1	// SPDX-License-Identifier: GPL-2.0-only
2	/*
3	* linux/fs/nfs/dir.c
4	*
5	* Copyright (C) 1992 Rick Sladkey
6	*
7	* nfs directory handling functions
8	*
9	* 10 Apr 1996 Added silly rename for unlink --okir
10	* 28 Sep 1996 Improved directory cache --okir
11	* 23 Aug 1997 Claus Heine claus@momo.math.rwth-aachen.de
12	* Re-implemented silly rename for unlink, newly implemented
13	* silly rename for nfs_rename() following the suggestions
14	* of Olaf Kirch (okir) found in this file.
15	* Following Linus comments on my original hack, this version
16	* depends only on the dcache stuff and doesn't touch the inode
17	* layer (iput() and friends).
18	* 6 Jun 1999 Cache readdir lookups in the page cache. -DaveM
19	*/
20
21	#include <linux/compat.h>
22	#include <linux/module.h>
23	#include <linux/time.h>
24	#include <linux/errno.h>
25	#include <linux/stat.h>
26	#include <linux/fcntl.h>
27	#include <linux/string.h>
28	#include <linux/kernel.h>
29	#include <linux/slab.h>
30	#include <linux/mm.h>
31	#include <linux/sunrpc/clnt.h>
32	#include <linux/nfs_fs.h>
33	#include <linux/nfs_mount.h>
34	#include <linux/pagemap.h>
35	#include <linux/pagevec.h>
36	#include <linux/namei.h>
37	#include <linux/mount.h>
38	#include <linux/swap.h>
39	#include <linux/sched.h>
40	#include <linux/kmemleak.h>
41	#include <linux/xattr.h>
42	#include <linux/hash.h>
43
44	#include "delegation.h"
45	#include "iostat.h"
46	#include "internal.h"
47	#include "fscache.h"
48
49	#include "nfstrace.h"
50
51	/* #define NFS_DEBUG_VERBOSE 1 */
52
53	static int nfs_opendir(struct inode *, struct file *);
54	static int nfs_closedir(struct inode *, struct file *);
55	static int nfs_readdir(struct file *, struct dir_context *);
56	static int nfs_fsync_dir(struct file *, loff_t, loff_t, int);
57	static loff_t nfs_llseek_dir(struct file *, loff_t, int);
58	static void nfs_readdir_clear_array(struct folio *);
59
60	const struct file_operations nfs_dir_operations = {
61	.llseek = nfs_llseek_dir,
62	.read = generic_read_dir,
63	.iterate_shared = nfs_readdir,
64	.open = nfs_opendir,
65	.release = nfs_closedir,
66	.fsync = nfs_fsync_dir,
67	};
68
69	const struct address_space_operations nfs_dir_aops = {
70	.free_folio = nfs_readdir_clear_array,
71	};
72
73	#define NFS_INIT_DTSIZE PAGE_SIZE
74
75	static struct nfs_open_dir_context *
76	alloc_nfs_open_dir_context(struct inode *dir)
77	{
78	struct nfs_inode *nfsi = NFS_I(inode: dir);
79	struct nfs_open_dir_context *ctx;
80
81	ctx = kzalloc(size: sizeof(*ctx), GFP_KERNEL_ACCOUNT);
82	if (ctx != NULL) {
83	ctx->attr_gencount = nfsi->attr_gencount;
84	ctx->dtsize = NFS_INIT_DTSIZE;
85	spin_lock(lock: &dir->i_lock);
86	if (list_empty(head: &nfsi->open_files) &&
87	(nfsi->cache_validity & NFS_INO_DATA_INVAL_DEFER))
88	nfs_set_cache_invalid(inode: dir,
89	NFS_INO_INVALID_DATA |
90	NFS_INO_REVAL_FORCED);
91	list_add_tail_rcu(new: &ctx->list, head: &nfsi->open_files);
92	memcpy(ctx->verf, nfsi->cookieverf, sizeof(ctx->verf));
93	spin_unlock(lock: &dir->i_lock);
94	return ctx;
95	}
96	return ERR_PTR(error: -ENOMEM);
97	}
98
99	static void put_nfs_open_dir_context(struct inode *dir, struct nfs_open_dir_context *ctx)
100	{
101	spin_lock(lock: &dir->i_lock);
102	list_del_rcu(entry: &ctx->list);
103	spin_unlock(lock: &dir->i_lock);
104	kfree_rcu(ctx, rcu_head);
105	}
106
107	/*
108	* Open file
109	*/
110	static int
111	nfs_opendir(struct inode *inode, struct file *filp)
112	{
113	int res = 0;
114	struct nfs_open_dir_context *ctx;
115
116	dfprintk(FILE, "NFS: open dir(%pD2)\n", filp);
117
118	nfs_inc_stats(inode, stat: NFSIOS_VFSOPEN);
119
120	ctx = alloc_nfs_open_dir_context(dir: inode);
121	if (IS_ERR(ptr: ctx)) {
122	res = PTR_ERR(ptr: ctx);
123	goto out;
124	}
125	filp->private_data = ctx;
126	out:
127	return res;
128	}
129
130	static int
131	nfs_closedir(struct inode *inode, struct file *filp)
132	{
133	put_nfs_open_dir_context(dir: file_inode(f: filp), ctx: filp->private_data);
134	return 0;
135	}
136
137	struct nfs_cache_array_entry {
138	u64 cookie;
139	u64 ino;
140	const char *name;
141	unsigned int name_len;
142	unsigned char d_type;
143	};
144
145	struct nfs_cache_array {
146	u64 change_attr;
147	u64 last_cookie;
148	unsigned int size;
149	unsigned char folio_full : 1,
150	folio_is_eof : 1,
151	cookies_are_ordered : 1;
152	struct nfs_cache_array_entry array[];
153	};
154
155	struct nfs_readdir_descriptor {
156	struct file *file;
157	struct folio *folio;
158	struct dir_context *ctx;
159	pgoff_t folio_index;
160	pgoff_t folio_index_max;
161	u64 dir_cookie;
162	u64 last_cookie;
163	loff_t current_index;
164
165	__be32 verf[NFS_DIR_VERIFIER_SIZE];
166	unsigned long dir_verifier;
167	unsigned long timestamp;
168	unsigned long gencount;
169	unsigned long attr_gencount;
170	unsigned int cache_entry_index;
171	unsigned int buffer_fills;
172	unsigned int dtsize;
173	bool clear_cache;
174	bool plus;
175	bool eob;
176	bool eof;
177	};
178
179	static void nfs_set_dtsize(struct nfs_readdir_descriptor *desc, unsigned int sz)
180	{
181	struct nfs_server *server = NFS_SERVER(inode: file_inode(f: desc->file));
182	unsigned int maxsize = server->dtsize;
183
184	if (sz > maxsize)
185	sz = maxsize;
186	if (sz < NFS_MIN_FILE_IO_SIZE)
187	sz = NFS_MIN_FILE_IO_SIZE;
188	desc->dtsize = sz;
189	}
190
191	static void nfs_shrink_dtsize(struct nfs_readdir_descriptor *desc)
192	{
193	nfs_set_dtsize(desc, sz: desc->dtsize >> 1);
194	}
195
196	static void nfs_grow_dtsize(struct nfs_readdir_descriptor *desc)
197	{
198	nfs_set_dtsize(desc, sz: desc->dtsize << 1);
199	}
200
201	static void nfs_readdir_folio_init_array(struct folio *folio, u64 last_cookie,
202	u64 change_attr)
203	{
204	struct nfs_cache_array *array;
205
206	array = kmap_local_folio(folio, offset: 0);
207	array->change_attr = change_attr;
208	array->last_cookie = last_cookie;
209	array->size = 0;
210	array->folio_full = 0;
211	array->folio_is_eof = 0;
212	array->cookies_are_ordered = 1;
213	kunmap_local(array);
214	}
215
216	/*
217	* we are freeing strings created by nfs_add_to_readdir_array()
218	*/
219	static void nfs_readdir_clear_array(struct folio *folio)
220	{
221	struct nfs_cache_array *array;
222	unsigned int i;
223
224	array = kmap_local_folio(folio, offset: 0);
225	for (i = 0; i < array->size; i++)
226	kfree(objp: array->array[i].name);
227	array->size = 0;
228	kunmap_local(array);
229	}
230
231	static void nfs_readdir_folio_reinit_array(struct folio *folio, u64 last_cookie,
232	u64 change_attr)
233	{
234	nfs_readdir_clear_array(folio);
235	nfs_readdir_folio_init_array(folio, last_cookie, change_attr);
236	}
237
238	static struct folio *
239	nfs_readdir_folio_array_alloc(u64 last_cookie, gfp_t gfp_flags)
240	{
241	struct folio *folio = folio_alloc(gfp: gfp_flags, order: 0);
242	if (folio)
243	nfs_readdir_folio_init_array(folio, last_cookie, change_attr: 0);
244	return folio;
245	}
246
247	static void nfs_readdir_folio_array_free(struct folio *folio)
248	{
249	if (folio) {
250	nfs_readdir_clear_array(folio);
251	folio_put(folio);
252	}
253	}
254
255	static u64 nfs_readdir_array_index_cookie(struct nfs_cache_array *array)
256	{
257	return array->size == 0 ? array->last_cookie : array->array[0].cookie;
258	}
259
260	static void nfs_readdir_array_set_eof(struct nfs_cache_array *array)
261	{
262	array->folio_is_eof = 1;
263	array->folio_full = 1;
264	}
265
266	static bool nfs_readdir_array_is_full(struct nfs_cache_array *array)
267	{
268	return array->folio_full;
269	}
270
271	/*
272	* the caller is responsible for freeing qstr.name
273	* when called by nfs_readdir_add_to_array, the strings will be freed in
274	* nfs_clear_readdir_array()
275	*/
276	static const char *nfs_readdir_copy_name(const char *name, unsigned int len)
277	{
278	const char *ret = kmemdup_nul(s: name, len, GFP_KERNEL);
279
280	/*
281	* Avoid a kmemleak false positive. The pointer to the name is stored
282	* in a page cache page which kmemleak does not scan.
283	*/
284	if (ret != NULL)
285	kmemleak_not_leak(ptr: ret);
286	return ret;
287	}
288
289	static size_t nfs_readdir_array_maxentries(void)
290	{
291	return (PAGE_SIZE - sizeof(struct nfs_cache_array)) /
292	sizeof(struct nfs_cache_array_entry);
293	}
294
295	/*
296	* Check that the next array entry lies entirely within the page bounds
297	*/
298	static int nfs_readdir_array_can_expand(struct nfs_cache_array *array)
299	{
300	if (array->folio_full)
301	return -ENOSPC;
302	if (array->size == nfs_readdir_array_maxentries()) {
303	array->folio_full = 1;
304	return -ENOSPC;
305	}
306	return 0;
307	}
308
309	static int nfs_readdir_folio_array_append(struct folio *folio,
310	const struct nfs_entry *entry,
311	u64 *cookie)
312	{
313	struct nfs_cache_array *array;
314	struct nfs_cache_array_entry *cache_entry;
315	const char *name;
316	int ret = -ENOMEM;
317
318	name = nfs_readdir_copy_name(name: entry->name, len: entry->len);
319
320	array = kmap_local_folio(folio, offset: 0);
321	if (!name)
322	goto out;
323	ret = nfs_readdir_array_can_expand(array);
324	if (ret) {
325	kfree(objp: name);
326	goto out;
327	}
328
329	cache_entry = &array->array[array->size];
330	cache_entry->cookie = array->last_cookie;
331	cache_entry->ino = entry->ino;
332	cache_entry->d_type = entry->d_type;
333	cache_entry->name_len = entry->len;
334	cache_entry->name = name;
335	array->last_cookie = entry->cookie;
336	if (array->last_cookie <= cache_entry->cookie)
337	array->cookies_are_ordered = 0;
338	array->size++;
339	if (entry->eof != 0)
340	nfs_readdir_array_set_eof(array);
341	out:
342	*cookie = array->last_cookie;
343	kunmap_local(array);
344	return ret;
345	}
346
347	#define NFS_READDIR_COOKIE_MASK (U32_MAX >> 14)
348	/*
349	* Hash algorithm allowing content addressible access to sequences
350	* of directory cookies. Content is addressed by the value of the
351	* cookie index of the first readdir entry in a page.
352	*
353	* We select only the first 18 bits to avoid issues with excessive
354	* memory use for the page cache XArray. 18 bits should allow the caching
355	* of 262144 pages of sequences of readdir entries. Since each page holds
356	* 127 readdir entries for a typical 64-bit system, that works out to a
357	* cache of ~ 33 million entries per directory.
