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 Software <alan@redhat.com>
   7  * TCP send fixes (C) 1998 Red Hat Software <alan@redhat.com>
   8  * TCP NFS related read + write fixes
   9  *  (C) 1999 Dave Airlie, University of Limerick, Ireland <airlied@linux.ie>
  10  *
  11  * Rewrite of larges part of the code in order to stabilize TCP stuff.
  12  * Fix behaviour when socket buffer is full.
  13  *  (C) 1999 Trond Myklebust <trond.myklebust@fys.uio.no>
  14  *
  15  * IP socket transport implementation, (C) 2005 Chuck Lever <cel@netapp.com>
  16  *
  17  * IPv6 support contributed by Gilles Quillard, Bull Open Source, 2005.
  18  *   <gilles.quillard@bull.net>
  19  */
  20
  21 #include <linux/types.h>
  22 #include <linux/slab.h>
  23 #include <linux/module.h>
  24 #include <linux/capability.h>
  25 #include <linux/pagemap.h>
  26 #include <linux/errno.h>
  27 #include <linux/socket.h>
  28 #include <linux/in.h>
  29 #include <linux/net.h>
  30 #include <linux/mm.h>
  31 #include <linux/udp.h>
  32 #include <linux/tcp.h>
  33 #include <linux/sunrpc/clnt.h>
  34 #include <linux/sunrpc/sched.h>
  35 #include <linux/sunrpc/xprtsock.h>
  36 #include <linux/file.h>
  37
  38 #include <net/sock.h>
  39 #include <net/checksum.h>
  40 #include <net/udp.h>
  41 #include <net/tcp.h>
  42
  43 /*
  44  * xprtsock tunables
  45  */
  46 unsigned int xprt_udp_slot_table_entries = RPC_DEF_SLOT_TABLE;
  47 unsigned int xprt_tcp_slot_table_entries = RPC_DEF_SLOT_TABLE;
  48
  49 unsigned int xprt_min_resvport = RPC_DEF_MIN_RESVPORT;
  50 unsigned int xprt_max_resvport = RPC_DEF_MAX_RESVPORT;
  51
  52 /*
  53  * We can register our own files under /proc/sys/sunrpc by
  54  * calling register_sysctl_table() again.  The files in that
  55  * directory become the union of all files registered there.
  56  *
  57  * We simply need to make sure that we don't collide with
  58  * someone else's file names!
  59  */
  60
  61 #ifdef RPC_DEBUG
  62
  63 static unsigned int min_slot_table_size = RPC_MIN_SLOT_TABLE;
  64 static unsigned int max_slot_table_size = RPC_MAX_SLOT_TABLE;
  65 static unsigned int xprt_min_resvport_limit = RPC_MIN_RESVPORT;
  66 static unsigned int xprt_max_resvport_limit = RPC_MAX_RESVPORT;
  67
  68 static struct ctl_table_header *sunrpc_table_header;
  69
  70 /*
  71  * FIXME: changing the UDP slot table size should also resize the UDP
  72  *        socket buffers for existing UDP transports
  73  */
  74 static ctl_table xs_tunables_table[] = {
  75         {
  76                 .ctl_name       = CTL_SLOTTABLE_UDP,
  77                 .procname       = "udp_slot_table_entries",
  78                 .data           = &xprt_udp_slot_table_entries,
  79                 .maxlen         = sizeof(unsigned int),
  80                 .mode           = 0644,
  81                 .proc_handler   = &proc_dointvec_minmax,
  82                 .strategy       = &sysctl_intvec,
  83                 .extra1         = &min_slot_table_size,
  84                 .extra2         = &max_slot_table_size
  85         },
  86         {
  87                 .ctl_name       = CTL_SLOTTABLE_TCP,
  88                 .procname       = "tcp_slot_table_entries",
  89                 .data           = &xprt_tcp_slot_table_entries,
  90                 .maxlen         = sizeof(unsigned int),
  91                 .mode           = 0644,
  92                 .proc_handler   = &proc_dointvec_minmax,
  93                 .strategy       = &sysctl_intvec,
  94                 .extra1         = &min_slot_table_size,
  95                 .extra2         = &max_slot_table_size
  96         },
  97         {
  98                 .ctl_name       = CTL_MIN_RESVPORT,
  99                 .procname       = "min_resvport",
 100                 .data           = &xprt_min_resvport,
 101                 .maxlen         = sizeof(unsigned int),
 102                 .mode           = 0644,
 103                 .proc_handler   = &proc_dointvec_minmax,
 104                 .strategy       = &sysctl_intvec,
 105                 .extra1         = &xprt_min_resvport_limit,
 106                 .extra2         = &xprt_max_resvport_limit
 107         },
 108         {
 109                 .ctl_name       = CTL_MAX_RESVPORT,
 110                 .procname       = "max_resvport",
 111                 .data           = &xprt_max_resvport,
 112                 .maxlen         = sizeof(unsigned int),
 113                 .mode           = 0644,
 114                 .proc_handler   = &proc_dointvec_minmax,
 115                 .strategy       = &sysctl_intvec,
 116                 .extra1         = &xprt_min_resvport_limit,
 117                 .extra2         = &xprt_max_resvport_limit
 118         },
 119         {
 120                 .ctl_name = 0,
 121         },
 122 };
 123
 124 static ctl_table sunrpc_table[] = {
 125         {
 126                 .ctl_name       = CTL_SUNRPC,
 127                 .procname       = "sunrpc",
 128                 .mode           = 0555,
 129                 .child          = xs_tunables_table
 130         },
 131         {
 132                 .ctl_name = 0,
 133         },
 134 };
 135
 136 #endif
 137
 138 /*
 139  * How many times to try sending a request on a socket before waiting
 140  * for the socket buffer to clear.
 141  */
 142 #define XS_SENDMSG_RETRY        (10U)
 143
 144 /*
 145  * Time out for an RPC UDP socket connect.  UDP socket connects are
 146  * synchronous, but we set a timeout anyway in case of resource
 147  * exhaustion on the local host.
 148  */
 149 #define XS_UDP_CONN_TO          (5U * HZ)
 150
 151 /*
 152  * Wait duration for an RPC TCP connection to be established.  Solaris
 153  * NFS over TCP uses 60 seconds, for example, which is in line with how
 154  * long a server takes to reboot.
 155  */
 156 #define XS_TCP_CONN_TO          (60U * HZ)
 157
 158 /*
 159  * Wait duration for a reply from the RPC portmapper.
 160  */
 161 #define XS_BIND_TO              (60U * HZ)
 162
 163 /*
 164  * Delay if a UDP socket connect error occurs.  This is most likely some
 165  * kind of resource problem on the local host.
 166  */
 167 #define XS_UDP_REEST_TO         (2U * HZ)
 168
 169 /*
 170  * The reestablish timeout allows clients to delay for a bit before attempting
 171  * to reconnect to a server that just dropped our connection.
 172  *
 173  * We implement an exponential backoff when trying to reestablish a TCP
 174  * transport connection with the server.  Some servers like to drop a TCP
 175  * connection when they are overworked, so we start with a short timeout and
 176  * increase over time if the server is down or not responding.
 177  */
 178 #define XS_TCP_INIT_REEST_TO    (3U * HZ)
 179 #define XS_TCP_MAX_REEST_TO     (5U * 60 * HZ)
 180
 181 /*
 182  * TCP idle timeout; client drops the transport socket if it is idle
 183  * for this long.  Note that we also timeout UDP sockets to prevent
 184  * holding port numbers when there is no RPC traffic.
