Linux Audio

Check our new training course

Loading...
v6.2
   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	struct work_struct work;
 
  65	spinlock_t lock;
  66	struct list_head freeme;
 
  67};
  68
  69static struct workqueue_struct *nfsd_filecache_wq __read_mostly;
  70
  71static struct kmem_cache		*nfsd_file_slab;
  72static struct kmem_cache		*nfsd_file_mark_slab;
 
  73static struct list_lru			nfsd_file_lru;
  74static unsigned long			nfsd_file_flags;
  75static struct fsnotify_group		*nfsd_file_fsnotify_group;
 
  76static struct delayed_work		nfsd_filecache_laundrette;
  77static struct rhashtable		nfsd_file_rhash_tbl
  78						____cacheline_aligned_in_smp;
  79
  80enum nfsd_file_lookup_type {
  81	NFSD_FILE_KEY_INODE,
  82	NFSD_FILE_KEY_FULL,
  83};
  84
  85struct nfsd_file_lookup_key {
  86	struct inode			*inode;
  87	struct net			*net;
  88	const struct cred		*cred;
  89	unsigned char			need;
  90	bool				gc;
  91	enum nfsd_file_lookup_type	type;
  92};
  93
  94/*
  95 * The returned hash value is based solely on the address of an in-code
  96 * inode, a pointer to a slab-allocated object. The entropy in such a
  97 * pointer is concentrated in its middle bits.
  98 */
  99static u32 nfsd_file_inode_hash(const struct inode *inode, u32 seed)
 100{
 101	unsigned long ptr = (unsigned long)inode;
 102	u32 k;
 103
 104	k = ptr >> L1_CACHE_SHIFT;
 105	k &= 0x00ffffff;
 106	return jhash2(&k, 1, seed);
 107}
 108
 109/**
 110 * nfsd_file_key_hashfn - Compute the hash value of a lookup key
 111 * @data: key on which to compute the hash value
 112 * @len: rhash table's key_len parameter (unused)
 113 * @seed: rhash table's random seed of the day
 114 *
 115 * Return value:
 116 *   Computed 32-bit hash value
 117 */
 118static u32 nfsd_file_key_hashfn(const void *data, u32 len, u32 seed)
 119{
 120	const struct nfsd_file_lookup_key *key = data;
 121
 122	return nfsd_file_inode_hash(key->inode, seed);
 123}
 124
 125/**
 126 * nfsd_file_obj_hashfn - Compute the hash value of an nfsd_file
 127 * @data: object on which to compute the hash value
 128 * @len: rhash table's key_len parameter (unused)
 129 * @seed: rhash table's random seed of the day
 130 *
 131 * Return value:
 132 *   Computed 32-bit hash value
 133 */
 134static u32 nfsd_file_obj_hashfn(const void *data, u32 len, u32 seed)
 135{
 136	const struct nfsd_file *nf = data;
 137
 138	return nfsd_file_inode_hash(nf->nf_inode, seed);
 139}
 140
 141static bool
 142nfsd_match_cred(const struct cred *c1, const struct cred *c2)
 143{
 144	int i;
 145
 146	if (!uid_eq(c1->fsuid, c2->fsuid))
 147		return false;
 148	if (!gid_eq(c1->fsgid, c2->fsgid))
 149		return false;
 150	if (c1->group_info == NULL || c2->group_info == NULL)
 151		return c1->group_info == c2->group_info;
 152	if (c1->group_info->ngroups != c2->group_info->ngroups)
 153		return false;
 154	for (i = 0; i < c1->group_info->ngroups; i++) {
 155		if (!gid_eq(c1->group_info->gid[i], c2->group_info->gid[i]))
 156			return false;
 157	}
 158	return true;
 159}
 160
 161/**
 162 * nfsd_file_obj_cmpfn - Match a cache item against search criteria
 163 * @arg: search criteria
 164 * @ptr: cache item to check
 165 *
 166 * Return values:
 167 *   %0 - Item matches search criteria
 168 *   %1 - Item does not match search criteria
 169 */
 170static int nfsd_file_obj_cmpfn(struct rhashtable_compare_arg *arg,
 171			       const void *ptr)
 172{
 173	const struct nfsd_file_lookup_key *key = arg->key;
 174	const struct nfsd_file *nf = ptr;
 175
 176	switch (key->type) {
 177	case NFSD_FILE_KEY_INODE:
 178		if (nf->nf_inode != key->inode)
 179			return 1;
 180		break;
 181	case NFSD_FILE_KEY_FULL:
 182		if (nf->nf_inode != key->inode)
 183			return 1;
 184		if (nf->nf_may != key->need)
 185			return 1;
 186		if (nf->nf_net != key->net)
 187			return 1;
 188		if (!nfsd_match_cred(nf->nf_cred, key->cred))
 189			return 1;
 190		if (!!test_bit(NFSD_FILE_GC, &nf->nf_flags) != key->gc)
 191			return 1;
 192		if (test_bit(NFSD_FILE_HASHED, &nf->nf_flags) == 0)
 193			return 1;
 194		break;
 195	}
 196	return 0;
 197}
 198
 199static const struct rhashtable_params nfsd_file_rhash_params = {
 200	.key_len		= sizeof_field(struct nfsd_file, nf_inode),
 201	.key_offset		= offsetof(struct nfsd_file, nf_inode),
 202	.head_offset		= offsetof(struct nfsd_file, nf_rhash),
 203	.hashfn			= nfsd_file_key_hashfn,
 204	.obj_hashfn		= nfsd_file_obj_hashfn,
 205	.obj_cmpfn		= nfsd_file_obj_cmpfn,
 206	/* Reduce resizing churn on light workloads */
 207	.min_size		= 512,		/* buckets */
 208	.automatic_shrinking	= true,
 209};
 210
 211static void
 212nfsd_file_schedule_laundrette(void)
 213{
 214	if (test_bit(NFSD_FILE_CACHE_UP, &nfsd_file_flags))
 215		queue_delayed_work(system_wq, &nfsd_filecache_laundrette,
 216				   NFSD_LAUNDRETTE_DELAY);
 
 
 
 
 217}
 218
 219static void
 220nfsd_file_slab_free(struct rcu_head *rcu)
 221{
 222	struct nfsd_file *nf = container_of(rcu, struct nfsd_file, nf_rcu);
 223
 224	put_cred(nf->nf_cred);
 225	kmem_cache_free(nfsd_file_slab, nf);
 226}
 227
 228static void
 229nfsd_file_mark_free(struct fsnotify_mark *mark)
 230{
 231	struct nfsd_file_mark *nfm = container_of(mark, struct nfsd_file_mark,
 232						  nfm_mark);
 233
 234	kmem_cache_free(nfsd_file_mark_slab, nfm);
 235}
 236
 237static struct nfsd_file_mark *
 238nfsd_file_mark_get(struct nfsd_file_mark *nfm)
 239{
 240	if (!refcount_inc_not_zero(&nfm->nfm_ref))
 241		return NULL;
 242	return nfm;
 243}
 244
 245static void
 246nfsd_file_mark_put(struct nfsd_file_mark *nfm)
 247{
 248	if (refcount_dec_and_test(&nfm->nfm_ref)) {
 249		fsnotify_destroy_mark(&nfm->nfm_mark, nfsd_file_fsnotify_group);
 250		fsnotify_put_mark(&nfm->nfm_mark);
 251	}
 252}
 253
 254static struct nfsd_file_mark *
 255nfsd_file_mark_find_or_create(struct nfsd_file *nf, struct inode *inode)
 256{
 257	int			err;
 258	struct fsnotify_mark	*mark;
 259	struct nfsd_file_mark	*nfm = NULL, *new;
 
 260
 261	do {
 262		fsnotify_group_lock(nfsd_file_fsnotify_group);
 263		mark = fsnotify_find_mark(&inode->i_fsnotify_marks,
 264					  nfsd_file_fsnotify_group);
 265		if (mark) {
 266			nfm = nfsd_file_mark_get(container_of(mark,
 267						 struct nfsd_file_mark,
 268						 nfm_mark));
 269			fsnotify_group_unlock(nfsd_file_fsnotify_group);
 270			if (nfm) {
 271				fsnotify_put_mark(mark);
 272				break;
 273			}
 274			/* Avoid soft lockup race with nfsd_file_mark_put() */
 275			fsnotify_destroy_mark(mark, nfsd_file_fsnotify_group);
 276			fsnotify_put_mark(mark);
 277		} else {
 278			fsnotify_group_unlock(nfsd_file_fsnotify_group);
 279		}
 280
 281		/* allocate a new nfm */
 282		new = kmem_cache_alloc(nfsd_file_mark_slab, GFP_KERNEL);
 283		if (!new)
 284			return NULL;
 285		fsnotify_init_mark(&new->nfm_mark, nfsd_file_fsnotify_group);
 286		new->nfm_mark.mask = FS_ATTRIB|FS_DELETE_SELF;
 287		refcount_set(&new->nfm_ref, 1);
 288
 289		err = fsnotify_add_inode_mark(&new->nfm_mark, inode, 0);
 290
 291		/*
 292		 * If the add was successful, then return the object.
 293		 * Otherwise, we need to put the reference we hold on the
 294		 * nfm_mark. The fsnotify code will take a reference and put
 295		 * it on failure, so we can't just free it directly. It's also
 296		 * not safe to call fsnotify_destroy_mark on it as the
 297		 * mark->group will be NULL. Thus, we can't let the nfm_ref
 298		 * counter drive the destruction at this point.
 299		 */
 300		if (likely(!err))
 301			nfm = new;
 302		else
 303			fsnotify_put_mark(&new->nfm_mark);
 304	} while (unlikely(err == -EEXIST));
 305
 306	return nfm;
 307}
 308
 309static struct nfsd_file *
 310nfsd_file_alloc(struct nfsd_file_lookup_key *key, unsigned int may)
 
 311{
 312	struct nfsd_file *nf;
 313
 314	nf = kmem_cache_alloc(nfsd_file_slab, GFP_KERNEL);
 315	if (nf) {
 
 316		INIT_LIST_HEAD(&nf->nf_lru);
 317		nf->nf_birthtime = ktime_get();
 318		nf->nf_file = NULL;
 319		nf->nf_cred = get_current_cred();
 320		nf->nf_net = key->net;
 321		nf->nf_flags = 0;
 322		__set_bit(NFSD_FILE_HASHED, &nf->nf_flags);
 323		__set_bit(NFSD_FILE_PENDING, &nf->nf_flags);
 324		if (key->gc)
 325			__set_bit(NFSD_FILE_GC, &nf->nf_flags);
 326		nf->nf_inode = key->inode;
 327		refcount_set(&nf->nf_ref, 1);
 328		nf->nf_may = key->need;
 
 
 
 
 
 
 329		nf->nf_mark = NULL;
 
 
 330	}
 331	return nf;
 332}
 333
 334static void
 335nfsd_file_fsync(struct nfsd_file *nf)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 336{
 337	struct file *file = nf->nf_file;
 338	int ret;
 339
 340	if (!file || !(file->f_mode & FMODE_WRITE))
 341		return;
 342	ret = vfs_fsync(file, 1);
 343	trace_nfsd_file_fsync(nf, ret);
 344	if (ret)
 345		nfsd_reset_write_verifier(net_generic(nf->nf_net, nfsd_net_id));
 346}
 347
 348static int
 349nfsd_file_check_write_error(struct nfsd_file *nf)
 350{
 351	struct file *file = nf->nf_file;
 352
 353	if (!file || !(file->f_mode & FMODE_WRITE))
 354		return 0;
 355	return filemap_check_wb_err(file->f_mapping, READ_ONCE(file->f_wb_err));
 356}
 357
 358static void
 359nfsd_file_hash_remove(struct nfsd_file *nf)
 360{
 
