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

   1 /*
   2  * linux/net/sunrpc/xprtsock.c
   3  *
   4  * Client-side transport implementation for sockets.
   5  *
   6  * TCP callback races fixes (C) 1998 Red Hat 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 (likely(status >= (int) req->rq_slen))
 591                 return 0;
 592
 593         /* Still some bytes left; set up for a retry later. */
 594         if (status > 0)
 595                 status = -EAGAIN;
 596
 597         switch (status) {
 598         case -ENETUNREACH:
 599         case -EPIPE:
 600         case -ECONNREFUSED:
 601                 /* When the server has died, an ICMP port unreachable message
 602                  * prompts ECONNREFUSED. */
 603                 break;
 604         case -EAGAIN:
 605                 xs_nospace(task);
 606                 break;
 607         default:
 608                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
 609                         -status);
 610                 break;
 611         }
 612
 613         return status;
 614 }
 615
 616 static inline void xs_encode_tcp_record_marker(struct xdr_buf *buf)
 617 {
 618         u32 reclen = buf->len - sizeof(rpc_fraghdr);
 619         rpc_fraghdr *base = buf->head[0].iov_base;
 620         *base = htonl(RPC_LAST_STREAM_FRAGMENT | reclen);
 621 }
 622
 623 /**
 624  * xs_tcp_send_request - write an RPC request to a TCP socket
 625  * @task: address of RPC task that manages the state of an RPC request
 626  *
 627  * Return values:
 628  *        0:    The request has been sent
 629  *   EAGAIN:    The socket was blocked, please call again later to
 630  *              complete the request
 631  * ENOTCONN:    Caller needs to invoke connect logic then call again
 632  *    other:    Some other error occured, the request was not sent
 633  *
 634  * XXX: In the case of soft timeouts, should we eventually give up
 635  *      if sendmsg is not able to make progress?
 636  */
 637 static int xs_tcp_send_request(struct rpc_task *task)
 638 {
 639         struct rpc_rqst *req = task->tk_rqstp;
 640         struct rpc_xprt *xprt = req->rq_xprt;
 641         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 642         struct xdr_buf *xdr = &req->rq_snd_buf;
 643         int status;
 644         unsigned int retry = 0;
 645
 646         xs_encode_tcp_record_marker(&req->rq_snd_buf);
 647
 648         xs_pktdump("packet data:",
 649                                 req->rq_svec->iov_base,
 650                                 req->rq_svec->iov_len);
 651
 652         /* Continue transmitting the packet/record. We must be careful
 653          * to cope with writespace callbacks arriving _after_ we have
 654          * called sendmsg(). */
 655         while (1) {
 656                 req->rq_xtime = jiffies;
 657                 status = xs_sendpages(transport->sock,
 658                                         NULL, 0, xdr, req->rq_bytes_sent);
 659
 660                 dprintk("RPC:       xs_tcp_send_request(%u) = %d\n",
 661                                 xdr->len - req->rq_bytes_sent, status);
 662
 663                 if (unlikely(status < 0))
 664                         break;
 665
 666                 /* If we've sent the entire packet, immediately
 667                  * reset the count of bytes sent. */
 668                 req->rq_bytes_sent += status;
 669                 task->tk_bytes_sent += status;
 670                 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
 671                         req->rq_bytes_sent = 0;
 672                         return 0;
 673                 }
 674
 675                 status = -EAGAIN;
 676                 if (retry++ > XS_SENDMSG_RETRY)
 677                         break;
 678         }
 679
 680         switch (status) {
 681         case -EAGAIN:
 682                 xs_nospace(task);
 683                 break;
 684         case -ECONNREFUSED:
 685         case -ECONNRESET:
 686         case -ENOTCONN:
 687         case -EPIPE:
 688                 status = -ENOTCONN;
 689                 break;
 690         default:
 691                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
 692                         -status);
 693                 xprt_disconnect(xprt);
 694                 break;
 695         }
 696
 697         return status;
 698 }
 699
 700 /**
 701  * xs_tcp_release_xprt - clean up after a tcp transmission
 702  * @xprt: transport
 703  * @task: rpc task
 704  *
 705  * This cleans up if an error causes us to abort the transmission of a request.
 706  * In this case, the socket may need to be reset in order to avoid confusing
 707  * the server.
 708  */
 709 static void xs_tcp_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
 710 {
 711         struct rpc_rqst *req;
 712
 713         if (task != xprt->snd_task)
 714                 return;
 715         if (task == NULL)
 716                 goto out_release;
 717         req = task->tk_rqstp;
 718         if (req->rq_bytes_sent == 0)
 719                 goto out_release;
 720         if (req->rq_bytes_sent == req->rq_snd_buf.len)
 721                 goto out_release;
 722         set_bit(XPRT_CLOSE_WAIT, &task->tk_xprt->state);
 723 out_release:
 724         xprt_release_xprt(xprt, task);
 725 }
 726
 727 /**
 728  * xs_close - close a socket
 729  * @xprt: transport
 730  *
 731  * This is used when all requests are complete; ie, no DRC state remains
 732  * on the server we want to save.
