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 /**
 771  * xs_close - close a socket
 772  * @xprt: transport
 773  *
 774  * This is used when all requests are complete; ie, no DRC state remains
 775  * on the server we want to save.
 776  */
 777 static void xs_close(struct rpc_xprt *xprt)
 778 {
 779         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 780         struct socket *sock = transport->sock;
 781         struct sock *sk = transport->inet;
 782
 783         if (!sk)
 784                 goto clear_close_wait;
 785
 786         dprintk("RPC:       xs_close xprt %p\n", xprt);
 787
 788         write_lock_bh(&sk->sk_callback_lock);
 789         transport->inet = NULL;
 790         transport->sock = NULL;
 791
 792         sk->sk_user_data = NULL;
 793
 794         xs_restore_old_callbacks(transport, sk);
 795         write_unlock_bh(&sk->sk_callback_lock);
 796
 797         sk->sk_no_check = 0;
 798
 799         sock_release(sock);
 800 clear_close_wait:
 801         smp_mb__before_clear_bit();
 802         clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
 803         clear_bit(XPRT_CLOSING, &xprt->state);
 804         smp_mb__after_clear_bit();
 805         xprt_disconnect_done(xprt);
 806 }
 807
 808 /**
 809  * xs_destroy - prepare to shutdown a transport
 810  * @xprt: doomed transport
 811  *
 812  */
 813 static void xs_destroy(struct rpc_xprt *xprt)
 814 {
 815         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 816
 817         dprintk("RPC:       xs_destroy xprt %p\n", xprt);
 818
 819         cancel_rearming_delayed_work(&transport->connect_worker);
 820
 821         xs_close(xprt);
 822         xs_free_peer_addresses(xprt);
 823         kfree(xprt->slot);
 824         kfree(xprt);
 825         module_put(THIS_MODULE);
 826 }
 827
 828 static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
 829 {
 830         return (struct rpc_xprt *) sk->sk_user_data;
 831 }
 832
 833 /**
 834  * xs_udp_data_ready - "data ready" callback for UDP sockets
 835  * @sk: socket with data to read
 836  * @len: how much data to read
 837  *
 838  */
 839 static void xs_udp_data_ready(struct sock *sk, int len)
 840 {
 841         struct rpc_task *task;
 842         struct rpc_xprt *xprt;
 843         struct rpc_rqst *rovr;
 844         struct sk_buff *skb;
 845         int err, repsize, copied;
 846         u32 _xid;
 847         __be32 *xp;
 848
 849         read_lock(&sk->sk_callback_lock);
 850         dprintk("RPC:       xs_udp_data_ready...\n");
 851         if (!(xprt = xprt_from_sock(sk)))
 852                 goto out;
 853
 854         if ((skb = skb_recv_datagram(sk, 0, 1, &err)) == NULL)
 855                 goto out;
 856
 857         if (xprt->shutdown)
 858                 goto dropit;
 859
 860         repsize = skb->len - sizeof(struct udphdr);
 861         if (repsize < 4) {
 862                 dprintk("RPC:       impossible RPC reply size %d!\n", repsize);
 863                 goto dropit;
 864         }
 865
 866         /* Copy the XID from the skb... */
 867         xp = skb_header_pointer(skb, sizeof(struct udphdr),
 868                                 sizeof(_xid), &_xid);
 869         if (xp == NULL)
 870                 goto dropit;
 871
 872         /* Look up and lock the request corresponding to the given XID */
 873         spin_lock(&xprt->transport_lock);
 874         rovr = xprt_lookup_rqst(xprt, *xp);
 875         if (!rovr)
 876                 goto out_unlock;
 877         task = rovr->rq_task;
 878
 879         if ((copied = rovr->rq_private_buf.buflen) > repsize)
 880                 copied = repsize;
 881
 882         /* Suck it into the iovec, verify checksum if not done by hw. */
 883         if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
 884                 UDPX_INC_STATS_BH(sk, UDP_MIB_INERRORS);
 885                 goto out_unlock;
 886         }
 887
 888         UDPX_INC_STATS_BH(sk, UDP_MIB_INDATAGRAMS);
 889
 890         /* Something worked... */
 891         dst_confirm(skb->dst);
 892
 893         xprt_adjust_cwnd(task, copied);
 894         xprt_update_rtt(task);
 895         xprt_complete_rqst(task, copied);
 896
 897  out_unlock:
 898         spin_unlock(&xprt->transport_lock);
 899  dropit:
 900         skb_free_datagram(sk, skb);
 901  out:
 902         read_unlock(&sk->sk_callback_lock);
 903 }
 904
 905 static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
 906 {
 907         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 908         size_t len, used;
 909         char *p;
 910
 911         p = ((char *) &transport->tcp_fraghdr) + transport->tcp_offset;
 912         len = sizeof(transport->tcp_fraghdr) - transport->tcp_offset;
 913         used = xdr_skb_read_bits(desc, p, len);
 914         transport->tcp_offset += used;
 915         if (used != len)
 916                 return;
 917
 918         transport->tcp_reclen = ntohl(transport->tcp_fraghdr);
 919         if (transport->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
 920                 transport->tcp_flags |= TCP_RCV_LAST_FRAG;
 921         else
 922                 transport->tcp_flags &= ~TCP_RCV_LAST_FRAG;
 923         transport->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
 924
 925         transport->tcp_flags &= ~TCP_RCV_COPY_FRAGHDR;
 926         transport->tcp_offset = 0;
 927
 928         /* Sanity check of the record length */
 929         if (unlikely(transport->tcp_reclen < 4)) {
 930                 dprintk("RPC:       invalid TCP record fragment length\n");
 931                 xprt_force_disconnect(xprt);
 932                 return;
 933         }
 934         dprintk("RPC:       reading TCP record fragment of length %d\n",
 935                         transport->tcp_reclen);
 936 }
 937
 938 static void xs_tcp_check_fraghdr(struct sock_xprt *transport)
 939 {
 940         if (transport->tcp_offset == transport->tcp_reclen) {
 941                 transport->tcp_flags |= TCP_RCV_COPY_FRAGHDR;
 942                 transport->tcp_offset = 0;
 943                 if (transport->tcp_flags & TCP_RCV_LAST_FRAG) {
 944                         transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
 945                         transport->tcp_flags |= TCP_RCV_COPY_XID;
 946                         transport->tcp_copied = 0;
 947                 }
 948         }
 949 }
 950
