net/sunrpc/xprtsock.c

   1 /*
   2  * linux/net/sunrpc/xprtsock.c
   3  *
   4  * Client-side transport implementation for sockets.
   5  *
   6  * TCP callback races fixes (C) 1998 Red Hat
   7  * TCP send fixes (C) 1998 Red Hat
   8  * TCP NFS related read + write fixes
   9  *  (C) 1999 Dave Airlie, University of Limerick, Ireland <airlied@linux.ie>
  10  *
  11  * Rewrite of larges part of the code in order to stabilize TCP stuff.
  12  * Fix behaviour when socket buffer is full.
  13  *  (C) 1999 Trond Myklebust <trond.myklebust@fys.uio.no>
  14  *
  15  * IP socket transport implementation, (C) 2005 Chuck Lever <cel@netapp.com>
  16  *
  17  * IPv6 support contributed by Gilles Quillard, Bull Open Source, 2005.
  18  *   <gilles.quillard@bull.net>
  19  */
  20
  21 #include <linux/types.h>
  22 #include <linux/slab.h>
  23 #include <linux/module.h>
  24 #include <linux/capability.h>
  25 #include <linux/pagemap.h>
  26 #include <linux/errno.h>
  27 #include <linux/socket.h>
  28 #include <linux/in.h>
  29 #include <linux/net.h>
  30 #include <linux/mm.h>
  31 #include <linux/udp.h>
  32 #include <linux/tcp.h>
  33 #include <linux/sunrpc/clnt.h>
  34 #include <linux/sunrpc/sched.h>
  35 #include <linux/sunrpc/xprtsock.h>
  36 #include <linux/file.h>
  37
  38 #include <net/sock.h>
  39 #include <net/checksum.h>
  40 #include <net/udp.h>
  41 #include <net/tcp.h>
  42
  43 /*
  44  * xprtsock tunables
  45  */
  46 unsigned int xprt_udp_slot_table_entries = RPC_DEF_SLOT_TABLE;
  47 unsigned int xprt_tcp_slot_table_entries = RPC_DEF_SLOT_TABLE;
  48
  49 unsigned int xprt_min_resvport = RPC_DEF_MIN_RESVPORT;
  50 unsigned int xprt_max_resvport = RPC_DEF_MAX_RESVPORT;
  51
  52 /*
  53  * We can register our own files under /proc/sys/sunrpc by
  54  * calling register_sysctl_table() again.  The files in that
  55  * directory become the union of all files registered there.
  56  *
  57  * We simply need to make sure that we don't collide with
  58  * someone else's file names!
  59  */
  60
  61 #ifdef RPC_DEBUG
  62
  63 static unsigned int min_slot_table_size = RPC_MIN_SLOT_TABLE;
  64 static unsigned int max_slot_table_size = RPC_MAX_SLOT_TABLE;
  65 static unsigned int xprt_min_resvport_limit = RPC_MIN_RESVPORT;
  66 static unsigned int xprt_max_resvport_limit = RPC_MAX_RESVPORT;
  67
  68 static struct ctl_table_header *sunrpc_table_header;
  69
  70 /*
  71  * FIXME: changing the UDP slot table size should also resize the UDP
  72  *        socket buffers for existing UDP transports
  73  */
  74 static ctl_table xs_tunables_table[] = {
  75         {
  76                 .ctl_name       = CTL_SLOTTABLE_UDP,
  77                 .procname       = "udp_slot_table_entries",
  78                 .data           = &xprt_udp_slot_table_entries,
  79                 .maxlen         = sizeof(unsigned int),
  80                 .mode           = 0644,
  81                 .proc_handler   = &proc_dointvec_minmax,
  82                 .strategy       = &sysctl_intvec,
  83                 .extra1         = &min_slot_table_size,
  84                 .extra2         = &max_slot_table_size
  85         },
  86         {
  87                 .ctl_name       = CTL_SLOTTABLE_TCP,
  88                 .procname       = "tcp_slot_table_entries",
  89                 .data           = &xprt_tcp_slot_table_entries,
  90                 .maxlen         = sizeof(unsigned int),
  91                 .mode           = 0644,
  92                 .proc_handler   = &proc_dointvec_minmax,
  93                 .strategy       = &sysctl_intvec,
  94                 .extra1         = &min_slot_table_size,
  95                 .extra2         = &max_slot_table_size
  96         },
  97         {
  98                 .ctl_name       = CTL_MIN_RESVPORT,
  99                 .procname       = "min_resvport",
 100                 .data           = &xprt_min_resvport,
 101                 .maxlen         = sizeof(unsigned int),
 102                 .mode           = 0644,
 103                 .proc_handler   = &proc_dointvec_minmax,
 104                 .strategy       = &sysctl_intvec,
 105                 .extra1         = &xprt_min_resvport_limit,
 106                 .extra2         = &xprt_max_resvport_limit
 107         },
 108         {
 109                 .ctl_name       = CTL_MAX_RESVPORT,
 110                 .procname       = "max_resvport",
 111                 .data           = &xprt_max_resvport,
 112                 .maxlen         = sizeof(unsigned int),
 113                 .mode           = 0644,
 114                 .proc_handler   = &proc_dointvec_minmax,
 115                 .strategy       = &sysctl_intvec,
 116                 .extra1         = &xprt_min_resvport_limit,
 117                 .extra2         = &xprt_max_resvport_limit
 118         },
 119         {
 120                 .ctl_name = 0,
 121         },
 122 };
 123
 124 static ctl_table sunrpc_table[] = {
 125         {
 126                 .ctl_name       = CTL_SUNRPC,
 127                 .procname       = "sunrpc",
 128                 .mode           = 0555,
 129                 .child          = xs_tunables_table
 130         },
 131         {
 132                 .ctl_name = 0,
 133         },
 134 };
 135
 136 #endif
 137
 138 /*
 139  * Time out for an RPC UDP socket connect.  UDP socket connects are
 140  * synchronous, but we set a timeout anyway in case of resource
 141  * exhaustion on the local host.
 142  */
 143 #define XS_UDP_CONN_TO          (5U * HZ)
 144
 145 /*
 146  * Wait duration for an RPC TCP connection to be established.  Solaris
 147  * NFS over TCP uses 60 seconds, for example, which is in line with how
 148  * long a server takes to reboot.
 149  */
 150 #define XS_TCP_CONN_TO          (60U * HZ)
 151
 152 /*
 153  * Wait duration for a reply from the RPC portmapper.
 154  */
 155 #define XS_BIND_TO              (60U * HZ)
 156
 157 /*
 158  * Delay if a UDP socket connect error occurs.  This is most likely some
 159  * kind of resource problem on the local host.
 160  */
 161 #define XS_UDP_REEST_TO         (2U * HZ)
 162
 163 /*
 164  * The reestablish timeout allows clients to delay for a bit before attempting
 165  * to reconnect to a server that just dropped our connection.
 166  *
 167  * We implement an exponential backoff when trying to reestablish a TCP
 168  * transport connection with the server.  Some servers like to drop a TCP
 169  * connection when they are overworked, so we start with a short timeout and
 170  * increase over time if the server is down or not responding.
 171  */
 172 #define XS_TCP_INIT_REEST_TO    (3U * HZ)
 173 #define XS_TCP_MAX_REEST_TO     (5U * 60 * HZ)
 174
 175 /*
 176  * TCP idle timeout; client drops the transport socket if it is idle
 177  * for this long.  Note that we also timeout UDP sockets to prevent
 178  * holding port numbers when there is no RPC traffic.
