SAFE public projects git trees. - safe/jmp/linux-2.6/blob - net/sunrpc/xprtsock.c

   1 /*
   2  * linux/net/sunrpc/xprtsock.c
   3  *
   4  * Client-side transport implementation for sockets.
   5  *
   6  * TCP callback races fixes (C) 1998 Red Hat
   7  * TCP send fixes (C) 1998 Red Hat
   8  * TCP NFS related read + write fixes
   9  *  (C) 1999 Dave Airlie, University of Limerick, Ireland <airlied@linux.ie>
  10  *
  11  * Rewrite of larges part of the code in order to stabilize TCP stuff.
  12  * Fix behaviour when socket buffer is full.
  13  *  (C) 1999 Trond Myklebust <trond.myklebust@fys.uio.no>
  14  *
  15  * IP socket transport implementation, (C) 2005 Chuck Lever <cel@netapp.com>
  16  *
  17  * IPv6 support contributed by Gilles Quillard, Bull Open Source, 2005.
  18  *   <gilles.quillard@bull.net>
  19  */
  20
  21 #include <linux/types.h>
  22 #include <linux/slab.h>
  23 #include <linux/module.h>
  24 #include <linux/capability.h>
  25 #include <linux/pagemap.h>
  26 #include <linux/errno.h>
  27 #include <linux/socket.h>
  28 #include <linux/in.h>
  29 #include <linux/net.h>
  30 #include <linux/mm.h>
  31 #include <linux/udp.h>
  32 #include <linux/tcp.h>
  33 #include <linux/sunrpc/clnt.h>
  34 #include <linux/sunrpc/sched.h>
  35 #include <linux/sunrpc/xprtsock.h>
  36 #include <linux/file.h>
  37
  38 #include <net/sock.h>
  39 #include <net/checksum.h>
  40 #include <net/udp.h>
  41 #include <net/tcp.h>
  42
  43 /*
  44  * xprtsock tunables
  45  */
  46 unsigned int xprt_udp_slot_table_entries = RPC_DEF_SLOT_TABLE;
  47 unsigned int xprt_tcp_slot_table_entries = RPC_DEF_SLOT_TABLE;
  48
  49 unsigned int xprt_min_resvport = RPC_DEF_MIN_RESVPORT;
  50 unsigned int xprt_max_resvport = RPC_DEF_MAX_RESVPORT;
  51
  52 /*
  53  * We can register our own files under /proc/sys/sunrpc by
  54  * calling register_sysctl_table() again.  The files in that
  55  * directory become the union of all files registered there.
  56  *
  57  * We simply need to make sure that we don't collide with
  58  * someone else's file names!
  59  */
  60
  61 #ifdef RPC_DEBUG
  62
  63 static unsigned int min_slot_table_size = RPC_MIN_SLOT_TABLE;
  64 static unsigned int max_slot_table_size = RPC_MAX_SLOT_TABLE;
  65 static unsigned int xprt_min_resvport_limit = RPC_MIN_RESVPORT;
  66 static unsigned int xprt_max_resvport_limit = RPC_MAX_RESVPORT;
  67
  68 static struct ctl_table_header *sunrpc_table_header;
  69
  70 /*
  71  * FIXME: changing the UDP slot table size should also resize the UDP
  72  *        socket buffers for existing UDP transports
  73  */
  74 static ctl_table xs_tunables_table[] = {
  75         {
  76                 .ctl_name       = CTL_SLOTTABLE_UDP,
  77                 .procname       = "udp_slot_table_entries",
  78                 .data           = &xprt_udp_slot_table_entries,
  79                 .maxlen         = sizeof(unsigned int),
  80                 .mode           = 0644,
  81                 .proc_handler   = &proc_dointvec_minmax,
  82                 .strategy       = &sysctl_intvec,
  83                 .extra1         = &min_slot_table_size,
  84                 .extra2         = &max_slot_table_size
  85         },
  86         {
  87                 .ctl_name       = CTL_SLOTTABLE_TCP,
  88                 .procname       = "tcp_slot_table_entries",
  89                 .data           = &xprt_tcp_slot_table_entries,
  90                 .maxlen         = sizeof(unsigned int),
  91                 .mode           = 0644,
  92                 .proc_handler   = &proc_dointvec_minmax,
  93                 .strategy       = &sysctl_intvec,
  94                 .extra1         = &min_slot_table_size,
  95                 .extra2         = &max_slot_table_size
  96         },
  97         {
  98                 .ctl_name       = CTL_MIN_RESVPORT,
  99                 .procname       = "min_resvport",
 100                 .data           = &xprt_min_resvport,
 101                 .maxlen         = sizeof(unsigned int),
 102                 .mode           = 0644,
 103                 .proc_handler   = &proc_dointvec_minmax,
 104                 .strategy       = &sysctl_intvec,
 105                 .extra1         = &xprt_min_resvport_limit,
 106                 .extra2         = &xprt_max_resvport_limit
 107         },
 108         {
 109                 .ctl_name       = CTL_MAX_RESVPORT,
 110                 .procname       = "max_resvport",
 111                 .data           = &xprt_max_resvport,
 112                 .maxlen         = sizeof(unsigned int),
 113                 .mode           = 0644,
 114                 .proc_handler   = &proc_dointvec_minmax,
 115                 .strategy       = &sysctl_intvec,
 116                 .extra1         = &xprt_min_resvport_limit,
 117                 .extra2         = &xprt_max_resvport_limit
 118         },
 119         {
 120                 .ctl_name = 0,
 121         },
 122 };
 123
 124 static ctl_table sunrpc_table[] = {
 125         {
 126                 .ctl_name       = CTL_SUNRPC,
 127                 .procname       = "sunrpc",
 128                 .mode           = 0555,
 129                 .child          = xs_tunables_table
 130         },
 131         {
 132                 .ctl_name = 0,
 133         },
 134 };
 135
 136 #endif
 137
 138 /*
 139  * Time out for an RPC UDP socket connect.  UDP socket connects are
 140  * synchronous, but we set a timeout anyway in case of resource
 141  * exhaustion on the local host.
 142  */
 143 #define XS_UDP_CONN_TO          (5U * HZ)
 144
 145 /*
 146  * Wait duration for an RPC TCP connection to be established.  Solaris
 147  * NFS over TCP uses 60 seconds, for example, which is in line with how
 148  * long a server takes to reboot.
 149  */
 150 #define XS_TCP_CONN_TO          (60U * HZ)
 151
 152 /*
 153  * Wait duration for a reply from the RPC portmapper.
 154  */
 155 #define XS_BIND_TO              (60U * HZ)
 156
 157 /*
 158  * Delay if a UDP socket connect error occurs.  This is most likely some
 159  * kind of resource problem on the local host.
 160  */
 161 #define XS_UDP_REEST_TO         (2U * HZ)
 162
 163 /*
 164  * The reestablish timeout allows clients to delay for a bit before attempting
 165  * to reconnect to a server that just dropped our connection.
 166  *
 167  * We implement an exponential backoff when trying to reestablish a TCP
 168  * transport connection with the server.  Some servers like to drop a TCP
 169  * connection when they are overworked, so we start with a short timeout and
 170  * increase over time if the server is down or not responding.
 171  */
 172 #define XS_TCP_INIT_REEST_TO    (3U * HZ)
 173 #define XS_TCP_MAX_REEST_TO     (5U * 60 * HZ)
 174
 175 /*
 176  * TCP idle timeout; client drops the transport socket if it is idle
 177  * for this long.  Note that we also timeout UDP sockets to prevent
 178  * holding port numbers when there is no RPC traffic.
