linux/net/sunrpc/svc_xprt.c
<<
>>
Prefs
   1/*
   2 * linux/net/sunrpc/svc_xprt.c
   3 *
   4 * Author: Tom Tucker <tom@opengridcomputing.com>
   5 */
   6
   7#include <linux/sched.h>
   8#include <linux/errno.h>
   9#include <linux/freezer.h>
  10#include <linux/kthread.h>
  11#include <linux/slab.h>
  12#include <net/sock.h>
  13#include <linux/sunrpc/stats.h>
  14#include <linux/sunrpc/svc_xprt.h>
  15#include <linux/sunrpc/svcsock.h>
  16#include <linux/sunrpc/xprt.h>
  17#include <linux/module.h>
  18
  19#define RPCDBG_FACILITY RPCDBG_SVCXPRT
  20
  21static struct svc_deferred_req *svc_deferred_dequeue(struct svc_xprt *xprt);
  22static int svc_deferred_recv(struct svc_rqst *rqstp);
  23static struct cache_deferred_req *svc_defer(struct cache_req *req);
  24static void svc_age_temp_xprts(unsigned long closure);
  25static void svc_delete_xprt(struct svc_xprt *xprt);
  26
  27/* apparently the "standard" is that clients close
  28 * idle connections after 5 minutes, servers after
  29 * 6 minutes
  30 *   http://www.connectathon.org/talks96/nfstcp.pdf
  31 */
  32static int svc_conn_age_period = 6*60;
  33
  34/* List of registered transport classes */
  35static DEFINE_SPINLOCK(svc_xprt_class_lock);
  36static LIST_HEAD(svc_xprt_class_list);
  37
  38/* SMP locking strategy:
  39 *
  40 *      svc_pool->sp_lock protects most of the fields of that pool.
  41 *      svc_serv->sv_lock protects sv_tempsocks, sv_permsocks, sv_tmpcnt.
  42 *      when both need to be taken (rare), svc_serv->sv_lock is first.
  43 *      BKL protects svc_serv->sv_nrthread.
  44 *      svc_sock->sk_lock protects the svc_sock->sk_deferred list
  45 *             and the ->sk_info_authunix cache.
  46 *
  47 *      The XPT_BUSY bit in xprt->xpt_flags prevents a transport being
  48 *      enqueued multiply. During normal transport processing this bit
  49 *      is set by svc_xprt_enqueue and cleared by svc_xprt_received.
  50 *      Providers should not manipulate this bit directly.
  51 *
  52 *      Some flags can be set to certain values at any time
  53 *      providing that certain rules are followed:
  54 *
  55 *      XPT_CONN, XPT_DATA:
  56 *              - Can be set or cleared at any time.
  57 *              - After a set, svc_xprt_enqueue must be called to enqueue
  58 *                the transport for processing.
  59 *              - After a clear, the transport must be read/accepted.
  60 *                If this succeeds, it must be set again.
  61 *      XPT_CLOSE:
  62 *              - Can set at any time. It is never cleared.
  63 *      XPT_DEAD:
  64 *              - Can only be set while XPT_BUSY is held which ensures
  65 *                that no other thread will be using the transport or will
  66 *                try to set XPT_DEAD.
  67 */
  68
  69int svc_reg_xprt_class(struct svc_xprt_class *xcl)
  70{
  71        struct svc_xprt_class *cl;
  72        int res = -EEXIST;
  73
  74        dprintk("svc: Adding svc transport class '%s'\n", xcl->xcl_name);
  75
  76        INIT_LIST_HEAD(&xcl->xcl_list);
  77        spin_lock(&svc_xprt_class_lock);
  78        /* Make sure there isn't already a class with the same name */
  79        list_for_each_entry(cl, &svc_xprt_class_list, xcl_list) {
  80                if (strcmp(xcl->xcl_name, cl->xcl_name) == 0)
  81                        goto out;
  82        }
  83        list_add_tail(&xcl->xcl_list, &svc_xprt_class_list);
  84        res = 0;
  85out:
  86        spin_unlock(&svc_xprt_class_lock);
  87        return res;
  88}
  89EXPORT_SYMBOL_GPL(svc_reg_xprt_class);
  90
  91void svc_unreg_xprt_class(struct svc_xprt_class *xcl)
  92{
  93        dprintk("svc: Removing svc transport class '%s'\n", xcl->xcl_name);
  94        spin_lock(&svc_xprt_class_lock);
  95        list_del_init(&xcl->xcl_list);
  96        spin_unlock(&svc_xprt_class_lock);
  97}
  98EXPORT_SYMBOL_GPL(svc_unreg_xprt_class);
  99
 100/*
 101 * Format the transport list for printing
 102 */
 103int svc_print_xprts(char *buf, int maxlen)
 104{
 105        struct svc_xprt_class *xcl;
 106        char tmpstr[80];
 107        int len = 0;
 108        buf[0] = '\0';
 109
 110        spin_lock(&svc_xprt_class_lock);
 111        list_for_each_entry(xcl, &svc_xprt_class_list, xcl_list) {
 112                int slen;
 113
 114                sprintf(tmpstr, "%s %d\n", xcl->xcl_name, xcl->xcl_max_payload);
 115                slen = strlen(tmpstr);
 116                if (len + slen > maxlen)
 117                        break;
 118                len += slen;
 119                strcat(buf, tmpstr);
 120        }
 121        spin_unlock(&svc_xprt_class_lock);
 122
 123        return len;
 124}
 125
 126static void svc_xprt_free(struct kref *kref)
 127{
 128        struct svc_xprt *xprt =
 129                container_of(kref, struct svc_xprt, xpt_ref);
 130        struct module *owner = xprt->xpt_class->xcl_owner;
 131        if (test_bit(XPT_CACHE_AUTH, &xprt->xpt_flags))
 132                svcauth_unix_info_release(xprt);
 133        put_net(xprt->xpt_net);
 134        /* See comment on corresponding get in xs_setup_bc_tcp(): */
 135        if (xprt->xpt_bc_xprt)
 136                xprt_put(xprt->xpt_bc_xprt);
 137        xprt->xpt_ops->xpo_free(xprt);
 138        module_put(owner);
 139}
 140
 141void svc_xprt_put(struct svc_xprt *xprt)
 142{
 143        kref_put(&xprt->xpt_ref, svc_xprt_free);
 144}
 145EXPORT_SYMBOL_GPL(svc_xprt_put);
 146
 147/*
 148 * Called by transport drivers to initialize the transport independent
 149 * portion of the transport instance.