 
 361	trace_nfsd_file_unhash(nf);
 362
 363	if (nfsd_file_check_write_error(nf))
 364		nfsd_reset_write_verifier(net_generic(nf->nf_net, nfsd_net_id));
 365	rhashtable_remove_fast(&nfsd_file_rhash_tbl, &nf->nf_rhash,
 366			       nfsd_file_rhash_params);
 
 367}
 368
 369static bool
 370nfsd_file_unhash(struct nfsd_file *nf)
 371{
 372	if (test_and_clear_bit(NFSD_FILE_HASHED, &nf->nf_flags)) {
 373		nfsd_file_hash_remove(nf);
 
 
 374		return true;
 375	}
 376	return false;
 377}
 378
 379static void
 380nfsd_file_free(struct nfsd_file *nf)
 
 
 
 381{
 382	s64 age = ktime_to_ms(ktime_sub(ktime_get(), nf->nf_birthtime));
 383
 384	trace_nfsd_file_free(nf);
 385
 386	this_cpu_inc(nfsd_file_releases);
 387	this_cpu_add(nfsd_file_total_age, age);
 388
 389	nfsd_file_unhash(nf);
 390
 391	/*
 392	 * We call fsync here in order to catch writeback errors. It's not
 393	 * strictly required by the protocol, but an nfsd_file could get
 394	 * evicted from the cache before a COMMIT comes in. If another
 395	 * task were to open that file in the interim and scrape the error,
 396	 * then the client may never see it. By calling fsync here, we ensure
 397	 * that writeback happens before the entry is freed, and that any
 398	 * errors reported result in the write verifier changing.
 399	 */
 400	nfsd_file_fsync(nf);
 401
 402	if (nf->nf_mark)
 403		nfsd_file_mark_put(nf->nf_mark);
 404	if (nf->nf_file) {
 405		get_file(nf->nf_file);
 406		filp_close(nf->nf_file, NULL);
 407		fput(nf->nf_file);
 408	}
 409
 410	/*
 411	 * If this item is still linked via nf_lru, that's a bug.
 412	 * WARN and leak it to preserve system stability.
 413	 */
 414	if (WARN_ON_ONCE(!list_empty(&nf->nf_lru)))
 415		return;
 416
 417	call_rcu(&nf->nf_rcu, nfsd_file_slab_free);
 
 418}
 419
 420static bool
 421nfsd_file_check_writeback(struct nfsd_file *nf)
 422{
 423	struct file *file = nf->nf_file;
 424	struct address_space *mapping;
 425
 426	if (!file || !(file->f_mode & FMODE_WRITE))
 427		return false;
 428	mapping = file->f_mapping;
 429	return mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) ||
 430		mapping_tagged(mapping, PAGECACHE_TAG_WRITEBACK);
 431}
 432
 433static bool nfsd_file_lru_add(struct nfsd_file *nf)
 
 434{
 
 
 435	set_bit(NFSD_FILE_REFERENCED, &nf->nf_flags);
 436	if (list_lru_add(&nfsd_file_lru, &nf->nf_lru)) {
 437		trace_nfsd_file_lru_add(nf);
 438		return true;
 439	}
 440	return false;
 441}
 442
 443static bool nfsd_file_lru_remove(struct nfsd_file *nf)
 444{
 445	if (list_lru_del(&nfsd_file_lru, &nf->nf_lru)) {
 446		trace_nfsd_file_lru_del(nf);
 447		return true;
 448	}
 449	return false;
 450}
 451
 452struct nfsd_file *
 453nfsd_file_get(struct nfsd_file *nf)
 454{
 455	if (likely(refcount_inc_not_zero(&nf->nf_ref)))
 456		return nf;
 457	return NULL;
 458}
 459
 460/**
 461 * nfsd_file_put - put the reference to a nfsd_file
 462 * @nf: nfsd_file of which to put the reference
 463 *
 464 * Put a reference to a nfsd_file. In the non-GC case, we just put the
 465 * reference immediately. In the GC case, if the reference would be
 466 * the last one, the put it on the LRU instead to be cleaned up later.
 467 */
 468void
 469nfsd_file_put(struct nfsd_file *nf)
 470{
 471	might_sleep();
 472	trace_nfsd_file_put(nf);
 473
 474	if (test_bit(NFSD_FILE_GC, &nf->nf_flags) &&
 475	    test_bit(NFSD_FILE_HASHED, &nf->nf_flags)) {
 476		/*
 477		 * If this is the last reference (nf_ref == 1), then try to
 478		 * transfer it to the LRU.
 479		 */
 480		if (refcount_dec_not_one(&nf->nf_ref))
 481			return;
 482
 483		/* Try to add it to the LRU.  If that fails, decrement. */
 484		if (nfsd_file_lru_add(nf)) {
 485			/* If it's still hashed, we're done */
 486			if (test_bit(NFSD_FILE_HASHED, &nf->nf_flags)) {
 487				nfsd_file_schedule_laundrette();
 488				return;
 489			}
 490
 491			/*
 492			 * We're racing with unhashing, so try to remove it from
 493			 * the LRU. If removal fails, then someone else already
 494			 * has our reference.
 495			 */
 496			if (!nfsd_file_lru_remove(nf))
 497				return;
 498		}
 499	}
 500	if (refcount_dec_and_test(&nf->nf_ref))
 501		nfsd_file_free(nf);
 502}
 503
 504static void
 505nfsd_file_dispose_list(struct list_head *dispose)
 506{
 
 507	struct nfsd_file *nf;
 508
 509	while (!list_empty(dispose)) {
 510		nf = list_first_entry(dispose, struct nfsd_file, nf_lru);
 511		list_del_init(&nf->nf_lru);
 512		nfsd_file_free(nf);
 
 
 
 513	}
 
 
 514}
 515
 516static void
 517nfsd_file_list_remove_disposal(struct list_head *dst,
 518		struct nfsd_fcache_disposal *l)
 519{
 520	spin_lock(&l->lock);
 521	list_splice_init(&l->freeme, dst);
 522	spin_unlock(&l->lock);
 523}
 524
 525static void
 526nfsd_file_list_add_disposal(struct list_head *files, struct net *net)
 527{
 528	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
 529	struct nfsd_fcache_disposal *l = nn->fcache_disposal;
 530
 531	spin_lock(&l->lock);
 532	list_splice_tail_init(files, &l->freeme);
 533	spin_unlock(&l->lock);
 534	queue_work(nfsd_filecache_wq, &l->work);
 
 
 
 
 
 
 
 535}
 536
 537static void
 538nfsd_file_list_add_pernet(struct list_head *dst, struct list_head *src,
 539		struct net *net)
 540{
 541	struct nfsd_file *nf, *tmp;
 542
 543	list_for_each_entry_safe(nf, tmp, src, nf_lru) {
 544		if (nf->nf_net == net)
 545			list_move_tail(&nf->nf_lru, dst);
 546	}
 547}
 548
 549static void
 550nfsd_file_dispose_list_delayed(struct list_head *dispose)
 551{
 552	LIST_HEAD(list);
 553	struct nfsd_file *nf;
 554
 555	while(!list_empty(dispose)) {
 556		nf = list_first_entry(dispose, struct nfsd_file, nf_lru);
 557		nfsd_file_list_add_pernet(&list, dispose, nf->nf_net);
 558		nfsd_file_list_add_disposal(&list, nf->nf_net);
 559	}
 560}
 561
 562/**
 563 * nfsd_file_lru_cb - Examine an entry on the LRU list
 564 * @item: LRU entry to examine
 565 * @lru: controlling LRU
 566 * @lock: LRU list lock (unused)
 567 * @arg: dispose list
 568 *
 569 * Return values:
 570 *   %LRU_REMOVED: @item was removed from the LRU
 571 *   %LRU_ROTATE: @item is to be moved to the LRU tail
 572 *   %LRU_SKIP: @item cannot be evicted
 573 */
 574static enum lru_status
 575nfsd_file_lru_cb(struct list_head *item, struct list_lru_one *lru,
 576		 spinlock_t *lock, void *arg)
 577	__releases(lock)
 578	__acquires(lock)
 579{
 580	struct list_head *head = arg;
 581	struct nfsd_file *nf = list_entry(item, struct nfsd_file, nf_lru);
 582
 583	/* We should only be dealing with GC entries here */
 584	WARN_ON_ONCE(!test_bit(NFSD_FILE_GC, &nf->nf_flags));
 
 
 
 
 
 
 
 
 
 
 585
 586	/*
 587	 * Don't throw out files that are still undergoing I/O or
 588	 * that have uncleared errors pending.
 589	 */
 590	if (nfsd_file_check_writeback(nf)) {
 591		trace_nfsd_file_gc_writeback(nf);
 592		return LRU_SKIP;
 593	}
 594
 595	/* If it was recently added to the list, skip it */
 596	if (test_and_clear_bit(NFSD_FILE_REFERENCED, &nf->nf_flags)) {
 597		trace_nfsd_file_gc_referenced(nf);
 598		return LRU_ROTATE;
 599	}
 600
 601	/*
 602	 * Put the reference held on behalf of the LRU. If it wasn't the last
 603	 * one, then just remove it from the LRU and ignore it.
 604	 */
 605	if (!refcount_dec_and_test(&nf->nf_ref)) {
 606		trace_nfsd_file_gc_in_use(nf);
 607		list_lru_isolate(lru, &nf->nf_lru);
 608		return LRU_REMOVED;
 609	}
 610
 611	/* Refcount went to zero. Unhash it and queue it to the dispose list */
 612	nfsd_file_unhash(nf);
 613	list_lru_isolate_move(lru, &nf->nf_lru, head);
 614	this_cpu_inc(nfsd_file_evictions);
 615	trace_nfsd_file_gc_disposed(nf);
 616	return LRU_REMOVED;
 
 
 617}
 618
 619static void
 620nfsd_file_gc(void)
 621{
 622	LIST_HEAD(dispose);
 
 623	unsigned long ret;
 624
 625	ret = list_lru_walk(&nfsd_file_lru, nfsd_file_lru_cb,
 626			    &dispose, list_lru_count(&nfsd_file_lru));
 627	trace_nfsd_file_gc_removed(ret, list_lru_count(&nfsd_file_lru));
 628	nfsd_file_dispose_list_delayed(&dispose);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 629}
 630
 631static void
 632nfsd_file_gc_worker(struct work_struct *work)
 633{
 634	nfsd_file_gc();
 635	if (list_lru_count(&nfsd_file_lru))
 636		nfsd_file_schedule_laundrette();
 637}
 638
 639static unsigned long
 640nfsd_file_lru_count(struct shrinker *s, struct shrink_control *sc)
 641{
 642	return list_lru_count(&nfsd_file_lru);
 643}
 644
 645static unsigned long
 646nfsd_file_lru_scan(struct shrinker *s, struct shrink_control *sc)
 647{
 648	LIST_HEAD(dispose);
 649	unsigned long ret;
 650
 651	ret = list_lru_shrink_walk(&nfsd_file_lru, sc,
 652				   nfsd_file_lru_cb, &dispose);
 653	trace_nfsd_file_shrinker_removed(ret, list_lru_count(&nfsd_file_lru));
 654	nfsd_file_dispose_list_delayed(&dispose);
 655	return ret;
 656}
 657
 658static struct shrinker	nfsd_file_shrinker = {
 659	.scan_objects = nfsd_file_lru_scan,
 660	.count_objects = nfsd_file_lru_count,
 661	.seeks = 1,
 662};
 663
 664/**
 665 * nfsd_file_cond_queue - conditionally unhash and queue a nfsd_file
 666 * @nf: nfsd_file to attempt to queue
 667 * @dispose: private list to queue successfully-put objects
 668 *
 669 * Unhash an nfsd_file, try to get a reference to it, and then put that
 670 * reference. If it's the last reference, queue it to the dispose list.
 671 */
 672static void
 673nfsd_file_cond_queue(struct nfsd_file *nf, struct list_head *dispose)
 674	__must_hold(RCU)
 675{
 676	int decrement = 1;
 677
 678	/* If we raced with someone else unhashing, ignore it */
 679	if (!nfsd_file_unhash(nf))
 680		return;
 681
 682	/* If we can't get a reference, ignore it */
 683	if (!nfsd_file_get(nf))
 684		return;
 685
 686	/* Extra decrement if we remove from the LRU */
 687	if (nfsd_file_lru_remove(nf))
 688		++decrement;
 689
 690	/* If refcount goes to 0, then put on the dispose list */
 691	if (refcount_sub_and_test(decrement, &nf->nf_ref)) {
 692		list_add(&nf->nf_lru, dispose);
 693		trace_nfsd_file_closing(nf);
 694	}
 
