Pileus Git - ~andy/linux/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 static inline void xs_encode_tcp_record_marker(struct xdr_buf *buf)
 618 {
 619         u32 reclen = buf->len - sizeof(rpc_fraghdr);
 620         rpc_fraghdr *base = buf->head[0].iov_base;
 621         *base = htonl(RPC_LAST_STREAM_FRAGMENT | reclen);
 622 }
 623
 624 /**
 625  * xs_tcp_send_request - write an RPC request to a TCP socket
 626  * @task: address of RPC task that manages the state of an RPC request
 627  *
 628  * Return values:
 629  *        0:    The request has been sent
 630  *   EAGAIN:    The socket was blocked, please call again later to
 631  *              complete the request
 632  * ENOTCONN:    Caller needs to invoke connect logic then call again
 633  *    other:    Some other error occured, the request was not sent
 634  *
 635  * XXX: In the case of soft timeouts, should we eventually give up
 636  *      if sendmsg is not able to make progress?
 637  */
 638 static int xs_tcp_send_request(struct rpc_task *task)
 639 {
 640         struct rpc_rqst *req = task->tk_rqstp;
 641         struct rpc_xprt *xprt = req->rq_xprt;
 642         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 643         struct xdr_buf *xdr = &req->rq_snd_buf;
 644         int status;
 645         unsigned int retry = 0;
 646
 647         xs_encode_tcp_record_marker(&req->rq_snd_buf);
 648
 649         xs_pktdump("packet data:",
 650                                 req->rq_svec->iov_base,
 651                                 req->rq_svec->iov_len);
 652
 653         /* Continue transmitting the packet/record. We must be careful
 654          * to cope with writespace callbacks arriving _after_ we have
 655          * called sendmsg(). */
 656         while (1) {
 657                 req->rq_xtime = jiffies;
 658                 status = xs_sendpages(transport->sock,
 659                                         NULL, 0, xdr, req->rq_bytes_sent);
 660
 661                 dprintk("RPC:       xs_tcp_send_request(%u) = %d\n",
 662                                 xdr->len - req->rq_bytes_sent, status);
 663
 664                 if (unlikely(status < 0))
 665                         break;
 666
 667                 /* If we've sent the entire packet, immediately
 668                  * reset the count of bytes sent. */
 669                 req->rq_bytes_sent += status;
 670                 task->tk_bytes_sent += status;
 671                 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
 672                         req->rq_bytes_sent = 0;
 673                         return 0;
 674                 }
 675
 676                 status = -EAGAIN;
 677                 if (retry++ > XS_SENDMSG_RETRY)
 678                         break;
 679         }
 680
 681         switch (status) {
 682         case -EAGAIN:
 683                 xs_nospace(task);
 684                 break;
 685         case -ECONNREFUSED:
 686         case -ECONNRESET:
 687         case -ENOTCONN:
 688         case -EPIPE:
 689                 status = -ENOTCONN;
 690                 break;
 691         default:
 692                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
 693                         -status);
 694                 xprt_disconnect(xprt);
 695                 break;
 696         }
 697
 698         return status;
 699 }
 700
 701 /**
 702  * xs_tcp_release_xprt - clean up after a tcp transmission
 703  * @xprt: transport
 704  * @task: rpc task
 705  *
 706  * This cleans up if an error causes us to abort the transmission of a request.
 707  * In this case, the socket may need to be reset in order to avoid confusing
 708  * the server.
 709  */
 710 static void xs_tcp_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
 711 {
 712         struct rpc_rqst *req;
 713
 714         if (task != xprt->snd_task)
 715                 return;
 716         if (task == NULL)
 717                 goto out_release;
 718         req = task->tk_rqstp;
 719         if (req->rq_bytes_sent == 0)
 720                 goto out_release;
 721         if (req->rq_bytes_sent == req->rq_snd_buf.len)
 722                 goto out_release;
 723         set_bit(XPRT_CLOSE_WAIT, &task->tk_xprt->state);
 724 out_release:
 725         xprt_release_xprt(xprt, task);
 726 }
 727
 728 /**
 729  * xs_close - close a socket
 730  * @xprt: transport
 731  *
 732  * This is used when all requests are complete; ie, no DRC state remains
 733  * on the server we want to save.
 734  */
 735 static void xs_close(struct rpc_xprt *xprt)
 736 {
 737         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 738         struct socket *sock = transport->sock;
 739         struct sock *sk = transport->inet;
 740
 741         if (!sk)
 742                 goto clear_close_wait;
 743
 744         dprintk("RPC:       xs_close xprt %p\n", xprt);
 745
 746         write_lock_bh(&sk->sk_callback_lock);
 747         transport->inet = NULL;
 748         transport->sock = NULL;
 749
 750         sk->sk_user_data = NULL;
 751         sk->sk_data_ready = transport->old_data_ready;
 752         sk->sk_state_change = transport->old_state_change;
 753         sk->sk_write_space = transport->old_write_space;
 754         write_unlock_bh(&sk->sk_callback_lock);
 755
 756         sk->sk_no_check = 0;
 757
 758         sock_release(sock);
 759 clear_close_wait:
 760         smp_mb__before_clear_bit();
 761         clear_bit(XPRT_CONNECTED, &xprt->state);
 762         clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
 763         clear_bit(XPRT_CLOSING, &xprt->state);
 764         smp_mb__after_clear_bit();
 765 }
 766
 767 /**
 768  * xs_destroy - prepare to shutdown a transport
 769  * @xprt: doomed transport
 770  *
 771  */
 772 static void xs_destroy(struct rpc_xprt *xprt)
 773 {
 774         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 775
 776         dprintk("RPC:       xs_destroy xprt %p\n", xprt);
 777
 778         cancel_rearming_delayed_work(&transport->connect_worker);
 779
 780         xprt_disconnect(xprt);
 781         xs_close(xprt);
 782         xs_free_peer_addresses(xprt);
 783         kfree(xprt->slot);
 784         kfree(xprt);
 785         module_put(THIS_MODULE);
 786 }
 787
 788 static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
 789 {
 790         return (struct rpc_xprt *) sk->sk_user_data;
 791 }
 792
 793 /**
 794  * xs_udp_data_ready - "data ready" callback for UDP sockets
 795  * @sk: socket with data to read
 796  * @len: how much data to read
 797  *
 798  */
 799 static void xs_udp_data_ready(struct sock *sk, int len)
 800 {
 801         struct rpc_task *task;
 802         struct rpc_xprt *xprt;
 803         struct rpc_rqst *rovr;
 804         struct sk_buff *skb;
 805         int err, repsize, copied;
 806         u32 _xid;
 807         __be32 *xp;
 808
 809         read_lock(&sk->sk_callback_lock);
