linux/net/sunrpc/svcsock.c
<<
>>
Prefs
   1/*
   2 * linux/net/sunrpc/svcsock.c
   3 *
   4 * These are the RPC server socket internals.
   5 *
   6 * The server scheduling algorithm does not always distribute the load
   7 * evenly when servicing a single client. May need to modify the
   8 * svc_xprt_enqueue procedure...
   9 *
  10 * TCP support is largely untested and may be a little slow. The problem
  11 * is that we currently do two separate recvfrom's, one for the 4-byte
  12 * record length, and the second for the actual record. This could possibly
  13 * be improved by always reading a minimum size of around 100 bytes and
  14 * tucking any superfluous bytes away in a temporary store. Still, that
  15 * leaves write requests out in the rain. An alternative may be to peek at
  16 * the first skb in the queue, and if it matches the next TCP sequence
  17 * number, to extract the record marker. Yuck.
  18 *
  19 * Copyright (C) 1995, 1996 Olaf Kirch <okir@monad.swb.de>
  20 */
  21
  22#include <linux/kernel.h>
  23#include <linux/sched.h>
  24#include <linux/errno.h>
  25#include <linux/fcntl.h>
  26#include <linux/net.h>
  27#include <linux/in.h>
  28#include <linux/inet.h>
  29#include <linux/udp.h>
  30#include <linux/tcp.h>
  31#include <linux/unistd.h>
  32#include <linux/slab.h>
  33#include <linux/netdevice.h>
  34#include <linux/skbuff.h>
  35#include <linux/file.h>
  36#include <linux/freezer.h>
  37#include <net/sock.h>
  38#include <net/checksum.h>
  39#include <net/ip.h>
  40#include <net/ipv6.h>
  41#include <net/tcp.h>
  42#include <net/tcp_states.h>
  43#include <asm/uaccess.h>
  44#include <asm/ioctls.h>
  45
  46#include <linux/sunrpc/types.h>
  47#include <linux/sunrpc/clnt.h>
  48#include <linux/sunrpc/xdr.h>
  49#include <linux/sunrpc/msg_prot.h>
  50#include <linux/sunrpc/svcsock.h>
  51#include <linux/sunrpc/stats.h>
  52#include <linux/sunrpc/xprt.h>
  53
  54#define RPCDBG_FACILITY RPCDBG_SVCXPRT
  55
  56
  57static struct svc_sock *svc_setup_socket(struct svc_serv *, struct socket *,
  58                                         int *errp, int flags);
  59static void             svc_udp_data_ready(struct sock *, int);
  60static int              svc_udp_recvfrom(struct svc_rqst *);
  61static int              svc_udp_sendto(struct svc_rqst *);
  62static void             svc_sock_detach(struct svc_xprt *);
  63static void             svc_tcp_sock_detach(struct svc_xprt *);
  64static void             svc_sock_free(struct svc_xprt *);
  65
  66static struct svc_xprt *svc_create_socket(struct svc_serv *, int,
  67                                          struct sockaddr *, int, int);
  68#ifdef CONFIG_DEBUG_LOCK_ALLOC
  69static struct lock_class_key svc_key[2];
  70static struct lock_class_key svc_slock_key[2];
  71
  72static void svc_reclassify_socket(struct socket *sock)
  73{
  74        struct sock *sk = sock->sk;
  75        BUG_ON(sock_owned_by_user(sk));
  76        switch (sk->sk_family) {
  77        case AF_INET:
  78                sock_lock_init_class_and_name(sk, "slock-AF_INET-NFSD",
  79                                              &svc_slock_key[0],
  80                                              "sk_xprt.xpt_lock-AF_INET-NFSD",
  81                                              &svc_key[0]);
  82                break;
  83
  84        case AF_INET6:
  85                sock_lock_init_class_and_name(sk, "slock-AF_INET6-NFSD",
  86                                              &svc_slock_key[1],
  87                                              "sk_xprt.xpt_lock-AF_INET6-NFSD",
  88                                              &svc_key[1]);
  89                break;
  90
  91        default:
  92                BUG();
  93        }
  94}
  95#else
  96static void svc_reclassify_socket(struct socket *sock)
  97{
  98}
  99#endif
 100
 101/*
 102 * Release an skbuff after use
 103 */
 104static void svc_release_skb(struct svc_rqst *rqstp)
 105{
 106        struct sk_buff *skb = rqstp->rq_xprt_ctxt;
 107
 108        if (skb) {
 109                struct svc_sock *svsk =
 110                        container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
 111                rqstp->rq_xprt_ctxt = NULL;
 112
 113                dprintk("svc: service %p, releasing skb %p\n", rqstp, skb);
 114                skb_free_datagram_locked(svsk->sk_sk, skb);
 115        }
 116}
 117
 118union svc_pktinfo_u {
 119        struct in_pktinfo pkti;
 120        struct in6_pktinfo pkti6;
 121};
 122#define SVC_PKTINFO_SPACE \
 123        CMSG_SPACE(sizeof(union svc_pktinfo_u))
 124
 125static void svc_set_cmsg_data(struct svc_rqst *rqstp, struct cmsghdr *cmh)
 126{
 127        struct svc_sock *svsk =
 128                container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
 129        switch (svsk->sk_sk->sk_family) {
 130        case AF_INET: {
 131                        struct in_pktinfo *pki = CMSG_DATA(cmh);
 132
 133                        cmh->cmsg_level = SOL_IP;
 134                        cmh->cmsg_type = IP_PKTINFO;
 135                        pki->ipi_ifindex = 0;
 136                        pki->ipi_spec_dst.s_addr = rqstp->rq_daddr.addr.s_addr;
 137                        cmh->cmsg_len = CMSG_LEN(sizeof(*pki));
 138                }
 139                break;
 140
 141        case AF_INET6: {
 142                        struct in6_pktinfo *pki = CMSG_DATA(cmh);
 143
 144                        cmh->cmsg_level = SOL_IPV6;
 145                        cmh->cmsg_type = IPV6_PKTINFO;
 146                        pki->ipi6_ifindex = 0;
 147                        ipv6_addr_copy(&pki->ipi6_addr,
 148                                        &rqstp->rq_daddr.addr6);
 149                        cmh->cmsg_len = CMSG_LEN(sizeof(*pki));
 150                }
 151                break;
 152        }
 153}
 154
 155/*
 156 * send routine intended to be shared by the fore- and back-channel
 157 */
 158int svc_send_common(struct socket *sock, struct xdr_buf *xdr,
 159                    struct page *headpage, unsigned long headoffset,
 160                    struct page *tailpage, unsigned long tailoffset)
 161{
 162        int             result;
 163        int             size;
 164        struct page     **ppage = xdr->pages;
 165        size_t          base = xdr->page_base;
 166        unsigned int    pglen = xdr->page_len;
 167        unsigned int    flags = MSG_MORE;
 168        int             slen;
 169        int             len = 0;
 170
 171        slen = xdr->len;
 172
 173        /* send head */
 174        if (slen == xdr->head[0].iov_len)
 175                flags = 0;
 176        len = kernel_sendpage(sock, headpage, headoffset,
 177                                  xdr->head[0].iov_len, flags);
 178        if (len != xdr->head[0].iov_len)
 179                goto out;
 180        slen -= xdr->head[0].iov_len;
 181        if (slen == 0)
 182                goto out;
 183
 184        /* send page data */
 185        size = PAGE_SIZE - base < pglen ? PAGE_SIZE - base : pglen;
 186        while (pglen > 0) {
 187                if (slen == size)
 188                        flags = 0;
 189                result = kernel_sendpage(sock, *ppage, base, size, flags);
 190                if (result > 0)
 191                        len += result;
 192                if (result != size)
 193                        goto out;
