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 #ifdef CONFIG_NFS_V4_1
  38 #include <linux/sunrpc/bc_xprt.h>
  39 #endif
  40
  41 #include <net/sock.h>
  42 #include <net/checksum.h>
  43 #include <net/udp.h>
  44 #include <net/tcp.h>
  45
  46 /*
  47  * xprtsock tunables
  48  */
  49 unsigned int xprt_udp_slot_table_entries = RPC_DEF_SLOT_TABLE;
  50 unsigned int xprt_tcp_slot_table_entries = RPC_DEF_SLOT_TABLE;
  51
  52 unsigned int xprt_min_resvport = RPC_DEF_MIN_RESVPORT;
  53 unsigned int xprt_max_resvport = RPC_DEF_MAX_RESVPORT;
  54
  55 #define XS_TCP_LINGER_TO        (15U * HZ)
  56 static unsigned int xs_tcp_fin_timeout __read_mostly = XS_TCP_LINGER_TO;
  57
  58 /*
  59  * We can register our own files under /proc/sys/sunrpc by
  60  * calling register_sysctl_table() again.  The files in that
  61  * directory become the union of all files registered there.
  62  *
  63  * We simply need to make sure that we don't collide with
  64  * someone else's file names!
  65  */
  66
  67 #ifdef RPC_DEBUG
  68
  69 static unsigned int min_slot_table_size = RPC_MIN_SLOT_TABLE;
  70 static unsigned int max_slot_table_size = RPC_MAX_SLOT_TABLE;
  71 static unsigned int xprt_min_resvport_limit = RPC_MIN_RESVPORT;
  72 static unsigned int xprt_max_resvport_limit = RPC_MAX_RESVPORT;
  73
  74 static struct ctl_table_header *sunrpc_table_header;
  75
  76 /*
  77  * FIXME: changing the UDP slot table size should also resize the UDP
  78  *        socket buffers for existing UDP transports
  79  */
  80 static ctl_table xs_tunables_table[] = {
  81         {
  82                 .ctl_name       = CTL_SLOTTABLE_UDP,
  83                 .procname       = "udp_slot_table_entries",
  84                 .data           = &xprt_udp_slot_table_entries,
  85                 .maxlen         = sizeof(unsigned int),
  86                 .mode           = 0644,
  87                 .proc_handler   = &proc_dointvec_minmax,
  88                 .strategy       = &sysctl_intvec,
  89                 .extra1         = &min_slot_table_size,
  90                 .extra2         = &max_slot_table_size
  91         },
  92         {
  93                 .ctl_name       = CTL_SLOTTABLE_TCP,
  94                 .procname       = "tcp_slot_table_entries",
  95                 .data           = &xprt_tcp_slot_table_entries,
  96                 .maxlen         = sizeof(unsigned int),
  97                 .mode           = 0644,
  98                 .proc_handler   = &proc_dointvec_minmax,
  99                 .strategy       = &sysctl_intvec,
 100                 .extra1         = &min_slot_table_size,
 101                 .extra2         = &max_slot_table_size
 102         },
 103         {
 104                 .ctl_name       = CTL_MIN_RESVPORT,
 105                 .procname       = "min_resvport",
 106                 .data           = &xprt_min_resvport,
 107                 .maxlen         = sizeof(unsigned int),
 108                 .mode           = 0644,
 109                 .proc_handler   = &proc_dointvec_minmax,
 110                 .strategy       = &sysctl_intvec,
 111                 .extra1         = &xprt_min_resvport_limit,
 112                 .extra2         = &xprt_max_resvport_limit
 113         },
 114         {
 115                 .ctl_name       = CTL_MAX_RESVPORT,
 116                 .procname       = "max_resvport",
 117                 .data           = &xprt_max_resvport,
 118                 .maxlen         = sizeof(unsigned int),
 119                 .mode           = 0644,
 120                 .proc_handler   = &proc_dointvec_minmax,
 121                 .strategy       = &sysctl_intvec,
 122                 .extra1         = &xprt_min_resvport_limit,
 123                 .extra2         = &xprt_max_resvport_limit
 124         },
 125         {
 126                 .procname       = "tcp_fin_timeout",
 127                 .data           = &xs_tcp_fin_timeout,
 128                 .maxlen         = sizeof(xs_tcp_fin_timeout),
 129                 .mode           = 0644,
 130                 .proc_handler   = &proc_dointvec_jiffies,
 131                 .strategy       = sysctl_jiffies
 132         },
 133         {
 134                 .ctl_name = 0,
 135         },
 136 };
 137
 138 static ctl_table sunrpc_table[] = {
 139         {
 140                 .ctl_name       = CTL_SUNRPC,
 141                 .procname       = "sunrpc",
 142                 .mode           = 0555,
 143                 .child          = xs_tunables_table
 144         },
 145         {
 146                 .ctl_name = 0,
 147         },
 148 };
 149
 150 #endif
 151
 152 /*
 153  * Time out for an RPC UDP socket connect.  UDP socket connects are
 154  * synchronous, but we set a timeout anyway in case of resource
 155  * exhaustion on the local host.
 156  */
 157 #define XS_UDP_CONN_TO          (5U * HZ)
 158
 159 /*
 160  * Wait duration for an RPC TCP connection to be established.  Solaris
 161  * NFS over TCP uses 60 seconds, for example, which is in line with how
 162  * long a server takes to reboot.
 163  */
 164 #define XS_TCP_CONN_TO          (60U * HZ)
 165
 166 /*
 167  * Wait duration for a reply from the RPC portmapper.
 168  */
 169 #define XS_BIND_TO              (60U * HZ)
 170
 171 /*
 172  * Delay if a UDP socket connect error occurs.  This is most likely some
 173  * kind of resource problem on the local host.
 174  */
 175 #define XS_UDP_REEST_TO         (2U * HZ)
 176
 177 /*
 178  * The reestablish timeout allows clients to delay for a bit before attempting
 179  * to reconnect to a server that just dropped our connection.
 180  *
 181  * We implement an exponential backoff when trying to reestablish a TCP
 182  * transport connection with the server.  Some servers like to drop a TCP
 183  * connection when they are overworked, so we start with a short timeout and
 184  * increase over time if the server is down or not responding.
 185  */
 186 #define XS_TCP_INIT_REEST_TO    (3U * HZ)
 187 #define XS_TCP_MAX_REEST_TO     (5U * 60 * HZ)
 188
 189 /*
 190  * TCP idle timeout; client drops the transport socket if it is idle
 191  * for this long.  Note that we also timeout UDP sockets to prevent
 192  * holding port numbers when there is no RPC traffic.
 193  */
 194 #define XS_IDLE_DISC_TO         (5U * 60 * HZ)
 195
 196 #ifdef RPC_DEBUG
 197 # undef  RPC_DEBUG_DATA
 198 # define RPCDBG_FACILITY        RPCDBG_TRANS
 199 #endif
 200
 201 #ifdef RPC_DEBUG_DATA
 202 static void xs_pktdump(char *msg, u32 *packet, unsigned int count)
 203 {
 204         u8 *buf = (u8 *) packet;
 205         int j;
 206
 207         dprintk("RPC:       %s\n", msg);
 208         for (j = 0; j < count && j < 128; j += 4) {
 209                 if (!(j & 31)) {
 210                         if (j)
 211                                 dprintk("\n");
 212                         dprintk("0x%04x ", j);
 213                 }
 214                 dprintk("%02x%02x%02x%02x ",
 215                         buf[j], buf[j+1], buf[j+2], buf[j+3]);
 216         }
 217         dprintk("\n");
 218 }
 219 #else
 220 static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count)
 221 {
 222         /* NOP */
 223 }
 224 #endif
 225
 226 struct sock_xprt {
 227         struct rpc_xprt         xprt;
 228
 229         /*
 230          * Network layer
 231          */
 232         struct socket *         sock;
 233         struct sock *           inet;
 234
 235         /*
 236          * State of TCP reply receive
 237          */
 238         __be32                  tcp_fraghdr,
 239                                 tcp_xid;
 240
 241         u32                     tcp_offset,
 242                                 tcp_reclen;
 243
 244         unsigned long           tcp_copied,
 245                                 tcp_flags;
 246
 247         /*
 248          * Connection of transports
 249          */
 250         struct delayed_work     connect_worker;
 251         struct sockaddr_storage addr;
 252         unsigned short          port;
 253
 254         /*
 255          * UDP socket buffer size parameters
 256          */
 257         size_t                  rcvsize,
 258                                 sndsize;
 259
 260         /*
 261          * Saved socket callback addresses
 262          */
 263         void                    (*old_data_ready)(struct sock *, int);
 264         void                    (*old_state_change)(struct sock *);
 265         void                    (*old_write_space)(struct sock *);
 266         void                    (*old_error_report)(struct sock *);
 267 };
 268
 269 /*
 270  * TCP receive state flags
 271  */
 272 #define TCP_RCV_LAST_FRAG       (1UL << 0)
 273 #define TCP_RCV_COPY_FRAGHDR    (1UL << 1)
 274 #define TCP_RCV_COPY_XID        (1UL << 2)
 275 #define TCP_RCV_COPY_DATA       (1UL << 3)
 276 #define TCP_RCV_READ_CALLDIR    (1UL << 4)
 277 #define TCP_RCV_COPY_CALLDIR    (1UL << 5)
 278
 279 /*
 280  * TCP RPC flags
 281  */
 282 #define TCP_RPC_REPLY           (1UL << 6)
 283
 284 static inline struct sockaddr *xs_addr(struct rpc_xprt *xprt)
 285 {
 286         return (struct sockaddr *) &xprt->addr;
 287 }
 288
 289 static inline struct sockaddr_in *xs_addr_in(struct rpc_xprt *xprt)
 290 {
 291         return (struct sockaddr_in *) &xprt->addr;
 292 }
 293
 294 static inline struct sockaddr_in6 *xs_addr_in6(struct rpc_xprt *xprt)
 295 {
 296         return (struct sockaddr_in6 *) &xprt->addr;
 297 }
 298
 299 static void xs_format_ipv4_peer_addresses(struct rpc_xprt *xprt,
 300                                           const char *protocol,
 301                                           const char *netid)
 302 {
 303         struct sockaddr_in *addr = xs_addr_in(xprt);
 304         char *buf;
 305
 306         buf = kzalloc(20, GFP_KERNEL);
 307         if (buf) {
 308                 snprintf(buf, 20, "%pI4", &addr->sin_addr.s_addr);
 309         }
 310         xprt->address_strings[RPC_DISPLAY_ADDR] = buf;
 311
 312         buf = kzalloc(8, GFP_KERNEL);
 313         if (buf) {
 314                 snprintf(buf, 8, "%u",
 315                                 ntohs(addr->sin_port));
 316         }
 317         xprt->address_strings[RPC_DISPLAY_PORT] = buf;
 318
 319         xprt->address_strings[RPC_DISPLAY_PROTO] = protocol;
 320
 321         buf = kzalloc(48, GFP_KERNEL);
 322         if (buf) {
 323                 snprintf(buf, 48, "addr=%pI4 port=%u proto=%s",
 324                         &addr->sin_addr.s_addr,
 325                         ntohs(addr->sin_port),
 326                         protocol);
 327         }
 328         xprt->address_strings[RPC_DISPLAY_ALL] = buf;
 329
 330         buf = kzalloc(10, GFP_KERNEL);
 331         if (buf) {
 332                 snprintf(buf, 10, "%02x%02x%02x%02x",
 333                                 NIPQUAD(addr->sin_addr.s_addr));
 334         }
 335         xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = buf;
 336
 337         buf = kzalloc(8, GFP_KERNEL);
 338         if (buf) {
 339                 snprintf(buf, 8, "%4hx",
 340                                 ntohs(addr->sin_port));
 341         }
 342         xprt->address_strings[RPC_DISPLAY_HEX_PORT] = buf;
 343
 344         xprt->address_strings[RPC_DISPLAY_NETID] = netid;
 345 }
 346
 347 static void xs_format_ipv6_peer_addresses(struct rpc_xprt *xprt,
 348                                           const char *protocol,
 349                                           const char *netid)
 350 {
 351         struct sockaddr_in6 *addr = xs_addr_in6(xprt);
 352         char *buf;
 353
 354         buf = kzalloc(40, GFP_KERNEL);
 355         if (buf) {
 356                 snprintf(buf, 40, "%pI6",&addr->sin6_addr);
