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_argpages.
 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_argp = kmalloc_node(serv->sv_xdrsize, GFP_KERNEL, node);
 618	if (!rqstp->rq_argp)
 619		goto out_enomem;
 620
 621	rqstp->rq_resp = kmalloc_node(serv->sv_xdrsize, GFP_KERNEL, node);
 622	if (!rqstp->rq_resp)
 623		goto out_enomem;
 624
 625	if (!svc_init_buffer(rqstp, serv->sv_max_mesg, node))
 626		goto out_enomem;
 627
 628	return rqstp;
 629out_enomem:
 630	svc_rqst_free(rqstp);
 631	return NULL;
 632}
 633EXPORT_SYMBOL_GPL(svc_rqst_alloc);
 634
 635struct svc_rqst *
 636svc_prepare_thread(struct svc_serv *serv, struct svc_pool *pool, int node)
 637{
 638	struct svc_rqst	*rqstp;
 639
 640	rqstp = svc_rqst_alloc(serv, pool, node);
 641	if (!rqstp)
 642		return ERR_PTR(-ENOMEM);
 643
 644	serv->sv_nrthreads++;
 645	spin_lock_bh(&pool->sp_lock);
 646	pool->sp_nrthreads++;
 647	list_add_rcu(&rqstp->rq_all, &pool->sp_all_threads);
 648	spin_unlock_bh(&pool->sp_lock);
 649	return rqstp;
 650}
 651EXPORT_SYMBOL_GPL(svc_prepare_thread);
 652
 653/*
 654 * Choose a pool in which to create a new thread, for svc_set_num_threads
 655 */
 656static inline struct svc_pool *
 657choose_pool(struct svc_serv *serv, struct svc_pool *pool, unsigned int *state)
 658{
 659	if (pool != NULL)
 660		return pool;
 661
 662	return &serv->sv_pools[(*state)++ % serv->sv_nrpools];
 663}
 664
 665/*
 666 * Choose a thread to kill, for svc_set_num_threads
 667 */
 668static inline struct task_struct *
 669choose_victim(struct svc_serv *serv, struct svc_pool *pool, unsigned int *state)
 670{
 671	unsigned int i;
 672	struct task_struct *task = NULL;
 673
 674	if (pool != NULL) {
 675		spin_lock_bh(&pool->sp_lock);
 676	} else {
 677		/* choose a pool in round-robin fashion */
 678		for (i = 0; i < serv->sv_nrpools; i++) {
 679			pool = &serv->sv_pools[--(*state) % serv->sv_nrpools];
 680			spin_lock_bh(&pool->sp_lock);
 681			if (!list_empty(&pool->sp_all_threads))
 682				goto found_pool;
 683			spin_unlock_bh(&pool->sp_lock);
 684		}
 685		return NULL;
 686	}
 687
 688found_pool:
 689	if (!list_empty(&pool->sp_all_threads)) {
 690		struct svc_rqst *rqstp;
 691
 692		/*
 693		 * Remove from the pool->sp_all_threads list
 694		 * so we don't try to kill it again.
 695		 */
 696		rqstp = list_entry(pool->sp_all_threads.next, struct svc_rqst, rq_all);
 697		set_bit(RQ_VICTIM, &rqstp->rq_flags);
 698		list_del_rcu(&rqstp->rq_all);
 699		task = rqstp->rq_task;
 700	}
 701	spin_unlock_bh(&pool->sp_lock);
 702
 703	return task;
 704}
 705
 706/* create new threads */
 707static int
 708svc_start_kthreads(struct svc_serv *serv, struct svc_pool *pool, int nrservs)
 709{
 710	struct svc_rqst	*rqstp;
 711	struct task_struct *task;
 712	struct svc_pool *chosen_pool;
 713	unsigned int state = serv->sv_nrthreads-1;
 714	int node;
 715
 716	do {
 717		nrservs--;
 718		chosen_pool = choose_pool(serv, pool, &state);
 719
 720		node = svc_pool_map_get_node(chosen_pool->sp_id);
 721		rqstp = svc_prepare_thread(serv, chosen_pool, node);
 722		if (IS_ERR(rqstp))
 723			return PTR_ERR(rqstp);
 724
 725		__module_get(serv->sv_ops->svo_module);
 726		task = kthread_create_on_node(serv->sv_ops->svo_function, rqstp,
 727					      node, "%s", serv->sv_name);
 728		if (IS_ERR(task)) {
 729			module_put(serv->sv_ops->svo_module);
 730			svc_exit_thread(rqstp);
 731			return PTR_ERR(task);
 732		}
 733
 734		rqstp->rq_task = task;
 735		if (serv->sv_nrpools > 1)
 736			svc_pool_map_set_cpumask(task, chosen_pool->sp_id);
 737
 738		svc_sock_update_bufs(serv);
 739		wake_up_process(task);
 740	} while (nrservs > 0);
 741
 742	return 0;
 743}
 744
 745
 746/* destroy old threads */
 747static int
 748svc_signal_kthreads(struct svc_serv *serv, struct svc_pool *pool, int nrservs)
 749{
 750	struct task_struct *task;
 751	unsigned int state = serv->sv_nrthreads-1;
 752
 753	/* destroy old threads */
 754	do {
 755		task = choose_victim(serv, pool, &state);
 756		if (task == NULL)
 757			break;
 758		send_sig(SIGINT, task, 1);
 759		nrservs++;
 760	} while (nrservs < 0);
 761
 762	return 0;
 763}
 764
 765/*
 766 * Create or destroy enough new threads to make the number
 767 * of threads the given number.  If `pool' is non-NULL, applies
 768 * only to threads in that pool, otherwise round-robins between
 769 * all pools.  Caller must ensure that mutual exclusion between this and
 770 * server startup or shutdown.