 194                slen -= size;
 195                pglen -= size;
 196                size = PAGE_SIZE < pglen ? PAGE_SIZE : pglen;
 197                base = 0;
 198                ppage++;
 199        }
 200
 201        /* send tail */
 202        if (xdr->tail[0].iov_len) {
 203                result = kernel_sendpage(sock, tailpage, tailoffset,
 204                                   xdr->tail[0].iov_len, 0);
 205                if (result > 0)
 206                        len += result;
 207        }
 208
 209out:
 210        return len;
 211}
 212
 213
 214/*
 215 * Generic sendto routine
 216 */
 217static int svc_sendto(struct svc_rqst *rqstp, struct xdr_buf *xdr)
 218{
 219        struct svc_sock *svsk =
 220                container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
 221        struct socket   *sock = svsk->sk_sock;
 222        union {
 223                struct cmsghdr  hdr;
 224                long            all[SVC_PKTINFO_SPACE / sizeof(long)];
 225        } buffer;
 226        struct cmsghdr *cmh = &buffer.hdr;
 227        int             len = 0;
 228        unsigned long tailoff;
 229        unsigned long headoff;
 230        RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
 231
 232        if (rqstp->rq_prot == IPPROTO_UDP) {
 233                struct msghdr msg = {
 234                        .msg_name       = &rqstp->rq_addr,
 235                        .msg_namelen    = rqstp->rq_addrlen,
 236                        .msg_control    = cmh,
 237                        .msg_controllen = sizeof(buffer),
 238                        .msg_flags      = MSG_MORE,
 239                };
 240
 241                svc_set_cmsg_data(rqstp, cmh);
 242
 243                if (sock_sendmsg(sock, &msg, 0) < 0)
 244                        goto out;
 245        }
 246
 247        tailoff = ((unsigned long)xdr->tail[0].iov_base) & (PAGE_SIZE-1);
 248        headoff = 0;
 249        len = svc_send_common(sock, xdr, rqstp->rq_respages[0], headoff,
 250                               rqstp->rq_respages[0], tailoff);
 251
 252out:
 253        dprintk("svc: socket %p sendto([%p %Zu... ], %d) = %d (addr %s)\n",
 254                svsk, xdr->head[0].iov_base, xdr->head[0].iov_len,
 255                xdr->len, len, svc_print_addr(rqstp, buf, sizeof(buf)));
 256
 257        return len;
 258}
 259
 260/*
 261 * Report socket names for nfsdfs
 262 */
 263static int svc_one_sock_name(struct svc_sock *svsk, char *buf, int remaining)
 264{
 265        const struct sock *sk = svsk->sk_sk;
 266        const char *proto_name = sk->sk_protocol == IPPROTO_UDP ?
 267                                                        "udp" : "tcp";
 268        int len;
 269
 270        switch (sk->sk_family) {
 271        case PF_INET:
 272                len = snprintf(buf, remaining, "ipv4 %s %pI4 %d\n",
 273                                proto_name,
 274                                &inet_sk(sk)->inet_rcv_saddr,
 275                                inet_sk(sk)->inet_num);
 276                break;
 277        case PF_INET6:
 278                len = snprintf(buf, remaining, "ipv6 %s %pI6 %d\n",
 279                                proto_name,
 280                                &inet6_sk(sk)->rcv_saddr,
 281                                inet_sk(sk)->inet_num);
 282                break;
 283        default:
 284                len = snprintf(buf, remaining, "*unknown-%d*\n",
 285                                sk->sk_family);
 286        }
 287
 288        if (len >= remaining) {
 289                *buf = '\0';
 290                return -ENAMETOOLONG;
 291        }
 292        return len;
 293}
 294
 295/**
 296 * svc_sock_names - construct a list of listener names in a string
 297 * @serv: pointer to RPC service
 298 * @buf: pointer to a buffer to fill in with socket names
 299 * @buflen: size of the buffer to be filled
 300 * @toclose: pointer to '\0'-terminated C string containing the name
 301 *              of a listener to be closed
 302 *
 303 * Fills in @buf with a '\n'-separated list of names of listener
 304 * sockets.  If @toclose is not NULL, the socket named by @toclose
 305 * is closed, and is not included in the output list.
 306 *
 307 * Returns positive length of the socket name string, or a negative
 308 * errno value on error.
 309 */
 310int svc_sock_names(struct svc_serv *serv, char *buf, const size_t buflen,
 311                   const char *toclose)
 312{
 313        struct svc_sock *svsk, *closesk = NULL;
 314        int len = 0;
 315
 316        if (!serv)
 317                return 0;
 318
 319        spin_lock_bh(&serv->sv_lock);
 320        list_for_each_entry(svsk, &serv->sv_permsocks, sk_xprt.xpt_list) {
 321                int onelen = svc_one_sock_name(svsk, buf + len, buflen - len);
 322                if (onelen < 0) {
 323                        len = onelen;
 324                        break;
 325                }
 326                if (toclose && strcmp(toclose, buf + len) == 0)
 327                        closesk = svsk;
 328                else
 329                        len += onelen;
 330        }
 331        spin_unlock_bh(&serv->sv_lock);
 332
 333        if (closesk)
 334                /* Should unregister with portmap, but you cannot
 335                 * unregister just one protocol...
 336                 */
 337                svc_close_xprt(&closesk->sk_xprt);
 338        else if (toclose)
 339                return -ENOENT;
 340        return len;
 341}
 342EXPORT_SYMBOL_GPL(svc_sock_names);
 343
 344/*
 345 * Check input queue length
 346 */
 347static int svc_recv_available(struct svc_sock *svsk)
 348{
 349        struct socket   *sock = svsk->sk_sock;
 350        int             avail, err;
 351
 352        err = kernel_sock_ioctl(sock, TIOCINQ, (unsigned long) &avail);
 353
 354        return (err >= 0)? avail : err;
 355}
 356
 357/*
 358 * Generic recvfrom routine.
 359 */
 360static int svc_recvfrom(struct svc_rqst *rqstp, struct kvec *iov, int nr,
 361                        int buflen)
 362{
 363        struct svc_sock *svsk =
 364                container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
 365        struct msghdr msg = {
 366                .msg_flags      = MSG_DONTWAIT,
 367        };
 368        int len;
 369
 370        rqstp->rq_xprt_hlen = 0;
 371
 372        len = kernel_recvmsg(svsk->sk_sock, &msg, iov, nr, buflen,
 373                                msg.msg_flags);
 374
 375        dprintk("svc: socket %p recvfrom(%p, %Zu) = %d\n",
 376                svsk, iov[0].iov_base, iov[0].iov_len, len);
 377        return len;
 378}
 379
 380/*
 381 * Set socket snd and rcv buffer lengths
 382 */
 383static void svc_sock_setbufsize(struct socket *sock, unsigned int snd,
 384                                unsigned int rcv)
 385{
 386#if 0
 387        mm_segment_t    oldfs;
 388        oldfs = get_fs(); set_fs(KERNEL_DS);
 389        sock_setsockopt(sock, SOL_SOCKET, SO_SNDBUF,
 390                        (char*)&snd, sizeof(snd));
 391        sock_setsockopt(sock, SOL_SOCKET, SO_RCVBUF,
 392                        (char*)&rcv, sizeof(rcv));
 393#else
 394        /* sock_setsockopt limits use to sysctl_?mem_max,
 395         * which isn't acceptable.  Until that is made conditional
 396         * on not having CAP_SYS_RESOURCE or similar, we go direct...
