git.oblomov.eu 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
   7  * TCP send fixes (C) 1998 Red Hat
   8  * TCP NFS related read + write fixes
   9  *  (C) 1999 Dave Airlie, University of Limerick, Ireland <airlied@linux.ie>
  10  *
  11  * Rewrite of larges part of the code in order to stabilize TCP stuff.
  12  * Fix behaviour when socket buffer is full.
  13  *  (C) 1999 Trond Myklebust <trond.myklebust@fys.uio.no>
  14  *
  15  * IP socket transport implementation, (C) 2005 Chuck Lever <cel@netapp.com>
  16  *
  17  * IPv6 support contributed by Gilles Quillard, Bull Open Source, 2005.
  18  *   <gilles.quillard@bull.net>
  19  */
  20
  21 #include <linux/types.h>
  22 #include <linux/slab.h>
  23 #include <linux/module.h>
  24 #include <linux/capability.h>
  25 #include <linux/pagemap.h>
  26 #include <linux/errno.h>
  27 #include <linux/socket.h>
  28 #include <linux/in.h>
  29 #include <linux/net.h>
  30 #include <linux/mm.h>
  31 #include <linux/udp.h>
  32 #include <linux/tcp.h>
  33 #include <linux/sunrpc/clnt.h>
  34 #include <linux/sunrpc/sched.h>
  35 #include <linux/sunrpc/xprtsock.h>
  36 #include <linux/file.h>
  37
  38 #include <net/sock.h>
  39 #include <net/checksum.h>
  40 #include <net/udp.h>
  41 #include <net/tcp.h>
  42
  43 /*
  44  * xprtsock tunables
  45  */
  46 unsigned int xprt_udp_slot_table_entries = RPC_DEF_SLOT_TABLE;
  47 unsigned int xprt_tcp_slot_table_entries = RPC_DEF_SLOT_TABLE;
  48
  49 unsigned int xprt_min_resvport = RPC_DEF_MIN_RESVPORT;
  50 unsigned int xprt_max_resvport = RPC_DEF_MAX_RESVPORT;
  51
  52 #define XS_TCP_LINGER_TO        (15U * HZ)
  53 static unsigned int xs_tcp_fin_timeout __read_mostly = XS_TCP_LINGER_TO;
  54
  55 /*
  56  * We can register our own files under /proc/sys/sunrpc by
  57  * calling register_sysctl_table() again.  The files in that
  58  * directory become the union of all files registered there.
  59  *
  60  * We simply need to make sure that we don't collide with
  61  * someone else's file names!
  62  */
  63
  64 #ifdef RPC_DEBUG
  65
  66 static unsigned int min_slot_table_size = RPC_MIN_SLOT_TABLE;
  67 static unsigned int max_slot_table_size = RPC_MAX_SLOT_TABLE;
  68 static unsigned int xprt_min_resvport_limit = RPC_MIN_RESVPORT;
  69 static unsigned int xprt_max_resvport_limit = RPC_MAX_RESVPORT;
  70
  71 static struct ctl_table_header *sunrpc_table_header;
  72
  73 /*
  74  * FIXME: changing the UDP slot table size should also resize the UDP
  75  *        socket buffers for existing UDP transports
  76  */
  77 static ctl_table xs_tunables_table[] = {
  78         {
  79                 .ctl_name       = CTL_SLOTTABLE_UDP,
  80                 .procname       = "udp_slot_table_entries",
  81                 .data           = &xprt_udp_slot_table_entries,
  82                 .maxlen         = sizeof(unsigned int),
  83                 .mode           = 0644,
  84                 .proc_handler   = &proc_dointvec_minmax,
  85                 .strategy       = &sysctl_intvec,
  86                 .extra1         = &min_slot_table_size,
  87                 .extra2         = &max_slot_table_size
  88         },
  89         {
  90                 .ctl_name       = CTL_SLOTTABLE_TCP,
  91                 .procname       = "tcp_slot_table_entries",
  92                 .data           = &xprt_tcp_slot_table_entries,
  93                 .maxlen         = sizeof(unsigned int),
  94                 .mode           = 0644,
  95                 .proc_handler   = &proc_dointvec_minmax,
  96                 .strategy       = &sysctl_intvec,
  97                 .extra1         = &min_slot_table_size,
  98                 .extra2         = &max_slot_table_size
  99         },
 100         {
 101                 .ctl_name       = CTL_MIN_RESVPORT,
 102                 .procname       = "min_resvport",
 103                 .data           = &xprt_min_resvport,
 104                 .maxlen         = sizeof(unsigned int),
 105                 .mode           = 0644,
 106                 .proc_handler   = &proc_dointvec_minmax,
 107                 .strategy       = &sysctl_intvec,
 108                 .extra1         = &xprt_min_resvport_limit,
 109                 .extra2         = &xprt_max_resvport_limit
 110         },
 111         {
 112                 .ctl_name       = CTL_MAX_RESVPORT,
 113                 .procname       = "max_resvport",
 114                 .data           = &xprt_max_resvport,
 115                 .maxlen         = sizeof(unsigned int),
 116                 .mode           = 0644,
 117                 .proc_handler   = &proc_dointvec_minmax,
 118                 .strategy       = &sysctl_intvec,
 119                 .extra1         = &xprt_min_resvport_limit,
 120                 .extra2         = &xprt_max_resvport_limit
 121         },
 122         {
 123                 .procname       = "tcp_fin_timeout",
 124                 .data           = &xs_tcp_fin_timeout,
 125                 .maxlen         = sizeof(xs_tcp_fin_timeout),
 126                 .mode           = 0644,
 127                 .proc_handler   = &proc_dointvec_jiffies,
 128                 .strategy       = sysctl_jiffies
 129         },
 130         {
 131                 .ctl_name = 0,
 132         },
 133 };
 134
 135 static ctl_table sunrpc_table[] = {
 136         {
 137                 .ctl_name       = CTL_SUNRPC,
 138                 .procname       = "sunrpc",
 139                 .mode           = 0555,
 140                 .child          = xs_tunables_table
 141         },
 142         {
 143                 .ctl_name = 0,
 144         },
 145 };
 146
 147 #endif
 148
 149 /*
 150  * Time out for an RPC UDP socket connect.  UDP socket connects are
 151  * synchronous, but we set a timeout anyway in case of resource
 152  * exhaustion on the local host.
 153  */
 154 #define XS_UDP_CONN_TO          (5U * HZ)
 155
 156 /*
 157  * Wait duration for an RPC TCP connection to be established.  Solaris
 158  * NFS over TCP uses 60 seconds, for example, which is in line with how
 159  * long a server takes to reboot.
 160  */
 161 #define XS_TCP_CONN_TO          (60U * HZ)
 162
 163 /*
 164  * Wait duration for a reply from the RPC portmapper.
 165  */
 166 #define XS_BIND_TO              (60U * HZ)
 167
 168 /*
 169  * Delay if a UDP socket connect error occurs.  This is most likely some
 170  * kind of resource problem on the local host.
 171  */
 172 #define XS_UDP_REEST_TO         (2U * HZ)
 173
 174 /*
 175  * The reestablish timeout allows clients to delay for a bit before attempting
 176  * to reconnect to a server that just dropped our connection.
 177  *
 178  * We implement an exponential backoff when trying to reestablish a TCP
 179  * transport connection with the server.  Some servers like to drop a TCP
 180  * connection when they are overworked, so we start with a short timeout and
 181  * increase over time if the server is down or not responding.
 182  */
 183 #define XS_TCP_INIT_REEST_TO    (3U * HZ)
 184 #define XS_TCP_MAX_REEST_TO     (5U * 60 * HZ)
 185
 186 /*
 187  * TCP idle timeout; client drops the transport socket if it is idle
 188  * for this long.  Note that we also timeout UDP sockets to prevent
 189  * holding port numbers when there is no RPC traffic.
