xref: /linux/fs/nfsd/filecache.c (revision c49731a04e7c14b2ccd146a7fddf92e78ccae143)
1 /*
2  * Open file cache.
3  *
4  * (c) 2015 - Jeff Layton <jeff.layton@primarydata.com>
5  */
6 
7 #include <linux/hash.h>
8 #include <linux/slab.h>
9 #include <linux/file.h>
10 #include <linux/pagemap.h>
11 #include <linux/sched.h>
12 #include <linux/list_lru.h>
13 #include <linux/fsnotify_backend.h>
14 #include <linux/fsnotify.h>
15 #include <linux/seq_file.h>
16 
17 #include "vfs.h"
18 #include "nfsd.h"
19 #include "nfsfh.h"
20 #include "netns.h"
21 #include "filecache.h"
22 #include "trace.h"
23 
24 #define NFSDDBG_FACILITY	NFSDDBG_FH
25 
26 /* FIXME: dynamically size this for the machine somehow? */
27 #define NFSD_FILE_HASH_BITS                   12
28 #define NFSD_FILE_HASH_SIZE                  (1 << NFSD_FILE_HASH_BITS)
29 #define NFSD_LAUNDRETTE_DELAY		     (2 * HZ)
30 
31 #define NFSD_FILE_SHUTDOWN		     (1)
32 #define NFSD_FILE_LRU_THRESHOLD		     (4096UL)
33 #define NFSD_FILE_LRU_LIMIT		     (NFSD_FILE_LRU_THRESHOLD << 2)
34 
35 /* We only care about NFSD_MAY_READ/WRITE for this cache */
36 #define NFSD_FILE_MAY_MASK	(NFSD_MAY_READ|NFSD_MAY_WRITE)
37 
38 struct nfsd_fcache_bucket {
39 	struct hlist_head	nfb_head;
40 	spinlock_t		nfb_lock;
41 	unsigned int		nfb_count;
42 	unsigned int		nfb_maxcount;
43 };
44 
45 static DEFINE_PER_CPU(unsigned long, nfsd_file_cache_hits);
46 
47 struct nfsd_fcache_disposal {
48 	struct work_struct work;
49 	spinlock_t lock;
50 	struct list_head freeme;
51 };
52 
53 static struct workqueue_struct *nfsd_filecache_wq __read_mostly;
54 
55 static struct kmem_cache		*nfsd_file_slab;
56 static struct kmem_cache		*nfsd_file_mark_slab;
57 static struct nfsd_fcache_bucket	*nfsd_file_hashtbl;
58 static struct list_lru			nfsd_file_lru;
59 static long				nfsd_file_lru_flags;
60 static struct fsnotify_group		*nfsd_file_fsnotify_group;
61 static atomic_long_t			nfsd_filecache_count;
62 static struct delayed_work		nfsd_filecache_laundrette;
63 
64 static void nfsd_file_gc(void);
65 
66 static void
67 nfsd_file_schedule_laundrette(void)
68 {
69 	long count = atomic_long_read(&nfsd_filecache_count);
70 
71 	if (count == 0 || test_bit(NFSD_FILE_SHUTDOWN, &nfsd_file_lru_flags))
72 		return;
73 
74 	queue_delayed_work(system_wq, &nfsd_filecache_laundrette,
75 			NFSD_LAUNDRETTE_DELAY);
76 }
77 
78 static void
79 nfsd_file_slab_free(struct rcu_head *rcu)
80 {
81 	struct nfsd_file *nf = container_of(rcu, struct nfsd_file, nf_rcu);
82 
83 	put_cred(nf->nf_cred);
84 	kmem_cache_free(nfsd_file_slab, nf);
85 }
86 
87 static void
88 nfsd_file_mark_free(struct fsnotify_mark *mark)
89 {
90 	struct nfsd_file_mark *nfm = container_of(mark, struct nfsd_file_mark,
91 						  nfm_mark);
92 
93 	kmem_cache_free(nfsd_file_mark_slab, nfm);
94 }
95 
96 static struct nfsd_file_mark *
97 nfsd_file_mark_get(struct nfsd_file_mark *nfm)
98 {
99 	if (!refcount_inc_not_zero(&nfm->nfm_ref))
100 		return NULL;
101 	return nfm;
102 }
103 
104 static void
105 nfsd_file_mark_put(struct nfsd_file_mark *nfm)
106 {
107 	if (refcount_dec_and_test(&nfm->nfm_ref)) {
108 		fsnotify_destroy_mark(&nfm->nfm_mark, nfsd_file_fsnotify_group);
109 		fsnotify_put_mark(&nfm->nfm_mark);
110 	}
111 }
112 
113 static struct nfsd_file_mark *
114 nfsd_file_mark_find_or_create(struct nfsd_file *nf)
115 {
116 	int			err;
117 	struct fsnotify_mark	*mark;
118 	struct nfsd_file_mark	*nfm = NULL, *new;
119 	struct inode *inode = nf->nf_inode;
120 
121 	do {
122 		fsnotify_group_lock(nfsd_file_fsnotify_group);
123 		mark = fsnotify_find_mark(&inode->i_fsnotify_marks,
124 					  nfsd_file_fsnotify_group);
125 		if (mark) {
126 			nfm = nfsd_file_mark_get(container_of(mark,
127 						 struct nfsd_file_mark,
128 						 nfm_mark));
129 			fsnotify_group_unlock(nfsd_file_fsnotify_group);
130 			if (nfm) {
131 				fsnotify_put_mark(mark);
132 				break;
133 			}
134 			/* Avoid soft lockup race with nfsd_file_mark_put() */
135 			fsnotify_destroy_mark(mark, nfsd_file_fsnotify_group);
136 			fsnotify_put_mark(mark);
137 		} else {
138 			fsnotify_group_unlock(nfsd_file_fsnotify_group);
139 		}
140 
141 		/* allocate a new nfm */
142 		new = kmem_cache_alloc(nfsd_file_mark_slab, GFP_KERNEL);
143 		if (!new)
144 			return NULL;
145 		fsnotify_init_mark(&new->nfm_mark, nfsd_file_fsnotify_group);
146 		new->nfm_mark.mask = FS_ATTRIB|FS_DELETE_SELF;
147 		refcount_set(&new->nfm_ref, 1);
148 
149 		err = fsnotify_add_inode_mark(&new->nfm_mark, inode, 0);
150 
151 		/*
152 		 * If the add was successful, then return the object.
