linux/net/sunrpc/svc_xprt.c
<<
>>
Prefs
   1// SPDX-License-Identifier: GPL-2.0-only
   2/*
   3 * linux/net/sunrpc/svc_xprt.c
   4 *
   5 * Author: Tom Tucker <tom@opengridcomputing.com>
   6 */
   7
   8#include <linux/sched.h>
   9#include <linux/errno.h>
  10#include <linux/freezer.h>
  11#include <linux/kthread.h>
  12#include <linux/slab.h>
  13#include <net/sock.h>
  14#include <linux/sunrpc/addr.h>
  15#include <linux/sunrpc/stats.h>
  16#include <linux/sunrpc/svc_xprt.h>
  17#include <linux/sunrpc/svcsock.h>
  18#include <linux/sunrpc/xprt.h>
  19#include <linux/module.h>
  20#include <linux/netdevice.h>
  21#include <trace/events/sunrpc.h>
  22
  23#define RPCDBG_FACILITY RPCDBG_SVCXPRT
  24
  25static unsigned int svc_rpc_per_connection_limit __read_mostly;
  26module_param(svc_rpc_per_connection_limit, uint, 0644);
  27
  28
  29static struct svc_deferred_req *svc_deferred_dequeue(struct svc_xprt *xprt);
  30static int svc_deferred_recv(struct svc_rqst *rqstp);
  31static struct cache_deferred_req *svc_defer(struct cache_req *req);
  32static void svc_age_temp_xprts(struct timer_list *t);
  33static void svc_delete_xprt(struct svc_xprt *xprt);
  34
  35/* apparently the "standard" is that clients close
  36 * idle connections after 5 minutes, servers after
  37 * 6 minutes
  38 *   http://nfsv4bat.org/Documents/ConnectAThon/1996/nfstcp.pdf
  39 */
  40static int svc_conn_age_period = 6*60;
  41
  42/* List of registered transport classes */
  43static DEFINE_SPINLOCK(svc_xprt_class_lock);
  44static LIST_HEAD(svc_xprt_class_list);
  45
  46/* SMP locking strategy:
  47 *
  48 *      svc_pool->sp_lock protects most of the fields of that pool.
  49 *      svc_serv->sv_lock protects sv_tempsocks, sv_permsocks, sv_tmpcnt.
  50 *      when both need to be taken (rare), svc_serv->sv_lock is first.
  51 *      The "service mutex" protects svc_serv->sv_nrthread.
  52 *      svc_sock->sk_lock protects the svc_sock->sk_deferred list
  53 *             and the ->sk_info_authunix cache.
  54 *
  55 *      The XPT_BUSY bit in xprt->xpt_flags prevents a transport being
  56 *      enqueued multiply. During normal transport processing this bit
  57 *      is set by svc_xprt_enqueue and cleared by svc_xprt_received.
  58 *      Providers should not manipulate this bit directly.
  59 *
  60 *      Some flags can be set to certain values at any time
  61 *      providing that certain rules are followed:
  62 *
  63 *      XPT_CONN, XPT_DATA:
  64 *              - Can be set or cleared at any time.
  65 *              - After a set, svc_xprt_enqueue must be called to enqueue
  66 *                the transport for processing.
  67 *              - After a clear, the transport must be read/accepted.
  68 *                If this succeeds, it must be set again.
  69 *      XPT_CLOSE:
  70 *              - Can set at any time. It is never cleared.
  71 *      XPT_DEAD:
  72 *              - Can only be set while XPT_BUSY is held which ensures
  73 *                that no other thread will be using the transport or will
  74 *                try to set XPT_DEAD.
  75 */
  76int svc_reg_xprt_class(struct svc_xprt_class *xcl)
  77{
  78        struct svc_xprt_class *cl;
  79        int res = -EEXIST;
  80
  81        dprintk("svc: Adding svc transport class '%s'\n", xcl->xcl_name);
  82
  83        INIT_LIST_HEAD(&xcl->xcl_list);
  84        spin_lock(&svc_xprt_class_lock);
  85        /* Make sure there isn't already a class with the same name */
  86        list_for_each_entry(cl, &svc_xprt_class_list, xcl_list) {
  87                if (strcmp(xcl->xcl_name, cl->xcl_name) == 0)
  88                        goto out;
  89        }
  90        list_add_tail(&xcl->xcl_list, &svc_xprt_class_list);
  91        res = 0;
  92out:
  93        spin_unlock(&svc_xprt_class_lock);
  94        return res;
  95}
  96EXPORT_SYMBOL_GPL(svc_reg_xprt_class);
  97
  98void svc_unreg_xprt_class(struct svc_xprt_class *xcl)
  99{
 100        dprintk("svc: Removing svc transport class '%s'\n", xcl->xcl_name);
 101        spin_lock(&svc_xprt_class_lock);
 102        list_del_init(&xcl->xcl_list);
 103        spin_unlock(&svc_xprt_class_lock);
 104}
 105EXPORT_SYMBOL_GPL(svc_unreg_xprt_class);
 106
 107/**
 108 * svc_print_xprts - Format the transport list for printing
 109 * @buf: target buffer for formatted address
 110 * @maxlen: length of target buffer
 111 *
 112 * Fills in @buf with a string containing a list of transport names, each name
 113 * terminated with '\n'. If the buffer is too small, some entries may be
 114 * missing, but it is guaranteed that all lines in the output buffer are
 115 * complete.
 116 *
 117 * Returns positive length of the filled-in string.
 118 */
 119int svc_print_xprts(char *buf, int maxlen)
 120{
 121        struct svc_xprt_class *xcl;
 122        char tmpstr[80];
 123        int len = 0;
 124        buf[0] = '\0';
 125
 126        spin_lock(&svc_xprt_class_lock);
 127        list_for_each_entry(xcl, &svc_xprt_class_list, xcl_list) {
 128                int slen;
 129
 130                slen = snprintf(tmpstr, sizeof(tmpstr), "%s %d\n",
 131                                xcl->xcl_name, xcl->xcl_max_payload);
 132                if (slen >= sizeof(tmpstr) || len + slen >= maxlen)
 133                        break;
 134                len += slen;
 135                strcat(buf, tmpstr);
 136        }
 137        spin_unlock(&svc_xprt_class_lock);
 138
 139        return len;
 140}
 141
 142/**
 143 * svc_xprt_deferred_close - Close a transport
 144 * @xprt: transport instance
 145 *
 146 * Used in contexts that need to defer the work of shutting down
 147 * the transport to an nfsd thread.
 148 */
 149void svc_xprt_deferred_close(struct svc_xprt *xprt)
 150{
 151        if (!test_and_set_bit(XPT_CLOSE, &xprt->xpt_flags))
 152                svc_xprt_enqueue(xprt);
 153}
 154EXPORT_SYMBOL_GPL(svc_xprt_deferred_close);
 155
 156static void svc_xprt_free(struct kref *kref)
 157{
 158        struct svc_xprt *xprt =
 159                container_of(kref, struct svc_xprt, xpt_ref);
 160        struct module *owner = xprt->xpt_class->xcl_owner;
 161        if (test_bit(XPT_CACHE_AUTH, &xprt->xpt_flags))
 162                svcauth_unix_info_release(xprt);
 163        put_cred(xprt->xpt_cred);
 164        put_net(xprt->xpt_net);
 165        /* See comment on corresponding get in xs_setup_bc_tcp(): */
 166        if (xprt->xpt_bc_xprt)
 167                xprt_put(xprt->xpt_bc_xprt);
 168        if (xprt->xpt_bc_xps)
 169                xprt_switch_put(xprt->xpt_bc_xps);
 170        trace_svc_xprt_free(xprt);
 171        xprt->xpt_ops->xpo_free(xprt);
 172        module_put(owner);
 173}
 174
 175void svc_xprt_put(struct svc_xprt *xprt)
 176{
 177        kref_put(&xprt->xpt_ref, svc_xprt_free);
 178}
 179EXPORT_SYMBOL_GPL(svc_xprt_put);
 180
 181/*
 182 * Called by transport drivers to initialize the transport independent
 183 * portion of the transport instance.
