svc_xprt.c

   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	xprt->xpt_qtime = ktime_get();
 492	lwq_enqueue(&xprt->xpt_ready, &pool->sp_xprts);
 493
 494	svc_pool_wake_idle_thread(pool);
 495}
 496EXPORT_SYMBOL_GPL(svc_xprt_enqueue);
 497
 498/*
 499 * Dequeue the first transport, if there is one.
 500 */
 501static struct svc_xprt *svc_xprt_dequeue(struct svc_pool *pool)
 502{
 503	struct svc_xprt	*xprt = NULL;
 504
 505	xprt = lwq_dequeue(&pool->sp_xprts, struct svc_xprt, xpt_ready);
 506	if (xprt)
 507		svc_xprt_get(xprt);
 508	return xprt;
 509}
 510
 511/**
 512 * svc_reserve - change the space reserved for the reply to a request.
 513 * @rqstp:  The request in question
 514 * @space: new max space to reserve
 515 *
 516 * Each request reserves some space on the output queue of the transport
 517 * to make sure the reply fits.  This function reduces that reserved
 518 * space to be the amount of space used already, plus @space.
 519 *
 520 */
 521void svc_reserve(struct svc_rqst *rqstp, int space)
 522{
 523	struct svc_xprt *xprt = rqstp->rq_xprt;
 524
 525	space += rqstp->rq_res.head[0].iov_len;
 526
 527	if (xprt && space < rqstp->rq_reserved) {
 528		atomic_sub((rqstp->rq_reserved - space), &xprt->xpt_reserved);
 529		rqstp->rq_reserved = space;
 530		smp_wmb(); /* See smp_rmb() in svc_xprt_ready() */
 531		svc_xprt_enqueue(xprt);
 532	}
 533}
 534EXPORT_SYMBOL_GPL(svc_reserve);
 535
 536static void free_deferred(struct svc_xprt *xprt, struct svc_deferred_req *dr)
 537{
 538	if (!dr)
 539		return;
 540
 541	xprt->xpt_ops->xpo_release_ctxt(xprt, dr->xprt_ctxt);
 542	kfree(dr);
 543}
 544
 545static void svc_xprt_release(struct svc_rqst *rqstp)
 546{
 547	struct svc_xprt	*xprt = rqstp->rq_xprt;
 548
 549	xprt->xpt_ops->xpo_release_ctxt(xprt, rqstp->rq_xprt_ctxt);
 550	rqstp->rq_xprt_ctxt = NULL;
 551
 552	free_deferred(xprt, rqstp->rq_deferred);
 553	rqstp->rq_deferred = NULL;
 554
 555	svc_rqst_release_pages(rqstp);
 556	rqstp->rq_res.page_len = 0;
 557	rqstp->rq_res.page_base = 0;
 558
 559	/* Reset response buffer and release
 560	 * the reservation.
 561	 * But first, check that enough space was reserved
 562	 * for the reply, otherwise we have a bug!
 563	 */
 564	if ((rqstp->rq_res.len) >  rqstp->rq_reserved)
 565		printk(KERN_ERR "RPC request reserved %d but used %d\n",
 566		       rqstp->rq_reserved,
 567		       rqstp->rq_res.len);
 568
 569	rqstp->rq_res.head[0].iov_len = 0;
 570	svc_reserve(rqstp, 0);
 571	svc_xprt_release_slot(rqstp);
 572	rqstp->rq_xprt = NULL;
 573	svc_xprt_put(xprt);
 574}
 575
 576/**
 577 * svc_wake_up - Wake up a service thread for non-transport work
 578 * @serv: RPC service
 579 *
 580 * Some svc_serv's will have occasional work to do, even when a xprt is not
 581 * waiting to be serviced. This function is there to "kick" a task in one of
 582 * those services so that it can wake up and do that work. Note that we only
 583 * bother with pool 0 as we don't need to wake up more than one thread for
 584 * this purpose.
 585 */
 586void svc_wake_up(struct svc_serv *serv)
 587{
 588	struct svc_pool *pool = &serv->sv_pools[0];
 589
 590	set_bit(SP_TASK_PENDING, &pool->sp_flags);
 591	svc_pool_wake_idle_thread(pool);
 592}
 593EXPORT_SYMBOL_GPL(svc_wake_up);
 594
 595int svc_port_is_privileged(struct sockaddr *sin)
 596{
 597	switch (sin->sa_family) {
 598	case AF_INET:
 599		return ntohs(((struct sockaddr_in *)sin)->sin_port)
 600			< PROT_SOCK;
 601	case AF_INET6:
 602		return ntohs(((struct sockaddr_in6 *)sin)->sin6_port)
 603			< PROT_SOCK;
 604	default:
 605		return 0;
 606	}
 607}
 608
 609/*
 610 * Make sure that we don't have too many connections that have not yet
 611 * demonstrated that they have access to the NFS server. If we have,
 612 * something must be dropped. It's not clear what will happen if we allow
 613 * "too many" connections, but when dealing with network-facing software,
 614 * we have to code defensively. Here we do that by imposing hard limits.
