Linux Audio

Check our new training course

Loading...
v5.4
   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", 20);
  92	case SVC_POOL_GLOBAL:
  93		return strlcpy(buf, "global", 20);
  94	case SVC_POOL_PERCPU:
  95		return strlcpy(buf, "percpu", 20);
  96	case SVC_POOL_PERNODE:
  97		return strlcpy(buf, "pernode", 20);
  98	default:
  99		return sprintf(buf, "%d", *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	return error;
 995}
 996
 997int svc_rpcbind_set_version(struct net *net,
 998			    const struct svc_program *progp,
 999			    u32 version, int family,
1000			    unsigned short proto,
1001			    unsigned short port)
1002{
1003	dprintk("svc: svc_register(%sv%d, %s, %u, %u)\n",
1004		progp->pg_name, version,
1005		proto == IPPROTO_UDP?  "udp" : "tcp",
1006		port, family);
1007
1008	return __svc_register(net, progp->pg_name, progp->pg_prog,
1009				version, family, proto, port);
1010
1011}
1012EXPORT_SYMBOL_GPL(svc_rpcbind_set_version);
1013
1014int svc_generic_rpcbind_set(struct net *net,
1015			    const struct svc_program *progp,
1016			    u32 version, int family,
1017			    unsigned short proto,
1018			    unsigned short port)
1019{
1020	const struct svc_version *vers = progp->pg_vers[version];
1021	int error;
1022
1023	if (vers == NULL)
1024		return 0;
1025
1026	if (vers->vs_hidden) {
1027		dprintk("svc: svc_register(%sv%d, %s, %u, %u)"
1028			" (but not telling portmap)\n",
1029			progp->pg_name, version,
1030			proto == IPPROTO_UDP?  "udp" : "tcp",
1031			port, family);
1032		return 0;
1033	}
1034
1035	/*
1036	 * Don't register a UDP port if we need congestion
1037	 * control.
1038	 */
1039	if (vers->vs_need_cong_ctrl && proto == IPPROTO_UDP)
1040		return 0;
1041
1042	error = svc_rpcbind_set_version(net, progp, version,
1043					family, proto, port);
1044
1045	return (vers->vs_rpcb_optnl) ? 0 : error;
1046}
1047EXPORT_SYMBOL_GPL(svc_generic_rpcbind_set);
1048
1049/**
1050 * svc_register - register an RPC service with the local portmapper
1051 * @serv: svc_serv struct for the service to register
1052 * @net: net namespace for the service to register
1053 * @family: protocol family of service's listener socket
1054 * @proto: transport protocol number to advertise
1055 * @port: port to advertise
1056 *
1057 * Service is registered for any address in the passed-in protocol family
1058 */
1059int svc_register(const struct svc_serv *serv, struct net *net,
1060		 const int family, const unsigned short proto,
1061		 const unsigned short port)
1062{
1063	struct svc_program	*progp;
1064	unsigned int		i;
1065	int			error = 0;
1066
1067	WARN_ON_ONCE(proto == 0 && port == 0);
1068	if (proto == 0 && port == 0)
1069		return -EINVAL;
1070
1071	for (progp = serv->sv_program; progp; progp = progp->pg_next) {
1072		for (i = 0; i < progp->pg_nvers; i++) {
1073
1074			error = progp->pg_rpcbind_set(net, progp, i,
1075					family, proto, port);
1076			if (error < 0) {
1077				printk(KERN_WARNING "svc: failed to register "
1078					"%sv%u RPC service (errno %d).\n",
1079					progp->pg_name, i, -error);
1080				break;
1081			}
1082		}
1083	}
1084
1085	return error;
1086}
1087
1088/*
1089 * If user space is running rpcbind, it should take the v4 UNSET
1090 * and clear everything for this [program, version].  If user space
1091 * is running portmap, it will reject the v4 UNSET, but won't have
1092 * any "inet6" entries anyway.  So a PMAP_UNSET should be sufficient
1093 * in this case to clear all existing entries for [program, version].
1094 */
1095static void __svc_unregister(struct net *net, const u32 program, const u32 version,
1096			     const char *progname)
1097{
1098	int error;
1099
1100	error = rpcb_v4_register(net, program, version, NULL, "");
1101
1102	/*
1103	 * User space didn't support rpcbind v4, so retry this
1104	 * request with the legacy rpcbind v2 protocol.
1105	 */
1106	if (error == -EPROTONOSUPPORT)
1107		error = rpcb_register(net, program, version, 0, 0);
1108
1109	dprintk("svc: %s(%sv%u), error %d\n",
1110			__func__, progname, version, error);
1111}
1112
1113/*
1114 * All netids, bind addresses and ports registered for [program, version]
1115 * are removed from the local rpcbind database (if the service is not
1116 * hidden) to make way for a new instance of the service.
1117 *
1118 * The result of unregistration is reported via dprintk for those who want
1119 * verification of the result, but is otherwise not important.
1120 */
1121static void svc_unregister(const struct svc_serv *serv, struct net *net)
1122{
 
1123	struct svc_program *progp;
1124	unsigned long flags;
1125	unsigned int i;
1126
1127	clear_thread_flag(TIF_SIGPENDING);
1128
1129	for (progp = serv->sv_program; progp; progp = progp->pg_next) {
1130		for (i = 0; i < progp->pg_nvers; i++) {
1131			if (progp->pg_vers[i] == NULL)
1132				continue;
1133			if (progp->pg_vers[i]->vs_hidden)
1134				continue;
1135
1136			dprintk("svc: attempting to unregister %sv%u\n",
1137				progp->pg_name, i);
1138			__svc_unregister(net, progp->pg_prog, i, progp->pg_name);
1139		}
1140	}
1141
1142	spin_lock_irqsave(&current->sighand->siglock, flags);
 
 
1143	recalc_sigpending();
1144	spin_unlock_irqrestore(&current->sighand->siglock, flags);
 
1145}
1146
1147/*
1148 * dprintk the given error with the address of the client that caused it.
1149 */
1150#if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
1151static __printf(2, 3)
1152void svc_printk(struct svc_rqst *rqstp, const char *fmt, ...)
1153{
1154	struct va_format vaf;
1155	va_list args;
1156	char 	buf[RPC_MAX_ADDRBUFLEN];
1157
1158	va_start(args, fmt);
1159
1160	vaf.fmt = fmt;
1161	vaf.va = &args;
1162
1163	dprintk("svc: %s: %pV", svc_print_addr(rqstp, buf, sizeof(buf)), &vaf);
1164
1165	va_end(args);
1166}
1167#else
1168static __printf(2,3) void svc_printk(struct svc_rqst *rqstp, const char *fmt, ...) {}
1169#endif
1170
1171__be32
1172svc_return_autherr(struct svc_rqst *rqstp, __be32 auth_err)
1173{
1174	set_bit(RQ_AUTHERR, &rqstp->rq_flags);
1175	return auth_err;
1176}
1177EXPORT_SYMBOL_GPL(svc_return_autherr);
1178
1179static __be32
1180svc_get_autherr(struct svc_rqst *rqstp, __be32 *statp)
1181{
1182	if (test_and_clear_bit(RQ_AUTHERR, &rqstp->rq_flags))
1183		return *statp;
1184	return rpc_auth_ok;
1185}
1186
1187static int
1188svc_generic_dispatch(struct svc_rqst *rqstp, __be32 *statp)
1189{
1190	struct kvec *argv = &rqstp->rq_arg.head[0];
1191	struct kvec *resv = &rqstp->rq_res.head[0];
1192	const struct svc_procedure *procp = rqstp->rq_procinfo;
1193
1194	/*
1195	 * Decode arguments
1196	 * XXX: why do we ignore the return value?
1197	 */
1198	if (procp->pc_decode &&
1199	    !procp->pc_decode(rqstp, argv->iov_base)) {
1200		*statp = rpc_garbage_args;
1201		return 1;
1202	}
1203
1204	*statp = procp->pc_func(rqstp);
1205
1206	if (*statp == rpc_drop_reply ||
1207	    test_bit(RQ_DROPME, &rqstp->rq_flags))
1208		return 0;
1209
1210	if (test_bit(RQ_AUTHERR, &rqstp->rq_flags))
1211		return 1;
1212
1213	if (*statp != rpc_success)
1214		return 1;
1215
1216	/* Encode reply */
1217	if (procp->pc_encode &&
1218	    !procp->pc_encode(rqstp, resv->iov_base + resv->iov_len)) {
1219		dprintk("svc: failed to encode reply\n");
1220		/* serv->sv_stats->rpcsystemerr++; */
1221		*statp = rpc_system_err;
1222	}
1223	return 1;
1224}
1225
1226__be32
1227svc_generic_init_request(struct svc_rqst *rqstp,
1228		const struct svc_program *progp,
1229		struct svc_process_info *ret)
1230{
1231	const struct svc_version *versp = NULL;	/* compiler food */
1232	const struct svc_procedure *procp = NULL;
1233
1234	if (rqstp->rq_vers >= progp->pg_nvers )
1235		goto err_bad_vers;
1236	versp = progp->pg_vers[rqstp->rq_vers];
1237	if (!versp)
1238		goto err_bad_vers;
1239
1240	/*
1241	 * Some protocol versions (namely NFSv4) require some form of
1242	 * congestion control.  (See RFC 7530 section 3.1 paragraph 2)
1243	 * In other words, UDP is not allowed. We mark those when setting
1244	 * up the svc_xprt, and verify that here.
1245	 *
1246	 * The spec is not very clear about what error should be returned
1247	 * when someone tries to access a server that is listening on UDP
1248	 * for lower versions. RPC_PROG_MISMATCH seems to be the closest
1249	 * fit.
1250	 */
1251	if (versp->vs_need_cong_ctrl && rqstp->rq_xprt &&
1252	    !test_bit(XPT_CONG_CTRL, &rqstp->rq_xprt->xpt_flags))
1253		goto err_bad_vers;
1254
1255	if (rqstp->rq_proc >= versp->vs_nproc)
1256		goto err_bad_proc;
1257	rqstp->rq_procinfo = procp = &versp->vs_proc[rqstp->rq_proc];
1258	if (!procp)
1259		goto err_bad_proc;
1260
1261	/* Initialize storage for argp and resp */
1262	memset(rqstp->rq_argp, 0, procp->pc_argsize);
1263	memset(rqstp->rq_resp, 0, procp->pc_ressize);
1264
1265	/* Bump per-procedure stats counter */
1266	versp->vs_count[rqstp->rq_proc]++;
1267
1268	ret->dispatch = versp->vs_dispatch;
1269	return rpc_success;
1270err_bad_vers:
1271	ret->mismatch.lovers = progp->pg_lovers;
1272	ret->mismatch.hivers = progp->pg_hivers;
1273	return rpc_prog_mismatch;
1274err_bad_proc:
1275	return rpc_proc_unavail;
1276}
1277EXPORT_SYMBOL_GPL(svc_generic_init_request);
1278
1279/*
1280 * Common routine for processing the RPC request.
1281 */
1282static int
1283svc_process_common(struct svc_rqst *rqstp, struct kvec *argv, struct kvec *resv)
1284{
 
