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  * Time out for an RPC UDP socket connect.  UDP socket connects are
 140  * synchronous, but we set a timeout anyway in case of resource
 141  * exhaustion on the local host.
 142  */
 143 #define XS_UDP_CONN_TO          (5U * HZ)
 144
 145 /*
 146  * Wait duration for an RPC TCP connection to be established.  Solaris
 147  * NFS over TCP uses 60 seconds, for example, which is in line with how
 148  * long a server takes to reboot.
 149  */
 150 #define XS_TCP_CONN_TO          (60U * HZ)
 151
 152 /*
 153  * Wait duration for a reply from the RPC portmapper.
 154  */
 155 #define XS_BIND_TO              (60U * HZ)
 156
 157 /*
 158  * Delay if a UDP socket connect error occurs.  This is most likely some
 159  * kind of resource problem on the local host.
 160  */
 161 #define XS_UDP_REEST_TO         (2U * HZ)
 162
 163 /*
 164  * The reestablish timeout allows clients to delay for a bit before attempting
 165  * to reconnect to a server that just dropped our connection.
 166  *
 167  * We implement an exponential backoff when trying to reestablish a TCP
 168  * transport connection with the server.  Some servers like to drop a TCP
 169  * connection when they are overworked, so we start with a short timeout and
 170  * increase over time if the server is down or not responding.
 171  */
 172 #define XS_TCP_INIT_REEST_TO    (3U * HZ)
 173 #define XS_TCP_MAX_REEST_TO     (5U * 60 * HZ)
 174
 175 /*
 176  * TCP idle timeout; client drops the transport socket if it is idle
 177  * for this long.  Note that we also timeout UDP sockets to prevent
 178  * holding port numbers when there is no RPC traffic.
 179  */
 180 #define XS_IDLE_DISC_TO         (5U * 60 * HZ)
 181
 182 #ifdef RPC_DEBUG
 183 # undef  RPC_DEBUG_DATA
 184 # define RPCDBG_FACILITY        RPCDBG_TRANS
 185 #endif
 186
 187 #ifdef RPC_DEBUG_DATA
 188 static void xs_pktdump(char *msg, u32 *packet, unsigned int count)
 189 {
 190         u8 *buf = (u8 *) packet;
 191         int j;
 192
 193         dprintk("RPC:       %s\n", msg);
 194         for (j = 0; j < count && j < 128; j += 4) {
 195                 if (!(j & 31)) {
 196                         if (j)
 197                                 dprintk("\n");
 198                         dprintk("0x%04x ", j);
 199                 }
 200                 dprintk("%02x%02x%02x%02x ",
 201                         buf[j], buf[j+1], buf[j+2], buf[j+3]);
 202         }
 203         dprintk("\n");
 204 }
 205 #else
 206 static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count)
 207 {
 208         /* NOP */
 209 }
 210 #endif
 211
 212 struct sock_xprt {
 213         struct rpc_xprt         xprt;
 214
 215         /*
 216          * Network layer
 217          */
 218         struct socket *         sock;
 219         struct sock *           inet;
 220
 221         /*
 222          * State of TCP reply receive
 223          */
 224         __be32                  tcp_fraghdr,
 225                                 tcp_xid;
 226
 227         u32                     tcp_offset,
 228                                 tcp_reclen;
 229
 230         unsigned long           tcp_copied,
 231                                 tcp_flags;
 232
 233         /*
 234          * Connection of transports
 235          */
 236         struct delayed_work     connect_worker;
 237         struct sockaddr_storage addr;
 238         unsigned short          port;
 239
 240         /*
 241          * UDP socket buffer size parameters
 242          */
 243         size_t                  rcvsize,
 244                                 sndsize;
 245
 246         /*
 247          * Saved socket callback addresses
 248          */
 249         void                    (*old_data_ready)(struct sock *, int);
 250         void                    (*old_state_change)(struct sock *);
 251         void                    (*old_write_space)(struct sock *);
 252 };
 253
 254 /*
 255  * TCP receive state flags
 256  */
 257 #define TCP_RCV_LAST_FRAG       (1UL << 0)
 258 #define TCP_RCV_COPY_FRAGHDR    (1UL << 1)
 259 #define TCP_RCV_COPY_XID        (1UL << 2)
 260 #define TCP_RCV_COPY_DATA       (1UL << 3)
 261
 262 static inline struct sockaddr *xs_addr(struct rpc_xprt *xprt)
 263 {
 264         return (struct sockaddr *) &xprt->addr;
 265 }
 266
 267 static inline struct sockaddr_in *xs_addr_in(struct rpc_xprt *xprt)
 268 {
 269         return (struct sockaddr_in *) &xprt->addr;
 270 }
 271
 272 static inline struct sockaddr_in6 *xs_addr_in6(struct rpc_xprt *xprt)
 273 {
 274         return (struct sockaddr_in6 *) &xprt->addr;
 275 }
 276
 277 static void xs_format_ipv4_peer_addresses(struct rpc_xprt *xprt,
 278                                           const char *protocol,
 279                                           const char *netid)
 280 {
 281         struct sockaddr_in *addr = xs_addr_in(xprt);
 282         char *buf;
 283
 284         buf = kzalloc(20, GFP_KERNEL);
 285         if (buf) {
 286                 snprintf(buf, 20, NIPQUAD_FMT,
 287                                 NIPQUAD(addr->sin_addr.s_addr));
 288         }
 289         xprt->address_strings[RPC_DISPLAY_ADDR] = buf;
 290
 291         buf = kzalloc(8, GFP_KERNEL);
 292         if (buf) {
 293                 snprintf(buf, 8, "%u",
 294                                 ntohs(addr->sin_port));
 295         }
 296         xprt->address_strings[RPC_DISPLAY_PORT] = buf;
 297
 298         xprt->address_strings[RPC_DISPLAY_PROTO] = protocol;
 299
 300         buf = kzalloc(48, GFP_KERNEL);
 301         if (buf) {
 302                 snprintf(buf, 48, "addr="NIPQUAD_FMT" port=%u proto=%s",
 303                         NIPQUAD(addr->sin_addr.s_addr),
 304                         ntohs(addr->sin_port),
 305                         protocol);
 306         }
 307         xprt->address_strings[RPC_DISPLAY_ALL] = buf;
 308
 309         buf = kzalloc(10, GFP_KERNEL);
 310         if (buf) {
 311                 snprintf(buf, 10, "%02x%02x%02x%02x",
 312                                 NIPQUAD(addr->sin_addr.s_addr));
 313         }
 314         xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = buf;
 315
 316         buf = kzalloc(8, GFP_KERNEL);
 317         if (buf) {
 318                 snprintf(buf, 8, "%4hx",
 319                                 ntohs(addr->sin_port));
 320         }
 321         xprt->address_strings[RPC_DISPLAY_HEX_PORT] = buf;
 322
 323         buf = kzalloc(30, GFP_KERNEL);
 324         if (buf) {
 325                 snprintf(buf, 30, NIPQUAD_FMT".%u.%u",
 326                                 NIPQUAD(addr->sin_addr.s_addr),
 327                                 ntohs(addr->sin_port) >> 8,
 328                                 ntohs(addr->sin_port) & 0xff);
 329         }
 330         xprt->address_strings[RPC_DISPLAY_UNIVERSAL_ADDR] = buf;
 331
 332         xprt->address_strings[RPC_DISPLAY_NETID] = netid;
 333 }
 334
 335 static void xs_format_ipv6_peer_addresses(struct rpc_xprt *xprt,
 336                                           const char *protocol,
 337                                           const char *netid)
 338 {
 339         struct sockaddr_in6 *addr = xs_addr_in6(xprt);
 340         char *buf;
 341
 342         buf = kzalloc(40, GFP_KERNEL);
 343         if (buf) {
 344                 snprintf(buf, 40, NIP6_FMT,
 345                                 NIP6(addr->sin6_addr));
 346         }
 347         xprt->address_strings[RPC_DISPLAY_ADDR] = buf;
 348
 349         buf = kzalloc(8, GFP_KERNEL);
 350         if (buf) {
 351                 snprintf(buf, 8, "%u",
 352                                 ntohs(addr->sin6_port));
 353         }
 354         xprt->address_strings[RPC_DISPLAY_PORT] = buf;
 355
 356         xprt->address_strings[RPC_DISPLAY_PROTO] = protocol;
 357
 358         buf = kzalloc(64, GFP_KERNEL);
 359         if (buf) {
 360                 snprintf(buf, 64, "addr="NIP6_FMT" port=%u proto=%s",
 361                                 NIP6(addr->sin6_addr),
 362                                 ntohs(addr->sin6_port),
 363                                 protocol);
 364         }
 365         xprt->address_strings[RPC_DISPLAY_ALL] = buf;
 366
 367         buf = kzalloc(36, GFP_KERNEL);
 368         if (buf) {
 369                 snprintf(buf, 36, NIP6_SEQFMT,
 370                                 NIP6(addr->sin6_addr));
 371         }
 372         xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = buf;
 373
 374         buf = kzalloc(8, GFP_KERNEL);
 375         if (buf) {
 376                 snprintf(buf, 8, "%4hx",
 377                                 ntohs(addr->sin6_port));
 378         }
 379         xprt->address_strings[RPC_DISPLAY_HEX_PORT] = buf;
 380
 381         buf = kzalloc(50, GFP_KERNEL);
 382         if (buf) {
 383                 snprintf(buf, 50, NIP6_FMT".%u.%u",
 384                                 NIP6(addr->sin6_addr),
 385                                 ntohs(addr->sin6_port) >> 8,
 386                                 ntohs(addr->sin6_port) & 0xff);
 387         }
 388         xprt->address_strings[RPC_DISPLAY_UNIVERSAL_ADDR] = buf;
 389
 390         xprt->address_strings[RPC_DISPLAY_NETID] = netid;
 391 }
 392
 393 static void xs_free_peer_addresses(struct rpc_xprt *xprt)
 394 {
 395         unsigned int i;
 396
 397         for (i = 0; i < RPC_DISPLAY_MAX; i++)
 398                 switch (i) {
 399                 case RPC_DISPLAY_PROTO:
 400                 case RPC_DISPLAY_NETID:
 401                         continue;
 402                 default:
 403                         kfree(xprt->address_strings[i]);
 404                 }
