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
   7  * TCP send fixes (C) 1998 Red Hat
   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         void                    (*old_error_report)(struct sock *);
 253 };
 254
 255 /*
 256  * TCP receive state flags
 257  */
 258 #define TCP_RCV_LAST_FRAG       (1UL << 0)
 259 #define TCP_RCV_COPY_FRAGHDR    (1UL << 1)
 260 #define TCP_RCV_COPY_XID        (1UL << 2)
 261 #define TCP_RCV_COPY_DATA       (1UL << 3)
 262
 263 static inline struct sockaddr *xs_addr(struct rpc_xprt *xprt)
 264 {
 265         return (struct sockaddr *) &xprt->addr;
 266 }
 267
 268 static inline struct sockaddr_in *xs_addr_in(struct rpc_xprt *xprt)
 269 {
 270         return (struct sockaddr_in *) &xprt->addr;
 271 }
 272
 273 static inline struct sockaddr_in6 *xs_addr_in6(struct rpc_xprt *xprt)
 274 {
 275         return (struct sockaddr_in6 *) &xprt->addr;
 276 }
 277
 278 static void xs_format_ipv4_peer_addresses(struct rpc_xprt *xprt,
 279                                           const char *protocol,
 280                                           const char *netid)
 281 {
 282         struct sockaddr_in *addr = xs_addr_in(xprt);
 283         char *buf;
 284
 285         buf = kzalloc(20, GFP_KERNEL);
 286         if (buf) {
 287                 snprintf(buf, 20, "%pI4", &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=%pI4 port=%u proto=%s",
 303                         &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, "%pI4.%u.%u",
 326                                 &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, "%pI6",&addr->sin6_addr);
 345         }
 346         xprt->address_strings[RPC_DISPLAY_ADDR] = buf;
 347
 348         buf = kzalloc(8, GFP_KERNEL);
 349         if (buf) {
 350                 snprintf(buf, 8, "%u",
 351                                 ntohs(addr->sin6_port));
 352         }
 353         xprt->address_strings[RPC_DISPLAY_PORT] = buf;
 354
 355         xprt->address_strings[RPC_DISPLAY_PROTO] = protocol;
 356
 357         buf = kzalloc(64, GFP_KERNEL);
 358         if (buf) {
 359                 snprintf(buf, 64, "addr=%pI6 port=%u proto=%s",
 360                                 &addr->sin6_addr,
 361                                 ntohs(addr->sin6_port),
 362                                 protocol);
 363         }
 364         xprt->address_strings[RPC_DISPLAY_ALL] = buf;
 365
 366         buf = kzalloc(36, GFP_KERNEL);
 367         if (buf)
 368                 snprintf(buf, 36, "%pi6", &addr->sin6_addr);
 369
 370         xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = buf;
 371
 372         buf = kzalloc(8, GFP_KERNEL);
 373         if (buf) {
 374                 snprintf(buf, 8, "%4hx",
 375                                 ntohs(addr->sin6_port));
 376         }
 377         xprt->address_strings[RPC_DISPLAY_HEX_PORT] = buf;
 378
 379         buf = kzalloc(50, GFP_KERNEL);
 380         if (buf) {
 381                 snprintf(buf, 50, "%pI6.%u.%u",
 382                          &addr->sin6_addr,
 383                          ntohs(addr->sin6_port) >> 8,
 384                          ntohs(addr->sin6_port) & 0xff);
 385         }
 386         xprt->address_strings[RPC_DISPLAY_UNIVERSAL_ADDR] = buf;
 387
 388         xprt->address_strings[RPC_DISPLAY_NETID] = netid;
 389 }
 390
 391 static void xs_free_peer_addresses(struct rpc_xprt *xprt)
 392 {
 393         unsigned int i;
 394
 395         for (i = 0; i < RPC_DISPLAY_MAX; i++)
 396                 switch (i) {
 397                 case RPC_DISPLAY_PROTO:
 398                 case RPC_DISPLAY_NETID:
 399                         continue;
 400                 default:
 401                         kfree(xprt->address_strings[i]);
 402                 }
 403 }
 404
 405 #define XS_SENDMSG_FLAGS        (MSG_DONTWAIT | MSG_NOSIGNAL)
 406
 407 static int xs_send_kvec(struct socket *sock, struct sockaddr *addr, int addrlen, struct kvec *vec, unsigned int base, int more)
 408 {
 409         struct msghdr msg = {
 410                 .msg_name       = addr,
 411                 .msg_namelen    = addrlen,
 412                 .msg_flags      = XS_SENDMSG_FLAGS | (more ? MSG_MORE : 0),
 413         };
 414         struct kvec iov = {
 415                 .iov_base       = vec->iov_base + base,
 416                 .iov_len        = vec->iov_len - base,
 417         };
 418
 419         if (iov.iov_len != 0)
 420                 return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
 421         return kernel_sendmsg(sock, &msg, NULL, 0, 0);
 422 }
 423
 424 static int xs_send_pagedata(struct socket *sock, struct xdr_buf *xdr, unsigned int base, int more)
 425 {
 426         struct page **ppage;
 427         unsigned int remainder;
 428         int err, sent = 0;
 429
 430         remainder = xdr->page_len - base;
 431         base += xdr->page_base;
 432         ppage = xdr->pages + (base >> PAGE_SHIFT);
 433         base &= ~PAGE_MASK;
 434         for(;;) {
 435                 unsigned int len = min_t(unsigned int, PAGE_SIZE - base, remainder);
 436                 int flags = XS_SENDMSG_FLAGS;
 437
 438                 remainder -= len;
 439                 if (remainder != 0 || more)
 440                         flags |= MSG_MORE;
 441                 err = sock->ops->sendpage(sock, *ppage, base, len, flags);
 442                 if (remainder == 0 || err != len)
 443                         break;
 444                 sent += err;
 445                 ppage++;
 446                 base = 0;
 447         }
 448         if (sent == 0)
 449                 return err;
 450         if (err > 0)
 451                 sent += err;
 452         return sent;
 453 }
 454
 455 /**
 456  * xs_sendpages - write pages directly to a socket
 457  * @sock: socket to send on
 458  * @addr: UDP only -- address of destination
 459  * @addrlen: UDP only -- length of destination address
 460  * @xdr: buffer containing this request
 461  * @base: starting position in the buffer
 462  *
 463  */
 464 static int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base)
 465 {
 466         unsigned int remainder = xdr->len - base;
 467         int err, sent = 0;
 468
 469         if (unlikely(!sock))
 470                 return -ENOTSOCK;
 471
 472         clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags);
 473         if (base != 0) {
 474                 addr = NULL;
 475                 addrlen = 0;
 476         }
 477
 478         if (base < xdr->head[0].iov_len || addr != NULL) {
 479                 unsigned int len = xdr->head[0].iov_len - base;
 480                 remainder -= len;
 481                 err = xs_send_kvec(sock, addr, addrlen, &xdr->head[0], base, remainder != 0);
 482                 if (remainder == 0 || err != len)
 483                         goto out;
 484                 sent += err;
 485                 base = 0;
 486         } else
 487                 base -= xdr->head[0].iov_len;
 488
 489         if (base < xdr->page_len) {
 490                 unsigned int len = xdr->page_len - base;
 491                 remainder -= len;
 492                 err = xs_send_pagedata(sock, xdr, base, remainder != 0);
 493                 if (remainder == 0 || err != len)
 494                         goto out;
 495                 sent += err;
 496                 base = 0;
 497         } else
 498                 base -= xdr->page_len;
 499
 500         if (base >= xdr->tail[0].iov_len)
 501                 return sent;
 502         err = xs_send_kvec(sock, NULL, 0, &xdr->tail[0], base, 0);
 503 out:
 504         if (sent == 0)
 505                 return err;
 506         if (err > 0)
 507                 sent += err;
 508         return sent;
 509 }
 510
 511 static void xs_nospace_callback(struct rpc_task *task)
 512 {
 513         struct sock_xprt *transport = container_of(task->tk_rqstp->rq_xprt, struct sock_xprt, xprt);
 514
 515         transport->inet->sk_write_pending--;
 516         clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 517 }
 518
 519 /**
 520  * xs_nospace - place task on wait queue if transmit was incomplete
 521  * @task: task to put to sleep
 522  *
 523  */
 524 static int xs_nospace(struct rpc_task *task)
 525 {
 526         struct rpc_rqst *req = task->tk_rqstp;
 527         struct rpc_xprt *xprt = req->rq_xprt;
 528         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 529         int ret = 0;
 530
 531         dprintk("RPC: %5u xmit incomplete (%u left of %u)\n",
 532                         task->tk_pid, req->rq_slen - req->rq_bytes_sent,
 533                         req->rq_slen);
 534
 535         /* Protect against races with write_space */
 536         spin_lock_bh(&xprt->transport_lock);
 537
 538         /* Don't race with disconnect */
 539         if (xprt_connected(xprt)) {
 540                 if (test_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags)) {
 541                         ret = -EAGAIN;
 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                 ret = -ENOTCONN;