 771 *
 772 * Destroying threads relies on the service threads filling in
 773 * rqstp->rq_task, which only the nfs ones do.  Assumes the serv
 774 * has been created using svc_create_pooled().
 775 *
 776 * Based on code that used to be in nfsd_svc() but tweaked
 777 * to be pool-aware.
 778 */
 779int
 780svc_set_num_threads(struct svc_serv *serv, struct svc_pool *pool, int nrservs)
 781{
 782	if (pool == NULL) {
 783		/* The -1 assumes caller has done a svc_get() */
 784		nrservs -= (serv->sv_nrthreads-1);
 785	} else {
 786		spin_lock_bh(&pool->sp_lock);
 787		nrservs -= pool->sp_nrthreads;
 788		spin_unlock_bh(&pool->sp_lock);
 789	}
 790
 791	if (nrservs > 0)
 792		return svc_start_kthreads(serv, pool, nrservs);
 793	if (nrservs < 0)
 794		return svc_signal_kthreads(serv, pool, nrservs);
 795	return 0;
 796}
 797EXPORT_SYMBOL_GPL(svc_set_num_threads);
 798
 799/* destroy old threads */
 800static int
 801svc_stop_kthreads(struct svc_serv *serv, struct svc_pool *pool, int nrservs)
 802{
 803	struct task_struct *task;
 804	unsigned int state = serv->sv_nrthreads-1;
 805
 806	/* destroy old threads */
 807	do {
 808		task = choose_victim(serv, pool, &state);
 809		if (task == NULL)
 810			break;
 811		kthread_stop(task);
 812		nrservs++;
 813	} while (nrservs < 0);
 814	return 0;
 815}
 816
 817int
 818svc_set_num_threads_sync(struct svc_serv *serv, struct svc_pool *pool, int nrservs)
 819{
 820	if (pool == NULL) {
 821		/* The -1 assumes caller has done a svc_get() */
 822		nrservs -= (serv->sv_nrthreads-1);
 823	} else {
 824		spin_lock_bh(&pool->sp_lock);
 825		nrservs -= pool->sp_nrthreads;
 826		spin_unlock_bh(&pool->sp_lock);
 827	}
 828
 829	if (nrservs > 0)
 830		return svc_start_kthreads(serv, pool, nrservs);
 831	if (nrservs < 0)
 832		return svc_stop_kthreads(serv, pool, nrservs);
 833	return 0;
 834}
 835EXPORT_SYMBOL_GPL(svc_set_num_threads_sync);
 836
 837/*
 838 * Called from a server thread as it's exiting. Caller must hold the "service
 839 * mutex" for the service.
 840 */
 841void
 842svc_rqst_free(struct svc_rqst *rqstp)
 843{
 844	svc_release_buffer(rqstp);
 845	kfree(rqstp->rq_resp);
 846	kfree(rqstp->rq_argp);
 847	kfree(rqstp->rq_auth_data);
 848	kfree_rcu(rqstp, rq_rcu_head);
 849}
 850EXPORT_SYMBOL_GPL(svc_rqst_free);
 851
 852void
 853svc_exit_thread(struct svc_rqst *rqstp)
 854{
 855	struct svc_serv	*serv = rqstp->rq_server;
 856	struct svc_pool	*pool = rqstp->rq_pool;
 857
 858	spin_lock_bh(&pool->sp_lock);
 859	pool->sp_nrthreads--;
 860	if (!test_and_set_bit(RQ_VICTIM, &rqstp->rq_flags))
 861		list_del_rcu(&rqstp->rq_all);
 862	spin_unlock_bh(&pool->sp_lock);
 863
 864	svc_rqst_free(rqstp);
 865
 866	/* Release the server */
 867	if (serv)
 868		svc_destroy(serv);
 869}
 870EXPORT_SYMBOL_GPL(svc_exit_thread);
 871
 872/*
 873 * Register an "inet" protocol family netid with the local
 874 * rpcbind daemon via an rpcbind v4 SET request.
 875 *
 876 * No netconfig infrastructure is available in the kernel, so
 877 * we map IP_ protocol numbers to netids by hand.
 878 *
 879 * Returns zero on success; a negative errno value is returned
 880 * if any error occurs.
 881 */
 882static int __svc_rpcb_register4(struct net *net, const u32 program,
 883				const u32 version,
 884				const unsigned short protocol,
 885				const unsigned short port)
 886{
 887	const struct sockaddr_in sin = {
 888		.sin_family		= AF_INET,
 889		.sin_addr.s_addr	= htonl(INADDR_ANY),
 890		.sin_port		= htons(port),
 891	};
 892	const char *netid;
 893	int error;
 894
 895	switch (protocol) {
 896	case IPPROTO_UDP:
 897		netid = RPCBIND_NETID_UDP;
 898		break;
 899	case IPPROTO_TCP:
 900		netid = RPCBIND_NETID_TCP;
 901		break;
 902	default:
 903		return -ENOPROTOOPT;
 904	}
 905
 906	error = rpcb_v4_register(net, program, version,
 907					(const struct sockaddr *)&sin, netid);
 908
 909	/*
 910	 * User space didn't support rpcbind v4, so retry this
 911	 * registration request with the legacy rpcbind v2 protocol.
