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 Software <alan@redhat.com>
   7  * TCP send fixes (C) 1998 Red Hat Software <alan@redhat.com>
   8  * TCP NFS related read + write fixes
   9  *  (C) 1999 Dave Airlie, University of Limerick, Ireland <airlied@linux.ie>
  10  *
  11  * Rewrite of larges part of the code in order to stabilize TCP stuff.
  12  * Fix behaviour when socket buffer is full.
  13  *  (C) 1999 Trond Myklebust <trond.myklebust@fys.uio.no>
  14  *
  15  * IP socket transport implementation, (C) 2005 Chuck Lever <cel@netapp.com>
  16  */
  17
  18 #include <linux/types.h>
  19 #include <linux/slab.h>
  20 #include <linux/capability.h>
  21 #include <linux/sched.h>
  22 #include <linux/pagemap.h>
  23 #include <linux/errno.h>
  24 #include <linux/socket.h>
  25 #include <linux/in.h>
  26 #include <linux/net.h>
  27 #include <linux/mm.h>
  28 #include <linux/udp.h>
  29 #include <linux/tcp.h>
  30 #include <linux/sunrpc/clnt.h>
  31 #include <linux/sunrpc/sched.h>
  32 #include <linux/file.h>
  33
  34 #include <net/sock.h>
  35 #include <net/checksum.h>
  36 #include <net/udp.h>
  37 #include <net/tcp.h>
  38
  39 /*
  40  * xprtsock tunables
  41  */
  42 unsigned int xprt_udp_slot_table_entries = RPC_DEF_SLOT_TABLE;
  43 unsigned int xprt_tcp_slot_table_entries = RPC_DEF_SLOT_TABLE;
  44
  45 unsigned int xprt_min_resvport = RPC_DEF_MIN_RESVPORT;
  46 unsigned int xprt_max_resvport = RPC_DEF_MAX_RESVPORT;
  47
  48 /*
  49  * How many times to try sending a request on a socket before waiting
  50  * for the socket buffer to clear.
  51  */
  52 #define XS_SENDMSG_RETRY        (10U)
  53
  54 /*
  55  * Time out for an RPC UDP socket connect.  UDP socket connects are
  56  * synchronous, but we set a timeout anyway in case of resource
  57  * exhaustion on the local host.
  58  */
  59 #define XS_UDP_CONN_TO          (5U * HZ)
  60
  61 /*
  62  * Wait duration for an RPC TCP connection to be established.  Solaris
  63  * NFS over TCP uses 60 seconds, for example, which is in line with how
  64  * long a server takes to reboot.
  65  */
  66 #define XS_TCP_CONN_TO          (60U * HZ)
  67
  68 /*
  69  * Wait duration for a reply from the RPC portmapper.
  70  */
  71 #define XS_BIND_TO              (60U * HZ)
  72
  73 /*
  74  * Delay if a UDP socket connect error occurs.  This is most likely some
  75  * kind of resource problem on the local host.
  76  */
  77 #define XS_UDP_REEST_TO         (2U * HZ)
  78
  79 /*
  80  * The reestablish timeout allows clients to delay for a bit before attempting
  81  * to reconnect to a server that just dropped our connection.
  82  *
  83  * We implement an exponential backoff when trying to reestablish a TCP
  84  * transport connection with the server.  Some servers like to drop a TCP
  85  * connection when they are overworked, so we start with a short timeout and
  86  * increase over time if the server is down or not responding.
  87  */
  88 #define XS_TCP_INIT_REEST_TO    (3U * HZ)
  89 #define XS_TCP_MAX_REEST_TO     (5U * 60 * HZ)
  90
  91 /*
  92  * TCP idle timeout; client drops the transport socket if it is idle
  93  * for this long.  Note that we also timeout UDP sockets to prevent
  94  * holding port numbers when there is no RPC traffic.
  95  */
  96 #define XS_IDLE_DISC_TO         (5U * 60 * HZ)
  97
  98 #ifdef RPC_DEBUG
  99 # undef  RPC_DEBUG_DATA
 100 # define RPCDBG_FACILITY        RPCDBG_TRANS
 101 #endif
 102
 103 #ifdef RPC_DEBUG_DATA
 104 static void xs_pktdump(char *msg, u32 *packet, unsigned int count)
 105 {
 106         u8 *buf = (u8 *) packet;
 107         int j;
 108
 109         dprintk("RPC:      %s\n", msg);
 110         for (j = 0; j < count && j < 128; j += 4) {
 111                 if (!(j & 31)) {
 112                         if (j)
 113                                 dprintk("\n");
 114                         dprintk("0x%04x ", j);
 115                 }
 116                 dprintk("%02x%02x%02x%02x ",
 117                         buf[j], buf[j+1], buf[j+2], buf[j+3]);
 118         }
 119         dprintk("\n");
 120 }
 121 #else
 122 static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count)
 123 {
 124         /* NOP */
 125 }
 126 #endif
 127
 128 struct sock_xprt {
 129         struct rpc_xprt         xprt;
 130
 131         /*
 132          * Network layer
 133          */
 134         struct socket *         sock;
 135         struct sock *           inet;
 136
 137         /*
 138          * State of TCP reply receive
 139          */
 140         __be32                  tcp_fraghdr,
 141                                 tcp_xid;
 142
 143         u32                     tcp_offset,
 144                                 tcp_reclen;
 145
 146         unsigned long           tcp_copied,
 147                                 tcp_flags;
 148
 149         /*
 150          * Connection of transports
 151          */
 152         struct work_struct      connect_worker;
 153         unsigned short          port;
 154
 155         /*
 156          * UDP socket buffer size parameters
 157          */
 158         size_t                  rcvsize,
 159                                 sndsize;
 160
 161         /*
 162          * Saved socket callback addresses
 163          */
 164         void                    (*old_data_ready)(struct sock *, int);
 165         void                    (*old_state_change)(struct sock *);
 166         void                    (*old_write_space)(struct sock *);
 167 };
 168
 169 /*
 170  * TCP receive state flags
 171  */
 172 #define TCP_RCV_LAST_FRAG       (1UL << 0)
 173 #define TCP_RCV_COPY_FRAGHDR    (1UL << 1)
 174 #define TCP_RCV_COPY_XID        (1UL << 2)
 175 #define TCP_RCV_COPY_DATA       (1UL << 3)
 176
 177 static void xs_format_peer_addresses(struct rpc_xprt *xprt)
 178 {
 179         struct sockaddr_in *addr = (struct sockaddr_in *) &xprt->addr;
 180         char *buf;
 181
 182         buf = kzalloc(20, GFP_KERNEL);
 183         if (buf) {
 184                 snprintf(buf, 20, "%u.%u.%u.%u",
 185                                 NIPQUAD(addr->sin_addr.s_addr));
 186         }
 187         xprt->address_strings[RPC_DISPLAY_ADDR] = buf;
 188
 189         buf = kzalloc(8, GFP_KERNEL);
 190         if (buf) {
 191                 snprintf(buf, 8, "%u",
 192                                 ntohs(addr->sin_port));
 193         }
 194         xprt->address_strings[RPC_DISPLAY_PORT] = buf;
 195
 196         if (xprt->prot == IPPROTO_UDP)
 197                 xprt->address_strings[RPC_DISPLAY_PROTO] = "udp";
 198         else
 199                 xprt->address_strings[RPC_DISPLAY_PROTO] = "tcp";
 200
 201         buf = kzalloc(48, GFP_KERNEL);
 202         if (buf) {
 203                 snprintf(buf, 48, "addr=%u.%u.%u.%u port=%u proto=%s",
 204                         NIPQUAD(addr->sin_addr.s_addr),
 205                         ntohs(addr->sin_port),
 206                         xprt->prot == IPPROTO_UDP ? "udp" : "tcp");
 207         }
 208         xprt->address_strings[RPC_DISPLAY_ALL] = buf;
 209 }
 210
 211 static void xs_free_peer_addresses(struct rpc_xprt *xprt)
 212 {
 213         kfree(xprt->address_strings[RPC_DISPLAY_ADDR]);
 214         kfree(xprt->address_strings[RPC_DISPLAY_PORT]);
 215         kfree(xprt->address_strings[RPC_DISPLAY_ALL]);
 216 }
 217
 218 #define XS_SENDMSG_FLAGS        (MSG_DONTWAIT | MSG_NOSIGNAL)
 219
 220 static int xs_send_kvec(struct socket *sock, struct sockaddr *addr, int addrlen, struct kvec *vec, unsigned int base, int more)
 221 {
 222         struct msghdr msg = {
 223                 .msg_name       = addr,
 224                 .msg_namelen    = addrlen,
 225                 .msg_flags      = XS_SENDMSG_FLAGS | (more ? MSG_MORE : 0),
 226         };
 227         struct kvec iov = {
 228                 .iov_base       = vec->iov_base + base,
