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, "%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         status = xs_sendpages(transport->sock,
 581                               xs_addr(xprt),
 582                               xprt->addrlen, xdr,
 583                               req->rq_bytes_sent);
 584
 585         dprintk("RPC:       xs_udp_send_request(%u) = %d\n",
 586                         xdr->len - req->rq_bytes_sent, status);
 587
 588         if (status >= 0) {
 589                 task->tk_bytes_sent += status;
 590                 if (status >= req->rq_slen)
 591                         return 0;
 592                 /* Still some bytes left; set up for a retry later. */
 593                 status = -EAGAIN;
 594         }
 595
 596         switch (status) {
 597         case -ENOTSOCK:
 598                 status = -ENOTCONN;
 599                 /* Should we call xs_close() here? */
 600                 break;
 601         case -EAGAIN:
 602                 xs_nospace(task);
 603                 break;
 604         case -ENETUNREACH:
 605         case -EPIPE:
 606         case -ECONNREFUSED:
 607                 /* When the server has died, an ICMP port unreachable message
 608                  * prompts ECONNREFUSED. */
 609                 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 610                 break;
 611         default:
 612                 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 613                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
 614                         -status);
 615         }
 616
 617         return status;
 618 }
 619
 620 /**
 621  * xs_tcp_shutdown - gracefully shut down a TCP socket
 622  * @xprt: transport
 623  *
 624  * Initiates a graceful shutdown of the TCP socket by calling the
 625  * equivalent of shutdown(SHUT_WR);
 626  */
 627 static void xs_tcp_shutdown(struct rpc_xprt *xprt)
 628 {
 629         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 630         struct socket *sock = transport->sock;
 631
 632         if (sock != NULL)
 633                 kernel_sock_shutdown(sock, SHUT_WR);
 634 }
 635
 636 static inline void xs_encode_tcp_record_marker(struct xdr_buf *buf)
 637 {
 638         u32 reclen = buf->len - sizeof(rpc_fraghdr);
 639         rpc_fraghdr *base = buf->head[0].iov_base;
 640         *base = htonl(RPC_LAST_STREAM_FRAGMENT | reclen);
 641 }
 642
 643 /**
 644  * xs_tcp_send_request - write an RPC request to a TCP socket
 645  * @task: address of RPC task that manages the state of an RPC request
 646  *
 647  * Return values:
 648  *        0:    The request has been sent
 649  *   EAGAIN:    The socket was blocked, please call again later to
 650  *              complete the request
 651  * ENOTCONN:    Caller needs to invoke connect logic then call again
 652  *    other:    Some other error occured, the request was not sent
 653  *
 654  * XXX: In the case of soft timeouts, should we eventually give up
 655  *      if sendmsg is not able to make progress?
 656  */
 657 static int xs_tcp_send_request(struct rpc_task *task)
 658 {
 659         struct rpc_rqst *req = task->tk_rqstp;
 660         struct rpc_xprt *xprt = req->rq_xprt;
 661         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 662         struct xdr_buf *xdr = &req->rq_snd_buf;
 663         int status;
 664
 665         xs_encode_tcp_record_marker(&req->rq_snd_buf);
 666
 667         xs_pktdump("packet data:",
 668                                 req->rq_svec->iov_base,
 669                                 req->rq_svec->iov_len);
 670
 671         /* Continue transmitting the packet/record. We must be careful
 672          * to cope with writespace callbacks arriving _after_ we have
 673          * called sendmsg(). */
 674         while (1) {
 675                 status = xs_sendpages(transport->sock,
 676                                         NULL, 0, xdr, req->rq_bytes_sent);
 677
 678                 dprintk("RPC:       xs_tcp_send_request(%u) = %d\n",
 679                                 xdr->len - req->rq_bytes_sent, status);
 680
 681                 if (unlikely(status < 0))
 682                         break;
 683
 684                 /* If we've sent the entire packet, immediately
 685                  * reset the count of bytes sent. */
 686                 req->rq_bytes_sent += status;
 687                 task->tk_bytes_sent += status;
 688                 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
 689                         req->rq_bytes_sent = 0;
 690                         return 0;
 691                 }
 692
 693                 if (status != 0)
 694                         continue;
 695                 status = -EAGAIN;
 696                 break;
 697         }
 698
 699         switch (status) {
 700         case -ENOTSOCK:
 701                 status = -ENOTCONN;
 702                 /* Should we call xs_close() here? */
 703                 break;
 704         case -EAGAIN:
 705                 xs_nospace(task);
 706                 break;
 707         case -ECONNRESET:
 708                 xs_tcp_shutdown(xprt);
 709         case -ECONNREFUSED:
 710         case -ENOTCONN:
 711         case -EPIPE:
 712                 status = -ENOTCONN;
 713                 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 714                 break;
 715         default:
 716                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
 717                         -status);
 718                 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 719                 xs_tcp_shutdown(xprt);
 720         }
 721
 722         return status;
 723 }
 724
 725 /**
 726  * xs_tcp_release_xprt - clean up after a tcp transmission
 727  * @xprt: transport
 728  * @task: rpc task
 729  *
 730  * This cleans up if an error causes us to abort the transmission of a request.
 731  * In this case, the socket may need to be reset in order to avoid confusing
 732  * the server.
 733  */
 734 static void xs_tcp_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
 735 {
 736         struct rpc_rqst *req;
 737
 738         if (task != xprt->snd_task)
 739                 return;
 740         if (task == NULL)
 741                 goto out_release;
 742         req = task->tk_rqstp;
 743         if (req->rq_bytes_sent == 0)
 744                 goto out_release;
 745         if (req->rq_bytes_sent == req->rq_snd_buf.len)
 746                 goto out_release;
 747         set_bit(XPRT_CLOSE_WAIT, &task->tk_xprt->state);
 748 out_release:
 749         xprt_release_xprt(xprt, task);
 750 }
 751
 752 static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk)
 753 {
 754         transport->old_data_ready = sk->sk_data_ready;
 755         transport->old_state_change = sk->sk_state_change;
 756         transport->old_write_space = sk->sk_write_space;
 757         transport->old_error_report = sk->sk_error_report;
 758 }
 759
 760 static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk)
 761 {
 762         sk->sk_data_ready = transport->old_data_ready;
 763         sk->sk_state_change = transport->old_state_change;
 764         sk->sk_write_space = transport->old_write_space;
 765         sk->sk_error_report = transport->old_error_report;
 766 }
 767
 768 /**
 769  * xs_close - close a socket
 770  * @xprt: transport
 771  *
 772  * This is used when all requests are complete; ie, no DRC state remains
 773  * on the server we want to save.
