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 -ENOTCONN;
 471
 472         clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags);
 473         if (base != 0) {
 474                 addr = NULL;
 475                 addrlen = 0;
 476         }
 477
 478         if (base < xdr->head[0].iov_len || addr != NULL) {
 479                 unsigned int len = xdr->head[0].iov_len - base;
 480                 remainder -= len;
 481                 err = xs_send_kvec(sock, addr, addrlen, &xdr->head[0], base, remainder != 0);
 482                 if (remainder == 0 || err != len)
 483                         goto out;
 484                 sent += err;
 485                 base = 0;
 486         } else
 487                 base -= xdr->head[0].iov_len;
 488
 489         if (base < xdr->page_len) {
 490                 unsigned int len = xdr->page_len - base;
 491                 remainder -= len;
 492                 err = xs_send_pagedata(sock, xdr, base, remainder != 0);
 493                 if (remainder == 0 || err != len)
 494                         goto out;
 495                 sent += err;
 496                 base = 0;
 497         } else
 498                 base -= xdr->page_len;
 499
 500         if (base >= xdr->tail[0].iov_len)
 501                 return sent;
 502         err = xs_send_kvec(sock, NULL, 0, &xdr->tail[0], base, 0);
 503 out:
 504         if (sent == 0)
 505                 return err;
 506         if (err > 0)
 507                 sent += err;
 508         return sent;
 509 }
 510
 511 static void xs_nospace_callback(struct rpc_task *task)
 512 {
 513         struct sock_xprt *transport = container_of(task->tk_rqstp->rq_xprt, struct sock_xprt, xprt);
 514
 515         transport->inet->sk_write_pending--;
 516         clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 517 }
 518
 519 /**
 520  * xs_nospace - place task on wait queue if transmit was incomplete
 521  * @task: task to put to sleep
 522  *
 523  */
 524 static void xs_nospace(struct rpc_task *task)
 525 {
 526         struct rpc_rqst *req = task->tk_rqstp;
 527         struct rpc_xprt *xprt = req->rq_xprt;
 528         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 529
 530         dprintk("RPC: %5u xmit incomplete (%u left of %u)\n",
 531                         task->tk_pid, req->rq_slen - req->rq_bytes_sent,
 532                         req->rq_slen);
 533
 534         /* Protect against races with write_space */
 535         spin_lock_bh(&xprt->transport_lock);
 536
 537         /* Don't race with disconnect */
 538         if (xprt_connected(xprt)) {
 539                 if (test_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags)) {
 540                         /*
 541                          * Notify TCP that we're limited by the application
 542                          * window size
 543                          */
 544                         set_bit(SOCK_NOSPACE, &transport->sock->flags);
 545                         transport->inet->sk_write_pending++;
 546                         /* ...and wait for more buffer space */
 547                         xprt_wait_for_buffer_space(task, xs_nospace_callback);
 548                 }
 549         } else {
 550                 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 551                 task->tk_status = -ENOTCONN;
 552         }
 553
 554         spin_unlock_bh(&xprt->transport_lock);
 555 }
 556
 557 /**
 558  * xs_udp_send_request - write an RPC request to a UDP socket
 559  * @task: address of RPC task that manages the state of an RPC request
 560  *
 561  * Return values:
 562  *        0:    The request has been sent
 563  *   EAGAIN:    The socket was blocked, please call again later to
 564  *              complete the request
 565  * ENOTCONN:    Caller needs to invoke connect logic then call again
 566  *    other:    Some other error occured, the request was not sent
 567  */
 568 static int xs_udp_send_request(struct rpc_task *task)
 569 {
 570         struct rpc_rqst *req = task->tk_rqstp;
 571         struct rpc_xprt *xprt = req->rq_xprt;
 572         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 573         struct xdr_buf *xdr = &req->rq_snd_buf;
 574         int status;
 575
 576         xs_pktdump("packet data:",
 577                                 req->rq_svec->iov_base,
 578                                 req->rq_svec->iov_len);
 579
 580         status = xs_sendpages(transport->sock,
 581                               xs_addr(xprt),
 582                               xprt->addrlen, xdr,
 583                               req->rq_bytes_sent);
 584
 585         dprintk("RPC:       xs_udp_send_request(%u) = %d\n",
 586                         xdr->len - req->rq_bytes_sent, status);
 587
 588         if (status >= 0) {
 589                 task->tk_bytes_sent += status;
 590                 if (status >= req->rq_slen)
 591                         return 0;
 592                 /* Still some bytes left; set up for a retry later. */
 593                 status = -EAGAIN;
 594         }
 595
 596         switch (status) {
 597         case -EAGAIN:
 598                 xs_nospace(task);
 599                 break;
 600         case -ENETUNREACH:
 601         case -EPIPE:
 602         case -ECONNREFUSED:
 603                 /* When the server has died, an ICMP port unreachable message
 604                  * prompts ECONNREFUSED. */
 605                 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 606                 break;
 607         default:
 608                 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 609                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
 610                         -status);
 611         }
 612
 613         return status;
 614 }
 615
 616 /**
 617  * xs_tcp_shutdown - gracefully shut down a TCP socket
 618  * @xprt: transport
 619  *
 620  * Initiates a graceful shutdown of the TCP socket by calling the
 621  * equivalent of shutdown(SHUT_WR);
 622  */
 623 static void xs_tcp_shutdown(struct rpc_xprt *xprt)
 624 {
 625         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 626         struct socket *sock = transport->sock;
 627
 628         if (sock != NULL)
 629                 kernel_sock_shutdown(sock, SHUT_WR);
 630 }
 631
 632 static inline void xs_encode_tcp_record_marker(struct xdr_buf *buf)
 633 {
 634         u32 reclen = buf->len - sizeof(rpc_fraghdr);
 635         rpc_fraghdr *base = buf->head[0].iov_base;
 636         *base = htonl(RPC_LAST_STREAM_FRAGMENT | reclen);
 637 }
 638
 639 /**
 640  * xs_tcp_send_request - write an RPC request to a TCP socket
 641  * @task: address of RPC task that manages the state of an RPC request
 642  *
 643  * Return values:
 644  *        0:    The request has been sent
 645  *   EAGAIN:    The socket was blocked, please call again later to
 646  *              complete the request
 647  * ENOTCONN:    Caller needs to invoke connect logic then call again
 648  *    other:    Some other error occured, the request was not sent
 649  *
 650  * XXX: In the case of soft timeouts, should we eventually give up
 651  *      if sendmsg is not able to make progress?
