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 static void xs_format_peer_addresses(struct rpc_xprt *xprt)
 133 {
 134         struct sockaddr_in *addr = (struct sockaddr_in *) &xprt->addr;
 135         char *buf;
 136
 137         buf = kzalloc(20, GFP_KERNEL);
 138         if (buf) {
 139                 snprintf(buf, 20, "%u.%u.%u.%u",
 140                                 NIPQUAD(addr->sin_addr.s_addr));
 141         }
 142         xprt->address_strings[RPC_DISPLAY_ADDR] = buf;
 143
 144         buf = kzalloc(8, GFP_KERNEL);
 145         if (buf) {
 146                 snprintf(buf, 8, "%u",
 147                                 ntohs(addr->sin_port));
 148         }
 149         xprt->address_strings[RPC_DISPLAY_PORT] = buf;
 150
 151         if (xprt->prot == IPPROTO_UDP)
 152                 xprt->address_strings[RPC_DISPLAY_PROTO] = "udp";
 153         else
 154                 xprt->address_strings[RPC_DISPLAY_PROTO] = "tcp";
 155
 156         buf = kzalloc(48, GFP_KERNEL);
 157         if (buf) {
 158                 snprintf(buf, 48, "addr=%u.%u.%u.%u port=%u proto=%s",
 159                         NIPQUAD(addr->sin_addr.s_addr),
 160                         ntohs(addr->sin_port),
 161                         xprt->prot == IPPROTO_UDP ? "udp" : "tcp");
 162         }
 163         xprt->address_strings[RPC_DISPLAY_ALL] = buf;
 164 }
 165
 166 static void xs_free_peer_addresses(struct rpc_xprt *xprt)
 167 {
 168         kfree(xprt->address_strings[RPC_DISPLAY_ADDR]);
 169         kfree(xprt->address_strings[RPC_DISPLAY_PORT]);
 170         kfree(xprt->address_strings[RPC_DISPLAY_ALL]);
 171 }
 172
 173 #define XS_SENDMSG_FLAGS        (MSG_DONTWAIT | MSG_NOSIGNAL)
 174
 175 static int xs_send_kvec(struct socket *sock, struct sockaddr *addr, int addrlen, struct kvec *vec, unsigned int base, int more)
 176 {
 177         struct msghdr msg = {
 178                 .msg_name       = addr,
 179                 .msg_namelen    = addrlen,
 180                 .msg_flags      = XS_SENDMSG_FLAGS | (more ? MSG_MORE : 0),
 181         };
 182         struct kvec iov = {
 183                 .iov_base       = vec->iov_base + base,
 184                 .iov_len        = vec->iov_len - base,
 185         };
 186
 187         if (iov.iov_len != 0)
 188                 return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
 189         return kernel_sendmsg(sock, &msg, NULL, 0, 0);
 190 }
 191
 192 static int xs_send_pagedata(struct socket *sock, struct xdr_buf *xdr, unsigned int base, int more)
 193 {
 194         struct page **ppage;
 195         unsigned int remainder;
 196         int err, sent = 0;
 197
 198         remainder = xdr->page_len - base;
 199         base += xdr->page_base;
 200         ppage = xdr->pages + (base >> PAGE_SHIFT);
 201         base &= ~PAGE_MASK;
 202         for(;;) {
 203                 unsigned int len = min_t(unsigned int, PAGE_SIZE - base, remainder);
 204                 int flags = XS_SENDMSG_FLAGS;
 205
 206                 remainder -= len;
 207                 if (remainder != 0 || more)
 208                         flags |= MSG_MORE;
 209                 err = sock->ops->sendpage(sock, *ppage, base, len, flags);
 210                 if (remainder == 0 || err != len)
 211                         break;
 212                 sent += err;
 213                 ppage++;
 214                 base = 0;
 215         }
 216         if (sent == 0)
 217                 return err;
 218         if (err > 0)
 219                 sent += err;
 220         return sent;
 221 }
 222
 223 /**
 224  * xs_sendpages - write pages directly to a socket
 225  * @sock: socket to send on
 226  * @addr: UDP only -- address of destination
 227  * @addrlen: UDP only -- length of destination address
 228  * @xdr: buffer containing this request
 229  * @base: starting position in the buffer
 230  *
 231  */
 232 static int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base)
 233 {
 234         unsigned int remainder = xdr->len - base;
 235         int err, sent = 0;
 236
 237         if (unlikely(!sock))
 238                 return -ENOTCONN;
 239
 240         clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags);
 241         if (base != 0) {
 242                 addr = NULL;
 243                 addrlen = 0;
 244         }
 245
 246         if (base < xdr->head[0].iov_len || addr != NULL) {
 247                 unsigned int len = xdr->head[0].iov_len - base;
 248                 remainder -= len;
 249                 err = xs_send_kvec(sock, addr, addrlen, &xdr->head[0], base, remainder != 0);
 250                 if (remainder == 0 || err != len)
 251                         goto out;
 252                 sent += err;
 253                 base = 0;
 254         } else
 255                 base -= xdr->head[0].iov_len;
 256
 257         if (base < xdr->page_len) {
 258                 unsigned int len = xdr->page_len - base;
 259                 remainder -= len;
 260                 err = xs_send_pagedata(sock, xdr, base, remainder != 0);
 261                 if (remainder == 0 || err != len)
 262                         goto out;
 263                 sent += err;
 264                 base = 0;
 265         } else
 266                 base -= xdr->page_len;
 267
 268         if (base >= xdr->tail[0].iov_len)
 269                 return sent;
 270         err = xs_send_kvec(sock, NULL, 0, &xdr->tail[0], base, 0);
 271 out:
 272         if (sent == 0)
 273                 return err;
 274         if (err > 0)
 275                 sent += err;
 276         return sent;
 277 }
 278
 279 /**
 280  * xs_nospace - place task on wait queue if transmit was incomplete
 281  * @task: task to put to sleep
 282  *
 283  */
 284 static void xs_nospace(struct rpc_task *task)
 285 {
 286         struct rpc_rqst *req = task->tk_rqstp;
 287         struct rpc_xprt *xprt = req->rq_xprt;
 288
 289         dprintk("RPC: %4d xmit incomplete (%u left of %u)\n",
 290                         task->tk_pid, req->rq_slen - req->rq_bytes_sent,
 291                         req->rq_slen);
 292
 293         if (test_bit(SOCK_ASYNC_NOSPACE, &xprt->sock->flags)) {
 294                 /* Protect against races with write_space */
 295                 spin_lock_bh(&xprt->transport_lock);
 296
 297                 /* Don't race with disconnect */
 298                 if (!xprt_connected(xprt))
 299                         task->tk_status = -ENOTCONN;
 300                 else if (test_bit(SOCK_NOSPACE, &xprt->sock->flags))
 301                         xprt_wait_for_buffer_space(task);
 302
