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  * Time out for an RPC UDP socket connect.  UDP socket connects are
 140  * synchronous, but we set a timeout anyway in case of resource
 141  * exhaustion on the local host.
 142  */
 143 #define XS_UDP_CONN_TO          (5U * HZ)
 144
 145 /*
 146  * Wait duration for an RPC TCP connection to be established.  Solaris
 147  * NFS over TCP uses 60 seconds, for example, which is in line with how
 148  * long a server takes to reboot.
 149  */
 150 #define XS_TCP_CONN_TO          (60U * HZ)
 151
 152 /*
 153  * Wait duration for a reply from the RPC portmapper.
 154  */
 155 #define XS_BIND_TO              (60U * HZ)
 156
 157 /*
 158  * Delay if a UDP socket connect error occurs.  This is most likely some
 159  * kind of resource problem on the local host.
 160  */
 161 #define XS_UDP_REEST_TO         (2U * HZ)
 162
 163 /*
 164  * The reestablish timeout allows clients to delay for a bit before attempting
 165  * to reconnect to a server that just dropped our connection.
 166  *
 167  * We implement an exponential backoff when trying to reestablish a TCP
 168  * transport connection with the server.  Some servers like to drop a TCP
 169  * connection when they are overworked, so we start with a short timeout and
 170  * increase over time if the server is down or not responding.
 171  */
 172 #define XS_TCP_INIT_REEST_TO    (3U * HZ)
 173 #define XS_TCP_MAX_REEST_TO     (5U * 60 * HZ)
 174
 175 /*
 176  * TCP idle timeout; client drops the transport socket if it is idle
 177  * for this long.  Note that we also timeout UDP sockets to prevent
 178  * holding port numbers when there is no RPC traffic.
 179  */
 180 #define XS_IDLE_DISC_TO         (5U * 60 * HZ)
 181
 182 #ifdef RPC_DEBUG
 183 # undef  RPC_DEBUG_DATA
 184 # define RPCDBG_FACILITY        RPCDBG_TRANS
 185 #endif
 186
 187 #ifdef RPC_DEBUG_DATA
 188 static void xs_pktdump(char *msg, u32 *packet, unsigned int count)
 189 {
 190         u8 *buf = (u8 *) packet;
 191         int j;
 192
 193         dprintk("RPC:       %s\n", msg);
 194         for (j = 0; j < count && j < 128; j += 4) {
 195                 if (!(j & 31)) {
 196                         if (j)
 197                                 dprintk("\n");
 198                         dprintk("0x%04x ", j);
 199                 }
 200                 dprintk("%02x%02x%02x%02x ",
 201                         buf[j], buf[j+1], buf[j+2], buf[j+3]);
 202         }
 203         dprintk("\n");
 204 }
 205 #else
 206 static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count)
 207 {
 208         /* NOP */
 209 }
 210 #endif
 211
 212 struct sock_xprt {
 213         struct rpc_xprt         xprt;
 214
 215         /*
 216          * Network layer
 217          */
 218         struct socket *         sock;
 219         struct sock *           inet;
 220
 221         /*
 222          * State of TCP reply receive
 223          */
 224         __be32                  tcp_fraghdr,
 225                                 tcp_xid;
 226
 227         u32                     tcp_offset,
 228                                 tcp_reclen;
 229
 230         unsigned long           tcp_copied,
 231                                 tcp_flags;
 232
 233         /*
 234          * Connection of transports
 235          */
 236         struct delayed_work     connect_worker;
 237         struct sockaddr_storage addr;
 238         unsigned short          port;
 239
 240         /*
 241          * UDP socket buffer size parameters
 242          */
 243         size_t                  rcvsize,
 244                                 sndsize;
 245
 246         /*
 247          * Saved socket callback addresses
 248          */
 249         void                    (*old_data_ready)(struct sock *, int);
 250         void                    (*old_state_change)(struct sock *);
 251         void                    (*old_write_space)(struct sock *);
 252 };
 253
 254 /*
 255  * TCP receive state flags
 256  */
 257 #define TCP_RCV_LAST_FRAG       (1UL << 0)
 258 #define TCP_RCV_COPY_FRAGHDR    (1UL << 1)
 259 #define TCP_RCV_COPY_XID        (1UL << 2)
 260 #define TCP_RCV_COPY_DATA       (1UL << 3)
 261
 262 static inline struct sockaddr *xs_addr(struct rpc_xprt *xprt)
 263 {
 264         return (struct sockaddr *) &xprt->addr;
 265 }
 266
 267 static inline struct sockaddr_in *xs_addr_in(struct rpc_xprt *xprt)
 268 {
 269         return (struct sockaddr_in *) &xprt->addr;
 270 }
 271
 272 static inline struct sockaddr_in6 *xs_addr_in6(struct rpc_xprt *xprt)
 273 {
 274         return (struct sockaddr_in6 *) &xprt->addr;
 275 }
 276
 277 static void xs_format_ipv4_peer_addresses(struct rpc_xprt *xprt,
 278                                           const char *protocol,
 279                                           const char *netid)
 280 {
 281         struct sockaddr_in *addr = xs_addr_in(xprt);
 282         char *buf;
 283
 284         buf = kzalloc(20, GFP_KERNEL);
 285         if (buf) {
 286                 snprintf(buf, 20, NIPQUAD_FMT,
 287                                 NIPQUAD(addr->sin_addr.s_addr));
 288         }
 289         xprt->address_strings[RPC_DISPLAY_ADDR] = buf;
 290
 291         buf = kzalloc(8, GFP_KERNEL);
 292         if (buf) {
 293                 snprintf(buf, 8, "%u",
 294                                 ntohs(addr->sin_port));
 295         }
 296         xprt->address_strings[RPC_DISPLAY_PORT] = buf;
 297
 298         xprt->address_strings[RPC_DISPLAY_PROTO] = protocol;
 299
 300         buf = kzalloc(48, GFP_KERNEL);
 301         if (buf) {
 302                 snprintf(buf, 48, "addr="NIPQUAD_FMT" port=%u proto=%s",
 303                         NIPQUAD(addr->sin_addr.s_addr),
 304                         ntohs(addr->sin_port),
 305                         protocol);
 306         }
 307         xprt->address_strings[RPC_DISPLAY_ALL] = buf;
 308
 309         buf = kzalloc(10, GFP_KERNEL);
 310         if (buf) {
 311                 snprintf(buf, 10, "%02x%02x%02x%02x",
 312                                 NIPQUAD(addr->sin_addr.s_addr));
 313         }
 314         xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = buf;
 315
 316         buf = kzalloc(8, GFP_KERNEL);
 317         if (buf) {
 318                 snprintf(buf, 8, "%4hx",
 319                                 ntohs(addr->sin_port));
 320         }
 321         xprt->address_strings[RPC_DISPLAY_HEX_PORT] = buf;
 322
 323         buf = kzalloc(30, GFP_KERNEL);
 324         if (buf) {
 325                 snprintf(buf, 30, NIPQUAD_FMT".%u.%u",
 326                                 NIPQUAD(addr->sin_addr.s_addr),
 327                                 ntohs(addr->sin_port) >> 8,
 328                                 ntohs(addr->sin_port) & 0xff);
 329         }
 330         xprt->address_strings[RPC_DISPLAY_UNIVERSAL_ADDR] = buf;
 331
 332         xprt->address_strings[RPC_DISPLAY_NETID] = netid;
 333 }
 334
 335 static void xs_format_ipv6_peer_addresses(struct rpc_xprt *xprt,
 336                                           const char *protocol,
 337                                           const char *netid)
 338 {
 339         struct sockaddr_in6 *addr = xs_addr_in6(xprt);
 340         char *buf;
 341
 342         buf = kzalloc(40, GFP_KERNEL);
 343         if (buf) {
 344                 snprintf(buf, 40, NIP6_FMT,
 345                                 NIP6(addr->sin6_addr));
 346         }
 347         xprt->address_strings[RPC_DISPLAY_ADDR] = buf;
 348
 349         buf = kzalloc(8, GFP_KERNEL);
 350         if (buf) {
 351                 snprintf(buf, 8, "%u",
 352                                 ntohs(addr->sin6_port));
 353         }
 354         xprt->address_strings[RPC_DISPLAY_PORT] = buf;
 355
 356         xprt->address_strings[RPC_DISPLAY_PROTO] = protocol;
 357
 358         buf = kzalloc(64, GFP_KERNEL);
 359         if (buf) {
 360                 snprintf(buf, 64, "addr="NIP6_FMT" port=%u proto=%s",
 361                                 NIP6(addr->sin6_addr),
 362                                 ntohs(addr->sin6_port),
 363                                 protocol);
 364         }
 365         xprt->address_strings[RPC_DISPLAY_ALL] = buf;
 366
 367         buf = kzalloc(36, GFP_KERNEL);
 368         if (buf) {
 369                 snprintf(buf, 36, NIP6_SEQFMT,
 370                                 NIP6(addr->sin6_addr));
 371         }
 372         xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = buf;
 373
 374         buf = kzalloc(8, GFP_KERNEL);
 375         if (buf) {
 376                 snprintf(buf, 8, "%4hx",
 377                                 ntohs(addr->sin6_port));
 378         }
 379         xprt->address_strings[RPC_DISPLAY_HEX_PORT] = buf;
 380
 381         buf = kzalloc(50, GFP_KERNEL);
 382         if (buf) {
 383                 snprintf(buf, 50, NIP6_FMT".%u.%u",
 384                                 NIP6(addr->sin6_addr),
 385                                 ntohs(addr->sin6_port) >> 8,
 386                                 ntohs(addr->sin6_port) & 0xff);
 387         }
 388         xprt->address_strings[RPC_DISPLAY_UNIVERSAL_ADDR] = buf;
 389
 390         xprt->address_strings[RPC_DISPLAY_NETID] = netid;
 391 }
 392
 393 static void xs_free_peer_addresses(struct rpc_xprt *xprt)
 394 {
 395         unsigned int i;
