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	static int nfs_do_create(struct inode *dir, struct dentry *dentry,
60	umode_t mode, int open_flags);
61
62	const struct file_operations nfs_dir_operations = {
63	.llseek = nfs_llseek_dir,
64	.read = generic_read_dir,
65	.iterate_shared = nfs_readdir,
66	.open = nfs_opendir,
67	.release = nfs_closedir,
68	.fsync = nfs_fsync_dir,
69	};
70
71	const struct address_space_operations nfs_dir_aops = {
72	.free_folio = nfs_readdir_clear_array,
73	};
74
75	#define NFS_INIT_DTSIZE PAGE_SIZE
76
77	static struct nfs_open_dir_context *
78	alloc_nfs_open_dir_context(struct inode *dir)
79	{
80	struct nfs_inode *nfsi = NFS_I(inode: dir);
81	struct nfs_open_dir_context *ctx;
82
83	ctx = kzalloc(sizeof(*ctx), GFP_KERNEL_ACCOUNT);
84	if (ctx != NULL) {
85	ctx->attr_gencount = nfsi->attr_gencount;
86	ctx->dtsize = NFS_INIT_DTSIZE;
87	spin_lock(lock: &dir->i_lock);
88	if (list_empty(head: &nfsi->open_files) &&
89	(nfsi->cache_validity & NFS_INO_DATA_INVAL_DEFER))
90	nfs_set_cache_invalid(inode: dir,
91	NFS_INO_INVALID_DATA |
92	NFS_INO_REVAL_FORCED);
93	list_add_tail_rcu(new: &ctx->list, head: &nfsi->open_files);
94	memcpy(ctx->verf, nfsi->cookieverf, sizeof(ctx->verf));
95	spin_unlock(lock: &dir->i_lock);
96	return ctx;
97	}
98	return ERR_PTR(error: -ENOMEM);
99	}
100
101	static void put_nfs_open_dir_context(struct inode *dir, struct nfs_open_dir_context *ctx)
102	{
103	spin_lock(lock: &dir->i_lock);
104	list_del_rcu(entry: &ctx->list);
105	spin_unlock(lock: &dir->i_lock);
106	kfree_rcu(ctx, rcu_head);
107	}
108
109	/*
110	* Open file
111	*/
112	static int
113	nfs_opendir(struct inode *inode, struct file *filp)
114	{
115	int res = 0;
116	struct nfs_open_dir_context *ctx;
117
118	dfprintk(FILE, "NFS: open dir(%pD2)\n", filp);
119
120	nfs_inc_stats(inode, stat: NFSIOS_VFSOPEN);
121
122	ctx = alloc_nfs_open_dir_context(dir: inode);
123	if (IS_ERR(ptr: ctx)) {
124	res = PTR_ERR(ptr: ctx);
125	goto out;
126	}
127	filp->private_data = ctx;
128	out:
129	return res;
130	}
131
132	static int
133	nfs_closedir(struct inode *inode, struct file *filp)
134	{
135	put_nfs_open_dir_context(dir: file_inode(f: filp), ctx: filp->private_data);
136	return 0;
137	}
138
139	struct nfs_cache_array_entry {
140	u64 cookie;
141	u64 ino;
142	const char *name;
143	unsigned int name_len;
144	unsigned char d_type;
145	};
146
147	struct nfs_cache_array {
148	u64 change_attr;
149	u64 last_cookie;
150	unsigned int size;
151	unsigned char folio_full : 1,
152	folio_is_eof : 1,
153	cookies_are_ordered : 1;
154	struct nfs_cache_array_entry array[] __counted_by(size);
155	};
156
157	struct nfs_readdir_descriptor {
158	struct file *file;
159	struct folio *folio;
160	struct dir_context *ctx;
161	pgoff_t folio_index;
162	pgoff_t folio_index_max;
163	u64 dir_cookie;
164	u64 last_cookie;
165	loff_t current_index;
166
167	__be32 verf[NFS_DIR_VERIFIER_SIZE];
168	unsigned long dir_verifier;
169	unsigned long timestamp;
170	unsigned long gencount;
171	unsigned long attr_gencount;
172	unsigned int cache_entry_index;
173	unsigned int buffer_fills;
174	unsigned int dtsize;
175	bool clear_cache;
176	bool plus;
177	bool eob;
178	bool eof;
179	};
180
181	static void nfs_set_dtsize(struct nfs_readdir_descriptor *desc, unsigned int sz)
182	{
183	struct nfs_server *server = NFS_SERVER(inode: file_inode(f: desc->file));
184	unsigned int maxsize = server->dtsize;
185
186	if (sz > maxsize)
187	sz = maxsize;
188	if (sz < NFS_MIN_FILE_IO_SIZE)
189	sz = NFS_MIN_FILE_IO_SIZE;
190	desc->dtsize = sz;
191	}
192
193	static void nfs_shrink_dtsize(struct nfs_readdir_descriptor *desc)
194	{
195	nfs_set_dtsize(desc, sz: desc->dtsize >> 1);
196	}
197
198	static void nfs_grow_dtsize(struct nfs_readdir_descriptor *desc)
199	{
200	nfs_set_dtsize(desc, sz: desc->dtsize << 1);
201	}
202
203	static void nfs_readdir_folio_init_array(struct folio *folio, u64 last_cookie,
204	u64 change_attr)
205	{
206	struct nfs_cache_array *array;
207
208	array = kmap_local_folio(folio, offset: 0);
209	array->change_attr = change_attr;
210	array->last_cookie = last_cookie;
211	array->size = 0;
212	array->folio_full = 0;
213	array->folio_is_eof = 0;
214	array->cookies_are_ordered = 1;
215	kunmap_local(array);
216	}
217
218	/*
219	* we are freeing strings created by nfs_add_to_readdir_array()
220	*/
221	static void nfs_readdir_clear_array(struct folio *folio)
222	{
223	struct nfs_cache_array *array;
224	unsigned int i;
225
226	array = kmap_local_folio(folio, offset: 0);
227	for (i = 0; i < array->size; i++)
228	kfree(objp: array->array[i].name);
229	array->size = 0;
230	kunmap_local(array);
231	}
232
233	static void nfs_readdir_folio_reinit_array(struct folio *folio, u64 last_cookie,
234	u64 change_attr)
235	{
236	nfs_readdir_clear_array(folio);
237	nfs_readdir_folio_init_array(folio, last_cookie, change_attr);
238	}
239
240	static struct folio *
241	nfs_readdir_folio_array_alloc(u64 last_cookie, gfp_t gfp_flags)
242	{
243	struct folio *folio = folio_alloc(gfp_flags, 0);
244	if (folio)
245	nfs_readdir_folio_init_array(folio, last_cookie, change_attr: 0);
246	return folio;
247	}
248
249	static void nfs_readdir_folio_array_free(struct folio *folio)
250	{
251	if (folio) {
252	nfs_readdir_clear_array(folio);
253	folio_put(folio);
254	}
255	}
256
257	static u64 nfs_readdir_array_index_cookie(struct nfs_cache_array *array)
258	{
259	return array->size == 0 ? array->last_cookie : array->array[0].cookie;
260	}
261
262	static void nfs_readdir_array_set_eof(struct nfs_cache_array *array)
263	{
264	array->folio_is_eof = 1;
265	array->folio_full = 1;
266	}
267
268	static bool nfs_readdir_array_is_full(struct nfs_cache_array *array)
269	{
270	return array->folio_full;
271	}
272
273	/*
274	* the caller is responsible for freeing qstr.name
275	* when called by nfs_readdir_add_to_array, the strings will be freed in
276	* nfs_clear_readdir_array()
277	*/
278	static const char *nfs_readdir_copy_name(const char *name, unsigned int len)
279	{
280	const char *ret = kmemdup_nul(s: name, len, GFP_KERNEL);
281
282	/*
283	* Avoid a kmemleak false positive. The pointer to the name is stored
284	* in a page cache page which kmemleak does not scan.
285	*/
286	if (ret != NULL)
287	kmemleak_not_leak(ptr: ret);
288	return ret;
289	}
290
291	static size_t nfs_readdir_array_maxentries(void)
292	{
293	return (PAGE_SIZE - sizeof(struct nfs_cache_array)) /
294	sizeof(struct nfs_cache_array_entry);
295	}
296
297	/*
298	* Check that the next array entry lies entirely within the page bounds
299	*/
300	static int nfs_readdir_array_can_expand(struct nfs_cache_array *array)
301	{
302	if (array->folio_full)
303	return -ENOSPC;
304	if (array->size == nfs_readdir_array_maxentries()) {
305	array->folio_full = 1;
306	return -ENOSPC;
307	}
308	return 0;
309	}
310
311	static int nfs_readdir_folio_array_append(struct folio *folio,
312	const struct nfs_entry *entry,
313	u64 *cookie)
314	{
315	struct nfs_cache_array *array;
316	struct nfs_cache_array_entry *cache_entry;
317	const char *name;
318	int ret = -ENOMEM;
319
320	name = nfs_readdir_copy_name(name: entry->name, len: entry->len);
321
322	array = kmap_local_folio(folio, offset: 0);
323	if (!name)
324	goto out;
325	ret = nfs_readdir_array_can_expand(array);
326	if (ret) {
327	kfree(objp: name);
328	goto out;
329	}
330
331	array->size++;
332	cache_entry = &array->array[array->size - 1];
333	cache_entry->cookie = array->last_cookie;
334	cache_entry->ino = entry->ino;
335	cache_entry->d_type = entry->d_type;
336	cache_entry->name_len = entry->len;
337	cache_entry->name = name;
338	array->last_cookie = entry->cookie;
339	if (array->last_cookie <= cache_entry->cookie)
340	array->cookies_are_ordered = 0;
341	if (entry->eof != 0)
342	nfs_readdir_array_set_eof(array);
343	out:
344	*cookie = array->last_cookie;
345	kunmap_local(array);
346	return ret;
347	}
348
349	#define NFS_READDIR_COOKIE_MASK (U32_MAX >> 14)
350	/*
351	* Hash algorithm allowing content addressible access to sequences
352	* of directory cookies. Content is addressed by the value of the
353	* cookie index of the first readdir entry in a page.
354	*
355	* We select only the first 18 bits to avoid issues with excessive
356	* memory use for the page cache XArray. 18 bits should allow the caching
357	* of 262144 pages of sequences of readdir entries. Since each page holds
358	* 127 readdir entries for a typical 64-bit system, that works out to a
359	* cache of ~ 33 million entries per directory.
360	*/
361	static pgoff_t nfs_readdir_folio_cookie_hash(u64 cookie)
362	{
363	if (cookie == 0)
364	return 0;
365	return hash_64(val: cookie, bits: 18);
366	}
367
368	static bool nfs_readdir_folio_validate(struct folio *folio, u64 last_cookie,
369	u64 change_attr)
370	{
371	struct nfs_cache_array *array = kmap_local_folio(folio, offset: 0);
372	int ret = true;
373
374	if (array->change_attr != change_attr)
375	ret = false;
376	if (nfs_readdir_array_index_cookie(array) != last_cookie)
377	ret = false;
378	kunmap_local(array);
379	return ret;
380	}
381
382	static void nfs_readdir_folio_unlock_and_put(struct folio *folio)
383	{
384	folio_unlock(folio);
385	folio_put(folio);
386	}
387
388	static void nfs_readdir_folio_init_and_validate(struct folio *folio, u64 cookie,
389	u64 change_attr)
390	{
391	if (folio_test_uptodate(folio)) {
392	if (nfs_readdir_folio_validate(folio, last_cookie: cookie, change_attr))
393	return;
394	nfs_readdir_clear_array(folio);
395	}
396	nfs_readdir_folio_init_array(folio, last_cookie: cookie, change_attr);
397	folio_mark_uptodate(folio);
398	}
399
400	static struct folio *nfs_readdir_folio_get_locked(struct address_space *mapping,
401	u64 cookie, u64 change_attr)
402	{
403	pgoff_t index = nfs_readdir_folio_cookie_hash(cookie);
404	struct folio *folio;
405
406	folio = filemap_grab_folio(mapping, index);
407	if (IS_ERR(ptr: folio))
408	return NULL;
409	nfs_readdir_folio_init_and_validate(folio, cookie, change_attr);
410	return folio;
411	}
412
413	static u64 nfs_readdir_folio_last_cookie(struct folio *folio)
414	{
415	struct nfs_cache_array *array;
416	u64 ret;
417
418	array = kmap_local_folio(folio, offset: 0);
419	ret = array->last_cookie;
420	kunmap_local(array);
421	return ret;
422	}
423
424	static bool nfs_readdir_folio_needs_filling(struct folio *folio)
425	{
426	struct nfs_cache_array *array;
427	bool ret;
428
429	array = kmap_local_folio(folio, offset: 0);
430	ret = !nfs_readdir_array_is_full(array);
431	kunmap_local(array);
432	return ret;
433	}
434
435	static void nfs_readdir_folio_set_eof(struct folio *folio)
436	{
437	struct nfs_cache_array *array;
438
439	array = kmap_local_folio(folio, offset: 0);
440	nfs_readdir_array_set_eof(array);
441	kunmap_local(array);
442	}
443
444	static struct folio *nfs_readdir_folio_get_next(struct address_space *mapping,
445	u64 cookie, u64 change_attr)
446	{
447	pgoff_t index = nfs_readdir_folio_cookie_hash(cookie);
448	struct folio *folio;
449
450	folio = __filemap_get_folio(mapping, index,
451	FGP_LOCK|FGP_CREAT|FGP_NOFS|FGP_NOWAIT,
452	gfp: mapping_gfp_mask(mapping));
453	if (IS_ERR(ptr: folio))
454	return NULL;
455	nfs_readdir_folio_init_and_validate(folio, cookie, change_attr);
456	if (nfs_readdir_folio_last_cookie(folio) != cookie)
457	nfs_readdir_folio_reinit_array(folio, last_cookie: cookie, change_attr);
458	return folio;
459	}
460
461	static inline
462	int is_32bit_api(void)
463	{
464	#ifdef CONFIG_COMPAT
465	return in_compat_syscall();
466	#else
467	return (BITS_PER_LONG == 32);
468	#endif
469	}
470
471	static
472	bool nfs_readdir_use_cookie(const struct file *filp)
473	{
474	if ((filp->f_mode & FMODE_32BITHASH) ||
475	(!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
476	return false;
477	return true;
478	}
479
480	static void nfs_readdir_seek_next_array(struct nfs_cache_array *array,
481	struct nfs_readdir_descriptor *desc)
482	{
483	if (array->folio_full) {
484	desc->last_cookie = array->last_cookie;
485	desc->current_index += array->size;
486	desc->cache_entry_index = 0;
487	desc->folio_index++;
488	} else
489	desc->last_cookie = nfs_readdir_array_index_cookie(array);
490	}
491
492	static void nfs_readdir_rewind_search(struct nfs_readdir_descriptor *desc)
493	{
494	desc->current_index = 0;
495	desc->last_cookie = 0;
496	desc->folio_index = 0;
497	}
498
499	static int nfs_readdir_search_for_pos(struct nfs_cache_array *array,
500	struct nfs_readdir_descriptor *desc)
501	{
502	loff_t diff = desc->ctx->pos - desc->current_index;
503	unsigned int index;
504
505	if (diff < 0)
506	goto out_eof;
507	if (diff >= array->size) {
508	if (array->folio_is_eof)
509	goto out_eof;
510	nfs_readdir_seek_next_array(array, desc);
511	return -EAGAIN;
512	}
513
514	index = (unsigned int)diff;
515	desc->dir_cookie = array->array[index].cookie;
516	desc->cache_entry_index = index;
517	return 0;
518	out_eof:
519	desc->eof = true;
520	return -EBADCOOKIE;
521	}
522
523	static bool nfs_readdir_array_cookie_in_range(struct nfs_cache_array *array,
524	u64 cookie)
525	{
526	if (!array->cookies_are_ordered)
527	return true;
528	/* Optimisation for monotonically increasing cookies */
529	if (cookie >= array->last_cookie)
530	return false;
531	if (array->size && cookie < array->array[0].cookie)
532	return false;
533	return true;
534	}
535
536	static int nfs_readdir_search_for_cookie(struct nfs_cache_array *array,
537	struct nfs_readdir_descriptor *desc)
538	{
539	unsigned int i;
540	int status = -EAGAIN;
541
542	if (!nfs_readdir_array_cookie_in_range(array, cookie: desc->dir_cookie))