 774  */
 775 static void xs_close(struct rpc_xprt *xprt)
 776 {
 777         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 778         struct socket *sock = transport->sock;
 779         struct sock *sk = transport->inet;
 780
 781         if (!sk)
 782                 goto clear_close_wait;
 783
 784         dprintk("RPC:       xs_close xprt %p\n", xprt);
 785
 786         write_lock_bh(&sk->sk_callback_lock);
 787         transport->inet = NULL;
 788         transport->sock = NULL;
 789
 790         sk->sk_user_data = NULL;
 791
 792         xs_restore_old_callbacks(transport, sk);
 793         write_unlock_bh(&sk->sk_callback_lock);
 794
 795         sk->sk_no_check = 0;
 796
 797         sock_release(sock);
 798 clear_close_wait:
 799         smp_mb__before_clear_bit();
 800         clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
 801         clear_bit(XPRT_CLOSING, &xprt->state);
 802         smp_mb__after_clear_bit();
 803         xprt_disconnect_done(xprt);
 804 }
 805
 806 /**
 807  * xs_destroy - prepare to shutdown a transport
 808  * @xprt: doomed transport
 809  *
 810  */
 811 static void xs_destroy(struct rpc_xprt *xprt)
 812 {
 813         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 814
 815         dprintk("RPC:       xs_destroy xprt %p\n", xprt);
 816
 817         cancel_rearming_delayed_work(&transport->connect_worker);
 818
 819         xs_close(xprt);
 820         xs_free_peer_addresses(xprt);
 821         kfree(xprt->slot);
 822         kfree(xprt);
 823         module_put(THIS_MODULE);
 824 }
 825
 826 static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
 827 {
 828         return (struct rpc_xprt *) sk->sk_user_data;
 829 }
 830
 831 /**
 832  * xs_udp_data_ready - "data ready" callback for UDP sockets
 833  * @sk: socket with data to read
 834  * @len: how much data to read
 835  *
 836  */
 837 static void xs_udp_data_ready(struct sock *sk, int len)
 838 {
 839         struct rpc_task *task;
 840         struct rpc_xprt *xprt;
 841         struct rpc_rqst *rovr;
 842         struct sk_buff *skb;
 843         int err, repsize, copied;
 844         u32 _xid;
 845         __be32 *xp;
 846
 847         read_lock(&sk->sk_callback_lock);
 848         dprintk("RPC:       xs_udp_data_ready...\n");
 849         if (!(xprt = xprt_from_sock(sk)))
 850                 goto out;
 851
 852         if ((skb = skb_recv_datagram(sk, 0, 1, &err)) == NULL)
 853                 goto out;
 854
 855         if (xprt->shutdown)
 856                 goto dropit;
 857
 858         repsize = skb->len - sizeof(struct udphdr);
 859         if (repsize < 4) {
 860                 dprintk("RPC:       impossible RPC reply size %d!\n", repsize);
 861                 goto dropit;
 862         }
 863
 864         /* Copy the XID from the skb... */
 865         xp = skb_header_pointer(skb, sizeof(struct udphdr),
 866                                 sizeof(_xid), &_xid);
 867         if (xp == NULL)
 868                 goto dropit;
 869
 870         /* Look up and lock the request corresponding to the given XID */
 871         spin_lock(&xprt->transport_lock);
 872         rovr = xprt_lookup_rqst(xprt, *xp);
 873         if (!rovr)
 874                 goto out_unlock;
 875         task = rovr->rq_task;
 876
 877         if ((copied = rovr->rq_private_buf.buflen) > repsize)
 878                 copied = repsize;
 879
 880         /* Suck it into the iovec, verify checksum if not done by hw. */
 881         if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
 882                 UDPX_INC_STATS_BH(sk, UDP_MIB_INERRORS);
 883                 goto out_unlock;
 884         }
 885
 886         UDPX_INC_STATS_BH(sk, UDP_MIB_INDATAGRAMS);
 887
 888         /* Something worked... */
 889         dst_confirm(skb->dst);
 890
 891         xprt_adjust_cwnd(task, copied);
 892         xprt_update_rtt(task);
 893         xprt_complete_rqst(task, copied);
 894
 895  out_unlock:
 896         spin_unlock(&xprt->transport_lock);
 897  dropit:
 898         skb_free_datagram(sk, skb);
 899  out:
 900         read_unlock(&sk->sk_callback_lock);
 901 }
 902
 903 static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
 904 {
 905         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 906         size_t len, used;
 907         char *p;
 908
 909         p = ((char *) &transport->tcp_fraghdr) + transport->tcp_offset;
 910         len = sizeof(transport->tcp_fraghdr) - transport->tcp_offset;
 911         used = xdr_skb_read_bits(desc, p, len);
 912         transport->tcp_offset += used;
 913         if (used != len)
 914                 return;
 915
 916         transport->tcp_reclen = ntohl(transport->tcp_fraghdr);
 917         if (transport->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
 918                 transport->tcp_flags |= TCP_RCV_LAST_FRAG;
 919         else
 920                 transport->tcp_flags &= ~TCP_RCV_LAST_FRAG;
 921         transport->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
 922
 923         transport->tcp_flags &= ~TCP_RCV_COPY_FRAGHDR;
 924         transport->tcp_offset = 0;
 925
 926         /* Sanity check of the record length */
 927         if (unlikely(transport->tcp_reclen < 4)) {
 928                 dprintk("RPC:       invalid TCP record fragment length\n");
 929                 xprt_force_disconnect(xprt);
 930                 return;
 931         }
 932         dprintk("RPC:       reading TCP record fragment of length %d\n",
 933                         transport->tcp_reclen);
 934 }
 935
 936 static void xs_tcp_check_fraghdr(struct sock_xprt *transport)
 937 {
 938         if (transport->tcp_offset == transport->tcp_reclen) {
 939                 transport->tcp_flags |= TCP_RCV_COPY_FRAGHDR;
 940                 transport->tcp_offset = 0;
 941                 if (transport->tcp_flags & TCP_RCV_LAST_FRAG) {
 942                         transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
 943                         transport->tcp_flags |= TCP_RCV_COPY_XID;
 944                         transport->tcp_copied = 0;
 945                 }
 946         }
 947 }
 948
 949 static inline void xs_tcp_read_xid(struct sock_xprt *transport, struct xdr_skb_reader *desc)
 950 {
 951         size_t len, used;
 952         char *p;
 953
 954         len = sizeof(transport->tcp_xid) - transport->tcp_offset;
 955         dprintk("RPC:       reading XID (%Zu bytes)\n", len);
 956         p = ((char *) &transport->tcp_xid) + transport->tcp_offset;
 957         used = xdr_skb_read_bits(desc, p, len);
 958         transport->tcp_offset += used;
 959         if (used != len)
 960                 return;
 961         transport->tcp_flags &= ~TCP_RCV_COPY_XID;
 962         transport->tcp_flags |= TCP_RCV_COPY_DATA;
 963         transport->tcp_copied = 4;
 964         dprintk("RPC:       reading reply for XID %08x\n",
 965                         ntohl(transport->tcp_xid));
 966         xs_tcp_check_fraghdr(transport);
 967 }
 968
 969 static inline void xs_tcp_read_request(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
 970 {
 971         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 972         struct rpc_rqst *req;
 973         struct xdr_buf *rcvbuf;
 974         size_t len;
 975         ssize_t r;
 976
 977         /* Find and lock the request corresponding to this xid */
 978         spin_lock(&xprt->transport_lock);
 979         req = xprt_lookup_rqst(xprt, transport->tcp_xid);
 980         if (!req) {
 981                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
 982                 dprintk("RPC:       XID %08x request not found!\n",
 983                                 ntohl(transport->tcp_xid));
 984                 spin_unlock(&xprt->transport_lock);
 985                 return;
 986         }
 987
 988         rcvbuf = &req->rq_private_buf;
 989         len = desc->count;
 990         if (len > transport->tcp_reclen - transport->tcp_offset) {
 991                 struct xdr_skb_reader my_desc;
 992
 993                 len = transport->tcp_reclen - transport->tcp_offset;
 994                 memcpy(&my_desc, desc, sizeof(my_desc));
 995                 my_desc.count = len;
 996                 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
 997                                           &my_desc, xdr_skb_read_bits);
 998                 desc->count -= r;
 999                 desc->offset += r;
1000         } else
1001                 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
1002                                           desc, xdr_skb_read_bits);
1003
1004         if (r > 0) {
1005                 transport->tcp_copied += r;
1006                 transport->tcp_offset += r;
1007         }
1008         if (r != len) {
1009                 /* Error when copying to the receive buffer,
1010                  * usually because we weren't able to allocate
1011                  * additional buffer pages. All we can do now
1012                  * is turn off TCP_RCV_COPY_DATA, so the request
1013                  * will not receive any additional updates,
1014                  * and time out.
