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                                           const char *protocol,
 285                                           const char *netid)
 286 {
 287         struct sockaddr_in *addr = xs_addr_in(xprt);
 288         char *buf;
 289
 290         buf = kzalloc(20, GFP_KERNEL);
 291         if (buf) {
 292                 snprintf(buf, 20, NIPQUAD_FMT,
 293                                 NIPQUAD(addr->sin_addr.s_addr));
 294         }
 295         xprt->address_strings[RPC_DISPLAY_ADDR] = buf;
 296
 297         buf = kzalloc(8, GFP_KERNEL);
 298         if (buf) {
 299                 snprintf(buf, 8, "%u",
 300                                 ntohs(addr->sin_port));
 301         }
 302         xprt->address_strings[RPC_DISPLAY_PORT] = buf;
 303
 304         xprt->address_strings[RPC_DISPLAY_PROTO] = protocol;
 305
 306         buf = kzalloc(48, GFP_KERNEL);
 307         if (buf) {
 308                 snprintf(buf, 48, "addr="NIPQUAD_FMT" port=%u proto=%s",
 309                         NIPQUAD(addr->sin_addr.s_addr),
 310                         ntohs(addr->sin_port),
 311                         protocol);
 312         }
 313         xprt->address_strings[RPC_DISPLAY_ALL] = buf;
 314
 315         buf = kzalloc(10, GFP_KERNEL);
 316         if (buf) {
 317                 snprintf(buf, 10, "%02x%02x%02x%02x",
 318                                 NIPQUAD(addr->sin_addr.s_addr));
 319         }
 320         xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = buf;
 321
 322         buf = kzalloc(8, GFP_KERNEL);
 323         if (buf) {
 324                 snprintf(buf, 8, "%4hx",
 325                                 ntohs(addr->sin_port));
 326         }
 327         xprt->address_strings[RPC_DISPLAY_HEX_PORT] = buf;
 328
 329         buf = kzalloc(30, GFP_KERNEL);
 330         if (buf) {
 331                 snprintf(buf, 30, NIPQUAD_FMT".%u.%u",
 332                                 NIPQUAD(addr->sin_addr.s_addr),
 333                                 ntohs(addr->sin_port) >> 8,
 334                                 ntohs(addr->sin_port) & 0xff);
 335         }
 336         xprt->address_strings[RPC_DISPLAY_UNIVERSAL_ADDR] = buf;
 337
 338         xprt->address_strings[RPC_DISPLAY_NETID] = netid;
 339 }
 340
 341 static void xs_format_ipv6_peer_addresses(struct rpc_xprt *xprt,
 342                                           const char *protocol,
 343                                           const char *netid)
 344 {
 345         struct sockaddr_in6 *addr = xs_addr_in6(xprt);
 346         char *buf;
 347
 348         buf = kzalloc(40, GFP_KERNEL);
 349         if (buf) {
 350                 snprintf(buf, 40, NIP6_FMT,
 351                                 NIP6(addr->sin6_addr));
 352         }
 353         xprt->address_strings[RPC_DISPLAY_ADDR] = buf;
 354
 355         buf = kzalloc(8, GFP_KERNEL);
 356         if (buf) {
 357                 snprintf(buf, 8, "%u",
 358                                 ntohs(addr->sin6_port));
 359         }
 360         xprt->address_strings[RPC_DISPLAY_PORT] = buf;
 361
 362         xprt->address_strings[RPC_DISPLAY_PROTO] = protocol;
 363
 364         buf = kzalloc(64, GFP_KERNEL);
 365         if (buf) {
 366                 snprintf(buf, 64, "addr="NIP6_FMT" port=%u proto=%s",
 367                                 NIP6(addr->sin6_addr),
 368                                 ntohs(addr->sin6_port),
 369                                 protocol);
 370         }
 371         xprt->address_strings[RPC_DISPLAY_ALL] = buf;
 372
 373         buf = kzalloc(36, GFP_KERNEL);
 374         if (buf) {
 375                 snprintf(buf, 36, NIP6_SEQFMT,
 376                                 NIP6(addr->sin6_addr));
 377         }
 378         xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = buf;
 379
 380         buf = kzalloc(8, GFP_KERNEL);
 381         if (buf) {
 382                 snprintf(buf, 8, "%4hx",
 383                                 ntohs(addr->sin6_port));
 384         }
 385         xprt->address_strings[RPC_DISPLAY_HEX_PORT] = buf;
 386
 387         buf = kzalloc(50, GFP_KERNEL);
 388         if (buf) {
 389                 snprintf(buf, 50, NIP6_FMT".%u.%u",
 390                                 NIP6(addr->sin6_addr),
 391                                 ntohs(addr->sin6_port) >> 8,
 392                                 ntohs(addr->sin6_port) & 0xff);
 393         }
 394         xprt->address_strings[RPC_DISPLAY_UNIVERSAL_ADDR] = buf;
 395
 396         xprt->address_strings[RPC_DISPLAY_NETID] = netid;
 397 }
 398
 399 static void xs_free_peer_addresses(struct rpc_xprt *xprt)
 400 {
 401         unsigned int i;
 402
 403         for (i = 0; i < RPC_DISPLAY_MAX; i++)
 404                 switch (i) {
 405                 case RPC_DISPLAY_PROTO:
 406                 case RPC_DISPLAY_NETID:
 407                         continue;
 408                 default:
 409                         kfree(xprt->address_strings[i]);
 410                 }
 411 }
 412
 413 #define XS_SENDMSG_FLAGS        (MSG_DONTWAIT | MSG_NOSIGNAL)
 414
 415 static int xs_send_kvec(struct socket *sock, struct sockaddr *addr, int addrlen, struct kvec *vec, unsigned int base, int more)
 416 {
 417         struct msghdr msg = {
 418                 .msg_name       = addr,
 419                 .msg_namelen    = addrlen,
 420                 .msg_flags      = XS_SENDMSG_FLAGS | (more ? MSG_MORE : 0),
 421         };
 422         struct kvec iov = {
 423                 .iov_base       = vec->iov_base + base,
 424                 .iov_len        = vec->iov_len - base,
 425         };
 426
 427         if (iov.iov_len != 0)
 428                 return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
 429         return kernel_sendmsg(sock, &msg, NULL, 0, 0);
 430 }
 431
 432 static int xs_send_pagedata(struct socket *sock, struct xdr_buf *xdr, unsigned int base, int more)
 433 {
 434         struct page **ppage;
 435         unsigned int remainder;
 436         int err, sent = 0;
 437
 438         remainder = xdr->page_len - base;
 439         base += xdr->page_base;
 440         ppage = xdr->pages + (base >> PAGE_SHIFT);
 441         base &= ~PAGE_MASK;
 442         for(;;) {
 443                 unsigned int len = min_t(unsigned int, PAGE_SIZE - base, remainder);
 444                 int flags = XS_SENDMSG_FLAGS;
 445
 446                 remainder -= len;
 447                 if (remainder != 0 || more)
 448                         flags |= MSG_MORE;
 449                 err = sock->ops->sendpage(sock, *ppage, base, len, flags);
 450                 if (remainder == 0 || err != len)
 451                         break;
 452                 sent += err;
 453                 ppage++;
 454                 base = 0;
 455         }
 456         if (sent == 0)
 457                 return err;
 458         if (err > 0)
 459                 sent += err;
 460         return sent;
 461 }
 462
 463 /**
 464  * xs_sendpages - write pages directly to a socket
 465  * @sock: socket to send on
 466  * @addr: UDP only -- address of destination
 467  * @addrlen: UDP only -- length of destination address
 468  * @xdr: buffer containing this request
 469  * @base: starting position in the buffer
 470  *
 471  */
 472 static int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base)
 473 {
 474         unsigned int remainder = xdr->len - base;
 475         int err, sent = 0;
 476
 477         if (unlikely(!sock))
 478                 return -ENOTCONN;
 479
 480         clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags);
 481         if (base != 0) {
 482                 addr = NULL;
 483                 addrlen = 0;
 484         }
 485
 486         if (base < xdr->head[0].iov_len || addr != NULL) {
 487                 unsigned int len = xdr->head[0].iov_len - base;
 488                 remainder -= len;
 489                 err = xs_send_kvec(sock, addr, addrlen, &xdr->head[0], base, remainder != 0);
 490                 if (remainder == 0 || err != len)
 491                         goto out;
 492                 sent += err;
 493                 base = 0;
 494         } else
 495                 base -= xdr->head[0].iov_len;
 496
 497         if (base < xdr->page_len) {
 498                 unsigned int len = xdr->page_len - base;