 184 */
 185void svc_xprt_init(struct net *net, struct svc_xprt_class *xcl,
 186                   struct svc_xprt *xprt, struct svc_serv *serv)
 187{
 188        memset(xprt, 0, sizeof(*xprt));
 189        xprt->xpt_class = xcl;
 190        xprt->xpt_ops = xcl->xcl_ops;
 191        kref_init(&xprt->xpt_ref);
 192        xprt->xpt_server = serv;
 193        INIT_LIST_HEAD(&xprt->xpt_list);
 194        INIT_LIST_HEAD(&xprt->xpt_ready);
 195        INIT_LIST_HEAD(&xprt->xpt_deferred);
 196        INIT_LIST_HEAD(&xprt->xpt_users);
 197        mutex_init(&xprt->xpt_mutex);
 198        spin_lock_init(&xprt->xpt_lock);
 199        set_bit(XPT_BUSY, &xprt->xpt_flags);
 200        xprt->xpt_net = get_net(net);
 201        strcpy(xprt->xpt_remotebuf, "uninitialized");
 202}
 203EXPORT_SYMBOL_GPL(svc_xprt_init);
 204
 205static struct svc_xprt *__svc_xpo_create(struct svc_xprt_class *xcl,
 206                                         struct svc_serv *serv,
 207                                         struct net *net,
 208                                         const int family,
 209                                         const unsigned short port,
 210                                         int flags)
 211{
 212        struct sockaddr_in sin = {
 213                .sin_family             = AF_INET,
 214                .sin_addr.s_addr        = htonl(INADDR_ANY),
 215                .sin_port               = htons(port),
 216        };
 217#if IS_ENABLED(CONFIG_IPV6)
 218        struct sockaddr_in6 sin6 = {
 219                .sin6_family            = AF_INET6,
 220                .sin6_addr              = IN6ADDR_ANY_INIT,
 221                .sin6_port              = htons(port),
 222        };
 223#endif
 224        struct svc_xprt *xprt;
 225        struct sockaddr *sap;
 226        size_t len;
 227
 228        switch (family) {
 229        case PF_INET:
 230                sap = (struct sockaddr *)&sin;
 231                len = sizeof(sin);
 232                break;
 233#if IS_ENABLED(CONFIG_IPV6)
 234        case PF_INET6:
 235                sap = (struct sockaddr *)&sin6;
 236                len = sizeof(sin6);
 237                break;
 238#endif
 239        default:
 240                return ERR_PTR(-EAFNOSUPPORT);
 241        }
 242
 243        xprt = xcl->xcl_ops->xpo_create(serv, net, sap, len, flags);
 244        if (IS_ERR(xprt))
 245                trace_svc_xprt_create_err(serv->sv_program->pg_name,
 246                                          xcl->xcl_name, sap, xprt);
 247        return xprt;
 248}
 249
 250/**
 251 * svc_xprt_received - start next receiver thread
 252 * @xprt: controlling transport
 253 *
 254 * The caller must hold the XPT_BUSY bit and must
 255 * not thereafter touch transport data.
 256 *
 257 * Note: XPT_DATA only gets cleared when a read-attempt finds no (or
 258 * insufficient) data.
 259 */
 260void svc_xprt_received(struct svc_xprt *xprt)
 261{
 262        if (!test_bit(XPT_BUSY, &xprt->xpt_flags)) {
 263                WARN_ONCE(1, "xprt=0x%p already busy!", xprt);
 264                return;
 265        }
 266
 267        trace_svc_xprt_received(xprt);
 268
 269        /* As soon as we clear busy, the xprt could be closed and
 270         * 'put', so we need a reference to call svc_enqueue_xprt with:
 271         */
 272        svc_xprt_get(xprt);
 273        smp_mb__before_atomic();
 274        clear_bit(XPT_BUSY, &xprt->xpt_flags);
 275        xprt->xpt_server->sv_ops->svo_enqueue_xprt(xprt);
 276        svc_xprt_put(xprt);
 277}
 278EXPORT_SYMBOL_GPL(svc_xprt_received);
 279
 280void svc_add_new_perm_xprt(struct svc_serv *serv, struct svc_xprt *new)
 281{
 282        clear_bit(XPT_TEMP, &new->xpt_flags);
 283        spin_lock_bh(&serv->sv_lock);
 284        list_add(&new->xpt_list, &serv->sv_permsocks);
 285        spin_unlock_bh(&serv->sv_lock);
 286        svc_xprt_received(new);
 287}
 288
 289static int _svc_create_xprt(struct svc_serv *serv, const char *xprt_name,
 290                            struct net *net, const int family,
 291                            const unsigned short port, int flags,
 292                            const struct cred *cred)
 293{
 294        struct svc_xprt_class *xcl;
 295
 296        spin_lock(&svc_xprt_class_lock);
 297        list_for_each_entry(xcl, &svc_xprt_class_list, xcl_list) {
 298                struct svc_xprt *newxprt;
 299                unsigned short newport;
 300
 301                if (strcmp(xprt_name, xcl->xcl_name))
 302                        continue;
 303
 304                if (!try_module_get(xcl->xcl_owner))
 305                        goto err;
 306
 307                spin_unlock(&svc_xprt_class_lock);
 308                newxprt = __svc_xpo_create(xcl, serv, net, family, port, flags);
 309                if (IS_ERR(newxprt)) {
 310                        module_put(xcl->xcl_owner);
 311                        return PTR_ERR(newxprt);
 312                }
 313                newxprt->xpt_cred = get_cred(cred);
 314                svc_add_new_perm_xprt(serv, newxprt);
 315                newport = svc_xprt_local_port(newxprt);
 316                return newport;
 317        }
 318 err:
 319        spin_unlock(&svc_xprt_class_lock);
 320        /* This errno is exposed to user space.  Provide a reasonable
 321         * perror msg for a bad transport. */
 322        return -EPROTONOSUPPORT;
 323}
 324
 325int svc_create_xprt(struct svc_serv *serv, const char *xprt_name,
 326                    struct net *net, const int family,
 327                    const unsigned short port, int flags,
 328                    const struct cred *cred)
 329{
 330        int err;
 331
 332        err = _svc_create_xprt(serv, xprt_name, net, family, port, flags, cred);
 333        if (err == -EPROTONOSUPPORT) {
 334                request_module("svc%s", xprt_name);
 335                err = _svc_create_xprt(serv, xprt_name, net, family, port, flags, cred);
 336        }
 337        return err;
 338}
 339EXPORT_SYMBOL_GPL(svc_create_xprt);
 340
 341/*
 342 * Copy the local and remote xprt addresses to the rqstp structure
 343 */
 344void svc_xprt_copy_addrs(struct svc_rqst *rqstp, struct svc_xprt *xprt)
 345{
 346        memcpy(&rqstp->rq_addr, &xprt->xpt_remote, xprt->xpt_remotelen);
 347        rqstp->rq_addrlen = xprt->xpt_remotelen;
 348
 349        /*
 350         * Destination address in request is needed for binding the
 351         * source address in RPC replies/callbacks later.
