Root/fs/inode.c

1/*
2 * linux/fs/inode.c
3 *
4 * (C) 1997 Linus Torvalds
5 */
6
7#include <linux/fs.h>
8#include <linux/mm.h>
9#include <linux/dcache.h>
10#include <linux/init.h>
11#include <linux/quotaops.h>
12#include <linux/slab.h>
13#include <linux/writeback.h>
14#include <linux/module.h>
15#include <linux/backing-dev.h>
16#include <linux/wait.h>
17#include <linux/hash.h>
18#include <linux/swap.h>
19#include <linux/security.h>
20#include <linux/ima.h>
21#include <linux/pagemap.h>
22#include <linux/cdev.h>
23#include <linux/bootmem.h>
24#include <linux/inotify.h>
25#include <linux/fsnotify.h>
26#include <linux/mount.h>
27#include <linux/async.h>
28#include <linux/posix_acl.h>
29
30/*
31 * This is needed for the following functions:
32 * - inode_has_buffers
33 * - invalidate_inode_buffers
34 * - invalidate_bdev
35 *
36 * FIXME: remove all knowledge of the buffer layer from this file
37 */
38#include <linux/buffer_head.h>
39
40/*
41 * New inode.c implementation.
42 *
43 * This implementation has the basic premise of trying
44 * to be extremely low-overhead and SMP-safe, yet be
45 * simple enough to be "obviously correct".
46 *
47 * Famous last words.
48 */
49
50/* inode dynamic allocation 1999, Andrea Arcangeli <andrea@suse.de> */
51
52/* #define INODE_PARANOIA 1 */
53/* #define INODE_DEBUG 1 */
54
55/*
56 * Inode lookup is no longer as critical as it used to be:
57 * most of the lookups are going to be through the dcache.
58 */
59#define I_HASHBITS i_hash_shift
60#define I_HASHMASK i_hash_mask
61
62static unsigned int i_hash_mask __read_mostly;
63static unsigned int i_hash_shift __read_mostly;
64
65/*
66 * Each inode can be on two separate lists. One is
67 * the hash list of the inode, used for lookups. The
68 * other linked list is the "type" list:
69 * "in_use" - valid inode, i_count > 0, i_nlink > 0
70 * "dirty" - as "in_use" but also dirty
71 * "unused" - valid inode, i_count = 0
72 *
73 * A "dirty" list is maintained for each super block,
74 * allowing for low-overhead inode sync() operations.
75 */
76
77LIST_HEAD(inode_in_use);
78LIST_HEAD(inode_unused);
79static struct hlist_head *inode_hashtable __read_mostly;
80
81/*
82 * A simple spinlock to protect the list manipulations.
83 *
84 * NOTE! You also have to own the lock if you change
85 * the i_state of an inode while it is in use..
86 */
87DEFINE_SPINLOCK(inode_lock);
88
89/*
90 * iprune_mutex provides exclusion between the kswapd or try_to_free_pages
91 * icache shrinking path, and the umount path. Without this exclusion,
92 * by the time prune_icache calls iput for the inode whose pages it has
93 * been invalidating, or by the time it calls clear_inode & destroy_inode
94 * from its final dispose_list, the struct super_block they refer to
95 * (for inode->i_sb->s_op) may already have been freed and reused.
96 */
97static DEFINE_MUTEX(iprune_mutex);
98
99/*
100 * Statistics gathering..
101 */
102struct inodes_stat_t inodes_stat;
103
104static struct kmem_cache *inode_cachep __read_mostly;
105
106static void wake_up_inode(struct inode *inode)
107{
108    /*
109     * Prevent speculative execution through spin_unlock(&inode_lock);
110     */
111    smp_mb();
112    wake_up_bit(&inode->i_state, __I_LOCK);
113}
114
115/**
116 * inode_init_always - perform inode structure intialisation
117 * @sb: superblock inode belongs to
118 * @inode: inode to initialise
119 *
120 * These are initializations that need to be done on every inode
121 * allocation as the fields are not initialised by slab allocation.
122 */
123int inode_init_always(struct super_block *sb, struct inode *inode)
124{
125    static const struct address_space_operations empty_aops;
126    static struct inode_operations empty_iops;
127    static const struct file_operations empty_fops;
128    struct address_space *const mapping = &inode->i_data;
129
130    inode->i_sb = sb;
131    inode->i_blkbits = sb->s_blocksize_bits;
132    inode->i_flags = 0;
133    atomic_set(&inode->i_count, 1);
134    inode->i_op = &empty_iops;
135    inode->i_fop = &empty_fops;
136    inode->i_nlink = 1;
137    inode->i_uid = 0;
138    inode->i_gid = 0;
139    atomic_set(&inode->i_writecount, 0);
140    inode->i_size = 0;
141    inode->i_blocks = 0;
142    inode->i_bytes = 0;
143    inode->i_generation = 0;
144#ifdef CONFIG_QUOTA
145    memset(&inode->i_dquot, 0, sizeof(inode->i_dquot));
146#endif
147    inode->i_pipe = NULL;
148    inode->i_bdev = NULL;
149    inode->i_cdev = NULL;
150    inode->i_rdev = 0;
151    inode->dirtied_when = 0;
152
153    if (security_inode_alloc(inode))
154        goto out;
155
156    /* allocate and initialize an i_integrity */
157    if (ima_inode_alloc(inode))
158        goto out_free_security;
159
160    spin_lock_init(&inode->i_lock);
161    lockdep_set_class(&inode->i_lock, &sb->s_type->i_lock_key);
162
163    mutex_init(&inode->i_mutex);
164    lockdep_set_class(&inode->i_mutex, &sb->s_type->i_mutex_key);
165
166    init_rwsem(&inode->i_alloc_sem);
167    lockdep_set_class(&inode->i_alloc_sem, &sb->s_type->i_alloc_sem_key);
168
169    mapping->a_ops = &empty_aops;
170    mapping->host = inode;
171    mapping->flags = 0;
172    mapping_set_gfp_mask(mapping, GFP_HIGHUSER_MOVABLE);
173    mapping->assoc_mapping = NULL;
174    mapping->backing_dev_info = &default_backing_dev_info;
175    mapping->writeback_index = 0;
176
177    /*
178     * If the block_device provides a backing_dev_info for client
179     * inodes then use that. Otherwise the inode share the bdev's
180     * backing_dev_info.
