xref: /openbmc/linux/fs/nfs/dir.c (revision 1ca42382afd67bf58523d36b00fb4ff487d8173b)
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/swap.h>
37 #include <linux/sched.h>
38 #include <linux/kmemleak.h>
39 #include <linux/xattr.h>
40 
41 #include "delegation.h"
42 #include "iostat.h"
43 #include "internal.h"
44 #include "fscache.h"
45 
46 #include "nfstrace.h"
47 
48 /* #define NFS_DEBUG_VERBOSE 1 */
49 
50 static int nfs_opendir(struct inode *, struct file *);
51 static int nfs_closedir(struct inode *, struct file *);
52 static int nfs_readdir(struct file *, struct dir_context *);
53 static int nfs_fsync_dir(struct file *, loff_t, loff_t, int);
54 static loff_t nfs_llseek_dir(struct file *, loff_t, int);
55 static void nfs_readdir_clear_array(struct page*);
56 
57 const struct file_operations nfs_dir_operations = {
58 	.llseek		= nfs_llseek_dir,
59 	.read		= generic_read_dir,
60 	.iterate	= nfs_readdir,
61 	.open		= nfs_opendir,
62 	.release	= nfs_closedir,
63 	.fsync		= nfs_fsync_dir,
64 };
65 
66 const struct address_space_operations nfs_dir_aops = {
67 	.freepage = nfs_readdir_clear_array,
68 };
69 
70 static struct nfs_open_dir_context *alloc_nfs_open_dir_context(struct inode *dir, struct rpc_cred *cred)
71 {
72 	struct nfs_open_dir_context *ctx;
73 	ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
74 	if (ctx != NULL) {
75 		ctx->duped = 0;
76 		ctx->attr_gencount = NFS_I(dir)->attr_gencount;
77 		ctx->dir_cookie = 0;
78 		ctx->dup_cookie = 0;
79 		ctx->cred = get_rpccred(cred);
80 		return ctx;
81 	}
82 	return  ERR_PTR(-ENOMEM);
83 }
84 
85 static void put_nfs_open_dir_context(struct nfs_open_dir_context *ctx)
86 {
87 	put_rpccred(ctx->cred);
88 	kfree(ctx);
89 }
90 
91 /*
92  * Open file
93  */
94 static int
95 nfs_opendir(struct inode *inode, struct file *filp)
96 {
97 	int res = 0;
98 	struct nfs_open_dir_context *ctx;
99 	struct rpc_cred *cred;
100 
101 	dfprintk(FILE, "NFS: open dir(%s/%s)\n",
102 			filp->f_path.dentry->d_parent->d_name.name,
103 			filp->f_path.dentry->d_name.name);
104 
105 	nfs_inc_stats(inode, NFSIOS_VFSOPEN);
106 
107 	cred = rpc_lookup_cred();
108 	if (IS_ERR(cred))
109 		return PTR_ERR(cred);
110 	ctx = alloc_nfs_open_dir_context(inode, cred);
111 	if (IS_ERR(ctx)) {
112 		res = PTR_ERR(ctx);
113 		goto out;
114 	}
115 	filp->private_data = ctx;
116 	if (filp->f_path.dentry == filp->f_path.mnt->mnt_root) {
117 		/* This is a mountpoint, so d_revalidate will never
118 		 * have been called, so we need to refresh the
119 		 * inode (for close-open consistency) ourselves.
120 		 */
121 		__nfs_revalidate_inode(NFS_SERVER(inode), inode);
122 	}
123 out:
124 	put_rpccred(cred);
125 	return res;
126 }
127 
128 static int
129 nfs_closedir(struct inode *inode, struct file *filp)
130 {
131 	put_nfs_open_dir_context(filp->private_data);
132 	return 0;
133 }
134 
135 struct nfs_cache_array_entry {
136 	u64 cookie;
137 	u64 ino;
138 	struct qstr string;
139 	unsigned char d_type;
140 };
141 
142 struct nfs_cache_array {
143 	int size;
144 	int eof_index;
145 	u64 last_cookie;
146 	struct nfs_cache_array_entry array[0];
147 };
148 
149 typedef int (*decode_dirent_t)(struct xdr_stream *, struct nfs_entry *, int);
150 typedef struct {
151 	struct file	*file;
152 	struct page	*page;
153 	struct dir_context *ctx;
154 	unsigned long	page_index;
155 	u64		*dir_cookie;
156 	u64		last_cookie;
157 	loff_t		current_index;
158 	decode_dirent_t	decode;
159 
160 	unsigned long	timestamp;
161 	unsigned long	gencount;
162 	unsigned int	cache_entry_index;
163 	unsigned int	plus:1;
164 	unsigned int	eof:1;
165 } nfs_readdir_descriptor_t;
166 
167 /*
168  * The caller is responsible for calling nfs_readdir_release_array(page)
169  */
170 static
171 struct nfs_cache_array *nfs_readdir_get_array(struct page *page)
172 {
173 	void *ptr;
174 	if (page == NULL)
175 		return ERR_PTR(-EIO);
176 	ptr = kmap(page);
177 	if (ptr == NULL)
178 		return ERR_PTR(-ENOMEM);
179 	return ptr;
180 }
181 
182 static
183 void nfs_readdir_release_array(struct page *page)
184 {
185 	kunmap(page);
186 }
187 
188 /*
189  * we are freeing strings created by nfs_add_to_readdir_array()
190  */
191 static
192 void nfs_readdir_clear_array(struct page *page)
193 {
194 	struct nfs_cache_array *array;
195 	int i;
196 
197 	array = kmap_atomic(page);
198 	for (i = 0; i < array->size; i++)
199 		kfree(array->array[i].string.name);
200 	kunmap_atomic(array);
201 }
202 
203 /*
204  * the caller is responsible for freeing qstr.name
205  * when called by nfs_readdir_add_to_array, the strings will be freed in
206  * nfs_clear_readdir_array()
207  */
208 static
209 int nfs_readdir_make_qstr(struct qstr *string, const char *name, unsigned int len)
210 {
211 	string->len = len;
212 	string->name = kmemdup(name, len, GFP_KERNEL);
213 	if (string->name == NULL)
214 		return -ENOMEM;
215 	/*
216 	 * Avoid a kmemleak false positive. The pointer to the name is stored
217 	 * in a page cache page which kmemleak does not scan.
218 	 */
219 	kmemleak_not_leak(string->name);
220 	string->hash = full_name_hash(name, len);
221 	return 0;
222 }
223 
224 static
225 int nfs_readdir_add_to_array(struct nfs_entry *entry, struct page *page)
226 {
227 	struct nfs_cache_array *array = nfs_readdir_get_array(page);
228 	struct nfs_cache_array_entry *cache_entry;
229 	int ret;
230 
231 	if (IS_ERR(array))
232 		return PTR_ERR(array);
233 
234 	cache_entry = &array->array[array->size];
235 
236 	/* Check that this entry lies within the page bounds */
237 	ret = -ENOSPC;
238 	if ((char *)&cache_entry[1] - (char *)page_address(page) > PAGE_SIZE)
239 		goto out;
240 
241 	cache_entry->cookie = entry->prev_cookie;
242 	cache_entry->ino = entry->ino;
243 	cache_entry->d_type = entry->d_type;
244 	ret = nfs_readdir_make_qstr(&cache_entry->string, entry->name, entry->len);
245 	if (ret)
246 		goto out;
247 	array->last_cookie = entry->cookie;
248 	array->size++;
249 	if (entry->eof != 0)
250 		array->eof_index = array->size;
251 out:
252 	nfs_readdir_release_array(page);
253 	return ret;
254 }
255 
256 static
257 int nfs_readdir_search_for_pos(struct nfs_cache_array *array, nfs_readdir_descriptor_t *desc)
258 {
259 	loff_t diff = desc->ctx->pos - desc->current_index;
260 	unsigned int index;
261 
262 	if (diff < 0)
263 		goto out_eof;
264 	if (diff >= array->size) {
265 		if (array->eof_index >= 0)
266 			goto out_eof;
267 		return -EAGAIN;
268 	}
269 
270 	index = (unsigned int)diff;
271 	*desc->dir_cookie = array->array[index].cookie;
272 	desc->cache_entry_index = index;
273 	return 0;
274 out_eof:
275 	desc->eof = 1;
276 	return -EBADCOOKIE;
277 }
278 
279 static
280 int nfs_readdir_search_for_cookie(struct nfs_cache_array *array, nfs_readdir_descriptor_t *desc)
281 {
282 	int i;
283 	loff_t new_pos;
284 	int status = -EAGAIN;
285 
286 	for (i = 0; i < array->size; i++) {
287 		if (array->array[i].cookie == *desc->dir_cookie) {
288 			struct nfs_inode *nfsi = NFS_I(file_inode(desc->file));
289 			struct nfs_open_dir_context *ctx = desc->file->private_data;
290 
291 			new_pos = desc->current_index + i;
292 			if (ctx->attr_gencount != nfsi->attr_gencount
293 			    || (nfsi->cache_validity & (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA))) {
294 				ctx->duped = 0;
295 				ctx->attr_gencount = nfsi->attr_gencount;
296 			} else if (new_pos < desc->ctx->pos) {
297 				if (ctx->duped > 0
298 				    && ctx->dup_cookie == *desc->dir_cookie) {
299 					if (printk_ratelimit()) {
300 						pr_notice("NFS: directory %s/%s contains a readdir loop."