 499                 remainder -= len;
 500                 err = xs_send_pagedata(sock, xdr, base, remainder != 0);
 501                 if (remainder == 0 || err != len)
 502                         goto out;
 503                 sent += err;
 504                 base = 0;
 505         } else
 506                 base -= xdr->page_len;
 507
 508         if (base >= xdr->tail[0].iov_len)
 509                 return sent;
 510         err = xs_send_kvec(sock, NULL, 0, &xdr->tail[0], base, 0);
 511 out:
 512         if (sent == 0)
 513                 return err;
 514         if (err > 0)
 515                 sent += err;
 516         return sent;
 517 }
 518
 519 /**
 520  * xs_nospace - place task on wait queue if transmit was incomplete
 521  * @task: task to put to sleep
 522  *
 523  */
 524 static void xs_nospace(struct rpc_task *task)
 525 {
 526         struct rpc_rqst *req = task->tk_rqstp;
 527         struct rpc_xprt *xprt = req->rq_xprt;
 528         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 529
 530         dprintk("RPC: %5u xmit incomplete (%u left of %u)\n",
 531                         task->tk_pid, req->rq_slen - req->rq_bytes_sent,
 532                         req->rq_slen);
 533
 534         if (test_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags)) {
 535                 /* Protect against races with write_space */
 536                 spin_lock_bh(&xprt->transport_lock);
 537
 538                 /* Don't race with disconnect */
 539                 if (!xprt_connected(xprt))
 540                         task->tk_status = -ENOTCONN;
 541                 else if (test_bit(SOCK_NOSPACE, &transport->sock->flags))
 542                         xprt_wait_for_buffer_space(task);
 543
 544                 spin_unlock_bh(&xprt->transport_lock);
 545         } else
 546                 /* Keep holding the socket if it is blocked */
 547                 rpc_delay(task, HZ>>4);
 548 }
 549
 550 /**
 551  * xs_udp_send_request - write an RPC request to a UDP socket
 552  * @task: address of RPC task that manages the state of an RPC request
 553  *
 554  * Return values:
 555  *        0:    The request has been sent
 556  *   EAGAIN:    The socket was blocked, please call again later to
 557  *              complete the request
 558  * ENOTCONN:    Caller needs to invoke connect logic then call again
 559  *    other:    Some other error occured, the request was not sent
 560  */
 561 static int xs_udp_send_request(struct rpc_task *task)
 562 {
 563         struct rpc_rqst *req = task->tk_rqstp;
 564         struct rpc_xprt *xprt = req->rq_xprt;
 565         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 566         struct xdr_buf *xdr = &req->rq_snd_buf;
 567         int status;
 568
 569         xs_pktdump("packet data:",
 570                                 req->rq_svec->iov_base,
 571                                 req->rq_svec->iov_len);
 572
 573         req->rq_xtime = jiffies;
 574         status = xs_sendpages(transport->sock,
 575                               xs_addr(xprt),
 576                               xprt->addrlen, xdr,
 577                               req->rq_bytes_sent);
 578
 579         dprintk("RPC:       xs_udp_send_request(%u) = %d\n",
 580                         xdr->len - req->rq_bytes_sent, status);
 581
 582         if (status >= 0) {
 583                 task->tk_bytes_sent += status;
 584                 if (status >= req->rq_slen)
 585                         return 0;
 586                 /* Still some bytes left; set up for a retry later. */
 587                 status = -EAGAIN;
 588         }
 589
 590         switch (status) {
 591         case -ENETUNREACH:
 592         case -EPIPE:
 593         case -ECONNREFUSED:
 594                 /* When the server has died, an ICMP port unreachable message
 595                  * prompts ECONNREFUSED. */
 596                 break;
 597         case -EAGAIN:
 598                 xs_nospace(task);
 599                 break;
 600         default:
 601                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
 602                         -status);
 603                 break;
 604         }
 605
 606         return status;
 607 }
 608
 609 /**
 610  * xs_tcp_shutdown - gracefully shut down a TCP socket
 611  * @xprt: transport
 612  *
 613  * Initiates a graceful shutdown of the TCP socket by calling the
 614  * equivalent of shutdown(SHUT_WR);
 615  */
 616 static void xs_tcp_shutdown(struct rpc_xprt *xprt)
 617 {
 618         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 619         struct socket *sock = transport->sock;
 620
 621         if (sock != NULL)
 622                 kernel_sock_shutdown(sock, SHUT_WR);
 623 }
 624
 625 static inline void xs_encode_tcp_record_marker(struct xdr_buf *buf)
 626 {
 627         u32 reclen = buf->len - sizeof(rpc_fraghdr);
 628         rpc_fraghdr *base = buf->head[0].iov_base;
 629         *base = htonl(RPC_LAST_STREAM_FRAGMENT | reclen);
 630 }
 631
 632 /**
 633  * xs_tcp_send_request - write an RPC request to a TCP socket
 634  * @task: address of RPC task that manages the state of an RPC request
 635  *
 636  * Return values:
 637  *        0:    The request has been sent
 638  *   EAGAIN:    The socket was blocked, please call again later to
 639  *              complete the request
 640  * ENOTCONN:    Caller needs to invoke connect logic then call again
 641  *    other:    Some other error occured, the request was not sent
 642  *
 643  * XXX: In the case of soft timeouts, should we eventually give up
 644  *      if sendmsg is not able to make progress?
 645  */
 646 static int xs_tcp_send_request(struct rpc_task *task)
 647 {
 648         struct rpc_rqst *req = task->tk_rqstp;
 649         struct rpc_xprt *xprt = req->rq_xprt;
 650         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 651         struct xdr_buf *xdr = &req->rq_snd_buf;
 652         int status;
 653         unsigned int retry = 0;
 654
 655         xs_encode_tcp_record_marker(&req->rq_snd_buf);
 656
 657         xs_pktdump("packet data:",
 658                                 req->rq_svec->iov_base,
 659                                 req->rq_svec->iov_len);
 660
 661         /* Continue transmitting the packet/record. We must be careful
 662          * to cope with writespace callbacks arriving _after_ we have
 663          * called sendmsg(). */
 664         while (1) {
 665                 req->rq_xtime = jiffies;
 666                 status = xs_sendpages(transport->sock,
 667                                         NULL, 0, xdr, req->rq_bytes_sent);
 668
 669                 dprintk("RPC:       xs_tcp_send_request(%u) = %d\n",
 670                                 xdr->len - req->rq_bytes_sent, status);
 671
 672                 if (unlikely(status < 0))
 673                         break;
 674
 675                 /* If we've sent the entire packet, immediately
 676                  * reset the count of bytes sent. */
 677                 req->rq_bytes_sent += status;
 678                 task->tk_bytes_sent += status;
 679                 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
 680                         req->rq_bytes_sent = 0;
 681                         return 0;
 682                 }
 683
 684                 status = -EAGAIN;
 685                 if (retry++ > XS_SENDMSG_RETRY)
 686                         break;
 687         }
 688
 689         switch (status) {
 690         case -EAGAIN:
 691                 xs_nospace(task);
 692                 break;
 693         case -ECONNREFUSED:
 694         case -ECONNRESET:
 695         case -ENOTCONN:
 696         case -EPIPE:
 697                 status = -ENOTCONN;
 698                 break;
 699         default:
 700                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
 701                         -status);
 702                 xs_tcp_shutdown(xprt);
 703                 break;
 704         }
 705
 706         return status;
 707 }
 708
 709 /**
 710  * xs_tcp_release_xprt - clean up after a tcp transmission
 711  * @xprt: transport
 712  * @task: rpc task
 713  *
 714  * This cleans up if an error causes us to abort the transmission of a request.
 715  * In this case, the socket may need to be reset in order to avoid confusing
 716  * the server.