 810         dprintk("RPC:       xs_udp_data_ready...\n");
 811         if (!(xprt = xprt_from_sock(sk)))
 812                 goto out;
 813
 814         if ((skb = skb_recv_datagram(sk, 0, 1, &err)) == NULL)
 815                 goto out;
 816
 817         if (xprt->shutdown)
 818                 goto dropit;
 819
 820         repsize = skb->len - sizeof(struct udphdr);
 821         if (repsize < 4) {
 822                 dprintk("RPC:       impossible RPC reply size %d!\n", repsize);
 823                 goto dropit;
 824         }
 825
 826         /* Copy the XID from the skb... */
 827         xp = skb_header_pointer(skb, sizeof(struct udphdr),
 828                                 sizeof(_xid), &_xid);
 829         if (xp == NULL)
 830                 goto dropit;
 831
 832         /* Look up and lock the request corresponding to the given XID */
 833         spin_lock(&xprt->transport_lock);
 834         rovr = xprt_lookup_rqst(xprt, *xp);
 835         if (!rovr)
 836                 goto out_unlock;
 837         task = rovr->rq_task;
 838
 839         if ((copied = rovr->rq_private_buf.buflen) > repsize)
 840                 copied = repsize;
 841
 842         /* Suck it into the iovec, verify checksum if not done by hw. */
 843         if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
 844                 UDPX_INC_STATS_BH(sk, UDP_MIB_INERRORS);
 845                 goto out_unlock;
 846         }
 847
 848         UDPX_INC_STATS_BH(sk, UDP_MIB_INDATAGRAMS);
 849
 850         /* Something worked... */
 851         dst_confirm(skb->dst);
 852
 853         xprt_adjust_cwnd(task, copied);
 854         xprt_update_rtt(task);
 855         xprt_complete_rqst(task, copied);
 856
 857  out_unlock:
 858         spin_unlock(&xprt->transport_lock);
 859  dropit:
 860         skb_free_datagram(sk, skb);
 861  out:
 862         read_unlock(&sk->sk_callback_lock);
 863 }
 864
 865 static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
 866 {
 867         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 868         size_t len, used;
 869         char *p;
 870
 871         p = ((char *) &transport->tcp_fraghdr) + transport->tcp_offset;
 872         len = sizeof(transport->tcp_fraghdr) - transport->tcp_offset;
 873         used = xdr_skb_read_bits(desc, p, len);
 874         transport->tcp_offset += used;
 875         if (used != len)
 876                 return;
 877
 878         transport->tcp_reclen = ntohl(transport->tcp_fraghdr);
 879         if (transport->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
 880                 transport->tcp_flags |= TCP_RCV_LAST_FRAG;
 881         else
 882                 transport->tcp_flags &= ~TCP_RCV_LAST_FRAG;
 883         transport->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
 884
 885         transport->tcp_flags &= ~TCP_RCV_COPY_FRAGHDR;
 886         transport->tcp_offset = 0;
 887
 888         /* Sanity check of the record length */
 889         if (unlikely(transport->tcp_reclen < 4)) {
 890                 dprintk("RPC:       invalid TCP record fragment length\n");
 891                 xprt_disconnect(xprt);
 892                 return;
 893         }
 894         dprintk("RPC:       reading TCP record fragment of length %d\n",
 895                         transport->tcp_reclen);
 896 }
 897
 898 static void xs_tcp_check_fraghdr(struct sock_xprt *transport)
 899 {
 900         if (transport->tcp_offset == transport->tcp_reclen) {
 901                 transport->tcp_flags |= TCP_RCV_COPY_FRAGHDR;
 902                 transport->tcp_offset = 0;
 903                 if (transport->tcp_flags & TCP_RCV_LAST_FRAG) {
 904                         transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
 905                         transport->tcp_flags |= TCP_RCV_COPY_XID;
 906                         transport->tcp_copied = 0;
 907                 }
 908         }
 909 }
 910
 911 static inline void xs_tcp_read_xid(struct sock_xprt *transport, struct xdr_skb_reader *desc)
 912 {
 913         size_t len, used;
 914         char *p;
 915
 916         len = sizeof(transport->tcp_xid) - transport->tcp_offset;
 917         dprintk("RPC:       reading XID (%Zu bytes)\n", len);
 918         p = ((char *) &transport->tcp_xid) + transport->tcp_offset;
 919         used = xdr_skb_read_bits(desc, p, len);
 920         transport->tcp_offset += used;
 921         if (used != len)
 922                 return;
 923         transport->tcp_flags &= ~TCP_RCV_COPY_XID;
 924         transport->tcp_flags |= TCP_RCV_COPY_DATA;
 925         transport->tcp_copied = 4;
 926         dprintk("RPC:       reading reply for XID %08x\n",
 927                         ntohl(transport->tcp_xid));
 928         xs_tcp_check_fraghdr(transport);
 929 }
 930
 931 static inline void xs_tcp_read_request(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
 932 {
 933         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 934         struct rpc_rqst *req;
 935         struct xdr_buf *rcvbuf;
 936         size_t len;
 937         ssize_t r;
 938
 939         /* Find and lock the request corresponding to this xid */
 940         spin_lock(&xprt->transport_lock);
 941         req = xprt_lookup_rqst(xprt, transport->tcp_xid);
 942         if (!req) {
 943                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
 944                 dprintk("RPC:       XID %08x request not found!\n",
 945                                 ntohl(transport->tcp_xid));
 946                 spin_unlock(&xprt->transport_lock);
 947                 return;
 948         }
 949
 950         rcvbuf = &req->rq_private_buf;
 951         len = desc->count;
 952         if (len > transport->tcp_reclen - transport->tcp_offset) {
 953                 struct xdr_skb_reader my_desc;
 954
 955                 len = transport->tcp_reclen - transport->tcp_offset;
 956                 memcpy(&my_desc, desc, sizeof(my_desc));
 957                 my_desc.count = len;
 958                 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
 959                                           &my_desc, xdr_skb_read_bits);
 960                 desc->count -= r;
 961                 desc->offset += r;
 962         } else
 963                 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
 964                                           desc, xdr_skb_read_bits);
 965
 966         if (r > 0) {
 967                 transport->tcp_copied += r;
 968                 transport->tcp_offset += r;
 969         }
 970         if (r != len) {
 971                 /* Error when copying to the receive buffer,
 972                  * usually because we weren't able to allocate
 973                  * additional buffer pages. All we can do now
 974                  * is turn off TCP_RCV_COPY_DATA, so the request
 975                  * will not receive any additional updates,
 976                  * and time out.
