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  * IPv6 support contributed by Gilles Quillard, Bull Open Source, 2005.
  18  *   <gilles.quillard@bull.net>
  19  */
  20
  21 #include <linux/types.h>
  22 #include <linux/slab.h>
  23 #include <linux/module.h>
  24 #include <linux/capability.h>
  25 #include <linux/pagemap.h>
  26 #include <linux/errno.h>
  27 #include <linux/socket.h>
  28 #include <linux/in.h>
  29 #include <linux/net.h>
  30 #include <linux/mm.h>
  31 #include <linux/udp.h>
  32 #include <linux/tcp.h>
  33 #include <linux/sunrpc/clnt.h>
  34 #include <linux/sunrpc/sched.h>
  35 #include <linux/sunrpc/xprtsock.h>
  36 #include <linux/file.h>
  37
  38 #include <net/sock.h>
  39 #include <net/checksum.h>
  40 #include <net/udp.h>
  41 #include <net/tcp.h>
  42
  43 /*
  44  * xprtsock tunables
  45  */
  46 unsigned int xprt_udp_slot_table_entries = RPC_DEF_SLOT_TABLE;
  47 unsigned int xprt_tcp_slot_table_entries = RPC_DEF_SLOT_TABLE;
  48
  49 unsigned int xprt_min_resvport = RPC_DEF_MIN_RESVPORT;
  50 unsigned int xprt_max_resvport = RPC_DEF_MAX_RESVPORT;
  51
  52 /*
  53  * We can register our own files under /proc/sys/sunrpc by
  54  * calling register_sysctl_table() again.  The files in that
  55  * directory become the union of all files registered there.
  56  *
  57  * We simply need to make sure that we don't collide with
  58  * someone else's file names!
  59  */
  60
  61 #ifdef RPC_DEBUG
  62
  63 static unsigned int min_slot_table_size = RPC_MIN_SLOT_TABLE;
  64 static unsigned int max_slot_table_size = RPC_MAX_SLOT_TABLE;
  65 static unsigned int xprt_min_resvport_limit = RPC_MIN_RESVPORT;
  66 static unsigned int xprt_max_resvport_limit = RPC_MAX_RESVPORT;
  67
  68 static struct ctl_table_header *sunrpc_table_header;
  69
  70 /*
  71  * FIXME: changing the UDP slot table size should also resize the UDP
  72  *        socket buffers for existing UDP transports
  73  */
  74 static ctl_table xs_tunables_table[] = {
  75         {
  76                 .ctl_name       = CTL_SLOTTABLE_UDP,
  77                 .procname       = "udp_slot_table_entries",
  78                 .data           = &xprt_udp_slot_table_entries,
  79                 .maxlen         = sizeof(unsigned int),
  80                 .mode           = 0644,
  81                 .proc_handler   = &proc_dointvec_minmax,
  82                 .strategy       = &sysctl_intvec,
  83                 .extra1         = &min_slot_table_size,
  84                 .extra2         = &max_slot_table_size
  85         },
  86         {
  87                 .ctl_name       = CTL_SLOTTABLE_TCP,
  88                 .procname       = "tcp_slot_table_entries",
  89                 .data           = &xprt_tcp_slot_table_entries,
  90                 .maxlen         = sizeof(unsigned int),
  91                 .mode           = 0644,
  92                 .proc_handler   = &proc_dointvec_minmax,
  93                 .strategy       = &sysctl_intvec,
  94                 .extra1         = &min_slot_table_size,
  95                 .extra2         = &max_slot_table_size
  96         },
  97         {
  98                 .ctl_name       = CTL_MIN_RESVPORT,
  99                 .procname       = "min_resvport",
 100                 .data           = &xprt_min_resvport,
 101                 .maxlen         = sizeof(unsigned int),
 102                 .mode           = 0644,
 103                 .proc_handler   = &proc_dointvec_minmax,
 104                 .strategy       = &sysctl_intvec,
 105                 .extra1         = &xprt_min_resvport_limit,
 106                 .extra2         = &xprt_max_resvport_limit
 107         },
 108         {
 109                 .ctl_name       = CTL_MAX_RESVPORT,
 110                 .procname       = "max_resvport",
 111                 .data           = &xprt_max_resvport,
 112                 .maxlen         = sizeof(unsigned int),
 113                 .mode           = 0644,
 114                 .proc_handler   = &proc_dointvec_minmax,
 115                 .strategy       = &sysctl_intvec,
 116                 .extra1         = &xprt_min_resvport_limit,
 117                 .extra2         = &xprt_max_resvport_limit
 118         },
 119         {
 120                 .ctl_name = 0,
 121         },
 122 };
 123
 124 static ctl_table sunrpc_table[] = {
 125         {
 126                 .ctl_name       = CTL_SUNRPC,
 127                 .procname       = "sunrpc",
 128                 .mode           = 0555,
 129                 .child          = xs_tunables_table
 130         },
 131         {
 132                 .ctl_name = 0,
 133         },
 134 };
 135
 136 #endif
 137
 138 /*
 139  * How many times to try sending a request on a socket before waiting
 140  * for the socket buffer to clear.
 141  */
 142 #define XS_SENDMSG_RETRY        (10U)
 143
 144 /*
 145  * Time out for an RPC UDP socket connect.  UDP socket connects are
 146  * synchronous, but we set a timeout anyway in case of resource
 147  * exhaustion on the local host.
 148  */
 149 #define XS_UDP_CONN_TO          (5U * HZ)
 150
 151 /*
 152  * Wait duration for an RPC TCP connection to be established.  Solaris
 153  * NFS over TCP uses 60 seconds, for example, which is in line with how
 154  * long a server takes to reboot.
 155  */
 156 #define XS_TCP_CONN_TO          (60U * HZ)
 157
 158 /*
 159  * Wait duration for a reply from the RPC portmapper.
 160  */
 161 #define XS_BIND_TO              (60U * HZ)
 162
 163 /*
 164  * Delay if a UDP socket connect error occurs.  This is most likely some
 165  * kind of resource problem on the local host.
 166  */
 167 #define XS_UDP_REEST_TO         (2U * HZ)
 168
 169 /*
 170  * The reestablish timeout allows clients to delay for a bit before attempting
 171  * to reconnect to a server that just dropped our connection.
 172  *
 173  * We implement an exponential backoff when trying to reestablish a TCP
 174  * transport connection with the server.  Some servers like to drop a TCP
 175  * connection when they are overworked, so we start with a short timeout and
 176  * increase over time if the server is down or not responding.
 177  */
 178 #define XS_TCP_INIT_REEST_TO    (3U * HZ)
 179 #define XS_TCP_MAX_REEST_TO     (5U * 60 * HZ)
 180
 181 /*
 182  * TCP idle timeout; client drops the transport socket if it is idle
 183  * for this long.  Note that we also timeout UDP sockets to prevent
 184  * holding port numbers when there is no RPC traffic.