 733  */
 734 static void xs_close(struct rpc_xprt *xprt)
 735 {
 736         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 737         struct socket *sock = transport->sock;
 738         struct sock *sk = transport->inet;
 739
 740         if (!sk)
 741                 goto clear_close_wait;
 742
 743         dprintk("RPC:       xs_close xprt %p\n", xprt);
 744
 745         write_lock_bh(&sk->sk_callback_lock);
 746         transport->inet = NULL;
 747         transport->sock = NULL;
 748
 749         sk->sk_user_data = NULL;
 750         sk->sk_data_ready = transport->old_data_ready;
 751         sk->sk_state_change = transport->old_state_change;
 752         sk->sk_write_space = transport->old_write_space;
 753         write_unlock_bh(&sk->sk_callback_lock);
 754
 755         sk->sk_no_check = 0;
 756
 757         sock_release(sock);
 758 clear_close_wait:
 759         smp_mb__before_clear_bit();
 760         clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
 761         smp_mb__after_clear_bit();
 762 }
 763
 764 /**
 765  * xs_destroy - prepare to shutdown a transport
 766  * @xprt: doomed transport
 767  *
 768  */
 769 static void xs_destroy(struct rpc_xprt *xprt)
 770 {
 771         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 772
 773         dprintk("RPC:       xs_destroy xprt %p\n", xprt);
 774
 775         cancel_rearming_delayed_work(&transport->connect_worker);
 776
 777         xprt_disconnect(xprt);
 778         xs_close(xprt);
 779         xs_free_peer_addresses(xprt);
 780         kfree(xprt->slot);
 781         kfree(xprt);
 782         module_put(THIS_MODULE);
 783 }
 784
 785 static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
 786 {
 787         return (struct rpc_xprt *) sk->sk_user_data;
 788 }
 789
 790 /**
 791  * xs_udp_data_ready - "data ready" callback for UDP sockets
 792  * @sk: socket with data to read
 793  * @len: how much data to read
 794  *
 795  */
 796 static void xs_udp_data_ready(struct sock *sk, int len)
 797 {
 798         struct rpc_task *task;
 799         struct rpc_xprt *xprt;
 800         struct rpc_rqst *rovr;
 801         struct sk_buff *skb;
 802         int err, repsize, copied;
 803         u32 _xid;
 804         __be32 *xp;
 805
 806         read_lock(&sk->sk_callback_lock);
 807         dprintk("RPC:       xs_udp_data_ready...\n");
 808         if (!(xprt = xprt_from_sock(sk)))
 809                 goto out;
 810
 811         if ((skb = skb_recv_datagram(sk, 0, 1, &err)) == NULL)
 812                 goto out;
 813
 814         if (xprt->shutdown)
 815                 goto dropit;
 816
 817         repsize = skb->len - sizeof(struct udphdr);
 818         if (repsize < 4) {
 819                 dprintk("RPC:       impossible RPC reply size %d!\n", repsize);
 820                 goto dropit;
 821         }
 822
 823         /* Copy the XID from the skb... */
 824         xp = skb_header_pointer(skb, sizeof(struct udphdr),
 825                                 sizeof(_xid), &_xid);
 826         if (xp == NULL)
 827                 goto dropit;
 828
 829         /* Look up and lock the request corresponding to the given XID */
 830         spin_lock(&xprt->transport_lock);
 831         rovr = xprt_lookup_rqst(xprt, *xp);
 832         if (!rovr)
 833                 goto out_unlock;
 834         task = rovr->rq_task;
 835
 836         if ((copied = rovr->rq_private_buf.buflen) > repsize)
 837                 copied = repsize;
 838
 839         /* Suck it into the iovec, verify checksum if not done by hw. */
 840         if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb))
 841                 goto out_unlock;
 842
 843         /* Something worked... */
 844         dst_confirm(skb->dst);
 845
 846         xprt_adjust_cwnd(task, copied);
 847         xprt_update_rtt(task);
 848         xprt_complete_rqst(task, copied);
 849
 850  out_unlock:
 851         spin_unlock(&xprt->transport_lock);
 852  dropit:
 853         skb_free_datagram(sk, skb);
 854  out:
 855         read_unlock(&sk->sk_callback_lock);
 856 }
 857
 858 static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
 859 {
 860         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 861         size_t len, used;
 862         char *p;
 863
 864         p = ((char *) &transport->tcp_fraghdr) + transport->tcp_offset;
 865         len = sizeof(transport->tcp_fraghdr) - transport->tcp_offset;
 866         used = xdr_skb_read_bits(desc, p, len);
 867         transport->tcp_offset += used;
 868         if (used != len)
 869                 return;
 870
 871         transport->tcp_reclen = ntohl(transport->tcp_fraghdr);
 872         if (transport->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
 873                 transport->tcp_flags |= TCP_RCV_LAST_FRAG;
 874         else
 875                 transport->tcp_flags &= ~TCP_RCV_LAST_FRAG;
 876         transport->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
 877
 878         transport->tcp_flags &= ~TCP_RCV_COPY_FRAGHDR;
 879         transport->tcp_offset = 0;
 880
 881         /* Sanity check of the record length */
 882         if (unlikely(transport->tcp_reclen < 4)) {
 883                 dprintk("RPC:       invalid TCP record fragment length\n");
 884                 xprt_disconnect(xprt);
 885                 return;
 886         }
 887         dprintk("RPC:       reading TCP record fragment of length %d\n",
 888                         transport->tcp_reclen);
 889 }
 890
 891 static void xs_tcp_check_fraghdr(struct sock_xprt *transport)
 892 {
 893         if (transport->tcp_offset == transport->tcp_reclen) {
 894                 transport->tcp_flags |= TCP_RCV_COPY_FRAGHDR;
 895                 transport->tcp_offset = 0;
 896                 if (transport->tcp_flags & TCP_RCV_LAST_FRAG) {
 897                         transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
 898                         transport->tcp_flags |= TCP_RCV_COPY_XID;
 899                         transport->tcp_copied = 0;
 900                 }
 901         }
 902 }
 903
 904 static inline void xs_tcp_read_xid(struct sock_xprt *transport, struct xdr_skb_reader *desc)
 905 {
 906         size_t len, used;
 907         char *p;
 908
 909         len = sizeof(transport->tcp_xid) - transport->tcp_offset;
 910         dprintk("RPC:       reading XID (%Zu bytes)\n", len);
 911         p = ((char *) &transport->tcp_xid) + transport->tcp_offset;
 912         used = xdr_skb_read_bits(desc, p, len);
 913         transport->tcp_offset += used;
 914         if (used != len)
 915                 return;
 916         transport->tcp_flags &= ~TCP_RCV_COPY_XID;
 917         transport->tcp_flags |= TCP_RCV_COPY_DATA;
 918         transport->tcp_copied = 4;
 919         dprintk("RPC:       reading reply for XID %08x\n",
 920                         ntohl(transport->tcp_xid));
 921         xs_tcp_check_fraghdr(transport);
 922 }
 923
 924 static inline void xs_tcp_read_request(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
 925 {
 926         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 927         struct rpc_rqst *req;
 928         struct xdr_buf *rcvbuf;
 929         size_t len;
 930         ssize_t r;
 931
 932         /* Find and lock the request corresponding to this xid */
 933         spin_lock(&xprt->transport_lock);
 934         req = xprt_lookup_rqst(xprt, transport->tcp_xid);
 935         if (!req) {
 936                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
 937                 dprintk("RPC:       XID %08x request not found!\n",
 938                                 ntohl(transport->tcp_xid));
 939                 spin_unlock(&xprt->transport_lock);
 940                 return;
 941         }
 942
 943         rcvbuf = &req->rq_private_buf;
 944         len = desc->count;
 945         if (len > transport->tcp_reclen - transport->tcp_offset) {
 946                 struct xdr_skb_reader my_desc;
 947
 948                 len = transport->tcp_reclen - transport->tcp_offset;
 949                 memcpy(&my_desc, desc, sizeof(my_desc));
 950                 my_desc.count = len;
 951                 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
 952                                           &my_desc, xdr_skb_read_bits);
 953                 desc->count -= r;
 954                 desc->offset += r;
 955         } else
 956                 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
 957                                           desc, xdr_skb_read_bits);
 958
 959         if (r > 0) {
 960                 transport->tcp_copied += r;
 961                 transport->tcp_offset += r;
 962         }
 963         if (r != len) {
 964                 /* Error when copying to the receive buffer,
 965                  * usually because we weren't able to allocate
 966                  * additional buffer pages. All we can do now
 967                  * is turn off TCP_RCV_COPY_DATA, so the request
 968                  * will not receive any additional updates,
 969                  * and time out.
