xref: /openbmc/linux/fs/nfs/dir.c (revision e868d61272caa648214046a096e5a6bfc068dc8c)
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/time.h>
21 #include <linux/errno.h>
22 #include <linux/stat.h>
23 #include <linux/fcntl.h>
24 #include <linux/string.h>
25 #include <linux/kernel.h>
26 #include <linux/slab.h>
27 #include <linux/mm.h>
28 #include <linux/sunrpc/clnt.h>
29 #include <linux/nfs_fs.h>
30 #include <linux/nfs_mount.h>
31 #include <linux/pagemap.h>
32 #include <linux/smp_lock.h>
33 #include <linux/pagevec.h>
34 #include <linux/namei.h>
35 #include <linux/mount.h>
36 
37 #include "nfs4_fs.h"
38 #include "delegation.h"
39 #include "iostat.h"
40 
41 /* #define NFS_DEBUG_VERBOSE 1 */
42 
43 static int nfs_opendir(struct inode *, struct file *);
44 static int nfs_readdir(struct file *, void *, filldir_t);
45 static struct dentry *nfs_lookup(struct inode *, struct dentry *, struct nameidata *);
46 static int nfs_create(struct inode *, struct dentry *, int, struct nameidata *);
47 static int nfs_mkdir(struct inode *, struct dentry *, int);
48 static int nfs_rmdir(struct inode *, struct dentry *);
49 static int nfs_unlink(struct inode *, struct dentry *);
50 static int nfs_symlink(struct inode *, struct dentry *, const char *);
51 static int nfs_link(struct dentry *, struct inode *, struct dentry *);
52 static int nfs_mknod(struct inode *, struct dentry *, int, dev_t);
53 static int nfs_rename(struct inode *, struct dentry *,
54 		      struct inode *, struct dentry *);
55 static int nfs_fsync_dir(struct file *, struct dentry *, int);
56 static loff_t nfs_llseek_dir(struct file *, loff_t, int);
57 
58 const struct file_operations nfs_dir_operations = {
59 	.llseek		= nfs_llseek_dir,
60 	.read		= generic_read_dir,
61 	.readdir	= nfs_readdir,
62 	.open		= nfs_opendir,
63 	.release	= nfs_release,
64 	.fsync		= nfs_fsync_dir,
65 };
66 
67 const struct inode_operations nfs_dir_inode_operations = {
68 	.create		= nfs_create,
69 	.lookup		= nfs_lookup,
70 	.link		= nfs_link,
71 	.unlink		= nfs_unlink,
72 	.symlink	= nfs_symlink,
73 	.mkdir		= nfs_mkdir,
74 	.rmdir		= nfs_rmdir,
75 	.mknod		= nfs_mknod,
76 	.rename		= nfs_rename,
77 	.permission	= nfs_permission,
78 	.getattr	= nfs_getattr,
79 	.setattr	= nfs_setattr,
80 };
81 
82 #ifdef CONFIG_NFS_V3
83 const struct inode_operations nfs3_dir_inode_operations = {
84 	.create		= nfs_create,
85 	.lookup		= nfs_lookup,
86 	.link		= nfs_link,
87 	.unlink		= nfs_unlink,
88 	.symlink	= nfs_symlink,
89 	.mkdir		= nfs_mkdir,
90 	.rmdir		= nfs_rmdir,
91 	.mknod		= nfs_mknod,
92 	.rename		= nfs_rename,
93 	.permission	= nfs_permission,
94 	.getattr	= nfs_getattr,
95 	.setattr	= nfs_setattr,
96 	.listxattr	= nfs3_listxattr,
97 	.getxattr	= nfs3_getxattr,
98 	.setxattr	= nfs3_setxattr,
99 	.removexattr	= nfs3_removexattr,
100 };
101 #endif  /* CONFIG_NFS_V3 */
102 
103 #ifdef CONFIG_NFS_V4
104 
105 static struct dentry *nfs_atomic_lookup(struct inode *, struct dentry *, struct nameidata *);
106 const struct inode_operations nfs4_dir_inode_operations = {
107 	.create		= nfs_create,
108 	.lookup		= nfs_atomic_lookup,
109 	.link		= nfs_link,
110 	.unlink		= nfs_unlink,
111 	.symlink	= nfs_symlink,
112 	.mkdir		= nfs_mkdir,
113 	.rmdir		= nfs_rmdir,
114 	.mknod		= nfs_mknod,
115 	.rename		= nfs_rename,
116 	.permission	= nfs_permission,
117 	.getattr	= nfs_getattr,
118 	.setattr	= nfs_setattr,
119 	.getxattr       = nfs4_getxattr,
120 	.setxattr       = nfs4_setxattr,
121 	.listxattr      = nfs4_listxattr,
122 };
123 
124 #endif /* CONFIG_NFS_V4 */
125 
126 /*
127  * Open file
128  */
129 static int
130 nfs_opendir(struct inode *inode, struct file *filp)
131 {
132 	int res;
133 
134 	dfprintk(VFS, "NFS: opendir(%s/%ld)\n",
135 			inode->i_sb->s_id, inode->i_ino);
136 
137 	lock_kernel();
138 	/* Call generic open code in order to cache credentials */
139 	res = nfs_open(inode, filp);
140 	unlock_kernel();
141 	return res;
142 }
143 
144 typedef __be32 * (*decode_dirent_t)(__be32 *, struct nfs_entry *, int);
145 typedef struct {
146 	struct file	*file;
147 	struct page	*page;
148 	unsigned long	page_index;
149 	__be32		*ptr;
150 	u64		*dir_cookie;
151 	loff_t		current_index;
152 	struct nfs_entry *entry;
153 	decode_dirent_t	decode;
154 	int		plus;
155 	int		error;
156 	unsigned long	timestamp;
157 	int		timestamp_valid;
158 } nfs_readdir_descriptor_t;
159 
160 /* Now we cache directories properly, by stuffing the dirent
161  * data directly in the page cache.