358	*/
359	static pgoff_t nfs_readdir_folio_cookie_hash(u64 cookie)
360	{
361	if (cookie == 0)
362	return 0;
363	return hash_64(val: cookie, bits: 18);
364	}
365
366	static bool nfs_readdir_folio_validate(struct folio *folio, u64 last_cookie,
367	u64 change_attr)
368	{
369	struct nfs_cache_array *array = kmap_local_folio(folio, offset: 0);
370	int ret = true;
371
372	if (array->change_attr != change_attr)
373	ret = false;
374	if (nfs_readdir_array_index_cookie(array) != last_cookie)
375	ret = false;
376	kunmap_local(array);
377	return ret;
378	}
379
380	static void nfs_readdir_folio_unlock_and_put(struct folio *folio)
381	{
382	folio_unlock(folio);
383	folio_put(folio);
384	}
385
386	static void nfs_readdir_folio_init_and_validate(struct folio *folio, u64 cookie,
387	u64 change_attr)
388	{
389	if (folio_test_uptodate(folio)) {
390	if (nfs_readdir_folio_validate(folio, last_cookie: cookie, change_attr))
391	return;
392	nfs_readdir_clear_array(folio);
393	}
394	nfs_readdir_folio_init_array(folio, last_cookie: cookie, change_attr);
395	folio_mark_uptodate(folio);
396	}
397
398	static struct folio *nfs_readdir_folio_get_locked(struct address_space *mapping,
399	u64 cookie, u64 change_attr)
400	{
401	pgoff_t index = nfs_readdir_folio_cookie_hash(cookie);
402	struct folio *folio;
403
404	folio = filemap_grab_folio(mapping, index);
405	if (IS_ERR(ptr: folio))
406	return NULL;
407	nfs_readdir_folio_init_and_validate(folio, cookie, change_attr);
408	return folio;
409	}
410
411	static u64 nfs_readdir_folio_last_cookie(struct folio *folio)
412	{
413	struct nfs_cache_array *array;
414	u64 ret;
415
416	array = kmap_local_folio(folio, offset: 0);
417	ret = array->last_cookie;
418	kunmap_local(array);
419	return ret;
420	}
421
422	static bool nfs_readdir_folio_needs_filling(struct folio *folio)
423	{
424	struct nfs_cache_array *array;
425	bool ret;
426
427	array = kmap_local_folio(folio, offset: 0);
428	ret = !nfs_readdir_array_is_full(array);
429	kunmap_local(array);
430	return ret;
431	}
432
433	static void nfs_readdir_folio_set_eof(struct folio *folio)
434	{
435	struct nfs_cache_array *array;
436
437	array = kmap_local_folio(folio, offset: 0);
438	nfs_readdir_array_set_eof(array);
439	kunmap_local(array);
440	}
441
442	static struct folio *nfs_readdir_folio_get_next(struct address_space *mapping,
443	u64 cookie, u64 change_attr)
444	{
445	pgoff_t index = nfs_readdir_folio_cookie_hash(cookie);
446	struct folio *folio;
447
448	folio = __filemap_get_folio(mapping, index,
449	FGP_LOCK|FGP_CREAT|FGP_NOFS|FGP_NOWAIT,
450	gfp: mapping_gfp_mask(mapping));
451	if (IS_ERR(ptr: folio))
452	return NULL;
453	nfs_readdir_folio_init_and_validate(folio, cookie, change_attr);
454	if (nfs_readdir_folio_last_cookie(folio) != cookie)
455	nfs_readdir_folio_reinit_array(folio, last_cookie: cookie, change_attr);
456	return folio;
457	}
458
459	static inline
460	int is_32bit_api(void)
461	{
462	#ifdef CONFIG_COMPAT
463	return in_compat_syscall();
464	#else
465	return (BITS_PER_LONG == 32);
466	#endif
467	}
468
469	static
470	bool nfs_readdir_use_cookie(const struct file *filp)
471	{
472	if ((filp->f_mode & FMODE_32BITHASH) ||
473	(!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
474	return false;
475	return true;
476	}
477
478	static void nfs_readdir_seek_next_array(struct nfs_cache_array *array,
479	struct nfs_readdir_descriptor *desc)
480	{
481	if (array->folio_full) {
482	desc->last_cookie = array->last_cookie;
483	desc->current_index += array->size;
484	desc->cache_entry_index = 0;
485	desc->folio_index++;
486	} else
487	desc->last_cookie = nfs_readdir_array_index_cookie(array);
488	}
489
490	static void nfs_readdir_rewind_search(struct nfs_readdir_descriptor *desc)
491	{
492	desc->current_index = 0;
493	desc->last_cookie = 0;
494	desc->folio_index = 0;
495	}
496
497	static int nfs_readdir_search_for_pos(struct nfs_cache_array *array,
498	struct nfs_readdir_descriptor *desc)
499	{
500	loff_t diff = desc->ctx->pos - desc->current_index;
501	unsigned int index;
502
503	if (diff < 0)
504	goto out_eof;
505	if (diff >= array->size) {
506	if (array->folio_is_eof)
507	goto out_eof;
508	nfs_readdir_seek_next_array(array, desc);
509	return -EAGAIN;
510	}
511
512	index = (unsigned int)diff;
513	desc->dir_cookie = array->array[index].cookie;
514	desc->cache_entry_index = index;
515	return 0;
516	out_eof:
517	desc->eof = true;
518	return -EBADCOOKIE;
519	}
520
521	static bool nfs_readdir_array_cookie_in_range(struct nfs_cache_array *array,
522	u64 cookie)
523	{
524	if (!array->cookies_are_ordered)
525	return true;
526	/* Optimisation for monotonically increasing cookies */
527	if (cookie >= array->last_cookie)
528	return false;
529	if (array->size && cookie < array->array[0].cookie)
530	return false;
531	return true;
532	}
533
534	static int nfs_readdir_search_for_cookie(struct nfs_cache_array *array,
535	struct nfs_readdir_descriptor *desc)
536	{
537	unsigned int i;
538	int status = -EAGAIN;
539
540	if (!nfs_readdir_array_cookie_in_range(array, cookie: desc->dir_cookie))
541	goto check_eof;
542
543	for (i = 0; i < array->size; i++) {
544	if (array->array[i].cookie == desc->dir_cookie) {
545	if (nfs_readdir_use_cookie(filp: desc->file))
546	desc->ctx->pos = desc->dir_cookie;
547	else
548	desc->ctx->pos = desc->current_index + i;
549	desc->cache_entry_index = i;
550	return 0;
551	}
552	}
553	check_eof:
554	if (array->folio_is_eof) {
555	status = -EBADCOOKIE;
556	if (desc->dir_cookie == array->last_cookie)
557	desc->eof = true;
558	} else
559	nfs_readdir_seek_next_array(array, desc);
560	return status;
561	}
562
563	static int nfs_readdir_search_array(struct nfs_readdir_descriptor *desc)
564	{
565	struct nfs_cache_array *array;
566	int status;
567
568	array = kmap_local_folio(folio: desc->folio, offset: 0);
569
570	if (desc->dir_cookie == 0)
571	status = nfs_readdir_search_for_pos(array, desc);
572	else
573	status = nfs_readdir_search_for_cookie(array, desc);
574
575	kunmap_local(array);
576	return status;
577	}
578
579	/* Fill a page with xdr information before transferring to the cache page */
580	static int nfs_readdir_xdr_filler(struct nfs_readdir_descriptor *desc,
581	__be32 *verf, u64 cookie,
582	struct page **pages, size_t bufsize,
583	__be32 *verf_res)
584	{
585	struct inode *inode = file_inode(f: desc->file);
586	struct nfs_readdir_arg arg = {
587	.dentry = file_dentry(file: desc->file),
588	.cred = desc->file->f_cred,
589	.verf = verf,
590	.cookie = cookie,
591	.pages = pages,
592	.page_len = bufsize,
593	.plus = desc->plus,
594	};
595	struct nfs_readdir_res res = {
596	.verf = verf_res,
597	};
598	unsigned long timestamp, gencount;
599	int error;
600
601	again:
602	timestamp = jiffies;
603	gencount = nfs_inc_attr_generation_counter();
604	desc->dir_verifier = nfs_save_change_attribute(dir: inode);
605	error = NFS_PROTO(inode)->readdir(&arg, &res);
606	if (error < 0) {
607	/* We requested READDIRPLUS, but the server doesn't grok it */
608	if (error == -ENOTSUPP && desc->plus) {
609	NFS_SERVER(inode)->caps &= ~NFS_CAP_READDIRPLUS;
610	desc->plus = arg.plus = false;
611	goto again;
612	}
613	goto error;
614	}
615	desc->timestamp = timestamp;
616	desc->gencount = gencount;
617	error:
618	return error;
619	}
620
621	static int xdr_decode(struct nfs_readdir_descriptor *desc,
622	struct nfs_entry *entry, struct xdr_stream *xdr)
623	{
624	struct inode *inode = file_inode(f: desc->file);
625	int error;
626
627	error = NFS_PROTO(inode)->decode_dirent(xdr, entry, desc->plus);
628	if (error)
629	return error;
630	entry->fattr->time_start = desc->timestamp;
631	entry->fattr->gencount = desc->gencount;
632	return 0;
633	}
634
635	/* Match file and dirent using either filehandle or fileid
636	* Note: caller is responsible for checking the fsid
637	*/
638	static
639	int nfs_same_file(struct dentry *dentry, struct nfs_entry *entry)
640	{
641	struct inode *inode;
642	struct nfs_inode *nfsi;
643
644	if (d_really_is_negative(dentry))
645	return 0;
646
647	inode = d_inode(dentry);
648	if (is_bad_inode(inode) || NFS_STALE(inode))
649	return 0;
650
651	nfsi = NFS_I(inode);
652	if (entry->fattr->fileid != nfsi->fileid)
653	return 0;
654	if (entry->fh->size && nfs_compare_fh(a: entry->fh, b: &nfsi->fh) != 0)
655	return 0;
656	return 1;
657	}
658
659	#define NFS_READDIR_CACHE_USAGE_THRESHOLD (8UL)
660
661	static bool nfs_use_readdirplus(struct inode *dir, struct dir_context *ctx,
662	unsigned int cache_hits,
663	unsigned int cache_misses)
664	{
665	if (!nfs_server_capable(inode: dir, NFS_CAP_READDIRPLUS))
666	return false;
667	if (ctx->pos == 0 ||
668	cache_hits + cache_misses > NFS_READDIR_CACHE_USAGE_THRESHOLD)
669	return true;
670	return false;
671	}
672
673	/*
674	* This function is called by the getattr code to request the
675	* use of readdirplus to accelerate any future lookups in the same
676	* directory.
677	*/
678	void nfs_readdir_record_entry_cache_hit(struct inode *dir)
679	{
680	struct nfs_inode *nfsi = NFS_I(inode: dir);
681	struct nfs_open_dir_context *ctx;
682
683	if (nfs_server_capable(inode: dir, NFS_CAP_READDIRPLUS) &&
684	S_ISDIR(dir->i_mode)) {
685	rcu_read_lock();
686	list_for_each_entry_rcu (ctx, &nfsi->open_files, list)
687	atomic_inc(v: &ctx->cache_hits);
688	rcu_read_unlock();
689	}
690	}
691
692	/*
693	* This function is mainly for use by nfs_getattr().
694	*
695	* If this is an 'ls -l', we want to force use of readdirplus.
696	*/
697	void nfs_readdir_record_entry_cache_miss(struct inode *dir)
698	{
699	struct nfs_inode *nfsi = NFS_I(inode: dir);
700	struct nfs_open_dir_context *ctx;
701
702	if (nfs_server_capable(inode: dir, NFS_CAP_READDIRPLUS) &&
703	S_ISDIR(dir->i_mode)) {
704	rcu_read_lock();
705	list_for_each_entry_rcu (ctx, &nfsi->open_files, list)
706	atomic_inc(v: &ctx->cache_misses);
707	rcu_read_unlock();
708	}
709	}
710
711	static void nfs_lookup_advise_force_readdirplus(struct inode *dir,
712	unsigned int flags)
713	{
714	if (nfs_server_capable(inode: dir, NFS_CAP_CASE_INSENSITIVE))
715	return;
716	if (flags & (LOOKUP_EXCL | LOOKUP_PARENT | LOOKUP_REVAL))
717	return;
718	nfs_readdir_record_entry_cache_miss(dir);
719	}
720
721	static
722	void nfs_prime_dcache(struct dentry *parent, struct nfs_entry *entry,
723	unsigned long dir_verifier)
724	{
725	struct qstr filename = QSTR_INIT(entry->name, entry->len);
726	DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
727	struct dentry *dentry;
728	struct dentry *alias;
729	struct inode *inode;
730	int status;
731
732	if (!(entry->fattr->valid & NFS_ATTR_FATTR_FILEID))
733	return;
734	if (!(entry->fattr->valid & NFS_ATTR_FATTR_FSID))
735	return;
736	if (filename.len == 0)
737	return;
738	/* Validate that the name doesn't contain any illegal '\0' */
739	if (strnlen(p: filename.name, maxlen: filename.len) != filename.len)
740	return;
741	/* ...or '/' */
742	if (strnchr(filename.name, filename.len, '/'))
743	return;
744	if (filename.name[0] == '.') {
745	if (filename.len == 1)
746	return;
747	if (filename.len == 2 && filename.name[1] == '.')
748	return;
749	}
750	filename.hash = full_name_hash(salt: parent, filename.name, filename.len);
751
752	dentry = d_lookup(parent, &filename);
753	again:
754	if (!dentry) {
755	dentry = d_alloc_parallel(parent, &filename, &wq);
756	if (IS_ERR(ptr: dentry))
757	return;
758	}
759	if (!d_in_lookup(dentry)) {
760	/* Is there a mountpoint here? If so, just exit */
761	if (!nfs_fsid_equal(a: &NFS_SB(s: dentry->d_sb)->fsid,
762	b: &entry->fattr->fsid))
763	goto out;
764	if (nfs_same_file(dentry, entry)) {
765	if (!entry->fh->size)
766	goto out;
767	nfs_set_verifier(dentry, verf: dir_verifier);
768	status = nfs_refresh_inode(d_inode(dentry), entry->fattr);
769	if (!status)
770	nfs_setsecurity(inode: d_inode(dentry), fattr: entry->fattr);
771	trace_nfs_readdir_lookup_revalidate(dir: d_inode(dentry: parent),
772	dentry, flags: 0, error: status);
773	goto out;
774	} else {
775	trace_nfs_readdir_lookup_revalidate_failed(
776	dir: d_inode(dentry: parent), dentry, flags: 0);
777	d_invalidate(dentry);
778	dput(dentry);
779	dentry = NULL;
780	goto again;
781	}
782	}
783	if (!entry->fh->size) {
784	d_lookup_done(dentry);
785	goto out;
786	}
787
788	inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr);
789	alias = d_splice_alias(inode, dentry);
790	d_lookup_done(dentry);
791	if (alias) {
792	if (IS_ERR(ptr: alias))
793	goto out;
794	dput(dentry);
795	dentry = alias;
796	}
797	nfs_set_verifier(dentry, verf: dir_verifier);
798	trace_nfs_readdir_lookup(dir: d_inode(dentry: parent), dentry, flags: 0);
799	out:
800	dput(dentry);
801	}
802
803	static int nfs_readdir_entry_decode(struct nfs_readdir_descriptor *desc,
804	struct nfs_entry *entry,
805	struct xdr_stream *stream)
806	{
807	int ret;
808
809	if (entry->fattr->label)
810	entry->fattr->label->len = NFS4_MAXLABELLEN;
811	ret = xdr_decode(desc, entry, xdr: stream);
812	if (ret || !desc->plus)
813	return ret;
814	nfs_prime_dcache(parent: file_dentry(file: desc->file), entry, dir_verifier: desc->dir_verifier);
815	return 0;
816	}
817
818	/* Perform conversion from xdr to cache array */
819	static int nfs_readdir_folio_filler(struct nfs_readdir_descriptor *desc,
820	struct nfs_entry *entry,
821	struct page **xdr_pages, unsigned int buflen,
822	struct folio **arrays, size_t narrays,
823	u64 change_attr)
824	{
825	struct address_space *mapping = desc->file->f_mapping;
826	struct folio *new, *folio = *arrays;
827	struct xdr_stream stream;
828	struct page *scratch;
829	struct xdr_buf buf;
830	u64 cookie;
831	int status;
832
833	scratch = alloc_page(GFP_KERNEL);
834	if (scratch == NULL)
835	return -ENOMEM;
836
837	xdr_init_decode_pages(xdr: &stream, buf: &buf, pages: xdr_pages, len: buflen);
838	xdr_set_scratch_page(xdr: &stream, page: scratch);
839
840	do {
841	status = nfs_readdir_entry_decode(desc, entry, stream: &stream);
842	if (status != 0)
843	break;
844
845	status = nfs_readdir_folio_array_append(folio, entry, cookie: &cookie);
846	if (status != -ENOSPC)
847	continue;
848
849	if (folio->mapping != mapping) {
850	if (!--narrays)
851	break;
852	new = nfs_readdir_folio_array_alloc(last_cookie: cookie, GFP_KERNEL);
853	if (!new)
854	break;
855	arrays++;
856	*arrays = folio = new;
857	} else {
858	new = nfs_readdir_folio_get_next(mapping, cookie,
859	change_attr);
860	if (!new)
861	break;
862	if (folio != *arrays)
863	nfs_readdir_folio_unlock_and_put(folio);
864	folio = new;
865	}
866	desc->folio_index_max++;
867	status = nfs_readdir_folio_array_append(folio, entry, cookie: &cookie);
868	} while (!status && !entry->eof);
869
870	switch (status) {
871	case -EBADCOOKIE:
872	if (!entry->eof)
873	break;
874	nfs_readdir_folio_set_eof(folio);
875	fallthrough;
876	case -EAGAIN:
877	status = 0;
878	break;
879	case -ENOSPC:
880	status = 0;
881	if (!desc->plus)
882	break;
883	while (!nfs_readdir_entry_decode(desc, entry, stream: &stream))
884	;
885	}
886
887	if (folio != *arrays)
888	nfs_readdir_folio_unlock_and_put(folio);
889
890	put_page(page: scratch);
891	return status;
892	}
893
894	static void nfs_readdir_free_pages(struct page **pages, size_t npages)
895	{
896	while (npages--)
897	put_page(page: pages[npages]);
898	kfree(objp: pages);
899	}
900
901	/*
902	* nfs_readdir_alloc_pages() will allocate pages that must be freed with a call
903	* to nfs_readdir_free_pages()
904	*/
905	static struct page **nfs_readdir_alloc_pages(size_t npages)
906	{
907	struct page **pages;
908	size_t i;
909
910	pages = kmalloc_array(n: npages, size: sizeof(*pages), GFP_KERNEL);
911	if (!pages)
912	return NULL;
913	for (i = 0; i < npages; i++) {
914	struct page *page = alloc_page(GFP_KERNEL);
915	if (page == NULL)
916	goto out_freepages;
917	pages[i] = page;
918	}
919	return pages;
920
921	out_freepages:
922	nfs_readdir_free_pages(pages, npages: i);
923	return NULL;
924	}
925
926	static int nfs_readdir_xdr_to_array(struct nfs_readdir_descriptor *desc,
927	__be32 *verf_arg, __be32 *verf_res,
928	struct folio **arrays, size_t narrays)
929	{
930	u64 change_attr;
931	struct page **pages;
932	struct folio *folio = *arrays;
933	struct nfs_entry *entry;
934	size_t array_size;
935	struct inode *inode = file_inode(f: desc->file);
936	unsigned int dtsize = desc->dtsize;
937	unsigned int pglen;
938	int status = -ENOMEM;
939
940	entry = kzalloc(size: sizeof(*entry), GFP_KERNEL);
941	if (!entry)
942	return -ENOMEM;
943	entry->cookie = nfs_readdir_folio_last_cookie(folio);
944	entry->fh = nfs_alloc_fhandle();
945	entry->fattr = nfs_alloc_fattr_with_label(server: NFS_SERVER(inode));
946	entry->server = NFS_SERVER(inode);
947	if (entry->fh == NULL || entry->fattr == NULL)
948	goto out;
949
950	array_size = (dtsize + PAGE_SIZE - 1) >> PAGE_SHIFT;
951	pages = nfs_readdir_alloc_pages(npages: array_size);
952	if (!pages)
953	goto out;
954
955	change_attr = inode_peek_iversion_raw(inode);
956	status = nfs_readdir_xdr_filler(desc, verf: verf_arg, cookie: entry->cookie, pages,
957	bufsize: dtsize, verf_res);
958	if (status < 0)
959	goto free_pages;
960
961	pglen = status;
962	if (pglen != 0)
963	status = nfs_readdir_folio_filler(desc, entry, xdr_pages: pages, buflen: pglen,
964	arrays, narrays, change_attr);
965	else
966	nfs_readdir_folio_set_eof(folio);
967	desc->buffer_fills++;
968
969	free_pages:
970	nfs_readdir_free_pages(pages, npages: array_size);
971	out:
972	nfs_free_fattr(fattr: entry->fattr);
973	nfs_free_fhandle(fh: entry->fh);
974	kfree(objp: entry);
975	return status;
976	}
977
978	static void nfs_readdir_folio_put(struct nfs_readdir_descriptor *desc)
979	{
980	folio_put(folio: desc->folio);
981	desc->folio = NULL;
982	}
983
984	static void
985	nfs_readdir_folio_unlock_and_put_cached(struct nfs_readdir_descriptor *desc)
986	{
987	folio_unlock(folio: desc->folio);
988	nfs_readdir_folio_put(desc);
989	}
990
991	static struct folio *
992	nfs_readdir_folio_get_cached(struct nfs_readdir_descriptor *desc)
993	{
994	struct address_space *mapping = desc->file->f_mapping;
995	u64 change_attr = inode_peek_iversion_raw(inode: mapping->host);
996	u64 cookie = desc->last_cookie;
997	struct folio *folio;
998
999	folio = nfs_readdir_folio_get_locked(mapping, cookie, change_attr);
1000	if (!folio)
1001	return NULL;
1002	if (desc->clear_cache && !nfs_readdir_folio_needs_filling(folio))
1003	nfs_readdir_folio_reinit_array(folio, last_cookie: cookie, change_attr);
1004	return folio;
1005	}
1006
1007	/*
1008	* Returns 0 if desc->dir_cookie was found on page desc->page_index
1009	* and locks the page to prevent removal from the page cache.