 179  */
 180 #define XS_IDLE_DISC_TO         (5U * 60 * HZ)
 181
 182 #ifdef RPC_DEBUG
 183 # undef  RPC_DEBUG_DATA
 184 # define RPCDBG_FACILITY        RPCDBG_TRANS
 185 #endif
 186
 187 #ifdef RPC_DEBUG_DATA
 188 static void xs_pktdump(char *msg, u32 *packet, unsigned int count)
 189 {
 190         u8 *buf = (u8 *) packet;
 191         int j;
 192
 193         dprintk("RPC:       %s\n", msg);
 194         for (j = 0; j < count && j < 128; j += 4) {
 195                 if (!(j & 31)) {
 196                         if (j)
 197                                 dprintk("\n");
 198                         dprintk("0x%04x ", j);
 199                 }
 200                 dprintk("%02x%02x%02x%02x ",
 201                         buf[j], buf[j+1], buf[j+2], buf[j+3]);
 202         }
 203         dprintk("\n");
 204 }
 205 #else
 206 static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count)
 207 {
 208         /* NOP */
 209 }
 210 #endif
 211
 212 struct sock_xprt {
 213         struct rpc_xprt         xprt;
 214
 215         /*
 216          * Network layer
 217          */
 218         struct socket *         sock;
 219         struct sock *           inet;
 220
 221         /*
 222          * State of TCP reply receive
 223          */
 224         __be32                  tcp_fraghdr,
 225                                 tcp_xid;
 226
 227         u32                     tcp_offset,
 228                                 tcp_reclen;
 229
 230         unsigned long           tcp_copied,
 231                                 tcp_flags;
 232
 233         /*
 234          * Connection of transports
 235          */
 236         struct delayed_work     connect_worker;
 237         struct sockaddr_storage addr;
 238         unsigned short          port;
 239
 240         /*
 241          * UDP socket buffer size parameters
 242          */
 243         size_t                  rcvsize,
 244                                 sndsize;
 245
 246         /*
 247          * Saved socket callback addresses
 248          */
 249         void                    (*old_data_ready)(struct sock *, int);
 250         void                    (*old_state_change)(struct sock *);
 251         void                    (*old_write_space)(struct sock *);
 252         void                    (*old_error_report)(struct sock *);
 253 };
 254
 255 /*
 256  * TCP receive state flags
 257  */
 258 #define TCP_RCV_LAST_FRAG       (1UL << 0)
 259 #define TCP_RCV_COPY_FRAGHDR    (1UL << 1)
 260 #define TCP_RCV_COPY_XID        (1UL << 2)
 261 #define TCP_RCV_COPY_DATA       (1UL << 3)
 262
 263 static inline struct sockaddr *xs_addr(struct rpc_xprt *xprt)
 264 {
 265         return (struct sockaddr *) &xprt->addr;
 266 }
 267
 268 static inline struct sockaddr_in *xs_addr_in(struct rpc_xprt *xprt)
 269 {
 270         return (struct sockaddr_in *) &xprt->addr;
 271 }
 272
 273 static inline struct sockaddr_in6 *xs_addr_in6(struct rpc_xprt *xprt)
 274 {
 275         return (struct sockaddr_in6 *) &xprt->addr;
 276 }
 277
 278 static void xs_format_ipv4_peer_addresses(struct rpc_xprt *xprt,
 279                                           const char *protocol,
 280                                           const char *netid)
 281 {
 282         struct sockaddr_in *addr = xs_addr_in(xprt);
 283         char *buf;
 284
 285         buf = kzalloc(20, GFP_KERNEL);
 286         if (buf) {
 287                 snprintf(buf, 20, NIPQUAD_FMT,
 288                                 NIPQUAD(addr->sin_addr.s_addr));
 289         }
 290         xprt->address_strings[RPC_DISPLAY_ADDR] = buf;
 291
 292         buf = kzalloc(8, GFP_KERNEL);
 293         if (buf) {
 294                 snprintf(buf, 8, "%u",
 295                                 ntohs(addr->sin_port));
 296         }
 297         xprt->address_strings[RPC_DISPLAY_PORT] = buf;
 298
 299         xprt->address_strings[RPC_DISPLAY_PROTO] = protocol;
 300
 301         buf = kzalloc(48, GFP_KERNEL);
 302         if (buf) {
 303                 snprintf(buf, 48, "addr="NIPQUAD_FMT" port=%u proto=%s",
 304                         NIPQUAD(addr->sin_addr.s_addr),
 305                         ntohs(addr->sin_port),
 306                         protocol);
 307         }
 308         xprt->address_strings[RPC_DISPLAY_ALL] = buf;
 309
 310         buf = kzalloc(10, GFP_KERNEL);
 311         if (buf) {
 312                 snprintf(buf, 10, "%02x%02x%02x%02x",
 313                                 NIPQUAD(addr->sin_addr.s_addr));
 314         }
 315         xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = buf;
 316
 317         buf = kzalloc(8, GFP_KERNEL);
 318         if (buf) {
 319                 snprintf(buf, 8, "%4hx",
 320                                 ntohs(addr->sin_port));
 321         }
 322         xprt->address_strings[RPC_DISPLAY_HEX_PORT] = buf;
 323
 324         buf = kzalloc(30, GFP_KERNEL);
 325         if (buf) {
 326                 snprintf(buf, 30, NIPQUAD_FMT".%u.%u",
 327                                 NIPQUAD(addr->sin_addr.s_addr),
 328                                 ntohs(addr->sin_port) >> 8,
 329                                 ntohs(addr->sin_port) & 0xff);
 330         }
 331         xprt->address_strings[RPC_DISPLAY_UNIVERSAL_ADDR] = buf;
 332
 333         xprt->address_strings[RPC_DISPLAY_NETID] = netid;
 334 }
 335
 336 static void xs_format_ipv6_peer_addresses(struct rpc_xprt *xprt,
 337                                           const char *protocol,
 338                                           const char *netid)
 339 {
 340         struct sockaddr_in6 *addr = xs_addr_in6(xprt);
 341         char *buf;
 342
 343         buf = kzalloc(40, GFP_KERNEL);
 344         if (buf) {
 345                 snprintf(buf, 40, NIP6_FMT,
 346                                 NIP6(addr->sin6_addr));
 347         }
 348         xprt->address_strings[RPC_DISPLAY_ADDR] = buf;
 349
 350         buf = kzalloc(8, GFP_KERNEL);
 351         if (buf) {
 352                 snprintf(buf, 8, "%u",
 353                                 ntohs(addr->sin6_port));
 354         }
 355         xprt->address_strings[RPC_DISPLAY_PORT] = buf;
 356
 357         xprt->address_strings[RPC_DISPLAY_PROTO] = protocol;
 358
 359         buf = kzalloc(64, GFP_KERNEL);
 360         if (buf) {
 361                 snprintf(buf, 64, "addr="NIP6_FMT" port=%u proto=%s",
 362                                 NIP6(addr->sin6_addr),
 363                                 ntohs(addr->sin6_port),
 364                                 protocol);
 365         }
 366         xprt->address_strings[RPC_DISPLAY_ALL] = buf;
 367
 368         buf = kzalloc(36, GFP_KERNEL);
 369         if (buf) {
 370                 snprintf(buf, 36, NIP6_SEQFMT,
 371                                 NIP6(addr->sin6_addr));
 372         }
 373         xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = buf;
 374
 375         buf = kzalloc(8, GFP_KERNEL);
 376         if (buf) {
 377                 snprintf(buf, 8, "%4hx",
 378                                 ntohs(addr->sin6_port));
 379         }
 380         xprt->address_strings[RPC_DISPLAY_HEX_PORT] = buf;
 381
 382         buf = kzalloc(50, GFP_KERNEL);
 383         if (buf) {
 384                 snprintf(buf, 50, NIP6_FMT".%u.%u",
 385                                 NIP6(addr->sin6_addr),
 386                                 ntohs(addr->sin6_port) >> 8,
 387                                 ntohs(addr->sin6_port) & 0xff);
 388         }
 389         xprt->address_strings[RPC_DISPLAY_UNIVERSAL_ADDR] = buf;
 390
 391         xprt->address_strings[RPC_DISPLAY_NETID] = netid;
 392 }
 393
 394 static void xs_free_peer_addresses(struct rpc_xprt *xprt)
 395 {
 396         unsigned int i;
 397
 398         for (i = 0; i < RPC_DISPLAY_MAX; i++)
 399                 switch (i) {
 400                 case RPC_DISPLAY_PROTO:
 401                 case RPC_DISPLAY_NETID:
 402                         continue;
 403                 default:
 404                         kfree(xprt->address_strings[i]);
 405                 }
 406 }
 407
 408 #define XS_SENDMSG_FLAGS        (MSG_DONTWAIT | MSG_NOSIGNAL)
 409
 410 static int xs_send_kvec(struct socket *sock, struct sockaddr *addr, int addrlen, struct kvec *vec, unsigned int base, int more)
 411 {
 412         struct msghdr msg = {
 413                 .msg_name       = addr,
 414                 .msg_namelen    = addrlen,
 415                 .msg_flags      = XS_SENDMSG_FLAGS | (more ? MSG_MORE : 0),
 416         };
 417         struct kvec iov = {
 418                 .iov_base       = vec->iov_base + base,
 419                 .iov_len        = vec->iov_len - base,
 420         };
 421
 422         if (iov.iov_len != 0)
 423                 return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
 424         return kernel_sendmsg(sock, &msg, NULL, 0, 0);
 425 }
 426
 427 static int xs_send_pagedata(struct socket *sock, struct xdr_buf *xdr, unsigned int base, int more)
 428 {
 429         struct page **ppage;
 430         unsigned int remainder;
 431         int err, sent = 0;
 432
 433         remainder = xdr->page_len - base;
 434         base += xdr->page_base;
 435         ppage = xdr->pages + (base >> PAGE_SHIFT);
 436         base &= ~PAGE_MASK;
 437         for(;;) {
 438                 unsigned int len = min_t(unsigned int, PAGE_SIZE - base, remainder);
 439                 int flags = XS_SENDMSG_FLAGS;
 440
 441                 remainder -= len;
 442                 if (remainder != 0 || more)
 443                         flags |= MSG_MORE;
 444                 err = sock->ops->sendpage(sock, *ppage, base, len, flags);
 445                 if (remainder == 0 || err != len)
 446                         break;
 447                 sent += err;
 448                 ppage++;
 449                 base = 0;
 450         }
 451         if (sent == 0)
 452                 return err;
 453         if (err > 0)
 454                 sent += err;
 455         return sent;
 456 }
 457
 458 /**
 459  * xs_sendpages - write pages directly to a socket
 460  * @sock: socket to send on
 461  * @addr: UDP only -- address of destination
 462  * @addrlen: UDP only -- length of destination address
 463  * @xdr: buffer containing this request
 464  * @base: starting position in the buffer
 465  *
 466  */
 467 static int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base)
 468 {
 469         unsigned int remainder = xdr->len - base;
 470         int err, sent = 0;
 471
 472         if (unlikely(!sock))
 473                 return -ENOTCONN;
 474
 475         clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags);
 476         if (base != 0) {
 477                 addr = NULL;
 478                 addrlen = 0;
 479         }
 480
 481         if (base < xdr->head[0].iov_len || addr != NULL) {
 482                 unsigned int len = xdr->head[0].iov_len - base;
 483                 remainder -= len;
 484                 err = xs_send_kvec(sock, addr, addrlen, &xdr->head[0], base, remainder != 0);