 229                 .iov_len        = vec->iov_len - base,
 230         };
 231
 232         if (iov.iov_len != 0)
 233                 return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
 234         return kernel_sendmsg(sock, &msg, NULL, 0, 0);
 235 }
 236
 237 static int xs_send_pagedata(struct socket *sock, struct xdr_buf *xdr, unsigned int base, int more)
 238 {
 239         struct page **ppage;
 240         unsigned int remainder;
 241         int err, sent = 0;
 242
 243         remainder = xdr->page_len - base;
 244         base += xdr->page_base;
 245         ppage = xdr->pages + (base >> PAGE_SHIFT);
 246         base &= ~PAGE_MASK;
 247         for(;;) {
 248                 unsigned int len = min_t(unsigned int, PAGE_SIZE - base, remainder);
 249                 int flags = XS_SENDMSG_FLAGS;
 250
 251                 remainder -= len;
 252                 if (remainder != 0 || more)
 253                         flags |= MSG_MORE;
 254                 err = sock->ops->sendpage(sock, *ppage, base, len, flags);
 255                 if (remainder == 0 || err != len)
 256                         break;
 257                 sent += err;
 258                 ppage++;
 259                 base = 0;
 260         }
 261         if (sent == 0)
 262                 return err;
 263         if (err > 0)
 264                 sent += err;
 265         return sent;
 266 }
 267
 268 /**
 269  * xs_sendpages - write pages directly to a socket
 270  * @sock: socket to send on
 271  * @addr: UDP only -- address of destination
 272  * @addrlen: UDP only -- length of destination address
 273  * @xdr: buffer containing this request
 274  * @base: starting position in the buffer
 275  *
 276  */
 277 static int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base)
 278 {
 279         unsigned int remainder = xdr->len - base;
 280         int err, sent = 0;
 281
 282         if (unlikely(!sock))
 283                 return -ENOTCONN;
 284
 285         clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags);
 286         if (base != 0) {
 287                 addr = NULL;
 288                 addrlen = 0;
 289         }
 290
 291         if (base < xdr->head[0].iov_len || addr != NULL) {
 292                 unsigned int len = xdr->head[0].iov_len - base;
 293                 remainder -= len;
 294                 err = xs_send_kvec(sock, addr, addrlen, &xdr->head[0], base, remainder != 0);
 295                 if (remainder == 0 || err != len)
 296                         goto out;
 297                 sent += err;
 298                 base = 0;
 299         } else
 300                 base -= xdr->head[0].iov_len;
 301
 302         if (base < xdr->page_len) {
 303                 unsigned int len = xdr->page_len - base;
 304                 remainder -= len;
 305                 err = xs_send_pagedata(sock, xdr, base, remainder != 0);
 306                 if (remainder == 0 || err != len)
 307                         goto out;
 308                 sent += err;
 309                 base = 0;
 310         } else
 311                 base -= xdr->page_len;
 312
 313         if (base >= xdr->tail[0].iov_len)
 314                 return sent;
 315         err = xs_send_kvec(sock, NULL, 0, &xdr->tail[0], base, 0);
 316 out:
 317         if (sent == 0)
 318                 return err;
 319         if (err > 0)
 320                 sent += err;
 321         return sent;
 322 }
 323
 324 /**
 325  * xs_nospace - place task on wait queue if transmit was incomplete
 326  * @task: task to put to sleep
 327  *
 328  */
 329 static void xs_nospace(struct rpc_task *task)
 330 {
 331         struct rpc_rqst *req = task->tk_rqstp;
 332         struct rpc_xprt *xprt = req->rq_xprt;
 333         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 334
 335         dprintk("RPC: %4d xmit incomplete (%u left of %u)\n",
 336                         task->tk_pid, req->rq_slen - req->rq_bytes_sent,
 337                         req->rq_slen);
 338
 339         if (test_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags)) {
 340                 /* Protect against races with write_space */
 341                 spin_lock_bh(&xprt->transport_lock);
 342
 343                 /* Don't race with disconnect */
 344                 if (!xprt_connected(xprt))
 345                         task->tk_status = -ENOTCONN;
 346                 else if (test_bit(SOCK_NOSPACE, &transport->sock->flags))
 347                         xprt_wait_for_buffer_space(task);
 348
 349                 spin_unlock_bh(&xprt->transport_lock);
 350         } else
 351                 /* Keep holding the socket if it is blocked */
 352                 rpc_delay(task, HZ>>4);
 353 }
 354
 355 /**
 356  * xs_udp_send_request - write an RPC request to a UDP socket
 357  * @task: address of RPC task that manages the state of an RPC request
 358  *
 359  * Return values:
 360  *        0:    The request has been sent
 361  *   EAGAIN:    The socket was blocked, please call again later to
 362  *              complete the request
 363  * ENOTCONN:    Caller needs to invoke connect logic then call again
 364  *    other:    Some other error occured, the request was not sent
 365  */
 366 static int xs_udp_send_request(struct rpc_task *task)
 367 {
 368         struct rpc_rqst *req = task->tk_rqstp;
 369         struct rpc_xprt *xprt = req->rq_xprt;
 370         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 371         struct xdr_buf *xdr = &req->rq_snd_buf;
 372         int status;
 373
 374         xs_pktdump("packet data:",
 375                                 req->rq_svec->iov_base,
 376                                 req->rq_svec->iov_len);
 377
 378         req->rq_xtime = jiffies;
 379         status = xs_sendpages(transport->sock,
 380                               (struct sockaddr *) &xprt->addr,
 381                               xprt->addrlen, xdr,
 382                               req->rq_bytes_sent);
 383
 384         dprintk("RPC:      xs_udp_send_request(%u) = %d\n",
 385                         xdr->len - req->rq_bytes_sent, status);
 386
 387         if (likely(status >= (int) req->rq_slen))
 388                 return 0;
 389
 390         /* Still some bytes left; set up for a retry later. */
 391         if (status > 0)
 392                 status = -EAGAIN;
 393
 394         switch (status) {
 395         case -ENETUNREACH:
 396         case -EPIPE:
 397         case -ECONNREFUSED:
 398                 /* When the server has died, an ICMP port unreachable message
 399                  * prompts ECONNREFUSED. */
 400                 break;
 401         case -EAGAIN:
 402                 xs_nospace(task);
 403                 break;
 404         default:
 405                 dprintk("RPC:      sendmsg returned unrecognized error %d\n",
 406                         -status);
 407                 break;
 408         }
 409
 410         return status;
 411 }
 412
 413 static inline void xs_encode_tcp_record_marker(struct xdr_buf *buf)
 414 {
 415         u32 reclen = buf->len - sizeof(rpc_fraghdr);
 416         rpc_fraghdr *base = buf->head[0].iov_base;
 417         *base = htonl(RPC_LAST_STREAM_FRAGMENT | reclen);
 418 }
 419
 420 /**
 421  * xs_tcp_send_request - write an RPC request to a TCP socket
 422  * @task: address of RPC task that manages the state of an RPC request
 423  *
 424  * Return values:
 425  *        0:    The request has been sent
 426  *   EAGAIN:    The socket was blocked, please call again later to
 427  *              complete the request
 428  * ENOTCONN:    Caller needs to invoke connect logic then call again
 429  *    other:    Some other error occured, the request was not sent
 430  *
 431  * XXX: In the case of soft timeouts, should we eventually give up
 432  *      if sendmsg is not able to make progress?