 150 */
 151void svc_xprt_init(struct net *net, struct svc_xprt_class *xcl,
 152                   struct svc_xprt *xprt, struct svc_serv *serv)
 153{
 154        memset(xprt, 0, sizeof(*xprt));
 155        xprt->xpt_class = xcl;
 156        xprt->xpt_ops = xcl->xcl_ops;
 157        kref_init(&xprt->xpt_ref);
 158        xprt->xpt_server = serv;
 159        INIT_LIST_HEAD(&xprt->xpt_list);
 160        INIT_LIST_HEAD(&xprt->xpt_ready);
 161        INIT_LIST_HEAD(&xprt->xpt_deferred);
 162        INIT_LIST_HEAD(&xprt->xpt_users);
 163        mutex_init(&xprt->xpt_mutex);
 164        spin_lock_init(&xprt->xpt_lock);
 165        set_bit(XPT_BUSY, &xprt->xpt_flags);
 166        rpc_init_wait_queue(&xprt->xpt_bc_pending, "xpt_bc_pending");
 167        xprt->xpt_net = get_net(net);
 168}
 169EXPORT_SYMBOL_GPL(svc_xprt_init);
 170
 171static struct svc_xprt *__svc_xpo_create(struct svc_xprt_class *xcl,
 172                                         struct svc_serv *serv,
 173                                         struct net *net,
 174                                         const int family,
 175                                         const unsigned short port,
 176                                         int flags)
 177{
 178        struct sockaddr_in sin = {
 179                .sin_family             = AF_INET,
 180                .sin_addr.s_addr        = htonl(INADDR_ANY),
 181                .sin_port               = htons(port),
 182        };
 183#if IS_ENABLED(CONFIG_IPV6)
 184        struct sockaddr_in6 sin6 = {
 185                .sin6_family            = AF_INET6,
 186                .sin6_addr              = IN6ADDR_ANY_INIT,
 187                .sin6_port              = htons(port),
 188        };
 189#endif
 190        struct sockaddr *sap;
 191        size_t len;
 192
 193        switch (family) {
 194        case PF_INET:
 195                sap = (struct sockaddr *)&sin;
 196                len = sizeof(sin);
 197                break;
 198#if IS_ENABLED(CONFIG_IPV6)
 199        case PF_INET6:
 200                sap = (struct sockaddr *)&sin6;
 201                len = sizeof(sin6);
 202                break;
 203#endif
 204        default:
 205                return ERR_PTR(-EAFNOSUPPORT);
 206        }
 207
 208        return xcl->xcl_ops->xpo_create(serv, net, sap, len, flags);
 209}
 210
 211/*
 212 * svc_xprt_received conditionally queues the transport for processing
 213 * by another thread. The caller must hold the XPT_BUSY bit and must
 214 * not thereafter touch transport data.
 215 *
 216 * Note: XPT_DATA only gets cleared when a read-attempt finds no (or
 217 * insufficient) data.
 218 */
 219static void svc_xprt_received(struct svc_xprt *xprt)
 220{
 221        WARN_ON_ONCE(!test_bit(XPT_BUSY, &xprt->xpt_flags));
 222        if (!test_bit(XPT_BUSY, &xprt->xpt_flags))
 223                return;
 224        /* As soon as we clear busy, the xprt could be closed and
 225         * 'put', so we need a reference to call svc_xprt_enqueue with:
 226         */
 227        svc_xprt_get(xprt);
 228        clear_bit(XPT_BUSY, &xprt->xpt_flags);
 229        svc_xprt_enqueue(xprt);
 230        svc_xprt_put(xprt);
 231}
 232
 233void svc_add_new_perm_xprt(struct svc_serv *serv, struct svc_xprt *new)
 234{
 235        clear_bit(XPT_TEMP, &new->xpt_flags);
 236        spin_lock_bh(&serv->sv_lock);
 237        list_add(&new->xpt_list, &serv->sv_permsocks);
 238        spin_unlock_bh(&serv->sv_lock);
 239        svc_xprt_received(new);
 240}
 241
 242int svc_create_xprt(struct svc_serv *serv, const char *xprt_name,
 243                    struct net *net, const int family,
 244                    const unsigned short port, int flags)
 245{
 246        struct svc_xprt_class *xcl;
 247
 248        dprintk("svc: creating transport %s[%d]\n", xprt_name, port);
 249        spin_lock(&svc_xprt_class_lock);
 250        list_for_each_entry(xcl, &svc_xprt_class_list, xcl_list) {
 251                struct svc_xprt *newxprt;
 252                unsigned short newport;
 253
 254                if (strcmp(xprt_name, xcl->xcl_name))
 255                        continue;
 256
 257                if (!try_module_get(xcl->xcl_owner))
 258                        goto err;
 259
 260                spin_unlock(&svc_xprt_class_lock);
 261                newxprt = __svc_xpo_create(xcl, serv, net, family, port, flags);
 262                if (IS_ERR(newxprt)) {
 263                        module_put(xcl->xcl_owner);
 264                        return PTR_ERR(newxprt);
 265                }
 266                svc_add_new_perm_xprt(serv, newxprt);
 267                newport = svc_xprt_local_port(newxprt);
 268                return newport;
 269        }
 270 err:
 271        spin_unlock(&svc_xprt_class_lock);
 272        dprintk("svc: transport %s not found\n", xprt_name);
 273
 274        /* This errno is exposed to user space.  Provide a reasonable
 275         * perror msg for a bad transport. */
 276        return -EPROTONOSUPPORT;
 277}
 278EXPORT_SYMBOL_GPL(svc_create_xprt);
 279
 280/*
 281 * Copy the local and remote xprt addresses to the rqstp structure
 282 */
 283void svc_xprt_copy_addrs(struct svc_rqst *rqstp, struct svc_xprt *xprt)
 284{
 285        memcpy(&rqstp->rq_addr, &xprt->xpt_remote, xprt->xpt_remotelen);
 286        rqstp->rq_addrlen = xprt->xpt_remotelen;
 287
 288        /*
 289         * Destination address in request is needed for binding the
 290         * source address in RPC replies/callbacks later.
 291         */
 292        memcpy(&rqstp->rq_daddr, &xprt->xpt_local, xprt->xpt_locallen);
 293        rqstp->rq_daddrlen = xprt->xpt_locallen;
 294}
 295EXPORT_SYMBOL_GPL(svc_xprt_copy_addrs);
 296
 297/**
 298 * svc_print_addr - Format rq_addr field for printing
 299 * @rqstp: svc_rqst struct containing address to print
 300 * @buf: target buffer for formatted address
 301 * @len: length of target buffer
 302 *
 303 */
 304char *svc_print_addr(struct svc_rqst *rqstp, char *buf, size_t len)
 305{
 306        return __svc_print_addr(svc_addr(rqstp), buf, len);
 307}
 308EXPORT_SYMBOL_GPL(svc_print_addr);
 309
 310/*
 311 * Queue up an idle server thread.  Must have pool->sp_lock held.