 912	 */
 913	if (error == -EPROTONOSUPPORT)
 914		error = rpcb_register(net, program, version, protocol, port);
 915
 916	return error;
 917}
 918
 919#if IS_ENABLED(CONFIG_IPV6)
 920/*
 921 * Register an "inet6" protocol family netid with the local
 922 * rpcbind daemon via an rpcbind v4 SET request.
 923 *
 924 * No netconfig infrastructure is available in the kernel, so
 925 * we map IP_ protocol numbers to netids by hand.
 926 *
 927 * Returns zero on success; a negative errno value is returned
 928 * if any error occurs.
 929 */
 930static int __svc_rpcb_register6(struct net *net, const u32 program,
 931				const u32 version,
 932				const unsigned short protocol,
 933				const unsigned short port)
 934{
 935	const struct sockaddr_in6 sin6 = {
 936		.sin6_family		= AF_INET6,
 937		.sin6_addr		= IN6ADDR_ANY_INIT,
 938		.sin6_port		= htons(port),
 939	};
 940	const char *netid;
 941	int error;
 942
 943	switch (protocol) {
 944	case IPPROTO_UDP:
 945		netid = RPCBIND_NETID_UDP6;
 946		break;
 947	case IPPROTO_TCP:
 948		netid = RPCBIND_NETID_TCP6;
 949		break;
 950	default:
 951		return -ENOPROTOOPT;
 952	}
 953
 954	error = rpcb_v4_register(net, program, version,
 955					(const struct sockaddr *)&sin6, netid);
 956
 957	/*
 958	 * User space didn't support rpcbind version 4, so we won't
 959	 * use a PF_INET6 listener.
 960	 */
 961	if (error == -EPROTONOSUPPORT)
 962		error = -EAFNOSUPPORT;
 963
 964	return error;
 965}
 966#endif	/* IS_ENABLED(CONFIG_IPV6) */
 967
 968/*
 969 * Register a kernel RPC service via rpcbind version 4.
 970 *
 971 * Returns zero on success; a negative errno value is returned
 972 * if any error occurs.
 973 */
 974static int __svc_register(struct net *net, const char *progname,
 975			  const u32 program, const u32 version,
 976			  const int family,
 977			  const unsigned short protocol,
 978			  const unsigned short port)
 979{
 980	int error = -EAFNOSUPPORT;
 981
 982	switch (family) {
 983	case PF_INET:
 984		error = __svc_rpcb_register4(net, program, version,
 985						protocol, port);
 986		break;
 987#if IS_ENABLED(CONFIG_IPV6)
 988	case PF_INET6:
 989		error = __svc_rpcb_register6(net, program, version,
 990						protocol, port);
 991#endif
 992	}
 993
 994	trace_svc_register(progname, version, protocol, port, family, error);
 995	return error;
 996}
 997
 998int svc_rpcbind_set_version(struct net *net,
 999			    const struct svc_program *progp,
1000			    u32 version, int family,
1001			    unsigned short proto,
1002			    unsigned short port)
1003{
 
 
 
 
 
1004	return __svc_register(net, progp->pg_name, progp->pg_prog,
1005				version, family, proto, port);
1006
1007}
1008EXPORT_SYMBOL_GPL(svc_rpcbind_set_version);
1009
1010int svc_generic_rpcbind_set(struct net *net,
1011			    const struct svc_program *progp,
1012			    u32 version, int family,
1013			    unsigned short proto,
1014			    unsigned short port)
1015{
1016	const struct svc_version *vers = progp->pg_vers[version];
1017	int error;
1018
1019	if (vers == NULL)
1020		return 0;
1021
1022	if (vers->vs_hidden) {
1023		trace_svc_noregister(progp->pg_name, version, proto,
1024				     port, family, 0);
 
 
 
1025		return 0;
1026	}
1027
1028	/*
1029	 * Don't register a UDP port if we need congestion
1030	 * control.
1031	 */
1032	if (vers->vs_need_cong_ctrl && proto == IPPROTO_UDP)
1033		return 0;
1034
1035	error = svc_rpcbind_set_version(net, progp, version,
1036					family, proto, port);
1037
1038	return (vers->vs_rpcb_optnl) ? 0 : error;
1039}
1040EXPORT_SYMBOL_GPL(svc_generic_rpcbind_set);
1041
1042/**
1043 * svc_register - register an RPC service with the local portmapper
1044 * @serv: svc_serv struct for the service to register
1045 * @net: net namespace for the service to register
1046 * @family: protocol family of service's listener socket
1047 * @proto: transport protocol number to advertise
1048 * @port: port to advertise
1049 *
1050 * Service is registered for any address in the passed-in protocol family
1051 */
1052int svc_register(const struct svc_serv *serv, struct net *net,
1053		 const int family, const unsigned short proto,
1054		 const unsigned short port)
1055{
1056	struct svc_program	*progp;
1057	unsigned int		i;
1058	int			error = 0;
1059
1060	WARN_ON_ONCE(proto == 0 && port == 0);
1061	if (proto == 0 && port == 0)
1062		return -EINVAL;
1063
1064	for (progp = serv->sv_program; progp; progp = progp->pg_next) {
1065		for (i = 0; i < progp->pg_nvers; i++) {
1066
1067			error = progp->pg_rpcbind_set(net, progp, i,
1068					family, proto, port);
1069			if (error < 0) {
1070				printk(KERN_WARNING "svc: failed to register "
1071					"%sv%u RPC service (errno %d).\n",
1072					progp->pg_name, i, -error);
1073				break;
1074			}
1075		}
1076	}
1077
1078	return error;
1079}
1080
1081/*
1082 * If user space is running rpcbind, it should take the v4 UNSET
1083 * and clear everything for this [program, version].  If user space
1084 * is running portmap, it will reject the v4 UNSET, but won't have
1085 * any "inet6" entries anyway.  So a PMAP_UNSET should be sufficient
1086 * in this case to clear all existing entries for [program, version].
