linux/net/sunrpc/xprtsock.c
<<
>>
Prefs
   1/*
   2 * linux/net/sunrpc/xprtsock.c
   3 *
   4 * Client-side transport implementation for sockets.
   5 *
   6 * TCP callback races fixes (C) 1998 Red Hat
   7 * TCP send fixes (C) 1998 Red Hat
   8 * TCP NFS related read + write fixes
   9 *  (C) 1999 Dave Airlie, University of Limerick, Ireland <airlied@linux.ie>
  10 *
  11 * Rewrite of larges part of the code in order to stabilize TCP stuff.
  12 * Fix behaviour when socket buffer is full.
  13 *  (C) 1999 Trond Myklebust <trond.myklebust@fys.uio.no>
  14 *
  15 * IP socket transport implementation, (C) 2005 Chuck Lever <cel@netapp.com>
  16 *
  17 * IPv6 support contributed by Gilles Quillard, Bull Open Source, 2005.
  18 *   <gilles.quillard@bull.net>
  19 */
  20
  21#include <linux/types.h>
  22#include <linux/string.h>
  23#include <linux/slab.h>
  24#include <linux/module.h>
  25#include <linux/capability.h>
  26#include <linux/pagemap.h>
  27#include <linux/errno.h>
  28#include <linux/socket.h>
  29#include <linux/in.h>
  30#include <linux/net.h>
  31#include <linux/mm.h>
  32#include <linux/un.h>
  33#include <linux/udp.h>
  34#include <linux/tcp.h>
  35#include <linux/sunrpc/clnt.h>
  36#include <linux/sunrpc/sched.h>
  37#include <linux/sunrpc/svcsock.h>
  38#include <linux/sunrpc/xprtsock.h>
  39#include <linux/file.h>
  40#ifdef CONFIG_SUNRPC_BACKCHANNEL
  41#include <linux/sunrpc/bc_xprt.h>
  42#endif
  43
  44#include <net/sock.h>
  45#include <net/checksum.h>
  46#include <net/udp.h>
  47#include <net/tcp.h>
  48
  49#include "sunrpc.h"
  50
  51static void xs_close(struct rpc_xprt *xprt);
  52
  53/*
  54 * xprtsock tunables
  55 */
  56static unsigned int xprt_udp_slot_table_entries = RPC_DEF_SLOT_TABLE;
  57static unsigned int xprt_tcp_slot_table_entries = RPC_MIN_SLOT_TABLE;
  58static unsigned int xprt_max_tcp_slot_table_entries = RPC_MAX_SLOT_TABLE;
  59
  60static unsigned int xprt_min_resvport = RPC_DEF_MIN_RESVPORT;
  61static unsigned int xprt_max_resvport = RPC_DEF_MAX_RESVPORT;
  62
  63#define XS_TCP_LINGER_TO        (15U * HZ)
  64static unsigned int xs_tcp_fin_timeout __read_mostly = XS_TCP_LINGER_TO;
  65
  66/*
  67 * We can register our own files under /proc/sys/sunrpc by
  68 * calling register_sysctl_table() again.  The files in that
  69 * directory become the union of all files registered there.
  70 *
  71 * We simply need to make sure that we don't collide with
  72 * someone else's file names!
  73 */
  74
  75#ifdef RPC_DEBUG
  76
  77static unsigned int min_slot_table_size = RPC_MIN_SLOT_TABLE;
  78static unsigned int max_slot_table_size = RPC_MAX_SLOT_TABLE;
  79static unsigned int max_tcp_slot_table_limit = RPC_MAX_SLOT_TABLE_LIMIT;
  80static unsigned int xprt_min_resvport_limit = RPC_MIN_RESVPORT;
  81static unsigned int xprt_max_resvport_limit = RPC_MAX_RESVPORT;
  82
  83static struct ctl_table_header *sunrpc_table_header;
  84
  85/*
  86 * FIXME: changing the UDP slot table size should also resize the UDP
  87 *        socket buffers for existing UDP transports
  88 */
  89static ctl_table xs_tunables_table[] = {
  90        {
  91                .procname       = "udp_slot_table_entries",
  92                .data           = &xprt_udp_slot_table_entries,
  93                .maxlen         = sizeof(unsigned int),
  94                .mode           = 0644,
  95                .proc_handler   = proc_dointvec_minmax,
  96                .extra1         = &min_slot_table_size,
  97                .extra2         = &max_slot_table_size
  98        },
  99        {
 100                .procname       = "tcp_slot_table_entries",
 101                .data           = &xprt_tcp_slot_table_entries,
 102                .maxlen         = sizeof(unsigned int),
 103                .mode           = 0644,
 104                .proc_handler   = proc_dointvec_minmax,
 105                .extra1         = &min_slot_table_size,
 106                .extra2         = &max_slot_table_size
 107        },
 108        {
 109                .procname       = "tcp_max_slot_table_entries",
 110                .data           = &xprt_max_tcp_slot_table_entries,
 111                .maxlen         = sizeof(unsigned int),
 112                .mode           = 0644,
 113                .proc_handler   = proc_dointvec_minmax,
 114                .extra1         = &min_slot_table_size,
 115                .extra2         = &max_tcp_slot_table_limit
 116        },
 117        {
 118                .procname       = "min_resvport",
 119                .data           = &xprt_min_resvport,
 120                .maxlen         = sizeof(unsigned int),
 121                .mode           = 0644,
 122                .proc_handler   = proc_dointvec_minmax,
 123                .extra1         = &xprt_min_resvport_limit,
 124                .extra2         = &xprt_max_resvport_limit
 125        },
 126        {
 127                .procname       = "max_resvport",
 128                .data           = &xprt_max_resvport,
 129                .maxlen         = sizeof(unsigned int),
 130                .mode           = 0644,
 131                .proc_handler   = proc_dointvec_minmax,
 132                .extra1         = &xprt_min_resvport_limit,
 133                .extra2         = &xprt_max_resvport_limit
 134        },
 135        {
 136                .procname       = "tcp_fin_timeout",
 137                .data           = &xs_tcp_fin_timeout,
 138                .maxlen         = sizeof(xs_tcp_fin_timeout),
 139                .mode           = 0644,
 140                .proc_handler   = proc_dointvec_jiffies,
 141        },
 142        { },
 143};
 144
 145static ctl_table sunrpc_table[] = {
 146        {
 147                .procname       = "sunrpc",
 148                .mode           = 0555,
 149                .child          = xs_tunables_table
 150        },
 151        { },
 152};
 153
 154#endif
 155
 156/*
 157 * Wait duration for a reply from the RPC portmapper.
 158 */
 159#define XS_BIND_TO              (60U * HZ)
 160
 161/*
 162 * Delay if a UDP socket connect error occurs.  This is most likely some
 163 * kind of resource problem on the local host.
 164 */
 165#define XS_UDP_REEST_TO         (2U * HZ)
 166
 167/*
 168 * The reestablish timeout allows clients to delay for a bit before attempting
 169 * to reconnect to a server that just dropped our connection.
 170 *
 171 * We implement an exponential backoff when trying to reestablish a TCP
 172 * transport connection with the server.  Some servers like to drop a TCP
 173 * connection when they are overworked, so we start with a short timeout and
 174 * increase over time if the server is down or not responding.
 175 */
 176#define XS_TCP_INIT_REEST_TO    (3U * HZ)
 177#define XS_TCP_MAX_REEST_TO     (5U * 60 * HZ)
 178
 179/*
 180 * TCP idle timeout; client drops the transport socket if it is idle
 181 * for this long.  Note that we also timeout UDP sockets to prevent
 182 * holding port numbers when there is no RPC traffic.
 183 */
 184#define XS_IDLE_DISC_TO         (5U * 60 * HZ)
 185
 186#ifdef RPC_DEBUG
 187# undef  RPC_DEBUG_DATA
 188# define RPCDBG_FACILITY        RPCDBG_TRANS
 189#endif
 190
 191#ifdef RPC_DEBUG_DATA
 192static void xs_pktdump(char *msg, u32 *packet, unsigned int count)
 193{
 194        u8 *buf = (u8 *) packet;
 195        int j;
 196
 197        dprintk("RPC:       %s\n", msg);
 198        for (j = 0; j < count && j < 128; j += 4) {
 199                if (!(j & 31)) {
 200                        if (j)
 201                                dprintk("\n");
 202                        dprintk("0x%04x ", j);
 203                }
 204                dprintk("%02x%02x%02x%02x ",
 205                        buf[j], buf[j+1], buf[j+2], buf[j+3]);
 206        }
 207        dprintk("\n");
 208}
 209#else
 210static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count)
 211{
 212        /* NOP */
 213}
 214#endif
 215
 216struct sock_xprt {
 217        struct rpc_xprt         xprt;
 218
 219        /*
 220         * Network layer
 221         */
 222        struct socket *         sock;
 223        struct sock *           inet;
 224
 225        /*
 226         * State of TCP reply receive
 227         */
 228        __be32                  tcp_fraghdr,
 229                                tcp_xid,
 230                                tcp_calldir;
 231
 232        u32                     tcp_offset,
 233                                tcp_reclen;
 234
 235        unsigned long           tcp_copied,
 236                                tcp_flags;
 237
 238        /*
 239         * Connection of transports
 240         */
 241        struct delayed_work     connect_worker;
 242        struct sockaddr_storage srcaddr;
 243        unsigned short          srcport;
 244
 245        /*
 246         * UDP socket buffer size parameters
 247         */
 248        size_t                  rcvsize,
 249                                sndsize;
 250
 251        /*
 252         * Saved socket callback addresses
 253         */
 254        void                    (*old_data_ready)(struct sock *, int);
 255        void                    (*old_state_change)(struct sock *);
 256        void                    (*old_write_space)(struct sock *);
 257};
 258
 259/*
 260 * TCP receive state flags
 261 */
 262#define TCP_RCV_LAST_FRAG       (1UL << 0)
 263#define TCP_RCV_COPY_FRAGHDR    (1UL << 1)
 264#define TCP_RCV_COPY_XID        (1UL << 2)
 265#define TCP_RCV_COPY_DATA       (1UL << 3)
 266#define TCP_RCV_READ_CALLDIR    (1UL << 4)
 267#define TCP_RCV_COPY_CALLDIR    (1UL << 5)
 268
 269/*
 270 * TCP RPC flags
 271 */
 272#define TCP_RPC_REPLY           (1UL << 6)
 273
 274static inline struct sockaddr *xs_addr(struct rpc_xprt *xprt)
 275{
 276        return (struct sockaddr *) &xprt->addr;
 277}
 278
 279static inline struct sockaddr_un *xs_addr_un(struct rpc_xprt *xprt)
 280{
 281        return (struct sockaddr_un *) &xprt->addr;
 282}
 283
 284static inline struct sockaddr_in *xs_addr_in(struct rpc_xprt *xprt)
 285{
 286        return (struct sockaddr_in *) &xprt->addr;
 287}
 288
 289static inline struct sockaddr_in6 *xs_addr_in6(struct rpc_xprt *xprt)
 290{
 291        return (struct sockaddr_in6 *) &xprt->addr;
 292}
 293
 294static void xs_format_common_peer_addresses(struct rpc_xprt *xprt)
 295{
 296        struct sockaddr *sap = xs_addr(xprt);
 297        struct sockaddr_in6 *sin6;
 298        struct sockaddr_in *sin;
 299        struct sockaddr_un *sun;
 300        char buf[128];
 301
 302        switch (sap->sa_family) {
 303        case AF_LOCAL:
 304                sun = xs_addr_un(xprt);
 305                strlcpy(buf, sun->sun_path, sizeof(buf));
 306                xprt->address_strings[RPC_DISPLAY_ADDR] =
 307                                                kstrdup(buf, GFP_KERNEL);
 308                break;
 309        case AF_INET:
 310                (void)rpc_ntop(sap, buf, sizeof(buf));
 311                xprt->address_strings[RPC_DISPLAY_ADDR] =
 312                                                kstrdup(buf, GFP_KERNEL);
 313                sin = xs_addr_in(xprt);
 314                snprintf(buf, sizeof(buf), "%08x", ntohl(sin->sin_addr.s_addr));
 315                break;
 316        case AF_INET6:
 317                (void)rpc_ntop(sap, buf, sizeof(buf));
 318                xprt->address_strings[RPC_DISPLAY_ADDR] =
 319                                                kstrdup(buf, GFP_KERNEL);
 320                sin6 = xs_addr_in6(xprt);
 321                snprintf(buf, sizeof(buf), "%pi6", &sin6->sin6_addr);
 322                break;
 323        default:
 324                BUG();
 325        }
 326
 327        xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = kstrdup(buf, GFP_KERNEL);
 328}
 329
 330static void xs_format_common_peer_ports(struct rpc_xprt *xprt)
 331{
 332        struct sockaddr *sap = xs_addr(xprt);
 333        char buf[128];
 334
 335        snprintf(buf, sizeof(buf), "%u", rpc_get_port(sap));
 336        xprt->address_strings[RPC_DISPLAY_PORT] = kstrdup(buf, GFP_KERNEL);
 337
 338        snprintf(buf, sizeof(buf), "%4hx", rpc_get_port(sap));
 339        xprt->address_strings[RPC_DISPLAY_HEX_PORT] = kstrdup(buf, GFP_KERNEL);
 340}
 341
 342static void xs_format_peer_addresses(struct rpc_xprt *xprt,
 343                                     const char *protocol,
 344                                     const char *netid)
 345{
 346        xprt->address_strings[RPC_DISPLAY_PROTO] = protocol;
 347        xprt->address_strings[RPC_DISPLAY_NETID] = netid;
 348        xs_format_common_peer_addresses(xprt);
 349        xs_format_common_peer_ports(xprt);
 350}
 351
 352static void xs_update_peer_port(struct rpc_xprt *xprt)
 353{
 354        kfree(xprt->address_strings[RPC_DISPLAY_HEX_PORT]);
 355        kfree(xprt->address_strings[RPC_DISPLAY_PORT]);
 356
 357        xs_format_common_peer_ports(xprt);
 358}
 359
 360static void xs_free_peer_addresses(struct rpc_xprt *xprt)
 361{
 362        unsigned int i;
 363
 364        for (i = 0; i < RPC_DISPLAY_MAX; i++)
 365                switch (i) {
 366                case RPC_DISPLAY_PROTO:
 367                case RPC_DISPLAY_NETID:
 368                        continue;
 369                default:
 370                        kfree(xprt->address_strings[i]);
 371                }
 372}
 373
 374#define XS_SENDMSG_FLAGS        (MSG_DONTWAIT | MSG_NOSIGNAL)
 375
 376static int xs_send_kvec(struct socket *sock, struct sockaddr *addr, int addrlen, struct kvec *vec, unsigned int base, int more)
 377{
 378        struct msghdr msg = {
 379                .msg_name       = addr,
 380                .msg_namelen    = addrlen,
 381                .msg_flags      = XS_SENDMSG_FLAGS | (more ? MSG_MORE : 0),
 382        };
 383        struct kvec iov = {
 384                .iov_base       = vec->iov_base + base,
