err.no Git - 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                                           const char *protocol,
 285                                           const char *netid)
 286 {
 287         struct sockaddr_in *addr = xs_addr_in(xprt);
 288         char *buf;
 289
 290         buf = kzalloc(20, GFP_KERNEL);
 291         if (buf) {
 292                 snprintf(buf, 20, NIPQUAD_FMT,
 293                                 NIPQUAD(addr->sin_addr.s_addr));
 294         }
 295         xprt->address_strings[RPC_DISPLAY_ADDR] = buf;
 296
 297         buf = kzalloc(8, GFP_KERNEL);
 298         if (buf) {
 299                 snprintf(buf, 8, "%u",
 300                                 ntohs(addr->sin_port));
 301         }
 302         xprt->address_strings[RPC_DISPLAY_PORT] = buf;
 303
 304         xprt->address_strings[RPC_DISPLAY_PROTO] = protocol;
 305
 306         buf = kzalloc(48, GFP_KERNEL);
 307         if (buf) {
 308                 snprintf(buf, 48, "addr="NIPQUAD_FMT" port=%u proto=%s",
 309                         NIPQUAD(addr->sin_addr.s_addr),
 310                         ntohs(addr->sin_port),
 311                         protocol);
 312         }
 313         xprt->address_strings[RPC_DISPLAY_ALL] = buf;
 314
 315         buf = kzalloc(10, GFP_KERNEL);
 316         if (buf) {
 317                 snprintf(buf, 10, "%02x%02x%02x%02x",
 318                                 NIPQUAD(addr->sin_addr.s_addr));
 319         }
 320         xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = buf;
 321
 322         buf = kzalloc(8, GFP_KERNEL);
 323         if (buf) {
 324                 snprintf(buf, 8, "%4hx",
 325                                 ntohs(addr->sin_port));
 326         }
 327         xprt->address_strings[RPC_DISPLAY_HEX_PORT] = buf;
 328
 329         buf = kzalloc(30, GFP_KERNEL);
 330         if (buf) {
 331                 snprintf(buf, 30, NIPQUAD_FMT".%u.%u",
 332                                 NIPQUAD(addr->sin_addr.s_addr),
 333                                 ntohs(addr->sin_port) >> 8,
 334                                 ntohs(addr->sin_port) & 0xff);
 335         }
 336         xprt->address_strings[RPC_DISPLAY_UNIVERSAL_ADDR] = buf;
 337
 338         xprt->address_strings[RPC_DISPLAY_NETID] = netid;
 339 }
 340
 341 static void xs_format_ipv6_peer_addresses(struct rpc_xprt *xprt,
 342                                           const char *protocol,
 343                                           const char *netid)
 344 {
 345         struct sockaddr_in6 *addr = xs_addr_in6(xprt);
 346         char *buf;
 347
 348         buf = kzalloc(40, GFP_KERNEL);
 349         if (buf) {
 350                 snprintf(buf, 40, NIP6_FMT,
 351                                 NIP6(addr->sin6_addr));
 352         }
 353         xprt->address_strings[RPC_DISPLAY_ADDR] = buf;
 354
 355         buf = kzalloc(8, GFP_KERNEL);
 356         if (buf) {
 357                 snprintf(buf, 8, "%u",
 358                                 ntohs(addr->sin6_port));
 359         }
 360         xprt->address_strings[RPC_DISPLAY_PORT] = buf;
 361
 362         xprt->address_strings[RPC_DISPLAY_PROTO] = protocol;
 363
 364         buf = kzalloc(64, GFP_KERNEL);
 365         if (buf) {
 366                 snprintf(buf, 64, "addr="NIP6_FMT" port=%u proto=%s",
 367                                 NIP6(addr->sin6_addr),
 368                                 ntohs(addr->sin6_port),
 369                                 protocol);
 370         }
 371         xprt->address_strings[RPC_DISPLAY_ALL] = buf;
 372
 373         buf = kzalloc(36, GFP_KERNEL);
 374         if (buf) {
 375                 snprintf(buf, 36, NIP6_SEQFMT,
 376                                 NIP6(addr->sin6_addr));
 377         }
 378         xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = buf;
 379
 380         buf = kzalloc(8, GFP_KERNEL);
 381         if (buf) {
 382                 snprintf(buf, 8, "%4hx",
 383                                 ntohs(addr->sin6_port));
 384         }
 385         xprt->address_strings[RPC_DISPLAY_HEX_PORT] = buf;
 386
 387         buf = kzalloc(50, GFP_KERNEL);
 388         if (buf) {
 389                 snprintf(buf, 50, NIP6_FMT".%u.%u",
 390                                 NIP6(addr->sin6_addr),
 391                                 ntohs(addr->sin6_port) >> 8,
 392                                 ntohs(addr->sin6_port) & 0xff);
 393         }
 394         xprt->address_strings[RPC_DISPLAY_UNIVERSAL_ADDR] = buf;
 395
 396         xprt->address_strings[RPC_DISPLAY_NETID] = netid;
 397 }
 398
 399 static void xs_free_peer_addresses(struct rpc_xprt *xprt)
 400 {
 401         unsigned int i;
 402
 403         for (i = 0; i < RPC_DISPLAY_MAX; i++)
 404                 switch (i) {
 405                 case RPC_DISPLAY_PROTO:
 406                 case RPC_DISPLAY_NETID:
 407                         continue;
 408                 default:
 409                         kfree(xprt->address_strings[i]);
 410                 }
 411 }
 412
 413 #define XS_SENDMSG_FLAGS        (MSG_DONTWAIT | MSG_NOSIGNAL)
 414
 415 static int xs_send_kvec(struct socket *sock, struct sockaddr *addr, int addrlen, struct kvec *vec, unsigned int base, int more)
 416 {
 417         struct msghdr msg = {
 418                 .msg_name       = addr,
 419                 .msg_namelen    = addrlen,
 420                 .msg_flags      = XS_SENDMSG_FLAGS | (more ? MSG_MORE : 0),
 421         };
 422         struct kvec iov = {
 423                 .iov_base       = vec->iov_base + base,
 424                 .iov_len        = vec->iov_len - base,
 425         };
 426
 427         if (iov.iov_len != 0)
 428                 return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
 429         return kernel_sendmsg(sock, &msg, NULL, 0, 0);
 430 }
 431
 432 static int xs_send_pagedata(struct socket *sock, struct xdr_buf *xdr, unsigned int base, int more)
 433 {
 434         struct page **ppage;
 435         unsigned int remainder;
 436         int err, sent = 0;
 437
 438         remainder = xdr->page_len - base;
 439         base += xdr->page_base;
 440         ppage = xdr->pages + (base >> PAGE_SHIFT);
 441         base &= ~PAGE_MASK;
 442         for(;;) {
 443                 unsigned int len = min_t(unsigned int, PAGE_SIZE - base, remainder);
 444                 int flags = XS_SENDMSG_FLAGS;
 445
 446                 remainder -= len;
 447                 if (remainder != 0 || more)
 448                         flags |= MSG_MORE;
 449                 err = sock->ops->sendpage(sock, *ppage, base, len, flags);
 450                 if (remainder == 0 || err != len)
 451                         break;
 452                 sent += err;
 453                 ppage++;
 454                 base = 0;
 455         }
 456         if (sent == 0)
 457                 return err;
 458         if (err > 0)
 459                 sent += err;
 460         return sent;
 461 }
 462
 463 /**
 464  * xs_sendpages - write pages directly to a socket
 465  * @sock: socket to send on
 466  * @addr: UDP only -- address of destination
 467  * @addrlen: UDP only -- length of destination address
 468  * @xdr: buffer containing this request
 469  * @base: starting position in the buffer
 470  *
 471  */
 472 static int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base)
 473 {
 474         unsigned int remainder = xdr->len - base;
 475         int err, sent = 0;
 476
 477         if (unlikely(!sock))
 478                 return -ENOTCONN;
 479
 480         clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags);
 481         if (base != 0) {
 482                 addr = NULL;
 483                 addrlen = 0;
 484         }
 485
 486         if (base < xdr->head[0].iov_len || addr != NULL) {
 487                 unsigned int len = xdr->head[0].iov_len - base;
 488                 remainder -= len;
 489                 err = xs_send_kvec(sock, addr, addrlen, &xdr->head[0], base, remainder != 0);
 490                 if (remainder == 0 || err != len)
 491                         goto out;
 492                 sent += err;
 493                 base = 0;
 494         } else
 495                 base -= xdr->head[0].iov_len;
 496
 497         if (base < xdr->page_len) {
 498                 unsigned int len = xdr->page_len - base;
 499                 remainder -= len;
 500                 err = xs_send_pagedata(sock, xdr, base, remainder != 0);
 501                 if (remainder == 0 || err != len)