 312 * Note: this is really a stack rather than a queue, so that we only
 313 * use as many different threads as we need, and the rest don't pollute
 314 * the cache.
 315 */
 316static void svc_thread_enqueue(struct svc_pool *pool, struct svc_rqst *rqstp)
 317{
 318        list_add(&rqstp->rq_list, &pool->sp_threads);
 319}
 320
 321/*
 322 * Dequeue an nfsd thread.  Must have pool->sp_lock held.
 323 */
 324static void svc_thread_dequeue(struct svc_pool *pool, struct svc_rqst *rqstp)
 325{
 326        list_del(&rqstp->rq_list);
 327}
 328
 329static bool svc_xprt_has_something_to_do(struct svc_xprt *xprt)
 330{
 331        if (xprt->xpt_flags & ((1<<XPT_CONN)|(1<<XPT_CLOSE)))
 332                return true;
 333        if (xprt->xpt_flags & ((1<<XPT_DATA)|(1<<XPT_DEFERRED)))
 334                return xprt->xpt_ops->xpo_has_wspace(xprt);
 335        return false;
 336}
 337
 338/*
 339 * Queue up a transport with data pending. If there are idle nfsd
 340 * processes, wake 'em up.
 341 *
 342 */
 343void svc_xprt_enqueue(struct svc_xprt *xprt)
 344{
 345        struct svc_pool *pool;
 346        struct svc_rqst *rqstp;
 347        int cpu;
 348
 349        if (!svc_xprt_has_something_to_do(xprt))
 350                return;
 351
 352        cpu = get_cpu();
 353        pool = svc_pool_for_cpu(xprt->xpt_server, cpu);
 354        put_cpu();
 355
 356        spin_lock_bh(&pool->sp_lock);
 357
 358        if (!list_empty(&pool->sp_threads) &&
 359            !list_empty(&pool->sp_sockets))
 360                printk(KERN_ERR
 361                       "svc_xprt_enqueue: "
 362                       "threads and transports both waiting??\n");
 363
 364        pool->sp_stats.packets++;
 365
 366        /* Mark transport as busy. It will remain in this state until
 367         * the provider calls svc_xprt_received. We update XPT_BUSY
 368         * atomically because it also guards against trying to enqueue
 369         * the transport twice.
 370         */
 371        if (test_and_set_bit(XPT_BUSY, &xprt->xpt_flags)) {
 372                /* Don't enqueue transport while already enqueued */
 373                dprintk("svc: transport %p busy, not enqueued\n", xprt);
 374                goto out_unlock;
 375        }
 376
 377        if (!list_empty(&pool->sp_threads)) {
 378                rqstp = list_entry(pool->sp_threads.next,
 379                                   struct svc_rqst,
 380                                   rq_list);
 381                dprintk("svc: transport %p served by daemon %p\n",
 382                        xprt, rqstp);
 383                svc_thread_dequeue(pool, rqstp);
 384                if (rqstp->rq_xprt)
 385                        printk(KERN_ERR
 386                                "svc_xprt_enqueue: server %p, rq_xprt=%p!\n",
 387                                rqstp, rqstp->rq_xprt);
 388                rqstp->rq_xprt = xprt;
 389                svc_xprt_get(xprt);
 390                pool->sp_stats.threads_woken++;
 391                wake_up(&rqstp->rq_wait);
 392        } else {
 393                dprintk("svc: transport %p put into queue\n", xprt);
 394                list_add_tail(&xprt->xpt_ready, &pool->sp_sockets);
 395                pool->sp_stats.sockets_queued++;
 396        }
 397
 398out_unlock:
 399        spin_unlock_bh(&pool->sp_lock);
 400}
 401EXPORT_SYMBOL_GPL(svc_xprt_enqueue);
 402
 403/*
 404 * Dequeue the first transport.  Must be called with the pool->sp_lock held.
 405 */
 406static struct svc_xprt *svc_xprt_dequeue(struct svc_pool *pool)
 407{
 408        struct svc_xprt *xprt;
 409
 410        if (list_empty(&pool->sp_sockets))
 411                return NULL;
 412
 413        xprt = list_entry(pool->sp_sockets.next,
 414                          struct svc_xprt, xpt_ready);
 415        list_del_init(&xprt->xpt_ready);
 416
 417        dprintk("svc: transport %p dequeued, inuse=%d\n",
 418                xprt, atomic_read(&xprt->xpt_ref.refcount));
 419
 420        return xprt;
 421}
 422
 423/**
 424 * svc_reserve - change the space reserved for the reply to a request.
 425 * @rqstp:  The request in question
 426 * @space: new max space to reserve
 427 *
 428 * Each request reserves some space on the output queue of the transport
 429 * to make sure the reply fits.  This function reduces that reserved
 430 * space to be the amount of space used already, plus @space.
 431 *
 432 */
 433void svc_reserve(struct svc_rqst *rqstp, int space)
 434{
 435        space += rqstp->rq_res.head[0].iov_len;
 436
 437        if (space < rqstp->rq_reserved) {
 438                struct svc_xprt *xprt = rqstp->rq_xprt;
 439                atomic_sub((rqstp->rq_reserved - space), &xprt->xpt_reserved);
 440                rqstp->rq_reserved = space;
 441
 442                svc_xprt_enqueue(xprt);
 443        }
 444}
 445EXPORT_SYMBOL_GPL(svc_reserve);
 446
 447static void svc_xprt_release(struct svc_rqst *rqstp)
 448{
 449        struct svc_xprt *xprt = rqstp->rq_xprt;
 450
 451        rqstp->rq_xprt->xpt_ops->xpo_release_rqst(rqstp);
 452
 453        kfree(rqstp->rq_deferred);
 454        rqstp->rq_deferred = NULL;
 455
 456        svc_free_res_pages(rqstp);
 457        rqstp->rq_res.page_len = 0;
 458        rqstp->rq_res.page_base = 0;
 459
 460        /* Reset response buffer and release
 461         * the reservation.
 462         * But first, check that enough space was reserved
 463         * for the reply, otherwise we have a bug!
 464         */
 465        if ((rqstp->rq_res.len) >  rqstp->rq_reserved)
 466                printk(KERN_ERR "RPC request reserved %d but used %d\n",
 467                       rqstp->rq_reserved,
 468                       rqstp->rq_res.len);
 469
 470        rqstp->rq_res.head[0].iov_len = 0;
 471        svc_reserve(rqstp, 0);
 472        rqstp->rq_xprt = NULL;
 473
 474        svc_xprt_put(xprt);
 475}
 476
 477/*
 478 * External function to wake up a server waiting for data
 479 * This really only makes sense for services like lockd
 480 * which have exactly one thread anyway.