1015                  * Any remaining data from this record will
1016                  * be discarded.
1017                  */
1018                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1019                 dprintk("RPC:       XID %08x truncated request\n",
1020                                 ntohl(transport->tcp_xid));
1021                 dprintk("RPC:       xprt = %p, tcp_copied = %lu, "
1022                                 "tcp_offset = %u, tcp_reclen = %u\n",
1023                                 xprt, transport->tcp_copied,
1024                                 transport->tcp_offset, transport->tcp_reclen);
1025                 goto out;
1026         }
1027
1028         dprintk("RPC:       XID %08x read %Zd bytes\n",
1029                         ntohl(transport->tcp_xid), r);
1030         dprintk("RPC:       xprt = %p, tcp_copied = %lu, tcp_offset = %u, "
1031                         "tcp_reclen = %u\n", xprt, transport->tcp_copied,
1032                         transport->tcp_offset, transport->tcp_reclen);
1033
1034         if (transport->tcp_copied == req->rq_private_buf.buflen)
1035                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1036         else if (transport->tcp_offset == transport->tcp_reclen) {
1037                 if (transport->tcp_flags & TCP_RCV_LAST_FRAG)
1038                         transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1039         }
1040
1041 out:
1042         if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1043                 xprt_complete_rqst(req->rq_task, transport->tcp_copied);
1044         spin_unlock(&xprt->transport_lock);
1045         xs_tcp_check_fraghdr(transport);
1046 }
1047
1048 static inline void xs_tcp_read_discard(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1049 {
1050         size_t len;
1051
1052         len = transport->tcp_reclen - transport->tcp_offset;
1053         if (len > desc->count)
1054                 len = desc->count;
1055         desc->count -= len;
1056         desc->offset += len;
1057         transport->tcp_offset += len;
1058         dprintk("RPC:       discarded %Zu bytes\n", len);
1059         xs_tcp_check_fraghdr(transport);
1060 }
1061
1062 static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
1063 {
1064         struct rpc_xprt *xprt = rd_desc->arg.data;
1065         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1066         struct xdr_skb_reader desc = {
1067                 .skb    = skb,
1068                 .offset = offset,
1069                 .count  = len,
1070         };
1071
1072         dprintk("RPC:       xs_tcp_data_recv started\n");
1073         do {
1074                 /* Read in a new fragment marker if necessary */
1075                 /* Can we ever really expect to get completely empty fragments? */
1076                 if (transport->tcp_flags & TCP_RCV_COPY_FRAGHDR) {
1077                         xs_tcp_read_fraghdr(xprt, &desc);
1078                         continue;
1079                 }
1080                 /* Read in the xid if necessary */
1081                 if (transport->tcp_flags & TCP_RCV_COPY_XID) {
1082                         xs_tcp_read_xid(transport, &desc);
1083                         continue;
1084                 }
1085                 /* Read in the request data */
1086                 if (transport->tcp_flags & TCP_RCV_COPY_DATA) {
1087                         xs_tcp_read_request(xprt, &desc);
1088                         continue;
1089                 }
1090                 /* Skip over any trailing bytes on short reads */
1091                 xs_tcp_read_discard(transport, &desc);
1092         } while (desc.count);
1093         dprintk("RPC:       xs_tcp_data_recv done\n");
1094         return len - desc.count;
1095 }
1096
1097 /**
1098  * xs_tcp_data_ready - "data ready" callback for TCP sockets
1099  * @sk: socket with data to read
1100  * @bytes: how much data to read
1101  *
1102  */
1103 static void xs_tcp_data_ready(struct sock *sk, int bytes)
1104 {
1105         struct rpc_xprt *xprt;
1106         read_descriptor_t rd_desc;
1107         int read;
1108
1109         dprintk("RPC:       xs_tcp_data_ready...\n");
1110
1111         read_lock(&sk->sk_callback_lock);
1112         if (!(xprt = xprt_from_sock(sk)))
1113                 goto out;
1114         if (xprt->shutdown)
1115                 goto out;
1116
1117         /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
1118         rd_desc.arg.data = xprt;
1119         do {
1120                 rd_desc.count = 65536;
1121                 read = tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
1122         } while (read > 0);
1123 out:
1124         read_unlock(&sk->sk_callback_lock);
1125 }
1126
1127 /**
1128  * xs_tcp_state_change - callback to handle TCP socket state changes
1129  * @sk: socket whose state has changed
1130  *
1131  */
1132 static void xs_tcp_state_change(struct sock *sk)
1133 {
1134         struct rpc_xprt *xprt;
1135
1136         read_lock(&sk->sk_callback_lock);
1137         if (!(xprt = xprt_from_sock(sk)))
1138                 goto out;
1139         dprintk("RPC:       xs_tcp_state_change client %p...\n", xprt);
1140         dprintk("RPC:       state %x conn %d dead %d zapped %d\n",
1141                         sk->sk_state, xprt_connected(xprt),
1142                         sock_flag(sk, SOCK_DEAD),
1143                         sock_flag(sk, SOCK_ZAPPED));
1144
1145         switch (sk->sk_state) {
1146         case TCP_ESTABLISHED:
1147                 spin_lock_bh(&xprt->transport_lock);
1148                 if (!xprt_test_and_set_connected(xprt)) {
1149                         struct sock_xprt *transport = container_of(xprt,
1150                                         struct sock_xprt, xprt);
1151
1152                         /* Reset TCP record info */
1153                         transport->tcp_offset = 0;
1154                         transport->tcp_reclen = 0;
1155                         transport->tcp_copied = 0;
1156                         transport->tcp_flags =
1157                                 TCP_RCV_COPY_FRAGHDR | TCP_RCV_COPY_XID;
1158
1159                         xprt_wake_pending_tasks(xprt, 0);
1160                 }
1161                 spin_unlock_bh(&xprt->transport_lock);
1162                 break;
1163         case TCP_FIN_WAIT1:
1164                 /* The client initiated a shutdown of the socket */
1165                 xprt->connect_cookie++;
1166                 xprt->reestablish_timeout = 0;
1167                 set_bit(XPRT_CLOSING, &xprt->state);