 502                         goto out;
 503                 sent += err;
 504                 base = 0;
 505         } else
 506                 base -= xdr->page_len;
 507
 508         if (base >= xdr->tail[0].iov_len)
 509                 return sent;
 510         err = xs_send_kvec(sock, NULL, 0, &xdr->tail[0], base, 0);
 511 out:
 512         if (sent == 0)
 513                 return err;
 514         if (err > 0)
 515                 sent += err;
 516         return sent;
 517 }
 518
 519 static void xs_nospace_callback(struct rpc_task *task)
 520 {
 521         struct sock_xprt *transport = container_of(task->tk_rqstp->rq_xprt, struct sock_xprt, xprt);
 522
 523         transport->inet->sk_write_pending--;
 524         clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 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         /* Protect against races with write_space */
 543         spin_lock_bh(&xprt->transport_lock);
 544
 545         /* Don't race with disconnect */
 546         if (xprt_connected(xprt)) {
 547                 if (test_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags)) {
 548                         /*
 549                          * Notify TCP that we're limited by the application
 550                          * window size
 551                          */
 552                         set_bit(SOCK_NOSPACE, &transport->sock->flags);
 553                         transport->inet->sk_write_pending++;
 554                         /* ...and wait for more buffer space */
 555                         xprt_wait_for_buffer_space(task, xs_nospace_callback);
 556                 }
 557         } else {
 558                 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 559                 task->tk_status = -ENOTCONN;
 560         }
 561
 562         spin_unlock_bh(&xprt->transport_lock);
 563 }
 564
 565 /**
 566  * xs_udp_send_request - write an RPC request to a UDP socket
 567  * @task: address of RPC task that manages the state of an RPC request
 568  *
 569  * Return values:
 570  *        0:    The request has been sent
 571  *   EAGAIN:    The socket was blocked, please call again later to
 572  *              complete the request
 573  * ENOTCONN:    Caller needs to invoke connect logic then call again
 574  *    other:    Some other error occured, the request was not sent
 575  */
 576 static int xs_udp_send_request(struct rpc_task *task)
 577 {
 578         struct rpc_rqst *req = task->tk_rqstp;
 579         struct rpc_xprt *xprt = req->rq_xprt;
 580         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 581         struct xdr_buf *xdr = &req->rq_snd_buf;
 582         int status;
 583
 584         xs_pktdump("packet data:",
 585                                 req->rq_svec->iov_base,
 586                                 req->rq_svec->iov_len);
 587
 588         req->rq_xtime = jiffies;
 589         status = xs_sendpages(transport->sock,
 590                               xs_addr(xprt),
 591                               xprt->addrlen, xdr,
 592                               req->rq_bytes_sent);
 593
 594         dprintk("RPC:       xs_udp_send_request(%u) = %d\n",
 595                         xdr->len - req->rq_bytes_sent, status);
 596
 597         if (status >= 0) {
 598                 task->tk_bytes_sent += status;
 599                 if (status >= req->rq_slen)
 600                         return 0;
 601                 /* Still some bytes left; set up for a retry later. */
 602                 status = -EAGAIN;
 603         }
 604
 605         switch (status) {
 606         case -EAGAIN:
 607                 xs_nospace(task);
 608                 break;
 609         case -ENETUNREACH:
 610         case -EPIPE:
 611         case -ECONNREFUSED:
 612                 /* When the server has died, an ICMP port unreachable message
 613                  * prompts ECONNREFUSED. */
 614                 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 615                 break;
 616         default:
 617                 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 618                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
 619                         -status);
 620         }
 621
 622         return status;
 623 }
 624
 625 /**
 626  * xs_tcp_shutdown - gracefully shut down a TCP socket
 627  * @xprt: transport
 628  *
 629  * Initiates a graceful shutdown of the TCP socket by calling the
 630  * equivalent of shutdown(SHUT_WR);
 631  */
 632 static void xs_tcp_shutdown(struct rpc_xprt *xprt)
 633 {
 634         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 635         struct socket *sock = transport->sock;
 636
 637         if (sock != NULL)
 638                 kernel_sock_shutdown(sock, SHUT_WR);
 639 }
 640
 641 static inline void xs_encode_tcp_record_marker(struct xdr_buf *buf)
 642 {
 643         u32 reclen = buf->len - sizeof(rpc_fraghdr);
 644         rpc_fraghdr *base = buf->head[0].iov_base;
 645         *base = htonl(RPC_LAST_STREAM_FRAGMENT | reclen);
 646 }
 647
 648 /**
 649  * xs_tcp_send_request - write an RPC request to a TCP socket
 650  * @task: address of RPC task that manages the state of an RPC request
 651  *
 652  * Return values:
 653  *        0:    The request has been sent
 654  *   EAGAIN:    The socket was blocked, please call again later to
 655  *              complete the request
 656  * ENOTCONN:    Caller needs to invoke connect logic then call again
 657  *    other:    Some other error occured, the request was not sent
 658  *
 659  * XXX: In the case of soft timeouts, should we eventually give up
 660  *      if sendmsg is not able to make progress?
 661  */
 662 static int xs_tcp_send_request(struct rpc_task *task)
 663 {
 664         struct rpc_rqst *req = task->tk_rqstp;
 665         struct rpc_xprt *xprt = req->rq_xprt;
 666         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 667         struct xdr_buf *xdr = &req->rq_snd_buf;
 668         int status;
 669         unsigned int retry = 0;
 670
 671         xs_encode_tcp_record_marker(&req->rq_snd_buf);
 672
 673         xs_pktdump("packet data:",
 674                                 req->rq_svec->iov_base,
 675                                 req->rq_svec->iov_len);
 676
 677         /* Continue transmitting the packet/record. We must be careful
 678          * to cope with writespace callbacks arriving _after_ we have
 679          * called sendmsg(). */
 680         while (1) {
 681                 req->rq_xtime = jiffies;
 682                 status = xs_sendpages(transport->sock,
 683                                         NULL, 0, xdr, req->rq_bytes_sent);
 684
 685                 dprintk("RPC:       xs_tcp_send_request(%u) = %d\n",
 686                                 xdr->len - req->rq_bytes_sent, status);
 687
 688                 if (unlikely(status < 0))
 689                         break;
 690
 691                 /* If we've sent the entire packet, immediately
 692                  * reset the count of bytes sent. */
 693                 req->rq_bytes_sent += status;
 694                 task->tk_bytes_sent += status;
 695                 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
 696                         req->rq_bytes_sent = 0;
 697                         return 0;
 698                 }
 699
 700                 status = -EAGAIN;
 701                 if (retry++ > XS_SENDMSG_RETRY)
 702                         break;
 703         }
 704
 705         switch (status) {
 706         case -EAGAIN:
 707                 xs_nospace(task);
 708                 break;
 709         case -ECONNREFUSED:
 710         case -ECONNRESET:
 711         case -ENOTCONN:
 712         case -EPIPE:
 713                 status = -ENOTCONN;
 714                 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 715                 break;
 716         default:
 717                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
 718                         -status);
 719                 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 720                 xs_tcp_shutdown(xprt);
 721         }
 722
 723         return status;
 724 }
 725
 726 /**
 727  * xs_tcp_release_xprt - clean up after a tcp transmission
 728  * @xprt: transport
 729  * @task: rpc task
 730  *
 731  * This cleans up if an error causes us to abort the transmission of a request.
 732  * In this case, the socket may need to be reset in order to avoid confusing
 733  * the server.