162  *
163  * Inode invalidation due to refresh etc. takes care of
164  * _everything_, no sloppy entry flushing logic, no extraneous
165  * copying, network direct to page cache, the way it was meant
166  * to be.
167  *
168  * NOTE: Dirent information verification is done always by the
169  *	 page-in of the RPC reply, nowhere else, this simplies
170  *	 things substantially.
171  */
172 static
173 int nfs_readdir_filler(nfs_readdir_descriptor_t *desc, struct page *page)
174 {
175 	struct file	*file = desc->file;
176 	struct inode	*inode = file->f_path.dentry->d_inode;
177 	struct rpc_cred	*cred = nfs_file_cred(file);
178 	unsigned long	timestamp;
179 	int		error;
180 
181 	dfprintk(DIRCACHE, "NFS: %s: reading cookie %Lu into page %lu\n",
182 			__FUNCTION__, (long long)desc->entry->cookie,
183 			page->index);
184 
185  again:
186 	timestamp = jiffies;
187 	error = NFS_PROTO(inode)->readdir(file->f_path.dentry, cred, desc->entry->cookie, page,
188 					  NFS_SERVER(inode)->dtsize, desc->plus);
189 	if (error < 0) {
190 		/* We requested READDIRPLUS, but the server doesn't grok it */
191 		if (error == -ENOTSUPP && desc->plus) {
192 			NFS_SERVER(inode)->caps &= ~NFS_CAP_READDIRPLUS;
193 			clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_FLAGS(inode));
194 			desc->plus = 0;
195 			goto again;
196 		}
197 		goto error;
198 	}
199 	desc->timestamp = timestamp;
200 	desc->timestamp_valid = 1;
201 	SetPageUptodate(page);
202 	spin_lock(&inode->i_lock);
203 	NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ATIME;
204 	spin_unlock(&inode->i_lock);
205 	/* Ensure consistent page alignment of the data.
206 	 * Note: assumes we have exclusive access to this mapping either
207 	 *	 through inode->i_mutex or some other mechanism.
208 	 */
209 	if (page->index == 0 && invalidate_inode_pages2_range(inode->i_mapping, PAGE_CACHE_SIZE, -1) < 0) {
210 		/* Should never happen */
211 		nfs_zap_mapping(inode, inode->i_mapping);
212 	}
213 	unlock_page(page);
214 	return 0;
215  error:
216 	SetPageError(page);
217 	unlock_page(page);
218 	nfs_zap_caches(inode);
219 	desc->error = error;
220 	return -EIO;
221 }
222 
223 static inline
224 int dir_decode(nfs_readdir_descriptor_t *desc)
225 {
226 	__be32	*p = desc->ptr;
227 	p = desc->decode(p, desc->entry, desc->plus);
228 	if (IS_ERR(p))
229 		return PTR_ERR(p);
230 	desc->ptr = p;
231 	if (desc->timestamp_valid)
232 		desc->entry->fattr->time_start = desc->timestamp;
233 	else
234 		desc->entry->fattr->valid &= ~NFS_ATTR_FATTR;
235 	return 0;
236 }
237 
238 static inline
239 void dir_page_release(nfs_readdir_descriptor_t *desc)
240 {
241 	kunmap(desc->page);
242 	page_cache_release(desc->page);
243 	desc->page = NULL;
244 	desc->ptr = NULL;
245 }
246 
247 /*
248  * Given a pointer to a buffer that has already been filled by a call
249  * to readdir, find the next entry with cookie '*desc->dir_cookie'.
250  *
251  * If the end of the buffer has been reached, return -EAGAIN, if not,
252  * return the offset within the buffer of the next entry to be
253  * read.
254  */
255 static inline
256 int find_dirent(nfs_readdir_descriptor_t *desc)
257 {
258 	struct nfs_entry *entry = desc->entry;
259 	int		loop_count = 0,
260 			status;
261 
262 	while((status = dir_decode(desc)) == 0) {
263 		dfprintk(DIRCACHE, "NFS: %s: examining cookie %Lu\n",
264 				__FUNCTION__, (unsigned long long)entry->cookie);
265 		if (entry->prev_cookie == *desc->dir_cookie)
266 			break;
267 		if (loop_count++ > 200) {
268 			loop_count = 0;
269 			schedule();
270 		}
271 	}
272 	return status;
273 }
274 
275 /*
276  * Given a pointer to a buffer that has already been filled by a call
277  * to readdir, find the entry at offset 'desc->file->f_pos'.
278  *
279  * If the end of the buffer has been reached, return -EAGAIN, if not,
280  * return the offset within the buffer of the next entry to be
281  * read.
282  */
283 static inline
284 int find_dirent_index(nfs_readdir_descriptor_t *desc)
285 {
286 	struct nfs_entry *entry = desc->entry;
287 	int		loop_count = 0,
288 			status;
289 
290 	for(;;) {
291 		status = dir_decode(desc);
292 		if (status)
293 			break;
294 
295 		dfprintk(DIRCACHE, "NFS: found cookie %Lu at index %Ld\n",
296 				(unsigned long long)entry->cookie, desc->current_index);
297 
298 		if (desc->file->f_pos == desc->current_index) {
299 			*desc->dir_cookie = entry->cookie;
300 			break;
301 		}
302 		desc->current_index++;
303 		if (loop_count++ > 200) {
304 			loop_count = 0;
305 			schedule();
306 		}
307 	}
308 	return status;
309 }
310 
311 /*
312  * Find the given page, and call find_dirent() or find_dirent_index in
313  * order to try to return the next entry.
314  */
315 static inline
316 int find_dirent_page(nfs_readdir_descriptor_t *desc)
317 {
318 	struct inode	*inode = desc->file->f_path.dentry->d_inode;
319 	struct page	*page;
320 	int		status;
321 
322 	dfprintk(DIRCACHE, "NFS: %s: searching page %ld for target %Lu\n",
323 			__FUNCTION__, desc->page_index,
324 			(long long) *desc->dir_cookie);
325 
326 	/* If we find the page in the page_cache, we cannot be sure
327 	 * how fresh the data is, so we will ignore readdir_plus attributes.
