 {
1127         struct rpc_xprt *xprt;
1128
1129         read_lock(&sk->sk_callback_lock);
1130         if (!(xprt = xprt_from_sock(sk)))
1131                 goto out;
1132         dprintk("RPC:       xs_tcp_state_change client %p...\n", xprt);
1133         dprintk("RPC:       state %x conn %d dead %d zapped %d\n",
1134                         sk->sk_state, xprt_connected(xprt),
1135                         sock_flag(sk, SOCK_DEAD),
1136                         sock_flag(sk, SOCK_ZAPPED));
1137
1138         switch (sk->sk_state) {
1139         case TCP_ESTABLISHED:
1140                 spin_lock_bh(&xprt->transport_lock);
1141                 if (!xprt_test_and_set_connected(xprt)) {
1142                         struct sock_xprt *transport = container_of(xprt,
1143                                         struct sock_xprt, xprt);
1144
1145                         /* Reset TCP record info */
1146                         transport->tcp_offset = 0;
1147                         transport->tcp_reclen = 0;
1148                         transport->tcp_copied = 0;
1149                         transport->tcp_flags =
1150                                 TCP_RCV_COPY_FRAGHDR | TCP_RCV_COPY_XID;
1151
1152                         xprt_wake_pending_tasks(xprt, 0);
1153                 }
1154                 spin_unlock_bh(&xprt->transport_lock);
1155                 break;
1156         case TCP_FIN_WAIT1:
1157                 /* The client initiated a shutdown of the socket */
1158                 xprt->connect_cookie++;
1159                 xprt->reestablish_timeout = 0;
1160                 set_bit(XPRT_CLOSING, &xprt->state);
1161                 smp_mb__before_clear_bit();
1162                 clear_bit(XPRT_CONNECTED, &xprt->state);
1163                 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1164                 smp_mb__after_clear_bit();
1165                 break;
1166         case TCP_CLOSE_WAIT:
1167                 /* The server initiated a shutdown of the socket */
1168                 set_bit(XPRT_CLOSING, &xprt->state);
1169                 xprt_force_disconnect(xprt);
1170         case TCP_SYN_SENT:
1171                 xprt->connect_cookie++;
1172         case TCP_CLOSING:
1173                 /*
1174                  * If the server closed down the connection, make sure that
1175                  * we back off before reconnecting
1176                  */
1177                 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1178                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1179                 break;
1180         case TCP_LAST_ACK:
1181                 smp_mb__before_clear_bit();
1182                 clear_bit(XPRT_CONNECTED, &xprt->state);
1183                 smp_mb__after_clear_bit();
1184                 break;
1185         case TCP_CLOSE:
1186                 smp_mb__before_clear_bit();
1187                 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1188                 clear_bit(XPRT_CLOSING, &xprt->state);
1189                 smp_mb__after_clear_bit();
1190                 /* Mark transport as closed and wake up all pending tasks */
1191                 xprt_disconnect_done(xprt);
1192         }
1193  out:
1194         read_unlock(&sk->sk_callback_lock);
1195 }
1196
1197 /**
1198  * xs_tcp_error_report - callback mainly for catching RST events
1199  * @sk: socket
1200  */
1201 static void xs_tcp_error_report(struct sock *sk)
1202 {
1203         struct rpc_xprt *xprt;
1204
1205         read_lock(&sk->sk_callback_lock);
1206         if (sk->sk_err != ECONNRESET || sk->sk_state != TCP_ESTABLISHED)
1207                 goto out;
1208         if (!(xprt = xprt_from_sock(sk)))
1209                 goto out;
1210         dprintk("RPC:       %s client %p...\n"
1211                         "RPC:       error %d\n",
1212                         __func__, xprt, sk->sk_err);
1213
1214         xprt_force_disconnect(xprt);
1215 out:
1216         read_unlock(&sk->sk_callback_lock);
1217 }
1218
1219 /**
1220  * xs_udp_write_space - callback invoked when socket buffer space
1221  *                             becomes available
1222  * @sk: socket whose state has changed
1223  *
1224  * Called when more output buffer space is available for this socket.
1225  * We try not to wake our writers until they can make "significant"
1226  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1227  * with a bunch of small requests.