 385                .iov_len        = vec->iov_len - base,
 386        };
 387
 388        if (iov.iov_len != 0)
 389                return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
 390        return kernel_sendmsg(sock, &msg, NULL, 0, 0);
 391}
 392
 393static int xs_send_pagedata(struct socket *sock, struct xdr_buf *xdr, unsigned int base, int more)
 394{
 395        struct page **ppage;
 396        unsigned int remainder;
 397        int err, sent = 0;
 398
 399        remainder = xdr->page_len - base;
 400        base += xdr->page_base;
 401        ppage = xdr->pages + (base >> PAGE_SHIFT);
 402        base &= ~PAGE_MASK;
 403        for(;;) {
 404                unsigned int len = min_t(unsigned int, PAGE_SIZE - base, remainder);
 405                int flags = XS_SENDMSG_FLAGS;
 406
 407                remainder -= len;
 408                if (remainder != 0 || more)
 409                        flags |= MSG_MORE;
 410                err = sock->ops->sendpage(sock, *ppage, base, len, flags);
 411                if (remainder == 0 || err != len)
 412                        break;
 413                sent += err;
 414                ppage++;
 415                base = 0;
 416        }
 417        if (sent == 0)
 418                return err;
 419        if (err > 0)
 420                sent += err;
 421        return sent;
 422}
 423
 424/**
 425 * xs_sendpages - write pages directly to a socket
 426 * @sock: socket to send on
 427 * @addr: UDP only -- address of destination
 428 * @addrlen: UDP only -- length of destination address
 429 * @xdr: buffer containing this request
 430 * @base: starting position in the buffer
 431 *
 432 */
 433static int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base)
 434{
 435        unsigned int remainder = xdr->len - base;
 436        int err, sent = 0;
 437
 438        if (unlikely(!sock))
 439                return -ENOTSOCK;
 440
 441        clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags);
 442        if (base != 0) {
 443                addr = NULL;
 444                addrlen = 0;
 445        }
 446
 447        if (base < xdr->head[0].iov_len || addr != NULL) {
 448                unsigned int len = xdr->head[0].iov_len - base;
 449                remainder -= len;
 450                err = xs_send_kvec(sock, addr, addrlen, &xdr->head[0], base, remainder != 0);
 451                if (remainder == 0 || err != len)
 452                        goto out;
 453                sent += err;
 454                base = 0;
 455        } else
 456                base -= xdr->head[0].iov_len;
 457
 458        if (base < xdr->page_len) {
 459                unsigned int len = xdr->page_len - base;
 460                remainder -= len;
 461                err = xs_send_pagedata(sock, xdr, base, remainder != 0);
 462                if (remainder == 0 || err != len)
 463                        goto out;
 464                sent += err;
 465                base = 0;
 466        } else
 467                base -= xdr->page_len;
 468
 469        if (base >= xdr->tail[0].iov_len)
 470                return sent;
 471        err = xs_send_kvec(sock, NULL, 0, &xdr->tail[0], base, 0);
 472out:
 473        if (sent == 0)
 474                return err;
 475        if (err > 0)
 476                sent += err;
 477        return sent;
 478}
 479
 480static void xs_nospace_callback(struct rpc_task *task)
 481{
 482        struct sock_xprt *transport = container_of(task->tk_rqstp->rq_xprt, struct sock_xprt, xprt);
 483
 484        transport->inet->sk_write_pending--;
 485        clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 486}
 487
 488/**
 489 * xs_nospace - place task on wait queue if transmit was incomplete
 490 * @task: task to put to sleep
 491 *
 492 */
 493static int xs_nospace(struct rpc_task *task)
 494{
 495        struct rpc_rqst *req = task->tk_rqstp;
 496        struct rpc_xprt *xprt = req->rq_xprt;
 497        struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 498        int ret = -EAGAIN;
 499
 500        dprintk("RPC: %5u xmit incomplete (%u left of %u)\n",
 501                        task->tk_pid, req->rq_slen - req->rq_bytes_sent,
 502                        req->rq_slen);
 503
 504        /* Protect against races with write_space */
 505        spin_lock_bh(&xprt->transport_lock);
 506
 507        /* Don't race with disconnect */
 508        if (xprt_connected(xprt)) {
 509                if (test_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags)) {
 510                        /*
 511                         * Notify TCP that we're limited by the application
 512                         * window size
 513                         */
 514                        set_bit(SOCK_NOSPACE, &transport->sock->flags);
 515                        transport->inet->sk_write_pending++;
 516                        /* ...and wait for more buffer space */
 517                        xprt_wait_for_buffer_space(task, xs_nospace_callback);
 518                }
 519        } else {
 520                clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 521                ret = -ENOTCONN;
 522        }
 523
 524        spin_unlock_bh(&xprt->transport_lock);
 525        return ret;
 526}
 527
 528/*
 529 * Construct a stream transport record marker in @buf.
 530 */
 531static inline void xs_encode_stream_record_marker(struct xdr_buf *buf)
 532{
 533        u32 reclen = buf->len - sizeof(rpc_fraghdr);
 534        rpc_fraghdr *base = buf->head[0].iov_base;
 535        *base = cpu_to_be32(RPC_LAST_STREAM_FRAGMENT | reclen);
 536}
 537
 538/**
 539 * xs_local_send_request - write an RPC request to an AF_LOCAL socket
 540 * @task: RPC task that manages the state of an RPC request
 541 *
 542 * Return values:
 543 *        0:    The request has been sent
 544 *   EAGAIN:    The socket was blocked, please call again later to
 545 *              complete the request
 546 * ENOTCONN:    Caller needs to invoke connect logic then call again
 547 *    other:    Some other error occured, the request was not sent
 548 */
 549static int xs_local_send_request(struct rpc_task *task)
 550{
 551        struct rpc_rqst *req = task->tk_rqstp;
 552        struct rpc_xprt *xprt = req->rq_xprt;
 553        struct sock_xprt *transport =
 554                                container_of(xprt, struct sock_xprt, xprt);
 555        struct xdr_buf *xdr = &req->rq_snd_buf;
 556        int status;
 557
 558        xs_encode_stream_record_marker(&req->rq_snd_buf);
 559
 560        xs_pktdump("packet data:",
 561                        req->rq_svec->iov_base, req->rq_svec->iov_len);
 562
 563        status = xs_sendpages(transport->sock, NULL, 0,
 564                                                xdr, req->rq_bytes_sent);
 565        dprintk("RPC:       %s(%u) = %d\n",
 566                        __func__, xdr->len - req->rq_bytes_sent, status);
 567        if (likely(status >= 0)) {
 568                req->rq_bytes_sent += status;
 569                req->rq_xmit_bytes_sent += status;
 570                if (likely(req->rq_bytes_sent >= req->rq_slen)) {
 571                        req->rq_bytes_sent = 0;
 572                        return 0;
 573                }
 574                status = -EAGAIN;
 575        }
 576
 577        switch (status) {
 578        case -EAGAIN:
 579                status = xs_nospace(task);
 580                break;
 581        default:
 582                dprintk("RPC:       sendmsg returned unrecognized error %d\n",
 583                        -status);
 584        case -EPIPE:
 585                xs_close(xprt);
 586                status = -ENOTCONN;
 587        }
 588
 589        return status;
 590}
 591
 592/**
 593 * xs_udp_send_request - write an RPC request to a UDP socket
 594 * @task: address of RPC task that manages the state of an RPC request
 595 *
 596 * Return values:
 597 *        0:    The request has been sent
 598 *   EAGAIN:    The socket was blocked, please call again later to
 599 *              complete the request
 600 * ENOTCONN:    Caller needs to invoke connect logic then call again
 601 *    other:    Some other error occurred, the request was not sent
 602 */
 603static int xs_udp_send_request(struct rpc_task *task)
 604{
 605        struct rpc_rqst *req = task->tk_rqstp;
 606        struct rpc_xprt *xprt = req->rq_xprt;
 607        struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 608        struct xdr_buf *xdr = &req->rq_snd_buf;
 609        int status;
 610
 611        xs_pktdump("packet data:",
 612                                req->rq_svec->iov_base,
 613                                req->rq_svec->iov_len);
 614
 615        if (!xprt_bound(xprt))
 616                return -ENOTCONN;
 617        status = xs_sendpages(transport->sock,
 618                              xs_addr(xprt),
 619                              xprt->addrlen, xdr,
 620                              req->rq_bytes_sent);
 621
 622        dprintk("RPC:       xs_udp_send_request(%u) = %d\n",
 623                        xdr->len - req->rq_bytes_sent, status);
 624
 625        if (status >= 0) {
 626                req->rq_xmit_bytes_sent += status;
 627                if (status >= req->rq_slen)
 628                        return 0;
 629                /* Still some bytes left; set up for a retry later. */
 630                status = -EAGAIN;
 631        }
 632
 633        switch (status) {
 634        case -ENOTSOCK:
 635                status = -ENOTCONN;
 636                /* Should we call xs_close() here? */
 637                break;
 638        case -EAGAIN:
 639                status = xs_nospace(task);
 640                break;
 641        default:
 642                dprintk("RPC:       sendmsg returned unrecognized error %d\n",
 643                        -status);
 644        case -ENETUNREACH:
 645        case -EPIPE:
 646        case -ECONNREFUSED:
 647                /* When the server has died, an ICMP port unreachable message
 648                 * prompts ECONNREFUSED. */
 649                clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 650        }
 651
 652        return status;
 653}
 654
 655/**
 656 * xs_tcp_shutdown - gracefully shut down a TCP socket
 657 * @xprt: transport
 658 *
 659 * Initiates a graceful shutdown of the TCP socket by calling the
 660 * equivalent of shutdown(SHUT_WR);
 661 */
 662static void xs_tcp_shutdown(struct rpc_xprt *xprt)
 663{
 664        struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 665        struct socket *sock = transport->sock;
 666
 667        if (sock != NULL)
 668                kernel_sock_shutdown(sock, SHUT_WR);
 669}
 670
 671/**
 672 * xs_tcp_send_request - write an RPC request to a TCP socket
 673 * @task: address of RPC task that manages the state of an RPC request
 674 *
 675 * Return values:
 676 *        0:    The request has been sent
 677 *   EAGAIN:    The socket was blocked, please call again later to
 678 *              complete the request
 679 * ENOTCONN:    Caller needs to invoke connect logic then call again
 680 *    other:    Some other error occurred, the request was not sent
 681 *
 682 * XXX: In the case of soft timeouts, should we eventually give up
 683 *      if sendmsg is not able to make progress?
 684 */
 685static int xs_tcp_send_request(struct rpc_task *task)
 686{
 687        struct rpc_rqst *req = task->tk_rqstp;
 688        struct rpc_xprt *xprt = req->rq_xprt;
 689        struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 690        struct xdr_buf *xdr = &req->rq_snd_buf;
 691        int status;
 692
 693        xs_encode_stream_record_marker(&req->rq_snd_buf);
 694
 695        xs_pktdump("packet data:",
 696                                req->rq_svec->iov_base,
 697                                req->rq_svec->iov_len);
 698
 699        /* Continue transmitting the packet/record. We must be careful
 700         * to cope with writespace callbacks arriving _after_ we have
 701         * called sendmsg(). */
 702        while (1) {
 703                status = xs_sendpages(transport->sock,
 704                                        NULL, 0, xdr, req->rq_bytes_sent);
 705
 706                dprintk("RPC:       xs_tcp_send_request(%u) = %d\n",
 707                                xdr->len - req->rq_bytes_sent, status);
 708
 709                if (unlikely(status < 0))
 710                        break;
 711
 712                /* If we've sent the entire packet, immediately
 713                 * reset the count of bytes sent. */
 714                req->rq_bytes_sent += status;
 715                req->rq_xmit_bytes_sent += status;
 716                if (likely(req->rq_bytes_sent >= req->rq_slen)) {
 717                        req->rq_bytes_sent = 0;
 718                        return 0;
 719                }
 720
 721                if (status != 0)
 722                        continue;
 723                status = -EAGAIN;
 724                break;
 725        }
 726
 727        switch (status) {
 728        case -ENOTSOCK:
 729                status = -ENOTCONN;
 730                /* Should we call xs_close() here? */
 731                break;
 732        case -EAGAIN:
 733                status = xs_nospace(task);
 734                break;
 735        default:
 736                dprintk("RPC:       sendmsg returned unrecognized error %d\n",
 737                        -status);
 738        case -ECONNRESET:
 739                xs_tcp_shutdown(xprt);
 740        case -ECONNREFUSED:
 741        case -ENOTCONN:
 742        case -EPIPE:
 743                clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 744        }
 745
 746        return status;
 747}
 748
 749/**
 750 * xs_tcp_release_xprt - clean up after a tcp transmission
 751 * @xprt: transport
 752 * @task: rpc task
 753 *
 754 * This cleans up if an error causes us to abort the transmission of a request.