 433  */
 434 static int xs_tcp_send_request(struct rpc_task *task)
 435 {
 436         struct rpc_rqst *req = task->tk_rqstp;
 437         struct rpc_xprt *xprt = req->rq_xprt;
 438         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 439         struct xdr_buf *xdr = &req->rq_snd_buf;
 440         int status, retry = 0;
 441
 442         xs_encode_tcp_record_marker(&req->rq_snd_buf);
 443
 444         xs_pktdump("packet data:",
 445                                 req->rq_svec->iov_base,
 446                                 req->rq_svec->iov_len);
 447
 448         /* Continue transmitting the packet/record. We must be careful
 449          * to cope with writespace callbacks arriving _after_ we have
 450          * called sendmsg(). */
 451         while (1) {
 452                 req->rq_xtime = jiffies;
 453                 status = xs_sendpages(transport->sock,
 454                                         NULL, 0, xdr, req->rq_bytes_sent);
 455
 456                 dprintk("RPC:      xs_tcp_send_request(%u) = %d\n",
 457                                 xdr->len - req->rq_bytes_sent, status);
 458
 459                 if (unlikely(status < 0))
 460                         break;
 461
 462                 /* If we've sent the entire packet, immediately
 463                  * reset the count of bytes sent. */
 464                 req->rq_bytes_sent += status;
 465                 task->tk_bytes_sent += status;
 466                 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
 467                         req->rq_bytes_sent = 0;
 468                         return 0;
 469                 }
 470
 471                 status = -EAGAIN;
 472                 if (retry++ > XS_SENDMSG_RETRY)
 473                         break;
 474         }
 475
 476         switch (status) {
 477         case -EAGAIN:
 478                 xs_nospace(task);
 479                 break;
 480         case -ECONNREFUSED:
 481         case -ECONNRESET:
 482         case -ENOTCONN:
 483         case -EPIPE:
 484                 status = -ENOTCONN;
 485                 break;
 486         default:
 487                 dprintk("RPC:      sendmsg returned unrecognized error %d\n",
 488                         -status);
 489                 xprt_disconnect(xprt);
 490                 break;
 491         }
 492
 493         return status;
 494 }
 495
 496 /**
 497  * xs_tcp_release_xprt - clean up after a tcp transmission
 498  * @xprt: transport
 499  * @task: rpc task
 500  *
 501  * This cleans up if an error causes us to abort the transmission of a request.
 502  * In this case, the socket may need to be reset in order to avoid confusing
 503  * the server.
 504  */
 505 static void xs_tcp_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
 506 {
 507         struct rpc_rqst *req;
 508
 509         if (task != xprt->snd_task)
 510                 return;
 511         if (task == NULL)
 512                 goto out_release;
 513         req = task->tk_rqstp;
 514         if (req->rq_bytes_sent == 0)
 515                 goto out_release;
 516         if (req->rq_bytes_sent == req->rq_snd_buf.len)
 517                 goto out_release;
 518         set_bit(XPRT_CLOSE_WAIT, &task->tk_xprt->state);
 519 out_release:
 520         xprt_release_xprt(xprt, task);
 521 }
 522
 523 /**
 524  * xs_close - close a socket
 525  * @xprt: transport
 526  *
 527  * This is used when all requests are complete; ie, no DRC state remains
 528  * on the server we want to save.