 615 *
 616 * There's no point in trying to do random drop here for DoS
 617 * prevention. The NFS clients does 1 reconnect in 15 seconds. An
 618 * attacker can easily beat that.
 619 *
 620 * The only somewhat efficient mechanism would be if drop old
 621 * connections from the same IP first. But right now we don't even
 622 * record the client IP in svc_sock.
 623 */
 624static void svc_check_conn_limits(struct svc_serv *serv)
 625{
 626	if (serv->sv_tmpcnt > XPT_MAX_TMP_CONN) {
 627		struct svc_xprt *xprt = NULL, *xprti;
 628		spin_lock_bh(&serv->sv_lock);
 629		if (!list_empty(&serv->sv_tempsocks)) {
 630			/*
 631			 * Always select the oldest connection. It's not fair,
 632			 * but nor is life.
 633			 */
 634			list_for_each_entry_reverse(xprti, &serv->sv_tempsocks,
 635						    xpt_list) {
 636				if (!test_bit(XPT_PEER_VALID, &xprti->xpt_flags)) {
 637					xprt = xprti;
 638					set_bit(XPT_CLOSE, &xprt->xpt_flags);
 639					svc_xprt_get(xprt);
 640					break;
 641				}
 642			}
 643		}
 644		spin_unlock_bh(&serv->sv_lock);
 645
 646		if (xprt) {
 647			svc_xprt_enqueue(xprt);
 648			svc_xprt_put(xprt);
 649		}
 650	}
 651}
 652
 653static bool svc_alloc_arg(struct svc_rqst *rqstp)
 654{
 655	struct xdr_buf *arg = &rqstp->rq_arg;
 656	unsigned long pages, filled, ret;
 657
 658	pages = rqstp->rq_maxpages;
 659	for (filled = 0; filled < pages; filled = ret) {
 660		ret = alloc_pages_bulk(GFP_KERNEL, pages, rqstp->rq_pages);
 661		if (ret > filled)
 662			/* Made progress, don't sleep yet */
 663			continue;
 664
 665		set_current_state(TASK_IDLE);
 666		if (svc_thread_should_stop(rqstp)) {
 667			set_current_state(TASK_RUNNING);
 668			return false;
 669		}
 670		trace_svc_alloc_arg_err(pages, ret);
 671		memalloc_retry_wait(GFP_KERNEL);
 672	}
 673	rqstp->rq_page_end = &rqstp->rq_pages[pages];
 674	rqstp->rq_pages[pages] = NULL; /* this might be seen in nfsd_splice_actor() */
 675
 676	/* Make arg->head point to first page and arg->pages point to rest */
 677	arg->head[0].iov_base = page_address(rqstp->rq_pages[0]);
 678	arg->head[0].iov_len = PAGE_SIZE;
 679	arg->pages = rqstp->rq_pages + 1;
 680	arg->page_base = 0;
 681	/* save at least one page for response */
 682	arg->page_len = (pages-2)*PAGE_SIZE;
 683	arg->len = (pages-1)*PAGE_SIZE;
 684	arg->tail[0].iov_len = 0;
 685
 686	rqstp->rq_xid = xdr_zero;
 687	return true;
 688}
 689
 690static bool
 691svc_thread_should_sleep(struct svc_rqst *rqstp)
 692{
 693	struct svc_pool		*pool = rqstp->rq_pool;
 694
 695	/* did someone call svc_wake_up? */
 696	if (test_bit(SP_TASK_PENDING, &pool->sp_flags))
 697		return false;
 698
 699	/* was a socket queued? */
 700	if (!lwq_empty(&pool->sp_xprts))
 701		return false;
 702
 703	/* are we shutting down? */
 704	if (svc_thread_should_stop(rqstp))
 705		return false;
 706
 707#if defined(CONFIG_SUNRPC_BACKCHANNEL)
 708	if (svc_is_backchannel(rqstp)) {
 709		if (!lwq_empty(&rqstp->rq_server->sv_cb_list))
 710			return false;
 711	}
 712#endif
 713
 714	return true;
 715}
 716
 717static void svc_thread_wait_for_work(struct svc_rqst *rqstp)
 718{
 719	struct svc_pool *pool = rqstp->rq_pool;
 720
 721	if (svc_thread_should_sleep(rqstp)) {
 722		set_current_state(TASK_IDLE | TASK_FREEZABLE);
 723		llist_add(&rqstp->rq_idle, &pool->sp_idle_threads);
 724		if (likely(svc_thread_should_sleep(rqstp)))
 725			schedule();
 726
 727		while (!llist_del_first_this(&pool->sp_idle_threads,
 728					     &rqstp->rq_idle)) {
 729			/* Work just became available.  This thread can only
 730			 * handle it after removing rqstp from the idle
 731			 * list. If that attempt failed, some other thread
 732			 * must have queued itself after finding no
 733			 * work to do, so that thread has taken responsibly
 734			 * for this new work.  This thread can safely sleep
 735			 * until woken again.