1285	struct svc_program	*progp;
1286	const struct svc_procedure *procp = NULL;
1287	struct svc_serv		*serv = rqstp->rq_server;
1288	struct svc_process_info process;
1289	__be32			*statp;
1290	u32			prog, vers;
1291	__be32			auth_stat, rpc_stat;
1292	int			auth_res;
1293	__be32			*reply_statp;
1294
1295	rpc_stat = rpc_success;
1296
1297	if (argv->iov_len < 6*4)
1298		goto err_short_len;
1299
1300	/* Will be turned off by GSS integrity and privacy services */
1301	set_bit(RQ_SPLICE_OK, &rqstp->rq_flags);
1302	/* Will be turned off only when NFSv4 Sessions are used */
1303	set_bit(RQ_USEDEFERRAL, &rqstp->rq_flags);
1304	clear_bit(RQ_DROPME, &rqstp->rq_flags);
1305
1306	svc_putu32(resv, rqstp->rq_xid);
 
 
 
1307
1308	vers = svc_getnl(argv);
1309
1310	/* First words of reply: */
1311	svc_putnl(resv, 1);		/* REPLY */
1312
1313	if (vers != 2)		/* RPC version number */
1314		goto err_bad_rpc;
1315
1316	/* Save position in case we later decide to reject: */
1317	reply_statp = resv->iov_base + resv->iov_len;
1318
1319	svc_putnl(resv, 0);		/* ACCEPT */
1320
1321	rqstp->rq_prog = prog = svc_getnl(argv);	/* program number */
1322	rqstp->rq_vers = svc_getnl(argv);	/* version number */
1323	rqstp->rq_proc = svc_getnl(argv);	/* procedure number */
1324
1325	for (progp = serv->sv_program; progp; progp = progp->pg_next)
1326		if (prog == progp->pg_prog)
1327			break;
1328
1329	/*
1330	 * Decode auth data, and add verifier to reply buffer.
1331	 * We do this before anything else in order to get a decent
1332	 * auth verifier.
1333	 */
1334	auth_res = svc_authenticate(rqstp, &auth_stat);
1335	/* Also give the program a chance to reject this call: */
1336	if (auth_res == SVC_OK && progp) {
1337		auth_stat = rpc_autherr_badcred;
1338		auth_res = progp->pg_authenticate(rqstp);
1339	}
1340	switch (auth_res) {
1341	case SVC_OK:
1342		break;
1343	case SVC_GARBAGE:
1344		goto err_garbage;
1345	case SVC_SYSERR:
1346		rpc_stat = rpc_system_err;
1347		goto err_bad;
1348	case SVC_DENIED:
1349		goto err_bad_auth;
1350	case SVC_CLOSE:
1351		goto close;
1352	case SVC_DROP:
1353		goto dropit;
1354	case SVC_COMPLETE:
1355		goto sendit;
 
 
 
1356	}
1357
1358	if (progp == NULL)
1359		goto err_bad_prog;
1360
1361	rpc_stat = progp->pg_init_request(rqstp, progp, &process);
1362	switch (rpc_stat) {
1363	case rpc_success:
1364		break;
1365	case rpc_prog_unavail:
1366		goto err_bad_prog;
1367	case rpc_prog_mismatch:
1368		goto err_bad_vers;
1369	case rpc_proc_unavail:
1370		goto err_bad_proc;
1371	}
1372
1373	procp = rqstp->rq_procinfo;
1374	/* Should this check go into the dispatcher? */
1375	if (!procp || !procp->pc_func)
1376		goto err_bad_proc;
1377
1378	/* Syntactic check complete */
1379	serv->sv_stats->rpccnt++;
1380	trace_svc_process(rqstp, progp->pg_name);
1381
1382	/* Build the reply header. */
1383	statp = resv->iov_base +resv->iov_len;
1384	svc_putnl(resv, RPC_SUCCESS);
1385
1386	/* un-reserve some of the out-queue now that we have a
1387	 * better idea of reply size
1388	 */
1389	if (procp->pc_xdrressize)
1390		svc_reserve_auth(rqstp, procp->pc_xdrressize<<2);
1391
1392	/* Call the function that processes the request. */
1393	if (!process.dispatch) {
1394		if (!svc_generic_dispatch(rqstp, statp))
1395			goto release_dropit;
1396		if (*statp == rpc_garbage_args)
1397			goto err_garbage;
1398		auth_stat = svc_get_autherr(rqstp, statp);
1399		if (auth_stat != rpc_auth_ok)
1400			goto err_release_bad_auth;
1401	} else {
1402		dprintk("svc: calling dispatcher\n");
1403		if (!process.dispatch(rqstp, statp))
1404			goto release_dropit; /* Release reply info */
1405	}
1406
1407	/* Check RPC status result */
1408	if (*statp != rpc_success)
1409		resv->iov_len = ((void*)statp)  - resv->iov_base + 4;
1410
1411	/* Release reply info */
1412	if (procp->pc_release)
1413		procp->pc_release(rqstp);
 
 
 
 
 
 
 
 
 
1414
1415	if (procp->pc_encode == NULL)
1416		goto dropit;
1417
1418 sendit:
1419	if (svc_authorise(rqstp))
1420		goto close;
1421	return 1;		/* Caller can now send it */
1422
1423release_dropit:
1424	if (procp->pc_release)
1425		procp->pc_release(rqstp);
1426 dropit:
1427	svc_authorise(rqstp);	/* doesn't hurt to call this twice */
1428	dprintk("svc: svc_process dropit\n");
1429	return 0;
1430
1431 close:
 
 
1432	if (rqstp->rq_xprt && test_bit(XPT_TEMP, &rqstp->rq_xprt->xpt_flags))
1433		svc_close_xprt(rqstp->rq_xprt);
1434	dprintk("svc: svc_process close\n");
1435	return 0;
1436
1437err_short_len:
1438	svc_printk(rqstp, "short len %zd, dropping request\n",
1439			argv->iov_len);
1440	goto close;
1441
1442err_bad_rpc:
1443	serv->sv_stats->rpcbadfmt++;
1444	svc_putnl(resv, 1);	/* REJECT */
1445	svc_putnl(resv, 0);	/* RPC_MISMATCH */
1446	svc_putnl(resv, 2);	/* Only RPCv2 supported */
1447	svc_putnl(resv, 2);
1448	goto sendit;
 
1449
1450err_release_bad_auth:
1451	if (procp->pc_release)
1452		procp->pc_release(rqstp);
1453err_bad_auth:
1454	dprintk("svc: authentication failed (%d)\n", ntohl(auth_stat));
 
1455	serv->sv_stats->rpcbadauth++;
1456	/* Restore write pointer to location of accept status: */
1457	xdr_ressize_check(rqstp, reply_statp);
1458	svc_putnl(resv, 1);	/* REJECT */
1459	svc_putnl(resv, 1);	/* AUTH_ERROR */
1460	svc_putnl(resv, ntohl(auth_stat));	/* status */
1461	goto sendit;
1462
1463err_bad_prog:
1464	dprintk("svc: unknown program %d\n", prog);
1465	serv->sv_stats->rpcbadfmt++;
1466	svc_putnl(resv, RPC_PROG_UNAVAIL);
1467	goto sendit;
1468
1469err_bad_vers:
1470	svc_printk(rqstp, "unknown version (%d for prog %d, %s)\n",
1471		       rqstp->rq_vers, rqstp->rq_prog, progp->pg_name);
1472
1473	serv->sv_stats->rpcbadfmt++;
1474	svc_putnl(resv, RPC_PROG_MISMATCH);
1475	svc_putnl(resv, process.mismatch.lovers);
1476	svc_putnl(resv, process.mismatch.hivers);
 
 
 
 
 
1477	goto sendit;
1478
1479err_bad_proc:
1480	svc_printk(rqstp, "unknown procedure (%d)\n", rqstp->rq_proc);
1481
1482	serv->sv_stats->rpcbadfmt++;
1483	svc_putnl(resv, RPC_PROC_UNAVAIL);
1484	goto sendit;
1485
1486err_garbage:
1487	svc_printk(rqstp, "failed to decode args\n");
1488
1489	rpc_stat = rpc_garbage_args;
1490err_bad:
1491	serv->sv_stats->rpcbadfmt++;
1492	svc_putnl(resv, ntohl(rpc_stat));
 
 
 
 
 
1493	goto sendit;
1494}
1495
1496/*
1497 * Process the RPC request.
 
 
1498 */
1499int
1500svc_process(struct svc_rqst *rqstp)
1501{
1502	struct kvec		*argv = &rqstp->rq_arg.head[0];
1503	struct kvec		*resv = &rqstp->rq_res.head[0];
1504	struct svc_serv		*serv = rqstp->rq_server;
1505	u32			dir;
 
 
 
 
 
1506
1507	/*
1508	 * Setup response xdr_buf.
1509	 * Initially it has just one page
1510	 */
1511	rqstp->rq_next_page = &rqstp->rq_respages[1];
1512	resv->iov_base = page_address(rqstp->rq_respages[0]);
1513	resv->iov_len = 0;
1514	rqstp->rq_res.pages = rqstp->rq_respages + 1;
1515	rqstp->rq_res.len = 0;
1516	rqstp->rq_res.page_base = 0;
1517	rqstp->rq_res.page_len = 0;
1518	rqstp->rq_res.buflen = PAGE_SIZE;
1519	rqstp->rq_res.tail[0].iov_base = NULL;
1520	rqstp->rq_res.tail[0].iov_len = 0;
1521
1522	dir  = svc_getnl(argv);
1523	if (dir != 0) {
1524		/* direction != CALL */
1525		svc_printk(rqstp, "bad direction %d, dropping request\n", dir);
1526		serv->sv_stats->rpcbadfmt++;
1527		goto out_drop;
1528	}
 
 
1529
1530	/* Reserve space for the record marker */
1531	if (rqstp->rq_prot == IPPROTO_TCP)
1532		svc_putnl(resv, 0);
1533
1534	/* Returns 1 for send, 0 for drop */
1535	if (likely(svc_process_common(rqstp, argv, resv)))
1536		return svc_send(rqstp);
1537
 
 
 
 
1538out_drop:
1539	svc_drop(rqstp);
1540	return 0;
1541}
1542EXPORT_SYMBOL_GPL(svc_process);
1543
1544#if defined(CONFIG_SUNRPC_BACKCHANNEL)
1545/*
1546 * Process a backchannel RPC request that arrived over an existing
1547 * outbound connection
 
 
1548 */
1549int
1550bc_svc_process(struct svc_serv *serv, struct rpc_rqst *req,
1551	       struct svc_rqst *rqstp)
1552{
1553	struct kvec	*argv = &rqstp->rq_arg.head[0];
1554	struct kvec	*resv = &rqstp->rq_res.head[0];
1555	struct rpc_task *task;
1556	int proc_error;
1557	int error;
1558
1559	dprintk("svc: %s(%p)\n", __func__, req);
1560
1561	/* Build the svc_rqst used by the common processing routine */
1562	rqstp->rq_xid = req->rq_xid;
1563	rqstp->rq_prot = req->rq_xprt->prot;
1564	rqstp->rq_server = serv;
1565	rqstp->rq_bc_net = req->rq_xprt->xprt_net;
1566
1567	rqstp->rq_addrlen = sizeof(req->rq_xprt->addr);
1568	memcpy(&rqstp->rq_addr, &req->rq_xprt->addr, rqstp->rq_addrlen);
1569	memcpy(&rqstp->rq_arg, &req->rq_rcv_buf, sizeof(rqstp->rq_arg));
1570	memcpy(&rqstp->rq_res, &req->rq_snd_buf, sizeof(rqstp->rq_res));
1571
1572	/* Adjust the argument buffer length */
1573	rqstp->rq_arg.len = req->rq_private_buf.len;
1574	if (rqstp->rq_arg.len <= rqstp->rq_arg.head[0].iov_len) {
1575		rqstp->rq_arg.head[0].iov_len = rqstp->rq_arg.len;
1576		rqstp->rq_arg.page_len = 0;
1577	} else if (rqstp->rq_arg.len <= rqstp->rq_arg.head[0].iov_len +
1578			rqstp->rq_arg.page_len)
1579		rqstp->rq_arg.page_len = rqstp->rq_arg.len -
1580			rqstp->rq_arg.head[0].iov_len;
1581	else
1582		rqstp->rq_arg.len = rqstp->rq_arg.head[0].iov_len +
1583			rqstp->rq_arg.page_len;
1584
1585	/* reset result send buffer "put" position */
1586	resv->iov_len = 0;
1587
1588	/*
1589	 * Skip the next two words because they've already been
1590	 * processed in the transport
1591	 */
1592	svc_getu32(argv);	/* XID */
1593	svc_getnl(argv);	/* CALLDIR */
 