181     */
182    if (sb->s_bdev) {
183        struct backing_dev_info *bdi;
184
185        bdi = sb->s_bdev->bd_inode_backing_dev_info;
186        if (!bdi)
187            bdi = sb->s_bdev->bd_inode->i_mapping->backing_dev_info;
188        mapping->backing_dev_info = bdi;
189    }
190    inode->i_private = NULL;
191    inode->i_mapping = mapping;
192#ifdef CONFIG_FS_POSIX_ACL
193    inode->i_acl = inode->i_default_acl = ACL_NOT_CACHED;
194#endif
195
196#ifdef CONFIG_FSNOTIFY
197    inode->i_fsnotify_mask = 0;
198#endif
199
200    return 0;
201
202out_free_security:
203    security_inode_free(inode);
204out:
205    return -ENOMEM;
206}
207EXPORT_SYMBOL(inode_init_always);
208
209static struct inode *alloc_inode(struct super_block *sb)
210{
211    struct inode *inode;
212
213    if (sb->s_op->alloc_inode)
214        inode = sb->s_op->alloc_inode(sb);
215    else
216        inode = kmem_cache_alloc(inode_cachep, GFP_KERNEL);
217
218    if (!inode)
219        return NULL;
220
221    if (unlikely(inode_init_always(sb, inode))) {
222        if (inode->i_sb->s_op->destroy_inode)
223            inode->i_sb->s_op->destroy_inode(inode);
224        else
225            kmem_cache_free(inode_cachep, inode);
226        return NULL;
227    }
228
229    return inode;
230}
231
232void __destroy_inode(struct inode *inode)
233{
234    BUG_ON(inode_has_buffers(inode));
235    ima_inode_free(inode);
236    security_inode_free(inode);
237    fsnotify_inode_delete(inode);
238#ifdef CONFIG_FS_POSIX_ACL
239    if (inode->i_acl && inode->i_acl != ACL_NOT_CACHED)
240        posix_acl_release(inode->i_acl);
241    if (inode->i_default_acl && inode->i_default_acl != ACL_NOT_CACHED)
242        posix_acl_release(inode->i_default_acl);
243#endif
244}
245EXPORT_SYMBOL(__destroy_inode);
246
247void destroy_inode(struct inode *inode)
248{
249    __destroy_inode(inode);
250    if (inode->i_sb->s_op->destroy_inode)
251        inode->i_sb->s_op->destroy_inode(inode);
252    else
253        kmem_cache_free(inode_cachep, (inode));
254}
255
256/*
257 * These are initializations that only need to be done
258 * once, because the fields are idempotent across use
259 * of the inode, so let the slab aware of that.
260 */
261void inode_init_once(struct inode *inode)
262{
263    memset(inode, 0, sizeof(*inode));
264    INIT_HLIST_NODE(&inode->i_hash);
265    INIT_LIST_HEAD(&inode->i_dentry);
266    INIT_LIST_HEAD(&inode->i_devices);
267    INIT_RADIX_TREE(&inode->i_data.page_tree, GFP_ATOMIC);
268    spin_lock_init(&inode->i_data.tree_lock);
269    spin_lock_init(&inode->i_data.i_mmap_lock);
270    INIT_LIST_HEAD(&inode->i_data.private_list);
271    spin_lock_init(&inode->i_data.private_lock);
272    INIT_RAW_PRIO_TREE_ROOT(&inode->i_data.i_mmap);
273    INIT_LIST_HEAD(&inode->i_data.i_mmap_nonlinear);
274    i_size_ordered_init(inode);
275#ifdef CONFIG_INOTIFY
276    INIT_LIST_HEAD(&inode->inotify_watches);
277    mutex_init(&inode->inotify_mutex);
278#endif
279#ifdef CONFIG_FSNOTIFY
280    INIT_HLIST_HEAD(&inode->i_fsnotify_mark_entries);
281#endif
282}
283EXPORT_SYMBOL(inode_init_once);
284
285static void init_once(void *foo)
286{
287    struct inode *inode = (struct inode *) foo;
288
289    inode_init_once(inode);
290}
291
292/*
293 * inode_lock must be held
294 */
295void __iget(struct inode *inode)
296{
297    if (atomic_read(&inode->i_count)) {
298        atomic_inc(&inode->i_count);
299        return;
300    }
301    atomic_inc(&inode->i_count);
302    if (!(inode->i_state & (I_DIRTY|I_SYNC)))
303        list_move(&inode->i_list, &inode_in_use);
304    inodes_stat.nr_unused--;
305}
306
307/**
308 * clear_inode - clear an inode
309 * @inode: inode to clear
310 *
311 * This is called by the filesystem to tell us
312 * that the inode is no longer useful. We just
313 * terminate it with extreme prejudice.
314 */
315void clear_inode(struct inode *inode)
316{
317    might_sleep();
318    invalidate_inode_buffers(inode);
319
320    BUG_ON(inode->i_data.nrpages);
321    BUG_ON(!(inode->i_state & I_FREEING));
322    BUG_ON(inode->i_state & I_CLEAR);
323    inode_sync_wait(inode);
324    vfs_dq_drop(inode);
325    if (inode->i_sb->s_op->clear_inode)
326        inode->i_sb->s_op->clear_inode(inode);
327    if (S_ISBLK(inode->i_mode) && inode->i_bdev)
328        bd_forget(inode);
329    if (S_ISCHR(inode->i_mode) && inode->i_cdev)
330        cd_forget(inode);
331    inode->i_state = I_CLEAR;
332}
333EXPORT_SYMBOL(clear_inode);
334
335/*
336 * dispose_list - dispose of the contents of a local list
337 * @head: the head of the list to free
338 *
339 * Dispose-list gets a local list with local inodes in it, so it doesn't
340 * need to worry about list corruption and SMP locks.
341 */
342static void dispose_list(struct list_head *head)
343{
344    int nr_disposed = 0;
345
346    while (!list_empty(head)) {
347        struct inode *inode;
348
349        inode = list_first_entry(head, struct inode, i_list);
350        list_del(&inode->i_list);
351
352        if (inode->i_data.nrpages)
353            truncate_inode_pages(&inode->i_data, 0);
354        clear_inode(inode);
355
356        spin_lock(&inode_lock);
357        hlist_del_init(&inode->i_hash);
358        list_del_init(&inode->i_sb_list);
359        spin_unlock(&inode_lock);
360
361        wake_up_inode(inode);
362        destroy_inode(inode);
363        nr_disposed++;
364    }
365    spin_lock(&inode_lock);
366    inodes_stat.nr_inodes -= nr_disposed;
367    spin_unlock(&inode_lock);
368}
369
370/*
371 * Invalidate all inodes for a device.
372 */
373static int invalidate_list(struct list_head *head, struct list_head *dispose)
374{
375    struct list_head *next;
376    int busy = 0, count = 0;
377
378    next = head->next;
379    for (;;) {
380        struct list_head *tmp = next;
381        struct inode *inode;
382
383        /*
384         * We can reschedule here without worrying about the list's
385         * consistency because the per-sb list of inodes must not
386         * change during umount anymore, and because iprune_mutex keeps
387         * shrink_icache_memory() away.