 357         }
 358         xprt->address_strings[RPC_DISPLAY_ADDR] = buf;
 359
 360         buf = kzalloc(8, GFP_KERNEL);
 361         if (buf) {
 362                 snprintf(buf, 8, "%u",
 363                                 ntohs(addr->sin6_port));
 364         }
 365         xprt->address_strings[RPC_DISPLAY_PORT] = buf;
 366
 367         xprt->address_strings[RPC_DISPLAY_PROTO] = protocol;
 368
 369         buf = kzalloc(64, GFP_KERNEL);
 370         if (buf) {
 371                 snprintf(buf, 64, "addr=%pI6 port=%u proto=%s",
 372                                 &addr->sin6_addr,
 373                                 ntohs(addr->sin6_port),
 374                                 protocol);
 375         }
 376         xprt->address_strings[RPC_DISPLAY_ALL] = buf;
 377
 378         buf = kzalloc(36, GFP_KERNEL);
 379         if (buf)
 380                 snprintf(buf, 36, "%pi6", &addr->sin6_addr);
 381
 382         xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = buf;
 383
 384         buf = kzalloc(8, GFP_KERNEL);
 385         if (buf) {
 386                 snprintf(buf, 8, "%4hx",
 387                                 ntohs(addr->sin6_port));
 388         }
 389         xprt->address_strings[RPC_DISPLAY_HEX_PORT] = buf;
 390
 391         xprt->address_strings[RPC_DISPLAY_NETID] = netid;
 392 }
 393
 394 static void xs_free_peer_addresses(struct rpc_xprt *xprt)
 395 {
 396         unsigned int i;
 397
 398         for (i = 0; i < RPC_DISPLAY_MAX; i++)
 399                 switch (i) {
 400                 case RPC_DISPLAY_PROTO:
 401                 case RPC_DISPLAY_NETID:
 402                         continue;
 403                 default:
 404                         kfree(xprt->address_strings[i]);
 405                 }
 406 }
 407
 408 #define XS_SENDMSG_FLAGS        (MSG_DONTWAIT | MSG_NOSIGNAL)
 409
 410 static int xs_send_kvec(struct socket *sock, struct sockaddr *addr, int addrlen, struct kvec *vec, unsigned int base, int more)
 411 {
 412         struct msghdr msg = {
 413                 .msg_name       = addr,
 414                 .msg_namelen    = addrlen,
 415                 .msg_flags      = XS_SENDMSG_FLAGS | (more ? MSG_MORE : 0),
 416         };
 417         struct kvec iov = {
 418                 .iov_base       = vec->iov_base + base,
 419                 .iov_len        = vec->iov_len - base,
 420         };
 421
 422         if (iov.iov_len != 0)
 423                 return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
 424         return kernel_sendmsg(sock, &msg, NULL, 0, 0);
 425 }
 426
 427 static int xs_send_pagedata(struct socket *sock, struct xdr_buf *xdr, unsigned int base, int more)
 428 {
 429         struct page **ppage;
 430         unsigned int remainder;
 431         int err, sent = 0;
 432
 433         remainder = xdr->page_len - base;
 434         base += xdr->page_base;
 435         ppage = xdr->pages + (base >> PAGE_SHIFT);
 436         base &= ~PAGE_MASK;
 437         for(;;) {
 438                 unsigned int len = min_t(unsigned int, PAGE_SIZE - base, remainder);
 439                 int flags = XS_SENDMSG_FLAGS;
 440
 441                 remainder -= len;
 442                 if (remainder != 0 || more)
 443                         flags |= MSG_MORE;
 444                 err = sock->ops->sendpage(sock, *ppage, base, len, flags);
 445                 if (remainder == 0 || err != len)
 446                         break;
 447                 sent += err;
 448                 ppage++;
 449                 base = 0;
 450         }
 451         if (sent == 0)
 452                 return err;
 453         if (err > 0)
 454                 sent += err;
 455         return sent;
 456 }
 457
 458 /**
 459  * xs_sendpages - write pages directly to a socket
 460  * @sock: socket to send on
 461  * @addr: UDP only -- address of destination
 462  * @addrlen: UDP only -- length of destination address
 463  * @xdr: buffer containing this request
 464  * @base: starting position in the buffer
 465  *
 466  */
 467 static int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base)
 468 {
 469         unsigned int remainder = xdr->len - base;
 470         int err, sent = 0;
 471
 472         if (unlikely(!sock))
 473                 return -ENOTSOCK;
 474
 475         clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags);
 476         if (base != 0) {
 477                 addr = NULL;
 478                 addrlen = 0;
 479         }
 480
 481         if (base < xdr->head[0].iov_len || addr != NULL) {
 482                 unsigned int len = xdr->head[0].iov_len - base;
 483                 remainder -= len;
 484                 err = xs_send_kvec(sock, addr, addrlen, &xdr->head[0], base, remainder != 0);
 485                 if (remainder == 0 || err != len)
 486                         goto out;
 487                 sent += err;
 488                 base = 0;
 489         } else
 490                 base -= xdr->head[0].iov_len;
 491
 492         if (base < xdr->page_len) {
 493                 unsigned int len = xdr->page_len - base;
 494                 remainder -= len;
 495                 err = xs_send_pagedata(sock, xdr, base, remainder != 0);
 496                 if (remainder == 0 || err != len)
 497                         goto out;
 498                 sent += err;
 499                 base = 0;
 500         } else
 501                 base -= xdr->page_len;
 502
 503         if (base >= xdr->tail[0].iov_len)
 504                 return sent;
 505         err = xs_send_kvec(sock, NULL, 0, &xdr->tail[0], base, 0);
 506 out:
 507         if (sent == 0)
 508                 return err;
 509         if (err > 0)
 510                 sent += err;
 511         return sent;
 512 }
 513
 514 static void xs_nospace_callback(struct rpc_task *task)
 515 {
 516         struct sock_xprt *transport = container_of(task->tk_rqstp->rq_xprt, struct sock_xprt, xprt);
 517
 518         transport->inet->sk_write_pending--;
 519         clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 520 }
 521
 522 /**
 523  * xs_nospace - place task on wait queue if transmit was incomplete
 524  * @task: task to put to sleep
 525  *
 526  */
 527 static int xs_nospace(struct rpc_task *task)
 528 {
 529         struct rpc_rqst *req = task->tk_rqstp;
 530         struct rpc_xprt *xprt = req->rq_xprt;
 531         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 532         int ret = 0;
 533
 534         dprintk("RPC: %5u xmit incomplete (%u left of %u)\n",
 535                         task->tk_pid, req->rq_slen - req->rq_bytes_sent,
 536                         req->rq_slen);
 537
 538         /* Protect against races with write_space */
 539         spin_lock_bh(&xprt->transport_lock);
 540
 541         /* Don't race with disconnect */
 542         if (xprt_connected(xprt)) {
 543                 if (test_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags)) {
 544                         ret = -EAGAIN;
 545                         /*
 546                          * Notify TCP that we're limited by the application
 547                          * window size
 548                          */
 549                         set_bit(SOCK_NOSPACE, &transport->sock->flags);
 550                         transport->inet->sk_write_pending++;
 551                         /* ...and wait for more buffer space */
 552                         xprt_wait_for_buffer_space(task, xs_nospace_callback);
 553                 }
 554         } else {
 555                 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 556                 ret = -ENOTCONN;
 557         }
 558
 559         spin_unlock_bh(&xprt->transport_lock);
 560         return ret;
 561 }
 562
 563 /**
 564  * xs_udp_send_request - write an RPC request to a UDP socket
 565  * @task: address of RPC task that manages the state of an RPC request
 566  *
 567  * Return values:
 568  *        0:    The request has been sent
 569  *   EAGAIN:    The socket was blocked, please call again later to
 570  *              complete the request
 571  * ENOTCONN:    Caller needs to invoke connect logic then call again
 572  *    other:    Some other error occured, the request was not sent
 573  */
 574 static int xs_udp_send_request(struct rpc_task *task)
 575 {
 576         struct rpc_rqst *req = task->tk_rqstp;
 577         struct rpc_xprt *xprt = req->rq_xprt;
 578         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 579         struct xdr_buf *xdr = &req->rq_snd_buf;
 580         int status;
 581
 582         xs_pktdump("packet data:",
 583                                 req->rq_svec->iov_base,
 584                                 req->rq_svec->iov_len);
 585
 586         if (!xprt_bound(xprt))
 587                 return -ENOTCONN;
 588         status = xs_sendpages(transport->sock,
 589                               xs_addr(xprt),
 590                               xprt->addrlen, xdr,
 591                               req->rq_bytes_sent);
 592
 593         dprintk("RPC:       xs_udp_send_request(%u) = %d\n",
 594                         xdr->len - req->rq_bytes_sent, status);
 595
 596         if (status >= 0) {
 597                 task->tk_bytes_sent += status;
 598                 if (status >= req->rq_slen)
 599                         return 0;
 600                 /* Still some bytes left; set up for a retry later. */
 601                 status = -EAGAIN;
 602         }
 603         if (!transport->sock)
 604                 goto out;
 605
 606         switch (status) {
 607         case -ENOTSOCK:
 608                 status = -ENOTCONN;
 609                 /* Should we call xs_close() here? */
 610                 break;
 611         case -EAGAIN:
 612                 status = xs_nospace(task);
 613                 break;
 614         default:
 615                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
 616                         -status);
 617         case -ENETUNREACH:
 618         case -EPIPE:
 619         case -ECONNREFUSED:
 620                 /* When the server has died, an ICMP port unreachable message
 621                  * prompts ECONNREFUSED. */
 622                 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 623         }
 624 out:
 625         return status;
 626 }
 627
 628 /**
 629  * xs_tcp_shutdown - gracefully shut down a TCP socket
 630  * @xprt: transport
 631  *
 632  * Initiates a graceful shutdown of the TCP socket by calling the
 633  * equivalent of shutdown(SHUT_WR);
 634  */
 635 static void xs_tcp_shutdown(struct rpc_xprt *xprt)
 636 {
 637         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 638         struct socket *sock = transport->sock;
 639
 640         if (sock != NULL)
 641                 kernel_sock_shutdown(sock, SHUT_WR);
 642 }
 643
 644 static inline void xs_encode_tcp_record_marker(struct xdr_buf *buf)
 645 {
 646         u32 reclen = buf->len - sizeof(rpc_fraghdr);
 647         rpc_fraghdr *base = buf->head[0].iov_base;
 648         *base = htonl(RPC_LAST_STREAM_FRAGMENT | reclen);
 649 }
 650
 651 /**
 652  * xs_tcp_send_request - write an RPC request to a TCP socket
 653  * @task: address of RPC task that manages the state of an RPC request
 654  *
 655  * Return values:
 656  *        0:    The request has been sent
 657  *   EAGAIN:    The socket was blocked, please call again later to
 658  *              complete the request
 659  * ENOTCONN:    Caller needs to invoke connect logic then call again
 660  *    other:    Some other error occured, the request was not sent
 661  *
 662  * XXX: In the case of soft timeouts, should we eventually give up
 663  *      if sendmsg is not able to make progress?
