4 * (c) 2015 - Jeff Layton <jeff.layton@primarydata.com>
7 #include <linux/hash.h>
8 #include <linux/slab.h>
9 #include <linux/file.h>
10 #include <linux/sched.h>
11 #include <linux/list_lru.h>
12 #include <linux/fsnotify_backend.h>
13 #include <linux/fsnotify.h>
14 #include <linux/seq_file.h>
20 #include "filecache.h"
23 #define NFSDDBG_FACILITY NFSDDBG_FH
25 /* FIXME: dynamically size this for the machine somehow? */
26 #define NFSD_FILE_HASH_BITS 12
27 #define NFSD_FILE_HASH_SIZE (1 << NFSD_FILE_HASH_BITS)
28 #define NFSD_LAUNDRETTE_DELAY (2 * HZ)
30 #define NFSD_FILE_LRU_RESCAN (0)
31 #define NFSD_FILE_SHUTDOWN (1)
32 #define NFSD_FILE_LRU_THRESHOLD (4096UL)
33 #define NFSD_FILE_LRU_LIMIT (NFSD_FILE_LRU_THRESHOLD << 2)
35 /* We only care about NFSD_MAY_READ/WRITE for this cache */
36 #define NFSD_FILE_MAY_MASK (NFSD_MAY_READ|NFSD_MAY_WRITE)
38 struct nfsd_fcache_bucket {
39 struct hlist_head nfb_head;
41 unsigned int nfb_count;
42 unsigned int nfb_maxcount;
45 static DEFINE_PER_CPU(unsigned long, nfsd_file_cache_hits);
47 struct nfsd_fcache_disposal {
48 struct list_head list;
49 struct work_struct work;
52 struct list_head freeme;
56 struct workqueue_struct *nfsd_filecache_wq __read_mostly;
58 static struct kmem_cache *nfsd_file_slab;
59 static struct kmem_cache *nfsd_file_mark_slab;
60 static struct nfsd_fcache_bucket *nfsd_file_hashtbl;
61 static struct list_lru nfsd_file_lru;
62 static long nfsd_file_lru_flags;
63 static struct fsnotify_group *nfsd_file_fsnotify_group;
64 static atomic_long_t nfsd_filecache_count;
65 static struct delayed_work nfsd_filecache_laundrette;
66 static DEFINE_SPINLOCK(laundrette_lock);
67 static LIST_HEAD(laundrettes);
69 static void nfsd_file_gc(void);
72 nfsd_file_schedule_laundrette(void)
74 long count = atomic_long_read(&nfsd_filecache_count);
76 if (count == 0 || test_bit(NFSD_FILE_SHUTDOWN, &nfsd_file_lru_flags))
79 queue_delayed_work(system_wq, &nfsd_filecache_laundrette,
80 NFSD_LAUNDRETTE_DELAY);
84 nfsd_file_slab_free(struct rcu_head *rcu)
86 struct nfsd_file *nf = container_of(rcu, struct nfsd_file, nf_rcu);
88 put_cred(nf->nf_cred);
89 kmem_cache_free(nfsd_file_slab, nf);
93 nfsd_file_mark_free(struct fsnotify_mark *mark)
95 struct nfsd_file_mark *nfm = container_of(mark, struct nfsd_file_mark,
98 kmem_cache_free(nfsd_file_mark_slab, nfm);
101 static struct nfsd_file_mark *
102 nfsd_file_mark_get(struct nfsd_file_mark *nfm)
104 if (!atomic_inc_not_zero(&nfm->nfm_ref))
110 nfsd_file_mark_put(struct nfsd_file_mark *nfm)
112 if (atomic_dec_and_test(&nfm->nfm_ref)) {
114 fsnotify_destroy_mark(&nfm->nfm_mark, nfsd_file_fsnotify_group);
115 fsnotify_put_mark(&nfm->nfm_mark);
119 static struct nfsd_file_mark *
120 nfsd_file_mark_find_or_create(struct nfsd_file *nf)
123 struct fsnotify_mark *mark;
124 struct nfsd_file_mark *nfm = NULL, *new;
125 struct inode *inode = nf->nf_inode;
128 mutex_lock(&nfsd_file_fsnotify_group->mark_mutex);
129 mark = fsnotify_find_mark(&inode->i_fsnotify_marks,
130 nfsd_file_fsnotify_group);
132 nfm = nfsd_file_mark_get(container_of(mark,
133 struct nfsd_file_mark,
135 mutex_unlock(&nfsd_file_fsnotify_group->mark_mutex);
137 fsnotify_put_mark(mark);
140 /* Avoid soft lockup race with nfsd_file_mark_put() */
141 fsnotify_destroy_mark(mark, nfsd_file_fsnotify_group);
142 fsnotify_put_mark(mark);
144 mutex_unlock(&nfsd_file_fsnotify_group->mark_mutex);
146 /* allocate a new nfm */
147 new = kmem_cache_alloc(nfsd_file_mark_slab, GFP_KERNEL);
150 fsnotify_init_mark(&new->nfm_mark, nfsd_file_fsnotify_group);
151 new->nfm_mark.mask = FS_ATTRIB|FS_DELETE_SELF;
152 atomic_set(&new->nfm_ref, 1);
154 err = fsnotify_add_inode_mark(&new->nfm_mark, inode, 0);
157 * If the add was successful, then return the object.