 529  */
 530 static void xs_close(struct rpc_xprt *xprt)
 531 {
 532         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 533         struct socket *sock = transport->sock;
 534         struct sock *sk = transport->inet;
 535
 536         if (!sk)
 537                 goto clear_close_wait;
 538
 539         dprintk("RPC:      xs_close xprt %p\n", xprt);
 540
 541         write_lock_bh(&sk->sk_callback_lock);
 542         transport->inet = NULL;
 543         transport->sock = NULL;
 544
 545         sk->sk_user_data = NULL;
 546         sk->sk_data_ready = transport->old_data_ready;
 547         sk->sk_state_change = transport->old_state_change;
 548         sk->sk_write_space = transport->old_write_space;
 549         write_unlock_bh(&sk->sk_callback_lock);
 550
 551         sk->sk_no_check = 0;
 552
 553         sock_release(sock);
 554 clear_close_wait:
 555         smp_mb__before_clear_bit();
 556         clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
 557         smp_mb__after_clear_bit();
 558 }
 559
 560 /**
 561  * xs_destroy - prepare to shutdown a transport
 562  * @xprt: doomed transport
 563  *
 564  */
 565 static void xs_destroy(struct rpc_xprt *xprt)
 566 {
 567         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 568
 569         dprintk("RPC:      xs_destroy xprt %p\n", xprt);
 570
 571         cancel_delayed_work(&transport->connect_worker);
 572         flush_scheduled_work();
 573
 574         xprt_disconnect(xprt);
 575         xs_close(xprt);
 576         xs_free_peer_addresses(xprt);
 577         kfree(xprt->slot);
 578         kfree(xprt);
 579 }
 580
 581 static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
 582 {
 583         return (struct rpc_xprt *) sk->sk_user_data;
 584 }
 585
 586 /**
 587  * xs_udp_data_ready - "data ready" callback for UDP sockets
 588  * @sk: socket with data to read
 589  * @len: how much data to read
 590  *
 591  */
 592 static void xs_udp_data_ready(struct sock *sk, int len)
 593 {
 594         struct rpc_task *task;
 595         struct rpc_xprt *xprt;
 596         struct rpc_rqst *rovr;
 597         struct sk_buff *skb;
 598         int err, repsize, copied;
 599         u32 _xid;
 600         __be32 *xp;
 601
 602         read_lock(&sk->sk_callback_lock);
 603         dprintk("RPC:      xs_udp_data_ready...\n");
 604         if (!(xprt = xprt_from_sock(sk)))
 605                 goto out;
 606
 607         if ((skb = skb_recv_datagram(sk, 0, 1, &err)) == NULL)
 608                 goto out;
 609
 610         if (xprt->shutdown)
 611                 goto dropit;
 612
 613         repsize = skb->len - sizeof(struct udphdr);
 614         if (repsize < 4) {
 615                 dprintk("RPC:      impossible RPC reply size %d!\n", repsize);
 616                 goto dropit;
 617         }
 618
 619         /* Copy the XID from the skb... */
 620         xp = skb_header_pointer(skb, sizeof(struct udphdr),
 621                                 sizeof(_xid), &_xid);
 622         if (xp == NULL)
 623                 goto dropit;
 624
 625         /* Look up and lock the request corresponding to the given XID */
 626         spin_lock(&xprt->transport_lock);
 627         rovr = xprt_lookup_rqst(xprt, *xp);
 628         if (!rovr)
 629                 goto out_unlock;
 630         task = rovr->rq_task;
 631
 632         if ((copied = rovr->rq_private_buf.buflen) > repsize)
 633                 copied = repsize;
 634
 635         /* Suck it into the iovec, verify checksum if not done by hw. */
 636         if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb))
 637                 goto out_unlock;
 638
 639         /* Something worked... */
 640         dst_confirm(skb->dst);
 641
 642         xprt_adjust_cwnd(task, copied);
 643         xprt_update_rtt(task);
 644         xprt_complete_rqst(task, copied);
 645
 646  out_unlock:
 647         spin_unlock(&xprt->transport_lock);
 648  dropit:
 649         skb_free_datagram(sk, skb);
 650  out:
 651         read_unlock(&sk->sk_callback_lock);
 652 }
 653
 654 static inline size_t xs_tcp_copy_data(skb_reader_t *desc, void *p, size_t len)
 655 {
 656         if (len > desc->count)
 657                 len = desc->count;
 658         if (skb_copy_bits(desc->skb, desc->offset, p, len)) {
 659                 dprintk("RPC:      failed to copy %zu bytes from skb. %zu bytes remain\n",
 660                                 len, desc->count);
 661                 return 0;
 662         }
 663         desc->offset += len;
 664         desc->count -= len;
 665         dprintk("RPC:      copied %zu bytes from skb. %zu bytes remain\n",
 666                         len, desc->count);
 667         return len;
 668 }
 669
 670 static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, skb_reader_t *desc)
 671 {
 672         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 673         size_t len, used;
 674         char *p;
 675
 676         p = ((char *) &transport->tcp_fraghdr) + transport->tcp_offset;
 677         len = sizeof(transport->tcp_fraghdr) - transport->tcp_offset;
 678         used = xs_tcp_copy_data(desc, p, len);
 679         transport->tcp_offset += used;
 680         if (used != len)
 681                 return;
 682
 683         transport->tcp_reclen = ntohl(transport->tcp_fraghdr);
 684         if (transport->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
 685                 transport->tcp_flags |= TCP_RCV_LAST_FRAG;
 686         else
 687                 transport->tcp_flags &= ~TCP_RCV_LAST_FRAG;
 688         transport->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
 689
 690         transport->tcp_flags &= ~TCP_RCV_COPY_FRAGHDR;
 691         transport->tcp_offset = 0;
 692
 693         /* Sanity check of the record length */
 694         if (unlikely(transport->tcp_reclen < 4)) {
 695                 dprintk("RPC:      invalid TCP record fragment length\n");
 696                 xprt_disconnect(xprt);
 697                 return;
 698         }
 699         dprintk("RPC:      reading TCP record fragment of length %d\n",
 700                         transport->tcp_reclen);
 701 }
 702
 703 static void xs_tcp_check_fraghdr(struct sock_xprt *transport)
 704 {
 705         if (transport->tcp_offset == transport->tcp_reclen) {
 706                 transport->tcp_flags |= TCP_RCV_COPY_FRAGHDR;
 707                 transport->tcp_offset = 0;
 708                 if (transport->tcp_flags & TCP_RCV_LAST_FRAG) {
 709                         transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
 710                         transport->tcp_flags |= TCP_RCV_COPY_XID;
 711                         transport->tcp_copied = 0;
 712                 }
 713         }
 714 }
 715
 716 static inline void xs_tcp_read_xid(struct sock_xprt *transport, skb_reader_t *desc)
 717 {
 718         size_t len, used;
 719         char *p;
 720
 721         len = sizeof(transport->tcp_xid) - transport->tcp_offset;
 722         dprintk("RPC:      reading XID (%Zu bytes)\n", len);
 723         p = ((char *) &transport->tcp_xid) + transport->tcp_offset;
 724         used = xs_tcp_copy_data(desc, p, len);
 725         transport->tcp_offset += used;
 726         if (used != len)
 727                 return;
 728         transport->tcp_flags &= ~TCP_RCV_COPY_XID;
 729         transport->tcp_flags |= TCP_RCV_COPY_DATA;
 730         transport->tcp_copied = 4;
 731         dprintk("RPC:      reading reply for XID %08x\n",
 732                         ntohl(transport->tcp_xid));
 733         xs_tcp_check_fraghdr(transport);
 734 }
 735
 736 static inline void xs_tcp_read_request(struct rpc_xprt *xprt, skb_reader_t *desc)
 737 {
 738         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 739         struct rpc_rqst *req;
 740         struct xdr_buf *rcvbuf;
 741         size_t len;
 742         ssize_t r;
 743
 744         /* Find and lock the request corresponding to this xid */
 745         spin_lock(&xprt->transport_lock);
 746         req = xprt_lookup_rqst(xprt, transport->tcp_xid);
 747         if (!req) {
 748                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
 749                 dprintk("RPC:      XID %08x request not found!\n",
 750                                 ntohl(transport->tcp_xid));
 751                 spin_unlock(&xprt->transport_lock);
 752                 return;
 753         }
 754
 755         rcvbuf = &req->rq_private_buf;
 756         len = desc->count;
 757         if (len > transport->tcp_reclen - transport->tcp_offset) {
 758                 skb_reader_t my_desc;
 759
 760                 len = transport->tcp_reclen - transport->tcp_offset;
 761                 memcpy(&my_desc, desc, sizeof(my_desc));
 762                 my_desc.count = len;
 763                 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
 764                                           &my_desc, xs_tcp_copy_data);
 765                 desc->count -= r;
 766                 desc->offset += r;
 767         } else
 768                 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
 769                                           desc, xs_tcp_copy_data);
 770
 771         if (r > 0) {
 772                 transport->tcp_copied += r;
 773                 transport->tcp_offset += r;
 774         }
 775         if (r != len) {
 776                 /* Error when copying to the receive buffer,
 777                  * usually because we weren't able to allocate
 778                  * additional buffer pages. All we can do now
 779                  * is turn off TCP_RCV_COPY_DATA, so the request
 780                  * will not receive any additional updates,
 781                  * and time out.