388         */
389        cond_resched_lock(&inode_lock);
390
391        next = next->next;
392        if (tmp == head)
393            break;
394        inode = list_entry(tmp, struct inode, i_sb_list);
395        if (inode->i_state & I_NEW)
396            continue;
397        invalidate_inode_buffers(inode);
398        if (!atomic_read(&inode->i_count)) {
399            list_move(&inode->i_list, dispose);
400            WARN_ON(inode->i_state & I_NEW);
401            inode->i_state |= I_FREEING;
402            count++;
403            continue;
404        }
405        busy = 1;
406    }
407    /* only unused inodes may be cached with i_count zero */
408    inodes_stat.nr_unused -= count;
409    return busy;
410}
411
412/**
413 * invalidate_inodes - discard the inodes on a device
414 * @sb: superblock
415 *
416 * Discard all of the inodes for a given superblock. If the discard
417 * fails because there are busy inodes then a non zero value is returned.
418 * If the discard is successful all the inodes have been discarded.
419 */
420int invalidate_inodes(struct super_block *sb)
421{
422    int busy;
423    LIST_HEAD(throw_away);
424
425    mutex_lock(&iprune_mutex);
426    spin_lock(&inode_lock);
427    inotify_unmount_inodes(&sb->s_inodes);
428    fsnotify_unmount_inodes(&sb->s_inodes);
429    busy = invalidate_list(&sb->s_inodes, &throw_away);
430    spin_unlock(&inode_lock);
431
432    dispose_list(&throw_away);
433    mutex_unlock(&iprune_mutex);
434
435    return busy;
436}
437EXPORT_SYMBOL(invalidate_inodes);
438
439static int can_unuse(struct inode *inode)
440{
441    if (inode->i_state)
442        return 0;
443    if (inode_has_buffers(inode))
444        return 0;
445    if (atomic_read(&inode->i_count))
446        return 0;
447    if (inode->i_data.nrpages)
448        return 0;
449    return 1;
450}
451
452/*
453 * Scan `goal' inodes on the unused list for freeable ones. They are moved to
454 * a temporary list and then are freed outside inode_lock by dispose_list().
455 *
456 * Any inodes which are pinned purely because of attached pagecache have their
457 * pagecache removed. We expect the final iput() on that inode to add it to
458 * the front of the inode_unused list. So look for it there and if the
459 * inode is still freeable, proceed. The right inode is found 99.9% of the
460 * time in testing on a 4-way.
461 *
462 * If the inode has metadata buffers attached to mapping->private_list then
463 * try to remove them.
464 */
465static void prune_icache(int nr_to_scan)
466{
467    LIST_HEAD(freeable);
468    int nr_pruned = 0;
469    int nr_scanned;
470    unsigned long reap = 0;
471
472    mutex_lock(&iprune_mutex);
473    spin_lock(&inode_lock);
474    for (nr_scanned = 0; nr_scanned < nr_to_scan; nr_scanned++) {
475        struct inode *inode;
476
477        if (list_empty(&inode_unused))
478            break;
479
480        inode = list_entry(inode_unused.prev, struct inode, i_list);
481
482        if (inode->i_state || atomic_read(&inode->i_count)) {
483            list_move(&inode->i_list, &inode_unused);
484            continue;
485        }
486        if (inode_has_buffers(inode) || inode->i_data.nrpages) {
487            __iget(inode);
488            spin_unlock(&inode_lock);
489            if (remove_inode_buffers(inode))
490                reap += invalidate_mapping_pages(&inode->i_data,
491                                0, -1);
492            iput(inode);
493            spin_lock(&inode_lock);
494
495            if (inode != list_entry(inode_unused.next,
496                        struct inode, i_list))
497                continue; /* wrong inode or list_empty */
498            if (!can_unuse(inode))
499                continue;
500        }
501        list_move(&inode->i_list, &freeable);
502        WARN_ON(inode->i_state & I_NEW);
503        inode->i_state |= I_FREEING;
504        nr_pruned++;
505    }
506    inodes_stat.nr_unused -= nr_pruned;
507    if (current_is_kswapd())
508        __count_vm_events(KSWAPD_INODESTEAL, reap);
509    else
510        __count_vm_events(PGINODESTEAL, reap);
511    spin_unlock(&inode_lock);
512
513    dispose_list(&freeable);
514    mutex_unlock(&iprune_mutex);
515}
516
517/*
518 * shrink_icache_memory() will attempt to reclaim some unused inodes. Here,
519 * "unused" means that no dentries are referring to the inodes: the files are
520 * not open and the dcache references to those inodes have already been
521 * reclaimed.
522 *
523 * This function is passed the number of inodes to scan, and it returns the
524 * total number of remaining possibly-reclaimable inodes.
525 */
526static int shrink_icache_memory(int nr, gfp_t gfp_mask)
527{
528    if (nr) {
529        /*
530         * Nasty deadlock avoidance. We may hold various FS locks,
531         * and we don't want to recurse into the FS that called us
532         * in clear_inode() and friends..
533         */
534        if (!(gfp_mask & __GFP_FS))
535            return -1;
536        prune_icache(nr);
537    }
538    return (inodes_stat.nr_unused / 100) * sysctl_vfs_cache_pressure;
539}
540
541static struct shrinker icache_shrinker = {
542    .shrink = shrink_icache_memory,
543    .seeks = DEFAULT_SEEKS,
544};
545
546static void __wait_on_freeing_inode(struct inode *inode);
547/*
548 * Called with the inode lock held.
549 * NOTE: we are not increasing the inode-refcount, you must call __iget()
550 * by hand after calling find_inode now! This simplifies iunique and won't
551 * add any additional branch in the common code.
552 */
553static struct inode *find_inode(struct super_block *sb,
554                struct hlist_head *head,
555                int (*test)(struct inode *, void *),
556                void *data)
557{
558    struct hlist_node *node;
559    struct inode *inode = NULL;
560
561repeat:
562    hlist_for_each_entry(inode, node, head, i_hash) {
563        if (inode->i_sb != sb)
564            continue;
565        if (!test(inode, data))
566            continue;
567        if (inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE)) {
568            __wait_on_freeing_inode(inode);
569            goto repeat;
570        }
571        break;
572    }
573    return node ? inode : NULL;
574}
575
576/*
577 * find_inode_fast is the fast path version of find_inode, see the comment at
578 * iget_locked for details.