 352         */
 353        memcpy(&rqstp->rq_daddr, &xprt->xpt_local, xprt->xpt_locallen);
 354        rqstp->rq_daddrlen = xprt->xpt_locallen;
 355}
 356EXPORT_SYMBOL_GPL(svc_xprt_copy_addrs);
 357
 358/**
 359 * svc_print_addr - Format rq_addr field for printing
 360 * @rqstp: svc_rqst struct containing address to print
 361 * @buf: target buffer for formatted address
 362 * @len: length of target buffer
 363 *
 364 */
 365char *svc_print_addr(struct svc_rqst *rqstp, char *buf, size_t len)
 366{
 367        return __svc_print_addr(svc_addr(rqstp), buf, len);
 368}
 369EXPORT_SYMBOL_GPL(svc_print_addr);
 370
 371static bool svc_xprt_slots_in_range(struct svc_xprt *xprt)
 372{
 373        unsigned int limit = svc_rpc_per_connection_limit;
 374        int nrqsts = atomic_read(&xprt->xpt_nr_rqsts);
 375
 376        return limit == 0 || (nrqsts >= 0 && nrqsts < limit);
 377}
 378
 379static bool svc_xprt_reserve_slot(struct svc_rqst *rqstp, struct svc_xprt *xprt)
 380{
 381        if (!test_bit(RQ_DATA, &rqstp->rq_flags)) {
 382                if (!svc_xprt_slots_in_range(xprt))
 383                        return false;
 384                atomic_inc(&xprt->xpt_nr_rqsts);
 385                set_bit(RQ_DATA, &rqstp->rq_flags);
 386        }
 387        return true;
 388}
 389
 390static void svc_xprt_release_slot(struct svc_rqst *rqstp)
 391{
 392        struct svc_xprt *xprt = rqstp->rq_xprt;
 393        if (test_and_clear_bit(RQ_DATA, &rqstp->rq_flags)) {
 394                atomic_dec(&xprt->xpt_nr_rqsts);
 395                smp_wmb(); /* See smp_rmb() in svc_xprt_ready() */
 396                svc_xprt_enqueue(xprt);
 397        }
 398}
 399
 400static bool svc_xprt_ready(struct svc_xprt *xprt)
 401{
 402        unsigned long xpt_flags;
 403
 404        /*
 405         * If another cpu has recently updated xpt_flags,
 406         * sk_sock->flags, xpt_reserved, or xpt_nr_rqsts, we need to
 407         * know about it; otherwise it's possible that both that cpu and
 408         * this one could call svc_xprt_enqueue() without either
 409         * svc_xprt_enqueue() recognizing that the conditions below
 410         * are satisfied, and we could stall indefinitely:
 411         */
 412        smp_rmb();
 413        xpt_flags = READ_ONCE(xprt->xpt_flags);
 414
 415        if (xpt_flags & (BIT(XPT_CONN) | BIT(XPT_CLOSE)))
 416                return true;
 417        if (xpt_flags & (BIT(XPT_DATA) | BIT(XPT_DEFERRED))) {
 418                if (xprt->xpt_ops->xpo_has_wspace(xprt) &&
 419                    svc_xprt_slots_in_range(xprt))
 420                        return true;
 421                trace_svc_xprt_no_write_space(xprt);
 422                return false;
 423        }
 424        return false;
 425}
 426
 427void svc_xprt_do_enqueue(struct svc_xprt *xprt)
 428{
 429        struct svc_pool *pool;
 430        struct svc_rqst *rqstp = NULL;
 431        int cpu;
 432
 433        if (!svc_xprt_ready(xprt))
 434                return;
 435
 436        /* Mark transport as busy. It will remain in this state until
 437         * the provider calls svc_xprt_received. We update XPT_BUSY
 438         * atomically because it also guards against trying to enqueue
 439         * the transport twice.
 440         */
 441        if (test_and_set_bit(XPT_BUSY, &xprt->xpt_flags))
 442                return;
 443
 444        cpu = get_cpu();
 445        pool = svc_pool_for_cpu(xprt->xpt_server, cpu);
 446
 447        atomic_long_inc(&pool->sp_stats.packets);
 448
 449        spin_lock_bh(&pool->sp_lock);
 450        list_add_tail(&xprt->xpt_ready, &pool->sp_sockets);
 451        pool->sp_stats.sockets_queued++;
 452        spin_unlock_bh(&pool->sp_lock);
 453
 454        /* find a thread for this xprt */
 455        rcu_read_lock();
 456        list_for_each_entry_rcu(rqstp, &pool->sp_all_threads, rq_all) {
 457                if (test_and_set_bit(RQ_BUSY, &rqstp->rq_flags))
 458                        continue;
 459                atomic_long_inc(&pool->sp_stats.threads_woken);
 460                rqstp->rq_qtime = ktime_get();
 461                wake_up_process(rqstp->rq_task);
 462                goto out_unlock;
 463        }
 464        set_bit(SP_CONGESTED, &pool->sp_flags);
 465        rqstp = NULL;
 466out_unlock:
 467        rcu_read_unlock();
 468        put_cpu();
 469        trace_svc_xprt_do_enqueue(xprt, rqstp);
 470}
 471EXPORT_SYMBOL_GPL(svc_xprt_do_enqueue);
 472
 473/*
 474 * Queue up a transport with data pending. If there are idle nfsd
 475 * processes, wake 'em up.
 476 *
 477 */
 478void svc_xprt_enqueue(struct svc_xprt *xprt)
 479{
 480        if (test_bit(XPT_BUSY, &xprt->xpt_flags))
 481                return;
 482        xprt->xpt_server->sv_ops->svo_enqueue_xprt(xprt);
 483}
 484EXPORT_SYMBOL_GPL(svc_xprt_enqueue);
 485
 486/*
 487 * Dequeue the first transport, if there is one.
 488 */
 489static struct svc_xprt *svc_xprt_dequeue(struct svc_pool *pool)
 490{
 491        struct svc_xprt *xprt = NULL;
 492
 493        if (list_empty(&pool->sp_sockets))
 494                goto out;
 495
 496        spin_lock_bh(&pool->sp_lock);
 497        if (likely(!list_empty(&pool->sp_sockets))) {
 498                xprt = list_first_entry(&pool->sp_sockets,
 499                                        struct svc_xprt, xpt_ready);
 500                list_del_init(&xprt->xpt_ready);
 501                svc_xprt_get(xprt);
 502        }
 503        spin_unlock_bh(&pool->sp_lock);
 504out:
 505        return xprt;
 506}
 507
 508/**
 509 * svc_reserve - change the space reserved for the reply to a request.
 510 * @rqstp:  The request in question
 511 * @space: new max space to reserve
 512 *
 513 * Each request reserves some space on the output queue of the transport
 514 * to make sure the reply fits.  This function reduces that reserved
 515 * space to be the amount of space used already, plus @space.