158 * Otherwise, we need to put the reference we hold on the
159 * nfm_mark. The fsnotify code will take a reference and put
160 * it on failure, so we can't just free it directly. It's also
161 * not safe to call fsnotify_destroy_mark on it as the
162 * mark->group will be NULL. Thus, we can't let the nfm_ref
163 * counter drive the destruction at this point.
168 fsnotify_put_mark(&new->nfm_mark);
169 } while (unlikely(err == -EEXIST));
174 static struct nfsd_file *
175 nfsd_file_alloc(struct inode *inode, unsigned int may, unsigned int hashval,
178 struct nfsd_file *nf;
180 nf = kmem_cache_alloc(nfsd_file_slab, GFP_KERNEL);
182 INIT_HLIST_NODE(&nf->nf_node);
183 INIT_LIST_HEAD(&nf->nf_lru);
185 nf->nf_cred = get_current_cred();
188 nf->nf_inode = inode;
189 nf->nf_hashval = hashval;
190 atomic_set(&nf->nf_ref, 1);
191 nf->nf_may = may & NFSD_FILE_MAY_MASK;
192 if (may & NFSD_MAY_NOT_BREAK_LEASE) {
193 if (may & NFSD_MAY_WRITE)
194 __set_bit(NFSD_FILE_BREAK_WRITE, &nf->nf_flags);
195 if (may & NFSD_MAY_READ)
196 __set_bit(NFSD_FILE_BREAK_READ, &nf->nf_flags);
199 trace_nfsd_file_alloc(nf);
205 nfsd_file_free(struct nfsd_file *nf)
209 trace_nfsd_file_put_final(nf);
211 nfsd_file_mark_put(nf->nf_mark);
213 get_file(nf->nf_file);
214 filp_close(nf->nf_file, NULL);
218 call_rcu(&nf->nf_rcu, nfsd_file_slab_free);
223 nfsd_file_check_writeback(struct nfsd_file *nf)
225 struct file *file = nf->nf_file;
226 struct address_space *mapping;
228 if (!file || !(file->f_mode & FMODE_WRITE))
230 mapping = file->f_mapping;
231 return mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) ||
232 mapping_tagged(mapping, PAGECACHE_TAG_WRITEBACK);
236 nfsd_file_check_write_error(struct nfsd_file *nf)
238 struct file *file = nf->nf_file;
240 if (!file || !(file->f_mode & FMODE_WRITE))
242 return filemap_check_wb_err(file->f_mapping, READ_ONCE(file->f_wb_err));
246 nfsd_file_in_use(struct nfsd_file *nf)
248 return nfsd_file_check_writeback(nf) ||
249 nfsd_file_check_write_error(nf);
253 nfsd_file_do_unhash(struct nfsd_file *nf)
255 lockdep_assert_held(&nfsd_file_hashtbl[nf->nf_hashval].nfb_lock);
257 trace_nfsd_file_unhash(nf);
259 if (nfsd_file_check_write_error(nf))
260 nfsd_reset_boot_verifier(net_generic(nf->nf_net, nfsd_net_id));
261 --nfsd_file_hashtbl[nf->nf_hashval].nfb_count;
262 hlist_del_rcu(&nf->nf_node);
263 atomic_long_dec(&nfsd_filecache_count);
267 nfsd_file_unhash(struct nfsd_file *nf)
269 if (test_and_clear_bit(NFSD_FILE_HASHED, &nf->nf_flags)) {
270 nfsd_file_do_unhash(nf);
271 if (!list_empty(&nf->nf_lru))
272 list_lru_del(&nfsd_file_lru, &nf->nf_lru);
279 * Return true if the file was unhashed.