 485                 if (remainder == 0 || err != len)
 486                         goto out;
 487                 sent += err;
 488                 base = 0;
 489         } else
 490                 base -= xdr->head[0].iov_len;
 491
 492         if (base < xdr->page_len) {
 493                 unsigned int len = xdr->page_len - base;
 494                 remainder -= len;
 495                 err = xs_send_pagedata(sock, xdr, base, remainder != 0);
 496                 if (remainder == 0 || err != len)
 497                         goto out;
 498                 sent += err;
 499                 base = 0;
 500         } else
 501                 base -= xdr->page_len;
 502
 503         if (base >= xdr->tail[0].iov_len)
 504                 return sent;
 505         err = xs_send_kvec(sock, NULL, 0, &xdr->tail[0], base, 0);
 506 out:
 507         if (sent == 0)
 508                 return err;
 509         if (err > 0)
 510                 sent += err;
 511         return sent;
 512 }
 513
 514 static void xs_nospace_callback(struct rpc_task *task)
 515 {
 516         struct sock_xprt *transport = container_of(task->tk_rqstp->rq_xprt, struct sock_xprt, xprt);
 517
 518         transport->inet->sk_write_pending--;
 519         clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 520 }
 521
 522 /**
 523  * xs_nospace - place task on wait queue if transmit was incomplete
 524  * @task: task to put to sleep
 525  *
 526  */
 527 static void xs_nospace(struct rpc_task *task)
 528 {
 529         struct rpc_rqst *req = task->tk_rqstp;
 530         struct rpc_xprt *xprt = req->rq_xprt;
 531         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 532
 533         dprintk("RPC: %5u xmit incomplete (%u left of %u)\n",
 534                         task->tk_pid, req->rq_slen - req->rq_bytes_sent,
 535                         req->rq_slen);
 536
 537         /* Protect against races with write_space */
 538         spin_lock_bh(&xprt->transport_lock);
 539
 540         /* Don't race with disconnect */
 541         if (xprt_connected(xprt)) {
 542                 if (test_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags)) {
 543                         /*
 544                          * Notify TCP that we're limited by the application
 545                          * window size
 546                          */
 547                         set_bit(SOCK_NOSPACE, &transport->sock->flags);
 548                         transport->inet->sk_write_pending++;
 549                         /* ...and wait for more buffer space */
 550                         xprt_wait_for_buffer_space(task, xs_nospace_callback);
 551                 }
 552         } else {
 553                 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 554                 task->tk_status = -ENOTCONN;
 555         }
 556
 557         spin_unlock_bh(&xprt->transport_lock);
 558 }
 559
 560 /**
 561  * xs_udp_send_request - write an RPC request to a UDP socket
 562  * @task: address of RPC task that manages the state of an RPC request
 563  *
 564  * Return values:
 565  *        0:    The request has been sent
 566  *   EAGAIN:    The socket was blocked, please call again later to
 567  *              complete the request
 568  * ENOTCONN:    Caller needs to invoke connect logic then call again
 569  *    other:    Some other error occured, the request was not sent
 570  */
 571 static int xs_udp_send_request(struct rpc_task *task)
 572 {
 573         struct rpc_rqst *req = task->tk_rqstp;
 574         struct rpc_xprt *xprt = req->rq_xprt;
 575         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 576         struct xdr_buf *xdr = &req->rq_snd_buf;
 577         int status;
 578
 579         xs_pktdump("packet data:",
 580                                 req->rq_svec->iov_base,
 581                                 req->rq_svec->iov_len);
 582
 583         status = xs_sendpages(transport->sock,
 584                               xs_addr(xprt),
 585                               xprt->addrlen, xdr,
 586                               req->rq_bytes_sent);
 587
 588         dprintk("RPC:       xs_udp_send_request(%u) = %d\n",
 589                         xdr->len - req->rq_bytes_sent, status);
 590
 591         if (status >= 0) {
 592                 task->tk_bytes_sent += status;
 593                 if (status >= req->rq_slen)
 594                         return 0;
 595                 /* Still some bytes left; set up for a retry later. */
 596                 status = -EAGAIN;
 597         }
 598
 599         switch (status) {
 600         case -EAGAIN:
 601                 xs_nospace(task);
 602                 break;
 603         case -ENETUNREACH:
 604         case -EPIPE:
 605         case -ECONNREFUSED:
 606                 /* When the server has died, an ICMP port unreachable message
 607                  * prompts ECONNREFUSED. */
 608                 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 609                 break;
 610         default:
 611                 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 612                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
 613                         -status);
 614         }
 615
 616         return status;
 617 }
 618
 619 /**
 620  * xs_tcp_shutdown - gracefully shut down a TCP socket
 621  * @xprt: transport
 622  *
 623  * Initiates a graceful shutdown of the TCP socket by calling the
 624  * equivalent of shutdown(SHUT_WR);
 625  */
 626 static void xs_tcp_shutdown(struct rpc_xprt *xprt)
 627 {
 628         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 629         struct socket *sock = transport->sock;
 630
 631         if (sock != NULL)
 632                 kernel_sock_shutdown(sock, SHUT_WR);
 633 }
 634
 635 static inline void xs_encode_tcp_record_marker(struct xdr_buf *buf)
 636 {
 637         u32 reclen = buf->len - sizeof(rpc_fraghdr);
 638         rpc_fraghdr *base = buf->head[0].iov_base;
 639         *base = htonl(RPC_LAST_STREAM_FRAGMENT | reclen);
 640 }
 641
 642 /**
 643  * xs_tcp_send_request - write an RPC request to a TCP socket
 644  * @task: address of RPC task that manages the state of an RPC request
 645  *
 646  * Return values:
 647  *        0:    The request has been sent
 648  *   EAGAIN:    The socket was blocked, please call again later to
 649  *              complete the request
 650  * ENOTCONN:    Caller needs to invoke connect logic then call again
 651  *    other:    Some other error occured, the request was not sent
 652  *
 653  * XXX: In the case of soft timeouts, should we eventually give up
 654  *      if sendmsg is not able to make progress?
 655  */
 656 static int xs_tcp_send_request(struct rpc_task *task)
 657 {
 658         struct rpc_rqst *req = task->tk_rqstp;
 659         struct rpc_xprt *xprt = req->rq_xprt;
 660         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 661         struct xdr_buf *xdr = &req->rq_snd_buf;
 662         int status;
 663
 664         xs_encode_tcp_record_marker(&req->rq_snd_buf);
 665
 666         xs_pktdump("packet data:",
 667                                 req->rq_svec->iov_base,
 668                                 req->rq_svec->iov_len);
 669
 670         /* Continue transmitting the packet/record. We must be careful
 671          * to cope with writespace callbacks arriving _after_ we have
 672          * called sendmsg(). */
 673         while (1) {
 674                 status = xs_sendpages(transport->sock,
 675                                         NULL, 0, xdr, req->rq_bytes_sent);
 676
 677                 dprintk("RPC:       xs_tcp_send_request(%u) = %d\n",
 678                                 xdr->len - req->rq_bytes_sent, status);
 679
 680                 if (unlikely(status < 0))
 681                         break;
 682
 683                 /* If we've sent the entire packet, immediately
 684                  * reset the count of bytes sent. */
 685                 req->rq_bytes_sent += status;
 686                 task->tk_bytes_sent += status;
 687                 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
 688                         req->rq_bytes_sent = 0;
 689                         return 0;
 690                 }
 691
 692                 if (status != 0)
 693                         continue;
 694                 status = -EAGAIN;
 695                 break;
 696         }
 697
 698         switch (status) {
 699         case -EAGAIN:
 700                 xs_nospace(task);
 701                 break;
 702         case -ECONNRESET:
 703                 xs_tcp_shutdown(xprt);
 704         case -ECONNREFUSED:
 705         case -ENOTCONN:
 706         case -EPIPE:
 707                 status = -ENOTCONN;
 708                 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 709                 break;
 710         default:
 711                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
 712                         -status);
 713                 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 714                 xs_tcp_shutdown(xprt);
 715         }
 716
 717         return status;
 718 }
 719
 720 /**
 721  * xs_tcp_release_xprt - clean up after a tcp transmission
 722  * @xprt: transport
 723  * @task: rpc task
 724  *
 725  * This cleans up if an error causes us to abort the transmission of a request.
 726  * In this case, the socket may need to be reset in order to avoid confusing
 727  * the server.
 728  */
 729 static void xs_tcp_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
 730 {
 731         struct rpc_rqst *req;
 732
 733         if (task != xprt->snd_task)
 734                 return;
 735         if (task == NULL)
 736                 goto out_release;
 737         req = task->tk_rqstp;
 738         if (req->rq_bytes_sent == 0)
 739                 goto out_release;
 740         if (req->rq_bytes_sent == req->rq_snd_buf.len)
 741                 goto out_release;
 742         set_bit(XPRT_CLOSE_WAIT, &task->tk_xprt->state);
 743 out_release:
 744         xprt_release_xprt(xprt, task);
 745 }
 746
 747 static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk)
 748 {
 749         transport->old_data_ready = sk->sk_data_ready;
 750         transport->old_state_change = sk->sk_state_change;
 751         transport->old_write_space = sk->sk_write_space;
 752         transport->old_error_report = sk->sk_error_report;
 753 }
 754
 755 static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk)
 756 {
 757         sk->sk_data_ready = transport->old_data_ready;
 758         sk->sk_state_change = transport->old_state_change;
 759         sk->sk_write_space = transport->old_write_space;
 760         sk->sk_error_report = transport->old_error_report;
 761 }
 762
 763 /**
 764  * xs_close - close a socket
 765  * @xprt: transport
 766  *
 767  * This is used when all requests are complete; ie, no DRC state remains
 768  * on the server we want to save.
