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 /*
  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 void xs_nospace(struct rpc_task *task)
 525 {
 526         struct rpc_rqst *req = task->tk_rqstp;
 527         struct rpc_xprt *xprt = req->rq_xprt;
 528         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 529
 530         dprintk("RPC: %5u xmit incomplete (%u left of %u)\n",
 531                         task->tk_pid, req->rq_slen - req->rq_bytes_sent,
 532                         req->rq_slen);
 533
 534         /* Protect against races with write_space */
 535         spin_lock_bh(&xprt->transport_lock);
 536
 537         /* Don't race with disconnect */
 538         if (xprt_connected(xprt)) {
 539                 if (test_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags)) {
 540                         /*
 541                          * Notify TCP that we're limited by the application
 542                          * window size
 543                          */
 544                         set_bit(SOCK_NOSPACE, &transport->sock->flags);
 545                         transport->inet->sk_write_pending++;
 546                         /* ...and wait for more buffer space */
 547                         xprt_wait_for_buffer_space(task, xs_nospace_callback);
 548                 }
 549         } else {
 550                 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 551                 task->tk_status = -ENOTCONN;
 552         }
 553
 554         spin_unlock_bh(&xprt->transport_lock);
 555 }
 556
 557 /**
 558  * xs_udp_send_request - write an RPC request to a UDP socket
 559  * @task: address of RPC task that manages the state of an RPC request
 560  *
 561  * Return values:
 562  *        0:    The request has been sent
 563  *   EAGAIN:    The socket was blocked, please call again later to
 564  *              complete the request
 565  * ENOTCONN:    Caller needs to invoke connect logic then call again
 566  *    other:    Some other error occured, the request was not sent
 567  */
 568 static int xs_udp_send_request(struct rpc_task *task)
 569 {
 570         struct rpc_rqst *req = task->tk_rqstp;
 571         struct rpc_xprt *xprt = req->rq_xprt;
 572         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 573         struct xdr_buf *xdr = &req->rq_snd_buf;
 574         int status;
 575
 576         xs_pktdump("packet data:",
 577                                 req->rq_svec->iov_base,
 578                                 req->rq_svec->iov_len);
 579
 580         if (!xprt_bound(xprt))
 581                 return -ENOTCONN;
 582         status = xs_sendpages(transport->sock,
 583                               xs_addr(xprt),
 584                               xprt->addrlen, xdr,
 585                               req->rq_bytes_sent);
 586
 587         dprintk("RPC:       xs_udp_send_request(%u) = %d\n",
 588                         xdr->len - req->rq_bytes_sent, status);
 589
 590         if (status >= 0) {
 591                 task->tk_bytes_sent += status;
 592                 if (status >= req->rq_slen)
 593                         return 0;
 594                 /* Still some bytes left; set up for a retry later. */
 595                 status = -EAGAIN;
 596         }
 597
 598         switch (status) {
 599         case -ENOTSOCK:
 600                 status = -ENOTCONN;
 601                 /* Should we call xs_close() here? */
 602                 break;
 603         case -EAGAIN:
 604                 xs_nospace(task);
 605                 break;
 606         case -ENETUNREACH:
 607         case -EPIPE:
 608         case -ECONNREFUSED:
 609                 /* When the server has died, an ICMP port unreachable message
 610                  * prompts ECONNREFUSED. */
 611                 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 612                 break;
 613         default:
 614                 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 615                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
 616                         -status);
 617         }
 618
 619         return status;
 620 }
 621
 622 /**
 623  * xs_tcp_shutdown - gracefully shut down a TCP socket
 624  * @xprt: transport
 625  *
 626  * Initiates a graceful shutdown of the TCP socket by calling the
 627  * equivalent of shutdown(SHUT_WR);
 628  */
 629 static void xs_tcp_shutdown(struct rpc_xprt *xprt)
 630 {
 631         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 632         struct socket *sock = transport->sock;
 633
 634         if (sock != NULL)
 635                 kernel_sock_shutdown(sock, SHUT_WR);
 636 }
 637
 638 static inline void xs_encode_tcp_record_marker(struct xdr_buf *buf)
 639 {
 640         u32 reclen = buf->len - sizeof(rpc_fraghdr);
 641         rpc_fraghdr *base = buf->head[0].iov_base;
 642         *base = htonl(RPC_LAST_STREAM_FRAGMENT | reclen);
 643 }
 644
 645 /**
 646  * xs_tcp_send_request - write an RPC request to a TCP socket
 647  * @task: address of RPC task that manages the state of an RPC request
 648  *
 649  * Return values:
 650  *        0:    The request has been sent
 651  *   EAGAIN:    The socket was blocked, please call again later to
 652  *              complete the request
 653  * ENOTCONN:    Caller needs to invoke connect logic then call again
 654  *    other:    Some other error occured, the request was not sent
 655  *
 656  * XXX: In the case of soft timeouts, should we eventually give up
 657  *      if sendmsg is not able to make progress?
 658  */
 659 static int xs_tcp_send_request(struct rpc_task *task)
 660 {
 661         struct rpc_rqst *req = task->tk_rqstp;
 662         struct rpc_xprt *xprt = req->rq_xprt;
 663         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 664         struct xdr_buf *xdr = &req->rq_snd_buf;
 665         int status;
 666
 667         xs_encode_tcp_record_marker(&req->rq_snd_buf);
 668
 669         xs_pktdump("packet data:",
 670                                 req->rq_svec->iov_base,
 671                                 req->rq_svec->iov_len);
 672
 673         /* Continue transmitting the packet/record. We must be careful
 674          * to cope with writespace callbacks arriving _after_ we have
 675          * called sendmsg(). */
 676         while (1) {
 677                 status = xs_sendpages(transport->sock,
 678                                         NULL, 0, xdr, req->rq_bytes_sent);
 679
 680                 dprintk("RPC:       xs_tcp_send_request(%u) = %d\n",
 681                                 xdr->len - req->rq_bytes_sent, status);
 682
 683                 if (unlikely(status < 0))
 684                         break;
 685
 686                 /* If we've sent the entire packet, immediately
 687                  * reset the count of bytes sent. */
 688                 req->rq_bytes_sent += status;
 689                 task->tk_bytes_sent += status;
 690                 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
 691                         req->rq_bytes_sent = 0;
 692                         return 0;
 693                 }
 694
 695                 if (status != 0)
 696                         continue;
 697                 status = -EAGAIN;
 698                 break;
 699         }
 700
 701         switch (status) {
 702         case -ENOTSOCK:
 703                 status = -ENOTCONN;
 704                 /* Should we call xs_close() here? */
 705                 break;
 706         case -EAGAIN:
 707                 xs_nospace(task);
 708                 break;
 709         case -ECONNRESET:
 710                 xs_tcp_shutdown(xprt);
 711         case -ECONNREFUSED:
 712         case -ENOTCONN:
 713         case -EPIPE:
 714                 status = -ENOTCONN;
 715                 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 716                 break;
 717         default:
 718                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
 719                         -status);
 720                 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 721                 xs_tcp_shutdown(xprt);
 722         }
 723
 724         return status;
 725 }
 726
 727 /**
 728  * xs_tcp_release_xprt - clean up after a tcp transmission
 729  * @xprt: transport
 730  * @task: rpc task
 731  *
 732  * This cleans up if an error causes us to abort the transmission of a request.