282 nfsd_file_unhash_and_release_locked(struct nfsd_file *nf, struct list_head *dispose)
284 lockdep_assert_held(&nfsd_file_hashtbl[nf->nf_hashval].nfb_lock);
286 trace_nfsd_file_unhash_and_release_locked(nf);
287 if (!nfsd_file_unhash(nf))
289 /* keep final reference for nfsd_file_lru_dispose */
290 if (atomic_add_unless(&nf->nf_ref, -1, 1))
293 list_add(&nf->nf_lru, dispose);
298 nfsd_file_put_noref(struct nfsd_file *nf)
301 trace_nfsd_file_put(nf);
303 count = atomic_dec_return(&nf->nf_ref);
305 WARN_ON(test_bit(NFSD_FILE_HASHED, &nf->nf_flags));
312 nfsd_file_put(struct nfsd_file *nf)
314 bool is_hashed = test_bit(NFSD_FILE_HASHED, &nf->nf_flags) != 0;
315 bool unused = !nfsd_file_in_use(nf);
317 set_bit(NFSD_FILE_REFERENCED, &nf->nf_flags);
318 if (nfsd_file_put_noref(nf) == 1 && is_hashed && unused)
319 nfsd_file_schedule_laundrette();
320 if (atomic_long_read(&nfsd_filecache_count) >= NFSD_FILE_LRU_LIMIT)
325 nfsd_file_get(struct nfsd_file *nf)
327 if (likely(atomic_inc_not_zero(&nf->nf_ref)))
333 nfsd_file_dispose_list(struct list_head *dispose)
335 struct nfsd_file *nf;
337 while(!list_empty(dispose)) {
338 nf = list_first_entry(dispose, struct nfsd_file, nf_lru);
339 list_del(&nf->nf_lru);
340 nfsd_file_put_noref(nf);
345 nfsd_file_dispose_list_sync(struct list_head *dispose)
348 struct nfsd_file *nf;
350 while(!list_empty(dispose)) {
351 nf = list_first_entry(dispose, struct nfsd_file, nf_lru);
352 list_del(&nf->nf_lru);
353 if (!atomic_dec_and_test(&nf->nf_ref))
355 if (nfsd_file_free(nf))
359 flush_delayed_fput();
363 nfsd_file_list_remove_disposal(struct list_head *dst,
364 struct nfsd_fcache_disposal *l)
367 list_splice_init(&l->freeme, dst);
368 spin_unlock(&l->lock);
372 nfsd_file_list_add_disposal(struct list_head *files, struct net *net)
374 struct nfsd_fcache_disposal *l;
377 list_for_each_entry_rcu(l, &laundrettes, list) {
380 list_splice_tail_init(files, &l->freeme);
381 spin_unlock(&l->lock);
382 queue_work(nfsd_filecache_wq, &l->work);
390 nfsd_file_list_add_pernet(struct list_head *dst, struct list_head *src,
393 struct nfsd_file *nf, *tmp;
395 list_for_each_entry_safe(nf, tmp, src, nf_lru) {
396 if (nf->nf_net == net)
397 list_move_tail(&nf->nf_lru, dst);
402 nfsd_file_dispose_list_delayed(struct list_head *dispose)
405 struct nfsd_file *nf;
407 while(!list_empty(dispose)) {
408 nf = list_first_entry(dispose, struct nfsd_file, nf_lru);
409 nfsd_file_list_add_pernet(&list, dispose, nf->nf_net);
410 nfsd_file_list_add_disposal(&list, nf->nf_net);
415 * Note this can deadlock with nfsd_file_cache_purge.