 190  */
 191 #define XS_IDLE_DISC_TO         (5U * 60 * HZ)
 192
 193 #ifdef RPC_DEBUG
 194 # undef  RPC_DEBUG_DATA
 195 # define RPCDBG_FACILITY        RPCDBG_TRANS
 196 #endif
 197
 198 #ifdef RPC_DEBUG_DATA
 199 static void xs_pktdump(char *msg, u32 *packet, unsigned int count)
 200 {
 201         u8 *buf = (u8 *) packet;
 202         int j;
 203
 204         dprintk("RPC:       %s\n", msg);
 205         for (j = 0; j < count && j < 128; j += 4) {
 206                 if (!(j & 31)) {
 207                         if (j)
 208                                 dprintk("\n");
 209                         dprintk("0x%04x ", j);
 210                 }
 211                 dprintk("%02x%02x%02x%02x ",
 212                         buf[j], buf[j+1], buf[j+2], buf[j+3]);
 213         }
 214         dprintk("\n");
 215 }
 216 #else
 217 static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count)
 218 {
 219         /* NOP */
 220 }
 221 #endif
 222
 223 struct sock_xprt {
 224         struct rpc_xprt         xprt;
 225
 226         /*
 227          * Network layer
 228          */
 229         struct socket *         sock;
 230         struct sock *           inet;
 231
 232         /*
 233          * State of TCP reply receive
 234          */
 235         __be32                  tcp_fraghdr,
 236                                 tcp_xid;
 237
 238         u32                     tcp_offset,
 239                                 tcp_reclen;
 240
 241         unsigned long           tcp_copied,
 242                                 tcp_flags;
 243
 244         /*
 245          * Connection of transports
 246          */
 247         struct delayed_work     connect_worker;
 248         struct sockaddr_storage addr;
 249         unsigned short          port;
 250
 251         /*
 252          * UDP socket buffer size parameters
 253          */
 254         size_t                  rcvsize,
 255                                 sndsize;
 256
 257         /*
 258          * Saved socket callback addresses
 259          */
 260         void                    (*old_data_ready)(struct sock *, int);
 261         void                    (*old_state_change)(struct sock *);
 262         void                    (*old_write_space)(struct sock *);
 263         void                    (*old_error_report)(struct sock *);
 264 };
 265
 266 /*
 267  * TCP receive state flags
 268  */
 269 #define TCP_RCV_LAST_FRAG       (1UL << 0)
 270 #define TCP_RCV_COPY_FRAGHDR    (1UL << 1)
 271 #define TCP_RCV_COPY_XID        (1UL << 2)
 272 #define TCP_RCV_COPY_DATA       (1UL << 3)
 273 #define TCP_RCV_READ_CALLDIR    (1UL << 4)
 274 #define TCP_RCV_COPY_CALLDIR    (1UL << 5)
 275
 276 /*
 277  * TCP RPC flags
 278  */
 279 #define TCP_RPC_REPLY           (1UL << 6)
 280
 281 static inline struct sockaddr *xs_addr(struct rpc_xprt *xprt)
 282 {
 283         return (struct sockaddr *) &xprt->addr;
 284 }
 285
 286 static inline struct sockaddr_in *xs_addr_in(struct rpc_xprt *xprt)
 287 {
 288         return (struct sockaddr_in *) &xprt->addr;
 289 }
 290
 291 static inline struct sockaddr_in6 *xs_addr_in6(struct rpc_xprt *xprt)
 292 {
 293         return (struct sockaddr_in6 *) &xprt->addr;
 294 }
 295
 296 static void xs_format_ipv4_peer_addresses(struct rpc_xprt *xprt,
 297                                           const char *protocol,
 298                                           const char *netid)
 299 {
 300         struct sockaddr_in *addr = xs_addr_in(xprt);
 301         char *buf;
 302
 303         buf = kzalloc(20, GFP_KERNEL);
 304         if (buf) {
 305                 snprintf(buf, 20, "%pI4", &addr->sin_addr.s_addr);
 306         }
 307         xprt->address_strings[RPC_DISPLAY_ADDR] = buf;
 308
 309         buf = kzalloc(8, GFP_KERNEL);
 310         if (buf) {
 311                 snprintf(buf, 8, "%u",
 312                                 ntohs(addr->sin_port));
 313         }
 314         xprt->address_strings[RPC_DISPLAY_PORT] = buf;
 315
 316         xprt->address_strings[RPC_DISPLAY_PROTO] = protocol;
 317
 318         buf = kzalloc(48, GFP_KERNEL);
 319         if (buf) {
 320                 snprintf(buf, 48, "addr=%pI4 port=%u proto=%s",
 321                         &addr->sin_addr.s_addr,
 322                         ntohs(addr->sin_port),
 323                         protocol);
 324         }
 325         xprt->address_strings[RPC_DISPLAY_ALL] = buf;
 326
 327         buf = kzalloc(10, GFP_KERNEL);
 328         if (buf) {
 329                 snprintf(buf, 10, "%02x%02x%02x%02x",
 330                                 NIPQUAD(addr->sin_addr.s_addr));
 331         }
 332         xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = buf;
 333
 334         buf = kzalloc(8, GFP_KERNEL);
 335         if (buf) {
 336                 snprintf(buf, 8, "%4hx",
 337                                 ntohs(addr->sin_port));
 338         }
 339         xprt->address_strings[RPC_DISPLAY_HEX_PORT] = buf;
 340
 341         buf = kzalloc(30, GFP_KERNEL);
 342         if (buf) {
 343                 snprintf(buf, 30, "%pI4.%u.%u",
 344                                 &addr->sin_addr.s_addr,
 345                                 ntohs(addr->sin_port) >> 8,
 346                                 ntohs(addr->sin_port) & 0xff);
 347         }
 348         xprt->address_strings[RPC_DISPLAY_UNIVERSAL_ADDR] = buf;
 349
 350         xprt->address_strings[RPC_DISPLAY_NETID] = netid;
 351 }
 352
 353 static void xs_format_ipv6_peer_addresses(struct rpc_xprt *xprt,
 354                                           const char *protocol,
 355                                           const char *netid)
 356 {
 357         struct sockaddr_in6 *addr = xs_addr_in6(xprt);
 358         char *buf;
 359
 360         buf = kzalloc(40, GFP_KERNEL);
 361         if (buf) {
 362                 snprintf(buf, 40, "%pI6",&addr->sin6_addr);
 363         }
 364         xprt->address_strings[RPC_DISPLAY_ADDR] = buf;
 365
 366         buf = kzalloc(8, GFP_KERNEL);
 367         if (buf) {
 368                 snprintf(buf, 8, "%u",
 369                                 ntohs(addr->sin6_port));
 370         }
 371         xprt->address_strings[RPC_DISPLAY_PORT] = buf;
 372
 373         xprt->address_strings[RPC_DISPLAY_PROTO] = protocol;
 374
 375         buf = kzalloc(64, GFP_KERNEL);
 376         if (buf) {
 377                 snprintf(buf, 64, "addr=%pI6 port=%u proto=%s",
 378                                 &addr->sin6_addr,
 379                                 ntohs(addr->sin6_port),
 380                                 protocol);
 381         }
 382         xprt->address_strings[RPC_DISPLAY_ALL] = buf;
 383
 384         buf = kzalloc(36, GFP_KERNEL);
 385         if (buf)
 386                 snprintf(buf, 36, "%pi6", &addr->sin6_addr);
 387
 388         xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = buf;
 389
 390         buf = kzalloc(8, GFP_KERNEL);
 391         if (buf) {
 392                 snprintf(buf, 8, "%4hx",
 393                                 ntohs(addr->sin6_port));
 394         }
 395         xprt->address_strings[RPC_DISPLAY_HEX_PORT] = buf;
 396
 397         buf = kzalloc(50, GFP_KERNEL);
 398         if (buf) {
 399                 snprintf(buf, 50, "%pI6.%u.%u",
 400                          &addr->sin6_addr,
 401                          ntohs(addr->sin6_port) >> 8,
 402                          ntohs(addr->sin6_port) & 0xff);
 403         }
 404         xprt->address_strings[RPC_DISPLAY_UNIVERSAL_ADDR] = buf;
 405
 406         xprt->address_strings[RPC_DISPLAY_NETID] = netid;
 407 }
 408
 409 static void xs_free_peer_addresses(struct rpc_xprt *xprt)
 410 {
 411         unsigned int i;
 412
 413         for (i = 0; i < RPC_DISPLAY_MAX; i++)
 414                 switch (i) {
 415                 case RPC_DISPLAY_PROTO:
 416                 case RPC_DISPLAY_NETID:
 417                         continue;
 418                 default:
 419                         kfree(xprt->address_strings[i]);
 420                 }
 421 }
 422
 423 #define XS_SENDMSG_FLAGS        (MSG_DONTWAIT | MSG_NOSIGNAL)
 424
 425 static int xs_send_kvec(struct socket *sock, struct sockaddr *addr, int addrlen, struct kvec *vec, unsigned int base, int more)
 426 {
 427         struct msghdr msg = {
 428                 .msg_name       = addr,
 429                 .msg_namelen    = addrlen,
 430                 .msg_flags      = XS_SENDMSG_FLAGS | (more ? MSG_MORE : 0),
 431         };
 432         struct kvec iov = {
 433                 .iov_base       = vec->iov_base + base,
 434                 .iov_len        = vec->iov_len - base,
 435         };
 436
 437         if (iov.iov_len != 0)
 438                 return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
 439         return kernel_sendmsg(sock, &msg, NULL, 0, 0);
 440 }
 441
 442 static int xs_send_pagedata(struct socket *sock, struct xdr_buf *xdr, unsigned int base, int more)
 443 {
 444         struct page **ppage;
 445         unsigned int remainder;
 446         int err, sent = 0;
 447
 448         remainder = xdr->page_len - base;
 449         base += xdr->page_base;
 450         ppage = xdr->pages + (base >> PAGE_SHIFT);
 451         base &= ~PAGE_MASK;
 452         for(;;) {
 453                 unsigned int len = min_t(unsigned int, PAGE_SIZE - base, remainder);
 454                 int flags = XS_SENDMSG_FLAGS;
 455
 456                 remainder -= len;
 457                 if (remainder != 0 || more)
 458                         flags |= MSG_MORE;
 459                 err = sock->ops->sendpage(sock, *ppage, base, len, flags);
 460                 if (remainder == 0 || err != len)
 461                         break;
 462                 sent += err;
 463                 ppage++;
 464                 base = 0;
 465         }
 466         if (sent == 0)
 467                 return err;
 468         if (err > 0)
 469                 sent += err;
 470         return sent;
 471 }
 472
 473 /**
 474  * xs_sendpages - write pages directly to a socket
 475  * @sock: socket to send on
 476  * @addr: UDP only -- address of destination
 477  * @addrlen: UDP only -- length of destination address
 478  * @xdr: buffer containing this request
 479  * @base: starting position in the buffer
 480  *
 481  */
 482 static int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base)
 483 {
 484         unsigned int remainder = xdr->len - base;
 485         int err, sent = 0;
 486
 487         if (unlikely(!sock))
 488                 return -ENOTSOCK;
 489
 490         clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags);
 491         if (base != 0) {
 492                 addr = NULL;
 493                 addrlen = 0;
 494         }
 495
 496         if (base < xdr->head[0].iov_len || addr != NULL) {
 497                 unsigned int len = xdr->head[0].iov_len - base;
 498                 remainder -= len;
 499                 err = xs_send_kvec(sock, addr, addrlen, &xdr->head[0], base, remainder != 0);
 500                 if (remainder == 0 || err != len)
 501                         goto out;
 502                 sent += err;
 503                 base = 0;
 504         } else
 505                 base -= xdr->head[0].iov_len;
 506
 507         if (base < xdr->page_len) {
 508                 unsigned int len = xdr->page_len - base;
 509                 remainder -= len;
 510                 err = xs_send_pagedata(sock, xdr, base, remainder != 0);
 511                 if (remainder == 0 || err != len)