 733  * In this case, the socket may need to be reset in order to avoid confusing
 734  * the server.
 735  */
 736 static void xs_tcp_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
 737 {
 738         struct rpc_rqst *req;
 739
 740         if (task != xprt->snd_task)
 741                 return;
 742         if (task == NULL)
 743                 goto out_release;
 744         req = task->tk_rqstp;
 745         if (req->rq_bytes_sent == 0)
 746                 goto out_release;
 747         if (req->rq_bytes_sent == req->rq_snd_buf.len)
 748                 goto out_release;
 749         set_bit(XPRT_CLOSE_WAIT, &task->tk_xprt->state);
 750 out_release:
 751         xprt_release_xprt(xprt, task);
 752 }
 753
 754 static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk)
 755 {
 756         transport->old_data_ready = sk->sk_data_ready;
 757         transport->old_state_change = sk->sk_state_change;
 758         transport->old_write_space = sk->sk_write_space;
 759         transport->old_error_report = sk->sk_error_report;
 760 }
 761
 762 static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk)
 763 {
 764         sk->sk_data_ready = transport->old_data_ready;
 765         sk->sk_state_change = transport->old_state_change;
 766         sk->sk_write_space = transport->old_write_space;
 767         sk->sk_error_report = transport->old_error_report;
 768 }
 769
 770 static void xs_reset_transport(struct sock_xprt *transport)
 771 {
 772         struct socket *sock = transport->sock;
 773         struct sock *sk = transport->inet;
 774
 775         if (sk == NULL)
 776                 return;
 777
 778         write_lock_bh(&sk->sk_callback_lock);
 779         transport->inet = NULL;
 780         transport->sock = NULL;
 781
 782         sk->sk_user_data = NULL;
 783
 784         xs_restore_old_callbacks(transport, sk);
 785         write_unlock_bh(&sk->sk_callback_lock);
 786
 787         sk->sk_no_check = 0;
 788
 789         sock_release(sock);
 790 }
 791
 792 /**
 793  * xs_close - close a socket
 794  * @xprt: transport
 795  *
 796  * This is used when all requests are complete; ie, no DRC state remains
 797  * on the server we want to save.
 798  */
 799 static void xs_close(struct rpc_xprt *xprt)
 800 {
 801         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 802
 803         dprintk("RPC:       xs_close xprt %p\n", xprt);
 804
 805         xs_reset_transport(transport);
 806
 807         smp_mb__before_clear_bit();
 808         clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
 809         clear_bit(XPRT_CLOSING, &xprt->state);
 810         smp_mb__after_clear_bit();
 811         xprt_disconnect_done(xprt);
 812 }
 813
 814 /**
 815  * xs_destroy - prepare to shutdown a transport
 816  * @xprt: doomed transport
 817  *
 818  */
 819 static void xs_destroy(struct rpc_xprt *xprt)
 820 {
 821         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 822
 823         dprintk("RPC:       xs_destroy xprt %p\n", xprt);
 824
 825         cancel_rearming_delayed_work(&transport->connect_worker);
 826
 827         xs_close(xprt);
 828         xs_free_peer_addresses(xprt);
 829         kfree(xprt->slot);
 830         kfree(xprt);
 831         module_put(THIS_MODULE);
 832 }
 833
 834 static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
 835 {
 836         return (struct rpc_xprt *) sk->sk_user_data;
 837 }
 838
 839 /**
 840  * xs_udp_data_ready - "data ready" callback for UDP sockets
 841  * @sk: socket with data to read
 842  * @len: how much data to read
 843  *
 844  */
 845 static void xs_udp_data_ready(struct sock *sk, int len)
 846 {
 847         struct rpc_task *task;
 848         struct rpc_xprt *xprt;
 849         struct rpc_rqst *rovr;
 850         struct sk_buff *skb;
 851         int err, repsize, copied;
 852         u32 _xid;
 853         __be32 *xp;
 854
 855         read_lock(&sk->sk_callback_lock);
 856         dprintk("RPC:       xs_udp_data_ready...\n");
 857         if (!(xprt = xprt_from_sock(sk)))
 858                 goto out;
 859
 860         if ((skb = skb_recv_datagram(sk, 0, 1, &err)) == NULL)
 861                 goto out;
 862
 863         if (xprt->shutdown)
 864                 goto dropit;
 865
 866         repsize = skb->len - sizeof(struct udphdr);
 867         if (repsize < 4) {
 868                 dprintk("RPC:       impossible RPC reply size %d!\n", repsize);
 869                 goto dropit;
 870         }
 871
 872         /* Copy the XID from the skb... */
 873         xp = skb_header_pointer(skb, sizeof(struct udphdr),
 874                                 sizeof(_xid), &_xid);
 875         if (xp == NULL)
 876                 goto dropit;
 877
 878         /* Look up and lock the request corresponding to the given XID */
 879         spin_lock(&xprt->transport_lock);
 880         rovr = xprt_lookup_rqst(xprt, *xp);
 881         if (!rovr)
 882                 goto out_unlock;
 883         task = rovr->rq_task;
 884
 885         if ((copied = rovr->rq_private_buf.buflen) > repsize)
 886                 copied = repsize;
 887
 888         /* Suck it into the iovec, verify checksum if not done by hw. */
 889         if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
 890                 UDPX_INC_STATS_BH(sk, UDP_MIB_INERRORS);
 891                 goto out_unlock;
 892         }
 893
 894         UDPX_INC_STATS_BH(sk, UDP_MIB_INDATAGRAMS);
 895
 896         /* Something worked... */
 897         dst_confirm(skb->dst);
 898
 899         xprt_adjust_cwnd(task, copied);
 900         xprt_update_rtt(task);
 901         xprt_complete_rqst(task, copied);
 902
 903  out_unlock:
 904         spin_unlock(&xprt->transport_lock);
 905  dropit:
 906         skb_free_datagram(sk, skb);
 907  out:
 908         read_unlock(&sk->sk_callback_lock);
 909 }
 910
 911 static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
 912 {
 913         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 914         size_t len, used;
 915         char *p;
 916
 917         p = ((char *) &transport->tcp_fraghdr) + transport->tcp_offset;
 918         len = sizeof(transport->tcp_fraghdr) - transport->tcp_offset;
 919         used = xdr_skb_read_bits(desc, p, len);
 920         transport->tcp_offset += used;
 921         if (used != len)
 922                 return;
 923
 924         transport->tcp_reclen = ntohl(transport->tcp_fraghdr);
 925         if (transport->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
 926                 transport->tcp_flags |= TCP_RCV_LAST_FRAG;
 927         else
 928                 transport->tcp_flags &= ~TCP_RCV_LAST_FRAG;
 929         transport->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
 930
 931         transport->tcp_flags &= ~TCP_RCV_COPY_FRAGHDR;
 932         transport->tcp_offset = 0;
 933
 934         /* Sanity check of the record length */
 935         if (unlikely(transport->tcp_reclen < 4)) {
 936                 dprintk("RPC:       invalid TCP record fragment length\n");
 937                 xprt_force_disconnect(xprt);
 938                 return;
 939         }
 940         dprintk("RPC:       reading TCP record fragment of length %d\n",