1010	*/
1011	static int find_and_lock_cache_page(struct nfs_readdir_descriptor *desc)
1012	{
1013	struct inode *inode = file_inode(f: desc->file);
1014	struct nfs_inode *nfsi = NFS_I(inode);
1015	__be32 verf[NFS_DIR_VERIFIER_SIZE];
1016	int res;
1017
1018	desc->folio = nfs_readdir_folio_get_cached(desc);
1019	if (!desc->folio)
1020	return -ENOMEM;
1021	if (nfs_readdir_folio_needs_filling(folio: desc->folio)) {
1022	/* Grow the dtsize if we had to go back for more pages */
1023	if (desc->folio_index == desc->folio_index_max)
1024	nfs_grow_dtsize(desc);
1025	desc->folio_index_max = desc->folio_index;
1026	trace_nfs_readdir_cache_fill(file: desc->file, verifier: nfsi->cookieverf,
1027	cookie: desc->last_cookie,
1028	page_index: desc->folio->index, dtsize: desc->dtsize);
1029	res = nfs_readdir_xdr_to_array(desc, verf_arg: nfsi->cookieverf, verf_res: verf,
1030	arrays: &desc->folio, narrays: 1);
1031	if (res < 0) {
1032	nfs_readdir_folio_unlock_and_put_cached(desc);
1033	trace_nfs_readdir_cache_fill_done(inode, error: res);
1034	if (res == -EBADCOOKIE || res == -ENOTSYNC) {
1035	invalidate_inode_pages2(mapping: desc->file->f_mapping);
1036	nfs_readdir_rewind_search(desc);
1037	trace_nfs_readdir_invalidate_cache_range(
1038	inode, range_start: 0, MAX_LFS_FILESIZE);
1039	return -EAGAIN;
1040	}
1041	return res;
1042	}
1043	/*
1044	* Set the cookie verifier if the page cache was empty
1045	*/
1046	if (desc->last_cookie == 0 &&
1047	memcmp(p: nfsi->cookieverf, q: verf, size: sizeof(nfsi->cookieverf))) {
1048	memcpy(nfsi->cookieverf, verf,
1049	sizeof(nfsi->cookieverf));
1050	invalidate_inode_pages2_range(mapping: desc->file->f_mapping, start: 1,
1051	end: -1);
1052	trace_nfs_readdir_invalidate_cache_range(
1053	inode, range_start: 1, MAX_LFS_FILESIZE);
1054	}
1055	desc->clear_cache = false;
1056	}
1057	res = nfs_readdir_search_array(desc);
1058	if (res == 0)
1059	return 0;
1060	nfs_readdir_folio_unlock_and_put_cached(desc);
1061	return res;
1062	}
1063
1064	/* Search for desc->dir_cookie from the beginning of the page cache */
1065	static int readdir_search_pagecache(struct nfs_readdir_descriptor *desc)
1066	{
1067	int res;
1068
1069	do {
1070	res = find_and_lock_cache_page(desc);
1071	} while (res == -EAGAIN);
1072	return res;
1073	}
1074
1075	#define NFS_READDIR_CACHE_MISS_THRESHOLD (16UL)
1076
1077	/*
1078	* Once we've found the start of the dirent within a page: fill 'er up...
1079	*/
1080	static void nfs_do_filldir(struct nfs_readdir_descriptor *desc,
1081	const __be32 *verf)
1082	{
1083	struct file *file = desc->file;
1084	struct nfs_cache_array *array;
1085	unsigned int i;
1086	bool first_emit = !desc->dir_cookie;
1087
1088	array = kmap_local_folio(folio: desc->folio, offset: 0);
1089	for (i = desc->cache_entry_index; i < array->size; i++) {
1090	struct nfs_cache_array_entry *ent;
1091
1092	/*
1093	* nfs_readdir_handle_cache_misses return force clear at
1094	* (cache_misses > NFS_READDIR_CACHE_MISS_THRESHOLD) for
1095	* readdir heuristic, NFS_READDIR_CACHE_MISS_THRESHOLD + 1
1096	* entries need be emitted here.
1097	*/
1098	if (first_emit && i > NFS_READDIR_CACHE_MISS_THRESHOLD + 2) {
1099	desc->eob = true;
1100	break;
1101	}
1102
1103	ent = &array->array[i];
1104	if (!dir_emit(ctx: desc->ctx, name: ent->name, namelen: ent->name_len,
1105	ino: nfs_compat_user_ino64(fileid: ent->ino), type: ent->d_type)) {
1106	desc->eob = true;
1107	break;
1108	}
1109	memcpy(desc->verf, verf, sizeof(desc->verf));
1110	if (i == array->size - 1) {
1111	desc->dir_cookie = array->last_cookie;
1112	nfs_readdir_seek_next_array(array, desc);
1113	} else {
1114	desc->dir_cookie = array->array[i + 1].cookie;
1115	desc->last_cookie = array->array[0].cookie;
1116	}
1117	if (nfs_readdir_use_cookie(filp: file))
1118	desc->ctx->pos = desc->dir_cookie;
1119	else
1120	desc->ctx->pos++;
1121	}
1122	if (array->folio_is_eof)
1123	desc->eof = !desc->eob;
1124
1125	kunmap_local(array);
1126	dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling ended @ cookie %llu\n",
1127	(unsigned long long)desc->dir_cookie);
1128	}
1129
1130	/*
1131	* If we cannot find a cookie in our cache, we suspect that this is
1132	* because it points to a deleted file, so we ask the server to return
1133	* whatever it thinks is the next entry. We then feed this to filldir.
1134	* If all goes well, we should then be able to find our way round the
1135	* cache on the next call to readdir_search_pagecache();
1136	*
1137	* NOTE: we cannot add the anonymous page to the pagecache because
1138	* the data it contains might not be page aligned. Besides,
1139	* we should already have a complete representation of the
1140	* directory in the page cache by the time we get here.
1141	*/
1142	static int uncached_readdir(struct nfs_readdir_descriptor *desc)
1143	{
1144	struct folio **arrays;
1145	size_t i, sz = 512;
1146	__be32 verf[NFS_DIR_VERIFIER_SIZE];
1147	int status = -ENOMEM;
1148
1149	dfprintk(DIRCACHE, "NFS: uncached_readdir() searching for cookie %llu\n",
1150	(unsigned long long)desc->dir_cookie);
1151
1152	arrays = kcalloc(n: sz, size: sizeof(*arrays), GFP_KERNEL);
1153	if (!arrays)
1154	goto out;
1155	arrays[0] = nfs_readdir_folio_array_alloc(last_cookie: desc->dir_cookie, GFP_KERNEL);
1156	if (!arrays[0])
1157	goto out;
1158
1159	desc->folio_index = 0;
1160	desc->cache_entry_index = 0;
1161	desc->last_cookie = desc->dir_cookie;
1162	desc->folio_index_max = 0;
1163
1164	trace_nfs_readdir_uncached(file: desc->file, verifier: desc->verf, cookie: desc->last_cookie,
1165	page_index: -1, dtsize: desc->dtsize);
1166
1167	status = nfs_readdir_xdr_to_array(desc, verf_arg: desc->verf, verf_res: verf, arrays, narrays: sz);
1168	if (status < 0) {
1169	trace_nfs_readdir_uncached_done(inode: file_inode(f: desc->file), error: status);
1170	goto out_free;
1171	}
1172
1173	for (i = 0; !desc->eob && i < sz && arrays[i]; i++) {
1174	desc->folio = arrays[i];
1175	nfs_do_filldir(desc, verf);
1176	}
1177	desc->folio = NULL;
1178
1179	/*
1180	* Grow the dtsize if we have to go back for more pages,
1181	* or shrink it if we're reading too many.
1182	*/
1183	if (!desc->eof) {
1184	if (!desc->eob)
1185	nfs_grow_dtsize(desc);
1186	else if (desc->buffer_fills == 1 &&
1187	i < (desc->folio_index_max >> 1))
1188	nfs_shrink_dtsize(desc);
1189	}
1190	out_free:
1191	for (i = 0; i < sz && arrays[i]; i++)
1192	nfs_readdir_folio_array_free(folio: arrays[i]);
1193	out:
1194	if (!nfs_readdir_use_cookie(filp: desc->file))
1195	nfs_readdir_rewind_search(desc);
1196	desc->folio_index_max = -1;
1197	kfree(objp: arrays);
1198	dfprintk(DIRCACHE, "NFS: %s: returns %d\n", __func__, status);
1199	return status;
1200	}
1201
1202	static bool nfs_readdir_handle_cache_misses(struct inode *inode,
1203	struct nfs_readdir_descriptor *desc,
1204	unsigned int cache_misses,
1205	bool force_clear)
1206	{
1207	if (desc->ctx->pos == 0 || !desc->plus)
1208	return false;
1209	if (cache_misses <= NFS_READDIR_CACHE_MISS_THRESHOLD && !force_clear)
1210	return false;
1211	trace_nfs_readdir_force_readdirplus(inode);
1212	return true;
1213	}
1214
1215	/* The file offset position represents the dirent entry number. A
1216	last cookie cache takes care of the common case of reading the
1217	whole directory.
1218	*/
1219	static int nfs_readdir(struct file *file, struct dir_context *ctx)
1220	{
1221	struct dentry *dentry = file_dentry(file);
1222	struct inode *inode = d_inode(dentry);
1223	struct nfs_inode *nfsi = NFS_I(inode);
1224	struct nfs_open_dir_context *dir_ctx = file->private_data;
1225	struct nfs_readdir_descriptor *desc;
1226	unsigned int cache_hits, cache_misses;
1227	bool force_clear;
1228	int res;
1229
1230	dfprintk(FILE, "NFS: readdir(%pD2) starting at cookie %llu\n",
1231	file, (long long)ctx->pos);
1232	nfs_inc_stats(inode, stat: NFSIOS_VFSGETDENTS);
1233
1234	/*
1235	* ctx->pos points to the dirent entry number.
1236	* *desc->dir_cookie has the cookie for the next entry. We have
1237	* to either find the entry with the appropriate number or
1238	* revalidate the cookie.
1239	*/
1240	nfs_revalidate_mapping(inode, mapping: file->f_mapping);
1241
1242	res = -ENOMEM;
1243	desc = kzalloc(size: sizeof(*desc), GFP_KERNEL);
1244	if (!desc)
1245	goto out;
1246	desc->file = file;
1247	desc->ctx = ctx;
1248	desc->folio_index_max = -1;
1249
1250	spin_lock(lock: &file->f_lock);
1251	desc->dir_cookie = dir_ctx->dir_cookie;
1252	desc->folio_index = dir_ctx->page_index;
1253	desc->last_cookie = dir_ctx->last_cookie;
1254	desc->attr_gencount = dir_ctx->attr_gencount;
1255	desc->eof = dir_ctx->eof;
1256	nfs_set_dtsize(desc, sz: dir_ctx->dtsize);
1257	memcpy(desc->verf, dir_ctx->verf, sizeof(desc->verf));
1258	cache_hits = atomic_xchg(v: &dir_ctx->cache_hits, new: 0);
1259	cache_misses = atomic_xchg(v: &dir_ctx->cache_misses, new: 0);
1260	force_clear = dir_ctx->force_clear;
1261	spin_unlock(lock: &file->f_lock);
1262
1263	if (desc->eof) {
1264	res = 0;
1265	goto out_free;
1266	}
1267
1268	desc->plus = nfs_use_readdirplus(dir: inode, ctx, cache_hits, cache_misses);
1269	force_clear = nfs_readdir_handle_cache_misses(inode, desc, cache_misses,
1270	force_clear);
1271	desc->clear_cache = force_clear;
1272
1273	do {
1274	res = readdir_search_pagecache(desc);
1275
1276	if (res == -EBADCOOKIE) {
1277	res = 0;
1278	/* This means either end of directory */
1279	if (desc->dir_cookie && !desc->eof) {
1280	/* Or that the server has 'lost' a cookie */
1281	res = uncached_readdir(desc);
1282	if (res == 0)
1283	continue;
1284	if (res == -EBADCOOKIE || res == -ENOTSYNC)
1285	res = 0;
1286	}
1287	break;
1288	}
1289	if (res == -ETOOSMALL && desc->plus) {
1290	nfs_zap_caches(inode);
1291	desc->plus = false;
1292	desc->eof = false;
1293	continue;
1294	}
1295	if (res < 0)
1296	break;
1297
1298	nfs_do_filldir(desc, verf: nfsi->cookieverf);
1299	nfs_readdir_folio_unlock_and_put_cached(desc);
1300	if (desc->folio_index == desc->folio_index_max)
1301	desc->clear_cache = force_clear;
1302	} while (!desc->eob && !desc->eof);
1303
1304	spin_lock(lock: &file->f_lock);
1305	dir_ctx->dir_cookie = desc->dir_cookie;
1306	dir_ctx->last_cookie = desc->last_cookie;
1307	dir_ctx->attr_gencount = desc->attr_gencount;
1308	dir_ctx->page_index = desc->folio_index;
1309	dir_ctx->force_clear = force_clear;
1310	dir_ctx->eof = desc->eof;
1311	dir_ctx->dtsize = desc->dtsize;
1312	memcpy(dir_ctx->verf, desc->verf, sizeof(dir_ctx->verf));
1313	spin_unlock(lock: &file->f_lock);
1314	out_free:
1315	kfree(objp: desc);
1316
1317	out:
1318	dfprintk(FILE, "NFS: readdir(%pD2) returns %d\n", file, res);
1319	return res;
1320	}
1321
1322	static loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int whence)
1323	{
1324	struct nfs_open_dir_context *dir_ctx = filp->private_data;
1325
1326	dfprintk(FILE, "NFS: llseek dir(%pD2, %lld, %d)\n",
1327	filp, offset, whence);
1328
1329	switch (whence) {
1330	default:
1331	return -EINVAL;
1332	case SEEK_SET:
1333	if (offset < 0)
1334	return -EINVAL;
1335	spin_lock(lock: &filp->f_lock);
1336	break;
1337	case SEEK_CUR:
1338	if (offset == 0)
1339	return filp->f_pos;
1340	spin_lock(lock: &filp->f_lock);
1341	offset += filp->f_pos;
1342	if (offset < 0) {
1343	spin_unlock(lock: &filp->f_lock);
1344	return -EINVAL;
1345	}
1346	}
1347	if (offset != filp->f_pos) {
1348	filp->f_pos = offset;
1349	dir_ctx->page_index = 0;
1350	if (!nfs_readdir_use_cookie(filp)) {
1351	dir_ctx->dir_cookie = 0;
1352	dir_ctx->last_cookie = 0;
1353	} else {
1354	dir_ctx->dir_cookie = offset;
1355	dir_ctx->last_cookie = offset;
1356	}
1357	dir_ctx->eof = false;
1358	}
1359	spin_unlock(lock: &filp->f_lock);
1360	return offset;
1361	}
1362
1363	/*
1364	* All directory operations under NFS are synchronous, so fsync()
1365	* is a dummy operation.
