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/addr.h>
  37 #include <linux/sunrpc/sched.h>
  38 #include <linux/sunrpc/svcsock.h>
  39 #include <linux/sunrpc/xprtsock.h>
  40 #include <linux/file.h>
  41 #ifdef CONFIG_SUNRPC_BACKCHANNEL
  42 #include <linux/sunrpc/bc_xprt.h>
  43 #endif
  44
  45 #include <net/sock.h>
  46 #include <net/checksum.h>
  47 #include <net/udp.h>
  48 #include <net/tcp.h>
  49
  50 #include "sunrpc.h"
  51
  52 static void xs_close(struct rpc_xprt *xprt);
  53
  54 /*
  55  * xprtsock tunables
  56  */
  57 static unsigned int xprt_udp_slot_table_entries = RPC_DEF_SLOT_TABLE;
  58 static unsigned int xprt_tcp_slot_table_entries = RPC_MIN_SLOT_TABLE;
  59 static unsigned int xprt_max_tcp_slot_table_entries = RPC_MAX_SLOT_TABLE;
  60
  61 static unsigned int xprt_min_resvport = RPC_DEF_MIN_RESVPORT;
  62 static unsigned int xprt_max_resvport = RPC_DEF_MAX_RESVPORT;
  63
  64 #define XS_TCP_LINGER_TO        (15U * HZ)
  65 static unsigned int xs_tcp_fin_timeout __read_mostly = XS_TCP_LINGER_TO;
  66
  67 /*
  68  * We can register our own files under /proc/sys/sunrpc by
  69  * calling register_sysctl_table() again.  The files in that
  70  * directory become the union of all files registered there.
  71  *
  72  * We simply need to make sure that we don't collide with
  73  * someone else's file names!
  74  */
  75
  76 #ifdef RPC_DEBUG
  77
  78 static unsigned int min_slot_table_size = RPC_MIN_SLOT_TABLE;
  79 static unsigned int max_slot_table_size = RPC_MAX_SLOT_TABLE;
  80 static unsigned int max_tcp_slot_table_limit = RPC_MAX_SLOT_TABLE_LIMIT;
  81 static unsigned int xprt_min_resvport_limit = RPC_MIN_RESVPORT;
  82 static unsigned int xprt_max_resvport_limit = RPC_MAX_RESVPORT;
  83
  84 static struct ctl_table_header *sunrpc_table_header;
  85
  86 /*
  87  * FIXME: changing the UDP slot table size should also resize the UDP
  88  *        socket buffers for existing UDP transports
  89  */
  90 static ctl_table xs_tunables_table[] = {
  91         {
  92                 .procname       = "udp_slot_table_entries",
  93                 .data           = &xprt_udp_slot_table_entries,
  94                 .maxlen         = sizeof(unsigned int),
  95                 .mode           = 0644,
  96                 .proc_handler   = proc_dointvec_minmax,
  97                 .extra1         = &min_slot_table_size,
  98                 .extra2         = &max_slot_table_size
  99         },
 100         {
 101                 .procname       = "tcp_slot_table_entries",
 102                 .data           = &xprt_tcp_slot_table_entries,
 103                 .maxlen         = sizeof(unsigned int),
 104                 .mode           = 0644,
 105                 .proc_handler   = proc_dointvec_minmax,
 106                 .extra1         = &min_slot_table_size,
 107                 .extra2         = &max_slot_table_size
 108         },
 109         {
 110                 .procname       = "tcp_max_slot_table_entries",
 111                 .data           = &xprt_max_tcp_slot_table_entries,
 112                 .maxlen         = sizeof(unsigned int),
 113                 .mode           = 0644,
 114                 .proc_handler   = proc_dointvec_minmax,
 115                 .extra1         = &min_slot_table_size,
 116                 .extra2         = &max_tcp_slot_table_limit
 117         },
 118         {
 119                 .procname       = "min_resvport",
 120                 .data           = &xprt_min_resvport,
 121                 .maxlen         = sizeof(unsigned int),
 122                 .mode           = 0644,
 123                 .proc_handler   = proc_dointvec_minmax,
 124                 .extra1         = &xprt_min_resvport_limit,
 125                 .extra2         = &xprt_max_resvport_limit
 126         },
 127         {
 128                 .procname       = "max_resvport",
 129                 .data           = &xprt_max_resvport,
 130                 .maxlen         = sizeof(unsigned int),
 131                 .mode           = 0644,
 132                 .proc_handler   = proc_dointvec_minmax,
 133                 .extra1         = &xprt_min_resvport_limit,
 134                 .extra2         = &xprt_max_resvport_limit
 135         },
 136         {
 137                 .procname       = "tcp_fin_timeout",
 138                 .data           = &xs_tcp_fin_timeout,
 139                 .maxlen         = sizeof(xs_tcp_fin_timeout),
 140                 .mode           = 0644,
 141                 .proc_handler   = proc_dointvec_jiffies,
 142         },
 143         { },
 144 };
 145
 146 static ctl_table sunrpc_table[] = {
 147         {
 148                 .procname       = "sunrpc",
 149                 .mode           = 0555,
 150                 .child          = xs_tunables_table
 151         },
 152         { },
 153 };
 154
 155 #endif
 156
 157 /*
 158  * Wait duration for a reply from the RPC portmapper.
 159  */
 160 #define XS_BIND_TO              (60U * HZ)
 161
 162 /*
 163  * Delay if a UDP socket connect error occurs.  This is most likely some
 164  * kind of resource problem on the local host.
 165  */
 166 #define XS_UDP_REEST_TO         (2U * HZ)
 167
 168 /*
 169  * The reestablish timeout allows clients to delay for a bit before attempting
 170  * to reconnect to a server that just dropped our connection.
 171  *
 172  * We implement an exponential backoff when trying to reestablish a TCP
 173  * transport connection with the server.  Some servers like to drop a TCP
 174  * connection when they are overworked, so we start with a short timeout and
 175  * increase over time if the server is down or not responding.
 176  */
 177 #define XS_TCP_INIT_REEST_TO    (3U * HZ)
 178 #define XS_TCP_MAX_REEST_TO     (5U * 60 * HZ)
 179
 180 /*
 181  * TCP idle timeout; client drops the transport socket if it is idle
 182  * for this long.  Note that we also timeout UDP sockets to prevent
 183  * holding port numbers when there is no RPC traffic.
 184  */
 185 #define XS_IDLE_DISC_TO         (5U * 60 * HZ)
 186
 187 #ifdef RPC_DEBUG
 188 # undef  RPC_DEBUG_DATA
 189 # define RPCDBG_FACILITY        RPCDBG_TRANS
 190 #endif
 191
 192 #ifdef RPC_DEBUG_DATA
 193 static void xs_pktdump(char *msg, u32 *packet, unsigned int count)
 194 {
 195         u8 *buf = (u8 *) packet;
 196         int j;
 197
 198         dprintk("RPC:       %s\n", msg);
 199         for (j = 0; j < count && j < 128; j += 4) {
 200                 if (!(j & 31)) {
 201                         if (j)
 202                                 dprintk("\n");
 203                         dprintk("0x%04x ", j);
 204                 }
 205                 dprintk("%02x%02x%02x%02x ",
 206                         buf[j], buf[j+1], buf[j+2], buf[j+3]);
 207         }
 208         dprintk("\n");
 209 }
 210 #else
 211 static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count)
 212 {
 213         /* NOP */
 214 }
 215 #endif
 216
 217 struct sock_xprt {
 218         struct rpc_xprt         xprt;
 219
 220         /*
 221          * Network layer
 222          */
 223         struct socket *         sock;
 224         struct sock *           inet;
 225
 226         /*
 227          * State of TCP reply receive
 228          */
 229         __be32                  tcp_fraghdr,
 230                                 tcp_xid,
 231                                 tcp_calldir;
 232
 233         u32                     tcp_offset,
 234                                 tcp_reclen;
 235
 236         unsigned long           tcp_copied,
 237                                 tcp_flags;
 238
 239         /*
 240          * Connection of transports
 241          */
 242         struct delayed_work     connect_worker;
 243         struct sockaddr_storage srcaddr;
 244         unsigned short          srcport;
 245
 246         /*
 247          * UDP socket buffer size parameters
 248          */
 249         size_t                  rcvsize,
 250                                 sndsize;
 251
 252         /*
 253          * Saved socket callback addresses
 254          */
 255         void                    (*old_data_ready)(struct sock *, int);
 256         void                    (*old_state_change)(struct sock *);
 257         void                    (*old_write_space)(struct sock *);
 258 };
 259
 260 /*
 261  * TCP receive state flags
 262  */
 263 #define TCP_RCV_LAST_FRAG       (1UL << 0)
 264 #define TCP_RCV_COPY_FRAGHDR    (1UL << 1)
 265 #define TCP_RCV_COPY_XID        (1UL << 2)
 266 #define TCP_RCV_COPY_DATA       (1UL << 3)
 267 #define TCP_RCV_READ_CALLDIR    (1UL << 4)
 268 #define TCP_RCV_COPY_CALLDIR    (1UL << 5)
 269
 270 /*
 271  * TCP RPC flags
 272  */
 273 #define TCP_RPC_REPLY           (1UL << 6)
 274
 275 static inline struct sockaddr *xs_addr(struct rpc_xprt *xprt)
 276 {
 277         return (struct sockaddr *) &xprt->addr;
 278 }
 279
 280 static inline struct sockaddr_un *xs_addr_un(struct rpc_xprt *xprt)
 281 {
 282         return (struct sockaddr_un *) &xprt->addr;
 283 }
 284
 285 static inline struct sockaddr_in *xs_addr_in(struct rpc_xprt *xprt)
 286 {
 287         return (struct sockaddr_in *) &xprt->addr;
 288 }
 289
 290 static inline struct sockaddr_in6 *xs_addr_in6(struct rpc_xprt *xprt)
 291 {
 292         return (struct sockaddr_in6 *) &xprt->addr;
 293 }
 294
 295 static void xs_format_common_peer_addresses(struct rpc_xprt *xprt)
 296 {
 297         struct sockaddr *sap = xs_addr(xprt);
 298         struct sockaddr_in6 *sin6;
 299         struct sockaddr_in *sin;
 300         struct sockaddr_un *sun;
 301         char buf[128];
 302
 303         switch (sap->sa_family) {
 304         case AF_LOCAL:
 305                 sun = xs_addr_un(xprt);
 306                 strlcpy(buf, sun->sun_path, sizeof(buf));
 307                 xprt->address_strings[RPC_DISPLAY_ADDR] =
 308                                                 kstrdup(buf, GFP_KERNEL);
 309                 break;
 310         case AF_INET:
 311                 (void)rpc_ntop(sap, buf, sizeof(buf));
 312                 xprt->address_strings[RPC_DISPLAY_ADDR] =
 313                                                 kstrdup(buf, GFP_KERNEL);
 314                 sin = xs_addr_in(xprt);
 315                 snprintf(buf, sizeof(buf), "%08x", ntohl(sin->sin_addr.s_addr));
 316                 break;
 317         case AF_INET6:
 318                 (void)rpc_ntop(sap, buf, sizeof(buf));
 319                 xprt->address_strings[RPC_DISPLAY_ADDR] =
 320                                                 kstrdup(buf, GFP_KERNEL);
 321                 sin6 = xs_addr_in6(xprt);
 322                 snprintf(buf, sizeof(buf), "%pi6", &sin6->sin6_addr);
 323                 break;
 324         default:
 325                 BUG();
 326         }
 327
 328         xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = kstrdup(buf, GFP_KERNEL);
 329 }
 330
 331 static void xs_format_common_peer_ports(struct rpc_xprt *xprt)
 332 {
 333         struct sockaddr *sap = xs_addr(xprt);
 334         char buf[128];
 335
 336         snprintf(buf, sizeof(buf), "%u", rpc_get_port(sap));
 337         xprt->address_strings[RPC_DISPLAY_PORT] = kstrdup(buf, GFP_KERNEL);
 338
 339         snprintf(buf, sizeof(buf), "%4hx", rpc_get_port(sap));
 340         xprt->address_strings[RPC_DISPLAY_HEX_PORT] = kstrdup(buf, GFP_KERNEL);
 341 }
 342
 343 static void xs_format_peer_addresses(struct rpc_xprt *xprt,
 344                                      const char *protocol,
 345                                      const char *netid)
 346 {
 347         xprt->address_strings[RPC_DISPLAY_PROTO] = protocol;
 348         xprt->address_strings[RPC_DISPLAY_NETID] = netid;
 349         xs_format_common_peer_addresses(xprt);
 350         xs_format_common_peer_ports(xprt);
 351 }
 352
 353 static void xs_update_peer_port(struct rpc_xprt *xprt)
 354 {
 355         kfree(xprt->address_strings[RPC_DISPLAY_HEX_PORT]);
 356         kfree(xprt->address_strings[RPC_DISPLAY_PORT]);
 357
 358         xs_format_common_peer_ports(xprt);
 359 }
 360
 361 static void xs_free_peer_addresses(struct rpc_xprt *xprt)
 362 {
 363         unsigned int i;
 364
 365         for (i = 0; i < RPC_DISPLAY_MAX; i++)
 366                 switch (i) {
 367                 case RPC_DISPLAY_PROTO:
 368                 case RPC_DISPLAY_NETID:
 369                         continue;
 370                 default:
 371                         kfree(xprt->address_strings[i]);
 372                 }
 373 }
 374
 375 #define XS_SENDMSG_FLAGS        (MSG_DONTWAIT | MSG_NOSIGNAL)
 376
 377 static int xs_send_kvec(struct socket *sock, struct sockaddr *addr, int addrlen, struct kvec *vec, unsigned int base, int more)
 378 {
 379         struct msghdr msg = {
 380                 .msg_name       = addr,
 381                 .msg_namelen    = addrlen,
 382                 .msg_flags      = XS_SENDMSG_FLAGS | (more ? MSG_MORE : 0),
 383         };
 384         struct kvec iov = {
 385                 .iov_base       = vec->iov_base + base,
 386                 .iov_len        = vec->iov_len - base,
 387         };
 388
 389         if (iov.iov_len != 0)
 390                 return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
 391         return kernel_sendmsg(sock, &msg, NULL, 0, 0);
 392 }
 393
 394 static int xs_send_pagedata(struct socket *sock, struct xdr_buf *xdr, unsigned int base, int more)
 395 {
 396         struct page **ppage;
 397         unsigned int remainder;
 398         int err, sent = 0;
 399
 400         remainder = xdr->page_len - base;
 401         base += xdr->page_base;
 402         ppage = xdr->pages + (base >> PAGE_SHIFT);