 185  */
 186 #define XS_IDLE_DISC_TO         (5U * 60 * HZ)
 187
 188 #ifdef RPC_DEBUG
 189 # undef  RPC_DEBUG_DATA
 190 # define RPCDBG_FACILITY        RPCDBG_TRANS
 191 #endif
 192
 193 #ifdef RPC_DEBUG_DATA
 194 static void xs_pktdump(char *msg, u32 *packet, unsigned int count)
 195 {
 196         u8 *buf = (u8 *) packet;
 197         int j;
 198
 199         dprintk("RPC:       %s\n", msg);
 200         for (j = 0; j < count && j < 128; j += 4) {
 201                 if (!(j & 31)) {
 202                         if (j)
 203                                 dprintk("\n");
 204                         dprintk("0x%04x ", j);
 205                 }
 206                 dprintk("%02x%02x%02x%02x ",
 207                         buf[j], buf[j+1], buf[j+2], buf[j+3]);
 208         }
 209         dprintk("\n");
 210 }
 211 #else
 212 static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count)
 213 {
 214         /* NOP */
 215 }
 216 #endif
 217
 218 struct sock_xprt {
 219         struct rpc_xprt         xprt;
 220
 221         /*
 222          * Network layer
 223          */
 224         struct socket *         sock;
 225         struct sock *           inet;
 226
 227         /*
 228          * State of TCP reply receive
 229          */
 230         __be32                  tcp_fraghdr,
 231                                 tcp_xid;
 232
 233         u32                     tcp_offset,
 234                                 tcp_reclen;
 235
 236         unsigned long           tcp_copied,
 237                                 tcp_flags;
 238
 239         /*
 240          * Connection of transports
 241          */
 242         struct delayed_work     connect_worker;
 243         struct sockaddr_storage addr;
 244         unsigned short          port;
 245
 246         /*
 247          * UDP socket buffer size parameters
 248          */
 249         size_t                  rcvsize,
 250                                 sndsize;
 251
 252         /*
 253          * Saved socket callback addresses
 254          */
 255         void                    (*old_data_ready)(struct sock *, int);
 256         void                    (*old_state_change)(struct sock *);
 257         void                    (*old_write_space)(struct sock *);
 258 };
 259
 260 /*
 261  * TCP receive state flags
 262  */
 263 #define TCP_RCV_LAST_FRAG       (1UL << 0)
 264 #define TCP_RCV_COPY_FRAGHDR    (1UL << 1)
 265 #define TCP_RCV_COPY_XID        (1UL << 2)
 266 #define TCP_RCV_COPY_DATA       (1UL << 3)
 267
 268 static inline struct sockaddr *xs_addr(struct rpc_xprt *xprt)
 269 {
 270         return (struct sockaddr *) &xprt->addr;
 271 }
 272
 273 static inline struct sockaddr_in *xs_addr_in(struct rpc_xprt *xprt)
 274 {
 275         return (struct sockaddr_in *) &xprt->addr;
 276 }
 277
 278 static inline struct sockaddr_in6 *xs_addr_in6(struct rpc_xprt *xprt)
 279 {
 280         return (struct sockaddr_in6 *) &xprt->addr;
 281 }
 282
 283 static void xs_format_ipv4_peer_addresses(struct rpc_xprt *xprt)
 284 {
 285         struct sockaddr_in *addr = xs_addr_in(xprt);
 286         char *buf;
 287
 288         buf = kzalloc(20, GFP_KERNEL);
 289         if (buf) {
 290                 snprintf(buf, 20, NIPQUAD_FMT,
 291                                 NIPQUAD(addr->sin_addr.s_addr));
 292         }
 293         xprt->address_strings[RPC_DISPLAY_ADDR] = buf;
 294
 295         buf = kzalloc(8, GFP_KERNEL);
 296         if (buf) {
 297                 snprintf(buf, 8, "%u",
 298                                 ntohs(addr->sin_port));
 299         }
 300         xprt->address_strings[RPC_DISPLAY_PORT] = buf;
 301
 302         buf = kzalloc(8, GFP_KERNEL);
 303         if (buf) {
 304                 if (xprt->prot == IPPROTO_UDP)
 305                         snprintf(buf, 8, "udp");
 306                 else
 307                         snprintf(buf, 8, "tcp");
 308         }
 309         xprt->address_strings[RPC_DISPLAY_PROTO] = buf;
 310
 311         buf = kzalloc(48, GFP_KERNEL);
 312         if (buf) {
 313                 snprintf(buf, 48, "addr="NIPQUAD_FMT" port=%u proto=%s",
 314                         NIPQUAD(addr->sin_addr.s_addr),
 315                         ntohs(addr->sin_port),
 316                         xprt->prot == IPPROTO_UDP ? "udp" : "tcp");
 317         }
 318         xprt->address_strings[RPC_DISPLAY_ALL] = buf;
 319
 320         buf = kzalloc(10, GFP_KERNEL);
 321         if (buf) {
 322                 snprintf(buf, 10, "%02x%02x%02x%02x",
 323                                 NIPQUAD(addr->sin_addr.s_addr));
 324         }
 325         xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = buf;
 326
 327         buf = kzalloc(8, GFP_KERNEL);
 328         if (buf) {
 329                 snprintf(buf, 8, "%4hx",
 330                                 ntohs(addr->sin_port));
 331         }
 332         xprt->address_strings[RPC_DISPLAY_HEX_PORT] = buf;
 333
 334         buf = kzalloc(30, GFP_KERNEL);
 335         if (buf) {
 336                 snprintf(buf, 30, NIPQUAD_FMT".%u.%u",
 337                                 NIPQUAD(addr->sin_addr.s_addr),
 338                                 ntohs(addr->sin_port) >> 8,
 339                                 ntohs(addr->sin_port) & 0xff);
 340         }
 341         xprt->address_strings[RPC_DISPLAY_UNIVERSAL_ADDR] = buf;
 342
 343         xprt->address_strings[RPC_DISPLAY_NETID] =
 344                 kstrdup(xprt->prot == IPPROTO_UDP ?
 345                         RPCBIND_NETID_UDP : RPCBIND_NETID_TCP, GFP_KERNEL);
 346 }
 347
 348 static void xs_format_ipv6_peer_addresses(struct rpc_xprt *xprt)
 349 {
 350         struct sockaddr_in6 *addr = xs_addr_in6(xprt);
 351         char *buf;
 352
 353         buf = kzalloc(40, GFP_KERNEL);
 354         if (buf) {
 355                 snprintf(buf, 40, NIP6_FMT,
 356                                 NIP6(addr->sin6_addr));
 357         }
 358         xprt->address_strings[RPC_DISPLAY_ADDR] = buf;
 359
 360         buf = kzalloc(8, GFP_KERNEL);
 361         if (buf) {
 362                 snprintf(buf, 8, "%u",
 363                                 ntohs(addr->sin6_port));
 364         }
 365         xprt->address_strings[RPC_DISPLAY_PORT] = buf;
 366
 367         buf = kzalloc(8, GFP_KERNEL);
 368         if (buf) {
 369                 if (xprt->prot == IPPROTO_UDP)
 370                         snprintf(buf, 8, "udp");
 371                 else
 372                         snprintf(buf, 8, "tcp");
 373         }
 374         xprt->address_strings[RPC_DISPLAY_PROTO] = buf;
 375
 376         buf = kzalloc(64, GFP_KERNEL);
 377         if (buf) {
 378                 snprintf(buf, 64, "addr="NIP6_FMT" port=%u proto=%s",
 379                                 NIP6(addr->sin6_addr),
 380                                 ntohs(addr->sin6_port),
 381                                 xprt->prot == IPPROTO_UDP ? "udp" : "tcp");
 382         }
 383         xprt->address_strings[RPC_DISPLAY_ALL] = buf;
 384
 385         buf = kzalloc(36, GFP_KERNEL);
 386         if (buf) {
 387                 snprintf(buf, 36, NIP6_SEQFMT,
 388                                 NIP6(addr->sin6_addr));
 389         }
 390         xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = buf;
 391
 392         buf = kzalloc(8, GFP_KERNEL);
 393         if (buf) {
 394                 snprintf(buf, 8, "%4hx",
 395                                 ntohs(addr->sin6_port));
 396         }
 397         xprt->address_strings[RPC_DISPLAY_HEX_PORT] = buf;
 398
 399         buf = kzalloc(50, GFP_KERNEL);
 400         if (buf) {
 401                 snprintf(buf, 50, NIP6_FMT".%u.%u",
 402                                 NIP6(addr->sin6_addr),
 403                                 ntohs(addr->sin6_port) >> 8,
 404                                 ntohs(addr->sin6_port) & 0xff);
 405         }
 406         xprt->address_strings[RPC_DISPLAY_UNIVERSAL_ADDR] = buf;
 407
 408         xprt->address_strings[RPC_DISPLAY_NETID] =
 409                 kstrdup(xprt->prot == IPPROTO_UDP ?
 410                         RPCBIND_NETID_UDP6 : RPCBIND_NETID_TCP6, GFP_KERNEL);
 411 }
 412
 413 static void xs_free_peer_addresses(struct rpc_xprt *xprt)
 414 {
 415         int i;
 416
 417         for (i = 0; i < RPC_DISPLAY_MAX; i++)
 418                 kfree(xprt->address_strings[i]);
 419 }
 420
 421 #define XS_SENDMSG_FLAGS        (MSG_DONTWAIT | MSG_NOSIGNAL)
 422
 423 static int xs_send_kvec(struct socket *sock, struct sockaddr *addr, int addrlen, struct kvec *vec, unsigned int base, int more)
 424 {
 425         struct msghdr msg = {
 426                 .msg_name       = addr,
 427                 .msg_namelen    = addrlen,
 428                 .msg_flags      = XS_SENDMSG_FLAGS | (more ? MSG_MORE : 0),
 429         };
 430         struct kvec iov = {
 431                 .iov_base       = vec->iov_base + base,
 432                 .iov_len        = vec->iov_len - base,
 433         };
 434
 435         if (iov.iov_len != 0)
 436                 return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
 437         return kernel_sendmsg(sock, &msg, NULL, 0, 0);
 438 }
 439
 440 static int xs_send_pagedata(struct socket *sock, struct xdr_buf *xdr, unsigned int base, int more)
 441 {
 442         struct page **ppage;
 443         unsigned int remainder;
 444         int err, sent = 0;
 445
 446         remainder = xdr->page_len - base;
 447         base += xdr->page_base;
 448         ppage = xdr->pages + (base >> PAGE_SHIFT);
 449         base &= ~PAGE_MASK;
 450         for(;;) {
 451                 unsigned int len = min_t(unsigned int, PAGE_SIZE - base, remainder);
 452                 int flags = XS_SENDMSG_FLAGS;
 453
 454                 remainder -= len;
 455                 if (remainder != 0 || more)
 456                         flags |= MSG_MORE;
 457                 err = sock->ops->sendpage(sock, *ppage, base, len, flags);
 458                 if (remainder == 0 || err != len)
 459                         break;
 460                 sent += err;
 461                 ppage++;
 462                 base = 0;
 463         }
 464         if (sent == 0)
 465                 return err;
 466         if (err > 0)
 467                 sent += err;
 468         return sent;
 469 }
 470
 471 /**
 472  * xs_sendpages - write pages directly to a socket
 473  * @sock: socket to send on