 185  */
 186 #define XS_IDLE_DISC_TO         (5U * 60 * HZ)
 187
 188 #ifdef RPC_DEBUG
 189 # undef  RPC_DEBUG_DATA
 190 # define RPCDBG_FACILITY        RPCDBG_TRANS
 191 #endif
 192
 193 #ifdef RPC_DEBUG_DATA
 194 static void xs_pktdump(char *msg, u32 *packet, unsigned int count)
 195 {
 196         u8 *buf = (u8 *) packet;
 197         int j;
 198
 199         dprintk("RPC:       %s\n", msg);
 200         for (j = 0; j < count && j < 128; j += 4) {
 201                 if (!(j & 31)) {
 202                         if (j)
 203                                 dprintk("\n");
 204                         dprintk("0x%04x ", j);
 205                 }
 206                 dprintk("%02x%02x%02x%02x ",
 207                         buf[j], buf[j+1], buf[j+2], buf[j+3]);
 208         }
 209         dprintk("\n");
 210 }
 211 #else
 212 static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count)
 213 {
 214         /* NOP */
 215 }
 216 #endif
 217
 218 struct sock_xprt {
 219         struct rpc_xprt         xprt;
 220
 221         /*
 222          * Network layer
 223          */
 224         struct socket *         sock;
 225         struct sock *           inet;
 226
 227         /*
 228          * State of TCP reply receive
 229          */
 230         __be32                  tcp_fraghdr,
 231                                 tcp_xid;
 232
 233         u32                     tcp_offset,
 234                                 tcp_reclen;
 235
 236         unsigned long           tcp_copied,
 237                                 tcp_flags;
 238
 239         /*
 240          * Connection of transports
 241          */
 242         struct delayed_work     connect_worker;
 243         struct sockaddr_storage addr;
 244         unsigned short          port;
 245
 246         /*
 247          * UDP socket buffer size parameters
 248          */
 249         size_t                  rcvsize,
 250                                 sndsize;
 251
 252         /*
 253          * Saved socket callback addresses
 254          */
 255         void                    (*old_data_ready)(struct sock *, int);
 256         void                    (*old_state_change)(struct sock *);
 257         void                    (*old_write_space)(struct sock *);
 258 };
 259
 260 /*
 261  * TCP receive state flags
 262  */
 263 #define TCP_RCV_LAST_FRAG       (1UL << 0)
 264 #define TCP_RCV_COPY_FRAGHDR    (1UL << 1)
 265 #define TCP_RCV_COPY_XID        (1UL << 2)
 266 #define TCP_RCV_COPY_DATA       (1UL << 3)
 267
 268 static inline struct sockaddr *xs_addr(struct rpc_xprt *xprt)
 269 {
 270         return (struct sockaddr *) &xprt->addr;
 271 }
 272
 273 static inline struct sockaddr_in *xs_addr_in(struct rpc_xprt *xprt)
 274 {
 275         return (struct sockaddr_in *) &xprt->addr;
 276 }
 277
 278 static inline struct sockaddr_in6 *xs_addr_in6(struct rpc_xprt *xprt)
 279 {
 280         return (struct sockaddr_in6 *) &xprt->addr;
 281 }
 282
 283 static void xs_format_ipv4_peer_addresses(struct rpc_xprt *xprt)
 284 {
 285         struct sockaddr_in *addr = xs_addr_in(xprt);
 286         char *buf;
 287
 288         buf = kzalloc(20, GFP_KERNEL);
 289         if (buf) {
 290                 snprintf(buf, 20, NIPQUAD_FMT,
 291                                 NIPQUAD(addr->sin_addr.s_addr));
 292         }
 293         xprt->address_strings[RPC_DISPLAY_ADDR] = buf;
 294
 295         buf = kzalloc(8, GFP_KERNEL);
 296         if (buf) {
 297                 snprintf(buf, 8, "%u",
 298                                 ntohs(addr->sin_port));
 299         }
 300         xprt->address_strings[RPC_DISPLAY_PORT] = buf;
 301
 302         buf = kzalloc(8, GFP_KERNEL);
 303         if (buf) {
 304                 if (xprt->prot == IPPROTO_UDP)
 305                         snprintf(buf, 8, "udp");
 306                 else
 307                         snprintf(buf, 8, "tcp");
 308         }
 309         xprt->address_strings[RPC_DISPLAY_PROTO] = buf;
 310
 311         buf = kzalloc(48, GFP_KERNEL);
 312         if (buf) {
 313                 snprintf(buf, 48, "addr="NIPQUAD_FMT" port=%u proto=%s",
 314                         NIPQUAD(addr->sin_addr.s_addr),
 315                         ntohs(addr->sin_port),
 316                         xprt->prot == IPPROTO_UDP ? "udp" : "tcp");
 317         }
 318         xprt->address_strings[RPC_DISPLAY_ALL] = buf;
 319
 320         buf = kzalloc(10, GFP_KERNEL);
 321         if (buf) {
 322                 snprintf(buf, 10, "%02x%02x%02x%02x",
 323                                 NIPQUAD(addr->sin_addr.s_addr));
 324         }
 325         xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = buf;
 326
 327         buf = kzalloc(8, GFP_KERNEL);
 328         if (buf) {
 329                 snprintf(buf, 8, "%4hx",
 330                                 ntohs(addr->sin_port));
 331         }
 332         xprt->address_strings[RPC_DISPLAY_HEX_PORT] = buf;
 333
 334         buf = kzalloc(30, GFP_KERNEL);
 335         if (buf) {
 336                 snprintf(buf, 30, NIPQUAD_FMT".%u.%u",
 337                                 NIPQUAD(addr->sin_addr.s_addr),
 338                                 ntohs(addr->sin_port) >> 8,
 339                                 ntohs(addr->sin_port) & 0xff);
 340         }
 341         xprt->address_strings[RPC_DISPLAY_UNIVERSAL_ADDR] = buf;
 342
 343         xprt->address_strings[RPC_DISPLAY_NETID] =
 344                 kstrdup(xprt->prot == IPPROTO_UDP ?
 345                         RPCBIND_NETID_UDP : RPCBIND_NETID_TCP, GFP_KERNEL);
 346 }
 347
 348 static void xs_format_ipv6_peer_addresses(struct rpc_xprt *xprt)
 349 {
 350         struct sockaddr_in6 *addr = xs_addr_in6(xprt);
 351         char *buf;
 352
 353         buf = kzalloc(40, GFP_KERNEL);
 354         if (buf) {
 355                 snprintf(buf, 40, NIP6_FMT,
 356                                 NIP6(addr->sin6_addr));
 357         }
 358         xprt->address_strings[RPC_DISPLAY_ADDR] = buf;
 359
 360         buf = kzalloc(8, GFP_KERNEL);
 361         if (buf) {
 362                 snprintf(buf, 8, "%u",
 363                                 ntohs(addr->sin6_port));
 364         }
 365         xprt->address_strings[RPC_DISPLAY_PORT] = buf;
 366
 367         buf = kzalloc(8, GFP_KERNEL);
 368         if (buf) {
 369                 if (xprt->prot == IPPROTO_UDP)
 370                         snprintf(buf, 8, "udp");
 371                 else
 372                         snprintf(buf, 8, "tcp");
 373         }
 374         xprt->address_strings[RPC_DISPLAY_PROTO] = buf;
 375
 376         buf = kzalloc(64, GFP_KERNEL);
 377         if (buf) {
 378                 snprintf(buf, 64, "addr="NIP6_FMT" port=%u proto=%s",
 379                                 NIP6(addr->sin6_addr),
 380                                 ntohs(addr->sin6_port),
 381                                 xprt->prot == IPPROTO_UDP ? "udp" : "tcp");
 382         }
 383         xprt->address_strings[RPC_DISPLAY_ALL] = buf;
 384
 385         buf = kzalloc(36, GFP_KERNEL);
 386         if (buf) {
 387                 snprintf(buf, 36, NIP6_SEQFMT,
 388                                 NIP6(addr->sin6_addr));
 389         }
 390         xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = buf;
 391
 392         buf = kzalloc(8, GFP_KERNEL);
 393         if (buf) {
 394                 snprintf(buf, 8, "%4hx",
 395                                 ntohs(addr->sin6_port));
 396         }
 397         xprt->address_strings[RPC_DISPLAY_HEX_PORT] = buf;
 398
 399         buf = kzalloc(50, GFP_KERNEL);
 400         if (buf) {
 401                 snprintf(buf, 50, NIP6_FMT".%u.%u",
 402                                 NIP6(addr->sin6_addr),
 403                                 ntohs(addr->sin6_port) >> 8,
 404                                 ntohs(addr->sin6_port) & 0xff);
 405         }
 406         xprt->address_strings[RPC_DISPLAY_UNIVERSAL_ADDR] = buf;
 407
 408         xprt->address_strings[RPC_DISPLAY_NETID] =
 409                 kstrdup(xprt->prot == IPPROTO_UDP ?