 554         }
 555
 556         spin_unlock_bh(&xprt->transport_lock);
 557         return ret;
 558 }
 559
 560 /**
 561  * xs_udp_send_request - write an RPC request to a UDP socket
 562  * @task: address of RPC task that manages the state of an RPC request
 563  *
 564  * Return values:
 565  *        0:    The request has been sent
 566  *   EAGAIN:    The socket was blocked, please call again later to
 567  *              complete the request
 568  * ENOTCONN:    Caller needs to invoke connect logic then call again
 569  *    other:    Some other error occured, the request was not sent
 570  */
 571 static int xs_udp_send_request(struct rpc_task *task)
 572 {
 573         struct rpc_rqst *req = task->tk_rqstp;
 574         struct rpc_xprt *xprt = req->rq_xprt;
 575         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 576         struct xdr_buf *xdr = &req->rq_snd_buf;
 577         int status;
 578
 579         xs_pktdump("packet data:",
 580                                 req->rq_svec->iov_base,
 581                                 req->rq_svec->iov_len);
 582
 583         if (!xprt_bound(xprt))
 584                 return -ENOTCONN;
 585         status = xs_sendpages(transport->sock,
 586                               xs_addr(xprt),
 587                               xprt->addrlen, xdr,
 588                               req->rq_bytes_sent);
 589
 590         dprintk("RPC:       xs_udp_send_request(%u) = %d\n",
 591                         xdr->len - req->rq_bytes_sent, status);
 592
 593         if (status >= 0) {
 594                 task->tk_bytes_sent += status;
 595                 if (status >= req->rq_slen)
 596                         return 0;
 597                 /* Still some bytes left; set up for a retry later. */
 598                 status = -EAGAIN;
 599         }
 600         if (!transport->sock)
 601                 goto out;
 602
 603         switch (status) {
 604         case -ENOTSOCK:
 605                 status = -ENOTCONN;
 606                 /* Should we call xs_close() here? */
 607                 break;
 608         case -EAGAIN:
 609                 status = xs_nospace(task);
 610                 break;
 611         default:
 612                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
 613                         -status);
 614         case -ENETUNREACH:
 615         case -EPIPE:
 616         case -ECONNREFUSED:
 617                 /* When the server has died, an ICMP port unreachable message
 618                  * prompts ECONNREFUSED. */
 619                 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 620         }
 621 out:
 622         return status;
 623 }
 624
 625 /**
 626  * xs_tcp_shutdown - gracefully shut down a TCP socket
 627  * @xprt: transport
 628  *
 629  * Initiates a graceful shutdown of the TCP socket by calling the
 630  * equivalent of shutdown(SHUT_WR);
 631  */
 632 static void xs_tcp_shutdown(struct rpc_xprt *xprt)
 633 {
 634         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 635         struct socket *sock = transport->sock;
 636
 637         if (sock != NULL)
 638                 kernel_sock_shutdown(sock, SHUT_WR);
 639 }
 640
 641 static inline void xs_encode_tcp_record_marker(struct xdr_buf *buf)
 642 {
 643         u32 reclen = buf->len - sizeof(rpc_fraghdr);
 644         rpc_fraghdr *base = buf->head[0].iov_base;
 645         *base = htonl(RPC_LAST_STREAM_FRAGMENT | reclen);
 646 }
 647
 648 /**
 649  * xs_tcp_send_request - write an RPC request to a TCP socket
 650  * @task: address of RPC task that manages the state of an RPC request
 651  *
 652  * Return values:
 653  *        0:    The request has been sent
 654  *   EAGAIN:    The socket was blocked, please call again later to
 655  *              complete the request
 656  * ENOTCONN:    Caller needs to invoke connect logic then call again
 657  *    other:    Some other error occured, the request was not sent
 658  *
 659  * XXX: In the case of soft timeouts, should we eventually give up
 660  *      if sendmsg is not able to make progress?
 661  */
 662 static int xs_tcp_send_request(struct rpc_task *task)
 663 {
 664         struct rpc_rqst *req = task->tk_rqstp;
 665         struct rpc_xprt *xprt = req->rq_xprt;
 666         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 667         struct xdr_buf *xdr = &req->rq_snd_buf;
 668         int status;
 669
 670         xs_encode_tcp_record_marker(&req->rq_snd_buf);
 671
 672         xs_pktdump("packet data:",
 673                                 req->rq_svec->iov_base,
 674                                 req->rq_svec->iov_len);
 675
 676         /* Continue transmitting the packet/record. We must be careful
 677          * to cope with writespace callbacks arriving _after_ we have
 678          * called sendmsg(). */
 679         while (1) {
 680                 status = xs_sendpages(transport->sock,
 681                                         NULL, 0, xdr, req->rq_bytes_sent);
 682
 683                 dprintk("RPC:       xs_tcp_send_request(%u) = %d\n",
 684                                 xdr->len - req->rq_bytes_sent, status);
 685
 686                 if (unlikely(status < 0))
 687                         break;
 688
 689                 /* If we've sent the entire packet, immediately
 690                  * reset the count of bytes sent. */
 691                 req->rq_bytes_sent += status;
 692                 task->tk_bytes_sent += status;
 693                 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
 694                         req->rq_bytes_sent = 0;
 695                         return 0;
 696                 }
 697
 698                 if (status != 0)
 699                         continue;
 700                 status = -EAGAIN;
 701                 break;
 702         }
 703         if (!transport->sock)
 704                 goto out;
 705
 706         switch (status) {
 707         case -ENOTSOCK:
 708                 status = -ENOTCONN;
 709                 /* Should we call xs_close() here? */
 710                 break;
 711         case -EAGAIN:
 712                 status = xs_nospace(task);
 713                 break;
 714         default:
 715                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
 716                         -status);
 717         case -ECONNRESET:
 718                 xs_tcp_shutdown(xprt);
 719         case -ECONNREFUSED:
 720         case -ENOTCONN:
 721         case -EPIPE:
 722                 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 723         }
 724 out:
 725         return status;
 726 }
 727
 728 /**
 729  * xs_tcp_release_xprt - clean up after a tcp transmission
 730  * @xprt: transport
 731  * @task: rpc task
 732  *
 733  * This cleans up if an error causes us to abort the transmission of a request.
 734  * In this case, the socket may need to be reset in order to avoid confusing
 735  * the server.
 736  */
 737 static void xs_tcp_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
 738 {
 739         struct rpc_rqst *req;
 740
 741         if (task != xprt->snd_task)
 742                 return;
 743         if (task == NULL)
 744                 goto out_release;
 745         req = task->tk_rqstp;
 746         if (req->rq_bytes_sent == 0)
 747                 goto out_release;
 748         if (req->rq_bytes_sent == req->rq_snd_buf.len)
 749                 goto out_release;
 750         set_bit(XPRT_CLOSE_WAIT, &task->tk_xprt->state);
 751 out_release:
 752         xprt_release_xprt(xprt, task);
 753 }
 754
 755 static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk)
 756 {
 757         transport->old_data_ready = sk->sk_data_ready;
 758         transport->old_state_change = sk->sk_state_change;
 759         transport->old_write_space = sk->sk_write_space;
 760         transport->old_error_report = sk->sk_error_report;
 761 }
 762
 763 static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk)
 764 {
 765         sk->sk_data_ready = transport->old_data_ready;
 766         sk->sk_state_change = transport->old_state_change;
 767         sk->sk_write_space = transport->old_write_space;
 768         sk->sk_error_report = transport->old_error_report;
 769 }
 770
 771 static void xs_reset_transport(struct sock_xprt *transport)
 772 {
 773         struct socket *sock = transport->sock;
 774         struct sock *sk = transport->inet;
 775
 776         if (sk == NULL)
 777                 return;
 778
 779         write_lock_bh(&sk->sk_callback_lock);
 780         transport->inet = NULL;
 781         transport->sock = NULL;
 782
 783         sk->sk_user_data = NULL;
 784
 785         xs_restore_old_callbacks(transport, sk);
 786         write_unlock_bh(&sk->sk_callback_lock);
 787
 788         sk->sk_no_check = 0;
 789
 790         sock_release(sock);
 791 }
 792
 793 /**
 794  * xs_close - close a socket
 795  * @xprt: transport
 796  *
 797  * This is used when all requests are complete; ie, no DRC state remains
 798  * on the server we want to save.