328 	 */
329 	desc->timestamp_valid = 0;
330 	page = read_cache_page(inode->i_mapping, desc->page_index,
331 			       (filler_t *)nfs_readdir_filler, desc);
332 	if (IS_ERR(page)) {
333 		status = PTR_ERR(page);
334 		goto out;
335 	}
336 
337 	/* NOTE: Someone else may have changed the READDIRPLUS flag */
338 	desc->page = page;
339 	desc->ptr = kmap(page);		/* matching kunmap in nfs_do_filldir */
340 	if (*desc->dir_cookie != 0)
341 		status = find_dirent(desc);
342 	else
343 		status = find_dirent_index(desc);
344 	if (status < 0)
345 		dir_page_release(desc);
346  out:
347 	dfprintk(DIRCACHE, "NFS: %s: returns %d\n", __FUNCTION__, status);
348 	return status;
349 }
350 
351 /*
352  * Recurse through the page cache pages, and return a
353  * filled nfs_entry structure of the next directory entry if possible.
354  *
355  * The target for the search is '*desc->dir_cookie' if non-0,
356  * 'desc->file->f_pos' otherwise
357  */
358 static inline
359 int readdir_search_pagecache(nfs_readdir_descriptor_t *desc)
360 {
361 	int		loop_count = 0;
362 	int		res;
363 
364 	/* Always search-by-index from the beginning of the cache */
365 	if (*desc->dir_cookie == 0) {
366 		dfprintk(DIRCACHE, "NFS: readdir_search_pagecache() searching for offset %Ld\n",
367 				(long long)desc->file->f_pos);
368 		desc->page_index = 0;
369 		desc->entry->cookie = desc->entry->prev_cookie = 0;
370 		desc->entry->eof = 0;
371 		desc->current_index = 0;
372 	} else
373 		dfprintk(DIRCACHE, "NFS: readdir_search_pagecache() searching for cookie %Lu\n",
374 				(unsigned long long)*desc->dir_cookie);
375 
376 	for (;;) {
377 		res = find_dirent_page(desc);
378 		if (res != -EAGAIN)
379 			break;
380 		/* Align to beginning of next page */
381 		desc->page_index ++;
382 		if (loop_count++ > 200) {
383 			loop_count = 0;
384 			schedule();
385 		}
386 	}
387 
388 	dfprintk(DIRCACHE, "NFS: %s: returns %d\n", __FUNCTION__, res);
389 	return res;
390 }
391 
392 static inline unsigned int dt_type(struct inode *inode)
393 {
394 	return (inode->i_mode >> 12) & 15;
395 }
396 
397 static struct dentry *nfs_readdir_lookup(nfs_readdir_descriptor_t *desc);
398 
399 /*
400  * Once we've found the start of the dirent within a page: fill 'er up...
401  */
402 static
403 int nfs_do_filldir(nfs_readdir_descriptor_t *desc, void *dirent,
404 		   filldir_t filldir)
405 {
406 	struct file	*file = desc->file;
407 	struct nfs_entry *entry = desc->entry;
408 	struct dentry	*dentry = NULL;
409 	unsigned long	fileid;
410 	int		loop_count = 0,
411 			res;
412 
413 	dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling starting @ cookie %Lu\n",
414 			(unsigned long long)entry->cookie);
415 
416 	for(;;) {
417 		unsigned d_type = DT_UNKNOWN;
418 		/* Note: entry->prev_cookie contains the cookie for
419 		 *	 retrieving the current dirent on the server */
420 		fileid = nfs_fileid_to_ino_t(entry->ino);
421 
422 		/* Get a dentry if we have one */
423 		if (dentry != NULL)
424 			dput(dentry);
425 		dentry = nfs_readdir_lookup(desc);
426 
427 		/* Use readdirplus info */
428 		if (dentry != NULL && dentry->d_inode != NULL) {
429 			d_type = dt_type(dentry->d_inode);
430 			fileid = dentry->d_inode->i_ino;
431 		}
432 
433 		res = filldir(dirent, entry->name, entry->len,
434 			      file->f_pos, fileid, d_type);
435 		if (res < 0)
436 			break;
437 		file->f_pos++;
438 		*desc->dir_cookie = entry->cookie;
439 		if (dir_decode(desc) != 0) {
440 			desc->page_index ++;
441 			break;
442 		}
443 		if (loop_count++ > 200) {
444 			loop_count = 0;
445 			schedule();
446 		}
447 	}
448 	dir_page_release(desc);
449 	if (dentry != NULL)
450 		dput(dentry);
451 	dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n",
452 			(unsigned long long)*desc->dir_cookie, res);
453 	return res;
454 }
455 
456 /*
457  * If we cannot find a cookie in our cache, we suspect that this is
458  * because it points to a deleted file, so we ask the server to return
459  * whatever it thinks is the next entry. We then feed this to filldir.
460  * If all goes well, we should then be able to find our way round the
461  * cache on the next call to readdir_search_pagecache();
462  *
463  * NOTE: we cannot add the anonymous page to the pagecache because
464  *	 the data it contains might not be page aligned. Besides,
465  *	 we should already have a complete representation of the
466  *	 directory in the page cache by the time we get here.