543	goto check_eof;
544
545	for (i = 0; i < array->size; i++) {
546	if (array->array[i].cookie == desc->dir_cookie) {
547	if (nfs_readdir_use_cookie(filp: desc->file))
548	desc->ctx->pos = desc->dir_cookie;
549	else
550	desc->ctx->pos = desc->current_index + i;
551	desc->cache_entry_index = i;
552	return 0;
553	}
554	}
555	check_eof:
556	if (array->folio_is_eof) {
557	status = -EBADCOOKIE;
558	if (desc->dir_cookie == array->last_cookie)
559	desc->eof = true;
560	} else
561	nfs_readdir_seek_next_array(array, desc);
562	return status;
563	}
564
565	static int nfs_readdir_search_array(struct nfs_readdir_descriptor *desc)
566	{
567	struct nfs_cache_array *array;
568	int status;
569
570	array = kmap_local_folio(folio: desc->folio, offset: 0);
571
572	if (desc->dir_cookie == 0)
573	status = nfs_readdir_search_for_pos(array, desc);
574	else
575	status = nfs_readdir_search_for_cookie(array, desc);
576
577	kunmap_local(array);
578	return status;
579	}
580
581	/* Fill a page with xdr information before transferring to the cache page */
582	static int nfs_readdir_xdr_filler(struct nfs_readdir_descriptor *desc,
583	__be32 *verf, u64 cookie,
584	struct page **pages, size_t bufsize,
585	__be32 *verf_res)
586	{
587	struct inode *inode = file_inode(f: desc->file);
588	struct nfs_readdir_arg arg = {
589	.dentry = file_dentry(file: desc->file),
590	.cred = desc->file->f_cred,
591	.verf = verf,
592	.cookie = cookie,
593	.pages = pages,
594	.page_len = bufsize,
595	.plus = desc->plus,
596	};
597	struct nfs_readdir_res res = {
598	.verf = verf_res,
599	};
600	unsigned long timestamp, gencount;
601	int error;
602
603	again:
604	timestamp = jiffies;
605	gencount = nfs_inc_attr_generation_counter();
606	desc->dir_verifier = nfs_save_change_attribute(dir: inode);
607	error = NFS_PROTO(inode)->readdir(&arg, &res);
608	if (error < 0) {
609	/* We requested READDIRPLUS, but the server doesn't grok it */
610	if (error == -ENOTSUPP && desc->plus) {
611	NFS_SERVER(inode)->caps &= ~NFS_CAP_READDIRPLUS;
612	desc->plus = arg.plus = false;
613	goto again;
614	}
615	goto error;
616	}
617	desc->timestamp = timestamp;
618	desc->gencount = gencount;
619	error:
620	return error;
621	}
622
623	static int xdr_decode(struct nfs_readdir_descriptor *desc,
624	struct nfs_entry *entry, struct xdr_stream *xdr)
625	{
626	struct inode *inode = file_inode(f: desc->file);
627	int error;
628
629	error = NFS_PROTO(inode)->decode_dirent(xdr, entry, desc->plus);
630	if (error)
631	return error;
632	entry->fattr->time_start = desc->timestamp;
633	entry->fattr->gencount = desc->gencount;
634	return 0;
635	}
636
637	/* Match file and dirent using either filehandle or fileid
638	* Note: caller is responsible for checking the fsid
639	*/
640	static
641	int nfs_same_file(struct dentry *dentry, struct nfs_entry *entry)
642	{
643	struct inode *inode;
644	struct nfs_inode *nfsi;
645
646	if (d_really_is_negative(dentry))
647	return 0;
648
649	inode = d_inode(dentry);
650	if (is_bad_inode(inode) || NFS_STALE(inode))
651	return 0;
652
653	nfsi = NFS_I(inode);
654	if (entry->fattr->fileid != nfsi->fileid)
655	return 0;
656	if (entry->fh->size && nfs_compare_fh(a: entry->fh, b: &nfsi->fh) != 0)
657	return 0;
658	return 1;
659	}
660
661	#define NFS_READDIR_CACHE_USAGE_THRESHOLD (8UL)
662
663	static bool nfs_use_readdirplus(struct inode *dir, struct dir_context *ctx,
664	unsigned int cache_hits,
665	unsigned int cache_misses)
666	{
667	if (!nfs_server_capable(inode: dir, NFS_CAP_READDIRPLUS))
668	return false;
669	if (NFS_SERVER(inode: dir)->flags & NFS_MOUNT_FORCE_RDIRPLUS)
670	return true;
671	if (ctx->pos == 0 ||
672	cache_hits + cache_misses > NFS_READDIR_CACHE_USAGE_THRESHOLD)
673	return true;
674	return false;
675	}
676
677	/*
678	* This function is called by the getattr code to request the
679	* use of readdirplus to accelerate any future lookups in the same
680	* directory.
681	*/
682	void nfs_readdir_record_entry_cache_hit(struct inode *dir)
683	{
684	struct nfs_inode *nfsi = NFS_I(inode: dir);
685	struct nfs_open_dir_context *ctx;
686
687	if (nfs_server_capable(inode: dir, NFS_CAP_READDIRPLUS) &&
688	S_ISDIR(dir->i_mode)) {
689	rcu_read_lock();
690	list_for_each_entry_rcu (ctx, &nfsi->open_files, list)
691	atomic_inc(v: &ctx->cache_hits);
692	rcu_read_unlock();
693	}
694	}
695
696	/*
697	* This function is mainly for use by nfs_getattr().
698	*
699	* If this is an 'ls -l', we want to force use of readdirplus.
700	*/
701	void nfs_readdir_record_entry_cache_miss(struct inode *dir)
702	{
703	struct nfs_inode *nfsi = NFS_I(inode: dir);
704	struct nfs_open_dir_context *ctx;
705
706	if (nfs_server_capable(inode: dir, NFS_CAP_READDIRPLUS) &&
707	S_ISDIR(dir->i_mode)) {
708	rcu_read_lock();
709	list_for_each_entry_rcu (ctx, &nfsi->open_files, list)
710	atomic_inc(v: &ctx->cache_misses);
711	rcu_read_unlock();
712	}
713	}
714
715	static void nfs_lookup_advise_force_readdirplus(struct inode *dir,
716	unsigned int flags)
717	{
718	if (nfs_server_capable(inode: dir, NFS_CAP_CASE_INSENSITIVE))
719	return;
720	if (flags & (LOOKUP_EXCL | LOOKUP_PARENT | LOOKUP_REVAL))
721	return;
722	nfs_readdir_record_entry_cache_miss(dir);
723	}
724
725	static
726	void nfs_prime_dcache(struct dentry *parent, struct nfs_entry *entry,
727	unsigned long dir_verifier)
728	{
729	struct qstr filename = QSTR_INIT(entry->name, entry->len);
730	DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
731	struct dentry *dentry;
732	struct dentry *alias;
733	struct inode *inode;
734	int status;
735
736	if (!(entry->fattr->valid & NFS_ATTR_FATTR_FILEID))
737	return;
738	if (!(entry->fattr->valid & NFS_ATTR_FATTR_FSID))
739	return;
740	if (filename.len == 0)
741	return;
742	/* Validate that the name doesn't contain any illegal '\0' */
743	if (strnlen(p: filename.name, maxlen: filename.len) != filename.len)
744	return;
745	/* ...or '/' */
746	if (strnchr(filename.name, filename.len, '/'))
747	return;
748	if (filename.name[0] == '.') {
749	if (filename.len == 1)
750	return;
751	if (filename.len == 2 && filename.name[1] == '.')
752	return;
753	}
754	filename.hash = full_name_hash(salt: parent, filename.name, filename.len);
755
756	dentry = d_lookup(parent, &filename);
757	again:
758	if (!dentry) {
759	dentry = d_alloc_parallel(parent, &filename, &wq);
760	if (IS_ERR(ptr: dentry))
761	return;
762	}
763	if (!d_in_lookup(dentry)) {
764	/* Is there a mountpoint here? If so, just exit */
765	if (!nfs_fsid_equal(a: &NFS_SB(s: dentry->d_sb)->fsid,
766	b: &entry->fattr->fsid))
767	goto out;
768	if (nfs_same_file(dentry, entry)) {
769	if (!entry->fh->size)
770	goto out;
771	nfs_set_verifier(dentry, verf: dir_verifier);
772	status = nfs_refresh_inode(d_inode(dentry), entry->fattr);
773	if (!status)
774	nfs_setsecurity(inode: d_inode(dentry), fattr: entry->fattr);
775	trace_nfs_readdir_lookup_revalidate(dir: d_inode(dentry: parent),
776	dentry, flags: 0, error: status);
777	goto out;
778	} else {
779	trace_nfs_readdir_lookup_revalidate_failed(
780	dir: d_inode(dentry: parent), dentry, flags: 0);
781	d_invalidate(dentry);
782	dput(dentry);
783	dentry = NULL;
784	goto again;
785	}
786	}
787	if (!entry->fh->size) {
788	d_lookup_done(dentry);
789	goto out;
790	}
791
792	inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr);
793	alias = d_splice_alias(inode, dentry);
794	d_lookup_done(dentry);
795	if (alias) {
796	if (IS_ERR(ptr: alias))
797	goto out;
798	dput(dentry);
799	dentry = alias;
800	}
801	nfs_set_verifier(dentry, verf: dir_verifier);
802	trace_nfs_readdir_lookup(dir: d_inode(dentry: parent), dentry, flags: 0);
803	out:
804	dput(dentry);
805	}
806
807	static int nfs_readdir_entry_decode(struct nfs_readdir_descriptor *desc,
808	struct nfs_entry *entry,
809	struct xdr_stream *stream)
810	{
811	int ret;
812
813	if (entry->fattr->label)
814	entry->fattr->label->len = NFS4_MAXLABELLEN;
815	ret = xdr_decode(desc, entry, xdr: stream);
816	if (ret || !desc->plus)
817	return ret;
818	nfs_prime_dcache(parent: file_dentry(file: desc->file), entry, dir_verifier: desc->dir_verifier);
819	return 0;
820	}
821
822	/* Perform conversion from xdr to cache array */
823	static int nfs_readdir_folio_filler(struct nfs_readdir_descriptor *desc,
824	struct nfs_entry *entry,
825	struct page **xdr_pages, unsigned int buflen,
826	struct folio **arrays, size_t narrays,
827	u64 change_attr)
828	{
829	struct address_space *mapping = desc->file->f_mapping;
830	struct folio *new, *folio = *arrays;
831	struct xdr_stream stream;
832	struct page *scratch;
833	struct xdr_buf buf;
834	u64 cookie;
835	int status;
836
837	scratch = alloc_page(GFP_KERNEL);
838	if (scratch == NULL)
839	return -ENOMEM;
840
841	xdr_init_decode_pages(xdr: &stream, buf: &buf, pages: xdr_pages, len: buflen);
842	xdr_set_scratch_page(xdr: &stream, page: scratch);
843
844	do {
845	status = nfs_readdir_entry_decode(desc, entry, stream: &stream);
846	if (status != 0)
847	break;
848
849	status = nfs_readdir_folio_array_append(folio, entry, cookie: &cookie);
850	if (status != -ENOSPC)
851	continue;
852
853	if (folio->mapping != mapping) {
854	if (!--narrays)
855	break;
856	new = nfs_readdir_folio_array_alloc(last_cookie: cookie, GFP_KERNEL);
857	if (!new)
858	break;
859	arrays++;
860	*arrays = folio = new;
861	} else {
862	new = nfs_readdir_folio_get_next(mapping, cookie,
863	change_attr);
864	if (!new)
865	break;
866	if (folio != *arrays)
867	nfs_readdir_folio_unlock_and_put(folio);
868	folio = new;
869	}
870	desc->folio_index_max++;
871	status = nfs_readdir_folio_array_append(folio, entry, cookie: &cookie);
872	} while (!status && !entry->eof);
873
874	switch (status) {
875	case -EBADCOOKIE:
876	if (!entry->eof)
877	break;
878	nfs_readdir_folio_set_eof(folio);
879	fallthrough;
880	case -EAGAIN:
881	status = 0;
882	break;
883	case -ENOSPC:
884	status = 0;
885	if (!desc->plus)
886	break;
887	while (!nfs_readdir_entry_decode(desc, entry, stream: &stream))
888	;
889	}
890
891	if (folio != *arrays)
892	nfs_readdir_folio_unlock_and_put(folio);
893
894	put_page(page: scratch);
895	return status;
896	}
897
898	static void nfs_readdir_free_pages(struct page **pages, size_t npages)
899	{
900	while (npages--)
901	put_page(page: pages[npages]);
902	kfree(objp: pages);
903	}
904
905	/*
906	* nfs_readdir_alloc_pages() will allocate pages that must be freed with a call
907	* to nfs_readdir_free_pages()
908	*/
909	static struct page **nfs_readdir_alloc_pages(size_t npages)
910	{
911	struct page **pages;
912	size_t i;
913
914	pages = kmalloc_array(npages, sizeof(*pages), GFP_KERNEL);
915	if (!pages)
916	return NULL;
917	for (i = 0; i < npages; i++) {
918	struct page *page = alloc_page(GFP_KERNEL);
919	if (page == NULL)
920	goto out_freepages;
921	pages[i] = page;
922	}
923	return pages;
924
925	out_freepages:
926	nfs_readdir_free_pages(pages, npages: i);
927	return NULL;
928	}
929
930	static int nfs_readdir_xdr_to_array(struct nfs_readdir_descriptor *desc,
931	__be32 *verf_arg, __be32 *verf_res,
932	struct folio **arrays, size_t narrays)
933	{
934	u64 change_attr;
935	struct page **pages;
936	struct folio *folio = *arrays;
937	struct nfs_entry *entry;
938	size_t array_size;
939	struct inode *inode = file_inode(f: desc->file);
940	unsigned int dtsize = desc->dtsize;
941	unsigned int pglen;
942	int status = -ENOMEM;
943
944	entry = kzalloc(sizeof(*entry), GFP_KERNEL);
945	if (!entry)
946	return -ENOMEM;
947	entry->cookie = nfs_readdir_folio_last_cookie(folio);
948	entry->fh = nfs_alloc_fhandle();
949	entry->fattr = nfs_alloc_fattr_with_label(server: NFS_SERVER(inode));
950	entry->server = NFS_SERVER(inode);
951	if (entry->fh == NULL || entry->fattr == NULL)
952	goto out;
953
954	array_size = (dtsize + PAGE_SIZE - 1) >> PAGE_SHIFT;
955	pages = nfs_readdir_alloc_pages(npages: array_size);
956	if (!pages)
957	goto out;
958
959	change_attr = inode_peek_iversion_raw(inode);
960	status = nfs_readdir_xdr_filler(desc, verf: verf_arg, cookie: entry->cookie, pages,
961	bufsize: dtsize, verf_res);
962	if (status < 0)
963	goto free_pages;
964
965	pglen = status;
966	if (pglen != 0)
967	status = nfs_readdir_folio_filler(desc, entry, xdr_pages: pages, buflen: pglen,
968	arrays, narrays, change_attr);
969	else
970	nfs_readdir_folio_set_eof(folio);
971	desc->buffer_fills++;
972
973	free_pages:
974	nfs_readdir_free_pages(pages, npages: array_size);
975	out:
976	nfs_free_fattr(fattr: entry->fattr);
977	nfs_free_fhandle(fh: entry->fh);
978	kfree(objp: entry);
979	return status;
980	}
981
982	static void nfs_readdir_folio_put(struct nfs_readdir_descriptor *desc)
983	{
984	folio_put(folio: desc->folio);
985	desc->folio = NULL;
986	}
987
988	static void
989	nfs_readdir_folio_unlock_and_put_cached(struct nfs_readdir_descriptor *desc)
990	{
991	folio_unlock(folio: desc->folio);
992	nfs_readdir_folio_put(desc);
993	}
994
995	static struct folio *
996	nfs_readdir_folio_get_cached(struct nfs_readdir_descriptor *desc)
997	{
998	struct address_space *mapping = desc->file->f_mapping;
999	u64 change_attr = inode_peek_iversion_raw(inode: mapping->host);
1000	u64 cookie = desc->last_cookie;
1001	struct folio *folio;
1002
1003	folio = nfs_readdir_folio_get_locked(mapping, cookie, change_attr);
1004	if (!folio)
1005	return NULL;
1006	if (desc->clear_cache && !nfs_readdir_folio_needs_filling(folio))
1007	nfs_readdir_folio_reinit_array(folio, last_cookie: cookie, change_attr);
1008	return folio;
1009	}
1010
1011	/*
1012	* Returns 0 if desc->dir_cookie was found on page desc->page_index
1013	* and locks the page to prevent removal from the page cache.