 755 * In this case, the socket may need to be reset in order to avoid confusing
 756 * the server.
 757 */
 758static void xs_tcp_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
 759{
 760        struct rpc_rqst *req;
 761
 762        if (task != xprt->snd_task)
 763                return;
 764        if (task == NULL)
 765                goto out_release;
 766        req = task->tk_rqstp;
 767        if (req == NULL)
 768                goto out_release;
 769        if (req->rq_bytes_sent == 0)
 770                goto out_release;
 771        if (req->rq_bytes_sent == req->rq_snd_buf.len)
 772                goto out_release;
 773        set_bit(XPRT_CLOSE_WAIT, &task->tk_xprt->state);
 774out_release:
 775        xprt_release_xprt(xprt, task);
 776}
 777
 778static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk)
 779{
 780        transport->old_data_ready = sk->sk_data_ready;
 781        transport->old_state_change = sk->sk_state_change;
 782        transport->old_write_space = sk->sk_write_space;
 783}
 784
 785static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk)
 786{
 787        sk->sk_data_ready = transport->old_data_ready;
 788        sk->sk_state_change = transport->old_state_change;
 789        sk->sk_write_space = transport->old_write_space;
 790}
 791
 792static void xs_reset_transport(struct sock_xprt *transport)
 793{
 794        struct socket *sock = transport->sock;
 795        struct sock *sk = transport->inet;
 796
 797        if (sk == NULL)
 798                return;
 799
 800        transport->srcport = 0;
 801
 802        write_lock_bh(&sk->sk_callback_lock);
 803        transport->inet = NULL;
 804        transport->sock = NULL;
 805
 806        sk->sk_user_data = NULL;
 807
 808        xs_restore_old_callbacks(transport, sk);
 809        write_unlock_bh(&sk->sk_callback_lock);
 810
 811        sk->sk_no_check = 0;
 812
 813        sock_release(sock);
 814}
 815
 816/**
 817 * xs_close - close a socket
 818 * @xprt: transport
 819 *
 820 * This is used when all requests are complete; ie, no DRC state remains
 821 * on the server we want to save.
 822 *
 823 * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with
 824 * xs_reset_transport() zeroing the socket from underneath a writer.
 825 */
 826static void xs_close(struct rpc_xprt *xprt)
 827{
 828        struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 829
 830        dprintk("RPC:       xs_close xprt %p\n", xprt);
 831
 832        xs_reset_transport(transport);
 833        xprt->reestablish_timeout = 0;
 834
 835        smp_mb__before_clear_bit();
 836        clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
 837        clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
 838        clear_bit(XPRT_CLOSING, &xprt->state);
 839        smp_mb__after_clear_bit();
 840        xprt_disconnect_done(xprt);
 841}
 842
 843static void xs_tcp_close(struct rpc_xprt *xprt)
 844{
 845        if (test_and_clear_bit(XPRT_CONNECTION_CLOSE, &xprt->state))
 846                xs_close(xprt);
 847        else
 848                xs_tcp_shutdown(xprt);
 849}
 850
 851/**
 852 * xs_destroy - prepare to shutdown a transport
 853 * @xprt: doomed transport
 854 *
 855 */
 856static void xs_destroy(struct rpc_xprt *xprt)
 857{
 858        struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 859
 860        dprintk("RPC:       xs_destroy xprt %p\n", xprt);
 861
 862        cancel_delayed_work_sync(&transport->connect_worker);
 863
 864        xs_close(xprt);
 865        xs_free_peer_addresses(xprt);
 866        xprt_free(xprt);
 867        module_put(THIS_MODULE);
 868}
 869
 870static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
 871{
 872        return (struct rpc_xprt *) sk->sk_user_data;
 873}
 874
 875static int xs_local_copy_to_xdr(struct xdr_buf *xdr, struct sk_buff *skb)
 876{
 877        struct xdr_skb_reader desc = {
 878                .skb            = skb,
 879                .offset         = sizeof(rpc_fraghdr),
 880                .count          = skb->len - sizeof(rpc_fraghdr),
 881        };
 882
 883        if (xdr_partial_copy_from_skb(xdr, 0, &desc, xdr_skb_read_bits) < 0)
 884                return -1;
 885        if (desc.count)
 886                return -1;
 887        return 0;
 888}
 889
 890/**
 891 * xs_local_data_ready - "data ready" callback for AF_LOCAL sockets
 892 * @sk: socket with data to read
 893 * @len: how much data to read
 894 *
 895 * Currently this assumes we can read the whole reply in a single gulp.
 896 */
 897static void xs_local_data_ready(struct sock *sk, int len)
 898{
 899        struct rpc_task *task;
 900        struct rpc_xprt *xprt;
 901        struct rpc_rqst *rovr;
 902        struct sk_buff *skb;
 903        int err, repsize, copied;
 904        u32 _xid;
 905        __be32 *xp;
 906
 907        read_lock_bh(&sk->sk_callback_lock);
 908        dprintk("RPC:       %s...\n", __func__);
 909        xprt = xprt_from_sock(sk);
 910        if (xprt == NULL)
 911                goto out;
 912
 913        skb = skb_recv_datagram(sk, 0, 1, &err);
 914        if (skb == NULL)
 915                goto out;
 916
 917        repsize = skb->len - sizeof(rpc_fraghdr);
 918        if (repsize < 4) {
 919                dprintk("RPC:       impossible RPC reply size %d\n", repsize);
 920                goto dropit;
 921        }
 922
 923        /* Copy the XID from the skb... */
 924        xp = skb_header_pointer(skb, sizeof(rpc_fraghdr), sizeof(_xid), &_xid);
 925        if (xp == NULL)
 926                goto dropit;
 927
 928        /* Look up and lock the request corresponding to the given XID */
 929        spin_lock(&xprt->transport_lock);
 930        rovr = xprt_lookup_rqst(xprt, *xp);
 931        if (!rovr)
 932                goto out_unlock;
 933        task = rovr->rq_task;
 934
 935        copied = rovr->rq_private_buf.buflen;
 936        if (copied > repsize)
 937                copied = repsize;
 938
 939        if (xs_local_copy_to_xdr(&rovr->rq_private_buf, skb)) {
 940                dprintk("RPC:       sk_buff copy failed\n");
 941                goto out_unlock;
 942        }
 943
 944        xprt_complete_rqst(task, copied);
 945
 946 out_unlock:
 947        spin_unlock(&xprt->transport_lock);
 948 dropit:
 949        skb_free_datagram(sk, skb);
 950 out:
 951        read_unlock_bh(&sk->sk_callback_lock);
 952}
 953
 954/**
 955 * xs_udp_data_ready - "data ready" callback for UDP sockets
 956 * @sk: socket with data to read
 957 * @len: how much data to read
 958 *
 959 */
 960static void xs_udp_data_ready(struct sock *sk, int len)
 961{
 962        struct rpc_task *task;
 963        struct rpc_xprt *xprt;
 964        struct rpc_rqst *rovr;
 965        struct sk_buff *skb;
 966        int err, repsize, copied;
 967        u32 _xid;
 968        __be32 *xp;
 969
 970        read_lock_bh(&sk->sk_callback_lock);
 971        dprintk("RPC:       xs_udp_data_ready...\n");
 972        if (!(xprt = xprt_from_sock(sk)))
 973                goto out;
 974
 975        if ((skb = skb_recv_datagram(sk, 0, 1, &err)) == NULL)
 976                goto out;
 977
 978        repsize = skb->len - sizeof(struct udphdr);
 979        if (repsize < 4) {
 980                dprintk("RPC:       impossible RPC reply size %d!\n", repsize);
 981                goto dropit;
 982        }
 983
 984        /* Copy the XID from the skb... */
 985        xp = skb_header_pointer(skb, sizeof(struct udphdr),
 986                                sizeof(_xid), &_xid);
 987        if (xp == NULL)
 988                goto dropit;
 989
 990        /* Look up and lock the request corresponding to the given XID */
 991        spin_lock(&xprt->transport_lock);
 992        rovr = xprt_lookup_rqst(xprt, *xp);
 993        if (!rovr)
 994                goto out_unlock;
 995        task = rovr->rq_task;
 996
 997        if ((copied = rovr->rq_private_buf.buflen) > repsize)
 998                copied = repsize;
 999
1000        /* Suck it into the iovec, verify checksum if not done by hw. */
1001        if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
1002                UDPX_INC_STATS_BH(sk, UDP_MIB_INERRORS);
1003                goto out_unlock;
1004        }
1005
1006        UDPX_INC_STATS_BH(sk, UDP_MIB_INDATAGRAMS);
1007
1008        xprt_adjust_cwnd(task, copied);
1009        xprt_complete_rqst(task, copied);
1010
1011 out_unlock:
1012        spin_unlock(&xprt->transport_lock);
1013 dropit:
1014        skb_free_datagram(sk, skb);
1015 out:
1016        read_unlock_bh(&sk->sk_callback_lock);
1017}
1018
1019/*
1020 * Helper function to force a TCP close if the server is sending
1021 * junk and/or it has put us in CLOSE_WAIT
1022 */
1023static void xs_tcp_force_close(struct rpc_xprt *xprt)
1024{
1025        set_bit(XPRT_CONNECTION_CLOSE, &xprt->state);
1026        xprt_force_disconnect(xprt);
1027}
1028
1029static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
1030{
1031        struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1032        size_t len, used;
1033        char *p;
1034
1035        p = ((char *) &transport->tcp_fraghdr) + transport->tcp_offset;
1036        len = sizeof(transport->tcp_fraghdr) - transport->tcp_offset;
1037        used = xdr_skb_read_bits(desc, p, len);
1038        transport->tcp_offset += used;
1039        if (used != len)
1040                return;
1041
1042        transport->tcp_reclen = ntohl(transport->tcp_fraghdr);
1043        if (transport->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
1044                transport->tcp_flags |= TCP_RCV_LAST_FRAG;
1045        else
1046                transport->tcp_flags &= ~TCP_RCV_LAST_FRAG;
1047        transport->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
1048
1049        transport->tcp_flags &= ~TCP_RCV_COPY_FRAGHDR;
1050        transport->tcp_offset = 0;
1051
1052        /* Sanity check of the record length */
1053        if (unlikely(transport->tcp_reclen < 8)) {
1054                dprintk("RPC:       invalid TCP record fragment length\n");
1055                xs_tcp_force_close(xprt);
1056                return;
1057        }
1058        dprintk("RPC:       reading TCP record fragment of length %d\n",
1059                        transport->tcp_reclen);
1060}
1061
1062static void xs_tcp_check_fraghdr(struct sock_xprt *transport)
1063{
1064        if (transport->tcp_offset == transport->tcp_reclen) {
1065                transport->tcp_flags |= TCP_RCV_COPY_FRAGHDR;
1066                transport->tcp_offset = 0;
1067                if (transport->tcp_flags & TCP_RCV_LAST_FRAG) {
1068                        transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1069                        transport->tcp_flags |= TCP_RCV_COPY_XID;
1070                        transport->tcp_copied = 0;
1071                }
1072        }
1073}
1074
1075static inline void xs_tcp_read_xid(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1076{
1077        size_t len, used;
1078        char *p;
1079
1080        len = sizeof(transport->tcp_xid) - transport->tcp_offset;
1081        dprintk("RPC:       reading XID (%Zu bytes)\n", len);
1082        p = ((char *) &transport->tcp_xid) + transport->tcp_offset;
1083        used = xdr_skb_read_bits(desc, p, len);
1084        transport->tcp_offset += used;
1085        if (used != len)
1086                return;
1087        transport->tcp_flags &= ~TCP_RCV_COPY_XID;
1088        transport->tcp_flags |= TCP_RCV_READ_CALLDIR;
1089        transport->tcp_copied = 4;
1090        dprintk("RPC:       reading %s XID %08x\n",
1091                        (transport->tcp_flags & TCP_RPC_REPLY) ? "reply for"
1092                                                              : "request with",
1093                        ntohl(transport->tcp_xid));
1094        xs_tcp_check_fraghdr(transport);
1095}
1096
1097static inline void xs_tcp_read_calldir(struct sock_xprt *transport,
1098                                       struct xdr_skb_reader *desc)
1099{
1100        size_t len, used;
1101        u32 offset;
1102        char *p;
1103
1104        /*
1105         * We want transport->tcp_offset to be 8 at the end of this routine
1106         * (4 bytes for the xid and 4 bytes for the call/reply flag).
1107         * When this function is called for the first time,
1108         * transport->tcp_offset is 4 (after having already read the xid).