 396
 397         for (i = 0; i < RPC_DISPLAY_MAX; i++)
 398                 switch (i) {
 399                 case RPC_DISPLAY_PROTO:
 400                 case RPC_DISPLAY_NETID:
 401                         continue;
 402                 default:
 403                         kfree(xprt->address_strings[i]);
 404                 }
 405 }
 406
 407 #define XS_SENDMSG_FLAGS        (MSG_DONTWAIT | MSG_NOSIGNAL)
 408
 409 static int xs_send_kvec(struct socket *sock, struct sockaddr *addr, int addrlen, struct kvec *vec, unsigned int base, int more)
 410 {
 411         struct msghdr msg = {
 412                 .msg_name       = addr,
 413                 .msg_namelen    = addrlen,
 414                 .msg_flags      = XS_SENDMSG_FLAGS | (more ? MSG_MORE : 0),
 415         };
 416         struct kvec iov = {
 417                 .iov_base       = vec->iov_base + base,
 418                 .iov_len        = vec->iov_len - base,
 419         };
 420
 421         if (iov.iov_len != 0)
 422                 return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
 423         return kernel_sendmsg(sock, &msg, NULL, 0, 0);
 424 }
 425
 426 static int xs_send_pagedata(struct socket *sock, struct xdr_buf *xdr, unsigned int base, int more)
 427 {
 428         struct page **ppage;
 429         unsigned int remainder;
 430         int err, sent = 0;
 431
 432         remainder = xdr->page_len - base;
 433         base += xdr->page_base;
 434         ppage = xdr->pages + (base >> PAGE_SHIFT);
 435         base &= ~PAGE_MASK;
 436         for(;;) {
 437                 unsigned int len = min_t(unsigned int, PAGE_SIZE - base, remainder);
 438                 int flags = XS_SENDMSG_FLAGS;
 439
 440                 remainder -= len;
 441                 if (remainder != 0 || more)
 442                         flags |= MSG_MORE;
 443                 err = sock->ops->sendpage(sock, *ppage, base, len, flags);
 444                 if (remainder == 0 || err != len)
 445                         break;
 446                 sent += err;
 447                 ppage++;
 448                 base = 0;
 449         }
 450         if (sent == 0)
 451                 return err;
 452         if (err > 0)
 453                 sent += err;
 454         return sent;
 455 }
 456
 457 /**
 458  * xs_sendpages - write pages directly to a socket
 459  * @sock: socket to send on
 460  * @addr: UDP only -- address of destination
 461  * @addrlen: UDP only -- length of destination address
 462  * @xdr: buffer containing this request
 463  * @base: starting position in the buffer
 464  *
 465  */
 466 static int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base)
 467 {
 468         unsigned int remainder = xdr->len - base;
 469         int err, sent = 0;
 470
 471         if (unlikely(!sock))
 472                 return -ENOTCONN;
 473
 474         clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags);
 475         if (base != 0) {
 476                 addr = NULL;
 477                 addrlen = 0;
 478         }
 479
 480         if (base < xdr->head[0].iov_len || addr != NULL) {
 481                 unsigned int len = xdr->head[0].iov_len - base;
 482                 remainder -= len;
 483                 err = xs_send_kvec(sock, addr, addrlen, &xdr->head[0], base, remainder != 0);
 484                 if (remainder == 0 || err != len)
 485                         goto out;
 486                 sent += err;
 487                 base = 0;
 488         } else
 489                 base -= xdr->head[0].iov_len;
 490
 491         if (base < xdr->page_len) {
 492                 unsigned int len = xdr->page_len - base;
 493                 remainder -= len;
 494                 err = xs_send_pagedata(sock, xdr, base, remainder != 0);
 495                 if (remainder == 0 || err != len)
 496                         goto out;
 497                 sent += err;
 498                 base = 0;
 499         } else
 500                 base -= xdr->page_len;
 501
 502         if (base >= xdr->tail[0].iov_len)
 503                 return sent;
 504         err = xs_send_kvec(sock, NULL, 0, &xdr->tail[0], base, 0);
 505 out:
 506         if (sent == 0)
 507                 return err;
 508         if (err > 0)
 509                 sent += err;
 510         return sent;
 511 }
 512
 513 static void xs_nospace_callback(struct rpc_task *task)
 514 {
 515         struct sock_xprt *transport = container_of(task->tk_rqstp->rq_xprt, struct sock_xprt, xprt);
 516
 517         transport->inet->sk_write_pending--;
 518         clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 519 }
 520
 521 /**
 522  * xs_nospace - place task on wait queue if transmit was incomplete
 523  * @task: task to put to sleep
 524  *
 525  */
 526 static void xs_nospace(struct rpc_task *task)
 527 {
 528         struct rpc_rqst *req = task->tk_rqstp;
 529         struct rpc_xprt *xprt = req->rq_xprt;
 530         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 531
 532         dprintk("RPC: %5u xmit incomplete (%u left of %u)\n",
 533                         task->tk_pid, req->rq_slen - req->rq_bytes_sent,
 534                         req->rq_slen);
 535
 536         /* Protect against races with write_space */
 537         spin_lock_bh(&xprt->transport_lock);
 538
 539         /* Don't race with disconnect */
 540         if (xprt_connected(xprt)) {
 541                 if (test_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags)) {
 542                         /*
 543                          * Notify TCP that we're limited by the application
 544                          * window size
 545                          */
 546                         set_bit(SOCK_NOSPACE, &transport->sock->flags);
 547                         transport->inet->sk_write_pending++;
 548                         /* ...and wait for more buffer space */
 549                         xprt_wait_for_buffer_space(task, xs_nospace_callback);
 550                 }
 551         } else {
 552                 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 553                 task->tk_status = -ENOTCONN;
 554         }
 555
 556         spin_unlock_bh(&xprt->transport_lock);
 557 }
 558
 559 /**
 560  * xs_udp_send_request - write an RPC request to a UDP socket
 561  * @task: address of RPC task that manages the state of an RPC request
 562  *
 563  * Return values:
 564  *        0:    The request has been sent
 565  *   EAGAIN:    The socket was blocked, please call again later to
 566  *              complete the request
 567  * ENOTCONN:    Caller needs to invoke connect logic then call again
 568  *    other:    Some other error occured, the request was not sent
 569  */
 570 static int xs_udp_send_request(struct rpc_task *task)
 571 {
 572         struct rpc_rqst *req = task->tk_rqstp;
 573         struct rpc_xprt *xprt = req->rq_xprt;
 574         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 575         struct xdr_buf *xdr = &req->rq_snd_buf;
 576         int status;
 577
 578         xs_pktdump("packet data:",
 579                                 req->rq_svec->iov_base,
 580                                 req->rq_svec->iov_len);
 581
 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 -EAGAIN:
 600                 xs_nospace(task);
 601                 break;
 602         case -ENETUNREACH:
 603         case -EPIPE:
 604         case -ECONNREFUSED:
 605                 /* When the server has died, an ICMP port unreachable message
 606                  * prompts ECONNREFUSED. */
 607                 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 608                 break;
 609         default:
 610                 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 611                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
 612                         -status);
 613         }
 614
 615         return status;
 616 }
 617
 618 /**
 619  * xs_tcp_shutdown - gracefully shut down a TCP socket
 620  * @xprt: transport
 621  *
 622  * Initiates a graceful shutdown of the TCP socket by calling the
 623  * equivalent of shutdown(SHUT_WR);
 624  */
 625 static void xs_tcp_shutdown(struct rpc_xprt *xprt)
 626 {
 627         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 628         struct socket *sock = transport->sock;
 629
 630         if (sock != NULL)
 631                 kernel_sock_shutdown(sock, SHUT_WR);
 632 }
 633
 634 static inline void xs_encode_tcp_record_marker(struct xdr_buf *buf)
 635 {
 636         u32 reclen = buf->len - sizeof(rpc_fraghdr);
 637         rpc_fraghdr *base = buf->head[0].iov_base;
 638         *base = htonl(RPC_LAST_STREAM_FRAGMENT | reclen);
 639 }
 640
 641 /**
 642  * xs_tcp_send_request - write an RPC request to a TCP socket
 643  * @task: address of RPC task that manages the state of an RPC request
 644  *
 645  * Return values:
 646  *        0:    The request has been sent
 647  *   EAGAIN:    The socket was blocked, please call again later to
 648  *              complete the request
 649  * ENOTCONN:    Caller needs to invoke connect logic then call again
 650  *    other:    Some other error occured, the request was not sent
 651  *
 652  * XXX: In the case of soft timeouts, should we eventually give up
 653  *      if sendmsg is not able to make progress?