 512                         goto out;
 513                 sent += err;
 514                 base = 0;
 515         } else
 516                 base -= xdr->page_len;
 517
 518         if (base >= xdr->tail[0].iov_len)
 519                 return sent;
 520         err = xs_send_kvec(sock, NULL, 0, &xdr->tail[0], base, 0);
 521 out:
 522         if (sent == 0)
 523                 return err;
 524         if (err > 0)
 525                 sent += err;
 526         return sent;
 527 }
 528
 529 static void xs_nospace_callback(struct rpc_task *task)
 530 {
 531         struct sock_xprt *transport = container_of(task->tk_rqstp->rq_xprt, struct sock_xprt, xprt);
 532
 533         transport->inet->sk_write_pending--;
 534         clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 535 }
 536
 537 /**
 538  * xs_nospace - place task on wait queue if transmit was incomplete
 539  * @task: task to put to sleep
 540  *
 541  */
 542 static int xs_nospace(struct rpc_task *task)
 543 {
 544         struct rpc_rqst *req = task->tk_rqstp;
 545         struct rpc_xprt *xprt = req->rq_xprt;
 546         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 547         int ret = 0;
 548
 549         dprintk("RPC: %5u xmit incomplete (%u left of %u)\n",
 550                         task->tk_pid, req->rq_slen - req->rq_bytes_sent,
 551                         req->rq_slen);
 552
 553         /* Protect against races with write_space */
 554         spin_lock_bh(&xprt->transport_lock);
 555
 556         /* Don't race with disconnect */
 557         if (xprt_connected(xprt)) {
 558                 if (test_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags)) {
 559                         ret = -EAGAIN;
 560                         /*
 561                          * Notify TCP that we're limited by the application
 562                          * window size
 563                          */
 564                         set_bit(SOCK_NOSPACE, &transport->sock->flags);
 565                         transport->inet->sk_write_pending++;
 566                         /* ...and wait for more buffer space */
 567                         xprt_wait_for_buffer_space(task, xs_nospace_callback);
 568                 }
 569         } else {
 570                 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 571                 ret = -ENOTCONN;
 572         }
 573
 574         spin_unlock_bh(&xprt->transport_lock);
 575         return ret;
 576 }
 577
 578 /**
 579  * xs_udp_send_request - write an RPC request to a UDP socket
 580  * @task: address of RPC task that manages the state of an RPC request
 581  *
 582  * Return values:
 583  *        0:    The request has been sent
 584  *   EAGAIN:    The socket was blocked, please call again later to
 585  *              complete the request
 586  * ENOTCONN:    Caller needs to invoke connect logic then call again
 587  *    other:    Some other error occured, the request was not sent
 588  */
 589 static int xs_udp_send_request(struct rpc_task *task)
 590 {
 591         struct rpc_rqst *req = task->tk_rqstp;
 592         struct rpc_xprt *xprt = req->rq_xprt;
 593         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 594         struct xdr_buf *xdr = &req->rq_snd_buf;
 595         int status;
 596
 597         xs_pktdump("packet data:",
 598                                 req->rq_svec->iov_base,
 599                                 req->rq_svec->iov_len);
 600
 601         if (!xprt_bound(xprt))
 602                 return -ENOTCONN;
 603         status = xs_sendpages(transport->sock,
 604                               xs_addr(xprt),
 605                               xprt->addrlen, xdr,
 606                               req->rq_bytes_sent);
 607
 608         dprintk("RPC:       xs_udp_send_request(%u) = %d\n",
 609                         xdr->len - req->rq_bytes_sent, status);
 610
 611         if (status >= 0) {
 612                 task->tk_bytes_sent += status;
 613                 if (status >= req->rq_slen)
 614                         return 0;
 615                 /* Still some bytes left; set up for a retry later. */
 616                 status = -EAGAIN;
 617         }
 618         if (!transport->sock)
 619                 goto out;
 620
 621         switch (status) {
 622         case -ENOTSOCK:
 623                 status = -ENOTCONN;
 624                 /* Should we call xs_close() here? */
 625                 break;
 626         case -EAGAIN:
 627                 status = xs_nospace(task);
 628                 break;
 629         default:
 630                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
 631                         -status);
 632         case -ENETUNREACH:
 633         case -EPIPE:
 634         case -ECONNREFUSED:
 635                 /* When the server has died, an ICMP port unreachable message
 636                  * prompts ECONNREFUSED. */
 637                 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 638         }
 639 out:
 640         return status;
 641 }
 642
 643 /**
 644  * xs_tcp_shutdown - gracefully shut down a TCP socket
 645  * @xprt: transport
 646  *
 647  * Initiates a graceful shutdown of the TCP socket by calling the
 648  * equivalent of shutdown(SHUT_WR);
 649  */
 650 static void xs_tcp_shutdown(struct rpc_xprt *xprt)
 651 {
 652         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 653         struct socket *sock = transport->sock;
 654
 655         if (sock != NULL)
 656                 kernel_sock_shutdown(sock, SHUT_WR);
 657 }
 658
 659 static inline void xs_encode_tcp_record_marker(struct xdr_buf *buf)
 660 {
 661         u32 reclen = buf->len - sizeof(rpc_fraghdr);
 662         rpc_fraghdr *base = buf->head[0].iov_base;
 663         *base = htonl(RPC_LAST_STREAM_FRAGMENT | reclen);
 664 }
 665
 666 /**
 667  * xs_tcp_send_request - write an RPC request to a TCP socket
 668  * @task: address of RPC task that manages the state of an RPC request
 669  *
 670  * Return values:
 671  *        0:    The request has been sent
 672  *   EAGAIN:    The socket was blocked, please call again later to
 673  *              complete the request
 674  * ENOTCONN:    Caller needs to invoke connect logic then call again
 675  *    other:    Some other error occured, the request was not sent
 676  *
 677  * XXX: In the case of soft timeouts, should we eventually give up
 678  *      if sendmsg is not able to make progress?
 679  */
 680 static int xs_tcp_send_request(struct rpc_task *task)
 681 {
 682         struct rpc_rqst *req = task->tk_rqstp;
 683         struct rpc_xprt *xprt = req->rq_xprt;
 684         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 685         struct xdr_buf *xdr = &req->rq_snd_buf;
 686         int status;
 687
 688         xs_encode_tcp_record_marker(&req->rq_snd_buf);
 689
 690         xs_pktdump("packet data:",
 691                                 req->rq_svec->iov_base,
 692                                 req->rq_svec->iov_len);
 693
 694         /* Continue transmitting the packet/record. We must be careful
 695          * to cope with writespace callbacks arriving _after_ we have
 696          * called sendmsg(). */
 697         while (1) {
 698                 status = xs_sendpages(transport->sock,
 699                                         NULL, 0, xdr, req->rq_bytes_sent);
 700
 701                 dprintk("RPC:       xs_tcp_send_request(%u) = %d\n",
 702                                 xdr->len - req->rq_bytes_sent, status);
 703
 704                 if (unlikely(status < 0))
 705                         break;
 706
 707                 /* If we've sent the entire packet, immediately
 708                  * reset the count of bytes sent. */
 709                 req->rq_bytes_sent += status;
 710                 task->tk_bytes_sent += status;
 711                 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
 712                         req->rq_bytes_sent = 0;
 713                         return 0;
 714                 }
 715
 716                 if (status != 0)
 717                         continue;
 718                 status = -EAGAIN;
 719                 break;
 720         }
 721         if (!transport->sock)
 722                 goto out;
 723
 724         switch (status) {
 725         case -ENOTSOCK:
 726                 status = -ENOTCONN;
 727                 /* Should we call xs_close() here? */
 728                 break;
 729         case -EAGAIN:
 730                 status = xs_nospace(task);
 731                 break;
 732         default:
 733                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
 734                         -status);
 735         case -ECONNRESET:
 736         case -EPIPE:
 737                 xs_tcp_shutdown(xprt);
 738         case -ECONNREFUSED:
 739         case -ENOTCONN:
 740                 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 741         }
 742 out:
 743         return status;
 744 }
 745
 746 /**
 747  * xs_tcp_release_xprt - clean up after a tcp transmission
 748  * @xprt: transport
 749  * @task: rpc task
 750  *
 751  * This cleans up if an error causes us to abort the transmission of a request.
 752  * In this case, the socket may need to be reset in order to avoid confusing
 753  * the server.
 754  */
 755 static void xs_tcp_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
 756 {
 757         struct rpc_rqst *req;
 758
 759         if (task != xprt->snd_task)
 760                 return;
 761         if (task == NULL)
 762                 goto out_release;
 763         req = task->tk_rqstp;
 764         if (req->rq_bytes_sent == 0)
 765                 goto out_release;
 766         if (req->rq_bytes_sent == req->rq_snd_buf.len)
 767                 goto out_release;
 768         set_bit(XPRT_CLOSE_WAIT, &task->tk_xprt->state);
 769 out_release:
 770         xprt_release_xprt(xprt, task);
 771 }
 772
 773 static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk)
 774 {
 775         transport->old_data_ready = sk->sk_data_ready;
 776         transport->old_state_change = sk->sk_state_change;
 777         transport->old_write_space = sk->sk_write_space;
 778         transport->old_error_report = sk->sk_error_report;
 779 }
 780
 781 static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk)
 782 {
 783         sk->sk_data_ready = transport->old_data_ready;
 784         sk->sk_state_change = transport->old_state_change;
 785         sk->sk_write_space = transport->old_write_space;
 786         sk->sk_error_report = transport->old_error_report;
 787 }
 788
 789 static void xs_reset_transport(struct sock_xprt *transport)
 790 {
 791         struct socket *sock = transport->sock;
 792         struct sock *sk = transport->inet;
 793
 794         if (sk == NULL)
 795                 return;
 796
 797         write_lock_bh(&sk->sk_callback_lock);
 798         transport->inet = NULL;
 799         transport->sock = NULL;
 800
 801         sk->sk_user_data = NULL;
 802
 803         xs_restore_old_callbacks(transport, sk);
 804         write_unlock_bh(&sk->sk_callback_lock);
 805
 806         sk->sk_no_check = 0;
 807
 808         sock_release(sock);
 809 }
 810
 811 /**
 812  * xs_close - close a socket
 813  * @xprt: transport
 814  *
 815  * This is used when all requests are complete; ie, no DRC state remains
 816  * on the server we want to save.
 817  *
 818  * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with
 819  * xs_reset_transport() zeroing the socket from underneath a writer.