1594
1595	/* Parse and execute the bc call */
1596	proc_error = svc_process_common(rqstp, argv, resv);
1597
1598	atomic_dec(&req->rq_xprt->bc_slot_count);
1599	if (!proc_error) {
1600		/* Processing error: drop the request */
1601		xprt_free_bc_request(req);
1602		error = -EINVAL;
1603		goto out;
1604	}
1605	/* Finally, send the reply synchronously */
1606	memcpy(&req->rq_snd_buf, &rqstp->rq_res, sizeof(req->rq_snd_buf));
1607	task = rpc_run_bc_task(req);
1608	if (IS_ERR(task)) {
1609		error = PTR_ERR(task);
1610		goto out;
 
1611	}
 
 
 
 
 
1612
1613	WARN_ON_ONCE(atomic_read(&task->tk_count) != 1);
1614	error = task->tk_status;
1615	rpc_put_task(task);
1616
1617out:
1618	dprintk("svc: %s(), error=%d\n", __func__, error);
1619	return error;
1620}
1621EXPORT_SYMBOL_GPL(bc_svc_process);
1622#endif /* CONFIG_SUNRPC_BACKCHANNEL */
1623
1624/*
1625 * Return (transport-specific) limit on the rpc payload.
 
 
 
 
1626 */
1627u32 svc_max_payload(const struct svc_rqst *rqstp)
1628{
1629	u32 max = rqstp->rq_xprt->xpt_class->xcl_max_payload;
1630
1631	if (rqstp->rq_server->sv_max_payload < max)
1632		max = rqstp->rq_server->sv_max_payload;
1633	return max;
1634}
1635EXPORT_SYMBOL_GPL(svc_max_payload);
1636
1637/**
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1638 * svc_fill_write_vector - Construct data argument for VFS write call
1639 * @rqstp: svc_rqst to operate on
1640 * @pages: list of pages containing data payload
1641 * @first: buffer containing first section of write payload
1642 * @total: total number of bytes of write payload
1643 *
1644 * Fills in rqstp::rq_vec, and returns the number of elements.
1645 */
1646unsigned int svc_fill_write_vector(struct svc_rqst *rqstp, struct page **pages,
1647				   struct kvec *first, size_t total)
1648{
 
 
1649	struct kvec *vec = rqstp->rq_vec;
 
1650	unsigned int i;
1651
1652	/* Some types of transport can present the write payload
1653	 * entirely in rq_arg.pages. In this case, @first is empty.
1654	 */
1655	i = 0;
1656	if (first->iov_len) {
1657		vec[i].iov_base = first->iov_base;
1658		vec[i].iov_len = min_t(size_t, total, first->iov_len);
1659		total -= vec[i].iov_len;
1660		++i;
1661	}
1662
1663	while (total) {
1664		vec[i].iov_base = page_address(*pages);
1665		vec[i].iov_len = min_t(size_t, total, PAGE_SIZE);
1666		total -= vec[i].iov_len;
1667		++i;
1668		++pages;
1669	}
1670
1671	WARN_ON_ONCE(i > ARRAY_SIZE(rqstp->rq_vec));
1672	return i;
1673}
1674EXPORT_SYMBOL_GPL(svc_fill_write_vector);
1675
1676/**
1677 * svc_fill_symlink_pathname - Construct pathname argument for VFS symlink call
1678 * @rqstp: svc_rqst to operate on
1679 * @first: buffer containing first section of pathname
1680 * @p: buffer containing remaining section of pathname
1681 * @total: total length of the pathname argument
1682 *
1683 * The VFS symlink API demands a NUL-terminated pathname in mapped memory.
1684 * Returns pointer to a NUL-terminated string, or an ERR_PTR. Caller must free
1685 * the returned string.
1686 */
1687char *svc_fill_symlink_pathname(struct svc_rqst *rqstp, struct kvec *first,
1688				void *p, size_t total)
1689{
1690	size_t len, remaining;
1691	char *result, *dst;
1692
1693	result = kmalloc(total + 1, GFP_KERNEL);
1694	if (!result)
1695		return ERR_PTR(-ESERVERFAULT);
1696
1697	dst = result;
1698	remaining = total;
1699
1700	len = min_t(size_t, total, first->iov_len);
1701	if (len) {
1702		memcpy(dst, first->iov_base, len);
1703		dst += len;
1704		remaining -= len;
1705	}
1706
1707	if (remaining) {
1708		len = min_t(size_t, remaining, PAGE_SIZE);
1709		memcpy(dst, p, len);
1710		dst += len;
1711	}
1712
1713	*dst = '\0';
1714
1715	/* Sanity check: Linux doesn't allow the pathname argument to
1716	 * contain a NUL byte.
1717	 */
1718	if (strlen(result) != total) {
1719		kfree(result);
1720		return ERR_PTR(-EINVAL);
1721	}
1722	return result;
1723}
1724EXPORT_SYMBOL_GPL(svc_fill_symlink_pathname);
v6.8
   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#include "fail.h"
  35
  36#define RPCDBG_FACILITY	RPCDBG_SVCDSP
  37
  38static void svc_unregister(const struct svc_serv *serv, struct net *net);
  39
 
 
  40#define SVC_POOL_DEFAULT	SVC_POOL_GLOBAL
  41
  42/*
  43 * Mode for mapping cpus to pools.
  44 */
  45enum {
  46	SVC_POOL_AUTO = -1,	/* choose one of the others */
  47	SVC_POOL_GLOBAL,	/* no mapping, just a single global pool
  48				 * (legacy & UP mode) */
  49	SVC_POOL_PERCPU,	/* one pool per cpu */
  50	SVC_POOL_PERNODE	/* one pool per numa node */
  51};
  52
  53/*
  54 * Structure for mapping cpus to pools and vice versa.
  55 * Setup once during sunrpc initialisation.
  56 */
  57
  58struct svc_pool_map {
  59	int count;			/* How many svc_servs use us */
  60	int mode;			/* Note: int not enum to avoid
  61					 * warnings about "enumeration value
  62					 * not handled in switch" */
  63	unsigned int npools;
  64	unsigned int *pool_to;		/* maps pool id to cpu or node */
  65	unsigned int *to_pool;		/* maps cpu or node to pool id */
  66};
  67
  68static struct svc_pool_map svc_pool_map = {
  69	.mode = SVC_POOL_DEFAULT
  70};
 
  71
  72static DEFINE_MUTEX(svc_pool_map_mutex);/* protects svc_pool_map.count only */
  73
  74static int
  75param_set_pool_mode(const char *val, const struct kernel_param *kp)
  76{
  77	int *ip = (int *)kp->arg;
  78	struct svc_pool_map *m = &svc_pool_map;
  79	int err;
  80
  81	mutex_lock(&svc_pool_map_mutex);
  82
  83	err = -EBUSY;
  84	if (m->count)
  85		goto out;
  86
  87	err = 0;
  88	if (!strncmp(val, "auto", 4))
  89		*ip = SVC_POOL_AUTO;
  90	else if (!strncmp(val, "global", 6))
  91		*ip = SVC_POOL_GLOBAL;
  92	else if (!strncmp(val, "percpu", 6))
  93		*ip = SVC_POOL_PERCPU;
  94	else if (!strncmp(val, "pernode", 7))
  95		*ip = SVC_POOL_PERNODE;
  96	else
  97		err = -EINVAL;
  98
  99out:
 100	mutex_unlock(&svc_pool_map_mutex);
 101	return err;
 102}
 103
 104static int
 105param_get_pool_mode(char *buf, const struct kernel_param *kp)
 106{
 107	int *ip = (int *)kp->arg;
 108
 109	switch (*ip)
 110	{
 111	case SVC_POOL_AUTO:
 112		return sysfs_emit(buf, "auto\n");
 113	case SVC_POOL_GLOBAL:
 114		return sysfs_emit(buf, "global\n");
 115	case SVC_POOL_PERCPU:
 116		return sysfs_emit(buf, "percpu\n");
 117	case SVC_POOL_PERNODE:
 118		return sysfs_emit(buf, "pernode\n");
 119	default:
 120		return sysfs_emit(buf, "%d\n", *ip);
 121	}
 122}
 123
 124module_param_call(pool_mode, param_set_pool_mode, param_get_pool_mode,
 125		 &svc_pool_map.mode, 0644);
 126
 127/*
 128 * Detect best pool mapping mode heuristically,
 129 * according to the machine's topology.
 130 */
 131static int
 132svc_pool_map_choose_mode(void)
 133{
 134	unsigned int node;
 135
 136	if (nr_online_nodes > 1) {
 137		/*
 138		 * Actually have multiple NUMA nodes,
 139		 * so split pools on NUMA node boundaries
 140		 */
 141		return SVC_POOL_PERNODE;
 142	}
 143
 144	node = first_online_node;
 145	if (nr_cpus_node(node) > 2) {
 146		/*
 147		 * Non-trivial SMP, or CONFIG_NUMA on
 148		 * non-NUMA hardware, e.g. with a generic
 149		 * x86_64 kernel on Xeons.  In this case we
 150		 * want to divide the pools on cpu boundaries.
 151		 */
 152		return SVC_POOL_PERCPU;
 153	}
 154
 155	/* default: one global pool */
 156	return SVC_POOL_GLOBAL;
 157}
 158
 159/*
 160 * Allocate the to_pool[] and pool_to[] arrays.
 161 * Returns 0 on success or an errno.
 162 */
 163static int
 164svc_pool_map_alloc_arrays(struct svc_pool_map *m, unsigned int maxpools)
 165{
 166	m->to_pool = kcalloc(maxpools, sizeof(unsigned int), GFP_KERNEL);
 167	if (!m->to_pool)
 168		goto fail;
 169	m->pool_to = kcalloc(maxpools, sizeof(unsigned int), GFP_KERNEL);
 170	if (!m->pool_to)
 171		goto fail_free;
 172
 173	return 0;
 174
 175fail_free:
 176	kfree(m->to_pool);
 177	m->to_pool = NULL;
 178fail:
 179	return -ENOMEM;
 180}
 181
 182/*
 183 * Initialise the pool map for SVC_POOL_PERCPU mode.
 184 * Returns number of pools or <0 on error.
 185 */
 186static int
 187svc_pool_map_init_percpu(struct svc_pool_map *m)
 188{
 189	unsigned int maxpools = nr_cpu_ids;
 190	unsigned int pidx = 0;
 191	unsigned int cpu;
 192	int err;
 193
 194	err = svc_pool_map_alloc_arrays(m, maxpools);
 195	if (err)
 196		return err;
 197
 198	for_each_online_cpu(cpu) {
 199		BUG_ON(pidx >= maxpools);
 200		m->to_pool[cpu] = pidx;
 201		m->pool_to[pidx] = cpu;
 202		pidx++;
 203	}
 204	/* cpus brought online later all get mapped to pool0, sorry */
 205
 206	return pidx;
 207};
 208
 209
 210/*
 211 * Initialise the pool map for SVC_POOL_PERNODE mode.
 212 * Returns number of pools or <0 on error.
 213 */
 214static int
 215svc_pool_map_init_pernode(struct svc_pool_map *m)
 216{
 217	unsigned int maxpools = nr_node_ids;
 218	unsigned int pidx = 0;
 219	unsigned int node;
 220	int err;
 221
 222	err = svc_pool_map_alloc_arrays(m, maxpools);
 223	if (err)
 224		return err;
 225
 226	for_each_node_with_cpus(node) {
 227		/* some architectures (e.g. SN2) have cpuless nodes */
 228		BUG_ON(pidx > maxpools);
 229		m->to_pool[node] = pidx;
 230		m->pool_to[pidx] = node;
 231		pidx++;
 232	}
 233	/* nodes brought online later all get mapped to pool0, sorry */
 234
 235	return pidx;
 236}
 237
 238
 239/*
 240 * Add a reference to the global map of cpus to pools (and
 241 * vice versa) if pools are in use.
 242 * Initialise the map if we're the first user.
 243 * Returns the number of pools. If this is '1', no reference
 244 * was taken.
 245 */
 246static unsigned int
 247svc_pool_map_get(void)
 248{
 249	struct svc_pool_map *m = &svc_pool_map;
 250	int npools = -1;
 251
 252	mutex_lock(&svc_pool_map_mutex);
 253
 254	if (m->count++) {
 255		mutex_unlock(&svc_pool_map_mutex);
 256		WARN_ON_ONCE(m->npools <= 1);
 257		return m->npools;
 258	}
 259
 260	if (m->mode == SVC_POOL_AUTO)
 261		m->mode = svc_pool_map_choose_mode();
 262
 263	switch (m->mode) {
 264	case SVC_POOL_PERCPU:
 265		npools = svc_pool_map_init_percpu(m);
 266		break;
 267	case SVC_POOL_PERNODE:
 268		npools = svc_pool_map_init_pernode(m);
 269		break;
 270	}
 271
 272	if (npools <= 0) {
 273		/* default, or memory allocation failure */
 274		npools = 1;
 275		m->mode = SVC_POOL_GLOBAL;
 276	}
 277	m->npools = npools;
 278
 279	if (npools == 1)
 280		/* service is unpooled, so doesn't hold a reference */
 281		m->count--;
 282
 283	mutex_unlock(&svc_pool_map_mutex);
 284	return npools;
 285}
 