 664  */
 665 static int xs_tcp_send_request(struct rpc_task *task)
 666 {
 667         struct rpc_rqst *req = task->tk_rqstp;
 668         struct rpc_xprt *xprt = req->rq_xprt;
 669         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 670         struct xdr_buf *xdr = &req->rq_snd_buf;
 671         int status;
 672
 673         xs_encode_tcp_record_marker(&req->rq_snd_buf);
 674
 675         xs_pktdump("packet data:",
 676                                 req->rq_svec->iov_base,
 677                                 req->rq_svec->iov_len);
 678
 679         /* Continue transmitting the packet/record. We must be careful
 680          * to cope with writespace callbacks arriving _after_ we have
 681          * called sendmsg(). */
 682         while (1) {
 683                 status = xs_sendpages(transport->sock,
 684                                         NULL, 0, xdr, req->rq_bytes_sent);
 685
 686                 dprintk("RPC:       xs_tcp_send_request(%u) = %d\n",
 687                                 xdr->len - req->rq_bytes_sent, status);
 688
 689                 if (unlikely(status < 0))
 690                         break;
 691
 692                 /* If we've sent the entire packet, immediately
 693                  * reset the count of bytes sent. */
 694                 req->rq_bytes_sent += status;
 695                 task->tk_bytes_sent += status;
 696                 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
 697                         req->rq_bytes_sent = 0;
 698                         return 0;
 699                 }
 700
 701                 if (status != 0)
 702                         continue;
 703                 status = -EAGAIN;
 704                 break;
 705         }
 706         if (!transport->sock)
 707                 goto out;
 708
 709         switch (status) {
 710         case -ENOTSOCK:
 711                 status = -ENOTCONN;
 712                 /* Should we call xs_close() here? */
 713                 break;
 714         case -EAGAIN:
 715                 status = xs_nospace(task);
 716                 break;
 717         default:
 718                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
 719                         -status);
 720         case -ECONNRESET:
 721         case -EPIPE:
 722                 xs_tcp_shutdown(xprt);
 723         case -ECONNREFUSED:
 724         case -ENOTCONN:
 725                 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 726         }
 727 out:
 728         return status;
 729 }
 730
 731 /**
 732  * xs_tcp_release_xprt - clean up after a tcp transmission
 733  * @xprt: transport
 734  * @task: rpc task
 735  *
 736  * This cleans up if an error causes us to abort the transmission of a request.
 737  * In this case, the socket may need to be reset in order to avoid confusing
 738  * the server.
 739  */
 740 static void xs_tcp_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
 741 {
 742         struct rpc_rqst *req;
 743
 744         if (task != xprt->snd_task)
 745                 return;
 746         if (task == NULL)
 747                 goto out_release;
 748         req = task->tk_rqstp;
 749         if (req->rq_bytes_sent == 0)
 750                 goto out_release;
 751         if (req->rq_bytes_sent == req->rq_snd_buf.len)
 752                 goto out_release;
 753         set_bit(XPRT_CLOSE_WAIT, &task->tk_xprt->state);
 754 out_release:
 755         xprt_release_xprt(xprt, task);
 756 }
 757
 758 static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk)
 759 {
 760         transport->old_data_ready = sk->sk_data_ready;
 761         transport->old_state_change = sk->sk_state_change;
 762         transport->old_write_space = sk->sk_write_space;
 763         transport->old_error_report = sk->sk_error_report;
 764 }
 765
 766 static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk)
 767 {
 768         sk->sk_data_ready = transport->old_data_ready;
 769         sk->sk_state_change = transport->old_state_change;
 770         sk->sk_write_space = transport->old_write_space;
 771         sk->sk_error_report = transport->old_error_report;
 772 }
 773
 774 static void xs_reset_transport(struct sock_xprt *transport)
 775 {
 776         struct socket *sock = transport->sock;
 777         struct sock *sk = transport->inet;
 778
 779         if (sk == NULL)
 780                 return;
 781
 782         write_lock_bh(&sk->sk_callback_lock);
 783         transport->inet = NULL;
 784         transport->sock = NULL;
 785
 786         sk->sk_user_data = NULL;
 787
 788         xs_restore_old_callbacks(transport, sk);
 789         write_unlock_bh(&sk->sk_callback_lock);
 790
 791         sk->sk_no_check = 0;
 792
 793         sock_release(sock);
 794 }
 795
 796 /**
 797  * xs_close - close a socket
 798  * @xprt: transport
 799  *
 800  * This is used when all requests are complete; ie, no DRC state remains
 801  * on the server we want to save.
 802  *
 803  * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with
 804  * xs_reset_transport() zeroing the socket from underneath a writer.
 805  */
 806 static void xs_close(struct rpc_xprt *xprt)
 807 {
 808         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 809
 810         dprintk("RPC:       xs_close xprt %p\n", xprt);
 811
 812         xs_reset_transport(transport);
 813
 814         smp_mb__before_clear_bit();
 815         clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
 816         clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
 817         clear_bit(XPRT_CLOSING, &xprt->state);
 818         smp_mb__after_clear_bit();
 819         xprt_disconnect_done(xprt);
 820 }
 821
 822 static void xs_tcp_close(struct rpc_xprt *xprt)
 823 {
 824         if (test_and_clear_bit(XPRT_CONNECTION_CLOSE, &xprt->state))
 825                 xs_close(xprt);
 826         else
 827                 xs_tcp_shutdown(xprt);
 828 }
 829
 830 /**
 831  * xs_destroy - prepare to shutdown a transport
 832  * @xprt: doomed transport
 833  *
 834  */
 835 static void xs_destroy(struct rpc_xprt *xprt)
 836 {
 837         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 838
 839         dprintk("RPC:       xs_destroy xprt %p\n", xprt);
 840
 841         cancel_rearming_delayed_work(&transport->connect_worker);
 842
 843         xs_close(xprt);
 844         xs_free_peer_addresses(xprt);
 845         kfree(xprt->slot);
 846         kfree(xprt);
 847         module_put(THIS_MODULE);
 848 }
 849
 850 static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
 851 {
 852         return (struct rpc_xprt *) sk->sk_user_data;
 853 }
 854
 855 /**
 856  * xs_udp_data_ready - "data ready" callback for UDP sockets
 857  * @sk: socket with data to read
 858  * @len: how much data to read
 859  *
 860  */
 861 static void xs_udp_data_ready(struct sock *sk, int len)
 862 {
 863         struct rpc_task *task;
 864         struct rpc_xprt *xprt;
 865         struct rpc_rqst *rovr;
 866         struct sk_buff *skb;
 867         int err, repsize, copied;
 868         u32 _xid;
 869         __be32 *xp;
 870
 871         read_lock(&sk->sk_callback_lock);
 872         dprintk("RPC:       xs_udp_data_ready...\n");
 873         if (!(xprt = xprt_from_sock(sk)))
 874                 goto out;
 875
 876         if ((skb = skb_recv_datagram(sk, 0, 1, &err)) == NULL)
 877                 goto out;
 878
 879         if (xprt->shutdown)
 880                 goto dropit;
 881
 882         repsize = skb->len - sizeof(struct udphdr);
 883         if (repsize < 4) {
 884                 dprintk("RPC:       impossible RPC reply size %d!\n", repsize);
 885                 goto dropit;
 886         }
 887
 888         /* Copy the XID from the skb... */
 889         xp = skb_header_pointer(skb, sizeof(struct udphdr),
 890                                 sizeof(_xid), &_xid);
 891         if (xp == NULL)
 892                 goto dropit;
 893
 894         /* Look up and lock the request corresponding to the given XID */
 895         spin_lock(&xprt->transport_lock);
 896         rovr = xprt_lookup_rqst(xprt, *xp);
 897         if (!rovr)
 898                 goto out_unlock;
 899         task = rovr->rq_task;
 900
 901         if ((copied = rovr->rq_private_buf.buflen) > repsize)
 902                 copied = repsize;
 903
 904         /* Suck it into the iovec, verify checksum if not done by hw. */
 905         if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
 906                 UDPX_INC_STATS_BH(sk, UDP_MIB_INERRORS);
 907                 goto out_unlock;
 908         }
 909
 910         UDPX_INC_STATS_BH(sk, UDP_MIB_INDATAGRAMS);
 911
 912         /* Something worked... */
 913         dst_confirm(skb_dst(skb));
 914
 915         xprt_adjust_cwnd(task, copied);
 916         xprt_update_rtt(task);
 917         xprt_complete_rqst(task, copied);
 918
 919  out_unlock:
 920         spin_unlock(&xprt->transport_lock);
 921  dropit:
 922         skb_free_datagram(sk, skb);
 923  out:
 924         read_unlock(&sk->sk_callback_lock);
 925 }
 926
 927 static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
 928 {
 929         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 930         size_t len, used;
 931         char *p;
 932
 933         p = ((char *) &transport->tcp_fraghdr) + transport->tcp_offset;
 934         len = sizeof(transport->tcp_fraghdr) - transport->tcp_offset;
 935         used = xdr_skb_read_bits(desc, p, len);
 936         transport->tcp_offset += used;
 937         if (used != len)
 938                 return;
 939
 940         transport->tcp_reclen = ntohl(transport->tcp_fraghdr);
 941         if (transport->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
 942                 transport->tcp_flags |= TCP_RCV_LAST_FRAG;
 943         else
 944                 transport->tcp_flags &= ~TCP_RCV_LAST_FRAG;
 945         transport->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
 946
 947         transport->tcp_flags &= ~TCP_RCV_COPY_FRAGHDR;
 948         transport->tcp_offset = 0;
 949
 950         /* Sanity check of the record length */
 951         if (unlikely(transport->tcp_reclen < 8)) {
 952                 dprintk("RPC:       invalid TCP record fragment length\n");
