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
   7  * TCP send fixes (C) 1998 Red Hat
   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/string.h>
  23 #include <linux/slab.h>
  24 #include <linux/module.h>
  25 #include <linux/capability.h>
  26 #include <linux/pagemap.h>
  27 #include <linux/errno.h>
  28 #include <linux/socket.h>
  29 #include <linux/in.h>
  30 #include <linux/net.h>
  31 #include <linux/mm.h>
  32 #include <linux/un.h>
  33 #include <linux/udp.h>
  34 #include <linux/tcp.h>
  35 #include <linux/sunrpc/clnt.h>
  36 #include <linux/sunrpc/sched.h>
  37 #include <linux/sunrpc/svcsock.h>
  38 #include <linux/sunrpc/xprtsock.h>
  39 #include <linux/file.h>
  40 #ifdef CONFIG_SUNRPC_BACKCHANNEL
  41 #include <linux/sunrpc/bc_xprt.h>
  42 #endif
  43
  44 #include <net/sock.h>
  45 #include <net/checksum.h>
  46 #include <net/udp.h>
  47 #include <net/tcp.h>
  48
  49 #include "sunrpc.h"
  50
  51 static void xs_close(struct rpc_xprt *xprt);
  52
  53 /*
  54  * xprtsock tunables
  55  */
  56 static unsigned int xprt_udp_slot_table_entries = RPC_DEF_SLOT_TABLE;
  57 static unsigned int xprt_tcp_slot_table_entries = RPC_MIN_SLOT_TABLE;
  58 static unsigned int xprt_max_tcp_slot_table_entries = RPC_MAX_SLOT_TABLE;
  59
  60 static unsigned int xprt_min_resvport = RPC_DEF_MIN_RESVPORT;
  61 static unsigned int xprt_max_resvport = RPC_DEF_MAX_RESVPORT;
  62
  63 #define XS_TCP_LINGER_TO        (15U * HZ)
  64 static unsigned int xs_tcp_fin_timeout __read_mostly = XS_TCP_LINGER_TO;
  65
  66 /*
  67  * We can register our own files under /proc/sys/sunrpc by
  68  * calling register_sysctl_table() again.  The files in that
  69  * directory become the union of all files registered there.
  70  *
  71  * We simply need to make sure that we don't collide with
  72  * someone else's file names!
  73  */
  74
  75 #ifdef RPC_DEBUG
  76
  77 static unsigned int min_slot_table_size = RPC_MIN_SLOT_TABLE;
  78 static unsigned int max_slot_table_size = RPC_MAX_SLOT_TABLE;
  79 static unsigned int max_tcp_slot_table_limit = RPC_MAX_SLOT_TABLE_LIMIT;
  80 static unsigned int xprt_min_resvport_limit = RPC_MIN_RESVPORT;
  81 static unsigned int xprt_max_resvport_limit = RPC_MAX_RESVPORT;
  82
  83 static struct ctl_table_header *sunrpc_table_header;
  84
  85 /*
  86  * FIXME: changing the UDP slot table size should also resize the UDP
  87  *        socket buffers for existing UDP transports
  88  */
  89 static ctl_table xs_tunables_table[] = {
  90         {
  91                 .procname       = "udp_slot_table_entries",
  92                 .data           = &xprt_udp_slot_table_entries,
  93                 .maxlen         = sizeof(unsigned int),
  94                 .mode           = 0644,
  95                 .proc_handler   = proc_dointvec_minmax,
  96                 .extra1         = &min_slot_table_size,
  97                 .extra2         = &max_slot_table_size
  98         },
  99         {
 100                 .procname       = "tcp_slot_table_entries",
 101                 .data           = &xprt_tcp_slot_table_entries,
 102                 .maxlen         = sizeof(unsigned int),
 103                 .mode           = 0644,
 104                 .proc_handler   = proc_dointvec_minmax,
 105                 .extra1         = &min_slot_table_size,
 106                 .extra2         = &max_slot_table_size
 107         },
 108         {
 109                 .procname       = "tcp_max_slot_table_entries",
 110                 .data           = &xprt_max_tcp_slot_table_entries,
 111                 .maxlen         = sizeof(unsigned int),
 112                 .mode           = 0644,
 113                 .proc_handler   = proc_dointvec_minmax,
 114                 .extra1         = &min_slot_table_size,
 115                 .extra2         = &max_tcp_slot_table_limit
 116         },
 117         {
 118                 .procname       = "min_resvport",
 119                 .data           = &xprt_min_resvport,
 120                 .maxlen         = sizeof(unsigned int),
 121                 .mode           = 0644,
 122                 .proc_handler   = proc_dointvec_minmax,
 123                 .extra1         = &xprt_min_resvport_limit,
 124                 .extra2         = &xprt_max_resvport_limit
 125         },
 126         {
 127                 .procname       = "max_resvport",
 128                 .data           = &xprt_max_resvport,
 129                 .maxlen         = sizeof(unsigned int),
 130                 .mode           = 0644,
 131                 .proc_handler   = proc_dointvec_minmax,
 132                 .extra1         = &xprt_min_resvport_limit,
 133                 .extra2         = &xprt_max_resvport_limit
 134         },
 135         {
 136                 .procname       = "tcp_fin_timeout",
 137                 .data           = &xs_tcp_fin_timeout,
 138                 .maxlen         = sizeof(xs_tcp_fin_timeout),
 139                 .mode           = 0644,
 140                 .proc_handler   = proc_dointvec_jiffies,
 141         },
 142         { },
 143 };
 144
 145 static ctl_table sunrpc_table[] = {
 146         {
 147                 .procname       = "sunrpc",
 148                 .mode           = 0555,
 149                 .child          = xs_tunables_table
 150         },
 151         { },
 152 };
 153
 154 #endif
 155
 156 /*
 157  * Wait duration for a reply from the RPC portmapper.
 158  */
 159 #define XS_BIND_TO              (60U * HZ)
 160
 161 /*
 162  * Delay if a UDP socket connect error occurs.  This is most likely some
 163  * kind of resource problem on the local host.
 164  */
 165 #define XS_UDP_REEST_TO         (2U * HZ)
 166
 167 /*
 168  * The reestablish timeout allows clients to delay for a bit before attempting
 169  * to reconnect to a server that just dropped our connection.
 170  *
 171  * We implement an exponential backoff when trying to reestablish a TCP
 172  * transport connection with the server.  Some servers like to drop a TCP
 173  * connection when they are overworked, so we start with a short timeout and
 174  * increase over time if the server is down or not responding.
 175  */
 176 #define XS_TCP_INIT_REEST_TO    (3U * HZ)
 177 #define XS_TCP_MAX_REEST_TO     (5U * 60 * HZ)
 178
 179 /*
 180  * TCP idle timeout; client drops the transport socket if it is idle
 181  * for this long.  Note that we also timeout UDP sockets to prevent
 182  * holding port numbers when there is no RPC traffic.
 183  */
 184 #define XS_IDLE_DISC_TO         (5U * 60 * HZ)
 185
 186 #ifdef RPC_DEBUG
 187 # undef  RPC_DEBUG_DATA
 188 # define RPCDBG_FACILITY        RPCDBG_TRANS
 189 #endif
 190
 191 #ifdef RPC_DEBUG_DATA
 192 static void xs_pktdump(char *msg, u32 *packet, unsigned int count)
 193 {
 194         u8 *buf = (u8 *) packet;
 195         int j;
 196
 197         dprintk("RPC:       %s\n", msg);
 198         for (j = 0; j < count && j < 128; j += 4) {
 199                 if (!(j & 31)) {
 200                         if (j)
 201                                 dprintk("\n");
 202                         dprintk("0x%04x ", j);
 203                 }
 204                 dprintk("%02x%02x%02x%02x ",
 205                         buf[j], buf[j+1], buf[j+2], buf[j+3]);
 206         }
 207         dprintk("\n");
 208 }
 209 #else
 210 static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count)
 211 {
 212         /* NOP */
 213 }
 214 #endif
 215
 216 struct sock_xprt {
 217         struct rpc_xprt         xprt;
 218
 219         /*
 220          * Network layer
 221          */
 222         struct socket *         sock;
 223         struct sock *           inet;
 224
 225         /*
 226          * State of TCP reply receive
 227          */
 228         __be32                  tcp_fraghdr,
 229                                 tcp_xid,
 230                                 tcp_calldir;
 231
 232         u32                     tcp_offset,
 233                                 tcp_reclen;
 234
 235         unsigned long           tcp_copied,
 236                                 tcp_flags;
 237
 238         /*
 239          * Connection of transports
 240          */
 241         struct delayed_work     connect_worker;
 242         struct sockaddr_storage srcaddr;
 243         unsigned short          srcport;
 244
 245         /*
 246          * UDP socket buffer size parameters
 247          */
 248         size_t                  rcvsize,
 249                                 sndsize;
 250
 251         /*
 252          * Saved socket callback addresses
 253          */
 254         void                    (*old_data_ready)(struct sock *, int);
 255         void                    (*old_state_change)(struct sock *);
 256         void                    (*old_write_space)(struct sock *);
 257 };
 258
 259 /*
 260  * TCP receive state flags
 261  */
 262 #define TCP_RCV_LAST_FRAG       (1UL << 0)
 263 #define TCP_RCV_COPY_FRAGHDR    (1UL << 1)
 264 #define TCP_RCV_COPY_XID        (1UL << 2)
 265 #define TCP_RCV_COPY_DATA       (1UL << 3)
 266 #define TCP_RCV_READ_CALLDIR    (1UL << 4)
 267 #define TCP_RCV_COPY_CALLDIR    (1UL << 5)
 268
 269 /*
 270  * TCP RPC flags
 271  */
 272 #define TCP_RPC_REPLY           (1UL << 6)
 273
 274 static inline struct sockaddr *xs_addr(struct rpc_xprt *xprt)
 275 {
 276         return (struct sockaddr *) &xprt->addr;
 277 }
 278
 279 static inline struct sockaddr_un *xs_addr_un(struct rpc_xprt *xprt)
 280 {
 281         return (struct sockaddr_un *) &xprt->addr;
 282 }
 283
 284 static inline struct sockaddr_in *xs_addr_in(struct rpc_xprt *xprt)
 285 {
 286         return (struct sockaddr_in *) &xprt->addr;
 287 }
 288
 289 static inline struct sockaddr_in6 *xs_addr_in6(struct rpc_xprt *xprt)
 290 {
 291         return (struct sockaddr_in6 *) &xprt->addr;
 292 }
 293
 294 static void xs_format_common_peer_addresses(struct rpc_xprt *xprt)
 295 {
 296         struct sockaddr *sap = xs_addr(xprt);
 297         struct sockaddr_in6 *sin6;
 298         struct sockaddr_in *sin;
 299         struct sockaddr_un *sun;
 300         char buf[128];
 301
 302         switch (sap->sa_family) {
 303         case AF_LOCAL:
 304                 sun = xs_addr_un(xprt);
 305                 strlcpy(buf, sun->sun_path, sizeof(buf));
 306                 xprt->address_strings[RPC_DISPLAY_ADDR] =
 307                                                 kstrdup(buf, GFP_KERNEL);
 308                 break;
 309         case AF_INET:
 310                 (void)rpc_ntop(sap, buf, sizeof(buf));
 311                 xprt->address_strings[RPC_DISPLAY_ADDR] =
 312                                                 kstrdup(buf, GFP_KERNEL);
 313                 sin = xs_addr_in(xprt);
 314                 snprintf(buf, sizeof(buf), "%08x", ntohl(sin->sin_addr.s_addr));
 315                 break;
 316         case AF_INET6:
 317                 (void)rpc_ntop(sap, buf, sizeof(buf));
 318                 xprt->address_strings[RPC_DISPLAY_ADDR] =
 319                                                 kstrdup(buf, GFP_KERNEL);
 320                 sin6 = xs_addr_in6(xprt);
 321                 snprintf(buf, sizeof(buf), "%pi6", &sin6->sin6_addr);
 322                 break;
 323         default:
 324                 BUG();
 325         }
 326
 327         xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = kstrdup(buf, GFP_KERNEL);
 328 }
 329
 330 static void xs_format_common_peer_ports(struct rpc_xprt *xprt)
 331 {
 332         struct sockaddr *sap = xs_addr(xprt);
 333         char buf[128];
 334
 335         snprintf(buf, sizeof(buf), "%u", rpc_get_port(sap));
 336         xprt->address_strings[RPC_DISPLAY_PORT] = kstrdup(buf, GFP_KERNEL);
 337
 338         snprintf(buf, sizeof(buf), "%4hx", rpc_get_port(sap));
 339         xprt->address_strings[RPC_DISPLAY_HEX_PORT] = kstrdup(buf, GFP_KERNEL);
 340 }
 341
 342 static void xs_format_peer_addresses(struct rpc_xprt *xprt,
 343                                      const char *protocol,
 344                                      const char *netid)
 345 {
 346         xprt->address_strings[RPC_DISPLAY_PROTO] = protocol;
 347         xprt->address_strings[RPC_DISPLAY_NETID] = netid;
 348         xs_format_common_peer_addresses(xprt);
 349         xs_format_common_peer_ports(xprt);
 350 }
 351
 352 static void xs_update_peer_port(struct rpc_xprt *xprt)
 353 {
 354         kfree(xprt->address_strings[RPC_DISPLAY_HEX_PORT]);
 355         kfree(xprt->address_strings[RPC_DISPLAY_PORT]);
 356
 357         xs_format_common_peer_ports(xprt);
 358 }
 359
 360 static void xs_free_peer_addresses(struct rpc_xprt *xprt)
 361 {
 362         unsigned int i;
 363
 364         for (i = 0; i < RPC_DISPLAY_MAX; i++)
 365                 switch (i) {
 366                 case RPC_DISPLAY_PROTO:
 367                 case RPC_DISPLAY_NETID:
 368                         continue;
 369                 default:
 370                         kfree(xprt->address_strings[i]);
 371                 }
 372 }
 373
 374 #define XS_SENDMSG_FLAGS        (MSG_DONTWAIT | MSG_NOSIGNAL)
 375
 376 static int xs_send_kvec(struct socket *sock, struct sockaddr *addr, int addrlen, struct kvec *vec, unsigned int base, int more)
 377 {
 378         struct msghdr msg = {
 379                 .msg_name       = addr,
 380                 .msg_namelen    = addrlen,
 381                 .msg_flags      = XS_SENDMSG_FLAGS | (more ? MSG_MORE : 0),
 382         };
 383         struct kvec iov = {
 384                 .iov_base       = vec->iov_base + base,
 385                 .iov_len        = vec->iov_len - base,
 386         };
 387
 388         if (iov.iov_len != 0)
 389                 return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
 390         return kernel_sendmsg(sock, &msg, NULL, 0, 0);
 391 }
 392
 393 static int xs_send_pagedata(struct socket *sock, struct xdr_buf *xdr, unsigned int base, int more)
 394 {
