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/module.h>
  25#include <linux/errno.h>
  26#include <linux/fcntl.h>
  27#include <linux/net.h>
  28#include <linux/in.h>
  29#include <linux/inet.h>
  30#include <linux/udp.h>
  31#include <linux/tcp.h>
  32#include <linux/unistd.h>
  33#include <linux/slab.h>
  34#include <linux/netdevice.h>
  35#include <linux/skbuff.h>
  36#include <linux/file.h>
  37#include <linux/freezer.h>
  38#include <net/sock.h>
  39#include <net/checksum.h>
  40#include <net/ip.h>
  41#include <net/ipv6.h>
  42#include <net/tcp.h>
  43#include <net/tcp_states.h>
  44#include <asm/uaccess.h>
  45#include <asm/ioctls.h>
  46#include <trace/events/skb.h>
  47
  48#include <linux/sunrpc/types.h>
  49#include <linux/sunrpc/clnt.h>
  50#include <linux/sunrpc/xdr.h>
  51#include <linux/sunrpc/msg_prot.h>
  52#include <linux/sunrpc/svcsock.h>
  53#include <linux/sunrpc/stats.h>
  54#include <linux/sunrpc/xprt.h>
  55
  56#include "sunrpc.h"
  57
  58#define RPCDBG_FACILITY RPCDBG_SVCXPRT
  59
  60
  61static struct svc_sock *svc_setup_socket(struct svc_serv *, struct socket *,
  62                                         int *errp, int flags);
  63static void             svc_udp_data_ready(struct sock *, int);
  64static int              svc_udp_recvfrom(struct svc_rqst *);
  65static int              svc_udp_sendto(struct svc_rqst *);
  66static void             svc_sock_detach(struct svc_xprt *);
  67static void             svc_tcp_sock_detach(struct svc_xprt *);
  68static void             svc_sock_free(struct svc_xprt *);
  69
  70static struct svc_xprt *svc_create_socket(struct svc_serv *, int,
  71                                          struct net *, struct sockaddr *,
  72                                          int, int);
  73#if defined(CONFIG_SUNRPC_BACKCHANNEL)
  74static struct svc_xprt *svc_bc_create_socket(struct svc_serv *, int,
  75                                             struct net *, struct sockaddr *,
  76                                             int, int);
  77static void svc_bc_sock_free(struct svc_xprt *xprt);
  78#endif /* CONFIG_SUNRPC_BACKCHANNEL */
  79
  80#ifdef CONFIG_DEBUG_LOCK_ALLOC
  81static struct lock_class_key svc_key[2];
  82static struct lock_class_key svc_slock_key[2];
  83
  84static void svc_reclassify_socket(struct socket *sock)
  85{
  86        struct sock *sk = sock->sk;
  87        BUG_ON(sock_owned_by_user(sk));
  88        switch (sk->sk_family) {
  89        case AF_INET:
  90                sock_lock_init_class_and_name(sk, "slock-AF_INET-NFSD",
  91                                              &svc_slock_key[0],
  92                                              "sk_xprt.xpt_lock-AF_INET-NFSD",
  93                                              &svc_key[0]);
  94                break;
  95
  96        case AF_INET6:
  97                sock_lock_init_class_and_name(sk, "slock-AF_INET6-NFSD",
  98                                              &svc_slock_key[1],
  99                                              "sk_xprt.xpt_lock-AF_INET6-NFSD",
 100                                              &svc_key[1]);
 101                break;
 102
 103        default:
 104                BUG();
 105        }
 106}
 107#else
 108static void svc_reclassify_socket(struct socket *sock)
 109{
 110}
 111#endif
 112
 113/*
 114 * Release an skbuff after use
 115 */
 116static void svc_release_skb(struct svc_rqst *rqstp)
 117{
 118        struct sk_buff *skb = rqstp->rq_xprt_ctxt;
 119
 120        if (skb) {
 121                struct svc_sock *svsk =
 122                        container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
 123                rqstp->rq_xprt_ctxt = NULL;
 124
 125                dprintk("svc: service %p, releasing skb %p\n", rqstp, skb);
 126                skb_free_datagram_locked(svsk->sk_sk, skb);
 127        }
 128}
 129
 130union svc_pktinfo_u {
 131        struct in_pktinfo pkti;
 132        struct in6_pktinfo pkti6;
 133};
 134#define SVC_PKTINFO_SPACE \
 135        CMSG_SPACE(sizeof(union svc_pktinfo_u))
 136
 137static void svc_set_cmsg_data(struct svc_rqst *rqstp, struct cmsghdr *cmh)
 138{
 139        struct svc_sock *svsk =
 140                container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
 141        switch (svsk->sk_sk->sk_family) {
 142        case AF_INET: {
 143                        struct in_pktinfo *pki = CMSG_DATA(cmh);
 144
 145                        cmh->cmsg_level = SOL_IP;
 146                        cmh->cmsg_type = IP_PKTINFO;
 147                        pki->ipi_ifindex = 0;
 148                        pki->ipi_spec_dst.s_addr =
 149                                 svc_daddr_in(rqstp)->sin_addr.s_addr;
 150                        cmh->cmsg_len = CMSG_LEN(sizeof(*pki));
 151                }
 152                break;
 153
 154        case AF_INET6: {
 155                        struct in6_pktinfo *pki = CMSG_DATA(cmh);
 156                        struct sockaddr_in6 *daddr = svc_daddr_in6(rqstp);
 157
 158                        cmh->cmsg_level = SOL_IPV6;
 159                        cmh->cmsg_type = IPV6_PKTINFO;
 160                        pki->ipi6_ifindex = daddr->sin6_scope_id;
 161                        pki->ipi6_addr = daddr->sin6_addr;
 162                        cmh->cmsg_len = CMSG_LEN(sizeof(*pki));
 163                }
 164                break;
 165        }
 166}
 167
 168/*
 169 * send routine intended to be shared by the fore- and back-channel
 170 */
 171int svc_send_common(struct socket *sock, struct xdr_buf *xdr,
 172                    struct page *headpage, unsigned long headoffset,
 173                    struct page *tailpage, unsigned long tailoffset)
 174{
 175        int             result;
 176        int             size;
 177        struct page     **ppage = xdr->pages;
 178        size_t          base = xdr->page_base;
 179        unsigned int    pglen = xdr->page_len;
 180        unsigned int    flags = MSG_MORE;
 181        int             slen;
 182        int             len = 0;
 183
 184        slen = xdr->len;
 185
 186        /* send head */
 187        if (slen == xdr->head[0].iov_len)
 188                flags = 0;
 189        len = kernel_sendpage(sock, headpage, headoffset,
 190                                  xdr->head[0].iov_len, flags);
 191        if (len != xdr->head[0].iov_len)
 192                goto out;
 193        slen -= xdr->head[0].iov_len;
 194        if (slen == 0)
 195                goto out;
 196
 197        /* send page data */
 198        size = PAGE_SIZE - base < pglen ? PAGE_SIZE - base : pglen;
 199        while (pglen > 0) {
 200                if (slen == size)
 201                        flags = 0;
 202                result = kernel_sendpage(sock, *ppage, base, size, flags);
 203                if (result > 0)
 204                        len += result;
 205                if (result != size)
 206                        goto out;
 207                slen -= size;
 208                pglen -= size;
 209                size = PAGE_SIZE < pglen ? PAGE_SIZE : pglen;
 210                base = 0;
 211                ppage++;
 212        }
 213
 214        /* send tail */
 215        if (xdr->tail[0].iov_len) {
 216                result = kernel_sendpage(sock, tailpage, tailoffset,
 217                                   xdr->tail[0].iov_len, 0);
 218                if (result > 0)
 219                        len += result;
 220        }
 221
 222out:
 223        return len;
 224}
 225
 226
 227/*
 228 * Generic sendto routine
 229 */
 230static int svc_sendto(struct svc_rqst *rqstp, struct xdr_buf *xdr)
 231{
 232        struct svc_sock *svsk =
 233                container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
 234        struct socket   *sock = svsk->sk_sock;
 235        union {
 236                struct cmsghdr  hdr;
 237                long            all[SVC_PKTINFO_SPACE / sizeof(long)];
 238        } buffer;
 239        struct cmsghdr *cmh = &buffer.hdr;
 240        int             len = 0;
 241        unsigned long tailoff;
 242        unsigned long headoff;
 243        RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
 244
 245        if (rqstp->rq_prot == IPPROTO_UDP) {
 246                struct msghdr msg = {
 247                        .msg_name       = &rqstp->rq_addr,
 248                        .msg_namelen    = rqstp->rq_addrlen,
 249                        .msg_control    = cmh,
 250                        .msg_controllen = sizeof(buffer),
 251                        .msg_flags      = MSG_MORE,
 252                };
 253
 254                svc_set_cmsg_data(rqstp, cmh);
 255
 256                if (sock_sendmsg(sock, &msg, 0) < 0)
 257                        goto out;
 258        }
 259
 260        tailoff = ((unsigned long)xdr->tail[0].iov_base) & (PAGE_SIZE-1);
 261        headoff = 0;
 262        len = svc_send_common(sock, xdr, rqstp->rq_respages[0], headoff,
 263                               rqstp->rq_respages[0], tailoff);
 264
 265out:
 266        dprintk("svc: socket %p sendto([%p %Zu... ], %d) = %d (addr %s)\n",
 267                svsk, xdr->head[0].iov_base, xdr->head[0].iov_len,
 268                xdr->len, len, svc_print_addr(rqstp, buf, sizeof(buf)));
 269
 270        return len;
 271}
 272
 273/*
 274 * Report socket names for nfsdfs
 275 */
 276static int svc_one_sock_name(struct svc_sock *svsk, char *buf, int remaining)
 277{
 278        const struct sock *sk = svsk->sk_sk;
 279        const char *proto_name = sk->sk_protocol == IPPROTO_UDP ?