 951 static inline void xs_tcp_read_xid(struct sock_xprt *transport, struct xdr_skb_reader *desc)
 952 {
 953         size_t len, used;
 954         char *p;
 955
 956         len = sizeof(transport->tcp_xid) - transport->tcp_offset;
 957         dprintk("RPC:       reading XID (%Zu bytes)\n", len);
 958         p = ((char *) &transport->tcp_xid) + transport->tcp_offset;
 959         used = xdr_skb_read_bits(desc, p, len);
 960         transport->tcp_offset += used;
 961         if (used != len)
 962                 return;
 963         transport->tcp_flags &= ~TCP_RCV_COPY_XID;
 964         transport->tcp_flags |= TCP_RCV_COPY_DATA;
 965         transport->tcp_copied = 4;
 966         dprintk("RPC:       reading reply for XID %08x\n",
 967                         ntohl(transport->tcp_xid));
 968         xs_tcp_check_fraghdr(transport);
 969 }
 970
 971 static inline void xs_tcp_read_request(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
 972 {
 973         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 974         struct rpc_rqst *req;
 975         struct xdr_buf *rcvbuf;
 976         size_t len;
 977         ssize_t r;
 978
 979         /* Find and lock the request corresponding to this xid */
 980         spin_lock(&xprt->transport_lock);
 981         req = xprt_lookup_rqst(xprt, transport->tcp_xid);
 982         if (!req) {
 983                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
 984                 dprintk("RPC:       XID %08x request not found!\n",
 985                                 ntohl(transport->tcp_xid));
 986                 spin_unlock(&xprt->transport_lock);
 987                 return;
 988         }
 989
 990         rcvbuf = &req->rq_private_buf;
 991         len = desc->count;
 992         if (len > transport->tcp_reclen - transport->tcp_offset) {
 993                 struct xdr_skb_reader my_desc;
 994
 995                 len = transport->tcp_reclen - transport->tcp_offset;
 996                 memcpy(&my_desc, desc, sizeof(my_desc));
 997                 my_desc.count = len;
 998                 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
 999                                           &my_desc, xdr_skb_read_bits);
1000                 desc->count -= r;
1001                 desc->offset += r;
1002         } else
1003                 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
1004                                           desc, xdr_skb_read_bits);
1005
1006         if (r > 0) {
1007                 transport->tcp_copied += r;
1008                 transport->tcp_offset += r;
1009         }
1010         if (r != len) {
1011                 /* Error when copying to the receive buffer,
1012                  * usually because we weren't able to allocate
1013                  * additional buffer pages. All we can do now
1014                  * is turn off TCP_RCV_COPY_DATA, so the request
1015                  * will not receive any additional updates,
1016                  * and time out.
1017                  * Any remaining data from this record will
1018                  * be discarded.
1019                  */
1020                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1021                 dprintk("RPC:       XID %08x truncated request\n",
1022                                 ntohl(transport->tcp_xid));
1023                 dprintk("RPC:       xprt = %p, tcp_copied = %lu, "
1024                                 "tcp_offset = %u, tcp_reclen = %u\n",
1025                                 xprt, transport->tcp_copied,
1026                                 transport->tcp_offset, transport->tcp_reclen);
1027                 goto out;
1028         }
1029
1030         dprintk("RPC:       XID %08x read %Zd bytes\n",
1031                         ntohl(transport->tcp_xid), r);
1032         dprintk("RPC:       xprt = %p, tcp_copied = %lu, tcp_offset = %u, "
1033                         "tcp_reclen = %u\n", xprt, transport->tcp_copied,
1034                         transport->tcp_offset, transport->tcp_reclen);
1035
1036         if (transport->tcp_copied == req->rq_private_buf.buflen)
1037                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1038         else if (transport->tcp_offset == transport->tcp_reclen) {
1039                 if (transport->tcp_flags & TCP_RCV_LAST_FRAG)
1040                         transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1041         }
1042
1043 out:
1044         if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1045                 xprt_complete_rqst(req->rq_task, transport->tcp_copied);
1046         spin_unlock(&xprt->transport_lock);
1047         xs_tcp_check_fraghdr(transport);
1048 }
1049
1050 static inline void xs_tcp_read_discard(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1051 {
1052         size_t len;
1053
1054         len = transport->tcp_reclen - transport->tcp_offset;
1055         if (len > desc->count)
1056                 len = desc->count;
1057         desc->count -= len;
1058         desc->offset += len;
1059         transport->tcp_offset += len;
1060         dprintk("RPC:       discarded %Zu bytes\n", len);
1061         xs_tcp_check_fraghdr(transport);
1062 }
1063
1064 static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
1065 {
1066         struct rpc_xprt *xprt = rd_desc->arg.data;
1067         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1068         struct xdr_skb_reader desc = {
1069                 .skb    = skb,
1070                 .offset = offset,
1071                 .count  = len,
1072         };
1073
1074         dprintk("RPC:       xs_tcp_data_recv started\n");
1075         do {
1076                 /* Read in a new fragment marker if necessary */
1077                 /* Can we ever really expect to get completely empty fragments? */
1078                 if (transport->tcp_flags & TCP_RCV_COPY_FRAGHDR) {
1079                         xs_tcp_read_fraghdr(xprt, &desc);
1080                         continue;
1081                 }
1082                 /* Read in the xid if necessary */
1083                 if (transport->tcp_flags & TCP_RCV_COPY_XID) {
1084                         xs_tcp_read_xid(transport, &desc);
1085                         continue;
1086                 }
1087                 /* Read in the request data */
1088                 if (transport->tcp_flags & TCP_RCV_COPY_DATA) {
1089                         xs_tcp_read_request(xprt, &desc);
1090                         continue;