1087 */
1088static void __svc_unregister(struct net *net, const u32 program, const u32 version,
1089			     const char *progname)
1090{
1091	int error;
1092
1093	error = rpcb_v4_register(net, program, version, NULL, "");
1094
1095	/*
1096	 * User space didn't support rpcbind v4, so retry this
1097	 * request with the legacy rpcbind v2 protocol.
1098	 */
1099	if (error == -EPROTONOSUPPORT)
1100		error = rpcb_register(net, program, version, 0, 0);
1101
1102	trace_svc_unregister(progname, version, error);
 
1103}
1104
1105/*
1106 * All netids, bind addresses and ports registered for [program, version]
1107 * are removed from the local rpcbind database (if the service is not
1108 * hidden) to make way for a new instance of the service.
1109 *
1110 * The result of unregistration is reported via dprintk for those who want
1111 * verification of the result, but is otherwise not important.
1112 */
1113static void svc_unregister(const struct svc_serv *serv, struct net *net)
1114{
1115	struct svc_program *progp;
1116	unsigned long flags;
1117	unsigned int i;
1118
1119	clear_thread_flag(TIF_SIGPENDING);
1120
1121	for (progp = serv->sv_program; progp; progp = progp->pg_next) {
1122		for (i = 0; i < progp->pg_nvers; i++) {
1123			if (progp->pg_vers[i] == NULL)
1124				continue;
1125			if (progp->pg_vers[i]->vs_hidden)
1126				continue;
 
 
 
1127			__svc_unregister(net, progp->pg_prog, i, progp->pg_name);
1128		}
1129	}
1130
1131	spin_lock_irqsave(&current->sighand->siglock, flags);
1132	recalc_sigpending();
1133	spin_unlock_irqrestore(&current->sighand->siglock, flags);
1134}
1135
1136/*
1137 * dprintk the given error with the address of the client that caused it.
1138 */
1139#if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
1140static __printf(2, 3)
1141void svc_printk(struct svc_rqst *rqstp, const char *fmt, ...)
1142{
1143	struct va_format vaf;
1144	va_list args;
1145	char 	buf[RPC_MAX_ADDRBUFLEN];
1146
1147	va_start(args, fmt);
1148
1149	vaf.fmt = fmt;
1150	vaf.va = &args;
1151
1152	dprintk("svc: %s: %pV", svc_print_addr(rqstp, buf, sizeof(buf)), &vaf);
1153
1154	va_end(args);
1155}
1156#else
1157static __printf(2,3) void svc_printk(struct svc_rqst *rqstp, const char *fmt, ...) {}
1158#endif
1159
1160__be32
1161svc_return_autherr(struct svc_rqst *rqstp, __be32 auth_err)
1162{
1163	set_bit(RQ_AUTHERR, &rqstp->rq_flags);
1164	return auth_err;
1165}
1166EXPORT_SYMBOL_GPL(svc_return_autherr);
1167
1168static __be32
1169svc_get_autherr(struct svc_rqst *rqstp, __be32 *statp)
1170{
1171	if (test_and_clear_bit(RQ_AUTHERR, &rqstp->rq_flags))
1172		return *statp;
1173	return rpc_auth_ok;
1174}
1175
1176static int
1177svc_generic_dispatch(struct svc_rqst *rqstp, __be32 *statp)
1178{
1179	struct kvec *argv = &rqstp->rq_arg.head[0];
1180	struct kvec *resv = &rqstp->rq_res.head[0];
1181	const struct svc_procedure *procp = rqstp->rq_procinfo;
1182
1183	/*
1184	 * Decode arguments
1185	 * XXX: why do we ignore the return value?