 280                                                        "udp" : "tcp";
 281        int len;
 282
 283        switch (sk->sk_family) {
 284        case PF_INET:
 285                len = snprintf(buf, remaining, "ipv4 %s %pI4 %d\n",
 286                                proto_name,
 287                                &inet_sk(sk)->inet_rcv_saddr,
 288                                inet_sk(sk)->inet_num);
 289                break;
 290        case PF_INET6:
 291                len = snprintf(buf, remaining, "ipv6 %s %pI6 %d\n",
 292                                proto_name,
 293                                &inet6_sk(sk)->rcv_saddr,
 294                                inet_sk(sk)->inet_num);
 295                break;
 296        default:
 297                len = snprintf(buf, remaining, "*unknown-%d*\n",
 298                                sk->sk_family);
 299        }
 300
 301        if (len >= remaining) {
 302                *buf = '\0';
 303                return -ENAMETOOLONG;
 304        }
 305        return len;
 306}
 307
 308/**
 309 * svc_sock_names - construct a list of listener names in a string
 310 * @serv: pointer to RPC service
 311 * @buf: pointer to a buffer to fill in with socket names
 312 * @buflen: size of the buffer to be filled
 313 * @toclose: pointer to '\0'-terminated C string containing the name
 314 *              of a listener to be closed
 315 *
 316 * Fills in @buf with a '\n'-separated list of names of listener
 317 * sockets.  If @toclose is not NULL, the socket named by @toclose
 318 * is closed, and is not included in the output list.
 319 *
 320 * Returns positive length of the socket name string, or a negative
 321 * errno value on error.
 322 */
 323int svc_sock_names(struct svc_serv *serv, char *buf, const size_t buflen,
 324                   const char *toclose)
 325{
 326        struct svc_sock *svsk, *closesk = NULL;
 327        int len = 0;
 328
 329        if (!serv)
 330                return 0;
 331
 332        spin_lock_bh(&serv->sv_lock);
 333        list_for_each_entry(svsk, &serv->sv_permsocks, sk_xprt.xpt_list) {
 334                int onelen = svc_one_sock_name(svsk, buf + len, buflen - len);
 335                if (onelen < 0) {
 336                        len = onelen;
 337                        break;
 338                }
 339                if (toclose && strcmp(toclose, buf + len) == 0) {
 340                        closesk = svsk;
 341                        svc_xprt_get(&closesk->sk_xprt);
 342                } else
 343                        len += onelen;
 344        }
 345        spin_unlock_bh(&serv->sv_lock);
 346
 347        if (closesk) {
 348                /* Should unregister with portmap, but you cannot
 349                 * unregister just one protocol...
 350                 */
 351                svc_close_xprt(&closesk->sk_xprt);
 352                svc_xprt_put(&closesk->sk_xprt);
 353        } else if (toclose)
 354                return -ENOENT;
 355        return len;
 356}
 357EXPORT_SYMBOL_GPL(svc_sock_names);
 358
 359/*
 360 * Check input queue length
 361 */
 362static int svc_recv_available(struct svc_sock *svsk)
 363{
 364        struct socket   *sock = svsk->sk_sock;
 365        int             avail, err;
 366
 367        err = kernel_sock_ioctl(sock, TIOCINQ, (unsigned long) &avail);
 368
 369        return (err >= 0)? avail : err;
 370}
 371
 372/*
 373 * Generic recvfrom routine.
 374 */
 375static int svc_recvfrom(struct svc_rqst *rqstp, struct kvec *iov, int nr,
 376                        int buflen)
 377{
 378        struct svc_sock *svsk =
 379                container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
 380        struct msghdr msg = {
 381                .msg_flags      = MSG_DONTWAIT,
 382        };
 383        int len;
 384
 385        rqstp->rq_xprt_hlen = 0;
 386
 387        len = kernel_recvmsg(svsk->sk_sock, &msg, iov, nr, buflen,
 388                                msg.msg_flags);
 389
 390        dprintk("svc: socket %p recvfrom(%p, %Zu) = %d\n",
 391                svsk, iov[0].iov_base, iov[0].iov_len, len);
 392        return len;
 393}
 394
 395static int svc_partial_recvfrom(struct svc_rqst *rqstp,
 396                                struct kvec *iov, int nr,
 397                                int buflen, unsigned int base)
 398{
 399        size_t save_iovlen;
 400        void *save_iovbase;
 401        unsigned int i;
 402        int ret;
 403
 404        if (base == 0)
 405                return svc_recvfrom(rqstp, iov, nr, buflen);
 406
 407        for (i = 0; i < nr; i++) {
 408                if (iov[i].iov_len > base)
 409                        break;
 410                base -= iov[i].iov_len;
 411        }
 412        save_iovlen = iov[i].iov_len;
 413        save_iovbase = iov[i].iov_base;
 414        iov[i].iov_len -= base;
 415        iov[i].iov_base += base;
 416        ret = svc_recvfrom(rqstp, &iov[i], nr - i, buflen);
 417        iov[i].iov_len = save_iovlen;
 418        iov[i].iov_base = save_iovbase;
 419        return ret;
 420}
 421
 422/*
 423 * Set socket snd and rcv buffer lengths
 424 */
 425static void svc_sock_setbufsize(struct socket *sock, unsigned int snd,
 426                                unsigned int rcv)
 427{
 428#if 0
 429        mm_segment_t    oldfs;
 430        oldfs = get_fs(); set_fs(KERNEL_DS);
 431        sock_setsockopt(sock, SOL_SOCKET, SO_SNDBUF,
 432                        (char*)&snd, sizeof(snd));
 433        sock_setsockopt(sock, SOL_SOCKET, SO_RCVBUF,
 434                        (char*)&rcv, sizeof(rcv));
 435#else
 436        /* sock_setsockopt limits use to sysctl_?mem_max,
 437         * which isn't acceptable.  Until that is made conditional
 438         * on not having CAP_SYS_RESOURCE or similar, we go direct...
 439         * DaveM said I could!