 516 *
 517 */
 518void svc_reserve(struct svc_rqst *rqstp, int space)
 519{
 520        struct svc_xprt *xprt = rqstp->rq_xprt;
 521
 522        space += rqstp->rq_res.head[0].iov_len;
 523
 524        if (xprt && space < rqstp->rq_reserved) {
 525                atomic_sub((rqstp->rq_reserved - space), &xprt->xpt_reserved);
 526                rqstp->rq_reserved = space;
 527                smp_wmb(); /* See smp_rmb() in svc_xprt_ready() */
 528                svc_xprt_enqueue(xprt);
 529        }
 530}
 531EXPORT_SYMBOL_GPL(svc_reserve);
 532
 533static void svc_xprt_release(struct svc_rqst *rqstp)
 534{
 535        struct svc_xprt *xprt = rqstp->rq_xprt;
 536
 537        xprt->xpt_ops->xpo_release_rqst(rqstp);
 538
 539        kfree(rqstp->rq_deferred);
 540        rqstp->rq_deferred = NULL;
 541
 542        svc_free_res_pages(rqstp);
 543        rqstp->rq_res.page_len = 0;
 544        rqstp->rq_res.page_base = 0;
 545
 546        /* Reset response buffer and release
 547         * the reservation.
 548         * But first, check that enough space was reserved
 549         * for the reply, otherwise we have a bug!
 550         */
 551        if ((rqstp->rq_res.len) >  rqstp->rq_reserved)
 552                printk(KERN_ERR "RPC request reserved %d but used %d\n",
 553                       rqstp->rq_reserved,
 554                       rqstp->rq_res.len);
 555
 556        rqstp->rq_res.head[0].iov_len = 0;
 557        svc_reserve(rqstp, 0);
 558        svc_xprt_release_slot(rqstp);
 559        rqstp->rq_xprt = NULL;
 560        svc_xprt_put(xprt);
 561}
 562
 563/*
 564 * Some svc_serv's will have occasional work to do, even when a xprt is not
 565 * waiting to be serviced. This function is there to "kick" a task in one of
 566 * those services so that it can wake up and do that work. Note that we only
 567 * bother with pool 0 as we don't need to wake up more than one thread for
 568 * this purpose.
 569 */
 570void svc_wake_up(struct svc_serv *serv)
 571{
 572        struct svc_rqst *rqstp;
 573        struct svc_pool *pool;
 574
 575        pool = &serv->sv_pools[0];
 576
 577        rcu_read_lock();
 578        list_for_each_entry_rcu(rqstp, &pool->sp_all_threads, rq_all) {
 579                /* skip any that aren't queued */
 580                if (test_bit(RQ_BUSY, &rqstp->rq_flags))
 581                        continue;
 582                rcu_read_unlock();
 583                wake_up_process(rqstp->rq_task);
 584                trace_svc_wake_up(rqstp->rq_task->pid);
 585                return;
 586        }
 587        rcu_read_unlock();
 588
 589        /* No free entries available */
 590        set_bit(SP_TASK_PENDING, &pool->sp_flags);
 591        smp_wmb();
 592        trace_svc_wake_up(0);
 593}
 594EXPORT_SYMBOL_GPL(svc_wake_up);
 595
 596int svc_port_is_privileged(struct sockaddr *sin)
 597{
 598        switch (sin->sa_family) {
 599        case AF_INET:
 600                return ntohs(((struct sockaddr_in *)sin)->sin_port)
 601                        < PROT_SOCK;
 602        case AF_INET6:
 603                return ntohs(((struct sockaddr_in6 *)sin)->sin6_port)
 604                        < PROT_SOCK;
 605        default:
 606                return 0;
 607        }
 608}
 609
 610/*
 611 * Make sure that we don't have too many active connections. If we have,
 612 * something must be dropped. It's not clear what will happen if we allow
 613 * "too many" connections, but when dealing with network-facing software,
 614 * we have to code defensively. Here we do that by imposing hard limits.
 615 *
 616 * There's no point in trying to do random drop here for DoS
 617 * prevention. The NFS clients does 1 reconnect in 15 seconds. An
 618 * attacker can easily beat that.
 619 *
 620 * The only somewhat efficient mechanism would be if drop old
 621 * connections from the same IP first. But right now we don't even
 622 * record the client IP in svc_sock.
 623 *
 624 * single-threaded services that expect a lot of clients will probably
 625 * need to set sv_maxconn to override the default value which is based
 626 * on the number of threads
 627 */
 628static void svc_check_conn_limits(struct svc_serv *serv)
 629{
 630        unsigned int limit = serv->sv_maxconn ? serv->sv_maxconn :
 631                                (serv->sv_nrthreads+3) * 20;
 632
 633        if (serv->sv_tmpcnt > limit) {
 634                struct svc_xprt *xprt = NULL;
 635                spin_lock_bh(&serv->sv_lock);
 636                if (!list_empty(&serv->sv_tempsocks)) {
 637                        /* Try to help the admin */
 638                        net_notice_ratelimited("%s: too many open connections, consider increasing the %s\n",
 639                                               serv->sv_name, serv->sv_maxconn ?
 640                                               "max number of connections" :
 641                                               "number of threads");
 642                        /*
 643                         * Always select the oldest connection. It's not fair,
 644                         * but so is life
 645                         */
 646                        xprt = list_entry(serv->sv_tempsocks.prev,
 647                                          struct svc_xprt,
 648                                          xpt_list);
 649                        set_bit(XPT_CLOSE, &xprt->xpt_flags);
 650                        svc_xprt_get(xprt);
 651                }
 652                spin_unlock_bh(&serv->sv_lock);
 653
 654                if (xprt) {
 655                        svc_xprt_enqueue(xprt);
 656                        svc_xprt_put(xprt);
 657                }
 658        }
 659}
 660
 661static int svc_alloc_arg(struct svc_rqst *rqstp)
 662{
 663        struct svc_serv *serv = rqstp->rq_server;
 664        struct xdr_buf *arg = &rqstp->rq_arg;
 665        unsigned long pages, filled;
 666
 667        pages = (serv->sv_max_mesg + 2 * PAGE_SIZE) >> PAGE_SHIFT;
 668        if (pages > RPCSVC_MAXPAGES) {
 669                pr_warn_once("svc: warning: pages=%lu > RPCSVC_MAXPAGES=%lu\n",
 670                             pages, RPCSVC_MAXPAGES);
 671                /* use as many pages as possible */
 672                pages = RPCSVC_MAXPAGES;
 673        }
 674
 675        for (;;) {
 676                filled = alloc_pages_bulk_array(GFP_KERNEL, pages,
 677                                                rqstp->rq_pages);
 678                if (filled == pages)
 679                        break;
 680
 681                set_current_state(TASK_INTERRUPTIBLE);
 682                if (signalled() || kthread_should_stop()) {
 683                        set_current_state(TASK_RUNNING);
 684                        return -EINTR;
 685                }
 686                schedule_timeout(msecs_to_jiffies(500));
 687        }
 688        rqstp->rq_page_end = &rqstp->rq_pages[pages];
 689        rqstp->rq_pages[pages] = NULL; /* this might be seen in nfsd_splice_actor() */
 690
 691        /* Make arg->head point to first page and arg->pages point to rest */
 692        arg->head[0].iov_base = page_address(rqstp->rq_pages[0]);
 693        arg->head[0].iov_len = PAGE_SIZE;
 694        arg->pages = rqstp->rq_pages + 1;
 695        arg->page_base = 0;
 696        /* save at least one page for response */
 697        arg->page_len = (pages-2)*PAGE_SIZE;
 698        arg->len = (pages-1)*PAGE_SIZE;
 699        arg->tail[0].iov_len = 0;
 700        return 0;
 701}
 702
 703static bool
 704rqst_should_sleep(struct svc_rqst *rqstp)
 705{
 706        struct svc_pool         *pool = rqstp->rq_pool;
 707
 708        /* did someone call svc_wake_up? */
 709        if (test_and_clear_bit(SP_TASK_PENDING, &pool->sp_flags))
 710                return false;
 711
 712        /* was a socket queued? */
 713        if (!list_empty(&pool->sp_sockets))
 714                return false;
 715
 716        /* are we shutting down? */
 717        if (signalled() || kthread_should_stop())
 718                return false;
 719
 720        /* are we freezing? */
 721        if (freezing(current))
 722                return false;
 723
 724        return true;
 725}
 726
 727static struct svc_xprt *svc_get_next_xprt(struct svc_rqst *rqstp, long timeout)
 728{
 729        struct svc_pool         *pool = rqstp->rq_pool;
 730        long                    time_left = 0;
 731
 732        /* rq_xprt should be clear on entry */
 733        WARN_ON_ONCE(rqstp->rq_xprt);
 734
 735        rqstp->rq_xprt = svc_xprt_dequeue(pool);
 736        if (rqstp->rq_xprt)
 737                goto out_found;
 738
 739        /*
 740         * We have to be able to interrupt this wait
 741         * to bring down the daemons ...
