SAFE public projects git trees. - safe/jmp/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/pagemap.h>
  22 #include <linux/errno.h>
  23 #include <linux/socket.h>
  24 #include <linux/in.h>
  25 #include <linux/net.h>
  26 #include <linux/mm.h>
  27 #include <linux/udp.h>
  28 #include <linux/tcp.h>
  29 #include <linux/sunrpc/clnt.h>
  30 #include <linux/sunrpc/sched.h>
  31 #include <linux/file.h>
  32
  33 #include <net/sock.h>
  34 #include <net/checksum.h>
  35 #include <net/udp.h>
  36 #include <net/tcp.h>
  37
  38 /*
  39  * xprtsock tunables
  40  */
  41 unsigned int xprt_udp_slot_table_entries = RPC_DEF_SLOT_TABLE;
  42 unsigned int xprt_tcp_slot_table_entries = RPC_DEF_SLOT_TABLE;
  43
  44 unsigned int xprt_min_resvport = RPC_DEF_MIN_RESVPORT;
  45 unsigned int xprt_max_resvport = RPC_DEF_MAX_RESVPORT;
  46
  47 /*
  48  * We can register our own files under /proc/sys/sunrpc by
  49  * calling register_sysctl_table() again.  The files in that
  50  * directory become the union of all files registered there.
  51  *
  52  * We simply need to make sure that we don't collide with
  53  * someone else's file names!
  54  */
  55
  56 #ifdef RPC_DEBUG
  57
  58 static unsigned int min_slot_table_size = RPC_MIN_SLOT_TABLE;
  59 static unsigned int max_slot_table_size = RPC_MAX_SLOT_TABLE;
  60 static unsigned int xprt_min_resvport_limit = RPC_MIN_RESVPORT;
  61 static unsigned int xprt_max_resvport_limit = RPC_MAX_RESVPORT;
  62
  63 static struct ctl_table_header *sunrpc_table_header;
  64
  65 /*
  66  * FIXME: changing the UDP slot table size should also resize the UDP
  67  *        socket buffers for existing UDP transports
  68  */
  69 static ctl_table xs_tunables_table[] = {
  70         {
  71                 .ctl_name       = CTL_SLOTTABLE_UDP,
  72                 .procname       = "udp_slot_table_entries",
  73                 .data           = &xprt_udp_slot_table_entries,
  74                 .maxlen         = sizeof(unsigned int),
  75                 .mode           = 0644,
  76                 .proc_handler   = &proc_dointvec_minmax,
  77                 .strategy       = &sysctl_intvec,
  78                 .extra1         = &min_slot_table_size,
  79                 .extra2         = &max_slot_table_size
  80         },
  81         {
  82                 .ctl_name       = CTL_SLOTTABLE_TCP,
  83                 .procname       = "tcp_slot_table_entries",
  84                 .data           = &xprt_tcp_slot_table_entries,
  85                 .maxlen         = sizeof(unsigned int),
  86                 .mode           = 0644,
  87                 .proc_handler   = &proc_dointvec_minmax,
  88                 .strategy       = &sysctl_intvec,
  89                 .extra1         = &min_slot_table_size,
  90                 .extra2         = &max_slot_table_size
  91         },
  92         {
  93                 .ctl_name       = CTL_MIN_RESVPORT,
  94                 .procname       = "min_resvport",
  95                 .data           = &xprt_min_resvport,
  96                 .maxlen         = sizeof(unsigned int),
  97                 .mode           = 0644,
  98                 .proc_handler   = &proc_dointvec_minmax,
  99                 .strategy       = &sysctl_intvec,
 100                 .extra1         = &xprt_min_resvport_limit,
 101                 .extra2         = &xprt_max_resvport_limit
 102         },
 103         {
 104                 .ctl_name       = CTL_MAX_RESVPORT,
 105                 .procname       = "max_resvport",
 106                 .data           = &xprt_max_resvport,
 107                 .maxlen         = sizeof(unsigned int),
 108                 .mode           = 0644,
 109                 .proc_handler   = &proc_dointvec_minmax,
 110                 .strategy       = &sysctl_intvec,
 111                 .extra1         = &xprt_min_resvport_limit,
 112                 .extra2         = &xprt_max_resvport_limit
 113         },
 114         {
 115                 .ctl_name = 0,
 116         },
 117 };
 118
 119 static ctl_table sunrpc_table[] = {
 120         {
 121                 .ctl_name       = CTL_SUNRPC,
 122                 .procname       = "sunrpc",
 123                 .mode           = 0555,
 124                 .child          = xs_tunables_table
 125         },
 126         {
 127                 .ctl_name = 0,
 128         },
 129 };
 130
 131 #endif
 132
 133 /*
 134  * How many times to try sending a request on a socket before waiting
 135  * for the socket buffer to clear.
 136  */
 137 #define XS_SENDMSG_RETRY        (10U)
 138
 139 /*
 140  * Time out for an RPC UDP socket connect.  UDP socket connects are
 141  * synchronous, but we set a timeout anyway in case of resource
 142  * exhaustion on the local host.
 143  */
 144 #define XS_UDP_CONN_TO          (5U * HZ)
 145
 146 /*
 147  * Wait duration for an RPC TCP connection to be established.  Solaris
 148  * NFS over TCP uses 60 seconds, for example, which is in line with how
 149  * long a server takes to reboot.
 150  */
 151 #define XS_TCP_CONN_TO          (60U * HZ)
 152
 153 /*
 154  * Wait duration for a reply from the RPC portmapper.
 155  */
 156 #define XS_BIND_TO              (60U * HZ)
 157
 158 /*
 159  * Delay if a UDP socket connect error occurs.  This is most likely some
 160  * kind of resource problem on the local host.
 161  */
 162 #define XS_UDP_REEST_TO         (2U * HZ)
 163
 164 /*
 165  * The reestablish timeout allows clients to delay for a bit before attempting
 166  * to reconnect to a server that just dropped our connection.
 167  *
 168  * We implement an exponential backoff when trying to reestablish a TCP
 169  * transport connection with the server.  Some servers like to drop a TCP
 170  * connection when they are overworked, so we start with a short timeout and
 171  * increase over time if the server is down or not responding.
 172  */
 173 #define XS_TCP_INIT_REEST_TO    (3U * HZ)
 174 #define XS_TCP_MAX_REEST_TO     (5U * 60 * HZ)
 175
 176 /*
 177  * TCP idle timeout; client drops the transport socket if it is idle
 178  * for this long.  Note that we also timeout UDP sockets to prevent
 179  * holding port numbers when there is no RPC traffic.