 695}
 696
 697/**
 698 * nfsd_file_queue_for_close: try to close out any open nfsd_files for an inode
 699 * @inode:   inode on which to close out nfsd_files
 700 * @dispose: list on which to gather nfsd_files to close out
 701 *
 702 * An nfsd_file represents a struct file being held open on behalf of nfsd. An
 703 * open file however can block other activity (such as leases), or cause
 704 * undesirable behavior (e.g. spurious silly-renames when reexporting NFS).
 705 *
 706 * This function is intended to find open nfsd_files when this sort of
 707 * conflicting access occurs and then attempt to close those files out.
 708 *
 709 * Populates the dispose list with entries that have already had their
 710 * refcounts go to zero. The actual free of an nfsd_file can be expensive,
 711 * so we leave it up to the caller whether it wants to wait or not.
 712 */
 713static void
 714nfsd_file_queue_for_close(struct inode *inode, struct list_head *dispose)
 715{
 716	struct nfsd_file_lookup_key key = {
 717		.type	= NFSD_FILE_KEY_INODE,
 718		.inode	= inode,
 719	};
 720	struct nfsd_file *nf;
 721
 722	rcu_read_lock();
 723	do {
 724		nf = rhashtable_lookup(&nfsd_file_rhash_tbl, &key,
 725				       nfsd_file_rhash_params);
 726		if (!nf)
 727			break;
 728		nfsd_file_cond_queue(nf, dispose);
 729	} while (1);
 730	rcu_read_unlock();
 731}
 732
 733/**
 734 * nfsd_file_close_inode - attempt a delayed close of a nfsd_file
 735 * @inode: inode of the file to attempt to remove
 736 *
 737 * Close out any open nfsd_files that can be reaped for @inode. The
 738 * actual freeing is deferred to the dispose_list_delayed infrastructure.
 739 *
 740 * This is used by the fsnotify callbacks and setlease notifier.
 741 */
 742static void
 743nfsd_file_close_inode(struct inode *inode)
 744{
 
 
 745	LIST_HEAD(dispose);
 746
 747	nfsd_file_queue_for_close(inode, &dispose);
 748	nfsd_file_dispose_list_delayed(&dispose);
 
 749}
 750
 751/**
 752 * nfsd_file_close_inode_sync - attempt to forcibly close a nfsd_file
 753 * @inode: inode of the file to attempt to remove
 754 *
 755 * Close out any open nfsd_files that can be reaped for @inode. The
 756 * nfsd_files are closed out synchronously.
 757 *
 758 * This is called from nfsd_rename and nfsd_unlink to avoid silly-renames
 759 * when reexporting NFS.
 760 */
 761void
 762nfsd_file_close_inode_sync(struct inode *inode)
 763{
 764	struct nfsd_file *nf;
 
 765	LIST_HEAD(dispose);
 766
 767	trace_nfsd_file_close(inode);
 768
 769	nfsd_file_queue_for_close(inode, &dispose);
 770	while (!list_empty(&dispose)) {
 771		nf = list_first_entry(&dispose, struct nfsd_file, nf_lru);
 772		list_del_init(&nf->nf_lru);
 773		nfsd_file_free(nf);
 774	}
 775	flush_delayed_fput();
 776}
 777
 778/**
 779 * nfsd_file_delayed_close - close unused nfsd_files
 780 * @work: dummy
 781 *
 782 * Walk the LRU list and destroy any entries that have not been used since
 783 * the last scan.
 
 
 784 */
 785static void
 786nfsd_file_delayed_close(struct work_struct *work)
 787{
 788	LIST_HEAD(head);
 789	struct nfsd_fcache_disposal *l = container_of(work,
 790			struct nfsd_fcache_disposal, work);
 791
 792	nfsd_file_list_remove_disposal(&head, l);
 793	nfsd_file_dispose_list(&head);
 794}
 795
 796static int
 797nfsd_file_lease_notifier_call(struct notifier_block *nb, unsigned long arg,
 798			    void *data)
 799{
 800	struct file_lock *fl = data;
 801
 802	/* Only close files for F_SETLEASE leases */
 803	if (fl->fl_flags & FL_LEASE)
 804		nfsd_file_close_inode(file_inode(fl->fl_file));
 805	return 0;
 806}
 807
 808static struct notifier_block nfsd_file_lease_notifier = {
 809	.notifier_call = nfsd_file_lease_notifier_call,
 810};
 811
 812static int
 813nfsd_file_fsnotify_handle_event(struct fsnotify_mark *mark, u32 mask,
 814				struct inode *inode, struct inode *dir,
 815				const struct qstr *name, u32 cookie)
 816{
 817	if (WARN_ON_ONCE(!inode))
 818		return 0;
 819
 820	trace_nfsd_file_fsnotify_handle_event(inode, mask);
 821
 822	/* Should be no marks on non-regular files */
 823	if (!S_ISREG(inode->i_mode)) {
 824		WARN_ON_ONCE(1);
 825		return 0;
 826	}
 827
 828	/* don't close files if this was not the last link */
 829	if (mask & FS_ATTRIB) {
 830		if (inode->i_nlink)
 831			return 0;
 832	}
 833
 834	nfsd_file_close_inode(inode);
 835	return 0;
 836}
 837
 838
 839static const struct fsnotify_ops nfsd_file_fsnotify_ops = {
 840	.handle_inode_event = nfsd_file_fsnotify_handle_event,
 841	.free_mark = nfsd_file_mark_free,
 842};
 843
 844int
 845nfsd_file_cache_init(void)
 846{
 847	int ret;
 
 848
 849	lockdep_assert_held(&nfsd_mutex);
 850	if (test_and_set_bit(NFSD_FILE_CACHE_UP, &nfsd_file_flags) == 1)
 851		return 0;
 852
 853	ret = rhashtable_init(&nfsd_file_rhash_tbl, &nfsd_file_rhash_params);
 854	if (ret)
 855		return ret;
 856
 857	ret = -ENOMEM;
 858	nfsd_filecache_wq = alloc_workqueue("nfsd_filecache", 0, 0);
 859	if (!nfsd_filecache_wq)
 860		goto out;
 861
 
 
 
 
 
 
 
 862	nfsd_file_slab = kmem_cache_create("nfsd_file",
 863				sizeof(struct nfsd_file), 0, 0, NULL);
 864	if (!nfsd_file_slab) {
 865		pr_err("nfsd: unable to create nfsd_file_slab\n");
 866		goto out_err;
 867	}
 868
 869	nfsd_file_mark_slab = kmem_cache_create("nfsd_file_mark",
 870					sizeof(struct nfsd_file_mark), 0, 0, NULL);
 871	if (!nfsd_file_mark_slab) {
 872		pr_err("nfsd: unable to create nfsd_file_mark_slab\n");
 873		goto out_err;
 874	}
 875
 876
 877	ret = list_lru_init(&nfsd_file_lru);
 878	if (ret) {
 879		pr_err("nfsd: failed to init nfsd_file_lru: %d\n", ret);
 880		goto out_err;
 881	}
 882
 883	ret = register_shrinker(&nfsd_file_shrinker, "nfsd-filecache");
 884	if (ret) {
 885		pr_err("nfsd: failed to register nfsd_file_shrinker: %d\n", ret);
 886		goto out_lru;
 887	}
 888
 889	ret = lease_register_notifier(&nfsd_file_lease_notifier);
 890	if (ret) {
 891		pr_err("nfsd: unable to register lease notifier: %d\n", ret);
 892		goto out_shrinker;
 893	}
 894
 895	nfsd_file_fsnotify_group = fsnotify_alloc_group(&nfsd_file_fsnotify_ops,
 896							FSNOTIFY_GROUP_NOFS);
 897	if (IS_ERR(nfsd_file_fsnotify_group)) {
 898		pr_err("nfsd: unable to create fsnotify group: %ld\n",
 899			PTR_ERR(nfsd_file_fsnotify_group));
 900		ret = PTR_ERR(nfsd_file_fsnotify_group);
 901		nfsd_file_fsnotify_group = NULL;
 902		goto out_notifier;
 903	}
 904
 
 
 
 
 
 905	INIT_DELAYED_WORK(&nfsd_filecache_laundrette, nfsd_file_gc_worker);
 906out:
 907	return ret;
 908out_notifier:
 909	lease_unregister_notifier(&nfsd_file_lease_notifier);
 910out_shrinker:
 911	unregister_shrinker(&nfsd_file_shrinker);
 912out_lru:
 913	list_lru_destroy(&nfsd_file_lru);
 914out_err:
 915	kmem_cache_destroy(nfsd_file_slab);
 916	nfsd_file_slab = NULL;
 917	kmem_cache_destroy(nfsd_file_mark_slab);
 918	nfsd_file_mark_slab = NULL;
 
 
 919	destroy_workqueue(nfsd_filecache_wq);
 920	nfsd_filecache_wq = NULL;
 921	rhashtable_destroy(&nfsd_file_rhash_tbl);
 922	goto out;
 923}
 924
 925/**
 926 * __nfsd_file_cache_purge: clean out the cache for shutdown
 927 * @net: net-namespace to shut down the cache (may be NULL)
 928 *
 929 * Walk the nfsd_file cache and close out any that match @net. If @net is NULL,
 930 * then close out everything. Called when an nfsd instance is being shut down.
 931 */
 932static void
 933__nfsd_file_cache_purge(struct net *net)
 934{
 935	struct rhashtable_iter iter;
 936	struct nfsd_file *nf;
 
 937	LIST_HEAD(dispose);
 
 938
 939	rhashtable_walk_enter(&nfsd_file_rhash_tbl, &iter);
 940	do {
 941		rhashtable_walk_start(&iter);
 942
 943		nf = rhashtable_walk_next(&iter);
 944		while (!IS_ERR_OR_NULL(nf)) {
 945			if (!net || nf->nf_net == net)
 946				nfsd_file_cond_queue(nf, &dispose);
 947			nf = rhashtable_walk_next(&iter);
 948		}
 949
 950		rhashtable_walk_stop(&iter);
 951	} while (nf == ERR_PTR(-EAGAIN));
 952	rhashtable_walk_exit(&iter);
 
 
 953
 954	nfsd_file_dispose_list(&dispose);
 
 
 
 
 
 
 
 
 955}
 956
 957static struct nfsd_fcache_disposal *
 958nfsd_alloc_fcache_disposal(void)
 959{
 960	struct nfsd_fcache_disposal *l;
 961
 962	l = kmalloc(sizeof(*l), GFP_KERNEL);
 963	if (!l)
 964		return NULL;
 965	INIT_WORK(&l->work, nfsd_file_delayed_close);
 