301 								"Please contact your server vendor.  "
302 								"The file: %s has duplicate cookie %llu\n",
303 								desc->file->f_dentry->d_parent->d_name.name,
304 								desc->file->f_dentry->d_name.name,
305 								array->array[i].string.name,
306 								*desc->dir_cookie);
307 					}
308 					status = -ELOOP;
309 					goto out;
310 				}
311 				ctx->dup_cookie = *desc->dir_cookie;
312 				ctx->duped = -1;
313 			}
314 			desc->ctx->pos = new_pos;
315 			desc->cache_entry_index = i;
316 			return 0;
317 		}
318 	}
319 	if (array->eof_index >= 0) {
320 		status = -EBADCOOKIE;
321 		if (*desc->dir_cookie == array->last_cookie)
322 			desc->eof = 1;
323 	}
324 out:
325 	return status;
326 }
327 
328 static
329 int nfs_readdir_search_array(nfs_readdir_descriptor_t *desc)
330 {
331 	struct nfs_cache_array *array;
332 	int status;
333 
334 	array = nfs_readdir_get_array(desc->page);
335 	if (IS_ERR(array)) {
336 		status = PTR_ERR(array);
337 		goto out;
338 	}
339 
340 	if (*desc->dir_cookie == 0)
341 		status = nfs_readdir_search_for_pos(array, desc);
342 	else
343 		status = nfs_readdir_search_for_cookie(array, desc);
344 
345 	if (status == -EAGAIN) {
346 		desc->last_cookie = array->last_cookie;
347 		desc->current_index += array->size;
348 		desc->page_index++;
349 	}
350 	nfs_readdir_release_array(desc->page);
351 out:
352 	return status;
353 }
354 
355 /* Fill a page with xdr information before transferring to the cache page */
356 static
357 int nfs_readdir_xdr_filler(struct page **pages, nfs_readdir_descriptor_t *desc,
358 			struct nfs_entry *entry, struct file *file, struct inode *inode)
359 {
360 	struct nfs_open_dir_context *ctx = file->private_data;
361 	struct rpc_cred	*cred = ctx->cred;
362 	unsigned long	timestamp, gencount;
363 	int		error;
364 
365  again:
366 	timestamp = jiffies;
367 	gencount = nfs_inc_attr_generation_counter();
368 	error = NFS_PROTO(inode)->readdir(file->f_path.dentry, cred, entry->cookie, pages,
369 					  NFS_SERVER(inode)->dtsize, desc->plus);
370 	if (error < 0) {
371 		/* We requested READDIRPLUS, but the server doesn't grok it */
372 		if (error == -ENOTSUPP && desc->plus) {
373 			NFS_SERVER(inode)->caps &= ~NFS_CAP_READDIRPLUS;
374 			clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
375 			desc->plus = 0;
376 			goto again;
377 		}
378 		goto error;
379 	}
380 	desc->timestamp = timestamp;
381 	desc->gencount = gencount;
382 error:
383 	return error;
384 }
385 
386 static int xdr_decode(nfs_readdir_descriptor_t *desc,
387 		      struct nfs_entry *entry, struct xdr_stream *xdr)
388 {
389 	int error;
390 
391 	error = desc->decode(xdr, entry, desc->plus);
392 	if (error)
393 		return error;
394 	entry->fattr->time_start = desc->timestamp;
395 	entry->fattr->gencount = desc->gencount;
396 	return 0;
397 }
398 
399 static
400 int nfs_same_file(struct dentry *dentry, struct nfs_entry *entry)
401 {
402 	if (dentry->d_inode == NULL)
403 		goto different;
404 	if (nfs_compare_fh(entry->fh, NFS_FH(dentry->d_inode)) != 0)
405 		goto different;
406 	return 1;
407 different:
408 	return 0;
409 }
410 
411 static
412 bool nfs_use_readdirplus(struct inode *dir, struct dir_context *ctx)
413 {
414 	if (!nfs_server_capable(dir, NFS_CAP_READDIRPLUS))
415 		return false;
416 	if (test_and_clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(dir)->flags))
417 		return true;
418 	if (ctx->pos == 0)
419 		return true;
420 	return false;
421 }
422 
423 /*
424  * This function is called by the lookup code to request the use of
425  * readdirplus to accelerate any future lookups in the same
426  * directory.
427  */
428 static
429 void nfs_advise_use_readdirplus(struct inode *dir)
430 {
431 	set_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(dir)->flags);
432 }
433 
434 static
435 void nfs_prime_dcache(struct dentry *parent, struct nfs_entry *entry)
436 {
437 	struct qstr filename = QSTR_INIT(entry->name, entry->len);
438 	struct dentry *dentry;
439 	struct dentry *alias;
440 	struct inode *dir = parent->d_inode;
441 	struct inode *inode;
442 	int status;
443 
444 	if (filename.name[0] == '.') {
445 		if (filename.len == 1)
446 			return;
447 		if (filename.len == 2 && filename.name[1] == '.')
448 			return;
449 	}
450 	filename.hash = full_name_hash(filename.name, filename.len);
451 
452 	dentry = d_lookup(parent, &filename);
453 	if (dentry != NULL) {
454 		if (nfs_same_file(dentry, entry)) {
455 			nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
456 			status = nfs_refresh_inode(dentry->d_inode, entry->fattr);
457 			if (!status)
458 				nfs_setsecurity(dentry->d_inode, entry->fattr, entry->label);
459 			goto out;
460 		} else {
461 			if (d_invalidate(dentry) != 0)
462 				goto out;
463 			dput(dentry);
464 		}
465 	}
466 
467 	dentry = d_alloc(parent, &filename);
468 	if (dentry == NULL)
469 		return;
470 
471 	inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr, entry->label);
472 	if (IS_ERR(inode))
473 		goto out;
474 
475 	alias = d_materialise_unique(dentry, inode);
476 	if (IS_ERR(alias))
477 		goto out;
478 	else if (alias) {
479 		nfs_set_verifier(alias, nfs_save_change_attribute(dir));
480 		dput(alias);
481 	} else
482 		nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
483 
484 out:
485 	dput(dentry);
486 }
487 
488 /* Perform conversion from xdr to cache array */
489 static
490 int nfs_readdir_page_filler(nfs_readdir_descriptor_t *desc, struct nfs_entry *entry,
491 				struct page **xdr_pages, struct page *page, unsigned int buflen)
492 {
493 	struct xdr_stream stream;
494 	struct xdr_buf buf;
495 	struct page *scratch;
496 	struct nfs_cache_array *array;
497 	unsigned int count = 0;
498 	int status;
499 
500 	scratch = alloc_page(GFP_KERNEL);
501 	if (scratch == NULL)
502 		return -ENOMEM;
503 
504 	xdr_init_decode_pages(&stream, &buf, xdr_pages, buflen);
505 	xdr_set_scratch_buffer(&stream, page_address(scratch), PAGE_SIZE);
506 
507 	do {
508 		status = xdr_decode(desc, entry, &stream);
509 		if (status != 0) {
510 			if (status == -EAGAIN)
511 				status = 0;
512 			break;
513 		}
514 
515 		count++;
516 
517 		if (desc->plus != 0)
518 			nfs_prime_dcache(desc->file->f_path.dentry, entry);
519 
520 		status = nfs_readdir_add_to_array(entry, page);
521 		if (status != 0)
522 			break;
523 	} while (!entry->eof);
524 
525 	if (count == 0 || (status == -EBADCOOKIE && entry->eof != 0)) {
526 		array = nfs_readdir_get_array(page);
527 		if (!IS_ERR(array)) {
528 			array->eof_index = array->size;
529 			status = 0;
530 			nfs_readdir_release_array(page);
531 		} else
532 			status = PTR_ERR(array);
533 	}
534 
535 	put_page(scratch);
536 	return status;
537 }
538 
539 static
540 void nfs_readdir_free_pagearray(struct page **pages, unsigned int npages)
541 {
542 	unsigned int i;
543 	for (i = 0; i < npages; i++)
544 		put_page(pages[i]);
545 }
546 
547 static
548 void nfs_readdir_free_large_page(void *ptr, struct page **pages,
549 		unsigned int npages)
550 {
551 	nfs_readdir_free_pagearray(pages, npages);
552 }
553 
554 /*
555  * nfs_readdir_large_page will allocate pages that must be freed with a call
556  * to nfs_readdir_free_large_page
557  */
558 static
559 int nfs_readdir_large_page(struct page **pages, unsigned int npages)
560 {
561 	unsigned int i;
562 
563 	for (i = 0; i < npages; i++) {
564 		struct page *page = alloc_page(GFP_KERNEL);
565 		if (page == NULL)
566 			goto out_freepages;
567 		pages[i] = page;
568 	}
569 	return 0;
570 
571 out_freepages:
572 	nfs_readdir_free_pagearray(pages, i);
573 	return -ENOMEM;
574 }
575 
576 static
577 int nfs_readdir_xdr_to_array(nfs_readdir_descriptor_t *desc, struct page *page, struct inode *inode)
578 {
579 	struct page *pages[NFS_MAX_READDIR_PAGES];
580 	void *pages_ptr = NULL;
581 	struct nfs_entry entry;
582 	struct file	*file = desc->file;
583 	struct nfs_cache_array *array;
584 	int status = -ENOMEM;
585 	unsigned int array_size = ARRAY_SIZE(pages);
586 
587 	entry.prev_cookie = 0;
588 	entry.cookie = desc->last_cookie;
589 	entry.eof = 0;
590 	entry.fh = nfs_alloc_fhandle();
591 	entry.fattr = nfs_alloc_fattr();
592 	entry.server = NFS_SERVER(inode);
593 	if (entry.fh == NULL || entry.fattr == NULL)
594 		goto out;
595 
596 	entry.label = nfs4_label_alloc(NFS_SERVER(inode), GFP_NOWAIT);
597 	if (IS_ERR(entry.label)) {
598 		status = PTR_ERR(entry.label);
599 		goto out;
600 	}
601 
602 	array = nfs_readdir_get_array(page);
603 	if (IS_ERR(array)) {
604 		status = PTR_ERR(array);
605 		goto out_label_free;
606 	}
607 	memset(array, 0, sizeof(struct nfs_cache_array));
608 	array->eof_index = -1;
609 
610 	status = nfs_readdir_large_page(pages, array_size);
611 	if (status < 0)
612 		goto out_release_array;
613 	do {
614 		unsigned int pglen;
615 		status = nfs_readdir_xdr_filler(pages, desc, &entry, file, inode);
616 
617 		if (status < 0)
618 			break;
619 		pglen = status;
620 		status = nfs_readdir_page_filler(desc, &entry, pages, page, pglen);
621 		if (status < 0) {
622 			if (status == -ENOSPC)
623 				status = 0;
624 			break;
625 		}
626 	} while (array->eof_index < 0);
627 
628 	nfs_readdir_free_large_page(pages_ptr, pages, array_size);
629 out_release_array:
630 	nfs_readdir_release_array(page);
631 out_label_free:
632 	nfs4_label_free(entry.label);
633 out:
634 	nfs_free_fattr(entry.fattr);
635 	nfs_free_fhandle(entry.fh);
636 	return status;
637 }
638 
639 /*
640  * Now we cache directories properly, by converting xdr information
641  * to an array that can be used for lookups later.  This results in
642  * fewer cache pages, since we can store more information on each page.