 405 }
 406
 407 #define XS_SENDMSG_FLAGS        (MSG_DONTWAIT | MSG_NOSIGNAL)
 408
 409 static int xs_send_kvec(struct socket *sock, struct sockaddr *addr, int addrlen, struct kvec *vec, unsigned int base, int more)
 410 {
 411         struct msghdr msg = {
 412                 .msg_name       = addr,
 413                 .msg_namelen    = addrlen,
 414                 .msg_flags      = XS_SENDMSG_FLAGS | (more ? MSG_MORE : 0),
 415         };
 416         struct kvec iov = {
 417                 .iov_base       = vec->iov_base + base,
 418                 .iov_len        = vec->iov_len - base,
 419         };
 420
 421         if (iov.iov_len != 0)
 422                 return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
 423         return kernel_sendmsg(sock, &msg, NULL, 0, 0);
 424 }
 425
 426 static int xs_send_pagedata(struct socket *sock, struct xdr_buf *xdr, unsigned int base, int more)
 427 {
 428         struct page **ppage;
 429         unsigned int remainder;
 430         int err, sent = 0;
 431
 432         remainder = xdr->page_len - base;
 433         base += xdr->page_base;
 434         ppage = xdr->pages + (base >> PAGE_SHIFT);
 435         base &= ~PAGE_MASK;
 436         for(;;) {
 437                 unsigned int len = min_t(unsigned int, PAGE_SIZE - base, remainder);
 438                 int flags = XS_SENDMSG_FLAGS;
 439
 440                 remainder -= len;
 441                 if (remainder != 0 || more)
 442                         flags |= MSG_MORE;
 443                 err = sock->ops->sendpage(sock, *ppage, base, len, flags);
 444                 if (remainder == 0 || err != len)
 445                         break;
 446                 sent += err;
 447                 ppage++;
 448                 base = 0;
 449         }
 450         if (sent == 0)
 451                 return err;
 452         if (err > 0)
 453                 sent += err;
 454         return sent;
 455 }
 456
 457 /**
 458  * xs_sendpages - write pages directly to a socket
 459  * @sock: socket to send on
 460  * @addr: UDP only -- address of destination
 461  * @addrlen: UDP only -- length of destination address
 462  * @xdr: buffer containing this request
 463  * @base: starting position in the buffer
 464  *
 465  */
 466 static int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base)
 467 {
 468         unsigned int remainder = xdr->len - base;
 469         int err, sent = 0;
 470
 471         if (unlikely(!sock))
 472                 return -ENOTCONN;
 473
 474         clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags);
 475         if (base != 0) {
 476                 addr = NULL;
 477                 addrlen = 0;
 478         }
 479
 480         if (base < xdr->head[0].iov_len || addr != NULL) {
 481                 unsigned int len = xdr->head[0].iov_len - base;
 482                 remainder -= len;
 483                 err = xs_send_kvec(sock, addr, addrlen, &xdr->head[0], base, remainder != 0);
 484                 if (remainder == 0 || err != len)
 485                         goto out;
 486                 sent += err;
 487                 base = 0;
 488         } else
 489                 base -= xdr->head[0].iov_len;
 490
 491         if (base < xdr->page_len) {
 492                 unsigned int len = xdr->page_len - base;
 493                 remainder -= len;
 494                 err = xs_send_pagedata(sock, xdr, base, remainder != 0);
 495                 if (remainder == 0 || err != len)
 496                         goto out;
 497                 sent += err;
 498                 base = 0;
 499         } else
 500                 base -= xdr->page_len;
 501
 502         if (base >= xdr->tail[0].iov_len)
 503                 return sent;
 504         err = xs_send_kvec(sock, NULL, 0, &xdr->tail[0], base, 0);
 505 out:
 506         if (sent == 0)
 507                 return err;
 508         if (err > 0)
 509                 sent += err;
 510         return sent;
 511 }
 512
 513 static void xs_nospace_callback(struct rpc_task *task)
 514 {
 515         struct sock_xprt *transport = container_of(task->tk_rqstp->rq_xprt, struct sock_xprt, xprt);
 516
 517         transport->inet->sk_write_pending--;
 518         clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 519 }
 520
 521 /**
 522  * xs_nospace - place task on wait queue if transmit was incomplete
 523  * @task: task to put to sleep
 524  *
 525  */
 526 static void xs_nospace(struct rpc_task *task)
 527 {
 528         struct rpc_rqst *req = task->tk_rqstp;
 529         struct rpc_xprt *xprt = req->rq_xprt;
 530         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 531
 532         dprintk("RPC: %5u xmit incomplete (%u left of %u)\n",
 533                         task->tk_pid, req->rq_slen - req->rq_bytes_sent,
 534                         req->rq_slen);
 535
 536         /* Protect against races with write_space */
 537         spin_lock_bh(&xprt->transport_lock);
 538
 539         /* Don't race with disconnect */
 540         if (xprt_connected(xprt)) {
 541                 if (test_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags)) {
 542                         /*
 543                          * Notify TCP that we're limited by the application
 544                          * window size
 545                          */
 546                         set_bit(SOCK_NOSPACE, &transport->sock->flags);
 547                         transport->inet->sk_write_pending++;
 548                         /* ...and wait for more buffer space */
 549                         xprt_wait_for_buffer_space(task, xs_nospace_callback);
 550                 }
 551         } else {
 552                 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 553                 task->tk_status = -ENOTCONN;
 554         }
 555
 556         spin_unlock_bh(&xprt->transport_lock);
 557 }
 558
 559 /**
 560  * xs_udp_send_request - write an RPC request to a UDP socket
 561  * @task: address of RPC task that manages the state of an RPC request
 562  *
 563  * Return values:
 564  *        0:    The request has been sent
 565  *   EAGAIN:    The socket was blocked, please call again later to
 566  *              complete the request
 567  * ENOTCONN:    Caller needs to invoke connect logic then call again
 568  *    other:    Some other error occured, the request was not sent
 569  */
 570 static int xs_udp_send_request(struct rpc_task *task)
 571 {
 572         struct rpc_rqst *req = task->tk_rqstp;
 573         struct rpc_xprt *xprt = req->rq_xprt;
 574         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 575         struct xdr_buf *xdr = &req->rq_snd_buf;
 576         int status;
 577
 578         xs_pktdump("packet data:",
 579                                 req->rq_svec->iov_base,
 580                                 req->rq_svec->iov_len);
 581
 582         req->rq_xtime = jiffies;
 583         status = xs_sendpages(transport->sock,
 584                               xs_addr(xprt),
 585                               xprt->addrlen, xdr,
 586                               req->rq_bytes_sent);
 587
 588         dprintk("RPC:       xs_udp_send_request(%u) = %d\n",
 589                         xdr->len - req->rq_bytes_sent, status);
 590
 591         if (status >= 0) {
 592                 task->tk_bytes_sent += status;
 593                 if (status >= req->rq_slen)
 594                         return 0;
 595                 /* Still some bytes left; set up for a retry later. */
 596                 status = -EAGAIN;
 597         }
 598
 599         switch (status) {
 600         case -EAGAIN:
 601                 xs_nospace(task);
 602                 break;
 603         case -ENETUNREACH:
 604         case -EPIPE:
 605         case -ECONNREFUSED:
 606                 /* When the server has died, an ICMP port unreachable message
 607                  * prompts ECONNREFUSED. */
 608                 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 609                 break;
 610         default:
 611                 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 612                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
 613                         -status);
 614         }
 615
 616         return status;
 617 }
 618
 619 /**
 620  * xs_tcp_shutdown - gracefully shut down a TCP socket
 621  * @xprt: transport
 622  *
 623  * Initiates a graceful shutdown of the TCP socket by calling the
 624  * equivalent of shutdown(SHUT_WR);
 625  */
 626 static void xs_tcp_shutdown(struct rpc_xprt *xprt)
 627 {
 628         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 629         struct socket *sock = transport->sock;
 630
 631         if (sock != NULL)
 632                 kernel_sock_shutdown(sock, SHUT_WR);
 633 }
 634
 635 static inline void xs_encode_tcp_record_marker(struct xdr_buf *buf)
 636 {
 637         u32 reclen = buf->len - sizeof(rpc_fraghdr);
 638         rpc_fraghdr *base = buf->head[0].iov_base;
 639         *base = htonl(RPC_LAST_STREAM_FRAGMENT | reclen);
 640 }
 641
 642 /**
 643  * xs_tcp_send_request - write an RPC request to a TCP socket
 644  * @task: address of RPC task that manages the state of an RPC request
 645  *
 646  * Return values:
 647  *        0:    The request has been sent
 648  *   EAGAIN:    The socket was blocked, please call again later to
 649  *              complete the request
 650  * ENOTCONN:    Caller needs to invoke connect logic then call again
 651  *    other:    Some other error occured, the request was not sent
 652  *
 653  * XXX: In the case of soft timeouts, should we eventually give up
 654  *      if sendmsg is not able to make progress?