 736			 */
 737			schedule();
 738			set_current_state(TASK_IDLE | TASK_FREEZABLE);
 739		}
 740		__set_current_state(TASK_RUNNING);
 741	} else {
 742		cond_resched();
 743	}
 744	try_to_freeze();
 745}
 746
 747static void svc_add_new_temp_xprt(struct svc_serv *serv, struct svc_xprt *newxpt)
 748{
 749	spin_lock_bh(&serv->sv_lock);
 750	set_bit(XPT_TEMP, &newxpt->xpt_flags);
 751	list_add(&newxpt->xpt_list, &serv->sv_tempsocks);
 752	serv->sv_tmpcnt++;
 753	if (serv->sv_temptimer.function == NULL) {
 754		/* setup timer to age temp transports */
 755		serv->sv_temptimer.function = svc_age_temp_xprts;
 756		mod_timer(&serv->sv_temptimer,
 757			  jiffies + svc_conn_age_period * HZ);
 758	}
 759	spin_unlock_bh(&serv->sv_lock);
 760	svc_xprt_received(newxpt);
 761}
 762
 763static void svc_handle_xprt(struct svc_rqst *rqstp, struct svc_xprt *xprt)
 764{
 765	struct svc_serv *serv = rqstp->rq_server;
 766	int len = 0;
 767
 768	if (test_bit(XPT_CLOSE, &xprt->xpt_flags)) {
 769		if (test_and_clear_bit(XPT_KILL_TEMP, &xprt->xpt_flags))
 770			xprt->xpt_ops->xpo_kill_temp_xprt(xprt);
 771		svc_delete_xprt(xprt);
 772		/* Leave XPT_BUSY set on the dead xprt: */
 773		goto out;
 774	}
 775	if (test_bit(XPT_LISTENER, &xprt->xpt_flags)) {
 776		struct svc_xprt *newxpt;
 777		/*
 778		 * We know this module_get will succeed because the
 779		 * listener holds a reference too
 780		 */
 781		__module_get(xprt->xpt_class->xcl_owner);
 782		svc_check_conn_limits(xprt->xpt_server);
 783		newxpt = xprt->xpt_ops->xpo_accept(xprt);
 784		if (newxpt) {
 785			newxpt->xpt_cred = get_cred(xprt->xpt_cred);
 786			svc_add_new_temp_xprt(serv, newxpt);
 787			trace_svc_xprt_accept(newxpt, serv->sv_name);
 788		} else {
 789			module_put(xprt->xpt_class->xcl_owner);
 790		}
 791		svc_xprt_received(xprt);
 792	} else if (test_bit(XPT_HANDSHAKE, &xprt->xpt_flags)) {
 793		xprt->xpt_ops->xpo_handshake(xprt);
 794		svc_xprt_received(xprt);
 795	} else if (svc_xprt_reserve_slot(rqstp, xprt)) {
 796		/* XPT_DATA|XPT_DEFERRED case: */
 797		rqstp->rq_deferred = svc_deferred_dequeue(xprt);
 798		if (rqstp->rq_deferred)
 799			len = svc_deferred_recv(rqstp);
 800		else
 801			len = xprt->xpt_ops->xpo_recvfrom(rqstp);
 802		rqstp->rq_reserved = serv->sv_max_mesg;
 803		atomic_add(rqstp->rq_reserved, &xprt->xpt_reserved);
 804		if (len <= 0)
 805			goto out;
 806
 807		trace_svc_xdr_recvfrom(&rqstp->rq_arg);
 808
 809		clear_bit(XPT_OLD, &xprt->xpt_flags);
 810
 811		rqstp->rq_chandle.defer = svc_defer;
 812
 813		if (serv->sv_stats)
 814			serv->sv_stats->netcnt++;
 815		percpu_counter_inc(&rqstp->rq_pool->sp_messages_arrived);
 816		rqstp->rq_stime = ktime_get();
 817		svc_process(rqstp);
 818	} else
 819		svc_xprt_received(xprt);
 820
 821out:
 822	rqstp->rq_res.len = 0;
 823	svc_xprt_release(rqstp);
 824}
 825
 826static void svc_thread_wake_next(struct svc_rqst *rqstp)
 827{
 828	if (!svc_thread_should_sleep(rqstp))
 829		/* More work pending after I dequeued some,
 830		 * wake another worker
 831		 */
 832		svc_pool_wake_idle_thread(rqstp->rq_pool);
 833}
 834
 835/**
 836 * svc_recv - Receive and process the next request on any transport
 837 * @rqstp: an idle RPC service thread
 838 *
 839 * This code is carefully organised not to touch any cachelines in
 840 * the shared svc_serv structure, only cachelines in the local
 841 * svc_pool.
