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, "%pI4", &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=%pI4 port=%u proto=%s",
 303                         &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, "%pI4.%u.%u",
 326                                 &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, "%pI6",&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=%pI6 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, "%pi6", &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, "%pI6.%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 -ENOTSOCK;
 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         if (!xprt_bound(xprt))
 581                 return -ENOTCONN;
 582         status = xs_sendpages(transport->sock,
 583                               xs_addr(xprt),
 584                               xprt->addrlen, xdr,
 585                               req->rq_bytes_sent);
 586
 587         dprintk("RPC:       xs_udp_send_request(%u) = %d\n",
 588                         xdr->len - req->rq_bytes_sent, status);
 589
 590         if (status >= 0) {
 591                 task->tk_bytes_sent += status;
 592                 if (status >= req->rq_slen)
 593                         return 0;
 594                 /* Still some bytes left; set up for a retry later. */
 595                 status = -EAGAIN;
 596         }
 597
 598         switch (status) {
 599         case -ENOTSOCK:
 600                 status = -ENOTCONN;
 601                 /* Should we call xs_close() here? */
 602                 break;
 603         case -EAGAIN:
 604                 xs_nospace(task);
 605                 break;
 606         case -ENETUNREACH:
 607         case -EPIPE:
 608         case -ECONNREFUSED:
 609                 /* When the server has died, an ICMP port unreachable message
 610                  * prompts ECONNREFUSED. */
 611                 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 612                 break;
 613         default:
 614                 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 615                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
 616                         -status);
 617         }
 618
 619         return status;
 620 }
 621
 622 /**
 623  * xs_tcp_shutdown - gracefully shut down a TCP socket
 624  * @xprt: transport
 625  *
 626  * Initiates a graceful shutdown of the TCP socket by calling the
 627  * equivalent of shutdown(SHUT_WR);
 628  */
 629 static void xs_tcp_shutdown(struct rpc_xprt *xprt)
 630 {
 631         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 632         struct socket *sock = transport->sock;
 633
 634         if (sock != NULL)
 635                 kernel_sock_shutdown(sock, SHUT_WR);
 636 }
 637
 638 static inline void xs_encode_tcp_record_marker(struct xdr_buf *buf)
 639 {
 640         u32 reclen = buf->len - sizeof(rpc_fraghdr);
 641         rpc_fraghdr *base = buf->head[0].iov_base;
 642         *base = htonl(RPC_LAST_STREAM_FRAGMENT | reclen);
 643 }
 644
 645 /**
 646  * xs_tcp_send_request - write an RPC request to a TCP socket
 647  * @task: address of RPC task that manages the state of an RPC request
 648  *
 649  * Return values:
 650  *        0:    The request has been sent
 651  *   EAGAIN:    The socket was blocked, please call again later to
 652  *              complete the request
 653  * ENOTCONN:    Caller needs to invoke connect logic then call again
 654  *    other:    Some other error occured, the request was not sent
 655  *
 656  * XXX: In the case of soft timeouts, should we eventually give up
 657  *      if sendmsg is not able to make progress?
 658  */
 659 static int xs_tcp_send_request(struct rpc_task *task)
 660 {
 661         struct rpc_rqst *req = task->tk_rqstp;
 662         struct rpc_xprt *xprt = req->rq_xprt;
 663         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 664         struct xdr_buf *xdr = &req->rq_snd_buf;
 665         int status;
 666
 667         xs_encode_tcp_record_marker(&req->rq_snd_buf);
 668
 669         xs_pktdump("packet data:",
 670                                 req->rq_svec->iov_base,
 671                                 req->rq_svec->iov_len);
 672
 673         /* Continue transmitting the packet/record. We must be careful
 674          * to cope with writespace callbacks arriving _after_ we have
 675          * called sendmsg(). */
 676         while (1) {
 677                 status = xs_sendpages(transport->sock,
 678                                         NULL, 0, xdr, req->rq_bytes_sent);
 679
 680                 dprintk("RPC:       xs_tcp_send_request(%u) = %d\n",
 681                                 xdr->len - req->rq_bytes_sent, status);
 682
 683                 if (unlikely(status < 0))
 684                         break;
 685
 686                 /* If we've sent the entire packet, immediately
 687                  * reset the count of bytes sent. */
 688                 req->rq_bytes_sent += status;
 689                 task->tk_bytes_sent += status;
 690                 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
 691                         req->rq_bytes_sent = 0;
 692                         return 0;
 693                 }
 694
 695                 if (status != 0)
 696                         continue;
 697                 status = -EAGAIN;
 698                 break;
 699         }
 700
 701         switch (status) {
 702         case -ENOTSOCK:
 703                 status = -ENOTCONN;
 704                 /* Should we call xs_close() here? */
 705                 break;
 706         case -EAGAIN:
 707                 xs_nospace(task);
 708                 break;
 709         case -ECONNRESET:
 710                 xs_tcp_shutdown(xprt);
 711         case -ECONNREFUSED:
 712         case -ENOTCONN:
 713         case -EPIPE:
 714                 status = -ENOTCONN;
 715                 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 716                 break;
 717         default:
 718                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
 719                         -status);
 720                 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 721                 xs_tcp_shutdown(xprt);
 722         }
 723
 724         return status;
 725 }
 726
 727 /**
 728  * xs_tcp_release_xprt - clean up after a tcp transmission
 729  * @xprt: transport
 730  * @task: rpc task
 731  *
 732  * This cleans up if an error causes us to abort the transmission of a request.
 733  * In this case, the socket may need to be reset in order to avoid confusing
 734  * the server.
 735  */
 736 static void xs_tcp_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
 737 {
 738         struct rpc_rqst *req;
 739
 740         if (task != xprt->snd_task)
 741                 return;
 742         if (task == NULL)
 743                 goto out_release;
 744         req = task->tk_rqstp;
 745         if (req->rq_bytes_sent == 0)
 746                 goto out_release;
 747         if (req->rq_bytes_sent == req->rq_snd_buf.len)
 748                 goto out_release;
 749         set_bit(XPRT_CLOSE_WAIT, &task->tk_xprt->state);
 750 out_release:
 751         xprt_release_xprt(xprt, task);
 752 }
 753
 754 static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk)
 755 {
 756         transport->old_data_ready = sk->sk_data_ready;
 757         transport->old_state_change = sk->sk_state_change;
 758         transport->old_write_space = sk->sk_write_space;
 759         transport->old_error_report = sk->sk_error_report;
 760 }
 761
 762 static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk)
 763 {
 764         sk->sk_data_ready = transport->old_data_ready;
 765         sk->sk_state_change = transport->old_state_change;
 766         sk->sk_write_space = transport->old_write_space;
 767         sk->sk_error_report = transport->old_error_report;
 768 }
 769
 770 /**
 771  * xs_close - close a socket
 772  * @xprt: transport
 773  *
 774  * This is used when all requests are complete; ie, no DRC state remains
 775  * on the server we want to save.