 403         base &= ~PAGE_MASK;
 404         for(;;) {
 405                 unsigned int len = min_t(unsigned int, PAGE_SIZE - base, remainder);
 406                 int flags = XS_SENDMSG_FLAGS;
 407
 408                 remainder -= len;
 409                 if (remainder != 0 || more)
 410                         flags |= MSG_MORE;
 411                 err = sock->ops->sendpage(sock, *ppage, base, len, flags);
 412                 if (remainder == 0 || err != len)
 413                         break;
 414                 sent += err;
 415                 ppage++;
 416                 base = 0;
 417         }
 418         if (sent == 0)
 419                 return err;
 420         if (err > 0)
 421                 sent += err;
 422         return sent;
 423 }
 424
 425 /**
 426  * xs_sendpages - write pages directly to a socket
 427  * @sock: socket to send on
 428  * @addr: UDP only -- address of destination
 429  * @addrlen: UDP only -- length of destination address
 430  * @xdr: buffer containing this request
 431  * @base: starting position in the buffer
 432  *
 433  */
 434 static int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base)
 435 {
 436         unsigned int remainder = xdr->len - base;
 437         int err, sent = 0;
 438
 439         if (unlikely(!sock))
 440                 return -ENOTSOCK;
 441
 442         clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags);
 443         if (base != 0) {
 444                 addr = NULL;
 445                 addrlen = 0;
 446         }
 447
 448         if (base < xdr->head[0].iov_len || addr != NULL) {
 449                 unsigned int len = xdr->head[0].iov_len - base;
 450                 remainder -= len;
 451                 err = xs_send_kvec(sock, addr, addrlen, &xdr->head[0], base, remainder != 0);
 452                 if (remainder == 0 || err != len)
 453                         goto out;
 454                 sent += err;
 455                 base = 0;
 456         } else
 457                 base -= xdr->head[0].iov_len;
 458
 459         if (base < xdr->page_len) {
 460                 unsigned int len = xdr->page_len - base;
 461                 remainder -= len;
 462                 err = xs_send_pagedata(sock, xdr, base, remainder != 0);
 463                 if (remainder == 0 || err != len)
 464                         goto out;
 465                 sent += err;
 466                 base = 0;
 467         } else
 468                 base -= xdr->page_len;
 469
 470         if (base >= xdr->tail[0].iov_len)
 471                 return sent;
 472         err = xs_send_kvec(sock, NULL, 0, &xdr->tail[0], base, 0);
 473 out:
 474         if (sent == 0)
 475                 return err;
 476         if (err > 0)
 477                 sent += err;
 478         return sent;
 479 }
 480
 481 static void xs_nospace_callback(struct rpc_task *task)
 482 {
 483         struct sock_xprt *transport = container_of(task->tk_rqstp->rq_xprt, struct sock_xprt, xprt);
 484
 485         transport->inet->sk_write_pending--;
 486         clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 487 }
 488
 489 /**
 490  * xs_nospace - place task on wait queue if transmit was incomplete
 491  * @task: task to put to sleep
 492  *
 493  */
 494 static int xs_nospace(struct rpc_task *task)
 495 {
 496         struct rpc_rqst *req = task->tk_rqstp;
 497         struct rpc_xprt *xprt = req->rq_xprt;
 498         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 499         int ret = -EAGAIN;
 500
 501         dprintk("RPC: %5u xmit incomplete (%u left of %u)\n",
 502                         task->tk_pid, req->rq_slen - req->rq_bytes_sent,
 503                         req->rq_slen);
 504
 505         /* Protect against races with write_space */
 506         spin_lock_bh(&xprt->transport_lock);
 507
 508         /* Don't race with disconnect */
 509         if (xprt_connected(xprt)) {
 510                 if (test_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags)) {
 511                         /*
 512                          * Notify TCP that we're limited by the application
 513                          * window size
 514                          */
 515                         set_bit(SOCK_NOSPACE, &transport->sock->flags);
 516                         transport->inet->sk_write_pending++;
 517                         /* ...and wait for more buffer space */
 518                         xprt_wait_for_buffer_space(task, xs_nospace_callback);
 519                 }
 520         } else {
 521                 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 522                 ret = -ENOTCONN;
 523         }
 524
 525         spin_unlock_bh(&xprt->transport_lock);
 526         return ret;
 527 }
 528
 529 /*
 530  * Construct a stream transport record marker in @buf.
 531  */
 532 static inline void xs_encode_stream_record_marker(struct xdr_buf *buf)
 533 {
 534         u32 reclen = buf->len - sizeof(rpc_fraghdr);
 535         rpc_fraghdr *base = buf->head[0].iov_base;
 536         *base = cpu_to_be32(RPC_LAST_STREAM_FRAGMENT | reclen);
 537 }
 538
 539 /**
 540  * xs_local_send_request - write an RPC request to an AF_LOCAL socket
 541  * @task: RPC task that manages the state of an RPC request
 542  *
 543  * Return values:
 544  *        0:    The request has been sent
 545  *   EAGAIN:    The socket was blocked, please call again later to
 546  *              complete the request
 547  * ENOTCONN:    Caller needs to invoke connect logic then call again
 548  *    other:    Some other error occured, the request was not sent
 549  */
 550 static int xs_local_send_request(struct rpc_task *task)
 551 {
 552         struct rpc_rqst *req = task->tk_rqstp;
 553         struct rpc_xprt *xprt = req->rq_xprt;
 554         struct sock_xprt *transport =
 555                                 container_of(xprt, struct sock_xprt, xprt);
 556         struct xdr_buf *xdr = &req->rq_snd_buf;
 557         int status;
 558
 559         xs_encode_stream_record_marker(&req->rq_snd_buf);
 560
 561         xs_pktdump("packet data:",
 562                         req->rq_svec->iov_base, req->rq_svec->iov_len);
 563
 564         status = xs_sendpages(transport->sock, NULL, 0,
 565                                                 xdr, req->rq_bytes_sent);
 566         dprintk("RPC:       %s(%u) = %d\n",
 567                         __func__, xdr->len - req->rq_bytes_sent, status);
 568         if (likely(status >= 0)) {
 569                 req->rq_bytes_sent += status;
 570                 req->rq_xmit_bytes_sent += status;
 571                 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
 572                         req->rq_bytes_sent = 0;
 573                         return 0;
 574                 }
 575                 status = -EAGAIN;
 576         }
 577
 578         switch (status) {
 579         case -EAGAIN:
 580                 status = xs_nospace(task);
 581                 break;
 582         default:
 583                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
 584                         -status);
 585         case -EPIPE:
 586                 xs_close(xprt);
 587                 status = -ENOTCONN;
 588         }
 589
 590         return status;
 591 }
 592
 593 /**
 594  * xs_udp_send_request - write an RPC request to a UDP socket
 595  * @task: address of RPC task that manages the state of an RPC request
 596  *
 597  * Return values:
 598  *        0:    The request has been sent
 599  *   EAGAIN:    The socket was blocked, please call again later to
 600  *              complete the request
 601  * ENOTCONN:    Caller needs to invoke connect logic then call again
 602  *    other:    Some other error occurred, the request was not sent
 603  */
 604 static int xs_udp_send_request(struct rpc_task *task)
 605 {
 606         struct rpc_rqst *req = task->tk_rqstp;
 607         struct rpc_xprt *xprt = req->rq_xprt;
 608         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 609         struct xdr_buf *xdr = &req->rq_snd_buf;
 610         int status;
 611
 612         xs_pktdump("packet data:",
 613                                 req->rq_svec->iov_base,
 614                                 req->rq_svec->iov_len);
 615
 616         if (!xprt_bound(xprt))
 617                 return -ENOTCONN;
 618         status = xs_sendpages(transport->sock,
 619                               xs_addr(xprt),
 620                               xprt->addrlen, xdr,
 621                               req->rq_bytes_sent);
 622
 623         dprintk("RPC:       xs_udp_send_request(%u) = %d\n",
 624                         xdr->len - req->rq_bytes_sent, status);
 625
 626         if (status >= 0) {
 627                 req->rq_xmit_bytes_sent += status;
 628                 if (status >= req->rq_slen)
 629                         return 0;
 630                 /* Still some bytes left; set up for a retry later. */
 631                 status = -EAGAIN;
 632         }
 633
 634         switch (status) {
 635         case -ENOTSOCK:
 636                 status = -ENOTCONN;
 637                 /* Should we call xs_close() here? */
 638                 break;
 639         case -EAGAIN:
 640                 status = xs_nospace(task);
 641                 break;
 642         default:
 643                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
 644                         -status);
 645         case -ENETUNREACH:
 646         case -EPIPE:
 647         case -ECONNREFUSED:
 648                 /* When the server has died, an ICMP port unreachable message
 649                  * prompts ECONNREFUSED. */
 650                 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 651         }
 652
 653         return status;
 654 }
 655
 656 /**
 657  * xs_tcp_shutdown - gracefully shut down a TCP socket
 658  * @xprt: transport
 659  *
 660  * Initiates a graceful shutdown of the TCP socket by calling the
 661  * equivalent of shutdown(SHUT_WR);
 662  */
 663 static void xs_tcp_shutdown(struct rpc_xprt *xprt)
 664 {
 665         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 666         struct socket *sock = transport->sock;
 667
 668         if (sock != NULL)
 669                 kernel_sock_shutdown(sock, SHUT_WR);
 670 }
 671
 672 /**
 673  * xs_tcp_send_request - write an RPC request to a TCP socket
 674  * @task: address of RPC task that manages the state of an RPC request
 675  *
 676  * Return values:
 677  *        0:    The request has been sent
 678  *   EAGAIN:    The socket was blocked, please call again later to
 679  *              complete the request
 680  * ENOTCONN:    Caller needs to invoke connect logic then call again
 681  *    other:    Some other error occurred, the request was not sent
 682  *
 683  * XXX: In the case of soft timeouts, should we eventually give up
 684  *      if sendmsg is not able to make progress?
 685  */
 686 static int xs_tcp_send_request(struct rpc_task *task)
 687 {
 688         struct rpc_rqst *req = task->tk_rqstp;
 689         struct rpc_xprt *xprt = req->rq_xprt;
 690         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 691         struct xdr_buf *xdr = &req->rq_snd_buf;
 692         int status;
 693
 694         xs_encode_stream_record_marker(&req->rq_snd_buf);
 695
 696         xs_pktdump("packet data:",
 697                                 req->rq_svec->iov_base,
 698                                 req->rq_svec->iov_len);
 699
 700         /* Continue transmitting the packet/record. We must be careful
 701          * to cope with writespace callbacks arriving _after_ we have
 702          * called sendmsg(). */
 703         while (1) {
 704                 status = xs_sendpages(transport->sock,
 705                                         NULL, 0, xdr, req->rq_bytes_sent);
 706
 707                 dprintk("RPC:       xs_tcp_send_request(%u) = %d\n",
 708                                 xdr->len - req->rq_bytes_sent, status);
 709
 710                 if (unlikely(status < 0))
 711                         break;
 712
 713                 /* If we've sent the entire packet, immediately
 714                  * reset the count of bytes sent. */
 715                 req->rq_bytes_sent += status;
 716                 req->rq_xmit_bytes_sent += status;
 717                 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
 718                         req->rq_bytes_sent = 0;
 719                         return 0;
 720                 }
 721
 722                 if (status != 0)
 723                         continue;
 724                 status = -EAGAIN;
 725                 break;
 726         }
 727
 728         switch (status) {
 729         case -ENOTSOCK:
 730                 status = -ENOTCONN;
 731                 /* Should we call xs_close() here? */
 732                 break;
 733         case -EAGAIN:
 734                 status = xs_nospace(task);
 735                 break;
 736         default:
 737                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
 738                         -status);
 739         case -ECONNRESET:
 740                 xs_tcp_shutdown(xprt);
 741         case -ECONNREFUSED:
 742         case -ENOTCONN:
 743         case -EPIPE:
 744                 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 745         }
 746
 747         return status;
 748 }
 749
 750 /**
 751  * xs_tcp_release_xprt - clean up after a tcp transmission
 752  * @xprt: transport
 753  * @task: rpc task
 754  *
 755  * This cleans up if an error causes us to abort the transmission of a request.
 756  * In this case, the socket may need to be reset in order to avoid confusing
 757  * the server.