 820  */
 821 static void xs_close(struct rpc_xprt *xprt)
 822 {
 823         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 824
 825         dprintk("RPC:       xs_close xprt %p\n", xprt);
 826
 827         xs_reset_transport(transport);
 828
 829         smp_mb__before_clear_bit();
 830         clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
 831         clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
 832         clear_bit(XPRT_CLOSING, &xprt->state);
 833         smp_mb__after_clear_bit();
 834         xprt_disconnect_done(xprt);
 835 }
 836
 837 static void xs_tcp_close(struct rpc_xprt *xprt)
 838 {
 839         if (test_and_clear_bit(XPRT_CONNECTION_CLOSE, &xprt->state))
 840                 xs_close(xprt);
 841         else
 842                 xs_tcp_shutdown(xprt);
 843 }
 844
 845 /**
 846  * xs_destroy - prepare to shutdown a transport
 847  * @xprt: doomed transport
 848  *
 849  */
 850 static void xs_destroy(struct rpc_xprt *xprt)
 851 {
 852         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 853
 854         dprintk("RPC:       xs_destroy xprt %p\n", xprt);
 855
 856         cancel_rearming_delayed_work(&transport->connect_worker);
 857
 858         xs_close(xprt);
 859         xs_free_peer_addresses(xprt);
 860         kfree(xprt->slot);
 861         kfree(xprt);
 862         module_put(THIS_MODULE);
 863 }
 864
 865 static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
 866 {
 867         return (struct rpc_xprt *) sk->sk_user_data;
 868 }
 869
 870 /**
 871  * xs_udp_data_ready - "data ready" callback for UDP sockets
 872  * @sk: socket with data to read
 873  * @len: how much data to read
 874  *
 875  */
 876 static void xs_udp_data_ready(struct sock *sk, int len)
 877 {
 878         struct rpc_task *task;
 879         struct rpc_xprt *xprt;
 880         struct rpc_rqst *rovr;
 881         struct sk_buff *skb;
 882         int err, repsize, copied;
 883         u32 _xid;
 884         __be32 *xp;
 885
 886         read_lock(&sk->sk_callback_lock);
 887         dprintk("RPC:       xs_udp_data_ready...\n");
 888         if (!(xprt = xprt_from_sock(sk)))
 889                 goto out;
 890
 891         if ((skb = skb_recv_datagram(sk, 0, 1, &err)) == NULL)
 892                 goto out;
 893
 894         if (xprt->shutdown)
 895                 goto dropit;
 896
 897         repsize = skb->len - sizeof(struct udphdr);
 898         if (repsize < 4) {
 899                 dprintk("RPC:       impossible RPC reply size %d!\n", repsize);
 900                 goto dropit;
 901         }
 902
 903         /* Copy the XID from the skb... */
 904         xp = skb_header_pointer(skb, sizeof(struct udphdr),
 905                                 sizeof(_xid), &_xid);
 906         if (xp == NULL)
 907                 goto dropit;
 908
 909         /* Look up and lock the request corresponding to the given XID */
 910         spin_lock(&xprt->transport_lock);
 911         rovr = xprt_lookup_rqst(xprt, *xp);
 912         if (!rovr)
 913                 goto out_unlock;
 914         task = rovr->rq_task;
 915
 916         if ((copied = rovr->rq_private_buf.buflen) > repsize)
 917                 copied = repsize;
 918
 919         /* Suck it into the iovec, verify checksum if not done by hw. */
 920         if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
 921                 UDPX_INC_STATS_BH(sk, UDP_MIB_INERRORS);
 922                 goto out_unlock;
 923         }
 924
 925         UDPX_INC_STATS_BH(sk, UDP_MIB_INDATAGRAMS);
 926
 927         /* Something worked... */
 928         dst_confirm(skb->dst);
 929
 930         xprt_adjust_cwnd(task, copied);
 931         xprt_update_rtt(task);
 932         xprt_complete_rqst(task, copied);
 933
 934  out_unlock:
 935         spin_unlock(&xprt->transport_lock);
 936  dropit:
 937         skb_free_datagram(sk, skb);
 938  out:
 939         read_unlock(&sk->sk_callback_lock);
 940 }
 941
 942 static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
 943 {
 944         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 945         size_t len, used;
 946         char *p;
 947
 948         p = ((char *) &transport->tcp_fraghdr) + transport->tcp_offset;
 949         len = sizeof(transport->tcp_fraghdr) - transport->tcp_offset;
 950         used = xdr_skb_read_bits(desc, p, len);
 951         transport->tcp_offset += used;
 952         if (used != len)
 953                 return;
 954
 955         transport->tcp_reclen = ntohl(transport->tcp_fraghdr);
 956         if (transport->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
 957                 transport->tcp_flags |= TCP_RCV_LAST_FRAG;
 958         else
 959                 transport->tcp_flags &= ~TCP_RCV_LAST_FRAG;
 960         transport->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
 961
 962         transport->tcp_flags &= ~TCP_RCV_COPY_FRAGHDR;
 963         transport->tcp_offset = 0;
 964
 965         /* Sanity check of the record length */
 966         if (unlikely(transport->tcp_reclen < 8)) {
 967                 dprintk("RPC:       invalid TCP record fragment length\n");
 968                 xprt_force_disconnect(xprt);
 969                 return;
 970         }
 971         dprintk("RPC:       reading TCP record fragment of length %d\n",
 972                         transport->tcp_reclen);
 973 }
 974
 975 static void xs_tcp_check_fraghdr(struct sock_xprt *transport)
 976 {
 977         if (transport->tcp_offset == transport->tcp_reclen) {
 978                 transport->tcp_flags |= TCP_RCV_COPY_FRAGHDR;
 979                 transport->tcp_offset = 0;
 980                 if (transport->tcp_flags & TCP_RCV_LAST_FRAG) {
 981                         transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
 982                         transport->tcp_flags |= TCP_RCV_COPY_XID;
 983                         transport->tcp_copied = 0;
 984                 }
 985         }
 986 }
 987
 988 static inline void xs_tcp_read_xid(struct sock_xprt *transport, struct xdr_skb_reader *desc)
 989 {
 990         size_t len, used;
 991         char *p;
 992
 993         len = sizeof(transport->tcp_xid) - transport->tcp_offset;
 994         dprintk("RPC:       reading XID (%Zu bytes)\n", len);
 995         p = ((char *) &transport->tcp_xid) + transport->tcp_offset;
 996         used = xdr_skb_read_bits(desc, p, len);
 997         transport->tcp_offset += used;
 998         if (used != len)
 999                 return;
1000         transport->tcp_flags &= ~TCP_RCV_COPY_XID;
1001         transport->tcp_flags |= TCP_RCV_READ_CALLDIR;
1002         transport->tcp_copied = 4;
1003         dprintk("RPC:       reading %s XID %08x\n",
1004                         (transport->tcp_flags & TCP_RPC_REPLY) ? "reply for"
1005                                                               : "request with",
1006                         ntohl(transport->tcp_xid));
1007         xs_tcp_check_fraghdr(transport);
1008 }
1009
1010 static inline void xs_tcp_read_calldir(struct sock_xprt *transport,
1011                                        struct xdr_skb_reader *desc)
1012 {
1013         size_t len, used;
1014         u32 offset;
1015         __be32  calldir;
1016
1017         /*
1018          * We want transport->tcp_offset to be 8 at the end of this routine
1019          * (4 bytes for the xid and 4 bytes for the call/reply flag).
1020          * When this function is called for the first time,
1021          * transport->tcp_offset is 4 (after having already read the xid).
1022          */
1023         offset = transport->tcp_offset - sizeof(transport->tcp_xid);
1024         len = sizeof(calldir) - offset;
1025         dprintk("RPC:       reading CALL/REPLY flag (%Zu bytes)\n", len);
1026         used = xdr_skb_read_bits(desc, &calldir, len);
1027         transport->tcp_offset += used;
1028         if (used != len)
1029                 return;
1030         transport->tcp_flags &= ~TCP_RCV_READ_CALLDIR;
1031         transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1032         transport->tcp_flags |= TCP_RCV_COPY_DATA;
1033         /*
1034          * We don't yet have the XDR buffer, so we will write the calldir
1035          * out after we get the buffer from the 'struct rpc_rqst'
1036          */
1037         if (ntohl(calldir) == RPC_REPLY)
1038                 transport->tcp_flags |= TCP_RPC_REPLY;
1039         else
1040                 transport->tcp_flags &= ~TCP_RPC_REPLY;
1041         dprintk("RPC:       reading %s CALL/REPLY flag %08x\n",
1042                         (transport->tcp_flags & TCP_RPC_REPLY) ?
1043                                 "reply for" : "request with", calldir);
1044         xs_tcp_check_fraghdr(transport);
1045 }
1046
1047 static inline void xs_tcp_read_request(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
1048 {
1049         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1050         struct rpc_rqst *req;
1051         struct xdr_buf *rcvbuf;
1052         size_t len;
1053         ssize_t r;
1054
1055         /* Find and lock the request corresponding to this xid */
1056         spin_lock(&xprt->transport_lock);
1057         req = xprt_lookup_rqst(xprt, transport->tcp_xid);
1058         if (!req) {
1059                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1060                 dprintk("RPC:       XID %08x request not found!\n",
1061                                 ntohl(transport->tcp_xid));
1062                 spin_unlock(&xprt->transport_lock);
1063                 return;
1064         }
1065
1066         rcvbuf = &req->rq_private_buf;
1067
1068         if (transport->tcp_flags & TCP_RCV_COPY_CALLDIR) {
1069                 /*
1070                  * Save the RPC direction in the XDR buffer
1071                  */
1072                 __be32  calldir = transport->tcp_flags & TCP_RPC_REPLY ?
1073                                         htonl(RPC_REPLY) : 0;
1074
1075                 memcpy(rcvbuf->head[0].iov_base + transport->tcp_copied,
1076                         &calldir, sizeof(calldir));
1077                 transport->tcp_copied += sizeof(calldir);
1078                 transport->tcp_flags &= ~TCP_RCV_COPY_CALLDIR;
1079         }
1080
1081         len = desc->count;
1082         if (len > transport->tcp_reclen - transport->tcp_offset) {
1083                 struct xdr_skb_reader my_desc;
1084
1085                 len = transport->tcp_reclen - transport->tcp_offset;
1086                 memcpy(&my_desc, desc, sizeof(my_desc));
1087                 my_desc.count = len;
1088                 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
1089                                           &my_desc, xdr_skb_read_bits);
1090                 desc->count -= r;
1091                 desc->offset += r;
1092         } else
1093                 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
1094                                           desc, xdr_skb_read_bits);
1095
1096         if (r > 0) {
1097                 transport->tcp_copied += r;
1098                 transport->tcp_offset += r;
1099         }
1100         if (r != len) {
1101                 /* Error when copying to the receive buffer,
1102                  * usually because we weren't able to allocate
1103                  * additional buffer pages. All we can do now
1104                  * is turn off TCP_RCV_COPY_DATA, so the request
1105                  * will not receive any additional updates,
1106                  * and time out.
1107                  * Any remaining data from this record will
1108                  * be discarded.