 655  */
 656 static int xs_tcp_send_request(struct rpc_task *task)
 657 {
 658         struct rpc_rqst *req = task->tk_rqstp;
 659         struct rpc_xprt *xprt = req->rq_xprt;
 660         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 661         struct xdr_buf *xdr = &req->rq_snd_buf;
 662         int status;
 663
 664         xs_encode_tcp_record_marker(&req->rq_snd_buf);
 665
 666         xs_pktdump("packet data:",
 667                                 req->rq_svec->iov_base,
 668                                 req->rq_svec->iov_len);
 669
 670         /* Continue transmitting the packet/record. We must be careful
 671          * to cope with writespace callbacks arriving _after_ we have
 672          * called sendmsg(). */
 673         while (1) {
 674                 req->rq_xtime = jiffies;
 675                 status = xs_sendpages(transport->sock,
 676                                         NULL, 0, xdr, req->rq_bytes_sent);
 677
 678                 dprintk("RPC:       xs_tcp_send_request(%u) = %d\n",
 679                                 xdr->len - req->rq_bytes_sent, status);
 680
 681                 if (unlikely(status < 0))
 682                         break;
 683
 684                 /* If we've sent the entire packet, immediately
 685                  * reset the count of bytes sent. */
 686                 req->rq_bytes_sent += status;
 687                 task->tk_bytes_sent += status;
 688                 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
 689                         req->rq_bytes_sent = 0;
 690                         return 0;
 691                 }
 692
 693                 if (status != 0)
 694                         continue;
 695                 status = -EAGAIN;
 696                 break;
 697         }
 698
 699         switch (status) {
 700         case -EAGAIN:
 701                 xs_nospace(task);
 702                 break;
 703         case -ECONNREFUSED:
 704         case -ECONNRESET:
 705         case -ENOTCONN:
 706         case -EPIPE:
 707                 status = -ENOTCONN;
 708                 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 709                 break;
 710         default:
 711                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
 712                         -status);
 713                 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 714                 xs_tcp_shutdown(xprt);
 715         }
 716
 717         return status;
 718 }
 719
 720 /**
 721  * xs_tcp_release_xprt - clean up after a tcp transmission
 722  * @xprt: transport
 723  * @task: rpc task
 724  *
 725  * This cleans up if an error causes us to abort the transmission of a request.
 726  * In this case, the socket may need to be reset in order to avoid confusing
 727  * the server.
 728  */
 729 static void xs_tcp_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
 730 {
 731         struct rpc_rqst *req;
 732
 733         if (task != xprt->snd_task)
 734                 return;
 735         if (task == NULL)
 736                 goto out_release;
 737         req = task->tk_rqstp;
 738         if (req->rq_bytes_sent == 0)
 739                 goto out_release;
 740         if (req->rq_bytes_sent == req->rq_snd_buf.len)
 741                 goto out_release;
 742         set_bit(XPRT_CLOSE_WAIT, &task->tk_xprt->state);
 743 out_release:
 744         xprt_release_xprt(xprt, task);
 745 }
 746
 747 /**
 748  * xs_close - close a socket
 749  * @xprt: transport
 750  *
 751  * This is used when all requests are complete; ie, no DRC state remains
 752  * on the server we want to save.
 753  */
 754 static void xs_close(struct rpc_xprt *xprt)
 755 {
 756         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 757         struct socket *sock = transport->sock;
 758         struct sock *sk = transport->inet;
 759
 760         if (!sk)
 761                 goto clear_close_wait;
 762
 763         dprintk("RPC:       xs_close xprt %p\n", xprt);
 764
 765         write_lock_bh(&sk->sk_callback_lock);
 766         transport->inet = NULL;
 767         transport->sock = NULL;
 768
 769         sk->sk_user_data = NULL;
 770         sk->sk_data_ready = transport->old_data_ready;
 771         sk->sk_state_change = transport->old_state_change;
 772         sk->sk_write_space = transport->old_write_space;
 773         write_unlock_bh(&sk->sk_callback_lock);
 774
 775         sk->sk_no_check = 0;
 776
 777         sock_release(sock);
 778 clear_close_wait:
 779         smp_mb__before_clear_bit();
 780         clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
 781         clear_bit(XPRT_CLOSING, &xprt->state);
 782         smp_mb__after_clear_bit();
 783         xprt_disconnect_done(xprt);
 784 }
 785
 786 /**
 787  * xs_destroy - prepare to shutdown a transport
 788  * @xprt: doomed transport
 789  *
 790  */
 791 static void xs_destroy(struct rpc_xprt *xprt)
 792 {
 793         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 794
 795         dprintk("RPC:       xs_destroy xprt %p\n", xprt);
 796
 797         cancel_rearming_delayed_work(&transport->connect_worker);
 798
 799         xs_close(xprt);
 800         xs_free_peer_addresses(xprt);
 801         kfree(xprt->slot);
 802         kfree(xprt);
 803         module_put(THIS_MODULE);
 804 }
 805
 806 static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
 807 {
 808         return (struct rpc_xprt *) sk->sk_user_data;
 809 }
 810
 811 /**
 812  * xs_udp_data_ready - "data ready" callback for UDP sockets
 813  * @sk: socket with data to read
 814  * @len: how much data to read
 815  *
 816  */
 817 static void xs_udp_data_ready(struct sock *sk, int len)
 818 {
 819         struct rpc_task *task;
 820         struct rpc_xprt *xprt;
 821         struct rpc_rqst *rovr;
 822         struct sk_buff *skb;
 823         int err, repsize, copied;
 824         u32 _xid;
 825         __be32 *xp;
 826
 827         read_lock(&sk->sk_callback_lock);
 828         dprintk("RPC:       xs_udp_data_ready...\n");
 829         if (!(xprt = xprt_from_sock(sk)))
 830                 goto out;
 831
 832         if ((skb = skb_recv_datagram(sk, 0, 1, &err)) == NULL)
 833                 goto out;
 834
 835         if (xprt->shutdown)
 836                 goto dropit;
 837
 838         repsize = skb->len - sizeof(struct udphdr);
 839         if (repsize < 4) {
 840                 dprintk("RPC:       impossible RPC reply size %d!\n", repsize);
 841                 goto dropit;
 842         }
 843
 844         /* Copy the XID from the skb... */
 845         xp = skb_header_pointer(skb, sizeof(struct udphdr),
 846                                 sizeof(_xid), &_xid);
 847         if (xp == NULL)
 848                 goto dropit;
 849
 850         /* Look up and lock the request corresponding to the given XID */
 851         spin_lock(&xprt->transport_lock);
 852         rovr = xprt_lookup_rqst(xprt, *xp);
 853         if (!rovr)
 854                 goto out_unlock;
 855         task = rovr->rq_task;
 856
 857         if ((copied = rovr->rq_private_buf.buflen) > repsize)
 858                 copied = repsize;
 859
 860         /* Suck it into the iovec, verify checksum if not done by hw. */
 861         if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
 862                 UDPX_INC_STATS_BH(sk, UDP_MIB_INERRORS);
 863                 goto out_unlock;
 864         }
 865
 866         UDPX_INC_STATS_BH(sk, UDP_MIB_INDATAGRAMS);
 867
 868         /* Something worked... */
 869         dst_confirm(skb->dst);
 870
 871         xprt_adjust_cwnd(task, copied);
 872         xprt_update_rtt(task);
 873         xprt_complete_rqst(task, copied);
 874
 875  out_unlock:
 876         spin_unlock(&xprt->transport_lock);
 877  dropit:
 878         skb_free_datagram(sk, skb);
 879  out:
 880         read_unlock(&sk->sk_callback_lock);
 881 }
 882
 883 static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
 884 {
 885         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 886         size_t len, used;
 887         char *p;
 888
 889         p = ((char *) &transport->tcp_fraghdr) + transport->tcp_offset;
 890         len = sizeof(transport->tcp_fraghdr) - transport->tcp_offset;
 891         used = xdr_skb_read_bits(desc, p, len);
 892         transport->tcp_offset += used;
 893         if (used != len)
 894                 return;
 895
 896         transport->tcp_reclen = ntohl(transport->tcp_fraghdr);