 758  */
 759 static void xs_tcp_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
 760 {
 761         struct rpc_rqst *req;
 762
 763         if (task != xprt->snd_task)
 764                 return;
 765         if (task == NULL)
 766                 goto out_release;
 767         req = task->tk_rqstp;
 768         if (req == NULL)
 769                 goto out_release;
 770         if (req->rq_bytes_sent == 0)
 771                 goto out_release;
 772         if (req->rq_bytes_sent == req->rq_snd_buf.len)
 773                 goto out_release;
 774         set_bit(XPRT_CLOSE_WAIT, &xprt->state);
 775 out_release:
 776         xprt_release_xprt(xprt, task);
 777 }
 778
 779 static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk)
 780 {
 781         transport->old_data_ready = sk->sk_data_ready;
 782         transport->old_state_change = sk->sk_state_change;
 783         transport->old_write_space = sk->sk_write_space;
 784 }
 785
 786 static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk)
 787 {
 788         sk->sk_data_ready = transport->old_data_ready;
 789         sk->sk_state_change = transport->old_state_change;
 790         sk->sk_write_space = transport->old_write_space;
 791 }
 792
 793 static void xs_reset_transport(struct sock_xprt *transport)
 794 {
 795         struct socket *sock = transport->sock;
 796         struct sock *sk = transport->inet;
 797
 798         if (sk == NULL)
 799                 return;
 800
 801         transport->srcport = 0;
 802
 803         write_lock_bh(&sk->sk_callback_lock);
 804         transport->inet = NULL;
 805         transport->sock = NULL;
 806
 807         sk->sk_user_data = NULL;
 808
 809         xs_restore_old_callbacks(transport, sk);
 810         write_unlock_bh(&sk->sk_callback_lock);
 811
 812         sk->sk_no_check = 0;
 813
 814         sock_release(sock);
 815 }
 816
 817 /**
 818  * xs_close - close a socket
 819  * @xprt: transport
 820  *
 821  * This is used when all requests are complete; ie, no DRC state remains
 822  * on the server we want to save.
 823  *
 824  * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with
 825  * xs_reset_transport() zeroing the socket from underneath a writer.
 826  */
 827 static void xs_close(struct rpc_xprt *xprt)
 828 {
 829         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 830
 831         dprintk("RPC:       xs_close xprt %p\n", xprt);
 832
 833         xs_reset_transport(transport);
 834         xprt->reestablish_timeout = 0;
 835
 836         smp_mb__before_clear_bit();
 837         clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
 838         clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
 839         clear_bit(XPRT_CLOSING, &xprt->state);
 840         smp_mb__after_clear_bit();
 841         xprt_disconnect_done(xprt);
 842 }
 843
 844 static void xs_tcp_close(struct rpc_xprt *xprt)
 845 {
 846         if (test_and_clear_bit(XPRT_CONNECTION_CLOSE, &xprt->state))
 847                 xs_close(xprt);
 848         else
 849                 xs_tcp_shutdown(xprt);
 850 }
 851
 852 /**
 853  * xs_destroy - prepare to shutdown a transport
 854  * @xprt: doomed transport
 855  *
 856  */
 857 static void xs_destroy(struct rpc_xprt *xprt)
 858 {
 859         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 860
 861         dprintk("RPC:       xs_destroy xprt %p\n", xprt);
 862
 863         cancel_delayed_work_sync(&transport->connect_worker);
 864
 865         xs_close(xprt);
 866         xs_free_peer_addresses(xprt);
 867         xprt_free(xprt);
 868         module_put(THIS_MODULE);
 869 }
 870
 871 static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
 872 {
 873         return (struct rpc_xprt *) sk->sk_user_data;
 874 }
 875
 876 static int xs_local_copy_to_xdr(struct xdr_buf *xdr, struct sk_buff *skb)
 877 {
 878         struct xdr_skb_reader desc = {
 879                 .skb            = skb,
 880                 .offset         = sizeof(rpc_fraghdr),
 881                 .count          = skb->len - sizeof(rpc_fraghdr),
 882         };
 883
 884         if (xdr_partial_copy_from_skb(xdr, 0, &desc, xdr_skb_read_bits) < 0)
 885                 return -1;
 886         if (desc.count)
 887                 return -1;
 888         return 0;
 889 }
 890
 891 /**
 892  * xs_local_data_ready - "data ready" callback for AF_LOCAL sockets
 893  * @sk: socket with data to read
 894  * @len: how much data to read
 895  *
 896  * Currently this assumes we can read the whole reply in a single gulp.
 897  */
 898 static void xs_local_data_ready(struct sock *sk, int len)
 899 {
 900         struct rpc_task *task;
 901         struct rpc_xprt *xprt;
 902         struct rpc_rqst *rovr;
 903         struct sk_buff *skb;
 904         int err, repsize, copied;
 905         u32 _xid;
 906         __be32 *xp;
 907
 908         read_lock_bh(&sk->sk_callback_lock);
 909         dprintk("RPC:       %s...\n", __func__);
 910         xprt = xprt_from_sock(sk);
 911         if (xprt == NULL)
 912                 goto out;
 913
 914         skb = skb_recv_datagram(sk, 0, 1, &err);
 915         if (skb == NULL)
 916                 goto out;
 917
 918         repsize = skb->len - sizeof(rpc_fraghdr);
 919         if (repsize < 4) {
 920                 dprintk("RPC:       impossible RPC reply size %d\n", repsize);
 921                 goto dropit;
 922         }
 923
 924         /* Copy the XID from the skb... */
 925         xp = skb_header_pointer(skb, sizeof(rpc_fraghdr), sizeof(_xid), &_xid);
 926         if (xp == NULL)
 927                 goto dropit;
 928
 929         /* Look up and lock the request corresponding to the given XID */
 930         spin_lock(&xprt->transport_lock);
 931         rovr = xprt_lookup_rqst(xprt, *xp);
 932         if (!rovr)
 933                 goto out_unlock;
 934         task = rovr->rq_task;
 935
 936         copied = rovr->rq_private_buf.buflen;
 937         if (copied > repsize)
 938                 copied = repsize;
 939
 940         if (xs_local_copy_to_xdr(&rovr->rq_private_buf, skb)) {
 941                 dprintk("RPC:       sk_buff copy failed\n");
 942                 goto out_unlock;
 943         }
 944
 945         xprt_complete_rqst(task, copied);
 946
 947  out_unlock:
 948         spin_unlock(&xprt->transport_lock);
 949  dropit:
 950         skb_free_datagram(sk, skb);
 951  out:
 952         read_unlock_bh(&sk->sk_callback_lock);
 953 }
 954
 955 /**
 956  * xs_udp_data_ready - "data ready" callback for UDP sockets
 957  * @sk: socket with data to read
 958  * @len: how much data to read
 959  *
 960  */
 961 static void xs_udp_data_ready(struct sock *sk, int len)
 962 {
 963         struct rpc_task *task;
 964         struct rpc_xprt *xprt;
 965         struct rpc_rqst *rovr;
 966         struct sk_buff *skb;
 967         int err, repsize, copied;
 968         u32 _xid;
 969         __be32 *xp;
 970
 971         read_lock_bh(&sk->sk_callback_lock);
 972         dprintk("RPC:       xs_udp_data_ready...\n");
 973         if (!(xprt = xprt_from_sock(sk)))
 974                 goto out;
 975
 976         if ((skb = skb_recv_datagram(sk, 0, 1, &err)) == NULL)
 977                 goto out;
 978
 979         repsize = skb->len - sizeof(struct udphdr);
 980         if (repsize < 4) {
 981                 dprintk("RPC:       impossible RPC reply size %d!\n", repsize);
 982                 goto dropit;
 983         }
 984
 985         /* Copy the XID from the skb... */
 986         xp = skb_header_pointer(skb, sizeof(struct udphdr),
 987                                 sizeof(_xid), &_xid);
 988         if (xp == NULL)
 989                 goto dropit;
 990
 991         /* Look up and lock the request corresponding to the given XID */
 992         spin_lock(&xprt->transport_lock);
 993         rovr = xprt_lookup_rqst(xprt, *xp);
 994         if (!rovr)
 995                 goto out_unlock;
 996         task = rovr->rq_task;
 997
 998         if ((copied = rovr->rq_private_buf.buflen) > repsize)
 999                 copied = repsize;
1000
1001         /* Suck it into the iovec, verify checksum if not done by hw. */
1002         if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
1003                 UDPX_INC_STATS_BH(sk, UDP_MIB_INERRORS);
1004                 goto out_unlock;
1005         }
1006
1007         UDPX_INC_STATS_BH(sk, UDP_MIB_INDATAGRAMS);
1008
1009         xprt_adjust_cwnd(xprt, task, copied);
1010         xprt_complete_rqst(task, copied);
1011
1012  out_unlock:
1013         spin_unlock(&xprt->transport_lock);
1014  dropit:
1015         skb_free_datagram(sk, skb);
1016  out:
1017         read_unlock_bh(&sk->sk_callback_lock);
1018 }
1019
1020 /*
1021  * Helper function to force a TCP close if the server is sending
1022  * junk and/or it has put us in CLOSE_WAIT
1023  */
1024 static void xs_tcp_force_close(struct rpc_xprt *xprt)
1025 {
1026         set_bit(XPRT_CONNECTION_CLOSE, &xprt->state);
1027         xprt_force_disconnect(xprt);
1028 }
1029
1030 static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
1031 {
1032         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1033         size_t len, used;
1034         char *p;
1035
1036         p = ((char *) &transport->tcp_fraghdr) + transport->tcp_offset;
1037         len = sizeof(transport->tcp_fraghdr) - transport->tcp_offset;
1038         used = xdr_skb_read_bits(desc, p, len);
1039         transport->tcp_offset += used;
1040         if (used != len)
1041                 return;
1042
1043         transport->tcp_reclen = ntohl(transport->tcp_fraghdr);
1044         if (transport->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
1045                 transport->tcp_flags |= TCP_RCV_LAST_FRAG;
1046         else
1047                 transport->tcp_flags &= ~TCP_RCV_LAST_FRAG;
1048         transport->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
1049
1050         transport->tcp_flags &= ~TCP_RCV_COPY_FRAGHDR;
1051         transport->tcp_offset = 0;
1052
1053         /* Sanity check of the record length */
1054         if (unlikely(transport->tcp_reclen < 8)) {
1055                 dprintk("RPC:       invalid TCP record fragment length\n");
1056                 xs_tcp_force_close(xprt);
1057                 return;
1058         }
1059         dprintk("RPC:       reading TCP record fragment of length %d\n",
1060                         transport->tcp_reclen);
1061 }
1062
1063 static void xs_tcp_check_fraghdr(struct sock_xprt *transport)
1064 {
1065         if (transport->tcp_offset == transport->tcp_reclen) {
1066                 transport->tcp_flags |= TCP_RCV_COPY_FRAGHDR;
1067                 transport->tcp_offset = 0;
1068                 if (transport->tcp_flags & TCP_RCV_LAST_FRAG) {
1069                         transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1070                         transport->tcp_flags |= TCP_RCV_COPY_XID;
1071                         transport->tcp_copied = 0;
1072                 }
1073         }
1074 }
1075
1076 static inline void xs_tcp_read_xid(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1077 {
1078         size_t len, used;
1079         char *p;
1080
1081         len = sizeof(transport->tcp_xid) - transport->tcp_offset;
1082         dprintk("RPC:       reading XID (%Zu bytes)\n", len);
1083         p = ((char *) &transport->tcp_xid) + transport->tcp_offset;
1084         used = xdr_skb_read_bits(desc, p, len);
1085         transport->tcp_offset += used;
1086         if (used != len)
1087                 return;
1088         transport->tcp_flags &= ~TCP_RCV_COPY_XID;
1089         transport->tcp_flags |= TCP_RCV_READ_CALLDIR;
1090         transport->tcp_copied = 4;
1091         dprintk("RPC:       reading %s XID %08x\n",
1092                         (transport->tcp_flags & TCP_RPC_REPLY) ? "reply for"
1093                                                               : "request with",
1094                         ntohl(transport->tcp_xid));
1095         xs_tcp_check_fraghdr(transport);
1096 }
1097
1098 static inline void xs_tcp_read_calldir(struct sock_xprt *transport,
1099                                        struct xdr_skb_reader *desc)
1100 {
1101         size_t len, used;
1102         u32 offset;
1103         char *p;
1104
1105         /*
1106          * We want transport->tcp_offset to be 8 at the end of this routine
1107          * (4 bytes for the xid and 4 bytes for the call/reply flag).
1108          * When this function is called for the first time,
1109          * transport->tcp_offset is 4 (after having already read the xid).
1110          */
1111         offset = transport->tcp_offset - sizeof(transport->tcp_xid);
1112         len = sizeof(transport->tcp_calldir) - offset;
1113         dprintk("RPC:       reading CALL/REPLY flag (%Zu bytes)\n", len);
1114         p = ((char *) &transport->tcp_calldir) + offset;
1115         used = xdr_skb_read_bits(desc, p, len);
1116         transport->tcp_offset += used;
1117         if (used != len)
1118                 return;
1119         transport->tcp_flags &= ~TCP_RCV_READ_CALLDIR;
1120         /*
1121          * We don't yet have the XDR buffer, so we will write the calldir
1122          * out after we get the buffer from the 'struct rpc_rqst'
1123          */
1124         switch (ntohl(transport->tcp_calldir)) {
1125         case RPC_REPLY:
1126                 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1127                 transport->tcp_flags |= TCP_RCV_COPY_DATA;
1128                 transport->tcp_flags |= TCP_RPC_REPLY;
1129                 break;
1130         case RPC_CALL:
1131                 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1132                 transport->tcp_flags |= TCP_RCV_COPY_DATA;
1133                 transport->tcp_flags &= ~TCP_RPC_REPLY;
1134                 break;
1135         default:
1136                 dprintk("RPC:       invalid request message type\n");
1137                 xs_tcp_force_close(&transport->xprt);
1138         }
1139         xs_tcp_check_fraghdr(transport);
1140 }
1141
1142 static inline void xs_tcp_read_common(struct rpc_xprt *xprt,
1143                                      struct xdr_skb_reader *desc,
1144                                      struct rpc_rqst *req)
1145 {
1146         struct sock_xprt *transport =
1147                                 container_of(xprt, struct sock_xprt, xprt);
1148         struct xdr_buf *rcvbuf;
1149         size_t len;
1150         ssize_t r;
1151
1152         rcvbuf = &req->rq_private_buf;
1153
1154         if (transport->tcp_flags & TCP_RCV_COPY_CALLDIR) {
1155                 /*
1156                  * Save the RPC direction in the XDR buffer
1157                  */
1158                 memcpy(rcvbuf->head[0].iov_base + transport->tcp_copied,
1159                         &transport->tcp_calldir,
1160                         sizeof(transport->tcp_calldir));
1161                 transport->tcp_copied += sizeof(transport->tcp_calldir);
1162                 transport->tcp_flags &= ~TCP_RCV_COPY_CALLDIR;
1163         }
1164
1165         len = desc->count;
1166         if (len > transport->tcp_reclen - transport->tcp_offset) {
1167                 struct xdr_skb_reader my_desc;
1168
1169                 len = transport->tcp_reclen - transport->tcp_offset;
1170                 memcpy(&my_desc, desc, sizeof(my_desc));
1171                 my_desc.count = len;
1172                 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
1173                                           &my_desc, xdr_skb_read_bits);
1174                 desc->count -= r;
1175                 desc->offset += r;
1176         } else
1177                 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
1178                                           desc, xdr_skb_read_bits);
1179
1180         if (r > 0) {
1181                 transport->tcp_copied += r;
1182                 transport->tcp_offset += r;
1183         }
1184         if (r != len) {
1185                 /* Error when copying to the receive buffer,
1186                  * usually because we weren't able to allocate
1187                  * additional buffer pages. All we can do now
1188                  * is turn off TCP_RCV_COPY_DATA, so the request
1189                  * will not receive any additional updates,
1190                  * and time out.
