개념 설명 전체 · v6.18.37 / fs/dcache.c

    1 // SPDX-License-Identifier: GPL-2.0-only
    2 /*
    3  * fs/dcache.c
    4  *
    5  * Complete reimplementation
    6  * (C) 1997 Thomas Schoebel-Theuer,
    7  * with heavy changes by Linus Torvalds
    8  */
    9 
   10 /*
   11  * Notes on the allocation strategy:
   12  *
   13  * The dcache is a master of the icache - whenever a dcache entry
   14  * exists, the inode will always exist. "iput()" is done either when
   15  * the dcache entry is deleted or garbage collected.
   16  */
   17 
   18 #include <linux/ratelimit.h>
   19 #include <linux/string.h>
   20 #include <linux/mm.h>
   21 #include <linux/fs.h>
   22 #include <linux/fscrypt.h>
   23 #include <linux/fsnotify.h>
   24 #include <linux/slab.h>
   25 #include <linux/init.h>
   26 #include <linux/hash.h>
   27 #include <linux/cache.h>
   28 #include <linux/export.h>
   29 #include <linux/security.h>
   30 #include <linux/seqlock.h>
   31 #include <linux/memblock.h>
   32 #include <linux/bit_spinlock.h>
   33 #include <linux/rculist_bl.h>
   34 #include <linux/list_lru.h>
   35 #include "internal.h"
   36 #include "mount.h"
   37 
   38 #include <asm/runtime-const.h>
   39 
   40 /*
   41  * Usage:
   42  * dcache->d_inode->i_lock protects:
   43  *   - i_dentry, d_u.d_alias, d_inode of aliases
   44  * dcache_hash_bucket lock protects:
   45  *   - the dcache hash table
   46  * s_roots bl list spinlock protects:
   47  *   - the s_roots list (see __d_drop)
   48  * dentry->d_sb->s_dentry_lru_lock protects:
   49  *   - the dcache lru lists and counters
   50  * d_lock protects:
   51  *   - d_flags
   52  *   - d_name
   53  *   - d_lru
   54  *   - d_count
   55  *   - d_unhashed()
   56  *   - d_parent and d_chilren
   57  *   - childrens' d_sib and d_parent
   58  *   - d_u.d_alias, d_inode
   59  *
   60  * Ordering:
   61  * dentry->d_inode->i_lock
   62  *   dentry->d_lock
   63  *     dentry->d_sb->s_dentry_lru_lock
   64  *     dcache_hash_bucket lock
   65  *     s_roots lock
   66  *
   67  * If there is an ancestor relationship:
   68  * dentry->d_parent->...->d_parent->d_lock
   69  *   ...
   70  *     dentry->d_parent->d_lock
   71  *       dentry->d_lock
   72  *
   73  * If no ancestor relationship:
   74  * arbitrary, since it's serialized on rename_lock
   75  */
   76 static int sysctl_vfs_cache_pressure __read_mostly = 100;
   77 static int sysctl_vfs_cache_pressure_denom __read_mostly = 100;
   78 
   79 unsigned long vfs_pressure_ratio(unsigned long val)
   80 {
   81 	return mult_frac(val, sysctl_vfs_cache_pressure, sysctl_vfs_cache_pressure_denom);
   82 }
   83 EXPORT_SYMBOL_GPL(vfs_pressure_ratio);
   84 
   85 __cacheline_aligned_in_smp DEFINE_SEQLOCK(rename_lock);
   86 
   87 EXPORT_SYMBOL(rename_lock);
   88 
   89 static struct kmem_cache *dentry_cache __ro_after_init;
   90 
   91 const struct qstr empty_name = QSTR_INIT("", 0);
   92 EXPORT_SYMBOL(empty_name);
   93 const struct qstr slash_name = QSTR_INIT("/", 1);
   94 EXPORT_SYMBOL(slash_name);
   95 const struct qstr dotdot_name = QSTR_INIT("..", 2);
   96 EXPORT_SYMBOL(dotdot_name);
   97 
   98 /*
   99  * This is the single most critical data structure when it comes
  100  * to the dcache: the hashtable for lookups. Somebody should try
  101  * to make this good - I've just made it work.
  102  *
  103  * This hash-function tries to avoid losing too many bits of hash
  104  * information, yet avoid using a prime hash-size or similar.
  105  *
  106  * Marking the variables "used" ensures that the compiler doesn't
  107  * optimize them away completely on architectures with runtime
  108  * constant infrastructure, this allows debuggers to see their
  109  * values. But updating these values has no effect on those arches.
  110  */
  111 
  112 static unsigned int d_hash_shift __ro_after_init __used;
  113 
  114 static struct hlist_bl_head *dentry_hashtable __ro_after_init __used;
  115 
  116 static inline struct hlist_bl_head *d_hash(unsigned long hashlen)
  117 {
  118 	return runtime_const_ptr(dentry_hashtable) +
  119 		runtime_const_shift_right_32(hashlen, d_hash_shift);
  120 }
  121 
  122 #define IN_LOOKUP_SHIFT 10
  123 static struct hlist_bl_head in_lookup_hashtable[1 << IN_LOOKUP_SHIFT];
  124 
  125 static inline struct hlist_bl_head *in_lookup_hash(const struct dentry *parent,
  126 					unsigned int hash)
  127 {
  128 	hash += (unsigned long) parent / L1_CACHE_BYTES;
  129 	return in_lookup_hashtable + hash_32(hash, IN_LOOKUP_SHIFT);
  130 }
  131 
  132 struct dentry_stat_t {
  133 	long nr_dentry;
  134 	long nr_unused;
  135 	long age_limit;		/* age in seconds */
  136 	long want_pages;	/* pages requested by system */
  137 	long nr_negative;	/* # of unused negative dentries */
  138 	long dummy;		/* Reserved for future use */
  139 };
  140 
  141 static DEFINE_PER_CPU(long, nr_dentry);
  142 static DEFINE_PER_CPU(long, nr_dentry_unused);
  143 static DEFINE_PER_CPU(long, nr_dentry_negative);
  144 static int dentry_negative_policy;
  145 
  146 #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
  147 /* Statistics gathering. */
  148 static struct dentry_stat_t dentry_stat = {
  149 	.age_limit = 45,
  150 };
  151 
  152 /*
  153  * Here we resort to our own counters instead of using generic per-cpu counters
  154  * for consistency with what the vfs inode code does. We are expected to harvest
  155  * better code and performance by having our own specialized counters.
  156  *
  157  * Please note that the loop is done over all possible CPUs, not over all online
  158  * CPUs. The reason for this is that we don't want to play games with CPUs going
  159  * on and off. If one of them goes off, we will just keep their counters.
  160  *
  161  * glommer: See cffbc8a for details, and if you ever intend to change this,
  162  * please update all vfs counters to match.
  163  */
  164 static long get_nr_dentry(void)
  165 {
  166 	int i;
  167 	long sum = 0;
  168 	for_each_possible_cpu(i)
  169 		sum += per_cpu(nr_dentry, i);
  170 	return sum < 0 ? 0 : sum;
  171 }
  172 
  173 static long get_nr_dentry_unused(void)
  174 {
  175 	int i;
  176 	long sum = 0;
  177 	for_each_possible_cpu(i)
  178 		sum += per_cpu(nr_dentry_unused, i);
  179 	return sum < 0 ? 0 : sum;
  180 }
  181 
  182 static long get_nr_dentry_negative(void)
  183 {
  184 	int i;
  185 	long sum = 0;
  186 
  187 	for_each_possible_cpu(i)
  188 		sum += per_cpu(nr_dentry_negative, i);
  189 	return sum < 0 ? 0 : sum;
  190 }
  191 
  192 static int proc_nr_dentry(const struct ctl_table *table, int write, void *buffer,
  193 			  size_t *lenp, loff_t *ppos)
  194 {
  195 	dentry_stat.nr_dentry = get_nr_dentry();
  196 	dentry_stat.nr_unused = get_nr_dentry_unused();
  197 	dentry_stat.nr_negative = get_nr_dentry_negative();
  198 	return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
  199 }
  200 
  201 static const struct ctl_table fs_dcache_sysctls[] = {
  202 	{
  203 		.procname	= "dentry-state",
  204 		.data		= &dentry_stat,
  205 		.maxlen		= 6*sizeof(long),
  206 		.mode		= 0444,
  207 		.proc_handler	= proc_nr_dentry,
  208 	},
  209 	{
  210 		.procname	= "dentry-negative",
  211 		.data		= &dentry_negative_policy,
  212 		.maxlen		= sizeof(dentry_negative_policy),
  213 		.mode		= 0644,
  214 		.proc_handler	= proc_dointvec_minmax,
  215 		.extra1		= SYSCTL_ZERO,
  216 		.extra2		= SYSCTL_ONE,
  217 	},
  218 };
  219 
  220 static const struct ctl_table vm_dcache_sysctls[] = {
  221 	{
  222 		.procname	= "vfs_cache_pressure",
  223 		.data		= &sysctl_vfs_cache_pressure,
  224 		.maxlen		= sizeof(sysctl_vfs_cache_pressure),
  225 		.mode		= 0644,
  226 		.proc_handler	= proc_dointvec_minmax,
  227 		.extra1		= SYSCTL_ZERO,
  228 	},
  229 	{
  230 		.procname	= "vfs_cache_pressure_denom",
  231 		.data		= &sysctl_vfs_cache_pressure_denom,
  232 		.maxlen		= sizeof(sysctl_vfs_cache_pressure_denom),
  233 		.mode		= 0644,
  234 		.proc_handler	= proc_dointvec_minmax,
  235 		.extra1		= SYSCTL_ONE_HUNDRED,
  236 	},
  237 };
  238 
  239 static int __init init_fs_dcache_sysctls(void)
  240 {
  241 	register_sysctl_init("vm", vm_dcache_sysctls);
  242 	register_sysctl_init("fs", fs_dcache_sysctls);
  243 	return 0;
  244 }
  245 fs_initcall(init_fs_dcache_sysctls);
  246 #endif
  247 
  248 /*
  249  * Compare 2 name strings, return 0 if they match, otherwise non-zero.
  250  * The strings are both count bytes long, and count is non-zero.
  251  */
  252 #ifdef CONFIG_DCACHE_WORD_ACCESS
  253 
  254 #include <asm/word-at-a-time.h>
  255 /*
  256  * NOTE! 'cs' and 'scount' come from a dentry, so it has a
  257  * aligned allocation for this particular component. We don't
  258  * strictly need the load_unaligned_zeropad() safety, but it
  259  * doesn't hurt either.
  260  *
  261  * In contrast, 'ct' and 'tcount' can be from a pathname, and do
  262  * need the careful unaligned handling.
  263  */
  264 static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount)
  265 {
  266 	unsigned long a,b,mask;
  267 
  268 	for (;;) {
  269 		a = read_word_at_a_time(cs);
  270 		b = load_unaligned_zeropad(ct);
  271 		if (tcount < sizeof(unsigned long))
  272 			break;
  273 		if (unlikely(a != b))
  274 			return 1;
  275 		cs += sizeof(unsigned long);
  276 		ct += sizeof(unsigned long);
  277 		tcount -= sizeof(unsigned long);
  278 		if (!tcount)
  279 			return 0;
  280 	}
  281 	mask = bytemask_from_count(tcount);
  282 	return unlikely(!!((a ^ b) & mask));
  283 }
  284 
  285 #else
  286 
  287 static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount)
  288 {
  289 	do {
  290 		if (*cs != *ct)
  291 			return 1;
  292 		cs++;
  293 		ct++;
  294 		tcount--;
  295 	} while (tcount);
  296 	return 0;
  297 }
  298 
  299 #endif
  300 
  301 static inline int dentry_cmp(const struct dentry *dentry, const unsigned char *ct, unsigned tcount)
  302 {
  303 	/*
  304 	 * Be careful about RCU walk racing with rename:
  305 	 * use 'READ_ONCE' to fetch the name pointer.
  306 	 *
  307 	 * NOTE! Even if a rename will mean that the length
  308 	 * was not loaded atomically, we don't care. The
  309 	 * RCU walk will check the sequence count eventually,
  310 	 * and catch it. And we won't overrun the buffer,
  311 	 * because we're reading the name pointer atomically,
  312 	 * and a dentry name is guaranteed to be properly
  313 	 * terminated with a NUL byte.
  314 	 *
  315 	 * End result: even if 'len' is wrong, we'll exit
  316 	 * early because the data cannot match (there can
  317 	 * be no NUL in the ct/tcount data)
  318 	 */
  319 	const unsigned char *cs = READ_ONCE(dentry->d_name.name);
  320 
  321 	return dentry_string_cmp(cs, ct, tcount);
  322 }
  323 
  324 /*
  325  * long names are allocated separately from dentry and never modified.
  326  * Refcounted, freeing is RCU-delayed.  See take_dentry_name_snapshot()
  327  * for the reason why ->count and ->head can't be combined into a union.
  328  * dentry_string_cmp() relies upon ->name[] being word-aligned.
  329  */
  330 struct external_name {
  331 	atomic_t count;
  332 	struct rcu_head head;
  333 	unsigned char name[] __aligned(sizeof(unsigned long));
  334 };
  335 
  336 static inline struct external_name *external_name(struct dentry *dentry)
  337 {
  338 	return container_of(dentry->d_name.name, struct external_name, name[0]);
  339 }
  340 
  341 static void __d_free(struct rcu_head *head)
  342 {
  343 	struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu);
  344 
  345 	kmem_cache_free(dentry_cache, dentry); 
  346 }
  347 
  348 static void __d_free_external(struct rcu_head *head)
  349 {
  350 	struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu);
  351 	kfree(external_name(dentry));
  352 	kmem_cache_free(dentry_cache, dentry);
  353 }
  354 
  355 static inline int dname_external(const struct dentry *dentry)
  356 {
  357 	return dentry->d_name.name != dentry->d_shortname.string;
  358 }
  359 
  360 void take_dentry_name_snapshot(struct name_snapshot *name, struct dentry *dentry)
  361 {
  362 	unsigned seq;
  363 	const unsigned char *s;
  364 
  365 	rcu_read_lock();
  366 retry:
  367 	seq = read_seqcount_begin(&dentry->d_seq);
  368 	s = READ_ONCE(dentry->d_name.name);
  369 	name->name.hash_len = dentry->d_name.hash_len;
  370 	name->name.name = name->inline_name.string;
  371 	if (likely(s == dentry->d_shortname.string)) {
  372 		name->inline_name = dentry->d_shortname;
  373 	} else {
  374 		struct external_name *p;
  375 		p = container_of(s, struct external_name, name[0]);
  376 		// get a valid reference
  377 		if (unlikely(!atomic_inc_not_zero(&p->count)))
  378 			goto retry;
  379 		name->name.name = s;
  380 	}
  381 	if (read_seqcount_retry(&dentry->d_seq, seq)) {
  382 		release_dentry_name_snapshot(name);
  383 		goto retry;
  384 	}
  385 	rcu_read_unlock();
  386 }
  387 EXPORT_SYMBOL(take_dentry_name_snapshot);
  388 
  389 void release_dentry_name_snapshot(struct name_snapshot *name)
  390 {
  391 	if (unlikely(name->name.name != name->inline_name.string)) {
  392 		struct external_name *p;
  393 		p = container_of(name->name.name, struct external_name, name[0]);
  394 		if (unlikely(atomic_dec_and_test(&p->count)))
  395 			kfree_rcu(p, head);
  396 	}
  397 }
  398 EXPORT_SYMBOL(release_dentry_name_snapshot);
  399 
  400 static inline void __d_set_inode_and_type(struct dentry *dentry,
  401 					  struct inode *inode,
  402 					  unsigned type_flags)
  403 {
  404 	unsigned flags;
  405 
  406 	dentry->d_inode = inode;
  407 	flags = READ_ONCE(dentry->d_flags);
  408 	flags &= ~DCACHE_ENTRY_TYPE;
  409 	flags |= type_flags;
  410 	smp_store_release(&dentry->d_flags, flags);
  411 }
  412 
  413 static inline void __d_clear_type_and_inode(struct dentry *dentry)
  414 {
  415 	unsigned flags = READ_ONCE(dentry->d_flags);
  416 
  417 	flags &= ~DCACHE_ENTRY_TYPE;
  418 	WRITE_ONCE(dentry->d_flags, flags);
  419 	dentry->d_inode = NULL;
  420 	/*
  421 	 * The negative counter only tracks dentries on the LRU. Don't inc if
  422 	 * d_lru is on another list.