 474  * @addr: UDP only -- address of destination
 475  * @addrlen: UDP only -- length of destination address
 476  * @xdr: buffer containing this request
 477  * @base: starting position in the buffer
 478  *
 479  */
 480 static int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base)
 481 {
 482         unsigned int remainder = xdr->len - base;
 483         int err, sent = 0;
 484
 485         if (unlikely(!sock))
 486                 return -ENOTCONN;
 487
 488         clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags);
 489         if (base != 0) {
 490                 addr = NULL;
 491                 addrlen = 0;
 492         }
 493
 494         if (base < xdr->head[0].iov_len || addr != NULL) {
 495                 unsigned int len = xdr->head[0].iov_len - base;
 496                 remainder -= len;
 497                 err = xs_send_kvec(sock, addr, addrlen, &xdr->head[0], base, remainder != 0);
 498                 if (remainder == 0 || err != len)
 499                         goto out;
 500                 sent += err;
 501                 base = 0;
 502         } else
 503                 base -= xdr->head[0].iov_len;
 504
 505         if (base < xdr->page_len) {
 506                 unsigned int len = xdr->page_len - base;
 507                 remainder -= len;
 508                 err = xs_send_pagedata(sock, xdr, base, remainder != 0);
 509                 if (remainder == 0 || err != len)
 510                         goto out;
 511                 sent += err;
 512                 base = 0;
 513         } else
 514                 base -= xdr->page_len;
 515
 516         if (base >= xdr->tail[0].iov_len)
 517                 return sent;
 518         err = xs_send_kvec(sock, NULL, 0, &xdr->tail[0], base, 0);
 519 out:
 520         if (sent == 0)
 521                 return err;
 522         if (err > 0)
 523                 sent += err;
 524         return sent;
 525 }
 526
 527 /**
 528  * xs_nospace - place task on wait queue if transmit was incomplete
 529  * @task: task to put to sleep
 530  *
 531  */
 532 static void xs_nospace(struct rpc_task *task)
 533 {
 534         struct rpc_rqst *req = task->tk_rqstp;
 535         struct rpc_xprt *xprt = req->rq_xprt;
 536         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 537
 538         dprintk("RPC: %5u xmit incomplete (%u left of %u)\n",
 539                         task->tk_pid, req->rq_slen - req->rq_bytes_sent,
 540                         req->rq_slen);
 541
 542         if (test_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags)) {
 543                 /* Protect against races with write_space */
 544                 spin_lock_bh(&xprt->transport_lock);
 545
 546                 /* Don't race with disconnect */
 547                 if (!xprt_connected(xprt))
 548                         task->tk_status = -ENOTCONN;
 549                 else if (test_bit(SOCK_NOSPACE, &transport->sock->flags))
 550                         xprt_wait_for_buffer_space(task);
 551
 552                 spin_unlock_bh(&xprt->transport_lock);
 553         } else
 554                 /* Keep holding the socket if it is blocked */
 555                 rpc_delay(task, HZ>>4);
 556 }
 557
 558 /**
 559  * xs_udp_send_request - write an RPC request to a UDP socket
 560  * @task: address of RPC task that manages the state of an RPC request
 561  *
 562  * Return values:
 563  *        0:    The request has been sent
 564  *   EAGAIN:    The socket was blocked, please call again later to
 565  *              complete the request
 566  * ENOTCONN:    Caller needs to invoke connect logic then call again
 567  *    other:    Some other error occured, the request was not sent
 568  */
 569 static int xs_udp_send_request(struct rpc_task *task)
 570 {
 571         struct rpc_rqst *req = task->tk_rqstp;
 572         struct rpc_xprt *xprt = req->rq_xprt;
 573         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 574         struct xdr_buf *xdr = &req->rq_snd_buf;
 575         int status;
 576
 577         xs_pktdump("packet data:",
 578                                 req->rq_svec->iov_base,
 579                                 req->rq_svec->iov_len);
 580
 581         req->rq_xtime = jiffies;
 582         status = xs_sendpages(transport->sock,
 583                               xs_addr(xprt),
 584                               xprt->addrlen, xdr,
 585                               req->rq_bytes_sent);
 586
 587         dprintk("RPC:       xs_udp_send_request(%u) = %d\n",
 588                         xdr->len - req->rq_bytes_sent, status);
 589
 590         if (status >= 0) {
 591                 task->tk_bytes_sent += status;
 592                 if (status >= req->rq_slen)
 593                         return 0;
 594                 /* Still some bytes left; set up for a retry later. */
 595                 status = -EAGAIN;
 596         }
 597
 598         switch (status) {
 599         case -ENETUNREACH:
 600         case -EPIPE:
 601         case -ECONNREFUSED:
 602                 /* When the server has died, an ICMP port unreachable message
 603                  * prompts ECONNREFUSED. */
 604                 break;
 605         case -EAGAIN:
 606                 xs_nospace(task);
 607                 break;
 608         default:
 609                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
 610                         -status);
 611                 break;
 612         }
 613
 614         return status;
 615 }
 616
 617 /**
 618  * xs_tcp_shutdown - gracefully shut down a TCP socket
 619  * @xprt: transport
 620  *
 621  * Initiates a graceful shutdown of the TCP socket by calling the
 622  * equivalent of shutdown(SHUT_WR);
 623  */
 624 static void xs_tcp_shutdown(struct rpc_xprt *xprt)
 625 {
 626         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 627         struct socket *sock = transport->sock;
 628
 629         if (sock != NULL)
 630                 kernel_sock_shutdown(sock, SHUT_WR);
 631 }
 632
 633 static inline void xs_encode_tcp_record_marker(struct xdr_buf *buf)
 634 {
 635         u32 reclen = buf->len - sizeof(rpc_fraghdr);
 636         rpc_fraghdr *base = buf->head[0].iov_base;
 637         *base = htonl(RPC_LAST_STREAM_FRAGMENT | reclen);
 638 }
 639
 640 /**
 641  * xs_tcp_send_request - write an RPC request to a TCP socket
 642  * @task: address of RPC task that manages the state of an RPC request
 643  *
 644  * Return values:
 645  *        0:    The request has been sent
 646  *   EAGAIN:    The socket was blocked, please call again later to
 647  *              complete the request
 648  * ENOTCONN:    Caller needs to invoke connect logic then call again
 649  *    other:    Some other error occured, the request was not sent
 650  *
 651  * XXX: In the case of soft timeouts, should we eventually give up
 652  *      if sendmsg is not able to make progress?
 653  */
 654 static int xs_tcp_send_request(struct rpc_task *task)
 655 {
 656         struct rpc_rqst *req = task->tk_rqstp;
 657         struct rpc_xprt *xprt = req->rq_xprt;
 658         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 659         struct xdr_buf *xdr = &req->rq_snd_buf;
 660         int status;
 661         unsigned int retry = 0;
 662
 663         xs_encode_tcp_record_marker(&req->rq_snd_buf);
 664
 665         xs_pktdump("packet data:",
 666                                 req->rq_svec->iov_base,
 667                                 req->rq_svec->iov_len);
 668
 669         /* Continue transmitting the packet/record. We must be careful
 670          * to cope with writespace callbacks arriving _after_ we have
 671          * called sendmsg(). */
 672         while (1) {
 673                 req->rq_xtime = jiffies;
 674                 status = xs_sendpages(transport->sock,
 675                                         NULL, 0, xdr, req->rq_bytes_sent);
 676
 677                 dprintk("RPC:       xs_tcp_send_request(%u) = %d\n",
 678                                 xdr->len - req->rq_bytes_sent, status);
 679
 680                 if (unlikely(status < 0))
 681                         break;
 682
 683                 /* If we've sent the entire packet, immediately
 684                  * reset the count of bytes sent. */
 685                 req->rq_bytes_sent += status;
 686                 task->tk_bytes_sent += status;
 687                 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
 688                         req->rq_bytes_sent = 0;
 689                         return 0;
 690                 }
 691
 692                 status = -EAGAIN;
 693                 if (retry++ > XS_SENDMSG_RETRY)
 694                         break;
 695         }
 696
 697         switch (status) {
 698         case -EAGAIN:
 699                 xs_nospace(task);
 700                 break;
 701         case -ECONNREFUSED:
 702         case -ECONNRESET:
 703         case -ENOTCONN:
 704         case -EPIPE:
 705                 status = -ENOTCONN;
 706                 break;
 707         default:
 708                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
 709                         -status);
 710                 xs_tcp_shutdown(xprt);
 711                 break;
 712         }
 713
 714         return status;
 715 }
 716
 717 /**
 718  * xs_tcp_release_xprt - clean up after a tcp transmission
 719  * @xprt: transport
 720  * @task: rpc task
 721  *
 722  * This cleans up if an error causes us to abort the transmission of a request.
 723  * In this case, the socket may need to be reset in order to avoid confusing
 724  * the server.
 725  */
 726 static void xs_tcp_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
 727 {
 728         struct rpc_rqst *req;
 729
 730         if (task != xprt->snd_task)
 731                 return;
 732         if (task == NULL)
 733                 goto out_release;
 734         req = task->tk_rqstp;
 735         if (req->rq_bytes_sent == 0)
 736                 goto out_release;
 737         if (req->rq_bytes_sent == req->rq_snd_buf.len)
 738                 goto out_release;
 739         set_bit(XPRT_CLOSE_WAIT, &task->tk_xprt->state);
 740 out_release:
 741         xprt_release_xprt(xprt, task);
 742 }
 743
 744 /**
 745  * xs_close - close a socket
 746  * @xprt: transport
 747  *
 748  * This is used when all requests are complete; ie, no DRC state remains
 749  * on the server we want to save.
