1// SPDX-License-Identifier: GPL-2.0-only
   2/*
   3 * linux/net/sunrpc/svc_xprt.c
   4 *
   5 * Author: Tom Tucker <tom@opengridcomputing.com>
   6 */
   7
   8#include <linux/sched.h>
   9#include <linux/sched/mm.h>
  10#include <linux/errno.h>
  11#include <linux/freezer.h>
 
  12#include <linux/slab.h>
  13#include <net/sock.h>
  14#include <linux/sunrpc/addr.h>
  15#include <linux/sunrpc/stats.h>
  16#include <linux/sunrpc/svc_xprt.h>
  17#include <linux/sunrpc/svcsock.h>
  18#include <linux/sunrpc/xprt.h>
  19#include <linux/sunrpc/bc_xprt.h>
  20#include <linux/module.h>
  21#include <linux/netdevice.h>
  22#include <trace/events/sunrpc.h>
  23
  24#define RPCDBG_FACILITY	RPCDBG_SVCXPRT
  25
  26static unsigned int svc_rpc_per_connection_limit __read_mostly;
  27module_param(svc_rpc_per_connection_limit, uint, 0644);
  28
  29
  30static struct svc_deferred_req *svc_deferred_dequeue(struct svc_xprt *xprt);
  31static int svc_deferred_recv(struct svc_rqst *rqstp);
  32static struct cache_deferred_req *svc_defer(struct cache_req *req);
  33static void svc_age_temp_xprts(struct timer_list *t);
  34static void svc_delete_xprt(struct svc_xprt *xprt);
  35
  36/* apparently the "standard" is that clients close
  37 * idle connections after 5 minutes, servers after
  38 * 6 minutes
  39 *   http://nfsv4bat.org/Documents/ConnectAThon/1996/nfstcp.pdf
  40 */
  41static int svc_conn_age_period = 6*60;
  42
  43/* List of registered transport classes */
  44static DEFINE_SPINLOCK(svc_xprt_class_lock);
  45static LIST_HEAD(svc_xprt_class_list);
  46
  47/* SMP locking strategy:
  48 *
 
  49 *	svc_serv->sv_lock protects sv_tempsocks, sv_permsocks, sv_tmpcnt.
  50 *	when both need to be taken (rare), svc_serv->sv_lock is first.
  51 *	The "service mutex" protects svc_serv->sv_nrthread.
  52 *	svc_sock->sk_lock protects the svc_sock->sk_deferred list
  53 *             and the ->sk_info_authunix cache.
  54 *
  55 *	The XPT_BUSY bit in xprt->xpt_flags prevents a transport being
  56 *	enqueued multiply. During normal transport processing this bit
  57 *	is set by svc_xprt_enqueue and cleared by svc_xprt_received.
  58 *	Providers should not manipulate this bit directly.
  59 *
  60 *	Some flags can be set to certain values at any time
  61 *	providing that certain rules are followed:
  62 *
  63 *	XPT_CONN, XPT_DATA:
  64 *		- Can be set or cleared at any time.
  65 *		- After a set, svc_xprt_enqueue must be called to enqueue
  66 *		  the transport for processing.
  67 *		- After a clear, the transport must be read/accepted.
  68 *		  If this succeeds, it must be set again.
  69 *	XPT_CLOSE:
  70 *		- Can set at any time. It is never cleared.
  71 *      XPT_DEAD:
  72 *		- Can only be set while XPT_BUSY is held which ensures
  73 *		  that no other thread will be using the transport or will
  74 *		  try to set XPT_DEAD.
  75 */
  76
  77/**
  78 * svc_reg_xprt_class - Register a server-side RPC transport class
  79 * @xcl: New transport class to be registered
  80 *
  81 * Returns zero on success; otherwise a negative errno is returned.
  82 */
  83int svc_reg_xprt_class(struct svc_xprt_class *xcl)
  84{
  85	struct svc_xprt_class *cl;
  86	int res = -EEXIST;
  87
 
 
  88	INIT_LIST_HEAD(&xcl->xcl_list);
  89	spin_lock(&svc_xprt_class_lock);
  90	/* Make sure there isn't already a class with the same name */
  91	list_for_each_entry(cl, &svc_xprt_class_list, xcl_list) {
  92		if (strcmp(xcl->xcl_name, cl->xcl_name) == 0)
  93			goto out;
  94	}
  95	list_add_tail(&xcl->xcl_list, &svc_xprt_class_list);
  96	res = 0;
  97out:
  98	spin_unlock(&svc_xprt_class_lock);
  99	return res;
 100}
 101EXPORT_SYMBOL_GPL(svc_reg_xprt_class);
 102
 103/**
 104 * svc_unreg_xprt_class - Unregister a server-side RPC transport class
 105 * @xcl: Transport class to be unregistered
 106 *
 107 */
 108void svc_unreg_xprt_class(struct svc_xprt_class *xcl)
 109{
 
 110	spin_lock(&svc_xprt_class_lock);
 111	list_del_init(&xcl->xcl_list);
 112	spin_unlock(&svc_xprt_class_lock);
 113}
 114EXPORT_SYMBOL_GPL(svc_unreg_xprt_class);
 115
 116/**
 117 * svc_print_xprts - Format the transport list for printing
 118 * @buf: target buffer for formatted address
 119 * @maxlen: length of target buffer
 120 *
 121 * Fills in @buf with a string containing a list of transport names, each name
 122 * terminated with '\n'. If the buffer is too small, some entries may be
 123 * missing, but it is guaranteed that all lines in the output buffer are
 124 * complete.
 125 *
 126 * Returns positive length of the filled-in string.
 127 */
 128int svc_print_xprts(char *buf, int maxlen)
 129{
 130	struct svc_xprt_class *xcl;
 131	char tmpstr[80];
 132	int len = 0;
 133	buf[0] = '\0';
 134
 135	spin_lock(&svc_xprt_class_lock);
 136	list_for_each_entry(xcl, &svc_xprt_class_list, xcl_list) {
 137		int slen;
 138
 139		slen = snprintf(tmpstr, sizeof(tmpstr), "%s %d\n",
 140				xcl->xcl_name, xcl->xcl_max_payload);
 141		if (slen >= sizeof(tmpstr) || len + slen >= maxlen)
 142			break;
 143		len += slen;
 144		strcat(buf, tmpstr);
 145	}
 146	spin_unlock(&svc_xprt_class_lock);
 147
 148	return len;
 149}
 150
 151/**
 152 * svc_xprt_deferred_close - Close a transport
 153 * @xprt: transport instance
 154 *
 155 * Used in contexts that need to defer the work of shutting down
 156 * the transport to an nfsd thread.
 157 */
 158void svc_xprt_deferred_close(struct svc_xprt *xprt)
 159{
 160	trace_svc_xprt_close(xprt);
 161	if (!test_and_set_bit(XPT_CLOSE, &xprt->xpt_flags))
 162		svc_xprt_enqueue(xprt);
 163}
 164EXPORT_SYMBOL_GPL(svc_xprt_deferred_close);
 165
 166static void svc_xprt_free(struct kref *kref)
 167{
 168	struct svc_xprt *xprt =
 169		container_of(kref, struct svc_xprt, xpt_ref);
 170	struct module *owner = xprt->xpt_class->xcl_owner;
 171	if (test_bit(XPT_CACHE_AUTH, &xprt->xpt_flags))
 172		svcauth_unix_info_release(xprt);
 173	put_cred(xprt->xpt_cred);
 174	put_net_track(xprt->xpt_net, &xprt->ns_tracker);
 175	/* See comment on corresponding get in xs_setup_bc_tcp(): */
 176	if (xprt->xpt_bc_xprt)
 177		xprt_put(xprt->xpt_bc_xprt);
 178	if (xprt->xpt_bc_xps)
 179		xprt_switch_put(xprt->xpt_bc_xps);
 180	trace_svc_xprt_free(xprt);
 181	xprt->xpt_ops->xpo_free(xprt);
 182	module_put(owner);
 183}
 184
 185void svc_xprt_put(struct svc_xprt *xprt)
 186{
 187	kref_put(&xprt->xpt_ref, svc_xprt_free);
 188}
 189EXPORT_SYMBOL_GPL(svc_xprt_put);
 190
 191/*
 192 * Called by transport drivers to initialize the transport independent
 193 * portion of the transport instance.
 194 */
 195void svc_xprt_init(struct net *net, struct svc_xprt_class *xcl,
 196		   struct svc_xprt *xprt, struct svc_serv *serv)
 197{
 198	memset(xprt, 0, sizeof(*xprt));
 199	xprt->xpt_class = xcl;
 200	xprt->xpt_ops = xcl->xcl_ops;
 201	kref_init(&xprt->xpt_ref);
 202	xprt->xpt_server = serv;
 203	INIT_LIST_HEAD(&xprt->xpt_list);
 