1109                  */
1110                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1111                 dprintk("RPC:       XID %08x truncated request\n",
1112                                 ntohl(transport->tcp_xid));
1113                 dprintk("RPC:       xprt = %p, tcp_copied = %lu, "
1114                                 "tcp_offset = %u, tcp_reclen = %u\n",
1115                                 xprt, transport->tcp_copied,
1116                                 transport->tcp_offset, transport->tcp_reclen);
1117                 goto out;
1118         }
1119
1120         dprintk("RPC:       XID %08x read %Zd bytes\n",
1121                         ntohl(transport->tcp_xid), r);
1122         dprintk("RPC:       xprt = %p, tcp_copied = %lu, tcp_offset = %u, "
1123                         "tcp_reclen = %u\n", xprt, transport->tcp_copied,
1124                         transport->tcp_offset, transport->tcp_reclen);
1125
1126         if (transport->tcp_copied == req->rq_private_buf.buflen)
1127                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1128         else if (transport->tcp_offset == transport->tcp_reclen) {
1129                 if (transport->tcp_flags & TCP_RCV_LAST_FRAG)
1130                         transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1131         }
1132
1133 out:
1134         if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1135                 xprt_complete_rqst(req->rq_task, transport->tcp_copied);
1136         spin_unlock(&xprt->transport_lock);
1137         xs_tcp_check_fraghdr(transport);
1138 }
1139
1140 static inline void xs_tcp_read_discard(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1141 {
1142         size_t len;
1143
1144         len = transport->tcp_reclen - transport->tcp_offset;
1145         if (len > desc->count)
1146                 len = desc->count;
1147         desc->count -= len;
1148         desc->offset += len;
1149         transport->tcp_offset += len;
1150         dprintk("RPC:       discarded %Zu bytes\n", len);
1151         xs_tcp_check_fraghdr(transport);
1152 }
1153
1154 static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
1155 {
1156         struct rpc_xprt *xprt = rd_desc->arg.data;
1157         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1158         struct xdr_skb_reader desc = {
1159                 .skb    = skb,
1160                 .offset = offset,
1161                 .count  = len,
1162         };
1163
1164         dprintk("RPC:       xs_tcp_data_recv started\n");
1165         do {
1166                 /* Read in a new fragment marker if necessary */
1167                 /* Can we ever really expect to get completely empty fragments? */
1168                 if (transport->tcp_flags & TCP_RCV_COPY_FRAGHDR) {
1169                         xs_tcp_read_fraghdr(xprt, &desc);
1170                         continue;
1171                 }
1172                 /* Read in the xid if necessary */
1173                 if (transport->tcp_flags & TCP_RCV_COPY_XID) {
1174                         xs_tcp_read_xid(transport, &desc);
1175                         continue;
1176                 }
1177                 /* Read in the call/reply flag */
1178                 if (transport->tcp_flags & TCP_RCV_READ_CALLDIR) {
1179                         xs_tcp_read_calldir(transport, &desc);
1180                         continue;
1181                 }
1182                 /* Read in the request data */
1183                 if (transport->tcp_flags & TCP_RCV_COPY_DATA) {
1184                         xs_tcp_read_request(xprt, &desc);
1185                         continue;
1186                 }
1187                 /* Skip over any trailing bytes on short reads */
1188                 xs_tcp_read_discard(transport, &desc);
1189         } while (desc.count);
1190         dprintk("RPC:       xs_tcp_data_recv done\n");
1191         return len - desc.count;
1192 }
1193
1194 /**
1195  * xs_tcp_data_ready - "data ready" callback for TCP sockets
1196  * @sk: socket with data to read
1197  * @bytes: how much data to read
1198  *
1199  */
1200 static void xs_tcp_data_ready(struct sock *sk, int bytes)
1201 {
1202         struct rpc_xprt *xprt;
1203         read_descriptor_t rd_desc;
1204         int read;
1205
1206         dprintk("RPC:       xs_tcp_data_ready...\n");
1207
1208         read_lock(&sk->sk_callback_lock);
1209         if (!(xprt = xprt_from_sock(sk)))
1210                 goto out;
1211         if (xprt->shutdown)
1212                 goto out;
1213
1214         /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
1215         rd_desc.arg.data = xprt;
1216         do {
1217                 rd_desc.count = 65536;
1218                 read = tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
1219         } while (read > 0);
1220 out:
1221         read_unlock(&sk->sk_callback_lock);
1222 }
1223
1224 /*
1225  * Do the equivalent of linger/linger2 handling for dealing with
1226  * broken servers that don't close the socket in a timely
1227  * fashion
1228  */
1229 static void xs_tcp_schedule_linger_timeout(struct rpc_xprt *xprt,
1230                 unsigned long timeout)
1231 {
1232         struct sock_xprt *transport;
1233
1234         if (xprt_test_and_set_connecting(xprt))
1235                 return;
1236         set_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1237         transport = container_of(xprt, struct sock_xprt, xprt);
1238         queue_delayed_work(rpciod_workqueue, &transport->connect_worker,
1239                            timeout);
1240 }
1241
1242 static void xs_tcp_cancel_linger_timeout(struct rpc_xprt *xprt)
1243 {
1244         struct sock_xprt *transport;
1245
1246         transport = container_of(xprt, struct sock_xprt, xprt);
1247
1248         if (!test_bit(XPRT_CONNECTION_ABORT, &xprt->state) ||
1249             !cancel_delayed_work(&transport->connect_worker))
1250                 return;
1251         clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1252         xprt_clear_connecting(xprt);
1253 }
1254
1255 static void xs_sock_mark_closed(struct rpc_xprt *xprt)
1256 {
1257         smp_mb__before_clear_bit();
1258         clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1259         clear_bit(XPRT_CLOSING, &xprt->state);
1260         smp_mb__after_clear_bit();
1261         /* Mark transport as closed and wake up all pending tasks */
1262         xprt_disconnect_done(xprt);
1263 }
1264
1265 /**
1266  * xs_tcp_state_change - callback to handle TCP socket state changes
1267  * @sk: socket whose state has changed
1268  *
1269  */
1270 static void xs_tcp_state_change(struct sock *sk)
1271 {
1272         struct rpc_xprt *xprt;
1273
1274         read_lock(&sk->sk_callback_lock);
1275         if (!(xprt = xprt_from_sock(sk)))
1276                 goto out;
1277         dprintk("RPC:       xs_tcp_state_change client %p...\n", xprt);
1278         dprintk("RPC:       state %x conn %d dead %d zapped %d\n",
1279                         sk->sk_state, xprt_connected(xprt),
1280                         sock_flag(sk, SOCK_DEAD),
1281                         sock_flag(sk, SOCK_ZAPPED));
1282
1283         switch (sk->sk_state) {
1284         case TCP_ESTABLISHED:
1285                 spin_lock_bh(&xprt->transport_lock);
1286                 if (!xprt_test_and_set_connected(xprt)) {
1287                         struct sock_xprt *transport = container_of(xprt,
1288                                         struct sock_xprt, xprt);
1289
1290                         /* Reset TCP record info */
1291                         transport->tcp_offset = 0;
1292                         transport->tcp_reclen = 0;
1293                         transport->tcp_copied = 0;
1294                         transport->tcp_flags =
1295                                 TCP_RCV_COPY_FRAGHDR | TCP_RCV_COPY_XID;
1296
1297                         xprt_wake_pending_tasks(xprt, -EAGAIN);
1298                 }
1299                 spin_unlock_bh(&xprt->transport_lock);
1300                 break;
1301         case TCP_FIN_WAIT1:
1302                 /* The client initiated a shutdown of the socket */
1303                 xprt->connect_cookie++;
1304                 xprt->reestablish_timeout = 0;
1305                 set_bit(XPRT_CLOSING, &xprt->state);
1306                 smp_mb__before_clear_bit();
1307                 clear_bit(XPRT_CONNECTED, &xprt->state);
1308                 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1309                 smp_mb__after_clear_bit();
1310                 xs_tcp_schedule_linger_timeout(xprt, xs_tcp_fin_timeout);
1311                 break;
1312         case TCP_CLOSE_WAIT:
1313                 /* The server initiated a shutdown of the socket */
1314                 xprt_force_disconnect(xprt);
1315         case TCP_SYN_SENT:
1316                 xprt->connect_cookie++;
1317         case TCP_CLOSING:
1318                 /*
1319                  * If the server closed down the connection, make sure that
1320                  * we back off before reconnecting
1321                  */
1322                 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1323                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1324                 break;
1325         case TCP_LAST_ACK:
1326                 set_bit(XPRT_CLOSING, &xprt->state);
1327                 xs_tcp_schedule_linger_timeout(xprt, xs_tcp_fin_timeout);
1328                 smp_mb__before_clear_bit();
1329                 clear_bit(XPRT_CONNECTED, &xprt->state);
1330                 smp_mb__after_clear_bit();
1331                 break;
1332         case TCP_CLOSE:
1333                 xs_tcp_cancel_linger_timeout(xprt);
1334                 xs_sock_mark_closed(xprt);
1335         }
1336  out:
1337         read_unlock(&sk->sk_callback_lock);
1338 }
1339
1340 /**
1341  * xs_error_report - callback mainly for catching socket errors
1342  * @sk: socket
1343  */
1344 static void xs_error_report(struct sock *sk)
1345 {
1346         struct rpc_xprt *xprt;
1347
1348         read_lock(&sk->sk_callback_lock);
1349         if (!(xprt = xprt_from_sock(sk)))
1350                 goto out;
1351         dprintk("RPC:       %s client %p...\n"
1352                         "RPC:       error %d\n",
1353                         __func__, xprt, sk->sk_err);
1354         xprt_wake_pending_tasks(xprt, -EAGAIN);
1355 out:
1356         read_unlock(&sk->sk_callback_lock);
1357 }
1358
1359 static void xs_write_space(struct sock *sk)
1360 {
1361         struct socket *sock;
1362         struct rpc_xprt *xprt;
1363
1364         if (unlikely(!(sock = sk->sk_socket)))
1365                 return;
1366         clear_bit(SOCK_NOSPACE, &sock->flags);
1367
1368         if (unlikely(!(xprt = xprt_from_sock(sk))))
1369                 return;
1370         if (test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags) == 0)
1371                 return;
1372
1373         xprt_write_space(xprt);
1374 }
1375
1376 /**
1377  * xs_udp_write_space - callback invoked when socket buffer space
1378  *                             becomes available
1379  * @sk: socket whose state has changed
1380  *
1381  * Called when more output buffer space is available for this socket.
1382  * We try not to wake our writers until they can make "significant"
1383  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1384  * with a bunch of small requests.
1385  */
1386 static void xs_udp_write_space(struct sock *sk)
1387 {
1388         read_lock(&sk->sk_callback_lock);
1389
1390         /* from net/core/sock.c:sock_def_write_space */
1391         if (sock_writeable(sk))
1392                 xs_write_space(sk);
1393
1394         read_unlock(&sk->sk_callback_lock);
1395 }
1396
1397 /**
1398  * xs_tcp_write_space - callback invoked when socket buffer space
1399  *                             becomes available
1400  * @sk: socket whose state has changed
1401  *
1402  * Called when more output buffer space is available for this socket.