 180  */
 181 #define XS_IDLE_DISC_TO         (5U * 60 * HZ)
 182
 183 #ifdef RPC_DEBUG
 184 # undef  RPC_DEBUG_DATA
 185 # define RPCDBG_FACILITY        RPCDBG_TRANS
 186 #endif
 187
 188 #ifdef RPC_DEBUG_DATA
 189 static void xs_pktdump(char *msg, u32 *packet, unsigned int count)
 190 {
 191         u8 *buf = (u8 *) packet;
 192         int j;
 193
 194         dprintk("RPC:       %s\n", msg);
 195         for (j = 0; j < count && j < 128; j += 4) {
 196                 if (!(j & 31)) {
 197                         if (j)
 198                                 dprintk("\n");
 199                         dprintk("0x%04x ", j);
 200                 }
 201                 dprintk("%02x%02x%02x%02x ",
 202                         buf[j], buf[j+1], buf[j+2], buf[j+3]);
 203         }
 204         dprintk("\n");
 205 }
 206 #else
 207 static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count)
 208 {
 209         /* NOP */
 210 }
 211 #endif
 212
 213 struct sock_xprt {
 214         struct rpc_xprt         xprt;
 215
 216         /*
 217          * Network layer
 218          */
 219         struct socket *         sock;
 220         struct sock *           inet;
 221
 222         /*
 223          * State of TCP reply receive
 224          */
 225         __be32                  tcp_fraghdr,
 226                                 tcp_xid;
 227
 228         u32                     tcp_offset,
 229                                 tcp_reclen;
 230
 231         unsigned long           tcp_copied,
 232                                 tcp_flags;
 233
 234         /*
 235          * Connection of transports
 236          */
 237         struct delayed_work     connect_worker;
 238         unsigned short          port;
 239
 240         /*
 241          * UDP socket buffer size parameters
 242          */
 243         size_t                  rcvsize,
 244                                 sndsize;
 245
 246         /*
 247          * Saved socket callback addresses
 248          */
 249         void                    (*old_data_ready)(struct sock *, int);
 250         void                    (*old_state_change)(struct sock *);
 251         void                    (*old_write_space)(struct sock *);
 252 };
 253
 254 /*
 255  * TCP receive state flags
 256  */
 257 #define TCP_RCV_LAST_FRAG       (1UL << 0)
 258 #define TCP_RCV_COPY_FRAGHDR    (1UL << 1)
 259 #define TCP_RCV_COPY_XID        (1UL << 2)
 260 #define TCP_RCV_COPY_DATA       (1UL << 3)
 261
 262 static void xs_format_peer_addresses(struct rpc_xprt *xprt)
 263 {
 264         struct sockaddr_in *addr = (struct sockaddr_in *) &xprt->addr;
 265         char *buf;
 266
 267         buf = kzalloc(20, GFP_KERNEL);
 268         if (buf) {
 269                 snprintf(buf, 20, "%u.%u.%u.%u",
 270                                 NIPQUAD(addr->sin_addr.s_addr));
 271         }
 272         xprt->address_strings[RPC_DISPLAY_ADDR] = buf;
 273
 274         buf = kzalloc(8, GFP_KERNEL);
 275         if (buf) {
 276                 snprintf(buf, 8, "%u",
 277                                 ntohs(addr->sin_port));
 278         }
 279         xprt->address_strings[RPC_DISPLAY_PORT] = buf;
 280
 281         if (xprt->prot == IPPROTO_UDP)
 282                 xprt->address_strings[RPC_DISPLAY_PROTO] = "udp";
 283         else
 284                 xprt->address_strings[RPC_DISPLAY_PROTO] = "tcp";
 285
 286         buf = kzalloc(48, GFP_KERNEL);
 287         if (buf) {
 288                 snprintf(buf, 48, "addr=%u.%u.%u.%u port=%u proto=%s",
 289                         NIPQUAD(addr->sin_addr.s_addr),
 290                         ntohs(addr->sin_port),
 291                         xprt->prot == IPPROTO_UDP ? "udp" : "tcp");
 292         }
 293         xprt->address_strings[RPC_DISPLAY_ALL] = buf;
 294 }
 295
 296 static void xs_free_peer_addresses(struct rpc_xprt *xprt)
 297 {
 298         kfree(xprt->address_strings[RPC_DISPLAY_ADDR]);
 299         kfree(xprt->address_strings[RPC_DISPLAY_PORT]);
 300         kfree(xprt->address_strings[RPC_DISPLAY_ALL]);
 301 }
 302
 303 #define XS_SENDMSG_FLAGS        (MSG_DONTWAIT | MSG_NOSIGNAL)
 304
 305 static int xs_send_kvec(struct socket *sock, struct sockaddr *addr, int addrlen, struct kvec *vec, unsigned int base, int more)
 306 {
 307         struct msghdr msg = {
 308                 .msg_name       = addr,
 309                 .msg_namelen    = addrlen,
 310                 .msg_flags      = XS_SENDMSG_FLAGS | (more ? MSG_MORE : 0),
 311         };
 312         struct kvec iov = {
 313                 .iov_base       = vec->iov_base + base,
 314                 .iov_len        = vec->iov_len - base,
 315         };
 316
 317         if (iov.iov_len != 0)
 318                 return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
 319         return kernel_sendmsg(sock, &msg, NULL, 0, 0);
 320 }
 321
 322 static int xs_send_pagedata(struct socket *sock, struct xdr_buf *xdr, unsigned int base, int more)
 323 {
 324         struct page **ppage;
 325         unsigned int remainder;
 326         int err, sent = 0;
 327
 328         remainder = xdr->page_len - base;
 329         base += xdr->page_base;
 330         ppage = xdr->pages + (base >> PAGE_SHIFT);
 331         base &= ~PAGE_MASK;
 332         for(;;) {
 333                 unsigned int len = min_t(unsigned int, PAGE_SIZE - base, remainder);
 334                 int flags = XS_SENDMSG_FLAGS;
 335
 336                 remainder -= len;
 337                 if (remainder != 0 || more)
 338                         flags |= MSG_MORE;
 339                 err = sock->ops->sendpage(sock, *ppage, base, len, flags);
 340                 if (remainder == 0 || err != len)
 341                         break;
 342                 sent += err;
 343                 ppage++;
 344                 base = 0;
 345         }
 346         if (sent == 0)
 347                 return err;
 348         if (err > 0)
 349                 sent += err;
 350         return sent;
 351 }
 352
 353 /**
 354  * xs_sendpages - write pages directly to a socket
 355  * @sock: socket to send on
 356  * @addr: UDP only -- address of destination
 357  * @addrlen: UDP only -- length of destination address
 358  * @xdr: buffer containing this request
 359  * @base: starting position in the buffer
 360  *
 361  */
 362 static int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base)
 363 {
 364         unsigned int remainder = xdr->len - base;
 365         int err, sent = 0;
 366
 367         if (unlikely(!sock))
 368                 return -ENOTCONN;
 369
 370         clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags);
 371         if (base != 0) {
 372                 addr = NULL;
 373                 addrlen = 0;
 374         }
 375
 376         if (base < xdr->head[0].iov_len || addr != NULL) {
 377                 unsigned int len = xdr->head[0].iov_len - base;
 378                 remainder -= len;
 379                 err = xs_send_kvec(sock, addr, addrlen, &xdr->head[0], base, remainder != 0);
 380                 if (remainder == 0 || err != len)
 381                         goto out;
 382                 sent += err;
 383                 base = 0;
 384         } else
 385                 base -= xdr->head[0].iov_len;
 386
 387         if (base < xdr->page_len) {
 388                 unsigned int len = xdr->page_len - base;
 389                 remainder -= len;
 390                 err = xs_send_pagedata(sock, xdr, base, remainder != 0);
 391                 if (remainder == 0 || err != len)
 392                         goto out;
 393                 sent += err;
 394                 base = 0;
 395         } else
 396                 base -= xdr->page_len;
 397
 398         if (base >= xdr->tail[0].iov_len)
 399                 return sent;
 400         err = xs_send_kvec(sock, NULL, 0, &xdr->tail[0], base, 0);
 401 out:
 402         if (sent == 0)
 403                 return err;
 404         if (err > 0)