 842 */
 843void svc_recv(struct svc_rqst *rqstp)
 844{
 845	struct svc_pool *pool = rqstp->rq_pool;
 846
 847	if (!svc_alloc_arg(rqstp))
 848		return;
 849
 850	svc_thread_wait_for_work(rqstp);
 851
 852	clear_bit(SP_TASK_PENDING, &pool->sp_flags);
 853
 854	if (svc_thread_should_stop(rqstp)) {
 855		svc_thread_wake_next(rqstp);
 856		return;
 857	}
 858
 859	rqstp->rq_xprt = svc_xprt_dequeue(pool);
 860	if (rqstp->rq_xprt) {
 861		struct svc_xprt *xprt = rqstp->rq_xprt;
 862
 863		svc_thread_wake_next(rqstp);
 864		/* Normally we will wait up to 5 seconds for any required
 865		 * cache information to be provided.  When there are no
 866		 * idle threads, we reduce the wait time.
 867		 */
 868		if (pool->sp_idle_threads.first)
 869			rqstp->rq_chandle.thread_wait = 5 * HZ;
 870		else
 871			rqstp->rq_chandle.thread_wait = 1 * HZ;
 872
 873		trace_svc_xprt_dequeue(rqstp);
 874		svc_handle_xprt(rqstp, xprt);
 875	}
 876
 877#if defined(CONFIG_SUNRPC_BACKCHANNEL)
 878	if (svc_is_backchannel(rqstp)) {
 879		struct svc_serv *serv = rqstp->rq_server;
 880		struct rpc_rqst *req;
 881
 882		req = lwq_dequeue(&serv->sv_cb_list,
 883				  struct rpc_rqst, rq_bc_list);
 884		if (req) {
 885			svc_thread_wake_next(rqstp);
 886			svc_process_bc(req, rqstp);
 887		}
 888	}
 889#endif
 890}
 891EXPORT_SYMBOL_GPL(svc_recv);
 892
 893/**
 894 * svc_send - Return reply to client
 895 * @rqstp: RPC transaction context
 896 *
 897 */
 898void svc_send(struct svc_rqst *rqstp)
 899{
 900	struct svc_xprt	*xprt;
 901	struct xdr_buf	*xb;
 902	int status;
 903
 904	xprt = rqstp->rq_xprt;
 905
 906	/* calculate over-all length */
 907	xb = &rqstp->rq_res;
 908	xb->len = xb->head[0].iov_len +
 909		xb->page_len +
 910		xb->tail[0].iov_len;
 911	trace_svc_xdr_sendto(rqstp->rq_xid, xb);
 912	trace_svc_stats_latency(rqstp);
 913
 914	status = xprt->xpt_ops->xpo_sendto(rqstp);
 915
 916	trace_svc_send(rqstp, status);
 917}
 918
 919/*
 920 * Timer function to close old temporary transports, using
 921 * a mark-and-sweep algorithm.
 922 */
 923static void svc_age_temp_xprts(struct timer_list *t)
 924{
 925	struct svc_serv *serv = timer_container_of(serv, t, sv_temptimer);
 926	struct svc_xprt *xprt;
 927	struct list_head *le, *next;
 928
 929	dprintk("svc_age_temp_xprts\n");
 930
 931	if (!spin_trylock_bh(&serv->sv_lock)) {
 932		/* busy, try again 1 sec later */
 933		dprintk("svc_age_temp_xprts: busy\n");
 934		mod_timer(&serv->sv_temptimer, jiffies + HZ);
 935		return;
 936	}
 937
 938	list_for_each_safe(le, next, &serv->sv_tempsocks) {
 939		xprt = list_entry(le, struct svc_xprt, xpt_list);
 940
 941		/* First time through, just mark it OLD. Second time
 942		 * through, close it. */
 943		if (!test_and_set_bit(XPT_OLD, &xprt->xpt_flags))
 944			continue;
 945		if (kref_read(&xprt->xpt_ref) > 1 ||
 946		    test_bit(XPT_BUSY, &xprt->xpt_flags))
 947			continue;
 948		list_del_init(le);
 949		set_bit(XPT_CLOSE, &xprt->xpt_flags);
 950		dprintk("queuing xprt %p for closing\n", xprt);
 951
 952		/* a thread will dequeue and close it soon */
 953		svc_xprt_enqueue(xprt);
 954	}
 955	spin_unlock_bh(&serv->sv_lock);
 956
 957	mod_timer(&serv->sv_temptimer, jiffies + svc_conn_age_period * HZ);
 958}
 959
 960/* Close temporary transports whose xpt_local matches server_addr immediately
 961 * instead of waiting for them to be picked up by the timer.
 962 *
 963 * This is meant to be called from a notifier_block that runs when an ip
 964 * address is deleted.
