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_SHUTDOWN (1)
31 #define NFSD_FILE_LRU_THRESHOLD (4096UL)
32 #define NFSD_FILE_LRU_LIMIT (NFSD_FILE_LRU_THRESHOLD << 2)
34 /* We only care about NFSD_MAY_READ/WRITE for this cache */
35 #define NFSD_FILE_MAY_MASK (NFSD_MAY_READ|NFSD_MAY_WRITE)
37 struct nfsd_fcache_bucket {
38 struct hlist_head nfb_head;
40 unsigned int nfb_count;
41 unsigned int nfb_maxcount;
44 static DEFINE_PER_CPU(unsigned long, nfsd_file_cache_hits);
46 struct nfsd_fcache_disposal {
47 struct list_head list;
48 struct work_struct work;
51 struct list_head freeme;
55 static struct workqueue_struct *nfsd_filecache_wq __read_mostly;
57 static struct kmem_cache *nfsd_file_slab;
58 static struct kmem_cache *nfsd_file_mark_slab;
59 static struct nfsd_fcache_bucket *nfsd_file_hashtbl;
60 static struct list_lru nfsd_file_lru;
61 static long nfsd_file_lru_flags;
62 static struct fsnotify_group *nfsd_file_fsnotify_group;
63 static atomic_long_t nfsd_filecache_count;
64 static struct delayed_work nfsd_filecache_laundrette;
65 static DEFINE_SPINLOCK(laundrette_lock);
66 static LIST_HEAD(laundrettes);
68 static void nfsd_file_gc(void);
71 nfsd_file_schedule_laundrette(void)
73 long count = atomic_long_read(&nfsd_filecache_count);
75 if (count == 0 || test_bit(NFSD_FILE_SHUTDOWN, &nfsd_file_lru_flags))
78 queue_delayed_work(system_wq, &nfsd_filecache_laundrette,
79 NFSD_LAUNDRETTE_DELAY);
83 nfsd_file_slab_free(struct rcu_head *rcu)
85 struct nfsd_file *nf = container_of(rcu, struct nfsd_file, nf_rcu);
87 put_cred(nf->nf_cred);
88 kmem_cache_free(nfsd_file_slab, nf);
92 nfsd_file_mark_free(struct fsnotify_mark *mark)
94 struct nfsd_file_mark *nfm = container_of(mark, struct nfsd_file_mark,
97 kmem_cache_free(nfsd_file_mark_slab, nfm);
100 static struct nfsd_file_mark *
101 nfsd_file_mark_get(struct nfsd_file_mark *nfm)
103 if (!refcount_inc_not_zero(&nfm->nfm_ref))
109 nfsd_file_mark_put(struct nfsd_file_mark *nfm)
111 if (refcount_dec_and_test(&nfm->nfm_ref)) {
112 fsnotify_destroy_mark(&nfm->nfm_mark, nfsd_file_fsnotify_group);
113 fsnotify_put_mark(&nfm->nfm_mark);
117 static struct nfsd_file_mark *
118 nfsd_file_mark_find_or_create(struct nfsd_file *nf)
121 struct fsnotify_mark *mark;
122 struct nfsd_file_mark *nfm = NULL, *new;
123 struct inode *inode = nf->nf_inode;
126 mutex_lock(&nfsd_file_fsnotify_group->mark_mutex);
127 mark = fsnotify_find_mark(&inode->i_fsnotify_marks,
128 nfsd_file_fsnotify_group);
130 nfm = nfsd_file_mark_get(container_of(mark,
131 struct nfsd_file_mark,
133 mutex_unlock(&nfsd_file_fsnotify_group->mark_mutex);
135 fsnotify_put_mark(mark);
138 /* Avoid soft lockup race with nfsd_file_mark_put() */
139 fsnotify_destroy_mark(mark, nfsd_file_fsnotify_group);
140 fsnotify_put_mark(mark);
142 mutex_unlock(&nfsd_file_fsnotify_group->mark_mutex);
144 /* allocate a new nfm */
145 new = kmem_cache_alloc(nfsd_file_mark_slab, GFP_KERNEL);
148 fsnotify_init_mark(&new->nfm_mark, nfsd_file_fsnotify_group);
149 new->nfm_mark.mask = FS_ATTRIB|FS_DELETE_SELF;
150 refcount_set(&new->nfm_ref, 1);
152 err = fsnotify_add_inode_mark(&new->nfm_mark, inode, 0);
155 * If the add was successful, then return the object.
