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_CLOSE_WAIT, &xprt->state);
 778         clear_bit(XPRT_CLOSING, &xprt->state);
 779         smp_mb__after_clear_bit();
 780         xprt_disconnect_done(xprt);
 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         xs_close(xprt);
 797         xs_free_peer_addresses(xprt);
 798         kfree(xprt->slot);
 799         kfree(xprt);
 800         module_put(THIS_MODULE);
 801 }
 802
 803 static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
 804 {
 805         return (struct rpc_xprt *) sk->sk_user_data;
 806 }
 807
 808 /**
 809  * xs_udp_data_ready - "data ready" callback for UDP sockets
 810  * @sk: socket with data to read
 811  * @len: how much data to read
 812  *
 813  */
 814 static void xs_udp_data_ready(struct sock *sk, int len)
 815 {
 816         struct rpc_task *task;
 817         struct rpc_xprt *xprt;
 818         struct rpc_rqst *rovr;
 819         struct sk_buff *skb;
 820         int err, repsize, copied;
 821         u32 _xid;
 822         __be32 *xp;
 823
 824         read_lock(&sk->sk_callback_lock);
 825         dprintk("RPC:       xs_udp_data_ready...\n");
 826         if (!(xprt = xprt_from_sock(sk)))
 827                 goto out;
 828
 829         if ((skb = skb_recv_datagram(sk, 0, 1, &err)) == NULL)
 830                 goto out;
 831
 832         if (xprt->shutdown)
 833                 goto dropit;
 834
 835         repsize = skb->len - sizeof(struct udphdr);
 836         if (repsize < 4) {
 837                 dprintk("RPC:       impossible RPC reply size %d!\n", repsize);
 838                 goto dropit;
 839         }
 840
 841         /* Copy the XID from the skb... */
 842         xp = skb_header_pointer(skb, sizeof(struct udphdr),
 843                                 sizeof(_xid), &_xid);
 844         if (xp == NULL)
 845                 goto dropit;
 846
 847         /* Look up and lock the request corresponding to the given XID */
 848         spin_lock(&xprt->transport_lock);
 849         rovr = xprt_lookup_rqst(xprt, *xp);
 850         if (!rovr)
 851                 goto out_unlock;
 852         task = rovr->rq_task;
 853
 854         if ((copied = rovr->rq_private_buf.buflen) > repsize)
 855                 copied = repsize;
 856
 857         /* Suck it into the iovec, verify checksum if not done by hw. */
 858         if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
 859                 UDPX_INC_STATS_BH(sk, UDP_MIB_INERRORS);
 860                 goto out_unlock;
 861         }
 862
 863         UDPX_INC_STATS_BH(sk, UDP_MIB_INDATAGRAMS);
 864
 865         /* Something worked... */
 866         dst_confirm(skb->dst);
 867
 868         xprt_adjust_cwnd(task, copied);
 869         xprt_update_rtt(task);
 870         xprt_complete_rqst(task, copied);
 871
 872  out_unlock:
 873         spin_unlock(&xprt->transport_lock);
 874  dropit:
 875         skb_free_datagram(sk, skb);
 876  out:
 877         read_unlock(&sk->sk_callback_lock);
 878 }
 879
 880 static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
 881 {
 882         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 883         size_t len, used;
 884         char *p;
 885
 886         p = ((char *) &transport->tcp_fraghdr) + transport->tcp_offset;
 887         len = sizeof(transport->tcp_fraghdr) - transport->tcp_offset;
 888         used = xdr_skb_read_bits(desc, p, len);
 889         transport->tcp_offset += used;
 890         if (used != len)
 891                 return;
 892
 893         transport->tcp_reclen = ntohl(transport->tcp_fraghdr);
 894         if (transport->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
 895                 transport->tcp_flags |= TCP_RCV_LAST_FRAG;
 896         else
 897                 transport->tcp_flags &= ~TCP_RCV_LAST_FRAG;
 898         transport->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
 899
 900         transport->tcp_flags &= ~TCP_RCV_COPY_FRAGHDR;
 901         transport->tcp_offset = 0;
 902
 903         /* Sanity check of the record length */
 904         if (unlikely(transport->tcp_reclen < 4)) {
 905                 dprintk("RPC:       invalid TCP record fragment length\n");
 906                 xprt_force_disconnect(xprt);
 907                 return;
 908         }
 909         dprintk("RPC:       reading TCP record fragment of length %d\n",
 910                         transport->tcp_reclen);
 911 }
 912
 913 static void xs_tcp_check_fraghdr(struct sock_xprt *transport)
 914 {
 915         if (transport->tcp_offset == transport->tcp_reclen) {
 916                 transport->tcp_flags |= TCP_RCV_COPY_FRAGHDR;
 917                 transport->tcp_offset = 0;
 918                 if (transport->tcp_flags & TCP_RCV_LAST_FRAG) {
 919                         transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
 920                         transport->tcp_flags |= TCP_RCV_COPY_XID;
 921                         transport->tcp_copied = 0;
 922                 }
 923         }
 924 }
 925
 926 static inline void xs_tcp_read_xid(struct sock_xprt *transport, struct xdr_skb_reader *desc)
 927 {
 928         size_t len, used;
 929         char *p;
 930
 931         len = sizeof(transport->tcp_xid) - transport->tcp_offset;
 932         dprintk("RPC:       reading XID (%Zu bytes)\n", len);
 933         p = ((char *) &transport->tcp_xid) + transport->tcp_offset;
 934         used = xdr_skb_read_bits(desc, p, len);
 935         transport->tcp_offset += used;
 936         if (used != len)
 937                 return;
 938         transport->tcp_flags &= ~TCP_RCV_COPY_XID;
 939         transport->tcp_flags |= TCP_RCV_COPY_DATA;
 940         transport->tcp_copied = 4;
 941         dprintk("RPC:       reading reply for XID %08x\n",
 942                         ntohl(transport->tcp_xid));
 943         xs_tcp_check_fraghdr(transport);
 944 }
 945
 946 static inline void xs_tcp_read_request(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
 947 {
 948         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 949         struct rpc_rqst *req;
 950         struct xdr_buf *rcvbuf;
 951         size_t len;
 952         ssize_t r;
 953
 954         /* Find and lock the request corresponding to this xid */
 955         spin_lock(&xprt->transport_lock);
 956         req = xprt_lookup_rqst(xprt, transport->tcp_xid);
 957         if (!req) {
 958                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