 966	spin_lock_init(&l->lock);
 967	INIT_LIST_HEAD(&l->freeme);
 968	return l;
 969}
 970
 971static void
 972nfsd_free_fcache_disposal(struct nfsd_fcache_disposal *l)
 973{
 
 974	cancel_work_sync(&l->work);
 975	nfsd_file_dispose_list(&l->freeme);
 976	kfree(l);
 977}
 978
 979static void
 980nfsd_free_fcache_disposal_net(struct net *net)
 981{
 982	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
 983	struct nfsd_fcache_disposal *l = nn->fcache_disposal;
 
 
 984
 985	nfsd_free_fcache_disposal(l);
 
 
 
 
 
 986}
 987
 988int
 989nfsd_file_cache_start_net(struct net *net)
 990{
 991	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
 992
 993	nn->fcache_disposal = nfsd_alloc_fcache_disposal();
 994	return nn->fcache_disposal ? 0 : -ENOMEM;
 
 
 
 995}
 996
 997/**
 998 * nfsd_file_cache_purge - Remove all cache items associated with @net
 999 * @net: target net namespace
1000 *
1001 */
1002void
1003nfsd_file_cache_purge(struct net *net)
1004{
1005	lockdep_assert_held(&nfsd_mutex);
1006	if (test_bit(NFSD_FILE_CACHE_UP, &nfsd_file_flags) == 1)
1007		__nfsd_file_cache_purge(net);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1008}
1009
1010void
1011nfsd_file_cache_shutdown_net(struct net *net)
1012{
1013	nfsd_file_cache_purge(net);
1014	nfsd_free_fcache_disposal_net(net);
1015}
1016
1017void
1018nfsd_file_cache_shutdown(void)
1019{
1020	int i;
1021
1022	lockdep_assert_held(&nfsd_mutex);
1023	if (test_and_clear_bit(NFSD_FILE_CACHE_UP, &nfsd_file_flags) == 0)
1024		return;
1025
1026	lease_unregister_notifier(&nfsd_file_lease_notifier);
1027	unregister_shrinker(&nfsd_file_shrinker);
1028	/*
1029	 * make sure all callers of nfsd_file_lru_cb are done before
1030	 * calling nfsd_file_cache_purge
1031	 */
1032	cancel_delayed_work_sync(&nfsd_filecache_laundrette);
1033	__nfsd_file_cache_purge(NULL);
1034	list_lru_destroy(&nfsd_file_lru);
1035	rcu_barrier();
1036	fsnotify_put_group(nfsd_file_fsnotify_group);
1037	nfsd_file_fsnotify_group = NULL;
1038	kmem_cache_destroy(nfsd_file_slab);
1039	nfsd_file_slab = NULL;
1040	fsnotify_wait_marks_destroyed();
1041	kmem_cache_destroy(nfsd_file_mark_slab);
1042	nfsd_file_mark_slab = NULL;
 
 
1043	destroy_workqueue(nfsd_filecache_wq);
1044	nfsd_filecache_wq = NULL;
1045	rhashtable_destroy(&nfsd_file_rhash_tbl);
1046
1047	for_each_possible_cpu(i) {
1048		per_cpu(nfsd_file_cache_hits, i) = 0;
1049		per_cpu(nfsd_file_acquisitions, i) = 0;
1050		per_cpu(nfsd_file_releases, i) = 0;
1051		per_cpu(nfsd_file_total_age, i) = 0;
1052		per_cpu(nfsd_file_evictions, i) = 0;
 
 
 
 
 
 
 
 
 
 
1053	}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1054}
1055
1056/**
1057 * nfsd_file_is_cached - are there any cached open files for this inode?
1058 * @inode: inode to check
1059 *
1060 * The lookup matches inodes in all net namespaces and is atomic wrt
1061 * nfsd_file_acquire().
1062 *
1063 * Return values:
1064 *   %true: filecache contains at least one file matching this inode
1065 *   %false: filecache contains no files matching this inode
1066 */
1067bool
1068nfsd_file_is_cached(struct inode *inode)
1069{
1070	struct nfsd_file_lookup_key key = {
1071		.type	= NFSD_FILE_KEY_INODE,
1072		.inode	= inode,
1073	};
1074	bool ret = false;
1075
1076	if (rhashtable_lookup_fast(&nfsd_file_rhash_tbl, &key,
1077				   nfsd_file_rhash_params) != NULL)
1078		ret = true;
1079	trace_nfsd_file_is_cached(inode, (int)ret);
 
 
 
 
 
 
1080	return ret;
1081}
1082
1083static __be32
1084nfsd_file_do_acquire(struct svc_rqst *rqstp, struct svc_fh *fhp,
1085		     unsigned int may_flags, struct file *file,
1086		     struct nfsd_file **pnf, bool want_gc)
1087{
1088	struct nfsd_file_lookup_key key = {
1089		.type	= NFSD_FILE_KEY_FULL,
1090		.need	= may_flags & NFSD_FILE_MAY_MASK,
1091		.net	= SVC_NET(rqstp),
1092		.gc	= want_gc,
1093	};
1094	bool open_retry = true;
1095	struct nfsd_file *nf;
1096	__be32 status;
1097	int ret;
1098
 
1099	status = fh_verify(rqstp, fhp, S_IFREG,
1100				may_flags|NFSD_MAY_OWNER_OVERRIDE);
1101	if (status != nfs_ok)
1102		return status;
1103	key.inode = d_inode(fhp->fh_dentry);
1104	key.cred = get_current_cred();
1105
 
 
1106retry:
1107	rcu_read_lock();
1108	nf = rhashtable_lookup(&nfsd_file_rhash_tbl, &key,
1109			       nfsd_file_rhash_params);
1110	if (nf)
1111		nf = nfsd_file_get(nf);
1112	rcu_read_unlock();
1113
1114	if (nf) {
1115		if (nfsd_file_lru_remove(nf))
1116			WARN_ON_ONCE(refcount_dec_and_test(&nf->nf_ref));
1117		goto wait_for_construction;
1118	}
1119
1120	nf = nfsd_file_alloc(&key, may_flags);
1121	if (!nf) {
1122		status = nfserr_jukebox;
1123		goto out_status;
 
1124	}
1125
1126	ret = rhashtable_lookup_insert_key(&nfsd_file_rhash_tbl,
1127					   &key, &nf->nf_rhash,
1128					   nfsd_file_rhash_params);
1129	if (likely(ret == 0))
1130		goto open_file;
1131
1132	nfsd_file_slab_free(&nf->nf_rcu);
1133	nf = NULL;
1134	if (ret == -EEXIST)
1135		goto retry;
1136	trace_nfsd_file_insert_err(rqstp, key.inode, may_flags, ret);
1137	status = nfserr_jukebox;
1138	goto out_status;
1139
1140wait_for_construction:
1141	wait_on_bit(&nf->nf_flags, NFSD_FILE_PENDING, TASK_UNINTERRUPTIBLE);
1142
1143	/* Did construction of this file fail? */
1144	if (!test_bit(NFSD_FILE_HASHED, &nf->nf_flags)) {
1145		trace_nfsd_file_cons_err(rqstp, key.inode, may_flags, nf);
1146		if (!open_retry) {
1147			status = nfserr_jukebox;
1148			goto out;
1149		}
1150		open_retry = false;
1151		if (refcount_dec_and_test(&nf->nf_ref))
1152			nfsd_file_free(nf);
1153		goto retry;
1154	}
1155
1156	this_cpu_inc(nfsd_file_cache_hits);
1157
1158	status = nfserrno(nfsd_open_break_lease(file_inode(nf->nf_file), may_flags));
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1159out:
1160	if (status == nfs_ok) {
1161		this_cpu_inc(nfsd_file_acquisitions);
1162		*pnf = nf;
1163	} else {
1164		if (refcount_dec_and_test(&nf->nf_ref))
1165			nfsd_file_free(nf);
1166		nf = NULL;
1167	}
1168
1169out_status:
1170	put_cred(key.cred);
1171	trace_nfsd_file_acquire(rqstp, key.inode, may_flags, nf, status);
1172	return status;
1173
1174open_file:
1175	trace_nfsd_file_alloc(nf);
1176	nf->nf_mark = nfsd_file_mark_find_or_create(nf, key.inode);
1177	if (nf->nf_mark) {
1178		if (file) {
1179			get_file(file);
1180			nf->nf_file = file;
1181			status = nfs_ok;
1182			trace_nfsd_file_opened(nf, status);
1183		} else {
1184			status = nfsd_open_verified(rqstp, fhp, may_flags,
1185						    &nf->nf_file);
1186			trace_nfsd_file_open(nf, status);
1187		}
1188	} else
 
 
 
 
 
1189		status = nfserr_jukebox;
1190	/*
1191	 * If construction failed, or we raced with a call to unlink()
1192	 * then unhash.
1193	 */
1194	if (status == nfs_ok && key.inode->i_nlink == 0)
1195		status = nfserr_jukebox;
1196	if (status != nfs_ok)
1197		nfsd_file_unhash(nf);
 
 
 
 
1198	clear_bit_unlock(NFSD_FILE_PENDING, &nf->nf_flags);
1199	smp_mb__after_atomic();
1200	wake_up_bit(&nf->nf_flags, NFSD_FILE_PENDING);
1201	goto out;
1202}
1203
1204/**
1205 * nfsd_file_acquire_gc - Get a struct nfsd_file with an open file
1206 * @rqstp: the RPC transaction being executed
1207 * @fhp: the NFS filehandle of the file to be opened
1208 * @may_flags: NFSD_MAY_ settings for the file
1209 * @pnf: OUT: new or found "struct nfsd_file" object
1210 *
1211 * The nfsd_file object returned by this API is reference-counted
1212 * and garbage-collected. The object is retained for a few
1213 * seconds after the final nfsd_file_put() in case the caller
1214 * wants to re-use it.
1215 *
1216 * Returns nfs_ok and sets @pnf on success; otherwise an nfsstat in
1217 * network byte order is returned.
1218 */
1219__be32
1220nfsd_file_acquire_gc(struct svc_rqst *rqstp, struct svc_fh *fhp,
1221		     unsigned int may_flags, struct nfsd_file **pnf)
1222{
1223	return nfsd_file_do_acquire(rqstp, fhp, may_flags, NULL, pnf, true);
1224}
1225
1226/**
1227 * nfsd_file_acquire - Get a struct nfsd_file with an open file
1228 * @rqstp: the RPC transaction being executed
1229 * @fhp: the NFS filehandle of the file to be opened
1230 * @may_flags: NFSD_MAY_ settings for the file
1231 * @pnf: OUT: new or found "struct nfsd_file" object
1232 *
1233 * The nfsd_file_object returned by this API is reference-counted
1234 * but not garbage-collected. The object is unhashed after the
1235 * final nfsd_file_put().
1236 *
1237 * Returns nfs_ok and sets @pnf on success; otherwise an nfsstat in
1238 * network byte order is returned.
1239 */
1240__be32
1241nfsd_file_acquire(struct svc_rqst *rqstp, struct svc_fh *fhp,
1242		  unsigned int may_flags, struct nfsd_file **pnf)
1243{
1244	return nfsd_file_do_acquire(rqstp, fhp, may_flags, NULL, pnf, false);
1245}
1246
1247/**
1248 * nfsd_file_acquire_opened - Get a struct nfsd_file using existing open file
1249 * @rqstp: the RPC transaction being executed
1250 * @fhp: the NFS filehandle of the file just created
1251 * @may_flags: NFSD_MAY_ settings for the file
1252 * @file: cached, already-open file (may be NULL)
1253 * @pnf: OUT: new or found "struct nfsd_file" object
1254 *
1255 * Acquire a nfsd_file object that is not GC'ed. If one doesn't already exist,
1256 * and @file is non-NULL, use it to instantiate a new nfsd_file instead of
1257 * opening a new one.
1258 *
1259 * Returns nfs_ok and sets @pnf on success; otherwise an nfsstat in
1260 * network byte order is returned.
1261 */
1262__be32
1263nfsd_file_acquire_opened(struct svc_rqst *rqstp, struct svc_fh *fhp,
1264			 unsigned int may_flags, struct file *file,
1265			 struct nfsd_file **pnf)
1266{
1267	return nfsd_file_do_acquire(rqstp, fhp, may_flags, file, pnf, false);
1268}
1269
1270/*
1271 * Note that fields may be added, removed or reordered in the future. Programs
1272 * scraping this file for info should test the labels to ensure they're
1273 * getting the correct field.
1274 */
1275int nfsd_file_cache_stats_show(struct seq_file *m, void *v)
1276{
1277	unsigned long releases = 0, evictions = 0;
1278	unsigned long hits = 0, acquisitions = 0;
1279	unsigned int i, count = 0, buckets = 0;
1280	unsigned long lru = 0, total_age = 0;
1281
1282	/* Serialize with server shutdown */
 