417 static enum lru_status
418 nfsd_file_lru_cb(struct list_head *item, struct list_lru_one *lru,
419 spinlock_t *lock, void *arg)
423 struct list_head *head = arg;
424 struct nfsd_file *nf = list_entry(item, struct nfsd_file, nf_lru);
427 * Do a lockless refcount check. The hashtable holds one reference, so
428 * we look to see if anything else has a reference, or if any have
429 * been put since the shrinker last ran. Those don't get unhashed and
432 * Note that in the put path, we set the flag and then decrement the
433 * counter. Here we check the counter and then test and clear the flag.
434 * That order is deliberate to ensure that we can do this locklessly.
436 if (atomic_read(&nf->nf_ref) > 1)
440 * Don't throw out files that are still undergoing I/O or
441 * that have uncleared errors pending.
443 if (nfsd_file_check_writeback(nf))
446 if (test_and_clear_bit(NFSD_FILE_REFERENCED, &nf->nf_flags))
449 if (!test_and_clear_bit(NFSD_FILE_HASHED, &nf->nf_flags))
452 list_lru_isolate_move(lru, &nf->nf_lru, head);
455 set_bit(NFSD_FILE_LRU_RESCAN, &nfsd_file_lru_flags);
461 nfsd_file_lru_walk_list(struct shrink_control *sc)
464 struct nfsd_file *nf;
468 ret = list_lru_shrink_walk(&nfsd_file_lru, sc,
469 nfsd_file_lru_cb, &head);
471 ret = list_lru_walk(&nfsd_file_lru,
474 list_for_each_entry(nf, &head, nf_lru) {
475 spin_lock(&nfsd_file_hashtbl[nf->nf_hashval].nfb_lock);
476 nfsd_file_do_unhash(nf);
477 spin_unlock(&nfsd_file_hashtbl[nf->nf_hashval].nfb_lock);
479 nfsd_file_dispose_list_delayed(&head);
486 nfsd_file_lru_walk_list(NULL);
490 nfsd_file_gc_worker(struct work_struct *work)
493 nfsd_file_schedule_laundrette();
497 nfsd_file_lru_count(struct shrinker *s, struct shrink_control *sc)
499 return list_lru_count(&nfsd_file_lru);
503 nfsd_file_lru_scan(struct shrinker *s, struct shrink_control *sc)
505 return nfsd_file_lru_walk_list(sc);
508 static struct shrinker nfsd_file_shrinker = {
509 .scan_objects = nfsd_file_lru_scan,
510 .count_objects = nfsd_file_lru_count,
515 __nfsd_file_close_inode(struct inode *inode, unsigned int hashval,
516 struct list_head *dispose)
518 struct nfsd_file *nf;
519 struct hlist_node *tmp;
521 spin_lock(&nfsd_file_hashtbl[hashval].nfb_lock);
522 hlist_for_each_entry_safe(nf, tmp, &nfsd_file_hashtbl[hashval].nfb_head, nf_node) {
523 if (inode == nf->nf_inode)
524 nfsd_file_unhash_and_release_locked(nf, dispose);
526 spin_unlock(&nfsd_file_hashtbl[hashval].nfb_lock);
530 * nfsd_file_close_inode_sync - attempt to forcibly close a nfsd_file
531 * @inode: inode of the file to attempt to remove
533 * Walk the whole hash bucket, looking for any files that correspond to "inode".
534 * If any do, then unhash them and put the hashtable reference to them and
535 * destroy any that had their last reference put. Also ensure that any of the
536 * fputs also have their final __fput done as well.
539 nfsd_file_close_inode_sync(struct inode *inode)
541 unsigned int hashval = (unsigned int)hash_long(inode->i_ino,
542 NFSD_FILE_HASH_BITS);
545 __nfsd_file_close_inode(inode, hashval, &dispose);
546 trace_nfsd_file_close_inode_sync(inode, hashval, !list_empty(&dispose));
547 nfsd_file_dispose_list_sync(&dispose);
551 * nfsd_file_close_inode_sync - attempt to forcibly close a nfsd_file
552 * @inode: inode of the file to attempt to remove
554 * Walk the whole hash bucket, looking for any files that correspond to "inode".
555 * If any do, then unhash them and put the hashtable reference to them and
556 * destroy any that had their last reference put.