1014	*/
1015	static int find_and_lock_cache_page(struct nfs_readdir_descriptor *desc)
1016	{
1017	struct inode *inode = file_inode(f: desc->file);
1018	struct nfs_inode *nfsi = NFS_I(inode);
1019	__be32 verf[NFS_DIR_VERIFIER_SIZE];
1020	int res;
1021
1022	desc->folio = nfs_readdir_folio_get_cached(desc);
1023	if (!desc->folio)
1024	return -ENOMEM;
1025	if (nfs_readdir_folio_needs_filling(folio: desc->folio)) {
1026	/* Grow the dtsize if we had to go back for more pages */
1027	if (desc->folio_index == desc->folio_index_max)
1028	nfs_grow_dtsize(desc);
1029	desc->folio_index_max = desc->folio_index;
1030	trace_nfs_readdir_cache_fill(file: desc->file, verifier: nfsi->cookieverf,
1031	cookie: desc->last_cookie,
1032	page_index: desc->folio->index, dtsize: desc->dtsize);
1033	res = nfs_readdir_xdr_to_array(desc, verf_arg: nfsi->cookieverf, verf_res: verf,
1034	arrays: &desc->folio, narrays: 1);
1035	if (res < 0) {
1036	nfs_readdir_folio_unlock_and_put_cached(desc);
1037	trace_nfs_readdir_cache_fill_done(inode, error: res);
1038	if (res == -EBADCOOKIE || res == -ENOTSYNC) {
1039	invalidate_inode_pages2(mapping: desc->file->f_mapping);
1040	nfs_readdir_rewind_search(desc);
1041	trace_nfs_readdir_invalidate_cache_range(
1042	inode, range_start: 0, MAX_LFS_FILESIZE);
1043	return -EAGAIN;
1044	}
1045	return res;
1046	}
1047	/*
1048	* Set the cookie verifier if the page cache was empty
1049	*/
1050	if (desc->last_cookie == 0 &&
1051	memcmp(p: nfsi->cookieverf, q: verf, size: sizeof(nfsi->cookieverf))) {
1052	memcpy(nfsi->cookieverf, verf,
1053	sizeof(nfsi->cookieverf));
1054	invalidate_inode_pages2_range(mapping: desc->file->f_mapping, start: 1,
1055	end: -1);
1056	trace_nfs_readdir_invalidate_cache_range(
1057	inode, range_start: 1, MAX_LFS_FILESIZE);
1058	}
1059	desc->clear_cache = false;
1060	}
1061	res = nfs_readdir_search_array(desc);
1062	if (res == 0)
1063	return 0;
1064	nfs_readdir_folio_unlock_and_put_cached(desc);
1065	return res;
1066	}
1067
1068	/* Search for desc->dir_cookie from the beginning of the page cache */
1069	static int readdir_search_pagecache(struct nfs_readdir_descriptor *desc)
1070	{
1071	int res;
1072
1073	do {
1074	res = find_and_lock_cache_page(desc);
1075	} while (res == -EAGAIN);
1076	return res;
1077	}
1078
1079	#define NFS_READDIR_CACHE_MISS_THRESHOLD (16UL)
1080
1081	/*
1082	* Once we've found the start of the dirent within a page: fill 'er up...
1083	*/
1084	static void nfs_do_filldir(struct nfs_readdir_descriptor *desc,
1085	const __be32 *verf)
1086	{
1087	struct file *file = desc->file;
1088	struct nfs_cache_array *array;
1089	unsigned int i;
1090	bool first_emit = !desc->dir_cookie;
1091
1092	array = kmap_local_folio(folio: desc->folio, offset: 0);
1093	for (i = desc->cache_entry_index; i < array->size; i++) {
1094	struct nfs_cache_array_entry *ent;
1095
1096	/*
1097	* nfs_readdir_handle_cache_misses return force clear at
1098	* (cache_misses > NFS_READDIR_CACHE_MISS_THRESHOLD) for
1099	* readdir heuristic, NFS_READDIR_CACHE_MISS_THRESHOLD + 1
1100	* entries need be emitted here.
1101	*/
1102	if (first_emit && i > NFS_READDIR_CACHE_MISS_THRESHOLD + 2) {
1103	desc->eob = true;
1104	break;
1105	}
1106
1107	ent = &array->array[i];
1108	if (!dir_emit(ctx: desc->ctx, name: ent->name, namelen: ent->name_len,
1109	ino: nfs_compat_user_ino64(fileid: ent->ino), type: ent->d_type)) {
1110	desc->eob = true;
1111	break;
1112	}
1113	memcpy(desc->verf, verf, sizeof(desc->verf));
1114	if (i == array->size - 1) {
1115	desc->dir_cookie = array->last_cookie;
1116	nfs_readdir_seek_next_array(array, desc);
1117	} else {
1118	desc->dir_cookie = array->array[i + 1].cookie;
1119	desc->last_cookie = array->array[0].cookie;
1120	}
1121	if (nfs_readdir_use_cookie(filp: file))
1122	desc->ctx->pos = desc->dir_cookie;
1123	else
1124	desc->ctx->pos++;
1125	}
1126	if (array->folio_is_eof)
1127	desc->eof = !desc->eob;
1128
1129	kunmap_local(array);
1130	dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling ended @ cookie %llu\n",
1131	(unsigned long long)desc->dir_cookie);
1132	}
1133
1134	/*
1135	* If we cannot find a cookie in our cache, we suspect that this is
1136	* because it points to a deleted file, so we ask the server to return
1137	* whatever it thinks is the next entry. We then feed this to filldir.
1138	* If all goes well, we should then be able to find our way round the
1139	* cache on the next call to readdir_search_pagecache();
1140	*
1141	* NOTE: we cannot add the anonymous page to the pagecache because
1142	* the data it contains might not be page aligned. Besides,
1143	* we should already have a complete representation of the
1144	* directory in the page cache by the time we get here.
1145	*/
1146	static int uncached_readdir(struct nfs_readdir_descriptor *desc)
1147	{
1148	struct folio **arrays;
1149	size_t i, sz = 512;
1150	__be32 verf[NFS_DIR_VERIFIER_SIZE];
1151	int status = -ENOMEM;
1152
1153	dfprintk(DIRCACHE, "NFS: uncached_readdir() searching for cookie %llu\n",
1154	(unsigned long long)desc->dir_cookie);
1155
1156	arrays = kcalloc(sz, sizeof(*arrays), GFP_KERNEL);
1157	if (!arrays)
1158	goto out;
1159	arrays[0] = nfs_readdir_folio_array_alloc(last_cookie: desc->dir_cookie, GFP_KERNEL);
1160	if (!arrays[0])
1161	goto out;
1162
1163	desc->folio_index = 0;
1164	desc->cache_entry_index = 0;
1165	desc->last_cookie = desc->dir_cookie;
1166	desc->folio_index_max = 0;
1167
1168	trace_nfs_readdir_uncached(file: desc->file, verifier: desc->verf, cookie: desc->last_cookie,
1169	page_index: -1, dtsize: desc->dtsize);
1170
1171	status = nfs_readdir_xdr_to_array(desc, verf_arg: desc->verf, verf_res: verf, arrays, narrays: sz);
1172	if (status < 0) {
1173	trace_nfs_readdir_uncached_done(inode: file_inode(f: desc->file), error: status);
1174	goto out_free;
1175	}
1176
1177	for (i = 0; !desc->eob && i < sz && arrays[i]; i++) {
1178	desc->folio = arrays[i];
1179	nfs_do_filldir(desc, verf);
1180	}
1181	desc->folio = NULL;
1182
1183	/*
1184	* Grow the dtsize if we have to go back for more pages,
1185	* or shrink it if we're reading too many.
1186	*/
1187	if (!desc->eof) {
1188	if (!desc->eob)
1189	nfs_grow_dtsize(desc);
1190	else if (desc->buffer_fills == 1 &&
1191	i < (desc->folio_index_max >> 1))
1192	nfs_shrink_dtsize(desc);
1193	}
1194	out_free:
1195	for (i = 0; i < sz && arrays[i]; i++)
1196	nfs_readdir_folio_array_free(folio: arrays[i]);
1197	out:
1198	if (!nfs_readdir_use_cookie(filp: desc->file))
1199	nfs_readdir_rewind_search(desc);
1200	desc->folio_index_max = -1;
1201	kfree(objp: arrays);
1202	dfprintk(DIRCACHE, "NFS: %s: returns %d\n", __func__, status);
1203	return status;
1204	}
1205
1206	static bool nfs_readdir_handle_cache_misses(struct inode *inode,
1207	struct nfs_readdir_descriptor *desc,
1208	unsigned int cache_misses,
1209	bool force_clear)
1210	{
1211	if (desc->ctx->pos == 0 || !desc->plus)
1212	return false;
1213	if (cache_misses <= NFS_READDIR_CACHE_MISS_THRESHOLD && !force_clear)
1214	return false;
1215	trace_nfs_readdir_force_readdirplus(inode);
1216	return true;
1217	}
1218
1219	/* The file offset position represents the dirent entry number. A
1220	last cookie cache takes care of the common case of reading the
1221	whole directory.
1222	*/
1223	static int nfs_readdir(struct file *file, struct dir_context *ctx)
1224	{
1225	struct dentry *dentry = file_dentry(file);
1226	struct inode *inode = d_inode(dentry);
1227	struct nfs_inode *nfsi = NFS_I(inode);
1228	struct nfs_open_dir_context *dir_ctx = file->private_data;
1229	struct nfs_readdir_descriptor *desc;
1230	unsigned int cache_hits, cache_misses;
1231	bool force_clear;
1232	int res;
1233
1234	dfprintk(FILE, "NFS: readdir(%pD2) starting at cookie %llu\n",
1235	file, (long long)ctx->pos);
1236	nfs_inc_stats(inode, stat: NFSIOS_VFSGETDENTS);
1237
1238	/*
1239	* ctx->pos points to the dirent entry number.
1240	* *desc->dir_cookie has the cookie for the next entry. We have
1241	* to either find the entry with the appropriate number or
1242	* revalidate the cookie.
1243	*/
1244	nfs_revalidate_mapping(inode, mapping: file->f_mapping);
1245
1246	res = -ENOMEM;
1247	desc = kzalloc(sizeof(*desc), GFP_KERNEL);
1248	if (!desc)
1249	goto out;
1250	desc->file = file;
1251	desc->ctx = ctx;
1252	desc->folio_index_max = -1;
1253
1254	spin_lock(lock: &file->f_lock);
1255	desc->dir_cookie = dir_ctx->dir_cookie;
1256	desc->folio_index = dir_ctx->page_index;
1257	desc->last_cookie = dir_ctx->last_cookie;
1258	desc->attr_gencount = dir_ctx->attr_gencount;
1259	desc->eof = dir_ctx->eof;
1260	nfs_set_dtsize(desc, sz: dir_ctx->dtsize);
1261	memcpy(desc->verf, dir_ctx->verf, sizeof(desc->verf));
1262	cache_hits = atomic_xchg(v: &dir_ctx->cache_hits, new: 0);
1263	cache_misses = atomic_xchg(v: &dir_ctx->cache_misses, new: 0);
1264	force_clear = dir_ctx->force_clear;
1265	spin_unlock(lock: &file->f_lock);
1266
1267	if (desc->eof) {
1268	res = 0;
1269	goto out_free;
1270	}
1271
1272	desc->plus = nfs_use_readdirplus(dir: inode, ctx, cache_hits, cache_misses);
1273	force_clear = nfs_readdir_handle_cache_misses(inode, desc, cache_misses,
1274	force_clear);
1275	desc->clear_cache = force_clear;
1276
1277	do {
1278	res = readdir_search_pagecache(desc);
1279
1280	if (res == -EBADCOOKIE) {
1281	res = 0;
1282	/* This means either end of directory */
1283	if (desc->dir_cookie && !desc->eof) {
1284	/* Or that the server has 'lost' a cookie */
1285	res = uncached_readdir(desc);
1286	if (res == 0)
1287	continue;
1288	if (res == -EBADCOOKIE || res == -ENOTSYNC)
1289	res = 0;
1290	}
1291	break;
1292	}
1293	if (res == -ETOOSMALL && desc->plus) {
1294	nfs_zap_caches(inode);
1295	desc->plus = false;
1296	desc->eof = false;
1297	continue;
1298	}
1299	if (res < 0)
1300	break;
1301
1302	nfs_do_filldir(desc, verf: nfsi->cookieverf);
1303	nfs_readdir_folio_unlock_and_put_cached(desc);
1304	if (desc->folio_index == desc->folio_index_max)
1305	desc->clear_cache = force_clear;
1306	} while (!desc->eob && !desc->eof);
1307
1308	spin_lock(lock: &file->f_lock);
1309	dir_ctx->dir_cookie = desc->dir_cookie;
1310	dir_ctx->last_cookie = desc->last_cookie;
1311	dir_ctx->attr_gencount = desc->attr_gencount;
1312	dir_ctx->page_index = desc->folio_index;
1313	dir_ctx->force_clear = force_clear;
1314	dir_ctx->eof = desc->eof;
1315	dir_ctx->dtsize = desc->dtsize;
1316	memcpy(dir_ctx->verf, desc->verf, sizeof(dir_ctx->verf));
1317	spin_unlock(lock: &file->f_lock);
1318	out_free:
1319	kfree(objp: desc);
1320
1321	out:
1322	dfprintk(FILE, "NFS: readdir(%pD2) returns %d\n", file, res);
1323	return res;
1324	}
1325
1326	static loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int whence)
1327	{
1328	struct nfs_open_dir_context *dir_ctx = filp->private_data;
1329
1330	dfprintk(FILE, "NFS: llseek dir(%pD2, %lld, %d)\n",
1331	filp, offset, whence);
1332
1333	switch (whence) {
1334	default:
1335	return -EINVAL;
1336	case SEEK_SET:
1337	if (offset < 0)
1338	return -EINVAL;
1339	spin_lock(lock: &filp->f_lock);
1340	break;
1341	case SEEK_CUR:
1342	if (offset == 0)
1343	return filp->f_pos;
1344	spin_lock(lock: &filp->f_lock);
1345	offset += filp->f_pos;
1346	if (offset < 0) {
1347	spin_unlock(lock: &filp->f_lock);
1348	return -EINVAL;
1349	}
1350	}
1351	if (offset != filp->f_pos) {
1352	filp->f_pos = offset;
1353	dir_ctx->page_index = 0;
1354	if (!nfs_readdir_use_cookie(filp)) {
1355	dir_ctx->dir_cookie = 0;
1356	dir_ctx->last_cookie = 0;
1357	} else {
1358	dir_ctx->dir_cookie = offset;
1359	dir_ctx->last_cookie = offset;
1360	}
1361	dir_ctx->eof = false;
1362	}
1363	spin_unlock(lock: &filp->f_lock);
1364	return offset;
1365	}
1366
1367	/*
1368	* All directory operations under NFS are synchronous, so fsync()
1369	* is a dummy operation.