 410                         RPCBIND_NETID_UDP6 : RPCBIND_NETID_TCP6, GFP_KERNEL);
 411 }
 412
 413 static void xs_free_peer_addresses(struct rpc_xprt *xprt)
 414 {
 415         int i;
 416
 417         for (i = 0; i < RPC_DISPLAY_MAX; i++)
 418                 kfree(xprt->address_strings[i]);
 419 }
 420
 421 #define XS_SENDMSG_FLAGS        (MSG_DONTWAIT | MSG_NOSIGNAL)
 422
 423 static int xs_send_kvec(struct socket *sock, struct sockaddr *addr, int addrlen, struct kvec *vec, unsigned int base, int more)
 424 {
 425         struct msghdr msg = {
 426                 .msg_name       = addr,
 427                 .msg_namelen    = addrlen,
 428                 .msg_flags      = XS_SENDMSG_FLAGS | (more ? MSG_MORE : 0),
 429         };
 430         struct kvec iov = {
 431                 .iov_base       = vec->iov_base + base,
 432                 .iov_len        = vec->iov_len - base,
 433         };
 434
 435         if (iov.iov_len != 0)
 436                 return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
 437         return kernel_sendmsg(sock, &msg, NULL, 0, 0);
 438 }
 439
 440 static int xs_send_pagedata(struct socket *sock, struct xdr_buf *xdr, unsigned int base, int more)
 441 {
 442         struct page **ppage;
 443         unsigned int remainder;
 444         int err, sent = 0;
 445
 446         remainder = xdr->page_len - base;
 447         base += xdr->page_base;
 448         ppage = xdr->pages + (base >> PAGE_SHIFT);
 449         base &= ~PAGE_MASK;
 450         for(;;) {
 451                 unsigned int len = min_t(unsigned int, PAGE_SIZE - base, remainder);
 452                 int flags = XS_SENDMSG_FLAGS;
 453
 454                 remainder -= len;
 455                 if (remainder != 0 || more)
 456                         flags |= MSG_MORE;
 457                 err = sock->ops->sendpage(sock, *ppage, base, len, flags);
 458                 if (remainder == 0 || err != len)
 459                         break;
 460                 sent += err;
 461                 ppage++;
 462                 base = 0;
 463         }
 464         if (sent == 0)
 465                 return err;
 466         if (err > 0)
 467                 sent += err;
 468         return sent;
 469 }
 470
 471 /**
 472  * xs_sendpages - write pages directly to a socket
 473  * @sock: socket to send on
 474  * @addr: UDP only -- address of destination
 475  * @addrlen: UDP only -- length of destination address
 476  * @xdr: buffer containing this request
 477  * @base: starting position in the buffer
 478  *
 479  */
 480 static int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base)
 481 {
 482         unsigned int remainder = xdr->len - base;
 483         int err, sent = 0;
 484
 485         if (unlikely(!sock))
 486                 return -ENOTCONN;
 487
 488         clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags);
 489         if (base != 0) {
 490                 addr = NULL;
 491                 addrlen = 0;
 492         }
 493
 494         if (base < xdr->head[0].iov_len || addr != NULL) {
 495                 unsigned int len = xdr->head[0].iov_len - base;
 496                 remainder -= len;
 497                 err = xs_send_kvec(sock, addr, addrlen, &xdr->head[0], base, remainder != 0);
 498                 if (remainder == 0 || err != len)
 499                         goto out;
 500                 sent += err;
 501                 base = 0;
 502         } else
 503                 base -= xdr->head[0].iov_len;
 504
 505         if (base < xdr->page_len) {
 506                 unsigned int len = xdr->page_len - base;
 507                 remainder -= len;
 508                 err = xs_send_pagedata(sock, xdr, base, remainder != 0);
 509                 if (remainder == 0 || err != len)
 510                         goto out;
 511                 sent += err;
 512                 base = 0;
 513         } else
 514                 base -= xdr->page_len;
 515
 516         if (base >= xdr->tail[0].iov_len)
 517                 return sent;
 518         err = xs_send_kvec(sock, NULL, 0, &xdr->tail[0], base, 0);
 519 out:
 520         if (sent == 0)
 521                 return err;
 522         if (err > 0)
 523                 sent += err;
 524         return sent;
 525 }
 526
 527 /**
 528  * xs_nospace - place task on wait queue if transmit was incomplete
 529  * @task: task to put to sleep
 530  *
 531  */
 532 static void xs_nospace(struct rpc_task *task)
 533 {
 534         struct rpc_rqst *req = task->tk_rqstp;
 535         struct rpc_xprt *xprt = req->rq_xprt;
 536         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 537
 538         dprintk("RPC: %5u xmit incomplete (%u left of %u)\n",
 539                         task->tk_pid, req->rq_slen - req->rq_bytes_sent,
 540                         req->rq_slen);
 541
 542         if (test_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags)) {
 543                 /* Protect against races with write_space */
 544                 spin_lock_bh(&xprt->transport_lock);
 545
 546                 /* Don't race with disconnect */
 547                 if (!xprt_connected(xprt))
 548                         task->tk_status = -ENOTCONN;
 549                 else if (test_bit(SOCK_NOSPACE, &transport->sock->flags))
 550                         xprt_wait_for_buffer_space(task);
 551
 552                 spin_unlock_bh(&xprt->transport_lock);
 553         } else
 554                 /* Keep holding the socket if it is blocked */
 555                 rpc_delay(task, HZ>>4);
 556 }
 557
 558 /**
 559  * xs_udp_send_request - write an RPC request to a UDP socket
 560  * @task: address of RPC task that manages the state of an RPC request
 561  *
 562  * Return values:
 563  *        0:    The request has been sent
 564  *   EAGAIN:    The socket was blocked, please call again later to
 565  *              complete the request
 566  * ENOTCONN:    Caller needs to invoke connect logic then call again
 567  *    other:    Some other error occured, the request was not sent
 568  */
 569 static int xs_udp_send_request(struct rpc_task *task)
 570 {
 571         struct rpc_rqst *req = task->tk_rqstp;
 572         struct rpc_xprt *xprt = req->rq_xprt;
 573         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 574         struct xdr_buf *xdr = &req->rq_snd_buf;
 575         int status;
 576
 577         xs_pktdump("packet data:",
 578                                 req->rq_svec->iov_base,
 579                                 req->rq_svec->iov_len);
 580
 581         req->rq_xtime = jiffies;
 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 -ENETUNREACH:
 600         case -EPIPE:
 601         case -ECONNREFUSED:
 602                 /* When the server has died, an ICMP port unreachable message
 603                  * prompts ECONNREFUSED. */
 604                 break;
 605         case -EAGAIN:
 606                 xs_nospace(task);
 607                 break;
 608         default:
 609                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
 610                         -status);
 611                 break;
 612         }
 613
 614         return status;
 615 }
 616
 617 static inline void xs_encode_tcp_record_marker(struct xdr_buf *buf)
 618 {
 619         u32 reclen = buf->len - sizeof(rpc_fraghdr);
 620         rpc_fraghdr *base = buf->head[0].iov_base;
 621         *base = htonl(RPC_LAST_STREAM_FRAGMENT | reclen);
 622 }
 623
 624 /**
 625  * xs_tcp_send_request - write an RPC request to a TCP socket
 626  * @task: address of RPC task that manages the state of an RPC request
 627  *
 628  * Return values:
 629  *        0:    The request has been sent
 630  *   EAGAIN:    The socket was blocked, please call again later to
 631  *              complete the request
 632  * ENOTCONN:    Caller needs to invoke connect logic then call again
 633  *    other:    Some other error occured, the request was not sent
 634  *
 635  * XXX: In the case of soft timeouts, should we eventually give up
 636  *      if sendmsg is not able to make progress?
 637  */
 638 static int xs_tcp_send_request(struct rpc_task *task)
 639 {
 640         struct rpc_rqst *req = task->tk_rqstp;
 641         struct rpc_xprt *xprt = req->rq_xprt;
 642         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 643         struct xdr_buf *xdr = &req->rq_snd_buf;
 644         int status;
 645         unsigned int retry = 0;
 646
 647         xs_encode_tcp_record_marker(&req->rq_snd_buf);
 648
 649         xs_pktdump("packet data:",
 650                                 req->rq_svec->iov_base,
 651                                 req->rq_svec->iov_len);
 652
 653         /* Continue transmitting the packet/record. We must be careful
 654          * to cope with writespace callbacks arriving _after_ we have
 655          * called sendmsg(). */
 656         while (1) {
 657                 req->rq_xtime = jiffies;
 658                 status = xs_sendpages(transport->sock,
 659                                         NULL, 0, xdr, req->rq_bytes_sent);
 660
 661                 dprintk("RPC:       xs_tcp_send_request(%u) = %d\n",
 662                                 xdr->len - req->rq_bytes_sent, status);
 663
 664                 if (unlikely(status < 0))
 665                         break;
 666
 667                 /* If we've sent the entire packet, immediately
 668                  * reset the count of bytes sent. */
 669                 req->rq_bytes_sent += status;
 670                 task->tk_bytes_sent += status;
 671                 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
 672                         req->rq_bytes_sent = 0;
 673                         return 0;
 674                 }
 675
 676                 status = -EAGAIN;
 677                 if (retry++ > XS_SENDMSG_RETRY)
 678                         break;
 679         }
 680
 681         switch (status) {
 682         case -EAGAIN:
 683                 xs_nospace(task);
 684                 break;
 685         case -ECONNREFUSED:
 686         case -ECONNRESET:
 687         case -ENOTCONN:
 688         case -EPIPE:
 689                 status = -ENOTCONN;
 690                 break;
 691         default:
 692                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
 693                         -status);
 694                 xprt_disconnect(xprt);
 695                 break;
 696         }
 697
 698         return status;
 699 }
 700
 701 /**
 702  * xs_tcp_release_xprt - clean up after a tcp transmission
 703  * @xprt: transport
 704  * @task: rpc task
 705  *
 706  * This cleans up if an error causes us to abort the transmission of a request.
 707  * In this case, the socket may need to be reset in order to avoid confusing
 708  * the server.
 709  */
 710 static void xs_tcp_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
 711 {
 712         struct rpc_rqst *req;
 713
 714         if (task != xprt->snd_task)
 715                 return;
 716         if (task == NULL)
 717                 goto out_release;
 718         req = task->tk_rqstp;
 719         if (req->rq_bytes_sent == 0)
 720                 goto out_release;
 721         if (req->rq_bytes_sent == req->rq_snd_buf.len)
 722                 goto out_release;
 723         set_bit(XPRT_CLOSE_WAIT, &task->tk_xprt->state);
 724 out_release:
 725         xprt_release_xprt(xprt, task);
 726 }
 727
 728 /**
 729  * xs_close - close a socket
 730  * @xprt: transport
 731  *
 732  * This is used when all requests are complete; ie, no DRC state remains
 733  * on the server we want to save.