1403  * We try not to wake our writers until they can make "significant"
1404  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1405  * with a bunch of small requests.
1406  */
1407 static void xs_tcp_write_space(struct sock *sk)
1408 {
1409         read_lock(&sk->sk_callback_lock);
1410
1411         /* from net/core/stream.c:sk_stream_write_space */
1412         if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk))
1413                 xs_write_space(sk);
1414
1415         read_unlock(&sk->sk_callback_lock);
1416 }
1417
1418 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
1419 {
1420         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1421         struct sock *sk = transport->inet;
1422
1423         if (transport->rcvsize) {
1424                 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1425                 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
1426         }
1427         if (transport->sndsize) {
1428                 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1429                 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1430                 sk->sk_write_space(sk);
1431         }
1432 }
1433
1434 /**
1435  * xs_udp_set_buffer_size - set send and receive limits
1436  * @xprt: generic transport
1437  * @sndsize: requested size of send buffer, in bytes
1438  * @rcvsize: requested size of receive buffer, in bytes
1439  *
1440  * Set socket send and receive buffer size limits.
1441  */
1442 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1443 {
1444         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1445
1446         transport->sndsize = 0;
1447         if (sndsize)
1448                 transport->sndsize = sndsize + 1024;
1449         transport->rcvsize = 0;
1450         if (rcvsize)
1451                 transport->rcvsize = rcvsize + 1024;
1452
1453         xs_udp_do_set_buffer_size(xprt);
1454 }
1455
1456 /**
1457  * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1458  * @task: task that timed out
1459  *
1460  * Adjust the congestion window after a retransmit timeout has occurred.
1461  */
1462 static void xs_udp_timer(struct rpc_task *task)
1463 {
1464         xprt_adjust_cwnd(task, -ETIMEDOUT);
1465 }
1466
1467 static unsigned short xs_get_random_port(void)
1468 {
1469         unsigned short range = xprt_max_resvport - xprt_min_resvport;
1470         unsigned short rand = (unsigned short) net_random() % range;
1471         return rand + xprt_min_resvport;
1472 }
1473
1474 /**
1475  * xs_set_port - reset the port number in the remote endpoint address
1476  * @xprt: generic transport
1477  * @port: new port number
1478  *
1479  */
1480 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1481 {
1482         struct sockaddr *addr = xs_addr(xprt);
1483
1484         dprintk("RPC:       setting port for xprt %p to %u\n", xprt, port);
1485
1486         switch (addr->sa_family) {
1487         case AF_INET:
1488                 ((struct sockaddr_in *)addr)->sin_port = htons(port);
1489                 break;
1490         case AF_INET6:
1491                 ((struct sockaddr_in6 *)addr)->sin6_port = htons(port);
1492                 break;
1493         default:
1494                 BUG();
1495         }
1496 }
1497
1498 static unsigned short xs_get_srcport(struct sock_xprt *transport, struct socket *sock)
1499 {
1500         unsigned short port = transport->port;
1501
1502         if (port == 0 && transport->xprt.resvport)
1503                 port = xs_get_random_port();
1504         return port;
1505 }
1506
1507 static unsigned short xs_next_srcport(struct sock_xprt *transport, struct socket *sock, unsigned short port)
1508 {
1509         if (transport->port != 0)
1510                 transport->port = 0;
1511         if (!transport->xprt.resvport)
1512                 return 0;
1513         if (port <= xprt_min_resvport || port > xprt_max_resvport)
1514                 return xprt_max_resvport;
1515         return --port;
1516 }
1517
1518 static int xs_bind4(struct sock_xprt *transport, struct socket *sock)
1519 {
1520         struct sockaddr_in myaddr = {
1521                 .sin_family = AF_INET,
1522         };
1523         struct sockaddr_in *sa;
1524         int err, nloop = 0;
1525         unsigned short port = xs_get_srcport(transport, sock);
1526         unsigned short last;
1527
1528         sa = (struct sockaddr_in *)&transport->addr;
1529         myaddr.sin_addr = sa->sin_addr;
1530         do {
1531                 myaddr.sin_port = htons(port);
1532                 err = kernel_bind(sock, (struct sockaddr *) &myaddr,
1533                                                 sizeof(myaddr));
1534                 if (port == 0)
1535                         break;
1536                 if (err == 0) {
1537                         transport->port = port;
1538                         break;
1539                 }
1540                 last = port;
1541                 port = xs_next_srcport(transport, sock, port);
1542                 if (port > last)
1543                         nloop++;
1544         } while (err == -EADDRINUSE && nloop != 2);
1545         dprintk("RPC:       %s %pI4:%u: %s (%d)\n",
1546                         __func__, &myaddr.sin_addr,
1547                         port, err ? "failed" : "ok", err);
1548         return err;
1549 }
1550
1551 static int xs_bind6(struct sock_xprt *transport, struct socket *sock)
1552 {
1553         struct sockaddr_in6 myaddr = {
1554                 .sin6_family = AF_INET6,
1555         };
1556         struct sockaddr_in6 *sa;
1557         int err, nloop = 0;
1558         unsigned short port = xs_get_srcport(transport, sock);
1559         unsigned short last;
1560
1561         sa = (struct sockaddr_in6 *)&transport->addr;
1562         myaddr.sin6_addr = sa->sin6_addr;
1563         do {
1564                 myaddr.sin6_port = htons(port);
1565                 err = kernel_bind(sock, (struct sockaddr *) &myaddr,
1566                                                 sizeof(myaddr));
1567                 if (port == 0)
1568                         break;
1569                 if (err == 0) {
1570                         transport->port = port;
1571                         break;
1572                 }
1573                 last = port;
1574                 port = xs_next_srcport(transport, sock, port);
1575                 if (port > last)
1576                         nloop++;
1577         } while (err == -EADDRINUSE && nloop != 2);
1578         dprintk("RPC:       xs_bind6 %pI6:%u: %s (%d)\n",
1579                 &myaddr.sin6_addr, port, err ? "failed" : "ok", err);
1580         return err;
1581 }
1582
1583 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1584 static struct lock_class_key xs_key[2];
1585 static struct lock_class_key xs_slock_key[2];
1586
1587 static inline void xs_reclassify_socket4(struct socket *sock)
1588 {
1589         struct sock *sk = sock->sk;
1590
1591         BUG_ON(sock_owned_by_user(sk));
1592         sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1593                 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1594 }
1595
1596 static inline void xs_reclassify_socket6(struct socket *sock)
1597 {
1598         struct sock *sk = sock->sk;
1599
1600         BUG_ON(sock_owned_by_user(sk));
1601         sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1602                 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
1603 }
1604 #else
1605 static inline void xs_reclassify_socket4(struct socket *sock)
1606 {
1607 }
1608
1609 static inline void xs_reclassify_socket6(struct socket *sock)
1610 {
1611 }
1612 #endif
1613
1614 static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
1615 {
1616         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1617
1618         if (!transport->inet) {
1619                 struct sock *sk = sock->sk;
1620
1621                 write_lock_bh(&sk->sk_callback_lock);
1622
1623                 xs_save_old_callbacks(transport, sk);
1624
1625                 sk->sk_user_data = xprt;
1626                 sk->sk_data_ready = xs_udp_data_ready;
1627                 sk->sk_write_space = xs_udp_write_space;
1628                 sk->sk_error_report = xs_error_report;
1629                 sk->sk_no_check = UDP_CSUM_NORCV;
1630                 sk->sk_allocation = GFP_ATOMIC;
1631
1632                 xprt_set_connected(xprt);
1633
1634                 /* Reset to new socket */
1635                 transport->sock = sock;
1636                 transport->inet = sk;
1637
1638                 write_unlock_bh(&sk->sk_callback_lock);
1639         }
1640         xs_udp_do_set_buffer_size(xprt);
1641 }
1642
1643 /**
1644  * xs_udp_connect_worker4 - set up a UDP socket
1645  * @work: RPC transport to connect
1646  *
1647  * Invoked by a work queue tasklet.
1648  */
1649 static void xs_udp_connect_worker4(struct work_struct *work)
1650 {
1651         struct sock_xprt *transport =
1652                 container_of(work, struct sock_xprt, connect_worker.work);
1653         struct rpc_xprt *xprt = &transport->xprt;
1654         struct socket *sock = transport->sock;
1655         int err, status = -EIO;
1656
1657         if (xprt->shutdown)
1658                 goto out;
1659
1660         /* Start by resetting any existing state */
1661         xs_reset_transport(transport);
1662
1663         err = sock_create_kern(PF_INET, SOCK_DGRAM, IPPROTO_UDP, &sock);
1664         if (err < 0) {
1665                 dprintk("RPC:       can't create UDP transport socket (%d).\n", -err);
1666                 goto out;
1667         }
1668         xs_reclassify_socket4(sock);
1669
1670         if (xs_bind4(transport, sock)) {
1671                 sock_release(sock);
1672                 goto out;
1673         }
1674
1675         dprintk("RPC:       worker connecting xprt %p to address: %s\n",
1676                         xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1677
1678         xs_udp_finish_connecting(xprt, sock);
1679         status = 0;
1680 out:
1681         xprt_clear_connecting(xprt);
1682         xprt_wake_pending_tasks(xprt, status);
1683 }
1684
1685 /**
1686  * xs_udp_connect_worker6 - set up a UDP socket
1687  * @work: RPC transport to connect
1688  *
1689  * Invoked by a work queue tasklet.
1690  */
1691 static void xs_udp_connect_worker6(struct work_struct *work)
1692 {
1693         struct sock_xprt *transport =
1694                 container_of(work, struct sock_xprt, connect_worker.work);
1695         struct rpc_xprt *xprt = &transport->xprt;
1696         struct socket *sock = transport->sock;
1697         int err, status = -EIO;
1698
1699         if (xprt->shutdown)
1700                 goto out;
1701
1702         /* Start by resetting any existing state */
1703         xs_reset_transport(transport);
1704
1705         err = sock_create_kern(PF_INET6, SOCK_DGRAM, IPPROTO_UDP, &sock);
1706         if (err < 0) {
1707                 dprintk("RPC:       can't create UDP transport socket (%d).\n", -err);
1708                 goto out;
1709         }
1710         xs_reclassify_socket6(sock);
1711
1712         if (xs_bind6(transport, sock) < 0) {
1713                 sock_release(sock);
1714                 goto out;
1715         }
1716
1717         dprintk("RPC:       worker connecting xprt %p to address: %s\n",
1718                         xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1719
1720         xs_udp_finish_connecting(xprt, sock);
1721         status = 0;
1722 out:
1723         xprt_clear_connecting(xprt);
1724         xprt_wake_pending_tasks(xprt, status);
1725 }
1726
1727 /*
1728  * We need to preserve the port number so the reply cache on the server can
1729  * find our cached RPC replies when we get around to reconnecting.