 179  */
 180 #define XS_IDLE_DISC_TO         (5U * 60 * HZ)
 181
 182 #ifdef RPC_DEBUG
 183 # undef  RPC_DEBUG_DATA
 184 # define RPCDBG_FACILITY        RPCDBG_TRANS
 185 #endif
 186
 187 #ifdef RPC_DEBUG_DATA
 188 static void xs_pktdump(char *msg, u32 *packet, unsigned int count)
 189 {
 190         u8 *buf = (u8 *) packet;
 191         int j;
 192
 193         dprintk("RPC:       %s\n", msg);
 194         for (j = 0; j < count && j < 128; j += 4) {
 195                 if (!(j & 31)) {
 196                         if (j)
 197                                 dprintk("\n");
 198                         dprintk("0x%04x ", j);
 199                 }
 200                 dprintk("%02x%02x%02x%02x ",
 201                         buf[j], buf[j+1], buf[j+2], buf[j+3]);
 202         }
 203         dprintk("\n");
 204 }
 205 #else
 206 static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count)
 207 {
 208         /* NOP */
 209 }
 210 #endif
 211
 212 struct sock_xprt {
 213         struct rpc_xprt         xprt;
 214
 215         /*
 216          * Network layer
 217          */
 218         struct socket *         sock;
 219         struct sock *           inet;
 220
 221         /*
 222          * State of TCP reply receive
 223          */
 224         __be32                  tcp_fraghdr,
 225                                 tcp_xid;
 226
 227         u32                     tcp_offset,
 228                                 tcp_reclen;
 229
 230         unsigned long           tcp_copied,
 231                                 tcp_flags;
 232
 233         /*
 234          * Connection of transports
 235          */
 236         struct delayed_work     connect_worker;
 237         struct sockaddr_storage addr;
 238         unsigned short          port;
 239
 240         /*
 241          * UDP socket buffer size parameters
 242          */
 243         size_t                  rcvsize,
 244                                 sndsize;
 245
 246         /*
 247          * Saved socket callback addresses
 248          */
 249         void                    (*old_data_ready)(struct sock *, int);
 250         void                    (*old_state_change)(struct sock *);
 251         void                    (*old_write_space)(struct sock *);
 252         void                    (*old_error_report)(struct sock *);
 253 };
 254
 255 /*
 256  * TCP receive state flags
 257  */
 258 #define TCP_RCV_LAST_FRAG       (1UL << 0)
 259 #define TCP_RCV_COPY_FRAGHDR    (1UL << 1)
 260 #define TCP_RCV_COPY_XID        (1UL << 2)
 261 #define TCP_RCV_COPY_DATA       (1UL << 3)
 262
 263 static inline struct sockaddr *xs_addr(struct rpc_xprt *xprt)
 264 {
 265         return (struct sockaddr *) &xprt->addr;
 266 }
 267
 268 static inline struct sockaddr_in *xs_addr_in(struct rpc_xprt *xprt)
 269 {
 270         return (struct sockaddr_in *) &xprt->addr;
 271 }
 272
 273 static inline struct sockaddr_in6 *xs_addr_in6(struct rpc_xprt *xprt)
 274 {
 275         return (struct sockaddr_in6 *) &xprt->addr;
 276 }
 277
 278 static void xs_format_ipv4_peer_addresses(struct rpc_xprt *xprt,
 279                                           const char *protocol,
 280                                           const char *netid)
 281 {
 282         struct sockaddr_in *addr = xs_addr_in(xprt);
 283         char *buf;
 284
 285         buf = kzalloc(20, GFP_KERNEL);
 286         if (buf) {
 287                 snprintf(buf, 20, "%pI4", &addr->sin_addr.s_addr);
 288         }
 289         xprt->address_strings[RPC_DISPLAY_ADDR] = buf;
 290
 291         buf = kzalloc(8, GFP_KERNEL);
 292         if (buf) {
 293                 snprintf(buf, 8, "%u",
 294                                 ntohs(addr->sin_port));
 295         }
 296         xprt->address_strings[RPC_DISPLAY_PORT] = buf;
 297
 298         xprt->address_strings[RPC_DISPLAY_PROTO] = protocol;
 299
 300         buf = kzalloc(48, GFP_KERNEL);
 301         if (buf) {
 302                 snprintf(buf, 48, "addr=%pI4 port=%u proto=%s",
 303                         &addr->sin_addr.s_addr,
 304                         ntohs(addr->sin_port),
 305                         protocol);
 306         }
 307         xprt->address_strings[RPC_DISPLAY_ALL] = buf;
 308
 309         buf = kzalloc(10, GFP_KERNEL);
 310         if (buf) {
 311                 snprintf(buf, 10, "%02x%02x%02x%02x",
 312                                 NIPQUAD(addr->sin_addr.s_addr));
 313         }
 314         xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = buf;
 315
 316         buf = kzalloc(8, GFP_KERNEL);
 317         if (buf) {
 318                 snprintf(buf, 8, "%4hx",
 319                                 ntohs(addr->sin_port));
 320         }
 321         xprt->address_strings[RPC_DISPLAY_HEX_PORT] = buf;
 322
 323         buf = kzalloc(30, GFP_KERNEL);
 324         if (buf) {
 325                 snprintf(buf, 30, "%pI4.%u.%u",
 326                                 &addr->sin_addr.s_addr,
 327                                 ntohs(addr->sin_port) >> 8,
 328                                 ntohs(addr->sin_port) & 0xff);
 329         }
 330         xprt->address_strings[RPC_DISPLAY_UNIVERSAL_ADDR] = buf;
 331
 332         xprt->address_strings[RPC_DISPLAY_NETID] = netid;
 333 }
 334
 335 static void xs_format_ipv6_peer_addresses(struct rpc_xprt *xprt,
 336                                           const char *protocol,
 337                                           const char *netid)
 338 {
 339         struct sockaddr_in6 *addr = xs_addr_in6(xprt);
 340         char *buf;
 341
 342         buf = kzalloc(40, GFP_KERNEL);
 343         if (buf) {
 344                 snprintf(buf, 40, "%pI6",&addr->sin6_addr);
 345         }
 346         xprt->address_strings[RPC_DISPLAY_ADDR] = buf;
 347
 348         buf = kzalloc(8, GFP_KERNEL);
 349         if (buf) {
 350                 snprintf(buf, 8, "%u",
 351                                 ntohs(addr->sin6_port));
 352         }
 353         xprt->address_strings[RPC_DISPLAY_PORT] = buf;
 354
 355         xprt->address_strings[RPC_DISPLAY_PROTO] = protocol;
 356
 357         buf = kzalloc(64, GFP_KERNEL);
 358         if (buf) {
 359                 snprintf(buf, 64, "addr=%pI6 port=%u proto=%s",
 360                                 &addr->sin6_addr,
 361                                 ntohs(addr->sin6_port),
 362                                 protocol);
 363         }
 364         xprt->address_strings[RPC_DISPLAY_ALL] = buf;
 365
 366         buf = kzalloc(36, GFP_KERNEL);
 367         if (buf)
 368                 snprintf(buf, 36, "%pi6", &addr->sin6_addr);
 369
 370         xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = buf;
 371
 372         buf = kzalloc(8, GFP_KERNEL);
 373         if (buf) {
 374                 snprintf(buf, 8, "%4hx",
 375                                 ntohs(addr->sin6_port));
 376         }
 377         xprt->address_strings[RPC_DISPLAY_HEX_PORT] = buf;
 378
 379         buf = kzalloc(50, GFP_KERNEL);
 380         if (buf) {
 381                 snprintf(buf, 50, "%pI6.%u.%u",
 382                          &addr->sin6_addr,
 383                          ntohs(addr->sin6_port) >> 8,
 384                          ntohs(addr->sin6_port) & 0xff);
 385         }
 386         xprt->address_strings[RPC_DISPLAY_UNIVERSAL_ADDR] = buf;
 387
 388         xprt->address_strings[RPC_DISPLAY_NETID] = netid;
 389 }
 390
 391 static void xs_free_peer_addresses(struct rpc_xprt *xprt)
 392 {
 393         unsigned int i;
 394
 395         for (i = 0; i < RPC_DISPLAY_MAX; i++)
 396                 switch (i) {
 397                 case RPC_DISPLAY_PROTO:
 398                 case RPC_DISPLAY_NETID:
 399                         continue;
 400                 default:
 401                         kfree(xprt->address_strings[i]);
 402                 }
 403 }
 404
 405 #define XS_SENDMSG_FLAGS        (MSG_DONTWAIT | MSG_NOSIGNAL)
 406
 407 static int xs_send_kvec(struct socket *sock, struct sockaddr *addr, int addrlen, struct kvec *vec, unsigned int base, int more)
 408 {
 409         struct msghdr msg = {
 410                 .msg_name       = addr,
 411                 .msg_namelen    = addrlen,
 412                 .msg_flags      = XS_SENDMSG_FLAGS | (more ? MSG_MORE : 0),
 413         };
 414         struct kvec iov = {
 415                 .iov_base       = vec->iov_base + base,
 416                 .iov_len        = vec->iov_len - base,
 417         };
 418
 419         if (iov.iov_len != 0)
 420                 return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
 421         return kernel_sendmsg(sock, &msg, NULL, 0, 0);
 422 }
 423
 424 static int xs_send_pagedata(struct socket *sock, struct xdr_buf *xdr, unsigned int base, int more)
 425 {
 426         struct page **ppage;
 427         unsigned int remainder;
 428         int err, sent = 0;
 429
 430         remainder = xdr->page_len - base;
 431         base += xdr->page_base;
 432         ppage = xdr->pages + (base >> PAGE_SHIFT);
 433         base &= ~PAGE_MASK;
 434         for(;;) {
 435                 unsigned int len = min_t(unsigned int, PAGE_SIZE - base, remainder);
 436                 int flags = XS_SENDMSG_FLAGS;
 437
 438                 remainder -= len;
 439                 if (remainder != 0 || more)
 440                         flags |= MSG_MORE;
 441                 err = sock->ops->sendpage(sock, *ppage, base, len, flags);
 442                 if (remainder == 0 || err != len)
 443                         break;
 444                 sent += err;
 445                 ppage++;
 446                 base = 0;
 447         }
 448         if (sent == 0)
 449                 return err;
 450         if (err > 0)
 451                 sent += err;
 452         return sent;
 453 }
 454
 455 /**
 456  * xs_sendpages - write pages directly to a socket
 457  * @sock: socket to send on
 458  * @addr: UDP only -- address of destination
 459  * @addrlen: UDP only -- length of destination address
 460  * @xdr: buffer containing this request
 461  * @base: starting position in the buffer
 462  *
 463  */
 464 static int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base)
 465 {
 466         unsigned int remainder = xdr->len - base;
 467         int err, sent = 0;
 468
 469         if (unlikely(!sock))
 470                 return -ENOTSOCK;
 471
 472         clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags);
 473         if (base != 0) {
 474                 addr = NULL;
 475                 addrlen = 0;
 476         }
 477
 478         if (base < xdr->head[0].iov_len || addr != NULL) {
 479                 unsigned int len = xdr->head[0].iov_len - base;
 480                 remainder -= len;
 481                 err = xs_send_kvec(sock, addr, addrlen, &xdr->head[0], base, remainder != 0);
 482                 if (remainder == 0 || err != len)
 483                         goto out;
 484                 sent += err;
 485                 base = 0;
 486         } else
 487                 base -= xdr->head[0].iov_len;
 488
 489         if (base < xdr->page_len) {
 490                 unsigned int len = xdr->page_len - base;
 491                 remainder -= len;
 492                 err = xs_send_pagedata(sock, xdr, base, remainder != 0);
 493                 if (remainder == 0 || err != len)