1109         */
1110        offset = transport->tcp_offset - sizeof(transport->tcp_xid);
1111        len = sizeof(transport->tcp_calldir) - offset;
1112        dprintk("RPC:       reading CALL/REPLY flag (%Zu bytes)\n", len);
1113        p = ((char *) &transport->tcp_calldir) + offset;
1114        used = xdr_skb_read_bits(desc, p, len);
1115        transport->tcp_offset += used;
1116        if (used != len)
1117                return;
1118        transport->tcp_flags &= ~TCP_RCV_READ_CALLDIR;
1119        /*
1120         * We don't yet have the XDR buffer, so we will write the calldir
1121         * out after we get the buffer from the 'struct rpc_rqst'
1122         */
1123        switch (ntohl(transport->tcp_calldir)) {
1124        case RPC_REPLY:
1125                transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1126                transport->tcp_flags |= TCP_RCV_COPY_DATA;
1127                transport->tcp_flags |= TCP_RPC_REPLY;
1128                break;
1129        case RPC_CALL:
1130                transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1131                transport->tcp_flags |= TCP_RCV_COPY_DATA;
1132                transport->tcp_flags &= ~TCP_RPC_REPLY;
1133                break;
1134        default:
1135                dprintk("RPC:       invalid request message type\n");
1136                xs_tcp_force_close(&transport->xprt);
1137        }
1138        xs_tcp_check_fraghdr(transport);
1139}
1140
1141static inline void xs_tcp_read_common(struct rpc_xprt *xprt,
1142                                     struct xdr_skb_reader *desc,
1143                                     struct rpc_rqst *req)
1144{
1145        struct sock_xprt *transport =
1146                                container_of(xprt, struct sock_xprt, xprt);
1147        struct xdr_buf *rcvbuf;
1148        size_t len;
1149        ssize_t r;
1150
1151        rcvbuf = &req->rq_private_buf;
1152
1153        if (transport->tcp_flags & TCP_RCV_COPY_CALLDIR) {
1154                /*
1155                 * Save the RPC direction in the XDR buffer
1156                 */
1157                memcpy(rcvbuf->head[0].iov_base + transport->tcp_copied,
1158                        &transport->tcp_calldir,
1159                        sizeof(transport->tcp_calldir));
1160                transport->tcp_copied += sizeof(transport->tcp_calldir);
1161                transport->tcp_flags &= ~TCP_RCV_COPY_CALLDIR;
1162        }
1163
1164        len = desc->count;
1165        if (len > transport->tcp_reclen - transport->tcp_offset) {
1166                struct xdr_skb_reader my_desc;
1167
1168                len = transport->tcp_reclen - transport->tcp_offset;
1169                memcpy(&my_desc, desc, sizeof(my_desc));
1170                my_desc.count = len;
1171                r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
1172                                          &my_desc, xdr_skb_read_bits);
1173                desc->count -= r;
1174                desc->offset += r;
1175        } else
1176                r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
1177                                          desc, xdr_skb_read_bits);
1178
1179        if (r > 0) {
1180                transport->tcp_copied += r;
1181                transport->tcp_offset += r;
1182        }
1183        if (r != len) {
1184                /* Error when copying to the receive buffer,
1185                 * usually because we weren't able to allocate
1186                 * additional buffer pages. All we can do now
1187                 * is turn off TCP_RCV_COPY_DATA, so the request
1188                 * will not receive any additional updates,
1189                 * and time out.
1190                 * Any remaining data from this record will
1191                 * be discarded.
1192                 */
1193                transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1194                dprintk("RPC:       XID %08x truncated request\n",
1195                                ntohl(transport->tcp_xid));
1196                dprintk("RPC:       xprt = %p, tcp_copied = %lu, "
1197                                "tcp_offset = %u, tcp_reclen = %u\n",
1198                                xprt, transport->tcp_copied,
1199                                transport->tcp_offset, transport->tcp_reclen);
1200                return;
1201        }
1202
1203        dprintk("RPC:       XID %08x read %Zd bytes\n",
1204                        ntohl(transport->tcp_xid), r);
1205        dprintk("RPC:       xprt = %p, tcp_copied = %lu, tcp_offset = %u, "
1206                        "tcp_reclen = %u\n", xprt, transport->tcp_copied,
1207                        transport->tcp_offset, transport->tcp_reclen);
1208
1209        if (transport->tcp_copied == req->rq_private_buf.buflen)
1210                transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1211        else if (transport->tcp_offset == transport->tcp_reclen) {
1212                if (transport->tcp_flags & TCP_RCV_LAST_FRAG)
1213                        transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1214        }
1215}
1216
1217/*
1218 * Finds the request corresponding to the RPC xid and invokes the common
1219 * tcp read code to read the data.
1220 */
1221static inline int xs_tcp_read_reply(struct rpc_xprt *xprt,
1222                                    struct xdr_skb_reader *desc)
1223{
1224        struct sock_xprt *transport =
1225                                container_of(xprt, struct sock_xprt, xprt);
1226        struct rpc_rqst *req;
1227
1228        dprintk("RPC:       read reply XID %08x\n", ntohl(transport->tcp_xid));
1229
1230        /* Find and lock the request corresponding to this xid */
1231        spin_lock(&xprt->transport_lock);
1232        req = xprt_lookup_rqst(xprt, transport->tcp_xid);
1233        if (!req) {
1234                dprintk("RPC:       XID %08x request not found!\n",
1235                                ntohl(transport->tcp_xid));
1236                spin_unlock(&xprt->transport_lock);
1237                return -1;
1238        }
1239
1240        xs_tcp_read_common(xprt, desc, req);
1241
1242        if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1243                xprt_complete_rqst(req->rq_task, transport->tcp_copied);
1244
1245        spin_unlock(&xprt->transport_lock);
1246        return 0;
1247}
1248
1249#if defined(CONFIG_SUNRPC_BACKCHANNEL)
1250/*
1251 * Obtains an rpc_rqst previously allocated and invokes the common
1252 * tcp read code to read the data.  The result is placed in the callback
1253 * queue.
1254 * If we're unable to obtain the rpc_rqst we schedule the closing of the
1255 * connection and return -1.
1256 */
1257static inline int xs_tcp_read_callback(struct rpc_xprt *xprt,
1258                                       struct xdr_skb_reader *desc)
1259{
1260        struct sock_xprt *transport =
1261                                container_of(xprt, struct sock_xprt, xprt);
1262        struct rpc_rqst *req;
1263
1264        req = xprt_alloc_bc_request(xprt);
1265        if (req == NULL) {
1266                printk(KERN_WARNING "Callback slot table overflowed\n");
1267                xprt_force_disconnect(xprt);
1268                return -1;
1269        }
1270
1271        req->rq_xid = transport->tcp_xid;
1272        dprintk("RPC:       read callback  XID %08x\n", ntohl(req->rq_xid));
1273        xs_tcp_read_common(xprt, desc, req);
1274
1275        if (!(transport->tcp_flags & TCP_RCV_COPY_DATA)) {
1276                struct svc_serv *bc_serv = xprt->bc_serv;
1277
1278                /*
1279                 * Add callback request to callback list.  The callback
1280                 * service sleeps on the sv_cb_waitq waiting for new
1281                 * requests.  Wake it up after adding enqueing the
1282                 * request.
1283                 */
1284                dprintk("RPC:       add callback request to list\n");
1285                spin_lock(&bc_serv->sv_cb_lock);
1286                list_add(&req->rq_bc_list, &bc_serv->sv_cb_list);
1287                spin_unlock(&bc_serv->sv_cb_lock);
1288                wake_up(&bc_serv->sv_cb_waitq);
1289        }
1290
1291        req->rq_private_buf.len = transport->tcp_copied;
1292
1293        return 0;
1294}
1295
1296static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1297                                        struct xdr_skb_reader *desc)
1298{
1299        struct sock_xprt *transport =
1300                                container_of(xprt, struct sock_xprt, xprt);
1301
1302        return (transport->tcp_flags & TCP_RPC_REPLY) ?
1303                xs_tcp_read_reply(xprt, desc) :
1304                xs_tcp_read_callback(xprt, desc);
1305}
1306#else
1307static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1308                                        struct xdr_skb_reader *desc)
1309{
1310        return xs_tcp_read_reply(xprt, desc);
1311}
1312#endif /* CONFIG_SUNRPC_BACKCHANNEL */
1313
1314/*
1315 * Read data off the transport.  This can be either an RPC_CALL or an
1316 * RPC_REPLY.  Relay the processing to helper functions.
1317 */
1318static void xs_tcp_read_data(struct rpc_xprt *xprt,
1319                                    struct xdr_skb_reader *desc)
1320{
1321        struct sock_xprt *transport =
1322                                container_of(xprt, struct sock_xprt, xprt);
1323
1324        if (_xs_tcp_read_data(xprt, desc) == 0)
1325                xs_tcp_check_fraghdr(transport);
1326        else {
1327                /*
1328                 * The transport_lock protects the request handling.
1329                 * There's no need to hold it to update the tcp_flags.
1330                 */
1331                transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1332        }
1333}
1334
1335static inline void xs_tcp_read_discard(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1336{
1337        size_t len;
1338
1339        len = transport->tcp_reclen - transport->tcp_offset;
1340        if (len > desc->count)
1341                len = desc->count;
1342        desc->count -= len;
1343        desc->offset += len;
1344        transport->tcp_offset += len;
1345        dprintk("RPC:       discarded %Zu bytes\n", len);
1346        xs_tcp_check_fraghdr(transport);
1347}
1348
1349static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
1350{
1351        struct rpc_xprt *xprt = rd_desc->arg.data;
1352        struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1353        struct xdr_skb_reader desc = {
1354                .skb    = skb,
1355                .offset = offset,
1356                .count  = len,
1357        };
1358
1359        dprintk("RPC:       xs_tcp_data_recv started\n");
1360        do {
1361                /* Read in a new fragment marker if necessary */
1362                /* Can we ever really expect to get completely empty fragments? */
1363                if (transport->tcp_flags & TCP_RCV_COPY_FRAGHDR) {
1364                        xs_tcp_read_fraghdr(xprt, &desc);
1365                        continue;
1366                }
1367                /* Read in the xid if necessary */
1368                if (transport->tcp_flags & TCP_RCV_COPY_XID) {
1369                        xs_tcp_read_xid(transport, &desc);
1370                        continue;
1371                }
1372                /* Read in the call/reply flag */
1373                if (transport->tcp_flags & TCP_RCV_READ_CALLDIR) {
1374                        xs_tcp_read_calldir(transport, &desc);
1375                        continue;
1376                }
1377                /* Read in the request data */
1378                if (transport->tcp_flags & TCP_RCV_COPY_DATA) {
1379                        xs_tcp_read_data(xprt, &desc);
1380                        continue;
1381                }
1382                /* Skip over any trailing bytes on short reads */
1383                xs_tcp_read_discard(transport, &desc);
1384        } while (desc.count);
1385        dprintk("RPC:       xs_tcp_data_recv done\n");
1386        return len - desc.count;
1387}
1388
1389/**
1390 * xs_tcp_data_ready - "data ready" callback for TCP sockets
1391 * @sk: socket with data to read
1392 * @bytes: how much data to read
1393 *
1394 */
1395static void xs_tcp_data_ready(struct sock *sk, int bytes)
1396{
1397        struct rpc_xprt *xprt;
1398        read_descriptor_t rd_desc;
1399        int read;
1400
1401        dprintk("RPC:       xs_tcp_data_ready...\n");
1402
1403        read_lock_bh(&sk->sk_callback_lock);
1404        if (!(xprt = xprt_from_sock(sk)))
1405                goto out;
1406        /* Any data means we had a useful conversation, so
1407         * the we don't need to delay the next reconnect
1408         */
1409        if (xprt->reestablish_timeout)
1410                xprt->reestablish_timeout = 0;
1411
1412        /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
1413        rd_desc.arg.data = xprt;
1414        do {
1415                rd_desc.count = 65536;
1416                read = tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
1417        } while (read > 0);
1418out:
1419        read_unlock_bh(&sk->sk_callback_lock);
1420}
1421
1422/*
1423 * Do the equivalent of linger/linger2 handling for dealing with
1424 * broken servers that don't close the socket in a timely
1425 * fashion
1426 */
1427static void xs_tcp_schedule_linger_timeout(struct rpc_xprt *xprt,
1428                unsigned long timeout)
1429{
1430        struct sock_xprt *transport;
1431
1432        if (xprt_test_and_set_connecting(xprt))
1433                return;
1434        set_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1435        transport = container_of(xprt, struct sock_xprt, xprt);
1436        queue_delayed_work(rpciod_workqueue, &transport->connect_worker,
1437                           timeout);
1438}
1439
1440static void xs_tcp_cancel_linger_timeout(struct rpc_xprt *xprt)
1441{
1442        struct sock_xprt *transport;
1443
1444        transport = container_of(xprt, struct sock_xprt, xprt);
1445
1446        if (!test_bit(XPRT_CONNECTION_ABORT, &xprt->state) ||
1447            !cancel_delayed_work(&transport->connect_worker))
1448                return;
1449        clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1450        xprt_clear_connecting(xprt);
1451}
1452
1453static void xs_sock_reset_connection_flags(struct rpc_xprt *xprt)
1454{
1455        smp_mb__before_clear_bit();
1456        clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1457        clear_bit(XPRT_CONNECTION_CLOSE, &xprt->state);
1458        clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1459        clear_bit(XPRT_CLOSING, &xprt->state);
1460        smp_mb__after_clear_bit();
1461}
1462
1463static void xs_sock_mark_closed(struct rpc_xprt *xprt)
1464{
1465        xs_sock_reset_connection_flags(xprt);
1466        /* Mark transport as closed and wake up all pending tasks */
1467        xprt_disconnect_done(xprt);
1468}
1469
1470/**
1471 * xs_tcp_state_change - callback to handle TCP socket state changes
1472 * @sk: socket whose state has changed
1473 *
1474 */
1475static void xs_tcp_state_change(struct sock *sk)
1476{
1477        struct rpc_xprt *xprt;
1478
1479        read_lock_bh(&sk->sk_callback_lock);
1480        if (!(xprt = xprt_from_sock(sk)))
1481                goto out;
1482        dprintk("RPC:       xs_tcp_state_change client %p...\n", xprt);
1483        dprintk("RPC:       state %x conn %d dead %d zapped %d sk_shutdown %d\n",
1484                        sk->sk_state, xprt_connected(xprt),
1485                        sock_flag(sk, SOCK_DEAD),
1486                        sock_flag(sk, SOCK_ZAPPED),
1487                        sk->sk_shutdown);
1488
1489        switch (sk->sk_state) {
1490        case TCP_ESTABLISHED:
1491                spin_lock(&xprt->transport_lock);
1492                if (!xprt_test_and_set_connected(xprt)) {
1493                        struct sock_xprt *transport = container_of(xprt,
1494                                        struct sock_xprt, xprt);
1495
1496                        /* Reset TCP record info */
1497                        transport->tcp_offset = 0;
1498                        transport->tcp_reclen = 0;
1499                        transport->tcp_copied = 0;
1500                        transport->tcp_flags =
1501                                TCP_RCV_COPY_FRAGHDR | TCP_RCV_COPY_XID;
1502
1503                        xprt_wake_pending_tasks(xprt, -EAGAIN);
1504                }
1505                spin_unlock(&xprt->transport_lock);
1506                break;
1507        case TCP_FIN_WAIT1:
1508                /* The client initiated a shutdown of the socket */
1509                xprt->connect_cookie++;
1510                xprt->reestablish_timeout = 0;
1511                set_bit(XPRT_CLOSING, &xprt->state);
1512                smp_mb__before_clear_bit();
1513                clear_bit(XPRT_CONNECTED, &xprt->state);
1514                clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1515                smp_mb__after_clear_bit();
1516                xs_tcp_schedule_linger_timeout(xprt, xs_tcp_fin_timeout);
1517                break;
1518        case TCP_CLOSE_WAIT:
1519                /* The server initiated a shutdown of the socket */
1520                xprt->connect_cookie++;
1521                clear_bit(XPRT_CONNECTED, &xprt->state);
1522                xs_tcp_force_close(xprt);
1523        case TCP_CLOSING:
1524                /*
1525                 * If the server closed down the connection, make sure that
1526                 * we back off before reconnecting
1527                 */
1528                if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1529                        xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1530                break;
1531        case TCP_LAST_ACK:
1532                set_bit(XPRT_CLOSING, &xprt->state);
1533                xs_tcp_schedule_linger_timeout(xprt, xs_tcp_fin_timeout);
1534                smp_mb__before_clear_bit();
1535                clear_bit(XPRT_CONNECTED, &xprt->state);
1536                smp_mb__after_clear_bit();
1537                break;
1538        case TCP_CLOSE:
1539                xs_tcp_cancel_linger_timeout(xprt);
1540                xs_sock_mark_closed(xprt);
1541        }
1542 out:
1543        read_unlock_bh(&sk->sk_callback_lock);
1544}
1545
1546static void xs_write_space(struct sock *sk)
1547{
1548        struct socket *sock;
1549        struct rpc_xprt *xprt;
1550
1551        if (unlikely(!(sock = sk->sk_socket)))
1552                return;
1553        clear_bit(SOCK_NOSPACE, &sock->flags);
1554
1555        if (unlikely(!(xprt = xprt_from_sock(sk))))
1556                return;
1557        if (test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags) == 0)
1558                return;
1559
1560        xprt_write_space(xprt);
1561}
1562
1563/**
1564 * xs_udp_write_space - callback invoked when socket buffer space
1565 *                             becomes available
1566 * @sk: socket whose state has changed
1567 *
1568 * Called when more output buffer space is available for this socket.
