xref: /linux/fs/nfs/dir.c (revision 6eb2fb3170549737207974c2c6ad34bcc2f3025e)
1 /*
2  *  linux/fs/nfs/dir.c
3  *
4  *  Copyright (C) 1992  Rick Sladkey
5  *
6  *  nfs directory handling functions
7  *
8  * 10 Apr 1996	Added silly rename for unlink	--okir
9  * 28 Sep 1996	Improved directory cache --okir
10  * 23 Aug 1997  Claus Heine claus@momo.math.rwth-aachen.de
11  *              Re-implemented silly rename for unlink, newly implemented
12  *              silly rename for nfs_rename() following the suggestions
13  *              of Olaf Kirch (okir) found in this file.
14  *              Following Linus comments on my original hack, this version
15  *              depends only on the dcache stuff and doesn't touch the inode
16  *              layer (iput() and friends).
17  *  6 Jun 1999	Cache readdir lookups in the page cache. -DaveM
18  */
19 
20 #include <linux/module.h>
21 #include <linux/time.h>
22 #include <linux/errno.h>
23 #include <linux/stat.h>
24 #include <linux/fcntl.h>
25 #include <linux/string.h>
26 #include <linux/kernel.h>
27 #include <linux/slab.h>
28 #include <linux/mm.h>
29 #include <linux/sunrpc/clnt.h>
30 #include <linux/nfs_fs.h>
31 #include <linux/nfs_mount.h>
32 #include <linux/pagemap.h>
33 #include <linux/pagevec.h>
34 #include <linux/namei.h>
35 #include <linux/mount.h>
36 #include <linux/sched.h>
37 #include <linux/kmemleak.h>
38 #include <linux/xattr.h>
39 
40 #include "delegation.h"
41 #include "iostat.h"
42 #include "internal.h"
43 #include "fscache.h"
44 
45 /* #define NFS_DEBUG_VERBOSE 1 */
46 
47 static int nfs_opendir(struct inode *, struct file *);
48 static int nfs_closedir(struct inode *, struct file *);
49 static int nfs_readdir(struct file *, void *, filldir_t);
50 static int nfs_fsync_dir(struct file *, loff_t, loff_t, int);
51 static loff_t nfs_llseek_dir(struct file *, loff_t, int);
52 static void nfs_readdir_clear_array(struct page*);
53 
54 const struct file_operations nfs_dir_operations = {
55 	.llseek		= nfs_llseek_dir,
56 	.read		= generic_read_dir,
57 	.readdir	= nfs_readdir,
58 	.open		= nfs_opendir,
59 	.release	= nfs_closedir,
60 	.fsync		= nfs_fsync_dir,
61 };
62 
63 const struct address_space_operations nfs_dir_aops = {
64 	.freepage = nfs_readdir_clear_array,
65 };
66 
67 static struct nfs_open_dir_context *alloc_nfs_open_dir_context(struct inode *dir, struct rpc_cred *cred)
68 {
69 	struct nfs_open_dir_context *ctx;
70 	ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
71 	if (ctx != NULL) {
72 		ctx->duped = 0;
73 		ctx->attr_gencount = NFS_I(dir)->attr_gencount;
74 		ctx->dir_cookie = 0;
75 		ctx->dup_cookie = 0;
76 		ctx->cred = get_rpccred(cred);
77 		return ctx;
78 	}
79 	return  ERR_PTR(-ENOMEM);
80 }
81 
82 static void put_nfs_open_dir_context(struct nfs_open_dir_context *ctx)
83 {
84 	put_rpccred(ctx->cred);
85 	kfree(ctx);
86 }
87 
88 /*
89  * Open file
90  */
91 static int
92 nfs_opendir(struct inode *inode, struct file *filp)
93 {
94 	int res = 0;
95 	struct nfs_open_dir_context *ctx;
96 	struct rpc_cred *cred;
97 
98 	dfprintk(FILE, "NFS: open dir(%s/%s)\n",
99 			filp->f_path.dentry->d_parent->d_name.name,
100 			filp->f_path.dentry->d_name.name);
101 
102 	nfs_inc_stats(inode, NFSIOS_VFSOPEN);
103 
104 	cred = rpc_lookup_cred();
105 	if (IS_ERR(cred))
106 		return PTR_ERR(cred);
107 	ctx = alloc_nfs_open_dir_context(inode, cred);
108 	if (IS_ERR(ctx)) {
109 		res = PTR_ERR(ctx);
110 		goto out;
111 	}
112 	filp->private_data = ctx;
113 	if (filp->f_path.dentry == filp->f_path.mnt->mnt_root) {
114 		/* This is a mountpoint, so d_revalidate will never
115 		 * have been called, so we need to refresh the
116 		 * inode (for close-open consistency) ourselves.
117 		 */
118 		__nfs_revalidate_inode(NFS_SERVER(inode), inode);
119 	}
120 out:
121 	put_rpccred(cred);
122 	return res;
123 }
124 
125 static int
126 nfs_closedir(struct inode *inode, struct file *filp)
127 {
128 	put_nfs_open_dir_context(filp->private_data);
129 	return 0;
130 }
131 
132 struct nfs_cache_array_entry {
133 	u64 cookie;
134 	u64 ino;
135 	struct qstr string;
136 	unsigned char d_type;
137 };
138 
139 struct nfs_cache_array {
140 	int size;
141 	int eof_index;
142 	u64 last_cookie;
143 	struct nfs_cache_array_entry array[0];
144 };
145 
146 typedef int (*decode_dirent_t)(struct xdr_stream *, struct nfs_entry *, int);
147 typedef struct {
148 	struct file	*file;
149 	struct page	*page;
150 	unsigned long	page_index;
151 	u64		*dir_cookie;
152 	u64		last_cookie;
153 	loff_t		current_index;
154 	decode_dirent_t	decode;
155 
156 	unsigned long	timestamp;
157 	unsigned long	gencount;
158 	unsigned int	cache_entry_index;
159 	unsigned int	plus:1;
160 	unsigned int	eof:1;
161 } nfs_readdir_descriptor_t;
162 
163 /*
164  * The caller is responsible for calling nfs_readdir_release_array(page)
165  */
166 static
167 struct nfs_cache_array *nfs_readdir_get_array(struct page *page)
168 {
169 	void *ptr;
170 	if (page == NULL)
171 		return ERR_PTR(-EIO);
172 	ptr = kmap(page);
173 	if (ptr == NULL)
174 		return ERR_PTR(-ENOMEM);
175 	return ptr;
176 }
177 
178 static
179 void nfs_readdir_release_array(struct page *page)
180 {
181 	kunmap(page);
182 }
183 
184 /*
185  * we are freeing strings created by nfs_add_to_readdir_array()
186  */
187 static
188 void nfs_readdir_clear_array(struct page *page)
189 {
190 	struct nfs_cache_array *array;
191 	int i;
192 
193 	array = kmap_atomic(page);
194 	for (i = 0; i < array->size; i++)
195 		kfree(array->array[i].string.name);
196 	kunmap_atomic(array);
197 }
198 
199 /*
200  * the caller is responsible for freeing qstr.name
201  * when called by nfs_readdir_add_to_array, the strings will be freed in
202  * nfs_clear_readdir_array()
203  */
204 static
205 int nfs_readdir_make_qstr(struct qstr *string, const char *name, unsigned int len)
206 {
207 	string->len = len;
208 	string->name = kmemdup(name, len, GFP_KERNEL);
209 	if (string->name == NULL)
210 		return -ENOMEM;
211 	/*
212 	 * Avoid a kmemleak false positive. The pointer to the name is stored
213 	 * in a page cache page which kmemleak does not scan.
214 	 */
215 	kmemleak_not_leak(string->name);
216 	string->hash = full_name_hash(name, len);
217 	return 0;
218 }
219 
220 static
221 int nfs_readdir_add_to_array(struct nfs_entry *entry, struct page *page)
222 {
223 	struct nfs_cache_array *array = nfs_readdir_get_array(page);
224 	struct nfs_cache_array_entry *cache_entry;
225 	int ret;
226 
227 	if (IS_ERR(array))
228 		return PTR_ERR(array);
229 
230 	cache_entry = &array->array[array->size];
231 
232 	/* Check that this entry lies within the page bounds */
233 	ret = -ENOSPC;
234 	if ((char *)&cache_entry[1] - (char *)page_address(page) > PAGE_SIZE)
235 		goto out;
236 
237 	cache_entry->cookie = entry->prev_cookie;
238 	cache_entry->ino = entry->ino;
239 	cache_entry->d_type = entry->d_type;
240 	ret = nfs_readdir_make_qstr(&cache_entry->string, entry->name, entry->len);
241 	if (ret)
242 		goto out;
243 	array->last_cookie = entry->cookie;
244 	array->size++;
245 	if (entry->eof != 0)
246 		array->eof_index = array->size;
247 out:
248 	nfs_readdir_release_array(page);
249 	return ret;
250 }
251 
252 static
253 int nfs_readdir_search_for_pos(struct nfs_cache_array *array, nfs_readdir_descriptor_t *desc)
254 {
255 	loff_t diff = desc->file->f_pos - desc->current_index;
256 	unsigned int index;
257 
258 	if (diff < 0)
259 		goto out_eof;
260 	if (diff >= array->size) {
261 		if (array->eof_index >= 0)
262 			goto out_eof;
263 		return -EAGAIN;
264 	}
265 
266 	index = (unsigned int)diff;
267 	*desc->dir_cookie = array->array[index].cookie;
268 	desc->cache_entry_index = index;
269 	return 0;
270 out_eof:
271 	desc->eof = 1;
272 	return -EBADCOOKIE;
273 }
274 
275 static
276 int nfs_readdir_search_for_cookie(struct nfs_cache_array *array, nfs_readdir_descriptor_t *desc)
277 {
278 	int i;
279 	loff_t new_pos;
280 	int status = -EAGAIN;
281 
282 	for (i = 0; i < array->size; i++) {
283 		if (array->array[i].cookie == *desc->dir_cookie) {
284 			struct nfs_inode *nfsi = NFS_I(file_inode(desc->file));
285 			struct nfs_open_dir_context *ctx = desc->file->private_data;
286 
287 			new_pos = desc->current_index + i;
288 			if (ctx->attr_gencount != nfsi->attr_gencount
289 			    || (nfsi->cache_validity & (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA))) {
290 				ctx->duped = 0;
291 				ctx->attr_gencount = nfsi->attr_gencount;
292 			} else if (new_pos < desc->file->f_pos) {
293 				if (ctx->duped > 0
294 				    && ctx->dup_cookie == *desc->dir_cookie) {
295 					if (printk_ratelimit()) {
296 						pr_notice("NFS: directory %s/%s contains a readdir loop."