153 		 * Otherwise, we need to put the reference we hold on the
154 		 * nfm_mark. The fsnotify code will take a reference and put
155 		 * it on failure, so we can't just free it directly. It's also
156 		 * not safe to call fsnotify_destroy_mark on it as the
157 		 * mark->group will be NULL. Thus, we can't let the nfm_ref
158 		 * counter drive the destruction at this point.
159 		 */
160 		if (likely(!err))
161 			nfm = new;
162 		else
163 			fsnotify_put_mark(&new->nfm_mark);
164 	} while (unlikely(err == -EEXIST));
165 
166 	return nfm;
167 }
168 
169 static struct nfsd_file *
170 nfsd_file_alloc(struct inode *inode, unsigned int may, unsigned int hashval,
171 		struct net *net)
172 {
173 	struct nfsd_file *nf;
174 
175 	nf = kmem_cache_alloc(nfsd_file_slab, GFP_KERNEL);
176 	if (nf) {
177 		INIT_HLIST_NODE(&nf->nf_node);
178 		INIT_LIST_HEAD(&nf->nf_lru);
179 		nf->nf_file = NULL;
180 		nf->nf_cred = get_current_cred();
181 		nf->nf_net = net;
182 		nf->nf_flags = 0;
183 		nf->nf_inode = inode;
184 		nf->nf_hashval = hashval;
185 		refcount_set(&nf->nf_ref, 1);
186 		nf->nf_may = may & NFSD_FILE_MAY_MASK;
187 		if (may & NFSD_MAY_NOT_BREAK_LEASE) {
188 			if (may & NFSD_MAY_WRITE)
189 				__set_bit(NFSD_FILE_BREAK_WRITE, &nf->nf_flags);
190 			if (may & NFSD_MAY_READ)
191 				__set_bit(NFSD_FILE_BREAK_READ, &nf->nf_flags);
192 		}
193 		nf->nf_mark = NULL;
194 		trace_nfsd_file_alloc(nf);
195 	}
196 	return nf;
197 }
198 
199 static bool
200 nfsd_file_free(struct nfsd_file *nf)
201 {
202 	bool flush = false;
203 
204 	trace_nfsd_file_put_final(nf);
205 	if (nf->nf_mark)
206 		nfsd_file_mark_put(nf->nf_mark);
207 	if (nf->nf_file) {
208 		get_file(nf->nf_file);
209 		filp_close(nf->nf_file, NULL);
210 		fput(nf->nf_file);
211 		flush = true;
212 	}
213 	call_rcu(&nf->nf_rcu, nfsd_file_slab_free);
214 	return flush;
215 }
216 
217 static bool
218 nfsd_file_check_writeback(struct nfsd_file *nf)
219 {
220 	struct file *file = nf->nf_file;
221 	struct address_space *mapping;
222 
223 	if (!file || !(file->f_mode & FMODE_WRITE))
224 		return false;
225 	mapping = file->f_mapping;
226 	return mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) ||
227 		mapping_tagged(mapping, PAGECACHE_TAG_WRITEBACK);
228 }
229 
230 static int
231 nfsd_file_check_write_error(struct nfsd_file *nf)
232 {
233 	struct file *file = nf->nf_file;
234 
235 	if (!file || !(file->f_mode & FMODE_WRITE))
236 		return 0;
237 	return filemap_check_wb_err(file->f_mapping, READ_ONCE(file->f_wb_err));
238 }
239 
240 static void
241 nfsd_file_flush(struct nfsd_file *nf)
242 {
243 	if (nf->nf_file && vfs_fsync(nf->nf_file, 1) != 0)
244 		nfsd_reset_write_verifier(net_generic(nf->nf_net, nfsd_net_id));
245 }
246 
247 static void
248 nfsd_file_do_unhash(struct nfsd_file *nf)
249 {
250 	lockdep_assert_held(&nfsd_file_hashtbl[nf->nf_hashval].nfb_lock);
251 
252 	trace_nfsd_file_unhash(nf);
253 
254 	if (nfsd_file_check_write_error(nf))
255 		nfsd_reset_write_verifier(net_generic(nf->nf_net, nfsd_net_id));
256 	--nfsd_file_hashtbl[nf->nf_hashval].nfb_count;
257 	hlist_del_rcu(&nf->nf_node);
258 	atomic_long_dec(&nfsd_filecache_count);
259 }
260 
261 static bool
262 nfsd_file_unhash(struct nfsd_file *nf)
263 {
264 	if (test_and_clear_bit(NFSD_FILE_HASHED, &nf->nf_flags)) {
265 		nfsd_file_do_unhash(nf);
266 		if (!list_empty(&nf->nf_lru))
267 			list_lru_del(&nfsd_file_lru, &nf->nf_lru);
268 		return true;
269 	}
270 	return false;
271 }
272 
273 /*
274  * Return true if the file was unhashed.