 494                         goto out;
 495                 sent += err;
 496                 base = 0;
 497         } else
 498                 base -= xdr->page_len;
 499
 500         if (base >= xdr->tail[0].iov_len)
 501                 return sent;
 502         err = xs_send_kvec(sock, NULL, 0, &xdr->tail[0], base, 0);
 503 out:
 504         if (sent == 0)
 505                 return err;
 506         if (err > 0)
 507                 sent += err;
 508         return sent;
 509 }
 510
 511 static void xs_nospace_callback(struct rpc_task *task)
 512 {
 513         struct sock_xprt *transport = container_of(task->tk_rqstp->rq_xprt, struct sock_xprt, xprt);
 514
 515         transport->inet->sk_write_pending--;
 516         clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 517 }
 518
 519 /**
 520  * xs_nospace - place task on wait queue if transmit was incomplete
 521  * @task: task to put to sleep
 522  *
 523  */
 524 static void xs_nospace(struct rpc_task *task)
 525 {
 526         struct rpc_rqst *req = task->tk_rqstp;
 527         struct rpc_xprt *xprt = req->rq_xprt;
 528         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 529
 530         dprintk("RPC: %5u xmit incomplete (%u left of %u)\n",
 531                         task->tk_pid, req->rq_slen - req->rq_bytes_sent,
 532                         req->rq_slen);
 533
 534         /* Protect against races with write_space */
 535         spin_lock_bh(&xprt->transport_lock);
 536
 537         /* Don't race with disconnect */
 538         if (xprt_connected(xprt)) {
 539                 if (test_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags)) {
 540                         /*
 541                          * Notify TCP that we're limited by the application
 542                          * window size
 543                          */
 544                         set_bit(SOCK_NOSPACE, &transport->sock->flags);
 545                         transport->inet->sk_write_pending++;
 546                         /* ...and wait for more buffer space */
 547                         xprt_wait_for_buffer_space(task, xs_nospace_callback);
 548                 }
 549         } else {
 550                 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 551                 task->tk_status = -ENOTCONN;
 552         }
 553
 554         spin_unlock_bh(&xprt->transport_lock);
 555 }
 556
 557 /**
 558  * xs_udp_send_request - write an RPC request to a UDP socket
 559  * @task: address of RPC task that manages the state of an RPC request
 560  *
 561  * Return values:
 562  *        0:    The request has been sent
 563  *   EAGAIN:    The socket was blocked, please call again later to
 564  *              complete the request
 565  * ENOTCONN:    Caller needs to invoke connect logic then call again
 566  *    other:    Some other error occured, the request was not sent
 567  */
 568 static int xs_udp_send_request(struct rpc_task *task)
 569 {
 570         struct rpc_rqst *req = task->tk_rqstp;
 571         struct rpc_xprt *xprt = req->rq_xprt;
 572         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 573         struct xdr_buf *xdr = &req->rq_snd_buf;
 574         int status;
 575
 576         xs_pktdump("packet data:",
 577                                 req->rq_svec->iov_base,
 578                                 req->rq_svec->iov_len);
 579
 580         if (!xprt_bound(xprt))
 581                 return -ENOTCONN;
 582         status = xs_sendpages(transport->sock,
 583                               xs_addr(xprt),
 584                               xprt->addrlen, xdr,
 585                               req->rq_bytes_sent);
 586
 587         dprintk("RPC:       xs_udp_send_request(%u) = %d\n",
 588                         xdr->len - req->rq_bytes_sent, status);
 589
 590         if (status >= 0) {
 591                 task->tk_bytes_sent += status;
 592                 if (status >= req->rq_slen)
 593                         return 0;
 594                 /* Still some bytes left; set up for a retry later. */
 595                 status = -EAGAIN;
 596         }
 597         if (!transport->sock)
 598                 goto out;
 599
 600         switch (status) {
 601         case -ENOTSOCK:
 602                 status = -ENOTCONN;
 603                 /* Should we call xs_close() here? */
 604                 break;
 605         case -EAGAIN:
 606                 xs_nospace(task);
 607                 break;
 608         default:
 609                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
 610                         -status);
 611         case -ENETUNREACH:
 612         case -EPIPE:
 613         case -ECONNREFUSED:
 614                 /* When the server has died, an ICMP port unreachable message
 615                  * prompts ECONNREFUSED. */
 616                 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 617         }
 618 out:
 619         return status;
 620 }
 621
 622 /**
 623  * xs_tcp_shutdown - gracefully shut down a TCP socket
 624  * @xprt: transport
 625  *
 626  * Initiates a graceful shutdown of the TCP socket by calling the
 627  * equivalent of shutdown(SHUT_WR);
 628  */
 629 static void xs_tcp_shutdown(struct rpc_xprt *xprt)
 630 {
 631         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 632         struct socket *sock = transport->sock;
 633
 634         if (sock != NULL)
 635                 kernel_sock_shutdown(sock, SHUT_WR);
 636 }
 637
 638 static inline void xs_encode_tcp_record_marker(struct xdr_buf *buf)
 639 {
 640         u32 reclen = buf->len - sizeof(rpc_fraghdr);
 641         rpc_fraghdr *base = buf->head[0].iov_base;
 642         *base = htonl(RPC_LAST_STREAM_FRAGMENT | reclen);
 643 }
 644
 645 /**
 646  * xs_tcp_send_request - write an RPC request to a TCP socket
 647  * @task: address of RPC task that manages the state of an RPC request
 648  *
 649  * Return values:
 650  *        0:    The request has been sent
 651  *   EAGAIN:    The socket was blocked, please call again later to
 652  *              complete the request
 653  * ENOTCONN:    Caller needs to invoke connect logic then call again
 654  *    other:    Some other error occured, the request was not sent
 655  *
 656  * XXX: In the case of soft timeouts, should we eventually give up
 657  *      if sendmsg is not able to make progress?
 658  */
 659 static int xs_tcp_send_request(struct rpc_task *task)
 660 {
 661         struct rpc_rqst *req = task->tk_rqstp;
 662         struct rpc_xprt *xprt = req->rq_xprt;
 663         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 664         struct xdr_buf *xdr = &req->rq_snd_buf;
 665         int status;
 666
 667         xs_encode_tcp_record_marker(&req->rq_snd_buf);
 668
 669         xs_pktdump("packet data:",
 670                                 req->rq_svec->iov_base,
 671                                 req->rq_svec->iov_len);
 672
 673         /* Continue transmitting the packet/record. We must be careful
 674          * to cope with writespace callbacks arriving _after_ we have
 675          * called sendmsg(). */
 676         while (1) {
 677                 status = xs_sendpages(transport->sock,
 678                                         NULL, 0, xdr, req->rq_bytes_sent);
 679
 680                 dprintk("RPC:       xs_tcp_send_request(%u) = %d\n",
 681                                 xdr->len - req->rq_bytes_sent, status);
 682
 683                 if (unlikely(status < 0))
 684                         break;
 685
 686                 /* If we've sent the entire packet, immediately
 687                  * reset the count of bytes sent. */
 688                 req->rq_bytes_sent += status;
 689                 task->tk_bytes_sent += status;
 690                 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
 691                         req->rq_bytes_sent = 0;
 692                         return 0;
 693                 }
 694
 695                 if (status != 0)
 696                         continue;
 697                 status = -EAGAIN;
 698                 break;
 699         }
 700         if (!transport->sock)
 701                 goto out;
 702
 703         switch (status) {
 704         case -ENOTSOCK:
 705                 status = -ENOTCONN;
 706                 /* Should we call xs_close() here? */
 707                 break;
 708         case -EAGAIN:
 709                 xs_nospace(task);
 710                 break;
 711         default:
 712                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
 713                         -status);
 714         case -ECONNRESET:
 715                 xs_tcp_shutdown(xprt);
 716         case -ECONNREFUSED:
 717         case -ENOTCONN:
 718         case -EPIPE:
 719                 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 720         }
 721 out:
 722         return status;
 723 }
 724
 725 /**
 726  * xs_tcp_release_xprt - clean up after a tcp transmission
 727  * @xprt: transport
 728  * @task: rpc task
 729  *
 730  * This cleans up if an error causes us to abort the transmission of a request.
 731  * In this case, the socket may need to be reset in order to avoid confusing
 732  * the server.
 733  */
 734 static void xs_tcp_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
 735 {
 736         struct rpc_rqst *req;
 737
 738         if (task != xprt->snd_task)
 739                 return;
 740         if (task == NULL)
 741                 goto out_release;
 742         req = task->tk_rqstp;
 743         if (req->rq_bytes_sent == 0)
 744                 goto out_release;
 745         if (req->rq_bytes_sent == req->rq_snd_buf.len)
 746                 goto out_release;
 747         set_bit(XPRT_CLOSE_WAIT, &task->tk_xprt->state);
 748 out_release:
 749         xprt_release_xprt(xprt, task);
 750 }
 751
 752 static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk)
 753 {
 754         transport->old_data_ready = sk->sk_data_ready;
 755         transport->old_state_change = sk->sk_state_change;
 756         transport->old_write_space = sk->sk_write_space;
 757         transport->old_error_report = sk->sk_error_report;
 758 }
 759
 760 static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk)
 761 {
 762         sk->sk_data_ready = transport->old_data_ready;
 763         sk->sk_state_change = transport->old_state_change;
 764         sk->sk_write_space = transport->old_write_space;
 765         sk->sk_error_report = transport->old_error_report;
 766 }
 767
 768 static void xs_reset_transport(struct sock_xprt *transport)
 769 {
 770         struct socket *sock = transport->sock;
 771         struct sock *sk = transport->inet;
 772
 773         if (sk == NULL)
 774                 return;
 775
 776         write_lock_bh(&sk->sk_callback_lock);
 777         transport->inet = NULL;
 778         transport->sock = NULL;
 779
 780         sk->sk_user_data = NULL;
 781
 782         xs_restore_old_callbacks(transport, sk);
 783         write_unlock_bh(&sk->sk_callback_lock);
 784
 785         sk->sk_no_check = 0;
 786
 787         sock_release(sock);
 788 }
 789
 790 /**
 791  * xs_close - close a socket
 792  * @xprt: transport
 793  *
 794  * This is used when all requests are complete; ie, no DRC state remains
 795  * on the server we want to save.