 941                         transport->tcp_reclen);
 942 }
 943
 944 static void xs_tcp_check_fraghdr(struct sock_xprt *transport)
 945 {
 946         if (transport->tcp_offset == transport->tcp_reclen) {
 947                 transport->tcp_flags |= TCP_RCV_COPY_FRAGHDR;
 948                 transport->tcp_offset = 0;
 949                 if (transport->tcp_flags & TCP_RCV_LAST_FRAG) {
 950                         transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
 951                         transport->tcp_flags |= TCP_RCV_COPY_XID;
 952                         transport->tcp_copied = 0;
 953                 }
 954         }
 955 }
 956
 957 static inline void xs_tcp_read_xid(struct sock_xprt *transport, struct xdr_skb_reader *desc)
 958 {
 959         size_t len, used;
 960         char *p;
 961
 962         len = sizeof(transport->tcp_xid) - transport->tcp_offset;
 963         dprintk("RPC:       reading XID (%Zu bytes)\n", len);
 964         p = ((char *) &transport->tcp_xid) + transport->tcp_offset;
 965         used = xdr_skb_read_bits(desc, p, len);
 966         transport->tcp_offset += used;
 967         if (used != len)
 968                 return;
 969         transport->tcp_flags &= ~TCP_RCV_COPY_XID;
 970         transport->tcp_flags |= TCP_RCV_COPY_DATA;
 971         transport->tcp_copied = 4;
 972         dprintk("RPC:       reading reply for XID %08x\n",
 973                         ntohl(transport->tcp_xid));
 974         xs_tcp_check_fraghdr(transport);
 975 }
 976
 977 static inline void xs_tcp_read_request(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
 978 {
 979         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 980         struct rpc_rqst *req;
 981         struct xdr_buf *rcvbuf;
 982         size_t len;
 983         ssize_t r;
 984
 985         /* Find and lock the request corresponding to this xid */
 986         spin_lock(&xprt->transport_lock);
 987         req = xprt_lookup_rqst(xprt, transport->tcp_xid);
 988         if (!req) {
 989                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
 990                 dprintk("RPC:       XID %08x request not found!\n",
 991                                 ntohl(transport->tcp_xid));
 992                 spin_unlock(&xprt->transport_lock);
 993                 return;
 994         }
 995
 996         rcvbuf = &req->rq_private_buf;
 997         len = desc->count;
 998         if (len > transport->tcp_reclen - transport->tcp_offset) {
 999                 struct xdr_skb_reader my_desc;
1000
1001                 len = transport->tcp_reclen - transport->tcp_offset;
1002                 memcpy(&my_desc, desc, sizeof(my_desc));
1003                 my_desc.count = len;
1004                 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
1005                                           &my_desc, xdr_skb_read_bits);
1006                 desc->count -= r;
1007                 desc->offset += r;
1008         } else
1009                 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
1010                                           desc, xdr_skb_read_bits);
1011
1012         if (r > 0) {
1013                 transport->tcp_copied += r;
1014                 transport->tcp_offset += r;
1015         }
1016         if (r != len) {
1017                 /* Error when copying to the receive buffer,
1018                  * usually because we weren't able to allocate
1019                  * additional buffer pages. All we can do now
1020                  * is turn off TCP_RCV_COPY_DATA, so the request
1021                  * will not receive any additional updates,
1022                  * and time out.
1023                  * Any remaining data from this record will
1024                  * be discarded.
1025                  */
1026                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1027                 dprintk("RPC:       XID %08x truncated request\n",
1028                                 ntohl(transport->tcp_xid));
1029                 dprintk("RPC:       xprt = %p, tcp_copied = %lu, "
1030                                 "tcp_offset = %u, tcp_reclen = %u\n",
1031                                 xprt, transport->tcp_copied,
1032                                 transport->tcp_offset, transport->tcp_reclen);
1033                 goto out;
1034         }
1035
1036         dprintk("RPC:       XID %08x read %Zd bytes\n",
1037                         ntohl(transport->tcp_xid), r);
1038         dprintk("RPC:       xprt = %p, tcp_copied = %lu, tcp_offset = %u, "
1039                         "tcp_reclen = %u\n", xprt, transport->tcp_copied,
1040                         transport->tcp_offset, transport->tcp_reclen);
1041
1042         if (transport->tcp_copied == req->rq_private_buf.buflen)
1043                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1044         else if (transport->tcp_offset == transport->tcp_reclen) {
1045                 if (transport->tcp_flags & TCP_RCV_LAST_FRAG)
1046                         transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1047         }
1048
1049 out:
1050         if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1051                 xprt_complete_rqst(req->rq_task, transport->tcp_copied);
1052         spin_unlock(&xprt->transport_lock);
1053         xs_tcp_check_fraghdr(transport);
1054 }
1055
1056 static inline void xs_tcp_read_discard(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1057 {
1058         size_t len;
1059
1060         len = transport->tcp_reclen - transport->tcp_offset;
1061         if (len > desc->count)
1062                 len = desc->count;
1063         desc->count -= len;
1064         desc->offset += len;
1065         transport->tcp_offset += len;
1066         dprintk("RPC:       discarded %Zu bytes\n", len);
1067         xs_tcp_check_fraghdr(transport);
1068 }
1069
1070 static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
1071 {
1072         struct rpc_xprt *xprt = rd_desc->arg.data;
1073         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1074         struct xdr_skb_reader desc = {
1075                 .skb    = skb,
1076                 .offset = offset,
1077                 .count  = len,
1078         };
1079
1080         dprintk("RPC:       xs_tcp_data_recv started\n");
1081         do {
1082                 /* Read in a new fragment marker if necessary */
1083                 /* Can we ever really expect to get completely empty fragments? */
1084                 if (transport->tcp_flags & TCP_RCV_COPY_FRAGHDR) {
1085                         xs_tcp_read_fraghdr(xprt, &desc);
1086                         continue;
1087                 }
1088                 /* Read in the xid if necessary */
1089                 if (transport->tcp_flags & TCP_RCV_COPY_XID) {
1090                         xs_tcp_read_xid(transport, &desc);
1091                         continue;
1092                 }
1093                 /* Read in the request data */
1094                 if (transport->tcp_flags & TCP_RCV_COPY_DATA) {
1095                         xs_tcp_read_request(xprt, &desc);
1096                         continue;