 776  */
 777 static void xs_close(struct rpc_xprt *xprt)
 778 {
 779         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 780         struct socket *sock = transport->sock;
 781         struct sock *sk = transport->inet;
 782
 783         if (!sk)
 784                 goto clear_close_wait;
 785
 786         dprintk("RPC:       xs_close xprt %p\n", xprt);
 787
 788         write_lock_bh(&sk->sk_callback_lock);
 789         transport->inet = NULL;
 790         transport->sock = NULL;
 791
 792         sk->sk_user_data = NULL;
 793
 794         xs_restore_old_callbacks(transport, sk);
 795         write_unlock_bh(&sk->sk_callback_lock);
 796
 797         sk->sk_no_check = 0;
 798
 799         sock_release(sock);
 800 clear_close_wait:
 801         smp_mb__before_clear_bit();
 802         clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
 803         clear_bit(XPRT_CLOSING, &xprt->state);
 804         smp_mb__after_clear_bit();
 805         xprt_disconnect_done(xprt);
 806 }
 807
 808 /**
 809  * xs_destroy - prepare to shutdown a transport
 810  * @xprt: doomed transport
 811  *
 812  */
 813 static void xs_destroy(struct rpc_xprt *xprt)
 814 {
 815         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 816
 817         dprintk("RPC:       xs_destroy xprt %p\n", xprt);
 818
 819         cancel_rearming_delayed_work(&transport->connect_worker);
 820
 821         xs_close(xprt);
 822         xs_free_peer_addresses(xprt);
 823         kfree(xprt->slot);
 824         kfree(xprt);
 825         module_put(THIS_MODULE);
 826 }
 827
 828 static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
 829 {
 830         return (struct rpc_xprt *) sk->sk_user_data;
 831 }
 832
 833 /**
 834  * xs_udp_data_ready - "data ready" callback for UDP sockets
 835  * @sk: socket with data to read
 836  * @len: how much data to read
 837  *
 838  */
 839 static void xs_udp_data_ready(struct sock *sk, int len)
 840 {
 841         struct rpc_task *task;
 842         struct rpc_xprt *xprt;
 843         struct rpc_rqst *rovr;
 844         struct sk_buff *skb;
 845         int err, repsize, copied;
 846         u32 _xid;
 847         __be32 *xp;
 848
 849         read_lock(&sk->sk_callback_lock);
 850         dprintk("RPC:       xs_udp_data_ready...\n");
 851         if (!(xprt = xprt_from_sock(sk)))
 852                 goto out;
 853
 854         if ((skb = skb_recv_datagram(sk, 0, 1, &err)) == NULL)
 855                 goto out;
 856
 857         if (xprt->shutdown)
 858                 goto dropit;
 859
 860         repsize = skb->len - sizeof(struct udphdr);
 861         if (repsize < 4) {
 862                 dprintk("RPC:       impossible RPC reply size %d!\n", repsize);
 863                 goto dropit;
 864         }
 865
 866         /* Copy the XID from the skb... */
 867         xp = skb_header_pointer(skb, sizeof(struct udphdr),
 868                                 sizeof(_xid), &_xid);
 869         if (xp == NULL)
 870                 goto dropit;
 871
 872         /* Look up and lock the request corresponding to the given XID */
 873         spin_lock(&xprt->transport_lock);
 874         rovr = xprt_lookup_rqst(xprt, *xp);
 875         if (!rovr)
 876                 goto out_unlock;
 877         task = rovr->rq_task;
 878
 879         if ((copied = rovr->rq_private_buf.buflen) > repsize)
 880                 copied = repsize;
 881
 882         /* Suck it into the iovec, verify checksum if not done by hw. */
 883         if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
 884                 UDPX_INC_STATS_BH(sk, UDP_MIB_INERRORS);
 885                 goto out_unlock;
 886         }
 887
 888         UDPX_INC_STATS_BH(sk, UDP_MIB_INDATAGRAMS);
 889
 890         /* Something worked... */
 891         dst_confirm(skb->dst);
 892
 893         xprt_adjust_cwnd(task, copied);
 894         xprt_update_rtt(task);
 895         xprt_complete_rqst(task, copied);
 896
 897  out_unlock:
 898         spin_unlock(&xprt->transport_lock);
 899  dropit:
 900         skb_free_datagram(sk, skb);
 901  out:
 902         read_unlock(&sk->sk_callback_lock);
 903 }
 904
 905 static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
 906 {
 907         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 908         size_t len, used;
 909         char *p;
 910
 911         p = ((char *) &transport->tcp_fraghdr) + transport->tcp_offset;
 912         len = sizeof(transport->tcp_fraghdr) - transport->tcp_offset;
 913         used = xdr_skb_read_bits(desc, p, len);
 914         transport->tcp_offset += used;
 915         if (used != len)
 916                 return;
 917
 918         transport->tcp_reclen = ntohl(transport->tcp_fraghdr);
 919         if (transport->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
 920                 transport->tcp_flags |= TCP_RCV_LAST_FRAG;
 921         else
 922                 transport->tcp_flags &= ~TCP_RCV_LAST_FRAG;
 923         transport->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
 924
 925         transport->tcp_flags &= ~TCP_RCV_COPY_FRAGHDR;
 926         transport->tcp_offset = 0;
 927
 928         /* Sanity check of the record length */
 929         if (unlikely(transport->tcp_reclen < 4)) {
 930                 dprintk("RPC:       invalid TCP record fragment length\n");
 931                 xprt_force_disconnect(xprt);
 932                 return;
 933         }
 934         dprintk("RPC:       reading TCP record fragment of length %d\n",
 935                         transport->tcp_reclen);
 936 }
 937
 938 static void xs_tcp_check_fraghdr(struct sock_xprt *transport)
 939 {
 940         if (transport->tcp_offset == transport->tcp_reclen) {
 941                 transport->tcp_flags |= TCP_RCV_COPY_FRAGHDR;
 942                 transport->tcp_offset = 0;
 943                 if (transport->tcp_flags & TCP_RCV_LAST_FRAG) {
 944                         transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
 945                         transport->tcp_flags |= TCP_RCV_COPY_XID;
 946                         transport->tcp_copied = 0;
 947                 }
 948         }
 949 }
 950
 951 static inline void xs_tcp_read_xid(struct sock_xprt *transport, struct xdr_skb_reader *desc)
 952 {
 953         size_t len, used;
 954         char *p;
 955
 956         len = sizeof(transport->tcp_xid) - transport->tcp_offset;
 957         dprintk("RPC:       reading XID (%Zu bytes)\n", len);
 958         p = ((char *) &transport->tcp_xid) + transport->tcp_offset;
 959         used = xdr_skb_read_bits(desc, p, len);
 960         transport->tcp_offset += used;
 961         if (used != len)
 962                 return;
 963         transport->tcp_flags &= ~TCP_RCV_COPY_XID;
 964         transport->tcp_flags |= TCP_RCV_COPY_DATA;
 965         transport->tcp_copied = 4;
 966         dprintk("RPC:       reading reply for XID %08x\n",
 967                         ntohl(transport->tcp_xid));
 968         xs_tcp_check_fraghdr(transport);
 969 }
 970
 971 static inline void xs_tcp_read_request(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
 972 {
 973         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 974         struct rpc_rqst *req;
 975         struct xdr_buf *rcvbuf;
 976         size_t len;
 977         ssize_t r;
 978
 979         /* Find and lock the request corresponding to this xid */
 980         spin_lock(&xprt->transport_lock);
 981         req = xprt_lookup_rqst(xprt, transport->tcp_xid);
 982         if (!req) {
 983                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
 984                 dprintk("RPC:       XID %08x request not found!\n",
 985                                 ntohl(transport->tcp_xid));
 986                 spin_unlock(&xprt->transport_lock);
 987                 return;
 988         }
 989
 990         rcvbuf = &req->rq_private_buf;
 991         len = desc->count;
 992         if (len > transport->tcp_reclen - transport->tcp_offset) {
 993                 struct xdr_skb_reader my_desc;
 994
 995                 len = transport->tcp_reclen - transport->tcp_offset;
 996                 memcpy(&my_desc, desc, sizeof(my_desc));
 997                 my_desc.count = len;
 998                 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
 999                                           &my_desc, xdr_skb_read_bits);
1000                 desc->count -= r;
1001                 desc->offset += r;
1002         } else
1003                 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
1004                                           desc, xdr_skb_read_bits);
1005
1006         if (r > 0) {
1007                 transport->tcp_copied += r;
1008                 transport->tcp_offset += r;
1009         }
1010         if (r != len) {
1011                 /* Error when copying to the receive buffer,
1012                  * usually because we weren't able to allocate
1013                  * additional buffer pages. All we can do now
1014                  * is turn off TCP_RCV_COPY_DATA, so the request
1015                  * will not receive any additional updates,
1016                  * and time out.