1186	 */
1187	if (procp->pc_decode &&
1188	    !procp->pc_decode(rqstp, argv->iov_base)) {
1189		*statp = rpc_garbage_args;
1190		return 1;
1191	}
1192
1193	*statp = procp->pc_func(rqstp);
1194
1195	if (*statp == rpc_drop_reply ||
1196	    test_bit(RQ_DROPME, &rqstp->rq_flags))
1197		return 0;
1198
1199	if (test_bit(RQ_AUTHERR, &rqstp->rq_flags))
1200		return 1;
1201
1202	if (*statp != rpc_success)
1203		return 1;
1204
1205	/* Encode reply */
1206	if (procp->pc_encode &&
1207	    !procp->pc_encode(rqstp, resv->iov_base + resv->iov_len)) {
1208		dprintk("svc: failed to encode reply\n");
1209		/* serv->sv_stats->rpcsystemerr++; */
1210		*statp = rpc_system_err;
1211	}
1212	return 1;
1213}
1214
1215__be32
1216svc_generic_init_request(struct svc_rqst *rqstp,
1217		const struct svc_program *progp,
1218		struct svc_process_info *ret)
1219{
1220	const struct svc_version *versp = NULL;	/* compiler food */
1221	const struct svc_procedure *procp = NULL;
1222
1223	if (rqstp->rq_vers >= progp->pg_nvers )
1224		goto err_bad_vers;
1225	versp = progp->pg_vers[rqstp->rq_vers];
1226	if (!versp)
1227		goto err_bad_vers;
1228
1229	/*
1230	 * Some protocol versions (namely NFSv4) require some form of
1231	 * congestion control.  (See RFC 7530 section 3.1 paragraph 2)
1232	 * In other words, UDP is not allowed. We mark those when setting
1233	 * up the svc_xprt, and verify that here.
1234	 *
1235	 * The spec is not very clear about what error should be returned
1236	 * when someone tries to access a server that is listening on UDP
1237	 * for lower versions. RPC_PROG_MISMATCH seems to be the closest
1238	 * fit.
1239	 */
1240	if (versp->vs_need_cong_ctrl && rqstp->rq_xprt &&
1241	    !test_bit(XPT_CONG_CTRL, &rqstp->rq_xprt->xpt_flags))
1242		goto err_bad_vers;
1243
1244	if (rqstp->rq_proc >= versp->vs_nproc)
1245		goto err_bad_proc;
1246	rqstp->rq_procinfo = procp = &versp->vs_proc[rqstp->rq_proc];
1247	if (!procp)
1248		goto err_bad_proc;
1249
1250	/* Initialize storage for argp and resp */
1251	memset(rqstp->rq_argp, 0, procp->pc_argsize);
1252	memset(rqstp->rq_resp, 0, procp->pc_ressize);
1253
1254	/* Bump per-procedure stats counter */
1255	versp->vs_count[rqstp->rq_proc]++;
1256
1257	ret->dispatch = versp->vs_dispatch;
1258	return rpc_success;
1259err_bad_vers:
1260	ret->mismatch.lovers = progp->pg_lovers;
1261	ret->mismatch.hivers = progp->pg_hivers;
1262	return rpc_prog_mismatch;
1263err_bad_proc:
1264	return rpc_proc_unavail;
1265}
1266EXPORT_SYMBOL_GPL(svc_generic_init_request);
1267
1268/*
1269 * Common routine for processing the RPC request.
1270 */
1271static int
1272svc_process_common(struct svc_rqst *rqstp, struct kvec *argv, struct kvec *resv)
1273{
1274	struct svc_program	*progp;
1275	const struct svc_procedure *procp = NULL;
1276	struct svc_serv		*serv = rqstp->rq_server;
1277	struct svc_process_info process;
1278	__be32			*statp;
1279	u32			prog, vers;
1280	__be32			auth_stat, rpc_stat;
1281	int			auth_res;
1282	__be32			*reply_statp;
1283
1284	rpc_stat = rpc_success;
1285
1286	if (argv->iov_len < 6*4)
1287		goto err_short_len;
1288
1289	/* Will be turned off by GSS integrity and privacy services */
1290	set_bit(RQ_SPLICE_OK, &rqstp->rq_flags);
1291	/* Will be turned off only when NFSv4 Sessions are used */
1292	set_bit(RQ_USEDEFERRAL, &rqstp->rq_flags);
1293	clear_bit(RQ_DROPME, &rqstp->rq_flags);
1294
1295	svc_putu32(resv, rqstp->rq_xid);
1296
1297	vers = svc_getnl(argv);
1298
1299	/* First words of reply: */
1300	svc_putnl(resv, 1);		/* REPLY */
1301
1302	if (vers != 2)		/* RPC version number */
1303		goto err_bad_rpc;
1304
1305	/* Save position in case we later decide to reject: */
1306	reply_statp = resv->iov_base + resv->iov_len;
1307
1308	svc_putnl(resv, 0);		/* ACCEPT */
1309
1310	rqstp->rq_prog = prog = svc_getnl(argv);	/* program number */
1311	rqstp->rq_vers = svc_getnl(argv);	/* version number */
1312	rqstp->rq_proc = svc_getnl(argv);	/* procedure number */
1313
1314	for (progp = serv->sv_program; progp; progp = progp->pg_next)
1315		if (prog == progp->pg_prog)
1316			break;
1317
1318	/*
1319	 * Decode auth data, and add verifier to reply buffer.
1320	 * We do this before anything else in order to get a decent
1321	 * auth verifier.