297 								"Please contact your server vendor.  "
298 								"The file: %s has duplicate cookie %llu\n",
299 								desc->file->f_dentry->d_parent->d_name.name,
300 								desc->file->f_dentry->d_name.name,
301 								array->array[i].string.name,
302 								*desc->dir_cookie);
303 					}
304 					status = -ELOOP;
305 					goto out;
306 				}
307 				ctx->dup_cookie = *desc->dir_cookie;
308 				ctx->duped = -1;
309 			}
310 			desc->file->f_pos = new_pos;
311 			desc->cache_entry_index = i;
312 			return 0;
313 		}
314 	}
315 	if (array->eof_index >= 0) {
316 		status = -EBADCOOKIE;
317 		if (*desc->dir_cookie == array->last_cookie)
318 			desc->eof = 1;
319 	}
320 out:
321 	return status;
322 }
323 
324 static
325 int nfs_readdir_search_array(nfs_readdir_descriptor_t *desc)
326 {
327 	struct nfs_cache_array *array;
328 	int status;
329 
330 	array = nfs_readdir_get_array(desc->page);
331 	if (IS_ERR(array)) {
332 		status = PTR_ERR(array);
333 		goto out;
334 	}
335 
336 	if (*desc->dir_cookie == 0)
337 		status = nfs_readdir_search_for_pos(array, desc);
338 	else
339 		status = nfs_readdir_search_for_cookie(array, desc);
340 
341 	if (status == -EAGAIN) {
342 		desc->last_cookie = array->last_cookie;
343 		desc->current_index += array->size;
344 		desc->page_index++;
345 	}
346 	nfs_readdir_release_array(desc->page);
347 out:
348 	return status;
349 }
350 
351 /* Fill a page with xdr information before transferring to the cache page */
352 static
353 int nfs_readdir_xdr_filler(struct page **pages, nfs_readdir_descriptor_t *desc,
354 			struct nfs_entry *entry, struct file *file, struct inode *inode)
355 {
356 	struct nfs_open_dir_context *ctx = file->private_data;
357 	struct rpc_cred	*cred = ctx->cred;
358 	unsigned long	timestamp, gencount;
359 	int		error;
360 
361  again:
362 	timestamp = jiffies;
363 	gencount = nfs_inc_attr_generation_counter();
364 	error = NFS_PROTO(inode)->readdir(file->f_path.dentry, cred, entry->cookie, pages,
365 					  NFS_SERVER(inode)->dtsize, desc->plus);
366 	if (error < 0) {
367 		/* We requested READDIRPLUS, but the server doesn't grok it */
368 		if (error == -ENOTSUPP && desc->plus) {
369 			NFS_SERVER(inode)->caps &= ~NFS_CAP_READDIRPLUS;
370 			clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
371 			desc->plus = 0;
372 			goto again;
373 		}
374 		goto error;
375 	}
376 	desc->timestamp = timestamp;
377 	desc->gencount = gencount;
378 error:
379 	return error;
380 }
381 
382 static int xdr_decode(nfs_readdir_descriptor_t *desc,
383 		      struct nfs_entry *entry, struct xdr_stream *xdr)
384 {
385 	int error;
386 
387 	error = desc->decode(xdr, entry, desc->plus);
388 	if (error)
389 		return error;
390 	entry->fattr->time_start = desc->timestamp;
391 	entry->fattr->gencount = desc->gencount;
392 	return 0;
393 }
394 
395 static
396 int nfs_same_file(struct dentry *dentry, struct nfs_entry *entry)
397 {
398 	if (dentry->d_inode == NULL)
399 		goto different;
400 	if (nfs_compare_fh(entry->fh, NFS_FH(dentry->d_inode)) != 0)
401 		goto different;
402 	return 1;
403 different:
404 	return 0;
405 }
406 
407 static
408 bool nfs_use_readdirplus(struct inode *dir, struct file *filp)
409 {
410 	if (!nfs_server_capable(dir, NFS_CAP_READDIRPLUS))
411 		return false;
412 	if (test_and_clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(dir)->flags))
413 		return true;
414 	if (filp->f_pos == 0)
415 		return true;
416 	return false;
417 }
418 
419 /*
420  * This function is called by the lookup code to request the use of
421  * readdirplus to accelerate any future lookups in the same
422  * directory.
423  */
424 static
425 void nfs_advise_use_readdirplus(struct inode *dir)
426 {
427 	set_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(dir)->flags);
428 }
429 
430 static
431 void nfs_prime_dcache(struct dentry *parent, struct nfs_entry *entry)
432 {
433 	struct qstr filename = QSTR_INIT(entry->name, entry->len);
434 	struct dentry *dentry;
435 	struct dentry *alias;
436 	struct inode *dir = parent->d_inode;
437 	struct inode *inode;
438 
439 	if (filename.name[0] == '.') {
440 		if (filename.len == 1)
441 			return;
442 		if (filename.len == 2 && filename.name[1] == '.')
443 			return;
444 	}
445 	filename.hash = full_name_hash(filename.name, filename.len);
446 
447 	dentry = d_lookup(parent, &filename);
448 	if (dentry != NULL) {
449 		if (nfs_same_file(dentry, entry)) {
450 			nfs_refresh_inode(dentry->d_inode, entry->fattr);
451 			goto out;
452 		} else {
453 			if (d_invalidate(dentry) != 0)
454 				goto out;
455 			dput(dentry);
456 		}
457 	}
458 
459 	dentry = d_alloc(parent, &filename);
460 	if (dentry == NULL)
461 		return;
462 
463 	inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr);
464 	if (IS_ERR(inode))
465 		goto out;
466 
467 	alias = d_materialise_unique(dentry, inode);
468 	if (IS_ERR(alias))
469 		goto out;
470 	else if (alias) {
471 		nfs_set_verifier(alias, nfs_save_change_attribute(dir));
472 		dput(alias);
473 	} else
474 		nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
475 
476 out:
477 	dput(dentry);
478 }
479 
480 /* Perform conversion from xdr to cache array */
481 static
482 int nfs_readdir_page_filler(nfs_readdir_descriptor_t *desc, struct nfs_entry *entry,
483 				struct page **xdr_pages, struct page *page, unsigned int buflen)
484 {
485 	struct xdr_stream stream;
486 	struct xdr_buf buf;
487 	struct page *scratch;
488 	struct nfs_cache_array *array;
489 	unsigned int count = 0;
490 	int status;
491 
492 	scratch = alloc_page(GFP_KERNEL);
493 	if (scratch == NULL)
494 		return -ENOMEM;
495 
496 	xdr_init_decode_pages(&stream, &buf, xdr_pages, buflen);
497 	xdr_set_scratch_buffer(&stream, page_address(scratch), PAGE_SIZE);
498 
499 	do {
500 		status = xdr_decode(desc, entry, &stream);
501 		if (status != 0) {
502 			if (status == -EAGAIN)
503 				status = 0;
504 			break;
505 		}
506 
507 		count++;
508 
509 		if (desc->plus != 0)
510 			nfs_prime_dcache(desc->file->f_path.dentry, entry);
511 
512 		status = nfs_readdir_add_to_array(entry, page);
513 		if (status != 0)
514 			break;
515 	} while (!entry->eof);
516 
517 	if (count == 0 || (status == -EBADCOOKIE && entry->eof != 0)) {
518 		array = nfs_readdir_get_array(page);
519 		if (!IS_ERR(array)) {
520 			array->eof_index = array->size;
521 			status = 0;
522 			nfs_readdir_release_array(page);
523 		} else
524 			status = PTR_ERR(array);
525 	}
526 
527 	put_page(scratch);
528 	return status;
529 }
530 
531 static
532 void nfs_readdir_free_pagearray(struct page **pages, unsigned int npages)
533 {
534 	unsigned int i;
535 	for (i = 0; i < npages; i++)
536 		put_page(pages[i]);
537 }
538 
539 static
540 void nfs_readdir_free_large_page(void *ptr, struct page **pages,
541 		unsigned int npages)
542 {
543 	nfs_readdir_free_pagearray(pages, npages);
544 }
545 
546 /*
547  * nfs_readdir_large_page will allocate pages that must be freed with a call
548  * to nfs_readdir_free_large_page
549  */
550 static
551 int nfs_readdir_large_page(struct page **pages, unsigned int npages)
552 {
553 	unsigned int i;
554 
555 	for (i = 0; i < npages; i++) {
556 		struct page *page = alloc_page(GFP_KERNEL);
557 		if (page == NULL)
558 			goto out_freepages;
559 		pages[i] = page;
560 	}
561 	return 0;
562 
563 out_freepages:
564 	nfs_readdir_free_pagearray(pages, i);
565 	return -ENOMEM;
566 }
567 
568 static
569 int nfs_readdir_xdr_to_array(nfs_readdir_descriptor_t *desc, struct page *page, struct inode *inode)
570 {
571 	struct page *pages[NFS_MAX_READDIR_PAGES];
572 	void *pages_ptr = NULL;
573 	struct nfs_entry entry;
574 	struct file	*file = desc->file;
575 	struct nfs_cache_array *array;
576 	int status = -ENOMEM;
577 	unsigned int array_size = ARRAY_SIZE(pages);
578 
579 	entry.prev_cookie = 0;
580 	entry.cookie = desc->last_cookie;
581 	entry.eof = 0;
582 	entry.fh = nfs_alloc_fhandle();
583 	entry.fattr = nfs_alloc_fattr();
584 	entry.server = NFS_SERVER(inode);
585 	if (entry.fh == NULL || entry.fattr == NULL)
586 		goto out;
587 
588 	array = nfs_readdir_get_array(page);
589 	if (IS_ERR(array)) {
590 		status = PTR_ERR(array);
591 		goto out;
592 	}
593 	memset(array, 0, sizeof(struct nfs_cache_array));
594 	array->eof_index = -1;
595 
596 	status = nfs_readdir_large_page(pages, array_size);
597 	if (status < 0)
598 		goto out_release_array;
599 	do {
600 		unsigned int pglen;
601 		status = nfs_readdir_xdr_filler(pages, desc, &entry, file, inode);
602 
603 		if (status < 0)
604 			break;
605 		pglen = status;
606 		status = nfs_readdir_page_filler(desc, &entry, pages, page, pglen);
607 		if (status < 0) {
608 			if (status == -ENOSPC)
609 				status = 0;
610 			break;
611 		}
612 	} while (array->eof_index < 0);
613 
614 	nfs_readdir_free_large_page(pages_ptr, pages, array_size);
615 out_release_array:
616 	nfs_readdir_release_array(page);
617 out:
618 	nfs_free_fattr(entry.fattr);
619 	nfs_free_fhandle(entry.fh);
620 	return status;
621 }
622 
623 /*
624  * Now we cache directories properly, by converting xdr information
625  * to an array that can be used for lookups later.  This results in
626  * fewer cache pages, since we can store more information on each page.