467  */
468 static inline
469 int uncached_readdir(nfs_readdir_descriptor_t *desc, void *dirent,
470 		     filldir_t filldir)
471 {
472 	struct file	*file = desc->file;
473 	struct inode	*inode = file->f_path.dentry->d_inode;
474 	struct rpc_cred	*cred = nfs_file_cred(file);
475 	struct page	*page = NULL;
476 	int		status;
477 	unsigned long	timestamp;
478 
479 	dfprintk(DIRCACHE, "NFS: uncached_readdir() searching for cookie %Lu\n",
480 			(unsigned long long)*desc->dir_cookie);
481 
482 	page = alloc_page(GFP_HIGHUSER);
483 	if (!page) {
484 		status = -ENOMEM;
485 		goto out;
486 	}
487 	timestamp = jiffies;
488 	desc->error = NFS_PROTO(inode)->readdir(file->f_path.dentry, cred, *desc->dir_cookie,
489 						page,
490 						NFS_SERVER(inode)->dtsize,
491 						desc->plus);
492 	spin_lock(&inode->i_lock);
493 	NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ATIME;
494 	spin_unlock(&inode->i_lock);
495 	desc->page = page;
496 	desc->ptr = kmap(page);		/* matching kunmap in nfs_do_filldir */
497 	if (desc->error >= 0) {
498 		desc->timestamp = timestamp;
499 		desc->timestamp_valid = 1;
500 		if ((status = dir_decode(desc)) == 0)
501 			desc->entry->prev_cookie = *desc->dir_cookie;
502 	} else
503 		status = -EIO;
504 	if (status < 0)
505 		goto out_release;
506 
507 	status = nfs_do_filldir(desc, dirent, filldir);
508 
509 	/* Reset read descriptor so it searches the page cache from
510 	 * the start upon the next call to readdir_search_pagecache() */
511 	desc->page_index = 0;
512 	desc->entry->cookie = desc->entry->prev_cookie = 0;
513 	desc->entry->eof = 0;
514  out:
515 	dfprintk(DIRCACHE, "NFS: %s: returns %d\n",
516 			__FUNCTION__, status);
517 	return status;
518  out_release:
519 	dir_page_release(desc);
520 	goto out;
521 }
522 
523 /* The file offset position represents the dirent entry number.  A
524    last cookie cache takes care of the common case of reading the
525    whole directory.
526  */
527 static int nfs_readdir(struct file *filp, void *dirent, filldir_t filldir)
528 {
529 	struct dentry	*dentry = filp->f_path.dentry;
530 	struct inode	*inode = dentry->d_inode;
531 	nfs_readdir_descriptor_t my_desc,
532 			*desc = &my_desc;
533 	struct nfs_entry my_entry;
534 	struct nfs_fh	 fh;
535 	struct nfs_fattr fattr;
536 	long		res;
537 
538 	dfprintk(VFS, "NFS: readdir(%s/%s) starting at cookie %Lu\n",
539 			dentry->d_parent->d_name.name, dentry->d_name.name,
540 			(long long)filp->f_pos);
541 	nfs_inc_stats(inode, NFSIOS_VFSGETDENTS);
542 
543 	lock_kernel();
544 
545 	res = nfs_revalidate_mapping_nolock(inode, filp->f_mapping);
546 	if (res < 0) {
547 		unlock_kernel();
548 		return res;
549 	}
550 
551 	/*
552 	 * filp->f_pos points to the dirent entry number.
553 	 * *desc->dir_cookie has the cookie for the next entry. We have
554 	 * to either find the entry with the appropriate number or
555 	 * revalidate the cookie.
556 	 */
557 	memset(desc, 0, sizeof(*desc));
558 
559 	desc->file = filp;
560 	desc->dir_cookie = &((struct nfs_open_context *)filp->private_data)->dir_cookie;
561 	desc->decode = NFS_PROTO(inode)->decode_dirent;
562 	desc->plus = NFS_USE_READDIRPLUS(inode);
563 
564 	my_entry.cookie = my_entry.prev_cookie = 0;
565 	my_entry.eof = 0;
566 	my_entry.fh = &fh;
567 	my_entry.fattr = &fattr;
568 	nfs_fattr_init(&fattr);
569 	desc->entry = &my_entry;
570 
571 	while(!desc->entry->eof) {
572 		res = readdir_search_pagecache(desc);
573 
574 		if (res == -EBADCOOKIE) {
575 			/* This means either end of directory */
576 			if (*desc->dir_cookie && desc->entry->cookie != *desc->dir_cookie) {
577 				/* Or that the server has 'lost' a cookie */
578 				res = uncached_readdir(desc, dirent, filldir);
579 				if (res >= 0)
580 					continue;
581 			}
582 			res = 0;
583 			break;
584 		}
585 		if (res == -ETOOSMALL && desc->plus) {
586 			clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_FLAGS(inode));
587 			nfs_zap_caches(inode);
588 			desc->plus = 0;
589 			desc->entry->eof = 0;
590 			continue;
591 		}
592 		if (res < 0)
593 			break;
594 
595 		res = nfs_do_filldir(desc, dirent, filldir);
596 		if (res < 0) {
597 			res = 0;
598 			break;
599 		}
600 	}
601 	unlock_kernel();
602 	if (res > 0)
603 		res = 0;
604 	dfprintk(VFS, "NFS: readdir(%s/%s) returns %ld\n",
605 			dentry->d_parent->d_name.name, dentry->d_name.name,
606 			res);
607 	return res;
608 }
609 
610 loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int origin)
611 {
612 	mutex_lock(&filp->f_path.dentry->d_inode->i_mutex);
613 	switch (origin) {
614 		case 1:
615 			offset += filp->f_pos;
616 		case 0:
617 			if (offset >= 0)
618 				break;
619 		default:
620 			offset = -EINVAL;
621 			goto out;
622 	}
623 	if (offset != filp->f_pos) {
624 		filp->f_pos = offset;
625 		((struct nfs_open_context *)filp->private_data)->dir_cookie = 0;
626 	}
627 out:
628 	mutex_unlock(&filp->f_path.dentry->d_inode->i_mutex);
629 	return offset;
630 }
631 
632 /*
633  * All directory operations under NFS are synchronous, so fsync()
634  * is a dummy operation.
635  */
636 int nfs_fsync_dir(struct file *filp, struct dentry *dentry, int datasync)
637 {
638 	dfprintk(VFS, "NFS: fsync_dir(%s/%s) datasync %d\n",
639 			dentry->d_parent->d_name.name, dentry->d_name.name,
640 			datasync);
641 
642 	return 0;
643 }
644 
645 /*
646  * A check for whether or not the parent directory has changed.