 970                  * Any remaining data from this record will
 971                  * be discarded.
 972                  */
 973                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
 974                 dprintk("RPC:       XID %08x truncated request\n",
 975                                 ntohl(transport->tcp_xid));
 976                 dprintk("RPC:       xprt = %p, tcp_copied = %lu, "
 977                                 "tcp_offset = %u, tcp_reclen = %u\n",
 978                                 xprt, transport->tcp_copied,
 979                                 transport->tcp_offset, transport->tcp_reclen);
 980                 goto out;
 981         }
 982
 983         dprintk("RPC:       XID %08x read %Zd bytes\n",
 984                         ntohl(transport->tcp_xid), r);
 985         dprintk("RPC:       xprt = %p, tcp_copied = %lu, tcp_offset = %u, "
 986                         "tcp_reclen = %u\n", xprt, transport->tcp_copied,
 987                         transport->tcp_offset, transport->tcp_reclen);
 988
 989         if (transport->tcp_copied == req->rq_private_buf.buflen)
 990                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
 991         else if (transport->tcp_offset == transport->tcp_reclen) {
 992                 if (transport->tcp_flags & TCP_RCV_LAST_FRAG)
 993                         transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
 994         }
 995
 996 out:
 997         if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
 998                 xprt_complete_rqst(req->rq_task, transport->tcp_copied);
 999         spin_unlock(&xprt->transport_lock);
1000         xs_tcp_check_fraghdr(transport);
1001 }
1002
1003 static inline void xs_tcp_read_discard(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1004 {
1005         size_t len;
1006
1007         len = transport->tcp_reclen - transport->tcp_offset;
1008         if (len > desc->count)
1009                 len = desc->count;
1010         desc->count -= len;
1011         desc->offset += len;
1012         transport->tcp_offset += len;
1013         dprintk("RPC:       discarded %Zu bytes\n", len);
1014         xs_tcp_check_fraghdr(transport);
1015 }
1016
1017 static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
1018 {
1019         struct rpc_xprt *xprt = rd_desc->arg.data;
1020         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1021         struct xdr_skb_reader desc = {
1022                 .skb    = skb,
1023                 .offset = offset,
1024                 .count  = len,
1025         };
1026
1027         dprintk("RPC:       xs_tcp_data_recv started\n");
1028         do {
1029                 /* Read in a new fragment marker if necessary */
1030                 /* Can we ever really expect to get completely empty fragments? */
1031                 if (transport->tcp_flags & TCP_RCV_COPY_FRAGHDR) {
1032                         xs_tcp_read_fraghdr(xprt, &desc);
1033                         continue;
1034                 }
1035                 /* Read in the xid if necessary */
1036                 if (transport->tcp_flags & TCP_RCV_COPY_XID) {
1037                         xs_tcp_read_xid(transport, &desc);
1038                         continue;
1039                 }
1040                 /* Read in the request data */
1041                 if (transport->tcp_flags & TCP_RCV_COPY_DATA) {
1042                         xs_tcp_read_request(xprt, &desc);
1043                         continue;
1044                 }
1045                 /* Skip over any trailing bytes on short reads */
1046                 xs_tcp_read_discard(transport, &desc);
1047         } while (desc.count);
1048         dprintk("RPC:       xs_tcp_data_recv done\n");
1049         return len - desc.count;
1050 }
1051
1052 /**
1053  * xs_tcp_data_ready - "data ready" callback for TCP sockets
1054  * @sk: socket with data to read
1055  * @bytes: how much data to read
1056  *
1057  */
1058 static void xs_tcp_data_ready(struct sock *sk, int bytes)
1059 {
1060         struct rpc_xprt *xprt;
1061         read_descriptor_t rd_desc;
1062
1063         dprintk("RPC:       xs_tcp_data_ready...\n");
1064
1065         read_lock(&sk->sk_callback_lock);
1066         if (!(xprt = xprt_from_sock(sk)))
1067                 goto out;
1068         if (xprt->shutdown)
1069                 goto out;
1070
1071         /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
1072         rd_desc.arg.data = xprt;
1073         rd_desc.count = 65536;
1074         tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
1075 out:
1076         read_unlock(&sk->sk_callback_lock);
1077 }
1078
1079 /**
1080  * xs_tcp_state_change - callback to handle TCP socket state changes
1081  * @sk: socket whose state has changed
1082  *
1083  */
1084 static void xs_tcp_state_change(struct sock *sk)
1085 {
1086         struct rpc_xprt *xprt;
1087
1088         read_lock(&sk->sk_callback_lock);
1089         if (!(xprt = xprt_from_sock(sk)))
1090                 goto out;
1091         dprintk("RPC:       xs_tcp_state_change client %p...\n", xprt);
1092         dprintk("RPC:       state %x conn %d dead %d zapped %d\n",
1093                         sk->sk_state, xprt_connected(xprt),
1094                         sock_flag(sk, SOCK_DEAD),
1095                         sock_flag(sk, SOCK_ZAPPED));
1096
1097         switch (sk->sk_state) {
1098         case TCP_ESTABLISHED:
1099                 spin_lock_bh(&xprt->transport_lock);
1100                 if (!xprt_test_and_set_connected(xprt)) {
1101                         struct sock_xprt *transport = container_of(xprt,
1102                                         struct sock_xprt, xprt);
1103
1104                         /* Reset TCP record info */
1105                         transport->tcp_offset = 0;
1106                         transport->tcp_reclen = 0;
1107                         transport->tcp_copied = 0;
1108                         transport->tcp_flags =
1109                                 TCP_RCV_COPY_FRAGHDR | TCP_RCV_COPY_XID;
1110
1111                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1112                         xprt_wake_pending_tasks(xprt, 0);
1113                 }
1114                 spin_unlock_bh(&xprt->transport_lock);
1115                 break;
1116         case TCP_SYN_SENT:
1117         case TCP_SYN_RECV:
1118                 break;
1119         case TCP_CLOSE_WAIT:
1120                 /* Try to schedule an autoclose RPC calls */
1121                 set_bit(XPRT_CLOSE_WAIT, &xprt->state);
1122                 if (test_and_set_bit(XPRT_LOCKED, &xprt->state) == 0)
1123                         queue_work(rpciod_workqueue, &xprt->task_cleanup);
1124         default:
1125                 xprt_disconnect(xprt);
1126         }
1127  out:
1128         read_unlock(&sk->sk_callback_lock);
1129 }
1130
1131 /**
1132  * xs_udp_write_space - callback invoked when socket buffer space
1133  *                             becomes available
1134  * @sk: socket whose state has changed
1135  *
1136  * Called when more output buffer space is available for this socket.