  423 	 */
  424 	if ((flags & (DCACHE_LRU_LIST|DCACHE_SHRINK_LIST)) == DCACHE_LRU_LIST)
  425 		this_cpu_inc(nr_dentry_negative);
  426 }
  427 
  428 static void dentry_free(struct dentry *dentry)
  429 {
  430 	WARN_ON(!hlist_unhashed(&dentry->d_u.d_alias));
  431 	if (unlikely(dname_external(dentry))) {
  432 		struct external_name *p = external_name(dentry);
  433 		if (likely(atomic_dec_and_test(&p->count))) {
  434 			call_rcu(&dentry->d_u.d_rcu, __d_free_external);
  435 			return;
  436 		}
  437 	}
  438 	/* if dentry was never visible to RCU, immediate free is OK */
  439 	if (dentry->d_flags & DCACHE_NORCU)
  440 		__d_free(&dentry->d_u.d_rcu);
  441 	else
  442 		call_rcu(&dentry->d_u.d_rcu, __d_free);
  443 }
  444 
  445 /*
  446  * Release the dentry's inode, using the filesystem
  447  * d_iput() operation if defined.
  448  */
  449 static void dentry_unlink_inode(struct dentry * dentry)
  450 	__releases(dentry->d_lock)
  451 	__releases(dentry->d_inode->i_lock)
  452 {
  453 	struct inode *inode = dentry->d_inode;
  454 
  455 	raw_write_seqcount_begin(&dentry->d_seq);
  456 	__d_clear_type_and_inode(dentry);
  457 	hlist_del_init(&dentry->d_u.d_alias);
  458 	raw_write_seqcount_end(&dentry->d_seq);
  459 	spin_unlock(&dentry->d_lock);
  460 	spin_unlock(&inode->i_lock);
  461 	if (!inode->i_nlink)
  462 		fsnotify_inoderemove(inode);
  463 	if (dentry->d_op && dentry->d_op->d_iput)
  464 		dentry->d_op->d_iput(dentry, inode);
  465 	else
  466 		iput(inode);
  467 }
  468 
  469 /*
  470  * The DCACHE_LRU_LIST bit is set whenever the 'd_lru' entry
  471  * is in use - which includes both the "real" per-superblock
  472  * LRU list _and_ the DCACHE_SHRINK_LIST use.
  473  *
  474  * The DCACHE_SHRINK_LIST bit is set whenever the dentry is
  475  * on the shrink list (ie not on the superblock LRU list).
  476  *
  477  * The per-cpu "nr_dentry_unused" counters are updated with
  478  * the DCACHE_LRU_LIST bit.
  479  *
  480  * The per-cpu "nr_dentry_negative" counters are only updated
  481  * when deleted from or added to the per-superblock LRU list, not
  482  * from/to the shrink list. That is to avoid an unneeded dec/inc
  483  * pair when moving from LRU to shrink list in select_collect().
  484  *
  485  * These helper functions make sure we always follow the
  486  * rules. d_lock must be held by the caller.
  487  */
  488 #define D_FLAG_VERIFY(dentry,x) WARN_ON_ONCE(((dentry)->d_flags & (DCACHE_LRU_LIST | DCACHE_SHRINK_LIST)) != (x))
  489 static void d_lru_add(struct dentry *dentry)
  490 {
  491 	D_FLAG_VERIFY(dentry, 0);
  492 	dentry->d_flags |= DCACHE_LRU_LIST;
  493 	this_cpu_inc(nr_dentry_unused);
  494 	if (d_is_negative(dentry))
  495 		this_cpu_inc(nr_dentry_negative);
  496 	WARN_ON_ONCE(!list_lru_add_obj(
  497 			&dentry->d_sb->s_dentry_lru, &dentry->d_lru));
  498 }
  499 
  500 static void d_lru_del(struct dentry *dentry)
  501 {
  502 	D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
  503 	dentry->d_flags &= ~DCACHE_LRU_LIST;
  504 	this_cpu_dec(nr_dentry_unused);
  505 	if (d_is_negative(dentry))
  506 		this_cpu_dec(nr_dentry_negative);
  507 	WARN_ON_ONCE(!list_lru_del_obj(
  508 			&dentry->d_sb->s_dentry_lru, &dentry->d_lru));
  509 }
  510 
  511 static void d_shrink_del(struct dentry *dentry)
  512 {
  513 	D_FLAG_VERIFY(dentry, DCACHE_SHRINK_LIST | DCACHE_LRU_LIST);
  514 	list_del_init(&dentry->d_lru);
  515 	dentry->d_flags &= ~(DCACHE_SHRINK_LIST | DCACHE_LRU_LIST);
  516 	this_cpu_dec(nr_dentry_unused);
  517 }
  518 
  519 static void d_shrink_add(struct dentry *dentry, struct list_head *list)
  520 {
  521 	D_FLAG_VERIFY(dentry, 0);
  522 	list_add(&dentry->d_lru, list);
  523 	dentry->d_flags |= DCACHE_SHRINK_LIST | DCACHE_LRU_LIST;
  524 	this_cpu_inc(nr_dentry_unused);
  525 }
  526 
  527 /*
  528  * These can only be called under the global LRU lock, ie during the
  529  * callback for freeing the LRU list. "isolate" removes it from the
  530  * LRU lists entirely, while shrink_move moves it to the indicated
  531  * private list.
  532  */
  533 static void d_lru_isolate(struct list_lru_one *lru, struct dentry *dentry)
  534 {
  535 	D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
  536 	dentry->d_flags &= ~DCACHE_LRU_LIST;
  537 	this_cpu_dec(nr_dentry_unused);
  538 	if (d_is_negative(dentry))
  539 		this_cpu_dec(nr_dentry_negative);
  540 	list_lru_isolate(lru, &dentry->d_lru);
  541 }
  542 
  543 static void d_lru_shrink_move(struct list_lru_one *lru, struct dentry *dentry,
  544 			      struct list_head *list)
  545 {
  546 	D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
  547 	dentry->d_flags |= DCACHE_SHRINK_LIST;
  548 	if (d_is_negative(dentry))
  549 		this_cpu_dec(nr_dentry_negative);
  550 	list_lru_isolate_move(lru, &dentry->d_lru, list);
  551 }
  552 
  553 static void ___d_drop(struct dentry *dentry)
  554 {
  555 	struct hlist_bl_head *b;
  556 	/*
  557 	 * Hashed dentries are normally on the dentry hashtable,
  558 	 * with the exception of those newly allocated by
  559 	 * d_obtain_root, which are always IS_ROOT:
  560 	 */
  561 	if (unlikely(IS_ROOT(dentry)))
  562 		b = &dentry->d_sb->s_roots;
  563 	else
  564 		b = d_hash(dentry->d_name.hash);
  565 
  566 	hlist_bl_lock(b);
  567 	__hlist_bl_del(&dentry->d_hash);
  568 	hlist_bl_unlock(b);
  569 }
  570 
  571 void __d_drop(struct dentry *dentry)
  572 {
  573 	if (!d_unhashed(dentry)) {
  574 		___d_drop(dentry);
  575 		dentry->d_hash.pprev = NULL;
  576 		write_seqcount_invalidate(&dentry->d_seq);
  577 	}
  578 }
  579 EXPORT_SYMBOL(__d_drop);
  580 
  581 /**
  582  * d_drop - drop a dentry
  583  * @dentry: dentry to drop
  584  *
  585  * d_drop() unhashes the entry from the parent dentry hashes, so that it won't
  586  * be found through a VFS lookup any more. Note that this is different from
  587  * deleting the dentry - d_delete will try to mark the dentry negative if
  588  * possible, giving a successful _negative_ lookup, while d_drop will
  589  * just make the cache lookup fail.
  590  *
  591  * d_drop() is used mainly for stuff that wants to invalidate a dentry for some
  592  * reason (NFS timeouts or autofs deletes).
  593  *
  594  * __d_drop requires dentry->d_lock
  595  *
  596  * ___d_drop doesn't mark dentry as "unhashed"
  597  * (dentry->d_hash.pprev will be LIST_POISON2, not NULL).
  598  */
  599 void d_drop(struct dentry *dentry)
  600 {
  601 	spin_lock(&dentry->d_lock);
  602 	__d_drop(dentry);
  603 	spin_unlock(&dentry->d_lock);
  604 }
  605 EXPORT_SYMBOL(d_drop);
  606 
  607 static inline void dentry_unlist(struct dentry *dentry)
  608 {
  609 	struct dentry *next;
  610 	/*
  611 	 * Inform d_walk() and shrink_dentry_list() that we are no longer
  612 	 * attached to the dentry tree
  613 	 */
  614 	dentry->d_flags |= DCACHE_DENTRY_KILLED;
  615 	if (unlikely(hlist_unhashed(&dentry->d_sib)))
  616 		return;
  617 	__hlist_del(&dentry->d_sib);
  618 	/*
  619 	 * Cursors can move around the list of children.  While we'd been
  620 	 * a normal list member, it didn't matter - ->d_sib.next would've
  621 	 * been updated.  However, from now on it won't be and for the
  622 	 * things like d_walk() it might end up with a nasty surprise.
  623 	 * Normally d_walk() doesn't care about cursors moving around -
  624 	 * ->d_lock on parent prevents that and since a cursor has no children
  625 	 * of its own, we get through it without ever unlocking the parent.
  626 	 * There is one exception, though - if we ascend from a child that
  627 	 * gets killed as soon as we unlock it, the next sibling is found
  628 	 * using the value left in its ->d_sib.next.  And if _that_
  629 	 * pointed to a cursor, and cursor got moved (e.g. by lseek())
  630 	 * before d_walk() regains parent->d_lock, we'll end up skipping
  631 	 * everything the cursor had been moved past.
  632 	 *
  633 	 * Solution: make sure that the pointer left behind in ->d_sib.next
  634 	 * points to something that won't be moving around.  I.e. skip the
  635 	 * cursors.
  636 	 */
  637 	while (dentry->d_sib.next) {
  638 		next = hlist_entry(dentry->d_sib.next, struct dentry, d_sib);
  639 		if (likely(!(next->d_flags & DCACHE_DENTRY_CURSOR)))
  640 			break;
  641 		dentry->d_sib.next = next->d_sib.next;
  642 	}
  643 }
  644 
  645 static struct dentry *__dentry_kill(struct dentry *dentry)
  646 {
  647 	struct dentry *parent = NULL;
  648 	bool can_free = true;
  649 
  650 	/*
  651 	 * The dentry is now unrecoverably dead to the world.
  652 	 */
  653 	lockref_mark_dead(&dentry->d_lockref);
  654 
  655 	/*
  656 	 * inform the fs via d_prune that this dentry is about to be
  657 	 * unhashed and destroyed.
  658 	 */
  659 	if (dentry->d_flags & DCACHE_OP_PRUNE)
  660 		dentry->d_op->d_prune(dentry);
  661 
  662 	if (dentry->d_flags & DCACHE_LRU_LIST) {
  663 		if (!(dentry->d_flags & DCACHE_SHRINK_LIST))
  664 			d_lru_del(dentry);
  665 	}
  666 	/* if it was on the hash then remove it */
  667 	__d_drop(dentry);
  668 	if (dentry->d_inode)
  669 		dentry_unlink_inode(dentry);
  670 	else
  671 		spin_unlock(&dentry->d_lock);
  672 	this_cpu_dec(nr_dentry);
  673 	if (dentry->d_op && dentry->d_op->d_release)
  674 		dentry->d_op->d_release(dentry);
  675 
  676 	cond_resched();
  677 	/* now that it's negative, ->d_parent is stable */
  678 	if (!IS_ROOT(dentry)) {
  679 		parent = dentry->d_parent;
  680 		spin_lock(&parent->d_lock);
  681 	}
  682 	spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
  683 	dentry_unlist(dentry);
  684 	if (dentry->d_flags & DCACHE_SHRINK_LIST)
  685 		can_free = false;
  686 	spin_unlock(&dentry->d_lock);
  687 	if (likely(can_free))
  688 		dentry_free(dentry);
  689 	if (parent && --parent->d_lockref.count) {
  690 		spin_unlock(&parent->d_lock);
  691 		return NULL;
  692 	}
  693 	return parent;
  694 }
  695 
  696 /*
  697  * Lock a dentry for feeding it to __dentry_kill().
  698  * Called under rcu_read_lock() and dentry->d_lock; the former
  699  * guarantees that nothing we access will be freed under us.
  700  * Note that dentry is *not* protected from concurrent dentry_kill(),
  701  * d_delete(), etc.
  702  *
  703  * Return false if dentry is busy.  Otherwise, return true and have
  704  * that dentry's inode locked.
  705  */
  706 
  707 static bool lock_for_kill(struct dentry *dentry)
  708 {
  709 	struct inode *inode = dentry->d_inode;
  710 
  711 	if (unlikely(dentry->d_lockref.count))
  712 		return false;
  713 
  714 	if (!inode || likely(spin_trylock(&inode->i_lock)))
  715 		return true;
  716 
  717 	do {
  718 		spin_unlock(&dentry->d_lock);
  719 		spin_lock(&inode->i_lock);
  720 		spin_lock(&dentry->d_lock);
  721 		if (likely(inode == dentry->d_inode))
  722 			break;
  723 		spin_unlock(&inode->i_lock);
  724 		inode = dentry->d_inode;
  725 	} while (inode);
  726 	if (likely(!dentry->d_lockref.count))
  727 		return true;
  728 	if (inode)
  729 		spin_unlock(&inode->i_lock);
  730 	return false;
  731 }
  732 
  733 /*
  734  * Decide if dentry is worth retaining.  Usually this is called with dentry
  735  * locked; if not locked, we are more limited and might not be able to tell
  736  * without a lock.  False in this case means "punt to locked path and recheck".
  737  *
  738  * In case we aren't locked, these predicates are not "stable". However, it is
  739  * sufficient that at some point after we dropped the reference the dentry was
  740  * hashed and the flags had the proper value. Other dentry users may have
  741  * re-gotten a reference to the dentry and change that, but our work is done -
  742  * we can leave the dentry around with a zero refcount.
  743  */
  744 static inline bool retain_dentry(struct dentry *dentry, bool locked)
  745 {
  746 	unsigned int d_flags;
  747 
  748 	smp_rmb();
  749 	d_flags = READ_ONCE(dentry->d_flags);
  750 
  751 	// Unreachable? Nobody would be able to look it up, no point retaining
  752 	if (unlikely(d_unhashed(dentry)))
  753 		return false;
  754 
  755 	// Same if it's disconnected
  756 	if (unlikely(d_flags & DCACHE_DISCONNECTED))
  757 		return false;
  758 
  759 	// ->d_delete() might tell us not to bother, but that requires
  760 	// ->d_lock; can't decide without it
  761 	if (unlikely(d_flags & DCACHE_OP_DELETE)) {
  762 		if (!locked || dentry->d_op->d_delete(dentry))
  763 			return false;
  764 	}
  765 
  766 	// Explicitly told not to bother
  767 	if (unlikely(d_flags & DCACHE_DONTCACHE))
  768 		return false;
  769 
  770 	// At this point it looks like we ought to keep it.  We also might
  771 	// need to do something - put it on LRU if it wasn't there already
  772 	// and mark it referenced if it was on LRU, but not marked yet.
  773 	// Unfortunately, both actions require ->d_lock, so in lockless
  774 	// case we'd have to punt rather than doing those.
  775 	if (unlikely(!(d_flags & DCACHE_LRU_LIST))) {
  776 		if (!locked)
  777 			return false;
  778 		d_lru_add(dentry);
  779 	} else if (unlikely(!(d_flags & DCACHE_REFERENCED))) {
  780 		if (!locked)
  781 			return false;
  782 		dentry->d_flags |= DCACHE_REFERENCED;
  783 	}
  784 	return true;
  785 }
  786 
  787 void d_mark_dontcache(struct inode *inode)
  788 {
  789 	struct dentry *de;
  790 
  791 	spin_lock(&inode->i_lock);
  792 	hlist_for_each_entry(de, &inode->i_dentry, d_u.d_alias) {
  793 		spin_lock(&de->d_lock);
  794 		de->d_flags |= DCACHE_DONTCACHE;
  795 		spin_unlock(&de->d_lock);
  796 	}
  797 	inode->i_state |= I_DONTCACHE;
  798 	spin_unlock(&inode->i_lock);
  799 }
  800 EXPORT_SYMBOL(d_mark_dontcache);
  801 
  802 /*
  803  * Try to do a lockless dput(), and return whether that was successful.
  804  *
  805  * If unsuccessful, we return false, having already taken the dentry lock.
  806  * In that case refcount is guaranteed to be zero and we have already
  807  * decided that it's not worth keeping around.
  808  *
  809  * The caller needs to hold the RCU read lock, so that the dentry is
  810  * guaranteed to stay around even if the refcount goes down to zero!
