linux/net/sunrpc/svc.c
<<
>>
Prefs
   1// SPDX-License-Identifier: GPL-2.0-only
   2/*
   3 * linux/net/sunrpc/svc.c
   4 *
   5 * High-level RPC service routines
   6 *
   7 * Copyright (C) 1995, 1996 Olaf Kirch <okir@monad.swb.de>
   8 *
   9 * Multiple threads pools and NUMAisation
  10 * Copyright (c) 2006 Silicon Graphics, Inc.
  11 * by Greg Banks <gnb@melbourne.sgi.com>
  12 */
  13
  14#include <linux/linkage.h>
  15#include <linux/sched/signal.h>
  16#include <linux/errno.h>
  17#include <linux/net.h>
  18#include <linux/in.h>
  19#include <linux/mm.h>
  20#include <linux/interrupt.h>
  21#include <linux/module.h>
  22#include <linux/kthread.h>
  23#include <linux/slab.h>
  24
  25#include <linux/sunrpc/types.h>
  26#include <linux/sunrpc/xdr.h>
  27#include <linux/sunrpc/stats.h>
  28#include <linux/sunrpc/svcsock.h>
  29#include <linux/sunrpc/clnt.h>
  30#include <linux/sunrpc/bc_xprt.h>
  31
  32#include <trace/events/sunrpc.h>
  33
  34#define RPCDBG_FACILITY RPCDBG_SVCDSP
  35
  36static void svc_unregister(const struct svc_serv *serv, struct net *net);
  37
  38#define svc_serv_is_pooled(serv)    ((serv)->sv_ops->svo_function)
  39
  40#define SVC_POOL_DEFAULT        SVC_POOL_GLOBAL
  41
  42/*
  43 * Structure for mapping cpus to pools and vice versa.
  44 * Setup once during sunrpc initialisation.
  45 */
  46struct svc_pool_map svc_pool_map = {
  47        .mode = SVC_POOL_DEFAULT
  48};
  49EXPORT_SYMBOL_GPL(svc_pool_map);
  50
  51static DEFINE_MUTEX(svc_pool_map_mutex);/* protects svc_pool_map.count only */
  52
  53static int
  54param_set_pool_mode(const char *val, const struct kernel_param *kp)
  55{
  56        int *ip = (int *)kp->arg;
  57        struct svc_pool_map *m = &svc_pool_map;
  58        int err;
  59
  60        mutex_lock(&svc_pool_map_mutex);
  61
  62        err = -EBUSY;
  63        if (m->count)
  64                goto out;
  65
  66        err = 0;
  67        if (!strncmp(val, "auto", 4))
  68                *ip = SVC_POOL_AUTO;
  69        else if (!strncmp(val, "global", 6))
  70                *ip = SVC_POOL_GLOBAL;
  71        else if (!strncmp(val, "percpu", 6))
  72                *ip = SVC_POOL_PERCPU;
  73        else if (!strncmp(val, "pernode", 7))
  74                *ip = SVC_POOL_PERNODE;
  75        else
  76                err = -EINVAL;
  77
  78out:
  79        mutex_unlock(&svc_pool_map_mutex);
  80        return err;
  81}
  82
  83static int
  84param_get_pool_mode(char *buf, const struct kernel_param *kp)
  85{
  86        int *ip = (int *)kp->arg;
  87
  88        switch (*ip)
  89        {
  90        case SVC_POOL_AUTO:
  91                return strlcpy(buf, "auto\n", 20);
  92        case SVC_POOL_GLOBAL:
  93                return strlcpy(buf, "global\n", 20);
  94        case SVC_POOL_PERCPU:
  95                return strlcpy(buf, "percpu\n", 20);
  96        case SVC_POOL_PERNODE:
  97                return strlcpy(buf, "pernode\n", 20);
  98        default:
  99                return sprintf(buf, "%d\n", *ip);
 100        }
 101}
 102
 103module_param_call(pool_mode, param_set_pool_mode, param_get_pool_mode,
 104                 &svc_pool_map.mode, 0644);
 105
 106/*
 107 * Detect best pool mapping mode heuristically,
 108 * according to the machine's topology.
 109 */
 110static int
 111svc_pool_map_choose_mode(void)
 112{
 113        unsigned int node;
 114
 115        if (nr_online_nodes > 1) {
 116                /*
 117                 * Actually have multiple NUMA nodes,
 118                 * so split pools on NUMA node boundaries
 119                 */
 120                return SVC_POOL_PERNODE;
 121        }
 122
 123        node = first_online_node;
 124        if (nr_cpus_node(node) > 2) {
 125                /*
 126                 * Non-trivial SMP, or CONFIG_NUMA on
 127                 * non-NUMA hardware, e.g. with a generic
 128                 * x86_64 kernel on Xeons.  In this case we
 129                 * want to divide the pools on cpu boundaries.
 130                 */
 131                return SVC_POOL_PERCPU;
 132        }
 133
 134        /* default: one global pool */
 135        return SVC_POOL_GLOBAL;
 136}
 137
 138/*
 139 * Allocate the to_pool[] and pool_to[] arrays.
 140 * Returns 0 on success or an errno.
 141 */
 142static int
 143svc_pool_map_alloc_arrays(struct svc_pool_map *m, unsigned int maxpools)
 144{
 145        m->to_pool = kcalloc(maxpools, sizeof(unsigned int), GFP_KERNEL);
 146        if (!m->to_pool)
 147                goto fail;
 148        m->pool_to = kcalloc(maxpools, sizeof(unsigned int), GFP_KERNEL);
 149        if (!m->pool_to)
 150                goto fail_free;
 151
 152        return 0;
 153
 154fail_free:
 155        kfree(m->to_pool);
 156        m->to_pool = NULL;
 157fail:
 158        return -ENOMEM;
 159}
 160
 161/*
 162 * Initialise the pool map for SVC_POOL_PERCPU mode.
 163 * Returns number of pools or <0 on error.
 164 */
 165static int
 166svc_pool_map_init_percpu(struct svc_pool_map *m)
 167{
 168        unsigned int maxpools = nr_cpu_ids;
 169        unsigned int pidx = 0;
 170        unsigned int cpu;
 171        int err;
 172
 173        err = svc_pool_map_alloc_arrays(m, maxpools);
 174        if (err)
 175                return err;
 176
 177        for_each_online_cpu(cpu) {
 178                BUG_ON(pidx >= maxpools);
 179                m->to_pool[cpu] = pidx;
 180                m->pool_to[pidx] = cpu;
 181                pidx++;
 182        }
 183        /* cpus brought online later all get mapped to pool0, sorry */
 184
 185        return pidx;
 186};
 187
 188
 189/*
 190 * Initialise the pool map for SVC_POOL_PERNODE mode.
 191 * Returns number of pools or <0 on error.
 192 */
 193static int
 194svc_pool_map_init_pernode(struct svc_pool_map *m)
 195{
 196        unsigned int maxpools = nr_node_ids;
 197        unsigned int pidx = 0;
 198        unsigned int node;
 199        int err;
 200
 201        err = svc_pool_map_alloc_arrays(m, maxpools);
 202        if (err)
 203                return err;
 204
 205        for_each_node_with_cpus(node) {
 206                /* some architectures (e.g. SN2) have cpuless nodes */
 207                BUG_ON(pidx > maxpools);
 208                m->to_pool[node] = pidx;
 209                m->pool_to[pidx] = node;
 210                pidx++;
 211        }
 212        /* nodes brought online later all get mapped to pool0, sorry */
 213
 214        return pidx;
 215}
 216
 217
 218/*
 219 * Add a reference to the global map of cpus to pools (and
 220 * vice versa).  Initialise the map if we're the first user.
 221 * Returns the number of pools.