 799  */
 800 static void xs_close(struct rpc_xprt *xprt)
 801 {
 802         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 803
 804         dprintk("RPC:       xs_close xprt %p\n", xprt);
 805
 806         xs_reset_transport(transport);
 807
 808         smp_mb__before_clear_bit();
 809         clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
 810         clear_bit(XPRT_CLOSING, &xprt->state);
 811         smp_mb__after_clear_bit();
 812         xprt_disconnect_done(xprt);
 813 }
 814
 815 /**
 816  * xs_destroy - prepare to shutdown a transport
 817  * @xprt: doomed transport
 818  *
 819  */
 820 static void xs_destroy(struct rpc_xprt *xprt)
 821 {
 822         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 823
 824         dprintk("RPC:       xs_destroy xprt %p\n", xprt);
 825
 826         cancel_rearming_delayed_work(&transport->connect_worker);
 827
 828         xs_close(xprt);
 829         xs_free_peer_addresses(xprt);
 830         kfree(xprt->slot);
 831         kfree(xprt);
 832         module_put(THIS_MODULE);
 833 }
 834
 835 static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
 836 {
 837         return (struct rpc_xprt *) sk->sk_user_data;
 838 }
 839
 840 /**
 841  * xs_udp_data_ready - "data ready" callback for UDP sockets
 842  * @sk: socket with data to read
 843  * @len: how much data to read
 844  *
 845  */
 846 static void xs_udp_data_ready(struct sock *sk, int len)
 847 {
 848         struct rpc_task *task;
 849         struct rpc_xprt *xprt;
 850         struct rpc_rqst *rovr;
 851         struct sk_buff *skb;
 852         int err, repsize, copied;
 853         u32 _xid;
 854         __be32 *xp;
 855
 856         read_lock(&sk->sk_callback_lock);
 857         dprintk("RPC:       xs_udp_data_ready...\n");
 858         if (!(xprt = xprt_from_sock(sk)))
 859                 goto out;
 860
 861         if ((skb = skb_recv_datagram(sk, 0, 1, &err)) == NULL)
 862                 goto out;
 863
 864         if (xprt->shutdown)
 865                 goto dropit;
 866
 867         repsize = skb->len - sizeof(struct udphdr);
 868         if (repsize < 4) {
 869                 dprintk("RPC:       impossible RPC reply size %d!\n", repsize);
 870                 goto dropit;
 871         }
 872
 873         /* Copy the XID from the skb... */
 874         xp = skb_header_pointer(skb, sizeof(struct udphdr),
 875                                 sizeof(_xid), &_xid);
 876         if (xp == NULL)
 877                 goto dropit;
 878
 879         /* Look up and lock the request corresponding to the given XID */
 880         spin_lock(&xprt->transport_lock);
 881         rovr = xprt_lookup_rqst(xprt, *xp);
 882         if (!rovr)
 883                 goto out_unlock;
 884         task = rovr->rq_task;
 885
 886         if ((copied = rovr->rq_private_buf.buflen) > repsize)
 887                 copied = repsize;
 888
 889         /* Suck it into the iovec, verify checksum if not done by hw. */
 890         if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
 891                 UDPX_INC_STATS_BH(sk, UDP_MIB_INERRORS);
 892                 goto out_unlock;
 893         }
 894
 895         UDPX_INC_STATS_BH(sk, UDP_MIB_INDATAGRAMS);
 896
 897         /* Something worked... */
 898         dst_confirm(skb->dst);
 899
 900         xprt_adjust_cwnd(task, copied);
 901         xprt_update_rtt(task);
 902         xprt_complete_rqst(task, copied);
 903
 904  out_unlock:
 905         spin_unlock(&xprt->transport_lock);
 906  dropit:
 907         skb_free_datagram(sk, skb);
 908  out:
 909         read_unlock(&sk->sk_callback_lock);
 910 }
 911
 912 static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
 913 {
 914         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 915         size_t len, used;
 916         char *p;
 917
 918         p = ((char *) &transport->tcp_fraghdr) + transport->tcp_offset;
 919         len = sizeof(transport->tcp_fraghdr) - transport->tcp_offset;
 920         used = xdr_skb_read_bits(desc, p, len);
 921         transport->tcp_offset += used;
 922         if (used != len)
 923                 return;
 924
 925         transport->tcp_reclen = ntohl(transport->tcp_fraghdr);
 926         if (transport->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
 927                 transport->tcp_flags |= TCP_RCV_LAST_FRAG;
 928         else
 929                 transport->tcp_flags &= ~TCP_RCV_LAST_FRAG;
 930         transport->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
 931
 932         transport->tcp_flags &= ~TCP_RCV_COPY_FRAGHDR;
 933         transport->tcp_offset = 0;
 934
 935         /* Sanity check of the record length */
 936         if (unlikely(transport->tcp_reclen < 4)) {
 937                 dprintk("RPC:       invalid TCP record fragment length\n");
 938                 xprt_force_disconnect(xprt);
 939                 return;
 940         }
 941         dprintk("RPC:       reading TCP record fragment of length %d\n",
 942                         transport->tcp_reclen);
 943 }
 944
 945 static void xs_tcp_check_fraghdr(struct sock_xprt *transport)
 946 {
 947         if (transport->tcp_offset == transport->tcp_reclen) {
 948                 transport->tcp_flags |= TCP_RCV_COPY_FRAGHDR;
 949                 transport->tcp_offset = 0;
 950                 if (transport->tcp_flags & TCP_RCV_LAST_FRAG) {
 951                         transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
 952                         transport->tcp_flags |= TCP_RCV_COPY_XID;
 953                         transport->tcp_copied = 0;
 954                 }
 955         }
 956 }
 957
 958 static inline void xs_tcp_read_xid(struct sock_xprt *transport, struct xdr_skb_reader *desc)
 959 {
 960         size_t len, used;
 961         char *p;
 962
 963         len = sizeof(transport->tcp_xid) - transport->tcp_offset;
 964         dprintk("RPC:       reading XID (%Zu bytes)\n", len);
 965         p = ((char *) &transport->tcp_xid) + transport->tcp_offset;
 966         used = xdr_skb_read_bits(desc, p, len);
 967         transport->tcp_offset += used;
 968         if (used != len)
 969                 return;
 970         transport->tcp_flags &= ~TCP_RCV_COPY_XID;
 971         transport->tcp_flags |= TCP_RCV_COPY_DATA;
 972         transport->tcp_copied = 4;
 973         dprintk("RPC:       reading reply for XID %08x\n",
 974                         ntohl(transport->tcp_xid));
 975         xs_tcp_check_fraghdr(transport);
 976 }
 977
 978 static inline void xs_tcp_read_request(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
 979 {
 980         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 981         struct rpc_rqst *req;
 982         struct xdr_buf *rcvbuf;
 983         size_t len;
 984         ssize_t r;
 985
 986         /* Find and lock the request corresponding to this xid */
 987         spin_lock(&xprt->transport_lock);
 988         req = xprt_lookup_rqst(xprt, transport->tcp_xid);
 989         if (!req) {
 990                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
 991                 dprintk("RPC:       XID %08x request not found!\n",
 992                                 ntohl(transport->tcp_xid));
 993                 spin_unlock(&xprt->transport_lock);
 994                 return;
 995         }
 996
 997         rcvbuf = &req->rq_private_buf;
 998         len = desc->count;
 999         if (len > transport->tcp_reclen - transport->tcp_offset) {
1000                 struct xdr_skb_reader my_desc;
1001
1002                 len = transport->tcp_reclen - transport->tcp_offset;
1003                 memcpy(&my_desc, desc, sizeof(my_desc));
1004                 my_desc.count = len;
1005                 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
1006                                           &my_desc, xdr_skb_read_bits);
1007                 desc->count -= r;
1008                 desc->offset += r;
1009         } else
1010                 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
1011                                           desc, xdr_skb_read_bits);
1012
1013         if (r > 0) {
1014                 transport->tcp_copied += r;
1015                 transport->tcp_offset += r;
1016         }
1017         if (r != len) {
1018                 /* Error when copying to the receive buffer,
1019                  * usually because we weren't able to allocate
1020                  * additional buffer pages. All we can do now
1021                  * is turn off TCP_RCV_COPY_DATA, so the request
1022                  * will not receive any additional updates,
1023                  * and time out.
