Linux Audio

Check our new training course

Loading...
v6.9.4
   1// SPDX-License-Identifier: GPL-2.0
   2/*
   3 * The NFSD open file cache.
   4 *
   5 * (c) 2015 - Jeff Layton <jeff.layton@primarydata.com>
   6 *
   7 * An nfsd_file object is a per-file collection of open state that binds
   8 * together:
   9 *   - a struct file *
  10 *   - a user credential
  11 *   - a network namespace
  12 *   - a read-ahead context
  13 *   - monitoring for writeback errors
  14 *
  15 * nfsd_file objects are reference-counted. Consumers acquire a new
  16 * object via the nfsd_file_acquire API. They manage their interest in
  17 * the acquired object, and hence the object's reference count, via
  18 * nfsd_file_get and nfsd_file_put. There are two varieties of nfsd_file
  19 * object:
  20 *
  21 *  * non-garbage-collected: When a consumer wants to precisely control
  22 *    the lifetime of a file's open state, it acquires a non-garbage-
  23 *    collected nfsd_file. The final nfsd_file_put releases the open
  24 *    state immediately.
  25 *
  26 *  * garbage-collected: When a consumer does not control the lifetime
  27 *    of open state, it acquires a garbage-collected nfsd_file. The
  28 *    final nfsd_file_put allows the open state to linger for a period
  29 *    during which it may be re-used.
  30 */
  31
  32#include <linux/hash.h>
  33#include <linux/slab.h>
  34#include <linux/file.h>
  35#include <linux/pagemap.h>
  36#include <linux/sched.h>
  37#include <linux/list_lru.h>
  38#include <linux/fsnotify_backend.h>
  39#include <linux/fsnotify.h>
  40#include <linux/seq_file.h>
  41#include <linux/rhashtable.h>
  42
  43#include "vfs.h"
  44#include "nfsd.h"
  45#include "nfsfh.h"
  46#include "netns.h"
  47#include "filecache.h"
  48#include "trace.h"
  49
  50#define NFSD_LAUNDRETTE_DELAY		     (2 * HZ)
  51
  52#define NFSD_FILE_CACHE_UP		     (0)
  53
  54/* We only care about NFSD_MAY_READ/WRITE for this cache */
  55#define NFSD_FILE_MAY_MASK	(NFSD_MAY_READ|NFSD_MAY_WRITE)
  56
  57static DEFINE_PER_CPU(unsigned long, nfsd_file_cache_hits);
  58static DEFINE_PER_CPU(unsigned long, nfsd_file_acquisitions);
 
  59static DEFINE_PER_CPU(unsigned long, nfsd_file_releases);
  60static DEFINE_PER_CPU(unsigned long, nfsd_file_total_age);
  61static DEFINE_PER_CPU(unsigned long, nfsd_file_evictions);
  62
  63struct nfsd_fcache_disposal {
  64	spinlock_t lock;
  65	struct list_head freeme;
  66};
  67
  68static struct kmem_cache		*nfsd_file_slab;
  69static struct kmem_cache		*nfsd_file_mark_slab;
  70static struct list_lru			nfsd_file_lru;
  71static unsigned long			nfsd_file_flags;
  72static struct fsnotify_group		*nfsd_file_fsnotify_group;
  73static struct delayed_work		nfsd_filecache_laundrette;
  74static struct rhltable			nfsd_file_rhltable
  75						____cacheline_aligned_in_smp;
  76
  77static bool
  78nfsd_match_cred(const struct cred *c1, const struct cred *c2)
  79{
  80	int i;
  81
  82	if (!uid_eq(c1->fsuid, c2->fsuid))
  83		return false;
  84	if (!gid_eq(c1->fsgid, c2->fsgid))
  85		return false;
  86	if (c1->group_info == NULL || c2->group_info == NULL)
  87		return c1->group_info == c2->group_info;
  88	if (c1->group_info->ngroups != c2->group_info->ngroups)
  89		return false;
  90	for (i = 0; i < c1->group_info->ngroups; i++) {
  91		if (!gid_eq(c1->group_info->gid[i], c2->group_info->gid[i]))
  92			return false;
  93	}
  94	return true;
  95}
  96
  97static const struct rhashtable_params nfsd_file_rhash_params = {
  98	.key_len		= sizeof_field(struct nfsd_file, nf_inode),
  99	.key_offset		= offsetof(struct nfsd_file, nf_inode),
 100	.head_offset		= offsetof(struct nfsd_file, nf_rlist),
 101
 102	/*
 103	 * Start with a single page hash table to reduce resizing churn
 104	 * on light workloads.
 105	 */
 106	.min_size		= 256,
 107	.automatic_shrinking	= true,
 108};
 109
 110static void
 111nfsd_file_schedule_laundrette(void)
 112{
 113	if (test_bit(NFSD_FILE_CACHE_UP, &nfsd_file_flags))
 114		queue_delayed_work(system_wq, &nfsd_filecache_laundrette,
 115				   NFSD_LAUNDRETTE_DELAY);
 116}
 117
 118static void
 119nfsd_file_slab_free(struct rcu_head *rcu)
 120{
 121	struct nfsd_file *nf = container_of(rcu, struct nfsd_file, nf_rcu);
 122
 123	put_cred(nf->nf_cred);
 124	kmem_cache_free(nfsd_file_slab, nf);
 125}
 126
 127static void
 128nfsd_file_mark_free(struct fsnotify_mark *mark)
 129{
 130	struct nfsd_file_mark *nfm = container_of(mark, struct nfsd_file_mark,
 131						  nfm_mark);
 132
 133	kmem_cache_free(nfsd_file_mark_slab, nfm);
 134}
 135
 136static struct nfsd_file_mark *
 137nfsd_file_mark_get(struct nfsd_file_mark *nfm)
 138{
 139	if (!refcount_inc_not_zero(&nfm->nfm_ref))
 140		return NULL;
 141	return nfm;
 142}
 143
 144static void
 145nfsd_file_mark_put(struct nfsd_file_mark *nfm)
 146{
 147	if (refcount_dec_and_test(&nfm->nfm_ref)) {
 148		fsnotify_destroy_mark(&nfm->nfm_mark, nfsd_file_fsnotify_group);
 149		fsnotify_put_mark(&nfm->nfm_mark);
 150	}
 151}
 152
 153static struct nfsd_file_mark *
 154nfsd_file_mark_find_or_create(struct nfsd_file *nf, struct inode *inode)
 155{
 156	int			err;
 157	struct fsnotify_mark	*mark;
 158	struct nfsd_file_mark	*nfm = NULL, *new;
 159
 160	do {
 161		fsnotify_group_lock(nfsd_file_fsnotify_group);
 162		mark = fsnotify_find_mark(&inode->i_fsnotify_marks,
 163					  nfsd_file_fsnotify_group);
 164		if (mark) {
 165			nfm = nfsd_file_mark_get(container_of(mark,
 166						 struct nfsd_file_mark,
 167						 nfm_mark));
 168			fsnotify_group_unlock(nfsd_file_fsnotify_group);
 169			if (nfm) {
 170				fsnotify_put_mark(mark);
 171				break;
 172			}
 173			/* Avoid soft lockup race with nfsd_file_mark_put() */
 174			fsnotify_destroy_mark(mark, nfsd_file_fsnotify_group);
 175			fsnotify_put_mark(mark);
 176		} else {
 177			fsnotify_group_unlock(nfsd_file_fsnotify_group);
 178		}
 179
 180		/* allocate a new nfm */
 181		new = kmem_cache_alloc(nfsd_file_mark_slab, GFP_KERNEL);
 182		if (!new)
 183			return NULL;
 184		fsnotify_init_mark(&new->nfm_mark, nfsd_file_fsnotify_group);
 185		new->nfm_mark.mask = FS_ATTRIB|FS_DELETE_SELF;
 186		refcount_set(&new->nfm_ref, 1);
 187
 188		err = fsnotify_add_inode_mark(&new->nfm_mark, inode, 0);
 189
 190		/*
 191		 * If the add was successful, then return the object.
 192		 * Otherwise, we need to put the reference we hold on the
 193		 * nfm_mark. The fsnotify code will take a reference and put
 194		 * it on failure, so we can't just free it directly. It's also
 195		 * not safe to call fsnotify_destroy_mark on it as the
 196		 * mark->group will be NULL. Thus, we can't let the nfm_ref
 197		 * counter drive the destruction at this point.
 198		 */
 199		if (likely(!err))
 200			nfm = new;
 201		else
 202			fsnotify_put_mark(&new->nfm_mark);
 203	} while (unlikely(err == -EEXIST));
 204
 205	return nfm;
 206}
 207
 208static struct nfsd_file *
 209nfsd_file_alloc(struct net *net, struct inode *inode, unsigned char need,
 210		bool want_gc)
 211{
 212	struct nfsd_file *nf;
 213
 214	nf = kmem_cache_alloc(nfsd_file_slab, GFP_KERNEL);
 215	if (unlikely(!nf))
 216		return NULL;
 217
 
 218	INIT_LIST_HEAD(&nf->nf_lru);
 
 219	nf->nf_birthtime = ktime_get();
 220	nf->nf_file = NULL;
 221	nf->nf_cred = get_current_cred();
 222	nf->nf_net = net;
 223	nf->nf_flags = want_gc ?
 224		BIT(NFSD_FILE_HASHED) | BIT(NFSD_FILE_PENDING) | BIT(NFSD_FILE_GC) :
 225		BIT(NFSD_FILE_HASHED) | BIT(NFSD_FILE_PENDING);
 226	nf->nf_inode = inode;
 227	refcount_set(&nf->nf_ref, 1);
 228	nf->nf_may = need;
 229	nf->nf_mark = NULL;
 230	return nf;
 231}
 232
 233/**
 234 * nfsd_file_check_write_error - check for writeback errors on a file
 235 * @nf: nfsd_file to check for writeback errors
 236 *
 237 * Check whether a nfsd_file has an unseen error. Reset the write
 238 * verifier if so.
 239 */
 240static void
 241nfsd_file_check_write_error(struct nfsd_file *nf)
 242{
 243	struct file *file = nf->nf_file;
 244
 245	if ((file->f_mode & FMODE_WRITE) &&
 246	    filemap_check_wb_err(file->f_mapping, READ_ONCE(file->f_wb_err)))
 247		nfsd_reset_write_verifier(net_generic(nf->nf_net, nfsd_net_id));
 248}
 249
 250static void
 251nfsd_file_hash_remove(struct nfsd_file *nf)
 252{
 253	trace_nfsd_file_unhash(nf);
 254	rhltable_remove(&nfsd_file_rhltable, &nf->nf_rlist,
 255			nfsd_file_rhash_params);
 256}
 257
 258static bool
 259nfsd_file_unhash(struct nfsd_file *nf)
 260{
 261	if (test_and_clear_bit(NFSD_FILE_HASHED, &nf->nf_flags)) {
 262		nfsd_file_hash_remove(nf);
 263		return true;
 264	}
 265	return false;
 266}
 267
 268static void
 269nfsd_file_free(struct nfsd_file *nf)
 270{
 271	s64 age = ktime_to_ms(ktime_sub(ktime_get(), nf->nf_birthtime));
 272
 273	trace_nfsd_file_free(nf);
 274
 275	this_cpu_inc(nfsd_file_releases);
 276	this_cpu_add(nfsd_file_total_age, age);
 277
 278	nfsd_file_unhash(nf);
 279	if (nf->nf_mark)
 280		nfsd_file_mark_put(nf->nf_mark);
 281	if (nf->nf_file) {
 282		nfsd_file_check_write_error(nf);
 283		nfsd_filp_close(nf->nf_file);
 284	}
 285
 286	/*
 287	 * If this item is still linked via nf_lru, that's a bug.
 288	 * WARN and leak it to preserve system stability.
 289	 */
 290	if (WARN_ON_ONCE(!list_empty(&nf->nf_lru)))
 291		return;
 292
 293	call_rcu(&nf->nf_rcu, nfsd_file_slab_free);
 294}
 295
 296static bool
 297nfsd_file_check_writeback(struct nfsd_file *nf)
 298{
 299	struct file *file = nf->nf_file;
 300	struct address_space *mapping;
 301
 302	/* File not open for write? */
 303	if (!(file->f_mode & FMODE_WRITE))
 304		return false;
 305
 306	/*
 307	 * Some filesystems (e.g. NFS) flush all dirty data on close.
 308	 * On others, there is no need to wait for writeback.
 309	 */
 310	if (!(file_inode(file)->i_sb->s_export_op->flags & EXPORT_OP_FLUSH_ON_CLOSE))
 311		return false;
 312
 313	mapping = file->f_mapping;
 314	return mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) ||
 315		mapping_tagged(mapping, PAGECACHE_TAG_WRITEBACK);
 316}
 317
 318
 319static bool nfsd_file_lru_add(struct nfsd_file *nf)
 320{
 321	set_bit(NFSD_FILE_REFERENCED, &nf->nf_flags);
 322	if (list_lru_add_obj(&nfsd_file_lru, &nf->nf_lru)) {
 323		trace_nfsd_file_lru_add(nf);
 324		return true;
 325	}
 326	return false;
 327}
 328
 329static bool nfsd_file_lru_remove(struct nfsd_file *nf)
 330{
 331	if (list_lru_del_obj(&nfsd_file_lru, &nf->nf_lru)) {
 332		trace_nfsd_file_lru_del(nf);
 333		return true;
 334	}
 335	return false;
 336}
 337
 338struct nfsd_file *
 339nfsd_file_get(struct nfsd_file *nf)
 340{
 341	if (nf && refcount_inc_not_zero(&nf->nf_ref))
 342		return nf;
 343	return NULL;
 344}
 345
 346/**
 347 * nfsd_file_put - put the reference to a nfsd_file
 348 * @nf: nfsd_file of which to put the reference
 349 *
 350 * Put a reference to a nfsd_file. In the non-GC case, we just put the
 351 * reference immediately. In the GC case, if the reference would be
 352 * the last one, the put it on the LRU instead to be cleaned up later.
 353 */
 354void
 355nfsd_file_put(struct nfsd_file *nf)
 356{
 357	might_sleep();
 358	trace_nfsd_file_put(nf);
 359
 360	if (test_bit(NFSD_FILE_GC, &nf->nf_flags) &&
 361	    test_bit(NFSD_FILE_HASHED, &nf->nf_flags)) {
 362		/*
 363		 * If this is the last reference (nf_ref == 1), then try to
 364		 * transfer it to the LRU.
 365		 */
 366		if (refcount_dec_not_one(&nf->nf_ref))
 367			return;
 368
 369		/* Try to add it to the LRU.  If that fails, decrement. */
 370		if (nfsd_file_lru_add(nf)) {
 371			/* If it's still hashed, we're done */
 372			if (test_bit(NFSD_FILE_HASHED, &nf->nf_flags)) {
 373				nfsd_file_schedule_laundrette();
 374				return;
 375			}
 376
 377			/*
 378			 * We're racing with unhashing, so try to remove it from
 379			 * the LRU. If removal fails, then someone else already
 380			 * has our reference.
 381			 */
 382			if (!nfsd_file_lru_remove(nf))
 383				return;
 384		}
 385	}
 386	if (refcount_dec_and_test(&nf->nf_ref))
 387		nfsd_file_free(nf);
 388}
 389
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 390static void
 391nfsd_file_dispose_list(struct list_head *dispose)
 392{
 393	struct nfsd_file *nf;
 394
 395	while (!list_empty(dispose)) {
 396		nf = list_first_entry(dispose, struct nfsd_file, nf_lru);
 397		list_del_init(&nf->nf_lru);
 398		nfsd_file_free(nf);
 399	}
 400}
 401
 402/**
 403 * nfsd_file_dispose_list_delayed - move list of dead files to net's freeme list
 404 * @dispose: list of nfsd_files to be disposed
 405 *
 406 * Transfers each file to the "freeme" list for its nfsd_net, to eventually
 407 * be disposed of by the per-net garbage collector.
 408 */
 409static void
 410nfsd_file_dispose_list_delayed(struct list_head *dispose)
 411{
 412	while(!list_empty(dispose)) {
 413		struct nfsd_file *nf = list_first_entry(dispose,
 414						struct nfsd_file, nf_lru);
 415		struct nfsd_net *nn = net_generic(nf->nf_net, nfsd_net_id);
 416		struct nfsd_fcache_disposal *l = nn->fcache_disposal;
 