1370	*/
1371	static int nfs_fsync_dir(struct file *filp, loff_t start, loff_t end,
1372	int datasync)
1373	{
1374	dfprintk(FILE, "NFS: fsync dir(%pD2) datasync %d\n", filp, datasync);
1375
1376	nfs_inc_stats(inode: file_inode(f: filp), stat: NFSIOS_VFSFSYNC);
1377	return 0;
1378	}
1379
1380	/**
1381	* nfs_force_lookup_revalidate - Mark the directory as having changed
1382	* @dir: pointer to directory inode
1383	*
1384	* This forces the revalidation code in nfs_lookup_revalidate() to do a
1385	* full lookup on all child dentries of 'dir' whenever a change occurs
1386	* on the server that might have invalidated our dcache.
1387	*
1388	* Note that we reserve bit '0' as a tag to let us know when a dentry
1389	* was revalidated while holding a delegation on its inode.
1390	*
1391	* The caller should be holding dir->i_lock
1392	*/
1393	void nfs_force_lookup_revalidate(struct inode *dir)
1394	{
1395	NFS_I(inode: dir)->cache_change_attribute += 2;
1396	}
1397	EXPORT_SYMBOL_GPL(nfs_force_lookup_revalidate);
1398
1399	/**
1400	* nfs_verify_change_attribute - Detects NFS remote directory changes
1401	* @dir: pointer to parent directory inode
1402	* @verf: previously saved change attribute
1403	*
1404	* Return "false" if the verifiers doesn't match the change attribute.
1405	* This would usually indicate that the directory contents have changed on
1406	* the server, and that any dentries need revalidating.
1407	*/
1408	static bool nfs_verify_change_attribute(struct inode *dir, unsigned long verf)
1409	{
1410	return (verf & ~1UL) == nfs_save_change_attribute(dir);
1411	}
1412
1413	static void nfs_set_verifier_delegated(unsigned long *verf)
1414	{
1415	*verf |= 1UL;
1416	}
1417
1418	#if IS_ENABLED(CONFIG_NFS_V4)
1419	static void nfs_unset_verifier_delegated(unsigned long *verf)
1420	{
1421	*verf &= ~1UL;
1422	}
1423	#endif /* IS_ENABLED(CONFIG_NFS_V4) */
1424
1425	static bool nfs_test_verifier_delegated(unsigned long verf)
1426	{
1427	return verf & 1;
1428	}
1429
1430	static bool nfs_verifier_is_delegated(struct dentry *dentry)
1431	{
1432	return nfs_test_verifier_delegated(verf: dentry->d_time);
1433	}
1434
1435	static void nfs_set_verifier_locked(struct dentry *dentry, unsigned long verf)
1436	{
1437	struct inode *inode = d_inode(dentry);
1438	struct inode *dir = d_inode_rcu(dentry: dentry->d_parent);
1439
1440	if (!dir || !nfs_verify_change_attribute(dir, verf))
1441	return;
1442	if (inode && NFS_PROTO(inode)->have_delegation(inode, FMODE_READ, 0))
1443	nfs_set_verifier_delegated(verf: &verf);
1444	dentry->d_time = verf;
1445	}
1446
1447	/**
1448	* nfs_set_verifier - save a parent directory verifier in the dentry
1449	* @dentry: pointer to dentry
1450	* @verf: verifier to save
1451	*
1452	* Saves the parent directory verifier in @dentry. If the inode has
1453	* a delegation, we also tag the dentry as having been revalidated
1454	* while holding a delegation so that we know we don't have to
1455	* look it up again after a directory change.
1456	*/
1457	void nfs_set_verifier(struct dentry *dentry, unsigned long verf)
1458	{
1459
1460	spin_lock(lock: &dentry->d_lock);
1461	nfs_set_verifier_locked(dentry, verf);
1462	spin_unlock(lock: &dentry->d_lock);
1463	}
1464	EXPORT_SYMBOL_GPL(nfs_set_verifier);
1465
1466	#if IS_ENABLED(CONFIG_NFS_V4)
1467	/**
1468	* nfs_clear_verifier_delegated - clear the dir verifier delegation tag
1469	* @inode: pointer to inode
1470	*
1471	* Iterates through the dentries in the inode alias list and clears
1472	* the tag used to indicate that the dentry has been revalidated
1473	* while holding a delegation.
1474	* This function is intended for use when the delegation is being
1475	* returned or revoked.
1476	*/
1477	void nfs_clear_verifier_delegated(struct inode *inode)
1478	{
1479	struct dentry *alias;
1480
1481	if (!inode)
1482	return;
1483	spin_lock(lock: &inode->i_lock);
1484	hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
1485	spin_lock(lock: &alias->d_lock);
1486	nfs_unset_verifier_delegated(verf: &alias->d_time);
1487	spin_unlock(lock: &alias->d_lock);
1488	}
1489	spin_unlock(lock: &inode->i_lock);
1490	}
1491	EXPORT_SYMBOL_GPL(nfs_clear_verifier_delegated);
1492	#endif /* IS_ENABLED(CONFIG_NFS_V4) */
1493
1494	static int nfs_dentry_verify_change(struct inode *dir, struct dentry *dentry)
1495	{
1496	if (nfs_server_capable(inode: dir, NFS_CAP_CASE_INSENSITIVE) &&
1497	d_really_is_negative(dentry))
1498	return dentry->d_time == inode_peek_iversion_raw(inode: dir);
1499	return nfs_verify_change_attribute(dir, verf: dentry->d_time);
1500	}
1501
1502	/*
1503	* A check for whether or not the parent directory has changed.
1504	* In the case it has, we assume that the dentries are untrustworthy
1505	* and may need to be looked up again.
1506	* If rcu_walk prevents us from performing a full check, return 0.
1507	*/
1508	static int nfs_check_verifier(struct inode *dir, struct dentry *dentry,
1509	int rcu_walk)
1510	{
1511	if (IS_ROOT(dentry))
1512	return 1;
1513	if (NFS_SERVER(inode: dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONE)
1514	return 0;
1515	if (!nfs_dentry_verify_change(dir, dentry))
1516	return 0;
1517	/* Revalidate nfsi->cache_change_attribute before we declare a match */
1518	if (nfs_mapping_need_revalidate_inode(inode: dir)) {
1519	if (rcu_walk)
1520	return 0;
1521	if (__nfs_revalidate_inode(NFS_SERVER(inode: dir), dir) < 0)
1522	return 0;
1523	}
1524	if (!nfs_dentry_verify_change(dir, dentry))
1525	return 0;
1526	return 1;
1527	}
1528
1529	/*
1530	* Use intent information to check whether or not we're going to do
1531	* an O_EXCL create using this path component.
1532	*/
1533	static int nfs_is_exclusive_create(struct inode *dir, unsigned int flags)
1534	{
1535	if (NFS_PROTO(inode: dir)->version == 2)
1536	return 0;
1537	return (flags & (LOOKUP_CREATE | LOOKUP_EXCL)) ==
1538	(LOOKUP_CREATE | LOOKUP_EXCL);
1539	}
1540
1541	/*
1542	* Inode and filehandle revalidation for lookups.
1543	*
1544	* We force revalidation in the cases where the VFS sets LOOKUP_REVAL,
1545	* or if the intent information indicates that we're about to open this
1546	* particular file and the "nocto" mount flag is not set.
1547	*
1548	*/
1549	static
1550	int nfs_lookup_verify_inode(struct inode *inode, unsigned int flags)
1551	{
1552	struct nfs_server *server = NFS_SERVER(inode);
1553	int ret;
1554
1555	if (IS_AUTOMOUNT(inode))
1556	return 0;
1557
1558	if (flags & LOOKUP_OPEN) {
1559	switch (inode->i_mode & S_IFMT) {
1560	case S_IFREG:
1561	/* A NFSv4 OPEN will revalidate later */
1562	if (server->caps & NFS_CAP_ATOMIC_OPEN)
1563	goto out;
1564	fallthrough;
1565	case S_IFDIR:
1566	if (server->flags & NFS_MOUNT_NOCTO)
1567	break;
1568	/* NFS close-to-open cache consistency validation */
1569	goto out_force;
1570	}
1571	}
1572
1573	/* VFS wants an on-the-wire revalidation */
1574	if (flags & LOOKUP_REVAL)
1575	goto out_force;
1576	out:
1577	if (inode->i_nlink > 0 ||
1578	(inode->i_nlink == 0 &&
1579	test_bit(NFS_INO_PRESERVE_UNLINKED, &NFS_I(inode)->flags)))
1580	return 0;
1581	else
1582	return -ESTALE;
1583	out_force:
1584	if (flags & LOOKUP_RCU)
1585	return -ECHILD;
1586	ret = __nfs_revalidate_inode(server, inode);
1587	if (ret != 0)
1588	return ret;
1589	goto out;
1590	}
1591
1592	static void nfs_mark_dir_for_revalidate(struct inode *inode)
1593	{
1594	spin_lock(lock: &inode->i_lock);
1595	nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE);
1596	spin_unlock(lock: &inode->i_lock);
1597	}
1598
1599	/*
1600	* We judge how long we want to trust negative
1601	* dentries by looking at the parent inode mtime.
1602	*
1603	* If parent mtime has changed, we revalidate, else we wait for a
1604	* period corresponding to the parent's attribute cache timeout value.
1605	*
1606	* If LOOKUP_RCU prevents us from performing a full check, return 1
1607	* suggesting a reval is needed.
1608	*
1609	* Note that when creating a new file, or looking up a rename target,
1610	* then it shouldn't be necessary to revalidate a negative dentry.
1611	*/
1612	static inline
1613	int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry,
1614	unsigned int flags)
1615	{
1616	if (flags & (LOOKUP_CREATE | LOOKUP_RENAME_TARGET))
1617	return 0;
1618	if (NFS_SERVER(inode: dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONEG)
1619	return 1;
1620	/* Case insensitive server? Revalidate negative dentries */
1621	if (nfs_server_capable(inode: dir, NFS_CAP_CASE_INSENSITIVE))
1622	return 1;
1623	return !nfs_check_verifier(dir, dentry, rcu_walk: flags & LOOKUP_RCU);
1624	}
1625
1626	static int
1627	nfs_lookup_revalidate_done(struct inode *dir, struct dentry *dentry,
1628	struct inode *inode, int error)
1629	{
1630	switch (error) {
1631	case 1:
1632	break;
1633	case -ETIMEDOUT:
1634	if (inode && (IS_ROOT(dentry) ||
1635	NFS_SERVER(inode)->flags & NFS_MOUNT_SOFTREVAL))
1636	error = 1;
1637	break;
1638	case -ESTALE:
1639	case -ENOENT:
1640	error = 0;
1641	fallthrough;
1642	default:
1643	/*
1644	* We can't d_drop the root of a disconnected tree:
1645	* its d_hash is on the s_anon list and d_drop() would hide
1646	* it from shrink_dcache_for_unmount(), leading to busy
1647	* inodes on unmount and further oopses.
1648	*/
1649	if (inode && IS_ROOT(dentry))
1650	error = 1;
1651	break;
1652	}
1653	trace_nfs_lookup_revalidate_exit(dir, dentry, flags: 0, error);
1654	return error;
1655	}
1656
1657	static int
1658	nfs_lookup_revalidate_negative(struct inode *dir, struct dentry *dentry,
1659	unsigned int flags)
1660	{
1661	int ret = 1;
1662	if (nfs_neg_need_reval(dir, dentry, flags)) {
1663	if (flags & LOOKUP_RCU)
1664	return -ECHILD;
1665	ret = 0;
1666	}
1667	return nfs_lookup_revalidate_done(dir, dentry, NULL, error: ret);
1668	}
1669
1670	static int
1671	nfs_lookup_revalidate_delegated(struct inode *dir, struct dentry *dentry,
1672	struct inode *inode)
1673	{
1674	nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1675	return nfs_lookup_revalidate_done(dir, dentry, inode, error: 1);
1676	}
1677
1678	static int nfs_lookup_revalidate_dentry(struct inode *dir, const struct qstr *name,
1679	struct dentry *dentry,
1680	struct inode *inode, unsigned int flags)
1681	{
1682	struct nfs_fh *fhandle;
1683	struct nfs_fattr *fattr;
1684	unsigned long dir_verifier;
1685	int ret;
1686
1687	trace_nfs_lookup_revalidate_enter(dir, dentry, flags);
1688
1689	ret = -ENOMEM;
1690	fhandle = nfs_alloc_fhandle();
1691	fattr = nfs_alloc_fattr_with_label(server: NFS_SERVER(inode));
1692	if (fhandle == NULL || fattr == NULL)
1693	goto out;
1694
1695	dir_verifier = nfs_save_change_attribute(dir);
1696	ret = NFS_PROTO(inode: dir)->lookup(dir, dentry, name, fhandle, fattr);
1697	if (ret < 0)
1698	goto out;
1699
1700	/* Request help from readdirplus */
1701	nfs_lookup_advise_force_readdirplus(dir, flags);
1702
1703	ret = 0;
1704	if (nfs_compare_fh(a: NFS_FH(inode), b: fhandle))
1705	goto out;
1706	if (nfs_refresh_inode(inode, fattr) < 0)
1707	goto out;
1708
1709	nfs_setsecurity(inode, fattr);
1710	nfs_set_verifier(dentry, dir_verifier);
1711
1712	ret = 1;
1713	out:
1714	nfs_free_fattr(fattr);
1715	nfs_free_fhandle(fh: fhandle);
1716
1717	/*
1718	* If the lookup failed despite the dentry change attribute being
1719	* a match, then we should revalidate the directory cache.
1720	*/
1721	if (!ret && nfs_dentry_verify_change(dir, dentry))
1722	nfs_mark_dir_for_revalidate(inode: dir);
1723	return nfs_lookup_revalidate_done(dir, dentry, inode, error: ret);
1724	}
1725
1726	/*
1727	* This is called every time the dcache has a lookup hit,
1728	* and we should check whether we can really trust that
1729	* lookup.
1730	*
1731	* NOTE! The hit can be a negative hit too, don't assume
1732	* we have an inode!
1733	*
1734	* If the parent directory is seen to have changed, we throw out the
1735	* cached dentry and do a new lookup.
1736	*/
1737	static int
1738	nfs_do_lookup_revalidate(struct inode *dir, const struct qstr *name,
1739	struct dentry *dentry, unsigned int flags)
1740	{
1741	struct inode *inode;
1742	int error = 0;
1743
1744	nfs_inc_stats(inode: dir, stat: NFSIOS_DENTRYREVALIDATE);
1745	inode = d_inode(dentry);
1746
1747	if (!inode)
1748	return nfs_lookup_revalidate_negative(dir, dentry, flags);
1749
1750	if (is_bad_inode(inode)) {
1751	dfprintk(LOOKUPCACHE, "%s: %pd2 has dud inode\n",
1752	__func__, dentry);
1753	goto out_bad;
1754	}
1755
1756	if ((flags & LOOKUP_RENAME_TARGET) && d_count(dentry) < 2 &&
1757	nfs_server_capable(inode: dir, NFS_CAP_CASE_INSENSITIVE))
1758	goto out_bad;
1759
1760	if (nfs_verifier_is_delegated(dentry))
1761	return nfs_lookup_revalidate_delegated(dir, dentry, inode);
1762
1763	/* Force a full look up iff the parent directory has changed */
1764	if (!(flags & (LOOKUP_EXCL | LOOKUP_REVAL)) &&
1765	nfs_check_verifier(dir, dentry, rcu_walk: flags & LOOKUP_RCU)) {
1766	error = nfs_lookup_verify_inode(inode, flags);
1767	if (error) {
1768	if (error == -ESTALE)
1769	nfs_mark_dir_for_revalidate(inode: dir);
1770	goto out_bad;
1771	}
1772	goto out_valid;
1773	}
1774
1775	if (flags & LOOKUP_RCU)
1776	return -ECHILD;
1777
1778	if (NFS_STALE(inode))
1779	goto out_bad;
1780
1781	return nfs_lookup_revalidate_dentry(dir, name, dentry, inode, flags);
1782	out_valid:
1783	return nfs_lookup_revalidate_done(dir, dentry, inode, error: 1);
1784	out_bad:
1785	if (flags & LOOKUP_RCU)
1786	return -ECHILD;
1787	return nfs_lookup_revalidate_done(dir, dentry, inode, error);
1788	}
1789
1790	static int
1791	__nfs_lookup_revalidate(struct dentry *dentry, unsigned int flags)
1792	{
1793	if (flags & LOOKUP_RCU) {
1794	if (dentry->d_fsdata == NFS_FSDATA_BLOCKED)
1795	return -ECHILD;
1796	} else {
1797	/* Wait for unlink to complete - see unblock_revalidate() */
1798	wait_var_event(&dentry->d_fsdata,
1799	smp_load_acquire(&dentry->d_fsdata)
1800	!= NFS_FSDATA_BLOCKED);
1801	}
1802	return 0;
1803	}
1804
1805	static int nfs_lookup_revalidate(struct inode *dir, const struct qstr *name,
1806	struct dentry *dentry, unsigned int flags)
1807	{
1808	if (__nfs_lookup_revalidate(dentry, flags))
1809	return -ECHILD;
1810	return nfs_do_lookup_revalidate(dir, name, dentry, flags);
1811	}
1812
1813	static void block_revalidate(struct dentry *dentry)
1814	{
1815	/* old devname - just in case */
1816	kfree(objp: dentry->d_fsdata);
1817
1818	/* Any new reference that could lead to an open
1819	* will take ->d_lock in lookup_open() -> d_lookup().