559 nfsd_file_close_inode(struct inode *inode)
561 unsigned int hashval = (unsigned int)hash_long(inode->i_ino,
562 NFSD_FILE_HASH_BITS);
565 __nfsd_file_close_inode(inode, hashval, &dispose);
566 trace_nfsd_file_close_inode(inode, hashval, !list_empty(&dispose));
567 nfsd_file_dispose_list_delayed(&dispose);
571 * nfsd_file_delayed_close - close unused nfsd_files
574 * Walk the LRU list and close any entries that have not been used since
577 * Note this can deadlock with nfsd_file_cache_purge.
580 nfsd_file_delayed_close(struct work_struct *work)
583 struct nfsd_fcache_disposal *l = container_of(work,
584 struct nfsd_fcache_disposal, work);
586 nfsd_file_list_remove_disposal(&head, l);
587 nfsd_file_dispose_list(&head);
591 nfsd_file_lease_notifier_call(struct notifier_block *nb, unsigned long arg,
594 struct file_lock *fl = data;
596 /* Only close files for F_SETLEASE leases */
597 if (fl->fl_flags & FL_LEASE)
598 nfsd_file_close_inode_sync(file_inode(fl->fl_file));
602 static struct notifier_block nfsd_file_lease_notifier = {
603 .notifier_call = nfsd_file_lease_notifier_call,
607 nfsd_file_fsnotify_handle_event(struct fsnotify_group *group,
609 u32 mask, const void *data, int data_type,
610 const struct qstr *file_name, u32 cookie,
611 struct fsnotify_iter_info *iter_info)
613 trace_nfsd_file_fsnotify_handle_event(inode, mask);
615 /* Should be no marks on non-regular files */
616 if (!S_ISREG(inode->i_mode)) {
621 /* don't close files if this was not the last link */
622 if (mask & FS_ATTRIB) {
627 nfsd_file_close_inode(inode);
632 static const struct fsnotify_ops nfsd_file_fsnotify_ops = {
633 .handle_event = nfsd_file_fsnotify_handle_event,
634 .free_mark = nfsd_file_mark_free,
638 nfsd_file_cache_init(void)
643 clear_bit(NFSD_FILE_SHUTDOWN, &nfsd_file_lru_flags);
645 if (nfsd_file_hashtbl)
648 nfsd_filecache_wq = alloc_workqueue("nfsd_filecache", 0, 0);
649 if (!nfsd_filecache_wq)
652 nfsd_file_hashtbl = kcalloc(NFSD_FILE_HASH_SIZE,
653 sizeof(*nfsd_file_hashtbl), GFP_KERNEL);
654 if (!nfsd_file_hashtbl) {
655 pr_err("nfsd: unable to allocate nfsd_file_hashtbl\n");
659 nfsd_file_slab = kmem_cache_create("nfsd_file",
660 sizeof(struct nfsd_file), 0, 0, NULL);
661 if (!nfsd_file_slab) {
662 pr_err("nfsd: unable to create nfsd_file_slab\n");
666 nfsd_file_mark_slab = kmem_cache_create("nfsd_file_mark",
667 sizeof(struct nfsd_file_mark), 0, 0, NULL);
668 if (!nfsd_file_mark_slab) {
669 pr_err("nfsd: unable to create nfsd_file_mark_slab\n");
674 ret = list_lru_init(&nfsd_file_lru);
676 pr_err("nfsd: failed to init nfsd_file_lru: %d\n", ret);
680 ret = register_shrinker(&nfsd_file_shrinker);
682 pr_err("nfsd: failed to register nfsd_file_shrinker: %d\n", ret);
686 ret = lease_register_notifier(&nfsd_file_lease_notifier);
688 pr_err("nfsd: unable to register lease notifier: %d\n", ret);
692 nfsd_file_fsnotify_group = fsnotify_alloc_group(&nfsd_file_fsnotify_ops);
693 if (IS_ERR(nfsd_file_fsnotify_group)) {
694 pr_err("nfsd: unable to create fsnotify group: %ld\n",
695 PTR_ERR(nfsd_file_fsnotify_group));
696 nfsd_file_fsnotify_group = NULL;
700 for (i = 0; i < NFSD_FILE_HASH_SIZE; i++) {
701 INIT_HLIST_HEAD(&nfsd_file_hashtbl[i].nfb_head);
702 spin_lock_init(&nfsd_file_hashtbl[i].nfb_lock);
705 INIT_DELAYED_WORK(&nfsd_filecache_laundrette, nfsd_file_gc_worker);
709 lease_unregister_notifier(&nfsd_file_lease_notifier);
711 unregister_shrinker(&nfsd_file_shrinker);
713 list_lru_destroy(&nfsd_file_lru);
715 kmem_cache_destroy(nfsd_file_slab);
716 nfsd_file_slab = NULL;
717 kmem_cache_destroy(nfsd_file_mark_slab);
718 nfsd_file_mark_slab = NULL;
719 kfree(nfsd_file_hashtbl);
720 nfsd_file_hashtbl = NULL;
721 destroy_workqueue(nfsd_filecache_wq);
722 nfsd_filecache_wq = NULL;
727 * Note this can deadlock with nfsd_file_lru_cb.