 734  */
 735 static void xs_tcp_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
 736 {
 737         struct rpc_rqst *req;
 738
 739         if (task != xprt->snd_task)
 740                 return;
 741         if (task == NULL)
 742                 goto out_release;
 743         req = task->tk_rqstp;
 744         if (req->rq_bytes_sent == 0)
 745                 goto out_release;
 746         if (req->rq_bytes_sent == req->rq_snd_buf.len)
 747                 goto out_release;
 748         set_bit(XPRT_CLOSE_WAIT, &task->tk_xprt->state);
 749 out_release:
 750         xprt_release_xprt(xprt, task);
 751 }
 752
 753 /**
 754  * xs_close - close a socket
 755  * @xprt: transport
 756  *
 757  * This is used when all requests are complete; ie, no DRC state remains
 758  * on the server we want to save.
 759  */
 760 static void xs_close(struct rpc_xprt *xprt)
 761 {
 762         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 763         struct socket *sock = transport->sock;
 764         struct sock *sk = transport->inet;
 765
 766         if (!sk)
 767                 goto clear_close_wait;
 768
 769         dprintk("RPC:       xs_close xprt %p\n", xprt);
 770
 771         write_lock_bh(&sk->sk_callback_lock);
 772         transport->inet = NULL;
 773         transport->sock = NULL;
 774
 775         sk->sk_user_data = NULL;
 776         sk->sk_data_ready = transport->old_data_ready;
 777         sk->sk_state_change = transport->old_state_change;
 778         sk->sk_write_space = transport->old_write_space;
 779         write_unlock_bh(&sk->sk_callback_lock);
 780
 781         sk->sk_no_check = 0;
 782
 783         sock_release(sock);
 784 clear_close_wait:
 785         smp_mb__before_clear_bit();
 786         clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
 787         clear_bit(XPRT_CLOSING, &xprt->state);
 788         smp_mb__after_clear_bit();
 789         xprt_disconnect_done(xprt);
 790 }
 791
 792 /**
 793  * xs_destroy - prepare to shutdown a transport
 794  * @xprt: doomed transport
 795  *
 796  */
 797 static void xs_destroy(struct rpc_xprt *xprt)
 798 {
 799         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 800
 801         dprintk("RPC:       xs_destroy xprt %p\n", xprt);
 802
 803         cancel_rearming_delayed_work(&transport->connect_worker);
 804
 805         xs_close(xprt);
 806         xs_free_peer_addresses(xprt);
 807         kfree(xprt->slot);
 808         kfree(xprt);
 809         module_put(THIS_MODULE);
 810 }
 811
 812 static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
 813 {
 814         return (struct rpc_xprt *) sk->sk_user_data;
 815 }
 816
 817 /**
 818  * xs_udp_data_ready - "data ready" callback for UDP sockets
 819  * @sk: socket with data to read
 820  * @len: how much data to read
 821  *
 822  */
 823 static void xs_udp_data_ready(struct sock *sk, int len)
 824 {
 825         struct rpc_task *task;
 826         struct rpc_xprt *xprt;
 827         struct rpc_rqst *rovr;
 828         struct sk_buff *skb;
 829         int err, repsize, copied;
 830         u32 _xid;
 831         __be32 *xp;
 832
 833         read_lock(&sk->sk_callback_lock);
 834         dprintk("RPC:       xs_udp_data_ready...\n");
 835         if (!(xprt = xprt_from_sock(sk)))
 836                 goto out;
 837
 838         if ((skb = skb_recv_datagram(sk, 0, 1, &err)) == NULL)
 839                 goto out;
 840
 841         if (xprt->shutdown)
 842                 goto dropit;
 843
 844         repsize = skb->len - sizeof(struct udphdr);
 845         if (repsize < 4) {
 846                 dprintk("RPC:       impossible RPC reply size %d!\n", repsize);
 847                 goto dropit;
 848         }
 849
 850         /* Copy the XID from the skb... */
 851         xp = skb_header_pointer(skb, sizeof(struct udphdr),
 852                                 sizeof(_xid), &_xid);
 853         if (xp == NULL)
 854                 goto dropit;
 855
 856         /* Look up and lock the request corresponding to the given XID */
 857         spin_lock(&xprt->transport_lock);
 858         rovr = xprt_lookup_rqst(xprt, *xp);
 859         if (!rovr)
 860                 goto out_unlock;
 861         task = rovr->rq_task;
 862
 863         if ((copied = rovr->rq_private_buf.buflen) > repsize)
 864                 copied = repsize;
 865
 866         /* Suck it into the iovec, verify checksum if not done by hw. */
 867         if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
 868                 UDPX_INC_STATS_BH(sk, UDP_MIB_INERRORS);
 869                 goto out_unlock;
 870         }
 871
 872         UDPX_INC_STATS_BH(sk, UDP_MIB_INDATAGRAMS);
 873
 874         /* Something worked... */
 875         dst_confirm(skb->dst);
 876
 877         xprt_adjust_cwnd(task, copied);
 878         xprt_update_rtt(task);
 879         xprt_complete_rqst(task, copied);
 880
 881  out_unlock:
 882         spin_unlock(&xprt->transport_lock);
 883  dropit:
 884         skb_free_datagram(sk, skb);
 885  out:
 886         read_unlock(&sk->sk_callback_lock);
 887 }
 888
 889 static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
 890 {
 891         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 892         size_t len, used;
 893         char *p;
 894
 895         p = ((char *) &transport->tcp_fraghdr) + transport->tcp_offset;
 896         len = sizeof(transport->tcp_fraghdr) - transport->tcp_offset;
 897         used = xdr_skb_read_bits(desc, p, len);
 898         transport->tcp_offset += used;
 899         if (used != len)
 900                 return;
 901
 902         transport->tcp_reclen = ntohl(transport->tcp_fraghdr);
 903         if (transport->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
 904                 transport->tcp_flags |= TCP_RCV_LAST_FRAG;
 905         else
 906                 transport->tcp_flags &= ~TCP_RCV_LAST_FRAG;
 907         transport->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
 908
 909         transport->tcp_flags &= ~TCP_RCV_COPY_FRAGHDR;
 910         transport->tcp_offset = 0;
 911
 912         /* Sanity check of the record length */
 913         if (unlikely(transport->tcp_reclen < 4)) {
 914                 dprintk("RPC:       invalid TCP record fragment length\n");
 915                 xprt_force_disconnect(xprt);
 916                 return;
 917         }
 918         dprintk("RPC:       reading TCP record fragment of length %d\n",
 919                         transport->tcp_reclen);
 920 }
 921
 922 static void xs_tcp_check_fraghdr(struct sock_xprt *transport)
 923 {
 924         if (transport->tcp_offset == transport->tcp_reclen) {
 925                 transport->tcp_flags |= TCP_RCV_COPY_FRAGHDR;
 926                 transport->tcp_offset = 0;
 927                 if (transport->tcp_flags & TCP_RCV_LAST_FRAG) {
 928                         transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
 929                         transport->tcp_flags |= TCP_RCV_COPY_XID;
 930                         transport->tcp_copied = 0;
 931                 }
 932         }
 933 }
 934
 935 static inline void xs_tcp_read_xid(struct sock_xprt *transport, struct xdr_skb_reader *desc)
 936 {
 937         size_t len, used;
 938         char *p;
 939
 940         len = sizeof(transport->tcp_xid) - transport->tcp_offset;
 941         dprintk("RPC:       reading XID (%Zu bytes)\n", len);
 942         p = ((char *) &transport->tcp_xid) + transport->tcp_offset;
 943         used = xdr_skb_read_bits(desc, p, len);
 944         transport->tcp_offset += used;
 945         if (used != len)
 946                 return;
 947         transport->tcp_flags &= ~TCP_RCV_COPY_XID;
 948         transport->tcp_flags |= TCP_RCV_COPY_DATA;
 949         transport->tcp_copied = 4;
 950         dprintk("RPC:       reading reply for XID %08x\n",
 951                         ntohl(transport->tcp_xid));
 952         xs_tcp_check_fraghdr(transport);
 953 }
 954
 955 static inline void xs_tcp_read_request(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
 956 {
 957         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 958         struct rpc_rqst *req;
 959         struct xdr_buf *rcvbuf;
 960         size_t len;
 961         ssize_t r;
 962
 963         /* Find and lock the request corresponding to this xid */
 964         spin_lock(&xprt->transport_lock);
 965         req = xprt_lookup_rqst(xprt, transport->tcp_xid);
 966         if (!req) {
 967                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
 968                 dprintk("RPC:       XID %08x request not found!\n",
 969                                 ntohl(transport->tcp_xid));
 970                 spin_unlock(&xprt->transport_lock);
 971                 return;
 972         }
 973
 974         rcvbuf = &req->rq_private_buf;
 975         len = desc->count;
 976         if (len > transport->tcp_reclen - transport->tcp_offset) {
 977                 struct xdr_skb_reader my_desc;
 978
 979                 len = transport->tcp_reclen - transport->tcp_offset;
 980                 memcpy(&my_desc, desc, sizeof(my_desc));
 981                 my_desc.count = len;
 982                 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
 983                                           &my_desc, xdr_skb_read_bits);
 984                 desc->count -= r;
 985                 desc->offset += r;
 986         } else
 987                 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
 988                                           desc, xdr_skb_read_bits);
 989
 990         if (r > 0) {
 991                 transport->tcp_copied += r;
 992                 transport->tcp_offset += r;
 993         }
 994         if (r != len) {
 995                 /* Error when copying to the receive buffer,
 996                  * usually because we weren't able to allocate
 997                  * additional buffer pages. All we can do now
 998                  * is turn off TCP_RCV_COPY_DATA, so the request
 999                  * will not receive any additional updates,
1000                  * and time out.