627  * We only need to convert from xdr once so future lookups are much simpler
628  */
629 static
630 int nfs_readdir_filler(nfs_readdir_descriptor_t *desc, struct page* page)
631 {
632 	struct inode	*inode = file_inode(desc->file);
633 	int ret;
634 
635 	ret = nfs_readdir_xdr_to_array(desc, page, inode);
636 	if (ret < 0)
637 		goto error;
638 	SetPageUptodate(page);
639 
640 	if (invalidate_inode_pages2_range(inode->i_mapping, page->index + 1, -1) < 0) {
641 		/* Should never happen */
642 		nfs_zap_mapping(inode, inode->i_mapping);
643 	}
644 	unlock_page(page);
645 	return 0;
646  error:
647 	unlock_page(page);
648 	return ret;
649 }
650 
651 static
652 void cache_page_release(nfs_readdir_descriptor_t *desc)
653 {
654 	if (!desc->page->mapping)
655 		nfs_readdir_clear_array(desc->page);
656 	page_cache_release(desc->page);
657 	desc->page = NULL;
658 }
659 
660 static
661 struct page *get_cache_page(nfs_readdir_descriptor_t *desc)
662 {
663 	return read_cache_page(file_inode(desc->file)->i_mapping,
664 			desc->page_index, (filler_t *)nfs_readdir_filler, desc);
665 }
666 
667 /*
668  * Returns 0 if desc->dir_cookie was found on page desc->page_index
669  */
670 static
671 int find_cache_page(nfs_readdir_descriptor_t *desc)
672 {
673 	int res;
674 
675 	desc->page = get_cache_page(desc);
676 	if (IS_ERR(desc->page))
677 		return PTR_ERR(desc->page);
678 
679 	res = nfs_readdir_search_array(desc);
680 	if (res != 0)
681 		cache_page_release(desc);
682 	return res;
683 }
684 
685 /* Search for desc->dir_cookie from the beginning of the page cache */
686 static inline
687 int readdir_search_pagecache(nfs_readdir_descriptor_t *desc)
688 {
689 	int res;
690 
691 	if (desc->page_index == 0) {
692 		desc->current_index = 0;
693 		desc->last_cookie = 0;
694 	}
695 	do {
696 		res = find_cache_page(desc);
697 	} while (res == -EAGAIN);
698 	return res;
699 }
700 
701 /*
702  * Once we've found the start of the dirent within a page: fill 'er up...
703  */
704 static
705 int nfs_do_filldir(nfs_readdir_descriptor_t *desc, void *dirent,
706 		   filldir_t filldir)
707 {
708 	struct file	*file = desc->file;
709 	int i = 0;
710 	int res = 0;
711 	struct nfs_cache_array *array = NULL;
712 	struct nfs_open_dir_context *ctx = file->private_data;
713 
714 	array = nfs_readdir_get_array(desc->page);
715 	if (IS_ERR(array)) {
716 		res = PTR_ERR(array);
717 		goto out;
718 	}
719 
720 	for (i = desc->cache_entry_index; i < array->size; i++) {
721 		struct nfs_cache_array_entry *ent;
722 
723 		ent = &array->array[i];
724 		if (filldir(dirent, ent->string.name, ent->string.len,
725 		    file->f_pos, nfs_compat_user_ino64(ent->ino),
726 		    ent->d_type) < 0) {
727 			desc->eof = 1;
728 			break;
729 		}
730 		file->f_pos++;
731 		if (i < (array->size-1))
732 			*desc->dir_cookie = array->array[i+1].cookie;
733 		else
734 			*desc->dir_cookie = array->last_cookie;
735 		if (ctx->duped != 0)
736 			ctx->duped = 1;
737 	}
738 	if (array->eof_index >= 0)
739 		desc->eof = 1;
740 
741 	nfs_readdir_release_array(desc->page);
742 out:
743 	cache_page_release(desc);
744 	dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n",
745 			(unsigned long long)*desc->dir_cookie, res);
746 	return res;
747 }
748 
749 /*
750  * If we cannot find a cookie in our cache, we suspect that this is
751  * because it points to a deleted file, so we ask the server to return
752  * whatever it thinks is the next entry. We then feed this to filldir.
753  * If all goes well, we should then be able to find our way round the
754  * cache on the next call to readdir_search_pagecache();
755  *
756  * NOTE: we cannot add the anonymous page to the pagecache because
757  *	 the data it contains might not be page aligned. Besides,
758  *	 we should already have a complete representation of the
759  *	 directory in the page cache by the time we get here.
760  */
761 static inline
762 int uncached_readdir(nfs_readdir_descriptor_t *desc, void *dirent,
763 		     filldir_t filldir)
764 {
765 	struct page	*page = NULL;
766 	int		status;
767 	struct inode *inode = file_inode(desc->file);
768 	struct nfs_open_dir_context *ctx = desc->file->private_data;
769 
770 	dfprintk(DIRCACHE, "NFS: uncached_readdir() searching for cookie %Lu\n",
771 			(unsigned long long)*desc->dir_cookie);
772 
773 	page = alloc_page(GFP_HIGHUSER);
774 	if (!page) {
775 		status = -ENOMEM;
776 		goto out;
777 	}
778 
779 	desc->page_index = 0;
780 	desc->last_cookie = *desc->dir_cookie;
781 	desc->page = page;
782 	ctx->duped = 0;
783 
784 	status = nfs_readdir_xdr_to_array(desc, page, inode);
785 	if (status < 0)
786 		goto out_release;
787 
788 	status = nfs_do_filldir(desc, dirent, filldir);
789 
790  out:
791 	dfprintk(DIRCACHE, "NFS: %s: returns %d\n",
792 			__func__, status);
793 	return status;
794  out_release:
795 	cache_page_release(desc);
796 	goto out;
797 }
798 
799 /* The file offset position represents the dirent entry number.  A
800    last cookie cache takes care of the common case of reading the
801    whole directory.
802  */
803 static int nfs_readdir(struct file *filp, void *dirent, filldir_t filldir)
804 {
805 	struct dentry	*dentry = filp->f_path.dentry;
806 	struct inode	*inode = dentry->d_inode;
807 	nfs_readdir_descriptor_t my_desc,
808 			*desc = &my_desc;
809 	struct nfs_open_dir_context *dir_ctx = filp->private_data;
810 	int res;
811 
812 	dfprintk(FILE, "NFS: readdir(%s/%s) starting at cookie %llu\n",
813 			dentry->d_parent->d_name.name, dentry->d_name.name,
814 			(long long)filp->f_pos);
815 	nfs_inc_stats(inode, NFSIOS_VFSGETDENTS);
816 
817 	/*
818 	 * filp->f_pos points to the dirent entry number.
819 	 * *desc->dir_cookie has the cookie for the next entry. We have
820 	 * to either find the entry with the appropriate number or
821 	 * revalidate the cookie.
822 	 */
823 	memset(desc, 0, sizeof(*desc));
824 
825 	desc->file = filp;
826 	desc->dir_cookie = &dir_ctx->dir_cookie;
827 	desc->decode = NFS_PROTO(inode)->decode_dirent;
828 	desc->plus = nfs_use_readdirplus(inode, filp) ? 1 : 0;
829 
830 	nfs_block_sillyrename(dentry);
831 	res = nfs_revalidate_mapping(inode, filp->f_mapping);
832 	if (res < 0)
833 		goto out;
834 
835 	do {
836 		res = readdir_search_pagecache(desc);
837 
838 		if (res == -EBADCOOKIE) {
839 			res = 0;
840 			/* This means either end of directory */
841 			if (*desc->dir_cookie && desc->eof == 0) {
842 				/* Or that the server has 'lost' a cookie */
843 				res = uncached_readdir(desc, dirent, filldir);
844 				if (res == 0)
845 					continue;
846 			}
847 			break;
848 		}
849 		if (res == -ETOOSMALL && desc->plus) {
850 			clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
851 			nfs_zap_caches(inode);
852 			desc->page_index = 0;
853 			desc->plus = 0;
854 			desc->eof = 0;
855 			continue;
856 		}
857 		if (res < 0)
858 			break;
859 
860 		res = nfs_do_filldir(desc, dirent, filldir);
861 		if (res < 0)
862 			break;
863 	} while (!desc->eof);
864 out:
865 	nfs_unblock_sillyrename(dentry);
866 	if (res > 0)
867 		res = 0;
868 	dfprintk(FILE, "NFS: readdir(%s/%s) returns %d\n",
869 			dentry->d_parent->d_name.name, dentry->d_name.name,
870 			res);
871 	return res;
872 }
873 
874 static loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int whence)
875 {
876 	struct dentry *dentry = filp->f_path.dentry;
877 	struct inode *inode = dentry->d_inode;
878 	struct nfs_open_dir_context *dir_ctx = filp->private_data;
879 
880 	dfprintk(FILE, "NFS: llseek dir(%s/%s, %lld, %d)\n",
881 			dentry->d_parent->d_name.name,
882 			dentry->d_name.name,
883 			offset, whence);
884 
885 	mutex_lock(&inode->i_mutex);
886 	switch (whence) {
887 		case 1:
888 			offset += filp->f_pos;
889 		case 0:
890 			if (offset >= 0)
891 				break;
892 		default:
893 			offset = -EINVAL;
894 			goto out;
895 	}
896 	if (offset != filp->f_pos) {
897 		filp->f_pos = offset;
898 		dir_ctx->dir_cookie = 0;
899 		dir_ctx->duped = 0;
900 	}
901 out:
902 	mutex_unlock(&inode->i_mutex);
903 	return offset;
904 }
905 
906 /*
907  * All directory operations under NFS are synchronous, so fsync()
908  * is a dummy operation.