1366	*/
1367	static int nfs_fsync_dir(struct file *filp, loff_t start, loff_t end,
1368	int datasync)
1369	{
1370	dfprintk(FILE, "NFS: fsync dir(%pD2) datasync %d\n", filp, datasync);
1371
1372	nfs_inc_stats(inode: file_inode(f: filp), stat: NFSIOS_VFSFSYNC);
1373	return 0;
1374	}
1375
1376	/**
1377	* nfs_force_lookup_revalidate - Mark the directory as having changed
1378	* @dir: pointer to directory inode
1379	*
1380	* This forces the revalidation code in nfs_lookup_revalidate() to do a
1381	* full lookup on all child dentries of 'dir' whenever a change occurs
1382	* on the server that might have invalidated our dcache.
1383	*
1384	* Note that we reserve bit '0' as a tag to let us know when a dentry
1385	* was revalidated while holding a delegation on its inode.
1386	*
1387	* The caller should be holding dir->i_lock
1388	*/
1389	void nfs_force_lookup_revalidate(struct inode *dir)
1390	{
1391	NFS_I(inode: dir)->cache_change_attribute += 2;
1392	}
1393	EXPORT_SYMBOL_GPL(nfs_force_lookup_revalidate);
1394
1395	/**
1396	* nfs_verify_change_attribute - Detects NFS remote directory changes
1397	* @dir: pointer to parent directory inode
1398	* @verf: previously saved change attribute
1399	*
1400	* Return "false" if the verifiers doesn't match the change attribute.
1401	* This would usually indicate that the directory contents have changed on
1402	* the server, and that any dentries need revalidating.
1403	*/
1404	static bool nfs_verify_change_attribute(struct inode *dir, unsigned long verf)
1405	{
1406	return (verf & ~1UL) == nfs_save_change_attribute(dir);
1407	}
1408
1409	static void nfs_set_verifier_delegated(unsigned long *verf)
1410	{
1411	*verf |= 1UL;
1412	}
1413
1414	#if IS_ENABLED(CONFIG_NFS_V4)
1415	static void nfs_unset_verifier_delegated(unsigned long *verf)
1416	{
1417	*verf &= ~1UL;
1418	}
1419	#endif /* IS_ENABLED(CONFIG_NFS_V4) */
1420
1421	static bool nfs_test_verifier_delegated(unsigned long verf)
1422	{
1423	return verf & 1;
1424	}
1425
1426	static bool nfs_verifier_is_delegated(struct dentry *dentry)
1427	{
1428	return nfs_test_verifier_delegated(verf: dentry->d_time);
1429	}
1430
1431	static void nfs_set_verifier_locked(struct dentry *dentry, unsigned long verf)
1432	{
1433	struct inode *inode = d_inode(dentry);
1434	struct inode *dir = d_inode(dentry: dentry->d_parent);
1435
1436	if (!nfs_verify_change_attribute(dir, verf))
1437	return;
1438	if (inode && NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
1439	nfs_set_verifier_delegated(verf: &verf);
1440	dentry->d_time = verf;
1441	}
1442
1443	/**
1444	* nfs_set_verifier - save a parent directory verifier in the dentry
1445	* @dentry: pointer to dentry
1446	* @verf: verifier to save
1447	*
1448	* Saves the parent directory verifier in @dentry. If the inode has
1449	* a delegation, we also tag the dentry as having been revalidated
1450	* while holding a delegation so that we know we don't have to
1451	* look it up again after a directory change.
1452	*/
1453	void nfs_set_verifier(struct dentry *dentry, unsigned long verf)
1454	{
1455
1456	spin_lock(lock: &dentry->d_lock);
1457	nfs_set_verifier_locked(dentry, verf);
1458	spin_unlock(lock: &dentry->d_lock);
1459	}
1460	EXPORT_SYMBOL_GPL(nfs_set_verifier);
1461
1462	#if IS_ENABLED(CONFIG_NFS_V4)
1463	/**
1464	* nfs_clear_verifier_delegated - clear the dir verifier delegation tag
1465	* @inode: pointer to inode
1466	*
1467	* Iterates through the dentries in the inode alias list and clears
1468	* the tag used to indicate that the dentry has been revalidated
1469	* while holding a delegation.
1470	* This function is intended for use when the delegation is being
1471	* returned or revoked.
1472	*/
1473	void nfs_clear_verifier_delegated(struct inode *inode)
1474	{
1475	struct dentry *alias;
1476
1477	if (!inode)
1478	return;
1479	spin_lock(lock: &inode->i_lock);
1480	hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
1481	spin_lock(lock: &alias->d_lock);
1482	nfs_unset_verifier_delegated(verf: &alias->d_time);
1483	spin_unlock(lock: &alias->d_lock);
1484	}
1485	spin_unlock(lock: &inode->i_lock);
1486	}
1487	EXPORT_SYMBOL_GPL(nfs_clear_verifier_delegated);
1488	#endif /* IS_ENABLED(CONFIG_NFS_V4) */
1489
1490	static int nfs_dentry_verify_change(struct inode *dir, struct dentry *dentry)
1491	{
1492	if (nfs_server_capable(inode: dir, NFS_CAP_CASE_INSENSITIVE) &&
1493	d_really_is_negative(dentry))
1494	return dentry->d_time == inode_peek_iversion_raw(inode: dir);
1495	return nfs_verify_change_attribute(dir, verf: dentry->d_time);
1496	}
1497
1498	/*
1499	* A check for whether or not the parent directory has changed.
1500	* In the case it has, we assume that the dentries are untrustworthy
1501	* and may need to be looked up again.
1502	* If rcu_walk prevents us from performing a full check, return 0.
1503	*/
1504	static int nfs_check_verifier(struct inode *dir, struct dentry *dentry,
1505	int rcu_walk)
1506	{
1507	if (IS_ROOT(dentry))
1508	return 1;
1509	if (NFS_SERVER(inode: dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONE)
1510	return 0;
1511	if (!nfs_dentry_verify_change(dir, dentry))
1512	return 0;
1513	/* Revalidate nfsi->cache_change_attribute before we declare a match */
1514	if (nfs_mapping_need_revalidate_inode(inode: dir)) {
1515	if (rcu_walk)
1516	return 0;
1517	if (__nfs_revalidate_inode(NFS_SERVER(inode: dir), dir) < 0)
1518	return 0;
1519	}
1520	if (!nfs_dentry_verify_change(dir, dentry))
1521	return 0;
1522	return 1;
1523	}
1524
1525	/*
1526	* Use intent information to check whether or not we're going to do
1527	* an O_EXCL create using this path component.
1528	*/
1529	static int nfs_is_exclusive_create(struct inode *dir, unsigned int flags)
1530	{
1531	if (NFS_PROTO(inode: dir)->version == 2)
1532	return 0;
1533	return flags & LOOKUP_EXCL;
1534	}
1535
1536	/*
1537	* Inode and filehandle revalidation for lookups.
1538	*
1539	* We force revalidation in the cases where the VFS sets LOOKUP_REVAL,
1540	* or if the intent information indicates that we're about to open this
1541	* particular file and the "nocto" mount flag is not set.
1542	*
1543	*/
1544	static
1545	int nfs_lookup_verify_inode(struct inode *inode, unsigned int flags)
1546	{
1547	struct nfs_server *server = NFS_SERVER(inode);
1548	int ret;
1549
1550	if (IS_AUTOMOUNT(inode))
1551	return 0;
1552
1553	if (flags & LOOKUP_OPEN) {
1554	switch (inode->i_mode & S_IFMT) {
1555	case S_IFREG:
1556	/* A NFSv4 OPEN will revalidate later */
1557	if (server->caps & NFS_CAP_ATOMIC_OPEN)
1558	goto out;
1559	fallthrough;
1560	case S_IFDIR:
1561	if (server->flags & NFS_MOUNT_NOCTO)
1562	break;
1563	/* NFS close-to-open cache consistency validation */
1564	goto out_force;
1565	}
1566	}
1567
1568	/* VFS wants an on-the-wire revalidation */
1569	if (flags & LOOKUP_REVAL)
1570	goto out_force;
1571	out:
1572	if (inode->i_nlink > 0 ||
1573	(inode->i_nlink == 0 &&
1574	test_bit(NFS_INO_PRESERVE_UNLINKED, &NFS_I(inode)->flags)))
1575	return 0;
1576	else
1577	return -ESTALE;
1578	out_force:
1579	if (flags & LOOKUP_RCU)
1580	return -ECHILD;
1581	ret = __nfs_revalidate_inode(server, inode);
1582	if (ret != 0)
1583	return ret;
1584	goto out;
1585	}
1586
1587	static void nfs_mark_dir_for_revalidate(struct inode *inode)
1588	{
1589	spin_lock(lock: &inode->i_lock);
1590	nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE);
1591	spin_unlock(lock: &inode->i_lock);
1592	}
1593
1594	/*
1595	* We judge how long we want to trust negative
1596	* dentries by looking at the parent inode mtime.
1597	*
1598	* If parent mtime has changed, we revalidate, else we wait for a
1599	* period corresponding to the parent's attribute cache timeout value.
1600	*
1601	* If LOOKUP_RCU prevents us from performing a full check, return 1
1602	* suggesting a reval is needed.
1603	*
1604	* Note that when creating a new file, or looking up a rename target,
1605	* then it shouldn't be necessary to revalidate a negative dentry.
1606	*/
1607	static inline
1608	int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry,
1609	unsigned int flags)
1610	{
1611	if (flags & (LOOKUP_CREATE | LOOKUP_RENAME_TARGET))
1612	return 0;
1613	if (NFS_SERVER(inode: dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONEG)
1614	return 1;
1615	/* Case insensitive server? Revalidate negative dentries */
1616	if (nfs_server_capable(inode: dir, NFS_CAP_CASE_INSENSITIVE))
1617	return 1;
1618	return !nfs_check_verifier(dir, dentry, rcu_walk: flags & LOOKUP_RCU);
1619	}
1620
1621	static int
1622	nfs_lookup_revalidate_done(struct inode *dir, struct dentry *dentry,
1623	struct inode *inode, int error)
1624	{
1625	switch (error) {
1626	case 1:
1627	break;
1628	case 0:
1629	/*
1630	* We can't d_drop the root of a disconnected tree:
1631	* its d_hash is on the s_anon list and d_drop() would hide
1632	* it from shrink_dcache_for_unmount(), leading to busy
1633	* inodes on unmount and further oopses.
1634	*/
1635	if (inode && IS_ROOT(dentry))
1636	error = 1;
1637	break;
1638	}
1639	trace_nfs_lookup_revalidate_exit(dir, dentry, flags: 0, error);
1640	return error;
1641	}
1642
1643	static int
1644	nfs_lookup_revalidate_negative(struct inode *dir, struct dentry *dentry,
1645	unsigned int flags)
1646	{
1647	int ret = 1;
1648	if (nfs_neg_need_reval(dir, dentry, flags)) {
1649	if (flags & LOOKUP_RCU)
1650	return -ECHILD;
1651	ret = 0;
1652	}
1653	return nfs_lookup_revalidate_done(dir, dentry, NULL, error: ret);
1654	}
1655
1656	static int
1657	nfs_lookup_revalidate_delegated(struct inode *dir, struct dentry *dentry,
1658	struct inode *inode)
1659	{
1660	nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1661	return nfs_lookup_revalidate_done(dir, dentry, inode, error: 1);
1662	}
1663
1664	static int nfs_lookup_revalidate_dentry(struct inode *dir,
1665	struct dentry *dentry,
1666	struct inode *inode, unsigned int flags)
1667	{
1668	struct nfs_fh *fhandle;
1669	struct nfs_fattr *fattr;
1670	unsigned long dir_verifier;
1671	int ret;
1672
1673	trace_nfs_lookup_revalidate_enter(dir, dentry, flags);
1674
1675	ret = -ENOMEM;
1676	fhandle = nfs_alloc_fhandle();
1677	fattr = nfs_alloc_fattr_with_label(server: NFS_SERVER(inode));
1678	if (fhandle == NULL || fattr == NULL)
1679	goto out;
1680
1681	dir_verifier = nfs_save_change_attribute(dir);
1682	ret = NFS_PROTO(inode: dir)->lookup(dir, dentry, fhandle, fattr);
1683	if (ret < 0) {
1684	switch (ret) {
1685	case -ESTALE:
1686	case -ENOENT:
1687	ret = 0;
1688	break;
1689	case -ETIMEDOUT:
1690	if (NFS_SERVER(inode)->flags & NFS_MOUNT_SOFTREVAL)
1691	ret = 1;
1692	}
1693	goto out;
1694	}
1695
1696	/* Request help from readdirplus */
1697	nfs_lookup_advise_force_readdirplus(dir, flags);
1698
1699	ret = 0;
1700	if (nfs_compare_fh(a: NFS_FH(inode), b: fhandle))
1701	goto out;
1702	if (nfs_refresh_inode(inode, fattr) < 0)
1703	goto out;
1704
1705	nfs_setsecurity(inode, fattr);
1706	nfs_set_verifier(dentry, dir_verifier);
1707
1708	ret = 1;
1709	out:
1710	nfs_free_fattr(fattr);
1711	nfs_free_fhandle(fh: fhandle);
1712
1713	/*
1714	* If the lookup failed despite the dentry change attribute being
1715	* a match, then we should revalidate the directory cache.
1716	*/
1717	if (!ret && nfs_dentry_verify_change(dir, dentry))
1718	nfs_mark_dir_for_revalidate(inode: dir);
1719	return nfs_lookup_revalidate_done(dir, dentry, inode, error: ret);
1720	}
1721
1722	/*
1723	* This is called every time the dcache has a lookup hit,
1724	* and we should check whether we can really trust that
1725	* lookup.