 654  */
 655 static int xs_tcp_send_request(struct rpc_task *task)
 656 {
 657         struct rpc_rqst *req = task->tk_rqstp;
 658         struct rpc_xprt *xprt = req->rq_xprt;
 659         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 660         struct xdr_buf *xdr = &req->rq_snd_buf;
 661         int status;
 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                 status = xs_sendpages(transport->sock,
 674                                         NULL, 0, xdr, req->rq_bytes_sent);
 675
 676                 dprintk("RPC:       xs_tcp_send_request(%u) = %d\n",
 677                                 xdr->len - req->rq_bytes_sent, status);
 678
 679                 if (unlikely(status < 0))
 680                         break;
 681
 682                 /* If we've sent the entire packet, immediately
 683                  * reset the count of bytes sent. */
 684                 req->rq_bytes_sent += status;
 685                 task->tk_bytes_sent += status;
 686                 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
 687                         req->rq_bytes_sent = 0;
 688                         return 0;
 689                 }
 690
 691                 if (status != 0)
 692                         continue;
 693                 status = -EAGAIN;
 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                 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 707                 break;
 708         default:
 709                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
 710                         -status);
 711                 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 712                 xs_tcp_shutdown(xprt);
 713         }
 714
 715         return status;
 716 }
 717
 718 /**
 719  * xs_tcp_release_xprt - clean up after a tcp transmission
 720  * @xprt: transport
 721  * @task: rpc task
 722  *
 723  * This cleans up if an error causes us to abort the transmission of a request.
 724  * In this case, the socket may need to be reset in order to avoid confusing
 725  * the server.
 726  */
 727 static void xs_tcp_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
 728 {
 729         struct rpc_rqst *req;
 730
 731         if (task != xprt->snd_task)
 732                 return;
 733         if (task == NULL)
 734                 goto out_release;
 735         req = task->tk_rqstp;
 736         if (req->rq_bytes_sent == 0)
 737                 goto out_release;
 738         if (req->rq_bytes_sent == req->rq_snd_buf.len)
 739                 goto out_release;
 740         set_bit(XPRT_CLOSE_WAIT, &task->tk_xprt->state);
 741 out_release:
 742         xprt_release_xprt(xprt, task);
 743 }
 744
 745 /**
 746  * xs_close - close a socket
 747  * @xprt: transport
 748  *
 749  * This is used when all requests are complete; ie, no DRC state remains
 750  * on the server we want to save.
 751  */
 752 static void xs_close(struct rpc_xprt *xprt)
 753 {
 754         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 755         struct socket *sock = transport->sock;
 756         struct sock *sk = transport->inet;
 757
 758         if (!sk)
 759                 goto clear_close_wait;
 760
 761         dprintk("RPC:       xs_close xprt %p\n", xprt);
 762
 763         write_lock_bh(&sk->sk_callback_lock);
 764         transport->inet = NULL;
 765         transport->sock = NULL;
 766
 767         sk->sk_user_data = NULL;
 768         sk->sk_data_ready = transport->old_data_ready;
 769         sk->sk_state_change = transport->old_state_change;
 770         sk->sk_write_space = transport->old_write_space;
 771         write_unlock_bh(&sk->sk_callback_lock);
 772
 773         sk->sk_no_check = 0;
 774
 775         sock_release(sock);
 776 clear_close_wait:
 777         smp_mb__before_clear_bit();
 778         clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
 779         clear_bit(XPRT_CLOSING, &xprt->state);
 780         smp_mb__after_clear_bit();
 781         xprt_disconnect_done(xprt);
 782 }
 783
 784 /**
 785  * xs_destroy - prepare to shutdown a transport
 786  * @xprt: doomed transport
 787  *
 788  */
 789 static void xs_destroy(struct rpc_xprt *xprt)
 790 {
 791         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 792
 793         dprintk("RPC:       xs_destroy xprt %p\n", xprt);
 794
 795         cancel_rearming_delayed_work(&transport->connect_worker);
 796
 797         xs_close(xprt);
 798         xs_free_peer_addresses(xprt);
 799         kfree(xprt->slot);
 800         kfree(xprt);
 801         module_put(THIS_MODULE);
 802 }
 803
 804 static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
 805 {
 806         return (struct rpc_xprt *) sk->sk_user_data;
 807 }
 808
 809 /**
 810  * xs_udp_data_ready - "data ready" callback for UDP sockets
 811  * @sk: socket with data to read
 812  * @len: how much data to read
 813  *
 814  */
 815 static void xs_udp_data_ready(struct sock *sk, int len)
 816 {
 817         struct rpc_task *task;
 818         struct rpc_xprt *xprt;
 819         struct rpc_rqst *rovr;
 820         struct sk_buff *skb;
 821         int err, repsize, copied;
 822         u32 _xid;
 823         __be32 *xp;
 824
 825         read_lock(&sk->sk_callback_lock);
 826         dprintk("RPC:       xs_udp_data_ready...\n");
 827         if (!(xprt = xprt_from_sock(sk)))
 828                 goto out;
 829
 830         if ((skb = skb_recv_datagram(sk, 0, 1, &err)) == NULL)
 831                 goto out;
 832
 833         if (xprt->shutdown)
 834                 goto dropit;
 835
 836         repsize = skb->len - sizeof(struct udphdr);
 837         if (repsize < 4) {
 838                 dprintk("RPC:       impossible RPC reply size %d!\n", repsize);
 839                 goto dropit;
 840         }
 841
 842         /* Copy the XID from the skb... */
 843         xp = skb_header_pointer(skb, sizeof(struct udphdr),
 844                                 sizeof(_xid), &_xid);
 845         if (xp == NULL)
 846                 goto dropit;
 847
 848         /* Look up and lock the request corresponding to the given XID */
 849         spin_lock(&xprt->transport_lock);
 850         rovr = xprt_lookup_rqst(xprt, *xp);
 851         if (!rovr)
 852                 goto out_unlock;
 853         task = rovr->rq_task;
 854
 855         if ((copied = rovr->rq_private_buf.buflen) > repsize)
 856                 copied = repsize;
 857
 858         /* Suck it into the iovec, verify checksum if not done by hw. */
 859         if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
 860                 UDPX_INC_STATS_BH(sk, UDP_MIB_INERRORS);
 861                 goto out_unlock;
 862         }
 863
 864         UDPX_INC_STATS_BH(sk, UDP_MIB_INDATAGRAMS);
 865
 866         /* Something worked... */
 867         dst_confirm(skb->dst);
 868
 869         xprt_adjust_cwnd(task, copied);
 870         xprt_update_rtt(task);
 871         xprt_complete_rqst(task, copied);
 872
 873  out_unlock:
 874         spin_unlock(&xprt->transport_lock);
 875  dropit:
 876         skb_free_datagram(sk, skb);
 877  out:
 878         read_unlock(&sk->sk_callback_lock);
 879 }
 880
 881 static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
 882 {
 883         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 884         size_t len, used;
 885         char *p;
 886
 887         p = ((char *) &transport->tcp_fraghdr) + transport->tcp_offset;
 888         len = sizeof(transport->tcp_fraghdr) - transport->tcp_offset;
 889         used = xdr_skb_read_bits(desc, p, len);
 890         transport->tcp_offset += used;
 891         if (used != len)
 892                 return;
 893
 894         transport->tcp_reclen = ntohl(transport->tcp_fraghdr);
 895         if (transport->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
 896                 transport->tcp_flags |= TCP_RCV_LAST_FRAG;
 897         else
 898                 transport->tcp_flags &= ~TCP_RCV_LAST_FRAG;
 899         transport->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
 900
 901         transport->tcp_flags &= ~TCP_RCV_COPY_FRAGHDR;
 902         transport->tcp_offset = 0;
 903
 904         /* Sanity check of the record length */
 905         if (unlikely(transport->tcp_reclen < 4)) {
 906                 dprintk("RPC:       invalid TCP record fragment length\n");
 907                 xprt_force_disconnect(xprt);
 908                 return;
 909         }
 910         dprintk("RPC:       reading TCP record fragment of length %d\n",
 911                         transport->tcp_reclen);
 912 }
 913
 914 static void xs_tcp_check_fraghdr(struct sock_xprt *transport)
 915 {
 916         if (transport->tcp_offset == transport->tcp_reclen) {
 917                 transport->tcp_flags |= TCP_RCV_COPY_FRAGHDR;
 918                 transport->tcp_offset = 0;
 919                 if (transport->tcp_flags & TCP_RCV_LAST_FRAG) {
 920                         transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
 921                         transport->tcp_flags |= TCP_RCV_COPY_XID;
 922                         transport->tcp_copied = 0;
 923                 }
 924         }
 925 }
 926
 927 static inline void xs_tcp_read_xid(struct sock_xprt *transport, struct xdr_skb_reader *desc)