1168                 smp_mb__before_clear_bit();
1169                 clear_bit(XPRT_CONNECTED, &xprt->state);
1170                 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1171                 smp_mb__after_clear_bit();
1172                 break;
1173         case TCP_CLOSE_WAIT:
1174                 /* The server initiated a shutdown of the socket */
1175                 set_bit(XPRT_CLOSING, &xprt->state);
1176                 xprt_force_disconnect(xprt);
1177         case TCP_SYN_SENT:
1178                 xprt->connect_cookie++;
1179         case TCP_CLOSING:
1180                 /*
1181                  * If the server closed down the connection, make sure that
1182                  * we back off before reconnecting
1183                  */
1184                 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1185                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1186                 break;
1187         case TCP_LAST_ACK:
1188                 smp_mb__before_clear_bit();
1189                 clear_bit(XPRT_CONNECTED, &xprt->state);
1190                 smp_mb__after_clear_bit();
1191                 break;
1192         case TCP_CLOSE:
1193                 smp_mb__before_clear_bit();
1194                 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1195                 clear_bit(XPRT_CLOSING, &xprt->state);
1196                 smp_mb__after_clear_bit();
1197                 /* Mark transport as closed and wake up all pending tasks */
1198                 xprt_disconnect_done(xprt);
1199         }
1200  out:
1201         read_unlock(&sk->sk_callback_lock);
1202 }
1203
1204 /**
1205  * xs_tcp_error_report - callback mainly for catching RST events
1206  * @sk: socket
1207  */
1208 static void xs_tcp_error_report(struct sock *sk)
1209 {
1210         struct rpc_xprt *xprt;
1211
1212         read_lock(&sk->sk_callback_lock);
1213         if (sk->sk_err != ECONNRESET || sk->sk_state != TCP_ESTABLISHED)
1214                 goto out;
1215         if (!(xprt = xprt_from_sock(sk)))
1216                 goto out;
1217         dprintk("RPC:       %s client %p...\n"
1218                         "RPC:       error %d\n",
1219                         __func__, xprt, sk->sk_err);
1220
1221         xprt_force_disconnect(xprt);
1222 out:
1223         read_unlock(&sk->sk_callback_lock);
1224 }
1225
1226 /**
1227  * xs_udp_write_space - callback invoked when socket buffer space
1228  *                             becomes available
1229  * @sk: socket whose state has changed
1230  *
1231  * Called when more output buffer space is available for this socket.
1232  * We try not to wake our writers until they can make "significant"
1233  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1234  * with a bunch of small requests.
1235  */
1236 static void xs_udp_write_space(struct sock *sk)
1237 {
1238         read_lock(&sk->sk_callback_lock);
1239
1240         /* from net/core/sock.c:sock_def_write_space */
1241         if (sock_writeable(sk)) {
1242                 struct socket *sock;
1243                 struct rpc_xprt *xprt;
1244
1245                 if (unlikely(!(sock = sk->sk_socket)))
1246                         goto out;
1247                 clear_bit(SOCK_NOSPACE, &sock->flags);
1248
1249                 if (unlikely(!(xprt = xprt_from_sock(sk))))
1250                         goto out;
1251                 if (test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags) == 0)
1252                         goto out;
1253
1254                 xprt_write_space(xprt);
1255         }
1256
1257  out:
1258         read_unlock(&sk->sk_callback_lock);
1259 }
1260
1261 /**
1262  * xs_tcp_write_space - callback invoked when socket buffer space
1263  *                             becomes available
1264  * @sk: socket whose state has changed
1265  *
1266  * Called when more output buffer space is available for this socket.
1267  * We try not to wake our writers until they can make "significant"
1268  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1269  * with a bunch of small requests.
1270  */
1271 static void xs_tcp_write_space(struct sock *sk)
1272 {
1273         read_lock(&sk->sk_callback_lock);
1274
1275         /* from net/core/stream.c:sk_stream_write_space */
1276         if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk)) {
1277                 struct socket *sock;
1278                 struct rpc_xprt *xprt;
1279
1280                 if (unlikely(!(sock = sk->sk_socket)))
1281                         goto out;
1282                 clear_bit(SOCK_NOSPACE, &sock->flags);
1283
1284                 if (unlikely(!(xprt = xprt_from_sock(sk))))
1285                         goto out;
1286                 if (test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags) == 0)
1287                         goto out;
1288
1289                 xprt_write_space(xprt);
1290         }
1291
1292  out:
1293         read_unlock(&sk->sk_callback_lock);
1294 }
1295
1296 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
1297 {
1298         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1299         struct sock *sk = transport->inet;
1300
1301         if (transport->rcvsize) {
1302                 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1303                 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
1304         }
1305         if (transport->sndsize) {
1306                 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1307                 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1308                 sk->sk_write_space(sk);
1309         }
1310 }
1311
1312 /**
1313  * xs_udp_set_buffer_size - set send and receive limits
1314  * @xprt: generic transport
1315  * @sndsize: requested size of send buffer, in bytes
1316  * @rcvsize: requested size of receive buffer, in bytes
1317  *
1318  * Set socket send and receive buffer size limits.
1319  */
1320 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1321 {
1322         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1323
1324         transport->sndsize = 0;
1325         if (sndsize)
1326                 transport->sndsize = sndsize + 1024;
1327         transport->rcvsize = 0;
1328         if (rcvsize)
1329                 transport->rcvsize = rcvsize + 1024;
1330
1331         xs_udp_do_set_buffer_size(xprt);
1332 }
1333
1334 /**
1335  * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1336  * @task: task that timed out
1337  *
1338  * Adjust the congestion window after a retransmit timeout has occurred.