 965 */
 966void svc_age_temp_xprts_now(struct svc_serv *serv, struct sockaddr *server_addr)
 967{
 968	struct svc_xprt *xprt;
 969	struct list_head *le, *next;
 970	LIST_HEAD(to_be_closed);
 971
 972	spin_lock_bh(&serv->sv_lock);
 973	list_for_each_safe(le, next, &serv->sv_tempsocks) {
 974		xprt = list_entry(le, struct svc_xprt, xpt_list);
 975		if (rpc_cmp_addr(server_addr, (struct sockaddr *)
 976				&xprt->xpt_local)) {
 977			dprintk("svc_age_temp_xprts_now: found %p\n", xprt);
 978			list_move(le, &to_be_closed);
 979		}
 980	}
 981	spin_unlock_bh(&serv->sv_lock);
 982
 983	while (!list_empty(&to_be_closed)) {
 984		le = to_be_closed.next;
 985		list_del_init(le);
 986		xprt = list_entry(le, struct svc_xprt, xpt_list);
 987		set_bit(XPT_CLOSE, &xprt->xpt_flags);
 988		set_bit(XPT_KILL_TEMP, &xprt->xpt_flags);
 989		dprintk("svc_age_temp_xprts_now: queuing xprt %p for closing\n",
 990				xprt);
 991		svc_xprt_enqueue(xprt);
 992	}
 993}
 994EXPORT_SYMBOL_GPL(svc_age_temp_xprts_now);
 995
 996static void call_xpt_users(struct svc_xprt *xprt)
 997{
 998	struct svc_xpt_user *u;
 999
1000	spin_lock(&xprt->xpt_lock);
1001	while (!list_empty(&xprt->xpt_users)) {
1002		u = list_first_entry(&xprt->xpt_users, struct svc_xpt_user, list);
1003		list_del_init(&u->list);
1004		u->callback(u);
1005	}
1006	spin_unlock(&xprt->xpt_lock);
1007}
1008
1009/*
1010 * Remove a dead transport
1011 */
1012static void svc_delete_xprt(struct svc_xprt *xprt)
1013{
1014	struct svc_serv	*serv = xprt->xpt_server;
1015	struct svc_deferred_req *dr;
1016
1017	/* unregister with rpcbind for when transport type is TCP or UDP.
1018	 */
1019	if (test_bit(XPT_RPCB_UNREG, &xprt->xpt_flags)) {
1020		struct svc_sock *svsk = container_of(xprt, struct svc_sock,
1021						     sk_xprt);
1022		struct socket *sock = svsk->sk_sock;
1023
1024		if (svc_register(serv, xprt->xpt_net, sock->sk->sk_family,
1025				 sock->sk->sk_protocol, 0) < 0)
1026			pr_warn("failed to unregister %s with rpcbind\n",
1027				xprt->xpt_class->xcl_name);
1028	}
1029
1030	if (test_and_set_bit(XPT_DEAD, &xprt->xpt_flags))
1031		return;
1032
1033	trace_svc_xprt_detach(xprt);
1034	xprt->xpt_ops->xpo_detach(xprt);
1035	if (xprt->xpt_bc_xprt)
1036		xprt->xpt_bc_xprt->ops->close(xprt->xpt_bc_xprt);
1037
1038	spin_lock_bh(&serv->sv_lock);
1039	list_del_init(&xprt->xpt_list);
1040	if (test_bit(XPT_TEMP, &xprt->xpt_flags) &&
1041	    !test_bit(XPT_PEER_VALID, &xprt->xpt_flags))
1042		serv->sv_tmpcnt--;
1043	spin_unlock_bh(&serv->sv_lock);
1044
1045	while ((dr = svc_deferred_dequeue(xprt)) != NULL)
1046		free_deferred(xprt, dr);
1047
1048	call_xpt_users(xprt);
1049	svc_xprt_put(xprt);
1050}
1051
1052/**
1053 * svc_xprt_close - Close a client connection
1054 * @xprt: transport to disconnect
1055 *
1056 */
1057void svc_xprt_close(struct svc_xprt *xprt)
1058{
1059	trace_svc_xprt_close(xprt);
1060	set_bit(XPT_CLOSE, &xprt->xpt_flags);
1061	if (test_and_set_bit(XPT_BUSY, &xprt->xpt_flags))
1062		/* someone else will have to effect the close */
1063		return;
1064	/*
1065	 * We expect svc_close_xprt() to work even when no threads are
1066	 * running (e.g., while configuring the server before starting
1067	 * any threads), so if the transport isn't busy, we delete
1068	 * it ourself:
1069	 */
1070	svc_delete_xprt(xprt);
1071}
1072EXPORT_SYMBOL_GPL(svc_xprt_close);
1073
1074static int svc_close_list(struct svc_serv *serv, struct list_head *xprt_list, struct net *net)
1075{
1076	struct svc_xprt *xprt;
1077	int ret = 0;
1078
1079	spin_lock_bh(&serv->sv_lock);
1080	list_for_each_entry(xprt, xprt_list, xpt_list) {
1081		if (xprt->xpt_net != net)
1082			continue;
1083		ret++;
1084		set_bit(XPT_CLOSE, &xprt->xpt_flags);
1085		svc_xprt_enqueue(xprt);
1086	}
1087	spin_unlock_bh(&serv->sv_lock);
1088	return ret;
1089}
1090
1091static void svc_clean_up_xprts(struct svc_serv *serv, struct net *net)
1092{
1093	struct svc_xprt *xprt;
1094	int i;
1095
1096	for (i = 0; i < serv->sv_nrpools; i++) {
1097		struct svc_pool *pool = &serv->sv_pools[i];
1098		struct llist_node *q, **t1, *t2;
1099
1100		q = lwq_dequeue_all(&pool->sp_xprts);
1101		lwq_for_each_safe(xprt, t1, t2, &q, xpt_ready) {