 769  */
 770 static void xs_close(struct rpc_xprt *xprt)
 771 {
 772         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 773         struct socket *sock = transport->sock;
 774         struct sock *sk = transport->inet;
 775
 776         if (!sk)
 777                 goto clear_close_wait;
 778
 779         dprintk("RPC:       xs_close xprt %p\n", xprt);
 780
 781         write_lock_bh(&sk->sk_callback_lock);
 782         transport->inet = NULL;
 783         transport->sock = NULL;
 784
 785         sk->sk_user_data = NULL;
 786
 787         xs_restore_old_callbacks(transport, sk);
 788         write_unlock_bh(&sk->sk_callback_lock);
 789
 790         sk->sk_no_check = 0;
 791
 792         sock_release(sock);
 793 clear_close_wait:
 794         smp_mb__before_clear_bit();
 795         clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
 796         clear_bit(XPRT_CLOSING, &xprt->state);
 797         smp_mb__after_clear_bit();
 798         xprt_disconnect_done(xprt);
 799 }
 800
 801 /**
 802  * xs_destroy - prepare to shutdown a transport
 803  * @xprt: doomed transport
 804  *
 805  */
 806 static void xs_destroy(struct rpc_xprt *xprt)
 807 {
 808         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 809
 810         dprintk("RPC:       xs_destroy xprt %p\n", xprt);
 811
 812         cancel_rearming_delayed_work(&transport->connect_worker);
 813
 814         xs_close(xprt);
 815         xs_free_peer_addresses(xprt);
 816         kfree(xprt->slot);
 817         kfree(xprt);
 818         module_put(THIS_MODULE);
 819 }
 820
 821 static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
 822 {
 823         return (struct rpc_xprt *) sk->sk_user_data;
 824 }
 825
 826 /**
 827  * xs_udp_data_ready - "data ready" callback for UDP sockets
 828  * @sk: socket with data to read
 829  * @len: how much data to read
 830  *
 831  */
 832 static void xs_udp_data_ready(struct sock *sk, int len)
 833 {
 834         struct rpc_task *task;
 835         struct rpc_xprt *xprt;
 836         struct rpc_rqst *rovr;
 837         struct sk_buff *skb;
 838         int err, repsize, copied;
 839         u32 _xid;
 840         __be32 *xp;
 841
 842         read_lock(&sk->sk_callback_lock);
 843         dprintk("RPC:       xs_udp_data_ready...\n");
 844         if (!(xprt = xprt_from_sock(sk)))
 845                 goto out;
 846
 847         if ((skb = skb_recv_datagram(sk, 0, 1, &err)) == NULL)
 848                 goto out;
 849
 850         if (xprt->shutdown)
 851                 goto dropit;
 852
 853         repsize = skb->len - sizeof(struct udphdr);
 854         if (repsize < 4) {
 855                 dprintk("RPC:       impossible RPC reply size %d!\n", repsize);
 856                 goto dropit;
 857         }
 858
 859         /* Copy the XID from the skb... */
 860         xp = skb_header_pointer(skb, sizeof(struct udphdr),
 861                                 sizeof(_xid), &_xid);
 862         if (xp == NULL)
 863                 goto dropit;
 864
 865         /* Look up and lock the request corresponding to the given XID */
 866         spin_lock(&xprt->transport_lock);
 867         rovr = xprt_lookup_rqst(xprt, *xp);
 868         if (!rovr)
 869                 goto out_unlock;
 870         task = rovr->rq_task;
 871
 872         if ((copied = rovr->rq_private_buf.buflen) > repsize)
 873                 copied = repsize;
 874
 875         /* Suck it into the iovec, verify checksum if not done by hw. */
 876         if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
 877                 UDPX_INC_STATS_BH(sk, UDP_MIB_INERRORS);
 878                 goto out_unlock;
 879         }
 880
 881         UDPX_INC_STATS_BH(sk, UDP_MIB_INDATAGRAMS);
 882
 883         /* Something worked... */
 884         dst_confirm(skb->dst);
 885
 886         xprt_adjust_cwnd(task, copied);
 887         xprt_update_rtt(task);
 888         xprt_complete_rqst(task, copied);
 889
 890  out_unlock:
 891         spin_unlock(&xprt->transport_lock);
 892  dropit:
 893         skb_free_datagram(sk, skb);
 894  out:
 895         read_unlock(&sk->sk_callback_lock);
 896 }
 897
 898 static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
 899 {
 900         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 901         size_t len, used;
 902         char *p;
 903
 904         p = ((char *) &transport->tcp_fraghdr) + transport->tcp_offset;
 905         len = sizeof(transport->tcp_fraghdr) - transport->tcp_offset;
 906         used = xdr_skb_read_bits(desc, p, len);
 907         transport->tcp_offset += used;
 908         if (used != len)
 909                 return;
 910
 911         transport->tcp_reclen = ntohl(transport->tcp_fraghdr);
 912         if (transport->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
 913                 transport->tcp_flags |= TCP_RCV_LAST_FRAG;
 914         else
 915                 transport->tcp_flags &= ~TCP_RCV_LAST_FRAG;
 916         transport->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
 917
 918         transport->tcp_flags &= ~TCP_RCV_COPY_FRAGHDR;
 919         transport->tcp_offset = 0;
 920
 921         /* Sanity check of the record length */
 922         if (unlikely(transport->tcp_reclen < 4)) {
 923                 dprintk("RPC:       invalid TCP record fragment length\n");
 924                 xprt_force_disconnect(xprt);
 925                 return;
 926         }
 927         dprintk("RPC:       reading TCP record fragment of length %d\n",
 928                         transport->tcp_reclen);
 929 }
 930
 931 static void xs_tcp_check_fraghdr(struct sock_xprt *transport)
 932 {
 933         if (transport->tcp_offset == transport->tcp_reclen) {
 934                 transport->tcp_flags |= TCP_RCV_COPY_FRAGHDR;
 935                 transport->tcp_offset = 0;
 936                 if (transport->tcp_flags & TCP_RCV_LAST_FRAG) {
 937                         transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
 938                         transport->tcp_flags |= TCP_RCV_COPY_XID;
 939                         transport->tcp_copied = 0;
 940                 }
 941         }
 942 }
 943
 944 static inline void xs_tcp_read_xid(struct sock_xprt *transport, struct xdr_skb_reader *desc)
 945 {
 946         size_t len, used;
 947         char *p;
 948
 949         len = sizeof(transport->tcp_xid) - transport->tcp_offset;
 950         dprintk("RPC:       reading XID (%Zu bytes)\n", len);
 951         p = ((char *) &transport->tcp_xid) + transport->tcp_offset;
 952         used = xdr_skb_read_bits(desc, p, len);
 953         transport->tcp_offset += used;
 954         if (used != len)
 955                 return;
 956         transport->tcp_flags &= ~TCP_RCV_COPY_XID;
 957         transport->tcp_flags |= TCP_RCV_COPY_DATA;
 958         transport->tcp_copied = 4;
 959         dprintk("RPC:       reading reply for XID %08x\n",
 960                         ntohl(transport->tcp_xid));
 961         xs_tcp_check_fraghdr(transport);
 962 }
 963
 964 static inline void xs_tcp_read_request(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
 965 {
 966         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 967         struct rpc_rqst *req;
 968         struct xdr_buf *rcvbuf;
 969         size_t len;
 970         ssize_t r;
 971
 972         /* Find and lock the request corresponding to this xid */
 973         spin_lock(&xprt->transport_lock);
 974         req = xprt_lookup_rqst(xprt, transport->tcp_xid);
 975         if (!req) {
 976                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
 977                 dprintk("RPC:       XID %08x request not found!\n",
 978                                 ntohl(transport->tcp_xid));
 979                 spin_unlock(&xprt->transport_lock);
 980                 return;
 981         }
 982
 983         rcvbuf = &req->rq_private_buf;
 984         len = desc->count;
 985         if (len > transport->tcp_reclen - transport->tcp_offset) {
 986                 struct xdr_skb_reader my_desc;
 987
 988                 len = transport->tcp_reclen - transport->tcp_offset;
 989                 memcpy(&my_desc, desc, sizeof(my_desc));
 990                 my_desc.count = len;
 991                 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
 992                                           &my_desc, xdr_skb_read_bits);
 993                 desc->count -= r;
 994                 desc->offset += r;
 995         } else
 996                 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
 997                                           desc, xdr_skb_read_bits);
 998
 999         if (r > 0) {
1000                 transport->tcp_copied += r;
1001                 transport->tcp_offset += r;
1002         }
1003         if (r != len) {
1004                 /* Error when copying to the receive buffer,
1005                  * usually because we weren't able to allocate
1006                  * additional buffer pages. All we can do now
1007                  * is turn off TCP_RCV_COPY_DATA, so the request
1008                  * will not receive any additional updates,
1009                  * and time out.
1010                  * Any remaining data from this record will
1011                  * be discarded.