730 nfsd_file_cache_purge(struct net *net)
733 struct nfsd_file *nf;
734 struct hlist_node *next;
738 if (!nfsd_file_hashtbl)
741 for (i = 0; i < NFSD_FILE_HASH_SIZE; i++) {
742 struct nfsd_fcache_bucket *nfb = &nfsd_file_hashtbl[i];
744 spin_lock(&nfb->nfb_lock);
745 hlist_for_each_entry_safe(nf, next, &nfb->nfb_head, nf_node) {
746 if (net && nf->nf_net != net)
748 del = nfsd_file_unhash_and_release_locked(nf, &dispose);
751 * Deadlock detected! Something marked this entry as
752 * unhased, but hasn't removed it from the hash list.
756 spin_unlock(&nfb->nfb_lock);
757 nfsd_file_dispose_list(&dispose);
761 static struct nfsd_fcache_disposal *
762 nfsd_alloc_fcache_disposal(struct net *net)
764 struct nfsd_fcache_disposal *l;
766 l = kmalloc(sizeof(*l), GFP_KERNEL);
769 INIT_WORK(&l->work, nfsd_file_delayed_close);
771 spin_lock_init(&l->lock);
772 INIT_LIST_HEAD(&l->freeme);
777 nfsd_free_fcache_disposal(struct nfsd_fcache_disposal *l)
779 rcu_assign_pointer(l->net, NULL);
780 cancel_work_sync(&l->work);
781 nfsd_file_dispose_list(&l->freeme);
786 nfsd_add_fcache_disposal(struct nfsd_fcache_disposal *l)
788 spin_lock(&laundrette_lock);
789 list_add_tail_rcu(&l->list, &laundrettes);
790 spin_unlock(&laundrette_lock);
794 nfsd_del_fcache_disposal(struct nfsd_fcache_disposal *l)
796 spin_lock(&laundrette_lock);
797 list_del_rcu(&l->list);
798 spin_unlock(&laundrette_lock);
802 nfsd_alloc_fcache_disposal_net(struct net *net)
804 struct nfsd_fcache_disposal *l;
806 l = nfsd_alloc_fcache_disposal(net);
809 nfsd_add_fcache_disposal(l);
814 nfsd_free_fcache_disposal_net(struct net *net)
816 struct nfsd_fcache_disposal *l;
819 list_for_each_entry_rcu(l, &laundrettes, list) {
822 nfsd_del_fcache_disposal(l);
824 nfsd_free_fcache_disposal(l);
831 nfsd_file_cache_start_net(struct net *net)
833 return nfsd_alloc_fcache_disposal_net(net);
837 nfsd_file_cache_shutdown_net(struct net *net)
839 nfsd_file_cache_purge(net);
840 nfsd_free_fcache_disposal_net(net);
844 nfsd_file_cache_shutdown(void)
848 set_bit(NFSD_FILE_SHUTDOWN, &nfsd_file_lru_flags);
850 lease_unregister_notifier(&nfsd_file_lease_notifier);
851 unregister_shrinker(&nfsd_file_shrinker);
853 * make sure all callers of nfsd_file_lru_cb are done before
854 * calling nfsd_file_cache_purge
856 cancel_delayed_work_sync(&nfsd_filecache_laundrette);
857 nfsd_file_cache_purge(NULL);
858 list_lru_destroy(&nfsd_file_lru);
860 fsnotify_put_group(nfsd_file_fsnotify_group);
861 nfsd_file_fsnotify_group = NULL;
862 kmem_cache_destroy(nfsd_file_slab);
863 nfsd_file_slab = NULL;
864 fsnotify_wait_marks_destroyed();
865 kmem_cache_destroy(nfsd_file_mark_slab);
866 nfsd_file_mark_slab = NULL;
867 kfree(nfsd_file_hashtbl);
868 nfsd_file_hashtbl = NULL;
869 destroy_workqueue(nfsd_filecache_wq);
870 nfsd_filecache_wq = NULL;