1191                  * Any remaining data from this record will
1192                  * be discarded.
1193                  */
1194                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1195                 dprintk("RPC:       XID %08x truncated request\n",
1196                                 ntohl(transport->tcp_xid));
1197                 dprintk("RPC:       xprt = %p, tcp_copied = %lu, "
1198                                 "tcp_offset = %u, tcp_reclen = %u\n",
1199                                 xprt, transport->tcp_copied,
1200                                 transport->tcp_offset, transport->tcp_reclen);
1201                 return;
1202         }
1203
1204         dprintk("RPC:       XID %08x read %Zd bytes\n",
1205                         ntohl(transport->tcp_xid), r);
1206         dprintk("RPC:       xprt = %p, tcp_copied = %lu, tcp_offset = %u, "
1207                         "tcp_reclen = %u\n", xprt, transport->tcp_copied,
1208                         transport->tcp_offset, transport->tcp_reclen);
1209
1210         if (transport->tcp_copied == req->rq_private_buf.buflen)
1211                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1212         else if (transport->tcp_offset == transport->tcp_reclen) {
1213                 if (transport->tcp_flags & TCP_RCV_LAST_FRAG)
1214                         transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1215         }
1216 }
1217
1218 /*
1219  * Finds the request corresponding to the RPC xid and invokes the common
1220  * tcp read code to read the data.
1221  */
1222 static inline int xs_tcp_read_reply(struct rpc_xprt *xprt,
1223                                     struct xdr_skb_reader *desc)
1224 {
1225         struct sock_xprt *transport =
1226                                 container_of(xprt, struct sock_xprt, xprt);
1227         struct rpc_rqst *req;
1228
1229         dprintk("RPC:       read reply XID %08x\n", ntohl(transport->tcp_xid));
1230
1231         /* Find and lock the request corresponding to this xid */
1232         spin_lock(&xprt->transport_lock);
1233         req = xprt_lookup_rqst(xprt, transport->tcp_xid);
1234         if (!req) {
1235                 dprintk("RPC:       XID %08x request not found!\n",
1236                                 ntohl(transport->tcp_xid));
1237                 spin_unlock(&xprt->transport_lock);
1238                 return -1;
1239         }
1240
1241         xs_tcp_read_common(xprt, desc, req);
1242
1243         if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1244                 xprt_complete_rqst(req->rq_task, transport->tcp_copied);
1245
1246         spin_unlock(&xprt->transport_lock);
1247         return 0;
1248 }
1249
1250 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1251 /*
1252  * Obtains an rpc_rqst previously allocated and invokes the common
1253  * tcp read code to read the data.  The result is placed in the callback
1254  * queue.
1255  * If we're unable to obtain the rpc_rqst we schedule the closing of the
1256  * connection and return -1.
1257  */
1258 static inline int xs_tcp_read_callback(struct rpc_xprt *xprt,
1259                                        struct xdr_skb_reader *desc)
1260 {
1261         struct sock_xprt *transport =
1262                                 container_of(xprt, struct sock_xprt, xprt);
1263         struct rpc_rqst *req;
1264
1265         req = xprt_alloc_bc_request(xprt);
1266         if (req == NULL) {
1267                 printk(KERN_WARNING "Callback slot table overflowed\n");
1268                 xprt_force_disconnect(xprt);
1269                 return -1;
1270         }
1271
1272         req->rq_xid = transport->tcp_xid;
1273         dprintk("RPC:       read callback  XID %08x\n", ntohl(req->rq_xid));
1274         xs_tcp_read_common(xprt, desc, req);
1275
1276         if (!(transport->tcp_flags & TCP_RCV_COPY_DATA)) {
1277                 struct svc_serv *bc_serv = xprt->bc_serv;
1278
1279                 /*
1280                  * Add callback request to callback list.  The callback
1281                  * service sleeps on the sv_cb_waitq waiting for new
1282                  * requests.  Wake it up after adding enqueing the
1283                  * request.
1284                  */
1285                 dprintk("RPC:       add callback request to list\n");
1286                 spin_lock(&bc_serv->sv_cb_lock);
1287                 list_add(&req->rq_bc_list, &bc_serv->sv_cb_list);
1288                 spin_unlock(&bc_serv->sv_cb_lock);
1289                 wake_up(&bc_serv->sv_cb_waitq);
1290         }
1291
1292         req->rq_private_buf.len = transport->tcp_copied;
1293
1294         return 0;
1295 }
1296
1297 static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1298                                         struct xdr_skb_reader *desc)
1299 {
1300         struct sock_xprt *transport =
1301                                 container_of(xprt, struct sock_xprt, xprt);
1302
1303         return (transport->tcp_flags & TCP_RPC_REPLY) ?
1304                 xs_tcp_read_reply(xprt, desc) :
1305                 xs_tcp_read_callback(xprt, desc);
1306 }
1307 #else
1308 static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1309                                         struct xdr_skb_reader *desc)
1310 {
1311         return xs_tcp_read_reply(xprt, desc);
1312 }
1313 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1314
1315 /*
1316  * Read data off the transport.  This can be either an RPC_CALL or an
1317  * RPC_REPLY.  Relay the processing to helper functions.
1318  */
1319 static void xs_tcp_read_data(struct rpc_xprt *xprt,
1320                                     struct xdr_skb_reader *desc)
1321 {
1322         struct sock_xprt *transport =
1323                                 container_of(xprt, struct sock_xprt, xprt);
1324
1325         if (_xs_tcp_read_data(xprt, desc) == 0)
1326                 xs_tcp_check_fraghdr(transport);
1327         else {
1328                 /*
1329                  * The transport_lock protects the request handling.
1330                  * There's no need to hold it to update the tcp_flags.
1331                  */
1332                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1333         }
1334 }
1335
1336 static inline void xs_tcp_read_discard(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1337 {
1338         size_t len;
1339
1340         len = transport->tcp_reclen - transport->tcp_offset;
1341         if (len > desc->count)
1342                 len = desc->count;
1343         desc->count -= len;
1344         desc->offset += len;
1345         transport->tcp_offset += len;
1346         dprintk("RPC:       discarded %Zu bytes\n", len);
1347         xs_tcp_check_fraghdr(transport);
1348 }
1349
1350 static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
1351 {
1352         struct rpc_xprt *xprt = rd_desc->arg.data;
1353         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1354         struct xdr_skb_reader desc = {
1355                 .skb    = skb,
1356                 .offset = offset,
1357                 .count  = len,
1358         };
1359
1360         dprintk("RPC:       xs_tcp_data_recv started\n");
1361         do {
1362                 /* Read in a new fragment marker if necessary */
1363                 /* Can we ever really expect to get completely empty fragments? */
1364                 if (transport->tcp_flags & TCP_RCV_COPY_FRAGHDR) {
1365                         xs_tcp_read_fraghdr(xprt, &desc);
1366                         continue;
1367                 }
1368                 /* Read in the xid if necessary */
1369                 if (transport->tcp_flags & TCP_RCV_COPY_XID) {
1370                         xs_tcp_read_xid(transport, &desc);
1371                         continue;
1372                 }
1373                 /* Read in the call/reply flag */
1374                 if (transport->tcp_flags & TCP_RCV_READ_CALLDIR) {
1375                         xs_tcp_read_calldir(transport, &desc);
1376                         continue;
1377                 }
1378                 /* Read in the request data */
1379                 if (transport->tcp_flags & TCP_RCV_COPY_DATA) {
1380                         xs_tcp_read_data(xprt, &desc);
1381                         continue;
1382                 }
1383                 /* Skip over any trailing bytes on short reads */
1384                 xs_tcp_read_discard(transport, &desc);
1385         } while (desc.count);
1386         dprintk("RPC:       xs_tcp_data_recv done\n");
1387         return len - desc.count;
1388 }
1389
1390 /**
1391  * xs_tcp_data_ready - "data ready" callback for TCP sockets
1392  * @sk: socket with data to read
1393  * @bytes: how much data to read
1394  *
1395  */
1396 static void xs_tcp_data_ready(struct sock *sk, int bytes)
1397 {
1398         struct rpc_xprt *xprt;
1399         read_descriptor_t rd_desc;
1400         int read;
1401
1402         dprintk("RPC:       xs_tcp_data_ready...\n");
1403
1404         read_lock_bh(&sk->sk_callback_lock);
1405         if (!(xprt = xprt_from_sock(sk)))
1406                 goto out;
1407         /* Any data means we had a useful conversation, so
1408          * the we don't need to delay the next reconnect
1409          */
1410         if (xprt->reestablish_timeout)
1411                 xprt->reestablish_timeout = 0;
1412
1413         /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
1414         rd_desc.arg.data = xprt;
1415         do {
1416                 rd_desc.count = 65536;
1417                 read = tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
1418         } while (read > 0);
1419 out:
1420         read_unlock_bh(&sk->sk_callback_lock);
1421 }
1422
1423 /*
1424  * Do the equivalent of linger/linger2 handling for dealing with
1425  * broken servers that don't close the socket in a timely
1426  * fashion
1427  */
1428 static void xs_tcp_schedule_linger_timeout(struct rpc_xprt *xprt,
1429                 unsigned long timeout)
1430 {
1431         struct sock_xprt *transport;
1432
1433         if (xprt_test_and_set_connecting(xprt))
1434                 return;
1435         set_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1436         transport = container_of(xprt, struct sock_xprt, xprt);
1437         queue_delayed_work(rpciod_workqueue, &transport->connect_worker,
1438                            timeout);
1439 }
1440
1441 static void xs_tcp_cancel_linger_timeout(struct rpc_xprt *xprt)
1442 {
1443         struct sock_xprt *transport;
1444
1445         transport = container_of(xprt, struct sock_xprt, xprt);
1446
1447         if (!test_bit(XPRT_CONNECTION_ABORT, &xprt->state) ||
1448             !cancel_delayed_work(&transport->connect_worker))
1449                 return;
1450         clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1451         xprt_clear_connecting(xprt);
1452 }
1453
1454 static void xs_sock_reset_connection_flags(struct rpc_xprt *xprt)
1455 {
1456         smp_mb__before_clear_bit();
1457         clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1458         clear_bit(XPRT_CONNECTION_CLOSE, &xprt->state);
1459         clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1460         clear_bit(XPRT_CLOSING, &xprt->state);
1461         smp_mb__after_clear_bit();
1462 }
1463
1464 static void xs_sock_mark_closed(struct rpc_xprt *xprt)
1465 {
1466         xs_sock_reset_connection_flags(xprt);
1467         /* Mark transport as closed and wake up all pending tasks */
1468         xprt_disconnect_done(xprt);
1469 }
1470
1471 /**
1472  * xs_tcp_state_change - callback to handle TCP socket state changes
1473  * @sk: socket whose state has changed
1474  *
1475  */
1476 static void xs_tcp_state_change(struct sock *sk)
1477 {
1478         struct rpc_xprt *xprt;
1479
1480         read_lock_bh(&sk->sk_callback_lock);
1481         if (!(xprt = xprt_from_sock(sk)))
1482                 goto out;
1483         dprintk("RPC:       xs_tcp_state_change client %p...\n", xprt);
1484         dprintk("RPC:       state %x conn %d dead %d zapped %d sk_shutdown %d\n",
1485                         sk->sk_state, xprt_connected(xprt),
1486                         sock_flag(sk, SOCK_DEAD),
1487                         sock_flag(sk, SOCK_ZAPPED),
1488                         sk->sk_shutdown);
1489
1490         switch (sk->sk_state) {
1491         case TCP_ESTABLISHED:
1492                 spin_lock(&xprt->transport_lock);
1493                 if (!xprt_test_and_set_connected(xprt)) {
1494                         struct sock_xprt *transport = container_of(xprt,
1495                                         struct sock_xprt, xprt);
1496
1497                         /* Reset TCP record info */
1498                         transport->tcp_offset = 0;
1499                         transport->tcp_reclen = 0;
1500                         transport->tcp_copied = 0;
1501                         transport->tcp_flags =
1502                                 TCP_RCV_COPY_FRAGHDR | TCP_RCV_COPY_XID;
1503
1504                         xprt_wake_pending_tasks(xprt, -EAGAIN);
1505                 }
1506                 spin_unlock(&xprt->transport_lock);
1507                 break;
1508         case TCP_FIN_WAIT1:
1509                 /* The client initiated a shutdown of the socket */
1510                 xprt->connect_cookie++;
1511                 xprt->reestablish_timeout = 0;
1512                 set_bit(XPRT_CLOSING, &xprt->state);
1513                 smp_mb__before_clear_bit();
1514                 clear_bit(XPRT_CONNECTED, &xprt->state);
1515                 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1516                 smp_mb__after_clear_bit();
1517                 xs_tcp_schedule_linger_timeout(xprt, xs_tcp_fin_timeout);
1518                 break;
1519         case TCP_CLOSE_WAIT:
1520                 /* The server initiated a shutdown of the socket */
1521                 xprt->connect_cookie++;
1522                 clear_bit(XPRT_CONNECTED, &xprt->state);
1523                 xs_tcp_force_close(xprt);
1524         case TCP_CLOSING:
1525                 /*
1526                  * If the server closed down the connection, make sure that
1527                  * we back off before reconnecting
1528                  */
1529                 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1530                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1531                 break;
1532         case TCP_LAST_ACK:
1533                 set_bit(XPRT_CLOSING, &xprt->state);
1534                 xs_tcp_schedule_linger_timeout(xprt, xs_tcp_fin_timeout);
1535                 smp_mb__before_clear_bit();
1536                 clear_bit(XPRT_CONNECTED, &xprt->state);
1537                 smp_mb__after_clear_bit();
1538                 break;
1539         case TCP_CLOSE:
1540                 xs_tcp_cancel_linger_timeout(xprt);
1541                 xs_sock_mark_closed(xprt);
1542         }
1543  out:
1544         read_unlock_bh(&sk->sk_callback_lock);
1545 }
1546
1547 static void xs_write_space(struct sock *sk)
1548 {
1549         struct socket *sock;
1550         struct rpc_xprt *xprt;
1551
1552         if (unlikely(!(sock = sk->sk_socket)))
1553                 return;
1554         clear_bit(SOCK_NOSPACE, &sock->flags);
1555
1556         if (unlikely(!(xprt = xprt_from_sock(sk))))
1557                 return;
1558         if (test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags) == 0)
1559                 return;
1560
1561         xprt_write_space(xprt);
1562 }
1563
1564 /**
1565  * xs_udp_write_space - callback invoked when socket buffer space
1566  *                             becomes available
1567  * @sk: socket whose state has changed
1568  *
1569  * Called when more output buffer space is available for this socket.