 286
 287/*
 288 * Drop a reference to the global map of cpus to pools, if
 289 * pools were in use, i.e. if npools > 1.
 290 * When the last reference is dropped, the map data is
 291 * freed; this allows the sysadmin to change the pool
 292 * mode using the pool_mode module option without
 293 * rebooting or re-loading sunrpc.ko.
 294 */
 295static void
 296svc_pool_map_put(int npools)
 297{
 298	struct svc_pool_map *m = &svc_pool_map;
 299
 300	if (npools <= 1)
 301		return;
 302	mutex_lock(&svc_pool_map_mutex);
 303
 304	if (!--m->count) {
 305		kfree(m->to_pool);
 306		m->to_pool = NULL;
 307		kfree(m->pool_to);
 308		m->pool_to = NULL;
 309		m->npools = 0;
 310	}
 311
 312	mutex_unlock(&svc_pool_map_mutex);
 313}
 
 314
 315static int svc_pool_map_get_node(unsigned int pidx)
 316{
 317	const struct svc_pool_map *m = &svc_pool_map;
 318
 319	if (m->count) {
 320		if (m->mode == SVC_POOL_PERCPU)
 321			return cpu_to_node(m->pool_to[pidx]);
 322		if (m->mode == SVC_POOL_PERNODE)
 323			return m->pool_to[pidx];
 324	}
 325	return NUMA_NO_NODE;
 326}
 327/*
 328 * Set the given thread's cpus_allowed mask so that it
 329 * will only run on cpus in the given pool.
 330 */
 331static inline void
 332svc_pool_map_set_cpumask(struct task_struct *task, unsigned int pidx)
 333{
 334	struct svc_pool_map *m = &svc_pool_map;
 335	unsigned int node = m->pool_to[pidx];
 336
 337	/*
 338	 * The caller checks for sv_nrpools > 1, which
 339	 * implies that we've been initialized.
 340	 */
 341	WARN_ON_ONCE(m->count == 0);
 342	if (m->count == 0)
 343		return;
 344
 345	switch (m->mode) {
 346	case SVC_POOL_PERCPU:
 347	{
 348		set_cpus_allowed_ptr(task, cpumask_of(node));
 349		break;
 350	}
 351	case SVC_POOL_PERNODE:
 352	{
 353		set_cpus_allowed_ptr(task, cpumask_of_node(node));
 354		break;
 355	}
 356	}
 357}
 358
 359/**
 360 * svc_pool_for_cpu - Select pool to run a thread on this cpu
 361 * @serv: An RPC service
 362 *
 363 * Use the active CPU and the svc_pool_map's mode setting to
 364 * select the svc thread pool to use. Once initialized, the
 365 * svc_pool_map does not change.
 366 *
 367 * Return value:
 368 *   A pointer to an svc_pool
 369 */
 370struct svc_pool *svc_pool_for_cpu(struct svc_serv *serv)
 
 371{
 372	struct svc_pool_map *m = &svc_pool_map;
 373	int cpu = raw_smp_processor_id();
 374	unsigned int pidx = 0;
 375
 376	if (serv->sv_nrpools <= 1)
 377		return serv->sv_pools;
 378
 379	switch (m->mode) {
 380	case SVC_POOL_PERCPU:
 381		pidx = m->to_pool[cpu];
 382		break;
 383	case SVC_POOL_PERNODE:
 384		pidx = m->to_pool[cpu_to_node(cpu)];
 385		break;
 
 
 
 
 386	}
 387
 388	return &serv->sv_pools[pidx % serv->sv_nrpools];
 389}
 390
 391int svc_rpcb_setup(struct svc_serv *serv, struct net *net)
 392{
 393	int err;
 394
 395	err = rpcb_create_local(net);
 396	if (err)
 397		return err;
 398
 399	/* Remove any stale portmap registrations */
 400	svc_unregister(serv, net);
 401	return 0;
 402}
 403EXPORT_SYMBOL_GPL(svc_rpcb_setup);
 404
 405void svc_rpcb_cleanup(struct svc_serv *serv, struct net *net)
 406{
 407	svc_unregister(serv, net);
 408	rpcb_put_local(net);
 409}
 410EXPORT_SYMBOL_GPL(svc_rpcb_cleanup);
 411
 412static int svc_uses_rpcbind(struct svc_serv *serv)
 413{
 414	struct svc_program	*progp;
 415	unsigned int		i;
 416
 417	for (progp = serv->sv_program; progp; progp = progp->pg_next) {
 418		for (i = 0; i < progp->pg_nvers; i++) {
 419			if (progp->pg_vers[i] == NULL)
 420				continue;
 421			if (!progp->pg_vers[i]->vs_hidden)
 422				return 1;
 423		}
 424	}
 425
 426	return 0;
 427}
 428
 429int svc_bind(struct svc_serv *serv, struct net *net)
 430{
 431	if (!svc_uses_rpcbind(serv))
 432		return 0;
 433	return svc_rpcb_setup(serv, net);
 434}
 435EXPORT_SYMBOL_GPL(svc_bind);
 436
 437#if defined(CONFIG_SUNRPC_BACKCHANNEL)
 438static void
 439__svc_init_bc(struct svc_serv *serv)
 440{
 441	lwq_init(&serv->sv_cb_list);
 
 
 442}
 443#else
 444static void
 445__svc_init_bc(struct svc_serv *serv)
 446{
 447}
 448#endif
 449
 450/*
 451 * Create an RPC service
 452 */
 453static struct svc_serv *
 454__svc_create(struct svc_program *prog, unsigned int bufsize, int npools,
 455	     int (*threadfn)(void *data))
 456{
 457	struct svc_serv	*serv;
 458	unsigned int vers;
 459	unsigned int xdrsize;
 460	unsigned int i;
 461
 462	if (!(serv = kzalloc(sizeof(*serv), GFP_KERNEL)))
 463		return NULL;
 464	serv->sv_name      = prog->pg_name;
 465	serv->sv_program   = prog;
 
 466	serv->sv_stats     = prog->pg_stats;
 467	if (bufsize > RPCSVC_MAXPAYLOAD)
 468		bufsize = RPCSVC_MAXPAYLOAD;
 469	serv->sv_max_payload = bufsize? bufsize : 4096;
 470	serv->sv_max_mesg  = roundup(serv->sv_max_payload + PAGE_SIZE, PAGE_SIZE);
 471	serv->sv_threadfn = threadfn;
 472	xdrsize = 0;
 473	while (prog) {
 474		prog->pg_lovers = prog->pg_nvers-1;
 475		for (vers=0; vers<prog->pg_nvers ; vers++)
 476			if (prog->pg_vers[vers]) {
 477				prog->pg_hivers = vers;
 478				if (prog->pg_lovers > vers)
 479					prog->pg_lovers = vers;
 480				if (prog->pg_vers[vers]->vs_xdrsize > xdrsize)
 481					xdrsize = prog->pg_vers[vers]->vs_xdrsize;
 482			}
 483		prog = prog->pg_next;
 484	}
 485	serv->sv_xdrsize   = xdrsize;
 486	INIT_LIST_HEAD(&serv->sv_tempsocks);
 487	INIT_LIST_HEAD(&serv->sv_permsocks);
 488	timer_setup(&serv->sv_temptimer, NULL, 0);
 489	spin_lock_init(&serv->sv_lock);
 490
 491	__svc_init_bc(serv);
 492
 493	serv->sv_nrpools = npools;
 494	serv->sv_pools =
 495		kcalloc(serv->sv_nrpools, sizeof(struct svc_pool),
 496			GFP_KERNEL);
 497	if (!serv->sv_pools) {
 498		kfree(serv);
 499		return NULL;
 500	}
 501
 502	for (i = 0; i < serv->sv_nrpools; i++) {
 503		struct svc_pool *pool = &serv->sv_pools[i];
 504
 505		dprintk("svc: initialising pool %u for %s\n",
 506				i, serv->sv_name);
 507
 508		pool->sp_id = i;
 509		lwq_init(&pool->sp_xprts);
 510		INIT_LIST_HEAD(&pool->sp_all_threads);
 511		init_llist_head(&pool->sp_idle_threads);
 512
 513		percpu_counter_init(&pool->sp_messages_arrived, 0, GFP_KERNEL);
 514		percpu_counter_init(&pool->sp_sockets_queued, 0, GFP_KERNEL);
 515		percpu_counter_init(&pool->sp_threads_woken, 0, GFP_KERNEL);
 516	}
 517
 518	return serv;
 519}
 520
 521/**
 522 * svc_create - Create an RPC service
 523 * @prog: the RPC program the new service will handle
 524 * @bufsize: maximum message size for @prog
 525 * @threadfn: a function to service RPC requests for @prog
 526 *
 527 * Returns an instantiated struct svc_serv object or NULL.
 528 */
 529struct svc_serv *svc_create(struct svc_program *prog, unsigned int bufsize,
 530			    int (*threadfn)(void *data))
 531{
 532	return __svc_create(prog, bufsize, 1, threadfn);
 533}
 534EXPORT_SYMBOL_GPL(svc_create);
 535
 536/**
 537 * svc_create_pooled - Create an RPC service with pooled threads
 538 * @prog: the RPC program the new service will handle
 539 * @bufsize: maximum message size for @prog
 540 * @threadfn: a function to service RPC requests for @prog
 541 *
 542 * Returns an instantiated struct svc_serv object or NULL.
 543 */
 544struct svc_serv *svc_create_pooled(struct svc_program *prog,
 545				   unsigned int bufsize,
 546				   int (*threadfn)(void *data))
 547{
 548	struct svc_serv *serv;
 549	unsigned int npools = svc_pool_map_get();
 550
 551	serv = __svc_create(prog, bufsize, npools, threadfn);
 552	if (!serv)
 553		goto out_err;
 554	return serv;
 555out_err:
 556	svc_pool_map_put(npools);
 557	return NULL;
 558}
 559EXPORT_SYMBOL_GPL(svc_create_pooled);
 560
 