1012                  */
1013                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1014                 dprintk("RPC:       XID %08x truncated request\n",
1015                                 ntohl(transport->tcp_xid));
1016                 dprintk("RPC:       xprt = %p, tcp_copied = %lu, "
1017                                 "tcp_offset = %u, tcp_reclen = %u\n",
1018                                 xprt, transport->tcp_copied,
1019                                 transport->tcp_offset, transport->tcp_reclen);
1020                 goto out;
1021         }
1022
1023         dprintk("RPC:       XID %08x read %Zd bytes\n",
1024                         ntohl(transport->tcp_xid), r);
1025         dprintk("RPC:       xprt = %p, tcp_copied = %lu, tcp_offset = %u, "
1026                         "tcp_reclen = %u\n", xprt, transport->tcp_copied,
1027                         transport->tcp_offset, transport->tcp_reclen);
1028
1029         if (transport->tcp_copied == req->rq_private_buf.buflen)
1030                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1031         else if (transport->tcp_offset == transport->tcp_reclen) {
1032                 if (transport->tcp_flags & TCP_RCV_LAST_FRAG)
1033                         transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1034         }
1035
1036 out:
1037         if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1038                 xprt_complete_rqst(req->rq_task, transport->tcp_copied);
1039         spin_unlock(&xprt->transport_lock);
1040         xs_tcp_check_fraghdr(transport);
1041 }
1042
1043 static inline void xs_tcp_read_discard(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1044 {
1045         size_t len;
1046
1047         len = transport->tcp_reclen - transport->tcp_offset;
1048         if (len > desc->count)
1049                 len = desc->count;
1050         desc->count -= len;
1051         desc->offset += len;
1052         transport->tcp_offset += len;
1053         dprintk("RPC:       discarded %Zu bytes\n", len);
1054         xs_tcp_check_fraghdr(transport);
1055 }
1056
1057 static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
1058 {
1059         struct rpc_xprt *xprt = rd_desc->arg.data;
1060         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1061         struct xdr_skb_reader desc = {
1062                 .skb    = skb,
1063                 .offset = offset,
1064                 .count  = len,
1065         };
1066
1067         dprintk("RPC:       xs_tcp_data_recv started\n");
1068         do {
1069                 /* Read in a new fragment marker if necessary */
1070                 /* Can we ever really expect to get completely empty fragments? */
1071                 if (transport->tcp_flags & TCP_RCV_COPY_FRAGHDR) {
1072                         xs_tcp_read_fraghdr(xprt, &desc);
1073                         continue;
1074                 }
1075                 /* Read in the xid if necessary */
1076                 if (transport->tcp_flags & TCP_RCV_COPY_XID) {
1077                         xs_tcp_read_xid(transport, &desc);
1078                         continue;
1079                 }
1080                 /* Read in the request data */
1081                 if (transport->tcp_flags & TCP_RCV_COPY_DATA) {
1082                         xs_tcp_read_request(xprt, &desc);
1083                         continue;
1084                 }
1085                 /* Skip over any trailing bytes on short reads */
1086                 xs_tcp_read_discard(transport, &desc);
1087         } while (desc.count);
1088         dprintk("RPC:       xs_tcp_data_recv done\n");
1089         return len - desc.count;
1090 }
1091
1092 /**
1093  * xs_tcp_data_ready - "data ready" callback for TCP sockets
1094  * @sk: socket with data to read
1095  * @bytes: how much data to read
1096  *
1097  */
1098 static void xs_tcp_data_ready(struct sock *sk, int bytes)
1099 {
1100         struct rpc_xprt *xprt;
1101         read_descriptor_t rd_desc;
1102         int read;
1103
1104         dprintk("RPC:       xs_tcp_data_ready...\n");
1105
1106         read_lock(&sk->sk_callback_lock);
1107         if (!(xprt = xprt_from_sock(sk)))
1108                 goto out;
1109         if (xprt->shutdown)
1110                 goto out;
1111
1112         /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
1113         rd_desc.arg.data = xprt;
1114         do {
1115                 rd_desc.count = 65536;
1116                 read = tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
1117         } while (read > 0);
1118 out:
1119         read_unlock(&sk->sk_callback_lock);
1120 }
1121
1122 /**
1123  * xs_tcp_state_change - callback to handle TCP socket state changes
1124  * @sk: socket whose state has changed
1125  *
1126  */
1127 static void xs_tcp_state_change(struct sock *sk)
1128 {
1129         struct rpc_xprt *xprt;
1130
1131         read_lock(&sk->sk_callback_lock);
1132         if (!(xprt = xprt_from_sock(sk)))
1133                 goto out;
1134         dprintk("RPC:       xs_tcp_state_change client %p...\n", xprt);
1135         dprintk("RPC:       state %x conn %d dead %d zapped %d\n",
1136                         sk->sk_state, xprt_connected(xprt),
1137                         sock_flag(sk, SOCK_DEAD),
1138                         sock_flag(sk, SOCK_ZAPPED));
1139
1140         switch (sk->sk_state) {
1141         case TCP_ESTABLISHED:
1142                 spin_lock_bh(&xprt->transport_lock);
1143                 if (!xprt_test_and_set_connected(xprt)) {
1144                         struct sock_xprt *transport = container_of(xprt,
1145                                         struct sock_xprt, xprt);
1146
1147                         /* Reset TCP record info */
1148                         transport->tcp_offset = 0;
1149                         transport->tcp_reclen = 0;
1150                         transport->tcp_copied = 0;
1151                         transport->tcp_flags =
1152                                 TCP_RCV_COPY_FRAGHDR | TCP_RCV_COPY_XID;
1153
1154                         xprt_wake_pending_tasks(xprt, 0);
1155                 }
1156                 spin_unlock_bh(&xprt->transport_lock);
1157                 break;
1158         case TCP_FIN_WAIT1:
1159                 /* The client initiated a shutdown of the socket */
1160                 xprt->connect_cookie++;
1161                 xprt->reestablish_timeout = 0;
1162                 set_bit(XPRT_CLOSING, &xprt->state);
1163                 smp_mb__before_clear_bit();
1164                 clear_bit(XPRT_CONNECTED, &xprt->state);
1165                 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1166                 smp_mb__after_clear_bit();
1167                 break;
1168         case TCP_CLOSE_WAIT:
1169                 /* The server initiated a shutdown of the socket */
1170                 set_bit(XPRT_CLOSING, &xprt->state);
1171                 xprt_force_disconnect(xprt);
1172         case TCP_SYN_SENT:
1173                 xprt->connect_cookie++;
1174         case TCP_CLOSING:
1175                 /*
1176                  * If the server closed down the connection, make sure that
1177                  * we back off before reconnecting
1178                  */
1179                 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1180                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1181                 break;
1182         case TCP_LAST_ACK:
1183                 smp_mb__before_clear_bit();
1184                 clear_bit(XPRT_CONNECTED, &xprt->state);
1185                 smp_mb__after_clear_bit();
1186                 break;
1187         case TCP_CLOSE:
1188                 smp_mb__before_clear_bit();
1189                 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1190                 clear_bit(XPRT_CLOSING, &xprt->state);
1191                 smp_mb__after_clear_bit();
1192                 /* Mark transport as closed and wake up all pending tasks */
1193                 xprt_disconnect_done(xprt);
1194         }
1195  out:
1196         read_unlock(&sk->sk_callback_lock);
1197 }
1198
1199 /**
1200  * xs_tcp_error_report - callback mainly for catching RST events
1201  * @sk: socket
1202  */
1203 static void xs_tcp_error_report(struct sock *sk)
1204 {
1205         struct rpc_xprt *xprt;
1206
1207         read_lock(&sk->sk_callback_lock);
1208         if (sk->sk_err != ECONNRESET || sk->sk_state != TCP_ESTABLISHED)
1209                 goto out;
1210         if (!(xprt = xprt_from_sock(sk)))
1211                 goto out;
1212         dprintk("RPC:       %s client %p...\n"
1213                         "RPC:       error %d\n",
1214                         __func__, xprt, sk->sk_err);
1215
1216         xprt_force_disconnect(xprt);
1217 out:
1218         read_unlock(&sk->sk_callback_lock);
1219 }
1220
1221 /**
1222  * xs_udp_write_space - callback invoked when socket buffer space
1223  *                             becomes available
1224  * @sk: socket whose state has changed
1225  *
1226  * Called when more output buffer space is available for this socket.
1227  * We try not to wake our writers until they can make "significant"
1228  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1229  * with a bunch of small requests.
1230  */
1231 static void xs_udp_write_space(struct sock *sk)
1232 {
1233         read_lock(&sk->sk_callback_lock);
1234
1235         /* from net/core/sock.c:sock_def_write_space */
1236         if (sock_writeable(sk)) {
1237                 struct socket *sock;
1238                 struct rpc_xprt *xprt;
1239
1240                 if (unlikely(!(sock = sk->sk_socket)))
1241                         goto out;
1242                 clear_bit(SOCK_NOSPACE, &sock->flags);
1243
1244                 if (unlikely(!(xprt = xprt_from_sock(sk))))
1245                         goto out;
1246                 if (test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags) == 0)
1247                         goto out;
1248
1249                 xprt_write_space(xprt);
1250         }
1251
1252  out:
1253         read_unlock(&sk->sk_callback_lock);
1254 }
1255
1256 /**
1257  * xs_tcp_write_space - callback invoked when socket buffer space
1258  *                             becomes available
1259  * @sk: socket whose state has changed
1260  *
1261  * Called when more output buffer space is available for this socket.
1262  * We try not to wake our writers until they can make "significant"
1263  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1264  * with a bunch of small requests.
1265  */
1266 static void xs_tcp_write_space(struct sock *sk)
1267 {
1268         read_lock(&sk->sk_callback_lock);
1269
1270         /* from net/core/stream.c:sk_stream_write_space */
1271         if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk)) {
1272                 struct socket *sock;
1273                 struct rpc_xprt *xprt;
1274
1275                 if (unlikely(!(sock = sk->sk_socket)))
1276                         goto out;
1277                 clear_bit(SOCK_NOSPACE, &sock->flags);
1278
1279                 if (unlikely(!(xprt = xprt_from_sock(sk))))
1280                         goto out;
1281                 if (test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags) == 0)
1282                         goto out;
1283
1284                 xprt_write_space(xprt);
1285         }
1286
1287  out:
1288         read_unlock(&sk->sk_callback_lock);
1289 }
1290
1291 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
1292 {
1293         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1294         struct sock *sk = transport->inet;
1295
1296         if (transport->rcvsize) {
1297                 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1298                 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
1299         }
1300         if (transport->sndsize) {
1301                 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1302                 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1303                 sk->sk_write_space(sk);
1304         }
1305 }
1306
1307 /**
1308  * xs_udp_set_buffer_size - set send and receive limits
1309  * @xprt: generic transport
1310  * @sndsize: requested size of send buffer, in bytes
1311  * @rcvsize: requested size of receive buffer, in bytes
1312  *
1313  * Set socket send and receive buffer size limits.