 204	INIT_LIST_HEAD(&xprt->xpt_deferred);
 205	INIT_LIST_HEAD(&xprt->xpt_users);
 206	mutex_init(&xprt->xpt_mutex);
 207	spin_lock_init(&xprt->xpt_lock);
 208	set_bit(XPT_BUSY, &xprt->xpt_flags);
 209	xprt->xpt_net = get_net_track(net, &xprt->ns_tracker, GFP_ATOMIC);
 210	strcpy(xprt->xpt_remotebuf, "uninitialized");
 211}
 212EXPORT_SYMBOL_GPL(svc_xprt_init);
 213
 214/**
 215 * svc_xprt_received - start next receiver thread
 216 * @xprt: controlling transport
 217 *
 218 * The caller must hold the XPT_BUSY bit and must
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 219 * not thereafter touch transport data.
 220 *
 221 * Note: XPT_DATA only gets cleared when a read-attempt finds no (or
 222 * insufficient) data.
 223 */
 224void svc_xprt_received(struct svc_xprt *xprt)
 225{
 226	if (!test_bit(XPT_BUSY, &xprt->xpt_flags)) {
 227		WARN_ONCE(1, "xprt=0x%p already busy!", xprt);
 228		return;
 229	}
 230
 231	/* As soon as we clear busy, the xprt could be closed and
 232	 * 'put', so we need a reference to call svc_xprt_enqueue with:
 233	 */
 234	svc_xprt_get(xprt);
 235	smp_mb__before_atomic();
 236	clear_bit(XPT_BUSY, &xprt->xpt_flags);
 237	svc_xprt_enqueue(xprt);
 238	svc_xprt_put(xprt);
 239}
 240EXPORT_SYMBOL_GPL(svc_xprt_received);
 241
 242void svc_add_new_perm_xprt(struct svc_serv *serv, struct svc_xprt *new)
 243{
 244	clear_bit(XPT_TEMP, &new->xpt_flags);
 245	spin_lock_bh(&serv->sv_lock);
 246	list_add(&new->xpt_list, &serv->sv_permsocks);
 247	spin_unlock_bh(&serv->sv_lock);
 248	svc_xprt_received(new);
 249}
 250
 251static int _svc_xprt_create(struct svc_serv *serv, const char *xprt_name,
 252			    struct net *net, struct sockaddr *sap,
 253			    size_t len, int flags, const struct cred *cred)
 254{
 255	struct svc_xprt_class *xcl;
 256
 
 257	spin_lock(&svc_xprt_class_lock);
 258	list_for_each_entry(xcl, &svc_xprt_class_list, xcl_list) {
 259		struct svc_xprt *newxprt;
 260		unsigned short newport;
 261
 262		if (strcmp(xprt_name, xcl->xcl_name))
 263			continue;
 264
 265		if (!try_module_get(xcl->xcl_owner))
 266			goto err;
 267
 268		spin_unlock(&svc_xprt_class_lock);
 269		newxprt = xcl->xcl_ops->xpo_create(serv, net, sap, len, flags);
 270		if (IS_ERR(newxprt)) {
 271			trace_svc_xprt_create_err(serv->sv_programs->pg_name,
 272						  xcl->xcl_name, sap, len,
 273						  newxprt);
 274			module_put(xcl->xcl_owner);
 275			return PTR_ERR(newxprt);
 276		}
 277		newxprt->xpt_cred = get_cred(cred);
 278		svc_add_new_perm_xprt(serv, newxprt);
 279		newport = svc_xprt_local_port(newxprt);
 280		return newport;
 281	}
 282 err:
 283	spin_unlock(&svc_xprt_class_lock);
 
 
 284	/* This errno is exposed to user space.  Provide a reasonable
 285	 * perror msg for a bad transport. */
 286	return -EPROTONOSUPPORT;
 287}
 288
 289/**
 290 * svc_xprt_create_from_sa - Add a new listener to @serv from socket address
 291 * @serv: target RPC service
 292 * @xprt_name: transport class name
 293 * @net: network namespace
 294 * @sap: socket address pointer
 295 * @flags: SVC_SOCK flags
 296 * @cred: credential to bind to this transport
 297 *
 298 * Return local xprt port on success or %-EPROTONOSUPPORT on failure
 299 */
 300int svc_xprt_create_from_sa(struct svc_serv *serv, const char *xprt_name,
 301			    struct net *net, struct sockaddr *sap,
 302			    int flags, const struct cred *cred)
 303{
 304	size_t len;
 305	int err;
 306
 307	switch (sap->sa_family) {
 308	case AF_INET:
 309		len = sizeof(struct sockaddr_in);
 310		break;
 311#if IS_ENABLED(CONFIG_IPV6)
 312	case AF_INET6:
 313		len = sizeof(struct sockaddr_in6);
 314		break;
 315#endif
 316	default:
 317		return -EAFNOSUPPORT;
 318	}
 319
 320	err = _svc_xprt_create(serv, xprt_name, net, sap, len, flags, cred);
 321	if (err == -EPROTONOSUPPORT) {
 322		request_module("svc%s", xprt_name);
 323		err = _svc_xprt_create(serv, xprt_name, net, sap, len, flags,
 324				       cred);
 325	}
 326
 327	return err;
 328}
 329EXPORT_SYMBOL_GPL(svc_xprt_create_from_sa);
 330
 331/**
 332 * svc_xprt_create - Add a new listener to @serv
 333 * @serv: target RPC service
 334 * @xprt_name: transport class name
 335 * @net: network namespace
 336 * @family: network address family
 337 * @port: listener port
 338 * @flags: SVC_SOCK flags
 339 * @cred: credential to bind to this transport
 340 *
 341 * Return local xprt port on success or %-EPROTONOSUPPORT on failure
 342 */
 343int svc_xprt_create(struct svc_serv *serv, const char *xprt_name,
 344		    struct net *net, const int family,
 345		    const unsigned short port, int flags,
 346		    const struct cred *cred)
 347{
 348	struct sockaddr_in sin = {
 349		.sin_family		= AF_INET,
 350		.sin_addr.s_addr	= htonl(INADDR_ANY),
 351		.sin_port		= htons(port),
 352	};
 353#if IS_ENABLED(CONFIG_IPV6)
 354	struct sockaddr_in6 sin6 = {
 355		.sin6_family		= AF_INET6,
 356		.sin6_addr		= IN6ADDR_ANY_INIT,
 357		.sin6_port		= htons(port),
 358	};
 359#endif
 360	struct sockaddr *sap;
 361
 362	switch (family) {
 363	case PF_INET:
 364		sap = (struct sockaddr *)&sin;
 365		break;
 366#if IS_ENABLED(CONFIG_IPV6)
 367	case PF_INET6:
 368		sap = (struct sockaddr *)&sin6;
 369		break;
 370#endif
 371	default:
 372		return -EAFNOSUPPORT;
 373	}
 374
 375	return svc_xprt_create_from_sa(serv, xprt_name, net, sap, flags, cred);
 376}
 377EXPORT_SYMBOL_GPL(svc_xprt_create);
 378
 379/*
 380 * Copy the local and remote xprt addresses to the rqstp structure
 381 */
 382void svc_xprt_copy_addrs(struct svc_rqst *rqstp, struct svc_xprt *xprt)
 383{
 384	memcpy(&rqstp->rq_addr, &xprt->xpt_remote, xprt->xpt_remotelen);
 385	rqstp->rq_addrlen = xprt->xpt_remotelen;
 386
 387	/*
 388	 * Destination address in request is needed for binding the
 389	 * source address in RPC replies/callbacks later.
 390	 */
 391	memcpy(&rqstp->rq_daddr, &xprt->xpt_local, xprt->xpt_locallen);
 392	rqstp->rq_daddrlen = xprt->xpt_locallen;
 393}
 394EXPORT_SYMBOL_GPL(svc_xprt_copy_addrs);
 395
 396/**
 397 * svc_print_addr - Format rq_addr field for printing
 398 * @rqstp: svc_rqst struct containing address to print
 399 * @buf: target buffer for formatted address
 400 * @len: length of target buffer
 401 *
 402 */
 403char *svc_print_addr(struct svc_rqst *rqstp, char *buf, size_t len)
 404{
 405	return __svc_print_addr(svc_addr(rqstp), buf, len);
 406}
 407EXPORT_SYMBOL_GPL(svc_print_addr);
 408
 409static bool svc_xprt_slots_in_range(struct svc_xprt *xprt)
 410{
 411	unsigned int limit = svc_rpc_per_connection_limit;
 412	int nrqsts = atomic_read(&xprt->xpt_nr_rqsts);
 413
 414	return limit == 0 || (nrqsts >= 0 && nrqsts < limit);
 