 481 */
 482void svc_wake_up(struct svc_serv *serv)
 483{
 484        struct svc_rqst *rqstp;
 485        unsigned int i;
 486        struct svc_pool *pool;
 487
 488        for (i = 0; i < serv->sv_nrpools; i++) {
 489                pool = &serv->sv_pools[i];
 490
 491                spin_lock_bh(&pool->sp_lock);
 492                if (!list_empty(&pool->sp_threads)) {
 493                        rqstp = list_entry(pool->sp_threads.next,
 494                                           struct svc_rqst,
 495                                           rq_list);
 496                        dprintk("svc: daemon %p woken up.\n", rqstp);
 497                        /*
 498                        svc_thread_dequeue(pool, rqstp);
 499                        rqstp->rq_xprt = NULL;
 500                         */
 501                        wake_up(&rqstp->rq_wait);
 502                }
 503                spin_unlock_bh(&pool->sp_lock);
 504        }
 505}
 506EXPORT_SYMBOL_GPL(svc_wake_up);
 507
 508int svc_port_is_privileged(struct sockaddr *sin)
 509{
 510        switch (sin->sa_family) {
 511        case AF_INET:
 512                return ntohs(((struct sockaddr_in *)sin)->sin_port)
 513                        < PROT_SOCK;
 514        case AF_INET6:
 515                return ntohs(((struct sockaddr_in6 *)sin)->sin6_port)
 516                        < PROT_SOCK;
 517        default:
 518                return 0;
 519        }
 520}
 521
 522/*
 523 * Make sure that we don't have too many active connections. If we have,
 524 * something must be dropped. It's not clear what will happen if we allow
 525 * "too many" connections, but when dealing with network-facing software,
 526 * we have to code defensively. Here we do that by imposing hard limits.
 527 *
 528 * There's no point in trying to do random drop here for DoS
 529 * prevention. The NFS clients does 1 reconnect in 15 seconds. An
 530 * attacker can easily beat that.
 531 *
 532 * The only somewhat efficient mechanism would be if drop old
 533 * connections from the same IP first. But right now we don't even
 534 * record the client IP in svc_sock.
 535 *
 536 * single-threaded services that expect a lot of clients will probably
 537 * need to set sv_maxconn to override the default value which is based
 538 * on the number of threads
 539 */
 540static void svc_check_conn_limits(struct svc_serv *serv)
 541{
 542        unsigned int limit = serv->sv_maxconn ? serv->sv_maxconn :
 543                                (serv->sv_nrthreads+3) * 20;
 544
 545        if (serv->sv_tmpcnt > limit) {
 546                struct svc_xprt *xprt = NULL;
 547                spin_lock_bh(&serv->sv_lock);
 548                if (!list_empty(&serv->sv_tempsocks)) {
 549                        /* Try to help the admin */
 550                        net_notice_ratelimited("%s: too many open connections, consider increasing the %s\n",
 551                                               serv->sv_name, serv->sv_maxconn ?
 552                                               "max number of connections" :
 553                                               "number of threads");
 554                        /*
 555                         * Always select the oldest connection. It's not fair,
 556                         * but so is life
 557                         */
 558                        xprt = list_entry(serv->sv_tempsocks.prev,
 559                                          struct svc_xprt,
 560                                          xpt_list);
 561                        set_bit(XPT_CLOSE, &xprt->xpt_flags);
 562                        svc_xprt_get(xprt);
 563                }
 564                spin_unlock_bh(&serv->sv_lock);
 565
 566                if (xprt) {
 567                        svc_xprt_enqueue(xprt);
 568                        svc_xprt_put(xprt);
 569                }
 570        }
 571}
 572
 573int svc_alloc_arg(struct svc_rqst *rqstp)
 574{
 575        struct svc_serv *serv = rqstp->rq_server;
 576        struct xdr_buf *arg;
 577        int pages;
 578        int i;
 579
 580        /* now allocate needed pages.  If we get a failure, sleep briefly */
 581        pages = (serv->sv_max_mesg + PAGE_SIZE) / PAGE_SIZE;
 582        WARN_ON_ONCE(pages >= RPCSVC_MAXPAGES);
 583        if (pages >= RPCSVC_MAXPAGES)
 584                /* use as many pages as possible */
 585                pages = RPCSVC_MAXPAGES - 1;
 586        for (i = 0; i < pages ; i++)
 587                while (rqstp->rq_pages[i] == NULL) {
 588                        struct page *p = alloc_page(GFP_KERNEL);
 589                        if (!p) {
 590                                set_current_state(TASK_INTERRUPTIBLE);
 591                                if (signalled() || kthread_should_stop()) {
 592                                        set_current_state(TASK_RUNNING);
 593                                        return -EINTR;
 594                                }
 595                                schedule_timeout(msecs_to_jiffies(500));
 596                        }
 597                        rqstp->rq_pages[i] = p;
 598                }
 599        rqstp->rq_pages[i++] = NULL; /* this might be seen in nfs_read_actor */
 600
 601        /* Make arg->head point to first page and arg->pages point to rest */
 602        arg = &rqstp->rq_arg;
 603        arg->head[0].iov_base = page_address(rqstp->rq_pages[0]);
 604        arg->head[0].iov_len = PAGE_SIZE;
 605        arg->pages = rqstp->rq_pages + 1;
 606        arg->page_base = 0;
 607        /* save at least one page for response */
 608        arg->page_len = (pages-2)*PAGE_SIZE;
 609        arg->len = (pages-1)*PAGE_SIZE;
 610        arg->tail[0].iov_len = 0;
 611        return 0;
 612}
 613
 614struct svc_xprt *svc_get_next_xprt(struct svc_rqst *rqstp, long timeout)
 615{
 616        struct svc_xprt *xprt;
 617        struct svc_pool         *pool = rqstp->rq_pool;
 618        DECLARE_WAITQUEUE(wait, current);
 619        long                    time_left;
 620
 621        /* Normally we will wait up to 5 seconds for any required
 622         * cache information to be provided.
 623         */
 624        rqstp->rq_chandle.thread_wait = 5*HZ;
 625
 626        spin_lock_bh(&pool->sp_lock);
 627        xprt = svc_xprt_dequeue(pool);
 628        if (xprt) {
 629                rqstp->rq_xprt = xprt;
 630                svc_xprt_get(xprt);
 631
 632                /* As there is a shortage of threads and this request
 633                 * had to be queued, don't allow the thread to wait so
 634                 * long for cache updates.
 635                 */
 636                rqstp->rq_chandle.thread_wait = 1*HZ;
 637        } else {
 638                /* No data pending. Go to sleep */
 639                svc_thread_enqueue(pool, rqstp);
 640
 641                /*
 642                 * We have to be able to interrupt this wait
 643                 * to bring down the daemons ...