 405                 sent += err;
 406         return sent;
 407 }
 408
 409 /**
 410  * xs_nospace - place task on wait queue if transmit was incomplete
 411  * @task: task to put to sleep
 412  *
 413  */
 414 static void xs_nospace(struct rpc_task *task)
 415 {
 416         struct rpc_rqst *req = task->tk_rqstp;
 417         struct rpc_xprt *xprt = req->rq_xprt;
 418         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 419
 420         dprintk("RPC: %5u xmit incomplete (%u left of %u)\n",
 421                         task->tk_pid, req->rq_slen - req->rq_bytes_sent,
 422                         req->rq_slen);
 423
 424         if (test_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags)) {
 425                 /* Protect against races with write_space */
 426                 spin_lock_bh(&xprt->transport_lock);
 427
 428                 /* Don't race with disconnect */
 429                 if (!xprt_connected(xprt))
 430                         task->tk_status = -ENOTCONN;
 431                 else if (test_bit(SOCK_NOSPACE, &transport->sock->flags))
 432                         xprt_wait_for_buffer_space(task);
 433
 434                 spin_unlock_bh(&xprt->transport_lock);
 435         } else
 436                 /* Keep holding the socket if it is blocked */
 437                 rpc_delay(task, HZ>>4);
 438 }
 439
 440 /**
 441  * xs_udp_send_request - write an RPC request to a UDP socket
 442  * @task: address of RPC task that manages the state of an RPC request
 443  *
 444  * Return values:
 445  *        0:    The request has been sent
 446  *   EAGAIN:    The socket was blocked, please call again later to
 447  *              complete the request
 448  * ENOTCONN:    Caller needs to invoke connect logic then call again
 449  *    other:    Some other error occured, the request was not sent
 450  */
 451 static int xs_udp_send_request(struct rpc_task *task)
 452 {
 453         struct rpc_rqst *req = task->tk_rqstp;
 454         struct rpc_xprt *xprt = req->rq_xprt;
 455         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 456         struct xdr_buf *xdr = &req->rq_snd_buf;
 457         int status;
 458
 459         xs_pktdump("packet data:",
 460                                 req->rq_svec->iov_base,
 461                                 req->rq_svec->iov_len);
 462
 463         req->rq_xtime = jiffies;
 464         status = xs_sendpages(transport->sock,
 465                               (struct sockaddr *) &xprt->addr,
 466                               xprt->addrlen, xdr,
 467                               req->rq_bytes_sent);
 468
 469         dprintk("RPC:       xs_udp_send_request(%u) = %d\n",
 470                         xdr->len - req->rq_bytes_sent, status);
 471
 472         if (likely(status >= (int) req->rq_slen))
 473                 return 0;
 474
 475         /* Still some bytes left; set up for a retry later. */
 476         if (status > 0)
 477                 status = -EAGAIN;
 478
 479         switch (status) {
 480         case -ENETUNREACH:
 481         case -EPIPE:
 482         case -ECONNREFUSED:
 483                 /* When the server has died, an ICMP port unreachable message
 484                  * prompts ECONNREFUSED. */
 485                 break;
 486         case -EAGAIN:
 487                 xs_nospace(task);
 488                 break;
 489         default:
 490                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
 491                         -status);
 492                 break;
 493         }
 494
 495         return status;
 496 }
 497
 498 static inline void xs_encode_tcp_record_marker(struct xdr_buf *buf)
 499 {
 500         u32 reclen = buf->len - sizeof(rpc_fraghdr);
 501         rpc_fraghdr *base = buf->head[0].iov_base;
 502         *base = htonl(RPC_LAST_STREAM_FRAGMENT | reclen);
 503 }
 504
 505 /**
 506  * xs_tcp_send_request - write an RPC request to a TCP socket
 507  * @task: address of RPC task that manages the state of an RPC request
 508  *
 509  * Return values:
 510  *        0:    The request has been sent
 511  *   EAGAIN:    The socket was blocked, please call again later to
 512  *              complete the request
 513  * ENOTCONN:    Caller needs to invoke connect logic then call again
 514  *    other:    Some other error occured, the request was not sent
 515  *
 516  * XXX: In the case of soft timeouts, should we eventually give up
 517  *      if sendmsg is not able to make progress?
 518  */
 519 static int xs_tcp_send_request(struct rpc_task *task)
 520 {
 521         struct rpc_rqst *req = task->tk_rqstp;
 522         struct rpc_xprt *xprt = req->rq_xprt;
 523         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 524         struct xdr_buf *xdr = &req->rq_snd_buf;
 525         int status, retry = 0;
 526
 527         xs_encode_tcp_record_marker(&req->rq_snd_buf);
 528
 529         xs_pktdump("packet data:",
 530                                 req->rq_svec->iov_base,
 531                                 req->rq_svec->iov_len);
 532
 533         /* Continue transmitting the packet/record. We must be careful
 534          * to cope with writespace callbacks arriving _after_ we have
 535          * called sendmsg(). */
 536         while (1) {
 537                 req->rq_xtime = jiffies;
 538                 status = xs_sendpages(transport->sock,
 539                                         NULL, 0, xdr, req->rq_bytes_sent);
 540
 541                 dprintk("RPC:       xs_tcp_send_request(%u) = %d\n",
 542                                 xdr->len - req->rq_bytes_sent, status);
 543
 544                 if (unlikely(status < 0))
 545                         break;
 546
 547                 /* If we've sent the entire packet, immediately
 548                  * reset the count of bytes sent. */
 549                 req->rq_bytes_sent += status;
 550                 task->tk_bytes_sent += status;
 551                 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
 552                         req->rq_bytes_sent = 0;
 553                         return 0;
 554                 }
 555
 556                 status = -EAGAIN;
 557                 if (retry++ > XS_SENDMSG_RETRY)
 558                         break;
 559         }
 560
 561         switch (status) {
 562         case -EAGAIN:
 563                 xs_nospace(task);
 564                 break;
 565         case -ECONNREFUSED:
 566         case -ECONNRESET:
 567         case -ENOTCONN:
 568         case -EPIPE:
 569                 status = -ENOTCONN;
 570                 break;
 571         default:
 572                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
 573                         -status);
 574                 xprt_disconnect(xprt);
 575                 break;
 576         }
 577
 578         return status;
 579 }
 580
 581 /**
 582  * xs_tcp_release_xprt - clean up after a tcp transmission
 583  * @xprt: transport
 584  * @task: rpc task
 585  *
 586  * This cleans up if an error causes us to abort the transmission of a request.
 587  * In this case, the socket may need to be reset in order to avoid confusing
 588  * the server.
 589  */
 590 static void xs_tcp_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
 591 {
 592         struct rpc_rqst *req;
 593
 594         if (task != xprt->snd_task)
 595                 return;
 596         if (task == NULL)
 597                 goto out_release;
 598         req = task->tk_rqstp;
 599         if (req->rq_bytes_sent == 0)
 600                 goto out_release;
 601         if (req->rq_bytes_sent == req->rq_snd_buf.len)
 602                 goto out_release;
 603         set_bit(XPRT_CLOSE_WAIT, &task->tk_xprt->state);
 604 out_release:
 605         xprt_release_xprt(xprt, task);
 606 }
 607
 608 /**
 609  * xs_close - close a socket
 610  * @xprt: transport
 611  *
 612  * This is used when all requests are complete; ie, no DRC state remains
 613  * on the server we want to save.
