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

   1 /*
   2  * linux/net/sunrpc/xprtsock.c
   3  *
   4  * Client-side transport implementation for sockets.
   5  *
   6  * TCP callback races fixes (C) 1998 Red Hat
   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 #define XS_TCP_LINGER_TO        (15U * HZ)
  53
  54 /*
  55  * We can register our own files under /proc/sys/sunrpc by
  56  * calling register_sysctl_table() again.  The files in that
  57  * directory become the union of all files registered there.
  58  *
  59  * We simply need to make sure that we don't collide with
  60  * someone else's file names!
  61  */
  62
  63 #ifdef RPC_DEBUG
  64
  65 static unsigned int min_slot_table_size = RPC_MIN_SLOT_TABLE;
  66 static unsigned int max_slot_table_size = RPC_MAX_SLOT_TABLE;
  67 static unsigned int xprt_min_resvport_limit = RPC_MIN_RESVPORT;
  68 static unsigned int xprt_max_resvport_limit = RPC_MAX_RESVPORT;
  69
  70 static struct ctl_table_header *sunrpc_table_header;
  71
  72 /*
  73  * FIXME: changing the UDP slot table size should also resize the UDP
  74  *        socket buffers for existing UDP transports
  75  */
  76 static ctl_table xs_tunables_table[] = {
  77         {
  78                 .ctl_name       = CTL_SLOTTABLE_UDP,
  79                 .procname       = "udp_slot_table_entries",
  80                 .data           = &xprt_udp_slot_table_entries,
  81                 .maxlen         = sizeof(unsigned int),
  82                 .mode           = 0644,
  83                 .proc_handler   = &proc_dointvec_minmax,
  84                 .strategy       = &sysctl_intvec,
  85                 .extra1         = &min_slot_table_size,
  86                 .extra2         = &max_slot_table_size
  87         },
  88         {
  89                 .ctl_name       = CTL_SLOTTABLE_TCP,
  90                 .procname       = "tcp_slot_table_entries",
  91                 .data           = &xprt_tcp_slot_table_entries,
  92                 .maxlen         = sizeof(unsigned int),
  93                 .mode           = 0644,
  94                 .proc_handler   = &proc_dointvec_minmax,
  95                 .strategy       = &sysctl_intvec,
  96                 .extra1         = &min_slot_table_size,
  97                 .extra2         = &max_slot_table_size
  98         },
  99         {
 100                 .ctl_name       = CTL_MIN_RESVPORT,
 101                 .procname       = "min_resvport",
 102                 .data           = &xprt_min_resvport,
 103                 .maxlen         = sizeof(unsigned int),
 104                 .mode           = 0644,
 105                 .proc_handler   = &proc_dointvec_minmax,
 106                 .strategy       = &sysctl_intvec,
 107                 .extra1         = &xprt_min_resvport_limit,
 108                 .extra2         = &xprt_max_resvport_limit
 109         },
 110         {
 111                 .ctl_name       = CTL_MAX_RESVPORT,
 112                 .procname       = "max_resvport",
 113                 .data           = &xprt_max_resvport,
 114                 .maxlen         = sizeof(unsigned int),
 115                 .mode           = 0644,
 116                 .proc_handler   = &proc_dointvec_minmax,
 117                 .strategy       = &sysctl_intvec,
 118                 .extra1         = &xprt_min_resvport_limit,
 119                 .extra2         = &xprt_max_resvport_limit
 120         },
 121         {
 122                 .ctl_name = 0,
 123         },
 124 };
 125
 126 static ctl_table sunrpc_table[] = {
 127         {
 128                 .ctl_name       = CTL_SUNRPC,
 129                 .procname       = "sunrpc",
 130                 .mode           = 0555,
 131                 .child          = xs_tunables_table
 132         },
 133         {
 134                 .ctl_name = 0,
 135         },
 136 };
 137
 138 #endif
 139
 140 /*
 141  * Time out for an RPC UDP socket connect.  UDP socket connects are
 142  * synchronous, but we set a timeout anyway in case of resource
 143  * exhaustion on the local host.
 144  */
 145 #define XS_UDP_CONN_TO          (5U * HZ)
 146
 147 /*
 148  * Wait duration for an RPC TCP connection to be established.  Solaris
 149  * NFS over TCP uses 60 seconds, for example, which is in line with how
 150  * long a server takes to reboot.
 151  */
 152 #define XS_TCP_CONN_TO          (60U * HZ)
 153
 154 /*
 155  * Wait duration for a reply from the RPC portmapper.
 156  */
 157 #define XS_BIND_TO              (60U * HZ)
 158
 159 /*
 160  * Delay if a UDP socket connect error occurs.  This is most likely some
 161  * kind of resource problem on the local host.
 162  */
 163 #define XS_UDP_REEST_TO         (2U * HZ)
 164
 165 /*
 166  * The reestablish timeout allows clients to delay for a bit before attempting
 167  * to reconnect to a server that just dropped our connection.
 168  *
 169  * We implement an exponential backoff when trying to reestablish a TCP
 170  * transport connection with the server.  Some servers like to drop a TCP
 171  * connection when they are overworked, so we start with a short timeout and
 172  * increase over time if the server is down or not responding.
 173  */
 174 #define XS_TCP_INIT_REEST_TO    (3U * HZ)
 175 #define XS_TCP_MAX_REEST_TO     (5U * 60 * HZ)
 176
 177 /*
 178  * TCP idle timeout; client drops the transport socket if it is idle
 179  * for this long.  Note that we also timeout UDP sockets to prevent
 180  * holding port numbers when there is no RPC traffic.
 181  */
 182 #define XS_IDLE_DISC_TO         (5U * 60 * HZ)
 183
 184 #ifdef RPC_DEBUG
 185 # undef  RPC_DEBUG_DATA
 186 # define RPCDBG_FACILITY        RPCDBG_TRANS
 187 #endif
 188
 189 #ifdef RPC_DEBUG_DATA
 190 static void xs_pktdump(char *msg, u32 *packet, unsigned int count)
 191 {
 192         u8 *buf = (u8 *) packet;
 193         int j;
 194
 195         dprintk("RPC:       %s\n", msg);
 196         for (j = 0; j < count && j < 128; j += 4) {
 197                 if (!(j & 31)) {
 198                         if (j)
 199                                 dprintk("\n");
 200                         dprintk("0x%04x ", j);
 201                 }
 202                 dprintk("%02x%02x%02x%02x ",
 203                         buf[j], buf[j+1], buf[j+2], buf[j+3]);
 204         }
 205         dprintk("\n");
 206 }
 207 #else
 208 static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count)
 209 {
 210         /* NOP */
 211 }
 212 #endif
 213
 214 struct sock_xprt {
 215         struct rpc_xprt         xprt;
 216
 217         /*
 218          * Network layer
 219          */
 220         struct socket *         sock;
 221         struct sock *           inet;
 222
 223         /*
 224          * State of TCP reply receive
 225          */
 226         __be32                  tcp_fraghdr,
 227                                 tcp_xid;
 228
 229         u32                     tcp_offset,
 230                                 tcp_reclen;
 231
 232         unsigned long           tcp_copied,
 233                                 tcp_flags;
 234
 235         /*
 236          * Connection of transports
 237          */
 238         struct delayed_work     connect_worker;
 239         struct sockaddr_storage addr;
 240         unsigned short          port;
 241
 242         /*
 243          * UDP socket buffer size parameters
 244          */
 245         size_t                  rcvsize,
 246                                 sndsize;
 247
 248         /*
 249          * Saved socket callback addresses
 250          */
 251         void                    (*old_data_ready)(struct sock *, int);
 252         void                    (*old_state_change)(struct sock *);
 253         void                    (*old_write_space)(struct sock *);
 254         void                    (*old_error_report)(struct sock *);
 255 };
 256
 257 /*
 258  * TCP receive state flags
 259  */
 260 #define TCP_RCV_LAST_FRAG       (1UL << 0)
 261 #define TCP_RCV_COPY_FRAGHDR    (1UL << 1)
 262 #define TCP_RCV_COPY_XID        (1UL << 2)
 263 #define TCP_RCV_COPY_DATA       (1UL << 3)
 264
 265 static inline struct sockaddr *xs_addr(struct rpc_xprt *xprt)
 266 {
 267         return (struct sockaddr *) &xprt->addr;
 268 }
 269
 270 static inline struct sockaddr_in *xs_addr_in(struct rpc_xprt *xprt)
 271 {
 272         return (struct sockaddr_in *) &xprt->addr;
 273 }
 274
 275 static inline struct sockaddr_in6 *xs_addr_in6(struct rpc_xprt *xprt)
 276 {
 277         return (struct sockaddr_in6 *) &xprt->addr;
 278 }
 279
 280 static void xs_format_ipv4_peer_addresses(struct rpc_xprt *xprt,
 281                                           const char *protocol,
 282                                           const char *netid)
 283 {
 284         struct sockaddr_in *addr = xs_addr_in(xprt);