 717  */
 718 static void xs_tcp_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
 719 {
 720         struct rpc_rqst *req;
 721
 722         if (task != xprt->snd_task)
 723                 return;
 724         if (task == NULL)
 725                 goto out_release;
 726         req = task->tk_rqstp;
 727         if (req->rq_bytes_sent == 0)
 728                 goto out_release;
 729         if (req->rq_bytes_sent == req->rq_snd_buf.len)
 730                 goto out_release;
 731         set_bit(XPRT_CLOSE_WAIT, &task->tk_xprt->state);
 732 out_release:
 733         xprt_release_xprt(xprt, task);
 734 }
 735
 736 /**
 737  * xs_close - close a socket
 738  * @xprt: transport
 739  *
 740  * This is used when all requests are complete; ie, no DRC state remains
 741  * on the server we want to save.
 742  */
 743 static void xs_close(struct rpc_xprt *xprt)
 744 {
 745         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 746         struct socket *sock = transport->sock;
 747         struct sock *sk = transport->inet;
 748
 749         if (!sk)
 750                 goto clear_close_wait;
 751
 752         dprintk("RPC:       xs_close xprt %p\n", xprt);
 753
 754         write_lock_bh(&sk->sk_callback_lock);
 755         transport->inet = NULL;
 756         transport->sock = NULL;
 757
 758         sk->sk_user_data = NULL;
 759         sk->sk_data_ready = transport->old_data_ready;
 760         sk->sk_state_change = transport->old_state_change;
 761         sk->sk_write_space = transport->old_write_space;
 762         write_unlock_bh(&sk->sk_callback_lock);
 763
 764         sk->sk_no_check = 0;
 765
 766         sock_release(sock);
 767 clear_close_wait:
 768         smp_mb__before_clear_bit();
 769         clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
 770         clear_bit(XPRT_CLOSING, &xprt->state);
 771         smp_mb__after_clear_bit();
 772         xprt_disconnect_done(xprt);
 773 }
 774
 775 /**
 776  * xs_destroy - prepare to shutdown a transport
 777  * @xprt: doomed transport
 778  *
 779  */
 780 static void xs_destroy(struct rpc_xprt *xprt)
 781 {
 782         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 783
 784         dprintk("RPC:       xs_destroy xprt %p\n", xprt);
 785
 786         cancel_rearming_delayed_work(&transport->connect_worker);
 787
 788         xs_close(xprt);
 789         xs_free_peer_addresses(xprt);
 790         kfree(xprt->slot);
 791         kfree(xprt);
 792         module_put(THIS_MODULE);
 793 }
 794
 795 static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
 796 {
 797         return (struct rpc_xprt *) sk->sk_user_data;
 798 }
 799
 800 /**
 801  * xs_udp_data_ready - "data ready" callback for UDP sockets
 802  * @sk: socket with data to read
 803  * @len: how much data to read
 804  *
 805  */
 806 static void xs_udp_data_ready(struct sock *sk, int len)
 807 {
 808         struct rpc_task *task;
 809         struct rpc_xprt *xprt;
 810         struct rpc_rqst *rovr;
 811         struct sk_buff *skb;
 812         int err, repsize, copied;
 813         u32 _xid;
 814         __be32 *xp;
 815
 816         read_lock(&sk->sk_callback_lock);
 817         dprintk("RPC:       xs_udp_data_ready...\n");
 818         if (!(xprt = xprt_from_sock(sk)))
 819                 goto out;
 820
 821         if ((skb = skb_recv_datagram(sk, 0, 1, &err)) == NULL)
 822                 goto out;
 823
 824         if (xprt->shutdown)
 825                 goto dropit;
 826
 827         repsize = skb->len - sizeof(struct udphdr);
 828         if (repsize < 4) {
 829                 dprintk("RPC:       impossible RPC reply size %d!\n", repsize);
 830                 goto dropit;
 831         }
 832
 833         /* Copy the XID from the skb... */
 834         xp = skb_header_pointer(skb, sizeof(struct udphdr),
 835                                 sizeof(_xid), &_xid);
 836         if (xp == NULL)
 837                 goto dropit;
 838
 839         /* Look up and lock the request corresponding to the given XID */
 840         spin_lock(&xprt->transport_lock);
 841         rovr = xprt_lookup_rqst(xprt, *xp);
 842         if (!rovr)
 843                 goto out_unlock;
 844         task = rovr->rq_task;
 845
 846         if ((copied = rovr->rq_private_buf.buflen) > repsize)
 847                 copied = repsize;
 848
 849         /* Suck it into the iovec, verify checksum if not done by hw. */
 850         if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
 851                 UDPX_INC_STATS_BH(sk, UDP_MIB_INERRORS);
 852                 goto out_unlock;
 853         }
 854
 855         UDPX_INC_STATS_BH(sk, UDP_MIB_INDATAGRAMS);
 856
 857         /* Something worked... */
 858         dst_confirm(skb->dst);
 859
 860         xprt_adjust_cwnd(task, copied);
 861         xprt_update_rtt(task);
 862         xprt_complete_rqst(task, copied);
 863
 864  out_unlock:
 865         spin_unlock(&xprt->transport_lock);
 866  dropit:
 867         skb_free_datagram(sk, skb);
 868  out:
 869         read_unlock(&sk->sk_callback_lock);
 870 }
 871
 872 static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
 873 {
 874         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 875         size_t len, used;
 876         char *p;
 877
 878         p = ((char *) &transport->tcp_fraghdr) + transport->tcp_offset;
 879         len = sizeof(transport->tcp_fraghdr) - transport->tcp_offset;
 880         used = xdr_skb_read_bits(desc, p, len);
 881         transport->tcp_offset += used;
 882         if (used != len)
 883                 return;
 884
 885         transport->tcp_reclen = ntohl(transport->tcp_fraghdr);
 886         if (transport->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
 887                 transport->tcp_flags |= TCP_RCV_LAST_FRAG;
 888         else
 889                 transport->tcp_flags &= ~TCP_RCV_LAST_FRAG;
 890         transport->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
 891
 892         transport->tcp_flags &= ~TCP_RCV_COPY_FRAGHDR;
 893         transport->tcp_offset = 0;
 894
 895         /* Sanity check of the record length */
 896         if (unlikely(transport->tcp_reclen < 4)) {
 897                 dprintk("RPC:       invalid TCP record fragment length\n");
 898                 xprt_force_disconnect(xprt);
 899                 return;
 900         }
 901         dprintk("RPC:       reading TCP record fragment of length %d\n",
 902                         transport->tcp_reclen);
 903 }
 904
 905 static void xs_tcp_check_fraghdr(struct sock_xprt *transport)
 906 {
 907         if (transport->tcp_offset == transport->tcp_reclen) {
 908                 transport->tcp_flags |= TCP_RCV_COPY_FRAGHDR;
 909                 transport->tcp_offset = 0;
 910                 if (transport->tcp_flags & TCP_RCV_LAST_FRAG) {
 911                         transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
 912                         transport->tcp_flags |= TCP_RCV_COPY_XID;
 913                         transport->tcp_copied = 0;
 914                 }
 915         }
 916 }
 917
 918 static inline void xs_tcp_read_xid(struct sock_xprt *transport, struct xdr_skb_reader *desc)
 919 {
 920         size_t len, used;
 921         char *p;
 922
 923         len = sizeof(transport->tcp_xid) - transport->tcp_offset;
 924         dprintk("RPC:       reading XID (%Zu bytes)\n", len);
 925         p = ((char *) &transport->tcp_xid) + transport->tcp_offset;
 926         used = xdr_skb_read_bits(desc, p, len);
 927         transport->tcp_offset += used;
 928         if (used != len)
 929                 return;
 930         transport->tcp_flags &= ~TCP_RCV_COPY_XID;
 931         transport->tcp_flags |= TCP_RCV_COPY_DATA;
 932         transport->tcp_copied = 4;
 933         dprintk("RPC:       reading reply for XID %08x\n",
 934                         ntohl(transport->tcp_xid));
 935         xs_tcp_check_fraghdr(transport);
 936 }
 937
 938 static inline void xs_tcp_read_request(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
 939 {
 940         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 941         struct rpc_rqst *req;
 942         struct xdr_buf *rcvbuf;
 943         size_t len;
 944         ssize_t r;
 945
 946         /* Find and lock the request corresponding to this xid */
 947         spin_lock(&xprt->transport_lock);
 948         req = xprt_lookup_rqst(xprt, transport->tcp_xid);
 949         if (!req) {
 950                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
 951                 dprintk("RPC:       XID %08x request not found!\n",
 952                                 ntohl(transport->tcp_xid));
 953                 spin_unlock(&xprt->transport_lock);
 954                 return;
 955         }
 956
 957         rcvbuf = &req->rq_private_buf;
 958         len = desc->count;
 959         if (len > transport->tcp_reclen - transport->tcp_offset) {
 960                 struct xdr_skb_reader my_desc;
 961
 962                 len = transport->tcp_reclen - transport->tcp_offset;
 963                 memcpy(&my_desc, desc, sizeof(my_desc));
 964                 my_desc.count = len;
 965                 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
 966                                           &my_desc, xdr_skb_read_bits);
 967                 desc->count -= r;
 968                 desc->offset += r;
 969         } else
 970                 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
 971                                           desc, xdr_skb_read_bits);
 972
 973         if (r > 0) {
 974                 transport->tcp_copied += r;
 975                 transport->tcp_offset += r;
 976         }
 977         if (r != len) {
 978                 /* Error when copying to the receive buffer,
 979                  * usually because we weren't able to allocate
 980                  * additional buffer pages. All we can do now
 981                  * is turn off TCP_RCV_COPY_DATA, so the request
 982                  * will not receive any additional updates,
 983                  * and time out.