909  */
910 static int nfs_fsync_dir(struct file *filp, loff_t start, loff_t end,
911 			 int datasync)
912 {
913 	struct dentry *dentry = filp->f_path.dentry;
914 	struct inode *inode = dentry->d_inode;
915 
916 	dfprintk(FILE, "NFS: fsync dir(%s/%s) datasync %d\n",
917 			dentry->d_parent->d_name.name, dentry->d_name.name,
918 			datasync);
919 
920 	mutex_lock(&inode->i_mutex);
921 	nfs_inc_stats(dentry->d_inode, NFSIOS_VFSFSYNC);
922 	mutex_unlock(&inode->i_mutex);
923 	return 0;
924 }
925 
926 /**
927  * nfs_force_lookup_revalidate - Mark the directory as having changed
928  * @dir - pointer to directory inode
929  *
930  * This forces the revalidation code in nfs_lookup_revalidate() to do a
931  * full lookup on all child dentries of 'dir' whenever a change occurs
932  * on the server that might have invalidated our dcache.
933  *
934  * The caller should be holding dir->i_lock
935  */
936 void nfs_force_lookup_revalidate(struct inode *dir)
937 {
938 	NFS_I(dir)->cache_change_attribute++;
939 }
940 EXPORT_SYMBOL_GPL(nfs_force_lookup_revalidate);
941 
942 /*
943  * A check for whether or not the parent directory has changed.
944  * In the case it has, we assume that the dentries are untrustworthy
945  * and may need to be looked up again.
946  */
947 static int nfs_check_verifier(struct inode *dir, struct dentry *dentry)
948 {
949 	if (IS_ROOT(dentry))
950 		return 1;
951 	if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONE)
952 		return 0;
953 	if (!nfs_verify_change_attribute(dir, dentry->d_time))
954 		return 0;
955 	/* Revalidate nfsi->cache_change_attribute before we declare a match */
956 	if (nfs_revalidate_inode(NFS_SERVER(dir), dir) < 0)
957 		return 0;
958 	if (!nfs_verify_change_attribute(dir, dentry->d_time))
959 		return 0;
960 	return 1;
961 }
962 
963 /*
964  * Use intent information to check whether or not we're going to do
965  * an O_EXCL create using this path component.
966  */
967 static int nfs_is_exclusive_create(struct inode *dir, unsigned int flags)
968 {
969 	if (NFS_PROTO(dir)->version == 2)
970 		return 0;
971 	return flags & LOOKUP_EXCL;
972 }
973 
974 /*
975  * Inode and filehandle revalidation for lookups.
976  *
977  * We force revalidation in the cases where the VFS sets LOOKUP_REVAL,
978  * or if the intent information indicates that we're about to open this
979  * particular file and the "nocto" mount flag is not set.
980  *
981  */
982 static
983 int nfs_lookup_verify_inode(struct inode *inode, unsigned int flags)
984 {
985 	struct nfs_server *server = NFS_SERVER(inode);
986 	int ret;
987 
988 	if (IS_AUTOMOUNT(inode))
989 		return 0;
990 	/* VFS wants an on-the-wire revalidation */
991 	if (flags & LOOKUP_REVAL)
992 		goto out_force;
993 	/* This is an open(2) */
994 	if ((flags & LOOKUP_OPEN) && !(server->flags & NFS_MOUNT_NOCTO) &&
995 	    (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)))
996 		goto out_force;
997 out:
998 	return (inode->i_nlink == 0) ? -ENOENT : 0;
999 out_force:
1000 	ret = __nfs_revalidate_inode(server, inode);
1001 	if (ret != 0)
1002 		return ret;
1003 	goto out;
1004 }
1005 
1006 /*
1007  * We judge how long we want to trust negative
1008  * dentries by looking at the parent inode mtime.
1009  *
1010  * If parent mtime has changed, we revalidate, else we wait for a
1011  * period corresponding to the parent's attribute cache timeout value.
1012  */
1013 static inline
1014 int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry,
1015 		       unsigned int flags)
1016 {
1017 	/* Don't revalidate a negative dentry if we're creating a new file */
1018 	if (flags & LOOKUP_CREATE)
1019 		return 0;
1020 	if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONEG)
1021 		return 1;
1022 	return !nfs_check_verifier(dir, dentry);
1023 }
1024 
1025 /*
1026  * This is called every time the dcache has a lookup hit,
1027  * and we should check whether we can really trust that
1028  * lookup.
1029  *
1030  * NOTE! The hit can be a negative hit too, don't assume
1031  * we have an inode!
1032  *
1033  * If the parent directory is seen to have changed, we throw out the
1034  * cached dentry and do a new lookup.
1035  */
1036 static int nfs_lookup_revalidate(struct dentry *dentry, unsigned int flags)
1037 {
1038 	struct inode *dir;
1039 	struct inode *inode;
1040 	struct dentry *parent;
1041 	struct nfs_fh *fhandle = NULL;
1042 	struct nfs_fattr *fattr = NULL;
1043 	int error;
1044 
1045 	if (flags & LOOKUP_RCU)
1046 		return -ECHILD;
1047 
1048 	parent = dget_parent(dentry);
1049 	dir = parent->d_inode;
1050 	nfs_inc_stats(dir, NFSIOS_DENTRYREVALIDATE);
1051 	inode = dentry->d_inode;
1052 
1053 	if (!inode) {
1054 		if (nfs_neg_need_reval(dir, dentry, flags))
1055 			goto out_bad;
1056 		goto out_valid_noent;
1057 	}
1058 
1059 	if (is_bad_inode(inode)) {
1060 		dfprintk(LOOKUPCACHE, "%s: %s/%s has dud inode\n",
1061 				__func__, dentry->d_parent->d_name.name,
1062 				dentry->d_name.name);
1063 		goto out_bad;
1064 	}
1065 
1066 	if (NFS_PROTO(dir)->have_delegation(inode, FMODE_READ))
1067 		goto out_set_verifier;
1068 
1069 	/* Force a full look up iff the parent directory has changed */
1070 	if (!nfs_is_exclusive_create(dir, flags) && nfs_check_verifier(dir, dentry)) {
1071 		if (nfs_lookup_verify_inode(inode, flags))
1072 			goto out_zap_parent;
1073 		goto out_valid;
1074 	}
1075 
1076 	if (NFS_STALE(inode))
1077 		goto out_bad;
1078 
1079 	error = -ENOMEM;
1080 	fhandle = nfs_alloc_fhandle();
1081 	fattr = nfs_alloc_fattr();
1082 	if (fhandle == NULL || fattr == NULL)
1083 		goto out_error;
1084 
1085 	error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr);
1086 	if (error)
1087 		goto out_bad;
1088 	if (nfs_compare_fh(NFS_FH(inode), fhandle))
1089 		goto out_bad;
1090 	if ((error = nfs_refresh_inode(inode, fattr)) != 0)
1091 		goto out_bad;
1092 
1093 	nfs_free_fattr(fattr);
1094 	nfs_free_fhandle(fhandle);
1095 out_set_verifier:
1096 	nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1097  out_valid:
1098 	/* Success: notify readdir to use READDIRPLUS */
1099 	nfs_advise_use_readdirplus(dir);
1100  out_valid_noent:
1101 	dput(parent);
1102 	dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is valid\n",
1103 			__func__, dentry->d_parent->d_name.name,
1104 			dentry->d_name.name);
1105 	return 1;
1106 out_zap_parent:
1107 	nfs_zap_caches(dir);
1108  out_bad:
1109 	nfs_free_fattr(fattr);
1110 	nfs_free_fhandle(fhandle);
1111 	nfs_mark_for_revalidate(dir);
1112 	if (inode && S_ISDIR(inode->i_mode)) {
1113 		/* Purge readdir caches. */
1114 		nfs_zap_caches(inode);
1115 		/* If we have submounts, don't unhash ! */
1116 		if (have_submounts(dentry))
1117 			goto out_valid;
1118 		if (dentry->d_flags & DCACHE_DISCONNECTED)
1119 			goto out_valid;
1120 		shrink_dcache_parent(dentry);
1121 	}
1122 	d_drop(dentry);
1123 	dput(parent);
1124 	dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is invalid\n",
1125 			__func__, dentry->d_parent->d_name.name,
1126 			dentry->d_name.name);
1127 	return 0;
1128 out_error:
1129 	nfs_free_fattr(fattr);
1130 	nfs_free_fhandle(fhandle);
1131 	dput(parent);
1132 	dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) lookup returned error %d\n",
1133 			__func__, dentry->d_parent->d_name.name,
1134 			dentry->d_name.name, error);
1135 	return error;
1136 }
1137 
1138 /*
1139  * A weaker form of d_revalidate for revalidating just the dentry->d_inode
1140  * when we don't really care about the dentry name. This is called when a
1141  * pathwalk ends on a dentry that was not found via a normal lookup in the
1142  * parent dir (e.g.: ".", "..", procfs symlinks or mountpoint traversals).
1143  *
1144  * In this situation, we just want to verify that the inode itself is OK
1145  * since the dentry might have changed on the server.
1146  */
1147 static int nfs_weak_revalidate(struct dentry *dentry, unsigned int flags)
1148 {
1149 	int error;
1150 	struct inode *inode = dentry->d_inode;
1151 
1152 	/*
1153 	 * I believe we can only get a negative dentry here in the case of a
1154 	 * procfs-style symlink. Just assume it's correct for now, but we may
1155 	 * eventually need to do something more here.
1156 	 */
1157 	if (!inode) {
1158 		dfprintk(LOOKUPCACHE, "%s: %s/%s has negative inode\n",
1159 				__func__, dentry->d_parent->d_name.name,
1160 				dentry->d_name.name);
1161 		return 1;
1162 	}
1163 
1164 	if (is_bad_inode(inode)) {
1165 		dfprintk(LOOKUPCACHE, "%s: %s/%s has dud inode\n",
1166 				__func__, dentry->d_parent->d_name.name,
1167 				dentry->d_name.name);
1168 		return 0;
1169 	}
1170 
1171 	error = nfs_revalidate_inode(NFS_SERVER(inode), inode);
1172 	dfprintk(LOOKUPCACHE, "NFS: %s: inode %lu is %s\n",
1173 			__func__, inode->i_ino, error ? "invalid" : "valid");
1174 	return !error;
1175 }
1176 
1177 /*
1178  * This is called from dput() when d_count is going to 0.