 285         char *buf;
 286
 287         buf = kzalloc(20, GFP_KERNEL);
 288         if (buf) {
 289                 snprintf(buf, 20, "%pI4", &addr->sin_addr.s_addr);
 290         }
 291         xprt->address_strings[RPC_DISPLAY_ADDR] = buf;
 292
 293         buf = kzalloc(8, GFP_KERNEL);
 294         if (buf) {
 295                 snprintf(buf, 8, "%u",
 296                                 ntohs(addr->sin_port));
 297         }
 298         xprt->address_strings[RPC_DISPLAY_PORT] = buf;
 299
 300         xprt->address_strings[RPC_DISPLAY_PROTO] = protocol;
 301
 302         buf = kzalloc(48, GFP_KERNEL);
 303         if (buf) {
 304                 snprintf(buf, 48, "addr=%pI4 port=%u proto=%s",
 305                         &addr->sin_addr.s_addr,
 306                         ntohs(addr->sin_port),
 307                         protocol);
 308         }
 309         xprt->address_strings[RPC_DISPLAY_ALL] = buf;
 310
 311         buf = kzalloc(10, GFP_KERNEL);
 312         if (buf) {
 313                 snprintf(buf, 10, "%02x%02x%02x%02x",
 314                                 NIPQUAD(addr->sin_addr.s_addr));
 315         }
 316         xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = buf;
 317
 318         buf = kzalloc(8, GFP_KERNEL);
 319         if (buf) {
 320                 snprintf(buf, 8, "%4hx",
 321                                 ntohs(addr->sin_port));
 322         }
 323         xprt->address_strings[RPC_DISPLAY_HEX_PORT] = buf;
 324
 325         buf = kzalloc(30, GFP_KERNEL);
 326         if (buf) {
 327                 snprintf(buf, 30, "%pI4.%u.%u",
 328                                 &addr->sin_addr.s_addr,
 329                                 ntohs(addr->sin_port) >> 8,
 330                                 ntohs(addr->sin_port) & 0xff);
 331         }
 332         xprt->address_strings[RPC_DISPLAY_UNIVERSAL_ADDR] = buf;
 333
 334         xprt->address_strings[RPC_DISPLAY_NETID] = netid;
 335 }
 336
 337 static void xs_format_ipv6_peer_addresses(struct rpc_xprt *xprt,
 338                                           const char *protocol,
 339                                           const char *netid)
 340 {
 341         struct sockaddr_in6 *addr = xs_addr_in6(xprt);
 342         char *buf;
 343
 344         buf = kzalloc(40, GFP_KERNEL);
 345         if (buf) {
 346                 snprintf(buf, 40, "%pI6",&addr->sin6_addr);
 347         }
 348         xprt->address_strings[RPC_DISPLAY_ADDR] = buf;
 349
 350         buf = kzalloc(8, GFP_KERNEL);
 351         if (buf) {
 352                 snprintf(buf, 8, "%u",
 353                                 ntohs(addr->sin6_port));
 354         }
 355         xprt->address_strings[RPC_DISPLAY_PORT] = buf;
 356
 357         xprt->address_strings[RPC_DISPLAY_PROTO] = protocol;
 358
 359         buf = kzalloc(64, GFP_KERNEL);
 360         if (buf) {
 361                 snprintf(buf, 64, "addr=%pI6 port=%u proto=%s",
 362                                 &addr->sin6_addr,
 363                                 ntohs(addr->sin6_port),
 364                                 protocol);
 365         }
 366         xprt->address_strings[RPC_DISPLAY_ALL] = buf;
 367
 368         buf = kzalloc(36, GFP_KERNEL);
 369         if (buf)
 370                 snprintf(buf, 36, "%pi6", &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, "%pI6.%u.%u",
 384                          &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 -ENOTSOCK;
 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 int 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         int ret = 0;
 532
 533         dprintk("RPC: %5u xmit incomplete (%u left of %u)\n",
 534                         task->tk_pid, req->rq_slen - req->rq_bytes_sent,
 535                         req->rq_slen);
 536
 537         /* Protect against races with write_space */
 538         spin_lock_bh(&xprt->transport_lock);
 539
 540         /* Don't race with disconnect */
 541         if (xprt_connected(xprt)) {
 542                 if (test_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags)) {
 543                         ret = -EAGAIN;
 544                         /*
 545                          * Notify TCP that we're limited by the application
 546                          * window size
 547                          */
 548                         set_bit(SOCK_NOSPACE, &transport->sock->flags);
 549                         transport->inet->sk_write_pending++;
 550                         /* ...and wait for more buffer space */
 551                         xprt_wait_for_buffer_space(task, xs_nospace_callback);
 552                 }
 553         } else {
 554                 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 555                 ret = -ENOTCONN;
 556         }
 557
 558         spin_unlock_bh(&xprt->transport_lock);
 559         return ret;
 560 }
 561
 562 /**
 563  * xs_udp_send_request - write an RPC request to a UDP socket
 564  * @task: address of RPC task that manages the state of an RPC request
 565  *
 566  * Return values:
 567  *        0:    The request has been sent
 568  *   EAGAIN:    The socket was blocked, please call again later to
 569  *              complete the request
 570  * ENOTCONN:    Caller needs to invoke connect logic then call again
 571  *    other:    Some other error occured, the request was not sent
 572  */
 573 static int xs_udp_send_request(struct rpc_task *task)
 574 {
 575         struct rpc_rqst *req = task->tk_rqstp;
 576         struct rpc_xprt *xprt = req->rq_xprt;
 577         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 578         struct xdr_buf *xdr = &req->rq_snd_buf;
 579         int status;
 580
 581         xs_pktdump("packet data:",
 582                                 req->rq_svec->iov_base,
 583                                 req->rq_svec->iov_len);
 584
 585         if (!xprt_bound(xprt))
 586                 return -ENOTCONN;
 587         status = xs_sendpages(transport->sock,
 588                               xs_addr(xprt),
 589                               xprt->addrlen, xdr,
 590                               req->rq_bytes_sent);
 591
 592         dprintk("RPC:       xs_udp_send_request(%u) = %d\n",
 593                         xdr->len - req->rq_bytes_sent, status);
 594
 595         if (status >= 0) {
 596                 task->tk_bytes_sent += status;
 597                 if (status >= req->rq_slen)
 598                         return 0;
 599                 /* Still some bytes left; set up for a retry later. */
 600                 status = -EAGAIN;
 601         }
 602         if (!transport->sock)
 603                 goto out;
 604
 605         switch (status) {
 606         case -ENOTSOCK:
 607                 status = -ENOTCONN;
 608                 /* Should we call xs_close() here? */
 609                 break;
 610         case -EAGAIN:
 611                 status = xs_nospace(task);
 612                 break;
 613         default:
 614                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
 615                         -status);
 616         case -ENETUNREACH:
 617         case -EPIPE:
 618         case -ECONNREFUSED:
 619                 /* When the server has died, an ICMP port unreachable message
 620                  * prompts ECONNREFUSED. */
 621                 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 622         }
 623 out:
 624         return status;
 625 }
 626
 627 /**
 628  * xs_tcp_shutdown - gracefully shut down a TCP socket
 629  * @xprt: transport
 630  *
 631  * Initiates a graceful shutdown of the TCP socket by calling the
 632  * equivalent of shutdown(SHUT_WR);
 633  */
 634 static void xs_tcp_shutdown(struct rpc_xprt *xprt)
 635 {
 636         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 637         struct socket *sock = transport->sock;
 638
 639         if (sock != NULL)
 640                 kernel_sock_shutdown(sock, SHUT_WR);
 641 }
 642
 643 static inline void xs_encode_tcp_record_marker(struct xdr_buf *buf)
 644 {
 645         u32 reclen = buf->len - sizeof(rpc_fraghdr);
 646         rpc_fraghdr *base = buf->head[0].iov_base;
 647         *base = htonl(RPC_LAST_STREAM_FRAGMENT | reclen);
 648 }
 649
 650 /**
 651  * xs_tcp_send_request - write an RPC request to a TCP socket
 652  * @task: address of RPC task that manages the state of an RPC request
 653  *
 654  * Return values:
 655  *        0:    The request has been sent
 656  *   EAGAIN:    The socket was blocked, please call again later to
 657  *              complete the request
 658  * ENOTCONN:    Caller needs to invoke connect logic then call again
 659  *    other:    Some other error occured, the request was not sent
 660  *
 661  * XXX: In the case of soft timeouts, should we eventually give up
 662  *      if sendmsg is not able to make progress?