 959                 dprintk("RPC:       XID %08x request not found!\n",
 960                                 ntohl(transport->tcp_xid));
 961                 spin_unlock(&xprt->transport_lock);
 962                 return;
 963         }
 964
 965         rcvbuf = &req->rq_private_buf;
 966         len = desc->count;
 967         if (len > transport->tcp_reclen - transport->tcp_offset) {
 968                 struct xdr_skb_reader my_desc;
 969
 970                 len = transport->tcp_reclen - transport->tcp_offset;
 971                 memcpy(&my_desc, desc, sizeof(my_desc));
 972                 my_desc.count = len;
 973                 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
 974                                           &my_desc, xdr_skb_read_bits);
 975                 desc->count -= r;
 976                 desc->offset += r;
 977         } else
 978                 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
 979                                           desc, xdr_skb_read_bits);
 980
 981         if (r > 0) {
 982                 transport->tcp_copied += r;
 983                 transport->tcp_offset += r;
 984         }
 985         if (r != len) {
 986                 /* Error when copying to the receive buffer,
 987                  * usually because we weren't able to allocate
 988                  * additional buffer pages. All we can do now
 989                  * is turn off TCP_RCV_COPY_DATA, so the request
 990                  * will not receive any additional updates,
 991                  * and time out.
 992                  * Any remaining data from this record will
 993                  * be discarded.
 994                  */
 995                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
 996                 dprintk("RPC:       XID %08x truncated request\n",
 997                                 ntohl(transport->tcp_xid));
 998                 dprintk("RPC:       xprt = %p, tcp_copied = %lu, "
 999                                 "tcp_offset = %u, tcp_reclen = %u\n",
1000                                 xprt, transport->tcp_copied,
1001                                 transport->tcp_offset, transport->tcp_reclen);
1002                 goto out;
1003         }
1004
1005         dprintk("RPC:       XID %08x read %Zd bytes\n",
1006                         ntohl(transport->tcp_xid), r);
1007         dprintk("RPC:       xprt = %p, tcp_copied = %lu, tcp_offset = %u, "
1008                         "tcp_reclen = %u\n", xprt, transport->tcp_copied,
1009                         transport->tcp_offset, transport->tcp_reclen);
1010
1011         if (transport->tcp_copied == req->rq_private_buf.buflen)
1012                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1013         else if (transport->tcp_offset == transport->tcp_reclen) {
1014                 if (transport->tcp_flags & TCP_RCV_LAST_FRAG)
1015                         transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1016         }
1017
1018 out:
1019         if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1020                 xprt_complete_rqst(req->rq_task, transport->tcp_copied);
1021         spin_unlock(&xprt->transport_lock);
1022         xs_tcp_check_fraghdr(transport);
1023 }
1024
1025 static inline void xs_tcp_read_discard(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1026 {
1027         size_t len;
1028
1029         len = transport->tcp_reclen - transport->tcp_offset;
1030         if (len > desc->count)
1031                 len = desc->count;
1032         desc->count -= len;
1033         desc->offset += len;
1034         transport->tcp_offset += len;
1035         dprintk("RPC:       discarded %Zu bytes\n", len);
1036         xs_tcp_check_fraghdr(transport);
1037 }
1038
1039 static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
1040 {
1041         struct rpc_xprt *xprt = rd_desc->arg.data;
1042         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1043         struct xdr_skb_reader desc = {
1044                 .skb    = skb,
1045                 .offset = offset,
1046                 .count  = len,
1047         };
1048
1049         dprintk("RPC:       xs_tcp_data_recv started\n");
1050         do {
1051                 /* Read in a new fragment marker if necessary */
1052                 /* Can we ever really expect to get completely empty fragments? */
1053                 if (transport->tcp_flags & TCP_RCV_COPY_FRAGHDR) {
1054                         xs_tcp_read_fraghdr(xprt, &desc);
1055                         continue;
1056                 }
1057                 /* Read in the xid if necessary */
1058                 if (transport->tcp_flags & TCP_RCV_COPY_XID) {
1059                         xs_tcp_read_xid(transport, &desc);
1060                         continue;
1061                 }
1062                 /* Read in the request data */
1063                 if (transport->tcp_flags & TCP_RCV_COPY_DATA) {
1064                         xs_tcp_read_request(xprt, &desc);
1065                         continue;
1066                 }
1067                 /* Skip over any trailing bytes on short reads */
1068                 xs_tcp_read_discard(transport, &desc);
1069         } while (desc.count);
1070         dprintk("RPC:       xs_tcp_data_recv done\n");
1071         return len - desc.count;
1072 }
1073
1074 /**
1075  * xs_tcp_data_ready - "data ready" callback for TCP sockets
1076  * @sk: socket with data to read
1077  * @bytes: how much data to read
1078  *
1079  */
1080 static void xs_tcp_data_ready(struct sock *sk, int bytes)
1081 {
1082         struct rpc_xprt *xprt;
1083         read_descriptor_t rd_desc;
1084
1085         dprintk("RPC:       xs_tcp_data_ready...\n");
1086
1087         read_lock(&sk->sk_callback_lock);
1088         if (!(xprt = xprt_from_sock(sk)))
1089                 goto out;
1090         if (xprt->shutdown)
1091                 goto out;
1092
1093         /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
1094         rd_desc.arg.data = xprt;
1095         rd_desc.count = 65536;
1096         tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
1097 out:
1098         read_unlock(&sk->sk_callback_lock);
1099 }
1100
1101 /**
1102  * xs_tcp_state_change - callback to handle TCP socket state changes
1103  * @sk: socket whose state has changed
1104  *
1105  */
1106 static void xs_tcp_state_change(struct sock *sk)
1107 {
1108         struct rpc_xprt *xprt;
1109
1110         read_lock(&sk->sk_callback_lock);