1017                  * Any remaining data from this record will
1018                  * be discarded.
1019                  */
1020                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1021                 dprintk("RPC:       XID %08x truncated request\n",
1022                                 ntohl(transport->tcp_xid));
1023                 dprintk("RPC:       xprt = %p, tcp_copied = %lu, "
1024                                 "tcp_offset = %u, tcp_reclen = %u\n",
1025                                 xprt, transport->tcp_copied,
1026                                 transport->tcp_offset, transport->tcp_reclen);
1027                 goto out;
1028         }
1029
1030         dprintk("RPC:       XID %08x read %Zd bytes\n",
1031                         ntohl(transport->tcp_xid), r);
1032         dprintk("RPC:       xprt = %p, tcp_copied = %lu, tcp_offset = %u, "
1033                         "tcp_reclen = %u\n", xprt, transport->tcp_copied,
1034                         transport->tcp_offset, transport->tcp_reclen);
1035
1036         if (transport->tcp_copied == req->rq_private_buf.buflen)
1037                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1038         else if (transport->tcp_offset == transport->tcp_reclen) {
1039                 if (transport->tcp_flags & TCP_RCV_LAST_FRAG)
1040                         transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1041         }
1042
1043 out:
1044         if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1045                 xprt_complete_rqst(req->rq_task, transport->tcp_copied);
1046         spin_unlock(&xprt->transport_lock);
1047         xs_tcp_check_fraghdr(transport);
1048 }
1049
1050 static inline void xs_tcp_read_discard(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1051 {
1052         size_t len;
1053
1054         len = transport->tcp_reclen - transport->tcp_offset;
1055         if (len > desc->count)
1056                 len = desc->count;
1057         desc->count -= len;
1058         desc->offset += len;
1059         transport->tcp_offset += len;
1060         dprintk("RPC:       discarded %Zu bytes\n", len);
1061         xs_tcp_check_fraghdr(transport);
1062 }
1063
1064 static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
1065 {
1066         struct rpc_xprt *xprt = rd_desc->arg.data;
1067         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1068         struct xdr_skb_reader desc = {
1069                 .skb    = skb,
1070                 .offset = offset,
1071                 .count  = len,
1072         };
1073
1074         dprintk("RPC:       xs_tcp_data_recv started\n");
1075         do {
1076                 /* Read in a new fragment marker if necessary */
1077                 /* Can we ever really expect to get completely empty fragments? */
1078                 if (transport->tcp_flags & TCP_RCV_COPY_FRAGHDR) {
1079                         xs_tcp_read_fraghdr(xprt, &desc);
1080                         continue;
1081                 }
1082                 /* Read in the xid if necessary */
1083                 if (transport->tcp_flags & TCP_RCV_COPY_XID) {
1084                         xs_tcp_read_xid(transport, &desc);
1085                         continue;
1086                 }
1087                 /* Read in the request data */
1088                 if (transport->tcp_flags & TCP_RCV_COPY_DATA) {
1089                         xs_tcp_read_request(xprt, &desc);
1090                         continue;
1091                 }
1092                 /* Skip over any trailing bytes on short reads */
1093                 xs_tcp_read_discard(transport, &desc);
1094         } while (desc.count);
1095         dprintk("RPC:       xs_tcp_data_recv done\n");
1096         return len - desc.count;
1097 }
1098
1099 /**
1100  * xs_tcp_data_ready - "data ready" callback for TCP sockets
1101  * @sk: socket with data to read
1102  * @bytes: how much data to read
1103  *
1104  */
1105 static void xs_tcp_data_ready(struct sock *sk, int bytes)
1106 {
1107         struct rpc_xprt *xprt;
1108         read_descriptor_t rd_desc;
1109         int read;
1110
1111         dprintk("RPC:       xs_tcp_data_ready...\n");
1112
1113         read_lock(&sk->sk_callback_lock);
1114         if (!(xprt = xprt_from_sock(sk)))
1115                 goto out;
1116         if (xprt->shutdown)
1117                 goto out;
1118
1119         /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
1120         rd_desc.arg.data = xprt;
1121         do {
1122                 rd_desc.count = 65536;
1123                 read = tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
1124         } while (read > 0);
1125 out:
1126         read_unlock(&sk->sk_callback_lock);
1127 }
1128
1129 /**
1130  * xs_tcp_state_change - callback to handle TCP socket state changes
1131  * @sk: socket whose state has changed
1132  *
1133  */
1134 static void xs_tcp_state_change(struct sock *sk)
1135 {
1136         struct rpc_xprt *xprt;
1137
1138         read_lock(&sk->sk_callback_lock);
1139         if (!(xprt = xprt_from_sock(sk)))
1140                 goto out;
1141         dprintk("RPC:       xs_tcp_state_change client %p...\n", xprt);
1142         dprintk("RPC:       state %x conn %d dead %d zapped %d\n",
1143                         sk->sk_state, xprt_connected(xprt),
1144                         sock_flag(sk, SOCK_DEAD),
1145                         sock_flag(sk, SOCK_ZAPPED));
1146
1147         switch (sk->sk_state) {
1148         case TCP_ESTABLISHED:
1149                 spin_lock_bh(&xprt->transport_lock);
1150                 if (!xprt_test_and_set_connected(xprt)) {
1151                         struct sock_xprt *transport = container_of(xprt,
1152                                         struct sock_xprt, xprt);
1153
1154                         /* Reset TCP record info */
1155                         transport->tcp_offset = 0;
1156                         transport->tcp_reclen = 0;
1157                         transport->tcp_copied = 0;
1158                         transport->tcp_flags =
1159                                 TCP_RCV_COPY_FRAGHDR | TCP_RCV_COPY_XID;
1160
1161                         xprt_wake_pending_tasks(xprt, 0);
1162                 }
1163                 spin_unlock_bh(&xprt->transport_lock);
1164                 break;
1165         case TCP_FIN_WAIT1:
1166                 /* The client initiated a shutdown of the socket */
1167                 xprt->connect_cookie++;
1168                 xprt->reestablish_timeout = 0;
1169                 set_bit(XPRT_CLOSING, &xprt->state);