 663  */
 664 static int xs_tcp_send_request(struct rpc_task *task)
 665 {
 666         struct rpc_rqst *req = task->tk_rqstp;
 667         struct rpc_xprt *xprt = req->rq_xprt;
 668         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 669         struct xdr_buf *xdr = &req->rq_snd_buf;
 670         int status;
 671
 672         xs_encode_tcp_record_marker(&req->rq_snd_buf);
 673
 674         xs_pktdump("packet data:",
 675                                 req->rq_svec->iov_base,
 676                                 req->rq_svec->iov_len);
 677
 678         /* Continue transmitting the packet/record. We must be careful
 679          * to cope with writespace callbacks arriving _after_ we have
 680          * called sendmsg(). */
 681         while (1) {
 682                 status = xs_sendpages(transport->sock,
 683                                         NULL, 0, xdr, req->rq_bytes_sent);
 684
 685                 dprintk("RPC:       xs_tcp_send_request(%u) = %d\n",
 686                                 xdr->len - req->rq_bytes_sent, status);
 687
 688                 if (unlikely(status < 0))
 689                         break;
 690
 691                 /* If we've sent the entire packet, immediately
 692                  * reset the count of bytes sent. */
 693                 req->rq_bytes_sent += status;
 694                 task->tk_bytes_sent += status;
 695                 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
 696                         req->rq_bytes_sent = 0;
 697                         return 0;
 698                 }
 699
 700                 if (status != 0)
 701                         continue;
 702                 status = -EAGAIN;
 703                 break;
 704         }
 705         if (!transport->sock)
 706                 goto out;
 707
 708         switch (status) {
 709         case -ENOTSOCK:
 710                 status = -ENOTCONN;
 711                 /* Should we call xs_close() here? */
 712                 break;
 713         case -EAGAIN:
 714                 status = xs_nospace(task);
 715                 break;
 716         default:
 717                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
 718                         -status);
 719         case -ECONNRESET:
 720                 xs_tcp_shutdown(xprt);
 721         case -ECONNREFUSED:
 722         case -ENOTCONN:
 723         case -EPIPE:
 724                 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 725         }
 726 out:
 727         return status;
 728 }
 729
 730 /**
 731  * xs_tcp_release_xprt - clean up after a tcp transmission
 732  * @xprt: transport
 733  * @task: rpc task
 734  *
 735  * This cleans up if an error causes us to abort the transmission of a request.
 736  * In this case, the socket may need to be reset in order to avoid confusing
 737  * the server.
 738  */
 739 static void xs_tcp_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
 740 {
 741         struct rpc_rqst *req;
 742
 743         if (task != xprt->snd_task)
 744                 return;
 745         if (task == NULL)
 746                 goto out_release;
 747         req = task->tk_rqstp;
 748         if (req->rq_bytes_sent == 0)
 749                 goto out_release;
 750         if (req->rq_bytes_sent == req->rq_snd_buf.len)
 751                 goto out_release;
 752         set_bit(XPRT_CLOSE_WAIT, &task->tk_xprt->state);
 753 out_release:
 754         xprt_release_xprt(xprt, task);
 755 }
 756
 757 static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk)
 758 {
 759         transport->old_data_ready = sk->sk_data_ready;
 760         transport->old_state_change = sk->sk_state_change;
 761         transport->old_write_space = sk->sk_write_space;
 762         transport->old_error_report = sk->sk_error_report;
 763 }
 764
 765 static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk)
 766 {
 767         sk->sk_data_ready = transport->old_data_ready;
 768         sk->sk_state_change = transport->old_state_change;
 769         sk->sk_write_space = transport->old_write_space;
 770         sk->sk_error_report = transport->old_error_report;
 771 }
 772
 773 static void xs_reset_transport(struct sock_xprt *transport)
 774 {
 775         struct socket *sock = transport->sock;
 776         struct sock *sk = transport->inet;
 777
 778         if (sk == NULL)
 779                 return;
 780
 781         write_lock_bh(&sk->sk_callback_lock);
 782         transport->inet = NULL;
 783         transport->sock = NULL;
 784
 785         sk->sk_user_data = NULL;
 786
 787         xs_restore_old_callbacks(transport, sk);
 788         write_unlock_bh(&sk->sk_callback_lock);
 789
 790         sk->sk_no_check = 0;
 791
 792         sock_release(sock);
 793 }
 794
 795 /**
 796  * xs_close - close a socket
 797  * @xprt: transport
 798  *
 799  * This is used when all requests are complete; ie, no DRC state remains
 800  * on the server we want to save.
 801  */
 802 static void xs_close(struct rpc_xprt *xprt)
 803 {
 804         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 805
 806         dprintk("RPC:       xs_close xprt %p\n", xprt);
 807
 808         xs_reset_transport(transport);
 809
 810         smp_mb__before_clear_bit();
 811         clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
 812         clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
 813         clear_bit(XPRT_CLOSING, &xprt->state);
 814         smp_mb__after_clear_bit();
 815         xprt_disconnect_done(xprt);
 816 }
 817
 818 /**
 819  * xs_destroy - prepare to shutdown a transport
 820  * @xprt: doomed transport
 821  *
 822  */
 823 static void xs_destroy(struct rpc_xprt *xprt)
 824 {
 825         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 826
 827         dprintk("RPC:       xs_destroy xprt %p\n", xprt);
 828
 829         cancel_rearming_delayed_work(&transport->connect_worker);
 830
 831         xs_close(xprt);
 832         xs_free_peer_addresses(xprt);
 833         kfree(xprt->slot);
 834         kfree(xprt);
 835         module_put(THIS_MODULE);
 836 }
 837
 838 static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
 839 {
 840         return (struct rpc_xprt *) sk->sk_user_data;
 841 }
 842
 843 /**
 844  * xs_udp_data_ready - "data ready" callback for UDP sockets
 845  * @sk: socket with data to read
 846  * @len: how much data to read
 847  *
 848  */
 849 static void xs_udp_data_ready(struct sock *sk, int len)
 850 {
 851         struct rpc_task *task;
 852         struct rpc_xprt *xprt;
 853         struct rpc_rqst *rovr;
 854         struct sk_buff *skb;
 855         int err, repsize, copied;
 856         u32 _xid;
 857         __be32 *xp;
 858
 859         read_lock(&sk->sk_callback_lock);
 860         dprintk("RPC:       xs_udp_data_ready...\n");
 861         if (!(xprt = xprt_from_sock(sk)))
 862                 goto out;
 863
 864         if ((skb = skb_recv_datagram(sk, 0, 1, &err)) == NULL)
 865                 goto out;
 866
 867         if (xprt->shutdown)
 868                 goto dropit;
 869
 870         repsize = skb->len - sizeof(struct udphdr);
 871         if (repsize < 4) {
 872                 dprintk("RPC:       impossible RPC reply size %d!\n", repsize);
 873                 goto dropit;
 874         }
 875
 876         /* Copy the XID from the skb... */
 877         xp = skb_header_pointer(skb, sizeof(struct udphdr),
 878                                 sizeof(_xid), &_xid);
 879         if (xp == NULL)
 880                 goto dropit;
 881
 882         /* Look up and lock the request corresponding to the given XID */
 883         spin_lock(&xprt->transport_lock);
 884         rovr = xprt_lookup_rqst(xprt, *xp);
 885         if (!rovr)
 886                 goto out_unlock;
 887         task = rovr->rq_task;
 888
 889         if ((copied = rovr->rq_private_buf.buflen) > repsize)
 890                 copied = repsize;
 891
 892         /* Suck it into the iovec, verify checksum if not done by hw. */
 893         if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
 894                 UDPX_INC_STATS_BH(sk, UDP_MIB_INERRORS);
 895                 goto out_unlock;
 896         }
 897
 898         UDPX_INC_STATS_BH(sk, UDP_MIB_INDATAGRAMS);
 899
 900         /* Something worked... */
 901         dst_confirm(skb->dst);
 902
 903         xprt_adjust_cwnd(task, copied);
 904         xprt_update_rtt(task);
 905         xprt_complete_rqst(task, copied);
 906
 907  out_unlock:
 908         spin_unlock(&xprt->transport_lock);
 909  dropit:
 910         skb_free_datagram(sk, skb);
 911  out:
 912         read_unlock(&sk->sk_callback_lock);
 913 }
 914
 915 static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
 916 {
 917         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 918         size_t len, used;
 919         char *p;
 920
 921         p = ((char *) &transport->tcp_fraghdr) + transport->tcp_offset;
 922         len = sizeof(transport->tcp_fraghdr) - transport->tcp_offset;
 923         used = xdr_skb_read_bits(desc, p, len);
 924         transport->tcp_offset += used;
 925         if (used != len)
 926                 return;
 927
 928         transport->tcp_reclen = ntohl(transport->tcp_fraghdr);
 929         if (transport->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
 930                 transport->tcp_flags |= TCP_RCV_LAST_FRAG;