1111         if (!(xprt = xprt_from_sock(sk)))
1112                 goto out;
1113         dprintk("RPC:       xs_tcp_state_change client %p...\n", xprt);
1114         dprintk("RPC:       state %x conn %d dead %d zapped %d\n",
1115                         sk->sk_state, xprt_connected(xprt),
1116                         sock_flag(sk, SOCK_DEAD),
1117                         sock_flag(sk, SOCK_ZAPPED));
1118
1119         switch (sk->sk_state) {
1120         case TCP_ESTABLISHED:
1121                 spin_lock_bh(&xprt->transport_lock);
1122                 if (!xprt_test_and_set_connected(xprt)) {
1123                         struct sock_xprt *transport = container_of(xprt,
1124                                         struct sock_xprt, xprt);
1125
1126                         /* Reset TCP record info */
1127                         transport->tcp_offset = 0;
1128                         transport->tcp_reclen = 0;
1129                         transport->tcp_copied = 0;
1130                         transport->tcp_flags =
1131                                 TCP_RCV_COPY_FRAGHDR | TCP_RCV_COPY_XID;
1132
1133                         xprt_wake_pending_tasks(xprt, 0);
1134                 }
1135                 spin_unlock_bh(&xprt->transport_lock);
1136                 break;
1137         case TCP_FIN_WAIT1:
1138                 /* The client initiated a shutdown of the socket */
1139                 xprt->reestablish_timeout = 0;
1140                 set_bit(XPRT_CLOSING, &xprt->state);
1141                 smp_mb__before_clear_bit();
1142                 clear_bit(XPRT_CONNECTED, &xprt->state);
1143                 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1144                 smp_mb__after_clear_bit();
1145                 break;
1146         case TCP_CLOSE_WAIT:
1147                 /* The server initiated a shutdown of the socket */
1148                 set_bit(XPRT_CLOSING, &xprt->state);
1149                 xprt_force_disconnect(xprt);
1150         case TCP_SYN_SENT:
1151         case TCP_CLOSING:
1152                 /*
1153                  * If the server closed down the connection, make sure that
1154                  * we back off before reconnecting
1155                  */
1156                 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1157                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1158                 break;
1159         case TCP_LAST_ACK:
1160                 smp_mb__before_clear_bit();
1161                 clear_bit(XPRT_CONNECTED, &xprt->state);
1162                 smp_mb__after_clear_bit();
1163                 break;
1164         case TCP_CLOSE:
1165                 smp_mb__before_clear_bit();
1166                 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1167                 clear_bit(XPRT_CLOSING, &xprt->state);
1168                 smp_mb__after_clear_bit();
1169                 /* Mark transport as closed and wake up all pending tasks */
1170                 xprt_disconnect_done(xprt);
1171         }
1172  out:
1173         read_unlock(&sk->sk_callback_lock);
1174 }
1175
1176 /**
1177  * xs_udp_write_space - callback invoked when socket buffer space
1178  *                             becomes available
1179  * @sk: socket whose state has changed
1180  *
1181  * Called when more output buffer space is available for this socket.
1182  * We try not to wake our writers until they can make "significant"
1183  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1184  * with a bunch of small requests.
1185  */
1186 static void xs_udp_write_space(struct sock *sk)
1187 {
1188         read_lock(&sk->sk_callback_lock);
1189
1190         /* from net/core/sock.c:sock_def_write_space */
1191         if (sock_writeable(sk)) {
1192                 struct socket *sock;
1193                 struct rpc_xprt *xprt;
1194
1195                 if (unlikely(!(sock = sk->sk_socket)))
1196                         goto out;
1197                 if (unlikely(!(xprt = xprt_from_sock(sk))))
1198                         goto out;
1199                 if (unlikely(!test_and_clear_bit(SOCK_NOSPACE, &sock->flags)))
1200                         goto out;
1201
1202                 xprt_write_space(xprt);
1203         }
1204
1205  out:
1206         read_unlock(&sk->sk_callback_lock);
1207 }
1208
1209 /**
1210  * xs_tcp_write_space - callback invoked when socket buffer space
1211  *                             becomes available
1212  * @sk: socket whose state has changed
1213  *
1214  * Called when more output buffer space is available for this socket.
1215  * We try not to wake our writers until they can make "significant"
1216  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1217  * with a bunch of small requests.
1218  */
1219 static void xs_tcp_write_space(struct sock *sk)
1220 {
1221         read_lock(&sk->sk_callback_lock);
1222
1223         /* from net/core/stream.c:sk_stream_write_space */
1224         if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk)) {
1225                 struct socket *sock;
1226                 struct rpc_xprt *xprt;
1227
1228                 if (unlikely(!(sock = sk->sk_socket)))
1229                         goto out;
1230                 if (unlikely(!(xprt = xprt_from_sock(sk))))
1231                         goto out;
1232                 if (unlikely(!test_and_clear_bit(SOCK_NOSPACE, &sock->flags)))
1233                         goto out;
1234
1235                 xprt_write_space(xprt);
1236         }
1237
1238  out:
1239         read_unlock(&sk->sk_callback_lock);
1240 }
1241
1242 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
1243 {
1244         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1245         struct sock *sk = transport->inet;
1246
1247         if (transport->rcvsize) {
1248                 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1249                 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
1250         }
1251         if (transport->sndsize) {
1252                 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1253                 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1254                 sk->sk_write_space(sk);
1255         }
1256 }
1257
1258 /**
1259  * xs_udp_set_buffer_size - set send and receive limits
1260  * @xprt: generic transport
1261  * @sndsize: requested size of send buffer, in bytes
1262  * @rcvsize: requested size of receive buffer, in bytes
1263  *
1264  * Set socket send and receive buffer size limits.