1339  */
1340 static void xs_udp_timer(struct rpc_task *task)
1341 {
1342         xprt_adjust_cwnd(task, -ETIMEDOUT);
1343 }
1344
1345 static unsigned short xs_get_random_port(void)
1346 {
1347         unsigned short range = xprt_max_resvport - xprt_min_resvport;
1348         unsigned short rand = (unsigned short) net_random() % range;
1349         return rand + xprt_min_resvport;
1350 }
1351
1352 /**
1353  * xs_set_port - reset the port number in the remote endpoint address
1354  * @xprt: generic transport
1355  * @port: new port number
1356  *
1357  */
1358 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1359 {
1360         struct sockaddr *addr = xs_addr(xprt);
1361
1362         dprintk("RPC:       setting port for xprt %p to %u\n", xprt, port);
1363
1364         switch (addr->sa_family) {
1365         case AF_INET:
1366                 ((struct sockaddr_in *)addr)->sin_port = htons(port);
1367                 break;
1368         case AF_INET6:
1369                 ((struct sockaddr_in6 *)addr)->sin6_port = htons(port);
1370                 break;
1371         default:
1372                 BUG();
1373         }
1374 }
1375
1376 static unsigned short xs_get_srcport(struct sock_xprt *transport, struct socket *sock)
1377 {
1378         unsigned short port = transport->port;
1379
1380         if (port == 0 && transport->xprt.resvport)
1381                 port = xs_get_random_port();
1382         return port;
1383 }
1384
1385 static unsigned short xs_next_srcport(struct sock_xprt *transport, struct socket *sock, unsigned short port)
1386 {
1387         if (transport->port != 0)
1388                 transport->port = 0;
1389         if (!transport->xprt.resvport)
1390                 return 0;
1391         if (port <= xprt_min_resvport || port > xprt_max_resvport)
1392                 return xprt_max_resvport;
1393         return --port;
1394 }
1395
1396 static int xs_bind4(struct sock_xprt *transport, struct socket *sock)
1397 {
1398         struct sockaddr_in myaddr = {
1399                 .sin_family = AF_INET,
1400         };
1401         struct sockaddr_in *sa;
1402         int err, nloop = 0;
1403         unsigned short port = xs_get_srcport(transport, sock);
1404         unsigned short last;
1405
1406         sa = (struct sockaddr_in *)&transport->addr;
1407         myaddr.sin_addr = sa->sin_addr;
1408         do {
1409                 myaddr.sin_port = htons(port);
1410                 err = kernel_bind(sock, (struct sockaddr *) &myaddr,
1411                                                 sizeof(myaddr));
1412                 if (port == 0)
1413                         break;
1414                 if (err == 0) {
1415                         transport->port = port;
1416                         break;
1417                 }
1418                 last = port;
1419                 port = xs_next_srcport(transport, sock, port);
1420                 if (port > last)
1421                         nloop++;
1422         } while (err == -EADDRINUSE && nloop != 2);
1423         dprintk("RPC:       %s %pI4:%u: %s (%d)\n",
1424                         __func__, &myaddr.sin_addr,
1425                         port, err ? "failed" : "ok", err);
1426         return err;
1427 }
1428
1429 static int xs_bind6(struct sock_xprt *transport, struct socket *sock)
1430 {
1431         struct sockaddr_in6 myaddr = {
1432                 .sin6_family = AF_INET6,
1433         };
1434         struct sockaddr_in6 *sa;
1435         int err, nloop = 0;
1436         unsigned short port = xs_get_srcport(transport, sock);
1437         unsigned short last;
1438
1439         sa = (struct sockaddr_in6 *)&transport->addr;
1440         myaddr.sin6_addr = sa->sin6_addr;
1441         do {
1442                 myaddr.sin6_port = htons(port);
1443                 err = kernel_bind(sock, (struct sockaddr *) &myaddr,
1444                                                 sizeof(myaddr));
1445                 if (port == 0)
1446                         break;
1447                 if (err == 0) {
1448                         transport->port = port;
1449                         break;
1450                 }
1451                 last = port;
1452                 port = xs_next_srcport(transport, sock, port);
1453                 if (port > last)
1454                         nloop++;
1455         } while (err == -EADDRINUSE && nloop != 2);
1456         dprintk("RPC:       xs_bind6 %pI6:%u: %s (%d)\n",
1457                 &myaddr.sin6_addr, port, err ? "failed" : "ok", err);
1458         return err;
1459 }
1460
1461 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1462 static struct lock_class_key xs_key[2];
1463 static struct lock_class_key xs_slock_key[2];
1464
1465 static inline void xs_reclassify_socket4(struct socket *sock)
1466 {
1467         struct sock *sk = sock->sk;
1468
1469         BUG_ON(sock_owned_by_user(sk));
1470         sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1471                 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1472 }
1473
1474 static inline void xs_reclassify_socket6(struct socket *sock)
1475 {
1476         struct sock *sk = sock->sk;
1477
1478         BUG_ON(sock_owned_by_user(sk));
1479         sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1480                 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
1481 }
1482 #else
1483 static inline void xs_reclassify_socket4(struct socket *sock)
1484 {
1485 }
1486
1487 static inline void xs_reclassify_socket6(struct socket *sock)
1488 {
1489 }
1490 #endif
1491
1492 static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
1493 {
1494         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1495
1496         if (!transport->inet) {
1497                 struct sock *sk = sock->sk;
1498
1499                 write_lock_bh(&sk->sk_callback_lock);
1500
1501                 xs_save_old_callbacks(transport, sk);
1502
1503                 sk->sk_user_data = xprt;
1504                 sk->sk_data_ready = xs_udp_data_ready;
1505                 sk->sk_write_space = xs_udp_write_space;
1506                 sk->sk_no_check = UDP_CSUM_NORCV;
1507                 sk->sk_allocation = GFP_ATOMIC;
1508
1509                 xprt_set_connected(xprt);
1510
1511                 /* Reset to new socket */
1512                 transport->sock = sock;
1513                 transport->inet = sk;
1514
1515                 write_unlock_bh(&sk->sk_callback_lock);
1516         }
1517         xs_udp_do_set_buffer_size(xprt);
1518 }
1519
1520 /**
1521  * xs_udp_connect_worker4 - set up a UDP socket
1522  * @work: RPC transport to connect
1523  *
1524  * Invoked by a work queue tasklet.
1525  */
1526 static void xs_udp_connect_worker4(struct work_struct *work)
1527 {
1528         struct sock_xprt *transport =
1529                 container_of(work, struct sock_xprt, connect_worker.work);
1530         struct rpc_xprt *xprt = &transport->xprt;
1531         struct socket *sock = transport->sock;
1532         int err, status = -EIO;
1533
1534         if (xprt->shutdown || !xprt_bound(xprt))
1535                 goto out;
1536
1537         /* Start by resetting any existing state */
1538         xs_close(xprt);
1539
1540         if ((err = sock_create_kern(PF_INET, SOCK_DGRAM, IPPROTO_UDP, &sock)) < 0) {
1541                 dprintk("RPC:       can't create UDP transport socket (%d).\n", -err);
1542                 goto out;
1543         }
1544         xs_reclassify_socket4(sock);
1545
1546         if (xs_bind4(transport, sock)) {
1547                 sock_release(sock);
1548                 goto out;
1549         }
1550
1551         dprintk("RPC:       worker connecting xprt %p to address: %s\n",
1552                         xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1553
1554         xs_udp_finish_connecting(xprt, sock);
1555         status = 0;
1556 out:
1557         xprt_wake_pending_tasks(xprt, status);
1558         xprt_clear_connecting(xprt);
1559 }
1560
1561 /**
1562  * xs_udp_connect_worker6 - set up a UDP socket
1563  * @work: RPC transport to connect
1564  *
1565  * Invoked by a work queue tasklet.