 928 {
 929         size_t len, used;
 930         char *p;
 931
 932         len = sizeof(transport->tcp_xid) - transport->tcp_offset;
 933         dprintk("RPC:       reading XID (%Zu bytes)\n", len);
 934         p = ((char *) &transport->tcp_xid) + transport->tcp_offset;
 935         used = xdr_skb_read_bits(desc, p, len);
 936         transport->tcp_offset += used;
 937         if (used != len)
 938                 return;
 939         transport->tcp_flags &= ~TCP_RCV_COPY_XID;
 940         transport->tcp_flags |= TCP_RCV_COPY_DATA;
 941         transport->tcp_copied = 4;
 942         dprintk("RPC:       reading reply for XID %08x\n",
 943                         ntohl(transport->tcp_xid));
 944         xs_tcp_check_fraghdr(transport);
 945 }
 946
 947 static inline void xs_tcp_read_request(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
 948 {
 949         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 950         struct rpc_rqst *req;
 951         struct xdr_buf *rcvbuf;
 952         size_t len;
 953         ssize_t r;
 954
 955         /* Find and lock the request corresponding to this xid */
 956         spin_lock(&xprt->transport_lock);
 957         req = xprt_lookup_rqst(xprt, transport->tcp_xid);
 958         if (!req) {
 959                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
 960                 dprintk("RPC:       XID %08x request not found!\n",
 961                                 ntohl(transport->tcp_xid));
 962                 spin_unlock(&xprt->transport_lock);
 963                 return;
 964         }
 965
 966         rcvbuf = &req->rq_private_buf;
 967         len = desc->count;
 968         if (len > transport->tcp_reclen - transport->tcp_offset) {
 969                 struct xdr_skb_reader my_desc;
 970
 971                 len = transport->tcp_reclen - transport->tcp_offset;
 972                 memcpy(&my_desc, desc, sizeof(my_desc));
 973                 my_desc.count = len;
 974                 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
 975                                           &my_desc, xdr_skb_read_bits);
 976                 desc->count -= r;
 977                 desc->offset += r;
 978         } else
 979                 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
 980                                           desc, xdr_skb_read_bits);
 981
 982         if (r > 0) {
 983                 transport->tcp_copied += r;
 984                 transport->tcp_offset += r;
 985         }
 986         if (r != len) {
 987                 /* Error when copying to the receive buffer,
 988                  * usually because we weren't able to allocate
 989                  * additional buffer pages. All we can do now
 990                  * is turn off TCP_RCV_COPY_DATA, so the request
 991                  * will not receive any additional updates,
 992                  * and time out.
 993                  * Any remaining data from this record will
 994                  * be discarded.
 995                  */
 996                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
 997                 dprintk("RPC:       XID %08x truncated request\n",
 998                                 ntohl(transport->tcp_xid));
 999                 dprintk("RPC:       xprt = %p, tcp_copied = %lu, "
1000                                 "tcp_offset = %u, tcp_reclen = %u\n",
1001                                 xprt, transport->tcp_copied,
1002                                 transport->tcp_offset, transport->tcp_reclen);
1003                 goto out;
1004         }
1005
1006         dprintk("RPC:       XID %08x read %Zd bytes\n",
1007                         ntohl(transport->tcp_xid), r);
1008         dprintk("RPC:       xprt = %p, tcp_copied = %lu, tcp_offset = %u, "
1009                         "tcp_reclen = %u\n", xprt, transport->tcp_copied,
1010                         transport->tcp_offset, transport->tcp_reclen);
1011
1012         if (transport->tcp_copied == req->rq_private_buf.buflen)
1013                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1014         else if (transport->tcp_offset == transport->tcp_reclen) {
1015                 if (transport->tcp_flags & TCP_RCV_LAST_FRAG)
1016                         transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1017         }
1018
1019 out:
1020         if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1021                 xprt_complete_rqst(req->rq_task, transport->tcp_copied);
1022         spin_unlock(&xprt->transport_lock);
1023         xs_tcp_check_fraghdr(transport);
1024 }
1025
1026 static inline void xs_tcp_read_discard(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1027 {
1028         size_t len;
1029
1030         len = transport->tcp_reclen - transport->tcp_offset;
1031         if (len > desc->count)
1032                 len = desc->count;
1033         desc->count -= len;
1034         desc->offset += len;
1035         transport->tcp_offset += len;
1036         dprintk("RPC:       discarded %Zu bytes\n", len);
1037         xs_tcp_check_fraghdr(transport);
1038 }
1039
1040 static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
1041 {
1042         struct rpc_xprt *xprt = rd_desc->arg.data;
1043         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1044         struct xdr_skb_reader desc = {
1045                 .skb    = skb,
1046                 .offset = offset,
1047                 .count  = len,
1048         };
1049
1050         dprintk("RPC:       xs_tcp_data_recv started\n");
1051         do {
1052                 /* Read in a new fragment marker if necessary */
1053                 /* Can we ever really expect to get completely empty fragments? */
1054                 if (transport->tcp_flags & TCP_RCV_COPY_FRAGHDR) {
1055                         xs_tcp_read_fraghdr(xprt, &desc);
1056                         continue;
1057                 }
1058                 /* Read in the xid if necessary */
1059                 if (transport->tcp_flags & TCP_RCV_COPY_XID) {
1060                         xs_tcp_read_xid(transport, &desc);
1061                         continue;
1062                 }
1063                 /* Read in the request data */
1064                 if (transport->tcp_flags & TCP_RCV_COPY_DATA) {
1065                         xs_tcp_read_request(xprt, &desc);
1066                         continue;
1067                 }
1068                 /* Skip over any trailing bytes on short reads */
1069                 xs_tcp_read_discard(transport, &desc);
1070         } while (desc.count);
1071         dprintk("RPC:       xs_tcp_data_recv done\n");
1072         return len - desc.count;
1073 }
1074
1075 /**
1076  * xs_tcp_data_ready - "data ready" callback for TCP sockets
1077  * @sk: socket with data to read
1078  * @bytes: how much data to read
1079  *
1080  */
1081 static void xs_tcp_data_ready(struct sock *sk, int bytes)
1082 {
1083         struct rpc_xprt *xprt;
1084         read_descriptor_t rd_desc;
1085         int read;
1086
1087         dprintk("RPC:       xs_tcp_data_ready...\n");
1088
1089         read_lock(&sk->sk_callback_lock);
1090         if (!(xprt = xprt_from_sock(sk)))
1091                 goto out;
1092         if (xprt->shutdown)
1093                 goto out;
1094
1095         /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
1096         rd_desc.arg.data = xprt;
1097         do {
1098                 rd_desc.count = 65536;
1099                 read = tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
1100         } while (read > 0);
1101 out:
1102         read_unlock(&sk->sk_callback_lock);
1103 }
1104
1105 /**
1106  * xs_tcp_state_change - callback to handle TCP socket state changes
1107  * @sk: socket whose state has changed
1108  *
1109  */
1110 static void xs_tcp_state_change(struct sock *sk)
1111 {
1112         struct rpc_xprt *xprt;
1113
1114         read_lock(&sk->sk_callback_lock);
1115         if (!(xprt = xprt_from_sock(sk)))
1116                 goto out;
1117         dprintk("RPC:       xs_tcp_state_change client %p...\n", xprt);
1118         dprintk("RPC:       state %x conn %d dead %d zapped %d\n",
1119                         sk->sk_state, xprt_connected(xprt),
1120                         sock_flag(sk, SOCK_DEAD),
1121                         sock_flag(sk, SOCK_ZAPPED));
1122
1123         switch (sk->sk_state) {
1124         case TCP_ESTABLISHED:
1125                 spin_lock_bh(&xprt->transport_lock);
1126                 if (!xprt_test_and_set_connected(xprt)) {
1127                         struct sock_xprt *transport = container_of(xprt,
1128                                         struct sock_xprt, xprt);
1129
1130                         /* Reset TCP record info */
1131                         transport->tcp_offset = 0;
1132                         transport->tcp_reclen = 0;
1133                         transport->tcp_copied = 0;
1134                         transport->tcp_flags =
1135                                 TCP_RCV_COPY_FRAGHDR | TCP_RCV_COPY_XID;
1136
1137                         xprt_wake_pending_tasks(xprt, 0);
1138                 }
1139                 spin_unlock_bh(&xprt->transport_lock);