 415}
 416
 417static bool svc_xprt_reserve_slot(struct svc_rqst *rqstp, struct svc_xprt *xprt)
 418{
 419	if (!test_bit(RQ_DATA, &rqstp->rq_flags)) {
 420		if (!svc_xprt_slots_in_range(xprt))
 421			return false;
 422		atomic_inc(&xprt->xpt_nr_rqsts);
 423		set_bit(RQ_DATA, &rqstp->rq_flags);
 424	}
 425	return true;
 426}
 427
 428static void svc_xprt_release_slot(struct svc_rqst *rqstp)
 429{
 430	struct svc_xprt	*xprt = rqstp->rq_xprt;
 431	if (test_and_clear_bit(RQ_DATA, &rqstp->rq_flags)) {
 432		atomic_dec(&xprt->xpt_nr_rqsts);
 433		smp_wmb(); /* See smp_rmb() in svc_xprt_ready() */
 434		svc_xprt_enqueue(xprt);
 
 
 
 
 
 435	}
 436}
 437
 438static bool svc_xprt_ready(struct svc_xprt *xprt)
 439{
 440	unsigned long xpt_flags;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 441
 442	/*
 443	 * If another cpu has recently updated xpt_flags,
 444	 * sk_sock->flags, xpt_reserved, or xpt_nr_rqsts, we need to
 445	 * know about it; otherwise it's possible that both that cpu and
 446	 * this one could call svc_xprt_enqueue() without either
 447	 * svc_xprt_enqueue() recognizing that the conditions below
 448	 * are satisfied, and we could stall indefinitely:
 449	 */
 450	smp_rmb();
 451	xpt_flags = READ_ONCE(xprt->xpt_flags);
 452
 453	trace_svc_xprt_enqueue(xprt, xpt_flags);
 454	if (xpt_flags & BIT(XPT_BUSY))
 455		return false;
 456	if (xpt_flags & (BIT(XPT_CONN) | BIT(XPT_CLOSE) | BIT(XPT_HANDSHAKE)))
 457		return true;
 458	if (xpt_flags & (BIT(XPT_DATA) | BIT(XPT_DEFERRED))) {
 459		if (xprt->xpt_ops->xpo_has_wspace(xprt) &&
 460		    svc_xprt_slots_in_range(xprt))
 461			return true;
 462		trace_svc_xprt_no_write_space(xprt);
 463		return false;
 464	}
 465	return false;
 
 
 
 466}
 
 467
 468/**
 469 * svc_xprt_enqueue - Queue a transport on an idle nfsd thread
 470 * @xprt: transport with data pending
 471 *
 472 */
 473void svc_xprt_enqueue(struct svc_xprt *xprt)
 474{
 475	struct svc_pool *pool;
 476
 477	if (!svc_xprt_ready(xprt))
 478		return;
 479
 480	/* Mark transport as busy. It will remain in this state until
 481	 * the provider calls svc_xprt_received. We update XPT_BUSY
 482	 * atomically because it also guards against trying to enqueue
 483	 * the transport twice.
 484	 */
 485	if (test_and_set_bit(XPT_BUSY, &xprt->xpt_flags))
 486		return;
 487
 488	pool = svc_pool_for_cpu(xprt->xpt_server);
 489
 490	percpu_counter_inc(&pool->sp_sockets_queued);
 491	lwq_enqueue(&xprt->xpt_ready, &pool->sp_xprts);
 492
 493	svc_pool_wake_idle_thread(pool);
 494}
 495EXPORT_SYMBOL_GPL(svc_xprt_enqueue);
 496
 497/*
 498 * Dequeue the first transport, if there is one.
 499 */
 500static struct svc_xprt *svc_xprt_dequeue(struct svc_pool *pool)
 501{
 502	struct svc_xprt	*xprt = NULL;
 503
 504	xprt = lwq_dequeue(&pool->sp_xprts, struct svc_xprt, xpt_ready);
 505	if (xprt)
 
 
 
 
 
 
 506		svc_xprt_get(xprt);
 
 
 
 
 
 
 
 507	return xprt;
 508}
 509
 510/**
 511 * svc_reserve - change the space reserved for the reply to a request.
 512 * @rqstp:  The request in question
 513 * @space: new max space to reserve
 514 *
 515 * Each request reserves some space on the output queue of the transport
 516 * to make sure the reply fits.  This function reduces that reserved
 517 * space to be the amount of space used already, plus @space.
 518 *
 519 */
 520void svc_reserve(struct svc_rqst *rqstp, int space)
 521{
 522	struct svc_xprt *xprt = rqstp->rq_xprt;
 523
 524	space += rqstp->rq_res.head[0].iov_len;
 525
 526	if (xprt && space < rqstp->rq_reserved) {
 
 527		atomic_sub((rqstp->rq_reserved - space), &xprt->xpt_reserved);
 528		rqstp->rq_reserved = space;
 529		smp_wmb(); /* See smp_rmb() in svc_xprt_ready() */
 