 984                  * Any remaining data from this record will
 985                  * be discarded.
 986                  */
 987                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
 988                 dprintk("RPC:       XID %08x truncated request\n",
 989                                 ntohl(transport->tcp_xid));
 990                 dprintk("RPC:       xprt = %p, tcp_copied = %lu, "
 991                                 "tcp_offset = %u, tcp_reclen = %u\n",
 992                                 xprt, transport->tcp_copied,
 993                                 transport->tcp_offset, transport->tcp_reclen);
 994                 goto out;
 995         }
 996
 997         dprintk("RPC:       XID %08x read %Zd bytes\n",
 998                         ntohl(transport->tcp_xid), r);
 999         dprintk("RPC:       xprt = %p, tcp_copied = %lu, tcp_offset = %u, "
1000                         "tcp_reclen = %u\n", xprt, transport->tcp_copied,
1001                         transport->tcp_offset, transport->tcp_reclen);
1002
1003         if (transport->tcp_copied == req->rq_private_buf.buflen)
1004                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1005         else if (transport->tcp_offset == transport->tcp_reclen) {
1006                 if (transport->tcp_flags & TCP_RCV_LAST_FRAG)
1007                         transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1008         }
1009
1010 out:
1011         if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1012                 xprt_complete_rqst(req->rq_task, transport->tcp_copied);
1013         spin_unlock(&xprt->transport_lock);
1014         xs_tcp_check_fraghdr(transport);
1015 }
1016
1017 static inline void xs_tcp_read_discard(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1018 {
1019         size_t len;
1020
1021         len = transport->tcp_reclen - transport->tcp_offset;
1022         if (len > desc->count)
1023                 len = desc->count;
1024         desc->count -= len;
1025         desc->offset += len;
1026         transport->tcp_offset += len;
1027         dprintk("RPC:       discarded %Zu bytes\n", len);
1028         xs_tcp_check_fraghdr(transport);
1029 }
1030
1031 static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
1032 {
1033         struct rpc_xprt *xprt = rd_desc->arg.data;
1034         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1035         struct xdr_skb_reader desc = {
1036                 .skb    = skb,
1037                 .offset = offset,
1038                 .count  = len,
1039         };
1040
1041         dprintk("RPC:       xs_tcp_data_recv started\n");
1042         do {
1043                 /* Read in a new fragment marker if necessary */
1044                 /* Can we ever really expect to get completely empty fragments? */
1045                 if (transport->tcp_flags & TCP_RCV_COPY_FRAGHDR) {
1046                         xs_tcp_read_fraghdr(xprt, &desc);
1047                         continue;
1048                 }
1049                 /* Read in the xid if necessary */
1050                 if (transport->tcp_flags & TCP_RCV_COPY_XID) {
1051                         xs_tcp_read_xid(transport, &desc);
1052                         continue;
1053                 }
1054                 /* Read in the request data */
1055                 if (transport->tcp_flags & TCP_RCV_COPY_DATA) {
1056                         xs_tcp_read_request(xprt, &desc);
1057                         continue;
1058                 }
1059                 /* Skip over any trailing bytes on short reads */
1060                 xs_tcp_read_discard(transport, &desc);
1061         } while (desc.count);
1062         dprintk("RPC:       xs_tcp_data_recv done\n");
1063         return len - desc.count;
1064 }
1065
1066 /**
1067  * xs_tcp_data_ready - "data ready" callback for TCP sockets
1068  * @sk: socket with data to read
1069  * @bytes: how much data to read
1070  *
1071  */
1072 static void xs_tcp_data_ready(struct sock *sk, int bytes)
1073 {
1074         struct rpc_xprt *xprt;
1075         read_descriptor_t rd_desc;
1076
1077         dprintk("RPC:       xs_tcp_data_ready...\n");
1078
1079         read_lock(&sk->sk_callback_lock);
1080         if (!(xprt = xprt_from_sock(sk)))
1081                 goto out;
1082         if (xprt->shutdown)
1083                 goto out;
1084
1085         /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
1086         rd_desc.arg.data = xprt;
1087         rd_desc.count = 65536;
1088         tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
1089 out:
1090         read_unlock(&sk->sk_callback_lock);
1091 }
1092
1093 /**
1094  * xs_tcp_state_change - callback to handle TCP socket state changes
1095  * @sk: socket whose state has changed
1096  *
1097  */
1098 static void xs_tcp_state_change(struct sock *sk)
1099 {
1100         struct rpc_xprt *xprt;
1101
1102         read_lock(&sk->sk_callback_lock);
1103         if (!(xprt = xprt_from_sock(sk)))
1104                 goto out;
1105         dprintk("RPC:       xs_tcp_state_change client %p...\n", xprt);
1106         dprintk("RPC:       state %x conn %d dead %d zapped %d\n",
1107                         sk->sk_state, xprt_connected(xprt),
1108                         sock_flag(sk, SOCK_DEAD),
1109                         sock_flag(sk, SOCK_ZAPPED));
1110
1111         switch (sk->sk_state) {
1112         case TCP_ESTABLISHED:
1113                 spin_lock_bh(&xprt->transport_lock);
1114                 if (!xprt_test_and_set_connected(xprt)) {
1115                         struct sock_xprt *transport = container_of(xprt,
1116                                         struct sock_xprt, xprt);
1117
1118                         /* Reset TCP record info */
1119                         transport->tcp_offset = 0;
1120                         transport->tcp_reclen = 0;
1121                         transport->tcp_copied = 0;
1122                         transport->tcp_flags =
1123                                 TCP_RCV_COPY_FRAGHDR | TCP_RCV_COPY_XID;
1124
1125                         xprt_wake_pending_tasks(xprt, 0);
1126                 }
1127                 spin_unlock_bh(&xprt->transport_lock);
1128                 break;
1129         case TCP_FIN_WAIT1:
1130                 /* The client initiated a shutdown of the socket */
1131                 xprt->reestablish_timeout = 0;
1132                 set_bit(XPRT_CLOSING, &xprt->state);
1133                 smp_mb__before_clear_bit();
1134                 clear_bit(XPRT_CONNECTED, &xprt->state);
1135                 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1136                 smp_mb__after_clear_bit();
1137                 break;
1138         case TCP_CLOSE_WAIT:
1139                 /* The server initiated a shutdown of the socket */
1140                 set_bit(XPRT_CLOSING, &xprt->state);
1141                 xprt_force_disconnect(xprt);
1142         case TCP_SYN_SENT:
1143         case TCP_CLOSING:
1144                 /*
1145                  * If the server closed down the connection, make sure that
1146                  * we back off before reconnecting
1147                  */
1148                 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1149                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1150                 break;
1151         case TCP_LAST_ACK:
1152                 smp_mb__before_clear_bit();
1153                 clear_bit(XPRT_CONNECTED, &xprt->state);
1154                 smp_mb__after_clear_bit();
1155                 break;
1156         case TCP_CLOSE:
1157                 smp_mb__before_clear_bit();
1158                 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1159                 clear_bit(XPRT_CLOSING, &xprt->state);
1160                 smp_mb__after_clear_bit();
1161                 /* Mark transport as closed and wake up all pending tasks */
1162                 xprt_disconnect_done(xprt);
1163         }
1164  out:
1165         read_unlock(&sk->sk_callback_lock);
1166 }
1167
1168 /**
1169  * xs_udp_write_space - callback invoked when socket buffer space
1170  *                             becomes available
1171  * @sk: socket whose state has changed
1172  *
1173  * Called when more output buffer space is available for this socket.