1228  */
1229 static void xs_udp_write_space(struct sock *sk)
1230 {
1231         read_lock(&sk->sk_callback_lock);
1232
1233         /* from net/core/sock.c:sock_def_write_space */
1234         if (sock_writeable(sk)) {
1235                 struct socket *sock;
1236                 struct rpc_xprt *xprt;
1237
1238                 if (unlikely(!(sock = sk->sk_socket)))
1239                         goto out;
1240                 clear_bit(SOCK_NOSPACE, &sock->flags);
1241
1242                 if (unlikely(!(xprt = xprt_from_sock(sk))))
1243                         goto out;
1244                 if (test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags) == 0)
1245                         goto out;
1246
1247                 xprt_write_space(xprt);
1248         }
1249
1250  out:
1251         read_unlock(&sk->sk_callback_lock);
1252 }
1253
1254 /**
1255  * xs_tcp_write_space - callback invoked when socket buffer space
1256  *                             becomes available
1257  * @sk: socket whose state has changed
1258  *
1259  * Called when more output buffer space is available for this socket.
1260  * We try not to wake our writers until they can make "significant"
1261  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1262  * with a bunch of small requests.
1263  */
1264 static void xs_tcp_write_space(struct sock *sk)
1265 {
1266         read_lock(&sk->sk_callback_lock);
1267
1268         /* from net/core/stream.c:sk_stream_write_space */
1269         if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk)) {
1270                 struct socket *sock;
1271                 struct rpc_xprt *xprt;
1272
1273                 if (unlikely(!(sock = sk->sk_socket)))
1274                         goto out;
1275                 clear_bit(SOCK_NOSPACE, &sock->flags);
1276
1277                 if (unlikely(!(xprt = xprt_from_sock(sk))))
1278                         goto out;
1279                 if (test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags) == 0)
1280                         goto out;
1281
1282                 xprt_write_space(xprt);
1283         }
1284
1285  out:
1286         read_unlock(&sk->sk_callback_lock);
1287 }
1288
1289 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
1290 {
1291         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1292         struct sock *sk = transport->inet;
1293
1294         if (transport->rcvsize) {
1295                 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1296                 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
1297         }
1298         if (transport->sndsize) {
1299                 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1300                 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1301                 sk->sk_write_space(sk);
1302         }
1303 }
1304
1305 /**
1306  * xs_udp_set_buffer_size - set send and receive limits
1307  * @xprt: generic transport
1308  * @sndsize: requested size of send buffer, in bytes
1309  * @rcvsize: requested size of receive buffer, in bytes
1310  *
1311  * Set socket send and receive buffer size limits.
1312  */
1313 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1314 {
1315         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1316
1317         transport->sndsize = 0;
1318         if (sndsize)
1319                 transport->sndsize = sndsize + 1024;
1320         transport->rcvsize = 0;
1321         if (rcvsize)
1322                 transport->rcvsize = rcvsize + 1024;
1323
1324         xs_udp_do_set_buffer_size(xprt);
1325 }
1326
1327 /**
1328  * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1329  * @task: task that timed out
1330  *
1331  * Adjust the congestion window after a retransmit timeout has occurred.
1332  */
1333 static void xs_udp_timer(struct rpc_task *task)
1334 {
1335         xprt_adjust_cwnd(task, -ETIMEDOUT);
1336 }
1337
1338 static unsigned short xs_get_random_port(void)
1339 {
1340         unsigned short range = xprt_max_resvport - xprt_min_resvport;
1341         unsigned short rand = (unsigned short) net_random() % range;
1342         return rand + xprt_min_resvport;
1343 }
1344
1345 /**
1346  * xs_set_port - reset the port number in the remote endpoint address
1347  * @xprt: generic transport
1348  * @port: new port number
1349  *
1350  */
1351 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1352 {
1353         struct sockaddr *addr = xs_addr(xprt);
1354
1355         dprintk("RPC:       setting port for xprt %p to %u\n", xprt, port);
1356
1357         switch (addr->sa_family) {
1358         case AF_INET:
1359                 ((struct sockaddr_in *)addr)->sin_port = htons(port);
1360                 break;
1361         case AF_INET6:
1362                 ((struct sockaddr_in6 *)addr)->sin6_port = htons(port);
1363                 break;
1364         default:
1365                 BUG();
1366         }
1367 }
1368
1369 static unsigned short xs_get_srcport(struct sock_xprt *transport, struct socket *sock)
1370 {
1371         unsigned short port = transport->port;
1372
1373         if (port == 0 && transport->xprt.resvport)
1374                 port = xs_get_random_port();
1375         return port;
1376 }
1377