1137  * We try not to wake our writers until they can make "significant"
1138  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1139  * with a bunch of small requests.
1140  */
1141 static void xs_udp_write_space(struct sock *sk)
1142 {
1143         read_lock(&sk->sk_callback_lock);
1144
1145         /* from net/core/sock.c:sock_def_write_space */
1146         if (sock_writeable(sk)) {
1147                 struct socket *sock;
1148                 struct rpc_xprt *xprt;
1149
1150                 if (unlikely(!(sock = sk->sk_socket)))
1151                         goto out;
1152                 if (unlikely(!(xprt = xprt_from_sock(sk))))
1153                         goto out;
1154                 if (unlikely(!test_and_clear_bit(SOCK_NOSPACE, &sock->flags)))
1155                         goto out;
1156
1157                 xprt_write_space(xprt);
1158         }
1159
1160  out:
1161         read_unlock(&sk->sk_callback_lock);
1162 }
1163
1164 /**
1165  * xs_tcp_write_space - callback invoked when socket buffer space
1166  *                             becomes available
1167  * @sk: socket whose state has changed
1168  *
1169  * Called when more output buffer space is available for this socket.
1170  * We try not to wake our writers until they can make "significant"
1171  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1172  * with a bunch of small requests.
1173  */
1174 static void xs_tcp_write_space(struct sock *sk)
1175 {
1176         read_lock(&sk->sk_callback_lock);
1177
1178         /* from net/core/stream.c:sk_stream_write_space */
1179         if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk)) {
1180                 struct socket *sock;
1181                 struct rpc_xprt *xprt;
1182
1183                 if (unlikely(!(sock = sk->sk_socket)))
1184                         goto out;
1185                 if (unlikely(!(xprt = xprt_from_sock(sk))))
1186                         goto out;
1187                 if (unlikely(!test_and_clear_bit(SOCK_NOSPACE, &sock->flags)))
1188                         goto out;
1189
1190                 xprt_write_space(xprt);
1191         }
1192
1193  out:
1194         read_unlock(&sk->sk_callback_lock);
1195 }
1196
1197 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
1198 {
1199         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1200         struct sock *sk = transport->inet;
1201
1202         if (transport->rcvsize) {
1203                 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1204                 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
1205         }
1206         if (transport->sndsize) {
1207                 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1208                 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1209                 sk->sk_write_space(sk);
1210         }
1211 }
1212
1213 /**
1214  * xs_udp_set_buffer_size - set send and receive limits
1215  * @xprt: generic transport
1216  * @sndsize: requested size of send buffer, in bytes
1217  * @rcvsize: requested size of receive buffer, in bytes
1218  *
1219  * Set socket send and receive buffer size limits.
1220  */
1221 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1222 {
1223         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1224
1225         transport->sndsize = 0;
1226         if (sndsize)
1227                 transport->sndsize = sndsize + 1024;
1228         transport->rcvsize = 0;
1229         if (rcvsize)
1230                 transport->rcvsize = rcvsize + 1024;
1231
1232         xs_udp_do_set_buffer_size(xprt);
1233 }
1234
1235 /**
1236  * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1237  * @task: task that timed out
1238  *
1239  * Adjust the congestion window after a retransmit timeout has occurred.