1265  */
1266 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1267 {
1268         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1269
1270         transport->sndsize = 0;
1271         if (sndsize)
1272                 transport->sndsize = sndsize + 1024;
1273         transport->rcvsize = 0;
1274         if (rcvsize)
1275                 transport->rcvsize = rcvsize + 1024;
1276
1277         xs_udp_do_set_buffer_size(xprt);
1278 }
1279
1280 /**
1281  * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1282  * @task: task that timed out
1283  *
1284  * Adjust the congestion window after a retransmit timeout has occurred.
1285  */
1286 static void xs_udp_timer(struct rpc_task *task)
1287 {
1288         xprt_adjust_cwnd(task, -ETIMEDOUT);
1289 }
1290
1291 static unsigned short xs_get_random_port(void)
1292 {
1293         unsigned short range = xprt_max_resvport - xprt_min_resvport;
1294         unsigned short rand = (unsigned short) net_random() % range;
1295         return rand + xprt_min_resvport;
1296 }
1297
1298 /**
1299  * xs_set_port - reset the port number in the remote endpoint address
1300  * @xprt: generic transport
1301  * @port: new port number
1302  *
1303  */
1304 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1305 {
1306         struct sockaddr *addr = xs_addr(xprt);
1307
1308         dprintk("RPC:       setting port for xprt %p to %u\n", xprt, port);
1309
1310         switch (addr->sa_family) {
1311         case AF_INET:
1312                 ((struct sockaddr_in *)addr)->sin_port = htons(port);
1313                 break;
1314         case AF_INET6:
1315                 ((struct sockaddr_in6 *)addr)->sin6_port = htons(port);
1316                 break;
1317         default:
1318                 BUG();
1319         }
1320 }
1321
1322 static unsigned short xs_get_srcport(struct sock_xprt *transport, struct socket *sock)
1323 {
1324         unsigned short port = transport->port;
1325
1326         if (port == 0 && transport->xprt.resvport)
1327                 port = xs_get_random_port();
1328         return port;
1329 }
1330
1331 static unsigned short xs_next_srcport(struct sock_xprt *transport, struct socket *sock, unsigned short port)
1332 {
1333         if (transport->port != 0)
1334                 transport->port = 0;
1335         if (!transport->xprt.resvport)
1336                 return 0;
1337         if (port <= xprt_min_resvport || port > xprt_max_resvport)
1338                 return xprt_max_resvport;
1339         return --port;
1340 }
1341
1342 static int xs_bind4(struct sock_xprt *transport, struct socket *sock)
1343 {
1344         struct sockaddr_in myaddr = {
1345                 .sin_family = AF_INET,
1346         };
1347         struct sockaddr_in *sa;
1348         int err, nloop = 0;
1349         unsigned short port = xs_get_srcport(transport, sock);
1350         unsigned short last;
1351
1352         sa = (struct sockaddr_in *)&transport->addr;
1353         myaddr.sin_addr = sa->sin_addr;
1354         do {
1355                 myaddr.sin_port = htons(port);
1356                 err = kernel_bind(sock, (struct sockaddr *) &myaddr,
1357                                                 sizeof(myaddr));
1358                 if (port == 0)
1359                         break;
1360                 if (err == 0) {
1361                         transport->port = port;
1362                         break;
1363                 }
1364                 last = port;
1365                 port = xs_next_srcport(transport, sock, port);
1366                 if (port > last)
1367                         nloop++;
1368         } while (err == -EADDRINUSE && nloop != 2);
1369         dprintk("RPC:       %s "NIPQUAD_FMT":%u: %s (%d)\n",
1370                         __FUNCTION__, NIPQUAD(myaddr.sin_addr),
1371                         port, err ? "failed" : "ok", err);
1372         return err;
1373 }
1374
1375 static int xs_bind6(struct sock_xprt *transport, struct socket *sock)
1376 {
1377         struct sockaddr_in6 myaddr = {
1378                 .sin6_family = AF_INET6,
1379         };
1380         struct sockaddr_in6 *sa;
1381         int err, nloop = 0;
1382         unsigned short port = xs_get_srcport(transport, sock);
1383         unsigned short last;
1384
1385         sa = (struct sockaddr_in6 *)&transport->addr;
1386         myaddr.sin6_addr = sa->sin6_addr;
1387         do {
1388                 myaddr.sin6_port = htons(port);
1389                 err = kernel_bind(sock, (struct sockaddr *) &myaddr,
1390                                                 sizeof(myaddr));
1391                 if (port == 0)
1392                         break;
1393                 if (err == 0) {
1394                         transport->port = port;
1395                         break;
1396                 }
1397                 last = port;
1398                 port = xs_next_srcport(transport, sock, port);
1399                 if (port > last)
1400                         nloop++;
1401         } while (err == -EADDRINUSE && nloop != 2);
1402         dprintk("RPC:       xs_bind6 "NIP6_FMT":%u: %s (%d)\n",
1403                 NIP6(myaddr.sin6_addr), port, err ? "failed" : "ok", err);
1404         return err;
1405 }
1406
1407 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1408 static struct lock_class_key xs_key[2];
1409 static struct lock_class_key xs_slock_key[2];
1410
1411 static inline void xs_reclassify_socket4(struct socket *sock)
1412 {
1413         struct sock *sk = sock->sk;
1414
1415         BUG_ON(sock_owned_by_user(sk));
1416         sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1417                 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1418 }
1419
1420 static inline void xs_reclassify_socket6(struct socket *sock)
1421 {
1422         struct sock *sk = sock->sk;
1423
1424         BUG_ON(sock_owned_by_user(sk));
1425         sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1426                 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
1427 }
1428 #else
1429 static inline void xs_reclassify_socket4(struct socket *sock)
1430 {
1431 }
1432
1433 static inline void xs_reclassify_socket6(struct socket *sock)
1434 {
1435 }
1436 #endif
1437
1438 static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
1439 {
1440         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1441
1442         if (!transport->inet) {
1443                 struct sock *sk = sock->sk;
1444
1445                 write_lock_bh(&sk->sk_callback_lock);
1446
1447                 sk->sk_user_data = xprt;
1448                 transport->old_data_ready = sk->sk_data_ready;
1449                 transport->old_state_change = sk->sk_state_change;
1450                 transport->old_write_space = sk->sk_write_space;
1451                 sk->sk_data_ready = xs_udp_data_ready;
1452                 sk->sk_write_space = xs_udp_write_space;
1453                 sk->sk_no_check = UDP_CSUM_NORCV;
1454                 sk->sk_allocation = GFP_ATOMIC;
1455
1456                 xprt_set_connected(xprt);
1457
1458                 /* Reset to new socket */
1459                 transport->sock = sock;
1460                 transport->inet = sk;
1461
1462                 write_unlock_bh(&sk->sk_callback_lock);
1463         }
1464         xs_udp_do_set_buffer_size(xprt);
1465 }
1466
1467 /**
1468  * xs_udp_connect_worker4 - set up a UDP socket
1469  * @work: RPC transport to connect
1470  *
1471  * Invoked by a work queue tasklet.