 644                 */
 645                set_current_state(TASK_INTERRUPTIBLE);
 646
 647                /*
 648                 * checking kthread_should_stop() here allows us to avoid
 649                 * locking and signalling when stopping kthreads that call
 650                 * svc_recv. If the thread has already been woken up, then
 651                 * we can exit here without sleeping. If not, then it
 652                 * it'll be woken up quickly during the schedule_timeout
 653                 */
 654                if (kthread_should_stop()) {
 655                        set_current_state(TASK_RUNNING);
 656                        spin_unlock_bh(&pool->sp_lock);
 657                        return ERR_PTR(-EINTR);
 658                }
 659
 660                add_wait_queue(&rqstp->rq_wait, &wait);
 661                spin_unlock_bh(&pool->sp_lock);
 662
 663                time_left = schedule_timeout(timeout);
 664
 665                try_to_freeze();
 666
 667                spin_lock_bh(&pool->sp_lock);
 668                remove_wait_queue(&rqstp->rq_wait, &wait);
 669                if (!time_left)
 670                        pool->sp_stats.threads_timedout++;
 671
 672                xprt = rqstp->rq_xprt;
 673                if (!xprt) {
 674                        svc_thread_dequeue(pool, rqstp);
 675                        spin_unlock_bh(&pool->sp_lock);
 676                        dprintk("svc: server %p, no data yet\n", rqstp);
 677                        if (signalled() || kthread_should_stop())
 678                                return ERR_PTR(-EINTR);
 679                        else
 680                                return ERR_PTR(-EAGAIN);
 681                }
 682        }
 683        spin_unlock_bh(&pool->sp_lock);
 684        return xprt;
 685}
 686
 687void svc_add_new_temp_xprt(struct svc_serv *serv, struct svc_xprt *newxpt)
 688{
 689        spin_lock_bh(&serv->sv_lock);
 690        set_bit(XPT_TEMP, &newxpt->xpt_flags);
 691        list_add(&newxpt->xpt_list, &serv->sv_tempsocks);
 692        serv->sv_tmpcnt++;
 693        if (serv->sv_temptimer.function == NULL) {
 694                /* setup timer to age temp transports */
 695                setup_timer(&serv->sv_temptimer, svc_age_temp_xprts,
 696                            (unsigned long)serv);
 697                mod_timer(&serv->sv_temptimer,
 698                          jiffies + svc_conn_age_period * HZ);
 699        }
 700        spin_unlock_bh(&serv->sv_lock);
 701        svc_xprt_received(newxpt);
 702}
 703
 704static int svc_handle_xprt(struct svc_rqst *rqstp, struct svc_xprt *xprt)
 705{
 706        struct svc_serv *serv = rqstp->rq_server;
 707        int len = 0;
 708
 709        if (test_bit(XPT_CLOSE, &xprt->xpt_flags)) {
 710                dprintk("svc_recv: found XPT_CLOSE\n");
 711                svc_delete_xprt(xprt);
 712                /* Leave XPT_BUSY set on the dead xprt: */
 713                return 0;
 714        }
 715        if (test_bit(XPT_LISTENER, &xprt->xpt_flags)) {
 716                struct svc_xprt *newxpt;
 717                /*
 718                 * We know this module_get will succeed because the
 719                 * listener holds a reference too
 720                 */
 721                __module_get(xprt->xpt_class->xcl_owner);
 722                svc_check_conn_limits(xprt->xpt_server);
 723                newxpt = xprt->xpt_ops->xpo_accept(xprt);
 724                if (newxpt)
 725                        svc_add_new_temp_xprt(serv, newxpt);
 726        } else if (xprt->xpt_ops->xpo_has_wspace(xprt)) {
 727                /* XPT_DATA|XPT_DEFERRED case: */
 728                dprintk("svc: server %p, pool %u, transport %p, inuse=%d\n",
 729                        rqstp, rqstp->rq_pool->sp_id, xprt,
 730                        atomic_read(&xprt->xpt_ref.refcount));
 731                rqstp->rq_deferred = svc_deferred_dequeue(xprt);
 732                if (rqstp->rq_deferred)
 733                        len = svc_deferred_recv(rqstp);
 734                else
 735                        len = xprt->xpt_ops->xpo_recvfrom(rqstp);
 736                dprintk("svc: got len=%d\n", len);
 737                rqstp->rq_reserved = serv->sv_max_mesg;
 738                atomic_add(rqstp->rq_reserved, &xprt->xpt_reserved);
 739        }
 740        /* clear XPT_BUSY: */
 741        svc_xprt_received(xprt);
 742        return len;
 743}
 744
 745/*
 746 * Receive the next request on any transport.  This code is carefully
 747 * organised not to touch any cachelines in the shared svc_serv
 748 * structure, only cachelines in the local svc_pool.
 749 */
 750int svc_recv(struct svc_rqst *rqstp, long timeout)
 751{
 752        struct svc_xprt         *xprt = NULL;
 753        struct svc_serv         *serv = rqstp->rq_server;
 754        int                     len, err;
 755
 756        dprintk("svc: server %p waiting for data (to = %ld)\n",
 757                rqstp, timeout);
 758
 759        if (rqstp->rq_xprt)
 760                printk(KERN_ERR
 761                        "svc_recv: service %p, transport not NULL!\n",
 762                         rqstp);
 763        if (waitqueue_active(&rqstp->rq_wait))
 764                printk(KERN_ERR
 765                        "svc_recv: service %p, wait queue active!\n",
 766                         rqstp);
 767
 768        err = svc_alloc_arg(rqstp);
 769        if (err)
 770                return err;
 771
 772        try_to_freeze();
 773        cond_resched();
 774        if (signalled() || kthread_should_stop())
 775                return -EINTR;
 776
 777        xprt = svc_get_next_xprt(rqstp, timeout);
 778        if (IS_ERR(xprt))
 779                return PTR_ERR(xprt);
 780
 781        len = svc_handle_xprt(rqstp, xprt);
 782
 783        /* No data, incomplete (TCP) read, or accept() */
 784        if (len <= 0)
 785                goto out;
 786
 787        clear_bit(XPT_OLD, &xprt->xpt_flags);
 788
 789        rqstp->rq_secure = svc_port_is_privileged(svc_addr(rqstp));
 790        rqstp->rq_chandle.defer = svc_defer;
 791
 792        if (serv->sv_stats)
 793                serv->sv_stats->netcnt++;
 794        return len;
 795out:
 796        rqstp->rq_res.len = 0;
 797        svc_xprt_release(rqstp);
 798        return -EAGAIN;
 799}
 800EXPORT_SYMBOL_GPL(svc_recv);
 801
 802/*
 803 * Drop request
 804 */
 805void svc_drop(struct svc_rqst *rqstp)
 806{
 807        dprintk("svc: xprt %p dropped request\n", rqstp->rq_xprt);
 808        svc_xprt_release(rqstp);
 809}
 810EXPORT_SYMBOL_GPL(svc_drop);
 811
 812/*
 813 * Return reply to client.