1001                  * Any remaining data from this record will
1002                  * be discarded.
1003                  */
1004                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1005                 dprintk("RPC:       XID %08x truncated request\n",
1006                                 ntohl(transport->tcp_xid));
1007                 dprintk("RPC:       xprt = %p, tcp_copied = %lu, "
1008                                 "tcp_offset = %u, tcp_reclen = %u\n",
1009                                 xprt, transport->tcp_copied,
1010                                 transport->tcp_offset, transport->tcp_reclen);
1011                 goto out;
1012         }
1013
1014         dprintk("RPC:       XID %08x read %Zd bytes\n",
1015                         ntohl(transport->tcp_xid), r);
1016         dprintk("RPC:       xprt = %p, tcp_copied = %lu, tcp_offset = %u, "
1017                         "tcp_reclen = %u\n", xprt, transport->tcp_copied,
1018                         transport->tcp_offset, transport->tcp_reclen);
1019
1020         if (transport->tcp_copied == req->rq_private_buf.buflen)
1021                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1022         else if (transport->tcp_offset == transport->tcp_reclen) {
1023                 if (transport->tcp_flags & TCP_RCV_LAST_FRAG)
1024                         transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1025         }
1026
1027 out:
1028         if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1029                 xprt_complete_rqst(req->rq_task, transport->tcp_copied);
1030         spin_unlock(&xprt->transport_lock);
1031         xs_tcp_check_fraghdr(transport);
1032 }
1033
1034 static inline void xs_tcp_read_discard(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1035 {
1036         size_t len;
1037
1038         len = transport->tcp_reclen - transport->tcp_offset;
1039         if (len > desc->count)
1040                 len = desc->count;
1041         desc->count -= len;
1042         desc->offset += len;
1043         transport->tcp_offset += len;
1044         dprintk("RPC:       discarded %Zu bytes\n", len);
1045         xs_tcp_check_fraghdr(transport);
1046 }
1047
1048 static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
1049 {
1050         struct rpc_xprt *xprt = rd_desc->arg.data;
1051         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1052         struct xdr_skb_reader desc = {
1053                 .skb    = skb,
1054                 .offset = offset,
1055                 .count  = len,
1056         };
1057
1058         dprintk("RPC:       xs_tcp_data_recv started\n");
1059         do {
1060                 /* Read in a new fragment marker if necessary */
1061                 /* Can we ever really expect to get completely empty fragments? */
1062                 if (transport->tcp_flags & TCP_RCV_COPY_FRAGHDR) {
1063                         xs_tcp_read_fraghdr(xprt, &desc);
1064                         continue;
1065                 }
1066                 /* Read in the xid if necessary */
1067                 if (transport->tcp_flags & TCP_RCV_COPY_XID) {
1068                         xs_tcp_read_xid(transport, &desc);
1069                         continue;
1070                 }
1071                 /* Read in the request data */
1072                 if (transport->tcp_flags & TCP_RCV_COPY_DATA) {
1073                         xs_tcp_read_request(xprt, &desc);
1074                         continue;
1075                 }
1076                 /* Skip over any trailing bytes on short reads */
1077                 xs_tcp_read_discard(transport, &desc);
1078         } while (desc.count);
1079         dprintk("RPC:       xs_tcp_data_recv done\n");
1080         return len - desc.count;
1081 }
1082
1083 /**
1084  * xs_tcp_data_ready - "data ready" callback for TCP sockets
1085  * @sk: socket with data to read
1086  * @bytes: how much data to read
1087  *
1088  */
1089 static void xs_tcp_data_ready(struct sock *sk, int bytes)
1090 {
1091         struct rpc_xprt *xprt;
1092         read_descriptor_t rd_desc;
1093         int read;
1094
1095         dprintk("RPC:       xs_tcp_data_ready...\n");
1096
1097         read_lock(&sk->sk_callback_lock);
1098         if (!(xprt = xprt_from_sock(sk)))
1099                 goto out;
1100         if (xprt->shutdown)
1101                 goto out;
1102
1103         /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
1104         rd_desc.arg.data = xprt;
1105         do {
1106                 rd_desc.count = 65536;
1107                 read = tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
1108         } while (read > 0);
1109 out:
1110         read_unlock(&sk->sk_callback_lock);
1111 }
1112
1113 /**
1114  * xs_tcp_state_change - callback to handle TCP socket state changes
1115  * @sk: socket whose state has changed
1116  *
1117  */
1118 static void xs_tcp_state_change(struct sock *sk)
1119 {
1120         struct rpc_xprt *xprt;
1121
1122         read_lock(&sk->sk_callback_lock);
1123         if (!(xprt = xprt_from_sock(sk)))
1124                 goto out;
1125         dprintk("RPC:       xs_tcp_state_change client %p...\n", xprt);
1126         dprintk("RPC:       state %x conn %d dead %d zapped %d\n",
1127                         sk->sk_state, xprt_connected(xprt),
1128                         sock_flag(sk, SOCK_DEAD),
1129                         sock_flag(sk, SOCK_ZAPPED));
1130
1131         switch (sk->sk_state) {
1132         case TCP_ESTABLISHED:
1133                 spin_lock_bh(&xprt->transport_lock);
1134                 if (!xprt_test_and_set_connected(xprt)) {
1135                         struct sock_xprt *transport = container_of(xprt,
1136                                         struct sock_xprt, xprt);
1137
1138                         /* Reset TCP record info */
1139                         transport->tcp_offset = 0;
1140                         transport->tcp_reclen = 0;
1141                         transport->tcp_copied = 0;
1142                         transport->tcp_flags =
1143                                 TCP_RCV_COPY_FRAGHDR | TCP_RCV_COPY_XID;
1144
1145                         xprt_wake_pending_tasks(xprt, 0);
1146                 }
1147                 spin_unlock_bh(&xprt->transport_lock);
1148                 break;
1149         case TCP_FIN_WAIT1:
1150                 /* The client initiated a shutdown of the socket */
1151                 xprt->reestablish_timeout = 0;
1152                 set_bit(XPRT_CLOSING, &xprt->state);
1153                 smp_mb__before_clear_bit();
1154                 clear_bit(XPRT_CONNECTED, &xprt->state);
1155                 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1156                 smp_mb__after_clear_bit();
1157                 break;
1158         case TCP_CLOSE_WAIT:
1159                 /* The server initiated a shutdown of the socket */
1160                 set_bit(XPRT_CLOSING, &xprt->state);
1161                 xprt_force_disconnect(xprt);
1162         case TCP_SYN_SENT:
1163         case TCP_CLOSING:
1164                 /*
1165                  * If the server closed down the connection, make sure that
1166                  * we back off before reconnecting
1167                  */
1168                 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1169                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1170                 break;
1171         case TCP_LAST_ACK:
1172                 smp_mb__before_clear_bit();
1173                 clear_bit(XPRT_CONNECTED, &xprt->state);
1174                 smp_mb__after_clear_bit();
1175                 break;
1176         case TCP_CLOSE:
1177                 smp_mb__before_clear_bit();
1178                 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1179                 clear_bit(XPRT_CLOSING, &xprt->state);
1180                 smp_mb__after_clear_bit();
1181                 /* Mark transport as closed and wake up all pending tasks */
1182                 xprt_disconnect_done(xprt);
1183         }
1184  out:
1185         read_unlock(&sk->sk_callback_lock);
1186 }
1187
1188 /**
1189  * xs_udp_write_space - callback invoked when socket buffer space
1190  *                             becomes available
1191  * @sk: socket whose state has changed
1192  *
1193  * Called when more output buffer space is available for this socket.