 614  */
 615 static void xs_close(struct rpc_xprt *xprt)
 616 {
 617         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 618         struct socket *sock = transport->sock;
 619         struct sock *sk = transport->inet;
 620
 621         if (!sk)
 622                 goto clear_close_wait;
 623
 624         dprintk("RPC:       xs_close xprt %p\n", xprt);
 625
 626         write_lock_bh(&sk->sk_callback_lock);
 627         transport->inet = NULL;
 628         transport->sock = NULL;
 629
 630         sk->sk_user_data = NULL;
 631         sk->sk_data_ready = transport->old_data_ready;
 632         sk->sk_state_change = transport->old_state_change;
 633         sk->sk_write_space = transport->old_write_space;
 634         write_unlock_bh(&sk->sk_callback_lock);
 635
 636         sk->sk_no_check = 0;
 637
 638         sock_release(sock);
 639 clear_close_wait:
 640         smp_mb__before_clear_bit();
 641         clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
 642         smp_mb__after_clear_bit();
 643 }
 644
 645 /**
 646  * xs_destroy - prepare to shutdown a transport
 647  * @xprt: doomed transport
 648  *
 649  */
 650 static void xs_destroy(struct rpc_xprt *xprt)
 651 {
 652         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 653
 654         dprintk("RPC:       xs_destroy xprt %p\n", xprt);
 655
 656         cancel_delayed_work(&transport->connect_worker);
 657         flush_scheduled_work();
 658
 659         xprt_disconnect(xprt);
 660         xs_close(xprt);
 661         xs_free_peer_addresses(xprt);
 662         kfree(xprt->slot);
 663         kfree(xprt);
 664 }
 665
 666 static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
 667 {
 668         return (struct rpc_xprt *) sk->sk_user_data;
 669 }
 670
 671 /**
 672  * xs_udp_data_ready - "data ready" callback for UDP sockets
 673  * @sk: socket with data to read
 674  * @len: how much data to read
 675  *
 676  */
 677 static void xs_udp_data_ready(struct sock *sk, int len)
 678 {
 679         struct rpc_task *task;
 680         struct rpc_xprt *xprt;
 681         struct rpc_rqst *rovr;
 682         struct sk_buff *skb;
 683         int err, repsize, copied;
 684         u32 _xid;
 685         __be32 *xp;
 686
 687         read_lock(&sk->sk_callback_lock);
 688         dprintk("RPC:       xs_udp_data_ready...\n");
 689         if (!(xprt = xprt_from_sock(sk)))
 690                 goto out;
 691
 692         if ((skb = skb_recv_datagram(sk, 0, 1, &err)) == NULL)
 693                 goto out;
 694
 695         if (xprt->shutdown)
 696                 goto dropit;
 697
 698         repsize = skb->len - sizeof(struct udphdr);
 699         if (repsize < 4) {
 700                 dprintk("RPC:       impossible RPC reply size %d!\n", repsize);
 701                 goto dropit;
 702         }
 703
 704         /* Copy the XID from the skb... */
 705         xp = skb_header_pointer(skb, sizeof(struct udphdr),
 706                                 sizeof(_xid), &_xid);
 707         if (xp == NULL)
 708                 goto dropit;
 709
 710         /* Look up and lock the request corresponding to the given XID */
 711         spin_lock(&xprt->transport_lock);
 712         rovr = xprt_lookup_rqst(xprt, *xp);
 713         if (!rovr)
 714                 goto out_unlock;
 715         task = rovr->rq_task;
 716
 717         if ((copied = rovr->rq_private_buf.buflen) > repsize)
 718                 copied = repsize;
 719
 720         /* Suck it into the iovec, verify checksum if not done by hw. */
 721         if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb))
 722                 goto out_unlock;
 723
 724         /* Something worked... */
 725         dst_confirm(skb->dst);
 726
 727         xprt_adjust_cwnd(task, copied);
 728         xprt_update_rtt(task);
 729         xprt_complete_rqst(task, copied);
 730
 731  out_unlock:
 732         spin_unlock(&xprt->transport_lock);
 733  dropit:
 734         skb_free_datagram(sk, skb);
 735  out:
 736         read_unlock(&sk->sk_callback_lock);
 737 }
 738
 739 static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
 740 {
 741         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 742         size_t len, used;
 743         char *p;
 744
 745         p = ((char *) &transport->tcp_fraghdr) + transport->tcp_offset;
 746         len = sizeof(transport->tcp_fraghdr) - transport->tcp_offset;
 747         used = xdr_skb_read_bits(desc, p, len);
 748         transport->tcp_offset += used;
 749         if (used != len)
 750                 return;
 751
 752         transport->tcp_reclen = ntohl(transport->tcp_fraghdr);
 753         if (transport->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
 754                 transport->tcp_flags |= TCP_RCV_LAST_FRAG;
 755         else
 756                 transport->tcp_flags &= ~TCP_RCV_LAST_FRAG;
 757         transport->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
 758
 759         transport->tcp_flags &= ~TCP_RCV_COPY_FRAGHDR;
 760         transport->tcp_offset = 0;
 761
 762         /* Sanity check of the record length */
 763         if (unlikely(transport->tcp_reclen < 4)) {
 764                 dprintk("RPC:       invalid TCP record fragment length\n");
 765                 xprt_disconnect(xprt);
 766                 return;
 767         }
 768         dprintk("RPC:       reading TCP record fragment of length %d\n",
 769                         transport->tcp_reclen);
 770 }
 771
 772 static void xs_tcp_check_fraghdr(struct sock_xprt *transport)
 773 {
 774         if (transport->tcp_offset == transport->tcp_reclen) {
 775                 transport->tcp_flags |= TCP_RCV_COPY_FRAGHDR;
 776                 transport->tcp_offset = 0;
 777                 if (transport->tcp_flags & TCP_RCV_LAST_FRAG) {
 778                         transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
 779                         transport->tcp_flags |= TCP_RCV_COPY_XID;
 780                         transport->tcp_copied = 0;
 781                 }
 782         }
 783 }
 784
 785 static inline void xs_tcp_read_xid(struct sock_xprt *transport, struct xdr_skb_reader *desc)
 786 {
 787         size_t len, used;
 788         char *p;
 789
 790         len = sizeof(transport->tcp_xid) - transport->tcp_offset;
 791         dprintk("RPC:       reading XID (%Zu bytes)\n", len);
 792         p = ((char *) &transport->tcp_xid) + transport->tcp_offset;
 793         used = xdr_skb_read_bits(desc, p, len);
 794         transport->tcp_offset += used;
 795         if (used != len)
 796                 return;
 797         transport->tcp_flags &= ~TCP_RCV_COPY_XID;
 798         transport->tcp_flags |= TCP_RCV_COPY_DATA;
 799         transport->tcp_copied = 4;
 800         dprintk("RPC:       reading reply for XID %08x\n",
 801                         ntohl(transport->tcp_xid));
 802         xs_tcp_check_fraghdr(transport);
 803 }
 804
 805 static inline void xs_tcp_read_request(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
 806 {
 807         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 808         struct rpc_rqst *req;
 809         struct xdr_buf *rcvbuf;
 810         size_t len;
 811         ssize_t r;
 812
 813         /* Find and lock the request corresponding to this xid */
 814         spin_lock(&xprt->transport_lock);
 815         req = xprt_lookup_rqst(xprt, transport->tcp_xid);
 816         if (!req) {
 817                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
 818                 dprintk("RPC:       XID %08x request not found!\n",
 819                                 ntohl(transport->tcp_xid));
 820                 spin_unlock(&xprt->transport_lock);
 821                 return;
 822         }
 823
 824         rcvbuf = &req->rq_private_buf;
 825         len = desc->count;
 826         if (len > transport->tcp_reclen - transport->tcp_offset) {
 827                 struct xdr_skb_reader my_desc;
 828
 829                 len = transport->tcp_reclen - transport->tcp_offset;
 830                 memcpy(&my_desc, desc, sizeof(my_desc));
 831                 my_desc.count = len;
 832                 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
 833                                           &my_desc, xdr_skb_read_bits);
 834                 desc->count -= r;
 835                 desc->offset += r;
 836         } else
 837                 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
 838                                           desc, xdr_skb_read_bits);
 839
 840         if (r > 0) {
 841                 transport->tcp_copied += r;
 842                 transport->tcp_offset += r;
 843         }
 844         if (r != len) {
 845                 /* Error when copying to the receive buffer,
 846                  * usually because we weren't able to allocate
 847                  * additional buffer pages. All we can do now
 848                  * is turn off TCP_RCV_COPY_DATA, so the request
 849                  * will not receive any additional updates,
 850                  * and time out.