1472  */
1473 static void xs_udp_connect_worker4(struct work_struct *work)
1474 {
1475         struct sock_xprt *transport =
1476                 container_of(work, struct sock_xprt, connect_worker.work);
1477         struct rpc_xprt *xprt = &transport->xprt;
1478         struct socket *sock = transport->sock;
1479         int err, status = -EIO;
1480
1481         if (xprt->shutdown || !xprt_bound(xprt))
1482                 goto out;
1483
1484         /* Start by resetting any existing state */
1485         xs_close(xprt);
1486
1487         if ((err = sock_create_kern(PF_INET, SOCK_DGRAM, IPPROTO_UDP, &sock)) < 0) {
1488                 dprintk("RPC:       can't create UDP transport socket (%d).\n", -err);
1489                 goto out;
1490         }
1491         xs_reclassify_socket4(sock);
1492
1493         if (xs_bind4(transport, sock)) {
1494                 sock_release(sock);
1495                 goto out;
1496         }
1497
1498         dprintk("RPC:       worker connecting xprt %p to address: %s\n",
1499                         xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1500
1501         xs_udp_finish_connecting(xprt, sock);
1502         status = 0;
1503 out:
1504         xprt_wake_pending_tasks(xprt, status);
1505         xprt_clear_connecting(xprt);
1506 }
1507
1508 /**
1509  * xs_udp_connect_worker6 - set up a UDP socket
1510  * @work: RPC transport to connect
1511  *
1512  * Invoked by a work queue tasklet.
1513  */
1514 static void xs_udp_connect_worker6(struct work_struct *work)
1515 {
1516         struct sock_xprt *transport =
1517                 container_of(work, struct sock_xprt, connect_worker.work);
1518         struct rpc_xprt *xprt = &transport->xprt;
1519         struct socket *sock = transport->sock;
1520         int err, status = -EIO;
1521
1522         if (xprt->shutdown || !xprt_bound(xprt))
1523                 goto out;
1524
1525         /* Start by resetting any existing state */
1526         xs_close(xprt);
1527
1528         if ((err = sock_create_kern(PF_INET6, SOCK_DGRAM, IPPROTO_UDP, &sock)) < 0) {
1529                 dprintk("RPC:       can't create UDP transport socket (%d).\n", -err);
1530                 goto out;
1531         }
1532         xs_reclassify_socket6(sock);
1533
1534         if (xs_bind6(transport, sock) < 0) {
1535                 sock_release(sock);
1536                 goto out;
1537         }
1538
1539         dprintk("RPC:       worker connecting xprt %p to address: %s\n",
1540                         xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1541
1542         xs_udp_finish_connecting(xprt, sock);
1543         status = 0;
1544 out:
1545         xprt_wake_pending_tasks(xprt, status);
1546         xprt_clear_connecting(xprt);
1547 }
1548
1549 /*
1550  * We need to preserve the port number so the reply cache on the server can
1551  * find our cached RPC replies when we get around to reconnecting.
1552  */
1553 static void xs_tcp_reuse_connection(struct rpc_xprt *xprt)
1554 {
1555         int result;
1556         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1557         struct sockaddr any;
1558
1559         dprintk("RPC:       disconnecting xprt %p to reuse port\n", xprt);
1560
1561         /*
1562          * Disconnect the transport socket by doing a connect operation
1563          * with AF_UNSPEC.  This should return immediately...
1564          */
1565         memset(&any, 0, sizeof(any));
1566         any.sa_family = AF_UNSPEC;
1567         result = kernel_connect(transport->sock, &any, sizeof(any), 0);
1568         if (result)
1569                 dprintk("RPC:       AF_UNSPEC connect return code %d\n",
1570                                 result);
1571 }
1572
1573 static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
1574 {
1575         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1576
1577         if (!transport->inet) {
1578                 struct sock *sk = sock->sk;
1579
1580                 write_lock_bh(&sk->sk_callback_lock);
1581
1582                 sk->sk_user_data = xprt;
1583                 transport->old_data_ready = sk->sk_data_ready;
1584                 transport->old_state_change = sk->sk_state_change;
1585                 transport->old_write_space = sk->sk_write_space;
1586                 sk->sk_data_ready = xs_tcp_data_ready;
1587                 sk->sk_state_change = xs_tcp_state_change;
1588                 sk->sk_write_space = xs_tcp_write_space;
1589                 sk->sk_allocation = GFP_ATOMIC;
1590
1591                 /* socket options */
1592                 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
1593                 sock_reset_flag(sk, SOCK_LINGER);
1594                 tcp_sk(sk)->linger2 = 0;
1595                 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
1596
1597                 xprt_clear_connected(xprt);
1598
1599                 /* Reset to new socket */
1600                 transport->sock = sock;
1601                 transport->inet = sk;
1602
1603                 write_unlock_bh(&sk->sk_callback_lock);
1604         }
1605
1606         /* Tell the socket layer to start connecting... */
1607         xprt->stat.connect_count++;
1608         xprt->stat.connect_start = jiffies;
1609         return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
1610 }
1611
1612 /**
1613  * xs_tcp_connect_worker4 - connect a TCP socket to a remote endpoint
1614  * @work: RPC transport to connect
1615  *
1616  * Invoked by a work queue tasklet.