156 * Otherwise, we need to put the reference we hold on the
157 * nfm_mark. The fsnotify code will take a reference and put
158 * it on failure, so we can't just free it directly. It's also
159 * not safe to call fsnotify_destroy_mark on it as the
160 * mark->group will be NULL. Thus, we can't let the nfm_ref
161 * counter drive the destruction at this point.
166 fsnotify_put_mark(&new->nfm_mark);
167 } while (unlikely(err == -EEXIST));
172 static struct nfsd_file *
173 nfsd_file_alloc(struct inode *inode, unsigned int may, unsigned int hashval,
176 struct nfsd_file *nf;
178 nf = kmem_cache_alloc(nfsd_file_slab, GFP_KERNEL);
180 INIT_HLIST_NODE(&nf->nf_node);
181 INIT_LIST_HEAD(&nf->nf_lru);
183 nf->nf_cred = get_current_cred();
186 nf->nf_inode = inode;
187 nf->nf_hashval = hashval;
188 refcount_set(&nf->nf_ref, 1);
189 nf->nf_may = may & NFSD_FILE_MAY_MASK;
191 trace_nfsd_file_alloc(nf);
197 nfsd_file_free(struct nfsd_file *nf)
201 trace_nfsd_file_put_final(nf);
203 nfsd_file_mark_put(nf->nf_mark);
205 get_file(nf->nf_file);
206 filp_close(nf->nf_file, NULL);
210 call_rcu(&nf->nf_rcu, nfsd_file_slab_free);
215 nfsd_file_check_writeback(struct nfsd_file *nf)
217 struct file *file = nf->nf_file;
218 struct address_space *mapping;
220 if (!file || !(file->f_mode & FMODE_WRITE))
222 mapping = file->f_mapping;
223 return mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) ||
224 mapping_tagged(mapping, PAGECACHE_TAG_WRITEBACK);
228 nfsd_file_check_write_error(struct nfsd_file *nf)
230 struct file *file = nf->nf_file;
232 if (!file || !(file->f_mode & FMODE_WRITE))
234 return filemap_check_wb_err(file->f_mapping, READ_ONCE(file->f_wb_err));
238 nfsd_file_do_unhash(struct nfsd_file *nf)
240 lockdep_assert_held(&nfsd_file_hashtbl[nf->nf_hashval].nfb_lock);
242 trace_nfsd_file_unhash(nf);
244 if (nfsd_file_check_write_error(nf))
245 nfsd_reset_boot_verifier(net_generic(nf->nf_net, nfsd_net_id));
246 --nfsd_file_hashtbl[nf->nf_hashval].nfb_count;
247 hlist_del_rcu(&nf->nf_node);
248 atomic_long_dec(&nfsd_filecache_count);
252 nfsd_file_unhash(struct nfsd_file *nf)
254 if (test_and_clear_bit(NFSD_FILE_HASHED, &nf->nf_flags)) {
255 nfsd_file_do_unhash(nf);
256 if (!list_empty(&nf->nf_lru))
257 list_lru_del(&nfsd_file_lru, &nf->nf_lru);
264 * Return true if the file was unhashed.