1820	* Holding this lock ensures we cannot race with
1821	* __nfs_lookup_revalidate() and removes and need
1822	* for further barriers.
1823	*/
1824	lockdep_assert_held(&dentry->d_lock);
1825
1826	dentry->d_fsdata = NFS_FSDATA_BLOCKED;
1827	}
1828
1829	static void unblock_revalidate(struct dentry *dentry)
1830	{
1831	/* store_release ensures wait_var_event() sees the update */
1832	smp_store_release(&dentry->d_fsdata, NULL);
1833	wake_up_var(var: &dentry->d_fsdata);
1834	}
1835
1836	/*
1837	* A weaker form of d_revalidate for revalidating just the d_inode(dentry)
1838	* when we don't really care about the dentry name. This is called when a
1839	* pathwalk ends on a dentry that was not found via a normal lookup in the
1840	* parent dir (e.g.: ".", "..", procfs symlinks or mountpoint traversals).
1841	*
1842	* In this situation, we just want to verify that the inode itself is OK
1843	* since the dentry might have changed on the server.
1844	*/
1845	static int nfs_weak_revalidate(struct dentry *dentry, unsigned int flags)
1846	{
1847	struct inode *inode = d_inode(dentry);
1848	int error = 0;
1849
1850	/*
1851	* I believe we can only get a negative dentry here in the case of a
1852	* procfs-style symlink. Just assume it's correct for now, but we may
1853	* eventually need to do something more here.
1854	*/
1855	if (!inode) {
1856	dfprintk(LOOKUPCACHE, "%s: %pd2 has negative inode\n",
1857	__func__, dentry);
1858	return 1;
1859	}
1860
1861	if (is_bad_inode(inode)) {
1862	dfprintk(LOOKUPCACHE, "%s: %pd2 has dud inode\n",
1863	__func__, dentry);
1864	return 0;
1865	}
1866
1867	error = nfs_lookup_verify_inode(inode, flags);
1868	dfprintk(LOOKUPCACHE, "NFS: %s: inode %lu is %s\n",
1869	__func__, inode->i_ino, error ? "invalid" : "valid");
1870	return !error;
1871	}
1872
1873	/*
1874	* This is called from dput() when d_count is going to 0.
1875	*/
1876	static int nfs_dentry_delete(const struct dentry *dentry)
1877	{
1878	dfprintk(VFS, "NFS: dentry_delete(%pd2, %x)\n",
1879	dentry, dentry->d_flags);
1880
1881	/* Unhash any dentry with a stale inode */
1882	if (d_really_is_positive(dentry) && NFS_STALE(inode: d_inode(dentry)))
1883	return 1;
1884
1885	if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1886	/* Unhash it, so that ->d_iput() would be called */
1887	return 1;
1888	}
1889	if (!(dentry->d_sb->s_flags & SB_ACTIVE)) {
1890	/* Unhash it, so that ancestors of killed async unlink
1891	* files will be cleaned up during umount */
1892	return 1;
1893	}
1894	return 0;
1895
1896	}
1897
1898	/* Ensure that we revalidate inode->i_nlink */
1899	static void nfs_drop_nlink(struct inode *inode)
1900	{
1901	spin_lock(lock: &inode->i_lock);
1902	/* drop the inode if we're reasonably sure this is the last link */
1903	if (inode->i_nlink > 0)
1904	drop_nlink(inode);
1905	NFS_I(inode)->attr_gencount = nfs_inc_attr_generation_counter();
1906	nfs_set_cache_invalid(
1907	inode, NFS_INO_INVALID_CHANGE | NFS_INO_INVALID_CTIME |
1908	NFS_INO_INVALID_NLINK);
1909	spin_unlock(lock: &inode->i_lock);
1910	}
1911
1912	/*
1913	* Called when the dentry loses inode.
1914	* We use it to clean up silly-renamed files.
1915	*/
1916	static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode)
1917	{
1918	if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1919	nfs_complete_unlink(dentry, inode);
1920	nfs_drop_nlink(inode);
1921	}
1922	iput(inode);
1923	}
1924
1925	static void nfs_d_release(struct dentry *dentry)
1926	{
1927	/* free cached devname value, if it survived that far */
1928	if (unlikely(dentry->d_fsdata)) {
1929	if (dentry->d_flags & DCACHE_NFSFS_RENAMED)
1930	WARN_ON(1);
1931	else
1932	kfree(objp: dentry->d_fsdata);
1933	}
1934	}
1935
1936	const struct dentry_operations nfs_dentry_operations = {
1937	.d_revalidate = nfs_lookup_revalidate,
1938	.d_weak_revalidate = nfs_weak_revalidate,
1939	.d_delete = nfs_dentry_delete,
1940	.d_iput = nfs_dentry_iput,
1941	.d_automount = nfs_d_automount,
1942	.d_release = nfs_d_release,
1943	};
1944	EXPORT_SYMBOL_GPL(nfs_dentry_operations);
1945
1946	struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
1947	{
1948	struct dentry *res;
1949	struct inode *inode = NULL;
1950	struct nfs_fh *fhandle = NULL;
1951	struct nfs_fattr *fattr = NULL;
1952	unsigned long dir_verifier;
1953	int error;
1954
1955	dfprintk(VFS, "NFS: lookup(%pd2)\n", dentry);
1956	nfs_inc_stats(inode: dir, stat: NFSIOS_VFSLOOKUP);
1957
1958	if (unlikely(dentry->d_name.len > NFS_SERVER(dir)->namelen))
1959	return ERR_PTR(error: -ENAMETOOLONG);
1960
1961	/*
1962	* If we're doing an exclusive create, optimize away the lookup
1963	* but don't hash the dentry.
1964	*/
1965	if (nfs_is_exclusive_create(dir, flags) || flags & LOOKUP_RENAME_TARGET)
1966	return NULL;
1967
1968	res = ERR_PTR(error: -ENOMEM);
1969	fhandle = nfs_alloc_fhandle();
1970	fattr = nfs_alloc_fattr_with_label(server: NFS_SERVER(inode: dir));
1971	if (fhandle == NULL || fattr == NULL)
1972	goto out;
1973
1974	dir_verifier = nfs_save_change_attribute(dir);
1975	trace_nfs_lookup_enter(dir, dentry, flags);
1976	error = NFS_PROTO(inode: dir)->lookup(dir, dentry, &dentry->d_name,
1977	fhandle, fattr);
1978	if (error == -ENOENT) {
1979	if (nfs_server_capable(inode: dir, NFS_CAP_CASE_INSENSITIVE))
1980	dir_verifier = inode_peek_iversion_raw(inode: dir);
1981	goto no_entry;
1982	}
1983	if (error < 0) {
1984	res = ERR_PTR(error);
1985	goto out;
1986	}
1987	inode = nfs_fhget(dentry->d_sb, fhandle, fattr);
1988	res = ERR_CAST(ptr: inode);
1989	if (IS_ERR(ptr: res))
1990	goto out;
1991
1992	/* Notify readdir to use READDIRPLUS */
1993	nfs_lookup_advise_force_readdirplus(dir, flags);
1994
1995	no_entry:
1996	res = d_splice_alias(inode, dentry);
1997	if (res != NULL) {
1998	if (IS_ERR(ptr: res))
1999	goto out;
2000	dentry = res;
2001	}
2002	nfs_set_verifier(dentry, dir_verifier);
2003	out:
2004	trace_nfs_lookup_exit(dir, dentry, flags, error: PTR_ERR_OR_ZERO(ptr: res));
2005	nfs_free_fattr(fattr);
2006	nfs_free_fhandle(fh: fhandle);
2007	return res;
2008	}
2009	EXPORT_SYMBOL_GPL(nfs_lookup);
2010
2011	void nfs_d_prune_case_insensitive_aliases(struct inode *inode)
2012	{
2013	/* Case insensitive server? Revalidate dentries */
2014	if (inode && nfs_server_capable(inode, NFS_CAP_CASE_INSENSITIVE))
2015	d_prune_aliases(inode);
2016	}
2017	EXPORT_SYMBOL_GPL(nfs_d_prune_case_insensitive_aliases);
2018
2019	#if IS_ENABLED(CONFIG_NFS_V4)
2020	static int nfs4_lookup_revalidate(struct inode *, const struct qstr *,
2021	struct dentry *, unsigned int);
2022
2023	const struct dentry_operations nfs4_dentry_operations = {
2024	.d_revalidate = nfs4_lookup_revalidate,
2025	.d_weak_revalidate = nfs_weak_revalidate,
2026	.d_delete = nfs_dentry_delete,
2027	.d_iput = nfs_dentry_iput,
2028	.d_automount = nfs_d_automount,
2029	.d_release = nfs_d_release,
2030	};
2031	EXPORT_SYMBOL_GPL(nfs4_dentry_operations);
2032
2033	static struct nfs_open_context *create_nfs_open_context(struct dentry *dentry, int open_flags, struct file *filp)
2034	{
2035	return alloc_nfs_open_context(dentry, f_mode: flags_to_mode(flags: open_flags), filp);
2036	}
2037
2038	static int do_open(struct inode *inode, struct file *filp)
2039	{
2040	nfs_fscache_open_file(inode, filp);
2041	return 0;
2042	}
2043
2044	static int nfs_finish_open(struct nfs_open_context *ctx,
2045	struct dentry *dentry,
2046	struct file *file, unsigned open_flags)
2047	{
2048	int err;
2049
2050	err = finish_open(file, dentry, open: do_open);
2051	if (err)
2052	goto out;
2053	if (S_ISREG(file_inode(file)->i_mode))
2054	nfs_file_set_open_context(filp: file, ctx);
2055	else
2056	err = -EOPENSTALE;
2057	out:
2058	return err;
2059	}
2060
2061	int nfs_atomic_open(struct inode *dir, struct dentry *dentry,
2062	struct file *file, unsigned open_flags,
2063	umode_t mode)
2064	{
2065	DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
2066	struct nfs_open_context *ctx;
2067	struct dentry *res;
2068	struct iattr attr = { .ia_valid = ATTR_OPEN };
2069	struct inode *inode;
2070	unsigned int lookup_flags = 0;
2071	unsigned long dir_verifier;
2072	bool switched = false;
2073	int created = 0;
2074	int err;
2075
2076	/* Expect a negative dentry */
2077	BUG_ON(d_inode(dentry));
2078
2079	dfprintk(VFS, "NFS: atomic_open(%s/%lu), %pd\n",
2080	dir->i_sb->s_id, dir->i_ino, dentry);
2081
2082	err = nfs_check_flags(open_flags);
2083	if (err)
2084	return err;
2085
2086	/* NFS only supports OPEN on regular files */
2087	if ((open_flags & O_DIRECTORY)) {
2088	if (!d_in_lookup(dentry)) {
2089	/*
2090	* Hashed negative dentry with O_DIRECTORY: dentry was
2091	* revalidated and is fine, no need to perform lookup
2092	* again
2093	*/
2094	return -ENOENT;
2095	}
2096	lookup_flags = LOOKUP_OPEN|LOOKUP_DIRECTORY;
2097	goto no_open;
2098	}
2099
2100	if (dentry->d_name.len > NFS_SERVER(inode: dir)->namelen)
2101	return -ENAMETOOLONG;
2102
2103	if (open_flags & O_CREAT) {
2104	struct nfs_server *server = NFS_SERVER(inode: dir);
2105
2106	if (!(server->attr_bitmask[2] & FATTR4_WORD2_MODE_UMASK))
2107	mode &= ~current_umask();
2108
2109	attr.ia_valid |= ATTR_MODE;
2110	attr.ia_mode = mode;
2111	}
2112	if (open_flags & O_TRUNC) {
2113	attr.ia_valid |= ATTR_SIZE;
2114	attr.ia_size = 0;
2115	}
2116
2117	if (!(open_flags & O_CREAT) && !d_in_lookup(dentry)) {
2118	d_drop(dentry);
2119	switched = true;
2120	dentry = d_alloc_parallel(dentry->d_parent,
2121	&dentry->d_name, &wq);
2122	if (IS_ERR(ptr: dentry))
2123	return PTR_ERR(ptr: dentry);
2124	if (unlikely(!d_in_lookup(dentry)))
2125	return finish_no_open(file, dentry);
2126	}
2127
2128	ctx = create_nfs_open_context(dentry, open_flags, filp: file);
2129	err = PTR_ERR(ptr: ctx);
2130	if (IS_ERR(ptr: ctx))
2131	goto out;
2132
2133	trace_nfs_atomic_open_enter(dir, ctx, flags: open_flags);
2134	inode = NFS_PROTO(inode: dir)->open_context(dir, ctx, open_flags, &attr, &created);
2135	if (created)
2136	file->f_mode |= FMODE_CREATED;
2137	if (IS_ERR(ptr: inode)) {
2138	err = PTR_ERR(ptr: inode);
2139	trace_nfs_atomic_open_exit(dir, ctx, flags: open_flags, error: err);
2140	put_nfs_open_context(ctx);
2141	d_drop(dentry);
2142	switch (err) {
2143	case -ENOENT:
2144	d_splice_alias(NULL, dentry);
2145	if (nfs_server_capable(inode: dir, NFS_CAP_CASE_INSENSITIVE))
2146	dir_verifier = inode_peek_iversion_raw(inode: dir);
2147	else
2148	dir_verifier = nfs_save_change_attribute(dir);
2149	nfs_set_verifier(dentry, dir_verifier);
2150	break;
2151	case -EISDIR:
2152	case -ENOTDIR:
2153	goto no_open;
2154	case -ELOOP:
2155	if (!(open_flags & O_NOFOLLOW))
2156	goto no_open;
2157	break;
2158	/* case -EINVAL: */
2159	default:
2160	break;
2161	}
2162	goto out;
2163	}
2164	file->f_mode |= FMODE_CAN_ODIRECT;
2165
2166	err = nfs_finish_open(ctx, dentry: ctx->dentry, file, open_flags);
2167	trace_nfs_atomic_open_exit(dir, ctx, flags: open_flags, error: err);
2168	put_nfs_open_context(ctx);
2169	out:
2170	if (unlikely(switched)) {
2171	d_lookup_done(dentry);
2172	dput(dentry);
2173	}
2174	return err;
2175
2176	no_open:
2177	res = nfs_lookup(dir, dentry, lookup_flags);
2178	if (!res) {
2179	inode = d_inode(dentry);
2180	if ((lookup_flags & LOOKUP_DIRECTORY) && inode &&
2181	!(S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)))
2182	res = ERR_PTR(error: -ENOTDIR);
2183	else if (inode && S_ISREG(inode->i_mode))
2184	res = ERR_PTR(error: -EOPENSTALE);
2185	} else if (!IS_ERR(ptr: res)) {
2186	inode = d_inode(dentry: res);
2187	if ((lookup_flags & LOOKUP_DIRECTORY) && inode &&
2188	!(S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))) {
2189	dput(res);
2190	res = ERR_PTR(error: -ENOTDIR);
2191	} else if (inode && S_ISREG(inode->i_mode)) {
2192	dput(res);
2193	res = ERR_PTR(error: -EOPENSTALE);
2194	}
2195	}
2196	if (switched) {
2197	d_lookup_done(dentry);
2198	if (!res)
2199	res = dentry;
2200	else
2201	dput(dentry);
2202	}
2203	if (IS_ERR(ptr: res))
2204	return PTR_ERR(ptr: res);
2205	return finish_no_open(file, dentry: res);
2206	}
2207	EXPORT_SYMBOL_GPL(nfs_atomic_open);
2208
2209	static int
2210	nfs4_lookup_revalidate(struct inode *dir, const struct qstr *name,
2211	struct dentry *dentry, unsigned int flags)
2212	{
2213	struct inode *inode;
2214
2215	if (__nfs_lookup_revalidate(dentry, flags))
2216	return -ECHILD;
2217
2218	trace_nfs_lookup_revalidate_enter(dir, dentry, flags);
2219
2220	if (!(flags & LOOKUP_OPEN) || (flags & LOOKUP_DIRECTORY))
2221	goto full_reval;
2222	if (d_mountpoint(dentry))
2223	goto full_reval;
2224
2225	inode = d_inode(dentry);
2226
2227	/* We can't create new files in nfs_open_revalidate(), so we
2228	* optimize away revalidation of negative dentries.