1617  */
1618 static void xs_tcp_connect_worker4(struct work_struct *work)
1619 {
1620         struct sock_xprt *transport =
1621                 container_of(work, struct sock_xprt, connect_worker.work);
1622         struct rpc_xprt *xprt = &transport->xprt;
1623         struct socket *sock = transport->sock;
1624         int err, status = -EIO;
1625
1626         if (xprt->shutdown || !xprt_bound(xprt))
1627                 goto out;
1628
1629         if (!sock) {
1630                 /* start from scratch */
1631                 if ((err = sock_create_kern(PF_INET, SOCK_STREAM, IPPROTO_TCP, &sock)) < 0) {
1632                         dprintk("RPC:       can't create TCP transport socket (%d).\n", -err);
1633                         goto out;
1634                 }
1635                 xs_reclassify_socket4(sock);
1636
1637                 if (xs_bind4(transport, sock) < 0) {
1638                         sock_release(sock);
1639                         goto out;
1640                 }
1641         } else
1642                 /* "close" the socket, preserving the local port */
1643                 xs_tcp_reuse_connection(xprt);
1644
1645         dprintk("RPC:       worker connecting xprt %p to address: %s\n",
1646                         xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1647
1648         status = xs_tcp_finish_connecting(xprt, sock);
1649         dprintk("RPC:       %p connect status %d connected %d sock state %d\n",
1650                         xprt, -status, xprt_connected(xprt),
1651                         sock->sk->sk_state);
1652         if (status < 0) {
1653                 switch (status) {
1654                         case -EINPROGRESS:
1655                         case -EALREADY:
1656                                 goto out_clear;
1657                         case -ECONNREFUSED:
1658                         case -ECONNRESET:
1659                                 /* retry with existing socket, after a delay */
1660                                 break;
1661                         default:
1662                                 /* get rid of existing socket, and retry */
1663                                 xs_tcp_shutdown(xprt);
1664                 }
1665         }
1666 out:
1667         xprt_wake_pending_tasks(xprt, status);
1668 out_clear:
1669         xprt_clear_connecting(xprt);
1670 }
1671
1672 /**
1673  * xs_tcp_connect_worker6 - connect a TCP socket to a remote endpoint
1674  * @work: RPC transport to connect
1675  *
1676  * Invoked by a work queue tasklet.
1677  */
1678 static void xs_tcp_connect_worker6(struct work_struct *work)
1679 {
1680         struct sock_xprt *transport =
1681                 container_of(work, struct sock_xprt, connect_worker.work);
1682         struct rpc_xprt *xprt = &transport->xprt;
1683         struct socket *sock = transport->sock;
1684         int err, status = -EIO;
1685
1686         if (xprt->shutdown || !xprt_bound(xprt))
1687                 goto out;
1688
1689         if (!sock) {
1690                 /* start from scratch */
1691                 if ((err = sock_create_kern(PF_INET6, SOCK_STREAM, IPPROTO_TCP, &sock)) < 0) {
1692                         dprintk("RPC:       can't create TCP transport socket (%d).\n", -err);
1693                         goto out;
1694                 }
1695                 xs_reclassify_socket6(sock);
1696
1697                 if (xs_bind6(transport, sock) < 0) {
1698                         sock_release(sock);
1699                         goto out;
1700                 }
1701         } else
1702                 /* "close" the socket, preserving the local port */
1703                 xs_tcp_reuse_connection(xprt);
1704
1705         dprintk("RPC:       worker connecting xprt %p to address: %s\n",
1706                         xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1707
1708         status = xs_tcp_finish_connecting(xprt, sock);
1709         dprintk("RPC:       %p connect status %d connected %d sock state %d\n",
1710                         xprt, -status, xprt_connected(xprt), sock->sk->sk_state);
1711         if (status < 0) {
1712                 switch (status) {
1713                         case -EINPROGRESS:
1714                         case -EALREADY:
1715                                 goto out_clear;
1716                         case -ECONNREFUSED:
1717                         case -ECONNRESET:
1718                                 /* retry with existing socket, after a delay */
1719                                 break;
1720                         default:
1721                                 /* get rid of existing socket, and retry */
1722                                 xs_tcp_shutdown(xprt);
1723                 }
1724         }
1725 out:
1726         xprt_wake_pending_tasks(xprt, status);
1727 out_clear:
1728         xprt_clear_connecting(xprt);
1729 }
1730
1731 /**
1732  * xs_connect - connect a socket to a remote endpoint
1733  * @task: address of RPC task that manages state of connect request
1734  *
1735  * TCP: If the remote end dropped the connection, delay reconnecting.
1736  *
1737  * UDP socket connects are synchronous, but we use a work queue anyway
1738  * to guarantee that even unprivileged user processes can set up a
1739  * socket on a privileged port.
1740  *
1741  * If a UDP socket connect fails, the delay behavior here prevents
1742  * retry floods (hard mounts).