 796  */
 797 static void xs_close(struct rpc_xprt *xprt)
 798 {
 799         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 800
 801         dprintk("RPC:       xs_close xprt %p\n", xprt);
 802
 803         xs_reset_transport(transport);
 804
 805         smp_mb__before_clear_bit();
 806         clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
 807         clear_bit(XPRT_CLOSING, &xprt->state);
 808         smp_mb__after_clear_bit();
 809         xprt_disconnect_done(xprt);
 810 }
 811
 812 /**
 813  * xs_destroy - prepare to shutdown a transport
 814  * @xprt: doomed transport
 815  *
 816  */
 817 static void xs_destroy(struct rpc_xprt *xprt)
 818 {
 819         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 820
 821         dprintk("RPC:       xs_destroy xprt %p\n", xprt);
 822
 823         cancel_rearming_delayed_work(&transport->connect_worker);
 824
 825         xs_close(xprt);
 826         xs_free_peer_addresses(xprt);
 827         kfree(xprt->slot);
 828         kfree(xprt);
 829         module_put(THIS_MODULE);
 830 }
 831
 832 static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
 833 {
 834         return (struct rpc_xprt *) sk->sk_user_data;
 835 }
 836
 837 /**
 838  * xs_udp_data_ready - "data ready" callback for UDP sockets
 839  * @sk: socket with data to read
 840  * @len: how much data to read
 841  *
 842  */
 843 static void xs_udp_data_ready(struct sock *sk, int len)
 844 {
 845         struct rpc_task *task;
 846         struct rpc_xprt *xprt;
 847         struct rpc_rqst *rovr;
 848         struct sk_buff *skb;
 849         int err, repsize, copied;
 850         u32 _xid;
 851         __be32 *xp;
 852
 853         read_lock(&sk->sk_callback_lock);
 854         dprintk("RPC:       xs_udp_data_ready...\n");
 855         if (!(xprt = xprt_from_sock(sk)))
 856                 goto out;
 857
 858         if ((skb = skb_recv_datagram(sk, 0, 1, &err)) == NULL)
 859                 goto out;
 860
 861         if (xprt->shutdown)
 862                 goto dropit;
 863
 864         repsize = skb->len - sizeof(struct udphdr);
 865         if (repsize < 4) {
 866                 dprintk("RPC:       impossible RPC reply size %d!\n", repsize);
 867                 goto dropit;
 868         }
 869
 870         /* Copy the XID from the skb... */
 871         xp = skb_header_pointer(skb, sizeof(struct udphdr),
 872                                 sizeof(_xid), &_xid);
 873         if (xp == NULL)
 874                 goto dropit;
 875
 876         /* Look up and lock the request corresponding to the given XID */
 877         spin_lock(&xprt->transport_lock);
 878         rovr = xprt_lookup_rqst(xprt, *xp);
 879         if (!rovr)
 880                 goto out_unlock;
 881         task = rovr->rq_task;
 882
 883         if ((copied = rovr->rq_private_buf.buflen) > repsize)
 884                 copied = repsize;
 885
 886         /* Suck it into the iovec, verify checksum if not done by hw. */
 887         if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
 888                 UDPX_INC_STATS_BH(sk, UDP_MIB_INERRORS);
 889                 goto out_unlock;
 890         }
 891
 892         UDPX_INC_STATS_BH(sk, UDP_MIB_INDATAGRAMS);
 893
 894         /* Something worked... */
 895         dst_confirm(skb->dst);
 896
 897         xprt_adjust_cwnd(task, copied);
 898         xprt_update_rtt(task);
 899         xprt_complete_rqst(task, copied);
 900
 901  out_unlock:
 902         spin_unlock(&xprt->transport_lock);
 903  dropit:
 904         skb_free_datagram(sk, skb);
 905  out:
 906         read_unlock(&sk->sk_callback_lock);
 907 }
 908
 909 static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
 910 {
 911         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 912         size_t len, used;
 913         char *p;
 914
 915         p = ((char *) &transport->tcp_fraghdr) + transport->tcp_offset;
 916         len = sizeof(transport->tcp_fraghdr) - transport->tcp_offset;
 917         used = xdr_skb_read_bits(desc, p, len);
 918         transport->tcp_offset += used;
 919         if (used != len)
 920                 return;
 921
 922         transport->tcp_reclen = ntohl(transport->tcp_fraghdr);
 923         if (transport->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
 924                 transport->tcp_flags |= TCP_RCV_LAST_FRAG;
 925         else
 926                 transport->tcp_flags &= ~TCP_RCV_LAST_FRAG;
 927         transport->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
 928
 929         transport->tcp_flags &= ~TCP_RCV_COPY_FRAGHDR;
 930         transport->tcp_offset = 0;
 931
 932         /* Sanity check of the record length */
 933         if (unlikely(transport->tcp_reclen < 4)) {
 934                 dprintk("RPC:       invalid TCP record fragment length\n");
 935                 xprt_force_disconnect(xprt);
 936                 return;
 937         }
 938         dprintk("RPC:       reading TCP record fragment of length %d\n",
 939                         transport->tcp_reclen);
 940 }
 941
 942 static void xs_tcp_check_fraghdr(struct sock_xprt *transport)
 943 {
 944         if (transport->tcp_offset == transport->tcp_reclen) {
 945                 transport->tcp_flags |= TCP_RCV_COPY_FRAGHDR;
 946                 transport->tcp_offset = 0;
 947                 if (transport->tcp_flags & TCP_RCV_LAST_FRAG) {
 948                         transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
 949                         transport->tcp_flags |= TCP_RCV_COPY_XID;
 950                         transport->tcp_copied = 0;
 951                 }
 952         }
 953 }
 954
 955 static inline void xs_tcp_read_xid(struct sock_xprt *transport, struct xdr_skb_reader *desc)
 956 {
 957         size_t len, used;
 958         char *p;
 959
 960         len = sizeof(transport->tcp_xid) - transport->tcp_offset;
 961         dprintk("RPC:       reading XID (%Zu bytes)\n", len);
 962         p = ((char *) &transport->tcp_xid) + transport->tcp_offset;
 963         used = xdr_skb_read_bits(desc, p, len);
 964         transport->tcp_offset += used;
 965         if (used != len)
 966                 return;
 967         transport->tcp_flags &= ~TCP_RCV_COPY_XID;
 968         transport->tcp_flags |= TCP_RCV_COPY_DATA;
 969         transport->tcp_copied = 4;
 970         dprintk("RPC:       reading reply for XID %08x\n",
 971                         ntohl(transport->tcp_xid));
 972         xs_tcp_check_fraghdr(transport);
 973 }
 974
 975 static inline void xs_tcp_read_request(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
 976 {
 977         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 978         struct rpc_rqst *req;
 979         struct xdr_buf *rcvbuf;
 980         size_t len;
 981         ssize_t r;
 982
 983         /* Find and lock the request corresponding to this xid */
 984         spin_lock(&xprt->transport_lock);
 985         req = xprt_lookup_rqst(xprt, transport->tcp_xid);
 986         if (!req) {
 987                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
 988                 dprintk("RPC:       XID %08x request not found!\n",
 989                                 ntohl(transport->tcp_xid));
 990                 spin_unlock(&xprt->transport_lock);
 991                 return;
 992         }
 993
 994         rcvbuf = &req->rq_private_buf;
 995         len = desc->count;
 996         if (len > transport->tcp_reclen - transport->tcp_offset) {
 997                 struct xdr_skb_reader my_desc;
 998
 999                 len = transport->tcp_reclen - transport->tcp_offset;
1000                 memcpy(&my_desc, desc, sizeof(my_desc));
1001                 my_desc.count = len;
1002                 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
1003                                           &my_desc, xdr_skb_read_bits);
1004                 desc->count -= r;
1005                 desc->offset += r;
1006         } else
1007                 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
1008                                           desc, xdr_skb_read_bits);
1009
1010         if (r > 0) {
1011                 transport->tcp_copied += r;
1012                 transport->tcp_offset += r;
1013         }
1014         if (r != len) {
1015                 /* Error when copying to the receive buffer,
1016                  * usually because we weren't able to allocate
1017                  * additional buffer pages. All we can do now
1018                  * is turn off TCP_RCV_COPY_DATA, so the request
1019                  * will not receive any additional updates,
1020                  * and time out.
1021                  * Any remaining data from this record will
1022                  * be discarded.