 734  */
 735 static void xs_close(struct rpc_xprt *xprt)
 736 {
 737         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 738         struct socket *sock = transport->sock;
 739         struct sock *sk = transport->inet;
 740
 741         if (!sk)
 742                 goto clear_close_wait;
 743
 744         dprintk("RPC:       xs_close xprt %p\n", xprt);
 745
 746         write_lock_bh(&sk->sk_callback_lock);
 747         transport->inet = NULL;
 748         transport->sock = NULL;
 749
 750         sk->sk_user_data = NULL;
 751         sk->sk_data_ready = transport->old_data_ready;
 752         sk->sk_state_change = transport->old_state_change;
 753         sk->sk_write_space = transport->old_write_space;
 754         write_unlock_bh(&sk->sk_callback_lock);
 755
 756         sk->sk_no_check = 0;
 757
 758         sock_release(sock);
 759 clear_close_wait:
 760         smp_mb__before_clear_bit();
 761         clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
 762         smp_mb__after_clear_bit();
 763 }
 764
 765 /**
 766  * xs_destroy - prepare to shutdown a transport
 767  * @xprt: doomed transport
 768  *
 769  */
 770 static void xs_destroy(struct rpc_xprt *xprt)
 771 {
 772         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 773
 774         dprintk("RPC:       xs_destroy xprt %p\n", xprt);
 775
 776         cancel_rearming_delayed_work(&transport->connect_worker);
 777
 778         xprt_disconnect(xprt);
 779         xs_close(xprt);
 780         xs_free_peer_addresses(xprt);
 781         kfree(xprt->slot);
 782         kfree(xprt);
 783         module_put(THIS_MODULE);
 784 }
 785
 786 static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
 787 {
 788         return (struct rpc_xprt *) sk->sk_user_data;
 789 }
 790
 791 /**
 792  * xs_udp_data_ready - "data ready" callback for UDP sockets
 793  * @sk: socket with data to read
 794  * @len: how much data to read
 795  *
 796  */
 797 static void xs_udp_data_ready(struct sock *sk, int len)
 798 {
 799         struct rpc_task *task;
 800         struct rpc_xprt *xprt;
 801         struct rpc_rqst *rovr;
 802         struct sk_buff *skb;
 803         int err, repsize, copied;
 804         u32 _xid;
 805         __be32 *xp;
 806
 807         read_lock(&sk->sk_callback_lock);
 808         dprintk("RPC:       xs_udp_data_ready...\n");
 809         if (!(xprt = xprt_from_sock(sk)))
 810                 goto out;
 811
 812         if ((skb = skb_recv_datagram(sk, 0, 1, &err)) == NULL)
 813                 goto out;
 814
 815         if (xprt->shutdown)
 816                 goto dropit;
 817
 818         repsize = skb->len - sizeof(struct udphdr);
 819         if (repsize < 4) {
 820                 dprintk("RPC:       impossible RPC reply size %d!\n", repsize);
 821                 goto dropit;
 822         }
 823
 824         /* Copy the XID from the skb... */
 825         xp = skb_header_pointer(skb, sizeof(struct udphdr),
 826                                 sizeof(_xid), &_xid);
 827         if (xp == NULL)
 828                 goto dropit;
 829
 830         /* Look up and lock the request corresponding to the given XID */
 831         spin_lock(&xprt->transport_lock);
 832         rovr = xprt_lookup_rqst(xprt, *xp);
 833         if (!rovr)
 834                 goto out_unlock;
 835         task = rovr->rq_task;
 836
 837         if ((copied = rovr->rq_private_buf.buflen) > repsize)
 838                 copied = repsize;
 839
 840         /* Suck it into the iovec, verify checksum if not done by hw. */
 841         if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
 842                 UDPX_INC_STATS_BH(sk, UDP_MIB_INERRORS);
 843                 goto out_unlock;
 844         }
 845
 846         UDPX_INC_STATS_BH(sk, UDP_MIB_INDATAGRAMS);
 847
 848         /* Something worked... */
 849         dst_confirm(skb->dst);
 850
 851         xprt_adjust_cwnd(task, copied);
 852         xprt_update_rtt(task);
 853         xprt_complete_rqst(task, copied);
 854
 855  out_unlock:
 856         spin_unlock(&xprt->transport_lock);
 857  dropit:
 858         skb_free_datagram(sk, skb);
 859  out:
 860         read_unlock(&sk->sk_callback_lock);
 861 }
 862
 863 static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
 864 {
 865         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 866         size_t len, used;
 867         char *p;
 868
 869         p = ((char *) &transport->tcp_fraghdr) + transport->tcp_offset;
 870         len = sizeof(transport->tcp_fraghdr) - transport->tcp_offset;
 871         used = xdr_skb_read_bits(desc, p, len);
 872         transport->tcp_offset += used;
 873         if (used != len)
 874                 return;
 875
 876         transport->tcp_reclen = ntohl(transport->tcp_fraghdr);
 877         if (transport->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
 878                 transport->tcp_flags |= TCP_RCV_LAST_FRAG;
 879         else
 880                 transport->tcp_flags &= ~TCP_RCV_LAST_FRAG;
 881         transport->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
 882
 883         transport->tcp_flags &= ~TCP_RCV_COPY_FRAGHDR;
 884         transport->tcp_offset = 0;
 885
 886         /* Sanity check of the record length */
 887         if (unlikely(transport->tcp_reclen < 4)) {
 888                 dprintk("RPC:       invalid TCP record fragment length\n");
 889                 xprt_disconnect(xprt);
 890                 return;
 891         }
 892         dprintk("RPC:       reading TCP record fragment of length %d\n",
 893                         transport->tcp_reclen);
 894 }
 895
 896 static void xs_tcp_check_fraghdr(struct sock_xprt *transport)
 897 {
 898         if (transport->tcp_offset == transport->tcp_reclen) {
 899                 transport->tcp_flags |= TCP_RCV_COPY_FRAGHDR;
 900                 transport->tcp_offset = 0;
 901                 if (transport->tcp_flags & TCP_RCV_LAST_FRAG) {
 902                         transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
 903                         transport->tcp_flags |= TCP_RCV_COPY_XID;
 904                         transport->tcp_copied = 0;
 905                 }
 906         }
 907 }
 908
 909 static inline void xs_tcp_read_xid(struct sock_xprt *transport, struct xdr_skb_reader *desc)
 910 {
 911         size_t len, used;
 912         char *p;
 913
 914         len = sizeof(transport->tcp_xid) - transport->tcp_offset;
 915         dprintk("RPC:       reading XID (%Zu bytes)\n", len);
 916         p = ((char *) &transport->tcp_xid) + transport->tcp_offset;
 917         used = xdr_skb_read_bits(desc, p, len);
 918         transport->tcp_offset += used;
 919         if (used != len)
 920                 return;
 921         transport->tcp_flags &= ~TCP_RCV_COPY_XID;
 922         transport->tcp_flags |= TCP_RCV_COPY_DATA;
 923         transport->tcp_copied = 4;
 924         dprintk("RPC:       reading reply for XID %08x\n",
 925                         ntohl(transport->tcp_xid));
 926         xs_tcp_check_fraghdr(transport);
 927 }
 928
 929 static inline void xs_tcp_read_request(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
 930 {
 931         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 932         struct rpc_rqst *req;
 933         struct xdr_buf *rcvbuf;
 934         size_t len;
 935         ssize_t r;
 936
 937         /* Find and lock the request corresponding to this xid */
 938         spin_lock(&xprt->transport_lock);
 939         req = xprt_lookup_rqst(xprt, transport->tcp_xid);
 940         if (!req) {
 941                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
 942                 dprintk("RPC:       XID %08x request not found!\n",
 943                                 ntohl(transport->tcp_xid));
 944                 spin_unlock(&xprt->transport_lock);
 945                 return;
 946         }
 947
 948         rcvbuf = &req->rq_private_buf;
 949         len = desc->count;
 950         if (len > transport->tcp_reclen - transport->tcp_offset) {
 951                 struct xdr_skb_reader my_desc;
 952
 953                 len = transport->tcp_reclen - transport->tcp_offset;
 954                 memcpy(&my_desc, desc, sizeof(my_desc));
 955                 my_desc.count = len;
 956                 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
 957                                           &my_desc, xdr_skb_read_bits);
 958                 desc->count -= r;
 959                 desc->offset += r;
 960         } else
 961                 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
 962                                           desc, xdr_skb_read_bits);
 963
 964         if (r > 0) {
 965                 transport->tcp_copied += r;
 966                 transport->tcp_offset += r;
 967         }
 968         if (r != len) {
 969                 /* Error when copying to the receive buffer,
 970                  * usually because we weren't able to allocate
 971                  * additional buffer pages. All we can do now
 972                  * is turn off TCP_RCV_COPY_DATA, so the request
 973                  * will not receive any additional updates,
 974                  * and time out.
 975                  * Any remaining data from this record will
 976                  * be discarded.