 652  */
 653 static int xs_tcp_send_request(struct rpc_task *task)
 654 {
 655         struct rpc_rqst *req = task->tk_rqstp;
 656         struct rpc_xprt *xprt = req->rq_xprt;
 657         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 658         struct xdr_buf *xdr = &req->rq_snd_buf;
 659         int status;
 660
 661         xs_encode_tcp_record_marker(&req->rq_snd_buf);
 662
 663         xs_pktdump("packet data:",
 664                                 req->rq_svec->iov_base,
 665                                 req->rq_svec->iov_len);
 666
 667         /* Continue transmitting the packet/record. We must be careful
 668          * to cope with writespace callbacks arriving _after_ we have
 669          * called sendmsg(). */
 670         while (1) {
 671                 status = xs_sendpages(transport->sock,
 672                                         NULL, 0, xdr, req->rq_bytes_sent);
 673
 674                 dprintk("RPC:       xs_tcp_send_request(%u) = %d\n",
 675                                 xdr->len - req->rq_bytes_sent, status);
 676
 677                 if (unlikely(status < 0))
 678                         break;
 679
 680                 /* If we've sent the entire packet, immediately
 681                  * reset the count of bytes sent. */
 682                 req->rq_bytes_sent += status;
 683                 task->tk_bytes_sent += status;
 684                 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
 685                         req->rq_bytes_sent = 0;
 686                         return 0;
 687                 }
 688
 689                 if (status != 0)
 690                         continue;
 691                 status = -EAGAIN;
 692                 break;
 693         }
 694
 695         switch (status) {
 696         case -EAGAIN:
 697                 xs_nospace(task);
 698                 break;
 699         case -ECONNRESET:
 700                 xs_tcp_shutdown(xprt);
 701         case -ECONNREFUSED:
 702         case -ENOTCONN:
 703         case -EPIPE:
 704                 status = -ENOTCONN;
 705                 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 706                 break;
 707         default:
 708                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
 709                         -status);
 710                 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 711                 xs_tcp_shutdown(xprt);
 712         }
 713
 714         return status;
 715 }
 716
 717 /**
 718  * xs_tcp_release_xprt - clean up after a tcp transmission
 719  * @xprt: transport
 720  * @task: rpc task
 721  *
 722  * This cleans up if an error causes us to abort the transmission of a request.
 723  * In this case, the socket may need to be reset in order to avoid confusing
 724  * the server.
 725  */
 726 static void xs_tcp_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
 727 {
 728         struct rpc_rqst *req;
 729
 730         if (task != xprt->snd_task)
 731                 return;
 732         if (task == NULL)
 733                 goto out_release;
 734         req = task->tk_rqstp;
 735         if (req->rq_bytes_sent == 0)
 736                 goto out_release;
 737         if (req->rq_bytes_sent == req->rq_snd_buf.len)
 738                 goto out_release;
 739         set_bit(XPRT_CLOSE_WAIT, &task->tk_xprt->state);
 740 out_release:
 741         xprt_release_xprt(xprt, task);
 742 }
 743
 744 static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk)
 745 {
 746         transport->old_data_ready = sk->sk_data_ready;
 747         transport->old_state_change = sk->sk_state_change;
 748         transport->old_write_space = sk->sk_write_space;
 749         transport->old_error_report = sk->sk_error_report;
 750 }
 751
 752 static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk)
 753 {
 754         sk->sk_data_ready = transport->old_data_ready;
 755         sk->sk_state_change = transport->old_state_change;
 756         sk->sk_write_space = transport->old_write_space;
 757         sk->sk_error_report = transport->old_error_report;
 758 }
 759
 760 /**
 761  * xs_close - close a socket
 762  * @xprt: transport
 763  *
 764  * This is used when all requests are complete; ie, no DRC state remains
 765  * on the server we want to save.
 766  */
 767 static void xs_close(struct rpc_xprt *xprt)
 768 {
 769         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 770         struct socket *sock = transport->sock;
 771         struct sock *sk = transport->inet;
 772
 773         if (!sk)
 774                 goto clear_close_wait;
 775
 776         dprintk("RPC:       xs_close xprt %p\n", xprt);
 777
 778         write_lock_bh(&sk->sk_callback_lock);
 779         transport->inet = NULL;
 780         transport->sock = NULL;
 781
 782         sk->sk_user_data = NULL;
 783
 784         xs_restore_old_callbacks(transport, sk);
 785         write_unlock_bh(&sk->sk_callback_lock);
 786
 787         sk->sk_no_check = 0;
 788
 789         sock_release(sock);
 790 clear_close_wait:
 791         smp_mb__before_clear_bit();
 792         clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
 793         clear_bit(XPRT_CLOSING, &xprt->state);
 794         smp_mb__after_clear_bit();
 795         xprt_disconnect_done(xprt);
 796 }
 797
 798 /**
 799  * xs_destroy - prepare to shutdown a transport
 800  * @xprt: doomed transport
 801  *
 802  */
 803 static void xs_destroy(struct rpc_xprt *xprt)
 804 {
 805         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 806
 807         dprintk("RPC:       xs_destroy xprt %p\n", xprt);
 808
 809         cancel_rearming_delayed_work(&transport->connect_worker);
 810
 811         xs_close(xprt);
 812         xs_free_peer_addresses(xprt);
 813         kfree(xprt->slot);
 814         kfree(xprt);
 815         module_put(THIS_MODULE);
 816 }
 817
 818 static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
 819 {
 820         return (struct rpc_xprt *) sk->sk_user_data;
 821 }
 822
 823 /**
 824  * xs_udp_data_ready - "data ready" callback for UDP sockets
 825  * @sk: socket with data to read
 826  * @len: how much data to read
 827  *
 828  */
 829 static void xs_udp_data_ready(struct sock *sk, int len)
 830 {
 831         struct rpc_task *task;
 832         struct rpc_xprt *xprt;
 833         struct rpc_rqst *rovr;
 834         struct sk_buff *skb;
 835         int err, repsize, copied;
 836         u32 _xid;
 837         __be32 *xp;
 838
 839         read_lock(&sk->sk_callback_lock);
 840         dprintk("RPC:       xs_udp_data_ready...\n");
 841         if (!(xprt = xprt_from_sock(sk)))
 842                 goto out;
 843
 844         if ((skb = skb_recv_datagram(sk, 0, 1, &err)) == NULL)
 845                 goto out;
 846
 847         if (xprt->shutdown)
 848                 goto dropit;
 849
 850         repsize = skb->len - sizeof(struct udphdr);
 851         if (repsize < 4) {
 852                 dprintk("RPC:       impossible RPC reply size %d!\n", repsize);
 853                 goto dropit;
 854         }
 855
 856         /* Copy the XID from the skb... */
 857         xp = skb_header_pointer(skb, sizeof(struct udphdr),
 858                                 sizeof(_xid), &_xid);
 859         if (xp == NULL)
 860                 goto dropit;
 861
 862         /* Look up and lock the request corresponding to the given XID */
 863         spin_lock(&xprt->transport_lock);
 864         rovr = xprt_lookup_rqst(xprt, *xp);
 865         if (!rovr)
 866                 goto out_unlock;
 867         task = rovr->rq_task;
 868
 869         if ((copied = rovr->rq_private_buf.buflen) > repsize)
 870                 copied = repsize;
 871
 872         /* Suck it into the iovec, verify checksum if not done by hw. */
 873         if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
 874                 UDPX_INC_STATS_BH(sk, UDP_MIB_INERRORS);
 875                 goto out_unlock;
 876         }
 877
 878         UDPX_INC_STATS_BH(sk, UDP_MIB_INDATAGRAMS);
 879
 880         /* Something worked... */
 881         dst_confirm(skb->dst);
 882
 883         xprt_adjust_cwnd(task, copied);
 884         xprt_update_rtt(task);
 885         xprt_complete_rqst(task, copied);
 886
 887  out_unlock:
 888         spin_unlock(&xprt->transport_lock);
 889  dropit:
 890         skb_free_datagram(sk, skb);
 891  out:
 892         read_unlock(&sk->sk_callback_lock);
 893 }
 894
 895 static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
 896 {
 897         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 898         size_t len, used;
 899         char *p;
 900