1170                 smp_mb__before_clear_bit();
1171                 clear_bit(XPRT_CONNECTED, &xprt->state);
1172                 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1173                 smp_mb__after_clear_bit();
1174                 break;
1175         case TCP_CLOSE_WAIT:
1176                 /* The server initiated a shutdown of the socket */
1177                 set_bit(XPRT_CLOSING, &xprt->state);
1178                 xprt_force_disconnect(xprt);
1179         case TCP_SYN_SENT:
1180                 xprt->connect_cookie++;
1181         case TCP_CLOSING:
1182                 /*
1183                  * If the server closed down the connection, make sure that
1184                  * we back off before reconnecting
1185                  */
1186                 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1187                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1188                 break;
1189         case TCP_LAST_ACK:
1190                 smp_mb__before_clear_bit();
1191                 clear_bit(XPRT_CONNECTED, &xprt->state);
1192                 smp_mb__after_clear_bit();
1193                 break;
1194         case TCP_CLOSE:
1195                 smp_mb__before_clear_bit();
1196                 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1197                 clear_bit(XPRT_CLOSING, &xprt->state);
1198                 smp_mb__after_clear_bit();
1199                 /* Mark transport as closed and wake up all pending tasks */
1200                 xprt_disconnect_done(xprt);
1201         }
1202  out:
1203         read_unlock(&sk->sk_callback_lock);
1204 }
1205
1206 /**
1207  * xs_tcp_error_report - callback mainly for catching RST events
1208  * @sk: socket
1209  */
1210 static void xs_tcp_error_report(struct sock *sk)
1211 {
1212         struct rpc_xprt *xprt;
1213
1214         read_lock(&sk->sk_callback_lock);
1215         if (sk->sk_err != ECONNRESET || sk->sk_state != TCP_ESTABLISHED)
1216                 goto out;
1217         if (!(xprt = xprt_from_sock(sk)))
1218                 goto out;
1219         dprintk("RPC:       %s client %p...\n"
1220                         "RPC:       error %d\n",
1221                         __func__, xprt, sk->sk_err);
1222
1223         xprt_force_disconnect(xprt);
1224 out:
1225         read_unlock(&sk->sk_callback_lock);
1226 }
1227
1228 /**
1229  * xs_udp_write_space - callback invoked when socket buffer space
1230  *                             becomes available
1231  * @sk: socket whose state has changed
1232  *
1233  * Called when more output buffer space is available for this socket.
1234  * We try not to wake our writers until they can make "significant"
1235  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1236  * with a bunch of small requests.
1237  */
1238 static void xs_udp_write_space(struct sock *sk)
1239 {
1240         read_lock(&sk->sk_callback_lock);
1241
1242         /* from net/core/sock.c:sock_def_write_space */
1243         if (sock_writeable(sk)) {
1244                 struct socket *sock;
1245                 struct rpc_xprt *xprt;
1246
1247                 if (unlikely(!(sock = sk->sk_socket)))
1248                         goto out;
1249                 clear_bit(SOCK_NOSPACE, &sock->flags);
1250
1251                 if (unlikely(!(xprt = xprt_from_sock(sk))))
1252                         goto out;
1253                 if (test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags) == 0)
1254                         goto out;
1255
1256                 xprt_write_space(xprt);
1257         }
1258
1259  out:
1260         read_unlock(&sk->sk_callback_lock);
1261 }
1262
1263 /**
1264  * xs_tcp_write_space - callback invoked when socket buffer space
1265  *                             becomes available
1266  * @sk: socket whose state has changed
1267  *
1268  * Called when more output buffer space is available for this socket.
1269  * We try not to wake our writers until they can make "significant"
1270  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1271  * with a bunch of small requests.
1272  */
1273 static void xs_tcp_write_space(struct sock *sk)
1274 {
1275         read_lock(&sk->sk_callback_lock);
1276
1277         /* from net/core/stream.c:sk_stream_write_space */
1278         if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk)) {
1279                 struct socket *sock;
1280                 struct rpc_xprt *xprt;
1281
1282                 if (unlikely(!(sock = sk->sk_socket)))
1283                         goto out;
1284                 clear_bit(SOCK_NOSPACE, &sock->flags);
1285
1286                 if (unlikely(!(xprt = xprt_from_sock(sk))))
1287                         goto out;
1288                 if (test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags) == 0)
1289                         goto out;
1290
1291                 xprt_write_space(xprt);
1292         }
1293
1294  out:
1295         read_unlock(&sk->sk_callback_lock);
1296 }
1297
1298 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
1299 {
1300         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1301         struct sock *sk = transport->inet;
1302
1303         if (transport->rcvsize) {
1304                 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1305                 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
1306         }
1307         if (transport->sndsize) {
1308                 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1309                 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1310                 sk->sk_write_space(sk);
1311         }
1312 }
1313
1314 /**
1315  * xs_udp_set_buffer_size - set send and receive limits
1316  * @xprt: generic transport
1317  * @sndsize: requested size of send buffer, in bytes
1318  * @rcvsize: requested size of receive buffer, in bytes
1319  *
1320  * Set socket send and receive buffer size limits.
1321  */
1322 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1323 {
1324         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1325
1326         transport->sndsize = 0;
1327         if (sndsize)
1328                 transport->sndsize = sndsize + 1024;
1329         transport->rcvsize = 0;
1330         if (rcvsize)
1331                 transport->rcvsize = rcvsize + 1024;
1332
1333         xs_udp_do_set_buffer_size(xprt);
1334 }
1335
1336 /**
1337  * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1338  * @task: task that timed out
1339  *
1340  * Adjust the congestion window after a retransmit timeout has occurred.