1174  * We try not to wake our writers until they can make "significant"
1175  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1176  * with a bunch of small requests.
1177  */
1178 static void xs_udp_write_space(struct sock *sk)
1179 {
1180         read_lock(&sk->sk_callback_lock);
1181
1182         /* from net/core/sock.c:sock_def_write_space */
1183         if (sock_writeable(sk)) {
1184                 struct socket *sock;
1185                 struct rpc_xprt *xprt;
1186
1187                 if (unlikely(!(sock = sk->sk_socket)))
1188                         goto out;
1189                 if (unlikely(!(xprt = xprt_from_sock(sk))))
1190                         goto out;
1191                 if (unlikely(!test_and_clear_bit(SOCK_NOSPACE, &sock->flags)))
1192                         goto out;
1193
1194                 xprt_write_space(xprt);
1195         }
1196
1197  out:
1198         read_unlock(&sk->sk_callback_lock);
1199 }
1200
1201 /**
1202  * xs_tcp_write_space - callback invoked when socket buffer space
1203  *                             becomes available
1204  * @sk: socket whose state has changed
1205  *
1206  * Called when more output buffer space is available for this socket.
1207  * We try not to wake our writers until they can make "significant"
1208  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1209  * with a bunch of small requests.
1210  */
1211 static void xs_tcp_write_space(struct sock *sk)
1212 {
1213         read_lock(&sk->sk_callback_lock);
1214
1215         /* from net/core/stream.c:sk_stream_write_space */
1216         if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk)) {
1217                 struct socket *sock;
1218                 struct rpc_xprt *xprt;
1219
1220                 if (unlikely(!(sock = sk->sk_socket)))
1221                         goto out;
1222                 if (unlikely(!(xprt = xprt_from_sock(sk))))
1223                         goto out;
1224                 if (unlikely(!test_and_clear_bit(SOCK_NOSPACE, &sock->flags)))
1225                         goto out;
1226
1227                 xprt_write_space(xprt);
1228         }
1229
1230  out:
1231         read_unlock(&sk->sk_callback_lock);
1232 }
1233
1234 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
1235 {
1236         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1237         struct sock *sk = transport->inet;
1238
1239         if (transport->rcvsize) {
1240                 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1241                 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
1242         }
1243         if (transport->sndsize) {
1244                 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1245                 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1246                 sk->sk_write_space(sk);
1247         }
1248 }
1249
1250 /**
1251  * xs_udp_set_buffer_size - set send and receive limits
1252  * @xprt: generic transport
1253  * @sndsize: requested size of send buffer, in bytes
1254  * @rcvsize: requested size of receive buffer, in bytes
1255  *
1256  * Set socket send and receive buffer size limits.
1257  */
1258 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1259 {
1260         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1261
1262         transport->sndsize = 0;
1263         if (sndsize)
1264                 transport->sndsize = sndsize + 1024;
1265         transport->rcvsize = 0;
1266         if (rcvsize)
1267                 transport->rcvsize = rcvsize + 1024;
1268
1269         xs_udp_do_set_buffer_size(xprt);
1270 }
1271
1272 /**
1273  * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1274  * @task: task that timed out
1275  *
1276  * Adjust the congestion window after a retransmit timeout has occurred.
1277  */
1278 static void xs_udp_timer(struct rpc_task *task)
1279 {
1280         xprt_adjust_cwnd(task, -ETIMEDOUT);
1281 }
1282
1283 static unsigned short xs_get_random_port(void)
1284 {
1285         unsigned short range = xprt_max_resvport - xprt_min_resvport;
1286         unsigned short rand = (unsigned short) net_random() % range;
1287         return rand + xprt_min_resvport;
1288 }
1289
1290 /**
1291  * xs_set_port - reset the port number in the remote endpoint address
1292  * @xprt: generic transport
1293  * @port: new port number
1294  *
1295  */
1296 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1297 {
1298         struct sockaddr *addr = xs_addr(xprt);
1299
1300         dprintk("RPC:       setting port for xprt %p to %u\n", xprt, port);
1301
1302         switch (addr->sa_family) {
1303         case AF_INET:
1304                 ((struct sockaddr_in *)addr)->sin_port = htons(port);
1305                 break;
1306         case AF_INET6:
1307                 ((struct sockaddr_in6 *)addr)->sin6_port = htons(port);
1308                 break;
1309         default:
1310                 BUG();
1311         }
1312 }
1313
1314 static unsigned short xs_get_srcport(struct sock_xprt *transport, struct socket *sock)
1315 {
1316         unsigned short port = transport->port;
1317
1318         if (port == 0 && transport->xprt.resvport)
1319                 port = xs_get_random_port();
1320         return port;
1321 }
1322
1323 static unsigned short xs_next_srcport(struct sock_xprt *transport, struct socket *sock, unsigned short port)
1324 {
1325         if (transport->port != 0)
1326                 transport->port = 0;
1327         if (!transport->xprt.resvport)
1328                 return 0;
1329         if (port <= xprt_min_resvport || port > xprt_max_resvport)
1330                 return xprt_max_resvport;
1331         return --port;
1332 }
1333
1334 static int xs_bind4(struct sock_xprt *transport, struct socket *sock)
1335 {
1336         struct sockaddr_in myaddr = {
1337                 .sin_family = AF_INET,
1338         };
1339         struct sockaddr_in *sa;
1340         int err, nloop = 0;
1341         unsigned short port = xs_get_srcport(transport, sock);
1342         unsigned short last;
1343
1344         sa = (struct sockaddr_in *)&transport->addr;
1345         myaddr.sin_addr = sa->sin_addr;
1346         do {
1347                 myaddr.sin_port = htons(port);
1348                 err = kernel_bind(sock, (struct sockaddr *) &myaddr,
1349                                                 sizeof(myaddr));
1350                 if (port == 0)
1351                         break;
1352                 if (err == 0) {
1353                         transport->port = port;
1354                         break;
1355                 }
1356                 last = port;
1357                 port = xs_next_srcport(transport, sock, port);
1358                 if (port > last)
1359                         nloop++;
1360         } while (err == -EADDRINUSE && nloop != 2);
1361         dprintk("RPC:       %s "NIPQUAD_FMT":%u: %s (%d)\n",
1362                         __FUNCTION__, NIPQUAD(myaddr.sin_addr),
1363                         port, err ? "failed" : "ok", err);
1364         return err;
1365 }
1366
1367 static int xs_bind6(struct sock_xprt *transport, struct socket *sock)
1368 {
1369         struct sockaddr_in6 myaddr = {
1370                 .sin6_family = AF_INET6,
1371         };
1372         struct sockaddr_in6 *sa;
1373         int err, nloop = 0;
1374         unsigned short port = xs_get_srcport(transport, sock);
1375         unsigned short last;
1376
1377         sa = (struct sockaddr_in6 *)&transport->addr;
1378         myaddr.sin6_addr = sa->sin6_addr;
1379         do {
1380                 myaddr.sin6_port = htons(port);
1381                 err = kernel_bind(sock, (struct sockaddr *) &myaddr,
1382                                                 sizeof(myaddr));
1383                 if (port == 0)
1384                         break;
1385                 if (err == 0) {
1386                         transport->port = port;
1387                         break;
1388                 }
1389                 last = port;
1390                 port = xs_next_srcport(transport, sock, port);
1391                 if (port > last)
1392                         nloop++;
1393         } while (err == -EADDRINUSE && nloop != 2);
1394         dprintk("RPC:       xs_bind6 "NIP6_FMT":%u: %s (%d)\n",
1395                 NIP6(myaddr.sin6_addr), port, err ? "failed" : "ok", err);
1396         return err;
1397 }
1398
1399 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1400 static struct lock_class_key xs_key[2];
1401 static struct lock_class_key xs_slock_key[2];
1402
1403 static inline void xs_reclassify_socket4(struct socket *sock)
1404 {
1405         struct sock *sk = sock->sk;
1406
1407         BUG_ON(sock_owned_by_user(sk));
1408         sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1409                 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1410 }
1411
1412 static inline void xs_reclassify_socket6(struct socket *sock)
1413 {
1414         struct sock *sk = sock->sk;
1415
1416         BUG_ON(sock_owned_by_user(sk));
1417         sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1418                 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
1419 }
1420 #else
1421 static inline void xs_reclassify_socket4(struct socket *sock)
1422 {
1423 }
1424
1425 static inline void xs_reclassify_socket6(struct socket *sock)
1426 {
1427 }
1428 #endif
1429
1430 static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
1431 {
1432         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1433
1434         if (!transport->inet) {
1435                 struct sock *sk = sock->sk;
1436
1437                 write_lock_bh(&sk->sk_callback_lock);
1438
1439                 sk->sk_user_data = xprt;
1440                 transport->old_data_ready = sk->sk_data_ready;
1441                 transport->old_state_change = sk->sk_state_change;
1442                 transport->old_write_space = sk->sk_write_space;
1443                 sk->sk_data_ready = xs_udp_data_ready;
1444                 sk->sk_write_space = xs_udp_write_space;
1445                 sk->sk_no_check = UDP_CSUM_NORCV;
1446                 sk->sk_allocation = GFP_ATOMIC;
1447
1448                 xprt_set_connected(xprt);
1449
1450                 /* Reset to new socket */
1451                 transport->sock = sock;
1452                 transport->inet = sk;
1453
1454                 write_unlock_bh(&sk->sk_callback_lock);
1455         }
1456         xs_udp_do_set_buffer_size(xprt);
1457 }
1458
1459 /**
1460  * xs_udp_connect_worker4 - set up a UDP socket
1461  * @work: RPC transport to connect
1462  *
1463  * Invoked by a work queue tasklet.