 742         */
 743        set_current_state(TASK_INTERRUPTIBLE);
 744        smp_mb__before_atomic();
 745        clear_bit(SP_CONGESTED, &pool->sp_flags);
 746        clear_bit(RQ_BUSY, &rqstp->rq_flags);
 747        smp_mb__after_atomic();
 748
 749        if (likely(rqst_should_sleep(rqstp)))
 750                time_left = schedule_timeout(timeout);
 751        else
 752                __set_current_state(TASK_RUNNING);
 753
 754        try_to_freeze();
 755
 756        set_bit(RQ_BUSY, &rqstp->rq_flags);
 757        smp_mb__after_atomic();
 758        rqstp->rq_xprt = svc_xprt_dequeue(pool);
 759        if (rqstp->rq_xprt)
 760                goto out_found;
 761
 762        if (!time_left)
 763                atomic_long_inc(&pool->sp_stats.threads_timedout);
 764
 765        if (signalled() || kthread_should_stop())
 766                return ERR_PTR(-EINTR);
 767        return ERR_PTR(-EAGAIN);
 768out_found:
 769        /* Normally we will wait up to 5 seconds for any required
 770         * cache information to be provided.
 771         */
 772        if (!test_bit(SP_CONGESTED, &pool->sp_flags))
 773                rqstp->rq_chandle.thread_wait = 5*HZ;
 774        else
 775                rqstp->rq_chandle.thread_wait = 1*HZ;
 776        trace_svc_xprt_dequeue(rqstp);
 777        return rqstp->rq_xprt;
 778}
 779
 780static void svc_add_new_temp_xprt(struct svc_serv *serv, struct svc_xprt *newxpt)
 781{
 782        spin_lock_bh(&serv->sv_lock);
 783        set_bit(XPT_TEMP, &newxpt->xpt_flags);
 784        list_add(&newxpt->xpt_list, &serv->sv_tempsocks);
 785        serv->sv_tmpcnt++;
 786        if (serv->sv_temptimer.function == NULL) {
 787                /* setup timer to age temp transports */
 788                serv->sv_temptimer.function = svc_age_temp_xprts;
 789                mod_timer(&serv->sv_temptimer,
 790                          jiffies + svc_conn_age_period * HZ);
 791        }
 792        spin_unlock_bh(&serv->sv_lock);
 793        svc_xprt_received(newxpt);
 794}
 795
 796static int svc_handle_xprt(struct svc_rqst *rqstp, struct svc_xprt *xprt)
 797{
 798        struct svc_serv *serv = rqstp->rq_server;
 799        int len = 0;
 800
 801        if (test_bit(XPT_CLOSE, &xprt->xpt_flags)) {
 802                if (test_and_clear_bit(XPT_KILL_TEMP, &xprt->xpt_flags))
 803                        xprt->xpt_ops->xpo_kill_temp_xprt(xprt);
 804                svc_delete_xprt(xprt);
 805                /* Leave XPT_BUSY set on the dead xprt: */
 806                goto out;
 807        }
 808        if (test_bit(XPT_LISTENER, &xprt->xpt_flags)) {
 809                struct svc_xprt *newxpt;
 810                /*
 811                 * We know this module_get will succeed because the
 812                 * listener holds a reference too
 813                 */
 814                __module_get(xprt->xpt_class->xcl_owner);
 815                svc_check_conn_limits(xprt->xpt_server);
 816                newxpt = xprt->xpt_ops->xpo_accept(xprt);
 817                if (newxpt) {
 818                        newxpt->xpt_cred = get_cred(xprt->xpt_cred);
 819                        svc_add_new_temp_xprt(serv, newxpt);
 820                        trace_svc_xprt_accept(newxpt, serv->sv_name);
 821                } else {
 822                        module_put(xprt->xpt_class->xcl_owner);
 823                }
 824                svc_xprt_received(xprt);
 825        } else if (svc_xprt_reserve_slot(rqstp, xprt)) {
 826                /* XPT_DATA|XPT_DEFERRED case: */
 827                dprintk("svc: server %p, pool %u, transport %p, inuse=%d\n",
 828                        rqstp, rqstp->rq_pool->sp_id, xprt,
 829                        kref_read(&xprt->xpt_ref));
 830                rqstp->rq_deferred = svc_deferred_dequeue(xprt);
 831                if (rqstp->rq_deferred)
 832                        len = svc_deferred_recv(rqstp);
 833                else
 834                        len = xprt->xpt_ops->xpo_recvfrom(rqstp);
 835                rqstp->rq_stime = ktime_get();
 836                rqstp->rq_reserved = serv->sv_max_mesg;
 837                atomic_add(rqstp->rq_reserved, &xprt->xpt_reserved);
 838        } else
 839                svc_xprt_received(xprt);
 840out:
 841        trace_svc_handle_xprt(xprt, len);
 842        return len;
 843}
 844
 845/*
 846 * Receive the next request on any transport.  This code is carefully
 847 * organised not to touch any cachelines in the shared svc_serv
 848 * structure, only cachelines in the local svc_pool.
 849 */
 850int svc_recv(struct svc_rqst *rqstp, long timeout)
 851{
 852        struct svc_xprt         *xprt = NULL;
 853        struct svc_serv         *serv = rqstp->rq_server;
 854        int                     len, err;
 855
 856        err = svc_alloc_arg(rqstp);
 857        if (err)
 858                goto out;
 859
 860        try_to_freeze();
 861        cond_resched();
 862        err = -EINTR;
 863        if (signalled() || kthread_should_stop())
 864                goto out;
 865
 866        xprt = svc_get_next_xprt(rqstp, timeout);
 867        if (IS_ERR(xprt)) {
 868                err = PTR_ERR(xprt);
 869                goto out;
 870        }
 871
 872        len = svc_handle_xprt(rqstp, xprt);
 873
 874        /* No data, incomplete (TCP) read, or accept() */
 875        err = -EAGAIN;
 876        if (len <= 0)
 877                goto out_release;
 878        trace_svc_xdr_recvfrom(&rqstp->rq_arg);
 879
 880        clear_bit(XPT_OLD, &xprt->xpt_flags);
 881
 882        xprt->xpt_ops->xpo_secure_port(rqstp);
 883        rqstp->rq_chandle.defer = svc_defer;
 884        rqstp->rq_xid = svc_getu32(&rqstp->rq_arg.head[0]);
 885
 886        if (serv->sv_stats)
 887                serv->sv_stats->netcnt++;
 888        return len;
 889out_release:
 890        rqstp->rq_res.len = 0;
 891        svc_xprt_release(rqstp);
 892out:
 893        return err;
 894}
 895EXPORT_SYMBOL_GPL(svc_recv);
 896
 897/*
 898 * Drop request
 899 */
 900void svc_drop(struct svc_rqst *rqstp)
 901{
 902        trace_svc_drop(rqstp);
 903        svc_xprt_release(rqstp);
 904}
 905EXPORT_SYMBOL_GPL(svc_drop);
 906
 907/*
 908 * Return reply to client.