 977                  */
 978                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
 979                 dprintk("RPC:       XID %08x truncated request\n",
 980                                 ntohl(transport->tcp_xid));
 981                 dprintk("RPC:       xprt = %p, tcp_copied = %lu, "
 982                                 "tcp_offset = %u, tcp_reclen = %u\n",
 983                                 xprt, transport->tcp_copied,
 984                                 transport->tcp_offset, transport->tcp_reclen);
 985                 goto out;
 986         }
 987
 988         dprintk("RPC:       XID %08x read %Zd bytes\n",
 989                         ntohl(transport->tcp_xid), r);
 990         dprintk("RPC:       xprt = %p, tcp_copied = %lu, tcp_offset = %u, "
 991                         "tcp_reclen = %u\n", xprt, transport->tcp_copied,
 992                         transport->tcp_offset, transport->tcp_reclen);
 993
 994         if (transport->tcp_copied == req->rq_private_buf.buflen)
 995                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
 996         else if (transport->tcp_offset == transport->tcp_reclen) {
 997                 if (transport->tcp_flags & TCP_RCV_LAST_FRAG)
 998                         transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
 999         }
1000
1001 out:
1002         if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1003                 xprt_complete_rqst(req->rq_task, transport->tcp_copied);
1004         spin_unlock(&xprt->transport_lock);
1005         xs_tcp_check_fraghdr(transport);
1006 }
1007
1008 static inline void xs_tcp_read_discard(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1009 {
1010         size_t len;
1011
1012         len = transport->tcp_reclen - transport->tcp_offset;
1013         if (len > desc->count)
1014                 len = desc->count;
1015         desc->count -= len;
1016         desc->offset += len;
1017         transport->tcp_offset += len;
1018         dprintk("RPC:       discarded %Zu bytes\n", len);
1019         xs_tcp_check_fraghdr(transport);
1020 }
1021
1022 static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
1023 {
1024         struct rpc_xprt *xprt = rd_desc->arg.data;
1025         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1026         struct xdr_skb_reader desc = {
1027                 .skb    = skb,
1028                 .offset = offset,
1029                 .count  = len,
1030         };
1031
1032         dprintk("RPC:       xs_tcp_data_recv started\n");
1033         do {
1034                 /* Read in a new fragment marker if necessary */
1035                 /* Can we ever really expect to get completely empty fragments? */
1036                 if (transport->tcp_flags & TCP_RCV_COPY_FRAGHDR) {
1037                         xs_tcp_read_fraghdr(xprt, &desc);
1038                         continue;
1039                 }
1040                 /* Read in the xid if necessary */
1041                 if (transport->tcp_flags & TCP_RCV_COPY_XID) {
1042                         xs_tcp_read_xid(transport, &desc);
1043                         continue;
1044                 }
1045                 /* Read in the request data */
1046                 if (transport->tcp_flags & TCP_RCV_COPY_DATA) {
1047                         xs_tcp_read_request(xprt, &desc);
1048                         continue;
1049                 }
1050                 /* Skip over any trailing bytes on short reads */
1051                 xs_tcp_read_discard(transport, &desc);
1052         } while (desc.count);
1053         dprintk("RPC:       xs_tcp_data_recv done\n");
1054         return len - desc.count;
1055 }
1056
1057 /**
1058  * xs_tcp_data_ready - "data ready" callback for TCP sockets
1059  * @sk: socket with data to read
1060  * @bytes: how much data to read
1061  *
1062  */
1063 static void xs_tcp_data_ready(struct sock *sk, int bytes)
1064 {
1065         struct rpc_xprt *xprt;
1066         read_descriptor_t rd_desc;
1067
1068         dprintk("RPC:       xs_tcp_data_ready...\n");
1069
1070         read_lock(&sk->sk_callback_lock);
1071         if (!(xprt = xprt_from_sock(sk)))
1072                 goto out;
1073         if (xprt->shutdown)
1074                 goto out;
1075
1076         /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
1077         rd_desc.arg.data = xprt;
1078         rd_desc.count = 65536;
1079         tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
1080 out:
1081         read_unlock(&sk->sk_callback_lock);
1082 }
1083
1084 /**
1085  * xs_tcp_state_change - callback to handle TCP socket state changes
1086  * @sk: socket whose state has changed
1087  *
1088  */
1089 static void xs_tcp_state_change(struct sock *sk)
1090 {
1091         struct rpc_xprt *xprt;
1092
1093         read_lock(&sk->sk_callback_lock);
1094         if (!(xprt = xprt_from_sock(sk)))
1095                 goto out;
1096         dprintk("RPC:       xs_tcp_state_change client %p...\n", xprt);
1097         dprintk("RPC:       state %x conn %d dead %d zapped %d\n",
1098                         sk->sk_state, xprt_connected(xprt),
1099                         sock_flag(sk, SOCK_DEAD),
1100                         sock_flag(sk, SOCK_ZAPPED));
1101
1102         switch (sk->sk_state) {
1103         case TCP_ESTABLISHED:
1104                 spin_lock_bh(&xprt->transport_lock);
1105                 if (!xprt_test_and_set_connected(xprt)) {
1106                         struct sock_xprt *transport = container_of(xprt,
1107                                         struct sock_xprt, xprt);
1108
1109                         /* Reset TCP record info */
1110                         transport->tcp_offset = 0;
1111                         transport->tcp_reclen = 0;
1112                         transport->tcp_copied = 0;
1113                         transport->tcp_flags =
1114                                 TCP_RCV_COPY_FRAGHDR | TCP_RCV_COPY_XID;
1115
1116                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1117                         xprt_wake_pending_tasks(xprt, 0);
1118                 }
1119                 spin_unlock_bh(&xprt->transport_lock);
1120                 break;
1121         case TCP_SYN_SENT:
1122         case TCP_SYN_RECV:
1123                 break;
1124         case TCP_CLOSE_WAIT:
1125                 xprt_force_disconnect(xprt);
1126         default:
1127                 xprt_disconnect(xprt);
1128         }
1129  out:
1130         read_unlock(&sk->sk_callback_lock);
1131 }
1132
1133 /**
1134  * xs_udp_write_space - callback invoked when socket buffer space
1135  *                             becomes available
1136  * @sk: socket whose state has changed
1137  *
1138  * Called when more output buffer space is available for this socket.
1139  * We try not to wake our writers until they can make "significant"
1140  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1141  * with a bunch of small requests.
1142  */
1143 static void xs_udp_write_space(struct sock *sk)
1144 {
1145         read_lock(&sk->sk_callback_lock);
1146
1147         /* from net/core/sock.c:sock_def_write_space */
1148         if (sock_writeable(sk)) {
1149                 struct socket *sock;
1150                 struct rpc_xprt *xprt;
1151
1152                 if (unlikely(!(sock = sk->sk_socket)))
1153                         goto out;
1154                 if (unlikely(!(xprt = xprt_from_sock(sk))))
1155                         goto out;
1156                 if (unlikely(!test_and_clear_bit(SOCK_NOSPACE, &sock->flags)))
1157                         goto out;
1158
1159                 xprt_write_space(xprt);
1160         }
1161
1162  out:
1163         read_unlock(&sk->sk_callback_lock);
1164 }
1165
1166 /**
1167  * xs_tcp_write_space - callback invoked when socket buffer space
1168  *                             becomes available
1169  * @sk: socket whose state has changed
1170  *
1171  * Called when more output buffer space is available for this socket.
1172  * We try not to wake our writers until they can make "significant"
1173  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1174  * with a bunch of small requests.
1175  */
1176 static void xs_tcp_write_space(struct sock *sk)
1177 {
1178         read_lock(&sk->sk_callback_lock);
1179
1180         /* from net/core/stream.c:sk_stream_write_space */
1181         if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk)) {
1182                 struct socket *sock;
1183                 struct rpc_xprt *xprt;
1184
1185                 if (unlikely(!(sock = sk->sk_socket)))
1186                         goto out;
1187                 if (unlikely(!(xprt = xprt_from_sock(sk))))
1188                         goto out;
1189                 if (unlikely(!test_and_clear_bit(SOCK_NOSPACE, &sock->flags)))
1190                         goto out;
1191
1192                 xprt_write_space(xprt);
1193         }
1194
1195  out:
1196         read_unlock(&sk->sk_callback_lock);
1197 }
1198
1199 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
1200 {
1201         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1202         struct sock *sk = transport->inet;
1203
1204         if (transport->rcvsize) {
1205                 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1206                 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
1207         }
1208         if (transport->sndsize) {
1209                 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1210                 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1211                 sk->sk_write_space(sk);
1212         }
1213 }
1214
1215 /**
1216  * xs_udp_set_buffer_size - set send and receive limits
1217  * @xprt: generic transport
1218  * @sndsize: requested size of send buffer, in bytes
1219  * @rcvsize: requested size of receive buffer, in bytes
1220  *
1221  * Set socket send and receive buffer size limits.
1222  */
1223 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1224 {
1225         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1226
1227         transport->sndsize = 0;
1228         if (sndsize)
1229                 transport->sndsize = sndsize + 1024;
1230         transport->rcvsize = 0;
1231         if (rcvsize)
1232                 transport->rcvsize = rcvsize + 1024;
1233
1234         xs_udp_do_set_buffer_size(xprt);
1235 }
1236
1237 /**
1238  * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1239  * @task: task that timed out
1240  *
1241  * Adjust the congestion window after a retransmit timeout has occurred.