 901         p = ((char *) &transport->tcp_fraghdr) + transport->tcp_offset;
 902         len = sizeof(transport->tcp_fraghdr) - transport->tcp_offset;
 903         used = xdr_skb_read_bits(desc, p, len);
 904         transport->tcp_offset += used;
 905         if (used != len)
 906                 return;
 907
 908         transport->tcp_reclen = ntohl(transport->tcp_fraghdr);
 909         if (transport->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
 910                 transport->tcp_flags |= TCP_RCV_LAST_FRAG;
 911         else
 912                 transport->tcp_flags &= ~TCP_RCV_LAST_FRAG;
 913         transport->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
 914
 915         transport->tcp_flags &= ~TCP_RCV_COPY_FRAGHDR;
 916         transport->tcp_offset = 0;
 917
 918         /* Sanity check of the record length */
 919         if (unlikely(transport->tcp_reclen < 4)) {
 920                 dprintk("RPC:       invalid TCP record fragment length\n");
 921                 xprt_force_disconnect(xprt);
 922                 return;
 923         }
 924         dprintk("RPC:       reading TCP record fragment of length %d\n",
 925                         transport->tcp_reclen);
 926 }
 927
 928 static void xs_tcp_check_fraghdr(struct sock_xprt *transport)
 929 {
 930         if (transport->tcp_offset == transport->tcp_reclen) {
 931                 transport->tcp_flags |= TCP_RCV_COPY_FRAGHDR;
 932                 transport->tcp_offset = 0;
 933                 if (transport->tcp_flags & TCP_RCV_LAST_FRAG) {
 934                         transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
 935                         transport->tcp_flags |= TCP_RCV_COPY_XID;
 936                         transport->tcp_copied = 0;
 937                 }
 938         }
 939 }
 940
 941 static inline void xs_tcp_read_xid(struct sock_xprt *transport, struct xdr_skb_reader *desc)
 942 {
 943         size_t len, used;
 944         char *p;
 945
 946         len = sizeof(transport->tcp_xid) - transport->tcp_offset;
 947         dprintk("RPC:       reading XID (%Zu bytes)\n", len);
 948         p = ((char *) &transport->tcp_xid) + transport->tcp_offset;
 949         used = xdr_skb_read_bits(desc, p, len);
 950         transport->tcp_offset += used;
 951         if (used != len)
 952                 return;
 953         transport->tcp_flags &= ~TCP_RCV_COPY_XID;
 954         transport->tcp_flags |= TCP_RCV_COPY_DATA;
 955         transport->tcp_copied = 4;
 956         dprintk("RPC:       reading reply for XID %08x\n",
 957                         ntohl(transport->tcp_xid));
 958         xs_tcp_check_fraghdr(transport);
 959 }
 960
 961 static inline void xs_tcp_read_request(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
 962 {
 963         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 964         struct rpc_rqst *req;
 965         struct xdr_buf *rcvbuf;
 966         size_t len;
 967         ssize_t r;
 968
 969         /* Find and lock the request corresponding to this xid */
 970         spin_lock(&xprt->transport_lock);
 971         req = xprt_lookup_rqst(xprt, transport->tcp_xid);
 972         if (!req) {
 973                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
 974                 dprintk("RPC:       XID %08x request not found!\n",
 975                                 ntohl(transport->tcp_xid));
 976                 spin_unlock(&xprt->transport_lock);
 977                 return;
 978         }
 979
 980         rcvbuf = &req->rq_private_buf;
 981         len = desc->count;
 982         if (len > transport->tcp_reclen - transport->tcp_offset) {
 983                 struct xdr_skb_reader my_desc;
 984
 985                 len = transport->tcp_reclen - transport->tcp_offset;
 986                 memcpy(&my_desc, desc, sizeof(my_desc));
 987                 my_desc.count = len;
 988                 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
 989                                           &my_desc, xdr_skb_read_bits);
 990                 desc->count -= r;
 991                 desc->offset += r;
 992         } else
 993                 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
 994                                           desc, xdr_skb_read_bits);
 995
 996         if (r > 0) {
 997                 transport->tcp_copied += r;
 998                 transport->tcp_offset += r;
 999         }
1000         if (r != len) {
1001                 /* Error when copying to the receive buffer,
1002                  * usually because we weren't able to allocate
1003                  * additional buffer pages. All we can do now
1004                  * is turn off TCP_RCV_COPY_DATA, so the request
1005                  * will not receive any additional updates,
1006                  * and time out.
1007                  * Any remaining data from this record will
1008                  * be discarded.
1009                  */
1010                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1011                 dprintk("RPC:       XID %08x truncated request\n",
1012                                 ntohl(transport->tcp_xid));
1013                 dprintk("RPC:       xprt = %p, tcp_copied = %lu, "
1014                                 "tcp_offset = %u, tcp_reclen = %u\n",
1015                                 xprt, transport->tcp_copied,
1016                                 transport->tcp_offset, transport->tcp_reclen);
1017                 goto out;
1018         }
1019
1020         dprintk("RPC:       XID %08x read %Zd bytes\n",
1021                         ntohl(transport->tcp_xid), r);
1022         dprintk("RPC:       xprt = %p, tcp_copied = %lu, tcp_offset = %u, "
1023                         "tcp_reclen = %u\n", xprt, transport->tcp_copied,
1024                         transport->tcp_offset, transport->tcp_reclen);
1025
1026         if (transport->tcp_copied == req->rq_private_buf.buflen)
1027                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1028         else if (transport->tcp_offset == transport->tcp_reclen) {
1029                 if (transport->tcp_flags & TCP_RCV_LAST_FRAG)
1030                         transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1031         }
1032
1033 out:
1034         if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1035                 xprt_complete_rqst(req->rq_task, transport->tcp_copied);
1036         spin_unlock(&xprt->transport_lock);
1037         xs_tcp_check_fraghdr(transport);
1038 }
1039
1040 static inline void xs_tcp_read_discard(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1041 {
1042         size_t len;
1043
1044         len = transport->tcp_reclen - transport->tcp_offset;
1045         if (len > desc->count)
1046                 len = desc->count;
1047         desc->count -= len;
1048         desc->offset += len;
1049         transport->tcp_offset += len;
1050         dprintk("RPC:       discarded %Zu bytes\n", len);
1051         xs_tcp_check_fraghdr(transport);
1052 }
1053
1054 static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
1055 {
1056         struct rpc_xprt *xprt = rd_desc->arg.data;
1057         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1058         struct xdr_skb_reader desc = {
1059                 .skb    = skb,
1060                 .offset = offset,
1061                 .count  = len,
1062         };
1063
1064         dprintk("RPC:       xs_tcp_data_recv started\n");
1065         do {
1066                 /* Read in a new fragment marker if necessary */
1067                 /* Can we ever really expect to get completely empty fragments? */
1068                 if (transport->tcp_flags & TCP_RCV_COPY_FRAGHDR) {
1069                         xs_tcp_read_fraghdr(xprt, &desc);
1070                         continue;
1071                 }
1072                 /* Read in the xid if necessary */
1073                 if (transport->tcp_flags & TCP_RCV_COPY_XID) {
1074                         xs_tcp_read_xid(transport, &desc);
1075                         continue;
1076                 }
1077                 /* Read in the request data */
1078                 if (transport->tcp_flags & TCP_RCV_COPY_DATA) {
1079                         xs_tcp_read_request(xprt, &desc);
1080                         continue;
1081                 }
1082                 /* Skip over any trailing bytes on short reads */
1083                 xs_tcp_read_discard(transport, &desc);
1084         } while (desc.count);
1085         dprintk("RPC:       xs_tcp_data_recv done\n");
1086         return len - desc.count;
1087 }
1088
1089 /**
1090  * xs_tcp_data_ready - "data ready" callback for TCP sockets