 303                 spin_unlock_bh(&xprt->transport_lock);
 304         } else
 305                 /* Keep holding the socket if it is blocked */
 306                 rpc_delay(task, HZ>>4);
 307 }
 308
 309 /**
 310  * xs_udp_send_request - write an RPC request to a UDP socket
 311  * @task: address of RPC task that manages the state of an RPC request
 312  *
 313  * Return values:
 314  *        0:    The request has been sent
 315  *   EAGAIN:    The socket was blocked, please call again later to
 316  *              complete the request
 317  * ENOTCONN:    Caller needs to invoke connect logic then call again
 318  *    other:    Some other error occured, the request was not sent
 319  */
 320 static int xs_udp_send_request(struct rpc_task *task)
 321 {
 322         struct rpc_rqst *req = task->tk_rqstp;
 323         struct rpc_xprt *xprt = req->rq_xprt;
 324         struct xdr_buf *xdr = &req->rq_snd_buf;
 325         int status;
 326
 327         xs_pktdump("packet data:",
 328                                 req->rq_svec->iov_base,
 329                                 req->rq_svec->iov_len);
 330
 331         req->rq_xtime = jiffies;
 332         status = xs_sendpages(xprt->sock, (struct sockaddr *) &xprt->addr,
 333                                 xprt->addrlen, xdr, req->rq_bytes_sent);
 334
 335         dprintk("RPC:      xs_udp_send_request(%u) = %d\n",
 336                         xdr->len - req->rq_bytes_sent, status);
 337
 338         if (likely(status >= (int) req->rq_slen))
 339                 return 0;
 340
 341         /* Still some bytes left; set up for a retry later. */
 342         if (status > 0)
 343                 status = -EAGAIN;
 344
 345         switch (status) {
 346         case -ENETUNREACH:
 347         case -EPIPE:
 348         case -ECONNREFUSED:
 349                 /* When the server has died, an ICMP port unreachable message
 350                  * prompts ECONNREFUSED. */
 351                 break;
 352         case -EAGAIN:
 353                 xs_nospace(task);
 354                 break;
 355         default:
 356                 dprintk("RPC:      sendmsg returned unrecognized error %d\n",
 357                         -status);
 358                 break;
 359         }
 360
 361         return status;
 362 }
 363
 364 static inline void xs_encode_tcp_record_marker(struct xdr_buf *buf)
 365 {
 366         u32 reclen = buf->len - sizeof(rpc_fraghdr);
 367         rpc_fraghdr *base = buf->head[0].iov_base;
 368         *base = htonl(RPC_LAST_STREAM_FRAGMENT | reclen);
 369 }
 370
 371 /**
 372  * xs_tcp_send_request - write an RPC request to a TCP socket
 373  * @task: address of RPC task that manages the state of an RPC request
 374  *
 375  * Return values:
 376  *        0:    The request has been sent
 377  *   EAGAIN:    The socket was blocked, please call again later to
 378  *              complete the request
 379  * ENOTCONN:    Caller needs to invoke connect logic then call again
 380  *    other:    Some other error occured, the request was not sent
 381  *
 382  * XXX: In the case of soft timeouts, should we eventually give up
 383  *      if sendmsg is not able to make progress?
 384  */
 385 static int xs_tcp_send_request(struct rpc_task *task)
 386 {
 387         struct rpc_rqst *req = task->tk_rqstp;
 388         struct rpc_xprt *xprt = req->rq_xprt;
 389         struct xdr_buf *xdr = &req->rq_snd_buf;
 390         int status, retry = 0;
 391
 392         xs_encode_tcp_record_marker(&req->rq_snd_buf);
 393
 394         xs_pktdump("packet data:",
 395                                 req->rq_svec->iov_base,
 396                                 req->rq_svec->iov_len);
 397
 398         /* Continue transmitting the packet/record. We must be careful
 399          * to cope with writespace callbacks arriving _after_ we have
 400          * called sendmsg(). */
 401         while (1) {
 402                 req->rq_xtime = jiffies;
 403                 status = xs_sendpages(xprt->sock, NULL, 0, xdr,
 404                                                 req->rq_bytes_sent);
 405
 406                 dprintk("RPC:      xs_tcp_send_request(%u) = %d\n",
 407                                 xdr->len - req->rq_bytes_sent, status);
 408
 409                 if (unlikely(status < 0))
 410                         break;
 411
 412                 /* If we've sent the entire packet, immediately
 413                  * reset the count of bytes sent. */
 414                 req->rq_bytes_sent += status;
 415                 task->tk_bytes_sent += status;
 416                 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
 417                         req->rq_bytes_sent = 0;
 418                         return 0;
 419                 }
 420
 421                 status = -EAGAIN;
 422                 if (retry++ > XS_SENDMSG_RETRY)
 423                         break;
 424         }
 425
 426         switch (status) {
 427         case -EAGAIN:
 428                 xs_nospace(task);
 429                 break;
 430         case -ECONNREFUSED:
 431         case -ECONNRESET:
 432         case -ENOTCONN:
 433         case -EPIPE:
 434                 status = -ENOTCONN;
 435                 break;
 436         default:
 437                 dprintk("RPC:      sendmsg returned unrecognized error %d\n",
 438                         -status);
 439                 xprt_disconnect(xprt);
 440                 break;
 441         }
 442
 443         return status;
 444 }
 445
 446 /**
 447  * xs_tcp_release_xprt - clean up after a tcp transmission
 448  * @xprt: transport
 449  * @task: rpc task
 450  *
 451  * This cleans up if an error causes us to abort the transmission of a request.
 452  * In this case, the socket may need to be reset in order to avoid confusing
 453  * the server.
 454  */
 455 static void xs_tcp_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
 456 {
 457         struct rpc_rqst *req;
 458
 459         if (task != xprt->snd_task)
 460                 return;
 461         if (task == NULL)
 462                 goto out_release;
 463         req = task->tk_rqstp;
 464         if (req->rq_bytes_sent == 0)
 465                 goto out_release;
 466         if (req->rq_bytes_sent == req->rq_snd_buf.len)
 467                 goto out_release;
 468         set_bit(XPRT_CLOSE_WAIT, &task->tk_xprt->state);
 469 out_release:
 470         xprt_release_xprt(xprt, task);
 471 }
 472
 473 /**
 474  * xs_close - close a socket
 475  * @xprt: transport
 476  *
 477  * This is used when all requests are complete; ie, no DRC state remains
 478  * on the server we want to save.