1743  */
1744 static void xs_connect(struct rpc_task *task)
1745 {
1746         struct rpc_xprt *xprt = task->tk_xprt;
1747         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1748
1749         if (xprt_test_and_set_connecting(xprt))
1750                 return;
1751
1752         if (transport->sock != NULL) {
1753                 dprintk("RPC:       xs_connect delayed xprt %p for %lu "
1754                                 "seconds\n",
1755                                 xprt, xprt->reestablish_timeout / HZ);
1756                 queue_delayed_work(rpciod_workqueue,
1757                                    &transport->connect_worker,
1758                                    xprt->reestablish_timeout);
1759                 xprt->reestablish_timeout <<= 1;
1760                 if (xprt->reestablish_timeout > XS_TCP_MAX_REEST_TO)
1761                         xprt->reestablish_timeout = XS_TCP_MAX_REEST_TO;
1762         } else {
1763                 dprintk("RPC:       xs_connect scheduled xprt %p\n", xprt);
1764                 queue_delayed_work(rpciod_workqueue,
1765                                    &transport->connect_worker, 0);
1766         }
1767 }
1768
1769 static void xs_tcp_connect(struct rpc_task *task)
1770 {
1771         struct rpc_xprt *xprt = task->tk_xprt;
1772
1773         /* Initiate graceful shutdown of the socket if not already done */
1774         if (test_bit(XPRT_CONNECTED, &xprt->state))
1775                 xs_tcp_shutdown(xprt);
1776         /* Exit if we need to wait for socket shutdown to complete */
1777         if (test_bit(XPRT_CLOSING, &xprt->state))
1778                 return;
1779         xs_connect(task);
1780 }
1781
1782 /**
1783  * xs_udp_print_stats - display UDP socket-specifc stats
1784  * @xprt: rpc_xprt struct containing statistics
1785  * @seq: output file
1786  *
1787  */
1788 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
1789 {
1790         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1791
1792         seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %Lu %Lu\n",
1793                         transport->port,
1794                         xprt->stat.bind_count,
1795                         xprt->stat.sends,
1796                         xprt->stat.recvs,
1797                         xprt->stat.bad_xids,
1798                         xprt->stat.req_u,
1799                         xprt->stat.bklog_u);
1800 }
1801
1802 /**
1803  * xs_tcp_print_stats - display TCP socket-specifc stats
1804  * @xprt: rpc_xprt struct containing statistics
1805  * @seq: output file
1806  *
1807  */
1808 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
1809 {
1810         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1811         long idle_time = 0;
1812
1813         if (xprt_connected(xprt))
1814                 idle_time = (long)(jiffies - xprt->last_used) / HZ;
1815
1816         seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu %Lu %Lu\n",
1817                         transport->port,
1818                         xprt->stat.bind_count,
1819                         xprt->stat.connect_count,
1820                         xprt->stat.connect_time,
1821                         idle_time,
1822                         xprt->stat.sends,
1823                         xprt->stat.recvs,
1824                         xprt->stat.bad_xids,
1825                         xprt->stat.req_u,
1826                         xprt->stat.bklog_u);
1827 }
1828
1829 static struct rpc_xprt_ops xs_udp_ops = {
1830         .set_buffer_size        = xs_udp_set_buffer_size,
1831         .reserve_xprt           = xprt_reserve_xprt_cong,
1832         .release_xprt           = xprt_release_xprt_cong,
1833         .rpcbind                = rpcb_getport_async,
1834         .set_port               = xs_set_port,
1835         .connect                = xs_connect,
1836         .buf_alloc              = rpc_malloc,
1837         .buf_free               = rpc_free,
1838         .send_request           = xs_udp_send_request,
1839         .set_retrans_timeout    = xprt_set_retrans_timeout_rtt,
1840         .timer                  = xs_udp_timer,
1841         .release_request        = xprt_release_rqst_cong,
1842         .close                  = xs_close,
1843         .destroy                = xs_destroy,
1844         .print_stats            = xs_udp_print_stats,
1845 };
1846
1847 static struct rpc_xprt_ops xs_tcp_ops = {
1848         .reserve_xprt           = xprt_reserve_xprt,
1849         .release_xprt           = xs_tcp_release_xprt,
1850         .rpcbind                = rpcb_getport_async,
1851         .set_port               = xs_set_port,
1852         .connect                = xs_tcp_connect,
1853         .buf_alloc              = rpc_malloc,
1854         .buf_free               = rpc_free,
1855         .send_request           = xs_tcp_send_request,
1856         .set_retrans_timeout    = xprt_set_retrans_timeout_def,
1857         .close                  = xs_tcp_shutdown,
1858         .destroy                = xs_destroy,
1859         .print_stats            = xs_tcp_print_stats,
1860 };
1861
1862 static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
1863                                       unsigned int slot_table_size)
1864 {
1865         struct rpc_xprt *xprt;
1866         struct sock_xprt *new;
1867
1868         if (args->addrlen > sizeof(xprt->addr)) {
1869                 dprintk("RPC:       xs_setup_xprt: address too large\n");
1870                 return ERR_PTR(-EBADF);
1871         }
1872
1873         new = kzalloc(sizeof(*new), GFP_KERNEL);
1874         if (new == NULL) {
1875                 dprintk("RPC:       xs_setup_xprt: couldn't allocate "
1876                                 "rpc_xprt\n");
1877                 return ERR_PTR(-ENOMEM);
1878         }
1879         xprt = &new->xprt;
1880
1881         xprt->max_reqs = slot_table_size;
1882         xprt->slot = kcalloc(xprt->max_reqs, sizeof(struct rpc_rqst), GFP_KERNEL);
1883         if (xprt->slot == NULL) {
1884                 kfree(xprt);
1885                 dprintk("RPC:       xs_setup_xprt: couldn't allocate slot "
1886                                 "table\n");
1887                 return ERR_PTR(-ENOMEM);
1888         }
1889
1890         memcpy(&xprt->addr, args->dstaddr, args->addrlen);
1891         xprt->addrlen = args->addrlen;
1892         if (args->srcaddr)