1097                 }
1098                 /* Skip over any trailing bytes on short reads */
1099                 xs_tcp_read_discard(transport, &desc);
1100         } while (desc.count);
1101         dprintk("RPC:       xs_tcp_data_recv done\n");
1102         return len - desc.count;
1103 }
1104
1105 /**
1106  * xs_tcp_data_ready - "data ready" callback for TCP sockets
1107  * @sk: socket with data to read
1108  * @bytes: how much data to read
1109  *
1110  */
1111 static void xs_tcp_data_ready(struct sock *sk, int bytes)
1112 {
1113         struct rpc_xprt *xprt;
1114         read_descriptor_t rd_desc;
1115         int read;
1116
1117         dprintk("RPC:       xs_tcp_data_ready...\n");
1118
1119         read_lock(&sk->sk_callback_lock);
1120         if (!(xprt = xprt_from_sock(sk)))
1121                 goto out;
1122         if (xprt->shutdown)
1123                 goto out;
1124
1125         /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
1126         rd_desc.arg.data = xprt;
1127         do {
1128                 rd_desc.count = 65536;
1129                 read = tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
1130         } while (read > 0);
1131 out:
1132         read_unlock(&sk->sk_callback_lock);
1133 }
1134
1135 /**
1136  * xs_tcp_state_change - callback to handle TCP socket state changes
1137  * @sk: socket whose state has changed
1138  *
1139  */
1140 static void xs_tcp_state_change(struct sock *sk)
1141 {
1142         struct rpc_xprt *xprt;
1143
1144         read_lock(&sk->sk_callback_lock);
1145         if (!(xprt = xprt_from_sock(sk)))
1146                 goto out;
1147         dprintk("RPC:       xs_tcp_state_change client %p...\n", xprt);
1148         dprintk("RPC:       state %x conn %d dead %d zapped %d\n",
1149                         sk->sk_state, xprt_connected(xprt),
1150                         sock_flag(sk, SOCK_DEAD),
1151                         sock_flag(sk, SOCK_ZAPPED));
1152
1153         switch (sk->sk_state) {
1154         case TCP_ESTABLISHED:
1155                 spin_lock_bh(&xprt->transport_lock);
1156                 if (!xprt_test_and_set_connected(xprt)) {
1157                         struct sock_xprt *transport = container_of(xprt,
1158                                         struct sock_xprt, xprt);
1159
1160                         /* Reset TCP record info */
1161                         transport->tcp_offset = 0;
1162                         transport->tcp_reclen = 0;
1163                         transport->tcp_copied = 0;
1164                         transport->tcp_flags =
1165                                 TCP_RCV_COPY_FRAGHDR | TCP_RCV_COPY_XID;
1166
1167                         xprt_wake_pending_tasks(xprt, 0);
1168                 }
1169                 spin_unlock_bh(&xprt->transport_lock);
1170                 break;
1171         case TCP_FIN_WAIT1:
1172                 /* The client initiated a shutdown of the socket */
1173                 xprt->connect_cookie++;
1174                 xprt->reestablish_timeout = 0;
1175                 set_bit(XPRT_CLOSING, &xprt->state);
1176                 smp_mb__before_clear_bit();
1177                 clear_bit(XPRT_CONNECTED, &xprt->state);
1178                 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1179                 smp_mb__after_clear_bit();
1180                 break;
1181         case TCP_CLOSE_WAIT:
1182                 /* The server initiated a shutdown of the socket */
1183                 xprt_force_disconnect(xprt);
1184         case TCP_SYN_SENT:
1185                 xprt->connect_cookie++;
1186         case TCP_CLOSING:
1187                 /*
1188                  * If the server closed down the connection, make sure that
1189                  * we back off before reconnecting
1190                  */
1191                 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1192                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1193                 break;
1194         case TCP_LAST_ACK:
1195                 set_bit(XPRT_CLOSING, &xprt->state);
1196                 smp_mb__before_clear_bit();
1197                 clear_bit(XPRT_CONNECTED, &xprt->state);
1198                 smp_mb__after_clear_bit();
1199                 break;
1200         case TCP_CLOSE:
1201                 smp_mb__before_clear_bit();
1202                 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1203                 clear_bit(XPRT_CLOSING, &xprt->state);
1204                 smp_mb__after_clear_bit();
1205                 /* Mark transport as closed and wake up all pending tasks */
1206                 xprt_disconnect_done(xprt);
1207         }
1208  out:
1209         read_unlock(&sk->sk_callback_lock);
1210 }
1211
1212 /**
1213  * xs_tcp_error_report - callback mainly for catching RST events
1214  * @sk: socket
1215  */
1216 static void xs_tcp_error_report(struct sock *sk)
1217 {
1218         struct rpc_xprt *xprt;
1219
1220         read_lock(&sk->sk_callback_lock);
1221         if (sk->sk_err != ECONNRESET || sk->sk_state != TCP_ESTABLISHED)
1222                 goto out;
1223         if (!(xprt = xprt_from_sock(sk)))
1224                 goto out;
1225         dprintk("RPC:       %s client %p...\n"
1226                         "RPC:       error %d\n",
1227                         __func__, xprt, sk->sk_err);
1228
1229         xprt_force_disconnect(xprt);
1230 out:
1231         read_unlock(&sk->sk_callback_lock);
1232 }
1233
1234 /**
1235  * xs_udp_write_space - callback invoked when socket buffer space
1236  *                             becomes available
1237  * @sk: socket whose state has changed
1238  *
1239  * Called when more output buffer space is available for this socket.
1240  * We try not to wake our writers until they can make "significant"
1241  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1242  * with a bunch of small requests.
1243  */
1244 static void xs_udp_write_space(struct sock *sk)
1245 {
1246         read_lock(&sk->sk_callback_lock);
1247
1248         /* from net/core/sock.c:sock_def_write_space */
1249         if (sock_writeable(sk)) {
1250                 struct socket *sock;
1251                 struct rpc_xprt *xprt;
1252
1253                 if (unlikely(!(sock = sk->sk_socket)))
1254                         goto out;
1255                 clear_bit(SOCK_NOSPACE, &sock->flags);
1256
1257                 if (unlikely(!(xprt = xprt_from_sock(sk))))
1258                         goto out;
1259                 if (test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags) == 0)
1260                         goto out;
1261
1262                 xprt_write_space(xprt);
1263         }
1264
1265  out:
1266         read_unlock(&sk->sk_callback_lock);
1267 }
1268
1269 /**
1270  * xs_tcp_write_space - callback invoked when socket buffer space
1271  *                             becomes available
1272  * @sk: socket whose state has changed
1273  *
1274  * Called when more output buffer space is available for this socket.
1275  * We try not to wake our writers until they can make "significant"
1276  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1277  * with a bunch of small requests.