1242  */
1243 static void xs_udp_timer(struct rpc_task *task)
1244 {
1245         xprt_adjust_cwnd(task, -ETIMEDOUT);
1246 }
1247
1248 static unsigned short xs_get_random_port(void)
1249 {
1250         unsigned short range = xprt_max_resvport - xprt_min_resvport;
1251         unsigned short rand = (unsigned short) net_random() % range;
1252         return rand + xprt_min_resvport;
1253 }
1254
1255 /**
1256  * xs_set_port - reset the port number in the remote endpoint address
1257  * @xprt: generic transport
1258  * @port: new port number
1259  *
1260  */
1261 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1262 {
1263         struct sockaddr *addr = xs_addr(xprt);
1264
1265         dprintk("RPC:       setting port for xprt %p to %u\n", xprt, port);
1266
1267         switch (addr->sa_family) {
1268         case AF_INET:
1269                 ((struct sockaddr_in *)addr)->sin_port = htons(port);
1270                 break;
1271         case AF_INET6:
1272                 ((struct sockaddr_in6 *)addr)->sin6_port = htons(port);
1273                 break;
1274         default:
1275                 BUG();
1276         }
1277 }
1278
1279 static int xs_bind4(struct sock_xprt *transport, struct socket *sock)
1280 {
1281         struct sockaddr_in myaddr = {
1282                 .sin_family = AF_INET,
1283         };
1284         struct sockaddr_in *sa;
1285         int err;
1286         unsigned short port = transport->port;
1287
1288         if (!transport->xprt.resvport)
1289                 port = 0;
1290         sa = (struct sockaddr_in *)&transport->addr;
1291         myaddr.sin_addr = sa->sin_addr;
1292         do {
1293                 myaddr.sin_port = htons(port);
1294                 err = kernel_bind(sock, (struct sockaddr *) &myaddr,
1295                                                 sizeof(myaddr));
1296                 if (!transport->xprt.resvport)
1297                         break;
1298                 if (err == 0) {
1299                         transport->port = port;
1300                         break;
1301                 }
1302                 if (port <= xprt_min_resvport)
1303                         port = xprt_max_resvport;
1304                 else
1305                         port--;
1306         } while (err == -EADDRINUSE && port != transport->port);
1307         dprintk("RPC:       %s "NIPQUAD_FMT":%u: %s (%d)\n",
1308                         __FUNCTION__, NIPQUAD(myaddr.sin_addr),
1309                         port, err ? "failed" : "ok", err);
1310         return err;
1311 }
1312
1313 static int xs_bind6(struct sock_xprt *transport, struct socket *sock)
1314 {
1315         struct sockaddr_in6 myaddr = {
1316                 .sin6_family = AF_INET6,
1317         };
1318         struct sockaddr_in6 *sa;
1319         int err;
1320         unsigned short port = transport->port;
1321
1322         if (!transport->xprt.resvport)
1323                 port = 0;
1324         sa = (struct sockaddr_in6 *)&transport->addr;
1325         myaddr.sin6_addr = sa->sin6_addr;
1326         do {
1327                 myaddr.sin6_port = htons(port);
1328                 err = kernel_bind(sock, (struct sockaddr *) &myaddr,
1329                                                 sizeof(myaddr));
1330                 if (!transport->xprt.resvport)
1331                         break;
1332                 if (err == 0) {
1333                         transport->port = port;
1334                         break;
1335                 }
1336                 if (port <= xprt_min_resvport)
1337                         port = xprt_max_resvport;
1338                 else
1339                         port--;
1340         } while (err == -EADDRINUSE && port != transport->port);
1341         dprintk("RPC:       xs_bind6 "NIP6_FMT":%u: %s (%d)\n",
1342                 NIP6(myaddr.sin6_addr), port, err ? "failed" : "ok", err);
1343         return err;
1344 }
1345
1346 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1347 static struct lock_class_key xs_key[2];
1348 static struct lock_class_key xs_slock_key[2];
1349
1350 static inline void xs_reclassify_socket4(struct socket *sock)
1351 {
1352         struct sock *sk = sock->sk;
1353
1354         BUG_ON(sock_owned_by_user(sk));
1355         sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1356                 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1357 }
1358
1359 static inline void xs_reclassify_socket6(struct socket *sock)
1360 {
1361         struct sock *sk = sock->sk;
1362
1363         BUG_ON(sock_owned_by_user(sk));
1364         sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1365                 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
1366 }
1367 #else
1368 static inline void xs_reclassify_socket4(struct socket *sock)
1369 {
1370 }
1371
1372 static inline void xs_reclassify_socket6(struct socket *sock)
1373 {
1374 }
1375 #endif
1376
1377 static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
1378 {
1379         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1380
1381         if (!transport->inet) {
1382                 struct sock *sk = sock->sk;
1383
1384                 write_lock_bh(&sk->sk_callback_lock);
1385
1386                 sk->sk_user_data = xprt;
1387                 transport->old_data_ready = sk->sk_data_ready;
1388                 transport->old_state_change = sk->sk_state_change;
1389                 transport->old_write_space = sk->sk_write_space;
1390                 sk->sk_data_ready = xs_udp_data_ready;
1391                 sk->sk_write_space = xs_udp_write_space;
1392                 sk->sk_no_check = UDP_CSUM_NORCV;
1393                 sk->sk_allocation = GFP_ATOMIC;
1394
1395                 xprt_set_connected(xprt);
1396
1397                 /* Reset to new socket */
1398                 transport->sock = sock;
1399                 transport->inet = sk;
1400
1401                 write_unlock_bh(&sk->sk_callback_lock);
1402         }
1403         xs_udp_do_set_buffer_size(xprt);
1404 }
1405
1406 /**
1407  * xs_udp_connect_worker4 - set up a UDP socket
1408  * @work: RPC transport to connect
1409  *
1410  * Invoked by a work queue tasklet.
1411  */
1412 static void xs_udp_connect_worker4(struct work_struct *work)
1413 {
1414         struct sock_xprt *transport =
1415                 container_of(work, struct sock_xprt, connect_worker.work);
1416         struct rpc_xprt *xprt = &transport->xprt;
1417         struct socket *sock = transport->sock;
1418         int err, status = -EIO;
1419
1420         if (xprt->shutdown || !xprt_bound(xprt))
1421                 goto out;
1422
1423         /* Start by resetting any existing state */
1424         xs_close(xprt);
1425
1426         if ((err = sock_create_kern(PF_INET, SOCK_DGRAM, IPPROTO_UDP, &sock)) < 0) {
1427                 dprintk("RPC:       can't create UDP transport socket (%d).\n", -err);
1428                 goto out;
1429         }
1430         xs_reclassify_socket4(sock);
1431
1432         if (xs_bind4(transport, sock)) {
1433                 sock_release(sock);
1434                 goto out;
1435         }
1436
1437         dprintk("RPC:       worker connecting xprt %p to address: %s\n",
1438                         xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1439
1440         xs_udp_finish_connecting(xprt, sock);
1441         status = 0;
1442 out:
1443         xprt_wake_pending_tasks(xprt, status);
1444         xprt_clear_connecting(xprt);
1445 }
1446
1447 /**
1448  * xs_udp_connect_worker6 - set up a UDP socket
1449  * @work: RPC transport to connect
1450  *
1451  * Invoked by a work queue tasklet.
1452  */
1453 static void xs_udp_connect_worker6(struct work_struct *work)
1454 {
1455         struct sock_xprt *transport =
1456                 container_of(work, struct sock_xprt, connect_worker.work);
1457         struct rpc_xprt *xprt = &transport->xprt;
1458         struct socket *sock = transport->sock;
1459         int err, status = -EIO;
1460
1461         if (xprt->shutdown || !xprt_bound(xprt))
1462                 goto out;
1463
1464         /* Start by resetting any existing state */
1465         xs_close(xprt);
1466
1467         if ((err = sock_create_kern(PF_INET6, SOCK_DGRAM, IPPROTO_UDP, &sock)) < 0) {
1468                 dprintk("RPC:       can't create UDP transport socket (%d).\n", -err);
1469                 goto out;
1470         }
1471         xs_reclassify_socket6(sock);
1472
1473         if (xs_bind6(transport, sock) < 0) {
1474                 sock_release(sock);
1475                 goto out;
1476         }
1477
1478         dprintk("RPC:       worker connecting xprt %p to address: %s\n",
1479                         xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1480
1481         xs_udp_finish_connecting(xprt, sock);
1482         status = 0;
1483 out:
1484         xprt_wake_pending_tasks(xprt, status);
1485         xprt_clear_connecting(xprt);
1486 }
1487
1488 /*
1489  * We need to preserve the port number so the reply cache on the server can
1490  * find our cached RPC replies when we get around to reconnecting.
1491  */
1492 static void xs_tcp_reuse_connection(struct rpc_xprt *xprt)
1493 {
1494         int result;
1495         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1496         struct sockaddr any;
1497
1498         dprintk("RPC:       disconnecting xprt %p to reuse port\n", xprt);
1499
1500         /*
1501          * Disconnect the transport socket by doing a connect operation
1502          * with AF_UNSPEC.  This should return immediately...