1023                  */
1024                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1025                 dprintk("RPC:       XID %08x truncated request\n",
1026                                 ntohl(transport->tcp_xid));
1027                 dprintk("RPC:       xprt = %p, tcp_copied = %lu, "
1028                                 "tcp_offset = %u, tcp_reclen = %u\n",
1029                                 xprt, transport->tcp_copied,
1030                                 transport->tcp_offset, transport->tcp_reclen);
1031                 goto out;
1032         }
1033
1034         dprintk("RPC:       XID %08x read %Zd bytes\n",
1035                         ntohl(transport->tcp_xid), r);
1036         dprintk("RPC:       xprt = %p, tcp_copied = %lu, tcp_offset = %u, "
1037                         "tcp_reclen = %u\n", xprt, transport->tcp_copied,
1038                         transport->tcp_offset, transport->tcp_reclen);
1039
1040         if (transport->tcp_copied == req->rq_private_buf.buflen)
1041                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1042         else if (transport->tcp_offset == transport->tcp_reclen) {
1043                 if (transport->tcp_flags & TCP_RCV_LAST_FRAG)
1044                         transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1045         }
1046
1047 out:
1048         if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1049                 xprt_complete_rqst(req->rq_task, transport->tcp_copied);
1050         spin_unlock(&xprt->transport_lock);
1051         xs_tcp_check_fraghdr(transport);
1052 }
1053
1054 static inline void xs_tcp_read_discard(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1055 {
1056         size_t len;
1057
1058         len = transport->tcp_reclen - transport->tcp_offset;
1059         if (len > desc->count)
1060                 len = desc->count;
1061         desc->count -= len;
1062         desc->offset += len;
1063         transport->tcp_offset += len;
1064         dprintk("RPC:       discarded %Zu bytes\n", len);
1065         xs_tcp_check_fraghdr(transport);
1066 }
1067
1068 static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
1069 {
1070         struct rpc_xprt *xprt = rd_desc->arg.data;
1071         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1072         struct xdr_skb_reader desc = {
1073                 .skb    = skb,
1074                 .offset = offset,
1075                 .count  = len,
1076         };
1077
1078         dprintk("RPC:       xs_tcp_data_recv started\n");
1079         do {
1080                 /* Read in a new fragment marker if necessary */
1081                 /* Can we ever really expect to get completely empty fragments? */
1082                 if (transport->tcp_flags & TCP_RCV_COPY_FRAGHDR) {
1083                         xs_tcp_read_fraghdr(xprt, &desc);
1084                         continue;
1085                 }
1086                 /* Read in the xid if necessary */
1087                 if (transport->tcp_flags & TCP_RCV_COPY_XID) {
1088                         xs_tcp_read_xid(transport, &desc);
1089                         continue;
1090                 }
1091                 /* Read in the request data */
1092                 if (transport->tcp_flags & TCP_RCV_COPY_DATA) {
1093                         xs_tcp_read_request(xprt, &desc);
1094                         continue;
1095                 }
1096                 /* Skip over any trailing bytes on short reads */
1097                 xs_tcp_read_discard(transport, &desc);
1098         } while (desc.count);
1099         dprintk("RPC:       xs_tcp_data_recv done\n");
1100         return len - desc.count;
1101 }
1102
1103 /**
1104  * xs_tcp_data_ready - "data ready" callback for TCP sockets
1105  * @sk: socket with data to read
1106  * @bytes: how much data to read
1107  *
1108  */
1109 static void xs_tcp_data_ready(struct sock *sk, int bytes)
1110 {
1111         struct rpc_xprt *xprt;
1112         read_descriptor_t rd_desc;
1113         int read;
1114
1115         dprintk("RPC:       xs_tcp_data_ready...\n");
1116
1117         read_lock(&sk->sk_callback_lock);
1118         if (!(xprt = xprt_from_sock(sk)))
1119                 goto out;
1120         if (xprt->shutdown)
1121                 goto out;
1122
1123         /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
1124         rd_desc.arg.data = xprt;
1125         do {
1126                 rd_desc.count = 65536;
1127                 read = tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
1128         } while (read > 0);
1129 out:
1130         read_unlock(&sk->sk_callback_lock);
1131 }
1132
1133 /**
1134  * xs_tcp_state_change - callback to handle TCP socket state changes
1135  * @sk: socket whose state has changed
1136  *
1137  */
1138 static void xs_tcp_state_change(struct sock *sk)
1139 {
1140         struct rpc_xprt *xprt;
1141
1142         read_lock(&sk->sk_callback_lock);
1143         if (!(xprt = xprt_from_sock(sk)))
1144                 goto out;
1145         dprintk("RPC:       xs_tcp_state_change client %p...\n", xprt);
1146         dprintk("RPC:       state %x conn %d dead %d zapped %d\n",
1147                         sk->sk_state, xprt_connected(xprt),
1148                         sock_flag(sk, SOCK_DEAD),
1149                         sock_flag(sk, SOCK_ZAPPED));
1150
1151         switch (sk->sk_state) {
1152         case TCP_ESTABLISHED:
1153                 spin_lock_bh(&xprt->transport_lock);
1154                 if (!xprt_test_and_set_connected(xprt)) {
1155                         struct sock_xprt *transport = container_of(xprt,
1156                                         struct sock_xprt, xprt);
1157
1158                         /* Reset TCP record info */
1159                         transport->tcp_offset = 0;
1160                         transport->tcp_reclen = 0;
1161                         transport->tcp_copied = 0;
1162                         transport->tcp_flags =
1163                                 TCP_RCV_COPY_FRAGHDR | TCP_RCV_COPY_XID;
1164
1165                         xprt_wake_pending_tasks(xprt, -EAGAIN);
1166                 }
1167                 spin_unlock_bh(&xprt->transport_lock);
1168                 break;
1169         case TCP_FIN_WAIT1:
1170                 /* The client initiated a shutdown of the socket */
1171                 xprt->connect_cookie++;
1172                 xprt->reestablish_timeout = 0;
1173                 set_bit(XPRT_CLOSING, &xprt->state);
1174                 smp_mb__before_clear_bit();
1175                 clear_bit(XPRT_CONNECTED, &xprt->state);
1176                 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1177                 smp_mb__after_clear_bit();
1178                 break;
1179         case TCP_CLOSE_WAIT:
1180                 /* The server initiated a shutdown of the socket */
1181                 xprt_force_disconnect(xprt);
1182         case TCP_SYN_SENT:
1183                 xprt->connect_cookie++;
1184         case TCP_CLOSING:
1185                 /*
1186                  * If the server closed down the connection, make sure that
1187                  * we back off before reconnecting
1188                  */
1189                 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1190                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1191                 break;
1192         case TCP_LAST_ACK:
1193                 set_bit(XPRT_CLOSING, &xprt->state);
1194                 smp_mb__before_clear_bit();
1195                 clear_bit(XPRT_CONNECTED, &xprt->state);
1196                 smp_mb__after_clear_bit();
1197                 break;
1198         case TCP_CLOSE:
1199                 smp_mb__before_clear_bit();
1200                 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1201                 clear_bit(XPRT_CLOSING, &xprt->state);
1202                 smp_mb__after_clear_bit();
1203                 /* Mark transport as closed and wake up all pending tasks */
1204                 xprt_disconnect_done(xprt);
1205         }
1206  out:
1207         read_unlock(&sk->sk_callback_lock);
1208 }
1209
1210 /**
1211  * xs_error_report - callback mainly for catching socket errors
1212  * @sk: socket
1213  */
1214 static void xs_error_report(struct sock *sk)
1215 {
1216         struct rpc_xprt *xprt;
1217
1218         read_lock(&sk->sk_callback_lock);
1219         if (!(xprt = xprt_from_sock(sk)))
1220                 goto out;
1221         dprintk("RPC:       %s client %p...\n"
1222                         "RPC:       error %d\n",
1223                         __func__, xprt, sk->sk_err);
1224         xprt_wake_pending_tasks(xprt, -EAGAIN);
1225 out:
1226         read_unlock(&sk->sk_callback_lock);
1227 }
1228
1229 /**
1230  * xs_udp_write_space - callback invoked when socket buffer space
1231  *                             becomes available
1232  * @sk: socket whose state has changed
1233  *
1234  * Called when more output buffer space is available for this socket.
1235  * We try not to wake our writers until they can make "significant"
1236  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1237  * with a bunch of small requests.
1238  */
1239 static void xs_udp_write_space(struct sock *sk)
1240 {
1241         read_lock(&sk->sk_callback_lock);
1242
1243         /* from net/core/sock.c:sock_def_write_space */
1244         if (sock_writeable(sk)) {
1245                 struct socket *sock;
1246                 struct rpc_xprt *xprt;
1247
1248                 if (unlikely(!(sock = sk->sk_socket)))
1249                         goto out;
1250                 clear_bit(SOCK_NOSPACE, &sock->flags);
1251
1252                 if (unlikely(!(xprt = xprt_from_sock(sk))))
1253                         goto out;
1254                 if (test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags) == 0)
1255                         goto out;
1256
1257                 xprt_write_space(xprt);
1258         }
1259
1260  out:
1261         read_unlock(&sk->sk_callback_lock);
1262 }
1263
1264 /**
1265  * xs_tcp_write_space - callback invoked when socket buffer space
1266  *                             becomes available
1267  * @sk: socket whose state has changed
1268  *
1269  * Called when more output buffer space is available for this socket.
1270  * We try not to wake our writers until they can make "significant"
1271  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1272  * with a bunch of small requests.
1273  */
1274 static void xs_tcp_write_space(struct sock *sk)
1275 {
1276         read_lock(&sk->sk_callback_lock);
1277
1278         /* from net/core/stream.c:sk_stream_write_space */
1279         if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk)) {
1280                 struct socket *sock;
1281                 struct rpc_xprt *xprt;
1282
1283                 if (unlikely(!(sock = sk->sk_socket)))
1284                         goto out;
1285                 clear_bit(SOCK_NOSPACE, &sock->flags);
1286
1287                 if (unlikely(!(xprt = xprt_from_sock(sk))))
1288                         goto out;
1289                 if (test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags) == 0)
1290                         goto out;
1291
1292                 xprt_write_space(xprt);
1293         }
1294
1295  out:
1296         read_unlock(&sk->sk_callback_lock);
1297 }
1298
1299 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
1300 {
1301         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1302         struct sock *sk = transport->inet;
1303
1304         if (transport->rcvsize) {
1305                 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1306                 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
1307         }
1308         if (transport->sndsize) {
1309                 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1310                 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1311                 sk->sk_write_space(sk);
1312         }
1313 }
1314
1315 /**
1316  * xs_udp_set_buffer_size - set send and receive limits
1317  * @xprt: generic transport
1318  * @sndsize: requested size of send buffer, in bytes
1319  * @rcvsize: requested size of receive buffer, in bytes
1320  *
1321  * Set socket send and receive buffer size limits.