1024                  * Any remaining data from this record will
1025                  * be discarded.
1026                  */
1027                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1028                 dprintk("RPC:       XID %08x truncated request\n",
1029                                 ntohl(transport->tcp_xid));
1030                 dprintk("RPC:       xprt = %p, tcp_copied = %lu, "
1031                                 "tcp_offset = %u, tcp_reclen = %u\n",
1032                                 xprt, transport->tcp_copied,
1033                                 transport->tcp_offset, transport->tcp_reclen);
1034                 goto out;
1035         }
1036
1037         dprintk("RPC:       XID %08x read %Zd bytes\n",
1038                         ntohl(transport->tcp_xid), r);
1039         dprintk("RPC:       xprt = %p, tcp_copied = %lu, tcp_offset = %u, "
1040                         "tcp_reclen = %u\n", xprt, transport->tcp_copied,
1041                         transport->tcp_offset, transport->tcp_reclen);
1042
1043         if (transport->tcp_copied == req->rq_private_buf.buflen)
1044                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1045         else if (transport->tcp_offset == transport->tcp_reclen) {
1046                 if (transport->tcp_flags & TCP_RCV_LAST_FRAG)
1047                         transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1048         }
1049
1050 out:
1051         if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1052                 xprt_complete_rqst(req->rq_task, transport->tcp_copied);
1053         spin_unlock(&xprt->transport_lock);
1054         xs_tcp_check_fraghdr(transport);
1055 }
1056
1057 static inline void xs_tcp_read_discard(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1058 {
1059         size_t len;
1060
1061         len = transport->tcp_reclen - transport->tcp_offset;
1062         if (len > desc->count)
1063                 len = desc->count;
1064         desc->count -= len;
1065         desc->offset += len;
1066         transport->tcp_offset += len;
1067         dprintk("RPC:       discarded %Zu bytes\n", len);
1068         xs_tcp_check_fraghdr(transport);
1069 }
1070
1071 static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
1072 {
1073         struct rpc_xprt *xprt = rd_desc->arg.data;
1074         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1075         struct xdr_skb_reader desc = {
1076                 .skb    = skb,
1077                 .offset = offset,
1078                 .count  = len,
1079         };
1080
1081         dprintk("RPC:       xs_tcp_data_recv started\n");
1082         do {
1083                 /* Read in a new fragment marker if necessary */
1084                 /* Can we ever really expect to get completely empty fragments? */
1085                 if (transport->tcp_flags & TCP_RCV_COPY_FRAGHDR) {
1086                         xs_tcp_read_fraghdr(xprt, &desc);
1087                         continue;
1088                 }
1089                 /* Read in the xid if necessary */
1090                 if (transport->tcp_flags & TCP_RCV_COPY_XID) {
1091                         xs_tcp_read_xid(transport, &desc);
1092                         continue;
1093                 }
1094                 /* Read in the request data */
1095                 if (transport->tcp_flags & TCP_RCV_COPY_DATA) {
1096                         xs_tcp_read_request(xprt, &desc);
1097                         continue;
1098                 }
1099                 /* Skip over any trailing bytes on short reads */
1100                 xs_tcp_read_discard(transport, &desc);
1101         } while (desc.count);
1102         dprintk("RPC:       xs_tcp_data_recv done\n");
1103         return len - desc.count;
1104 }
1105
1106 /**
1107  * xs_tcp_data_ready - "data ready" callback for TCP sockets
1108  * @sk: socket with data to read
1109  * @bytes: how much data to read
1110  *
1111  */
1112 static void xs_tcp_data_ready(struct sock *sk, int bytes)
1113 {
1114         struct rpc_xprt *xprt;
1115         read_descriptor_t rd_desc;
1116         int read;
1117
1118         dprintk("RPC:       xs_tcp_data_ready...\n");
1119
1120         read_lock(&sk->sk_callback_lock);
1121         if (!(xprt = xprt_from_sock(sk)))
1122                 goto out;
1123         if (xprt->shutdown)
1124                 goto out;
1125
1126         /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
1127         rd_desc.arg.data = xprt;
1128         do {
1129                 rd_desc.count = 65536;
1130                 read = tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
1131         } while (read > 0);
1132 out:
1133         read_unlock(&sk->sk_callback_lock);
1134 }
1135
1136 /**
1137  * xs_tcp_state_change - callback to handle TCP socket state changes
1138  * @sk: socket whose state has changed
1139  *
1140  */
1141 static void xs_tcp_state_change(struct sock *sk)
1142 {
1143         struct rpc_xprt *xprt;
1144
1145         read_lock(&sk->sk_callback_lock);
1146         if (!(xprt = xprt_from_sock(sk)))
1147                 goto out;
1148         dprintk("RPC:       xs_tcp_state_change client %p...\n", xprt);
1149         dprintk("RPC:       state %x conn %d dead %d zapped %d\n",
1150                         sk->sk_state, xprt_connected(xprt),
1151                         sock_flag(sk, SOCK_DEAD),
1152                         sock_flag(sk, SOCK_ZAPPED));
1153
1154         switch (sk->sk_state) {
1155         case TCP_ESTABLISHED:
1156                 spin_lock_bh(&xprt->transport_lock);
1157                 if (!xprt_test_and_set_connected(xprt)) {
1158                         struct sock_xprt *transport = container_of(xprt,
1159                                         struct sock_xprt, xprt);
1160
1161                         /* Reset TCP record info */
1162                         transport->tcp_offset = 0;
1163                         transport->tcp_reclen = 0;
1164                         transport->tcp_copied = 0;
1165                         transport->tcp_flags =
1166                                 TCP_RCV_COPY_FRAGHDR | TCP_RCV_COPY_XID;
1167
1168                         xprt_wake_pending_tasks(xprt, -EAGAIN);
1169                 }
1170                 spin_unlock_bh(&xprt->transport_lock);
1171                 break;
1172         case TCP_FIN_WAIT1:
1173                 /* The client initiated a shutdown of the socket */
1174                 xprt->connect_cookie++;
1175                 xprt->reestablish_timeout = 0;
1176                 set_bit(XPRT_CLOSING, &xprt->state);
1177                 smp_mb__before_clear_bit();
1178                 clear_bit(XPRT_CONNECTED, &xprt->state);
1179                 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1180                 smp_mb__after_clear_bit();
1181                 break;
1182         case TCP_CLOSE_WAIT:
1183                 /* The server initiated a shutdown of the socket */
1184                 xprt_force_disconnect(xprt);
1185         case TCP_SYN_SENT:
1186                 xprt->connect_cookie++;
1187         case TCP_CLOSING:
1188                 /*
1189                  * If the server closed down the connection, make sure that
1190                  * we back off before reconnecting
1191                  */
1192                 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1193                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1194                 break;
1195         case TCP_LAST_ACK:
1196                 set_bit(XPRT_CLOSING, &xprt->state);
1197                 smp_mb__before_clear_bit();
1198                 clear_bit(XPRT_CONNECTED, &xprt->state);
1199                 smp_mb__after_clear_bit();
1200                 break;
1201         case TCP_CLOSE:
1202                 smp_mb__before_clear_bit();
1203                 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1204                 clear_bit(XPRT_CLOSING, &xprt->state);
1205                 smp_mb__after_clear_bit();
1206                 /* Mark transport as closed and wake up all pending tasks */
1207                 xprt_disconnect_done(xprt);
1208         }
1209  out:
1210         read_unlock(&sk->sk_callback_lock);
1211 }
1212
1213 /**
1214  * xs_error_report - callback mainly for catching socket errors
1215  * @sk: socket
1216  */
1217 static void xs_error_report(struct sock *sk)
1218 {
1219         struct rpc_xprt *xprt;
1220
1221         read_lock(&sk->sk_callback_lock);
1222         if (!(xprt = xprt_from_sock(sk)))
1223                 goto out;
1224         dprintk("RPC:       %s client %p...\n"
1225                         "RPC:       error %d\n",
1226                         __func__, xprt, sk->sk_err);
1227         xprt_wake_pending_tasks(xprt, -EAGAIN);
1228 out:
1229         read_unlock(&sk->sk_callback_lock);
1230 }
1231
1232 /**
1233  * xs_udp_write_space - callback invoked when socket buffer space
1234  *                             becomes available
1235  * @sk: socket whose state has changed
1236  *
1237  * Called when more output buffer space is available for this socket.