643  * We only need to convert from xdr once so future lookups are much simpler
644  */
645 static
646 int nfs_readdir_filler(nfs_readdir_descriptor_t *desc, struct page* page)
647 {
648 	struct inode	*inode = file_inode(desc->file);
649 	int ret;
650 
651 	ret = nfs_readdir_xdr_to_array(desc, page, inode);
652 	if (ret < 0)
653 		goto error;
654 	SetPageUptodate(page);
655 
656 	if (invalidate_inode_pages2_range(inode->i_mapping, page->index + 1, -1) < 0) {
657 		/* Should never happen */
658 		nfs_zap_mapping(inode, inode->i_mapping);
659 	}
660 	unlock_page(page);
661 	return 0;
662  error:
663 	unlock_page(page);
664 	return ret;
665 }
666 
667 static
668 void cache_page_release(nfs_readdir_descriptor_t *desc)
669 {
670 	if (!desc->page->mapping)
671 		nfs_readdir_clear_array(desc->page);
672 	page_cache_release(desc->page);
673 	desc->page = NULL;
674 }
675 
676 static
677 struct page *get_cache_page(nfs_readdir_descriptor_t *desc)
678 {
679 	return read_cache_page(file_inode(desc->file)->i_mapping,
680 			desc->page_index, (filler_t *)nfs_readdir_filler, desc);
681 }
682 
683 /*
684  * Returns 0 if desc->dir_cookie was found on page desc->page_index
685  */
686 static
687 int find_cache_page(nfs_readdir_descriptor_t *desc)
688 {
689 	int res;
690 
691 	desc->page = get_cache_page(desc);
692 	if (IS_ERR(desc->page))
693 		return PTR_ERR(desc->page);
694 
695 	res = nfs_readdir_search_array(desc);
696 	if (res != 0)
697 		cache_page_release(desc);
698 	return res;
699 }
700 
701 /* Search for desc->dir_cookie from the beginning of the page cache */
702 static inline
703 int readdir_search_pagecache(nfs_readdir_descriptor_t *desc)
704 {
705 	int res;
706 
707 	if (desc->page_index == 0) {
708 		desc->current_index = 0;
709 		desc->last_cookie = 0;
710 	}
711 	do {
712 		res = find_cache_page(desc);
713 	} while (res == -EAGAIN);
714 	return res;
715 }
716 
717 /*
718  * Once we've found the start of the dirent within a page: fill 'er up...
719  */
720 static
721 int nfs_do_filldir(nfs_readdir_descriptor_t *desc)
722 {
723 	struct file	*file = desc->file;
724 	int i = 0;
725 	int res = 0;
726 	struct nfs_cache_array *array = NULL;
727 	struct nfs_open_dir_context *ctx = file->private_data;
728 
729 	array = nfs_readdir_get_array(desc->page);
730 	if (IS_ERR(array)) {
731 		res = PTR_ERR(array);
732 		goto out;
733 	}
734 
735 	for (i = desc->cache_entry_index; i < array->size; i++) {
736 		struct nfs_cache_array_entry *ent;
737 
738 		ent = &array->array[i];
739 		if (!dir_emit(desc->ctx, ent->string.name, ent->string.len,
740 		    nfs_compat_user_ino64(ent->ino), ent->d_type)) {
741 			desc->eof = 1;
742 			break;
743 		}
744 		desc->ctx->pos++;
745 		if (i < (array->size-1))
746 			*desc->dir_cookie = array->array[i+1].cookie;
747 		else
748 			*desc->dir_cookie = array->last_cookie;
749 		if (ctx->duped != 0)
750 			ctx->duped = 1;
751 	}
752 	if (array->eof_index >= 0)
753 		desc->eof = 1;
754 
755 	nfs_readdir_release_array(desc->page);
756 out:
757 	cache_page_release(desc);
758 	dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n",
759 			(unsigned long long)*desc->dir_cookie, res);
760 	return res;
761 }
762 
763 /*
764  * If we cannot find a cookie in our cache, we suspect that this is
765  * because it points to a deleted file, so we ask the server to return
766  * whatever it thinks is the next entry. We then feed this to filldir.
767  * If all goes well, we should then be able to find our way round the
768  * cache on the next call to readdir_search_pagecache();
769  *
770  * NOTE: we cannot add the anonymous page to the pagecache because
771  *	 the data it contains might not be page aligned. Besides,
772  *	 we should already have a complete representation of the
773  *	 directory in the page cache by the time we get here.
774  */
775 static inline
776 int uncached_readdir(nfs_readdir_descriptor_t *desc)
777 {
778 	struct page	*page = NULL;
779 	int		status;
780 	struct inode *inode = file_inode(desc->file);
781 	struct nfs_open_dir_context *ctx = desc->file->private_data;
782 
783 	dfprintk(DIRCACHE, "NFS: uncached_readdir() searching for cookie %Lu\n",
784 			(unsigned long long)*desc->dir_cookie);
785 
786 	page = alloc_page(GFP_HIGHUSER);
787 	if (!page) {
788 		status = -ENOMEM;
789 		goto out;
790 	}
791 
792 	desc->page_index = 0;
793 	desc->last_cookie = *desc->dir_cookie;
794 	desc->page = page;
795 	ctx->duped = 0;
796 
797 	status = nfs_readdir_xdr_to_array(desc, page, inode);
798 	if (status < 0)
799 		goto out_release;
800 
801 	status = nfs_do_filldir(desc);
802 
803  out:
804 	dfprintk(DIRCACHE, "NFS: %s: returns %d\n",
805 			__func__, status);
806 	return status;
807  out_release:
808 	cache_page_release(desc);
809 	goto out;
810 }
811 
812 /* The file offset position represents the dirent entry number.  A
813    last cookie cache takes care of the common case of reading the
814    whole directory.
815  */
816 static int nfs_readdir(struct file *file, struct dir_context *ctx)
817 {
818 	struct dentry	*dentry = file->f_path.dentry;
819 	struct inode	*inode = dentry->d_inode;
820 	nfs_readdir_descriptor_t my_desc,
821 			*desc = &my_desc;
822 	struct nfs_open_dir_context *dir_ctx = file->private_data;
823 	int res = 0;
824 
825 	dfprintk(FILE, "NFS: readdir(%s/%s) starting at cookie %llu\n",
826 			dentry->d_parent->d_name.name, dentry->d_name.name,
827 			(long long)ctx->pos);
828 	nfs_inc_stats(inode, NFSIOS_VFSGETDENTS);
829 
830 	/*
831 	 * ctx->pos points to the dirent entry number.
832 	 * *desc->dir_cookie has the cookie for the next entry. We have
833 	 * to either find the entry with the appropriate number or
834 	 * revalidate the cookie.
835 	 */
836 	memset(desc, 0, sizeof(*desc));
837 
838 	desc->file = file;
839 	desc->ctx = ctx;
840 	desc->dir_cookie = &dir_ctx->dir_cookie;
841 	desc->decode = NFS_PROTO(inode)->decode_dirent;
842 	desc->plus = nfs_use_readdirplus(inode, ctx) ? 1 : 0;
843 
844 	nfs_block_sillyrename(dentry);
845 	if (ctx->pos == 0 || nfs_attribute_cache_expired(inode))
846 		res = nfs_revalidate_mapping(inode, file->f_mapping);
847 	if (res < 0)
848 		goto out;
849 
850 	do {
851 		res = readdir_search_pagecache(desc);
852 
853 		if (res == -EBADCOOKIE) {
854 			res = 0;
855 			/* This means either end of directory */
856 			if (*desc->dir_cookie && desc->eof == 0) {
857 				/* Or that the server has 'lost' a cookie */
858 				res = uncached_readdir(desc);
859 				if (res == 0)
860 					continue;
861 			}
862 			break;
863 		}
864 		if (res == -ETOOSMALL && desc->plus) {
865 			clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
866 			nfs_zap_caches(inode);
867 			desc->page_index = 0;
868 			desc->plus = 0;
869 			desc->eof = 0;
870 			continue;
871 		}
872 		if (res < 0)
873 			break;
874 
875 		res = nfs_do_filldir(desc);
876 		if (res < 0)
877 			break;
878 	} while (!desc->eof);
879 out:
880 	nfs_unblock_sillyrename(dentry);
881 	if (res > 0)
882 		res = 0;
883 	dfprintk(FILE, "NFS: readdir(%s/%s) returns %d\n",
884 			dentry->d_parent->d_name.name, dentry->d_name.name,
885 			res);
886 	return res;
887 }
888 
889 static loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int whence)
890 {
891 	struct dentry *dentry = filp->f_path.dentry;
892 	struct inode *inode = dentry->d_inode;
893 	struct nfs_open_dir_context *dir_ctx = filp->private_data;
894 
895 	dfprintk(FILE, "NFS: llseek dir(%s/%s, %lld, %d)\n",
896 			dentry->d_parent->d_name.name,
897 			dentry->d_name.name,
898 			offset, whence);
899 
900 	mutex_lock(&inode->i_mutex);
901 	switch (whence) {
902 		case 1:
903 			offset += filp->f_pos;
904 		case 0:
905 			if (offset >= 0)
906 				break;
907 		default:
908 			offset = -EINVAL;
909 			goto out;
910 	}
911 	if (offset != filp->f_pos) {
912 		filp->f_pos = offset;
913 		dir_ctx->dir_cookie = 0;
914 		dir_ctx->duped = 0;
915 	}
916 out:
917 	mutex_unlock(&inode->i_mutex);
918 	return offset;
919 }
920 
921 /*
922  * All directory operations under NFS are synchronous, so fsync()
923  * is a dummy operation.
924  */
925 static int nfs_fsync_dir(struct file *filp, loff_t start, loff_t end,
926 			 int datasync)
927 {
928 	struct dentry *dentry = filp->f_path.dentry;
929 	struct inode *inode = dentry->d_inode;
930 
931 	dfprintk(FILE, "NFS: fsync dir(%s/%s) datasync %d\n",
932 			dentry->d_parent->d_name.name, dentry->d_name.name,
933 			datasync);
934 
935 	mutex_lock(&inode->i_mutex);
936 	nfs_inc_stats(dentry->d_inode, NFSIOS_VFSFSYNC);
937 	mutex_unlock(&inode->i_mutex);
938 	return 0;
939 }
940 
941 /**
942  * nfs_force_lookup_revalidate - Mark the directory as having changed
943  * @dir - pointer to directory inode
944  *
945  * This forces the revalidation code in nfs_lookup_revalidate() to do a
946  * full lookup on all child dentries of 'dir' whenever a change occurs
947  * on the server that might have invalidated our dcache.
948  *
949  * The caller should be holding dir->i_lock
950  */
951 void nfs_force_lookup_revalidate(struct inode *dir)
952 {
953 	NFS_I(dir)->cache_change_attribute++;
954 }
955 EXPORT_SYMBOL_GPL(nfs_force_lookup_revalidate);
956 
957 /*
958  * A check for whether or not the parent directory has changed.
959  * In the case it has, we assume that the dentries are untrustworthy
960  * and may need to be looked up again.