1566  */
1567 static void xs_udp_connect_worker6(struct work_struct *work)
1568 {
1569         struct sock_xprt *transport =
1570                 container_of(work, struct sock_xprt, connect_worker.work);
1571         struct rpc_xprt *xprt = &transport->xprt;
1572         struct socket *sock = transport->sock;
1573         int err, status = -EIO;
1574
1575         if (xprt->shutdown || !xprt_bound(xprt))
1576                 goto out;
1577
1578         /* Start by resetting any existing state */
1579         xs_close(xprt);
1580
1581         if ((err = sock_create_kern(PF_INET6, SOCK_DGRAM, IPPROTO_UDP, &sock)) < 0) {
1582                 dprintk("RPC:       can't create UDP transport socket (%d).\n", -err);
1583                 goto out;
1584         }
1585         xs_reclassify_socket6(sock);
1586
1587         if (xs_bind6(transport, sock) < 0) {
1588                 sock_release(sock);
1589                 goto out;
1590         }
1591
1592         dprintk("RPC:       worker connecting xprt %p to address: %s\n",
1593                         xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1594
1595         xs_udp_finish_connecting(xprt, sock);
1596         status = 0;
1597 out:
1598         xprt_wake_pending_tasks(xprt, status);
1599         xprt_clear_connecting(xprt);
1600 }
1601
1602 /*
1603  * We need to preserve the port number so the reply cache on the server can
1604  * find our cached RPC replies when we get around to reconnecting.
1605  */
1606 static void xs_tcp_reuse_connection(struct rpc_xprt *xprt)
1607 {
1608         int result;
1609         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1610         struct sockaddr any;
1611
1612         dprintk("RPC:       disconnecting xprt %p to reuse port\n", xprt);
1613
1614         /*
1615          * Disconnect the transport socket by doing a connect operation
1616          * with AF_UNSPEC.  This should return immediately...
1617          */
1618         memset(&any, 0, sizeof(any));
1619         any.sa_family = AF_UNSPEC;
1620         result = kernel_connect(transport->sock, &any, sizeof(any), 0);
1621         if (result)
1622                 dprintk("RPC:       AF_UNSPEC connect return code %d\n",
1623                                 result);
1624 }
1625
1626 static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
1627 {
1628         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1629
1630         if (!transport->inet) {
1631                 struct sock *sk = sock->sk;
1632
1633                 write_lock_bh(&sk->sk_callback_lock);
1634
1635                 xs_save_old_callbacks(transport, sk);
1636
1637                 sk->sk_user_data = xprt;
1638                 sk->sk_data_ready = xs_tcp_data_ready;
1639                 sk->sk_state_change = xs_tcp_state_change;
1640                 sk->sk_write_space = xs_tcp_write_space;
1641                 sk->sk_error_report = xs_tcp_error_report;
1642                 sk->sk_allocation = GFP_ATOMIC;
1643
1644                 /* socket options */
1645                 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
1646                 sock_reset_flag(sk, SOCK_LINGER);
1647                 tcp_sk(sk)->linger2 = 0;
1648                 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
1649
1650                 xprt_clear_connected(xprt);
1651
1652                 /* Reset to new socket */
1653                 transport->sock = sock;
1654                 transport->inet = sk;
1655
1656                 write_unlock_bh(&sk->sk_callback_lock);
1657         }
1658
1659         /* Tell the socket layer to start connecting... */
1660         xprt->stat.connect_count++;
1661         xprt->stat.connect_start = jiffies;
1662         return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
1663 }
1664
1665 /**
1666  * xs_tcp_connect_worker4 - connect a TCP socket to a remote endpoint
1667  * @work: RPC transport to connect
1668  *
1669  * Invoked by a work queue tasklet.
1670  */
1671 static void xs_tcp_connect_worker4(struct work_struct *work)
1672 {
1673         struct sock_xprt *transport =
1674                 container_of(work, struct sock_xprt, connect_worker.work);
1675         struct rpc_xprt *xprt = &transport->xprt;
1676         struct socket *sock = transport->sock;
1677         int err, status = -EIO;
1678
1679         if (xprt->shutdown || !xprt_bound(xprt))
1680                 goto out;
1681
1682         if (!sock) {
1683                 /* start from scratch */
1684                 if ((err = sock_create_kern(PF_INET, SOCK_STREAM, IPPROTO_TCP, &sock)) < 0) {
1685                         dprintk("RPC:       can't create TCP transport socket (%d).\n", -err);
1686                         goto out;
1687                 }
1688                 xs_reclassify_socket4(sock);
1689
1690                 if (xs_bind4(transport, sock) < 0) {
1691                         sock_release(sock);
1692                         goto out;
1693                 }
1694         } else
1695                 /* "close" the socket, preserving the local port */
1696                 xs_tcp_reuse_connection(xprt);
1697
1698         dprintk("RPC:       worker connecting xprt %p to address: %s\n",
1699                         xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1700
1701         status = xs_tcp_finish_connecting(xprt, sock);
1702         dprintk("RPC:       %p connect status %d connected %d sock state %d\n",
1703                         xprt, -status, xprt_connected(xprt),
1704                         sock->sk->sk_state);
1705         if (status < 0) {
1706                 switch (status) {
1707                         case -EINPROGRESS:
1708                         case -EALREADY:
1709                                 goto out_clear;
1710                         case -ECONNREFUSED:
1711                         case -ECONNRESET:
1712                                 /* retry with existing socket, after a delay */
1713                                 break;
1714                         default:
1715                                 /* get rid of existing socket, and retry */
1716                                 xs_tcp_shutdown(xprt);
1717                 }
1718         }
1719 out:
1720         xprt_wake_pending_tasks(xprt, status);
1721 out_clear:
1722         xprt_clear_connecting(xprt);
1723 }
1724
1725 /**
1726  * xs_tcp_connect_worker6 - connect a TCP socket to a remote endpoint
1727  * @work: RPC transport to connect
1728  *
1729  * Invoked by a work queue tasklet.