1278  */
1279 static void xs_tcp_write_space(struct sock *sk)
1280 {
1281         read_lock(&sk->sk_callback_lock);
1282
1283         /* from net/core/stream.c:sk_stream_write_space */
1284         if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk)) {
1285                 struct socket *sock;
1286                 struct rpc_xprt *xprt;
1287
1288                 if (unlikely(!(sock = sk->sk_socket)))
1289                         goto out;
1290                 clear_bit(SOCK_NOSPACE, &sock->flags);
1291
1292                 if (unlikely(!(xprt = xprt_from_sock(sk))))
1293                         goto out;
1294                 if (test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags) == 0)
1295                         goto out;
1296
1297                 xprt_write_space(xprt);
1298         }
1299
1300  out:
1301         read_unlock(&sk->sk_callback_lock);
1302 }
1303
1304 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
1305 {
1306         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1307         struct sock *sk = transport->inet;
1308
1309         if (transport->rcvsize) {
1310                 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1311                 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
1312         }
1313         if (transport->sndsize) {
1314                 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1315                 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1316                 sk->sk_write_space(sk);
1317         }
1318 }
1319
1320 /**
1321  * xs_udp_set_buffer_size - set send and receive limits
1322  * @xprt: generic transport
1323  * @sndsize: requested size of send buffer, in bytes
1324  * @rcvsize: requested size of receive buffer, in bytes
1325  *
1326  * Set socket send and receive buffer size limits.
1327  */
1328 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1329 {
1330         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1331
1332         transport->sndsize = 0;
1333         if (sndsize)
1334                 transport->sndsize = sndsize + 1024;
1335         transport->rcvsize = 0;
1336         if (rcvsize)
1337                 transport->rcvsize = rcvsize + 1024;
1338
1339         xs_udp_do_set_buffer_size(xprt);
1340 }
1341
1342 /**
1343  * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1344  * @task: task that timed out
1345  *
1346  * Adjust the congestion window after a retransmit timeout has occurred.
1347  */
1348 static void xs_udp_timer(struct rpc_task *task)
1349 {
1350         xprt_adjust_cwnd(task, -ETIMEDOUT);
1351 }
1352
1353 static unsigned short xs_get_random_port(void)
1354 {
1355         unsigned short range = xprt_max_resvport - xprt_min_resvport;
1356         unsigned short rand = (unsigned short) net_random() % range;
1357         return rand + xprt_min_resvport;
1358 }
1359
1360 /**
1361  * xs_set_port - reset the port number in the remote endpoint address
1362  * @xprt: generic transport
1363  * @port: new port number
1364  *
1365  */
1366 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1367 {
1368         struct sockaddr *addr = xs_addr(xprt);
1369
1370         dprintk("RPC:       setting port for xprt %p to %u\n", xprt, port);
1371
1372         switch (addr->sa_family) {
1373         case AF_INET:
1374                 ((struct sockaddr_in *)addr)->sin_port = htons(port);
1375                 break;
1376         case AF_INET6:
1377                 ((struct sockaddr_in6 *)addr)->sin6_port = htons(port);
1378                 break;
1379         default:
1380                 BUG();
1381         }
1382 }
1383
1384 static unsigned short xs_get_srcport(struct sock_xprt *transport, struct socket *sock)
1385 {
1386         unsigned short port = transport->port;
1387
1388         if (port == 0 && transport->xprt.resvport)
1389                 port = xs_get_random_port();
1390         return port;
1391 }
1392
1393 static unsigned short xs_next_srcport(struct sock_xprt *transport, struct socket *sock, unsigned short port)
1394 {
1395         if (transport->port != 0)
1396                 transport->port = 0;
1397         if (!transport->xprt.resvport)
1398                 return 0;
1399         if (port <= xprt_min_resvport || port > xprt_max_resvport)
1400                 return xprt_max_resvport;
1401         return --port;
1402 }
1403
1404 static int xs_bind4(struct sock_xprt *transport, struct socket *sock)
1405 {
1406         struct sockaddr_in myaddr = {
1407                 .sin_family = AF_INET,
1408         };
1409         struct sockaddr_in *sa;
1410         int err, nloop = 0;
1411         unsigned short port = xs_get_srcport(transport, sock);
1412         unsigned short last;
1413
1414         sa = (struct sockaddr_in *)&transport->addr;
1415         myaddr.sin_addr = sa->sin_addr;
1416         do {
1417                 myaddr.sin_port = htons(port);
1418                 err = kernel_bind(sock, (struct sockaddr *) &myaddr,
1419                                                 sizeof(myaddr));
1420                 if (port == 0)
1421                         break;
1422                 if (err == 0) {
1423                         transport->port = port;
1424                         break;
1425                 }
1426                 last = port;
1427                 port = xs_next_srcport(transport, sock, port);
1428                 if (port > last)
1429                         nloop++;
1430         } while (err == -EADDRINUSE && nloop != 2);
1431         dprintk("RPC:       %s %pI4:%u: %s (%d)\n",
1432                         __func__, &myaddr.sin_addr,
1433                         port, err ? "failed" : "ok", err);
1434         return err;
1435 }
1436
1437 static int xs_bind6(struct sock_xprt *transport, struct socket *sock)
1438 {
1439         struct sockaddr_in6 myaddr = {
1440                 .sin6_family = AF_INET6,
1441         };
1442         struct sockaddr_in6 *sa;
1443         int err, nloop = 0;
1444         unsigned short port = xs_get_srcport(transport, sock);
1445         unsigned short last;
1446
1447         sa = (struct sockaddr_in6 *)&transport->addr;
1448         myaddr.sin6_addr = sa->sin6_addr;
1449         do {
1450                 myaddr.sin6_port = htons(port);
1451                 err = kernel_bind(sock, (struct sockaddr *) &myaddr,
1452                                                 sizeof(myaddr));
1453                 if (port == 0)
1454                         break;
1455                 if (err == 0) {
1456                         transport->port = port;
1457                         break;
1458                 }
1459                 last = port;
1460                 port = xs_next_srcport(transport, sock, port);
1461                 if (port > last)
1462                         nloop++;
1463         } while (err == -EADDRINUSE && nloop != 2);
1464         dprintk("RPC:       xs_bind6 %pI6:%u: %s (%d)\n",
1465                 &myaddr.sin6_addr, port, err ? "failed" : "ok", err);
1466         return err;
1467 }
1468
1469 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1470 static struct lock_class_key xs_key[2];
1471 static struct lock_class_key xs_slock_key[2];
1472
1473 static inline void xs_reclassify_socket4(struct socket *sock)
1474 {
1475         struct sock *sk = sock->sk;
1476
1477         BUG_ON(sock_owned_by_user(sk));
1478         sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1479                 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1480 }
1481
1482 static inline void xs_reclassify_socket6(struct socket *sock)
1483 {
1484         struct sock *sk = sock->sk;
1485
1486         BUG_ON(sock_owned_by_user(sk));
1487         sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1488                 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
1489 }
1490 #else
1491 static inline void xs_reclassify_socket4(struct socket *sock)
1492 {
1493 }
1494
1495 static inline void xs_reclassify_socket6(struct socket *sock)
1496 {
1497 }
1498 #endif
1499
1500 static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
1501 {
1502         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1503
1504         if (!transport->inet) {
1505                 struct sock *sk = sock->sk;
1506
1507                 write_lock_bh(&sk->sk_callback_lock);
1508
1509                 xs_save_old_callbacks(transport, sk);
1510
1511                 sk->sk_user_data = xprt;
1512                 sk->sk_data_ready = xs_udp_data_ready;
1513                 sk->sk_write_space = xs_udp_write_space;
1514                 sk->sk_no_check = UDP_CSUM_NORCV;
1515                 sk->sk_allocation = GFP_ATOMIC;
1516
1517                 xprt_set_connected(xprt);
1518
1519                 /* Reset to new socket */
1520                 transport->sock = sock;
1521                 transport->inet = sk;
1522
1523                 write_unlock_bh(&sk->sk_callback_lock);
1524         }
1525         xs_udp_do_set_buffer_size(xprt);
1526 }
1527
1528 /**
1529  * xs_udp_connect_worker4 - set up a UDP socket
1530  * @work: RPC transport to connect
1531  *
1532  * Invoked by a work queue tasklet.