1341  */
1342 static void xs_udp_timer(struct rpc_task *task)
1343 {
1344         xprt_adjust_cwnd(task, -ETIMEDOUT);
1345 }
1346
1347 static unsigned short xs_get_random_port(void)
1348 {
1349         unsigned short range = xprt_max_resvport - xprt_min_resvport;
1350         unsigned short rand = (unsigned short) net_random() % range;
1351         return rand + xprt_min_resvport;
1352 }
1353
1354 /**
1355  * xs_set_port - reset the port number in the remote endpoint address
1356  * @xprt: generic transport
1357  * @port: new port number
1358  *
1359  */
1360 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1361 {
1362         struct sockaddr *addr = xs_addr(xprt);
1363
1364         dprintk("RPC:       setting port for xprt %p to %u\n", xprt, port);
1365
1366         switch (addr->sa_family) {
1367         case AF_INET:
1368                 ((struct sockaddr_in *)addr)->sin_port = htons(port);
1369                 break;
1370         case AF_INET6:
1371                 ((struct sockaddr_in6 *)addr)->sin6_port = htons(port);
1372                 break;
1373         default:
1374                 BUG();
1375         }
1376 }
1377
1378 static unsigned short xs_get_srcport(struct sock_xprt *transport, struct socket *sock)
1379 {
1380         unsigned short port = transport->port;
1381
1382         if (port == 0 && transport->xprt.resvport)
1383                 port = xs_get_random_port();
1384         return port;
1385 }
1386
1387 static unsigned short xs_next_srcport(struct sock_xprt *transport, struct socket *sock, unsigned short port)
1388 {
1389         if (transport->port != 0)
1390                 transport->port = 0;
1391         if (!transport->xprt.resvport)
1392                 return 0;
1393         if (port <= xprt_min_resvport || port > xprt_max_resvport)
1394                 return xprt_max_resvport;
1395         return --port;
1396 }
1397
1398 static int xs_bind4(struct sock_xprt *transport, struct socket *sock)
1399 {
1400         struct sockaddr_in myaddr = {
1401                 .sin_family = AF_INET,
1402         };
1403         struct sockaddr_in *sa;
1404         int err, nloop = 0;
1405         unsigned short port = xs_get_srcport(transport, sock);
1406         unsigned short last;
1407
1408         sa = (struct sockaddr_in *)&transport->addr;
1409         myaddr.sin_addr = sa->sin_addr;
1410         do {
1411                 myaddr.sin_port = htons(port);
1412                 err = kernel_bind(sock, (struct sockaddr *) &myaddr,
1413                                                 sizeof(myaddr));
1414                 if (port == 0)
1415                         break;
1416                 if (err == 0) {
1417                         transport->port = port;
1418                         break;
1419                 }
1420                 last = port;
1421                 port = xs_next_srcport(transport, sock, port);
1422                 if (port > last)
1423                         nloop++;
1424         } while (err == -EADDRINUSE && nloop != 2);
1425         dprintk("RPC:       %s %pI4:%u: %s (%d)\n",
1426                         __func__, &myaddr.sin_addr,
1427                         port, err ? "failed" : "ok", err);
1428         return err;
1429 }
1430
1431 static int xs_bind6(struct sock_xprt *transport, struct socket *sock)
1432 {
1433         struct sockaddr_in6 myaddr = {
1434                 .sin6_family = AF_INET6,
1435         };
1436         struct sockaddr_in6 *sa;
1437         int err, nloop = 0;
1438         unsigned short port = xs_get_srcport(transport, sock);
1439         unsigned short last;
1440
1441         sa = (struct sockaddr_in6 *)&transport->addr;
1442         myaddr.sin6_addr = sa->sin6_addr;
1443         do {
1444                 myaddr.sin6_port = htons(port);
1445                 err = kernel_bind(sock, (struct sockaddr *) &myaddr,
1446                                                 sizeof(myaddr));
1447                 if (port == 0)
1448                         break;
1449                 if (err == 0) {
1450                         transport->port = port;
1451                         break;
1452                 }
1453                 last = port;
1454                 port = xs_next_srcport(transport, sock, port);
1455                 if (port > last)
1456                         nloop++;
1457         } while (err == -EADDRINUSE && nloop != 2);
1458         dprintk("RPC:       xs_bind6 %pI6:%u: %s (%d)\n",
1459                 &myaddr.sin6_addr, port, err ? "failed" : "ok", err);
1460         return err;
1461 }
1462
1463 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1464 static struct lock_class_key xs_key[2];
1465 static struct lock_class_key xs_slock_key[2];
1466
1467 static inline void xs_reclassify_socket4(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_INET-RPC",
1473                 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1474 }
1475
1476 static inline void xs_reclassify_socket6(struct socket *sock)
1477 {
1478         struct sock *sk = sock->sk;
1479
1480         BUG_ON(sock_owned_by_user(sk));
1481         sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1482                 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
1483 }
1484 #else
1485 static inline void xs_reclassify_socket4(struct socket *sock)
1486 {
1487 }
1488
1489 static inline void xs_reclassify_socket6(struct socket *sock)
1490 {
1491 }
1492 #endif
1493
1494 static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
1495 {
1496         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1497
1498         if (!transport->inet) {
1499                 struct sock *sk = sock->sk;
1500
1501                 write_lock_bh(&sk->sk_callback_lock);
1502
1503                 xs_save_old_callbacks(transport, sk);
1504
1505                 sk->sk_user_data = xprt;
1506                 sk->sk_data_ready = xs_udp_data_ready;
1507                 sk->sk_write_space = xs_udp_write_space;
1508                 sk->sk_no_check = UDP_CSUM_NORCV;
1509                 sk->sk_allocation = GFP_ATOMIC;
1510
1511                 xprt_set_connected(xprt);
1512
1513                 /* Reset to new socket */
1514                 transport->sock = sock;
1515                 transport->inet = sk;
1516
1517                 write_unlock_bh(&sk->sk_callback_lock);
1518         }
1519         xs_udp_do_set_buffer_size(xprt);
1520 }
1521
1522 /**
1523  * xs_udp_connect_worker4 - set up a UDP socket
1524  * @work: RPC transport to connect
1525  *
1526  * Invoked by a work queue tasklet.
1527  */
1528 static void xs_udp_connect_worker4(struct work_struct *work)
1529 {
1530         struct sock_xprt *transport =
1531                 container_of(work, struct sock_xprt, connect_worker.work);
1532         struct rpc_xprt *xprt = &transport->xprt;
1533         struct socket *sock = transport->sock;
1534         int err, status = -EIO;
1535
1536         if (xprt->shutdown)
1537                 goto out;
1538
1539         /* Start by resetting any existing state */
1540         xs_close(xprt);
1541
1542         if ((err = sock_create_kern(PF_INET, SOCK_DGRAM, IPPROTO_UDP, &sock)) < 0) {
1543                 dprintk("RPC:       can't create UDP transport socket (%d).\n", -err);
1544                 goto out;
1545         }
1546         xs_reclassify_socket4(sock);
1547
1548         if (xs_bind4(transport, sock)) {
1549                 sock_release(sock);
1550                 goto out;
1551         }
1552
1553         dprintk("RPC:       worker connecting xprt %p to address: %s\n",
1554                         xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1555
1556         xs_udp_finish_connecting(xprt, sock);
1557         status = 0;
1558 out:
1559         xprt_wake_pending_tasks(xprt, status);
1560         xprt_clear_connecting(xprt);
1561 }
1562
1563 /**
1564  * xs_udp_connect_worker6 - set up a UDP socket
1565  * @work: RPC transport to connect
1566  *
1567  * Invoked by a work queue tasklet.