579 */
580static struct inode *find_inode_fast(struct super_block *sb,
581                struct hlist_head *head, unsigned long ino)
582{
583    struct hlist_node *node;
584    struct inode *inode = NULL;
585
586repeat:
587    hlist_for_each_entry(inode, node, head, i_hash) {
588        if (inode->i_ino != ino)
589            continue;
590        if (inode->i_sb != sb)
591            continue;
592        if (inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE)) {
593            __wait_on_freeing_inode(inode);
594            goto repeat;
595        }
596        break;
597    }
598    return node ? inode : NULL;
599}
600
601static unsigned long hash(struct super_block *sb, unsigned long hashval)
602{
603    unsigned long tmp;
604
605    tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
606            L1_CACHE_BYTES;
607    tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> I_HASHBITS);
608    return tmp & I_HASHMASK;
609}
610
611static inline void
612__inode_add_to_lists(struct super_block *sb, struct hlist_head *head,
613            struct inode *inode)
614{
615    inodes_stat.nr_inodes++;
616    list_add(&inode->i_list, &inode_in_use);
617    list_add(&inode->i_sb_list, &sb->s_inodes);
618    if (head)
619        hlist_add_head(&inode->i_hash, head);
620}
621
622/**
623 * inode_add_to_lists - add a new inode to relevant lists
624 * @sb: superblock inode belongs to
625 * @inode: inode to mark in use
626 *
627 * When an inode is allocated it needs to be accounted for, added to the in use
628 * list, the owning superblock and the inode hash. This needs to be done under
629 * the inode_lock, so export a function to do this rather than the inode lock
630 * itself. We calculate the hash list to add to here so it is all internal
631 * which requires the caller to have already set up the inode number in the
632 * inode to add.
633 */
634void inode_add_to_lists(struct super_block *sb, struct inode *inode)
635{
636    struct hlist_head *head = inode_hashtable + hash(sb, inode->i_ino);
637
638    spin_lock(&inode_lock);
639    __inode_add_to_lists(sb, head, inode);
640    spin_unlock(&inode_lock);
641}
642EXPORT_SYMBOL_GPL(inode_add_to_lists);
643
644/**
645 * new_inode - obtain an inode
646 * @sb: superblock
647 *
648 * Allocates a new inode for given superblock. The default gfp_mask
649 * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE.
650 * If HIGHMEM pages are unsuitable or it is known that pages allocated
651 * for the page cache are not reclaimable or migratable,
652 * mapping_set_gfp_mask() must be called with suitable flags on the
653 * newly created inode's mapping
654 *
655 */
656struct inode *new_inode(struct super_block *sb)
657{
658    /*
659     * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
660     * error if st_ino won't fit in target struct field. Use 32bit counter
661     * here to attempt to avoid that.
662     */
663    static unsigned int last_ino;
664    struct inode *inode;
665
666    spin_lock_prefetch(&inode_lock);
667
668    inode = alloc_inode(sb);
669    if (inode) {
670        spin_lock(&inode_lock);
671        __inode_add_to_lists(sb, NULL, inode);
672        inode->i_ino = ++last_ino;
673        inode->i_state = 0;
674        spin_unlock(&inode_lock);
675    }
676    return inode;
677}
678EXPORT_SYMBOL(new_inode);
679
680void unlock_new_inode(struct inode *inode)
681{
682#ifdef CONFIG_DEBUG_LOCK_ALLOC
683    if (inode->i_mode & S_IFDIR) {
684        struct file_system_type *type = inode->i_sb->s_type;
685
686        /* Set new key only if filesystem hasn't already changed it */
687        if (!lockdep_match_class(&inode->i_mutex,
688            &type->i_mutex_key)) {
689            /*
690             * ensure nobody is actually holding i_mutex
691             */
692            mutex_destroy(&inode->i_mutex);
693            mutex_init(&inode->i_mutex);
694            lockdep_set_class(&inode->i_mutex,
695                      &type->i_mutex_dir_key);
696        }
697    }
698#endif
699    /*
700     * This is special! We do not need the spinlock
701     * when clearing I_LOCK, because we're guaranteed
702     * that nobody else tries to do anything about the
703     * state of the inode when it is locked, as we
704     * just created it (so there can be no old holders
705     * that haven't tested I_LOCK).
706     */
707    WARN_ON((inode->i_state & (I_LOCK|I_NEW)) != (I_LOCK|I_NEW));
708    inode->i_state &= ~(I_LOCK|I_NEW);
709    wake_up_inode(inode);
710}
711EXPORT_SYMBOL(unlock_new_inode);
712
713/*
714 * This is called without the inode lock held.. Be careful.
715 *
716 * We no longer cache the sb_flags in i_flags - see fs.h
717 * -- rmk@arm.uk.linux.org
718 */
719static struct inode *get_new_inode(struct super_block *sb,
720                struct hlist_head *head,
721                int (*test)(struct inode *, void *),
722                int (*set)(struct inode *, void *),
723                void *data)
724{
725    struct inode *inode;
726
727    inode = alloc_inode(sb);
728    if (inode) {
729        struct inode *old;
730
731        spin_lock(&inode_lock);
732        /* We released the lock, so.. */
733        old = find_inode(sb, head, test, data);
734        if (!old) {
735            if (set(inode, data))
736                goto set_failed;
737
738            __inode_add_to_lists(sb, head, inode);
739            inode->i_state = I_LOCK|I_NEW;
740            spin_unlock(&inode_lock);
741
742            /* Return the locked inode with I_NEW set, the
743             * caller is responsible for filling in the contents
744             */
745            return inode;
746        }
747
748        /*
749         * Uhhuh, somebody else created the same inode under
750         * us. Use the old inode instead of the one we just
751         * allocated.
752         */
753        __iget(old);
754        spin_unlock(&inode_lock);
755        destroy_inode(inode);
756        inode = old;
757        wait_on_inode(inode);
758    }
759    return inode;
760
761set_failed:
762    spin_unlock(&inode_lock);
763    destroy_inode(inode);
764    return NULL;
765}
766
767/*
768 * get_new_inode_fast is the fast path version of get_new_inode, see the
769 * comment at iget_locked for details.
770 */
771static struct inode *get_new_inode_fast(struct super_block *sb,
772                struct hlist_head *head, unsigned long ino)
773{
774    struct inode *inode;
775
776    inode = alloc_inode(sb);
777    if (inode) {
778        struct inode *old;
779
780        spin_lock(&inode_lock);
781        /* We released the lock, so.. */
782        old = find_inode_fast(sb, head, ino);
783        if (!old) {
784            inode->i_ino = ino;
785            __inode_add_to_lists(sb, head, inode);
786            inode->i_state = I_LOCK|I_NEW;
787            spin_unlock(&inode_lock);
788
789            /* Return the locked inode with I_NEW set, the
790             * caller is responsible for filling in the contents
791             */
792            return inode;
793        }
794
795        /*
796         * Uhhuh, somebody else created the same inode under
797         * us. Use the old inode instead of the one we just
798         * allocated.