1102			if (xprt->xpt_net == net) {
1103				set_bit(XPT_CLOSE, &xprt->xpt_flags);
1104				svc_delete_xprt(xprt);
1105				xprt = NULL;
1106			}
1107		}
1108
1109		if (q)
1110			lwq_enqueue_batch(q, &pool->sp_xprts);
1111	}
1112}
1113
1114/**
1115 * svc_xprt_destroy_all - Destroy transports associated with @serv
1116 * @serv: RPC service to be shut down
1117 * @net: target network namespace
1118 * @unregister: true if it is OK to unregister the destroyed xprts
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			  bool unregister)
1133{
1134	int delay = 0;
1135
1136	while (svc_close_list(serv, &serv->sv_permsocks, net) +
1137	       svc_close_list(serv, &serv->sv_tempsocks, net)) {
1138
1139		svc_clean_up_xprts(serv, net);
1140		msleep(delay++);
1141	}
1142
1143	if (unregister)
1144		svc_rpcb_cleanup(serv, net);
1145}
1146EXPORT_SYMBOL_GPL(svc_xprt_destroy_all);
1147
1148/*
1149 * Handle defer and revisit of requests
1150 */
1151
1152static void svc_revisit(struct cache_deferred_req *dreq, int too_many)
1153{
1154	struct svc_deferred_req *dr =
1155		container_of(dreq, struct svc_deferred_req, handle);
1156	struct svc_xprt *xprt = dr->xprt;
1157
1158	spin_lock(&xprt->xpt_lock);
1159	set_bit(XPT_DEFERRED, &xprt->xpt_flags);
1160	if (too_many || test_bit(XPT_DEAD, &xprt->xpt_flags)) {
1161		spin_unlock(&xprt->xpt_lock);
1162		trace_svc_defer_drop(dr);
1163		free_deferred(xprt, dr);
1164		svc_xprt_put(xprt);
1165		return;
1166	}
1167	dr->xprt = NULL;
1168	list_add(&dr->handle.recent, &xprt->xpt_deferred);
1169	spin_unlock(&xprt->xpt_lock);
1170	trace_svc_defer_queue(dr);
1171	svc_xprt_enqueue(xprt);
1172	svc_xprt_put(xprt);
1173}
1174
1175/*
1176 * Save the request off for later processing. The request buffer looks
1177 * like this:
1178 *
1179 * <xprt-header><rpc-header><rpc-pagelist><rpc-tail>
1180 *
1181 * This code can only handle requests that consist of an xprt-header
1182 * and rpc-header.
1183 */
1184static struct cache_deferred_req *svc_defer(struct cache_req *req)
1185{
1186	struct svc_rqst *rqstp = container_of(req, struct svc_rqst, rq_chandle);
1187	struct svc_deferred_req *dr;
1188
1189	if (rqstp->rq_arg.page_len || !test_bit(RQ_USEDEFERRAL, &rqstp->rq_flags))
1190		return NULL; /* if more than a page, give up FIXME */
1191	if (rqstp->rq_deferred) {
1192		dr = rqstp->rq_deferred;
1193		rqstp->rq_deferred = NULL;
1194	} else {
1195		size_t skip;
1196		size_t size;
1197		/* FIXME maybe discard if size too large */
1198		size = sizeof(struct svc_deferred_req) + rqstp->rq_arg.len;
1199		dr = kmalloc(size, GFP_KERNEL);
1200		if (dr == NULL)
1201			return NULL;
1202
1203		dr->handle.owner = rqstp->rq_server;
1204		dr->prot = rqstp->rq_prot;
1205		memcpy(&dr->addr, &rqstp->rq_addr, rqstp->rq_addrlen);
1206		dr->addrlen = rqstp->rq_addrlen;
1207		dr->daddr = rqstp->rq_daddr;
1208		dr->argslen = rqstp->rq_arg.len >> 2;
1209
1210		/* back up head to the start of the buffer and copy */
1211		skip = rqstp->rq_arg.len - rqstp->rq_arg.head[0].iov_len;
1212		memcpy(dr->args, rqstp->rq_arg.head[0].iov_base - skip,
1213		       dr->argslen << 2);
1214	}
1215	dr->xprt_ctxt = rqstp->rq_xprt_ctxt;
1216	rqstp->rq_xprt_ctxt = NULL;
1217	trace_svc_defer(rqstp);
1218	svc_xprt_get(rqstp->rq_xprt);
1219	dr->xprt = rqstp->rq_xprt;
1220	set_bit(RQ_DROPME, &rqstp->rq_flags);
1221
1222	dr->handle.revisit = svc_revisit;
1223	return &dr->handle;
1224}
1225
1226/*
1227 * recv data from a deferred request into an active one
1228 */
1229static noinline int svc_deferred_recv(struct svc_rqst *rqstp)
1230{
1231	struct svc_deferred_req *dr = rqstp->rq_deferred;
1232
1233	trace_svc_defer_recv(dr);
1234
1235	/* setup iov_base past transport header */
1236	rqstp->rq_arg.head[0].iov_base = dr->args;
1237	/* The iov_len does not include the transport header bytes */
1238	rqstp->rq_arg.head[0].iov_len = dr->argslen << 2;
1239	rqstp->rq_arg.page_len = 0;
1240	/* The rq_arg.len includes the transport header bytes */
1241	rqstp->rq_arg.len     = dr->argslen << 2;
1242	rqstp->rq_prot        = dr->prot;
1243	memcpy(&rqstp->rq_addr, &dr->addr, dr->addrlen);
1244	rqstp->rq_addrlen     = dr->addrlen;
1245	/* Save off transport header len in case we get deferred again */