1179  */
1180 static int nfs_dentry_delete(const struct dentry *dentry)
1181 {
1182 	dfprintk(VFS, "NFS: dentry_delete(%s/%s, %x)\n",
1183 		dentry->d_parent->d_name.name, dentry->d_name.name,
1184 		dentry->d_flags);
1185 
1186 	/* Unhash any dentry with a stale inode */
1187 	if (dentry->d_inode != NULL && NFS_STALE(dentry->d_inode))
1188 		return 1;
1189 
1190 	if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1191 		/* Unhash it, so that ->d_iput() would be called */
1192 		return 1;
1193 	}
1194 	if (!(dentry->d_sb->s_flags & MS_ACTIVE)) {
1195 		/* Unhash it, so that ancestors of killed async unlink
1196 		 * files will be cleaned up during umount */
1197 		return 1;
1198 	}
1199 	return 0;
1200 
1201 }
1202 
1203 /* Ensure that we revalidate inode->i_nlink */
1204 static void nfs_drop_nlink(struct inode *inode)
1205 {
1206 	spin_lock(&inode->i_lock);
1207 	/* drop the inode if we're reasonably sure this is the last link */
1208 	if (inode->i_nlink == 1)
1209 		clear_nlink(inode);
1210 	NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ATTR;
1211 	spin_unlock(&inode->i_lock);
1212 }
1213 
1214 /*
1215  * Called when the dentry loses inode.
1216  * We use it to clean up silly-renamed files.
1217  */
1218 static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode)
1219 {
1220 	if (S_ISDIR(inode->i_mode))
1221 		/* drop any readdir cache as it could easily be old */
1222 		NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA;
1223 
1224 	if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1225 		nfs_complete_unlink(dentry, inode);
1226 		nfs_drop_nlink(inode);
1227 	}
1228 	iput(inode);
1229 }
1230 
1231 static void nfs_d_release(struct dentry *dentry)
1232 {
1233 	/* free cached devname value, if it survived that far */
1234 	if (unlikely(dentry->d_fsdata)) {
1235 		if (dentry->d_flags & DCACHE_NFSFS_RENAMED)
1236 			WARN_ON(1);
1237 		else
1238 			kfree(dentry->d_fsdata);
1239 	}
1240 }
1241 
1242 const struct dentry_operations nfs_dentry_operations = {
1243 	.d_revalidate	= nfs_lookup_revalidate,
1244 	.d_weak_revalidate	= nfs_weak_revalidate,
1245 	.d_delete	= nfs_dentry_delete,
1246 	.d_iput		= nfs_dentry_iput,
1247 	.d_automount	= nfs_d_automount,
1248 	.d_release	= nfs_d_release,
1249 };
1250 EXPORT_SYMBOL_GPL(nfs_dentry_operations);
1251 
1252 struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
1253 {
1254 	struct dentry *res;
1255 	struct dentry *parent;
1256 	struct inode *inode = NULL;
1257 	struct nfs_fh *fhandle = NULL;
1258 	struct nfs_fattr *fattr = NULL;
1259 	int error;
1260 
1261 	dfprintk(VFS, "NFS: lookup(%s/%s)\n",
1262 		dentry->d_parent->d_name.name, dentry->d_name.name);
1263 	nfs_inc_stats(dir, NFSIOS_VFSLOOKUP);
1264 
1265 	res = ERR_PTR(-ENAMETOOLONG);
1266 	if (dentry->d_name.len > NFS_SERVER(dir)->namelen)
1267 		goto out;
1268 
1269 	/*
1270 	 * If we're doing an exclusive create, optimize away the lookup
1271 	 * but don't hash the dentry.
1272 	 */
1273 	if (nfs_is_exclusive_create(dir, flags)) {
1274 		d_instantiate(dentry, NULL);
1275 		res = NULL;
1276 		goto out;
1277 	}
1278 
1279 	res = ERR_PTR(-ENOMEM);
1280 	fhandle = nfs_alloc_fhandle();
1281 	fattr = nfs_alloc_fattr();
1282 	if (fhandle == NULL || fattr == NULL)
1283 		goto out;
1284 
1285 	parent = dentry->d_parent;
1286 	/* Protect against concurrent sillydeletes */
1287 	nfs_block_sillyrename(parent);
1288 	error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr);
1289 	if (error == -ENOENT)
1290 		goto no_entry;
1291 	if (error < 0) {
1292 		res = ERR_PTR(error);
1293 		goto out_unblock_sillyrename;
1294 	}
1295 	inode = nfs_fhget(dentry->d_sb, fhandle, fattr);
1296 	res = ERR_CAST(inode);
1297 	if (IS_ERR(res))
1298 		goto out_unblock_sillyrename;
1299 
1300 	/* Success: notify readdir to use READDIRPLUS */
1301 	nfs_advise_use_readdirplus(dir);
1302 
1303 no_entry:
1304 	res = d_materialise_unique(dentry, inode);
1305 	if (res != NULL) {
1306 		if (IS_ERR(res))
1307 			goto out_unblock_sillyrename;
1308 		dentry = res;
1309 	}
1310 	nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1311 out_unblock_sillyrename:
1312 	nfs_unblock_sillyrename(parent);
1313 out:
1314 	nfs_free_fattr(fattr);
1315 	nfs_free_fhandle(fhandle);
1316 	return res;
1317 }
1318 EXPORT_SYMBOL_GPL(nfs_lookup);
1319 
1320 #if IS_ENABLED(CONFIG_NFS_V4)
1321 static int nfs4_lookup_revalidate(struct dentry *, unsigned int);
1322 
1323 const struct dentry_operations nfs4_dentry_operations = {
1324 	.d_revalidate	= nfs4_lookup_revalidate,
1325 	.d_delete	= nfs_dentry_delete,
1326 	.d_iput		= nfs_dentry_iput,
1327 	.d_automount	= nfs_d_automount,
1328 	.d_release	= nfs_d_release,
1329 };
1330 EXPORT_SYMBOL_GPL(nfs4_dentry_operations);
1331 
1332 static fmode_t flags_to_mode(int flags)
1333 {
1334 	fmode_t res = (__force fmode_t)flags & FMODE_EXEC;
1335 	if ((flags & O_ACCMODE) != O_WRONLY)
1336 		res |= FMODE_READ;
1337 	if ((flags & O_ACCMODE) != O_RDONLY)
1338 		res |= FMODE_WRITE;
1339 	return res;
1340 }
1341 
1342 static struct nfs_open_context *create_nfs_open_context(struct dentry *dentry, int open_flags)
1343 {
1344 	return alloc_nfs_open_context(dentry, flags_to_mode(open_flags));
1345 }
1346 
1347 static int do_open(struct inode *inode, struct file *filp)
1348 {
1349 	nfs_fscache_set_inode_cookie(inode, filp);
1350 	return 0;
1351 }
1352 
1353 static int nfs_finish_open(struct nfs_open_context *ctx,
1354 			   struct dentry *dentry,
1355 			   struct file *file, unsigned open_flags,
1356 			   int *opened)
1357 {
1358 	int err;
1359 
1360 	if (ctx->dentry != dentry) {
1361 		dput(ctx->dentry);
1362 		ctx->dentry = dget(dentry);
1363 	}
1364 
1365 	/* If the open_intent is for execute, we have an extra check to make */
1366 	if (ctx->mode & FMODE_EXEC) {
1367 		err = nfs_may_open(dentry->d_inode, ctx->cred, open_flags);
1368 		if (err < 0)
1369 			goto out;
1370 	}
1371 
1372 	err = finish_open(file, dentry, do_open, opened);
1373 	if (err)
1374 		goto out;
1375 	nfs_file_set_open_context(file, ctx);
1376 
1377 out:
1378 	put_nfs_open_context(ctx);
1379 	return err;
1380 }
1381 
1382 int nfs_atomic_open(struct inode *dir, struct dentry *dentry,
1383 		    struct file *file, unsigned open_flags,
1384 		    umode_t mode, int *opened)
1385 {
1386 	struct nfs_open_context *ctx;
1387 	struct dentry *res;
1388 	struct iattr attr = { .ia_valid = ATTR_OPEN };
1389 	struct inode *inode;
1390 	int err;
1391 
1392 	/* Expect a negative dentry */
1393 	BUG_ON(dentry->d_inode);
1394 
1395 	dfprintk(VFS, "NFS: atomic_open(%s/%ld), %s\n",
1396 			dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1397 
1398 	/* NFS only supports OPEN on regular files */
1399 	if ((open_flags & O_DIRECTORY)) {
1400 		if (!d_unhashed(dentry)) {
1401 			/*
1402 			 * Hashed negative dentry with O_DIRECTORY: dentry was
1403 			 * revalidated and is fine, no need to perform lookup
1404 			 * again
1405 			 */
1406 			return -ENOENT;
1407 		}
1408 		goto no_open;
1409 	}
1410 
1411 	if (dentry->d_name.len > NFS_SERVER(dir)->namelen)
1412 		return -ENAMETOOLONG;
1413 
1414 	if (open_flags & O_CREAT) {
1415 		attr.ia_valid |= ATTR_MODE;
1416 		attr.ia_mode = mode & ~current_umask();
1417 	}
1418 	if (open_flags & O_TRUNC) {
1419 		attr.ia_valid |= ATTR_SIZE;
1420 		attr.ia_size = 0;
1421 	}
1422 
1423 	ctx = create_nfs_open_context(dentry, open_flags);
1424 	err = PTR_ERR(ctx);
1425 	if (IS_ERR(ctx))
1426 		goto out;
1427 
1428 	nfs_block_sillyrename(dentry->d_parent);
1429 	inode = NFS_PROTO(dir)->open_context(dir, ctx, open_flags, &attr);
1430 	d_drop(dentry);
1431 	if (IS_ERR(inode)) {
1432 		nfs_unblock_sillyrename(dentry->d_parent);
1433 		put_nfs_open_context(ctx);
1434 		err = PTR_ERR(inode);
1435 		switch (err) {
1436 		case -ENOENT:
1437 			d_add(dentry, NULL);
1438 			break;
1439 		case -EISDIR:
1440 		case -ENOTDIR:
1441 			goto no_open;
1442 		case -ELOOP:
1443 			if (!(open_flags & O_NOFOLLOW))
1444 				goto no_open;
1445 			break;
1446 			/* case -EINVAL: */
1447 		default:
1448 			break;
1449 		}
1450 		goto out;
1451 	}
1452 	res = d_add_unique(dentry, inode);
1453 	if (res != NULL)
1454 		dentry = res;
1455 
1456 	nfs_unblock_sillyrename(dentry->d_parent);
1457 	nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1458 
1459 	err = nfs_finish_open(ctx, dentry, file, open_flags, opened);
1460 
1461 	dput(res);
1462 out:
1463 	return err;
1464 
1465 no_open:
1466 	res = nfs_lookup(dir, dentry, 0);
1467 	err = PTR_ERR(res);
1468 	if (IS_ERR(res))
1469 		goto out;
1470 
1471 	return finish_no_open(file, res);
1472 }
1473 EXPORT_SYMBOL_GPL(nfs_atomic_open);
1474 
1475 static int nfs4_lookup_revalidate(struct dentry *dentry, unsigned int flags)
1476 {
1477 	struct dentry *parent = NULL;
1478 	struct inode *inode;
1479 	struct inode *dir;
1480 	int ret = 0;
1481 
1482 	if (flags & LOOKUP_RCU)
1483 		return -ECHILD;
1484 
1485 	if (!(flags & LOOKUP_OPEN) || (flags & LOOKUP_DIRECTORY))
1486 		goto no_open;
1487 	if (d_mountpoint(dentry))
1488 		goto no_open;
1489 	if (NFS_SB(dentry->d_sb)->caps & NFS_CAP_ATOMIC_OPEN_V1)
1490 		goto no_open;
1491 
1492 	inode = dentry->d_inode;
1493 	parent = dget_parent(dentry);
1494 	dir = parent->d_inode;
1495 
1496 	/* We can't create new files in nfs_open_revalidate(), so we
1497 	 * optimize away revalidation of negative dentries.