1140                 break;
1141         case TCP_FIN_WAIT1:
1142                 /* The client initiated a shutdown of the socket */
1143                 xprt->connect_cookie++;
1144                 xprt->reestablish_timeout = 0;
1145                 set_bit(XPRT_CLOSING, &xprt->state);
1146                 smp_mb__before_clear_bit();
1147                 clear_bit(XPRT_CONNECTED, &xprt->state);
1148                 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1149                 smp_mb__after_clear_bit();
1150                 break;
1151         case TCP_CLOSE_WAIT:
1152                 /* The server initiated a shutdown of the socket */
1153                 set_bit(XPRT_CLOSING, &xprt->state);
1154                 xprt_force_disconnect(xprt);
1155         case TCP_SYN_SENT:
1156                 xprt->connect_cookie++;
1157         case TCP_CLOSING:
1158                 /*
1159                  * If the server closed down the connection, make sure that
1160                  * we back off before reconnecting
1161                  */
1162                 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1163                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1164                 break;
1165         case TCP_LAST_ACK:
1166                 smp_mb__before_clear_bit();
1167                 clear_bit(XPRT_CONNECTED, &xprt->state);
1168                 smp_mb__after_clear_bit();
1169                 break;
1170         case TCP_CLOSE:
1171                 smp_mb__before_clear_bit();
1172                 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1173                 clear_bit(XPRT_CLOSING, &xprt->state);
1174                 smp_mb__after_clear_bit();
1175                 /* Mark transport as closed and wake up all pending tasks */
1176                 xprt_disconnect_done(xprt);
1177         }
1178  out:
1179         read_unlock(&sk->sk_callback_lock);
1180 }
1181
1182 /**
1183  * xs_udp_write_space - callback invoked when socket buffer space
1184  *                             becomes available
1185  * @sk: socket whose state has changed
1186  *
1187  * Called when more output buffer space is available for this socket.
1188  * We try not to wake our writers until they can make "significant"
1189  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1190  * with a bunch of small requests.
1191  */
1192 static void xs_udp_write_space(struct sock *sk)
1193 {
1194         read_lock(&sk->sk_callback_lock);
1195
1196         /* from net/core/sock.c:sock_def_write_space */
1197         if (sock_writeable(sk)) {
1198                 struct socket *sock;
1199                 struct rpc_xprt *xprt;
1200
1201                 if (unlikely(!(sock = sk->sk_socket)))
1202                         goto out;
1203                 clear_bit(SOCK_NOSPACE, &sock->flags);
1204
1205                 if (unlikely(!(xprt = xprt_from_sock(sk))))
1206                         goto out;
1207                 if (test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags) == 0)
1208                         goto out;
1209
1210                 xprt_write_space(xprt);
1211         }
1212
1213  out:
1214         read_unlock(&sk->sk_callback_lock);
1215 }
1216
1217 /**
1218  * xs_tcp_write_space - callback invoked when socket buffer space
1219  *                             becomes available
1220  * @sk: socket whose state has changed
1221  *
1222  * Called when more output buffer space is available for this socket.
1223  * We try not to wake our writers until they can make "significant"
1224  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1225  * with a bunch of small requests.
1226  */
1227 static void xs_tcp_write_space(struct sock *sk)
1228 {
1229         read_lock(&sk->sk_callback_lock);
1230
1231         /* from net/core/stream.c:sk_stream_write_space */
1232         if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk)) {
1233                 struct socket *sock;
1234                 struct rpc_xprt *xprt;
1235
1236                 if (unlikely(!(sock = sk->sk_socket)))
1237                         goto out;
1238                 clear_bit(SOCK_NOSPACE, &sock->flags);
1239
1240                 if (unlikely(!(xprt = xprt_from_sock(sk))))
1241                         goto out;
1242                 if (test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags) == 0)
1243                         goto out;
1244
1245                 xprt_write_space(xprt);
1246         }
1247
1248  out:
1249         read_unlock(&sk->sk_callback_lock);
1250 }
1251
1252 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
1253 {
1254         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1255         struct sock *sk = transport->inet;
1256
1257         if (transport->rcvsize) {
1258                 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1259                 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
1260         }
1261         if (transport->sndsize) {
1262                 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1263                 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1264                 sk->sk_write_space(sk);
1265         }
1266 }
1267
1268 /**
1269  * xs_udp_set_buffer_size - set send and receive limits
1270  * @xprt: generic transport
1271  * @sndsize: requested size of send buffer, in bytes
1272  * @rcvsize: requested size of receive buffer, in bytes
1273  *
1274  * Set socket send and receive buffer size limits.
1275  */
1276 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1277 {
1278         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1279
1280         transport->sndsize = 0;
1281         if (sndsize)
1282                 transport->sndsize = sndsize + 1024;
1283         transport->rcvsize = 0;
1284         if (rcvsize)
1285                 transport->rcvsize = rcvsize + 1024;
1286
1287         xs_udp_do_set_buffer_size(xprt);
1288 }
1289
1290 /**
1291  * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1292  * @task: task that timed out
1293  *
1294  * Adjust the congestion window after a retransmit timeout has occurred.
1295  */
1296 static void xs_udp_timer(struct rpc_task *task)
1297 {
1298         xprt_adjust_cwnd(task, -ETIMEDOUT);
1299 }
1300
1301 static unsigned short xs_get_random_port(void)
1302 {
1303         unsigned short range = xprt_max_resvport - xprt_min_resvport;
1304         unsigned short rand = (unsigned short) net_random() % range;
1305         return rand + xprt_min_resvport;
1306 }
1307
1308 /**
1309  * xs_set_port - reset the port number in the remote endpoint address
1310  * @xprt: generic transport
1311  * @port: new port number
1312  *
1313  */
1314 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1315 {
1316         struct sockaddr *addr = xs_addr(xprt);
1317
1318         dprintk("RPC:       setting port for xprt %p to %u\n", xprt, port);
1319
1320         switch (addr->sa_family) {
1321         case AF_INET:
1322                 ((struct sockaddr_in *)addr)->sin_port = htons(port);
1323                 break;
1324         case AF_INET6:
1325                 ((struct sockaddr_in6 *)addr)->sin6_port = htons(port);
1326                 break;
1327         default:
1328                 BUG();
1329         }
1330 }
1331
1332 static unsigned short xs_get_srcport(struct sock_xprt *transport, struct socket *sock)
1333 {
1334         unsigned short port = transport->port;
1335
1336         if (port == 0 && transport->xprt.resvport)
1337                 port = xs_get_random_port();
1338         return port;
1339 }
1340
1341 static unsigned short xs_next_srcport(struct sock_xprt *transport, struct socket *sock, unsigned short port)
1342 {
1343         if (transport->port != 0)
1344                 transport->port = 0;
1345         if (!transport->xprt.resvport)
1346                 return 0;
1347         if (port <= xprt_min_resvport || port > xprt_max_resvport)
1348                 return xprt_max_resvport;
1349         return --port;
1350 }
1351
1352 static int xs_bind4(struct sock_xprt *transport, struct socket *sock)
1353 {
1354         struct sockaddr_in myaddr = {
1355                 .sin_family = AF_INET,
1356         };
1357         struct sockaddr_in *sa;
1358         int err, nloop = 0;
1359         unsigned short port = xs_get_srcport(transport, sock);
1360         unsigned short last;
1361
1362         sa = (struct sockaddr_in *)&transport->addr;
1363         myaddr.sin_addr = sa->sin_addr;
1364         do {
1365                 myaddr.sin_port = htons(port);
1366                 err = kernel_bind(sock, (struct sockaddr *) &myaddr,
1367                                                 sizeof(myaddr));
1368                 if (port == 0)
1369                         break;
1370                 if (err == 0) {
1371                         transport->port = port;
1372                         break;
1373                 }
1374                 last = port;
1375                 port = xs_next_srcport(transport, sock, port);
1376                 if (port > last)
1377                         nloop++;
1378         } while (err == -EADDRINUSE && nloop != 2);
1379         dprintk("RPC:       %s "NIPQUAD_FMT":%u: %s (%d)\n",