 395         struct page **ppage;
 396         unsigned int remainder;
 397         int err, sent = 0;
 398
 399         remainder = xdr->page_len - base;
 400         base += xdr->page_base;
 401         ppage = xdr->pages + (base >> PAGE_SHIFT);
 402         base &= ~PAGE_MASK;
 403         for(;;) {
 404                 unsigned int len = min_t(unsigned int, PAGE_SIZE - base, remainder);
 405                 int flags = XS_SENDMSG_FLAGS;
 406
 407                 remainder -= len;
 408                 if (remainder != 0 || more)
 409                         flags |= MSG_MORE;
 410                 err = sock->ops->sendpage(sock, *ppage, base, len, flags);
 411                 if (remainder == 0 || err != len)
 412                         break;
 413                 sent += err;
 414                 ppage++;
 415                 base = 0;
 416         }
 417         if (sent == 0)
 418                 return err;
 419         if (err > 0)
 420                 sent += err;
 421         return sent;
 422 }
 423
 424 /**
 425  * xs_sendpages - write pages directly to a socket
 426  * @sock: socket to send on
 427  * @addr: UDP only -- address of destination
 428  * @addrlen: UDP only -- length of destination address
 429  * @xdr: buffer containing this request
 430  * @base: starting position in the buffer
 431  *
 432  */
 433 static int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base)
 434 {
 435         unsigned int remainder = xdr->len - base;
 436         int err, sent = 0;
 437
 438         if (unlikely(!sock))
 439                 return -ENOTSOCK;
 440
 441         clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags);
 442         if (base != 0) {
 443                 addr = NULL;
 444                 addrlen = 0;
 445         }
 446
 447         if (base < xdr->head[0].iov_len || addr != NULL) {
 448                 unsigned int len = xdr->head[0].iov_len - base;
 449                 remainder -= len;
 450                 err = xs_send_kvec(sock, addr, addrlen, &xdr->head[0], base, remainder != 0);
 451                 if (remainder == 0 || err != len)
 452                         goto out;
 453                 sent += err;
 454                 base = 0;
 455         } else
 456                 base -= xdr->head[0].iov_len;
 457
 458         if (base < xdr->page_len) {
 459                 unsigned int len = xdr->page_len - base;
 460                 remainder -= len;
 461                 err = xs_send_pagedata(sock, xdr, base, remainder != 0);
 462                 if (remainder == 0 || err != len)
 463                         goto out;
 464                 sent += err;
 465                 base = 0;
 466         } else
 467                 base -= xdr->page_len;
 468
 469         if (base >= xdr->tail[0].iov_len)
 470                 return sent;
 471         err = xs_send_kvec(sock, NULL, 0, &xdr->tail[0], base, 0);
 472 out:
 473         if (sent == 0)
 474                 return err;
 475         if (err > 0)
 476                 sent += err;
 477         return sent;
 478 }
 479
 480 static void xs_nospace_callback(struct rpc_task *task)
 481 {
 482         struct sock_xprt *transport = container_of(task->tk_rqstp->rq_xprt, struct sock_xprt, xprt);
 483
 484         transport->inet->sk_write_pending--;
 485         clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 486 }
 487
 488 /**
 489  * xs_nospace - place task on wait queue if transmit was incomplete
 490  * @task: task to put to sleep
 491  *
 492  */
 493 static int xs_nospace(struct rpc_task *task)
 494 {
 495         struct rpc_rqst *req = task->tk_rqstp;
 496         struct rpc_xprt *xprt = req->rq_xprt;
 497         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 498         int ret = -EAGAIN;
 499
 500         dprintk("RPC: %5u xmit incomplete (%u left of %u)\n",
 501                         task->tk_pid, req->rq_slen - req->rq_bytes_sent,
 502                         req->rq_slen);
 503
 504         /* Protect against races with write_space */
 505         spin_lock_bh(&xprt->transport_lock);
 506
 507         /* Don't race with disconnect */
 508         if (xprt_connected(xprt)) {
 509                 if (test_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags)) {
 510                         /*
 511                          * Notify TCP that we're limited by the application
 512                          * window size
 513                          */
 514                         set_bit(SOCK_NOSPACE, &transport->sock->flags);
 515                         transport->inet->sk_write_pending++;
 516                         /* ...and wait for more buffer space */
 517                         xprt_wait_for_buffer_space(task, xs_nospace_callback);
 518                 }
 519         } else {
 520                 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 521                 ret = -ENOTCONN;
 522         }
 523
 524         spin_unlock_bh(&xprt->transport_lock);
 525         return ret;
 526 }
 527
 528 /*
 529  * Construct a stream transport record marker in @buf.
 530  */
 531 static inline void xs_encode_stream_record_marker(struct xdr_buf *buf)
 532 {
 533         u32 reclen = buf->len - sizeof(rpc_fraghdr);
 534         rpc_fraghdr *base = buf->head[0].iov_base;
 535         *base = cpu_to_be32(RPC_LAST_STREAM_FRAGMENT | reclen);
 536 }
 537
 538 /**
 539  * xs_local_send_request - write an RPC request to an AF_LOCAL socket
 540  * @task: RPC task that manages the state of an RPC request
 541  *
 542  * Return values:
 543  *        0:    The request has been sent
 544  *   EAGAIN:    The socket was blocked, please call again later to
 545  *              complete the request
 546  * ENOTCONN:    Caller needs to invoke connect logic then call again
 547  *    other:    Some other error occured, the request was not sent
 548  */
 549 static int xs_local_send_request(struct rpc_task *task)
 550 {
 551         struct rpc_rqst *req = task->tk_rqstp;
 552         struct rpc_xprt *xprt = req->rq_xprt;
 553         struct sock_xprt *transport =
 554                                 container_of(xprt, struct sock_xprt, xprt);
 555         struct xdr_buf *xdr = &req->rq_snd_buf;
 556         int status;
 557
 558         xs_encode_stream_record_marker(&req->rq_snd_buf);
 559
 560         xs_pktdump("packet data:",
 561                         req->rq_svec->iov_base, req->rq_svec->iov_len);
 562
 563         status = xs_sendpages(transport->sock, NULL, 0,
 564                                                 xdr, req->rq_bytes_sent);
 565         dprintk("RPC:       %s(%u) = %d\n",
 566                         __func__, xdr->len - req->rq_bytes_sent, status);
 567         if (likely(status >= 0)) {
 568                 req->rq_bytes_sent += status;
 569                 req->rq_xmit_bytes_sent += status;
 570                 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
 571                         req->rq_bytes_sent = 0;
 572                         return 0;
 573                 }
 574                 status = -EAGAIN;
 575         }
 576
 577         switch (status) {
 578         case -EAGAIN:
 579                 status = xs_nospace(task);
 580                 break;
 581         default:
 582                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
 583                         -status);
 584         case -EPIPE:
 585                 xs_close(xprt);
 586                 status = -ENOTCONN;
 587         }
 588
 589         return status;
 590 }
 591
 592 /**
 593  * xs_udp_send_request - write an RPC request to a UDP socket
 594  * @task: address of RPC task that manages the state of an RPC request
 595  *
 596  * Return values:
 597  *        0:    The request has been sent
 598  *   EAGAIN:    The socket was blocked, please call again later to
 599  *              complete the request
 600  * ENOTCONN:    Caller needs to invoke connect logic then call again
 601  *    other:    Some other error occurred, the request was not sent
 602  */
 603 static int xs_udp_send_request(struct rpc_task *task)
 604 {
 605         struct rpc_rqst *req = task->tk_rqstp;
 606         struct rpc_xprt *xprt = req->rq_xprt;
 607         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 608         struct xdr_buf *xdr = &req->rq_snd_buf;
 609         int status;
 610
 611         xs_pktdump("packet data:",
 612                                 req->rq_svec->iov_base,
 613                                 req->rq_svec->iov_len);
 614
 615         if (!xprt_bound(xprt))
 616                 return -ENOTCONN;
 617         status = xs_sendpages(transport->sock,
 618                               xs_addr(xprt),
 619                               xprt->addrlen, xdr,
 620                               req->rq_bytes_sent);
 621
 622         dprintk("RPC:       xs_udp_send_request(%u) = %d\n",
 623                         xdr->len - req->rq_bytes_sent, status);
 624
 625         if (status >= 0) {
 626                 req->rq_xmit_bytes_sent += status;
 627                 if (status >= req->rq_slen)
 628                         return 0;
 629                 /* Still some bytes left; set up for a retry later. */
 630                 status = -EAGAIN;
 631         }
 632
 633         switch (status) {
 634         case -ENOTSOCK:
 635                 status = -ENOTCONN;
 636                 /* Should we call xs_close() here? */
 637                 break;
 638         case -EAGAIN:
 639                 status = xs_nospace(task);
 640                 break;
 641         default:
 642                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
 643                         -status);
 644         case -ENETUNREACH:
 645         case -EPIPE:
 646         case -ECONNREFUSED:
 647                 /* When the server has died, an ICMP port unreachable message
 648                  * prompts ECONNREFUSED. */
 649                 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 650         }
 651
 652         return status;
 653 }
 654
 655 /**
 656  * xs_tcp_shutdown - gracefully shut down a TCP socket
 657  * @xprt: transport
 658  *
 659  * Initiates a graceful shutdown of the TCP socket by calling the
 660  * equivalent of shutdown(SHUT_WR);
 661  */
 662 static void xs_tcp_shutdown(struct rpc_xprt *xprt)
 663 {
 664         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 665         struct socket *sock = transport->sock;
 666
 667         if (sock != NULL)
 668                 kernel_sock_shutdown(sock, SHUT_WR);
 669 }
 670
 671 /**
 672  * xs_tcp_send_request - write an RPC request to a TCP socket
 673  * @task: address of RPC task that manages the state of an RPC request
 674  *
 675  * Return values:
 676  *        0:    The request has been sent
 677  *   EAGAIN:    The socket was blocked, please call again later to
 678  *              complete the request
 679  * ENOTCONN:    Caller needs to invoke connect logic then call again
 680  *    other:    Some other error occurred, the request was not sent
 681  *
 682  * XXX: In the case of soft timeouts, should we eventually give up
 683  *      if sendmsg is not able to make progress?
 684  */
 685 static int xs_tcp_send_request(struct rpc_task *task)
 686 {
 687         struct rpc_rqst *req = task->tk_rqstp;
 688         struct rpc_xprt *xprt = req->rq_xprt;
 689         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 690         struct xdr_buf *xdr = &req->rq_snd_buf;
 691         int status;
 692
 693         xs_encode_stream_record_marker(&req->rq_snd_buf);
 694
 695         xs_pktdump("packet data:",
 696                                 req->rq_svec->iov_base,
 697                                 req->rq_svec->iov_len);
 698
 699         /* Continue transmitting the packet/record. We must be careful
 700          * to cope with writespace callbacks arriving _after_ we have
 701          * called sendmsg(). */
 702         while (1) {
 703                 status = xs_sendpages(transport->sock,
 704                                         NULL, 0, xdr, req->rq_bytes_sent);
 705
 706                 dprintk("RPC:       xs_tcp_send_request(%u) = %d\n",
 707                                 xdr->len - req->rq_bytes_sent, status);
 708
 709                 if (unlikely(status < 0))
 710                         break;
 711
 712                 /* If we've sent the entire packet, immediately
 713                  * reset the count of bytes sent. */
 714                 req->rq_bytes_sent += status;
 715                 req->rq_xmit_bytes_sent += status;
 716                 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
 717                         req->rq_bytes_sent = 0;
 718                         return 0;
 719                 }
 720
 721                 if (status != 0)
 722                         continue;
 723                 status = -EAGAIN;
 724                 break;
 725         }
 726
 727         switch (status) {
 728         case -ENOTSOCK:
 729                 status = -ENOTCONN;
 730                 /* Should we call xs_close() here? */
 731                 break;
 732         case -EAGAIN:
 733                 status = xs_nospace(task);
 734                 break;
 735         default:
 736                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
 737                         -status);
 738         case -ECONNRESET:
 739                 xs_tcp_shutdown(xprt);
 740         case -ECONNREFUSED:
 741         case -ENOTCONN:
 742         case -EPIPE:
 743                 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 744         }
 745
 746         return status;
 747 }
 748
 749 /**
 750  * xs_tcp_release_xprt - clean up after a tcp transmission
 751  * @xprt: transport
 752  * @task: rpc task
 753  *
 754  * This cleans up if an error causes us to abort the transmission of a request.
 755  * In this case, the socket may need to be reset in order to avoid confusing
 756  * the server.