 782                  * Any remaining data from this record will
 783                  * be discarded.
 784                  */
 785                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
 786                 dprintk("RPC:      XID %08x truncated request\n",
 787                                 ntohl(transport->tcp_xid));
 788                 dprintk("RPC:      xprt = %p, tcp_copied = %lu, tcp_offset = %u, tcp_reclen = %u\n",
 789                                 xprt, transport->tcp_copied, transport->tcp_offset,
 790                                         transport->tcp_reclen);
 791                 goto out;
 792         }
 793
 794         dprintk("RPC:      XID %08x read %Zd bytes\n",
 795                         ntohl(transport->tcp_xid), r);
 796         dprintk("RPC:      xprt = %p, tcp_copied = %lu, tcp_offset = %u, tcp_reclen = %u\n",
 797                         xprt, transport->tcp_copied, transport->tcp_offset,
 798                                 transport->tcp_reclen);
 799
 800         if (transport->tcp_copied == req->rq_private_buf.buflen)
 801                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
 802         else if (transport->tcp_offset == transport->tcp_reclen) {
 803                 if (transport->tcp_flags & TCP_RCV_LAST_FRAG)
 804                         transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
 805         }
 806
 807 out:
 808         if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
 809                 xprt_complete_rqst(req->rq_task, transport->tcp_copied);
 810         spin_unlock(&xprt->transport_lock);
 811         xs_tcp_check_fraghdr(transport);
 812 }
 813
 814 static inline void xs_tcp_read_discard(struct sock_xprt *transport, skb_reader_t *desc)
 815 {
 816         size_t len;
 817
 818         len = transport->tcp_reclen - transport->tcp_offset;
 819         if (len > desc->count)
 820                 len = desc->count;
 821         desc->count -= len;
 822         desc->offset += len;
 823         transport->tcp_offset += len;
 824         dprintk("RPC:      discarded %Zu bytes\n", len);
 825         xs_tcp_check_fraghdr(transport);
 826 }
 827
 828 static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
 829 {
 830         struct rpc_xprt *xprt = rd_desc->arg.data;
 831         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 832         skb_reader_t desc = {
 833                 .skb    = skb,
 834                 .offset = offset,
 835                 .count  = len,
 836         };
 837
 838         dprintk("RPC:      xs_tcp_data_recv started\n");
 839         do {
 840                 /* Read in a new fragment marker if necessary */
 841                 /* Can we ever really expect to get completely empty fragments? */
 842                 if (transport->tcp_flags & TCP_RCV_COPY_FRAGHDR) {
 843                         xs_tcp_read_fraghdr(xprt, &desc);
 844                         continue;
 845                 }
 846                 /* Read in the xid if necessary */
 847                 if (transport->tcp_flags & TCP_RCV_COPY_XID) {
 848                         xs_tcp_read_xid(transport, &desc);
 849                         continue;
 850                 }
 851                 /* Read in the request data */
 852                 if (transport->tcp_flags & TCP_RCV_COPY_DATA) {
 853                         xs_tcp_read_request(xprt, &desc);
 854                         continue;
 855                 }
 856                 /* Skip over any trailing bytes on short reads */
 857                 xs_tcp_read_discard(transport, &desc);
 858         } while (desc.count);
 859         dprintk("RPC:      xs_tcp_data_recv done\n");
 860         return len - desc.count;
 861 }
 862
 863 /**
 864  * xs_tcp_data_ready - "data ready" callback for TCP sockets
 865  * @sk: socket with data to read
 866  * @bytes: how much data to read
 867  *
 868  */
 869 static void xs_tcp_data_ready(struct sock *sk, int bytes)
 870 {
 871         struct rpc_xprt *xprt;
 872         read_descriptor_t rd_desc;
 873
 874         read_lock(&sk->sk_callback_lock);
 875         dprintk("RPC:      xs_tcp_data_ready...\n");
 876         if (!(xprt = xprt_from_sock(sk)))
 877                 goto out;
 878         if (xprt->shutdown)
 879                 goto out;
 880
 881         /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
 882         rd_desc.arg.data = xprt;
 883         rd_desc.count = 65536;
 884         tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
 885 out:
 886         read_unlock(&sk->sk_callback_lock);
 887 }
 888
 889 /**
 890  * xs_tcp_state_change - callback to handle TCP socket state changes
 891  * @sk: socket whose state has changed
 892  *
 893  */
 894 static void xs_tcp_state_change(struct sock *sk)
 895 {
 896         struct rpc_xprt *xprt;
 897
 898         read_lock(&sk->sk_callback_lock);
 899         if (!(xprt = xprt_from_sock(sk)))
 900                 goto out;
 901         dprintk("RPC:      xs_tcp_state_change client %p...\n", xprt);
 902         dprintk("RPC:      state %x conn %d dead %d zapped %d\n",
 903                                 sk->sk_state, xprt_connected(xprt),
 904                                 sock_flag(sk, SOCK_DEAD),
 905                                 sock_flag(sk, SOCK_ZAPPED));
 906
 907         switch (sk->sk_state) {
 908         case TCP_ESTABLISHED:
 909                 spin_lock_bh(&xprt->transport_lock);
 910                 if (!xprt_test_and_set_connected(xprt)) {
 911                         struct sock_xprt *transport = container_of(xprt,
 912                                         struct sock_xprt, xprt);
 913
 914                         /* Reset TCP record info */
 915                         transport->tcp_offset = 0;
 916                         transport->tcp_reclen = 0;
 917                         transport->tcp_copied = 0;
 918                         transport->tcp_flags =
 919                                 TCP_RCV_COPY_FRAGHDR | TCP_RCV_COPY_XID;
 920
 921                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
 922                         xprt_wake_pending_tasks(xprt, 0);
 923                 }
 924                 spin_unlock_bh(&xprt->transport_lock);
 925                 break;
 926         case TCP_SYN_SENT:
 927         case TCP_SYN_RECV:
 928                 break;
 929         case TCP_CLOSE_WAIT:
 930                 /* Try to schedule an autoclose RPC calls */
 931                 set_bit(XPRT_CLOSE_WAIT, &xprt->state);
 932                 if (test_and_set_bit(XPRT_LOCKED, &xprt->state) == 0)
 933                         schedule_work(&xprt->task_cleanup);
 934         default:
 935                 xprt_disconnect(xprt);
 936         }
 937  out:
 938         read_unlock(&sk->sk_callback_lock);
 939 }
 940
 941 /**
 942  * xs_udp_write_space - callback invoked when socket buffer space
 943  *                             becomes available
 944  * @sk: socket whose state has changed
 945  *
 946  * Called when more output buffer space is available for this socket.
 947  * We try not to wake our writers until they can make "significant"
 948  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
 949  * with a bunch of small requests.