 440         */
 441        lock_sock(sock->sk);
 442        sock->sk->sk_sndbuf = snd * 2;
 443        sock->sk->sk_rcvbuf = rcv * 2;
 444        sock->sk->sk_write_space(sock->sk);
 445        release_sock(sock->sk);
 446#endif
 447}
 448/*
 449 * INET callback when data has been received on the socket.
 450 */
 451static void svc_udp_data_ready(struct sock *sk, int count)
 452{
 453        struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
 454        wait_queue_head_t *wq = sk_sleep(sk);
 455
 456        if (svsk) {
 457                dprintk("svc: socket %p(inet %p), count=%d, busy=%d\n",
 458                        svsk, sk, count,
 459                        test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags));
 460                set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
 461                svc_xprt_enqueue(&svsk->sk_xprt);
 462        }
 463        if (wq && waitqueue_active(wq))
 464                wake_up_interruptible(wq);
 465}
 466
 467/*
 468 * INET callback when space is newly available on the socket.
 469 */
 470static void svc_write_space(struct sock *sk)
 471{
 472        struct svc_sock *svsk = (struct svc_sock *)(sk->sk_user_data);
 473        wait_queue_head_t *wq = sk_sleep(sk);
 474
 475        if (svsk) {
 476                dprintk("svc: socket %p(inet %p), write_space busy=%d\n",
 477                        svsk, sk, test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags));
 478                svc_xprt_enqueue(&svsk->sk_xprt);
 479        }
 480
 481        if (wq && waitqueue_active(wq)) {
 482                dprintk("RPC svc_write_space: someone sleeping on %p\n",
 483                       svsk);
 484                wake_up_interruptible(wq);
 485        }
 486}
 487
 488static void svc_tcp_write_space(struct sock *sk)
 489{
 490        struct socket *sock = sk->sk_socket;
 491
 492        if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk) && sock)
 493                clear_bit(SOCK_NOSPACE, &sock->flags);
 494        svc_write_space(sk);
 495}
 496
 497/*
 498 * See net/ipv6/ip_sockglue.c : ip_cmsg_recv_pktinfo
 499 */
 500static int svc_udp_get_dest_address4(struct svc_rqst *rqstp,
 501                                     struct cmsghdr *cmh)
 502{
 503        struct in_pktinfo *pki = CMSG_DATA(cmh);
 504        struct sockaddr_in *daddr = svc_daddr_in(rqstp);
 505
 506        if (cmh->cmsg_type != IP_PKTINFO)
 507                return 0;
 508
 509        daddr->sin_family = AF_INET;
 510        daddr->sin_addr.s_addr = pki->ipi_spec_dst.s_addr;
 511        return 1;
 512}
 513
 514/*
 515 * See net/ipv6/datagram.c : datagram_recv_ctl
 516 */
 517static int svc_udp_get_dest_address6(struct svc_rqst *rqstp,
 518                                     struct cmsghdr *cmh)
 519{
 520        struct in6_pktinfo *pki = CMSG_DATA(cmh);
 521        struct sockaddr_in6 *daddr = svc_daddr_in6(rqstp);
 522
 523        if (cmh->cmsg_type != IPV6_PKTINFO)
 524                return 0;
 525
 526        daddr->sin6_family = AF_INET6;
 527        daddr->sin6_addr = pki->ipi6_addr;
 528        daddr->sin6_scope_id = pki->ipi6_ifindex;
 529        return 1;
 530}
 531
 532/*
 533 * Copy the UDP datagram's destination address to the rqstp structure.
 534 * The 'destination' address in this case is the address to which the
 535 * peer sent the datagram, i.e. our local address. For multihomed
 536 * hosts, this can change from msg to msg. Note that only the IP
 537 * address changes, the port number should remain the same.
 538 */
 539static int svc_udp_get_dest_address(struct svc_rqst *rqstp,
 540                                    struct cmsghdr *cmh)
 541{
 542        switch (cmh->cmsg_level) {
 543        case SOL_IP:
 544                return svc_udp_get_dest_address4(rqstp, cmh);
 545        case SOL_IPV6:
 546                return svc_udp_get_dest_address6(rqstp, cmh);
 547        }
 548
 549        return 0;
 550}
 551
 552/*
 553 * Receive a datagram from a UDP socket.
 554 */
 555static int svc_udp_recvfrom(struct svc_rqst *rqstp)
 556{
 557        struct svc_sock *svsk =
 558                container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
 559        struct svc_serv *serv = svsk->sk_xprt.xpt_server;
 560        struct sk_buff  *skb;
 561        union {
 562                struct cmsghdr  hdr;
 563                long            all[SVC_PKTINFO_SPACE / sizeof(long)];
 564        } buffer;
 565        struct cmsghdr *cmh = &buffer.hdr;
 566        struct msghdr msg = {
 567                .msg_name = svc_addr(rqstp),
 568                .msg_control = cmh,
 569                .msg_controllen = sizeof(buffer),
 570                .msg_flags = MSG_DONTWAIT,
 571        };
 572        size_t len;
 573        int err;
 574
 575        if (test_and_clear_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags))
 576            /* udp sockets need large rcvbuf as all pending
 577             * requests are still in that buffer.  sndbuf must
 578             * also be large enough that there is enough space
 579             * for one reply per thread.  We count all threads
 580             * rather than threads in a particular pool, which
 581             * provides an upper bound on the number of threads
 582             * which will access the socket.
 583             */
 584            svc_sock_setbufsize(svsk->sk_sock,
 585                                (serv->sv_nrthreads+3) * serv->sv_max_mesg,
 586                                (serv->sv_nrthreads+3) * serv->sv_max_mesg);
 587
 588        clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
 589        skb = NULL;
 590        err = kernel_recvmsg(svsk->sk_sock, &msg, NULL,
 591                             0, 0, MSG_PEEK | MSG_DONTWAIT);
 592        if (err >= 0)
 593                skb = skb_recv_datagram(svsk->sk_sk, 0, 1, &err);
 594
 595        if (skb == NULL) {
 596                if (err != -EAGAIN) {
 597                        /* possibly an icmp error */
 598                        dprintk("svc: recvfrom returned error %d\n", -err);
 599                        set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
 600                }
 601                return -EAGAIN;
 602        }
 603        len = svc_addr_len(svc_addr(rqstp));
 604        if (len == 0)
 605                return -EAFNOSUPPORT;
 606        rqstp->rq_addrlen = len;
 607        if (skb->tstamp.tv64 == 0) {
 608                skb->tstamp = ktime_get_real();
 609                /* Don't enable netstamp, sunrpc doesn't
 610                   need that much accuracy */
 611        }
 612        svsk->sk_sk->sk_stamp = skb->tstamp;
 613        set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); /* there may be more data... */
 614
 615        len  = skb->len - sizeof(struct udphdr);
 616        rqstp->rq_arg.len = len;
 617
 618        rqstp->rq_prot = IPPROTO_UDP;
 619
 620        if (!svc_udp_get_dest_address(rqstp, cmh)) {
 621                net_warn_ratelimited("svc: received unknown control message %d/%d; dropping RPC reply datagram\n",
 622                                     cmh->cmsg_level, cmh->cmsg_type);
 623out_free:
 624                trace_kfree_skb(skb, svc_udp_recvfrom);
 625                skb_free_datagram_locked(svsk->sk_sk, skb);
 626                return 0;
 627        }
 628        rqstp->rq_daddrlen = svc_addr_len(svc_daddr(rqstp));
 629
 630        if (skb_is_nonlinear(skb)) {
 631                /* we have to copy */
 632                local_bh_disable();
 633                if (csum_partial_copy_to_xdr(&rqstp->rq_arg, skb)) {
 634                        local_bh_enable();
 635                        /* checksum error */
 636                        goto out_free;
 637                }
 638                local_bh_enable();
 639                skb_free_datagram_locked(svsk->sk_sk, skb);
 640        } else {
 641                /* we can use it in-place */
 642                rqstp->rq_arg.head[0].iov_base = skb->data +
 643                        sizeof(struct udphdr);
 644                rqstp->rq_arg.head[0].iov_len = len;
 645                if (skb_checksum_complete(skb))
 646                        goto out_free;
 647                rqstp->rq_xprt_ctxt = skb;
 648        }
 649
 650        rqstp->rq_arg.page_base = 0;
 651        if (len <= rqstp->rq_arg.head[0].iov_len) {
 652                rqstp->rq_arg.head[0].iov_len = len;
 653                rqstp->rq_arg.page_len = 0;
 654                rqstp->rq_respages = rqstp->rq_pages+1;
 655        } else {
 656                rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len;
 657                rqstp->rq_respages = rqstp->rq_pages + 1 +
 658                        DIV_ROUND_UP(rqstp->rq_arg.page_len, PAGE_SIZE);
 659        }
 660
 661        if (serv->sv_stats)
 662                serv->sv_stats->netudpcnt++;
 663
 664        return len;
 665}
 666
 667static int
 668svc_udp_sendto(struct svc_rqst *rqstp)
 669{
 670        int             error;
 671
 672        error = svc_sendto(rqstp, &rqstp->rq_res);
 673        if (error == -ECONNREFUSED)
 674                /* ICMP error on earlier request. */
 675                error = svc_sendto(rqstp, &rqstp->rq_res);
 676
 677        return error;
 678}
 679
 680static void svc_udp_prep_reply_hdr(struct svc_rqst *rqstp)
 681{
 682}
 683
 684static int svc_udp_has_wspace(struct svc_xprt *xprt)
 685{
 686        struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
 687        struct svc_serv *serv = xprt->xpt_server;
 688        unsigned long required;
 689
 690        /*
 691         * Set the SOCK_NOSPACE flag before checking the available
 692         * sock space.