1498 	 */
1499 	if (inode == NULL) {
1500 		if (!nfs_neg_need_reval(dir, dentry, flags))
1501 			ret = 1;
1502 		goto out;
1503 	}
1504 
1505 	/* NFS only supports OPEN on regular files */
1506 	if (!S_ISREG(inode->i_mode))
1507 		goto no_open_dput;
1508 	/* We cannot do exclusive creation on a positive dentry */
1509 	if (flags & LOOKUP_EXCL)
1510 		goto no_open_dput;
1511 
1512 	/* Let f_op->open() actually open (and revalidate) the file */
1513 	ret = 1;
1514 
1515 out:
1516 	dput(parent);
1517 	return ret;
1518 
1519 no_open_dput:
1520 	dput(parent);
1521 no_open:
1522 	return nfs_lookup_revalidate(dentry, flags);
1523 }
1524 
1525 #endif /* CONFIG_NFSV4 */
1526 
1527 /*
1528  * Code common to create, mkdir, and mknod.
1529  */
1530 int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle,
1531 				struct nfs_fattr *fattr)
1532 {
1533 	struct dentry *parent = dget_parent(dentry);
1534 	struct inode *dir = parent->d_inode;
1535 	struct inode *inode;
1536 	int error = -EACCES;
1537 
1538 	d_drop(dentry);
1539 
1540 	/* We may have been initialized further down */
1541 	if (dentry->d_inode)
1542 		goto out;
1543 	if (fhandle->size == 0) {
1544 		error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr);
1545 		if (error)
1546 			goto out_error;
1547 	}
1548 	nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1549 	if (!(fattr->valid & NFS_ATTR_FATTR)) {
1550 		struct nfs_server *server = NFS_SB(dentry->d_sb);
1551 		error = server->nfs_client->rpc_ops->getattr(server, fhandle, fattr);
1552 		if (error < 0)
1553 			goto out_error;
1554 	}
1555 	inode = nfs_fhget(dentry->d_sb, fhandle, fattr);
1556 	error = PTR_ERR(inode);
1557 	if (IS_ERR(inode))
1558 		goto out_error;
1559 	d_add(dentry, inode);
1560 out:
1561 	dput(parent);
1562 	return 0;
1563 out_error:
1564 	nfs_mark_for_revalidate(dir);
1565 	dput(parent);
1566 	return error;
1567 }
1568 EXPORT_SYMBOL_GPL(nfs_instantiate);
1569 
1570 /*
1571  * Following a failed create operation, we drop the dentry rather
1572  * than retain a negative dentry. This avoids a problem in the event
1573  * that the operation succeeded on the server, but an error in the
1574  * reply path made it appear to have failed.
1575  */
1576 int nfs_create(struct inode *dir, struct dentry *dentry,
1577 		umode_t mode, bool excl)
1578 {
1579 	struct iattr attr;
1580 	int open_flags = excl ? O_CREAT | O_EXCL : O_CREAT;
1581 	int error;
1582 
1583 	dfprintk(VFS, "NFS: create(%s/%ld), %s\n",
1584 			dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1585 
1586 	attr.ia_mode = mode;
1587 	attr.ia_valid = ATTR_MODE;
1588 
1589 	error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags);
1590 	if (error != 0)
1591 		goto out_err;
1592 	return 0;
1593 out_err:
1594 	d_drop(dentry);
1595 	return error;
1596 }
1597 EXPORT_SYMBOL_GPL(nfs_create);
1598 
1599 /*
1600  * See comments for nfs_proc_create regarding failed operations.
1601  */
1602 int
1603 nfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t rdev)
1604 {
1605 	struct iattr attr;
1606 	int status;
1607 
1608 	dfprintk(VFS, "NFS: mknod(%s/%ld), %s\n",
1609 			dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1610 
1611 	if (!new_valid_dev(rdev))
1612 		return -EINVAL;
1613 
1614 	attr.ia_mode = mode;
1615 	attr.ia_valid = ATTR_MODE;
1616 
1617 	status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev);
1618 	if (status != 0)
1619 		goto out_err;
1620 	return 0;
1621 out_err:
1622 	d_drop(dentry);
1623 	return status;
1624 }
1625 EXPORT_SYMBOL_GPL(nfs_mknod);
1626 
1627 /*
1628  * See comments for nfs_proc_create regarding failed operations.
1629  */
1630 int nfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
1631 {
1632 	struct iattr attr;
1633 	int error;
1634 
1635 	dfprintk(VFS, "NFS: mkdir(%s/%ld), %s\n",
1636 			dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1637 
1638 	attr.ia_valid = ATTR_MODE;
1639 	attr.ia_mode = mode | S_IFDIR;
1640 
1641 	error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr);
1642 	if (error != 0)
1643 		goto out_err;
1644 	return 0;
1645 out_err:
1646 	d_drop(dentry);
1647 	return error;
1648 }
1649 EXPORT_SYMBOL_GPL(nfs_mkdir);
1650 
1651 static void nfs_dentry_handle_enoent(struct dentry *dentry)
1652 {
1653 	if (dentry->d_inode != NULL && !d_unhashed(dentry))
1654 		d_delete(dentry);
1655 }
1656 
1657 int nfs_rmdir(struct inode *dir, struct dentry *dentry)
1658 {
1659 	int error;
1660 
1661 	dfprintk(VFS, "NFS: rmdir(%s/%ld), %s\n",
1662 			dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1663 
1664 	error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
1665 	/* Ensure the VFS deletes this inode */
1666 	if (error == 0 && dentry->d_inode != NULL)
1667 		clear_nlink(dentry->d_inode);
1668 	else if (error == -ENOENT)
1669 		nfs_dentry_handle_enoent(dentry);
1670 
1671 	return error;
1672 }
1673 EXPORT_SYMBOL_GPL(nfs_rmdir);
1674 
1675 /*
1676  * Remove a file after making sure there are no pending writes,
1677  * and after checking that the file has only one user.
1678  *
1679  * We invalidate the attribute cache and free the inode prior to the operation
1680  * to avoid possible races if the server reuses the inode.
1681  */
1682 static int nfs_safe_remove(struct dentry *dentry)
1683 {
1684 	struct inode *dir = dentry->d_parent->d_inode;
1685 	struct inode *inode = dentry->d_inode;
1686 	int error = -EBUSY;
1687 
1688 	dfprintk(VFS, "NFS: safe_remove(%s/%s)\n",
1689 		dentry->d_parent->d_name.name, dentry->d_name.name);
1690 
1691 	/* If the dentry was sillyrenamed, we simply call d_delete() */
1692 	if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1693 		error = 0;
1694 		goto out;
1695 	}
1696 
1697 	if (inode != NULL) {
1698 		NFS_PROTO(inode)->return_delegation(inode);
1699 		error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1700 		if (error == 0)
1701 			nfs_drop_nlink(inode);
1702 	} else
1703 		error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1704 	if (error == -ENOENT)
1705 		nfs_dentry_handle_enoent(dentry);
1706 out:
1707 	return error;
1708 }
1709 
1710 /*  We do silly rename. In case sillyrename() returns -EBUSY, the inode
1711  *  belongs to an active ".nfs..." file and we return -EBUSY.
1712  *
1713  *  If sillyrename() returns 0, we do nothing, otherwise we unlink.
1714  */
1715 int nfs_unlink(struct inode *dir, struct dentry *dentry)
1716 {
1717 	int error;
1718 	int need_rehash = 0;
1719 
1720 	dfprintk(VFS, "NFS: unlink(%s/%ld, %s)\n", dir->i_sb->s_id,
1721 		dir->i_ino, dentry->d_name.name);
1722 
1723 	spin_lock(&dentry->d_lock);
1724 	if (dentry->d_count > 1) {
1725 		spin_unlock(&dentry->d_lock);
1726 		/* Start asynchronous writeout of the inode */
1727 		write_inode_now(dentry->d_inode, 0);
1728 		error = nfs_sillyrename(dir, dentry);
1729 		return error;
1730 	}
1731 	if (!d_unhashed(dentry)) {
1732 		__d_drop(dentry);
1733 		need_rehash = 1;
1734 	}
1735 	spin_unlock(&dentry->d_lock);
1736 	error = nfs_safe_remove(dentry);
1737 	if (!error || error == -ENOENT) {
1738 		nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1739 	} else if (need_rehash)
1740 		d_rehash(dentry);
1741 	return error;
1742 }
1743 EXPORT_SYMBOL_GPL(nfs_unlink);
1744 
1745 /*
1746  * To create a symbolic link, most file systems instantiate a new inode,
1747  * add a page to it containing the path, then write it out to the disk
1748  * using prepare_write/commit_write.
1749  *
1750  * Unfortunately the NFS client can't create the in-core inode first
1751  * because it needs a file handle to create an in-core inode (see
1752  * fs/nfs/inode.c:nfs_fhget).  We only have a file handle *after* the
1753  * symlink request has completed on the server.
1754  *
1755  * So instead we allocate a raw page, copy the symname into it, then do
1756  * the SYMLINK request with the page as the buffer.  If it succeeds, we
1757  * now have a new file handle and can instantiate an in-core NFS inode
1758  * and move the raw page into its mapping.
