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