1380                         __func__, NIPQUAD(myaddr.sin_addr),
1381                         port, err ? "failed" : "ok", err);
1382         return err;
1383 }
1384
1385 static int xs_bind6(struct sock_xprt *transport, struct socket *sock)
1386 {
1387         struct sockaddr_in6 myaddr = {
1388                 .sin6_family = AF_INET6,
1389         };
1390         struct sockaddr_in6 *sa;
1391         int err, nloop = 0;
1392         unsigned short port = xs_get_srcport(transport, sock);
1393         unsigned short last;
1394
1395         sa = (struct sockaddr_in6 *)&transport->addr;
1396         myaddr.sin6_addr = sa->sin6_addr;
1397         do {
1398                 myaddr.sin6_port = htons(port);
1399                 err = kernel_bind(sock, (struct sockaddr *) &myaddr,
1400                                                 sizeof(myaddr));
1401                 if (port == 0)
1402                         break;
1403                 if (err == 0) {
1404                         transport->port = port;
1405                         break;
1406                 }
1407                 last = port;
1408                 port = xs_next_srcport(transport, sock, port);
1409                 if (port > last)
1410                         nloop++;
1411         } while (err == -EADDRINUSE && nloop != 2);
1412         dprintk("RPC:       xs_bind6 "NIP6_FMT":%u: %s (%d)\n",
1413                 NIP6(myaddr.sin6_addr), port, err ? "failed" : "ok", err);
1414         return err;
1415 }
1416
1417 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1418 static struct lock_class_key xs_key[2];
1419 static struct lock_class_key xs_slock_key[2];
1420
1421 static inline void xs_reclassify_socket4(struct socket *sock)
1422 {
1423         struct sock *sk = sock->sk;
1424
1425         BUG_ON(sock_owned_by_user(sk));
1426         sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1427                 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1428 }
1429
1430 static inline void xs_reclassify_socket6(struct socket *sock)
1431 {
1432         struct sock *sk = sock->sk;
1433
1434         BUG_ON(sock_owned_by_user(sk));
1435         sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1436                 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
1437 }
1438 #else
1439 static inline void xs_reclassify_socket4(struct socket *sock)
1440 {
1441 }
1442
1443 static inline void xs_reclassify_socket6(struct socket *sock)
1444 {
1445 }
1446 #endif
1447
1448 static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
1449 {
1450         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1451
1452         if (!transport->inet) {
1453                 struct sock *sk = sock->sk;
1454
1455                 write_lock_bh(&sk->sk_callback_lock);
1456
1457                 sk->sk_user_data = xprt;
1458                 transport->old_data_ready = sk->sk_data_ready;
1459                 transport->old_state_change = sk->sk_state_change;
1460                 transport->old_write_space = sk->sk_write_space;
1461                 sk->sk_data_ready = xs_udp_data_ready;
1462                 sk->sk_write_space = xs_udp_write_space;
1463                 sk->sk_no_check = UDP_CSUM_NORCV;
1464                 sk->sk_allocation = GFP_ATOMIC;
1465
1466                 xprt_set_connected(xprt);
1467
1468                 /* Reset to new socket */
1469                 transport->sock = sock;
1470                 transport->inet = sk;
1471
1472                 write_unlock_bh(&sk->sk_callback_lock);
1473         }
1474         xs_udp_do_set_buffer_size(xprt);
1475 }
1476
1477 /**
1478  * xs_udp_connect_worker4 - set up a UDP socket
1479  * @work: RPC transport to connect
1480  *
1481  * Invoked by a work queue tasklet.
1482  */
1483 static void xs_udp_connect_worker4(struct work_struct *work)
1484 {
1485         struct sock_xprt *transport =
1486                 container_of(work, struct sock_xprt, connect_worker.work);
1487         struct rpc_xprt *xprt = &transport->xprt;
1488         struct socket *sock = transport->sock;
1489         int err, status = -EIO;
1490
1491         if (xprt->shutdown || !xprt_bound(xprt))
1492                 goto out;
1493
1494         /* Start by resetting any existing state */
1495         xs_close(xprt);
1496
1497         if ((err = sock_create_kern(PF_INET, SOCK_DGRAM, IPPROTO_UDP, &sock)) < 0) {
1498                 dprintk("RPC:       can't create UDP transport socket (%d).\n", -err);
1499                 goto out;
1500         }
1501         xs_reclassify_socket4(sock);
1502
1503         if (xs_bind4(transport, sock)) {
1504                 sock_release(sock);
1505                 goto out;
1506         }
1507
1508         dprintk("RPC:       worker connecting xprt %p to address: %s\n",
1509                         xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1510
1511         xs_udp_finish_connecting(xprt, sock);
1512         status = 0;
1513 out:
1514         xprt_wake_pending_tasks(xprt, status);
1515         xprt_clear_connecting(xprt);
1516 }
1517
1518 /**
1519  * xs_udp_connect_worker6 - set up a UDP socket
1520  * @work: RPC transport to connect
1521  *
1522  * Invoked by a work queue tasklet.
1523  */
1524 static void xs_udp_connect_worker6(struct work_struct *work)
1525 {
1526         struct sock_xprt *transport =
1527                 container_of(work, struct sock_xprt, connect_worker.work);
1528         struct rpc_xprt *xprt = &transport->xprt;
1529         struct socket *sock = transport->sock;
1530         int err, status = -EIO;
1531
1532         if (xprt->shutdown || !xprt_bound(xprt))
1533                 goto out;
1534
1535         /* Start by resetting any existing state */
1536         xs_close(xprt);
1537
1538         if ((err = sock_create_kern(PF_INET6, SOCK_DGRAM, IPPROTO_UDP, &sock)) < 0) {
1539                 dprintk("RPC:       can't create UDP transport socket (%d).\n", -err);
1540                 goto out;
1541         }
1542         xs_reclassify_socket6(sock);
1543
1544         if (xs_bind6(transport, sock) < 0) {
1545                 sock_release(sock);
1546                 goto out;
1547         }
1548
1549         dprintk("RPC:       worker connecting xprt %p to address: %s\n",
1550                         xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1551
1552         xs_udp_finish_connecting(xprt, sock);
1553         status = 0;
1554 out:
1555         xprt_wake_pending_tasks(xprt, status);
1556         xprt_clear_connecting(xprt);
1557 }
1558
1559 /*
1560  * We need to preserve the port number so the reply cache on the server can
1561  * find our cached RPC replies when we get around to reconnecting.
1562  */
1563 static void xs_tcp_reuse_connection(struct rpc_xprt *xprt)
1564 {
1565         int result;
1566         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1567         struct sockaddr any;
1568
1569         dprintk("RPC:       disconnecting xprt %p to reuse port\n", xprt);
1570
1571         /*
1572          * Disconnect the transport socket by doing a connect operation
1573          * with AF_UNSPEC.  This should return immediately...
1574          */
1575         memset(&any, 0, sizeof(any));
1576         any.sa_family = AF_UNSPEC;
1577         result = kernel_connect(transport->sock, &any, sizeof(any), 0);
1578         if (result)
1579                 dprintk("RPC:       AF_UNSPEC connect return code %d\n",
1580                                 result);
1581 }
1582
1583 static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
1584 {
1585         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1586
1587         if (!transport->inet) {
1588                 struct sock *sk = sock->sk;
1589
1590                 write_lock_bh(&sk->sk_callback_lock);
1591
1592                 sk->sk_user_data = xprt;
1593                 transport->old_data_ready = sk->sk_data_ready;
1594                 transport->old_state_change = sk->sk_state_change;
1595                 transport->old_write_space = sk->sk_write_space;
1596                 sk->sk_data_ready = xs_tcp_data_ready;
1597                 sk->sk_state_change = xs_tcp_state_change;
1598                 sk->sk_write_space = xs_tcp_write_space;
1599                 sk->sk_allocation = GFP_ATOMIC;
1600
1601                 /* socket options */
1602                 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
1603                 sock_reset_flag(sk, SOCK_LINGER);
1604                 tcp_sk(sk)->linger2 = 0;
1605                 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
1606
1607                 xprt_clear_connected(xprt);
1608
1609                 /* Reset to new socket */
1610                 transport->sock = sock;
1611                 transport->inet = sk;
1612
1613                 write_unlock_bh(&sk->sk_callback_lock);
1614         }
1615
1616         /* Tell the socket layer to start connecting... */
1617         xprt->stat.connect_count++;
1618         xprt->stat.connect_start = jiffies;
1619         return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
1620 }
1621
1622 /**
1623  * xs_tcp_connect_worker4 - connect a TCP socket to a remote endpoint
1624  * @work: RPC transport to connect
1625  *
1626  * Invoked by a work queue tasklet.