1503          */
1504         memset(&any, 0, sizeof(any));
1505         any.sa_family = AF_UNSPEC;
1506         result = kernel_connect(transport->sock, &any, sizeof(any), 0);
1507         if (result)
1508                 dprintk("RPC:       AF_UNSPEC connect return code %d\n",
1509                                 result);
1510 }
1511
1512 static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
1513 {
1514         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1515
1516         if (!transport->inet) {
1517                 struct sock *sk = sock->sk;
1518
1519                 write_lock_bh(&sk->sk_callback_lock);
1520
1521                 sk->sk_user_data = xprt;
1522                 transport->old_data_ready = sk->sk_data_ready;
1523                 transport->old_state_change = sk->sk_state_change;
1524                 transport->old_write_space = sk->sk_write_space;
1525                 sk->sk_data_ready = xs_tcp_data_ready;
1526                 sk->sk_state_change = xs_tcp_state_change;
1527                 sk->sk_write_space = xs_tcp_write_space;
1528                 sk->sk_allocation = GFP_ATOMIC;
1529
1530                 /* socket options */
1531                 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
1532                 sock_reset_flag(sk, SOCK_LINGER);
1533                 tcp_sk(sk)->linger2 = 0;
1534                 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
1535
1536                 xprt_clear_connected(xprt);
1537
1538                 /* Reset to new socket */
1539                 transport->sock = sock;
1540                 transport->inet = sk;
1541
1542                 write_unlock_bh(&sk->sk_callback_lock);
1543         }
1544
1545         /* Tell the socket layer to start connecting... */
1546         xprt->stat.connect_count++;
1547         xprt->stat.connect_start = jiffies;
1548         return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
1549 }
1550
1551 /**
1552  * xs_tcp_connect_worker4 - connect a TCP socket to a remote endpoint
1553  * @work: RPC transport to connect
1554  *
1555  * Invoked by a work queue tasklet.
1556  */
1557 static void xs_tcp_connect_worker4(struct work_struct *work)
1558 {
1559         struct sock_xprt *transport =
1560                 container_of(work, struct sock_xprt, connect_worker.work);
1561         struct rpc_xprt *xprt = &transport->xprt;
1562         struct socket *sock = transport->sock;
1563         int err, status = -EIO;
1564
1565         if (xprt->shutdown || !xprt_bound(xprt))
1566                 goto out;
1567
1568         if (!sock) {
1569                 /* start from scratch */
1570                 if ((err = sock_create_kern(PF_INET, SOCK_STREAM, IPPROTO_TCP, &sock)) < 0) {
1571                         dprintk("RPC:       can't create TCP transport socket (%d).\n", -err);
1572                         goto out;
1573                 }
1574                 xs_reclassify_socket4(sock);
1575
1576                 if (xs_bind4(transport, sock) < 0) {
1577                         sock_release(sock);
1578                         goto out;
1579                 }
1580         } else
1581                 /* "close" the socket, preserving the local port */
1582                 xs_tcp_reuse_connection(xprt);
1583
1584         dprintk("RPC:       worker connecting xprt %p to address: %s\n",
1585                         xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1586
1587         status = xs_tcp_finish_connecting(xprt, sock);
1588         dprintk("RPC:       %p connect status %d connected %d sock state %d\n",
1589                         xprt, -status, xprt_connected(xprt),
1590                         sock->sk->sk_state);
1591         if (status < 0) {
1592                 switch (status) {
1593                         case -EINPROGRESS:
1594                         case -EALREADY:
1595                                 goto out_clear;
1596                         case -ECONNREFUSED:
1597                         case -ECONNRESET:
1598                                 /* retry with existing socket, after a delay */
1599                                 break;
1600                         default:
1601                                 /* get rid of existing socket, and retry */
1602                                 xs_close(xprt);
1603                                 break;
1604                 }
1605         }
1606 out:
1607         xprt_wake_pending_tasks(xprt, status);
1608 out_clear:
1609         xprt_clear_connecting(xprt);
1610 }
1611
1612 /**
1613  * xs_tcp_connect_worker6 - connect a TCP socket to a remote endpoint
1614  * @work: RPC transport to connect
1615  *
1616  * Invoked by a work queue tasklet.
1617  */
1618 static void xs_tcp_connect_worker6(struct work_struct *work)
1619 {
1620         struct sock_xprt *transport =
1621                 container_of(work, struct sock_xprt, connect_worker.work);
1622         struct rpc_xprt *xprt = &transport->xprt;
1623         struct socket *sock = transport->sock;
1624         int err, status = -EIO;
1625
1626         if (xprt->shutdown || !xprt_bound(xprt))
1627                 goto out;
1628
1629         if (!sock) {
1630                 /* start from scratch */
1631                 if ((err = sock_create_kern(PF_INET6, SOCK_STREAM, IPPROTO_TCP, &sock)) < 0) {
1632                         dprintk("RPC:       can't create TCP transport socket (%d).\n", -err);
1633                         goto out;
1634                 }
1635                 xs_reclassify_socket6(sock);
1636
1637                 if (xs_bind6(transport, sock) < 0) {
1638                         sock_release(sock);
1639                         goto out;
1640                 }
1641         } else
1642                 /* "close" the socket, preserving the local port */
1643                 xs_tcp_reuse_connection(xprt);
1644
1645         dprintk("RPC:       worker connecting xprt %p to address: %s\n",
1646                         xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1647
1648         status = xs_tcp_finish_connecting(xprt, sock);
1649         dprintk("RPC:       %p connect status %d connected %d sock state %d\n",
1650                         xprt, -status, xprt_connected(xprt), sock->sk->sk_state);
1651         if (status < 0) {
1652                 switch (status) {
1653                         case -EINPROGRESS:
1654                         case -EALREADY:
1655                                 goto out_clear;
1656                         case -ECONNREFUSED:
1657                         case -ECONNRESET:
1658                                 /* retry with existing socket, after a delay */
1659                                 break;
1660                         default:
1661                                 /* get rid of existing socket, and retry */
1662                                 xs_close(xprt);
1663                                 break;
1664                 }
1665         }
1666 out:
1667         xprt_wake_pending_tasks(xprt, status);
1668 out_clear:
1669         xprt_clear_connecting(xprt);
1670 }
1671
1672 /**
1673  * xs_connect - connect a socket to a remote endpoint
1674  * @task: address of RPC task that manages state of connect request
1675  *
1676  * TCP: If the remote end dropped the connection, delay reconnecting.
1677  *
1678  * UDP socket connects are synchronous, but we use a work queue anyway
1679  * to guarantee that even unprivileged user processes can set up a
1680  * socket on a privileged port.
1681  *
1682  * If a UDP socket connect fails, the delay behavior here prevents
1683  * retry floods (hard mounts).