 931         else
 932                 transport->tcp_flags &= ~TCP_RCV_LAST_FRAG;
 933         transport->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
 934
 935         transport->tcp_flags &= ~TCP_RCV_COPY_FRAGHDR;
 936         transport->tcp_offset = 0;
 937
 938         /* Sanity check of the record length */
 939         if (unlikely(transport->tcp_reclen < 4)) {
 940                 dprintk("RPC:       invalid TCP record fragment length\n");
 941                 xprt_force_disconnect(xprt);
 942                 return;
 943         }
 944         dprintk("RPC:       reading TCP record fragment of length %d\n",
 945                         transport->tcp_reclen);
 946 }
 947
 948 static void xs_tcp_check_fraghdr(struct sock_xprt *transport)
 949 {
 950         if (transport->tcp_offset == transport->tcp_reclen) {
 951                 transport->tcp_flags |= TCP_RCV_COPY_FRAGHDR;
 952                 transport->tcp_offset = 0;
 953                 if (transport->tcp_flags & TCP_RCV_LAST_FRAG) {
 954                         transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
 955                         transport->tcp_flags |= TCP_RCV_COPY_XID;
 956                         transport->tcp_copied = 0;
 957                 }
 958         }
 959 }
 960
 961 static inline void xs_tcp_read_xid(struct sock_xprt *transport, struct xdr_skb_reader *desc)
 962 {
 963         size_t len, used;
 964         char *p;
 965
 966         len = sizeof(transport->tcp_xid) - transport->tcp_offset;
 967         dprintk("RPC:       reading XID (%Zu bytes)\n", len);
 968         p = ((char *) &transport->tcp_xid) + transport->tcp_offset;
 969         used = xdr_skb_read_bits(desc, p, len);
 970         transport->tcp_offset += used;
 971         if (used != len)
 972                 return;
 973         transport->tcp_flags &= ~TCP_RCV_COPY_XID;
 974         transport->tcp_flags |= TCP_RCV_COPY_DATA;
 975         transport->tcp_copied = 4;
 976         dprintk("RPC:       reading reply for XID %08x\n",
 977                         ntohl(transport->tcp_xid));
 978         xs_tcp_check_fraghdr(transport);
 979 }
 980
 981 static inline void xs_tcp_read_request(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
 982 {
 983         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 984         struct rpc_rqst *req;
 985         struct xdr_buf *rcvbuf;
 986         size_t len;
 987         ssize_t r;
 988
 989         /* Find and lock the request corresponding to this xid */
 990         spin_lock(&xprt->transport_lock);
 991         req = xprt_lookup_rqst(xprt, transport->tcp_xid);
 992         if (!req) {
 993                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
 994                 dprintk("RPC:       XID %08x request not found!\n",
 995                                 ntohl(transport->tcp_xid));
 996                 spin_unlock(&xprt->transport_lock);
 997                 return;
 998         }
 999
1000         rcvbuf = &req->rq_private_buf;
1001         len = desc->count;
1002         if (len > transport->tcp_reclen - transport->tcp_offset) {
1003                 struct xdr_skb_reader my_desc;
1004
1005                 len = transport->tcp_reclen - transport->tcp_offset;
1006                 memcpy(&my_desc, desc, sizeof(my_desc));
1007                 my_desc.count = len;
1008                 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
1009                                           &my_desc, xdr_skb_read_bits);
1010                 desc->count -= r;
1011                 desc->offset += r;
1012         } else
1013                 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
1014                                           desc, xdr_skb_read_bits);
1015
1016         if (r > 0) {
1017                 transport->tcp_copied += r;
1018                 transport->tcp_offset += r;
1019         }
1020         if (r != len) {
1021                 /* Error when copying to the receive buffer,
1022                  * usually because we weren't able to allocate
1023                  * additional buffer pages. All we can do now
1024                  * is turn off TCP_RCV_COPY_DATA, so the request
1025                  * will not receive any additional updates,
1026                  * and time out.
1027                  * Any remaining data from this record will
1028                  * be discarded.
1029                  */
1030                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1031                 dprintk("RPC:       XID %08x truncated request\n",
1032                                 ntohl(transport->tcp_xid));
1033                 dprintk("RPC:       xprt = %p, tcp_copied = %lu, "
1034                                 "tcp_offset = %u, tcp_reclen = %u\n",
1035                                 xprt, transport->tcp_copied,
1036                                 transport->tcp_offset, transport->tcp_reclen);
1037                 goto out;
1038         }
1039
1040         dprintk("RPC:       XID %08x read %Zd bytes\n",
1041                         ntohl(transport->tcp_xid), r);
1042         dprintk("RPC:       xprt = %p, tcp_copied = %lu, tcp_offset = %u, "
1043                         "tcp_reclen = %u\n", xprt, transport->tcp_copied,
1044                         transport->tcp_offset, transport->tcp_reclen);
1045
1046         if (transport->tcp_copied == req->rq_private_buf.buflen)
1047                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1048         else if (transport->tcp_offset == transport->tcp_reclen) {
1049                 if (transport->tcp_flags & TCP_RCV_LAST_FRAG)
1050                         transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1051         }
1052
1053 out:
1054         if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1055                 xprt_complete_rqst(req->rq_task, transport->tcp_copied);
1056         spin_unlock(&xprt->transport_lock);
1057         xs_tcp_check_fraghdr(transport);
1058 }
1059
1060 static inline void xs_tcp_read_discard(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1061 {
1062         size_t len;
1063
1064         len = transport->tcp_reclen - transport->tcp_offset;
1065         if (len > desc->count)
1066                 len = desc->count;
1067         desc->count -= len;
1068         desc->offset += len;
1069         transport->tcp_offset += len;
1070         dprintk("RPC:       discarded %Zu bytes\n", len);
1071         xs_tcp_check_fraghdr(transport);
1072 }
1073
1074 static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
1075 {
1076         struct rpc_xprt *xprt = rd_desc->arg.data;
1077         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1078         struct xdr_skb_reader desc = {
1079                 .skb    = skb,
1080                 .offset = offset,
1081                 .count  = len,
1082         };
1083
1084         dprintk("RPC:       xs_tcp_data_recv started\n");
1085         do {
1086                 /* Read in a new fragment marker if necessary */
1087                 /* Can we ever really expect to get completely empty fragments? */
1088                 if (transport->tcp_flags & TCP_RCV_COPY_FRAGHDR) {
1089                         xs_tcp_read_fraghdr(xprt, &desc);
1090                         continue;
1091                 }
1092                 /* Read in the xid if necessary */
1093                 if (transport->tcp_flags & TCP_RCV_COPY_XID) {
1094                         xs_tcp_read_xid(transport, &desc);
1095                         continue;
1096                 }
1097                 /* Read in the request data */
1098                 if (transport->tcp_flags & TCP_RCV_COPY_DATA) {
1099                         xs_tcp_read_request(xprt, &desc);
1100                         continue;
1101                 }
1102                 /* Skip over any trailing bytes on short reads */
1103                 xs_tcp_read_discard(transport, &desc);
1104         } while (desc.count);
1105         dprintk("RPC:       xs_tcp_data_recv done\n");
1106         return len - desc.count;
1107 }
1108
1109 /**
1110  * xs_tcp_data_ready - "data ready" callback for TCP sockets
1111  * @sk: socket with data to read
1112  * @bytes: how much data to read
1113  *
1114  */
1115 static void xs_tcp_data_ready(struct sock *sk, int bytes)
1116 {
1117         struct rpc_xprt *xprt;
1118         read_descriptor_t rd_desc;
1119         int read;
1120
1121         dprintk("RPC:       xs_tcp_data_ready...\n");
1122
1123         read_lock(&sk->sk_callback_lock);
1124         if (!(xprt = xprt_from_sock(sk)))
1125                 goto out;
1126         if (xprt->shutdown)
1127                 goto out;
1128
1129         /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
1130         rd_desc.arg.data = xprt;
1131         do {
1132                 rd_desc.count = 65536;
1133                 read = tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
1134         } while (read > 0);
1135 out:
1136         read_unlock(&sk->sk_callback_lock);
1137 }
1138
1139 /*
1140  * Do the equivalent of linger/linger2 handling for dealing with
1141  * broken servers that don't close the socket in a timely
1142  * fashion
1143  */
1144 static void xs_tcp_schedule_linger_timeout(struct rpc_xprt *xprt,
1145                 unsigned long timeout)
1146 {
1147         struct sock_xprt *transport;
1148
1149         if (xprt_test_and_set_connecting(xprt))