 897         if (transport->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
 898                 transport->tcp_flags |= TCP_RCV_LAST_FRAG;
 899         else
 900                 transport->tcp_flags &= ~TCP_RCV_LAST_FRAG;
 901         transport->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
 902
 903         transport->tcp_flags &= ~TCP_RCV_COPY_FRAGHDR;
 904         transport->tcp_offset = 0;
 905
 906         /* Sanity check of the record length */
 907         if (unlikely(transport->tcp_reclen < 4)) {
 908                 dprintk("RPC:       invalid TCP record fragment length\n");
 909                 xprt_force_disconnect(xprt);
 910                 return;
 911         }
 912         dprintk("RPC:       reading TCP record fragment of length %d\n",
 913                         transport->tcp_reclen);
 914 }
 915
 916 static void xs_tcp_check_fraghdr(struct sock_xprt *transport)
 917 {
 918         if (transport->tcp_offset == transport->tcp_reclen) {
 919                 transport->tcp_flags |= TCP_RCV_COPY_FRAGHDR;
 920                 transport->tcp_offset = 0;
 921                 if (transport->tcp_flags & TCP_RCV_LAST_FRAG) {
 922                         transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
 923                         transport->tcp_flags |= TCP_RCV_COPY_XID;
 924                         transport->tcp_copied = 0;
 925                 }
 926         }
 927 }
 928
 929 static inline void xs_tcp_read_xid(struct sock_xprt *transport, struct xdr_skb_reader *desc)
 930 {
 931         size_t len, used;
 932         char *p;
 933
 934         len = sizeof(transport->tcp_xid) - transport->tcp_offset;
 935         dprintk("RPC:       reading XID (%Zu bytes)\n", len);
 936         p = ((char *) &transport->tcp_xid) + transport->tcp_offset;
 937         used = xdr_skb_read_bits(desc, p, len);
 938         transport->tcp_offset += used;
 939         if (used != len)
 940                 return;
 941         transport->tcp_flags &= ~TCP_RCV_COPY_XID;
 942         transport->tcp_flags |= TCP_RCV_COPY_DATA;
 943         transport->tcp_copied = 4;
 944         dprintk("RPC:       reading reply for XID %08x\n",
 945                         ntohl(transport->tcp_xid));
 946         xs_tcp_check_fraghdr(transport);
 947 }
 948
 949 static inline void xs_tcp_read_request(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
 950 {
 951         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 952         struct rpc_rqst *req;
 953         struct xdr_buf *rcvbuf;
 954         size_t len;
 955         ssize_t r;
 956
 957         /* Find and lock the request corresponding to this xid */
 958         spin_lock(&xprt->transport_lock);
 959         req = xprt_lookup_rqst(xprt, transport->tcp_xid);
 960         if (!req) {
 961                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
 962                 dprintk("RPC:       XID %08x request not found!\n",
 963                                 ntohl(transport->tcp_xid));
 964                 spin_unlock(&xprt->transport_lock);
 965                 return;
 966         }
 967
 968         rcvbuf = &req->rq_private_buf;
 969         len = desc->count;
 970         if (len > transport->tcp_reclen - transport->tcp_offset) {
 971                 struct xdr_skb_reader my_desc;
 972
 973                 len = transport->tcp_reclen - transport->tcp_offset;
 974                 memcpy(&my_desc, desc, sizeof(my_desc));
 975                 my_desc.count = len;
 976                 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
 977                                           &my_desc, xdr_skb_read_bits);
 978                 desc->count -= r;
 979                 desc->offset += r;
 980         } else
 981                 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
 982                                           desc, xdr_skb_read_bits);
 983
 984         if (r > 0) {
 985                 transport->tcp_copied += r;
 986                 transport->tcp_offset += r;
 987         }
 988         if (r != len) {
 989                 /* Error when copying to the receive buffer,
 990                  * usually because we weren't able to allocate
 991                  * additional buffer pages. All we can do now
 992                  * is turn off TCP_RCV_COPY_DATA, so the request
 993                  * will not receive any additional updates,
 994                  * and time out.
 995                  * Any remaining data from this record will
 996                  * be discarded.
 997                  */
 998                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
 999                 dprintk("RPC:       XID %08x truncated request\n",
1000                                 ntohl(transport->tcp_xid));
1001                 dprintk("RPC:       xprt = %p, tcp_copied = %lu, "
1002                                 "tcp_offset = %u, tcp_reclen = %u\n",
1003                                 xprt, transport->tcp_copied,
1004                                 transport->tcp_offset, transport->tcp_reclen);
1005                 goto out;
1006         }
1007
1008         dprintk("RPC:       XID %08x read %Zd bytes\n",
1009                         ntohl(transport->tcp_xid), r);
1010         dprintk("RPC:       xprt = %p, tcp_copied = %lu, tcp_offset = %u, "
1011                         "tcp_reclen = %u\n", xprt, transport->tcp_copied,
1012                         transport->tcp_offset, transport->tcp_reclen);
1013
1014         if (transport->tcp_copied == req->rq_private_buf.buflen)
1015                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1016         else if (transport->tcp_offset == transport->tcp_reclen) {
1017                 if (transport->tcp_flags & TCP_RCV_LAST_FRAG)
1018                         transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1019         }
1020
1021 out:
1022         if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1023                 xprt_complete_rqst(req->rq_task, transport->tcp_copied);
1024         spin_unlock(&xprt->transport_lock);
1025         xs_tcp_check_fraghdr(transport);
1026 }
1027
1028 static inline void xs_tcp_read_discard(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1029 {
1030         size_t len;
1031
1032         len = transport->tcp_reclen - transport->tcp_offset;
1033         if (len > desc->count)
1034                 len = desc->count;
1035         desc->count -= len;
1036         desc->offset += len;
1037         transport->tcp_offset += len;
1038         dprintk("RPC:       discarded %Zu bytes\n", len);
1039         xs_tcp_check_fraghdr(transport);
1040 }
1041
1042 static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
1043 {
1044         struct rpc_xprt *xprt = rd_desc->arg.data;
1045         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1046         struct xdr_skb_reader desc = {
1047                 .skb    = skb,
1048                 .offset = offset,
1049                 .count  = len,
1050         };
1051
1052         dprintk("RPC:       xs_tcp_data_recv started\n");
1053         do {
1054                 /* Read in a new fragment marker if necessary */
1055                 /* Can we ever really expect to get completely empty fragments? */
1056                 if (transport->tcp_flags & TCP_RCV_COPY_FRAGHDR) {
1057                         xs_tcp_read_fraghdr(xprt, &desc);
1058                         continue;
1059                 }
1060                 /* Read in the xid if necessary */
1061                 if (transport->tcp_flags & TCP_RCV_COPY_XID) {
1062                         xs_tcp_read_xid(transport, &desc);
1063                         continue;
1064                 }
1065                 /* Read in the request data */
1066                 if (transport->tcp_flags & TCP_RCV_COPY_DATA) {
1067                         xs_tcp_read_request(xprt, &desc);
1068                         continue;
1069                 }
1070                 /* Skip over any trailing bytes on short reads */
1071                 xs_tcp_read_discard(transport, &desc);
1072         } while (desc.count);
1073         dprintk("RPC:       xs_tcp_data_recv done\n");
1074         return len - desc.count;
1075 }
1076
1077 /**
1078  * xs_tcp_data_ready - "data ready" callback for TCP sockets
1079  * @sk: socket with data to read
1080  * @bytes: how much data to read
1081  *
1082  */