 814 */
 815int svc_send(struct svc_rqst *rqstp)
 816{
 817        struct svc_xprt *xprt;
 818        int             len;
 819        struct xdr_buf  *xb;
 820
 821        xprt = rqstp->rq_xprt;
 822        if (!xprt)
 823                return -EFAULT;
 824
 825        /* release the receive skb before sending the reply */
 826        rqstp->rq_xprt->xpt_ops->xpo_release_rqst(rqstp);
 827
 828        /* calculate over-all length */
 829        xb = &rqstp->rq_res;
 830        xb->len = xb->head[0].iov_len +
 831                xb->page_len +
 832                xb->tail[0].iov_len;
 833
 834        /* Grab mutex to serialize outgoing data. */
 835        mutex_lock(&xprt->xpt_mutex);
 836        if (test_bit(XPT_DEAD, &xprt->xpt_flags)
 837                        || test_bit(XPT_CLOSE, &xprt->xpt_flags))
 838                len = -ENOTCONN;
 839        else
 840                len = xprt->xpt_ops->xpo_sendto(rqstp);
 841        mutex_unlock(&xprt->xpt_mutex);
 842        rpc_wake_up(&xprt->xpt_bc_pending);
 843        svc_xprt_release(rqstp);
 844
 845        if (len == -ECONNREFUSED || len == -ENOTCONN || len == -EAGAIN)
 846                return 0;
 847        return len;
 848}
 849
 850/*
 851 * Timer function to close old temporary transports, using
 852 * a mark-and-sweep algorithm.
 853 */
 854static void svc_age_temp_xprts(unsigned long closure)
 855{
 856        struct svc_serv *serv = (struct svc_serv *)closure;
 857        struct svc_xprt *xprt;
 858        struct list_head *le, *next;
 859
 860        dprintk("svc_age_temp_xprts\n");
 861
 862        if (!spin_trylock_bh(&serv->sv_lock)) {
 863                /* busy, try again 1 sec later */
 864                dprintk("svc_age_temp_xprts: busy\n");
 865                mod_timer(&serv->sv_temptimer, jiffies + HZ);
 866                return;
 867        }
 868
 869        list_for_each_safe(le, next, &serv->sv_tempsocks) {
 870                xprt = list_entry(le, struct svc_xprt, xpt_list);
 871
 872                /* First time through, just mark it OLD. Second time
 873                 * through, close it. */
 874                if (!test_and_set_bit(XPT_OLD, &xprt->xpt_flags))
 875                        continue;
 876                if (atomic_read(&xprt->xpt_ref.refcount) > 1 ||
 877                    test_bit(XPT_BUSY, &xprt->xpt_flags))
 878                        continue;
 879                list_del_init(le);
 880                set_bit(XPT_CLOSE, &xprt->xpt_flags);
 881                set_bit(XPT_DETACHED, &xprt->xpt_flags);
 882                dprintk("queuing xprt %p for closing\n", xprt);
 883
 884                /* a thread will dequeue and close it soon */
 885                svc_xprt_enqueue(xprt);
 886        }
 887        spin_unlock_bh(&serv->sv_lock);
 888
 889        mod_timer(&serv->sv_temptimer, jiffies + svc_conn_age_period * HZ);
 890}
 891
 892static void call_xpt_users(struct svc_xprt *xprt)
 893{
 894        struct svc_xpt_user *u;
 895
 896        spin_lock(&xprt->xpt_lock);
 897        while (!list_empty(&xprt->xpt_users)) {
 898                u = list_first_entry(&xprt->xpt_users, struct svc_xpt_user, list);
 899                list_del(&u->list);
 900                u->callback(u);
 901        }
 902        spin_unlock(&xprt->xpt_lock);
 903}
 904
 905/*
 906 * Remove a dead transport
 907 */
 908static void svc_delete_xprt(struct svc_xprt *xprt)
 909{
 910        struct svc_serv *serv = xprt->xpt_server;
 911        struct svc_deferred_req *dr;
 912
 913        /* Only do this once */
 914        if (test_and_set_bit(XPT_DEAD, &xprt->xpt_flags))
 915                BUG();
 916
 917        dprintk("svc: svc_delete_xprt(%p)\n", xprt);
 918        xprt->xpt_ops->xpo_detach(xprt);
 919
 920        spin_lock_bh(&serv->sv_lock);
 921        if (!test_and_set_bit(XPT_DETACHED, &xprt->xpt_flags))
 922                list_del_init(&xprt->xpt_list);
 923        WARN_ON_ONCE(!list_empty(&xprt->xpt_ready));
 924        if (test_bit(XPT_TEMP, &xprt->xpt_flags))
 925                serv->sv_tmpcnt--;
 926        spin_unlock_bh(&serv->sv_lock);
 927
 928        while ((dr = svc_deferred_dequeue(xprt)) != NULL)
 929                kfree(dr);
 930
 931        call_xpt_users(xprt);
 932        svc_xprt_put(xprt);
 933}
 934
 935void svc_close_xprt(struct svc_xprt *xprt)
 936{
 937        set_bit(XPT_CLOSE, &xprt->xpt_flags);
 938        if (test_and_set_bit(XPT_BUSY, &xprt->xpt_flags))
 939                /* someone else will have to effect the close */
 940                return;
 941        /*
 942         * We expect svc_close_xprt() to work even when no threads are
 943         * running (e.g., while configuring the server before starting
 944         * any threads), so if the transport isn't busy, we delete
 945         * it ourself:
 946         */
 947        svc_delete_xprt(xprt);
 948}
 949EXPORT_SYMBOL_GPL(svc_close_xprt);
 950
 951static int svc_close_list(struct svc_serv *serv, struct list_head *xprt_list, struct net *net)
 952{
 953        struct svc_xprt *xprt;
 954        int ret = 0;
 955
 956        spin_lock(&serv->sv_lock);
 957        list_for_each_entry(xprt, xprt_list, xpt_list) {
 958                if (xprt->xpt_net != net)
 959                        continue;
 960                ret++;
 961                set_bit(XPT_CLOSE, &xprt->xpt_flags);
 962                svc_xprt_enqueue(xprt);
 963        }
 964        spin_unlock(&serv->sv_lock);
 965        return ret;
 966}
 967
 968static struct svc_xprt *svc_dequeue_net(struct svc_serv *serv, struct net *net)
 969{
 970        struct svc_pool *pool;
 971        struct svc_xprt *xprt;
 972        struct svc_xprt *tmp;
 973        int i;
 974
 975        for (i = 0; i < serv->sv_nrpools; i++) {
 976                pool = &serv->sv_pools[i];
 977
 978                spin_lock_bh(&pool->sp_lock);
 979                list_for_each_entry_safe(xprt, tmp, &pool->sp_sockets, xpt_ready) {
 980                        if (xprt->xpt_net != net)
 981                                continue;
 982                        list_del_init(&xprt->xpt_ready);
 983                        spin_unlock_bh(&pool->sp_lock);
 984                        return xprt;
 985                }
 986                spin_unlock_bh(&pool->sp_lock);
 987        }
 988        return NULL;
 989}
 990
 991static void svc_clean_up_xprts(struct svc_serv *serv, struct net *net)
 992{
 993        struct svc_xprt *xprt;
 994
 995        while ((xprt = svc_dequeue_net(serv, net))) {
 996                set_bit(XPT_CLOSE, &xprt->xpt_flags);
 997                svc_delete_xprt(xprt);
 998        }
 999}
1000
1001/*
1002 * Server threads may still be running (especially in the case where the
1003 * service is still running in other network namespaces).