1150                 return;
1151         set_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1152         transport = container_of(xprt, struct sock_xprt, xprt);
1153         queue_delayed_work(rpciod_workqueue, &transport->connect_worker,
1154                            timeout);
1155 }
1156
1157 static void xs_tcp_cancel_linger_timeout(struct rpc_xprt *xprt)
1158 {
1159         struct sock_xprt *transport;
1160
1161         transport = container_of(xprt, struct sock_xprt, xprt);
1162
1163         if (!test_bit(XPRT_CONNECTION_ABORT, &xprt->state) ||
1164             !cancel_delayed_work(&transport->connect_worker))
1165                 return;
1166         clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1167         xprt_clear_connecting(xprt);
1168 }
1169
1170 static void xs_sock_mark_closed(struct rpc_xprt *xprt)
1171 {
1172         smp_mb__before_clear_bit();
1173         clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1174         clear_bit(XPRT_CLOSING, &xprt->state);
1175         smp_mb__after_clear_bit();
1176         /* Mark transport as closed and wake up all pending tasks */
1177         xprt_disconnect_done(xprt);
1178 }
1179
1180 /**
1181  * xs_tcp_state_change - callback to handle TCP socket state changes
1182  * @sk: socket whose state has changed
1183  *
1184  */
1185 static void xs_tcp_state_change(struct sock *sk)
1186 {
1187         struct rpc_xprt *xprt;
1188
1189         read_lock(&sk->sk_callback_lock);
1190         if (!(xprt = xprt_from_sock(sk)))
1191                 goto out;
1192         dprintk("RPC:       xs_tcp_state_change client %p...\n", xprt);
1193         dprintk("RPC:       state %x conn %d dead %d zapped %d\n",
1194                         sk->sk_state, xprt_connected(xprt),
1195                         sock_flag(sk, SOCK_DEAD),
1196                         sock_flag(sk, SOCK_ZAPPED));
1197
1198         switch (sk->sk_state) {
1199         case TCP_ESTABLISHED:
1200                 spin_lock_bh(&xprt->transport_lock);
1201                 if (!xprt_test_and_set_connected(xprt)) {
1202                         struct sock_xprt *transport = container_of(xprt,
1203                                         struct sock_xprt, xprt);
1204
1205                         /* Reset TCP record info */
1206                         transport->tcp_offset = 0;
1207                         transport->tcp_reclen = 0;
1208                         transport->tcp_copied = 0;
1209                         transport->tcp_flags =
1210                                 TCP_RCV_COPY_FRAGHDR | TCP_RCV_COPY_XID;
1211
1212                         xprt_wake_pending_tasks(xprt, -EAGAIN);
1213                 }
1214                 spin_unlock_bh(&xprt->transport_lock);
1215                 break;
1216         case TCP_FIN_WAIT1:
1217                 /* The client initiated a shutdown of the socket */
1218                 xprt->connect_cookie++;
1219                 xprt->reestablish_timeout = 0;
1220                 set_bit(XPRT_CLOSING, &xprt->state);
1221                 smp_mb__before_clear_bit();
1222                 clear_bit(XPRT_CONNECTED, &xprt->state);
1223                 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1224                 smp_mb__after_clear_bit();
1225                 xs_tcp_schedule_linger_timeout(xprt, XS_TCP_LINGER_TO);
1226                 break;
1227         case TCP_CLOSE_WAIT:
1228                 /* The server initiated a shutdown of the socket */
1229                 xprt_force_disconnect(xprt);
1230         case TCP_SYN_SENT:
1231                 xprt->connect_cookie++;
1232         case TCP_CLOSING:
1233                 /*
1234                  * If the server closed down the connection, make sure that
1235                  * we back off before reconnecting
1236                  */
1237                 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1238                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1239                 break;
1240         case TCP_LAST_ACK:
1241                 set_bit(XPRT_CLOSING, &xprt->state);
1242                 xs_tcp_schedule_linger_timeout(xprt, XS_TCP_LINGER_TO);
1243                 smp_mb__before_clear_bit();
1244                 clear_bit(XPRT_CONNECTED, &xprt->state);
1245                 smp_mb__after_clear_bit();
1246                 break;
1247         case TCP_CLOSE:
1248                 xs_tcp_cancel_linger_timeout(xprt);
1249                 xs_sock_mark_closed(xprt);
1250         }
1251  out:
1252         read_unlock(&sk->sk_callback_lock);
1253 }
1254
1255 /**
1256  * xs_error_report - callback mainly for catching socket errors
1257  * @sk: socket
1258  */
1259 static void xs_error_report(struct sock *sk)
1260 {
1261         struct rpc_xprt *xprt;
1262
1263         read_lock(&sk->sk_callback_lock);
1264         if (!(xprt = xprt_from_sock(sk)))
1265                 goto out;
1266         dprintk("RPC:       %s client %p...\n"
1267                         "RPC:       error %d\n",
1268                         __func__, xprt, sk->sk_err);
1269         xprt_wake_pending_tasks(xprt, -EAGAIN);
1270 out:
1271         read_unlock(&sk->sk_callback_lock);
1272 }
1273
1274 /**
1275  * xs_udp_write_space - callback invoked when socket buffer space
1276  *                             becomes available
1277  * @sk: socket whose state has changed
1278  *
1279  * Called when more output buffer space is available for this socket.
1280  * We try not to wake our writers until they can make "significant"
1281  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1282  * with a bunch of small requests.
1283  */
1284 static void xs_udp_write_space(struct sock *sk)
1285 {
1286         read_lock(&sk->sk_callback_lock);
1287
1288         /* from net/core/sock.c:sock_def_write_space */
1289         if (sock_writeable(sk)) {
1290                 struct socket *sock;
1291                 struct rpc_xprt *xprt;
1292
1293                 if (unlikely(!(sock = sk->sk_socket)))
1294                         goto out;
1295                 clear_bit(SOCK_NOSPACE, &sock->flags);
1296
1297                 if (unlikely(!(xprt = xprt_from_sock(sk))))
1298                         goto out;
1299                 if (test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags) == 0)
1300                         goto out;
1301
1302                 xprt_write_space(xprt);
1303         }
1304
1305  out:
1306         read_unlock(&sk->sk_callback_lock);
1307 }
1308
1309 /**
1310  * xs_tcp_write_space - callback invoked when socket buffer space
1311  *                             becomes available
1312  * @sk: socket whose state has changed
1313  *
1314  * Called when more output buffer space is available for this socket.
1315  * We try not to wake our writers until they can make "significant"
1316  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1317  * with a bunch of small requests.
1318  */
1319 static void xs_tcp_write_space(struct sock *sk)
1320 {
1321         read_lock(&sk->sk_callback_lock);
1322
1323         /* from net/core/stream.c:sk_stream_write_space */
1324         if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk)) {
1325                 struct socket *sock;
1326                 struct rpc_xprt *xprt;
1327
1328                 if (unlikely(!(sock = sk->sk_socket)))
1329                         goto out;
1330                 clear_bit(SOCK_NOSPACE, &sock->flags);
1331
1332                 if (unlikely(!(xprt = xprt_from_sock(sk))))
1333                         goto out;
1334                 if (test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags) == 0)
1335                         goto out;
1336
1337                 xprt_write_space(xprt);
1338         }
1339
1340  out:
1341         read_unlock(&sk->sk_callback_lock);
1342 }
1343
1344 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
1345 {
1346         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1347         struct sock *sk = transport->inet;
1348
1349         if (transport->rcvsize) {
1350                 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1351                 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
1352         }
1353         if (transport->sndsize) {
1354                 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1355                 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1356                 sk->sk_write_space(sk);
1357         }
1358 }
1359
1360 /**
1361  * xs_udp_set_buffer_size - set send and receive limits
1362  * @xprt: generic transport
1363  * @sndsize: requested size of send buffer, in bytes
1364  * @rcvsize: requested size of receive buffer, in bytes
1365  *
1366  * Set socket send and receive buffer size limits.
1367  */
1368 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1369 {
1370         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1371
1372         transport->sndsize = 0;
1373         if (sndsize)
1374                 transport->sndsize = sndsize + 1024;
1375         transport->rcvsize = 0;
1376         if (rcvsize)
1377                 transport->rcvsize = rcvsize + 1024;
1378
1379         xs_udp_do_set_buffer_size(xprt);
1380 }
1381
1382 /**
1383  * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1384  * @task: task that timed out
1385  *
1386  * Adjust the congestion window after a retransmit timeout has occurred.