 417
 418		spin_lock(&l->lock);
 419		list_move_tail(&nf->nf_lru, &l->freeme);
 420		spin_unlock(&l->lock);
 421		svc_wake_up(nn->nfsd_serv);
 
 
 
 
 
 
 
 
 422	}
 423}
 424
 425/**
 426 * nfsd_file_net_dispose - deal with nfsd_files waiting to be disposed.
 427 * @nn: nfsd_net in which to find files to be disposed.
 428 *
 429 * When files held open for nfsv3 are removed from the filecache, whether
 430 * due to memory pressure or garbage collection, they are queued to
 431 * a per-net-ns queue.  This function completes the disposal, either
 432 * directly or by waking another nfsd thread to help with the work.
 433 */
 434void nfsd_file_net_dispose(struct nfsd_net *nn)
 435{
 436	struct nfsd_fcache_disposal *l = nn->fcache_disposal;
 437
 438	if (!list_empty(&l->freeme)) {
 439		LIST_HEAD(dispose);
 440		int i;
 441
 442		spin_lock(&l->lock);
 443		for (i = 0; i < 8 && !list_empty(&l->freeme); i++)
 444			list_move(l->freeme.next, &dispose);
 445		spin_unlock(&l->lock);
 446		if (!list_empty(&l->freeme))
 447			/* Wake up another thread to share the work
 448			 * *before* doing any actual disposing.
 449			 */
 450			svc_wake_up(nn->nfsd_serv);
 451		nfsd_file_dispose_list(&dispose);
 452	}
 453}
 454
 455/**
 456 * nfsd_file_lru_cb - Examine an entry on the LRU list
 457 * @item: LRU entry to examine
 458 * @lru: controlling LRU
 459 * @lock: LRU list lock (unused)
 460 * @arg: dispose list
 461 *
 462 * Return values:
 463 *   %LRU_REMOVED: @item was removed from the LRU
 464 *   %LRU_ROTATE: @item is to be moved to the LRU tail
 465 *   %LRU_SKIP: @item cannot be evicted
 466 */
 467static enum lru_status
 468nfsd_file_lru_cb(struct list_head *item, struct list_lru_one *lru,
 469		 spinlock_t *lock, void *arg)
 470	__releases(lock)
 471	__acquires(lock)
 472{
 473	struct list_head *head = arg;
 474	struct nfsd_file *nf = list_entry(item, struct nfsd_file, nf_lru);
 475
 476	/* We should only be dealing with GC entries here */
 477	WARN_ON_ONCE(!test_bit(NFSD_FILE_GC, &nf->nf_flags));
 478
 479	/*
 480	 * Don't throw out files that are still undergoing I/O or
 481	 * that have uncleared errors pending.
 482	 */
 483	if (nfsd_file_check_writeback(nf)) {
 484		trace_nfsd_file_gc_writeback(nf);
 485		return LRU_SKIP;
 486	}
 487
 488	/* If it was recently added to the list, skip it */
 489	if (test_and_clear_bit(NFSD_FILE_REFERENCED, &nf->nf_flags)) {
 490		trace_nfsd_file_gc_referenced(nf);
 491		return LRU_ROTATE;
 492	}
 493
 494	/*
 495	 * Put the reference held on behalf of the LRU. If it wasn't the last
 496	 * one, then just remove it from the LRU and ignore it.
 497	 */
 498	if (!refcount_dec_and_test(&nf->nf_ref)) {
 499		trace_nfsd_file_gc_in_use(nf);
 500		list_lru_isolate(lru, &nf->nf_lru);
 501		return LRU_REMOVED;
 502	}
 503
 504	/* Refcount went to zero. Unhash it and queue it to the dispose list */
 505	nfsd_file_unhash(nf);
 506	list_lru_isolate_move(lru, &nf->nf_lru, head);
 
 507	this_cpu_inc(nfsd_file_evictions);
 508	trace_nfsd_file_gc_disposed(nf);
 509	return LRU_REMOVED;
 510}
 511
 512static void
 513nfsd_file_gc(void)
 514{
 515	LIST_HEAD(dispose);
 516	unsigned long ret;
 517
 518	ret = list_lru_walk(&nfsd_file_lru, nfsd_file_lru_cb,
 519			    &dispose, list_lru_count(&nfsd_file_lru));
 520	trace_nfsd_file_gc_removed(ret, list_lru_count(&nfsd_file_lru));
 521	nfsd_file_dispose_list_delayed(&dispose);
 522}
 523
 524static void
 525nfsd_file_gc_worker(struct work_struct *work)
 526{
 527	nfsd_file_gc();
 528	if (list_lru_count(&nfsd_file_lru))
 529		nfsd_file_schedule_laundrette();
 530}
 531
 532static unsigned long
 533nfsd_file_lru_count(struct shrinker *s, struct shrink_control *sc)
 534{
 535	return list_lru_count(&nfsd_file_lru);
 536}
 537
 538static unsigned long
 539nfsd_file_lru_scan(struct shrinker *s, struct shrink_control *sc)
 540{
 541	LIST_HEAD(dispose);
 542	unsigned long ret;
 543
 544	ret = list_lru_shrink_walk(&nfsd_file_lru, sc,
 545				   nfsd_file_lru_cb, &dispose);
 546	trace_nfsd_file_shrinker_removed(ret, list_lru_count(&nfsd_file_lru));
 547	nfsd_file_dispose_list_delayed(&dispose);
 548	return ret;
 549}
 550
 551static struct shrinker *nfsd_file_shrinker;
 552
 553/**
 554 * nfsd_file_cond_queue - conditionally unhash and queue a nfsd_file
 555 * @nf: nfsd_file to attempt to queue
 556 * @dispose: private list to queue successfully-put objects
 557 *
 558 * Unhash an nfsd_file, try to get a reference to it, and then put that
 559 * reference. If it's the last reference, queue it to the dispose list.
 560 */
 561static void
 562nfsd_file_cond_queue(struct nfsd_file *nf, struct list_head *dispose)
 563	__must_hold(RCU)
 564{
 565	int decrement = 1;
 566
 567	/* If we raced with someone else unhashing, ignore it */
 568	if (!nfsd_file_unhash(nf))
 569		return;
 570
 571	/* If we can't get a reference, ignore it */
 572	if (!nfsd_file_get(nf))
 573		return;
 574
 575	/* Extra decrement if we remove from the LRU */
 576	if (nfsd_file_lru_remove(nf))
 577		++decrement;
 578
 579	/* If refcount goes to 0, then put on the dispose list */
 580	if (refcount_sub_and_test(decrement, &nf->nf_ref)) {
 581		list_add(&nf->nf_lru, dispose);
 582		trace_nfsd_file_closing(nf);
 583	}
 584}
 585
 586/**
 587 * nfsd_file_queue_for_close: try to close out any open nfsd_files for an inode
 588 * @inode:   inode on which to close out nfsd_files
 589 * @dispose: list on which to gather nfsd_files to close out
 590 *
 591 * An nfsd_file represents a struct file being held open on behalf of nfsd.
 592 * An open file however can block other activity (such as leases), or cause
 593 * undesirable behavior (e.g. spurious silly-renames when reexporting NFS).
 594 *
 595 * This function is intended to find open nfsd_files when this sort of
 596 * conflicting access occurs and then attempt to close those files out.
 597 *
 598 * Populates the dispose list with entries that have already had their
 599 * refcounts go to zero. The actual free of an nfsd_file can be expensive,
 600 * so we leave it up to the caller whether it wants to wait or not.
 601 */
 602static void
 603nfsd_file_queue_for_close(struct inode *inode, struct list_head *dispose)
 604{
 605	struct rhlist_head *tmp, *list;
 606	struct nfsd_file *nf;
 607
 608	rcu_read_lock();
 609	list = rhltable_lookup(&nfsd_file_rhltable, &inode,
 610			       nfsd_file_rhash_params);
 611	rhl_for_each_entry_rcu(nf, tmp, list, nf_rlist) {
 612		if (!test_bit(NFSD_FILE_GC, &nf->nf_flags))
 613			continue;
 614		nfsd_file_cond_queue(nf, dispose);
 615	}
 616	rcu_read_unlock();
 617}
 618
 619/**
 620 * nfsd_file_close_inode - attempt a delayed close of a nfsd_file
 621 * @inode: inode of the file to attempt to remove
 622 *
 623 * Close out any open nfsd_files that can be reaped for @inode. The
 624 * actual freeing is deferred to the dispose_list_delayed infrastructure.
 625 *
 626 * This is used by the fsnotify callbacks and setlease notifier.
 627 */
 628static void
 629nfsd_file_close_inode(struct inode *inode)
 630{
 631	LIST_HEAD(dispose);
 632
 633	nfsd_file_queue_for_close(inode, &dispose);
 634	nfsd_file_dispose_list_delayed(&dispose);
 635}
 636
 637/**
 638 * nfsd_file_close_inode_sync - attempt to forcibly close a nfsd_file
 639 * @inode: inode of the file to attempt to remove
 640 *
 641 * Close out any open nfsd_files that can be reaped for @inode. The
 642 * nfsd_files are closed out synchronously.
 643 *
 644 * This is called from nfsd_rename and nfsd_unlink to avoid silly-renames
 645 * when reexporting NFS.
 646 */
 647void
 648nfsd_file_close_inode_sync(struct inode *inode)
 649{
 650	struct nfsd_file *nf;
 651	LIST_HEAD(dispose);
 652
 653	trace_nfsd_file_close(inode);
 654
 655	nfsd_file_queue_for_close(inode, &dispose);
 656	while (!list_empty(&dispose)) {
 657		nf = list_first_entry(&dispose, struct nfsd_file, nf_lru);
 658		list_del_init(&nf->nf_lru);
 659		nfsd_file_free(nf);
 660	}
 661}
 662
 663static int
 664nfsd_file_lease_notifier_call(struct notifier_block *nb, unsigned long arg,
 665			    void *data)
 666{
 667	struct file_lock *fl = data;
 668
 669	/* Only close files for F_SETLEASE leases */
 670	if (fl->c.flc_flags & FL_LEASE)
 671		nfsd_file_close_inode(file_inode(fl->c.flc_file));
 672	return 0;
 673}
 674
 675static struct notifier_block nfsd_file_lease_notifier = {
 676	.notifier_call = nfsd_file_lease_notifier_call,
 677};
 678
 679static int
 680nfsd_file_fsnotify_handle_event(struct fsnotify_mark *mark, u32 mask,
 681				struct inode *inode, struct inode *dir,
 682				const struct qstr *name, u32 cookie)
 683{
 684	if (WARN_ON_ONCE(!inode))
 685		return 0;
 686
 687	trace_nfsd_file_fsnotify_handle_event(inode, mask);
 688
 689	/* Should be no marks on non-regular files */
 690	if (!S_ISREG(inode->i_mode)) {
 691		WARN_ON_ONCE(1);
 692		return 0;
 693	}
 694
 695	/* don't close files if this was not the last link */
 696	if (mask & FS_ATTRIB) {
 697		if (inode->i_nlink)
 698			return 0;
 699	}
 700
 701	nfsd_file_close_inode(inode);
 702	return 0;
 703}
 704
 705
 706static const struct fsnotify_ops nfsd_file_fsnotify_ops = {
 707	.handle_inode_event = nfsd_file_fsnotify_handle_event,
 708	.free_mark = nfsd_file_mark_free,
 709};
 710
 711int
 712nfsd_file_cache_init(void)
 713{
 714	int ret;
 715
 716	lockdep_assert_held(&nfsd_mutex);
 717	if (test_and_set_bit(NFSD_FILE_CACHE_UP, &nfsd_file_flags) == 1)
 718		return 0;
 719
 720	ret = rhltable_init(&nfsd_file_rhltable, &nfsd_file_rhash_params);
 721	if (ret)
 722		return ret;
 723
 724	ret = -ENOMEM;
 725	nfsd_file_slab = KMEM_CACHE(nfsd_file, 0);
 726	if (!nfsd_file_slab) {
 727		pr_err("nfsd: unable to create nfsd_file_slab\n");
 728		goto out_err;
 729	}
 730
 731	nfsd_file_mark_slab = KMEM_CACHE(nfsd_file_mark, 0);
 732	if (!nfsd_file_mark_slab) {
 733		pr_err("nfsd: unable to create nfsd_file_mark_slab\n");
 734		goto out_err;
 735	}
 736
 737	ret = list_lru_init(&nfsd_file_lru);
 738	if (ret) {
 739		pr_err("nfsd: failed to init nfsd_file_lru: %d\n", ret);
 740		goto out_err;
 741	}
 742
 743	nfsd_file_shrinker = shrinker_alloc(0, "nfsd-filecache");
 744	if (!nfsd_file_shrinker) {
 745		ret = -ENOMEM;
 746		pr_err("nfsd: failed to allocate nfsd_file_shrinker\n");
 747		goto out_lru;
 748	}
 749
 750	nfsd_file_shrinker->count_objects = nfsd_file_lru_count;
 751	nfsd_file_shrinker->scan_objects = nfsd_file_lru_scan;
 752	nfsd_file_shrinker->seeks = 1;
 753
 754	shrinker_register(nfsd_file_shrinker);
 755
 756	ret = lease_register_notifier(&nfsd_file_lease_notifier);
 757	if (ret) {
 758		pr_err("nfsd: unable to register lease notifier: %d\n", ret);
 759		goto out_shrinker;
 760	}
 761
 762	nfsd_file_fsnotify_group = fsnotify_alloc_group(&nfsd_file_fsnotify_ops,
 763							FSNOTIFY_GROUP_NOFS);
 764	if (IS_ERR(nfsd_file_fsnotify_group)) {
 765		pr_err("nfsd: unable to create fsnotify group: %ld\n",
 766			PTR_ERR(nfsd_file_fsnotify_group));
 767		ret = PTR_ERR(nfsd_file_fsnotify_group);
 768		nfsd_file_fsnotify_group = NULL;
 769		goto out_notifier;
 770	}
 771
 772	INIT_DELAYED_WORK(&nfsd_filecache_laundrette, nfsd_file_gc_worker);
 773out:
 