1378 static unsigned short xs_next_srcport(struct sock_xprt *transport, struct socket *sock, unsigned short port)
1379 {
1380         if (transport->port != 0)
1381                 transport->port = 0;
1382         if (!transport->xprt.resvport)
1383                 return 0;
1384         if (port <= xprt_min_resvport || port > xprt_max_resvport)
1385                 return xprt_max_resvport;
1386         return --port;
1387 }
1388
1389 static int xs_bind4(struct sock_xprt *transport, struct socket *sock)
1390 {
1391         struct sockaddr_in myaddr = {
1392                 .sin_family = AF_INET,
1393         };
1394         struct sockaddr_in *sa;
1395         int err, nloop = 0;
1396         unsigned short port = xs_get_srcport(transport, sock);
1397         unsigned short last;
1398
1399         sa = (struct sockaddr_in *)&transport->addr;
1400         myaddr.sin_addr = sa->sin_addr;
1401         do {
1402                 myaddr.sin_port = htons(port);
1403                 err = kernel_bind(sock, (struct sockaddr *) &myaddr,
1404                                                 sizeof(myaddr));
1405                 if (port == 0)
1406                         break;
1407                 if (err == 0) {
1408                         transport->port = port;
1409                         break;
1410                 }
1411                 last = port;
1412                 port = xs_next_srcport(transport, sock, port);
1413                 if (port > last)
1414                         nloop++;
1415         } while (err == -EADDRINUSE && nloop != 2);
1416         dprintk("RPC:       %s "NIPQUAD_FMT":%u: %s (%d)\n",
1417                         __func__, NIPQUAD(myaddr.sin_addr),
1418                         port, err ? "failed" : "ok", err);
1419         return err;
1420 }
1421
1422 static int xs_bind6(struct sock_xprt *transport, struct socket *sock)
1423 {
1424         struct sockaddr_in6 myaddr = {
1425                 .sin6_family = AF_INET6,
1426         };
1427         struct sockaddr_in6 *sa;
1428         int err, nloop = 0;
1429         unsigned short port = xs_get_srcport(transport, sock);
1430         unsigned short last;
1431
1432         sa = (struct sockaddr_in6 *)&transport->addr;
1433         myaddr.sin6_addr = sa->sin6_addr;
1434         do {
1435                 myaddr.sin6_port = htons(port);
1436                 err = kernel_bind(sock, (struct sockaddr *) &myaddr,
1437                                                 sizeof(myaddr));
1438                 if (port == 0)
1439                         break;
1440                 if (err == 0) {
1441                         transport->port = port;
1442                         break;
1443                 }
1444                 last = port;
1445                 port = xs_next_srcport(transport, sock, port);
1446                 if (port > last)
1447                         nloop++;
1448         } while (err == -EADDRINUSE && nloop != 2);
1449         dprintk("RPC:       xs_bind6 %pI6:%u: %s (%d)\n",
1450                 &myaddr.sin6_addr, port, err ? "failed" : "ok", err);
1451         return err;
1452 }
1453
1454 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1455 static struct lock_class_key xs_key[2];
1456 static struct lock_class_key xs_slock_key[2];
1457
1458 static inline void xs_reclassify_socket4(struct socket *sock)
1459 {
1460         struct sock *sk = sock->sk;
1461
1462         BUG_ON(sock_owned_by_user(sk));
1463         sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1464                 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1465 }
1466
1467 static inline void xs_reclassify_socket6(struct socket *sock)
1468 {
1469         struct sock *sk = sock->sk;
1470
1471         BUG_ON(sock_owned_by_user(sk));
1472         sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1473                 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
1474 }
1475 #else
1476 static inline void xs_reclassify_socket4(struct socket *sock)
1477 {
1478 }
1479
1480 static inline void xs_reclassify_socket6(struct socket *sock)
1481 {
1482 }
1483 #endif
1484
1485 static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
1486 {
1487         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1488
1489         if (!transport->inet) {
1490                 struct sock *sk = sock->sk;
1491
1492                 write_lock_bh(&sk->sk_callback_lock);
1493
1494                 xs_save_old_callbacks(transport, sk);
1495
1496                 sk->sk_user_data = xprt;
1497                 sk->sk_data_ready = xs_udp_data_ready;
1498                 sk->sk_write_space = xs_udp_write_space;
1499                 sk->sk_no_check = UDP_CSUM_NORCV;
1500                 sk->sk_allocation = GFP_ATOMIC;
1501
1502                 xprt_set_connected(xprt);
1503
1504                 /* Reset to new socket */
1505                 transport->sock = sock;
1506                 transport->inet = sk;
1507
1508                 write_unlock_bh(&sk->sk_callback_lock);
1509         }
1510         xs_udp_do_set_buffer_size(xprt);
1511 }
1512
1513 /**
1514  * xs_udp_connect_worker4 - set up a UDP socket
1515  * @work: RPC transport to connect
1516  *
1517  * Invoked by a work queue tasklet.