 
 
 
1283	mutex_lock(&nfsd_mutex);
1284	if (test_bit(NFSD_FILE_CACHE_UP, &nfsd_file_flags) == 1) {
1285		struct bucket_table *tbl;
1286		struct rhashtable *ht;
1287
1288		lru = list_lru_count(&nfsd_file_lru);
1289
1290		rcu_read_lock();
1291		ht = &nfsd_file_rhash_tbl;
1292		count = atomic_read(&ht->nelems);
1293		tbl = rht_dereference_rcu(ht->tbl, ht);
1294		buckets = tbl->size;
1295		rcu_read_unlock();
1296	}
1297	mutex_unlock(&nfsd_mutex);
1298
1299	for_each_possible_cpu(i) {
1300		hits += per_cpu(nfsd_file_cache_hits, i);
1301		acquisitions += per_cpu(nfsd_file_acquisitions, i);
1302		releases += per_cpu(nfsd_file_releases, i);
1303		total_age += per_cpu(nfsd_file_total_age, i);
1304		evictions += per_cpu(nfsd_file_evictions, i);
1305	}
1306
1307	seq_printf(m, "total entries: %u\n", count);
1308	seq_printf(m, "hash buckets:  %u\n", buckets);
1309	seq_printf(m, "lru entries:   %lu\n", lru);
1310	seq_printf(m, "cache hits:    %lu\n", hits);
1311	seq_printf(m, "acquisitions:  %lu\n", acquisitions);
1312	seq_printf(m, "releases:      %lu\n", releases);
1313	seq_printf(m, "evictions:     %lu\n", evictions);
1314	if (releases)
1315		seq_printf(m, "mean age (ms): %ld\n", total_age / releases);
1316	else
1317		seq_printf(m, "mean age (ms): -\n");
1318	return 0;
 
 
 
 
 
1319}
v5.14.15
 
   1/*
   2 * Open file cache.
   3 *
   4 * (c) 2015 - Jeff Layton <jeff.layton@primarydata.com>
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
   5 */
   6
   7#include <linux/hash.h>
   8#include <linux/slab.h>
   9#include <linux/file.h>
 
  10#include <linux/sched.h>
  11#include <linux/list_lru.h>
  12#include <linux/fsnotify_backend.h>
  13#include <linux/fsnotify.h>
  14#include <linux/seq_file.h>
 
  15
  16#include "vfs.h"
  17#include "nfsd.h"
  18#include "nfsfh.h"
  19#include "netns.h"
  20#include "filecache.h"
  21#include "trace.h"
  22
  23#define NFSDDBG_FACILITY	NFSDDBG_FH
  24
  25/* FIXME: dynamically size this for the machine somehow? */
  26#define NFSD_FILE_HASH_BITS                   12
  27#define NFSD_FILE_HASH_SIZE                  (1 << NFSD_FILE_HASH_BITS)
  28#define NFSD_LAUNDRETTE_DELAY		     (2 * HZ)
  29
  30#define NFSD_FILE_SHUTDOWN		     (1)
  31#define NFSD_FILE_LRU_THRESHOLD		     (4096UL)
  32#define NFSD_FILE_LRU_LIMIT		     (NFSD_FILE_LRU_THRESHOLD << 2)
  33
  34/* We only care about NFSD_MAY_READ/WRITE for this cache */
  35#define NFSD_FILE_MAY_MASK	(NFSD_MAY_READ|NFSD_MAY_WRITE)
  36
  37struct nfsd_fcache_bucket {
  38	struct hlist_head	nfb_head;
  39	spinlock_t		nfb_lock;
  40	unsigned int		nfb_count;
  41	unsigned int		nfb_maxcount;
  42};
  43
  44static DEFINE_PER_CPU(unsigned long, nfsd_file_cache_hits);
 
 
 
 
  45
  46struct nfsd_fcache_disposal {
  47	struct list_head list;
  48	struct work_struct work;
  49	struct net *net;
  50	spinlock_t lock;
  51	struct list_head freeme;
  52	struct rcu_head rcu;
  53};
  54
  55static struct workqueue_struct *nfsd_filecache_wq __read_mostly;
  56
  57static struct kmem_cache		*nfsd_file_slab;
  58static struct kmem_cache		*nfsd_file_mark_slab;
  59static struct nfsd_fcache_bucket	*nfsd_file_hashtbl;
  60static struct list_lru			nfsd_file_lru;
  61static long				nfsd_file_lru_flags;
  62static struct fsnotify_group		*nfsd_file_fsnotify_group;
  63static atomic_long_t			nfsd_filecache_count;
  64static struct delayed_work		nfsd_filecache_laundrette;
  65static DEFINE_SPINLOCK(laundrette_lock);
  66static LIST_HEAD(laundrettes);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
  67
  68static void nfsd_file_gc(void);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
  69
  70static void
  71nfsd_file_schedule_laundrette(void)
  72{
  73	long count = atomic_long_read(&nfsd_filecache_count);
  74
  75	if (count == 0 || test_bit(NFSD_FILE_SHUTDOWN, &nfsd_file_lru_flags))
  76		return;
  77
  78	queue_delayed_work(system_wq, &nfsd_filecache_laundrette,
  79			NFSD_LAUNDRETTE_DELAY);
  80}
  81
  82static void
  83nfsd_file_slab_free(struct rcu_head *rcu)
  84{
  85	struct nfsd_file *nf = container_of(rcu, struct nfsd_file, nf_rcu);
  86
  87	put_cred(nf->nf_cred);
  88	kmem_cache_free(nfsd_file_slab, nf);
  89}
  90
  91static void
  92nfsd_file_mark_free(struct fsnotify_mark *mark)
  93{
  94	struct nfsd_file_mark *nfm = container_of(mark, struct nfsd_file_mark,
  95						  nfm_mark);
  96
  97	kmem_cache_free(nfsd_file_mark_slab, nfm);
  98}
  99
 100static struct nfsd_file_mark *
 101nfsd_file_mark_get(struct nfsd_file_mark *nfm)
 102{
 103	if (!refcount_inc_not_zero(&nfm->nfm_ref))
 104		return NULL;
 105	return nfm;
 106}
 107
 108static void
 109nfsd_file_mark_put(struct nfsd_file_mark *nfm)
 110{
 111	if (refcount_dec_and_test(&nfm->nfm_ref)) {
 112		fsnotify_destroy_mark(&nfm->nfm_mark, nfsd_file_fsnotify_group);
 113		fsnotify_put_mark(&nfm->nfm_mark);
 114	}
 115}
 116
 117static struct nfsd_file_mark *
 118nfsd_file_mark_find_or_create(struct nfsd_file *nf)
 119{
 120	int			err;
 121	struct fsnotify_mark	*mark;
 122	struct nfsd_file_mark	*nfm = NULL, *new;
 123	struct inode *inode = nf->nf_inode;
 124
 125	do {
 126		mutex_lock(&nfsd_file_fsnotify_group->mark_mutex);
 127		mark = fsnotify_find_mark(&inode->i_fsnotify_marks,
 128				nfsd_file_fsnotify_group);
 129		if (mark) {
 130			nfm = nfsd_file_mark_get(container_of(mark,
 131						 struct nfsd_file_mark,
 132						 nfm_mark));
 133			mutex_unlock(&nfsd_file_fsnotify_group->mark_mutex);
 134			if (nfm) {
 135				fsnotify_put_mark(mark);
 136				break;
 137			}
 138			/* Avoid soft lockup race with nfsd_file_mark_put() */
 139			fsnotify_destroy_mark(mark, nfsd_file_fsnotify_group);
 140			fsnotify_put_mark(mark);
 141		} else
 142			mutex_unlock(&nfsd_file_fsnotify_group->mark_mutex);
 
 143
 144		/* allocate a new nfm */
 145		new = kmem_cache_alloc(nfsd_file_mark_slab, GFP_KERNEL);
 146		if (!new)
 147			return NULL;
 148		fsnotify_init_mark(&new->nfm_mark, nfsd_file_fsnotify_group);
 149		new->nfm_mark.mask = FS_ATTRIB|FS_DELETE_SELF;
 150		refcount_set(&new->nfm_ref, 1);
 151
 152		err = fsnotify_add_inode_mark(&new->nfm_mark, inode, 0);
 153
 154		/*
 155		 * If the add was successful, then return the object.
 156		 * Otherwise, we need to put the reference we hold on the
 157		 * nfm_mark. The fsnotify code will take a reference and put
 158		 * it on failure, so we can't just free it directly. It's also
 159		 * not safe to call fsnotify_destroy_mark on it as the
 160		 * mark->group will be NULL. Thus, we can't let the nfm_ref
 161		 * counter drive the destruction at this point.
 162		 */
 163		if (likely(!err))
 164			nfm = new;
 165		else
 166			fsnotify_put_mark(&new->nfm_mark);
 167	} while (unlikely(err == -EEXIST));
 168
 169	return nfm;
 170}
 171
 172static struct nfsd_file *
 173nfsd_file_alloc(struct inode *inode, unsigned int may, unsigned int hashval,
 174		struct net *net)
 175{
 176	struct nfsd_file *nf;
 177
 178	nf = kmem_cache_alloc(nfsd_file_slab, GFP_KERNEL);
 179	if (nf) {
 180		INIT_HLIST_NODE(&nf->nf_node);
 181		INIT_LIST_HEAD(&nf->nf_lru);
 
 182		nf->nf_file = NULL;
 183		nf->nf_cred = get_current_cred();
 184		nf->nf_net = net;
 185		nf->nf_flags = 0;
 186		nf->nf_inode = inode;
 187		nf->nf_hashval = hashval;
 
 
 