961  */
962 static int nfs_check_verifier(struct inode *dir, struct dentry *dentry)
963 {
964 	if (IS_ROOT(dentry))
965 		return 1;
966 	if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONE)
967 		return 0;
968 	if (!nfs_verify_change_attribute(dir, dentry->d_time))
969 		return 0;
970 	/* Revalidate nfsi->cache_change_attribute before we declare a match */
971 	if (nfs_revalidate_inode(NFS_SERVER(dir), dir) < 0)
972 		return 0;
973 	if (!nfs_verify_change_attribute(dir, dentry->d_time))
974 		return 0;
975 	return 1;
976 }
977 
978 /*
979  * Use intent information to check whether or not we're going to do
980  * an O_EXCL create using this path component.
981  */
982 static int nfs_is_exclusive_create(struct inode *dir, unsigned int flags)
983 {
984 	if (NFS_PROTO(dir)->version == 2)
985 		return 0;
986 	return flags & LOOKUP_EXCL;
987 }
988 
989 /*
990  * Inode and filehandle revalidation for lookups.
991  *
992  * We force revalidation in the cases where the VFS sets LOOKUP_REVAL,
993  * or if the intent information indicates that we're about to open this
994  * particular file and the "nocto" mount flag is not set.
995  *
996  */
997 static
998 int nfs_lookup_verify_inode(struct inode *inode, unsigned int flags)
999 {
1000 	struct nfs_server *server = NFS_SERVER(inode);
1001 	int ret;
1002 
1003 	if (IS_AUTOMOUNT(inode))
1004 		return 0;
1005 	/* VFS wants an on-the-wire revalidation */
1006 	if (flags & LOOKUP_REVAL)
1007 		goto out_force;
1008 	/* This is an open(2) */
1009 	if ((flags & LOOKUP_OPEN) && !(server->flags & NFS_MOUNT_NOCTO) &&
1010 	    (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)))
1011 		goto out_force;
1012 out:
1013 	return (inode->i_nlink == 0) ? -ENOENT : 0;
1014 out_force:
1015 	ret = __nfs_revalidate_inode(server, inode);
1016 	if (ret != 0)
1017 		return ret;
1018 	goto out;
1019 }
1020 
1021 /*
1022  * We judge how long we want to trust negative
1023  * dentries by looking at the parent inode mtime.
1024  *
1025  * If parent mtime has changed, we revalidate, else we wait for a
1026  * period corresponding to the parent's attribute cache timeout value.
1027  */
1028 static inline
1029 int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry,
1030 		       unsigned int flags)
1031 {
1032 	/* Don't revalidate a negative dentry if we're creating a new file */
1033 	if (flags & LOOKUP_CREATE)
1034 		return 0;
1035 	if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONEG)
1036 		return 1;
1037 	return !nfs_check_verifier(dir, dentry);
1038 }
1039 
1040 /*
1041  * This is called every time the dcache has a lookup hit,
1042  * and we should check whether we can really trust that
1043  * lookup.
1044  *
1045  * NOTE! The hit can be a negative hit too, don't assume
1046  * we have an inode!
1047  *
1048  * If the parent directory is seen to have changed, we throw out the
1049  * cached dentry and do a new lookup.
1050  */
1051 static int nfs_lookup_revalidate(struct dentry *dentry, unsigned int flags)
1052 {
1053 	struct inode *dir;
1054 	struct inode *inode;
1055 	struct dentry *parent;
1056 	struct nfs_fh *fhandle = NULL;
1057 	struct nfs_fattr *fattr = NULL;
1058 	struct nfs4_label *label = NULL;
1059 	int error;
1060 
1061 	if (flags & LOOKUP_RCU)
1062 		return -ECHILD;
1063 
1064 	parent = dget_parent(dentry);
1065 	dir = parent->d_inode;
1066 	nfs_inc_stats(dir, NFSIOS_DENTRYREVALIDATE);
1067 	inode = dentry->d_inode;
1068 
1069 	if (!inode) {
1070 		if (nfs_neg_need_reval(dir, dentry, flags))
1071 			goto out_bad;
1072 		goto out_valid_noent;
1073 	}
1074 
1075 	if (is_bad_inode(inode)) {
1076 		dfprintk(LOOKUPCACHE, "%s: %s/%s has dud inode\n",
1077 				__func__, dentry->d_parent->d_name.name,
1078 				dentry->d_name.name);
1079 		goto out_bad;
1080 	}
1081 
1082 	if (NFS_PROTO(dir)->have_delegation(inode, FMODE_READ))
1083 		goto out_set_verifier;
1084 
1085 	/* Force a full look up iff the parent directory has changed */
1086 	if (!nfs_is_exclusive_create(dir, flags) && nfs_check_verifier(dir, dentry)) {
1087 		if (nfs_lookup_verify_inode(inode, flags))
1088 			goto out_zap_parent;
1089 		goto out_valid;
1090 	}
1091 
1092 	if (NFS_STALE(inode))
1093 		goto out_bad;
1094 
1095 	error = -ENOMEM;
1096 	fhandle = nfs_alloc_fhandle();
1097 	fattr = nfs_alloc_fattr();
1098 	if (fhandle == NULL || fattr == NULL)
1099 		goto out_error;
1100 
1101 	label = nfs4_label_alloc(NFS_SERVER(inode), GFP_NOWAIT);
1102 	if (IS_ERR(label))
1103 		goto out_error;
1104 
1105 	trace_nfs_lookup_revalidate_enter(dir, dentry, flags);
1106 	error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr, label);
1107 	trace_nfs_lookup_revalidate_exit(dir, dentry, flags, error);
1108 	if (error)
1109 		goto out_bad;
1110 	if (nfs_compare_fh(NFS_FH(inode), fhandle))
1111 		goto out_bad;
1112 	if ((error = nfs_refresh_inode(inode, fattr)) != 0)
1113 		goto out_bad;
1114 
1115 	nfs_setsecurity(inode, fattr, label);
1116 
1117 	nfs_free_fattr(fattr);
1118 	nfs_free_fhandle(fhandle);
1119 	nfs4_label_free(label);
1120 
1121 out_set_verifier:
1122 	nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1123  out_valid:
1124 	/* Success: notify readdir to use READDIRPLUS */
1125 	nfs_advise_use_readdirplus(dir);
1126  out_valid_noent:
1127 	dput(parent);
1128 	dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is valid\n",
1129 			__func__, dentry->d_parent->d_name.name,
1130 			dentry->d_name.name);
1131 	return 1;
1132 out_zap_parent:
1133 	nfs_zap_caches(dir);
1134  out_bad:
1135 	nfs_free_fattr(fattr);
1136 	nfs_free_fhandle(fhandle);
1137 	nfs4_label_free(label);
1138 	nfs_mark_for_revalidate(dir);
1139 	if (inode && S_ISDIR(inode->i_mode)) {
1140 		/* Purge readdir caches. */
1141 		nfs_zap_caches(inode);
1142 		/* If we have submounts, don't unhash ! */
1143 		if (have_submounts(dentry))
1144 			goto out_valid;
1145 		if (dentry->d_flags & DCACHE_DISCONNECTED)
1146 			goto out_valid;
1147 		shrink_dcache_parent(dentry);
1148 	}
1149 	d_drop(dentry);
1150 	dput(parent);
1151 	dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is invalid\n",
1152 			__func__, dentry->d_parent->d_name.name,
1153 			dentry->d_name.name);
1154 	return 0;
1155 out_error:
1156 	nfs_free_fattr(fattr);
1157 	nfs_free_fhandle(fhandle);
1158 	nfs4_label_free(label);
1159 	dput(parent);
1160 	dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) lookup returned error %d\n",
1161 			__func__, dentry->d_parent->d_name.name,
1162 			dentry->d_name.name, error);
1163 	return error;
1164 }
1165 
1166 /*
1167  * A weaker form of d_revalidate for revalidating just the dentry->d_inode
1168  * when we don't really care about the dentry name. This is called when a
1169  * pathwalk ends on a dentry that was not found via a normal lookup in the
1170  * parent dir (e.g.: ".", "..", procfs symlinks or mountpoint traversals).
1171  *
1172  * In this situation, we just want to verify that the inode itself is OK
1173  * since the dentry might have changed on the server.
1174  */
1175 static int nfs_weak_revalidate(struct dentry *dentry, unsigned int flags)
1176 {
1177 	int error;
1178 	struct inode *inode = dentry->d_inode;
1179 
1180 	/*
1181 	 * I believe we can only get a negative dentry here in the case of a
1182 	 * procfs-style symlink. Just assume it's correct for now, but we may
1183 	 * eventually need to do something more here.
1184 	 */
1185 	if (!inode) {
1186 		dfprintk(LOOKUPCACHE, "%s: %s/%s has negative inode\n",
1187 				__func__, dentry->d_parent->d_name.name,
1188 				dentry->d_name.name);
1189 		return 1;
1190 	}
1191 
1192 	if (is_bad_inode(inode)) {
1193 		dfprintk(LOOKUPCACHE, "%s: %s/%s has dud inode\n",
1194 				__func__, dentry->d_parent->d_name.name,
1195 				dentry->d_name.name);
1196 		return 0;
1197 	}
1198 
1199 	error = nfs_revalidate_inode(NFS_SERVER(inode), inode);
1200 	dfprintk(LOOKUPCACHE, "NFS: %s: inode %lu is %s\n",
1201 			__func__, inode->i_ino, error ? "invalid" : "valid");
1202 	return !error;
1203 }
1204 
1205 /*
1206  * This is called from dput() when d_count is going to 0.
1207  */
1208 static int nfs_dentry_delete(const struct dentry *dentry)
1209 {
1210 	dfprintk(VFS, "NFS: dentry_delete(%s/%s, %x)\n",
1211 		dentry->d_parent->d_name.name, dentry->d_name.name,
1212 		dentry->d_flags);
1213 
1214 	/* Unhash any dentry with a stale inode */
1215 	if (dentry->d_inode != NULL && NFS_STALE(dentry->d_inode))
1216 		return 1;
1217 
1218 	if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1219 		/* Unhash it, so that ->d_iput() would be called */
1220 		return 1;
1221 	}
1222 	if (!(dentry->d_sb->s_flags & MS_ACTIVE)) {
1223 		/* Unhash it, so that ancestors of killed async unlink
1224 		 * files will be cleaned up during umount */
1225 		return 1;
1226 	}
1227 	return 0;
1228 
1229 }
1230 
1231 /* Ensure that we revalidate inode->i_nlink */
1232 static void nfs_drop_nlink(struct inode *inode)
1233 {
1234 	spin_lock(&inode->i_lock);
1235 	/* drop the inode if we're reasonably sure this is the last link */
1236 	if (inode->i_nlink == 1)
1237 		clear_nlink(inode);
1238 	NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ATTR;
1239 	spin_unlock(&inode->i_lock);
1240 }
1241 
1242 /*
1243  * Called when the dentry loses inode.