  811  */
  812 static inline bool fast_dput(struct dentry *dentry)
  813 {
  814 	int ret;
  815 
  816 	/*
  817 	 * try to decrement the lockref optimistically.
  818 	 */
  819 	ret = lockref_put_return(&dentry->d_lockref);
  820 
  821 	/*
  822 	 * If the lockref_put_return() failed due to the lock being held
  823 	 * by somebody else, the fast path has failed. We will need to
  824 	 * get the lock, and then check the count again.
  825 	 */
  826 	if (unlikely(ret < 0)) {
  827 		spin_lock(&dentry->d_lock);
  828 		if (WARN_ON_ONCE(dentry->d_lockref.count <= 0)) {
  829 			spin_unlock(&dentry->d_lock);
  830 			return true;
  831 		}
  832 		dentry->d_lockref.count--;
  833 		goto locked;
  834 	}
  835 
  836 	/*
  837 	 * If we weren't the last ref, we're done.
  838 	 */
  839 	if (ret)
  840 		return true;
  841 
  842 	/*
  843 	 * Can we decide that decrement of refcount is all we needed without
  844 	 * taking the lock?  There's a very common case when it's all we need -
  845 	 * dentry looks like it ought to be retained and there's nothing else
  846 	 * to do.
  847 	 */
  848 	if (retain_dentry(dentry, false))
  849 		return true;
  850 
  851 	/*
  852 	 * Either not worth retaining or we can't tell without the lock.
  853 	 * Get the lock, then.  We've already decremented the refcount to 0,
  854 	 * but we'll need to re-check the situation after getting the lock.
  855 	 */
  856 	spin_lock(&dentry->d_lock);
  857 
  858 	/*
  859 	 * Did somebody else grab a reference to it in the meantime, and
  860 	 * we're no longer the last user after all? Alternatively, somebody
  861 	 * else could have killed it and marked it dead. Either way, we
  862 	 * don't need to do anything else.
  863 	 */
  864 locked:
  865 	if (dentry->d_lockref.count || retain_dentry(dentry, true)) {
  866 		spin_unlock(&dentry->d_lock);
  867 		return true;
  868 	}
  869 	return false;
  870 }
  871 
  872 
  873 /* 
  874  * This is dput
  875  *
  876  * This is complicated by the fact that we do not want to put
  877  * dentries that are no longer on any hash chain on the unused
  878  * list: we'd much rather just get rid of them immediately.
  879  *
  880  * However, that implies that we have to traverse the dentry
  881  * tree upwards to the parents which might _also_ now be
  882  * scheduled for deletion (it may have been only waiting for
  883  * its last child to go away).
  884  *
  885  * This tail recursion is done by hand as we don't want to depend
  886  * on the compiler to always get this right (gcc generally doesn't).
  887  * Real recursion would eat up our stack space.
  888  */
  889 
  890 /*
  891  * dput - release a dentry
  892  * @dentry: dentry to release 
  893  *
  894  * Release a dentry. This will drop the usage count and if appropriate
  895  * call the dentry unlink method as well as removing it from the queues and
  896  * releasing its resources. If the parent dentries were scheduled for release
  897  * they too may now get deleted.
  898  */
  899 void dput(struct dentry *dentry)
  900 {
  901 	if (!dentry)
  902 		return;
  903 	might_sleep();
  904 	rcu_read_lock();
  905 	if (likely(fast_dput(dentry))) {
  906 		rcu_read_unlock();
  907 		return;
  908 	}
  909 	while (lock_for_kill(dentry)) {
  910 		rcu_read_unlock();
  911 		dentry = __dentry_kill(dentry);
  912 		if (!dentry)
  913 			return;
  914 		if (retain_dentry(dentry, true)) {
  915 			spin_unlock(&dentry->d_lock);
  916 			return;
  917 		}
  918 		rcu_read_lock();
  919 	}
  920 	rcu_read_unlock();
  921 	spin_unlock(&dentry->d_lock);
  922 }
  923 EXPORT_SYMBOL(dput);
  924 
  925 static void to_shrink_list(struct dentry *dentry, struct list_head *list)
  926 __must_hold(&dentry->d_lock)
  927 {
  928 	if (!(dentry->d_flags & DCACHE_SHRINK_LIST)) {
  929 		if (dentry->d_flags & DCACHE_LRU_LIST)
  930 			d_lru_del(dentry);
  931 		d_shrink_add(dentry, list);
  932 	}
  933 }
  934 
  935 void dput_to_list(struct dentry *dentry, struct list_head *list)
  936 {
  937 	rcu_read_lock();
  938 	if (likely(fast_dput(dentry))) {
  939 		rcu_read_unlock();
  940 		return;
  941 	}
  942 	rcu_read_unlock();
  943 	to_shrink_list(dentry, list);
  944 	spin_unlock(&dentry->d_lock);
  945 }
  946 
  947 struct dentry *dget_parent(struct dentry *dentry)
  948 {
  949 	int gotref;
  950 	struct dentry *ret;
  951 	unsigned seq;
  952 
  953 	/*
  954 	 * Do optimistic parent lookup without any
  955 	 * locking.
  956 	 */
  957 	rcu_read_lock();
  958 	seq = raw_seqcount_begin(&dentry->d_seq);
  959 	ret = READ_ONCE(dentry->d_parent);
  960 	gotref = lockref_get_not_zero(&ret->d_lockref);
  961 	rcu_read_unlock();
  962 	if (likely(gotref)) {
  963 		if (!read_seqcount_retry(&dentry->d_seq, seq))
  964 			return ret;
  965 		dput(ret);
  966 	}
  967 
  968 repeat:
  969 	/*
  970 	 * Don't need rcu_dereference because we re-check it was correct under
  971 	 * the lock.
  972 	 */
  973 	rcu_read_lock();
  974 	ret = dentry->d_parent;
  975 	spin_lock(&ret->d_lock);
  976 	if (unlikely(ret != dentry->d_parent)) {
  977 		spin_unlock(&ret->d_lock);
  978 		rcu_read_unlock();
  979 		goto repeat;
  980 	}
  981 	rcu_read_unlock();
  982 	BUG_ON(!ret->d_lockref.count);
  983 	ret->d_lockref.count++;
  984 	spin_unlock(&ret->d_lock);
  985 	return ret;
  986 }
  987 EXPORT_SYMBOL(dget_parent);
  988 
  989 static struct dentry * __d_find_any_alias(struct inode *inode)
  990 {
  991 	struct dentry *alias;
  992 
  993 	if (hlist_empty(&inode->i_dentry))
  994 		return NULL;
  995 	alias = hlist_entry(inode->i_dentry.first, struct dentry, d_u.d_alias);
  996 	lockref_get(&alias->d_lockref);
  997 	return alias;
  998 }
  999 
 1000 /**
 1001  * d_find_any_alias - find any alias for a given inode
 1002  * @inode: inode to find an alias for
 1003  *
 1004  * If any aliases exist for the given inode, take and return a
 1005  * reference for one of them.  If no aliases exist, return %NULL.
 1006  */
 1007 struct dentry *d_find_any_alias(struct inode *inode)
 1008 {
 1009 	struct dentry *de;
 1010 
 1011 	spin_lock(&inode->i_lock);
 1012 	de = __d_find_any_alias(inode);
 1013 	spin_unlock(&inode->i_lock);
 1014 	return de;
 1015 }
 1016 EXPORT_SYMBOL(d_find_any_alias);
 1017 
 1018 static struct dentry *__d_find_alias(struct inode *inode)
 1019 {
 1020 	struct dentry *alias;
 1021 
 1022 	if (S_ISDIR(inode->i_mode))
 1023 		return __d_find_any_alias(inode);
 1024 
 1025 	hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
 1026 		spin_lock(&alias->d_lock);
 1027  		if (!d_unhashed(alias)) {
 1028 			dget_dlock(alias);
 1029 			spin_unlock(&alias->d_lock);
 1030 			return alias;
 1031 		}
 1032 		spin_unlock(&alias->d_lock);
 1033 	}
 1034 	return NULL;
 1035 }
 1036 
 1037 /**
 1038  * d_find_alias - grab a hashed alias of inode
 1039  * @inode: inode in question
 1040  *
 1041  * If inode has a hashed alias, or is a directory and has any alias,
 1042  * acquire the reference to alias and return it. Otherwise return NULL.
 1043  * Notice that if inode is a directory there can be only one alias and
 1044  * it can be unhashed only if it has no children, or if it is the root
 1045  * of a filesystem, or if the directory was renamed and d_revalidate
 1046  * was the first vfs operation to notice.
 1047  *
 1048  * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
 1049  * any other hashed alias over that one.
 1050  */
 1051 struct dentry *d_find_alias(struct inode *inode)
 1052 {
 1053 	struct dentry *de = NULL;
 1054 
 1055 	if (!hlist_empty(&inode->i_dentry)) {
 1056 		spin_lock(&inode->i_lock);
 1057 		de = __d_find_alias(inode);
 1058 		spin_unlock(&inode->i_lock);
 1059 	}
 1060 	return de;
 1061 }
 1062 EXPORT_SYMBOL(d_find_alias);
 1063 
 1064 /*
 1065  *  Caller MUST be holding rcu_read_lock() and be guaranteed
 1066  *  that inode won't get freed until rcu_read_unlock().
 1067  */
 1068 struct dentry *d_find_alias_rcu(struct inode *inode)
 1069 {
 1070 	struct hlist_head *l = &inode->i_dentry;
 1071 	struct dentry *de = NULL;
 1072 
 1073 	spin_lock(&inode->i_lock);
 1074 	// ->i_dentry and ->i_rcu are colocated, but the latter won't be
 1075 	// used without having I_FREEING set, which means no aliases left
 1076 	if (likely(!(inode->i_state & I_FREEING) && !hlist_empty(l))) {
 1077 		if (S_ISDIR(inode->i_mode)) {
 1078 			de = hlist_entry(l->first, struct dentry, d_u.d_alias);
 1079 		} else {
 1080 			hlist_for_each_entry(de, l, d_u.d_alias)
 1081 				if (!d_unhashed(de))
 1082 					break;
 1083 		}
 1084 	}
 1085 	spin_unlock(&inode->i_lock);
 1086 	return de;
 1087 }
 1088 
 1089 /*
 1090  *	Try to kill dentries associated with this inode.
 1091  * WARNING: you must own a reference to inode.
 1092  */
 1093 void d_prune_aliases(struct inode *inode)
 1094 {
 1095 	LIST_HEAD(dispose);
 1096 	struct dentry *dentry;
 1097 
 1098 	spin_lock(&inode->i_lock);
 1099 	hlist_for_each_entry(dentry, &inode->i_dentry, d_u.d_alias) {
 1100 		spin_lock(&dentry->d_lock);
 1101 		if (!dentry->d_lockref.count)
 1102 			to_shrink_list(dentry, &dispose);
 1103 		spin_unlock(&dentry->d_lock);
 1104 	}
 1105 	spin_unlock(&inode->i_lock);
 1106 	shrink_dentry_list(&dispose);
 1107 }
 1108 EXPORT_SYMBOL(d_prune_aliases);
 1109 
 1110 static inline void shrink_kill(struct dentry *victim)
 1111 {
 1112 	do {
 1113 		rcu_read_unlock();
 1114 		victim = __dentry_kill(victim);
 1115 		rcu_read_lock();
 1116 	} while (victim && lock_for_kill(victim));
 1117 	rcu_read_unlock();
 1118 	if (victim)
 1119 		spin_unlock(&victim->d_lock);
 1120 }
 1121 
 1122 void shrink_dentry_list(struct list_head *list)
 1123 {
 1124 	while (!list_empty(list)) {
 1125 		struct dentry *dentry;
 1126 
 1127 		dentry = list_entry(list->prev, struct dentry, d_lru);
 1128 		spin_lock(&dentry->d_lock);
 1129 		rcu_read_lock();
 1130 		if (!lock_for_kill(dentry)) {
 1131 			bool can_free;
 1132 			rcu_read_unlock();
 1133 			d_shrink_del(dentry);
 1134 			can_free = dentry->d_flags & DCACHE_DENTRY_KILLED;
 1135 			spin_unlock(&dentry->d_lock);
 1136 			if (can_free)
 1137 				dentry_free(dentry);
 1138 			continue;
 1139 		}
 1140 		d_shrink_del(dentry);
 1141 		shrink_kill(dentry);
 1142 	}
 1143 }
 1144 
 1145 static enum lru_status dentry_lru_isolate(struct list_head *item,
 1146 		struct list_lru_one *lru, void *arg)
 1147 {
 1148 	struct list_head *freeable = arg;
 1149 	struct dentry	*dentry = container_of(item, struct dentry, d_lru);
 1150 
 1151 
 1152 	/*
 1153 	 * we are inverting the lru lock/dentry->d_lock here,
 1154 	 * so use a trylock. If we fail to get the lock, just skip
 1155 	 * it
 1156 	 */
 1157 	if (!spin_trylock(&dentry->d_lock))
 1158 		return LRU_SKIP;
 1159 
 1160 	/*
 1161 	 * Referenced dentries are still in use. If they have active
 1162 	 * counts, just remove them from the LRU. Otherwise give them
 1163 	 * another pass through the LRU.
 1164 	 */
 1165 	if (dentry->d_lockref.count) {
 1166 		d_lru_isolate(lru, dentry);
 1167 		spin_unlock(&dentry->d_lock);
 1168 		return LRU_REMOVED;
 1169 	}
 1170 
 1171 	if (dentry->d_flags & DCACHE_REFERENCED) {
 1172 		dentry->d_flags &= ~DCACHE_REFERENCED;
 1173 		spin_unlock(&dentry->d_lock);
 1174 
 1175 		/*
 1176 		 * The list move itself will be made by the common LRU code. At
 1177 		 * this point, we've dropped the dentry->d_lock but keep the
 1178 		 * lru lock. This is safe to do, since every list movement is
 1179 		 * protected by the lru lock even if both locks are held.
 1180 		 *
 1181 		 * This is guaranteed by the fact that all LRU management
 1182 		 * functions are intermediated by the LRU API calls like
 1183 		 * list_lru_add_obj and list_lru_del_obj. List movement in this file
 1184 		 * only ever occur through this functions or through callbacks
 1185 		 * like this one, that are called from the LRU API.
 1186 		 *
 1187 		 * The only exceptions to this are functions like
 1188 		 * shrink_dentry_list, and code that first checks for the
 1189 		 * DCACHE_SHRINK_LIST flag.  Those are guaranteed to be
 1190 		 * operating only with stack provided lists after they are
 1191 		 * properly isolated from the main list.  It is thus, always a
 1192 		 * local access.
 1193 		 */
 1194 		return LRU_ROTATE;
 1195 	}
 1196 
 1197 	d_lru_shrink_move(lru, dentry, freeable);
 1198 	spin_unlock(&dentry->d_lock);
 1199 
 1200 	return LRU_REMOVED;
 1201 }
 1202 
 1203 /**
 1204  * prune_dcache_sb - shrink the dcache
 1205  * @sb: superblock
 1206  * @sc: shrink control, passed to list_lru_shrink_walk()
 1207  *
 1208  * Attempt to shrink the superblock dcache LRU by @sc->nr_to_scan entries. This
 1209  * is done when we need more memory and called from the superblock shrinker
 1210  * function.
 1211  *
 1212  * This function may fail to free any resources if all the dentries are in
 1213  * use.
 1214  */
 1215 long prune_dcache_sb(struct super_block *sb, struct shrink_control *sc)
 1216 {
 1217 	LIST_HEAD(dispose);
 1218 	long freed;
 1219 
 1220 	freed = list_lru_shrink_walk(&sb->s_dentry_lru, sc,
 1221 				     dentry_lru_isolate, &dispose);
 1222 	shrink_dentry_list(&dispose);
 1223 	return freed;
 1224 }
 1225 
 1226 static enum lru_status dentry_lru_isolate_shrink(struct list_head *item,
 1227 		struct list_lru_one *lru, void *arg)
 1228 {
 1229 	struct list_head *freeable = arg;
 1230 	struct dentry	*dentry = container_of(item, struct dentry, d_lru);
 1231 
 1232 	/*
 1233 	 * we are inverting the lru lock/dentry->d_lock here,
 1234 	 * so use a trylock. If we fail to get the lock, just skip
 1235 	 * it
 1236 	 */
 1237 	if (!spin_trylock(&dentry->d_lock))
 1238 		return LRU_SKIP;
 1239 
 1240 	d_lru_shrink_move(lru, dentry, freeable);
 1241 	spin_unlock(&dentry->d_lock);
 1242 
 1243 	return LRU_REMOVED;
 1244 }
 1245 
 1246 
 1247 /**
 1248  * shrink_dcache_sb - shrink dcache for a superblock
 1249  * @sb: superblock
 1250  *
 1251  * Shrink the dcache for the specified super block. This is used to free
 1252  * the dcache before unmounting a file system.