1322	 */
1323	auth_res = svc_authenticate(rqstp, &auth_stat);
1324	/* Also give the program a chance to reject this call: */
1325	if (auth_res == SVC_OK && progp) {
1326		auth_stat = rpc_autherr_badcred;
1327		auth_res = progp->pg_authenticate(rqstp);
1328	}
1329	if (auth_res != SVC_OK)
1330		trace_svc_authenticate(rqstp, auth_res, auth_stat);
1331	switch (auth_res) {
1332	case SVC_OK:
1333		break;
1334	case SVC_GARBAGE:
1335		goto err_garbage;
1336	case SVC_SYSERR:
1337		rpc_stat = rpc_system_err;
1338		goto err_bad;
1339	case SVC_DENIED:
1340		goto err_bad_auth;
1341	case SVC_CLOSE:
1342		goto close;
1343	case SVC_DROP:
1344		goto dropit;
1345	case SVC_COMPLETE:
1346		goto sendit;
1347	}
1348
1349	if (progp == NULL)
1350		goto err_bad_prog;
1351
1352	rpc_stat = progp->pg_init_request(rqstp, progp, &process);
1353	switch (rpc_stat) {
1354	case rpc_success:
1355		break;
1356	case rpc_prog_unavail:
1357		goto err_bad_prog;
1358	case rpc_prog_mismatch:
1359		goto err_bad_vers;
1360	case rpc_proc_unavail:
1361		goto err_bad_proc;
1362	}
1363
1364	procp = rqstp->rq_procinfo;
1365	/* Should this check go into the dispatcher? */
1366	if (!procp || !procp->pc_func)
1367		goto err_bad_proc;
1368
1369	/* Syntactic check complete */
1370	serv->sv_stats->rpccnt++;
1371	trace_svc_process(rqstp, progp->pg_name);
1372
1373	/* Build the reply header. */
1374	statp = resv->iov_base +resv->iov_len;
1375	svc_putnl(resv, RPC_SUCCESS);
1376
1377	/* un-reserve some of the out-queue now that we have a
1378	 * better idea of reply size
1379	 */
1380	if (procp->pc_xdrressize)
1381		svc_reserve_auth(rqstp, procp->pc_xdrressize<<2);
1382
1383	/* Call the function that processes the request. */
1384	if (!process.dispatch) {
1385		if (!svc_generic_dispatch(rqstp, statp))
1386			goto release_dropit;
1387		if (*statp == rpc_garbage_args)
1388			goto err_garbage;
1389		auth_stat = svc_get_autherr(rqstp, statp);
1390		if (auth_stat != rpc_auth_ok)
1391			goto err_release_bad_auth;
1392	} else {
1393		dprintk("svc: calling dispatcher\n");
1394		if (!process.dispatch(rqstp, statp))
1395			goto release_dropit; /* Release reply info */
1396	}
1397
1398	/* Check RPC status result */
1399	if (*statp != rpc_success)
1400		resv->iov_len = ((void*)statp)  - resv->iov_base + 4;
1401
1402	/* Release reply info */
1403	if (procp->pc_release)
1404		procp->pc_release(rqstp);
1405
1406	if (procp->pc_encode == NULL)
1407		goto dropit;
1408
1409 sendit:
1410	if (svc_authorise(rqstp))
1411		goto close;
1412	return 1;		/* Caller can now send it */
1413
1414release_dropit:
1415	if (procp->pc_release)
1416		procp->pc_release(rqstp);
1417 dropit:
1418	svc_authorise(rqstp);	/* doesn't hurt to call this twice */
1419	dprintk("svc: svc_process dropit\n");
1420	return 0;
1421
1422 close:
1423	if (rqstp->rq_xprt && test_bit(XPT_TEMP, &rqstp->rq_xprt->xpt_flags))
1424		svc_close_xprt(rqstp->rq_xprt);
1425	dprintk("svc: svc_process close\n");
1426	return 0;
1427
1428err_short_len:
1429	svc_printk(rqstp, "short len %zd, dropping request\n",
1430			argv->iov_len);
1431	goto close;
1432
1433err_bad_rpc:
1434	serv->sv_stats->rpcbadfmt++;
1435	svc_putnl(resv, 1);	/* REJECT */
1436	svc_putnl(resv, 0);	/* RPC_MISMATCH */
1437	svc_putnl(resv, 2);	/* Only RPCv2 supported */
1438	svc_putnl(resv, 2);
1439	goto sendit;
1440
1441err_release_bad_auth:
1442	if (procp->pc_release)
1443		procp->pc_release(rqstp);
1444err_bad_auth:
1445	dprintk("svc: authentication failed (%d)\n", ntohl(auth_stat));
1446	serv->sv_stats->rpcbadauth++;
1447	/* Restore write pointer to location of accept status: */
1448	xdr_ressize_check(rqstp, reply_statp);
1449	svc_putnl(resv, 1);	/* REJECT */
1450	svc_putnl(resv, 1);	/* AUTH_ERROR */
1451	svc_putnl(resv, ntohl(auth_stat));	/* status */
1452	goto sendit;
1453
1454err_bad_prog:
1455	dprintk("svc: unknown program %d\n", prog);
1456	serv->sv_stats->rpcbadfmt++;
1457	svc_putnl(resv, RPC_PROG_UNAVAIL);
1458	goto sendit;
1459
1460err_bad_vers:
1461	svc_printk(rqstp, "unknown version (%d for prog %d, %s)\n",
1462		       rqstp->rq_vers, rqstp->rq_prog, progp->pg_name);
1463
1464	serv->sv_stats->rpcbadfmt++;
1465	svc_putnl(resv, RPC_PROG_MISMATCH);
1466	svc_putnl(resv, process.mismatch.lovers);
1467	svc_putnl(resv, process.mismatch.hivers);
1468	goto sendit;
1469
1470err_bad_proc:
1471	svc_printk(rqstp, "unknown procedure (%d)\n", rqstp->rq_proc);
1472
1473	serv->sv_stats->rpcbadfmt++;
1474	svc_putnl(resv, RPC_PROC_UNAVAIL);
1475	goto sendit;
1476
1477err_garbage:
1478	svc_printk(rqstp, "failed to decode args\n");
1479
1480	rpc_stat = rpc_garbage_args;
1481err_bad:
1482	serv->sv_stats->rpcbadfmt++;
1483	svc_putnl(resv, ntohl(rpc_stat));
1484	goto sendit;
1485}
1486
1487/*
1488 * Process the RPC request.