1314  */
1315 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1316 {
1317         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1318
1319         transport->sndsize = 0;
1320         if (sndsize)
1321                 transport->sndsize = sndsize + 1024;
1322         transport->rcvsize = 0;
1323         if (rcvsize)
1324                 transport->rcvsize = rcvsize + 1024;
1325
1326         xs_udp_do_set_buffer_size(xprt);
1327 }
1328
1329 /**
1330  * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1331  * @task: task that timed out
1332  *
1333  * Adjust the congestion window after a retransmit timeout has occurred.
1334  */
1335 static void xs_udp_timer(struct rpc_task *task)
1336 {
1337         xprt_adjust_cwnd(task, -ETIMEDOUT);
1338 }
1339
1340 static unsigned short xs_get_random_port(void)
1341 {
1342         unsigned short range = xprt_max_resvport - xprt_min_resvport;
1343         unsigned short rand = (unsigned short) net_random() % range;
1344         return rand + xprt_min_resvport;
1345 }
1346
1347 /**
1348  * xs_set_port - reset the port number in the remote endpoint address
1349  * @xprt: generic transport
1350  * @port: new port number
1351  *
1352  */
1353 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1354 {
1355         struct sockaddr *addr = xs_addr(xprt);
1356
1357         dprintk("RPC:       setting port for xprt %p to %u\n", xprt, port);
1358
1359         switch (addr->sa_family) {
1360         case AF_INET:
1361                 ((struct sockaddr_in *)addr)->sin_port = htons(port);
1362                 break;
1363         case AF_INET6:
1364                 ((struct sockaddr_in6 *)addr)->sin6_port = htons(port);
1365                 break;
1366         default:
1367                 BUG();
1368         }
1369 }
1370
1371 static unsigned short xs_get_srcport(struct sock_xprt *transport, struct socket *sock)
1372 {
1373         unsigned short port = transport->port;
1374
1375         if (port == 0 && transport->xprt.resvport)
1376                 port = xs_get_random_port();
1377         return port;
1378 }
1379
1380 static unsigned short xs_next_srcport(struct sock_xprt *transport, struct socket *sock, unsigned short port)
1381 {
1382         if (transport->port != 0)
1383                 transport->port = 0;
1384         if (!transport->xprt.resvport)
1385                 return 0;
1386         if (port <= xprt_min_resvport || port > xprt_max_resvport)
1387                 return xprt_max_resvport;
1388         return --port;
1389 }
1390
1391 static int xs_bind4(struct sock_xprt *transport, struct socket *sock)
1392 {
1393         struct sockaddr_in myaddr = {
1394                 .sin_family = AF_INET,
1395         };
1396         struct sockaddr_in *sa;
1397         int err, nloop = 0;
1398         unsigned short port = xs_get_srcport(transport, sock);
1399         unsigned short last;
1400
1401         sa = (struct sockaddr_in *)&transport->addr;
1402         myaddr.sin_addr = sa->sin_addr;
1403         do {
1404                 myaddr.sin_port = htons(port);
1405                 err = kernel_bind(sock, (struct sockaddr *) &myaddr,
1406                                                 sizeof(myaddr));
1407                 if (port == 0)
1408                         break;
1409                 if (err == 0) {
1410                         transport->port = port;
1411                         break;
1412                 }
1413                 last = port;
1414                 port = xs_next_srcport(transport, sock, port);
1415                 if (port > last)
1416                         nloop++;
1417         } while (err == -EADDRINUSE && nloop != 2);
1418         dprintk("RPC:       %s "NIPQUAD_FMT":%u: %s (%d)\n",
1419                         __func__, NIPQUAD(myaddr.sin_addr),
1420                         port, err ? "failed" : "ok", err);
1421         return err;
1422 }
1423
1424 static int xs_bind6(struct sock_xprt *transport, struct socket *sock)
1425 {
1426         struct sockaddr_in6 myaddr = {
1427                 .sin6_family = AF_INET6,
1428         };
1429         struct sockaddr_in6 *sa;
1430         int err, nloop = 0;
1431         unsigned short port = xs_get_srcport(transport, sock);
1432         unsigned short last;
1433
1434         sa = (struct sockaddr_in6 *)&transport->addr;
1435         myaddr.sin6_addr = sa->sin6_addr;
1436         do {
1437                 myaddr.sin6_port = htons(port);
1438                 err = kernel_bind(sock, (struct sockaddr *) &myaddr,
1439                                                 sizeof(myaddr));
1440                 if (port == 0)
1441                         break;
1442                 if (err == 0) {
1443                         transport->port = port;
1444                         break;
1445                 }
1446                 last = port;
1447                 port = xs_next_srcport(transport, sock, port);
1448                 if (port > last)
1449                         nloop++;
1450         } while (err == -EADDRINUSE && nloop != 2);
1451         dprintk("RPC:       xs_bind6 "NIP6_FMT":%u: %s (%d)\n",
1452                 NIP6(myaddr.sin6_addr), port, err ? "failed" : "ok", err);
1453         return err;
1454 }
1455
1456 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1457 static struct lock_class_key xs_key[2];
1458 static struct lock_class_key xs_slock_key[2];
1459
1460 static inline void xs_reclassify_socket4(struct socket *sock)
1461 {
1462         struct sock *sk = sock->sk;
1463
1464         BUG_ON(sock_owned_by_user(sk));
1465         sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1466                 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1467 }
1468
1469 static inline void xs_reclassify_socket6(struct socket *sock)
1470 {
1471         struct sock *sk = sock->sk;
1472
1473         BUG_ON(sock_owned_by_user(sk));
1474         sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1475                 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
1476 }
1477 #else
1478 static inline void xs_reclassify_socket4(struct socket *sock)
1479 {
1480 }
1481
1482 static inline void xs_reclassify_socket6(struct socket *sock)
1483 {
1484 }
1485 #endif
1486
1487 static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
1488 {
1489         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1490
1491         if (!transport->inet) {
1492                 struct sock *sk = sock->sk;
1493
1494                 write_lock_bh(&sk->sk_callback_lock);
1495
1496                 xs_save_old_callbacks(transport, sk);
1497
1498                 sk->sk_user_data = xprt;
1499                 sk->sk_data_ready = xs_udp_data_ready;
1500                 sk->sk_write_space = xs_udp_write_space;
1501                 sk->sk_no_check = UDP_CSUM_NORCV;
1502                 sk->sk_allocation = GFP_ATOMIC;
1503
1504                 xprt_set_connected(xprt);
1505
1506                 /* Reset to new socket */
1507                 transport->sock = sock;
1508                 transport->inet = sk;
1509
1510                 write_unlock_bh(&sk->sk_callback_lock);
1511         }
1512         xs_udp_do_set_buffer_size(xprt);
1513 }
1514
1515 /**
1516  * xs_udp_connect_worker4 - set up a UDP socket
1517  * @work: RPC transport to connect
1518  *
1519  * Invoked by a work queue tasklet.
1520  */
1521 static void xs_udp_connect_worker4(struct work_struct *work)
1522 {
1523         struct sock_xprt *transport =
1524                 container_of(work, struct sock_xprt, connect_worker.work);
1525         struct rpc_xprt *xprt = &transport->xprt;
1526         struct socket *sock = transport->sock;
1527         int err, status = -EIO;
1528
1529         if (xprt->shutdown || !xprt_bound(xprt))
1530                 goto out;
1531
1532         /* Start by resetting any existing state */
1533         xs_close(xprt);
1534
1535         if ((err = sock_create_kern(PF_INET, SOCK_DGRAM, IPPROTO_UDP, &sock)) < 0) {
1536                 dprintk("RPC:       can't create UDP transport socket (%d).\n", -err);
1537                 goto out;
1538         }
1539         xs_reclassify_socket4(sock);
1540
1541         if (xs_bind4(transport, sock)) {
1542                 sock_release(sock);
1543                 goto out;
1544         }
1545
1546         dprintk("RPC:       worker connecting xprt %p to address: %s\n",
1547                         xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1548
1549         xs_udp_finish_connecting(xprt, sock);
1550         status = 0;
1551 out:
1552         xprt_wake_pending_tasks(xprt, status);
1553         xprt_clear_connecting(xprt);
1554 }
1555
1556 /**
1557  * xs_udp_connect_worker6 - set up a UDP socket
1558  * @work: RPC transport to connect
1559  *
1560  * Invoked by a work queue tasklet.
1561  */
1562 static void xs_udp_connect_worker6(struct work_struct *work)
1563 {
1564         struct sock_xprt *transport =
1565                 container_of(work, struct sock_xprt, connect_worker.work);
1566         struct rpc_xprt *xprt = &transport->xprt;
1567         struct socket *sock = transport->sock;
1568         int err, status = -EIO;
1569
1570         if (xprt->shutdown || !xprt_bound(xprt))
1571                 goto out;
1572
1573         /* Start by resetting any existing state */
1574         xs_close(xprt);
1575
1576         if ((err = sock_create_kern(PF_INET6, SOCK_DGRAM, IPPROTO_UDP, &sock)) < 0) {
1577                 dprintk("RPC:       can't create UDP transport socket (%d).\n", -err);
1578                 goto out;
1579         }
1580         xs_reclassify_socket6(sock);
1581
1582         if (xs_bind6(transport, sock) < 0) {
1583                 sock_release(sock);
1584                 goto out;
1585         }
1586
1587         dprintk("RPC:       worker connecting xprt %p to address: %s\n",
1588                         xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1589
1590         xs_udp_finish_connecting(xprt, sock);
1591         status = 0;
1592 out:
1593         xprt_wake_pending_tasks(xprt, status);
1594         xprt_clear_connecting(xprt);
1595 }
1596
1597 /*
1598  * We need to preserve the port number so the reply cache on the server can
1599  * find our cached RPC replies when we get around to reconnecting.