799         */
800        __iget(old);
801        spin_unlock(&inode_lock);
802        destroy_inode(inode);
803        inode = old;
804        wait_on_inode(inode);
805    }
806    return inode;
807}
808
809/**
810 * iunique - get a unique inode number
811 * @sb: superblock
812 * @max_reserved: highest reserved inode number
813 *
814 * Obtain an inode number that is unique on the system for a given
815 * superblock. This is used by file systems that have no natural
816 * permanent inode numbering system. An inode number is returned that
817 * is higher than the reserved limit but unique.
818 *
819 * BUGS:
820 * With a large number of inodes live on the file system this function
821 * currently becomes quite slow.
822 */
823ino_t iunique(struct super_block *sb, ino_t max_reserved)
824{
825    /*
826     * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
827     * error if st_ino won't fit in target struct field. Use 32bit counter
828     * here to attempt to avoid that.
829     */
830    static unsigned int counter;
831    struct inode *inode;
832    struct hlist_head *head;
833    ino_t res;
834
835    spin_lock(&inode_lock);
836    do {
837        if (counter <= max_reserved)
838            counter = max_reserved + 1;
839        res = counter++;
840        head = inode_hashtable + hash(sb, res);
841        inode = find_inode_fast(sb, head, res);
842    } while (inode != NULL);
843    spin_unlock(&inode_lock);
844
845    return res;
846}
847EXPORT_SYMBOL(iunique);
848
849struct inode *igrab(struct inode *inode)
850{
851    spin_lock(&inode_lock);
852    if (!(inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE)))
853        __iget(inode);
854    else
855        /*
856         * Handle the case where s_op->clear_inode is not been
857         * called yet, and somebody is calling igrab
858         * while the inode is getting freed.
859         */
860        inode = NULL;
861    spin_unlock(&inode_lock);
862    return inode;
863}
864EXPORT_SYMBOL(igrab);
865
866/**
867 * ifind - internal function, you want ilookup5() or iget5().
868 * @sb: super block of file system to search
869 * @head: the head of the list to search
870 * @test: callback used for comparisons between inodes
871 * @data: opaque data pointer to pass to @test
872 * @wait: if true wait for the inode to be unlocked, if false do not
873 *
874 * ifind() searches for the inode specified by @data in the inode
875 * cache. This is a generalized version of ifind_fast() for file systems where
876 * the inode number is not sufficient for unique identification of an inode.
877 *
878 * If the inode is in the cache, the inode is returned with an incremented
879 * reference count.
880 *
881 * Otherwise NULL is returned.
882 *
883 * Note, @test is called with the inode_lock held, so can't sleep.
884 */
885static struct inode *ifind(struct super_block *sb,
886        struct hlist_head *head, int (*test)(struct inode *, void *),
887        void *data, const int wait)
888{
889    struct inode *inode;
890
891    spin_lock(&inode_lock);
892    inode = find_inode(sb, head, test, data);
893    if (inode) {
894        __iget(inode);
895        spin_unlock(&inode_lock);
896        if (likely(wait))
897            wait_on_inode(inode);
898        return inode;
899    }
900    spin_unlock(&inode_lock);
901    return NULL;
902}
903
904/**
905 * ifind_fast - internal function, you want ilookup() or iget().
906 * @sb: super block of file system to search
907 * @head: head of the list to search
908 * @ino: inode number to search for
909 *
910 * ifind_fast() searches for the inode @ino in the inode cache. This is for
911 * file systems where the inode number is sufficient for unique identification
912 * of an inode.
913 *
914 * If the inode is in the cache, the inode is returned with an incremented
915 * reference count.
916 *
917 * Otherwise NULL is returned.
918 */
919static struct inode *ifind_fast(struct super_block *sb,
920        struct hlist_head *head, unsigned long ino)
921{
922    struct inode *inode;
923
924    spin_lock(&inode_lock);
925    inode = find_inode_fast(sb, head, ino);
926    if (inode) {
927        __iget(inode);
928        spin_unlock(&inode_lock);
929        wait_on_inode(inode);
930        return inode;
931    }
932    spin_unlock(&inode_lock);
933    return NULL;
934}
935
936/**
937 * ilookup5_nowait - search for an inode in the inode cache
938 * @sb: super block of file system to search
939 * @hashval: hash value (usually inode number) to search for
940 * @test: callback used for comparisons between inodes
941 * @data: opaque data pointer to pass to @test
942 *
943 * ilookup5() uses ifind() to search for the inode specified by @hashval and
944 * @data in the inode cache. This is a generalized version of ilookup() for
945 * file systems where the inode number is not sufficient for unique
946 * identification of an inode.
947 *
948 * If the inode is in the cache, the inode is returned with an incremented
949 * reference count. Note, the inode lock is not waited upon so you have to be
950 * very careful what you do with the returned inode. You probably should be
951 * using ilookup5() instead.
952 *
953 * Otherwise NULL is returned.
954 *
955 * Note, @test is called with the inode_lock held, so can't sleep.
956 */
957struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval,
958        int (*test)(struct inode *, void *), void *data)
959{
960    struct hlist_head *head = inode_hashtable + hash(sb, hashval);
961
962    return ifind(sb, head, test, data, 0);
963}
964EXPORT_SYMBOL(ilookup5_nowait);
965
966/**
967 * ilookup5 - search for an inode in the inode cache
968 * @sb: super block of file system to search
969 * @hashval: hash value (usually inode number) to search for
970 * @test: callback used for comparisons between inodes
971 * @data: opaque data pointer to pass to @test
972 *
973 * ilookup5() uses ifind() to search for the inode specified by @hashval and
974 * @data in the inode cache. This is a generalized version of ilookup() for
975 * file systems where the inode number is not sufficient for unique
976 * identification of an inode.
977 *
978 * If the inode is in the cache, the inode lock is waited upon and the inode is
979 * returned with an incremented reference count.
980 *
981 * Otherwise NULL is returned.
982 *
983 * Note, @test is called with the inode_lock held, so can't sleep.
984 */
985struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
986        int (*test)(struct inode *, void *), void *data)
987{
988    struct hlist_head *head = inode_hashtable + hash(sb, hashval);
989
990    return ifind(sb, head, test, data, 1);
991}
992EXPORT_SYMBOL(ilookup5);
993
994/**
995 * ilookup - search for an inode in the inode cache
996 * @sb: super block of file system to search
997 * @ino: inode number to search for
998 *
999 * ilookup() uses ifind_fast() to search for the inode @ino in the inode cache.
1000 * This is for file systems where the inode number is sufficient for unique
1001 * identification of an inode.
1002 *
1003 * If the inode is in the cache, the inode is returned with an incremented
1004 * reference count.
1005 *
1006 * Otherwise NULL is returned.