 
 774	return ret;
 775out_notifier:
 776	lease_unregister_notifier(&nfsd_file_lease_notifier);
 777out_shrinker:
 778	shrinker_free(nfsd_file_shrinker);
 779out_lru:
 780	list_lru_destroy(&nfsd_file_lru);
 781out_err:
 782	kmem_cache_destroy(nfsd_file_slab);
 783	nfsd_file_slab = NULL;
 784	kmem_cache_destroy(nfsd_file_mark_slab);
 785	nfsd_file_mark_slab = NULL;
 786	rhltable_destroy(&nfsd_file_rhltable);
 787	goto out;
 788}
 789
 790/**
 791 * __nfsd_file_cache_purge: clean out the cache for shutdown
 792 * @net: net-namespace to shut down the cache (may be NULL)
 793 *
 794 * Walk the nfsd_file cache and close out any that match @net. If @net is NULL,
 795 * then close out everything. Called when an nfsd instance is being shut down,
 796 * and when the exports table is flushed.
 797 */
 798static void
 799__nfsd_file_cache_purge(struct net *net)
 800{
 801	struct rhashtable_iter iter;
 802	struct nfsd_file *nf;
 803	LIST_HEAD(dispose);
 804
 805	rhltable_walk_enter(&nfsd_file_rhltable, &iter);
 806	do {
 807		rhashtable_walk_start(&iter);
 808
 809		nf = rhashtable_walk_next(&iter);
 810		while (!IS_ERR_OR_NULL(nf)) {
 811			if (!net || nf->nf_net == net)
 812				nfsd_file_cond_queue(nf, &dispose);
 813			nf = rhashtable_walk_next(&iter);
 814		}
 815
 816		rhashtable_walk_stop(&iter);
 817	} while (nf == ERR_PTR(-EAGAIN));
 818	rhashtable_walk_exit(&iter);
 819
 820	nfsd_file_dispose_list(&dispose);
 821}
 822
 823static struct nfsd_fcache_disposal *
 824nfsd_alloc_fcache_disposal(void)
 825{
 826	struct nfsd_fcache_disposal *l;
 827
 828	l = kmalloc(sizeof(*l), GFP_KERNEL);
 829	if (!l)
 830		return NULL;
 831	spin_lock_init(&l->lock);
 832	INIT_LIST_HEAD(&l->freeme);
 833	return l;
 834}
 835
 836static void
 837nfsd_free_fcache_disposal(struct nfsd_fcache_disposal *l)
 838{
 839	nfsd_file_dispose_list(&l->freeme);
 840	kfree(l);
 841}
 842
 843static void
 844nfsd_free_fcache_disposal_net(struct net *net)
 845{
 846	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
 847	struct nfsd_fcache_disposal *l = nn->fcache_disposal;
 848
 849	nfsd_free_fcache_disposal(l);
 850}
 851
 852int
 853nfsd_file_cache_start_net(struct net *net)
 854{
 855	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
 856
 857	nn->fcache_disposal = nfsd_alloc_fcache_disposal();
 858	return nn->fcache_disposal ? 0 : -ENOMEM;
 859}
 860
 861/**
 862 * nfsd_file_cache_purge - Remove all cache items associated with @net
 863 * @net: target net namespace
 864 *
 865 */
 866void
 867nfsd_file_cache_purge(struct net *net)
 868{
 869	lockdep_assert_held(&nfsd_mutex);
 870	if (test_bit(NFSD_FILE_CACHE_UP, &nfsd_file_flags) == 1)
 871		__nfsd_file_cache_purge(net);
 872}
 873
 874void
 875nfsd_file_cache_shutdown_net(struct net *net)
 876{
 877	nfsd_file_cache_purge(net);
 878	nfsd_free_fcache_disposal_net(net);
 879}
 880
 881void
 882nfsd_file_cache_shutdown(void)
 883{
 884	int i;
 885
 886	lockdep_assert_held(&nfsd_mutex);
 887	if (test_and_clear_bit(NFSD_FILE_CACHE_UP, &nfsd_file_flags) == 0)
 888		return;
 889
 890	lease_unregister_notifier(&nfsd_file_lease_notifier);
 891	shrinker_free(nfsd_file_shrinker);
 892	/*
 893	 * make sure all callers of nfsd_file_lru_cb are done before
 894	 * calling nfsd_file_cache_purge
 895	 */
 896	cancel_delayed_work_sync(&nfsd_filecache_laundrette);
 897	__nfsd_file_cache_purge(NULL);
 898	list_lru_destroy(&nfsd_file_lru);
 899	rcu_barrier();
 900	fsnotify_put_group(nfsd_file_fsnotify_group);
 901	nfsd_file_fsnotify_group = NULL;
 902	kmem_cache_destroy(nfsd_file_slab);
 903	nfsd_file_slab = NULL;
 904	fsnotify_wait_marks_destroyed();
 905	kmem_cache_destroy(nfsd_file_mark_slab);
 906	nfsd_file_mark_slab = NULL;
 907	rhltable_destroy(&nfsd_file_rhltable);
 908
 909	for_each_possible_cpu(i) {
 910		per_cpu(nfsd_file_cache_hits, i) = 0;
 911		per_cpu(nfsd_file_acquisitions, i) = 0;
 
 912		per_cpu(nfsd_file_releases, i) = 0;
 913		per_cpu(nfsd_file_total_age, i) = 0;
 914		per_cpu(nfsd_file_evictions, i) = 0;
 915	}
 916}
 917
 918static struct nfsd_file *
 919nfsd_file_lookup_locked(const struct net *net, const struct cred *cred,
 920			struct inode *inode, unsigned char need,
 921			bool want_gc)
 922{
 923	struct rhlist_head *tmp, *list;
 924	struct nfsd_file *nf;
 925
 926	list = rhltable_lookup(&nfsd_file_rhltable, &inode,
 927			       nfsd_file_rhash_params);
 928	rhl_for_each_entry_rcu(nf, tmp, list, nf_rlist) {
 929		if (nf->nf_may != need)
 930			continue;
 931		if (nf->nf_net != net)
 932			continue;
 933		if (!nfsd_match_cred(nf->nf_cred, cred))
 934			continue;
 935		if (test_bit(NFSD_FILE_GC, &nf->nf_flags) != want_gc)
 936			continue;
 937		if (test_bit(NFSD_FILE_HASHED, &nf->nf_flags) == 0)
 938			continue;
 939
 940		if (!nfsd_file_get(nf))
 941			continue;
 942		return nf;
 943	}
 944	return NULL;
 945}
 946
 947/**
 948 * nfsd_file_is_cached - are there any cached open files for this inode?
 949 * @inode: inode to check
 950 *
 951 * The lookup matches inodes in all net namespaces and is atomic wrt
 952 * nfsd_file_acquire().
 953 *
 954 * Return values:
 955 *   %true: filecache contains at least one file matching this inode
 956 *   %false: filecache contains no files matching this inode
 957 */
 958bool
 959nfsd_file_is_cached(struct inode *inode)
 960{
 961	struct rhlist_head *tmp, *list;
 962	struct nfsd_file *nf;
 963	bool ret = false;
 964
 965	rcu_read_lock();
 966	list = rhltable_lookup(&nfsd_file_rhltable, &inode,
 967			       nfsd_file_rhash_params);
 968	rhl_for_each_entry_rcu(nf, tmp, list, nf_rlist)
 969		if (test_bit(NFSD_FILE_GC, &nf->nf_flags)) {
 970			ret = true;
 971			break;
 972		}
 973	rcu_read_unlock();
 974
 975	trace_nfsd_file_is_cached(inode, (int)ret);
 976	return ret;
 977}
 978
 979static __be32
 980nfsd_file_do_acquire(struct svc_rqst *rqstp, struct svc_fh *fhp,
 
 
 