1464  */
1465 static void xs_udp_connect_worker4(struct work_struct *work)
1466 {
1467         struct sock_xprt *transport =
1468                 container_of(work, struct sock_xprt, connect_worker.work);
1469         struct rpc_xprt *xprt = &transport->xprt;
1470         struct socket *sock = transport->sock;
1471         int err, status = -EIO;
1472
1473         if (xprt->shutdown || !xprt_bound(xprt))
1474                 goto out;
1475
1476         /* Start by resetting any existing state */
1477         xs_close(xprt);
1478
1479         if ((err = sock_create_kern(PF_INET, SOCK_DGRAM, IPPROTO_UDP, &sock)) < 0) {
1480                 dprintk("RPC:       can't create UDP transport socket (%d).\n", -err);
1481                 goto out;
1482         }
1483         xs_reclassify_socket4(sock);
1484
1485         if (xs_bind4(transport, sock)) {
1486                 sock_release(sock);
1487                 goto out;
1488         }
1489
1490         dprintk("RPC:       worker connecting xprt %p to address: %s\n",
1491                         xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1492
1493         xs_udp_finish_connecting(xprt, sock);
1494         status = 0;
1495 out:
1496         xprt_wake_pending_tasks(xprt, status);
1497         xprt_clear_connecting(xprt);
1498 }
1499
1500 /**
1501  * xs_udp_connect_worker6 - set up a UDP socket
1502  * @work: RPC transport to connect
1503  *
1504  * Invoked by a work queue tasklet.
1505  */
1506 static void xs_udp_connect_worker6(struct work_struct *work)
1507 {
1508         struct sock_xprt *transport =
1509                 container_of(work, struct sock_xprt, connect_worker.work);
1510         struct rpc_xprt *xprt = &transport->xprt;
1511         struct socket *sock = transport->sock;
1512         int err, status = -EIO;
1513
1514         if (xprt->shutdown || !xprt_bound(xprt))
1515                 goto out;
1516
1517         /* Start by resetting any existing state */
1518         xs_close(xprt);
1519
1520         if ((err = sock_create_kern(PF_INET6, SOCK_DGRAM, IPPROTO_UDP, &sock)) < 0) {
1521                 dprintk("RPC:       can't create UDP transport socket (%d).\n", -err);
1522                 goto out;
1523         }
1524         xs_reclassify_socket6(sock);
1525
1526         if (xs_bind6(transport, sock) < 0) {
1527                 sock_release(sock);
1528                 goto out;
1529         }
1530
1531         dprintk("RPC:       worker connecting xprt %p to address: %s\n",
1532                         xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1533
1534         xs_udp_finish_connecting(xprt, sock);
1535         status = 0;
1536 out:
1537         xprt_wake_pending_tasks(xprt, status);
1538         xprt_clear_connecting(xprt);
1539 }
1540
1541 /*
1542  * We need to preserve the port number so the reply cache on the server can
1543  * find our cached RPC replies when we get around to reconnecting.
1544  */
1545 static void xs_tcp_reuse_connection(struct rpc_xprt *xprt)
1546 {
1547         int result;
1548         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1549         struct sockaddr any;
1550
1551         dprintk("RPC:       disconnecting xprt %p to reuse port\n", xprt);
1552
1553         /*
1554          * Disconnect the transport socket by doing a connect operation
1555          * with AF_UNSPEC.  This should return immediately...
1556          */
1557         memset(&any, 0, sizeof(any));
1558         any.sa_family = AF_UNSPEC;
1559         result = kernel_connect(transport->sock, &any, sizeof(any), 0);
1560         if (result)
1561                 dprintk("RPC:       AF_UNSPEC connect return code %d\n",
1562                                 result);
1563 }
1564
1565 static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
1566 {
1567         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1568
1569         if (!transport->inet) {
1570                 struct sock *sk = sock->sk;
1571
1572                 write_lock_bh(&sk->sk_callback_lock);
1573
1574                 sk->sk_user_data = xprt;
1575                 transport->old_data_ready = sk->sk_data_ready;
1576                 transport->old_state_change = sk->sk_state_change;
1577                 transport->old_write_space = sk->sk_write_space;
1578                 sk->sk_data_ready = xs_tcp_data_ready;
1579                 sk->sk_state_change = xs_tcp_state_change;
1580                 sk->sk_write_space = xs_tcp_write_space;
1581                 sk->sk_allocation = GFP_ATOMIC;
1582
1583                 /* socket options */
1584                 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
1585                 sock_reset_flag(sk, SOCK_LINGER);
1586                 tcp_sk(sk)->linger2 = 0;
1587                 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
1588
1589                 xprt_clear_connected(xprt);
1590
1591                 /* Reset to new socket */
1592                 transport->sock = sock;
1593                 transport->inet = sk;
1594
1595                 write_unlock_bh(&sk->sk_callback_lock);
1596         }
1597
1598         /* Tell the socket layer to start connecting... */
1599         xprt->stat.connect_count++;
1600         xprt->stat.connect_start = jiffies;
1601         return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
1602 }
1603
1604 /**
1605  * xs_tcp_connect_worker4 - connect a TCP socket to a remote endpoint
1606  * @work: RPC transport to connect
1607  *
1608  * Invoked by a work queue tasklet.