1194  * We try not to wake our writers until they can make "significant"
1195  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1196  * with a bunch of small requests.
1197  */
1198 static void xs_udp_write_space(struct sock *sk)
1199 {
1200         read_lock(&sk->sk_callback_lock);
1201
1202         /* from net/core/sock.c:sock_def_write_space */
1203         if (sock_writeable(sk)) {
1204                 struct socket *sock;
1205                 struct rpc_xprt *xprt;
1206
1207                 if (unlikely(!(sock = sk->sk_socket)))
1208                         goto out;
1209                 clear_bit(SOCK_NOSPACE, &sock->flags);
1210
1211                 if (unlikely(!(xprt = xprt_from_sock(sk))))
1212                         goto out;
1213                 if (test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags) == 0)
1214                         goto out;
1215
1216                 xprt_write_space(xprt);
1217         }
1218
1219  out:
1220         read_unlock(&sk->sk_callback_lock);
1221 }
1222
1223 /**
1224  * xs_tcp_write_space - callback invoked when socket buffer space
1225  *                             becomes available
1226  * @sk: socket whose state has changed
1227  *
1228  * Called when more output buffer space is available for this socket.
1229  * We try not to wake our writers until they can make "significant"
1230  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1231  * with a bunch of small requests.
1232  */
1233 static void xs_tcp_write_space(struct sock *sk)
1234 {
1235         read_lock(&sk->sk_callback_lock);
1236
1237         /* from net/core/stream.c:sk_stream_write_space */
1238         if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk)) {
1239                 struct socket *sock;
1240                 struct rpc_xprt *xprt;
1241
1242                 if (unlikely(!(sock = sk->sk_socket)))
1243                         goto out;
1244                 clear_bit(SOCK_NOSPACE, &sock->flags);
1245
1246                 if (unlikely(!(xprt = xprt_from_sock(sk))))
1247                         goto out;
1248                 if (test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags) == 0)
1249                         goto out;
1250
1251                 xprt_write_space(xprt);
1252         }
1253
1254  out:
1255         read_unlock(&sk->sk_callback_lock);
1256 }
1257
1258 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
1259 {
1260         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1261         struct sock *sk = transport->inet;
1262
1263         if (transport->rcvsize) {
1264                 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1265                 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
1266         }
1267         if (transport->sndsize) {
1268                 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1269                 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1270                 sk->sk_write_space(sk);
1271         }
1272 }
1273
1274 /**
1275  * xs_udp_set_buffer_size - set send and receive limits
1276  * @xprt: generic transport
1277  * @sndsize: requested size of send buffer, in bytes
1278  * @rcvsize: requested size of receive buffer, in bytes
1279  *
1280  * Set socket send and receive buffer size limits.
1281  */
1282 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1283 {
1284         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1285
1286         transport->sndsize = 0;
1287         if (sndsize)
1288                 transport->sndsize = sndsize + 1024;
1289         transport->rcvsize = 0;
1290         if (rcvsize)
1291                 transport->rcvsize = rcvsize + 1024;
1292
1293         xs_udp_do_set_buffer_size(xprt);
1294 }
1295
1296 /**
1297  * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1298  * @task: task that timed out
1299  *
1300  * Adjust the congestion window after a retransmit timeout has occurred.
1301  */
1302 static void xs_udp_timer(struct rpc_task *task)
1303 {
1304         xprt_adjust_cwnd(task, -ETIMEDOUT);
1305 }
1306
1307 static unsigned short xs_get_random_port(void)
1308 {
1309         unsigned short range = xprt_max_resvport - xprt_min_resvport;
1310         unsigned short rand = (unsigned short) net_random() % range;
1311         return rand + xprt_min_resvport;
1312 }
1313
1314 /**
1315  * xs_set_port - reset the port number in the remote endpoint address
1316  * @xprt: generic transport
1317  * @port: new port number
1318  *
1319  */
1320 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1321 {
1322         struct sockaddr *addr = xs_addr(xprt);
1323
1324         dprintk("RPC:       setting port for xprt %p to %u\n", xprt, port);
1325
1326         switch (addr->sa_family) {
1327         case AF_INET:
1328                 ((struct sockaddr_in *)addr)->sin_port = htons(port);
1329                 break;
1330         case AF_INET6:
1331                 ((struct sockaddr_in6 *)addr)->sin6_port = htons(port);
1332                 break;
1333         default:
1334                 BUG();
1335         }
1336 }
1337
1338 static unsigned short xs_get_srcport(struct sock_xprt *transport, struct socket *sock)
1339 {
1340         unsigned short port = transport->port;
1341
1342         if (port == 0 && transport->xprt.resvport)
1343                 port = xs_get_random_port();
1344         return port;
1345 }
1346
1347 static unsigned short xs_next_srcport(struct sock_xprt *transport, struct socket *sock, unsigned short port)
1348 {
1349         if (transport->port != 0)
1350                 transport->port = 0;
1351         if (!transport->xprt.resvport)
1352                 return 0;
1353         if (port <= xprt_min_resvport || port > xprt_max_resvport)
1354                 return xprt_max_resvport;
1355         return --port;
1356 }
1357
1358 static int xs_bind4(struct sock_xprt *transport, struct socket *sock)
1359 {
1360         struct sockaddr_in myaddr = {
1361                 .sin_family = AF_INET,
1362         };
1363         struct sockaddr_in *sa;
1364         int err, nloop = 0;
1365         unsigned short port = xs_get_srcport(transport, sock);
1366         unsigned short last;
1367
1368         sa = (struct sockaddr_in *)&transport->addr;
1369         myaddr.sin_addr = sa->sin_addr;
1370         do {
1371                 myaddr.sin_port = htons(port);
1372                 err = kernel_bind(sock, (struct sockaddr *) &myaddr,
1373                                                 sizeof(myaddr));
1374                 if (port == 0)
1375                         break;
1376                 if (err == 0) {
1377                         transport->port = port;
1378                         break;
1379                 }
1380                 last = port;
1381                 port = xs_next_srcport(transport, sock, port);
1382                 if (port > last)
1383                         nloop++;
1384         } while (err == -EADDRINUSE && nloop != 2);
1385         dprintk("RPC:       %s "NIPQUAD_FMT":%u: %s (%d)\n",
1386                         __FUNCTION__, NIPQUAD(myaddr.sin_addr),
1387                         port, err ? "failed" : "ok", err);
1388         return err;
1389 }
1390
1391 static int xs_bind6(struct sock_xprt *transport, struct socket *sock)
1392 {
1393         struct sockaddr_in6 myaddr = {
1394                 .sin6_family = AF_INET6,
1395         };
1396         struct sockaddr_in6 *sa;
1397         int err, nloop = 0;
1398         unsigned short port = xs_get_srcport(transport, sock);
1399         unsigned short last;
1400
1401         sa = (struct sockaddr_in6 *)&transport->addr;
1402         myaddr.sin6_addr = sa->sin6_addr;
1403         do {
1404                 myaddr.sin6_port = htons(port);
1405                 err = kernel_bind(sock, (struct sockaddr *) &myaddr,
1406                                                 sizeof(myaddr));
1407                 if (port == 0)
1408                         break;
1409                 if (err == 0) {
1410                         transport->port = port;
1411                         break;
1412                 }
1413                 last = port;
1414                 port = xs_next_srcport(transport, sock, port);
1415                 if (port > last)
1416                         nloop++;
1417         } while (err == -EADDRINUSE && nloop != 2);
1418         dprintk("RPC:       xs_bind6 "NIP6_FMT":%u: %s (%d)\n",
1419                 NIP6(myaddr.sin6_addr), port, err ? "failed" : "ok", err);
1420         return err;
1421 }
1422
1423 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1424 static struct lock_class_key xs_key[2];
1425 static struct lock_class_key xs_slock_key[2];
1426
1427 static inline void xs_reclassify_socket4(struct socket *sock)
1428 {
1429         struct sock *sk = sock->sk;
1430
1431         BUG_ON(sock_owned_by_user(sk));
1432         sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1433                 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1434 }
1435
1436 static inline void xs_reclassify_socket6(struct socket *sock)
1437 {
1438         struct sock *sk = sock->sk;
1439
1440         BUG_ON(sock_owned_by_user(sk));
1441         sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1442                 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
1443 }
1444 #else
1445 static inline void xs_reclassify_socket4(struct socket *sock)
1446 {
1447 }
1448
1449 static inline void xs_reclassify_socket6(struct socket *sock)
1450 {
1451 }
1452 #endif
1453
1454 static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
1455 {
1456         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1457
1458         if (!transport->inet) {
1459                 struct sock *sk = sock->sk;
1460
1461                 write_lock_bh(&sk->sk_callback_lock);
1462
1463                 sk->sk_user_data = xprt;
1464                 transport->old_data_ready = sk->sk_data_ready;
1465                 transport->old_state_change = sk->sk_state_change;
1466                 transport->old_write_space = sk->sk_write_space;
1467                 sk->sk_data_ready = xs_udp_data_ready;
1468                 sk->sk_write_space = xs_udp_write_space;
1469                 sk->sk_no_check = UDP_CSUM_NORCV;
1470                 sk->sk_allocation = GFP_ATOMIC;
1471
1472                 xprt_set_connected(xprt);
1473
1474                 /* Reset to new socket */
1475                 transport->sock = sock;
1476                 transport->inet = sk;
1477
1478                 write_unlock_bh(&sk->sk_callback_lock);
1479         }
1480         xs_udp_do_set_buffer_size(xprt);
1481 }
1482
1483 /**
1484  * xs_udp_connect_worker4 - set up a UDP socket
1485  * @work: RPC transport to connect
1486  *
1487  * Invoked by a work queue tasklet.