 750  */
 751 static void xs_close(struct rpc_xprt *xprt)
 752 {
 753         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 754         struct socket *sock = transport->sock;
 755         struct sock *sk = transport->inet;
 756
 757         if (!sk)
 758                 goto clear_close_wait;
 759
 760         dprintk("RPC:       xs_close xprt %p\n", xprt);
 761
 762         write_lock_bh(&sk->sk_callback_lock);
 763         transport->inet = NULL;
 764         transport->sock = NULL;
 765
 766         sk->sk_user_data = NULL;
 767         sk->sk_data_ready = transport->old_data_ready;
 768         sk->sk_state_change = transport->old_state_change;
 769         sk->sk_write_space = transport->old_write_space;
 770         write_unlock_bh(&sk->sk_callback_lock);
 771
 772         sk->sk_no_check = 0;
 773
 774         sock_release(sock);
 775 clear_close_wait:
 776         smp_mb__before_clear_bit();
 777         clear_bit(XPRT_CONNECTED, &xprt->state);
 778         clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
 779         clear_bit(XPRT_CLOSING, &xprt->state);
 780         smp_mb__after_clear_bit();
 781 }
 782
 783 /**
 784  * xs_destroy - prepare to shutdown a transport
 785  * @xprt: doomed transport
 786  *
 787  */
 788 static void xs_destroy(struct rpc_xprt *xprt)
 789 {
 790         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 791
 792         dprintk("RPC:       xs_destroy xprt %p\n", xprt);
 793
 794         cancel_rearming_delayed_work(&transport->connect_worker);
 795
 796         xprt_disconnect(xprt);
 797         xs_close(xprt);
 798         xs_free_peer_addresses(xprt);
 799         kfree(xprt->slot);
 800         kfree(xprt);
 801         module_put(THIS_MODULE);
 802 }
 803
 804 static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
 805 {
 806         return (struct rpc_xprt *) sk->sk_user_data;
 807 }
 808
 809 /**
 810  * xs_udp_data_ready - "data ready" callback for UDP sockets
 811  * @sk: socket with data to read
 812  * @len: how much data to read
 813  *
 814  */
 815 static void xs_udp_data_ready(struct sock *sk, int len)
 816 {
 817         struct rpc_task *task;
 818         struct rpc_xprt *xprt;
 819         struct rpc_rqst *rovr;
 820         struct sk_buff *skb;
 821         int err, repsize, copied;
 822         u32 _xid;
 823         __be32 *xp;
 824
 825         read_lock(&sk->sk_callback_lock);
 826         dprintk("RPC:       xs_udp_data_ready...\n");
 827         if (!(xprt = xprt_from_sock(sk)))
 828                 goto out;
 829
 830         if ((skb = skb_recv_datagram(sk, 0, 1, &err)) == NULL)
 831                 goto out;
 832
 833         if (xprt->shutdown)
 834                 goto dropit;
 835
 836         repsize = skb->len - sizeof(struct udphdr);
 837         if (repsize < 4) {
 838                 dprintk("RPC:       impossible RPC reply size %d!\n", repsize);
 839                 goto dropit;
 840         }
 841
 842         /* Copy the XID from the skb... */
 843         xp = skb_header_pointer(skb, sizeof(struct udphdr),
 844                                 sizeof(_xid), &_xid);
 845         if (xp == NULL)
 846                 goto dropit;
 847
 848         /* Look up and lock the request corresponding to the given XID */
 849         spin_lock(&xprt->transport_lock);
 850         rovr = xprt_lookup_rqst(xprt, *xp);
 851         if (!rovr)
 852                 goto out_unlock;
 853         task = rovr->rq_task;
 854
 855         if ((copied = rovr->rq_private_buf.buflen) > repsize)
 856                 copied = repsize;
 857
 858         /* Suck it into the iovec, verify checksum if not done by hw. */
 859         if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
 860                 UDPX_INC_STATS_BH(sk, UDP_MIB_INERRORS);
 861                 goto out_unlock;
 862         }
 863
 864         UDPX_INC_STATS_BH(sk, UDP_MIB_INDATAGRAMS);
 865
 866         /* Something worked... */
 867         dst_confirm(skb->dst);
 868
 869         xprt_adjust_cwnd(task, copied);
 870         xprt_update_rtt(task);
 871         xprt_complete_rqst(task, copied);
 872
 873  out_unlock:
 874         spin_unlock(&xprt->transport_lock);
 875  dropit:
 876         skb_free_datagram(sk, skb);
 877  out:
 878         read_unlock(&sk->sk_callback_lock);
 879 }
 880
 881 static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
 882 {
 883         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 884         size_t len, used;
 885         char *p;
 886
 887         p = ((char *) &transport->tcp_fraghdr) + transport->tcp_offset;
 888         len = sizeof(transport->tcp_fraghdr) - transport->tcp_offset;
 889         used = xdr_skb_read_bits(desc, p, len);
 890         transport->tcp_offset += used;
 891         if (used != len)
 892                 return;
 893
 894         transport->tcp_reclen = ntohl(transport->tcp_fraghdr);
 895         if (transport->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
 896                 transport->tcp_flags |= TCP_RCV_LAST_FRAG;
 897         else
 898                 transport->tcp_flags &= ~TCP_RCV_LAST_FRAG;
 899         transport->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
 900
 901         transport->tcp_flags &= ~TCP_RCV_COPY_FRAGHDR;
 902         transport->tcp_offset = 0;
 903
 904         /* Sanity check of the record length */
 905         if (unlikely(transport->tcp_reclen < 4)) {
 906                 dprintk("RPC:       invalid TCP record fragment length\n");
 907                 xprt_disconnect(xprt);
 908                 return;
 909         }
 910         dprintk("RPC:       reading TCP record fragment of length %d\n",
 911                         transport->tcp_reclen);
 912 }
 913
 914 static void xs_tcp_check_fraghdr(struct sock_xprt *transport)
 915 {
 916         if (transport->tcp_offset == transport->tcp_reclen) {
 917                 transport->tcp_flags |= TCP_RCV_COPY_FRAGHDR;
 918                 transport->tcp_offset = 0;
 919                 if (transport->tcp_flags & TCP_RCV_LAST_FRAG) {
 920                         transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
 921                         transport->tcp_flags |= TCP_RCV_COPY_XID;
 922                         transport->tcp_copied = 0;
 923                 }
 924         }
 925 }
 926
 927 static inline void xs_tcp_read_xid(struct sock_xprt *transport, struct xdr_skb_reader *desc)
 928 {
 929         size_t len, used;
 930         char *p;
 931
 932         len = sizeof(transport->tcp_xid) - transport->tcp_offset;
 933         dprintk("RPC:       reading XID (%Zu bytes)\n", len);
 934         p = ((char *) &transport->tcp_xid) + transport->tcp_offset;
 935         used = xdr_skb_read_bits(desc, p, len);
 936         transport->tcp_offset += used;
 937         if (used != len)
 938                 return;
 939         transport->tcp_flags &= ~TCP_RCV_COPY_XID;
 940         transport->tcp_flags |= TCP_RCV_COPY_DATA;
 941         transport->tcp_copied = 4;
 942         dprintk("RPC:       reading reply for XID %08x\n",
 943                         ntohl(transport->tcp_xid));
 944         xs_tcp_check_fraghdr(transport);
 945 }
 946
 947 static inline void xs_tcp_read_request(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
 948 {
 949         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 950         struct rpc_rqst *req;
 951         struct xdr_buf *rcvbuf;
 952         size_t len;
 953         ssize_t r;
 954
 955         /* Find and lock the request corresponding to this xid */
 956         spin_lock(&xprt->transport_lock);
 957         req = xprt_lookup_rqst(xprt, transport->tcp_xid);
 958         if (!req) {
 959                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
 960                 dprintk("RPC:       XID %08x request not found!\n",
 961                                 ntohl(transport->tcp_xid));
 962                 spin_unlock(&xprt->transport_lock);
 963                 return;
 964         }
 965
 966         rcvbuf = &req->rq_private_buf;
 967         len = desc->count;
 968         if (len > transport->tcp_reclen - transport->tcp_offset) {
 969                 struct xdr_skb_reader my_desc;
 970
 971                 len = transport->tcp_reclen - transport->tcp_offset;
 972                 memcpy(&my_desc, desc, sizeof(my_desc));
 973                 my_desc.count = len;
 974                 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
 975                                           &my_desc, xdr_skb_read_bits);
 976                 desc->count -= r;
 977                 desc->offset += r;
 978         } else
 979                 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
 980                                           desc, xdr_skb_read_bits);
 981
 982         if (r > 0) {
 983                 transport->tcp_copied += r;
 984                 transport->tcp_offset += r;
 985         }
 986         if (r != len) {
 987                 /* Error when copying to the receive buffer,
 988                  * usually because we weren't able to allocate
 989                  * additional buffer pages. All we can do now
 990                  * is turn off TCP_RCV_COPY_DATA, so the request
 991                  * will not receive any additional updates,
 992                  * and time out.
 993                  * Any remaining data from this record will
 994                  * be discarded.