1244  * We use it to clean up silly-renamed files.
1245  */
1246 static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode)
1247 {
1248 	if (S_ISDIR(inode->i_mode))
1249 		/* drop any readdir cache as it could easily be old */
1250 		NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA;
1251 
1252 	if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1253 		nfs_complete_unlink(dentry, inode);
1254 		nfs_drop_nlink(inode);
1255 	}
1256 	iput(inode);
1257 }
1258 
1259 static void nfs_d_release(struct dentry *dentry)
1260 {
1261 	/* free cached devname value, if it survived that far */
1262 	if (unlikely(dentry->d_fsdata)) {
1263 		if (dentry->d_flags & DCACHE_NFSFS_RENAMED)
1264 			WARN_ON(1);
1265 		else
1266 			kfree(dentry->d_fsdata);
1267 	}
1268 }
1269 
1270 const struct dentry_operations nfs_dentry_operations = {
1271 	.d_revalidate	= nfs_lookup_revalidate,
1272 	.d_weak_revalidate	= nfs_weak_revalidate,
1273 	.d_delete	= nfs_dentry_delete,
1274 	.d_iput		= nfs_dentry_iput,
1275 	.d_automount	= nfs_d_automount,
1276 	.d_release	= nfs_d_release,
1277 };
1278 EXPORT_SYMBOL_GPL(nfs_dentry_operations);
1279 
1280 struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
1281 {
1282 	struct dentry *res;
1283 	struct dentry *parent;
1284 	struct inode *inode = NULL;
1285 	struct nfs_fh *fhandle = NULL;
1286 	struct nfs_fattr *fattr = NULL;
1287 	struct nfs4_label *label = NULL;
1288 	int error;
1289 
1290 	dfprintk(VFS, "NFS: lookup(%s/%s)\n",
1291 		dentry->d_parent->d_name.name, dentry->d_name.name);
1292 	nfs_inc_stats(dir, NFSIOS_VFSLOOKUP);
1293 
1294 	res = ERR_PTR(-ENAMETOOLONG);
1295 	if (dentry->d_name.len > NFS_SERVER(dir)->namelen)
1296 		goto out;
1297 
1298 	/*
1299 	 * If we're doing an exclusive create, optimize away the lookup
1300 	 * but don't hash the dentry.
1301 	 */
1302 	if (nfs_is_exclusive_create(dir, flags)) {
1303 		d_instantiate(dentry, NULL);
1304 		res = NULL;
1305 		goto out;
1306 	}
1307 
1308 	res = ERR_PTR(-ENOMEM);
1309 	fhandle = nfs_alloc_fhandle();
1310 	fattr = nfs_alloc_fattr();
1311 	if (fhandle == NULL || fattr == NULL)
1312 		goto out;
1313 
1314 	label = nfs4_label_alloc(NFS_SERVER(dir), GFP_NOWAIT);
1315 	if (IS_ERR(label))
1316 		goto out;
1317 
1318 	parent = dentry->d_parent;
1319 	/* Protect against concurrent sillydeletes */
1320 	trace_nfs_lookup_enter(dir, dentry, flags);
1321 	nfs_block_sillyrename(parent);
1322 	error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr, label);
1323 	if (error == -ENOENT)
1324 		goto no_entry;
1325 	if (error < 0) {
1326 		res = ERR_PTR(error);
1327 		goto out_unblock_sillyrename;
1328 	}
1329 	inode = nfs_fhget(dentry->d_sb, fhandle, fattr, label);
1330 	res = ERR_CAST(inode);
1331 	if (IS_ERR(res))
1332 		goto out_unblock_sillyrename;
1333 
1334 	/* Success: notify readdir to use READDIRPLUS */
1335 	nfs_advise_use_readdirplus(dir);
1336 
1337 no_entry:
1338 	res = d_materialise_unique(dentry, inode);
1339 	if (res != NULL) {
1340 		if (IS_ERR(res))
1341 			goto out_unblock_sillyrename;
1342 		dentry = res;
1343 	}
1344 	nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1345 out_unblock_sillyrename:
1346 	nfs_unblock_sillyrename(parent);
1347 	trace_nfs_lookup_exit(dir, dentry, flags, error);
1348 	nfs4_label_free(label);
1349 out:
1350 	nfs_free_fattr(fattr);
1351 	nfs_free_fhandle(fhandle);
1352 	return res;
1353 }
1354 EXPORT_SYMBOL_GPL(nfs_lookup);
1355 
1356 #if IS_ENABLED(CONFIG_NFS_V4)
1357 static int nfs4_lookup_revalidate(struct dentry *, unsigned int);
1358 
1359 const struct dentry_operations nfs4_dentry_operations = {
1360 	.d_revalidate	= nfs4_lookup_revalidate,
1361 	.d_delete	= nfs_dentry_delete,
1362 	.d_iput		= nfs_dentry_iput,
1363 	.d_automount	= nfs_d_automount,
1364 	.d_release	= nfs_d_release,
1365 };
1366 EXPORT_SYMBOL_GPL(nfs4_dentry_operations);
1367 
1368 static fmode_t flags_to_mode(int flags)
1369 {
1370 	fmode_t res = (__force fmode_t)flags & FMODE_EXEC;
1371 	if ((flags & O_ACCMODE) != O_WRONLY)
1372 		res |= FMODE_READ;
1373 	if ((flags & O_ACCMODE) != O_RDONLY)
1374 		res |= FMODE_WRITE;
1375 	return res;
1376 }
1377 
1378 static struct nfs_open_context *create_nfs_open_context(struct dentry *dentry, int open_flags)
1379 {
1380 	return alloc_nfs_open_context(dentry, flags_to_mode(open_flags));
1381 }
1382 
1383 static int do_open(struct inode *inode, struct file *filp)
1384 {
1385 	nfs_fscache_set_inode_cookie(inode, filp);
1386 	return 0;
1387 }
1388 
1389 static int nfs_finish_open(struct nfs_open_context *ctx,
1390 			   struct dentry *dentry,
1391 			   struct file *file, unsigned open_flags,
1392 			   int *opened)
1393 {
1394 	int err;
1395 
1396 	err = finish_open(file, dentry, do_open, opened);
1397 	if (err)
1398 		goto out;
1399 	nfs_file_set_open_context(file, ctx);
1400 
1401 out:
1402 	put_nfs_open_context(ctx);
1403 	return err;
1404 }
1405 
1406 int nfs_atomic_open(struct inode *dir, struct dentry *dentry,
1407 		    struct file *file, unsigned open_flags,
1408 		    umode_t mode, int *opened)
1409 {
1410 	struct nfs_open_context *ctx;
1411 	struct dentry *res;
1412 	struct iattr attr = { .ia_valid = ATTR_OPEN };
1413 	struct inode *inode;
1414 	unsigned int lookup_flags = 0;
1415 	int err;
1416 
1417 	/* Expect a negative dentry */
1418 	BUG_ON(dentry->d_inode);
1419 
1420 	dfprintk(VFS, "NFS: atomic_open(%s/%ld), %s\n",
1421 			dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1422 
1423 	err = nfs_check_flags(open_flags);
1424 	if (err)
1425 		return err;
1426 
1427 	/* NFS only supports OPEN on regular files */
1428 	if ((open_flags & O_DIRECTORY)) {
1429 		if (!d_unhashed(dentry)) {
1430 			/*
1431 			 * Hashed negative dentry with O_DIRECTORY: dentry was
1432 			 * revalidated and is fine, no need to perform lookup
1433 			 * again
1434 			 */
1435 			return -ENOENT;
1436 		}
1437 		lookup_flags = LOOKUP_OPEN|LOOKUP_DIRECTORY;
1438 		goto no_open;
1439 	}
1440 
1441 	if (dentry->d_name.len > NFS_SERVER(dir)->namelen)
1442 		return -ENAMETOOLONG;
1443 
1444 	if (open_flags & O_CREAT) {
1445 		attr.ia_valid |= ATTR_MODE;
1446 		attr.ia_mode = mode & ~current_umask();
1447 	}
1448 	if (open_flags & O_TRUNC) {
1449 		attr.ia_valid |= ATTR_SIZE;
1450 		attr.ia_size = 0;
1451 	}
1452 
1453 	ctx = create_nfs_open_context(dentry, open_flags);
1454 	err = PTR_ERR(ctx);
1455 	if (IS_ERR(ctx))
1456 		goto out;
1457 
1458 	trace_nfs_atomic_open_enter(dir, ctx, open_flags);
1459 	nfs_block_sillyrename(dentry->d_parent);
1460 	inode = NFS_PROTO(dir)->open_context(dir, ctx, open_flags, &attr);
1461 	nfs_unblock_sillyrename(dentry->d_parent);
1462 	if (IS_ERR(inode)) {
1463 		put_nfs_open_context(ctx);
1464 		err = PTR_ERR(inode);
1465 		trace_nfs_atomic_open_exit(dir, ctx, open_flags, err);
1466 		switch (err) {
1467 		case -ENOENT:
1468 			d_drop(dentry);
1469 			d_add(dentry, NULL);
1470 			break;
1471 		case -EISDIR:
1472 		case -ENOTDIR:
1473 			goto no_open;
1474 		case -ELOOP:
1475 			if (!(open_flags & O_NOFOLLOW))
1476 				goto no_open;
1477 			break;
1478 			/* case -EINVAL: */
1479 		default:
1480 			break;
1481 		}
1482 		goto out;
1483 	}
1484 
1485 	err = nfs_finish_open(ctx, ctx->dentry, file, open_flags, opened);
1486 	trace_nfs_atomic_open_exit(dir, ctx, open_flags, err);
1487 out:
1488 	return err;
1489 
1490 no_open:
1491 	res = nfs_lookup(dir, dentry, lookup_flags);
1492 	err = PTR_ERR(res);
1493 	if (IS_ERR(res))
1494 		goto out;
1495 
1496 	return finish_no_open(file, res);
1497 }
1498 EXPORT_SYMBOL_GPL(nfs_atomic_open);
1499 
1500 static int nfs4_lookup_revalidate(struct dentry *dentry, unsigned int flags)
1501 {
1502 	struct dentry *parent = NULL;
1503 	struct inode *inode;
1504 	struct inode *dir;
1505 	int ret = 0;
1506 
1507 	if (flags & LOOKUP_RCU)
1508 		return -ECHILD;
1509 
1510 	if (!(flags & LOOKUP_OPEN) || (flags & LOOKUP_DIRECTORY))
1511 		goto no_open;
1512 	if (d_mountpoint(dentry))
1513 		goto no_open;
1514 	if (NFS_SB(dentry->d_sb)->caps & NFS_CAP_ATOMIC_OPEN_V1)
1515 		goto no_open;
1516 
1517 	inode = dentry->d_inode;
1518 	parent = dget_parent(dentry);
1519 	dir = parent->d_inode;
1520 
1521 	/* We can't create new files in nfs_open_revalidate(), so we
1522 	 * optimize away revalidation of negative dentries.