1004 *
1005 * So we shut down sockets the same way we would on a running server, by
1006 * setting XPT_CLOSE, enqueuing, and letting a thread pick it up to do
1007 * the close.  In the case there are no such other threads,
1008 * threads running, svc_clean_up_xprts() does a simple version of a
1009 * server's main event loop, and in the case where there are other
1010 * threads, we may need to wait a little while and then check again to
1011 * see if they're done.
1012 */
1013void svc_close_net(struct svc_serv *serv, struct net *net)
1014{
1015        int delay = 0;
1016
1017        while (svc_close_list(serv, &serv->sv_permsocks, net) +
1018               svc_close_list(serv, &serv->sv_tempsocks, net)) {
1019
1020                svc_clean_up_xprts(serv, net);
1021                msleep(delay++);
1022        }
1023}
1024
1025/*
1026 * Handle defer and revisit of requests
1027 */
1028
1029static void svc_revisit(struct cache_deferred_req *dreq, int too_many)
1030{
1031        struct svc_deferred_req *dr =
1032                container_of(dreq, struct svc_deferred_req, handle);
1033        struct svc_xprt *xprt = dr->xprt;
1034
1035        spin_lock(&xprt->xpt_lock);
1036        set_bit(XPT_DEFERRED, &xprt->xpt_flags);
1037        if (too_many || test_bit(XPT_DEAD, &xprt->xpt_flags)) {
1038                spin_unlock(&xprt->xpt_lock);
1039                dprintk("revisit canceled\n");
1040                svc_xprt_put(xprt);
1041                kfree(dr);
1042                return;
1043        }
1044        dprintk("revisit queued\n");
1045        dr->xprt = NULL;
1046        list_add(&dr->handle.recent, &xprt->xpt_deferred);
1047        spin_unlock(&xprt->xpt_lock);
1048        svc_xprt_enqueue(xprt);
1049        svc_xprt_put(xprt);
1050}
1051
1052/*
1053 * Save the request off for later processing. The request buffer looks
1054 * like this:
1055 *
1056 * <xprt-header><rpc-header><rpc-pagelist><rpc-tail>
1057 *
1058 * This code can only handle requests that consist of an xprt-header
1059 * and rpc-header.
1060 */
1061static struct cache_deferred_req *svc_defer(struct cache_req *req)
1062{
1063        struct svc_rqst *rqstp = container_of(req, struct svc_rqst, rq_chandle);
1064        struct svc_deferred_req *dr;
1065
1066        if (rqstp->rq_arg.page_len || !rqstp->rq_usedeferral)
1067                return NULL; /* if more than a page, give up FIXME */
1068        if (rqstp->rq_deferred) {
1069                dr = rqstp->rq_deferred;
1070                rqstp->rq_deferred = NULL;
1071        } else {
1072                size_t skip;
1073                size_t size;
1074                /* FIXME maybe discard if size too large */
1075                size = sizeof(struct svc_deferred_req) + rqstp->rq_arg.len;
1076                dr = kmalloc(size, GFP_KERNEL);
1077                if (dr == NULL)
1078                        return NULL;
1079
1080                dr->handle.owner = rqstp->rq_server;
1081                dr->prot = rqstp->rq_prot;
1082                memcpy(&dr->addr, &rqstp->rq_addr, rqstp->rq_addrlen);
1083                dr->addrlen = rqstp->rq_addrlen;
1084                dr->daddr = rqstp->rq_daddr;
1085                dr->argslen = rqstp->rq_arg.len >> 2;
1086                dr->xprt_hlen = rqstp->rq_xprt_hlen;
1087
1088                /* back up head to the start of the buffer and copy */
1089                skip = rqstp->rq_arg.len - rqstp->rq_arg.head[0].iov_len;
1090                memcpy(dr->args, rqstp->rq_arg.head[0].iov_base - skip,
1091                       dr->argslen << 2);
1092        }
1093        svc_xprt_get(rqstp->rq_xprt);
1094        dr->xprt = rqstp->rq_xprt;
1095        rqstp->rq_dropme = true;
1096
1097        dr->handle.revisit = svc_revisit;
1098        return &dr->handle;
1099}
1100
1101/*
1102 * recv data from a deferred request into an active one
1103 */
1104static int svc_deferred_recv(struct svc_rqst *rqstp)
1105{
1106        struct svc_deferred_req *dr = rqstp->rq_deferred;
1107
1108        /* setup iov_base past transport header */
1109        rqstp->rq_arg.head[0].iov_base = dr->args + (dr->xprt_hlen>>2);
1110        /* The iov_len does not include the transport header bytes */
1111        rqstp->rq_arg.head[0].iov_len = (dr->argslen<<2) - dr->xprt_hlen;
1112        rqstp->rq_arg.page_len = 0;
1113        /* The rq_arg.len includes the transport header bytes */
1114        rqstp->rq_arg.len     = dr->argslen<<2;
1115        rqstp->rq_prot        = dr->prot;
1116        memcpy(&rqstp->rq_addr, &dr->addr, dr->addrlen);
1117        rqstp->rq_addrlen     = dr->addrlen;
1118        /* Save off transport header len in case we get deferred again */
1119        rqstp->rq_xprt_hlen   = dr->xprt_hlen;
1120        rqstp->rq_daddr       = dr->daddr;
1121        rqstp->rq_respages    = rqstp->rq_pages;
1122        return (dr->argslen<<2) - dr->xprt_hlen;
1123}
1124
1125
1126static struct svc_deferred_req *svc_deferred_dequeue(struct svc_xprt *xprt)
1127{
1128        struct svc_deferred_req *dr = NULL;
1129
1130        if (!test_bit(XPT_DEFERRED, &xprt->xpt_flags))
1131                return NULL;
1132        spin_lock(&xprt->xpt_lock);
1133        if (!list_empty(&xprt->xpt_deferred)) {
1134                dr = list_entry(xprt->xpt_deferred.next,
1135                                struct svc_deferred_req,
1136                                handle.recent);
1137                list_del_init(&dr->handle.recent);
1138        } else
1139                clear_bit(XPT_DEFERRED, &xprt->xpt_flags);
1140        spin_unlock(&xprt->xpt_lock);
1141        return dr;
1142}
1143
1144/**
1145 * svc_find_xprt - find an RPC transport instance
1146 * @serv: pointer to svc_serv to search
1147 * @xcl_name: C string containing transport's class name
1148 * @net: owner net pointer
1149 * @af: Address family of transport's local address
1150 * @port: transport's IP port number
1151 *
1152 * Return the transport instance pointer for the endpoint accepting
1153 * connections/peer traffic from the specified transport class,
1154 * address family and port.