1007 */
1008struct inode *ilookup(struct super_block *sb, unsigned long ino)
1009{
1010    struct hlist_head *head = inode_hashtable + hash(sb, ino);
1011
1012    return ifind_fast(sb, head, ino);
1013}
1014EXPORT_SYMBOL(ilookup);
1015
1016/**
1017 * iget5_locked - obtain an inode from a mounted file system
1018 * @sb: super block of file system
1019 * @hashval: hash value (usually inode number) to get
1020 * @test: callback used for comparisons between inodes
1021 * @set: callback used to initialize a new struct inode
1022 * @data: opaque data pointer to pass to @test and @set
1023 *
1024 * iget5_locked() uses ifind() to search for the inode specified by @hashval
1025 * and @data in the inode cache and if present it is returned with an increased
1026 * reference count. This is a generalized version of iget_locked() for file
1027 * systems where the inode number is not sufficient for unique identification
1028 * of an inode.
1029 *
1030 * If the inode is not in cache, get_new_inode() is called to allocate a new
1031 * inode and this is returned locked, hashed, and with the I_NEW flag set. The
1032 * file system gets to fill it in before unlocking it via unlock_new_inode().
1033 *
1034 * Note both @test and @set are called with the inode_lock held, so can't sleep.
1035 */
1036struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
1037        int (*test)(struct inode *, void *),
1038        int (*set)(struct inode *, void *), void *data)
1039{
1040    struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1041    struct inode *inode;
1042
1043    inode = ifind(sb, head, test, data, 1);
1044    if (inode)
1045        return inode;
1046    /*
1047     * get_new_inode() will do the right thing, re-trying the search
1048     * in case it had to block at any point.
1049     */
1050    return get_new_inode(sb, head, test, set, data);
1051}
1052EXPORT_SYMBOL(iget5_locked);
1053
1054/**
1055 * iget_locked - obtain an inode from a mounted file system
1056 * @sb: super block of file system
1057 * @ino: inode number to get
1058 *
1059 * iget_locked() uses ifind_fast() to search for the inode specified by @ino in
1060 * the inode cache and if present it is returned with an increased reference
1061 * count. This is for file systems where the inode number is sufficient for
1062 * unique identification of an inode.
1063 *
1064 * If the inode is not in cache, get_new_inode_fast() is called to allocate a
1065 * new inode and this is returned locked, hashed, and with the I_NEW flag set.
1066 * The file system gets to fill it in before unlocking it via
1067 * unlock_new_inode().
1068 */
1069struct inode *iget_locked(struct super_block *sb, unsigned long ino)
1070{
1071    struct hlist_head *head = inode_hashtable + hash(sb, ino);
1072    struct inode *inode;
1073
1074    inode = ifind_fast(sb, head, ino);
1075    if (inode)
1076        return inode;
1077    /*
1078     * get_new_inode_fast() will do the right thing, re-trying the search
1079     * in case it had to block at any point.
1080     */
1081    return get_new_inode_fast(sb, head, ino);
1082}
1083EXPORT_SYMBOL(iget_locked);
1084
1085int insert_inode_locked(struct inode *inode)
1086{
1087    struct super_block *sb = inode->i_sb;
1088    ino_t ino = inode->i_ino;
1089    struct hlist_head *head = inode_hashtable + hash(sb, ino);
1090
1091    inode->i_state |= I_LOCK|I_NEW;
1092    while (1) {
1093        struct hlist_node *node;
1094        struct inode *old = NULL;
1095        spin_lock(&inode_lock);
1096        hlist_for_each_entry(old, node, head, i_hash) {
1097            if (old->i_ino != ino)
1098                continue;
1099            if (old->i_sb != sb)
1100                continue;
1101            if (old->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE))
1102                continue;
1103            break;
1104        }
1105        if (likely(!node)) {
1106            hlist_add_head(&inode->i_hash, head);
1107            spin_unlock(&inode_lock);
1108            return 0;
1109        }
1110        __iget(old);
1111        spin_unlock(&inode_lock);
1112        wait_on_inode(old);
1113        if (unlikely(!hlist_unhashed(&old->i_hash))) {
1114            iput(old);
1115            return -EBUSY;
1116        }
1117        iput(old);
1118    }
1119}
1120EXPORT_SYMBOL(insert_inode_locked);
1121
1122int insert_inode_locked4(struct inode *inode, unsigned long hashval,
1123        int (*test)(struct inode *, void *), void *data)
1124{
1125    struct super_block *sb = inode->i_sb;
1126    struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1127
1128    inode->i_state |= I_LOCK|I_NEW;
1129
1130    while (1) {
1131        struct hlist_node *node;
1132        struct inode *old = NULL;
1133
1134        spin_lock(&inode_lock);
1135        hlist_for_each_entry(old, node, head, i_hash) {
1136            if (old->i_sb != sb)
1137                continue;
1138            if (!test(old, data))
1139                continue;
1140            if (old->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE))
1141                continue;
1142            break;
1143        }
1144        if (likely(!node)) {
1145            hlist_add_head(&inode->i_hash, head);
1146            spin_unlock(&inode_lock);
1147            return 0;
1148        }
1149        __iget(old);
1150        spin_unlock(&inode_lock);
1151        wait_on_inode(old);
1152        if (unlikely(!hlist_unhashed(&old->i_hash))) {
1153            iput(old);
1154            return -EBUSY;
1155        }
1156        iput(old);
1157    }
1158}
1159EXPORT_SYMBOL(insert_inode_locked4);
1160
1161/**
1162 * __insert_inode_hash - hash an inode
1163 * @inode: unhashed inode
1164 * @hashval: unsigned long value used to locate this object in the
1165 * inode_hashtable.
1166 *
1167 * Add an inode to the inode hash for this superblock.
1168 */
1169void __insert_inode_hash(struct inode *inode, unsigned long hashval)
1170{
1171    struct hlist_head *head = inode_hashtable + hash(inode->i_sb, hashval);
1172    spin_lock(&inode_lock);
1173    hlist_add_head(&inode->i_hash, head);
1174    spin_unlock(&inode_lock);
1175}
1176EXPORT_SYMBOL(__insert_inode_hash);
1177
1178/**
1179 * remove_inode_hash - remove an inode from the hash
1180 * @inode: inode to unhash
1181 *
1182 * Remove an inode from the superblock.
1183 */
1184void remove_inode_hash(struct inode *inode)
1185{
1186    spin_lock(&inode_lock);
1187    hlist_del_init(&inode->i_hash);
1188    spin_unlock(&inode_lock);
1189}
1190EXPORT_SYMBOL(remove_inode_hash);
1191
1192/*
1193 * Tell the filesystem that this inode is no longer of any interest and should
1194 * be completely destroyed.
1195 *
1196 * We leave the inode in the inode hash table until *after* the filesystem's
1197 * ->delete_inode completes. This ensures that an iget (such as nfsd might
1198 * instigate) will always find up-to-date information either in the hash or on
1199 * disk.