1091  * @sk: socket with data to read
1092  * @bytes: how much data to read
1093  *
1094  */
1095 static void xs_tcp_data_ready(struct sock *sk, int bytes)
1096 {
1097         struct rpc_xprt *xprt;
1098         read_descriptor_t rd_desc;
1099         int read;
1100
1101         dprintk("RPC:       xs_tcp_data_ready...\n");
1102
1103         read_lock(&sk->sk_callback_lock);
1104         if (!(xprt = xprt_from_sock(sk)))
1105                 goto out;
1106         if (xprt->shutdown)
1107                 goto out;
1108
1109         /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
1110         rd_desc.arg.data = xprt;
1111         do {
1112                 rd_desc.count = 65536;
1113                 read = tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
1114         } while (read > 0);
1115 out:
1116         read_unlock(&sk->sk_callback_lock);
1117 }
1118
1119 /**
1120  * xs_tcp_state_change - callback to handle TCP socket state changes
1121  * @sk: socket whose state has changed
1122  *
1123  */
1124 static void xs_tcp_state_change(struct sock *sk)
1125 {
1126         struct rpc_xprt *xprt;
1127
1128         read_lock(&sk->sk_callback_lock);
1129         if (!(xprt = xprt_from_sock(sk)))
1130                 goto out;
1131         dprintk("RPC:       xs_tcp_state_change client %p...\n", xprt);
1132         dprintk("RPC:       state %x conn %d dead %d zapped %d\n",
1133                         sk->sk_state, xprt_connected(xprt),
1134                         sock_flag(sk, SOCK_DEAD),
1135                         sock_flag(sk, SOCK_ZAPPED));
1136
1137         switch (sk->sk_state) {
1138         case TCP_ESTABLISHED:
1139                 spin_lock_bh(&xprt->transport_lock);
1140                 if (!xprt_test_and_set_connected(xprt)) {
1141                         struct sock_xprt *transport = container_of(xprt,
1142                                         struct sock_xprt, xprt);
1143
1144                         /* Reset TCP record info */
1145                         transport->tcp_offset = 0;
1146                         transport->tcp_reclen = 0;
1147                         transport->tcp_copied = 0;
1148                         transport->tcp_flags =
1149                                 TCP_RCV_COPY_FRAGHDR | TCP_RCV_COPY_XID;
1150
1151                         xprt_wake_pending_tasks(xprt, 0);
1152                 }
1153                 spin_unlock_bh(&xprt->transport_lock);
1154                 break;
1155         case TCP_FIN_WAIT1:
1156                 /* The client initiated a shutdown of the socket */
1157                 xprt->connect_cookie++;
1158                 xprt->reestablish_timeout = 0;
1159                 set_bit(XPRT_CLOSING, &xprt->state);
1160                 smp_mb__before_clear_bit();
1161                 clear_bit(XPRT_CONNECTED, &xprt->state);
1162                 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1163                 smp_mb__after_clear_bit();
1164                 break;
1165         case TCP_CLOSE_WAIT:
1166                 /* The server initiated a shutdown of the socket */
1167                 set_bit(XPRT_CLOSING, &xprt->state);
1168                 xprt_force_disconnect(xprt);
1169         case TCP_SYN_SENT:
1170                 xprt->connect_cookie++;
1171         case TCP_CLOSING:
1172                 /*
1173                  * If the server closed down the connection, make sure that
1174                  * we back off before reconnecting
1175                  */
1176                 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1177                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1178                 break;
1179         case TCP_LAST_ACK:
1180                 smp_mb__before_clear_bit();
1181                 clear_bit(XPRT_CONNECTED, &xprt->state);
1182                 smp_mb__after_clear_bit();
1183                 break;
1184         case TCP_CLOSE:
1185                 smp_mb__before_clear_bit();
1186                 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1187                 clear_bit(XPRT_CLOSING, &xprt->state);
1188                 smp_mb__after_clear_bit();
1189                 /* Mark transport as closed and wake up all pending tasks */
1190                 xprt_disconnect_done(xprt);
1191         }
1192  out:
1193         read_unlock(&sk->sk_callback_lock);
1194 }
1195
1196 /**
1197  * xs_tcp_error_report - callback mainly for catching RST events
1198  * @sk: socket
1199  */
1200 static void xs_tcp_error_report(struct sock *sk)
1201 {
1202         struct rpc_xprt *xprt;
1203
1204         read_lock(&sk->sk_callback_lock);
1205         if (sk->sk_err != ECONNRESET || sk->sk_state != TCP_ESTABLISHED)
1206                 goto out;
1207         if (!(xprt = xprt_from_sock(sk)))
1208                 goto out;
1209         dprintk("RPC:       %s client %p...\n"
1210                         "RPC:       error %d\n",
1211                         __func__, xprt, sk->sk_err);
1212
1213         xprt_force_disconnect(xprt);
1214 out:
1215         read_unlock(&sk->sk_callback_lock);
1216 }
1217
1218 /**
1219  * xs_udp_write_space - callback invoked when socket buffer space
1220  *                             becomes available
1221  * @sk: socket whose state has changed
1222  *
1223  * Called when more output buffer space is available for this socket.
1224  * We try not to wake our writers until they can make "significant"
1225  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1226  * with a bunch of small requests.
1227  */
1228 static void xs_udp_write_space(struct sock *sk)
1229 {
1230         read_lock(&sk->sk_callback_lock);
1231
1232         /* from net/core/sock.c:sock_def_write_space */
1233         if (sock_writeable(sk)) {
1234                 struct socket *sock;
1235                 struct rpc_xprt *xprt;
1236
1237                 if (unlikely(!(sock = sk->sk_socket)))
1238                         goto out;
1239                 clear_bit(SOCK_NOSPACE, &sock->flags);
1240
1241                 if (unlikely(!(xprt = xprt_from_sock(sk))))
1242                         goto out;
1243                 if (test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags) == 0)
1244                         goto out;
1245
1246                 xprt_write_space(xprt);
1247         }
1248
1249  out:
1250         read_unlock(&sk->sk_callback_lock);
1251 }
1252
1253 /**
1254  * xs_tcp_write_space - callback invoked when socket buffer space
1255  *                             becomes available
1256  * @sk: socket whose state has changed
1257  *
1258  * Called when more output buffer space is available for this socket.
1259  * We try not to wake our writers until they can make "significant"
1260  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1261  * with a bunch of small requests.
1262  */
1263 static void xs_tcp_write_space(struct sock *sk)
1264 {
1265         read_lock(&sk->sk_callback_lock);
1266
1267         /* from net/core/stream.c:sk_stream_write_space */
1268         if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk)) {
1269                 struct socket *sock;
1270                 struct rpc_xprt *xprt;
1271
1272                 if (unlikely(!(sock = sk->sk_socket)))
1273                         goto out;
1274                 clear_bit(SOCK_NOSPACE, &sock->flags);
1275
1276                 if (unlikely(!(xprt = xprt_from_sock(sk))))
1277                         goto out;
1278                 if (test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags) == 0)
1279                         goto out;
1280
1281                 xprt_write_space(xprt);
1282         }
1283
1284  out:
1285         read_unlock(&sk->sk_callback_lock);
1286 }
1287
1288 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
1289 {
1290         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1291         struct sock *sk = transport->inet;
1292
1293         if (transport->rcvsize) {
1294                 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1295                 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
1296         }
1297         if (transport->sndsize) {
1298                 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1299                 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1300                 sk->sk_write_space(sk);
1301         }
1302 }
1303
1304 /**
1305  * xs_udp_set_buffer_size - set send and receive limits
1306  * @xprt: generic transport
1307  * @sndsize: requested size of send buffer, in bytes
1308  * @rcvsize: requested size of receive buffer, in bytes
1309  *
1310  * Set socket send and receive buffer size limits.