 479  */
 480 static void xs_close(struct rpc_xprt *xprt)
 481 {
 482         struct socket *sock = xprt->sock;
 483         struct sock *sk = xprt->inet;
 484
 485         if (!sk)
 486                 goto clear_close_wait;
 487
 488         dprintk("RPC:      xs_close xprt %p\n", xprt);
 489
 490         write_lock_bh(&sk->sk_callback_lock);
 491         xprt->inet = NULL;
 492         xprt->sock = NULL;
 493
 494         sk->sk_user_data = NULL;
 495         sk->sk_data_ready = xprt->old_data_ready;
 496         sk->sk_state_change = xprt->old_state_change;
 497         sk->sk_write_space = xprt->old_write_space;
 498         write_unlock_bh(&sk->sk_callback_lock);
 499
 500         sk->sk_no_check = 0;
 501
 502         sock_release(sock);
 503 clear_close_wait:
 504         smp_mb__before_clear_bit();
 505         clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
 506         smp_mb__after_clear_bit();
 507 }
 508
 509 /**
 510  * xs_destroy - prepare to shutdown a transport
 511  * @xprt: doomed transport
 512  *
 513  */
 514 static void xs_destroy(struct rpc_xprt *xprt)
 515 {
 516         dprintk("RPC:      xs_destroy xprt %p\n", xprt);
 517
 518         cancel_delayed_work(&xprt->connect_worker);
 519         flush_scheduled_work();
 520
 521         xprt_disconnect(xprt);
 522         xs_close(xprt);
 523         xs_free_peer_addresses(xprt);
 524         kfree(xprt->slot);
 525         kfree(xprt);
 526 }
 527
 528 static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
 529 {
 530         return (struct rpc_xprt *) sk->sk_user_data;
 531 }
 532
 533 /**
 534  * xs_udp_data_ready - "data ready" callback for UDP sockets
 535  * @sk: socket with data to read
 536  * @len: how much data to read
 537  *
 538  */
 539 static void xs_udp_data_ready(struct sock *sk, int len)
 540 {
 541         struct rpc_task *task;
 542         struct rpc_xprt *xprt;
 543         struct rpc_rqst *rovr;
 544         struct sk_buff *skb;
 545         int err, repsize, copied;
 546         u32 _xid;
 547         __be32 *xp;
 548
 549         read_lock(&sk->sk_callback_lock);
 550         dprintk("RPC:      xs_udp_data_ready...\n");
 551         if (!(xprt = xprt_from_sock(sk)))
 552                 goto out;
 553
 554         if ((skb = skb_recv_datagram(sk, 0, 1, &err)) == NULL)
 555                 goto out;
 556
 557         if (xprt->shutdown)
 558                 goto dropit;
 559
 560         repsize = skb->len - sizeof(struct udphdr);
 561         if (repsize < 4) {
 562                 dprintk("RPC:      impossible RPC reply size %d!\n", repsize);
 563                 goto dropit;
 564         }
 565
 566         /* Copy the XID from the skb... */
 567         xp = skb_header_pointer(skb, sizeof(struct udphdr),
 568                                 sizeof(_xid), &_xid);
 569         if (xp == NULL)
 570                 goto dropit;
 571
 572         /* Look up and lock the request corresponding to the given XID */
 573         spin_lock(&xprt->transport_lock);
 574         rovr = xprt_lookup_rqst(xprt, *xp);
 575         if (!rovr)
 576                 goto out_unlock;
 577         task = rovr->rq_task;
 578
 579         if ((copied = rovr->rq_private_buf.buflen) > repsize)
 580                 copied = repsize;
 581
 582         /* Suck it into the iovec, verify checksum if not done by hw. */
 583         if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb))
 584                 goto out_unlock;
 585
 586         /* Something worked... */
 587         dst_confirm(skb->dst);
 588
 589         xprt_adjust_cwnd(task, copied);
 590         xprt_update_rtt(task);
 591         xprt_complete_rqst(task, copied);
 592
 593  out_unlock:
 594         spin_unlock(&xprt->transport_lock);
 595  dropit:
 596         skb_free_datagram(sk, skb);
 597  out:
 598         read_unlock(&sk->sk_callback_lock);
 599 }
 600
 601 static inline size_t xs_tcp_copy_data(skb_reader_t *desc, void *p, size_t len)
 602 {
 603         if (len > desc->count)
 604                 len = desc->count;
 605         if (skb_copy_bits(desc->skb, desc->offset, p, len)) {
 606                 dprintk("RPC:      failed to copy %zu bytes from skb. %zu bytes remain\n",
 607                                 len, desc->count);
 608                 return 0;
 609         }
 610         desc->offset += len;
 611         desc->count -= len;
 612         dprintk("RPC:      copied %zu bytes from skb. %zu bytes remain\n",
 613                         len, desc->count);
 614         return len;
 615 }
 616
 617 static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, skb_reader_t *desc)
 618 {
 619         size_t len, used;
 620         char *p;
 621
 622         p = ((char *) &xprt->tcp_recm) + xprt->tcp_offset;
 623         len = sizeof(xprt->tcp_recm) - xprt->tcp_offset;
 624         used = xs_tcp_copy_data(desc, p, len);
 625         xprt->tcp_offset += used;
 626         if (used != len)
 627                 return;
 628
 629         xprt->tcp_reclen = ntohl(xprt->tcp_recm);
 630         if (xprt->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
 631                 xprt->tcp_flags |= XPRT_LAST_FRAG;
 632         else
 633                 xprt->tcp_flags &= ~XPRT_LAST_FRAG;
 634         xprt->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
 635
 636         xprt->tcp_flags &= ~XPRT_COPY_RECM;
 637         xprt->tcp_offset = 0;
 638
 639         /* Sanity check of the record length */
 640         if (unlikely(xprt->tcp_reclen < 4)) {
 641                 dprintk("RPC:      invalid TCP record fragment length\n");
 642                 xprt_disconnect(xprt);
 643                 return;
 644         }
 645         dprintk("RPC:      reading TCP record fragment of length %d\n",
 646                         xprt->tcp_reclen);
 647 }
 648
 649 static void xs_tcp_check_recm(struct rpc_xprt *xprt)
 650 {
 651         dprintk("RPC:      xprt = %p, tcp_copied = %lu, tcp_offset = %u, tcp_reclen = %u, tcp_flags = %lx\n",
 652                         xprt, xprt->tcp_copied, xprt->tcp_offset, xprt->tcp_reclen, xprt->tcp_flags);
 653         if (xprt->tcp_offset == xprt->tcp_reclen) {
 654                 xprt->tcp_flags |= XPRT_COPY_RECM;
 655                 xprt->tcp_offset = 0;
 656                 if (xprt->tcp_flags & XPRT_LAST_FRAG) {