 757  */
 758 static void xs_tcp_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
 759 {
 760         struct rpc_rqst *req;
 761
 762         if (task != xprt->snd_task)
 763                 return;
 764         if (task == NULL)
 765                 goto out_release;
 766         req = task->tk_rqstp;
 767         if (req == NULL)
 768                 goto out_release;
 769         if (req->rq_bytes_sent == 0)
 770                 goto out_release;
 771         if (req->rq_bytes_sent == req->rq_snd_buf.len)
 772                 goto out_release;
 773         set_bit(XPRT_CLOSE_WAIT, &xprt->state);
 774 out_release:
 775         xprt_release_xprt(xprt, task);
 776 }
 777
 778 static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk)
 779 {
 780         transport->old_data_ready = sk->sk_data_ready;
 781         transport->old_state_change = sk->sk_state_change;
 782         transport->old_write_space = sk->sk_write_space;
 783 }
 784
 785 static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk)
 786 {
 787         sk->sk_data_ready = transport->old_data_ready;
 788         sk->sk_state_change = transport->old_state_change;
 789         sk->sk_write_space = transport->old_write_space;
 790 }
 791
 792 static void xs_reset_transport(struct sock_xprt *transport)
 793 {
 794         struct socket *sock = transport->sock;
 795         struct sock *sk = transport->inet;
 796
 797         if (sk == NULL)
 798                 return;
 799
 800         transport->srcport = 0;
 801
 802         write_lock_bh(&sk->sk_callback_lock);
 803         transport->inet = NULL;
 804         transport->sock = NULL;
 805
 806         sk->sk_user_data = NULL;
 807
 808         xs_restore_old_callbacks(transport, sk);
 809         write_unlock_bh(&sk->sk_callback_lock);
 810
 811         sk->sk_no_check = 0;
 812
 813         sock_release(sock);
 814 }
 815
 816 /**
 817  * xs_close - close a socket
 818  * @xprt: transport
 819  *
 820  * This is used when all requests are complete; ie, no DRC state remains
 821  * on the server we want to save.
 822  *
 823  * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with
 824  * xs_reset_transport() zeroing the socket from underneath a writer.
 825  */
 826 static void xs_close(struct rpc_xprt *xprt)
 827 {
 828         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 829
 830         dprintk("RPC:       xs_close xprt %p\n", xprt);
 831
 832         xs_reset_transport(transport);
 833         xprt->reestablish_timeout = 0;
 834
 835         smp_mb__before_clear_bit();
 836         clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
 837         clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
 838         clear_bit(XPRT_CLOSING, &xprt->state);
 839         smp_mb__after_clear_bit();
 840         xprt_disconnect_done(xprt);
 841 }
 842
 843 static void xs_tcp_close(struct rpc_xprt *xprt)
 844 {
 845         if (test_and_clear_bit(XPRT_CONNECTION_CLOSE, &xprt->state))
 846                 xs_close(xprt);
 847         else
 848                 xs_tcp_shutdown(xprt);
 849 }
 850
 851 /**
 852  * xs_destroy - prepare to shutdown a transport
 853  * @xprt: doomed transport
 854  *
 855  */
 856 static void xs_destroy(struct rpc_xprt *xprt)
 857 {
 858         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 859
 860         dprintk("RPC:       xs_destroy xprt %p\n", xprt);
 861
 862         cancel_delayed_work_sync(&transport->connect_worker);
 863
 864         xs_close(xprt);
 865         xs_free_peer_addresses(xprt);
 866         xprt_free(xprt);
 867         module_put(THIS_MODULE);
 868 }
 869
 870 static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
 871 {
 872         return (struct rpc_xprt *) sk->sk_user_data;
 873 }
 874
 875 static int xs_local_copy_to_xdr(struct xdr_buf *xdr, struct sk_buff *skb)
 876 {
 877         struct xdr_skb_reader desc = {
 878                 .skb            = skb,
 879                 .offset         = sizeof(rpc_fraghdr),
 880                 .count          = skb->len - sizeof(rpc_fraghdr),
 881         };
 882
 883         if (xdr_partial_copy_from_skb(xdr, 0, &desc, xdr_skb_read_bits) < 0)
 884                 return -1;
 885         if (desc.count)
 886                 return -1;
 887         return 0;
 888 }
 889
 890 /**
 891  * xs_local_data_ready - "data ready" callback for AF_LOCAL sockets
 892  * @sk: socket with data to read
 893  * @len: how much data to read
 894  *
 895  * Currently this assumes we can read the whole reply in a single gulp.
 896  */
 897 static void xs_local_data_ready(struct sock *sk, int len)
 898 {
 899         struct rpc_task *task;
 900         struct rpc_xprt *xprt;
 901         struct rpc_rqst *rovr;
 902         struct sk_buff *skb;
 903         int err, repsize, copied;
 904         u32 _xid;
 905         __be32 *xp;
 906
 907         read_lock_bh(&sk->sk_callback_lock);
 908         dprintk("RPC:       %s...\n", __func__);
 909         xprt = xprt_from_sock(sk);
 910         if (xprt == NULL)
 911                 goto out;
 912
 913         skb = skb_recv_datagram(sk, 0, 1, &err);
 914         if (skb == NULL)
 915                 goto out;
 916
 917         repsize = skb->len - sizeof(rpc_fraghdr);
 918         if (repsize < 4) {
 919                 dprintk("RPC:       impossible RPC reply size %d\n", repsize);
 920                 goto dropit;
 921         }
 922
 923         /* Copy the XID from the skb... */
 924         xp = skb_header_pointer(skb, sizeof(rpc_fraghdr), sizeof(_xid), &_xid);
 925         if (xp == NULL)
 926                 goto dropit;
 927
 928         /* Look up and lock the request corresponding to the given XID */
 929         spin_lock(&xprt->transport_lock);
 930         rovr = xprt_lookup_rqst(xprt, *xp);
 931         if (!rovr)
 932                 goto out_unlock;
 933         task = rovr->rq_task;
 934
 935         copied = rovr->rq_private_buf.buflen;
 936         if (copied > repsize)
 937                 copied = repsize;
 938
 939         if (xs_local_copy_to_xdr(&rovr->rq_private_buf, skb)) {
 940                 dprintk("RPC:       sk_buff copy failed\n");
 941                 goto out_unlock;
 942         }
 943
 944         xprt_complete_rqst(task, copied);
 945
 946  out_unlock:
 947         spin_unlock(&xprt->transport_lock);
 948  dropit:
 949         skb_free_datagram(sk, skb);
 950  out:
 951         read_unlock_bh(&sk->sk_callback_lock);
 952 }
 953
 954 /**
 955  * xs_udp_data_ready - "data ready" callback for UDP sockets
 956  * @sk: socket with data to read
 957  * @len: how much data to read
 958  *
 959  */
 960 static void xs_udp_data_ready(struct sock *sk, int len)
 961 {
 962         struct rpc_task *task;
 963         struct rpc_xprt *xprt;
 964         struct rpc_rqst *rovr;
 965         struct sk_buff *skb;
 966         int err, repsize, copied;
 967         u32 _xid;
 968         __be32 *xp;
 969
 970         read_lock_bh(&sk->sk_callback_lock);
 971         dprintk("RPC:       xs_udp_data_ready...\n");
 972         if (!(xprt = xprt_from_sock(sk)))
 973                 goto out;
 974
 975         if ((skb = skb_recv_datagram(sk, 0, 1, &err)) == NULL)
 976                 goto out;
 977
 978         repsize = skb->len - sizeof(struct udphdr);
 979         if (repsize < 4) {
 980                 dprintk("RPC:       impossible RPC reply size %d!\n", repsize);
 981                 goto dropit;
 982         }
 983
 984         /* Copy the XID from the skb... */
 985         xp = skb_header_pointer(skb, sizeof(struct udphdr),
 986                                 sizeof(_xid), &_xid);
 987         if (xp == NULL)
 988                 goto dropit;
 989
 990         /* Look up and lock the request corresponding to the given XID */
 991         spin_lock(&xprt->transport_lock);
 992         rovr = xprt_lookup_rqst(xprt, *xp);
 993         if (!rovr)
 994                 goto out_unlock;
 995         task = rovr->rq_task;
 996
 997         if ((copied = rovr->rq_private_buf.buflen) > repsize)
 998                 copied = repsize;
 999
1000         /* Suck it into the iovec, verify checksum if not done by hw. */
1001         if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
1002                 UDPX_INC_STATS_BH(sk, UDP_MIB_INERRORS);
1003                 goto out_unlock;
1004         }
1005
1006         UDPX_INC_STATS_BH(sk, UDP_MIB_INDATAGRAMS);
1007
1008         xprt_adjust_cwnd(xprt, task, copied);
1009         xprt_complete_rqst(task, copied);
1010
1011  out_unlock:
1012         spin_unlock(&xprt->transport_lock);
1013  dropit:
1014         skb_free_datagram(sk, skb);
1015  out:
1016         read_unlock_bh(&sk->sk_callback_lock);
1017 }
1018
1019 /*
1020  * Helper function to force a TCP close if the server is sending
1021  * junk and/or it has put us in CLOSE_WAIT
1022  */
1023 static void xs_tcp_force_close(struct rpc_xprt *xprt)
1024 {
1025         set_bit(XPRT_CONNECTION_CLOSE, &xprt->state);
1026         xprt_force_disconnect(xprt);
1027 }
1028
1029 static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
1030 {
1031         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1032         size_t len, used;
1033         char *p;
1034
1035         p = ((char *) &transport->tcp_fraghdr) + transport->tcp_offset;
1036         len = sizeof(transport->tcp_fraghdr) - transport->tcp_offset;
1037         used = xdr_skb_read_bits(desc, p, len);
1038         transport->tcp_offset += used;
1039         if (used != len)
1040                 return;
1041
1042         transport->tcp_reclen = ntohl(transport->tcp_fraghdr);
1043         if (transport->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
1044                 transport->tcp_flags |= TCP_RCV_LAST_FRAG;
1045         else
1046                 transport->tcp_flags &= ~TCP_RCV_LAST_FRAG;
1047         transport->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
1048
1049         transport->tcp_flags &= ~TCP_RCV_COPY_FRAGHDR;
1050         transport->tcp_offset = 0;
1051
1052         /* Sanity check of the record length */
1053         if (unlikely(transport->tcp_reclen < 8)) {
1054                 dprintk("RPC:       invalid TCP record fragment length\n");
1055                 xs_tcp_force_close(xprt);
1056                 return;
1057         }
1058         dprintk("RPC:       reading TCP record fragment of length %d\n",
1059                         transport->tcp_reclen);
1060 }
1061
1062 static void xs_tcp_check_fraghdr(struct sock_xprt *transport)
1063 {
1064         if (transport->tcp_offset == transport->tcp_reclen) {
1065                 transport->tcp_flags |= TCP_RCV_COPY_FRAGHDR;
1066                 transport->tcp_offset = 0;
1067                 if (transport->tcp_flags & TCP_RCV_LAST_FRAG) {
1068                         transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1069                         transport->tcp_flags |= TCP_RCV_COPY_XID;
1070                         transport->tcp_copied = 0;
1071                 }
1072         }
1073 }
1074
1075 static inline void xs_tcp_read_xid(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1076 {
1077         size_t len, used;
1078         char *p;
1079
1080         len = sizeof(transport->tcp_xid) - transport->tcp_offset;
1081         dprintk("RPC:       reading XID (%Zu bytes)\n", len);
1082         p = ((char *) &transport->tcp_xid) + transport->tcp_offset;
1083         used = xdr_skb_read_bits(desc, p, len);
1084         transport->tcp_offset += used;
1085         if (used != len)
1086                 return;
1087         transport->tcp_flags &= ~TCP_RCV_COPY_XID;
1088         transport->tcp_flags |= TCP_RCV_READ_CALLDIR;
1089         transport->tcp_copied = 4;
1090         dprintk("RPC:       reading %s XID %08x\n",
1091                         (transport->tcp_flags & TCP_RPC_REPLY) ? "reply for"
1092                                                               : "request with",
1093                         ntohl(transport->tcp_xid));
1094         xs_tcp_check_fraghdr(transport);
1095 }
1096
1097 static inline void xs_tcp_read_calldir(struct sock_xprt *transport,
1098                                        struct xdr_skb_reader *desc)
1099 {
1100         size_t len, used;
1101         u32 offset;
1102         char *p;
1103
1104         /*
1105          * We want transport->tcp_offset to be 8 at the end of this routine
1106          * (4 bytes for the xid and 4 bytes for the call/reply flag).
1107          * When this function is called for the first time,
1108          * transport->tcp_offset is 4 (after having already read the xid).