 977                  * Any remaining data from this record will
 978                  * be discarded.
 979                  */
 980                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
 981                 dprintk("RPC:       XID %08x truncated request\n",
 982                                 ntohl(transport->tcp_xid));
 983                 dprintk("RPC:       xprt = %p, tcp_copied = %lu, "
 984                                 "tcp_offset = %u, tcp_reclen = %u\n",
 985                                 xprt, transport->tcp_copied,
 986                                 transport->tcp_offset, transport->tcp_reclen);
 987                 goto out;
 988         }
 989
 990         dprintk("RPC:       XID %08x read %Zd bytes\n",
 991                         ntohl(transport->tcp_xid), r);
 992         dprintk("RPC:       xprt = %p, tcp_copied = %lu, tcp_offset = %u, "
 993                         "tcp_reclen = %u\n", xprt, transport->tcp_copied,
 994                         transport->tcp_offset, transport->tcp_reclen);
 995
 996         if (transport->tcp_copied == req->rq_private_buf.buflen)
 997                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
 998         else if (transport->tcp_offset == transport->tcp_reclen) {
 999                 if (transport->tcp_flags & TCP_RCV_LAST_FRAG)
1000                         transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1001         }
1002
1003 out:
1004         if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1005                 xprt_complete_rqst(req->rq_task, transport->tcp_copied);
1006         spin_unlock(&xprt->transport_lock);
1007         xs_tcp_check_fraghdr(transport);
1008 }
1009
1010 static inline void xs_tcp_read_discard(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1011 {
1012         size_t len;
1013
1014         len = transport->tcp_reclen - transport->tcp_offset;
1015         if (len > desc->count)
1016                 len = desc->count;
1017         desc->count -= len;
1018         desc->offset += len;
1019         transport->tcp_offset += len;
1020         dprintk("RPC:       discarded %Zu bytes\n", len);
1021         xs_tcp_check_fraghdr(transport);
1022 }
1023
1024 static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
1025 {
1026         struct rpc_xprt *xprt = rd_desc->arg.data;
1027         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1028         struct xdr_skb_reader desc = {
1029                 .skb    = skb,
1030                 .offset = offset,
1031                 .count  = len,
1032         };
1033
1034         dprintk("RPC:       xs_tcp_data_recv started\n");
1035         do {
1036                 /* Read in a new fragment marker if necessary */
1037                 /* Can we ever really expect to get completely empty fragments? */
1038                 if (transport->tcp_flags & TCP_RCV_COPY_FRAGHDR) {
1039                         xs_tcp_read_fraghdr(xprt, &desc);
1040                         continue;
1041                 }
1042                 /* Read in the xid if necessary */
1043                 if (transport->tcp_flags & TCP_RCV_COPY_XID) {
1044                         xs_tcp_read_xid(transport, &desc);
1045                         continue;
1046                 }
1047                 /* Read in the request data */
1048                 if (transport->tcp_flags & TCP_RCV_COPY_DATA) {
1049                         xs_tcp_read_request(xprt, &desc);
1050                         continue;
1051                 }
1052                 /* Skip over any trailing bytes on short reads */
1053                 xs_tcp_read_discard(transport, &desc);
1054         } while (desc.count);
1055         dprintk("RPC:       xs_tcp_data_recv done\n");
1056         return len - desc.count;
1057 }
1058
1059 /**
1060  * xs_tcp_data_ready - "data ready" callback for TCP sockets
1061  * @sk: socket with data to read
1062  * @bytes: how much data to read
1063  *
1064  */
1065 static void xs_tcp_data_ready(struct sock *sk, int bytes)
1066 {
1067         struct rpc_xprt *xprt;
1068         read_descriptor_t rd_desc;
1069
1070         dprintk("RPC:       xs_tcp_data_ready...\n");
1071
1072         read_lock(&sk->sk_callback_lock);
1073         if (!(xprt = xprt_from_sock(sk)))
1074                 goto out;
1075         if (xprt->shutdown)
1076                 goto out;
1077
1078         /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
1079         rd_desc.arg.data = xprt;
1080         rd_desc.count = 65536;
1081         tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
1082 out:
1083         read_unlock(&sk->sk_callback_lock);
1084 }
1085
1086 /**
1087  * xs_tcp_state_change - callback to handle TCP socket state changes
1088  * @sk: socket whose state has changed
1089  *
1090  */
1091 static void xs_tcp_state_change(struct sock *sk)
1092 {
1093         struct rpc_xprt *xprt;
1094
1095         read_lock(&sk->sk_callback_lock);
1096         if (!(xprt = xprt_from_sock(sk)))
1097                 goto out;
1098         dprintk("RPC:       xs_tcp_state_change client %p...\n", xprt);
1099         dprintk("RPC:       state %x conn %d dead %d zapped %d\n",
1100                         sk->sk_state, xprt_connected(xprt),
1101                         sock_flag(sk, SOCK_DEAD),
1102                         sock_flag(sk, SOCK_ZAPPED));
1103
1104         switch (sk->sk_state) {
1105         case TCP_ESTABLISHED:
1106                 spin_lock_bh(&xprt->transport_lock);
1107                 if (!xprt_test_and_set_connected(xprt)) {
1108                         struct sock_xprt *transport = container_of(xprt,
1109                                         struct sock_xprt, xprt);
1110
1111                         /* Reset TCP record info */
1112                         transport->tcp_offset = 0;
1113                         transport->tcp_reclen = 0;
1114                         transport->tcp_copied = 0;
1115                         transport->tcp_flags =
1116                                 TCP_RCV_COPY_FRAGHDR | TCP_RCV_COPY_XID;
1117
1118                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1119                         xprt_wake_pending_tasks(xprt, 0);
1120                 }
1121                 spin_unlock_bh(&xprt->transport_lock);
1122                 break;
1123         case TCP_FIN_WAIT1:
1124                 /* The client initiated a shutdown of the socket */
1125                 set_bit(XPRT_CLOSING, &xprt->state);
1126                 smp_mb__before_clear_bit();
1127                 clear_bit(XPRT_CONNECTED, &xprt->state);
1128                 smp_mb__after_clear_bit();
1129                 break;
1130         case TCP_CLOSE_WAIT:
1131                 /* The server initiated a shutdown of the socket */
1132                 set_bit(XPRT_CLOSING, &xprt->state);
1133                 xprt_force_disconnect(xprt);
1134                 break;
1135         case TCP_LAST_ACK:
1136                 smp_mb__before_clear_bit();
1137                 clear_bit(XPRT_CONNECTED, &xprt->state);
1138                 smp_mb__after_clear_bit();
1139                 break;
1140         case TCP_CLOSE:
1141                 smp_mb__before_clear_bit();
1142                 clear_bit(XPRT_CLOSING, &xprt->state);
1143                 smp_mb__after_clear_bit();
1144                 /* Mark transport as closed and wake up all pending tasks */
1145                 xprt_disconnect(xprt);
1146         }
1147  out:
1148         read_unlock(&sk->sk_callback_lock);
1149 }
1150
1151 /**
1152  * xs_udp_write_space - callback invoked when socket buffer space
1153  *                             becomes available
1154  * @sk: socket whose state has changed
1155  *
1156  * Called when more output buffer space is available for this socket.