 188		refcount_set(&nf->nf_ref, 1);
 189		nf->nf_may = may & NFSD_FILE_MAY_MASK;
 190		if (may & NFSD_MAY_NOT_BREAK_LEASE) {
 191			if (may & NFSD_MAY_WRITE)
 192				__set_bit(NFSD_FILE_BREAK_WRITE, &nf->nf_flags);
 193			if (may & NFSD_MAY_READ)
 194				__set_bit(NFSD_FILE_BREAK_READ, &nf->nf_flags);
 195		}
 196		nf->nf_mark = NULL;
 197		init_rwsem(&nf->nf_rwsem);
 198		trace_nfsd_file_alloc(nf);
 199	}
 200	return nf;
 201}
 202
 203static bool
 204nfsd_file_free(struct nfsd_file *nf)
 205{
 206	bool flush = false;
 207
 208	trace_nfsd_file_put_final(nf);
 209	if (nf->nf_mark)
 210		nfsd_file_mark_put(nf->nf_mark);
 211	if (nf->nf_file) {
 212		get_file(nf->nf_file);
 213		filp_close(nf->nf_file, NULL);
 214		fput(nf->nf_file);
 215		flush = true;
 216	}
 217	call_rcu(&nf->nf_rcu, nfsd_file_slab_free);
 218	return flush;
 219}
 220
 221static bool
 222nfsd_file_check_writeback(struct nfsd_file *nf)
 223{
 224	struct file *file = nf->nf_file;
 225	struct address_space *mapping;
 226
 227	if (!file || !(file->f_mode & FMODE_WRITE))
 228		return false;
 229	mapping = file->f_mapping;
 230	return mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) ||
 231		mapping_tagged(mapping, PAGECACHE_TAG_WRITEBACK);
 
 232}
 233
 234static int
 235nfsd_file_check_write_error(struct nfsd_file *nf)
 236{
 237	struct file *file = nf->nf_file;
 238
 239	if (!file || !(file->f_mode & FMODE_WRITE))
 240		return 0;
 241	return filemap_check_wb_err(file->f_mapping, READ_ONCE(file->f_wb_err));
 242}
 243
 244static void
 245nfsd_file_do_unhash(struct nfsd_file *nf)
 246{
 247	lockdep_assert_held(&nfsd_file_hashtbl[nf->nf_hashval].nfb_lock);
 248
 249	trace_nfsd_file_unhash(nf);
 250
 251	if (nfsd_file_check_write_error(nf))
 252		nfsd_reset_boot_verifier(net_generic(nf->nf_net, nfsd_net_id));
 253	--nfsd_file_hashtbl[nf->nf_hashval].nfb_count;
 254	hlist_del_rcu(&nf->nf_node);
 255	atomic_long_dec(&nfsd_filecache_count);
 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_do_unhash(nf);
 263		if (!list_empty(&nf->nf_lru))
 264			list_lru_del(&nfsd_file_lru, &nf->nf_lru);
 265		return true;
 266	}
 267	return false;
 268}
 269
 270/*
 271 * Return true if the file was unhashed.
 272 */
 273static bool
 274nfsd_file_unhash_and_release_locked(struct nfsd_file *nf, struct list_head *dispose)
 275{
 276	lockdep_assert_held(&nfsd_file_hashtbl[nf->nf_hashval].nfb_lock);
 
 
 
 
 
 
 
 277
 278	trace_nfsd_file_unhash_and_release_locked(nf);
 279	if (!nfsd_file_unhash(nf))
 280		return false;
 281	/* keep final reference for nfsd_file_lru_dispose */
 282	if (refcount_dec_not_one(&nf->nf_ref))
 283		return true;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 284
 285	list_add(&nf->nf_lru, dispose);
 286	return true;
 287}
 288
 289static void
 290nfsd_file_put_noref(struct nfsd_file *nf)
 291{
 292	trace_nfsd_file_put(nf);
 
 293
 294	if (refcount_dec_and_test(&nf->nf_ref)) {
 295		WARN_ON(test_bit(NFSD_FILE_HASHED, &nf->nf_flags));
 296		nfsd_file_free(nf);
 297	}
 
 298}
 299
 300void
 301nfsd_file_put(struct nfsd_file *nf)
 302{
 303	bool is_hashed;
 304
 305	set_bit(NFSD_FILE_REFERENCED, &nf->nf_flags);
 306	if (refcount_read(&nf->nf_ref) > 2 || !nf->nf_file) {
 307		nfsd_file_put_noref(nf);
 308		return;
 309	}
 
 
 310
 311	filemap_flush(nf->nf_file->f_mapping);
 312	is_hashed = test_bit(NFSD_FILE_HASHED, &nf->nf_flags) != 0;
 313	nfsd_file_put_noref(nf);
 314	if (is_hashed)
 315		nfsd_file_schedule_laundrette();
 316	if (atomic_long_read(&nfsd_filecache_count) >= NFSD_FILE_LRU_LIMIT)
 317		nfsd_file_gc();
 318}
 319
 320struct nfsd_file *
 321nfsd_file_get(struct nfsd_file *nf)
 322{
 323	if (likely(refcount_inc_not_zero(&nf->nf_ref)))
 324		return nf;
 325	return NULL;
 326}
 327
 328static void
 329nfsd_file_dispose_list(struct list_head *dispose)
 
 
 
 
 
 
 
 
 330{
 331	struct nfsd_file *nf;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 332
 333	while(!list_empty(dispose)) {
 334		nf = list_first_entry(dispose, struct nfsd_file, nf_lru);
 335		list_del(&nf->nf_lru);
 336		nfsd_file_put_noref(nf);
 
 
 
 
 337	}
 
 
 338}
 339
 340static void
 341nfsd_file_dispose_list_sync(struct list_head *dispose)
 342{
 343	bool flush = false;
 344	struct nfsd_file *nf;
 345
 346	while(!list_empty(dispose)) {
 347		nf = list_first_entry(dispose, struct nfsd_file, nf_lru);
 348		list_del(&nf->nf_lru);
 349		if (!refcount_dec_and_test(&nf->nf_ref))
 350			continue;
 351		if (nfsd_file_free(nf))
 352			flush = true;
 353	}
 354	if (flush)
 355		flush_delayed_fput();
 356}
 357
 358static void
 359nfsd_file_list_remove_disposal(struct list_head *dst,
 360		struct nfsd_fcache_disposal *l)
 361{
 362	spin_lock(&l->lock);
 363	list_splice_init(&l->freeme, dst);
 364	spin_unlock(&l->lock);
 365}
 366
 367static void
 368nfsd_file_list_add_disposal(struct list_head *files, struct net *net)
 369{
 370	struct nfsd_fcache_disposal *l;
 
 371
 372	rcu_read_lock();
 373	list_for_each_entry_rcu(l, &laundrettes, list) {
 374		if (l->net == net) {
 375			spin_lock(&l->lock);
 376			list_splice_tail_init(files, &l->freeme);
 377			spin_unlock(&l->lock);
 378			queue_work(nfsd_filecache_wq, &l->work);
 379			break;
 380		}
 381	}
 382	rcu_read_unlock();
 383}
 384
 385static void
 386nfsd_file_list_add_pernet(struct list_head *dst, struct list_head *src,
 387		struct net *net)
 388{
 389	struct nfsd_file *nf, *tmp;
 390
 391	list_for_each_entry_safe(nf, tmp, src, nf_lru) {
 392		if (nf->nf_net == net)
 393			list_move_tail(&nf->nf_lru, dst);
 394	}
 395}
 396
 397static void
 398nfsd_file_dispose_list_delayed(struct list_head *dispose)
 399{
 400	LIST_HEAD(list);
 401	struct nfsd_file *nf;
 402
 403	while(!list_empty(dispose)) {
 404		nf = list_first_entry(dispose, struct nfsd_file, nf_lru);
 405		nfsd_file_list_add_pernet(&list, dispose, nf->nf_net);
 406		nfsd_file_list_add_disposal(&list, nf->nf_net);
 407	}
 408}
 409
 410/*
 411 * Note this can deadlock with nfsd_file_cache_purge.
 
 
 
 
 
 
 
 
 
 412 */
 413static enum lru_status
 414nfsd_file_lru_cb(struct list_head *item, struct list_lru_one *lru,
 415		 spinlock_t *lock, void *arg)
 416	__releases(lock)
 417	__acquires(lock)
 418{
 419	struct list_head *head = arg;
 420	struct nfsd_file *nf = list_entry(item, struct nfsd_file, nf_lru);
 421
 422	/*
 423	 * Do a lockless refcount check. The hashtable holds one reference, so
 424	 * we look to see if anything else has a reference, or if any have
 425	 * been put since the shrinker last ran. Those don't get unhashed and
 426	 * released.
 427	 *
 428	 * Note that in the put path, we set the flag and then decrement the
 429	 * counter. Here we check the counter and then test and clear the flag.
 430	 * That order is deliberate to ensure that we can do this locklessly.
 431	 */
 432	if (refcount_read(&nf->nf_ref) > 1)
 433		goto out_skip;
 434
 435	/*
 436	 * Don't throw out files that are still undergoing I/O or
 437	 * that have uncleared errors pending.
 438	 */
 439	if (nfsd_file_check_writeback(nf))
 440		goto out_skip;
 
 
 441
 442	if (test_and_clear_bit(NFSD_FILE_REFERENCED, &nf->nf_flags))
 443		goto out_skip;
 
 
 
 444
 445	if (!test_and_clear_bit(NFSD_FILE_HASHED, &nf->nf_flags))
 446		goto out_skip;
 
 
 
 
 
 
 
 447
 
 
 448	list_lru_isolate_move(lru, &nf->nf_lru, head);
 
 
 449	return LRU_REMOVED;
 450out_skip:
 451	return LRU_SKIP;
 452}
 453
 454static unsigned long
 455nfsd_file_lru_walk_list(struct shrink_control *sc)
 456{
 457	LIST_HEAD(head);
 458	struct nfsd_file *nf;
 459	unsigned long ret;
 460
 461	if (sc)
 462		ret = list_lru_shrink_walk(&nfsd_file_lru, sc,
 463				nfsd_file_lru_cb, &head);
 464	else
 465		ret = list_lru_walk(&nfsd_file_lru,
 466				nfsd_file_lru_cb,
 467				&head, LONG_MAX);
 468	list_for_each_entry(nf, &head, nf_lru) {
 469		spin_lock(&nfsd_file_hashtbl[nf->nf_hashval].nfb_lock);
 470		nfsd_file_do_unhash(nf);
 471		spin_unlock(&nfsd_file_hashtbl[nf->nf_hashval].nfb_lock);
 472	}
 473	nfsd_file_dispose_list_delayed(&head);
 474	return ret;
 475}
 476
 477static void
 478nfsd_file_gc(void)
 479{
 480	nfsd_file_lru_walk_list(NULL);
 481}
 482
 483static void
 484nfsd_file_gc_worker(struct work_struct *work)
 485{
 486	nfsd_file_gc();
 487	nfsd_file_schedule_laundrette();
 
 488}
 489
 490static unsigned long
 491nfsd_file_lru_count(struct shrinker *s, struct shrink_control *sc)
 492{
 493	return list_lru_count(&nfsd_file_lru);
 494}
 495
 496static unsigned long
 497nfsd_file_lru_scan(struct shrinker *s, struct shrink_control *sc)
 498{
 499	return nfsd_file_lru_walk_list(sc);
 
 
 
 
 
 
 
 500}
 501
 502static struct shrinker	nfsd_file_shrinker = {
 503	.scan_objects = nfsd_file_lru_scan,
 504	.count_objects = nfsd_file_lru_count,
 505	.seeks = 1,
 506};
 507
 
 
 
 
 
 
 
 
 508static void
 509__nfsd_file_close_inode(struct inode *inode, unsigned int hashval,
 510			struct list_head *dispose)
 511{
 512	struct nfsd_file	*nf;
 513	struct hlist_node	*tmp;
 
 
 
 514
 515	spin_lock(&nfsd_file_hashtbl[hashval].nfb_lock);
 516	hlist_for_each_entry_safe(nf, tmp, &nfsd_file_hashtbl[hashval].nfb_head, nf_node) {
 517		if (inode == nf->nf_inode)
 518			nfsd_file_unhash_and_release_locked(nf, dispose);
 
 
 
 
 
 
 
 
 519	}
 520	spin_unlock(&nfsd_file_hashtbl[hashval].nfb_lock);
 521}
 522
 523/**
 524 * nfsd_file_close_inode_sync - attempt to forcibly close a nfsd_file
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 525 * @inode: inode of the file to attempt to remove
 526 *
 527 * Walk the whole hash bucket, looking for any files that correspond to "inode".
 528 * If any do, then unhash them and put the hashtable reference to them and
 529 * destroy any that had their last reference put. Also ensure that any of the
 530 * fputs also have their final __fput done as well.
 531 */
 532void
 533nfsd_file_close_inode_sync(struct inode *inode)
 534{
 535	unsigned int		hashval = (unsigned int)hash_long(inode->i_ino,
 536						NFSD_FILE_HASH_BITS);
 537	LIST_HEAD(dispose);
 538
 539	__nfsd_file_close_inode(inode, hashval, &dispose);
 540	trace_nfsd_file_close_inode_sync(inode, hashval, !list_empty(&dispose));
 541	nfsd_file_dispose_list_sync(&dispose);
 542}
 543
 544/**
 545 * nfsd_file_close_inode_sync - attempt to forcibly close a nfsd_file
 546 * @inode: inode of the file to attempt to remove
 547 *
 548 * Walk the whole hash bucket, looking for any files that correspond to "inode".
 549 * If any do, then unhash them and put the hashtable reference to them and
 550 * destroy any that had their last reference put.
 