1609  */
1610 static void xs_tcp_connect_worker4(struct work_struct *work)
1611 {
1612         struct sock_xprt *transport =
1613                 container_of(work, struct sock_xprt, connect_worker.work);
1614         struct rpc_xprt *xprt = &transport->xprt;
1615         struct socket *sock = transport->sock;
1616         int err, status = -EIO;
1617
1618         if (xprt->shutdown || !xprt_bound(xprt))
1619                 goto out;
1620
1621         if (!sock) {
1622                 /* start from scratch */
1623                 if ((err = sock_create_kern(PF_INET, SOCK_STREAM, IPPROTO_TCP, &sock)) < 0) {
1624                         dprintk("RPC:       can't create TCP transport socket (%d).\n", -err);
1625                         goto out;
1626                 }
1627                 xs_reclassify_socket4(sock);
1628
1629                 if (xs_bind4(transport, sock) < 0) {
1630                         sock_release(sock);
1631                         goto out;
1632                 }
1633         } else
1634                 /* "close" the socket, preserving the local port */
1635                 xs_tcp_reuse_connection(xprt);
1636
1637         dprintk("RPC:       worker connecting xprt %p to address: %s\n",
1638                         xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1639
1640         status = xs_tcp_finish_connecting(xprt, sock);
1641         dprintk("RPC:       %p connect status %d connected %d sock state %d\n",
1642                         xprt, -status, xprt_connected(xprt),
1643                         sock->sk->sk_state);
1644         if (status < 0) {
1645                 switch (status) {
1646                         case -EINPROGRESS:
1647                         case -EALREADY:
1648                                 goto out_clear;
1649                         case -ECONNREFUSED:
1650                         case -ECONNRESET:
1651                                 /* retry with existing socket, after a delay */
1652                                 break;
1653                         default:
1654                                 /* get rid of existing socket, and retry */
1655                                 xs_tcp_shutdown(xprt);
1656                 }
1657         }
1658 out:
1659         xprt_wake_pending_tasks(xprt, status);
1660 out_clear:
1661         xprt_clear_connecting(xprt);
1662 }
1663
1664 /**
1665  * xs_tcp_connect_worker6 - connect a TCP socket to a remote endpoint
1666  * @work: RPC transport to connect
1667  *
1668  * Invoked by a work queue tasklet.
1669  */
1670 static void xs_tcp_connect_worker6(struct work_struct *work)
1671 {
1672         struct sock_xprt *transport =
1673                 container_of(work, struct sock_xprt, connect_worker.work);
1674         struct rpc_xprt *xprt = &transport->xprt;
1675         struct socket *sock = transport->sock;
1676         int err, status = -EIO;
1677
1678         if (xprt->shutdown || !xprt_bound(xprt))
1679                 goto out;
1680
1681         if (!sock) {
1682                 /* start from scratch */
1683                 if ((err = sock_create_kern(PF_INET6, SOCK_STREAM, IPPROTO_TCP, &sock)) < 0) {
1684                         dprintk("RPC:       can't create TCP transport socket (%d).\n", -err);
1685                         goto out;
1686                 }
1687                 xs_reclassify_socket6(sock);
1688
1689                 if (xs_bind6(transport, sock) < 0) {
1690                         sock_release(sock);
1691                         goto out;
1692                 }
1693         } else
1694                 /* "close" the socket, preserving the local port */
1695                 xs_tcp_reuse_connection(xprt);
1696
1697         dprintk("RPC:       worker connecting xprt %p to address: %s\n",
1698                         xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1699
1700         status = xs_tcp_finish_connecting(xprt, sock);
1701         dprintk("RPC:       %p connect status %d connected %d sock state %d\n",
1702                         xprt, -status, xprt_connected(xprt), sock->sk->sk_state);
1703         if (status < 0) {
1704                 switch (status) {
1705                         case -EINPROGRESS:
1706                         case -EALREADY:
1707                                 goto out_clear;
1708                         case -ECONNREFUSED:
1709                         case -ECONNRESET:
1710                                 /* retry with existing socket, after a delay */
1711                                 break;
1712                         default:
1713                                 /* get rid of existing socket, and retry */
1714                                 xs_tcp_shutdown(xprt);
1715                 }
1716         }
1717 out:
1718         xprt_wake_pending_tasks(xprt, status);
1719 out_clear:
1720         xprt_clear_connecting(xprt);
1721 }
1722
1723 /**
1724  * xs_connect - connect a socket to a remote endpoint
1725  * @task: address of RPC task that manages state of connect request
1726  *
1727  * TCP: If the remote end dropped the connection, delay reconnecting.
1728  *
1729  * UDP socket connects are synchronous, but we use a work queue anyway
1730  * to guarantee that even unprivileged user processes can set up a
1731  * socket on a privileged port.
1732  *
1733  * If a UDP socket connect fails, the delay behavior here prevents
1734  * retry floods (hard mounts).
1735  */
1736 static void xs_connect(struct rpc_task *task)
1737 {
1738         struct rpc_xprt *xprt = task->tk_xprt;
1739         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1740
1741         if (xprt_test_and_set_connecting(xprt))
1742                 return;
1743
1744         if (transport->sock != NULL) {
1745                 dprintk("RPC:       xs_connect delayed xprt %p for %lu "
1746                                 "seconds\n",
1747                                 xprt, xprt->reestablish_timeout / HZ);
1748                 queue_delayed_work(rpciod_workqueue,
1749                                    &transport->connect_worker,
1750                                    xprt->reestablish_timeout);
1751                 xprt->reestablish_timeout <<= 1;
1752                 if (xprt->reestablish_timeout > XS_TCP_MAX_REEST_TO)
1753                         xprt->reestablish_timeout = XS_TCP_MAX_REEST_TO;
1754         } else {
1755                 dprintk("RPC:       xs_connect scheduled xprt %p\n", xprt);
1756                 queue_delayed_work(rpciod_workqueue,
1757                                    &transport->connect_worker, 0);
1758         }
1759 }
1760
1761 static void xs_tcp_connect(struct rpc_task *task)
1762 {
1763         struct rpc_xprt *xprt = task->tk_xprt;
1764
1765         /* Initiate graceful shutdown of the socket if not already done */
1766         if (test_bit(XPRT_CONNECTED, &xprt->state))
1767                 xs_tcp_shutdown(xprt);
1768         /* Exit if we need to wait for socket shutdown to complete */
1769         if (test_bit(XPRT_CLOSING, &xprt->state))
1770                 return;
1771         xs_connect(task);
1772 }
1773
1774 /**
1775  * xs_udp_print_stats - display UDP socket-specifc stats
1776  * @xprt: rpc_xprt struct containing statistics
1777  * @seq: output file
1778  *
1779  */
1780 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
1781 {
1782         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1783
1784         seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %Lu %Lu\n",
1785                         transport->port,
1786                         xprt->stat.bind_count,
1787                         xprt->stat.sends,
1788                         xprt->stat.recvs,
1789                         xprt->stat.bad_xids,
1790                         xprt->stat.req_u,
1791                         xprt->stat.bklog_u);
1792 }
1793
1794 /**
1795  * xs_tcp_print_stats - display TCP socket-specifc stats
1796  * @xprt: rpc_xprt struct containing statistics
1797  * @seq: output file
1798  *
1799  */
1800 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
1801 {
1802         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1803         long idle_time = 0;
1804
1805         if (xprt_connected(xprt))
1806                 idle_time = (long)(jiffies - xprt->last_used) / HZ;
1807
1808         seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu %Lu %Lu\n",
1809                         transport->port,
1810                         xprt->stat.bind_count,
1811                         xprt->stat.connect_count,
1812                         xprt->stat.connect_time,
1813                         idle_time,
1814                         xprt->stat.sends,
1815                         xprt->stat.recvs,
1816                         xprt->stat.bad_xids,
1817                         xprt->stat.req_u,
1818                         xprt->stat.bklog_u);
1819 }
1820
1821 static struct rpc_xprt_ops xs_udp_ops = {
1822         .set_buffer_size        = xs_udp_set_buffer_size,
1823         .reserve_xprt           = xprt_reserve_xprt_cong,
1824         .release_xprt           = xprt_release_xprt_cong,
1825         .rpcbind                = rpcb_getport_async,
1826         .set_port               = xs_set_port,
1827         .connect                = xs_connect,