1726	*
1727	* NOTE! The hit can be a negative hit too, don't assume
1728	* we have an inode!
1729	*
1730	* If the parent directory is seen to have changed, we throw out the
1731	* cached dentry and do a new lookup.
1732	*/
1733	static int
1734	nfs_do_lookup_revalidate(struct inode *dir, struct dentry *dentry,
1735	unsigned int flags)
1736	{
1737	struct inode *inode;
1738	int error;
1739
1740	nfs_inc_stats(inode: dir, stat: NFSIOS_DENTRYREVALIDATE);
1741	inode = d_inode(dentry);
1742
1743	if (!inode)
1744	return nfs_lookup_revalidate_negative(dir, dentry, flags);
1745
1746	if (is_bad_inode(inode)) {
1747	dfprintk(LOOKUPCACHE, "%s: %pd2 has dud inode\n",
1748	__func__, dentry);
1749	goto out_bad;
1750	}
1751
1752	if ((flags & LOOKUP_RENAME_TARGET) && d_count(dentry) < 2 &&
1753	nfs_server_capable(inode: dir, NFS_CAP_CASE_INSENSITIVE))
1754	goto out_bad;
1755
1756	if (nfs_verifier_is_delegated(dentry))
1757	return nfs_lookup_revalidate_delegated(dir, dentry, inode);
1758
1759	/* Force a full look up iff the parent directory has changed */
1760	if (!(flags & (LOOKUP_EXCL | LOOKUP_REVAL)) &&
1761	nfs_check_verifier(dir, dentry, rcu_walk: flags & LOOKUP_RCU)) {
1762	error = nfs_lookup_verify_inode(inode, flags);
1763	if (error) {
1764	if (error == -ESTALE)
1765	nfs_mark_dir_for_revalidate(inode: dir);
1766	goto out_bad;
1767	}
1768	goto out_valid;
1769	}
1770
1771	if (flags & LOOKUP_RCU)
1772	return -ECHILD;
1773
1774	if (NFS_STALE(inode))
1775	goto out_bad;
1776
1777	return nfs_lookup_revalidate_dentry(dir, dentry, inode, flags);
1778	out_valid:
1779	return nfs_lookup_revalidate_done(dir, dentry, inode, error: 1);
1780	out_bad:
1781	if (flags & LOOKUP_RCU)
1782	return -ECHILD;
1783	return nfs_lookup_revalidate_done(dir, dentry, inode, error: 0);
1784	}
1785
1786	static int
1787	__nfs_lookup_revalidate(struct dentry *dentry, unsigned int flags,
1788	int (*reval)(struct inode *, struct dentry *, unsigned int))
1789	{
1790	struct dentry *parent;
1791	struct inode *dir;
1792	int ret;
1793
1794	if (flags & LOOKUP_RCU) {
1795	if (dentry->d_fsdata == NFS_FSDATA_BLOCKED)
1796	return -ECHILD;
1797	parent = READ_ONCE(dentry->d_parent);
1798	dir = d_inode_rcu(dentry: parent);
1799	if (!dir)
1800	return -ECHILD;
1801	ret = reval(dir, dentry, flags);
1802	if (parent != READ_ONCE(dentry->d_parent))
1803	return -ECHILD;
1804	} else {
1805	/* Wait for unlink to complete */
1806	wait_var_event(&dentry->d_fsdata,
1807	dentry->d_fsdata != NFS_FSDATA_BLOCKED);
1808	parent = dget_parent(dentry);
1809	ret = reval(d_inode(dentry: parent), dentry, flags);
1810	dput(parent);
1811	}
1812	return ret;
1813	}
1814
1815	static int nfs_lookup_revalidate(struct dentry *dentry, unsigned int flags)
1816	{
1817	return __nfs_lookup_revalidate(dentry, flags, reval: nfs_do_lookup_revalidate);
1818	}
1819
1820	/*
1821	* A weaker form of d_revalidate for revalidating just the d_inode(dentry)
1822	* when we don't really care about the dentry name. This is called when a
1823	* pathwalk ends on a dentry that was not found via a normal lookup in the
1824	* parent dir (e.g.: ".", "..", procfs symlinks or mountpoint traversals).
1825	*
1826	* In this situation, we just want to verify that the inode itself is OK
1827	* since the dentry might have changed on the server.
1828	*/
1829	static int nfs_weak_revalidate(struct dentry *dentry, unsigned int flags)
1830	{
1831	struct inode *inode = d_inode(dentry);
1832	int error = 0;
1833
1834	/*
1835	* I believe we can only get a negative dentry here in the case of a
1836	* procfs-style symlink. Just assume it's correct for now, but we may
1837	* eventually need to do something more here.
1838	*/
1839	if (!inode) {
1840	dfprintk(LOOKUPCACHE, "%s: %pd2 has negative inode\n",
1841	__func__, dentry);
1842	return 1;
1843	}
1844
1845	if (is_bad_inode(inode)) {
1846	dfprintk(LOOKUPCACHE, "%s: %pd2 has dud inode\n",
1847	__func__, dentry);
1848	return 0;
1849	}
1850
1851	error = nfs_lookup_verify_inode(inode, flags);
1852	dfprintk(LOOKUPCACHE, "NFS: %s: inode %lu is %s\n",
1853	__func__, inode->i_ino, error ? "invalid" : "valid");
1854	return !error;
1855	}
1856
1857	/*
1858	* This is called from dput() when d_count is going to 0.
1859	*/
1860	static int nfs_dentry_delete(const struct dentry *dentry)
1861	{
1862	dfprintk(VFS, "NFS: dentry_delete(%pd2, %x)\n",
1863	dentry, dentry->d_flags);
1864
1865	/* Unhash any dentry with a stale inode */
1866	if (d_really_is_positive(dentry) && NFS_STALE(inode: d_inode(dentry)))
1867	return 1;
1868
1869	if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1870	/* Unhash it, so that ->d_iput() would be called */
1871	return 1;
1872	}
1873	if (!(dentry->d_sb->s_flags & SB_ACTIVE)) {
1874	/* Unhash it, so that ancestors of killed async unlink
1875	* files will be cleaned up during umount */
1876	return 1;
1877	}
1878	return 0;
1879
1880	}
1881
1882	/* Ensure that we revalidate inode->i_nlink */
1883	static void nfs_drop_nlink(struct inode *inode)
1884	{
1885	spin_lock(lock: &inode->i_lock);
1886	/* drop the inode if we're reasonably sure this is the last link */
1887	if (inode->i_nlink > 0)
1888	drop_nlink(inode);
1889	NFS_I(inode)->attr_gencount = nfs_inc_attr_generation_counter();
1890	nfs_set_cache_invalid(
1891	inode, NFS_INO_INVALID_CHANGE | NFS_INO_INVALID_CTIME |
1892	NFS_INO_INVALID_NLINK);
1893	spin_unlock(lock: &inode->i_lock);
1894	}
1895
1896	/*
1897	* Called when the dentry loses inode.
1898	* We use it to clean up silly-renamed files.
1899	*/
1900	static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode)
1901	{
1902	if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1903	nfs_complete_unlink(dentry, inode);
1904	nfs_drop_nlink(inode);
1905	}
1906	iput(inode);
1907	}
1908
1909	static void nfs_d_release(struct dentry *dentry)
1910	{
1911	/* free cached devname value, if it survived that far */
1912	if (unlikely(dentry->d_fsdata)) {
1913	if (dentry->d_flags & DCACHE_NFSFS_RENAMED)
1914	WARN_ON(1);
1915	else
1916	kfree(objp: dentry->d_fsdata);
1917	}
1918	}
1919
1920	const struct dentry_operations nfs_dentry_operations = {
1921	.d_revalidate = nfs_lookup_revalidate,
1922	.d_weak_revalidate = nfs_weak_revalidate,
1923	.d_delete = nfs_dentry_delete,
1924	.d_iput = nfs_dentry_iput,
1925	.d_automount = nfs_d_automount,
1926	.d_release = nfs_d_release,
1927	};
1928	EXPORT_SYMBOL_GPL(nfs_dentry_operations);
1929
1930	struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
1931	{
1932	struct dentry *res;
1933	struct inode *inode = NULL;
1934	struct nfs_fh *fhandle = NULL;
1935	struct nfs_fattr *fattr = NULL;
1936	unsigned long dir_verifier;
1937	int error;
1938
1939	dfprintk(VFS, "NFS: lookup(%pd2)\n", dentry);
1940	nfs_inc_stats(inode: dir, stat: NFSIOS_VFSLOOKUP);
1941
1942	if (unlikely(dentry->d_name.len > NFS_SERVER(dir)->namelen))
1943	return ERR_PTR(error: -ENAMETOOLONG);
1944
1945	/*
1946	* If we're doing an exclusive create, optimize away the lookup
1947	* but don't hash the dentry.
1948	*/
1949	if (nfs_is_exclusive_create(dir, flags) || flags & LOOKUP_RENAME_TARGET)
1950	return NULL;
1951
1952	res = ERR_PTR(error: -ENOMEM);
1953	fhandle = nfs_alloc_fhandle();
1954	fattr = nfs_alloc_fattr_with_label(server: NFS_SERVER(inode: dir));
1955	if (fhandle == NULL || fattr == NULL)
1956	goto out;
1957
1958	dir_verifier = nfs_save_change_attribute(dir);
1959	trace_nfs_lookup_enter(dir, dentry, flags);
1960	error = NFS_PROTO(inode: dir)->lookup(dir, dentry, fhandle, fattr);
1961	if (error == -ENOENT) {
1962	if (nfs_server_capable(inode: dir, NFS_CAP_CASE_INSENSITIVE))
1963	dir_verifier = inode_peek_iversion_raw(inode: dir);
1964	goto no_entry;
1965	}
1966	if (error < 0) {
1967	res = ERR_PTR(error);
1968	goto out;
1969	}
1970	inode = nfs_fhget(dentry->d_sb, fhandle, fattr);
1971	res = ERR_CAST(ptr: inode);
1972	if (IS_ERR(ptr: res))
1973	goto out;
1974
1975	/* Notify readdir to use READDIRPLUS */
1976	nfs_lookup_advise_force_readdirplus(dir, flags);
1977
1978	no_entry:
1979	res = d_splice_alias(inode, dentry);
1980	if (res != NULL) {
1981	if (IS_ERR(ptr: res))
1982	goto out;
1983	dentry = res;
1984	}
1985	nfs_set_verifier(dentry, dir_verifier);
1986	out:
1987	trace_nfs_lookup_exit(dir, dentry, flags, error: PTR_ERR_OR_ZERO(ptr: res));
1988	nfs_free_fattr(fattr);
1989	nfs_free_fhandle(fh: fhandle);
1990	return res;
1991	}
1992	EXPORT_SYMBOL_GPL(nfs_lookup);
1993
1994	void nfs_d_prune_case_insensitive_aliases(struct inode *inode)
1995	{
1996	/* Case insensitive server? Revalidate dentries */
1997	if (inode && nfs_server_capable(inode, NFS_CAP_CASE_INSENSITIVE))
1998	d_prune_aliases(inode);
1999	}
2000	EXPORT_SYMBOL_GPL(nfs_d_prune_case_insensitive_aliases);
2001
2002	#if IS_ENABLED(CONFIG_NFS_V4)
2003	static int nfs4_lookup_revalidate(struct dentry *, unsigned int);
2004
2005	const struct dentry_operations nfs4_dentry_operations = {
2006	.d_revalidate = nfs4_lookup_revalidate,
2007	.d_weak_revalidate = nfs_weak_revalidate,
2008	.d_delete = nfs_dentry_delete,
2009	.d_iput = nfs_dentry_iput,
2010	.d_automount = nfs_d_automount,
2011	.d_release = nfs_d_release,
2012	};
2013	EXPORT_SYMBOL_GPL(nfs4_dentry_operations);
2014
2015	static struct nfs_open_context *create_nfs_open_context(struct dentry *dentry, int open_flags, struct file *filp)
2016	{
2017	return alloc_nfs_open_context(dentry, f_mode: flags_to_mode(flags: open_flags), filp);
2018	}
2019
2020	static int do_open(struct inode *inode, struct file *filp)
2021	{
2022	nfs_fscache_open_file(inode, filp);
2023	return 0;
2024	}
2025
2026	static int nfs_finish_open(struct nfs_open_context *ctx,
2027	struct dentry *dentry,
2028	struct file *file, unsigned open_flags)
2029	{
2030	int err;
2031
2032	err = finish_open(file, dentry, open: do_open);
2033	if (err)
2034	goto out;
2035	if (S_ISREG(file_inode(file)->i_mode))
2036	nfs_file_set_open_context(filp: file, ctx);
2037	else
2038	err = -EOPENSTALE;
2039	out:
2040	return err;
2041	}
2042
2043	int nfs_atomic_open(struct inode *dir, struct dentry *dentry,
2044	struct file *file, unsigned open_flags,
2045	umode_t mode)
2046	{
2047	DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
2048	struct nfs_open_context *ctx;
2049	struct dentry *res;
2050	struct iattr attr = { .ia_valid = ATTR_OPEN };
2051	struct inode *inode;
2052	unsigned int lookup_flags = 0;
2053	unsigned long dir_verifier;
2054	bool switched = false;
2055	int created = 0;
2056	int err;
2057
2058	/* Expect a negative dentry */
2059	BUG_ON(d_inode(dentry));
2060
2061	dfprintk(VFS, "NFS: atomic_open(%s/%lu), %pd\n",
2062	dir->i_sb->s_id, dir->i_ino, dentry);
2063
2064	err = nfs_check_flags(open_flags);
2065	if (err)
2066	return err;
2067
2068	/* NFS only supports OPEN on regular files */
2069	if ((open_flags & O_DIRECTORY)) {
2070	if (!d_in_lookup(dentry)) {
2071	/*
2072	* Hashed negative dentry with O_DIRECTORY: dentry was
2073	* revalidated and is fine, no need to perform lookup
2074	* again
2075	*/
2076	return -ENOENT;
2077	}
2078	lookup_flags = LOOKUP_OPEN|LOOKUP_DIRECTORY;
2079	goto no_open;
2080	}
2081
2082	if (dentry->d_name.len > NFS_SERVER(inode: dir)->namelen)
2083	return -ENAMETOOLONG;
2084
2085	if (open_flags & O_CREAT) {
2086	struct nfs_server *server = NFS_SERVER(inode: dir);
2087
2088	if (!(server->attr_bitmask[2] & FATTR4_WORD2_MODE_UMASK))
2089	mode &= ~current_umask();
2090
2091	attr.ia_valid |= ATTR_MODE;
2092	attr.ia_mode = mode;
2093	}
2094	if (open_flags & O_TRUNC) {
2095	attr.ia_valid |= ATTR_SIZE;
2096	attr.ia_size = 0;
2097	}
2098
2099	if (!(open_flags & O_CREAT) && !d_in_lookup(dentry)) {
2100	d_drop(dentry);
2101	switched = true;
2102	dentry = d_alloc_parallel(dentry->d_parent,
2103	&dentry->d_name, &wq);
2104	if (IS_ERR(ptr: dentry))
2105	return PTR_ERR(ptr: dentry);
2106	if (unlikely(!d_in_lookup(dentry)))
2107	return finish_no_open(file, dentry);
2108	}
2109
2110	ctx = create_nfs_open_context(dentry, open_flags, filp: file);
2111	err = PTR_ERR(ptr: ctx);
2112	if (IS_ERR(ptr: ctx))
2113	goto out;
2114
2115	trace_nfs_atomic_open_enter(dir, ctx, flags: open_flags);
2116	inode = NFS_PROTO(inode: dir)->open_context(dir, ctx, open_flags, &attr, &created);
2117	if (created)
2118	file->f_mode |= FMODE_CREATED;
2119	if (IS_ERR(ptr: inode)) {
2120	err = PTR_ERR(ptr: inode);
2121	trace_nfs_atomic_open_exit(dir, ctx, flags: open_flags, error: err);
2122	put_nfs_open_context(ctx);
2123	d_drop(dentry);
2124	switch (err) {
2125	case -ENOENT:
2126	d_splice_alias(NULL, dentry);
2127	if (nfs_server_capable(inode: dir, NFS_CAP_CASE_INSENSITIVE))
2128	dir_verifier = inode_peek_iversion_raw(inode: dir);
2129	else
2130	dir_verifier = nfs_save_change_attribute(dir);
2131	nfs_set_verifier(dentry, dir_verifier);
2132	break;
2133	case -EISDIR:
2134	case -ENOTDIR:
2135	goto no_open;
2136	case -ELOOP:
2137	if (!(open_flags & O_NOFOLLOW))
2138	goto no_open;
2139	break;
2140	/* case -EINVAL: */
2141	default:
2142	break;
2143	}
2144	goto out;
2145	}
2146	file->f_mode |= FMODE_CAN_ODIRECT;
2147
2148	err = nfs_finish_open(ctx, dentry: ctx->dentry, file, open_flags);
2149	trace_nfs_atomic_open_exit(dir, ctx, flags: open_flags, error: err);
2150	put_nfs_open_context(ctx);
2151	out:
2152	if (unlikely(switched)) {
2153	d_lookup_done(dentry);
2154	dput(dentry);
2155	}
2156	return err;
2157
2158	no_open:
2159	res = nfs_lookup(dir, dentry, lookup_flags);
2160	if (!res) {
2161	inode = d_inode(dentry);
2162	if ((lookup_flags & LOOKUP_DIRECTORY) && inode &&
2163	!(S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)))
2164	res = ERR_PTR(error: -ENOTDIR);
2165	else if (inode && S_ISREG(inode->i_mode))
2166	res = ERR_PTR(error: -EOPENSTALE);
2167	} else if (!IS_ERR(ptr: res)) {
2168	inode = d_inode(dentry: res);
2169	if ((lookup_flags & LOOKUP_DIRECTORY) && inode &&
2170	!(S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))) {
2171	dput(res);
2172	res = ERR_PTR(error: -ENOTDIR);
2173	} else if (inode && S_ISREG(inode->i_mode)) {
2174	dput(res);
2175	res = ERR_PTR(error: -EOPENSTALE);
2176	}
2177	}
2178	if (switched) {
2179	d_lookup_done(dentry);
2180	if (!res)
2181	res = dentry;
2182	else
2183	dput(dentry);
2184	}
2185	if (IS_ERR(ptr: res))
2186	return PTR_ERR(ptr: res);
2187	return finish_no_open(file, dentry: res);
2188	}
2189	EXPORT_SYMBOL_GPL(nfs_atomic_open);
2190
2191	static int
2192	nfs4_do_lookup_revalidate(struct inode *dir, struct dentry *dentry,
2193	unsigned int flags)
2194	{
2195	struct inode *inode;
2196
2197	if (!(flags & LOOKUP_OPEN) || (flags & LOOKUP_DIRECTORY))
2198	goto full_reval;
2199	if (d_mountpoint(dentry))
2200	goto full_reval;
2201
2202	inode = d_inode(dentry);
2203
2204	/* We can't create new files in nfs_open_revalidate(), so we
2205	* optimize away revalidation of negative dentries.