1568  */
1569 static void xs_udp_connect_worker6(struct work_struct *work)
1570 {
1571         struct sock_xprt *transport =
1572                 container_of(work, struct sock_xprt, connect_worker.work);
1573         struct rpc_xprt *xprt = &transport->xprt;
1574         struct socket *sock = transport->sock;
1575         int err, status = -EIO;
1576
1577         if (xprt->shutdown)
1578                 goto out;
1579
1580         /* Start by resetting any existing state */
1581         xs_close(xprt);
1582
1583         if ((err = sock_create_kern(PF_INET6, SOCK_DGRAM, IPPROTO_UDP, &sock)) < 0) {
1584                 dprintk("RPC:       can't create UDP transport socket (%d).\n", -err);
1585                 goto out;
1586         }
1587         xs_reclassify_socket6(sock);
1588
1589         if (xs_bind6(transport, sock) < 0) {
1590                 sock_release(sock);
1591                 goto out;
1592         }
1593
1594         dprintk("RPC:       worker connecting xprt %p to address: %s\n",
1595                         xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1596
1597         xs_udp_finish_connecting(xprt, sock);
1598         status = 0;
1599 out:
1600         xprt_wake_pending_tasks(xprt, status);
1601         xprt_clear_connecting(xprt);
1602 }
1603
1604 /*
1605  * We need to preserve the port number so the reply cache on the server can
1606  * find our cached RPC replies when we get around to reconnecting.
1607  */
1608 static void xs_tcp_reuse_connection(struct rpc_xprt *xprt)
1609 {
1610         int result;
1611         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1612         struct sockaddr any;
1613
1614         dprintk("RPC:       disconnecting xprt %p to reuse port\n", xprt);
1615
1616         /*
1617          * Disconnect the transport socket by doing a connect operation
1618          * with AF_UNSPEC.  This should return immediately...
1619          */
1620         memset(&any, 0, sizeof(any));
1621         any.sa_family = AF_UNSPEC;
1622         result = kernel_connect(transport->sock, &any, sizeof(any), 0);
1623         if (result)
1624                 dprintk("RPC:       AF_UNSPEC connect return code %d\n",
1625                                 result);
1626 }
1627
1628 static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
1629 {
1630         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1631
1632         if (!transport->inet) {
1633                 struct sock *sk = sock->sk;
1634
1635                 write_lock_bh(&sk->sk_callback_lock);
1636
1637                 xs_save_old_callbacks(transport, sk);
1638
1639                 sk->sk_user_data = xprt;
1640                 sk->sk_data_ready = xs_tcp_data_ready;
1641                 sk->sk_state_change = xs_tcp_state_change;
1642                 sk->sk_write_space = xs_tcp_write_space;
1643                 sk->sk_error_report = xs_tcp_error_report;
1644                 sk->sk_allocation = GFP_ATOMIC;
1645
1646                 /* socket options */
1647                 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
1648                 sock_reset_flag(sk, SOCK_LINGER);
1649                 tcp_sk(sk)->linger2 = 0;
1650                 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
1651
1652                 xprt_clear_connected(xprt);
1653
1654                 /* Reset to new socket */
1655                 transport->sock = sock;
1656                 transport->inet = sk;
1657
1658                 write_unlock_bh(&sk->sk_callback_lock);
1659         }
1660
1661         if (!xprt_bound(xprt))
1662                 return -ENOTCONN;
1663
1664         /* Tell the socket layer to start connecting... */
1665         xprt->stat.connect_count++;
1666         xprt->stat.connect_start = jiffies;
1667         return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
1668 }
1669
1670 /**
1671  * xs_tcp_connect_worker4 - connect a TCP socket to a remote endpoint
1672  * @work: RPC transport to connect
1673  *
1674  * Invoked by a work queue tasklet.
1675  */
1676 static void xs_tcp_connect_worker4(struct work_struct *work)
1677 {
1678         struct sock_xprt *transport =
1679                 container_of(work, struct sock_xprt, connect_worker.work);
1680         struct rpc_xprt *xprt = &transport->xprt;
1681         struct socket *sock = transport->sock;
1682         int err, status = -EIO;
1683
1684         if (xprt->shutdown)
1685                 goto out;
1686
1687         if (!sock) {
1688                 /* start from scratch */
1689                 if ((err = sock_create_kern(PF_INET, SOCK_STREAM, IPPROTO_TCP, &sock)) < 0) {
1690                         dprintk("RPC:       can't create TCP transport socket (%d).\n", -err);
1691                         goto out;
1692                 }
1693                 xs_reclassify_socket4(sock);
1694
1695                 if (xs_bind4(transport, sock) < 0) {
1696                         sock_release(sock);
1697                         goto out;
1698                 }
1699         } else
1700                 /* "close" the socket, preserving the local port */
1701                 xs_tcp_reuse_connection(xprt);
1702
1703         dprintk("RPC:       worker connecting xprt %p to address: %s\n",
1704                         xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1705
1706         status = xs_tcp_finish_connecting(xprt, sock);
1707         dprintk("RPC:       %p connect status %d connected %d sock state %d\n",
1708                         xprt, -status, xprt_connected(xprt),
1709                         sock->sk->sk_state);
1710         if (status < 0) {
1711                 switch (status) {
1712                         case -EINPROGRESS:
1713                         case -EALREADY:
1714                                 goto out_clear;
1715                         case -ECONNREFUSED:
1716                         case -ECONNRESET:
1717                                 /* retry with existing socket, after a delay */
1718                                 break;
1719                         default:
1720                                 /* get rid of existing socket, and retry */
1721                                 xs_tcp_shutdown(xprt);
1722                 }
1723         }
1724 out:
1725         xprt_wake_pending_tasks(xprt, status);
1726 out_clear:
1727         xprt_clear_connecting(xprt);
1728 }
1729
1730 /**
1731  * xs_tcp_connect_worker6 - connect a TCP socket to a remote endpoint
1732  * @work: RPC transport to connect
1733  *
1734  * Invoked by a work queue tasklet.