 
 530		svc_xprt_enqueue(xprt);
 531	}
 532}
 533EXPORT_SYMBOL_GPL(svc_reserve);
 534
 535static void free_deferred(struct svc_xprt *xprt, struct svc_deferred_req *dr)
 536{
 537	if (!dr)
 538		return;
 539
 540	xprt->xpt_ops->xpo_release_ctxt(xprt, dr->xprt_ctxt);
 541	kfree(dr);
 542}
 543
 544static void svc_xprt_release(struct svc_rqst *rqstp)
 545{
 546	struct svc_xprt	*xprt = rqstp->rq_xprt;
 547
 548	xprt->xpt_ops->xpo_release_ctxt(xprt, rqstp->rq_xprt_ctxt);
 549	rqstp->rq_xprt_ctxt = NULL;
 550
 551	free_deferred(xprt, rqstp->rq_deferred);
 552	rqstp->rq_deferred = NULL;
 553
 554	svc_rqst_release_pages(rqstp);
 555	rqstp->rq_res.page_len = 0;
 556	rqstp->rq_res.page_base = 0;
 557
 558	/* Reset response buffer and release
 559	 * the reservation.
 560	 * But first, check that enough space was reserved
 561	 * for the reply, otherwise we have a bug!
 562	 */
 563	if ((rqstp->rq_res.len) >  rqstp->rq_reserved)
 564		printk(KERN_ERR "RPC request reserved %d but used %d\n",
 565		       rqstp->rq_reserved,
 566		       rqstp->rq_res.len);
 567
 568	rqstp->rq_res.head[0].iov_len = 0;
 569	svc_reserve(rqstp, 0);
 570	svc_xprt_release_slot(rqstp);
 571	rqstp->rq_xprt = NULL;
 
 572	svc_xprt_put(xprt);
 573}
 574
 575/**
 576 * svc_wake_up - Wake up a service thread for non-transport work
 577 * @serv: RPC service
 578 *
 579 * Some svc_serv's will have occasional work to do, even when a xprt is not
 580 * waiting to be serviced. This function is there to "kick" a task in one of
 581 * those services so that it can wake up and do that work. Note that we only
 582 * bother with pool 0 as we don't need to wake up more than one thread for
 583 * this purpose.
 584 */
 585void svc_wake_up(struct svc_serv *serv)
 586{
 587	struct svc_pool *pool = &serv->sv_pools[0];
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 588
 
 589	set_bit(SP_TASK_PENDING, &pool->sp_flags);
 590	svc_pool_wake_idle_thread(pool);
 
 591}
 592EXPORT_SYMBOL_GPL(svc_wake_up);
 593
 594int svc_port_is_privileged(struct sockaddr *sin)
 595{
 596	switch (sin->sa_family) {
 597	case AF_INET:
 598		return ntohs(((struct sockaddr_in *)sin)->sin_port)
 599			< PROT_SOCK;
 600	case AF_INET6:
 601		return ntohs(((struct sockaddr_in6 *)sin)->sin6_port)
 602			< PROT_SOCK;
 603	default:
 604		return 0;
 605	}
 606}
 607
 608/*
 609 * Make sure that we don't have too many active connections. If we have,
 610 * something must be dropped. It's not clear what will happen if we allow
 611 * "too many" connections, but when dealing with network-facing software,
 612 * we have to code defensively. Here we do that by imposing hard limits.
 613 *
 614 * There's no point in trying to do random drop here for DoS
 615 * prevention. The NFS clients does 1 reconnect in 15 seconds. An
 616 * attacker can easily beat that.
 617 *
 618 * The only somewhat efficient mechanism would be if drop old
 619 * connections from the same IP first. But right now we don't even
 620 * record the client IP in svc_sock.
 621 *
 622 * single-threaded services that expect a lot of clients will probably
 623 * need to set sv_maxconn to override the default value which is based
 624 * on the number of threads
 625 */
 626static void svc_check_conn_limits(struct svc_serv *serv)
 627{
 628	unsigned int limit = serv->sv_maxconn ? serv->sv_maxconn :
 629				(serv->sv_nrthreads+3) * 20;
 630
 631	if (serv->sv_tmpcnt > limit) {
 632		struct svc_xprt *xprt = NULL;
 633		spin_lock_bh(&serv->sv_lock);
 634		if (!list_empty(&serv->sv_tempsocks)) {
 635			/* Try to help the admin */
 636			net_notice_ratelimited("%s: too many open connections, consider increasing the %s\n",
 637					       serv->sv_name, serv->sv_maxconn ?
 638					       "max number of connections" :
 639					       "number of threads");
 640			/*
 641			 * Always select the oldest connection. It's not fair,
 642			 * but so is life
 643			 */
 644			xprt = list_entry(serv->sv_tempsocks.prev,
 645					  struct svc_xprt,
 646					  xpt_list);
 647			set_bit(XPT_CLOSE, &xprt->xpt_flags);
 648			svc_xprt_get(xprt);
 649		}
 650		spin_unlock_bh(&serv->sv_lock);
 651
 652		if (xprt) {
 653			svc_xprt_enqueue(xprt);
 654			svc_xprt_put(xprt);
 655		}
 656	}
 657}
 658
 659static bool svc_alloc_arg(struct svc_rqst *rqstp)
 660{
 661	struct svc_serv *serv = rqstp->rq_server;
 662	struct xdr_buf *arg = &rqstp->rq_arg;
 663	unsigned long pages, filled, ret;
 
 664
 665	pages = (serv->sv_max_mesg + 2 * PAGE_SIZE) >> PAGE_SHIFT;
 666	if (pages > RPCSVC_MAXPAGES) {
 667		pr_warn_once("svc: warning: pages=%lu > RPCSVC_MAXPAGES=%lu\n",
 668			     pages, RPCSVC_MAXPAGES);
 669		/* use as many pages as possible */
 670		pages = RPCSVC_MAXPAGES;
 671	}
 672
 673	for (filled = 0; filled < pages; filled = ret) {
 674		ret = alloc_pages_bulk_array(GFP_KERNEL, pages,
 675					     rqstp->rq_pages);
 676		if (ret > filled)
 677			/* Made progress, don't sleep yet */
 678			continue;
 679
 680		set_current_state(TASK_IDLE);
 681		if (svc_thread_should_stop(rqstp)) {
 682			set_current_state(TASK_RUNNING);
 683			return false;
 684		}
 685		trace_svc_alloc_arg_err(pages, ret);
 686		memalloc_retry_wait(GFP_KERNEL);
 687	}
 688	rqstp->rq_page_end = &rqstp->rq_pages[pages];
 689	rqstp->rq_pages[pages] = NULL; /* this might be seen in nfsd_splice_actor() */
 690
 691	/* Make arg->head point to first page and arg->pages point to rest */
 
 692	arg->head[0].iov_base = page_address(rqstp->rq_pages[0]);
 693	arg->head[0].iov_len = PAGE_SIZE;
 694	arg->pages = rqstp->rq_pages + 1;
 695	arg->page_base = 0;
 696	/* save at least one page for response */
 697	arg->page_len = (pages-2)*PAGE_SIZE;
 698	arg->len = (pages-1)*PAGE_SIZE;
 699	arg->tail[0].iov_len = 0;
 700
 701	rqstp->rq_xid = xdr_zero;
 702	return true;
 703}
 704
 705static bool
 706svc_thread_should_sleep(struct svc_rqst *rqstp)
 707{
 708	struct svc_pool		*pool = rqstp->rq_pool;
 709
 710	/* did someone call svc_wake_up? */
 711	if (test_bit(SP_TASK_PENDING, &pool->sp_flags))
 712		return false;
 713
 714	/* was a socket queued? */
 715	if (!lwq_empty(&pool->sp_xprts))
 716		return false;
 717
 718	/* are we shutting down? */
 719	if (svc_thread_should_stop(rqstp))
 720		return false;
 721
 722#if defined(CONFIG_SUNRPC_BACKCHANNEL)
 723	if (svc_is_backchannel(rqstp)) {
 724		if (!lwq_empty(&rqstp->rq_server->sv_cb_list))
 725			return false;
 726	}
 727#endif
 728
 729	return true;
 730}
 731
 732static void svc_thread_wait_for_work(struct svc_rqst *rqstp)
 733{
 734	struct svc_pool *pool = rqstp->rq_pool;
 