 851                  * Any remaining data from this record will
 852                  * be discarded.
 853                  */
 854                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
 855                 dprintk("RPC:       XID %08x truncated request\n",
 856                                 ntohl(transport->tcp_xid));
 857                 dprintk("RPC:       xprt = %p, tcp_copied = %lu, "
 858                                 "tcp_offset = %u, tcp_reclen = %u\n",
 859                                 xprt, transport->tcp_copied,
 860                                 transport->tcp_offset, transport->tcp_reclen);
 861                 goto out;
 862         }
 863
 864         dprintk("RPC:       XID %08x read %Zd bytes\n",
 865                         ntohl(transport->tcp_xid), r);
 866         dprintk("RPC:       xprt = %p, tcp_copied = %lu, tcp_offset = %u, "
 867                         "tcp_reclen = %u\n", xprt, transport->tcp_copied,
 868                         transport->tcp_offset, transport->tcp_reclen);
 869
 870         if (transport->tcp_copied == req->rq_private_buf.buflen)
 871                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
 872         else if (transport->tcp_offset == transport->tcp_reclen) {
 873                 if (transport->tcp_flags & TCP_RCV_LAST_FRAG)
 874                         transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
 875         }
 876
 877 out:
 878         if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
 879                 xprt_complete_rqst(req->rq_task, transport->tcp_copied);
 880         spin_unlock(&xprt->transport_lock);
 881         xs_tcp_check_fraghdr(transport);
 882 }
 883
 884 static inline void xs_tcp_read_discard(struct sock_xprt *transport, struct xdr_skb_reader *desc)
 885 {
 886         size_t len;
 887
 888         len = transport->tcp_reclen - transport->tcp_offset;
 889         if (len > desc->count)
 890                 len = desc->count;
 891         desc->count -= len;
 892         desc->offset += len;
 893         transport->tcp_offset += len;
 894         dprintk("RPC:       discarded %Zu bytes\n", len);
 895         xs_tcp_check_fraghdr(transport);
 896 }
 897
 898 static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
 899 {
 900         struct rpc_xprt *xprt = rd_desc->arg.data;
 901         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 902         struct xdr_skb_reader desc = {
 903                 .skb    = skb,
 904                 .offset = offset,
 905                 .count  = len,
 906         };
 907
 908         dprintk("RPC:       xs_tcp_data_recv started\n");
 909         do {
 910                 /* Read in a new fragment marker if necessary */
 911                 /* Can we ever really expect to get completely empty fragments? */
 912                 if (transport->tcp_flags & TCP_RCV_COPY_FRAGHDR) {
 913                         xs_tcp_read_fraghdr(xprt, &desc);
 914                         continue;
 915                 }
 916                 /* Read in the xid if necessary */
 917                 if (transport->tcp_flags & TCP_RCV_COPY_XID) {
 918                         xs_tcp_read_xid(transport, &desc);
 919                         continue;
 920                 }
 921                 /* Read in the request data */
 922                 if (transport->tcp_flags & TCP_RCV_COPY_DATA) {
 923                         xs_tcp_read_request(xprt, &desc);
 924                         continue;
 925                 }
 926                 /* Skip over any trailing bytes on short reads */
 927                 xs_tcp_read_discard(transport, &desc);
 928         } while (desc.count);
 929         dprintk("RPC:       xs_tcp_data_recv done\n");
 930         return len - desc.count;
 931 }
 932
 933 /**
 934  * xs_tcp_data_ready - "data ready" callback for TCP sockets
 935  * @sk: socket with data to read
 936  * @bytes: how much data to read
 937  *
 938  */
 939 static void xs_tcp_data_ready(struct sock *sk, int bytes)
 940 {
 941         struct rpc_xprt *xprt;
 942         read_descriptor_t rd_desc;
 943
 944         dprintk("RPC:       xs_tcp_data_ready...\n");
 945
 946         read_lock(&sk->sk_callback_lock);
 947         if (!(xprt = xprt_from_sock(sk)))
 948                 goto out;
 949         if (xprt->shutdown)
 950                 goto out;
 951
 952         /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
 953         rd_desc.arg.data = xprt;
 954         rd_desc.count = 65536;
 955         tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
 956 out:
 957         read_unlock(&sk->sk_callback_lock);
 958 }
 959
 960 /**
 961  * xs_tcp_state_change - callback to handle TCP socket state changes
 962  * @sk: socket whose state has changed
 963  *
 964  */
 965 static void xs_tcp_state_change(struct sock *sk)
 966 {
 967         struct rpc_xprt *xprt;
 968
 969         read_lock(&sk->sk_callback_lock);
 970         if (!(xprt = xprt_from_sock(sk)))
 971                 goto out;
 972         dprintk("RPC:       xs_tcp_state_change client %p...\n", xprt);
 973         dprintk("RPC:       state %x conn %d dead %d zapped %d\n",
 974                         sk->sk_state, xprt_connected(xprt),
 975                         sock_flag(sk, SOCK_DEAD),
 976                         sock_flag(sk, SOCK_ZAPPED));
 977
 978         switch (sk->sk_state) {
 979         case TCP_ESTABLISHED:
 980                 spin_lock_bh(&xprt->transport_lock);
 981                 if (!xprt_test_and_set_connected(xprt)) {
 982                         struct sock_xprt *transport = container_of(xprt,
 983                                         struct sock_xprt, xprt);
 984
 985                         /* Reset TCP record info */
 986                         transport->tcp_offset = 0;
 987                         transport->tcp_reclen = 0;
 988                         transport->tcp_copied = 0;
 989                         transport->tcp_flags =
 990                                 TCP_RCV_COPY_FRAGHDR | TCP_RCV_COPY_XID;
 991
 992                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
 993                         xprt_wake_pending_tasks(xprt, 0);
 994                 }
 995                 spin_unlock_bh(&xprt->transport_lock);
 996                 break;
 997         case TCP_SYN_SENT:
 998         case TCP_SYN_RECV:
 999                 break;
1000         case TCP_CLOSE_WAIT:
1001                 /* Try to schedule an autoclose RPC calls */
1002                 set_bit(XPRT_CLOSE_WAIT, &xprt->state);
1003                 if (test_and_set_bit(XPRT_LOCKED, &xprt->state) == 0)
1004                         schedule_work(&xprt->task_cleanup);
1005         default:
1006                 xprt_disconnect(xprt);
1007         }
1008  out:
1009         read_unlock(&sk->sk_callback_lock);
1010 }
1011
1012 /**
1013  * xs_udp_write_space - callback invoked when socket buffer space
1014  *                             becomes available
1015  * @sk: socket whose state has changed
1016  *
1017  * Called when more output buffer space is available for this socket.
1018  * We try not to wake our writers until they can make "significant"
1019  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1020  * with a bunch of small requests.
1021  */
1022 static void xs_udp_write_space(struct sock *sk)
1023 {
1024         read_lock(&sk->sk_callback_lock);
1025
1026         /* from net/core/sock.c:sock_def_write_space */
1027         if (sock_writeable(sk)) {
1028                 struct socket *sock;
1029                 struct rpc_xprt *xprt;
1030
1031                 if (unlikely(!(sock = sk->sk_socket)))
1032                         goto out;
1033                 if (unlikely(!(xprt = xprt_from_sock(sk))))
1034                         goto out;
1035                 if (unlikely(!test_and_clear_bit(SOCK_NOSPACE, &sock->flags)))
1036                         goto out;
1037
1038                 xprt_write_space(xprt);
1039         }
1040
1041  out:
1042         read_unlock(&sk->sk_callback_lock);
1043 }
1044
1045 /**
1046  * xs_tcp_write_space - callback invoked when socket buffer space
1047  *                             becomes available
1048  * @sk: socket whose state has changed
1049  *
1050  * Called when more output buffer space is available for this socket.