1083 static void xs_tcp_data_ready(struct sock *sk, int bytes)
1084 {
1085         struct rpc_xprt *xprt;
1086         read_descriptor_t rd_desc;
1087         int read;
1088
1089         dprintk("RPC:       xs_tcp_data_ready...\n");
1090
1091         read_lock(&sk->sk_callback_lock);
1092         if (!(xprt = xprt_from_sock(sk)))
1093                 goto out;
1094         if (xprt->shutdown)
1095                 goto out;
1096
1097         /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
1098         rd_desc.arg.data = xprt;
1099         do {
1100                 rd_desc.count = 65536;
1101                 read = tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
1102         } while (read > 0);
1103 out:
1104         read_unlock(&sk->sk_callback_lock);
1105 }
1106
1107 /**
1108  * xs_tcp_state_change - callback to handle TCP socket state changes
1109  * @sk: socket whose state has changed
1110  *
1111  */
1112 static void xs_tcp_state_change(struct sock *sk)
1113 {
1114         struct rpc_xprt *xprt;
1115
1116         read_lock(&sk->sk_callback_lock);
1117         if (!(xprt = xprt_from_sock(sk)))
1118                 goto out;
1119         dprintk("RPC:       xs_tcp_state_change client %p...\n", xprt);
1120         dprintk("RPC:       state %x conn %d dead %d zapped %d\n",
1121                         sk->sk_state, xprt_connected(xprt),
1122                         sock_flag(sk, SOCK_DEAD),
1123                         sock_flag(sk, SOCK_ZAPPED));
1124
1125         switch (sk->sk_state) {
1126         case TCP_ESTABLISHED:
1127                 spin_lock_bh(&xprt->transport_lock);
1128                 if (!xprt_test_and_set_connected(xprt)) {
1129                         struct sock_xprt *transport = container_of(xprt,
1130                                         struct sock_xprt, xprt);
1131
1132                         /* Reset TCP record info */
1133                         transport->tcp_offset = 0;
1134                         transport->tcp_reclen = 0;
1135                         transport->tcp_copied = 0;
1136                         transport->tcp_flags =
1137                                 TCP_RCV_COPY_FRAGHDR | TCP_RCV_COPY_XID;
1138
1139                         xprt_wake_pending_tasks(xprt, 0);
1140                 }
1141                 spin_unlock_bh(&xprt->transport_lock);
1142                 break;
1143         case TCP_FIN_WAIT1:
1144                 /* The client initiated a shutdown of the socket */
1145                 xprt->connect_cookie++;
1146                 xprt->reestablish_timeout = 0;
1147                 set_bit(XPRT_CLOSING, &xprt->state);
1148                 smp_mb__before_clear_bit();
1149                 clear_bit(XPRT_CONNECTED, &xprt->state);
1150                 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1151                 smp_mb__after_clear_bit();
1152                 break;
1153         case TCP_CLOSE_WAIT:
1154                 /* The server initiated a shutdown of the socket */
1155                 set_bit(XPRT_CLOSING, &xprt->state);
1156                 xprt_force_disconnect(xprt);
1157         case TCP_SYN_SENT:
1158                 xprt->connect_cookie++;
1159         case TCP_CLOSING:
1160                 /*
1161                  * If the server closed down the connection, make sure that
1162                  * we back off before reconnecting
1163                  */
1164                 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1165                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1166                 break;
1167         case TCP_LAST_ACK:
1168                 smp_mb__before_clear_bit();
1169                 clear_bit(XPRT_CONNECTED, &xprt->state);
1170                 smp_mb__after_clear_bit();
1171                 break;
1172         case TCP_CLOSE:
1173                 smp_mb__before_clear_bit();
1174                 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1175                 clear_bit(XPRT_CLOSING, &xprt->state);
1176                 smp_mb__after_clear_bit();
1177                 /* Mark transport as closed and wake up all pending tasks */
1178                 xprt_disconnect_done(xprt);
1179         }
1180  out:
1181         read_unlock(&sk->sk_callback_lock);
1182 }
1183
1184 /**
1185  * xs_udp_write_space - callback invoked when socket buffer space
1186  *                             becomes available
1187  * @sk: socket whose state has changed
1188  *
1189  * Called when more output buffer space is available for this socket.
1190  * We try not to wake our writers until they can make "significant"
1191  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1192  * with a bunch of small requests.
1193  */
1194 static void xs_udp_write_space(struct sock *sk)
1195 {
1196         read_lock(&sk->sk_callback_lock);
1197
1198         /* from net/core/sock.c:sock_def_write_space */
1199         if (sock_writeable(sk)) {
1200                 struct socket *sock;
1201                 struct rpc_xprt *xprt;
1202
1203                 if (unlikely(!(sock = sk->sk_socket)))
1204                         goto out;
1205                 clear_bit(SOCK_NOSPACE, &sock->flags);
1206
1207                 if (unlikely(!(xprt = xprt_from_sock(sk))))
1208                         goto out;
1209                 if (test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags) == 0)
1210                         goto out;
1211
1212                 xprt_write_space(xprt);
1213         }
1214
1215  out:
1216         read_unlock(&sk->sk_callback_lock);
1217 }
1218
1219 /**
1220  * xs_tcp_write_space - callback invoked when socket buffer space
1221  *                             becomes available
1222  * @sk: socket whose state has changed
1223  *
1224  * Called when more output buffer space is available for this socket.
1225  * We try not to wake our writers until they can make "significant"
1226  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1227  * with a bunch of small requests.
1228  */
1229 static void xs_tcp_write_space(struct sock *sk)
1230 {
1231         read_lock(&sk->sk_callback_lock);
1232
1233         /* from net/core/stream.c:sk_stream_write_space */
1234         if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk)) {
1235                 struct socket *sock;
1236                 struct rpc_xprt *xprt;
1237
1238                 if (unlikely(!(sock = sk->sk_socket)))
1239                         goto out;
1240                 clear_bit(SOCK_NOSPACE, &sock->flags);
1241
1242                 if (unlikely(!(xprt = xprt_from_sock(sk))))
1243                         goto out;
1244                 if (test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags) == 0)
1245                         goto out;
1246
1247                 xprt_write_space(xprt);
1248         }
1249
1250  out:
1251         read_unlock(&sk->sk_callback_lock);
1252 }
1253
1254 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
1255 {
1256         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1257         struct sock *sk = transport->inet;
1258
1259         if (transport->rcvsize) {
1260                 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1261                 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
1262         }
1263         if (transport->sndsize) {
1264                 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1265                 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1266                 sk->sk_write_space(sk);
1267         }
1268 }
1269
1270 /**
1271  * xs_udp_set_buffer_size - set send and receive limits
1272  * @xprt: generic transport
1273  * @sndsize: requested size of send buffer, in bytes
1274  * @rcvsize: requested size of receive buffer, in bytes
1275  *
1276  * Set socket send and receive buffer size limits.
1277  */
1278 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1279 {
1280         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1281
1282         transport->sndsize = 0;
1283         if (sndsize)
1284                 transport->sndsize = sndsize + 1024;
1285         transport->rcvsize = 0;
1286         if (rcvsize)
1287                 transport->rcvsize = rcvsize + 1024;
1288
1289         xs_udp_do_set_buffer_size(xprt);
1290 }
1291
1292 /**
1293  * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1294  * @task: task that timed out
1295  *
1296  * Adjust the congestion window after a retransmit timeout has occurred.