 
 735
 736	if (svc_thread_should_sleep(rqstp)) {
 737		set_current_state(TASK_IDLE | TASK_FREEZABLE);
 738		llist_add(&rqstp->rq_idle, &pool->sp_idle_threads);
 739		if (likely(svc_thread_should_sleep(rqstp)))
 740			schedule();
 741
 742		while (!llist_del_first_this(&pool->sp_idle_threads,
 743					     &rqstp->rq_idle)) {
 744			/* Work just became available.  This thread can only
 745			 * handle it after removing rqstp from the idle
 746			 * list. If that attempt failed, some other thread
 747			 * must have queued itself after finding no
 748			 * work to do, so that thread has taken responsibly
 749			 * for this new work.  This thread can safely sleep
 750			 * until woken again.
 751			 */
 752			schedule();
 753			set_current_state(TASK_IDLE | TASK_FREEZABLE);
 754		}
 755		__set_current_state(TASK_RUNNING);
 756	} else {
 757		cond_resched();
 758	}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 759	try_to_freeze();
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 760}
 761
 762static void svc_add_new_temp_xprt(struct svc_serv *serv, struct svc_xprt *newxpt)
 763{
 764	spin_lock_bh(&serv->sv_lock);
 765	set_bit(XPT_TEMP, &newxpt->xpt_flags);
 766	list_add(&newxpt->xpt_list, &serv->sv_tempsocks);
 767	serv->sv_tmpcnt++;
 768	if (serv->sv_temptimer.function == NULL) {
 769		/* setup timer to age temp transports */
 770		serv->sv_temptimer.function = svc_age_temp_xprts;
 
 771		mod_timer(&serv->sv_temptimer,
 772			  jiffies + svc_conn_age_period * HZ);
 773	}
 774	spin_unlock_bh(&serv->sv_lock);
 775	svc_xprt_received(newxpt);
 776}
 777
 778static void svc_handle_xprt(struct svc_rqst *rqstp, struct svc_xprt *xprt)
 779{
 780	struct svc_serv *serv = rqstp->rq_server;
 781	int len = 0;
 782
 783	if (test_bit(XPT_CLOSE, &xprt->xpt_flags)) {
 784		if (test_and_clear_bit(XPT_KILL_TEMP, &xprt->xpt_flags))
 785			xprt->xpt_ops->xpo_kill_temp_xprt(xprt);
 786		svc_delete_xprt(xprt);
 787		/* Leave XPT_BUSY set on the dead xprt: */
 788		goto out;
 789	}
 790	if (test_bit(XPT_LISTENER, &xprt->xpt_flags)) {
 791		struct svc_xprt *newxpt;
 792		/*
 793		 * We know this module_get will succeed because the
 794		 * listener holds a reference too
 795		 */
 796		__module_get(xprt->xpt_class->xcl_owner);
 797		svc_check_conn_limits(xprt->xpt_server);
 798		newxpt = xprt->xpt_ops->xpo_accept(xprt);
 799		if (newxpt) {
 800			newxpt->xpt_cred = get_cred(xprt->xpt_cred);
 801			svc_add_new_temp_xprt(serv, newxpt);
 802			trace_svc_xprt_accept(newxpt, serv->sv_name);
 803		} else {
 804			module_put(xprt->xpt_class->xcl_owner);
 805		}
 806		svc_xprt_received(xprt);
 807	} else if (test_bit(XPT_HANDSHAKE, &xprt->xpt_flags)) {
 808		xprt->xpt_ops->xpo_handshake(xprt);
 809		svc_xprt_received(xprt);
 810	} else if (svc_xprt_reserve_slot(rqstp, xprt)) {
 811		/* XPT_DATA|XPT_DEFERRED case: */
 
 
 
 812		rqstp->rq_deferred = svc_deferred_dequeue(xprt);
 813		if (rqstp->rq_deferred)
 814			len = svc_deferred_recv(rqstp);
 815		else
 816			len = xprt->xpt_ops->xpo_recvfrom(rqstp);
 
 817		rqstp->rq_reserved = serv->sv_max_mesg;
 818		atomic_add(rqstp->rq_reserved, &xprt->xpt_reserved);
 819		if (len <= 0)
 820			goto out;
 821
 822		trace_svc_xdr_recvfrom(&rqstp->rq_arg);
 823
 824		clear_bit(XPT_OLD, &xprt->xpt_flags);
 825
 826		rqstp->rq_chandle.defer = svc_defer;
 827
 828		if (serv->sv_stats)
 829			serv->sv_stats->netcnt++;
 830		percpu_counter_inc(&rqstp->rq_pool->sp_messages_arrived);
 831		rqstp->rq_stime = ktime_get();
 832		svc_process(rqstp);
 833	} else
 834		svc_xprt_received(xprt);
 835
 836out:
 837	rqstp->rq_res.len = 0;
 838	svc_xprt_release(rqstp);
 839}
 840
 841static void svc_thread_wake_next(struct svc_rqst *rqstp)
 842{
 843	if (!svc_thread_should_sleep(rqstp))
 844		/* More work pending after I dequeued some,
 845		 * wake another worker
 846		 */
 847		svc_pool_wake_idle_thread(rqstp->rq_pool);
 848}
 849
 850/**
 851 * svc_recv - Receive and process the next request on any transport
 852 * @rqstp: an idle RPC service thread
 853 *
 854 * This code is carefully organised not to touch any cachelines in
 855 * the shared svc_serv structure, only cachelines in the local
 856 * svc_pool.
 857 */
 858void svc_recv(struct svc_rqst *rqstp)
 859{
 860	struct svc_pool *pool = rqstp->rq_pool;
 
 
 861
 862	if (!svc_alloc_arg(rqstp))
 863		return;
 864
 865	svc_thread_wait_for_work(rqstp);
 
 
 
 866
 867	clear_bit(SP_TASK_PENDING, &pool->sp_flags);
 868
 869	if (svc_thread_should_stop(rqstp)) {
 870		svc_thread_wake_next(rqstp);
 871		return;
 872	}
 873
 874	rqstp->rq_xprt = svc_xprt_dequeue(pool);
 875	if (rqstp->rq_xprt) {
 876		struct svc_xprt *xprt = rqstp->rq_xprt;
 877
 878		svc_thread_wake_next(rqstp);
 879		/* Normally we will wait up to 5 seconds for any required
 880		 * cache information to be provided.  When there are no
 881		 * idle threads, we reduce the wait time.
 882		 */
 883		if (pool->sp_idle_threads.first)
 884			rqstp->rq_chandle.thread_wait = 5 * HZ;
 885		else
 886			rqstp->rq_chandle.thread_wait = 1 * HZ;
 887
 888		trace_svc_xprt_dequeue(rqstp);
 889		svc_handle_xprt(rqstp, xprt);
 
 
 890	}
 891
 892#if defined(CONFIG_SUNRPC_BACKCHANNEL)
 893	if (svc_is_backchannel(rqstp)) {
 894		struct svc_serv *serv = rqstp->rq_server;
 895		struct rpc_rqst *req;
 896
 897		req = lwq_dequeue(&serv->sv_cb_list,
 898				  struct rpc_rqst, rq_bc_list);
 899		if (req) {
 900			svc_thread_wake_next(rqstp);
 901			svc_process_bc(req, rqstp);
 902		}
 903	}
 904#endif
 
 
 
 
 
 
 
 
 
 
 
 
 
 905}
 906EXPORT_SYMBOL_GPL(svc_recv);
 907
 908/**
 909 * svc_send - Return reply to client
 910 * @rqstp: RPC transaction context
 911 *
 912 */
 913void svc_send(struct svc_rqst *rqstp)
 
 
 
 
 
 
 
 
 
 
 914{
 915	struct svc_xprt	*xprt;
 