1109          */
1110         offset = transport->tcp_offset - sizeof(transport->tcp_xid);
1111         len = sizeof(transport->tcp_calldir) - offset;
1112         dprintk("RPC:       reading CALL/REPLY flag (%Zu bytes)\n", len);
1113         p = ((char *) &transport->tcp_calldir) + offset;
1114         used = xdr_skb_read_bits(desc, p, len);
1115         transport->tcp_offset += used;
1116         if (used != len)
1117                 return;
1118         transport->tcp_flags &= ~TCP_RCV_READ_CALLDIR;
1119         /*
1120          * We don't yet have the XDR buffer, so we will write the calldir
1121          * out after we get the buffer from the 'struct rpc_rqst'
1122          */
1123         switch (ntohl(transport->tcp_calldir)) {
1124         case RPC_REPLY:
1125                 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1126                 transport->tcp_flags |= TCP_RCV_COPY_DATA;
1127                 transport->tcp_flags |= TCP_RPC_REPLY;
1128                 break;
1129         case RPC_CALL:
1130                 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1131                 transport->tcp_flags |= TCP_RCV_COPY_DATA;
1132                 transport->tcp_flags &= ~TCP_RPC_REPLY;
1133                 break;
1134         default:
1135                 dprintk("RPC:       invalid request message type\n");
1136                 xs_tcp_force_close(&transport->xprt);
1137         }
1138         xs_tcp_check_fraghdr(transport);
1139 }
1140
1141 static inline void xs_tcp_read_common(struct rpc_xprt *xprt,
1142                                      struct xdr_skb_reader *desc,
1143                                      struct rpc_rqst *req)
1144 {
1145         struct sock_xprt *transport =
1146                                 container_of(xprt, struct sock_xprt, xprt);
1147         struct xdr_buf *rcvbuf;
1148         size_t len;
1149         ssize_t r;
1150
1151         rcvbuf = &req->rq_private_buf;
1152
1153         if (transport->tcp_flags & TCP_RCV_COPY_CALLDIR) {
1154                 /*
1155                  * Save the RPC direction in the XDR buffer
1156                  */
1157                 memcpy(rcvbuf->head[0].iov_base + transport->tcp_copied,
1158                         &transport->tcp_calldir,
1159                         sizeof(transport->tcp_calldir));
1160                 transport->tcp_copied += sizeof(transport->tcp_calldir);
1161                 transport->tcp_flags &= ~TCP_RCV_COPY_CALLDIR;
1162         }
1163
1164         len = desc->count;
1165         if (len > transport->tcp_reclen - transport->tcp_offset) {
1166                 struct xdr_skb_reader my_desc;
1167
1168                 len = transport->tcp_reclen - transport->tcp_offset;
1169                 memcpy(&my_desc, desc, sizeof(my_desc));
1170                 my_desc.count = len;
1171                 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
1172                                           &my_desc, xdr_skb_read_bits);
1173                 desc->count -= r;
1174                 desc->offset += r;
1175         } else
1176                 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
1177                                           desc, xdr_skb_read_bits);
1178
1179         if (r > 0) {
1180                 transport->tcp_copied += r;
1181                 transport->tcp_offset += r;
1182         }
1183         if (r != len) {
1184                 /* Error when copying to the receive buffer,
1185                  * usually because we weren't able to allocate
1186                  * additional buffer pages. All we can do now
1187                  * is turn off TCP_RCV_COPY_DATA, so the request
1188                  * will not receive any additional updates,
1189                  * and time out.
1190                  * Any remaining data from this record will
1191                  * be discarded.
1192                  */
1193                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1194                 dprintk("RPC:       XID %08x truncated request\n",
1195                                 ntohl(transport->tcp_xid));
1196                 dprintk("RPC:       xprt = %p, tcp_copied = %lu, "
1197                                 "tcp_offset = %u, tcp_reclen = %u\n",
1198                                 xprt, transport->tcp_copied,
1199                                 transport->tcp_offset, transport->tcp_reclen);
1200                 return;
1201         }
1202
1203         dprintk("RPC:       XID %08x read %Zd bytes\n",
1204                         ntohl(transport->tcp_xid), r);
1205         dprintk("RPC:       xprt = %p, tcp_copied = %lu, tcp_offset = %u, "
1206                         "tcp_reclen = %u\n", xprt, transport->tcp_copied,
1207                         transport->tcp_offset, transport->tcp_reclen);
1208
1209         if (transport->tcp_copied == req->rq_private_buf.buflen)
1210                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1211         else if (transport->tcp_offset == transport->tcp_reclen) {
1212                 if (transport->tcp_flags & TCP_RCV_LAST_FRAG)
1213                         transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1214         }
1215 }
1216
1217 /*
1218  * Finds the request corresponding to the RPC xid and invokes the common
1219  * tcp read code to read the data.
1220  */
1221 static inline int xs_tcp_read_reply(struct rpc_xprt *xprt,
1222                                     struct xdr_skb_reader *desc)
1223 {
1224         struct sock_xprt *transport =
1225                                 container_of(xprt, struct sock_xprt, xprt);
1226         struct rpc_rqst *req;
1227
1228         dprintk("RPC:       read reply XID %08x\n", ntohl(transport->tcp_xid));
1229
1230         /* Find and lock the request corresponding to this xid */
1231         spin_lock(&xprt->transport_lock);
1232         req = xprt_lookup_rqst(xprt, transport->tcp_xid);
1233         if (!req) {
1234                 dprintk("RPC:       XID %08x request not found!\n",
1235                                 ntohl(transport->tcp_xid));
1236                 spin_unlock(&xprt->transport_lock);
1237                 return -1;
1238         }
1239
1240         xs_tcp_read_common(xprt, desc, req);
1241
1242         if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1243                 xprt_complete_rqst(req->rq_task, transport->tcp_copied);
1244
1245         spin_unlock(&xprt->transport_lock);
1246         return 0;
1247 }
1248
1249 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1250 /*
1251  * Obtains an rpc_rqst previously allocated and invokes the common
1252  * tcp read code to read the data.  The result is placed in the callback
1253  * queue.
1254  * If we're unable to obtain the rpc_rqst we schedule the closing of the
1255  * connection and return -1.
1256  */
1257 static inline int xs_tcp_read_callback(struct rpc_xprt *xprt,
1258                                        struct xdr_skb_reader *desc)
1259 {
1260         struct sock_xprt *transport =
1261                                 container_of(xprt, struct sock_xprt, xprt);
1262         struct rpc_rqst *req;
1263
1264         req = xprt_alloc_bc_request(xprt);
1265         if (req == NULL) {
1266                 printk(KERN_WARNING "Callback slot table overflowed\n");
1267                 xprt_force_disconnect(xprt);
1268                 return -1;
1269         }
1270
1271         req->rq_xid = transport->tcp_xid;
1272         dprintk("RPC:       read callback  XID %08x\n", ntohl(req->rq_xid));
1273         xs_tcp_read_common(xprt, desc, req);
1274
1275         if (!(transport->tcp_flags & TCP_RCV_COPY_DATA)) {
1276                 struct svc_serv *bc_serv = xprt->bc_serv;
1277
1278                 /*
1279                  * Add callback request to callback list.  The callback
1280                  * service sleeps on the sv_cb_waitq waiting for new
1281                  * requests.  Wake it up after adding enqueing the
1282                  * request.
1283                  */
1284                 dprintk("RPC:       add callback request to list\n");
1285                 spin_lock(&bc_serv->sv_cb_lock);
1286                 list_add(&req->rq_bc_list, &bc_serv->sv_cb_list);
1287                 spin_unlock(&bc_serv->sv_cb_lock);
1288                 wake_up(&bc_serv->sv_cb_waitq);
1289         }
1290
1291         req->rq_private_buf.len = transport->tcp_copied;
1292
1293         return 0;
1294 }
1295
1296 static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1297                                         struct xdr_skb_reader *desc)
1298 {
1299         struct sock_xprt *transport =
1300                                 container_of(xprt, struct sock_xprt, xprt);
1301
1302         return (transport->tcp_flags & TCP_RPC_REPLY) ?
1303                 xs_tcp_read_reply(xprt, desc) :
1304                 xs_tcp_read_callback(xprt, desc);
1305 }
1306 #else
1307 static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1308                                         struct xdr_skb_reader *desc)
1309 {
1310         return xs_tcp_read_reply(xprt, desc);
1311 }
1312 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1313
1314 /*
1315  * Read data off the transport.  This can be either an RPC_CALL or an
1316  * RPC_REPLY.  Relay the processing to helper functions.
1317  */
1318 static void xs_tcp_read_data(struct rpc_xprt *xprt,
1319                                     struct xdr_skb_reader *desc)
1320 {
1321         struct sock_xprt *transport =
1322                                 container_of(xprt, struct sock_xprt, xprt);
1323
1324         if (_xs_tcp_read_data(xprt, desc) == 0)
1325                 xs_tcp_check_fraghdr(transport);
1326         else {
1327                 /*
1328                  * The transport_lock protects the request handling.
1329                  * There's no need to hold it to update the tcp_flags.
1330                  */
1331                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1332         }
1333 }
1334
1335 static inline void xs_tcp_read_discard(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1336 {
1337         size_t len;
1338
1339         len = transport->tcp_reclen - transport->tcp_offset;
1340         if (len > desc->count)
1341                 len = desc->count;
1342         desc->count -= len;
1343         desc->offset += len;
1344         transport->tcp_offset += len;
1345         dprintk("RPC:       discarded %Zu bytes\n", len);
1346         xs_tcp_check_fraghdr(transport);
1347 }
1348
1349 static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
1350 {
1351         struct rpc_xprt *xprt = rd_desc->arg.data;
1352         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1353         struct xdr_skb_reader desc = {
1354                 .skb    = skb,
1355                 .offset = offset,
1356                 .count  = len,
1357         };
1358
1359         dprintk("RPC:       xs_tcp_data_recv started\n");
1360         do {
1361                 /* Read in a new fragment marker if necessary */
1362                 /* Can we ever really expect to get completely empty fragments? */
1363                 if (transport->tcp_flags & TCP_RCV_COPY_FRAGHDR) {
1364                         xs_tcp_read_fraghdr(xprt, &desc);
1365                         continue;
1366                 }
1367                 /* Read in the xid if necessary */
1368                 if (transport->tcp_flags & TCP_RCV_COPY_XID) {
1369                         xs_tcp_read_xid(transport, &desc);
1370                         continue;
1371                 }
1372                 /* Read in the call/reply flag */
1373                 if (transport->tcp_flags & TCP_RCV_READ_CALLDIR) {
1374                         xs_tcp_read_calldir(transport, &desc);
1375                         continue;
1376                 }
1377                 /* Read in the request data */
1378                 if (transport->tcp_flags & TCP_RCV_COPY_DATA) {
1379                         xs_tcp_read_data(xprt, &desc);
1380                         continue;
1381                 }
1382                 /* Skip over any trailing bytes on short reads */
1383                 xs_tcp_read_discard(transport, &desc);
1384         } while (desc.count);
1385         dprintk("RPC:       xs_tcp_data_recv done\n");
1386         return len - desc.count;
1387 }
1388
1389 /**
1390  * xs_tcp_data_ready - "data ready" callback for TCP sockets
1391  * @sk: socket with data to read
1392  * @bytes: how much data to read
1393  *
1394  */
1395 static void xs_tcp_data_ready(struct sock *sk, int bytes)
1396 {
1397         struct rpc_xprt *xprt;
1398         read_descriptor_t rd_desc;
1399         int read;
1400
1401         dprintk("RPC:       xs_tcp_data_ready...\n");
1402
1403         read_lock_bh(&sk->sk_callback_lock);
1404         if (!(xprt = xprt_from_sock(sk)))
1405                 goto out;
1406         /* Any data means we had a useful conversation, so
1407          * the we don't need to delay the next reconnect
1408          */
1409         if (xprt->reestablish_timeout)
1410                 xprt->reestablish_timeout = 0;
1411
1412         /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
1413         rd_desc.arg.data = xprt;
1414         do {
1415                 rd_desc.count = 65536;
1416                 read = tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
1417         } while (read > 0);
1418 out:
1419         read_unlock_bh(&sk->sk_callback_lock);
1420 }
1421
1422 /*
1423  * Do the equivalent of linger/linger2 handling for dealing with
1424  * broken servers that don't close the socket in a timely
1425  * fashion
1426  */
1427 static void xs_tcp_schedule_linger_timeout(struct rpc_xprt *xprt,
1428                 unsigned long timeout)
1429 {
1430         struct sock_xprt *transport;
1431
1432         if (xprt_test_and_set_connecting(xprt))
1433                 return;
1434         set_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1435         transport = container_of(xprt, struct sock_xprt, xprt);
1436         queue_delayed_work(rpciod_workqueue, &transport->connect_worker,
1437                            timeout);
1438 }
1439
1440 static void xs_tcp_cancel_linger_timeout(struct rpc_xprt *xprt)
1441 {
1442         struct sock_xprt *transport;
1443
1444         transport = container_of(xprt, struct sock_xprt, xprt);
1445
1446         if (!test_bit(XPRT_CONNECTION_ABORT, &xprt->state) ||
1447             !cancel_delayed_work(&transport->connect_worker))
1448                 return;
1449         clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1450         xprt_clear_connecting(xprt);
1451 }
1452
1453 static void xs_sock_reset_connection_flags(struct rpc_xprt *xprt)
1454 {
1455         smp_mb__before_clear_bit();
1456         clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1457         clear_bit(XPRT_CONNECTION_CLOSE, &xprt->state);
1458         clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1459         clear_bit(XPRT_CLOSING, &xprt->state);
1460         smp_mb__after_clear_bit();
1461 }
1462
1463 static void xs_sock_mark_closed(struct rpc_xprt *xprt)
1464 {
1465         xs_sock_reset_connection_flags(xprt);
1466         /* Mark transport as closed and wake up all pending tasks */
1467         xprt_disconnect_done(xprt);
1468 }
1469
1470 /**
1471  * xs_tcp_state_change - callback to handle TCP socket state changes
1472  * @sk: socket whose state has changed
1473  *
1474  */
1475 static void xs_tcp_state_change(struct sock *sk)
1476 {
1477         struct rpc_xprt *xprt;
1478
1479         read_lock_bh(&sk->sk_callback_lock);
1480         if (!(xprt = xprt_from_sock(sk)))
1481                 goto out;
1482         dprintk("RPC:       xs_tcp_state_change client %p...\n", xprt);
1483         dprintk("RPC:       state %x conn %d dead %d zapped %d sk_shutdown %d\n",
1484                         sk->sk_state, xprt_connected(xprt),
1485                         sock_flag(sk, SOCK_DEAD),
1486                         sock_flag(sk, SOCK_ZAPPED),
1487                         sk->sk_shutdown);
1488
1489         switch (sk->sk_state) {
1490         case TCP_ESTABLISHED:
1491                 spin_lock(&xprt->transport_lock);
1492                 if (!xprt_test_and_set_connected(xprt)) {
1493                         struct sock_xprt *transport = container_of(xprt,
1494                                         struct sock_xprt, xprt);
1495
1496                         /* Reset TCP record info */
1497                         transport->tcp_offset = 0;
1498                         transport->tcp_reclen = 0;
1499                         transport->tcp_copied = 0;
1500                         transport->tcp_flags =
1501                                 TCP_RCV_COPY_FRAGHDR | TCP_RCV_COPY_XID;
1502
1503                         xprt_wake_pending_tasks(xprt, -EAGAIN);
1504                 }
1505                 spin_unlock(&xprt->transport_lock);
1506                 break;
1507         case TCP_FIN_WAIT1:
1508                 /* The client initiated a shutdown of the socket */
1509                 xprt->connect_cookie++;
1510                 xprt->reestablish_timeout = 0;
1511                 set_bit(XPRT_CLOSING, &xprt->state);
1512                 smp_mb__before_clear_bit();
1513                 clear_bit(XPRT_CONNECTED, &xprt->state);
1514                 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1515                 smp_mb__after_clear_bit();
1516                 xs_tcp_schedule_linger_timeout(xprt, xs_tcp_fin_timeout);
1517                 break;
1518         case TCP_CLOSE_WAIT:
1519                 /* The server initiated a shutdown of the socket */
1520                 xprt->connect_cookie++;
1521                 clear_bit(XPRT_CONNECTED, &xprt->state);
1522                 xs_tcp_force_close(xprt);
1523         case TCP_CLOSING:
1524                 /*
1525                  * If the server closed down the connection, make sure that
1526                  * we back off before reconnecting
1527                  */
1528                 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1529                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1530                 break;
1531         case TCP_LAST_ACK:
1532                 set_bit(XPRT_CLOSING, &xprt->state);
1533                 xs_tcp_schedule_linger_timeout(xprt, xs_tcp_fin_timeout);
1534                 smp_mb__before_clear_bit();
1535                 clear_bit(XPRT_CONNECTED, &xprt->state);
1536                 smp_mb__after_clear_bit();
1537                 break;
1538         case TCP_CLOSE:
1539                 xs_tcp_cancel_linger_timeout(xprt);
1540                 xs_sock_mark_closed(xprt);
1541         }
1542  out:
1543         read_unlock_bh(&sk->sk_callback_lock);
1544 }
1545
1546 static void xs_write_space(struct sock *sk)
1547 {
1548         struct socket *sock;
1549         struct rpc_xprt *xprt;
1550
1551         if (unlikely(!(sock = sk->sk_socket)))
1552                 return;
1553         clear_bit(SOCK_NOSPACE, &sock->flags);
1554
1555         if (unlikely(!(xprt = xprt_from_sock(sk))))
1556                 return;
1557         if (test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags) == 0)
1558                 return;
1559
1560         xprt_write_space(xprt);
1561 }
1562
1563 /**
1564  * xs_udp_write_space - callback invoked when socket buffer space
1565  *                             becomes available
1566  * @sk: socket whose state has changed
1567  *
1568  * Called when more output buffer space is available for this socket.