647  * In the case it has, we assume that the dentries are untrustworthy
648  * and may need to be looked up again.
649  */
650 static int nfs_check_verifier(struct inode *dir, struct dentry *dentry)
651 {
652 	unsigned long verf;
653 
654 	if (IS_ROOT(dentry))
655 		return 1;
656 	verf = (unsigned long)dentry->d_fsdata;
657 	if (nfs_caches_unstable(dir)
658 			|| verf != NFS_I(dir)->cache_change_attribute)
659 		return 0;
660 	return 1;
661 }
662 
663 static inline void nfs_set_verifier(struct dentry * dentry, unsigned long verf)
664 {
665 	dentry->d_fsdata = (void *)verf;
666 }
667 
668 static void nfs_refresh_verifier(struct dentry * dentry, unsigned long verf)
669 {
670 	nfs_set_verifier(dentry, verf);
671 }
672 
673 /*
674  * Whenever an NFS operation succeeds, we know that the dentry
675  * is valid, so we update the revalidation timestamp.
676  */
677 static inline void nfs_renew_times(struct dentry * dentry)
678 {
679 	dentry->d_time = jiffies;
680 }
681 
682 /*
683  * Return the intent data that applies to this particular path component
684  *
685  * Note that the current set of intents only apply to the very last
686  * component of the path.
687  * We check for this using LOOKUP_CONTINUE and LOOKUP_PARENT.
688  */
689 static inline unsigned int nfs_lookup_check_intent(struct nameidata *nd, unsigned int mask)
690 {
691 	if (nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))
692 		return 0;
693 	return nd->flags & mask;
694 }
695 
696 /*
697  * Inode and filehandle revalidation for lookups.
698  *
699  * We force revalidation in the cases where the VFS sets LOOKUP_REVAL,
700  * or if the intent information indicates that we're about to open this
701  * particular file and the "nocto" mount flag is not set.
702  *
703  */
704 static inline
705 int nfs_lookup_verify_inode(struct inode *inode, struct nameidata *nd)
706 {
707 	struct nfs_server *server = NFS_SERVER(inode);
708 
709 	if (nd != NULL) {
710 		/* VFS wants an on-the-wire revalidation */
711 		if (nd->flags & LOOKUP_REVAL)
712 			goto out_force;
713 		/* This is an open(2) */
714 		if (nfs_lookup_check_intent(nd, LOOKUP_OPEN) != 0 &&
715 				!(server->flags & NFS_MOUNT_NOCTO) &&
716 				(S_ISREG(inode->i_mode) ||
717 				 S_ISDIR(inode->i_mode)))
718 			goto out_force;
719 	}
720 	return nfs_revalidate_inode(server, inode);
721 out_force:
722 	return __nfs_revalidate_inode(server, inode);
723 }
724 
725 /*
726  * We judge how long we want to trust negative
727  * dentries by looking at the parent inode mtime.
728  *
729  * If parent mtime has changed, we revalidate, else we wait for a
730  * period corresponding to the parent's attribute cache timeout value.
731  */
732 static inline
733 int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry,
734 		       struct nameidata *nd)
735 {
736 	/* Don't revalidate a negative dentry if we're creating a new file */
737 	if (nd != NULL && nfs_lookup_check_intent(nd, LOOKUP_CREATE) != 0)
738 		return 0;
739 	return !nfs_check_verifier(dir, dentry);
740 }
741 
742 /*
743  * This is called every time the dcache has a lookup hit,
744  * and we should check whether we can really trust that
745  * lookup.
746  *
747  * NOTE! The hit can be a negative hit too, don't assume
748  * we have an inode!
749  *
750  * If the parent directory is seen to have changed, we throw out the
751  * cached dentry and do a new lookup.
752  */
753 static int nfs_lookup_revalidate(struct dentry * dentry, struct nameidata *nd)
754 {
755 	struct inode *dir;
756 	struct inode *inode;
757 	struct dentry *parent;
758 	int error;
759 	struct nfs_fh fhandle;
760 	struct nfs_fattr fattr;
761 	unsigned long verifier;
762 
763 	parent = dget_parent(dentry);
764 	lock_kernel();
765 	dir = parent->d_inode;
766 	nfs_inc_stats(dir, NFSIOS_DENTRYREVALIDATE);
767 	inode = dentry->d_inode;
768 
769 	/* Revalidate parent directory attribute cache */
770 	if (nfs_revalidate_inode(NFS_SERVER(dir), dir) < 0)
771 		goto out_zap_parent;
772 
773 	if (!inode) {
774 		if (nfs_neg_need_reval(dir, dentry, nd))
775 			goto out_bad;
776 		goto out_valid;
777 	}
778 
779 	if (is_bad_inode(inode)) {
780 		dfprintk(LOOKUPCACHE, "%s: %s/%s has dud inode\n",
781 				__FUNCTION__, dentry->d_parent->d_name.name,
782 				dentry->d_name.name);
783 		goto out_bad;
784 	}
785 
786 	/* Force a full look up iff the parent directory has changed */
787 	if (nfs_check_verifier(dir, dentry)) {
788 		if (nfs_lookup_verify_inode(inode, nd))
789 			goto out_zap_parent;
790 		goto out_valid;
791 	}
792 
793 	if (NFS_STALE(inode))
794 		goto out_bad;
795 
796 	verifier = nfs_save_change_attribute(dir);
797 	error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, &fhandle, &fattr);
798 	if (error)
799 		goto out_bad;
800 	if (nfs_compare_fh(NFS_FH(inode), &fhandle))
801 		goto out_bad;
802 	if ((error = nfs_refresh_inode(inode, &fattr)) != 0)
803 		goto out_bad;
804 
805 	nfs_renew_times(dentry);
806 	nfs_refresh_verifier(dentry, verifier);
807  out_valid:
808 	unlock_kernel();
809 	dput(parent);
810 	dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is valid\n",
811 			__FUNCTION__, dentry->d_parent->d_name.name,
812 			dentry->d_name.name);
813 	return 1;
814 out_zap_parent:
815 	nfs_zap_caches(dir);
816  out_bad:
817 	NFS_CACHEINV(dir);
818 	if (inode && S_ISDIR(inode->i_mode)) {
819 		/* Purge readdir caches. */
820 		nfs_zap_caches(inode);
821 		/* If we have submounts, don't unhash ! */
822 		if (have_submounts(dentry))
823 			goto out_valid;
824 		shrink_dcache_parent(dentry);
825 	}
826 	d_drop(dentry);
827 	unlock_kernel();
828 	dput(parent);
829 	dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is invalid\n",
830 			__FUNCTION__, dentry->d_parent->d_name.name,
831 			dentry->d_name.name);
832 	return 0;
833 }
834 
835 /*
836  * This is called from dput() when d_count is going to 0.