 995                  */
 996                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
 997                 dprintk("RPC:       XID %08x truncated request\n",
 998                                 ntohl(transport->tcp_xid));
 999                 dprintk("RPC:       xprt = %p, tcp_copied = %lu, "
1000                                 "tcp_offset = %u, tcp_reclen = %u\n",
1001                                 xprt, transport->tcp_copied,
1002                                 transport->tcp_offset, transport->tcp_reclen);
1003                 goto out;
1004         }
1005
1006         dprintk("RPC:       XID %08x read %Zd bytes\n",
1007                         ntohl(transport->tcp_xid), r);
1008         dprintk("RPC:       xprt = %p, tcp_copied = %lu, tcp_offset = %u, "
1009                         "tcp_reclen = %u\n", xprt, transport->tcp_copied,
1010                         transport->tcp_offset, transport->tcp_reclen);
1011
1012         if (transport->tcp_copied == req->rq_private_buf.buflen)
1013                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1014         else if (transport->tcp_offset == transport->tcp_reclen) {
1015                 if (transport->tcp_flags & TCP_RCV_LAST_FRAG)
1016                         transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1017         }
1018
1019 out:
1020         if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1021                 xprt_complete_rqst(req->rq_task, transport->tcp_copied);
1022         spin_unlock(&xprt->transport_lock);
1023         xs_tcp_check_fraghdr(transport);
1024 }
1025
1026 static inline void xs_tcp_read_discard(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1027 {
1028         size_t len;
1029
1030         len = transport->tcp_reclen - transport->tcp_offset;
1031         if (len > desc->count)
1032                 len = desc->count;
1033         desc->count -= len;
1034         desc->offset += len;
1035         transport->tcp_offset += len;
1036         dprintk("RPC:       discarded %Zu bytes\n", len);
1037         xs_tcp_check_fraghdr(transport);
1038 }
1039
1040 static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
1041 {
1042         struct rpc_xprt *xprt = rd_desc->arg.data;
1043         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1044         struct xdr_skb_reader desc = {
1045                 .skb    = skb,
1046                 .offset = offset,
1047                 .count  = len,
1048         };
1049
1050         dprintk("RPC:       xs_tcp_data_recv started\n");
1051         do {
1052                 /* Read in a new fragment marker if necessary */
1053                 /* Can we ever really expect to get completely empty fragments? */
1054                 if (transport->tcp_flags & TCP_RCV_COPY_FRAGHDR) {
1055                         xs_tcp_read_fraghdr(xprt, &desc);
1056                         continue;
1057                 }
1058                 /* Read in the xid if necessary */
1059                 if (transport->tcp_flags & TCP_RCV_COPY_XID) {
1060                         xs_tcp_read_xid(transport, &desc);
1061                         continue;
1062                 }
1063                 /* Read in the request data */
1064                 if (transport->tcp_flags & TCP_RCV_COPY_DATA) {
1065                         xs_tcp_read_request(xprt, &desc);
1066                         continue;
1067                 }
1068                 /* Skip over any trailing bytes on short reads */
1069                 xs_tcp_read_discard(transport, &desc);
1070         } while (desc.count);
1071         dprintk("RPC:       xs_tcp_data_recv done\n");
1072         return len - desc.count;
1073 }
1074
1075 /**
1076  * xs_tcp_data_ready - "data ready" callback for TCP sockets
1077  * @sk: socket with data to read
1078  * @bytes: how much data to read
1079  *
1080  */
1081 static void xs_tcp_data_ready(struct sock *sk, int bytes)
1082 {
1083         struct rpc_xprt *xprt;
1084         read_descriptor_t rd_desc;
1085
1086         dprintk("RPC:       xs_tcp_data_ready...\n");
1087
1088         read_lock(&sk->sk_callback_lock);
1089         if (!(xprt = xprt_from_sock(sk)))
1090                 goto out;
1091         if (xprt->shutdown)
1092                 goto out;
1093
1094         /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
1095         rd_desc.arg.data = xprt;
1096         rd_desc.count = 65536;
1097         tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
1098 out:
1099         read_unlock(&sk->sk_callback_lock);
1100 }
1101
1102 /**
1103  * xs_tcp_state_change - callback to handle TCP socket state changes
1104  * @sk: socket whose state has changed
1105  *
1106  */
1107 static void xs_tcp_state_change(struct sock *sk)
1108 {
1109         struct rpc_xprt *xprt;
1110
1111         read_lock(&sk->sk_callback_lock);
1112         if (!(xprt = xprt_from_sock(sk)))
1113                 goto out;
1114         dprintk("RPC:       xs_tcp_state_change client %p...\n", xprt);
1115         dprintk("RPC:       state %x conn %d dead %d zapped %d\n",
1116                         sk->sk_state, xprt_connected(xprt),
1117                         sock_flag(sk, SOCK_DEAD),
1118                         sock_flag(sk, SOCK_ZAPPED));
1119
1120         switch (sk->sk_state) {
1121         case TCP_ESTABLISHED:
1122                 spin_lock_bh(&xprt->transport_lock);
1123                 if (!xprt_test_and_set_connected(xprt)) {
1124                         struct sock_xprt *transport = container_of(xprt,
1125                                         struct sock_xprt, xprt);
1126
1127                         /* Reset TCP record info */
1128                         transport->tcp_offset = 0;
1129                         transport->tcp_reclen = 0;
1130                         transport->tcp_copied = 0;
1131                         transport->tcp_flags =
1132                                 TCP_RCV_COPY_FRAGHDR | TCP_RCV_COPY_XID;
1133
1134                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1135                         xprt_wake_pending_tasks(xprt, 0);
1136                 }
1137                 spin_unlock_bh(&xprt->transport_lock);
1138                 break;
1139         case TCP_FIN_WAIT1:
1140                 /* The client initiated a shutdown of the socket */
1141                 set_bit(XPRT_CLOSING, &xprt->state);
1142                 smp_mb__before_clear_bit();
1143                 clear_bit(XPRT_CONNECTED, &xprt->state);
1144                 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1145                 smp_mb__after_clear_bit();
1146                 break;
1147         case TCP_CLOSE_WAIT:
1148                 /* The server initiated a shutdown of the socket */
1149                 set_bit(XPRT_CLOSING, &xprt->state);
1150                 xprt_force_disconnect(xprt);
1151                 break;
1152         case TCP_LAST_ACK:
1153                 smp_mb__before_clear_bit();
1154                 clear_bit(XPRT_CONNECTED, &xprt->state);
1155                 smp_mb__after_clear_bit();
1156                 break;
1157         case TCP_CLOSE:
1158                 smp_mb__before_clear_bit();
1159                 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1160                 clear_bit(XPRT_CLOSING, &xprt->state);
1161                 smp_mb__after_clear_bit();
1162                 /* Mark transport as closed and wake up all pending tasks */
1163                 xprt_disconnect(xprt);
1164         }
1165  out:
1166         read_unlock(&sk->sk_callback_lock);
1167 }
1168
1169 /**
1170  * xs_udp_write_space - callback invoked when socket buffer space
1171  *                             becomes available
1172  * @sk: socket whose state has changed
1173  *
1174  * Called when more output buffer space is available for this socket.
1175  * We try not to wake our writers until they can make "significant"
1176  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1177  * with a bunch of small requests.
1178  */
1179 static void xs_udp_write_space(struct sock *sk)
1180 {
1181         read_lock(&sk->sk_callback_lock);
1182
1183         /* from net/core/sock.c:sock_def_write_space */
1184         if (sock_writeable(sk)) {
1185                 struct socket *sock;
1186                 struct rpc_xprt *xprt;
1187
1188                 if (unlikely(!(sock = sk->sk_socket)))
1189                         goto out;
1190                 if (unlikely(!(xprt = xprt_from_sock(sk))))
1191                         goto out;
1192                 if (unlikely(!test_and_clear_bit(SOCK_NOSPACE, &sock->flags)))
1193                         goto out;
1194
1195                 xprt_write_space(xprt);
1196         }
1197
1198  out:
1199         read_unlock(&sk->sk_callback_lock);
1200 }
1201
1202 /**
1203  * xs_tcp_write_space - callback invoked when socket buffer space
1204  *                             becomes available
1205  * @sk: socket whose state has changed
1206  *
1207  * Called when more output buffer space is available for this socket.
1208  * We try not to wake our writers until they can make "significant"
1209  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1210  * with a bunch of small requests.
1211  */
1212 static void xs_tcp_write_space(struct sock *sk)
1213 {
1214         read_lock(&sk->sk_callback_lock);
1215
1216         /* from net/core/stream.c:sk_stream_write_space */
1217         if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk)) {
1218                 struct socket *sock;
1219                 struct rpc_xprt *xprt;
1220
1221                 if (unlikely(!(sock = sk->sk_socket)))
1222                         goto out;
1223                 if (unlikely(!(xprt = xprt_from_sock(sk))))
1224                         goto out;
1225                 if (unlikely(!test_and_clear_bit(SOCK_NOSPACE, &sock->flags)))
1226                         goto out;
1227
1228                 xprt_write_space(xprt);
1229         }
1230
1231  out:
1232         read_unlock(&sk->sk_callback_lock);
1233 }
1234
1235 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
1236 {
1237         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1238         struct sock *sk = transport->inet;
1239
1240         if (transport->rcvsize) {
1241                 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1242                 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
1243         }
1244         if (transport->sndsize) {
1245                 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1246                 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1247                 sk->sk_write_space(sk);
1248         }
1249 }
1250
1251 /**
1252  * xs_udp_set_buffer_size - set send and receive limits
1253  * @xprt: generic transport
1254  * @sndsize: requested size of send buffer, in bytes
1255  * @rcvsize: requested size of receive buffer, in bytes
1256  *
1257  * Set socket send and receive buffer size limits.
1258  */
1259 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1260 {
1261         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1262
1263         transport->sndsize = 0;
1264         if (sndsize)
1265                 transport->sndsize = sndsize + 1024;
1266         transport->rcvsize = 0;
1267         if (rcvsize)
1268                 transport->rcvsize = rcvsize + 1024;
1269
1270         xs_udp_do_set_buffer_size(xprt);
1271 }
1272
1273 /**
1274  * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1275  * @task: task that timed out
1276  *
1277  * Adjust the congestion window after a retransmit timeout has occurred.