1523 	 */
1524 	if (inode == NULL) {
1525 		if (!nfs_neg_need_reval(dir, dentry, flags))
1526 			ret = 1;
1527 		goto out;
1528 	}
1529 
1530 	/* NFS only supports OPEN on regular files */
1531 	if (!S_ISREG(inode->i_mode))
1532 		goto no_open_dput;
1533 	/* We cannot do exclusive creation on a positive dentry */
1534 	if (flags & LOOKUP_EXCL)
1535 		goto no_open_dput;
1536 
1537 	/* Let f_op->open() actually open (and revalidate) the file */
1538 	ret = 1;
1539 
1540 out:
1541 	dput(parent);
1542 	return ret;
1543 
1544 no_open_dput:
1545 	dput(parent);
1546 no_open:
1547 	return nfs_lookup_revalidate(dentry, flags);
1548 }
1549 
1550 #endif /* CONFIG_NFSV4 */
1551 
1552 /*
1553  * Code common to create, mkdir, and mknod.
1554  */
1555 int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle,
1556 				struct nfs_fattr *fattr,
1557 				struct nfs4_label *label)
1558 {
1559 	struct dentry *parent = dget_parent(dentry);
1560 	struct inode *dir = parent->d_inode;
1561 	struct inode *inode;
1562 	int error = -EACCES;
1563 
1564 	d_drop(dentry);
1565 
1566 	/* We may have been initialized further down */
1567 	if (dentry->d_inode)
1568 		goto out;
1569 	if (fhandle->size == 0) {
1570 		error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr, NULL);
1571 		if (error)
1572 			goto out_error;
1573 	}
1574 	nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1575 	if (!(fattr->valid & NFS_ATTR_FATTR)) {
1576 		struct nfs_server *server = NFS_SB(dentry->d_sb);
1577 		error = server->nfs_client->rpc_ops->getattr(server, fhandle, fattr, NULL);
1578 		if (error < 0)
1579 			goto out_error;
1580 	}
1581 	inode = nfs_fhget(dentry->d_sb, fhandle, fattr, label);
1582 	error = PTR_ERR(inode);
1583 	if (IS_ERR(inode))
1584 		goto out_error;
1585 	d_add(dentry, inode);
1586 out:
1587 	dput(parent);
1588 	return 0;
1589 out_error:
1590 	nfs_mark_for_revalidate(dir);
1591 	dput(parent);
1592 	return error;
1593 }
1594 EXPORT_SYMBOL_GPL(nfs_instantiate);
1595 
1596 /*
1597  * Following a failed create operation, we drop the dentry rather
1598  * than retain a negative dentry. This avoids a problem in the event
1599  * that the operation succeeded on the server, but an error in the
1600  * reply path made it appear to have failed.
1601  */
1602 int nfs_create(struct inode *dir, struct dentry *dentry,
1603 		umode_t mode, bool excl)
1604 {
1605 	struct iattr attr;
1606 	int open_flags = excl ? O_CREAT | O_EXCL : O_CREAT;
1607 	int error;
1608 
1609 	dfprintk(VFS, "NFS: create(%s/%ld), %s\n",
1610 			dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1611 
1612 	attr.ia_mode = mode;
1613 	attr.ia_valid = ATTR_MODE;
1614 
1615 	trace_nfs_create_enter(dir, dentry, open_flags);
1616 	error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags);
1617 	trace_nfs_create_exit(dir, dentry, open_flags, error);
1618 	if (error != 0)
1619 		goto out_err;
1620 	return 0;
1621 out_err:
1622 	d_drop(dentry);
1623 	return error;
1624 }
1625 EXPORT_SYMBOL_GPL(nfs_create);
1626 
1627 /*
1628  * See comments for nfs_proc_create regarding failed operations.
1629  */
1630 int
1631 nfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t rdev)
1632 {
1633 	struct iattr attr;
1634 	int status;
1635 
1636 	dfprintk(VFS, "NFS: mknod(%s/%ld), %s\n",
1637 			dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1638 
1639 	if (!new_valid_dev(rdev))
1640 		return -EINVAL;
1641 
1642 	attr.ia_mode = mode;
1643 	attr.ia_valid = ATTR_MODE;
1644 
1645 	trace_nfs_mknod_enter(dir, dentry);
1646 	status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev);
1647 	trace_nfs_mknod_exit(dir, dentry, status);
1648 	if (status != 0)
1649 		goto out_err;
1650 	return 0;
1651 out_err:
1652 	d_drop(dentry);
1653 	return status;
1654 }
1655 EXPORT_SYMBOL_GPL(nfs_mknod);
1656 
1657 /*
1658  * See comments for nfs_proc_create regarding failed operations.
1659  */
1660 int nfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
1661 {
1662 	struct iattr attr;
1663 	int error;
1664 
1665 	dfprintk(VFS, "NFS: mkdir(%s/%ld), %s\n",
1666 			dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1667 
1668 	attr.ia_valid = ATTR_MODE;
1669 	attr.ia_mode = mode | S_IFDIR;
1670 
1671 	trace_nfs_mkdir_enter(dir, dentry);
1672 	error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr);
1673 	trace_nfs_mkdir_exit(dir, dentry, error);
1674 	if (error != 0)
1675 		goto out_err;
1676 	return 0;
1677 out_err:
1678 	d_drop(dentry);
1679 	return error;
1680 }
1681 EXPORT_SYMBOL_GPL(nfs_mkdir);
1682 
1683 static void nfs_dentry_handle_enoent(struct dentry *dentry)
1684 {
1685 	if (dentry->d_inode != NULL && !d_unhashed(dentry))
1686 		d_delete(dentry);
1687 }
1688 
1689 int nfs_rmdir(struct inode *dir, struct dentry *dentry)
1690 {
1691 	int error;
1692 
1693 	dfprintk(VFS, "NFS: rmdir(%s/%ld), %s\n",
1694 			dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1695 
1696 	trace_nfs_rmdir_enter(dir, dentry);
1697 	error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
1698 	/* Ensure the VFS deletes this inode */
1699 	if (error == 0 && dentry->d_inode != NULL)
1700 		clear_nlink(dentry->d_inode);
1701 	else if (error == -ENOENT)
1702 		nfs_dentry_handle_enoent(dentry);
1703 	trace_nfs_rmdir_exit(dir, dentry, error);
1704 
1705 	return error;
1706 }
1707 EXPORT_SYMBOL_GPL(nfs_rmdir);
1708 
1709 /*
1710  * Remove a file after making sure there are no pending writes,
1711  * and after checking that the file has only one user.
1712  *
1713  * We invalidate the attribute cache and free the inode prior to the operation
1714  * to avoid possible races if the server reuses the inode.
1715  */
1716 static int nfs_safe_remove(struct dentry *dentry)
1717 {
1718 	struct inode *dir = dentry->d_parent->d_inode;
1719 	struct inode *inode = dentry->d_inode;
1720 	int error = -EBUSY;
1721 
1722 	dfprintk(VFS, "NFS: safe_remove(%s/%s)\n",
1723 		dentry->d_parent->d_name.name, dentry->d_name.name);
1724 
1725 	/* If the dentry was sillyrenamed, we simply call d_delete() */
1726 	if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1727 		error = 0;
1728 		goto out;
1729 	}
1730 
1731 	trace_nfs_remove_enter(dir, dentry);
1732 	if (inode != NULL) {
1733 		NFS_PROTO(inode)->return_delegation(inode);
1734 		error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1735 		if (error == 0)
1736 			nfs_drop_nlink(inode);
1737 	} else
1738 		error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1739 	if (error == -ENOENT)
1740 		nfs_dentry_handle_enoent(dentry);
1741 	trace_nfs_remove_exit(dir, dentry, error);
1742 out:
1743 	return error;
1744 }
1745 
1746 /*  We do silly rename. In case sillyrename() returns -EBUSY, the inode
1747  *  belongs to an active ".nfs..." file and we return -EBUSY.
1748  *
1749  *  If sillyrename() returns 0, we do nothing, otherwise we unlink.
1750  */
1751 int nfs_unlink(struct inode *dir, struct dentry *dentry)
1752 {
1753 	int error;
1754 	int need_rehash = 0;
1755 
1756 	dfprintk(VFS, "NFS: unlink(%s/%ld, %s)\n", dir->i_sb->s_id,
1757 		dir->i_ino, dentry->d_name.name);
1758 
1759 	trace_nfs_unlink_enter(dir, dentry);
1760 	spin_lock(&dentry->d_lock);
1761 	if (d_count(dentry) > 1) {
1762 		spin_unlock(&dentry->d_lock);
1763 		/* Start asynchronous writeout of the inode */
1764 		write_inode_now(dentry->d_inode, 0);
1765 		error = nfs_sillyrename(dir, dentry);
1766 		goto out;
1767 	}
1768 	if (!d_unhashed(dentry)) {
1769 		__d_drop(dentry);
1770 		need_rehash = 1;
1771 	}
1772 	spin_unlock(&dentry->d_lock);
1773 	error = nfs_safe_remove(dentry);
1774 	if (!error || error == -ENOENT) {
1775 		nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1776 	} else if (need_rehash)
1777 		d_rehash(dentry);
1778 out:
1779 	trace_nfs_unlink_exit(dir, dentry, error);
1780 	return error;
1781 }
1782 EXPORT_SYMBOL_GPL(nfs_unlink);
1783 
1784 /*
1785  * To create a symbolic link, most file systems instantiate a new inode,
1786  * add a page to it containing the path, then write it out to the disk
1787  * using prepare_write/commit_write.
1788  *
1789  * Unfortunately the NFS client can't create the in-core inode first
1790  * because it needs a file handle to create an in-core inode (see
1791  * fs/nfs/inode.c:nfs_fhget).  We only have a file handle *after* the
1792  * symlink request has completed on the server.
1793  *
1794  * So instead we allocate a raw page, copy the symname into it, then do
1795  * the SYMLINK request with the page as the buffer.  If it succeeds, we
1796  * now have a new file handle and can instantiate an in-core NFS inode
1797  * and move the raw page into its mapping.