1730  */
1731 static void xs_tcp_connect_worker6(struct work_struct *work)
1732 {
1733         struct sock_xprt *transport =
1734                 container_of(work, struct sock_xprt, connect_worker.work);
1735         struct rpc_xprt *xprt = &transport->xprt;
1736         struct socket *sock = transport->sock;
1737         int err, status = -EIO;
1738
1739         if (xprt->shutdown || !xprt_bound(xprt))
1740                 goto out;
1741
1742         if (!sock) {
1743                 /* start from scratch */
1744                 if ((err = sock_create_kern(PF_INET6, SOCK_STREAM, IPPROTO_TCP, &sock)) < 0) {
1745                         dprintk("RPC:       can't create TCP transport socket (%d).\n", -err);
1746                         goto out;
1747                 }
1748                 xs_reclassify_socket6(sock);
1749
1750                 if (xs_bind6(transport, sock) < 0) {
1751                         sock_release(sock);
1752                         goto out;
1753                 }
1754         } else
1755                 /* "close" the socket, preserving the local port */
1756                 xs_tcp_reuse_connection(xprt);
1757
1758         dprintk("RPC:       worker connecting xprt %p to address: %s\n",
1759                         xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1760
1761         status = xs_tcp_finish_connecting(xprt, sock);
1762         dprintk("RPC:       %p connect status %d connected %d sock state %d\n",
1763                         xprt, -status, xprt_connected(xprt), sock->sk->sk_state);
1764         if (status < 0) {
1765                 switch (status) {
1766                         case -EINPROGRESS:
1767                         case -EALREADY:
1768                                 goto out_clear;
1769                         case -ECONNREFUSED:
1770                         case -ECONNRESET:
1771                                 /* retry with existing socket, after a delay */
1772                                 break;
1773                         default:
1774                                 /* get rid of existing socket, and retry */
1775                                 xs_tcp_shutdown(xprt);
1776                 }
1777         }
1778 out:
1779         xprt_wake_pending_tasks(xprt, status);
1780 out_clear:
1781         xprt_clear_connecting(xprt);
1782 }
1783
1784 /**
1785  * xs_connect - connect a socket to a remote endpoint
1786  * @task: address of RPC task that manages state of connect request
1787  *
1788  * TCP: If the remote end dropped the connection, delay reconnecting.
1789  *
1790  * UDP socket connects are synchronous, but we use a work queue anyway
1791  * to guarantee that even unprivileged user processes can set up a
1792  * socket on a privileged port.
1793  *
1794  * If a UDP socket connect fails, the delay behavior here prevents
1795  * retry floods (hard mounts).
1796  */
1797 static void xs_connect(struct rpc_task *task)
1798 {
1799         struct rpc_xprt *xprt = task->tk_xprt;
1800         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1801
1802         if (xprt_test_and_set_connecting(xprt))
1803                 return;
1804
1805         if (transport->sock != NULL) {
1806                 dprintk("RPC:       xs_connect delayed xprt %p for %lu "
1807                                 "seconds\n",
1808                                 xprt, xprt->reestablish_timeout / HZ);
1809                 queue_delayed_work(rpciod_workqueue,
1810                                    &transport->connect_worker,
1811                                    xprt->reestablish_timeout);
1812                 xprt->reestablish_timeout <<= 1;
1813                 if (xprt->reestablish_timeout > XS_TCP_MAX_REEST_TO)
1814                         xprt->reestablish_timeout = XS_TCP_MAX_REEST_TO;
1815         } else {
1816                 dprintk("RPC:       xs_connect scheduled xprt %p\n", xprt);
1817                 queue_delayed_work(rpciod_workqueue,
1818                                    &transport->connect_worker, 0);
1819         }
1820 }
1821
1822 static void xs_tcp_connect(struct rpc_task *task)
1823 {
1824         struct rpc_xprt *xprt = task->tk_xprt;
1825
1826         /* Initiate graceful shutdown of the socket if not already done */
1827         if (test_bit(XPRT_CONNECTED, &xprt->state))
1828                 xs_tcp_shutdown(xprt);
1829         /* Exit if we need to wait for socket shutdown to complete */
1830         if (test_bit(XPRT_CLOSING, &xprt->state))
1831                 return;
1832         xs_connect(task);
1833 }
1834
1835 /**
1836  * xs_udp_print_stats - display UDP socket-specifc stats
1837  * @xprt: rpc_xprt struct containing statistics
1838  * @seq: output file
1839  *
1840  */
1841 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
1842 {
1843         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1844
1845         seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %Lu %Lu\n",
1846                         transport->port,
1847                         xprt->stat.bind_count,
1848                         xprt->stat.sends,
1849                         xprt->stat.recvs,
1850                         xprt->stat.bad_xids,
1851                         xprt->stat.req_u,
1852                         xprt->stat.bklog_u);
1853 }
1854
1855 /**
1856  * xs_tcp_print_stats - display TCP socket-specifc stats
1857  * @xprt: rpc_xprt struct containing statistics
1858  * @seq: output file
1859  *
1860  */
1861 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
1862 {
1863         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1864         long idle_time = 0;
1865
1866         if (xprt_connected(xprt))
1867                 idle_time = (long)(jiffies - xprt->last_used) / HZ;
1868
1869         seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu %Lu %Lu\n",
1870                         transport->port,
1871                         xprt->stat.bind_count,
1872                         xprt->stat.connect_count,
1873                         xprt->stat.connect_time,
1874                         idle_time,
1875                         xprt->stat.sends,
1876                         xprt->stat.recvs,
1877                         xprt->stat.bad_xids,
1878                         xprt->stat.req_u,
1879                         xprt->stat.bklog_u);
1880 }
1881
1882 static struct rpc_xprt_ops xs_udp_ops = {
1883         .set_buffer_size        = xs_udp_set_buffer_size,
1884         .reserve_xprt           = xprt_reserve_xprt_cong,
1885         .release_xprt           = xprt_release_xprt_cong,
1886         .rpcbind                = rpcb_getport_async,
1887         .set_port               = xs_set_port,
1888         .connect                = xs_connect,
1889         .buf_alloc              = rpc_malloc,
1890         .buf_free               = rpc_free,
1891         .send_request           = xs_udp_send_request,
1892         .set_retrans_timeout    = xprt_set_retrans_timeout_rtt,
1893         .timer                  = xs_udp_timer,
1894         .release_request        = xprt_release_rqst_cong,
1895         .close                  = xs_close,
1896         .destroy                = xs_destroy,
1897         .print_stats            = xs_udp_print_stats,
1898 };
1899
1900 static struct rpc_xprt_ops xs_tcp_ops = {
1901         .reserve_xprt           = xprt_reserve_xprt,
1902         .release_xprt           = xs_tcp_release_xprt,
1903         .rpcbind                = rpcb_getport_async,
1904         .set_port               = xs_set_port,
1905         .connect                = xs_tcp_connect,
1906         .buf_alloc              = rpc_malloc,
1907         .buf_free               = rpc_free,
1908         .send_request           = xs_tcp_send_request,
1909         .set_retrans_timeout    = xprt_set_retrans_timeout_def,
1910         .close                  = xs_tcp_shutdown,
1911         .destroy                = xs_destroy,
1912         .print_stats            = xs_tcp_print_stats,
1913 };
1914