 222 */
 223unsigned int
 224svc_pool_map_get(void)
 225{
 226        struct svc_pool_map *m = &svc_pool_map;
 227        int npools = -1;
 228
 229        mutex_lock(&svc_pool_map_mutex);
 230
 231        if (m->count++) {
 232                mutex_unlock(&svc_pool_map_mutex);
 233                return m->npools;
 234        }
 235
 236        if (m->mode == SVC_POOL_AUTO)
 237                m->mode = svc_pool_map_choose_mode();
 238
 239        switch (m->mode) {
 240        case SVC_POOL_PERCPU:
 241                npools = svc_pool_map_init_percpu(m);
 242                break;
 243        case SVC_POOL_PERNODE:
 244                npools = svc_pool_map_init_pernode(m);
 245                break;
 246        }
 247
 248        if (npools < 0) {
 249                /* default, or memory allocation failure */
 250                npools = 1;
 251                m->mode = SVC_POOL_GLOBAL;
 252        }
 253        m->npools = npools;
 254
 255        mutex_unlock(&svc_pool_map_mutex);
 256        return m->npools;
 257}
 258EXPORT_SYMBOL_GPL(svc_pool_map_get);
 259
 260/*
 261 * Drop a reference to the global map of cpus to pools.
 262 * When the last reference is dropped, the map data is
 263 * freed; this allows the sysadmin to change the pool
 264 * mode using the pool_mode module option without
 265 * rebooting or re-loading sunrpc.ko.
 266 */
 267void
 268svc_pool_map_put(void)
 269{
 270        struct svc_pool_map *m = &svc_pool_map;
 271
 272        mutex_lock(&svc_pool_map_mutex);
 273
 274        if (!--m->count) {
 275                kfree(m->to_pool);
 276                m->to_pool = NULL;
 277                kfree(m->pool_to);
 278                m->pool_to = NULL;
 279                m->npools = 0;
 280        }
 281
 282        mutex_unlock(&svc_pool_map_mutex);
 283}
 284EXPORT_SYMBOL_GPL(svc_pool_map_put);
 285
 286static int svc_pool_map_get_node(unsigned int pidx)
 287{
 288        const struct svc_pool_map *m = &svc_pool_map;
 289
 290        if (m->count) {
 291                if (m->mode == SVC_POOL_PERCPU)
 292                        return cpu_to_node(m->pool_to[pidx]);
 293                if (m->mode == SVC_POOL_PERNODE)
 294                        return m->pool_to[pidx];
 295        }
 296        return NUMA_NO_NODE;
 297}
 298/*
 299 * Set the given thread's cpus_allowed mask so that it
 300 * will only run on cpus in the given pool.
 301 */
 302static inline void
 303svc_pool_map_set_cpumask(struct task_struct *task, unsigned int pidx)
 304{
 305        struct svc_pool_map *m = &svc_pool_map;
 306        unsigned int node = m->pool_to[pidx];
 307
 308        /*
 309         * The caller checks for sv_nrpools > 1, which
 310         * implies that we've been initialized.
 311         */
 312        WARN_ON_ONCE(m->count == 0);
 313        if (m->count == 0)
 314                return;
 315
 316        switch (m->mode) {
 317        case SVC_POOL_PERCPU:
 318        {
 319                set_cpus_allowed_ptr(task, cpumask_of(node));
 320                break;
 321        }
 322        case SVC_POOL_PERNODE:
 323        {
 324                set_cpus_allowed_ptr(task, cpumask_of_node(node));
 325                break;
 326        }
 327        }
 328}
 329
 330/*
 331 * Use the mapping mode to choose a pool for a given CPU.
 332 * Used when enqueueing an incoming RPC.  Always returns
 333 * a non-NULL pool pointer.
 334 */
 335struct svc_pool *
 336svc_pool_for_cpu(struct svc_serv *serv, int cpu)
 337{
 338        struct svc_pool_map *m = &svc_pool_map;
 339        unsigned int pidx = 0;
 340
 341        /*
 342         * An uninitialised map happens in a pure client when
 343         * lockd is brought up, so silently treat it the
 344         * same as SVC_POOL_GLOBAL.
 345         */
 346        if (svc_serv_is_pooled(serv)) {
 347                switch (m->mode) {
 348                case SVC_POOL_PERCPU:
 349                        pidx = m->to_pool[cpu];
 350                        break;
 351                case SVC_POOL_PERNODE:
 352                        pidx = m->to_pool[cpu_to_node(cpu)];
 353                        break;
 354                }
 355        }
 356        return &serv->sv_pools[pidx % serv->sv_nrpools];
 357}
 358
 359int svc_rpcb_setup(struct svc_serv *serv, struct net *net)
 360{
 361        int err;
 362
 363        err = rpcb_create_local(net);
 364        if (err)
 365                return err;
 366
 367        /* Remove any stale portmap registrations */
 368        svc_unregister(serv, net);
 369        return 0;
 370}
 371EXPORT_SYMBOL_GPL(svc_rpcb_setup);
 372
 373void svc_rpcb_cleanup(struct svc_serv *serv, struct net *net)
 374{
 375        svc_unregister(serv, net);
 376        rpcb_put_local(net);
 377}
 378EXPORT_SYMBOL_GPL(svc_rpcb_cleanup);
 379
 380static int svc_uses_rpcbind(struct svc_serv *serv)
 381{
 382        struct svc_program      *progp;
 383        unsigned int            i;
 384
 385        for (progp = serv->sv_program; progp; progp = progp->pg_next) {
 386                for (i = 0; i < progp->pg_nvers; i++) {
 387                        if (progp->pg_vers[i] == NULL)
 388                                continue;
 389                        if (!progp->pg_vers[i]->vs_hidden)
 390                                return 1;
 391                }
 392        }
 393
 394        return 0;
 395}
 396
 397int svc_bind(struct svc_serv *serv, struct net *net)
 398{
 399        if (!svc_uses_rpcbind(serv))
 400                return 0;
 401        return svc_rpcb_setup(serv, net);
 402}
 403EXPORT_SYMBOL_GPL(svc_bind);
 404
 405#if defined(CONFIG_SUNRPC_BACKCHANNEL)
 406static void
 407__svc_init_bc(struct svc_serv *serv)
 408{
 409        INIT_LIST_HEAD(&serv->sv_cb_list);
 410        spin_lock_init(&serv->sv_cb_lock);
 411        init_waitqueue_head(&serv->sv_cb_waitq);
 412}
 413#else
 414static void
 415__svc_init_bc(struct svc_serv *serv)
 416{
 417}
 418#endif
 419
 420/*
 421 * Create an RPC service
 422 */
 423static struct svc_serv *
 424__svc_create(struct svc_program *prog, unsigned int bufsize, int npools,
 425             const struct svc_serv_ops *ops)
 426{
 427        struct svc_serv *serv;
 428        unsigned int vers;
 429        unsigned int xdrsize;
 430        unsigned int i;
 431
 432        if (!(serv = kzalloc(sizeof(*serv), GFP_KERNEL)))
 433                return NULL;
 434        serv->sv_name      = prog->pg_name;
 435        serv->sv_program   = prog;
 436        serv->sv_nrthreads = 1;
 437        serv->sv_stats     = prog->pg_stats;
 438        if (bufsize > RPCSVC_MAXPAYLOAD)
 439                bufsize = RPCSVC_MAXPAYLOAD;
 440        serv->sv_max_payload = bufsize? bufsize : 4096;
 441        serv->sv_max_mesg  = roundup(serv->sv_max_payload + PAGE_SIZE, PAGE_SIZE);
 442        serv->sv_ops = ops;
 443        xdrsize = 0;
 444        while (prog) {
 445                prog->pg_lovers = prog->pg_nvers-1;
 446                for (vers=0; vers<prog->pg_nvers ; vers++)
 447                        if (prog->pg_vers[vers]) {
 448                                prog->pg_hivers = vers;
 449                                if (prog->pg_lovers > vers)
 450                                        prog->pg_lovers = vers;
 451                                if (prog->pg_vers[vers]->vs_xdrsize > xdrsize)
 452                                        xdrsize = prog->pg_vers[vers]->vs_xdrsize;
 453                        }
 454                prog = prog->pg_next;
 455        }
 456        serv->sv_xdrsize   = xdrsize;
 457        INIT_LIST_HEAD(&serv->sv_tempsocks);
 458        INIT_LIST_HEAD(&serv->sv_permsocks);
 459        timer_setup(&serv->sv_temptimer, NULL, 0);
 460        spin_lock_init(&serv->sv_lock);
 461
 462        __svc_init_bc(serv);