 693         */
 694        set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
 695        required = atomic_read(&svsk->sk_xprt.xpt_reserved) + serv->sv_max_mesg;
 696        if (required*2 > sock_wspace(svsk->sk_sk))
 697                return 0;
 698        clear_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
 699        return 1;
 700}
 701
 702static struct svc_xprt *svc_udp_accept(struct svc_xprt *xprt)
 703{
 704        BUG();
 705        return NULL;
 706}
 707
 708static struct svc_xprt *svc_udp_create(struct svc_serv *serv,
 709                                       struct net *net,
 710                                       struct sockaddr *sa, int salen,
 711                                       int flags)
 712{
 713        return svc_create_socket(serv, IPPROTO_UDP, net, sa, salen, flags);
 714}
 715
 716static struct svc_xprt_ops svc_udp_ops = {
 717        .xpo_create = svc_udp_create,
 718        .xpo_recvfrom = svc_udp_recvfrom,
 719        .xpo_sendto = svc_udp_sendto,
 720        .xpo_release_rqst = svc_release_skb,
 721        .xpo_detach = svc_sock_detach,
 722        .xpo_free = svc_sock_free,
 723        .xpo_prep_reply_hdr = svc_udp_prep_reply_hdr,
 724        .xpo_has_wspace = svc_udp_has_wspace,
 725        .xpo_accept = svc_udp_accept,
 726};
 727
 728static struct svc_xprt_class svc_udp_class = {
 729        .xcl_name = "udp",
 730        .xcl_owner = THIS_MODULE,
 731        .xcl_ops = &svc_udp_ops,
 732        .xcl_max_payload = RPCSVC_MAXPAYLOAD_UDP,
 733};
 734
 735static void svc_udp_init(struct svc_sock *svsk, struct svc_serv *serv)
 736{
 737        int err, level, optname, one = 1;
 738
 739        svc_xprt_init(sock_net(svsk->sk_sock->sk), &svc_udp_class,
 740                      &svsk->sk_xprt, serv);
 741        clear_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
 742        svsk->sk_sk->sk_data_ready = svc_udp_data_ready;
 743        svsk->sk_sk->sk_write_space = svc_write_space;
 744
 745        /* initialise setting must have enough space to
 746         * receive and respond to one request.
 747         * svc_udp_recvfrom will re-adjust if necessary
 748         */
 749        svc_sock_setbufsize(svsk->sk_sock,
 750                            3 * svsk->sk_xprt.xpt_server->sv_max_mesg,
 751                            3 * svsk->sk_xprt.xpt_server->sv_max_mesg);
 752
 753        /* data might have come in before data_ready set up */
 754        set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
 755        set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
 756
 757        /* make sure we get destination address info */
 758        switch (svsk->sk_sk->sk_family) {
 759        case AF_INET:
 760                level = SOL_IP;
 761                optname = IP_PKTINFO;
 762                break;
 763        case AF_INET6:
 764                level = SOL_IPV6;
 765                optname = IPV6_RECVPKTINFO;
 766                break;
 767        default:
 768                BUG();
 769        }
 770        err = kernel_setsockopt(svsk->sk_sock, level, optname,
 771                                        (char *)&one, sizeof(one));
 772        dprintk("svc: kernel_setsockopt returned %d\n", err);
 773}
 774
 775/*
 776 * A data_ready event on a listening socket means there's a connection
 777 * pending. Do not use state_change as a substitute for it.
 778 */
 779static void svc_tcp_listen_data_ready(struct sock *sk, int count_unused)
 780{
 781        struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
 782        wait_queue_head_t *wq;
 783
 784        dprintk("svc: socket %p TCP (listen) state change %d\n",
 785                sk, sk->sk_state);
 786
 787        /*
 788         * This callback may called twice when a new connection
 789         * is established as a child socket inherits everything
 790         * from a parent LISTEN socket.
 791         * 1) data_ready method of the parent socket will be called
 792         *    when one of child sockets become ESTABLISHED.
 793         * 2) data_ready method of the child socket may be called
 794         *    when it receives data before the socket is accepted.
 795         * In case of 2, we should ignore it silently.
 796         */
 797        if (sk->sk_state == TCP_LISTEN) {
 798                if (svsk) {
 799                        set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
 800                        svc_xprt_enqueue(&svsk->sk_xprt);
 801                } else
 802                        printk("svc: socket %p: no user data\n", sk);
 803        }
 804
 805        wq = sk_sleep(sk);
 806        if (wq && waitqueue_active(wq))
 807                wake_up_interruptible_all(wq);
 808}
 809
 810/*
 811 * A state change on a connected socket means it's dying or dead.
 812 */
 813static void svc_tcp_state_change(struct sock *sk)
 814{
 815        struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
 816        wait_queue_head_t *wq = sk_sleep(sk);
 817
 818        dprintk("svc: socket %p TCP (connected) state change %d (svsk %p)\n",
 819                sk, sk->sk_state, sk->sk_user_data);
 820
 821        if (!svsk)
 822                printk("svc: socket %p: no user data\n", sk);
 823        else {
 824                set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
 825                svc_xprt_enqueue(&svsk->sk_xprt);
 826        }
 827        if (wq && waitqueue_active(wq))
 828                wake_up_interruptible_all(wq);
 829}
 830
 831static void svc_tcp_data_ready(struct sock *sk, int count)
 832{
 833        struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
 834        wait_queue_head_t *wq = sk_sleep(sk);
 835
 836        dprintk("svc: socket %p TCP data ready (svsk %p)\n",
 837                sk, sk->sk_user_data);
 838        if (svsk) {
 839                set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
 840                svc_xprt_enqueue(&svsk->sk_xprt);
 841        }
 842        if (wq && waitqueue_active(wq))
 843                wake_up_interruptible(wq);
 844}
 845
 846/*
 847 * Accept a TCP connection
 848 */
 849static struct svc_xprt *svc_tcp_accept(struct svc_xprt *xprt)
 850{
 851        struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
 852        struct sockaddr_storage addr;
 853        struct sockaddr *sin = (struct sockaddr *) &addr;
 854        struct svc_serv *serv = svsk->sk_xprt.xpt_server;
 855        struct socket   *sock = svsk->sk_sock;
 856        struct socket   *newsock;
 857        struct svc_sock *newsvsk;
 858        int             err, slen;
 859        RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
 860
 861        dprintk("svc: tcp_accept %p sock %p\n", svsk, sock);
 862        if (!sock)
 863                return NULL;
 864
 865        clear_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
 866        err = kernel_accept(sock, &newsock, O_NONBLOCK);
 867        if (err < 0) {
 868                if (err == -ENOMEM)
 869                        printk(KERN_WARNING "%s: no more sockets!\n",
 870                               serv->sv_name);
 871                else if (err != -EAGAIN)
 872                        net_warn_ratelimited("%s: accept failed (err %d)!\n",
 873                                             serv->sv_name, -err);
 874                return NULL;
 875        }
 876        set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
 877
 878        err = kernel_getpeername(newsock, sin, &slen);
 879        if (err < 0) {
 880                net_warn_ratelimited("%s: peername failed (err %d)!\n",
 881                                     serv->sv_name, -err);
 882                goto failed;            /* aborted connection or whatever */
 883        }
 884
 885        /* Ideally, we would want to reject connections from unauthorized
 886         * hosts here, but when we get encryption, the IP of the host won't
 887         * tell us anything.  For now just warn about unpriv connections.