267 nfsd_file_unhash_and_release_locked(struct nfsd_file *nf, struct list_head *dispose)
269 lockdep_assert_held(&nfsd_file_hashtbl[nf->nf_hashval].nfb_lock);
271 trace_nfsd_file_unhash_and_release_locked(nf);
272 if (!nfsd_file_unhash(nf))
274 /* keep final reference for nfsd_file_lru_dispose */
275 if (refcount_dec_not_one(&nf->nf_ref))
278 list_add(&nf->nf_lru, dispose);
283 nfsd_file_put_noref(struct nfsd_file *nf)
285 trace_nfsd_file_put(nf);
287 if (refcount_dec_and_test(&nf->nf_ref)) {
288 WARN_ON(test_bit(NFSD_FILE_HASHED, &nf->nf_flags));
294 nfsd_file_put(struct nfsd_file *nf)
298 set_bit(NFSD_FILE_REFERENCED, &nf->nf_flags);
299 if (refcount_read(&nf->nf_ref) > 2 || !nf->nf_file) {
300 nfsd_file_put_noref(nf);
304 filemap_flush(nf->nf_file->f_mapping);
305 is_hashed = test_bit(NFSD_FILE_HASHED, &nf->nf_flags) != 0;
306 nfsd_file_put_noref(nf);
308 nfsd_file_schedule_laundrette();
309 if (atomic_long_read(&nfsd_filecache_count) >= NFSD_FILE_LRU_LIMIT)
314 nfsd_file_get(struct nfsd_file *nf)
316 if (likely(refcount_inc_not_zero(&nf->nf_ref)))
322 nfsd_file_dispose_list(struct list_head *dispose)
324 struct nfsd_file *nf;
326 while(!list_empty(dispose)) {
327 nf = list_first_entry(dispose, struct nfsd_file, nf_lru);
328 list_del(&nf->nf_lru);
329 nfsd_file_put_noref(nf);
334 nfsd_file_dispose_list_sync(struct list_head *dispose)
337 struct nfsd_file *nf;
339 while(!list_empty(dispose)) {
340 nf = list_first_entry(dispose, struct nfsd_file, nf_lru);
341 list_del(&nf->nf_lru);
342 if (!refcount_dec_and_test(&nf->nf_ref))
344 if (nfsd_file_free(nf))
348 flush_delayed_fput();
352 nfsd_file_list_remove_disposal(struct list_head *dst,
353 struct nfsd_fcache_disposal *l)
356 list_splice_init(&l->freeme, dst);
357 spin_unlock(&l->lock);
361 nfsd_file_list_add_disposal(struct list_head *files, struct net *net)
363 struct nfsd_fcache_disposal *l;
366 list_for_each_entry_rcu(l, &laundrettes, list) {
369 list_splice_tail_init(files, &l->freeme);
370 spin_unlock(&l->lock);
371 queue_work(nfsd_filecache_wq, &l->work);
379 nfsd_file_list_add_pernet(struct list_head *dst, struct list_head *src,
382 struct nfsd_file *nf, *tmp;
384 list_for_each_entry_safe(nf, tmp, src, nf_lru) {
385 if (nf->nf_net == net)
386 list_move_tail(&nf->nf_lru, dst);
391 nfsd_file_dispose_list_delayed(struct list_head *dispose)
394 struct nfsd_file *nf;
396 while(!list_empty(dispose)) {
397 nf = list_first_entry(dispose, struct nfsd_file, nf_lru);
398 nfsd_file_list_add_pernet(&list, dispose, nf->nf_net);
399 nfsd_file_list_add_disposal(&list, nf->nf_net);
404 * Note this can deadlock with nfsd_file_cache_purge.
406 static enum lru_status
407 nfsd_file_lru_cb(struct list_head *item, struct list_lru_one *lru,
408 spinlock_t *lock, void *arg)
412 struct list_head *head = arg;
413 struct nfsd_file *nf = list_entry(item, struct nfsd_file, nf_lru);
416 * Do a lockless refcount check. The hashtable holds one reference, so
417 * we look to see if anything else has a reference, or if any have
418 * been put since the shrinker last ran. Those don't get unhashed and
421 * Note that in the put path, we set the flag and then decrement the
422 * counter. Here we check the counter and then test and clear the flag.
423 * That order is deliberate to ensure that we can do this locklessly.
425 if (refcount_read(&nf->nf_ref) > 1)
429 * Don't throw out files that are still undergoing I/O or
430 * that have uncleared errors pending.