 
 
 
 
 
 
 
 
 561/*
 562 * Destroy an RPC service. Should be called with appropriate locking to
 563 * protect sv_permsocks and sv_tempsocks.
 564 */
 565void
 566svc_destroy(struct svc_serv **servp)
 567{
 568	struct svc_serv *serv = *servp;
 569	unsigned int i;
 570
 571	*servp = NULL;
 
 
 
 
 
 
 
 572
 573	dprintk("svc: svc_destroy(%s)\n", serv->sv_program->pg_name);
 574	timer_shutdown_sync(&serv->sv_temptimer);
 575
 576	/*
 577	 * Remaining transports at this point are not expected.
 
 578	 */
 579	WARN_ONCE(!list_empty(&serv->sv_permsocks),
 580		  "SVC: permsocks remain for %s\n", serv->sv_program->pg_name);
 581	WARN_ONCE(!list_empty(&serv->sv_tempsocks),
 582		  "SVC: tempsocks remain for %s\n", serv->sv_program->pg_name);
 583
 584	cache_clean_deferred(serv);
 585
 586	svc_pool_map_put(serv->sv_nrpools);
 
 587
 588	for (i = 0; i < serv->sv_nrpools; i++) {
 589		struct svc_pool *pool = &serv->sv_pools[i];
 590
 591		percpu_counter_destroy(&pool->sp_messages_arrived);
 592		percpu_counter_destroy(&pool->sp_sockets_queued);
 593		percpu_counter_destroy(&pool->sp_threads_woken);
 594	}
 595	kfree(serv->sv_pools);
 596	kfree(serv);
 597}
 598EXPORT_SYMBOL_GPL(svc_destroy);
 599
 600static bool
 
 
 
 
 601svc_init_buffer(struct svc_rqst *rqstp, unsigned int size, int node)
 602{
 603	unsigned long pages, ret;
 604
 605	/* bc_xprt uses fore channel allocated buffers */
 606	if (svc_is_backchannel(rqstp))
 607		return true;
 608
 609	pages = size / PAGE_SIZE + 1; /* extra page as we hold both request and reply.
 610				       * We assume one is at most one page
 611				       */
 
 612	WARN_ON_ONCE(pages > RPCSVC_MAXPAGES);
 613	if (pages > RPCSVC_MAXPAGES)
 614		pages = RPCSVC_MAXPAGES;
 615
 616	ret = alloc_pages_bulk_array_node(GFP_KERNEL, node, pages,
 617					  rqstp->rq_pages);
 618	return ret == pages;
 
 
 
 
 619}
 620
 621/*
 622 * Release an RPC server buffer
 623 */
 624static void
 625svc_release_buffer(struct svc_rqst *rqstp)
 626{
 627	unsigned int i;
 628
 629	for (i = 0; i < ARRAY_SIZE(rqstp->rq_pages); i++)
 630		if (rqstp->rq_pages[i])
 631			put_page(rqstp->rq_pages[i]);
 632}
 633
 634struct svc_rqst *
 635svc_rqst_alloc(struct svc_serv *serv, struct svc_pool *pool, int node)
 636{
 637	struct svc_rqst	*rqstp;
 638
 639	rqstp = kzalloc_node(sizeof(*rqstp), GFP_KERNEL, node);
 640	if (!rqstp)
 641		return rqstp;
 642
 643	folio_batch_init(&rqstp->rq_fbatch);
 644
 645	rqstp->rq_server = serv;
 646	rqstp->rq_pool = pool;
 647
 648	rqstp->rq_scratch_page = alloc_pages_node(node, GFP_KERNEL, 0);
 649	if (!rqstp->rq_scratch_page)
 650		goto out_enomem;
 651
 652	rqstp->rq_argp = kmalloc_node(serv->sv_xdrsize, GFP_KERNEL, node);
 653	if (!rqstp->rq_argp)
 654		goto out_enomem;
 655
 656	rqstp->rq_resp = kmalloc_node(serv->sv_xdrsize, GFP_KERNEL, node);
 657	if (!rqstp->rq_resp)
 658		goto out_enomem;
 659
 660	if (!svc_init_buffer(rqstp, serv->sv_max_mesg, node))
 661		goto out_enomem;
 662
 663	return rqstp;
 664out_enomem:
 665	svc_rqst_free(rqstp);
 666	return NULL;
 667}
 668EXPORT_SYMBOL_GPL(svc_rqst_alloc);
 669
 670static struct svc_rqst *
 671svc_prepare_thread(struct svc_serv *serv, struct svc_pool *pool, int node)
 672{
 673	struct svc_rqst	*rqstp;
 674
 675	rqstp = svc_rqst_alloc(serv, pool, node);
 676	if (!rqstp)
 677		return ERR_PTR(-ENOMEM);
 678
 679	spin_lock_bh(&serv->sv_lock);
 680	serv->sv_nrthreads += 1;
 681	spin_unlock_bh(&serv->sv_lock);
 682
 683	atomic_inc(&pool->sp_nrthreads);
 684
 685	/* Protected by whatever lock the service uses when calling
 686	 * svc_set_num_threads()
 687	 */
 688	list_add_rcu(&rqstp->rq_all, &pool->sp_all_threads);
 689
 690	return rqstp;
 691}
 
 692
 693/**
 694 * svc_pool_wake_idle_thread - Awaken an idle thread in @pool
 695 * @pool: service thread pool
 696 *
 697 * Can be called from soft IRQ or process context. Finding an idle
 698 * service thread and marking it BUSY is atomic with respect to
 699 * other calls to svc_pool_wake_idle_thread().
 700 *
 701 */
 702void svc_pool_wake_idle_thread(struct svc_pool *pool)
 
 703{
 704	struct svc_rqst	*rqstp;
 705	struct llist_node *ln;
 706
 707	rcu_read_lock();
 708	ln = READ_ONCE(pool->sp_idle_threads.first);
 709	if (ln) {
 710		rqstp = llist_entry(ln, struct svc_rqst, rq_idle);
 711		WRITE_ONCE(rqstp->rq_qtime, ktime_get());
 712		if (!task_is_running(rqstp->rq_task)) {
 713			wake_up_process(rqstp->rq_task);
 714			trace_svc_wake_up(rqstp->rq_task->pid);
 715			percpu_counter_inc(&pool->sp_threads_woken);
 716		}
 717		rcu_read_unlock();
 718		return;
 719	}
 720	rcu_read_unlock();
 721
 
 722}
 723EXPORT_SYMBOL_GPL(svc_pool_wake_idle_thread);
 724
 725static struct svc_pool *
 726svc_pool_next(struct svc_serv *serv, struct svc_pool *pool, unsigned int *state)
 
 
 
 727{
 728	return pool ? pool : &serv->sv_pools[(*state)++ % serv->sv_nrpools];
 729}
 730
 731static struct svc_pool *
 732svc_pool_victim(struct svc_serv *serv, struct svc_pool *target_pool,
 733		unsigned int *state)
 734{
 735	struct svc_pool *pool;
 736	unsigned int i;
 737
 738retry:
 739	pool = target_pool;
 740
 741	if (pool != NULL) {
 742		if (atomic_inc_not_zero(&pool->sp_nrthreads))
 743			goto found_pool;
 744		return NULL;
 745	} else {
 
 746		for (i = 0; i < serv->sv_nrpools; i++) {
 747			pool = &serv->sv_pools[--(*state) % serv->sv_nrpools];
 748			if (atomic_inc_not_zero(&pool->sp_nrthreads))
 
 749				goto found_pool;
 
 750		}
 751		return NULL;
 752	}
 753
 754found_pool:
 755	set_bit(SP_VICTIM_REMAINS, &pool->sp_flags);
 756	set_bit(SP_NEED_VICTIM, &pool->sp_flags);
 757	if (!atomic_dec_and_test(&pool->sp_nrthreads))
 758		return pool;
 759	/* Nothing left in this pool any more */
 760	clear_bit(SP_NEED_VICTIM, &pool->sp_flags);
 761	clear_bit(SP_VICTIM_REMAINS, &pool->sp_flags);
 762	goto retry;
 
 
 
 
 
 
 
 763}
 764
 
 765static int
 766svc_start_kthreads(struct svc_serv *serv, struct svc_pool *pool, int nrservs)
 767{
 768	struct svc_rqst	*rqstp;
 769	struct task_struct *task;
 770	struct svc_pool *chosen_pool;
 771	unsigned int state = serv->sv_nrthreads-1;
 772	int node;
 773
 774	do {
 775		nrservs--;
 776		chosen_pool = svc_pool_next(serv, pool, &state);
 
 777		node = svc_pool_map_get_node(chosen_pool->sp_id);
 778
 779		rqstp = svc_prepare_thread(serv, chosen_pool, node);
 780		if (IS_ERR(rqstp))
 781			return PTR_ERR(rqstp);
 782		task = kthread_create_on_node(serv->sv_threadfn, rqstp,
 
 
 783					      node, "%s", serv->sv_name);
 784		if (IS_ERR(task)) {
 
 785			svc_exit_thread(rqstp);
 786			return PTR_ERR(task);
 787		}
 788
 789		rqstp->rq_task = task;
 790		if (serv->sv_nrpools > 1)
 791			svc_pool_map_set_cpumask(task, chosen_pool->sp_id);
 792
 793		svc_sock_update_bufs(serv);
 794		wake_up_process(task);
 795	} while (nrservs > 0);
 796
 797	return 0;
 798}
 799
 
 
 800static int
 801svc_stop_kthreads(struct svc_serv *serv, struct svc_pool *pool, int nrservs)
 802{
 
 803	unsigned int state = serv->sv_nrthreads-1;
 804	struct svc_pool *victim;
 805
 
 806	do {
 807		victim = svc_pool_victim(serv, pool, &state);
 808		if (!victim)
 809			break;
 810		svc_pool_wake_idle_thread(victim);
 811		wait_on_bit(&victim->sp_flags, SP_VICTIM_REMAINS,
 812			    TASK_IDLE);
 813		nrservs++;
 814	} while (nrservs < 0);
 