1518  */
1519 static void xs_udp_connect_worker4(struct work_struct *work)
1520 {
1521         struct sock_xprt *transport =
1522                 container_of(work, struct sock_xprt, connect_worker.work);
1523         struct rpc_xprt *xprt = &transport->xprt;
1524         struct socket *sock = transport->sock;
1525         int err, status = -EIO;
1526
1527         if (xprt->shutdown || !xprt_bound(xprt))
1528                 goto out;
1529
1530         /* Start by resetting any existing state */
1531         xs_close(xprt);
1532
1533         if ((err = sock_create_kern(PF_INET, SOCK_DGRAM, IPPROTO_UDP, &sock)) < 0) {
1534                 dprintk("RPC:       can't create UDP transport socket (%d).\n", -err);
1535                 goto out;
1536         }
1537         xs_reclassify_socket4(sock);
1538
1539         if (xs_bind4(transport, sock)) {
1540                 sock_release(sock);
1541                 goto out;
1542         }
1543
1544         dprintk("RPC:       worker connecting xprt %p to address: %s\n",
1545                         xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1546
1547         xs_udp_finish_connecting(xprt, sock);
1548         status = 0;
1549 out:
1550         xprt_wake_pending_tasks(xprt, status);
1551         xprt_clear_connecting(xprt);
1552 }
1553
1554 /**
1555  * xs_udp_connect_worker6 - set up a UDP socket
1556  * @work: RPC transport to connect
1557  *
1558  * Invoked by a work queue tasklet.
1559  */
1560 static void xs_udp_connect_worker6(struct work_struct *work)
1561 {
1562         struct sock_xprt *transport =
1563                 container_of(work, struct sock_xprt, connect_worker.work);
1564         struct rpc_xprt *xprt = &transport->xprt;
1565         struct socket *sock = transport->sock;
1566         int err, status = -EIO;
1567
1568         if (xprt->shutdown || !xprt_bound(xprt))
1569                 goto out;
1570
1571         /* Start by resetting any existing state */
1572         xs_close(xprt);
1573
1574         if ((err = sock_create_kern(PF_INET6, SOCK_DGRAM, IPPROTO_UDP, &sock)) < 0) {
1575                 dprintk("RPC:       can't create UDP transport socket (%d).\n", -err);
1576                 goto out;
1577         }
1578         xs_reclassify_socket6(sock);
1579
1580         if (xs_bind6(transport, sock) < 0) {
1581                 sock_release(sock);
1582                 goto out;
1583         }
1584
1585         dprintk("RPC:       worker connecting xprt %p to address: %s\n",
1586                         xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1587
1588         xs_udp_finish_connecting(xprt, sock);
1589         status = 0;
1590 out:
1591         xprt_wake_pending_tasks(xprt, status);
1592         xprt_clear_connecting(xprt);
1593 }
1594
1595 /*
1596  * We need to preserve the port number so the reply cache on the server can
1597  * find our cached RPC replies when we get around to reconnecting.
1598  */
1599 static void xs_tcp_reuse_connection(struct rpc_xprt *xprt)
1600 {
1601         int result;
1602         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1603         struct sockaddr any;
1604
1605         dprintk("RPC:       disconnecting xprt %p to reuse port\n", xprt);
1606
1607         /*
1608          * Disconnect the transport socket by doing a connect operation
1609          * with AF_UNSPEC.  This should return immediately...
1610          */
1611         memset(&any, 0, sizeof(any));
1612         any.sa_family = AF_UNSPEC;
1613         result = kernel_connect(transport->sock, &any, sizeof(any), 0);
1614         if (result)
1615                 dprintk("RPC:       AF_UNSPEC connect return code %d\n",
1616                                 result);
1617 }
1618
1619 static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
1620 {
1621         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1622
1623         if (!transport->inet) {
1624                 struct sock *sk = sock->sk;
1625
1626                 write_lock_bh(&sk->sk_callback_lock);
1627
1628                 xs_save_old_callbacks(transport, sk);
1629
1630                 sk->sk_user_data = xprt;
1631                 sk->sk_data_ready = xs_tcp_data_ready;
1632                 sk->sk_state_change = xs_tcp_state_change;
1633                 sk->sk_write_space = xs_tcp_write_space;
1634                 sk->sk_error_report = xs_tcp_error_report;
1635                 sk->sk_allocation = GFP_ATOMIC;
1636
1637                 /* socket options */
1638                 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
1639                 sock_reset_flag(sk, SOCK_LINGER);
1640                 tcp_sk(sk)->linger2 = 0;
1641                 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
1642
1643                 xprt_clear_connected(xprt);
1644
1645                 /* Reset to new socket */
1646                 transport->sock = sock;
1647                 transport->inet = sk;
1648
1649                 write_unlock_bh(&sk->sk_callback_lock);
1650         }
1651
1652         /* Tell the socket layer to start connecting... */
1653         xprt->stat.connect_count++;
1654         xprt->stat.connect_start = jiffies;
1655         return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
1656 }
1657
1658 /**
1659  * xs_tcp_connect_worker4 - connect a TCP socket to a remote endpoint
1660  * @work: RPC transport to connect
1661  *
1662  * Invoked by a work queue tasklet.