1051  * We try not to wake our writers until they can make "significant"
1052  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1053  * with a bunch of small requests.
1054  */
1055 static void xs_tcp_write_space(struct sock *sk)
1056 {
1057         read_lock(&sk->sk_callback_lock);
1058
1059         /* from net/core/stream.c:sk_stream_write_space */
1060         if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk)) {
1061                 struct socket *sock;
1062                 struct rpc_xprt *xprt;
1063
1064                 if (unlikely(!(sock = sk->sk_socket)))
1065                         goto out;
1066                 if (unlikely(!(xprt = xprt_from_sock(sk))))
1067                         goto out;
1068                 if (unlikely(!test_and_clear_bit(SOCK_NOSPACE, &sock->flags)))
1069                         goto out;
1070
1071                 xprt_write_space(xprt);
1072         }
1073
1074  out:
1075         read_unlock(&sk->sk_callback_lock);
1076 }
1077
1078 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
1079 {
1080         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1081         struct sock *sk = transport->inet;
1082
1083         if (transport->rcvsize) {
1084                 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1085                 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
1086         }
1087         if (transport->sndsize) {
1088                 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1089                 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1090                 sk->sk_write_space(sk);
1091         }
1092 }
1093
1094 /**
1095  * xs_udp_set_buffer_size - set send and receive limits
1096  * @xprt: generic transport
1097  * @sndsize: requested size of send buffer, in bytes
1098  * @rcvsize: requested size of receive buffer, in bytes
1099  *
1100  * Set socket send and receive buffer size limits.
1101  */
1102 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1103 {
1104         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1105
1106         transport->sndsize = 0;
1107         if (sndsize)
1108                 transport->sndsize = sndsize + 1024;
1109         transport->rcvsize = 0;
1110         if (rcvsize)
1111                 transport->rcvsize = rcvsize + 1024;
1112
1113         xs_udp_do_set_buffer_size(xprt);
1114 }
1115
1116 /**
1117  * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1118  * @task: task that timed out
1119  *
1120  * Adjust the congestion window after a retransmit timeout has occurred.
1121  */
1122 static void xs_udp_timer(struct rpc_task *task)
1123 {
1124         xprt_adjust_cwnd(task, -ETIMEDOUT);
1125 }
1126
1127 static unsigned short xs_get_random_port(void)
1128 {
1129         unsigned short range = xprt_max_resvport - xprt_min_resvport;
1130         unsigned short rand = (unsigned short) net_random() % range;
1131         return rand + xprt_min_resvport;
1132 }
1133
1134 /**
1135  * xs_set_port - reset the port number in the remote endpoint address
1136  * @xprt: generic transport
1137  * @port: new port number
1138  *
1139  */
1140 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1141 {
1142         struct sockaddr_in *sap = (struct sockaddr_in *) &xprt->addr;
1143
1144         dprintk("RPC:       setting port for xprt %p to %u\n", xprt, port);
1145
1146         sap->sin_port = htons(port);
1147 }
1148
1149 static int xs_bindresvport(struct sock_xprt *transport, struct socket *sock)
1150 {
1151         struct sockaddr_in myaddr = {
1152                 .sin_family = AF_INET,
1153         };
1154         int err;
1155         unsigned short port = transport->port;
1156
1157         do {
1158                 myaddr.sin_port = htons(port);
1159                 err = kernel_bind(sock, (struct sockaddr *) &myaddr,
1160                                                 sizeof(myaddr));
1161                 if (err == 0) {
1162                         transport->port = port;
1163                         dprintk("RPC:       xs_bindresvport bound to port %u\n",
1164                                         port);
1165                         return 0;
1166                 }
1167                 if (port <= xprt_min_resvport)
1168                         port = xprt_max_resvport;
1169                 else
1170                         port--;
1171         } while (err == -EADDRINUSE && port != transport->port);
1172
1173         dprintk("RPC:       can't bind to reserved port (%d).\n", -err);
1174         return err;
1175 }
1176
1177 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1178 static struct lock_class_key xs_key[2];
1179 static struct lock_class_key xs_slock_key[2];
1180
1181 static inline void xs_reclassify_socket(struct socket *sock)
1182 {
1183         struct sock *sk = sock->sk;
1184         BUG_ON(sk->sk_lock.owner != NULL);
1185         switch (sk->sk_family) {
1186         case AF_INET:
1187                 sock_lock_init_class_and_name(sk, "slock-AF_INET-NFS",
1188                         &xs_slock_key[0], "sk_lock-AF_INET-NFS", &xs_key[0]);
1189                 break;
1190
1191         case AF_INET6:
1192                 sock_lock_init_class_and_name(sk, "slock-AF_INET6-NFS",
1193                         &xs_slock_key[1], "sk_lock-AF_INET6-NFS", &xs_key[1]);
1194                 break;
1195
1196         default:
1197                 BUG();
1198         }
1199 }
1200 #else
1201 static inline void xs_reclassify_socket(struct socket *sock)
1202 {
1203 }
1204 #endif
1205
1206 /**
1207  * xs_udp_connect_worker - set up a UDP socket
1208  * @work: RPC transport to connect
1209  *
1210  * Invoked by a work queue tasklet.
1211  */
1212 static void xs_udp_connect_worker(struct work_struct *work)
1213 {
1214         struct sock_xprt *transport =
1215                 container_of(work, struct sock_xprt, connect_worker.work);
1216         struct rpc_xprt *xprt = &transport->xprt;
1217         struct socket *sock = transport->sock;
1218         int err, status = -EIO;
1219
1220         if (xprt->shutdown || !xprt_bound(xprt))
1221                 goto out;
1222
1223         /* Start by resetting any existing state */
1224         xs_close(xprt);
1225
1226         if ((err = sock_create_kern(PF_INET, SOCK_DGRAM, IPPROTO_UDP, &sock)) < 0) {
1227                 dprintk("RPC:       can't create UDP transport socket (%d).\n", -err);
1228                 goto out;
1229         }
1230         xs_reclassify_socket(sock);
1231
1232         if (xprt->resvport && xs_bindresvport(transport, sock) < 0) {
1233                 sock_release(sock);
1234                 goto out;
1235         }
1236
1237         dprintk("RPC:       worker connecting xprt %p to address: %s\n",
1238                         xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1239
1240         if (!transport->inet) {
1241                 struct sock *sk = sock->sk;
1242
1243                 write_lock_bh(&sk->sk_callback_lock);
1244
1245                 sk->sk_user_data = xprt;
1246                 transport->old_data_ready = sk->sk_data_ready;
1247                 transport->old_state_change = sk->sk_state_change;
1248                 transport->old_write_space = sk->sk_write_space;
1249                 sk->sk_data_ready = xs_udp_data_ready;
1250                 sk->sk_write_space = xs_udp_write_space;
1251                 sk->sk_no_check = UDP_CSUM_NORCV;
1252                 sk->sk_allocation = GFP_ATOMIC;
1253
1254                 xprt_set_connected(xprt);
1255
1256                 /* Reset to new socket */
1257                 transport->sock = sock;
1258                 transport->inet = sk;
1259
1260                 write_unlock_bh(&sk->sk_callback_lock);
1261         }
1262         xs_udp_do_set_buffer_size(xprt);
1263         status = 0;
1264 out:
1265         xprt_wake_pending_tasks(xprt, status);
1266         xprt_clear_connecting(xprt);
1267 }
1268
1269 /*
1270  * We need to preserve the port number so the reply cache on the server can
1271  * find our cached RPC replies when we get around to reconnecting.