 397         * DaveM said I could!
 398         */
 399        lock_sock(sock->sk);
 400        sock->sk->sk_sndbuf = snd * 2;
 401        sock->sk->sk_rcvbuf = rcv * 2;
 402        sock->sk->sk_userlocks |= SOCK_SNDBUF_LOCK|SOCK_RCVBUF_LOCK;
 403        sock->sk->sk_write_space(sock->sk);
 404        release_sock(sock->sk);
 405#endif
 406}
 407/*
 408 * INET callback when data has been received on the socket.
 409 */
 410static void svc_udp_data_ready(struct sock *sk, int count)
 411{
 412        struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
 413
 414        if (svsk) {
 415                dprintk("svc: socket %p(inet %p), count=%d, busy=%d\n",
 416                        svsk, sk, count,
 417                        test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags));
 418                set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
 419                svc_xprt_enqueue(&svsk->sk_xprt);
 420        }
 421        if (sk_sleep(sk) && waitqueue_active(sk_sleep(sk)))
 422                wake_up_interruptible(sk_sleep(sk));
 423}
 424
 425/*
 426 * INET callback when space is newly available on the socket.
 427 */
 428static void svc_write_space(struct sock *sk)
 429{
 430        struct svc_sock *svsk = (struct svc_sock *)(sk->sk_user_data);
 431
 432        if (svsk) {
 433                dprintk("svc: socket %p(inet %p), write_space busy=%d\n",
 434                        svsk, sk, test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags));
 435                svc_xprt_enqueue(&svsk->sk_xprt);
 436        }
 437
 438        if (sk_sleep(sk) && waitqueue_active(sk_sleep(sk))) {
 439                dprintk("RPC svc_write_space: someone sleeping on %p\n",
 440                       svsk);
 441                wake_up_interruptible(sk_sleep(sk));
 442        }
 443}
 444
 445static void svc_tcp_write_space(struct sock *sk)
 446{
 447        struct socket *sock = sk->sk_socket;
 448
 449        if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk) && sock)
 450                clear_bit(SOCK_NOSPACE, &sock->flags);
 451        svc_write_space(sk);
 452}
 453
 454/*
 455 * See net/ipv6/ip_sockglue.c : ip_cmsg_recv_pktinfo
 456 */
 457static int svc_udp_get_dest_address4(struct svc_rqst *rqstp,
 458                                     struct cmsghdr *cmh)
 459{
 460        struct in_pktinfo *pki = CMSG_DATA(cmh);
 461        if (cmh->cmsg_type != IP_PKTINFO)
 462                return 0;
 463        rqstp->rq_daddr.addr.s_addr = pki->ipi_spec_dst.s_addr;
 464        return 1;
 465}
 466
 467/*
 468 * See net/ipv6/datagram.c : datagram_recv_ctl
 469 */
 470static int svc_udp_get_dest_address6(struct svc_rqst *rqstp,
 471                                     struct cmsghdr *cmh)
 472{
 473        struct in6_pktinfo *pki = CMSG_DATA(cmh);
 474        if (cmh->cmsg_type != IPV6_PKTINFO)
 475                return 0;
 476        ipv6_addr_copy(&rqstp->rq_daddr.addr6, &pki->ipi6_addr);
 477        return 1;
 478}
 479
 480/*
 481 * Copy the UDP datagram's destination address to the rqstp structure.
 482 * The 'destination' address in this case is the address to which the
 483 * peer sent the datagram, i.e. our local address. For multihomed
 484 * hosts, this can change from msg to msg. Note that only the IP
 485 * address changes, the port number should remain the same.
 486 */
 487static int svc_udp_get_dest_address(struct svc_rqst *rqstp,
 488                                    struct cmsghdr *cmh)
 489{
 490        switch (cmh->cmsg_level) {
 491        case SOL_IP:
 492                return svc_udp_get_dest_address4(rqstp, cmh);
 493        case SOL_IPV6:
 494                return svc_udp_get_dest_address6(rqstp, cmh);
 495        }
 496
 497        return 0;
 498}
 499
 500/*
 501 * Receive a datagram from a UDP socket.
 502 */
 503static int svc_udp_recvfrom(struct svc_rqst *rqstp)
 504{
 505        struct svc_sock *svsk =
 506                container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
 507        struct svc_serv *serv = svsk->sk_xprt.xpt_server;
 508        struct sk_buff  *skb;
 509        union {
 510                struct cmsghdr  hdr;
 511                long            all[SVC_PKTINFO_SPACE / sizeof(long)];
 512        } buffer;
 513        struct cmsghdr *cmh = &buffer.hdr;
 514        struct msghdr msg = {
 515                .msg_name = svc_addr(rqstp),
 516                .msg_control = cmh,
 517                .msg_controllen = sizeof(buffer),
 518                .msg_flags = MSG_DONTWAIT,
 519        };
 520        size_t len;
 521        int err;
 522
 523        if (test_and_clear_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags))
 524            /* udp sockets need large rcvbuf as all pending
 525             * requests are still in that buffer.  sndbuf must
 526             * also be large enough that there is enough space
 527             * for one reply per thread.  We count all threads
 528             * rather than threads in a particular pool, which
 529             * provides an upper bound on the number of threads
 530             * which will access the socket.