432 if (nfsd_file_check_writeback(nf))
435 if (test_and_clear_bit(NFSD_FILE_REFERENCED, &nf->nf_flags))
438 if (!test_and_clear_bit(NFSD_FILE_HASHED, &nf->nf_flags))
441 list_lru_isolate_move(lru, &nf->nf_lru, head);
448 nfsd_file_lru_walk_list(struct shrink_control *sc)
451 struct nfsd_file *nf;
455 ret = list_lru_shrink_walk(&nfsd_file_lru, sc,
456 nfsd_file_lru_cb, &head);
458 ret = list_lru_walk(&nfsd_file_lru,
461 list_for_each_entry(nf, &head, nf_lru) {
462 spin_lock(&nfsd_file_hashtbl[nf->nf_hashval].nfb_lock);
463 nfsd_file_do_unhash(nf);
464 spin_unlock(&nfsd_file_hashtbl[nf->nf_hashval].nfb_lock);
466 nfsd_file_dispose_list_delayed(&head);
473 nfsd_file_lru_walk_list(NULL);
477 nfsd_file_gc_worker(struct work_struct *work)
480 nfsd_file_schedule_laundrette();
484 nfsd_file_lru_count(struct shrinker *s, struct shrink_control *sc)
486 return list_lru_count(&nfsd_file_lru);
490 nfsd_file_lru_scan(struct shrinker *s, struct shrink_control *sc)
492 return nfsd_file_lru_walk_list(sc);
495 static struct shrinker nfsd_file_shrinker = {
496 .scan_objects = nfsd_file_lru_scan,
497 .count_objects = nfsd_file_lru_count,
502 __nfsd_file_close_inode(struct inode *inode, unsigned int hashval,
503 struct list_head *dispose)
505 struct nfsd_file *nf;
506 struct hlist_node *tmp;
508 spin_lock(&nfsd_file_hashtbl[hashval].nfb_lock);
509 hlist_for_each_entry_safe(nf, tmp, &nfsd_file_hashtbl[hashval].nfb_head, nf_node) {
510 if (inode == nf->nf_inode)
511 nfsd_file_unhash_and_release_locked(nf, dispose);
513 spin_unlock(&nfsd_file_hashtbl[hashval].nfb_lock);
517 * nfsd_file_close_inode_sync - attempt to forcibly close a nfsd_file
518 * @inode: inode of the file to attempt to remove
520 * Walk the whole hash bucket, looking for any files that correspond to "inode".
521 * If any do, then unhash them and put the hashtable reference to them and
522 * destroy any that had their last reference put. Also ensure that any of the
523 * fputs also have their final __fput done as well.
526 nfsd_file_close_inode_sync(struct inode *inode)
528 unsigned int hashval = (unsigned int)hash_long(inode->i_ino,
529 NFSD_FILE_HASH_BITS);
532 __nfsd_file_close_inode(inode, hashval, &dispose);
533 trace_nfsd_file_close_inode_sync(inode, hashval, !list_empty(&dispose));
534 nfsd_file_dispose_list_sync(&dispose);
538 * nfsd_file_close_inode - attempt a delayed close of a nfsd_file
539 * @inode: inode of the file to attempt to remove
541 * Walk the whole hash bucket, looking for any files that correspond to "inode".
542 * If any do, then unhash them and put the hashtable reference to them and
543 * destroy any that had their last reference put.
546 nfsd_file_close_inode(struct inode *inode)
548 unsigned int hashval = (unsigned int)hash_long(inode->i_ino,
549 NFSD_FILE_HASH_BITS);
552 __nfsd_file_close_inode(inode, hashval, &dispose);
553 trace_nfsd_file_close_inode(inode, hashval, !list_empty(&dispose));
554 nfsd_file_dispose_list_delayed(&dispose);
558 * nfsd_file_delayed_close - close unused nfsd_files
561 * Walk the LRU list and close any entries that have not been used since
564 * Note this can deadlock with nfsd_file_cache_purge.