 981		     unsigned int may_flags, struct file *file,
 982		     struct nfsd_file **pnf, bool want_gc)
 983{
 984	unsigned char need = may_flags & NFSD_FILE_MAY_MASK;
 985	struct net *net = SVC_NET(rqstp);
 986	struct nfsd_file *new, *nf;
 987	bool stale_retry = true;
 988	bool open_retry = true;
 989	struct inode *inode;
 990	__be32 status;
 991	int ret;
 992
 993retry:
 994	status = fh_verify(rqstp, fhp, S_IFREG,
 995				may_flags|NFSD_MAY_OWNER_OVERRIDE);
 
 
 
 
 
 996	if (status != nfs_ok)
 997		return status;
 998	inode = d_inode(fhp->fh_dentry);
 999
1000	rcu_read_lock();
1001	nf = nfsd_file_lookup_locked(net, current_cred(), inode, need, want_gc);
1002	rcu_read_unlock();
1003
1004	if (nf) {
1005		/*
1006		 * If the nf is on the LRU then it holds an extra reference
1007		 * that must be put if it's removed. It had better not be
1008		 * the last one however, since we should hold another.
1009		 */
1010		if (nfsd_file_lru_remove(nf))
1011			WARN_ON_ONCE(refcount_dec_and_test(&nf->nf_ref));
1012		goto wait_for_construction;
1013	}
1014
1015	new = nfsd_file_alloc(net, inode, need, want_gc);
1016	if (!new) {
1017		status = nfserr_jukebox;
1018		goto out;
1019	}
1020
1021	rcu_read_lock();
1022	spin_lock(&inode->i_lock);
1023	nf = nfsd_file_lookup_locked(net, current_cred(), inode, need, want_gc);
1024	if (unlikely(nf)) {
1025		spin_unlock(&inode->i_lock);
1026		rcu_read_unlock();
1027		nfsd_file_slab_free(&new->nf_rcu);
1028		goto wait_for_construction;
1029	}
1030	nf = new;
1031	ret = rhltable_insert(&nfsd_file_rhltable, &nf->nf_rlist,
1032			      nfsd_file_rhash_params);
1033	spin_unlock(&inode->i_lock);
1034	rcu_read_unlock();
1035	if (likely(ret == 0))
1036		goto open_file;
1037
1038	if (ret == -EEXIST)
1039		goto retry;
1040	trace_nfsd_file_insert_err(rqstp, inode, may_flags, ret);
1041	status = nfserr_jukebox;
1042	goto construction_err;
1043
1044wait_for_construction:
1045	wait_on_bit(&nf->nf_flags, NFSD_FILE_PENDING, TASK_UNINTERRUPTIBLE);
1046
1047	/* Did construction of this file fail? */
1048	if (!test_bit(NFSD_FILE_HASHED, &nf->nf_flags)) {
1049		trace_nfsd_file_cons_err(rqstp, inode, may_flags, nf);
1050		if (!open_retry) {
1051			status = nfserr_jukebox;
1052			goto construction_err;
1053		}
 
1054		open_retry = false;
1055		fh_put(fhp);
1056		goto retry;
1057	}
1058	this_cpu_inc(nfsd_file_cache_hits);
1059
1060	status = nfserrno(nfsd_open_break_lease(file_inode(nf->nf_file), may_flags));
1061	if (status != nfs_ok) {
1062		nfsd_file_put(nf);
1063		nf = NULL;
1064	}
1065
1066out:
1067	if (status == nfs_ok) {
1068		this_cpu_inc(nfsd_file_acquisitions);
1069		nfsd_file_check_write_error(nf);
1070		*pnf = nf;
1071	}
1072	trace_nfsd_file_acquire(rqstp, inode, may_flags, nf, status);
1073	return status;
1074
1075open_file:
1076	trace_nfsd_file_alloc(nf);
1077	nf->nf_mark = nfsd_file_mark_find_or_create(nf, inode);
1078	if (nf->nf_mark) {
1079		if (file) {
1080			get_file(file);
1081			nf->nf_file = file;
1082			status = nfs_ok;
1083			trace_nfsd_file_opened(nf, status);
1084		} else {
1085			ret = nfsd_open_verified(rqstp, fhp, may_flags,
1086						 &nf->nf_file);
1087			if (ret == -EOPENSTALE && stale_retry) {
1088				stale_retry = false;
1089				nfsd_file_unhash(nf);
1090				clear_and_wake_up_bit(NFSD_FILE_PENDING,
1091						      &nf->nf_flags);
1092				if (refcount_dec_and_test(&nf->nf_ref))
1093					nfsd_file_free(nf);
1094				nf = NULL;
1095				fh_put(fhp);
1096				goto retry;
1097			}
1098			status = nfserrno(ret);
1099			trace_nfsd_file_open(nf, status);
1100		}
1101	} else
1102		status = nfserr_jukebox;
1103	/*
1104	 * If construction failed, or we raced with a call to unlink()
1105	 * then unhash.
1106	 */
1107	if (status != nfs_ok || inode->i_nlink == 0)
1108		nfsd_file_unhash(nf);
1109	clear_and_wake_up_bit(NFSD_FILE_PENDING, &nf->nf_flags);
1110	if (status == nfs_ok)
1111		goto out;
1112
1113construction_err:
1114	if (refcount_dec_and_test(&nf->nf_ref))
1115		nfsd_file_free(nf);
1116	nf = NULL;
1117	goto out;
1118}
1119
1120/**
1121 * nfsd_file_acquire_gc - Get a struct nfsd_file with an open file
1122 * @rqstp: the RPC transaction being executed
1123 * @fhp: the NFS filehandle of the file to be opened
1124 * @may_flags: NFSD_MAY_ settings for the file
1125 * @pnf: OUT: new or found "struct nfsd_file" object
1126 *
1127 * The nfsd_file object returned by this API is reference-counted
1128 * and garbage-collected. The object is retained for a few
1129 * seconds after the final nfsd_file_put() in case the caller
1130 * wants to re-use it.
1131 *
1132 * Return values:
1133 *   %nfs_ok - @pnf points to an nfsd_file with its reference
1134 *   count boosted.
1135 *
1136 * On error, an nfsstat value in network byte order is returned.
1137 */
1138__be32
1139nfsd_file_acquire_gc(struct svc_rqst *rqstp, struct svc_fh *fhp,
1140		     unsigned int may_flags, struct nfsd_file **pnf)
1141{
1142	return nfsd_file_do_acquire(rqstp, fhp, may_flags, NULL, pnf, true);
 
1143}
1144
1145/**
1146 * nfsd_file_acquire - Get a struct nfsd_file with an open file
1147 * @rqstp: the RPC transaction being executed
1148 * @fhp: the NFS filehandle of the file to be opened
1149 * @may_flags: NFSD_MAY_ settings for the file
1150 * @pnf: OUT: new or found "struct nfsd_file" object
1151 *
1152 * The nfsd_file_object returned by this API is reference-counted
1153 * but not garbage-collected. The object is unhashed after the
1154 * final nfsd_file_put().
1155 *
1156 * Return values:
1157 *   %nfs_ok - @pnf points to an nfsd_file with its reference
1158 *   count boosted.
1159 *
1160 * On error, an nfsstat value in network byte order is returned.
1161 */
1162__be32
1163nfsd_file_acquire(struct svc_rqst *rqstp, struct svc_fh *fhp,
1164		  unsigned int may_flags, struct nfsd_file **pnf)
1165{
1166	return nfsd_file_do_acquire(rqstp, fhp, may_flags, NULL, pnf, false);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1167}
1168
1169/**
1170 * nfsd_file_acquire_opened - Get a struct nfsd_file using existing open file
1171 * @rqstp: the RPC transaction being executed
1172 * @fhp: the NFS filehandle of the file just created
1173 * @may_flags: NFSD_MAY_ settings for the file
1174 * @file: cached, already-open file (may be NULL)
1175 * @pnf: OUT: new or found "struct nfsd_file" object
1176 *
1177 * Acquire a nfsd_file object that is not GC'ed. If one doesn't already exist,
1178 * and @file is non-NULL, use it to instantiate a new nfsd_file instead of
1179 * opening a new one.
1180 *
1181 * Return values:
1182 *   %nfs_ok - @pnf points to an nfsd_file with its reference
1183 *   count boosted.
1184 *
1185 * On error, an nfsstat value in network byte order is returned.
1186 */
1187__be32
1188nfsd_file_acquire_opened(struct svc_rqst *rqstp, struct svc_fh *fhp,
1189			 unsigned int may_flags, struct file *file,
1190			 struct nfsd_file **pnf)
1191{
1192	return nfsd_file_do_acquire(rqstp, fhp, may_flags, file, pnf, false);
 
1193}
1194
1195/*
1196 * Note that fields may be added, removed or reordered in the future. Programs
1197 * scraping this file for info should test the labels to ensure they're
1198 * getting the correct field.
1199 */
1200int nfsd_file_cache_stats_show(struct seq_file *m, void *v)
1201{
1202	unsigned long releases = 0, evictions = 0;
1203	unsigned long hits = 0, acquisitions = 0;
1204	unsigned int i, count = 0, buckets = 0;
1205	unsigned long lru = 0, total_age = 0;
1206
1207	/* Serialize with server shutdown */
1208	mutex_lock(&nfsd_mutex);
1209	if (test_bit(NFSD_FILE_CACHE_UP, &nfsd_file_flags) == 1) {
1210		struct bucket_table *tbl;
1211		struct rhashtable *ht;
1212
1213		lru = list_lru_count(&nfsd_file_lru);
1214
1215		rcu_read_lock();
1216		ht = &nfsd_file_rhltable.ht;
1217		count = atomic_read(&ht->nelems);
1218		tbl = rht_dereference_rcu(ht->tbl, ht);
1219		buckets = tbl->size;
1220		rcu_read_unlock();
1221	}
1222	mutex_unlock(&nfsd_mutex);
1223
1224	for_each_possible_cpu(i) {
1225		hits += per_cpu(nfsd_file_cache_hits, i);
1226		acquisitions += per_cpu(nfsd_file_acquisitions, i);
 
1227		releases += per_cpu(nfsd_file_releases, i);
1228		total_age += per_cpu(nfsd_file_total_age, i);
1229		evictions += per_cpu(nfsd_file_evictions, i);
1230	}
1231
1232	seq_printf(m, "total inodes:  %u\n", count);
1233	seq_printf(m, "hash buckets:  %u\n", buckets);
1234	seq_printf(m, "lru entries:   %lu\n", lru);
1235	seq_printf(m, "cache hits:    %lu\n", hits);
1236	seq_printf(m, "acquisitions:  %lu\n", acquisitions);
 