1570  * We try not to wake our writers until they can make "significant"
1571  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1572  * with a bunch of small requests.
1573  */
1574 static void xs_udp_write_space(struct sock *sk)
1575 {
1576         read_lock_bh(&sk->sk_callback_lock);
1577
1578         /* from net/core/sock.c:sock_def_write_space */
1579         if (sock_writeable(sk))
1580                 xs_write_space(sk);
1581
1582         read_unlock_bh(&sk->sk_callback_lock);
1583 }
1584
1585 /**
1586  * xs_tcp_write_space - callback invoked when socket buffer space
1587  *                             becomes available
1588  * @sk: socket whose state has changed
1589  *
1590  * Called when more output buffer space is available for this socket.
1591  * We try not to wake our writers until they can make "significant"
1592  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1593  * with a bunch of small requests.
1594  */
1595 static void xs_tcp_write_space(struct sock *sk)
1596 {
1597         read_lock_bh(&sk->sk_callback_lock);
1598
1599         /* from net/core/stream.c:sk_stream_write_space */
1600         if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk))
1601                 xs_write_space(sk);
1602
1603         read_unlock_bh(&sk->sk_callback_lock);
1604 }
1605
1606 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
1607 {
1608         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1609         struct sock *sk = transport->inet;
1610
1611         if (transport->rcvsize) {
1612                 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1613                 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
1614         }
1615         if (transport->sndsize) {
1616                 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1617                 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1618                 sk->sk_write_space(sk);
1619         }
1620 }
1621
1622 /**
1623  * xs_udp_set_buffer_size - set send and receive limits
1624  * @xprt: generic transport
1625  * @sndsize: requested size of send buffer, in bytes
1626  * @rcvsize: requested size of receive buffer, in bytes
1627  *
1628  * Set socket send and receive buffer size limits.
1629  */
1630 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1631 {
1632         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1633
1634         transport->sndsize = 0;
1635         if (sndsize)
1636                 transport->sndsize = sndsize + 1024;
1637         transport->rcvsize = 0;
1638         if (rcvsize)
1639                 transport->rcvsize = rcvsize + 1024;
1640
1641         xs_udp_do_set_buffer_size(xprt);
1642 }
1643
1644 /**
1645  * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1646  * @task: task that timed out
1647  *
1648  * Adjust the congestion window after a retransmit timeout has occurred.
1649  */
1650 static void xs_udp_timer(struct rpc_xprt *xprt, struct rpc_task *task)
1651 {
1652         xprt_adjust_cwnd(xprt, task, -ETIMEDOUT);
1653 }
1654
1655 static unsigned short xs_get_random_port(void)
1656 {
1657         unsigned short range = xprt_max_resvport - xprt_min_resvport;
1658         unsigned short rand = (unsigned short) net_random() % range;
1659         return rand + xprt_min_resvport;
1660 }
1661
1662 /**
1663  * xs_set_port - reset the port number in the remote endpoint address
1664  * @xprt: generic transport
1665  * @port: new port number
1666  *
1667  */
1668 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1669 {
1670         dprintk("RPC:       setting port for xprt %p to %u\n", xprt, port);
1671
1672         rpc_set_port(xs_addr(xprt), port);
1673         xs_update_peer_port(xprt);
1674 }
1675
1676 static unsigned short xs_get_srcport(struct sock_xprt *transport)
1677 {
1678         unsigned short port = transport->srcport;
1679
1680         if (port == 0 && transport->xprt.resvport)
1681                 port = xs_get_random_port();
1682         return port;
1683 }
1684
1685 static unsigned short xs_next_srcport(struct sock_xprt *transport, unsigned short port)
1686 {
1687         if (transport->srcport != 0)
1688                 transport->srcport = 0;
1689         if (!transport->xprt.resvport)
1690                 return 0;
1691         if (port <= xprt_min_resvport || port > xprt_max_resvport)
1692                 return xprt_max_resvport;
1693         return --port;
1694 }
1695 static int xs_bind(struct sock_xprt *transport, struct socket *sock)
1696 {
1697         struct sockaddr_storage myaddr;
1698         int err, nloop = 0;
1699         unsigned short port = xs_get_srcport(transport);
1700         unsigned short last;
1701
1702         memcpy(&myaddr, &transport->srcaddr, transport->xprt.addrlen);
1703         do {
1704                 rpc_set_port((struct sockaddr *)&myaddr, port);
1705                 err = kernel_bind(sock, (struct sockaddr *)&myaddr,
1706                                 transport->xprt.addrlen);
1707                 if (port == 0)
1708                         break;
1709                 if (err == 0) {
1710                         transport->srcport = port;
1711                         break;
1712                 }
1713                 last = port;
1714                 port = xs_next_srcport(transport, port);
1715                 if (port > last)
1716                         nloop++;
1717         } while (err == -EADDRINUSE && nloop != 2);
1718
1719         if (myaddr.ss_family == AF_INET)
1720                 dprintk("RPC:       %s %pI4:%u: %s (%d)\n", __func__,
1721                                 &((struct sockaddr_in *)&myaddr)->sin_addr,
1722                                 port, err ? "failed" : "ok", err);
1723         else
1724                 dprintk("RPC:       %s %pI6:%u: %s (%d)\n", __func__,
1725                                 &((struct sockaddr_in6 *)&myaddr)->sin6_addr,
1726                                 port, err ? "failed" : "ok", err);
1727         return err;
1728 }
1729
1730 /*
1731  * We don't support autobind on AF_LOCAL sockets
1732  */
1733 static void xs_local_rpcbind(struct rpc_task *task)
1734 {
1735         rcu_read_lock();
1736         xprt_set_bound(rcu_dereference(task->tk_client->cl_xprt));
1737         rcu_read_unlock();
1738 }
1739
1740 static void xs_local_set_port(struct rpc_xprt *xprt, unsigned short port)
1741 {
1742 }
1743
1744 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1745 static struct lock_class_key xs_key[2];
1746 static struct lock_class_key xs_slock_key[2];
1747
1748 static inline void xs_reclassify_socketu(struct socket *sock)
1749 {
1750         struct sock *sk = sock->sk;
1751
1752         sock_lock_init_class_and_name(sk, "slock-AF_LOCAL-RPC",
1753                 &xs_slock_key[1], "sk_lock-AF_LOCAL-RPC", &xs_key[1]);
1754 }
1755
1756 static inline void xs_reclassify_socket4(struct socket *sock)
1757 {
1758         struct sock *sk = sock->sk;
1759
1760         sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1761                 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1762 }
1763
1764 static inline void xs_reclassify_socket6(struct socket *sock)
1765 {
1766         struct sock *sk = sock->sk;
1767
1768         sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1769                 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
1770 }
1771
1772 static inline void xs_reclassify_socket(int family, struct socket *sock)
1773 {
1774         WARN_ON_ONCE(sock_owned_by_user(sock->sk));
1775         if (sock_owned_by_user(sock->sk))
1776                 return;
1777
1778         switch (family) {
1779         case AF_LOCAL:
1780                 xs_reclassify_socketu(sock);
1781                 break;
1782         case AF_INET:
1783                 xs_reclassify_socket4(sock);
1784                 break;
1785         case AF_INET6:
1786                 xs_reclassify_socket6(sock);
1787                 break;
1788         }
1789 }
1790 #else
1791 static inline void xs_reclassify_socketu(struct socket *sock)
1792 {
1793 }
1794
1795 static inline void xs_reclassify_socket4(struct socket *sock)
1796 {
1797 }
1798
1799 static inline void xs_reclassify_socket6(struct socket *sock)
1800 {
1801 }
1802
1803 static inline void xs_reclassify_socket(int family, struct socket *sock)
1804 {
1805 }
1806 #endif
1807
1808 static struct socket *xs_create_sock(struct rpc_xprt *xprt,
1809                 struct sock_xprt *transport, int family, int type, int protocol)
1810 {
1811         struct socket *sock;
1812         int err;
1813
1814         err = __sock_create(xprt->xprt_net, family, type, protocol, &sock, 1);
1815         if (err < 0) {
1816                 dprintk("RPC:       can't create %d transport socket (%d).\n",
1817                                 protocol, -err);
1818                 goto out;
1819         }
1820         xs_reclassify_socket(family, sock);
1821
1822         err = xs_bind(transport, sock);
1823         if (err) {
1824                 sock_release(sock);
1825                 goto out;
1826         }
1827
1828         return sock;
1829 out:
1830         return ERR_PTR(err);
1831 }
1832
1833 static int xs_local_finish_connecting(struct rpc_xprt *xprt,
1834                                       struct socket *sock)
1835 {
1836         struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1837                                                                         xprt);
1838
1839         if (!transport->inet) {
1840                 struct sock *sk = sock->sk;
1841
1842                 write_lock_bh(&sk->sk_callback_lock);
1843
1844                 xs_save_old_callbacks(transport, sk);
1845
1846                 sk->sk_user_data = xprt;
1847                 sk->sk_data_ready = xs_local_data_ready;
1848                 sk->sk_write_space = xs_udp_write_space;
1849                 sk->sk_allocation = GFP_ATOMIC;
1850
1851                 xprt_clear_connected(xprt);
1852
1853                 /* Reset to new socket */
1854                 transport->sock = sock;
1855                 transport->inet = sk;
1856
1857                 write_unlock_bh(&sk->sk_callback_lock);
1858         }
1859
1860         /* Tell the socket layer to start connecting... */
1861         xprt->stat.connect_count++;
1862         xprt->stat.connect_start = jiffies;
1863         return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, 0);
1864 }
1865
1866 /**
1867  * xs_local_setup_socket - create AF_LOCAL socket, connect to a local endpoint
1868  * @xprt: RPC transport to connect
1869  * @transport: socket transport to connect
1870  * @create_sock: function to create a socket of the correct type
1871  */
1872 static int xs_local_setup_socket(struct sock_xprt *transport)
1873 {
1874         struct rpc_xprt *xprt = &transport->xprt;
1875         struct socket *sock;
1876         int status = -EIO;
1877
1878         current->flags |= PF_FSTRANS;
1879
1880         clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1881         status = __sock_create(xprt->xprt_net, AF_LOCAL,
1882                                         SOCK_STREAM, 0, &sock, 1);
1883         if (status < 0) {
1884                 dprintk("RPC:       can't create AF_LOCAL "
1885                         "transport socket (%d).\n", -status);
1886                 goto out;
1887         }
1888         xs_reclassify_socketu(sock);
1889
1890         dprintk("RPC:       worker connecting xprt %p via AF_LOCAL to %s\n",
1891                         xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1892
1893         status = xs_local_finish_connecting(xprt, sock);
1894         switch (status) {
1895         case 0:
1896                 dprintk("RPC:       xprt %p connected to %s\n",
1897                                 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1898                 xprt_set_connected(xprt);
1899                 break;
1900         case -ENOENT:
1901                 dprintk("RPC:       xprt %p: socket %s does not exist\n",
1902                                 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1903                 break;
1904         case -ECONNREFUSED:
1905                 dprintk("RPC:       xprt %p: connection refused for %s\n",
1906                                 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1907                 break;
1908         default:
1909                 printk(KERN_ERR "%s: unhandled error (%d) connecting to %s\n",
1910                                 __func__, -status,
1911                                 xprt->address_strings[RPC_DISPLAY_ADDR]);
1912         }
1913
1914 out:
1915         xprt_clear_connecting(xprt);
1916         xprt_wake_pending_tasks(xprt, status);
1917         current->flags &= ~PF_FSTRANS;
1918         return status;
1919 }
1920
1921 static void xs_local_connect(struct rpc_xprt *xprt, struct rpc_task *task)
1922 {
1923         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1924         int ret;
1925
1926          if (RPC_IS_ASYNC(task)) {
1927                 /*
1928                  * We want the AF_LOCAL connect to be resolved in the
1929                  * filesystem namespace of the process making the rpc
1930                  * call.  Thus we connect synchronously.