1278  */
1279 static void xs_udp_timer(struct rpc_task *task)
1280 {
1281         xprt_adjust_cwnd(task, -ETIMEDOUT);
1282 }
1283
1284 static unsigned short xs_get_random_port(void)
1285 {
1286         unsigned short range = xprt_max_resvport - xprt_min_resvport;
1287         unsigned short rand = (unsigned short) net_random() % range;
1288         return rand + xprt_min_resvport;
1289 }
1290
1291 /**
1292  * xs_set_port - reset the port number in the remote endpoint address
1293  * @xprt: generic transport
1294  * @port: new port number
1295  *
1296  */
1297 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1298 {
1299         struct sockaddr *addr = xs_addr(xprt);
1300
1301         dprintk("RPC:       setting port for xprt %p to %u\n", xprt, port);
1302
1303         switch (addr->sa_family) {
1304         case AF_INET:
1305                 ((struct sockaddr_in *)addr)->sin_port = htons(port);
1306                 break;
1307         case AF_INET6:
1308                 ((struct sockaddr_in6 *)addr)->sin6_port = htons(port);
1309                 break;
1310         default:
1311                 BUG();
1312         }
1313 }
1314
1315 static unsigned short xs_get_srcport(struct sock_xprt *transport, struct socket *sock)
1316 {
1317         unsigned short port = transport->port;
1318
1319         if (port == 0 && transport->xprt.resvport)
1320                 port = xs_get_random_port();
1321         return port;
1322 }
1323
1324 static unsigned short xs_next_srcport(struct sock_xprt *transport, struct socket *sock, unsigned short port)
1325 {
1326         if (transport->port != 0)
1327                 transport->port = 0;
1328         if (!transport->xprt.resvport)
1329                 return 0;
1330         if (port <= xprt_min_resvport || port > xprt_max_resvport)
1331                 return xprt_max_resvport;
1332         return --port;
1333 }
1334
1335 static int xs_bind4(struct sock_xprt *transport, struct socket *sock)
1336 {
1337         struct sockaddr_in myaddr = {
1338                 .sin_family = AF_INET,
1339         };
1340         struct sockaddr_in *sa;
1341         int err, nloop = 0;
1342         unsigned short port = xs_get_srcport(transport, sock);
1343         unsigned short last;
1344
1345         sa = (struct sockaddr_in *)&transport->addr;
1346         myaddr.sin_addr = sa->sin_addr;
1347         do {
1348                 myaddr.sin_port = htons(port);
1349                 err = kernel_bind(sock, (struct sockaddr *) &myaddr,
1350                                                 sizeof(myaddr));
1351                 if (port == 0)
1352                         break;
1353                 if (err == 0) {
1354                         transport->port = port;
1355                         break;
1356                 }
1357                 last = port;
1358                 port = xs_next_srcport(transport, sock, port);
1359                 if (port > last)
1360                         nloop++;
1361         } while (err == -EADDRINUSE && nloop != 2);
1362         dprintk("RPC:       %s "NIPQUAD_FMT":%u: %s (%d)\n",
1363                         __FUNCTION__, NIPQUAD(myaddr.sin_addr),
1364                         port, err ? "failed" : "ok", err);
1365         return err;
1366 }
1367
1368 static int xs_bind6(struct sock_xprt *transport, struct socket *sock)
1369 {
1370         struct sockaddr_in6 myaddr = {
1371                 .sin6_family = AF_INET6,
1372         };
1373         struct sockaddr_in6 *sa;
1374         int err, nloop = 0;
1375         unsigned short port = xs_get_srcport(transport, sock);
1376         unsigned short last;
1377
1378         sa = (struct sockaddr_in6 *)&transport->addr;
1379         myaddr.sin6_addr = sa->sin6_addr;
1380         do {
1381                 myaddr.sin6_port = htons(port);
1382                 err = kernel_bind(sock, (struct sockaddr *) &myaddr,
1383                                                 sizeof(myaddr));
1384                 if (port == 0)
1385                         break;
1386                 if (err == 0) {
1387                         transport->port = port;
1388                         break;
1389                 }
1390                 last = port;
1391                 port = xs_next_srcport(transport, sock, port);
1392                 if (port > last)
1393                         nloop++;
1394         } while (err == -EADDRINUSE && nloop != 2);
1395         dprintk("RPC:       xs_bind6 "NIP6_FMT":%u: %s (%d)\n",
1396                 NIP6(myaddr.sin6_addr), port, err ? "failed" : "ok", err);
1397         return err;
1398 }
1399
1400 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1401 static struct lock_class_key xs_key[2];
1402 static struct lock_class_key xs_slock_key[2];
1403
1404 static inline void xs_reclassify_socket4(struct socket *sock)
1405 {
1406         struct sock *sk = sock->sk;
1407
1408         BUG_ON(sock_owned_by_user(sk));
1409         sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1410                 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1411 }
1412
1413 static inline void xs_reclassify_socket6(struct socket *sock)
1414 {
1415         struct sock *sk = sock->sk;
1416
1417         BUG_ON(sock_owned_by_user(sk));
1418         sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1419                 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
1420 }
1421 #else
1422 static inline void xs_reclassify_socket4(struct socket *sock)
1423 {
1424 }
1425
1426 static inline void xs_reclassify_socket6(struct socket *sock)
1427 {
1428 }
1429 #endif
1430
1431 static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
1432 {
1433         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1434
1435         if (!transport->inet) {
1436                 struct sock *sk = sock->sk;
1437
1438                 write_lock_bh(&sk->sk_callback_lock);
1439
1440                 sk->sk_user_data = xprt;
1441                 transport->old_data_ready = sk->sk_data_ready;
1442                 transport->old_state_change = sk->sk_state_change;
1443                 transport->old_write_space = sk->sk_write_space;
1444                 sk->sk_data_ready = xs_udp_data_ready;
1445                 sk->sk_write_space = xs_udp_write_space;
1446                 sk->sk_no_check = UDP_CSUM_NORCV;
1447                 sk->sk_allocation = GFP_ATOMIC;
1448
1449                 xprt_set_connected(xprt);
1450
1451                 /* Reset to new socket */
1452                 transport->sock = sock;
1453                 transport->inet = sk;
1454
1455                 write_unlock_bh(&sk->sk_callback_lock);
1456         }
1457         xs_udp_do_set_buffer_size(xprt);
1458 }
1459
1460 /**
1461  * xs_udp_connect_worker4 - set up a UDP socket
1462  * @work: RPC transport to connect
1463  *
1464  * Invoked by a work queue tasklet.
1465  */
1466 static void xs_udp_connect_worker4(struct work_struct *work)
1467 {
1468         struct sock_xprt *transport =
1469                 container_of(work, struct sock_xprt, connect_worker.work);
1470         struct rpc_xprt *xprt = &transport->xprt;
1471         struct socket *sock = transport->sock;
1472         int err, status = -EIO;
1473
1474         if (xprt->shutdown || !xprt_bound(xprt))
1475                 goto out;
1476
1477         /* Start by resetting any existing state */
1478         xs_close(xprt);
1479
1480         if ((err = sock_create_kern(PF_INET, SOCK_DGRAM, IPPROTO_UDP, &sock)) < 0) {
1481                 dprintk("RPC:       can't create UDP transport socket (%d).\n", -err);
1482                 goto out;
1483         }
1484         xs_reclassify_socket4(sock);
1485
1486         if (xs_bind4(transport, sock)) {
1487                 sock_release(sock);
1488                 goto out;
1489         }
1490
1491         dprintk("RPC:       worker connecting xprt %p to address: %s\n",
1492                         xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1493
1494         xs_udp_finish_connecting(xprt, sock);
1495         status = 0;
1496 out:
1497         xprt_wake_pending_tasks(xprt, status);
1498         xprt_clear_connecting(xprt);
1499 }
1500
1501 /**
1502  * xs_udp_connect_worker6 - set up a UDP socket
1503  * @work: RPC transport to connect
1504  *
1505  * Invoked by a work queue tasklet.
1506  */
1507 static void xs_udp_connect_worker6(struct work_struct *work)
1508 {
1509         struct sock_xprt *transport =
1510                 container_of(work, struct sock_xprt, connect_worker.work);
1511         struct rpc_xprt *xprt = &transport->xprt;
1512         struct socket *sock = transport->sock;
1513         int err, status = -EIO;
1514
1515         if (xprt->shutdown || !xprt_bound(xprt))
1516                 goto out;
1517
1518         /* Start by resetting any existing state */
1519         xs_close(xprt);
1520
1521         if ((err = sock_create_kern(PF_INET6, SOCK_DGRAM, IPPROTO_UDP, &sock)) < 0) {
1522                 dprintk("RPC:       can't create UDP transport socket (%d).\n", -err);
1523                 goto out;
1524         }
1525         xs_reclassify_socket6(sock);
1526
1527         if (xs_bind6(transport, sock) < 0) {
1528                 sock_release(sock);
1529                 goto out;
1530         }
1531
1532         dprintk("RPC:       worker connecting xprt %p to address: %s\n",
1533                         xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1534
1535         xs_udp_finish_connecting(xprt, sock);
1536         status = 0;
1537 out:
1538         xprt_wake_pending_tasks(xprt, status);
1539         xprt_clear_connecting(xprt);
1540 }
1541
1542 /*
1543  * We need to preserve the port number so the reply cache on the server can
1544  * find our cached RPC replies when we get around to reconnecting.
1545  */
1546 static void xs_tcp_reuse_connection(struct rpc_xprt *xprt)
1547 {
1548         int result;
1549         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1550         struct sockaddr any;
1551
1552         dprintk("RPC:       disconnecting xprt %p to reuse port\n", xprt);
1553
1554         /*
1555          * Disconnect the transport socket by doing a connect operation
1556          * with AF_UNSPEC.  This should return immediately...