1297  */
1298 static void xs_udp_timer(struct rpc_task *task)
1299 {
1300         xprt_adjust_cwnd(task, -ETIMEDOUT);
1301 }
1302
1303 static unsigned short xs_get_random_port(void)
1304 {
1305         unsigned short range = xprt_max_resvport - xprt_min_resvport;
1306         unsigned short rand = (unsigned short) net_random() % range;
1307         return rand + xprt_min_resvport;
1308 }
1309
1310 /**
1311  * xs_set_port - reset the port number in the remote endpoint address
1312  * @xprt: generic transport
1313  * @port: new port number
1314  *
1315  */
1316 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1317 {
1318         struct sockaddr *addr = xs_addr(xprt);
1319
1320         dprintk("RPC:       setting port for xprt %p to %u\n", xprt, port);
1321
1322         switch (addr->sa_family) {
1323         case AF_INET:
1324                 ((struct sockaddr_in *)addr)->sin_port = htons(port);
1325                 break;
1326         case AF_INET6:
1327                 ((struct sockaddr_in6 *)addr)->sin6_port = htons(port);
1328                 break;
1329         default:
1330                 BUG();
1331         }
1332 }
1333
1334 static unsigned short xs_get_srcport(struct sock_xprt *transport, struct socket *sock)
1335 {
1336         unsigned short port = transport->port;
1337
1338         if (port == 0 && transport->xprt.resvport)
1339                 port = xs_get_random_port();
1340         return port;
1341 }
1342
1343 static unsigned short xs_next_srcport(struct sock_xprt *transport, struct socket *sock, unsigned short port)
1344 {
1345         if (transport->port != 0)
1346                 transport->port = 0;
1347         if (!transport->xprt.resvport)
1348                 return 0;
1349         if (port <= xprt_min_resvport || port > xprt_max_resvport)
1350                 return xprt_max_resvport;
1351         return --port;
1352 }
1353
1354 static int xs_bind4(struct sock_xprt *transport, struct socket *sock)
1355 {
1356         struct sockaddr_in myaddr = {
1357                 .sin_family = AF_INET,
1358         };
1359         struct sockaddr_in *sa;
1360         int err, nloop = 0;
1361         unsigned short port = xs_get_srcport(transport, sock);
1362         unsigned short last;
1363
1364         sa = (struct sockaddr_in *)&transport->addr;
1365         myaddr.sin_addr = sa->sin_addr;
1366         do {
1367                 myaddr.sin_port = htons(port);
1368                 err = kernel_bind(sock, (struct sockaddr *) &myaddr,
1369                                                 sizeof(myaddr));
1370                 if (port == 0)
1371                         break;
1372                 if (err == 0) {
1373                         transport->port = port;
1374                         break;
1375                 }
1376                 last = port;
1377                 port = xs_next_srcport(transport, sock, port);
1378                 if (port > last)
1379                         nloop++;
1380         } while (err == -EADDRINUSE && nloop != 2);
1381         dprintk("RPC:       %s "NIPQUAD_FMT":%u: %s (%d)\n",
1382                         __func__, NIPQUAD(myaddr.sin_addr),
1383                         port, err ? "failed" : "ok", err);
1384         return err;
1385 }
1386
1387 static int xs_bind6(struct sock_xprt *transport, struct socket *sock)
1388 {
1389         struct sockaddr_in6 myaddr = {
1390                 .sin6_family = AF_INET6,
1391         };
1392         struct sockaddr_in6 *sa;
1393         int err, nloop = 0;
1394         unsigned short port = xs_get_srcport(transport, sock);
1395         unsigned short last;
1396
1397         sa = (struct sockaddr_in6 *)&transport->addr;
1398         myaddr.sin6_addr = sa->sin6_addr;
1399         do {
1400                 myaddr.sin6_port = htons(port);
1401                 err = kernel_bind(sock, (struct sockaddr *) &myaddr,
1402                                                 sizeof(myaddr));
1403                 if (port == 0)
1404                         break;
1405                 if (err == 0) {
1406                         transport->port = port;
1407                         break;
1408                 }
1409                 last = port;
1410                 port = xs_next_srcport(transport, sock, port);
1411                 if (port > last)
1412                         nloop++;
1413         } while (err == -EADDRINUSE && nloop != 2);
1414         dprintk("RPC:       xs_bind6 "NIP6_FMT":%u: %s (%d)\n",
1415                 NIP6(myaddr.sin6_addr), port, err ? "failed" : "ok", err);
1416         return err;
1417 }
1418
1419 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1420 static struct lock_class_key xs_key[2];
1421 static struct lock_class_key xs_slock_key[2];
1422
1423 static inline void xs_reclassify_socket4(struct socket *sock)
1424 {
1425         struct sock *sk = sock->sk;
1426
1427         BUG_ON(sock_owned_by_user(sk));
1428         sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1429                 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1430 }
1431
1432 static inline void xs_reclassify_socket6(struct socket *sock)
1433 {
1434         struct sock *sk = sock->sk;
1435
1436         BUG_ON(sock_owned_by_user(sk));
1437         sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1438                 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
1439 }
1440 #else
1441 static inline void xs_reclassify_socket4(struct socket *sock)
1442 {
1443 }
1444
1445 static inline void xs_reclassify_socket6(struct socket *sock)
1446 {
1447 }
1448 #endif
1449
1450 static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
1451 {
1452         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1453
1454         if (!transport->inet) {
1455                 struct sock *sk = sock->sk;
1456
1457                 write_lock_bh(&sk->sk_callback_lock);
1458
1459                 sk->sk_user_data = xprt;
1460                 transport->old_data_ready = sk->sk_data_ready;
1461                 transport->old_state_change = sk->sk_state_change;
1462                 transport->old_write_space = sk->sk_write_space;
1463                 sk->sk_data_ready = xs_udp_data_ready;
1464                 sk->sk_write_space = xs_udp_write_space;
1465                 sk->sk_no_check = UDP_CSUM_NORCV;
1466                 sk->sk_allocation = GFP_ATOMIC;
1467
1468                 xprt_set_connected(xprt);
1469
1470                 /* Reset to new socket */
1471                 transport->sock = sock;
1472                 transport->inet = sk;
1473
1474                 write_unlock_bh(&sk->sk_callback_lock);
1475         }
1476         xs_udp_do_set_buffer_size(xprt);
1477 }
1478
1479 /**
1480  * xs_udp_connect_worker4 - set up a UDP socket
1481  * @work: RPC transport to connect
1482  *
1483  * Invoked by a work queue tasklet.
1484  */
1485 static void xs_udp_connect_worker4(struct work_struct *work)
1486 {
1487         struct sock_xprt *transport =
1488                 container_of(work, struct sock_xprt, connect_worker.work);
1489         struct rpc_xprt *xprt = &transport->xprt;
1490         struct socket *sock = transport->sock;
1491         int err, status = -EIO;
1492
1493         if (xprt->shutdown || !xprt_bound(xprt))
1494                 goto out;
1495
1496         /* Start by resetting any existing state */
1497         xs_close(xprt);
1498
1499         if ((err = sock_create_kern(PF_INET, SOCK_DGRAM, IPPROTO_UDP, &sock)) < 0) {
1500                 dprintk("RPC:       can't create UDP transport socket (%d).\n", -err);
1501                 goto out;
1502         }
1503         xs_reclassify_socket4(sock);
1504
1505         if (xs_bind4(transport, sock)) {
1506                 sock_release(sock);
1507                 goto out;
1508         }
1509
1510         dprintk("RPC:       worker connecting xprt %p to address: %s\n",
1511                         xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1512
1513         xs_udp_finish_connecting(xprt, sock);
1514         status = 0;
1515 out:
1516         xprt_wake_pending_tasks(xprt, status);
1517         xprt_clear_connecting(xprt);
1518 }
1519
1520 /**
1521  * xs_udp_connect_worker6 - set up a UDP socket
1522  * @work: RPC transport to connect
1523  *
1524  * Invoked by a work queue tasklet.
1525  */
1526 static void xs_udp_connect_worker6(struct work_struct *work)
1527 {
1528         struct sock_xprt *transport =
1529                 container_of(work, struct sock_xprt, connect_worker.work);
1530         struct rpc_xprt *xprt = &transport->xprt;
1531         struct socket *sock = transport->sock;
1532         int err, status = -EIO;
1533
1534         if (xprt->shutdown || !xprt_bound(xprt))
1535                 goto out;
1536
1537         /* Start by resetting any existing state */
1538         xs_close(xprt);
1539
1540         if ((err = sock_create_kern(PF_INET6, SOCK_DGRAM, IPPROTO_UDP, &sock)) < 0) {
1541                 dprintk("RPC:       can't create UDP transport socket (%d).\n", -err);
1542                 goto out;
1543         }
1544         xs_reclassify_socket6(sock);
1545
1546         if (xs_bind6(transport, sock) < 0) {
1547                 sock_release(sock);
1548                 goto out;
1549         }
1550
1551         dprintk("RPC:       worker connecting xprt %p to address: %s\n",
1552                         xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1553
1554         xs_udp_finish_connecting(xprt, sock);
1555         status = 0;
1556 out:
1557         xprt_wake_pending_tasks(xprt, status);
1558         xprt_clear_connecting(xprt);
1559 }
1560
1561 /*
1562  * We need to preserve the port number so the reply cache on the server can
1563  * find our cached RPC replies when we get around to reconnecting.
1564  */
1565 static void xs_tcp_reuse_connection(struct rpc_xprt *xprt)
1566 {
1567         int result;
1568         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1569         struct sockaddr any;
1570
1571         dprintk("RPC:       disconnecting xprt %p to reuse port\n", xprt);
1572
1573         /*
1574          * Disconnect the transport socket by doing a connect operation
1575          * with AF_UNSPEC.  This should return immediately...
1576          */
1577         memset(&any, 0, sizeof(any));
1578         any.sa_family = AF_UNSPEC;
1579         result = kernel_connect(transport->sock, &any, sizeof(any), 0);
1580         if (result)
1581                 dprintk("RPC:       AF_UNSPEC connect return code %d\n",
1582                                 result);
1583 }
1584
1585 static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
1586 {
1587         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1588
1589         if (!transport->inet) {
1590                 struct sock *sk = sock->sk;
1591
1592                 write_lock_bh(&sk->sk_callback_lock);
1593
1594                 sk->sk_user_data = xprt;
1595                 transport->old_data_ready = sk->sk_data_ready;
1596                 transport->old_state_change = sk->sk_state_change;
1597                 transport->old_write_space = sk->sk_write_space;
1598                 sk->sk_data_ready = xs_tcp_data_ready;
1599                 sk->sk_state_change = xs_tcp_state_change;
1600                 sk->sk_write_space = xs_tcp_write_space;
1601                 sk->sk_allocation = GFP_ATOMIC;
1602
1603                 /* socket options */
1604                 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
1605                 sock_reset_flag(sk, SOCK_LINGER);
1606                 tcp_sk(sk)->linger2 = 0;
1607                 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
1608
1609                 xprt_clear_connected(xprt);
1610
1611                 /* Reset to new socket */
1612                 transport->sock = sock;
1613                 transport->inet = sk;
1614
1615                 write_unlock_bh(&sk->sk_callback_lock);
1616         }
1617
1618         /* Tell the socket layer to start connecting... */
1619         xprt->stat.connect_count++;
1620         xprt->stat.connect_start = jiffies;
1621         return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
1622 }
1623
1624 /**
1625  * xs_tcp_connect_worker4 - connect a TCP socket to a remote endpoint
1626  * @work: RPC transport to connect
1627  *
1628  * Invoked by a work queue tasklet.