1246	rqstp->rq_daddr       = dr->daddr;
1247	rqstp->rq_respages    = rqstp->rq_pages;
1248	rqstp->rq_xprt_ctxt   = dr->xprt_ctxt;
1249
1250	dr->xprt_ctxt = NULL;
1251	svc_xprt_received(rqstp->rq_xprt);
1252	return dr->argslen << 2;
1253}
1254
1255
1256static struct svc_deferred_req *svc_deferred_dequeue(struct svc_xprt *xprt)
1257{
1258	struct svc_deferred_req *dr = NULL;
1259
1260	if (!test_bit(XPT_DEFERRED, &xprt->xpt_flags))
1261		return NULL;
1262	spin_lock(&xprt->xpt_lock);
1263	if (!list_empty(&xprt->xpt_deferred)) {
1264		dr = list_entry(xprt->xpt_deferred.next,
1265				struct svc_deferred_req,
1266				handle.recent);
1267		list_del_init(&dr->handle.recent);
1268	} else
1269		clear_bit(XPT_DEFERRED, &xprt->xpt_flags);
1270	spin_unlock(&xprt->xpt_lock);
1271	return dr;
1272}
1273
1274/**
1275 * svc_find_listener - find an RPC transport instance
1276 * @serv: pointer to svc_serv to search
1277 * @xcl_name: C string containing transport's class name
1278 * @net: owner net pointer
1279 * @sa: sockaddr containing address
1280 *
1281 * Return the transport instance pointer for the endpoint accepting
1282 * connections/peer traffic from the specified transport class,
1283 * and matching sockaddr.
1284 */
1285struct svc_xprt *svc_find_listener(struct svc_serv *serv, const char *xcl_name,
1286				   struct net *net, const struct sockaddr *sa)
1287{
1288	struct svc_xprt *xprt;
1289	struct svc_xprt *found = NULL;
1290
1291	spin_lock_bh(&serv->sv_lock);
1292	list_for_each_entry(xprt, &serv->sv_permsocks, xpt_list) {
1293		if (xprt->xpt_net != net)
1294			continue;
1295		if (strcmp(xprt->xpt_class->xcl_name, xcl_name))
1296			continue;
1297		if (!rpc_cmp_addr_port(sa, (struct sockaddr *)&xprt->xpt_local))
1298			continue;
1299		found = xprt;
1300		svc_xprt_get(xprt);
1301		break;
1302	}
1303	spin_unlock_bh(&serv->sv_lock);
1304	return found;
1305}
1306EXPORT_SYMBOL_GPL(svc_find_listener);
1307
1308/**
1309 * svc_find_xprt - find an RPC transport instance
1310 * @serv: pointer to svc_serv to search
1311 * @xcl_name: C string containing transport's class name
1312 * @net: owner net pointer
1313 * @af: Address family of transport's local address
1314 * @port: transport's IP port number
1315 *
1316 * Return the transport instance pointer for the endpoint accepting
1317 * connections/peer traffic from the specified transport class,
1318 * address family and port.
1319 *
1320 * Specifying 0 for the address family or port is effectively a
1321 * wild-card, and will result in matching the first transport in the
1322 * service's list that has a matching class name.
1323 */
1324struct svc_xprt *svc_find_xprt(struct svc_serv *serv, const char *xcl_name,
1325			       struct net *net, const sa_family_t af,
1326			       const unsigned short port)
1327{
1328	struct svc_xprt *xprt;
1329	struct svc_xprt *found = NULL;
1330
1331	/* Sanity check the args */
1332	if (serv == NULL || xcl_name == NULL)
1333		return found;
1334
1335	spin_lock_bh(&serv->sv_lock);
1336	list_for_each_entry(xprt, &serv->sv_permsocks, xpt_list) {
1337		if (xprt->xpt_net != net)
1338			continue;
1339		if (strcmp(xprt->xpt_class->xcl_name, xcl_name))
1340			continue;
1341		if (af != AF_UNSPEC && af != xprt->xpt_local.ss_family)
1342			continue;
1343		if (port != 0 && port != svc_xprt_local_port(xprt))
1344			continue;
1345		found = xprt;
1346		svc_xprt_get(xprt);
1347		break;
1348	}
1349	spin_unlock_bh(&serv->sv_lock);
1350	return found;
1351}
1352EXPORT_SYMBOL_GPL(svc_find_xprt);
1353
1354static int svc_one_xprt_name(const struct svc_xprt *xprt,
1355			     char *pos, int remaining)
1356{
1357	int len;
1358
1359	len = snprintf(pos, remaining, "%s %u\n",
1360			xprt->xpt_class->xcl_name,
1361			svc_xprt_local_port(xprt));
1362	if (len >= remaining)
1363		return -ENAMETOOLONG;
1364	return len;
1365}
1366
1367/**
1368 * svc_xprt_names - format a buffer with a list of transport names
1369 * @serv: pointer to an RPC service
1370 * @buf: pointer to a buffer to be filled in
1371 * @buflen: length of buffer to be filled in
1372 *
1373 * Fills in @buf with a string containing a list of transport names,
1374 * each name terminated with '\n'.