1240  */
1241 static void xs_udp_timer(struct rpc_task *task)
1242 {
1243         xprt_adjust_cwnd(task, -ETIMEDOUT);
1244 }
1245
1246 static unsigned short xs_get_random_port(void)
1247 {
1248         unsigned short range = xprt_max_resvport - xprt_min_resvport;
1249         unsigned short rand = (unsigned short) net_random() % range;
1250         return rand + xprt_min_resvport;
1251 }
1252
1253 /**
1254  * xs_set_port - reset the port number in the remote endpoint address
1255  * @xprt: generic transport
1256  * @port: new port number
1257  *
1258  */
1259 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1260 {
1261         struct sockaddr *addr = xs_addr(xprt);
1262
1263         dprintk("RPC:       setting port for xprt %p to %u\n", xprt, port);
1264
1265         switch (addr->sa_family) {
1266         case AF_INET:
1267                 ((struct sockaddr_in *)addr)->sin_port = htons(port);
1268                 break;
1269         case AF_INET6:
1270                 ((struct sockaddr_in6 *)addr)->sin6_port = htons(port);
1271                 break;
1272         default:
1273                 BUG();
1274         }
1275 }
1276
1277 static int xs_bind4(struct sock_xprt *transport, struct socket *sock)
1278 {
1279         struct sockaddr_in myaddr = {
1280                 .sin_family = AF_INET,
1281         };
1282         struct sockaddr_in *sa;
1283         int err;
1284         unsigned short port = transport->port;
1285
1286         if (!transport->xprt.resvport)
1287                 port = 0;
1288         sa = (struct sockaddr_in *)&transport->addr;
1289         myaddr.sin_addr = sa->sin_addr;
1290         do {
1291                 myaddr.sin_port = htons(port);
1292                 err = kernel_bind(sock, (struct sockaddr *) &myaddr,
1293                                                 sizeof(myaddr));
1294                 if (!transport->xprt.resvport)
1295                         break;
1296                 if (err == 0) {
1297                         transport->port = port;
1298                         break;
1299                 }
1300                 if (port <= xprt_min_resvport)
1301                         port = xprt_max_resvport;
1302                 else
1303                         port--;
1304         } while (err == -EADDRINUSE && port != transport->port);
1305         dprintk("RPC:       %s "NIPQUAD_FMT":%u: %s (%d)\n",
1306                         __FUNCTION__, NIPQUAD(myaddr.sin_addr),
1307                         port, err ? "failed" : "ok", err);
1308         return err;
1309 }
1310
1311 static int xs_bind6(struct sock_xprt *transport, struct socket *sock)
1312 {
1313         struct sockaddr_in6 myaddr = {
1314                 .sin6_family = AF_INET6,
1315         };
1316         struct sockaddr_in6 *sa;
1317         int err;
1318         unsigned short port = transport->port;
1319
1320         if (!transport->xprt.resvport)
1321                 port = 0;
1322         sa = (struct sockaddr_in6 *)&transport->addr;
1323         myaddr.sin6_addr = sa->sin6_addr;
1324         do {
1325                 myaddr.sin6_port = htons(port);
1326                 err = kernel_bind(sock, (struct sockaddr *) &myaddr,
1327                                                 sizeof(myaddr));
1328                 if (!transport->xprt.resvport)
1329                         break;
1330                 if (err == 0) {
1331                         transport->port = port;
1332                         break;
1333                 }
1334                 if (port <= xprt_min_resvport)
1335                         port = xprt_max_resvport;
1336                 else
1337                         port--;
1338         } while (err == -EADDRINUSE && port != transport->port);
1339         dprintk("RPC:       xs_bind6 "NIP6_FMT":%u: %s (%d)\n",
1340                 NIP6(myaddr.sin6_addr), port, err ? "failed" : "ok", err);
1341         return err;
1342 }
1343
1344 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1345 static struct lock_class_key xs_key[2];
1346 static struct lock_class_key xs_slock_key[2];
1347
1348 static inline void xs_reclassify_socket4(struct socket *sock)
1349 {
1350         struct sock *sk = sock->sk;
1351
1352         BUG_ON(sk->sk_lock.owner != NULL);
1353         sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1354                 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1355 }
1356
1357 static inline void xs_reclassify_socket6(struct socket *sock)
1358 {
1359         struct sock *sk = sock->sk;
1360
1361         BUG_ON(sk->sk_lock.owner != NULL);
1362         sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1363                 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
1364 }
1365 #else
1366 static inline void xs_reclassify_socket4(struct socket *sock)
1367 {
1368 }
1369
1370 static inline void xs_reclassify_socket6(struct socket *sock)
1371 {
1372 }
1373 #endif
1374
1375 static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
1376 {
1377         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1378
1379         if (!transport->inet) {
1380                 struct sock *sk = sock->sk;
1381
1382                 write_lock_bh(&sk->sk_callback_lock);
1383
1384                 sk->sk_user_data = xprt;
1385                 transport->old_data_ready = sk->sk_data_ready;
1386                 transport->old_state_change = sk->sk_state_change;
1387                 transport->old_write_space = sk->sk_write_space;
1388                 sk->sk_data_ready = xs_udp_data_ready;
1389                 sk->sk_write_space = xs_udp_write_space;
1390                 sk->sk_no_check = UDP_CSUM_NORCV;
1391                 sk->sk_allocation = GFP_ATOMIC;
1392
1393                 xprt_set_connected(xprt);
1394
1395                 /* Reset to new socket */
1396                 transport->sock = sock;
1397                 transport->inet = sk;
1398
1399                 write_unlock_bh(&sk->sk_callback_lock);
1400         }
1401         xs_udp_do_set_buffer_size(xprt);
1402 }
1403
1404 /**
1405  * xs_udp_connect_worker4 - set up a UDP socket
1406  * @work: RPC transport to connect
1407  *
1408  * Invoked by a work queue tasklet.
1409  */
1410 static void xs_udp_connect_worker4(struct work_struct *work)
1411 {
1412         struct sock_xprt *transport =
1413                 container_of(work, struct sock_xprt, connect_worker.work);
1414         struct rpc_xprt *xprt = &transport->xprt;
1415         struct socket *sock = transport->sock;
1416         int err, status = -EIO;
1417
1418         if (xprt->shutdown || !xprt_bound(xprt))
1419                 goto out;
1420
1421         /* Start by resetting any existing state */
1422         xs_close(xprt);
1423
1424         if ((err = sock_create_kern(PF_INET, SOCK_DGRAM, IPPROTO_UDP, &sock)) < 0) {
1425                 dprintk("RPC:       can't create UDP transport socket (%d).\n", -err);
1426                 goto out;
1427         }
1428         xs_reclassify_socket4(sock);
1429
1430         if (xs_bind4(transport, sock)) {
1431                 sock_release(sock);
1432                 goto out;
1433         }
1434
1435         dprintk("RPC:       worker connecting xprt %p to address: %s\n",
1436                         xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1437
1438         xs_udp_finish_connecting(xprt, sock);
1439         status = 0;
1440 out:
1441         xprt_wake_pending_tasks(xprt, status);
1442         xprt_clear_connecting(xprt);
1443 }
1444
1445 /**
1446  * xs_udp_connect_worker6 - set up a UDP socket
1447  * @work: RPC transport to connect
1448  *
1449  * Invoked by a work queue tasklet.
1450  */
1451 static void xs_udp_connect_worker6(struct work_struct *work)
1452 {
1453         struct sock_xprt *transport =
1454                 container_of(work, struct sock_xprt, connect_worker.work);
1455         struct rpc_xprt *xprt = &transport->xprt;
1456         struct socket *sock = transport->sock;
1457         int err, status = -EIO;
1458
1459         if (xprt->shutdown || !xprt_bound(xprt))
1460                 goto out;
1461
1462         /* Start by resetting any existing state */
1463         xs_close(xprt);
1464
1465         if ((err = sock_create_kern(PF_INET6, SOCK_DGRAM, IPPROTO_UDP, &sock)) < 0) {
1466                 dprintk("RPC:       can't create UDP transport socket (%d).\n", -err);
1467                 goto out;
1468         }
1469         xs_reclassify_socket6(sock);
1470
1471         if (xs_bind6(transport, sock) < 0) {
1472                 sock_release(sock);
1473                 goto out;
1474         }
1475
1476         dprintk("RPC:       worker connecting xprt %p to address: %s\n",
1477                         xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1478
1479         xs_udp_finish_connecting(xprt, sock);
1480         status = 0;
1481 out:
1482         xprt_wake_pending_tasks(xprt, status);
1483         xprt_clear_connecting(xprt);
1484 }
1485
1486 /*
1487  * We need to preserve the port number so the reply cache on the server can
1488  * find our cached RPC replies when we get around to reconnecting.