 950  */
 951 static void xs_udp_write_space(struct sock *sk)
 952 {
 953         read_lock(&sk->sk_callback_lock);
 954
 955         /* from net/core/sock.c:sock_def_write_space */
 956         if (sock_writeable(sk)) {
 957                 struct socket *sock;
 958                 struct rpc_xprt *xprt;
 959
 960                 if (unlikely(!(sock = sk->sk_socket)))
 961                         goto out;
 962                 if (unlikely(!(xprt = xprt_from_sock(sk))))
 963                         goto out;
 964                 if (unlikely(!test_and_clear_bit(SOCK_NOSPACE, &sock->flags)))
 965                         goto out;
 966
 967                 xprt_write_space(xprt);
 968         }
 969
 970  out:
 971         read_unlock(&sk->sk_callback_lock);
 972 }
 973
 974 /**
 975  * xs_tcp_write_space - callback invoked when socket buffer space
 976  *                             becomes available
 977  * @sk: socket whose state has changed
 978  *
 979  * Called when more output buffer space is available for this socket.
 980  * We try not to wake our writers until they can make "significant"
 981  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
 982  * with a bunch of small requests.
 983  */
 984 static void xs_tcp_write_space(struct sock *sk)
 985 {
 986         read_lock(&sk->sk_callback_lock);
 987
 988         /* from net/core/stream.c:sk_stream_write_space */
 989         if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk)) {
 990                 struct socket *sock;
 991                 struct rpc_xprt *xprt;
 992
 993                 if (unlikely(!(sock = sk->sk_socket)))
 994                         goto out;
 995                 if (unlikely(!(xprt = xprt_from_sock(sk))))
 996                         goto out;
 997                 if (unlikely(!test_and_clear_bit(SOCK_NOSPACE, &sock->flags)))
 998                         goto out;
 999
1000                 xprt_write_space(xprt);
1001         }
1002
1003  out:
1004         read_unlock(&sk->sk_callback_lock);
1005 }
1006
1007 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
1008 {
1009         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1010         struct sock *sk = transport->inet;
1011
1012         if (transport->rcvsize) {
1013                 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1014                 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
1015         }
1016         if (transport->sndsize) {
1017                 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1018                 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1019                 sk->sk_write_space(sk);
1020         }
1021 }
1022
1023 /**
1024  * xs_udp_set_buffer_size - set send and receive limits
1025  * @xprt: generic transport
1026  * @sndsize: requested size of send buffer, in bytes
1027  * @rcvsize: requested size of receive buffer, in bytes
1028  *
1029  * Set socket send and receive buffer size limits.
1030  */
1031 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1032 {
1033         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1034
1035         transport->sndsize = 0;
1036         if (sndsize)
1037                 transport->sndsize = sndsize + 1024;
1038         transport->rcvsize = 0;
1039         if (rcvsize)
1040                 transport->rcvsize = rcvsize + 1024;
1041
1042         xs_udp_do_set_buffer_size(xprt);
1043 }
1044
1045 /**
1046  * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1047  * @task: task that timed out
1048  *
1049  * Adjust the congestion window after a retransmit timeout has occurred.
1050  */
1051 static void xs_udp_timer(struct rpc_task *task)
1052 {
1053         xprt_adjust_cwnd(task, -ETIMEDOUT);
1054 }
1055
1056 static unsigned short xs_get_random_port(void)
1057 {
1058         unsigned short range = xprt_max_resvport - xprt_min_resvport;
1059         unsigned short rand = (unsigned short) net_random() % range;
1060         return rand + xprt_min_resvport;
1061 }
1062
1063 /**
1064  * xs_set_port - reset the port number in the remote endpoint address
1065  * @xprt: generic transport
1066  * @port: new port number
1067  *
1068  */
1069 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1070 {
1071         struct sockaddr_in *sap = (struct sockaddr_in *) &xprt->addr;
1072
1073         dprintk("RPC:      setting port for xprt %p to %u\n", xprt, port);
1074
1075         sap->sin_port = htons(port);
1076 }
1077
1078 static int xs_bindresvport(struct sock_xprt *transport, struct socket *sock)
1079 {
1080         struct sockaddr_in myaddr = {
1081                 .sin_family = AF_INET,
1082         };
1083         int err;
1084         unsigned short port = transport->port;
1085
1086         do {
1087                 myaddr.sin_port = htons(port);
1088                 err = kernel_bind(sock, (struct sockaddr *) &myaddr,
1089                                                 sizeof(myaddr));
1090                 if (err == 0) {
1091                         transport->port = port;
1092                         dprintk("RPC:      xs_bindresvport bound to port %u\n",
1093                                         port);
1094                         return 0;
1095                 }
1096                 if (port <= xprt_min_resvport)
1097                         port = xprt_max_resvport;
1098                 else
1099                         port--;
1100         } while (err == -EADDRINUSE && port != transport->port);
1101
1102         dprintk("RPC:      can't bind to reserved port (%d).\n", -err);
1103         return err;
1104 }
1105
1106 /**
1107  * xs_udp_connect_worker - set up a UDP socket
1108  * @args: RPC transport to connect
1109  *
1110  * Invoked by a work queue tasklet.
1111  */
1112 static void xs_udp_connect_worker(void *args)
1113 {
1114         struct sock_xprt *transport = (struct sock_xprt *)args;
1115         struct rpc_xprt *xprt = &transport->xprt;
1116         struct socket *sock = transport->sock;
1117         int err, status = -EIO;
1118
1119         if (xprt->shutdown || !xprt_bound(xprt))
1120                 goto out;
1121
1122         /* Start by resetting any existing state */
1123         xs_close(xprt);
1124
1125         if ((err = sock_create_kern(PF_INET, SOCK_DGRAM, IPPROTO_UDP, &sock)) < 0) {
1126                 dprintk("RPC:      can't create UDP transport socket (%d).\n", -err);
1127                 goto out;
1128         }
1129
1130         if (xprt->resvport && xs_bindresvport(transport, sock) < 0) {
1131                 sock_release(sock);
1132                 goto out;
1133         }
1134
1135         dprintk("RPC:      worker connecting xprt %p to address: %s\n",
1136                         xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1137
1138         if (!transport->inet) {
1139                 struct sock *sk = sock->sk;
1140
1141                 write_lock_bh(&sk->sk_callback_lock);
1142
1143                 sk->sk_user_data = xprt;
1144                 transport->old_data_ready = sk->sk_data_ready;
1145                 transport->old_state_change = sk->sk_state_change;
1146                 transport->old_write_space = sk->sk_write_space;
1147                 sk->sk_data_ready = xs_udp_data_ready;
1148                 sk->sk_write_space = xs_udp_write_space;
1149                 sk->sk_no_check = UDP_CSUM_NORCV;
1150                 sk->sk_allocation = GFP_ATOMIC;
1151
1152                 xprt_set_connected(xprt);
1153
1154                 /* Reset to new socket */
1155                 transport->sock = sock;
1156                 transport->inet = sk;
1157
1158                 write_unlock_bh(&sk->sk_callback_lock);
1159         }
1160         xs_udp_do_set_buffer_size(xprt);
1161         status = 0;
1162 out:
1163         xprt_wake_pending_tasks(xprt, status);
1164         xprt_clear_connecting(xprt);
1165 }
1166
1167 /*
1168  * We need to preserve the port number so the reply cache on the server can
1169  * find our cached RPC replies when we get around to reconnecting.