1091                 }
1092                 /* Skip over any trailing bytes on short reads */
1093                 xs_tcp_read_discard(transport, &desc);
1094         } while (desc.count);
1095         dprintk("RPC:       xs_tcp_data_recv done\n");
1096         return len - desc.count;
1097 }
1098
1099 /**
1100  * xs_tcp_data_ready - "data ready" callback for TCP sockets
1101  * @sk: socket with data to read
1102  * @bytes: how much data to read
1103  *
1104  */
1105 static void xs_tcp_data_ready(struct sock *sk, int bytes)
1106 {
1107         struct rpc_xprt *xprt;
1108         read_descriptor_t rd_desc;
1109         int read;
1110
1111         dprintk("RPC:       xs_tcp_data_ready...\n");
1112
1113         read_lock(&sk->sk_callback_lock);
1114         if (!(xprt = xprt_from_sock(sk)))
1115                 goto out;
1116         if (xprt->shutdown)
1117                 goto out;
1118
1119         /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
1120         rd_desc.arg.data = xprt;
1121         do {
1122                 rd_desc.count = 65536;
1123                 read = tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
1124         } while (read > 0);
1125 out:
1126         read_unlock(&sk->sk_callback_lock);
1127 }
1128
1129 /**
1130  * xs_tcp_state_change - callback to handle TCP socket state changes
1131  * @sk: socket whose state has changed
1132  *
1133  */
1134 static void xs_tcp_state_change(struct sock *sk)
1135 {
1136         struct rpc_xprt *xprt;
1137
1138         read_lock(&sk->sk_callback_lock);
1139         if (!(xprt = xprt_from_sock(sk)))
1140                 goto out;
1141         dprintk("RPC:       xs_tcp_state_change client %p...\n", xprt);
1142         dprintk("RPC:       state %x conn %d dead %d zapped %d\n",
1143                         sk->sk_state, xprt_connected(xprt),
1144                         sock_flag(sk, SOCK_DEAD),
1145                         sock_flag(sk, SOCK_ZAPPED));
1146
1147         switch (sk->sk_state) {
1148         case TCP_ESTABLISHED:
1149                 spin_lock_bh(&xprt->transport_lock);
1150                 if (!xprt_test_and_set_connected(xprt)) {
1151                         struct sock_xprt *transport = container_of(xprt,
1152                                         struct sock_xprt, xprt);
1153
1154                         /* Reset TCP record info */
1155                         transport->tcp_offset = 0;
1156                         transport->tcp_reclen = 0;
1157                         transport->tcp_copied = 0;
1158                         transport->tcp_flags =
1159                                 TCP_RCV_COPY_FRAGHDR | TCP_RCV_COPY_XID;
1160
1161                         xprt_wake_pending_tasks(xprt, 0);
1162                 }
1163                 spin_unlock_bh(&xprt->transport_lock);
1164                 break;
1165         case TCP_FIN_WAIT1:
1166                 /* The client initiated a shutdown of the socket */
1167                 xprt->connect_cookie++;
1168                 xprt->reestablish_timeout = 0;
1169                 set_bit(XPRT_CLOSING, &xprt->state);
1170                 smp_mb__before_clear_bit();
1171                 clear_bit(XPRT_CONNECTED, &xprt->state);
1172                 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1173                 smp_mb__after_clear_bit();
1174                 break;
1175         case TCP_CLOSE_WAIT:
1176                 /* The server initiated a shutdown of the socket */
1177                 set_bit(XPRT_CLOSING, &xprt->state);
1178                 xprt_force_disconnect(xprt);
1179         case TCP_SYN_SENT:
1180                 xprt->connect_cookie++;
1181         case TCP_CLOSING:
1182                 /*
1183                  * If the server closed down the connection, make sure that
1184                  * we back off before reconnecting
1185                  */
1186                 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1187                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1188                 break;
1189         case TCP_LAST_ACK:
1190                 smp_mb__before_clear_bit();
1191                 clear_bit(XPRT_CONNECTED, &xprt->state);
1192                 smp_mb__after_clear_bit();
1193                 break;
1194         case TCP_CLOSE:
1195                 smp_mb__before_clear_bit();
1196                 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1197                 clear_bit(XPRT_CLOSING, &xprt->state);
1198                 smp_mb__after_clear_bit();
1199                 /* Mark transport as closed and wake up all pending tasks */
1200                 xprt_disconnect_done(xprt);
1201         }
1202  out:
1203         read_unlock(&sk->sk_callback_lock);
1204 }
1205
1206 /**
1207  * xs_tcp_error_report - callback mainly for catching RST events
1208  * @sk: socket
1209  */
1210 static void xs_tcp_error_report(struct sock *sk)
1211 {
1212         struct rpc_xprt *xprt;
1213
1214         read_lock(&sk->sk_callback_lock);
1215         if (sk->sk_err != ECONNRESET || sk->sk_state != TCP_ESTABLISHED)
1216                 goto out;
1217         if (!(xprt = xprt_from_sock(sk)))
1218                 goto out;
1219         dprintk("RPC:       %s client %p...\n"
1220                         "RPC:       error %d\n",
1221                         __func__, xprt, sk->sk_err);
1222
1223         xprt_force_disconnect(xprt);
1224 out:
1225         read_unlock(&sk->sk_callback_lock);
1226 }
1227
1228 static void xs_write_space(struct sock *sk)
1229 {
1230         struct socket *sock;
1231         struct rpc_xprt *xprt;
1232
1233         if (unlikely(!(sock = sk->sk_socket)))
1234                 return;
1235         clear_bit(SOCK_NOSPACE, &sock->flags);
1236
1237         if (unlikely(!(xprt = xprt_from_sock(sk))))
1238                 return;
1239         if (test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags) == 0)
1240                 return;
1241
1242         xprt_write_space(xprt);
1243 }
1244
1245 /**
1246  * xs_udp_write_space - callback invoked when socket buffer space
1247  *                             becomes available
1248  * @sk: socket whose state has changed
1249  *
1250  * Called when more output buffer space is available for this socket.
1251  * We try not to wake our writers until they can make "significant"
1252  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1253  * with a bunch of small requests.