2206	*/
2207	if (inode == NULL)
2208	goto full_reval;
2209
2210	if (nfs_verifier_is_delegated(dentry))
2211	return nfs_lookup_revalidate_delegated(dir, dentry, inode);
2212
2213	/* NFS only supports OPEN on regular files */
2214	if (!S_ISREG(inode->i_mode))
2215	goto full_reval;
2216
2217	/* We cannot do exclusive creation on a positive dentry */
2218	if (flags & (LOOKUP_EXCL | LOOKUP_REVAL))
2219	goto reval_dentry;
2220
2221	/* Check if the directory changed */
2222	if (!nfs_check_verifier(dir, dentry, rcu_walk: flags & LOOKUP_RCU))
2223	goto reval_dentry;
2224
2225	/* Let f_op->open() actually open (and revalidate) the file */
2226	return 1;
2227	reval_dentry:
2228	if (flags & LOOKUP_RCU)
2229	return -ECHILD;
2230	return nfs_lookup_revalidate_dentry(dir, dentry, inode, flags);
2231
2232	full_reval:
2233	return nfs_do_lookup_revalidate(dir, dentry, flags);
2234	}
2235
2236	static int nfs4_lookup_revalidate(struct dentry *dentry, unsigned int flags)
2237	{
2238	return __nfs_lookup_revalidate(dentry, flags,
2239	reval: nfs4_do_lookup_revalidate);
2240	}
2241
2242	#endif /* CONFIG_NFSV4 */
2243
2244	struct dentry *
2245	nfs_add_or_obtain(struct dentry *dentry, struct nfs_fh *fhandle,
2246	struct nfs_fattr *fattr)
2247	{
2248	struct dentry *parent = dget_parent(dentry);
2249	struct inode *dir = d_inode(dentry: parent);
2250	struct inode *inode;
2251	struct dentry *d;
2252	int error;
2253
2254	d_drop(dentry);
2255
2256	if (fhandle->size == 0) {
2257	error = NFS_PROTO(inode: dir)->lookup(dir, dentry, fhandle, fattr);
2258	if (error)
2259	goto out_error;
2260	}
2261	nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
2262	if (!(fattr->valid & NFS_ATTR_FATTR)) {
2263	struct nfs_server *server = NFS_SB(s: dentry->d_sb);
2264	error = server->nfs_client->rpc_ops->getattr(server, fhandle,
2265	fattr, NULL);
2266	if (error < 0)
2267	goto out_error;
2268	}
2269	inode = nfs_fhget(dentry->d_sb, fhandle, fattr);
2270	d = d_splice_alias(inode, dentry);
2271	out:
2272	dput(parent);
2273	return d;
2274	out_error:
2275	d = ERR_PTR(error);
2276	goto out;
2277	}
2278	EXPORT_SYMBOL_GPL(nfs_add_or_obtain);
2279
2280	/*
2281	* Code common to create, mkdir, and mknod.
2282	*/
2283	int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle,
2284	struct nfs_fattr *fattr)
2285	{
2286	struct dentry *d;
2287
2288	d = nfs_add_or_obtain(dentry, fhandle, fattr);
2289	if (IS_ERR(ptr: d))
2290	return PTR_ERR(ptr: d);
2291
2292	/* Callers don't care */
2293	dput(d);
2294	return 0;
2295	}
2296	EXPORT_SYMBOL_GPL(nfs_instantiate);
2297
2298	/*
2299	* Following a failed create operation, we drop the dentry rather
2300	* than retain a negative dentry. This avoids a problem in the event
2301	* that the operation succeeded on the server, but an error in the
2302	* reply path made it appear to have failed.
2303	*/
2304	int nfs_create(struct mnt_idmap *idmap, struct inode *dir,
2305	struct dentry *dentry, umode_t mode, bool excl)
2306	{
2307	struct iattr attr;
2308	int open_flags = excl ? O_CREAT | O_EXCL : O_CREAT;
2309	int error;
2310
2311	dfprintk(VFS, "NFS: create(%s/%lu), %pd\n",
2312	dir->i_sb->s_id, dir->i_ino, dentry);
2313
2314	attr.ia_mode = mode;
2315	attr.ia_valid = ATTR_MODE;
2316
2317	trace_nfs_create_enter(dir, dentry, flags: open_flags);
2318	error = NFS_PROTO(inode: dir)->create(dir, dentry, &attr, open_flags);
2319	trace_nfs_create_exit(dir, dentry, flags: open_flags, error);
2320	if (error != 0)
2321	goto out_err;
2322	return 0;
2323	out_err:
2324	d_drop(dentry);
2325	return error;
2326	}
2327	EXPORT_SYMBOL_GPL(nfs_create);
2328
2329	/*
2330	* See comments for nfs_proc_create regarding failed operations.
2331	*/
2332	int
2333	nfs_mknod(struct mnt_idmap *idmap, struct inode *dir,
2334	struct dentry *dentry, umode_t mode, dev_t rdev)
2335	{
2336	struct iattr attr;
2337	int status;
2338
2339	dfprintk(VFS, "NFS: mknod(%s/%lu), %pd\n",
2340	dir->i_sb->s_id, dir->i_ino, dentry);
2341
2342	attr.ia_mode = mode;
2343	attr.ia_valid = ATTR_MODE;
2344
2345	trace_nfs_mknod_enter(dir, dentry);
2346	status = NFS_PROTO(inode: dir)->mknod(dir, dentry, &attr, rdev);
2347	trace_nfs_mknod_exit(dir, dentry, error: status);
2348	if (status != 0)
2349	goto out_err;
2350	return 0;
2351	out_err:
2352	d_drop(dentry);
2353	return status;
2354	}
2355	EXPORT_SYMBOL_GPL(nfs_mknod);
2356
2357	/*
2358	* See comments for nfs_proc_create regarding failed operations.
2359	*/
2360	int nfs_mkdir(struct mnt_idmap *idmap, struct inode *dir,
2361	struct dentry *dentry, umode_t mode)
2362	{
2363	struct iattr attr;
2364	int error;
2365
2366	dfprintk(VFS, "NFS: mkdir(%s/%lu), %pd\n",
2367	dir->i_sb->s_id, dir->i_ino, dentry);
2368
2369	attr.ia_valid = ATTR_MODE;
2370	attr.ia_mode = mode | S_IFDIR;
2371
2372	trace_nfs_mkdir_enter(dir, dentry);
2373	error = NFS_PROTO(inode: dir)->mkdir(dir, dentry, &attr);
2374	trace_nfs_mkdir_exit(dir, dentry, error);
2375	if (error != 0)
2376	goto out_err;
2377	return 0;
2378	out_err:
2379	d_drop(dentry);
2380	return error;
2381	}
2382	EXPORT_SYMBOL_GPL(nfs_mkdir);
2383
2384	static void nfs_dentry_handle_enoent(struct dentry *dentry)
2385	{
2386	if (simple_positive(dentry))
2387	d_delete(dentry);
2388	}
2389
2390	static void nfs_dentry_remove_handle_error(struct inode *dir,
2391	struct dentry *dentry, int error)
2392	{
2393	switch (error) {
2394	case -ENOENT:
2395	if (d_really_is_positive(dentry))
2396	d_delete(dentry);
2397	nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
2398	break;
2399	case 0:
2400	nfs_d_prune_case_insensitive_aliases(d_inode(dentry));
2401	nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
2402	}
2403	}
2404
2405	int nfs_rmdir(struct inode *dir, struct dentry *dentry)
2406	{
2407	int error;
2408
2409	dfprintk(VFS, "NFS: rmdir(%s/%lu), %pd\n",
2410	dir->i_sb->s_id, dir->i_ino, dentry);
2411
2412	trace_nfs_rmdir_enter(dir, dentry);
2413	if (d_really_is_positive(dentry)) {
2414	down_write(sem: &NFS_I(inode: d_inode(dentry))->rmdir_sem);
2415	error = NFS_PROTO(inode: dir)->rmdir(dir, &dentry->d_name);
2416	/* Ensure the VFS deletes this inode */
2417	switch (error) {
2418	case 0:
2419	clear_nlink(inode: d_inode(dentry));
2420	break;
2421	case -ENOENT:
2422	nfs_dentry_handle_enoent(dentry);
2423	}
2424	up_write(sem: &NFS_I(inode: d_inode(dentry))->rmdir_sem);
2425	} else
2426	error = NFS_PROTO(inode: dir)->rmdir(dir, &dentry->d_name);
2427	nfs_dentry_remove_handle_error(dir, dentry, error);
2428	trace_nfs_rmdir_exit(dir, dentry, error);
2429
2430	return error;
2431	}
2432	EXPORT_SYMBOL_GPL(nfs_rmdir);
2433
2434	/*
2435	* Remove a file after making sure there are no pending writes,
2436	* and after checking that the file has only one user.
2437	*
2438	* We invalidate the attribute cache and free the inode prior to the operation
2439	* to avoid possible races if the server reuses the inode.
2440	*/
2441	static int nfs_safe_remove(struct dentry *dentry)
2442	{
2443	struct inode *dir = d_inode(dentry: dentry->d_parent);
2444	struct inode *inode = d_inode(dentry);
2445	int error = -EBUSY;
2446
2447	dfprintk(VFS, "NFS: safe_remove(%pd2)\n", dentry);
2448
2449	/* If the dentry was sillyrenamed, we simply call d_delete() */
2450	if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
2451	error = 0;
2452	goto out;
2453	}
2454
2455	trace_nfs_remove_enter(dir, dentry);
2456	if (inode != NULL) {
2457	error = NFS_PROTO(inode: dir)->remove(dir, dentry);
2458	if (error == 0)
2459	nfs_drop_nlink(inode);
2460	} else
2461	error = NFS_PROTO(inode: dir)->remove(dir, dentry);
2462	if (error == -ENOENT)
2463	nfs_dentry_handle_enoent(dentry);
2464	trace_nfs_remove_exit(dir, dentry, error);
2465	out:
2466	return error;
2467	}
2468
2469	/* We do silly rename. In case sillyrename() returns -EBUSY, the inode
2470	* belongs to an active ".nfs..." file and we return -EBUSY.
2471	*
2472	* If sillyrename() returns 0, we do nothing, otherwise we unlink.
2473	*/
2474	int nfs_unlink(struct inode *dir, struct dentry *dentry)
2475	{
2476	int error;
2477
2478	dfprintk(VFS, "NFS: unlink(%s/%lu, %pd)\n", dir->i_sb->s_id,
2479	dir->i_ino, dentry);
2480
2481	trace_nfs_unlink_enter(dir, dentry);
2482	spin_lock(lock: &dentry->d_lock);
2483	if (d_count(dentry) > 1 && !test_bit(NFS_INO_PRESERVE_UNLINKED,
2484	&NFS_I(d_inode(dentry))->flags)) {
2485	spin_unlock(lock: &dentry->d_lock);
2486	/* Start asynchronous writeout of the inode */
2487	write_inode_now(d_inode(dentry), sync: 0);
2488	error = nfs_sillyrename(dir, dentry);
2489	goto out;
2490	}
2491	/* We must prevent any concurrent open until the unlink
2492	* completes. ->d_revalidate will wait for ->d_fsdata
2493	* to clear. We set it here to ensure no lookup succeeds until
2494	* the unlink is complete on the server.
2495	*/
2496	error = -ETXTBSY;
2497	if (WARN_ON(dentry->d_flags & DCACHE_NFSFS_RENAMED) ||
2498	WARN_ON(dentry->d_fsdata == NFS_FSDATA_BLOCKED)) {
2499	spin_unlock(lock: &dentry->d_lock);
2500	goto out;
2501	}
2502	/* old devname */
2503	kfree(objp: dentry->d_fsdata);
2504	dentry->d_fsdata = NFS_FSDATA_BLOCKED;
2505
2506	spin_unlock(lock: &dentry->d_lock);
2507	error = nfs_safe_remove(dentry);
2508	nfs_dentry_remove_handle_error(dir, dentry, error);
2509	dentry->d_fsdata = NULL;
2510	wake_up_var(var: &dentry->d_fsdata);
2511	out:
2512	trace_nfs_unlink_exit(dir, dentry, error);
2513	return error;
2514	}
2515	EXPORT_SYMBOL_GPL(nfs_unlink);
2516
2517	/*
2518	* To create a symbolic link, most file systems instantiate a new inode,
2519	* add a page to it containing the path, then write it out to the disk
2520	* using prepare_write/commit_write.
2521	*
2522	* Unfortunately the NFS client can't create the in-core inode first
2523	* because it needs a file handle to create an in-core inode (see
2524	* fs/nfs/inode.c:nfs_fhget). We only have a file handle *after* the
2525	* symlink request has completed on the server.