1569 * We try not to wake our writers until they can make "significant"
1570 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1571 * with a bunch of small requests.
1572 */
1573static void xs_udp_write_space(struct sock *sk)
1574{
1575        read_lock_bh(&sk->sk_callback_lock);
1576
1577        /* from net/core/sock.c:sock_def_write_space */
1578        if (sock_writeable(sk))
1579                xs_write_space(sk);
1580
1581        read_unlock_bh(&sk->sk_callback_lock);
1582}
1583
1584/**
1585 * xs_tcp_write_space - callback invoked when socket buffer space
1586 *                             becomes available
1587 * @sk: socket whose state has changed
1588 *
1589 * Called when more output buffer space is available for this socket.
1590 * We try not to wake our writers until they can make "significant"
1591 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1592 * with a bunch of small requests.
1593 */
1594static void xs_tcp_write_space(struct sock *sk)
1595{
1596        read_lock_bh(&sk->sk_callback_lock);
1597
1598        /* from net/core/stream.c:sk_stream_write_space */
1599        if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk))
1600                xs_write_space(sk);
1601
1602        read_unlock_bh(&sk->sk_callback_lock);
1603}
1604
1605static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
1606{
1607        struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1608        struct sock *sk = transport->inet;
1609
1610        if (transport->rcvsize) {
1611                sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1612                sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
1613        }
1614        if (transport->sndsize) {
1615                sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1616                sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1617                sk->sk_write_space(sk);
1618        }
1619}
1620
1621/**
1622 * xs_udp_set_buffer_size - set send and receive limits
1623 * @xprt: generic transport
1624 * @sndsize: requested size of send buffer, in bytes
1625 * @rcvsize: requested size of receive buffer, in bytes
1626 *
1627 * Set socket send and receive buffer size limits.
1628 */
1629static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1630{
1631        struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1632
1633        transport->sndsize = 0;
1634        if (sndsize)
1635                transport->sndsize = sndsize + 1024;
1636        transport->rcvsize = 0;
1637        if (rcvsize)
1638                transport->rcvsize = rcvsize + 1024;
1639
1640        xs_udp_do_set_buffer_size(xprt);
1641}
1642
1643/**
1644 * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1645 * @task: task that timed out
1646 *
1647 * Adjust the congestion window after a retransmit timeout has occurred.
1648 */
1649static void xs_udp_timer(struct rpc_task *task)
1650{
1651        xprt_adjust_cwnd(task, -ETIMEDOUT);
1652}
1653
1654static unsigned short xs_get_random_port(void)
1655{
1656        unsigned short range = xprt_max_resvport - xprt_min_resvport;
1657        unsigned short rand = (unsigned short) net_random() % range;
1658        return rand + xprt_min_resvport;
1659}
1660
1661/**
1662 * xs_set_port - reset the port number in the remote endpoint address
1663 * @xprt: generic transport
1664 * @port: new port number
1665 *
1666 */
1667static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1668{
1669        dprintk("RPC:       setting port for xprt %p to %u\n", xprt, port);
1670
1671        rpc_set_port(xs_addr(xprt), port);
1672        xs_update_peer_port(xprt);
1673}
1674
1675static unsigned short xs_get_srcport(struct sock_xprt *transport)
1676{
1677        unsigned short port = transport->srcport;
1678
1679        if (port == 0 && transport->xprt.resvport)
1680                port = xs_get_random_port();
1681        return port;
1682}
1683
1684static unsigned short xs_next_srcport(struct sock_xprt *transport, unsigned short port)
1685{
1686        if (transport->srcport != 0)
1687                transport->srcport = 0;
1688        if (!transport->xprt.resvport)
1689                return 0;
1690        if (port <= xprt_min_resvport || port > xprt_max_resvport)
1691                return xprt_max_resvport;
1692        return --port;
1693}
1694static int xs_bind(struct sock_xprt *transport, struct socket *sock)
1695{
1696        struct sockaddr_storage myaddr;
1697        int err, nloop = 0;
1698        unsigned short port = xs_get_srcport(transport);
1699        unsigned short last;
1700
1701        memcpy(&myaddr, &transport->srcaddr, transport->xprt.addrlen);
1702        do {
1703                rpc_set_port((struct sockaddr *)&myaddr, port);
1704                err = kernel_bind(sock, (struct sockaddr *)&myaddr,
1705                                transport->xprt.addrlen);
1706                if (port == 0)
1707                        break;
1708                if (err == 0) {
1709                        transport->srcport = port;
1710                        break;
1711                }
1712                last = port;
1713                port = xs_next_srcport(transport, port);
1714                if (port > last)
1715                        nloop++;
1716        } while (err == -EADDRINUSE && nloop != 2);
1717
1718        if (myaddr.ss_family == AF_INET)
1719                dprintk("RPC:       %s %pI4:%u: %s (%d)\n", __func__,
1720                                &((struct sockaddr_in *)&myaddr)->sin_addr,
1721                                port, err ? "failed" : "ok", err);
1722        else
1723                dprintk("RPC:       %s %pI6:%u: %s (%d)\n", __func__,
1724                                &((struct sockaddr_in6 *)&myaddr)->sin6_addr,
1725                                port, err ? "failed" : "ok", err);
1726        return err;
1727}
1728
1729/*
1730 * We don't support autobind on AF_LOCAL sockets
1731 */
1732static void xs_local_rpcbind(struct rpc_task *task)
1733{
1734        xprt_set_bound(task->tk_xprt);
1735}
1736
1737static void xs_local_set_port(struct rpc_xprt *xprt, unsigned short port)
1738{
1739}
1740
1741#ifdef CONFIG_DEBUG_LOCK_ALLOC
1742static struct lock_class_key xs_key[2];
1743static struct lock_class_key xs_slock_key[2];
1744
1745static inline void xs_reclassify_socketu(struct socket *sock)
1746{
1747        struct sock *sk = sock->sk;
1748
1749        sock_lock_init_class_and_name(sk, "slock-AF_LOCAL-RPC",
1750                &xs_slock_key[1], "sk_lock-AF_LOCAL-RPC", &xs_key[1]);
1751}
1752
1753static inline void xs_reclassify_socket4(struct socket *sock)
1754{
1755        struct sock *sk = sock->sk;
1756
1757        sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1758                &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1759}
1760
1761static inline void xs_reclassify_socket6(struct socket *sock)
1762{
1763        struct sock *sk = sock->sk;
1764
1765        sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1766                &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
1767}
1768
1769static inline void xs_reclassify_socket(int family, struct socket *sock)
1770{
1771        WARN_ON_ONCE(sock_owned_by_user(sock->sk));
1772        if (sock_owned_by_user(sock->sk))
1773                return;
1774
1775        switch (family) {
1776        case AF_LOCAL:
1777                xs_reclassify_socketu(sock);
1778                break;
1779        case AF_INET:
1780                xs_reclassify_socket4(sock);
1781                break;
1782        case AF_INET6:
1783                xs_reclassify_socket6(sock);
1784                break;
1785        }
1786}
1787#else
1788static inline void xs_reclassify_socketu(struct socket *sock)
1789{
1790}
1791
1792static inline void xs_reclassify_socket4(struct socket *sock)
1793{
1794}
1795
1796static inline void xs_reclassify_socket6(struct socket *sock)
1797{
1798}
1799
1800static inline void xs_reclassify_socket(int family, struct socket *sock)
1801{
1802}
1803#endif
1804
1805static struct socket *xs_create_sock(struct rpc_xprt *xprt,
1806                struct sock_xprt *transport, int family, int type, int protocol)
1807{
1808        struct socket *sock;
1809        int err;
1810
1811        err = __sock_create(xprt->xprt_net, family, type, protocol, &sock, 1);
1812        if (err < 0) {
1813                dprintk("RPC:       can't create %d transport socket (%d).\n",
1814                                protocol, -err);
1815                goto out;
1816        }
1817        xs_reclassify_socket(family, sock);
1818
1819        err = xs_bind(transport, sock);
1820        if (err) {
1821                sock_release(sock);
1822                goto out;
1823        }
1824
1825        return sock;
1826out:
1827        return ERR_PTR(err);
1828}
1829
1830static int xs_local_finish_connecting(struct rpc_xprt *xprt,
1831                                      struct socket *sock)
1832{
1833        struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1834                                                                        xprt);
1835
1836        if (!transport->inet) {
1837                struct sock *sk = sock->sk;
1838
1839                write_lock_bh(&sk->sk_callback_lock);
1840
1841                xs_save_old_callbacks(transport, sk);
1842
1843                sk->sk_user_data = xprt;
1844                sk->sk_data_ready = xs_local_data_ready;
1845                sk->sk_write_space = xs_udp_write_space;
1846                sk->sk_allocation = GFP_ATOMIC;
1847
1848                xprt_clear_connected(xprt);
1849
1850                /* Reset to new socket */
1851                transport->sock = sock;
1852                transport->inet = sk;
1853
1854                write_unlock_bh(&sk->sk_callback_lock);
1855        }
1856
1857        /* Tell the socket layer to start connecting... */
1858        xprt->stat.connect_count++;
1859        xprt->stat.connect_start = jiffies;
1860        return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, 0);
1861}
1862
1863/**
1864 * xs_local_setup_socket - create AF_LOCAL socket, connect to a local endpoint
1865 * @xprt: RPC transport to connect
1866 * @transport: socket transport to connect
1867 * @create_sock: function to create a socket of the correct type
1868 *
1869 * Invoked by a work queue tasklet.