 463
 464        serv->sv_nrpools = npools;
 465        serv->sv_pools =
 466                kcalloc(serv->sv_nrpools, sizeof(struct svc_pool),
 467                        GFP_KERNEL);
 468        if (!serv->sv_pools) {
 469                kfree(serv);
 470                return NULL;
 471        }
 472
 473        for (i = 0; i < serv->sv_nrpools; i++) {
 474                struct svc_pool *pool = &serv->sv_pools[i];
 475
 476                dprintk("svc: initialising pool %u for %s\n",
 477                                i, serv->sv_name);
 478
 479                pool->sp_id = i;
 480                INIT_LIST_HEAD(&pool->sp_sockets);
 481                INIT_LIST_HEAD(&pool->sp_all_threads);
 482                spin_lock_init(&pool->sp_lock);
 483        }
 484
 485        return serv;
 486}
 487
 488struct svc_serv *
 489svc_create(struct svc_program *prog, unsigned int bufsize,
 490           const struct svc_serv_ops *ops)
 491{
 492        return __svc_create(prog, bufsize, /*npools*/1, ops);
 493}
 494EXPORT_SYMBOL_GPL(svc_create);
 495
 496struct svc_serv *
 497svc_create_pooled(struct svc_program *prog, unsigned int bufsize,
 498                  const struct svc_serv_ops *ops)
 499{
 500        struct svc_serv *serv;
 501        unsigned int npools = svc_pool_map_get();
 502
 503        serv = __svc_create(prog, bufsize, npools, ops);
 504        if (!serv)
 505                goto out_err;
 506        return serv;
 507out_err:
 508        svc_pool_map_put();
 509        return NULL;
 510}
 511EXPORT_SYMBOL_GPL(svc_create_pooled);
 512
 513void svc_shutdown_net(struct svc_serv *serv, struct net *net)
 514{
 515        svc_close_net(serv, net);
 516
 517        if (serv->sv_ops->svo_shutdown)
 518                serv->sv_ops->svo_shutdown(serv, net);
 519}
 520EXPORT_SYMBOL_GPL(svc_shutdown_net);
 521
 522/*
 523 * Destroy an RPC service. Should be called with appropriate locking to
 524 * protect the sv_nrthreads, sv_permsocks and sv_tempsocks.
 525 */
 526void
 527svc_destroy(struct svc_serv *serv)
 528{
 529        dprintk("svc: svc_destroy(%s, %d)\n",
 530                                serv->sv_program->pg_name,
 531                                serv->sv_nrthreads);
 532
 533        if (serv->sv_nrthreads) {
 534                if (--(serv->sv_nrthreads) != 0) {
 535                        svc_sock_update_bufs(serv);
 536                        return;
 537                }
 538        } else
 539                printk("svc_destroy: no threads for serv=%p!\n", serv);
 540
 541        del_timer_sync(&serv->sv_temptimer);
 542
 543        /*
 544         * The last user is gone and thus all sockets have to be destroyed to
 545         * the point. Check this.
 546         */
 547        BUG_ON(!list_empty(&serv->sv_permsocks));
 548        BUG_ON(!list_empty(&serv->sv_tempsocks));
 549
 550        cache_clean_deferred(serv);
 551
 552        if (svc_serv_is_pooled(serv))
 553                svc_pool_map_put();
 554
 555        kfree(serv->sv_pools);
 556        kfree(serv);
 557}
 558EXPORT_SYMBOL_GPL(svc_destroy);
 559
 560/*
 561 * Allocate an RPC server's buffer space.
 562 * We allocate pages and place them in rq_pages.
 563 */
 564static int
 565svc_init_buffer(struct svc_rqst *rqstp, unsigned int size, int node)
 566{
 567        unsigned int pages, arghi;
 568
 569        /* bc_xprt uses fore channel allocated buffers */
 570        if (svc_is_backchannel(rqstp))
 571                return 1;
 572
 573        pages = size / PAGE_SIZE + 1; /* extra page as we hold both request and reply.
 574                                       * We assume one is at most one page
 575                                       */
 576        arghi = 0;
 577        WARN_ON_ONCE(pages > RPCSVC_MAXPAGES);
 578        if (pages > RPCSVC_MAXPAGES)
 579                pages = RPCSVC_MAXPAGES;
 580        while (pages) {
 581                struct page *p = alloc_pages_node(node, GFP_KERNEL, 0);
 582                if (!p)
 583                        break;
 584                rqstp->rq_pages[arghi++] = p;
 585                pages--;
 586        }
 587        return pages == 0;
 588}
 589
 590/*
 591 * Release an RPC server buffer
 592 */
 593static void
 594svc_release_buffer(struct svc_rqst *rqstp)
 595{
 596        unsigned int i;
 597
 598        for (i = 0; i < ARRAY_SIZE(rqstp->rq_pages); i++)
 599                if (rqstp->rq_pages[i])
 600                        put_page(rqstp->rq_pages[i]);
 601}
 602
 603struct svc_rqst *
 604svc_rqst_alloc(struct svc_serv *serv, struct svc_pool *pool, int node)
 605{
 606        struct svc_rqst *rqstp;
 607
 608        rqstp = kzalloc_node(sizeof(*rqstp), GFP_KERNEL, node);
 609        if (!rqstp)
 610                return rqstp;
 611
 612        __set_bit(RQ_BUSY, &rqstp->rq_flags);
 613        spin_lock_init(&rqstp->rq_lock);
 614        rqstp->rq_server = serv;
 615        rqstp->rq_pool = pool;
 616
 617        rqstp->rq_scratch_page = alloc_pages_node(node, GFP_KERNEL, 0);
 618        if (!rqstp->rq_scratch_page)
 619                goto out_enomem;
 620
 621        rqstp->rq_argp = kmalloc_node(serv->sv_xdrsize, GFP_KERNEL, node);
 622        if (!rqstp->rq_argp)
 623                goto out_enomem;
 624
 625        rqstp->rq_resp = kmalloc_node(serv->sv_xdrsize, GFP_KERNEL, node);
 626        if (!rqstp->rq_resp)
 627                goto out_enomem;
 628
 629        if (!svc_init_buffer(rqstp, serv->sv_max_mesg, node))
 630                goto out_enomem;
 631
 632        return rqstp;
 633out_enomem:
 634        svc_rqst_free(rqstp);
 635        return NULL;
 636}
 637EXPORT_SYMBOL_GPL(svc_rqst_alloc);
 638
 639struct svc_rqst *
 640svc_prepare_thread(struct svc_serv *serv, struct svc_pool *pool, int node)
 641{
 642        struct svc_rqst *rqstp;
 643
 644        rqstp = svc_rqst_alloc(serv, pool, node);
 645        if (!rqstp)
 646                return ERR_PTR(-ENOMEM);
 647
 648        serv->sv_nrthreads++;
 649        spin_lock_bh(&pool->sp_lock);
 650        pool->sp_nrthreads++;
 651        list_add_rcu(&rqstp->rq_all, &pool->sp_all_threads);
 652        spin_unlock_bh(&pool->sp_lock);
 653        return rqstp;
 654}
 655EXPORT_SYMBOL_GPL(svc_prepare_thread);
 656
 657/*
 658 * Choose a pool in which to create a new thread, for svc_set_num_threads
 659 */
 660static inline struct svc_pool *
 661choose_pool(struct svc_serv *serv, struct svc_pool *pool, unsigned int *state)
 662{
 663        if (pool != NULL)
 664                return pool;
 665
 666        return &serv->sv_pools[(*state)++ % serv->sv_nrpools];
 667}
 668
 669/*
 670 * Choose a thread to kill, for svc_set_num_threads
 671 */
 672static inline struct task_struct *
 673choose_victim(struct svc_serv *serv, struct svc_pool *pool, unsigned int *state)
 674{
 675        unsigned int i;
 676        struct task_struct *task = NULL;
 677
 678        if (pool != NULL) {
 679                spin_lock_bh(&pool->sp_lock);
 680        } else {
 681                /* choose a pool in round-robin fashion */
 682                for (i = 0; i < serv->sv_nrpools; i++) {
 683                        pool = &serv->sv_pools[--(*state) % serv->sv_nrpools];
 684                        spin_lock_bh(&pool->sp_lock);
 685                        if (!list_empty(&pool->sp_all_threads))
 686                                goto found_pool;
 687                        spin_unlock_bh(&pool->sp_lock);
 688                }
 689                return NULL;
 690        }
 691
 692found_pool:
 693        if (!list_empty(&pool->sp_all_threads)) {
 694                struct svc_rqst *rqstp;
 695
 696                /*
 697                 * Remove from the pool->sp_all_threads list
 698                 * so we don't try to kill it again.