1730  */
1731 static void xs_abort_connection(struct rpc_xprt *xprt, struct sock_xprt *transport)
1732 {
1733         int result;
1734         struct sockaddr any;
1735
1736         dprintk("RPC:       disconnecting xprt %p to reuse port\n", xprt);
1737
1738         /*
1739          * Disconnect the transport socket by doing a connect operation
1740          * with AF_UNSPEC.  This should return immediately...
1741          */
1742         memset(&any, 0, sizeof(any));
1743         any.sa_family = AF_UNSPEC;
1744         result = kernel_connect(transport->sock, &any, sizeof(any), 0);
1745         if (!result)
1746                 xs_sock_mark_closed(xprt);
1747         else
1748                 dprintk("RPC:       AF_UNSPEC connect return code %d\n",
1749                                 result);
1750 }
1751
1752 static void xs_tcp_reuse_connection(struct rpc_xprt *xprt, struct sock_xprt *transport)
1753 {
1754         unsigned int state = transport->inet->sk_state;
1755
1756         if (state == TCP_CLOSE && transport->sock->state == SS_UNCONNECTED)
1757                 return;
1758         if ((1 << state) & (TCPF_ESTABLISHED|TCPF_SYN_SENT))
1759                 return;
1760         xs_abort_connection(xprt, transport);
1761 }
1762
1763 static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
1764 {
1765         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1766
1767         if (!transport->inet) {
1768                 struct sock *sk = sock->sk;
1769
1770                 write_lock_bh(&sk->sk_callback_lock);
1771
1772                 xs_save_old_callbacks(transport, sk);
1773
1774                 sk->sk_user_data = xprt;
1775                 sk->sk_data_ready = xs_tcp_data_ready;
1776                 sk->sk_state_change = xs_tcp_state_change;
1777                 sk->sk_write_space = xs_tcp_write_space;
1778                 sk->sk_error_report = xs_error_report;
1779                 sk->sk_allocation = GFP_ATOMIC;
1780
1781                 /* socket options */
1782                 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
1783                 sock_reset_flag(sk, SOCK_LINGER);
1784                 tcp_sk(sk)->linger2 = 0;
1785                 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
1786
1787                 xprt_clear_connected(xprt);
1788
1789                 /* Reset to new socket */
1790                 transport->sock = sock;
1791                 transport->inet = sk;
1792
1793                 write_unlock_bh(&sk->sk_callback_lock);
1794         }
1795
1796         if (!xprt_bound(xprt))
1797                 return -ENOTCONN;
1798
1799         /* Tell the socket layer to start connecting... */
1800         xprt->stat.connect_count++;
1801         xprt->stat.connect_start = jiffies;
1802         return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
1803 }
1804
1805 /**
1806  * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint
1807  * @xprt: RPC transport to connect
1808  * @transport: socket transport to connect
1809  * @create_sock: function to create a socket of the correct type
1810  *
1811  * Invoked by a work queue tasklet.
1812  */
1813 static void xs_tcp_setup_socket(struct rpc_xprt *xprt,
1814                 struct sock_xprt *transport,
1815                 struct socket *(*create_sock)(struct rpc_xprt *,
1816                         struct sock_xprt *))
1817 {
1818         struct socket *sock = transport->sock;
1819         int status = -EIO;
1820
1821         if (xprt->shutdown)
1822                 goto out;
1823
1824         if (!sock) {
1825                 clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1826                 sock = create_sock(xprt, transport);
1827                 if (IS_ERR(sock)) {
1828                         status = PTR_ERR(sock);
1829                         goto out;
1830                 }
1831         } else {
1832                 int abort_and_exit;
1833
1834                 abort_and_exit = test_and_clear_bit(XPRT_CONNECTION_ABORT,
1835                                 &xprt->state);
1836                 /* "close" the socket, preserving the local port */
1837                 xs_tcp_reuse_connection(xprt, transport);
1838
1839                 if (abort_and_exit)
1840                         goto out_eagain;
1841         }
1842
1843         dprintk("RPC:       worker connecting xprt %p to address: %s\n",
1844                         xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1845
1846         status = xs_tcp_finish_connecting(xprt, sock);
1847         dprintk("RPC:       %p connect status %d connected %d sock state %d\n",
1848                         xprt, -status, xprt_connected(xprt),
1849                         sock->sk->sk_state);
1850         switch (status) {
1851         default:
1852                 printk("%s: connect returned unhandled error %d\n",
1853                         __func__, status);
1854         case -EADDRNOTAVAIL:
1855                 /* We're probably in TIME_WAIT. Get rid of existing socket,
1856                  * and retry
1857                  */
1858                 set_bit(XPRT_CONNECTION_CLOSE, &xprt->state);
1859                 xprt_force_disconnect(xprt);
1860         case -ECONNREFUSED:
1861         case -ECONNRESET:
1862         case -ENETUNREACH:
1863                 /* retry with existing socket, after a delay */
1864         case 0:
1865         case -EINPROGRESS:
1866         case -EALREADY:
1867                 xprt_clear_connecting(xprt);
1868                 return;
1869         }
1870 out_eagain:
1871         status = -EAGAIN;
1872 out:
1873         xprt_clear_connecting(xprt);
1874         xprt_wake_pending_tasks(xprt, status);
1875 }
1876
1877 static struct socket *xs_create_tcp_sock4(struct rpc_xprt *xprt,
1878                 struct sock_xprt *transport)
1879 {
1880         struct socket *sock;
1881         int err;
1882
1883         /* start from scratch */
1884         err = sock_create_kern(PF_INET, SOCK_STREAM, IPPROTO_TCP, &sock);
1885         if (err < 0) {
1886                 dprintk("RPC:       can't create TCP transport socket (%d).\n",
1887                                 -err);
1888                 goto out_err;
1889         }
1890         xs_reclassify_socket4(sock);
1891
1892         if (xs_bind4(transport, sock) < 0) {
1893                 sock_release(sock);
1894                 goto out_err;
1895         }
1896         return sock;
1897 out_err:
1898         return ERR_PTR(-EIO);
1899 }
1900
1901 /**
1902  * xs_tcp_connect_worker4 - connect a TCP socket to a remote endpoint
1903  * @work: RPC transport to connect
1904  *
1905  * Invoked by a work queue tasklet.
1906  */
1907 static void xs_tcp_connect_worker4(struct work_struct *work)
1908 {
1909         struct sock_xprt *transport =
1910                 container_of(work, struct sock_xprt, connect_worker.work);
1911         struct rpc_xprt *xprt = &transport->xprt;
1912
1913         xs_tcp_setup_socket(xprt, transport, xs_create_tcp_sock4);
1914 }
1915
1916 static struct socket *xs_create_tcp_sock6(struct rpc_xprt *xprt,
1917                 struct sock_xprt *transport)
1918 {
1919         struct socket *sock;
1920         int err;
1921
1922         /* start from scratch */
1923         err = sock_create_kern(PF_INET6, SOCK_STREAM, IPPROTO_TCP, &sock);
1924         if (err < 0) {
1925                 dprintk("RPC:       can't create TCP transport socket (%d).\n",
1926                                 -err);
1927                 goto out_err;
1928         }
1929         xs_reclassify_socket6(sock);
1930
1931         if (xs_bind6(transport, sock) < 0) {
1932                 sock_release(sock);
1933                 goto out_err;
1934         }
1935         return sock;
1936 out_err:
1937         return ERR_PTR(-EIO);
1938 }
1939
1940 /**
1941  * xs_tcp_connect_worker6 - connect a TCP socket to a remote endpoint
1942  * @work: RPC transport to connect
1943  *
1944  * Invoked by a work queue tasklet.
1945  */
1946 static void xs_tcp_connect_worker6(struct work_struct *work)
1947 {
1948         struct sock_xprt *transport =
1949                 container_of(work, struct sock_xprt, connect_worker.work);
1950         struct rpc_xprt *xprt = &transport->xprt;
1951
1952         xs_tcp_setup_socket(xprt, transport, xs_create_tcp_sock6);
1953 }
1954
1955 /**
1956  * xs_connect - connect a socket to a remote endpoint
1957  * @task: address of RPC task that manages state of connect request
1958  *
1959  * TCP: If the remote end dropped the connection, delay reconnecting.
1960  *
1961  * UDP socket connects are synchronous, but we use a work queue anyway
1962  * to guarantee that even unprivileged user processes can set up a
1963  * socket on a privileged port.
1964  *
1965  * If a UDP socket connect fails, the delay behavior here prevents
1966  * retry floods (hard mounts).