1870 */
1871static void xs_local_setup_socket(struct work_struct *work)
1872{
1873        struct sock_xprt *transport =
1874                container_of(work, struct sock_xprt, connect_worker.work);
1875        struct rpc_xprt *xprt = &transport->xprt;
1876        struct socket *sock;
1877        int status = -EIO;
1878
1879        current->flags |= PF_FSTRANS;
1880
1881        clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1882        status = __sock_create(xprt->xprt_net, AF_LOCAL,
1883                                        SOCK_STREAM, 0, &sock, 1);
1884        if (status < 0) {
1885                dprintk("RPC:       can't create AF_LOCAL "
1886                        "transport socket (%d).\n", -status);
1887                goto out;
1888        }
1889        xs_reclassify_socketu(sock);
1890
1891        dprintk("RPC:       worker connecting xprt %p via AF_LOCAL to %s\n",
1892                        xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1893
1894        status = xs_local_finish_connecting(xprt, sock);
1895        switch (status) {
1896        case 0:
1897                dprintk("RPC:       xprt %p connected to %s\n",
1898                                xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1899                xprt_set_connected(xprt);
1900                break;
1901        case -ENOENT:
1902                dprintk("RPC:       xprt %p: socket %s does not exist\n",
1903                                xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1904                break;
1905        case -ECONNREFUSED:
1906                dprintk("RPC:       xprt %p: connection refused for %s\n",
1907                                xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1908                break;
1909        default:
1910                printk(KERN_ERR "%s: unhandled error (%d) connecting to %s\n",
1911                                __func__, -status,
1912                                xprt->address_strings[RPC_DISPLAY_ADDR]);
1913        }
1914
1915out:
1916        xprt_clear_connecting(xprt);
1917        xprt_wake_pending_tasks(xprt, status);
1918        current->flags &= ~PF_FSTRANS;
1919}
1920
1921#ifdef CONFIG_SUNRPC_SWAP
1922static void xs_set_memalloc(struct rpc_xprt *xprt)
1923{
1924        struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1925                        xprt);
1926
1927        if (xprt->swapper)
1928                sk_set_memalloc(transport->inet);
1929}
1930
1931/**
1932 * xs_swapper - Tag this transport as being used for swap.
1933 * @xprt: transport to tag
1934 * @enable: enable/disable
1935 *
1936 */
1937int xs_swapper(struct rpc_xprt *xprt, int enable)
1938{
1939        struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1940                        xprt);
1941        int err = 0;
1942
1943        if (enable) {
1944                xprt->swapper++;
1945                xs_set_memalloc(xprt);
1946        } else if (xprt->swapper) {
1947                xprt->swapper--;
1948                sk_clear_memalloc(transport->inet);
1949        }
1950
1951        return err;
1952}
1953EXPORT_SYMBOL_GPL(xs_swapper);
1954#else
1955static void xs_set_memalloc(struct rpc_xprt *xprt)
1956{
1957}
1958#endif
1959
1960static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
1961{
1962        struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1963
1964        if (!transport->inet) {
1965                struct sock *sk = sock->sk;
1966
1967                write_lock_bh(&sk->sk_callback_lock);
1968
1969                xs_save_old_callbacks(transport, sk);
1970
1971                sk->sk_user_data = xprt;
1972                sk->sk_data_ready = xs_udp_data_ready;
1973                sk->sk_write_space = xs_udp_write_space;
1974                sk->sk_no_check = UDP_CSUM_NORCV;
1975                sk->sk_allocation = GFP_ATOMIC;
1976
1977                xprt_set_connected(xprt);
1978
1979                /* Reset to new socket */
1980                transport->sock = sock;
1981                transport->inet = sk;
1982
1983                xs_set_memalloc(xprt);
1984
1985                write_unlock_bh(&sk->sk_callback_lock);
1986        }
1987        xs_udp_do_set_buffer_size(xprt);
1988}
1989
1990static void xs_udp_setup_socket(struct work_struct *work)
1991{
1992        struct sock_xprt *transport =
1993                container_of(work, struct sock_xprt, connect_worker.work);
1994        struct rpc_xprt *xprt = &transport->xprt;
1995        struct socket *sock = transport->sock;
1996        int status = -EIO;
1997
1998        current->flags |= PF_FSTRANS;
1999
2000        /* Start by resetting any existing state */
2001        xs_reset_transport(transport);
2002        sock = xs_create_sock(xprt, transport,
2003                        xs_addr(xprt)->sa_family, SOCK_DGRAM, IPPROTO_UDP);
2004        if (IS_ERR(sock))
2005                goto out;
2006
2007        dprintk("RPC:       worker connecting xprt %p via %s to "
2008                                "%s (port %s)\n", xprt,
2009                        xprt->address_strings[RPC_DISPLAY_PROTO],
2010                        xprt->address_strings[RPC_DISPLAY_ADDR],
2011                        xprt->address_strings[RPC_DISPLAY_PORT]);
2012
2013        xs_udp_finish_connecting(xprt, sock);
2014        status = 0;
2015out:
2016        xprt_clear_connecting(xprt);
2017        xprt_wake_pending_tasks(xprt, status);
2018        current->flags &= ~PF_FSTRANS;
2019}
2020
2021/*
2022 * We need to preserve the port number so the reply cache on the server can
2023 * find our cached RPC replies when we get around to reconnecting.
2024 */
2025static void xs_abort_connection(struct sock_xprt *transport)
2026{
2027        int result;
2028        struct sockaddr any;
2029
2030        dprintk("RPC:       disconnecting xprt %p to reuse port\n", transport);
2031
2032        /*
2033         * Disconnect the transport socket by doing a connect operation
2034         * with AF_UNSPEC.  This should return immediately...
2035         */
2036        memset(&any, 0, sizeof(any));
2037        any.sa_family = AF_UNSPEC;
2038        result = kernel_connect(transport->sock, &any, sizeof(any), 0);
2039        if (!result)
2040                xs_sock_reset_connection_flags(&transport->xprt);
2041        dprintk("RPC:       AF_UNSPEC connect return code %d\n", result);
2042}
2043
2044static void xs_tcp_reuse_connection(struct sock_xprt *transport)
2045{
2046        unsigned int state = transport->inet->sk_state;
2047
2048        if (state == TCP_CLOSE && transport->sock->state == SS_UNCONNECTED) {
2049                /* we don't need to abort the connection if the socket
2050                 * hasn't undergone a shutdown
2051                 */
2052                if (transport->inet->sk_shutdown == 0)
2053                        return;
2054                dprintk("RPC:       %s: TCP_CLOSEd and sk_shutdown set to %d\n",
2055                                __func__, transport->inet->sk_shutdown);
2056        }
2057        if ((1 << state) & (TCPF_ESTABLISHED|TCPF_SYN_SENT)) {
2058                /* we don't need to abort the connection if the socket
2059                 * hasn't undergone a shutdown
2060                 */
2061                if (transport->inet->sk_shutdown == 0)
2062                        return;
2063                dprintk("RPC:       %s: ESTABLISHED/SYN_SENT "
2064                                "sk_shutdown set to %d\n",
2065                                __func__, transport->inet->sk_shutdown);
2066        }
2067        xs_abort_connection(transport);
2068}
2069
2070static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2071{
2072        struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2073        int ret = -ENOTCONN;
2074
2075        if (!transport->inet) {
2076                struct sock *sk = sock->sk;
2077
2078                write_lock_bh(&sk->sk_callback_lock);
2079
2080                xs_save_old_callbacks(transport, sk);
2081
2082                sk->sk_user_data = xprt;
2083                sk->sk_data_ready = xs_tcp_data_ready;
2084                sk->sk_state_change = xs_tcp_state_change;
2085                sk->sk_write_space = xs_tcp_write_space;
2086                sk->sk_allocation = GFP_ATOMIC;
2087
2088                /* socket options */
2089                sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
2090                sock_reset_flag(sk, SOCK_LINGER);
2091                tcp_sk(sk)->linger2 = 0;
2092                tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
2093
2094                xprt_clear_connected(xprt);
2095
2096                /* Reset to new socket */
2097                transport->sock = sock;
2098                transport->inet = sk;
2099
2100                write_unlock_bh(&sk->sk_callback_lock);
2101        }
2102
2103        if (!xprt_bound(xprt))
2104                goto out;
2105
2106        xs_set_memalloc(xprt);
2107
2108        /* Tell the socket layer to start connecting... */
2109        xprt->stat.connect_count++;
2110        xprt->stat.connect_start = jiffies;
2111        ret = kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
2112        switch (ret) {
2113        case 0:
2114        case -EINPROGRESS:
2115                /* SYN_SENT! */
2116                xprt->connect_cookie++;
2117                if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2118                        xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2119        }
2120out:
2121        return ret;
2122}
2123
2124/**
2125 * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint
2126 * @xprt: RPC transport to connect
2127 * @transport: socket transport to connect
2128 * @create_sock: function to create a socket of the correct type
2129 *
2130 * Invoked by a work queue tasklet.
2131 */
2132static void xs_tcp_setup_socket(struct work_struct *work)
2133{
2134        struct sock_xprt *transport =
2135                container_of(work, struct sock_xprt, connect_worker.work);
2136        struct socket *sock = transport->sock;
2137        struct rpc_xprt *xprt = &transport->xprt;
2138        int status = -EIO;
2139
2140        current->flags |= PF_FSTRANS;
2141
2142        if (!sock) {
2143                clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
2144                sock = xs_create_sock(xprt, transport,
2145                                xs_addr(xprt)->sa_family, SOCK_STREAM, IPPROTO_TCP);
2146                if (IS_ERR(sock)) {
2147                        status = PTR_ERR(sock);
2148                        goto out;
2149                }
2150        } else {
2151                int abort_and_exit;
2152
2153                abort_and_exit = test_and_clear_bit(XPRT_CONNECTION_ABORT,
2154                                &xprt->state);
2155                /* "close" the socket, preserving the local port */
2156                xs_tcp_reuse_connection(transport);
2157
2158                if (abort_and_exit)
2159                        goto out_eagain;
2160        }
2161
2162        dprintk("RPC:       worker connecting xprt %p via %s to "
2163                                "%s (port %s)\n", xprt,
2164                        xprt->address_strings[RPC_DISPLAY_PROTO],
2165                        xprt->address_strings[RPC_DISPLAY_ADDR],
2166                        xprt->address_strings[RPC_DISPLAY_PORT]);
2167
2168        status = xs_tcp_finish_connecting(xprt, sock);
2169        dprintk("RPC:       %p connect status %d connected %d sock state %d\n",
2170                        xprt, -status, xprt_connected(xprt),
2171                        sock->sk->sk_state);
2172        switch (status) {
2173        default:
2174                printk("%s: connect returned unhandled error %d\n",
2175                        __func__, status);
2176        case -EADDRNOTAVAIL:
2177                /* We're probably in TIME_WAIT. Get rid of existing socket,
2178                 * and retry
2179                 */
2180                xs_tcp_force_close(xprt);
2181                break;
2182        case -ECONNREFUSED:
2183        case -ECONNRESET:
2184        case -ENETUNREACH:
2185                /* retry with existing socket, after a delay */
2186        case 0:
2187        case -EINPROGRESS:
2188        case -EALREADY:
2189                xprt_clear_connecting(xprt);
2190                current->flags &= ~PF_FSTRANS;
2191                return;
2192        case -EINVAL:
2193                /* Happens, for instance, if the user specified a link
2194                 * local IPv6 address without a scope-id.
2195                 */
2196                goto out;
2197        }
2198out_eagain:
2199        status = -EAGAIN;
2200out:
2201        xprt_clear_connecting(xprt);
2202        xprt_wake_pending_tasks(xprt, status);
2203        current->flags &= ~PF_FSTRANS;
2204}
2205
2206/**
2207 * xs_connect - connect a socket to a remote endpoint
2208 * @task: address of RPC task that manages state of connect request
2209 *
2210 * TCP: If the remote end dropped the connection, delay reconnecting.
2211 *
2212 * UDP socket connects are synchronous, but we use a work queue anyway
2213 * to guarantee that even unprivileged user processes can set up a
2214 * socket on a privileged port.
2215 *
2216 * If a UDP socket connect fails, the delay behavior here prevents
2217 * retry floods (hard mounts).
2218 */
2219static void xs_connect(struct rpc_task *task)
2220{
2221        struct rpc_xprt *xprt = task->tk_xprt;
2222        struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2223
2224        if (transport->sock != NULL && !RPC_IS_SOFTCONN(task)) {
2225                dprintk("RPC:       xs_connect delayed xprt %p for %lu "
2226                                "seconds\n",
2227                                xprt, xprt->reestablish_timeout / HZ);
2228                queue_delayed_work(rpciod_workqueue,
2229                                   &transport->connect_worker,
2230                                   xprt->reestablish_timeout);
2231                xprt->reestablish_timeout <<= 1;
2232                if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2233                        xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2234                if (xprt->reestablish_timeout > XS_TCP_MAX_REEST_TO)
2235                        xprt->reestablish_timeout = XS_TCP_MAX_REEST_TO;
2236        } else {
2237                dprintk("RPC:       xs_connect scheduled xprt %p\n", xprt);
2238                queue_delayed_work(rpciod_workqueue,
2239                                   &transport->connect_worker, 0);
2240        }
2241}
2242
2243/**
2244 * xs_local_print_stats - display AF_LOCAL socket-specifc stats
2245 * @xprt: rpc_xprt struct containing statistics
2246 * @seq: output file
2247 *
2248 */
2249static void xs_local_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2250{
2251        long idle_time = 0;
2252
2253        if (xprt_connected(xprt))
2254                idle_time = (long)(jiffies - xprt->last_used) / HZ;
2255
2256        seq_printf(seq, "\txprt:\tlocal %lu %lu %lu %ld %lu %lu %lu "
2257                        "%llu %llu %lu %llu %llu\n",
2258                        xprt->stat.bind_count,
2259                        xprt->stat.connect_count,
2260                        xprt->stat.connect_time,
2261                        idle_time,
2262                        xprt->stat.sends,
2263                        xprt->stat.recvs,
2264                        xprt->stat.bad_xids,
2265                        xprt->stat.req_u,
2266                        xprt->stat.bklog_u,
2267                        xprt->stat.max_slots,
2268                        xprt->stat.sending_u,
2269                        xprt->stat.pending_u);
2270}
2271
2272/**
2273 * xs_udp_print_stats - display UDP socket-specifc stats
2274 * @xprt: rpc_xprt struct containing statistics
2275 * @seq: output file
2276 *
2277 */
2278static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2279{
2280        struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2281
2282        seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %llu %llu "
2283                        "%lu %llu %llu\n",
2284                        transport->srcport,
2285                        xprt->stat.bind_count,
2286                        xprt->stat.sends,
2287                        xprt->stat.recvs,
2288                        xprt->stat.bad_xids,
2289                        xprt->stat.req_u,
2290                        xprt->stat.bklog_u,
2291                        xprt->stat.max_slots,
2292                        xprt->stat.sending_u,
2293                        xprt->stat.pending_u);
2294}
2295
2296/**
2297 * xs_tcp_print_stats - display TCP socket-specifc stats
2298 * @xprt: rpc_xprt struct containing statistics
2299 * @seq: output file
2300 *
2301 */
2302static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2303{
2304        struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2305        long idle_time = 0;
2306
2307        if (xprt_connected(xprt))
2308                idle_time = (long)(jiffies - xprt->last_used) / HZ;
2309
2310        seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu "
2311                        "%llu %llu %lu %llu %llu\n",
2312                        transport->srcport,
2313                        xprt->stat.bind_count,
2314                        xprt->stat.connect_count,
2315                        xprt->stat.connect_time,
2316                        idle_time,
2317                        xprt->stat.sends,
2318                        xprt->stat.recvs,
2319                        xprt->stat.bad_xids,
2320                        xprt->stat.req_u,
2321                        xprt->stat.bklog_u,
2322                        xprt->stat.max_slots,
2323                        xprt->stat.sending_u,
2324                        xprt->stat.pending_u);
2325}
2326
2327/*
2328 * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason
2329 * we allocate pages instead doing a kmalloc like rpc_malloc is because we want
2330 * to use the server side send routines.