1828         .buf_alloc              = rpc_malloc,
1829         .buf_free               = rpc_free,
1830         .send_request           = xs_udp_send_request,
1831         .set_retrans_timeout    = xprt_set_retrans_timeout_rtt,
1832         .timer                  = xs_udp_timer,
1833         .release_request        = xprt_release_rqst_cong,
1834         .close                  = xs_close,
1835         .destroy                = xs_destroy,
1836         .print_stats            = xs_udp_print_stats,
1837 };
1838
1839 static struct rpc_xprt_ops xs_tcp_ops = {
1840         .reserve_xprt           = xprt_reserve_xprt,
1841         .release_xprt           = xs_tcp_release_xprt,
1842         .rpcbind                = rpcb_getport_async,
1843         .set_port               = xs_set_port,
1844         .connect                = xs_tcp_connect,
1845         .buf_alloc              = rpc_malloc,
1846         .buf_free               = rpc_free,
1847         .send_request           = xs_tcp_send_request,
1848         .set_retrans_timeout    = xprt_set_retrans_timeout_def,
1849         .close                  = xs_tcp_shutdown,
1850         .destroy                = xs_destroy,
1851         .print_stats            = xs_tcp_print_stats,
1852 };
1853
1854 static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
1855                                       unsigned int slot_table_size)
1856 {
1857         struct rpc_xprt *xprt;
1858         struct sock_xprt *new;
1859
1860         if (args->addrlen > sizeof(xprt->addr)) {
1861                 dprintk("RPC:       xs_setup_xprt: address too large\n");
1862                 return ERR_PTR(-EBADF);
1863         }
1864
1865         new = kzalloc(sizeof(*new), GFP_KERNEL);
1866         if (new == NULL) {
1867                 dprintk("RPC:       xs_setup_xprt: couldn't allocate "
1868                                 "rpc_xprt\n");
1869                 return ERR_PTR(-ENOMEM);
1870         }
1871         xprt = &new->xprt;
1872
1873         xprt->max_reqs = slot_table_size;
1874         xprt->slot = kcalloc(xprt->max_reqs, sizeof(struct rpc_rqst), GFP_KERNEL);
1875         if (xprt->slot == NULL) {
1876                 kfree(xprt);
1877                 dprintk("RPC:       xs_setup_xprt: couldn't allocate slot "
1878                                 "table\n");
1879                 return ERR_PTR(-ENOMEM);
1880         }
1881
1882         memcpy(&xprt->addr, args->dstaddr, args->addrlen);
1883         xprt->addrlen = args->addrlen;
1884         if (args->srcaddr)
1885                 memcpy(&new->addr, args->srcaddr, args->addrlen);
1886
1887         return xprt;
1888 }
1889
1890 static const struct rpc_timeout xs_udp_default_timeout = {
1891         .to_initval = 5 * HZ,
1892         .to_maxval = 30 * HZ,
1893         .to_increment = 5 * HZ,
1894         .to_retries = 5,
1895 };
1896
1897 /**
1898  * xs_setup_udp - Set up transport to use a UDP socket
1899  * @args: rpc transport creation arguments
1900  *
1901  */
1902 static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
1903 {
1904         struct sockaddr *addr = args->dstaddr;
1905         struct rpc_xprt *xprt;
1906         struct sock_xprt *transport;
1907
1908         xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries);
1909         if (IS_ERR(xprt))
1910                 return xprt;
1911         transport = container_of(xprt, struct sock_xprt, xprt);
1912
1913         xprt->prot = IPPROTO_UDP;
1914         xprt->tsh_size = 0;
1915         /* XXX: header size can vary due to auth type, IPv6, etc. */
1916         xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
1917
1918         xprt->bind_timeout = XS_BIND_TO;
1919         xprt->connect_timeout = XS_UDP_CONN_TO;
1920         xprt->reestablish_timeout = XS_UDP_REEST_TO;
1921         xprt->idle_timeout = XS_IDLE_DISC_TO;
1922
1923         xprt->ops = &xs_udp_ops;
1924
1925         xprt->timeout = &xs_udp_default_timeout;
1926
1927         switch (addr->sa_family) {
1928         case AF_INET:
1929                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
1930                         xprt_set_bound(xprt);
1931
1932                 INIT_DELAYED_WORK(&transport->connect_worker,
1933                                         xs_udp_connect_worker4);
1934                 xs_format_ipv4_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
1935                 break;
1936         case AF_INET6:
1937                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
1938                         xprt_set_bound(xprt);
1939
1940                 INIT_DELAYED_WORK(&transport->connect_worker,
1941                                         xs_udp_connect_worker6);
1942                 xs_format_ipv6_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
1943                 break;
1944         default:
1945                 kfree(xprt);
1946                 return ERR_PTR(-EAFNOSUPPORT);
1947         }
1948
1949         dprintk("RPC:       set up transport to address %s\n",
1950                         xprt->address_strings[RPC_DISPLAY_ALL]);
1951
1952         if (try_module_get(THIS_MODULE))
1953                 return xprt;
1954
1955         kfree(xprt->slot);
1956         kfree(xprt);
1957         return ERR_PTR(-EINVAL);
1958 }
1959
1960 static const struct rpc_timeout xs_tcp_default_timeout = {
1961         .to_initval = 60 * HZ,
1962         .to_maxval = 60 * HZ,
1963         .to_retries = 2,
1964 };
1965
1966 /**
1967  * xs_setup_tcp - Set up transport to use a TCP socket
1968  * @args: rpc transport creation arguments
1969  *
1970  */
1971 static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
1972 {
1973         struct sockaddr *addr = args->dstaddr;
1974         struct rpc_xprt *xprt;
1975         struct sock_xprt *transport;
1976
1977         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries);
1978         if (IS_ERR(xprt))
1979                 return xprt;
1980         transport = container_of(xprt, struct sock_xprt, xprt);
1981
1982         xprt->prot = IPPROTO_TCP;
1983         xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
1984         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
1985
1986         xprt->bind_timeout = XS_BIND_TO;
1987         xprt->connect_timeout = XS_TCP_CONN_TO;
1988         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1989         xprt->idle_timeout = XS_IDLE_DISC_TO;
1990
1991         xprt->ops = &xs_tcp_ops;
1992         xprt->timeout = &xs_tcp_default_timeout;
1993
1994         switch (addr->sa_family) {
1995         case AF_INET:
1996                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
1997                         xprt_set_bound(xprt);
1998
1999                 INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_connect_worker4);
2000                 xs_format_ipv4_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
2001                 break;
2002         case AF_INET6:
2003                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2004                         xprt_set_bound(xprt);
2005
2006                 INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_connect_worker6);
2007                 xs_format_ipv6_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
2008                 break;
2009         default:
2010                 kfree(xprt);
2011                 return ERR_PTR(-EAFNOSUPPORT);
2012         }
2013
2014         dprintk("RPC:       set up transport to address %s\n",
2015                         xprt->address_strings[RPC_DISPLAY_ALL]);
2016
2017         if (try_module_get(THIS_MODULE))
2018                 return xprt;
2019
2020         kfree(xprt->slot);
2021         kfree(xprt);
2022         return ERR_PTR(-EINVAL);
2023 }
2024
2025 static struct xprt_class        xs_udp_transport = {
2026         .list           = LIST_HEAD_INIT(xs_udp_transport.list),
2027         .name           = "udp",
2028         .owner          = THIS_MODULE,
2029         .ident          = IPPROTO_UDP,
2030         .setup          = xs_setup_udp,
2031 };
2032
2033 static struct xprt_class        xs_tcp_transport = {
2034         .list           = LIST_HEAD_INIT(xs_tcp_transport.list),
2035         .name           = "tcp",
2036         .owner          = THIS_MODULE,
2037         .ident          = IPPROTO_TCP,
2038         .setup          = xs_setup_tcp,
2039 };
2040
2041 /**
2042  * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
2043  *
2044  */
2045 int init_socket_xprt(void)
2046 {
2047 #ifdef RPC_DEBUG
2048         if (!sunrpc_table_header)
2049                 sunrpc_table_header = register_sysctl_table(sunrpc_table);
2050 #endif
2051
2052         xprt_register_transport(&xs_udp_transport);
2053         xprt_register_transport(&xs_tcp_transport);
2054
2055         return 0;
2056 }
2057
2058 /**
2059  * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
2060  *
2061  */
2062 void cleanup_socket_xprt(void)
2063 {
2064 #ifdef RPC_DEBUG
2065         if (sunrpc_table_header) {
2066                 unregister_sysctl_table(sunrpc_table_header);
2067                 sunrpc_table_header = NULL;
2068         }
2069 #endif
2070
2071         xprt_unregister_transport(&xs_udp_transport);
2072         xprt_unregister_transport(&xs_tcp_transport);
2073 }