 1253  */
 1254 void shrink_dcache_sb(struct super_block *sb)
 1255 {
 1256 	do {
 1257 		LIST_HEAD(dispose);
 1258 
 1259 		list_lru_walk(&sb->s_dentry_lru,
 1260 			dentry_lru_isolate_shrink, &dispose, 1024);
 1261 		shrink_dentry_list(&dispose);
 1262 	} while (list_lru_count(&sb->s_dentry_lru) > 0);
 1263 }
 1264 EXPORT_SYMBOL(shrink_dcache_sb);
 1265 
 1266 /**
 1267  * enum d_walk_ret - action to talke during tree walk
 1268  * @D_WALK_CONTINUE:	contrinue walk
 1269  * @D_WALK_QUIT:	quit walk
 1270  * @D_WALK_NORETRY:	quit when retry is needed
 1271  * @D_WALK_SKIP:	skip this dentry and its children
 1272  */
 1273 enum d_walk_ret {
 1274 	D_WALK_CONTINUE,
 1275 	D_WALK_QUIT,
 1276 	D_WALK_NORETRY,
 1277 	D_WALK_SKIP,
 1278 };
 1279 
 1280 /**
 1281  * d_walk - walk the dentry tree
 1282  * @parent:	start of walk
 1283  * @data:	data passed to @enter() and @finish()
 1284  * @enter:	callback when first entering the dentry
 1285  *
 1286  * The @enter() callbacks are called with d_lock held.
 1287  */
 1288 static void d_walk(struct dentry *parent, void *data,
 1289 		   enum d_walk_ret (*enter)(void *, struct dentry *))
 1290 {
 1291 	struct dentry *this_parent, *dentry;
 1292 	unsigned seq = 0;
 1293 	enum d_walk_ret ret;
 1294 	bool retry = true;
 1295 
 1296 again:
 1297 	read_seqbegin_or_lock(&rename_lock, &seq);
 1298 	this_parent = parent;
 1299 	spin_lock(&this_parent->d_lock);
 1300 
 1301 	ret = enter(data, this_parent);
 1302 	switch (ret) {
 1303 	case D_WALK_CONTINUE:
 1304 		break;
 1305 	case D_WALK_QUIT:
 1306 	case D_WALK_SKIP:
 1307 		goto out_unlock;
 1308 	case D_WALK_NORETRY:
 1309 		retry = false;
 1310 		break;
 1311 	}
 1312 repeat:
 1313 	dentry = d_first_child(this_parent);
 1314 resume:
 1315 	hlist_for_each_entry_from(dentry, d_sib) {
 1316 		if (unlikely(dentry->d_flags & DCACHE_DENTRY_CURSOR))
 1317 			continue;
 1318 
 1319 		spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
 1320 
 1321 		ret = enter(data, dentry);
 1322 		switch (ret) {
 1323 		case D_WALK_CONTINUE:
 1324 			break;
 1325 		case D_WALK_QUIT:
 1326 			spin_unlock(&dentry->d_lock);
 1327 			goto out_unlock;
 1328 		case D_WALK_NORETRY:
 1329 			retry = false;
 1330 			break;
 1331 		case D_WALK_SKIP:
 1332 			spin_unlock(&dentry->d_lock);
 1333 			continue;
 1334 		}
 1335 
 1336 		if (!hlist_empty(&dentry->d_children)) {
 1337 			spin_unlock(&this_parent->d_lock);
 1338 			spin_release(&dentry->d_lock.dep_map, _RET_IP_);
 1339 			this_parent = dentry;
 1340 			spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
 1341 			goto repeat;
 1342 		}
 1343 		spin_unlock(&dentry->d_lock);
 1344 	}
 1345 	/*
 1346 	 * All done at this level ... ascend and resume the search.
 1347 	 */
 1348 	rcu_read_lock();
 1349 ascend:
 1350 	if (this_parent != parent) {
 1351 		dentry = this_parent;
 1352 		this_parent = dentry->d_parent;
 1353 
 1354 		spin_unlock(&dentry->d_lock);
 1355 		spin_lock(&this_parent->d_lock);
 1356 
 1357 		/* might go back up the wrong parent if we have had a rename. */
 1358 		if (need_seqretry(&rename_lock, seq))
 1359 			goto rename_retry;
 1360 		/* go into the first sibling still alive */
 1361 		hlist_for_each_entry_continue(dentry, d_sib) {
 1362 			if (likely(!(dentry->d_flags & DCACHE_DENTRY_KILLED))) {
 1363 				rcu_read_unlock();
 1364 				goto resume;
 1365 			}
 1366 		}
 1367 		goto ascend;
 1368 	}
 1369 	if (need_seqretry(&rename_lock, seq))
 1370 		goto rename_retry;
 1371 	rcu_read_unlock();
 1372 
 1373 out_unlock:
 1374 	spin_unlock(&this_parent->d_lock);
 1375 	done_seqretry(&rename_lock, seq);
 1376 	return;
 1377 
 1378 rename_retry:
 1379 	spin_unlock(&this_parent->d_lock);
 1380 	rcu_read_unlock();
 1381 	BUG_ON(seq & 1);
 1382 	if (!retry)
 1383 		return;
 1384 	seq = 1;
 1385 	goto again;
 1386 }
 1387 
 1388 struct check_mount {
 1389 	struct vfsmount *mnt;
 1390 	unsigned int mounted;
 1391 };
 1392 
 1393 /* locks: mount_locked_reader && dentry->d_lock */
 1394 static enum d_walk_ret path_check_mount(void *data, struct dentry *dentry)
 1395 {
 1396 	struct check_mount *info = data;
 1397 	struct path path = { .mnt = info->mnt, .dentry = dentry };
 1398 
 1399 	if (likely(!d_mountpoint(dentry)))
 1400 		return D_WALK_CONTINUE;
 1401 	if (__path_is_mountpoint(&path)) {
 1402 		info->mounted = 1;
 1403 		return D_WALK_QUIT;
 1404 	}
 1405 	return D_WALK_CONTINUE;
 1406 }
 1407 
 1408 /**
 1409  * path_has_submounts - check for mounts over a dentry in the
 1410  *                      current namespace.
 1411  * @parent: path to check.
 1412  *
 1413  * Return true if the parent or its subdirectories contain
 1414  * a mount point in the current namespace.
 1415  */
 1416 int path_has_submounts(const struct path *parent)
 1417 {
 1418 	struct check_mount data = { .mnt = parent->mnt, .mounted = 0 };
 1419 
 1420 	guard(mount_locked_reader)();
 1421 	d_walk(parent->dentry, &data, path_check_mount);
 1422 
 1423 	return data.mounted;
 1424 }
 1425 EXPORT_SYMBOL(path_has_submounts);
 1426 
 1427 /*
 1428  * Called by mount code to set a mountpoint and check if the mountpoint is
 1429  * reachable (e.g. NFS can unhash a directory dentry and then the complete
 1430  * subtree can become unreachable).
 1431  *
 1432  * Only one of d_invalidate() and d_set_mounted() must succeed.  For
 1433  * this reason take rename_lock and d_lock on dentry and ancestors.
 1434  */
 1435 int d_set_mounted(struct dentry *dentry)
 1436 {
 1437 	struct dentry *p;
 1438 	int ret = -ENOENT;
 1439 	read_seqlock_excl(&rename_lock);
 1440 	for (p = dentry->d_parent; !IS_ROOT(p); p = p->d_parent) {
 1441 		/* Need exclusion wrt. d_invalidate() */
 1442 		spin_lock(&p->d_lock);
 1443 		if (unlikely(d_unhashed(p))) {
 1444 			spin_unlock(&p->d_lock);
 1445 			goto out;
 1446 		}
 1447 		spin_unlock(&p->d_lock);
 1448 	}
 1449 	spin_lock(&dentry->d_lock);
 1450 	if (!d_unlinked(dentry)) {
 1451 		ret = -EBUSY;
 1452 		if (!d_mountpoint(dentry)) {
 1453 			dentry->d_flags |= DCACHE_MOUNTED;
 1454 			ret = 0;
 1455 		}
 1456 	}
 1457  	spin_unlock(&dentry->d_lock);
 1458 out:
 1459 	read_sequnlock_excl(&rename_lock);
 1460 	return ret;
 1461 }
 1462 
 1463 /*
 1464  * Search the dentry child list of the specified parent,
 1465  * and move any unused dentries to the end of the unused
 1466  * list for prune_dcache(). We descend to the next level
 1467  * whenever the d_children list is non-empty and continue
 1468  * searching.
 1469  *
 1470  * It returns zero iff there are no unused children,
 1471  * otherwise  it returns the number of children moved to
 1472  * the end of the unused list. This may not be the total
 1473  * number of unused children, because select_parent can
 1474  * drop the lock and return early due to latency
 1475  * constraints.
 1476  */
 1477 
 1478 struct select_data {
 1479 	struct dentry *start;
 1480 	union {
 1481 		long found;
 1482 		struct dentry *victim;
 1483 	};
 1484 	struct list_head dispose;
 1485 };
 1486 
 1487 static enum d_walk_ret select_collect(void *_data, struct dentry *dentry)
 1488 {
 1489 	struct select_data *data = _data;
 1490 	enum d_walk_ret ret = D_WALK_CONTINUE;
 1491 
 1492 	if (data->start == dentry)
 1493 		goto out;
 1494 
 1495 	if (dentry->d_flags & DCACHE_SHRINK_LIST) {
 1496 		data->found++;
 1497 	} else if (!dentry->d_lockref.count) {
 1498 		to_shrink_list(dentry, &data->dispose);
 1499 		data->found++;
 1500 	} else if (dentry->d_lockref.count < 0) {
 1501 		data->found++;
 1502 	}
 1503 	/*
 1504 	 * We can return to the caller if we have found some (this
 1505 	 * ensures forward progress). We'll be coming back to find
 1506 	 * the rest.
 1507 	 */
 1508 	if (!list_empty(&data->dispose))
 1509 		ret = need_resched() ? D_WALK_QUIT : D_WALK_NORETRY;
 1510 out:
 1511 	return ret;
 1512 }
 1513 
 1514 static enum d_walk_ret select_collect2(void *_data, struct dentry *dentry)
 1515 {
 1516 	struct select_data *data = _data;
 1517 	enum d_walk_ret ret = D_WALK_CONTINUE;
 1518 
 1519 	if (data->start == dentry)
 1520 		goto out;
 1521 
 1522 	if (!dentry->d_lockref.count) {
 1523 		if (dentry->d_flags & DCACHE_SHRINK_LIST) {
 1524 			rcu_read_lock();
 1525 			data->victim = dentry;
 1526 			return D_WALK_QUIT;
 1527 		}
 1528 		to_shrink_list(dentry, &data->dispose);
 1529 	}
 1530 	/*
 1531 	 * We can return to the caller if we have found some (this
 1532 	 * ensures forward progress). We'll be coming back to find
 1533 	 * the rest.
 1534 	 */
 1535 	if (!list_empty(&data->dispose))
 1536 		ret = need_resched() ? D_WALK_QUIT : D_WALK_NORETRY;
 1537 out:
 1538 	return ret;
 1539 }
 1540 
 1541 /**
 1542  * shrink_dcache_parent - prune dcache
 1543  * @parent: parent of entries to prune
 1544  *
 1545  * Prune the dcache to remove unused children of the parent dentry.
 1546  */
 1547 void shrink_dcache_parent(struct dentry *parent)
 1548 {
 1549 	for (;;) {
 1550 		struct select_data data = {.start = parent};
 1551 
 1552 		INIT_LIST_HEAD(&data.dispose);
 1553 		d_walk(parent, &data, select_collect);
 1554 
 1555 		if (!list_empty(&data.dispose)) {
 1556 			shrink_dentry_list(&data.dispose);
 1557 			continue;
 1558 		}
 1559 
 1560 		cond_resched();
 1561 		if (!data.found)
 1562 			break;
 1563 		data.victim = NULL;
 1564 		d_walk(parent, &data, select_collect2);
 1565 		if (data.victim) {
 1566 			spin_lock(&data.victim->d_lock);
 1567 			if (!lock_for_kill(data.victim)) {
 1568 				spin_unlock(&data.victim->d_lock);
 1569 				rcu_read_unlock();
 1570 			} else {
 1571 				shrink_kill(data.victim);
 1572 			}
 1573 		}
 1574 		if (!list_empty(&data.dispose))
 1575 			shrink_dentry_list(&data.dispose);
 1576 	}
 1577 }
 1578 EXPORT_SYMBOL(shrink_dcache_parent);
 1579 
 1580 static enum d_walk_ret umount_check(void *_data, struct dentry *dentry)
 1581 {
 1582 	/* it has busy descendents; complain about those instead */
 1583 	if (!hlist_empty(&dentry->d_children))
 1584 		return D_WALK_CONTINUE;
 1585 
 1586 	/* root with refcount 1 is fine */
 1587 	if (dentry == _data && dentry->d_lockref.count == 1)
 1588 		return D_WALK_CONTINUE;
 1589 
 1590 	WARN(1, "BUG: Dentry %p{i=%lx,n=%pd} "
 1591 			" still in use (%d) [unmount of %s %s]\n",
 1592 		       dentry,
 1593 		       dentry->d_inode ?
 1594 		       dentry->d_inode->i_ino : 0UL,
 1595 		       dentry,
 1596 		       dentry->d_lockref.count,
 1597 		       dentry->d_sb->s_type->name,
 1598 		       dentry->d_sb->s_id);
 1599 	return D_WALK_CONTINUE;
 1600 }
 1601 
 1602 static void do_one_tree(struct dentry *dentry)
 1603 {
 1604 	shrink_dcache_parent(dentry);
 1605 	d_walk(dentry, dentry, umount_check);
 1606 	d_drop(dentry);
 1607 	dput(dentry);
 1608 }
 1609 
 1610 /*
 1611  * destroy the dentries attached to a superblock on unmounting
 1612  */
 1613 void shrink_dcache_for_umount(struct super_block *sb)
 1614 {
 1615 	struct dentry *dentry;
 1616 
 1617 	rwsem_assert_held_write(&sb->s_umount);
 1618 
 1619 	dentry = sb->s_root;
 1620 	sb->s_root = NULL;
 1621 	do_one_tree(dentry);
 1622 
 1623 	while (!hlist_bl_empty(&sb->s_roots)) {
 1624 		dentry = dget(hlist_bl_entry(hlist_bl_first(&sb->s_roots), struct dentry, d_hash));
 1625 		do_one_tree(dentry);
 1626 	}
 1627 }
 1628 
 1629 static enum d_walk_ret find_submount(void *_data, struct dentry *dentry)
 1630 {
 1631 	struct dentry **victim = _data;
 1632 	if (d_mountpoint(dentry)) {
 1633 		*victim = dget_dlock(dentry);
 1634 		return D_WALK_QUIT;
 1635 	}
 1636 	return D_WALK_CONTINUE;
 1637 }
 1638 
 1639 /**
 1640  * d_invalidate - detach submounts, prune dcache, and drop
 1641  * @dentry: dentry to invalidate (aka detach, prune and drop)
 1642  */
 1643 void d_invalidate(struct dentry *dentry)
 1644 {
 1645 	bool had_submounts = false;
 1646 	spin_lock(&dentry->d_lock);
 1647 	if (d_unhashed(dentry)) {
 1648 		spin_unlock(&dentry->d_lock);
 1649 		return;
 1650 	}
 1651 	__d_drop(dentry);
 1652 	spin_unlock(&dentry->d_lock);
 1653 
 1654 	/* Negative dentries can be dropped without further checks */
 1655 	if (!dentry->d_inode)
 1656 		return;
 1657 
 1658 	shrink_dcache_parent(dentry);
 1659 	for (;;) {
 1660 		struct dentry *victim = NULL;
 1661 		d_walk(dentry, &victim, find_submount);
 1662 		if (!victim) {
 1663 			if (had_submounts)
 1664 				shrink_dcache_parent(dentry);
 1665 			return;
 1666 		}
 1667 		had_submounts = true;
 1668 		detach_mounts(victim);
 1669 		dput(victim);
 1670 	}
 1671 }
 1672 EXPORT_SYMBOL(d_invalidate);
 1673 
 1674 /**
 1675  * __d_alloc	-	allocate a dcache entry
 1676  * @sb: filesystem it will belong to
 1677  * @name: qstr of the name
 1678  *
 1679  * Allocates a dentry. It returns %NULL if there is insufficient memory
 1680  * available. On a success the dentry is returned. The name passed in is
 1681  * copied and the copy passed in may be reused after this call.