 815	return 0;
 816}
 817
 818/**
 819 * svc_set_num_threads - adjust number of threads per RPC service
 820 * @serv: RPC service to adjust
 821 * @pool: Specific pool from which to choose threads, or NULL
 822 * @nrservs: New number of threads for @serv (0 or less means kill all threads)
 823 *
 824 * Create or destroy threads to make the number of threads for @serv the
 825 * given number. If @pool is non-NULL, change only threads in that pool;
 826 * otherwise, round-robin between all pools for @serv. @serv's
 827 * sv_nrthreads is adjusted for each thread created or destroyed.
 828 *
 829 * Caller must ensure mutual exclusion between this and server startup or
 830 * shutdown.
 831 *
 832 * Returns zero on success or a negative errno if an error occurred while
 833 * starting a thread.
 834 */
 835int
 836svc_set_num_threads(struct svc_serv *serv, struct svc_pool *pool, int nrservs)
 837{
 838	if (!pool)
 839		nrservs -= serv->sv_nrthreads;
 840	else
 841		nrservs -= atomic_read(&pool->sp_nrthreads);
 
 
 
 
 842
 843	if (nrservs > 0)
 844		return svc_start_kthreads(serv, pool, nrservs);
 845	if (nrservs < 0)
 846		return svc_stop_kthreads(serv, pool, nrservs);
 847	return 0;
 848}
 849EXPORT_SYMBOL_GPL(svc_set_num_threads);
 850
 851/**
 852 * svc_rqst_replace_page - Replace one page in rq_pages[]
 853 * @rqstp: svc_rqst with pages to replace
 854 * @page: replacement page
 855 *
 856 * When replacing a page in rq_pages, batch the release of the
 857 * replaced pages to avoid hammering the page allocator.
 858 *
 859 * Return values:
 860 *   %true: page replaced
 861 *   %false: array bounds checking failed
 862 */
 863bool svc_rqst_replace_page(struct svc_rqst *rqstp, struct page *page)
 864{
 865	struct page **begin = rqstp->rq_pages;
 866	struct page **end = &rqstp->rq_pages[RPCSVC_MAXPAGES];
 867
 868	if (unlikely(rqstp->rq_next_page < begin || rqstp->rq_next_page > end)) {
 869		trace_svc_replace_page_err(rqstp);
 870		return false;
 871	}
 872
 873	if (*rqstp->rq_next_page) {
 874		if (!folio_batch_add(&rqstp->rq_fbatch,
 875				page_folio(*rqstp->rq_next_page)))
 876			__folio_batch_release(&rqstp->rq_fbatch);
 877	}
 878
 879	get_page(page);
 880	*(rqstp->rq_next_page++) = page;
 881	return true;
 882}
 883EXPORT_SYMBOL_GPL(svc_rqst_replace_page);
 884
 885/**
 886 * svc_rqst_release_pages - Release Reply buffer pages
 887 * @rqstp: RPC transaction context
 888 *
 889 * Release response pages that might still be in flight after
 890 * svc_send, and any spliced filesystem-owned pages.
 891 */
 892void svc_rqst_release_pages(struct svc_rqst *rqstp)
 893{
 894	int i, count = rqstp->rq_next_page - rqstp->rq_respages;
 
 
 
 
 
 
 
 895
 896	if (count) {
 897		release_pages(rqstp->rq_respages, count);
 898		for (i = 0; i < count; i++)
 899			rqstp->rq_respages[i] = NULL;
 900	}
 901}
 
 902
 903/*
 904 * Called from a server thread as it's exiting. Caller must hold the "service
 905 * mutex" for the service.
 906 */
 907void
 908svc_rqst_free(struct svc_rqst *rqstp)
 909{
 910	folio_batch_release(&rqstp->rq_fbatch);
 911	svc_release_buffer(rqstp);
 912	if (rqstp->rq_scratch_page)
 913		put_page(rqstp->rq_scratch_page);
 914	kfree(rqstp->rq_resp);
 915	kfree(rqstp->rq_argp);
 916	kfree(rqstp->rq_auth_data);
 917	kfree_rcu(rqstp, rq_rcu_head);
 918}
 919EXPORT_SYMBOL_GPL(svc_rqst_free);
 920
 921void
 922svc_exit_thread(struct svc_rqst *rqstp)
 923{
 924	struct svc_serv	*serv = rqstp->rq_server;
 925	struct svc_pool	*pool = rqstp->rq_pool;
 926
 927	list_del_rcu(&rqstp->rq_all);
 928
 929	atomic_dec(&pool->sp_nrthreads);
 930
 931	spin_lock_bh(&serv->sv_lock);
 932	serv->sv_nrthreads -= 1;
 933	spin_unlock_bh(&serv->sv_lock);
 934	svc_sock_update_bufs(serv);
 935
 936	svc_rqst_free(rqstp);
 937
 938	clear_and_wake_up_bit(SP_VICTIM_REMAINS, &pool->sp_flags);
 
 
 939}
 940EXPORT_SYMBOL_GPL(svc_exit_thread);
 941
 942/*
 943 * Register an "inet" protocol family netid with the local
 944 * rpcbind daemon via an rpcbind v4 SET request.
 945 *
 946 * No netconfig infrastructure is available in the kernel, so
 947 * we map IP_ protocol numbers to netids by hand.
 948 *
 949 * Returns zero on success; a negative errno value is returned
 950 * if any error occurs.
 951 */
 952static int __svc_rpcb_register4(struct net *net, const u32 program,
 953				const u32 version,
 954				const unsigned short protocol,
 955				const unsigned short port)
 956{
 957	const struct sockaddr_in sin = {
 958		.sin_family		= AF_INET,
 959		.sin_addr.s_addr	= htonl(INADDR_ANY),
 960		.sin_port		= htons(port),
 961	};
 962	const char *netid;
 963	int error;
 964
 965	switch (protocol) {
 966	case IPPROTO_UDP:
 967		netid = RPCBIND_NETID_UDP;
 968		break;
 969	case IPPROTO_TCP:
 970		netid = RPCBIND_NETID_TCP;
 971		break;
 972	default:
 973		return -ENOPROTOOPT;
 974	}
 975
 976	error = rpcb_v4_register(net, program, version,
 977					(const struct sockaddr *)&sin, netid);
 978
 979	/*
 980	 * User space didn't support rpcbind v4, so retry this
 981	 * registration request with the legacy rpcbind v2 protocol.
 982	 */
 983	if (error == -EPROTONOSUPPORT)
 984		error = rpcb_register(net, program, version, protocol, port);
 985
 986	return error;
 987}
 988
 989#if IS_ENABLED(CONFIG_IPV6)
 990/*
 991 * Register an "inet6" protocol family netid with the local
 992 * rpcbind daemon via an rpcbind v4 SET request.
 993 *
 994 * No netconfig infrastructure is available in the kernel, so
 995 * we map IP_ protocol numbers to netids by hand.
 996 *
 997 * Returns zero on success; a negative errno value is returned
 998 * if any error occurs.
 999 */
1000static int __svc_rpcb_register6(struct net *net, const u32 program,
1001				const u32 version,
1002				const unsigned short protocol,
1003				const unsigned short port)
1004{
1005	const struct sockaddr_in6 sin6 = {
1006		.sin6_family		= AF_INET6,
1007		.sin6_addr		= IN6ADDR_ANY_INIT,
1008		.sin6_port		= htons(port),
1009	};
1010	const char *netid;
1011	int error;
1012
1013	switch (protocol) {
1014	case IPPROTO_UDP:
1015		netid = RPCBIND_NETID_UDP6;
1016		break;
1017	case IPPROTO_TCP:
1018		netid = RPCBIND_NETID_TCP6;
1019		break;
1020	default:
1021		return -ENOPROTOOPT;
1022	}
1023
1024	error = rpcb_v4_register(net, program, version,
1025					(const struct sockaddr *)&sin6, netid);
1026
1027	/*
1028	 * User space didn't support rpcbind version 4, so we won't
1029	 * use a PF_INET6 listener.
1030	 */
1031	if (error == -EPROTONOSUPPORT)
1032		error = -EAFNOSUPPORT;
1033
1034	return error;
1035}
1036#endif	/* IS_ENABLED(CONFIG_IPV6) */
1037
1038/*
1039 * Register a kernel RPC service via rpcbind version 4.
1040 *
1041 * Returns zero on success; a negative errno value is returned
1042 * if any error occurs.
1043 */
1044static int __svc_register(struct net *net, const char *progname,
1045			  const u32 program, const u32 version,
1046			  const int family,
1047			  const unsigned short protocol,
1048			  const unsigned short port)
1049{
1050	int error = -EAFNOSUPPORT;
1051
1052	switch (family) {
1053	case PF_INET:
1054		error = __svc_rpcb_register4(net, program, version,
1055						protocol, port);
1056		break;
1057#if IS_ENABLED(CONFIG_IPV6)
1058	case PF_INET6:
1059		error = __svc_rpcb_register6(net, program, version,
1060						protocol, port);
1061#endif
1062	}
1063
1064	trace_svc_register(progname, version, family, protocol, port, error);
1065	return error;
1066}
1067
1068int svc_rpcbind_set_version(struct net *net,
1069			    const struct svc_program *progp,
1070			    u32 version, int family,
1071			    unsigned short proto,
1072			    unsigned short port)
1073{
 
 
 
 
 
1074	return __svc_register(net, progp->pg_name, progp->pg_prog,
1075				version, family, proto, port);
1076
1077}
1078EXPORT_SYMBOL_GPL(svc_rpcbind_set_version);
1079
1080int svc_generic_rpcbind_set(struct net *net,
1081			    const struct svc_program *progp,
1082			    u32 version, int family,
1083			    unsigned short proto,
1084			    unsigned short port)
1085{
1086	const struct svc_version *vers = progp->pg_vers[version];
1087	int error;
1088
1089	if (vers == NULL)
1090		return 0;
1091
1092	if (vers->vs_hidden) {
1093		trace_svc_noregister(progp->pg_name, version, proto,
1094				     port, family, 0);
 
 
 
1095		return 0;
1096	}
1097
1098	/*
1099	 * Don't register a UDP port if we need congestion
1100	 * control.
1101	 */
1102	if (vers->vs_need_cong_ctrl && proto == IPPROTO_UDP)
1103		return 0;
1104
1105	error = svc_rpcbind_set_version(net, progp, version,
1106					family, proto, port);
1107
1108	return (vers->vs_rpcb_optnl) ? 0 : error;
1109}
1110EXPORT_SYMBOL_GPL(svc_generic_rpcbind_set);
1111
1112/**
1113 * svc_register - register an RPC service with the local portmapper
1114 * @serv: svc_serv struct for the service to register
1115 * @net: net namespace for the service to register
1116 * @family: protocol family of service's listener socket
1117 * @proto: transport protocol number to advertise
1118 * @port: port to advertise
1119 *
1120 * Service is registered for any address in the passed-in protocol family
1121 */
1122int svc_register(const struct svc_serv *serv, struct net *net,
1123		 const int family, const unsigned short proto,
1124		 const unsigned short port)
1125{
1126	struct svc_program	*progp;
1127	unsigned int		i;
1128	int			error = 0;
1129
1130	WARN_ON_ONCE(proto == 0 && port == 0);
1131	if (proto == 0 && port == 0)
1132		return -EINVAL;
1133
1134	for (progp = serv->sv_program; progp; progp = progp->pg_next) {
1135		for (i = 0; i < progp->pg_nvers; i++) {
1136
1137			error = progp->pg_rpcbind_set(net, progp, i,
1138					family, proto, port);
1139			if (error < 0) {
1140				printk(KERN_WARNING "svc: failed to register "
1141					"%sv%u RPC service (errno %d).\n",
1142					progp->pg_name, i, -error);
1143				break;
1144			}
1145		}
1146	}
1147
1148	return error;
1149}
1150
1151/*
1152 * If user space is running rpcbind, it should take the v4 UNSET
1153 * and clear everything for this [program, version].  If user space
1154 * is running portmap, it will reject the v4 UNSET, but won't have
1155 * any "inet6" entries anyway.  So a PMAP_UNSET should be sufficient
1156 * in this case to clear all existing entries for [program, version].
1157 */
1158static void __svc_unregister(struct net *net, const u32 program, const u32 version,
1159			     const char *progname)
1160{
1161	int error;
1162
1163	error = rpcb_v4_register(net, program, version, NULL, "");
1164
1165	/*
1166	 * User space didn't support rpcbind v4, so retry this
1167	 * request with the legacy rpcbind v2 protocol.
1168	 */
1169	if (error == -EPROTONOSUPPORT)
1170		error = rpcb_register(net, program, version, 0, 0);
1171
1172	trace_svc_unregister(progname, version, error);
 