 657                         xprt->tcp_flags &= ~XPRT_COPY_DATA;
 658                         xprt->tcp_flags |= XPRT_COPY_XID;
 659                         xprt->tcp_copied = 0;
 660                 }
 661         }
 662 }
 663
 664 static inline void xs_tcp_read_xid(struct rpc_xprt *xprt, skb_reader_t *desc)
 665 {
 666         size_t len, used;
 667         char *p;
 668
 669         len = sizeof(xprt->tcp_xid) - xprt->tcp_offset;
 670         dprintk("RPC:      reading XID (%Zu bytes)\n", len);
 671         p = ((char *) &xprt->tcp_xid) + xprt->tcp_offset;
 672         used = xs_tcp_copy_data(desc, p, len);
 673         xprt->tcp_offset += used;
 674         if (used != len)
 675                 return;
 676         xprt->tcp_flags &= ~XPRT_COPY_XID;
 677         xprt->tcp_flags |= XPRT_COPY_DATA;
 678         xprt->tcp_copied = 4;
 679         dprintk("RPC:      reading reply for XID %08x\n",
 680                                                 ntohl(xprt->tcp_xid));
 681         xs_tcp_check_recm(xprt);
 682 }
 683
 684 static inline void xs_tcp_read_request(struct rpc_xprt *xprt, skb_reader_t *desc)
 685 {
 686         struct rpc_rqst *req;
 687         struct xdr_buf *rcvbuf;
 688         size_t len;
 689         ssize_t r;
 690
 691         /* Find and lock the request corresponding to this xid */
 692         spin_lock(&xprt->transport_lock);
 693         req = xprt_lookup_rqst(xprt, xprt->tcp_xid);
 694         if (!req) {
 695                 xprt->tcp_flags &= ~XPRT_COPY_DATA;
 696                 dprintk("RPC:      XID %08x request not found!\n",
 697                                 ntohl(xprt->tcp_xid));
 698                 spin_unlock(&xprt->transport_lock);
 699                 return;
 700         }
 701
 702         rcvbuf = &req->rq_private_buf;
 703         len = desc->count;
 704         if (len > xprt->tcp_reclen - xprt->tcp_offset) {
 705                 skb_reader_t my_desc;
 706
 707                 len = xprt->tcp_reclen - xprt->tcp_offset;
 708                 memcpy(&my_desc, desc, sizeof(my_desc));
 709                 my_desc.count = len;
 710                 r = xdr_partial_copy_from_skb(rcvbuf, xprt->tcp_copied,
 711                                           &my_desc, xs_tcp_copy_data);
 712                 desc->count -= r;
 713                 desc->offset += r;
 714         } else
 715                 r = xdr_partial_copy_from_skb(rcvbuf, xprt->tcp_copied,
 716                                           desc, xs_tcp_copy_data);
 717
 718         if (r > 0) {
 719                 xprt->tcp_copied += r;
 720                 xprt->tcp_offset += r;
 721         }
 722         if (r != len) {
 723                 /* Error when copying to the receive buffer,
 724                  * usually because we weren't able to allocate
 725                  * additional buffer pages. All we can do now
 726                  * is turn off XPRT_COPY_DATA, so the request
 727                  * will not receive any additional updates,
 728                  * and time out.
 729                  * Any remaining data from this record will
 730                  * be discarded.
 731                  */
 732                 xprt->tcp_flags &= ~XPRT_COPY_DATA;
 733                 dprintk("RPC:      XID %08x truncated request\n",
 734                                 ntohl(xprt->tcp_xid));
 735                 dprintk("RPC:      xprt = %p, tcp_copied = %lu, tcp_offset = %u, tcp_reclen = %u\n",
 736                                 xprt, xprt->tcp_copied, xprt->tcp_offset, xprt->tcp_reclen);
 737                 goto out;
 738         }
 739
 740         dprintk("RPC:      XID %08x read %Zd bytes\n",
 741                         ntohl(xprt->tcp_xid), r);
 742         dprintk("RPC:      xprt = %p, tcp_copied = %lu, tcp_offset = %u, tcp_reclen = %u\n",
 743                         xprt, xprt->tcp_copied, xprt->tcp_offset, xprt->tcp_reclen);
 744
 745         if (xprt->tcp_copied == req->rq_private_buf.buflen)
 746                 xprt->tcp_flags &= ~XPRT_COPY_DATA;
 747         else if (xprt->tcp_offset == xprt->tcp_reclen) {
 748                 if (xprt->tcp_flags & XPRT_LAST_FRAG)
 749                         xprt->tcp_flags &= ~XPRT_COPY_DATA;
 750         }
 751
 752 out:
 753         if (!(xprt->tcp_flags & XPRT_COPY_DATA))
 754                 xprt_complete_rqst(req->rq_task, xprt->tcp_copied);
 755         spin_unlock(&xprt->transport_lock);
 756         xs_tcp_check_recm(xprt);
 757 }
 758
 759 static inline void xs_tcp_read_discard(struct rpc_xprt *xprt, skb_reader_t *desc)
 760 {
 761         size_t len;
 762
 763         len = xprt->tcp_reclen - xprt->tcp_offset;
 764         if (len > desc->count)
 765                 len = desc->count;
 766         desc->count -= len;
 767         desc->offset += len;
 768         xprt->tcp_offset += len;
 769         dprintk("RPC:      discarded %Zu bytes\n", len);
 770         xs_tcp_check_recm(xprt);
 771 }
 772
 773 static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
 774 {
 775         struct rpc_xprt *xprt = rd_desc->arg.data;
 776         skb_reader_t desc = {
 777                 .skb    = skb,
 778                 .offset = offset,
 779                 .count  = len,
 780                 .csum   = 0
 781         };
 782
 783         dprintk("RPC:      xs_tcp_data_recv started\n");
 784         do {
 785                 /* Read in a new fragment marker if necessary */
 786                 /* Can we ever really expect to get completely empty fragments? */
 787                 if (xprt->tcp_flags & XPRT_COPY_RECM) {
 788                         xs_tcp_read_fraghdr(xprt, &desc);
 789                         continue;
 790                 }
 791                 /* Read in the xid if necessary */
 792                 if (xprt->tcp_flags & XPRT_COPY_XID) {
 793                         xs_tcp_read_xid(xprt, &desc);
 794                         continue;
 795                 }
 796                 /* Read in the request data */
 797                 if (xprt->tcp_flags & XPRT_COPY_DATA) {
 798                         xs_tcp_read_request(xprt, &desc);
 799                         continue;
 800                 }
 801                 /* Skip over any trailing bytes on short reads */
 802                 xs_tcp_read_discard(xprt, &desc);