1311  */
1312 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1313 {
1314         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1315
1316         transport->sndsize = 0;
1317         if (sndsize)
1318                 transport->sndsize = sndsize + 1024;
1319         transport->rcvsize = 0;
1320         if (rcvsize)
1321                 transport->rcvsize = rcvsize + 1024;
1322
1323         xs_udp_do_set_buffer_size(xprt);
1324 }
1325
1326 /**
1327  * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1328  * @task: task that timed out
1329  *
1330  * Adjust the congestion window after a retransmit timeout has occurred.
1331  */
1332 static void xs_udp_timer(struct rpc_task *task)
1333 {
1334         xprt_adjust_cwnd(task, -ETIMEDOUT);
1335 }
1336
1337 static unsigned short xs_get_random_port(void)
1338 {
1339         unsigned short range = xprt_max_resvport - xprt_min_resvport;
1340         unsigned short rand = (unsigned short) net_random() % range;
1341         return rand + xprt_min_resvport;
1342 }
1343
1344 /**
1345  * xs_set_port - reset the port number in the remote endpoint address
1346  * @xprt: generic transport
1347  * @port: new port number
1348  *
1349  */
1350 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1351 {
1352         struct sockaddr *addr = xs_addr(xprt);
1353
1354         dprintk("RPC:       setting port for xprt %p to %u\n", xprt, port);
1355
1356         switch (addr->sa_family) {
1357         case AF_INET:
1358                 ((struct sockaddr_in *)addr)->sin_port = htons(port);
1359                 break;
1360         case AF_INET6:
1361                 ((struct sockaddr_in6 *)addr)->sin6_port = htons(port);
1362                 break;
1363         default:
1364                 BUG();
1365         }
1366 }
1367
1368 static unsigned short xs_get_srcport(struct sock_xprt *transport, struct socket *sock)
1369 {
1370         unsigned short port = transport->port;
1371
1372         if (port == 0 && transport->xprt.resvport)
1373                 port = xs_get_random_port();
1374         return port;
1375 }
1376
1377 static unsigned short xs_next_srcport(struct sock_xprt *transport, struct socket *sock, unsigned short port)
1378 {
1379         if (transport->port != 0)
1380                 transport->port = 0;
1381         if (!transport->xprt.resvport)
1382                 return 0;
1383         if (port <= xprt_min_resvport || port > xprt_max_resvport)
1384                 return xprt_max_resvport;
1385         return --port;
1386 }
1387
1388 static int xs_bind4(struct sock_xprt *transport, struct socket *sock)
1389 {
1390         struct sockaddr_in myaddr = {
1391                 .sin_family = AF_INET,
1392         };
1393         struct sockaddr_in *sa;
1394         int err, nloop = 0;
1395         unsigned short port = xs_get_srcport(transport, sock);
1396         unsigned short last;
1397
1398         sa = (struct sockaddr_in *)&transport->addr;
1399         myaddr.sin_addr = sa->sin_addr;
1400         do {
1401                 myaddr.sin_port = htons(port);
1402                 err = kernel_bind(sock, (struct sockaddr *) &myaddr,
1403                                                 sizeof(myaddr));
1404                 if (port == 0)
1405                         break;
1406                 if (err == 0) {
1407                         transport->port = port;
1408                         break;
1409                 }
1410                 last = port;
1411                 port = xs_next_srcport(transport, sock, port);
1412                 if (port > last)
1413                         nloop++;
1414         } while (err == -EADDRINUSE && nloop != 2);
1415         dprintk("RPC:       %s %pI4:%u: %s (%d)\n",
1416                         __func__, &myaddr.sin_addr,
1417                         port, err ? "failed" : "ok", err);
1418         return err;
1419 }
1420
1421 static int xs_bind6(struct sock_xprt *transport, struct socket *sock)
1422 {
1423         struct sockaddr_in6 myaddr = {
1424                 .sin6_family = AF_INET6,
1425         };
1426         struct sockaddr_in6 *sa;
1427         int err, nloop = 0;
1428         unsigned short port = xs_get_srcport(transport, sock);
1429         unsigned short last;
1430
1431         sa = (struct sockaddr_in6 *)&transport->addr;
1432         myaddr.sin6_addr = sa->sin6_addr;
1433         do {
1434                 myaddr.sin6_port = htons(port);
1435                 err = kernel_bind(sock, (struct sockaddr *) &myaddr,
1436                                                 sizeof(myaddr));
1437                 if (port == 0)
1438                         break;
1439                 if (err == 0) {
1440                         transport->port = port;
1441                         break;
1442                 }
1443                 last = port;
1444                 port = xs_next_srcport(transport, sock, port);
1445                 if (port > last)
1446                         nloop++;
1447         } while (err == -EADDRINUSE && nloop != 2);
1448         dprintk("RPC:       xs_bind6 %pI6:%u: %s (%d)\n",
1449                 &myaddr.sin6_addr, port, err ? "failed" : "ok", err);
1450         return err;
1451 }
1452
1453 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1454 static struct lock_class_key xs_key[2];
1455 static struct lock_class_key xs_slock_key[2];
1456
1457 static inline void xs_reclassify_socket4(struct socket *sock)
1458 {
1459         struct sock *sk = sock->sk;
1460
1461         BUG_ON(sock_owned_by_user(sk));
1462         sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1463                 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1464 }
1465
1466 static inline void xs_reclassify_socket6(struct socket *sock)
1467 {
1468         struct sock *sk = sock->sk;
1469
1470         BUG_ON(sock_owned_by_user(sk));
1471         sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1472                 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
1473 }
1474 #else
1475 static inline void xs_reclassify_socket4(struct socket *sock)
1476 {
1477 }
1478
1479 static inline void xs_reclassify_socket6(struct socket *sock)
1480 {
1481 }
1482 #endif
1483
1484 static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
1485 {
1486         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1487
1488         if (!transport->inet) {
1489                 struct sock *sk = sock->sk;
1490
1491                 write_lock_bh(&sk->sk_callback_lock);
1492
1493                 xs_save_old_callbacks(transport, sk);
1494
1495                 sk->sk_user_data = xprt;
1496                 sk->sk_data_ready = xs_udp_data_ready;
1497                 sk->sk_write_space = xs_udp_write_space;
1498                 sk->sk_no_check = UDP_CSUM_NORCV;
1499                 sk->sk_allocation = GFP_ATOMIC;
1500
1501                 xprt_set_connected(xprt);
1502
1503                 /* Reset to new socket */
1504                 transport->sock = sock;
1505                 transport->inet = sk;
1506
1507                 write_unlock_bh(&sk->sk_callback_lock);
1508         }
1509         xs_udp_do_set_buffer_size(xprt);
1510 }
1511
1512 /**
1513  * xs_udp_connect_worker4 - set up a UDP socket
1514  * @work: RPC transport to connect
1515  *
1516  * Invoked by a work queue tasklet.