1173}
1174
1175/*
1176 * All netids, bind addresses and ports registered for [program, version]
1177 * are removed from the local rpcbind database (if the service is not
1178 * hidden) to make way for a new instance of the service.
1179 *
1180 * The result of unregistration is reported via dprintk for those who want
1181 * verification of the result, but is otherwise not important.
1182 */
1183static void svc_unregister(const struct svc_serv *serv, struct net *net)
1184{
1185	struct sighand_struct *sighand;
1186	struct svc_program *progp;
1187	unsigned long flags;
1188	unsigned int i;
1189
1190	clear_thread_flag(TIF_SIGPENDING);
1191
1192	for (progp = serv->sv_program; progp; progp = progp->pg_next) {
1193		for (i = 0; i < progp->pg_nvers; i++) {
1194			if (progp->pg_vers[i] == NULL)
1195				continue;
1196			if (progp->pg_vers[i]->vs_hidden)
1197				continue;
 
 
 
1198			__svc_unregister(net, progp->pg_prog, i, progp->pg_name);
1199		}
1200	}
1201
1202	rcu_read_lock();
1203	sighand = rcu_dereference(current->sighand);
1204	spin_lock_irqsave(&sighand->siglock, flags);
1205	recalc_sigpending();
1206	spin_unlock_irqrestore(&sighand->siglock, flags);
1207	rcu_read_unlock();
1208}
1209
1210/*
1211 * dprintk the given error with the address of the client that caused it.
1212 */
1213#if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
1214static __printf(2, 3)
1215void svc_printk(struct svc_rqst *rqstp, const char *fmt, ...)
1216{
1217	struct va_format vaf;
1218	va_list args;
1219	char 	buf[RPC_MAX_ADDRBUFLEN];
1220
1221	va_start(args, fmt);
1222
1223	vaf.fmt = fmt;
1224	vaf.va = &args;
1225
1226	dprintk("svc: %s: %pV", svc_print_addr(rqstp, buf, sizeof(buf)), &vaf);
1227
1228	va_end(args);
1229}
1230#else
1231static __printf(2,3) void svc_printk(struct svc_rqst *rqstp, const char *fmt, ...) {}
1232#endif
1233
1234__be32
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1235svc_generic_init_request(struct svc_rqst *rqstp,
1236		const struct svc_program *progp,
1237		struct svc_process_info *ret)
1238{
1239	const struct svc_version *versp = NULL;	/* compiler food */
1240	const struct svc_procedure *procp = NULL;
1241
1242	if (rqstp->rq_vers >= progp->pg_nvers )
1243		goto err_bad_vers;
1244	versp = progp->pg_vers[rqstp->rq_vers];
1245	if (!versp)
1246		goto err_bad_vers;
1247
1248	/*
1249	 * Some protocol versions (namely NFSv4) require some form of
1250	 * congestion control.  (See RFC 7530 section 3.1 paragraph 2)
1251	 * In other words, UDP is not allowed. We mark those when setting
1252	 * up the svc_xprt, and verify that here.
1253	 *
1254	 * The spec is not very clear about what error should be returned
1255	 * when someone tries to access a server that is listening on UDP
1256	 * for lower versions. RPC_PROG_MISMATCH seems to be the closest
1257	 * fit.
1258	 */
1259	if (versp->vs_need_cong_ctrl && rqstp->rq_xprt &&
1260	    !test_bit(XPT_CONG_CTRL, &rqstp->rq_xprt->xpt_flags))
1261		goto err_bad_vers;
1262
1263	if (rqstp->rq_proc >= versp->vs_nproc)
1264		goto err_bad_proc;
1265	rqstp->rq_procinfo = procp = &versp->vs_proc[rqstp->rq_proc];
1266	if (!procp)
1267		goto err_bad_proc;
1268
1269	/* Initialize storage for argp and resp */
1270	memset(rqstp->rq_argp, 0, procp->pc_argzero);
1271	memset(rqstp->rq_resp, 0, procp->pc_ressize);
1272
1273	/* Bump per-procedure stats counter */
1274	this_cpu_inc(versp->vs_count[rqstp->rq_proc]);
1275
1276	ret->dispatch = versp->vs_dispatch;
1277	return rpc_success;
1278err_bad_vers:
1279	ret->mismatch.lovers = progp->pg_lovers;
1280	ret->mismatch.hivers = progp->pg_hivers;
1281	return rpc_prog_mismatch;
1282err_bad_proc:
1283	return rpc_proc_unavail;
1284}
1285EXPORT_SYMBOL_GPL(svc_generic_init_request);
1286
1287/*
1288 * Common routine for processing the RPC request.
1289 */
1290static int
1291svc_process_common(struct svc_rqst *rqstp)
1292{
1293	struct xdr_stream	*xdr = &rqstp->rq_res_stream;
1294	struct svc_program	*progp;
1295	const struct svc_procedure *procp = NULL;
1296	struct svc_serv		*serv = rqstp->rq_server;
1297	struct svc_process_info process;
1298	enum svc_auth_status	auth_res;
1299	unsigned int		aoffset;
1300	int			rc;
1301	__be32			*p;
 
 
 
 
 
 
1302
 
 
1303	/* Will be turned off only when NFSv4 Sessions are used */
1304	set_bit(RQ_USEDEFERRAL, &rqstp->rq_flags);
1305	clear_bit(RQ_DROPME, &rqstp->rq_flags);
1306
1307	/* Construct the first words of the reply: */
1308	svcxdr_init_encode(rqstp);
1309	xdr_stream_encode_be32(xdr, rqstp->rq_xid);
1310	xdr_stream_encode_be32(xdr, rpc_reply);
1311
1312	p = xdr_inline_decode(&rqstp->rq_arg_stream, XDR_UNIT * 4);
1313	if (unlikely(!p))
1314		goto err_short_len;
1315	if (*p++ != cpu_to_be32(RPC_VERSION))
 
 
1316		goto err_bad_rpc;
1317
1318	xdr_stream_encode_be32(xdr, rpc_msg_accepted);
 
1319
1320	rqstp->rq_prog = be32_to_cpup(p++);
1321	rqstp->rq_vers = be32_to_cpup(p++);
1322	rqstp->rq_proc = be32_to_cpup(p);
 
 
1323
1324	for (progp = serv->sv_program; progp; progp = progp->pg_next)
1325		if (rqstp->rq_prog == progp->pg_prog)
1326			break;
1327
1328	/*
1329	 * Decode auth data, and add verifier to reply buffer.
1330	 * We do this before anything else in order to get a decent
1331	 * auth verifier.
1332	 */
1333	auth_res = svc_authenticate(rqstp);
1334	/* Also give the program a chance to reject this call: */
1335	if (auth_res == SVC_OK && progp)
 
1336		auth_res = progp->pg_authenticate(rqstp);
1337	trace_svc_authenticate(rqstp, auth_res);
1338	switch (auth_res) {
1339	case SVC_OK:
1340		break;
1341	case SVC_GARBAGE:
1342		goto err_garbage_args;
1343	case SVC_SYSERR:
1344		goto err_system_err;
 
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	default:
1354		pr_warn_once("Unexpected svc_auth_status (%d)\n", auth_res);
1355		goto err_system_err;
1356	}
1357
1358	if (progp == NULL)
1359		goto err_bad_prog;
1360
1361	switch (progp->pg_init_request(rqstp, progp, &process)) {
 
1362	case rpc_success:
1363		break;
1364	case rpc_prog_unavail:
1365		goto err_bad_prog;
1366	case rpc_prog_mismatch:
1367		goto err_bad_vers;
1368	case rpc_proc_unavail:
1369		goto err_bad_proc;
1370	}
1371
1372	procp = rqstp->rq_procinfo;
1373	/* Should this check go into the dispatcher? */
1374	if (!procp || !procp->pc_func)
1375		goto err_bad_proc;
1376
1377	/* Syntactic check complete */
1378	serv->sv_stats->rpccnt++;
1379	trace_svc_process(rqstp, progp->pg_name);
1380
1381	aoffset = xdr_stream_pos(xdr);
 
 
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	rc = process.dispatch(rqstp);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1391	if (procp->pc_release)
1392		procp->pc_release(rqstp);
1393	xdr_finish_decode(xdr);
1394
1395	if (!rc)
1396		goto dropit;
1397	if (rqstp->rq_auth_stat != rpc_auth_ok)
1398		goto err_bad_auth;
1399
1400	if (*rqstp->rq_accept_statp != rpc_success)
1401		xdr_truncate_encode(xdr, aoffset);
1402
1403	if (procp->pc_encode == NULL)
1404		goto dropit;
1405
1406 sendit:
1407	if (svc_authorise(rqstp))
1408		goto close_xprt;
1409	return 1;		/* Caller can now send it */
1410
 
 
 
1411 dropit:
1412	svc_authorise(rqstp);	/* doesn't hurt to call this twice */
1413	dprintk("svc: svc_process dropit\n");
1414	return 0;
1415
1416 close:
1417	svc_authorise(rqstp);
1418close_xprt:
1419	if (rqstp->rq_xprt && test_bit(XPT_TEMP, &rqstp->rq_xprt->xpt_flags))
1420		svc_xprt_close(rqstp->rq_xprt);
1421	dprintk("svc: svc_process close\n");
1422	return 0;
1423
1424err_short_len:
1425	svc_printk(rqstp, "short len %u, dropping request\n",
1426		   rqstp->rq_arg.len);
1427	goto close_xprt;
1428
1429err_bad_rpc:
1430	serv->sv_stats->rpcbadfmt++;
1431	xdr_stream_encode_u32(xdr, RPC_MSG_DENIED);
1432	xdr_stream_encode_u32(xdr, RPC_MISMATCH);
1433	/* Only RPCv2 supported */
1434	xdr_stream_encode_u32(xdr, RPC_VERSION);
1435	xdr_stream_encode_u32(xdr, RPC_VERSION);
1436	return 1;	/* don't wrap */
1437
 
 
 