1759  */
1760 int nfs_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1761 {
1762 	struct pagevec lru_pvec;
1763 	struct page *page;
1764 	char *kaddr;
1765 	struct iattr attr;
1766 	unsigned int pathlen = strlen(symname);
1767 	int error;
1768 
1769 	dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s)\n", dir->i_sb->s_id,
1770 		dir->i_ino, dentry->d_name.name, symname);
1771 
1772 	if (pathlen > PAGE_SIZE)
1773 		return -ENAMETOOLONG;
1774 
1775 	attr.ia_mode = S_IFLNK | S_IRWXUGO;
1776 	attr.ia_valid = ATTR_MODE;
1777 
1778 	page = alloc_page(GFP_HIGHUSER);
1779 	if (!page)
1780 		return -ENOMEM;
1781 
1782 	kaddr = kmap_atomic(page);
1783 	memcpy(kaddr, symname, pathlen);
1784 	if (pathlen < PAGE_SIZE)
1785 		memset(kaddr + pathlen, 0, PAGE_SIZE - pathlen);
1786 	kunmap_atomic(kaddr);
1787 
1788 	error = NFS_PROTO(dir)->symlink(dir, dentry, page, pathlen, &attr);
1789 	if (error != 0) {
1790 		dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s) error %d\n",
1791 			dir->i_sb->s_id, dir->i_ino,
1792 			dentry->d_name.name, symname, error);
1793 		d_drop(dentry);
1794 		__free_page(page);
1795 		return error;
1796 	}
1797 
1798 	/*
1799 	 * No big deal if we can't add this page to the page cache here.
1800 	 * READLINK will get the missing page from the server if needed.
1801 	 */
1802 	pagevec_init(&lru_pvec, 0);
1803 	if (!add_to_page_cache(page, dentry->d_inode->i_mapping, 0,
1804 							GFP_KERNEL)) {
1805 		pagevec_add(&lru_pvec, page);
1806 		pagevec_lru_add_file(&lru_pvec);
1807 		SetPageUptodate(page);
1808 		unlock_page(page);
1809 	} else
1810 		__free_page(page);
1811 
1812 	return 0;
1813 }
1814 EXPORT_SYMBOL_GPL(nfs_symlink);
1815 
1816 int
1817 nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1818 {
1819 	struct inode *inode = old_dentry->d_inode;
1820 	int error;
1821 
1822 	dfprintk(VFS, "NFS: link(%s/%s -> %s/%s)\n",
1823 		old_dentry->d_parent->d_name.name, old_dentry->d_name.name,
1824 		dentry->d_parent->d_name.name, dentry->d_name.name);
1825 
1826 	NFS_PROTO(inode)->return_delegation(inode);
1827 
1828 	d_drop(dentry);
1829 	error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name);
1830 	if (error == 0) {
1831 		ihold(inode);
1832 		d_add(dentry, inode);
1833 	}
1834 	return error;
1835 }
1836 EXPORT_SYMBOL_GPL(nfs_link);
1837 
1838 /*
1839  * RENAME
1840  * FIXME: Some nfsds, like the Linux user space nfsd, may generate a
1841  * different file handle for the same inode after a rename (e.g. when
1842  * moving to a different directory). A fail-safe method to do so would
1843  * be to look up old_dir/old_name, create a link to new_dir/new_name and
1844  * rename the old file using the sillyrename stuff. This way, the original
1845  * file in old_dir will go away when the last process iput()s the inode.
1846  *
1847  * FIXED.
1848  *
1849  * It actually works quite well. One needs to have the possibility for
1850  * at least one ".nfs..." file in each directory the file ever gets
1851  * moved or linked to which happens automagically with the new
1852  * implementation that only depends on the dcache stuff instead of
1853  * using the inode layer
1854  *
1855  * Unfortunately, things are a little more complicated than indicated
1856  * above. For a cross-directory move, we want to make sure we can get
1857  * rid of the old inode after the operation.  This means there must be
1858  * no pending writes (if it's a file), and the use count must be 1.
1859  * If these conditions are met, we can drop the dentries before doing
1860  * the rename.
1861  */
1862 int nfs_rename(struct inode *old_dir, struct dentry *old_dentry,
1863 		      struct inode *new_dir, struct dentry *new_dentry)
1864 {
1865 	struct inode *old_inode = old_dentry->d_inode;
1866 	struct inode *new_inode = new_dentry->d_inode;
1867 	struct dentry *dentry = NULL, *rehash = NULL;
1868 	int error = -EBUSY;
1869 
1870 	dfprintk(VFS, "NFS: rename(%s/%s -> %s/%s, ct=%d)\n",
1871 		 old_dentry->d_parent->d_name.name, old_dentry->d_name.name,
1872 		 new_dentry->d_parent->d_name.name, new_dentry->d_name.name,
1873 		 new_dentry->d_count);
1874 
1875 	/*
1876 	 * For non-directories, check whether the target is busy and if so,
1877 	 * make a copy of the dentry and then do a silly-rename. If the
1878 	 * silly-rename succeeds, the copied dentry is hashed and becomes
1879 	 * the new target.
1880 	 */
1881 	if (new_inode && !S_ISDIR(new_inode->i_mode)) {
1882 		/*
1883 		 * To prevent any new references to the target during the
1884 		 * rename, we unhash the dentry in advance.
1885 		 */
1886 		if (!d_unhashed(new_dentry)) {
1887 			d_drop(new_dentry);
1888 			rehash = new_dentry;
1889 		}
1890 
1891 		if (new_dentry->d_count > 2) {
1892 			int err;
1893 
1894 			/* copy the target dentry's name */
1895 			dentry = d_alloc(new_dentry->d_parent,
1896 					 &new_dentry->d_name);
1897 			if (!dentry)
1898 				goto out;
1899 
1900 			/* silly-rename the existing target ... */
1901 			err = nfs_sillyrename(new_dir, new_dentry);
1902 			if (err)
1903 				goto out;
1904 
1905 			new_dentry = dentry;
1906 			rehash = NULL;
1907 			new_inode = NULL;
1908 		}
1909 	}
1910 
1911 	NFS_PROTO(old_inode)->return_delegation(old_inode);
1912 	if (new_inode != NULL)
1913 		NFS_PROTO(new_inode)->return_delegation(new_inode);
1914 
1915 	error = NFS_PROTO(old_dir)->rename(old_dir, &old_dentry->d_name,
1916 					   new_dir, &new_dentry->d_name);
1917 	nfs_mark_for_revalidate(old_inode);
1918 out:
1919 	if (rehash)
1920 		d_rehash(rehash);
1921 	if (!error) {
1922 		if (new_inode != NULL)
1923 			nfs_drop_nlink(new_inode);
1924 		d_move(old_dentry, new_dentry);
1925 		nfs_set_verifier(new_dentry,
1926 					nfs_save_change_attribute(new_dir));
1927 	} else if (error == -ENOENT)
1928 		nfs_dentry_handle_enoent(old_dentry);
1929 
1930 	/* new dentry created? */
1931 	if (dentry)
1932 		dput(dentry);
1933 	return error;
1934 }
1935 EXPORT_SYMBOL_GPL(nfs_rename);
1936 
1937 static DEFINE_SPINLOCK(nfs_access_lru_lock);
1938 static LIST_HEAD(nfs_access_lru_list);
1939 static atomic_long_t nfs_access_nr_entries;
1940 
1941 static void nfs_access_free_entry(struct nfs_access_entry *entry)
1942 {
1943 	put_rpccred(entry->cred);
1944 	kfree(entry);
1945 	smp_mb__before_atomic_dec();
1946 	atomic_long_dec(&nfs_access_nr_entries);
1947 	smp_mb__after_atomic_dec();
1948 }
1949 
1950 static void nfs_access_free_list(struct list_head *head)
1951 {
1952 	struct nfs_access_entry *cache;
1953 
1954 	while (!list_empty(head)) {
1955 		cache = list_entry(head->next, struct nfs_access_entry, lru);
1956 		list_del(&cache->lru);
1957 		nfs_access_free_entry(cache);
1958 	}
1959 }
1960 
1961 int nfs_access_cache_shrinker(struct shrinker *shrink,
1962 			      struct shrink_control *sc)
1963 {
1964 	LIST_HEAD(head);
1965 	struct nfs_inode *nfsi, *next;
1966 	struct nfs_access_entry *cache;
1967 	int nr_to_scan = sc->nr_to_scan;
1968 	gfp_t gfp_mask = sc->gfp_mask;
1969 
1970 	if ((gfp_mask & GFP_KERNEL) != GFP_KERNEL)
1971 		return (nr_to_scan == 0) ? 0 : -1;
1972 
1973 	spin_lock(&nfs_access_lru_lock);
1974 	list_for_each_entry_safe(nfsi, next, &nfs_access_lru_list, access_cache_inode_lru) {
1975 		struct inode *inode;
1976 
1977 		if (nr_to_scan-- == 0)
1978 			break;
1979 		inode = &nfsi->vfs_inode;
1980 		spin_lock(&inode->i_lock);
1981 		if (list_empty(&nfsi->access_cache_entry_lru))
1982 			goto remove_lru_entry;
1983 		cache = list_entry(nfsi->access_cache_entry_lru.next,
1984 				struct nfs_access_entry, lru);
1985 		list_move(&cache->lru, &head);
1986 		rb_erase(&cache->rb_node, &nfsi->access_cache);
1987 		if (!list_empty(&nfsi->access_cache_entry_lru))
1988 			list_move_tail(&nfsi->access_cache_inode_lru,
1989 					&nfs_access_lru_list);
1990 		else {
1991 remove_lru_entry:
1992 			list_del_init(&nfsi->access_cache_inode_lru);
1993 			smp_mb__before_clear_bit();
1994 			clear_bit(NFS_INO_ACL_LRU_SET, &nfsi->flags);
1995 			smp_mb__after_clear_bit();
1996 		}
1997 		spin_unlock(&inode->i_lock);
1998 	}
1999 	spin_unlock(&nfs_access_lru_lock);
2000 	nfs_access_free_list(&head);
2001 	return (atomic_long_read(&nfs_access_nr_entries) / 100) * sysctl_vfs_cache_pressure;
2002 }
2003 
2004 static void __nfs_access_zap_cache(struct nfs_inode *nfsi, struct list_head *head)
2005 {
2006 	struct rb_root *root_node = &nfsi->access_cache;
2007 	struct rb_node *n;
2008 	struct nfs_access_entry *entry;
2009 
2010 	/* Unhook entries from the cache */