1798  */
1799 int nfs_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1800 {
1801 	struct page *page;
1802 	char *kaddr;
1803 	struct iattr attr;
1804 	unsigned int pathlen = strlen(symname);
1805 	int error;
1806 
1807 	dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s)\n", dir->i_sb->s_id,
1808 		dir->i_ino, dentry->d_name.name, symname);
1809 
1810 	if (pathlen > PAGE_SIZE)
1811 		return -ENAMETOOLONG;
1812 
1813 	attr.ia_mode = S_IFLNK | S_IRWXUGO;
1814 	attr.ia_valid = ATTR_MODE;
1815 
1816 	page = alloc_page(GFP_HIGHUSER);
1817 	if (!page)
1818 		return -ENOMEM;
1819 
1820 	kaddr = kmap_atomic(page);
1821 	memcpy(kaddr, symname, pathlen);
1822 	if (pathlen < PAGE_SIZE)
1823 		memset(kaddr + pathlen, 0, PAGE_SIZE - pathlen);
1824 	kunmap_atomic(kaddr);
1825 
1826 	trace_nfs_symlink_enter(dir, dentry);
1827 	error = NFS_PROTO(dir)->symlink(dir, dentry, page, pathlen, &attr);
1828 	trace_nfs_symlink_exit(dir, dentry, error);
1829 	if (error != 0) {
1830 		dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s) error %d\n",
1831 			dir->i_sb->s_id, dir->i_ino,
1832 			dentry->d_name.name, symname, error);
1833 		d_drop(dentry);
1834 		__free_page(page);
1835 		return error;
1836 	}
1837 
1838 	/*
1839 	 * No big deal if we can't add this page to the page cache here.
1840 	 * READLINK will get the missing page from the server if needed.
1841 	 */
1842 	if (!add_to_page_cache_lru(page, dentry->d_inode->i_mapping, 0,
1843 							GFP_KERNEL)) {
1844 		SetPageUptodate(page);
1845 		unlock_page(page);
1846 	} else
1847 		__free_page(page);
1848 
1849 	return 0;
1850 }
1851 EXPORT_SYMBOL_GPL(nfs_symlink);
1852 
1853 int
1854 nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1855 {
1856 	struct inode *inode = old_dentry->d_inode;
1857 	int error;
1858 
1859 	dfprintk(VFS, "NFS: link(%s/%s -> %s/%s)\n",
1860 		old_dentry->d_parent->d_name.name, old_dentry->d_name.name,
1861 		dentry->d_parent->d_name.name, dentry->d_name.name);
1862 
1863 	NFS_PROTO(inode)->return_delegation(inode);
1864 
1865 	d_drop(dentry);
1866 	error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name);
1867 	if (error == 0) {
1868 		ihold(inode);
1869 		d_add(dentry, inode);
1870 	}
1871 	return error;
1872 }
1873 EXPORT_SYMBOL_GPL(nfs_link);
1874 
1875 /*
1876  * RENAME
1877  * FIXME: Some nfsds, like the Linux user space nfsd, may generate a
1878  * different file handle for the same inode after a rename (e.g. when
1879  * moving to a different directory). A fail-safe method to do so would
1880  * be to look up old_dir/old_name, create a link to new_dir/new_name and
1881  * rename the old file using the sillyrename stuff. This way, the original
1882  * file in old_dir will go away when the last process iput()s the inode.
1883  *
1884  * FIXED.
1885  *
1886  * It actually works quite well. One needs to have the possibility for
1887  * at least one ".nfs..." file in each directory the file ever gets
1888  * moved or linked to which happens automagically with the new
1889  * implementation that only depends on the dcache stuff instead of
1890  * using the inode layer
1891  *
1892  * Unfortunately, things are a little more complicated than indicated
1893  * above. For a cross-directory move, we want to make sure we can get
1894  * rid of the old inode after the operation.  This means there must be
1895  * no pending writes (if it's a file), and the use count must be 1.
1896  * If these conditions are met, we can drop the dentries before doing
1897  * the rename.
1898  */
1899 int nfs_rename(struct inode *old_dir, struct dentry *old_dentry,
1900 		      struct inode *new_dir, struct dentry *new_dentry)
1901 {
1902 	struct inode *old_inode = old_dentry->d_inode;
1903 	struct inode *new_inode = new_dentry->d_inode;
1904 	struct dentry *dentry = NULL, *rehash = NULL;
1905 	int error = -EBUSY;
1906 
1907 	dfprintk(VFS, "NFS: rename(%s/%s -> %s/%s, ct=%d)\n",
1908 		 old_dentry->d_parent->d_name.name, old_dentry->d_name.name,
1909 		 new_dentry->d_parent->d_name.name, new_dentry->d_name.name,
1910 		 d_count(new_dentry));
1911 
1912 	/*
1913 	 * For non-directories, check whether the target is busy and if so,
1914 	 * make a copy of the dentry and then do a silly-rename. If the
1915 	 * silly-rename succeeds, the copied dentry is hashed and becomes
1916 	 * the new target.
1917 	 */
1918 	if (new_inode && !S_ISDIR(new_inode->i_mode)) {
1919 		/*
1920 		 * To prevent any new references to the target during the
1921 		 * rename, we unhash the dentry in advance.
1922 		 */
1923 		if (!d_unhashed(new_dentry)) {
1924 			d_drop(new_dentry);
1925 			rehash = new_dentry;
1926 		}
1927 
1928 		if (d_count(new_dentry) > 2) {
1929 			int err;
1930 
1931 			/* copy the target dentry's name */
1932 			dentry = d_alloc(new_dentry->d_parent,
1933 					 &new_dentry->d_name);
1934 			if (!dentry)
1935 				goto out;
1936 
1937 			/* silly-rename the existing target ... */
1938 			err = nfs_sillyrename(new_dir, new_dentry);
1939 			if (err)
1940 				goto out;
1941 
1942 			new_dentry = dentry;
1943 			rehash = NULL;
1944 			new_inode = NULL;
1945 		}
1946 	}
1947 
1948 	NFS_PROTO(old_inode)->return_delegation(old_inode);
1949 	if (new_inode != NULL)
1950 		NFS_PROTO(new_inode)->return_delegation(new_inode);
1951 
1952 	error = NFS_PROTO(old_dir)->rename(old_dir, &old_dentry->d_name,
1953 					   new_dir, &new_dentry->d_name);
1954 	nfs_mark_for_revalidate(old_inode);
1955 out:
1956 	if (rehash)
1957 		d_rehash(rehash);
1958 	if (!error) {
1959 		if (new_inode != NULL)
1960 			nfs_drop_nlink(new_inode);
1961 		d_move(old_dentry, new_dentry);
1962 		nfs_set_verifier(new_dentry,
1963 					nfs_save_change_attribute(new_dir));
1964 	} else if (error == -ENOENT)
1965 		nfs_dentry_handle_enoent(old_dentry);
1966 
1967 	/* new dentry created? */
1968 	if (dentry)
1969 		dput(dentry);
1970 	return error;
1971 }
1972 EXPORT_SYMBOL_GPL(nfs_rename);
1973 
1974 static DEFINE_SPINLOCK(nfs_access_lru_lock);
1975 static LIST_HEAD(nfs_access_lru_list);
1976 static atomic_long_t nfs_access_nr_entries;
1977 
1978 static void nfs_access_free_entry(struct nfs_access_entry *entry)
1979 {
1980 	put_rpccred(entry->cred);
1981 	kfree(entry);
1982 	smp_mb__before_atomic_dec();
1983 	atomic_long_dec(&nfs_access_nr_entries);
1984 	smp_mb__after_atomic_dec();
1985 }
1986 
1987 static void nfs_access_free_list(struct list_head *head)
1988 {
1989 	struct nfs_access_entry *cache;
1990 
1991 	while (!list_empty(head)) {
1992 		cache = list_entry(head->next, struct nfs_access_entry, lru);
1993 		list_del(&cache->lru);
1994 		nfs_access_free_entry(cache);
1995 	}
1996 }
1997 
1998 int nfs_access_cache_shrinker(struct shrinker *shrink,
1999 			      struct shrink_control *sc)
2000 {
2001 	LIST_HEAD(head);
2002 	struct nfs_inode *nfsi, *next;
2003 	struct nfs_access_entry *cache;
2004 	int nr_to_scan = sc->nr_to_scan;
2005 	gfp_t gfp_mask = sc->gfp_mask;
2006 
2007 	if ((gfp_mask & GFP_KERNEL) != GFP_KERNEL)
2008 		return (nr_to_scan == 0) ? 0 : -1;
2009 
2010 	spin_lock(&nfs_access_lru_lock);
2011 	list_for_each_entry_safe(nfsi, next, &nfs_access_lru_list, access_cache_inode_lru) {
2012 		struct inode *inode;
2013 
2014 		if (nr_to_scan-- == 0)
2015 			break;
2016 		inode = &nfsi->vfs_inode;
2017 		spin_lock(&inode->i_lock);
2018 		if (list_empty(&nfsi->access_cache_entry_lru))
2019 			goto remove_lru_entry;
2020 		cache = list_entry(nfsi->access_cache_entry_lru.next,
2021 				struct nfs_access_entry, lru);
2022 		list_move(&cache->lru, &head);
2023 		rb_erase(&cache->rb_node, &nfsi->access_cache);
2024 		if (!list_empty(&nfsi->access_cache_entry_lru))
2025 			list_move_tail(&nfsi->access_cache_inode_lru,
2026 					&nfs_access_lru_list);
2027 		else {
2028 remove_lru_entry:
2029 			list_del_init(&nfsi->access_cache_inode_lru);
2030 			smp_mb__before_clear_bit();
2031 			clear_bit(NFS_INO_ACL_LRU_SET, &nfsi->flags);
2032 			smp_mb__after_clear_bit();
2033 		}
2034 		spin_unlock(&inode->i_lock);
2035 	}
2036 	spin_unlock(&nfs_access_lru_lock);
2037 	nfs_access_free_list(&head);
2038 	return (atomic_long_read(&nfs_access_nr_entries) / 100) * sysctl_vfs_cache_pressure;
2039 }
2040 
2041 static void __nfs_access_zap_cache(struct nfs_inode *nfsi, struct list_head *head)
2042 {
2043 	struct rb_root *root_node = &nfsi->access_cache;
2044 	struct rb_node *n;
2045 	struct nfs_access_entry *entry;
2046 
2047 	/* Unhook entries from the cache */