567 nfsd_file_delayed_close(struct work_struct *work)
570 struct nfsd_fcache_disposal *l = container_of(work,
571 struct nfsd_fcache_disposal, work);
573 nfsd_file_list_remove_disposal(&head, l);
574 nfsd_file_dispose_list(&head);
578 nfsd_file_lease_notifier_call(struct notifier_block *nb, unsigned long arg,
581 struct file_lock *fl = data;
583 /* Only close files for F_SETLEASE leases */
584 if (fl->fl_flags & FL_LEASE)
585 nfsd_file_close_inode_sync(file_inode(fl->fl_file));
589 static struct notifier_block nfsd_file_lease_notifier = {
590 .notifier_call = nfsd_file_lease_notifier_call,
594 nfsd_file_fsnotify_handle_event(struct fsnotify_mark *mark, u32 mask,
595 struct inode *inode, struct inode *dir,
596 const struct qstr *name, u32 cookie)
598 trace_nfsd_file_fsnotify_handle_event(inode, mask);
600 /* Should be no marks on non-regular files */
601 if (!S_ISREG(inode->i_mode)) {
606 /* don't close files if this was not the last link */
607 if (mask & FS_ATTRIB) {
612 nfsd_file_close_inode(inode);
617 static const struct fsnotify_ops nfsd_file_fsnotify_ops = {
618 .handle_inode_event = nfsd_file_fsnotify_handle_event,
619 .free_mark = nfsd_file_mark_free,
623 nfsd_file_cache_init(void)
628 clear_bit(NFSD_FILE_SHUTDOWN, &nfsd_file_lru_flags);
630 if (nfsd_file_hashtbl)
633 nfsd_filecache_wq = alloc_workqueue("nfsd_filecache", 0, 0);
634 if (!nfsd_filecache_wq)
637 nfsd_file_hashtbl = kvcalloc(NFSD_FILE_HASH_SIZE,
638 sizeof(*nfsd_file_hashtbl), GFP_KERNEL);
639 if (!nfsd_file_hashtbl) {
640 pr_err("nfsd: unable to allocate nfsd_file_hashtbl\n");
644 nfsd_file_slab = kmem_cache_create("nfsd_file",
645 sizeof(struct nfsd_file), 0, 0, NULL);
646 if (!nfsd_file_slab) {
647 pr_err("nfsd: unable to create nfsd_file_slab\n");
651 nfsd_file_mark_slab = kmem_cache_create("nfsd_file_mark",
652 sizeof(struct nfsd_file_mark), 0, 0, NULL);
653 if (!nfsd_file_mark_slab) {
654 pr_err("nfsd: unable to create nfsd_file_mark_slab\n");
659 ret = list_lru_init(&nfsd_file_lru);
661 pr_err("nfsd: failed to init nfsd_file_lru: %d\n", ret);
665 ret = register_shrinker(&nfsd_file_shrinker);
667 pr_err("nfsd: failed to register nfsd_file_shrinker: %d\n", ret);
671 ret = lease_register_notifier(&nfsd_file_lease_notifier);
673 pr_err("nfsd: unable to register lease notifier: %d\n", ret);
677 nfsd_file_fsnotify_group = fsnotify_alloc_group(&nfsd_file_fsnotify_ops);
678 if (IS_ERR(nfsd_file_fsnotify_group)) {
679 pr_err("nfsd: unable to create fsnotify group: %ld\n",
680 PTR_ERR(nfsd_file_fsnotify_group));
681 ret = PTR_ERR(nfsd_file_fsnotify_group);
682 nfsd_file_fsnotify_group = NULL;
686 for (i = 0; i < NFSD_FILE_HASH_SIZE; i++) {
687 INIT_HLIST_HEAD(&nfsd_file_hashtbl[i].nfb_head);
688 spin_lock_init(&nfsd_file_hashtbl[i].nfb_lock);
691 INIT_DELAYED_WORK(&nfsd_filecache_laundrette, nfsd_file_gc_worker);
695 lease_unregister_notifier(&nfsd_file_lease_notifier);
697 unregister_shrinker(&nfsd_file_shrinker);
699 list_lru_destroy(&nfsd_file_lru);
701 kmem_cache_destroy(nfsd_file_slab);
702 nfsd_file_slab = NULL;
703 kmem_cache_destroy(nfsd_file_mark_slab);
704 nfsd_file_mark_slab = NULL;
705 kvfree(nfsd_file_hashtbl);
706 nfsd_file_hashtbl = NULL;
707 destroy_workqueue(nfsd_filecache_wq);
708 nfsd_filecache_wq = NULL;
713 * Note this can deadlock with nfsd_file_lru_cb.