 953                 xprt_force_disconnect(xprt);
 954                 return;
 955         }
 956         dprintk("RPC:       reading TCP record fragment of length %d\n",
 957                         transport->tcp_reclen);
 958 }
 959
 960 static void xs_tcp_check_fraghdr(struct sock_xprt *transport)
 961 {
 962         if (transport->tcp_offset == transport->tcp_reclen) {
 963                 transport->tcp_flags |= TCP_RCV_COPY_FRAGHDR;
 964                 transport->tcp_offset = 0;
 965                 if (transport->tcp_flags & TCP_RCV_LAST_FRAG) {
 966                         transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
 967                         transport->tcp_flags |= TCP_RCV_COPY_XID;
 968                         transport->tcp_copied = 0;
 969                 }
 970         }
 971 }
 972
 973 static inline void xs_tcp_read_xid(struct sock_xprt *transport, struct xdr_skb_reader *desc)
 974 {
 975         size_t len, used;
 976         char *p;
 977
 978         len = sizeof(transport->tcp_xid) - transport->tcp_offset;
 979         dprintk("RPC:       reading XID (%Zu bytes)\n", len);
 980         p = ((char *) &transport->tcp_xid) + transport->tcp_offset;
 981         used = xdr_skb_read_bits(desc, p, len);
 982         transport->tcp_offset += used;
 983         if (used != len)
 984                 return;
 985         transport->tcp_flags &= ~TCP_RCV_COPY_XID;
 986         transport->tcp_flags |= TCP_RCV_READ_CALLDIR;
 987         transport->tcp_copied = 4;
 988         dprintk("RPC:       reading %s XID %08x\n",
 989                         (transport->tcp_flags & TCP_RPC_REPLY) ? "reply for"
 990                                                               : "request with",
 991                         ntohl(transport->tcp_xid));
 992         xs_tcp_check_fraghdr(transport);
 993 }
 994
 995 static inline void xs_tcp_read_calldir(struct sock_xprt *transport,
 996                                        struct xdr_skb_reader *desc)
 997 {
 998         size_t len, used;
 999         u32 offset;
1000         __be32  calldir;
1001
1002         /*
1003          * We want transport->tcp_offset to be 8 at the end of this routine
1004          * (4 bytes for the xid and 4 bytes for the call/reply flag).
1005          * When this function is called for the first time,
1006          * transport->tcp_offset is 4 (after having already read the xid).
1007          */
1008         offset = transport->tcp_offset - sizeof(transport->tcp_xid);
1009         len = sizeof(calldir) - offset;
1010         dprintk("RPC:       reading CALL/REPLY flag (%Zu bytes)\n", len);
1011         used = xdr_skb_read_bits(desc, &calldir, len);
1012         transport->tcp_offset += used;
1013         if (used != len)
1014                 return;
1015         transport->tcp_flags &= ~TCP_RCV_READ_CALLDIR;
1016         transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1017         transport->tcp_flags |= TCP_RCV_COPY_DATA;
1018         /*
1019          * We don't yet have the XDR buffer, so we will write the calldir
1020          * out after we get the buffer from the 'struct rpc_rqst'
1021          */
1022         if (ntohl(calldir) == RPC_REPLY)
1023                 transport->tcp_flags |= TCP_RPC_REPLY;
1024         else
1025                 transport->tcp_flags &= ~TCP_RPC_REPLY;
1026         dprintk("RPC:       reading %s CALL/REPLY flag %08x\n",
1027                         (transport->tcp_flags & TCP_RPC_REPLY) ?
1028                                 "reply for" : "request with", calldir);
1029         xs_tcp_check_fraghdr(transport);
1030 }
1031
1032 static inline void xs_tcp_read_common(struct rpc_xprt *xprt,
1033                                      struct xdr_skb_reader *desc,
1034                                      struct rpc_rqst *req)
1035 {
1036         struct sock_xprt *transport =
1037                                 container_of(xprt, struct sock_xprt, xprt);
1038         struct xdr_buf *rcvbuf;
1039         size_t len;
1040         ssize_t r;
1041
1042         rcvbuf = &req->rq_private_buf;
1043
1044         if (transport->tcp_flags & TCP_RCV_COPY_CALLDIR) {
1045                 /*
1046                  * Save the RPC direction in the XDR buffer
1047                  */
1048                 __be32  calldir = transport->tcp_flags & TCP_RPC_REPLY ?
1049                                         htonl(RPC_REPLY) : 0;
1050
1051                 memcpy(rcvbuf->head[0].iov_base + transport->tcp_copied,
1052                         &calldir, sizeof(calldir));
1053                 transport->tcp_copied += sizeof(calldir);
1054                 transport->tcp_flags &= ~TCP_RCV_COPY_CALLDIR;
1055         }
1056
1057         len = desc->count;
1058         if (len > transport->tcp_reclen - transport->tcp_offset) {
1059                 struct xdr_skb_reader my_desc;
1060
1061                 len = transport->tcp_reclen - transport->tcp_offset;
1062                 memcpy(&my_desc, desc, sizeof(my_desc));
1063                 my_desc.count = len;
1064                 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
1065                                           &my_desc, xdr_skb_read_bits);
1066                 desc->count -= r;
1067                 desc->offset += r;
1068         } else
1069                 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
1070                                           desc, xdr_skb_read_bits);
1071
1072         if (r > 0) {
1073                 transport->tcp_copied += r;
1074                 transport->tcp_offset += r;
1075         }
1076         if (r != len) {
1077                 /* Error when copying to the receive buffer,
1078                  * usually because we weren't able to allocate
1079                  * additional buffer pages. All we can do now
1080                  * is turn off TCP_RCV_COPY_DATA, so the request
1081                  * will not receive any additional updates,
1082                  * and time out.
1083                  * Any remaining data from this record will
1084                  * be discarded.
1085                  */
1086                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1087                 dprintk("RPC:       XID %08x truncated request\n",
1088                                 ntohl(transport->tcp_xid));
1089                 dprintk("RPC:       xprt = %p, tcp_copied = %lu, "
1090                                 "tcp_offset = %u, tcp_reclen = %u\n",
1091                                 xprt, transport->tcp_copied,
1092                                 transport->tcp_offset, transport->tcp_reclen);
1093                 return;
1094         }
1095
1096         dprintk("RPC:       XID %08x read %Zd bytes\n",
1097                         ntohl(transport->tcp_xid), r);
1098         dprintk("RPC:       xprt = %p, tcp_copied = %lu, tcp_offset = %u, "
1099                         "tcp_reclen = %u\n", xprt, transport->tcp_copied,
1100                         transport->tcp_offset, transport->tcp_reclen);
1101
1102         if (transport->tcp_copied == req->rq_private_buf.buflen)
1103                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1104         else if (transport->tcp_offset == transport->tcp_reclen) {
1105                 if (transport->tcp_flags & TCP_RCV_LAST_FRAG)
1106                         transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1107         }
1108
1109         return;
1110 }
1111
1112 /*
1113  * Finds the request corresponding to the RPC xid and invokes the common
1114  * tcp read code to read the data.
1115  */
1116 static inline int xs_tcp_read_reply(struct rpc_xprt *xprt,
1117                                     struct xdr_skb_reader *desc)
1118 {
1119         struct sock_xprt *transport =
1120                                 container_of(xprt, struct sock_xprt, xprt);
1121         struct rpc_rqst *req;
1122
1123         dprintk("RPC:       read reply XID %08x\n", ntohl(transport->tcp_xid));
1124
1125         /* Find and lock the request corresponding to this xid */
1126         spin_lock(&xprt->transport_lock);
1127         req = xprt_lookup_rqst(xprt, transport->tcp_xid);
1128         if (!req) {
1129                 dprintk("RPC:       XID %08x request not found!\n",
1130                                 ntohl(transport->tcp_xid));
1131                 spin_unlock(&xprt->transport_lock);
1132                 return -1;
1133         }
1134
1135         xs_tcp_read_common(xprt, desc, req);
1136
1137         if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1138                 xprt_complete_rqst(req->rq_task, transport->tcp_copied);
1139
1140         spin_unlock(&xprt->transport_lock);
1141         return 0;
1142 }
1143
1144 #if defined(CONFIG_NFS_V4_1)
1145 /*
1146  * Obtains an rpc_rqst previously allocated and invokes the common
1147  * tcp read code to read the data.  The result is placed in the callback
1148  * queue.
1149  * If we're unable to obtain the rpc_rqst we schedule the closing of the
1150  * connection and return -1.
1151  */
1152 static inline int xs_tcp_read_callback(struct rpc_xprt *xprt,
1153                                        struct xdr_skb_reader *desc)
1154 {
1155         struct sock_xprt *transport =
1156                                 container_of(xprt, struct sock_xprt, xprt);
1157         struct rpc_rqst *req;
1158
1159         req = xprt_alloc_bc_request(xprt);
1160         if (req == NULL) {
1161                 printk(KERN_WARNING "Callback slot table overflowed\n");
1162                 xprt_force_disconnect(xprt);
1163                 return -1;
1164         }
1165
1166         req->rq_xid = transport->tcp_xid;
1167         dprintk("RPC:       read callback  XID %08x\n", ntohl(req->rq_xid));
1168         xs_tcp_read_common(xprt, desc, req);
1169
1170         if (!(transport->tcp_flags & TCP_RCV_COPY_DATA)) {
1171                 struct svc_serv *bc_serv = xprt->bc_serv;
1172
1173                 /*
1174                  * Add callback request to callback list.  The callback
1175                  * service sleeps on the sv_cb_waitq waiting for new
1176                  * requests.  Wake it up after adding enqueing the
1177                  * request.