 
 551 */
 552static void
 553nfsd_file_close_inode(struct inode *inode)
 554{
 555	unsigned int		hashval = (unsigned int)hash_long(inode->i_ino,
 556						NFSD_FILE_HASH_BITS);
 557	LIST_HEAD(dispose);
 558
 559	__nfsd_file_close_inode(inode, hashval, &dispose);
 560	trace_nfsd_file_close_inode(inode, hashval, !list_empty(&dispose));
 561	nfsd_file_dispose_list_delayed(&dispose);
 
 
 
 
 
 
 562}
 563
 564/**
 565 * nfsd_file_delayed_close - close unused nfsd_files
 566 * @work: dummy
 567 *
 568 * Walk the LRU list and close any entries that have not been used since
 569 * the last scan.
 570 *
 571 * Note this can deadlock with nfsd_file_cache_purge.
 572 */
 573static void
 574nfsd_file_delayed_close(struct work_struct *work)
 575{
 576	LIST_HEAD(head);
 577	struct nfsd_fcache_disposal *l = container_of(work,
 578			struct nfsd_fcache_disposal, work);
 579
 580	nfsd_file_list_remove_disposal(&head, l);
 581	nfsd_file_dispose_list(&head);
 582}
 583
 584static int
 585nfsd_file_lease_notifier_call(struct notifier_block *nb, unsigned long arg,
 586			    void *data)
 587{
 588	struct file_lock *fl = data;
 589
 590	/* Only close files for F_SETLEASE leases */
 591	if (fl->fl_flags & FL_LEASE)
 592		nfsd_file_close_inode_sync(file_inode(fl->fl_file));
 593	return 0;
 594}
 595
 596static struct notifier_block nfsd_file_lease_notifier = {
 597	.notifier_call = nfsd_file_lease_notifier_call,
 598};
 599
 600static int
 601nfsd_file_fsnotify_handle_event(struct fsnotify_mark *mark, u32 mask,
 602				struct inode *inode, struct inode *dir,
 603				const struct qstr *name, u32 cookie)
 604{
 
 
 
 605	trace_nfsd_file_fsnotify_handle_event(inode, mask);
 606
 607	/* Should be no marks on non-regular files */
 608	if (!S_ISREG(inode->i_mode)) {
 609		WARN_ON_ONCE(1);
 610		return 0;
 611	}
 612
 613	/* don't close files if this was not the last link */
 614	if (mask & FS_ATTRIB) {
 615		if (inode->i_nlink)
 616			return 0;
 617	}
 618
 619	nfsd_file_close_inode(inode);
 620	return 0;
 621}
 622
 623
 624static const struct fsnotify_ops nfsd_file_fsnotify_ops = {
 625	.handle_inode_event = nfsd_file_fsnotify_handle_event,
 626	.free_mark = nfsd_file_mark_free,
 627};
 628
 629int
 630nfsd_file_cache_init(void)
 631{
 632	int		ret = -ENOMEM;
 633	unsigned int	i;
 634
 635	clear_bit(NFSD_FILE_SHUTDOWN, &nfsd_file_lru_flags);
 
 
 636
 637	if (nfsd_file_hashtbl)
 638		return 0;
 
 639
 
 640	nfsd_filecache_wq = alloc_workqueue("nfsd_filecache", 0, 0);
 641	if (!nfsd_filecache_wq)
 642		goto out;
 643
 644	nfsd_file_hashtbl = kcalloc(NFSD_FILE_HASH_SIZE,
 645				sizeof(*nfsd_file_hashtbl), GFP_KERNEL);
 646	if (!nfsd_file_hashtbl) {
 647		pr_err("nfsd: unable to allocate nfsd_file_hashtbl\n");
 648		goto out_err;
 649	}
 650
 651	nfsd_file_slab = kmem_cache_create("nfsd_file",
 652				sizeof(struct nfsd_file), 0, 0, NULL);
 653	if (!nfsd_file_slab) {
 654		pr_err("nfsd: unable to create nfsd_file_slab\n");
 655		goto out_err;
 656	}
 657
 658	nfsd_file_mark_slab = kmem_cache_create("nfsd_file_mark",
 659					sizeof(struct nfsd_file_mark), 0, 0, NULL);
 660	if (!nfsd_file_mark_slab) {
 661		pr_err("nfsd: unable to create nfsd_file_mark_slab\n");
 662		goto out_err;
 663	}
 664
 665
 666	ret = list_lru_init(&nfsd_file_lru);
 667	if (ret) {
 668		pr_err("nfsd: failed to init nfsd_file_lru: %d\n", ret);
 669		goto out_err;
 670	}
 671
 672	ret = register_shrinker(&nfsd_file_shrinker);
 673	if (ret) {
 674		pr_err("nfsd: failed to register nfsd_file_shrinker: %d\n", ret);
 675		goto out_lru;
 676	}
 677
 678	ret = lease_register_notifier(&nfsd_file_lease_notifier);
 679	if (ret) {
 680		pr_err("nfsd: unable to register lease notifier: %d\n", ret);
 681		goto out_shrinker;
 682	}
 683
 684	nfsd_file_fsnotify_group = fsnotify_alloc_group(&nfsd_file_fsnotify_ops);
 
 685	if (IS_ERR(nfsd_file_fsnotify_group)) {
 686		pr_err("nfsd: unable to create fsnotify group: %ld\n",
 687			PTR_ERR(nfsd_file_fsnotify_group));
 688		ret = PTR_ERR(nfsd_file_fsnotify_group);
 689		nfsd_file_fsnotify_group = NULL;
 690		goto out_notifier;
 691	}
 692
 693	for (i = 0; i < NFSD_FILE_HASH_SIZE; i++) {
 694		INIT_HLIST_HEAD(&nfsd_file_hashtbl[i].nfb_head);
 695		spin_lock_init(&nfsd_file_hashtbl[i].nfb_lock);
 696	}
 697
 698	INIT_DELAYED_WORK(&nfsd_filecache_laundrette, nfsd_file_gc_worker);
 699out:
 700	return ret;
 701out_notifier:
 702	lease_unregister_notifier(&nfsd_file_lease_notifier);
 703out_shrinker:
 704	unregister_shrinker(&nfsd_file_shrinker);
 705out_lru:
 706	list_lru_destroy(&nfsd_file_lru);
 707out_err:
 708	kmem_cache_destroy(nfsd_file_slab);
 709	nfsd_file_slab = NULL;
 710	kmem_cache_destroy(nfsd_file_mark_slab);
 711	nfsd_file_mark_slab = NULL;
 712	kfree(nfsd_file_hashtbl);
 713	nfsd_file_hashtbl = NULL;
 714	destroy_workqueue(nfsd_filecache_wq);
 715	nfsd_filecache_wq = NULL;
 
 716	goto out;
 717}
 718
 719/*
 720 * Note this can deadlock with nfsd_file_lru_cb.
 
 
 
 
 721 */
 722void
 723nfsd_file_cache_purge(struct net *net)
 724{
 725	unsigned int		i;
 726	struct nfsd_file	*nf;
 727	struct hlist_node	*next;
 728	LIST_HEAD(dispose);
 729	bool del;
 730
 731	if (!nfsd_file_hashtbl)
 732		return;
 
 733
 734	for (i = 0; i < NFSD_FILE_HASH_SIZE; i++) {
 735		struct nfsd_fcache_bucket *nfb = &nfsd_file_hashtbl[i];
 
 
 
 
 736
 737		spin_lock(&nfb->nfb_lock);
 738		hlist_for_each_entry_safe(nf, next, &nfb->nfb_head, nf_node) {
 739			if (net && nf->nf_net != net)
 740				continue;
 741			del = nfsd_file_unhash_and_release_locked(nf, &dispose);
 742
 743			/*
 744			 * Deadlock detected! Something marked this entry as
 745			 * unhased, but hasn't removed it from the hash list.
 746			 */
 747			WARN_ON_ONCE(!del);
 748		}
 749		spin_unlock(&nfb->nfb_lock);
 750		nfsd_file_dispose_list(&dispose);
 751	}
 752}
 753
 754static struct nfsd_fcache_disposal *
 755nfsd_alloc_fcache_disposal(struct net *net)
 756{
 757	struct nfsd_fcache_disposal *l;
 758
 759	l = kmalloc(sizeof(*l), GFP_KERNEL);
 760	if (!l)
 761		return NULL;
 762	INIT_WORK(&l->work, nfsd_file_delayed_close);
 763	l->net = net;
 764	spin_lock_init(&l->lock);
 765	INIT_LIST_HEAD(&l->freeme);
 766	return l;
 767}
 768
 769static void
 770nfsd_free_fcache_disposal(struct nfsd_fcache_disposal *l)
 771{
 772	rcu_assign_pointer(l->net, NULL);
 773	cancel_work_sync(&l->work);
 774	nfsd_file_dispose_list(&l->freeme);
 775	kfree_rcu(l, rcu);
 776}
 777
 778static void
 779nfsd_add_fcache_disposal(struct nfsd_fcache_disposal *l)
 780{
 781	spin_lock(&laundrette_lock);
 782	list_add_tail_rcu(&l->list, &laundrettes);
 783	spin_unlock(&laundrette_lock);
 784}
 785
 786static void
 787nfsd_del_fcache_disposal(struct nfsd_fcache_disposal *l)
 788{
 789	spin_lock(&laundrette_lock);
 790	list_del_rcu(&l->list);
 791	spin_unlock(&laundrette_lock);
 792}
 793
 794static int
 795nfsd_alloc_fcache_disposal_net(struct net *net)
 796{
 797	struct nfsd_fcache_disposal *l;
 798
 799	l = nfsd_alloc_fcache_disposal(net);
 800	if (!l)
 801		return -ENOMEM;
 802	nfsd_add_fcache_disposal(l);
 803	return 0;
 804}
 805
 806static void
 807nfsd_free_fcache_disposal_net(struct net *net)
 
 
 
 
 
 808{
 809	struct nfsd_fcache_disposal *l;
 810
 811	rcu_read_lock();
 812	list_for_each_entry_rcu(l, &laundrettes, list) {
 813		if (l->net != net)
 814			continue;
 815		nfsd_del_fcache_disposal(l);
 816		rcu_read_unlock();
 817		nfsd_free_fcache_disposal(l);
 818		return;
 819	}
 820	rcu_read_unlock();
 821}
 822
 823int
 824nfsd_file_cache_start_net(struct net *net)
 825{
 826	return nfsd_alloc_fcache_disposal_net(net);
 827}
 828
 829void
 830nfsd_file_cache_shutdown_net(struct net *net)
 831{
 832	nfsd_file_cache_purge(net);
 833	nfsd_free_fcache_disposal_net(net);
 834}
 835
 836void
 837nfsd_file_cache_shutdown(void)
 838{
 839	set_bit(NFSD_FILE_SHUTDOWN, &nfsd_file_lru_flags);
 