2229	*/
2230	if (inode == NULL)
2231	goto full_reval;
2232
2233	if (nfs_verifier_is_delegated(dentry))
2234	return nfs_lookup_revalidate_delegated(dir, dentry, inode);
2235
2236	/* NFS only supports OPEN on regular files */
2237	if (!S_ISREG(inode->i_mode))
2238	goto full_reval;
2239
2240	/* We cannot do exclusive creation on a positive dentry */
2241	if (flags & (LOOKUP_EXCL | LOOKUP_REVAL))
2242	goto reval_dentry;
2243
2244	/* Check if the directory changed */
2245	if (!nfs_check_verifier(dir, dentry, rcu_walk: flags & LOOKUP_RCU))
2246	goto reval_dentry;
2247
2248	/* Let f_op->open() actually open (and revalidate) the file */
2249	return 1;
2250	reval_dentry:
2251	if (flags & LOOKUP_RCU)
2252	return -ECHILD;
2253	return nfs_lookup_revalidate_dentry(dir, name, dentry, inode, flags);
2254
2255	full_reval:
2256	return nfs_do_lookup_revalidate(dir, name, dentry, flags);
2257	}
2258
2259	#endif /* CONFIG_NFSV4 */
2260
2261	int nfs_atomic_open_v23(struct inode *dir, struct dentry *dentry,
2262	struct file *file, unsigned int open_flags,
2263	umode_t mode)
2264	{
2265
2266	/* Same as look+open from lookup_open(), but with different O_TRUNC
2267	* handling.
2268	*/
2269	int error = 0;
2270
2271	if (dentry->d_name.len > NFS_SERVER(inode: dir)->namelen)
2272	return -ENAMETOOLONG;
2273
2274	if (open_flags & O_CREAT) {
2275	file->f_mode |= FMODE_CREATED;
2276	error = nfs_do_create(dir, dentry, mode, open_flags);
2277	if (error)
2278	return error;
2279	return finish_open(file, dentry, NULL);
2280	} else if (d_in_lookup(dentry)) {
2281	/* The only flags nfs_lookup considers are
2282	* LOOKUP_EXCL and LOOKUP_RENAME_TARGET, and
2283	* we want those to be zero so the lookup isn't skipped.
2284	*/
2285	struct dentry *res = nfs_lookup(dir, dentry, 0);
2286
2287	d_lookup_done(dentry);
2288	if (unlikely(res)) {
2289	if (IS_ERR(ptr: res))
2290	return PTR_ERR(ptr: res);
2291	return finish_no_open(file, dentry: res);
2292	}
2293	}
2294	return finish_no_open(file, NULL);
2295
2296	}
2297	EXPORT_SYMBOL_GPL(nfs_atomic_open_v23);
2298
2299	struct dentry *
2300	nfs_add_or_obtain(struct dentry *dentry, struct nfs_fh *fhandle,
2301	struct nfs_fattr *fattr)
2302	{
2303	struct dentry *parent = dget_parent(dentry);
2304	struct inode *dir = d_inode(dentry: parent);
2305	struct inode *inode;
2306	struct dentry *d;
2307	int error;
2308
2309	d_drop(dentry);
2310
2311	if (fhandle->size == 0) {
2312	error = NFS_PROTO(inode: dir)->lookup(dir, dentry, &dentry->d_name,
2313	fhandle, fattr);
2314	if (error)
2315	goto out_error;
2316	}
2317	nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
2318	if (!(fattr->valid & NFS_ATTR_FATTR)) {
2319	struct nfs_server *server = NFS_SB(s: dentry->d_sb);
2320	error = server->nfs_client->rpc_ops->getattr(server, fhandle,
2321	fattr, NULL);
2322	if (error < 0)
2323	goto out_error;
2324	}
2325	inode = nfs_fhget(dentry->d_sb, fhandle, fattr);
2326	d = d_splice_alias(inode, dentry);
2327	out:
2328	dput(parent);
2329	return d;
2330	out_error:
2331	d = ERR_PTR(error);
2332	goto out;
2333	}
2334	EXPORT_SYMBOL_GPL(nfs_add_or_obtain);
2335
2336	/*
2337	* Code common to create, mkdir, and mknod.
2338	*/
2339	int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle,
2340	struct nfs_fattr *fattr)
2341	{
2342	struct dentry *d;
2343
2344	d = nfs_add_or_obtain(dentry, fhandle, fattr);
2345	if (IS_ERR(ptr: d))
2346	return PTR_ERR(ptr: d);
2347
2348	/* Callers don't care */
2349	dput(d);
2350	return 0;
2351	}
2352	EXPORT_SYMBOL_GPL(nfs_instantiate);
2353
2354	/*
2355	* Following a failed create operation, we drop the dentry rather
2356	* than retain a negative dentry. This avoids a problem in the event
2357	* that the operation succeeded on the server, but an error in the
2358	* reply path made it appear to have failed.
2359	*/
2360	static int nfs_do_create(struct inode *dir, struct dentry *dentry,
2361	umode_t mode, int open_flags)
2362	{
2363	struct iattr attr;
2364	int error;
2365
2366	open_flags |= O_CREAT;
2367
2368	dfprintk(VFS, "NFS: create(%s/%lu), %pd\n",
2369	dir->i_sb->s_id, dir->i_ino, dentry);
2370
2371	attr.ia_mode = mode;
2372	attr.ia_valid = ATTR_MODE;
2373	if (open_flags & O_TRUNC) {
2374	attr.ia_size = 0;
2375	attr.ia_valid |= ATTR_SIZE;
2376	}
2377
2378	trace_nfs_create_enter(dir, dentry, flags: open_flags);
2379	error = NFS_PROTO(inode: dir)->create(dir, dentry, &attr, open_flags);
2380	trace_nfs_create_exit(dir, dentry, flags: open_flags, error);
2381	if (error != 0)
2382	goto out_err;
2383	return 0;
2384	out_err:
2385	d_drop(dentry);
2386	return error;
2387	}
2388
2389	int nfs_create(struct mnt_idmap *idmap, struct inode *dir,
2390	struct dentry *dentry, umode_t mode, bool excl)
2391	{
2392	return nfs_do_create(dir, dentry, mode, open_flags: excl ? O_EXCL : 0);
2393	}
2394	EXPORT_SYMBOL_GPL(nfs_create);
2395
2396	/*
2397	* See comments for nfs_proc_create regarding failed operations.
2398	*/
2399	int
2400	nfs_mknod(struct mnt_idmap *idmap, struct inode *dir,
2401	struct dentry *dentry, umode_t mode, dev_t rdev)
2402	{
2403	struct iattr attr;
2404	int status;
2405
2406	dfprintk(VFS, "NFS: mknod(%s/%lu), %pd\n",
2407	dir->i_sb->s_id, dir->i_ino, dentry);
2408
2409	attr.ia_mode = mode;
2410	attr.ia_valid = ATTR_MODE;
2411
2412	trace_nfs_mknod_enter(dir, dentry);
2413	status = NFS_PROTO(inode: dir)->mknod(dir, dentry, &attr, rdev);
2414	trace_nfs_mknod_exit(dir, dentry, error: status);
2415	if (status != 0)
2416	goto out_err;
2417	return 0;
2418	out_err:
2419	d_drop(dentry);
2420	return status;
2421	}
2422	EXPORT_SYMBOL_GPL(nfs_mknod);
2423
2424	/*
2425	* See comments for nfs_proc_create regarding failed operations.
2426	*/
2427	struct dentry *nfs_mkdir(struct mnt_idmap *idmap, struct inode *dir,
2428	struct dentry *dentry, umode_t mode)
2429	{
2430	struct iattr attr;
2431	struct dentry *ret;
2432
2433	dfprintk(VFS, "NFS: mkdir(%s/%lu), %pd\n",
2434	dir->i_sb->s_id, dir->i_ino, dentry);
2435
2436	attr.ia_valid = ATTR_MODE;
2437	attr.ia_mode = mode | S_IFDIR;
2438
2439	trace_nfs_mkdir_enter(dir, dentry);
2440	ret = NFS_PROTO(inode: dir)->mkdir(dir, dentry, &attr);
2441	trace_nfs_mkdir_exit(dir, dentry, error: PTR_ERR_OR_ZERO(ptr: ret));
2442	return ret;
2443	}
2444	EXPORT_SYMBOL_GPL(nfs_mkdir);
2445
2446	static void nfs_dentry_handle_enoent(struct dentry *dentry)
2447	{
2448	if (simple_positive(dentry))
2449	d_delete(dentry);
2450	}
2451
2452	static void nfs_dentry_remove_handle_error(struct inode *dir,
2453	struct dentry *dentry, int error)
2454	{
2455	switch (error) {
2456	case -ENOENT:
2457	if (d_really_is_positive(dentry))
2458	d_delete(dentry);
2459	nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
2460	break;
2461	case 0:
2462	nfs_d_prune_case_insensitive_aliases(d_inode(dentry));
2463	nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
2464	}
2465	}
2466
2467	int nfs_rmdir(struct inode *dir, struct dentry *dentry)
2468	{
2469	int error;
2470
2471	dfprintk(VFS, "NFS: rmdir(%s/%lu), %pd\n",
2472	dir->i_sb->s_id, dir->i_ino, dentry);
2473
2474	trace_nfs_rmdir_enter(dir, dentry);
2475	if (d_really_is_positive(dentry)) {
2476	down_write(sem: &NFS_I(inode: d_inode(dentry))->rmdir_sem);
2477	error = NFS_PROTO(inode: dir)->rmdir(dir, &dentry->d_name);
2478	/* Ensure the VFS deletes this inode */
2479	switch (error) {
2480	case 0:
2481	clear_nlink(inode: d_inode(dentry));
2482	break;
2483	case -ENOENT:
2484	nfs_dentry_handle_enoent(dentry);
2485	}
2486	up_write(sem: &NFS_I(inode: d_inode(dentry))->rmdir_sem);
2487	} else
2488	error = NFS_PROTO(inode: dir)->rmdir(dir, &dentry->d_name);
2489	nfs_dentry_remove_handle_error(dir, dentry, error);
2490	trace_nfs_rmdir_exit(dir, dentry, error);
2491
2492	return error;
2493	}
2494	EXPORT_SYMBOL_GPL(nfs_rmdir);
2495
2496	/*
2497	* Remove a file after making sure there are no pending writes,
2498	* and after checking that the file has only one user.
2499	*
2500	* We invalidate the attribute cache and free the inode prior to the operation
2501	* to avoid possible races if the server reuses the inode.
2502	*/
2503	static int nfs_safe_remove(struct dentry *dentry)
2504	{
2505	struct inode *dir = d_inode(dentry: dentry->d_parent);
2506	struct inode *inode = d_inode(dentry);
2507	int error = -EBUSY;
2508
2509	dfprintk(VFS, "NFS: safe_remove(%pd2)\n", dentry);
2510
2511	/* If the dentry was sillyrenamed, we simply call d_delete() */
2512	if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
2513	error = 0;
2514	goto out;
2515	}
2516
2517	trace_nfs_remove_enter(dir, dentry);
2518	if (inode != NULL) {
2519	error = NFS_PROTO(inode: dir)->remove(dir, dentry);
2520	if (error == 0)
2521	nfs_drop_nlink(inode);
2522	} else
2523	error = NFS_PROTO(inode: dir)->remove(dir, dentry);
2524	if (error == -ENOENT)
2525	nfs_dentry_handle_enoent(dentry);
2526	trace_nfs_remove_exit(dir, dentry, error);
2527	out:
2528	return error;
2529	}
2530
2531	/* We do silly rename. In case sillyrename() returns -EBUSY, the inode
2532	* belongs to an active ".nfs..." file and we return -EBUSY.
2533	*
2534	* If sillyrename() returns 0, we do nothing, otherwise we unlink.
2535	*/
2536	int nfs_unlink(struct inode *dir, struct dentry *dentry)
2537	{
2538	int error;
2539
2540	dfprintk(VFS, "NFS: unlink(%s/%lu, %pd)\n", dir->i_sb->s_id,
2541	dir->i_ino, dentry);
2542
2543	trace_nfs_unlink_enter(dir, dentry);
2544	spin_lock(lock: &dentry->d_lock);
2545	if (d_count(dentry) > 1 && !test_bit(NFS_INO_PRESERVE_UNLINKED,
2546	&NFS_I(d_inode(dentry))->flags)) {
2547	spin_unlock(lock: &dentry->d_lock);
2548	/* Start asynchronous writeout of the inode */
2549	write_inode_now(d_inode(dentry), sync: 0);
2550	error = nfs_sillyrename(dir, dentry);
2551	goto out;
2552	}
2553	/* We must prevent any concurrent open until the unlink
2554	* completes. ->d_revalidate will wait for ->d_fsdata
2555	* to clear. We set it here to ensure no lookup succeeds until
2556	* the unlink is complete on the server.
2557	*/
2558	error = -ETXTBSY;
2559	if (WARN_ON(dentry->d_flags & DCACHE_NFSFS_RENAMED) ||
2560	WARN_ON(dentry->d_fsdata == NFS_FSDATA_BLOCKED)) {
2561	spin_unlock(lock: &dentry->d_lock);
2562	goto out;
2563	}
2564	block_revalidate(dentry);
2565
2566	spin_unlock(lock: &dentry->d_lock);
2567	error = nfs_safe_remove(dentry);
2568	nfs_dentry_remove_handle_error(dir, dentry, error);
2569	unblock_revalidate(dentry);
2570	out:
2571	trace_nfs_unlink_exit(dir, dentry, error);
2572	return error;
2573	}
2574	EXPORT_SYMBOL_GPL(nfs_unlink);
2575
2576	/*
2577	* To create a symbolic link, most file systems instantiate a new inode,
2578	* add a page to it containing the path, then write it out to the disk
2579	* using prepare_write/commit_write.
2580	*
2581	* Unfortunately the NFS client can't create the in-core inode first
2582	* because it needs a file handle to create an in-core inode (see
2583	* fs/nfs/inode.c:nfs_fhget). We only have a file handle *after* the
2584	* symlink request has completed on the server.
2585	*
2586	* So instead we allocate a raw page, copy the symname into it, then do
2587	* the SYMLINK request with the page as the buffer. If it succeeds, we
2588	* now have a new file handle and can instantiate an in-core NFS inode
2589	* and move the raw page into its mapping.