1375 *
1376 * Returns positive length of the filled-in string on success; otherwise
1377 * a negative errno value is returned if an error occurs.
1378 */
1379int svc_xprt_names(struct svc_serv *serv, char *buf, const int buflen)
1380{
1381	struct svc_xprt *xprt;
1382	int len, totlen;
1383	char *pos;
1384
1385	/* Sanity check args */
1386	if (!serv)
1387		return 0;
1388
1389	spin_lock_bh(&serv->sv_lock);
1390
1391	pos = buf;
1392	totlen = 0;
1393	list_for_each_entry(xprt, &serv->sv_permsocks, xpt_list) {
1394		len = svc_one_xprt_name(xprt, pos, buflen - totlen);
1395		if (len < 0) {
1396			*buf = '\0';
1397			totlen = len;
1398		}
1399		if (len <= 0)
1400			break;
1401
1402		pos += len;
1403		totlen += len;
1404	}
1405
1406	spin_unlock_bh(&serv->sv_lock);
1407	return totlen;
1408}
1409EXPORT_SYMBOL_GPL(svc_xprt_names);
1410
1411/*----------------------------------------------------------------------------*/
1412
1413static void *svc_pool_stats_start(struct seq_file *m, loff_t *pos)
1414{
1415	unsigned int pidx = (unsigned int)*pos;
1416	struct svc_info *si = m->private;
1417
1418	dprintk("svc_pool_stats_start, *pidx=%u\n", pidx);
1419
1420	mutex_lock(si->mutex);
1421
1422	if (!pidx)
1423		return SEQ_START_TOKEN;
1424	if (!si->serv)
1425		return NULL;
1426	return pidx > si->serv->sv_nrpools ? NULL
1427		: &si->serv->sv_pools[pidx - 1];
1428}
1429
1430static void *svc_pool_stats_next(struct seq_file *m, void *p, loff_t *pos)
1431{
1432	struct svc_pool *pool = p;
1433	struct svc_info *si = m->private;
1434	struct svc_serv *serv = si->serv;
1435
1436	dprintk("svc_pool_stats_next, *pos=%llu\n", *pos);
1437
1438	if (!serv) {
1439		pool = NULL;
1440	} else if (p == SEQ_START_TOKEN) {
1441		pool = &serv->sv_pools[0];
1442	} else {
1443		unsigned int pidx = (pool - &serv->sv_pools[0]);
1444		if (pidx < serv->sv_nrpools-1)
1445			pool = &serv->sv_pools[pidx+1];
1446		else
1447			pool = NULL;
1448	}
1449	++*pos;
1450	return pool;
1451}
1452
1453static void svc_pool_stats_stop(struct seq_file *m, void *p)
1454{
1455	struct svc_info *si = m->private;
1456
1457	mutex_unlock(si->mutex);
1458}
1459
1460static int svc_pool_stats_show(struct seq_file *m, void *p)
1461{
1462	struct svc_pool *pool = p;
1463
1464	if (p == SEQ_START_TOKEN) {
1465		seq_puts(m, "# pool packets-arrived sockets-enqueued threads-woken threads-timedout\n");
1466		return 0;
1467	}
1468
1469	seq_printf(m, "%u %llu %llu %llu 0\n",
1470		   pool->sp_id,
1471		   percpu_counter_sum_positive(&pool->sp_messages_arrived),
1472		   percpu_counter_sum_positive(&pool->sp_sockets_queued),
1473		   percpu_counter_sum_positive(&pool->sp_threads_woken));
1474
1475	return 0;
1476}
1477
1478static const struct seq_operations svc_pool_stats_seq_ops = {
1479	.start	= svc_pool_stats_start,
1480	.next	= svc_pool_stats_next,
1481	.stop	= svc_pool_stats_stop,
1482	.show	= svc_pool_stats_show,
1483};
1484
1485int svc_pool_stats_open(struct svc_info *info, struct file *file)
1486{
1487	struct seq_file *seq;
1488	int err;
1489
1490	err = seq_open(file, &svc_pool_stats_seq_ops);
1491	if (err)
1492		return err;
1493	seq = file->private_data;
1494	seq->private = info;
1495
1496	return 0;
1497}
1498EXPORT_SYMBOL(svc_pool_stats_open);
1499
1500/*----------------------------------------------------------------------------*/