1200 *
1201 * I_FREEING is set so that no-one will take a new reference to the inode while
1202 * it is being deleted.
1203 */
1204void generic_delete_inode(struct inode *inode)
1205{
1206    const struct super_operations *op = inode->i_sb->s_op;
1207
1208    list_del_init(&inode->i_list);
1209    list_del_init(&inode->i_sb_list);
1210    WARN_ON(inode->i_state & I_NEW);
1211    inode->i_state |= I_FREEING;
1212    inodes_stat.nr_inodes--;
1213    spin_unlock(&inode_lock);
1214
1215    security_inode_delete(inode);
1216
1217    if (op->delete_inode) {
1218        void (*delete)(struct inode *) = op->delete_inode;
1219        if (!is_bad_inode(inode))
1220            vfs_dq_init(inode);
1221        /* Filesystems implementing their own
1222         * s_op->delete_inode are required to call
1223         * truncate_inode_pages and clear_inode()
1224         * internally */
1225        delete(inode);
1226    } else {
1227        truncate_inode_pages(&inode->i_data, 0);
1228        clear_inode(inode);
1229    }
1230    spin_lock(&inode_lock);
1231    hlist_del_init(&inode->i_hash);
1232    spin_unlock(&inode_lock);
1233    wake_up_inode(inode);
1234    BUG_ON(inode->i_state != I_CLEAR);
1235    destroy_inode(inode);
1236}
1237EXPORT_SYMBOL(generic_delete_inode);
1238
1239static void generic_forget_inode(struct inode *inode)
1240{
1241    struct super_block *sb = inode->i_sb;
1242
1243    if (!hlist_unhashed(&inode->i_hash)) {
1244        if (!(inode->i_state & (I_DIRTY|I_SYNC)))
1245            list_move(&inode->i_list, &inode_unused);
1246        inodes_stat.nr_unused++;
1247        if (sb->s_flags & MS_ACTIVE) {
1248            spin_unlock(&inode_lock);
1249            return;
1250        }
1251        WARN_ON(inode->i_state & I_NEW);
1252        inode->i_state |= I_WILL_FREE;
1253        spin_unlock(&inode_lock);
1254        write_inode_now(inode, 1);
1255        spin_lock(&inode_lock);
1256        WARN_ON(inode->i_state & I_NEW);
1257        inode->i_state &= ~I_WILL_FREE;
1258        inodes_stat.nr_unused--;
1259        hlist_del_init(&inode->i_hash);
1260    }
1261    list_del_init(&inode->i_list);
1262    list_del_init(&inode->i_sb_list);
1263    WARN_ON(inode->i_state & I_NEW);
1264    inode->i_state |= I_FREEING;
1265    inodes_stat.nr_inodes--;
1266    spin_unlock(&inode_lock);
1267    if (inode->i_data.nrpages)
1268        truncate_inode_pages(&inode->i_data, 0);
1269    clear_inode(inode);
1270    wake_up_inode(inode);
1271    destroy_inode(inode);
1272}
1273
1274/*
1275 * Normal UNIX filesystem behaviour: delete the
1276 * inode when the usage count drops to zero, and
1277 * i_nlink is zero.
1278 */
1279void generic_drop_inode(struct inode *inode)
1280{
1281    if (!inode->i_nlink)
1282        generic_delete_inode(inode);
1283    else
1284        generic_forget_inode(inode);
1285}
1286EXPORT_SYMBOL_GPL(generic_drop_inode);
1287
1288/*
1289 * Called when we're dropping the last reference
1290 * to an inode.
1291 *
1292 * Call the FS "drop()" function, defaulting to
1293 * the legacy UNIX filesystem behaviour..
1294 *
1295 * NOTE! NOTE! NOTE! We're called with the inode lock
1296 * held, and the drop function is supposed to release
1297 * the lock!
1298 */
1299static inline void iput_final(struct inode *inode)
1300{
1301    const struct super_operations *op = inode->i_sb->s_op;
1302    void (*drop)(struct inode *) = generic_drop_inode;
1303
1304    if (op && op->drop_inode)
1305        drop = op->drop_inode;
1306    drop(inode);
1307}
1308
1309/**
1310 * iput - put an inode
1311 * @inode: inode to put
1312 *
1313 * Puts an inode, dropping its usage count. If the inode use count hits
1314 * zero, the inode is then freed and may also be destroyed.
1315 *
1316 * Consequently, iput() can sleep.
1317 */
1318void iput(struct inode *inode)
1319{
1320    if (inode) {
1321        BUG_ON(inode->i_state == I_CLEAR);
1322
1323        if (atomic_dec_and_lock(&inode->i_count, &inode_lock))
1324            iput_final(inode);
1325    }
1326}
1327EXPORT_SYMBOL(iput);
1328
1329/**
1330 * bmap - find a block number in a file
1331 * @inode: inode of file
1332 * @block: block to find
1333 *
1334 * Returns the block number on the device holding the inode that
1335 * is the disk block number for the block of the file requested.
1336 * That is, asked for block 4 of inode 1 the function will return the
1337 * disk block relative to the disk start that holds that block of the
1338 * file.
1339 */
1340sector_t bmap(struct inode *inode, sector_t block)
1341{
1342    sector_t res = 0;
1343    if (inode->i_mapping->a_ops->bmap)
1344        res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block);
1345    return res;
1346}
1347EXPORT_SYMBOL(bmap);
1348
1349/*
1350 * With relative atime, only update atime if the previous atime is
1351 * earlier than either the ctime or mtime or if at least a day has
1352 * passed since the last atime update.
1353 */
1354static int relatime_need_update(struct vfsmount *mnt, struct inode *inode,
1355                 struct timespec now)
1356{
1357
1358    if (!(mnt->mnt_flags & MNT_RELATIME))
1359        return 1;
1360    /*
1361     * Is mtime younger than atime? If yes, update atime:
1362     */
1363    if (timespec_compare(&inode->i_mtime, &inode->i_atime) >= 0)
1364        return 1;
1365    /*
1366     * Is ctime younger than atime? If yes, update atime:
1367     */
1368    if (timespec_compare(&inode->i_ctime, &inode->i_atime) >= 0)
1369        return 1;
1370
1371    /*
1372     * Is the previous atime value older than a day? If yes,
1373     * update atime:
1374     */
1375    if ((long)(now.tv_sec - inode->i_atime.tv_sec) >= 24*60*60)
1376        return 1;
1377    /*
1378     * Good, we can skip the atime update:
1379     */
1380    return 0;
1381}
1382
1383/**
1384 * touch_atime - update the access time
1385 * @mnt: mount the inode is accessed on
1386 * @dentry: dentry accessed
1387 *
1388 * Update the accessed time on an inode and mark it for writeback.
1389 * This function automatically handles read only file systems and media,
1390 * as well as the "noatime" flag and inode specific "noatime" markers.