1627  */
1628 static void xs_tcp_connect_worker4(struct work_struct *work)
1629 {
1630         struct sock_xprt *transport =
1631                 container_of(work, struct sock_xprt, connect_worker.work);
1632         struct rpc_xprt *xprt = &transport->xprt;
1633         struct socket *sock = transport->sock;
1634         int err, status = -EIO;
1635
1636         if (xprt->shutdown || !xprt_bound(xprt))
1637                 goto out;
1638
1639         if (!sock) {
1640                 /* start from scratch */
1641                 if ((err = sock_create_kern(PF_INET, SOCK_STREAM, IPPROTO_TCP, &sock)) < 0) {
1642                         dprintk("RPC:       can't create TCP transport socket (%d).\n", -err);
1643                         goto out;
1644                 }
1645                 xs_reclassify_socket4(sock);
1646
1647                 if (xs_bind4(transport, sock) < 0) {
1648                         sock_release(sock);
1649                         goto out;
1650                 }
1651         } else
1652                 /* "close" the socket, preserving the local port */
1653                 xs_tcp_reuse_connection(xprt);
1654
1655         dprintk("RPC:       worker connecting xprt %p to address: %s\n",
1656                         xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1657
1658         status = xs_tcp_finish_connecting(xprt, sock);
1659         dprintk("RPC:       %p connect status %d connected %d sock state %d\n",
1660                         xprt, -status, xprt_connected(xprt),
1661                         sock->sk->sk_state);
1662         if (status < 0) {
1663                 switch (status) {
1664                         case -EINPROGRESS:
1665                         case -EALREADY:
1666                                 goto out_clear;
1667                         case -ECONNREFUSED:
1668                         case -ECONNRESET:
1669                                 /* retry with existing socket, after a delay */
1670                                 break;
1671                         default:
1672                                 /* get rid of existing socket, and retry */
1673                                 xs_tcp_shutdown(xprt);
1674                 }
1675         }
1676 out:
1677         xprt_wake_pending_tasks(xprt, status);
1678 out_clear:
1679         xprt_clear_connecting(xprt);
1680 }
1681
1682 /**
1683  * xs_tcp_connect_worker6 - connect a TCP socket to a remote endpoint
1684  * @work: RPC transport to connect
1685  *
1686  * Invoked by a work queue tasklet.
1687  */
1688 static void xs_tcp_connect_worker6(struct work_struct *work)
1689 {
1690         struct sock_xprt *transport =
1691                 container_of(work, struct sock_xprt, connect_worker.work);
1692         struct rpc_xprt *xprt = &transport->xprt;
1693         struct socket *sock = transport->sock;
1694         int err, status = -EIO;
1695
1696         if (xprt->shutdown || !xprt_bound(xprt))
1697                 goto out;
1698
1699         if (!sock) {
1700                 /* start from scratch */
1701                 if ((err = sock_create_kern(PF_INET6, SOCK_STREAM, IPPROTO_TCP, &sock)) < 0) {
1702                         dprintk("RPC:       can't create TCP transport socket (%d).\n", -err);
1703                         goto out;
1704                 }
1705                 xs_reclassify_socket6(sock);
1706
1707                 if (xs_bind6(transport, sock) < 0) {
1708                         sock_release(sock);
1709                         goto out;
1710                 }
1711         } else
1712                 /* "close" the socket, preserving the local port */
1713                 xs_tcp_reuse_connection(xprt);
1714
1715         dprintk("RPC:       worker connecting xprt %p to address: %s\n",
1716                         xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1717
1718         status = xs_tcp_finish_connecting(xprt, sock);
1719         dprintk("RPC:       %p connect status %d connected %d sock state %d\n",
1720                         xprt, -status, xprt_connected(xprt), sock->sk->sk_state);
1721         if (status < 0) {
1722                 switch (status) {
1723                         case -EINPROGRESS:
1724                         case -EALREADY:
1725                                 goto out_clear;
1726                         case -ECONNREFUSED:
1727                         case -ECONNRESET:
1728                                 /* retry with existing socket, after a delay */
1729                                 break;
1730                         default:
1731                                 /* get rid of existing socket, and retry */
1732                                 xs_tcp_shutdown(xprt);
1733                 }
1734         }
1735 out:
1736         xprt_wake_pending_tasks(xprt, status);
1737 out_clear:
1738         xprt_clear_connecting(xprt);
1739 }
1740
1741 /**
1742  * xs_connect - connect a socket to a remote endpoint
1743  * @task: address of RPC task that manages state of connect request
1744  *
1745  * TCP: If the remote end dropped the connection, delay reconnecting.
1746  *
1747  * UDP socket connects are synchronous, but we use a work queue anyway
1748  * to guarantee that even unprivileged user processes can set up a
1749  * socket on a privileged port.
1750  *
1751  * If a UDP socket connect fails, the delay behavior here prevents
1752  * retry floods (hard mounts).
1753  */
1754 static void xs_connect(struct rpc_task *task)
1755 {
1756         struct rpc_xprt *xprt = task->tk_xprt;
1757         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1758
1759         if (xprt_test_and_set_connecting(xprt))
1760                 return;
1761
1762         if (transport->sock != NULL) {
1763                 dprintk("RPC:       xs_connect delayed xprt %p for %lu "
1764                                 "seconds\n",
1765                                 xprt, xprt->reestablish_timeout / HZ);
1766                 queue_delayed_work(rpciod_workqueue,
1767                                    &transport->connect_worker,
1768                                    xprt->reestablish_timeout);
1769                 xprt->reestablish_timeout <<= 1;
1770                 if (xprt->reestablish_timeout > XS_TCP_MAX_REEST_TO)
1771                         xprt->reestablish_timeout = XS_TCP_MAX_REEST_TO;
1772         } else {
1773                 dprintk("RPC:       xs_connect scheduled xprt %p\n", xprt);
1774                 queue_delayed_work(rpciod_workqueue,
1775                                    &transport->connect_worker, 0);
1776         }
1777 }
1778
1779 static void xs_tcp_connect(struct rpc_task *task)
1780 {
1781         struct rpc_xprt *xprt = task->tk_xprt;
1782
1783         /* Initiate graceful shutdown of the socket if not already done */
1784         if (test_bit(XPRT_CONNECTED, &xprt->state))
1785                 xs_tcp_shutdown(xprt);
1786         /* Exit if we need to wait for socket shutdown to complete */
1787         if (test_bit(XPRT_CLOSING, &xprt->state))
1788                 return;
1789         xs_connect(task);
1790 }
1791
1792 /**
1793  * xs_udp_print_stats - display UDP socket-specifc stats
1794  * @xprt: rpc_xprt struct containing statistics
1795  * @seq: output file
1796  *
1797  */
1798 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
1799 {
1800         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1801
1802         seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %Lu %Lu\n",
1803                         transport->port,
1804                         xprt->stat.bind_count,
1805                         xprt->stat.sends,
1806                         xprt->stat.recvs,
1807                         xprt->stat.bad_xids,
1808                         xprt->stat.req_u,
1809                         xprt->stat.bklog_u);
1810 }
1811
1812 /**
1813  * xs_tcp_print_stats - display TCP socket-specifc stats
1814  * @xprt: rpc_xprt struct containing statistics
1815  * @seq: output file
1816  *
1817  */
1818 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
1819 {
1820         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1821         long idle_time = 0;
1822
1823         if (xprt_connected(xprt))
1824                 idle_time = (long)(jiffies - xprt->last_used) / HZ;
1825
1826         seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu %Lu %Lu\n",
1827                         transport->port,
1828                         xprt->stat.bind_count,
1829                         xprt->stat.connect_count,
1830                         xprt->stat.connect_time,
1831                         idle_time,
1832                         xprt->stat.sends,
1833                         xprt->stat.recvs,
1834                         xprt->stat.bad_xids,
1835                         xprt->stat.req_u,
1836                         xprt->stat.bklog_u);
1837 }
1838
1839 static struct rpc_xprt_ops xs_udp_ops = {
1840         .set_buffer_size        = xs_udp_set_buffer_size,
1841         .reserve_xprt           = xprt_reserve_xprt_cong,
1842         .release_xprt           = xprt_release_xprt_cong,
1843         .rpcbind                = rpcb_getport_async,
1844         .set_port               = xs_set_port,
1845         .connect                = xs_connect,