1600  */
1601 static void xs_tcp_reuse_connection(struct rpc_xprt *xprt)
1602 {
1603         int result;
1604         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1605         struct sockaddr any;
1606
1607         dprintk("RPC:       disconnecting xprt %p to reuse port\n", xprt);
1608
1609         /*
1610          * Disconnect the transport socket by doing a connect operation
1611          * with AF_UNSPEC.  This should return immediately...
1612          */
1613         memset(&any, 0, sizeof(any));
1614         any.sa_family = AF_UNSPEC;
1615         result = kernel_connect(transport->sock, &any, sizeof(any), 0);
1616         if (result)
1617                 dprintk("RPC:       AF_UNSPEC connect return code %d\n",
1618                                 result);
1619 }
1620
1621 static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
1622 {
1623         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1624
1625         if (!transport->inet) {
1626                 struct sock *sk = sock->sk;
1627
1628                 write_lock_bh(&sk->sk_callback_lock);
1629
1630                 xs_save_old_callbacks(transport, sk);
1631
1632                 sk->sk_user_data = xprt;
1633                 sk->sk_data_ready = xs_tcp_data_ready;
1634                 sk->sk_state_change = xs_tcp_state_change;
1635                 sk->sk_write_space = xs_tcp_write_space;
1636                 sk->sk_error_report = xs_tcp_error_report;
1637                 sk->sk_allocation = GFP_ATOMIC;
1638
1639                 /* socket options */
1640                 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
1641                 sock_reset_flag(sk, SOCK_LINGER);
1642                 tcp_sk(sk)->linger2 = 0;
1643                 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
1644
1645                 xprt_clear_connected(xprt);
1646
1647                 /* Reset to new socket */
1648                 transport->sock = sock;
1649                 transport->inet = sk;
1650
1651                 write_unlock_bh(&sk->sk_callback_lock);
1652         }
1653
1654         /* Tell the socket layer to start connecting... */
1655         xprt->stat.connect_count++;
1656         xprt->stat.connect_start = jiffies;
1657         return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
1658 }
1659
1660 /**
1661  * xs_tcp_connect_worker4 - connect a TCP socket to a remote endpoint
1662  * @work: RPC transport to connect
1663  *
1664  * Invoked by a work queue tasklet.
1665  */
1666 static void xs_tcp_connect_worker4(struct work_struct *work)
1667 {
1668         struct sock_xprt *transport =
1669                 container_of(work, struct sock_xprt, connect_worker.work);
1670         struct rpc_xprt *xprt = &transport->xprt;
1671         struct socket *sock = transport->sock;
1672         int err, status = -EIO;
1673
1674         if (xprt->shutdown || !xprt_bound(xprt))
1675                 goto out;
1676
1677         if (!sock) {
1678                 /* start from scratch */
1679                 if ((err = sock_create_kern(PF_INET, SOCK_STREAM, IPPROTO_TCP, &sock)) < 0) {
1680                         dprintk("RPC:       can't create TCP transport socket (%d).\n", -err);
1681                         goto out;
1682                 }
1683                 xs_reclassify_socket4(sock);
1684
1685                 if (xs_bind4(transport, sock) < 0) {
1686                         sock_release(sock);
1687                         goto out;
1688                 }
1689         } else
1690                 /* "close" the socket, preserving the local port */
1691                 xs_tcp_reuse_connection(xprt);
1692
1693         dprintk("RPC:       worker connecting xprt %p to address: %s\n",
1694                         xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1695
1696         status = xs_tcp_finish_connecting(xprt, sock);
1697         dprintk("RPC:       %p connect status %d connected %d sock state %d\n",
1698                         xprt, -status, xprt_connected(xprt),
1699                         sock->sk->sk_state);
1700         if (status < 0) {
1701                 switch (status) {
1702                         case -EINPROGRESS:
1703                         case -EALREADY:
1704                                 goto out_clear;
1705                         case -ECONNREFUSED:
1706                         case -ECONNRESET:
1707                                 /* retry with existing socket, after a delay */
1708                                 break;
1709                         default:
1710                                 /* get rid of existing socket, and retry */
1711                                 xs_tcp_shutdown(xprt);
1712                 }
1713         }
1714 out:
1715         xprt_wake_pending_tasks(xprt, status);
1716 out_clear:
1717         xprt_clear_connecting(xprt);
1718 }
1719
1720 /**
1721  * xs_tcp_connect_worker6 - connect a TCP socket to a remote endpoint
1722  * @work: RPC transport to connect
1723  *
1724  * Invoked by a work queue tasklet.
1725  */
1726 static void xs_tcp_connect_worker6(struct work_struct *work)
1727 {
1728         struct sock_xprt *transport =
1729                 container_of(work, struct sock_xprt, connect_worker.work);
1730         struct rpc_xprt *xprt = &transport->xprt;
1731         struct socket *sock = transport->sock;
1732         int err, status = -EIO;
1733
1734         if (xprt->shutdown || !xprt_bound(xprt))
1735                 goto out;
1736
1737         if (!sock) {
1738                 /* start from scratch */
1739                 if ((err = sock_create_kern(PF_INET6, SOCK_STREAM, IPPROTO_TCP, &sock)) < 0) {
1740                         dprintk("RPC:       can't create TCP transport socket (%d).\n", -err);
1741                         goto out;
1742                 }
1743                 xs_reclassify_socket6(sock);
1744
1745                 if (xs_bind6(transport, sock) < 0) {
1746                         sock_release(sock);
1747                         goto out;
1748                 }
1749         } else
1750                 /* "close" the socket, preserving the local port */
1751                 xs_tcp_reuse_connection(xprt);
1752
1753         dprintk("RPC:       worker connecting xprt %p to address: %s\n",
1754                         xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1755
1756         status = xs_tcp_finish_connecting(xprt, sock);
1757         dprintk("RPC:       %p connect status %d connected %d sock state %d\n",
1758                         xprt, -status, xprt_connected(xprt), sock->sk->sk_state);
1759         if (status < 0) {
1760                 switch (status) {
1761                         case -EINPROGRESS:
1762                         case -EALREADY:
1763                                 goto out_clear;
1764                         case -ECONNREFUSED:
1765                         case -ECONNRESET:
1766                                 /* retry with existing socket, after a delay */
1767                                 break;
1768                         default:
1769                                 /* get rid of existing socket, and retry */
1770                                 xs_tcp_shutdown(xprt);
1771                 }
1772         }
1773 out:
1774         xprt_wake_pending_tasks(xprt, status);
1775 out_clear:
1776         xprt_clear_connecting(xprt);
1777 }
1778
1779 /**
1780  * xs_connect - connect a socket to a remote endpoint
1781  * @task: address of RPC task that manages state of connect request
1782  *
1783  * TCP: If the remote end dropped the connection, delay reconnecting.
1784  *
1785  * UDP socket connects are synchronous, but we use a work queue anyway
1786  * to guarantee that even unprivileged user processes can set up a
1787  * socket on a privileged port.
1788  *
1789  * If a UDP socket connect fails, the delay behavior here prevents
1790  * retry floods (hard mounts).