1387  */
1388 static void xs_udp_timer(struct rpc_task *task)
1389 {
1390         xprt_adjust_cwnd(task, -ETIMEDOUT);
1391 }
1392
1393 static unsigned short xs_get_random_port(void)
1394 {
1395         unsigned short range = xprt_max_resvport - xprt_min_resvport;
1396         unsigned short rand = (unsigned short) net_random() % range;
1397         return rand + xprt_min_resvport;
1398 }
1399
1400 /**
1401  * xs_set_port - reset the port number in the remote endpoint address
1402  * @xprt: generic transport
1403  * @port: new port number
1404  *
1405  */
1406 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1407 {
1408         struct sockaddr *addr = xs_addr(xprt);
1409
1410         dprintk("RPC:       setting port for xprt %p to %u\n", xprt, port);
1411
1412         switch (addr->sa_family) {
1413         case AF_INET:
1414                 ((struct sockaddr_in *)addr)->sin_port = htons(port);
1415                 break;
1416         case AF_INET6:
1417                 ((struct sockaddr_in6 *)addr)->sin6_port = htons(port);
1418                 break;
1419         default:
1420                 BUG();
1421         }
1422 }
1423
1424 static unsigned short xs_get_srcport(struct sock_xprt *transport, struct socket *sock)
1425 {
1426         unsigned short port = transport->port;
1427
1428         if (port == 0 && transport->xprt.resvport)
1429                 port = xs_get_random_port();
1430         return port;
1431 }
1432
1433 static unsigned short xs_next_srcport(struct sock_xprt *transport, struct socket *sock, unsigned short port)
1434 {
1435         if (transport->port != 0)
1436                 transport->port = 0;
1437         if (!transport->xprt.resvport)
1438                 return 0;
1439         if (port <= xprt_min_resvport || port > xprt_max_resvport)
1440                 return xprt_max_resvport;
1441         return --port;
1442 }
1443
1444 static int xs_bind4(struct sock_xprt *transport, struct socket *sock)
1445 {
1446         struct sockaddr_in myaddr = {
1447                 .sin_family = AF_INET,
1448         };
1449         struct sockaddr_in *sa;
1450         int err, nloop = 0;
1451         unsigned short port = xs_get_srcport(transport, sock);
1452         unsigned short last;
1453
1454         sa = (struct sockaddr_in *)&transport->addr;
1455         myaddr.sin_addr = sa->sin_addr;
1456         do {
1457                 myaddr.sin_port = htons(port);
1458                 err = kernel_bind(sock, (struct sockaddr *) &myaddr,
1459                                                 sizeof(myaddr));
1460                 if (port == 0)
1461                         break;
1462                 if (err == 0) {
1463                         transport->port = port;
1464                         break;
1465                 }
1466                 last = port;
1467                 port = xs_next_srcport(transport, sock, port);
1468                 if (port > last)
1469                         nloop++;
1470         } while (err == -EADDRINUSE && nloop != 2);
1471         dprintk("RPC:       %s %pI4:%u: %s (%d)\n",
1472                         __func__, &myaddr.sin_addr,
1473                         port, err ? "failed" : "ok", err);
1474         return err;
1475 }
1476
1477 static int xs_bind6(struct sock_xprt *transport, struct socket *sock)
1478 {
1479         struct sockaddr_in6 myaddr = {
1480                 .sin6_family = AF_INET6,
1481         };
1482         struct sockaddr_in6 *sa;
1483         int err, nloop = 0;
1484         unsigned short port = xs_get_srcport(transport, sock);
1485         unsigned short last;
1486
1487         sa = (struct sockaddr_in6 *)&transport->addr;
1488         myaddr.sin6_addr = sa->sin6_addr;
1489         do {
1490                 myaddr.sin6_port = htons(port);
1491                 err = kernel_bind(sock, (struct sockaddr *) &myaddr,
1492                                                 sizeof(myaddr));
1493                 if (port == 0)
1494                         break;
1495                 if (err == 0) {
1496                         transport->port = port;
1497                         break;
1498                 }
1499                 last = port;
1500                 port = xs_next_srcport(transport, sock, port);
1501                 if (port > last)
1502                         nloop++;
1503         } while (err == -EADDRINUSE && nloop != 2);
1504         dprintk("RPC:       xs_bind6 %pI6:%u: %s (%d)\n",
1505                 &myaddr.sin6_addr, port, err ? "failed" : "ok", err);
1506         return err;
1507 }
1508
1509 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1510 static struct lock_class_key xs_key[2];
1511 static struct lock_class_key xs_slock_key[2];
1512
1513 static inline void xs_reclassify_socket4(struct socket *sock)
1514 {
1515         struct sock *sk = sock->sk;
1516
1517         BUG_ON(sock_owned_by_user(sk));
1518         sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1519                 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1520 }
1521
1522 static inline void xs_reclassify_socket6(struct socket *sock)
1523 {
1524         struct sock *sk = sock->sk;
1525
1526         BUG_ON(sock_owned_by_user(sk));
1527         sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1528                 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
1529 }
1530 #else
1531 static inline void xs_reclassify_socket4(struct socket *sock)
1532 {
1533 }
1534
1535 static inline void xs_reclassify_socket6(struct socket *sock)
1536 {
1537 }
1538 #endif
1539
1540 static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
1541 {
1542         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1543
1544         if (!transport->inet) {
1545                 struct sock *sk = sock->sk;
1546
1547                 write_lock_bh(&sk->sk_callback_lock);
1548
1549                 xs_save_old_callbacks(transport, sk);
1550
1551                 sk->sk_user_data = xprt;
1552                 sk->sk_data_ready = xs_udp_data_ready;
1553                 sk->sk_write_space = xs_udp_write_space;
1554                 sk->sk_error_report = xs_error_report;
1555                 sk->sk_no_check = UDP_CSUM_NORCV;
1556                 sk->sk_allocation = GFP_ATOMIC;
1557
1558                 xprt_set_connected(xprt);
1559
1560                 /* Reset to new socket */
1561                 transport->sock = sock;
1562                 transport->inet = sk;
1563
1564                 write_unlock_bh(&sk->sk_callback_lock);
1565         }
1566         xs_udp_do_set_buffer_size(xprt);
1567 }
1568
1569 /**
1570  * xs_udp_connect_worker4 - 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_worker4(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_INET, 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_socket4(sock);
1595
1596         if (xs_bind4(transport, sock)) {
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_clear_connecting(xprt);
1608         xprt_wake_pending_tasks(xprt, status);
1609 }
1610
1611 /**
1612  * xs_udp_connect_worker6 - set up a UDP socket
1613  * @work: RPC transport to connect
1614  *
1615  * Invoked by a work queue tasklet.
1616  */
1617 static void xs_udp_connect_worker6(struct work_struct *work)
1618 {
1619         struct sock_xprt *transport =
1620                 container_of(work, struct sock_xprt, connect_worker.work);
1621         struct rpc_xprt *xprt = &transport->xprt;
1622         struct socket *sock = transport->sock;
1623         int err, status = -EIO;
1624
1625         if (xprt->shutdown)
1626                 goto out;
1627
1628         /* Start by resetting any existing state */
1629         xs_reset_transport(transport);
1630
1631         err = sock_create_kern(PF_INET6, SOCK_DGRAM, IPPROTO_UDP, &sock);
1632         if (err < 0) {
1633                 dprintk("RPC:       can't create UDP transport socket (%d).\n", -err);
1634                 goto out;
1635         }
1636         xs_reclassify_socket6(sock);
1637
1638         if (xs_bind6(transport, sock) < 0) {
1639                 sock_release(sock);
1640                 goto out;
1641         }
1642
1643         dprintk("RPC:       worker connecting xprt %p to address: %s\n",
1644                         xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1645
1646         xs_udp_finish_connecting(xprt, sock);
1647         status = 0;
1648 out:
1649         xprt_clear_connecting(xprt);
1650         xprt_wake_pending_tasks(xprt, status);
1651 }
1652
1653 /*
1654  * We need to preserve the port number so the reply cache on the server can
1655  * find our cached RPC replies when we get around to reconnecting.
1656  */
1657 static void xs_abort_connection(struct rpc_xprt *xprt, struct sock_xprt *transport)
1658 {
1659         int result;
1660         struct sockaddr any;
1661
1662         dprintk("RPC:       disconnecting xprt %p to reuse port\n", xprt);
1663
1664         /*
1665          * Disconnect the transport socket by doing a connect operation
1666          * with AF_UNSPEC.  This should return immediately...
1667          */
1668         memset(&any, 0, sizeof(any));
1669         any.sa_family = AF_UNSPEC;
1670         result = kernel_connect(transport->sock, &any, sizeof(any), 0);
1671         if (!result)
1672                 xs_sock_mark_closed(xprt);
1673         else
1674                 dprintk("RPC:       AF_UNSPEC connect return code %d\n",
1675                                 result);
1676 }
1677
1678 static void xs_tcp_reuse_connection(struct rpc_xprt *xprt, struct sock_xprt *transport)
1679 {
1680         unsigned int state = transport->inet->sk_state;
1681
1682         if (state == TCP_CLOSE && transport->sock->state == SS_UNCONNECTED)
1683                 return;
1684         if ((1 << state) & (TCPF_ESTABLISHED|TCPF_SYN_SENT))
1685                 return;
1686         xs_abort_connection(xprt, transport);
1687 }
1688
1689 static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
1690 {
1691         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1692
1693         if (!transport->inet) {
1694                 struct sock *sk = sock->sk;
1695
1696                 write_lock_bh(&sk->sk_callback_lock);
1697
1698                 xs_save_old_callbacks(transport, sk);
1699
1700                 sk->sk_user_data = xprt;
1701                 sk->sk_data_ready = xs_tcp_data_ready;
1702                 sk->sk_state_change = xs_tcp_state_change;
1703                 sk->sk_write_space = xs_tcp_write_space;
1704                 sk->sk_error_report = xs_error_report;
1705                 sk->sk_allocation = GFP_ATOMIC;
1706
1707                 /* socket options */
1708                 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
1709                 sock_reset_flag(sk, SOCK_LINGER);
1710                 tcp_sk(sk)->linger2 = 0;
1711                 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
1712
1713                 xprt_clear_connected(xprt);
1714
1715                 /* Reset to new socket */
1716                 transport->sock = sock;
1717                 transport->inet = sk;
1718
1719                 write_unlock_bh(&sk->sk_callback_lock);
1720         }
1721
1722         if (!xprt_bound(xprt))
1723                 return -ENOTCONN;
1724
1725         /* Tell the socket layer to start connecting... */
1726         xprt->stat.connect_count++;
1727         xprt->stat.connect_start = jiffies;
1728         return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
1729 }
1730
1731 /**
1732  * xs_tcp_connect_worker4 - connect a TCP socket to a remote endpoint
1733  * @work: RPC transport to connect
1734  *
1735  * Invoked by a work queue tasklet.