1157  * We try not to wake our writers until they can make "significant"
1158  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1159  * with a bunch of small requests.
1160  */
1161 static void xs_udp_write_space(struct sock *sk)
1162 {
1163         read_lock(&sk->sk_callback_lock);
1164
1165         /* from net/core/sock.c:sock_def_write_space */
1166         if (sock_writeable(sk)) {
1167                 struct socket *sock;
1168                 struct rpc_xprt *xprt;
1169
1170                 if (unlikely(!(sock = sk->sk_socket)))
1171                         goto out;
1172                 if (unlikely(!(xprt = xprt_from_sock(sk))))
1173                         goto out;
1174                 if (unlikely(!test_and_clear_bit(SOCK_NOSPACE, &sock->flags)))
1175                         goto out;
1176
1177                 xprt_write_space(xprt);
1178         }
1179
1180  out:
1181         read_unlock(&sk->sk_callback_lock);
1182 }
1183
1184 /**
1185  * xs_tcp_write_space - callback invoked when socket buffer space
1186  *                             becomes available
1187  * @sk: socket whose state has changed
1188  *
1189  * Called when more output buffer space is available for this socket.
1190  * We try not to wake our writers until they can make "significant"
1191  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1192  * with a bunch of small requests.
1193  */
1194 static void xs_tcp_write_space(struct sock *sk)
1195 {
1196         read_lock(&sk->sk_callback_lock);
1197
1198         /* from net/core/stream.c:sk_stream_write_space */
1199         if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk)) {
1200                 struct socket *sock;
1201                 struct rpc_xprt *xprt;
1202
1203                 if (unlikely(!(sock = sk->sk_socket)))
1204                         goto out;
1205                 if (unlikely(!(xprt = xprt_from_sock(sk))))
1206                         goto out;
1207                 if (unlikely(!test_and_clear_bit(SOCK_NOSPACE, &sock->flags)))
1208                         goto out;
1209
1210                 xprt_write_space(xprt);
1211         }
1212
1213  out:
1214         read_unlock(&sk->sk_callback_lock);
1215 }
1216
1217 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
1218 {
1219         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1220         struct sock *sk = transport->inet;
1221
1222         if (transport->rcvsize) {
1223                 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1224                 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
1225         }
1226         if (transport->sndsize) {
1227                 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1228                 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1229                 sk->sk_write_space(sk);
1230         }
1231 }
1232
1233 /**
1234  * xs_udp_set_buffer_size - set send and receive limits
1235  * @xprt: generic transport
1236  * @sndsize: requested size of send buffer, in bytes
1237  * @rcvsize: requested size of receive buffer, in bytes
1238  *
1239  * Set socket send and receive buffer size limits.
1240  */
1241 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1242 {
1243         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1244
1245         transport->sndsize = 0;
1246         if (sndsize)
1247                 transport->sndsize = sndsize + 1024;
1248         transport->rcvsize = 0;
1249         if (rcvsize)
1250                 transport->rcvsize = rcvsize + 1024;
1251
1252         xs_udp_do_set_buffer_size(xprt);
1253 }
1254
1255 /**
1256  * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1257  * @task: task that timed out
1258  *
1259  * Adjust the congestion window after a retransmit timeout has occurred.
1260  */
1261 static void xs_udp_timer(struct rpc_task *task)
1262 {
1263         xprt_adjust_cwnd(task, -ETIMEDOUT);
1264 }
1265
1266 static unsigned short xs_get_random_port(void)
1267 {
1268         unsigned short range = xprt_max_resvport - xprt_min_resvport;
1269         unsigned short rand = (unsigned short) net_random() % range;
1270         return rand + xprt_min_resvport;
1271 }
1272
1273 /**
1274  * xs_set_port - reset the port number in the remote endpoint address
1275  * @xprt: generic transport
1276  * @port: new port number
1277  *
1278  */
1279 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1280 {
1281         struct sockaddr *addr = xs_addr(xprt);
1282
1283         dprintk("RPC:       setting port for xprt %p to %u\n", xprt, port);
1284
1285         switch (addr->sa_family) {
1286         case AF_INET:
1287                 ((struct sockaddr_in *)addr)->sin_port = htons(port);
1288                 break;
1289         case AF_INET6:
1290                 ((struct sockaddr_in6 *)addr)->sin6_port = htons(port);
1291                 break;
1292         default:
1293                 BUG();
1294         }
1295 }
1296
1297 static unsigned short xs_get_srcport(struct sock_xprt *transport, struct socket *sock)
1298 {
1299         unsigned short port = transport->port;
1300
1301         if (port == 0 && transport->xprt.resvport)
1302                 port = xs_get_random_port();
1303         return port;
1304 }
1305
1306 static unsigned short xs_next_srcport(struct sock_xprt *transport, struct socket *sock, unsigned short port)
1307 {
1308         if (transport->port != 0)
1309                 transport->port = 0;
1310         if (!transport->xprt.resvport)
1311                 return 0;
1312         if (port <= xprt_min_resvport || port > xprt_max_resvport)
1313                 return xprt_max_resvport;
1314         return --port;
1315 }
1316
1317 static int xs_bind4(struct sock_xprt *transport, struct socket *sock)
1318 {
1319         struct sockaddr_in myaddr = {
1320                 .sin_family = AF_INET,
1321         };
1322         struct sockaddr_in *sa;
1323         int err, nloop = 0;
1324         unsigned short port = xs_get_srcport(transport, sock);
1325         unsigned short last;
1326
1327         sa = (struct sockaddr_in *)&transport->addr;
1328         myaddr.sin_addr = sa->sin_addr;
1329         do {
1330                 myaddr.sin_port = htons(port);
1331                 err = kernel_bind(sock, (struct sockaddr *) &myaddr,
1332                                                 sizeof(myaddr));
1333                 if (port == 0)
1334                         break;
1335                 if (err == 0) {
1336                         transport->port = port;
1337                         break;
1338                 }
1339                 last = port;
1340                 port = xs_next_srcport(transport, sock, port);
1341                 if (port > last)
1342                         nloop++;
1343         } while (err == -EADDRINUSE && nloop != 2);