1488  */
1489 static void xs_udp_connect_worker4(struct work_struct *work)
1490 {
1491         struct sock_xprt *transport =
1492                 container_of(work, struct sock_xprt, connect_worker.work);
1493         struct rpc_xprt *xprt = &transport->xprt;
1494         struct socket *sock = transport->sock;
1495         int err, status = -EIO;
1496
1497         if (xprt->shutdown || !xprt_bound(xprt))
1498                 goto out;
1499
1500         /* Start by resetting any existing state */
1501         xs_close(xprt);
1502
1503         if ((err = sock_create_kern(PF_INET, SOCK_DGRAM, IPPROTO_UDP, &sock)) < 0) {
1504                 dprintk("RPC:       can't create UDP transport socket (%d).\n", -err);
1505                 goto out;
1506         }
1507         xs_reclassify_socket4(sock);
1508
1509         if (xs_bind4(transport, sock)) {
1510                 sock_release(sock);
1511                 goto out;
1512         }
1513
1514         dprintk("RPC:       worker connecting xprt %p to address: %s\n",
1515                         xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1516
1517         xs_udp_finish_connecting(xprt, sock);
1518         status = 0;
1519 out:
1520         xprt_wake_pending_tasks(xprt, status);
1521         xprt_clear_connecting(xprt);
1522 }
1523
1524 /**
1525  * xs_udp_connect_worker6 - set up a UDP socket
1526  * @work: RPC transport to connect
1527  *
1528  * Invoked by a work queue tasklet.
1529  */
1530 static void xs_udp_connect_worker6(struct work_struct *work)
1531 {
1532         struct sock_xprt *transport =
1533                 container_of(work, struct sock_xprt, connect_worker.work);
1534         struct rpc_xprt *xprt = &transport->xprt;
1535         struct socket *sock = transport->sock;
1536         int err, status = -EIO;
1537
1538         if (xprt->shutdown || !xprt_bound(xprt))
1539                 goto out;
1540
1541         /* Start by resetting any existing state */
1542         xs_close(xprt);
1543
1544         if ((err = sock_create_kern(PF_INET6, SOCK_DGRAM, IPPROTO_UDP, &sock)) < 0) {
1545                 dprintk("RPC:       can't create UDP transport socket (%d).\n", -err);
1546                 goto out;
1547         }
1548         xs_reclassify_socket6(sock);
1549
1550         if (xs_bind6(transport, sock) < 0) {
1551                 sock_release(sock);
1552                 goto out;
1553         }
1554
1555         dprintk("RPC:       worker connecting xprt %p to address: %s\n",
1556                         xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1557
1558         xs_udp_finish_connecting(xprt, sock);
1559         status = 0;
1560 out:
1561         xprt_wake_pending_tasks(xprt, status);
1562         xprt_clear_connecting(xprt);
1563 }
1564
1565 /*
1566  * We need to preserve the port number so the reply cache on the server can
1567  * find our cached RPC replies when we get around to reconnecting.
1568  */
1569 static void xs_tcp_reuse_connection(struct rpc_xprt *xprt)
1570 {
1571         int result;
1572         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1573         struct sockaddr any;
1574
1575         dprintk("RPC:       disconnecting xprt %p to reuse port\n", xprt);
1576
1577         /*
1578          * Disconnect the transport socket by doing a connect operation
1579          * with AF_UNSPEC.  This should return immediately...
1580          */
1581         memset(&any, 0, sizeof(any));
1582         any.sa_family = AF_UNSPEC;
1583         result = kernel_connect(transport->sock, &any, sizeof(any), 0);
1584         if (result)
1585                 dprintk("RPC:       AF_UNSPEC connect return code %d\n",
1586                                 result);
1587 }
1588
1589 static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
1590 {
1591         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1592
1593         if (!transport->inet) {
1594                 struct sock *sk = sock->sk;
1595
1596                 write_lock_bh(&sk->sk_callback_lock);
1597
1598                 sk->sk_user_data = xprt;
1599                 transport->old_data_ready = sk->sk_data_ready;
1600                 transport->old_state_change = sk->sk_state_change;
1601                 transport->old_write_space = sk->sk_write_space;
1602                 sk->sk_data_ready = xs_tcp_data_ready;
1603                 sk->sk_state_change = xs_tcp_state_change;
1604                 sk->sk_write_space = xs_tcp_write_space;
1605                 sk->sk_allocation = GFP_ATOMIC;
1606
1607                 /* socket options */
1608                 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
1609                 sock_reset_flag(sk, SOCK_LINGER);
1610                 tcp_sk(sk)->linger2 = 0;
1611                 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
1612
1613                 xprt_clear_connected(xprt);
1614
1615                 /* Reset to new socket */
1616                 transport->sock = sock;
1617                 transport->inet = sk;
1618
1619                 write_unlock_bh(&sk->sk_callback_lock);
1620         }
1621
1622         /* Tell the socket layer to start connecting... */
1623         xprt->stat.connect_count++;
1624         xprt->stat.connect_start = jiffies;
1625         return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
1626 }
1627
1628 /**
1629  * xs_tcp_connect_worker4 - connect a TCP socket to a remote endpoint
1630  * @work: RPC transport to connect
1631  *
1632  * Invoked by a work queue tasklet.