1155 *
1156 * Specifying 0 for the address family or port is effectively a
1157 * wild-card, and will result in matching the first transport in the
1158 * service's list that has a matching class name.
1159 */
1160struct svc_xprt *svc_find_xprt(struct svc_serv *serv, const char *xcl_name,
1161                               struct net *net, const sa_family_t af,
1162                               const unsigned short port)
1163{
1164        struct svc_xprt *xprt;
1165        struct svc_xprt *found = NULL;
1166
1167        /* Sanity check the args */
1168        if (serv == NULL || xcl_name == NULL)
1169                return found;
1170
1171        spin_lock_bh(&serv->sv_lock);
1172        list_for_each_entry(xprt, &serv->sv_permsocks, xpt_list) {
1173                if (xprt->xpt_net != net)
1174                        continue;
1175                if (strcmp(xprt->xpt_class->xcl_name, xcl_name))
1176                        continue;
1177                if (af != AF_UNSPEC && af != xprt->xpt_local.ss_family)
1178                        continue;
1179                if (port != 0 && port != svc_xprt_local_port(xprt))
1180                        continue;
1181                found = xprt;
1182                svc_xprt_get(xprt);
1183                break;
1184        }
1185        spin_unlock_bh(&serv->sv_lock);
1186        return found;
1187}
1188EXPORT_SYMBOL_GPL(svc_find_xprt);
1189
1190static int svc_one_xprt_name(const struct svc_xprt *xprt,
1191                             char *pos, int remaining)
1192{
1193        int len;
1194
1195        len = snprintf(pos, remaining, "%s %u\n",
1196                        xprt->xpt_class->xcl_name,
1197                        svc_xprt_local_port(xprt));
1198        if (len >= remaining)
1199                return -ENAMETOOLONG;
1200        return len;
1201}
1202
1203/**
1204 * svc_xprt_names - format a buffer with a list of transport names
1205 * @serv: pointer to an RPC service
1206 * @buf: pointer to a buffer to be filled in
1207 * @buflen: length of buffer to be filled in
1208 *
1209 * Fills in @buf with a string containing a list of transport names,
1210 * each name terminated with '\n'.
1211 *
1212 * Returns positive length of the filled-in string on success; otherwise
1213 * a negative errno value is returned if an error occurs.
1214 */
1215int svc_xprt_names(struct svc_serv *serv, char *buf, const int buflen)
1216{
1217        struct svc_xprt *xprt;
1218        int len, totlen;
1219        char *pos;
1220
1221        /* Sanity check args */
1222        if (!serv)
1223                return 0;
1224
1225        spin_lock_bh(&serv->sv_lock);
1226
1227        pos = buf;
1228        totlen = 0;
1229        list_for_each_entry(xprt, &serv->sv_permsocks, xpt_list) {
1230                len = svc_one_xprt_name(xprt, pos, buflen - totlen);
1231                if (len < 0) {
1232                        *buf = '\0';
1233                        totlen = len;
1234                }
1235                if (len <= 0)
1236                        break;
1237
1238                pos += len;
1239                totlen += len;
1240        }
1241
1242        spin_unlock_bh(&serv->sv_lock);
1243        return totlen;
1244}
1245EXPORT_SYMBOL_GPL(svc_xprt_names);
1246
1247
1248/*----------------------------------------------------------------------------*/
1249
1250static void *svc_pool_stats_start(struct seq_file *m, loff_t *pos)
1251{
1252        unsigned int pidx = (unsigned int)*pos;
1253        struct svc_serv *serv = m->private;
1254
1255        dprintk("svc_pool_stats_start, *pidx=%u\n", pidx);
1256
1257        if (!pidx)
1258                return SEQ_START_TOKEN;
1259        return (pidx > serv->sv_nrpools ? NULL : &serv->sv_pools[pidx-1]);
1260}
1261
1262static void *svc_pool_stats_next(struct seq_file *m, void *p, loff_t *pos)
1263{
1264        struct svc_pool *pool = p;
1265        struct svc_serv *serv = m->private;
1266
1267        dprintk("svc_pool_stats_next, *pos=%llu\n", *pos);
1268
1269        if (p == SEQ_START_TOKEN) {
1270                pool = &serv->sv_pools[0];
1271        } else {
1272                unsigned int pidx = (pool - &serv->sv_pools[0]);
1273                if (pidx < serv->sv_nrpools-1)
1274                        pool = &serv->sv_pools[pidx+1];
1275                else
1276                        pool = NULL;
1277        }
1278        ++*pos;
1279        return pool;
1280}
1281
1282static void svc_pool_stats_stop(struct seq_file *m, void *p)
1283{
1284}
1285
1286static int svc_pool_stats_show(struct seq_file *m, void *p)
1287{
1288        struct svc_pool *pool = p;
1289
1290        if (p == SEQ_START_TOKEN) {
1291                seq_puts(m, "# pool packets-arrived sockets-enqueued threads-woken threads-timedout\n");
1292                return 0;
1293        }
1294
1295        seq_printf(m, "%u %lu %lu %lu %lu\n",
1296                pool->sp_id,
1297                pool->sp_stats.packets,
1298                pool->sp_stats.sockets_queued,
1299                pool->sp_stats.threads_woken,
1300                pool->sp_stats.threads_timedout);
1301
1302        return 0;
1303}
1304
1305static const struct seq_operations svc_pool_stats_seq_ops = {
1306        .start  = svc_pool_stats_start,
1307        .next   = svc_pool_stats_next,
1308        .stop   = svc_pool_stats_stop,
1309        .show   = svc_pool_stats_show,
1310};
1311
1312int svc_pool_stats_open(struct svc_serv *serv, struct file *file)
1313{
1314        int err;
1315
1316        err = seq_open(file, &svc_pool_stats_seq_ops);
1317        if (!err)
1318                ((struct seq_file *) file->private_data)->private = serv;
1319        return err;
1320}
1321EXPORT_SYMBOL(svc_pool_stats_open);
1322
1323/*----------------------------------------------------------------------------*/
1324