 888         */
 889        if (!svc_port_is_privileged(sin)) {
 890                dprintk(KERN_WARNING
 891                        "%s: connect from unprivileged port: %s\n",
 892                        serv->sv_name,
 893                        __svc_print_addr(sin, buf, sizeof(buf)));
 894        }
 895        dprintk("%s: connect from %s\n", serv->sv_name,
 896                __svc_print_addr(sin, buf, sizeof(buf)));
 897
 898        /* make sure that a write doesn't block forever when
 899         * low on memory
 900         */
 901        newsock->sk->sk_sndtimeo = HZ*30;
 902
 903        if (!(newsvsk = svc_setup_socket(serv, newsock, &err,
 904                                 (SVC_SOCK_ANONYMOUS | SVC_SOCK_TEMPORARY))))
 905                goto failed;
 906        svc_xprt_set_remote(&newsvsk->sk_xprt, sin, slen);
 907        err = kernel_getsockname(newsock, sin, &slen);
 908        if (unlikely(err < 0)) {
 909                dprintk("svc_tcp_accept: kernel_getsockname error %d\n", -err);
 910                slen = offsetof(struct sockaddr, sa_data);
 911        }
 912        svc_xprt_set_local(&newsvsk->sk_xprt, sin, slen);
 913
 914        if (serv->sv_stats)
 915                serv->sv_stats->nettcpconn++;
 916
 917        return &newsvsk->sk_xprt;
 918
 919failed:
 920        sock_release(newsock);
 921        return NULL;
 922}
 923
 924static unsigned int svc_tcp_restore_pages(struct svc_sock *svsk, struct svc_rqst *rqstp)
 925{
 926        unsigned int i, len, npages;
 927
 928        if (svsk->sk_tcplen <= sizeof(rpc_fraghdr))
 929                return 0;
 930        len = svsk->sk_tcplen - sizeof(rpc_fraghdr);
 931        npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
 932        for (i = 0; i < npages; i++) {
 933                if (rqstp->rq_pages[i] != NULL)
 934                        put_page(rqstp->rq_pages[i]);
 935                BUG_ON(svsk->sk_pages[i] == NULL);
 936                rqstp->rq_pages[i] = svsk->sk_pages[i];
 937                svsk->sk_pages[i] = NULL;
 938        }
 939        rqstp->rq_arg.head[0].iov_base = page_address(rqstp->rq_pages[0]);
 940        return len;
 941}
 942
 943static void svc_tcp_save_pages(struct svc_sock *svsk, struct svc_rqst *rqstp)
 944{
 945        unsigned int i, len, npages;
 946
 947        if (svsk->sk_tcplen <= sizeof(rpc_fraghdr))
 948                return;
 949        len = svsk->sk_tcplen - sizeof(rpc_fraghdr);
 950        npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
 951        for (i = 0; i < npages; i++) {
 952                svsk->sk_pages[i] = rqstp->rq_pages[i];
 953                rqstp->rq_pages[i] = NULL;
 954        }
 955}
 956
 957static void svc_tcp_clear_pages(struct svc_sock *svsk)
 958{
 959        unsigned int i, len, npages;
 960
 961        if (svsk->sk_tcplen <= sizeof(rpc_fraghdr))
 962                goto out;
 963        len = svsk->sk_tcplen - sizeof(rpc_fraghdr);
 964        npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
 965        for (i = 0; i < npages; i++) {
 966                BUG_ON(svsk->sk_pages[i] == NULL);
 967                put_page(svsk->sk_pages[i]);
 968                svsk->sk_pages[i] = NULL;
 969        }
 970out:
 971        svsk->sk_tcplen = 0;
 972}
 973
 974/*
 975 * Receive data.
 976 * If we haven't gotten the record length yet, get the next four bytes.
 977 * Otherwise try to gobble up as much as possible up to the complete
 978 * record length.
 979 */
 980static int svc_tcp_recv_record(struct svc_sock *svsk, struct svc_rqst *rqstp)
 981{
 982        struct svc_serv *serv = svsk->sk_xprt.xpt_server;
 983        unsigned int want;
 984        int len;
 985
 986        clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
 987
 988        if (svsk->sk_tcplen < sizeof(rpc_fraghdr)) {
 989                struct kvec     iov;
 990
 991                want = sizeof(rpc_fraghdr) - svsk->sk_tcplen;
 992                iov.iov_base = ((char *) &svsk->sk_reclen) + svsk->sk_tcplen;
 993                iov.iov_len  = want;
 994                if ((len = svc_recvfrom(rqstp, &iov, 1, want)) < 0)
 995                        goto error;
 996                svsk->sk_tcplen += len;
 997
 998                if (len < want) {
 999                        dprintk("svc: short recvfrom while reading record "
1000                                "length (%d of %d)\n", len, want);
1001                        return -EAGAIN;
1002                }
1003
1004                svsk->sk_reclen = ntohl(svsk->sk_reclen);
1005                if (!(svsk->sk_reclen & RPC_LAST_STREAM_FRAGMENT)) {
1006                        /* FIXME: technically, a record can be fragmented,
1007                         *  and non-terminal fragments will not have the top
1008                         *  bit set in the fragment length header.
1009                         *  But apparently no known nfs clients send fragmented
1010                         *  records. */
1011                        net_notice_ratelimited("RPC: multiple fragments per record not supported\n");
1012                        goto err_delete;
1013                }
1014
1015                svsk->sk_reclen &= RPC_FRAGMENT_SIZE_MASK;
1016                dprintk("svc: TCP record, %d bytes\n", svsk->sk_reclen);
1017                if (svsk->sk_reclen > serv->sv_max_mesg) {
1018                        net_notice_ratelimited("RPC: fragment too large: 0x%08lx\n",
1019                                               (unsigned long)svsk->sk_reclen);
1020                        goto err_delete;
1021                }
1022        }
1023
1024        if (svsk->sk_reclen < 8)
1025                goto err_delete; /* client is nuts. */
1026
1027        len = svsk->sk_reclen;
1028
1029        return len;
1030error:
1031        dprintk("RPC: TCP recv_record got %d\n", len);
1032        return len;
1033err_delete:
1034        set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
1035        return -EAGAIN;
1036}
1037
1038static int receive_cb_reply(struct svc_sock *svsk, struct svc_rqst *rqstp)
1039{
1040        struct rpc_xprt *bc_xprt = svsk->sk_xprt.xpt_bc_xprt;
1041        struct rpc_rqst *req = NULL;
1042        struct kvec *src, *dst;
1043        __be32 *p = (__be32 *)rqstp->rq_arg.head[0].iov_base;
1044        __be32 xid;
1045        __be32 calldir;
1046
1047        xid = *p++;
1048        calldir = *p;
1049
1050        if (bc_xprt)
1051                req = xprt_lookup_rqst(bc_xprt, xid);
1052
1053        if (!req) {
1054                printk(KERN_NOTICE
1055                        "%s: Got unrecognized reply: "
1056                        "calldir 0x%x xpt_bc_xprt %p xid %08x\n",
1057                        __func__, ntohl(calldir),
1058                        bc_xprt, xid);
1059                return -EAGAIN;
1060        }
1061
1062        memcpy(&req->rq_private_buf, &req->rq_rcv_buf, sizeof(struct xdr_buf));
1063        /*
1064         * XXX!: cheating for now!  Only copying HEAD.