1557          */
1558         memset(&any, 0, sizeof(any));
1559         any.sa_family = AF_UNSPEC;
1560         result = kernel_connect(transport->sock, &any, sizeof(any), 0);
1561         if (result)
1562                 dprintk("RPC:       AF_UNSPEC connect return code %d\n",
1563                                 result);
1564 }
1565
1566 static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
1567 {
1568         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1569
1570         if (!transport->inet) {
1571                 struct sock *sk = sock->sk;
1572
1573                 write_lock_bh(&sk->sk_callback_lock);
1574
1575                 sk->sk_user_data = xprt;
1576                 transport->old_data_ready = sk->sk_data_ready;
1577                 transport->old_state_change = sk->sk_state_change;
1578                 transport->old_write_space = sk->sk_write_space;
1579                 sk->sk_data_ready = xs_tcp_data_ready;
1580                 sk->sk_state_change = xs_tcp_state_change;
1581                 sk->sk_write_space = xs_tcp_write_space;
1582                 sk->sk_allocation = GFP_ATOMIC;
1583
1584                 /* socket options */
1585                 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
1586                 sock_reset_flag(sk, SOCK_LINGER);
1587                 tcp_sk(sk)->linger2 = 0;
1588                 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
1589
1590                 xprt_clear_connected(xprt);
1591
1592                 /* Reset to new socket */
1593                 transport->sock = sock;
1594                 transport->inet = sk;
1595
1596                 write_unlock_bh(&sk->sk_callback_lock);
1597         }
1598
1599         /* Tell the socket layer to start connecting... */
1600         xprt->stat.connect_count++;
1601         xprt->stat.connect_start = jiffies;
1602         return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
1603 }
1604
1605 /**
1606  * xs_tcp_connect_worker4 - connect a TCP socket to a remote endpoint
1607  * @work: RPC transport to connect
1608  *
1609  * Invoked by a work queue tasklet.
1610  */
1611 static void xs_tcp_connect_worker4(struct work_struct *work)
1612 {
1613         struct sock_xprt *transport =
1614                 container_of(work, struct sock_xprt, connect_worker.work);
1615         struct rpc_xprt *xprt = &transport->xprt;
1616         struct socket *sock = transport->sock;
1617         int err, status = -EIO;
1618
1619         if (xprt->shutdown || !xprt_bound(xprt))
1620                 goto out;
1621
1622         if (!sock) {
1623                 /* start from scratch */
1624                 if ((err = sock_create_kern(PF_INET, SOCK_STREAM, IPPROTO_TCP, &sock)) < 0) {
1625                         dprintk("RPC:       can't create TCP transport socket (%d).\n", -err);
1626                         goto out;
1627                 }
1628                 xs_reclassify_socket4(sock);
1629
1630                 if (xs_bind4(transport, sock) < 0) {
1631                         sock_release(sock);
1632                         goto out;
1633                 }
1634         } else
1635                 /* "close" the socket, preserving the local port */
1636                 xs_tcp_reuse_connection(xprt);
1637
1638         dprintk("RPC:       worker connecting xprt %p to address: %s\n",
1639                         xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1640
1641         status = xs_tcp_finish_connecting(xprt, sock);
1642         dprintk("RPC:       %p connect status %d connected %d sock state %d\n",
1643                         xprt, -status, xprt_connected(xprt),
1644                         sock->sk->sk_state);
1645         if (status < 0) {
1646                 switch (status) {
1647                         case -EINPROGRESS:
1648                         case -EALREADY:
1649                                 goto out_clear;
1650                         case -ECONNREFUSED:
1651                         case -ECONNRESET:
1652                                 /* retry with existing socket, after a delay */
1653                                 break;
1654                         default:
1655                                 /* get rid of existing socket, and retry */
1656                                 xs_tcp_shutdown(xprt);
1657                 }
1658         }
1659 out:
1660         xprt_wake_pending_tasks(xprt, status);
1661 out_clear:
1662         xprt_clear_connecting(xprt);
1663 }
1664
1665 /**
1666  * xs_tcp_connect_worker6 - connect a TCP socket to a remote endpoint
1667  * @work: RPC transport to connect
1668  *
1669  * Invoked by a work queue tasklet.
1670  */
1671 static void xs_tcp_connect_worker6(struct work_struct *work)
1672 {
1673         struct sock_xprt *transport =
1674                 container_of(work, struct sock_xprt, connect_worker.work);
1675         struct rpc_xprt *xprt = &transport->xprt;
1676         struct socket *sock = transport->sock;
1677         int err, status = -EIO;
1678
1679         if (xprt->shutdown || !xprt_bound(xprt))
1680                 goto out;
1681
1682         if (!sock) {
1683                 /* start from scratch */
1684                 if ((err = sock_create_kern(PF_INET6, SOCK_STREAM, IPPROTO_TCP, &sock)) < 0) {
1685                         dprintk("RPC:       can't create TCP transport socket (%d).\n", -err);
1686                         goto out;
1687                 }
1688                 xs_reclassify_socket6(sock);
1689
1690                 if (xs_bind6(transport, sock) < 0) {
1691                         sock_release(sock);
1692                         goto out;
1693                 }
1694         } else
1695                 /* "close" the socket, preserving the local port */
1696                 xs_tcp_reuse_connection(xprt);
1697
1698         dprintk("RPC:       worker connecting xprt %p to address: %s\n",
1699                         xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1700
1701         status = xs_tcp_finish_connecting(xprt, sock);
1702         dprintk("RPC:       %p connect status %d connected %d sock state %d\n",
1703                         xprt, -status, xprt_connected(xprt), sock->sk->sk_state);
1704         if (status < 0) {
1705                 switch (status) {
1706                         case -EINPROGRESS:
1707                         case -EALREADY:
1708                                 goto out_clear;
1709                         case -ECONNREFUSED:
1710                         case -ECONNRESET:
1711                                 /* retry with existing socket, after a delay */
1712                                 break;
1713                         default:
1714                                 /* get rid of existing socket, and retry */
1715                                 xs_tcp_shutdown(xprt);
1716                 }
1717         }
1718 out:
1719         xprt_wake_pending_tasks(xprt, status);
1720 out_clear:
1721         xprt_clear_connecting(xprt);
1722 }
1723
1724 /**
1725  * xs_connect - connect a socket to a remote endpoint
1726  * @task: address of RPC task that manages state of connect request
1727  *
1728  * TCP: If the remote end dropped the connection, delay reconnecting.
1729  *
1730  * UDP socket connects are synchronous, but we use a work queue anyway
1731  * to guarantee that even unprivileged user processes can set up a
1732  * socket on a privileged port.
1733  *
1734  * If a UDP socket connect fails, the delay behavior here prevents
1735  * retry floods (hard mounts).