275  */
276 static bool
277 nfsd_file_unhash_and_release_locked(struct nfsd_file *nf, struct list_head *dispose)
278 {
279 	lockdep_assert_held(&nfsd_file_hashtbl[nf->nf_hashval].nfb_lock);
280 
281 	trace_nfsd_file_unhash_and_release_locked(nf);
282 	if (!nfsd_file_unhash(nf))
283 		return false;
284 	/* keep final reference for nfsd_file_lru_dispose */
285 	if (refcount_dec_not_one(&nf->nf_ref))
286 		return true;
287 
288 	list_add(&nf->nf_lru, dispose);
289 	return true;
290 }
291 
292 static void
293 nfsd_file_put_noref(struct nfsd_file *nf)
294 {
295 	trace_nfsd_file_put(nf);
296 
297 	if (refcount_dec_and_test(&nf->nf_ref)) {
298 		WARN_ON(test_bit(NFSD_FILE_HASHED, &nf->nf_flags));
299 		nfsd_file_free(nf);
300 	}
301 }
302 
303 void
304 nfsd_file_put(struct nfsd_file *nf)
305 {
306 	might_sleep();
307 
308 	set_bit(NFSD_FILE_REFERENCED, &nf->nf_flags);
309 	if (test_bit(NFSD_FILE_HASHED, &nf->nf_flags) == 0) {
310 		nfsd_file_flush(nf);
311 		nfsd_file_put_noref(nf);
312 	} else if (nf->nf_file) {
313 		nfsd_file_put_noref(nf);
314 		nfsd_file_schedule_laundrette();
315 	} else
316 		nfsd_file_put_noref(nf);
317 
318 	if (atomic_long_read(&nfsd_filecache_count) >= NFSD_FILE_LRU_LIMIT)
319 		nfsd_file_gc();
320 }
321 
322 struct nfsd_file *
323 nfsd_file_get(struct nfsd_file *nf)
324 {
325 	if (likely(refcount_inc_not_zero(&nf->nf_ref)))
326 		return nf;
327 	return NULL;
328 }
329 
330 static void
331 nfsd_file_dispose_list(struct list_head *dispose)
332 {
333 	struct nfsd_file *nf;
334 
335 	while(!list_empty(dispose)) {
336 		nf = list_first_entry(dispose, struct nfsd_file, nf_lru);
337 		list_del(&nf->nf_lru);
338 		nfsd_file_flush(nf);
339 		nfsd_file_put_noref(nf);
340 	}
341 }
342 
343 static void
344 nfsd_file_dispose_list_sync(struct list_head *dispose)
345 {
346 	bool flush = false;
347 	struct nfsd_file *nf;
348 
349 	while(!list_empty(dispose)) {
350 		nf = list_first_entry(dispose, struct nfsd_file, nf_lru);
351 		list_del(&nf->nf_lru);
352 		nfsd_file_flush(nf);
353 		if (!refcount_dec_and_test(&nf->nf_ref))
354 			continue;
355 		if (nfsd_file_free(nf))
356 			flush = true;
357 	}
358 	if (flush)
359 		flush_delayed_fput();
360 }
361 
362 static void
363 nfsd_file_list_remove_disposal(struct list_head *dst,
364 		struct nfsd_fcache_disposal *l)
365 {
366 	spin_lock(&l->lock);
367 	list_splice_init(&l->freeme, dst);
368 	spin_unlock(&l->lock);
369 }
370 
371 static void
372 nfsd_file_list_add_disposal(struct list_head *files, struct net *net)
373 {
374 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
375 	struct nfsd_fcache_disposal *l = nn->fcache_disposal;
376 
377 	spin_lock(&l->lock);
378 	list_splice_tail_init(files, &l->freeme);
379 	spin_unlock(&l->lock);
380 	queue_work(nfsd_filecache_wq, &l->work);
381 }
382 
383 static void
384 nfsd_file_list_add_pernet(struct list_head *dst, struct list_head *src,
385 		struct net *net)
386 {
387 	struct nfsd_file *nf, *tmp;
388 
389 	list_for_each_entry_safe(nf, tmp, src, nf_lru) {
390 		if (nf->nf_net == net)
391 			list_move_tail(&nf->nf_lru, dst);
392 	}
393 }
394 
395 static void
396 nfsd_file_dispose_list_delayed(struct list_head *dispose)
397 {
398 	LIST_HEAD(list);
399 	struct nfsd_file *nf;
400 
401 	while(!list_empty(dispose)) {
402 		nf = list_first_entry(dispose, struct nfsd_file, nf_lru);
403 		nfsd_file_list_add_pernet(&list, dispose, nf->nf_net);
404 		nfsd_file_list_add_disposal(&list, nf->nf_net);
405 	}
406 }
407 
408 /*
409  * Note this can deadlock with nfsd_file_cache_purge.
410  */
411 static enum lru_status
412 nfsd_file_lru_cb(struct list_head *item, struct list_lru_one *lru,
413 		 spinlock_t *lock, void *arg)
414 	__releases(lock)
415 	__acquires(lock)
416 {
417 	struct list_head *head = arg;
418 	struct nfsd_file *nf = list_entry(item, struct nfsd_file, nf_lru);
419 
420 	/*
421 	 * Do a lockless refcount check. The hashtable holds one reference, so
422 	 * we look to see if anything else has a reference, or if any have
423 	 * been put since the shrinker last ran. Those don't get unhashed and
424 	 * released.