1344         dprintk("RPC:       %s "NIPQUAD_FMT":%u: %s (%d)\n",
1345                         __FUNCTION__, NIPQUAD(myaddr.sin_addr),
1346                         port, err ? "failed" : "ok", err);
1347         return err;
1348 }
1349
1350 static int xs_bind6(struct sock_xprt *transport, struct socket *sock)
1351 {
1352         struct sockaddr_in6 myaddr = {
1353                 .sin6_family = AF_INET6,
1354         };
1355         struct sockaddr_in6 *sa;
1356         int err, nloop = 0;
1357         unsigned short port = xs_get_srcport(transport, sock);
1358         unsigned short last;
1359
1360         sa = (struct sockaddr_in6 *)&transport->addr;
1361         myaddr.sin6_addr = sa->sin6_addr;
1362         do {
1363                 myaddr.sin6_port = htons(port);
1364                 err = kernel_bind(sock, (struct sockaddr *) &myaddr,
1365                                                 sizeof(myaddr));
1366                 if (port == 0)
1367                         break;
1368                 if (err == 0) {
1369                         transport->port = port;
1370                         break;
1371                 }
1372                 last = port;
1373                 port = xs_next_srcport(transport, sock, port);
1374                 if (port > last)
1375                         nloop++;
1376         } while (err == -EADDRINUSE && nloop != 2);
1377         dprintk("RPC:       xs_bind6 "NIP6_FMT":%u: %s (%d)\n",
1378                 NIP6(myaddr.sin6_addr), port, err ? "failed" : "ok", err);
1379         return err;
1380 }
1381
1382 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1383 static struct lock_class_key xs_key[2];
1384 static struct lock_class_key xs_slock_key[2];
1385
1386 static inline void xs_reclassify_socket4(struct socket *sock)
1387 {
1388         struct sock *sk = sock->sk;
1389
1390         BUG_ON(sock_owned_by_user(sk));
1391         sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1392                 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1393 }
1394
1395 static inline void xs_reclassify_socket6(struct socket *sock)
1396 {
1397         struct sock *sk = sock->sk;
1398
1399         BUG_ON(sock_owned_by_user(sk));
1400         sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1401                 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
1402 }
1403 #else
1404 static inline void xs_reclassify_socket4(struct socket *sock)
1405 {
1406 }
1407
1408 static inline void xs_reclassify_socket6(struct socket *sock)
1409 {
1410 }
1411 #endif
1412
1413 static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
1414 {
1415         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1416
1417         if (!transport->inet) {
1418                 struct sock *sk = sock->sk;
1419
1420                 write_lock_bh(&sk->sk_callback_lock);
1421
1422                 sk->sk_user_data = xprt;
1423                 transport->old_data_ready = sk->sk_data_ready;
1424                 transport->old_state_change = sk->sk_state_change;
1425                 transport->old_write_space = sk->sk_write_space;
1426                 sk->sk_data_ready = xs_udp_data_ready;
1427                 sk->sk_write_space = xs_udp_write_space;
1428                 sk->sk_no_check = UDP_CSUM_NORCV;
1429                 sk->sk_allocation = GFP_ATOMIC;
1430
1431                 xprt_set_connected(xprt);
1432
1433                 /* Reset to new socket */
1434                 transport->sock = sock;
1435                 transport->inet = sk;
1436
1437                 write_unlock_bh(&sk->sk_callback_lock);
1438         }
1439         xs_udp_do_set_buffer_size(xprt);
1440 }
1441
1442 /**
1443  * xs_udp_connect_worker4 - set up a UDP socket
1444  * @work: RPC transport to connect
1445  *
1446  * Invoked by a work queue tasklet.
1447  */
1448 static void xs_udp_connect_worker4(struct work_struct *work)
1449 {
1450         struct sock_xprt *transport =
1451                 container_of(work, struct sock_xprt, connect_worker.work);
1452         struct rpc_xprt *xprt = &transport->xprt;
1453         struct socket *sock = transport->sock;
1454         int err, status = -EIO;
1455
1456         if (xprt->shutdown || !xprt_bound(xprt))
1457                 goto out;
1458
1459         /* Start by resetting any existing state */
1460         xs_close(xprt);
1461
1462         if ((err = sock_create_kern(PF_INET, SOCK_DGRAM, IPPROTO_UDP, &sock)) < 0) {
1463                 dprintk("RPC:       can't create UDP transport socket (%d).\n", -err);
1464                 goto out;
1465         }
1466         xs_reclassify_socket4(sock);
1467
1468         if (xs_bind4(transport, sock)) {
1469                 sock_release(sock);
1470                 goto out;
1471         }
1472
1473         dprintk("RPC:       worker connecting xprt %p to address: %s\n",
1474                         xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1475
1476         xs_udp_finish_connecting(xprt, sock);
1477         status = 0;
1478 out:
1479         xprt_wake_pending_tasks(xprt, status);
1480         xprt_clear_connecting(xprt);
1481 }
1482
1483 /**
1484  * xs_udp_connect_worker6 - set up a UDP socket
1485  * @work: RPC transport to connect
1486  *
1487  * Invoked by a work queue tasklet.
1488  */
1489 static void xs_udp_connect_worker6(struct work_struct *work)
1490 {
1491         struct sock_xprt *transport =
1492                 container_of(work, struct sock_xprt, connect_worker.work);
1493         struct rpc_xprt *xprt = &transport->xprt;
1494         struct socket *sock = transport->sock;
1495         int err, status = -EIO;
1496
1497         if (xprt->shutdown || !xprt_bound(xprt))
1498                 goto out;
1499
1500         /* Start by resetting any existing state */
1501         xs_close(xprt);
1502
1503         if ((err = sock_create_kern(PF_INET6, SOCK_DGRAM, IPPROTO_UDP, &sock)) < 0) {
1504                 dprintk("RPC:       can't create UDP transport socket (%d).\n", -err);
1505                 goto out;
1506         }
1507         xs_reclassify_socket6(sock);
1508
1509         if (xs_bind6(transport, sock) < 0) {
1510                 sock_release(sock);
1511                 goto out;
1512         }
1513
1514         dprintk("RPC:       worker connecting xprt %p to address: %s\n",
1515                         xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1516
1517         xs_udp_finish_connecting(xprt, sock);
1518         status = 0;
1519 out:
1520         xprt_wake_pending_tasks(xprt, status);
1521         xprt_clear_connecting(xprt);
1522 }
1523
1524 /*
1525  * We need to preserve the port number so the reply cache on the server can
1526  * find our cached RPC replies when we get around to reconnecting.