1533  */
1534 static void xs_udp_connect_worker4(struct work_struct *work)
1535 {
1536         struct sock_xprt *transport =
1537                 container_of(work, struct sock_xprt, connect_worker.work);
1538         struct rpc_xprt *xprt = &transport->xprt;
1539         struct socket *sock = transport->sock;
1540         int err, status = -EIO;
1541
1542         if (xprt->shutdown)
1543                 goto out;
1544
1545         /* Start by resetting any existing state */
1546         xs_reset_transport(transport);
1547
1548         err = sock_create_kern(PF_INET, SOCK_DGRAM, IPPROTO_UDP, &sock);
1549         if (err < 0) {
1550                 dprintk("RPC:       can't create UDP transport socket (%d).\n", -err);
1551                 goto out;
1552         }
1553         xs_reclassify_socket4(sock);
1554
1555         if (xs_bind4(transport, sock)) {
1556                 sock_release(sock);
1557                 goto out;
1558         }
1559
1560         dprintk("RPC:       worker connecting xprt %p to address: %s\n",
1561                         xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1562
1563         xs_udp_finish_connecting(xprt, sock);
1564         status = 0;
1565 out:
1566         xprt_wake_pending_tasks(xprt, status);
1567         xprt_clear_connecting(xprt);
1568 }
1569
1570 /**
1571  * xs_udp_connect_worker6 - set up a UDP socket
1572  * @work: RPC transport to connect
1573  *
1574  * Invoked by a work queue tasklet.
1575  */
1576 static void xs_udp_connect_worker6(struct work_struct *work)
1577 {
1578         struct sock_xprt *transport =
1579                 container_of(work, struct sock_xprt, connect_worker.work);
1580         struct rpc_xprt *xprt = &transport->xprt;
1581         struct socket *sock = transport->sock;
1582         int err, status = -EIO;
1583
1584         if (xprt->shutdown)
1585                 goto out;
1586
1587         /* Start by resetting any existing state */
1588         xs_reset_transport(transport);
1589
1590         err = sock_create_kern(PF_INET6, SOCK_DGRAM, IPPROTO_UDP, &sock);
1591         if (err < 0) {
1592                 dprintk("RPC:       can't create UDP transport socket (%d).\n", -err);
1593                 goto out;
1594         }
1595         xs_reclassify_socket6(sock);
1596
1597         if (xs_bind6(transport, sock) < 0) {
1598                 sock_release(sock);
1599                 goto out;
1600         }
1601
1602         dprintk("RPC:       worker connecting xprt %p to address: %s\n",
1603                         xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1604
1605         xs_udp_finish_connecting(xprt, sock);
1606         status = 0;
1607 out:
1608         xprt_wake_pending_tasks(xprt, status);
1609         xprt_clear_connecting(xprt);
1610 }
1611
1612 /*
1613  * We need to preserve the port number so the reply cache on the server can
1614  * find our cached RPC replies when we get around to reconnecting.
1615  */
1616 static void xs_abort_connection(struct rpc_xprt *xprt, struct sock_xprt *transport)
1617 {
1618         int result;
1619         struct sockaddr any;
1620
1621         dprintk("RPC:       disconnecting xprt %p to reuse port\n", xprt);
1622
1623         /*
1624          * Disconnect the transport socket by doing a connect operation
1625          * with AF_UNSPEC.  This should return immediately...
1626          */
1627         memset(&any, 0, sizeof(any));
1628         any.sa_family = AF_UNSPEC;
1629         result = kernel_connect(transport->sock, &any, sizeof(any), 0);
1630         if (result)
1631                 dprintk("RPC:       AF_UNSPEC connect return code %d\n",
1632                                 result);
1633 }
1634
1635 static void xs_tcp_reuse_connection(struct rpc_xprt *xprt, struct sock_xprt *transport)
1636 {
1637         unsigned int state = transport->inet->sk_state;
1638
1639         if (state == TCP_CLOSE && transport->sock->state == SS_UNCONNECTED)
1640                 return;
1641         if ((1 << state) & (TCPF_ESTABLISHED|TCPF_SYN_SENT))
1642                 return;
1643         xs_abort_connection(xprt, transport);
1644 }
1645
1646 static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
1647 {
1648         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1649
1650         if (!transport->inet) {
1651                 struct sock *sk = sock->sk;
1652
1653                 write_lock_bh(&sk->sk_callback_lock);
1654
1655                 xs_save_old_callbacks(transport, sk);
1656
1657                 sk->sk_user_data = xprt;
1658                 sk->sk_data_ready = xs_tcp_data_ready;
1659                 sk->sk_state_change = xs_tcp_state_change;
1660                 sk->sk_write_space = xs_tcp_write_space;
1661                 sk->sk_error_report = xs_tcp_error_report;
1662                 sk->sk_allocation = GFP_ATOMIC;
1663
1664                 /* socket options */
1665                 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
1666                 sock_reset_flag(sk, SOCK_LINGER);
1667                 tcp_sk(sk)->linger2 = 0;
1668                 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
1669
1670                 xprt_clear_connected(xprt);
1671
1672                 /* Reset to new socket */
1673                 transport->sock = sock;
1674                 transport->inet = sk;
1675
1676                 write_unlock_bh(&sk->sk_callback_lock);
1677         }
1678
1679         if (!xprt_bound(xprt))
1680                 return -ENOTCONN;
1681
1682         /* Tell the socket layer to start connecting... */
1683         xprt->stat.connect_count++;
1684         xprt->stat.connect_start = jiffies;
1685         return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
1686 }
1687
1688 /**
1689  * xs_tcp_connect_worker4 - connect a TCP socket to a remote endpoint
1690  * @work: RPC transport to connect
1691  *
1692  * Invoked by a work queue tasklet.