1272  */
1273 static void xs_tcp_reuse_connection(struct rpc_xprt *xprt)
1274 {
1275         int result;
1276         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1277         struct sockaddr any;
1278
1279         dprintk("RPC:       disconnecting xprt %p to reuse port\n", xprt);
1280
1281         /*
1282          * Disconnect the transport socket by doing a connect operation
1283          * with AF_UNSPEC.  This should return immediately...
1284          */
1285         memset(&any, 0, sizeof(any));
1286         any.sa_family = AF_UNSPEC;
1287         result = kernel_connect(transport->sock, &any, sizeof(any), 0);
1288         if (result)
1289                 dprintk("RPC:       AF_UNSPEC connect return code %d\n",
1290                                 result);
1291 }
1292
1293 /**
1294  * xs_tcp_connect_worker - connect a TCP socket to a remote endpoint
1295  * @work: RPC transport to connect
1296  *
1297  * Invoked by a work queue tasklet.
1298  */
1299 static void xs_tcp_connect_worker(struct work_struct *work)
1300 {
1301         struct sock_xprt *transport =
1302                 container_of(work, struct sock_xprt, connect_worker.work);
1303         struct rpc_xprt *xprt = &transport->xprt;
1304         struct socket *sock = transport->sock;
1305         int err, status = -EIO;
1306
1307         if (xprt->shutdown || !xprt_bound(xprt))
1308                 goto out;
1309
1310         if (!sock) {
1311                 /* start from scratch */
1312                 if ((err = sock_create_kern(PF_INET, SOCK_STREAM, IPPROTO_TCP, &sock)) < 0) {
1313                         dprintk("RPC:       can't create TCP transport "
1314                                         "socket (%d).\n", -err);
1315                         goto out;
1316                 }
1317                 xs_reclassify_socket(sock);
1318
1319                 if (xprt->resvport && xs_bindresvport(transport, sock) < 0) {
1320                         sock_release(sock);
1321                         goto out;
1322                 }
1323         } else
1324                 /* "close" the socket, preserving the local port */
1325                 xs_tcp_reuse_connection(xprt);
1326
1327         dprintk("RPC:       worker connecting xprt %p to address: %s\n",
1328                         xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1329
1330         if (!transport->inet) {
1331                 struct sock *sk = sock->sk;
1332
1333                 write_lock_bh(&sk->sk_callback_lock);
1334
1335                 sk->sk_user_data = xprt;
1336                 transport->old_data_ready = sk->sk_data_ready;
1337                 transport->old_state_change = sk->sk_state_change;
1338                 transport->old_write_space = sk->sk_write_space;
1339                 sk->sk_data_ready = xs_tcp_data_ready;
1340                 sk->sk_state_change = xs_tcp_state_change;
1341                 sk->sk_write_space = xs_tcp_write_space;
1342                 sk->sk_allocation = GFP_ATOMIC;
1343
1344                 /* socket options */
1345                 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
1346                 sock_reset_flag(sk, SOCK_LINGER);
1347                 tcp_sk(sk)->linger2 = 0;
1348                 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
1349
1350                 xprt_clear_connected(xprt);
1351
1352                 /* Reset to new socket */
1353                 transport->sock = sock;
1354                 transport->inet = sk;
1355
1356                 write_unlock_bh(&sk->sk_callback_lock);
1357         }
1358
1359         /* Tell the socket layer to start connecting... */
1360         xprt->stat.connect_count++;
1361         xprt->stat.connect_start = jiffies;
1362         status = kernel_connect(sock, (struct sockaddr *) &xprt->addr,
1363                         xprt->addrlen, O_NONBLOCK);
1364         dprintk("RPC:       %p connect status %d connected %d sock state %d\n",
1365                         xprt, -status, xprt_connected(xprt),
1366                         sock->sk->sk_state);
1367         if (status < 0) {
1368                 switch (status) {
1369                         case -EINPROGRESS:
1370                         case -EALREADY:
1371                                 goto out_clear;
1372                         case -ECONNREFUSED:
1373                         case -ECONNRESET:
1374                                 /* retry with existing socket, after a delay */
1375                                 break;
1376                         default:
1377                                 /* get rid of existing socket, and retry */
1378                                 xs_close(xprt);
1379                                 break;
1380                 }
1381         }
1382 out:
1383         xprt_wake_pending_tasks(xprt, status);
1384 out_clear:
1385         xprt_clear_connecting(xprt);
1386 }
1387
1388 /**
1389  * xs_connect - connect a socket to a remote endpoint
1390  * @task: address of RPC task that manages state of connect request
1391  *
1392  * TCP: If the remote end dropped the connection, delay reconnecting.
1393  *
1394  * UDP socket connects are synchronous, but we use a work queue anyway
1395  * to guarantee that even unprivileged user processes can set up a
1396  * socket on a privileged port.
1397  *
1398  * If a UDP socket connect fails, the delay behavior here prevents
1399  * retry floods (hard mounts).
1400  */
1401 static void xs_connect(struct rpc_task *task)
1402 {
1403         struct rpc_xprt *xprt = task->tk_xprt;
1404         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1405
1406         if (xprt_test_and_set_connecting(xprt))
1407                 return;
1408
1409         if (transport->sock != NULL) {
1410                 dprintk("RPC:       xs_connect delayed xprt %p for %lu "
1411                                 "seconds\n",
1412                                 xprt, xprt->reestablish_timeout / HZ);
1413                 schedule_delayed_work(&transport->connect_worker,
1414                                         xprt->reestablish_timeout);
1415                 xprt->reestablish_timeout <<= 1;
1416                 if (xprt->reestablish_timeout > XS_TCP_MAX_REEST_TO)
1417                         xprt->reestablish_timeout = XS_TCP_MAX_REEST_TO;
1418         } else {
1419                 dprintk("RPC:       xs_connect scheduled xprt %p\n", xprt);
1420                 schedule_delayed_work(&transport->connect_worker, 0);
1421
1422                 /* flush_scheduled_work can sleep... */
1423                 if (!RPC_IS_ASYNC(task))
1424                         flush_scheduled_work();
1425         }
1426 }
1427
1428 /**
1429  * xs_udp_print_stats - display UDP socket-specifc stats
1430  * @xprt: rpc_xprt struct containing statistics
1431  * @seq: output file
1432  *
1433  */
1434 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
1435 {
1436         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1437
1438         seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %Lu %Lu\n",
1439                         transport->port,
1440                         xprt->stat.bind_count,
1441                         xprt->stat.sends,
1442                         xprt->stat.recvs,
1443                         xprt->stat.bad_xids,
1444                         xprt->stat.req_u,
1445                         xprt->stat.bklog_u);
1446 }
1447
1448 /**
1449  * xs_tcp_print_stats - display TCP socket-specifc stats
1450  * @xprt: rpc_xprt struct containing statistics
1451  * @seq: output file
1452  *
1453  */
1454 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
1455 {
1456         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1457         long idle_time = 0;