 1682  */
 1683  
 1684 static struct dentry *__d_alloc(struct super_block *sb, const struct qstr *name)
 1685 {
 1686 	struct dentry *dentry;
 1687 	char *dname;
 1688 	int err;
 1689 
 1690 	dentry = kmem_cache_alloc_lru(dentry_cache, &sb->s_dentry_lru,
 1691 				      GFP_KERNEL);
 1692 	if (!dentry)
 1693 		return NULL;
 1694 
 1695 	/*
 1696 	 * We guarantee that the inline name is always NUL-terminated.
 1697 	 * This way the memcpy() done by the name switching in rename
 1698 	 * will still always have a NUL at the end, even if we might
 1699 	 * be overwriting an internal NUL character
 1700 	 */
 1701 	dentry->d_shortname.string[DNAME_INLINE_LEN-1] = 0;
 1702 	if (unlikely(!name)) {
 1703 		name = &slash_name;
 1704 		dname = dentry->d_shortname.string;
 1705 	} else if (name->len > DNAME_INLINE_LEN-1) {
 1706 		size_t size = offsetof(struct external_name, name[1]);
 1707 		struct external_name *p = kmalloc(size + name->len,
 1708 						  GFP_KERNEL_ACCOUNT |
 1709 						  __GFP_RECLAIMABLE);
 1710 		if (!p) {
 1711 			kmem_cache_free(dentry_cache, dentry); 
 1712 			return NULL;
 1713 		}
 1714 		atomic_set(&p->count, 1);
 1715 		dname = p->name;
 1716 	} else  {
 1717 		dname = dentry->d_shortname.string;
 1718 	}	
 1719 
 1720 	dentry->__d_name.len = name->len;
 1721 	dentry->__d_name.hash = name->hash;
 1722 	memcpy(dname, name->name, name->len);
 1723 	dname[name->len] = 0;
 1724 
 1725 	/* Make sure we always see the terminating NUL character */
 1726 	smp_store_release(&dentry->__d_name.name, dname); /* ^^^ */
 1727 
 1728 	dentry->d_flags = 0;
 1729 	lockref_init(&dentry->d_lockref);
 1730 	seqcount_spinlock_init(&dentry->d_seq, &dentry->d_lock);
 1731 	dentry->d_inode = NULL;
 1732 	dentry->d_parent = dentry;
 1733 	dentry->d_sb = sb;
 1734 	dentry->d_op = sb->__s_d_op;
 1735 	dentry->d_flags = sb->s_d_flags;
 1736 	dentry->d_fsdata = NULL;
 1737 	INIT_HLIST_BL_NODE(&dentry->d_hash);
 1738 	INIT_LIST_HEAD(&dentry->d_lru);
 1739 	INIT_HLIST_HEAD(&dentry->d_children);
 1740 	INIT_HLIST_NODE(&dentry->d_u.d_alias);
 1741 	INIT_HLIST_NODE(&dentry->d_sib);
 1742 
 1743 	if (dentry->d_op && dentry->d_op->d_init) {
 1744 		err = dentry->d_op->d_init(dentry);
 1745 		if (err) {
 1746 			if (dname_external(dentry))
 1747 				kfree(external_name(dentry));
 1748 			kmem_cache_free(dentry_cache, dentry);
 1749 			return NULL;
 1750 		}
 1751 	}
 1752 
 1753 	this_cpu_inc(nr_dentry);
 1754 
 1755 	return dentry;
 1756 }
 1757 
 1758 /**
 1759  * d_alloc	-	allocate a dcache entry
 1760  * @parent: parent of entry to allocate
 1761  * @name: qstr of the name
 1762  *
 1763  * Allocates a dentry. It returns %NULL if there is insufficient memory
 1764  * available. On a success the dentry is returned. The name passed in is
 1765  * copied and the copy passed in may be reused after this call.
 1766  */
 1767 struct dentry *d_alloc(struct dentry * parent, const struct qstr *name)
 1768 {
 1769 	struct dentry *dentry = __d_alloc(parent->d_sb, name);
 1770 	if (!dentry)
 1771 		return NULL;
 1772 	spin_lock(&parent->d_lock);
 1773 	/*
 1774 	 * don't need child lock because it is not subject
 1775 	 * to concurrency here
 1776 	 */
 1777 	dentry->d_parent = dget_dlock(parent);
 1778 	hlist_add_head(&dentry->d_sib, &parent->d_children);
 1779 	spin_unlock(&parent->d_lock);
 1780 
 1781 	return dentry;
 1782 }
 1783 EXPORT_SYMBOL(d_alloc);
 1784 
 1785 struct dentry *d_alloc_anon(struct super_block *sb)
 1786 {
 1787 	return __d_alloc(sb, NULL);
 1788 }
 1789 EXPORT_SYMBOL(d_alloc_anon);
 1790 
 1791 struct dentry *d_alloc_cursor(struct dentry * parent)
 1792 {
 1793 	struct dentry *dentry = d_alloc_anon(parent->d_sb);
 1794 	if (dentry) {
 1795 		dentry->d_flags |= DCACHE_DENTRY_CURSOR;
 1796 		dentry->d_parent = dget(parent);
 1797 	}
 1798 	return dentry;
 1799 }
 1800 
 1801 /**
 1802  * d_alloc_pseudo - allocate a dentry (for lookup-less filesystems)
 1803  * @sb: the superblock
 1804  * @name: qstr of the name
 1805  *
 1806  * For a filesystem that just pins its dentries in memory and never
 1807  * performs lookups at all, return an unhashed IS_ROOT dentry.
 1808  * This is used for pipes, sockets et.al. - the stuff that should
 1809  * never be anyone's children or parents.  Unlike all other
 1810  * dentries, these will not have RCU delay between dropping the
 1811  * last reference and freeing them.
 1812  *
 1813  * The only user is alloc_file_pseudo() and that's what should
 1814  * be considered a public interface.  Don't use directly.
 1815  */
 1816 struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name)
 1817 {
 1818 	static const struct dentry_operations anon_ops = {
 1819 		.d_dname = simple_dname
 1820 	};
 1821 	struct dentry *dentry = __d_alloc(sb, name);
 1822 	if (likely(dentry)) {
 1823 		dentry->d_flags |= DCACHE_NORCU;
 1824 		/* d_op_flags(&anon_ops) is 0 */
 1825 		if (!dentry->d_op)
 1826 			dentry->d_op = &anon_ops;
 1827 	}
 1828 	return dentry;
 1829 }
 1830 
 1831 struct dentry *d_alloc_name(struct dentry *parent, const char *name)
 1832 {
 1833 	struct qstr q;
 1834 
 1835 	q.name = name;
 1836 	q.hash_len = hashlen_string(parent, name);
 1837 	return d_alloc(parent, &q);
 1838 }
 1839 EXPORT_SYMBOL(d_alloc_name);
 1840 
 1841 #define DCACHE_OP_FLAGS \
 1842 	(DCACHE_OP_HASH | DCACHE_OP_COMPARE | DCACHE_OP_REVALIDATE | \
 1843 	 DCACHE_OP_WEAK_REVALIDATE | DCACHE_OP_DELETE | DCACHE_OP_PRUNE | \
 1844 	 DCACHE_OP_REAL)
 1845 
 1846 static unsigned int d_op_flags(const struct dentry_operations *op)
 1847 {
 1848 	unsigned int flags = 0;
 1849 	if (op) {
 1850 		if (op->d_hash)
 1851 			flags |= DCACHE_OP_HASH;
 1852 		if (op->d_compare)
 1853 			flags |= DCACHE_OP_COMPARE;
 1854 		if (op->d_revalidate)
 1855 			flags |= DCACHE_OP_REVALIDATE;
 1856 		if (op->d_weak_revalidate)
 1857 			flags |= DCACHE_OP_WEAK_REVALIDATE;
 1858 		if (op->d_delete)
 1859 			flags |= DCACHE_OP_DELETE;
 1860 		if (op->d_prune)
 1861 			flags |= DCACHE_OP_PRUNE;
 1862 		if (op->d_real)
 1863 			flags |= DCACHE_OP_REAL;
 1864 	}
 1865 	return flags;
 1866 }
 1867 
 1868 static void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op)
 1869 {
 1870 	unsigned int flags = d_op_flags(op);
 1871 	WARN_ON_ONCE(dentry->d_op);
 1872 	WARN_ON_ONCE(dentry->d_flags & DCACHE_OP_FLAGS);
 1873 	dentry->d_op = op;
 1874 	if (flags)
 1875 		dentry->d_flags |= flags;
 1876 }
 1877 
 1878 void set_default_d_op(struct super_block *s, const struct dentry_operations *ops)
 1879 {
 1880 	unsigned int flags = d_op_flags(ops);
 1881 	s->__s_d_op = ops;
 1882 	s->s_d_flags = (s->s_d_flags & ~DCACHE_OP_FLAGS) | flags;
 1883 }
 1884 EXPORT_SYMBOL(set_default_d_op);
 1885 
 1886 static unsigned d_flags_for_inode(struct inode *inode)
 1887 {
 1888 	unsigned add_flags = DCACHE_REGULAR_TYPE;
 1889 
 1890 	if (!inode)
 1891 		return DCACHE_MISS_TYPE;
 1892 
 1893 	if (S_ISDIR(inode->i_mode)) {
 1894 		add_flags = DCACHE_DIRECTORY_TYPE;
 1895 		if (unlikely(!(inode->i_opflags & IOP_LOOKUP))) {
 1896 			if (unlikely(!inode->i_op->lookup))
 1897 				add_flags = DCACHE_AUTODIR_TYPE;
 1898 			else
 1899 				inode->i_opflags |= IOP_LOOKUP;
 1900 		}
 1901 		goto type_determined;
 1902 	}
 1903 
 1904 	if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) {
 1905 		if (unlikely(inode->i_op->get_link)) {
 1906 			add_flags = DCACHE_SYMLINK_TYPE;
 1907 			goto type_determined;
 1908 		}
 1909 		inode->i_opflags |= IOP_NOFOLLOW;
 1910 	}
 1911 
 1912 	if (unlikely(!S_ISREG(inode->i_mode)))
 1913 		add_flags = DCACHE_SPECIAL_TYPE;
 1914 
 1915 type_determined:
 1916 	if (unlikely(IS_AUTOMOUNT(inode)))
 1917 		add_flags |= DCACHE_NEED_AUTOMOUNT;
 1918 	return add_flags;
 1919 }
 1920 
 1921 static void __d_instantiate(struct dentry *dentry, struct inode *inode)
 1922 {
 1923 	unsigned add_flags = d_flags_for_inode(inode);
 1924 	WARN_ON(d_in_lookup(dentry));
 1925 
 1926 	spin_lock(&dentry->d_lock);
 1927 	/*
 1928 	 * The negative counter only tracks dentries on the LRU. Don't dec if
 1929 	 * d_lru is on another list.
 1930 	 */
 1931 	if ((dentry->d_flags &
 1932 	     (DCACHE_LRU_LIST|DCACHE_SHRINK_LIST)) == DCACHE_LRU_LIST)
 1933 		this_cpu_dec(nr_dentry_negative);
 1934 	hlist_add_head(&dentry->d_u.d_alias, &inode->i_dentry);
 1935 	raw_write_seqcount_begin(&dentry->d_seq);
 1936 	__d_set_inode_and_type(dentry, inode, add_flags);
 1937 	raw_write_seqcount_end(&dentry->d_seq);
 1938 	fsnotify_update_flags(dentry);
 1939 	spin_unlock(&dentry->d_lock);
 1940 }
 1941 
 1942 /**
 1943  * d_instantiate - fill in inode information for a dentry
 1944  * @entry: dentry to complete
 1945  * @inode: inode to attach to this dentry
 1946  *
 1947  * Fill in inode information in the entry.
 1948  *
 1949  * This turns negative dentries into productive full members
 1950  * of society.
 1951  *
 1952  * NOTE! This assumes that the inode count has been incremented
 1953  * (or otherwise set) by the caller to indicate that it is now
 1954  * in use by the dcache.
 1955  */
 1956  
 1957 void d_instantiate(struct dentry *entry, struct inode * inode)
 1958 {
 1959 	BUG_ON(!hlist_unhashed(&entry->d_u.d_alias));
 1960 	if (inode) {
 1961 		security_d_instantiate(entry, inode);
 1962 		spin_lock(&inode->i_lock);
 1963 		__d_instantiate(entry, inode);
 1964 		spin_unlock(&inode->i_lock);
 1965 	}
 1966 }
 1967 EXPORT_SYMBOL(d_instantiate);
 1968 
 1969 /*
 1970  * This should be equivalent to d_instantiate() + unlock_new_inode(),
 1971  * with lockdep-related part of unlock_new_inode() done before
 1972  * anything else.  Use that instead of open-coding d_instantiate()/
 1973  * unlock_new_inode() combinations.
 1974  */
 1975 void d_instantiate_new(struct dentry *entry, struct inode *inode)
 1976 {
 1977 	BUG_ON(!hlist_unhashed(&entry->d_u.d_alias));
 1978 	BUG_ON(!inode);
 1979 	lockdep_annotate_inode_mutex_key(inode);
 1980 	security_d_instantiate(entry, inode);
 1981 	spin_lock(&inode->i_lock);
 1982 	__d_instantiate(entry, inode);
 1983 	WARN_ON(!(inode->i_state & I_NEW));
 1984 	inode->i_state &= ~I_NEW & ~I_CREATING;
 1985 	/*
 1986 	 * Pairs with the barrier in prepare_to_wait_event() to make sure
 1987 	 * ___wait_var_event() either sees the bit cleared or
 1988 	 * waitqueue_active() check in wake_up_var() sees the waiter.
 1989 	 */
 1990 	smp_mb();
 1991 	inode_wake_up_bit(inode, __I_NEW);
 1992 	spin_unlock(&inode->i_lock);
 1993 }
 1994 EXPORT_SYMBOL(d_instantiate_new);
 1995 
 1996 struct dentry *d_make_root(struct inode *root_inode)
 1997 {
 1998 	struct dentry *res = NULL;
 1999 
 2000 	if (root_inode) {
 2001 		res = d_alloc_anon(root_inode->i_sb);
 2002 		if (res)
 2003 			d_instantiate(res, root_inode);
 2004 		else
 2005 			iput(root_inode);
 2006 	}
 2007 	return res;
 2008 }
 2009 EXPORT_SYMBOL(d_make_root);
 2010 
 2011 static struct dentry *__d_obtain_alias(struct inode *inode, bool disconnected)
 2012 {
 2013 	struct super_block *sb;
 2014 	struct dentry *new, *res;
 2015 
 2016 	if (!inode)
 2017 		return ERR_PTR(-ESTALE);
 2018 	if (IS_ERR(inode))
 2019 		return ERR_CAST(inode);
 2020 
 2021 	sb = inode->i_sb;
 2022 
 2023 	res = d_find_any_alias(inode); /* existing alias? */
 2024 	if (res)
 2025 		goto out;
 2026 
 2027 	new = d_alloc_anon(sb);
 2028 	if (!new) {
 2029 		res = ERR_PTR(-ENOMEM);
 2030 		goto out;
 2031 	}
 2032 
 2033 	security_d_instantiate(new, inode);
 2034 	spin_lock(&inode->i_lock);
 2035 	res = __d_find_any_alias(inode); /* recheck under lock */
 2036 	if (likely(!res)) { /* still no alias, attach a disconnected dentry */
 2037 		unsigned add_flags = d_flags_for_inode(inode);
 2038 
 2039 		if (disconnected)
 2040 			add_flags |= DCACHE_DISCONNECTED;
 2041 
 2042 		spin_lock(&new->d_lock);
 2043 		__d_set_inode_and_type(new, inode, add_flags);
 2044 		hlist_add_head(&new->d_u.d_alias, &inode->i_dentry);
 2045 		if (!disconnected) {
 2046 			hlist_bl_lock(&sb->s_roots);
 2047 			hlist_bl_add_head(&new->d_hash, &sb->s_roots);
 2048 			hlist_bl_unlock(&sb->s_roots);
 2049 		}
 2050 		spin_unlock(&new->d_lock);
 2051 		spin_unlock(&inode->i_lock);
 2052 		inode = NULL; /* consumed by new->d_inode */
 2053 		res = new;
 2054 	} else {
 2055 		spin_unlock(&inode->i_lock);
 2056 		dput(new);
 2057 	}
 2058 
 2059  out:
 2060 	iput(inode);
 2061 	return res;
 2062 }
 2063 
 2064 /**
 2065  * d_obtain_alias - find or allocate a DISCONNECTED dentry for a given inode
 2066  * @inode: inode to allocate the dentry for
 2067  *
 2068  * Obtain a dentry for an inode resulting from NFS filehandle conversion or
 2069  * similar open by handle operations.  The returned dentry may be anonymous,
 2070  * or may have a full name (if the inode was already in the cache).