 531             */
 532            svc_sock_setbufsize(svsk->sk_sock,
 533                                (serv->sv_nrthreads+3) * serv->sv_max_mesg,
 534                                (serv->sv_nrthreads+3) * serv->sv_max_mesg);
 535
 536        clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
 537        skb = NULL;
 538        err = kernel_recvmsg(svsk->sk_sock, &msg, NULL,
 539                             0, 0, MSG_PEEK | MSG_DONTWAIT);
 540        if (err >= 0)
 541                skb = skb_recv_datagram(svsk->sk_sk, 0, 1, &err);
 542
 543        if (skb == NULL) {
 544                if (err != -EAGAIN) {
 545                        /* possibly an icmp error */
 546                        dprintk("svc: recvfrom returned error %d\n", -err);
 547                        set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
 548                }
 549                return -EAGAIN;
 550        }
 551        len = svc_addr_len(svc_addr(rqstp));
 552        if (len == 0)
 553                return -EAFNOSUPPORT;
 554        rqstp->rq_addrlen = len;
 555        if (skb->tstamp.tv64 == 0) {
 556                skb->tstamp = ktime_get_real();
 557                /* Don't enable netstamp, sunrpc doesn't
 558                   need that much accuracy */
 559        }
 560        svsk->sk_sk->sk_stamp = skb->tstamp;
 561        set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); /* there may be more data... */
 562
 563        len  = skb->len - sizeof(struct udphdr);
 564        rqstp->rq_arg.len = len;
 565
 566        rqstp->rq_prot = IPPROTO_UDP;
 567
 568        if (!svc_udp_get_dest_address(rqstp, cmh)) {
 569                if (net_ratelimit())
 570                        printk(KERN_WARNING
 571                                "svc: received unknown control message %d/%d; "
 572                                "dropping RPC reply datagram\n",
 573                                        cmh->cmsg_level, cmh->cmsg_type);
 574                skb_free_datagram_locked(svsk->sk_sk, skb);
 575                return 0;
 576        }
 577
 578        if (skb_is_nonlinear(skb)) {
 579                /* we have to copy */
 580                local_bh_disable();
 581                if (csum_partial_copy_to_xdr(&rqstp->rq_arg, skb)) {
 582                        local_bh_enable();
 583                        /* checksum error */
 584                        skb_free_datagram_locked(svsk->sk_sk, skb);
 585                        return 0;
 586                }
 587                local_bh_enable();
 588                skb_free_datagram_locked(svsk->sk_sk, skb);
 589        } else {
 590                /* we can use it in-place */
 591                rqstp->rq_arg.head[0].iov_base = skb->data +
 592                        sizeof(struct udphdr);
 593                rqstp->rq_arg.head[0].iov_len = len;
 594                if (skb_checksum_complete(skb)) {
 595                        skb_free_datagram_locked(svsk->sk_sk, skb);
 596                        return 0;
 597                }
 598                rqstp->rq_xprt_ctxt = skb;
 599        }
 600
 601        rqstp->rq_arg.page_base = 0;
 602        if (len <= rqstp->rq_arg.head[0].iov_len) {
 603                rqstp->rq_arg.head[0].iov_len = len;
 604                rqstp->rq_arg.page_len = 0;
 605                rqstp->rq_respages = rqstp->rq_pages+1;
 606        } else {
 607                rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len;
 608                rqstp->rq_respages = rqstp->rq_pages + 1 +
 609                        DIV_ROUND_UP(rqstp->rq_arg.page_len, PAGE_SIZE);
 610        }
 611
 612        if (serv->sv_stats)
 613                serv->sv_stats->netudpcnt++;
 614
 615        return len;
 616}
 617
 618static int
 619svc_udp_sendto(struct svc_rqst *rqstp)
 620{
 621        int             error;
 622
 623        error = svc_sendto(rqstp, &rqstp->rq_res);
 624        if (error == -ECONNREFUSED)
 625                /* ICMP error on earlier request. */
 626                error = svc_sendto(rqstp, &rqstp->rq_res);
 627
 628        return error;
 629}
 630
 631static void svc_udp_prep_reply_hdr(struct svc_rqst *rqstp)
 632{
 633}
 634
 635static int svc_udp_has_wspace(struct svc_xprt *xprt)
 636{
 637        struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
 638        struct svc_serv *serv = xprt->xpt_server;
 639        unsigned long required;
 640
 641        /*
 642         * Set the SOCK_NOSPACE flag before checking the available
 643         * sock space.
 644         */
 645        set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
 646        required = atomic_read(&svsk->sk_xprt.xpt_reserved) + serv->sv_max_mesg;
 647        if (required*2 > sock_wspace(svsk->sk_sk))
 648                return 0;
 649        clear_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
 650        return 1;
 651}
 652
 653static struct svc_xprt *svc_udp_accept(struct svc_xprt *xprt)
 654{
 655        BUG();
 656        return NULL;
 657}
 658
 659static struct svc_xprt *svc_udp_create(struct svc_serv *serv,
 660                                       struct sockaddr *sa, int salen,
 661                                       int flags)
 662{
 663        return svc_create_socket(serv, IPPROTO_UDP, sa, salen, flags);
 664}
 665
 666static struct svc_xprt_ops svc_udp_ops = {
 667        .xpo_create = svc_udp_create,
 668        .xpo_recvfrom = svc_udp_recvfrom,
 669        .xpo_sendto = svc_udp_sendto,
 670        .xpo_release_rqst = svc_release_skb,
 671        .xpo_detach = svc_sock_detach,
 672        .xpo_free = svc_sock_free,
 673        .xpo_prep_reply_hdr = svc_udp_prep_reply_hdr,
 674        .xpo_has_wspace = svc_udp_has_wspace,
 675        .xpo_accept = svc_udp_accept,
 676};
 677
 678static struct svc_xprt_class svc_udp_class = {
 679        .xcl_name = "udp",
 680        .xcl_owner = THIS_MODULE,
 681        .xcl_ops = &svc_udp_ops,
 682        .xcl_max_payload = RPCSVC_MAXPAYLOAD_UDP,
 683};
 684
 685static void svc_udp_init(struct svc_sock *svsk, struct svc_serv *serv)
 686{
 687        int err, level, optname, one = 1;
 688
 689        svc_xprt_init(&svc_udp_class, &svsk->sk_xprt, serv);
 690        clear_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
 691        svsk->sk_sk->sk_data_ready = svc_udp_data_ready;
 692        svsk->sk_sk->sk_write_space = svc_write_space;
 693
 694        /* initialise setting must have enough space to
 695         * receive and respond to one request.
 696         * svc_udp_recvfrom will re-adjust if necessary
 697         */
 698        svc_sock_setbufsize(svsk->sk_sock,
 699                            3 * svsk->sk_xprt.xpt_server->sv_max_mesg,
 700                            3 * svsk->sk_xprt.xpt_server->sv_max_mesg);
 701
 702        /* data might have come in before data_ready set up */
 703        set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
 704        set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
 705
 706        /* make sure we get destination address info */
 707        switch (svsk->sk_sk->sk_family) {
 708        case AF_INET:
 709                level = SOL_IP;
 710                optname = IP_PKTINFO;
 711                break;
 712        case AF_INET6:
 713                level = SOL_IPV6;
 714                optname = IPV6_RECVPKTINFO;
 715                break;
 716        default:
 717                BUG();
 718        }
 719        err = kernel_setsockopt(svsk->sk_sock, level, optname,
 720                                        (char *)&one, sizeof(one));
 721        dprintk("svc: kernel_setsockopt returned %d\n", err);
 722}
 723
 724/*
 725 * A data_ready event on a listening socket means there's a connection
 726 * pending. Do not use state_change as a substitute for it.
 727 */
 728static void svc_tcp_listen_data_ready(struct sock *sk, int count_unused)
 729{
 730        struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
 731
 732        dprintk("svc: socket %p TCP (listen) state change %d\n",
 733                sk, sk->sk_state);
 734
 735        /*
 736         * This callback may called twice when a new connection
 737         * is established as a child socket inherits everything
 738         * from a parent LISTEN socket.
 739         * 1) data_ready method of the parent socket will be called
 740         *    when one of child sockets become ESTABLISHED.
 741         * 2) data_ready method of the child socket may be called
 742         *    when it receives data before the socket is accepted.
 743         * In case of 2, we should ignore it silently.
 744         */
 745        if (sk->sk_state == TCP_LISTEN) {
 746                if (svsk) {
 747                        set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
 748                        svc_xprt_enqueue(&svsk->sk_xprt);
 749                } else
 750                        printk("svc: socket %p: no user data\n", sk);
 751        }
 752
 753        if (sk_sleep(sk) && waitqueue_active(sk_sleep(sk)))
 754                wake_up_interruptible_all(sk_sleep(sk));
 755}
 756
 757/*
 758 * A state change on a connected socket means it's dying or dead.