1527  */
1528 static void xs_tcp_reuse_connection(struct rpc_xprt *xprt)
1529 {
1530         int result;
1531         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1532         struct sockaddr any;
1533
1534         dprintk("RPC:       disconnecting xprt %p to reuse port\n", xprt);
1535
1536         /*
1537          * Disconnect the transport socket by doing a connect operation
1538          * with AF_UNSPEC.  This should return immediately...
1539          */
1540         memset(&any, 0, sizeof(any));
1541         any.sa_family = AF_UNSPEC;
1542         result = kernel_connect(transport->sock, &any, sizeof(any), 0);
1543         if (result)
1544                 dprintk("RPC:       AF_UNSPEC connect return code %d\n",
1545                                 result);
1546 }
1547
1548 static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
1549 {
1550         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1551
1552         if (!transport->inet) {
1553                 struct sock *sk = sock->sk;
1554
1555                 write_lock_bh(&sk->sk_callback_lock);
1556
1557                 sk->sk_user_data = xprt;
1558                 transport->old_data_ready = sk->sk_data_ready;
1559                 transport->old_state_change = sk->sk_state_change;
1560                 transport->old_write_space = sk->sk_write_space;
1561                 sk->sk_data_ready = xs_tcp_data_ready;
1562                 sk->sk_state_change = xs_tcp_state_change;
1563                 sk->sk_write_space = xs_tcp_write_space;
1564                 sk->sk_allocation = GFP_ATOMIC;
1565
1566                 /* socket options */
1567                 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
1568                 sock_reset_flag(sk, SOCK_LINGER);
1569                 tcp_sk(sk)->linger2 = 0;
1570                 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
1571
1572                 xprt_clear_connected(xprt);
1573
1574                 /* Reset to new socket */
1575                 transport->sock = sock;
1576                 transport->inet = sk;
1577
1578                 write_unlock_bh(&sk->sk_callback_lock);
1579         }
1580
1581         /* Tell the socket layer to start connecting... */
1582         xprt->stat.connect_count++;
1583         xprt->stat.connect_start = jiffies;
1584         return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
1585 }
1586
1587 /**
1588  * xs_tcp_connect_worker4 - connect a TCP socket to a remote endpoint
1589  * @work: RPC transport to connect
1590  *
1591  * Invoked by a work queue tasklet.
1592  */
1593 static void xs_tcp_connect_worker4(struct work_struct *work)
1594 {
1595         struct sock_xprt *transport =
1596                 container_of(work, struct sock_xprt, connect_worker.work);
1597         struct rpc_xprt *xprt = &transport->xprt;
1598         struct socket *sock = transport->sock;
1599         int err, status = -EIO;
1600
1601         if (xprt->shutdown || !xprt_bound(xprt))
1602                 goto out;
1603
1604         if (!sock) {
1605                 /* start from scratch */
1606                 if ((err = sock_create_kern(PF_INET, SOCK_STREAM, IPPROTO_TCP, &sock)) < 0) {
1607                         dprintk("RPC:       can't create TCP transport socket (%d).\n", -err);
1608                         goto out;
1609                 }
1610                 xs_reclassify_socket4(sock);
1611
1612                 if (xs_bind4(transport, sock) < 0) {
1613                         sock_release(sock);
1614                         goto out;
1615                 }
1616         } else
1617                 /* "close" the socket, preserving the local port */
1618                 xs_tcp_reuse_connection(xprt);
1619
1620         dprintk("RPC:       worker connecting xprt %p to address: %s\n",
1621                         xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1622
1623         status = xs_tcp_finish_connecting(xprt, sock);
1624         dprintk("RPC:       %p connect status %d connected %d sock state %d\n",
1625                         xprt, -status, xprt_connected(xprt),
1626                         sock->sk->sk_state);
1627         if (status < 0) {
1628                 switch (status) {
1629                         case -EINPROGRESS:
1630                         case -EALREADY:
1631                                 goto out_clear;
1632                         case -ECONNREFUSED:
1633                         case -ECONNRESET:
1634                                 /* retry with existing socket, after a delay */
1635                                 break;
1636                         default:
1637                                 /* get rid of existing socket, and retry */
1638                                 xs_close(xprt);
1639                                 break;
1640                 }
1641         }
1642 out:
1643         xprt_wake_pending_tasks(xprt, status);
1644 out_clear:
1645         xprt_clear_connecting(xprt);
1646 }
1647
1648 /**
1649  * xs_tcp_connect_worker6 - connect a TCP socket to a remote endpoint
1650  * @work: RPC transport to connect
1651  *
1652  * Invoked by a work queue tasklet.
1653  */
1654 static void xs_tcp_connect_worker6(struct work_struct *work)
1655 {
1656         struct sock_xprt *transport =
1657                 container_of(work, struct sock_xprt, connect_worker.work);
1658         struct rpc_xprt *xprt = &transport->xprt;
1659         struct socket *sock = transport->sock;
1660         int err, status = -EIO;
1661
1662         if (xprt->shutdown || !xprt_bound(xprt))
1663                 goto out;
1664
1665         if (!sock) {
1666                 /* start from scratch */
1667                 if ((err = sock_create_kern(PF_INET6, SOCK_STREAM, IPPROTO_TCP, &sock)) < 0) {
1668                         dprintk("RPC:       can't create TCP transport socket (%d).\n", -err);
1669                         goto out;
1670                 }
1671                 xs_reclassify_socket6(sock);
1672
1673                 if (xs_bind6(transport, sock) < 0) {
1674                         sock_release(sock);
1675                         goto out;
1676                 }
1677         } else
1678                 /* "close" the socket, preserving the local port */
1679                 xs_tcp_reuse_connection(xprt);
1680
1681         dprintk("RPC:       worker connecting xprt %p to address: %s\n",
1682                         xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1683
1684         status = xs_tcp_finish_connecting(xprt, sock);
1685         dprintk("RPC:       %p connect status %d connected %d sock state %d\n",
1686                         xprt, -status, xprt_connected(xprt), sock->sk->sk_state);
1687         if (status < 0) {
1688                 switch (status) {
1689                         case -EINPROGRESS:
1690                         case -EALREADY:
1691                                 goto out_clear;
1692                         case -ECONNREFUSED:
1693                         case -ECONNRESET:
1694                                 /* retry with existing socket, after a delay */
1695                                 break;
1696                         default:
1697                                 /* get rid of existing socket, and retry */
1698                                 xs_close(xprt);
1699                                 break;
1700                 }
1701         }
1702 out:
1703         xprt_wake_pending_tasks(xprt, status);
1704 out_clear:
1705         xprt_clear_connecting(xprt);
1706 }
1707
1708 /**
1709  * xs_connect - connect a socket to a remote endpoint
1710  * @task: address of RPC task that manages state of connect request
1711  *
1712  * TCP: If the remote end dropped the connection, delay reconnecting.