425 	 *
426 	 * Note that in the put path, we set the flag and then decrement the
427 	 * counter. Here we check the counter and then test and clear the flag.
428 	 * That order is deliberate to ensure that we can do this locklessly.
429 	 */
430 	if (refcount_read(&nf->nf_ref) > 1)
431 		goto out_skip;
432 
433 	/*
434 	 * Don't throw out files that are still undergoing I/O or
435 	 * that have uncleared errors pending.
436 	 */
437 	if (nfsd_file_check_writeback(nf))
438 		goto out_skip;
439 
440 	if (test_and_clear_bit(NFSD_FILE_REFERENCED, &nf->nf_flags))
441 		goto out_skip;
442 
443 	if (!test_and_clear_bit(NFSD_FILE_HASHED, &nf->nf_flags))
444 		goto out_skip;
445 
446 	list_lru_isolate_move(lru, &nf->nf_lru, head);
447 	return LRU_REMOVED;
448 out_skip:
449 	return LRU_SKIP;
450 }
451 
452 static unsigned long
453 nfsd_file_lru_walk_list(struct shrink_control *sc)
454 {
455 	LIST_HEAD(head);
456 	struct nfsd_file *nf;
457 	unsigned long ret;
458 
459 	if (sc)
460 		ret = list_lru_shrink_walk(&nfsd_file_lru, sc,
461 				nfsd_file_lru_cb, &head);
462 	else
463 		ret = list_lru_walk(&nfsd_file_lru,
464 				nfsd_file_lru_cb,
465 				&head, LONG_MAX);
466 	list_for_each_entry(nf, &head, nf_lru) {
467 		spin_lock(&nfsd_file_hashtbl[nf->nf_hashval].nfb_lock);
468 		nfsd_file_do_unhash(nf);
469 		spin_unlock(&nfsd_file_hashtbl[nf->nf_hashval].nfb_lock);
470 	}
471 	nfsd_file_dispose_list_delayed(&head);
472 	return ret;
473 }
474 
475 static void
476 nfsd_file_gc(void)
477 {
478 	nfsd_file_lru_walk_list(NULL);
479 }
480 
481 static void
482 nfsd_file_gc_worker(struct work_struct *work)
483 {
484 	nfsd_file_gc();
485 	nfsd_file_schedule_laundrette();
486 }
487 
488 static unsigned long
489 nfsd_file_lru_count(struct shrinker *s, struct shrink_control *sc)
490 {
491 	return list_lru_count(&nfsd_file_lru);
492 }
493 
494 static unsigned long
495 nfsd_file_lru_scan(struct shrinker *s, struct shrink_control *sc)
496 {
497 	return nfsd_file_lru_walk_list(sc);
498 }
499 
500 static struct shrinker	nfsd_file_shrinker = {
501 	.scan_objects = nfsd_file_lru_scan,
502 	.count_objects = nfsd_file_lru_count,
503 	.seeks = 1,
504 };
505 
506 static void
507 __nfsd_file_close_inode(struct inode *inode, unsigned int hashval,
508 			struct list_head *dispose)
509 {
510 	struct nfsd_file	*nf;
511 	struct hlist_node	*tmp;
512 
513 	spin_lock(&nfsd_file_hashtbl[hashval].nfb_lock);
514 	hlist_for_each_entry_safe(nf, tmp, &nfsd_file_hashtbl[hashval].nfb_head, nf_node) {
515 		if (inode == nf->nf_inode)
516 			nfsd_file_unhash_and_release_locked(nf, dispose);
517 	}
518 	spin_unlock(&nfsd_file_hashtbl[hashval].nfb_lock);
519 }
520 
521 /**
522  * nfsd_file_close_inode_sync - attempt to forcibly close a nfsd_file
523  * @inode: inode of the file to attempt to remove
524  *
525  * Walk the whole hash bucket, looking for any files that correspond to "inode".
526  * If any do, then unhash them and put the hashtable reference to them and
527  * destroy any that had their last reference put. Also ensure that any of the
528  * fputs also have their final __fput done as well.
529  */
530 void
531 nfsd_file_close_inode_sync(struct inode *inode)
532 {
533 	unsigned int		hashval = (unsigned int)hash_long(inode->i_ino,
534 						NFSD_FILE_HASH_BITS);
535 	LIST_HEAD(dispose);
536 
537 	__nfsd_file_close_inode(inode, hashval, &dispose);
538 	trace_nfsd_file_close_inode_sync(inode, hashval, !list_empty(&dispose));
539 	nfsd_file_dispose_list_sync(&dispose);
540 }
541 
542 /**
543  * nfsd_file_close_inode - attempt a delayed close of a nfsd_file
544  * @inode: inode of the file to attempt to remove
545  *
546  * Walk the whole hash bucket, looking for any files that correspond to "inode".
547  * If any do, then unhash them and put the hashtable reference to them and
548  * destroy any that had their last reference put.
549  */
550 static void
551 nfsd_file_close_inode(struct inode *inode)
552 {
553 	unsigned int		hashval = (unsigned int)hash_long(inode->i_ino,
554 						NFSD_FILE_HASH_BITS);
555 	LIST_HEAD(dispose);
556 
557 	__nfsd_file_close_inode(inode, hashval, &dispose);
558 	trace_nfsd_file_close_inode(inode, hashval, !list_empty(&dispose));
559 	nfsd_file_dispose_list_delayed(&dispose);
560 }
561 
562 /**
563  * nfsd_file_delayed_close - close unused nfsd_files
564  * @work: dummy
565  *
566  * Walk the LRU list and close any entries that have not been used since
567  * the last scan.