716 nfsd_file_cache_purge(struct net *net)
719 struct nfsd_file *nf;
720 struct hlist_node *next;
724 if (!nfsd_file_hashtbl)
727 for (i = 0; i < NFSD_FILE_HASH_SIZE; i++) {
728 struct nfsd_fcache_bucket *nfb = &nfsd_file_hashtbl[i];
730 spin_lock(&nfb->nfb_lock);
731 hlist_for_each_entry_safe(nf, next, &nfb->nfb_head, nf_node) {
732 if (net && nf->nf_net != net)
734 del = nfsd_file_unhash_and_release_locked(nf, &dispose);
737 * Deadlock detected! Something marked this entry as
738 * unhased, but hasn't removed it from the hash list.
742 spin_unlock(&nfb->nfb_lock);
743 nfsd_file_dispose_list(&dispose);
747 static struct nfsd_fcache_disposal *
748 nfsd_alloc_fcache_disposal(struct net *net)
750 struct nfsd_fcache_disposal *l;
752 l = kmalloc(sizeof(*l), GFP_KERNEL);
755 INIT_WORK(&l->work, nfsd_file_delayed_close);
757 spin_lock_init(&l->lock);
758 INIT_LIST_HEAD(&l->freeme);
763 nfsd_free_fcache_disposal(struct nfsd_fcache_disposal *l)
765 rcu_assign_pointer(l->net, NULL);
766 cancel_work_sync(&l->work);
767 nfsd_file_dispose_list(&l->freeme);
772 nfsd_add_fcache_disposal(struct nfsd_fcache_disposal *l)
774 spin_lock(&laundrette_lock);
775 list_add_tail_rcu(&l->list, &laundrettes);
776 spin_unlock(&laundrette_lock);
780 nfsd_del_fcache_disposal(struct nfsd_fcache_disposal *l)
782 spin_lock(&laundrette_lock);
783 list_del_rcu(&l->list);
784 spin_unlock(&laundrette_lock);
788 nfsd_alloc_fcache_disposal_net(struct net *net)
790 struct nfsd_fcache_disposal *l;
792 l = nfsd_alloc_fcache_disposal(net);
795 nfsd_add_fcache_disposal(l);
800 nfsd_free_fcache_disposal_net(struct net *net)
802 struct nfsd_fcache_disposal *l;
805 list_for_each_entry_rcu(l, &laundrettes, list) {
808 nfsd_del_fcache_disposal(l);
810 nfsd_free_fcache_disposal(l);
817 nfsd_file_cache_start_net(struct net *net)
819 return nfsd_alloc_fcache_disposal_net(net);
823 nfsd_file_cache_shutdown_net(struct net *net)
825 nfsd_file_cache_purge(net);
826 nfsd_free_fcache_disposal_net(net);
830 nfsd_file_cache_shutdown(void)
832 set_bit(NFSD_FILE_SHUTDOWN, &nfsd_file_lru_flags);
834 lease_unregister_notifier(&nfsd_file_lease_notifier);
835 unregister_shrinker(&nfsd_file_shrinker);
837 * make sure all callers of nfsd_file_lru_cb are done before
838 * calling nfsd_file_cache_purge
840 cancel_delayed_work_sync(&nfsd_filecache_laundrette);
841 nfsd_file_cache_purge(NULL);
842 list_lru_destroy(&nfsd_file_lru);
844 fsnotify_put_group(nfsd_file_fsnotify_group);
845 nfsd_file_fsnotify_group = NULL;
846 kmem_cache_destroy(nfsd_file_slab);
847 nfsd_file_slab = NULL;
848 fsnotify_wait_marks_destroyed();
849 kmem_cache_destroy(nfsd_file_mark_slab);
850 nfsd_file_mark_slab = NULL;
851 kvfree(nfsd_file_hashtbl);
852 nfsd_file_hashtbl = NULL;