1967  */
1968 static void xs_connect(struct rpc_task *task)
1969 {
1970         struct rpc_xprt *xprt = task->tk_xprt;
1971         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1972
1973         if (xprt_test_and_set_connecting(xprt))
1974                 return;
1975
1976         if (transport->sock != NULL) {
1977                 dprintk("RPC:       xs_connect delayed xprt %p for %lu "
1978                                 "seconds\n",
1979                                 xprt, xprt->reestablish_timeout / HZ);
1980                 queue_delayed_work(rpciod_workqueue,
1981                                    &transport->connect_worker,
1982                                    xprt->reestablish_timeout);
1983                 xprt->reestablish_timeout <<= 1;
1984                 if (xprt->reestablish_timeout > XS_TCP_MAX_REEST_TO)
1985                         xprt->reestablish_timeout = XS_TCP_MAX_REEST_TO;
1986         } else {
1987                 dprintk("RPC:       xs_connect scheduled xprt %p\n", xprt);
1988                 queue_delayed_work(rpciod_workqueue,
1989                                    &transport->connect_worker, 0);
1990         }
1991 }
1992
1993 static void xs_tcp_connect(struct rpc_task *task)
1994 {
1995         struct rpc_xprt *xprt = task->tk_xprt;
1996
1997         /* Exit if we need to wait for socket shutdown to complete */
1998         if (test_bit(XPRT_CLOSING, &xprt->state))
1999                 return;
2000         xs_connect(task);
2001 }
2002
2003 /**
2004  * xs_udp_print_stats - display UDP socket-specifc stats
2005  * @xprt: rpc_xprt struct containing statistics
2006  * @seq: output file
2007  *
2008  */
2009 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2010 {
2011         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2012
2013         seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %Lu %Lu\n",
2014                         transport->port,
2015                         xprt->stat.bind_count,
2016                         xprt->stat.sends,
2017                         xprt->stat.recvs,
2018                         xprt->stat.bad_xids,
2019                         xprt->stat.req_u,
2020                         xprt->stat.bklog_u);
2021 }
2022
2023 /**
2024  * xs_tcp_print_stats - display TCP socket-specifc stats
2025  * @xprt: rpc_xprt struct containing statistics
2026  * @seq: output file
2027  *
2028  */
2029 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2030 {
2031         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2032         long idle_time = 0;
2033
2034         if (xprt_connected(xprt))
2035                 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2036
2037         seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu %Lu %Lu\n",
2038                         transport->port,
2039                         xprt->stat.bind_count,
2040                         xprt->stat.connect_count,
2041                         xprt->stat.connect_time,
2042                         idle_time,
2043                         xprt->stat.sends,
2044                         xprt->stat.recvs,
2045                         xprt->stat.bad_xids,
2046                         xprt->stat.req_u,
2047                         xprt->stat.bklog_u);
2048 }
2049
2050 static struct rpc_xprt_ops xs_udp_ops = {
2051         .set_buffer_size        = xs_udp_set_buffer_size,
2052         .reserve_xprt           = xprt_reserve_xprt_cong,
2053         .release_xprt           = xprt_release_xprt_cong,
2054         .rpcbind                = rpcb_getport_async,
2055         .set_port               = xs_set_port,
2056         .connect                = xs_connect,
2057         .buf_alloc              = rpc_malloc,
2058         .buf_free               = rpc_free,
2059         .send_request           = xs_udp_send_request,
2060         .set_retrans_timeout    = xprt_set_retrans_timeout_rtt,
2061         .timer                  = xs_udp_timer,
2062         .release_request        = xprt_release_rqst_cong,
2063         .close                  = xs_close,
2064         .destroy                = xs_destroy,
2065         .print_stats            = xs_udp_print_stats,
2066 };
2067
2068 static struct rpc_xprt_ops xs_tcp_ops = {
2069         .reserve_xprt           = xprt_reserve_xprt,
2070         .release_xprt           = xs_tcp_release_xprt,
2071         .rpcbind                = rpcb_getport_async,
2072         .set_port               = xs_set_port,
2073         .connect                = xs_tcp_connect,
2074         .buf_alloc              = rpc_malloc,
2075         .buf_free               = rpc_free,
2076         .send_request           = xs_tcp_send_request,
2077         .set_retrans_timeout    = xprt_set_retrans_timeout_def,
2078         .close                  = xs_tcp_close,
2079         .destroy                = xs_destroy,
2080         .print_stats            = xs_tcp_print_stats,
2081 };
2082
2083 static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
2084                                       unsigned int slot_table_size)
2085 {
2086         struct rpc_xprt *xprt;
2087         struct sock_xprt *new;
2088
2089         if (args->addrlen > sizeof(xprt->addr)) {
2090                 dprintk("RPC:       xs_setup_xprt: address too large\n");
2091                 return ERR_PTR(-EBADF);
2092         }
2093
2094         new = kzalloc(sizeof(*new), GFP_KERNEL);
2095         if (new == NULL) {
2096                 dprintk("RPC:       xs_setup_xprt: couldn't allocate "
2097                                 "rpc_xprt\n");
2098                 return ERR_PTR(-ENOMEM);
2099         }
2100         xprt = &new->xprt;
2101
2102         xprt->max_reqs = slot_table_size;
2103         xprt->slot = kcalloc(xprt->max_reqs, sizeof(struct rpc_rqst), GFP_KERNEL);
2104         if (xprt->slot == NULL) {
2105                 kfree(xprt);
2106                 dprintk("RPC:       xs_setup_xprt: couldn't allocate slot "
2107                                 "table\n");
2108                 return ERR_PTR(-ENOMEM);
2109         }
2110
2111         memcpy(&xprt->addr, args->dstaddr, args->addrlen);
2112         xprt->addrlen = args->addrlen;
2113         if (args->srcaddr)
2114                 memcpy(&new->addr, args->srcaddr, args->addrlen);
2115
2116         return xprt;
2117 }
2118
2119 static const struct rpc_timeout xs_udp_default_timeout = {
2120         .to_initval = 5 * HZ,
2121         .to_maxval = 30 * HZ,
2122         .to_increment = 5 * HZ,
2123         .to_retries = 5,
2124 };
2125
2126 /**
2127  * xs_setup_udp - Set up transport to use a UDP socket
2128  * @args: rpc transport creation arguments
2129  *
2130  */
2131 static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
2132 {
2133         struct sockaddr *addr = args->dstaddr;
2134         struct rpc_xprt *xprt;
2135         struct sock_xprt *transport;
2136
2137         xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries);
2138         if (IS_ERR(xprt))
2139                 return xprt;
2140         transport = container_of(xprt, struct sock_xprt, xprt);
2141
2142         xprt->prot = IPPROTO_UDP;
2143         xprt->tsh_size = 0;
2144         /* XXX: header size can vary due to auth type, IPv6, etc. */
2145         xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
2146
2147         xprt->bind_timeout = XS_BIND_TO;
2148         xprt->connect_timeout = XS_UDP_CONN_TO;
2149         xprt->reestablish_timeout = XS_UDP_REEST_TO;
2150         xprt->idle_timeout = XS_IDLE_DISC_TO;
2151
2152         xprt->ops = &xs_udp_ops;
2153
2154         xprt->timeout = &xs_udp_default_timeout;
2155
2156         switch (addr->sa_family) {
2157         case AF_INET:
2158                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2159                         xprt_set_bound(xprt);
2160
2161                 INIT_DELAYED_WORK(&transport->connect_worker,
2162                                         xs_udp_connect_worker4);
2163                 xs_format_ipv4_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
2164                 break;
2165         case AF_INET6:
2166                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2167                         xprt_set_bound(xprt);
2168
2169                 INIT_DELAYED_WORK(&transport->connect_worker,
2170                                         xs_udp_connect_worker6);
2171                 xs_format_ipv6_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
2172                 break;
2173         default:
2174                 kfree(xprt);
2175                 return ERR_PTR(-EAFNOSUPPORT);
2176         }
2177
2178         dprintk("RPC:       set up transport to address %s\n",
2179                         xprt->address_strings[RPC_DISPLAY_ALL]);
2180
2181         if (try_module_get(THIS_MODULE))
2182                 return xprt;
2183
2184         kfree(xprt->slot);
2185         kfree(xprt);
2186         return ERR_PTR(-EINVAL);
2187 }
2188
2189 static const struct rpc_timeout xs_tcp_default_timeout = {
2190         .to_initval = 60 * HZ,
2191         .to_maxval = 60 * HZ,
2192         .to_retries = 2,
2193 };
2194
2195 /**
2196  * xs_setup_tcp - Set up transport to use a TCP socket
2197  * @args: rpc transport creation arguments
2198  *
2199  */
2200 static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
2201 {
2202         struct sockaddr *addr = args->dstaddr;
2203         struct rpc_xprt *xprt;
2204         struct sock_xprt *transport;
2205
2206         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries);
2207         if (IS_ERR(xprt))
2208                 return xprt;
2209         transport = container_of(xprt, struct sock_xprt, xprt);
2210
2211         xprt->prot = IPPROTO_TCP;
2212         xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2213         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2214
2215         xprt->bind_timeout = XS_BIND_TO;
2216         xprt->connect_timeout = XS_TCP_CONN_TO;
2217         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2218         xprt->idle_timeout = XS_IDLE_DISC_TO;
2219
2220         xprt->ops = &xs_tcp_ops;
2221         xprt->timeout = &xs_tcp_default_timeout;
2222
2223         switch (addr->sa_family) {
2224         case AF_INET:
2225                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2226                         xprt_set_bound(xprt);
2227
2228                 INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_connect_worker4);
2229                 xs_format_ipv4_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
2230                 break;
2231         case AF_INET6:
2232                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2233                         xprt_set_bound(xprt);
2234
2235                 INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_connect_worker6);
2236                 xs_format_ipv6_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
2237                 break;
2238         default:
2239                 kfree(xprt);
2240                 return ERR_PTR(-EAFNOSUPPORT);
2241         }
2242
2243         dprintk("RPC:       set up transport to address %s\n",
2244                         xprt->address_strings[RPC_DISPLAY_ALL]);
2245
2246         if (try_module_get(THIS_MODULE))
2247                 return xprt;
2248
2249         kfree(xprt->slot);
2250         kfree(xprt);
2251         return ERR_PTR(-EINVAL);
2252 }
2253
2254 static struct xprt_class        xs_udp_transport = {
2255         .list           = LIST_HEAD_INIT(xs_udp_transport.list),
2256         .name           = "udp",
2257         .owner          = THIS_MODULE,
2258         .ident          = IPPROTO_UDP,
2259         .setup          = xs_setup_udp,
2260 };
2261
2262 static struct xprt_class        xs_tcp_transport = {
2263         .list           = LIST_HEAD_INIT(xs_tcp_transport.list),
2264         .name           = "tcp",
2265         .owner          = THIS_MODULE,
2266         .ident          = IPPROTO_TCP,
2267         .setup          = xs_setup_tcp,
2268 };
2269
2270 /**
2271  * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
2272  *
2273  */
2274 int init_socket_xprt(void)
2275 {
2276 #ifdef RPC_DEBUG
2277         if (!sunrpc_table_header)
2278                 sunrpc_table_header = register_sysctl_table(sunrpc_table);
2279 #endif
2280
2281         xprt_register_transport(&xs_udp_transport);
2282         xprt_register_transport(&xs_tcp_transport);
2283
2284         return 0;
2285 }
2286
2287 /**
2288  * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
2289  *
2290  */
2291 void cleanup_socket_xprt(void)
2292 {
2293 #ifdef RPC_DEBUG
2294         if (sunrpc_table_header) {
2295                 unregister_sysctl_table(sunrpc_table_header);
2296                 sunrpc_table_header = NULL;
2297         }
2298 #endif
2299
2300         xprt_unregister_transport(&xs_udp_transport);
2301         xprt_unregister_transport(&xs_tcp_transport);
2302 }