568  *
569  * Note this can deadlock with nfsd_file_cache_purge.
570  */
571 static void
572 nfsd_file_delayed_close(struct work_struct *work)
573 {
574 	LIST_HEAD(head);
575 	struct nfsd_fcache_disposal *l = container_of(work,
576 			struct nfsd_fcache_disposal, work);
577 
578 	nfsd_file_list_remove_disposal(&head, l);
579 	nfsd_file_dispose_list(&head);
580 }
581 
582 static int
583 nfsd_file_lease_notifier_call(struct notifier_block *nb, unsigned long arg,
584 			    void *data)
585 {
586 	struct file_lock *fl = data;
587 
588 	/* Only close files for F_SETLEASE leases */
589 	if (fl->fl_flags & FL_LEASE)
590 		nfsd_file_close_inode_sync(file_inode(fl->fl_file));
591 	return 0;
592 }
593 
594 static struct notifier_block nfsd_file_lease_notifier = {
595 	.notifier_call = nfsd_file_lease_notifier_call,
596 };
597 
598 static int
599 nfsd_file_fsnotify_handle_event(struct fsnotify_mark *mark, u32 mask,
600 				struct inode *inode, struct inode *dir,
601 				const struct qstr *name, u32 cookie)
602 {
603 	if (WARN_ON_ONCE(!inode))
604 		return 0;
605 
606 	trace_nfsd_file_fsnotify_handle_event(inode, mask);
607 
608 	/* Should be no marks on non-regular files */
609 	if (!S_ISREG(inode->i_mode)) {
610 		WARN_ON_ONCE(1);
611 		return 0;
612 	}
613 
614 	/* don't close files if this was not the last link */
615 	if (mask & FS_ATTRIB) {
616 		if (inode->i_nlink)
617 			return 0;
618 	}
619 
620 	nfsd_file_close_inode(inode);
621 	return 0;
622 }
623 
624 
625 static const struct fsnotify_ops nfsd_file_fsnotify_ops = {
626 	.handle_inode_event = nfsd_file_fsnotify_handle_event,
627 	.free_mark = nfsd_file_mark_free,
628 };
629 
630 int
631 nfsd_file_cache_init(void)
632 {
633 	int		ret = -ENOMEM;
634 	unsigned int	i;
635 
636 	clear_bit(NFSD_FILE_SHUTDOWN, &nfsd_file_lru_flags);
637 
638 	if (nfsd_file_hashtbl)
639 		return 0;
640 
641 	nfsd_filecache_wq = alloc_workqueue("nfsd_filecache", 0, 0);
642 	if (!nfsd_filecache_wq)
643 		goto out;
644 
645 	nfsd_file_hashtbl = kvcalloc(NFSD_FILE_HASH_SIZE,
646 				sizeof(*nfsd_file_hashtbl), GFP_KERNEL);
647 	if (!nfsd_file_hashtbl) {
648 		pr_err("nfsd: unable to allocate nfsd_file_hashtbl\n");
649 		goto out_err;
650 	}
651 
652 	nfsd_file_slab = kmem_cache_create("nfsd_file",
653 				sizeof(struct nfsd_file), 0, 0, NULL);
654 	if (!nfsd_file_slab) {
655 		pr_err("nfsd: unable to create nfsd_file_slab\n");
656 		goto out_err;
657 	}
658 
659 	nfsd_file_mark_slab = kmem_cache_create("nfsd_file_mark",
660 					sizeof(struct nfsd_file_mark), 0, 0, NULL);
661 	if (!nfsd_file_mark_slab) {
662 		pr_err("nfsd: unable to create nfsd_file_mark_slab\n");
663 		goto out_err;
664 	}
665 
666 
667 	ret = list_lru_init(&nfsd_file_lru);
668 	if (ret) {
669 		pr_err("nfsd: failed to init nfsd_file_lru: %d\n", ret);
670 		goto out_err;
671 	}
672 
673 	ret = register_shrinker(&nfsd_file_shrinker);
674 	if (ret) {
675 		pr_err("nfsd: failed to register nfsd_file_shrinker: %d\n", ret);
676 		goto out_lru;
677 	}
678 
679 	ret = lease_register_notifier(&nfsd_file_lease_notifier);
680 	if (ret) {
681 		pr_err("nfsd: unable to register lease notifier: %d\n", ret);
682 		goto out_shrinker;
683 	}
684 
685 	nfsd_file_fsnotify_group = fsnotify_alloc_group(&nfsd_file_fsnotify_ops,
686 							FSNOTIFY_GROUP_NOFS);
687 	if (IS_ERR(nfsd_file_fsnotify_group)) {
688 		pr_err("nfsd: unable to create fsnotify group: %ld\n",
689 			PTR_ERR(nfsd_file_fsnotify_group));
690 		ret = PTR_ERR(nfsd_file_fsnotify_group);
691 		nfsd_file_fsnotify_group = NULL;
692 		goto out_notifier;
693 	}
694 
695 	for (i = 0; i < NFSD_FILE_HASH_SIZE; i++) {
696 		INIT_HLIST_HEAD(&nfsd_file_hashtbl[i].nfb_head);
697 		spin_lock_init(&nfsd_file_hashtbl[i].nfb_lock);
698 	}
699 
700 	INIT_DELAYED_WORK(&nfsd_filecache_laundrette, nfsd_file_gc_worker);
701 out:
702 	return ret;
703 out_notifier:
704 	lease_unregister_notifier(&nfsd_file_lease_notifier);
705 out_shrinker:
706 	unregister_shrinker(&nfsd_file_shrinker);
707 out_lru:
708 	list_lru_destroy(&nfsd_file_lru);
709 out_err:
710 	kmem_cache_destroy(nfsd_file_slab);
711 	nfsd_file_slab = NULL;
712 	kmem_cache_destroy(nfsd_file_mark_slab);
713 	nfsd_file_mark_slab = NULL;
714 	kvfree(nfsd_file_hashtbl);
715 	nfsd_file_hashtbl = NULL;
716 	destroy_workqueue(nfsd_filecache_wq);
717 	nfsd_filecache_wq = NULL;
718 	goto out;
719 }
720 
721 /*
722  * Note this can deadlock with nfsd_file_lru_cb.