 759 */
 760static void svc_tcp_state_change(struct sock *sk)
 761{
 762        struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
 763
 764        dprintk("svc: socket %p TCP (connected) state change %d (svsk %p)\n",
 765                sk, sk->sk_state, sk->sk_user_data);
 766
 767        if (!svsk)
 768                printk("svc: socket %p: no user data\n", sk);
 769        else {
 770                set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
 771                svc_xprt_enqueue(&svsk->sk_xprt);
 772        }
 773        if (sk_sleep(sk) && waitqueue_active(sk_sleep(sk)))
 774                wake_up_interruptible_all(sk_sleep(sk));
 775}
 776
 777static void svc_tcp_data_ready(struct sock *sk, int count)
 778{
 779        struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
 780
 781        dprintk("svc: socket %p TCP data ready (svsk %p)\n",
 782                sk, sk->sk_user_data);
 783        if (svsk) {
 784                set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
 785                svc_xprt_enqueue(&svsk->sk_xprt);
 786        }
 787        if (sk_sleep(sk) && waitqueue_active(sk_sleep(sk)))
 788                wake_up_interruptible(sk_sleep(sk));
 789}
 790
 791/*
 792 * Accept a TCP connection
 793 */
 794static struct svc_xprt *svc_tcp_accept(struct svc_xprt *xprt)
 795{
 796        struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
 797        struct sockaddr_storage addr;
 798        struct sockaddr *sin = (struct sockaddr *) &addr;
 799        struct svc_serv *serv = svsk->sk_xprt.xpt_server;
 800        struct socket   *sock = svsk->sk_sock;
 801        struct socket   *newsock;
 802        struct svc_sock *newsvsk;
 803        int             err, slen;
 804        RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
 805
 806        dprintk("svc: tcp_accept %p sock %p\n", svsk, sock);
 807        if (!sock)
 808                return NULL;
 809
 810        clear_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
 811        err = kernel_accept(sock, &newsock, O_NONBLOCK);
 812        if (err < 0) {
 813                if (err == -ENOMEM)
 814                        printk(KERN_WARNING "%s: no more sockets!\n",
 815                               serv->sv_name);
 816                else if (err != -EAGAIN && net_ratelimit())
 817                        printk(KERN_WARNING "%s: accept failed (err %d)!\n",
 818                                   serv->sv_name, -err);
 819                return NULL;
 820        }
 821        set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
 822
 823        err = kernel_getpeername(newsock, sin, &slen);
 824        if (err < 0) {
 825                if (net_ratelimit())
 826                        printk(KERN_WARNING "%s: peername failed (err %d)!\n",
 827                                   serv->sv_name, -err);
 828                goto failed;            /* aborted connection or whatever */
 829        }
 830
 831        /* Ideally, we would want to reject connections from unauthorized
 832         * hosts here, but when we get encryption, the IP of the host won't
 833         * tell us anything.  For now just warn about unpriv connections.
 834         */
 835        if (!svc_port_is_privileged(sin)) {
 836                dprintk(KERN_WARNING
 837                        "%s: connect from unprivileged port: %s\n",
 838                        serv->sv_name,
 839                        __svc_print_addr(sin, buf, sizeof(buf)));
 840        }
 841        dprintk("%s: connect from %s\n", serv->sv_name,
 842                __svc_print_addr(sin, buf, sizeof(buf)));
 843
 844        /* make sure that a write doesn't block forever when
 845         * low on memory
 846         */
 847        newsock->sk->sk_sndtimeo = HZ*30;
 848
 849        if (!(newsvsk = svc_setup_socket(serv, newsock, &err,
 850                                 (SVC_SOCK_ANONYMOUS | SVC_SOCK_TEMPORARY))))
 851                goto failed;
 852        svc_xprt_set_remote(&newsvsk->sk_xprt, sin, slen);
 853        err = kernel_getsockname(newsock, sin, &slen);
 854        if (unlikely(err < 0)) {
 855                dprintk("svc_tcp_accept: kernel_getsockname error %d\n", -err);
 856                slen = offsetof(struct sockaddr, sa_data);
 857        }
 858        svc_xprt_set_local(&newsvsk->sk_xprt, sin, slen);
 859
 860        if (serv->sv_stats)
 861                serv->sv_stats->nettcpconn++;
 862
 863        return &newsvsk->sk_xprt;
 864
 865failed:
 866        sock_release(newsock);
 867        return NULL;
 868}
 869
 870/*
 871 * Receive data.
 872 * If we haven't gotten the record length yet, get the next four bytes.
 873 * Otherwise try to gobble up as much as possible up to the complete
 874 * record length.
 875 */
 876static int svc_tcp_recv_record(struct svc_sock *svsk, struct svc_rqst *rqstp)
 877{
 878        struct svc_serv *serv = svsk->sk_xprt.xpt_server;
 879        int len;
 880
 881        if (test_and_clear_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags))
 882                /* sndbuf needs to have room for one request
 883                 * per thread, otherwise we can stall even when the
 884                 * network isn't a bottleneck.
 885                 *
 886                 * We count all threads rather than threads in a
 887                 * particular pool, which provides an upper bound
 888                 * on the number of threads which will access the socket.
 889                 *
 890                 * rcvbuf just needs to be able to hold a few requests.
 891                 * Normally they will be removed from the queue
 892                 * as soon a a complete request arrives.
 893                 */
 894                svc_sock_setbufsize(svsk->sk_sock,
 895                                    (serv->sv_nrthreads+3) * serv->sv_max_mesg,
 896                                    3 * serv->sv_max_mesg);
 897
 898        clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
 899
 900        if (svsk->sk_tcplen < sizeof(rpc_fraghdr)) {
 901                int             want = sizeof(rpc_fraghdr) - svsk->sk_tcplen;
 902                struct kvec     iov;
 903
 904                iov.iov_base = ((char *) &svsk->sk_reclen) + svsk->sk_tcplen;
 905                iov.iov_len  = want;
 906                if ((len = svc_recvfrom(rqstp, &iov, 1, want)) < 0)
 907                        goto error;
 908                svsk->sk_tcplen += len;
 909
 910                if (len < want) {
 911                        dprintk("svc: short recvfrom while reading record "
 912                                "length (%d of %d)\n", len, want);
 913                        goto err_again; /* record header not complete */
 914                }
 915
 916                svsk->sk_reclen = ntohl(svsk->sk_reclen);
 917                if (!(svsk->sk_reclen & RPC_LAST_STREAM_FRAGMENT)) {
 918                        /* FIXME: technically, a record can be fragmented,
 919                         *  and non-terminal fragments will not have the top
 920                         *  bit set in the fragment length header.