1663  */
1664 static void xs_tcp_connect_worker4(struct work_struct *work)
1665 {
1666         struct sock_xprt *transport =
1667                 container_of(work, struct sock_xprt, connect_worker.work);
1668         struct rpc_xprt *xprt = &transport->xprt;
1669         struct socket *sock = transport->sock;
1670         int err, status = -EIO;
1671
1672         if (xprt->shutdown || !xprt_bound(xprt))
1673                 goto out;
1674
1675         if (!sock) {
1676                 /* start from scratch */
1677                 if ((err = sock_create_kern(PF_INET, SOCK_STREAM, IPPROTO_TCP, &sock)) < 0) {
1678                         dprintk("RPC:       can't create TCP transport socket (%d).\n", -err);
1679                         goto out;
1680                 }
1681                 xs_reclassify_socket4(sock);
1682
1683                 if (xs_bind4(transport, sock) < 0) {
1684                         sock_release(sock);
1685                         goto out;
1686                 }
1687         } else
1688                 /* "close" the socket, preserving the local port */
1689                 xs_tcp_reuse_connection(xprt);
1690
1691         dprintk("RPC:       worker connecting xprt %p to address: %s\n",
1692                         xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1693
1694         status = xs_tcp_finish_connecting(xprt, sock);
1695         dprintk("RPC:       %p connect status %d connected %d sock state %d\n",
1696                         xprt, -status, xprt_connected(xprt),
1697                         sock->sk->sk_state);
1698         if (status < 0) {
1699                 switch (status) {
1700                         case -EINPROGRESS:
1701                         case -EALREADY:
1702                                 goto out_clear;
1703                         case -ECONNREFUSED:
1704                         case -ECONNRESET:
1705                                 /* retry with existing socket, after a delay */
1706                                 break;
1707                         default:
1708                                 /* get rid of existing socket, and retry */
1709                                 xs_tcp_shutdown(xprt);
1710                 }
1711         }
1712 out:
1713         xprt_wake_pending_tasks(xprt, status);
1714 out_clear:
1715         xprt_clear_connecting(xprt);
1716 }
1717
1718 /**
1719  * xs_tcp_connect_worker6 - connect a TCP socket to a remote endpoint
1720  * @work: RPC transport to connect
1721  *
1722  * Invoked by a work queue tasklet.
1723  */
1724 static void xs_tcp_connect_worker6(struct work_struct *work)
1725 {
1726         struct sock_xprt *transport =
1727                 container_of(work, struct sock_xprt, connect_worker.work);
1728         struct rpc_xprt *xprt = &transport->xprt;
1729         struct socket *sock = transport->sock;
1730         int err, status = -EIO;
1731
1732         if (xprt->shutdown || !xprt_bound(xprt))
1733                 goto out;
1734
1735         if (!sock) {
1736                 /* start from scratch */
1737                 if ((err = sock_create_kern(PF_INET6, SOCK_STREAM, IPPROTO_TCP, &sock)) < 0) {
1738                         dprintk("RPC:       can't create TCP transport socket (%d).\n", -err);
1739                         goto out;
1740                 }
1741                 xs_reclassify_socket6(sock);
1742
1743                 if (xs_bind6(transport, sock) < 0) {
1744                         sock_release(sock);
1745                         goto out;
1746                 }
1747         } else
1748                 /* "close" the socket, preserving the local port */
1749                 xs_tcp_reuse_connection(xprt);
1750
1751         dprintk("RPC:       worker connecting xprt %p to address: %s\n",
1752                         xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1753
1754         status = xs_tcp_finish_connecting(xprt, sock);
1755         dprintk("RPC:       %p connect status %d connected %d sock state %d\n",
1756                         xprt, -status, xprt_connected(xprt), sock->sk->sk_state);
1757         if (status < 0) {
1758                 switch (status) {
1759                         case -EINPROGRESS:
1760                         case -EALREADY:
1761                                 goto out_clear;
1762                         case -ECONNREFUSED:
1763                         case -ECONNRESET:
1764                                 /* retry with existing socket, after a delay */
1765                                 break;
1766                         default:
1767                                 /* get rid of existing socket, and retry */
1768                                 xs_tcp_shutdown(xprt);
1769                 }
1770         }
1771 out:
1772         xprt_wake_pending_tasks(xprt, status);
1773 out_clear:
1774         xprt_clear_connecting(xprt);
1775 }
1776
1777 /**
1778  * xs_connect - connect a socket to a remote endpoint
1779  * @task: address of RPC task that manages state of connect request
1780  *
1781  * TCP: If the remote end dropped the connection, delay reconnecting.