1170  */
1171 static void xs_tcp_reuse_connection(struct rpc_xprt *xprt)
1172 {
1173         int result;
1174         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1175         struct sockaddr any;
1176
1177         dprintk("RPC:      disconnecting xprt %p to reuse port\n", xprt);
1178
1179         /*
1180          * Disconnect the transport socket by doing a connect operation
1181          * with AF_UNSPEC.  This should return immediately...
1182          */
1183         memset(&any, 0, sizeof(any));
1184         any.sa_family = AF_UNSPEC;
1185         result = kernel_connect(transport->sock, &any, sizeof(any), 0);
1186         if (result)
1187                 dprintk("RPC:      AF_UNSPEC connect return code %d\n",
1188                                 result);
1189 }
1190
1191 /**
1192  * xs_tcp_connect_worker - connect a TCP socket to a remote endpoint
1193  * @args: RPC transport to connect
1194  *
1195  * Invoked by a work queue tasklet.
1196  */
1197 static void xs_tcp_connect_worker(void *args)
1198 {
1199         struct sock_xprt *transport = (struct sock_xprt *)args;
1200         struct rpc_xprt *xprt = &transport->xprt;
1201         struct socket *sock = transport->sock;
1202         int err, status = -EIO;
1203
1204         if (xprt->shutdown || !xprt_bound(xprt))
1205                 goto out;
1206
1207         if (!sock) {
1208                 /* start from scratch */
1209                 if ((err = sock_create_kern(PF_INET, SOCK_STREAM, IPPROTO_TCP, &sock)) < 0) {
1210                         dprintk("RPC:      can't create TCP transport socket (%d).\n", -err);
1211                         goto out;
1212                 }
1213
1214                 if (xprt->resvport && xs_bindresvport(transport, sock) < 0) {
1215                         sock_release(sock);
1216                         goto out;
1217                 }
1218         } else
1219                 /* "close" the socket, preserving the local port */
1220                 xs_tcp_reuse_connection(xprt);
1221
1222         dprintk("RPC:      worker connecting xprt %p to address: %s\n",
1223                         xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1224
1225         if (!transport->inet) {
1226                 struct sock *sk = sock->sk;
1227
1228                 write_lock_bh(&sk->sk_callback_lock);
1229
1230                 sk->sk_user_data = xprt;
1231                 transport->old_data_ready = sk->sk_data_ready;
1232                 transport->old_state_change = sk->sk_state_change;
1233                 transport->old_write_space = sk->sk_write_space;
1234                 sk->sk_data_ready = xs_tcp_data_ready;
1235                 sk->sk_state_change = xs_tcp_state_change;
1236                 sk->sk_write_space = xs_tcp_write_space;
1237                 sk->sk_allocation = GFP_ATOMIC;
1238
1239                 /* socket options */
1240                 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
1241                 sock_reset_flag(sk, SOCK_LINGER);
1242                 tcp_sk(sk)->linger2 = 0;
1243                 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
1244
1245                 xprt_clear_connected(xprt);
1246
1247                 /* Reset to new socket */
1248                 transport->sock = sock;
1249                 transport->inet = sk;
1250
1251                 write_unlock_bh(&sk->sk_callback_lock);
1252         }
1253
1254         /* Tell the socket layer to start connecting... */
1255         xprt->stat.connect_count++;
1256         xprt->stat.connect_start = jiffies;
1257         status = kernel_connect(sock, (struct sockaddr *) &xprt->addr,
1258                         xprt->addrlen, O_NONBLOCK);
1259         dprintk("RPC: %p  connect status %d connected %d sock state %d\n",
1260                         xprt, -status, xprt_connected(xprt), sock->sk->sk_state);
1261         if (status < 0) {
1262                 switch (status) {
1263                         case -EINPROGRESS:
1264                         case -EALREADY:
1265                                 goto out_clear;
1266                         case -ECONNREFUSED:
1267                         case -ECONNRESET:
1268                                 /* retry with existing socket, after a delay */
1269                                 break;
1270                         default:
1271                                 /* get rid of existing socket, and retry */
1272                                 xs_close(xprt);
1273                                 break;
1274                 }
1275         }
1276 out:
1277         xprt_wake_pending_tasks(xprt, status);
1278 out_clear:
1279         xprt_clear_connecting(xprt);
1280 }
1281
1282 /**
1283  * xs_connect - connect a socket to a remote endpoint
1284  * @task: address of RPC task that manages state of connect request
1285  *
1286  * TCP: If the remote end dropped the connection, delay reconnecting.
1287  *
1288  * UDP socket connects are synchronous, but we use a work queue anyway
1289  * to guarantee that even unprivileged user processes can set up a
1290  * socket on a privileged port.
1291  *
1292  * If a UDP socket connect fails, the delay behavior here prevents
1293  * retry floods (hard mounts).