1489 */
1490int
1491svc_process(struct svc_rqst *rqstp)
1492{
1493	struct kvec		*argv = &rqstp->rq_arg.head[0];
1494	struct kvec		*resv = &rqstp->rq_res.head[0];
1495	struct svc_serv		*serv = rqstp->rq_server;
1496	u32			dir;
1497
1498	/*
1499	 * Setup response xdr_buf.
1500	 * Initially it has just one page
1501	 */
1502	rqstp->rq_next_page = &rqstp->rq_respages[1];
1503	resv->iov_base = page_address(rqstp->rq_respages[0]);
1504	resv->iov_len = 0;
1505	rqstp->rq_res.pages = rqstp->rq_respages + 1;
1506	rqstp->rq_res.len = 0;
1507	rqstp->rq_res.page_base = 0;
1508	rqstp->rq_res.page_len = 0;
1509	rqstp->rq_res.buflen = PAGE_SIZE;
1510	rqstp->rq_res.tail[0].iov_base = NULL;
1511	rqstp->rq_res.tail[0].iov_len = 0;
1512
1513	dir  = svc_getnl(argv);
1514	if (dir != 0) {
1515		/* direction != CALL */
1516		svc_printk(rqstp, "bad direction %d, dropping request\n", dir);
1517		serv->sv_stats->rpcbadfmt++;
1518		goto out_drop;
1519	}
1520
 
 
 
 
1521	/* Returns 1 for send, 0 for drop */
1522	if (likely(svc_process_common(rqstp, argv, resv)))
1523		return svc_send(rqstp);
1524
1525out_drop:
1526	svc_drop(rqstp);
1527	return 0;
1528}
1529EXPORT_SYMBOL_GPL(svc_process);
1530
1531#if defined(CONFIG_SUNRPC_BACKCHANNEL)
1532/*
1533 * Process a backchannel RPC request that arrived over an existing
1534 * outbound connection
1535 */
1536int
1537bc_svc_process(struct svc_serv *serv, struct rpc_rqst *req,
1538	       struct svc_rqst *rqstp)
1539{
1540	struct kvec	*argv = &rqstp->rq_arg.head[0];
1541	struct kvec	*resv = &rqstp->rq_res.head[0];
1542	struct rpc_task *task;
1543	int proc_error;
1544	int error;
1545
1546	dprintk("svc: %s(%p)\n", __func__, req);
1547
1548	/* Build the svc_rqst used by the common processing routine */
1549	rqstp->rq_xid = req->rq_xid;
1550	rqstp->rq_prot = req->rq_xprt->prot;
1551	rqstp->rq_server = serv;
1552	rqstp->rq_bc_net = req->rq_xprt->xprt_net;
1553
1554	rqstp->rq_addrlen = sizeof(req->rq_xprt->addr);
1555	memcpy(&rqstp->rq_addr, &req->rq_xprt->addr, rqstp->rq_addrlen);
1556	memcpy(&rqstp->rq_arg, &req->rq_rcv_buf, sizeof(rqstp->rq_arg));
1557	memcpy(&rqstp->rq_res, &req->rq_snd_buf, sizeof(rqstp->rq_res));
1558
1559	/* Adjust the argument buffer length */
1560	rqstp->rq_arg.len = req->rq_private_buf.len;
1561	if (rqstp->rq_arg.len <= rqstp->rq_arg.head[0].iov_len) {
1562		rqstp->rq_arg.head[0].iov_len = rqstp->rq_arg.len;
1563		rqstp->rq_arg.page_len = 0;
1564	} else if (rqstp->rq_arg.len <= rqstp->rq_arg.head[0].iov_len +
1565			rqstp->rq_arg.page_len)
1566		rqstp->rq_arg.page_len = rqstp->rq_arg.len -
1567			rqstp->rq_arg.head[0].iov_len;
1568	else
1569		rqstp->rq_arg.len = rqstp->rq_arg.head[0].iov_len +
1570			rqstp->rq_arg.page_len;
1571
1572	/* reset result send buffer "put" position */
1573	resv->iov_len = 0;
1574
1575	/*
1576	 * Skip the next two words because they've already been
1577	 * processed in the transport
1578	 */
1579	svc_getu32(argv);	/* XID */
1580	svc_getnl(argv);	/* CALLDIR */
1581
1582	/* Parse and execute the bc call */
1583	proc_error = svc_process_common(rqstp, argv, resv);
1584
1585	atomic_dec(&req->rq_xprt->bc_slot_count);
1586	if (!proc_error) {
1587		/* Processing error: drop the request */
1588		xprt_free_bc_request(req);
1589		error = -EINVAL;
1590		goto out;
1591	}
1592	/* Finally, send the reply synchronously */
1593	memcpy(&req->rq_snd_buf, &rqstp->rq_res, sizeof(req->rq_snd_buf));
1594	task = rpc_run_bc_task(req);
1595	if (IS_ERR(task)) {
1596		error = PTR_ERR(task);
1597		goto out;
1598	}
1599
1600	WARN_ON_ONCE(atomic_read(&task->tk_count) != 1);
1601	error = task->tk_status;
1602	rpc_put_task(task);
1603
1604out:
1605	dprintk("svc: %s(), error=%d\n", __func__, error);
1606	return error;
1607}
1608EXPORT_SYMBOL_GPL(bc_svc_process);
1609#endif /* CONFIG_SUNRPC_BACKCHANNEL */
1610
1611/*
1612 * Return (transport-specific) limit on the rpc payload.