1237	seq_printf(m, "releases:      %lu\n", releases);
1238	seq_printf(m, "evictions:     %lu\n", evictions);
1239	if (releases)
1240		seq_printf(m, "mean age (ms): %ld\n", total_age / releases);
1241	else
1242		seq_printf(m, "mean age (ms): -\n");
1243	return 0;
1244}
v6.13.7
   1// SPDX-License-Identifier: GPL-2.0
   2/*
   3 * The NFSD open file cache.
   4 *
   5 * (c) 2015 - Jeff Layton <jeff.layton@primarydata.com>
   6 *
   7 * An nfsd_file object is a per-file collection of open state that binds
   8 * together:
   9 *   - a struct file *
  10 *   - a user credential
  11 *   - a network namespace
  12 *   - a read-ahead context
  13 *   - monitoring for writeback errors
  14 *
  15 * nfsd_file objects are reference-counted. Consumers acquire a new
  16 * object via the nfsd_file_acquire API. They manage their interest in
  17 * the acquired object, and hence the object's reference count, via
  18 * nfsd_file_get and nfsd_file_put. There are two varieties of nfsd_file
  19 * object:
  20 *
  21 *  * non-garbage-collected: When a consumer wants to precisely control
  22 *    the lifetime of a file's open state, it acquires a non-garbage-
  23 *    collected nfsd_file. The final nfsd_file_put releases the open
  24 *    state immediately.
  25 *
  26 *  * garbage-collected: When a consumer does not control the lifetime
  27 *    of open state, it acquires a garbage-collected nfsd_file. The
  28 *    final nfsd_file_put allows the open state to linger for a period
  29 *    during which it may be re-used.
  30 */
  31
  32#include <linux/hash.h>
  33#include <linux/slab.h>
  34#include <linux/file.h>
  35#include <linux/pagemap.h>
  36#include <linux/sched.h>
  37#include <linux/list_lru.h>
  38#include <linux/fsnotify_backend.h>
  39#include <linux/fsnotify.h>
  40#include <linux/seq_file.h>
  41#include <linux/rhashtable.h>
  42
  43#include "vfs.h"
  44#include "nfsd.h"
  45#include "nfsfh.h"
  46#include "netns.h"
  47#include "filecache.h"
  48#include "trace.h"
  49
  50#define NFSD_LAUNDRETTE_DELAY		     (2 * HZ)
  51
  52#define NFSD_FILE_CACHE_UP		     (0)
  53
  54/* We only care about NFSD_MAY_READ/WRITE for this cache */
  55#define NFSD_FILE_MAY_MASK	(NFSD_MAY_READ|NFSD_MAY_WRITE|NFSD_MAY_LOCALIO)
  56
  57static DEFINE_PER_CPU(unsigned long, nfsd_file_cache_hits);
  58static DEFINE_PER_CPU(unsigned long, nfsd_file_acquisitions);
  59static DEFINE_PER_CPU(unsigned long, nfsd_file_allocations);
  60static DEFINE_PER_CPU(unsigned long, nfsd_file_releases);
  61static DEFINE_PER_CPU(unsigned long, nfsd_file_total_age);
  62static DEFINE_PER_CPU(unsigned long, nfsd_file_evictions);
  63
  64struct nfsd_fcache_disposal {
  65	spinlock_t lock;
  66	struct list_head freeme;
  67};
  68
  69static struct kmem_cache		*nfsd_file_slab;
  70static struct kmem_cache		*nfsd_file_mark_slab;
  71static struct list_lru			nfsd_file_lru;
  72static unsigned long			nfsd_file_flags;
  73static struct fsnotify_group		*nfsd_file_fsnotify_group;
  74static struct delayed_work		nfsd_filecache_laundrette;
  75static struct rhltable			nfsd_file_rhltable
  76						____cacheline_aligned_in_smp;
  77
  78static bool
  79nfsd_match_cred(const struct cred *c1, const struct cred *c2)
  80{
  81	int i;
  82
  83	if (!uid_eq(c1->fsuid, c2->fsuid))
  84		return false;
  85	if (!gid_eq(c1->fsgid, c2->fsgid))
  86		return false;
  87	if (c1->group_info == NULL || c2->group_info == NULL)
  88		return c1->group_info == c2->group_info;
  89	if (c1->group_info->ngroups != c2->group_info->ngroups)
  90		return false;
  91	for (i = 0; i < c1->group_info->ngroups; i++) {
  92		if (!gid_eq(c1->group_info->gid[i], c2->group_info->gid[i]))
  93			return false;
  94	}
  95	return true;
  96}
  97
  98static const struct rhashtable_params nfsd_file_rhash_params = {
  99	.key_len		= sizeof_field(struct nfsd_file, nf_inode),
 100	.key_offset		= offsetof(struct nfsd_file, nf_inode),
 101	.head_offset		= offsetof(struct nfsd_file, nf_rlist),
 102
 103	/*
 104	 * Start with a single page hash table to reduce resizing churn
 105	 * on light workloads.
 106	 */
 107	.min_size		= 256,
 108	.automatic_shrinking	= true,
 109};
 110
 111static void
 112nfsd_file_schedule_laundrette(void)
 113{
 114	if (test_bit(NFSD_FILE_CACHE_UP, &nfsd_file_flags))
 115		queue_delayed_work(system_unbound_wq, &nfsd_filecache_laundrette,
 116				   NFSD_LAUNDRETTE_DELAY);
 117}
 118
 119static void
 120nfsd_file_slab_free(struct rcu_head *rcu)
 121{
 122	struct nfsd_file *nf = container_of(rcu, struct nfsd_file, nf_rcu);
 123
 124	put_cred(nf->nf_cred);
 125	kmem_cache_free(nfsd_file_slab, nf);
 126}
 127
 128static void
 129nfsd_file_mark_free(struct fsnotify_mark *mark)
 130{
 131	struct nfsd_file_mark *nfm = container_of(mark, struct nfsd_file_mark,
 132						  nfm_mark);
 133
 134	kmem_cache_free(nfsd_file_mark_slab, nfm);
 135}
 136
 137static struct nfsd_file_mark *
 138nfsd_file_mark_get(struct nfsd_file_mark *nfm)
 139{
 140	if (!refcount_inc_not_zero(&nfm->nfm_ref))
 141		return NULL;
 142	return nfm;
 143}
 144
 145static void
 146nfsd_file_mark_put(struct nfsd_file_mark *nfm)
 147{
 148	if (refcount_dec_and_test(&nfm->nfm_ref)) {
 149		fsnotify_destroy_mark(&nfm->nfm_mark, nfsd_file_fsnotify_group);
 150		fsnotify_put_mark(&nfm->nfm_mark);
 151	}
 152}
 153
 154static struct nfsd_file_mark *
 155nfsd_file_mark_find_or_create(struct inode *inode)
 156{
 157	int			err;
 158	struct fsnotify_mark	*mark;
 159	struct nfsd_file_mark	*nfm = NULL, *new;
 160
 161	do {
 162		fsnotify_group_lock(nfsd_file_fsnotify_group);
 163		mark = fsnotify_find_inode_mark(inode,
 164						nfsd_file_fsnotify_group);
 165		if (mark) {
 166			nfm = nfsd_file_mark_get(container_of(mark,
 167						 struct nfsd_file_mark,
 168						 nfm_mark));
 169			fsnotify_group_unlock(nfsd_file_fsnotify_group);
 170			if (nfm) {
 171				fsnotify_put_mark(mark);
 172				break;
 173			}
 174			/* Avoid soft lockup race with nfsd_file_mark_put() */
 175			fsnotify_destroy_mark(mark, nfsd_file_fsnotify_group);
 176			fsnotify_put_mark(mark);
 177		} else {
 178			fsnotify_group_unlock(nfsd_file_fsnotify_group);
 179		}
 180
 181		/* allocate a new nfm */
 182		new = kmem_cache_alloc(nfsd_file_mark_slab, GFP_KERNEL);
 183		if (!new)
 184			return NULL;
 185		fsnotify_init_mark(&new->nfm_mark, nfsd_file_fsnotify_group);
 186		new->nfm_mark.mask = FS_ATTRIB|FS_DELETE_SELF;
 187		refcount_set(&new->nfm_ref, 1);
 188
 189		err = fsnotify_add_inode_mark(&new->nfm_mark, inode, 0);
 190
 191		/*
 192		 * If the add was successful, then return the object.
 193		 * Otherwise, we need to put the reference we hold on the
 194		 * nfm_mark. The fsnotify code will take a reference and put
 195		 * it on failure, so we can't just free it directly. It's also
 196		 * not safe to call fsnotify_destroy_mark on it as the
 197		 * mark->group will be NULL. Thus, we can't let the nfm_ref
 198		 * counter drive the destruction at this point.
 199		 */
 200		if (likely(!err))
 201			nfm = new;
 202		else
 203			fsnotify_put_mark(&new->nfm_mark);
 204	} while (unlikely(err == -EEXIST));
 205
 206	return nfm;
 207}
 208
 209static struct nfsd_file *
 210nfsd_file_alloc(struct net *net, struct inode *inode, unsigned char need,
 211		bool want_gc)
 212{
 213	struct nfsd_file *nf;
 214
 215	nf = kmem_cache_alloc(nfsd_file_slab, GFP_KERNEL);
 216	if (unlikely(!nf))
 217		return NULL;
 218
 219	this_cpu_inc(nfsd_file_allocations);
 220	INIT_LIST_HEAD(&nf->nf_lru);
 221	INIT_LIST_HEAD(&nf->nf_gc);
 222	nf->nf_birthtime = ktime_get();
 223	nf->nf_file = NULL;
 224	nf->nf_cred = get_current_cred();
 225	nf->nf_net = net;
 226	nf->nf_flags = want_gc ?
 227		BIT(NFSD_FILE_HASHED) | BIT(NFSD_FILE_PENDING) | BIT(NFSD_FILE_GC) :
 228		BIT(NFSD_FILE_HASHED) | BIT(NFSD_FILE_PENDING);
 229	nf->nf_inode = inode;
 230	refcount_set(&nf->nf_ref, 1);
 231	nf->nf_may = need;
 232	nf->nf_mark = NULL;
 233	return nf;
 234}
 235
 236/**
 237 * nfsd_file_check_write_error - check for writeback errors on a file
 238 * @nf: nfsd_file to check for writeback errors
 239 *
 240 * Check whether a nfsd_file has an unseen error. Reset the write
 241 * verifier if so.
 242 */
 243static void
 244nfsd_file_check_write_error(struct nfsd_file *nf)
 245{
 246	struct file *file = nf->nf_file;
 247
 248	if ((file->f_mode & FMODE_WRITE) &&
 249	    filemap_check_wb_err(file->f_mapping, READ_ONCE(file->f_wb_err)))
 250		nfsd_reset_write_verifier(net_generic(nf->nf_net, nfsd_net_id));
 251}
 252
 253static void
 254nfsd_file_hash_remove(struct nfsd_file *nf)
 255{
 256	trace_nfsd_file_unhash(nf);
 257	rhltable_remove(&nfsd_file_rhltable, &nf->nf_rlist,
 258			nfsd_file_rhash_params);
 259}
 260
 261static bool
 262nfsd_file_unhash(struct nfsd_file *nf)
 263{
 264	if (test_and_clear_bit(NFSD_FILE_HASHED, &nf->nf_flags)) {
 265		nfsd_file_hash_remove(nf);
 266		return true;
 267	}
 268	return false;
 269}
 270
 271static void
 272nfsd_file_free(struct nfsd_file *nf)
 273{
 274	s64 age = ktime_to_ms(ktime_sub(ktime_get(), nf->nf_birthtime));
 275
 276	trace_nfsd_file_free(nf);
 277
 278	this_cpu_inc(nfsd_file_releases);
 279	this_cpu_add(nfsd_file_total_age, age);
 280
 281	nfsd_file_unhash(nf);
 282	if (nf->nf_mark)
 283		nfsd_file_mark_put(nf->nf_mark);
 284	if (nf->nf_file) {
 285		nfsd_file_check_write_error(nf);
 286		nfsd_filp_close(nf->nf_file);
 287	}
 288
 289	/*
 290	 * If this item is still linked via nf_lru, that's a bug.
 291	 * WARN and leak it to preserve system stability.
 292	 */
 293	if (WARN_ON_ONCE(!list_empty(&nf->nf_lru)))
 294		return;
 295
 296	call_rcu(&nf->nf_rcu, nfsd_file_slab_free);
 297}
 298
 299static bool
 300nfsd_file_check_writeback(struct nfsd_file *nf)
 301{
 302	struct file *file = nf->nf_file;
 303	struct address_space *mapping;
 304
 305	/* File not open for write? */
 306	if (!(file->f_mode & FMODE_WRITE))
 307		return false;
 308
 309	/*
 310	 * Some filesystems (e.g. NFS) flush all dirty data on close.
 311	 * On others, there is no need to wait for writeback.
 312	 */
 313	if (!(file_inode(file)->i_sb->s_export_op->flags & EXPORT_OP_FLUSH_ON_CLOSE))
 314		return false;
 315
 316	mapping = file->f_mapping;
 317	return mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) ||
 318		mapping_tagged(mapping, PAGECACHE_TAG_WRITEBACK);
 319}
 320
 321
 322static bool nfsd_file_lru_add(struct nfsd_file *nf)
 323{
 324	set_bit(NFSD_FILE_REFERENCED, &nf->nf_flags);
 325	if (list_lru_add_obj(&nfsd_file_lru, &nf->nf_lru)) {
 326		trace_nfsd_file_lru_add(nf);
 327		return true;
 328	}
 329	return false;
 330}
 331
 332static bool nfsd_file_lru_remove(struct nfsd_file *nf)
 333{
 334	if (list_lru_del_obj(&nfsd_file_lru, &nf->nf_lru)) {
 335		trace_nfsd_file_lru_del(nf);
 336		return true;
 337	}
 338	return false;
 339}
 340
 341struct nfsd_file *
 342nfsd_file_get(struct nfsd_file *nf)
 343{
 344	if (nf && refcount_inc_not_zero(&nf->nf_ref))
 345		return nf;
 346	return NULL;
 347}
 348
 349/**
 350 * nfsd_file_put - put the reference to a nfsd_file
 351 * @nf: nfsd_file of which to put the reference
 352 *
 353 * Put a reference to a nfsd_file. In the non-GC case, we just put the
 354 * reference immediately. In the GC case, if the reference would be
 355 * the last one, the put it on the LRU instead to be cleaned up later.
 356 */
 357void
 358nfsd_file_put(struct nfsd_file *nf)
 359{
 360	might_sleep();
 361	trace_nfsd_file_put(nf);
 362
 363	if (test_bit(NFSD_FILE_GC, &nf->nf_flags) &&
 364	    test_bit(NFSD_FILE_HASHED, &nf->nf_flags)) {
 365		/*
 366		 * If this is the last reference (nf_ref == 1), then try to
 367		 * transfer it to the LRU.
 368		 */
 369		if (refcount_dec_not_one(&nf->nf_ref))
 370			return;
 371
 372		/* Try to add it to the LRU.  If that fails, decrement. */
 373		if (nfsd_file_lru_add(nf)) {
 374			/* If it's still hashed, we're done */
 375			if (test_bit(NFSD_FILE_HASHED, &nf->nf_flags)) {
 376				nfsd_file_schedule_laundrette();
 377				return;
 378			}
 379
 380			/*
 381			 * We're racing with unhashing, so try to remove it from
 382			 * the LRU. If removal fails, then someone else already
 383			 * has our reference.
 384			 */
 385			if (!nfsd_file_lru_remove(nf))
 386				return;
 387		}
 388	}
 389	if (refcount_dec_and_test(&nf->nf_ref))
 390		nfsd_file_free(nf);
 391}
 392
 393/**
 394 * nfsd_file_put_local - put nfsd_file reference and arm nfsd_serv_put in caller
 395 * @nf: nfsd_file of which to put the reference
 396 *
 397 * First save the associated net to return to caller, then put
 398 * the reference of the nfsd_file.
 399 */
 400struct net *
 401nfsd_file_put_local(struct nfsd_file *nf)
 402{
 403	struct net *net = nf->nf_net;
 404
 405	nfsd_file_put(nf);
 406	return net;
 407}
 408
 409/**
 410 * nfsd_file_file - get the backing file of an nfsd_file
 411 * @nf: nfsd_file of which to access the backing file.
 412 *
 413 * Return backing file for @nf.
 414 */
 415struct file *
 416nfsd_file_file(struct nfsd_file *nf)
 417{
 418	return nf->nf_file;
 419}
 420
 421static void
 422nfsd_file_dispose_list(struct list_head *dispose)
 423{
 424	struct nfsd_file *nf;
 425
 426	while (!list_empty(dispose)) {
 427		nf = list_first_entry(dispose, struct nfsd_file, nf_gc);
 428		list_del_init(&nf->nf_gc);
 429		nfsd_file_free(nf);
 430	}
 431}
 432
 433/**
 434 * nfsd_file_dispose_list_delayed - move list of dead files to net's freeme list
 435 * @dispose: list of nfsd_files to be disposed
 436 *
 437 * Transfers each file to the "freeme" list for its nfsd_net, to eventually
 438 * be disposed of by the per-net garbage collector.
 439 */
 440static void
 441nfsd_file_dispose_list_delayed(struct list_head *dispose)
 442{
 443	while(!list_empty(dispose)) {
 444		struct nfsd_file *nf = list_first_entry(dispose,
 445						struct nfsd_file, nf_gc);
 446		struct nfsd_net *nn = net_generic(nf->nf_net, nfsd_net_id);
 447		struct nfsd_fcache_disposal *l = nn->fcache_disposal;
 448		struct svc_serv *serv;
 449
 450		spin_lock(&l->lock);
 451		list_move_tail(&nf->nf_gc, &l->freeme);
 452		spin_unlock(&l->lock);
 453
 454		/*
 455		 * The filecache laundrette is shut down after the
 456		 * nn->nfsd_serv pointer is cleared, but before the
 457		 * svc_serv is freed.
 458		 */
 459		serv = nn->nfsd_serv;
 460		if (serv)
 461			svc_wake_up(serv);
 462	}
 463}
 464
 465/**
 466 * nfsd_file_net_dispose - deal with nfsd_files waiting to be disposed.
 467 * @nn: nfsd_net in which to find files to be disposed.
 468 *
 469 * When files held open for nfsv3 are removed from the filecache, whether
 470 * due to memory pressure or garbage collection, they are queued to
 471 * a per-net-ns queue.  This function completes the disposal, either
 472 * directly or by waking another nfsd thread to help with the work.
 473 */
 474void nfsd_file_net_dispose(struct nfsd_net *nn)
 475{
 476	struct nfsd_fcache_disposal *l = nn->fcache_disposal;
 477
 478	if (!list_empty(&l->freeme)) {
 479		LIST_HEAD(dispose);
 480		int i;
 481
 482		spin_lock(&l->lock);
 483		for (i = 0; i < 8 && !list_empty(&l->freeme); i++)
 484			list_move(l->freeme.next, &dispose);
 485		spin_unlock(&l->lock);
 486		if (!list_empty(&l->freeme))
 487			/* Wake up another thread to share the work
 488			 * *before* doing any actual disposing.
 489			 */
 490			svc_wake_up(nn->nfsd_serv);
 491		nfsd_file_dispose_list(&dispose);
 492	}
 493}
 494
 495/**
 496 * nfsd_file_lru_cb - Examine an entry on the LRU list
 497 * @item: LRU entry to examine
 498 * @lru: controlling LRU
 