837  */
838 static int nfs_dentry_delete(struct dentry *dentry)
839 {
840 	dfprintk(VFS, "NFS: dentry_delete(%s/%s, %x)\n",
841 		dentry->d_parent->d_name.name, dentry->d_name.name,
842 		dentry->d_flags);
843 
844 	if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
845 		/* Unhash it, so that ->d_iput() would be called */
846 		return 1;
847 	}
848 	if (!(dentry->d_sb->s_flags & MS_ACTIVE)) {
849 		/* Unhash it, so that ancestors of killed async unlink
850 		 * files will be cleaned up during umount */
851 		return 1;
852 	}
853 	return 0;
854 
855 }
856 
857 /*
858  * Called when the dentry loses inode.
859  * We use it to clean up silly-renamed files.
860  */
861 static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode)
862 {
863 	nfs_inode_return_delegation(inode);
864 	if (S_ISDIR(inode->i_mode))
865 		/* drop any readdir cache as it could easily be old */
866 		NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA;
867 
868 	if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
869 		lock_kernel();
870 		drop_nlink(inode);
871 		nfs_complete_unlink(dentry);
872 		unlock_kernel();
873 	}
874 	/* When creating a negative dentry, we want to renew d_time */
875 	nfs_renew_times(dentry);
876 	iput(inode);
877 }
878 
879 struct dentry_operations nfs_dentry_operations = {
880 	.d_revalidate	= nfs_lookup_revalidate,
881 	.d_delete	= nfs_dentry_delete,
882 	.d_iput		= nfs_dentry_iput,
883 };
884 
885 /*
886  * Use intent information to check whether or not we're going to do
887  * an O_EXCL create using this path component.
888  */
889 static inline
890 int nfs_is_exclusive_create(struct inode *dir, struct nameidata *nd)
891 {
892 	if (NFS_PROTO(dir)->version == 2)
893 		return 0;
894 	if (nd == NULL || nfs_lookup_check_intent(nd, LOOKUP_CREATE) == 0)
895 		return 0;
896 	return (nd->intent.open.flags & O_EXCL) != 0;
897 }
898 
899 static inline int nfs_reval_fsid(struct vfsmount *mnt, struct inode *dir,
900 				 struct nfs_fh *fh, struct nfs_fattr *fattr)
901 {
902 	struct nfs_server *server = NFS_SERVER(dir);
903 
904 	if (!nfs_fsid_equal(&server->fsid, &fattr->fsid))
905 		/* Revalidate fsid on root dir */
906 		return __nfs_revalidate_inode(server, mnt->mnt_root->d_inode);
907 	return 0;
908 }
909 
910 static struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
911 {
912 	struct dentry *res;
913 	struct inode *inode = NULL;
914 	int error;
915 	struct nfs_fh fhandle;
916 	struct nfs_fattr fattr;
917 
918 	dfprintk(VFS, "NFS: lookup(%s/%s)\n",
919 		dentry->d_parent->d_name.name, dentry->d_name.name);
920 	nfs_inc_stats(dir, NFSIOS_VFSLOOKUP);
921 
922 	res = ERR_PTR(-ENAMETOOLONG);
923 	if (dentry->d_name.len > NFS_SERVER(dir)->namelen)
924 		goto out;
925 
926 	res = ERR_PTR(-ENOMEM);
927 	dentry->d_op = NFS_PROTO(dir)->dentry_ops;
928 
929 	lock_kernel();
930 
931 	/*
932 	 * If we're doing an exclusive create, optimize away the lookup
933 	 * but don't hash the dentry.
