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