1791  */
1792 static void xs_connect(struct rpc_task *task)
1793 {
1794         struct rpc_xprt *xprt = task->tk_xprt;
1795         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1796
1797         if (xprt_test_and_set_connecting(xprt))
1798                 return;
1799
1800         if (transport->sock != NULL) {
1801                 dprintk("RPC:       xs_connect delayed xprt %p for %lu "
1802                                 "seconds\n",
1803                                 xprt, xprt->reestablish_timeout / HZ);
1804                 queue_delayed_work(rpciod_workqueue,
1805                                    &transport->connect_worker,
1806                                    xprt->reestablish_timeout);
1807                 xprt->reestablish_timeout <<= 1;
1808                 if (xprt->reestablish_timeout > XS_TCP_MAX_REEST_TO)
1809                         xprt->reestablish_timeout = XS_TCP_MAX_REEST_TO;
1810         } else {
1811                 dprintk("RPC:       xs_connect scheduled xprt %p\n", xprt);
1812                 queue_delayed_work(rpciod_workqueue,
1813                                    &transport->connect_worker, 0);
1814         }
1815 }
1816
1817 static void xs_tcp_connect(struct rpc_task *task)
1818 {
1819         struct rpc_xprt *xprt = task->tk_xprt;
1820
1821         /* Initiate graceful shutdown of the socket if not already done */
1822         if (test_bit(XPRT_CONNECTED, &xprt->state))
1823                 xs_tcp_shutdown(xprt);
1824         /* Exit if we need to wait for socket shutdown to complete */
1825         if (test_bit(XPRT_CLOSING, &xprt->state))
1826                 return;
1827         xs_connect(task);
1828 }
1829
1830 /**
1831  * xs_udp_print_stats - display UDP socket-specifc stats
1832  * @xprt: rpc_xprt struct containing statistics
1833  * @seq: output file
1834  *
1835  */
1836 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
1837 {
1838         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1839
1840         seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %Lu %Lu\n",
1841                         transport->port,
1842                         xprt->stat.bind_count,
1843                         xprt->stat.sends,
1844                         xprt->stat.recvs,
1845                         xprt->stat.bad_xids,
1846                         xprt->stat.req_u,
1847                         xprt->stat.bklog_u);
1848 }
1849
1850 /**
1851  * xs_tcp_print_stats - display TCP socket-specifc stats
1852  * @xprt: rpc_xprt struct containing statistics
1853  * @seq: output file
1854  *
1855  */
1856 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
1857 {
1858         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1859         long idle_time = 0;
1860
1861         if (xprt_connected(xprt))
1862                 idle_time = (long)(jiffies - xprt->last_used) / HZ;
1863
1864         seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu %Lu %Lu\n",
1865                         transport->port,
1866                         xprt->stat.bind_count,
1867                         xprt->stat.connect_count,
1868                         xprt->stat.connect_time,
1869                         idle_time,
1870                         xprt->stat.sends,
1871                         xprt->stat.recvs,
1872                         xprt->stat.bad_xids,
1873                         xprt->stat.req_u,
1874                         xprt->stat.bklog_u);
1875 }
1876
1877 static struct rpc_xprt_ops xs_udp_ops = {
1878         .set_buffer_size        = xs_udp_set_buffer_size,
1879         .reserve_xprt           = xprt_reserve_xprt_cong,
1880         .release_xprt           = xprt_release_xprt_cong,
1881         .rpcbind                = rpcb_getport_async,
1882         .set_port               = xs_set_port,
1883         .connect                = xs_connect,
1884         .buf_alloc              = rpc_malloc,
1885         .buf_free               = rpc_free,
1886         .send_request           = xs_udp_send_request,
1887         .set_retrans_timeout    = xprt_set_retrans_timeout_rtt,
1888         .timer                  = xs_udp_timer,
1889         .release_request        = xprt_release_rqst_cong,
1890         .close                  = xs_close,
1891         .destroy                = xs_destroy,
1892         .print_stats            = xs_udp_print_stats,
1893 };
1894
1895 static struct rpc_xprt_ops xs_tcp_ops = {
1896         .reserve_xprt           = xprt_reserve_xprt,
1897         .release_xprt           = xs_tcp_release_xprt,
1898         .rpcbind                = rpcb_getport_async,
1899         .set_port               = xs_set_port,
1900         .connect                = xs_tcp_connect,
1901         .buf_alloc              = rpc_malloc,
1902         .buf_free               = rpc_free,
1903         .send_request           = xs_tcp_send_request,
1904         .set_retrans_timeout    = xprt_set_retrans_timeout_def,
1905         .close                  = xs_tcp_shutdown,
1906         .destroy                = xs_destroy,
1907         .print_stats            = xs_tcp_print_stats,
1908 };
1909
1910 static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
1911                                       unsigned int slot_table_size)
1912 {
1913         struct rpc_xprt *xprt;
1914         struct sock_xprt *new;
1915
1916         if (args->addrlen > sizeof(xprt->addr)) {
1917                 dprintk("RPC:       xs_setup_xprt: address too large\n");
1918                 return ERR_PTR(-EBADF);
1919         }
1920
1921         new = kzalloc(sizeof(*new), GFP_KERNEL);
1922         if (new == NULL) {
1923                 dprintk("RPC:       xs_setup_xprt: couldn't allocate "
1924                                 "rpc_xprt\n");
1925                 return ERR_PTR(-ENOMEM);
1926         }
1927         xprt = &new->xprt;
1928
1929         xprt->max_reqs = slot_table_size;
1930         xprt->slot = kcalloc(xprt->max_reqs, sizeof(struct rpc_rqst), GFP_KERNEL);
1931         if (xprt->slot == NULL) {
1932                 kfree(xprt);
1933                 dprintk("RPC:       xs_setup_xprt: couldn't allocate slot "
1934                                 "table\n");
1935                 return ERR_PTR(-ENOMEM);
1936         }
1937
1938         memcpy(&xprt->addr, args->dstaddr, args->addrlen);
1939         xprt->addrlen = args->addrlen;
1940         if (args->srcaddr)
1941                 memcpy(&new->addr, args->srcaddr, args->addrlen);
1942
1943         return xprt;
1944 }
1945
1946 static const struct rpc_timeout xs_udp_default_timeout = {
1947         .to_initval = 5 * HZ,
1948         .to_maxval = 30 * HZ,
1949         .to_increment = 5 * HZ,
1950         .to_retries = 5,
1951 };
1952
1953 /**
1954  * xs_setup_udp - Set up transport to use a UDP socket
1955  * @args: rpc transport creation arguments
1956  *
1957  */
1958 static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
1959 {
1960         struct sockaddr *addr = args->dstaddr;
1961         struct rpc_xprt *xprt;
1962         struct sock_xprt *transport;
1963
1964         xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries);
1965         if (IS_ERR(xprt))
1966                 return xprt;
1967         transport = container_of(xprt, struct sock_xprt, xprt);
1968
1969         xprt->prot = IPPROTO_UDP;
1970         xprt->tsh_size = 0;
1971         /* XXX: header size can vary due to auth type, IPv6, etc. */
1972         xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
1973
1974         xprt->bind_timeout = XS_BIND_TO;
1975         xprt->connect_timeout = XS_UDP_CONN_TO;
1976         xprt->reestablish_timeout = XS_UDP_REEST_TO;
1977         xprt->idle_timeout = XS_IDLE_DISC_TO;
1978
1979         xprt->ops = &xs_udp_ops;
1980
1981         xprt->timeout = &xs_udp_default_timeout;
1982
1983         switch (addr->sa_family) {
1984         case AF_INET:
1985                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
1986                         xprt_set_bound(xprt);
1987
1988                 INIT_DELAYED_WORK(&transport->connect_worker,
1989                                         xs_udp_connect_worker4);
1990                 xs_format_ipv4_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
1991                 break;
1992         case AF_INET6:
1993                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
1994                         xprt_set_bound(xprt);
1995
1996                 INIT_DELAYED_WORK(&transport->connect_worker,
1997                                         xs_udp_connect_worker6);
1998                 xs_format_ipv6_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
1999                 break;
2000         default:
2001                 kfree(xprt);
2002                 return ERR_PTR(-EAFNOSUPPORT);
2003         }
2004
2005         dprintk("RPC:       set up transport to address %s\n",
2006                         xprt->address_strings[RPC_DISPLAY_ALL]);
2007
2008         if (try_module_get(THIS_MODULE))
2009                 return xprt;
2010
2011         kfree(xprt->slot);
2012         kfree(xprt);
2013         return ERR_PTR(-EINVAL);
2014 }
2015
2016 static const struct rpc_timeout xs_tcp_default_timeout = {
2017         .to_initval = 60 * HZ,
2018         .to_maxval = 60 * HZ,
2019         .to_retries = 2,
2020 };
2021
2022 /**
2023  * xs_setup_tcp - Set up transport to use a TCP socket
2024  * @args: rpc transport creation arguments
2025  *
2026  */
2027 static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
2028 {
2029         struct sockaddr *addr = args->dstaddr;
2030         struct rpc_xprt *xprt;
2031         struct sock_xprt *transport;
2032
2033         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries);
2034         if (IS_ERR(xprt))
2035                 return xprt;
2036         transport = container_of(xprt, struct sock_xprt, xprt);
2037
2038         xprt->prot = IPPROTO_TCP;
2039         xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2040         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2041
2042         xprt->bind_timeout = XS_BIND_TO;
2043         xprt->connect_timeout = XS_TCP_CONN_TO;
2044         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2045         xprt->idle_timeout = XS_IDLE_DISC_TO;
2046
2047         xprt->ops = &xs_tcp_ops;
2048         xprt->timeout = &xs_tcp_default_timeout;
2049
2050         switch (addr->sa_family) {
2051         case AF_INET:
2052                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2053                         xprt_set_bound(xprt);
2054
2055                 INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_connect_worker4);
2056                 xs_format_ipv4_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
2057                 break;
2058         case AF_INET6:
2059                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2060                         xprt_set_bound(xprt);
2061
2062                 INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_connect_worker6);
2063                 xs_format_ipv6_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
2064                 break;
2065         default:
2066                 kfree(xprt);
2067                 return ERR_PTR(-EAFNOSUPPORT);
2068         }
2069
2070         dprintk("RPC:       set up transport to address %s\n",
2071                         xprt->address_strings[RPC_DISPLAY_ALL]);
2072
2073         if (try_module_get(THIS_MODULE))
2074                 return xprt;
2075
2076         kfree(xprt->slot);
2077         kfree(xprt);
2078         return ERR_PTR(-EINVAL);
2079 }
2080
2081 static struct xprt_class        xs_udp_transport = {
2082         .list           = LIST_HEAD_INIT(xs_udp_transport.list),
2083         .name           = "udp",
2084         .owner          = THIS_MODULE,
2085         .ident          = IPPROTO_UDP,
2086         .setup          = xs_setup_udp,
2087 };
2088
2089 static struct xprt_class        xs_tcp_transport = {
2090         .list           = LIST_HEAD_INIT(xs_tcp_transport.list),
2091         .name           = "tcp",
2092         .owner          = THIS_MODULE,
2093         .ident          = IPPROTO_TCP,
2094         .setup          = xs_setup_tcp,
2095 };
2096
2097 /**
2098  * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
2099  *
2100  */
2101 int init_socket_xprt(void)
2102 {
2103 #ifdef RPC_DEBUG
2104         if (!sunrpc_table_header)
2105                 sunrpc_table_header = register_sysctl_table(sunrpc_table);
2106 #endif
2107
2108         xprt_register_transport(&xs_udp_transport);
2109         xprt_register_transport(&xs_tcp_transport);
2110
2111         return 0;
2112 }
2113
2114 /**
2115  * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
2116  *
2117  */
2118 void cleanup_socket_xprt(void)
2119 {
2120 #ifdef RPC_DEBUG
2121         if (sunrpc_table_header) {
2122                 unregister_sysctl_table(sunrpc_table_header);
2123                 sunrpc_table_header = NULL;
2124         }
2125 #endif
2126
2127         xprt_unregister_transport(&xs_udp_transport);
2128         xprt_unregister_transport(&xs_tcp_transport);
2129 }