1458
1459         if (xprt_connected(xprt))
1460                 idle_time = (long)(jiffies - xprt->last_used) / HZ;
1461
1462         seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu %Lu %Lu\n",
1463                         transport->port,
1464                         xprt->stat.bind_count,
1465                         xprt->stat.connect_count,
1466                         xprt->stat.connect_time,
1467                         idle_time,
1468                         xprt->stat.sends,
1469                         xprt->stat.recvs,
1470                         xprt->stat.bad_xids,
1471                         xprt->stat.req_u,
1472                         xprt->stat.bklog_u);
1473 }
1474
1475 static struct rpc_xprt_ops xs_udp_ops = {
1476         .set_buffer_size        = xs_udp_set_buffer_size,
1477         .reserve_xprt           = xprt_reserve_xprt_cong,
1478         .release_xprt           = xprt_release_xprt_cong,
1479         .rpcbind                = rpc_getport,
1480         .set_port               = xs_set_port,
1481         .connect                = xs_connect,
1482         .buf_alloc              = rpc_malloc,
1483         .buf_free               = rpc_free,
1484         .send_request           = xs_udp_send_request,
1485         .set_retrans_timeout    = xprt_set_retrans_timeout_rtt,
1486         .timer                  = xs_udp_timer,
1487         .release_request        = xprt_release_rqst_cong,
1488         .close                  = xs_close,
1489         .destroy                = xs_destroy,
1490         .print_stats            = xs_udp_print_stats,
1491 };
1492
1493 static struct rpc_xprt_ops xs_tcp_ops = {
1494         .reserve_xprt           = xprt_reserve_xprt,
1495         .release_xprt           = xs_tcp_release_xprt,
1496         .rpcbind                = rpc_getport,
1497         .set_port               = xs_set_port,
1498         .connect                = xs_connect,
1499         .buf_alloc              = rpc_malloc,
1500         .buf_free               = rpc_free,
1501         .send_request           = xs_tcp_send_request,
1502         .set_retrans_timeout    = xprt_set_retrans_timeout_def,
1503         .close                  = xs_close,
1504         .destroy                = xs_destroy,
1505         .print_stats            = xs_tcp_print_stats,
1506 };
1507
1508 static struct rpc_xprt *xs_setup_xprt(struct sockaddr *addr, size_t addrlen, unsigned int slot_table_size)
1509 {
1510         struct rpc_xprt *xprt;
1511         struct sock_xprt *new;
1512
1513         if (addrlen > sizeof(xprt->addr)) {
1514                 dprintk("RPC:       xs_setup_xprt: address too large\n");
1515                 return ERR_PTR(-EBADF);
1516         }
1517
1518         new = kzalloc(sizeof(*new), GFP_KERNEL);
1519         if (new == NULL) {
1520                 dprintk("RPC:       xs_setup_xprt: couldn't allocate "
1521                                 "rpc_xprt\n");
1522                 return ERR_PTR(-ENOMEM);
1523         }
1524         xprt = &new->xprt;
1525
1526         xprt->max_reqs = slot_table_size;
1527         xprt->slot = kcalloc(xprt->max_reqs, sizeof(struct rpc_rqst), GFP_KERNEL);
1528         if (xprt->slot == NULL) {
1529                 kfree(xprt);
1530                 dprintk("RPC:       xs_setup_xprt: couldn't allocate slot "
1531                                 "table\n");
1532                 return ERR_PTR(-ENOMEM);
1533         }
1534
1535         memcpy(&xprt->addr, addr, addrlen);
1536         xprt->addrlen = addrlen;
1537         new->port = xs_get_random_port();
1538
1539         return xprt;
1540 }
1541
1542 /**
1543  * xs_setup_udp - Set up transport to use a UDP socket
1544  * @addr: address of remote server
1545  * @addrlen: length of address in bytes
1546  * @to:   timeout parameters
1547  *
1548  */
1549 struct rpc_xprt *xs_setup_udp(struct sockaddr *addr, size_t addrlen, struct rpc_timeout *to)
1550 {
1551         struct rpc_xprt *xprt;
1552         struct sock_xprt *transport;
1553
1554         xprt = xs_setup_xprt(addr, addrlen, xprt_udp_slot_table_entries);
1555         if (IS_ERR(xprt))
1556                 return xprt;
1557         transport = container_of(xprt, struct sock_xprt, xprt);
1558
1559         if (ntohs(((struct sockaddr_in *)addr)->sin_port) != 0)
1560                 xprt_set_bound(xprt);
1561
1562         xprt->prot = IPPROTO_UDP;
1563         xprt->tsh_size = 0;
1564         /* XXX: header size can vary due to auth type, IPv6, etc. */
1565         xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
1566
1567         INIT_DELAYED_WORK(&transport->connect_worker, xs_udp_connect_worker);
1568         xprt->bind_timeout = XS_BIND_TO;
1569         xprt->connect_timeout = XS_UDP_CONN_TO;
1570         xprt->reestablish_timeout = XS_UDP_REEST_TO;
1571         xprt->idle_timeout = XS_IDLE_DISC_TO;
1572
1573         xprt->ops = &xs_udp_ops;
1574
1575         if (to)
1576                 xprt->timeout = *to;
1577         else
1578                 xprt_set_timeout(&xprt->timeout, 5, 5 * HZ);
1579
1580         xs_format_peer_addresses(xprt);
1581         dprintk("RPC:       set up transport to address %s\n",
1582                         xprt->address_strings[RPC_DISPLAY_ALL]);
1583
1584         return xprt;
1585 }
1586
1587 /**
1588  * xs_setup_tcp - Set up transport to use a TCP socket
1589  * @addr: address of remote server
1590  * @addrlen: length of address in bytes
1591  * @to: timeout parameters
1592  *
1593  */
1594 struct rpc_xprt *xs_setup_tcp(struct sockaddr *addr, size_t addrlen, struct rpc_timeout *to)
1595 {
1596         struct rpc_xprt *xprt;
1597         struct sock_xprt *transport;
1598
1599         xprt = xs_setup_xprt(addr, addrlen, xprt_tcp_slot_table_entries);
1600         if (IS_ERR(xprt))
1601                 return xprt;
1602         transport = container_of(xprt, struct sock_xprt, xprt);
1603
1604         if (ntohs(((struct sockaddr_in *)addr)->sin_port) != 0)
1605                 xprt_set_bound(xprt);
1606
1607         xprt->prot = IPPROTO_TCP;
1608         xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
1609         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
1610
1611         INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_connect_worker);
1612         xprt->bind_timeout = XS_BIND_TO;
1613         xprt->connect_timeout = XS_TCP_CONN_TO;
1614         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1615         xprt->idle_timeout = XS_IDLE_DISC_TO;
1616
1617         xprt->ops = &xs_tcp_ops;
1618
1619         if (to)
1620                 xprt->timeout = *to;
1621         else
1622                 xprt_set_timeout(&xprt->timeout, 2, 60 * HZ);
1623
1624         xs_format_peer_addresses(xprt);
1625         dprintk("RPC:       set up transport to address %s\n",
1626                         xprt->address_strings[RPC_DISPLAY_ALL]);
1627
1628         return xprt;
1629 }
1630
1631 /**
1632  * init_socket_xprt - set up xprtsock's sysctls
1633  *
1634  */
1635 int init_socket_xprt(void)
1636 {
1637 #ifdef RPC_DEBUG
1638         if (!sunrpc_table_header) {
1639                 sunrpc_table_header = register_sysctl_table(sunrpc_table, 1);
1640 #ifdef CONFIG_PROC_FS
1641                 if (sunrpc_table[0].de)
1642                         sunrpc_table[0].de->owner = THIS_MODULE;
1643 #endif
1644         }
1645 #endif
1646
1647         return 0;
1648 }
1649
1650 /**
1651  * cleanup_socket_xprt - remove xprtsock's sysctls
1652  *
1653  */
1654 void cleanup_socket_xprt(void)
1655 {
1656 #ifdef RPC_DEBUG
1657         if (sunrpc_table_header) {
1658                 unregister_sysctl_table(sunrpc_table_header);
1659                 sunrpc_table_header = NULL;
1660         }
1661 #endif
1662 }