2331 */
2332static void *bc_malloc(struct rpc_task *task, size_t size)
2333{
2334        struct page *page;
2335        struct rpc_buffer *buf;
2336
2337        WARN_ON_ONCE(size > PAGE_SIZE - sizeof(struct rpc_buffer));
2338        if (size > PAGE_SIZE - sizeof(struct rpc_buffer))
2339                return NULL;
2340
2341        page = alloc_page(GFP_KERNEL);
2342        if (!page)
2343                return NULL;
2344
2345        buf = page_address(page);
2346        buf->len = PAGE_SIZE;
2347
2348        return buf->data;
2349}
2350
2351/*
2352 * Free the space allocated in the bc_alloc routine
2353 */
2354static void bc_free(void *buffer)
2355{
2356        struct rpc_buffer *buf;
2357
2358        if (!buffer)
2359                return;
2360
2361        buf = container_of(buffer, struct rpc_buffer, data);
2362        free_page((unsigned long)buf);
2363}
2364
2365/*
2366 * Use the svc_sock to send the callback. Must be called with svsk->sk_mutex
2367 * held. Borrows heavily from svc_tcp_sendto and xs_tcp_send_request.
2368 */
2369static int bc_sendto(struct rpc_rqst *req)
2370{
2371        int len;
2372        struct xdr_buf *xbufp = &req->rq_snd_buf;
2373        struct rpc_xprt *xprt = req->rq_xprt;
2374        struct sock_xprt *transport =
2375                                container_of(xprt, struct sock_xprt, xprt);
2376        struct socket *sock = transport->sock;
2377        unsigned long headoff;
2378        unsigned long tailoff;
2379
2380        xs_encode_stream_record_marker(xbufp);
2381
2382        tailoff = (unsigned long)xbufp->tail[0].iov_base & ~PAGE_MASK;
2383        headoff = (unsigned long)xbufp->head[0].iov_base & ~PAGE_MASK;
2384        len = svc_send_common(sock, xbufp,
2385                              virt_to_page(xbufp->head[0].iov_base), headoff,
2386                              xbufp->tail[0].iov_base, tailoff);
2387
2388        if (len != xbufp->len) {
2389                printk(KERN_NOTICE "Error sending entire callback!\n");
2390                len = -EAGAIN;
2391        }
2392
2393        return len;
2394}
2395
2396/*
2397 * The send routine. Borrows from svc_send
2398 */
2399static int bc_send_request(struct rpc_task *task)
2400{
2401        struct rpc_rqst *req = task->tk_rqstp;
2402        struct svc_xprt *xprt;
2403        u32                     len;
2404
2405        dprintk("sending request with xid: %08x\n", ntohl(req->rq_xid));
2406        /*
2407         * Get the server socket associated with this callback xprt
2408         */
2409        xprt = req->rq_xprt->bc_xprt;
2410
2411        /*
2412         * Grab the mutex to serialize data as the connection is shared
2413         * with the fore channel
2414         */
2415        if (!mutex_trylock(&xprt->xpt_mutex)) {
2416                rpc_sleep_on(&xprt->xpt_bc_pending, task, NULL);
2417                if (!mutex_trylock(&xprt->xpt_mutex))
2418                        return -EAGAIN;
2419                rpc_wake_up_queued_task(&xprt->xpt_bc_pending, task);
2420        }
2421        if (test_bit(XPT_DEAD, &xprt->xpt_flags))
2422                len = -ENOTCONN;
2423        else
2424                len = bc_sendto(req);
2425        mutex_unlock(&xprt->xpt_mutex);
2426
2427        if (len > 0)
2428                len = 0;
2429
2430        return len;
2431}
2432
2433/*
2434 * The close routine. Since this is client initiated, we do nothing
2435 */
2436
2437static void bc_close(struct rpc_xprt *xprt)
2438{
2439}
2440
2441/*
2442 * The xprt destroy routine. Again, because this connection is client
2443 * initiated, we do nothing
2444 */
2445
2446static void bc_destroy(struct rpc_xprt *xprt)
2447{
2448}
2449
2450static struct rpc_xprt_ops xs_local_ops = {
2451        .reserve_xprt           = xprt_reserve_xprt,
2452        .release_xprt           = xs_tcp_release_xprt,
2453        .alloc_slot             = xprt_alloc_slot,
2454        .rpcbind                = xs_local_rpcbind,
2455        .set_port               = xs_local_set_port,
2456        .connect                = xs_connect,
2457        .buf_alloc              = rpc_malloc,
2458        .buf_free               = rpc_free,
2459        .send_request           = xs_local_send_request,
2460        .set_retrans_timeout    = xprt_set_retrans_timeout_def,
2461        .close                  = xs_close,
2462        .destroy                = xs_destroy,
2463        .print_stats            = xs_local_print_stats,
2464};
2465
2466static struct rpc_xprt_ops xs_udp_ops = {
2467        .set_buffer_size        = xs_udp_set_buffer_size,
2468        .reserve_xprt           = xprt_reserve_xprt_cong,
2469        .release_xprt           = xprt_release_xprt_cong,
2470        .alloc_slot             = xprt_alloc_slot,
2471        .rpcbind                = rpcb_getport_async,
2472        .set_port               = xs_set_port,
2473        .connect                = xs_connect,
2474        .buf_alloc              = rpc_malloc,
2475        .buf_free               = rpc_free,
2476        .send_request           = xs_udp_send_request,
2477        .set_retrans_timeout    = xprt_set_retrans_timeout_rtt,
2478        .timer                  = xs_udp_timer,
2479        .release_request        = xprt_release_rqst_cong,
2480        .close                  = xs_close,
2481        .destroy                = xs_destroy,
2482        .print_stats            = xs_udp_print_stats,
2483};
2484
2485static struct rpc_xprt_ops xs_tcp_ops = {
2486        .reserve_xprt           = xprt_reserve_xprt,
2487        .release_xprt           = xs_tcp_release_xprt,
2488        .alloc_slot             = xprt_lock_and_alloc_slot,
2489        .rpcbind                = rpcb_getport_async,
2490        .set_port               = xs_set_port,
2491        .connect                = xs_connect,
2492        .buf_alloc              = rpc_malloc,
2493        .buf_free               = rpc_free,
2494        .send_request           = xs_tcp_send_request,
2495        .set_retrans_timeout    = xprt_set_retrans_timeout_def,
2496        .close                  = xs_tcp_close,
2497        .destroy                = xs_destroy,
2498        .print_stats            = xs_tcp_print_stats,
2499};
2500
2501/*
2502 * The rpc_xprt_ops for the server backchannel
2503 */
2504
2505static struct rpc_xprt_ops bc_tcp_ops = {
2506        .reserve_xprt           = xprt_reserve_xprt,
2507        .release_xprt           = xprt_release_xprt,
2508        .alloc_slot             = xprt_alloc_slot,
2509        .rpcbind                = xs_local_rpcbind,
2510        .buf_alloc              = bc_malloc,
2511        .buf_free               = bc_free,
2512        .send_request           = bc_send_request,
2513        .set_retrans_timeout    = xprt_set_retrans_timeout_def,
2514        .close                  = bc_close,
2515        .destroy                = bc_destroy,
2516        .print_stats            = xs_tcp_print_stats,
2517};
2518
2519static int xs_init_anyaddr(const int family, struct sockaddr *sap)
2520{
2521        static const struct sockaddr_in sin = {
2522                .sin_family             = AF_INET,
2523                .sin_addr.s_addr        = htonl(INADDR_ANY),
2524        };
2525        static const struct sockaddr_in6 sin6 = {
2526                .sin6_family            = AF_INET6,
2527                .sin6_addr              = IN6ADDR_ANY_INIT,
2528        };
2529
2530        switch (family) {
2531        case AF_LOCAL:
2532                break;
2533        case AF_INET:
2534                memcpy(sap, &sin, sizeof(sin));
2535                break;
2536        case AF_INET6:
2537                memcpy(sap, &sin6, sizeof(sin6));
2538                break;
2539        default:
2540                dprintk("RPC:       %s: Bad address family\n", __func__);
2541                return -EAFNOSUPPORT;
2542        }
2543        return 0;
2544}
2545
2546static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
2547                                      unsigned int slot_table_size,
2548                                      unsigned int max_slot_table_size)
2549{
2550        struct rpc_xprt *xprt;
2551        struct sock_xprt *new;
2552
2553        if (args->addrlen > sizeof(xprt->addr)) {
2554                dprintk("RPC:       xs_setup_xprt: address too large\n");
2555                return ERR_PTR(-EBADF);
2556        }
2557
2558        xprt = xprt_alloc(args->net, sizeof(*new), slot_table_size,
2559                        max_slot_table_size);
2560        if (xprt == NULL) {
2561                dprintk("RPC:       xs_setup_xprt: couldn't allocate "
2562                                "rpc_xprt\n");
2563                return ERR_PTR(-ENOMEM);
2564        }
2565
2566        new = container_of(xprt, struct sock_xprt, xprt);
2567        memcpy(&xprt->addr, args->dstaddr, args->addrlen);
2568        xprt->addrlen = args->addrlen;
2569        if (args->srcaddr)
2570                memcpy(&new->srcaddr, args->srcaddr, args->addrlen);
2571        else {
2572                int err;
2573                err = xs_init_anyaddr(args->dstaddr->sa_family,
2574                                        (struct sockaddr *)&new->srcaddr);
2575                if (err != 0) {
2576                        xprt_free(xprt);
2577                        return ERR_PTR(err);
2578                }
2579        }
2580
2581        return xprt;
2582}
2583
2584static const struct rpc_timeout xs_local_default_timeout = {
2585        .to_initval = 10 * HZ,
2586        .to_maxval = 10 * HZ,
2587        .to_retries = 2,
2588};
2589
2590/**
2591 * xs_setup_local - Set up transport to use an AF_LOCAL socket
2592 * @args: rpc transport creation arguments
2593 *
2594 * AF_LOCAL is a "tpi_cots_ord" transport, just like TCP
2595 */
2596static struct rpc_xprt *xs_setup_local(struct xprt_create *args)
2597{
2598        struct sockaddr_un *sun = (struct sockaddr_un *)args->dstaddr;
2599        struct sock_xprt *transport;
2600        struct rpc_xprt *xprt;
2601        struct rpc_xprt *ret;
2602
2603        xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2604                        xprt_max_tcp_slot_table_entries);
2605        if (IS_ERR(xprt))
2606                return xprt;
2607        transport = container_of(xprt, struct sock_xprt, xprt);
2608
2609        xprt->prot = 0;
2610        xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2611        xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2612
2613        xprt->bind_timeout = XS_BIND_TO;
2614        xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2615        xprt->idle_timeout = XS_IDLE_DISC_TO;
2616
2617        xprt->ops = &xs_local_ops;
2618        xprt->timeout = &xs_local_default_timeout;
2619
2620        switch (sun->sun_family) {
2621        case AF_LOCAL:
2622                if (sun->sun_path[0] != '/') {
2623                        dprintk("RPC:       bad AF_LOCAL address: %s\n",
2624                                        sun->sun_path);
2625                        ret = ERR_PTR(-EINVAL);
2626                        goto out_err;
2627                }
2628                xprt_set_bound(xprt);
2629                INIT_DELAYED_WORK(&transport->connect_worker,
2630                                        xs_local_setup_socket);
2631                xs_format_peer_addresses(xprt, "local", RPCBIND_NETID_LOCAL);
2632                break;
2633        default:
2634                ret = ERR_PTR(-EAFNOSUPPORT);
2635                goto out_err;
2636        }
2637
2638        dprintk("RPC:       set up xprt to %s via AF_LOCAL\n",
2639                        xprt->address_strings[RPC_DISPLAY_ADDR]);
2640
2641        if (try_module_get(THIS_MODULE))
2642                return xprt;
2643        ret = ERR_PTR(-EINVAL);
2644out_err:
2645        xprt_free(xprt);
2646        return ret;
2647}
2648
2649static const struct rpc_timeout xs_udp_default_timeout = {
2650        .to_initval = 5 * HZ,
2651        .to_maxval = 30 * HZ,
2652        .to_increment = 5 * HZ,
2653        .to_retries = 5,
2654};
2655
2656/**
2657 * xs_setup_udp - Set up transport to use a UDP socket
2658 * @args: rpc transport creation arguments
2659 *
2660 */
2661static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
2662{
2663        struct sockaddr *addr = args->dstaddr;
2664        struct rpc_xprt *xprt;
2665        struct sock_xprt *transport;
2666        struct rpc_xprt *ret;
2667
2668        xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries,
2669                        xprt_udp_slot_table_entries);
2670        if (IS_ERR(xprt))
2671                return xprt;
2672        transport = container_of(xprt, struct sock_xprt, xprt);
2673
2674        xprt->prot = IPPROTO_UDP;
2675        xprt->tsh_size = 0;
2676        /* XXX: header size can vary due to auth type, IPv6, etc. */
2677        xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
2678
2679        xprt->bind_timeout = XS_BIND_TO;
2680        xprt->reestablish_timeout = XS_UDP_REEST_TO;
2681        xprt->idle_timeout = XS_IDLE_DISC_TO;
2682
2683        xprt->ops = &xs_udp_ops;
2684
2685        xprt->timeout = &xs_udp_default_timeout;
2686
2687        switch (addr->sa_family) {