1735  */
1736 static void xs_tcp_connect_worker6(struct work_struct *work)
1737 {
1738         struct sock_xprt *transport =
1739                 container_of(work, struct sock_xprt, connect_worker.work);
1740         struct rpc_xprt *xprt = &transport->xprt;
1741         struct socket *sock = transport->sock;
1742         int err, status = -EIO;
1743
1744         if (xprt->shutdown)
1745                 goto out;
1746
1747         if (!sock) {
1748                 /* start from scratch */
1749                 if ((err = sock_create_kern(PF_INET6, SOCK_STREAM, IPPROTO_TCP, &sock)) < 0) {
1750                         dprintk("RPC:       can't create TCP transport socket (%d).\n", -err);
1751                         goto out;
1752                 }
1753                 xs_reclassify_socket6(sock);
1754
1755                 if (xs_bind6(transport, sock) < 0) {
1756                         sock_release(sock);
1757                         goto out;
1758                 }
1759         } else
1760                 /* "close" the socket, preserving the local port */
1761                 xs_tcp_reuse_connection(xprt);
1762
1763         dprintk("RPC:       worker connecting xprt %p to address: %s\n",
1764                         xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1765
1766         status = xs_tcp_finish_connecting(xprt, sock);
1767         dprintk("RPC:       %p connect status %d connected %d sock state %d\n",
1768                         xprt, -status, xprt_connected(xprt), sock->sk->sk_state);
1769         if (status < 0) {
1770                 switch (status) {
1771                         case -EINPROGRESS:
1772                         case -EALREADY:
1773                                 goto out_clear;
1774                         case -ECONNREFUSED:
1775                         case -ECONNRESET:
1776                                 /* retry with existing socket, after a delay */
1777                                 break;
1778                         default:
1779                                 /* get rid of existing socket, and retry */
1780                                 xs_tcp_shutdown(xprt);
1781                 }
1782         }
1783 out:
1784         xprt_wake_pending_tasks(xprt, status);
1785 out_clear:
1786         xprt_clear_connecting(xprt);
1787 }
1788
1789 /**
1790  * xs_connect - connect a socket to a remote endpoint
1791  * @task: address of RPC task that manages state of connect request
1792  *
1793  * TCP: If the remote end dropped the connection, delay reconnecting.
1794  *
1795  * UDP socket connects are synchronous, but we use a work queue anyway
1796  * to guarantee that even unprivileged user processes can set up a
1797  * socket on a privileged port.
1798  *
1799  * If a UDP socket connect fails, the delay behavior here prevents
1800  * retry floods (hard mounts).
1801  */
1802 static void xs_connect(struct rpc_task *task)
1803 {
1804         struct rpc_xprt *xprt = task->tk_xprt;
1805         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1806
1807         if (xprt_test_and_set_connecting(xprt))
1808                 return;
1809
1810         if (transport->sock != NULL) {
1811                 dprintk("RPC:       xs_connect delayed xprt %p for %lu "
1812                                 "seconds\n",
1813                                 xprt, xprt->reestablish_timeout / HZ);
1814                 queue_delayed_work(rpciod_workqueue,
1815                                    &transport->connect_worker,
1816                                    xprt->reestablish_timeout);
1817                 xprt->reestablish_timeout <<= 1;
1818                 if (xprt->reestablish_timeout > XS_TCP_MAX_REEST_TO)
1819                         xprt->reestablish_timeout = XS_TCP_MAX_REEST_TO;
1820         } else {
1821                 dprintk("RPC:       xs_connect scheduled xprt %p\n", xprt);
1822                 queue_delayed_work(rpciod_workqueue,
1823                                    &transport->connect_worker, 0);
1824         }
1825 }
1826
1827 static void xs_tcp_connect(struct rpc_task *task)
1828 {
1829         struct rpc_xprt *xprt = task->tk_xprt;
1830
1831         /* Initiate graceful shutdown of the socket if not already done */
1832         if (test_bit(XPRT_CONNECTED, &xprt->state))
1833                 xs_tcp_shutdown(xprt);
1834         /* Exit if we need to wait for socket shutdown to complete */
1835         if (test_bit(XPRT_CLOSING, &xprt->state))
1836                 return;
1837         xs_connect(task);
1838 }
1839
1840 /**
1841  * xs_udp_print_stats - display UDP socket-specifc stats
1842  * @xprt: rpc_xprt struct containing statistics
1843  * @seq: output file
1844  *
1845  */
1846 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
1847 {
1848         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1849
1850         seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %Lu %Lu\n",
1851                         transport->port,
1852                         xprt->stat.bind_count,
1853                         xprt->stat.sends,
1854                         xprt->stat.recvs,
1855                         xprt->stat.bad_xids,
1856                         xprt->stat.req_u,
1857                         xprt->stat.bklog_u);
1858 }
1859
1860 /**
1861  * xs_tcp_print_stats - display TCP socket-specifc stats
1862  * @xprt: rpc_xprt struct containing statistics
1863  * @seq: output file
1864  *
1865  */
1866 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
1867 {
1868         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1869         long idle_time = 0;
1870
1871         if (xprt_connected(xprt))
1872                 idle_time = (long)(jiffies - xprt->last_used) / HZ;
1873
1874         seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu %Lu %Lu\n",
1875                         transport->port,
1876                         xprt->stat.bind_count,
1877                         xprt->stat.connect_count,
1878                         xprt->stat.connect_time,
1879                         idle_time,
1880                         xprt->stat.sends,
1881                         xprt->stat.recvs,
1882                         xprt->stat.bad_xids,
1883                         xprt->stat.req_u,
1884                         xprt->stat.bklog_u);
1885 }
1886
1887 static struct rpc_xprt_ops xs_udp_ops = {
1888         .set_buffer_size        = xs_udp_set_buffer_size,
1889         .reserve_xprt           = xprt_reserve_xprt_cong,
1890         .release_xprt           = xprt_release_xprt_cong,
1891         .rpcbind                = rpcb_getport_async,
1892         .set_port               = xs_set_port,
1893         .connect                = xs_connect,
1894         .buf_alloc              = rpc_malloc,
1895         .buf_free               = rpc_free,
1896         .send_request           = xs_udp_send_request,
1897         .set_retrans_timeout    = xprt_set_retrans_timeout_rtt,
1898         .timer                  = xs_udp_timer,
1899         .release_request        = xprt_release_rqst_cong,
1900         .close                  = xs_close,
1901         .destroy                = xs_destroy,
1902         .print_stats            = xs_udp_print_stats,
1903 };
1904
1905 static struct rpc_xprt_ops xs_tcp_ops = {
1906         .reserve_xprt           = xprt_reserve_xprt,
1907         .release_xprt           = xs_tcp_release_xprt,
1908         .rpcbind                = rpcb_getport_async,
1909         .set_port               = xs_set_port,
1910         .connect                = xs_tcp_connect,
1911         .buf_alloc              = rpc_malloc,
1912         .buf_free               = rpc_free,
1913         .send_request           = xs_tcp_send_request,
1914         .set_retrans_timeout    = xprt_set_retrans_timeout_def,
1915         .close                  = xs_tcp_shutdown,
1916         .destroy                = xs_destroy,
1917         .print_stats            = xs_tcp_print_stats,
1918 };
1919