723  */
724 void
725 nfsd_file_cache_purge(struct net *net)
726 {
727 	unsigned int		i;
728 	struct nfsd_file	*nf;
729 	struct hlist_node	*next;
730 	LIST_HEAD(dispose);
731 	bool del;
732 
733 	if (!nfsd_file_hashtbl)
734 		return;
735 
736 	for (i = 0; i < NFSD_FILE_HASH_SIZE; i++) {
737 		struct nfsd_fcache_bucket *nfb = &nfsd_file_hashtbl[i];
738 
739 		spin_lock(&nfb->nfb_lock);
740 		hlist_for_each_entry_safe(nf, next, &nfb->nfb_head, nf_node) {
741 			if (net && nf->nf_net != net)
742 				continue;
743 			del = nfsd_file_unhash_and_release_locked(nf, &dispose);
744 
745 			/*
746 			 * Deadlock detected! Something marked this entry as
747 			 * unhased, but hasn't removed it from the hash list.
748 			 */
749 			WARN_ON_ONCE(!del);
750 		}
751 		spin_unlock(&nfb->nfb_lock);
752 		nfsd_file_dispose_list(&dispose);
753 	}
754 }
755 
756 static struct nfsd_fcache_disposal *
757 nfsd_alloc_fcache_disposal(void)
758 {
759 	struct nfsd_fcache_disposal *l;
760 
761 	l = kmalloc(sizeof(*l), GFP_KERNEL);
762 	if (!l)
763 		return NULL;
764 	INIT_WORK(&l->work, nfsd_file_delayed_close);
765 	spin_lock_init(&l->lock);
766 	INIT_LIST_HEAD(&l->freeme);
767 	return l;
768 }
769 
770 static void
771 nfsd_free_fcache_disposal(struct nfsd_fcache_disposal *l)
772 {
773 	cancel_work_sync(&l->work);
774 	nfsd_file_dispose_list(&l->freeme);
775 	kfree(l);
776 }
777 
778 static void
779 nfsd_free_fcache_disposal_net(struct net *net)
780 {
781 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
782 	struct nfsd_fcache_disposal *l = nn->fcache_disposal;
783 
784 	nfsd_free_fcache_disposal(l);
785 }
786 
787 int
788 nfsd_file_cache_start_net(struct net *net)
789 {
790 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
791 
792 	nn->fcache_disposal = nfsd_alloc_fcache_disposal();
793 	return nn->fcache_disposal ? 0 : -ENOMEM;
794 }
795 
796 void
797 nfsd_file_cache_shutdown_net(struct net *net)
798 {
799 	nfsd_file_cache_purge(net);
800 	nfsd_free_fcache_disposal_net(net);
801 }
802 
803 void
804 nfsd_file_cache_shutdown(void)
805 {
806 	set_bit(NFSD_FILE_SHUTDOWN, &nfsd_file_lru_flags);
807 
808 	lease_unregister_notifier(&nfsd_file_lease_notifier);
809 	unregister_shrinker(&nfsd_file_shrinker);
810 	/*
811 	 * make sure all callers of nfsd_file_lru_cb are done before
812 	 * calling nfsd_file_cache_purge
813 	 */
814 	cancel_delayed_work_sync(&nfsd_filecache_laundrette);
815 	nfsd_file_cache_purge(NULL);
816 	list_lru_destroy(&nfsd_file_lru);
817 	rcu_barrier();
818 	fsnotify_put_group(nfsd_file_fsnotify_group);
819 	nfsd_file_fsnotify_group = NULL;
820 	kmem_cache_destroy(nfsd_file_slab);
821 	nfsd_file_slab = NULL;
822 	fsnotify_wait_marks_destroyed();
823 	kmem_cache_destroy(nfsd_file_mark_slab);
824 	nfsd_file_mark_slab = NULL;
825 	kvfree(nfsd_file_hashtbl);
826 	nfsd_file_hashtbl = NULL;
827 	destroy_workqueue(nfsd_filecache_wq);
828 	nfsd_filecache_wq = NULL;
829 }
830 
831 static bool
832 nfsd_match_cred(const struct cred *c1, const struct cred *c2)
833 {
834 	int i;
835 
836 	if (!uid_eq(c1->fsuid, c2->fsuid))
837 		return false;
838 	if (!gid_eq(c1->fsgid, c2->fsgid))
839 		return false;
840 	if (c1->group_info == NULL || c2->group_info == NULL)
841 		return c1->group_info == c2->group_info;
842 	if (c1->group_info->ngroups != c2->group_info->ngroups)
843 		return false;
844 	for (i = 0; i < c1->group_info->ngroups; i++) {
845 		if (!gid_eq(c1->group_info->gid[i], c2->group_info->gid[i]))
846 			return false;
847 	}
848 	return true;
849 }
850 
851 static struct nfsd_file *
852 nfsd_file_find_locked(struct inode *inode, unsigned int may_flags,
853 			unsigned int hashval, struct net *net)
854 {
855 	struct nfsd_file *nf;
856 	unsigned char need = may_flags & NFSD_FILE_MAY_MASK;
857 
858 	hlist_for_each_entry_rcu(nf, &nfsd_file_hashtbl[hashval].nfb_head,
859 				 nf_node, lockdep_is_held(&nfsd_file_hashtbl[hashval].nfb_lock)) {
860 		if (nf->nf_may != need)
861 			continue;
862 		if (nf->nf_inode != inode)
863 			continue;
864 		if (nf->nf_net != net)
865 			continue;
866 		if (!nfsd_match_cred(nf->nf_cred, current_cred()))
867 			continue;
868 		if (!test_bit(NFSD_FILE_HASHED, &nf->nf_flags))
869 			continue;
870 		if (nfsd_file_get(nf) != NULL)
871 			return nf;
872 	}
873 	return NULL;
874 }
875 
876 /**
877  * nfsd_file_is_cached - are there any cached open files for this fh?