853 destroy_workqueue(nfsd_filecache_wq);
854 nfsd_filecache_wq = NULL;
858 nfsd_match_cred(const struct cred *c1, const struct cred *c2)
862 if (!uid_eq(c1->fsuid, c2->fsuid))
864 if (!gid_eq(c1->fsgid, c2->fsgid))
866 if (c1->group_info == NULL || c2->group_info == NULL)
867 return c1->group_info == c2->group_info;
868 if (c1->group_info->ngroups != c2->group_info->ngroups)
870 for (i = 0; i < c1->group_info->ngroups; i++) {
871 if (!gid_eq(c1->group_info->gid[i], c2->group_info->gid[i]))
877 static struct nfsd_file *
878 nfsd_file_find_locked(struct inode *inode, unsigned int may_flags,
879 unsigned int hashval, struct net *net)
881 struct nfsd_file *nf;
882 unsigned char need = may_flags & NFSD_FILE_MAY_MASK;
884 hlist_for_each_entry_rcu(nf, &nfsd_file_hashtbl[hashval].nfb_head,
885 nf_node, lockdep_is_held(&nfsd_file_hashtbl[hashval].nfb_lock)) {
886 if (nf->nf_may != need)
888 if (nf->nf_inode != inode)
890 if (nf->nf_net != net)
892 if (!nfsd_match_cred(nf->nf_cred, current_cred()))
894 if (!test_bit(NFSD_FILE_HASHED, &nf->nf_flags))
896 if (nfsd_file_get(nf) != NULL)
903 * nfsd_file_is_cached - are there any cached open files for this fh?
904 * @inode: inode of the file to check
906 * Scan the hashtable for open files that match this fh. Returns true if there
907 * are any, and false if not.
910 nfsd_file_is_cached(struct inode *inode)
913 struct nfsd_file *nf;
914 unsigned int hashval;
916 hashval = (unsigned int)hash_long(inode->i_ino, NFSD_FILE_HASH_BITS);
919 hlist_for_each_entry_rcu(nf, &nfsd_file_hashtbl[hashval].nfb_head,
921 if (inode == nf->nf_inode) {
927 trace_nfsd_file_is_cached(inode, hashval, (int)ret);
932 nfsd_file_acquire(struct svc_rqst *rqstp, struct svc_fh *fhp,
933 unsigned int may_flags, struct nfsd_file **pnf)
936 struct net *net = SVC_NET(rqstp);
937 struct nfsd_file *nf, *new;
939 unsigned int hashval;
942 /* FIXME: skip this if fh_dentry is already set? */
943 status = fh_verify(rqstp, fhp, S_IFREG,
944 may_flags|NFSD_MAY_OWNER_OVERRIDE);
945 if (status != nfs_ok)
948 inode = d_inode(fhp->fh_dentry);
949 hashval = (unsigned int)hash_long(inode->i_ino, NFSD_FILE_HASH_BITS);
952 nf = nfsd_file_find_locked(inode, may_flags, hashval, net);
955 goto wait_for_construction;
957 new = nfsd_file_alloc(inode, may_flags, hashval, net);
959 trace_nfsd_file_acquire(rqstp, hashval, inode, may_flags,
960 NULL, nfserr_jukebox);
961 return nfserr_jukebox;
964 spin_lock(&nfsd_file_hashtbl[hashval].nfb_lock);
965 nf = nfsd_file_find_locked(inode, may_flags, hashval, net);
968 spin_unlock(&nfsd_file_hashtbl[hashval].nfb_lock);
969 nfsd_file_slab_free(&new->nf_rcu);
971 wait_for_construction:
972 wait_on_bit(&nf->nf_flags, NFSD_FILE_PENDING, TASK_UNINTERRUPTIBLE);