1489  */
1490 static void xs_tcp_reuse_connection(struct rpc_xprt *xprt)
1491 {
1492         int result;
1493         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1494         struct sockaddr any;
1495
1496         dprintk("RPC:       disconnecting xprt %p to reuse port\n", xprt);
1497
1498         /*
1499          * Disconnect the transport socket by doing a connect operation
1500          * with AF_UNSPEC.  This should return immediately...
1501          */
1502         memset(&any, 0, sizeof(any));
1503         any.sa_family = AF_UNSPEC;
1504         result = kernel_connect(transport->sock, &any, sizeof(any), 0);
1505         if (result)
1506                 dprintk("RPC:       AF_UNSPEC connect return code %d\n",
1507                                 result);
1508 }
1509
1510 static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
1511 {
1512         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1513
1514         if (!transport->inet) {
1515                 struct sock *sk = sock->sk;
1516
1517                 write_lock_bh(&sk->sk_callback_lock);
1518
1519                 sk->sk_user_data = xprt;
1520                 transport->old_data_ready = sk->sk_data_ready;
1521                 transport->old_state_change = sk->sk_state_change;
1522                 transport->old_write_space = sk->sk_write_space;
1523                 sk->sk_data_ready = xs_tcp_data_ready;
1524                 sk->sk_state_change = xs_tcp_state_change;
1525                 sk->sk_write_space = xs_tcp_write_space;
1526                 sk->sk_allocation = GFP_ATOMIC;
1527
1528                 /* socket options */
1529                 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
1530                 sock_reset_flag(sk, SOCK_LINGER);
1531                 tcp_sk(sk)->linger2 = 0;
1532                 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
1533
1534                 xprt_clear_connected(xprt);
1535
1536                 /* Reset to new socket */
1537                 transport->sock = sock;
1538                 transport->inet = sk;
1539
1540                 write_unlock_bh(&sk->sk_callback_lock);
1541         }
1542
1543         /* Tell the socket layer to start connecting... */
1544         xprt->stat.connect_count++;
1545         xprt->stat.connect_start = jiffies;
1546         return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
1547 }
1548
1549 /**
1550  * xs_tcp_connect_worker4 - connect a TCP socket to a remote endpoint
1551  * @work: RPC transport to connect
1552  *
1553  * Invoked by a work queue tasklet.
1554  */
1555 static void xs_tcp_connect_worker4(struct work_struct *work)
1556 {
1557         struct sock_xprt *transport =
1558                 container_of(work, struct sock_xprt, connect_worker.work);
1559         struct rpc_xprt *xprt = &transport->xprt;
1560         struct socket *sock = transport->sock;
1561         int err, status = -EIO;
1562
1563         if (xprt->shutdown || !xprt_bound(xprt))
1564                 goto out;
1565
1566         if (!sock) {
1567                 /* start from scratch */
1568                 if ((err = sock_create_kern(PF_INET, SOCK_STREAM, IPPROTO_TCP, &sock)) < 0) {
1569                         dprintk("RPC:       can't create TCP transport socket (%d).\n", -err);
1570                         goto out;
1571                 }
1572                 xs_reclassify_socket4(sock);
1573
1574                 if (xs_bind4(transport, sock) < 0) {
1575                         sock_release(sock);
1576                         goto out;
1577                 }
1578         } else
1579                 /* "close" the socket, preserving the local port */
1580                 xs_tcp_reuse_connection(xprt);
1581
1582         dprintk("RPC:       worker connecting xprt %p to address: %s\n",
1583                         xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1584
1585         status = xs_tcp_finish_connecting(xprt, sock);
1586         dprintk("RPC:       %p connect status %d connected %d sock state %d\n",
1587                         xprt, -status, xprt_connected(xprt),
1588                         sock->sk->sk_state);
1589         if (status < 0) {
1590                 switch (status) {
1591                         case -EINPROGRESS:
1592                         case -EALREADY:
1593                                 goto out_clear;
1594                         case -ECONNREFUSED:
1595                         case -ECONNRESET:
1596                                 /* retry with existing socket, after a delay */
1597                                 break;
1598                         default:
1599                                 /* get rid of existing socket, and retry */
1600                                 xs_close(xprt);
1601                                 break;
1602                 }
1603         }
1604 out:
1605         xprt_wake_pending_tasks(xprt, status);
1606 out_clear:
1607         xprt_clear_connecting(xprt);
1608 }
1609
1610 /**
1611  * xs_tcp_connect_worker6 - connect a TCP socket to a remote endpoint
1612  * @work: RPC transport to connect
1613  *
1614  * Invoked by a work queue tasklet.
1615  */
1616 static void xs_tcp_connect_worker6(struct work_struct *work)
1617 {
1618         struct sock_xprt *transport =
1619                 container_of(work, struct sock_xprt, connect_worker.work);
1620         struct rpc_xprt *xprt = &transport->xprt;
1621         struct socket *sock = transport->sock;
1622         int err, status = -EIO;
1623
1624         if (xprt->shutdown || !xprt_bound(xprt))
1625                 goto out;
1626
1627         if (!sock) {
1628                 /* start from scratch */
1629                 if ((err = sock_create_kern(PF_INET6, SOCK_STREAM, IPPROTO_TCP, &sock)) < 0) {
1630                         dprintk("RPC:       can't create TCP transport socket (%d).\n", -err);
1631                         goto out;
1632                 }
1633                 xs_reclassify_socket6(sock);
1634
1635                 if (xs_bind6(transport, sock) < 0) {
1636                         sock_release(sock);
1637                         goto out;
1638                 }
1639         } else
1640                 /* "close" the socket, preserving the local port */
1641                 xs_tcp_reuse_connection(xprt);
1642
1643         dprintk("RPC:       worker connecting xprt %p to address: %s\n",
1644                         xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1645
1646         status = xs_tcp_finish_connecting(xprt, sock);
1647         dprintk("RPC:       %p connect status %d connected %d sock state %d\n",
1648                         xprt, -status, xprt_connected(xprt), sock->sk->sk_state);
1649         if (status < 0) {
1650                 switch (status) {
1651                         case -EINPROGRESS:
1652                         case -EALREADY:
1653                                 goto out_clear;
1654                         case -ECONNREFUSED:
1655                         case -ECONNRESET:
1656                                 /* retry with existing socket, after a delay */
1657                                 break;
1658                         default:
1659                                 /* get rid of existing socket, and retry */
1660                                 xs_close(xprt);
1661                                 break;
1662                 }
1663         }
1664 out:
1665         xprt_wake_pending_tasks(xprt, status);
1666 out_clear:
1667         xprt_clear_connecting(xprt);
1668 }
1669
1670 /**
1671  * xs_connect - connect a socket to a remote endpoint
1672  * @task: address of RPC task that manages state of connect request
1673  *
1674  * TCP: If the remote end dropped the connection, delay reconnecting.