2526	*
2527	* So instead we allocate a raw page, copy the symname into it, then do
2528	* the SYMLINK request with the page as the buffer. If it succeeds, we
2529	* now have a new file handle and can instantiate an in-core NFS inode
2530	* and move the raw page into its mapping.
2531	*/
2532	int nfs_symlink(struct mnt_idmap *idmap, struct inode *dir,
2533	struct dentry *dentry, const char *symname)
2534	{
2535	struct folio *folio;
2536	char *kaddr;
2537	struct iattr attr;
2538	unsigned int pathlen = strlen(symname);
2539	int error;
2540
2541	dfprintk(VFS, "NFS: symlink(%s/%lu, %pd, %s)\n", dir->i_sb->s_id,
2542	dir->i_ino, dentry, symname);
2543
2544	if (pathlen > PAGE_SIZE)
2545	return -ENAMETOOLONG;
2546
2547	attr.ia_mode = S_IFLNK | S_IRWXUGO;
2548	attr.ia_valid = ATTR_MODE;
2549
2550	folio = folio_alloc(GFP_USER, order: 0);
2551	if (!folio)
2552	return -ENOMEM;
2553
2554	kaddr = folio_address(folio);
2555	memcpy(kaddr, symname, pathlen);
2556	if (pathlen < PAGE_SIZE)
2557	memset(kaddr + pathlen, 0, PAGE_SIZE - pathlen);
2558
2559	trace_nfs_symlink_enter(dir, dentry);
2560	error = NFS_PROTO(inode: dir)->symlink(dir, dentry, folio, pathlen, &attr);
2561	trace_nfs_symlink_exit(dir, dentry, error);
2562	if (error != 0) {
2563	dfprintk(VFS, "NFS: symlink(%s/%lu, %pd, %s) error %d\n",
2564	dir->i_sb->s_id, dir->i_ino,
2565	dentry, symname, error);
2566	d_drop(dentry);
2567	folio_put(folio);
2568	return error;
2569	}
2570
2571	nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
2572
2573	/*
2574	* No big deal if we can't add this page to the page cache here.
2575	* READLINK will get the missing page from the server if needed.
2576	*/
2577	if (filemap_add_folio(mapping: d_inode(dentry)->i_mapping, folio, index: 0,
2578	GFP_KERNEL) == 0) {
2579	folio_mark_uptodate(folio);
2580	folio_unlock(folio);
2581	}
2582
2583	folio_put(folio);
2584	return 0;
2585	}
2586	EXPORT_SYMBOL_GPL(nfs_symlink);
2587
2588	int
2589	nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
2590	{
2591	struct inode *inode = d_inode(dentry: old_dentry);
2592	int error;
2593
2594	dfprintk(VFS, "NFS: link(%pd2 -> %pd2)\n",
2595	old_dentry, dentry);
2596
2597	trace_nfs_link_enter(inode, dir, dentry);
2598	d_drop(dentry);
2599	if (S_ISREG(inode->i_mode))
2600	nfs_sync_inode(inode);
2601	error = NFS_PROTO(inode: dir)->link(inode, dir, &dentry->d_name);
2602	if (error == 0) {
2603	nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
2604	ihold(inode);
2605	d_add(dentry, inode);
2606	}
2607	trace_nfs_link_exit(inode, dir, dentry, error);
2608	return error;
2609	}
2610	EXPORT_SYMBOL_GPL(nfs_link);
2611
2612	static void
2613	nfs_unblock_rename(struct rpc_task *task, struct nfs_renamedata *data)
2614	{
2615	struct dentry *new_dentry = data->new_dentry;
2616
2617	new_dentry->d_fsdata = NULL;
2618	wake_up_var(var: &new_dentry->d_fsdata);
2619	}
2620
2621	/*
2622	* RENAME
2623	* FIXME: Some nfsds, like the Linux user space nfsd, may generate a
2624	* different file handle for the same inode after a rename (e.g. when
2625	* moving to a different directory). A fail-safe method to do so would
2626	* be to look up old_dir/old_name, create a link to new_dir/new_name and
2627	* rename the old file using the sillyrename stuff. This way, the original
2628	* file in old_dir will go away when the last process iput()s the inode.
2629	*
2630	* FIXED.
2631	*
2632	* It actually works quite well. One needs to have the possibility for
2633	* at least one ".nfs..." file in each directory the file ever gets
2634	* moved or linked to which happens automagically with the new
2635	* implementation that only depends on the dcache stuff instead of
2636	* using the inode layer
2637	*
2638	* Unfortunately, things are a little more complicated than indicated
2639	* above. For a cross-directory move, we want to make sure we can get
2640	* rid of the old inode after the operation. This means there must be
2641	* no pending writes (if it's a file), and the use count must be 1.
2642	* If these conditions are met, we can drop the dentries before doing
2643	* the rename.
2644	*/
2645	int nfs_rename(struct mnt_idmap *idmap, struct inode *old_dir,
2646	struct dentry *old_dentry, struct inode *new_dir,
2647	struct dentry *new_dentry, unsigned int flags)
2648	{
2649	struct inode *old_inode = d_inode(dentry: old_dentry);
2650	struct inode *new_inode = d_inode(dentry: new_dentry);
2651	struct dentry *dentry = NULL;
2652	struct rpc_task *task;
2653	bool must_unblock = false;
2654	int error = -EBUSY;
2655
2656	if (flags)
2657	return -EINVAL;
2658
2659	dfprintk(VFS, "NFS: rename(%pd2 -> %pd2, ct=%d)\n",
2660	old_dentry, new_dentry,
2661	d_count(new_dentry));
2662
2663	trace_nfs_rename_enter(old_dir, old_dentry, new_dir, new_dentry);
2664	/*
2665	* For non-directories, check whether the target is busy and if so,
2666	* make a copy of the dentry and then do a silly-rename. If the
2667	* silly-rename succeeds, the copied dentry is hashed and becomes
2668	* the new target.
2669	*/
2670	if (new_inode && !S_ISDIR(new_inode->i_mode)) {
2671	/* We must prevent any concurrent open until the unlink
2672	* completes. ->d_revalidate will wait for ->d_fsdata
2673	* to clear. We set it here to ensure no lookup succeeds until
2674	* the unlink is complete on the server.
2675	*/
2676	error = -ETXTBSY;
2677	if (WARN_ON(new_dentry->d_flags & DCACHE_NFSFS_RENAMED) ||
2678	WARN_ON(new_dentry->d_fsdata == NFS_FSDATA_BLOCKED))
2679	goto out;
2680	if (new_dentry->d_fsdata) {
2681	/* old devname */
2682	kfree(objp: new_dentry->d_fsdata);
2683	new_dentry->d_fsdata = NULL;
2684	}
2685
2686	spin_lock(lock: &new_dentry->d_lock);
2687	if (d_count(dentry: new_dentry) > 2) {
2688	int err;
2689
2690	spin_unlock(lock: &new_dentry->d_lock);
2691
2692	/* copy the target dentry's name */
2693	dentry = d_alloc(new_dentry->d_parent,
2694	&new_dentry->d_name);
2695	if (!dentry)
2696	goto out;
2697
2698	/* silly-rename the existing target ... */
2699	err = nfs_sillyrename(dir: new_dir, dentry: new_dentry);
2700	if (err)
2701	goto out;
2702
2703	new_dentry = dentry;
2704	new_inode = NULL;
2705	} else {
2706	new_dentry->d_fsdata = NFS_FSDATA_BLOCKED;
2707	must_unblock = true;
2708	spin_unlock(lock: &new_dentry->d_lock);
2709	}
2710
2711	}
2712
2713	if (S_ISREG(old_inode->i_mode))
2714	nfs_sync_inode(inode: old_inode);
2715	task = nfs_async_rename(old_dir, new_dir, old_dentry, new_dentry,
2716	complete: must_unblock ? nfs_unblock_rename : NULL);
2717	if (IS_ERR(ptr: task)) {
2718	error = PTR_ERR(ptr: task);
2719	goto out;
2720	}
2721
2722	error = rpc_wait_for_completion_task(task);
2723	if (error != 0) {
2724	((struct nfs_renamedata *)task->tk_calldata)->cancelled = 1;
2725	/* Paired with the atomic_dec_and_test() barrier in rpc_do_put_task() */
2726	smp_wmb();
2727	} else
2728	error = task->tk_status;
2729	rpc_put_task(task);
2730	/* Ensure the inode attributes are revalidated */
2731	if (error == 0) {
2732	spin_lock(lock: &old_inode->i_lock);
2733	NFS_I(inode: old_inode)->attr_gencount = nfs_inc_attr_generation_counter();
2734	nfs_set_cache_invalid(inode: old_inode, NFS_INO_INVALID_CHANGE |
2735	NFS_INO_INVALID_CTIME |
2736	NFS_INO_REVAL_FORCED);
2737	spin_unlock(lock: &old_inode->i_lock);
2738	}
2739	out:
2740	trace_nfs_rename_exit(old_dir, old_dentry,
2741	new_dir, new_dentry, error);
2742	if (!error) {
2743	if (new_inode != NULL)
2744	nfs_drop_nlink(inode: new_inode);
2745	/*
2746	* The d_move() should be here instead of in an async RPC completion
2747	* handler because we need the proper locks to move the dentry. If
2748	* we're interrupted by a signal, the async RPC completion handler
2749	* should mark the directories for revalidation.
2750	*/
2751	d_move(old_dentry, new_dentry);
2752	nfs_set_verifier(old_dentry,
2753	nfs_save_change_attribute(dir: new_dir));
2754	} else if (error == -ENOENT)
2755	nfs_dentry_handle_enoent(dentry: old_dentry);
2756
2757	/* new dentry created? */
2758	if (dentry)
2759	dput(dentry);
2760	return error;
2761	}
2762	EXPORT_SYMBOL_GPL(nfs_rename);
2763
2764	static DEFINE_SPINLOCK(nfs_access_lru_lock);
2765	static LIST_HEAD(nfs_access_lru_list);
2766	static atomic_long_t nfs_access_nr_entries;
2767
2768	static unsigned long nfs_access_max_cachesize = 4*1024*1024;
2769	module_param(nfs_access_max_cachesize, ulong, 0644);
2770	MODULE_PARM_DESC(nfs_access_max_cachesize, "NFS access maximum total cache length");
2771
2772	static void nfs_access_free_entry(struct nfs_access_entry *entry)
2773	{
2774	put_group_info(entry->group_info);
2775	kfree_rcu(entry, rcu_head);
2776	smp_mb__before_atomic();
2777	atomic_long_dec(v: &nfs_access_nr_entries);
2778	smp_mb__after_atomic();
2779	}
2780
2781	static void nfs_access_free_list(struct list_head *head)
2782	{
2783	struct nfs_access_entry *cache;
2784
2785	while (!list_empty(head)) {
2786	cache = list_entry(head->next, struct nfs_access_entry, lru);
2787	list_del(entry: &cache->lru);
2788	nfs_access_free_entry(entry: cache);
2789	}
2790	}
2791
2792	static unsigned long
2793	nfs_do_access_cache_scan(unsigned int nr_to_scan)
2794	{
2795	LIST_HEAD(head);
2796	struct nfs_inode *nfsi, *next;
2797	struct nfs_access_entry *cache;
2798	long freed = 0;
2799
2800	spin_lock(lock: &nfs_access_lru_lock);
2801	list_for_each_entry_safe(nfsi, next, &nfs_access_lru_list, access_cache_inode_lru) {
2802	struct inode *inode;
2803
2804	if (nr_to_scan-- == 0)
2805	break;
2806	inode = &nfsi->vfs_inode;
2807	spin_lock(lock: &inode->i_lock);
2808	if (list_empty(head: &nfsi->access_cache_entry_lru))
2809	goto remove_lru_entry;
2810	cache = list_entry(nfsi->access_cache_entry_lru.next,
2811	struct nfs_access_entry, lru);
2812	list_move(list: &cache->lru, head: &head);
2813	rb_erase(&cache->rb_node, &nfsi->access_cache);
2814	freed++;
2815	if (!list_empty(head: &nfsi->access_cache_entry_lru))
2816	list_move_tail(list: &nfsi->access_cache_inode_lru,
2817	head: &nfs_access_lru_list);
2818	else {
2819	remove_lru_entry:
2820	list_del_init(entry: &nfsi->access_cache_inode_lru);
2821	smp_mb__before_atomic();
2822	clear_bit(NFS_INO_ACL_LRU_SET, addr: &nfsi->flags);
2823	smp_mb__after_atomic();
2824	}
2825	spin_unlock(lock: &inode->i_lock);
2826	}
2827	spin_unlock(lock: &nfs_access_lru_lock);
2828	nfs_access_free_list(head: &head);
2829	return freed;
2830	}
2831
2832	unsigned long
2833	nfs_access_cache_scan(struct shrinker *shrink, struct shrink_control *sc)
2834	{
2835	int nr_to_scan = sc->nr_to_scan;
2836	gfp_t gfp_mask = sc->gfp_mask;
2837
2838	if ((gfp_mask & GFP_KERNEL) != GFP_KERNEL)
2839	return SHRINK_STOP;
2840	return nfs_do_access_cache_scan(nr_to_scan);
2841	}
2842
2843
2844	unsigned long
2845	nfs_access_cache_count(struct shrinker *shrink, struct shrink_control *sc)
2846	{
2847	return vfs_pressure_ratio(val: atomic_long_read(v: &nfs_access_nr_entries));
2848	}
2849
2850	static void
2851	nfs_access_cache_enforce_limit(void)
2852	{
2853	long nr_entries = atomic_long_read(v: &nfs_access_nr_entries);
2854	unsigned long diff;
2855	unsigned int nr_to_scan;
2856
2857	if (nr_entries < 0 || nr_entries <= nfs_access_max_cachesize)
2858	return;
2859	nr_to_scan = 100;
2860	diff = nr_entries - nfs_access_max_cachesize;
2861	if (diff < nr_to_scan)
2862	nr_to_scan = diff;
2863	nfs_do_access_cache_scan(nr_to_scan);
2864	}
2865
2866	static void __nfs_access_zap_cache(struct nfs_inode *nfsi, struct list_head *head)
2867	{
2868	struct rb_root *root_node = &nfsi->access_cache;
2869	struct rb_node *n;
2870	struct nfs_access_entry *entry;
2871
2872	/* Unhook entries from the cache */
2873	while ((n = rb_first(root_node)) != NULL) {
2874	entry = rb_entry(n, struct nfs_access_entry, rb_node);
2875	rb_erase(n, root_node);
2876	list_move(list: &entry->lru, head);
2877	}
2878	nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS;
2879	}
2880
2881	void nfs_access_zap_cache(struct inode *inode)
2882	{
2883	LIST_HEAD(head);
2884
2885	if (test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags) == 0)
2886	return;
2887	/* Remove from global LRU init */
2888	spin_lock(lock: &nfs_access_lru_lock);
2889	if (test_and_clear_bit(NFS_INO_ACL_LRU_SET, addr: &NFS_I(inode)->flags))
2890	list_del_init(entry: &NFS_I(inode)->access_cache_inode_lru);
2891
2892	spin_lock(lock: &inode->i_lock);
2893	__nfs_access_zap_cache(nfsi: NFS_I(inode), head: &head);
2894	spin_unlock(lock: &inode->i_lock);
2895	spin_unlock(lock: &nfs_access_lru_lock);
2896	nfs_access_free_list(head: &head);
2897	}
2898	EXPORT_SYMBOL_GPL(nfs_access_zap_cache);
2899
2900	static int access_cmp(const struct cred *a, const struct nfs_access_entry *b)
2901	{
2902	struct group_info *ga, *gb;
2903	int g;
2904
2905	if (uid_lt(left: a->fsuid, right: b->fsuid))
2906	return -1;
2907	if (uid_gt(left: a->fsuid, right: b->fsuid))
2908	return 1;
2909
2910	if (gid_lt(left: a->fsgid, right: b->fsgid))
2911	return -1;
2912	if (gid_gt(left: a->fsgid, right: b->fsgid))
2913	return 1;
2914
2915	ga = a->group_info;
2916	gb = b->group_info;
2917	if (ga == gb)
2918	return 0;
2919	if (ga == NULL)
2920	return -1;
2921	if (gb == NULL)
2922	return 1;
2923	if (ga->ngroups < gb->ngroups)
2924	return -1;
2925	if (ga->ngroups > gb->ngroups)
2926	return 1;
2927
2928	for (g = 0; g < ga->ngroups; g++) {
2929	if (gid_lt(left: ga->gid[g], right: gb->gid[g]))
2930	return -1;
2931	if (gid_gt(left: ga->gid[g], right: gb->gid[g]))
2932	return 1;
2933	}
2934	return 0;
2935	}
2936
2937	static struct nfs_access_entry *nfs_access_search_rbtree(struct inode *inode, const struct cred *cred)
2938	{
2939	struct rb_node *n = NFS_I(inode)->access_cache.rb_node;
2940
2941	while (n != NULL) {
2942	struct nfs_access_entry *entry =
2943	rb_entry(n, struct nfs_access_entry, rb_node);
2944	int cmp = access_cmp(a: cred, b: entry);
2945
2946	if (cmp < 0)
2947	n = n->rb_left;
2948	else if (cmp > 0)
2949	n = n->rb_right;
2950	else
2951	return entry;
2952	}
2953	return NULL;
2954	}
2955
2956	static u64 nfs_access_login_time(const struct task_struct *task,
2957	const struct cred *cred)
2958	{
2959	const struct task_struct *parent;
2960	const struct cred *pcred;
2961	u64 ret;
2962
2963	rcu_read_lock();
2964	for (;;) {
2965	parent = rcu_dereference(task->real_parent);
2966	pcred = rcu_dereference(parent->cred);
2967	if (parent == task || cred_fscmp(pcred, cred) != 0)
2968	break;
2969	task = parent;
2970	}
2971	ret = task->start_time;
2972	rcu_read_unlock();
2973	return ret;
2974	}
2975
2976	static int nfs_access_get_cached_locked(struct inode *inode, const struct cred *cred, u32 *mask, bool may_block)
2977	{
2978	struct nfs_inode *nfsi = NFS_I(inode);
2979	u64 login_time = nfs_access_login_time(current, cred);
2980	struct nfs_access_entry *cache;
2981	bool retry = true;
2982	int err;
2983
2984	spin_lock(lock: &inode->i_lock);
2985	for(;;) {
2986	if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS)
2987	goto out_zap;
2988	cache = nfs_access_search_rbtree(inode, cred);
2989	err = -ENOENT;
2990	if (cache == NULL)
2991	goto out;
2992	/* Found an entry, is our attribute cache valid? */
2993	if (!nfs_check_cache_invalid(inode, NFS_INO_INVALID_ACCESS))
2994	break;
2995	if (!retry)
2996	break;
2997	err = -ECHILD;
2998	if (!may_block)
2999	goto out;
3000	spin_unlock(lock: &inode->i_lock);
3001	err = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
3002	if (err)
3003	return err;
3004	spin_lock(lock: &inode->i_lock);
3005	retry = false;
3006	}
3007	err = -ENOENT;
3008	if ((s64)(login_time - cache->timestamp) > 0)
3009	goto out;
3010	*mask = cache->mask;
3011	list_move_tail(list: &cache->lru, head: &nfsi->access_cache_entry_lru);
3012	err = 0;
3013	out:
3014	spin_unlock(lock: &inode->i_lock);
3015	return err;
3016	out_zap:
3017	spin_unlock(lock: &inode->i_lock);
3018	nfs_access_zap_cache(inode);
3019	return -ENOENT;
3020	}
3021
3022	static int nfs_access_get_cached_rcu(struct inode *inode, const struct cred *cred, u32 *mask)
3023	{
3024	/* Only check the most recently returned cache entry,
3025	* but do it without locking.