1782  *
1783  * UDP socket connects are synchronous, but we use a work queue anyway
1784  * to guarantee that even unprivileged user processes can set up a
1785  * socket on a privileged port.
1786  *
1787  * If a UDP socket connect fails, the delay behavior here prevents
1788  * retry floods (hard mounts).
1789  */
1790 static void xs_connect(struct rpc_task *task)
1791 {
1792         struct rpc_xprt *xprt = task->tk_xprt;
1793         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1794
1795         if (xprt_test_and_set_connecting(xprt))
1796                 return;
1797
1798         if (transport->sock != NULL) {
1799                 dprintk("RPC:       xs_connect delayed xprt %p for %lu "
1800                                 "seconds\n",
1801                                 xprt, xprt->reestablish_timeout / HZ);
1802                 queue_delayed_work(rpciod_workqueue,
1803                                    &transport->connect_worker,
1804                                    xprt->reestablish_timeout);
1805                 xprt->reestablish_timeout <<= 1;
1806                 if (xprt->reestablish_timeout > XS_TCP_MAX_REEST_TO)
1807                         xprt->reestablish_timeout = XS_TCP_MAX_REEST_TO;
1808         } else {
1809                 dprintk("RPC:       xs_connect scheduled xprt %p\n", xprt);
1810                 queue_delayed_work(rpciod_workqueue,
1811                                    &transport->connect_worker, 0);
1812         }
1813 }
1814
1815 static void xs_tcp_connect(struct rpc_task *task)
1816 {
1817         struct rpc_xprt *xprt = task->tk_xprt;
1818
1819         /* Initiate graceful shutdown of the socket if not already done */
1820         if (test_bit(XPRT_CONNECTED, &xprt->state))
1821                 xs_tcp_shutdown(xprt);
1822         /* Exit if we need to wait for socket shutdown to complete */
1823         if (test_bit(XPRT_CLOSING, &xprt->state))
1824                 return;
1825         xs_connect(task);
1826 }
1827
1828 /**
1829  * xs_udp_print_stats - display UDP socket-specifc stats
1830  * @xprt: rpc_xprt struct containing statistics
1831  * @seq: output file
1832  *
1833  */
1834 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
1835 {
1836         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1837
1838         seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %Lu %Lu\n",
1839                         transport->port,
1840                         xprt->stat.bind_count,
1841                         xprt->stat.sends,
1842                         xprt->stat.recvs,
1843                         xprt->stat.bad_xids,
1844                         xprt->stat.req_u,
1845                         xprt->stat.bklog_u);
1846 }
1847
1848 /**
1849  * xs_tcp_print_stats - display TCP socket-specifc stats
1850  * @xprt: rpc_xprt struct containing statistics
1851  * @seq: output file
1852  *
1853  */
1854 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
1855 {
1856         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1857         long idle_time = 0;
1858
1859         if (xprt_connected(xprt))
1860                 idle_time = (long)(jiffies - xprt->last_used) / HZ;
1861
1862         seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu %Lu %Lu\n",
1863                         transport->port,
1864                         xprt->stat.bind_count,
1865                         xprt->stat.connect_count,
1866                         xprt->stat.connect_time,
1867                         idle_time,
1868                         xprt->stat.sends,
1869                         xprt->stat.recvs,
1870                         xprt->stat.bad_xids,
1871                         xprt->stat.req_u,
1872                         xprt->stat.bklog_u);
1873 }
1874
1875 static struct rpc_xprt_ops xs_udp_ops = {
1876         .set_buffer_size        = xs_udp_set_buffer_size,
1877         .reserve_xprt           = xprt_reserve_xprt_cong,
1878         .release_xprt           = xprt_release_xprt_cong,
1879         .rpcbind                = rpcb_getport_async,
1880         .set_port               = xs_set_port,
1881         .connect                = xs_connect,
1882         .buf_alloc              = rpc_malloc,
1883         .buf_free               = rpc_free,
1884         .send_request           = xs_udp_send_request,
1885         .set_retrans_timeout    = xprt_set_retrans_timeout_rtt,
1886         .timer                  = xs_udp_timer,
1887         .release_request        = xprt_release_rqst_cong,