 699                 */
 700                rqstp = list_entry(pool->sp_all_threads.next, struct svc_rqst, rq_all);
 701                set_bit(RQ_VICTIM, &rqstp->rq_flags);
 702                list_del_rcu(&rqstp->rq_all);
 703                task = rqstp->rq_task;
 704        }
 705        spin_unlock_bh(&pool->sp_lock);
 706
 707        return task;
 708}
 709
 710/* create new threads */
 711static int
 712svc_start_kthreads(struct svc_serv *serv, struct svc_pool *pool, int nrservs)
 713{
 714        struct svc_rqst *rqstp;
 715        struct task_struct *task;
 716        struct svc_pool *chosen_pool;
 717        unsigned int state = serv->sv_nrthreads-1;
 718        int node;
 719
 720        do {
 721                nrservs--;
 722                chosen_pool = choose_pool(serv, pool, &state);
 723
 724                node = svc_pool_map_get_node(chosen_pool->sp_id);
 725                rqstp = svc_prepare_thread(serv, chosen_pool, node);
 726                if (IS_ERR(rqstp))
 727                        return PTR_ERR(rqstp);
 728
 729                __module_get(serv->sv_ops->svo_module);
 730                task = kthread_create_on_node(serv->sv_ops->svo_function, rqstp,
 731                                              node, "%s", serv->sv_name);
 732                if (IS_ERR(task)) {
 733                        module_put(serv->sv_ops->svo_module);
 734                        svc_exit_thread(rqstp);
 735                        return PTR_ERR(task);
 736                }
 737
 738                rqstp->rq_task = task;
 739                if (serv->sv_nrpools > 1)
 740                        svc_pool_map_set_cpumask(task, chosen_pool->sp_id);
 741
 742                svc_sock_update_bufs(serv);
 743                wake_up_process(task);
 744        } while (nrservs > 0);
 745
 746        return 0;
 747}
 748
 749
 750/* destroy old threads */
 751static int
 752svc_signal_kthreads(struct svc_serv *serv, struct svc_pool *pool, int nrservs)
 753{
 754        struct task_struct *task;
 755        unsigned int state = serv->sv_nrthreads-1;
 756
 757        /* destroy old threads */
 758        do {
 759                task = choose_victim(serv, pool, &state);
 760                if (task == NULL)
 761                        break;
 762                send_sig(SIGINT, task, 1);
 763                nrservs++;
 764        } while (nrservs < 0);
 765
 766        return 0;
 767}
 768
 769/*
 770 * Create or destroy enough new threads to make the number
 771 * of threads the given number.  If `pool' is non-NULL, applies
 772 * only to threads in that pool, otherwise round-robins between
 773 * all pools.  Caller must ensure that mutual exclusion between this and
 774 * server startup or shutdown.
 775 *
 776 * Destroying threads relies on the service threads filling in
 777 * rqstp->rq_task, which only the nfs ones do.  Assumes the serv
 778 * has been created using svc_create_pooled().
 779 *
 780 * Based on code that used to be in nfsd_svc() but tweaked
 781 * to be pool-aware.
 782 */
 783int
 784svc_set_num_threads(struct svc_serv *serv, struct svc_pool *pool, int nrservs)
 785{
 786        if (pool == NULL) {
 787                /* The -1 assumes caller has done a svc_get() */
 788                nrservs -= (serv->sv_nrthreads-1);
 789        } else {
 790                spin_lock_bh(&pool->sp_lock);
 791                nrservs -= pool->sp_nrthreads;
 792                spin_unlock_bh(&pool->sp_lock);
 793        }
 794
 795        if (nrservs > 0)
 796                return svc_start_kthreads(serv, pool, nrservs);
 797        if (nrservs < 0)
 798                return svc_signal_kthreads(serv, pool, nrservs);
 799        return 0;
 800}
 801EXPORT_SYMBOL_GPL(svc_set_num_threads);
 802
 803/* destroy old threads */
 804static int
 805svc_stop_kthreads(struct svc_serv *serv, struct svc_pool *pool, int nrservs)
 806{
 807        struct task_struct *task;
 808        unsigned int state = serv->sv_nrthreads-1;
 809
 810        /* destroy old threads */
 811        do {
 812                task = choose_victim(serv, pool, &state);
 813                if (task == NULL)
 814                        break;
 815                kthread_stop(task);
 816                nrservs++;
 817        } while (nrservs < 0);
 818        return 0;
 819}
 820
 821int
 822svc_set_num_threads_sync(struct svc_serv *serv, struct svc_pool *pool, int nrservs)
 823{
 824        if (pool == NULL) {
 825                /* The -1 assumes caller has done a svc_get() */
 826                nrservs -= (serv->sv_nrthreads-1);
 827        } else {
 828                spin_lock_bh(&pool->sp_lock);
 829                nrservs -= pool->sp_nrthreads;
 830                spin_unlock_bh(&pool->sp_lock);
 831        }
 832
 833        if (nrservs > 0)
 834                return svc_start_kthreads(serv, pool, nrservs);
 835        if (nrservs < 0)
 836                return svc_stop_kthreads(serv, pool, nrservs);
 837        return 0;
 838}
 839EXPORT_SYMBOL_GPL(svc_set_num_threads_sync);
 840
 841/*
 842 * Called from a server thread as it's exiting. Caller must hold the "service
 843 * mutex" for the service.
 844 */
 845void
 846svc_rqst_free(struct svc_rqst *rqstp)
 847{
 848        svc_release_buffer(rqstp);
 849        if (rqstp->rq_scratch_page)
 850                put_page(rqstp->rq_scratch_page);
 851        kfree(rqstp->rq_resp);
 852        kfree(rqstp->rq_argp);
 853        kfree(rqstp->rq_auth_data);
 854        kfree_rcu(rqstp, rq_rcu_head);
 855}
 856EXPORT_SYMBOL_GPL(svc_rqst_free);
 857
 858void
 859svc_exit_thread(struct svc_rqst *rqstp)
 860{
 861        struct svc_serv *serv = rqstp->rq_server;
 862        struct svc_pool *pool = rqstp->rq_pool;
 863
 864        spin_lock_bh(&pool->sp_lock);
 865        pool->sp_nrthreads--;
 866        if (!test_and_set_bit(RQ_VICTIM, &rqstp->rq_flags))
 867                list_del_rcu(&rqstp->rq_all);
 868        spin_unlock_bh(&pool->sp_lock);
 869
 870        svc_rqst_free(rqstp);
 871
 872        /* Release the server */
 873        if (serv)
 874                svc_destroy(serv);
 875}
 876EXPORT_SYMBOL_GPL(svc_exit_thread);
 877
 878/*
 879 * Register an "inet" protocol family netid with the local
 880 * rpcbind daemon via an rpcbind v4 SET request.
 881 *
 882 * No netconfig infrastructure is available in the kernel, so
 883 * we map IP_ protocol numbers to netids by hand.
 884 *
 885 * Returns zero on success; a negative errno value is returned
 886 * if any error occurs.