1238  * We try not to wake our writers until they can make "significant"
1239  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1240  * with a bunch of small requests.
1241  */
1242 static void xs_udp_write_space(struct sock *sk)
1243 {
1244         read_lock(&sk->sk_callback_lock);
1245
1246         /* from net/core/sock.c:sock_def_write_space */
1247         if (sock_writeable(sk)) {
1248                 struct socket *sock;
1249                 struct rpc_xprt *xprt;
1250
1251                 if (unlikely(!(sock = sk->sk_socket)))
1252                         goto out;
1253                 clear_bit(SOCK_NOSPACE, &sock->flags);
1254
1255                 if (unlikely(!(xprt = xprt_from_sock(sk))))
1256                         goto out;
1257                 if (test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags) == 0)
1258                         goto out;
1259
1260                 xprt_write_space(xprt);
1261         }
1262
1263  out:
1264         read_unlock(&sk->sk_callback_lock);
1265 }
1266
1267 /**
1268  * xs_tcp_write_space - callback invoked when socket buffer space
1269  *                             becomes available
1270  * @sk: socket whose state has changed
1271  *
1272  * Called when more output buffer space is available for this socket.
1273  * We try not to wake our writers until they can make "significant"
1274  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1275  * with a bunch of small requests.
1276  */
1277 static void xs_tcp_write_space(struct sock *sk)
1278 {
1279         read_lock(&sk->sk_callback_lock);
1280
1281         /* from net/core/stream.c:sk_stream_write_space */
1282         if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk)) {
1283                 struct socket *sock;
1284                 struct rpc_xprt *xprt;
1285
1286                 if (unlikely(!(sock = sk->sk_socket)))
1287                         goto out;
1288                 clear_bit(SOCK_NOSPACE, &sock->flags);
1289
1290                 if (unlikely(!(xprt = xprt_from_sock(sk))))
1291                         goto out;
1292                 if (test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags) == 0)
1293                         goto out;
1294
1295                 xprt_write_space(xprt);
1296         }
1297
1298  out:
1299         read_unlock(&sk->sk_callback_lock);
1300 }
1301
1302 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
1303 {
1304         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1305         struct sock *sk = transport->inet;
1306
1307         if (transport->rcvsize) {
1308                 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1309                 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
1310         }
1311         if (transport->sndsize) {
1312                 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1313                 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1314                 sk->sk_write_space(sk);
1315         }
1316 }
1317
1318 /**
1319  * xs_udp_set_buffer_size - set send and receive limits
1320  * @xprt: generic transport
1321  * @sndsize: requested size of send buffer, in bytes
1322  * @rcvsize: requested size of receive buffer, in bytes
1323  *
1324  * Set socket send and receive buffer size limits.
1325  */
1326 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1327 {
1328         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1329
1330         transport->sndsize = 0;
1331         if (sndsize)
1332                 transport->sndsize = sndsize + 1024;
1333         transport->rcvsize = 0;
1334         if (rcvsize)
1335                 transport->rcvsize = rcvsize + 1024;
1336
1337         xs_udp_do_set_buffer_size(xprt);
1338 }
1339
1340 /**
1341  * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1342  * @task: task that timed out
1343  *
1344  * Adjust the congestion window after a retransmit timeout has occurred.
1345  */
1346 static void xs_udp_timer(struct rpc_task *task)
1347 {
1348         xprt_adjust_cwnd(task, -ETIMEDOUT);
1349 }
1350
1351 static unsigned short xs_get_random_port(void)
1352 {
1353         unsigned short range = xprt_max_resvport - xprt_min_resvport;
1354         unsigned short rand = (unsigned short) net_random() % range;
1355         return rand + xprt_min_resvport;
1356 }
1357
1358 /**
1359  * xs_set_port - reset the port number in the remote endpoint address
1360  * @xprt: generic transport
1361  * @port: new port number
1362  *
1363  */
1364 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1365 {
1366         struct sockaddr *addr = xs_addr(xprt);
1367
1368         dprintk("RPC:       setting port for xprt %p to %u\n", xprt, port);
1369
1370         switch (addr->sa_family) {
1371         case AF_INET:
1372                 ((struct sockaddr_in *)addr)->sin_port = htons(port);
1373                 break;
1374         case AF_INET6:
1375                 ((struct sockaddr_in6 *)addr)->sin6_port = htons(port);
1376                 break;
1377         default:
1378                 BUG();
1379         }
1380 }
1381
1382 static unsigned short xs_get_srcport(struct sock_xprt *transport, struct socket *sock)
1383 {
1384         unsigned short port = transport->port;
1385
1386         if (port == 0 && transport->xprt.resvport)
1387                 port = xs_get_random_port();
1388         return port;
1389 }
1390
1391 static unsigned short xs_next_srcport(struct sock_xprt *transport, struct socket *sock, unsigned short port)
1392 {
1393         if (transport->port != 0)
1394                 transport->port = 0;
1395         if (!transport->xprt.resvport)
1396                 return 0;
1397         if (port <= xprt_min_resvport || port > xprt_max_resvport)
1398                 return xprt_max_resvport;
1399         return --port;
1400 }
1401
1402 static int xs_bind4(struct sock_xprt *transport, struct socket *sock)
1403 {
1404         struct sockaddr_in myaddr = {
1405                 .sin_family = AF_INET,
1406         };
1407         struct sockaddr_in *sa;
1408         int err, nloop = 0;
1409         unsigned short port = xs_get_srcport(transport, sock);
1410         unsigned short last;
1411
1412         sa = (struct sockaddr_in *)&transport->addr;
1413         myaddr.sin_addr = sa->sin_addr;
1414         do {
1415                 myaddr.sin_port = htons(port);
1416                 err = kernel_bind(sock, (struct sockaddr *) &myaddr,
1417                                                 sizeof(myaddr));
1418                 if (port == 0)
1419                         break;
1420                 if (err == 0) {
1421                         transport->port = port;
1422                         break;
1423                 }
1424                 last = port;
1425                 port = xs_next_srcport(transport, sock, port);
1426                 if (port > last)
1427                         nloop++;
1428         } while (err == -EADDRINUSE && nloop != 2);
1429         dprintk("RPC:       %s %pI4:%u: %s (%d)\n",
1430                         __func__, &myaddr.sin_addr,
1431                         port, err ? "failed" : "ok", err);
1432         return err;
1433 }
1434
1435 static int xs_bind6(struct sock_xprt *transport, struct socket *sock)
1436 {
1437         struct sockaddr_in6 myaddr = {
1438                 .sin6_family = AF_INET6,
1439         };
1440         struct sockaddr_in6 *sa;
1441         int err, nloop = 0;
1442         unsigned short port = xs_get_srcport(transport, sock);
1443         unsigned short last;
1444
1445         sa = (struct sockaddr_in6 *)&transport->addr;
1446         myaddr.sin6_addr = sa->sin6_addr;
1447         do {
1448                 myaddr.sin6_port = htons(port);
1449                 err = kernel_bind(sock, (struct sockaddr *) &myaddr,
1450                                                 sizeof(myaddr));
1451                 if (port == 0)
1452                         break;
1453                 if (err == 0) {
1454                         transport->port = port;
1455                         break;
1456                 }
1457                 last = port;
1458                 port = xs_next_srcport(transport, sock, port);
1459                 if (port > last)
1460                         nloop++;
1461         } while (err == -EADDRINUSE && nloop != 2);
1462         dprintk("RPC:       xs_bind6 %pI6:%u: %s (%d)\n",
1463                 &myaddr.sin6_addr, port, err ? "failed" : "ok", err);
1464         return err;
1465 }
1466
1467 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1468 static struct lock_class_key xs_key[2];
1469 static struct lock_class_key xs_slock_key[2];
1470
1471 static inline void xs_reclassify_socket4(struct socket *sock)
1472 {
1473         struct sock *sk = sock->sk;
1474
1475         BUG_ON(sock_owned_by_user(sk));
1476         sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1477                 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1478 }
1479
1480 static inline void xs_reclassify_socket6(struct socket *sock)
1481 {
1482         struct sock *sk = sock->sk;
1483
1484         BUG_ON(sock_owned_by_user(sk));
1485         sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1486                 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
1487 }
1488 #else
1489 static inline void xs_reclassify_socket4(struct socket *sock)
1490 {
1491 }
1492
1493 static inline void xs_reclassify_socket6(struct socket *sock)
1494 {
1495 }
1496 #endif
1497
1498 static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
1499 {
1500         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1501
1502         if (!transport->inet) {
1503                 struct sock *sk = sock->sk;
1504
1505                 write_lock_bh(&sk->sk_callback_lock);
1506
1507                 xs_save_old_callbacks(transport, sk);
1508
1509                 sk->sk_user_data = xprt;
1510                 sk->sk_data_ready = xs_udp_data_ready;
1511                 sk->sk_write_space = xs_udp_write_space;
1512                 sk->sk_error_report = xs_error_report;
1513                 sk->sk_no_check = UDP_CSUM_NORCV;
1514                 sk->sk_allocation = GFP_ATOMIC;
1515
1516                 xprt_set_connected(xprt);
1517
1518                 /* Reset to new socket */
1519                 transport->sock = sock;
1520                 transport->inet = sk;
1521
1522                 write_unlock_bh(&sk->sk_callback_lock);
1523         }
1524         xs_udp_do_set_buffer_size(xprt);
1525 }
1526
1527 /**
1528  * xs_udp_connect_worker4 - set up a UDP socket
1529  * @work: RPC transport to connect
1530  *
1531  * Invoked by a work queue tasklet.