1517  */
1518 static void xs_udp_connect_worker4(struct work_struct *work)
1519 {
1520         struct sock_xprt *transport =
1521                 container_of(work, struct sock_xprt, connect_worker.work);
1522         struct rpc_xprt *xprt = &transport->xprt;
1523         struct socket *sock = transport->sock;
1524         int err, status = -EIO;
1525
1526         if (xprt->shutdown || !xprt_bound(xprt))
1527                 goto out;
1528
1529         /* Start by resetting any existing state */
1530         xs_close(xprt);
1531
1532         if ((err = sock_create_kern(PF_INET, SOCK_DGRAM, IPPROTO_UDP, &sock)) < 0) {
1533                 dprintk("RPC:       can't create UDP transport socket (%d).\n", -err);
1534                 goto out;
1535         }
1536         xs_reclassify_socket4(sock);
1537
1538         if (xs_bind4(transport, sock)) {
1539                 sock_release(sock);
1540                 goto out;
1541         }
1542
1543         dprintk("RPC:       worker connecting xprt %p to address: %s\n",
1544                         xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1545
1546         xs_udp_finish_connecting(xprt, sock);
1547         status = 0;
1548 out:
1549         xprt_wake_pending_tasks(xprt, status);
1550         xprt_clear_connecting(xprt);
1551 }
1552
1553 /**
1554  * xs_udp_connect_worker6 - set up a UDP socket
1555  * @work: RPC transport to connect
1556  *
1557  * Invoked by a work queue tasklet.
1558  */
1559 static void xs_udp_connect_worker6(struct work_struct *work)
1560 {
1561         struct sock_xprt *transport =
1562                 container_of(work, struct sock_xprt, connect_worker.work);
1563         struct rpc_xprt *xprt = &transport->xprt;
1564         struct socket *sock = transport->sock;
1565         int err, status = -EIO;
1566
1567         if (xprt->shutdown || !xprt_bound(xprt))
1568                 goto out;
1569
1570         /* Start by resetting any existing state */
1571         xs_close(xprt);
1572
1573         if ((err = sock_create_kern(PF_INET6, SOCK_DGRAM, IPPROTO_UDP, &sock)) < 0) {
1574                 dprintk("RPC:       can't create UDP transport socket (%d).\n", -err);
1575                 goto out;
1576         }
1577         xs_reclassify_socket6(sock);
1578
1579         if (xs_bind6(transport, sock) < 0) {
1580                 sock_release(sock);
1581                 goto out;
1582         }
1583
1584         dprintk("RPC:       worker connecting xprt %p to address: %s\n",
1585                         xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1586
1587         xs_udp_finish_connecting(xprt, sock);
1588         status = 0;
1589 out:
1590         xprt_wake_pending_tasks(xprt, status);
1591         xprt_clear_connecting(xprt);
1592 }
1593
1594 /*
1595  * We need to preserve the port number so the reply cache on the server can
1596  * find our cached RPC replies when we get around to reconnecting.
1597  */
1598 static void xs_tcp_reuse_connection(struct rpc_xprt *xprt)
1599 {
1600         int result;
1601         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1602         struct sockaddr any;
1603
1604         dprintk("RPC:       disconnecting xprt %p to reuse port\n", xprt);
1605
1606         /*
1607          * Disconnect the transport socket by doing a connect operation
1608          * with AF_UNSPEC.  This should return immediately...
1609          */
1610         memset(&any, 0, sizeof(any));
1611         any.sa_family = AF_UNSPEC;
1612         result = kernel_connect(transport->sock, &any, sizeof(any), 0);
1613         if (result)
1614                 dprintk("RPC:       AF_UNSPEC connect return code %d\n",
1615                                 result);
1616 }
1617
1618 static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
1619 {
1620         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1621
1622         if (!transport->inet) {
1623                 struct sock *sk = sock->sk;
1624
1625                 write_lock_bh(&sk->sk_callback_lock);
1626
1627                 xs_save_old_callbacks(transport, sk);
1628
1629                 sk->sk_user_data = xprt;
1630                 sk->sk_data_ready = xs_tcp_data_ready;
1631                 sk->sk_state_change = xs_tcp_state_change;
1632                 sk->sk_write_space = xs_tcp_write_space;
1633                 sk->sk_error_report = xs_tcp_error_report;
1634                 sk->sk_allocation = GFP_ATOMIC;
1635
1636                 /* socket options */
1637                 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
1638                 sock_reset_flag(sk, SOCK_LINGER);
1639                 tcp_sk(sk)->linger2 = 0;
1640                 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
1641
1642                 xprt_clear_connected(xprt);
1643
1644                 /* Reset to new socket */
1645                 transport->sock = sock;
1646                 transport->inet = sk;
1647
1648                 write_unlock_bh(&sk->sk_callback_lock);
1649         }
1650
1651         /* Tell the socket layer to start connecting... */
1652         xprt->stat.connect_count++;
1653         xprt->stat.connect_start = jiffies;
1654         return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
1655 }
1656
1657 /**
1658  * xs_tcp_connect_worker4 - connect a TCP socket to a remote endpoint
1659  * @work: RPC transport to connect
1660  *
1661  * Invoked by a work queue tasklet.
1662  */
1663 static void xs_tcp_connect_worker4(struct work_struct *work)
1664 {
1665         struct sock_xprt *transport =
1666                 container_of(work, struct sock_xprt, connect_worker.work);
1667         struct rpc_xprt *xprt = &transport->xprt;
1668         struct socket *sock = transport->sock;
1669         int err, status = -EIO;
1670
1671         if (xprt->shutdown || !xprt_bound(xprt))
1672                 goto out;
1673
1674         if (!sock) {
1675                 /* start from scratch */
1676                 if ((err = sock_create_kern(PF_INET, SOCK_STREAM, IPPROTO_TCP, &sock)) < 0) {
1677                         dprintk("RPC:       can't create TCP transport socket (%d).\n", -err);
1678                         goto out;
1679                 }
1680                 xs_reclassify_socket4(sock);
1681
1682                 if (xs_bind4(transport, sock) < 0) {
1683                         sock_release(sock);
1684                         goto out;
1685                 }
1686         } else
1687                 /* "close" the socket, preserving the local port */
1688                 xs_tcp_reuse_connection(xprt);
1689
1690         dprintk("RPC:       worker connecting xprt %p to address: %s\n",
1691                         xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1692
1693         status = xs_tcp_finish_connecting(xprt, sock);
1694         dprintk("RPC:       %p connect status %d connected %d sock state %d\n",
1695                         xprt, -status, xprt_connected(xprt),
1696                         sock->sk->sk_state);
1697         if (status < 0) {
1698                 switch (status) {
1699                         case -EINPROGRESS:
1700                         case -EALREADY:
1701                                 goto out_clear;
1702                         case -ECONNREFUSED:
1703                         case -ECONNRESET:
1704                                 /* retry with existing socket, after a delay */
1705                                 break;
1706                         default:
1707                                 /* get rid of existing socket, and retry */
1708                                 xs_tcp_shutdown(xprt);
1709                 }
1710         }
1711 out:
1712         xprt_wake_pending_tasks(xprt, status);
1713 out_clear:
1714         xprt_clear_connecting(xprt);
1715 }
1716
1717 /**
1718  * xs_tcp_connect_worker6 - connect a TCP socket to a remote endpoint
1719  * @work: RPC transport to connect
1720  *
1721  * Invoked by a work queue tasklet.