1065         * But we know this is good enough for now (in fact, for any
1066         * callback reply in the forseeable future).
1067         */
1068        dst = &req->rq_private_buf.head[0];
1069        src = &rqstp->rq_arg.head[0];
1070        if (dst->iov_len < src->iov_len)
1071                return -EAGAIN; /* whatever; just giving up. */
1072        memcpy(dst->iov_base, src->iov_base, src->iov_len);
1073        xprt_complete_rqst(req->rq_task, svsk->sk_reclen);
1074        rqstp->rq_arg.len = 0;
1075        return 0;
1076}
1077
1078static int copy_pages_to_kvecs(struct kvec *vec, struct page **pages, int len)
1079{
1080        int i = 0;
1081        int t = 0;
1082
1083        while (t < len) {
1084                vec[i].iov_base = page_address(pages[i]);
1085                vec[i].iov_len = PAGE_SIZE;
1086                i++;
1087                t += PAGE_SIZE;
1088        }
1089        return i;
1090}
1091
1092
1093/*
1094 * Receive data from a TCP socket.
1095 */
1096static int svc_tcp_recvfrom(struct svc_rqst *rqstp)
1097{
1098        struct svc_sock *svsk =
1099                container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
1100        struct svc_serv *serv = svsk->sk_xprt.xpt_server;
1101        int             len;
1102        struct kvec *vec;
1103        unsigned int want, base;
1104        __be32 *p;
1105        __be32 calldir;
1106        int pnum;
1107
1108        dprintk("svc: tcp_recv %p data %d conn %d close %d\n",
1109                svsk, test_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags),
1110                test_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags),
1111                test_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags));
1112
1113        len = svc_tcp_recv_record(svsk, rqstp);
1114        if (len < 0)
1115                goto error;
1116
1117        base = svc_tcp_restore_pages(svsk, rqstp);
1118        want = svsk->sk_reclen - base;
1119
1120        vec = rqstp->rq_vec;
1121
1122        pnum = copy_pages_to_kvecs(&vec[0], &rqstp->rq_pages[0],
1123                                                svsk->sk_reclen);
1124
1125        rqstp->rq_respages = &rqstp->rq_pages[pnum];
1126
1127        /* Now receive data */
1128        len = svc_partial_recvfrom(rqstp, vec, pnum, want, base);
1129        if (len >= 0)
1130                svsk->sk_tcplen += len;
1131        if (len != want) {
1132                svc_tcp_save_pages(svsk, rqstp);
1133                if (len < 0 && len != -EAGAIN)
1134                        goto err_other;
1135                dprintk("svc: incomplete TCP record (%d of %d)\n",
1136                        svsk->sk_tcplen, svsk->sk_reclen);
1137                goto err_noclose;
1138        }
1139
1140        rqstp->rq_arg.len = svsk->sk_reclen;
1141        rqstp->rq_arg.page_base = 0;
1142        if (rqstp->rq_arg.len <= rqstp->rq_arg.head[0].iov_len) {
1143                rqstp->rq_arg.head[0].iov_len = rqstp->rq_arg.len;
1144                rqstp->rq_arg.page_len = 0;
1145        } else
1146                rqstp->rq_arg.page_len = rqstp->rq_arg.len - rqstp->rq_arg.head[0].iov_len;
1147
1148        rqstp->rq_xprt_ctxt   = NULL;
1149        rqstp->rq_prot        = IPPROTO_TCP;
1150
1151        p = (__be32 *)rqstp->rq_arg.head[0].iov_base;
1152        calldir = p[1];
1153        if (calldir)
1154                len = receive_cb_reply(svsk, rqstp);
1155
1156        /* Reset TCP read info */
1157        svsk->sk_reclen = 0;
1158        svsk->sk_tcplen = 0;
1159        /* If we have more data, signal svc_xprt_enqueue() to try again */
1160        if (svc_recv_available(svsk) > sizeof(rpc_fraghdr))
1161                set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
1162
1163        if (len < 0)
1164                goto error;
1165
1166        svc_xprt_copy_addrs(rqstp, &svsk->sk_xprt);
1167        if (serv->sv_stats)
1168                serv->sv_stats->nettcpcnt++;
1169
1170        dprintk("svc: TCP complete record (%d bytes)\n", rqstp->rq_arg.len);
1171        return rqstp->rq_arg.len;
1172
1173error:
1174        if (len != -EAGAIN)
1175                goto err_other;
1176        dprintk("RPC: TCP recvfrom got EAGAIN\n");
1177        return -EAGAIN;
1178err_other:
1179        printk(KERN_NOTICE "%s: recvfrom returned errno %d\n",
1180               svsk->sk_xprt.xpt_server->sv_name, -len);
1181        set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
1182err_noclose:
1183        return -EAGAIN; /* record not complete */
1184}
1185
1186/*
1187 * Send out data on TCP socket.
1188 */
1189static int svc_tcp_sendto(struct svc_rqst *rqstp)
1190{
1191        struct xdr_buf  *xbufp = &rqstp->rq_res;
1192        int sent;
1193        __be32 reclen;
1194
1195        /* Set up the first element of the reply kvec.
1196         * Any other kvecs that may be in use have been taken
1197         * care of by the server implementation itself.
1198         */
1199        reclen = htonl(0x80000000|((xbufp->len ) - 4));
1200        memcpy(xbufp->head[0].iov_base, &reclen, 4);
1201
1202        sent = svc_sendto(rqstp, &rqstp->rq_res);
1203        if (sent != xbufp->len) {
1204                printk(KERN_NOTICE
1205                       "rpc-srv/tcp: %s: %s %d when sending %d bytes "
1206                       "- shutting down socket\n",
1207                       rqstp->rq_xprt->xpt_server->sv_name,
1208                       (sent<0)?"got error":"sent only",
1209                       sent, xbufp->len);
1210                set_bit(XPT_CLOSE, &rqstp->rq_xprt->xpt_flags);
1211                svc_xprt_enqueue(rqstp->rq_xprt);
1212                sent = -EAGAIN;
1213        }
1214        return sent;
1215}
1216
1217/*
1218 * Setup response header. TCP has a 4B record length field.