1178                  */
1179                 dprintk("RPC:       add callback request to list\n");
1180                 spin_lock(&bc_serv->sv_cb_lock);
1181                 list_add(&req->rq_bc_list, &bc_serv->sv_cb_list);
1182                 spin_unlock(&bc_serv->sv_cb_lock);
1183                 wake_up(&bc_serv->sv_cb_waitq);
1184         }
1185
1186         req->rq_private_buf.len = transport->tcp_copied;
1187
1188         return 0;
1189 }
1190
1191 static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1192                                         struct xdr_skb_reader *desc)
1193 {
1194         struct sock_xprt *transport =
1195                                 container_of(xprt, struct sock_xprt, xprt);
1196
1197         return (transport->tcp_flags & TCP_RPC_REPLY) ?
1198                 xs_tcp_read_reply(xprt, desc) :
1199                 xs_tcp_read_callback(xprt, desc);
1200 }
1201 #else
1202 static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1203                                         struct xdr_skb_reader *desc)
1204 {
1205         return xs_tcp_read_reply(xprt, desc);
1206 }
1207 #endif /* CONFIG_NFS_V4_1 */
1208
1209 /*
1210  * Read data off the transport.  This can be either an RPC_CALL or an
1211  * RPC_REPLY.  Relay the processing to helper functions.
1212  */
1213 static void xs_tcp_read_data(struct rpc_xprt *xprt,
1214                                     struct xdr_skb_reader *desc)
1215 {
1216         struct sock_xprt *transport =
1217                                 container_of(xprt, struct sock_xprt, xprt);
1218
1219         if (_xs_tcp_read_data(xprt, desc) == 0)
1220                 xs_tcp_check_fraghdr(transport);
1221         else {
1222                 /*
1223                  * The transport_lock protects the request handling.
1224                  * There's no need to hold it to update the tcp_flags.
1225                  */
1226                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1227         }
1228 }
1229
1230 static inline void xs_tcp_read_discard(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1231 {
1232         size_t len;
1233
1234         len = transport->tcp_reclen - transport->tcp_offset;
1235         if (len > desc->count)
1236                 len = desc->count;
1237         desc->count -= len;
1238         desc->offset += len;
1239         transport->tcp_offset += len;
1240         dprintk("RPC:       discarded %Zu bytes\n", len);
1241         xs_tcp_check_fraghdr(transport);
1242 }
1243
1244 static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
1245 {
1246         struct rpc_xprt *xprt = rd_desc->arg.data;
1247         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1248         struct xdr_skb_reader desc = {
1249                 .skb    = skb,
1250                 .offset = offset,
1251                 .count  = len,
1252         };
1253
1254         dprintk("RPC:       xs_tcp_data_recv started\n");
1255         do {
1256                 /* Read in a new fragment marker if necessary */
1257                 /* Can we ever really expect to get completely empty fragments? */
1258                 if (transport->tcp_flags & TCP_RCV_COPY_FRAGHDR) {
1259                         xs_tcp_read_fraghdr(xprt, &desc);
1260                         continue;
1261                 }
1262                 /* Read in the xid if necessary */
1263                 if (transport->tcp_flags & TCP_RCV_COPY_XID) {
1264                         xs_tcp_read_xid(transport, &desc);
1265                         continue;
1266                 }
1267                 /* Read in the call/reply flag */
1268                 if (transport->tcp_flags & TCP_RCV_READ_CALLDIR) {
1269                         xs_tcp_read_calldir(transport, &desc);
1270                         continue;
1271                 }
1272                 /* Read in the request data */
1273                 if (transport->tcp_flags & TCP_RCV_COPY_DATA) {
1274                         xs_tcp_read_data(xprt, &desc);
1275                         continue;
1276                 }
1277                 /* Skip over any trailing bytes on short reads */
1278                 xs_tcp_read_discard(transport, &desc);
1279         } while (desc.count);
1280         dprintk("RPC:       xs_tcp_data_recv done\n");
1281         return len - desc.count;
1282 }
1283
1284 /**
1285  * xs_tcp_data_ready - "data ready" callback for TCP sockets
1286  * @sk: socket with data to read
1287  * @bytes: how much data to read
1288  *
1289  */
1290 static void xs_tcp_data_ready(struct sock *sk, int bytes)
1291 {
1292         struct rpc_xprt *xprt;
1293         read_descriptor_t rd_desc;
1294         int read;
1295
1296         dprintk("RPC:       xs_tcp_data_ready...\n");
1297
1298         read_lock(&sk->sk_callback_lock);
1299         if (!(xprt = xprt_from_sock(sk)))
1300                 goto out;
1301         if (xprt->shutdown)
1302                 goto out;
1303
1304         /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
1305         rd_desc.arg.data = xprt;
1306         do {
1307                 rd_desc.count = 65536;
1308                 read = tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
1309         } while (read > 0);
1310 out:
1311         read_unlock(&sk->sk_callback_lock);
1312 }
1313
1314 /*
1315  * Do the equivalent of linger/linger2 handling for dealing with
1316  * broken servers that don't close the socket in a timely
1317  * fashion
1318  */
1319 static void xs_tcp_schedule_linger_timeout(struct rpc_xprt *xprt,
1320                 unsigned long timeout)
1321 {
1322         struct sock_xprt *transport;
1323
1324         if (xprt_test_and_set_connecting(xprt))
1325                 return;
1326         set_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1327         transport = container_of(xprt, struct sock_xprt, xprt);
1328         queue_delayed_work(rpciod_workqueue, &transport->connect_worker,
1329                            timeout);
1330 }
1331
1332 static void xs_tcp_cancel_linger_timeout(struct rpc_xprt *xprt)
1333 {
1334         struct sock_xprt *transport;
1335
1336         transport = container_of(xprt, struct sock_xprt, xprt);
1337
1338         if (!test_bit(XPRT_CONNECTION_ABORT, &xprt->state) ||
1339             !cancel_delayed_work(&transport->connect_worker))
1340                 return;
1341         clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1342         xprt_clear_connecting(xprt);
1343 }
1344
1345 static void xs_sock_mark_closed(struct rpc_xprt *xprt)
1346 {
1347         smp_mb__before_clear_bit();
1348         clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1349         clear_bit(XPRT_CLOSING, &xprt->state);
1350         smp_mb__after_clear_bit();
1351         /* Mark transport as closed and wake up all pending tasks */
1352         xprt_disconnect_done(xprt);
1353 }
1354
1355 /**
1356  * xs_tcp_state_change - callback to handle TCP socket state changes
1357  * @sk: socket whose state has changed
1358  *
1359  */
1360 static void xs_tcp_state_change(struct sock *sk)
1361 {
1362         struct rpc_xprt *xprt;
1363
1364         read_lock(&sk->sk_callback_lock);
1365         if (!(xprt = xprt_from_sock(sk)))
1366                 goto out;
1367         dprintk("RPC:       xs_tcp_state_change client %p...\n", xprt);
1368         dprintk("RPC:       state %x conn %d dead %d zapped %d\n",
1369                         sk->sk_state, xprt_connected(xprt),
1370                         sock_flag(sk, SOCK_DEAD),
1371                         sock_flag(sk, SOCK_ZAPPED));
1372
1373         switch (sk->sk_state) {
1374         case TCP_ESTABLISHED:
1375                 spin_lock_bh(&xprt->transport_lock);
1376                 if (!xprt_test_and_set_connected(xprt)) {
1377                         struct sock_xprt *transport = container_of(xprt,
1378                                         struct sock_xprt, xprt);
1379
1380                         /* Reset TCP record info */
1381                         transport->tcp_offset = 0;
1382                         transport->tcp_reclen = 0;
1383                         transport->tcp_copied = 0;
1384                         transport->tcp_flags =
1385                                 TCP_RCV_COPY_FRAGHDR | TCP_RCV_COPY_XID;
1386
1387                         xprt_wake_pending_tasks(xprt, -EAGAIN);
1388                 }
1389                 spin_unlock_bh(&xprt->transport_lock);
1390                 break;
1391         case TCP_FIN_WAIT1:
1392                 /* The client initiated a shutdown of the socket */
1393                 xprt->connect_cookie++;
1394                 xprt->reestablish_timeout = 0;
1395                 set_bit(XPRT_CLOSING, &xprt->state);
1396                 smp_mb__before_clear_bit();
1397                 clear_bit(XPRT_CONNECTED, &xprt->state);
1398                 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1399                 smp_mb__after_clear_bit();
1400                 xs_tcp_schedule_linger_timeout(xprt, xs_tcp_fin_timeout);
1401                 break;
1402         case TCP_CLOSE_WAIT:
1403                 /* The server initiated a shutdown of the socket */
1404                 xprt_force_disconnect(xprt);
1405         case TCP_SYN_SENT:
1406                 xprt->connect_cookie++;
1407         case TCP_CLOSING:
1408                 /*
1409                  * If the server closed down the connection, make sure that
1410                  * we back off before reconnecting
1411                  */
1412                 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1413                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1414                 break;
1415         case TCP_LAST_ACK:
1416                 set_bit(XPRT_CLOSING, &xprt->state);
1417                 xs_tcp_schedule_linger_timeout(xprt, xs_tcp_fin_timeout);
1418                 smp_mb__before_clear_bit();
1419                 clear_bit(XPRT_CONNECTED, &xprt->state);
1420                 smp_mb__after_clear_bit();
1421                 break;
1422         case TCP_CLOSE:
1423                 xs_tcp_cancel_linger_timeout(xprt);
1424                 xs_sock_mark_closed(xprt);
1425         }
1426  out:
1427         read_unlock(&sk->sk_callback_lock);
1428 }
1429
1430 /**
1431  * xs_error_report - callback mainly for catching socket errors
1432  * @sk: socket
1433  */
1434 static void xs_error_report(struct sock *sk)
1435 {
1436         struct rpc_xprt *xprt;
1437
1438         read_lock(&sk->sk_callback_lock);
1439         if (!(xprt = xprt_from_sock(sk)))
1440                 goto out;
1441         dprintk("RPC:       %s client %p...\n"
1442                         "RPC:       error %d\n",
1443                         __func__, xprt, sk->sk_err);
1444         xprt_wake_pending_tasks(xprt, -EAGAIN);
1445 out:
1446         read_unlock(&sk->sk_callback_lock);
1447 }
1448
1449 static void xs_write_space(struct sock *sk)
1450 {
1451         struct socket *sock;
1452         struct rpc_xprt *xprt;
1453
1454         if (unlikely(!(sock = sk->sk_socket)))
1455                 return;
1456         clear_bit(SOCK_NOSPACE, &sock->flags);
1457
1458         if (unlikely(!(xprt = xprt_from_sock(sk))))
1459                 return;
1460         if (test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags) == 0)
1461                 return;
1462
1463         xprt_write_space(xprt);
1464 }
1465
1466 /**
1467  * xs_udp_write_space - callback invoked when socket buffer space
1468  *                             becomes available
1469  * @sk: socket whose state has changed
1470  *
1471  * Called when more output buffer space is available for this socket.