1629  */
1630 static void xs_tcp_connect_worker4(struct work_struct *work)
1631 {
1632         struct sock_xprt *transport =
1633                 container_of(work, struct sock_xprt, connect_worker.work);
1634         struct rpc_xprt *xprt = &transport->xprt;
1635         struct socket *sock = transport->sock;
1636         int err, status = -EIO;
1637
1638         if (xprt->shutdown || !xprt_bound(xprt))
1639                 goto out;
1640
1641         if (!sock) {
1642                 /* start from scratch */
1643                 if ((err = sock_create_kern(PF_INET, SOCK_STREAM, IPPROTO_TCP, &sock)) < 0) {
1644                         dprintk("RPC:       can't create TCP transport socket (%d).\n", -err);
1645                         goto out;
1646                 }
1647                 xs_reclassify_socket4(sock);
1648
1649                 if (xs_bind4(transport, sock) < 0) {
1650                         sock_release(sock);
1651                         goto out;
1652                 }
1653         } else
1654                 /* "close" the socket, preserving the local port */
1655                 xs_tcp_reuse_connection(xprt);
1656
1657         dprintk("RPC:       worker connecting xprt %p to address: %s\n",
1658                         xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1659
1660         status = xs_tcp_finish_connecting(xprt, sock);
1661         dprintk("RPC:       %p connect status %d connected %d sock state %d\n",
1662                         xprt, -status, xprt_connected(xprt),
1663                         sock->sk->sk_state);
1664         if (status < 0) {
1665                 switch (status) {
1666                         case -EINPROGRESS:
1667                         case -EALREADY:
1668                                 goto out_clear;
1669                         case -ECONNREFUSED:
1670                         case -ECONNRESET:
1671                                 /* retry with existing socket, after a delay */
1672                                 break;
1673                         default:
1674                                 /* get rid of existing socket, and retry */
1675                                 xs_tcp_shutdown(xprt);
1676                 }
1677         }
1678 out:
1679         xprt_wake_pending_tasks(xprt, status);
1680 out_clear:
1681         xprt_clear_connecting(xprt);
1682 }
1683
1684 /**
1685  * xs_tcp_connect_worker6 - connect a TCP socket to a remote endpoint
1686  * @work: RPC transport to connect
1687  *
1688  * Invoked by a work queue tasklet.
1689  */
1690 static void xs_tcp_connect_worker6(struct work_struct *work)
1691 {
1692         struct sock_xprt *transport =
1693                 container_of(work, struct sock_xprt, connect_worker.work);
1694         struct rpc_xprt *xprt = &transport->xprt;
1695         struct socket *sock = transport->sock;
1696         int err, status = -EIO;
1697
1698         if (xprt->shutdown || !xprt_bound(xprt))
1699                 goto out;
1700
1701         if (!sock) {
1702                 /* start from scratch */
1703                 if ((err = sock_create_kern(PF_INET6, SOCK_STREAM, IPPROTO_TCP, &sock)) < 0) {
1704                         dprintk("RPC:       can't create TCP transport socket (%d).\n", -err);
1705                         goto out;
1706                 }
1707                 xs_reclassify_socket6(sock);
1708
1709                 if (xs_bind6(transport, sock) < 0) {
1710                         sock_release(sock);
1711                         goto out;
1712                 }
1713         } else
1714                 /* "close" the socket, preserving the local port */
1715                 xs_tcp_reuse_connection(xprt);
1716
1717         dprintk("RPC:       worker connecting xprt %p to address: %s\n",
1718                         xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1719
1720         status = xs_tcp_finish_connecting(xprt, sock);
1721         dprintk("RPC:       %p connect status %d connected %d sock state %d\n",
1722                         xprt, -status, xprt_connected(xprt), sock->sk->sk_state);
1723         if (status < 0) {
1724                 switch (status) {
1725                         case -EINPROGRESS:
1726                         case -EALREADY:
1727                                 goto out_clear;
1728                         case -ECONNREFUSED:
1729                         case -ECONNRESET:
1730                                 /* retry with existing socket, after a delay */
1731                                 break;
1732                         default:
1733                                 /* get rid of existing socket, and retry */
1734                                 xs_tcp_shutdown(xprt);
1735                 }
1736         }
1737 out:
1738         xprt_wake_pending_tasks(xprt, status);
1739 out_clear:
1740         xprt_clear_connecting(xprt);
1741 }
1742
1743 /**
1744  * xs_connect - connect a socket to a remote endpoint
1745  * @task: address of RPC task that manages state of connect request
1746  *
1747  * TCP: If the remote end dropped the connection, delay reconnecting.
1748  *
1749  * UDP socket connects are synchronous, but we use a work queue anyway
1750  * to guarantee that even unprivileged user processes can set up a
1751  * socket on a privileged port.
1752  *
1753  * If a UDP socket connect fails, the delay behavior here prevents
1754  * retry floods (hard mounts).
1755  */
1756 static void xs_connect(struct rpc_task *task)
1757 {
1758         struct rpc_xprt *xprt = task->tk_xprt;
1759         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1760
1761         if (xprt_test_and_set_connecting(xprt))
1762                 return;
1763
1764         if (transport->sock != NULL) {
1765                 dprintk("RPC:       xs_connect delayed xprt %p for %lu "
1766                                 "seconds\n",
1767                                 xprt, xprt->reestablish_timeout / HZ);
1768                 queue_delayed_work(rpciod_workqueue,
1769                                    &transport->connect_worker,
1770                                    xprt->reestablish_timeout);
1771                 xprt->reestablish_timeout <<= 1;
1772                 if (xprt->reestablish_timeout > XS_TCP_MAX_REEST_TO)
1773                         xprt->reestablish_timeout = XS_TCP_MAX_REEST_TO;
1774         } else {
1775                 dprintk("RPC:       xs_connect scheduled xprt %p\n", xprt);
1776                 queue_delayed_work(rpciod_workqueue,
1777                                    &transport->connect_worker, 0);
1778         }
1779 }
1780
1781 static void xs_tcp_connect(struct rpc_task *task)
1782 {
1783         struct rpc_xprt *xprt = task->tk_xprt;
1784
1785         /* Initiate graceful shutdown of the socket if not already done */
1786         if (test_bit(XPRT_CONNECTED, &xprt->state))
1787                 xs_tcp_shutdown(xprt);
1788         /* Exit if we need to wait for socket shutdown to complete */
1789         if (test_bit(XPRT_CLOSING, &xprt->state))
1790                 return;
1791         xs_connect(task);
1792 }
1793
1794 /**
1795  * xs_udp_print_stats - display UDP socket-specifc stats
1796  * @xprt: rpc_xprt struct containing statistics
1797  * @seq: output file
1798  *
1799  */
1800 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
1801 {
1802         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1803
1804         seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %Lu %Lu\n",
1805                         transport->port,
1806                         xprt->stat.bind_count,
1807                         xprt->stat.sends,
1808                         xprt->stat.recvs,
1809                         xprt->stat.bad_xids,
1810                         xprt->stat.req_u,
1811                         xprt->stat.bklog_u);
1812 }
1813
1814 /**
1815  * xs_tcp_print_stats - display TCP socket-specifc stats
1816  * @xprt: rpc_xprt struct containing statistics
1817  * @seq: output file
1818  *
1819  */
1820 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
1821 {
1822         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1823         long idle_time = 0;
1824
1825         if (xprt_connected(xprt))
1826                 idle_time = (long)(jiffies - xprt->last_used) / HZ;
1827
1828         seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu %Lu %Lu\n",
1829                         transport->port,
1830                         xprt->stat.bind_count,
1831                         xprt->stat.connect_count,
1832                         xprt->stat.connect_time,
1833                         idle_time,
1834                         xprt->stat.sends,
1835                         xprt->stat.recvs,
1836                         xprt->stat.bad_xids,
1837                         xprt->stat.req_u,
1838                         xprt->stat.bklog_u);
1839 }
1840
1841 static struct rpc_xprt_ops xs_udp_ops = {
1842         .set_buffer_size        = xs_udp_set_buffer_size,
1843         .reserve_xprt           = xprt_reserve_xprt_cong,
1844         .release_xprt           = xprt_release_xprt_cong,
1845         .rpcbind                = rpcb_getport_async,
1846         .set_port               = xs_set_port,
1847         .connect                = xs_connect,