1693  */
1694 static void xs_tcp_connect_worker4(struct work_struct *work)
1695 {
1696         struct sock_xprt *transport =
1697                 container_of(work, struct sock_xprt, connect_worker.work);
1698         struct rpc_xprt *xprt = &transport->xprt;
1699         struct socket *sock = transport->sock;
1700         int err, status = -EIO;
1701
1702         if (xprt->shutdown)
1703                 goto out;
1704
1705         if (!sock) {
1706                 /* start from scratch */
1707                 if ((err = sock_create_kern(PF_INET, SOCK_STREAM, IPPROTO_TCP, &sock)) < 0) {
1708                         dprintk("RPC:       can't create TCP transport socket (%d).\n", -err);
1709                         goto out;
1710                 }
1711                 xs_reclassify_socket4(sock);
1712
1713                 if (xs_bind4(transport, sock) < 0) {
1714                         sock_release(sock);
1715                         goto out;
1716                 }
1717         } else
1718                 /* "close" the socket, preserving the local port */
1719                 xs_tcp_reuse_connection(xprt, transport);
1720
1721         dprintk("RPC:       worker connecting xprt %p to address: %s\n",
1722                         xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1723
1724         status = xs_tcp_finish_connecting(xprt, sock);
1725         dprintk("RPC:       %p connect status %d connected %d sock state %d\n",
1726                         xprt, -status, xprt_connected(xprt),
1727                         sock->sk->sk_state);
1728         if (status < 0) {
1729                 switch (status) {
1730                         case -EINPROGRESS:
1731                         case -EALREADY:
1732                                 goto out_clear;
1733                         case -ECONNREFUSED:
1734                         case -ECONNRESET:
1735                                 /* retry with existing socket, after a delay */
1736                                 break;
1737                         default:
1738                                 /* get rid of existing socket, and retry */
1739                                 xs_tcp_shutdown(xprt);
1740                 }
1741         }
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         if (status < 0) {
1788                 switch (status) {
1789                         case -EINPROGRESS:
1790                         case -EALREADY:
1791                                 goto out_clear;
1792                         case -ECONNREFUSED:
1793                         case -ECONNRESET:
1794                                 /* retry with existing socket, after a delay */
1795                                 break;
1796                         default:
1797                                 /* get rid of existing socket, and retry */
1798                                 xs_tcp_shutdown(xprt);
1799                 }
1800         }
1801 out:
1802         xprt_wake_pending_tasks(xprt, status);
1803 out_clear:
1804         xprt_clear_connecting(xprt);
1805 }
1806
1807 /**
1808  * xs_connect - connect a socket to a remote endpoint
1809  * @task: address of RPC task that manages state of connect request
1810  *
1811  * TCP: If the remote end dropped the connection, delay reconnecting.
1812  *
1813  * UDP socket connects are synchronous, but we use a work queue anyway
1814  * to guarantee that even unprivileged user processes can set up a
1815  * socket on a privileged port.
1816  *
1817  * If a UDP socket connect fails, the delay behavior here prevents
1818  * retry floods (hard mounts).
1819  */
1820 static void xs_connect(struct rpc_task *task)
1821 {
1822         struct rpc_xprt *xprt = task->tk_xprt;
1823         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1824
1825         if (xprt_test_and_set_connecting(xprt))
1826                 return;
1827
1828         if (transport->sock != NULL) {
1829                 dprintk("RPC:       xs_connect delayed xprt %p for %lu "
1830                                 "seconds\n",
1831                                 xprt, xprt->reestablish_timeout / HZ);
1832                 queue_delayed_work(rpciod_workqueue,
1833                                    &transport->connect_worker,
1834                                    xprt->reestablish_timeout);
1835                 xprt->reestablish_timeout <<= 1;
1836                 if (xprt->reestablish_timeout > XS_TCP_MAX_REEST_TO)
1837                         xprt->reestablish_timeout = XS_TCP_MAX_REEST_TO;
1838         } else {
1839                 dprintk("RPC:       xs_connect scheduled xprt %p\n", xprt);
1840                 queue_delayed_work(rpciod_workqueue,
1841                                    &transport->connect_worker, 0);
1842         }
1843 }
1844
1845 static void xs_tcp_connect(struct rpc_task *task)
1846 {
1847         struct rpc_xprt *xprt = task->tk_xprt;
1848
1849         /* Exit if we need to wait for socket shutdown to complete */
1850         if (test_bit(XPRT_CLOSING, &xprt->state))
1851                 return;
1852         xs_connect(task);
1853 }
1854
1855 /**
1856  * xs_udp_print_stats - display UDP socket-specifc stats
1857  * @xprt: rpc_xprt struct containing statistics
1858  * @seq: output file
1859  *
1860  */
1861 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
1862 {
1863         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1864
1865         seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %Lu %Lu\n",
1866                         transport->port,
1867                         xprt->stat.bind_count,
1868                         xprt->stat.sends,
1869                         xprt->stat.recvs,
1870                         xprt->stat.bad_xids,
1871                         xprt->stat.req_u,
1872                         xprt->stat.bklog_u);
1873 }
1874
1875 /**
1876  * xs_tcp_print_stats - display TCP socket-specifc stats
1877  * @xprt: rpc_xprt struct containing statistics
1878  * @seq: output file
1879  *
1880  */
1881 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
1882 {
1883         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1884         long idle_time = 0;
1885
1886         if (xprt_connected(xprt))
1887                 idle_time = (long)(jiffies - xprt->last_used) / HZ;
1888
1889         seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu %Lu %Lu\n",
1890                         transport->port,
1891                         xprt->stat.bind_count,
1892                         xprt->stat.connect_count,
1893                         xprt->stat.connect_time,
1894                         idle_time,
1895                         xprt->stat.sends,
1896                         xprt->stat.recvs,
1897                         xprt->stat.bad_xids,
1898                         xprt->stat.req_u,
1899                         xprt->stat.bklog_u);
1900 }
1901
1902 static struct rpc_xprt_ops xs_udp_ops = {
1903         .set_buffer_size        = xs_udp_set_buffer_size,
1904         .reserve_xprt           = xprt_reserve_xprt_cong,
1905         .release_xprt           = xprt_release_xprt_cong,
1906         .rpcbind                = rpcb_getport_async,
1907         .set_port               = xs_set_port,
1908         .connect                = xs_connect,
1909         .buf_alloc              = rpc_malloc,
1910         .buf_free               = rpc_free,
1911         .send_request           = xs_udp_send_request,
1912         .set_retrans_timeout    = xprt_set_retrans_timeout_rtt,
1913         .timer                  = xs_udp_timer,
1914         .release_request        = xprt_release_rqst_cong,
1915         .close                  = xs_close,
1916         .destroy                = xs_destroy,
1917         .print_stats            = xs_udp_print_stats,
1918 };
1919
1920 static struct rpc_xprt_ops xs_tcp_ops = {
1921         .reserve_xprt           = xprt_reserve_xprt,
1922         .release_xprt           = xs_tcp_release_xprt,
1923         .rpcbind                = rpcb_getport_async,
1924         .set_port               = xs_set_port,