 499 * @arg: dispose list
 500 *
 501 * Return values:
 502 *   %LRU_REMOVED: @item was removed from the LRU
 503 *   %LRU_ROTATE: @item is to be moved to the LRU tail
 504 *   %LRU_SKIP: @item cannot be evicted
 505 */
 506static enum lru_status
 507nfsd_file_lru_cb(struct list_head *item, struct list_lru_one *lru,
 508		 void *arg)
 
 
 509{
 510	struct list_head *head = arg;
 511	struct nfsd_file *nf = list_entry(item, struct nfsd_file, nf_lru);
 512
 513	/* We should only be dealing with GC entries here */
 514	WARN_ON_ONCE(!test_bit(NFSD_FILE_GC, &nf->nf_flags));
 515
 516	/*
 517	 * Don't throw out files that are still undergoing I/O or
 518	 * that have uncleared errors pending.
 519	 */
 520	if (nfsd_file_check_writeback(nf)) {
 521		trace_nfsd_file_gc_writeback(nf);
 522		return LRU_SKIP;
 523	}
 524
 525	/* If it was recently added to the list, skip it */
 526	if (test_and_clear_bit(NFSD_FILE_REFERENCED, &nf->nf_flags)) {
 527		trace_nfsd_file_gc_referenced(nf);
 528		return LRU_ROTATE;
 529	}
 530
 531	/*
 532	 * Put the reference held on behalf of the LRU. If it wasn't the last
 533	 * one, then just remove it from the LRU and ignore it.
 534	 */
 535	if (!refcount_dec_and_test(&nf->nf_ref)) {
 536		trace_nfsd_file_gc_in_use(nf);
 537		list_lru_isolate(lru, &nf->nf_lru);
 538		return LRU_REMOVED;
 539	}
 540
 541	/* Refcount went to zero. Unhash it and queue it to the dispose list */
 542	nfsd_file_unhash(nf);
 543	list_lru_isolate(lru, &nf->nf_lru);
 544	list_add(&nf->nf_gc, head);
 545	this_cpu_inc(nfsd_file_evictions);
 546	trace_nfsd_file_gc_disposed(nf);
 547	return LRU_REMOVED;
 548}
 549
 550static void
 551nfsd_file_gc(void)
 552{
 553	LIST_HEAD(dispose);
 554	unsigned long ret;
 555
 556	ret = list_lru_walk(&nfsd_file_lru, nfsd_file_lru_cb,
 557			    &dispose, list_lru_count(&nfsd_file_lru));
 558	trace_nfsd_file_gc_removed(ret, list_lru_count(&nfsd_file_lru));
 559	nfsd_file_dispose_list_delayed(&dispose);
 560}
 561
 562static void
 563nfsd_file_gc_worker(struct work_struct *work)
 564{
 565	nfsd_file_gc();
 566	if (list_lru_count(&nfsd_file_lru))
 567		nfsd_file_schedule_laundrette();
 568}
 569
 570static unsigned long
 571nfsd_file_lru_count(struct shrinker *s, struct shrink_control *sc)
 572{
 573	return list_lru_count(&nfsd_file_lru);
 574}
 575
 576static unsigned long
 577nfsd_file_lru_scan(struct shrinker *s, struct shrink_control *sc)
 578{
 579	LIST_HEAD(dispose);
 580	unsigned long ret;
 581
 582	ret = list_lru_shrink_walk(&nfsd_file_lru, sc,
 583				   nfsd_file_lru_cb, &dispose);
 584	trace_nfsd_file_shrinker_removed(ret, list_lru_count(&nfsd_file_lru));
 585	nfsd_file_dispose_list_delayed(&dispose);
 586	return ret;
 587}
 588
 589static struct shrinker *nfsd_file_shrinker;
 590
 591/**
 592 * nfsd_file_cond_queue - conditionally unhash and queue a nfsd_file
 593 * @nf: nfsd_file to attempt to queue
 594 * @dispose: private list to queue successfully-put objects
 595 *
 596 * Unhash an nfsd_file, try to get a reference to it, and then put that
 597 * reference. If it's the last reference, queue it to the dispose list.
 598 */
 599static void
 600nfsd_file_cond_queue(struct nfsd_file *nf, struct list_head *dispose)
 601	__must_hold(RCU)
 602{
 603	int decrement = 1;
 604
 605	/* If we raced with someone else unhashing, ignore it */
 606	if (!nfsd_file_unhash(nf))
 607		return;
 608
 609	/* If we can't get a reference, ignore it */
 610	if (!nfsd_file_get(nf))
 611		return;
 612
 613	/* Extra decrement if we remove from the LRU */
 614	if (nfsd_file_lru_remove(nf))
 615		++decrement;
 616
 617	/* If refcount goes to 0, then put on the dispose list */
 618	if (refcount_sub_and_test(decrement, &nf->nf_ref)) {
 619		list_add(&nf->nf_gc, dispose);
 620		trace_nfsd_file_closing(nf);
 621	}
 622}
 623
 624/**
 625 * nfsd_file_queue_for_close: try to close out any open nfsd_files for an inode
 626 * @inode:   inode on which to close out nfsd_files
 627 * @dispose: list on which to gather nfsd_files to close out
 628 *
 629 * An nfsd_file represents a struct file being held open on behalf of nfsd.
 630 * An open file however can block other activity (such as leases), or cause
 631 * undesirable behavior (e.g. spurious silly-renames when reexporting NFS).
 632 *
 633 * This function is intended to find open nfsd_files when this sort of
 634 * conflicting access occurs and then attempt to close those files out.
 635 *
 636 * Populates the dispose list with entries that have already had their
 637 * refcounts go to zero. The actual free of an nfsd_file can be expensive,
 638 * so we leave it up to the caller whether it wants to wait or not.
 639 */
 640static void
 641nfsd_file_queue_for_close(struct inode *inode, struct list_head *dispose)
 642{
 643	struct rhlist_head *tmp, *list;
 644	struct nfsd_file *nf;
 645
 646	rcu_read_lock();
 647	list = rhltable_lookup(&nfsd_file_rhltable, &inode,
 648			       nfsd_file_rhash_params);
 649	rhl_for_each_entry_rcu(nf, tmp, list, nf_rlist) {
 650		if (!test_bit(NFSD_FILE_GC, &nf->nf_flags))
 651			continue;
 652		nfsd_file_cond_queue(nf, dispose);
 653	}
 654	rcu_read_unlock();
 655}
 656
 657/**
 658 * nfsd_file_close_inode - attempt a delayed close of a nfsd_file
 659 * @inode: inode of the file to attempt to remove
 660 *
 661 * Close out any open nfsd_files that can be reaped for @inode. The
 662 * actual freeing is deferred to the dispose_list_delayed infrastructure.
 663 *
 664 * This is used by the fsnotify callbacks and setlease notifier.
 665 */
 666static void
 667nfsd_file_close_inode(struct inode *inode)
 668{
 669	LIST_HEAD(dispose);
 670
 671	nfsd_file_queue_for_close(inode, &dispose);
 672	nfsd_file_dispose_list_delayed(&dispose);
 673}
 674
 675/**
 676 * nfsd_file_close_inode_sync - attempt to forcibly close a nfsd_file
 677 * @inode: inode of the file to attempt to remove
 678 *
 679 * Close out any open nfsd_files that can be reaped for @inode. The
 680 * nfsd_files are closed out synchronously.
 681 *
 682 * This is called from nfsd_rename and nfsd_unlink to avoid silly-renames
 683 * when reexporting NFS.
 684 */
 685void
 686nfsd_file_close_inode_sync(struct inode *inode)
 687{
 688	struct nfsd_file *nf;
 689	LIST_HEAD(dispose);
 690
 691	trace_nfsd_file_close(inode);
 692
 693	nfsd_file_queue_for_close(inode, &dispose);
 694	while (!list_empty(&dispose)) {
 695		nf = list_first_entry(&dispose, struct nfsd_file, nf_gc);
 696		list_del_init(&nf->nf_gc);
 697		nfsd_file_free(nf);
 698	}
 699}
 700
 701static int
 702nfsd_file_lease_notifier_call(struct notifier_block *nb, unsigned long arg,
 703			    void *data)
 704{
 705	struct file_lease *fl = data;
 706
 707	/* Only close files for F_SETLEASE leases */
 708	if (fl->c.flc_flags & FL_LEASE)
 709		nfsd_file_close_inode(file_inode(fl->c.flc_file));
 710	return 0;
 711}
 712
 713static struct notifier_block nfsd_file_lease_notifier = {
 714	.notifier_call = nfsd_file_lease_notifier_call,
 715};
 716
 717static int
 718nfsd_file_fsnotify_handle_event(struct fsnotify_mark *mark, u32 mask,
 719				struct inode *inode, struct inode *dir,
 720				const struct qstr *name, u32 cookie)
 721{
 722	if (WARN_ON_ONCE(!inode))
 723		return 0;
 724
 725	trace_nfsd_file_fsnotify_handle_event(inode, mask);
 726
 727	/* Should be no marks on non-regular files */
 728	if (!S_ISREG(inode->i_mode)) {
 729		WARN_ON_ONCE(1);
 730		return 0;
 731	}
 732
 733	/* don't close files if this was not the last link */
 734	if (mask & FS_ATTRIB) {
 735		if (inode->i_nlink)
 736			return 0;
 737	}
 738
 739	nfsd_file_close_inode(inode);
 740	return 0;
 741}
 742
 743
 744static const struct fsnotify_ops nfsd_file_fsnotify_ops = {
 745	.handle_inode_event = nfsd_file_fsnotify_handle_event,
 746	.free_mark = nfsd_file_mark_free,
 747};
 748
 749int
 750nfsd_file_cache_init(void)
 751{
 752	int ret;
 753
 754	lockdep_assert_held(&nfsd_mutex);
 755	if (test_and_set_bit(NFSD_FILE_CACHE_UP, &nfsd_file_flags) == 1)