1254  */
1255 static void xs_udp_write_space(struct sock *sk)
1256 {
1257         read_lock(&sk->sk_callback_lock);
1258
1259         /* from net/core/sock.c:sock_def_write_space */
1260         if (sock_writeable(sk))
1261                 xs_write_space(sk);
1262
1263         read_unlock(&sk->sk_callback_lock);
1264 }
1265
1266 /**
1267  * xs_tcp_write_space - callback invoked when socket buffer space
1268  *                             becomes available
1269  * @sk: socket whose state has changed
1270  *
1271  * Called when more output buffer space is available for this socket.
1272  * We try not to wake our writers until they can make "significant"
1273  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1274  * with a bunch of small requests.
1275  */
1276 static void xs_tcp_write_space(struct sock *sk)
1277 {
1278         read_lock(&sk->sk_callback_lock);
1279
1280         /* from net/core/stream.c:sk_stream_write_space */
1281         if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk))
1282                 xs_write_space(sk);
1283
1284         read_unlock(&sk->sk_callback_lock);
1285 }
1286
1287 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
1288 {
1289         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1290         struct sock *sk = transport->inet;
1291
1292         if (transport->rcvsize) {
1293                 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1294                 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
1295         }
1296         if (transport->sndsize) {
1297                 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1298                 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1299                 sk->sk_write_space(sk);
1300         }
1301 }
1302
1303 /**
1304  * xs_udp_set_buffer_size - set send and receive limits
1305  * @xprt: generic transport
1306  * @sndsize: requested size of send buffer, in bytes
1307  * @rcvsize: requested size of receive buffer, in bytes
1308  *
1309  * Set socket send and receive buffer size limits.
1310  */
1311 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1312 {
1313         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1314
1315         transport->sndsize = 0;
1316         if (sndsize)
1317                 transport->sndsize = sndsize + 1024;
1318         transport->rcvsize = 0;
1319         if (rcvsize)
1320                 transport->rcvsize = rcvsize + 1024;
1321
1322         xs_udp_do_set_buffer_size(xprt);
1323 }
1324
1325 /**
1326  * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1327  * @task: task that timed out
1328  *
1329  * Adjust the congestion window after a retransmit timeout has occurred.
1330  */
1331 static void xs_udp_timer(struct rpc_task *task)
1332 {
1333         xprt_adjust_cwnd(task, -ETIMEDOUT);
1334 }
1335
1336 static unsigned short xs_get_random_port(void)
1337 {
1338         unsigned short range = xprt_max_resvport - xprt_min_resvport;
1339         unsigned short rand = (unsigned short) net_random() % range;
1340         return rand + xprt_min_resvport;
1341 }
1342
1343 /**
1344  * xs_set_port - reset the port number in the remote endpoint address
1345  * @xprt: generic transport
1346  * @port: new port number
1347  *
1348  */
1349 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1350 {
1351         struct sockaddr *addr = xs_addr(xprt);
1352
1353         dprintk("RPC:       setting port for xprt %p to %u\n", xprt, port);
1354
1355         switch (addr->sa_family) {
1356         case AF_INET:
1357                 ((struct sockaddr_in *)addr)->sin_port = htons(port);
1358                 break;
1359         case AF_INET6:
1360                 ((struct sockaddr_in6 *)addr)->sin6_port = htons(port);
1361                 break;
1362         default:
1363                 BUG();
1364         }
1365 }
1366
1367 static unsigned short xs_get_srcport(struct sock_xprt *transport, struct socket *sock)
1368 {
1369         unsigned short port = transport->port;
1370
1371         if (port == 0 && transport->xprt.resvport)
1372                 port = xs_get_random_port();
1373         return port;
1374 }
1375
1376 static unsigned short xs_next_srcport(struct sock_xprt *transport, struct socket *sock, unsigned short port)
1377 {
1378         if (transport->port != 0)
1379                 transport->port = 0;
1380         if (!transport->xprt.resvport)
1381                 return 0;
1382         if (port <= xprt_min_resvport || port > xprt_max_resvport)
1383                 return xprt_max_resvport;
1384         return --port;
1385 }
1386
1387 static int xs_bind4(struct sock_xprt *transport, struct socket *sock)
1388 {
1389         struct sockaddr_in myaddr = {
1390                 .sin_family = AF_INET,
1391         };
1392         struct sockaddr_in *sa;
1393         int err, nloop = 0;
1394         unsigned short port = xs_get_srcport(transport, sock);
1395         unsigned short last;
1396
1397         sa = (struct sockaddr_in *)&transport->addr;
1398         myaddr.sin_addr = sa->sin_addr;
1399         do {
1400                 myaddr.sin_port = htons(port);
1401                 err = kernel_bind(sock, (struct sockaddr *) &myaddr,
1402                                                 sizeof(myaddr));
1403                 if (port == 0)
1404                         break;
1405                 if (err == 0) {
1406                         transport->port = port;
1407                         break;
1408                 }
1409                 last = port;
1410                 port = xs_next_srcport(transport, sock, port);
1411                 if (port > last)
1412                         nloop++;
1413         } while (err == -EADDRINUSE && nloop != 2);
1414         dprintk("RPC:       %s %pI4:%u: %s (%d)\n",
1415                         __func__, &myaddr.sin_addr,
1416                         port, err ? "failed" : "ok", err);
1417         return err;
1418 }
1419
1420 static int xs_bind6(struct sock_xprt *transport, struct socket *sock)
1421 {
1422         struct sockaddr_in6 myaddr = {
1423                 .sin6_family = AF_INET6,
1424         };
1425         struct sockaddr_in6 *sa;
1426         int err, nloop = 0;
1427         unsigned short port = xs_get_srcport(transport, sock);
1428         unsigned short last;
1429
1430         sa = (struct sockaddr_in6 *)&transport->addr;