1893                 memcpy(&new->addr, args->srcaddr, args->addrlen);
1894
1895         return xprt;
1896 }
1897
1898 /**
1899  * xs_setup_udp - Set up transport to use a UDP socket
1900  * @args: rpc transport creation arguments
1901  *
1902  */
1903 static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
1904 {
1905         struct sockaddr *addr = args->dstaddr;
1906         struct rpc_xprt *xprt;
1907         struct sock_xprt *transport;
1908
1909         xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries);
1910         if (IS_ERR(xprt))
1911                 return xprt;
1912         transport = container_of(xprt, struct sock_xprt, xprt);
1913
1914         xprt->prot = IPPROTO_UDP;
1915         xprt->tsh_size = 0;
1916         /* XXX: header size can vary due to auth type, IPv6, etc. */
1917         xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
1918
1919         xprt->bind_timeout = XS_BIND_TO;
1920         xprt->connect_timeout = XS_UDP_CONN_TO;
1921         xprt->reestablish_timeout = XS_UDP_REEST_TO;
1922         xprt->idle_timeout = XS_IDLE_DISC_TO;
1923
1924         xprt->ops = &xs_udp_ops;
1925
1926         if (args->timeout)
1927                 xprt->timeout = *args->timeout;
1928         else
1929                 xprt_set_timeout(&xprt->timeout, 5, 5 * HZ);
1930
1931         switch (addr->sa_family) {
1932         case AF_INET:
1933                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
1934                         xprt_set_bound(xprt);
1935
1936                 INIT_DELAYED_WORK(&transport->connect_worker,
1937                                         xs_udp_connect_worker4);
1938                 xs_format_ipv4_peer_addresses(xprt);
1939                 break;
1940         case AF_INET6:
1941                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
1942                         xprt_set_bound(xprt);
1943
1944                 INIT_DELAYED_WORK(&transport->connect_worker,
1945                                         xs_udp_connect_worker6);
1946                 xs_format_ipv6_peer_addresses(xprt);
1947                 break;
1948         default:
1949                 kfree(xprt);
1950                 return ERR_PTR(-EAFNOSUPPORT);
1951         }
1952
1953         dprintk("RPC:       set up transport to address %s\n",
1954                         xprt->address_strings[RPC_DISPLAY_ALL]);
1955
1956         if (try_module_get(THIS_MODULE))
1957                 return xprt;
1958
1959         kfree(xprt->slot);
1960         kfree(xprt);
1961         return ERR_PTR(-EINVAL);
1962 }
1963
1964 /**
1965  * xs_setup_tcp - Set up transport to use a TCP socket
1966  * @args: rpc transport creation arguments
1967  *
1968  */
1969 static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
1970 {
1971         struct sockaddr *addr = args->dstaddr;
1972         struct rpc_xprt *xprt;
1973         struct sock_xprt *transport;
1974
1975         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries);
1976         if (IS_ERR(xprt))
1977                 return xprt;
1978         transport = container_of(xprt, struct sock_xprt, xprt);
1979
1980         xprt->prot = IPPROTO_TCP;
1981         xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
1982         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
1983
1984         xprt->bind_timeout = XS_BIND_TO;
1985         xprt->connect_timeout = XS_TCP_CONN_TO;
1986         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1987         xprt->idle_timeout = XS_IDLE_DISC_TO;
1988
1989         xprt->ops = &xs_tcp_ops;
1990
1991         if (args->timeout)
1992                 xprt->timeout = *args->timeout;
1993         else
1994                 xprt_set_timeout(&xprt->timeout, 2, 60 * HZ);
1995
1996         switch (addr->sa_family) {
1997         case AF_INET:
1998                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
1999                         xprt_set_bound(xprt);
2000
2001                 INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_connect_worker4);
2002                 xs_format_ipv4_peer_addresses(xprt);
2003                 break;
2004         case AF_INET6:
2005                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2006                         xprt_set_bound(xprt);
2007
2008                 INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_connect_worker6);
2009                 xs_format_ipv6_peer_addresses(xprt);
2010                 break;
2011         default:
2012                 kfree(xprt);
2013                 return ERR_PTR(-EAFNOSUPPORT);
2014         }
2015
2016         dprintk("RPC:       set up transport to address %s\n",
2017                         xprt->address_strings[RPC_DISPLAY_ALL]);
2018
2019         if (try_module_get(THIS_MODULE))
2020                 return xprt;
2021
2022         kfree(xprt->slot);
2023         kfree(xprt);
2024         return ERR_PTR(-EINVAL);
2025 }
2026
2027 static struct xprt_class        xs_udp_transport = {
2028         .list           = LIST_HEAD_INIT(xs_udp_transport.list),
2029         .name           = "udp",
2030         .owner          = THIS_MODULE,
2031         .ident          = IPPROTO_UDP,
2032         .setup          = xs_setup_udp,
2033 };
2034
2035 static struct xprt_class        xs_tcp_transport = {
2036         .list           = LIST_HEAD_INIT(xs_tcp_transport.list),
2037         .name           = "tcp",
2038         .owner          = THIS_MODULE,
2039         .ident          = IPPROTO_TCP,
2040         .setup          = xs_setup_tcp,
2041 };
2042
2043 /**
2044  * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
2045  *
2046  */
2047 int init_socket_xprt(void)
2048 {
2049 #ifdef RPC_DEBUG
2050         if (!sunrpc_table_header)
2051                 sunrpc_table_header = register_sysctl_table(sunrpc_table);
2052 #endif
2053
2054         xprt_register_transport(&xs_udp_transport);
2055         xprt_register_transport(&xs_tcp_transport);
2056
2057         return 0;
2058 }
2059
2060 /**
2061  * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
2062  *
2063  */
2064 void cleanup_socket_xprt(void)
2065 {
2066 #ifdef RPC_DEBUG
2067         if (sunrpc_table_header) {
2068                 unregister_sysctl_table(sunrpc_table_header);
2069                 sunrpc_table_header = NULL;
2070         }
2071 #endif
2072
2073         xprt_unregister_transport(&xs_udp_transport);
2074         xprt_unregister_transport(&xs_tcp_transport);
2075 }