 887 */
 888static int __svc_rpcb_register4(struct net *net, const u32 program,
 889                                const u32 version,
 890                                const unsigned short protocol,
 891                                const unsigned short port)
 892{
 893        const struct sockaddr_in sin = {
 894                .sin_family             = AF_INET,
 895                .sin_addr.s_addr        = htonl(INADDR_ANY),
 896                .sin_port               = htons(port),
 897        };
 898        const char *netid;
 899        int error;
 900
 901        switch (protocol) {
 902        case IPPROTO_UDP:
 903                netid = RPCBIND_NETID_UDP;
 904                break;
 905        case IPPROTO_TCP:
 906                netid = RPCBIND_NETID_TCP;
 907                break;
 908        default:
 909                return -ENOPROTOOPT;
 910        }
 911
 912        error = rpcb_v4_register(net, program, version,
 913                                        (const struct sockaddr *)&sin, netid);
 914
 915        /*
 916         * User space didn't support rpcbind v4, so retry this
 917         * registration request with the legacy rpcbind v2 protocol.
 918         */
 919        if (error == -EPROTONOSUPPORT)
 920                error = rpcb_register(net, program, version, protocol, port);
 921
 922        return error;
 923}
 924
 925#if IS_ENABLED(CONFIG_IPV6)
 926/*
 927 * Register an "inet6" protocol family netid with the local
 928 * rpcbind daemon via an rpcbind v4 SET request.
 929 *
 930 * No netconfig infrastructure is available in the kernel, so
 931 * we map IP_ protocol numbers to netids by hand.
 932 *
 933 * Returns zero on success; a negative errno value is returned
 934 * if any error occurs.
 935 */
 936static int __svc_rpcb_register6(struct net *net, const u32 program,
 937                                const u32 version,
 938                                const unsigned short protocol,
 939                                const unsigned short port)
 940{
 941        const struct sockaddr_in6 sin6 = {
 942                .sin6_family            = AF_INET6,
 943                .sin6_addr              = IN6ADDR_ANY_INIT,
 944                .sin6_port              = htons(port),
 945        };
 946        const char *netid;
 947        int error;
 948
 949        switch (protocol) {
 950        case IPPROTO_UDP:
 951                netid = RPCBIND_NETID_UDP6;
 952                break;
 953        case IPPROTO_TCP:
 954                netid = RPCBIND_NETID_TCP6;
 955                break;
 956        default:
 957                return -ENOPROTOOPT;
 958        }
 959
 960        error = rpcb_v4_register(net, program, version,
 961                                        (const struct sockaddr *)&sin6, netid);
 962
 963        /*
 964         * User space didn't support rpcbind version 4, so we won't
 965         * use a PF_INET6 listener.
 966         */
 967        if (error == -EPROTONOSUPPORT)
 968                error = -EAFNOSUPPORT;
 969
 970        return error;
 971}
 972#endif  /* IS_ENABLED(CONFIG_IPV6) */
 973
 974/*
 975 * Register a kernel RPC service via rpcbind version 4.
 976 *
 977 * Returns zero on success; a negative errno value is returned
 978 * if any error occurs.
 979 */
 980static int __svc_register(struct net *net, const char *progname,
 981                          const u32 program, const u32 version,
 982                          const int family,
 983                          const unsigned short protocol,
 984                          const unsigned short port)
 985{
 986        int error = -EAFNOSUPPORT;
 987
 988        switch (family) {
 989        case PF_INET:
 990                error = __svc_rpcb_register4(net, program, version,
 991                                                protocol, port);
 992                break;
 993#if IS_ENABLED(CONFIG_IPV6)
 994        case PF_INET6:
 995                error = __svc_rpcb_register6(net, program, version,
 996                                                protocol, port);
 997#endif
 998        }
 999
1000        trace_svc_register(progname, version, protocol, port, family, error);
1001        return error;
1002}
1003
1004int svc_rpcbind_set_version(struct net *net,
1005                            const struct svc_program *progp,
1006                            u32 version, int family,
1007                            unsigned short proto,
1008                            unsigned short port)
1009{
1010        return __svc_register(net, progp->pg_name, progp->pg_prog,
1011                                version, family, proto, port);
1012
1013}
1014EXPORT_SYMBOL_GPL(svc_rpcbind_set_version);
1015
1016int svc_generic_rpcbind_set(struct net *net,
1017                            const struct svc_program *progp,
1018                            u32 version, int family,
1019                            unsigned short proto,
1020                            unsigned short port)
1021{
1022        const struct svc_version *vers = progp->pg_vers[version];
1023        int error;
1024
1025        if (vers == NULL)
1026                return 0;
1027
1028        if (vers->vs_hidden) {
1029                trace_svc_noregister(progp->pg_name, version, proto,
1030                                     port, family, 0);
1031                return 0;
1032        }
1033
1034        /*
1035         * Don't register a UDP port if we need congestion
1036         * control.
1037         */
1038        if (vers->vs_need_cong_ctrl && proto == IPPROTO_UDP)
1039                return 0;
1040
1041        error = svc_rpcbind_set_version(net, progp, version,
1042                                        family, proto, port);
1043
1044        return (vers->vs_rpcb_optnl) ? 0 : error;
1045}
1046EXPORT_SYMBOL_GPL(svc_generic_rpcbind_set);
1047
1048/**
1049 * svc_register - register an RPC service with the local portmapper
1050 * @serv: svc_serv struct for the service to register
1051 * @net: net namespace for the service to register
1052 * @family: protocol family of service's listener socket
1053 * @proto: transport protocol number to advertise
1054 * @port: port to advertise
1055 *
1056 * Service is registered for any address in the passed-in protocol family
1057 */
1058int svc_register(const struct svc_serv *serv, struct net *net,
1059                 const int family, const unsigned short proto,
1060                 const unsigned short port)
1061{
1062        struct svc_program      *progp;
1063        unsigned int            i;
1064        int                     error = 0;
1065
1066        WARN_ON_ONCE(proto == 0 && port == 0);
1067        if (proto == 0 && port == 0)
1068                return -EINVAL;
1069
1070        for (progp = serv->sv_program; progp; progp = progp->pg_next) {
1071                for (i = 0; i < progp->pg_nvers; i++) {
1072
1073                        error = progp->pg_rpcbind_set(net, progp, i,
1074                                        family, proto, port);
1075                        if (error < 0) {
1076                                printk(KERN_WARNING "svc: failed to register "
1077                                        "%sv%u RPC service (errno %d).\n",
1078                                        progp->pg_name, i, -error);
1079                                break;
1080                        }
1081                }
1082        }
1083
1084        return error;
1085}
1086
1087/*
1088 * If user space is running rpcbind, it should take the v4 UNSET
1089 * and clear everything for this [program, version].  If user space
1090 * is running portmap, it will reject the v4 UNSET, but won't have
1091 * any "inet6" entries anyway.  So a PMAP_UNSET should be sufficient
1092 * in this case to clear all existing entries for [program, version].
1093 */
1094static void __svc_unregister(struct net *net, const u32 program, const u32 version,
1095                             const char *progname)
1096{
1097        int error;
1098
1099        error = rpcb_v4_register(net, program, version, NULL, "");
1100
1101        /*
1102         * User space didn't support rpcbind v4, so retry this
1103         * request with the legacy rpcbind v2 protocol.
1104         */
1105        if (error == -EPROTONOSUPPORT)
1106                error = rpcb_register(net, program, version, 0, 0);
1107
1108        trace_svc_unregister(progname, version, error);
1109}
1110
1111/*
1112 * All netids, bind addresses and ports registered for [program, version]
1113 * are removed from the local rpcbind database (if the service is not
1114 * hidden) to make way for a new instance of the service.
1115 *
1116 * The result of unregistration is reported via dprintk for those who want
1117 * verification of the result, but is otherwise not important.