1294  */
1295 static void xs_connect(struct rpc_task *task)
1296 {
1297         struct rpc_xprt *xprt = task->tk_xprt;
1298         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1299
1300         if (xprt_test_and_set_connecting(xprt))
1301                 return;
1302
1303         if (transport->sock != NULL) {
1304                 dprintk("RPC:      xs_connect delayed xprt %p for %lu seconds\n",
1305                                 xprt, xprt->reestablish_timeout / HZ);
1306                 schedule_delayed_work(&transport->connect_worker,
1307                                         xprt->reestablish_timeout);
1308                 xprt->reestablish_timeout <<= 1;
1309                 if (xprt->reestablish_timeout > XS_TCP_MAX_REEST_TO)
1310                         xprt->reestablish_timeout = XS_TCP_MAX_REEST_TO;
1311         } else {
1312                 dprintk("RPC:      xs_connect scheduled xprt %p\n", xprt);
1313                 schedule_work(&transport->connect_worker);
1314
1315                 /* flush_scheduled_work can sleep... */
1316                 if (!RPC_IS_ASYNC(task))
1317                         flush_scheduled_work();
1318         }
1319 }
1320
1321 /**
1322  * xs_udp_print_stats - display UDP socket-specifc stats
1323  * @xprt: rpc_xprt struct containing statistics
1324  * @seq: output file
1325  *
1326  */
1327 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
1328 {
1329         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1330
1331         seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %Lu %Lu\n",
1332                         transport->port,
1333                         xprt->stat.bind_count,
1334                         xprt->stat.sends,
1335                         xprt->stat.recvs,
1336                         xprt->stat.bad_xids,
1337                         xprt->stat.req_u,
1338                         xprt->stat.bklog_u);
1339 }
1340
1341 /**
1342  * xs_tcp_print_stats - display TCP socket-specifc stats
1343  * @xprt: rpc_xprt struct containing statistics
1344  * @seq: output file
1345  *
1346  */
1347 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
1348 {
1349         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1350         long idle_time = 0;
1351
1352         if (xprt_connected(xprt))
1353                 idle_time = (long)(jiffies - xprt->last_used) / HZ;
1354
1355         seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu %Lu %Lu\n",
1356                         transport->port,
1357                         xprt->stat.bind_count,
1358                         xprt->stat.connect_count,
1359                         xprt->stat.connect_time,
1360                         idle_time,
1361                         xprt->stat.sends,
1362                         xprt->stat.recvs,
1363                         xprt->stat.bad_xids,
1364                         xprt->stat.req_u,
1365                         xprt->stat.bklog_u);
1366 }
1367
1368 static struct rpc_xprt_ops xs_udp_ops = {
1369         .set_buffer_size        = xs_udp_set_buffer_size,
1370         .reserve_xprt           = xprt_reserve_xprt_cong,
1371         .release_xprt           = xprt_release_xprt_cong,
1372         .rpcbind                = rpc_getport,
1373         .set_port               = xs_set_port,
1374         .connect                = xs_connect,
1375         .buf_alloc              = rpc_malloc,
1376         .buf_free               = rpc_free,
1377         .send_request           = xs_udp_send_request,
1378         .set_retrans_timeout    = xprt_set_retrans_timeout_rtt,
1379         .timer                  = xs_udp_timer,
1380         .release_request        = xprt_release_rqst_cong,
1381         .close                  = xs_close,
1382         .destroy                = xs_destroy,
1383         .print_stats            = xs_udp_print_stats,
1384 };
1385
1386 static struct rpc_xprt_ops xs_tcp_ops = {
1387         .reserve_xprt           = xprt_reserve_xprt,
1388         .release_xprt           = xs_tcp_release_xprt,
1389         .rpcbind                = rpc_getport,
1390         .set_port               = xs_set_port,
1391         .connect                = xs_connect,
1392         .buf_alloc              = rpc_malloc,
1393         .buf_free               = rpc_free,
1394         .send_request           = xs_tcp_send_request,
1395         .set_retrans_timeout    = xprt_set_retrans_timeout_def,
1396         .close                  = xs_close,
1397         .destroy                = xs_destroy,
1398         .print_stats            = xs_tcp_print_stats,
1399 };
1400
1401 static struct rpc_xprt *xs_setup_xprt(struct sockaddr *addr, size_t addrlen, unsigned int slot_table_size)
1402 {
1403         struct rpc_xprt *xprt;
1404         struct sock_xprt *new;
1405
1406         if (addrlen > sizeof(xprt->addr)) {
1407                 dprintk("RPC:      xs_setup_xprt: address too large\n");
1408                 return ERR_PTR(-EBADF);
1409         }
1410
1411         new = kzalloc(sizeof(*new), GFP_KERNEL);
1412         if (new == NULL) {
1413                 dprintk("RPC:      xs_setup_xprt: couldn't allocate rpc_xprt\n");
1414                 return ERR_PTR(-ENOMEM);
1415         }
1416         xprt = &new->xprt;
1417
1418         xprt->max_reqs = slot_table_size;
1419         xprt->slot = kcalloc(xprt->max_reqs, sizeof(struct rpc_rqst), GFP_KERNEL);
1420         if (xprt->slot == NULL) {
1421                 kfree(xprt);
1422                 dprintk("RPC:      xs_setup_xprt: couldn't allocate slot table\n");
1423                 return ERR_PTR(-ENOMEM);
1424         }
1425
1426         memcpy(&xprt->addr, addr, addrlen);
1427         xprt->addrlen = addrlen;
1428         new->port = xs_get_random_port();
1429
1430         return xprt;
1431 }
1432
1433 /**
1434  * xs_setup_udp - Set up transport to use a UDP socket
1435  * @addr: address of remote server
1436  * @addrlen: length of address in bytes
1437  * @to:   timeout parameters
1438  *
1439  */
1440 struct rpc_xprt *xs_setup_udp(struct sockaddr *addr, size_t addrlen, struct rpc_timeout *to)
1441 {
1442         struct rpc_xprt *xprt;
1443         struct sock_xprt *transport;
1444
1445         xprt = xs_setup_xprt(addr, addrlen, xprt_udp_slot_table_entries);
1446         if (IS_ERR(xprt))
1447                 return xprt;
1448         transport = container_of(xprt, struct sock_xprt, xprt);
1449
1450         if (ntohs(((struct sockaddr_in *)addr)->sin_port) != 0)
1451                 xprt_set_bound(xprt);
1452
1453         xprt->prot = IPPROTO_UDP;
1454         xprt->tsh_size = 0;
1455         /* XXX: header size can vary due to auth type, IPv6, etc. */
1456         xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
1457
1458         INIT_WORK(&transport->connect_worker, xs_udp_connect_worker, transport);
1459         xprt->bind_timeout = XS_BIND_TO;
1460         xprt->connect_timeout = XS_UDP_CONN_TO;
1461         xprt->reestablish_timeout = XS_UDP_REEST_TO;
1462         xprt->idle_timeout = XS_IDLE_DISC_TO;
1463
1464         xprt->ops = &xs_udp_ops;
1465
1466         if (to)
1467                 xprt->timeout = *to;
1468         else
1469                 xprt_set_timeout(&xprt->timeout, 5, 5 * HZ);
1470
1471         xs_format_peer_addresses(xprt);
1472         dprintk("RPC:      set up transport to address %s\n",
1473                         xprt->address_strings[RPC_DISPLAY_ALL]);
1474
1475         return xprt;
1476 }
1477
1478 /**
1479  * xs_setup_tcp - Set up transport to use a TCP socket
1480  * @addr: address of remote server
1481  * @addrlen: length of address in bytes
1482  * @to: timeout parameters
1483  *
1484  */
1485 struct rpc_xprt *xs_setup_tcp(struct sockaddr *addr, size_t addrlen, struct rpc_timeout *to)
1486 {
1487         struct rpc_xprt *xprt;
1488         struct sock_xprt *transport;
1489
1490         xprt = xs_setup_xprt(addr, addrlen, xprt_tcp_slot_table_entries);
1491         if (IS_ERR(xprt))
1492                 return xprt;
1493         transport = container_of(xprt, struct sock_xprt, xprt);
1494
1495         if (ntohs(((struct sockaddr_in *)addr)->sin_port) != 0)
1496                 xprt_set_bound(xprt);
1497
1498         xprt->prot = IPPROTO_TCP;
1499         xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
1500         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
1501
1502         INIT_WORK(&transport->connect_worker, xs_tcp_connect_worker, transport);
1503         xprt->bind_timeout = XS_BIND_TO;
1504         xprt->connect_timeout = XS_TCP_CONN_TO;
1505         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1506         xprt->idle_timeout = XS_IDLE_DISC_TO;
1507
1508         xprt->ops = &xs_tcp_ops;
1509
1510         if (to)
1511                 xprt->timeout = *to;
1512         else
1513                 xprt_set_timeout(&xprt->timeout, 2, 60 * HZ);
1514
1515         xs_format_peer_addresses(xprt);
1516         dprintk("RPC:      set up transport to address %s\n",
1517                         xprt->address_strings[RPC_DISPLAY_ALL]);
1518
1519         return xprt;
1520 }