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