1613 */
1614u32 svc_max_payload(const struct svc_rqst *rqstp)
1615{
1616	u32 max = rqstp->rq_xprt->xpt_class->xcl_max_payload;
1617
1618	if (rqstp->rq_server->sv_max_payload < max)
1619		max = rqstp->rq_server->sv_max_payload;
1620	return max;
1621}
1622EXPORT_SYMBOL_GPL(svc_max_payload);
1623
1624/**
1625 * svc_encode_read_payload - mark a range of bytes as a READ payload
1626 * @rqstp: svc_rqst to operate on
1627 * @offset: payload's byte offset in rqstp->rq_res
1628 * @length: size of payload, in bytes
1629 *
1630 * Returns zero on success, or a negative errno if a permanent
1631 * error occurred.
1632 */
1633int svc_encode_read_payload(struct svc_rqst *rqstp, unsigned int offset,
1634			    unsigned int length)
1635{
1636	return rqstp->rq_xprt->xpt_ops->xpo_read_payload(rqstp, offset, length);
1637}
1638EXPORT_SYMBOL_GPL(svc_encode_read_payload);
1639
1640/**
1641 * svc_fill_write_vector - Construct data argument for VFS write call
1642 * @rqstp: svc_rqst to operate on
1643 * @pages: list of pages containing data payload
1644 * @first: buffer containing first section of write payload
1645 * @total: total number of bytes of write payload
1646 *
1647 * Fills in rqstp::rq_vec, and returns the number of elements.
1648 */
1649unsigned int svc_fill_write_vector(struct svc_rqst *rqstp, struct page **pages,
1650				   struct kvec *first, size_t total)
1651{
1652	struct kvec *vec = rqstp->rq_vec;
1653	unsigned int i;
1654
1655	/* Some types of transport can present the write payload
1656	 * entirely in rq_arg.pages. In this case, @first is empty.
1657	 */
1658	i = 0;
1659	if (first->iov_len) {
1660		vec[i].iov_base = first->iov_base;
1661		vec[i].iov_len = min_t(size_t, total, first->iov_len);
1662		total -= vec[i].iov_len;
1663		++i;
1664	}
1665
1666	while (total) {
1667		vec[i].iov_base = page_address(*pages);
1668		vec[i].iov_len = min_t(size_t, total, PAGE_SIZE);
1669		total -= vec[i].iov_len;
1670		++i;
1671		++pages;
1672	}
1673
1674	WARN_ON_ONCE(i > ARRAY_SIZE(rqstp->rq_vec));
1675	return i;
1676}
1677EXPORT_SYMBOL_GPL(svc_fill_write_vector);
1678
1679/**
1680 * svc_fill_symlink_pathname - Construct pathname argument for VFS symlink call
1681 * @rqstp: svc_rqst to operate on
1682 * @first: buffer containing first section of pathname
1683 * @p: buffer containing remaining section of pathname
1684 * @total: total length of the pathname argument
1685 *
1686 * The VFS symlink API demands a NUL-terminated pathname in mapped memory.
1687 * Returns pointer to a NUL-terminated string, or an ERR_PTR. Caller must free
1688 * the returned string.
1689 */
1690char *svc_fill_symlink_pathname(struct svc_rqst *rqstp, struct kvec *first,
1691				void *p, size_t total)
1692{
1693	size_t len, remaining;
1694	char *result, *dst;
1695
1696	result = kmalloc(total + 1, GFP_KERNEL);
1697	if (!result)
1698		return ERR_PTR(-ESERVERFAULT);
1699
1700	dst = result;
1701	remaining = total;
1702
1703	len = min_t(size_t, total, first->iov_len);
1704	if (len) {
1705		memcpy(dst, first->iov_base, len);
1706		dst += len;
1707		remaining -= len;
1708	}
1709
1710	if (remaining) {
1711		len = min_t(size_t, remaining, PAGE_SIZE);
1712		memcpy(dst, p, len);
1713		dst += len;
1714	}
1715
1716	*dst = '\0';
1717
1718	/* Sanity check: Linux doesn't allow the pathname argument to
1719	 * contain a NUL byte.
1720	 */
1721	if (strlen(result) != total) {
1722		kfree(result);
1723		return ERR_PTR(-EINVAL);
1724	}
1725	return result;
1726}
1727EXPORT_SYMBOL_GPL(svc_fill_symlink_pathname);