1888         .close                  = xs_close,
1889         .destroy                = xs_destroy,
1890         .print_stats            = xs_udp_print_stats,
1891 };
1892
1893 static struct rpc_xprt_ops xs_tcp_ops = {
1894         .reserve_xprt           = xprt_reserve_xprt,
1895         .release_xprt           = xs_tcp_release_xprt,
1896         .rpcbind                = rpcb_getport_async,
1897         .set_port               = xs_set_port,
1898         .connect                = xs_tcp_connect,
1899         .buf_alloc              = rpc_malloc,
1900         .buf_free               = rpc_free,
1901         .send_request           = xs_tcp_send_request,
1902         .set_retrans_timeout    = xprt_set_retrans_timeout_def,
1903         .close                  = xs_tcp_shutdown,
1904         .destroy                = xs_destroy,
1905         .print_stats            = xs_tcp_print_stats,
1906 };
1907
1908 static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
1909                                       unsigned int slot_table_size)
1910 {
1911         struct rpc_xprt *xprt;
1912         struct sock_xprt *new;
1913
1914         if (args->addrlen > sizeof(xprt->addr)) {
1915                 dprintk("RPC:       xs_setup_xprt: address too large\n");
1916                 return ERR_PTR(-EBADF);
1917         }
1918
1919         new = kzalloc(sizeof(*new), GFP_KERNEL);
1920         if (new == NULL) {
1921                 dprintk("RPC:       xs_setup_xprt: couldn't allocate "
1922                                 "rpc_xprt\n");
1923                 return ERR_PTR(-ENOMEM);
1924         }
1925         xprt = &new->xprt;
1926
1927         xprt->max_reqs = slot_table_size;
1928         xprt->slot = kcalloc(xprt->max_reqs, sizeof(struct rpc_rqst), GFP_KERNEL);
1929         if (xprt->slot == NULL) {
1930                 kfree(xprt);
1931                 dprintk("RPC:       xs_setup_xprt: couldn't allocate slot "
1932                                 "table\n");
1933                 return ERR_PTR(-ENOMEM);
1934         }
1935
1936         memcpy(&xprt->addr, args->dstaddr, args->addrlen);
1937         xprt->addrlen = args->addrlen;
1938         if (args->srcaddr)
1939                 memcpy(&new->addr, args->srcaddr, args->addrlen);
1940
1941         return xprt;
1942 }
1943
1944 static const struct rpc_timeout xs_udp_default_timeout = {
1945         .to_initval = 5 * HZ,
1946         .to_maxval = 30 * HZ,
1947         .to_increment = 5 * HZ,
1948         .to_retries = 5,
1949 };
1950
1951 /**
1952  * xs_setup_udp - Set up transport to use a UDP socket
1953  * @args: rpc transport creation arguments
1954  *
1955  */
1956 static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
1957 {
1958         struct sockaddr *addr = args->dstaddr;
1959         struct rpc_xprt *xprt;
1960         struct sock_xprt *transport;
1961
1962         xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries);
1963         if (IS_ERR(xprt))
1964                 return xprt;
1965         transport = container_of(xprt, struct sock_xprt, xprt);
1966
1967         xprt->prot = IPPROTO_UDP;
1968         xprt->tsh_size = 0;
1969         /* XXX: header size can vary due to auth type, IPv6, etc. */
1970         xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
1971
1972         xprt->bind_timeout = XS_BIND_TO;
1973         xprt->connect_timeout = XS_UDP_CONN_TO;
1974         xprt->reestablish_timeout = XS_UDP_REEST_TO;
1975         xprt->idle_timeout = XS_IDLE_DISC_TO;
1976
1977         xprt->ops = &xs_udp_ops;
1978
1979         xprt->timeout = &xs_udp_default_timeout;
1980
1981         switch (addr->sa_family) {
1982         case AF_INET:
1983                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
1984                         xprt_set_bound(xprt);
1985
1986                 INIT_DELAYED_WORK(&transport->connect_worker,
1987                                         xs_udp_connect_worker4);
1988                 xs_format_ipv4_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
1989                 break;
1990         case AF_INET6:
1991                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
1992                         xprt_set_bound(xprt);
1993
1994                 INIT_DELAYED_WORK(&transport->connect_worker,
1995                                         xs_udp_connect_worker6);
1996                 xs_format_ipv6_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