1736  */
1737 static void xs_tcp_connect_worker4(struct work_struct *work)
1738 {
1739         struct sock_xprt *transport =
1740                 container_of(work, struct sock_xprt, connect_worker.work);
1741         struct rpc_xprt *xprt = &transport->xprt;
1742         struct socket *sock = transport->sock;
1743         int err, status = -EIO;
1744
1745         if (xprt->shutdown)
1746                 goto out;
1747
1748         if (!sock) {
1749                 clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1750                 /* start from scratch */
1751                 if ((err = sock_create_kern(PF_INET, SOCK_STREAM, IPPROTO_TCP, &sock)) < 0) {
1752                         dprintk("RPC:       can't create TCP transport socket (%d).\n", -err);
1753                         goto out;
1754                 }
1755                 xs_reclassify_socket4(sock);
1756
1757                 if (xs_bind4(transport, sock) < 0) {
1758                         sock_release(sock);
1759                         goto out;
1760                 }
1761         } else {
1762                 int abort_and_exit;
1763
1764                 abort_and_exit = test_and_clear_bit(XPRT_CONNECTION_ABORT,
1765                                 &xprt->state);
1766                 /* "close" the socket, preserving the local port */
1767                 xs_tcp_reuse_connection(xprt, transport);
1768
1769                 if (abort_and_exit)
1770                         goto out_eagain;
1771         }
1772
1773         dprintk("RPC:       worker connecting xprt %p to address: %s\n",
1774                         xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1775
1776         status = xs_tcp_finish_connecting(xprt, sock);
1777         dprintk("RPC:       %p connect status %d connected %d sock state %d\n",
1778                         xprt, -status, xprt_connected(xprt),
1779                         sock->sk->sk_state);
1780         switch (status) {
1781         case -ECONNREFUSED:
1782         case -ECONNRESET:
1783         case -ENETUNREACH:
1784                 /* retry with existing socket, after a delay */
1785         case 0:
1786         case -EINPROGRESS:
1787         case -EALREADY:
1788                 xprt_clear_connecting(xprt);
1789                 return;
1790         }
1791         /* get rid of existing socket, and retry */
1792         xs_tcp_shutdown(xprt);
1793         printk("%s: connect returned unhandled error %d\n",
1794                         __func__, status);
1795 out_eagain:
1796         status = -EAGAIN;
1797 out:
1798         xprt_clear_connecting(xprt);
1799         xprt_wake_pending_tasks(xprt, status);
1800 }
1801
1802 /**
1803  * xs_tcp_connect_worker6 - connect a TCP socket to a remote endpoint
1804  * @work: RPC transport to connect
1805  *
1806  * Invoked by a work queue tasklet.
1807  */
1808 static void xs_tcp_connect_worker6(struct work_struct *work)
1809 {
1810         struct sock_xprt *transport =
1811                 container_of(work, struct sock_xprt, connect_worker.work);
1812         struct rpc_xprt *xprt = &transport->xprt;
1813         struct socket *sock = transport->sock;
1814         int err, status = -EIO;
1815
1816         if (xprt->shutdown)
1817                 goto out;
1818
1819         if (!sock) {
1820                 clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1821                 /* start from scratch */
1822                 if ((err = sock_create_kern(PF_INET6, SOCK_STREAM, IPPROTO_TCP, &sock)) < 0) {
1823                         dprintk("RPC:       can't create TCP transport socket (%d).\n", -err);
1824                         goto out;
1825                 }
1826                 xs_reclassify_socket6(sock);
1827
1828                 if (xs_bind6(transport, sock) < 0) {
1829                         sock_release(sock);
1830                         goto out;
1831                 }
1832         } else {
1833                 int abort_and_exit;
1834
1835                 abort_and_exit = test_and_clear_bit(XPRT_CONNECTION_ABORT,
1836                                 &xprt->state);
1837                 /* "close" the socket, preserving the local port */
1838                 xs_tcp_reuse_connection(xprt, transport);
1839
1840                 if (abort_and_exit)
1841                         goto out_eagain;
1842         }
1843
1844         dprintk("RPC:       worker connecting xprt %p to address: %s\n",
1845                         xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1846
1847         status = xs_tcp_finish_connecting(xprt, sock);
1848         dprintk("RPC:       %p connect status %d connected %d sock state %d\n",
1849                         xprt, -status, xprt_connected(xprt), sock->sk->sk_state);
1850         switch (status) {
1851         case -ECONNREFUSED:
1852         case -ECONNRESET:
1853         case -ENETUNREACH:
1854                 /* retry with existing socket, after a delay */
1855         case 0:
1856         case -EINPROGRESS:
1857         case -EALREADY:
1858                 xprt_clear_connecting(xprt);
1859                 return;
1860         }
1861         /* get rid of existing socket, and retry */
1862         xs_tcp_shutdown(xprt);
1863         printk("%s: connect returned unhandled error %d\n",
1864                         __func__, status);
1865 out_eagain:
1866         status = -EAGAIN;
1867 out:
1868         xprt_clear_connecting(xprt);
1869         xprt_wake_pending_tasks(xprt, status);
1870 }
1871
1872 /**
1873  * xs_connect - connect a socket to a remote endpoint
1874  * @task: address of RPC task that manages state of connect request
1875  *
1876  * TCP: If the remote end dropped the connection, delay reconnecting.
1877  *
1878  * UDP socket connects are synchronous, but we use a work queue anyway
1879  * to guarantee that even unprivileged user processes can set up a
1880  * socket on a privileged port.
1881  *
1882  * If a UDP socket connect fails, the delay behavior here prevents
1883  * retry floods (hard mounts).
1884  */
1885 static void xs_connect(struct rpc_task *task)
1886 {
1887         struct rpc_xprt *xprt = task->tk_xprt;
1888         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1889
1890         if (xprt_test_and_set_connecting(xprt))
1891                 return;
1892
1893         if (transport->sock != NULL) {
1894                 dprintk("RPC:       xs_connect delayed xprt %p for %lu "
1895                                 "seconds\n",
1896                                 xprt, xprt->reestablish_timeout / HZ);
1897                 queue_delayed_work(rpciod_workqueue,
1898                                    &transport->connect_worker,
1899                                    xprt->reestablish_timeout);
1900                 xprt->reestablish_timeout <<= 1;
1901                 if (xprt->reestablish_timeout > XS_TCP_MAX_REEST_TO)
1902                         xprt->reestablish_timeout = XS_TCP_MAX_REEST_TO;
1903         } else {
1904                 dprintk("RPC:       xs_connect scheduled xprt %p\n", xprt);
1905                 queue_delayed_work(rpciod_workqueue,
1906                                    &transport->connect_worker, 0);
1907         }
1908 }
1909
1910 static void xs_tcp_connect(struct rpc_task *task)
1911 {
1912         struct rpc_xprt *xprt = task->tk_xprt;
1913
1914         /* Exit if we need to wait for socket shutdown to complete */
1915         if (test_bit(XPRT_CLOSING, &xprt->state))
1916                 return;
1917         xs_connect(task);
1918 }
1919
1920 /**
1921  * xs_udp_print_stats - display UDP socket-specifc stats
1922  * @xprt: rpc_xprt struct containing statistics
1923  * @seq: output file
1924  *
1925  */
1926 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
1927 {
1928         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1929
1930         seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %Lu %Lu\n",
1931                         transport->port,
1932                         xprt->stat.bind_count,
1933                         xprt->stat.sends,
1934                         xprt->stat.recvs,
1935                         xprt->stat.bad_xids,
1936                         xprt->stat.req_u,
1937                         xprt->stat.bklog_u);
1938 }
1939
1940 /**
1941  * xs_tcp_print_stats - display TCP socket-specifc stats
1942  * @xprt: rpc_xprt struct containing statistics
1943  * @seq: output file
1944  *
1945  */
1946 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
1947 {
1948         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1949         long idle_time = 0;
1950
1951         if (xprt_connected(xprt))
1952                 idle_time = (long)(jiffies - xprt->last_used) / HZ;
1953
1954         seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu %Lu %Lu\n",
1955                         transport->port,
1956                         xprt->stat.bind_count,
1957                         xprt->stat.connect_count,
1958                         xprt->stat.connect_time,
1959                         idle_time,
1960                         xprt->stat.sends,
1961                         xprt->stat.recvs,
1962                         xprt->stat.bad_xids,
1963                         xprt->stat.req_u,
1964                         xprt->stat.bklog_u);
1965 }
1966
1967 static struct rpc_xprt_ops xs_udp_ops = {
1968         .set_buffer_size        = xs_udp_set_buffer_size,
1969         .reserve_xprt           = xprt_reserve_xprt_cong,