1431         myaddr.sin6_addr = sa->sin6_addr;
1432         do {
1433                 myaddr.sin6_port = htons(port);
1434                 err = kernel_bind(sock, (struct sockaddr *) &myaddr,
1435                                                 sizeof(myaddr));
1436                 if (port == 0)
1437                         break;
1438                 if (err == 0) {
1439                         transport->port = port;
1440                         break;
1441                 }
1442                 last = port;
1443                 port = xs_next_srcport(transport, sock, port);
1444                 if (port > last)
1445                         nloop++;
1446         } while (err == -EADDRINUSE && nloop != 2);
1447         dprintk("RPC:       xs_bind6 %pI6:%u: %s (%d)\n",
1448                 &myaddr.sin6_addr, port, err ? "failed" : "ok", err);
1449         return err;
1450 }
1451
1452 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1453 static struct lock_class_key xs_key[2];
1454 static struct lock_class_key xs_slock_key[2];
1455
1456 static inline void xs_reclassify_socket4(struct socket *sock)
1457 {
1458         struct sock *sk = sock->sk;
1459
1460         BUG_ON(sock_owned_by_user(sk));
1461         sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1462                 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1463 }
1464
1465 static inline void xs_reclassify_socket6(struct socket *sock)
1466 {
1467         struct sock *sk = sock->sk;
1468
1469         BUG_ON(sock_owned_by_user(sk));
1470         sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1471                 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
1472 }
1473 #else
1474 static inline void xs_reclassify_socket4(struct socket *sock)
1475 {
1476 }
1477
1478 static inline void xs_reclassify_socket6(struct socket *sock)
1479 {
1480 }
1481 #endif
1482
1483 static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
1484 {
1485         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1486
1487         if (!transport->inet) {
1488                 struct sock *sk = sock->sk;
1489
1490                 write_lock_bh(&sk->sk_callback_lock);
1491
1492                 xs_save_old_callbacks(transport, sk);
1493
1494                 sk->sk_user_data = xprt;
1495                 sk->sk_data_ready = xs_udp_data_ready;
1496                 sk->sk_write_space = xs_udp_write_space;
1497                 sk->sk_no_check = UDP_CSUM_NORCV;
1498                 sk->sk_allocation = GFP_ATOMIC;
1499
1500                 xprt_set_connected(xprt);
1501
1502                 /* Reset to new socket */
1503                 transport->sock = sock;
1504                 transport->inet = sk;
1505
1506                 write_unlock_bh(&sk->sk_callback_lock);
1507         }
1508         xs_udp_do_set_buffer_size(xprt);
1509 }
1510
1511 /**
1512  * xs_udp_connect_worker4 - set up a UDP socket
1513  * @work: RPC transport to connect
1514  *
1515  * Invoked by a work queue tasklet.
1516  */
1517 static void xs_udp_connect_worker4(struct work_struct *work)
1518 {
1519         struct sock_xprt *transport =
1520                 container_of(work, struct sock_xprt, connect_worker.work);
1521         struct rpc_xprt *xprt = &transport->xprt;
1522         struct socket *sock = transport->sock;
1523         int err, status = -EIO;
1524
1525         if (xprt->shutdown)
1526                 goto out;
1527
1528         /* Start by resetting any existing state */
1529         xs_close(xprt);
1530
1531         if ((err = sock_create_kern(PF_INET, SOCK_DGRAM, IPPROTO_UDP, &sock)) < 0) {
1532                 dprintk("RPC:       can't create UDP transport socket (%d).\n", -err);
1533                 goto out;
1534         }
1535         xs_reclassify_socket4(sock);
1536
1537         if (xs_bind4(transport, sock)) {
1538                 sock_release(sock);
1539                 goto out;
1540         }
1541
1542         dprintk("RPC:       worker connecting xprt %p to address: %s\n",
1543                         xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1544
1545         xs_udp_finish_connecting(xprt, sock);
1546         status = 0;
1547 out:
1548         xprt_wake_pending_tasks(xprt, status);
1549         xprt_clear_connecting(xprt);
1550 }
1551
1552 /**
1553  * xs_udp_connect_worker6 - set up a UDP socket
1554  * @work: RPC transport to connect
1555  *
1556  * Invoked by a work queue tasklet.
1557  */
1558 static void xs_udp_connect_worker6(struct work_struct *work)
1559 {
1560         struct sock_xprt *transport =
1561                 container_of(work, struct sock_xprt, connect_worker.work);
1562         struct rpc_xprt *xprt = &transport->xprt;
1563         struct socket *sock = transport->sock;
1564         int err, status = -EIO;
1565
1566         if (xprt->shutdown)
1567                 goto out;
1568
1569         /* Start by resetting any existing state */
1570         xs_close(xprt);
1571
1572         if ((err = sock_create_kern(PF_INET6, SOCK_DGRAM, IPPROTO_UDP, &sock)) < 0) {
1573                 dprintk("RPC:       can't create UDP transport socket (%d).\n", -err);
1574                 goto out;
1575         }
1576         xs_reclassify_socket6(sock);
1577
1578         if (xs_bind6(transport, sock) < 0) {
1579                 sock_release(sock);
1580                 goto out;
1581         }
1582
1583         dprintk("RPC:       worker connecting xprt %p to address: %s\n",
1584                         xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1585
1586         xs_udp_finish_connecting(xprt, sock);
1587         status = 0;
1588 out:
1589         xprt_wake_pending_tasks(xprt, status);
1590         xprt_clear_connecting(xprt);
1591 }
1592
1593 /*
1594  * We need to preserve the port number so the reply cache on the server can
1595  * find our cached RPC replies when we get around to reconnecting.
1596  */
1597 static void xs_tcp_reuse_connection(struct rpc_xprt *xprt)
1598 {
1599         int result;
1600         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1601         struct sockaddr any;
1602
1603         dprintk("RPC:       disconnecting xprt %p to reuse port\n", xprt);
1604
1605         /*
1606          * Disconnect the transport socket by doing a connect operation
1607          * with AF_UNSPEC.  This should return immediately...