1722  */
1723 static void xs_tcp_connect_worker6(struct work_struct *work)
1724 {
1725         struct sock_xprt *transport =
1726                 container_of(work, struct sock_xprt, connect_worker.work);
1727         struct rpc_xprt *xprt = &transport->xprt;
1728         struct socket *sock = transport->sock;
1729         int err, status = -EIO;
1730
1731         if (xprt->shutdown || !xprt_bound(xprt))
1732                 goto out;
1733
1734         if (!sock) {
1735                 /* start from scratch */
1736                 if ((err = sock_create_kern(PF_INET6, SOCK_STREAM, IPPROTO_TCP, &sock)) < 0) {
1737                         dprintk("RPC:       can't create TCP transport socket (%d).\n", -err);
1738                         goto out;
1739                 }
1740                 xs_reclassify_socket6(sock);
1741
1742                 if (xs_bind6(transport, sock) < 0) {
1743                         sock_release(sock);
1744                         goto out;
1745                 }
1746         } else
1747                 /* "close" the socket, preserving the local port */
1748                 xs_tcp_reuse_connection(xprt);
1749
1750         dprintk("RPC:       worker connecting xprt %p to address: %s\n",
1751                         xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1752
1753         status = xs_tcp_finish_connecting(xprt, sock);
1754         dprintk("RPC:       %p connect status %d connected %d sock state %d\n",
1755                         xprt, -status, xprt_connected(xprt), sock->sk->sk_state);
1756         if (status < 0) {
1757                 switch (status) {
1758                         case -EINPROGRESS:
1759                         case -EALREADY:
1760                                 goto out_clear;
1761                         case -ECONNREFUSED:
1762                         case -ECONNRESET:
1763                                 /* retry with existing socket, after a delay */
1764                                 break;
1765                         default:
1766                                 /* get rid of existing socket, and retry */
1767                                 xs_tcp_shutdown(xprt);
1768                 }
1769         }
1770 out:
1771         xprt_wake_pending_tasks(xprt, status);
1772 out_clear:
1773         xprt_clear_connecting(xprt);
1774 }
1775
1776 /**
1777  * xs_connect - connect a socket to a remote endpoint
1778  * @task: address of RPC task that manages state of connect request
1779  *
1780  * TCP: If the remote end dropped the connection, delay reconnecting.
1781  *
1782  * UDP socket connects are synchronous, but we use a work queue anyway
1783  * to guarantee that even unprivileged user processes can set up a
1784  * socket on a privileged port.
1785  *
1786  * If a UDP socket connect fails, the delay behavior here prevents
1787  * retry floods (hard mounts).
1788  */
1789 static void xs_connect(struct rpc_task *task)
1790 {
1791         struct rpc_xprt *xprt = task->tk_xprt;
1792         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1793
1794         if (xprt_test_and_set_connecting(xprt))
1795                 return;
1796
1797         if (transport->sock != NULL) {
1798                 dprintk("RPC:       xs_connect delayed xprt %p for %lu "
1799                                 "seconds\n",
1800                                 xprt, xprt->reestablish_timeout / HZ);
1801                 queue_delayed_work(rpciod_workqueue,
1802                                    &transport->connect_worker,
1803                                    xprt->reestablish_timeout);
1804                 xprt->reestablish_timeout <<= 1;
1805                 if (xprt->reestablish_timeout > XS_TCP_MAX_REEST_TO)
1806                         xprt->reestablish_timeout = XS_TCP_MAX_REEST_TO;
1807         } else {
1808                 dprintk("RPC:       xs_connect scheduled xprt %p\n", xprt);
1809                 queue_delayed_work(rpciod_workqueue,
1810                                    &transport->connect_worker, 0);
1811         }
1812 }
1813
1814 static void xs_tcp_connect(struct rpc_task *task)
1815 {
1816         struct rpc_xprt *xprt = task->tk_xprt;
1817
1818         /* Initiate graceful shutdown of the socket if not already done */
1819         if (test_bit(XPRT_CONNECTED, &xprt->state))
1820                 xs_tcp_shutdown(xprt);
1821         /* Exit if we need to wait for socket shutdown to complete */
1822         if (test_bit(XPRT_CLOSING, &xprt->state))
1823                 return;
1824         xs_connect(task);
1825 }
1826
1827 /**
1828  * xs_udp_print_stats - display UDP socket-specifc stats
1829  * @xprt: rpc_xprt struct containing statistics
1830  * @seq: output file
1831  *
1832  */
1833 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
1834 {
1835         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1836
1837         seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %Lu %Lu\n",
1838                         transport->port,
1839                         xprt->stat.bind_count,
1840                         xprt->stat.sends,
1841                         xprt->stat.recvs,
1842                         xprt->stat.bad_xids,
1843                         xprt->stat.req_u,
1844                         xprt->stat.bklog_u);
1845 }
1846
1847 /**
1848  * xs_tcp_print_stats - display TCP socket-specifc stats
1849  * @xprt: rpc_xprt struct containing statistics
1850  * @seq: output file
1851  *
1852  */
1853 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
1854 {
1855         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1856         long idle_time = 0;
1857
1858         if (xprt_connected(xprt))
1859                 idle_time = (long)(jiffies - xprt->last_used) / HZ;
1860
1861         seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu %Lu %Lu\n",
1862                         transport->port,
1863                         xprt->stat.bind_count,
1864                         xprt->stat.connect_count,
1865                         xprt->stat.connect_time,
1866                         idle_time,
1867                         xprt->stat.sends,
1868                         xprt->stat.recvs,
1869                         xprt->stat.bad_xids,
1870                         xprt->stat.req_u,
1871                         xprt->stat.bklog_u);
1872 }
1873
1874 static struct rpc_xprt_ops xs_udp_ops = {
1875         .set_buffer_size        = xs_udp_set_buffer_size,
1876         .reserve_xprt           = xprt_reserve_xprt_cong,
1877         .release_xprt           = xprt_release_xprt_cong,
1878         .rpcbind                = rpcb_getport_async,
1879         .set_port               = xs_set_port,
1880         .connect                = xs_connect,
1881         .buf_alloc              = rpc_malloc,
1882         .buf_free               = rpc_free,
1883         .send_request           = xs_udp_send_request,
1884         .set_retrans_timeout    = xprt_set_retrans_timeout_rtt,
1885         .timer                  = xs_udp_timer,
1886         .release_request        = xprt_release_rqst_cong,
1887         .close                  = xs_close,
1888         .destroy                = xs_destroy,
1889         .print_stats            = xs_udp_print_stats,
1890 };
1891
1892 static struct rpc_xprt_ops xs_tcp_ops = {
1893         .reserve_xprt           = xprt_reserve_xprt,
1894         .release_xprt           = xs_tcp_release_xprt,
1895         .rpcbind                = rpcb_getport_async,
1896         .set_port               = xs_set_port,
1897         .connect                = xs_tcp_connect,
1898         .buf_alloc              = rpc_malloc,
1899         .buf_free               = rpc_free,
1900         .send_request           = xs_tcp_send_request,
1901         .set_retrans_timeout    = xprt_set_retrans_timeout_def,
1902         .close                  = xs_tcp_shutdown,
1903         .destroy                = xs_destroy,
1904         .print_stats            = xs_tcp_print_stats,
1905 };
1906