1472  * We try not to wake our writers until they can make "significant"
1473  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1474  * with a bunch of small requests.
1475  */
1476 static void xs_udp_write_space(struct sock *sk)
1477 {
1478         read_lock(&sk->sk_callback_lock);
1479
1480         /* from net/core/sock.c:sock_def_write_space */
1481         if (sock_writeable(sk))
1482                 xs_write_space(sk);
1483
1484         read_unlock(&sk->sk_callback_lock);
1485 }
1486
1487 /**
1488  * xs_tcp_write_space - callback invoked when socket buffer space
1489  *                             becomes available
1490  * @sk: socket whose state has changed
1491  *
1492  * Called when more output buffer space is available for this socket.
1493  * We try not to wake our writers until they can make "significant"
1494  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1495  * with a bunch of small requests.
1496  */
1497 static void xs_tcp_write_space(struct sock *sk)
1498 {
1499         read_lock(&sk->sk_callback_lock);
1500
1501         /* from net/core/stream.c:sk_stream_write_space */
1502         if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk))
1503                 xs_write_space(sk);
1504
1505         read_unlock(&sk->sk_callback_lock);
1506 }
1507
1508 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
1509 {
1510         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1511         struct sock *sk = transport->inet;
1512
1513         if (transport->rcvsize) {
1514                 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1515                 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
1516         }
1517         if (transport->sndsize) {
1518                 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1519                 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1520                 sk->sk_write_space(sk);
1521         }
1522 }
1523
1524 /**
1525  * xs_udp_set_buffer_size - set send and receive limits
1526  * @xprt: generic transport
1527  * @sndsize: requested size of send buffer, in bytes
1528  * @rcvsize: requested size of receive buffer, in bytes
1529  *
1530  * Set socket send and receive buffer size limits.
1531  */
1532 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1533 {
1534         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1535
1536         transport->sndsize = 0;
1537         if (sndsize)
1538                 transport->sndsize = sndsize + 1024;
1539         transport->rcvsize = 0;
1540         if (rcvsize)
1541                 transport->rcvsize = rcvsize + 1024;
1542
1543         xs_udp_do_set_buffer_size(xprt);
1544 }
1545
1546 /**
1547  * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1548  * @task: task that timed out
1549  *
1550  * Adjust the congestion window after a retransmit timeout has occurred.
1551  */
1552 static void xs_udp_timer(struct rpc_task *task)
1553 {
1554         xprt_adjust_cwnd(task, -ETIMEDOUT);
1555 }
1556
1557 static unsigned short xs_get_random_port(void)
1558 {
1559         unsigned short range = xprt_max_resvport - xprt_min_resvport;
1560         unsigned short rand = (unsigned short) net_random() % range;
1561         return rand + xprt_min_resvport;
1562 }
1563
1564 /**
1565  * xs_set_port - reset the port number in the remote endpoint address
1566  * @xprt: generic transport
1567  * @port: new port number
1568  *
1569  */
1570 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1571 {
1572         struct sockaddr *addr = xs_addr(xprt);
1573
1574         dprintk("RPC:       setting port for xprt %p to %u\n", xprt, port);
1575
1576         switch (addr->sa_family) {
1577         case AF_INET:
1578                 ((struct sockaddr_in *)addr)->sin_port = htons(port);
1579                 break;
1580         case AF_INET6:
1581                 ((struct sockaddr_in6 *)addr)->sin6_port = htons(port);
1582                 break;
1583         default:
1584                 BUG();
1585         }
1586 }
1587
1588 static unsigned short xs_get_srcport(struct sock_xprt *transport, struct socket *sock)
1589 {
1590         unsigned short port = transport->port;
1591
1592         if (port == 0 && transport->xprt.resvport)
1593                 port = xs_get_random_port();
1594         return port;
1595 }
1596
1597 static unsigned short xs_next_srcport(struct sock_xprt *transport, struct socket *sock, unsigned short port)
1598 {
1599         if (transport->port != 0)
1600                 transport->port = 0;
1601         if (!transport->xprt.resvport)
1602                 return 0;
1603         if (port <= xprt_min_resvport || port > xprt_max_resvport)
1604                 return xprt_max_resvport;
1605         return --port;
1606 }
1607
1608 static int xs_bind4(struct sock_xprt *transport, struct socket *sock)
1609 {
1610         struct sockaddr_in myaddr = {
1611                 .sin_family = AF_INET,
1612         };
1613         struct sockaddr_in *sa;
1614         int err, nloop = 0;
1615         unsigned short port = xs_get_srcport(transport, sock);
1616         unsigned short last;
1617
1618         sa = (struct sockaddr_in *)&transport->addr;
1619         myaddr.sin_addr = sa->sin_addr;
1620         do {
1621                 myaddr.sin_port = htons(port);
1622                 err = kernel_bind(sock, (struct sockaddr *) &myaddr,
1623                                                 sizeof(myaddr));
1624                 if (port == 0)
1625                         break;
1626                 if (err == 0) {
1627                         transport->port = port;
1628                         break;
1629                 }
1630                 last = port;
1631                 port = xs_next_srcport(transport, sock, port);
1632                 if (port > last)
1633                         nloop++;
1634         } while (err == -EADDRINUSE && nloop != 2);
1635         dprintk("RPC:       %s %pI4:%u: %s (%d)\n",
1636                         __func__, &myaddr.sin_addr,
1637                         port, err ? "failed" : "ok", err);
1638         return err;
1639 }
1640
1641 static int xs_bind6(struct sock_xprt *transport, struct socket *sock)
1642 {
1643         struct sockaddr_in6 myaddr = {
1644                 .sin6_family = AF_INET6,
1645         };
1646         struct sockaddr_in6 *sa;
1647         int err, nloop = 0;
1648         unsigned short port = xs_get_srcport(transport, sock);
1649         unsigned short last;
1650
1651         sa = (struct sockaddr_in6 *)&transport->addr;
1652         myaddr.sin6_addr = sa->sin6_addr;
1653         do {
1654                 myaddr.sin6_port = htons(port);
1655                 err = kernel_bind(sock, (struct sockaddr *) &myaddr,
1656                                                 sizeof(myaddr));
1657                 if (port == 0)
1658                         break;
1659                 if (err == 0) {
1660                         transport->port = port;
1661                         break;
1662                 }
1663                 last = port;
1664                 port = xs_next_srcport(transport, sock, port);
1665                 if (port > last)
1666                         nloop++;
1667         } while (err == -EADDRINUSE && nloop != 2);
1668         dprintk("RPC:       xs_bind6 %pI6:%u: %s (%d)\n",
1669                 &myaddr.sin6_addr, port, err ? "failed" : "ok", err);
1670         return err;
1671 }
1672
1673 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1674 static struct lock_class_key xs_key[2];
1675 static struct lock_class_key xs_slock_key[2];
1676
1677 static inline void xs_reclassify_socket4(struct socket *sock)
1678 {
1679         struct sock *sk = sock->sk;
1680
1681         BUG_ON(sock_owned_by_user(sk));
1682         sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1683                 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1684 }
1685
1686 static inline void xs_reclassify_socket6(struct socket *sock)
1687 {
1688         struct sock *sk = sock->sk;
1689
1690         BUG_ON(sock_owned_by_user(sk));
1691         sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1692                 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
1693 }
1694 #else
1695 static inline void xs_reclassify_socket4(struct socket *sock)
1696 {
1697 }
1698
1699 static inline void xs_reclassify_socket6(struct socket *sock)
1700 {
1701 }
1702 #endif
1703
1704 static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
1705 {
1706         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1707
1708         if (!transport->inet) {
1709                 struct sock *sk = sock->sk;
1710
1711                 write_lock_bh(&sk->sk_callback_lock);
1712
1713                 xs_save_old_callbacks(transport, sk);
1714
1715                 sk->sk_user_data = xprt;
1716                 sk->sk_data_ready = xs_udp_data_ready;
1717                 sk->sk_write_space = xs_udp_write_space;
1718                 sk->sk_error_report = xs_error_report;
1719                 sk->sk_no_check = UDP_CSUM_NORCV;
1720                 sk->sk_allocation = GFP_ATOMIC;
1721
1722                 xprt_set_connected(xprt);
1723
1724                 /* Reset to new socket */
1725                 transport->sock = sock;
1726                 transport->inet = sk;
1727
1728                 write_unlock_bh(&sk->sk_callback_lock);
1729         }
1730         xs_udp_do_set_buffer_size(xprt);
1731 }
1732
1733 /**
1734  * xs_udp_connect_worker4 - set up a UDP socket
1735  * @work: RPC transport to connect
1736  *
1737  * Invoked by a work queue tasklet.
1738  */
1739 static void xs_udp_connect_worker4(struct work_struct *work)
1740 {
1741         struct sock_xprt *transport =
1742                 container_of(work, struct sock_xprt, connect_worker.work);
1743         struct rpc_xprt *xprt = &transport->xprt;
1744         struct socket *sock = transport->sock;
1745         int err, status = -EIO;
1746
1747         if (xprt->shutdown)
1748                 goto out;
1749
1750         /* Start by resetting any existing state */
1751         xs_reset_transport(transport);
1752
1753         err = sock_create_kern(PF_INET, SOCK_DGRAM, IPPROTO_UDP, &sock);
1754         if (err < 0) {
1755                 dprintk("RPC:       can't create UDP transport socket (%d).\n", -err);
1756                 goto out;
1757         }
1758         xs_reclassify_socket4(sock);
1759
1760         if (xs_bind4(transport, sock)) {
1761                 sock_release(sock);
1762                 goto out;
1763         }
1764
1765         dprintk("RPC:       worker connecting xprt %p to address: %s\n",
1766                         xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1767
1768         xs_udp_finish_connecting(xprt, sock);
1769         status = 0;
1770 out:
1771         xprt_clear_connecting(xprt);
1772         xprt_wake_pending_tasks(xprt, status);
1773 }
1774
1775 /**
1776  * xs_udp_connect_worker6 - set up a UDP socket
1777  * @work: RPC transport to connect
1778  *
1779  * Invoked by a work queue tasklet.