1608          */
1609         memset(&any, 0, sizeof(any));
1610         any.sa_family = AF_UNSPEC;
1611         result = kernel_connect(transport->sock, &any, sizeof(any), 0);
1612         if (result)
1613                 dprintk("RPC:       AF_UNSPEC connect return code %d\n",
1614                                 result);
1615 }
1616
1617 static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
1618 {
1619         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1620
1621         if (!transport->inet) {
1622                 struct sock *sk = sock->sk;
1623
1624                 write_lock_bh(&sk->sk_callback_lock);
1625
1626                 xs_save_old_callbacks(transport, sk);
1627
1628                 sk->sk_user_data = xprt;
1629                 sk->sk_data_ready = xs_tcp_data_ready;
1630                 sk->sk_state_change = xs_tcp_state_change;
1631                 sk->sk_write_space = xs_tcp_write_space;
1632                 sk->sk_error_report = xs_tcp_error_report;
1633                 sk->sk_allocation = GFP_ATOMIC;
1634
1635                 /* socket options */
1636                 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
1637                 sock_reset_flag(sk, SOCK_LINGER);
1638                 tcp_sk(sk)->linger2 = 0;
1639                 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
1640
1641                 xprt_clear_connected(xprt);
1642
1643                 /* Reset to new socket */
1644                 transport->sock = sock;
1645                 transport->inet = sk;
1646
1647                 write_unlock_bh(&sk->sk_callback_lock);
1648         }
1649
1650         if (!xprt_bound(xprt))
1651                 return -ENOTCONN;
1652
1653         /* Tell the socket layer to start connecting... */
1654         xprt->stat.connect_count++;
1655         xprt->stat.connect_start = jiffies;
1656         return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
1657 }
1658
1659 /**
1660  * xs_tcp_connect_worker4 - connect a TCP socket to a remote endpoint
1661  * @work: RPC transport to connect
1662  *
1663  * Invoked by a work queue tasklet.
1664  */
1665 static void xs_tcp_connect_worker4(struct work_struct *work)
1666 {
1667         struct sock_xprt *transport =
1668                 container_of(work, struct sock_xprt, connect_worker.work);
1669         struct rpc_xprt *xprt = &transport->xprt;
1670         struct socket *sock = transport->sock;
1671         int err, status = -EIO;
1672
1673         if (xprt->shutdown)
1674                 goto out;
1675
1676         if (!sock) {
1677                 /* start from scratch */
1678                 if ((err = sock_create_kern(PF_INET, SOCK_STREAM, IPPROTO_TCP, &sock)) < 0) {
1679                         dprintk("RPC:       can't create TCP transport socket (%d).\n", -err);
1680                         goto out;
1681                 }
1682                 xs_reclassify_socket4(sock);
1683
1684                 if (xs_bind4(transport, sock) < 0) {
1685                         sock_release(sock);
1686                         goto out;
1687                 }
1688         } else
1689                 /* "close" the socket, preserving the local port */
1690                 xs_tcp_reuse_connection(xprt);
1691
1692         dprintk("RPC:       worker connecting xprt %p to address: %s\n",
1693                         xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1694
1695         status = xs_tcp_finish_connecting(xprt, sock);
1696         dprintk("RPC:       %p connect status %d connected %d sock state %d\n",
1697                         xprt, -status, xprt_connected(xprt),
1698                         sock->sk->sk_state);
1699         if (status < 0) {
1700                 switch (status) {
1701                         case -EINPROGRESS:
1702                         case -EALREADY:
1703                                 goto out_clear;
1704                         case -ECONNREFUSED:
1705                         case -ECONNRESET:
1706                                 /* retry with existing socket, after a delay */
1707                                 break;
1708                         default:
1709                                 /* get rid of existing socket, and retry */
1710                                 xs_tcp_shutdown(xprt);
1711                 }
1712         }
1713 out:
1714         xprt_wake_pending_tasks(xprt, status);
1715 out_clear:
1716         xprt_clear_connecting(xprt);
1717 }
1718
1719 /**
1720  * xs_tcp_connect_worker6 - connect a TCP socket to a remote endpoint
1721  * @work: RPC transport to connect
1722  *
1723  * Invoked by a work queue tasklet.
1724  */
1725 static void xs_tcp_connect_worker6(struct work_struct *work)
1726 {
1727         struct sock_xprt *transport =
1728                 container_of(work, struct sock_xprt, connect_worker.work);
1729         struct rpc_xprt *xprt = &transport->xprt;
1730         struct socket *sock = transport->sock;
1731         int err, status = -EIO;
1732
1733         if (xprt->shutdown)
1734                 goto out;
1735
1736         if (!sock) {
1737                 /* start from scratch */
1738                 if ((err = sock_create_kern(PF_INET6, SOCK_STREAM, IPPROTO_TCP, &sock)) < 0) {
1739                         dprintk("RPC:       can't create TCP transport socket (%d).\n", -err);
1740                         goto out;
1741                 }
1742                 xs_reclassify_socket6(sock);
1743
1744                 if (xs_bind6(transport, sock) < 0) {
1745                         sock_release(sock);
1746                         goto out;
1747                 }
1748         } else
1749                 /* "close" the socket, preserving the local port */
1750                 xs_tcp_reuse_connection(xprt);
1751
1752         dprintk("RPC:       worker connecting xprt %p to address: %s\n",
1753                         xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1754
1755         status = xs_tcp_finish_connecting(xprt, sock);
1756         dprintk("RPC:       %p connect status %d connected %d sock state %d\n",
1757                         xprt, -status, xprt_connected(xprt), sock->sk->sk_state);
1758         if (status < 0) {
1759                 switch (status) {
1760                         case -EINPROGRESS:
1761                         case -EALREADY:
1762                                 goto out_clear;
1763                         case -ECONNREFUSED:
1764                         case -ECONNRESET:
1765                                 /* retry with existing socket, after a delay */
1766                                 break;
1767                         default:
1768                                 /* get rid of existing socket, and retry */
1769                                 xs_tcp_shutdown(xprt);
1770                 }
1771         }
1772 out:
1773         xprt_wake_pending_tasks(xprt, status);
1774 out_clear:
1775         xprt_clear_connecting(xprt);
1776 }
1777
1778 /**
1779  * xs_connect - connect a socket to a remote endpoint
1780  * @task: address of RPC task that manages state of connect request
1781  *
1782  * TCP: If the remote end dropped the connection, delay reconnecting.
1783  *
1784  * UDP socket connects are synchronous, but we use a work queue anyway
1785  * to guarantee that even unprivileged user processes can set up a
1786  * socket on a privileged port.
1787  *
1788  * If a UDP socket connect fails, the delay behavior here prevents
1789  * retry floods (hard mounts).