1970         .release_xprt           = xprt_release_xprt_cong,
1971         .rpcbind                = rpcb_getport_async,
1972         .set_port               = xs_set_port,
1973         .connect                = xs_connect,
1974         .buf_alloc              = rpc_malloc,
1975         .buf_free               = rpc_free,
1976         .send_request           = xs_udp_send_request,
1977         .set_retrans_timeout    = xprt_set_retrans_timeout_rtt,
1978         .timer                  = xs_udp_timer,
1979         .release_request        = xprt_release_rqst_cong,
1980         .close                  = xs_close,
1981         .destroy                = xs_destroy,
1982         .print_stats            = xs_udp_print_stats,
1983 };
1984
1985 static struct rpc_xprt_ops xs_tcp_ops = {
1986         .reserve_xprt           = xprt_reserve_xprt,
1987         .release_xprt           = xs_tcp_release_xprt,
1988         .rpcbind                = rpcb_getport_async,
1989         .set_port               = xs_set_port,
1990         .connect                = xs_tcp_connect,
1991         .buf_alloc              = rpc_malloc,
1992         .buf_free               = rpc_free,
1993         .send_request           = xs_tcp_send_request,
1994         .set_retrans_timeout    = xprt_set_retrans_timeout_def,
1995         .close                  = xs_tcp_shutdown,
1996         .destroy                = xs_destroy,
1997         .print_stats            = xs_tcp_print_stats,
1998 };
1999
2000 static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
2001                                       unsigned int slot_table_size)
2002 {
2003         struct rpc_xprt *xprt;
2004         struct sock_xprt *new;
2005
2006         if (args->addrlen > sizeof(xprt->addr)) {
2007                 dprintk("RPC:       xs_setup_xprt: address too large\n");
2008                 return ERR_PTR(-EBADF);
2009         }
2010
2011         new = kzalloc(sizeof(*new), GFP_KERNEL);
2012         if (new == NULL) {
2013                 dprintk("RPC:       xs_setup_xprt: couldn't allocate "
2014                                 "rpc_xprt\n");
2015                 return ERR_PTR(-ENOMEM);
2016         }
2017         xprt = &new->xprt;
2018
2019         xprt->max_reqs = slot_table_size;
2020         xprt->slot = kcalloc(xprt->max_reqs, sizeof(struct rpc_rqst), GFP_KERNEL);
2021         if (xprt->slot == NULL) {
2022                 kfree(xprt);
2023                 dprintk("RPC:       xs_setup_xprt: couldn't allocate slot "
2024                                 "table\n");
2025                 return ERR_PTR(-ENOMEM);
2026         }
2027
2028         memcpy(&xprt->addr, args->dstaddr, args->addrlen);
2029         xprt->addrlen = args->addrlen;
2030         if (args->srcaddr)
2031                 memcpy(&new->addr, args->srcaddr, args->addrlen);
2032
2033         return xprt;
2034 }
2035
2036 static const struct rpc_timeout xs_udp_default_timeout = {
2037         .to_initval = 5 * HZ,
2038         .to_maxval = 30 * HZ,
2039         .to_increment = 5 * HZ,
2040         .to_retries = 5,
2041 };
2042
2043 /**
2044  * xs_setup_udp - Set up transport to use a UDP socket
2045  * @args: rpc transport creation arguments
2046  *
2047  */
2048 static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
2049 {
2050         struct sockaddr *addr = args->dstaddr;
2051         struct rpc_xprt *xprt;
2052         struct sock_xprt *transport;
2053
2054         xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries);
2055         if (IS_ERR(xprt))
2056                 return xprt;
2057         transport = container_of(xprt, struct sock_xprt, xprt);
2058
2059         xprt->prot = IPPROTO_UDP;
2060         xprt->tsh_size = 0;
2061         /* XXX: header size can vary due to auth type, IPv6, etc. */
2062         xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
2063
2064         xprt->bind_timeout = XS_BIND_TO;
2065         xprt->connect_timeout = XS_UDP_CONN_TO;
2066         xprt->reestablish_timeout = XS_UDP_REEST_TO;
2067         xprt->idle_timeout = XS_IDLE_DISC_TO;
2068
2069         xprt->ops = &xs_udp_ops;
2070
2071         xprt->timeout = &xs_udp_default_timeout;
2072
2073         switch (addr->sa_family) {
2074         case AF_INET:
2075                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2076                         xprt_set_bound(xprt);
2077
2078                 INIT_DELAYED_WORK(&transport->connect_worker,
2079                                         xs_udp_connect_worker4);
2080                 xs_format_ipv4_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
2081                 break;
2082         case AF_INET6:
2083                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2084                         xprt_set_bound(xprt);
2085
2086                 INIT_DELAYED_WORK(&transport->connect_worker,
2087                                         xs_udp_connect_worker6);
2088                 xs_format_ipv6_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
2089                 break;
2090         default:
2091                 kfree(xprt);
2092                 return ERR_PTR(-EAFNOSUPPORT);
2093         }
2094
2095         dprintk("RPC:       set up transport to address %s\n",
2096                         xprt->address_strings[RPC_DISPLAY_ALL]);
2097
2098         if (try_module_get(THIS_MODULE))
2099                 return xprt;
2100
2101         kfree(xprt->slot);
2102         kfree(xprt);
2103         return ERR_PTR(-EINVAL);
2104 }
2105
2106 static const struct rpc_timeout xs_tcp_default_timeout = {
2107         .to_initval = 60 * HZ,
2108         .to_maxval = 60 * HZ,
2109         .to_retries = 2,
2110 };
2111
2112 /**
2113  * xs_setup_tcp - Set up transport to use a TCP socket
2114  * @args: rpc transport creation arguments
2115  *
2116  */
2117 static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
2118 {
2119         struct sockaddr *addr = args->dstaddr;
2120         struct rpc_xprt *xprt;
2121         struct sock_xprt *transport;
2122
2123         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries);
2124         if (IS_ERR(xprt))
2125                 return xprt;
2126         transport = container_of(xprt, struct sock_xprt, xprt);
2127
2128         xprt->prot = IPPROTO_TCP;
2129         xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2130         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2131
2132         xprt->bind_timeout = XS_BIND_TO;
2133         xprt->connect_timeout = XS_TCP_CONN_TO;
2134         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2135         xprt->idle_timeout = XS_IDLE_DISC_TO;
2136
2137         xprt->ops = &xs_tcp_ops;
2138         xprt->timeout = &xs_tcp_default_timeout;
2139
2140         switch (addr->sa_family) {
2141         case AF_INET:
2142                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2143                         xprt_set_bound(xprt);
2144
2145                 INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_connect_worker4);
2146                 xs_format_ipv4_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
2147                 break;
2148         case AF_INET6:
2149                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2150                         xprt_set_bound(xprt);
2151
2152                 INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_connect_worker6);
2153                 xs_format_ipv6_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
2154                 break;
2155         default:
2156                 kfree(xprt);
2157                 return ERR_PTR(-EAFNOSUPPORT);
2158         }
2159
2160         dprintk("RPC:       set up transport to address %s\n",
2161                         xprt->address_strings[RPC_DISPLAY_ALL]);
2162
2163         if (try_module_get(THIS_MODULE))
2164                 return xprt;
2165
2166         kfree(xprt->slot);
2167         kfree(xprt);
2168         return ERR_PTR(-EINVAL);
2169 }
2170
2171 static struct xprt_class        xs_udp_transport = {
2172         .list           = LIST_HEAD_INIT(xs_udp_transport.list),
2173         .name           = "udp",
2174         .owner          = THIS_MODULE,
2175         .ident          = IPPROTO_UDP,
2176         .setup          = xs_setup_udp,
2177 };
2178
2179 static struct xprt_class        xs_tcp_transport = {
2180         .list           = LIST_HEAD_INIT(xs_tcp_transport.list),
2181         .name           = "tcp",
2182         .owner          = THIS_MODULE,
2183         .ident          = IPPROTO_TCP,
2184         .setup          = xs_setup_tcp,
2185 };
2186
2187 /**
2188  * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
2189  *
2190  */
2191 int init_socket_xprt(void)
2192 {
2193 #ifdef RPC_DEBUG
2194         if (!sunrpc_table_header)
2195                 sunrpc_table_header = register_sysctl_table(sunrpc_table);
2196 #endif
2197
2198         xprt_register_transport(&xs_udp_transport);
2199         xprt_register_transport(&xs_tcp_transport);
2200
2201         return 0;
2202 }
2203
2204 /**
2205  * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
2206  *
2207  */
2208 void cleanup_socket_xprt(void)
2209 {
2210 #ifdef RPC_DEBUG
2211         if (sunrpc_table_header) {
2212                 unregister_sysctl_table(sunrpc_table_header);
2213                 sunrpc_table_header = NULL;
2214         }
2215 #endif
2216
2217         xprt_unregister_transport(&xs_udp_transport);
2218         xprt_unregister_transport(&xs_tcp_transport);
2219 }