 756		return 0;
 757
 758	ret = rhltable_init(&nfsd_file_rhltable, &nfsd_file_rhash_params);
 759	if (ret)
 760		goto out;
 761
 762	ret = -ENOMEM;
 763	nfsd_file_slab = KMEM_CACHE(nfsd_file, 0);
 764	if (!nfsd_file_slab) {
 765		pr_err("nfsd: unable to create nfsd_file_slab\n");
 766		goto out_err;
 767	}
 768
 769	nfsd_file_mark_slab = KMEM_CACHE(nfsd_file_mark, 0);
 770	if (!nfsd_file_mark_slab) {
 771		pr_err("nfsd: unable to create nfsd_file_mark_slab\n");
 772		goto out_err;
 773	}
 774
 775	ret = list_lru_init(&nfsd_file_lru);
 776	if (ret) {
 777		pr_err("nfsd: failed to init nfsd_file_lru: %d\n", ret);
 778		goto out_err;
 779	}
 780
 781	nfsd_file_shrinker = shrinker_alloc(0, "nfsd-filecache");
 782	if (!nfsd_file_shrinker) {
 783		ret = -ENOMEM;
 784		pr_err("nfsd: failed to allocate nfsd_file_shrinker\n");
 785		goto out_lru;
 786	}
 787
 788	nfsd_file_shrinker->count_objects = nfsd_file_lru_count;
 789	nfsd_file_shrinker->scan_objects = nfsd_file_lru_scan;
 790	nfsd_file_shrinker->seeks = 1;
 791
 792	shrinker_register(nfsd_file_shrinker);
 793
 794	ret = lease_register_notifier(&nfsd_file_lease_notifier);
 795	if (ret) {
 796		pr_err("nfsd: unable to register lease notifier: %d\n", ret);
 797		goto out_shrinker;
 798	}
 799
 800	nfsd_file_fsnotify_group = fsnotify_alloc_group(&nfsd_file_fsnotify_ops,
 801							0);
 802	if (IS_ERR(nfsd_file_fsnotify_group)) {
 803		pr_err("nfsd: unable to create fsnotify group: %ld\n",
 804			PTR_ERR(nfsd_file_fsnotify_group));
 805		ret = PTR_ERR(nfsd_file_fsnotify_group);
 806		nfsd_file_fsnotify_group = NULL;
 807		goto out_notifier;
 808	}
 809
 810	INIT_DELAYED_WORK(&nfsd_filecache_laundrette, nfsd_file_gc_worker);
 811out:
 812	if (ret)
 813		clear_bit(NFSD_FILE_CACHE_UP, &nfsd_file_flags);
 814	return ret;
 815out_notifier:
 816	lease_unregister_notifier(&nfsd_file_lease_notifier);
 817out_shrinker:
 818	shrinker_free(nfsd_file_shrinker);
 819out_lru:
 820	list_lru_destroy(&nfsd_file_lru);
 821out_err:
 822	kmem_cache_destroy(nfsd_file_slab);
 823	nfsd_file_slab = NULL;
 824	kmem_cache_destroy(nfsd_file_mark_slab);
 825	nfsd_file_mark_slab = NULL;
 826	rhltable_destroy(&nfsd_file_rhltable);
 827	goto out;
 828}
 829
 830/**
 831 * __nfsd_file_cache_purge: clean out the cache for shutdown
 832 * @net: net-namespace to shut down the cache (may be NULL)
 833 *
 834 * Walk the nfsd_file cache and close out any that match @net. If @net is NULL,
 835 * then close out everything. Called when an nfsd instance is being shut down,
 836 * and when the exports table is flushed.
 837 */
 838static void
 839__nfsd_file_cache_purge(struct net *net)
 840{
 841	struct rhashtable_iter iter;
 842	struct nfsd_file *nf;
 843	LIST_HEAD(dispose);
 844
 845	rhltable_walk_enter(&nfsd_file_rhltable, &iter);
 846	do {
 847		rhashtable_walk_start(&iter);
 848
 849		nf = rhashtable_walk_next(&iter);
 850		while (!IS_ERR_OR_NULL(nf)) {
 851			if (!net || nf->nf_net == net)
 852				nfsd_file_cond_queue(nf, &dispose);
 853			nf = rhashtable_walk_next(&iter);
 854		}
 855
 856		rhashtable_walk_stop(&iter);
 857	} while (nf == ERR_PTR(-EAGAIN));
 858	rhashtable_walk_exit(&iter);
 859
 860	nfsd_file_dispose_list(&dispose);
 861}
 862
 863static struct nfsd_fcache_disposal *
 864nfsd_alloc_fcache_disposal(void)
 865{
 866	struct nfsd_fcache_disposal *l;
 867
 868	l = kmalloc(sizeof(*l), GFP_KERNEL);
 869	if (!l)
 870		return NULL;
 871	spin_lock_init(&l->lock);
 872	INIT_LIST_HEAD(&l->freeme);
 873	return l;
 874}
 875
 876static void
 877nfsd_free_fcache_disposal(struct nfsd_fcache_disposal *l)
 878{
 879	nfsd_file_dispose_list(&l->freeme);
 880	kfree(l);
 881}
 882
 883static void
 884nfsd_free_fcache_disposal_net(struct net *net)
 885{
 886	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
 887	struct nfsd_fcache_disposal *l = nn->fcache_disposal;
 888
 889	nfsd_free_fcache_disposal(l);
 890}
 891
 892int
 893nfsd_file_cache_start_net(struct net *net)
 894{
 895	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
 896
 897	nn->fcache_disposal = nfsd_alloc_fcache_disposal();
 898	return nn->fcache_disposal ? 0 : -ENOMEM;
 899}
 900
 901/**
 902 * nfsd_file_cache_purge - Remove all cache items associated with @net
 903 * @net: target net namespace
 904 *
 905 */
 906void
 907nfsd_file_cache_purge(struct net *net)
 908{
 909	lockdep_assert_held(&nfsd_mutex);
 910	if (test_bit(NFSD_FILE_CACHE_UP, &nfsd_file_flags) == 1)
 911		__nfsd_file_cache_purge(net);
 912}
 913
 914void
 915nfsd_file_cache_shutdown_net(struct net *net)
 916{
 917	nfsd_file_cache_purge(net);
 918	nfsd_free_fcache_disposal_net(net);
 919}
 920
 921void
 922nfsd_file_cache_shutdown(void)
 923{
 924	int i;
 925
 926	lockdep_assert_held(&nfsd_mutex);
 927	if (test_and_clear_bit(NFSD_FILE_CACHE_UP, &nfsd_file_flags) == 0)
 928		return;
 929
 930	lease_unregister_notifier(&nfsd_file_lease_notifier);
 931	shrinker_free(nfsd_file_shrinker);
 932	/*
 933	 * make sure all callers of nfsd_file_lru_cb are done before
 934	 * calling nfsd_file_cache_purge
 935	 */
 936	cancel_delayed_work_sync(&nfsd_filecache_laundrette);
 937	__nfsd_file_cache_purge(NULL);
 938	list_lru_destroy(&nfsd_file_lru);
 939	rcu_barrier();
 940	fsnotify_put_group(nfsd_file_fsnotify_group);
 941	nfsd_file_fsnotify_group = NULL;
 942	kmem_cache_destroy(nfsd_file_slab);
 943	nfsd_file_slab = NULL;
 944	fsnotify_wait_marks_destroyed();
 945	kmem_cache_destroy(nfsd_file_mark_slab);
 946	nfsd_file_mark_slab = NULL;
 947	rhltable_destroy(&nfsd_file_rhltable);
 948
 949	for_each_possible_cpu(i) {
 950		per_cpu(nfsd_file_cache_hits, i) = 0;
 951		per_cpu(nfsd_file_acquisitions, i) = 0;
 952		per_cpu(nfsd_file_allocations, i) = 0;
 953		per_cpu(nfsd_file_releases, i) = 0;
 954		per_cpu(nfsd_file_total_age, i) = 0;
 955		per_cpu(nfsd_file_evictions, i) = 0;
 956	}
 957}
 958
 959static struct nfsd_file *
 960nfsd_file_lookup_locked(const struct net *net, const struct cred *cred,
 961			struct inode *inode, unsigned char need,
 962			bool want_gc)
 963{
 964	struct rhlist_head *tmp, *list;
 965	struct nfsd_file *nf;
 966
 967	list = rhltable_lookup(&nfsd_file_rhltable, &inode,
 968			       nfsd_file_rhash_params);
 969	rhl_for_each_entry_rcu(nf, tmp, list, nf_rlist) {
 970		if (nf->nf_may != need)
 971			continue;
 972		if (nf->nf_net != net)
 973			continue;
 974		if (!nfsd_match_cred(nf->nf_cred, cred))
 975			continue;
 976		if (test_bit(NFSD_FILE_GC, &nf->nf_flags) != want_gc)
 977			continue;
 978		if (test_bit(NFSD_FILE_HASHED, &nf->nf_flags) == 0)
 979			continue;
 980
 981		if (!nfsd_file_get(nf))
 982			continue;
 983		return nf;
 984	}
 985	return NULL;
 986}
 987
 988/**
 989 * nfsd_file_is_cached - are there any cached open files for this inode?
 990 * @inode: inode to check
 991 *
 992 * The lookup matches inodes in all net namespaces and is atomic wrt
 993 * nfsd_file_acquire().
 994 *
 995 * Return values:
 996 *   %true: filecache contains at least one file matching this inode
 997 *   %false: filecache contains no files matching this inode
 998 */
 999bool
1000nfsd_file_is_cached(struct inode *inode)
1001{
1002	struct rhlist_head *tmp, *list;
1003	struct nfsd_file *nf;
1004	bool ret = false;
1005
1006	rcu_read_lock();
1007	list = rhltable_lookup(&nfsd_file_rhltable, &inode,
1008			       nfsd_file_rhash_params);
1009	rhl_for_each_entry_rcu(nf, tmp, list, nf_rlist)
1010		if (test_bit(NFSD_FILE_GC, &nf->nf_flags)) {
1011			ret = true;
1012			break;
1013		}
1014	rcu_read_unlock();
1015
1016	trace_nfsd_file_is_cached(inode, (int)ret);
1017	return ret;
1018}
1019
1020static __be32
1021nfsd_file_do_acquire(struct svc_rqst *rqstp, struct net *net,
1022		     struct svc_cred *cred,
1023		     struct auth_domain *client,
1024		     struct svc_fh *fhp,
1025		     unsigned int may_flags, struct file *file,
1026		     struct nfsd_file **pnf, bool want_gc)
1027{
1028	unsigned char need = may_flags & NFSD_FILE_MAY_MASK;
 