1846         .buf_alloc              = rpc_malloc,
1847         .buf_free               = rpc_free,
1848         .send_request           = xs_udp_send_request,
1849         .set_retrans_timeout    = xprt_set_retrans_timeout_rtt,
1850         .timer                  = xs_udp_timer,
1851         .release_request        = xprt_release_rqst_cong,
1852         .close                  = xs_close,
1853         .destroy                = xs_destroy,
1854         .print_stats            = xs_udp_print_stats,
1855 };
1856
1857 static struct rpc_xprt_ops xs_tcp_ops = {
1858         .reserve_xprt           = xprt_reserve_xprt,
1859         .release_xprt           = xs_tcp_release_xprt,
1860         .rpcbind                = rpcb_getport_async,
1861         .set_port               = xs_set_port,
1862         .connect                = xs_tcp_connect,
1863         .buf_alloc              = rpc_malloc,
1864         .buf_free               = rpc_free,
1865         .send_request           = xs_tcp_send_request,
1866         .set_retrans_timeout    = xprt_set_retrans_timeout_def,
1867         .close                  = xs_tcp_shutdown,
1868         .destroy                = xs_destroy,
1869         .print_stats            = xs_tcp_print_stats,
1870 };
1871
1872 static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
1873                                       unsigned int slot_table_size)
1874 {
1875         struct rpc_xprt *xprt;
1876         struct sock_xprt *new;
1877
1878         if (args->addrlen > sizeof(xprt->addr)) {
1879                 dprintk("RPC:       xs_setup_xprt: address too large\n");
1880                 return ERR_PTR(-EBADF);
1881         }
1882
1883         new = kzalloc(sizeof(*new), GFP_KERNEL);
1884         if (new == NULL) {
1885                 dprintk("RPC:       xs_setup_xprt: couldn't allocate "
1886                                 "rpc_xprt\n");
1887                 return ERR_PTR(-ENOMEM);
1888         }
1889         xprt = &new->xprt;
1890
1891         xprt->max_reqs = slot_table_size;
1892         xprt->slot = kcalloc(xprt->max_reqs, sizeof(struct rpc_rqst), GFP_KERNEL);
1893         if (xprt->slot == NULL) {
1894                 kfree(xprt);
1895                 dprintk("RPC:       xs_setup_xprt: couldn't allocate slot "
1896                                 "table\n");
1897                 return ERR_PTR(-ENOMEM);
1898         }
1899
1900         memcpy(&xprt->addr, args->dstaddr, args->addrlen);
1901         xprt->addrlen = args->addrlen;
1902         if (args->srcaddr)
1903                 memcpy(&new->addr, args->srcaddr, args->addrlen);
1904
1905         return xprt;
1906 }
1907
1908 static const struct rpc_timeout xs_udp_default_timeout = {
1909         .to_initval = 5 * HZ,
1910         .to_maxval = 30 * HZ,
1911         .to_increment = 5 * HZ,
1912         .to_retries = 5,
1913 };
1914
1915 /**
1916  * xs_setup_udp - Set up transport to use a UDP socket
1917  * @args: rpc transport creation arguments
1918  *
1919  */
1920 static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
1921 {
1922         struct sockaddr *addr = args->dstaddr;
1923         struct rpc_xprt *xprt;
1924         struct sock_xprt *transport;
1925
1926         xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries);
1927         if (IS_ERR(xprt))
1928                 return xprt;
1929         transport = container_of(xprt, struct sock_xprt, xprt);
1930
1931         xprt->prot = IPPROTO_UDP;
1932         xprt->tsh_size = 0;
1933         /* XXX: header size can vary due to auth type, IPv6, etc. */
1934         xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
1935
1936         xprt->bind_timeout = XS_BIND_TO;
1937         xprt->connect_timeout = XS_UDP_CONN_TO;
1938         xprt->reestablish_timeout = XS_UDP_REEST_TO;
1939         xprt->idle_timeout = XS_IDLE_DISC_TO;
1940
1941         xprt->ops = &xs_udp_ops;
1942
1943         xprt->timeout = &xs_udp_default_timeout;
1944
1945         switch (addr->sa_family) {
1946         case AF_INET:
1947                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
1948                         xprt_set_bound(xprt);
1949
1950                 INIT_DELAYED_WORK(&transport->connect_worker,
1951                                         xs_udp_connect_worker4);
1952                 xs_format_ipv4_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
1953                 break;
1954         case AF_INET6:
1955                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
1956                         xprt_set_bound(xprt);
1957
1958                 INIT_DELAYED_WORK(&transport->connect_worker,
1959                                         xs_udp_connect_worker6);
1960                 xs_format_ipv6_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
1961                 break;
1962         default:
1963                 kfree(xprt);
1964                 return ERR_PTR(-EAFNOSUPPORT);
1965         }
1966
1967         dprintk("RPC:       set up transport to address %s\n",
1968                         xprt->address_strings[RPC_DISPLAY_ALL]);
1969
1970         if (try_module_get(THIS_MODULE))
1971                 return xprt;
1972
1973         kfree(xprt->slot);
1974         kfree(xprt);
1975         return ERR_PTR(-EINVAL);
1976 }
1977
1978 static const struct rpc_timeout xs_tcp_default_timeout = {
1979         .to_initval = 60 * HZ,
1980         .to_maxval = 60 * HZ,
1981         .to_retries = 2,
1982 };
1983
1984 /**
1985  * xs_setup_tcp - Set up transport to use a TCP socket
1986  * @args: rpc transport creation arguments
1987  *
1988  */
1989 static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
1990 {
1991         struct sockaddr *addr = args->dstaddr;
1992         struct rpc_xprt *xprt;
1993         struct sock_xprt *transport;
1994
1995         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries);
1996         if (IS_ERR(xprt))
1997                 return xprt;
1998         transport = container_of(xprt, struct sock_xprt, xprt);
1999
2000         xprt->prot = IPPROTO_TCP;
2001         xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2002         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2003
2004         xprt->bind_timeout = XS_BIND_TO;
2005         xprt->connect_timeout = XS_TCP_CONN_TO;
2006         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2007         xprt->idle_timeout = XS_IDLE_DISC_TO;
2008
2009         xprt->ops = &xs_tcp_ops;
2010         xprt->timeout = &xs_tcp_default_timeout;
2011
2012         switch (addr->sa_family) {
2013         case AF_INET:
2014                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2015                         xprt_set_bound(xprt);
2016
2017                 INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_connect_worker4);
2018                 xs_format_ipv4_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
2019                 break;
2020         case AF_INET6:
2021                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2022                         xprt_set_bound(xprt);
2023
2024                 INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_connect_worker6);
2025                 xs_format_ipv6_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
2026                 break;
2027         default:
2028                 kfree(xprt);
2029                 return ERR_PTR(-EAFNOSUPPORT);
2030         }
2031
2032         dprintk("RPC:       set up transport to address %s\n",
2033                         xprt->address_strings[RPC_DISPLAY_ALL]);
2034
2035         if (try_module_get(THIS_MODULE))
2036                 return xprt;
2037
2038         kfree(xprt->slot);
2039         kfree(xprt);
2040         return ERR_PTR(-EINVAL);
2041 }
2042
2043 static struct xprt_class        xs_udp_transport = {
2044         .list           = LIST_HEAD_INIT(xs_udp_transport.list),
2045         .name           = "udp",
2046         .owner          = THIS_MODULE,
2047         .ident          = IPPROTO_UDP,
2048         .setup          = xs_setup_udp,
2049 };
2050
2051 static struct xprt_class        xs_tcp_transport = {
2052         .list           = LIST_HEAD_INIT(xs_tcp_transport.list),
2053         .name           = "tcp",
2054         .owner          = THIS_MODULE,
2055         .ident          = IPPROTO_TCP,
2056         .setup          = xs_setup_tcp,
2057 };
2058
2059 /**
2060  * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
2061  *
2062  */
2063 int init_socket_xprt(void)
2064 {
2065 #ifdef RPC_DEBUG
2066         if (!sunrpc_table_header)
2067                 sunrpc_table_header = register_sysctl_table(sunrpc_table);
2068 #endif
2069
2070         xprt_register_transport(&xs_udp_transport);
2071         xprt_register_transport(&xs_tcp_transport);
2072
2073         return 0;
2074 }
2075
2076 /**
2077  * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
2078  *
2079  */
2080 void cleanup_socket_xprt(void)
2081 {
2082 #ifdef RPC_DEBUG
2083         if (sunrpc_table_header) {
2084                 unregister_sysctl_table(sunrpc_table_header);
2085                 sunrpc_table_header = NULL;
2086         }
2087 #endif
2088
2089         xprt_unregister_transport(&xs_udp_transport);
2090         xprt_unregister_transport(&xs_tcp_transport);
2091 }