 909 */
 910int svc_send(struct svc_rqst *rqstp)
 911{
 912        struct svc_xprt *xprt;
 913        int             len = -EFAULT;
 914        struct xdr_buf  *xb;
 915
 916        xprt = rqstp->rq_xprt;
 917        if (!xprt)
 918                goto out;
 919
 920        /* calculate over-all length */
 921        xb = &rqstp->rq_res;
 922        xb->len = xb->head[0].iov_len +
 923                xb->page_len +
 924                xb->tail[0].iov_len;
 925        trace_svc_xdr_sendto(rqstp->rq_xid, xb);
 926        trace_svc_stats_latency(rqstp);
 927
 928        len = xprt->xpt_ops->xpo_sendto(rqstp);
 929
 930        trace_svc_send(rqstp, len);
 931        svc_xprt_release(rqstp);
 932
 933        if (len == -ECONNREFUSED || len == -ENOTCONN || len == -EAGAIN)
 934                len = 0;
 935out:
 936        return len;
 937}
 938
 939/*
 940 * Timer function to close old temporary transports, using
 941 * a mark-and-sweep algorithm.
 942 */
 943static void svc_age_temp_xprts(struct timer_list *t)
 944{
 945        struct svc_serv *serv = from_timer(serv, t, sv_temptimer);
 946        struct svc_xprt *xprt;
 947        struct list_head *le, *next;
 948
 949        dprintk("svc_age_temp_xprts\n");
 950
 951        if (!spin_trylock_bh(&serv->sv_lock)) {
 952                /* busy, try again 1 sec later */
 953                dprintk("svc_age_temp_xprts: busy\n");
 954                mod_timer(&serv->sv_temptimer, jiffies + HZ);
 955                return;
 956        }
 957
 958        list_for_each_safe(le, next, &serv->sv_tempsocks) {
 959                xprt = list_entry(le, struct svc_xprt, xpt_list);
 960
 961                /* First time through, just mark it OLD. Second time
 962                 * through, close it. */
 963                if (!test_and_set_bit(XPT_OLD, &xprt->xpt_flags))
 964                        continue;
 965                if (kref_read(&xprt->xpt_ref) > 1 ||
 966                    test_bit(XPT_BUSY, &xprt->xpt_flags))
 967                        continue;
 968                list_del_init(le);
 969                set_bit(XPT_CLOSE, &xprt->xpt_flags);
 970                dprintk("queuing xprt %p for closing\n", xprt);
 971
 972                /* a thread will dequeue and close it soon */
 973                svc_xprt_enqueue(xprt);
 974        }
 975        spin_unlock_bh(&serv->sv_lock);
 976
 977        mod_timer(&serv->sv_temptimer, jiffies + svc_conn_age_period * HZ);
 978}
 979
 980/* Close temporary transports whose xpt_local matches server_addr immediately
 981 * instead of waiting for them to be picked up by the timer.
 982 *
 983 * This is meant to be called from a notifier_block that runs when an ip
 984 * address is deleted.
 985 */
 986void svc_age_temp_xprts_now(struct svc_serv *serv, struct sockaddr *server_addr)
 987{
 988        struct svc_xprt *xprt;
 989        struct list_head *le, *next;
 990        LIST_HEAD(to_be_closed);
 991
 992        spin_lock_bh(&serv->sv_lock);
 993        list_for_each_safe(le, next, &serv->sv_tempsocks) {
 994                xprt = list_entry(le, struct svc_xprt, xpt_list);
 995                if (rpc_cmp_addr(server_addr, (struct sockaddr *)
 996                                &xprt->xpt_local)) {
 997                        dprintk("svc_age_temp_xprts_now: found %p\n", xprt);
 998                        list_move(le, &to_be_closed);
 999                }
1000        }
1001        spin_unlock_bh(&serv->sv_lock);
1002
1003        while (!list_empty(&to_be_closed)) {
1004                le = to_be_closed.next;
1005                list_del_init(le);
1006                xprt = list_entry(le, struct svc_xprt, xpt_list);
1007                set_bit(XPT_CLOSE, &xprt->xpt_flags);
1008                set_bit(XPT_KILL_TEMP, &xprt->xpt_flags);
1009                dprintk("svc_age_temp_xprts_now: queuing xprt %p for closing\n",
1010                                xprt);
1011                svc_xprt_enqueue(xprt);
1012        }
1013}
1014EXPORT_SYMBOL_GPL(svc_age_temp_xprts_now);
1015
1016static void call_xpt_users(struct svc_xprt *xprt)
1017{
1018        struct svc_xpt_user *u;
1019
1020        spin_lock(&xprt->xpt_lock);
1021        while (!list_empty(&xprt->xpt_users)) {
1022                u = list_first_entry(&xprt->xpt_users, struct svc_xpt_user, list);
1023                list_del_init(&u->list);
1024                u->callback(u);
1025        }
1026        spin_unlock(&xprt->xpt_lock);
1027}
1028
1029/*
1030 * Remove a dead transport
1031 */
1032static void svc_delete_xprt(struct svc_xprt *xprt)
1033{
1034        struct svc_serv *serv = xprt->xpt_server;
1035        struct svc_deferred_req *dr;
1036
1037        if (test_and_set_bit(XPT_DEAD, &xprt->xpt_flags))
1038                return;
1039
1040        trace_svc_xprt_detach(xprt);
1041        xprt->xpt_ops->xpo_detach(xprt);
1042        if (xprt->xpt_bc_xprt)
1043                xprt->xpt_bc_xprt->ops->close(xprt->xpt_bc_xprt);
1044
1045        spin_lock_bh(&serv->sv_lock);
1046        list_del_init(&xprt->xpt_list);
1047        WARN_ON_ONCE(!list_empty(&xprt->xpt_ready));
1048        if (test_bit(XPT_TEMP, &xprt->xpt_flags))
1049                serv->sv_tmpcnt--;
1050        spin_unlock_bh(&serv->sv_lock);
1051
1052        while ((dr = svc_deferred_dequeue(xprt)) != NULL)
1053                kfree(dr);
1054
1055        call_xpt_users(xprt);
1056        svc_xprt_put(xprt);
1057}
1058
1059void svc_close_xprt(struct svc_xprt *xprt)
1060{
1061        trace_svc_xprt_close(xprt);
1062        set_bit(XPT_CLOSE, &xprt->xpt_flags);
1063        if (test_and_set_bit(XPT_BUSY, &xprt->xpt_flags))
1064                /* someone else will have to effect the close */
1065                return;
1066        /*
1067         * We expect svc_close_xprt() to work even when no threads are
1068         * running (e.g., while configuring the server before starting
1069         * any threads), so if the transport isn't busy, we delete
1070         * it ourself:
1071         */
1072        svc_delete_xprt(xprt);
1073}
1074EXPORT_SYMBOL_GPL(svc_close_xprt);
1075
1076static int svc_close_list(struct svc_serv *serv, struct list_head *xprt_list, struct net *net)
1077{
1078        struct svc_xprt *xprt;
1079        int ret = 0;
1080
1081        spin_lock_bh(&serv->sv_lock);
1082        list_for_each_entry(xprt, xprt_list, xpt_list) {
1083                if (xprt->xpt_net != net)
1084                        continue;
1085                ret++;
1086                set_bit(XPT_CLOSE, &xprt->xpt_flags);
1087                svc_xprt_enqueue(xprt);
1088        }
1089        spin_unlock_bh(&serv->sv_lock);
1090        return ret;
1091}
1092
1093static struct svc_xprt *svc_dequeue_net(struct svc_serv *serv, struct net *net)
1094{
1095        struct svc_pool *pool;
1096        struct svc_xprt *xprt;
1097        struct svc_xprt *tmp;
1098        int i;
1099
1100        for (i = 0; i < serv->sv_nrpools; i++) {
1101                pool = &serv->sv_pools[i];
1102
1103                spin_lock_bh(&pool->sp_lock);
1104                list_for_each_entry_safe(xprt, tmp, &pool->sp_sockets, xpt_ready) {
1105                        if (xprt->xpt_net != net)
1106                                continue;
1107                        list_del_init(&xprt->xpt_ready);
1108                        spin_unlock_bh(&pool->sp_lock);
1109                        return xprt;
1110                }
1111                spin_unlock_bh(&pool->sp_lock);
1112        }
1113        return NULL;
1114}
1115
1116static void svc_clean_up_xprts(struct svc_serv *serv, struct net *net)
1117{
1118        struct svc_xprt *xprt;
1119
1120        while ((xprt = svc_dequeue_net(serv, net))) {
1121                set_bit(XPT_CLOSE, &xprt->xpt_flags);
1122                svc_delete_xprt(xprt);
1123        }
1124}
1125
1126/*
1127 * Server threads may still be running (especially in the case where the
1128 * service is still running in other network namespaces).