1569  * We try not to wake our writers until they can make "significant"
1570  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1571  * with a bunch of small requests.
1572  */
1573 static void xs_udp_write_space(struct sock *sk)
1574 {
1575         read_lock_bh(&sk->sk_callback_lock);
1576
1577         /* from net/core/sock.c:sock_def_write_space */
1578         if (sock_writeable(sk))
1579                 xs_write_space(sk);
1580
1581         read_unlock_bh(&sk->sk_callback_lock);
1582 }
1583
1584 /**
1585  * xs_tcp_write_space - callback invoked when socket buffer space
1586  *                             becomes available
1587  * @sk: socket whose state has changed
1588  *
1589  * Called when more output buffer space is available for this socket.
1590  * We try not to wake our writers until they can make "significant"
1591  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1592  * with a bunch of small requests.
1593  */
1594 static void xs_tcp_write_space(struct sock *sk)
1595 {
1596         read_lock_bh(&sk->sk_callback_lock);
1597
1598         /* from net/core/stream.c:sk_stream_write_space */
1599         if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk))
1600                 xs_write_space(sk);
1601
1602         read_unlock_bh(&sk->sk_callback_lock);
1603 }
1604
1605 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
1606 {
1607         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1608         struct sock *sk = transport->inet;
1609
1610         if (transport->rcvsize) {
1611                 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1612                 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
1613         }
1614         if (transport->sndsize) {
1615                 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1616                 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1617                 sk->sk_write_space(sk);
1618         }
1619 }
1620
1621 /**
1622  * xs_udp_set_buffer_size - set send and receive limits
1623  * @xprt: generic transport
1624  * @sndsize: requested size of send buffer, in bytes
1625  * @rcvsize: requested size of receive buffer, in bytes
1626  *
1627  * Set socket send and receive buffer size limits.
1628  */
1629 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1630 {
1631         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1632
1633         transport->sndsize = 0;
1634         if (sndsize)
1635                 transport->sndsize = sndsize + 1024;
1636         transport->rcvsize = 0;
1637         if (rcvsize)
1638                 transport->rcvsize = rcvsize + 1024;
1639
1640         xs_udp_do_set_buffer_size(xprt);
1641 }
1642
1643 /**
1644  * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1645  * @task: task that timed out
1646  *
1647  * Adjust the congestion window after a retransmit timeout has occurred.
1648  */
1649 static void xs_udp_timer(struct rpc_xprt *xprt, struct rpc_task *task)
1650 {
1651         xprt_adjust_cwnd(xprt, task, -ETIMEDOUT);
1652 }
1653
1654 static unsigned short xs_get_random_port(void)
1655 {
1656         unsigned short range = xprt_max_resvport - xprt_min_resvport;
1657         unsigned short rand = (unsigned short) net_random() % range;
1658         return rand + xprt_min_resvport;
1659 }
1660
1661 /**
1662  * xs_set_port - reset the port number in the remote endpoint address
1663  * @xprt: generic transport
1664  * @port: new port number
1665  *
1666  */
1667 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1668 {
1669         dprintk("RPC:       setting port for xprt %p to %u\n", xprt, port);
1670
1671         rpc_set_port(xs_addr(xprt), port);
1672         xs_update_peer_port(xprt);
1673 }
1674
1675 static unsigned short xs_get_srcport(struct sock_xprt *transport)
1676 {
1677         unsigned short port = transport->srcport;
1678
1679         if (port == 0 && transport->xprt.resvport)
1680                 port = xs_get_random_port();
1681         return port;
1682 }
1683
1684 static unsigned short xs_next_srcport(struct sock_xprt *transport, unsigned short port)
1685 {
1686         if (transport->srcport != 0)
1687                 transport->srcport = 0;
1688         if (!transport->xprt.resvport)
1689                 return 0;
1690         if (port <= xprt_min_resvport || port > xprt_max_resvport)
1691                 return xprt_max_resvport;
1692         return --port;
1693 }
1694 static int xs_bind(struct sock_xprt *transport, struct socket *sock)
1695 {
1696         struct sockaddr_storage myaddr;
1697         int err, nloop = 0;
1698         unsigned short port = xs_get_srcport(transport);
1699         unsigned short last;
1700
1701         memcpy(&myaddr, &transport->srcaddr, transport->xprt.addrlen);
1702         do {
1703                 rpc_set_port((struct sockaddr *)&myaddr, port);
1704                 err = kernel_bind(sock, (struct sockaddr *)&myaddr,
1705                                 transport->xprt.addrlen);
1706                 if (port == 0)
1707                         break;
1708                 if (err == 0) {
1709                         transport->srcport = port;
1710                         break;
1711                 }
1712                 last = port;
1713                 port = xs_next_srcport(transport, port);
1714                 if (port > last)
1715                         nloop++;
1716         } while (err == -EADDRINUSE && nloop != 2);
1717
1718         if (myaddr.ss_family == AF_INET)
1719                 dprintk("RPC:       %s %pI4:%u: %s (%d)\n", __func__,
1720                                 &((struct sockaddr_in *)&myaddr)->sin_addr,
1721                                 port, err ? "failed" : "ok", err);
1722         else
1723                 dprintk("RPC:       %s %pI6:%u: %s (%d)\n", __func__,
1724                                 &((struct sockaddr_in6 *)&myaddr)->sin6_addr,
1725                                 port, err ? "failed" : "ok", err);
1726         return err;
1727 }
1728
1729 /*
1730  * We don't support autobind on AF_LOCAL sockets
1731  */
1732 static void xs_local_rpcbind(struct rpc_task *task)
1733 {
1734         rcu_read_lock();
1735         xprt_set_bound(rcu_dereference(task->tk_client->cl_xprt));
1736         rcu_read_unlock();
1737 }
1738
1739 static void xs_local_set_port(struct rpc_xprt *xprt, unsigned short port)
1740 {
1741 }
1742
1743 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1744 static struct lock_class_key xs_key[2];
1745 static struct lock_class_key xs_slock_key[2];
1746
1747 static inline void xs_reclassify_socketu(struct socket *sock)
1748 {
1749         struct sock *sk = sock->sk;
1750
1751         sock_lock_init_class_and_name(sk, "slock-AF_LOCAL-RPC",
1752                 &xs_slock_key[1], "sk_lock-AF_LOCAL-RPC", &xs_key[1]);
1753 }
1754
1755 static inline void xs_reclassify_socket4(struct socket *sock)
1756 {
1757         struct sock *sk = sock->sk;
1758
1759         sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1760                 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1761 }
1762
1763 static inline void xs_reclassify_socket6(struct socket *sock)
1764 {
1765         struct sock *sk = sock->sk;
1766
1767         sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1768                 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
1769 }
1770
1771 static inline void xs_reclassify_socket(int family, struct socket *sock)
1772 {
1773         WARN_ON_ONCE(sock_owned_by_user(sock->sk));
1774         if (sock_owned_by_user(sock->sk))
1775                 return;
1776
1777         switch (family) {
1778         case AF_LOCAL:
1779                 xs_reclassify_socketu(sock);
1780                 break;
1781         case AF_INET:
1782                 xs_reclassify_socket4(sock);
1783                 break;
1784         case AF_INET6:
1785                 xs_reclassify_socket6(sock);
1786                 break;
1787         }
1788 }
1789 #else
1790 static inline void xs_reclassify_socketu(struct socket *sock)
1791 {
1792 }
1793
1794 static inline void xs_reclassify_socket4(struct socket *sock)
1795 {
1796 }
1797
1798 static inline void xs_reclassify_socket6(struct socket *sock)
1799 {
1800 }
1801
1802 static inline void xs_reclassify_socket(int family, struct socket *sock)
1803 {
1804 }
1805 #endif
1806
1807 static struct socket *xs_create_sock(struct rpc_xprt *xprt,
1808                 struct sock_xprt *transport, int family, int type, int protocol)
1809 {
1810         struct socket *sock;
1811         int err;
1812
1813         err = __sock_create(xprt->xprt_net, family, type, protocol, &sock, 1);
1814         if (err < 0) {
1815                 dprintk("RPC:       can't create %d transport socket (%d).\n",
1816                                 protocol, -err);
1817                 goto out;
1818         }
1819         xs_reclassify_socket(family, sock);
1820
1821         err = xs_bind(transport, sock);
1822         if (err) {
1823                 sock_release(sock);
1824                 goto out;
1825         }
1826
1827         return sock;
1828 out:
1829         return ERR_PTR(err);
1830 }
1831
1832 static int xs_local_finish_connecting(struct rpc_xprt *xprt,
1833                                       struct socket *sock)
1834 {
1835         struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1836                                                                         xprt);
1837
1838         if (!transport->inet) {
1839                 struct sock *sk = sock->sk;
1840
1841                 write_lock_bh(&sk->sk_callback_lock);
1842
1843                 xs_save_old_callbacks(transport, sk);
1844
1845                 sk->sk_user_data = xprt;
1846                 sk->sk_data_ready = xs_local_data_ready;
1847                 sk->sk_write_space = xs_udp_write_space;
1848                 sk->sk_allocation = GFP_ATOMIC;
1849
1850                 xprt_clear_connected(xprt);
1851
1852                 /* Reset to new socket */
1853                 transport->sock = sock;
1854                 transport->inet = sk;
1855
1856                 write_unlock_bh(&sk->sk_callback_lock);
1857         }
1858
1859         /* Tell the socket layer to start connecting... */
1860         xprt->stat.connect_count++;
1861         xprt->stat.connect_start = jiffies;
1862         return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, 0);
1863 }
1864
1865 /**
1866  * xs_local_setup_socket - create AF_LOCAL socket, connect to a local endpoint
1867  * @xprt: RPC transport to connect
1868  * @transport: socket transport to connect
1869  * @create_sock: function to create a socket of the correct type
1870  *
1871  * Invoked by a work queue tasklet.