934 	 */
935 	if (nfs_is_exclusive_create(dir, nd)) {
936 		d_instantiate(dentry, NULL);
937 		res = NULL;
938 		goto out_unlock;
939 	}
940 
941 	error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, &fhandle, &fattr);
942 	if (error == -ENOENT)
943 		goto no_entry;
944 	if (error < 0) {
945 		res = ERR_PTR(error);
946 		goto out_unlock;
947 	}
948 	error = nfs_reval_fsid(nd->mnt, dir, &fhandle, &fattr);
949 	if (error < 0) {
950 		res = ERR_PTR(error);
951 		goto out_unlock;
952 	}
953 	inode = nfs_fhget(dentry->d_sb, &fhandle, &fattr);
954 	res = (struct dentry *)inode;
955 	if (IS_ERR(res))
956 		goto out_unlock;
957 
958 no_entry:
959 	res = d_materialise_unique(dentry, inode);
960 	if (res != NULL) {
961 		struct dentry *parent;
962 		if (IS_ERR(res))
963 			goto out_unlock;
964 		/* Was a directory renamed! */
965 		parent = dget_parent(res);
966 		if (!IS_ROOT(parent))
967 			nfs_mark_for_revalidate(parent->d_inode);
968 		dput(parent);
969 		dentry = res;
970 	}
971 	nfs_renew_times(dentry);
972 	nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
973 out_unlock:
974 	unlock_kernel();
975 out:
976 	return res;
977 }
978 
979 #ifdef CONFIG_NFS_V4
980 static int nfs_open_revalidate(struct dentry *, struct nameidata *);
981 
982 struct dentry_operations nfs4_dentry_operations = {
983 	.d_revalidate	= nfs_open_revalidate,
984 	.d_delete	= nfs_dentry_delete,
985 	.d_iput		= nfs_dentry_iput,
986 };
987 
988 /*
989  * Use intent information to determine whether we need to substitute
990  * the NFSv4-style stateful OPEN for the LOOKUP call
991  */
992 static int is_atomic_open(struct inode *dir, struct nameidata *nd)
993 {
994 	if (nd == NULL || nfs_lookup_check_intent(nd, LOOKUP_OPEN) == 0)
995 		return 0;
996 	/* NFS does not (yet) have a stateful open for directories */
997 	if (nd->flags & LOOKUP_DIRECTORY)
998 		return 0;
999 	/* Are we trying to write to a read only partition? */
1000 	if (IS_RDONLY(dir) && (nd->intent.open.flags & (O_CREAT|O_TRUNC|FMODE_WRITE)))
1001 		return 0;
1002 	return 1;
1003 }
1004 
1005 static struct dentry *nfs_atomic_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd)
1006 {
1007 	struct dentry *res = NULL;
1008 	int error;
1009 
1010 	dfprintk(VFS, "NFS: atomic_lookup(%s/%ld), %s\n",
1011 			dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1012 
1013 	/* Check that we are indeed trying to open this file */
1014 	if (!is_atomic_open(dir, nd))
1015 		goto no_open;
1016 
1017 	if (dentry->d_name.len > NFS_SERVER(dir)->namelen) {
1018 		res = ERR_PTR(-ENAMETOOLONG);
1019 		goto out;
1020 	}
1021 	dentry->d_op = NFS_PROTO(dir)->dentry_ops;
1022 
1023 	/* Let vfs_create() deal with O_EXCL */
1024 	if (nd->intent.open.flags & O_EXCL) {
1025 		d_add(dentry, NULL);
1026 		goto out;
1027 	}
1028 
1029 	/* Open the file on the server */
1030 	lock_kernel();
1031 	/* Revalidate parent directory attribute cache */
1032 	error = nfs_revalidate_inode(NFS_SERVER(dir), dir);
1033 	if (error < 0) {
1034 		res = ERR_PTR(error);
1035 		unlock_kernel();
1036 		goto out;
1037 	}
1038 
1039 	if (nd->intent.open.flags & O_CREAT) {
1040 		nfs_begin_data_update(dir);
1041 		res = nfs4_atomic_open(dir, dentry, nd);
1042 		nfs_end_data_update(dir);
1043 	} else
1044 		res = nfs4_atomic_open(dir, dentry, nd);
1045 	unlock_kernel();
1046 	if (IS_ERR(res)) {
1047 		error = PTR_ERR(res);
1048 		switch (error) {
1049 			/* Make a negative dentry */
1050 			case -ENOENT:
1051 				res = NULL;
1052 				goto out;
1053 			/* This turned out not to be a regular file */
1054 			case -EISDIR:
1055 			case -ENOTDIR:
1056 				goto no_open;
1057 			case -ELOOP:
1058 				if (!(nd->intent.open.flags & O_NOFOLLOW))
1059 					goto no_open;
1060 			/* case -EINVAL: */
1061 			default:
1062 				goto out;
1063 		}
1064 	} else if (res != NULL)
1065 		dentry = res;
1066 	nfs_renew_times(dentry);
1067 	nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1068 out:
1069 	return res;
1070 no_open:
1071 	return nfs_lookup(dir, dentry, nd);
1072 }
1073 
1074 static int nfs_open_revalidate(struct dentry *dentry, struct nameidata *nd)
1075 {
1076 	struct dentry *parent = NULL;
1077 	struct inode *inode = dentry->d_inode;
1078 	struct inode *dir;
1079 	unsigned long verifier;
1080 	int openflags, ret = 0;
1081 
1082 	parent = dget_parent(dentry);
1083 	dir = parent->d_inode;
1084 	if (!is_atomic_open(dir, nd))
1085 		goto no_open;
1086 	/* We can't create new files in nfs_open_revalidate(), so we
1087 	 * optimize away revalidation of negative dentries.
1088 	 */
1089 	if (inode == NULL)
1090 		goto out;
1091 	/* NFS only supports OPEN on regular files */
1092 	if (!S_ISREG(inode->i_mode))
1093 		goto no_open;
1094 	openflags = nd->intent.open.flags;
1095 	/* We cannot do exclusive creation on a positive dentry */
1096 	if ((openflags & (O_CREAT|O_EXCL)) == (O_CREAT|O_EXCL))
1097 		goto no_open;
1098 	/* We can't create new files, or truncate existing ones here */
1099 	openflags &= ~(O_CREAT|O_TRUNC);
1100 
1101 	/*
1102 	 * Note: we're not holding inode->i_mutex and so may be racing with
1103 	 * operations that change the directory. We therefore save the
1104 	 * change attribute *before* we do the RPC call.
1105 	 */
1106 	lock_kernel();
1107 	verifier = nfs_save_change_attribute(dir);
1108 	ret = nfs4_open_revalidate(dir, dentry, openflags, nd);
1109 	if (!ret)
1110 		nfs_refresh_verifier(dentry, verifier);
1111 	unlock_kernel();
1112 out:
1113 	dput(parent);
1114 	if (!ret)
1115 		d_drop(dentry);
1116 	return ret;
1117 no_open:
1118 	dput(parent);
1119 	if (inode != NULL && nfs_have_delegation(inode, FMODE_READ))
1120 		return 1;
1121 	return nfs_lookup_revalidate(dentry, nd);
1122 }
1123 #endif /* CONFIG_NFSV4 */
1124 
1125 static struct dentry *nfs_readdir_lookup(nfs_readdir_descriptor_t *desc)
1126 {
1127 	struct dentry *parent = desc->file->f_path.dentry;
1128 	struct inode *dir = parent->d_inode;
1129 	struct nfs_entry *entry = desc->entry;
1130 	struct dentry *dentry, *alias;
1131 	struct qstr name = {
1132 		.name = entry->name,
1133 		.len = entry->len,
1134 	};
1135 	struct inode *inode;
1136 
1137 	switch (name.len) {
1138 		case 2:
1139 			if (name.name[0] == '.' && name.name[1] == '.')