1118 */
1119static void svc_unregister(const struct svc_serv *serv, struct net *net)
1120{
1121        struct svc_program *progp;
1122        unsigned long flags;
1123        unsigned int i;
1124
1125        clear_thread_flag(TIF_SIGPENDING);
1126
1127        for (progp = serv->sv_program; progp; progp = progp->pg_next) {
1128                for (i = 0; i < progp->pg_nvers; i++) {
1129                        if (progp->pg_vers[i] == NULL)
1130                                continue;
1131                        if (progp->pg_vers[i]->vs_hidden)
1132                                continue;
1133                        __svc_unregister(net, progp->pg_prog, i, progp->pg_name);
1134                }
1135        }
1136
1137        spin_lock_irqsave(&current->sighand->siglock, flags);
1138        recalc_sigpending();
1139        spin_unlock_irqrestore(&current->sighand->siglock, flags);
1140}
1141
1142/*
1143 * dprintk the given error with the address of the client that caused it.
1144 */
1145#if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
1146static __printf(2, 3)
1147void svc_printk(struct svc_rqst *rqstp, const char *fmt, ...)
1148{
1149        struct va_format vaf;
1150        va_list args;
1151        char    buf[RPC_MAX_ADDRBUFLEN];
1152
1153        va_start(args, fmt);
1154
1155        vaf.fmt = fmt;
1156        vaf.va = &args;
1157
1158        dprintk("svc: %s: %pV", svc_print_addr(rqstp, buf, sizeof(buf)), &vaf);
1159
1160        va_end(args);
1161}
1162#else
1163static __printf(2,3) void svc_printk(struct svc_rqst *rqstp, const char *fmt, ...) {}
1164#endif
1165
1166__be32
1167svc_return_autherr(struct svc_rqst *rqstp, __be32 auth_err)
1168{
1169        set_bit(RQ_AUTHERR, &rqstp->rq_flags);
1170        return auth_err;
1171}
1172EXPORT_SYMBOL_GPL(svc_return_autherr);
1173
1174static __be32
1175svc_get_autherr(struct svc_rqst *rqstp, __be32 *statp)
1176{
1177        if (test_and_clear_bit(RQ_AUTHERR, &rqstp->rq_flags))
1178                return *statp;
1179        return rpc_auth_ok;
1180}
1181
1182static int
1183svc_generic_dispatch(struct svc_rqst *rqstp, __be32 *statp)
1184{
1185        struct kvec *argv = &rqstp->rq_arg.head[0];
1186        struct kvec *resv = &rqstp->rq_res.head[0];
1187        const struct svc_procedure *procp = rqstp->rq_procinfo;
1188
1189        /*
1190         * Decode arguments
1191         * XXX: why do we ignore the return value?
1192         */
1193        if (procp->pc_decode &&
1194            !procp->pc_decode(rqstp, argv->iov_base)) {
1195                *statp = rpc_garbage_args;
1196                return 1;
1197        }
1198
1199        *statp = procp->pc_func(rqstp);
1200
1201        if (*statp == rpc_drop_reply ||
1202            test_bit(RQ_DROPME, &rqstp->rq_flags))
1203                return 0;
1204
1205        if (test_bit(RQ_AUTHERR, &rqstp->rq_flags))
1206                return 1;
1207
1208        if (*statp != rpc_success)
1209                return 1;
1210
1211        /* Encode reply */
1212        if (procp->pc_encode &&
1213            !procp->pc_encode(rqstp, resv->iov_base + resv->iov_len)) {
1214                dprintk("svc: failed to encode reply\n");
1215                /* serv->sv_stats->rpcsystemerr++; */
1216                *statp = rpc_system_err;
1217        }
1218        return 1;
1219}
1220
1221__be32
1222svc_generic_init_request(struct svc_rqst *rqstp,
1223                const struct svc_program *progp,
1224                struct svc_process_info *ret)
1225{
1226        const struct svc_version *versp = NULL; /* compiler food */
1227        const struct svc_procedure *procp = NULL;
1228
1229        if (rqstp->rq_vers >= progp->pg_nvers )
1230                goto err_bad_vers;
1231        versp = progp->pg_vers[rqstp->rq_vers];
1232        if (!versp)
1233                goto err_bad_vers;
1234
1235        /*
1236         * Some protocol versions (namely NFSv4) require some form of
1237         * congestion control.  (See RFC 7530 section 3.1 paragraph 2)
1238         * In other words, UDP is not allowed. We mark those when setting
1239         * up the svc_xprt, and verify that here.
1240         *
1241         * The spec is not very clear about what error should be returned
1242         * when someone tries to access a server that is listening on UDP
1243         * for lower versions. RPC_PROG_MISMATCH seems to be the closest
1244         * fit.
1245         */
1246        if (versp->vs_need_cong_ctrl && rqstp->rq_xprt &&
1247            !test_bit(XPT_CONG_CTRL, &rqstp->rq_xprt->xpt_flags))
1248                goto err_bad_vers;
1249
1250        if (rqstp->rq_proc >= versp->vs_nproc)
1251                goto err_bad_proc;
1252        rqstp->rq_procinfo = procp = &versp->vs_proc[rqstp->rq_proc];
1253        if (!procp)
1254                goto err_bad_proc;
1255
1256        /* Initialize storage for argp and resp */
1257        memset(rqstp->rq_argp, 0, procp->pc_argsize);
1258        memset(rqstp->rq_resp, 0, procp->pc_ressize);
1259
1260        /* Bump per-procedure stats counter */
1261        versp->vs_count[rqstp->rq_proc]++;
1262
1263        ret->dispatch = versp->vs_dispatch;
1264        return rpc_success;
1265err_bad_vers:
1266        ret->mismatch.lovers = progp->pg_lovers;
1267        ret->mismatch.hivers = progp->pg_hivers;
1268        return rpc_prog_mismatch;
1269err_bad_proc:
1270        return rpc_proc_unavail;
1271}
1272EXPORT_SYMBOL_GPL(svc_generic_init_request);
1273
1274/*
1275 * Common routine for processing the RPC request.
1276 */
1277static int
1278svc_process_common(struct svc_rqst *rqstp, struct kvec *argv, struct kvec *resv)
1279{
1280        struct svc_program      *progp;
1281        const struct svc_procedure *procp = NULL;
1282        struct svc_serv         *serv = rqstp->rq_server;
1283        struct svc_process_info process;
1284        __be32                  *statp;
1285        u32                     prog, vers;
1286        __be32                  auth_stat, rpc_stat;
1287        int                     auth_res;
1288        __be32                  *reply_statp;
1289
1290        rpc_stat = rpc_success;
1291
1292        if (argv->iov_len < 6*4)
1293                goto err_short_len;
1294
1295        /* Will be turned off by GSS integrity and privacy services */
1296        set_bit(RQ_SPLICE_OK, &rqstp->rq_flags);
1297        /* Will be turned off only when NFSv4 Sessions are used */
1298        set_bit(RQ_USEDEFERRAL, &rqstp->rq_flags);
1299        clear_bit(RQ_DROPME, &rqstp->rq_flags);
1300
1301        svc_putu32(resv, rqstp->rq_xid);
1302
1303        vers = svc_getnl(argv);
1304
1305        /* First words of reply: */
1306        svc_putnl(resv, 1);             /* REPLY */
1307
1308        if (vers != 2)          /* RPC version number */
1309                goto err_bad_rpc;
1310
1311        /* Save position in case we later decide to reject: */
1312        reply_statp = resv->iov_base + resv->iov_len;
1313
1314        svc_putnl(resv, 0);             /* ACCEPT */
1315
1316        rqstp->rq_prog = prog = svc_getnl(argv);        /* program number */
1317        rqstp->rq_vers = svc_getnl(argv);       /* version number */
1318        rqstp->rq_proc = svc_getnl(argv);       /* procedure number */
1319
1320        for (progp = serv->sv_program; progp; progp = progp->pg_next)
1321                if (prog == progp->pg_prog)
1322                        break;
1323
1324        /*
1325         * Decode auth data, and add verifier to reply buffer.
1326         * We do this before anything else in order to get a decent
1327         * auth verifier.