2011 	while ((n = rb_first(root_node)) != NULL) {
2012 		entry = rb_entry(n, struct nfs_access_entry, rb_node);
2013 		rb_erase(n, root_node);
2014 		list_move(&entry->lru, head);
2015 	}
2016 	nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS;
2017 }
2018 
2019 void nfs_access_zap_cache(struct inode *inode)
2020 {
2021 	LIST_HEAD(head);
2022 
2023 	if (test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags) == 0)
2024 		return;
2025 	/* Remove from global LRU init */
2026 	spin_lock(&nfs_access_lru_lock);
2027 	if (test_and_clear_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
2028 		list_del_init(&NFS_I(inode)->access_cache_inode_lru);
2029 
2030 	spin_lock(&inode->i_lock);
2031 	__nfs_access_zap_cache(NFS_I(inode), &head);
2032 	spin_unlock(&inode->i_lock);
2033 	spin_unlock(&nfs_access_lru_lock);
2034 	nfs_access_free_list(&head);
2035 }
2036 EXPORT_SYMBOL_GPL(nfs_access_zap_cache);
2037 
2038 static struct nfs_access_entry *nfs_access_search_rbtree(struct inode *inode, struct rpc_cred *cred)
2039 {
2040 	struct rb_node *n = NFS_I(inode)->access_cache.rb_node;
2041 	struct nfs_access_entry *entry;
2042 
2043 	while (n != NULL) {
2044 		entry = rb_entry(n, struct nfs_access_entry, rb_node);
2045 
2046 		if (cred < entry->cred)
2047 			n = n->rb_left;
2048 		else if (cred > entry->cred)
2049 			n = n->rb_right;
2050 		else
2051 			return entry;
2052 	}
2053 	return NULL;
2054 }
2055 
2056 static int nfs_access_get_cached(struct inode *inode, struct rpc_cred *cred, struct nfs_access_entry *res)
2057 {
2058 	struct nfs_inode *nfsi = NFS_I(inode);
2059 	struct nfs_access_entry *cache;
2060 	int err = -ENOENT;
2061 
2062 	spin_lock(&inode->i_lock);
2063 	if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS)
2064 		goto out_zap;
2065 	cache = nfs_access_search_rbtree(inode, cred);
2066 	if (cache == NULL)
2067 		goto out;
2068 	if (!nfs_have_delegated_attributes(inode) &&
2069 	    !time_in_range_open(jiffies, cache->jiffies, cache->jiffies + nfsi->attrtimeo))
2070 		goto out_stale;
2071 	res->jiffies = cache->jiffies;
2072 	res->cred = cache->cred;
2073 	res->mask = cache->mask;
2074 	list_move_tail(&cache->lru, &nfsi->access_cache_entry_lru);
2075 	err = 0;
2076 out:
2077 	spin_unlock(&inode->i_lock);
2078 	return err;
2079 out_stale:
2080 	rb_erase(&cache->rb_node, &nfsi->access_cache);
2081 	list_del(&cache->lru);
2082 	spin_unlock(&inode->i_lock);
2083 	nfs_access_free_entry(cache);
2084 	return -ENOENT;
2085 out_zap:
2086 	spin_unlock(&inode->i_lock);
2087 	nfs_access_zap_cache(inode);
2088 	return -ENOENT;
2089 }
2090 
2091 static void nfs_access_add_rbtree(struct inode *inode, struct nfs_access_entry *set)
2092 {
2093 	struct nfs_inode *nfsi = NFS_I(inode);
2094 	struct rb_root *root_node = &nfsi->access_cache;
2095 	struct rb_node **p = &root_node->rb_node;
2096 	struct rb_node *parent = NULL;
2097 	struct nfs_access_entry *entry;
2098 
2099 	spin_lock(&inode->i_lock);
2100 	while (*p != NULL) {
2101 		parent = *p;
2102 		entry = rb_entry(parent, struct nfs_access_entry, rb_node);
2103 
2104 		if (set->cred < entry->cred)
2105 			p = &parent->rb_left;
2106 		else if (set->cred > entry->cred)
2107 			p = &parent->rb_right;
2108 		else
2109 			goto found;
2110 	}
2111 	rb_link_node(&set->rb_node, parent, p);
2112 	rb_insert_color(&set->rb_node, root_node);
2113 	list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
2114 	spin_unlock(&inode->i_lock);
2115 	return;
2116 found:
2117 	rb_replace_node(parent, &set->rb_node, root_node);
2118 	list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
2119 	list_del(&entry->lru);
2120 	spin_unlock(&inode->i_lock);
2121 	nfs_access_free_entry(entry);
2122 }
2123 
2124 void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set)
2125 {
2126 	struct nfs_access_entry *cache = kmalloc(sizeof(*cache), GFP_KERNEL);
2127 	if (cache == NULL)
2128 		return;
2129 	RB_CLEAR_NODE(&cache->rb_node);
2130 	cache->jiffies = set->jiffies;
2131 	cache->cred = get_rpccred(set->cred);
2132 	cache->mask = set->mask;
2133 
2134 	nfs_access_add_rbtree(inode, cache);
2135 
2136 	/* Update accounting */
2137 	smp_mb__before_atomic_inc();
2138 	atomic_long_inc(&nfs_access_nr_entries);
2139 	smp_mb__after_atomic_inc();
2140 
2141 	/* Add inode to global LRU list */
2142 	if (!test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) {
2143 		spin_lock(&nfs_access_lru_lock);
2144 		if (!test_and_set_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
2145 			list_add_tail(&NFS_I(inode)->access_cache_inode_lru,
2146 					&nfs_access_lru_list);
2147 		spin_unlock(&nfs_access_lru_lock);
2148 	}
2149 }
2150 EXPORT_SYMBOL_GPL(nfs_access_add_cache);
2151 
2152 void nfs_access_set_mask(struct nfs_access_entry *entry, u32 access_result)
2153 {
2154 	entry->mask = 0;
2155 	if (access_result & NFS4_ACCESS_READ)
2156 		entry->mask |= MAY_READ;
2157 	if (access_result &
2158 	    (NFS4_ACCESS_MODIFY | NFS4_ACCESS_EXTEND | NFS4_ACCESS_DELETE))
2159 		entry->mask |= MAY_WRITE;
2160 	if (access_result & (NFS4_ACCESS_LOOKUP|NFS4_ACCESS_EXECUTE))
2161 		entry->mask |= MAY_EXEC;
2162 }
2163 EXPORT_SYMBOL_GPL(nfs_access_set_mask);
2164 
2165 static int nfs_do_access(struct inode *inode, struct rpc_cred *cred, int mask)
2166 {
2167 	struct nfs_access_entry cache;
2168 	int status;
2169 
2170 	status = nfs_access_get_cached(inode, cred, &cache);
2171 	if (status == 0)
2172 		goto out;
2173 
2174 	/* Be clever: ask server to check for all possible rights */
2175 	cache.mask = MAY_EXEC | MAY_WRITE | MAY_READ;
2176 	cache.cred = cred;
2177 	cache.jiffies = jiffies;
2178 	status = NFS_PROTO(inode)->access(inode, &cache);
2179 	if (status != 0) {
2180 		if (status == -ESTALE) {
2181 			nfs_zap_caches(inode);
2182 			if (!S_ISDIR(inode->i_mode))
2183 				set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
2184 		}
2185 		return status;
2186 	}
2187 	nfs_access_add_cache(inode, &cache);
2188 out:
2189 	if ((mask & ~cache.mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
2190 		return 0;
2191 	return -EACCES;
2192 }
2193 
2194 static int nfs_open_permission_mask(int openflags)
2195 {
2196 	int mask = 0;
2197 
2198 	if (openflags & __FMODE_EXEC) {
2199 		/* ONLY check exec rights */
2200 		mask = MAY_EXEC;
2201 	} else {
2202 		if ((openflags & O_ACCMODE) != O_WRONLY)
2203 			mask |= MAY_READ;
2204 		if ((openflags & O_ACCMODE) != O_RDONLY)
2205 			mask |= MAY_WRITE;
2206 	}
2207 
2208 	return mask;
2209 }
2210 
2211 int nfs_may_open(struct inode *inode, struct rpc_cred *cred, int openflags)
2212 {
2213 	return nfs_do_access(inode, cred, nfs_open_permission_mask(openflags));
2214 }
2215 EXPORT_SYMBOL_GPL(nfs_may_open);
2216 
2217 int nfs_permission(struct inode *inode, int mask)
2218 {
2219 	struct rpc_cred *cred;
2220 	int res = 0;
2221 
2222 	if (mask & MAY_NOT_BLOCK)
2223 		return -ECHILD;
2224 
2225 	nfs_inc_stats(inode, NFSIOS_VFSACCESS);
2226 
2227 	if ((mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
2228 		goto out;
2229 	/* Is this sys_access() ? */
2230 	if (mask & (MAY_ACCESS | MAY_CHDIR))
2231 		goto force_lookup;
2232 
2233 	switch (inode->i_mode & S_IFMT) {
2234 		case S_IFLNK:
2235 			goto out;
2236 		case S_IFREG:
2237 			/* NFSv4 has atomic_open... */
2238 			if (nfs_server_capable(inode, NFS_CAP_ATOMIC_OPEN)
2239 					&& (mask & MAY_OPEN)
2240 					&& !(mask & MAY_EXEC))
2241 				goto out;
2242 			break;
2243 		case S_IFDIR:
2244 			/*
2245 			 * Optimize away all write operations, since the server
2246 			 * will check permissions when we perform the op.
2247 			 */
2248 			if ((mask & MAY_WRITE) && !(mask & MAY_READ))
2249 				goto out;
2250 	}
2251 
2252 force_lookup:
2253 	if (!NFS_PROTO(inode)->access)
2254 		goto out_notsup;
2255 
2256 	cred = rpc_lookup_cred();
2257 	if (!IS_ERR(cred)) {
2258 		res = nfs_do_access(inode, cred, mask);
2259 		put_rpccred(cred);
2260 	} else
2261 		res = PTR_ERR(cred);
2262 out:
2263 	if (!res && (mask & MAY_EXEC) && !execute_ok(inode))
2264 		res = -EACCES;
2265 
2266 	dfprintk(VFS, "NFS: permission(%s/%ld), mask=0x%x, res=%d\n",
2267 		inode->i_sb->s_id, inode->i_ino, mask, res);
2268 	return res;
2269 out_notsup:
2270 	res = nfs_revalidate_inode(NFS_SERVER(inode), inode);
2271 	if (res == 0)
2272 		res = generic_permission(inode, mask);
2273 	goto out;
2274 }
2275 EXPORT_SYMBOL_GPL(nfs_permission);
2276 
2277 /*
2278  * Local variables:
2279  *  version-control: t
2280  *  kept-new-versions: 5
2281  * End:
2282  */
2283