1848         .buf_alloc              = rpc_malloc,
1849         .buf_free               = rpc_free,
1850         .send_request           = xs_udp_send_request,
1851         .set_retrans_timeout    = xprt_set_retrans_timeout_rtt,
1852         .timer                  = xs_udp_timer,
1853         .release_request        = xprt_release_rqst_cong,
1854         .close                  = xs_close,
1855         .destroy                = xs_destroy,
1856         .print_stats            = xs_udp_print_stats,
1857 };
1858
1859 static struct rpc_xprt_ops xs_tcp_ops = {
1860         .reserve_xprt           = xprt_reserve_xprt,
1861         .release_xprt           = xs_tcp_release_xprt,
1862         .rpcbind                = rpcb_getport_async,
1863         .set_port               = xs_set_port,
1864         .connect                = xs_tcp_connect,
1865         .buf_alloc              = rpc_malloc,
1866         .buf_free               = rpc_free,
1867         .send_request           = xs_tcp_send_request,
1868         .set_retrans_timeout    = xprt_set_retrans_timeout_def,
1869         .close                  = xs_tcp_shutdown,
1870         .destroy                = xs_destroy,
1871         .print_stats            = xs_tcp_print_stats,
1872 };
1873
1874 static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
1875                                       unsigned int slot_table_size)
1876 {
1877         struct rpc_xprt *xprt;
1878         struct sock_xprt *new;
1879
1880         if (args->addrlen > sizeof(xprt->addr)) {
1881                 dprintk("RPC:       xs_setup_xprt: address too large\n");
1882                 return ERR_PTR(-EBADF);
1883         }
1884
1885         new = kzalloc(sizeof(*new), GFP_KERNEL);
1886         if (new == NULL) {
1887                 dprintk("RPC:       xs_setup_xprt: couldn't allocate "
1888                                 "rpc_xprt\n");
1889                 return ERR_PTR(-ENOMEM);
1890         }
1891         xprt = &new->xprt;
1892
1893         xprt->max_reqs = slot_table_size;
1894         xprt->slot = kcalloc(xprt->max_reqs, sizeof(struct rpc_rqst), GFP_KERNEL);
1895         if (xprt->slot == NULL) {
1896                 kfree(xprt);
1897                 dprintk("RPC:       xs_setup_xprt: couldn't allocate slot "
1898                                 "table\n");
1899                 return ERR_PTR(-ENOMEM);
1900         }
1901
1902         memcpy(&xprt->addr, args->dstaddr, args->addrlen);
1903         xprt->addrlen = args->addrlen;
1904         if (args->srcaddr)
1905                 memcpy(&new->addr, args->srcaddr, args->addrlen);
1906
1907         return xprt;
1908 }
1909
1910 static const struct rpc_timeout xs_udp_default_timeout = {
1911         .to_initval = 5 * HZ,
1912         .to_maxval = 30 * HZ,
1913         .to_increment = 5 * HZ,
1914         .to_retries = 5,
1915 };
1916
1917 /**
1918  * xs_setup_udp - Set up transport to use a UDP socket
1919  * @args: rpc transport creation arguments
1920  *
1921  */
1922 static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
1923 {
1924         struct sockaddr *addr = args->dstaddr;
1925         struct rpc_xprt *xprt;
1926         struct sock_xprt *transport;
1927
1928         xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries);
1929         if (IS_ERR(xprt))
1930                 return xprt;
1931         transport = container_of(xprt, struct sock_xprt, xprt);
1932
1933         xprt->prot = IPPROTO_UDP;
1934         xprt->tsh_size = 0;
1935         /* XXX: header size can vary due to auth type, IPv6, etc. */
1936         xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
1937
1938         xprt->bind_timeout = XS_BIND_TO;
1939         xprt->connect_timeout = XS_UDP_CONN_TO;
1940         xprt->reestablish_timeout = XS_UDP_REEST_TO;
1941         xprt->idle_timeout = XS_IDLE_DISC_TO;
1942
1943         xprt->ops = &xs_udp_ops;
1944
1945         xprt->timeout = &xs_udp_default_timeout;
1946
1947         switch (addr->sa_family) {
1948         case AF_INET:
1949                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
1950                         xprt_set_bound(xprt);
1951
1952                 INIT_DELAYED_WORK(&transport->connect_worker,
1953                                         xs_udp_connect_worker4);
1954                 xs_format_ipv4_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
1955                 break;
1956         case AF_INET6:
1957                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
1958                         xprt_set_bound(xprt);
1959
1960                 INIT_DELAYED_WORK(&transport->connect_worker,
1961                                         xs_udp_connect_worker6);
1962                 xs_format_ipv6_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
1963                 break;
1964         default:
1965                 kfree(xprt);
1966                 return ERR_PTR(-EAFNOSUPPORT);
1967         }
1968
1969         dprintk("RPC:       set up transport to address %s\n",
1970                         xprt->address_strings[RPC_DISPLAY_ALL]);
1971
1972         if (try_module_get(THIS_MODULE))
1973                 return xprt;
1974
1975         kfree(xprt->slot);
1976         kfree(xprt);
1977         return ERR_PTR(-EINVAL);
1978 }
1979
1980 static const struct rpc_timeout xs_tcp_default_timeout = {
1981         .to_initval = 60 * HZ,
1982         .to_maxval = 60 * HZ,
1983         .to_retries = 2,
1984 };
1985
1986 /**
1987  * xs_setup_tcp - Set up transport to use a TCP socket
1988  * @args: rpc transport creation arguments
1989  *
1990  */
1991 static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
1992 {
1993         struct sockaddr *addr = args->dstaddr;
1994         struct rpc_xprt *xprt;
1995         struct sock_xprt *transport;
1996
1997         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries);
1998         if (IS_ERR(xprt))
1999                 return xprt;
2000         transport = container_of(xprt, struct sock_xprt, xprt);
2001
2002         xprt->prot = IPPROTO_TCP;
2003         xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2004         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2005
2006         xprt->bind_timeout = XS_BIND_TO;
2007         xprt->connect_timeout = XS_TCP_CONN_TO;
2008         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2009         xprt->idle_timeout = XS_IDLE_DISC_TO;
2010
2011         xprt->ops = &xs_tcp_ops;
2012         xprt->timeout = &xs_tcp_default_timeout;
2013
2014         switch (addr->sa_family) {
2015         case AF_INET:
2016                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2017                         xprt_set_bound(xprt);
2018
2019                 INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_connect_worker4);
2020                 xs_format_ipv4_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
2021                 break;
2022         case AF_INET6:
2023                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2024                         xprt_set_bound(xprt);
2025
2026                 INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_connect_worker6);
2027                 xs_format_ipv6_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
2028                 break;
2029         default:
2030                 kfree(xprt);
2031                 return ERR_PTR(-EAFNOSUPPORT);
2032         }
2033
2034         dprintk("RPC:       set up transport to address %s\n",
2035                         xprt->address_strings[RPC_DISPLAY_ALL]);
2036
2037         if (try_module_get(THIS_MODULE))
2038                 return xprt;
2039
2040         kfree(xprt->slot);
2041         kfree(xprt);
2042         return ERR_PTR(-EINVAL);
2043 }
2044
2045 static struct xprt_class        xs_udp_transport = {
2046         .list           = LIST_HEAD_INIT(xs_udp_transport.list),
2047         .name           = "udp",
2048         .owner          = THIS_MODULE,
2049         .ident          = IPPROTO_UDP,
2050         .setup          = xs_setup_udp,
2051 };
2052
2053 static struct xprt_class        xs_tcp_transport = {
2054         .list           = LIST_HEAD_INIT(xs_tcp_transport.list),
2055         .name           = "tcp",
2056         .owner          = THIS_MODULE,
2057         .ident          = IPPROTO_TCP,
2058         .setup          = xs_setup_tcp,
2059 };
2060
2061 /**
2062  * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
2063  *
2064  */
2065 int init_socket_xprt(void)
2066 {
2067 #ifdef RPC_DEBUG
2068         if (!sunrpc_table_header)
2069                 sunrpc_table_header = register_sysctl_table(sunrpc_table);
2070 #endif
2071
2072         xprt_register_transport(&xs_udp_transport);
2073         xprt_register_transport(&xs_tcp_transport);
2074
2075         return 0;
2076 }
2077
2078 /**
2079  * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
2080  *
2081  */
2082 void cleanup_socket_xprt(void)
2083 {
2084 #ifdef RPC_DEBUG
2085         if (sunrpc_table_header) {
2086                 unregister_sysctl_table(sunrpc_table_header);
2087                 sunrpc_table_header = NULL;
2088         }
2089 #endif
2090
2091         xprt_unregister_transport(&xs_udp_transport);
2092         xprt_unregister_transport(&xs_tcp_transport);
2093 }