1029	struct nfsd_file *new, *nf;
1030	bool stale_retry = true;
1031	bool open_retry = true;
1032	struct inode *inode;
1033	__be32 status;
1034	int ret;
1035
1036retry:
1037	if (rqstp) {
1038		status = fh_verify(rqstp, fhp, S_IFREG,
1039				   may_flags|NFSD_MAY_OWNER_OVERRIDE);
1040	} else {
1041		status = fh_verify_local(net, cred, client, fhp, S_IFREG,
1042					 may_flags|NFSD_MAY_OWNER_OVERRIDE);
1043	}
1044	if (status != nfs_ok)
1045		return status;
1046	inode = d_inode(fhp->fh_dentry);
1047
1048	rcu_read_lock();
1049	nf = nfsd_file_lookup_locked(net, current_cred(), inode, need, want_gc);
1050	rcu_read_unlock();
1051
1052	if (nf) {
1053		/*
1054		 * If the nf is on the LRU then it holds an extra reference
1055		 * that must be put if it's removed. It had better not be
1056		 * the last one however, since we should hold another.
1057		 */
1058		if (nfsd_file_lru_remove(nf))
1059			refcount_dec(&nf->nf_ref);
1060		goto wait_for_construction;
1061	}
1062
1063	new = nfsd_file_alloc(net, inode, need, want_gc);
1064	if (!new) {
1065		status = nfserr_jukebox;
1066		goto out;
1067	}
1068
1069	rcu_read_lock();
1070	spin_lock(&inode->i_lock);
1071	nf = nfsd_file_lookup_locked(net, current_cred(), inode, need, want_gc);
1072	if (unlikely(nf)) {
1073		spin_unlock(&inode->i_lock);
1074		rcu_read_unlock();
1075		nfsd_file_free(new);
1076		goto wait_for_construction;
1077	}
1078	nf = new;
1079	ret = rhltable_insert(&nfsd_file_rhltable, &nf->nf_rlist,
1080			      nfsd_file_rhash_params);
1081	spin_unlock(&inode->i_lock);
1082	rcu_read_unlock();
1083	if (likely(ret == 0))
1084		goto open_file;
1085
 
 
1086	trace_nfsd_file_insert_err(rqstp, inode, may_flags, ret);
1087	status = nfserr_jukebox;
1088	goto construction_err;
1089
1090wait_for_construction:
1091	wait_on_bit(&nf->nf_flags, NFSD_FILE_PENDING, TASK_UNINTERRUPTIBLE);
1092
1093	/* Did construction of this file fail? */
1094	if (!test_bit(NFSD_FILE_HASHED, &nf->nf_flags)) {
1095		trace_nfsd_file_cons_err(rqstp, inode, may_flags, nf);
1096		if (!open_retry) {
1097			status = nfserr_jukebox;
1098			goto construction_err;
1099		}
1100		nfsd_file_put(nf);
1101		open_retry = false;
1102		fh_put(fhp);
1103		goto retry;
1104	}
1105	this_cpu_inc(nfsd_file_cache_hits);
1106
1107	status = nfserrno(nfsd_open_break_lease(file_inode(nf->nf_file), may_flags));
1108	if (status != nfs_ok) {
1109		nfsd_file_put(nf);
1110		nf = NULL;
1111	}
1112
1113out:
1114	if (status == nfs_ok) {
1115		this_cpu_inc(nfsd_file_acquisitions);
1116		nfsd_file_check_write_error(nf);
1117		*pnf = nf;
1118	}
1119	trace_nfsd_file_acquire(rqstp, inode, may_flags, nf, status);
1120	return status;
1121
1122open_file:
1123	trace_nfsd_file_alloc(nf);
1124	nf->nf_mark = nfsd_file_mark_find_or_create(inode);
1125	if (nf->nf_mark) {
1126		if (file) {
1127			get_file(file);
1128			nf->nf_file = file;
1129			status = nfs_ok;
1130			trace_nfsd_file_opened(nf, status);
1131		} else {
1132			ret = nfsd_open_verified(fhp, may_flags, &nf->nf_file);
 
1133			if (ret == -EOPENSTALE && stale_retry) {
1134				stale_retry = false;
1135				nfsd_file_unhash(nf);
1136				clear_and_wake_up_bit(NFSD_FILE_PENDING,
1137						      &nf->nf_flags);
1138				if (refcount_dec_and_test(&nf->nf_ref))
1139					nfsd_file_free(nf);
1140				nf = NULL;
1141				fh_put(fhp);
1142				goto retry;
1143			}
1144			status = nfserrno(ret);
1145			trace_nfsd_file_open(nf, status);
1146		}
1147	} else
1148		status = nfserr_jukebox;
1149	/*
1150	 * If construction failed, or we raced with a call to unlink()
1151	 * then unhash.
1152	 */
1153	if (status != nfs_ok || inode->i_nlink == 0)
1154		nfsd_file_unhash(nf);
1155	clear_and_wake_up_bit(NFSD_FILE_PENDING, &nf->nf_flags);
1156	if (status == nfs_ok)
1157		goto out;
1158
1159construction_err:
1160	if (refcount_dec_and_test(&nf->nf_ref))
1161		nfsd_file_free(nf);
1162	nf = NULL;
1163	goto out;
1164}
1165
1166/**
1167 * nfsd_file_acquire_gc - Get a struct nfsd_file with an open file
1168 * @rqstp: the RPC transaction being executed
1169 * @fhp: the NFS filehandle of the file to be opened
1170 * @may_flags: NFSD_MAY_ settings for the file
1171 * @pnf: OUT: new or found "struct nfsd_file" object
1172 *
1173 * The nfsd_file object returned by this API is reference-counted
1174 * and garbage-collected. The object is retained for a few
1175 * seconds after the final nfsd_file_put() in case the caller
1176 * wants to re-use it.
1177 *
1178 * Return values:
1179 *   %nfs_ok - @pnf points to an nfsd_file with its reference
1180 *   count boosted.
1181 *
1182 * On error, an nfsstat value in network byte order is returned.
1183 */
1184__be32
1185nfsd_file_acquire_gc(struct svc_rqst *rqstp, struct svc_fh *fhp,
1186		     unsigned int may_flags, struct nfsd_file **pnf)
1187{
1188	return nfsd_file_do_acquire(rqstp, SVC_NET(rqstp), NULL, NULL,
1189				    fhp, may_flags, NULL, pnf, true);
1190}
1191
1192/**
1193 * nfsd_file_acquire - Get a struct nfsd_file with an open file
1194 * @rqstp: the RPC transaction being executed
1195 * @fhp: the NFS filehandle of the file to be opened
1196 * @may_flags: NFSD_MAY_ settings for the file
1197 * @pnf: OUT: new or found "struct nfsd_file" object
1198 *
1199 * The nfsd_file_object returned by this API is reference-counted
1200 * but not garbage-collected. The object is unhashed after the
1201 * final nfsd_file_put().
1202 *
1203 * Return values:
1204 *   %nfs_ok - @pnf points to an nfsd_file with its reference
1205 *   count boosted.
1206 *
1207 * On error, an nfsstat value in network byte order is returned.
1208 */
1209__be32
1210nfsd_file_acquire(struct svc_rqst *rqstp, struct svc_fh *fhp,
1211		  unsigned int may_flags, struct nfsd_file **pnf)
1212{
1213	return nfsd_file_do_acquire(rqstp, SVC_NET(rqstp), NULL, NULL,
1214				    fhp, may_flags, NULL, pnf, false);
1215}
1216
1217/**
1218 * nfsd_file_acquire_local - Get a struct nfsd_file with an open file for localio
1219 * @net: The network namespace in which to perform a lookup
1220 * @cred: the user credential with which to validate access
1221 * @client: the auth_domain for LOCALIO lookup
1222 * @fhp: the NFS filehandle of the file to be opened
1223 * @may_flags: NFSD_MAY_ settings for the file
1224 * @pnf: OUT: new or found "struct nfsd_file" object
1225 *
1226 * This file lookup interface provide access to a file given the
1227 * filehandle and credential.  No connection-based authorisation
1228 * is performed and in that way it is quite different to other
1229 * file access mediated by nfsd.  It allows a kernel module such as the NFS
1230 * client to reach across network and filesystem namespaces to access
1231 * a file.  The security implications of this should be carefully
1232 * considered before use.
1233 *
1234 * The nfsd_file object returned by this API is reference-counted
1235 * and garbage-collected. The object is retained for a few
1236 * seconds after the final nfsd_file_put() in case the caller
1237 * wants to re-use it.
1238 *
1239 * Return values:
1240 *   %nfs_ok - @pnf points to an nfsd_file with its reference
1241 *   count boosted.
1242 *
1243 * On error, an nfsstat value in network byte order is returned.
1244 */
1245__be32
1246nfsd_file_acquire_local(struct net *net, struct svc_cred *cred,
1247			struct auth_domain *client, struct svc_fh *fhp,
1248			unsigned int may_flags, struct nfsd_file **pnf)
1249{
1250	/*
1251	 * Save creds before calling nfsd_file_do_acquire() (which calls
1252	 * nfsd_setuser). Important because caller (LOCALIO) is from
1253	 * client context.
1254	 */
1255	const struct cred *save_cred = get_current_cred();
1256	__be32 beres;
1257
1258	beres = nfsd_file_do_acquire(NULL, net, cred, client,
1259				     fhp, may_flags, NULL, pnf, true);
1260	revert_creds(save_cred);
1261	return beres;
1262}
1263
1264/**
1265 * nfsd_file_acquire_opened - Get a struct nfsd_file using existing open file
1266 * @rqstp: the RPC transaction being executed
1267 * @fhp: the NFS filehandle of the file just created
1268 * @may_flags: NFSD_MAY_ settings for the file
1269 * @file: cached, already-open file (may be NULL)
1270 * @pnf: OUT: new or found "struct nfsd_file" object
1271 *
1272 * Acquire a nfsd_file object that is not GC'ed. If one doesn't already exist,
1273 * and @file is non-NULL, use it to instantiate a new nfsd_file instead of
1274 * opening a new one.
1275 *
1276 * Return values:
1277 *   %nfs_ok - @pnf points to an nfsd_file with its reference
1278 *   count boosted.
1279 *
1280 * On error, an nfsstat value in network byte order is returned.
1281 */
1282__be32
1283nfsd_file_acquire_opened(struct svc_rqst *rqstp, struct svc_fh *fhp,
1284			 unsigned int may_flags, struct file *file,
1285			 struct nfsd_file **pnf)
1286{
1287	return nfsd_file_do_acquire(rqstp, SVC_NET(rqstp), NULL, NULL,
1288				    fhp, may_flags, file, pnf, false);
1289}
1290
1291/*
1292 * Note that fields may be added, removed or reordered in the future. Programs
1293 * scraping this file for info should test the labels to ensure they're
1294 * getting the correct field.
1295 */
1296int nfsd_file_cache_stats_show(struct seq_file *m, void *v)
1297{
1298	unsigned long allocations = 0, releases = 0, evictions = 0;
1299	unsigned long hits = 0, acquisitions = 0;
1300	unsigned int i, count = 0, buckets = 0;
1301	unsigned long lru = 0, total_age = 0;
1302
1303	/* Serialize with server shutdown */
1304	mutex_lock(&nfsd_mutex);
1305	if (test_bit(NFSD_FILE_CACHE_UP, &nfsd_file_flags) == 1) {
1306		struct bucket_table *tbl;
1307		struct rhashtable *ht;
1308
1309		lru = list_lru_count(&nfsd_file_lru);
1310
1311		rcu_read_lock();
1312		ht = &nfsd_file_rhltable.ht;
1313		count = atomic_read(&ht->nelems);
1314		tbl = rht_dereference_rcu(ht->tbl, ht);
1315		buckets = tbl->size;
1316		rcu_read_unlock();
1317	}
1318	mutex_unlock(&nfsd_mutex);
1319
1320	for_each_possible_cpu(i) {
1321		hits += per_cpu(nfsd_file_cache_hits, i);
1322		acquisitions += per_cpu(nfsd_file_acquisitions, i);
1323		allocations += per_cpu(nfsd_file_allocations, i);
1324		releases += per_cpu(nfsd_file_releases, i);
1325		total_age += per_cpu(nfsd_file_total_age, i);
1326		evictions += per_cpu(nfsd_file_evictions, i);
1327	}
1328
1329	seq_printf(m, "total inodes:  %u\n", count);
1330	seq_printf(m, "hash buckets:  %u\n", buckets);
1331	seq_printf(m, "lru entries:   %lu\n", lru);
1332	seq_printf(m, "cache hits:    %lu\n", hits);
1333	seq_printf(m, "acquisitions:  %lu\n", acquisitions);
1334	seq_printf(m, "allocations:   %lu\n", allocations);
1335	seq_printf(m, "releases:      %lu\n", releases);
1336	seq_printf(m, "evictions:     %lu\n", evictions);
1337	if (releases)
1338		seq_printf(m, "mean age (ms): %ld\n", total_age / releases);
1339	else
1340		seq_printf(m, "mean age (ms): -\n");
1341	return 0;
1342}