874 nfsd_match_cred(const struct cred *c1, const struct cred *c2)
878 if (!uid_eq(c1->fsuid, c2->fsuid))
880 if (!gid_eq(c1->fsgid, c2->fsgid))
882 if (c1->group_info == NULL || c2->group_info == NULL)
883 return c1->group_info == c2->group_info;
884 if (c1->group_info->ngroups != c2->group_info->ngroups)
886 for (i = 0; i < c1->group_info->ngroups; i++) {
887 if (!gid_eq(c1->group_info->gid[i], c2->group_info->gid[i]))
893 static struct nfsd_file *
894 nfsd_file_find_locked(struct inode *inode, unsigned int may_flags,
895 unsigned int hashval, struct net *net)
897 struct nfsd_file *nf;
898 unsigned char need = may_flags & NFSD_FILE_MAY_MASK;
900 hlist_for_each_entry_rcu(nf, &nfsd_file_hashtbl[hashval].nfb_head,
902 if ((need & nf->nf_may) != need)
904 if (nf->nf_inode != inode)
906 if (nf->nf_net != net)
908 if (!nfsd_match_cred(nf->nf_cred, current_cred()))
910 if (!test_bit(NFSD_FILE_HASHED, &nf->nf_flags))
912 if (nfsd_file_get(nf) != NULL)
919 * nfsd_file_is_cached - are there any cached open files for this fh?
920 * @inode: inode of the file to check
922 * Scan the hashtable for open files that match this fh. Returns true if there
923 * are any, and false if not.
926 nfsd_file_is_cached(struct inode *inode)
929 struct nfsd_file *nf;
930 unsigned int hashval;
932 hashval = (unsigned int)hash_long(inode->i_ino, NFSD_FILE_HASH_BITS);
935 hlist_for_each_entry_rcu(nf, &nfsd_file_hashtbl[hashval].nfb_head,
937 if (inode == nf->nf_inode) {
943 trace_nfsd_file_is_cached(inode, hashval, (int)ret);
948 nfsd_file_acquire(struct svc_rqst *rqstp, struct svc_fh *fhp,
949 unsigned int may_flags, struct nfsd_file **pnf)
952 struct net *net = SVC_NET(rqstp);
953 struct nfsd_file *nf, *new;
955 unsigned int hashval;
958 /* FIXME: skip this if fh_dentry is already set? */
959 status = fh_verify(rqstp, fhp, S_IFREG,
960 may_flags|NFSD_MAY_OWNER_OVERRIDE);
961 if (status != nfs_ok)
964 inode = d_inode(fhp->fh_dentry);
965 hashval = (unsigned int)hash_long(inode->i_ino, NFSD_FILE_HASH_BITS);
968 nf = nfsd_file_find_locked(inode, may_flags, hashval, net);
971 goto wait_for_construction;
973 new = nfsd_file_alloc(inode, may_flags, hashval, net);
975 trace_nfsd_file_acquire(rqstp, hashval, inode, may_flags,
976 NULL, nfserr_jukebox);
977 return nfserr_jukebox;
980 spin_lock(&nfsd_file_hashtbl[hashval].nfb_lock);
981 nf = nfsd_file_find_locked(inode, may_flags, hashval, net);
984 spin_unlock(&nfsd_file_hashtbl[hashval].nfb_lock);
985 nfsd_file_slab_free(&new->nf_rcu);
987 wait_for_construction:
988 wait_on_bit(&nf->nf_flags, NFSD_FILE_PENDING, TASK_UNINTERRUPTIBLE);
990 /* Did construction of this file fail? */
991 if (!test_bit(NFSD_FILE_HASHED, &nf->nf_flags)) {
993 status = nfserr_jukebox;
997 nfsd_file_put_noref(nf);
1001 this_cpu_inc(nfsd_file_cache_hits);
1003 if (!(may_flags & NFSD_MAY_NOT_BREAK_LEASE)) {
