git.oblomov.eu Git - linux-2.6/blob - net/sunrpc/xprtsock.c

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