1920 static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
1921                                       unsigned int slot_table_size)
1922 {
1923         struct rpc_xprt *xprt;
1924         struct sock_xprt *new;
1925
1926         if (args->addrlen > sizeof(xprt->addr)) {
1927                 dprintk("RPC:       xs_setup_xprt: address too large\n");
1928                 return ERR_PTR(-EBADF);
1929         }
1930
1931         new = kzalloc(sizeof(*new), GFP_KERNEL);
1932         if (new == NULL) {
1933                 dprintk("RPC:       xs_setup_xprt: couldn't allocate "
1934                                 "rpc_xprt\n");
1935                 return ERR_PTR(-ENOMEM);
1936         }
1937         xprt = &new->xprt;
1938
1939         xprt->max_reqs = slot_table_size;
1940         xprt->slot = kcalloc(xprt->max_reqs, sizeof(struct rpc_rqst), GFP_KERNEL);
1941         if (xprt->slot == NULL) {
1942                 kfree(xprt);
1943                 dprintk("RPC:       xs_setup_xprt: couldn't allocate slot "
1944                                 "table\n");
1945                 return ERR_PTR(-ENOMEM);
1946         }
1947
1948         memcpy(&xprt->addr, args->dstaddr, args->addrlen);
1949         xprt->addrlen = args->addrlen;
1950         if (args->srcaddr)
1951                 memcpy(&new->addr, args->srcaddr, args->addrlen);
1952
1953         return xprt;
1954 }
1955
1956 static const struct rpc_timeout xs_udp_default_timeout = {
1957         .to_initval = 5 * HZ,
1958         .to_maxval = 30 * HZ,
1959         .to_increment = 5 * HZ,
1960         .to_retries = 5,
1961 };
1962
1963 /**
1964  * xs_setup_udp - Set up transport to use a UDP socket
1965  * @args: rpc transport creation arguments
1966  *
1967  */
1968 static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
1969 {
1970         struct sockaddr *addr = args->dstaddr;
1971         struct rpc_xprt *xprt;
1972         struct sock_xprt *transport;
1973
1974         xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries);
1975         if (IS_ERR(xprt))
1976                 return xprt;
1977         transport = container_of(xprt, struct sock_xprt, xprt);
1978
1979         xprt->prot = IPPROTO_UDP;
1980         xprt->tsh_size = 0;
1981         /* XXX: header size can vary due to auth type, IPv6, etc. */
1982         xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
1983
1984         xprt->bind_timeout = XS_BIND_TO;
1985         xprt->connect_timeout = XS_UDP_CONN_TO;
1986         xprt->reestablish_timeout = XS_UDP_REEST_TO;
1987         xprt->idle_timeout = XS_IDLE_DISC_TO;
1988
1989         xprt->ops = &xs_udp_ops;
1990
1991         xprt->timeout = &xs_udp_default_timeout;
1992
1993         switch (addr->sa_family) {
1994         case AF_INET:
1995                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
1996                         xprt_set_bound(xprt);
1997
1998                 INIT_DELAYED_WORK(&transport->connect_worker,
1999                                         xs_udp_connect_worker4);
2000                 xs_format_ipv4_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
2001                 break;
2002         case AF_INET6:
2003                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2004                         xprt_set_bound(xprt);
2005
2006                 INIT_DELAYED_WORK(&transport->connect_worker,
2007                                         xs_udp_connect_worker6);
2008                 xs_format_ipv6_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
2009                 break;
2010         default:
2011                 kfree(xprt);
2012                 return ERR_PTR(-EAFNOSUPPORT);
2013         }
2014
2015         dprintk("RPC:       set up transport to address %s\n",
2016                         xprt->address_strings[RPC_DISPLAY_ALL]);
2017
2018         if (try_module_get(THIS_MODULE))
2019                 return xprt;
2020
2021         kfree(xprt->slot);
2022         kfree(xprt);
2023         return ERR_PTR(-EINVAL);
2024 }
2025
2026 static const struct rpc_timeout xs_tcp_default_timeout = {
2027         .to_initval = 60 * HZ,
2028         .to_maxval = 60 * HZ,
2029         .to_retries = 2,
2030 };
2031
2032 /**
2033  * xs_setup_tcp - Set up transport to use a TCP socket
2034  * @args: rpc transport creation arguments
2035  *
2036  */
2037 static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
2038 {
2039         struct sockaddr *addr = args->dstaddr;
2040         struct rpc_xprt *xprt;
2041         struct sock_xprt *transport;
2042
2043         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries);
2044         if (IS_ERR(xprt))
2045                 return xprt;
2046         transport = container_of(xprt, struct sock_xprt, xprt);
2047
2048         xprt->prot = IPPROTO_TCP;
2049         xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2050         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2051
2052         xprt->bind_timeout = XS_BIND_TO;
2053         xprt->connect_timeout = XS_TCP_CONN_TO;
2054         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2055         xprt->idle_timeout = XS_IDLE_DISC_TO;
2056
2057         xprt->ops = &xs_tcp_ops;
2058         xprt->timeout = &xs_tcp_default_timeout;
2059
2060         switch (addr->sa_family) {
2061         case AF_INET:
2062                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2063                         xprt_set_bound(xprt);
2064
2065                 INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_connect_worker4);
2066                 xs_format_ipv4_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
2067                 break;
2068         case AF_INET6:
2069                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2070                         xprt_set_bound(xprt);
2071
2072                 INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_connect_worker6);
2073                 xs_format_ipv6_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
2074                 break;
2075         default:
2076                 kfree(xprt);
2077                 return ERR_PTR(-EAFNOSUPPORT);
2078         }
2079
2080         dprintk("RPC:       set up transport to address %s\n",
2081                         xprt->address_strings[RPC_DISPLAY_ALL]);
2082
2083         if (try_module_get(THIS_MODULE))
2084                 return xprt;
2085
2086         kfree(xprt->slot);
2087         kfree(xprt);
2088         return ERR_PTR(-EINVAL);
2089 }
2090
2091 static struct xprt_class        xs_udp_transport = {
2092         .list           = LIST_HEAD_INIT(xs_udp_transport.list),
2093         .name           = "udp",
2094         .owner          = THIS_MODULE,
2095         .ident          = IPPROTO_UDP,
2096         .setup          = xs_setup_udp,
2097 };
2098
2099 static struct xprt_class        xs_tcp_transport = {
2100         .list           = LIST_HEAD_INIT(xs_tcp_transport.list),
2101         .name           = "tcp",
2102         .owner          = THIS_MODULE,
2103         .ident          = IPPROTO_TCP,
2104         .setup          = xs_setup_tcp,
2105 };
2106
2107 /**
2108  * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
2109  *
2110  */
2111 int init_socket_xprt(void)
2112 {
2113 #ifdef RPC_DEBUG
2114         if (!sunrpc_table_header)
2115                 sunrpc_table_header = register_sysctl_table(sunrpc_table);
2116 #endif
2117
2118         xprt_register_transport(&xs_udp_transport);
2119         xprt_register_transport(&xs_tcp_transport);
2120
2121         return 0;
2122 }
2123
2124 /**
2125  * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
2126  *
2127  */
2128 void cleanup_socket_xprt(void)
2129 {
2130 #ifdef RPC_DEBUG
2131         if (sunrpc_table_header) {
2132                 unregister_sysctl_table(sunrpc_table_header);
2133                 sunrpc_table_header = NULL;
2134         }
2135 #endif
2136
2137         xprt_unregister_transport(&xs_udp_transport);
2138         xprt_unregister_transport(&xs_tcp_transport);
2139 }