1713  *
1714  * UDP socket connects are synchronous, but we use a work queue anyway
1715  * to guarantee that even unprivileged user processes can set up a
1716  * socket on a privileged port.
1717  *
1718  * If a UDP socket connect fails, the delay behavior here prevents
1719  * retry floods (hard mounts).
1720  */
1721 static void xs_connect(struct rpc_task *task)
1722 {
1723         struct rpc_xprt *xprt = task->tk_xprt;
1724         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1725
1726         if (xprt_test_and_set_connecting(xprt))
1727                 return;
1728
1729         if (transport->sock != NULL) {
1730                 dprintk("RPC:       xs_connect delayed xprt %p for %lu "
1731                                 "seconds\n",
1732                                 xprt, xprt->reestablish_timeout / HZ);
1733                 queue_delayed_work(rpciod_workqueue,
1734                                    &transport->connect_worker,
1735                                    xprt->reestablish_timeout);
1736                 xprt->reestablish_timeout <<= 1;
1737                 if (xprt->reestablish_timeout > XS_TCP_MAX_REEST_TO)
1738                         xprt->reestablish_timeout = XS_TCP_MAX_REEST_TO;
1739         } else {
1740                 dprintk("RPC:       xs_connect scheduled xprt %p\n", xprt);
1741                 queue_delayed_work(rpciod_workqueue,
1742                                    &transport->connect_worker, 0);
1743         }
1744 }
1745
1746 /**
1747  * xs_udp_print_stats - display UDP socket-specifc stats
1748  * @xprt: rpc_xprt struct containing statistics
1749  * @seq: output file
1750  *
1751  */
1752 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
1753 {
1754         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1755
1756         seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %Lu %Lu\n",
1757                         transport->port,
1758                         xprt->stat.bind_count,
1759                         xprt->stat.sends,
1760                         xprt->stat.recvs,
1761                         xprt->stat.bad_xids,
1762                         xprt->stat.req_u,
1763                         xprt->stat.bklog_u);
1764 }
1765
1766 /**
1767  * xs_tcp_print_stats - display TCP socket-specifc stats
1768  * @xprt: rpc_xprt struct containing statistics
1769  * @seq: output file
1770  *
1771  */
1772 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
1773 {
1774         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1775         long idle_time = 0;
1776
1777         if (xprt_connected(xprt))
1778                 idle_time = (long)(jiffies - xprt->last_used) / HZ;
1779
1780         seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu %Lu %Lu\n",
1781                         transport->port,
1782                         xprt->stat.bind_count,
1783                         xprt->stat.connect_count,
1784                         xprt->stat.connect_time,
1785                         idle_time,
1786                         xprt->stat.sends,
1787                         xprt->stat.recvs,
1788                         xprt->stat.bad_xids,
1789                         xprt->stat.req_u,
1790                         xprt->stat.bklog_u);
1791 }
1792
1793 static struct rpc_xprt_ops xs_udp_ops = {
1794         .set_buffer_size        = xs_udp_set_buffer_size,
1795         .reserve_xprt           = xprt_reserve_xprt_cong,
1796         .release_xprt           = xprt_release_xprt_cong,
1797         .rpcbind                = rpcb_getport_async,
1798         .set_port               = xs_set_port,
1799         .connect                = xs_connect,
1800         .buf_alloc              = rpc_malloc,
1801         .buf_free               = rpc_free,
1802         .send_request           = xs_udp_send_request,
1803         .set_retrans_timeout    = xprt_set_retrans_timeout_rtt,
1804         .timer                  = xs_udp_timer,
1805         .release_request        = xprt_release_rqst_cong,
1806         .close                  = xs_close,
1807         .destroy                = xs_destroy,
1808         .print_stats            = xs_udp_print_stats,
1809 };
1810
1811 static struct rpc_xprt_ops xs_tcp_ops = {
1812         .reserve_xprt           = xprt_reserve_xprt,
1813         .release_xprt           = xs_tcp_release_xprt,
1814         .rpcbind                = rpcb_getport_async,
1815         .set_port               = xs_set_port,
1816         .connect                = xs_connect,
1817         .buf_alloc              = rpc_malloc,
1818         .buf_free               = rpc_free,
1819         .send_request           = xs_tcp_send_request,
1820         .set_retrans_timeout    = xprt_set_retrans_timeout_def,
1821         .close                  = xs_close,
1822         .destroy                = xs_destroy,
1823         .print_stats            = xs_tcp_print_stats,
1824 };
1825
1826 static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
1827                                       unsigned int slot_table_size)
1828 {
1829         struct rpc_xprt *xprt;
1830         struct sock_xprt *new;
1831
1832         if (args->addrlen > sizeof(xprt->addr)) {
1833                 dprintk("RPC:       xs_setup_xprt: address too large\n");
1834                 return ERR_PTR(-EBADF);
1835         }
1836
1837         new = kzalloc(sizeof(*new), GFP_KERNEL);
1838         if (new == NULL) {
1839                 dprintk("RPC:       xs_setup_xprt: couldn't allocate "
1840                                 "rpc_xprt\n");
1841                 return ERR_PTR(-ENOMEM);
1842         }
1843         xprt = &new->xprt;
1844
1845         xprt->max_reqs = slot_table_size;
1846         xprt->slot = kcalloc(xprt->max_reqs, sizeof(struct rpc_rqst), GFP_KERNEL);
1847         if (xprt->slot == NULL) {
1848                 kfree(xprt);
1849                 dprintk("RPC:       xs_setup_xprt: couldn't allocate slot "
1850                                 "table\n");
1851                 return ERR_PTR(-ENOMEM);
1852         }
1853
1854         memcpy(&xprt->addr, args->dstaddr, args->addrlen);
1855         xprt->addrlen = args->addrlen;
1856         if (args->srcaddr)
1857                 memcpy(&new->addr, args->srcaddr, args->addrlen);
1858
1859         return xprt;
1860 }
1861
1862 /**