878  * @inode: inode of the file to check
879  *
880  * Scan the hashtable for open files that match this fh. Returns true if there
881  * are any, and false if not.
882  */
883 bool
884 nfsd_file_is_cached(struct inode *inode)
885 {
886 	bool			ret = false;
887 	struct nfsd_file	*nf;
888 	unsigned int		hashval;
889 
890         hashval = (unsigned int)hash_long(inode->i_ino, NFSD_FILE_HASH_BITS);
891 
892 	rcu_read_lock();
893 	hlist_for_each_entry_rcu(nf, &nfsd_file_hashtbl[hashval].nfb_head,
894 				 nf_node) {
895 		if (inode == nf->nf_inode) {
896 			ret = true;
897 			break;
898 		}
899 	}
900 	rcu_read_unlock();
901 	trace_nfsd_file_is_cached(inode, hashval, (int)ret);
902 	return ret;
903 }
904 
905 static __be32
906 nfsd_do_file_acquire(struct svc_rqst *rqstp, struct svc_fh *fhp,
907 		     unsigned int may_flags, struct nfsd_file **pnf, bool open)
908 {
909 	__be32	status;
910 	struct net *net = SVC_NET(rqstp);
911 	struct nfsd_file *nf, *new;
912 	struct inode *inode;
913 	unsigned int hashval;
914 	bool retry = true;
915 
916 	/* FIXME: skip this if fh_dentry is already set? */
917 	status = fh_verify(rqstp, fhp, S_IFREG,
918 				may_flags|NFSD_MAY_OWNER_OVERRIDE);
919 	if (status != nfs_ok)
920 		return status;
921 
922 	inode = d_inode(fhp->fh_dentry);
923 	hashval = (unsigned int)hash_long(inode->i_ino, NFSD_FILE_HASH_BITS);
924 retry:
925 	rcu_read_lock();
926 	nf = nfsd_file_find_locked(inode, may_flags, hashval, net);
927 	rcu_read_unlock();
928 	if (nf)
929 		goto wait_for_construction;
930 
931 	new = nfsd_file_alloc(inode, may_flags, hashval, net);
932 	if (!new) {
933 		trace_nfsd_file_acquire(rqstp, hashval, inode, may_flags,
934 					NULL, nfserr_jukebox);
935 		return nfserr_jukebox;
936 	}
937 
938 	spin_lock(&nfsd_file_hashtbl[hashval].nfb_lock);
939 	nf = nfsd_file_find_locked(inode, may_flags, hashval, net);
940 	if (nf == NULL)
941 		goto open_file;
942 	spin_unlock(&nfsd_file_hashtbl[hashval].nfb_lock);
943 	nfsd_file_slab_free(&new->nf_rcu);
944 
945 wait_for_construction:
946 	wait_on_bit(&nf->nf_flags, NFSD_FILE_PENDING, TASK_UNINTERRUPTIBLE);
947 
948 	/* Did construction of this file fail? */
949 	if (!test_bit(NFSD_FILE_HASHED, &nf->nf_flags)) {
950 		if (!retry) {
951 			status = nfserr_jukebox;
952 			goto out;
953 		}
954 		retry = false;
955 		nfsd_file_put_noref(nf);
956 		goto retry;
957 	}
958 
959 	this_cpu_inc(nfsd_file_cache_hits);
960 
961 	if (!(may_flags & NFSD_MAY_NOT_BREAK_LEASE)) {
962 		bool write = (may_flags & NFSD_MAY_WRITE);
963 
964 		if (test_bit(NFSD_FILE_BREAK_READ, &nf->nf_flags) ||
965 		    (test_bit(NFSD_FILE_BREAK_WRITE, &nf->nf_flags) && write)) {
966 			status = nfserrno(nfsd_open_break_lease(
967 					file_inode(nf->nf_file), may_flags));
968 			if (status == nfs_ok) {
969 				clear_bit(NFSD_FILE_BREAK_READ, &nf->nf_flags);
970 				if (write)
971 					clear_bit(NFSD_FILE_BREAK_WRITE,
972 						  &nf->nf_flags);
973 			}
974 		}
975 	}
976 out:
977 	if (status == nfs_ok) {
978 		*pnf = nf;
979 	} else {
980 		nfsd_file_put(nf);
981 		nf = NULL;
982 	}
983 
984 	trace_nfsd_file_acquire(rqstp, hashval, inode, may_flags, nf, status);
985 	return status;
986 open_file:
987 	nf = new;
988 	/* Take reference for the hashtable */
989 	refcount_inc(&nf->nf_ref);
990 	__set_bit(NFSD_FILE_HASHED, &nf->nf_flags);
991 	__set_bit(NFSD_FILE_PENDING, &nf->nf_flags);
992 	list_lru_add(&nfsd_file_lru, &nf->nf_lru);
993 	hlist_add_head_rcu(&nf->nf_node, &nfsd_file_hashtbl[hashval].nfb_head);
994 	++nfsd_file_hashtbl[hashval].nfb_count;
995 	nfsd_file_hashtbl[hashval].nfb_maxcount = max(nfsd_file_hashtbl[hashval].nfb_maxcount,
996 			nfsd_file_hashtbl[hashval].nfb_count);
997 	spin_unlock(&nfsd_file_hashtbl[hashval].nfb_lock);
998 	if (atomic_long_inc_return(&nfsd_filecache_count) >= NFSD_FILE_LRU_THRESHOLD)
999 		nfsd_file_gc();
1000 
1001 	nf->nf_mark = nfsd_file_mark_find_or_create(nf);
1002 	if (nf->nf_mark) {
1003 		if (open) {
1004 			status = nfsd_open_verified(rqstp, fhp, may_flags,
1005 						    &nf->nf_file);
1006 			trace_nfsd_file_open(nf, status);
1007 		} else
1008 			status = nfs_ok;
1009 	} else
1010 		status = nfserr_jukebox;
1011 	/*
1012 	 * If construction failed, or we raced with a call to unlink()
1013 	 * then unhash.