 803         } while (desc.count);
 804         dprintk("RPC:      xs_tcp_data_recv done\n");
 805         return len - desc.count;
 806 }
 807
 808 /**
 809  * xs_tcp_data_ready - "data ready" callback for TCP sockets
 810  * @sk: socket with data to read
 811  * @bytes: how much data to read
 812  *
 813  */
 814 static void xs_tcp_data_ready(struct sock *sk, int bytes)
 815 {
 816         struct rpc_xprt *xprt;
 817         read_descriptor_t rd_desc;
 818
 819         read_lock(&sk->sk_callback_lock);
 820         dprintk("RPC:      xs_tcp_data_ready...\n");
 821         if (!(xprt = xprt_from_sock(sk)))
 822                 goto out;
 823         if (xprt->shutdown)
 824                 goto out;
 825
 826         /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
 827         rd_desc.arg.data = xprt;
 828         rd_desc.count = 65536;
 829         tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
 830 out:
 831         read_unlock(&sk->sk_callback_lock);
 832 }
 833
 834 /**
 835  * xs_tcp_state_change - callback to handle TCP socket state changes
 836  * @sk: socket whose state has changed
 837  *
 838  */
 839 static void xs_tcp_state_change(struct sock *sk)
 840 {
 841         struct rpc_xprt *xprt;
 842
 843         read_lock(&sk->sk_callback_lock);
 844         if (!(xprt = xprt_from_sock(sk)))
 845                 goto out;
 846         dprintk("RPC:      xs_tcp_state_change client %p...\n", xprt);
 847         dprintk("RPC:      state %x conn %d dead %d zapped %d\n",
 848                                 sk->sk_state, xprt_connected(xprt),
 849                                 sock_flag(sk, SOCK_DEAD),
 850                                 sock_flag(sk, SOCK_ZAPPED));
 851
 852         switch (sk->sk_state) {
 853         case TCP_ESTABLISHED:
 854                 spin_lock_bh(&xprt->transport_lock);
 855                 if (!xprt_test_and_set_connected(xprt)) {
 856                         /* Reset TCP record info */
 857                         xprt->tcp_offset = 0;
 858                         xprt->tcp_reclen = 0;
 859                         xprt->tcp_copied = 0;
 860                         xprt->tcp_flags = XPRT_COPY_RECM | XPRT_COPY_XID;
 861                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
 862                         xprt_wake_pending_tasks(xprt, 0);
 863                 }
 864                 spin_unlock_bh(&xprt->transport_lock);
 865                 break;
 866         case TCP_SYN_SENT:
 867         case TCP_SYN_RECV:
 868                 break;
 869         case TCP_CLOSE_WAIT:
 870                 /* Try to schedule an autoclose RPC calls */
 871                 set_bit(XPRT_CLOSE_WAIT, &xprt->state);
 872                 if (test_and_set_bit(XPRT_LOCKED, &xprt->state) == 0)
 873                         schedule_work(&xprt->task_cleanup);
 874         default:
 875                 xprt_disconnect(xprt);
 876         }
 877  out:
 878         read_unlock(&sk->sk_callback_lock);
 879 }
 880
 881 /**
 882  * xs_udp_write_space - callback invoked when socket buffer space
 883  *                             becomes available
 884  * @sk: socket whose state has changed
 885  *
 886  * Called when more output buffer space is available for this socket.
 887  * We try not to wake our writers until they can make "significant"
 888  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
 889  * with a bunch of small requests.
 890  */
 891 static void xs_udp_write_space(struct sock *sk)
 892 {
 893         read_lock(&sk->sk_callback_lock);
 894
 895         /* from net/core/sock.c:sock_def_write_space */
 896         if (sock_writeable(sk)) {
 897                 struct socket *sock;
 898                 struct rpc_xprt *xprt;
 899
 900                 if (unlikely(!(sock = sk->sk_socket)))
 901                         goto out;
 902                 if (unlikely(!(xprt = xprt_from_sock(sk))))
 903                         goto out;
 904                 if (unlikely(!test_and_clear_bit(SOCK_NOSPACE, &sock->flags)))
 905                         goto out;
 906
 907                 xprt_write_space(xprt);
 908         }
 909
 910  out:
 911         read_unlock(&sk->sk_callback_lock);
 912 }
 913
 914 /**
 915  * xs_tcp_write_space - callback invoked when socket buffer space
 916  *                             becomes available
 917  * @sk: socket whose state has changed
 918  *
 919  * Called when more output buffer space is available for this socket.
 920  * We try not to wake our writers until they can make "significant"
 921  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
 922  * with a bunch of small requests.
 923  */
 924 static void xs_tcp_write_space(struct sock *sk)
 925 {
 926         read_lock(&sk->sk_callback_lock);
 927
 928         /* from net/core/stream.c:sk_stream_write_space */
 929         if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk)) {
 930                 struct socket *sock;
 931                 struct rpc_xprt *xprt;
 932
 933                 if (unlikely(!(sock = sk->sk_socket)))
 934                         goto out;
 935                 if (unlikely(!(xprt = xprt_from_sock(sk))))
 936                         goto out;
 937                 if (unlikely(!test_and_clear_bit(SOCK_NOSPACE, &sock->flags)))
 938                         goto out;
 939
 940                 xprt_write_space(xprt);
 941         }
 942
 943  out:
 944         read_unlock(&sk->sk_callback_lock);
 945 }
 946
 947 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
 948 {
 949         struct sock *sk = xprt->inet;
 950
 951         if (xprt->rcvsize) {
 952                 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
 953                 sk->sk_rcvbuf = xprt->rcvsize * xprt->max_reqs *  2;
 954         }
 955         if (xprt->sndsize) {
 956                 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
 957                 sk->sk_sndbuf = xprt->sndsize * xprt->max_reqs * 2;
 958                 sk->sk_write_space(sk);
 959         }
 960 }
 961
 962 /**
 963  * xs_udp_set_buffer_size - set send and receive limits
 964  * @xprt: generic transport
 965  * @sndsize: requested size of send buffer, in bytes
 966  * @rcvsize: requested size of receive buffer, in bytes
 967  *
 968  * Set socket send and receive buffer size limits.