1328         */
1329        auth_res = svc_authenticate(rqstp, &auth_stat);
1330        /* Also give the program a chance to reject this call: */
1331        if (auth_res == SVC_OK && progp) {
1332                auth_stat = rpc_autherr_badcred;
1333                auth_res = progp->pg_authenticate(rqstp);
1334        }
1335        if (auth_res != SVC_OK)
1336                trace_svc_authenticate(rqstp, auth_res, auth_stat);
1337        switch (auth_res) {
1338        case SVC_OK:
1339                break;
1340        case SVC_GARBAGE:
1341                goto err_garbage;
1342        case SVC_SYSERR:
1343                rpc_stat = rpc_system_err;
1344                goto err_bad;
1345        case SVC_DENIED:
1346                goto err_bad_auth;
1347        case SVC_CLOSE:
1348                goto close;
1349        case SVC_DROP:
1350                goto dropit;
1351        case SVC_COMPLETE:
1352                goto sendit;
1353        }
1354
1355        if (progp == NULL)
1356                goto err_bad_prog;
1357
1358        rpc_stat = progp->pg_init_request(rqstp, progp, &process);
1359        switch (rpc_stat) {
1360        case rpc_success:
1361                break;
1362        case rpc_prog_unavail:
1363                goto err_bad_prog;
1364        case rpc_prog_mismatch:
1365                goto err_bad_vers;
1366        case rpc_proc_unavail:
1367                goto err_bad_proc;
1368        }
1369
1370        procp = rqstp->rq_procinfo;
1371        /* Should this check go into the dispatcher? */
1372        if (!procp || !procp->pc_func)
1373                goto err_bad_proc;
1374
1375        /* Syntactic check complete */
1376        serv->sv_stats->rpccnt++;
1377        trace_svc_process(rqstp, progp->pg_name);
1378
1379        /* Build the reply header. */
1380        statp = resv->iov_base +resv->iov_len;
1381        svc_putnl(resv, RPC_SUCCESS);
1382
1383        /* un-reserve some of the out-queue now that we have a
1384         * better idea of reply size
1385         */
1386        if (procp->pc_xdrressize)
1387                svc_reserve_auth(rqstp, procp->pc_xdrressize<<2);
1388
1389        /* Call the function that processes the request. */
1390        if (!process.dispatch) {
1391                if (!svc_generic_dispatch(rqstp, statp))
1392                        goto release_dropit;
1393                if (*statp == rpc_garbage_args)
1394                        goto err_garbage;
1395                auth_stat = svc_get_autherr(rqstp, statp);
1396                if (auth_stat != rpc_auth_ok)
1397                        goto err_release_bad_auth;
1398        } else {
1399                dprintk("svc: calling dispatcher\n");
1400                if (!process.dispatch(rqstp, statp))
1401                        goto release_dropit; /* Release reply info */
1402        }
1403
1404        /* Check RPC status result */
1405        if (*statp != rpc_success)
1406                resv->iov_len = ((void*)statp)  - resv->iov_base + 4;
1407
1408        /* Release reply info */
1409        if (procp->pc_release)
1410                procp->pc_release(rqstp);
1411
1412        if (procp->pc_encode == NULL)
1413                goto dropit;
1414
1415 sendit:
1416        if (svc_authorise(rqstp))
1417                goto close_xprt;
1418        return 1;               /* Caller can now send it */
1419
1420release_dropit:
1421        if (procp->pc_release)
1422                procp->pc_release(rqstp);
1423 dropit:
1424        svc_authorise(rqstp);   /* doesn't hurt to call this twice */
1425        dprintk("svc: svc_process dropit\n");
1426        return 0;
1427
1428 close:
1429        svc_authorise(rqstp);
1430close_xprt:
1431        if (rqstp->rq_xprt && test_bit(XPT_TEMP, &rqstp->rq_xprt->xpt_flags))
1432                svc_close_xprt(rqstp->rq_xprt);
1433        dprintk("svc: svc_process close\n");
1434        return 0;
1435
1436err_short_len:
1437        svc_printk(rqstp, "short len %zd, dropping request\n",
1438                        argv->iov_len);
1439        goto close_xprt;
1440
1441err_bad_rpc:
1442        serv->sv_stats->rpcbadfmt++;
1443        svc_putnl(resv, 1);     /* REJECT */
1444        svc_putnl(resv, 0);     /* RPC_MISMATCH */
1445        svc_putnl(resv, 2);     /* Only RPCv2 supported */
1446        svc_putnl(resv, 2);
1447        goto sendit;
1448
1449err_release_bad_auth:
1450        if (procp->pc_release)
1451                procp->pc_release(rqstp);
1452err_bad_auth:
1453        dprintk("svc: authentication failed (%d)\n", ntohl(auth_stat));
1454        serv->sv_stats->rpcbadauth++;
1455        /* Restore write pointer to location of accept status: */
1456        xdr_ressize_check(rqstp, reply_statp);
1457        svc_putnl(resv, 1);     /* REJECT */
1458        svc_putnl(resv, 1);     /* AUTH_ERROR */
1459        svc_putnl(resv, ntohl(auth_stat));      /* status */
1460        goto sendit;
1461
1462err_bad_prog:
1463        dprintk("svc: unknown program %d\n", prog);
1464        serv->sv_stats->rpcbadfmt++;
1465        svc_putnl(resv, RPC_PROG_UNAVAIL);
1466        goto sendit;
1467
1468err_bad_vers:
1469        svc_printk(rqstp, "unknown version (%d for prog %d, %s)\n",
1470                       rqstp->rq_vers, rqstp->rq_prog, progp->pg_name);
1471
1472        serv->sv_stats->rpcbadfmt++;
1473        svc_putnl(resv, RPC_PROG_MISMATCH);
1474        svc_putnl(resv, process.mismatch.lovers);
1475        svc_putnl(resv, process.mismatch.hivers);
1476        goto sendit;
1477
1478err_bad_proc:
1479        svc_printk(rqstp, "unknown procedure (%d)\n", rqstp->rq_proc);
1480
1481        serv->sv_stats->rpcbadfmt++;
1482        svc_putnl(resv, RPC_PROC_UNAVAIL);
1483        goto sendit;
1484
1485err_garbage:
1486        svc_printk(rqstp, "failed to decode args\n");
1487
1488        rpc_stat = rpc_garbage_args;
1489err_bad:
1490        serv->sv_stats->rpcbadfmt++;
1491        svc_putnl(resv, ntohl(rpc_stat));
1492        goto sendit;
1493}
1494
1495/*
1496 * Process the RPC request.
1497 */
1498int
1499svc_process(struct svc_rqst *rqstp)
1500{
1501        struct kvec             *argv = &rqstp->rq_arg.head[0];
1502        struct kvec             *resv = &rqstp->rq_res.head[0];
1503        struct svc_serv         *serv = rqstp->rq_server;
1504        u32                     dir;
1505
1506        /*
1507         * Setup response xdr_buf.