 2071  *
 2072  * When called on a directory inode, we must ensure that the inode only ever
 2073  * has one dentry.  If a dentry is found, that is returned instead of
 2074  * allocating a new one.
 2075  *
 2076  * On successful return, the reference to the inode has been transferred
 2077  * to the dentry.  In case of an error the reference on the inode is released.
 2078  * To make it easier to use in export operations a %NULL or IS_ERR inode may
 2079  * be passed in and the error will be propagated to the return value,
 2080  * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
 2081  */
 2082 struct dentry *d_obtain_alias(struct inode *inode)
 2083 {
 2084 	return __d_obtain_alias(inode, true);
 2085 }
 2086 EXPORT_SYMBOL(d_obtain_alias);
 2087 
 2088 /**
 2089  * d_obtain_root - find or allocate a dentry for a given inode
 2090  * @inode: inode to allocate the dentry for
 2091  *
 2092  * Obtain an IS_ROOT dentry for the root of a filesystem.
 2093  *
 2094  * We must ensure that directory inodes only ever have one dentry.  If a
 2095  * dentry is found, that is returned instead of allocating a new one.
 2096  *
 2097  * On successful return, the reference to the inode has been transferred
 2098  * to the dentry.  In case of an error the reference on the inode is
 2099  * released.  A %NULL or IS_ERR inode may be passed in and will be the
 2100  * error will be propagate to the return value, with a %NULL @inode
 2101  * replaced by ERR_PTR(-ESTALE).
 2102  */
 2103 struct dentry *d_obtain_root(struct inode *inode)
 2104 {
 2105 	return __d_obtain_alias(inode, false);
 2106 }
 2107 EXPORT_SYMBOL(d_obtain_root);
 2108 
 2109 /**
 2110  * d_add_ci - lookup or allocate new dentry with case-exact name
 2111  * @dentry: the negative dentry that was passed to the parent's lookup func
 2112  * @inode:  the inode case-insensitive lookup has found
 2113  * @name:   the case-exact name to be associated with the returned dentry
 2114  *
 2115  * This is to avoid filling the dcache with case-insensitive names to the
 2116  * same inode, only the actual correct case is stored in the dcache for
 2117  * case-insensitive filesystems.
 2118  *
 2119  * For a case-insensitive lookup match and if the case-exact dentry
 2120  * already exists in the dcache, use it and return it.
 2121  *
 2122  * If no entry exists with the exact case name, allocate new dentry with
 2123  * the exact case, and return the spliced entry.
 2124  */
 2125 struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode,
 2126 			struct qstr *name)
 2127 {
 2128 	struct dentry *found, *res;
 2129 
 2130 	/*
 2131 	 * First check if a dentry matching the name already exists,
 2132 	 * if not go ahead and create it now.
 2133 	 */
 2134 	found = d_hash_and_lookup(dentry->d_parent, name);
 2135 	if (found) {
 2136 		iput(inode);
 2137 		return found;
 2138 	}
 2139 	if (d_in_lookup(dentry)) {
 2140 		found = d_alloc_parallel(dentry->d_parent, name,
 2141 					dentry->d_wait);
 2142 		if (IS_ERR(found) || !d_in_lookup(found)) {
 2143 			iput(inode);
 2144 			return found;
 2145 		}
 2146 	} else {
 2147 		found = d_alloc(dentry->d_parent, name);
 2148 		if (!found) {
 2149 			iput(inode);
 2150 			return ERR_PTR(-ENOMEM);
 2151 		} 
 2152 	}
 2153 	res = d_splice_alias(inode, found);
 2154 	if (res) {
 2155 		d_lookup_done(found);
 2156 		dput(found);
 2157 		return res;
 2158 	}
 2159 	return found;
 2160 }
 2161 EXPORT_SYMBOL(d_add_ci);
 2162 
 2163 /**
 2164  * d_same_name - compare dentry name with case-exact name
 2165  * @dentry: the negative dentry that was passed to the parent's lookup func
 2166  * @parent: parent dentry
 2167  * @name:   the case-exact name to be associated with the returned dentry
 2168  *
 2169  * Return: true if names are same, or false
 2170  */
 2171 bool d_same_name(const struct dentry *dentry, const struct dentry *parent,
 2172 		 const struct qstr *name)
 2173 {
 2174 	if (likely(!(parent->d_flags & DCACHE_OP_COMPARE))) {
 2175 		if (dentry->d_name.len != name->len)
 2176 			return false;
 2177 		return dentry_cmp(dentry, name->name, name->len) == 0;
 2178 	}
 2179 	return parent->d_op->d_compare(dentry,
 2180 				       dentry->d_name.len, dentry->d_name.name,
 2181 				       name) == 0;
 2182 }
 2183 EXPORT_SYMBOL_GPL(d_same_name);
 2184 
 2185 /*
 2186  * This is __d_lookup_rcu() when the parent dentry has
 2187  * DCACHE_OP_COMPARE, which makes things much nastier.
 2188  */
 2189 static noinline struct dentry *__d_lookup_rcu_op_compare(
 2190 	const struct dentry *parent,
 2191 	const struct qstr *name,
 2192 	unsigned *seqp)
 2193 {
 2194 	u64 hashlen = name->hash_len;
 2195 	struct hlist_bl_head *b = d_hash(hashlen);
 2196 	struct hlist_bl_node *node;
 2197 	struct dentry *dentry;
 2198 
 2199 	hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
 2200 		int tlen;
 2201 		const char *tname;
 2202 		unsigned seq;
 2203 
 2204 seqretry:
 2205 		seq = raw_seqcount_begin(&dentry->d_seq);
 2206 		if (dentry->d_parent != parent)
 2207 			continue;
 2208 		if (d_unhashed(dentry))
 2209 			continue;
 2210 		if (dentry->d_name.hash != hashlen_hash(hashlen))
 2211 			continue;
 2212 		tlen = dentry->d_name.len;
 2213 		tname = dentry->d_name.name;
 2214 		/* we want a consistent (name,len) pair */
 2215 		if (read_seqcount_retry(&dentry->d_seq, seq)) {
 2216 			cpu_relax();
 2217 			goto seqretry;
 2218 		}
 2219 		if (parent->d_op->d_compare(dentry, tlen, tname, name) != 0)
 2220 			continue;
 2221 		*seqp = seq;
 2222 		return dentry;
 2223 	}
 2224 	return NULL;
 2225 }
 2226 
 2227 /**
 2228  * __d_lookup_rcu - search for a dentry (racy, store-free)
 2229  * @parent: parent dentry
 2230  * @name: qstr of name we wish to find
 2231  * @seqp: returns d_seq value at the point where the dentry was found
 2232  * Returns: dentry, or NULL
 2233  *
 2234  * __d_lookup_rcu is the dcache lookup function for rcu-walk name
 2235  * resolution (store-free path walking) design described in
 2236  * Documentation/filesystems/path-lookup.txt.
 2237  *
 2238  * This is not to be used outside core vfs.
 2239  *
 2240  * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock
 2241  * held, and rcu_read_lock held. The returned dentry must not be stored into
 2242  * without taking d_lock and checking d_seq sequence count against @seq
 2243  * returned here.
 2244  *
 2245  * Alternatively, __d_lookup_rcu may be called again to look up the child of
 2246  * the returned dentry, so long as its parent's seqlock is checked after the
 2247  * child is looked up. Thus, an interlocking stepping of sequence lock checks
 2248  * is formed, giving integrity down the path walk.
 2249  *
 2250  * NOTE! The caller *has* to check the resulting dentry against the sequence
 2251  * number we've returned before using any of the resulting dentry state!
 2252  */
 2253 struct dentry *__d_lookup_rcu(const struct dentry *parent,
 2254 				const struct qstr *name,
 2255 				unsigned *seqp)
 2256 {
 2257 	u64 hashlen = name->hash_len;
 2258 	const unsigned char *str = name->name;
 2259 	struct hlist_bl_head *b = d_hash(hashlen);
 2260 	struct hlist_bl_node *node;
 2261 	struct dentry *dentry;
 2262 
 2263 	/*
 2264 	 * Note: There is significant duplication with __d_lookup_rcu which is
 2265 	 * required to prevent single threaded performance regressions
 2266 	 * especially on architectures where smp_rmb (in seqcounts) are costly.
 2267 	 * Keep the two functions in sync.
 2268 	 */
 2269 
 2270 	if (unlikely(parent->d_flags & DCACHE_OP_COMPARE))
 2271 		return __d_lookup_rcu_op_compare(parent, name, seqp);
 2272 
 2273 	/*
 2274 	 * The hash list is protected using RCU.
 2275 	 *
 2276 	 * Carefully use d_seq when comparing a candidate dentry, to avoid
 2277 	 * races with d_move().
 2278 	 *
 2279 	 * It is possible that concurrent renames can mess up our list
 2280 	 * walk here and result in missing our dentry, resulting in the
 2281 	 * false-negative result. d_lookup() protects against concurrent
 2282 	 * renames using rename_lock seqlock.
 2283 	 *
 2284 	 * See Documentation/filesystems/path-lookup.txt for more details.
 2285 	 */
 2286 	hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
 2287 		unsigned seq;
 2288 
 2289 		/*
 2290 		 * The dentry sequence count protects us from concurrent
 2291 		 * renames, and thus protects parent and name fields.
 2292 		 *
 2293 		 * The caller must perform a seqcount check in order
 2294 		 * to do anything useful with the returned dentry.
 2295 		 *
 2296 		 * NOTE! We do a "raw" seqcount_begin here. That means that
 2297 		 * we don't wait for the sequence count to stabilize if it
 2298 		 * is in the middle of a sequence change. If we do the slow
 2299 		 * dentry compare, we will do seqretries until it is stable,
 2300 		 * and if we end up with a successful lookup, we actually
 2301 		 * want to exit RCU lookup anyway.
 2302 		 *
 2303 		 * Note that raw_seqcount_begin still *does* smp_rmb(), so
 2304 		 * we are still guaranteed NUL-termination of ->d_name.name.
 2305 		 */
 2306 		seq = raw_seqcount_begin(&dentry->d_seq);
 2307 		if (dentry->d_parent != parent)
 2308 			continue;
 2309 		if (d_unhashed(dentry))
 2310 			continue;
 2311 		if (dentry->d_name.hash_len != hashlen)
 2312 			continue;
 2313 		if (dentry_cmp(dentry, str, hashlen_len(hashlen)) != 0)
 2314 			continue;
 2315 		*seqp = seq;
 2316 		return dentry;
 2317 	}
 2318 	return NULL;
 2319 }
 2320 
 2321 /**
 2322  * d_lookup - search for a dentry
 2323  * @parent: parent dentry
 2324  * @name: qstr of name we wish to find
 2325  * Returns: dentry, or NULL
 2326  *
 2327  * d_lookup searches the children of the parent dentry for the name in
 2328  * question. If the dentry is found its reference count is incremented and the
 2329  * dentry is returned. The caller must use dput to free the entry when it has
 2330  * finished using it. %NULL is returned if the dentry does not exist.
 2331  */
 2332 struct dentry *d_lookup(const struct dentry *parent, const struct qstr *name)
 2333 {
 2334 	struct dentry *dentry;
 2335 	unsigned seq;
 2336 
 2337 	do {
 2338 		seq = read_seqbegin(&rename_lock);
 2339 		dentry = __d_lookup(parent, name);
 2340 		if (dentry)
 2341 			break;
 2342 	} while (read_seqretry(&rename_lock, seq));
 2343 	return dentry;
 2344 }
 2345 EXPORT_SYMBOL(d_lookup);
 2346 
 2347 /**
 2348  * __d_lookup - search for a dentry (racy)
 2349  * @parent: parent dentry
 2350  * @name: qstr of name we wish to find
 2351  * Returns: dentry, or NULL
 2352  *
 2353  * __d_lookup is like d_lookup, however it may (rarely) return a
 2354  * false-negative result due to unrelated rename activity.
 2355  *
 2356  * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
 2357  * however it must be used carefully, eg. with a following d_lookup in
 2358  * the case of failure.
 2359  *
 2360  * __d_lookup callers must be commented.
 2361  */
 2362 struct dentry *__d_lookup(const struct dentry *parent, const struct qstr *name)
 2363 {
 2364 	unsigned int hash = name->hash;
 2365 	struct hlist_bl_head *b = d_hash(hash);
 2366 	struct hlist_bl_node *node;
 2367 	struct dentry *found = NULL;
 2368 	struct dentry *dentry;
 2369 
 2370 	/*
 2371 	 * Note: There is significant duplication with __d_lookup_rcu which is
 2372 	 * required to prevent single threaded performance regressions
 2373 	 * especially on architectures where smp_rmb (in seqcounts) are costly.
 2374 	 * Keep the two functions in sync.
 2375 	 */
 2376 
 2377 	/*
 2378 	 * The hash list is protected using RCU.
 2379 	 *
 2380 	 * Take d_lock when comparing a candidate dentry, to avoid races
 2381 	 * with d_move().
 2382 	 *
 2383 	 * It is possible that concurrent renames can mess up our list
 2384 	 * walk here and result in missing our dentry, resulting in the
 2385 	 * false-negative result. d_lookup() protects against concurrent
 2386 	 * renames using rename_lock seqlock.
 2387 	 *
 2388 	 * See Documentation/filesystems/path-lookup.txt for more details.
 2389 	 */
 2390 	rcu_read_lock();
 2391 	
 2392 	hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
 2393 
 2394 		if (dentry->d_name.hash != hash)
 2395 			continue;
 2396 
 2397 		spin_lock(&dentry->d_lock);
 2398 		if (dentry->d_parent != parent)
 2399 			goto next;
 2400 		if (d_unhashed(dentry))
 2401 			goto next;
 2402 
 2403 		if (!d_same_name(dentry, parent, name))
 2404 			goto next;
 2405 
 2406 		dentry->d_lockref.count++;
 2407 		found = dentry;
 2408 		spin_unlock(&dentry->d_lock);
 2409 		break;
 2410 next:
 2411 		spin_unlock(&dentry->d_lock);
 2412  	}
 2413  	rcu_read_unlock();
 2414 
 2415  	return found;
 2416 }
 2417 
 2418 /**
 2419  * d_hash_and_lookup - hash the qstr then search for a dentry
 2420  * @dir: Directory to search in
 2421  * @name: qstr of name we wish to find
 2422  *
 2423  * On lookup failure NULL is returned; on bad name - ERR_PTR(-error)
 2424  */
 2425 struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name)
 2426 {
 2427 	/*
 2428 	 * Check for a fs-specific hash function. Note that we must
 2429 	 * calculate the standard hash first, as the d_op->d_hash()
 2430 	 * routine may choose to leave the hash value unchanged.
 2431 	 */
 2432 	name->hash = full_name_hash(dir, name->name, name->len);
 2433 	if (dir->d_flags & DCACHE_OP_HASH) {
 2434 		int err = dir->d_op->d_hash(dir, name);
 2435 		if (unlikely(err < 0))
 2436 			return ERR_PTR(err);
 2437 	}
 2438 	return d_lookup(dir, name);
 2439 }
 2440 
 2441 /*
 2442  * When a file is deleted, we have two options:
 2443  * - turn this dentry into a negative dentry
 2444  * - unhash this dentry and free it.
 2445  *
 2446  * Usually, we want to just turn this into
 2447  * a negative dentry, but if anybody else is
 2448  * currently using the dentry or the inode
 2449  * we can't do that and we fall back on removing
 2450  * it from the hash queues and waiting for
 2451  * it to be deleted later when it has no users
 2452  */
 2453  
 2454 /**
 2455  * d_delete - delete a dentry
 2456  * @dentry: The dentry to delete
 2457  *
 2458  * Turn the dentry into a negative dentry if possible, otherwise
 2459  * remove it from the hash queues so it can be deleted later
 2460  */
 2461  
 2462 void d_delete(struct dentry * dentry)
 2463 {
 2464 	struct inode *inode = dentry->d_inode;
 2465 
 2466 	spin_lock(&inode->i_lock);
 2467 	spin_lock(&dentry->d_lock);
 2468 	/*
 2469 	 * Are we the only user?