 969  */
 970 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
 971 {
 972         xprt->sndsize = 0;
 973         if (sndsize)
 974                 xprt->sndsize = sndsize + 1024;
 975         xprt->rcvsize = 0;
 976         if (rcvsize)
 977                 xprt->rcvsize = rcvsize + 1024;
 978
 979         xs_udp_do_set_buffer_size(xprt);
 980 }
 981
 982 /**
 983  * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
 984  * @task: task that timed out
 985  *
 986  * Adjust the congestion window after a retransmit timeout has occurred.
 987  */
 988 static void xs_udp_timer(struct rpc_task *task)
 989 {
 990         xprt_adjust_cwnd(task, -ETIMEDOUT);
 991 }
 992
 993 static unsigned short xs_get_random_port(void)
 994 {
 995         unsigned short range = xprt_max_resvport - xprt_min_resvport;
 996         unsigned short rand = (unsigned short) net_random() % range;
 997         return rand + xprt_min_resvport;
 998 }
 999
1000 /**
1001  * xs_print_peer_address - format an IPv4 address for printing
1002  * @xprt: generic transport
1003  * @format: flags field indicating which parts of the address to render
1004  */
1005 static char *xs_print_peer_address(struct rpc_xprt *xprt, enum rpc_display_format_t format)
1006 {
1007         if (xprt->address_strings[format] != NULL)
1008                 return xprt->address_strings[format];
1009         else
1010                 return "unprintable";
1011 }
1012
1013 /**
1014  * xs_set_port - reset the port number in the remote endpoint address
1015  * @xprt: generic transport
1016  * @port: new port number
1017  *
1018  */
1019 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1020 {
1021         struct sockaddr_in *sap = (struct sockaddr_in *) &xprt->addr;
1022
1023         dprintk("RPC:      setting port for xprt %p to %u\n", xprt, port);
1024
1025         sap->sin_port = htons(port);
1026 }
1027
1028 static int xs_bindresvport(struct rpc_xprt *xprt, struct socket *sock)
1029 {
1030         struct sockaddr_in myaddr = {
1031                 .sin_family = AF_INET,
1032         };
1033         int err;
1034         unsigned short port = xprt->port;
1035
1036         do {
1037                 myaddr.sin_port = htons(port);
1038                 err = kernel_bind(sock, (struct sockaddr *) &myaddr,
1039                                                 sizeof(myaddr));
1040                 if (err == 0) {
1041                         xprt->port = port;
1042                         dprintk("RPC:      xs_bindresvport bound to port %u\n",
1043                                         port);
1044                         return 0;
1045                 }
1046                 if (port <= xprt_min_resvport)
1047                         port = xprt_max_resvport;
1048                 else
1049                         port--;
1050         } while (err == -EADDRINUSE && port != xprt->port);
1051
1052         dprintk("RPC:      can't bind to reserved port (%d).\n", -err);
1053         return err;
1054 }
1055
1056 /**
1057  * xs_udp_connect_worker - set up a UDP socket
1058  * @args: RPC transport to connect
1059  *
1060  * Invoked by a work queue tasklet.
1061  */
1062 static void xs_udp_connect_worker(void *args)
1063 {
1064         struct rpc_xprt *xprt = (struct rpc_xprt *) args;
1065         struct socket *sock = xprt->sock;
1066         int err, status = -EIO;
1067
1068         if (xprt->shutdown || !xprt_bound(xprt))
1069                 goto out;
1070
1071         /* Start by resetting any existing state */
1072         xs_close(xprt);
1073
1074         if ((err = sock_create_kern(PF_INET, SOCK_DGRAM, IPPROTO_UDP, &sock)) < 0) {
1075                 dprintk("RPC:      can't create UDP transport socket (%d).\n", -err);
1076                 goto out;
1077         }
1078
1079         if (xprt->resvport && xs_bindresvport(xprt, sock) < 0) {
1080                 sock_release(sock);
1081                 goto out;
1082         }
1083
1084         dprintk("RPC:      worker connecting xprt %p to address: %s\n",
1085                         xprt, xs_print_peer_address(xprt, RPC_DISPLAY_ALL));
1086
1087         if (!xprt->inet) {
1088                 struct sock *sk = sock->sk;
1089
1090                 write_lock_bh(&sk->sk_callback_lock);
1091
1092                 sk->sk_user_data = xprt;
1093                 xprt->old_data_ready = sk->sk_data_ready;
1094                 xprt->old_state_change = sk->sk_state_change;
1095                 xprt->old_write_space = sk->sk_write_space;
1096                 sk->sk_data_ready = xs_udp_data_ready;
1097                 sk->sk_write_space = xs_udp_write_space;
1098                 sk->sk_no_check = UDP_CSUM_NORCV;
1099                 sk->sk_allocation = GFP_ATOMIC;
1100
1101                 xprt_set_connected(xprt);
1102
1103                 /* Reset to new socket */
1104                 xprt->sock = sock;
1105                 xprt->inet = sk;
1106
1107                 write_unlock_bh(&sk->sk_callback_lock);
1108         }
1109         xs_udp_do_set_buffer_size(xprt);
1110         status = 0;
1111 out:
1112         xprt_wake_pending_tasks(xprt, status);
1113         xprt_clear_connecting(xprt);
1114 }
1115
1116 /*
1117  * We need to preserve the port number so the reply cache on the server can
1118  * find our cached RPC replies when we get around to reconnecting.
1119  */
1120 static void xs_tcp_reuse_connection(struct rpc_xprt *xprt)
1121 {
1122         int result;
1123         struct socket *sock = xprt->sock;
1124         struct sockaddr any;
1125
1126         dprintk("RPC:      disconnecting xprt %p to reuse port\n", xprt);
1127
1128         /*
1129          * Disconnect the transport socket by doing a connect operation
1130          * with AF_UNSPEC.  This should return immediately...
1131          */
1132         memset(&any, 0, sizeof(any));
1133         any.sa_family = AF_UNSPEC;
1134         result = kernel_connect(sock, &any, sizeof(any), 0);
1135         if (result)
1136                 dprintk("RPC:      AF_UNSPEC connect return code %d\n",
1137                                 result);
1138 }
1139
1140 /**
1141  * xs_tcp_connect_worker - connect a TCP socket to a remote endpoint
1142  * @args: RPC transport to connect
1143  *
1144  * Invoked by a work queue tasklet.