1129 *
1130 * So we shut down sockets the same way we would on a running server, by
1131 * setting XPT_CLOSE, enqueuing, and letting a thread pick it up to do
1132 * the close.  In the case there are no such other threads,
1133 * threads running, svc_clean_up_xprts() does a simple version of a
1134 * server's main event loop, and in the case where there are other
1135 * threads, we may need to wait a little while and then check again to
1136 * see if they're done.
1137 */
1138void svc_close_net(struct svc_serv *serv, struct net *net)
1139{
1140        int delay = 0;
1141
1142        while (svc_close_list(serv, &serv->sv_permsocks, net) +
1143               svc_close_list(serv, &serv->sv_tempsocks, net)) {
1144
1145                svc_clean_up_xprts(serv, net);
1146                msleep(delay++);
1147        }
1148}
1149
1150/*
1151 * Handle defer and revisit of requests
1152 */
1153
1154static void svc_revisit(struct cache_deferred_req *dreq, int too_many)
1155{
1156        struct svc_deferred_req *dr =
1157                container_of(dreq, struct svc_deferred_req, handle);
1158        struct svc_xprt *xprt = dr->xprt;
1159
1160        spin_lock(&xprt->xpt_lock);
1161        set_bit(XPT_DEFERRED, &xprt->xpt_flags);
1162        if (too_many || test_bit(XPT_DEAD, &xprt->xpt_flags)) {
1163                spin_unlock(&xprt->xpt_lock);
1164                trace_svc_defer_drop(dr);
1165                svc_xprt_put(xprt);
1166                kfree(dr);
1167                return;
1168        }
1169        dr->xprt = NULL;
1170        list_add(&dr->handle.recent, &xprt->xpt_deferred);
1171        spin_unlock(&xprt->xpt_lock);
1172        trace_svc_defer_queue(dr);
1173        svc_xprt_enqueue(xprt);
1174        svc_xprt_put(xprt);
1175}
1176
1177/*
1178 * Save the request off for later processing. The request buffer looks
1179 * like this:
1180 *
1181 * <xprt-header><rpc-header><rpc-pagelist><rpc-tail>
1182 *
1183 * This code can only handle requests that consist of an xprt-header
1184 * and rpc-header.
1185 */
1186static struct cache_deferred_req *svc_defer(struct cache_req *req)
1187{
1188        struct svc_rqst *rqstp = container_of(req, struct svc_rqst, rq_chandle);
1189        struct svc_deferred_req *dr;
1190
1191        if (rqstp->rq_arg.page_len || !test_bit(RQ_USEDEFERRAL, &rqstp->rq_flags))
1192                return NULL; /* if more than a page, give up FIXME */
1193        if (rqstp->rq_deferred) {
1194                dr = rqstp->rq_deferred;
1195                rqstp->rq_deferred = NULL;
1196        } else {
1197                size_t skip;
1198                size_t size;
1199                /* FIXME maybe discard if size too large */
1200                size = sizeof(struct svc_deferred_req) + rqstp->rq_arg.len;
1201                dr = kmalloc(size, GFP_KERNEL);
1202                if (dr == NULL)
1203                        return NULL;
1204
1205                dr->handle.owner = rqstp->rq_server;
1206                dr->prot = rqstp->rq_prot;
1207                memcpy(&dr->addr, &rqstp->rq_addr, rqstp->rq_addrlen);
1208                dr->addrlen = rqstp->rq_addrlen;
1209                dr->daddr = rqstp->rq_daddr;
1210                dr->argslen = rqstp->rq_arg.len >> 2;
1211                dr->xprt_hlen = rqstp->rq_xprt_hlen;
1212
1213                /* back up head to the start of the buffer and copy */
1214                skip = rqstp->rq_arg.len - rqstp->rq_arg.head[0].iov_len;
1215                memcpy(dr->args, rqstp->rq_arg.head[0].iov_base - skip,
1216                       dr->argslen << 2);
1217        }
1218        trace_svc_defer(rqstp);
1219        svc_xprt_get(rqstp->rq_xprt);
1220        dr->xprt = rqstp->rq_xprt;
1221        set_bit(RQ_DROPME, &rqstp->rq_flags);
1222
1223        dr->handle.revisit = svc_revisit;
1224        return &dr->handle;
1225}
1226
1227/*
1228 * recv data from a deferred request into an active one
1229 */
1230static noinline int svc_deferred_recv(struct svc_rqst *rqstp)
1231{
1232        struct svc_deferred_req *dr = rqstp->rq_deferred;
1233
1234        trace_svc_defer_recv(dr);
1235
1236        /* setup iov_base past transport header */
1237        rqstp->rq_arg.head[0].iov_base = dr->args + (dr->xprt_hlen>>2);
1238        /* The iov_len does not include the transport header bytes */
1239        rqstp->rq_arg.head[0].iov_len = (dr->argslen<<2) - dr->xprt_hlen;
1240        rqstp->rq_arg.page_len = 0;
1241        /* The rq_arg.len includes the transport header bytes */
1242        rqstp->rq_arg.len     = dr->argslen<<2;
1243        rqstp->rq_prot        = dr->prot;
1244        memcpy(&rqstp->rq_addr, &dr->addr, dr->addrlen);
1245        rqstp->rq_addrlen     = dr->addrlen;
1246        /* Save off transport header len in case we get deferred again */
1247        rqstp->rq_xprt_hlen   = dr->xprt_hlen;
1248        rqstp->rq_daddr       = dr->daddr;
1249        rqstp->rq_respages    = rqstp->rq_pages;
1250        svc_xprt_received(rqstp->rq_xprt);
1251        return (dr->argslen<<2) - dr->xprt_hlen;
1252}
1253
1254
1255static struct svc_deferred_req *svc_deferred_dequeue(struct svc_xprt *xprt)
1256{
1257        struct svc_deferred_req *dr = NULL;
1258
1259        if (!test_bit(XPT_DEFERRED, &xprt->xpt_flags))
1260                return NULL;
1261        spin_lock(&xprt->xpt_lock);
1262        if (!list_empty(&xprt->xpt_deferred)) {
1263                dr = list_entry(xprt->xpt_deferred.next,
1264                                struct svc_deferred_req,
1265                                handle.recent);
1266                list_del_init(&dr->handle.recent);
1267        } else
1268                clear_bit(XPT_DEFERRED, &xprt->xpt_flags);
1269        spin_unlock(&xprt->xpt_lock);
1270        return dr;
1271}
1272
1273/**
1274 * svc_find_xprt - find an RPC transport instance
1275 * @serv: pointer to svc_serv to search
1276 * @xcl_name: C string containing transport's class name
1277 * @net: owner net pointer
1278 * @af: Address family of transport's local address
1279 * @port: transport's IP port number
1280 *
1281 * Return the transport instance pointer for the endpoint accepting
1282 * connections/peer traffic from the specified transport class,
1283 * address family and port.