1633  */
1634 static void xs_tcp_connect_worker4(struct work_struct *work)
1635 {
1636         struct sock_xprt *transport =
1637                 container_of(work, struct sock_xprt, connect_worker.work);
1638         struct rpc_xprt *xprt = &transport->xprt;
1639         struct socket *sock = transport->sock;
1640         int err, status = -EIO;
1641
1642         if (xprt->shutdown || !xprt_bound(xprt))
1643                 goto out;
1644
1645         if (!sock) {
1646                 /* start from scratch */
1647                 if ((err = sock_create_kern(PF_INET, SOCK_STREAM, IPPROTO_TCP, &sock)) < 0) {
1648                         dprintk("RPC:       can't create TCP transport socket (%d).\n", -err);
1649                         goto out;
1650                 }
1651                 xs_reclassify_socket4(sock);
1652
1653                 if (xs_bind4(transport, sock) < 0) {
1654                         sock_release(sock);
1655                         goto out;
1656                 }
1657         } else
1658                 /* "close" the socket, preserving the local port */
1659                 xs_tcp_reuse_connection(xprt);
1660
1661         dprintk("RPC:       worker connecting xprt %p to address: %s\n",
1662                         xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1663
1664         status = xs_tcp_finish_connecting(xprt, sock);
1665         dprintk("RPC:       %p connect status %d connected %d sock state %d\n",
1666                         xprt, -status, xprt_connected(xprt),
1667                         sock->sk->sk_state);
1668         if (status < 0) {
1669                 switch (status) {
1670                         case -EINPROGRESS:
1671                         case -EALREADY:
1672                                 goto out_clear;
1673                         case -ECONNREFUSED:
1674                         case -ECONNRESET:
1675                                 /* retry with existing socket, after a delay */
1676                                 break;
1677                         default:
1678                                 /* get rid of existing socket, and retry */
1679                                 xs_tcp_shutdown(xprt);
1680                 }
1681         }
1682 out:
1683         xprt_wake_pending_tasks(xprt, status);
1684 out_clear:
1685         xprt_clear_connecting(xprt);
1686 }
1687
1688 /**
1689  * xs_tcp_connect_worker6 - connect a TCP socket to a remote endpoint
1690  * @work: RPC transport to connect
1691  *
1692  * Invoked by a work queue tasklet.
1693  */
1694 static void xs_tcp_connect_worker6(struct work_struct *work)
1695 {
1696         struct sock_xprt *transport =
1697                 container_of(work, struct sock_xprt, connect_worker.work);
1698         struct rpc_xprt *xprt = &transport->xprt;
1699         struct socket *sock = transport->sock;
1700         int err, status = -EIO;
1701
1702         if (xprt->shutdown || !xprt_bound(xprt))
1703                 goto out;
1704
1705         if (!sock) {
1706                 /* start from scratch */
1707                 if ((err = sock_create_kern(PF_INET6, SOCK_STREAM, IPPROTO_TCP, &sock)) < 0) {
1708                         dprintk("RPC:       can't create TCP transport socket (%d).\n", -err);
1709                         goto out;
1710                 }
1711                 xs_reclassify_socket6(sock);
1712
1713                 if (xs_bind6(transport, sock) < 0) {
1714                         sock_release(sock);
1715                         goto out;
1716                 }
1717         } else
1718                 /* "close" the socket, preserving the local port */
1719                 xs_tcp_reuse_connection(xprt);
1720
1721         dprintk("RPC:       worker connecting xprt %p to address: %s\n",
1722                         xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1723
1724         status = xs_tcp_finish_connecting(xprt, sock);
1725         dprintk("RPC:       %p connect status %d connected %d sock state %d\n",
1726                         xprt, -status, xprt_connected(xprt), sock->sk->sk_state);
1727         if (status < 0) {
1728                 switch (status) {
1729                         case -EINPROGRESS:
1730                         case -EALREADY:
1731                                 goto out_clear;
1732                         case -ECONNREFUSED:
1733                         case -ECONNRESET:
1734                                 /* retry with existing socket, after a delay */
1735                                 break;
1736                         default:
1737                                 /* get rid of existing socket, and retry */
1738                                 xs_tcp_shutdown(xprt);
1739                 }
1740         }
1741 out:
1742         xprt_wake_pending_tasks(xprt, status);
1743 out_clear:
1744         xprt_clear_connecting(xprt);
1745 }
1746
1747 /**
1748  * xs_connect - connect a socket to a remote endpoint
1749  * @task: address of RPC task that manages state of connect request
1750  *
1751  * TCP: If the remote end dropped the connection, delay reconnecting.
1752  *
1753  * UDP socket connects are synchronous, but we use a work queue anyway
1754  * to guarantee that even unprivileged user processes can set up a
1755  * socket on a privileged port.
1756  *
1757  * If a UDP socket connect fails, the delay behavior here prevents
1758  * retry floods (hard mounts).
1759  */
1760 static void xs_connect(struct rpc_task *task)
1761 {
1762         struct rpc_xprt *xprt = task->tk_xprt;
1763         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1764
1765         if (xprt_test_and_set_connecting(xprt))
1766                 return;
1767
1768         if (transport->sock != NULL) {
1769                 dprintk("RPC:       xs_connect delayed xprt %p for %lu "
1770                                 "seconds\n",
1771                                 xprt, xprt->reestablish_timeout / HZ);
1772                 queue_delayed_work(rpciod_workqueue,
1773                                    &transport->connect_worker,
1774                                    xprt->reestablish_timeout);
1775                 xprt->reestablish_timeout <<= 1;
1776                 if (xprt->reestablish_timeout > XS_TCP_MAX_REEST_TO)
1777                         xprt->reestablish_timeout = XS_TCP_MAX_REEST_TO;
1778         } else {
1779                 dprintk("RPC:       xs_connect scheduled xprt %p\n", xprt);
1780                 queue_delayed_work(rpciod_workqueue,
1781                                    &transport->connect_worker, 0);
1782         }
1783 }
1784
1785 static void xs_tcp_connect(struct rpc_task *task)
1786 {
1787         struct rpc_xprt *xprt = task->tk_xprt;
1788
1789         /* Initiate graceful shutdown of the socket if not already done */
1790         if (test_bit(XPRT_CONNECTED, &xprt->state))
1791                 xs_tcp_shutdown(xprt);
1792         /* Exit if we need to wait for socket shutdown to complete */
1793         if (test_bit(XPRT_CLOSING, &xprt->state))
1794                 return;
1795         xs_connect(task);
1796 }
1797
1798 /**
1799  * xs_udp_print_stats - display UDP socket-specifc stats
1800  * @xprt: rpc_xprt struct containing statistics
1801  * @seq: output file
1802  *
1803  */
1804 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
1805 {
1806         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1807
1808         seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %Lu %Lu\n",
1809                         transport->port,
1810                         xprt->stat.bind_count,
1811                         xprt->stat.sends,
1812                         xprt->stat.recvs,
1813                         xprt->stat.bad_xids,
1814                         xprt->stat.req_u,
1815                         xprt->stat.bklog_u);
1816 }
1817
1818 /**
1819  * xs_tcp_print_stats - display TCP socket-specifc stats
1820  * @xprt: rpc_xprt struct containing statistics
1821  * @seq: output file
1822  *
1823  */
1824 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
1825 {
1826         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1827         long idle_time = 0;
1828
1829         if (xprt_connected(xprt))
1830                 idle_time = (long)(jiffies - xprt->last_used) / HZ;
1831
1832         seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu %Lu %Lu\n",
1833                         transport->port,
1834                         xprt->stat.bind_count,
1835                         xprt->stat.connect_count,
1836                         xprt->stat.connect_time,
1837                         idle_time,
1838                         xprt->stat.sends,
1839                         xprt->stat.recvs,
1840                         xprt->stat.bad_xids,
1841                         xprt->stat.req_u,
1842                         xprt->stat.bklog_u);
1843 }
1844
1845 static struct rpc_xprt_ops xs_udp_ops = {
1846         .set_buffer_size        = xs_udp_set_buffer_size,
1847         .reserve_xprt           = xprt_reserve_xprt_cong,
1848         .release_xprt           = xprt_release_xprt_cong,
1849         .rpcbind                = rpcb_getport_async,
1850         .set_port               = xs_set_port,
1851         .connect                = xs_connect,