1925         .connect                = xs_tcp_connect,
1926         .buf_alloc              = rpc_malloc,
1927         .buf_free               = rpc_free,
1928         .send_request           = xs_tcp_send_request,
1929         .set_retrans_timeout    = xprt_set_retrans_timeout_def,
1930         .close                  = xs_tcp_shutdown,
1931         .destroy                = xs_destroy,
1932         .print_stats            = xs_tcp_print_stats,
1933 };
1934
1935 static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
1936                                       unsigned int slot_table_size)
1937 {
1938         struct rpc_xprt *xprt;
1939         struct sock_xprt *new;
1940
1941         if (args->addrlen > sizeof(xprt->addr)) {
1942                 dprintk("RPC:       xs_setup_xprt: address too large\n");
1943                 return ERR_PTR(-EBADF);
1944         }
1945
1946         new = kzalloc(sizeof(*new), GFP_KERNEL);
1947         if (new == NULL) {
1948                 dprintk("RPC:       xs_setup_xprt: couldn't allocate "
1949                                 "rpc_xprt\n");
1950                 return ERR_PTR(-ENOMEM);
1951         }
1952         xprt = &new->xprt;
1953
1954         xprt->max_reqs = slot_table_size;
1955         xprt->slot = kcalloc(xprt->max_reqs, sizeof(struct rpc_rqst), GFP_KERNEL);
1956         if (xprt->slot == NULL) {
1957                 kfree(xprt);
1958                 dprintk("RPC:       xs_setup_xprt: couldn't allocate slot "
1959                                 "table\n");
1960                 return ERR_PTR(-ENOMEM);
1961         }
1962
1963         memcpy(&xprt->addr, args->dstaddr, args->addrlen);
1964         xprt->addrlen = args->addrlen;
1965         if (args->srcaddr)
1966                 memcpy(&new->addr, args->srcaddr, args->addrlen);
1967
1968         return xprt;
1969 }
1970
1971 static const struct rpc_timeout xs_udp_default_timeout = {
1972         .to_initval = 5 * HZ,
1973         .to_maxval = 30 * HZ,
1974         .to_increment = 5 * HZ,
1975         .to_retries = 5,
1976 };
1977
1978 /**
1979  * xs_setup_udp - Set up transport to use a UDP socket
1980  * @args: rpc transport creation arguments
1981  *
1982  */
1983 static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
1984 {
1985         struct sockaddr *addr = args->dstaddr;
1986         struct rpc_xprt *xprt;
1987         struct sock_xprt *transport;
1988
1989         xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries);
1990         if (IS_ERR(xprt))
1991                 return xprt;
1992         transport = container_of(xprt, struct sock_xprt, xprt);
1993
1994         xprt->prot = IPPROTO_UDP;
1995         xprt->tsh_size = 0;
1996         /* XXX: header size can vary due to auth type, IPv6, etc. */
1997         xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
1998
1999         xprt->bind_timeout = XS_BIND_TO;
2000         xprt->connect_timeout = XS_UDP_CONN_TO;
2001         xprt->reestablish_timeout = XS_UDP_REEST_TO;
2002         xprt->idle_timeout = XS_IDLE_DISC_TO;
2003
2004         xprt->ops = &xs_udp_ops;
2005
2006         xprt->timeout = &xs_udp_default_timeout;
2007
2008         switch (addr->sa_family) {
2009         case AF_INET:
2010                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2011                         xprt_set_bound(xprt);
2012
2013                 INIT_DELAYED_WORK(&transport->connect_worker,
2014                                         xs_udp_connect_worker4);
2015                 xs_format_ipv4_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
2016                 break;
2017         case AF_INET6:
2018                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2019                         xprt_set_bound(xprt);
2020
2021                 INIT_DELAYED_WORK(&transport->connect_worker,
2022                                         xs_udp_connect_worker6);
2023                 xs_format_ipv6_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
2024                 break;
2025         default:
2026                 kfree(xprt);
2027                 return ERR_PTR(-EAFNOSUPPORT);
2028         }
2029
2030         dprintk("RPC:       set up transport to address %s\n",
2031                         xprt->address_strings[RPC_DISPLAY_ALL]);
2032
2033         if (try_module_get(THIS_MODULE))
2034                 return xprt;
2035
2036         kfree(xprt->slot);
2037         kfree(xprt);
2038         return ERR_PTR(-EINVAL);
2039 }
2040
2041 static const struct rpc_timeout xs_tcp_default_timeout = {
2042         .to_initval = 60 * HZ,
2043         .to_maxval = 60 * HZ,
2044         .to_retries = 2,
2045 };
2046
2047 /**
2048  * xs_setup_tcp - Set up transport to use a TCP socket
2049  * @args: rpc transport creation arguments
2050  *
2051  */
2052 static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
2053 {
2054         struct sockaddr *addr = args->dstaddr;
2055         struct rpc_xprt *xprt;
2056         struct sock_xprt *transport;
2057
2058         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries);
2059         if (IS_ERR(xprt))
2060                 return xprt;
2061         transport = container_of(xprt, struct sock_xprt, xprt);
2062
2063         xprt->prot = IPPROTO_TCP;
2064         xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2065         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2066
2067         xprt->bind_timeout = XS_BIND_TO;
2068         xprt->connect_timeout = XS_TCP_CONN_TO;
2069         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2070         xprt->idle_timeout = XS_IDLE_DISC_TO;
2071
2072         xprt->ops = &xs_tcp_ops;
2073         xprt->timeout = &xs_tcp_default_timeout;
2074
2075         switch (addr->sa_family) {
2076         case AF_INET:
2077                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2078                         xprt_set_bound(xprt);
2079
2080                 INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_connect_worker4);
2081                 xs_format_ipv4_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
2082                 break;
2083         case AF_INET6:
2084                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2085                         xprt_set_bound(xprt);
2086
2087                 INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_connect_worker6);
2088                 xs_format_ipv6_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
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 struct xprt_class        xs_udp_transport = {
2107         .list           = LIST_HEAD_INIT(xs_udp_transport.list),
2108         .name           = "udp",
2109         .owner          = THIS_MODULE,
2110         .ident          = IPPROTO_UDP,
2111         .setup          = xs_setup_udp,
2112 };
2113
2114 static struct xprt_class        xs_tcp_transport = {
2115         .list           = LIST_HEAD_INIT(xs_tcp_transport.list),
2116         .name           = "tcp",
2117         .owner          = THIS_MODULE,
2118         .ident          = IPPROTO_TCP,
2119         .setup          = xs_setup_tcp,
2120 };
2121
2122 /**
2123  * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
2124  *
2125  */
2126 int init_socket_xprt(void)
2127 {
2128 #ifdef RPC_DEBUG
2129         if (!sunrpc_table_header)
2130                 sunrpc_table_header = register_sysctl_table(sunrpc_table);
2131 #endif
2132
2133         xprt_register_transport(&xs_udp_transport);
2134         xprt_register_transport(&xs_tcp_transport);
2135
2136         return 0;
2137 }
2138
2139 /**
2140  * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
2141  *
2142  */
2143 void cleanup_socket_xprt(void)
2144 {
2145 #ifdef RPC_DEBUG
2146         if (sunrpc_table_header) {
2147                 unregister_sysctl_table(sunrpc_table_header);
2148                 sunrpc_table_header = NULL;
2149         }
2150 #endif
2151
2152         xprt_unregister_transport(&xs_udp_transport);
2153         xprt_unregister_transport(&xs_tcp_transport);
2154 }