 921                         *  But apparently no known nfs clients send fragmented
 922                         *  records. */
 923                        if (net_ratelimit())
 924                                printk(KERN_NOTICE "RPC: multiple fragments "
 925                                        "per record not supported\n");
 926                        goto err_delete;
 927                }
 928
 929                svsk->sk_reclen &= RPC_FRAGMENT_SIZE_MASK;
 930                dprintk("svc: TCP record, %d bytes\n", svsk->sk_reclen);
 931                if (svsk->sk_reclen > serv->sv_max_mesg) {
 932                        if (net_ratelimit())
 933                                printk(KERN_NOTICE "RPC: "
 934                                        "fragment too large: 0x%08lx\n",
 935                                        (unsigned long)svsk->sk_reclen);
 936                        goto err_delete;
 937                }
 938        }
 939
 940        /* Check whether enough data is available */
 941        len = svc_recv_available(svsk);
 942        if (len < 0)
 943                goto error;
 944
 945        if (len < svsk->sk_reclen) {
 946                dprintk("svc: incomplete TCP record (%d of %d)\n",
 947                        len, svsk->sk_reclen);
 948                goto err_again; /* record not complete */
 949        }
 950        len = svsk->sk_reclen;
 951        set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
 952
 953        return len;
 954 error:
 955        if (len == -EAGAIN)
 956                dprintk("RPC: TCP recv_record got EAGAIN\n");
 957        return len;
 958 err_delete:
 959        set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
 960 err_again:
 961        return -EAGAIN;
 962}
 963
 964static int svc_process_calldir(struct svc_sock *svsk, struct svc_rqst *rqstp,
 965                               struct rpc_rqst **reqpp, struct kvec *vec)
 966{
 967        struct rpc_rqst *req = NULL;
 968        u32 *p;
 969        u32 xid;
 970        u32 calldir;
 971        int len;
 972
 973        len = svc_recvfrom(rqstp, vec, 1, 8);
 974        if (len < 0)
 975                goto error;
 976
 977        p = (u32 *)rqstp->rq_arg.head[0].iov_base;
 978        xid = *p++;
 979        calldir = *p;
 980
 981        if (calldir == 0) {
 982                /* REQUEST is the most common case */
 983                vec[0] = rqstp->rq_arg.head[0];
 984        } else {
 985                /* REPLY */
 986                if (svsk->sk_bc_xprt)
 987                        req = xprt_lookup_rqst(svsk->sk_bc_xprt, xid);
 988
 989                if (!req) {
 990                        printk(KERN_NOTICE
 991                                "%s: Got unrecognized reply: "
 992                                "calldir 0x%x sk_bc_xprt %p xid %08x\n",
 993                                __func__, ntohl(calldir),
 994                                svsk->sk_bc_xprt, xid);
 995                        vec[0] = rqstp->rq_arg.head[0];
 996                        goto out;
 997                }
 998
 999                memcpy(&req->rq_private_buf, &req->rq_rcv_buf,
1000                       sizeof(struct xdr_buf));
1001                /* copy the xid and call direction */
1002                memcpy(req->rq_private_buf.head[0].iov_base,
1003                       rqstp->rq_arg.head[0].iov_base, 8);
1004                vec[0] = req->rq_private_buf.head[0];
1005        }
1006 out:
1007        vec[0].iov_base += 8;
1008        vec[0].iov_len -= 8;
1009        len = svsk->sk_reclen - 8;
1010 error:
1011        *reqpp = req;
1012        return len;
1013}
1014
1015/*
1016 * Receive data from a TCP socket.
1017 */
1018static int svc_tcp_recvfrom(struct svc_rqst *rqstp)
1019{
1020        struct svc_sock *svsk =
1021                container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
1022        struct svc_serv *serv = svsk->sk_xprt.xpt_server;
1023        int             len;
1024        struct kvec *vec;
1025        int pnum, vlen;
1026        struct rpc_rqst *req = NULL;
1027
1028        dprintk("svc: tcp_recv %p data %d conn %d close %d\n",
1029                svsk, test_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags),
1030                test_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags),
1031                test_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags));
1032
1033        len = svc_tcp_recv_record(svsk, rqstp);
1034        if (len < 0)
1035                goto error;
1036
1037        vec = rqstp->rq_vec;
1038        vec[0] = rqstp->rq_arg.head[0];
1039        vlen = PAGE_SIZE;
1040
1041        /*
1042         * We have enough data for the whole tcp record. Let's try and read the
1043         * first 8 bytes to get the xid and the call direction. We can use this
1044         * to figure out if this is a call or a reply to a callback. If
1045         * sk_reclen is < 8 (xid and calldir), then this is a malformed packet.
1046         * In that case, don't bother with the calldir and just read the data.
1047         * It will be rejected in svc_process.
1048         */
1049        if (len >= 8) {
1050                len = svc_process_calldir(svsk, rqstp, &req, vec);
1051                if (len < 0)
1052                        goto err_again;
1053                vlen -= 8;
1054        }
1055
1056        pnum = 1;
1057        while (vlen < len) {
1058                vec[pnum].iov_base = (req) ?
1059                        page_address(req->rq_private_buf.pages[pnum - 1]) :
1060                        page_address(rqstp->rq_pages[pnum]);
1061                vec[pnum].iov_len = PAGE_SIZE;
1062                pnum++;
1063                vlen += PAGE_SIZE;
1064        }
1065        rqstp->rq_respages = &rqstp->rq_pages[pnum];
1066
1067        /* Now receive data */
1068        len = svc_recvfrom(rqstp, vec, pnum, len);
1069        if (len < 0)
1070                goto err_again;
1071
1072        /*
1073         * Account for the 8 bytes we read earlier
1074         */
1075        len += 8;
1076
1077        if (req) {
1078                xprt_complete_rqst(req->rq_task, len);
1079                len = 0;
1080                goto out;
1081        }
1082        dprintk("svc: TCP complete record (%d bytes)\n", len);
1083        rqstp->rq_arg.len = len;
1084        rqstp->rq_arg.page_base = 0;
1085        if (len <= rqstp->rq_arg.head[0].iov_len) {
1086                rqstp->rq_arg.head[0].iov_len = len;
1087                rqstp->rq_arg.page_len = 0;
1088        } else {
1089                rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len;
1090        }
1091
1092        rqstp->rq_xprt_ctxt   = NULL;
1093        rqstp->rq_prot        = IPPROTO_TCP;
1094
1095out:
1096        /* Reset TCP read info */
1097        svsk->sk_reclen = 0;
1098        svsk->sk_tcplen = 0;
1099
1100        svc_xprt_copy_addrs(rqstp, &svsk->sk_xprt);
1101        if (serv->sv_stats)
1102                serv->sv_stats->nettcpcnt++;
1103
1104        return len;
1105
1106err_again:
1107        if (len == -EAGAIN) {
1108                dprintk("RPC: TCP recvfrom got EAGAIN\n");
1109                return len;
1110        }
1111error:
1112        if (len != -EAGAIN) {
1113                printk(KERN_NOTICE "%s: recvfrom returned errno %d\n",
1114                       svsk->sk_xprt.xpt_server->sv_name, -len);
1115                set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
1116        }
1117        return -EAGAIN;
1118}
1119
1120/*
1121 * Send out data on TCP socket.
1122 */
1123static int svc_tcp_sendto(struct svc_rqst *rqstp)
1124{
1125        struct xdr_buf  *xbufp = &rqstp->rq_res;
1126        int sent;
1127        __be32 reclen;
1128
1129        /* Set up the first element of the reply kvec.
1130         * Any other kvecs that may be in use have been taken
1131         * care of by the server implementation itself.
1132         */
1133        reclen = htonl(0x80000000|((xbufp->len ) - 4));
1134        memcpy(xbufp->head[0].iov_base, &reclen, 4);
1135
1136        if (test_bit(XPT_DEAD, &rqstp->rq_xprt->xpt_flags))
1137                return -ENOTCONN;
1138
1139        sent = svc_sendto(rqstp, &rqstp->rq_res);
1140        if (sent != xbufp->len) {
1141                printk(KERN_NOTICE
1142                       "rpc-srv/tcp: %s: %s %d when sending %d bytes "
1143                       "- shutting down socket\n",
1144                       rqstp->rq_xprt->xpt_server->sv_name,
1145                       (sent<0)?"got error":"sent only",
1146                       sent, xbufp->len);
1147                set_bit(XPT_CLOSE, &rqstp->rq_xprt->xpt_flags);
1148                svc_xprt_enqueue(rqstp->rq_xprt);
1149                sent = -EAGAIN;
1150        }
1151        return sent;
1152}
1153
1154/*
1155 * Setup response header. TCP has a 4B record length field.