1145  */
1146 static void xs_tcp_connect_worker(void *args)
1147 {
1148         struct rpc_xprt *xprt = (struct rpc_xprt *)args;
1149         struct socket *sock = xprt->sock;
1150         int err, status = -EIO;
1151
1152         if (xprt->shutdown || !xprt_bound(xprt))
1153                 goto out;
1154
1155         if (!xprt->sock) {
1156                 /* start from scratch */
1157                 if ((err = sock_create_kern(PF_INET, SOCK_STREAM, IPPROTO_TCP, &sock)) < 0) {
1158                         dprintk("RPC:      can't create TCP transport socket (%d).\n", -err);
1159                         goto out;
1160                 }
1161
1162                 if (xprt->resvport && xs_bindresvport(xprt, sock) < 0) {
1163                         sock_release(sock);
1164                         goto out;
1165                 }
1166         } else
1167                 /* "close" the socket, preserving the local port */
1168                 xs_tcp_reuse_connection(xprt);
1169
1170         dprintk("RPC:      worker connecting xprt %p to address: %s\n",
1171                         xprt, xs_print_peer_address(xprt, RPC_DISPLAY_ALL));
1172
1173         if (!xprt->inet) {
1174                 struct sock *sk = sock->sk;
1175
1176                 write_lock_bh(&sk->sk_callback_lock);
1177
1178                 sk->sk_user_data = xprt;
1179                 xprt->old_data_ready = sk->sk_data_ready;
1180                 xprt->old_state_change = sk->sk_state_change;
1181                 xprt->old_write_space = sk->sk_write_space;
1182                 sk->sk_data_ready = xs_tcp_data_ready;
1183                 sk->sk_state_change = xs_tcp_state_change;
1184                 sk->sk_write_space = xs_tcp_write_space;
1185                 sk->sk_allocation = GFP_ATOMIC;
1186
1187                 /* socket options */
1188                 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
1189                 sock_reset_flag(sk, SOCK_LINGER);
1190                 tcp_sk(sk)->linger2 = 0;
1191                 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
1192
1193                 xprt_clear_connected(xprt);
1194
1195                 /* Reset to new socket */
1196                 xprt->sock = sock;
1197                 xprt->inet = sk;
1198
1199                 write_unlock_bh(&sk->sk_callback_lock);
1200         }
1201
1202         /* Tell the socket layer to start connecting... */
1203         xprt->stat.connect_count++;
1204         xprt->stat.connect_start = jiffies;
1205         status = kernel_connect(sock, (struct sockaddr *) &xprt->addr,
1206                         xprt->addrlen, O_NONBLOCK);
1207         dprintk("RPC: %p  connect status %d connected %d sock state %d\n",
1208                         xprt, -status, xprt_connected(xprt), sock->sk->sk_state);
1209         if (status < 0) {
1210                 switch (status) {
1211                         case -EINPROGRESS:
1212                         case -EALREADY:
1213                                 goto out_clear;
1214                         case -ECONNREFUSED:
1215                         case -ECONNRESET:
1216                                 /* retry with existing socket, after a delay */
1217                                 break;
1218                         default:
1219                                 /* get rid of existing socket, and retry */
1220                                 xs_close(xprt);
1221                                 break;
1222                 }
1223         }
1224 out:
1225         xprt_wake_pending_tasks(xprt, status);
1226 out_clear:
1227         xprt_clear_connecting(xprt);
1228 }
1229
1230 /**
1231  * xs_connect - connect a socket to a remote endpoint
1232  * @task: address of RPC task that manages state of connect request
1233  *
1234  * TCP: If the remote end dropped the connection, delay reconnecting.
1235  *
1236  * UDP socket connects are synchronous, but we use a work queue anyway
1237  * to guarantee that even unprivileged user processes can set up a
1238  * socket on a privileged port.
1239  *
1240  * If a UDP socket connect fails, the delay behavior here prevents
1241  * retry floods (hard mounts).
1242  */
1243 static void xs_connect(struct rpc_task *task)
1244 {
1245         struct rpc_xprt *xprt = task->tk_xprt;
1246
1247         if (xprt_test_and_set_connecting(xprt))
1248                 return;
1249
1250         if (xprt->sock != NULL) {
1251                 dprintk("RPC:      xs_connect delayed xprt %p for %lu seconds\n",
1252                                 xprt, xprt->reestablish_timeout / HZ);
1253                 schedule_delayed_work(&xprt->connect_worker,
1254                                         xprt->reestablish_timeout);
1255                 xprt->reestablish_timeout <<= 1;
1256                 if (xprt->reestablish_timeout > XS_TCP_MAX_REEST_TO)
1257                         xprt->reestablish_timeout = XS_TCP_MAX_REEST_TO;
1258         } else {
1259                 dprintk("RPC:      xs_connect scheduled xprt %p\n", xprt);
1260                 schedule_work(&xprt->connect_worker);
1261
1262                 /* flush_scheduled_work can sleep... */
1263                 if (!RPC_IS_ASYNC(task))
1264                         flush_scheduled_work();
1265         }
1266 }
1267
1268 /**
1269  * xs_udp_print_stats - display UDP socket-specifc stats
1270  * @xprt: rpc_xprt struct containing statistics
1271  * @seq: output file
1272  *
1273  */
1274 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
1275 {
1276         seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %Lu %Lu\n",
1277                         xprt->port,
1278                         xprt->stat.bind_count,
1279                         xprt->stat.sends,
1280                         xprt->stat.recvs,
1281                         xprt->stat.bad_xids,
1282                         xprt->stat.req_u,
1283                         xprt->stat.bklog_u);
1284 }
1285
1286 /**
1287  * xs_tcp_print_stats - display TCP socket-specifc stats
1288  * @xprt: rpc_xprt struct containing statistics
1289  * @seq: output file
1290  *
1291  */
1292 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
1293 {
1294         long idle_time = 0;
1295
1296         if (xprt_connected(xprt))
1297                 idle_time = (long)(jiffies - xprt->last_used) / HZ;
1298
1299         seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu %Lu %Lu\n",
1300                         xprt->port,