2048 	while ((n = rb_first(root_node)) != NULL) {
2049 		entry = rb_entry(n, struct nfs_access_entry, rb_node);
2050 		rb_erase(n, root_node);
2051 		list_move(&entry->lru, head);
2052 	}
2053 	nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS;
2054 }
2055 
2056 void nfs_access_zap_cache(struct inode *inode)
2057 {
2058 	LIST_HEAD(head);
2059 
2060 	if (test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags) == 0)
2061 		return;
2062 	/* Remove from global LRU init */
2063 	spin_lock(&nfs_access_lru_lock);
2064 	if (test_and_clear_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
2065 		list_del_init(&NFS_I(inode)->access_cache_inode_lru);
2066 
2067 	spin_lock(&inode->i_lock);
2068 	__nfs_access_zap_cache(NFS_I(inode), &head);
2069 	spin_unlock(&inode->i_lock);
2070 	spin_unlock(&nfs_access_lru_lock);
2071 	nfs_access_free_list(&head);
2072 }
2073 EXPORT_SYMBOL_GPL(nfs_access_zap_cache);
2074 
2075 static struct nfs_access_entry *nfs_access_search_rbtree(struct inode *inode, struct rpc_cred *cred)
2076 {
2077 	struct rb_node *n = NFS_I(inode)->access_cache.rb_node;
2078 	struct nfs_access_entry *entry;
2079 
2080 	while (n != NULL) {
2081 		entry = rb_entry(n, struct nfs_access_entry, rb_node);
2082 
2083 		if (cred < entry->cred)
2084 			n = n->rb_left;
2085 		else if (cred > entry->cred)
2086 			n = n->rb_right;
2087 		else
2088 			return entry;
2089 	}
2090 	return NULL;
2091 }
2092 
2093 static int nfs_access_get_cached(struct inode *inode, struct rpc_cred *cred, struct nfs_access_entry *res)
2094 {
2095 	struct nfs_inode *nfsi = NFS_I(inode);
2096 	struct nfs_access_entry *cache;
2097 	int err = -ENOENT;
2098 
2099 	spin_lock(&inode->i_lock);
2100 	if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS)
2101 		goto out_zap;
2102 	cache = nfs_access_search_rbtree(inode, cred);
2103 	if (cache == NULL)
2104 		goto out;
2105 	if (!nfs_have_delegated_attributes(inode) &&
2106 	    !time_in_range_open(jiffies, cache->jiffies, cache->jiffies + nfsi->attrtimeo))
2107 		goto out_stale;
2108 	res->jiffies = cache->jiffies;
2109 	res->cred = cache->cred;
2110 	res->mask = cache->mask;
2111 	list_move_tail(&cache->lru, &nfsi->access_cache_entry_lru);
2112 	err = 0;
2113 out:
2114 	spin_unlock(&inode->i_lock);
2115 	return err;
2116 out_stale:
2117 	rb_erase(&cache->rb_node, &nfsi->access_cache);
2118 	list_del(&cache->lru);
2119 	spin_unlock(&inode->i_lock);
2120 	nfs_access_free_entry(cache);
2121 	return -ENOENT;
2122 out_zap:
2123 	spin_unlock(&inode->i_lock);
2124 	nfs_access_zap_cache(inode);
2125 	return -ENOENT;
2126 }
2127 
2128 static void nfs_access_add_rbtree(struct inode *inode, struct nfs_access_entry *set)
2129 {
2130 	struct nfs_inode *nfsi = NFS_I(inode);
2131 	struct rb_root *root_node = &nfsi->access_cache;
2132 	struct rb_node **p = &root_node->rb_node;
2133 	struct rb_node *parent = NULL;
2134 	struct nfs_access_entry *entry;
2135 
2136 	spin_lock(&inode->i_lock);
2137 	while (*p != NULL) {
2138 		parent = *p;
2139 		entry = rb_entry(parent, struct nfs_access_entry, rb_node);
2140 
2141 		if (set->cred < entry->cred)
2142 			p = &parent->rb_left;
2143 		else if (set->cred > entry->cred)
2144 			p = &parent->rb_right;
2145 		else
2146 			goto found;
2147 	}
2148 	rb_link_node(&set->rb_node, parent, p);
2149 	rb_insert_color(&set->rb_node, root_node);
2150 	list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
2151 	spin_unlock(&inode->i_lock);
2152 	return;
2153 found:
2154 	rb_replace_node(parent, &set->rb_node, root_node);
2155 	list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
2156 	list_del(&entry->lru);
2157 	spin_unlock(&inode->i_lock);
2158 	nfs_access_free_entry(entry);
2159 }
2160 
2161 void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set)
2162 {
2163 	struct nfs_access_entry *cache = kmalloc(sizeof(*cache), GFP_KERNEL);
2164 	if (cache == NULL)
2165 		return;
2166 	RB_CLEAR_NODE(&cache->rb_node);
2167 	cache->jiffies = set->jiffies;
2168 	cache->cred = get_rpccred(set->cred);
2169 	cache->mask = set->mask;
2170 
2171 	nfs_access_add_rbtree(inode, cache);
2172 
2173 	/* Update accounting */
2174 	smp_mb__before_atomic_inc();
2175 	atomic_long_inc(&nfs_access_nr_entries);
2176 	smp_mb__after_atomic_inc();
2177 
2178 	/* Add inode to global LRU list */
2179 	if (!test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) {
2180 		spin_lock(&nfs_access_lru_lock);
2181 		if (!test_and_set_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
2182 			list_add_tail(&NFS_I(inode)->access_cache_inode_lru,
2183 					&nfs_access_lru_list);
2184 		spin_unlock(&nfs_access_lru_lock);
2185 	}
2186 }
2187 EXPORT_SYMBOL_GPL(nfs_access_add_cache);
2188 
2189 void nfs_access_set_mask(struct nfs_access_entry *entry, u32 access_result)
2190 {
2191 	entry->mask = 0;
2192 	if (access_result & NFS4_ACCESS_READ)
2193 		entry->mask |= MAY_READ;
2194 	if (access_result &
2195 	    (NFS4_ACCESS_MODIFY | NFS4_ACCESS_EXTEND | NFS4_ACCESS_DELETE))
2196 		entry->mask |= MAY_WRITE;
2197 	if (access_result & (NFS4_ACCESS_LOOKUP|NFS4_ACCESS_EXECUTE))
2198 		entry->mask |= MAY_EXEC;
2199 }
2200 EXPORT_SYMBOL_GPL(nfs_access_set_mask);
2201 
2202 static int nfs_do_access(struct inode *inode, struct rpc_cred *cred, int mask)
2203 {
2204 	struct nfs_access_entry cache;
2205 	int status;
2206 
2207 	trace_nfs_access_enter(inode);
2208 
2209 	status = nfs_access_get_cached(inode, cred, &cache);
2210 	if (status == 0)
2211 		goto out_cached;
2212 
2213 	/* Be clever: ask server to check for all possible rights */
2214 	cache.mask = MAY_EXEC | MAY_WRITE | MAY_READ;
2215 	cache.cred = cred;
2216 	cache.jiffies = jiffies;
2217 	status = NFS_PROTO(inode)->access(inode, &cache);
2218 	if (status != 0) {
2219 		if (status == -ESTALE) {
2220 			nfs_zap_caches(inode);
2221 			if (!S_ISDIR(inode->i_mode))
2222 				set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
2223 		}
2224 		goto out;
2225 	}
2226 	nfs_access_add_cache(inode, &cache);
2227 out_cached:
2228 	if ((mask & ~cache.mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) != 0)
2229 		status = -EACCES;
2230 out:
2231 	trace_nfs_access_exit(inode, status);
2232 	return status;
2233 }
2234 
2235 static int nfs_open_permission_mask(int openflags)
2236 {
2237 	int mask = 0;
2238 
2239 	if (openflags & __FMODE_EXEC) {
2240 		/* ONLY check exec rights */
2241 		mask = MAY_EXEC;
2242 	} else {
2243 		if ((openflags & O_ACCMODE) != O_WRONLY)
2244 			mask |= MAY_READ;
2245 		if ((openflags & O_ACCMODE) != O_RDONLY)
2246 			mask |= MAY_WRITE;
2247 	}
2248 
2249 	return mask;
2250 }
2251 
2252 int nfs_may_open(struct inode *inode, struct rpc_cred *cred, int openflags)
2253 {
2254 	return nfs_do_access(inode, cred, nfs_open_permission_mask(openflags));
2255 }
2256 EXPORT_SYMBOL_GPL(nfs_may_open);
2257 
2258 int nfs_permission(struct inode *inode, int mask)
2259 {
2260 	struct rpc_cred *cred;
2261 	int res = 0;
2262 
2263 	if (mask & MAY_NOT_BLOCK)
2264 		return -ECHILD;
2265 
2266 	nfs_inc_stats(inode, NFSIOS_VFSACCESS);
2267 
2268 	if ((mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
2269 		goto out;
2270 	/* Is this sys_access() ? */
2271 	if (mask & (MAY_ACCESS | MAY_CHDIR))
2272 		goto force_lookup;
2273 
2274 	switch (inode->i_mode & S_IFMT) {
2275 		case S_IFLNK:
2276 			goto out;
2277 		case S_IFREG:
2278 			break;
2279 		case S_IFDIR:
2280 			/*
2281 			 * Optimize away all write operations, since the server
2282 			 * will check permissions when we perform the op.
2283 			 */
2284 			if ((mask & MAY_WRITE) && !(mask & MAY_READ))
2285 				goto out;
2286 	}
2287 
2288 force_lookup:
2289 	if (!NFS_PROTO(inode)->access)
2290 		goto out_notsup;
2291 
2292 	cred = rpc_lookup_cred();
2293 	if (!IS_ERR(cred)) {
2294 		res = nfs_do_access(inode, cred, mask);
2295 		put_rpccred(cred);
2296 	} else
2297 		res = PTR_ERR(cred);
2298 out:
2299 	if (!res && (mask & MAY_EXEC) && !execute_ok(inode))
2300 		res = -EACCES;
2301 
2302 	dfprintk(VFS, "NFS: permission(%s/%ld), mask=0x%x, res=%d\n",
2303 		inode->i_sb->s_id, inode->i_ino, mask, res);
2304 	return res;
2305 out_notsup:
2306 	res = nfs_revalidate_inode(NFS_SERVER(inode), inode);
2307 	if (res == 0)
2308 		res = generic_permission(inode, mask);
2309 	goto out;
2310 }
2311 EXPORT_SYMBOL_GPL(nfs_permission);
2312 
2313 /*
2314  * Local variables:
2315  *  version-control: t
2316  *  kept-new-versions: 5
2317  * End:
2318  */
2319