1014 	 */
1015 	if (status != nfs_ok || inode->i_nlink == 0) {
1016 		bool do_free;
1017 		spin_lock(&nfsd_file_hashtbl[hashval].nfb_lock);
1018 		do_free = nfsd_file_unhash(nf);
1019 		spin_unlock(&nfsd_file_hashtbl[hashval].nfb_lock);
1020 		if (do_free)
1021 			nfsd_file_put_noref(nf);
1022 	}
1023 	clear_bit_unlock(NFSD_FILE_PENDING, &nf->nf_flags);
1024 	smp_mb__after_atomic();
1025 	wake_up_bit(&nf->nf_flags, NFSD_FILE_PENDING);
1026 	goto out;
1027 }
1028 
1029 /**
1030  * nfsd_file_acquire - Get a struct nfsd_file with an open file
1031  * @rqstp: the RPC transaction being executed
1032  * @fhp: the NFS filehandle of the file to be opened
1033  * @may_flags: NFSD_MAY_ settings for the file
1034  * @pnf: OUT: new or found "struct nfsd_file" object
1035  *
1036  * Returns nfs_ok and sets @pnf on success; otherwise an nfsstat in
1037  * network byte order is returned.
1038  */
1039 __be32
1040 nfsd_file_acquire(struct svc_rqst *rqstp, struct svc_fh *fhp,
1041 		  unsigned int may_flags, struct nfsd_file **pnf)
1042 {
1043 	return nfsd_do_file_acquire(rqstp, fhp, may_flags, pnf, true);
1044 }
1045 
1046 /**
1047  * nfsd_file_create - Get a struct nfsd_file, do not open
1048  * @rqstp: the RPC transaction being executed
1049  * @fhp: the NFS filehandle of the file just created
1050  * @may_flags: NFSD_MAY_ settings for the file
1051  * @pnf: OUT: new or found "struct nfsd_file" object
1052  *
1053  * Returns nfs_ok and sets @pnf on success; otherwise an nfsstat in
1054  * network byte order is returned.
1055  */
1056 __be32
1057 nfsd_file_create(struct svc_rqst *rqstp, struct svc_fh *fhp,
1058 		 unsigned int may_flags, struct nfsd_file **pnf)
1059 {
1060 	return nfsd_do_file_acquire(rqstp, fhp, may_flags, pnf, false);
1061 }
1062 
1063 /*
1064  * Note that fields may be added, removed or reordered in the future. Programs
1065  * scraping this file for info should test the labels to ensure they're
1066  * getting the correct field.
1067  */
1068 static int nfsd_file_cache_stats_show(struct seq_file *m, void *v)
1069 {
1070 	unsigned int i, count = 0, longest = 0;
1071 	unsigned long hits = 0;
1072 
1073 	/*
1074 	 * No need for spinlocks here since we're not terribly interested in
1075 	 * accuracy. We do take the nfsd_mutex simply to ensure that we
1076 	 * don't end up racing with server shutdown
1077 	 */
1078 	mutex_lock(&nfsd_mutex);
1079 	if (nfsd_file_hashtbl) {
1080 		for (i = 0; i < NFSD_FILE_HASH_SIZE; i++) {
1081 			count += nfsd_file_hashtbl[i].nfb_count;
1082 			longest = max(longest, nfsd_file_hashtbl[i].nfb_count);
1083 		}
1084 	}
1085 	mutex_unlock(&nfsd_mutex);
1086 
1087 	for_each_possible_cpu(i)
1088 		hits += per_cpu(nfsd_file_cache_hits, i);
1089 
1090 	seq_printf(m, "total entries: %u\n", count);
1091 	seq_printf(m, "longest chain: %u\n", longest);
1092 	seq_printf(m, "cache hits:    %lu\n", hits);
1093 	return 0;
1094 }
1095 
1096 int nfsd_file_cache_stats_open(struct inode *inode, struct file *file)
1097 {
1098 	return single_open(file, nfsd_file_cache_stats_show, NULL);
1099 }
1100