1931                  *
1932                  * If we want to support asynchronous AF_LOCAL calls,
1933                  * we'll need to figure out how to pass a namespace to
1934                  * connect.
1935                  */
1936                 rpc_exit(task, -ENOTCONN);
1937                 return;
1938         }
1939         ret = xs_local_setup_socket(transport);
1940         if (ret && !RPC_IS_SOFTCONN(task))
1941                 msleep_interruptible(15000);
1942 }
1943
1944 #ifdef CONFIG_SUNRPC_SWAP
1945 static void xs_set_memalloc(struct rpc_xprt *xprt)
1946 {
1947         struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1948                         xprt);
1949
1950         if (xprt->swapper)
1951                 sk_set_memalloc(transport->inet);
1952 }
1953
1954 /**
1955  * xs_swapper - Tag this transport as being used for swap.
1956  * @xprt: transport to tag
1957  * @enable: enable/disable
1958  *
1959  */
1960 int xs_swapper(struct rpc_xprt *xprt, int enable)
1961 {
1962         struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1963                         xprt);
1964         int err = 0;
1965
1966         if (enable) {
1967                 xprt->swapper++;
1968                 xs_set_memalloc(xprt);
1969         } else if (xprt->swapper) {
1970                 xprt->swapper--;
1971                 sk_clear_memalloc(transport->inet);
1972         }
1973
1974         return err;
1975 }
1976 EXPORT_SYMBOL_GPL(xs_swapper);
1977 #else
1978 static void xs_set_memalloc(struct rpc_xprt *xprt)
1979 {
1980 }
1981 #endif
1982
1983 static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
1984 {
1985         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1986
1987         if (!transport->inet) {
1988                 struct sock *sk = sock->sk;
1989
1990                 write_lock_bh(&sk->sk_callback_lock);
1991
1992                 xs_save_old_callbacks(transport, sk);
1993
1994                 sk->sk_user_data = xprt;
1995                 sk->sk_data_ready = xs_udp_data_ready;
1996                 sk->sk_write_space = xs_udp_write_space;
1997                 sk->sk_no_check = UDP_CSUM_NORCV;
1998                 sk->sk_allocation = GFP_ATOMIC;
1999
2000                 xprt_set_connected(xprt);
2001
2002                 /* Reset to new socket */
2003                 transport->sock = sock;
2004                 transport->inet = sk;
2005
2006                 xs_set_memalloc(xprt);
2007
2008                 write_unlock_bh(&sk->sk_callback_lock);
2009         }
2010         xs_udp_do_set_buffer_size(xprt);
2011 }
2012
2013 static void xs_udp_setup_socket(struct work_struct *work)
2014 {
2015         struct sock_xprt *transport =
2016                 container_of(work, struct sock_xprt, connect_worker.work);
2017         struct rpc_xprt *xprt = &transport->xprt;
2018         struct socket *sock = transport->sock;
2019         int status = -EIO;
2020
2021         current->flags |= PF_FSTRANS;
2022
2023         /* Start by resetting any existing state */
2024         xs_reset_transport(transport);
2025         sock = xs_create_sock(xprt, transport,
2026                         xs_addr(xprt)->sa_family, SOCK_DGRAM, IPPROTO_UDP);
2027         if (IS_ERR(sock))
2028                 goto out;
2029
2030         dprintk("RPC:       worker connecting xprt %p via %s to "
2031                                 "%s (port %s)\n", xprt,
2032                         xprt->address_strings[RPC_DISPLAY_PROTO],
2033                         xprt->address_strings[RPC_DISPLAY_ADDR],
2034                         xprt->address_strings[RPC_DISPLAY_PORT]);
2035
2036         xs_udp_finish_connecting(xprt, sock);
2037         status = 0;
2038 out:
2039         xprt_clear_connecting(xprt);
2040         xprt_wake_pending_tasks(xprt, status);
2041         current->flags &= ~PF_FSTRANS;
2042 }
2043
2044 /*
2045  * We need to preserve the port number so the reply cache on the server can
2046  * find our cached RPC replies when we get around to reconnecting.
2047  */
2048 static void xs_abort_connection(struct sock_xprt *transport)
2049 {
2050         int result;
2051         struct sockaddr any;
2052
2053         dprintk("RPC:       disconnecting xprt %p to reuse port\n", transport);
2054
2055         /*
2056          * Disconnect the transport socket by doing a connect operation
2057          * with AF_UNSPEC.  This should return immediately...
2058          */
2059         memset(&any, 0, sizeof(any));
2060         any.sa_family = AF_UNSPEC;
2061         result = kernel_connect(transport->sock, &any, sizeof(any), 0);
2062         if (!result)
2063                 xs_sock_reset_connection_flags(&transport->xprt);
2064         dprintk("RPC:       AF_UNSPEC connect return code %d\n", result);
2065 }
2066
2067 static void xs_tcp_reuse_connection(struct sock_xprt *transport)
2068 {
2069         unsigned int state = transport->inet->sk_state;
2070
2071         if (state == TCP_CLOSE && transport->sock->state == SS_UNCONNECTED) {
2072                 /* we don't need to abort the connection if the socket
2073                  * hasn't undergone a shutdown
2074                  */
2075                 if (transport->inet->sk_shutdown == 0)
2076                         return;
2077                 dprintk("RPC:       %s: TCP_CLOSEd and sk_shutdown set to %d\n",
2078                                 __func__, transport->inet->sk_shutdown);
2079         }
2080         if ((1 << state) & (TCPF_ESTABLISHED|TCPF_SYN_SENT)) {
2081                 /* we don't need to abort the connection if the socket
2082                  * hasn't undergone a shutdown
2083                  */
2084                 if (transport->inet->sk_shutdown == 0)
2085                         return;
2086                 dprintk("RPC:       %s: ESTABLISHED/SYN_SENT "
2087                                 "sk_shutdown set to %d\n",
2088                                 __func__, transport->inet->sk_shutdown);
2089         }
2090         xs_abort_connection(transport);
2091 }
2092
2093 static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2094 {
2095         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2096         int ret = -ENOTCONN;
2097
2098         if (!transport->inet) {
2099                 struct sock *sk = sock->sk;
2100
2101                 write_lock_bh(&sk->sk_callback_lock);
2102
2103                 xs_save_old_callbacks(transport, sk);
2104
2105                 sk->sk_user_data = xprt;
2106                 sk->sk_data_ready = xs_tcp_data_ready;
2107                 sk->sk_state_change = xs_tcp_state_change;
2108                 sk->sk_write_space = xs_tcp_write_space;
2109                 sk->sk_allocation = GFP_ATOMIC;
2110
2111                 /* socket options */
2112                 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
2113                 sock_reset_flag(sk, SOCK_LINGER);
2114                 tcp_sk(sk)->linger2 = 0;
2115                 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
2116
2117                 xprt_clear_connected(xprt);
2118
2119                 /* Reset to new socket */
2120                 transport->sock = sock;
2121                 transport->inet = sk;
2122
2123                 write_unlock_bh(&sk->sk_callback_lock);
2124         }
2125
2126         if (!xprt_bound(xprt))
2127                 goto out;
2128
2129         xs_set_memalloc(xprt);
2130
2131         /* Tell the socket layer to start connecting... */
2132         xprt->stat.connect_count++;
2133         xprt->stat.connect_start = jiffies;
2134         ret = kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
2135         switch (ret) {
2136         case 0:
2137         case -EINPROGRESS:
2138                 /* SYN_SENT! */
2139                 xprt->connect_cookie++;
2140                 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2141                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2142         }
2143 out:
2144         return ret;
2145 }
2146
2147 /**
2148  * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint
2149  * @xprt: RPC transport to connect
2150  * @transport: socket transport to connect
2151  * @create_sock: function to create a socket of the correct type
2152  *
2153  * Invoked by a work queue tasklet.
2154  */
2155 static void xs_tcp_setup_socket(struct work_struct *work)
2156 {
2157         struct sock_xprt *transport =
2158                 container_of(work, struct sock_xprt, connect_worker.work);
2159         struct socket *sock = transport->sock;
2160         struct rpc_xprt *xprt = &transport->xprt;
2161         int status = -EIO;
2162
2163         current->flags |= PF_FSTRANS;
2164
2165         if (!sock) {
2166                 clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
2167                 sock = xs_create_sock(xprt, transport,
2168                                 xs_addr(xprt)->sa_family, SOCK_STREAM, IPPROTO_TCP);
2169                 if (IS_ERR(sock)) {
2170                         status = PTR_ERR(sock);
2171                         goto out;
2172                 }
2173         } else {
2174                 int abort_and_exit;
2175
2176                 abort_and_exit = test_and_clear_bit(XPRT_CONNECTION_ABORT,
2177                                 &xprt->state);
2178                 /* "close" the socket, preserving the local port */
2179                 xs_tcp_reuse_connection(transport);
2180
2181                 if (abort_and_exit)
2182                         goto out_eagain;
2183         }
2184
2185         dprintk("RPC:       worker connecting xprt %p via %s to "
2186                                 "%s (port %s)\n", xprt,
2187                         xprt->address_strings[RPC_DISPLAY_PROTO],
2188                         xprt->address_strings[RPC_DISPLAY_ADDR],
2189                         xprt->address_strings[RPC_DISPLAY_PORT]);
2190
2191         status = xs_tcp_finish_connecting(xprt, sock);
2192         dprintk("RPC:       %p connect status %d connected %d sock state %d\n",
2193                         xprt, -status, xprt_connected(xprt),
2194                         sock->sk->sk_state);
2195         switch (status) {
2196         default:
2197                 printk("%s: connect returned unhandled error %d\n",
2198                         __func__, status);
2199         case -EADDRNOTAVAIL:
2200                 /* We're probably in TIME_WAIT. Get rid of existing socket,
2201                  * and retry
2202                  */
2203                 xs_tcp_force_close(xprt);
2204                 break;
2205         case -ECONNREFUSED:
2206         case -ECONNRESET:
2207         case -ENETUNREACH:
2208                 /* retry with existing socket, after a delay */
2209         case 0:
2210         case -EINPROGRESS:
2211         case -EALREADY:
2212                 xprt_clear_connecting(xprt);
2213                 current->flags &= ~PF_FSTRANS;
2214                 return;
2215         case -EINVAL:
2216                 /* Happens, for instance, if the user specified a link
2217                  * local IPv6 address without a scope-id.
2218                  */
2219                 goto out;
2220         }
2221 out_eagain:
2222         status = -EAGAIN;
2223 out:
2224         xprt_clear_connecting(xprt);
2225         xprt_wake_pending_tasks(xprt, status);
2226         current->flags &= ~PF_FSTRANS;
2227 }
2228
2229 /**
2230  * xs_connect - connect a socket to a remote endpoint
2231  * @xprt: pointer to transport structure
2232  * @task: address of RPC task that manages state of connect request
2233  *
2234  * TCP: If the remote end dropped the connection, delay reconnecting.
2235  *
2236  * UDP socket connects are synchronous, but we use a work queue anyway
2237  * to guarantee that even unprivileged user processes can set up a
2238  * socket on a privileged port.
2239  *
2240  * If a UDP socket connect fails, the delay behavior here prevents
2241  * retry floods (hard mounts).
2242  */
2243 static void xs_connect(struct rpc_xprt *xprt, struct rpc_task *task)
2244 {
2245         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2246
2247         if (transport->sock != NULL && !RPC_IS_SOFTCONN(task)) {
2248                 dprintk("RPC:       xs_connect delayed xprt %p for %lu "
2249                                 "seconds\n",
2250                                 xprt, xprt->reestablish_timeout / HZ);
2251                 queue_delayed_work(rpciod_workqueue,
2252                                    &transport->connect_worker,
2253                                    xprt->reestablish_timeout);
2254                 xprt->reestablish_timeout <<= 1;
2255                 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2256                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2257                 if (xprt->reestablish_timeout > XS_TCP_MAX_REEST_TO)
2258                         xprt->reestablish_timeout = XS_TCP_MAX_REEST_TO;
2259         } else {
2260                 dprintk("RPC:       xs_connect scheduled xprt %p\n", xprt);
2261                 queue_delayed_work(rpciod_workqueue,
2262                                    &transport->connect_worker, 0);
2263         }
2264 }
2265
2266 /**
2267  * xs_local_print_stats - display AF_LOCAL socket-specifc stats
2268  * @xprt: rpc_xprt struct containing statistics
2269  * @seq: output file
2270  *
2271  */
2272 static void xs_local_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2273 {
2274         long idle_time = 0;
2275
2276         if (xprt_connected(xprt))
2277                 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2278
2279         seq_printf(seq, "\txprt:\tlocal %lu %lu %lu %ld %lu %lu %lu "
2280                         "%llu %llu %lu %llu %llu\n",
2281                         xprt->stat.bind_count,
2282                         xprt->stat.connect_count,
2283                         xprt->stat.connect_time,
2284                         idle_time,
2285                         xprt->stat.sends,
2286                         xprt->stat.recvs,
2287                         xprt->stat.bad_xids,
2288                         xprt->stat.req_u,
2289                         xprt->stat.bklog_u,
2290                         xprt->stat.max_slots,
2291                         xprt->stat.sending_u,
2292                         xprt->stat.pending_u);
2293 }
2294
2295 /**
2296  * xs_udp_print_stats - display UDP socket-specifc stats
2297  * @xprt: rpc_xprt struct containing statistics
2298  * @seq: output file
2299  *
2300  */
2301 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2302 {
2303         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2304
2305         seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %llu %llu "
2306                         "%lu %llu %llu\n",
2307                         transport->srcport,
2308                         xprt->stat.bind_count,
2309                         xprt->stat.sends,
2310                         xprt->stat.recvs,
2311                         xprt->stat.bad_xids,
2312                         xprt->stat.req_u,
2313                         xprt->stat.bklog_u,
2314                         xprt->stat.max_slots,
2315                         xprt->stat.sending_u,
2316                         xprt->stat.pending_u);
2317 }
2318
2319 /**
2320  * xs_tcp_print_stats - display TCP socket-specifc stats
2321  * @xprt: rpc_xprt struct containing statistics
2322  * @seq: output file
2323  *
2324  */
2325 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2326 {
2327         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2328         long idle_time = 0;
2329
2330         if (xprt_connected(xprt))
2331                 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2332
2333         seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu "
2334                         "%llu %llu %lu %llu %llu\n",
2335                         transport->srcport,
2336                         xprt->stat.bind_count,
2337                         xprt->stat.connect_count,
2338                         xprt->stat.connect_time,
2339                         idle_time,
2340                         xprt->stat.sends,
2341                         xprt->stat.recvs,
2342                         xprt->stat.bad_xids,
2343                         xprt->stat.req_u,
2344                         xprt->stat.bklog_u,
2345                         xprt->stat.max_slots,
2346                         xprt->stat.sending_u,
2347                         xprt->stat.pending_u);
2348 }
2349
2350 /*
2351  * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason
2352  * we allocate pages instead doing a kmalloc like rpc_malloc is because we want
2353  * to use the server side send routines.