 2470 	 */
 2471 	if (dentry->d_lockref.count == 1) {
 2472 		if (dentry_negative_policy)
 2473 			__d_drop(dentry);
 2474 		dentry->d_flags &= ~DCACHE_CANT_MOUNT;
 2475 		dentry_unlink_inode(dentry);
 2476 	} else {
 2477 		__d_drop(dentry);
 2478 		spin_unlock(&dentry->d_lock);
 2479 		spin_unlock(&inode->i_lock);
 2480 	}
 2481 }
 2482 EXPORT_SYMBOL(d_delete);
 2483 
 2484 static void __d_rehash(struct dentry *entry)
 2485 {
 2486 	struct hlist_bl_head *b = d_hash(entry->d_name.hash);
 2487 
 2488 	hlist_bl_lock(b);
 2489 	hlist_bl_add_head_rcu(&entry->d_hash, b);
 2490 	hlist_bl_unlock(b);
 2491 }
 2492 
 2493 /**
 2494  * d_rehash	- add an entry back to the hash
 2495  * @entry: dentry to add to the hash
 2496  *
 2497  * Adds a dentry to the hash according to its name.
 2498  */
 2499  
 2500 void d_rehash(struct dentry * entry)
 2501 {
 2502 	spin_lock(&entry->d_lock);
 2503 	__d_rehash(entry);
 2504 	spin_unlock(&entry->d_lock);
 2505 }
 2506 EXPORT_SYMBOL(d_rehash);
 2507 
 2508 static inline unsigned start_dir_add(struct inode *dir)
 2509 {
 2510 	preempt_disable_nested();
 2511 	for (;;) {
 2512 		unsigned n = READ_ONCE(dir->i_dir_seq);
 2513 		if (!(n & 1) && try_cmpxchg(&dir->i_dir_seq, &n, n + 1))
 2514 			return n;
 2515 		cpu_relax();
 2516 	}
 2517 }
 2518 
 2519 static inline void end_dir_add(struct inode *dir, unsigned int n,
 2520 			       wait_queue_head_t *d_wait)
 2521 {
 2522 	smp_store_release(&dir->i_dir_seq, n + 2);
 2523 	preempt_enable_nested();
 2524 	if (wq_has_sleeper(d_wait))
 2525 		wake_up_all(d_wait);
 2526 }
 2527 
 2528 static void d_wait_lookup(struct dentry *dentry)
 2529 {
 2530 	if (d_in_lookup(dentry)) {
 2531 		DECLARE_WAITQUEUE(wait, current);
 2532 		add_wait_queue(dentry->d_wait, &wait);
 2533 		do {
 2534 			set_current_state(TASK_UNINTERRUPTIBLE);
 2535 			spin_unlock(&dentry->d_lock);
 2536 			schedule();
 2537 			spin_lock(&dentry->d_lock);
 2538 		} while (d_in_lookup(dentry));
 2539 	}
 2540 }
 2541 
 2542 struct dentry *d_alloc_parallel(struct dentry *parent,
 2543 				const struct qstr *name,
 2544 				wait_queue_head_t *wq)
 2545 {
 2546 	unsigned int hash = name->hash;
 2547 	struct hlist_bl_head *b = in_lookup_hash(parent, hash);
 2548 	struct hlist_bl_node *node;
 2549 	struct dentry *new = __d_alloc(parent->d_sb, name);
 2550 	struct dentry *dentry;
 2551 	unsigned seq, r_seq, d_seq;
 2552 
 2553 	if (unlikely(!new))
 2554 		return ERR_PTR(-ENOMEM);
 2555 
 2556 	new->d_flags |= DCACHE_PAR_LOOKUP;
 2557 	spin_lock(&parent->d_lock);
 2558 	new->d_parent = dget_dlock(parent);
 2559 	hlist_add_head(&new->d_sib, &parent->d_children);
 2560 	if (parent->d_flags & DCACHE_DISCONNECTED)
 2561 		new->d_flags |= DCACHE_DISCONNECTED;
 2562 	spin_unlock(&parent->d_lock);
 2563 
 2564 retry:
 2565 	rcu_read_lock();
 2566 	seq = smp_load_acquire(&parent->d_inode->i_dir_seq);
 2567 	r_seq = read_seqbegin(&rename_lock);
 2568 	dentry = __d_lookup_rcu(parent, name, &d_seq);
 2569 	if (unlikely(dentry)) {
 2570 		if (!lockref_get_not_dead(&dentry->d_lockref)) {
 2571 			rcu_read_unlock();
 2572 			goto retry;
 2573 		}
 2574 		if (read_seqcount_retry(&dentry->d_seq, d_seq)) {
 2575 			rcu_read_unlock();
 2576 			dput(dentry);
 2577 			goto retry;
 2578 		}
 2579 		rcu_read_unlock();
 2580 		dput(new);
 2581 		return dentry;
 2582 	}
 2583 	if (unlikely(read_seqretry(&rename_lock, r_seq))) {
 2584 		rcu_read_unlock();
 2585 		goto retry;
 2586 	}
 2587 
 2588 	if (unlikely(seq & 1)) {
 2589 		rcu_read_unlock();
 2590 		goto retry;
 2591 	}
 2592 
 2593 	hlist_bl_lock(b);
 2594 	if (unlikely(READ_ONCE(parent->d_inode->i_dir_seq) != seq)) {
 2595 		hlist_bl_unlock(b);
 2596 		rcu_read_unlock();
 2597 		goto retry;
 2598 	}
 2599 	/*
 2600 	 * No changes for the parent since the beginning of d_lookup().
 2601 	 * Since all removals from the chain happen with hlist_bl_lock(),
 2602 	 * any potential in-lookup matches are going to stay here until
 2603 	 * we unlock the chain.  All fields are stable in everything
 2604 	 * we encounter.
 2605 	 */
 2606 	hlist_bl_for_each_entry(dentry, node, b, d_u.d_in_lookup_hash) {
 2607 		if (dentry->d_name.hash != hash)
 2608 			continue;
 2609 		if (dentry->d_parent != parent)
 2610 			continue;
 2611 		if (!d_same_name(dentry, parent, name))
 2612 			continue;
 2613 		hlist_bl_unlock(b);
 2614 		/* now we can try to grab a reference */
 2615 		if (!lockref_get_not_dead(&dentry->d_lockref)) {
 2616 			rcu_read_unlock();
 2617 			goto retry;
 2618 		}
 2619 
 2620 		rcu_read_unlock();
 2621 		/*
 2622 		 * somebody is likely to be still doing lookup for it;
 2623 		 * wait for them to finish
 2624 		 */
 2625 		spin_lock(&dentry->d_lock);
 2626 		d_wait_lookup(dentry);
 2627 		/*
 2628 		 * it's not in-lookup anymore; in principle we should repeat
 2629 		 * everything from dcache lookup, but it's likely to be what
 2630 		 * d_lookup() would've found anyway.  If it is, just return it;
 2631 		 * otherwise we really have to repeat the whole thing.
 2632 		 */
 2633 		if (unlikely(dentry->d_name.hash != hash))
 2634 			goto mismatch;
 2635 		if (unlikely(dentry->d_parent != parent))
 2636 			goto mismatch;
 2637 		if (unlikely(d_unhashed(dentry)))
 2638 			goto mismatch;
 2639 		if (unlikely(!d_same_name(dentry, parent, name)))
 2640 			goto mismatch;
 2641 		/* OK, it *is* a hashed match; return it */
 2642 		spin_unlock(&dentry->d_lock);
 2643 		dput(new);
 2644 		return dentry;
 2645 	}
 2646 	rcu_read_unlock();
 2647 	new->d_wait = wq;
 2648 	hlist_bl_add_head(&new->d_u.d_in_lookup_hash, b);
 2649 	hlist_bl_unlock(b);
 2650 	return new;
 2651 mismatch:
 2652 	spin_unlock(&dentry->d_lock);
 2653 	dput(dentry);
 2654 	goto retry;
 2655 }
 2656 EXPORT_SYMBOL(d_alloc_parallel);
 2657 
 2658 /*
 2659  * - Unhash the dentry
 2660  * - Retrieve and clear the waitqueue head in dentry
 2661  * - Return the waitqueue head
 2662  */
 2663 static wait_queue_head_t *__d_lookup_unhash(struct dentry *dentry)
 2664 {
 2665 	wait_queue_head_t *d_wait;
 2666 	struct hlist_bl_head *b;
 2667 
 2668 	lockdep_assert_held(&dentry->d_lock);
 2669 
 2670 	b = in_lookup_hash(dentry->d_parent, dentry->d_name.hash);
 2671 	hlist_bl_lock(b);
 2672 	dentry->d_flags &= ~DCACHE_PAR_LOOKUP;
 2673 	__hlist_bl_del(&dentry->d_u.d_in_lookup_hash);
 2674 	d_wait = dentry->d_wait;
 2675 	dentry->d_wait = NULL;
 2676 	hlist_bl_unlock(b);
 2677 	INIT_HLIST_NODE(&dentry->d_u.d_alias);
 2678 	INIT_LIST_HEAD(&dentry->d_lru);
 2679 	return d_wait;
 2680 }
 2681 
 2682 void __d_lookup_unhash_wake(struct dentry *dentry)
 2683 {
 2684 	spin_lock(&dentry->d_lock);
 2685 	wake_up_all(__d_lookup_unhash(dentry));
 2686 	spin_unlock(&dentry->d_lock);
 2687 }
 2688 EXPORT_SYMBOL(__d_lookup_unhash_wake);
 2689 
 2690 /* inode->i_lock held if inode is non-NULL */
 2691 
 2692 static inline void __d_add(struct dentry *dentry, struct inode *inode,
 2693 			   const struct dentry_operations *ops)
 2694 {
 2695 	wait_queue_head_t *d_wait;
 2696 	struct inode *dir = NULL;
 2697 	unsigned n;
 2698 	spin_lock(&dentry->d_lock);
 2699 	if (unlikely(d_in_lookup(dentry))) {
 2700 		dir = dentry->d_parent->d_inode;
 2701 		n = start_dir_add(dir);
 2702 		d_wait = __d_lookup_unhash(dentry);
 2703 	}
 2704 	if (unlikely(ops))
 2705 		d_set_d_op(dentry, ops);
 2706 	if (inode) {
 2707 		unsigned add_flags = d_flags_for_inode(inode);
 2708 		hlist_add_head(&dentry->d_u.d_alias, &inode->i_dentry);
 2709 		raw_write_seqcount_begin(&dentry->d_seq);
 2710 		__d_set_inode_and_type(dentry, inode, add_flags);
 2711 		raw_write_seqcount_end(&dentry->d_seq);
 2712 		fsnotify_update_flags(dentry);
 2713 	}
 2714 	__d_rehash(dentry);
 2715 	if (dir)
 2716 		end_dir_add(dir, n, d_wait);
 2717 	spin_unlock(&dentry->d_lock);
 2718 	if (inode)
 2719 		spin_unlock(&inode->i_lock);
 2720 }
 2721 
 2722 /**
 2723  * d_add - add dentry to hash queues
 2724  * @entry: dentry to add
 2725  * @inode: The inode to attach to this dentry
 2726  *
 2727  * This adds the entry to the hash queues and initializes @inode.
 2728  * The entry was actually filled in earlier during d_alloc().
 2729  */
 2730 
 2731 void d_add(struct dentry *entry, struct inode *inode)
 2732 {
 2733 	if (inode) {
 2734 		security_d_instantiate(entry, inode);
 2735 		spin_lock(&inode->i_lock);
 2736 	}
 2737 	__d_add(entry, inode, NULL);
 2738 }
 2739 EXPORT_SYMBOL(d_add);
 2740 
 2741 static void swap_names(struct dentry *dentry, struct dentry *target)
 2742 {
 2743 	if (unlikely(dname_external(target))) {
 2744 		if (unlikely(dname_external(dentry))) {
 2745 			/*
 2746 			 * Both external: swap the pointers
 2747 			 */
 2748 			swap(target->__d_name.name, dentry->__d_name.name);
 2749 		} else {
 2750 			/*
 2751 			 * dentry:internal, target:external.  Steal target's
 2752 			 * storage and make target internal.
 2753 			 */
 2754 			dentry->__d_name.name = target->__d_name.name;
 2755 			target->d_shortname = dentry->d_shortname;
 2756 			target->__d_name.name = target->d_shortname.string;
 2757 		}
 2758 	} else {
 2759 		if (unlikely(dname_external(dentry))) {
 2760 			/*
 2761 			 * dentry:external, target:internal.  Give dentry's
 2762 			 * storage to target and make dentry internal
 2763 			 */
 2764 			target->__d_name.name = dentry->__d_name.name;
 2765 			dentry->d_shortname = target->d_shortname;
 2766 			dentry->__d_name.name = dentry->d_shortname.string;
 2767 		} else {
 2768 			/*
 2769 			 * Both are internal.
 2770 			 */
 2771 			for (int i = 0; i < DNAME_INLINE_WORDS; i++)
 2772 				swap(dentry->d_shortname.words[i],
 2773 				     target->d_shortname.words[i]);
 2774 		}
 2775 	}
 2776 	swap(dentry->__d_name.hash_len, target->__d_name.hash_len);
 2777 }
 2778 
 2779 static void copy_name(struct dentry *dentry, struct dentry *target)
 2780 {
 2781 	struct external_name *old_name = NULL;
 2782 	if (unlikely(dname_external(dentry)))
 2783 		old_name = external_name(dentry);
 2784 	if (unlikely(dname_external(target))) {
 2785 		atomic_inc(&external_name(target)->count);
 2786 		dentry->__d_name = target->__d_name;
 2787 	} else {
 2788 		dentry->d_shortname = target->d_shortname;
 2789 		dentry->__d_name.name = dentry->d_shortname.string;
 2790 		dentry->__d_name.hash_len = target->__d_name.hash_len;
 2791 	}
 2792 	if (old_name && likely(atomic_dec_and_test(&old_name->count)))
 2793 		kfree_rcu(old_name, head);
 2794 }
 2795 
 2796 /*
 2797  * __d_move - move a dentry
 2798  * @dentry: entry to move
 2799  * @target: new dentry
 2800  * @exchange: exchange the two dentries
 2801  *
 2802  * Update the dcache to reflect the move of a file name. Negative dcache
 2803  * entries should not be moved in this way. Caller must hold rename_lock, the
 2804  * i_rwsem of the source and target directories (exclusively), and the sb->
 2805  * s_vfs_rename_mutex if they differ. See lock_rename().
 2806  */
 2807 static void __d_move(struct dentry *dentry, struct dentry *target,
 2808 		     bool exchange)
 2809 {
 2810 	struct dentry *old_parent, *p;
 2811 	wait_queue_head_t *d_wait;
 2812 	struct inode *dir = NULL;
 2813 	unsigned n;
 2814 
 2815 	WARN_ON(!dentry->d_inode);
 2816 	if (WARN_ON(dentry == target))
 2817 		return;
 2818 
 2819 	BUG_ON(d_ancestor(target, dentry));
 2820 	old_parent = dentry->d_parent;
 2821 	p = d_ancestor(old_parent, target);
 2822 	if (IS_ROOT(dentry)) {
 2823 		BUG_ON(p);
 2824 		spin_lock(&target->d_parent->d_lock);
 2825 	} else if (!p) {
 2826 		/* target is not a descendent of dentry->d_parent */
 2827 		spin_lock(&target->d_parent->d_lock);
 2828 		spin_lock_nested(&old_parent->d_lock, DENTRY_D_LOCK_NESTED);
 2829 	} else {
 2830 		BUG_ON(p == dentry);
 2831 		spin_lock(&old_parent->d_lock);
 2832 		if (p != target)
 2833 			spin_lock_nested(&target->d_parent->d_lock,
 2834 					DENTRY_D_LOCK_NESTED);
 2835 	}
 2836 	spin_lock_nested(&dentry->d_lock, 2);
 2837 	spin_lock_nested(&target->d_lock, 3);
 2838 
 2839 	if (unlikely(d_in_lookup(target))) {
 2840 		dir = target->d_parent->d_inode;
 2841 		n = start_dir_add(dir);
 2842 		d_wait = __d_lookup_unhash(target);
 2843 	}
 2844 
 2845 	write_seqcount_begin(&dentry->d_seq);
 2846 	write_seqcount_begin_nested(&target->d_seq, DENTRY_D_LOCK_NESTED);
 2847 
 2848 	/* unhash both */
 2849 	if (!d_unhashed(dentry))
 2850 		___d_drop(dentry);
 2851 	if (!d_unhashed(target))
 2852 		___d_drop(target);
 2853 
 2854 	/* ... and switch them in the tree */
 2855 	dentry->d_parent = target->d_parent;
 2856 	if (!exchange) {
 2857 		copy_name(dentry, target);
 2858 		target->d_hash.pprev = NULL;
 2859 		dentry->d_parent->d_lockref.count++;
 2860 		if (dentry != old_parent) /* wasn't IS_ROOT */
 2861 			WARN_ON(!--old_parent->d_lockref.count);
 2862 	} else {
 2863 		target->d_parent = old_parent;
 2864 		swap_names(dentry, target);
 2865 		if (!hlist_unhashed(&target->d_sib))
 2866 			__hlist_del(&target->d_sib);
 2867 		hlist_add_head(&target->d_sib, &target->d_parent->d_children);
 2868 		__d_rehash(target);
 2869 		fsnotify_update_flags(target);
 2870 	}
 2871 	if (!hlist_unhashed(&dentry->d_sib))
 2872 		__hlist_del(&dentry->d_sib);
 2873 	hlist_add_head(&dentry->d_sib, &dentry->d_parent->d_children);
 2874 	__d_rehash(dentry);
 2875 	fsnotify_update_flags(dentry);
 2876 	fscrypt_handle_d_move(dentry);
 2877 
 2878 	write_seqcount_end(&target->d_seq);
 2879 	write_seqcount_end(&dentry->d_seq);
 2880 
 2881 	if (dir)
 2882 		end_dir_add(dir, n, d_wait);
 2883 
 2884 	if (dentry->d_parent != old_parent)
 2885 		spin_unlock(&dentry->d_parent->d_lock);
 2886 	if (dentry != old_parent)
 2887 		spin_unlock(&old_parent->d_lock);
 2888 	spin_unlock(&target->d_lock);
 2889 	spin_unlock(&dentry->d_lock);
 2890 }
 2891 
 2892 /*
 2893  * d_move - move a dentry
 2894  * @dentry: entry to move
 2895  * @target: new dentry
 2896  *
 2897  * Update the dcache to reflect the move of a file name. Negative
 2898  * dcache entries should not be moved in this way. See the locking
 2899  * requirements for __d_move.