1219 */
1220static void svc_tcp_prep_reply_hdr(struct svc_rqst *rqstp)
1221{
1222        struct kvec *resv = &rqstp->rq_res.head[0];
1223
1224        /* tcp needs a space for the record length... */
1225        svc_putnl(resv, 0);
1226}
1227
1228static int svc_tcp_has_wspace(struct svc_xprt *xprt)
1229{
1230        struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1231        struct svc_serv *serv = svsk->sk_xprt.xpt_server;
1232        int required;
1233
1234        if (test_bit(XPT_LISTENER, &xprt->xpt_flags))
1235                return 1;
1236        required = atomic_read(&xprt->xpt_reserved) + serv->sv_max_mesg;
1237        if (sk_stream_wspace(svsk->sk_sk) >= required)
1238                return 1;
1239        set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
1240        return 0;
1241}
1242
1243static struct svc_xprt *svc_tcp_create(struct svc_serv *serv,
1244                                       struct net *net,
1245                                       struct sockaddr *sa, int salen,
1246                                       int flags)
1247{
1248        return svc_create_socket(serv, IPPROTO_TCP, net, sa, salen, flags);
1249}
1250
1251#if defined(CONFIG_SUNRPC_BACKCHANNEL)
1252static struct svc_xprt *svc_bc_create_socket(struct svc_serv *, int,
1253                                             struct net *, struct sockaddr *,
1254                                             int, int);
1255static void svc_bc_sock_free(struct svc_xprt *xprt);
1256
1257static struct svc_xprt *svc_bc_tcp_create(struct svc_serv *serv,
1258                                       struct net *net,
1259                                       struct sockaddr *sa, int salen,
1260                                       int flags)
1261{
1262        return svc_bc_create_socket(serv, IPPROTO_TCP, net, sa, salen, flags);
1263}
1264
1265static void svc_bc_tcp_sock_detach(struct svc_xprt *xprt)
1266{
1267}
1268
1269static struct svc_xprt_ops svc_tcp_bc_ops = {
1270        .xpo_create = svc_bc_tcp_create,
1271        .xpo_detach = svc_bc_tcp_sock_detach,
1272        .xpo_free = svc_bc_sock_free,
1273        .xpo_prep_reply_hdr = svc_tcp_prep_reply_hdr,
1274};
1275
1276static struct svc_xprt_class svc_tcp_bc_class = {
1277        .xcl_name = "tcp-bc",
1278        .xcl_owner = THIS_MODULE,
1279        .xcl_ops = &svc_tcp_bc_ops,
1280        .xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP,
1281};
1282
1283static void svc_init_bc_xprt_sock(void)
1284{
1285        svc_reg_xprt_class(&svc_tcp_bc_class);
1286}
1287
1288static void svc_cleanup_bc_xprt_sock(void)
1289{
1290        svc_unreg_xprt_class(&svc_tcp_bc_class);
1291}
1292#else /* CONFIG_SUNRPC_BACKCHANNEL */
1293static void svc_init_bc_xprt_sock(void)
1294{
1295}
1296
1297static void svc_cleanup_bc_xprt_sock(void)
1298{
1299}
1300#endif /* CONFIG_SUNRPC_BACKCHANNEL */
1301
1302static struct svc_xprt_ops svc_tcp_ops = {
1303        .xpo_create = svc_tcp_create,
1304        .xpo_recvfrom = svc_tcp_recvfrom,
1305        .xpo_sendto = svc_tcp_sendto,
1306        .xpo_release_rqst = svc_release_skb,
1307        .xpo_detach = svc_tcp_sock_detach,
1308        .xpo_free = svc_sock_free,
1309        .xpo_prep_reply_hdr = svc_tcp_prep_reply_hdr,
1310        .xpo_has_wspace = svc_tcp_has_wspace,
1311        .xpo_accept = svc_tcp_accept,
1312};
1313
1314static struct svc_xprt_class svc_tcp_class = {
1315        .xcl_name = "tcp",
1316        .xcl_owner = THIS_MODULE,
1317        .xcl_ops = &svc_tcp_ops,
1318        .xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP,
1319};
1320
1321void svc_init_xprt_sock(void)
1322{
1323        svc_reg_xprt_class(&svc_tcp_class);
1324        svc_reg_xprt_class(&svc_udp_class);
1325        svc_init_bc_xprt_sock();
1326}
1327
1328void svc_cleanup_xprt_sock(void)
1329{
1330        svc_unreg_xprt_class(&svc_tcp_class);
1331        svc_unreg_xprt_class(&svc_udp_class);
1332        svc_cleanup_bc_xprt_sock();
1333}
1334
1335static void svc_tcp_init(struct svc_sock *svsk, struct svc_serv *serv)
1336{
1337        struct sock     *sk = svsk->sk_sk;
1338
1339        svc_xprt_init(sock_net(svsk->sk_sock->sk), &svc_tcp_class,
1340                      &svsk->sk_xprt, serv);
1341        set_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
1342        if (sk->sk_state == TCP_LISTEN) {
1343                dprintk("setting up TCP socket for listening\n");
1344                set_bit(XPT_LISTENER, &svsk->sk_xprt.xpt_flags);
1345                sk->sk_data_ready = svc_tcp_listen_data_ready;
1346                set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
1347        } else {
1348                dprintk("setting up TCP socket for reading\n");
1349                sk->sk_state_change = svc_tcp_state_change;
1350                sk->sk_data_ready = svc_tcp_data_ready;
1351                sk->sk_write_space = svc_tcp_write_space;
1352
1353                svsk->sk_reclen = 0;
1354                svsk->sk_tcplen = 0;
1355                memset(&svsk->sk_pages[0], 0, sizeof(svsk->sk_pages));
1356
1357                tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
1358
1359                set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
1360                if (sk->sk_state != TCP_ESTABLISHED)
1361                        set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
1362        }
1363}
1364
1365void svc_sock_update_bufs(struct svc_serv *serv)
1366{
1367        /*
1368         * The number of server threads has changed. Update
1369         * rcvbuf and sndbuf accordingly on all sockets
1370         */
1371        struct svc_sock *svsk;
1372
1373        spin_lock_bh(&serv->sv_lock);
1374        list_for_each_entry(svsk, &serv->sv_permsocks, sk_xprt.xpt_list)
1375                set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
1376        spin_unlock_bh(&serv->sv_lock);
1377}
1378EXPORT_SYMBOL_GPL(svc_sock_update_bufs);
1379
1380/*
1381 * Initialize socket for RPC use and create svc_sock struct
1382 * XXX: May want to setsockopt SO_SNDBUF and SO_RCVBUF.
1383 */
1384static struct svc_sock *svc_setup_socket(struct svc_serv *serv,
1385                                                struct socket *sock,
1386                                                int *errp, int flags)
1387{
1388        struct svc_sock *svsk;
1389        struct sock     *inet;
1390        int             pmap_register = !(flags & SVC_SOCK_ANONYMOUS);
1391
1392        dprintk("svc: svc_setup_socket %p\n", sock);
1393        if (!(svsk = kzalloc(sizeof(*svsk), GFP_KERNEL))) {
1394                *errp = -ENOMEM;
1395                return NULL;
1396        }
1397
1398        inet = sock->sk;
1399
1400        /* Register socket with portmapper */
1401        if (*errp >= 0 && pmap_register)
1402                *errp = svc_register(serv, sock_net(sock->sk), inet->sk_family,
1403                                     inet->sk_protocol,
1404                                     ntohs(inet_sk(inet)->inet_sport));
1405
1406        if (*errp < 0) {
1407                kfree(svsk);
1408                return NULL;
1409        }
1410
1411        inet->sk_user_data = svsk;
1412        svsk->sk_sock = sock;
1413        svsk->sk_sk = inet;
1414        svsk->sk_ostate = inet->sk_state_change;
1415        svsk->sk_odata = inet->sk_data_ready;
1416        svsk->sk_owspace = inet->sk_write_space;
1417
1418        /* Initialize the socket */
1419        if (sock->type == SOCK_DGRAM)
1420                svc_udp_init(svsk, serv);
1421        else {
1422                /* initialise setting must have enough space to
1423                 * receive and respond to one request.
1424                 */
1425                svc_sock_setbufsize(svsk->sk_sock, 4 * serv->sv_max_mesg,
1426                                        4 * serv->sv_max_mesg);
1427                svc_tcp_init(svsk, serv);
1428        }
1429
1430        dprintk("svc: svc_setup_socket created %p (inet %p)\n",
1431                                svsk, svsk->sk_sk);
1432
1433        return svsk;
1434}
1435
1436/**
1437 * svc_addsock - add a listener socket to an RPC service
1438 * @serv: pointer to RPC service to which to add a new listener
1439 * @fd: file descriptor of the new listener
1440 * @name_return: pointer to buffer to fill in with name of listener
1441 * @len: size of the buffer
1442 *
1443 * Fills in socket name and returns positive length of name if successful.
1444 * Name is terminated with '\n'.  On error, returns a negative errno
1445 * value.