1532  */
1533 static void xs_udp_connect_worker4(struct work_struct *work)
1534 {
1535         struct sock_xprt *transport =
1536                 container_of(work, struct sock_xprt, connect_worker.work);
1537         struct rpc_xprt *xprt = &transport->xprt;
1538         struct socket *sock = transport->sock;
1539         int err, status = -EIO;
1540
1541         if (xprt->shutdown)
1542                 goto out;
1543
1544         /* Start by resetting any existing state */
1545         xs_reset_transport(transport);
1546
1547         err = sock_create_kern(PF_INET, SOCK_DGRAM, IPPROTO_UDP, &sock);
1548         if (err < 0) {
1549                 dprintk("RPC:       can't create UDP transport socket (%d).\n", -err);
1550                 goto out;
1551         }
1552         xs_reclassify_socket4(sock);
1553
1554         if (xs_bind4(transport, sock)) {
1555                 sock_release(sock);
1556                 goto out;
1557         }
1558
1559         dprintk("RPC:       worker connecting xprt %p to address: %s\n",
1560                         xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1561
1562         xs_udp_finish_connecting(xprt, sock);
1563         status = 0;
1564 out:
1565         xprt_wake_pending_tasks(xprt, status);
1566         xprt_clear_connecting(xprt);
1567 }
1568
1569 /**
1570  * xs_udp_connect_worker6 - set up a UDP socket
1571  * @work: RPC transport to connect
1572  *
1573  * Invoked by a work queue tasklet.
1574  */
1575 static void xs_udp_connect_worker6(struct work_struct *work)
1576 {
1577         struct sock_xprt *transport =
1578                 container_of(work, struct sock_xprt, connect_worker.work);
1579         struct rpc_xprt *xprt = &transport->xprt;
1580         struct socket *sock = transport->sock;
1581         int err, status = -EIO;
1582
1583         if (xprt->shutdown)
1584                 goto out;
1585
1586         /* Start by resetting any existing state */
1587         xs_reset_transport(transport);
1588
1589         err = sock_create_kern(PF_INET6, SOCK_DGRAM, IPPROTO_UDP, &sock);
1590         if (err < 0) {
1591                 dprintk("RPC:       can't create UDP transport socket (%d).\n", -err);
1592                 goto out;
1593         }
1594         xs_reclassify_socket6(sock);
1595
1596         if (xs_bind6(transport, sock) < 0) {
1597                 sock_release(sock);
1598                 goto out;
1599         }
1600
1601         dprintk("RPC:       worker connecting xprt %p to address: %s\n",
1602                         xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1603
1604         xs_udp_finish_connecting(xprt, sock);
1605         status = 0;
1606 out:
1607         xprt_wake_pending_tasks(xprt, status);
1608         xprt_clear_connecting(xprt);
1609 }
1610
1611 /*
1612  * We need to preserve the port number so the reply cache on the server can
1613  * find our cached RPC replies when we get around to reconnecting.
1614  */
1615 static void xs_abort_connection(struct rpc_xprt *xprt, struct sock_xprt *transport)
1616 {
1617         int result;
1618         struct sockaddr any;
1619
1620         dprintk("RPC:       disconnecting xprt %p to reuse port\n", xprt);
1621
1622         /*
1623          * Disconnect the transport socket by doing a connect operation
1624          * with AF_UNSPEC.  This should return immediately...
1625          */
1626         memset(&any, 0, sizeof(any));
1627         any.sa_family = AF_UNSPEC;
1628         result = kernel_connect(transport->sock, &any, sizeof(any), 0);
1629         if (result)
1630                 dprintk("RPC:       AF_UNSPEC connect return code %d\n",
1631                                 result);
1632 }
1633
1634 static void xs_tcp_reuse_connection(struct rpc_xprt *xprt, struct sock_xprt *transport)
1635 {
1636         unsigned int state = transport->inet->sk_state;
1637
1638         if (state == TCP_CLOSE && transport->sock->state == SS_UNCONNECTED)
1639                 return;
1640         if ((1 << state) & (TCPF_ESTABLISHED|TCPF_SYN_SENT))
1641                 return;
1642         xs_abort_connection(xprt, transport);
1643 }
1644
1645 static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
1646 {
1647         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1648
1649         if (!transport->inet) {
1650                 struct sock *sk = sock->sk;
1651
1652                 write_lock_bh(&sk->sk_callback_lock);
1653
1654                 xs_save_old_callbacks(transport, sk);
1655
1656                 sk->sk_user_data = xprt;
1657                 sk->sk_data_ready = xs_tcp_data_ready;
1658                 sk->sk_state_change = xs_tcp_state_change;
1659                 sk->sk_write_space = xs_tcp_write_space;
1660                 sk->sk_error_report = xs_error_report;
1661                 sk->sk_allocation = GFP_ATOMIC;
1662
1663                 /* socket options */
1664                 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
1665                 sock_reset_flag(sk, SOCK_LINGER);
1666                 tcp_sk(sk)->linger2 = 0;
1667                 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
1668
1669                 xprt_clear_connected(xprt);
1670
1671                 /* Reset to new socket */
1672                 transport->sock = sock;
1673                 transport->inet = sk;
1674
1675                 write_unlock_bh(&sk->sk_callback_lock);
1676         }
1677
1678         if (!xprt_bound(xprt))
1679                 return -ENOTCONN;
1680
1681         /* Tell the socket layer to start connecting... */
1682         xprt->stat.connect_count++;
1683         xprt->stat.connect_start = jiffies;
1684         return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
1685 }
1686
1687 /**
1688  * xs_tcp_connect_worker4 - connect a TCP socket to a remote endpoint
1689  * @work: RPC transport to connect
1690  *
1691  * Invoked by a work queue tasklet.