1322  */
1323 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1324 {
1325         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1326
1327         transport->sndsize = 0;
1328         if (sndsize)
1329                 transport->sndsize = sndsize + 1024;
1330         transport->rcvsize = 0;
1331         if (rcvsize)
1332                 transport->rcvsize = rcvsize + 1024;
1333
1334         xs_udp_do_set_buffer_size(xprt);
1335 }
1336
1337 /**
1338  * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1339  * @task: task that timed out
1340  *
1341  * Adjust the congestion window after a retransmit timeout has occurred.
1342  */
1343 static void xs_udp_timer(struct rpc_task *task)
1344 {
1345         xprt_adjust_cwnd(task, -ETIMEDOUT);
1346 }
1347
1348 static unsigned short xs_get_random_port(void)
1349 {
1350         unsigned short range = xprt_max_resvport - xprt_min_resvport;
1351         unsigned short rand = (unsigned short) net_random() % range;
1352         return rand + xprt_min_resvport;
1353 }
1354
1355 /**
1356  * xs_set_port - reset the port number in the remote endpoint address
1357  * @xprt: generic transport
1358  * @port: new port number
1359  *
1360  */
1361 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1362 {
1363         struct sockaddr *addr = xs_addr(xprt);
1364
1365         dprintk("RPC:       setting port for xprt %p to %u\n", xprt, port);
1366
1367         switch (addr->sa_family) {
1368         case AF_INET:
1369                 ((struct sockaddr_in *)addr)->sin_port = htons(port);
1370                 break;
1371         case AF_INET6:
1372                 ((struct sockaddr_in6 *)addr)->sin6_port = htons(port);
1373                 break;
1374         default:
1375                 BUG();
1376         }
1377 }
1378
1379 static unsigned short xs_get_srcport(struct sock_xprt *transport, struct socket *sock)
1380 {
1381         unsigned short port = transport->port;
1382
1383         if (port == 0 && transport->xprt.resvport)
1384                 port = xs_get_random_port();
1385         return port;
1386 }
1387
1388 static unsigned short xs_next_srcport(struct sock_xprt *transport, struct socket *sock, unsigned short port)
1389 {
1390         if (transport->port != 0)
1391                 transport->port = 0;
1392         if (!transport->xprt.resvport)
1393                 return 0;
1394         if (port <= xprt_min_resvport || port > xprt_max_resvport)
1395                 return xprt_max_resvport;
1396         return --port;
1397 }
1398
1399 static int xs_bind4(struct sock_xprt *transport, struct socket *sock)
1400 {
1401         struct sockaddr_in myaddr = {
1402                 .sin_family = AF_INET,
1403         };
1404         struct sockaddr_in *sa;
1405         int err, nloop = 0;
1406         unsigned short port = xs_get_srcport(transport, sock);
1407         unsigned short last;
1408
1409         sa = (struct sockaddr_in *)&transport->addr;
1410         myaddr.sin_addr = sa->sin_addr;
1411         do {
1412                 myaddr.sin_port = htons(port);
1413                 err = kernel_bind(sock, (struct sockaddr *) &myaddr,
1414                                                 sizeof(myaddr));
1415                 if (port == 0)
1416                         break;
1417                 if (err == 0) {
1418                         transport->port = port;
1419                         break;
1420                 }
1421                 last = port;
1422                 port = xs_next_srcport(transport, sock, port);
1423                 if (port > last)
1424                         nloop++;
1425         } while (err == -EADDRINUSE && nloop != 2);
1426         dprintk("RPC:       %s %pI4:%u: %s (%d)\n",
1427                         __func__, &myaddr.sin_addr,
1428                         port, err ? "failed" : "ok", err);
1429         return err;
1430 }
1431
1432 static int xs_bind6(struct sock_xprt *transport, struct socket *sock)
1433 {
1434         struct sockaddr_in6 myaddr = {
1435                 .sin6_family = AF_INET6,
1436         };
1437         struct sockaddr_in6 *sa;
1438         int err, nloop = 0;
1439         unsigned short port = xs_get_srcport(transport, sock);
1440         unsigned short last;
1441
1442         sa = (struct sockaddr_in6 *)&transport->addr;
1443         myaddr.sin6_addr = sa->sin6_addr;
1444         do {
1445                 myaddr.sin6_port = htons(port);
1446                 err = kernel_bind(sock, (struct sockaddr *) &myaddr,
1447                                                 sizeof(myaddr));
1448                 if (port == 0)
1449                         break;
1450                 if (err == 0) {
1451                         transport->port = port;
1452                         break;
1453                 }
1454                 last = port;
1455                 port = xs_next_srcport(transport, sock, port);
1456                 if (port > last)
1457                         nloop++;
1458         } while (err == -EADDRINUSE && nloop != 2);
1459         dprintk("RPC:       xs_bind6 %pI6:%u: %s (%d)\n",
1460                 &myaddr.sin6_addr, port, err ? "failed" : "ok", err);
1461         return err;
1462 }
1463
1464 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1465 static struct lock_class_key xs_key[2];
1466 static struct lock_class_key xs_slock_key[2];
1467
1468 static inline void xs_reclassify_socket4(struct socket *sock)
1469 {
1470         struct sock *sk = sock->sk;
1471
1472         BUG_ON(sock_owned_by_user(sk));
1473         sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1474                 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1475 }
1476
1477 static inline void xs_reclassify_socket6(struct socket *sock)
1478 {
1479         struct sock *sk = sock->sk;
1480
1481         BUG_ON(sock_owned_by_user(sk));
1482         sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1483                 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
1484 }
1485 #else
1486 static inline void xs_reclassify_socket4(struct socket *sock)
1487 {
1488 }
1489
1490 static inline void xs_reclassify_socket6(struct socket *sock)
1491 {
1492 }
1493 #endif
1494
1495 static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
1496 {
1497         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1498
1499         if (!transport->inet) {
1500                 struct sock *sk = sock->sk;
1501
1502                 write_lock_bh(&sk->sk_callback_lock);
1503
1504                 xs_save_old_callbacks(transport, sk);
1505
1506                 sk->sk_user_data = xprt;
1507                 sk->sk_data_ready = xs_udp_data_ready;
1508                 sk->sk_write_space = xs_udp_write_space;
1509                 sk->sk_error_report = xs_error_report;
1510                 sk->sk_no_check = UDP_CSUM_NORCV;
1511                 sk->sk_allocation = GFP_ATOMIC;
1512
1513                 xprt_set_connected(xprt);
1514
1515                 /* Reset to new socket */
1516                 transport->sock = sock;
1517                 transport->inet = sk;
1518
1519                 write_unlock_bh(&sk->sk_callback_lock);
1520         }
1521         xs_udp_do_set_buffer_size(xprt);
1522 }
1523
1524 /**
1525  * xs_udp_connect_worker4 - set up a UDP socket
1526  * @work: RPC transport to connect
1527  *
1528  * Invoked by a work queue tasklet.
1529  */
1530 static void xs_udp_connect_worker4(struct work_struct *work)
1531 {
1532         struct sock_xprt *transport =
1533                 container_of(work, struct sock_xprt, connect_worker.work);
1534         struct rpc_xprt *xprt = &transport->xprt;
1535         struct socket *sock = transport->sock;
1536         int err, status = -EIO;
1537
1538         if (xprt->shutdown)
1539                 goto out;
1540
1541         /* Start by resetting any existing state */
1542         xs_reset_transport(transport);
1543
1544         err = sock_create_kern(PF_INET, SOCK_DGRAM, IPPROTO_UDP, &sock);
1545         if (err < 0) {
1546                 dprintk("RPC:       can't create UDP transport socket (%d).\n", -err);
1547                 goto out;
1548         }
1549         xs_reclassify_socket4(sock);
1550
1551         if (xs_bind4(transport, sock)) {
1552                 sock_release(sock);
1553                 goto out;
1554         }
1555
1556         dprintk("RPC:       worker connecting xprt %p to address: %s\n",
1557                         xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1558
1559         xs_udp_finish_connecting(xprt, sock);
1560         status = 0;
1561 out:
1562         xprt_wake_pending_tasks(xprt, status);
1563         xprt_clear_connecting(xprt);
1564 }
1565
1566 /**
1567  * xs_udp_connect_worker6 - set up a UDP socket
1568  * @work: RPC transport to connect
1569  *
1570  * Invoked by a work queue tasklet.
1571  */
1572 static void xs_udp_connect_worker6(struct work_struct *work)
1573 {
1574         struct sock_xprt *transport =
1575                 container_of(work, struct sock_xprt, connect_worker.work);
1576         struct rpc_xprt *xprt = &transport->xprt;
1577         struct socket *sock = transport->sock;
1578         int err, status = -EIO;
1579
1580         if (xprt->shutdown)
1581                 goto out;
1582
1583         /* Start by resetting any existing state */
1584         xs_reset_transport(transport);
1585
1586         err = sock_create_kern(PF_INET6, SOCK_DGRAM, IPPROTO_UDP, &sock);
1587         if (err < 0) {
1588                 dprintk("RPC:       can't create UDP transport socket (%d).\n", -err);
1589                 goto out;
1590         }
1591         xs_reclassify_socket6(sock);
1592
1593         if (xs_bind6(transport, sock) < 0) {
1594                 sock_release(sock);
1595                 goto out;
1596         }
1597
1598         dprintk("RPC:       worker connecting xprt %p to address: %s\n",
1599                         xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1600
1601         xs_udp_finish_connecting(xprt, sock);
1602         status = 0;
1603 out:
1604         xprt_wake_pending_tasks(xprt, status);
1605         xprt_clear_connecting(xprt);
1606 }
1607
1608 /*
1609  * We need to preserve the port number so the reply cache on the server can
1610  * find our cached RPC replies when we get around to reconnecting.