1508         * Initially it has just one page
1509         */
1510        rqstp->rq_next_page = &rqstp->rq_respages[1];
1511        resv->iov_base = page_address(rqstp->rq_respages[0]);
1512        resv->iov_len = 0;
1513        rqstp->rq_res.pages = rqstp->rq_respages + 1;
1514        rqstp->rq_res.len = 0;
1515        rqstp->rq_res.page_base = 0;
1516        rqstp->rq_res.page_len = 0;
1517        rqstp->rq_res.buflen = PAGE_SIZE;
1518        rqstp->rq_res.tail[0].iov_base = NULL;
1519        rqstp->rq_res.tail[0].iov_len = 0;
1520
1521        dir  = svc_getnl(argv);
1522        if (dir != 0) {
1523                /* direction != CALL */
1524                svc_printk(rqstp, "bad direction %d, dropping request\n", dir);
1525                serv->sv_stats->rpcbadfmt++;
1526                goto out_drop;
1527        }
1528
1529        /* Returns 1 for send, 0 for drop */
1530        if (likely(svc_process_common(rqstp, argv, resv)))
1531                return svc_send(rqstp);
1532
1533out_drop:
1534        svc_drop(rqstp);
1535        return 0;
1536}
1537EXPORT_SYMBOL_GPL(svc_process);
1538
1539#if defined(CONFIG_SUNRPC_BACKCHANNEL)
1540/*
1541 * Process a backchannel RPC request that arrived over an existing
1542 * outbound connection
1543 */
1544int
1545bc_svc_process(struct svc_serv *serv, struct rpc_rqst *req,
1546               struct svc_rqst *rqstp)
1547{
1548        struct kvec     *argv = &rqstp->rq_arg.head[0];
1549        struct kvec     *resv = &rqstp->rq_res.head[0];
1550        struct rpc_task *task;
1551        int proc_error;
1552        int error;
1553
1554        dprintk("svc: %s(%p)\n", __func__, req);
1555
1556        /* Build the svc_rqst used by the common processing routine */
1557        rqstp->rq_xid = req->rq_xid;
1558        rqstp->rq_prot = req->rq_xprt->prot;
1559        rqstp->rq_server = serv;
1560        rqstp->rq_bc_net = req->rq_xprt->xprt_net;
1561
1562        rqstp->rq_addrlen = sizeof(req->rq_xprt->addr);
1563        memcpy(&rqstp->rq_addr, &req->rq_xprt->addr, rqstp->rq_addrlen);
1564        memcpy(&rqstp->rq_arg, &req->rq_rcv_buf, sizeof(rqstp->rq_arg));
1565        memcpy(&rqstp->rq_res, &req->rq_snd_buf, sizeof(rqstp->rq_res));
1566
1567        /* Adjust the argument buffer length */
1568        rqstp->rq_arg.len = req->rq_private_buf.len;
1569        if (rqstp->rq_arg.len <= rqstp->rq_arg.head[0].iov_len) {
1570                rqstp->rq_arg.head[0].iov_len = rqstp->rq_arg.len;
1571                rqstp->rq_arg.page_len = 0;
1572        } else if (rqstp->rq_arg.len <= rqstp->rq_arg.head[0].iov_len +
1573                        rqstp->rq_arg.page_len)
1574                rqstp->rq_arg.page_len = rqstp->rq_arg.len -
1575                        rqstp->rq_arg.head[0].iov_len;
1576        else
1577                rqstp->rq_arg.len = rqstp->rq_arg.head[0].iov_len +
1578                        rqstp->rq_arg.page_len;
1579
1580        /* reset result send buffer "put" position */
1581        resv->iov_len = 0;
1582
1583        /*
1584         * Skip the next two words because they've already been
1585         * processed in the transport
1586         */
1587        svc_getu32(argv);       /* XID */
1588        svc_getnl(argv);        /* CALLDIR */
1589
1590        /* Parse and execute the bc call */
1591        proc_error = svc_process_common(rqstp, argv, resv);
1592
1593        atomic_dec(&req->rq_xprt->bc_slot_count);
1594        if (!proc_error) {
1595                /* Processing error: drop the request */
1596                xprt_free_bc_request(req);
1597                error = -EINVAL;
1598                goto out;
1599        }
1600        /* Finally, send the reply synchronously */
1601        memcpy(&req->rq_snd_buf, &rqstp->rq_res, sizeof(req->rq_snd_buf));
1602        task = rpc_run_bc_task(req);
1603        if (IS_ERR(task)) {
1604                error = PTR_ERR(task);
1605                goto out;
1606        }
1607
1608        WARN_ON_ONCE(atomic_read(&task->tk_count) != 1);
1609        error = task->tk_status;
1610        rpc_put_task(task);
1611
1612out:
1613        dprintk("svc: %s(), error=%d\n", __func__, error);
1614        return error;
1615}
1616EXPORT_SYMBOL_GPL(bc_svc_process);
1617#endif /* CONFIG_SUNRPC_BACKCHANNEL */
1618
1619/*
1620 * Return (transport-specific) limit on the rpc payload.
1621 */
1622u32 svc_max_payload(const struct svc_rqst *rqstp)
1623{
1624        u32 max = rqstp->rq_xprt->xpt_class->xcl_max_payload;
1625
1626        if (rqstp->rq_server->sv_max_payload < max)
1627                max = rqstp->rq_server->sv_max_payload;
1628        return max;
1629}
1630EXPORT_SYMBOL_GPL(svc_max_payload);
1631
1632/**
1633 * svc_encode_result_payload - mark a range of bytes as a result payload
1634 * @rqstp: svc_rqst to operate on
1635 * @offset: payload's byte offset in rqstp->rq_res
1636 * @length: size of payload, in bytes
1637 *
1638 * Returns zero on success, or a negative errno if a permanent
1639 * error occurred.
1640 */
1641int svc_encode_result_payload(struct svc_rqst *rqstp, unsigned int offset,
1642                              unsigned int length)
1643{
1644        return rqstp->rq_xprt->xpt_ops->xpo_result_payload(rqstp, offset,
1645                                                           length);
1646}
1647EXPORT_SYMBOL_GPL(svc_encode_result_payload);
1648
1649/**
1650 * svc_fill_write_vector - Construct data argument for VFS write call
1651 * @rqstp: svc_rqst to operate on
1652 * @pages: list of pages containing data payload
1653 * @first: buffer containing first section of write payload
1654 * @total: total number of bytes of write payload
1655 *
1656 * Fills in rqstp::rq_vec, and returns the number of elements.
1657 */
1658unsigned int svc_fill_write_vector(struct svc_rqst *rqstp, struct page **pages,
1659                                   struct kvec *first, size_t total)
1660{
1661        struct kvec *vec = rqstp->rq_vec;
1662        unsigned int i;
1663
1664        /* Some types of transport can present the write payload
1665         * entirely in rq_arg.pages. In this case, @first is empty.
1666         */
1667        i = 0;
1668        if (first->iov_len) {
1669                vec[i].iov_base = first->iov_base;
1670                vec[i].iov_len = min_t(size_t, total, first->iov_len);
1671                total -= vec[i].iov_len;
1672                ++i;
1673        }
1674
1675        while (total) {
1676                vec[i].iov_base = page_address(*pages);
1677                vec[i].iov_len = min_t(size_t, total, PAGE_SIZE);
1678                total -= vec[i].iov_len;
1679                ++i;
1680                ++pages;
1681        }
1682
1683        WARN_ON_ONCE(i > ARRAY_SIZE(rqstp->rq_vec));
1684        return i;
1685}
1686EXPORT_SYMBOL_GPL(svc_fill_write_vector);
1687
1688/**
1689 * svc_fill_symlink_pathname - Construct pathname argument for VFS symlink call
1690 * @rqstp: svc_rqst to operate on
1691 * @first: buffer containing first section of pathname
1692 * @p: buffer containing remaining section of pathname
1693 * @total: total length of the pathname argument
1694 *
1695 * The VFS symlink API demands a NUL-terminated pathname in mapped memory.
1696 * Returns pointer to a NUL-terminated string, or an ERR_PTR. Caller must free
1697 * the returned string.
1698 */
1699char *svc_fill_symlink_pathname(struct svc_rqst *rqstp, struct kvec *first,
1700                                void *p, size_t total)
1701{
1702        size_t len, remaining;
1703        char *result, *dst;
1704
1705        result = kmalloc(total + 1, GFP_KERNEL);
1706        if (!result)
1707                return ERR_PTR(-ESERVERFAULT);
1708
1709        dst = result;
1710        remaining = total;
1711
1712        len = min_t(size_t, total, first->iov_len);
1713        if (len) {
1714                memcpy(dst, first->iov_base, len);
1715                dst += len;
1716                remaining -= len;
1717        }
1718
1719        if (remaining) {
1720                len = min_t(size_t, remaining, PAGE_SIZE);
1721                memcpy(dst, p, len);
1722                dst += len;
1723        }
1724
1725        *dst = '\0';
1726
1727        /* Sanity check: Linux doesn't allow the pathname argument to
1728         * contain a NUL byte.
1729         */
1730        if (strlen(result) != total) {
1731                kfree(result);
1732                return ERR_PTR(-EINVAL);
1733        }
1734        return result;
1735}
1736EXPORT_SYMBOL_GPL(svc_fill_symlink_pathname);
1737