 
 
 
 840
 841	lease_unregister_notifier(&nfsd_file_lease_notifier);
 842	unregister_shrinker(&nfsd_file_shrinker);
 843	/*
 844	 * make sure all callers of nfsd_file_lru_cb are done before
 845	 * calling nfsd_file_cache_purge
 846	 */
 847	cancel_delayed_work_sync(&nfsd_filecache_laundrette);
 848	nfsd_file_cache_purge(NULL);
 849	list_lru_destroy(&nfsd_file_lru);
 850	rcu_barrier();
 851	fsnotify_put_group(nfsd_file_fsnotify_group);
 852	nfsd_file_fsnotify_group = NULL;
 853	kmem_cache_destroy(nfsd_file_slab);
 854	nfsd_file_slab = NULL;
 855	fsnotify_wait_marks_destroyed();
 856	kmem_cache_destroy(nfsd_file_mark_slab);
 857	nfsd_file_mark_slab = NULL;
 858	kfree(nfsd_file_hashtbl);
 859	nfsd_file_hashtbl = NULL;
 860	destroy_workqueue(nfsd_filecache_wq);
 861	nfsd_filecache_wq = NULL;
 862}
 863
 864static bool
 865nfsd_match_cred(const struct cred *c1, const struct cred *c2)
 866{
 867	int i;
 868
 869	if (!uid_eq(c1->fsuid, c2->fsuid))
 870		return false;
 871	if (!gid_eq(c1->fsgid, c2->fsgid))
 872		return false;
 873	if (c1->group_info == NULL || c2->group_info == NULL)
 874		return c1->group_info == c2->group_info;
 875	if (c1->group_info->ngroups != c2->group_info->ngroups)
 876		return false;
 877	for (i = 0; i < c1->group_info->ngroups; i++) {
 878		if (!gid_eq(c1->group_info->gid[i], c2->group_info->gid[i]))
 879			return false;
 880	}
 881	return true;
 882}
 883
 884static struct nfsd_file *
 885nfsd_file_find_locked(struct inode *inode, unsigned int may_flags,
 886			unsigned int hashval, struct net *net)
 887{
 888	struct nfsd_file *nf;
 889	unsigned char need = may_flags & NFSD_FILE_MAY_MASK;
 890
 891	hlist_for_each_entry_rcu(nf, &nfsd_file_hashtbl[hashval].nfb_head,
 892				 nf_node, lockdep_is_held(&nfsd_file_hashtbl[hashval].nfb_lock)) {
 893		if (nf->nf_may != need)
 894			continue;
 895		if (nf->nf_inode != inode)
 896			continue;
 897		if (nf->nf_net != net)
 898			continue;
 899		if (!nfsd_match_cred(nf->nf_cred, current_cred()))
 900			continue;
 901		if (!test_bit(NFSD_FILE_HASHED, &nf->nf_flags))
 902			continue;
 903		if (nfsd_file_get(nf) != NULL)
 904			return nf;
 905	}
 906	return NULL;
 907}
 908
 909/**
 910 * nfsd_file_is_cached - are there any cached open files for this fh?
 911 * @inode: inode of the file to check
 
 
 
 912 *
 913 * Scan the hashtable for open files that match this fh. Returns true if there
 914 * are any, and false if not.
 
 915 */
 916bool
 917nfsd_file_is_cached(struct inode *inode)
 918{
 919	bool			ret = false;
 920	struct nfsd_file	*nf;
 921	unsigned int		hashval;
 922
 923        hashval = (unsigned int)hash_long(inode->i_ino, NFSD_FILE_HASH_BITS);
 924
 925	rcu_read_lock();
 926	hlist_for_each_entry_rcu(nf, &nfsd_file_hashtbl[hashval].nfb_head,
 927				 nf_node) {
 928		if (inode == nf->nf_inode) {
 929			ret = true;
 930			break;
 931		}
 932	}
 933	rcu_read_unlock();
 934	trace_nfsd_file_is_cached(inode, hashval, (int)ret);
 935	return ret;
 936}
 937
 938__be32
 939nfsd_file_acquire(struct svc_rqst *rqstp, struct svc_fh *fhp,
 940		  unsigned int may_flags, struct nfsd_file **pnf)
 941{
 942	__be32	status;
 943	struct net *net = SVC_NET(rqstp);
 944	struct nfsd_file *nf, *new;
 945	struct inode *inode;
 946	unsigned int hashval;
 947	bool retry = true;
 
 
 
 
 
 948
 949	/* FIXME: skip this if fh_dentry is already set? */
 950	status = fh_verify(rqstp, fhp, S_IFREG,
 951				may_flags|NFSD_MAY_OWNER_OVERRIDE);
 952	if (status != nfs_ok)
 953		return status;
 
 
 954
 955	inode = d_inode(fhp->fh_dentry);
 956	hashval = (unsigned int)hash_long(inode->i_ino, NFSD_FILE_HASH_BITS);
 957retry:
 958	rcu_read_lock();
 959	nf = nfsd_file_find_locked(inode, may_flags, hashval, net);
 
 
 
 960	rcu_read_unlock();
 961	if (nf)
 
 
 
 962		goto wait_for_construction;
 
 963
 964	new = nfsd_file_alloc(inode, may_flags, hashval, net);
 965	if (!new) {
 966		trace_nfsd_file_acquire(rqstp, hashval, inode, may_flags,
 967					NULL, nfserr_jukebox);
 968		return nfserr_jukebox;
 969	}
 970
 971	spin_lock(&nfsd_file_hashtbl[hashval].nfb_lock);
 972	nf = nfsd_file_find_locked(inode, may_flags, hashval, net);
 973	if (nf == NULL)
 
 974		goto open_file;
 975	spin_unlock(&nfsd_file_hashtbl[hashval].nfb_lock);
 976	nfsd_file_slab_free(&new->nf_rcu);
 
 
 
 
 
 
 977
 978wait_for_construction:
 979	wait_on_bit(&nf->nf_flags, NFSD_FILE_PENDING, TASK_UNINTERRUPTIBLE);
 980
 981	/* Did construction of this file fail? */
 982	if (!test_bit(NFSD_FILE_HASHED, &nf->nf_flags)) {
 983		if (!retry) {
 
 984			status = nfserr_jukebox;
 985			goto out;
 986		}
 987		retry = false;
 988		nfsd_file_put_noref(nf);
 
 989		goto retry;
 990	}
 991
 992	this_cpu_inc(nfsd_file_cache_hits);
 993
 994	if (!(may_flags & NFSD_MAY_NOT_BREAK_LEASE)) {
 995		bool write = (may_flags & NFSD_MAY_WRITE);
 996
 997		if (test_bit(NFSD_FILE_BREAK_READ, &nf->nf_flags) ||
 998		    (test_bit(NFSD_FILE_BREAK_WRITE, &nf->nf_flags) && write)) {
 999			status = nfserrno(nfsd_open_break_lease(
1000					file_inode(nf->nf_file), may_flags));
1001			if (status == nfs_ok) {
1002				clear_bit(NFSD_FILE_BREAK_READ, &nf->nf_flags);
1003				if (write)
1004					clear_bit(NFSD_FILE_BREAK_WRITE,
1005						  &nf->nf_flags);
1006			}
1007		}
1008	}
1009out:
1010	if (status == nfs_ok) {
 
1011		*pnf = nf;
1012	} else {
1013		nfsd_file_put(nf);
 
1014		nf = NULL;
1015	}
1016
1017	trace_nfsd_file_acquire(rqstp, hashval, inode, may_flags, nf, status);
 
 
1018	return status;
 
1019open_file:
1020	nf = new;
1021	/* Take reference for the hashtable */
1022	refcount_inc(&nf->nf_ref);
1023	__set_bit(NFSD_FILE_HASHED, &nf->nf_flags);
1024	__set_bit(NFSD_FILE_PENDING, &nf->nf_flags);
1025	list_lru_add(&nfsd_file_lru, &nf->nf_lru);
1026	hlist_add_head_rcu(&nf->nf_node, &nfsd_file_hashtbl[hashval].nfb_head);
1027	++nfsd_file_hashtbl[hashval].nfb_count;
1028	nfsd_file_hashtbl[hashval].nfb_maxcount = max(nfsd_file_hashtbl[hashval].nfb_maxcount,
1029			nfsd_file_hashtbl[hashval].nfb_count);
1030	spin_unlock(&nfsd_file_hashtbl[hashval].nfb_lock);
1031	if (atomic_long_inc_return(&nfsd_filecache_count) >= NFSD_FILE_LRU_THRESHOLD)
1032		nfsd_file_gc();
1033
1034	nf->nf_mark = nfsd_file_mark_find_or_create(nf);
1035	if (nf->nf_mark)
1036		status = nfsd_open_verified(rqstp, fhp, S_IFREG,
1037				may_flags, &nf->nf_file);
1038	else
1039		status = nfserr_jukebox;
1040	/*
1041	 * If construction failed, or we raced with a call to unlink()
1042	 * then unhash.
1043	 */
1044	if (status != nfs_ok || inode->i_nlink == 0) {
1045		bool do_free;
1046		spin_lock(&nfsd_file_hashtbl[hashval].nfb_lock);
1047		do_free = nfsd_file_unhash(nf);
1048		spin_unlock(&nfsd_file_hashtbl[hashval].nfb_lock);
1049		if (do_free)
1050			nfsd_file_put_noref(nf);
1051	}
1052	clear_bit_unlock(NFSD_FILE_PENDING, &nf->nf_flags);
1053	smp_mb__after_atomic();
1054	wake_up_bit(&nf->nf_flags, NFSD_FILE_PENDING);
1055	goto out;
1056}
1057
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1058/*
1059 * Note that fields may be added, removed or reordered in the future. Programs
1060 * scraping this file for info should test the labels to ensure they're
1061 * getting the correct field.
1062 */
1063static int nfsd_file_cache_stats_show(struct seq_file *m, void *v)
1064{
1065	unsigned int i, count = 0, longest = 0;
1066	unsigned long hits = 0;
 
 
1067
1068	/*
1069	 * No need for spinlocks here since we're not terribly interested in
1070	 * accuracy. We do take the nfsd_mutex simply to ensure that we
1071	 * don't end up racing with server shutdown
1072	 */
1073	mutex_lock(&nfsd_mutex);
1074	if (nfsd_file_hashtbl) {
1075		for (i = 0; i < NFSD_FILE_HASH_SIZE; i++) {
1076			count += nfsd_file_hashtbl[i].nfb_count;
1077			longest = max(longest, nfsd_file_hashtbl[i].nfb_count);
1078		}
 
 
 
 
 
 
 
1079	}
1080	mutex_unlock(&nfsd_mutex);
1081
1082	for_each_possible_cpu(i)
1083		hits += per_cpu(nfsd_file_cache_hits, i);
 
 
 
 
 
1084
1085	seq_printf(m, "total entries: %u\n", count);
1086	seq_printf(m, "longest chain: %u\n", longest);
 
1087	seq_printf(m, "cache hits:    %lu\n", hits);
 
 
 
 
 
 
 
1088	return 0;
1089}
1090
1091int nfsd_file_cache_stats_open(struct inode *inode, struct file *file)
1092{
1093	return single_open(file, nfsd_file_cache_stats_show, NULL);
1094}