1611  */
1612 static void xs_abort_connection(struct rpc_xprt *xprt, struct sock_xprt *transport)
1613 {
1614         int result;
1615         struct sockaddr any;
1616
1617         dprintk("RPC:       disconnecting xprt %p to reuse port\n", xprt);
1618
1619         /*
1620          * Disconnect the transport socket by doing a connect operation
1621          * with AF_UNSPEC.  This should return immediately...
1622          */
1623         memset(&any, 0, sizeof(any));
1624         any.sa_family = AF_UNSPEC;
1625         result = kernel_connect(transport->sock, &any, sizeof(any), 0);
1626         if (result)
1627                 dprintk("RPC:       AF_UNSPEC connect return code %d\n",
1628                                 result);
1629 }
1630
1631 static void xs_tcp_reuse_connection(struct rpc_xprt *xprt, struct sock_xprt *transport)
1632 {
1633         unsigned int state = transport->inet->sk_state;
1634
1635         if (state == TCP_CLOSE && transport->sock->state == SS_UNCONNECTED)
1636                 return;
1637         if ((1 << state) & (TCPF_ESTABLISHED|TCPF_SYN_SENT))
1638                 return;
1639         xs_abort_connection(xprt, transport);
1640 }
1641
1642 static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
1643 {
1644         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1645
1646         if (!transport->inet) {
1647                 struct sock *sk = sock->sk;
1648
1649                 write_lock_bh(&sk->sk_callback_lock);
1650
1651                 xs_save_old_callbacks(transport, sk);
1652
1653                 sk->sk_user_data = xprt;
1654                 sk->sk_data_ready = xs_tcp_data_ready;
1655                 sk->sk_state_change = xs_tcp_state_change;
1656                 sk->sk_write_space = xs_tcp_write_space;
1657                 sk->sk_error_report = xs_error_report;
1658                 sk->sk_allocation = GFP_ATOMIC;
1659
1660                 /* socket options */
1661                 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
1662                 sock_reset_flag(sk, SOCK_LINGER);
1663                 tcp_sk(sk)->linger2 = 0;
1664                 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
1665
1666                 xprt_clear_connected(xprt);
1667
1668                 /* Reset to new socket */
1669                 transport->sock = sock;
1670                 transport->inet = sk;
1671
1672                 write_unlock_bh(&sk->sk_callback_lock);
1673         }
1674
1675         if (!xprt_bound(xprt))
1676                 return -ENOTCONN;
1677
1678         /* Tell the socket layer to start connecting... */
1679         xprt->stat.connect_count++;
1680         xprt->stat.connect_start = jiffies;
1681         return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
1682 }
1683
1684 /**
1685  * xs_tcp_connect_worker4 - connect a TCP socket to a remote endpoint
1686  * @work: RPC transport to connect
1687  *
1688  * Invoked by a work queue tasklet.
1689  */
1690 static void xs_tcp_connect_worker4(struct work_struct *work)
1691 {
1692         struct sock_xprt *transport =
1693                 container_of(work, struct sock_xprt, connect_worker.work);
1694         struct rpc_xprt *xprt = &transport->xprt;
1695         struct socket *sock = transport->sock;
1696         int err, status = -EIO;
1697
1698         if (xprt->shutdown)
1699                 goto out;
1700
1701         if (!sock) {
1702                 /* start from scratch */
1703                 if ((err = sock_create_kern(PF_INET, SOCK_STREAM, IPPROTO_TCP, &sock)) < 0) {
1704                         dprintk("RPC:       can't create TCP transport socket (%d).\n", -err);
1705                         goto out;
1706                 }
1707                 xs_reclassify_socket4(sock);
1708
1709                 if (xs_bind4(transport, sock) < 0) {
1710                         sock_release(sock);
1711                         goto out;
1712                 }
1713         } else
1714                 /* "close" the socket, preserving the local port */
1715                 xs_tcp_reuse_connection(xprt, transport);
1716
1717         dprintk("RPC:       worker connecting xprt %p to address: %s\n",
1718                         xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1719
1720         status = xs_tcp_finish_connecting(xprt, sock);
1721         dprintk("RPC:       %p connect status %d connected %d sock state %d\n",
1722                         xprt, -status, xprt_connected(xprt),
1723                         sock->sk->sk_state);
1724         switch (status) {
1725         case -ECONNREFUSED:
1726         case -ECONNRESET:
1727         case -ENETUNREACH:
1728                 /* retry with existing socket, after a delay */
1729         case 0:
1730         case -EINPROGRESS:
1731         case -EALREADY:
1732                 goto out_clear;
1733         }
1734         /* get rid of existing socket, and retry */
1735         xs_tcp_shutdown(xprt);
1736         printk("%s: connect returned unhandled error %d\n",
1737                         __func__, status);
1738         status = -EAGAIN;
1739 out:
1740         xprt_wake_pending_tasks(xprt, status);
1741 out_clear:
1742         xprt_clear_connecting(xprt);
1743 }
1744
1745 /**
1746  * xs_tcp_connect_worker6 - connect a TCP socket to a remote endpoint
1747  * @work: RPC transport to connect
1748  *
1749  * Invoked by a work queue tasklet.
1750  */
1751 static void xs_tcp_connect_worker6(struct work_struct *work)
1752 {
1753         struct sock_xprt *transport =
1754                 container_of(work, struct sock_xprt, connect_worker.work);
1755         struct rpc_xprt *xprt = &transport->xprt;
1756         struct socket *sock = transport->sock;
1757         int err, status = -EIO;
1758
1759         if (xprt->shutdown)
1760                 goto out;
1761
1762         if (!sock) {
1763                 /* start from scratch */
1764                 if ((err = sock_create_kern(PF_INET6, SOCK_STREAM, IPPROTO_TCP, &sock)) < 0) {
1765                         dprintk("RPC:       can't create TCP transport socket (%d).\n", -err);
1766                         goto out;
1767                 }
1768                 xs_reclassify_socket6(sock);
1769
1770                 if (xs_bind6(transport, sock) < 0) {
1771                         sock_release(sock);
1772                         goto out;
1773                 }
1774         } else
1775                 /* "close" the socket, preserving the local port */
1776                 xs_tcp_reuse_connection(xprt, transport);
1777
1778         dprintk("RPC:       worker connecting xprt %p to address: %s\n",
1779                         xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1780
1781         status = xs_tcp_finish_connecting(xprt, sock);
1782         dprintk("RPC:       %p connect status %d connected %d sock state %d\n",
1783                         xprt, -status, xprt_connected(xprt), sock->sk->sk_state);
1784         switch (status) {
1785         case -ECONNREFUSED:
1786         case -ECONNRESET:
1787         case -ENETUNREACH:
1788                 /* retry with existing socket, after a delay */
1789         case 0:
1790         case -EINPROGRESS:
1791         case -EALREADY:
1792                 goto out_clear;
1793         }
1794         /* get rid of existing socket, and retry */
1795         xs_tcp_shutdown(xprt);
1796         printk("%s: connect returned unhandled error %d\n",
1797                         __func__, status);
1798         status = -EAGAIN;
1799 out:
1800         xprt_wake_pending_tasks(xprt, status);
1801 out_clear:
1802         xprt_clear_connecting(xprt);
1803 }
1804
1805 /**
1806  * xs_connect - connect a socket to a remote endpoint
1807  * @task: address of RPC task that manages state of connect request
1808  *
1809  * TCP: If the remote end dropped the connection, delay reconnecting.
1810  *
1811  * UDP socket connects are synchronous, but we use a work queue anyway
1812  * to guarantee that even unprivileged user processes can set up a
1813  * socket on a privileged port.
1814  *
1815  * If a UDP socket connect fails, the delay behavior here prevents
1816  * retry floods (hard mounts).