3026	*/
3027	struct nfs_inode *nfsi = NFS_I(inode);
3028	u64 login_time = nfs_access_login_time(current, cred);
3029	struct nfs_access_entry *cache;
3030	int err = -ECHILD;
3031	struct list_head *lh;
3032
3033	rcu_read_lock();
3034	if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS)
3035	goto out;
3036	lh = rcu_dereference(list_tail_rcu(&nfsi->access_cache_entry_lru));
3037	cache = list_entry(lh, struct nfs_access_entry, lru);
3038	if (lh == &nfsi->access_cache_entry_lru ||
3039	access_cmp(a: cred, b: cache) != 0)
3040	cache = NULL;
3041	if (cache == NULL)
3042	goto out;
3043	if ((s64)(login_time - cache->timestamp) > 0)
3044	goto out;
3045	if (nfs_check_cache_invalid(inode, NFS_INO_INVALID_ACCESS))
3046	goto out;
3047	*mask = cache->mask;
3048	err = 0;
3049	out:
3050	rcu_read_unlock();
3051	return err;
3052	}
3053
3054	int nfs_access_get_cached(struct inode *inode, const struct cred *cred,
3055	u32 *mask, bool may_block)
3056	{
3057	int status;
3058
3059	status = nfs_access_get_cached_rcu(inode, cred, mask);
3060	if (status != 0)
3061	status = nfs_access_get_cached_locked(inode, cred, mask,
3062	may_block);
3063
3064	return status;
3065	}
3066	EXPORT_SYMBOL_GPL(nfs_access_get_cached);
3067
3068	static void nfs_access_add_rbtree(struct inode *inode,
3069	struct nfs_access_entry *set,
3070	const struct cred *cred)
3071	{
3072	struct nfs_inode *nfsi = NFS_I(inode);
3073	struct rb_root *root_node = &nfsi->access_cache;
3074	struct rb_node **p = &root_node->rb_node;
3075	struct rb_node *parent = NULL;
3076	struct nfs_access_entry *entry;
3077	int cmp;
3078
3079	spin_lock(lock: &inode->i_lock);
3080	while (*p != NULL) {
3081	parent = *p;
3082	entry = rb_entry(parent, struct nfs_access_entry, rb_node);
3083	cmp = access_cmp(a: cred, b: entry);
3084
3085	if (cmp < 0)
3086	p = &parent->rb_left;
3087	else if (cmp > 0)
3088	p = &parent->rb_right;
3089	else
3090	goto found;
3091	}
3092	rb_link_node(node: &set->rb_node, parent, rb_link: p);
3093	rb_insert_color(&set->rb_node, root_node);
3094	list_add_tail(new: &set->lru, head: &nfsi->access_cache_entry_lru);
3095	spin_unlock(lock: &inode->i_lock);
3096	return;
3097	found:
3098	rb_replace_node(victim: parent, new: &set->rb_node, root: root_node);
3099	list_add_tail(new: &set->lru, head: &nfsi->access_cache_entry_lru);
3100	list_del(entry: &entry->lru);
3101	spin_unlock(lock: &inode->i_lock);
3102	nfs_access_free_entry(entry);
3103	}
3104
3105	void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set,
3106	const struct cred *cred)
3107	{
3108	struct nfs_access_entry *cache = kmalloc(size: sizeof(*cache), GFP_KERNEL);
3109	if (cache == NULL)
3110	return;
3111	RB_CLEAR_NODE(&cache->rb_node);
3112	cache->fsuid = cred->fsuid;
3113	cache->fsgid = cred->fsgid;
3114	cache->group_info = get_group_info(gi: cred->group_info);
3115	cache->mask = set->mask;
3116	cache->timestamp = ktime_get_ns();
3117
3118	/* The above field assignments must be visible
3119	* before this item appears on the lru. We cannot easily
3120	* use rcu_assign_pointer, so just force the memory barrier.
3121	*/
3122	smp_wmb();
3123	nfs_access_add_rbtree(inode, set: cache, cred);
3124
3125	/* Update accounting */
3126	smp_mb__before_atomic();
3127	atomic_long_inc(v: &nfs_access_nr_entries);
3128	smp_mb__after_atomic();
3129
3130	/* Add inode to global LRU list */
3131	if (!test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) {
3132	spin_lock(lock: &nfs_access_lru_lock);
3133	if (!test_and_set_bit(NFS_INO_ACL_LRU_SET, addr: &NFS_I(inode)->flags))
3134	list_add_tail(new: &NFS_I(inode)->access_cache_inode_lru,
3135	head: &nfs_access_lru_list);
3136	spin_unlock(lock: &nfs_access_lru_lock);
3137	}
3138	nfs_access_cache_enforce_limit();
3139	}
3140	EXPORT_SYMBOL_GPL(nfs_access_add_cache);
3141
3142	#define NFS_MAY_READ (NFS_ACCESS_READ)
3143	#define NFS_MAY_WRITE (NFS_ACCESS_MODIFY | \
3144	NFS_ACCESS_EXTEND | \
3145	NFS_ACCESS_DELETE)
3146	#define NFS_FILE_MAY_WRITE (NFS_ACCESS_MODIFY | \
3147	NFS_ACCESS_EXTEND)
3148	#define NFS_DIR_MAY_WRITE NFS_MAY_WRITE
3149	#define NFS_MAY_LOOKUP (NFS_ACCESS_LOOKUP)
3150	#define NFS_MAY_EXECUTE (NFS_ACCESS_EXECUTE)
3151	static int
3152	nfs_access_calc_mask(u32 access_result, umode_t umode)
3153	{
3154	int mask = 0;
3155
3156	if (access_result & NFS_MAY_READ)
3157	mask |= MAY_READ;
3158	if (S_ISDIR(umode)) {
3159	if ((access_result & NFS_DIR_MAY_WRITE) == NFS_DIR_MAY_WRITE)
3160	mask |= MAY_WRITE;
3161	if ((access_result & NFS_MAY_LOOKUP) == NFS_MAY_LOOKUP)
3162	mask |= MAY_EXEC;
3163	} else if (S_ISREG(umode)) {
3164	if ((access_result & NFS_FILE_MAY_WRITE) == NFS_FILE_MAY_WRITE)
3165	mask |= MAY_WRITE;
3166	if ((access_result & NFS_MAY_EXECUTE) == NFS_MAY_EXECUTE)
3167	mask |= MAY_EXEC;
3168	} else if (access_result & NFS_MAY_WRITE)
3169	mask |= MAY_WRITE;
3170	return mask;
3171	}
3172
3173	void nfs_access_set_mask(struct nfs_access_entry *entry, u32 access_result)
3174	{
3175	entry->mask = access_result;
3176	}
3177	EXPORT_SYMBOL_GPL(nfs_access_set_mask);
3178
3179	static int nfs_do_access(struct inode *inode, const struct cred *cred, int mask)
3180	{
3181	struct nfs_access_entry cache;
3182	bool may_block = (mask & MAY_NOT_BLOCK) == 0;
3183	int cache_mask = -1;
3184	int status;
3185
3186	trace_nfs_access_enter(inode);
3187
3188	status = nfs_access_get_cached(inode, cred, &cache.mask, may_block);
3189	if (status == 0)
3190	goto out_cached;
3191
3192	status = -ECHILD;
3193	if (!may_block)
3194	goto out;
3195
3196	/*
3197	* Determine which access bits we want to ask for...
3198	*/
3199	cache.mask = NFS_ACCESS_READ | NFS_ACCESS_MODIFY | NFS_ACCESS_EXTEND |
3200	nfs_access_xattr_mask(server: NFS_SERVER(inode));
3201	if (S_ISDIR(inode->i_mode))
3202	cache.mask |= NFS_ACCESS_DELETE | NFS_ACCESS_LOOKUP;
3203	else
3204	cache.mask |= NFS_ACCESS_EXECUTE;
3205	status = NFS_PROTO(inode)->access(inode, &cache, cred);
3206	if (status != 0) {
3207	if (status == -ESTALE) {
3208	if (!S_ISDIR(inode->i_mode))
3209	nfs_set_inode_stale(inode);
3210	else
3211	nfs_zap_caches(inode);
3212	}
3213	goto out;
3214	}
3215	nfs_access_add_cache(inode, &cache, cred);
3216	out_cached:
3217	cache_mask = nfs_access_calc_mask(access_result: cache.mask, umode: inode->i_mode);
3218	if ((mask & ~cache_mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) != 0)
3219	status = -EACCES;
3220	out:
3221	trace_nfs_access_exit(inode, mask, permitted: cache_mask, error: status);
3222	return status;
3223	}
3224
3225	static int nfs_open_permission_mask(int openflags)
3226	{
3227	int mask = 0;
3228
3229	if (openflags & __FMODE_EXEC) {
3230	/* ONLY check exec rights */
3231	mask = MAY_EXEC;
3232	} else {
3233	if ((openflags & O_ACCMODE) != O_WRONLY)
3234	mask |= MAY_READ;
3235	if ((openflags & O_ACCMODE) != O_RDONLY)
3236	mask |= MAY_WRITE;
3237	}
3238
3239	return mask;
3240	}
3241
3242	int nfs_may_open(struct inode *inode, const struct cred *cred, int openflags)
3243	{
3244	return nfs_do_access(inode, cred, mask: nfs_open_permission_mask(openflags));
3245	}
3246	EXPORT_SYMBOL_GPL(nfs_may_open);
3247
3248	static int nfs_execute_ok(struct inode *inode, int mask)
3249	{
3250	struct nfs_server *server = NFS_SERVER(inode);
3251	int ret = 0;
3252
3253	if (S_ISDIR(inode->i_mode))
3254	return 0;
3255	if (nfs_check_cache_invalid(inode, NFS_INO_INVALID_MODE)) {
3256	if (mask & MAY_NOT_BLOCK)
3257	return -ECHILD;
3258	ret = __nfs_revalidate_inode(server, inode);
3259	}
3260	if (ret == 0 && !execute_ok(inode))
3261	ret = -EACCES;
3262	return ret;
3263	}
3264
3265	int nfs_permission(struct mnt_idmap *idmap,
3266	struct inode *inode,
3267	int mask)
3268	{
3269	const struct cred *cred = current_cred();
3270	int res = 0;
3271
3272	nfs_inc_stats(inode, stat: NFSIOS_VFSACCESS);
3273
3274	if ((mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
3275	goto out;
3276	/* Is this sys_access() ? */
3277	if (mask & (MAY_ACCESS | MAY_CHDIR))
3278	goto force_lookup;
3279
3280	switch (inode->i_mode & S_IFMT) {
3281	case S_IFLNK:
3282	goto out;
3283	case S_IFREG:
3284	if ((mask & MAY_OPEN) &&
3285	nfs_server_capable(inode, NFS_CAP_ATOMIC_OPEN))
3286	return 0;
3287	break;
3288	case S_IFDIR:
3289	/*
3290	* Optimize away all write operations, since the server
3291	* will check permissions when we perform the op.
3292	*/
3293	if ((mask & MAY_WRITE) && !(mask & MAY_READ))
3294	goto out;
3295	}
3296
3297	force_lookup:
3298	if (!NFS_PROTO(inode)->access)
3299	goto out_notsup;
3300
3301	res = nfs_do_access(inode, cred, mask);
3302	out:
3303	if (!res && (mask & MAY_EXEC))
3304	res = nfs_execute_ok(inode, mask);
3305
3306	dfprintk(VFS, "NFS: permission(%s/%lu), mask=0x%x, res=%d\n",
3307	inode->i_sb->s_id, inode->i_ino, mask, res);
3308	return res;
3309	out_notsup:
3310	if (mask & MAY_NOT_BLOCK)
3311	return -ECHILD;
3312
3313	res = nfs_revalidate_inode(inode, NFS_INO_INVALID_MODE |
3314	NFS_INO_INVALID_OTHER);
3315	if (res == 0)
3316	res = generic_permission(&nop_mnt_idmap, inode, mask);
3317	goto out;
3318	}
3319	EXPORT_SYMBOL_GPL(nfs_permission);
3320