1692  */
1693 static void xs_tcp_connect_worker4(struct work_struct *work)
1694 {
1695         struct sock_xprt *transport =
1696                 container_of(work, struct sock_xprt, connect_worker.work);
1697         struct rpc_xprt *xprt = &transport->xprt;
1698         struct socket *sock = transport->sock;
1699         int err, status = -EIO;
1700
1701         if (xprt->shutdown)
1702                 goto out;
1703
1704         if (!sock) {
1705                 /* start from scratch */
1706                 if ((err = sock_create_kern(PF_INET, SOCK_STREAM, IPPROTO_TCP, &sock)) < 0) {
1707                         dprintk("RPC:       can't create TCP transport socket (%d).\n", -err);
1708                         goto out;
1709                 }
1710                 xs_reclassify_socket4(sock);
1711
1712                 if (xs_bind4(transport, sock) < 0) {
1713                         sock_release(sock);
1714                         goto out;
1715                 }
1716         } else
1717                 /* "close" the socket, preserving the local port */
1718                 xs_tcp_reuse_connection(xprt, transport);
1719
1720         dprintk("RPC:       worker connecting xprt %p to address: %s\n",
1721                         xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1722
1723         status = xs_tcp_finish_connecting(xprt, sock);
1724         dprintk("RPC:       %p connect status %d connected %d sock state %d\n",
1725                         xprt, -status, xprt_connected(xprt),
1726                         sock->sk->sk_state);
1727         switch (status) {
1728         case -ECONNREFUSED:
1729         case -ECONNRESET:
1730         case -ENETUNREACH:
1731                 /* retry with existing socket, after a delay */
1732         case 0:
1733         case -EINPROGRESS:
1734         case -EALREADY:
1735                 goto out_clear;
1736         }
1737         /* get rid of existing socket, and retry */
1738         xs_tcp_shutdown(xprt);
1739         printk("%s: connect returned unhandled error %d\n",
1740                         __func__, status);
1741         status = -EAGAIN;
1742 out:
1743         xprt_wake_pending_tasks(xprt, status);
1744 out_clear:
1745         xprt_clear_connecting(xprt);
1746 }
1747
1748 /**
1749  * xs_tcp_connect_worker6 - connect a TCP socket to a remote endpoint
1750  * @work: RPC transport to connect
1751  *
1752  * Invoked by a work queue tasklet.
1753  */
1754 static void xs_tcp_connect_worker6(struct work_struct *work)
1755 {
1756         struct sock_xprt *transport =
1757                 container_of(work, struct sock_xprt, connect_worker.work);
1758         struct rpc_xprt *xprt = &transport->xprt;
1759         struct socket *sock = transport->sock;
1760         int err, status = -EIO;
1761
1762         if (xprt->shutdown)
1763                 goto out;
1764
1765         if (!sock) {
1766                 /* start from scratch */
1767                 if ((err = sock_create_kern(PF_INET6, SOCK_STREAM, IPPROTO_TCP, &sock)) < 0) {
1768                         dprintk("RPC:       can't create TCP transport socket (%d).\n", -err);
1769                         goto out;
1770                 }
1771                 xs_reclassify_socket6(sock);
1772
1773                 if (xs_bind6(transport, sock) < 0) {
1774                         sock_release(sock);
1775                         goto out;
1776                 }
1777         } else
1778                 /* "close" the socket, preserving the local port */
1779                 xs_tcp_reuse_connection(xprt, transport);
1780
1781         dprintk("RPC:       worker connecting xprt %p to address: %s\n",
1782                         xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1783
1784         status = xs_tcp_finish_connecting(xprt, sock);
1785         dprintk("RPC:       %p connect status %d connected %d sock state %d\n",
1786                         xprt, -status, xprt_connected(xprt), sock->sk->sk_state);
1787         switch (status) {
1788         case -ECONNREFUSED:
1789         case -ECONNRESET:
1790         case -ENETUNREACH:
1791                 /* retry with existing socket, after a delay */
1792         case 0:
1793         case -EINPROGRESS:
1794         case -EALREADY:
1795                 goto out_clear;
1796         }
1797         /* get rid of existing socket, and retry */
1798         xs_tcp_shutdown(xprt);
1799         printk("%s: connect returned unhandled error %d\n",
1800                         __func__, status);
1801         status = -EAGAIN;
1802 out:
1803         xprt_wake_pending_tasks(xprt, status);
1804 out_clear:
1805         xprt_clear_connecting(xprt);
1806 }
1807
1808 /**
1809  * xs_connect - connect a socket to a remote endpoint
1810  * @task: address of RPC task that manages state of connect request
1811  *
1812  * TCP: If the remote end dropped the connection, delay reconnecting.
1813  *
1814  * UDP socket connects are synchronous, but we use a work queue anyway
1815  * to guarantee that even unprivileged user processes can set up a
1816  * socket on a privileged port.
1817  *
1818  * If a UDP socket connect fails, the delay behavior here prevents
1819  * retry floods (hard mounts).
1820  */
1821 static void xs_connect(struct rpc_task *task)
1822 {
1823         struct rpc_xprt *xprt = task->tk_xprt;
1824         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1825
1826         if (xprt_test_and_set_connecting(xprt))
1827                 return;
1828
1829         if (transport->sock != NULL) {
1830                 dprintk("RPC:       xs_connect delayed xprt %p for %lu "
1831                                 "seconds\n",
1832                                 xprt, xprt->reestablish_timeout / HZ);
1833                 queue_delayed_work(rpciod_workqueue,
1834                                    &transport->connect_worker,
1835                                    xprt->reestablish_timeout);
1836                 xprt->reestablish_timeout <<= 1;
1837                 if (xprt->reestablish_timeout > XS_TCP_MAX_REEST_TO)
1838                         xprt->reestablish_timeout = XS_TCP_MAX_REEST_TO;
1839         } else {
1840                 dprintk("RPC:       xs_connect scheduled xprt %p\n", xprt);
1841                 queue_delayed_work(rpciod_workqueue,
1842                                    &transport->connect_worker, 0);
1843         }
1844 }
1845
1846 static void xs_tcp_connect(struct rpc_task *task)
1847 {
1848         struct rpc_xprt *xprt = task->tk_xprt;
1849
1850         /* Exit if we need to wait for socket shutdown to complete */
1851         if (test_bit(XPRT_CLOSING, &xprt->state))
1852                 return;
1853         xs_connect(task);
1854 }
1855
1856 /**
1857  * xs_udp_print_stats - display UDP socket-specifc stats
1858  * @xprt: rpc_xprt struct containing statistics
1859  * @seq: output file
1860  *
1861  */
1862 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
1863 {
1864         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1865
1866         seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %Lu %Lu\n",
1867                         transport->port,
1868                         xprt->stat.bind_count,
1869                         xprt->stat.sends,
1870                         xprt->stat.recvs,
1871                         xprt->stat.bad_xids,
1872                         xprt->stat.req_u,
1873                         xprt->stat.bklog_u);
1874 }
1875
1876 /**
1877  * xs_tcp_print_stats - display TCP socket-specifc stats
1878  * @xprt: rpc_xprt struct containing statistics
1879  * @seq: output file
1880  *
1881  */
1882 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
1883 {
1884         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1885         long idle_time = 0;
1886
1887         if (xprt_connected(xprt))
1888                 idle_time = (long)(jiffies - xprt->last_used) / HZ;
1889
1890         seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu %Lu %Lu\n",
1891                         transport->port,
1892                         xprt->stat.bind_count,
1893                         xprt->stat.connect_count,
1894                         xprt->stat.connect_time,
1895                         idle_time,
1896                         xprt->stat.sends,
1897                         xprt->stat.recvs,
1898                         xprt->stat.bad_xids,
1899                         xprt->stat.req_u,
1900                         xprt->stat.bklog_u);
1901 }
1902
1903 static struct rpc_xprt_ops xs_udp_ops = {
1904         .set_buffer_size        = xs_udp_set_buffer_size,
1905         .reserve_xprt           = xprt_reserve_xprt_cong,
1906         .release_xprt           = xprt_release_xprt_cong,
1907         .rpcbind                = rpcb_getport_async,
1908         .set_port               = xs_set_port,
1909         .connect                = xs_connect,
1910         .buf_alloc              = rpc_malloc,
1911         .buf_free               = rpc_free,
1912         .send_request           = xs_udp_send_request,