2688        case AF_INET:
2689                if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2690                        xprt_set_bound(xprt);
2691
2692                INIT_DELAYED_WORK(&transport->connect_worker,
2693                                        xs_udp_setup_socket);
2694                xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
2695                break;
2696        case AF_INET6:
2697                if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2698                        xprt_set_bound(xprt);
2699
2700                INIT_DELAYED_WORK(&transport->connect_worker,
2701                                        xs_udp_setup_socket);
2702                xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
2703                break;
2704        default:
2705                ret = ERR_PTR(-EAFNOSUPPORT);
2706                goto out_err;
2707        }
2708
2709        if (xprt_bound(xprt))
2710                dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
2711                                xprt->address_strings[RPC_DISPLAY_ADDR],
2712                                xprt->address_strings[RPC_DISPLAY_PORT],
2713                                xprt->address_strings[RPC_DISPLAY_PROTO]);
2714        else
2715                dprintk("RPC:       set up xprt to %s (autobind) via %s\n",
2716                                xprt->address_strings[RPC_DISPLAY_ADDR],
2717                                xprt->address_strings[RPC_DISPLAY_PROTO]);
2718
2719        if (try_module_get(THIS_MODULE))
2720                return xprt;
2721        ret = ERR_PTR(-EINVAL);
2722out_err:
2723        xprt_free(xprt);
2724        return ret;
2725}
2726
2727static const struct rpc_timeout xs_tcp_default_timeout = {
2728        .to_initval = 60 * HZ,
2729        .to_maxval = 60 * HZ,
2730        .to_retries = 2,
2731};
2732
2733/**
2734 * xs_setup_tcp - Set up transport to use a TCP socket
2735 * @args: rpc transport creation arguments
2736 *
2737 */
2738static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
2739{
2740        struct sockaddr *addr = args->dstaddr;
2741        struct rpc_xprt *xprt;
2742        struct sock_xprt *transport;
2743        struct rpc_xprt *ret;
2744
2745        xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2746                        xprt_max_tcp_slot_table_entries);
2747        if (IS_ERR(xprt))
2748                return xprt;
2749        transport = container_of(xprt, struct sock_xprt, xprt);
2750
2751        xprt->prot = IPPROTO_TCP;
2752        xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2753        xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2754
2755        xprt->bind_timeout = XS_BIND_TO;
2756        xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2757        xprt->idle_timeout = XS_IDLE_DISC_TO;
2758
2759        xprt->ops = &xs_tcp_ops;
2760        xprt->timeout = &xs_tcp_default_timeout;
2761
2762        switch (addr->sa_family) {
2763        case AF_INET:
2764                if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2765                        xprt_set_bound(xprt);
2766
2767                INIT_DELAYED_WORK(&transport->connect_worker,
2768                                        xs_tcp_setup_socket);
2769                xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
2770                break;
2771        case AF_INET6:
2772                if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2773                        xprt_set_bound(xprt);
2774
2775                INIT_DELAYED_WORK(&transport->connect_worker,
2776                                        xs_tcp_setup_socket);
2777                xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
2778                break;
2779        default:
2780                ret = ERR_PTR(-EAFNOSUPPORT);
2781                goto out_err;
2782        }
2783
2784        if (xprt_bound(xprt))
2785                dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
2786                                xprt->address_strings[RPC_DISPLAY_ADDR],
2787                                xprt->address_strings[RPC_DISPLAY_PORT],
2788                                xprt->address_strings[RPC_DISPLAY_PROTO]);
2789        else
2790                dprintk("RPC:       set up xprt to %s (autobind) via %s\n",
2791                                xprt->address_strings[RPC_DISPLAY_ADDR],
2792                                xprt->address_strings[RPC_DISPLAY_PROTO]);
2793
2794
2795        if (try_module_get(THIS_MODULE))
2796                return xprt;
2797        ret = ERR_PTR(-EINVAL);
2798out_err:
2799        xprt_free(xprt);
2800        return ret;
2801}
2802
2803/**
2804 * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket
2805 * @args: rpc transport creation arguments
2806 *
2807 */
2808static struct rpc_xprt *xs_setup_bc_tcp(struct xprt_create *args)
2809{
2810        struct sockaddr *addr = args->dstaddr;
2811        struct rpc_xprt *xprt;
2812        struct sock_xprt *transport;
2813        struct svc_sock *bc_sock;
2814        struct rpc_xprt *ret;
2815
2816        if (args->bc_xprt->xpt_bc_xprt) {
2817                /*
2818                 * This server connection already has a backchannel
2819                 * export; we can't create a new one, as we wouldn't be
2820                 * able to match replies based on xid any more.  So,
2821                 * reuse the already-existing one:
2822                 */
2823                 return args->bc_xprt->xpt_bc_xprt;
2824        }
2825        xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2826                        xprt_tcp_slot_table_entries);
2827        if (IS_ERR(xprt))
2828                return xprt;
2829        transport = container_of(xprt, struct sock_xprt, xprt);
2830
2831        xprt->prot = IPPROTO_TCP;
2832        xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2833        xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2834        xprt->timeout = &xs_tcp_default_timeout;
2835
2836        /* backchannel */
2837        xprt_set_bound(xprt);
2838        xprt->bind_timeout = 0;
2839        xprt->reestablish_timeout = 0;
2840        xprt->idle_timeout = 0;
2841
2842        xprt->ops = &bc_tcp_ops;
2843
2844        switch (addr->sa_family) {
2845        case AF_INET:
2846                xs_format_peer_addresses(xprt, "tcp",
2847                                         RPCBIND_NETID_TCP);
2848                break;
2849        case AF_INET6:
2850                xs_format_peer_addresses(xprt, "tcp",
2851                                   RPCBIND_NETID_TCP6);
2852                break;
2853        default:
2854                ret = ERR_PTR(-EAFNOSUPPORT);
2855                goto out_err;
2856        }
2857
2858        dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
2859                        xprt->address_strings[RPC_DISPLAY_ADDR],
2860                        xprt->address_strings[RPC_DISPLAY_PORT],
2861                        xprt->address_strings[RPC_DISPLAY_PROTO]);
2862
2863        /*
2864         * Once we've associated a backchannel xprt with a connection,
2865         * we want to keep it around as long as long as the connection
2866         * lasts, in case we need to start using it for a backchannel
2867         * again; this reference won't be dropped until bc_xprt is
2868         * destroyed.
2869         */
2870        xprt_get(xprt);
2871        args->bc_xprt->xpt_bc_xprt = xprt;
2872        xprt->bc_xprt = args->bc_xprt;
2873        bc_sock = container_of(args->bc_xprt, struct svc_sock, sk_xprt);
2874        transport->sock = bc_sock->sk_sock;
2875        transport->inet = bc_sock->sk_sk;
2876
2877        /*
2878         * Since we don't want connections for the backchannel, we set
2879         * the xprt status to connected
2880         */
2881        xprt_set_connected(xprt);
2882
2883
2884        if (try_module_get(THIS_MODULE))
2885                return xprt;
2886        xprt_put(xprt);
2887        ret = ERR_PTR(-EINVAL);
2888out_err:
2889        xprt_free(xprt);
2890        return ret;
2891}
2892
2893static struct xprt_class        xs_local_transport = {
2894        .list           = LIST_HEAD_INIT(xs_local_transport.list),
2895        .name           = "named UNIX socket",
2896        .owner          = THIS_MODULE,
2897        .ident          = XPRT_TRANSPORT_LOCAL,
2898        .setup          = xs_setup_local,
2899};
2900
2901static struct xprt_class        xs_udp_transport = {
2902        .list           = LIST_HEAD_INIT(xs_udp_transport.list),
2903        .name           = "udp",
2904        .owner          = THIS_MODULE,
2905        .ident          = XPRT_TRANSPORT_UDP,
2906        .setup          = xs_setup_udp,
2907};
2908
2909static struct xprt_class        xs_tcp_transport = {
2910        .list           = LIST_HEAD_INIT(xs_tcp_transport.list),
2911        .name           = "tcp",
2912        .owner          = THIS_MODULE,
2913        .ident          = XPRT_TRANSPORT_TCP,
2914        .setup          = xs_setup_tcp,
2915};
2916
2917static struct xprt_class        xs_bc_tcp_transport = {
2918        .list           = LIST_HEAD_INIT(xs_bc_tcp_transport.list),
2919        .name           = "tcp NFSv4.1 backchannel",
2920        .owner          = THIS_MODULE,
2921        .ident          = XPRT_TRANSPORT_BC_TCP,
2922        .setup          = xs_setup_bc_tcp,
2923};
2924
2925/**
2926 * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
2927 *
2928 */
2929int init_socket_xprt(void)
2930{
2931#ifdef RPC_DEBUG
2932        if (!sunrpc_table_header)
2933                sunrpc_table_header = register_sysctl_table(sunrpc_table);
2934#endif
2935
2936        xprt_register_transport(&xs_local_transport);
2937        xprt_register_transport(&xs_udp_transport);
2938        xprt_register_transport(&xs_tcp_transport);
2939        xprt_register_transport(&xs_bc_tcp_transport);
2940
2941        return 0;
2942}
2943
2944/**
2945 * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
2946 *
2947 */
2948void cleanup_socket_xprt(void)
2949{
2950#ifdef RPC_DEBUG
2951        if (sunrpc_table_header) {
2952                unregister_sysctl_table(sunrpc_table_header);
2953                sunrpc_table_header = NULL;
2954        }
2955#endif
2956
2957        xprt_unregister_transport(&xs_local_transport);
2958        xprt_unregister_transport(&xs_udp_transport);
2959        xprt_unregister_transport(&xs_tcp_transport);
2960        xprt_unregister_transport(&xs_bc_tcp_transport);
2961}
2962
2963static int param_set_uint_minmax(const char *val,
2964                const struct kernel_param *kp,
2965                unsigned int min, unsigned int max)
2966{
2967        unsigned long num;
2968        int ret;
2969
2970        if (!val)
2971                return -EINVAL;
2972        ret = strict_strtoul(val, 0, &num);
2973        if (ret == -EINVAL || num < min || num > max)
2974                return -EINVAL;
2975        *((unsigned int *)kp->arg) = num;
2976        return 0;
2977}
2978
2979static int param_set_portnr(const char *val, const struct kernel_param *kp)
2980{
2981        return param_set_uint_minmax(val, kp,
2982                        RPC_MIN_RESVPORT,
2983                        RPC_MAX_RESVPORT);
2984}
2985
2986static struct kernel_param_ops param_ops_portnr = {
2987        .set = param_set_portnr,
2988        .get = param_get_uint,
2989};
2990
2991#define param_check_portnr(name, p) \
2992        __param_check(name, p, unsigned int);
2993
2994module_param_named(min_resvport, xprt_min_resvport, portnr, 0644);
2995module_param_named(max_resvport, xprt_max_resvport, portnr, 0644);
2996
2997static int param_set_slot_table_size(const char *val,
2998                                     const struct kernel_param *kp)
2999{
3000        return param_set_uint_minmax(val, kp,
3001                        RPC_MIN_SLOT_TABLE,
3002                        RPC_MAX_SLOT_TABLE);
3003}
3004
3005static struct kernel_param_ops param_ops_slot_table_size = {
3006        .set = param_set_slot_table_size,
3007        .get = param_get_uint,
3008};
3009
3010#define param_check_slot_table_size(name, p) \
3011        __param_check(name, p, unsigned int);
3012
3013static int param_set_max_slot_table_size(const char *val,
3014                                     const struct kernel_param *kp)
3015{
3016        return param_set_uint_minmax(val, kp,
3017                        RPC_MIN_SLOT_TABLE,
3018                        RPC_MAX_SLOT_TABLE_LIMIT);
3019}
3020
3021static struct kernel_param_ops param_ops_max_slot_table_size = {
3022        .set = param_set_max_slot_table_size,
3023        .get = param_get_uint,
3024};
3025
3026#define param_check_max_slot_table_size(name, p) \
3027        __param_check(name, p, unsigned int);
3028
3029module_param_named(tcp_slot_table_entries, xprt_tcp_slot_table_entries,
3030                   slot_table_size, 0644);
3031module_param_named(tcp_max_slot_table_entries, xprt_max_tcp_slot_table_entries,
3032                   max_slot_table_size, 0644);
3033module_param_named(udp_slot_table_entries, xprt_udp_slot_table_entries,
3034                   slot_table_size, 0644);
3035
3036