1284 *
1285 * Specifying 0 for the address family or port is effectively a
1286 * wild-card, and will result in matching the first transport in the
1287 * service's list that has a matching class name.
1288 */
1289struct svc_xprt *svc_find_xprt(struct svc_serv *serv, const char *xcl_name,
1290                               struct net *net, const sa_family_t af,
1291                               const unsigned short port)
1292{
1293        struct svc_xprt *xprt;
1294        struct svc_xprt *found = NULL;
1295
1296        /* Sanity check the args */
1297        if (serv == NULL || xcl_name == NULL)
1298                return found;
1299
1300        spin_lock_bh(&serv->sv_lock);
1301        list_for_each_entry(xprt, &serv->sv_permsocks, xpt_list) {
1302                if (xprt->xpt_net != net)
1303                        continue;
1304                if (strcmp(xprt->xpt_class->xcl_name, xcl_name))
1305                        continue;
1306                if (af != AF_UNSPEC && af != xprt->xpt_local.ss_family)
1307                        continue;
1308                if (port != 0 && port != svc_xprt_local_port(xprt))
1309                        continue;
1310                found = xprt;
1311                svc_xprt_get(xprt);
1312                break;
1313        }
1314        spin_unlock_bh(&serv->sv_lock);
1315        return found;
1316}
1317EXPORT_SYMBOL_GPL(svc_find_xprt);
1318
1319static int svc_one_xprt_name(const struct svc_xprt *xprt,
1320                             char *pos, int remaining)
1321{
1322        int len;
1323
1324        len = snprintf(pos, remaining, "%s %u\n",
1325                        xprt->xpt_class->xcl_name,
1326                        svc_xprt_local_port(xprt));
1327        if (len >= remaining)
1328                return -ENAMETOOLONG;
1329        return len;
1330}
1331
1332/**
1333 * svc_xprt_names - format a buffer with a list of transport names
1334 * @serv: pointer to an RPC service
1335 * @buf: pointer to a buffer to be filled in
1336 * @buflen: length of buffer to be filled in
1337 *
1338 * Fills in @buf with a string containing a list of transport names,
1339 * each name terminated with '\n'.
1340 *
1341 * Returns positive length of the filled-in string on success; otherwise
1342 * a negative errno value is returned if an error occurs.
1343 */
1344int svc_xprt_names(struct svc_serv *serv, char *buf, const int buflen)
1345{
1346        struct svc_xprt *xprt;
1347        int len, totlen;
1348        char *pos;
1349
1350        /* Sanity check args */
1351        if (!serv)
1352                return 0;
1353
1354        spin_lock_bh(&serv->sv_lock);
1355
1356        pos = buf;
1357        totlen = 0;
1358        list_for_each_entry(xprt, &serv->sv_permsocks, xpt_list) {
1359                len = svc_one_xprt_name(xprt, pos, buflen - totlen);
1360                if (len < 0) {
1361                        *buf = '\0';
1362                        totlen = len;
1363                }
1364                if (len <= 0)
1365                        break;
1366
1367                pos += len;
1368                totlen += len;
1369        }
1370
1371        spin_unlock_bh(&serv->sv_lock);
1372        return totlen;
1373}
1374EXPORT_SYMBOL_GPL(svc_xprt_names);
1375
1376
1377/*----------------------------------------------------------------------------*/
1378
1379static void *svc_pool_stats_start(struct seq_file *m, loff_t *pos)
1380{
1381        unsigned int pidx = (unsigned int)*pos;
1382        struct svc_serv *serv = m->private;
1383
1384        dprintk("svc_pool_stats_start, *pidx=%u\n", pidx);
1385
1386        if (!pidx)
1387                return SEQ_START_TOKEN;
1388        return (pidx > serv->sv_nrpools ? NULL : &serv->sv_pools[pidx-1]);
1389}
1390
1391static void *svc_pool_stats_next(struct seq_file *m, void *p, loff_t *pos)
1392{
1393        struct svc_pool *pool = p;
1394        struct svc_serv *serv = m->private;
1395
1396        dprintk("svc_pool_stats_next, *pos=%llu\n", *pos);
1397
1398        if (p == SEQ_START_TOKEN) {
1399                pool = &serv->sv_pools[0];
1400        } else {
1401                unsigned int pidx = (pool - &serv->sv_pools[0]);
1402                if (pidx < serv->sv_nrpools-1)
1403                        pool = &serv->sv_pools[pidx+1];
1404                else
1405                        pool = NULL;
1406        }
1407        ++*pos;
1408        return pool;
1409}
1410
1411static void svc_pool_stats_stop(struct seq_file *m, void *p)
1412{
1413}
1414
1415static int svc_pool_stats_show(struct seq_file *m, void *p)
1416{
1417        struct svc_pool *pool = p;
1418
1419        if (p == SEQ_START_TOKEN) {
1420                seq_puts(m, "# pool packets-arrived sockets-enqueued threads-woken threads-timedout\n");
1421                return 0;
1422        }
1423
1424        seq_printf(m, "%u %lu %lu %lu %lu\n",
1425                pool->sp_id,
1426                (unsigned long)atomic_long_read(&pool->sp_stats.packets),
1427                pool->sp_stats.sockets_queued,
1428                (unsigned long)atomic_long_read(&pool->sp_stats.threads_woken),
1429                (unsigned long)atomic_long_read(&pool->sp_stats.threads_timedout));
1430
1431        return 0;
1432}
1433
1434static const struct seq_operations svc_pool_stats_seq_ops = {
1435        .start  = svc_pool_stats_start,
1436        .next   = svc_pool_stats_next,
1437        .stop   = svc_pool_stats_stop,
1438        .show   = svc_pool_stats_show,
1439};
1440
1441int svc_pool_stats_open(struct svc_serv *serv, struct file *file)
1442{
1443        int err;
1444
1445        err = seq_open(file, &svc_pool_stats_seq_ops);
1446        if (!err)
1447                ((struct seq_file *) file->private_data)->private = serv;
1448        return err;
1449}
1450EXPORT_SYMBOL(svc_pool_stats_open);
1451
1452/*----------------------------------------------------------------------------*/
1453