1872  */
1873 static void xs_local_setup_socket(struct work_struct *work)
1874 {
1875         struct sock_xprt *transport =
1876                 container_of(work, struct sock_xprt, connect_worker.work);
1877         struct rpc_xprt *xprt = &transport->xprt;
1878         struct socket *sock;
1879         int status = -EIO;
1880
1881         current->flags |= PF_FSTRANS;
1882
1883         clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1884         status = __sock_create(xprt->xprt_net, AF_LOCAL,
1885                                         SOCK_STREAM, 0, &sock, 1);
1886         if (status < 0) {
1887                 dprintk("RPC:       can't create AF_LOCAL "
1888                         "transport socket (%d).\n", -status);
1889                 goto out;
1890         }
1891         xs_reclassify_socketu(sock);
1892
1893         dprintk("RPC:       worker connecting xprt %p via AF_LOCAL to %s\n",
1894                         xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1895
1896         status = xs_local_finish_connecting(xprt, sock);
1897         switch (status) {
1898         case 0:
1899                 dprintk("RPC:       xprt %p connected to %s\n",
1900                                 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1901                 xprt_set_connected(xprt);
1902                 break;
1903         case -ENOENT:
1904                 dprintk("RPC:       xprt %p: socket %s does not exist\n",
1905                                 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1906                 break;
1907         case -ECONNREFUSED:
1908                 dprintk("RPC:       xprt %p: connection refused for %s\n",
1909                                 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1910                 break;
1911         default:
1912                 printk(KERN_ERR "%s: unhandled error (%d) connecting to %s\n",
1913                                 __func__, -status,
1914                                 xprt->address_strings[RPC_DISPLAY_ADDR]);
1915         }
1916
1917 out:
1918         xprt_clear_connecting(xprt);
1919         xprt_wake_pending_tasks(xprt, status);
1920         current->flags &= ~PF_FSTRANS;
1921 }
1922
1923 #ifdef CONFIG_SUNRPC_SWAP
1924 static void xs_set_memalloc(struct rpc_xprt *xprt)
1925 {
1926         struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1927                         xprt);
1928
1929         if (xprt->swapper)
1930                 sk_set_memalloc(transport->inet);
1931 }
1932
1933 /**
1934  * xs_swapper - Tag this transport as being used for swap.
1935  * @xprt: transport to tag
1936  * @enable: enable/disable
1937  *
1938  */
1939 int xs_swapper(struct rpc_xprt *xprt, int enable)
1940 {
1941         struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1942                         xprt);
1943         int err = 0;
1944
1945         if (enable) {
1946                 xprt->swapper++;
1947                 xs_set_memalloc(xprt);
1948         } else if (xprt->swapper) {
1949                 xprt->swapper--;
1950                 sk_clear_memalloc(transport->inet);
1951         }
1952
1953         return err;
1954 }
1955 EXPORT_SYMBOL_GPL(xs_swapper);
1956 #else
1957 static void xs_set_memalloc(struct rpc_xprt *xprt)
1958 {
1959 }
1960 #endif
1961
1962 static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
1963 {
1964         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1965
1966         if (!transport->inet) {
1967                 struct sock *sk = sock->sk;
1968
1969                 write_lock_bh(&sk->sk_callback_lock);
1970
1971                 xs_save_old_callbacks(transport, sk);
1972
1973                 sk->sk_user_data = xprt;
1974                 sk->sk_data_ready = xs_udp_data_ready;
1975                 sk->sk_write_space = xs_udp_write_space;
1976                 sk->sk_no_check = UDP_CSUM_NORCV;
1977                 sk->sk_allocation = GFP_ATOMIC;
1978
1979                 xprt_set_connected(xprt);
1980
1981                 /* Reset to new socket */
1982                 transport->sock = sock;
1983                 transport->inet = sk;
1984
1985                 xs_set_memalloc(xprt);
1986
1987                 write_unlock_bh(&sk->sk_callback_lock);
1988         }
1989         xs_udp_do_set_buffer_size(xprt);
1990 }
1991
1992 static void xs_udp_setup_socket(struct work_struct *work)
1993 {
1994         struct sock_xprt *transport =
1995                 container_of(work, struct sock_xprt, connect_worker.work);
1996         struct rpc_xprt *xprt = &transport->xprt;
1997         struct socket *sock = transport->sock;
1998         int status = -EIO;
1999
2000         current->flags |= PF_FSTRANS;
2001
2002         /* Start by resetting any existing state */
2003         xs_reset_transport(transport);
2004         sock = xs_create_sock(xprt, transport,
2005                         xs_addr(xprt)->sa_family, SOCK_DGRAM, IPPROTO_UDP);
2006         if (IS_ERR(sock))
2007                 goto out;
2008
2009         dprintk("RPC:       worker connecting xprt %p via %s to "
2010                                 "%s (port %s)\n", xprt,
2011                         xprt->address_strings[RPC_DISPLAY_PROTO],
2012                         xprt->address_strings[RPC_DISPLAY_ADDR],
2013                         xprt->address_strings[RPC_DISPLAY_PORT]);
2014
2015         xs_udp_finish_connecting(xprt, sock);
2016         status = 0;
2017 out:
2018         xprt_clear_connecting(xprt);
2019         xprt_wake_pending_tasks(xprt, status);
2020         current->flags &= ~PF_FSTRANS;
2021 }
2022
2023 /*
2024  * We need to preserve the port number so the reply cache on the server can
2025  * find our cached RPC replies when we get around to reconnecting.
2026  */
2027 static void xs_abort_connection(struct sock_xprt *transport)
2028 {
2029         int result;
2030         struct sockaddr any;
2031
2032         dprintk("RPC:       disconnecting xprt %p to reuse port\n", transport);
2033
2034         /*
2035          * Disconnect the transport socket by doing a connect operation
2036          * with AF_UNSPEC.  This should return immediately...
2037          */
2038         memset(&any, 0, sizeof(any));
2039         any.sa_family = AF_UNSPEC;
2040         result = kernel_connect(transport->sock, &any, sizeof(any), 0);
2041         if (!result)
2042                 xs_sock_reset_connection_flags(&transport->xprt);
2043         dprintk("RPC:       AF_UNSPEC connect return code %d\n", result);
2044 }
2045
2046 static void xs_tcp_reuse_connection(struct sock_xprt *transport)
2047 {
2048         unsigned int state = transport->inet->sk_state;
2049
2050         if (state == TCP_CLOSE && transport->sock->state == SS_UNCONNECTED) {
2051                 /* we don't need to abort the connection if the socket
2052                  * hasn't undergone a shutdown
2053                  */
2054                 if (transport->inet->sk_shutdown == 0)
2055                         return;
2056                 dprintk("RPC:       %s: TCP_CLOSEd and sk_shutdown set to %d\n",
2057                                 __func__, transport->inet->sk_shutdown);
2058         }
2059         if ((1 << state) & (TCPF_ESTABLISHED|TCPF_SYN_SENT)) {
2060                 /* we don't need to abort the connection if the socket
2061                  * hasn't undergone a shutdown
2062                  */
2063                 if (transport->inet->sk_shutdown == 0)
2064                         return;
2065                 dprintk("RPC:       %s: ESTABLISHED/SYN_SENT "
2066                                 "sk_shutdown set to %d\n",
2067                                 __func__, transport->inet->sk_shutdown);
2068         }
2069         xs_abort_connection(transport);
2070 }
2071
2072 static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2073 {
2074         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2075         int ret = -ENOTCONN;
2076
2077         if (!transport->inet) {
2078                 struct sock *sk = sock->sk;
2079
2080                 write_lock_bh(&sk->sk_callback_lock);
2081
2082                 xs_save_old_callbacks(transport, sk);
2083
2084                 sk->sk_user_data = xprt;
2085                 sk->sk_data_ready = xs_tcp_data_ready;
2086                 sk->sk_state_change = xs_tcp_state_change;
2087                 sk->sk_write_space = xs_tcp_write_space;
2088                 sk->sk_allocation = GFP_ATOMIC;
2089
2090                 /* socket options */
2091                 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
2092                 sock_reset_flag(sk, SOCK_LINGER);
2093                 tcp_sk(sk)->linger2 = 0;
2094                 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
2095
2096                 xprt_clear_connected(xprt);
2097
2098                 /* Reset to new socket */
2099                 transport->sock = sock;
2100                 transport->inet = sk;
2101
2102                 write_unlock_bh(&sk->sk_callback_lock);
2103         }
2104
2105         if (!xprt_bound(xprt))
2106                 goto out;
2107
2108         xs_set_memalloc(xprt);
2109
2110         /* Tell the socket layer to start connecting... */
2111         xprt->stat.connect_count++;
2112         xprt->stat.connect_start = jiffies;
2113         ret = kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
2114         switch (ret) {
2115         case 0:
2116         case -EINPROGRESS:
2117                 /* SYN_SENT! */
2118                 xprt->connect_cookie++;
2119                 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2120                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2121         }
2122 out:
2123         return ret;
2124 }
2125
2126 /**
2127  * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint
2128  * @xprt: RPC transport to connect
2129  * @transport: socket transport to connect
2130  * @create_sock: function to create a socket of the correct type
2131  *
2132  * Invoked by a work queue tasklet.
2133  */
2134 static void xs_tcp_setup_socket(struct work_struct *work)
2135 {
2136         struct sock_xprt *transport =
2137                 container_of(work, struct sock_xprt, connect_worker.work);
2138         struct socket *sock = transport->sock;
2139         struct rpc_xprt *xprt = &transport->xprt;
2140         int status = -EIO;
2141
2142         current->flags |= PF_FSTRANS;
2143
2144         if (!sock) {
2145                 clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
2146                 sock = xs_create_sock(xprt, transport,
2147                                 xs_addr(xprt)->sa_family, SOCK_STREAM, IPPROTO_TCP);
2148                 if (IS_ERR(sock)) {
2149                         status = PTR_ERR(sock);
2150                         goto out;
2151                 }
2152         } else {
2153                 int abort_and_exit;
2154
2155                 abort_and_exit = test_and_clear_bit(XPRT_CONNECTION_ABORT,
2156                                 &xprt->state);
2157                 /* "close" the socket, preserving the local port */
2158                 xs_tcp_reuse_connection(transport);
2159
2160                 if (abort_and_exit)
2161                         goto out_eagain;
2162         }
2163
2164         dprintk("RPC:       worker connecting xprt %p via %s to "
2165                                 "%s (port %s)\n", xprt,
2166                         xprt->address_strings[RPC_DISPLAY_PROTO],
2167                         xprt->address_strings[RPC_DISPLAY_ADDR],
2168                         xprt->address_strings[RPC_DISPLAY_PORT]);
2169
2170         status = xs_tcp_finish_connecting(xprt, sock);
2171         dprintk("RPC:       %p connect status %d connected %d sock state %d\n",
2172                         xprt, -status, xprt_connected(xprt),
2173                         sock->sk->sk_state);
2174         switch (status) {
2175         default:
2176                 printk("%s: connect returned unhandled error %d\n",
2177                         __func__, status);
2178         case -EADDRNOTAVAIL:
2179                 /* We're probably in TIME_WAIT. Get rid of existing socket,
2180                  * and retry
2181                  */
2182                 xs_tcp_force_close(xprt);
2183                 break;
2184         case -ECONNREFUSED:
2185         case -ECONNRESET:
2186         case -ENETUNREACH:
2187                 /* retry with existing socket, after a delay */
2188         case 0:
2189         case -EINPROGRESS:
2190         case -EALREADY:
2191                 xprt_clear_connecting(xprt);
2192                 current->flags &= ~PF_FSTRANS;
2193                 return;
2194         case -EINVAL:
2195                 /* Happens, for instance, if the user specified a link
2196                  * local IPv6 address without a scope-id.
2197                  */
2198                 goto out;
2199         }
2200 out_eagain:
2201         status = -EAGAIN;
2202 out:
2203         xprt_clear_connecting(xprt);
2204         xprt_wake_pending_tasks(xprt, status);
2205         current->flags &= ~PF_FSTRANS;
2206 }
2207
2208 /**
2209  * xs_connect - connect a socket to a remote endpoint
2210  * @xprt: pointer to transport structure
2211  * @task: address of RPC task that manages state of connect request
2212  *
2213  * TCP: If the remote end dropped the connection, delay reconnecting.
2214  *
2215  * UDP socket connects are synchronous, but we use a work queue anyway
2216  * to guarantee that even unprivileged user processes can set up a
2217  * socket on a privileged port.
2218  *
2219  * If a UDP socket connect fails, the delay behavior here prevents
2220  * retry floods (hard mounts).
2221  */
2222 static void xs_connect(struct rpc_xprt *xprt, struct rpc_task *task)
2223 {
2224         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2225
2226         if (transport->sock != NULL && !RPC_IS_SOFTCONN(task)) {
2227                 dprintk("RPC:       xs_connect delayed xprt %p for %lu "
2228                                 "seconds\n",
2229                                 xprt, xprt->reestablish_timeout / HZ);
2230                 queue_delayed_work(rpciod_workqueue,
2231                                    &transport->connect_worker,
2232                                    xprt->reestablish_timeout);
2233                 xprt->reestablish_timeout <<= 1;
2234                 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2235                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2236                 if (xprt->reestablish_timeout > XS_TCP_MAX_REEST_TO)
2237                         xprt->reestablish_timeout = XS_TCP_MAX_REEST_TO;
2238         } else {
2239                 dprintk("RPC:       xs_connect scheduled xprt %p\n", xprt);
2240                 queue_delayed_work(rpciod_workqueue,
2241                                    &transport->connect_worker, 0);
2242         }
2243 }
2244
2245 /**
2246  * xs_local_print_stats - display AF_LOCAL socket-specifc stats
2247  * @xprt: rpc_xprt struct containing statistics
2248  * @seq: output file
2249  *
2250  */
2251 static void xs_local_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2252 {
2253         long idle_time = 0;
2254
2255         if (xprt_connected(xprt))
2256                 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2257
2258         seq_printf(seq, "\txprt:\tlocal %lu %lu %lu %ld %lu %lu %lu "
2259                         "%llu %llu %lu %llu %llu\n",
2260                         xprt->stat.bind_count,
2261                         xprt->stat.connect_count,
2262                         xprt->stat.connect_time,
2263                         idle_time,
2264                         xprt->stat.sends,
2265                         xprt->stat.recvs,
2266                         xprt->stat.bad_xids,
2267                         xprt->stat.req_u,
2268                         xprt->stat.bklog_u,
2269                         xprt->stat.max_slots,
2270                         xprt->stat.sending_u,
2271                         xprt->stat.pending_u);
2272 }
2273
2274 /**
2275  * xs_udp_print_stats - display UDP socket-specifc stats
2276  * @xprt: rpc_xprt struct containing statistics
2277  * @seq: output file
2278  *
2279  */
2280 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2281 {
2282         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2283
2284         seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %llu %llu "
2285                         "%lu %llu %llu\n",
2286                         transport->srcport,
2287                         xprt->stat.bind_count,
2288                         xprt->stat.sends,
2289                         xprt->stat.recvs,
2290                         xprt->stat.bad_xids,
2291                         xprt->stat.req_u,
2292                         xprt->stat.bklog_u,
2293                         xprt->stat.max_slots,
2294                         xprt->stat.sending_u,
2295                         xprt->stat.pending_u);
2296 }
2297
2298 /**
2299  * xs_tcp_print_stats - display TCP socket-specifc stats
2300  * @xprt: rpc_xprt struct containing statistics
2301  * @seq: output file
2302  *
2303  */
2304 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2305 {
2306         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2307         long idle_time = 0;
2308
2309         if (xprt_connected(xprt))
2310                 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2311
2312         seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu "
2313                         "%llu %llu %lu %llu %llu\n",
2314                         transport->srcport,
2315                         xprt->stat.bind_count,
2316                         xprt->stat.connect_count,
2317                         xprt->stat.connect_time,
2318                         idle_time,
2319                         xprt->stat.sends,
2320                         xprt->stat.recvs,
2321                         xprt->stat.bad_xids,
2322                         xprt->stat.req_u,
2323                         xprt->stat.bklog_u,
2324                         xprt->stat.max_slots,
2325                         xprt->stat.sending_u,
2326                         xprt->stat.pending_u);
2327 }
2328
2329 /*
2330  * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason
2331  * we allocate pages instead doing a kmalloc like rpc_malloc is because we want
2332  * to use the server side send routines.