2590	*/
2591	int nfs_symlink(struct mnt_idmap *idmap, struct inode *dir,
2592	struct dentry *dentry, const char *symname)
2593	{
2594	struct folio *folio;
2595	char *kaddr;
2596	struct iattr attr;
2597	unsigned int pathlen = strlen(symname);
2598	int error;
2599
2600	dfprintk(VFS, "NFS: symlink(%s/%lu, %pd, %s)\n", dir->i_sb->s_id,
2601	dir->i_ino, dentry, symname);
2602
2603	if (pathlen > PAGE_SIZE)
2604	return -ENAMETOOLONG;
2605
2606	attr.ia_mode = S_IFLNK | S_IRWXUGO;
2607	attr.ia_valid = ATTR_MODE;
2608
2609	folio = folio_alloc(GFP_USER, 0);
2610	if (!folio)
2611	return -ENOMEM;
2612
2613	kaddr = folio_address(folio);
2614	memcpy(kaddr, symname, pathlen);
2615	if (pathlen < PAGE_SIZE)
2616	memset(kaddr + pathlen, 0, PAGE_SIZE - pathlen);
2617
2618	trace_nfs_symlink_enter(dir, dentry);
2619	error = NFS_PROTO(inode: dir)->symlink(dir, dentry, folio, pathlen, &attr);
2620	trace_nfs_symlink_exit(dir, dentry, error);
2621	if (error != 0) {
2622	dfprintk(VFS, "NFS: symlink(%s/%lu, %pd, %s) error %d\n",
2623	dir->i_sb->s_id, dir->i_ino,
2624	dentry, symname, error);
2625	d_drop(dentry);
2626	folio_put(folio);
2627	return error;
2628	}
2629
2630	nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
2631
2632	/*
2633	* No big deal if we can't add this page to the page cache here.
2634	* READLINK will get the missing page from the server if needed.
2635	*/
2636	if (filemap_add_folio(mapping: d_inode(dentry)->i_mapping, folio, index: 0,
2637	GFP_KERNEL) == 0) {
2638	folio_mark_uptodate(folio);
2639	folio_unlock(folio);
2640	}
2641
2642	folio_put(folio);
2643	return 0;
2644	}
2645	EXPORT_SYMBOL_GPL(nfs_symlink);
2646
2647	int
2648	nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
2649	{
2650	struct inode *inode = d_inode(dentry: old_dentry);
2651	int error;
2652
2653	dfprintk(VFS, "NFS: link(%pd2 -> %pd2)\n",
2654	old_dentry, dentry);
2655
2656	trace_nfs_link_enter(inode, dir, dentry);
2657	d_drop(dentry);
2658	if (S_ISREG(inode->i_mode))
2659	nfs_sync_inode(inode);
2660	error = NFS_PROTO(inode: dir)->link(inode, dir, &dentry->d_name);
2661	if (error == 0) {
2662	nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
2663	ihold(inode);
2664	d_add(dentry, inode);
2665	}
2666	trace_nfs_link_exit(inode, dir, dentry, error);
2667	return error;
2668	}
2669	EXPORT_SYMBOL_GPL(nfs_link);
2670
2671	static void
2672	nfs_unblock_rename(struct rpc_task *task, struct nfs_renamedata *data)
2673	{
2674	struct dentry *new_dentry = data->new_dentry;
2675
2676	unblock_revalidate(dentry: new_dentry);
2677	}
2678
2679	static bool nfs_rename_is_unsafe_cross_dir(struct dentry *old_dentry,
2680	struct dentry *new_dentry)
2681	{
2682	struct nfs_server *server = NFS_SB(s: old_dentry->d_sb);
2683
2684	if (old_dentry->d_parent != new_dentry->d_parent)
2685	return false;
2686	if (server->fh_expire_type & NFS_FH_RENAME_UNSAFE)
2687	return !(server->fh_expire_type & NFS_FH_NOEXPIRE_WITH_OPEN);
2688	return true;
2689	}
2690
2691	/*
2692	* RENAME
2693	* FIXME: Some nfsds, like the Linux user space nfsd, may generate a
2694	* different file handle for the same inode after a rename (e.g. when
2695	* moving to a different directory). A fail-safe method to do so would
2696	* be to look up old_dir/old_name, create a link to new_dir/new_name and
2697	* rename the old file using the sillyrename stuff. This way, the original
2698	* file in old_dir will go away when the last process iput()s the inode.
2699	*
2700	* FIXED.
2701	*
2702	* It actually works quite well. One needs to have the possibility for
2703	* at least one ".nfs..." file in each directory the file ever gets
2704	* moved or linked to which happens automagically with the new
2705	* implementation that only depends on the dcache stuff instead of
2706	* using the inode layer
2707	*
2708	* Unfortunately, things are a little more complicated than indicated
2709	* above. For a cross-directory move, we want to make sure we can get
2710	* rid of the old inode after the operation. This means there must be
2711	* no pending writes (if it's a file), and the use count must be 1.
2712	* If these conditions are met, we can drop the dentries before doing
2713	* the rename.
2714	*/
2715	int nfs_rename(struct mnt_idmap *idmap, struct inode *old_dir,
2716	struct dentry *old_dentry, struct inode *new_dir,
2717	struct dentry *new_dentry, unsigned int flags)
2718	{
2719	struct inode *old_inode = d_inode(dentry: old_dentry);
2720	struct inode *new_inode = d_inode(dentry: new_dentry);
2721	struct dentry *dentry = NULL;
2722	struct rpc_task *task;
2723	bool must_unblock = false;
2724	int error = -EBUSY;
2725
2726	if (flags)
2727	return -EINVAL;
2728
2729	dfprintk(VFS, "NFS: rename(%pd2 -> %pd2, ct=%d)\n",
2730	old_dentry, new_dentry,
2731	d_count(new_dentry));
2732
2733	trace_nfs_rename_enter(old_dir, old_dentry, new_dir, new_dentry);
2734	/*
2735	* For non-directories, check whether the target is busy and if so,
2736	* make a copy of the dentry and then do a silly-rename. If the
2737	* silly-rename succeeds, the copied dentry is hashed and becomes
2738	* the new target.
2739	*/
2740	if (new_inode && !S_ISDIR(new_inode->i_mode)) {
2741	/* We must prevent any concurrent open until the unlink
2742	* completes. ->d_revalidate will wait for ->d_fsdata
2743	* to clear. We set it here to ensure no lookup succeeds until
2744	* the unlink is complete on the server.
2745	*/
2746	error = -ETXTBSY;
2747	if (WARN_ON(new_dentry->d_flags & DCACHE_NFSFS_RENAMED) ||
2748	WARN_ON(new_dentry->d_fsdata == NFS_FSDATA_BLOCKED))
2749	goto out;
2750
2751	spin_lock(lock: &new_dentry->d_lock);
2752	if (d_count(dentry: new_dentry) > 2) {
2753	int err;
2754
2755	spin_unlock(lock: &new_dentry->d_lock);
2756
2757	/* copy the target dentry's name */
2758	dentry = d_alloc(new_dentry->d_parent,
2759	&new_dentry->d_name);
2760	if (!dentry)
2761	goto out;
2762
2763	/* silly-rename the existing target ... */
2764	err = nfs_sillyrename(dir: new_dir, dentry: new_dentry);
2765	if (err)
2766	goto out;
2767
2768	new_dentry = dentry;
2769	new_inode = NULL;
2770	} else {
2771	block_revalidate(dentry: new_dentry);
2772	must_unblock = true;
2773	spin_unlock(lock: &new_dentry->d_lock);
2774	}
2775
2776	}
2777
2778	if (S_ISREG(old_inode->i_mode) &&
2779	nfs_rename_is_unsafe_cross_dir(old_dentry, new_dentry))
2780	nfs_sync_inode(inode: old_inode);
2781	task = nfs_async_rename(old_dir, new_dir, old_dentry, new_dentry,
2782	complete: must_unblock ? nfs_unblock_rename : NULL);
2783	if (IS_ERR(ptr: task)) {
2784	if (must_unblock)
2785	unblock_revalidate(dentry: new_dentry);
2786	error = PTR_ERR(ptr: task);
2787	goto out;
2788	}
2789
2790	error = rpc_wait_for_completion_task(task);
2791	if (error != 0) {
2792	((struct nfs_renamedata *)task->tk_calldata)->cancelled = 1;
2793	/* Paired with the atomic_dec_and_test() barrier in rpc_do_put_task() */
2794	smp_wmb();
2795	} else
2796	error = task->tk_status;
2797	rpc_put_task(task);
2798	/* Ensure the inode attributes are revalidated */
2799	if (error == 0) {
2800	spin_lock(lock: &old_inode->i_lock);
2801	NFS_I(inode: old_inode)->attr_gencount = nfs_inc_attr_generation_counter();
2802	nfs_set_cache_invalid(inode: old_inode, NFS_INO_INVALID_CHANGE |
2803	NFS_INO_INVALID_CTIME |
2804	NFS_INO_REVAL_FORCED);
2805	spin_unlock(lock: &old_inode->i_lock);
2806	}
2807	out:
2808	trace_nfs_rename_exit(old_dir, old_dentry,
2809	new_dir, new_dentry, error);
2810	if (!error) {
2811	if (new_inode != NULL)
2812	nfs_drop_nlink(inode: new_inode);
2813	/*
2814	* The d_move() should be here instead of in an async RPC completion
2815	* handler because we need the proper locks to move the dentry. If
2816	* we're interrupted by a signal, the async RPC completion handler
2817	* should mark the directories for revalidation.
2818	*/
2819	d_move(old_dentry, new_dentry);
2820	nfs_set_verifier(old_dentry,
2821	nfs_save_change_attribute(dir: new_dir));
2822	} else if (error == -ENOENT)
2823	nfs_dentry_handle_enoent(dentry: old_dentry);
2824
2825	/* new dentry created? */
2826	if (dentry)
2827	dput(dentry);
2828	return error;
2829	}
2830	EXPORT_SYMBOL_GPL(nfs_rename);
2831
2832	static DEFINE_SPINLOCK(nfs_access_lru_lock);
2833	static LIST_HEAD(nfs_access_lru_list);
2834	static atomic_long_t nfs_access_nr_entries;
2835
2836	static unsigned long nfs_access_max_cachesize = 4*1024*1024;
2837	module_param(nfs_access_max_cachesize, ulong, 0644);
2838	MODULE_PARM_DESC(nfs_access_max_cachesize, "NFS access maximum total cache length");
2839
2840	static void nfs_access_free_entry(struct nfs_access_entry *entry)
2841	{
2842	put_group_info(entry->group_info);
2843	kfree_rcu(entry, rcu_head);
2844	smp_mb__before_atomic();
2845	atomic_long_dec(v: &nfs_access_nr_entries);
2846	smp_mb__after_atomic();
2847	}
2848
2849	static void nfs_access_free_list(struct list_head *head)
2850	{
2851	struct nfs_access_entry *cache;
2852
2853	while (!list_empty(head)) {
2854	cache = list_entry(head->next, struct nfs_access_entry, lru);
2855	list_del(entry: &cache->lru);
2856	nfs_access_free_entry(entry: cache);
2857	}
2858	}
2859
2860	static unsigned long
2861	nfs_do_access_cache_scan(unsigned int nr_to_scan)
2862	{
2863	LIST_HEAD(head);
2864	struct nfs_inode *nfsi, *next;
2865	struct nfs_access_entry *cache;
2866	long freed = 0;
2867
2868	spin_lock(lock: &nfs_access_lru_lock);
2869	list_for_each_entry_safe(nfsi, next, &nfs_access_lru_list, access_cache_inode_lru) {
2870	struct inode *inode;
2871
2872	if (nr_to_scan-- == 0)
2873	break;
2874	inode = &nfsi->vfs_inode;
2875	spin_lock(lock: &inode->i_lock);
2876	if (list_empty(head: &nfsi->access_cache_entry_lru))
2877	goto remove_lru_entry;
2878	cache = list_entry(nfsi->access_cache_entry_lru.next,
2879	struct nfs_access_entry, lru);
2880	list_move(list: &cache->lru, head: &head);
2881	rb_erase(&cache->rb_node, &nfsi->access_cache);
2882	freed++;
2883	if (!list_empty(head: &nfsi->access_cache_entry_lru))
2884	list_move_tail(list: &nfsi->access_cache_inode_lru,
2885	head: &nfs_access_lru_list);
2886	else {
2887	remove_lru_entry:
2888	list_del_init(entry: &nfsi->access_cache_inode_lru);
2889	smp_mb__before_atomic();
2890	clear_bit(NFS_INO_ACL_LRU_SET, addr: &nfsi->flags);
2891	smp_mb__after_atomic();
2892	}
2893	spin_unlock(lock: &inode->i_lock);
2894	}
2895	spin_unlock(lock: &nfs_access_lru_lock);
2896	nfs_access_free_list(head: &head);
2897	return freed;
2898	}
2899
2900	unsigned long
2901	nfs_access_cache_scan(struct shrinker *shrink, struct shrink_control *sc)
2902	{
2903	int nr_to_scan = sc->nr_to_scan;
2904	gfp_t gfp_mask = sc->gfp_mask;
2905
2906	if ((gfp_mask & GFP_KERNEL) != GFP_KERNEL)
2907	return SHRINK_STOP;
2908	return nfs_do_access_cache_scan(nr_to_scan);
2909	}
2910
2911
2912	unsigned long
2913	nfs_access_cache_count(struct shrinker *shrink, struct shrink_control *sc)
2914	{
2915	return vfs_pressure_ratio(val: atomic_long_read(v: &nfs_access_nr_entries));
2916	}
2917
2918	static void
2919	nfs_access_cache_enforce_limit(void)
2920	{
2921	long nr_entries = atomic_long_read(v: &nfs_access_nr_entries);
2922	unsigned long diff;
2923	unsigned int nr_to_scan;
2924
2925	if (nr_entries < 0 || nr_entries <= nfs_access_max_cachesize)
2926	return;
2927	nr_to_scan = 100;
2928	diff = nr_entries - nfs_access_max_cachesize;
2929	if (diff < nr_to_scan)
2930	nr_to_scan = diff;
2931	nfs_do_access_cache_scan(nr_to_scan);
2932	}
2933
2934	static void __nfs_access_zap_cache(struct nfs_inode *nfsi, struct list_head *head)
2935	{
2936	struct rb_root *root_node = &nfsi->access_cache;
2937	struct rb_node *n;
2938	struct nfs_access_entry *entry;
2939
2940	/* Unhook entries from the cache */
2941	while ((n = rb_first(root_node)) != NULL) {
2942	entry = rb_entry(n, struct nfs_access_entry, rb_node);
2943	rb_erase(n, root_node);
2944	list_move(list: &entry->lru, head);
2945	}
2946	nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS;
2947	}
2948
2949	void nfs_access_zap_cache(struct inode *inode)
2950	{
2951	LIST_HEAD(head);
2952
2953	if (test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags) == 0)
2954	return;
2955	/* Remove from global LRU init */
2956	spin_lock(lock: &nfs_access_lru_lock);
2957	if (test_and_clear_bit(NFS_INO_ACL_LRU_SET, addr: &NFS_I(inode)->flags))
2958	list_del_init(entry: &NFS_I(inode)->access_cache_inode_lru);
2959
2960	spin_lock(lock: &inode->i_lock);
2961	__nfs_access_zap_cache(nfsi: NFS_I(inode), head: &head);
2962	spin_unlock(lock: &inode->i_lock);
2963	spin_unlock(lock: &nfs_access_lru_lock);
2964	nfs_access_free_list(head: &head);
2965	}
2966	EXPORT_SYMBOL_GPL(nfs_access_zap_cache);
2967
2968	static int access_cmp(const struct cred *a, const struct nfs_access_entry *b)
2969	{
2970	struct group_info *ga, *gb;
2971	int g;
2972
2973	if (uid_lt(left: a->fsuid, right: b->fsuid))
2974	return -1;
2975	if (uid_gt(left: a->fsuid, right: b->fsuid))
2976	return 1;
2977
2978	if (gid_lt(left: a->fsgid, right: b->fsgid))
2979	return -1;
2980	if (gid_gt(left: a->fsgid, right: b->fsgid))
2981	return 1;
2982
2983	ga = a->group_info;
2984	gb = b->group_info;
2985	if (ga == gb)
2986	return 0;
2987	if (ga == NULL)
2988	return -1;
2989	if (gb == NULL)
2990	return 1;
2991	if (ga->ngroups < gb->ngroups)
2992	return -1;
2993	if (ga->ngroups > gb->ngroups)
2994	return 1;
2995
2996	for (g = 0; g < ga->ngroups; g++) {
2997	if (gid_lt(left: ga->gid[g], right: gb->gid[g]))
2998	return -1;
2999	if (gid_gt(left: ga->gid[g], right: gb->gid[g]))
3000	return 1;
3001	}
3002	return 0;
3003	}
3004
3005	static struct nfs_access_entry *nfs_access_search_rbtree(struct inode *inode, const struct cred *cred)
3006	{
3007	struct rb_node *n = NFS_I(inode)->access_cache.rb_node;
3008
3009	while (n != NULL) {
3010	struct nfs_access_entry *entry =
3011	rb_entry(n, struct nfs_access_entry, rb_node);
3012	int cmp = access_cmp(a: cred, b: entry);
3013
3014	if (cmp < 0)
3015	n = n->rb_left;
3016	else if (cmp > 0)
3017	n = n->rb_right;
3018	else
3019	return entry;
3020	}
3021	return NULL;
3022	}
3023
3024	static u64 nfs_access_login_time(const struct task_struct *task,
3025	const struct cred *cred)
3026	{
3027	const struct task_struct *parent;
3028	const struct cred *pcred;
3029	u64 ret;
3030
3031	rcu_read_lock();
3032	for (;;) {
3033	parent = rcu_dereference(task->real_parent);
3034	pcred = __task_cred(parent);
3035	if (parent == task || cred_fscmp(pcred, cred) != 0)
3036	break;
3037	task = parent;
3038	}
3039	ret = task->start_time;
3040	rcu_read_unlock();
3041	return ret;
3042	}
3043
3044	static int nfs_access_get_cached_locked(struct inode *inode, const struct cred *cred, u32 *mask, bool may_block)
3045	{
3046	struct nfs_inode *nfsi = NFS_I(inode);
3047	u64 login_time = nfs_access_login_time(current, cred);
3048	struct nfs_access_entry *cache;
3049	bool retry = true;
3050	int err;
3051
3052	spin_lock(lock: &inode->i_lock);
3053	for(;;) {
3054	if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS)
3055	goto out_zap;
3056	cache = nfs_access_search_rbtree(inode, cred);
3057	err = -ENOENT;
3058	if (cache == NULL)
3059	goto out;
3060	/* Found an entry, is our attribute cache valid? */
3061	if (!nfs_check_cache_invalid(inode, NFS_INO_INVALID_ACCESS))
3062	break;
3063	if (!retry)
3064	break;
3065	err = -ECHILD;
3066	if (!may_block)
3067	goto out;
3068	spin_unlock(lock: &inode->i_lock);
3069	err = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
3070	if (err)
3071	return err;
3072	spin_lock(lock: &inode->i_lock);
3073	retry = false;
3074	}
3075	err = -ENOENT;
3076	if ((s64)(login_time - cache->timestamp) > 0)
3077	goto out;
3078	*mask = cache->mask;
3079	list_move_tail(list: &cache->lru, head: &nfsi->access_cache_entry_lru);
3080	err = 0;
3081	out:
3082	spin_unlock(lock: &inode->i_lock);
3083	return err;
3084	out_zap:
3085	spin_unlock(lock: &inode->i_lock);
3086	nfs_access_zap_cache(inode);
3087	return -ENOENT;
3088	}
3089
3090	static int nfs_access_get_cached_rcu(struct inode *inode, const struct cred *cred, u32 *mask)
3091	{
3092	/* Only check the most recently returned cache entry,
3093	* but do it without locking.