1852         .buf_alloc              = rpc_malloc,
1853         .buf_free               = rpc_free,
1854         .send_request           = xs_udp_send_request,
1855         .set_retrans_timeout    = xprt_set_retrans_timeout_rtt,
1856         .timer                  = xs_udp_timer,
1857         .release_request        = xprt_release_rqst_cong,
1858         .close                  = xs_close,
1859         .destroy                = xs_destroy,
1860         .print_stats            = xs_udp_print_stats,
1861 };
1862
1863 static struct rpc_xprt_ops xs_tcp_ops = {
1864         .reserve_xprt           = xprt_reserve_xprt,
1865         .release_xprt           = xs_tcp_release_xprt,
1866         .rpcbind                = rpcb_getport_async,
1867         .set_port               = xs_set_port,
1868         .connect                = xs_tcp_connect,
1869         .buf_alloc              = rpc_malloc,
1870         .buf_free               = rpc_free,
1871         .send_request           = xs_tcp_send_request,
1872         .set_retrans_timeout    = xprt_set_retrans_timeout_def,
1873         .close                  = xs_tcp_shutdown,
1874         .destroy                = xs_destroy,
1875         .print_stats            = xs_tcp_print_stats,
1876 };
1877
1878 static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
1879                                       unsigned int slot_table_size)
1880 {
1881         struct rpc_xprt *xprt;
1882         struct sock_xprt *new;
1883
1884         if (args->addrlen > sizeof(xprt->addr)) {
1885                 dprintk("RPC:       xs_setup_xprt: address too large\n");
1886                 return ERR_PTR(-EBADF);
1887         }
1888
1889         new = kzalloc(sizeof(*new), GFP_KERNEL);
1890         if (new == NULL) {
1891                 dprintk("RPC:       xs_setup_xprt: couldn't allocate "
1892                                 "rpc_xprt\n");
1893                 return ERR_PTR(-ENOMEM);
1894         }
1895         xprt = &new->xprt;
1896
1897         xprt->max_reqs = slot_table_size;
1898         xprt->slot = kcalloc(xprt->max_reqs, sizeof(struct rpc_rqst), GFP_KERNEL);
1899         if (xprt->slot == NULL) {
1900                 kfree(xprt);
1901                 dprintk("RPC:       xs_setup_xprt: couldn't allocate slot "
1902                                 "table\n");
1903                 return ERR_PTR(-ENOMEM);
1904         }
1905
1906         memcpy(&xprt->addr, args->dstaddr, args->addrlen);
1907         xprt->addrlen = args->addrlen;
1908         if (args->srcaddr)
1909                 memcpy(&new->addr, args->srcaddr, args->addrlen);
1910
1911         return xprt;
1912 }
1913
1914 static const struct rpc_timeout xs_udp_default_timeout = {
1915         .to_initval = 5 * HZ,
1916         .to_maxval = 30 * HZ,
1917         .to_increment = 5 * HZ,
1918         .to_retries = 5,
1919 };
1920
1921 /**
1922  * xs_setup_udp - Set up transport to use a UDP socket
1923  * @args: rpc transport creation arguments
1924  *
1925  */
1926 static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
1927 {
1928         struct sockaddr *addr = args->dstaddr;
1929         struct rpc_xprt *xprt;
1930         struct sock_xprt *transport;
1931
1932         xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries);
1933         if (IS_ERR(xprt))
1934                 return xprt;
1935         transport = container_of(xprt, struct sock_xprt, xprt);
1936
1937         xprt->prot = IPPROTO_UDP;
1938         xprt->tsh_size = 0;
1939         /* XXX: header size can vary due to auth type, IPv6, etc. */
1940         xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
1941
1942         xprt->bind_timeout = XS_BIND_TO;
1943         xprt->connect_timeout = XS_UDP_CONN_TO;
1944         xprt->reestablish_timeout = XS_UDP_REEST_TO;
1945         xprt->idle_timeout = XS_IDLE_DISC_TO;
1946
1947         xprt->ops = &xs_udp_ops;
1948
1949         xprt->timeout = &xs_udp_default_timeout;
1950
1951         switch (addr->sa_family) {
1952         case AF_INET:
1953                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
1954                         xprt_set_bound(xprt);
1955
1956                 INIT_DELAYED_WORK(&transport->connect_worker,
1957                                         xs_udp_connect_worker4);
1958                 xs_format_ipv4_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
1959                 break;
1960         case AF_INET6:
1961                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
1962                         xprt_set_bound(xprt);
1963
1964                 INIT_DELAYED_WORK(&transport->connect_worker,
1965                                         xs_udp_connect_worker6);
1966                 xs_format_ipv6_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
1967                 break;
1968         default:
1969                 kfree(xprt);
1970                 return ERR_PTR(-EAFNOSUPPORT);
1971         }
1972
1973         dprintk("RPC:       set up transport to address %s\n",
1974                         xprt->address_strings[RPC_DISPLAY_ALL]);
1975
1976         if (try_module_get(THIS_MODULE))
1977                 return xprt;
1978
1979         kfree(xprt->slot);
1980         kfree(xprt);
1981         return ERR_PTR(-EINVAL);
1982 }
1983
1984 static const struct rpc_timeout xs_tcp_default_timeout = {
1985         .to_initval = 60 * HZ,
1986         .to_maxval = 60 * HZ,
1987         .to_retries = 2,
1988 };
1989
1990 /**
1991  * xs_setup_tcp - Set up transport to use a TCP socket
1992  * @args: rpc transport creation arguments
1993  *
1994  */
1995 static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
1996 {
1997         struct sockaddr *addr = args->dstaddr;
1998         struct rpc_xprt *xprt;
1999         struct sock_xprt *transport;
2000
2001         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries);
2002         if (IS_ERR(xprt))
2003                 return xprt;
2004         transport = container_of(xprt, struct sock_xprt, xprt);
2005
2006         xprt->prot = IPPROTO_TCP;
2007         xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2008         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2009
2010         xprt->bind_timeout = XS_BIND_TO;
2011         xprt->connect_timeout = XS_TCP_CONN_TO;
2012         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2013         xprt->idle_timeout = XS_IDLE_DISC_TO;
2014
2015         xprt->ops = &xs_tcp_ops;
2016         xprt->timeout = &xs_tcp_default_timeout;
2017
2018         switch (addr->sa_family) {
2019         case AF_INET:
2020                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2021                         xprt_set_bound(xprt);
2022
2023                 INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_connect_worker4);
2024                 xs_format_ipv4_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
2025                 break;
2026         case AF_INET6:
2027                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2028                         xprt_set_bound(xprt);
2029
2030                 INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_connect_worker6);
2031                 xs_format_ipv6_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
2032                 break;
2033         default:
2034                 kfree(xprt);
2035                 return ERR_PTR(-EAFNOSUPPORT);
2036         }
2037
2038         dprintk("RPC:       set up transport to address %s\n",
2039                         xprt->address_strings[RPC_DISPLAY_ALL]);
2040
2041         if (try_module_get(THIS_MODULE))
2042                 return xprt;
2043
2044         kfree(xprt->slot);
2045         kfree(xprt);
2046         return ERR_PTR(-EINVAL);
2047 }
2048
2049 static struct xprt_class        xs_udp_transport = {
2050         .list           = LIST_HEAD_INIT(xs_udp_transport.list),
2051         .name           = "udp",
2052         .owner          = THIS_MODULE,
2053         .ident          = IPPROTO_UDP,
2054         .setup          = xs_setup_udp,
2055 };
2056
2057 static struct xprt_class        xs_tcp_transport = {
2058         .list           = LIST_HEAD_INIT(xs_tcp_transport.list),
2059         .name           = "tcp",
2060         .owner          = THIS_MODULE,
2061         .ident          = IPPROTO_TCP,
2062         .setup          = xs_setup_tcp,
2063 };
2064
2065 /**
2066  * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
2067  *
2068  */
2069 int init_socket_xprt(void)
2070 {
2071 #ifdef RPC_DEBUG
2072         if (!sunrpc_table_header)
2073                 sunrpc_table_header = register_sysctl_table(sunrpc_table);
2074 #endif
2075
2076         xprt_register_transport(&xs_udp_transport);
2077         xprt_register_transport(&xs_tcp_transport);
2078
2079         return 0;
2080 }
2081
2082 /**
2083  * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
2084  *
2085  */
2086 void cleanup_socket_xprt(void)
2087 {
2088 #ifdef RPC_DEBUG
2089         if (sunrpc_table_header) {
2090                 unregister_sysctl_table(sunrpc_table_header);
2091                 sunrpc_table_header = NULL;
2092         }
2093 #endif
2094
2095         xprt_unregister_transport(&xs_udp_transport);
2096         xprt_unregister_transport(&xs_tcp_transport);
2097 }