1863  * xs_setup_udp - Set up transport to use a UDP socket
1864  * @args: rpc transport creation arguments
1865  *
1866  */
1867 static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
1868 {
1869         struct sockaddr *addr = args->dstaddr;
1870         struct rpc_xprt *xprt;
1871         struct sock_xprt *transport;
1872
1873         xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries);
1874         if (IS_ERR(xprt))
1875                 return xprt;
1876         transport = container_of(xprt, struct sock_xprt, xprt);
1877
1878         xprt->prot = IPPROTO_UDP;
1879         xprt->tsh_size = 0;
1880         /* XXX: header size can vary due to auth type, IPv6, etc. */
1881         xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
1882
1883         xprt->bind_timeout = XS_BIND_TO;
1884         xprt->connect_timeout = XS_UDP_CONN_TO;
1885         xprt->reestablish_timeout = XS_UDP_REEST_TO;
1886         xprt->idle_timeout = XS_IDLE_DISC_TO;
1887
1888         xprt->ops = &xs_udp_ops;
1889
1890         if (args->timeout)
1891                 xprt->timeout = *args->timeout;
1892         else
1893                 xprt_set_timeout(&xprt->timeout, 5, 5 * HZ);
1894
1895         switch (addr->sa_family) {
1896         case AF_INET:
1897                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
1898                         xprt_set_bound(xprt);
1899
1900                 INIT_DELAYED_WORK(&transport->connect_worker,
1901                                         xs_udp_connect_worker4);
1902                 xs_format_ipv4_peer_addresses(xprt);
1903                 break;
1904         case AF_INET6:
1905                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
1906                         xprt_set_bound(xprt);
1907
1908                 INIT_DELAYED_WORK(&transport->connect_worker,
1909                                         xs_udp_connect_worker6);
1910                 xs_format_ipv6_peer_addresses(xprt);
1911                 break;
1912         default:
1913                 kfree(xprt);
1914                 return ERR_PTR(-EAFNOSUPPORT);
1915         }
1916
1917         dprintk("RPC:       set up transport to address %s\n",
1918                         xprt->address_strings[RPC_DISPLAY_ALL]);
1919
1920         if (try_module_get(THIS_MODULE))
1921                 return xprt;
1922
1923         kfree(xprt->slot);
1924         kfree(xprt);
1925         return ERR_PTR(-EINVAL);
1926 }
1927
1928 /**
1929  * xs_setup_tcp - Set up transport to use a TCP socket
1930  * @args: rpc transport creation arguments
1931  *
1932  */
1933 static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
1934 {
1935         struct sockaddr *addr = args->dstaddr;
1936         struct rpc_xprt *xprt;
1937         struct sock_xprt *transport;
1938
1939         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries);
1940         if (IS_ERR(xprt))
1941                 return xprt;
1942         transport = container_of(xprt, struct sock_xprt, xprt);
1943
1944         xprt->prot = IPPROTO_TCP;
1945         xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
1946         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
1947
1948         xprt->bind_timeout = XS_BIND_TO;
1949         xprt->connect_timeout = XS_TCP_CONN_TO;
1950         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1951         xprt->idle_timeout = XS_IDLE_DISC_TO;
1952
1953         xprt->ops = &xs_tcp_ops;
1954
1955         if (args->timeout)
1956                 xprt->timeout = *args->timeout;
1957         else
1958                 xprt_set_timeout(&xprt->timeout, 2, 60 * HZ);
1959
1960         switch (addr->sa_family) {
1961         case AF_INET:
1962                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
1963                         xprt_set_bound(xprt);
1964
1965                 INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_connect_worker4);
1966                 xs_format_ipv4_peer_addresses(xprt);
1967                 break;
1968         case AF_INET6:
1969                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
1970                         xprt_set_bound(xprt);
1971
1972                 INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_connect_worker6);
1973                 xs_format_ipv6_peer_addresses(xprt);
1974                 break;
1975         default:
1976                 kfree(xprt);
1977                 return ERR_PTR(-EAFNOSUPPORT);
1978         }
1979
1980         dprintk("RPC:       set up transport to address %s\n",
1981                         xprt->address_strings[RPC_DISPLAY_ALL]);
1982
1983         if (try_module_get(THIS_MODULE))
1984                 return xprt;
1985
1986         kfree(xprt->slot);
1987         kfree(xprt);
1988         return ERR_PTR(-EINVAL);
1989 }
1990
1991 static struct xprt_class        xs_udp_transport = {
1992         .list           = LIST_HEAD_INIT(xs_udp_transport.list),
1993         .name           = "udp",
1994         .owner          = THIS_MODULE,
1995         .ident          = IPPROTO_UDP,
1996         .setup          = xs_setup_udp,
1997 };
1998
1999 static struct xprt_class        xs_tcp_transport = {
2000         .list           = LIST_HEAD_INIT(xs_tcp_transport.list),
2001         .name           = "tcp",
2002         .owner          = THIS_MODULE,
2003         .ident          = IPPROTO_TCP,
2004         .setup          = xs_setup_tcp,
2005 };
2006
2007 /**
2008  * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
2009  *
2010  */
2011 int init_socket_xprt(void)
2012 {
2013 #ifdef RPC_DEBUG
2014         if (!sunrpc_table_header)
2015                 sunrpc_table_header = register_sysctl_table(sunrpc_table);
2016 #endif
2017
2018         xprt_register_transport(&xs_udp_transport);
2019         xprt_register_transport(&xs_tcp_transport);
2020
2021         return 0;
2022 }
2023
2024 /**
2025  * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
2026  *
2027  */
2028 void cleanup_socket_xprt(void)
2029 {
2030 #ifdef RPC_DEBUG
2031         if (sunrpc_table_header) {
2032                 unregister_sysctl_table(sunrpc_table_header);
2033                 sunrpc_table_header = NULL;
2034         }
2035 #endif
2036
2037         xprt_unregister_transport(&xs_udp_transport);
2038         xprt_unregister_transport(&xs_tcp_transport);
2039 }