1907 static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
1908                                       unsigned int slot_table_size)
1909 {
1910         struct rpc_xprt *xprt;
1911         struct sock_xprt *new;
1912
1913         if (args->addrlen > sizeof(xprt->addr)) {
1914                 dprintk("RPC:       xs_setup_xprt: address too large\n");
1915                 return ERR_PTR(-EBADF);
1916         }
1917
1918         new = kzalloc(sizeof(*new), GFP_KERNEL);
1919         if (new == NULL) {
1920                 dprintk("RPC:       xs_setup_xprt: couldn't allocate "
1921                                 "rpc_xprt\n");
1922                 return ERR_PTR(-ENOMEM);
1923         }
1924         xprt = &new->xprt;
1925
1926         xprt->max_reqs = slot_table_size;
1927         xprt->slot = kcalloc(xprt->max_reqs, sizeof(struct rpc_rqst), GFP_KERNEL);
1928         if (xprt->slot == NULL) {
1929                 kfree(xprt);
1930                 dprintk("RPC:       xs_setup_xprt: couldn't allocate slot "
1931                                 "table\n");
1932                 return ERR_PTR(-ENOMEM);
1933         }
1934
1935         memcpy(&xprt->addr, args->dstaddr, args->addrlen);
1936         xprt->addrlen = args->addrlen;
1937         if (args->srcaddr)
1938                 memcpy(&new->addr, args->srcaddr, args->addrlen);
1939
1940         return xprt;
1941 }
1942
1943 static const struct rpc_timeout xs_udp_default_timeout = {
1944         .to_initval = 5 * HZ,
1945         .to_maxval = 30 * HZ,
1946         .to_increment = 5 * HZ,
1947         .to_retries = 5,
1948 };
1949
1950 /**
1951  * xs_setup_udp - Set up transport to use a UDP socket
1952  * @args: rpc transport creation arguments
1953  *
1954  */
1955 static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
1956 {
1957         struct sockaddr *addr = args->dstaddr;
1958         struct rpc_xprt *xprt;
1959         struct sock_xprt *transport;
1960
1961         xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries);
1962         if (IS_ERR(xprt))
1963                 return xprt;
1964         transport = container_of(xprt, struct sock_xprt, xprt);
1965
1966         xprt->prot = IPPROTO_UDP;
1967         xprt->tsh_size = 0;
1968         /* XXX: header size can vary due to auth type, IPv6, etc. */
1969         xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
1970
1971         xprt->bind_timeout = XS_BIND_TO;
1972         xprt->connect_timeout = XS_UDP_CONN_TO;
1973         xprt->reestablish_timeout = XS_UDP_REEST_TO;
1974         xprt->idle_timeout = XS_IDLE_DISC_TO;
1975
1976         xprt->ops = &xs_udp_ops;
1977
1978         xprt->timeout = &xs_udp_default_timeout;
1979
1980         switch (addr->sa_family) {
1981         case AF_INET:
1982                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
1983                         xprt_set_bound(xprt);
1984
1985                 INIT_DELAYED_WORK(&transport->connect_worker,
1986                                         xs_udp_connect_worker4);
1987                 xs_format_ipv4_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
1988                 break;
1989         case AF_INET6:
1990                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
1991                         xprt_set_bound(xprt);
1992
1993                 INIT_DELAYED_WORK(&transport->connect_worker,
1994                                         xs_udp_connect_worker6);
1995                 xs_format_ipv6_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
1996                 break;
1997         default:
1998                 kfree(xprt);
1999                 return ERR_PTR(-EAFNOSUPPORT);
2000         }
2001
2002         dprintk("RPC:       set up transport to address %s\n",
2003                         xprt->address_strings[RPC_DISPLAY_ALL]);
2004
2005         if (try_module_get(THIS_MODULE))
2006                 return xprt;
2007
2008         kfree(xprt->slot);
2009         kfree(xprt);
2010         return ERR_PTR(-EINVAL);
2011 }
2012
2013 static const struct rpc_timeout xs_tcp_default_timeout = {
2014         .to_initval = 60 * HZ,
2015         .to_maxval = 60 * HZ,
2016         .to_retries = 2,
2017 };
2018
2019 /**
2020  * xs_setup_tcp - Set up transport to use a TCP socket
2021  * @args: rpc transport creation arguments
2022  *
2023  */
2024 static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
2025 {
2026         struct sockaddr *addr = args->dstaddr;
2027         struct rpc_xprt *xprt;
2028         struct sock_xprt *transport;
2029
2030         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries);
2031         if (IS_ERR(xprt))
2032                 return xprt;
2033         transport = container_of(xprt, struct sock_xprt, xprt);
2034
2035         xprt->prot = IPPROTO_TCP;
2036         xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2037         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2038
2039         xprt->bind_timeout = XS_BIND_TO;
2040         xprt->connect_timeout = XS_TCP_CONN_TO;
2041         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2042         xprt->idle_timeout = XS_IDLE_DISC_TO;
2043
2044         xprt->ops = &xs_tcp_ops;
2045         xprt->timeout = &xs_tcp_default_timeout;
2046
2047         switch (addr->sa_family) {
2048         case AF_INET:
2049                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2050                         xprt_set_bound(xprt);
2051
2052                 INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_connect_worker4);
2053                 xs_format_ipv4_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
2054                 break;
2055         case AF_INET6:
2056                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2057                         xprt_set_bound(xprt);
2058
2059                 INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_connect_worker6);
2060                 xs_format_ipv6_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
2061                 break;
2062         default:
2063                 kfree(xprt);
2064                 return ERR_PTR(-EAFNOSUPPORT);
2065         }
2066
2067         dprintk("RPC:       set up transport to address %s\n",
2068                         xprt->address_strings[RPC_DISPLAY_ALL]);
2069
2070         if (try_module_get(THIS_MODULE))
2071                 return xprt;
2072
2073         kfree(xprt->slot);
2074         kfree(xprt);
2075         return ERR_PTR(-EINVAL);
2076 }
2077
2078 static struct xprt_class        xs_udp_transport = {
2079         .list           = LIST_HEAD_INIT(xs_udp_transport.list),
2080         .name           = "udp",
2081         .owner          = THIS_MODULE,
2082         .ident          = IPPROTO_UDP,
2083         .setup          = xs_setup_udp,
2084 };
2085
2086 static struct xprt_class        xs_tcp_transport = {
2087         .list           = LIST_HEAD_INIT(xs_tcp_transport.list),
2088         .name           = "tcp",
2089         .owner          = THIS_MODULE,
2090         .ident          = IPPROTO_TCP,
2091         .setup          = xs_setup_tcp,
2092 };
2093
2094 /**
2095  * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
2096  *
2097  */
2098 int init_socket_xprt(void)
2099 {
2100 #ifdef RPC_DEBUG
2101         if (!sunrpc_table_header)
2102                 sunrpc_table_header = register_sysctl_table(sunrpc_table);
2103 #endif
2104
2105         xprt_register_transport(&xs_udp_transport);
2106         xprt_register_transport(&xs_tcp_transport);
2107
2108         return 0;
2109 }
2110
2111 /**
2112  * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
2113  *
2114  */
2115 void cleanup_socket_xprt(void)
2116 {
2117 #ifdef RPC_DEBUG
2118         if (sunrpc_table_header) {
2119                 unregister_sysctl_table(sunrpc_table_header);
2120                 sunrpc_table_header = NULL;
2121         }
2122 #endif
2123
2124         xprt_unregister_transport(&xs_udp_transport);
2125         xprt_unregister_transport(&xs_tcp_transport);
2126 }