1391 */
1392void touch_atime(struct vfsmount *mnt, struct dentry *dentry)
1393{
1394    struct inode *inode = dentry->d_inode;
1395    struct timespec now;
1396
1397    if (mnt_want_write(mnt))
1398        return;
1399    if (inode->i_flags & S_NOATIME)
1400        goto out;
1401    if (IS_NOATIME(inode))
1402        goto out;
1403    if ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode))
1404        goto out;
1405
1406    if (mnt->mnt_flags & MNT_NOATIME)
1407        goto out;
1408    if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
1409        goto out;
1410
1411    now = current_fs_time(inode->i_sb);
1412
1413    if (!relatime_need_update(mnt, inode, now))
1414        goto out;
1415
1416    if (timespec_equal(&inode->i_atime, &now))
1417        goto out;
1418
1419    inode->i_atime = now;
1420    mark_inode_dirty_sync(inode);
1421out:
1422    mnt_drop_write(mnt);
1423}
1424EXPORT_SYMBOL(touch_atime);
1425
1426/**
1427 * file_update_time - update mtime and ctime time
1428 * @file: file accessed
1429 *
1430 * Update the mtime and ctime members of an inode and mark the inode
1431 * for writeback. Note that this function is meant exclusively for
1432 * usage in the file write path of filesystems, and filesystems may
1433 * choose to explicitly ignore update via this function with the
1434 * S_NOCMTIME inode flag, e.g. for network filesystem where these
1435 * timestamps are handled by the server.
1436 */
1437
1438void file_update_time(struct file *file)
1439{
1440    struct inode *inode = file->f_path.dentry->d_inode;
1441    struct timespec now;
1442    int sync_it = 0;
1443    int err;
1444
1445    if (IS_NOCMTIME(inode))
1446        return;
1447
1448    err = mnt_want_write_file(file);
1449    if (err)
1450        return;
1451
1452    now = current_fs_time(inode->i_sb);
1453    if (!timespec_equal(&inode->i_mtime, &now)) {
1454        inode->i_mtime = now;
1455        sync_it = 1;
1456    }
1457
1458    if (!timespec_equal(&inode->i_ctime, &now)) {
1459        inode->i_ctime = now;
1460        sync_it = 1;
1461    }
1462
1463    if (IS_I_VERSION(inode)) {
1464        inode_inc_iversion(inode);
1465        sync_it = 1;
1466    }
1467
1468    if (sync_it)
1469        mark_inode_dirty_sync(inode);
1470    mnt_drop_write(file->f_path.mnt);
1471}
1472EXPORT_SYMBOL(file_update_time);
1473
1474int inode_needs_sync(struct inode *inode)
1475{
1476    if (IS_SYNC(inode))
1477        return 1;
1478    if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
1479        return 1;
1480    return 0;
1481}
1482EXPORT_SYMBOL(inode_needs_sync);
1483
1484int inode_wait(void *word)
1485{
1486    schedule();
1487    return 0;
1488}
1489EXPORT_SYMBOL(inode_wait);
1490
1491/*
1492 * If we try to find an inode in the inode hash while it is being
1493 * deleted, we have to wait until the filesystem completes its
1494 * deletion before reporting that it isn't found. This function waits
1495 * until the deletion _might_ have completed. Callers are responsible
1496 * to recheck inode state.
1497 *
1498 * It doesn't matter if I_LOCK is not set initially, a call to
1499 * wake_up_inode() after removing from the hash list will DTRT.
1500 *
1501 * This is called with inode_lock held.
1502 */
1503static void __wait_on_freeing_inode(struct inode *inode)
1504{
1505    wait_queue_head_t *wq;
1506    DEFINE_WAIT_BIT(wait, &inode->i_state, __I_LOCK);
1507    wq = bit_waitqueue(&inode->i_state, __I_LOCK);
1508    prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
1509    spin_unlock(&inode_lock);
1510    schedule();
1511    finish_wait(wq, &wait.wait);
1512    spin_lock(&inode_lock);
1513}
1514
1515static __initdata unsigned long ihash_entries;
1516static int __init set_ihash_entries(char *str)
1517{
1518    if (!str)
1519        return 0;
1520    ihash_entries = simple_strtoul(str, &str, 0);
1521    return 1;
1522}
1523__setup("ihash_entries=", set_ihash_entries);
1524
1525/*
1526 * Initialize the waitqueues and inode hash table.
1527 */
1528void __init inode_init_early(void)
1529{
1530    int loop;
1531
1532    /* If hashes are distributed across NUMA nodes, defer
1533     * hash allocation until vmalloc space is available.
1534     */
1535    if (hashdist)
1536        return;
1537
1538    inode_hashtable =
1539        alloc_large_system_hash("Inode-cache",
1540                    sizeof(struct hlist_head),
1541                    ihash_entries,
1542                    14,
1543                    HASH_EARLY,
1544                    &i_hash_shift,
1545                    &i_hash_mask,
1546                    0);
1547
1548    for (loop = 0; loop < (1 << i_hash_shift); loop++)
1549        INIT_HLIST_HEAD(&inode_hashtable[loop]);
1550}
1551
1552void __init inode_init(void)
1553{
1554    int loop;
1555
1556    /* inode slab cache */
1557    inode_cachep = kmem_cache_create("inode_cache",
1558                     sizeof(struct inode),
1559                     0,
1560                     (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
1561                     SLAB_MEM_SPREAD),
1562                     init_once);
1563    register_shrinker(&icache_shrinker);
1564
1565    /* Hash may have been set up in inode_init_early */
1566    if (!hashdist)
1567        return;
1568
1569    inode_hashtable =
1570        alloc_large_system_hash("Inode-cache",
1571                    sizeof(struct hlist_head),
1572                    ihash_entries,
1573                    14,
1574                    0,
1575                    &i_hash_shift,
1576                    &i_hash_mask,
1577                    0);
1578
1579    for (loop = 0; loop < (1 << i_hash_shift); loop++)
1580        INIT_HLIST_HEAD(&inode_hashtable[loop]);
1581}
1582
1583void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
1584{
1585    inode->i_mode = mode;
1586    if (S_ISCHR(mode)) {
1587        inode->i_fop = &def_chr_fops;
1588        inode->i_rdev = rdev;
1589    } else if (S_ISBLK(mode)) {
1590        inode->i_fop = &def_blk_fops;
1591        inode->i_rdev = rdev;
1592    } else if (S_ISFIFO(mode))
1593        inode->i_fop = &def_fifo_fops;
1594    else if (S_ISSOCK(mode))
1595        inode->i_fop = &bad_sock_fops;
1596    else
1597        printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o)\n",
1598               mode);
1599}
1600EXPORT_SYMBOL(init_special_inode);
1601

Archive Download this file



interactive