1675  *
1676  * UDP socket connects are synchronous, but we use a work queue anyway
1677  * to guarantee that even unprivileged user processes can set up a
1678  * socket on a privileged port.
1679  *
1680  * If a UDP socket connect fails, the delay behavior here prevents
1681  * retry floods (hard mounts).
1682  */
1683 static void xs_connect(struct rpc_task *task)
1684 {
1685         struct rpc_xprt *xprt = task->tk_xprt;
1686         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1687
1688         if (xprt_test_and_set_connecting(xprt))
1689                 return;
1690
1691         if (transport->sock != NULL) {
1692                 dprintk("RPC:       xs_connect delayed xprt %p for %lu "
1693                                 "seconds\n",
1694                                 xprt, xprt->reestablish_timeout / HZ);
1695                 queue_delayed_work(rpciod_workqueue,
1696                                    &transport->connect_worker,
1697                                    xprt->reestablish_timeout);
1698                 xprt->reestablish_timeout <<= 1;
1699                 if (xprt->reestablish_timeout > XS_TCP_MAX_REEST_TO)
1700                         xprt->reestablish_timeout = XS_TCP_MAX_REEST_TO;
1701         } else {
1702                 dprintk("RPC:       xs_connect scheduled xprt %p\n", xprt);
1703                 queue_delayed_work(rpciod_workqueue,
1704                                    &transport->connect_worker, 0);
1705         }
1706 }
1707
1708 /**
1709  * xs_udp_print_stats - display UDP socket-specifc stats
1710  * @xprt: rpc_xprt struct containing statistics
1711  * @seq: output file
1712  *
1713  */
1714 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
1715 {
1716         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1717
1718         seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %Lu %Lu\n",
1719                         transport->port,
1720                         xprt->stat.bind_count,
1721                         xprt->stat.sends,
1722                         xprt->stat.recvs,
1723                         xprt->stat.bad_xids,
1724                         xprt->stat.req_u,
1725                         xprt->stat.bklog_u);
1726 }
1727
1728 /**
1729  * xs_tcp_print_stats - display TCP socket-specifc stats
1730  * @xprt: rpc_xprt struct containing statistics
1731  * @seq: output file
1732  *
1733  */
1734 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
1735 {
1736         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1737         long idle_time = 0;
1738
1739         if (xprt_connected(xprt))
1740                 idle_time = (long)(jiffies - xprt->last_used) / HZ;
1741
1742         seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu %Lu %Lu\n",
1743                         transport->port,
1744                         xprt->stat.bind_count,
1745                         xprt->stat.connect_count,
1746                         xprt->stat.connect_time,
1747                         idle_time,
1748                         xprt->stat.sends,
1749                         xprt->stat.recvs,
1750                         xprt->stat.bad_xids,
1751                         xprt->stat.req_u,
1752                         xprt->stat.bklog_u);
1753 }
1754
1755 static struct rpc_xprt_ops xs_udp_ops = {
1756         .set_buffer_size        = xs_udp_set_buffer_size,
1757         .reserve_xprt           = xprt_reserve_xprt_cong,
1758         .release_xprt           = xprt_release_xprt_cong,
1759         .rpcbind                = rpcb_getport_async,
1760         .set_port               = xs_set_port,
1761         .connect                = xs_connect,
1762         .buf_alloc              = rpc_malloc,
1763         .buf_free               = rpc_free,
1764         .send_request           = xs_udp_send_request,
1765         .set_retrans_timeout    = xprt_set_retrans_timeout_rtt,
1766         .timer                  = xs_udp_timer,
1767         .release_request        = xprt_release_rqst_cong,
1768         .close                  = xs_close,
1769         .destroy                = xs_destroy,
1770         .print_stats            = xs_udp_print_stats,
1771 };
1772
1773 static struct rpc_xprt_ops xs_tcp_ops = {
1774         .reserve_xprt           = xprt_reserve_xprt,
1775         .release_xprt           = xs_tcp_release_xprt,
1776         .rpcbind                = rpcb_getport_async,
1777         .set_port               = xs_set_port,
1778         .connect                = xs_connect,
1779         .buf_alloc              = rpc_malloc,
1780         .buf_free               = rpc_free,
1781         .send_request           = xs_tcp_send_request,
1782         .set_retrans_timeout    = xprt_set_retrans_timeout_def,
1783         .close                  = xs_close,
1784         .destroy                = xs_destroy,
1785         .print_stats            = xs_tcp_print_stats,
1786 };
1787
1788 static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
1789                                       unsigned int slot_table_size)
1790 {
1791         struct rpc_xprt *xprt;
1792         struct sock_xprt *new;
1793
1794         if (args->addrlen > sizeof(xprt->addr)) {
1795                 dprintk("RPC:       xs_setup_xprt: address too large\n");
1796                 return ERR_PTR(-EBADF);
1797         }
1798
1799         new = kzalloc(sizeof(*new), GFP_KERNEL);
1800         if (new == NULL) {
1801                 dprintk("RPC:       xs_setup_xprt: couldn't allocate "
1802                                 "rpc_xprt\n");
1803                 return ERR_PTR(-ENOMEM);
1804         }
1805         xprt = &new->xprt;
1806
1807         xprt->max_reqs = slot_table_size;
1808         xprt->slot = kcalloc(xprt->max_reqs, sizeof(struct rpc_rqst), GFP_KERNEL);
1809         if (xprt->slot == NULL) {
1810                 kfree(xprt);
1811                 dprintk("RPC:       xs_setup_xprt: couldn't allocate slot "
1812                                 "table\n");
1813                 return ERR_PTR(-ENOMEM);
1814         }
1815
1816         memcpy(&xprt->addr, args->dstaddr, args->addrlen);
1817         xprt->addrlen = args->addrlen;
1818         if (args->srcaddr)
1819                 memcpy(&new->addr, args->srcaddr, args->addrlen);