1736  */
1737 static void xs_connect(struct rpc_task *task)
1738 {
1739         struct rpc_xprt *xprt = task->tk_xprt;
1740         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1741
1742         if (xprt_test_and_set_connecting(xprt))
1743                 return;
1744
1745         if (transport->sock != NULL) {
1746                 dprintk("RPC:       xs_connect delayed xprt %p for %lu "
1747                                 "seconds\n",
1748                                 xprt, xprt->reestablish_timeout / HZ);
1749                 queue_delayed_work(rpciod_workqueue,
1750                                    &transport->connect_worker,
1751                                    xprt->reestablish_timeout);
1752                 xprt->reestablish_timeout <<= 1;
1753                 if (xprt->reestablish_timeout > XS_TCP_MAX_REEST_TO)
1754                         xprt->reestablish_timeout = XS_TCP_MAX_REEST_TO;
1755         } else {
1756                 dprintk("RPC:       xs_connect scheduled xprt %p\n", xprt);
1757                 queue_delayed_work(rpciod_workqueue,
1758                                    &transport->connect_worker, 0);
1759         }
1760 }
1761
1762 static void xs_tcp_connect(struct rpc_task *task)
1763 {
1764         struct rpc_xprt *xprt = task->tk_xprt;
1765
1766         /* Initiate graceful shutdown of the socket if not already done */
1767         if (test_bit(XPRT_CONNECTED, &xprt->state))
1768                 xs_tcp_shutdown(xprt);
1769         /* Exit if we need to wait for socket shutdown to complete */
1770         if (test_bit(XPRT_CLOSING, &xprt->state))
1771                 return;
1772         xs_connect(task);
1773 }
1774
1775 /**
1776  * xs_udp_print_stats - display UDP socket-specifc stats
1777  * @xprt: rpc_xprt struct containing statistics
1778  * @seq: output file
1779  *
1780  */
1781 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
1782 {
1783         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1784
1785         seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %Lu %Lu\n",
1786                         transport->port,
1787                         xprt->stat.bind_count,
1788                         xprt->stat.sends,
1789                         xprt->stat.recvs,
1790                         xprt->stat.bad_xids,
1791                         xprt->stat.req_u,
1792                         xprt->stat.bklog_u);
1793 }
1794
1795 /**
1796  * xs_tcp_print_stats - display TCP socket-specifc stats
1797  * @xprt: rpc_xprt struct containing statistics
1798  * @seq: output file
1799  *
1800  */
1801 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
1802 {
1803         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1804         long idle_time = 0;
1805
1806         if (xprt_connected(xprt))
1807                 idle_time = (long)(jiffies - xprt->last_used) / HZ;
1808
1809         seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu %Lu %Lu\n",
1810                         transport->port,
1811                         xprt->stat.bind_count,
1812                         xprt->stat.connect_count,
1813                         xprt->stat.connect_time,
1814                         idle_time,
1815                         xprt->stat.sends,
1816                         xprt->stat.recvs,
1817                         xprt->stat.bad_xids,
1818                         xprt->stat.req_u,
1819                         xprt->stat.bklog_u);
1820 }
1821
1822 static struct rpc_xprt_ops xs_udp_ops = {
1823         .set_buffer_size        = xs_udp_set_buffer_size,
1824         .reserve_xprt           = xprt_reserve_xprt_cong,
1825         .release_xprt           = xprt_release_xprt_cong,
1826         .rpcbind                = rpcb_getport_async,
1827         .set_port               = xs_set_port,
1828         .connect                = xs_connect,
1829         .buf_alloc              = rpc_malloc,
1830         .buf_free               = rpc_free,
1831         .send_request           = xs_udp_send_request,
1832         .set_retrans_timeout    = xprt_set_retrans_timeout_rtt,
1833         .timer                  = xs_udp_timer,
1834         .release_request        = xprt_release_rqst_cong,
1835         .close                  = xs_close,
1836         .destroy                = xs_destroy,
1837         .print_stats            = xs_udp_print_stats,
1838 };
1839
1840 static struct rpc_xprt_ops xs_tcp_ops = {
1841         .reserve_xprt           = xprt_reserve_xprt,
1842         .release_xprt           = xs_tcp_release_xprt,
1843         .rpcbind                = rpcb_getport_async,
1844         .set_port               = xs_set_port,
1845         .connect                = xs_tcp_connect,
1846         .buf_alloc              = rpc_malloc,
1847         .buf_free               = rpc_free,
1848         .send_request           = xs_tcp_send_request,
1849         .set_retrans_timeout    = xprt_set_retrans_timeout_def,
1850         .close                  = xs_tcp_shutdown,
1851         .destroy                = xs_destroy,
1852         .print_stats            = xs_tcp_print_stats,
1853 };
1854
1855 static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
1856                                       unsigned int slot_table_size)
1857 {
1858         struct rpc_xprt *xprt;
1859         struct sock_xprt *new;
1860
1861         if (args->addrlen > sizeof(xprt->addr)) {
1862                 dprintk("RPC:       xs_setup_xprt: address too large\n");
1863                 return ERR_PTR(-EBADF);
1864         }
1865
1866         new = kzalloc(sizeof(*new), GFP_KERNEL);
1867         if (new == NULL) {
1868                 dprintk("RPC:       xs_setup_xprt: couldn't allocate "
1869                                 "rpc_xprt\n");
1870                 return ERR_PTR(-ENOMEM);
1871         }
1872         xprt = &new->xprt;
1873
1874         xprt->max_reqs = slot_table_size;
1875         xprt->slot = kcalloc(xprt->max_reqs, sizeof(struct rpc_rqst), GFP_KERNEL);
1876         if (xprt->slot == NULL) {
1877                 kfree(xprt);
1878                 dprintk("RPC:       xs_setup_xprt: couldn't allocate slot "
1879                                 "table\n");
1880                 return ERR_PTR(-ENOMEM);
1881         }
1882
1883         memcpy(&xprt->addr, args->dstaddr, args->addrlen);
1884         xprt->addrlen = args->addrlen;
1885         if (args->srcaddr)
1886                 memcpy(&new->addr, args->srcaddr, args->addrlen);
1887
1888         return xprt;
1889 }
1890
1891 /**
1892  * xs_setup_udp - Set up transport to use a UDP socket
1893  * @args: rpc transport creation arguments
1894  *
1895  */
1896 static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
1897 {
1898         struct sockaddr *addr = args->dstaddr;
1899         struct rpc_xprt *xprt;
1900         struct sock_xprt *transport;
1901
1902         xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries);
1903         if (IS_ERR(xprt))
1904                 return xprt;
1905         transport = container_of(xprt, struct sock_xprt, xprt);
1906
1907         xprt->prot = IPPROTO_UDP;
1908         xprt->tsh_size = 0;
1909         /* XXX: header size can vary due to auth type, IPv6, etc. */
1910         xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
1911
1912         xprt->bind_timeout = XS_BIND_TO;
1913         xprt->connect_timeout = XS_UDP_CONN_TO;
1914         xprt->reestablish_timeout = XS_UDP_REEST_TO;
1915         xprt->idle_timeout = XS_IDLE_DISC_TO;
1916
1917         xprt->ops = &xs_udp_ops;
1918
1919         if (args->timeout)
1920                 xprt->timeout = *args->timeout;
1921         else
1922                 xprt_set_timeout(&xprt->timeout, 5, 5 * HZ);
1923
1924         switch (addr->sa_family) {
1925         case AF_INET:
1926                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
1927                         xprt_set_bound(xprt);
1928
1929                 INIT_DELAYED_WORK(&transport->connect_worker,
1930                                         xs_udp_connect_worker4);
1931                 xs_format_ipv4_peer_addresses(xprt);
1932                 break;
1933         case AF_INET6:
1934                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
1935                         xprt_set_bound(xprt);
1936
1937                 INIT_DELAYED_WORK(&transport->connect_worker,
1938                                         xs_udp_connect_worker6);
1939                 xs_format_ipv6_peer_addresses(xprt);
1940                 break;
1941         default:
1942                 kfree(xprt);
1943                 return ERR_PTR(-EAFNOSUPPORT);
1944         }
1945
1946         dprintk("RPC:       set up transport to address %s\n",
1947                         xprt->address_strings[RPC_DISPLAY_ALL]);
1948
1949         if (try_module_get(THIS_MODULE))
1950                 return xprt;
1951
1952         kfree(xprt->slot);
1953         kfree(xprt);
1954         return ERR_PTR(-EINVAL);
1955 }
1956
1957 /**
1958  * xs_setup_tcp - Set up transport to use a TCP socket
1959  * @args: rpc transport creation arguments
1960  *
1961  */
1962 static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
1963 {
1964         struct sockaddr *addr = args->dstaddr;
1965         struct rpc_xprt *xprt;
1966         struct sock_xprt *transport;
1967
1968         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries);
1969         if (IS_ERR(xprt))
1970                 return xprt;
1971         transport = container_of(xprt, struct sock_xprt, xprt);
1972
1973         xprt->prot = IPPROTO_TCP;
1974         xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
1975         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
1976
1977         xprt->bind_timeout = XS_BIND_TO;
1978         xprt->connect_timeout = XS_TCP_CONN_TO;
1979         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1980         xprt->idle_timeout = XS_IDLE_DISC_TO;
1981
1982         xprt->ops = &xs_tcp_ops;
1983
1984         if (args->timeout)
1985                 xprt->timeout = *args->timeout;
1986         else
1987                 xprt_set_timeout(&xprt->timeout, 2, 60 * HZ);
1988
1989         switch (addr->sa_family) {
1990         case AF_INET:
1991                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
1992                         xprt_set_bound(xprt);
1993
1994                 INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_connect_worker4);
1995                 xs_format_ipv4_peer_addresses(xprt);
1996                 break;
1997         case AF_INET6:
1998                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
1999                         xprt_set_bound(xprt);
2000
2001                 INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_connect_worker6);
2002                 xs_format_ipv6_peer_addresses(xprt);
2003                 break;
2004         default:
2005                 kfree(xprt);
2006                 return ERR_PTR(-EAFNOSUPPORT);
2007         }
2008
2009         dprintk("RPC:       set up transport to address %s\n",
2010                         xprt->address_strings[RPC_DISPLAY_ALL]);
2011
2012         if (try_module_get(THIS_MODULE))
2013                 return xprt;
2014
2015         kfree(xprt->slot);
2016         kfree(xprt);
2017         return ERR_PTR(-EINVAL);
2018 }
2019
2020 static struct xprt_class        xs_udp_transport = {
2021         .list           = LIST_HEAD_INIT(xs_udp_transport.list),
2022         .name           = "udp",
2023         .owner          = THIS_MODULE,
2024         .ident          = IPPROTO_UDP,
2025         .setup          = xs_setup_udp,
2026 };
2027
2028 static struct xprt_class        xs_tcp_transport = {
2029         .list           = LIST_HEAD_INIT(xs_tcp_transport.list),
2030         .name           = "tcp",
2031         .owner          = THIS_MODULE,
2032         .ident          = IPPROTO_TCP,
2033         .setup          = xs_setup_tcp,
2034 };
2035
2036 /**
2037  * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
2038  *
2039  */
2040 int init_socket_xprt(void)
2041 {
2042 #ifdef RPC_DEBUG
2043         if (!sunrpc_table_header)
2044                 sunrpc_table_header = register_sysctl_table(sunrpc_table);
2045 #endif
2046
2047         xprt_register_transport(&xs_udp_transport);
2048         xprt_register_transport(&xs_tcp_transport);
2049
2050         return 0;
2051 }
2052
2053 /**
2054  * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
2055  *
2056  */
2057 void cleanup_socket_xprt(void)
2058 {
2059 #ifdef RPC_DEBUG
2060         if (sunrpc_table_header) {
2061                 unregister_sysctl_table(sunrpc_table_header);
2062                 sunrpc_table_header = NULL;
2063         }
2064 #endif
2065
2066         xprt_unregister_transport(&xs_udp_transport);
2067         xprt_unregister_transport(&xs_tcp_transport);
2068 }