 916	struct xdr_buf	*xb;
 917	int status;
 918
 919	xprt = rqstp->rq_xprt;
 
 
 
 
 
 920
 921	/* calculate over-all length */
 922	xb = &rqstp->rq_res;
 923	xb->len = xb->head[0].iov_len +
 924		xb->page_len +
 925		xb->tail[0].iov_len;
 926	trace_svc_xdr_sendto(rqstp->rq_xid, xb);
 927	trace_svc_stats_latency(rqstp);
 928
 929	status = xprt->xpt_ops->xpo_sendto(rqstp);
 
 
 
 
 
 
 
 
 
 930
 931	trace_svc_send(rqstp, status);
 
 
 
 
 932}
 933
 934/*
 935 * Timer function to close old temporary transports, using
 936 * a mark-and-sweep algorithm.
 937 */
 938static void svc_age_temp_xprts(struct timer_list *t)
 939{
 940	struct svc_serv *serv = from_timer(serv, t, sv_temptimer);
 941	struct svc_xprt *xprt;
 942	struct list_head *le, *next;
 943
 944	dprintk("svc_age_temp_xprts\n");
 945
 946	if (!spin_trylock_bh(&serv->sv_lock)) {
 947		/* busy, try again 1 sec later */
 948		dprintk("svc_age_temp_xprts: busy\n");
 949		mod_timer(&serv->sv_temptimer, jiffies + HZ);
 950		return;
 951	}
 952
 953	list_for_each_safe(le, next, &serv->sv_tempsocks) {
 954		xprt = list_entry(le, struct svc_xprt, xpt_list);
 955
 956		/* First time through, just mark it OLD. Second time
 957		 * through, close it. */
 958		if (!test_and_set_bit(XPT_OLD, &xprt->xpt_flags))
 959			continue;
 960		if (kref_read(&xprt->xpt_ref) > 1 ||
 961		    test_bit(XPT_BUSY, &xprt->xpt_flags))
 962			continue;
 963		list_del_init(le);
 964		set_bit(XPT_CLOSE, &xprt->xpt_flags);
 965		dprintk("queuing xprt %p for closing\n", xprt);
 966
 967		/* a thread will dequeue and close it soon */
 968		svc_xprt_enqueue(xprt);
 969	}
 970	spin_unlock_bh(&serv->sv_lock);
 971
 972	mod_timer(&serv->sv_temptimer, jiffies + svc_conn_age_period * HZ);
 973}
 974
 975/* Close temporary transports whose xpt_local matches server_addr immediately
 976 * instead of waiting for them to be picked up by the timer.
 977 *
 978 * This is meant to be called from a notifier_block that runs when an ip
 979 * address is deleted.
 980 */
 981void svc_age_temp_xprts_now(struct svc_serv *serv, struct sockaddr *server_addr)
 982{
 983	struct svc_xprt *xprt;
 
 
 984	struct list_head *le, *next;
 985	LIST_HEAD(to_be_closed);
 
 
 
 
 986
 987	spin_lock_bh(&serv->sv_lock);
 988	list_for_each_safe(le, next, &serv->sv_tempsocks) {
 989		xprt = list_entry(le, struct svc_xprt, xpt_list);
 990		if (rpc_cmp_addr(server_addr, (struct sockaddr *)
 991				&xprt->xpt_local)) {
 992			dprintk("svc_age_temp_xprts_now: found %p\n", xprt);
 993			list_move(le, &to_be_closed);
 994		}
 995	}
 996	spin_unlock_bh(&serv->sv_lock);
 997
 998	while (!list_empty(&to_be_closed)) {
 999		le = to_be_closed.next;
1000		list_del_init(le);
1001		xprt = list_entry(le, struct svc_xprt, xpt_list);
1002		set_bit(XPT_CLOSE, &xprt->xpt_flags);
1003		set_bit(XPT_KILL_TEMP, &xprt->xpt_flags);
1004		dprintk("svc_age_temp_xprts_now: queuing xprt %p for closing\n",
1005				xprt);
1006		svc_xprt_enqueue(xprt);
 
1007	}
1008}
1009EXPORT_SYMBOL_GPL(svc_age_temp_xprts_now);
1010
1011static void call_xpt_users(struct svc_xprt *xprt)
1012{
1013	struct svc_xpt_user *u;
1014
1015	spin_lock(&xprt->xpt_lock);
1016	while (!list_empty(&xprt->xpt_users)) {
1017		u = list_first_entry(&xprt->xpt_users, struct svc_xpt_user, list);
1018		list_del_init(&u->list);
1019		u->callback(u);
1020	}
1021	spin_unlock(&xprt->xpt_lock);
1022}
1023
1024/*
1025 * Remove a dead transport
1026 */
1027static void svc_delete_xprt(struct svc_xprt *xprt)
1028{
1029	struct svc_serv	*serv = xprt->xpt_server;
1030	struct svc_deferred_req *dr;
1031
 
1032	if (test_and_set_bit(XPT_DEAD, &xprt->xpt_flags))
1033		return;
1034
1035	trace_svc_xprt_detach(xprt);
1036	xprt->xpt_ops->xpo_detach(xprt);
1037	if (xprt->xpt_bc_xprt)
1038		xprt->xpt_bc_xprt->ops->close(xprt->xpt_bc_xprt);
1039
1040	spin_lock_bh(&serv->sv_lock);
1041	list_del_init(&xprt->xpt_list);
 
1042	if (test_bit(XPT_TEMP, &xprt->xpt_flags))
1043		serv->sv_tmpcnt--;
1044	spin_unlock_bh(&serv->sv_lock);
1045
1046	while ((dr = svc_deferred_dequeue(xprt)) != NULL)
1047		free_deferred(xprt, dr);
1048
1049	call_xpt_users(xprt);
1050	svc_xprt_put(xprt);
1051}
1052
1053/**
1054 * svc_xprt_close - Close a client connection
1055 * @xprt: transport to disconnect
1056 *
1057 */
1058void svc_xprt_close(struct svc_xprt *xprt)
1059{
1060	trace_svc_xprt_close(xprt);
1061	set_bit(XPT_CLOSE, &xprt->xpt_flags);
1062	if (test_and_set_bit(XPT_BUSY, &xprt->xpt_flags))
1063		/* someone else will have to effect the close */
1064		return;
1065	/*
1066	 * We expect svc_close_xprt() to work even when no threads are
1067	 * running (e.g., while configuring the server before starting
1068	 * any threads), so if the transport isn't busy, we delete
1069	 * it ourself:
1070	 */
1071	svc_delete_xprt(xprt);
1072}
1073EXPORT_SYMBOL_GPL(svc_xprt_close);
1074
1075static int svc_close_list(struct svc_serv *serv, struct list_head *xprt_list, struct net *net)
1076{
1077	struct svc_xprt *xprt;
1078	int ret = 0;
1079
1080	spin_lock_bh(&serv->sv_lock);
1081	list_for_each_entry(xprt, xprt_list, xpt_list) {
1082		if (xprt->xpt_net != net)
1083			continue;
1084		ret++;
1085		set_bit(XPT_CLOSE, &xprt->xpt_flags);
1086		svc_xprt_enqueue(xprt);
1087	}
1088	spin_unlock_bh(&serv->sv_lock);
1089	return ret;
1090}
1091
1092static void svc_clean_up_xprts(struct svc_serv *serv, struct net *net)
1093{
 
1094	struct svc_xprt *xprt;
 
1095	int i;
1096
1097	for (i = 0; i < serv->sv_nrpools; i++) {
1098		struct svc_pool *pool = &serv->sv_pools[i];
1099		struct llist_node *q, **t1, *t2;
1100
1101		q = lwq_dequeue_all(&pool->sp_xprts);
1102		lwq_for_each_safe(xprt, t1, t2, &q, xpt_ready) {
1103			if (xprt->xpt_net == net) {
1104				set_bit(XPT_CLOSE, &xprt->xpt_flags);
1105				svc_delete_xprt(xprt);
1106				xprt = NULL;
1107			}
1108		}
 