1301                         xprt->stat.bind_count,
1302                         xprt->stat.connect_count,
1303                         xprt->stat.connect_time,
1304                         idle_time,
1305                         xprt->stat.sends,
1306                         xprt->stat.recvs,
1307                         xprt->stat.bad_xids,
1308                         xprt->stat.req_u,
1309                         xprt->stat.bklog_u);
1310 }
1311
1312 static struct rpc_xprt_ops xs_udp_ops = {
1313         .set_buffer_size        = xs_udp_set_buffer_size,
1314         .print_addr             = xs_print_peer_address,
1315         .reserve_xprt           = xprt_reserve_xprt_cong,
1316         .release_xprt           = xprt_release_xprt_cong,
1317         .rpcbind                = rpc_getport,
1318         .set_port               = xs_set_port,
1319         .connect                = xs_connect,
1320         .buf_alloc              = rpc_malloc,
1321         .buf_free               = rpc_free,
1322         .send_request           = xs_udp_send_request,
1323         .set_retrans_timeout    = xprt_set_retrans_timeout_rtt,
1324         .timer                  = xs_udp_timer,
1325         .release_request        = xprt_release_rqst_cong,
1326         .close                  = xs_close,
1327         .destroy                = xs_destroy,
1328         .print_stats            = xs_udp_print_stats,
1329 };
1330
1331 static struct rpc_xprt_ops xs_tcp_ops = {
1332         .print_addr             = xs_print_peer_address,
1333         .reserve_xprt           = xprt_reserve_xprt,
1334         .release_xprt           = xs_tcp_release_xprt,
1335         .rpcbind                = rpc_getport,
1336         .set_port               = xs_set_port,
1337         .connect                = xs_connect,
1338         .buf_alloc              = rpc_malloc,
1339         .buf_free               = rpc_free,
1340         .send_request           = xs_tcp_send_request,
1341         .set_retrans_timeout    = xprt_set_retrans_timeout_def,
1342         .close                  = xs_close,
1343         .destroy                = xs_destroy,
1344         .print_stats            = xs_tcp_print_stats,
1345 };
1346
1347 static struct rpc_xprt *xs_setup_xprt(struct sockaddr *addr, size_t addrlen, unsigned int slot_table_size)
1348 {
1349         struct rpc_xprt *xprt;
1350         struct sock_xprt *new;
1351
1352         if (addrlen > sizeof(xprt->addr)) {
1353                 dprintk("RPC:      xs_setup_xprt: address too large\n");
1354                 return ERR_PTR(-EBADF);
1355         }
1356
1357         new = kzalloc(sizeof(*new), GFP_KERNEL);
1358         if (new == NULL) {
1359                 dprintk("RPC:      xs_setup_xprt: couldn't allocate rpc_xprt\n");
1360                 return ERR_PTR(-ENOMEM);
1361         }
1362         xprt = &new->xprt;
1363
1364         xprt->max_reqs = slot_table_size;
1365         xprt->slot = kcalloc(xprt->max_reqs, sizeof(struct rpc_rqst), GFP_KERNEL);
1366         if (xprt->slot == NULL) {
1367                 kfree(xprt);
1368                 dprintk("RPC:      xs_setup_xprt: couldn't allocate slot table\n");
1369                 return ERR_PTR(-ENOMEM);
1370         }
1371
1372         memcpy(&xprt->addr, addr, addrlen);
1373         xprt->addrlen = addrlen;
1374         xprt->port = xs_get_random_port();
1375
1376         return xprt;
1377 }
1378
1379 /**
1380  * xs_setup_udp - Set up transport to use a UDP socket
1381  * @addr: address of remote server
1382  * @addrlen: length of address in bytes
1383  * @to:   timeout parameters
1384  *
1385  */
1386 struct rpc_xprt *xs_setup_udp(struct sockaddr *addr, size_t addrlen, struct rpc_timeout *to)
1387 {
1388         struct rpc_xprt *xprt;
1389
1390         xprt = xs_setup_xprt(addr, addrlen, xprt_udp_slot_table_entries);
1391         if (IS_ERR(xprt))
1392                 return xprt;
1393
1394         if (ntohs(((struct sockaddr_in *)addr)->sin_port) != 0)
1395                 xprt_set_bound(xprt);
1396
1397         xprt->prot = IPPROTO_UDP;
1398         xprt->tsh_size = 0;
1399         /* XXX: header size can vary due to auth type, IPv6, etc. */
1400         xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
1401
1402         INIT_WORK(&xprt->connect_worker, xs_udp_connect_worker, xprt);
1403         xprt->bind_timeout = XS_BIND_TO;
1404         xprt->connect_timeout = XS_UDP_CONN_TO;
1405         xprt->reestablish_timeout = XS_UDP_REEST_TO;
1406         xprt->idle_timeout = XS_IDLE_DISC_TO;
1407
1408         xprt->ops = &xs_udp_ops;
1409
1410         if (to)
1411                 xprt->timeout = *to;
1412         else
1413                 xprt_set_timeout(&xprt->timeout, 5, 5 * HZ);
1414
1415         xs_format_peer_addresses(xprt);
1416         dprintk("RPC:      set up transport to address %s\n",
1417                         xs_print_peer_address(xprt, RPC_DISPLAY_ALL));
1418
1419         return xprt;
1420 }
1421
1422 /**
1423  * xs_setup_tcp - Set up transport to use a TCP socket
1424  * @addr: address of remote server
1425  * @addrlen: length of address in bytes
1426  * @to: timeout parameters
1427  *
1428  */
1429 struct rpc_xprt *xs_setup_tcp(struct sockaddr *addr, size_t addrlen, struct rpc_timeout *to)
1430 {
1431         struct rpc_xprt *xprt;
1432
1433         xprt = xs_setup_xprt(addr, addrlen, xprt_tcp_slot_table_entries);
1434         if (IS_ERR(xprt))
1435                 return xprt;
1436
1437         if (ntohs(((struct sockaddr_in *)addr)->sin_port) != 0)
1438                 xprt_set_bound(xprt);
1439
1440         xprt->prot = IPPROTO_TCP;
1441         xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
1442         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
1443
1444         INIT_WORK(&xprt->connect_worker, xs_tcp_connect_worker, xprt);
1445         xprt->bind_timeout = XS_BIND_TO;
1446         xprt->connect_timeout = XS_TCP_CONN_TO;
1447         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1448         xprt->idle_timeout = XS_IDLE_DISC_TO;
1449
1450         xprt->ops = &xs_tcp_ops;
1451
1452         if (to)
1453                 xprt->timeout = *to;
1454         else
1455                 xprt_set_timeout(&xprt->timeout, 2, 60 * HZ);
1456
1457         xs_format_peer_addresses(xprt);
1458         dprintk("RPC:      set up transport to address %s\n",
1459                         xs_print_peer_address(xprt, RPC_DISPLAY_ALL));
1460
1461         return xprt;
1462 }