2354  */
2355 static void *bc_malloc(struct rpc_task *task, size_t size)
2356 {
2357         struct page *page;
2358         struct rpc_buffer *buf;
2359
2360         WARN_ON_ONCE(size > PAGE_SIZE - sizeof(struct rpc_buffer));
2361         if (size > PAGE_SIZE - sizeof(struct rpc_buffer))
2362                 return NULL;
2363
2364         page = alloc_page(GFP_KERNEL);
2365         if (!page)
2366                 return NULL;
2367
2368         buf = page_address(page);
2369         buf->len = PAGE_SIZE;
2370
2371         return buf->data;
2372 }
2373
2374 /*
2375  * Free the space allocated in the bc_alloc routine
2376  */
2377 static void bc_free(void *buffer)
2378 {
2379         struct rpc_buffer *buf;
2380
2381         if (!buffer)
2382                 return;
2383
2384         buf = container_of(buffer, struct rpc_buffer, data);
2385         free_page((unsigned long)buf);
2386 }
2387
2388 /*
2389  * Use the svc_sock to send the callback. Must be called with svsk->sk_mutex
2390  * held. Borrows heavily from svc_tcp_sendto and xs_tcp_send_request.
2391  */
2392 static int bc_sendto(struct rpc_rqst *req)
2393 {
2394         int len;
2395         struct xdr_buf *xbufp = &req->rq_snd_buf;
2396         struct rpc_xprt *xprt = req->rq_xprt;
2397         struct sock_xprt *transport =
2398                                 container_of(xprt, struct sock_xprt, xprt);
2399         struct socket *sock = transport->sock;
2400         unsigned long headoff;
2401         unsigned long tailoff;
2402
2403         xs_encode_stream_record_marker(xbufp);
2404
2405         tailoff = (unsigned long)xbufp->tail[0].iov_base & ~PAGE_MASK;
2406         headoff = (unsigned long)xbufp->head[0].iov_base & ~PAGE_MASK;
2407         len = svc_send_common(sock, xbufp,
2408                               virt_to_page(xbufp->head[0].iov_base), headoff,
2409                               xbufp->tail[0].iov_base, tailoff);
2410
2411         if (len != xbufp->len) {
2412                 printk(KERN_NOTICE "Error sending entire callback!\n");
2413                 len = -EAGAIN;
2414         }
2415
2416         return len;
2417 }
2418
2419 /*
2420  * The send routine. Borrows from svc_send
2421  */
2422 static int bc_send_request(struct rpc_task *task)
2423 {
2424         struct rpc_rqst *req = task->tk_rqstp;
2425         struct svc_xprt *xprt;
2426         u32                     len;
2427
2428         dprintk("sending request with xid: %08x\n", ntohl(req->rq_xid));
2429         /*
2430          * Get the server socket associated with this callback xprt
2431          */
2432         xprt = req->rq_xprt->bc_xprt;
2433
2434         /*
2435          * Grab the mutex to serialize data as the connection is shared
2436          * with the fore channel
2437          */
2438         if (!mutex_trylock(&xprt->xpt_mutex)) {
2439                 rpc_sleep_on(&xprt->xpt_bc_pending, task, NULL);
2440                 if (!mutex_trylock(&xprt->xpt_mutex))
2441                         return -EAGAIN;
2442                 rpc_wake_up_queued_task(&xprt->xpt_bc_pending, task);
2443         }
2444         if (test_bit(XPT_DEAD, &xprt->xpt_flags))
2445                 len = -ENOTCONN;
2446         else
2447                 len = bc_sendto(req);
2448         mutex_unlock(&xprt->xpt_mutex);
2449
2450         if (len > 0)
2451                 len = 0;
2452
2453         return len;
2454 }
2455
2456 /*
2457  * The close routine. Since this is client initiated, we do nothing
2458  */
2459
2460 static void bc_close(struct rpc_xprt *xprt)
2461 {
2462 }
2463
2464 /*
2465  * The xprt destroy routine. Again, because this connection is client
2466  * initiated, we do nothing
2467  */
2468
2469 static void bc_destroy(struct rpc_xprt *xprt)
2470 {
2471 }
2472
2473 static struct rpc_xprt_ops xs_local_ops = {
2474         .reserve_xprt           = xprt_reserve_xprt,
2475         .release_xprt           = xs_tcp_release_xprt,
2476         .alloc_slot             = xprt_alloc_slot,
2477         .rpcbind                = xs_local_rpcbind,
2478         .set_port               = xs_local_set_port,
2479         .connect                = xs_local_connect,
2480         .buf_alloc              = rpc_malloc,
2481         .buf_free               = rpc_free,
2482         .send_request           = xs_local_send_request,
2483         .set_retrans_timeout    = xprt_set_retrans_timeout_def,
2484         .close                  = xs_close,
2485         .destroy                = xs_destroy,
2486         .print_stats            = xs_local_print_stats,
2487 };
2488
2489 static struct rpc_xprt_ops xs_udp_ops = {
2490         .set_buffer_size        = xs_udp_set_buffer_size,
2491         .reserve_xprt           = xprt_reserve_xprt_cong,
2492         .release_xprt           = xprt_release_xprt_cong,
2493         .alloc_slot             = xprt_alloc_slot,
2494         .rpcbind                = rpcb_getport_async,
2495         .set_port               = xs_set_port,
2496         .connect                = xs_connect,
2497         .buf_alloc              = rpc_malloc,
2498         .buf_free               = rpc_free,
2499         .send_request           = xs_udp_send_request,
2500         .set_retrans_timeout    = xprt_set_retrans_timeout_rtt,
2501         .timer                  = xs_udp_timer,
2502         .release_request        = xprt_release_rqst_cong,
2503         .close                  = xs_close,
2504         .destroy                = xs_destroy,
2505         .print_stats            = xs_udp_print_stats,
2506 };
2507
2508 static struct rpc_xprt_ops xs_tcp_ops = {
2509         .reserve_xprt           = xprt_reserve_xprt,
2510         .release_xprt           = xs_tcp_release_xprt,
2511         .alloc_slot             = xprt_lock_and_alloc_slot,
2512         .rpcbind                = rpcb_getport_async,
2513         .set_port               = xs_set_port,
2514         .connect                = xs_connect,
2515         .buf_alloc              = rpc_malloc,
2516         .buf_free               = rpc_free,
2517         .send_request           = xs_tcp_send_request,
2518         .set_retrans_timeout    = xprt_set_retrans_timeout_def,
2519         .close                  = xs_tcp_close,
2520         .destroy                = xs_destroy,
2521         .print_stats            = xs_tcp_print_stats,
2522 };
2523
2524 /*
2525  * The rpc_xprt_ops for the server backchannel
2526  */
2527
2528 static struct rpc_xprt_ops bc_tcp_ops = {
2529         .reserve_xprt           = xprt_reserve_xprt,
2530         .release_xprt           = xprt_release_xprt,
2531         .alloc_slot             = xprt_alloc_slot,
2532         .rpcbind                = xs_local_rpcbind,
2533         .buf_alloc              = bc_malloc,
2534         .buf_free               = bc_free,
2535         .send_request           = bc_send_request,
2536         .set_retrans_timeout    = xprt_set_retrans_timeout_def,
2537         .close                  = bc_close,
2538         .destroy                = bc_destroy,
2539         .print_stats            = xs_tcp_print_stats,
2540 };
2541
2542 static int xs_init_anyaddr(const int family, struct sockaddr *sap)
2543 {
2544         static const struct sockaddr_in sin = {
2545                 .sin_family             = AF_INET,
2546                 .sin_addr.s_addr        = htonl(INADDR_ANY),
2547         };
2548         static const struct sockaddr_in6 sin6 = {
2549                 .sin6_family            = AF_INET6,
2550                 .sin6_addr              = IN6ADDR_ANY_INIT,
2551         };
2552
2553         switch (family) {
2554         case AF_LOCAL:
2555                 break;
2556         case AF_INET:
2557                 memcpy(sap, &sin, sizeof(sin));
2558                 break;
2559         case AF_INET6:
2560                 memcpy(sap, &sin6, sizeof(sin6));
2561                 break;
2562         default:
2563                 dprintk("RPC:       %s: Bad address family\n", __func__);
2564                 return -EAFNOSUPPORT;
2565         }
2566         return 0;
2567 }
2568
2569 static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
2570                                       unsigned int slot_table_size,
2571                                       unsigned int max_slot_table_size)
2572 {
2573         struct rpc_xprt *xprt;
2574         struct sock_xprt *new;
2575
2576         if (args->addrlen > sizeof(xprt->addr)) {
2577                 dprintk("RPC:       xs_setup_xprt: address too large\n");
2578                 return ERR_PTR(-EBADF);
2579         }
2580
2581         xprt = xprt_alloc(args->net, sizeof(*new), slot_table_size,
2582                         max_slot_table_size);
2583         if (xprt == NULL) {
2584                 dprintk("RPC:       xs_setup_xprt: couldn't allocate "
2585                                 "rpc_xprt\n");
2586                 return ERR_PTR(-ENOMEM);
2587         }
2588
2589         new = container_of(xprt, struct sock_xprt, xprt);
2590         memcpy(&xprt->addr, args->dstaddr, args->addrlen);
2591         xprt->addrlen = args->addrlen;
2592         if (args->srcaddr)
2593                 memcpy(&new->srcaddr, args->srcaddr, args->addrlen);
2594         else {
2595                 int err;
2596                 err = xs_init_anyaddr(args->dstaddr->sa_family,
2597                                         (struct sockaddr *)&new->srcaddr);
2598                 if (err != 0) {
2599                         xprt_free(xprt);
2600                         return ERR_PTR(err);
2601                 }
2602         }
2603
2604         return xprt;
2605 }
2606
2607 static const struct rpc_timeout xs_local_default_timeout = {
2608         .to_initval = 10 * HZ,
2609         .to_maxval = 10 * HZ,
2610         .to_retries = 2,
2611 };
2612
2613 /**
2614  * xs_setup_local - Set up transport to use an AF_LOCAL socket
2615  * @args: rpc transport creation arguments
2616  *
2617  * AF_LOCAL is a "tpi_cots_ord" transport, just like TCP
2618  */
2619 static struct rpc_xprt *xs_setup_local(struct xprt_create *args)
2620 {
2621         struct sockaddr_un *sun = (struct sockaddr_un *)args->dstaddr;
2622         struct sock_xprt *transport;
2623         struct rpc_xprt *xprt;
2624         struct rpc_xprt *ret;
2625
2626         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2627                         xprt_max_tcp_slot_table_entries);
2628         if (IS_ERR(xprt))
2629                 return xprt;
2630         transport = container_of(xprt, struct sock_xprt, xprt);
2631
2632         xprt->prot = 0;
2633         xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2634         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2635
2636         xprt->bind_timeout = XS_BIND_TO;
2637         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2638         xprt->idle_timeout = XS_IDLE_DISC_TO;
2639
2640         xprt->ops = &xs_local_ops;
2641         xprt->timeout = &xs_local_default_timeout;
2642
2643         switch (sun->sun_family) {
2644         case AF_LOCAL:
2645                 if (sun->sun_path[0] != '/') {
2646                         dprintk("RPC:       bad AF_LOCAL address: %s\n",
2647                                         sun->sun_path);
2648                         ret = ERR_PTR(-EINVAL);
2649                         goto out_err;
2650                 }
2651                 xprt_set_bound(xprt);
2652                 xs_format_peer_addresses(xprt, "local", RPCBIND_NETID_LOCAL);
2653                 break;
2654         default:
2655                 ret = ERR_PTR(-EAFNOSUPPORT);
2656                 goto out_err;
2657         }
2658
2659         dprintk("RPC:       set up xprt to %s via AF_LOCAL\n",
2660                         xprt->address_strings[RPC_DISPLAY_ADDR]);
2661
2662         if (try_module_get(THIS_MODULE))
2663                 return xprt;
2664         ret = ERR_PTR(-EINVAL);
2665 out_err:
2666         xprt_free(xprt);
2667         return ret;
2668 }
2669
2670 static const struct rpc_timeout xs_udp_default_timeout = {
2671         .to_initval = 5 * HZ,
2672         .to_maxval = 30 * HZ,
2673         .to_increment = 5 * HZ,
2674         .to_retries = 5,
2675 };
2676
2677 /**
2678  * xs_setup_udp - Set up transport to use a UDP socket
2679  * @args: rpc transport creation arguments
2680  *
2681  */
2682 static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
2683 {
2684         struct sockaddr *addr = args->dstaddr;
2685         struct rpc_xprt *xprt;
2686         struct sock_xprt *transport;
2687         struct rpc_xprt *ret;
2688
2689         xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries,
2690                         xprt_udp_slot_table_entries);
2691         if (IS_ERR(xprt))
2692                 return xprt;
2693         transport = container_of(xprt, struct sock_xprt, xprt);
2694
2695         xprt->prot = IPPROTO_UDP;
2696         xprt->tsh_size = 0;
2697         /* XXX: header size can vary due to auth type, IPv6, etc. */
2698         xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
2699
2700         xprt->bind_timeout = XS_BIND_TO;
2701         xprt->reestablish_timeout = XS_UDP_REEST_TO;
2702         xprt->idle_timeout = XS_IDLE_DISC_TO;
2703
2704         xprt->ops = &xs_udp_ops;
2705
2706         xprt->timeout = &xs_udp_default_timeout;
2707
2708         switch (addr->sa_family) {
2709         case AF_INET:
2710                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2711                         xprt_set_bound(xprt);
2712
2713                 INIT_DELAYED_WORK(&transport->connect_worker,
2714                                         xs_udp_setup_socket);
2715                 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
2716                 break;
2717         case AF_INET6:
2718                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2719                         xprt_set_bound(xprt);
2720
2721                 INIT_DELAYED_WORK(&transport->connect_worker,
2722                                         xs_udp_setup_socket);
2723                 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
2724                 break;
2725         default:
2726                 ret = ERR_PTR(-EAFNOSUPPORT);
2727                 goto out_err;
2728         }
2729
2730         if (xprt_bound(xprt))
2731                 dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
2732                                 xprt->address_strings[RPC_DISPLAY_ADDR],
2733                                 xprt->address_strings[RPC_DISPLAY_PORT],
2734                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
2735         else
2736                 dprintk("RPC:       set up xprt to %s (autobind) via %s\n",
2737                                 xprt->address_strings[RPC_DISPLAY_ADDR],
2738                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
2739
2740         if (try_module_get(THIS_MODULE))
2741                 return xprt;
2742         ret = ERR_PTR(-EINVAL);
2743 out_err:
2744         xprt_free(xprt);
2745         return ret;
2746 }
2747
2748 static const struct rpc_timeout xs_tcp_default_timeout = {
2749         .to_initval = 60 * HZ,
2750         .to_maxval = 60 * HZ,
2751         .to_retries = 2,
2752 };
2753
2754 /**
2755  * xs_setup_tcp - Set up transport to use a TCP socket
2756  * @args: rpc transport creation arguments
2757  *
2758  */
2759 static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
2760 {
2761         struct sockaddr *addr = args->dstaddr;
2762         struct rpc_xprt *xprt;
2763         struct sock_xprt *transport;
2764         struct rpc_xprt *ret;
2765
2766         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2767                         xprt_max_tcp_slot_table_entries);
2768         if (IS_ERR(xprt))
2769                 return xprt;
2770         transport = container_of(xprt, struct sock_xprt, xprt);
2771
2772         xprt->prot = IPPROTO_TCP;
2773         xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2774         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2775
2776         xprt->bind_timeout = XS_BIND_TO;
2777         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2778         xprt->idle_timeout = XS_IDLE_DISC_TO;
2779
2780         xprt->ops = &xs_tcp_ops;
2781         xprt->timeout = &xs_tcp_default_timeout;
2782
2783         switch (addr->sa_family) {
2784         case AF_INET:
2785                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2786                         xprt_set_bound(xprt);
2787
2788                 INIT_DELAYED_WORK(&transport->connect_worker,
2789                                         xs_tcp_setup_socket);
2790                 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
2791                 break;
2792         case AF_INET6:
2793                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2794                         xprt_set_bound(xprt);
2795
2796                 INIT_DELAYED_WORK(&transport->connect_worker,
2797                                         xs_tcp_setup_socket);
2798                 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
2799                 break;
2800         default:
2801                 ret = ERR_PTR(-EAFNOSUPPORT);
2802                 goto out_err;
2803         }
2804
2805         if (xprt_bound(xprt))
2806                 dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
2807                                 xprt->address_strings[RPC_DISPLAY_ADDR],
2808                                 xprt->address_strings[RPC_DISPLAY_PORT],
2809                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
2810         else
2811                 dprintk("RPC:       set up xprt to %s (autobind) via %s\n",
2812                                 xprt->address_strings[RPC_DISPLAY_ADDR],
2813                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
2814
2815
2816         if (try_module_get(THIS_MODULE))
2817                 return xprt;
2818         ret = ERR_PTR(-EINVAL);
2819 out_err:
2820         xprt_free(xprt);
2821         return ret;
2822 }
2823
2824 /**
2825  * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket
2826  * @args: rpc transport creation arguments
2827  *
2828  */
2829 static struct rpc_xprt *xs_setup_bc_tcp(struct xprt_create *args)
2830 {
2831         struct sockaddr *addr = args->dstaddr;
2832         struct rpc_xprt *xprt;
2833         struct sock_xprt *transport;
2834         struct svc_sock *bc_sock;
2835         struct rpc_xprt *ret;
2836
2837         if (args->bc_xprt->xpt_bc_xprt) {
2838                 /*
2839                  * This server connection already has a backchannel
2840                  * export; we can't create a new one, as we wouldn't be
2841                  * able to match replies based on xid any more.  So,
2842                  * reuse the already-existing one:
2843                  */
2844                  return args->bc_xprt->xpt_bc_xprt;
2845         }
2846         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2847                         xprt_tcp_slot_table_entries);
2848         if (IS_ERR(xprt))
2849                 return xprt;
2850         transport = container_of(xprt, struct sock_xprt, xprt);
2851
2852         xprt->prot = IPPROTO_TCP;
2853         xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2854         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2855         xprt->timeout = &xs_tcp_default_timeout;
2856
2857         /* backchannel */
2858         xprt_set_bound(xprt);
2859         xprt->bind_timeout = 0;
2860         xprt->reestablish_timeout = 0;
2861         xprt->idle_timeout = 0;
2862
2863         xprt->ops = &bc_tcp_ops;
2864
2865         switch (addr->sa_family) {
2866         case AF_INET:
2867                 xs_format_peer_addresses(xprt, "tcp",
2868                                          RPCBIND_NETID_TCP);
2869                 break;
2870         case AF_INET6:
2871                 xs_format_peer_addresses(xprt, "tcp",
2872                                    RPCBIND_NETID_TCP6);
2873                 break;
2874         default:
2875                 ret = ERR_PTR(-EAFNOSUPPORT);
2876                 goto out_err;
2877         }
2878
2879         dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
2880                         xprt->address_strings[RPC_DISPLAY_ADDR],
2881                         xprt->address_strings[RPC_DISPLAY_PORT],
2882                         xprt->address_strings[RPC_DISPLAY_PROTO]);
2883
2884         /*
2885          * Once we've associated a backchannel xprt with a connection,
2886          * we want to keep it around as long as long as the connection
2887          * lasts, in case we need to start using it for a backchannel
2888          * again; this reference won't be dropped until bc_xprt is
2889          * destroyed.
2890          */
2891         xprt_get(xprt);
2892         args->bc_xprt->xpt_bc_xprt = xprt;
2893         xprt->bc_xprt = args->bc_xprt;
2894         bc_sock = container_of(args->bc_xprt, struct svc_sock, sk_xprt);
2895         transport->sock = bc_sock->sk_sock;
2896         transport->inet = bc_sock->sk_sk;
2897
2898         /*
2899          * Since we don't want connections for the backchannel, we set
2900          * the xprt status to connected
2901          */
2902         xprt_set_connected(xprt);
2903
2904
2905         if (try_module_get(THIS_MODULE))
2906                 return xprt;
2907         xprt_put(xprt);
2908         ret = ERR_PTR(-EINVAL);
2909 out_err:
2910         xprt_free(xprt);
2911         return ret;
2912 }
2913
2914 static struct xprt_class        xs_local_transport = {
2915         .list           = LIST_HEAD_INIT(xs_local_transport.list),
2916         .name           = "named UNIX socket",
2917         .owner          = THIS_MODULE,
2918         .ident          = XPRT_TRANSPORT_LOCAL,
2919         .setup          = xs_setup_local,
2920 };
2921
2922 static struct xprt_class        xs_udp_transport = {
2923         .list           = LIST_HEAD_INIT(xs_udp_transport.list),
2924         .name           = "udp",
2925         .owner          = THIS_MODULE,
2926         .ident          = XPRT_TRANSPORT_UDP,
2927         .setup          = xs_setup_udp,
2928 };
2929
2930 static struct xprt_class        xs_tcp_transport = {
2931         .list           = LIST_HEAD_INIT(xs_tcp_transport.list),
2932         .name           = "tcp",
2933         .owner          = THIS_MODULE,
2934         .ident          = XPRT_TRANSPORT_TCP,
2935         .setup          = xs_setup_tcp,
2936 };
2937
2938 static struct xprt_class        xs_bc_tcp_transport = {
2939         .list           = LIST_HEAD_INIT(xs_bc_tcp_transport.list),
2940         .name           = "tcp NFSv4.1 backchannel",
2941         .owner          = THIS_MODULE,
2942         .ident          = XPRT_TRANSPORT_BC_TCP,
2943         .setup          = xs_setup_bc_tcp,
2944 };
2945
2946 /**
2947  * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
2948  *
2949  */
2950 int init_socket_xprt(void)
2951 {
2952 #ifdef RPC_DEBUG
2953         if (!sunrpc_table_header)
2954                 sunrpc_table_header = register_sysctl_table(sunrpc_table);
2955 #endif
2956
2957         xprt_register_transport(&xs_local_transport);
2958         xprt_register_transport(&xs_udp_transport);
2959         xprt_register_transport(&xs_tcp_transport);
2960         xprt_register_transport(&xs_bc_tcp_transport);
2961
2962         return 0;
2963 }
2964
2965 /**
2966  * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
2967  *
2968  */
2969 void cleanup_socket_xprt(void)
2970 {
2971 #ifdef RPC_DEBUG
2972         if (sunrpc_table_header) {
2973                 unregister_sysctl_table(sunrpc_table_header);
2974                 sunrpc_table_header = NULL;
2975         }
2976 #endif
2977
2978         xprt_unregister_transport(&xs_local_transport);
2979         xprt_unregister_transport(&xs_udp_transport);
2980         xprt_unregister_transport(&xs_tcp_transport);
2981         xprt_unregister_transport(&xs_bc_tcp_transport);
2982 }
2983
2984 static int param_set_uint_minmax(const char *val,
2985                 const struct kernel_param *kp,
2986                 unsigned int min, unsigned int max)
2987 {
2988         unsigned long num;
2989         int ret;
2990
2991         if (!val)
2992                 return -EINVAL;
2993         ret = strict_strtoul(val, 0, &num);
2994         if (ret == -EINVAL || num < min || num > max)
2995                 return -EINVAL;
2996         *((unsigned int *)kp->arg) = num;
2997         return 0;
2998 }
2999
3000 static int param_set_portnr(const char *val, const struct kernel_param *kp)
3001 {
3002         return param_set_uint_minmax(val, kp,
3003                         RPC_MIN_RESVPORT,
3004                         RPC_MAX_RESVPORT);
3005 }
3006
3007 static struct kernel_param_ops param_ops_portnr = {
3008         .set = param_set_portnr,
3009         .get = param_get_uint,
3010 };
3011
3012 #define param_check_portnr(name, p) \
3013         __param_check(name, p, unsigned int);
3014
3015 module_param_named(min_resvport, xprt_min_resvport, portnr, 0644);
3016 module_param_named(max_resvport, xprt_max_resvport, portnr, 0644);
3017
3018 static int param_set_slot_table_size(const char *val,
3019                                      const struct kernel_param *kp)
3020 {
3021         return param_set_uint_minmax(val, kp,
3022                         RPC_MIN_SLOT_TABLE,
3023                         RPC_MAX_SLOT_TABLE);
3024 }
3025
3026 static struct kernel_param_ops param_ops_slot_table_size = {
3027         .set = param_set_slot_table_size,
3028         .get = param_get_uint,
3029 };
3030
3031 #define param_check_slot_table_size(name, p) \
3032         __param_check(name, p, unsigned int);
3033
3034 static int param_set_max_slot_table_size(const char *val,
3035                                      const struct kernel_param *kp)
3036 {
3037         return param_set_uint_minmax(val, kp,
3038                         RPC_MIN_SLOT_TABLE,
3039                         RPC_MAX_SLOT_TABLE_LIMIT);
3040 }
3041
3042 static struct kernel_param_ops param_ops_max_slot_table_size = {
3043         .set = param_set_max_slot_table_size,
3044         .get = param_get_uint,
3045 };
3046
3047 #define param_check_max_slot_table_size(name, p) \
3048         __param_check(name, p, unsigned int);
3049
3050 module_param_named(tcp_slot_table_entries, xprt_tcp_slot_table_entries,
3051                    slot_table_size, 0644);
3052 module_param_named(tcp_max_slot_table_entries, xprt_max_tcp_slot_table_entries,
3053                    max_slot_table_size, 0644);
3054 module_param_named(udp_slot_table_entries, xprt_udp_slot_table_entries,
3055                    slot_table_size, 0644);
3056