1156 */
1157static void svc_tcp_prep_reply_hdr(struct svc_rqst *rqstp)
1158{
1159        struct kvec *resv = &rqstp->rq_res.head[0];
1160
1161        /* tcp needs a space for the record length... */
1162        svc_putnl(resv, 0);
1163}
1164
1165static int svc_tcp_has_wspace(struct svc_xprt *xprt)
1166{
1167        struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1168        struct svc_serv *serv = svsk->sk_xprt.xpt_server;
1169        int required;
1170
1171        if (test_bit(XPT_LISTENER, &xprt->xpt_flags))
1172                return 1;
1173        required = atomic_read(&xprt->xpt_reserved) + serv->sv_max_mesg;
1174        if (sk_stream_wspace(svsk->sk_sk) >= required)
1175                return 1;
1176        set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
1177        return 0;
1178}
1179
1180static struct svc_xprt *svc_tcp_create(struct svc_serv *serv,
1181                                       struct sockaddr *sa, int salen,
1182                                       int flags)
1183{
1184        return svc_create_socket(serv, IPPROTO_TCP, sa, salen, flags);
1185}
1186
1187static struct svc_xprt_ops svc_tcp_ops = {
1188        .xpo_create = svc_tcp_create,
1189        .xpo_recvfrom = svc_tcp_recvfrom,
1190        .xpo_sendto = svc_tcp_sendto,
1191        .xpo_release_rqst = svc_release_skb,
1192        .xpo_detach = svc_tcp_sock_detach,
1193        .xpo_free = svc_sock_free,
1194        .xpo_prep_reply_hdr = svc_tcp_prep_reply_hdr,
1195        .xpo_has_wspace = svc_tcp_has_wspace,
1196        .xpo_accept = svc_tcp_accept,
1197};
1198
1199static struct svc_xprt_class svc_tcp_class = {
1200        .xcl_name = "tcp",
1201        .xcl_owner = THIS_MODULE,
1202        .xcl_ops = &svc_tcp_ops,
1203        .xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP,
1204};
1205
1206void svc_init_xprt_sock(void)
1207{
1208        svc_reg_xprt_class(&svc_tcp_class);
1209        svc_reg_xprt_class(&svc_udp_class);
1210}
1211
1212void svc_cleanup_xprt_sock(void)
1213{
1214        svc_unreg_xprt_class(&svc_tcp_class);
1215        svc_unreg_xprt_class(&svc_udp_class);
1216}
1217
1218static void svc_tcp_init(struct svc_sock *svsk, struct svc_serv *serv)
1219{
1220        struct sock     *sk = svsk->sk_sk;
1221
1222        svc_xprt_init(&svc_tcp_class, &svsk->sk_xprt, serv);
1223        set_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
1224        if (sk->sk_state == TCP_LISTEN) {
1225                dprintk("setting up TCP socket for listening\n");
1226                set_bit(XPT_LISTENER, &svsk->sk_xprt.xpt_flags);
1227                sk->sk_data_ready = svc_tcp_listen_data_ready;
1228                set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
1229        } else {
1230                dprintk("setting up TCP socket for reading\n");
1231                sk->sk_state_change = svc_tcp_state_change;
1232                sk->sk_data_ready = svc_tcp_data_ready;
1233                sk->sk_write_space = svc_tcp_write_space;
1234
1235                svsk->sk_reclen = 0;
1236                svsk->sk_tcplen = 0;
1237
1238                tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
1239
1240                /* initialise setting must have enough space to
1241                 * receive and respond to one request.
1242                 * svc_tcp_recvfrom will re-adjust if necessary
1243                 */
1244                svc_sock_setbufsize(svsk->sk_sock,
1245                                    3 * svsk->sk_xprt.xpt_server->sv_max_mesg,
1246                                    3 * svsk->sk_xprt.xpt_server->sv_max_mesg);
1247
1248                set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
1249                set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
1250                if (sk->sk_state != TCP_ESTABLISHED)
1251                        set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
1252        }
1253}
1254
1255void svc_sock_update_bufs(struct svc_serv *serv)
1256{
1257        /*
1258         * The number of server threads has changed. Update
1259         * rcvbuf and sndbuf accordingly on all sockets
1260         */
1261        struct list_head *le;
1262
1263        spin_lock_bh(&serv->sv_lock);
1264        list_for_each(le, &serv->sv_permsocks) {
1265                struct svc_sock *svsk =
1266                        list_entry(le, struct svc_sock, sk_xprt.xpt_list);
1267                set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
1268        }
1269        list_for_each(le, &serv->sv_tempsocks) {
1270                struct svc_sock *svsk =
1271                        list_entry(le, struct svc_sock, sk_xprt.xpt_list);
1272                set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
1273        }
1274        spin_unlock_bh(&serv->sv_lock);
1275}
1276EXPORT_SYMBOL_GPL(svc_sock_update_bufs);
1277
1278/*
1279 * Initialize socket for RPC use and create svc_sock struct
1280 * XXX: May want to setsockopt SO_SNDBUF and SO_RCVBUF.
1281 */
1282static struct svc_sock *svc_setup_socket(struct svc_serv *serv,
1283                                                struct socket *sock,
1284                                                int *errp, int flags)
1285{
1286        struct svc_sock *svsk;
1287        struct sock     *inet;
1288        int             pmap_register = !(flags & SVC_SOCK_ANONYMOUS);
1289
1290        dprintk("svc: svc_setup_socket %p\n", sock);
1291        if (!(svsk = kzalloc(sizeof(*svsk), GFP_KERNEL))) {
1292                *errp = -ENOMEM;
1293                return NULL;
1294        }
1295
1296        inet = sock->sk;
1297
1298        /* Register socket with portmapper */
1299        if (*errp >= 0 && pmap_register)
1300                *errp = svc_register(serv, inet->sk_family, inet->sk_protocol,
1301                                     ntohs(inet_sk(inet)->inet_sport));
1302
1303        if (*errp < 0) {
1304                kfree(svsk);
1305                return NULL;
1306        }
1307
1308        inet->sk_user_data = svsk;
1309        svsk->sk_sock = sock;
1310        svsk->sk_sk = inet;
1311        svsk->sk_ostate = inet->sk_state_change;
1312        svsk->sk_odata = inet->sk_data_ready;
1313        svsk->sk_owspace = inet->sk_write_space;
1314
1315        /* Initialize the socket */
1316        if (sock->type == SOCK_DGRAM)
1317                svc_udp_init(svsk, serv);
1318        else
1319                svc_tcp_init(svsk, serv);
1320
1321        dprintk("svc: svc_setup_socket created %p (inet %p)\n",
1322                                svsk, svsk->sk_sk);
1323
1324        return svsk;
1325}
1326
1327/**
1328 * svc_addsock - add a listener socket to an RPC service
1329 * @serv: pointer to RPC service to which to add a new listener
1330 * @fd: file descriptor of the new listener
1331 * @name_return: pointer to buffer to fill in with name of listener
1332 * @len: size of the buffer
1333 *
1334 * Fills in socket name and returns positive length of name if successful.
1335 * Name is terminated with '\n'.  On error, returns a negative errno
1336 * value.