3094	*/
3095	struct nfs_inode *nfsi = NFS_I(inode);
3096	u64 login_time = nfs_access_login_time(current, cred);
3097	struct nfs_access_entry *cache;
3098	int err = -ECHILD;
3099	struct list_head *lh;
3100
3101	rcu_read_lock();
3102	if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS)
3103	goto out;
3104	lh = rcu_dereference(list_tail_rcu(&nfsi->access_cache_entry_lru));
3105	cache = list_entry(lh, struct nfs_access_entry, lru);
3106	if (lh == &nfsi->access_cache_entry_lru ||
3107	access_cmp(a: cred, b: cache) != 0)
3108	cache = NULL;
3109	if (cache == NULL)
3110	goto out;
3111	if ((s64)(login_time - cache->timestamp) > 0)
3112	goto out;
3113	if (nfs_check_cache_invalid(inode, NFS_INO_INVALID_ACCESS))
3114	goto out;
3115	*mask = cache->mask;
3116	err = 0;
3117	out:
3118	rcu_read_unlock();
3119	return err;
3120	}
3121
3122	int nfs_access_get_cached(struct inode *inode, const struct cred *cred,
3123	u32 *mask, bool may_block)
3124	{
3125	int status;
3126
3127	status = nfs_access_get_cached_rcu(inode, cred, mask);
3128	if (status != 0)
3129	status = nfs_access_get_cached_locked(inode, cred, mask,
3130	may_block);
3131
3132	return status;
3133	}
3134	EXPORT_SYMBOL_GPL(nfs_access_get_cached);
3135
3136	static void nfs_access_add_rbtree(struct inode *inode,
3137	struct nfs_access_entry *set,
3138	const struct cred *cred)
3139	{
3140	struct nfs_inode *nfsi = NFS_I(inode);
3141	struct rb_root *root_node = &nfsi->access_cache;
3142	struct rb_node **p = &root_node->rb_node;
3143	struct rb_node *parent = NULL;
3144	struct nfs_access_entry *entry;
3145	int cmp;
3146
3147	spin_lock(lock: &inode->i_lock);
3148	while (*p != NULL) {
3149	parent = *p;
3150	entry = rb_entry(parent, struct nfs_access_entry, rb_node);
3151	cmp = access_cmp(a: cred, b: entry);
3152
3153	if (cmp < 0)
3154	p = &parent->rb_left;
3155	else if (cmp > 0)
3156	p = &parent->rb_right;
3157	else
3158	goto found;
3159	}
3160	rb_link_node(node: &set->rb_node, parent, rb_link: p);
3161	rb_insert_color(&set->rb_node, root_node);
3162	list_add_tail(new: &set->lru, head: &nfsi->access_cache_entry_lru);
3163	spin_unlock(lock: &inode->i_lock);
3164	return;
3165	found:
3166	rb_replace_node(victim: parent, new: &set->rb_node, root: root_node);
3167	list_add_tail(new: &set->lru, head: &nfsi->access_cache_entry_lru);
3168	list_del(entry: &entry->lru);
3169	spin_unlock(lock: &inode->i_lock);
3170	nfs_access_free_entry(entry);
3171	}
3172
3173	void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set,
3174	const struct cred *cred)
3175	{
3176	struct nfs_access_entry *cache = kmalloc(sizeof(*cache), GFP_KERNEL);
3177	if (cache == NULL)
3178	return;
3179	RB_CLEAR_NODE(&cache->rb_node);
3180	cache->fsuid = cred->fsuid;
3181	cache->fsgid = cred->fsgid;
3182	cache->group_info = get_group_info(gi: cred->group_info);
3183	cache->mask = set->mask;
3184	cache->timestamp = ktime_get_ns();
3185
3186	/* The above field assignments must be visible
3187	* before this item appears on the lru. We cannot easily
3188	* use rcu_assign_pointer, so just force the memory barrier.
3189	*/
3190	smp_wmb();
3191	nfs_access_add_rbtree(inode, set: cache, cred);
3192
3193	/* Update accounting */
3194	smp_mb__before_atomic();
3195	atomic_long_inc(v: &nfs_access_nr_entries);
3196	smp_mb__after_atomic();
3197
3198	/* Add inode to global LRU list */
3199	if (!test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) {
3200	spin_lock(lock: &nfs_access_lru_lock);
3201	if (!test_and_set_bit(NFS_INO_ACL_LRU_SET, addr: &NFS_I(inode)->flags))
3202	list_add_tail(new: &NFS_I(inode)->access_cache_inode_lru,
3203	head: &nfs_access_lru_list);
3204	spin_unlock(lock: &nfs_access_lru_lock);
3205	}
3206	nfs_access_cache_enforce_limit();
3207	}
3208	EXPORT_SYMBOL_GPL(nfs_access_add_cache);
3209
3210	#define NFS_MAY_READ (NFS_ACCESS_READ)
3211	#define NFS_MAY_WRITE (NFS_ACCESS_MODIFY | \
3212	NFS_ACCESS_EXTEND | \
3213	NFS_ACCESS_DELETE)
3214	#define NFS_FILE_MAY_WRITE (NFS_ACCESS_MODIFY | \
3215	NFS_ACCESS_EXTEND)
3216	#define NFS_DIR_MAY_WRITE NFS_MAY_WRITE
3217	#define NFS_MAY_LOOKUP (NFS_ACCESS_LOOKUP)
3218	#define NFS_MAY_EXECUTE (NFS_ACCESS_EXECUTE)
3219	static int
3220	nfs_access_calc_mask(u32 access_result, umode_t umode)
3221	{
3222	int mask = 0;
3223
3224	if (access_result & NFS_MAY_READ)
3225	mask |= MAY_READ;
3226	if (S_ISDIR(umode)) {
3227	if ((access_result & NFS_DIR_MAY_WRITE) == NFS_DIR_MAY_WRITE)
3228	mask |= MAY_WRITE;
3229	if ((access_result & NFS_MAY_LOOKUP) == NFS_MAY_LOOKUP)
3230	mask |= MAY_EXEC;
3231	} else if (S_ISREG(umode)) {
3232	if ((access_result & NFS_FILE_MAY_WRITE) == NFS_FILE_MAY_WRITE)
3233	mask |= MAY_WRITE;
3234	if ((access_result & NFS_MAY_EXECUTE) == NFS_MAY_EXECUTE)
3235	mask |= MAY_EXEC;
3236	} else if (access_result & NFS_MAY_WRITE)
3237	mask |= MAY_WRITE;
3238	return mask;
3239	}
3240
3241	void nfs_access_set_mask(struct nfs_access_entry *entry, u32 access_result)
3242	{
3243	entry->mask = access_result;
3244	}
3245	EXPORT_SYMBOL_GPL(nfs_access_set_mask);
3246
3247	static int nfs_do_access(struct inode *inode, const struct cred *cred, int mask)
3248	{
3249	struct nfs_access_entry cache;
3250	bool may_block = (mask & MAY_NOT_BLOCK) == 0;
3251	int cache_mask = -1;
3252	int status;
3253
3254	trace_nfs_access_enter(inode);
3255
3256	status = nfs_access_get_cached(inode, cred, &cache.mask, may_block);
3257	if (status == 0)
3258	goto out_cached;
3259
3260	status = -ECHILD;
3261	if (!may_block)
3262	goto out;
3263
3264	/*
3265	* Determine which access bits we want to ask for...
3266	*/
3267	cache.mask = NFS_ACCESS_READ | NFS_ACCESS_MODIFY | NFS_ACCESS_EXTEND |
3268	nfs_access_xattr_mask(server: NFS_SERVER(inode));
3269	if (S_ISDIR(inode->i_mode))
3270	cache.mask |= NFS_ACCESS_DELETE | NFS_ACCESS_LOOKUP;
3271	else
3272	cache.mask |= NFS_ACCESS_EXECUTE;
3273	status = NFS_PROTO(inode)->access(inode, &cache, cred);
3274	if (status != 0) {
3275	if (status == -ESTALE) {
3276	if (!S_ISDIR(inode->i_mode))
3277	nfs_set_inode_stale(inode);
3278	else
3279	nfs_zap_caches(inode);
3280	}
3281	goto out;
3282	}
3283	nfs_access_add_cache(inode, &cache, cred);
3284	out_cached:
3285	cache_mask = nfs_access_calc_mask(access_result: cache.mask, umode: inode->i_mode);
3286	if ((mask & ~cache_mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) != 0)
3287	status = -EACCES;
3288	out:
3289	trace_nfs_access_exit(inode, mask, permitted: cache_mask, error: status);
3290	return status;
3291	}
3292
3293	static int nfs_open_permission_mask(int openflags)
3294	{
3295	int mask = 0;
3296
3297	if (openflags & __FMODE_EXEC) {
3298	/* ONLY check exec rights */
3299	mask = MAY_EXEC;
3300	} else {
3301	if ((openflags & O_ACCMODE) != O_WRONLY)
3302	mask |= MAY_READ;
3303	if ((openflags & O_ACCMODE) != O_RDONLY)
3304	mask |= MAY_WRITE;
3305	}
3306
3307	return mask;
3308	}
3309
3310	int nfs_may_open(struct inode *inode, const struct cred *cred, int openflags)
3311	{
3312	return nfs_do_access(inode, cred, mask: nfs_open_permission_mask(openflags));
3313	}
3314	EXPORT_SYMBOL_GPL(nfs_may_open);
3315
3316	static int nfs_execute_ok(struct inode *inode, int mask)
3317	{
3318	struct nfs_server *server = NFS_SERVER(inode);
3319	int ret = 0;
3320
3321	if (S_ISDIR(inode->i_mode))
3322	return 0;
3323	if (nfs_check_cache_invalid(inode, NFS_INO_INVALID_MODE)) {
3324	if (mask & MAY_NOT_BLOCK)
3325	return -ECHILD;
3326	ret = __nfs_revalidate_inode(server, inode);
3327	}
3328	if (ret == 0 && !execute_ok(inode))
3329	ret = -EACCES;
3330	return ret;
3331	}
3332
3333	int nfs_permission(struct mnt_idmap *idmap,
3334	struct inode *inode,
3335	int mask)
3336	{
3337	const struct cred *cred = current_cred();
3338	int res = 0;
3339
3340	nfs_inc_stats(inode, stat: NFSIOS_VFSACCESS);
3341
3342	if ((mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
3343	goto out;
3344	/* Is this sys_access() ? */
3345	if (mask & (MAY_ACCESS | MAY_CHDIR))
3346	goto force_lookup;
3347
3348	switch (inode->i_mode & S_IFMT) {
3349	case S_IFLNK:
3350	goto out;
3351	case S_IFREG:
3352	if ((mask & MAY_OPEN) &&
3353	nfs_server_capable(inode, NFS_CAP_ATOMIC_OPEN))
3354	return 0;
3355	break;
3356	case S_IFDIR:
3357	/*
3358	* Optimize away all write operations, since the server
3359	* will check permissions when we perform the op.
3360	*/
3361	if ((mask & MAY_WRITE) && !(mask & MAY_READ))
3362	goto out;
3363	}
3364
3365	force_lookup:
3366	if (!NFS_PROTO(inode)->access)
3367	goto out_notsup;
3368
3369	res = nfs_do_access(inode, cred, mask);
3370	out:
3371	if (!res && (mask & MAY_EXEC))
3372	res = nfs_execute_ok(inode, mask);
3373
3374	dfprintk(VFS, "NFS: permission(%s/%lu), mask=0x%x, res=%d\n",
3375	inode->i_sb->s_id, inode->i_ino, mask, res);
3376	return res;
3377	out_notsup:
3378	if (mask & MAY_NOT_BLOCK)
3379	return -ECHILD;
3380
3381	res = nfs_revalidate_inode(inode, NFS_INO_INVALID_MODE |
3382	NFS_INO_INVALID_OTHER);
3383	if (res == 0)
3384	res = generic_permission(&nop_mnt_idmap, inode, mask);
3385	goto out;
3386	}
3387	EXPORT_SYMBOL_GPL(nfs_permission);
3388