974 /* Did construction of this file fail? */
975 if (!test_bit(NFSD_FILE_HASHED, &nf->nf_flags)) {
977 status = nfserr_jukebox;
981 nfsd_file_put_noref(nf);
985 this_cpu_inc(nfsd_file_cache_hits);
987 status = nfserrno(nfsd_open_break_lease(file_inode(nf->nf_file), may_flags));
989 if (status == nfs_ok) {
996 trace_nfsd_file_acquire(rqstp, hashval, inode, may_flags, nf, status);
1000 /* Take reference for the hashtable */
1001 refcount_inc(&nf->nf_ref);
1002 __set_bit(NFSD_FILE_HASHED, &nf->nf_flags);
1003 __set_bit(NFSD_FILE_PENDING, &nf->nf_flags);
1004 list_lru_add(&nfsd_file_lru, &nf->nf_lru);
1005 hlist_add_head_rcu(&nf->nf_node, &nfsd_file_hashtbl[hashval].nfb_head);
1006 ++nfsd_file_hashtbl[hashval].nfb_count;
1007 nfsd_file_hashtbl[hashval].nfb_maxcount = max(nfsd_file_hashtbl[hashval].nfb_maxcount,
1008 nfsd_file_hashtbl[hashval].nfb_count);
1009 spin_unlock(&nfsd_file_hashtbl[hashval].nfb_lock);
1010 if (atomic_long_inc_return(&nfsd_filecache_count) >= NFSD_FILE_LRU_THRESHOLD)
1013 nf->nf_mark = nfsd_file_mark_find_or_create(nf);
1015 status = nfsd_open_verified(rqstp, fhp, S_IFREG,
1016 may_flags, &nf->nf_file);
1018 status = nfserr_jukebox;
1020 * If construction failed, or we raced with a call to unlink()
1023 if (status != nfs_ok || inode->i_nlink == 0) {
1025 spin_lock(&nfsd_file_hashtbl[hashval].nfb_lock);
1026 do_free = nfsd_file_unhash(nf);
1027 spin_unlock(&nfsd_file_hashtbl[hashval].nfb_lock);
1029 nfsd_file_put_noref(nf);
1031 clear_bit_unlock(NFSD_FILE_PENDING, &nf->nf_flags);
1032 smp_mb__after_atomic();
1033 wake_up_bit(&nf->nf_flags, NFSD_FILE_PENDING);
1038 * Note that fields may be added, removed or reordered in the future. Programs
1039 * scraping this file for info should test the labels to ensure they're
1040 * getting the correct field.
1042 static int nfsd_file_cache_stats_show(struct seq_file *m, void *v)
1044 unsigned int i, count = 0, longest = 0;
1045 unsigned long hits = 0;
1048 * No need for spinlocks here since we're not terribly interested in
1049 * accuracy. We do take the nfsd_mutex simply to ensure that we
1050 * don't end up racing with server shutdown
1052 mutex_lock(&nfsd_mutex);
1053 if (nfsd_file_hashtbl) {
1054 for (i = 0; i < NFSD_FILE_HASH_SIZE; i++) {
1055 count += nfsd_file_hashtbl[i].nfb_count;
1056 longest = max(longest, nfsd_file_hashtbl[i].nfb_count);
1059 mutex_unlock(&nfsd_mutex);
1061 for_each_possible_cpu(i)
1062 hits += per_cpu(nfsd_file_cache_hits, i);
1064 seq_printf(m, "total entries: %u\n", count);
1065 seq_printf(m, "longest chain: %u\n", longest);
1066 seq_printf(m, "cache hits: %lu\n", hits);
1070 int nfsd_file_cache_stats_open(struct inode *inode, struct file *file)
1072 return single_open(file, nfsd_file_cache_stats_show, NULL);