1780  */
1781 static void xs_udp_connect_worker6(struct work_struct *work)
1782 {
1783         struct sock_xprt *transport =
1784                 container_of(work, struct sock_xprt, connect_worker.work);
1785         struct rpc_xprt *xprt = &transport->xprt;
1786         struct socket *sock = transport->sock;
1787         int err, status = -EIO;
1788
1789         if (xprt->shutdown)
1790                 goto out;
1791
1792         /* Start by resetting any existing state */
1793         xs_reset_transport(transport);
1794
1795         err = sock_create_kern(PF_INET6, SOCK_DGRAM, IPPROTO_UDP, &sock);
1796         if (err < 0) {
1797                 dprintk("RPC:       can't create UDP transport socket (%d).\n", -err);
1798                 goto out;
1799         }
1800         xs_reclassify_socket6(sock);
1801
1802         if (xs_bind6(transport, sock) < 0) {
1803                 sock_release(sock);
1804                 goto out;
1805         }
1806
1807         dprintk("RPC:       worker connecting xprt %p to address: %s\n",
1808                         xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1809
1810         xs_udp_finish_connecting(xprt, sock);
1811         status = 0;
1812 out:
1813         xprt_clear_connecting(xprt);
1814         xprt_wake_pending_tasks(xprt, status);
1815 }
1816
1817 /*
1818  * We need to preserve the port number so the reply cache on the server can
1819  * find our cached RPC replies when we get around to reconnecting.
1820  */
1821 static void xs_abort_connection(struct rpc_xprt *xprt, struct sock_xprt *transport)
1822 {
1823         int result;
1824         struct sockaddr any;
1825
1826         dprintk("RPC:       disconnecting xprt %p to reuse port\n", xprt);
1827
1828         /*
1829          * Disconnect the transport socket by doing a connect operation
1830          * with AF_UNSPEC.  This should return immediately...
1831          */
1832         memset(&any, 0, sizeof(any));
1833         any.sa_family = AF_UNSPEC;
1834         result = kernel_connect(transport->sock, &any, sizeof(any), 0);
1835         if (!result)
1836                 xs_sock_mark_closed(xprt);
1837         else
1838                 dprintk("RPC:       AF_UNSPEC connect return code %d\n",
1839                                 result);
1840 }
1841
1842 static void xs_tcp_reuse_connection(struct rpc_xprt *xprt, struct sock_xprt *transport)
1843 {
1844         unsigned int state = transport->inet->sk_state;
1845
1846         if (state == TCP_CLOSE && transport->sock->state == SS_UNCONNECTED)
1847                 return;
1848         if ((1 << state) & (TCPF_ESTABLISHED|TCPF_SYN_SENT))
1849                 return;
1850         xs_abort_connection(xprt, transport);
1851 }
1852
1853 static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
1854 {
1855         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1856
1857         if (!transport->inet) {
1858                 struct sock *sk = sock->sk;
1859
1860                 write_lock_bh(&sk->sk_callback_lock);
1861
1862                 xs_save_old_callbacks(transport, sk);
1863
1864                 sk->sk_user_data = xprt;
1865                 sk->sk_data_ready = xs_tcp_data_ready;
1866                 sk->sk_state_change = xs_tcp_state_change;
1867                 sk->sk_write_space = xs_tcp_write_space;
1868                 sk->sk_error_report = xs_error_report;
1869                 sk->sk_allocation = GFP_ATOMIC;
1870
1871                 /* socket options */
1872                 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
1873                 sock_reset_flag(sk, SOCK_LINGER);
1874                 tcp_sk(sk)->linger2 = 0;
1875                 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
1876
1877                 xprt_clear_connected(xprt);
1878
1879                 /* Reset to new socket */
1880                 transport->sock = sock;
1881                 transport->inet = sk;
1882
1883                 write_unlock_bh(&sk->sk_callback_lock);
1884         }
1885
1886         if (!xprt_bound(xprt))
1887                 return -ENOTCONN;
1888
1889         /* Tell the socket layer to start connecting... */
1890         xprt->stat.connect_count++;
1891         xprt->stat.connect_start = jiffies;
1892         return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
1893 }
1894
1895 /**
1896  * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint
1897  * @xprt: RPC transport to connect
1898  * @transport: socket transport to connect
1899  * @create_sock: function to create a socket of the correct type
1900  *
1901  * Invoked by a work queue tasklet.
1902  */
1903 static void xs_tcp_setup_socket(struct rpc_xprt *xprt,
1904                 struct sock_xprt *transport,
1905                 struct socket *(*create_sock)(struct rpc_xprt *,
1906                         struct sock_xprt *))
1907 {
1908         struct socket *sock = transport->sock;
1909         int status = -EIO;
1910
1911         if (xprt->shutdown)
1912                 goto out;
1913
1914         if (!sock) {
1915                 clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1916                 sock = create_sock(xprt, transport);
1917                 if (IS_ERR(sock)) {
1918                         status = PTR_ERR(sock);
1919                         goto out;
1920                 }
1921         } else {
1922                 int abort_and_exit;
1923
1924                 abort_and_exit = test_and_clear_bit(XPRT_CONNECTION_ABORT,
1925                                 &xprt->state);
1926                 /* "close" the socket, preserving the local port */
1927                 xs_tcp_reuse_connection(xprt, transport);
1928
1929                 if (abort_and_exit)
1930                         goto out_eagain;
1931         }
1932
1933         dprintk("RPC:       worker connecting xprt %p to address: %s\n",
1934                         xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1935
1936         status = xs_tcp_finish_connecting(xprt, sock);
1937         dprintk("RPC:       %p connect status %d connected %d sock state %d\n",
1938                         xprt, -status, xprt_connected(xprt),
1939                         sock->sk->sk_state);
1940         switch (status) {
1941         default:
1942                 printk("%s: connect returned unhandled error %d\n",
1943                         __func__, status);
1944         case -EADDRNOTAVAIL:
1945                 /* We're probably in TIME_WAIT. Get rid of existing socket,
1946                  * and retry
1947                  */
1948                 set_bit(XPRT_CONNECTION_CLOSE, &xprt->state);
1949                 xprt_force_disconnect(xprt);
1950                 break;
1951         case -ECONNREFUSED:
1952         case -ECONNRESET:
1953         case -ENETUNREACH:
1954                 /* retry with existing socket, after a delay */
1955         case 0:
1956         case -EINPROGRESS:
1957         case -EALREADY:
1958                 xprt_clear_connecting(xprt);
1959                 return;
1960         }
1961 out_eagain:
1962         status = -EAGAIN;
1963 out:
1964         xprt_clear_connecting(xprt);
1965         xprt_wake_pending_tasks(xprt, status);
1966 }
1967
1968 static struct socket *xs_create_tcp_sock4(struct rpc_xprt *xprt,
1969                 struct sock_xprt *transport)
1970 {
1971         struct socket *sock;
1972         int err;
1973
1974         /* start from scratch */
1975         err = sock_create_kern(PF_INET, SOCK_STREAM, IPPROTO_TCP, &sock);
1976         if (err < 0) {
1977                 dprintk("RPC:       can't create TCP transport socket (%d).\n",
1978                                 -err);
1979                 goto out_err;
1980         }
1981         xs_reclassify_socket4(sock);
1982
1983         if (xs_bind4(transport, sock) < 0) {
1984                 sock_release(sock);
1985                 goto out_err;
1986         }
1987         return sock;
1988 out_err:
1989         return ERR_PTR(-EIO);
1990 }
1991
1992 /**
1993  * xs_tcp_connect_worker4 - connect a TCP socket to a remote endpoint
1994  * @work: RPC transport to connect
1995  *
1996  * Invoked by a work queue tasklet.
1997  */
1998 static void xs_tcp_connect_worker4(struct work_struct *work)
1999 {
2000         struct sock_xprt *transport =
2001                 container_of(work, struct sock_xprt, connect_worker.work);
2002         struct rpc_xprt *xprt = &transport->xprt;
2003
2004         xs_tcp_setup_socket(xprt, transport, xs_create_tcp_sock4);
2005 }
2006
2007 static struct socket *xs_create_tcp_sock6(struct rpc_xprt *xprt,
2008                 struct sock_xprt *transport)
2009 {
2010         struct socket *sock;
2011         int err;
2012
2013         /* start from scratch */
2014         err = sock_create_kern(PF_INET6, SOCK_STREAM, IPPROTO_TCP, &sock);
2015         if (err < 0) {
2016                 dprintk("RPC:       can't create TCP transport socket (%d).\n",
2017                                 -err);
2018                 goto out_err;
2019         }
2020         xs_reclassify_socket6(sock);
2021
2022         if (xs_bind6(transport, sock) < 0) {
2023                 sock_release(sock);
2024                 goto out_err;
2025         }
2026         return sock;
2027 out_err:
2028         return ERR_PTR(-EIO);
2029 }
2030
2031 /**
2032  * xs_tcp_connect_worker6 - connect a TCP socket to a remote endpoint
2033  * @work: RPC transport to connect
2034  *
2035  * Invoked by a work queue tasklet.
2036  */
2037 static void xs_tcp_connect_worker6(struct work_struct *work)
2038 {
2039         struct sock_xprt *transport =
2040                 container_of(work, struct sock_xprt, connect_worker.work);
2041         struct rpc_xprt *xprt = &transport->xprt;
2042
2043         xs_tcp_setup_socket(xprt, transport, xs_create_tcp_sock6);
2044 }
2045
2046 /**
2047  * xs_connect - connect a socket to a remote endpoint
2048  * @task: address of RPC task that manages state of connect request
2049  *
2050  * TCP: If the remote end dropped the connection, delay reconnecting.
2051  *
2052  * UDP socket connects are synchronous, but we use a work queue anyway
2053  * to guarantee that even unprivileged user processes can set up a
2054  * socket on a privileged port.
2055  *
2056  * If a UDP socket connect fails, the delay behavior here prevents
2057  * retry floods (hard mounts).