2333  */
2334 static void *bc_malloc(struct rpc_task *task, size_t size)
2335 {
2336         struct page *page;
2337         struct rpc_buffer *buf;
2338
2339         WARN_ON_ONCE(size > PAGE_SIZE - sizeof(struct rpc_buffer));
2340         if (size > PAGE_SIZE - sizeof(struct rpc_buffer))
2341                 return NULL;
2342
2343         page = alloc_page(GFP_KERNEL);
2344         if (!page)
2345                 return NULL;
2346
2347         buf = page_address(page);
2348         buf->len = PAGE_SIZE;
2349
2350         return buf->data;
2351 }
2352
2353 /*
2354  * Free the space allocated in the bc_alloc routine
2355  */
2356 static void bc_free(void *buffer)
2357 {
2358         struct rpc_buffer *buf;
2359
2360         if (!buffer)
2361                 return;
2362
2363         buf = container_of(buffer, struct rpc_buffer, data);
2364         free_page((unsigned long)buf);
2365 }
2366
2367 /*
2368  * Use the svc_sock to send the callback. Must be called with svsk->sk_mutex
2369  * held. Borrows heavily from svc_tcp_sendto and xs_tcp_send_request.
2370  */
2371 static int bc_sendto(struct rpc_rqst *req)
2372 {
2373         int len;
2374         struct xdr_buf *xbufp = &req->rq_snd_buf;
2375         struct rpc_xprt *xprt = req->rq_xprt;
2376         struct sock_xprt *transport =
2377                                 container_of(xprt, struct sock_xprt, xprt);
2378         struct socket *sock = transport->sock;
2379         unsigned long headoff;
2380         unsigned long tailoff;
2381
2382         xs_encode_stream_record_marker(xbufp);
2383
2384         tailoff = (unsigned long)xbufp->tail[0].iov_base & ~PAGE_MASK;
2385         headoff = (unsigned long)xbufp->head[0].iov_base & ~PAGE_MASK;
2386         len = svc_send_common(sock, xbufp,
2387                               virt_to_page(xbufp->head[0].iov_base), headoff,
2388                               xbufp->tail[0].iov_base, tailoff);
2389
2390         if (len != xbufp->len) {
2391                 printk(KERN_NOTICE "Error sending entire callback!\n");
2392                 len = -EAGAIN;
2393         }
2394
2395         return len;
2396 }
2397
2398 /*
2399  * The send routine. Borrows from svc_send
2400  */
2401 static int bc_send_request(struct rpc_task *task)
2402 {
2403         struct rpc_rqst *req = task->tk_rqstp;
2404         struct svc_xprt *xprt;
2405         u32                     len;
2406
2407         dprintk("sending request with xid: %08x\n", ntohl(req->rq_xid));
2408         /*
2409          * Get the server socket associated with this callback xprt
2410          */
2411         xprt = req->rq_xprt->bc_xprt;
2412
2413         /*
2414          * Grab the mutex to serialize data as the connection is shared
2415          * with the fore channel
2416          */
2417         if (!mutex_trylock(&xprt->xpt_mutex)) {
2418                 rpc_sleep_on(&xprt->xpt_bc_pending, task, NULL);
2419                 if (!mutex_trylock(&xprt->xpt_mutex))
2420                         return -EAGAIN;
2421                 rpc_wake_up_queued_task(&xprt->xpt_bc_pending, task);
2422         }
2423         if (test_bit(XPT_DEAD, &xprt->xpt_flags))
2424                 len = -ENOTCONN;
2425         else
2426                 len = bc_sendto(req);
2427         mutex_unlock(&xprt->xpt_mutex);
2428
2429         if (len > 0)
2430                 len = 0;
2431
2432         return len;
2433 }
2434
2435 /*
2436  * The close routine. Since this is client initiated, we do nothing
2437  */
2438
2439 static void bc_close(struct rpc_xprt *xprt)
2440 {
2441 }
2442
2443 /*
2444  * The xprt destroy routine. Again, because this connection is client
2445  * initiated, we do nothing
2446  */
2447
2448 static void bc_destroy(struct rpc_xprt *xprt)
2449 {
2450 }
2451
2452 static struct rpc_xprt_ops xs_local_ops = {
2453         .reserve_xprt           = xprt_reserve_xprt,
2454         .release_xprt           = xs_tcp_release_xprt,
2455         .alloc_slot             = xprt_alloc_slot,
2456         .rpcbind                = xs_local_rpcbind,
2457         .set_port               = xs_local_set_port,
2458         .connect                = xs_connect,
2459         .buf_alloc              = rpc_malloc,
2460         .buf_free               = rpc_free,
2461         .send_request           = xs_local_send_request,
2462         .set_retrans_timeout    = xprt_set_retrans_timeout_def,
2463         .close                  = xs_close,
2464         .destroy                = xs_destroy,
2465         .print_stats            = xs_local_print_stats,
2466 };
2467
2468 static struct rpc_xprt_ops xs_udp_ops = {
2469         .set_buffer_size        = xs_udp_set_buffer_size,
2470         .reserve_xprt           = xprt_reserve_xprt_cong,
2471         .release_xprt           = xprt_release_xprt_cong,
2472         .alloc_slot             = xprt_alloc_slot,
2473         .rpcbind                = rpcb_getport_async,
2474         .set_port               = xs_set_port,
2475         .connect                = xs_connect,
2476         .buf_alloc              = rpc_malloc,
2477         .buf_free               = rpc_free,
2478         .send_request           = xs_udp_send_request,
2479         .set_retrans_timeout    = xprt_set_retrans_timeout_rtt,
2480         .timer                  = xs_udp_timer,
2481         .release_request        = xprt_release_rqst_cong,
2482         .close                  = xs_close,
2483         .destroy                = xs_destroy,
2484         .print_stats            = xs_udp_print_stats,
2485 };
2486
2487 static struct rpc_xprt_ops xs_tcp_ops = {
2488         .reserve_xprt           = xprt_reserve_xprt,
2489         .release_xprt           = xs_tcp_release_xprt,
2490         .alloc_slot             = xprt_lock_and_alloc_slot,
2491         .rpcbind                = rpcb_getport_async,
2492         .set_port               = xs_set_port,
2493         .connect                = xs_connect,
2494         .buf_alloc              = rpc_malloc,
2495         .buf_free               = rpc_free,
2496         .send_request           = xs_tcp_send_request,
2497         .set_retrans_timeout    = xprt_set_retrans_timeout_def,
2498         .close                  = xs_tcp_close,
2499         .destroy                = xs_destroy,
2500         .print_stats            = xs_tcp_print_stats,
2501 };
2502
2503 /*
2504  * The rpc_xprt_ops for the server backchannel
2505  */
2506
2507 static struct rpc_xprt_ops bc_tcp_ops = {
2508         .reserve_xprt           = xprt_reserve_xprt,
2509         .release_xprt           = xprt_release_xprt,
2510         .alloc_slot             = xprt_alloc_slot,
2511         .rpcbind                = xs_local_rpcbind,
2512         .buf_alloc              = bc_malloc,
2513         .buf_free               = bc_free,
2514         .send_request           = bc_send_request,
2515         .set_retrans_timeout    = xprt_set_retrans_timeout_def,
2516         .close                  = bc_close,
2517         .destroy                = bc_destroy,
2518         .print_stats            = xs_tcp_print_stats,
2519 };
2520
2521 static int xs_init_anyaddr(const int family, struct sockaddr *sap)
2522 {
2523         static const struct sockaddr_in sin = {
2524                 .sin_family             = AF_INET,
2525                 .sin_addr.s_addr        = htonl(INADDR_ANY),
2526         };
2527         static const struct sockaddr_in6 sin6 = {
2528                 .sin6_family            = AF_INET6,
2529                 .sin6_addr              = IN6ADDR_ANY_INIT,
2530         };
2531
2532         switch (family) {
2533         case AF_LOCAL:
2534                 break;
2535         case AF_INET:
2536                 memcpy(sap, &sin, sizeof(sin));
2537                 break;
2538         case AF_INET6:
2539                 memcpy(sap, &sin6, sizeof(sin6));
2540                 break;
2541         default:
2542                 dprintk("RPC:       %s: Bad address family\n", __func__);
2543                 return -EAFNOSUPPORT;
2544         }
2545         return 0;
2546 }
2547
2548 static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
2549                                       unsigned int slot_table_size,
2550                                       unsigned int max_slot_table_size)
2551 {
2552         struct rpc_xprt *xprt;
2553         struct sock_xprt *new;
2554
2555         if (args->addrlen > sizeof(xprt->addr)) {
2556                 dprintk("RPC:       xs_setup_xprt: address too large\n");
2557                 return ERR_PTR(-EBADF);
2558         }
2559
2560         xprt = xprt_alloc(args->net, sizeof(*new), slot_table_size,
2561                         max_slot_table_size);
2562         if (xprt == NULL) {
2563                 dprintk("RPC:       xs_setup_xprt: couldn't allocate "
2564                                 "rpc_xprt\n");
2565                 return ERR_PTR(-ENOMEM);
2566         }
2567
2568         new = container_of(xprt, struct sock_xprt, xprt);
2569         memcpy(&xprt->addr, args->dstaddr, args->addrlen);
2570         xprt->addrlen = args->addrlen;
2571         if (args->srcaddr)
2572                 memcpy(&new->srcaddr, args->srcaddr, args->addrlen);
2573         else {
2574                 int err;
2575                 err = xs_init_anyaddr(args->dstaddr->sa_family,
2576                                         (struct sockaddr *)&new->srcaddr);
2577                 if (err != 0) {
2578                         xprt_free(xprt);
2579                         return ERR_PTR(err);
2580                 }
2581         }
2582
2583         return xprt;
2584 }
2585
2586 static const struct rpc_timeout xs_local_default_timeout = {
2587         .to_initval = 10 * HZ,
2588         .to_maxval = 10 * HZ,
2589         .to_retries = 2,
2590 };
2591
2592 /**
2593  * xs_setup_local - Set up transport to use an AF_LOCAL socket
2594  * @args: rpc transport creation arguments
2595  *
2596  * AF_LOCAL is a "tpi_cots_ord" transport, just like TCP
2597  */
2598 static struct rpc_xprt *xs_setup_local(struct xprt_create *args)
2599 {
2600         struct sockaddr_un *sun = (struct sockaddr_un *)args->dstaddr;
2601         struct sock_xprt *transport;
2602         struct rpc_xprt *xprt;
2603         struct rpc_xprt *ret;
2604
2605         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2606                         xprt_max_tcp_slot_table_entries);
2607         if (IS_ERR(xprt))
2608                 return xprt;
2609         transport = container_of(xprt, struct sock_xprt, xprt);
2610
2611         xprt->prot = 0;
2612         xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2613         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2614
2615         xprt->bind_timeout = XS_BIND_TO;
2616         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2617         xprt->idle_timeout = XS_IDLE_DISC_TO;
2618
2619         xprt->ops = &xs_local_ops;
2620         xprt->timeout = &xs_local_default_timeout;
2621
2622         switch (sun->sun_family) {
2623         case AF_LOCAL:
2624                 if (sun->sun_path[0] != '/') {
2625                         dprintk("RPC:       bad AF_LOCAL address: %s\n",
2626                                         sun->sun_path);
2627                         ret = ERR_PTR(-EINVAL);
2628                         goto out_err;
2629                 }
2630                 xprt_set_bound(xprt);
2631                 INIT_DELAYED_WORK(&transport->connect_worker,
2632                                         xs_local_setup_socket);
2633                 xs_format_peer_addresses(xprt, "local", RPCBIND_NETID_LOCAL);
2634                 break;
2635         default:
2636                 ret = ERR_PTR(-EAFNOSUPPORT);
2637                 goto out_err;
2638         }
2639
2640         dprintk("RPC:       set up xprt to %s via AF_LOCAL\n",
2641                         xprt->address_strings[RPC_DISPLAY_ADDR]);
2642
2643         if (try_module_get(THIS_MODULE))
2644                 return xprt;
2645         ret = ERR_PTR(-EINVAL);
2646 out_err:
2647         xprt_free(xprt);
2648         return ret;
2649 }
2650
2651 static const struct rpc_timeout xs_udp_default_timeout = {
2652         .to_initval = 5 * HZ,
2653         .to_maxval = 30 * HZ,
2654         .to_increment = 5 * HZ,
2655         .to_retries = 5,
2656 };
2657
2658 /**
2659  * xs_setup_udp - Set up transport to use a UDP socket
2660  * @args: rpc transport creation arguments
2661  *
2662  */
2663 static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
2664 {
2665         struct sockaddr *addr = args->dstaddr;
2666         struct rpc_xprt *xprt;
2667         struct sock_xprt *transport;
2668         struct rpc_xprt *ret;
2669
2670         xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries,
2671                         xprt_udp_slot_table_entries);
2672         if (IS_ERR(xprt))
2673                 return xprt;
2674         transport = container_of(xprt, struct sock_xprt, xprt);
2675
2676         xprt->prot = IPPROTO_UDP;
2677         xprt->tsh_size = 0;
2678         /* XXX: header size can vary due to auth type, IPv6, etc. */
2679         xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
2680
2681         xprt->bind_timeout = XS_BIND_TO;
2682         xprt->reestablish_timeout = XS_UDP_REEST_TO;
2683         xprt->idle_timeout = XS_IDLE_DISC_TO;
2684
2685         xprt->ops = &xs_udp_ops;
2686
2687         xprt->timeout = &xs_udp_default_timeout;
2688
2689         switch (addr->sa_family) {