1438err_bad_auth:
1439	dprintk("svc: authentication failed (%d)\n",
1440		be32_to_cpu(rqstp->rq_auth_stat));
1441	serv->sv_stats->rpcbadauth++;
1442	/* Restore write pointer to location of reply status: */
1443	xdr_truncate_encode(xdr, XDR_UNIT * 2);
1444	xdr_stream_encode_u32(xdr, RPC_MSG_DENIED);
1445	xdr_stream_encode_u32(xdr, RPC_AUTH_ERROR);
1446	xdr_stream_encode_be32(xdr, rqstp->rq_auth_stat);
1447	goto sendit;
1448
1449err_bad_prog:
1450	dprintk("svc: unknown program %d\n", rqstp->rq_prog);
1451	serv->sv_stats->rpcbadfmt++;
1452	*rqstp->rq_accept_statp = rpc_prog_unavail;
1453	goto sendit;
1454
1455err_bad_vers:
1456	svc_printk(rqstp, "unknown version (%d for prog %d, %s)\n",
1457		       rqstp->rq_vers, rqstp->rq_prog, progp->pg_name);
1458
1459	serv->sv_stats->rpcbadfmt++;
1460	*rqstp->rq_accept_statp = rpc_prog_mismatch;
1461
1462	/*
1463	 * svc_authenticate() has already added the verifier and
1464	 * advanced the stream just past rq_accept_statp.
1465	 */
1466	xdr_stream_encode_u32(xdr, process.mismatch.lovers);
1467	xdr_stream_encode_u32(xdr, 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	*rqstp->rq_accept_statp = rpc_proc_unavail;
1475	goto sendit;
1476
1477err_garbage_args:
1478	svc_printk(rqstp, "failed to decode RPC header\n");
1479
 
 
1480	serv->sv_stats->rpcbadfmt++;
1481	*rqstp->rq_accept_statp = rpc_garbage_args;
1482	goto sendit;
1483
1484err_system_err:
1485	serv->sv_stats->rpcbadfmt++;
1486	*rqstp->rq_accept_statp = rpc_system_err;
1487	goto sendit;
1488}
1489
1490/**
1491 * svc_process - Execute one RPC transaction
1492 * @rqstp: RPC transaction context
1493 *
1494 */
1495void svc_process(struct svc_rqst *rqstp)
 
1496{
 
1497	struct kvec		*resv = &rqstp->rq_res.head[0];
1498	__be32 *p;
1499
1500#if IS_ENABLED(CONFIG_FAIL_SUNRPC)
1501	if (!fail_sunrpc.ignore_server_disconnect &&
1502	    should_fail(&fail_sunrpc.attr, 1))
1503		svc_xprt_deferred_close(rqstp->rq_xprt);
1504#endif
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_next_page;
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	svcxdr_init_decode(rqstp);
1522	p = xdr_inline_decode(&rqstp->rq_arg_stream, XDR_UNIT * 2);
1523	if (unlikely(!p))
 
 
1524		goto out_drop;
1525	rqstp->rq_xid = *p++;
1526	if (unlikely(*p != rpc_call))
1527		goto out_baddir;
1528
1529	if (!svc_process_common(rqstp))
1530		goto out_drop;
1531	svc_send(rqstp);
1532	return;
 
 
 
1533
1534out_baddir:
1535	svc_printk(rqstp, "bad direction 0x%08x, dropping request\n",
1536		   be32_to_cpu(*p));
1537	rqstp->rq_server->sv_stats->rpcbadfmt++;
1538out_drop:
1539	svc_drop(rqstp);
 
1540}
 
1541
1542#if defined(CONFIG_SUNRPC_BACKCHANNEL)
1543/**
1544 * svc_process_bc - process a reverse-direction RPC request
1545 * @req: RPC request to be used for client-side processing
1546 * @rqstp: server-side execution context
1547 *
1548 */
1549void svc_process_bc(struct rpc_rqst *req, struct svc_rqst *rqstp)
 
 
1550{
 
 
1551	struct rpc_task *task;
1552	int proc_error;
1553	struct rpc_timeout timeout;
 
 
1554
1555	/* Build the svc_rqst used by the common processing routine */
1556	rqstp->rq_xid = req->rq_xid;
1557	rqstp->rq_prot = req->rq_xprt->prot;
 
1558	rqstp->rq_bc_net = req->rq_xprt->xprt_net;
1559
1560	rqstp->rq_addrlen = sizeof(req->rq_xprt->addr);
1561	memcpy(&rqstp->rq_addr, &req->rq_xprt->addr, rqstp->rq_addrlen);
1562	memcpy(&rqstp->rq_arg, &req->rq_rcv_buf, sizeof(rqstp->rq_arg));
1563	memcpy(&rqstp->rq_res, &req->rq_snd_buf, sizeof(rqstp->rq_res));
1564
1565	/* Adjust the argument buffer length */
1566	rqstp->rq_arg.len = req->rq_private_buf.len;
1567	if (rqstp->rq_arg.len <= rqstp->rq_arg.head[0].iov_len) {
1568		rqstp->rq_arg.head[0].iov_len = rqstp->rq_arg.len;
1569		rqstp->rq_arg.page_len = 0;
1570	} else if (rqstp->rq_arg.len <= rqstp->rq_arg.head[0].iov_len +
1571			rqstp->rq_arg.page_len)
1572		rqstp->rq_arg.page_len = rqstp->rq_arg.len -
1573			rqstp->rq_arg.head[0].iov_len;
1574	else
1575		rqstp->rq_arg.len = rqstp->rq_arg.head[0].iov_len +
1576			rqstp->rq_arg.page_len;
1577
1578	/* Reset the response buffer */
1579	rqstp->rq_res.head[0].iov_len = 0;
1580
1581	/*
1582	 * Skip the XID and calldir fields because they've already
1583	 * been processed by the caller.
1584	 */
1585	svcxdr_init_decode(rqstp);
1586	if (!xdr_inline_decode(&rqstp->rq_arg_stream, XDR_UNIT * 2))
1587		return;
1588
1589	/* Parse and execute the bc call */
1590	proc_error = svc_process_common(rqstp);
1591
1592	atomic_dec(&req->rq_xprt->bc_slot_count);
1593	if (!proc_error) {
1594		/* Processing error: drop the request */
1595		xprt_free_bc_request(req);
1596		return;
 
1597	}
1598	/* Finally, send the reply synchronously */
1599	if (rqstp->bc_to_initval > 0) {
1600		timeout.to_initval = rqstp->bc_to_initval;
1601		timeout.to_retries = rqstp->bc_to_retries;
1602	} else {
1603		timeout.to_initval = req->rq_xprt->timeout->to_initval;
1604		timeout.to_retries = req->rq_xprt->timeout->to_retries;
1605	}
1606	memcpy(&req->rq_snd_buf, &rqstp->rq_res, sizeof(req->rq_snd_buf));
1607	task = rpc_run_bc_task(req, &timeout);
1608
1609	if (IS_ERR(task))
1610		return;
1611
1612	WARN_ON_ONCE(atomic_read(&task->tk_count) != 1);
 
1613	rpc_put_task(task);
 
 
 
 
1614}
1615EXPORT_SYMBOL_GPL(svc_process_bc);
1616#endif /* CONFIG_SUNRPC_BACKCHANNEL */
1617
1618/**
1619 * svc_max_payload - Return transport-specific limit on the RPC payload
1620 * @rqstp: RPC transaction context
1621 *
1622 * Returns the maximum number of payload bytes the current transport
1623 * allows.
1624 */
1625u32 svc_max_payload(const struct svc_rqst *rqstp)
1626{
1627	u32 max = rqstp->rq_xprt->xpt_class->xcl_max_payload;
1628
1629	if (rqstp->rq_server->sv_max_payload < max)
1630		max = rqstp->rq_server->sv_max_payload;
1631	return max;
1632}
1633EXPORT_SYMBOL_GPL(svc_max_payload);
1634
1635/**
1636 * svc_proc_name - Return RPC procedure name in string form
1637 * @rqstp: svc_rqst to operate on
1638 *
1639 * Return value:
1640 *   Pointer to a NUL-terminated string
1641 */
1642const char *svc_proc_name(const struct svc_rqst *rqstp)
1643{
1644	if (rqstp && rqstp->rq_procinfo)
1645		return rqstp->rq_procinfo->pc_name;
1646	return "unknown";
1647}
1648
1649
1650/**
1651 * svc_encode_result_payload - mark a range of bytes as a result payload
1652 * @rqstp: svc_rqst to operate on
1653 * @offset: payload's byte offset in rqstp->rq_res
1654 * @length: size of payload, in bytes
1655 *
1656 * Returns zero on success, or a negative errno if a permanent
1657 * error occurred.
1658 */
1659int svc_encode_result_payload(struct svc_rqst *rqstp, unsigned int offset,
1660			      unsigned int length)
1661{
1662	return rqstp->rq_xprt->xpt_ops->xpo_result_payload(rqstp, offset,
1663							   length);
1664}
1665EXPORT_SYMBOL_GPL(svc_encode_result_payload);
1666
1667/**
1668 * svc_fill_write_vector - Construct data argument for VFS write call
1669 * @rqstp: svc_rqst to operate on
1670 * @payload: xdr_buf containing only the write data payload
 
 
1671 *
1672 * Fills in rqstp::rq_vec, and returns the number of elements.
1673 */
1674unsigned int svc_fill_write_vector(struct svc_rqst *rqstp,
1675				   struct xdr_buf *payload)
1676{
1677	struct page **pages = payload->pages;
1678	struct kvec *first = payload->head;
1679	struct kvec *vec = rqstp->rq_vec;
1680	size_t total = payload->len;
1681	unsigned int i;
1682
1683	/* Some types of transport can present the write payload
1684	 * entirely in rq_arg.pages. In this case, @first is empty.
1685	 */
1686	i = 0;
1687	if (first->iov_len) {
1688		vec[i].iov_base = first->iov_base;
1689		vec[i].iov_len = min_t(size_t, total, first->iov_len);
1690		total -= vec[i].iov_len;
1691		++i;
1692	}
1693
1694	while (total) {
1695		vec[i].iov_base = page_address(*pages);
1696		vec[i].iov_len = min_t(size_t, total, PAGE_SIZE);
1697		total -= vec[i].iov_len;
1698		++i;
1699		++pages;
1700	}
1701
1702	WARN_ON_ONCE(i > ARRAY_SIZE(rqstp->rq_vec));
1703	return i;
1704}
1705EXPORT_SYMBOL_GPL(svc_fill_write_vector);
1706
1707/**
1708 * svc_fill_symlink_pathname - Construct pathname argument for VFS symlink call
1709 * @rqstp: svc_rqst to operate on
1710 * @first: buffer containing first section of pathname
1711 * @p: buffer containing remaining section of pathname
1712 * @total: total length of the pathname argument
1713 *
1714 * The VFS symlink API demands a NUL-terminated pathname in mapped memory.
1715 * Returns pointer to a NUL-terminated string, or an ERR_PTR. Caller must free
1716 * the returned string.
1717 */
1718char *svc_fill_symlink_pathname(struct svc_rqst *rqstp, struct kvec *first,
1719				void *p, size_t total)
1720{
1721	size_t len, remaining;
1722	char *result, *dst;
1723
1724	result = kmalloc(total + 1, GFP_KERNEL);
1725	if (!result)
1726		return ERR_PTR(-ESERVERFAULT);
1727
1728	dst = result;
1729	remaining = total;
1730
1731	len = min_t(size_t, total, first->iov_len);
1732	if (len) {
1733		memcpy(dst, first->iov_base, len);
1734		dst += len;
1735		remaining -= len;
1736	}
1737
1738	if (remaining) {
1739		len = min_t(size_t, remaining, PAGE_SIZE);
1740		memcpy(dst, p, len);
1741		dst += len;
1742	}
1743
1744	*dst = '\0';
1745
1746	/* Sanity check: Linux doesn't allow the pathname argument to
1747	 * contain a NUL byte.
1748	 */
1749	if (strlen(result) != total) {
1750		kfree(result);
1751		return ERR_PTR(-EINVAL);
1752	}
1753	return result;
1754}
1755EXPORT_SYMBOL_GPL(svc_fill_symlink_pathname);