1446 */
1447int svc_addsock(struct svc_serv *serv, const int fd, char *name_return,
1448                const size_t len)
1449{
1450        int err = 0;
1451        struct socket *so = sockfd_lookup(fd, &err);
1452        struct svc_sock *svsk = NULL;
1453
1454        if (!so)
1455                return err;
1456        if ((so->sk->sk_family != PF_INET) && (so->sk->sk_family != PF_INET6))
1457                err =  -EAFNOSUPPORT;
1458        else if (so->sk->sk_protocol != IPPROTO_TCP &&
1459            so->sk->sk_protocol != IPPROTO_UDP)
1460                err =  -EPROTONOSUPPORT;
1461        else if (so->state > SS_UNCONNECTED)
1462                err = -EISCONN;
1463        else {
1464                if (!try_module_get(THIS_MODULE))
1465                        err = -ENOENT;
1466                else
1467                        svsk = svc_setup_socket(serv, so, &err,
1468                                                SVC_SOCK_DEFAULTS);
1469                if (svsk) {
1470                        struct sockaddr_storage addr;
1471                        struct sockaddr *sin = (struct sockaddr *)&addr;
1472                        int salen;
1473                        if (kernel_getsockname(svsk->sk_sock, sin, &salen) == 0)
1474                                svc_xprt_set_local(&svsk->sk_xprt, sin, salen);
1475                        clear_bit(XPT_TEMP, &svsk->sk_xprt.xpt_flags);
1476                        spin_lock_bh(&serv->sv_lock);
1477                        list_add(&svsk->sk_xprt.xpt_list, &serv->sv_permsocks);
1478                        spin_unlock_bh(&serv->sv_lock);
1479                        svc_xprt_received(&svsk->sk_xprt);
1480                        err = 0;
1481                } else
1482                        module_put(THIS_MODULE);
1483        }
1484        if (err) {
1485                sockfd_put(so);
1486                return err;
1487        }
1488        return svc_one_sock_name(svsk, name_return, len);
1489}
1490EXPORT_SYMBOL_GPL(svc_addsock);
1491
1492/*
1493 * Create socket for RPC service.
1494 */
1495static struct svc_xprt *svc_create_socket(struct svc_serv *serv,
1496                                          int protocol,
1497                                          struct net *net,
1498                                          struct sockaddr *sin, int len,
1499                                          int flags)
1500{
1501        struct svc_sock *svsk;
1502        struct socket   *sock;
1503        int             error;
1504        int             type;
1505        struct sockaddr_storage addr;
1506        struct sockaddr *newsin = (struct sockaddr *)&addr;
1507        int             newlen;
1508        int             family;
1509        int             val;
1510        RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
1511
1512        dprintk("svc: svc_create_socket(%s, %d, %s)\n",
1513                        serv->sv_program->pg_name, protocol,
1514                        __svc_print_addr(sin, buf, sizeof(buf)));
1515
1516        if (protocol != IPPROTO_UDP && protocol != IPPROTO_TCP) {
1517                printk(KERN_WARNING "svc: only UDP and TCP "
1518                                "sockets supported\n");
1519                return ERR_PTR(-EINVAL);
1520        }
1521
1522        type = (protocol == IPPROTO_UDP)? SOCK_DGRAM : SOCK_STREAM;
1523        switch (sin->sa_family) {
1524        case AF_INET6:
1525                family = PF_INET6;
1526                break;
1527        case AF_INET:
1528                family = PF_INET;
1529                break;
1530        default:
1531                return ERR_PTR(-EINVAL);
1532        }
1533
1534        error = __sock_create(net, family, type, protocol, &sock, 1);
1535        if (error < 0)
1536                return ERR_PTR(error);
1537
1538        svc_reclassify_socket(sock);
1539
1540        /*
1541         * If this is an PF_INET6 listener, we want to avoid
1542         * getting requests from IPv4 remotes.  Those should
1543         * be shunted to a PF_INET listener via rpcbind.
1544         */
1545        val = 1;
1546        if (family == PF_INET6)
1547                kernel_setsockopt(sock, SOL_IPV6, IPV6_V6ONLY,
1548                                        (char *)&val, sizeof(val));
1549
1550        if (type == SOCK_STREAM)
1551                sock->sk->sk_reuse = SK_CAN_REUSE; /* allow address reuse */
1552        error = kernel_bind(sock, sin, len);
1553        if (error < 0)
1554                goto bummer;
1555
1556        newlen = len;
1557        error = kernel_getsockname(sock, newsin, &newlen);
1558        if (error < 0)
1559                goto bummer;
1560
1561        if (protocol == IPPROTO_TCP) {
1562                if ((error = kernel_listen(sock, 64)) < 0)
1563                        goto bummer;
1564        }
1565
1566        if ((svsk = svc_setup_socket(serv, sock, &error, flags)) != NULL) {
1567                svc_xprt_set_local(&svsk->sk_xprt, newsin, newlen);
1568                return (struct svc_xprt *)svsk;
1569        }
1570
1571bummer:
1572        dprintk("svc: svc_create_socket error = %d\n", -error);
1573        sock_release(sock);
1574        return ERR_PTR(error);
1575}
1576
1577/*
1578 * Detach the svc_sock from the socket so that no
1579 * more callbacks occur.
1580 */
1581static void svc_sock_detach(struct svc_xprt *xprt)
1582{
1583        struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1584        struct sock *sk = svsk->sk_sk;
1585        wait_queue_head_t *wq;
1586
1587        dprintk("svc: svc_sock_detach(%p)\n", svsk);
1588
1589        /* put back the old socket callbacks */
1590        sk->sk_state_change = svsk->sk_ostate;
1591        sk->sk_data_ready = svsk->sk_odata;
1592        sk->sk_write_space = svsk->sk_owspace;
1593
1594        wq = sk_sleep(sk);
1595        if (wq && waitqueue_active(wq))
1596                wake_up_interruptible(wq);
1597}
1598
1599/*
1600 * Disconnect the socket, and reset the callbacks
1601 */
1602static void svc_tcp_sock_detach(struct svc_xprt *xprt)
1603{
1604        struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1605
1606        dprintk("svc: svc_tcp_sock_detach(%p)\n", svsk);
1607
1608        svc_sock_detach(xprt);
1609
1610        if (!test_bit(XPT_LISTENER, &xprt->xpt_flags)) {
1611                svc_tcp_clear_pages(svsk);
1612                kernel_sock_shutdown(svsk->sk_sock, SHUT_RDWR);
1613        }
1614}
1615
1616/*
1617 * Free the svc_sock's socket resources and the svc_sock itself.
1618 */
1619static void svc_sock_free(struct svc_xprt *xprt)
1620{
1621        struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1622        dprintk("svc: svc_sock_free(%p)\n", svsk);
1623
1624        if (svsk->sk_sock->file)
1625                sockfd_put(svsk->sk_sock);
1626        else
1627                sock_release(svsk->sk_sock);
1628        kfree(svsk);
1629}
1630
1631#if defined(CONFIG_SUNRPC_BACKCHANNEL)
1632/*
1633 * Create a back channel svc_xprt which shares the fore channel socket.
1634 */
1635static struct svc_xprt *svc_bc_create_socket(struct svc_serv *serv,
1636                                             int protocol,
1637                                             struct net *net,
1638                                             struct sockaddr *sin, int len,
1639                                             int flags)
1640{
1641        struct svc_sock *svsk;
1642        struct svc_xprt *xprt;
1643
1644        if (protocol != IPPROTO_TCP) {
1645                printk(KERN_WARNING "svc: only TCP sockets"
1646                        " supported on shared back channel\n");
1647                return ERR_PTR(-EINVAL);
1648        }
1649
1650        svsk = kzalloc(sizeof(*svsk), GFP_KERNEL);
1651        if (!svsk)
1652                return ERR_PTR(-ENOMEM);
1653
1654        xprt = &svsk->sk_xprt;
1655        svc_xprt_init(net, &svc_tcp_bc_class, xprt, serv);
1656
1657        serv->sv_bc_xprt = xprt;
1658
1659        return xprt;
1660}
1661
1662/*
1663 * Free a back channel svc_sock.
1664 */
1665static void svc_bc_sock_free(struct svc_xprt *xprt)
1666{
1667        if (xprt)
1668                kfree(container_of(xprt, struct svc_sock, sk_xprt));
1669}
1670#endif /* CONFIG_SUNRPC_BACKCHANNEL */
1671
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.