 
 
 
1109
1110		if (q)
1111			lwq_enqueue_batch(q, &pool->sp_xprts);
 
 
 
 
 
1112	}
1113}
1114
1115/**
1116 * svc_xprt_destroy_all - Destroy transports associated with @serv
1117 * @serv: RPC service to be shut down
1118 * @net: target network namespace
1119 *
1120 * Server threads may still be running (especially in the case where the
1121 * service is still running in other network namespaces).
1122 *
1123 * So we shut down sockets the same way we would on a running server, by
1124 * setting XPT_CLOSE, enqueuing, and letting a thread pick it up to do
1125 * the close.  In the case there are no such other threads,
1126 * threads running, svc_clean_up_xprts() does a simple version of a
1127 * server's main event loop, and in the case where there are other
1128 * threads, we may need to wait a little while and then check again to
1129 * see if they're done.
1130 */
1131void svc_xprt_destroy_all(struct svc_serv *serv, struct net *net)
1132{
1133	int delay = 0;
1134
1135	while (svc_close_list(serv, &serv->sv_permsocks, net) +
1136	       svc_close_list(serv, &serv->sv_tempsocks, net)) {
1137
1138		svc_clean_up_xprts(serv, net);
1139		msleep(delay++);
1140	}
1141}
1142EXPORT_SYMBOL_GPL(svc_xprt_destroy_all);
1143
1144/*
1145 * Handle defer and revisit of requests
1146 */
1147
1148static void svc_revisit(struct cache_deferred_req *dreq, int too_many)
1149{
1150	struct svc_deferred_req *dr =
1151		container_of(dreq, struct svc_deferred_req, handle);
1152	struct svc_xprt *xprt = dr->xprt;
1153
1154	spin_lock(&xprt->xpt_lock);
1155	set_bit(XPT_DEFERRED, &xprt->xpt_flags);
1156	if (too_many || test_bit(XPT_DEAD, &xprt->xpt_flags)) {
1157		spin_unlock(&xprt->xpt_lock);
1158		trace_svc_defer_drop(dr);
1159		free_deferred(xprt, dr);
1160		svc_xprt_put(xprt);
 
1161		return;
1162	}
 
1163	dr->xprt = NULL;
1164	list_add(&dr->handle.recent, &xprt->xpt_deferred);
1165	spin_unlock(&xprt->xpt_lock);
1166	trace_svc_defer_queue(dr);
1167	svc_xprt_enqueue(xprt);
1168	svc_xprt_put(xprt);
1169}
1170
1171/*
1172 * Save the request off for later processing. The request buffer looks
1173 * like this:
1174 *
1175 * <xprt-header><rpc-header><rpc-pagelist><rpc-tail>
1176 *
1177 * This code can only handle requests that consist of an xprt-header
1178 * and rpc-header.
1179 */
1180static struct cache_deferred_req *svc_defer(struct cache_req *req)
1181{
1182	struct svc_rqst *rqstp = container_of(req, struct svc_rqst, rq_chandle);
1183	struct svc_deferred_req *dr;
1184
1185	if (rqstp->rq_arg.page_len || !test_bit(RQ_USEDEFERRAL, &rqstp->rq_flags))
1186		return NULL; /* if more than a page, give up FIXME */
1187	if (rqstp->rq_deferred) {
1188		dr = rqstp->rq_deferred;
1189		rqstp->rq_deferred = NULL;
1190	} else {
1191		size_t skip;
1192		size_t size;
1193		/* FIXME maybe discard if size too large */
1194		size = sizeof(struct svc_deferred_req) + rqstp->rq_arg.len;
1195		dr = kmalloc(size, GFP_KERNEL);
1196		if (dr == NULL)
1197			return NULL;
1198
1199		dr->handle.owner = rqstp->rq_server;
1200		dr->prot = rqstp->rq_prot;
1201		memcpy(&dr->addr, &rqstp->rq_addr, rqstp->rq_addrlen);
1202		dr->addrlen = rqstp->rq_addrlen;
1203		dr->daddr = rqstp->rq_daddr;
1204		dr->argslen = rqstp->rq_arg.len >> 2;
 
1205
1206		/* back up head to the start of the buffer and copy */
1207		skip = rqstp->rq_arg.len - rqstp->rq_arg.head[0].iov_len;
1208		memcpy(dr->args, rqstp->rq_arg.head[0].iov_base - skip,
1209		       dr->argslen << 2);
1210	}
1211	dr->xprt_ctxt = rqstp->rq_xprt_ctxt;
1212	rqstp->rq_xprt_ctxt = NULL;
1213	trace_svc_defer(rqstp);
1214	svc_xprt_get(rqstp->rq_xprt);
1215	dr->xprt = rqstp->rq_xprt;
1216	set_bit(RQ_DROPME, &rqstp->rq_flags);
1217
1218	dr->handle.revisit = svc_revisit;
1219	return &dr->handle;
1220}
1221
1222/*
1223 * recv data from a deferred request into an active one
1224 */
1225static noinline int svc_deferred_recv(struct svc_rqst *rqstp)
1226{
1227	struct svc_deferred_req *dr = rqstp->rq_deferred;
1228
1229	trace_svc_defer_recv(dr);
1230
1231	/* setup iov_base past transport header */
1232	rqstp->rq_arg.head[0].iov_base = dr->args;
1233	/* The iov_len does not include the transport header bytes */
1234	rqstp->rq_arg.head[0].iov_len = dr->argslen << 2;
1235	rqstp->rq_arg.page_len = 0;
1236	/* The rq_arg.len includes the transport header bytes */
1237	rqstp->rq_arg.len     = dr->argslen << 2;
1238	rqstp->rq_prot        = dr->prot;
1239	memcpy(&rqstp->rq_addr, &dr->addr, dr->addrlen);
1240	rqstp->rq_addrlen     = dr->addrlen;
1241	/* Save off transport header len in case we get deferred again */
 