1004 bool write = (may_flags & NFSD_MAY_WRITE);
1006 if (test_bit(NFSD_FILE_BREAK_READ, &nf->nf_flags) ||
1007 (test_bit(NFSD_FILE_BREAK_WRITE, &nf->nf_flags) && write)) {
1008 status = nfserrno(nfsd_open_break_lease(
1009 file_inode(nf->nf_file), may_flags));
1010 if (status == nfs_ok) {
1011 clear_bit(NFSD_FILE_BREAK_READ, &nf->nf_flags);
1013 clear_bit(NFSD_FILE_BREAK_WRITE,
1019 if (status == nfs_ok) {
1026 trace_nfsd_file_acquire(rqstp, hashval, inode, may_flags, nf, status);
1030 /* Take reference for the hashtable */
1031 atomic_inc(&nf->nf_ref);
1032 __set_bit(NFSD_FILE_HASHED, &nf->nf_flags);
1033 __set_bit(NFSD_FILE_PENDING, &nf->nf_flags);
1034 list_lru_add(&nfsd_file_lru, &nf->nf_lru);
1035 hlist_add_head_rcu(&nf->nf_node, &nfsd_file_hashtbl[hashval].nfb_head);
1036 ++nfsd_file_hashtbl[hashval].nfb_count;
1037 nfsd_file_hashtbl[hashval].nfb_maxcount = max(nfsd_file_hashtbl[hashval].nfb_maxcount,
1038 nfsd_file_hashtbl[hashval].nfb_count);
1039 spin_unlock(&nfsd_file_hashtbl[hashval].nfb_lock);
1040 if (atomic_long_inc_return(&nfsd_filecache_count) >= NFSD_FILE_LRU_THRESHOLD)
1043 nf->nf_mark = nfsd_file_mark_find_or_create(nf);
1045 status = nfsd_open_verified(rqstp, fhp, S_IFREG,
1046 may_flags, &nf->nf_file);
1048 status = nfserr_jukebox;
1050 * If construction failed, or we raced with a call to unlink()
1053 if (status != nfs_ok || inode->i_nlink == 0) {
1055 spin_lock(&nfsd_file_hashtbl[hashval].nfb_lock);
1056 do_free = nfsd_file_unhash(nf);
1057 spin_unlock(&nfsd_file_hashtbl[hashval].nfb_lock);
1059 nfsd_file_put_noref(nf);
1061 clear_bit_unlock(NFSD_FILE_PENDING, &nf->nf_flags);
1062 smp_mb__after_atomic();
1063 wake_up_bit(&nf->nf_flags, NFSD_FILE_PENDING);
1068 * Note that fields may be added, removed or reordered in the future. Programs
1069 * scraping this file for info should test the labels to ensure they're
1070 * getting the correct field.
1072 static int nfsd_file_cache_stats_show(struct seq_file *m, void *v)
1074 unsigned int i, count = 0, longest = 0;
1075 unsigned long hits = 0;
1078 * No need for spinlocks here since we're not terribly interested in
1079 * accuracy. We do take the nfsd_mutex simply to ensure that we
1080 * don't end up racing with server shutdown
1082 mutex_lock(&nfsd_mutex);
1083 if (nfsd_file_hashtbl) {
1084 for (i = 0; i < NFSD_FILE_HASH_SIZE; i++) {
1085 count += nfsd_file_hashtbl[i].nfb_count;
1086 longest = max(longest, nfsd_file_hashtbl[i].nfb_count);
1089 mutex_unlock(&nfsd_mutex);
1091 for_each_possible_cpu(i)
1092 hits += per_cpu(nfsd_file_cache_hits, i);
1094 seq_printf(m, "total entries: %u\n", count);
1095 seq_printf(m, "longest chain: %u\n", longest);
1096 seq_printf(m, "cache hits: %lu\n", hits);
1100 int nfsd_file_cache_stats_open(struct inode *inode, struct file *file)
1102 return single_open(file, nfsd_file_cache_stats_show, NULL);