1817  */
1818 static void xs_connect(struct rpc_task *task)
1819 {
1820         struct rpc_xprt *xprt = task->tk_xprt;
1821         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1822
1823         if (xprt_test_and_set_connecting(xprt))
1824                 return;
1825
1826         if (transport->sock != NULL) {
1827                 dprintk("RPC:       xs_connect delayed xprt %p for %lu "
1828                                 "seconds\n",
1829                                 xprt, xprt->reestablish_timeout / HZ);
1830                 queue_delayed_work(rpciod_workqueue,
1831                                    &transport->connect_worker,
1832                                    xprt->reestablish_timeout);
1833                 xprt->reestablish_timeout <<= 1;
1834                 if (xprt->reestablish_timeout > XS_TCP_MAX_REEST_TO)
1835                         xprt->reestablish_timeout = XS_TCP_MAX_REEST_TO;
1836         } else {
1837                 dprintk("RPC:       xs_connect scheduled xprt %p\n", xprt);
1838                 queue_delayed_work(rpciod_workqueue,
1839                                    &transport->connect_worker, 0);
1840         }
1841 }
1842
1843 static void xs_tcp_connect(struct rpc_task *task)
1844 {
1845         struct rpc_xprt *xprt = task->tk_xprt;
1846
1847         /* Exit if we need to wait for socket shutdown to complete */
1848         if (test_bit(XPRT_CLOSING, &xprt->state))
1849                 return;
1850         xs_connect(task);
1851 }
1852
1853 /**
1854  * xs_udp_print_stats - display UDP socket-specifc stats
1855  * @xprt: rpc_xprt struct containing statistics
1856  * @seq: output file
1857  *
1858  */
1859 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
1860 {
1861         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1862
1863         seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %Lu %Lu\n",
1864                         transport->port,
1865                         xprt->stat.bind_count,
1866                         xprt->stat.sends,
1867                         xprt->stat.recvs,
1868                         xprt->stat.bad_xids,
1869                         xprt->stat.req_u,
1870                         xprt->stat.bklog_u);
1871 }
1872
1873 /**
1874  * xs_tcp_print_stats - display TCP socket-specifc stats
1875  * @xprt: rpc_xprt struct containing statistics
1876  * @seq: output file
1877  *
1878  */
1879 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
1880 {
1881         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1882         long idle_time = 0;
1883
1884         if (xprt_connected(xprt))
1885                 idle_time = (long)(jiffies - xprt->last_used) / HZ;
1886
1887         seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu %Lu %Lu\n",
1888                         transport->port,
1889                         xprt->stat.bind_count,
1890                         xprt->stat.connect_count,
1891                         xprt->stat.connect_time,
1892                         idle_time,
1893                         xprt->stat.sends,
1894                         xprt->stat.recvs,
1895                         xprt->stat.bad_xids,
1896                         xprt->stat.req_u,
1897                         xprt->stat.bklog_u);
1898 }
1899
1900 static struct rpc_xprt_ops xs_udp_ops = {
1901         .set_buffer_size        = xs_udp_set_buffer_size,
1902         .reserve_xprt           = xprt_reserve_xprt_cong,
1903         .release_xprt           = xprt_release_xprt_cong,
1904         .rpcbind                = rpcb_getport_async,
1905         .set_port               = xs_set_port,
1906         .connect                = xs_connect,
1907         .buf_alloc              = rpc_malloc,
1908         .buf_free               = rpc_free,
1909         .send_request           = xs_udp_send_request,
1910         .set_retrans_timeout    = xprt_set_retrans_timeout_rtt,
1911         .timer                  = xs_udp_timer,
1912         .release_request        = xprt_release_rqst_cong,
1913         .close                  = xs_close,
1914         .destroy                = xs_destroy,
1915         .print_stats            = xs_udp_print_stats,
1916 };
1917
1918 static struct rpc_xprt_ops xs_tcp_ops = {
1919         .reserve_xprt           = xprt_reserve_xprt,
1920         .release_xprt           = xs_tcp_release_xprt,
1921         .rpcbind                = rpcb_getport_async,
1922         .set_port               = xs_set_port,
1923         .connect                = xs_tcp_connect,
1924         .buf_alloc              = rpc_malloc,
1925         .buf_free               = rpc_free,
1926         .send_request           = xs_tcp_send_request,
1927         .set_retrans_timeout    = xprt_set_retrans_timeout_def,
1928         .close                  = xs_tcp_shutdown,
1929         .destroy                = xs_destroy,
1930         .print_stats            = xs_tcp_print_stats,
1931 };
1932
1933 static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
1934                                       unsigned int slot_table_size)
1935 {
1936         struct rpc_xprt *xprt;
1937         struct sock_xprt *new;
1938
1939         if (args->addrlen > sizeof(xprt->addr)) {
1940                 dprintk("RPC:       xs_setup_xprt: address too large\n");
1941                 return ERR_PTR(-EBADF);
1942         }
1943
1944         new = kzalloc(sizeof(*new), GFP_KERNEL);
1945         if (new == NULL) {
1946                 dprintk("RPC:       xs_setup_xprt: couldn't allocate "
1947                                 "rpc_xprt\n");
1948                 return ERR_PTR(-ENOMEM);
1949         }
1950         xprt = &new->xprt;
1951
1952         xprt->max_reqs = slot_table_size;
1953         xprt->slot = kcalloc(xprt->max_reqs, sizeof(struct rpc_rqst), GFP_KERNEL);
1954         if (xprt->slot == NULL) {
1955                 kfree(xprt);
1956                 dprintk("RPC:       xs_setup_xprt: couldn't allocate slot "
1957                                 "table\n");
1958                 return ERR_PTR(-ENOMEM);
1959         }
1960
1961         memcpy(&xprt->addr, args->dstaddr, args->addrlen);
1962         xprt->addrlen = args->addrlen;
1963         if (args->srcaddr)
1964                 memcpy(&new->addr, args->srcaddr, args->addrlen);
1965
1966         return xprt;
1967 }
1968
1969 static const struct rpc_timeout xs_udp_default_timeout = {
1970         .to_initval = 5 * HZ,
1971         .to_maxval = 30 * HZ,
1972         .to_increment = 5 * HZ,
1973         .to_retries = 5,
1974 };
1975
1976 /**
1977  * xs_setup_udp - Set up transport to use a UDP socket
1978  * @args: rpc transport creation arguments
1979  *
1980  */
1981 static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
1982 {
1983         struct sockaddr *addr = args->dstaddr;
1984         struct rpc_xprt *xprt;
1985         struct sock_xprt *transport;
1986
1987         xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries);
1988         if (IS_ERR(xprt))
1989                 return xprt;
1990         transport = container_of(xprt, struct sock_xprt, xprt);
1991
1992         xprt->prot = IPPROTO_UDP;
1993         xprt->tsh_size = 0;
1994         /* XXX: header size can vary due to auth type, IPv6, etc. */
1995         xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
1996
1997         xprt->bind_timeout = XS_BIND_TO;
1998         xprt->connect_timeout = XS_UDP_CONN_TO;
1999         xprt->reestablish_timeout = XS_UDP_REEST_TO;
2000         xprt->idle_timeout = XS_IDLE_DISC_TO;
2001
2002         xprt->ops = &xs_udp_ops;
2003
2004         xprt->timeout = &xs_udp_default_timeout;
2005
2006         switch (addr->sa_family) {
2007         case AF_INET:
2008                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2009                         xprt_set_bound(xprt);
2010
2011                 INIT_DELAYED_WORK(&transport->connect_worker,
2012                                         xs_udp_connect_worker4);
2013                 xs_format_ipv4_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
2014                 break;
2015         case AF_INET6:
2016                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2017                         xprt_set_bound(xprt);
2018
2019                 INIT_DELAYED_WORK(&transport->connect_worker,
2020                                         xs_udp_connect_worker6);
2021                 xs_format_ipv6_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
2022                 break;
2023         default:
2024                 kfree(xprt);
2025                 return ERR_PTR(-EAFNOSUPPORT);
2026         }
2027
2028         dprintk("RPC:       set up transport to address %s\n",
2029                         xprt->address_strings[RPC_DISPLAY_ALL]);
2030
2031         if (try_module_get(THIS_MODULE))
2032                 return xprt;
2033
2034         kfree(xprt->slot);
2035         kfree(xprt);
2036         return ERR_PTR(-EINVAL);
2037 }
2038
2039 static const struct rpc_timeout xs_tcp_default_timeout = {
2040         .to_initval = 60 * HZ,
2041         .to_maxval = 60 * HZ,
2042         .to_retries = 2,
2043 };
2044
2045 /**
2046  * xs_setup_tcp - Set up transport to use a TCP socket
2047  * @args: rpc transport creation arguments
2048  *
2049  */
2050 static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
2051 {
2052         struct sockaddr *addr = args->dstaddr;
2053         struct rpc_xprt *xprt;
2054         struct sock_xprt *transport;
2055
2056         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries);
2057         if (IS_ERR(xprt))
2058                 return xprt;
2059         transport = container_of(xprt, struct sock_xprt, xprt);
2060
2061         xprt->prot = IPPROTO_TCP;
2062         xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2063         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2064
2065         xprt->bind_timeout = XS_BIND_TO;
2066         xprt->connect_timeout = XS_TCP_CONN_TO;
2067         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2068         xprt->idle_timeout = XS_IDLE_DISC_TO;
2069
2070         xprt->ops = &xs_tcp_ops;
2071         xprt->timeout = &xs_tcp_default_timeout;
2072
2073         switch (addr->sa_family) {
2074         case AF_INET:
2075                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2076                         xprt_set_bound(xprt);
2077
2078                 INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_connect_worker4);
2079                 xs_format_ipv4_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
2080                 break;
2081         case AF_INET6:
2082                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2083                         xprt_set_bound(xprt);
2084
2085                 INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_connect_worker6);
2086                 xs_format_ipv6_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
2087                 break;
2088         default:
2089                 kfree(xprt);
2090                 return ERR_PTR(-EAFNOSUPPORT);
2091         }
2092
2093         dprintk("RPC:       set up transport to address %s\n",
2094                         xprt->address_strings[RPC_DISPLAY_ALL]);
2095
2096         if (try_module_get(THIS_MODULE))
2097                 return xprt;
2098
2099         kfree(xprt->slot);
2100         kfree(xprt);
2101         return ERR_PTR(-EINVAL);
2102 }
2103
2104 static struct xprt_class        xs_udp_transport = {
2105         .list           = LIST_HEAD_INIT(xs_udp_transport.list),
2106         .name           = "udp",
2107         .owner          = THIS_MODULE,
2108         .ident          = IPPROTO_UDP,
2109         .setup          = xs_setup_udp,
2110 };
2111
2112 static struct xprt_class        xs_tcp_transport = {
2113         .list           = LIST_HEAD_INIT(xs_tcp_transport.list),
2114         .name           = "tcp",
2115         .owner          = THIS_MODULE,
2116         .ident          = IPPROTO_TCP,
2117         .setup          = xs_setup_tcp,
2118 };
2119
2120 /**
2121  * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
2122  *
2123  */
2124 int init_socket_xprt(void)
2125 {
2126 #ifdef RPC_DEBUG
2127         if (!sunrpc_table_header)
2128                 sunrpc_table_header = register_sysctl_table(sunrpc_table);
2129 #endif
2130
2131         xprt_register_transport(&xs_udp_transport);
2132         xprt_register_transport(&xs_tcp_transport);
2133
2134         return 0;
2135 }
2136
2137 /**
2138  * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
2139  *
2140  */
2141 void cleanup_socket_xprt(void)
2142 {
2143 #ifdef RPC_DEBUG
2144         if (sunrpc_table_header) {
2145                 unregister_sysctl_table(sunrpc_table_header);
2146                 sunrpc_table_header = NULL;
2147         }
2148 #endif
2149
2150         xprt_unregister_transport(&xs_udp_transport);
2151         xprt_unregister_transport(&xs_tcp_transport);
2152 }