1242	rqstp->rq_daddr       = dr->daddr;
1243	rqstp->rq_respages    = rqstp->rq_pages;
1244	rqstp->rq_xprt_ctxt   = dr->xprt_ctxt;
1245
1246	dr->xprt_ctxt = NULL;
1247	svc_xprt_received(rqstp->rq_xprt);
1248	return dr->argslen << 2;
1249}
1250
1251
1252static struct svc_deferred_req *svc_deferred_dequeue(struct svc_xprt *xprt)
1253{
1254	struct svc_deferred_req *dr = NULL;
1255
1256	if (!test_bit(XPT_DEFERRED, &xprt->xpt_flags))
1257		return NULL;
1258	spin_lock(&xprt->xpt_lock);
1259	if (!list_empty(&xprt->xpt_deferred)) {
1260		dr = list_entry(xprt->xpt_deferred.next,
1261				struct svc_deferred_req,
1262				handle.recent);
1263		list_del_init(&dr->handle.recent);
1264	} else
1265		clear_bit(XPT_DEFERRED, &xprt->xpt_flags);
1266	spin_unlock(&xprt->xpt_lock);
1267	return dr;
1268}
1269
1270/**
1271 * svc_find_listener - find an RPC transport instance
1272 * @serv: pointer to svc_serv to search
1273 * @xcl_name: C string containing transport's class name
1274 * @net: owner net pointer
1275 * @sa: sockaddr containing address
1276 *
1277 * Return the transport instance pointer for the endpoint accepting
1278 * connections/peer traffic from the specified transport class,
1279 * and matching sockaddr.
1280 */
1281struct svc_xprt *svc_find_listener(struct svc_serv *serv, const char *xcl_name,
1282				   struct net *net, const struct sockaddr *sa)
1283{
1284	struct svc_xprt *xprt;
1285	struct svc_xprt *found = NULL;
1286
1287	spin_lock_bh(&serv->sv_lock);
1288	list_for_each_entry(xprt, &serv->sv_permsocks, xpt_list) {
1289		if (xprt->xpt_net != net)
1290			continue;
1291		if (strcmp(xprt->xpt_class->xcl_name, xcl_name))
1292			continue;
1293		if (!rpc_cmp_addr_port(sa, (struct sockaddr *)&xprt->xpt_local))
1294			continue;
1295		found = xprt;
1296		svc_xprt_get(xprt);
1297		break;
1298	}
1299	spin_unlock_bh(&serv->sv_lock);
1300	return found;
1301}
1302EXPORT_SYMBOL_GPL(svc_find_listener);
1303
1304/**
1305 * svc_find_xprt - find an RPC transport instance
1306 * @serv: pointer to svc_serv to search
1307 * @xcl_name: C string containing transport's class name
1308 * @net: owner net pointer
1309 * @af: Address family of transport's local address
1310 * @port: transport's IP port number
1311 *
1312 * Return the transport instance pointer for the endpoint accepting
1313 * connections/peer traffic from the specified transport class,
1314 * address family and port.
1315 *
1316 * Specifying 0 for the address family or port is effectively a
1317 * wild-card, and will result in matching the first transport in the
1318 * service's list that has a matching class name.
1319 */
1320struct svc_xprt *svc_find_xprt(struct svc_serv *serv, const char *xcl_name,
1321			       struct net *net, const sa_family_t af,
1322			       const unsigned short port)
1323{
1324	struct svc_xprt *xprt;
1325	struct svc_xprt *found = NULL;
1326
1327	/* Sanity check the args */
1328	if (serv == NULL || xcl_name == NULL)
1329		return found;
1330
1331	spin_lock_bh(&serv->sv_lock);
1332	list_for_each_entry(xprt, &serv->sv_permsocks, xpt_list) {
1333		if (xprt->xpt_net != net)
1334			continue;
1335		if (strcmp(xprt->xpt_class->xcl_name, xcl_name))
1336			continue;
1337		if (af != AF_UNSPEC && af != xprt->xpt_local.ss_family)
1338			continue;
1339		if (port != 0 && port != svc_xprt_local_port(xprt))
1340			continue;
1341		found = xprt;
1342		svc_xprt_get(xprt);
1343		break;
1344	}
1345	spin_unlock_bh(&serv->sv_lock);
1346	return found;
1347}
1348EXPORT_SYMBOL_GPL(svc_find_xprt);
1349
1350static int svc_one_xprt_name(const struct svc_xprt *xprt,
1351			     char *pos, int remaining)
1352{
1353	int len;
1354
1355	len = snprintf(pos, remaining, "%s %u\n",
1356			xprt->xpt_class->xcl_name,
1357			svc_xprt_local_port(xprt));
1358	if (len >= remaining)
1359		return -ENAMETOOLONG;
1360	return len;
1361}
1362
1363/**
1364 * svc_xprt_names - format a buffer with a list of transport names
1365 * @serv: pointer to an RPC service
1366 * @buf: pointer to a buffer to be filled in
1367 * @buflen: length of buffer to be filled in
1368 *
1369 * Fills in @buf with a string containing a list of transport names,
1370 * each name terminated with '\n'.
1371 *
1372 * Returns positive length of the filled-in string on success; otherwise
1373 * a negative errno value is returned if an error occurs.
1374 */
1375int svc_xprt_names(struct svc_serv *serv, char *buf, const int buflen)
1376{
1377	struct svc_xprt *xprt;
1378	int len, totlen;
1379	char *pos;
1380
1381	/* Sanity check args */
1382	if (!serv)
1383		return 0;
1384
1385	spin_lock_bh(&serv->sv_lock);
1386
1387	pos = buf;
1388	totlen = 0;
1389	list_for_each_entry(xprt, &serv->sv_permsocks, xpt_list) {
1390		len = svc_one_xprt_name(xprt, pos, buflen - totlen);
1391		if (len < 0) {
1392			*buf = '\0';
1393			totlen = len;
1394		}
1395		if (len <= 0)
1396			break;
1397
1398		pos += len;
1399		totlen += len;
1400	}
1401
1402	spin_unlock_bh(&serv->sv_lock);
1403	return totlen;
1404}
1405EXPORT_SYMBOL_GPL(svc_xprt_names);
1406
 
1407/*----------------------------------------------------------------------------*/
1408
1409static void *svc_pool_stats_start(struct seq_file *m, loff_t *pos)
1410{
1411	unsigned int pidx = (unsigned int)*pos;
1412	struct svc_info *si = m->private;
1413
1414	dprintk("svc_pool_stats_start, *pidx=%u\n", pidx);
1415
1416	mutex_lock(si->mutex);
1417
1418	if (!pidx)
1419		return SEQ_START_TOKEN;
1420	if (!si->serv)
1421		return NULL;
1422	return pidx > si->serv->sv_nrpools ? NULL
1423		: &si->serv->sv_pools[pidx - 1];
1424}
1425
1426static void *svc_pool_stats_next(struct seq_file *m, void *p, loff_t *pos)
1427{
1428	struct svc_pool *pool = p;
1429	struct svc_info *si = m->private;
1430	struct svc_serv *serv = si->serv;
1431
1432	dprintk("svc_pool_stats_next, *pos=%llu\n", *pos);
1433
1434	if (!serv) {
1435		pool = NULL;
1436	} else if (p == SEQ_START_TOKEN) {
1437		pool = &serv->sv_pools[0];
1438	} else {
1439		unsigned int pidx = (pool - &serv->sv_pools[0]);
1440		if (pidx < serv->sv_nrpools-1)
1441			pool = &serv->sv_pools[pidx+1];
1442		else
1443			pool = NULL;
1444	}
1445	++*pos;
1446	return pool;
1447}
1448
1449static void svc_pool_stats_stop(struct seq_file *m, void *p)
1450{
1451	struct svc_info *si = m->private;
1452
1453	mutex_unlock(si->mutex);
1454}
1455
1456static int svc_pool_stats_show(struct seq_file *m, void *p)
1457{
1458	struct svc_pool *pool = p;
1459
1460	if (p == SEQ_START_TOKEN) {
1461		seq_puts(m, "# pool packets-arrived sockets-enqueued threads-woken threads-timedout\n");
1462		return 0;
1463	}
1464
1465	seq_printf(m, "%u %llu %llu %llu 0\n",
1466		   pool->sp_id,
1467		   percpu_counter_sum_positive(&pool->sp_messages_arrived),
1468		   percpu_counter_sum_positive(&pool->sp_sockets_queued),
1469		   percpu_counter_sum_positive(&pool->sp_threads_woken));
 
1470
1471	return 0;
1472}
1473
1474static const struct seq_operations svc_pool_stats_seq_ops = {
1475	.start	= svc_pool_stats_start,
1476	.next	= svc_pool_stats_next,
1477	.stop	= svc_pool_stats_stop,
1478	.show	= svc_pool_stats_show,
1479};
1480
1481int svc_pool_stats_open(struct svc_info *info, struct file *file)
1482{
1483	struct seq_file *seq;
1484	int err;
1485
1486	err = seq_open(file, &svc_pool_stats_seq_ops);
1487	if (err)
1488		return err;
1489	seq = file->private_data;
1490	seq->private = info;
1491
1492	return 0;
1493}
1494EXPORT_SYMBOL(svc_pool_stats_open);
1495
1496/*----------------------------------------------------------------------------*/