1337 */
1338int svc_addsock(struct svc_serv *serv, const int fd, char *name_return,
1339                const size_t len)
1340{
1341        int err = 0;
1342        struct socket *so = sockfd_lookup(fd, &err);
1343        struct svc_sock *svsk = NULL;
1344
1345        if (!so)
1346                return err;
1347        if ((so->sk->sk_family != PF_INET) && (so->sk->sk_family != PF_INET6))
1348                err =  -EAFNOSUPPORT;
1349        else if (so->sk->sk_protocol != IPPROTO_TCP &&
1350            so->sk->sk_protocol != IPPROTO_UDP)
1351                err =  -EPROTONOSUPPORT;
1352        else if (so->state > SS_UNCONNECTED)
1353                err = -EISCONN;
1354        else {
1355                if (!try_module_get(THIS_MODULE))
1356                        err = -ENOENT;
1357                else
1358                        svsk = svc_setup_socket(serv, so, &err,
1359                                                SVC_SOCK_DEFAULTS);
1360                if (svsk) {
1361                        struct sockaddr_storage addr;
1362                        struct sockaddr *sin = (struct sockaddr *)&addr;
1363                        int salen;
1364                        if (kernel_getsockname(svsk->sk_sock, sin, &salen) == 0)
1365                                svc_xprt_set_local(&svsk->sk_xprt, sin, salen);
1366                        clear_bit(XPT_TEMP, &svsk->sk_xprt.xpt_flags);
1367                        spin_lock_bh(&serv->sv_lock);
1368                        list_add(&svsk->sk_xprt.xpt_list, &serv->sv_permsocks);
1369                        spin_unlock_bh(&serv->sv_lock);
1370                        svc_xprt_received(&svsk->sk_xprt);
1371                        err = 0;
1372                } else
1373                        module_put(THIS_MODULE);
1374        }
1375        if (err) {
1376                sockfd_put(so);
1377                return err;
1378        }
1379        return svc_one_sock_name(svsk, name_return, len);
1380}
1381EXPORT_SYMBOL_GPL(svc_addsock);
1382
1383/*
1384 * Create socket for RPC service.
1385 */
1386static struct svc_xprt *svc_create_socket(struct svc_serv *serv,
1387                                          int protocol,
1388                                          struct sockaddr *sin, int len,
1389                                          int flags)
1390{
1391        struct svc_sock *svsk;
1392        struct socket   *sock;
1393        int             error;
1394        int             type;
1395        struct sockaddr_storage addr;
1396        struct sockaddr *newsin = (struct sockaddr *)&addr;
1397        int             newlen;
1398        int             family;
1399        int             val;
1400        RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
1401
1402        dprintk("svc: svc_create_socket(%s, %d, %s)\n",
1403                        serv->sv_program->pg_name, protocol,
1404                        __svc_print_addr(sin, buf, sizeof(buf)));
1405
1406        if (protocol != IPPROTO_UDP && protocol != IPPROTO_TCP) {
1407                printk(KERN_WARNING "svc: only UDP and TCP "
1408                                "sockets supported\n");
1409                return ERR_PTR(-EINVAL);
1410        }
1411
1412        type = (protocol == IPPROTO_UDP)? SOCK_DGRAM : SOCK_STREAM;
1413        switch (sin->sa_family) {
1414        case AF_INET6:
1415                family = PF_INET6;
1416                break;
1417        case AF_INET:
1418                family = PF_INET;
1419                break;
1420        default:
1421                return ERR_PTR(-EINVAL);
1422        }
1423
1424        error = sock_create_kern(family, type, protocol, &sock);
1425        if (error < 0)
1426                return ERR_PTR(error);
1427
1428        svc_reclassify_socket(sock);
1429
1430        /*
1431         * If this is an PF_INET6 listener, we want to avoid
1432         * getting requests from IPv4 remotes.  Those should
1433         * be shunted to a PF_INET listener via rpcbind.
1434         */
1435        val = 1;
1436        if (family == PF_INET6)
1437                kernel_setsockopt(sock, SOL_IPV6, IPV6_V6ONLY,
1438                                        (char *)&val, sizeof(val));
1439
1440        if (type == SOCK_STREAM)
1441                sock->sk->sk_reuse = 1;         /* allow address reuse */
1442        error = kernel_bind(sock, sin, len);
1443        if (error < 0)
1444                goto bummer;
1445
1446        newlen = len;
1447        error = kernel_getsockname(sock, newsin, &newlen);
1448        if (error < 0)
1449                goto bummer;
1450
1451        if (protocol == IPPROTO_TCP) {
1452                if ((error = kernel_listen(sock, 64)) < 0)
1453                        goto bummer;
1454        }
1455
1456        if ((svsk = svc_setup_socket(serv, sock, &error, flags)) != NULL) {
1457                svc_xprt_set_local(&svsk->sk_xprt, newsin, newlen);
1458                return (struct svc_xprt *)svsk;
1459        }
1460
1461bummer:
1462        dprintk("svc: svc_create_socket error = %d\n", -error);
1463        sock_release(sock);
1464        return ERR_PTR(error);
1465}
1466
1467/*
1468 * Detach the svc_sock from the socket so that no
1469 * more callbacks occur.
1470 */
1471static void svc_sock_detach(struct svc_xprt *xprt)
1472{
1473        struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1474        struct sock *sk = svsk->sk_sk;
1475
1476        dprintk("svc: svc_sock_detach(%p)\n", svsk);
1477
1478        /* put back the old socket callbacks */
1479        sk->sk_state_change = svsk->sk_ostate;
1480        sk->sk_data_ready = svsk->sk_odata;
1481        sk->sk_write_space = svsk->sk_owspace;
1482
1483        if (sk_sleep(sk) && waitqueue_active(sk_sleep(sk)))
1484                wake_up_interruptible(sk_sleep(sk));
1485}
1486
1487/*
1488 * Disconnect the socket, and reset the callbacks
1489 */
1490static void svc_tcp_sock_detach(struct svc_xprt *xprt)
1491{
1492        struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1493
1494        dprintk("svc: svc_tcp_sock_detach(%p)\n", svsk);
1495
1496        svc_sock_detach(xprt);
1497
1498        if (!test_bit(XPT_LISTENER, &xprt->xpt_flags))
1499                kernel_sock_shutdown(svsk->sk_sock, SHUT_RDWR);
1500}
1501
1502/*
1503 * Free the svc_sock's socket resources and the svc_sock itself.
1504 */
1505static void svc_sock_free(struct svc_xprt *xprt)
1506{
1507        struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1508        dprintk("svc: svc_sock_free(%p)\n", svsk);
1509
1510        if (svsk->sk_sock->file)
1511                sockfd_put(svsk->sk_sock);
1512        else
1513                sock_release(svsk->sk_sock);
1514        kfree(svsk);
1515}
1516
1517/*
1518 * Create a svc_xprt.
1519 *
1520 * For internal use only (e.g. nfsv4.1 backchannel).
1521 * Callers should typically use the xpo_create() method.
1522 */
1523struct svc_xprt *svc_sock_create(struct svc_serv *serv, int prot)
1524{
1525        struct svc_sock *svsk;
1526        struct svc_xprt *xprt = NULL;
1527
1528        dprintk("svc: %s\n", __func__);
1529        svsk = kzalloc(sizeof(*svsk), GFP_KERNEL);
1530        if (!svsk)
1531                goto out;
1532
1533        xprt = &svsk->sk_xprt;
1534        if (prot == IPPROTO_TCP)
1535                svc_xprt_init(&svc_tcp_class, xprt, serv);
1536        else if (prot == IPPROTO_UDP)
1537                svc_xprt_init(&svc_udp_class, xprt, serv);
1538        else
1539                BUG();
1540out:
1541        dprintk("svc: %s return %p\n", __func__, xprt);
1542        return xprt;
1543}
1544EXPORT_SYMBOL_GPL(svc_sock_create);
1545
1546/*
1547 * Destroy a svc_sock.
1548 */
1549void svc_sock_destroy(struct svc_xprt *xprt)
1550{
1551        if (xprt)
1552                kfree(container_of(xprt, struct svc_sock, sk_xprt));
1553}
1554EXPORT_SYMBOL_GPL(svc_sock_destroy);
1555