1684  */
1685 static void xs_connect(struct rpc_task *task)
1686 {
1687         struct rpc_xprt *xprt = task->tk_xprt;
1688         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1689
1690         if (xprt_test_and_set_connecting(xprt))
1691                 return;
1692
1693         if (transport->sock != NULL) {
1694                 dprintk("RPC:       xs_connect delayed xprt %p for %lu "
1695                                 "seconds\n",
1696                                 xprt, xprt->reestablish_timeout / HZ);
1697                 queue_delayed_work(rpciod_workqueue,
1698                                    &transport->connect_worker,
1699                                    xprt->reestablish_timeout);
1700                 xprt->reestablish_timeout <<= 1;
1701                 if (xprt->reestablish_timeout > XS_TCP_MAX_REEST_TO)
1702                         xprt->reestablish_timeout = XS_TCP_MAX_REEST_TO;
1703         } else {
1704                 dprintk("RPC:       xs_connect scheduled xprt %p\n", xprt);
1705                 queue_delayed_work(rpciod_workqueue,
1706                                    &transport->connect_worker, 0);
1707         }
1708 }
1709
1710 /**
1711  * xs_udp_print_stats - display UDP socket-specifc stats
1712  * @xprt: rpc_xprt struct containing statistics
1713  * @seq: output file
1714  *
1715  */
1716 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
1717 {
1718         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1719
1720         seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %Lu %Lu\n",
1721                         transport->port,
1722                         xprt->stat.bind_count,
1723                         xprt->stat.sends,
1724                         xprt->stat.recvs,
1725                         xprt->stat.bad_xids,
1726                         xprt->stat.req_u,
1727                         xprt->stat.bklog_u);
1728 }
1729
1730 /**
1731  * xs_tcp_print_stats - display TCP socket-specifc stats
1732  * @xprt: rpc_xprt struct containing statistics
1733  * @seq: output file
1734  *
1735  */
1736 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
1737 {
1738         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1739         long idle_time = 0;
1740
1741         if (xprt_connected(xprt))
1742                 idle_time = (long)(jiffies - xprt->last_used) / HZ;
1743
1744         seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu %Lu %Lu\n",
1745                         transport->port,
1746                         xprt->stat.bind_count,
1747                         xprt->stat.connect_count,
1748                         xprt->stat.connect_time,
1749                         idle_time,
1750                         xprt->stat.sends,
1751                         xprt->stat.recvs,
1752                         xprt->stat.bad_xids,
1753                         xprt->stat.req_u,
1754                         xprt->stat.bklog_u);
1755 }
1756
1757 static struct rpc_xprt_ops xs_udp_ops = {
1758         .set_buffer_size        = xs_udp_set_buffer_size,
1759         .reserve_xprt           = xprt_reserve_xprt_cong,
1760         .release_xprt           = xprt_release_xprt_cong,
1761         .rpcbind                = rpcb_getport_async,
1762         .set_port               = xs_set_port,
1763         .connect                = xs_connect,
1764         .buf_alloc              = rpc_malloc,
1765         .buf_free               = rpc_free,
1766         .send_request           = xs_udp_send_request,
1767         .set_retrans_timeout    = xprt_set_retrans_timeout_rtt,
1768         .timer                  = xs_udp_timer,
1769         .release_request        = xprt_release_rqst_cong,
1770         .close                  = xs_close,
1771         .destroy                = xs_destroy,
1772         .print_stats            = xs_udp_print_stats,
1773 };
1774
1775 static struct rpc_xprt_ops xs_tcp_ops = {
1776         .reserve_xprt           = xprt_reserve_xprt,
1777         .release_xprt           = xs_tcp_release_xprt,
1778         .rpcbind                = rpcb_getport_async,
1779         .set_port               = xs_set_port,
1780         .connect                = xs_connect,
1781         .buf_alloc              = rpc_malloc,
1782         .buf_free               = rpc_free,
1783         .send_request           = xs_tcp_send_request,
1784         .set_retrans_timeout    = xprt_set_retrans_timeout_def,
1785         .close                  = xs_close,
1786         .destroy                = xs_destroy,
1787         .print_stats            = xs_tcp_print_stats,
1788 };
1789
1790 static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
1791                                       unsigned int slot_table_size)
1792 {
1793         struct rpc_xprt *xprt;
1794         struct sock_xprt *new;
1795
1796         if (args->addrlen > sizeof(xprt->addr)) {
1797                 dprintk("RPC:       xs_setup_xprt: address too large\n");
1798                 return ERR_PTR(-EBADF);
1799         }
1800
1801         new = kzalloc(sizeof(*new), GFP_KERNEL);
1802         if (new == NULL) {
1803                 dprintk("RPC:       xs_setup_xprt: couldn't allocate "
1804                                 "rpc_xprt\n");
1805                 return ERR_PTR(-ENOMEM);
1806         }
1807         xprt = &new->xprt;
1808
1809         xprt->max_reqs = slot_table_size;
1810         xprt->slot = kcalloc(xprt->max_reqs, sizeof(struct rpc_rqst), GFP_KERNEL);
1811         if (xprt->slot == NULL) {
1812                 kfree(xprt);
1813                 dprintk("RPC:       xs_setup_xprt: couldn't allocate slot "
1814                                 "table\n");
1815                 return ERR_PTR(-ENOMEM);
1816         }
1817
1818         memcpy(&xprt->addr, args->dstaddr, args->addrlen);
1819         xprt->addrlen = args->addrlen;
1820         if (args->srcaddr)
1821                 memcpy(&new->addr, args->srcaddr, args->addrlen);
1822         new->port = xs_get_random_port();
1823
1824         return xprt;
1825 }
1826
1827 /**
1828  * xs_setup_udp - Set up transport to use a UDP socket
1829  * @args: rpc transport creation arguments
1830  *
1831  */
1832 static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
1833 {
1834         struct sockaddr *addr = args->dstaddr;
1835         struct rpc_xprt *xprt;
1836         struct sock_xprt *transport;
1837
1838         xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries);
1839         if (IS_ERR(xprt))
1840                 return xprt;
1841         transport = container_of(xprt, struct sock_xprt, xprt);
1842
1843         xprt->prot = IPPROTO_UDP;
1844         xprt->tsh_size = 0;
1845         /* XXX: header size can vary due to auth type, IPv6, etc. */
1846         xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
1847
1848         xprt->bind_timeout = XS_BIND_TO;
1849         xprt->connect_timeout = XS_UDP_CONN_TO;
1850         xprt->reestablish_timeout = XS_UDP_REEST_TO;
1851         xprt->idle_timeout = XS_IDLE_DISC_TO;
1852
1853         xprt->ops = &xs_udp_ops;
1854
1855         if (args->timeout)
1856                 xprt->timeout = *args->timeout;
1857         else
1858                 xprt_set_timeout(&xprt->timeout, 5, 5 * HZ);
1859
1860         switch (addr->sa_family) {
1861         case AF_INET:
1862                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
1863                         xprt_set_bound(xprt);
1864
1865                 INIT_DELAYED_WORK(&transport->connect_worker,
1866                                         xs_udp_connect_worker4);
1867                 xs_format_ipv4_peer_addresses(xprt);
1868                 break;
1869         case AF_INET6:
1870                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
1871                         xprt_set_bound(xprt);
1872
1873                 INIT_DELAYED_WORK(&transport->connect_worker,
1874                                         xs_udp_connect_worker6);
1875                 xs_format_ipv6_peer_addresses(xprt);
1876                 break;
1877         default:
1878                 kfree(xprt);
1879                 return ERR_PTR(-EAFNOSUPPORT);
1880         }
1881
1882         dprintk("RPC:       set up transport to address %s\n",
1883                         xprt->address_strings[RPC_DISPLAY_ALL]);
1884
1885         if (try_module_get(THIS_MODULE))
1886                 return xprt;
1887
1888         kfree(xprt->slot);
1889         kfree(xprt);
1890         return ERR_PTR(-EINVAL);
1891 }
1892
1893 /**
1894  * xs_setup_tcp - Set up transport to use a TCP socket
1895  * @args: rpc transport creation arguments
1896  *
1897  */
1898 static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
1899 {
1900         struct sockaddr *addr = args->dstaddr;
1901         struct rpc_xprt *xprt;
1902         struct sock_xprt *transport;
1903
1904         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries);
1905         if (IS_ERR(xprt))
1906                 return xprt;
1907         transport = container_of(xprt, struct sock_xprt, xprt);
1908
1909         xprt->prot = IPPROTO_TCP;
1910         xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
1911         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
1912
1913         xprt->bind_timeout = XS_BIND_TO;
1914         xprt->connect_timeout = XS_TCP_CONN_TO;
1915         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1916         xprt->idle_timeout = XS_IDLE_DISC_TO;
1917
1918         xprt->ops = &xs_tcp_ops;
1919
1920         if (args->timeout)
1921                 xprt->timeout = *args->timeout;
1922         else
1923                 xprt_set_timeout(&xprt->timeout, 2, 60 * HZ);
1924
1925         switch (addr->sa_family) {
1926         case AF_INET:
1927                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
1928                         xprt_set_bound(xprt);
1929
1930                 INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_connect_worker4);
1931                 xs_format_ipv4_peer_addresses(xprt);
1932                 break;
1933         case AF_INET6:
1934                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
1935                         xprt_set_bound(xprt);
1936
1937                 INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_connect_worker6);
1938                 xs_format_ipv6_peer_addresses(xprt);
1939                 break;
1940         default:
1941                 kfree(xprt);
1942                 return ERR_PTR(-EAFNOSUPPORT);
1943         }
1944
1945         dprintk("RPC:       set up transport to address %s\n",
1946                         xprt->address_strings[RPC_DISPLAY_ALL]);
1947
1948         if (try_module_get(THIS_MODULE))
1949                 return xprt;
1950
1951         kfree(xprt->slot);
1952         kfree(xprt);
1953         return ERR_PTR(-EINVAL);
1954 }
1955
1956 static struct xprt_class        xs_udp_transport = {
1957         .list           = LIST_HEAD_INIT(xs_udp_transport.list),
1958         .name           = "udp",
1959         .owner          = THIS_MODULE,
1960         .ident          = IPPROTO_UDP,
1961         .setup          = xs_setup_udp,
1962 };
1963
1964 static struct xprt_class        xs_tcp_transport = {
1965         .list           = LIST_HEAD_INIT(xs_tcp_transport.list),
1966         .name           = "tcp",
1967         .owner          = THIS_MODULE,
1968         .ident          = IPPROTO_TCP,
1969         .setup          = xs_setup_tcp,
1970 };
1971
1972 /**
1973  * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
1974  *
1975  */
1976 int init_socket_xprt(void)
1977 {
1978 #ifdef RPC_DEBUG
1979         if (!sunrpc_table_header)
1980                 sunrpc_table_header = register_sysctl_table(sunrpc_table);
1981 #endif
1982
1983         xprt_register_transport(&xs_udp_transport);
1984         xprt_register_transport(&xs_tcp_transport);
1985
1986         return 0;
1987 }
1988
1989 /**
1990  * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
1991  *
1992  */
1993 void cleanup_socket_xprt(void)
1994 {
1995 #ifdef RPC_DEBUG
1996         if (sunrpc_table_header) {
1997                 unregister_sysctl_table(sunrpc_table_header);
1998                 sunrpc_table_header = NULL;
1999         }
2000 #endif
2001
2002         xprt_unregister_transport(&xs_udp_transport);
2003         xprt_unregister_transport(&xs_tcp_transport);
2004 }