 2900  */
 2901 void d_move(struct dentry *dentry, struct dentry *target)
 2902 {
 2903 	write_seqlock(&rename_lock);
 2904 	__d_move(dentry, target, false);
 2905 	write_sequnlock(&rename_lock);
 2906 }
 2907 EXPORT_SYMBOL(d_move);
 2908 
 2909 /*
 2910  * d_exchange - exchange two dentries
 2911  * @dentry1: first dentry
 2912  * @dentry2: second dentry
 2913  */
 2914 void d_exchange(struct dentry *dentry1, struct dentry *dentry2)
 2915 {
 2916 	write_seqlock(&rename_lock);
 2917 
 2918 	WARN_ON(!dentry1->d_inode);
 2919 	WARN_ON(!dentry2->d_inode);
 2920 	WARN_ON(IS_ROOT(dentry1));
 2921 	WARN_ON(IS_ROOT(dentry2));
 2922 
 2923 	__d_move(dentry1, dentry2, true);
 2924 
 2925 	write_sequnlock(&rename_lock);
 2926 }
 2927 EXPORT_SYMBOL(d_exchange);
 2928 
 2929 /**
 2930  * d_ancestor - search for an ancestor
 2931  * @p1: ancestor dentry
 2932  * @p2: child dentry
 2933  *
 2934  * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
 2935  * an ancestor of p2, else NULL.
 2936  */
 2937 struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2)
 2938 {
 2939 	struct dentry *p;
 2940 
 2941 	for (p = p2; !IS_ROOT(p); p = p->d_parent) {
 2942 		if (p->d_parent == p1)
 2943 			return p;
 2944 	}
 2945 	return NULL;
 2946 }
 2947 
 2948 /*
 2949  * This helper attempts to cope with remotely renamed directories
 2950  *
 2951  * It assumes that the caller is already holding
 2952  * dentry->d_parent->d_inode->i_rwsem, and rename_lock
 2953  *
 2954  * Note: If ever the locking in lock_rename() changes, then please
 2955  * remember to update this too...
 2956  */
 2957 static int __d_unalias(struct dentry *dentry, struct dentry *alias)
 2958 {
 2959 	struct mutex *m1 = NULL;
 2960 	struct rw_semaphore *m2 = NULL;
 2961 	int ret = -ESTALE;
 2962 
 2963 	/* If alias and dentry share a parent, then no extra locks required */
 2964 	if (alias->d_parent == dentry->d_parent)
 2965 		goto out_unalias;
 2966 
 2967 	/* See lock_rename() */
 2968 	if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex))
 2969 		goto out_err;
 2970 	m1 = &dentry->d_sb->s_vfs_rename_mutex;
 2971 	if (!inode_trylock_shared(alias->d_parent->d_inode))
 2972 		goto out_err;
 2973 	m2 = &alias->d_parent->d_inode->i_rwsem;
 2974 out_unalias:
 2975 	if (alias->d_op && alias->d_op->d_unalias_trylock &&
 2976 	    !alias->d_op->d_unalias_trylock(alias))
 2977 		goto out_err;
 2978 	__d_move(alias, dentry, false);
 2979 	if (alias->d_op && alias->d_op->d_unalias_unlock)
 2980 		alias->d_op->d_unalias_unlock(alias);
 2981 	ret = 0;
 2982 out_err:
 2983 	if (m2)
 2984 		up_read(m2);
 2985 	if (m1)
 2986 		mutex_unlock(m1);
 2987 	return ret;
 2988 }
 2989 
 2990 struct dentry *d_splice_alias_ops(struct inode *inode, struct dentry *dentry,
 2991 				  const struct dentry_operations *ops)
 2992 {
 2993 	if (IS_ERR(inode))
 2994 		return ERR_CAST(inode);
 2995 
 2996 	BUG_ON(!d_unhashed(dentry));
 2997 
 2998 	if (!inode)
 2999 		goto out;
 3000 
 3001 	security_d_instantiate(dentry, inode);
 3002 	spin_lock(&inode->i_lock);
 3003 	if (S_ISDIR(inode->i_mode)) {
 3004 		struct dentry *new = __d_find_any_alias(inode);
 3005 		if (unlikely(new)) {
 3006 			/* The reference to new ensures it remains an alias */
 3007 			spin_unlock(&inode->i_lock);
 3008 			write_seqlock(&rename_lock);
 3009 			if (unlikely(d_ancestor(new, dentry))) {
 3010 				write_sequnlock(&rename_lock);
 3011 				dput(new);
 3012 				new = ERR_PTR(-ELOOP);
 3013 				pr_warn_ratelimited(
 3014 					"VFS: Lookup of '%s' in %s %s"
 3015 					" would have caused loop\n",
 3016 					dentry->d_name.name,
 3017 					inode->i_sb->s_type->name,
 3018 					inode->i_sb->s_id);
 3019 			} else if (!IS_ROOT(new)) {
 3020 				struct dentry *old_parent = dget(new->d_parent);
 3021 				int err = __d_unalias(dentry, new);
 3022 				write_sequnlock(&rename_lock);
 3023 				if (err) {
 3024 					dput(new);
 3025 					new = ERR_PTR(err);
 3026 				}
 3027 				dput(old_parent);
 3028 			} else {
 3029 				__d_move(new, dentry, false);
 3030 				write_sequnlock(&rename_lock);
 3031 			}
 3032 			iput(inode);
 3033 			return new;
 3034 		}
 3035 	}
 3036 out:
 3037 	__d_add(dentry, inode, ops);
 3038 	return NULL;
 3039 }
 3040 
 3041 /**
 3042  * d_splice_alias - splice a disconnected dentry into the tree if one exists
 3043  * @inode:  the inode which may have a disconnected dentry
 3044  * @dentry: a negative dentry which we want to point to the inode.
 3045  *
 3046  * If inode is a directory and has an IS_ROOT alias, then d_move that in
 3047  * place of the given dentry and return it, else simply d_add the inode
 3048  * to the dentry and return NULL.
 3049  *
 3050  * If a non-IS_ROOT directory is found, the filesystem is corrupt, and
 3051  * we should error out: directories can't have multiple aliases.
 3052  *
 3053  * This is needed in the lookup routine of any filesystem that is exportable
 3054  * (via knfsd) so that we can build dcache paths to directories effectively.
 3055  *
 3056  * If a dentry was found and moved, then it is returned.  Otherwise NULL
 3057  * is returned.  This matches the expected return value of ->lookup.
 3058  *
 3059  * Cluster filesystems may call this function with a negative, hashed dentry.
 3060  * In that case, we know that the inode will be a regular file, and also this
 3061  * will only occur during atomic_open. So we need to check for the dentry
 3062  * being already hashed only in the final case.
 3063  */
 3064 struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry)
 3065 {
 3066 	return d_splice_alias_ops(inode, dentry, NULL);
 3067 }
 3068 EXPORT_SYMBOL(d_splice_alias);
 3069 
 3070 /*
 3071  * Test whether new_dentry is a subdirectory of old_dentry.
 3072  *
 3073  * Trivially implemented using the dcache structure
 3074  */
 3075 
 3076 /**
 3077  * is_subdir - is new dentry a subdirectory of old_dentry
 3078  * @new_dentry: new dentry
 3079  * @old_dentry: old dentry
 3080  *
 3081  * Returns true if new_dentry is a subdirectory of the parent (at any depth).
 3082  * Returns false otherwise.
 3083  * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
 3084  */
 3085   
 3086 bool is_subdir(struct dentry *new_dentry, struct dentry *old_dentry)
 3087 {
 3088 	bool subdir;
 3089 	unsigned seq;
 3090 
 3091 	if (new_dentry == old_dentry)
 3092 		return true;
 3093 
 3094 	/* Access d_parent under rcu as d_move() may change it. */
 3095 	rcu_read_lock();
 3096 	seq = read_seqbegin(&rename_lock);
 3097 	subdir = d_ancestor(old_dentry, new_dentry);
 3098 	 /* Try lockless once... */
 3099 	if (read_seqretry(&rename_lock, seq)) {
 3100 		/* ...else acquire lock for progress even on deep chains. */
 3101 		read_seqlock_excl(&rename_lock);
 3102 		subdir = d_ancestor(old_dentry, new_dentry);
 3103 		read_sequnlock_excl(&rename_lock);
 3104 	}
 3105 	rcu_read_unlock();
 3106 	return subdir;
 3107 }
 3108 EXPORT_SYMBOL(is_subdir);
 3109 
 3110 static enum d_walk_ret d_genocide_kill(void *data, struct dentry *dentry)
 3111 {
 3112 	struct dentry *root = data;
 3113 	if (dentry != root) {
 3114 		if (d_unhashed(dentry) || !dentry->d_inode)
 3115 			return D_WALK_SKIP;
 3116 
 3117 		if (!(dentry->d_flags & DCACHE_GENOCIDE)) {
 3118 			dentry->d_flags |= DCACHE_GENOCIDE;
 3119 			dentry->d_lockref.count--;
 3120 		}
 3121 	}
 3122 	return D_WALK_CONTINUE;
 3123 }
 3124 
 3125 void d_genocide(struct dentry *parent)
 3126 {
 3127 	d_walk(parent, parent, d_genocide_kill);
 3128 }
 3129 
 3130 void d_mark_tmpfile(struct file *file, struct inode *inode)
 3131 {
 3132 	struct dentry *dentry = file->f_path.dentry;
 3133 
 3134 	BUG_ON(dname_external(dentry) ||
 3135 		!hlist_unhashed(&dentry->d_u.d_alias) ||
 3136 		!d_unlinked(dentry));
 3137 	spin_lock(&dentry->d_parent->d_lock);
 3138 	spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
 3139 	dentry->__d_name.len = sprintf(dentry->d_shortname.string, "#%llu",
 3140 				(unsigned long long)inode->i_ino);
 3141 	spin_unlock(&dentry->d_lock);
 3142 	spin_unlock(&dentry->d_parent->d_lock);
 3143 }
 3144 EXPORT_SYMBOL(d_mark_tmpfile);
 3145 
 3146 void d_tmpfile(struct file *file, struct inode *inode)
 3147 {
 3148 	struct dentry *dentry = file->f_path.dentry;
 3149 
 3150 	inode_dec_link_count(inode);
 3151 	d_mark_tmpfile(file, inode);
 3152 	d_instantiate(dentry, inode);
 3153 }
 3154 EXPORT_SYMBOL(d_tmpfile);
 3155 
 3156 /*
 3157  * Obtain inode number of the parent dentry.
 3158  */
 3159 ino_t d_parent_ino(struct dentry *dentry)
 3160 {
 3161 	struct dentry *parent;
 3162 	struct inode *iparent;
 3163 	unsigned seq;
 3164 	ino_t ret;
 3165 
 3166 	scoped_guard(rcu) {
 3167 		seq = raw_seqcount_begin(&dentry->d_seq);
 3168 		parent = READ_ONCE(dentry->d_parent);
 3169 		iparent = d_inode_rcu(parent);
 3170 		if (likely(iparent)) {
 3171 			ret = iparent->i_ino;
 3172 			if (!read_seqcount_retry(&dentry->d_seq, seq))
 3173 				return ret;
 3174 		}
 3175 	}
 3176 
 3177 	spin_lock(&dentry->d_lock);
 3178 	ret = dentry->d_parent->d_inode->i_ino;
 3179 	spin_unlock(&dentry->d_lock);
 3180 	return ret;
 3181 }
 3182 EXPORT_SYMBOL(d_parent_ino);
 3183 
 3184 static __initdata unsigned long dhash_entries;
 3185 static int __init set_dhash_entries(char *str)
 3186 {
 3187 	if (!str)
 3188 		return 0;
 3189 	dhash_entries = simple_strtoul(str, &str, 0);
 3190 	return 1;
 3191 }
 3192 __setup("dhash_entries=", set_dhash_entries);
 3193 
 3194 static void __init dcache_init_early(void)
 3195 {
 3196 	/* If hashes are distributed across NUMA nodes, defer
 3197 	 * hash allocation until vmalloc space is available.
 3198 	 */
 3199 	if (hashdist)
 3200 		return;
 3201 
 3202 	dentry_hashtable =
 3203 		alloc_large_system_hash("Dentry cache",
 3204 					sizeof(struct hlist_bl_head),
 3205 					dhash_entries,
 3206 					13,
 3207 					HASH_EARLY | HASH_ZERO,
 3208 					&d_hash_shift,
 3209 					NULL,
 3210 					2,
 3211 					0);
 3212 	d_hash_shift = 32 - d_hash_shift;
 3213 
 3214 	runtime_const_init(shift, d_hash_shift);
 3215 	runtime_const_init(ptr, dentry_hashtable);
 3216 }
 3217 
 3218 static void __init dcache_init(void)
 3219 {
 3220 	/*
 3221 	 * A constructor could be added for stable state like the lists,
 3222 	 * but it is probably not worth it because of the cache nature
 3223 	 * of the dcache.
 3224 	 */
 3225 	dentry_cache = KMEM_CACHE_USERCOPY(dentry,
 3226 		SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_ACCOUNT,
 3227 		d_shortname.string);
 3228 
 3229 	/* Hash may have been set up in dcache_init_early */
 3230 	if (!hashdist)
 3231 		return;
 3232 
 3233 	dentry_hashtable =
 3234 		alloc_large_system_hash("Dentry cache",
 3235 					sizeof(struct hlist_bl_head),
 3236 					dhash_entries,
 3237 					13,
 3238 					HASH_ZERO,
 3239 					&d_hash_shift,
 3240 					NULL,
 3241 					2,
 3242 					0);
 3243 	d_hash_shift = 32 - d_hash_shift;
 3244 
 3245 	runtime_const_init(shift, d_hash_shift);
 3246 	runtime_const_init(ptr, dentry_hashtable);
 3247 }
 3248 
 3249 /* SLAB cache for __getname() consumers */
 3250 struct kmem_cache *names_cachep __ro_after_init;
 3251 EXPORT_SYMBOL(names_cachep);
 3252 
 3253 void __init vfs_caches_init_early(void)
 3254 {
 3255 	int i;
 3256 
 3257 	for (i = 0; i < ARRAY_SIZE(in_lookup_hashtable); i++)
 3258 		INIT_HLIST_BL_HEAD(&in_lookup_hashtable[i]);
 3259 
 3260 	dcache_init_early();
 3261 	inode_init_early();
 3262 }
 3263 
 3264 void __init vfs_caches_init(void)
 3265 {
 3266 	names_cachep = kmem_cache_create_usercopy("names_cache", PATH_MAX, 0,
 3267 			SLAB_HWCACHE_ALIGN|SLAB_PANIC, 0, PATH_MAX, NULL);
 3268 
 3269 	dcache_init();
 3270 	inode_init();
 3271 	files_init();
 3272 	files_maxfiles_init();
 3273 	mnt_init();
 3274 	bdev_cache_init();
 3275 	chrdev_init();
 3276 }