1913         .set_retrans_timeout    = xprt_set_retrans_timeout_rtt,
1914         .timer                  = xs_udp_timer,
1915         .release_request        = xprt_release_rqst_cong,
1916         .close                  = xs_close,
1917         .destroy                = xs_destroy,
1918         .print_stats            = xs_udp_print_stats,
1919 };
1920
1921 static struct rpc_xprt_ops xs_tcp_ops = {
1922         .reserve_xprt           = xprt_reserve_xprt,
1923         .release_xprt           = xs_tcp_release_xprt,
1924         .rpcbind                = rpcb_getport_async,
1925         .set_port               = xs_set_port,
1926         .connect                = xs_tcp_connect,
1927         .buf_alloc              = rpc_malloc,
1928         .buf_free               = rpc_free,
1929         .send_request           = xs_tcp_send_request,
1930         .set_retrans_timeout    = xprt_set_retrans_timeout_def,
1931         .close                  = xs_tcp_shutdown,
1932         .destroy                = xs_destroy,
1933         .print_stats            = xs_tcp_print_stats,
1934 };
1935
1936 static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
1937                                       unsigned int slot_table_size)
1938 {
1939         struct rpc_xprt *xprt;
1940         struct sock_xprt *new;
1941
1942         if (args->addrlen > sizeof(xprt->addr)) {
1943                 dprintk("RPC:       xs_setup_xprt: address too large\n");
1944                 return ERR_PTR(-EBADF);
1945         }
1946
1947         new = kzalloc(sizeof(*new), GFP_KERNEL);
1948         if (new == NULL) {
1949                 dprintk("RPC:       xs_setup_xprt: couldn't allocate "
1950                                 "rpc_xprt\n");
1951                 return ERR_PTR(-ENOMEM);
1952         }
1953         xprt = &new->xprt;
1954
1955         xprt->max_reqs = slot_table_size;
1956         xprt->slot = kcalloc(xprt->max_reqs, sizeof(struct rpc_rqst), GFP_KERNEL);
1957         if (xprt->slot == NULL) {
1958                 kfree(xprt);
1959                 dprintk("RPC:       xs_setup_xprt: couldn't allocate slot "
1960                                 "table\n");
1961                 return ERR_PTR(-ENOMEM);
1962         }
1963
1964         memcpy(&xprt->addr, args->dstaddr, args->addrlen);
1965         xprt->addrlen = args->addrlen;
1966         if (args->srcaddr)
1967                 memcpy(&new->addr, args->srcaddr, args->addrlen);
1968
1969         return xprt;
1970 }
1971
1972 static const struct rpc_timeout xs_udp_default_timeout = {
1973         .to_initval = 5 * HZ,
1974         .to_maxval = 30 * HZ,
1975         .to_increment = 5 * HZ,
1976         .to_retries = 5,
1977 };
1978
1979 /**
1980  * xs_setup_udp - Set up transport to use a UDP socket
1981  * @args: rpc transport creation arguments
1982  *
1983  */
1984 static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
1985 {
1986         struct sockaddr *addr = args->dstaddr;
1987         struct rpc_xprt *xprt;
1988         struct sock_xprt *transport;
1989
1990         xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries);
1991         if (IS_ERR(xprt))
1992                 return xprt;
1993         transport = container_of(xprt, struct sock_xprt, xprt);
1994
1995         xprt->prot = IPPROTO_UDP;
1996         xprt->tsh_size = 0;
1997         /* XXX: header size can vary due to auth type, IPv6, etc. */
1998         xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
1999
2000         xprt->bind_timeout = XS_BIND_TO;
2001         xprt->connect_timeout = XS_UDP_CONN_TO;
2002         xprt->reestablish_timeout = XS_UDP_REEST_TO;
2003         xprt->idle_timeout = XS_IDLE_DISC_TO;
2004
2005         xprt->ops = &xs_udp_ops;
2006
2007         xprt->timeout = &xs_udp_default_timeout;
2008
2009         switch (addr->sa_family) {
2010         case AF_INET:
2011                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2012                         xprt_set_bound(xprt);
2013
2014                 INIT_DELAYED_WORK(&transport->connect_worker,
2015                                         xs_udp_connect_worker4);
2016                 xs_format_ipv4_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
2017                 break;
2018         case AF_INET6:
2019                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2020                         xprt_set_bound(xprt);
2021
2022                 INIT_DELAYED_WORK(&transport->connect_worker,
2023                                         xs_udp_connect_worker6);
2024                 xs_format_ipv6_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
2025                 break;
2026         default:
2027                 kfree(xprt);
2028                 return ERR_PTR(-EAFNOSUPPORT);
2029         }
2030
2031         dprintk("RPC:       set up transport to address %s\n",
2032                         xprt->address_strings[RPC_DISPLAY_ALL]);
2033
2034         if (try_module_get(THIS_MODULE))
2035                 return xprt;
2036
2037         kfree(xprt->slot);
2038         kfree(xprt);
2039         return ERR_PTR(-EINVAL);
2040 }
2041
2042 static const struct rpc_timeout xs_tcp_default_timeout = {
2043         .to_initval = 60 * HZ,
2044         .to_maxval = 60 * HZ,
2045         .to_retries = 2,
2046 };
2047
2048 /**
2049  * xs_setup_tcp - Set up transport to use a TCP socket
2050  * @args: rpc transport creation arguments
2051  *
2052  */
2053 static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
2054 {
2055         struct sockaddr *addr = args->dstaddr;
2056         struct rpc_xprt *xprt;
2057         struct sock_xprt *transport;
2058
2059         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries);
2060         if (IS_ERR(xprt))
2061                 return xprt;
2062         transport = container_of(xprt, struct sock_xprt, xprt);
2063
2064         xprt->prot = IPPROTO_TCP;
2065         xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2066         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2067
2068         xprt->bind_timeout = XS_BIND_TO;
2069         xprt->connect_timeout = XS_TCP_CONN_TO;
2070         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2071         xprt->idle_timeout = XS_IDLE_DISC_TO;
2072
2073         xprt->ops = &xs_tcp_ops;
2074         xprt->timeout = &xs_tcp_default_timeout;
2075
2076         switch (addr->sa_family) {
2077         case AF_INET:
2078                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2079                         xprt_set_bound(xprt);
2080
2081                 INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_connect_worker4);
2082                 xs_format_ipv4_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
2083                 break;
2084         case AF_INET6:
2085                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2086                         xprt_set_bound(xprt);
2087
2088                 INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_connect_worker6);
2089                 xs_format_ipv6_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
2090                 break;
2091         default:
2092                 kfree(xprt);
2093                 return ERR_PTR(-EAFNOSUPPORT);
2094         }
2095
2096         dprintk("RPC:       set up transport to address %s\n",
2097                         xprt->address_strings[RPC_DISPLAY_ALL]);
2098
2099         if (try_module_get(THIS_MODULE))
2100                 return xprt;
2101
2102         kfree(xprt->slot);
2103         kfree(xprt);
2104         return ERR_PTR(-EINVAL);
2105 }
2106
2107 static struct xprt_class        xs_udp_transport = {
2108         .list           = LIST_HEAD_INIT(xs_udp_transport.list),
2109         .name           = "udp",
2110         .owner          = THIS_MODULE,
2111         .ident          = IPPROTO_UDP,
2112         .setup          = xs_setup_udp,
2113 };
2114
2115 static struct xprt_class        xs_tcp_transport = {
2116         .list           = LIST_HEAD_INIT(xs_tcp_transport.list),
2117         .name           = "tcp",
2118         .owner          = THIS_MODULE,
2119         .ident          = IPPROTO_TCP,
2120         .setup          = xs_setup_tcp,
2121 };
2122
2123 /**
2124  * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
2125  *
2126  */
2127 int init_socket_xprt(void)
2128 {
2129 #ifdef RPC_DEBUG
2130         if (!sunrpc_table_header)
2131                 sunrpc_table_header = register_sysctl_table(sunrpc_table);
2132 #endif
2133
2134         xprt_register_transport(&xs_udp_transport);
2135         xprt_register_transport(&xs_tcp_transport);
2136
2137         return 0;
2138 }
2139
2140 /**
2141  * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
2142  *
2143  */
2144 void cleanup_socket_xprt(void)
2145 {
2146 #ifdef RPC_DEBUG
2147         if (sunrpc_table_header) {
2148                 unregister_sysctl_table(sunrpc_table_header);
2149                 sunrpc_table_header = NULL;
2150         }
2151 #endif
2152
2153         xprt_unregister_transport(&xs_udp_transport);
2154         xprt_unregister_transport(&xs_tcp_transport);
2155 }