1790  */
1791 static void xs_connect(struct rpc_task *task)
1792 {
1793         struct rpc_xprt *xprt = task->tk_xprt;
1794         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1795
1796         if (xprt_test_and_set_connecting(xprt))
1797                 return;
1798
1799         if (transport->sock != NULL) {
1800                 dprintk("RPC:       xs_connect delayed xprt %p for %lu "
1801                                 "seconds\n",
1802                                 xprt, xprt->reestablish_timeout / HZ);
1803                 queue_delayed_work(rpciod_workqueue,
1804                                    &transport->connect_worker,
1805                                    xprt->reestablish_timeout);
1806                 xprt->reestablish_timeout <<= 1;
1807                 if (xprt->reestablish_timeout > XS_TCP_MAX_REEST_TO)
1808                         xprt->reestablish_timeout = XS_TCP_MAX_REEST_TO;
1809         } else {
1810                 dprintk("RPC:       xs_connect scheduled xprt %p\n", xprt);
1811                 queue_delayed_work(rpciod_workqueue,
1812                                    &transport->connect_worker, 0);
1813         }
1814 }
1815
1816 static void xs_tcp_connect(struct rpc_task *task)
1817 {
1818         struct rpc_xprt *xprt = task->tk_xprt;
1819
1820         /* Initiate graceful shutdown of the socket if not already done */
1821         if (test_bit(XPRT_CONNECTED, &xprt->state))
1822                 xs_tcp_shutdown(xprt);
1823         /* Exit if we need to wait for socket shutdown to complete */
1824         if (test_bit(XPRT_CLOSING, &xprt->state))
1825                 return;
1826         xs_connect(task);
1827 }
1828
1829 /**
1830  * xs_udp_print_stats - display UDP socket-specifc stats
1831  * @xprt: rpc_xprt struct containing statistics
1832  * @seq: output file
1833  *
1834  */
1835 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
1836 {
1837         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1838
1839         seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %Lu %Lu\n",
1840                         transport->port,
1841                         xprt->stat.bind_count,
1842                         xprt->stat.sends,
1843                         xprt->stat.recvs,
1844                         xprt->stat.bad_xids,
1845                         xprt->stat.req_u,
1846                         xprt->stat.bklog_u);
1847 }
1848
1849 /**
1850  * xs_tcp_print_stats - display TCP socket-specifc stats
1851  * @xprt: rpc_xprt struct containing statistics
1852  * @seq: output file
1853  *
1854  */
1855 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
1856 {
1857         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1858         long idle_time = 0;
1859
1860         if (xprt_connected(xprt))
1861                 idle_time = (long)(jiffies - xprt->last_used) / HZ;
1862
1863         seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu %Lu %Lu\n",
1864                         transport->port,
1865                         xprt->stat.bind_count,
1866                         xprt->stat.connect_count,
1867                         xprt->stat.connect_time,
1868                         idle_time,
1869                         xprt->stat.sends,
1870                         xprt->stat.recvs,
1871                         xprt->stat.bad_xids,
1872                         xprt->stat.req_u,
1873                         xprt->stat.bklog_u);
1874 }
1875
1876 static struct rpc_xprt_ops xs_udp_ops = {
1877         .set_buffer_size        = xs_udp_set_buffer_size,
1878         .reserve_xprt           = xprt_reserve_xprt_cong,
1879         .release_xprt           = xprt_release_xprt_cong,
1880         .rpcbind                = rpcb_getport_async,
1881         .set_port               = xs_set_port,
1882         .connect                = xs_connect,
1883         .buf_alloc              = rpc_malloc,
1884         .buf_free               = rpc_free,
1885         .send_request           = xs_udp_send_request,
1886         .set_retrans_timeout    = xprt_set_retrans_timeout_rtt,
1887         .timer                  = xs_udp_timer,
1888         .release_request        = xprt_release_rqst_cong,
1889         .close                  = xs_close,
1890         .destroy                = xs_destroy,
1891         .print_stats            = xs_udp_print_stats,
1892 };
1893
1894 static struct rpc_xprt_ops xs_tcp_ops = {
1895         .reserve_xprt           = xprt_reserve_xprt,
1896         .release_xprt           = xs_tcp_release_xprt,
1897         .rpcbind                = rpcb_getport_async,
1898         .set_port               = xs_set_port,
1899         .connect                = xs_tcp_connect,
1900         .buf_alloc              = rpc_malloc,
1901         .buf_free               = rpc_free,
1902         .send_request           = xs_tcp_send_request,
1903         .set_retrans_timeout    = xprt_set_retrans_timeout_def,
1904         .close                  = xs_tcp_shutdown,
1905         .destroy                = xs_destroy,
1906         .print_stats            = xs_tcp_print_stats,
1907 };
1908
1909 static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
1910                                       unsigned int slot_table_size)
1911 {
1912         struct rpc_xprt *xprt;
1913         struct sock_xprt *new;
1914
1915         if (args->addrlen > sizeof(xprt->addr)) {
1916                 dprintk("RPC:       xs_setup_xprt: address too large\n");
1917                 return ERR_PTR(-EBADF);
1918         }
1919
1920         new = kzalloc(sizeof(*new), GFP_KERNEL);
1921         if (new == NULL) {
1922                 dprintk("RPC:       xs_setup_xprt: couldn't allocate "
1923                                 "rpc_xprt\n");
1924                 return ERR_PTR(-ENOMEM);
1925         }
1926         xprt = &new->xprt;
1927
1928         xprt->max_reqs = slot_table_size;
1929         xprt->slot = kcalloc(xprt->max_reqs, sizeof(struct rpc_rqst), GFP_KERNEL);
1930         if (xprt->slot == NULL) {
1931                 kfree(xprt);
1932                 dprintk("RPC:       xs_setup_xprt: couldn't allocate slot "
1933                                 "table\n");
1934                 return ERR_PTR(-ENOMEM);
1935         }
1936
1937         memcpy(&xprt->addr, args->dstaddr, args->addrlen);
1938         xprt->addrlen = args->addrlen;
1939         if (args->srcaddr)
1940                 memcpy(&new->addr, args->srcaddr, args->addrlen);
1941
1942         return xprt;
1943 }
1944
1945 static const struct rpc_timeout xs_udp_default_timeout = {