1820         new->port = xs_get_random_port();
1821
1822         return xprt;
1823 }
1824
1825 /**
1826  * xs_setup_udp - Set up transport to use a UDP socket
1827  * @args: rpc transport creation arguments
1828  *
1829  */
1830 struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
1831 {
1832         struct sockaddr *addr = args->dstaddr;
1833         struct rpc_xprt *xprt;
1834         struct sock_xprt *transport;
1835
1836         xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries);
1837         if (IS_ERR(xprt))
1838                 return xprt;
1839         transport = container_of(xprt, struct sock_xprt, xprt);
1840
1841         xprt->prot = IPPROTO_UDP;
1842         xprt->tsh_size = 0;
1843         /* XXX: header size can vary due to auth type, IPv6, etc. */
1844         xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
1845
1846         xprt->bind_timeout = XS_BIND_TO;
1847         xprt->connect_timeout = XS_UDP_CONN_TO;
1848         xprt->reestablish_timeout = XS_UDP_REEST_TO;
1849         xprt->idle_timeout = XS_IDLE_DISC_TO;
1850
1851         xprt->ops = &xs_udp_ops;
1852
1853         if (args->timeout)
1854                 xprt->timeout = *args->timeout;
1855         else
1856                 xprt_set_timeout(&xprt->timeout, 5, 5 * HZ);
1857
1858         switch (addr->sa_family) {
1859         case AF_INET:
1860                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
1861                         xprt_set_bound(xprt);
1862
1863                 INIT_DELAYED_WORK(&transport->connect_worker,
1864                                         xs_udp_connect_worker4);
1865                 xs_format_ipv4_peer_addresses(xprt);
1866                 break;
1867         case AF_INET6:
1868                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
1869                         xprt_set_bound(xprt);
1870
1871                 INIT_DELAYED_WORK(&transport->connect_worker,
1872                                         xs_udp_connect_worker6);
1873                 xs_format_ipv6_peer_addresses(xprt);
1874                 break;
1875         default:
1876                 kfree(xprt);
1877                 return ERR_PTR(-EAFNOSUPPORT);
1878         }
1879
1880         dprintk("RPC:       set up transport to address %s\n",
1881                         xprt->address_strings[RPC_DISPLAY_ALL]);
1882
1883         if (try_module_get(THIS_MODULE))
1884                 return xprt;
1885
1886         kfree(xprt->slot);
1887         kfree(xprt);
1888         return ERR_PTR(-EINVAL);
1889 }
1890
1891 /**
1892  * xs_setup_tcp - Set up transport to use a TCP socket
1893  * @args: rpc transport creation arguments
1894  *
1895  */
1896 struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
1897 {
1898         struct sockaddr *addr = args->dstaddr;
1899         struct rpc_xprt *xprt;
1900         struct sock_xprt *transport;
1901
1902         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries);
1903         if (IS_ERR(xprt))
1904                 return xprt;
1905         transport = container_of(xprt, struct sock_xprt, xprt);
1906
1907         xprt->prot = IPPROTO_TCP;
1908         xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
1909         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
1910
1911         xprt->bind_timeout = XS_BIND_TO;
1912         xprt->connect_timeout = XS_TCP_CONN_TO;
1913         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1914         xprt->idle_timeout = XS_IDLE_DISC_TO;
1915
1916         xprt->ops = &xs_tcp_ops;
1917
1918         if (args->timeout)
1919                 xprt->timeout = *args->timeout;
1920         else
1921                 xprt_set_timeout(&xprt->timeout, 2, 60 * HZ);
1922
1923         switch (addr->sa_family) {
1924         case AF_INET:
1925                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
1926                         xprt_set_bound(xprt);
1927
1928                 INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_connect_worker4);
1929                 xs_format_ipv4_peer_addresses(xprt);
1930                 break;
1931         case AF_INET6:
1932                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
1933                         xprt_set_bound(xprt);
1934
1935                 INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_connect_worker6);
1936                 xs_format_ipv6_peer_addresses(xprt);
1937                 break;
1938         default:
1939                 kfree(xprt);
1940                 return ERR_PTR(-EAFNOSUPPORT);
1941         }
1942
1943         dprintk("RPC:       set up transport to address %s\n",
1944                         xprt->address_strings[RPC_DISPLAY_ALL]);
1945
1946         if (try_module_get(THIS_MODULE))
1947                 return xprt;
1948
1949         kfree(xprt->slot);
1950         kfree(xprt);
1951         return ERR_PTR(-EINVAL);
1952 }
1953
1954 static struct xprt_class        xs_udp_transport = {
1955         .list           = LIST_HEAD_INIT(xs_udp_transport.list),
1956         .name           = "udp",
1957         .owner          = THIS_MODULE,
1958         .family         = AF_INET,
1959         .protocol       = 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         .family         = AF_INET,
1968         .protocol       = IPPROTO_TCP,
1969         .setup          = xs_setup_tcp,
1970 };
1971
1972 /**
1973  * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
1974  *
1975  */
1976 int init_socket_xprt(void)
1977 {
1978 #ifdef RPC_DEBUG
1979         if (!sunrpc_table_header)
1980                 sunrpc_table_header = register_sysctl_table(sunrpc_table);
1981 #endif
1982
1983         xprt_register_transport(&xs_udp_transport);
1984         xprt_register_transport(&xs_tcp_transport);
1985
1986         return 0;
1987 }
1988
1989 /**
1990  * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
1991  *
1992  */
1993 void cleanup_socket_xprt(void)
1994 {
1995 #ifdef RPC_DEBUG
1996         if (sunrpc_table_header) {
1997                 unregister_sysctl_table(sunrpc_table_header);
1998                 sunrpc_table_header = NULL;
1999         }
2000 #endif
2001
2002         xprt_unregister_transport(&xs_udp_transport);
2003         xprt_unregister_transport(&xs_tcp_transport);
2004 }