2690         case AF_INET:
2691                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2692                         xprt_set_bound(xprt);
2693
2694                 INIT_DELAYED_WORK(&transport->connect_worker,
2695                                         xs_udp_setup_socket);
2696                 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
2697                 break;
2698         case AF_INET6:
2699                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2700                         xprt_set_bound(xprt);
2701
2702                 INIT_DELAYED_WORK(&transport->connect_worker,
2703                                         xs_udp_setup_socket);
2704                 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
2705                 break;
2706         default:
2707                 ret = ERR_PTR(-EAFNOSUPPORT);
2708                 goto out_err;
2709         }
2710
2711         if (xprt_bound(xprt))
2712                 dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
2713                                 xprt->address_strings[RPC_DISPLAY_ADDR],
2714                                 xprt->address_strings[RPC_DISPLAY_PORT],
2715                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
2716         else
2717                 dprintk("RPC:       set up xprt to %s (autobind) via %s\n",
2718                                 xprt->address_strings[RPC_DISPLAY_ADDR],
2719                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
2720
2721         if (try_module_get(THIS_MODULE))
2722                 return xprt;
2723         ret = ERR_PTR(-EINVAL);
2724 out_err:
2725         xprt_free(xprt);
2726         return ret;
2727 }
2728
2729 static const struct rpc_timeout xs_tcp_default_timeout = {
2730         .to_initval = 60 * HZ,
2731         .to_maxval = 60 * HZ,
2732         .to_retries = 2,
2733 };
2734
2735 /**
2736  * xs_setup_tcp - Set up transport to use a TCP socket
2737  * @args: rpc transport creation arguments
2738  *
2739  */
2740 static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
2741 {
2742         struct sockaddr *addr = args->dstaddr;
2743         struct rpc_xprt *xprt;
2744         struct sock_xprt *transport;
2745         struct rpc_xprt *ret;
2746
2747         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2748                         xprt_max_tcp_slot_table_entries);
2749         if (IS_ERR(xprt))
2750                 return xprt;
2751         transport = container_of(xprt, struct sock_xprt, xprt);
2752
2753         xprt->prot = IPPROTO_TCP;
2754         xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2755         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2756
2757         xprt->bind_timeout = XS_BIND_TO;
2758         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2759         xprt->idle_timeout = XS_IDLE_DISC_TO;
2760
2761         xprt->ops = &xs_tcp_ops;
2762         xprt->timeout = &xs_tcp_default_timeout;
2763
2764         switch (addr->sa_family) {
2765         case AF_INET:
2766                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2767                         xprt_set_bound(xprt);
2768
2769                 INIT_DELAYED_WORK(&transport->connect_worker,
2770                                         xs_tcp_setup_socket);
2771                 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
2772                 break;
2773         case AF_INET6:
2774                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2775                         xprt_set_bound(xprt);
2776
2777                 INIT_DELAYED_WORK(&transport->connect_worker,
2778                                         xs_tcp_setup_socket);
2779                 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
2780                 break;
2781         default:
2782                 ret = ERR_PTR(-EAFNOSUPPORT);
2783                 goto out_err;
2784         }
2785
2786         if (xprt_bound(xprt))
2787                 dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
2788                                 xprt->address_strings[RPC_DISPLAY_ADDR],
2789                                 xprt->address_strings[RPC_DISPLAY_PORT],
2790                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
2791         else
2792                 dprintk("RPC:       set up xprt to %s (autobind) via %s\n",
2793                                 xprt->address_strings[RPC_DISPLAY_ADDR],
2794                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
2795
2796
2797         if (try_module_get(THIS_MODULE))
2798                 return xprt;
2799         ret = ERR_PTR(-EINVAL);
2800 out_err:
2801         xprt_free(xprt);
2802         return ret;
2803 }
2804
2805 /**
2806  * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket
2807  * @args: rpc transport creation arguments
2808  *
2809  */
2810 static struct rpc_xprt *xs_setup_bc_tcp(struct xprt_create *args)
2811 {
2812         struct sockaddr *addr = args->dstaddr;
2813         struct rpc_xprt *xprt;
2814         struct sock_xprt *transport;
2815         struct svc_sock *bc_sock;
2816         struct rpc_xprt *ret;
2817
2818         if (args->bc_xprt->xpt_bc_xprt) {
2819                 /*
2820                  * This server connection already has a backchannel
2821                  * export; we can't create a new one, as we wouldn't be
2822                  * able to match replies based on xid any more.  So,
2823                  * reuse the already-existing one:
2824                  */
2825                  return args->bc_xprt->xpt_bc_xprt;
2826         }
2827         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2828                         xprt_tcp_slot_table_entries);
2829         if (IS_ERR(xprt))
2830                 return xprt;
2831         transport = container_of(xprt, struct sock_xprt, xprt);
2832
2833         xprt->prot = IPPROTO_TCP;
2834         xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2835         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2836         xprt->timeout = &xs_tcp_default_timeout;
2837
2838         /* backchannel */
2839         xprt_set_bound(xprt);
2840         xprt->bind_timeout = 0;
2841         xprt->reestablish_timeout = 0;
2842         xprt->idle_timeout = 0;
2843
2844         xprt->ops = &bc_tcp_ops;
2845
2846         switch (addr->sa_family) {
2847         case AF_INET:
2848                 xs_format_peer_addresses(xprt, "tcp",
2849                                          RPCBIND_NETID_TCP);
2850                 break;
2851         case AF_INET6:
2852                 xs_format_peer_addresses(xprt, "tcp",
2853                                    RPCBIND_NETID_TCP6);
2854                 break;
2855         default:
2856                 ret = ERR_PTR(-EAFNOSUPPORT);
2857                 goto out_err;
2858         }
2859
2860         dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
2861                         xprt->address_strings[RPC_DISPLAY_ADDR],
2862                         xprt->address_strings[RPC_DISPLAY_PORT],
2863                         xprt->address_strings[RPC_DISPLAY_PROTO]);
2864
2865         /*
2866          * Once we've associated a backchannel xprt with a connection,
2867          * we want to keep it around as long as long as the connection
2868          * lasts, in case we need to start using it for a backchannel
2869          * again; this reference won't be dropped until bc_xprt is
2870          * destroyed.
2871          */
2872         xprt_get(xprt);
2873         args->bc_xprt->xpt_bc_xprt = xprt;
2874         xprt->bc_xprt = args->bc_xprt;
2875         bc_sock = container_of(args->bc_xprt, struct svc_sock, sk_xprt);
2876         transport->sock = bc_sock->sk_sock;
2877         transport->inet = bc_sock->sk_sk;
2878
2879         /*
2880          * Since we don't want connections for the backchannel, we set
2881          * the xprt status to connected
2882          */
2883         xprt_set_connected(xprt);
2884
2885
2886         if (try_module_get(THIS_MODULE))
2887                 return xprt;
2888         xprt_put(xprt);
2889         ret = ERR_PTR(-EINVAL);
2890 out_err:
2891         xprt_free(xprt);
2892         return ret;
2893 }
2894
2895 static struct xprt_class        xs_local_transport = {
2896         .list           = LIST_HEAD_INIT(xs_local_transport.list),
2897         .name           = "named UNIX socket",
2898         .owner          = THIS_MODULE,
2899         .ident          = XPRT_TRANSPORT_LOCAL,
2900         .setup          = xs_setup_local,
2901 };
2902
2903 static struct xprt_class        xs_udp_transport = {
2904         .list           = LIST_HEAD_INIT(xs_udp_transport.list),
2905         .name           = "udp",
2906         .owner          = THIS_MODULE,
2907         .ident          = XPRT_TRANSPORT_UDP,
2908         .setup          = xs_setup_udp,
2909 };
2910
2911 static struct xprt_class        xs_tcp_transport = {
2912         .list           = LIST_HEAD_INIT(xs_tcp_transport.list),
2913         .name           = "tcp",
2914         .owner          = THIS_MODULE,
2915         .ident          = XPRT_TRANSPORT_TCP,
2916         .setup          = xs_setup_tcp,
2917 };
2918
2919 static struct xprt_class        xs_bc_tcp_transport = {
2920         .list           = LIST_HEAD_INIT(xs_bc_tcp_transport.list),
2921         .name           = "tcp NFSv4.1 backchannel",
2922         .owner          = THIS_MODULE,
2923         .ident          = XPRT_TRANSPORT_BC_TCP,
2924         .setup          = xs_setup_bc_tcp,
2925 };
2926
2927 /**
2928  * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
2929  *
2930  */
2931 int init_socket_xprt(void)
2932 {
2933 #ifdef RPC_DEBUG
2934         if (!sunrpc_table_header)
2935                 sunrpc_table_header = register_sysctl_table(sunrpc_table);
2936 #endif
2937
2938         xprt_register_transport(&xs_local_transport);
2939         xprt_register_transport(&xs_udp_transport);
2940         xprt_register_transport(&xs_tcp_transport);
2941         xprt_register_transport(&xs_bc_tcp_transport);
2942
2943         return 0;
2944 }
2945
2946 /**
2947  * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
2948  *
2949  */
2950 void cleanup_socket_xprt(void)
2951 {
2952 #ifdef RPC_DEBUG
2953         if (sunrpc_table_header) {
2954                 unregister_sysctl_table(sunrpc_table_header);
2955                 sunrpc_table_header = NULL;
2956         }
2957 #endif
2958
2959         xprt_unregister_transport(&xs_local_transport);
2960         xprt_unregister_transport(&xs_udp_transport);
2961         xprt_unregister_transport(&xs_tcp_transport);
2962         xprt_unregister_transport(&xs_bc_tcp_transport);
2963 }
2964
2965 static int param_set_uint_minmax(const char *val,
2966                 const struct kernel_param *kp,
2967                 unsigned int min, unsigned int max)
2968 {
2969         unsigned long num;
2970         int ret;
2971
2972         if (!val)
2973                 return -EINVAL;
2974         ret = strict_strtoul(val, 0, &num);
2975         if (ret == -EINVAL || num < min || num > max)
2976                 return -EINVAL;
2977         *((unsigned int *)kp->arg) = num;
2978         return 0;
2979 }
2980
2981 static int param_set_portnr(const char *val, const struct kernel_param *kp)
2982 {
2983         return param_set_uint_minmax(val, kp,
2984                         RPC_MIN_RESVPORT,
2985                         RPC_MAX_RESVPORT);
2986 }
2987
2988 static struct kernel_param_ops param_ops_portnr = {
2989         .set = param_set_portnr,
2990         .get = param_get_uint,
2991 };
2992
2993 #define param_check_portnr(name, p) \
2994         __param_check(name, p, unsigned int);
2995
2996 module_param_named(min_resvport, xprt_min_resvport, portnr, 0644);
2997 module_param_named(max_resvport, xprt_max_resvport, portnr, 0644);
2998
2999 static int param_set_slot_table_size(const char *val,
3000                                      const struct kernel_param *kp)
3001 {
3002         return param_set_uint_minmax(val, kp,
3003                         RPC_MIN_SLOT_TABLE,
3004                         RPC_MAX_SLOT_TABLE);
3005 }
3006
3007 static struct kernel_param_ops param_ops_slot_table_size = {
3008         .set = param_set_slot_table_size,
3009         .get = param_get_uint,
3010 };
3011
3012 #define param_check_slot_table_size(name, p) \
3013         __param_check(name, p, unsigned int);
3014
3015 static int param_set_max_slot_table_size(const char *val,
3016                                      const struct kernel_param *kp)
3017 {
3018         return param_set_uint_minmax(val, kp,
3019                         RPC_MIN_SLOT_TABLE,
3020                         RPC_MAX_SLOT_TABLE_LIMIT);
3021 }
3022
3023 static struct kernel_param_ops param_ops_max_slot_table_size = {
3024         .set = param_set_max_slot_table_size,
3025         .get = param_get_uint,
3026 };
3027
3028 #define param_check_max_slot_table_size(name, p) \
3029         __param_check(name, p, unsigned int);
3030
3031 module_param_named(tcp_slot_table_entries, xprt_tcp_slot_table_entries,
3032                    slot_table_size, 0644);
3033 module_param_named(tcp_max_slot_table_entries, xprt_max_tcp_slot_table_entries,
3034                    max_slot_table_size, 0644);
3035 module_param_named(udp_slot_table_entries, xprt_udp_slot_table_entries,
3036                    slot_table_size, 0644);
3037