1997                 break;
1998         default:
1999                 kfree(xprt);
2000                 return ERR_PTR(-EAFNOSUPPORT);
2001         }
2002
2003         dprintk("RPC:       set up transport to address %s\n",
2004                         xprt->address_strings[RPC_DISPLAY_ALL]);
2005
2006         if (try_module_get(THIS_MODULE))
2007                 return xprt;
2008
2009         kfree(xprt->slot);
2010         kfree(xprt);
2011         return ERR_PTR(-EINVAL);
2012 }
2013
2014 static const struct rpc_timeout xs_tcp_default_timeout = {
2015         .to_initval = 60 * HZ,
2016         .to_maxval = 60 * HZ,
2017         .to_retries = 2,
2018 };
2019
2020 /**
2021  * xs_setup_tcp - Set up transport to use a TCP socket
2022  * @args: rpc transport creation arguments
2023  *
2024  */
2025 static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
2026 {
2027         struct sockaddr *addr = args->dstaddr;
2028         struct rpc_xprt *xprt;
2029         struct sock_xprt *transport;
2030
2031         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries);
2032         if (IS_ERR(xprt))
2033                 return xprt;
2034         transport = container_of(xprt, struct sock_xprt, xprt);
2035
2036         xprt->prot = IPPROTO_TCP;
2037         xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2038         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2039
2040         xprt->bind_timeout = XS_BIND_TO;
2041         xprt->connect_timeout = XS_TCP_CONN_TO;
2042         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2043         xprt->idle_timeout = XS_IDLE_DISC_TO;
2044
2045         xprt->ops = &xs_tcp_ops;
2046         xprt->timeout = &xs_tcp_default_timeout;
2047
2048         switch (addr->sa_family) {
2049         case AF_INET:
2050                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2051                         xprt_set_bound(xprt);
2052
2053                 INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_connect_worker4);
2054                 xs_format_ipv4_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
2055                 break;
2056         case AF_INET6:
2057                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2058                         xprt_set_bound(xprt);
2059
2060                 INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_connect_worker6);
2061                 xs_format_ipv6_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
2062                 break;
2063         default:
2064                 kfree(xprt);
2065                 return ERR_PTR(-EAFNOSUPPORT);
2066         }
2067
2068         dprintk("RPC:       set up transport to address %s\n",
2069                         xprt->address_strings[RPC_DISPLAY_ALL]);
2070
2071         if (try_module_get(THIS_MODULE))
2072                 return xprt;
2073
2074         kfree(xprt->slot);
2075         kfree(xprt);
2076         return ERR_PTR(-EINVAL);
2077 }
2078
2079 static struct xprt_class        xs_udp_transport = {
2080         .list           = LIST_HEAD_INIT(xs_udp_transport.list),
2081         .name           = "udp",
2082         .owner          = THIS_MODULE,
2083         .ident          = IPPROTO_UDP,
2084         .setup          = xs_setup_udp,
2085 };
2086
2087 static struct xprt_class        xs_tcp_transport = {
2088         .list           = LIST_HEAD_INIT(xs_tcp_transport.list),
2089         .name           = "tcp",
2090         .owner          = THIS_MODULE,
2091         .ident          = IPPROTO_TCP,
2092         .setup          = xs_setup_tcp,
2093 };
2094
2095 /**
2096  * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
2097  *
2098  */
2099 int init_socket_xprt(void)
2100 {
2101 #ifdef RPC_DEBUG
2102         if (!sunrpc_table_header)
2103                 sunrpc_table_header = register_sysctl_table(sunrpc_table);
2104 #endif
2105
2106         xprt_register_transport(&xs_udp_transport);
2107         xprt_register_transport(&xs_tcp_transport);
2108
2109         return 0;
2110 }
2111
2112 /**
2113  * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
2114  *
2115  */
2116 void cleanup_socket_xprt(void)
2117 {
2118 #ifdef RPC_DEBUG
2119         if (sunrpc_table_header) {
2120                 unregister_sysctl_table(sunrpc_table_header);
2121                 sunrpc_table_header = NULL;
2122         }
2123 #endif
2124
2125         xprt_unregister_transport(&xs_udp_transport);
2126         xprt_unregister_transport(&xs_tcp_transport);
2127 }