1946         .to_initval = 5 * HZ,
1947         .to_maxval = 30 * HZ,
1948         .to_increment = 5 * HZ,
1949         .to_retries = 5,
1950 };
1951
1952 /**
1953  * xs_setup_udp - Set up transport to use a UDP socket
1954  * @args: rpc transport creation arguments
1955  *
1956  */
1957 static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
1958 {
1959         struct sockaddr *addr = args->dstaddr;
1960         struct rpc_xprt *xprt;
1961         struct sock_xprt *transport;
1962
1963         xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries);
1964         if (IS_ERR(xprt))
1965                 return xprt;
1966         transport = container_of(xprt, struct sock_xprt, xprt);
1967
1968         xprt->prot = IPPROTO_UDP;
1969         xprt->tsh_size = 0;
1970         /* XXX: header size can vary due to auth type, IPv6, etc. */
1971         xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
1972
1973         xprt->bind_timeout = XS_BIND_TO;
1974         xprt->connect_timeout = XS_UDP_CONN_TO;
1975         xprt->reestablish_timeout = XS_UDP_REEST_TO;
1976         xprt->idle_timeout = XS_IDLE_DISC_TO;
1977
1978         xprt->ops = &xs_udp_ops;
1979
1980         xprt->timeout = &xs_udp_default_timeout;
1981
1982         switch (addr->sa_family) {
1983         case AF_INET:
1984                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
1985                         xprt_set_bound(xprt);
1986
1987                 INIT_DELAYED_WORK(&transport->connect_worker,
1988                                         xs_udp_connect_worker4);
1989                 xs_format_ipv4_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
1990                 break;
1991         case AF_INET6:
1992                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
1993                         xprt_set_bound(xprt);
1994
1995                 INIT_DELAYED_WORK(&transport->connect_worker,
1996                                         xs_udp_connect_worker6);
1997                 xs_format_ipv6_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
1998                 break;
1999         default:
2000                 kfree(xprt);
2001                 return ERR_PTR(-EAFNOSUPPORT);
2002         }
2003
2004         dprintk("RPC:       set up transport to address %s\n",
2005                         xprt->address_strings[RPC_DISPLAY_ALL]);
2006
2007         if (try_module_get(THIS_MODULE))
2008                 return xprt;
2009
2010         kfree(xprt->slot);
2011         kfree(xprt);
2012         return ERR_PTR(-EINVAL);
2013 }
2014
2015 static const struct rpc_timeout xs_tcp_default_timeout = {
2016         .to_initval = 60 * HZ,
2017         .to_maxval = 60 * HZ,
2018         .to_retries = 2,
2019 };
2020
2021 /**
2022  * xs_setup_tcp - Set up transport to use a TCP socket
2023  * @args: rpc transport creation arguments
2024  *
2025  */
2026 static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
2027 {
2028         struct sockaddr *addr = args->dstaddr;
2029         struct rpc_xprt *xprt;
2030         struct sock_xprt *transport;
2031
2032         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries);
2033         if (IS_ERR(xprt))
2034                 return xprt;
2035         transport = container_of(xprt, struct sock_xprt, xprt);
2036
2037         xprt->prot = IPPROTO_TCP;
2038         xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2039         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2040
2041         xprt->bind_timeout = XS_BIND_TO;
2042         xprt->connect_timeout = XS_TCP_CONN_TO;
2043         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2044         xprt->idle_timeout = XS_IDLE_DISC_TO;
2045
2046         xprt->ops = &xs_tcp_ops;
2047         xprt->timeout = &xs_tcp_default_timeout;
2048
2049         switch (addr->sa_family) {
2050         case AF_INET:
2051                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2052                         xprt_set_bound(xprt);
2053
2054                 INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_connect_worker4);
2055                 xs_format_ipv4_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
2056                 break;
2057         case AF_INET6:
2058                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2059                         xprt_set_bound(xprt);
2060
2061                 INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_connect_worker6);
2062                 xs_format_ipv6_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
2063                 break;
2064         default:
2065                 kfree(xprt);
2066                 return ERR_PTR(-EAFNOSUPPORT);
2067         }
2068
2069         dprintk("RPC:       set up transport to address %s\n",
2070                         xprt->address_strings[RPC_DISPLAY_ALL]);
2071
2072         if (try_module_get(THIS_MODULE))
2073                 return xprt;
2074
2075         kfree(xprt->slot);
2076         kfree(xprt);
2077         return ERR_PTR(-EINVAL);
2078 }
2079
2080 static struct xprt_class        xs_udp_transport = {
2081         .list           = LIST_HEAD_INIT(xs_udp_transport.list),
2082         .name           = "udp",
2083         .owner          = THIS_MODULE,
2084         .ident          = IPPROTO_UDP,
2085         .setup          = xs_setup_udp,
2086 };
2087
2088 static struct xprt_class        xs_tcp_transport = {
2089         .list           = LIST_HEAD_INIT(xs_tcp_transport.list),
2090         .name           = "tcp",
2091         .owner          = THIS_MODULE,
2092         .ident          = IPPROTO_TCP,
2093         .setup          = xs_setup_tcp,
2094 };
2095
2096 /**
2097  * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
2098  *
2099  */
2100 int init_socket_xprt(void)
2101 {
2102 #ifdef RPC_DEBUG
2103         if (!sunrpc_table_header)
2104                 sunrpc_table_header = register_sysctl_table(sunrpc_table);
2105 #endif
2106
2107         xprt_register_transport(&xs_udp_transport);
2108         xprt_register_transport(&xs_tcp_transport);
2109
2110         return 0;
2111 }
2112
2113 /**
2114  * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
2115  *
2116  */
2117 void cleanup_socket_xprt(void)
2118 {
2119 #ifdef RPC_DEBUG
2120         if (sunrpc_table_header) {
2121                 unregister_sysctl_table(sunrpc_table_header);
2122                 sunrpc_table_header = NULL;
2123         }
2124 #endif
2125
2126         xprt_unregister_transport(&xs_udp_transport);
2127         xprt_unregister_transport(&xs_tcp_transport);
2128 }