2058  */
2059 static void xs_connect(struct rpc_task *task)
2060 {
2061         struct rpc_xprt *xprt = task->tk_xprt;
2062         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2063
2064         if (xprt_test_and_set_connecting(xprt))
2065                 return;
2066
2067         if (transport->sock != NULL) {
2068                 dprintk("RPC:       xs_connect delayed xprt %p for %lu "
2069                                 "seconds\n",
2070                                 xprt, xprt->reestablish_timeout / HZ);
2071                 queue_delayed_work(rpciod_workqueue,
2072                                    &transport->connect_worker,
2073                                    xprt->reestablish_timeout);
2074                 xprt->reestablish_timeout <<= 1;
2075                 if (xprt->reestablish_timeout > XS_TCP_MAX_REEST_TO)
2076                         xprt->reestablish_timeout = XS_TCP_MAX_REEST_TO;
2077         } else {
2078                 dprintk("RPC:       xs_connect scheduled xprt %p\n", xprt);
2079                 queue_delayed_work(rpciod_workqueue,
2080                                    &transport->connect_worker, 0);
2081         }
2082 }
2083
2084 static void xs_tcp_connect(struct rpc_task *task)
2085 {
2086         struct rpc_xprt *xprt = task->tk_xprt;
2087
2088         /* Exit if we need to wait for socket shutdown to complete */
2089         if (test_bit(XPRT_CLOSING, &xprt->state))
2090                 return;
2091         xs_connect(task);
2092 }
2093
2094 /**
2095  * xs_udp_print_stats - display UDP socket-specifc stats
2096  * @xprt: rpc_xprt struct containing statistics
2097  * @seq: output file
2098  *
2099  */
2100 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2101 {
2102         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2103
2104         seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %Lu %Lu\n",
2105                         transport->port,
2106                         xprt->stat.bind_count,
2107                         xprt->stat.sends,
2108                         xprt->stat.recvs,
2109                         xprt->stat.bad_xids,
2110                         xprt->stat.req_u,
2111                         xprt->stat.bklog_u);
2112 }
2113
2114 /**
2115  * xs_tcp_print_stats - display TCP socket-specifc stats
2116  * @xprt: rpc_xprt struct containing statistics
2117  * @seq: output file
2118  *
2119  */
2120 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2121 {
2122         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2123         long idle_time = 0;
2124
2125         if (xprt_connected(xprt))
2126                 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2127
2128         seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu %Lu %Lu\n",
2129                         transport->port,
2130                         xprt->stat.bind_count,
2131                         xprt->stat.connect_count,
2132                         xprt->stat.connect_time,
2133                         idle_time,
2134                         xprt->stat.sends,
2135                         xprt->stat.recvs,
2136                         xprt->stat.bad_xids,
2137                         xprt->stat.req_u,
2138                         xprt->stat.bklog_u);
2139 }
2140
2141 static struct rpc_xprt_ops xs_udp_ops = {
2142         .set_buffer_size        = xs_udp_set_buffer_size,
2143         .reserve_xprt           = xprt_reserve_xprt_cong,
2144         .release_xprt           = xprt_release_xprt_cong,
2145         .rpcbind                = rpcb_getport_async,
2146         .set_port               = xs_set_port,
2147         .connect                = xs_connect,
2148         .buf_alloc              = rpc_malloc,
2149         .buf_free               = rpc_free,
2150         .send_request           = xs_udp_send_request,
2151         .set_retrans_timeout    = xprt_set_retrans_timeout_rtt,
2152         .timer                  = xs_udp_timer,
2153         .release_request        = xprt_release_rqst_cong,
2154         .close                  = xs_close,
2155         .destroy                = xs_destroy,
2156         .print_stats            = xs_udp_print_stats,
2157 };
2158
2159 static struct rpc_xprt_ops xs_tcp_ops = {
2160         .reserve_xprt           = xprt_reserve_xprt,
2161         .release_xprt           = xs_tcp_release_xprt,
2162         .rpcbind                = rpcb_getport_async,
2163         .set_port               = xs_set_port,
2164         .connect                = xs_tcp_connect,
2165         .buf_alloc              = rpc_malloc,
2166         .buf_free               = rpc_free,
2167         .send_request           = xs_tcp_send_request,
2168         .set_retrans_timeout    = xprt_set_retrans_timeout_def,
2169 #if defined(CONFIG_NFS_V4_1)
2170         .release_request        = bc_release_request,
2171 #endif /* CONFIG_NFS_V4_1 */
2172         .close                  = xs_tcp_close,
2173         .destroy                = xs_destroy,
2174         .print_stats            = xs_tcp_print_stats,
2175 };
2176
2177 static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
2178                                       unsigned int slot_table_size)
2179 {
2180         struct rpc_xprt *xprt;
2181         struct sock_xprt *new;
2182
2183         if (args->addrlen > sizeof(xprt->addr)) {
2184                 dprintk("RPC:       xs_setup_xprt: address too large\n");
2185                 return ERR_PTR(-EBADF);
2186         }
2187
2188         new = kzalloc(sizeof(*new), GFP_KERNEL);
2189         if (new == NULL) {
2190                 dprintk("RPC:       xs_setup_xprt: couldn't allocate "
2191                                 "rpc_xprt\n");
2192                 return ERR_PTR(-ENOMEM);
2193         }
2194         xprt = &new->xprt;
2195
2196         xprt->max_reqs = slot_table_size;
2197         xprt->slot = kcalloc(xprt->max_reqs, sizeof(struct rpc_rqst), GFP_KERNEL);
2198         if (xprt->slot == NULL) {
2199                 kfree(xprt);
2200                 dprintk("RPC:       xs_setup_xprt: couldn't allocate slot "
2201                                 "table\n");
2202                 return ERR_PTR(-ENOMEM);
2203         }
2204
2205         memcpy(&xprt->addr, args->dstaddr, args->addrlen);
2206         xprt->addrlen = args->addrlen;
2207         if (args->srcaddr)
2208                 memcpy(&new->addr, args->srcaddr, args->addrlen);
2209
2210         return xprt;
2211 }
2212
2213 static const struct rpc_timeout xs_udp_default_timeout = {
2214         .to_initval = 5 * HZ,
2215         .to_maxval = 30 * HZ,
2216         .to_increment = 5 * HZ,
2217         .to_retries = 5,
2218 };
2219
2220 /**
2221  * xs_setup_udp - Set up transport to use a UDP socket
2222  * @args: rpc transport creation arguments
2223  *
2224  */
2225 static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
2226 {
2227         struct sockaddr *addr = args->dstaddr;
2228         struct rpc_xprt *xprt;
2229         struct sock_xprt *transport;
2230
2231         xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries);
2232         if (IS_ERR(xprt))
2233                 return xprt;
2234         transport = container_of(xprt, struct sock_xprt, xprt);
2235
2236         xprt->prot = IPPROTO_UDP;
2237         xprt->tsh_size = 0;
2238         /* XXX: header size can vary due to auth type, IPv6, etc. */
2239         xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
2240
2241         xprt->bind_timeout = XS_BIND_TO;
2242         xprt->connect_timeout = XS_UDP_CONN_TO;
2243         xprt->reestablish_timeout = XS_UDP_REEST_TO;
2244         xprt->idle_timeout = XS_IDLE_DISC_TO;
2245
2246         xprt->ops = &xs_udp_ops;
2247
2248         xprt->timeout = &xs_udp_default_timeout;
2249
2250         switch (addr->sa_family) {
2251         case AF_INET:
2252                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2253                         xprt_set_bound(xprt);
2254
2255                 INIT_DELAYED_WORK(&transport->connect_worker,
2256                                         xs_udp_connect_worker4);
2257                 xs_format_ipv4_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
2258                 break;
2259         case AF_INET6:
2260                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2261                         xprt_set_bound(xprt);
2262
2263                 INIT_DELAYED_WORK(&transport->connect_worker,
2264                                         xs_udp_connect_worker6);
2265                 xs_format_ipv6_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
2266                 break;
2267         default:
2268                 kfree(xprt);
2269                 return ERR_PTR(-EAFNOSUPPORT);
2270         }
2271
2272         dprintk("RPC:       set up transport to address %s\n",
2273                         xprt->address_strings[RPC_DISPLAY_ALL]);
2274
2275         if (try_module_get(THIS_MODULE))
2276                 return xprt;
2277
2278         kfree(xprt->slot);
2279         kfree(xprt);
2280         return ERR_PTR(-EINVAL);
2281 }
2282
2283 static const struct rpc_timeout xs_tcp_default_timeout = {
2284         .to_initval = 60 * HZ,
2285         .to_maxval = 60 * HZ,
2286         .to_retries = 2,
2287 };
2288
2289 /**
2290  * xs_setup_tcp - Set up transport to use a TCP socket
2291  * @args: rpc transport creation arguments
2292  *
2293  */
2294 static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
2295 {
2296         struct sockaddr *addr = args->dstaddr;
2297         struct rpc_xprt *xprt;
2298         struct sock_xprt *transport;
2299
2300         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries);
2301         if (IS_ERR(xprt))
2302                 return xprt;
2303         transport = container_of(xprt, struct sock_xprt, xprt);
2304
2305         xprt->prot = IPPROTO_TCP;
2306         xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2307         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2308
2309         xprt->bind_timeout = XS_BIND_TO;
2310         xprt->connect_timeout = XS_TCP_CONN_TO;
2311         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2312         xprt->idle_timeout = XS_IDLE_DISC_TO;
2313
2314         xprt->ops = &xs_tcp_ops;
2315         xprt->timeout = &xs_tcp_default_timeout;
2316
2317         switch (addr->sa_family) {
2318         case AF_INET:
2319                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2320                         xprt_set_bound(xprt);
2321
2322                 INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_connect_worker4);
2323                 xs_format_ipv4_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
2324                 break;
2325         case AF_INET6:
2326                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2327                         xprt_set_bound(xprt);
2328
2329                 INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_connect_worker6);
2330                 xs_format_ipv6_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
2331                 break;
2332         default:
2333                 kfree(xprt);
2334                 return ERR_PTR(-EAFNOSUPPORT);
2335         }
2336
2337         dprintk("RPC:       set up transport to address %s\n",
2338                         xprt->address_strings[RPC_DISPLAY_ALL]);
2339
2340         if (try_module_get(THIS_MODULE))
2341                 return xprt;
2342
2343         kfree(xprt->slot);
2344         kfree(xprt);
2345         return ERR_PTR(-EINVAL);
2346 }
2347
2348 static struct xprt_class        xs_udp_transport = {
2349         .list           = LIST_HEAD_INIT(xs_udp_transport.list),
2350         .name           = "udp",
2351         .owner          = THIS_MODULE,
2352         .ident          = IPPROTO_UDP,
2353         .setup          = xs_setup_udp,
2354 };
2355
2356 static struct xprt_class        xs_tcp_transport = {
2357         .list           = LIST_HEAD_INIT(xs_tcp_transport.list),
2358         .name           = "tcp",
2359         .owner          = THIS_MODULE,
2360         .ident          = IPPROTO_TCP,
2361         .setup          = xs_setup_tcp,
2362 };
2363
2364 /**
2365  * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
2366  *
2367  */
2368 int init_socket_xprt(void)
2369 {
2370 #ifdef RPC_DEBUG
2371         if (!sunrpc_table_header)
2372                 sunrpc_table_header = register_sysctl_table(sunrpc_table);
2373 #endif
2374
2375         xprt_register_transport(&xs_udp_transport);
2376         xprt_register_transport(&xs_tcp_transport);
2377
2378         return 0;
2379 }
2380
2381 /**
2382  * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
2383  *
2384  */
2385 void cleanup_socket_xprt(void)
2386 {
2387 #ifdef RPC_DEBUG
2388         if (sunrpc_table_header) {
2389                 unregister_sysctl_table(sunrpc_table_header);
2390                 sunrpc_table_header = NULL;
2391         }
2392 #endif
2393
2394         xprt_unregister_transport(&xs_udp_transport);
2395         xprt_unregister_transport(&xs_tcp_transport);
2396 }