1915 static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
1916                                       unsigned int slot_table_size)
1917 {
1918         struct rpc_xprt *xprt;
1919         struct sock_xprt *new;
1920
1921         if (args->addrlen > sizeof(xprt->addr)) {
1922                 dprintk("RPC:       xs_setup_xprt: address too large\n");
1923                 return ERR_PTR(-EBADF);
1924         }
1925
1926         new = kzalloc(sizeof(*new), GFP_KERNEL);
1927         if (new == NULL) {
1928                 dprintk("RPC:       xs_setup_xprt: couldn't allocate "
1929                                 "rpc_xprt\n");
1930                 return ERR_PTR(-ENOMEM);
1931         }
1932         xprt = &new->xprt;
1933
1934         xprt->max_reqs = slot_table_size;
1935         xprt->slot = kcalloc(xprt->max_reqs, sizeof(struct rpc_rqst), GFP_KERNEL);
1936         if (xprt->slot == NULL) {
1937                 kfree(xprt);
1938                 dprintk("RPC:       xs_setup_xprt: couldn't allocate slot "
1939                                 "table\n");
1940                 return ERR_PTR(-ENOMEM);
1941         }
1942
1943         memcpy(&xprt->addr, args->dstaddr, args->addrlen);
1944         xprt->addrlen = args->addrlen;
1945         if (args->srcaddr)
1946                 memcpy(&new->addr, args->srcaddr, args->addrlen);
1947
1948         return xprt;
1949 }
1950
1951 static const struct rpc_timeout xs_udp_default_timeout = {
1952         .to_initval = 5 * HZ,
1953         .to_maxval = 30 * HZ,
1954         .to_increment = 5 * HZ,
1955         .to_retries = 5,
1956 };
1957
1958 /**
1959  * xs_setup_udp - Set up transport to use a UDP socket
1960  * @args: rpc transport creation arguments
1961  *
1962  */
1963 static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
1964 {
1965         struct sockaddr *addr = args->dstaddr;
1966         struct rpc_xprt *xprt;
1967         struct sock_xprt *transport;
1968
1969         xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries);
1970         if (IS_ERR(xprt))
1971                 return xprt;
1972         transport = container_of(xprt, struct sock_xprt, xprt);
1973
1974         xprt->prot = IPPROTO_UDP;
1975         xprt->tsh_size = 0;
1976         /* XXX: header size can vary due to auth type, IPv6, etc. */
1977         xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
1978
1979         xprt->bind_timeout = XS_BIND_TO;
1980         xprt->connect_timeout = XS_UDP_CONN_TO;
1981         xprt->reestablish_timeout = XS_UDP_REEST_TO;
1982         xprt->idle_timeout = XS_IDLE_DISC_TO;
1983
1984         xprt->ops = &xs_udp_ops;
1985
1986         xprt->timeout = &xs_udp_default_timeout;
1987
1988         switch (addr->sa_family) {
1989         case AF_INET:
1990                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
1991                         xprt_set_bound(xprt);
1992
1993                 INIT_DELAYED_WORK(&transport->connect_worker,
1994                                         xs_udp_connect_worker4);
1995                 xs_format_ipv4_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
1996                 break;
1997         case AF_INET6:
1998                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
1999                         xprt_set_bound(xprt);
2000
2001                 INIT_DELAYED_WORK(&transport->connect_worker,
2002                                         xs_udp_connect_worker6);
2003                 xs_format_ipv6_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
2004                 break;
2005         default:
2006                 kfree(xprt);
2007                 return ERR_PTR(-EAFNOSUPPORT);
2008         }
2009
2010         dprintk("RPC:       set up transport to address %s\n",
2011                         xprt->address_strings[RPC_DISPLAY_ALL]);
2012
2013         if (try_module_get(THIS_MODULE))
2014                 return xprt;
2015
2016         kfree(xprt->slot);
2017         kfree(xprt);
2018         return ERR_PTR(-EINVAL);
2019 }
2020
2021 static const struct rpc_timeout xs_tcp_default_timeout = {
2022         .to_initval = 60 * HZ,
2023         .to_maxval = 60 * HZ,
2024         .to_retries = 2,
2025 };
2026
2027 /**
2028  * xs_setup_tcp - Set up transport to use a TCP socket
2029  * @args: rpc transport creation arguments
2030  *
2031  */
2032 static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
2033 {
2034         struct sockaddr *addr = args->dstaddr;
2035         struct rpc_xprt *xprt;
2036         struct sock_xprt *transport;
2037
2038         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries);
2039         if (IS_ERR(xprt))
2040                 return xprt;
2041         transport = container_of(xprt, struct sock_xprt, xprt);
2042
2043         xprt->prot = IPPROTO_TCP;
2044         xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2045         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2046
2047         xprt->bind_timeout = XS_BIND_TO;
2048         xprt->connect_timeout = XS_TCP_CONN_TO;
2049         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2050         xprt->idle_timeout = XS_IDLE_DISC_TO;
2051
2052         xprt->ops = &xs_tcp_ops;
2053         xprt->timeout = &xs_tcp_default_timeout;
2054
2055         switch (addr->sa_family) {
2056         case AF_INET:
2057                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2058                         xprt_set_bound(xprt);
2059
2060                 INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_connect_worker4);
2061                 xs_format_ipv4_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
2062                 break;
2063         case AF_INET6:
2064                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2065                         xprt_set_bound(xprt);
2066
2067                 INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_connect_worker6);
2068                 xs_format_ipv6_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
2069                 break;
2070         default:
2071                 kfree(xprt);
2072                 return ERR_PTR(-EAFNOSUPPORT);
2073         }
2074
2075         dprintk("RPC:       set up transport to address %s\n",
2076                         xprt->address_strings[RPC_DISPLAY_ALL]);
2077
2078         if (try_module_get(THIS_MODULE))
2079                 return xprt;
2080
2081         kfree(xprt->slot);
2082         kfree(xprt);
2083         return ERR_PTR(-EINVAL);
2084 }
2085
2086 static struct xprt_class        xs_udp_transport = {
2087         .list           = LIST_HEAD_INIT(xs_udp_transport.list),
2088         .name           = "udp",
2089         .owner          = THIS_MODULE,
2090         .ident          = IPPROTO_UDP,
2091         .setup          = xs_setup_udp,
2092 };
2093
2094 static struct xprt_class        xs_tcp_transport = {
2095         .list           = LIST_HEAD_INIT(xs_tcp_transport.list),
2096         .name           = "tcp",
2097         .owner          = THIS_MODULE,
2098         .ident          = IPPROTO_TCP,
2099         .setup          = xs_setup_tcp,
2100 };
2101
2102 /**
2103  * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
2104  *
2105  */
2106 int init_socket_xprt(void)
2107 {
2108 #ifdef RPC_DEBUG
2109         if (!sunrpc_table_header)
2110                 sunrpc_table_header = register_sysctl_table(sunrpc_table);
2111 #endif
2112
2113         xprt_register_transport(&xs_udp_transport);
2114         xprt_register_transport(&xs_tcp_transport);
2115
2116         return 0;
2117 }
2118
2119 /**
2120  * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
2121  *
2122  */
2123 void cleanup_socket_xprt(void)
2124 {
2125 #ifdef RPC_DEBUG
2126         if (sunrpc_table_header) {
2127                 unregister_sysctl_table(sunrpc_table_header);
2128                 sunrpc_table_header = NULL;
2129         }
2130 #endif
2131
2132         xprt_unregister_transport(&xs_udp_transport);
2133         xprt_unregister_transport(&xs_tcp_transport);
2134 }