1140 				return dget_parent(parent);
1141 			break;
1142 		case 1:
1143 			if (name.name[0] == '.')
1144 				return dget(parent);
1145 	}
1146 	name.hash = full_name_hash(name.name, name.len);
1147 	dentry = d_lookup(parent, &name);
1148 	if (dentry != NULL) {
1149 		/* Is this a positive dentry that matches the readdir info? */
1150 		if (dentry->d_inode != NULL &&
1151 				(NFS_FILEID(dentry->d_inode) == entry->ino ||
1152 				d_mountpoint(dentry))) {
1153 			if (!desc->plus || entry->fh->size == 0)
1154 				return dentry;
1155 			if (nfs_compare_fh(NFS_FH(dentry->d_inode),
1156 						entry->fh) == 0)
1157 				goto out_renew;
1158 		}
1159 		/* No, so d_drop to allow one to be created */
1160 		d_drop(dentry);
1161 		dput(dentry);
1162 	}
1163 	if (!desc->plus || !(entry->fattr->valid & NFS_ATTR_FATTR))
1164 		return NULL;
1165 	/* Note: caller is already holding the dir->i_mutex! */
1166 	dentry = d_alloc(parent, &name);
1167 	if (dentry == NULL)
1168 		return NULL;
1169 	dentry->d_op = NFS_PROTO(dir)->dentry_ops;
1170 	inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr);
1171 	if (IS_ERR(inode)) {
1172 		dput(dentry);
1173 		return NULL;
1174 	}
1175 
1176 	alias = d_materialise_unique(dentry, inode);
1177 	if (alias != NULL) {
1178 		dput(dentry);
1179 		if (IS_ERR(alias))
1180 			return NULL;
1181 		dentry = alias;
1182 	}
1183 
1184 	nfs_renew_times(dentry);
1185 	nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1186 	return dentry;
1187 out_renew:
1188 	nfs_renew_times(dentry);
1189 	nfs_refresh_verifier(dentry, nfs_save_change_attribute(dir));
1190 	return dentry;
1191 }
1192 
1193 /*
1194  * Code common to create, mkdir, and mknod.
1195  */
1196 int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle,
1197 				struct nfs_fattr *fattr)
1198 {
1199 	struct inode *inode;
1200 	int error = -EACCES;
1201 
1202 	/* We may have been initialized further down */
1203 	if (dentry->d_inode)
1204 		return 0;
1205 	if (fhandle->size == 0) {
1206 		struct inode *dir = dentry->d_parent->d_inode;
1207 		error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr);
1208 		if (error)
1209 			return error;
1210 	}
1211 	if (!(fattr->valid & NFS_ATTR_FATTR)) {
1212 		struct nfs_server *server = NFS_SB(dentry->d_sb);
1213 		error = server->nfs_client->rpc_ops->getattr(server, fhandle, fattr);
1214 		if (error < 0)
1215 			return error;
1216 	}
1217 	inode = nfs_fhget(dentry->d_sb, fhandle, fattr);
1218 	error = PTR_ERR(inode);
1219 	if (IS_ERR(inode))
1220 		return error;
1221 	d_instantiate(dentry, inode);
1222 	if (d_unhashed(dentry))
1223 		d_rehash(dentry);
1224 	return 0;
1225 }
1226 
1227 /*
1228  * Following a failed create operation, we drop the dentry rather
1229  * than retain a negative dentry. This avoids a problem in the event
1230  * that the operation succeeded on the server, but an error in the
1231  * reply path made it appear to have failed.
1232  */
1233 static int nfs_create(struct inode *dir, struct dentry *dentry, int mode,
1234 		struct nameidata *nd)
1235 {
1236 	struct iattr attr;
1237 	int error;
1238 	int open_flags = 0;
1239 
1240 	dfprintk(VFS, "NFS: create(%s/%ld), %s\n",
1241 			dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1242 
1243 	attr.ia_mode = mode;
1244 	attr.ia_valid = ATTR_MODE;
1245 
1246 	if (nd && (nd->flags & LOOKUP_CREATE))
1247 		open_flags = nd->intent.open.flags;
1248 
1249 	lock_kernel();
1250 	nfs_begin_data_update(dir);
1251 	error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags, nd);
1252 	nfs_end_data_update(dir);
1253 	if (error != 0)
1254 		goto out_err;
1255 	nfs_renew_times(dentry);
1256 	nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1257 	unlock_kernel();
1258 	return 0;
1259 out_err:
1260 	unlock_kernel();
1261 	d_drop(dentry);
1262 	return error;
1263 }
1264 
1265 /*
1266  * See comments for nfs_proc_create regarding failed operations.
1267  */
1268 static int
1269 nfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t rdev)
1270 {
1271 	struct iattr attr;
1272 	int status;
1273 
1274 	dfprintk(VFS, "NFS: mknod(%s/%ld), %s\n",
1275 			dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1276 
1277 	if (!new_valid_dev(rdev))
1278 		return -EINVAL;
1279 
1280 	attr.ia_mode = mode;
1281 	attr.ia_valid = ATTR_MODE;
1282 
1283 	lock_kernel();
1284 	nfs_begin_data_update(dir);
1285 	status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev);
1286 	nfs_end_data_update(dir);
1287 	if (status != 0)
1288 		goto out_err;
1289 	nfs_renew_times(dentry);
1290 	nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1291 	unlock_kernel();
1292 	return 0;
1293 out_err:
1294 	unlock_kernel();
1295 	d_drop(dentry);
1296 	return status;
1297 }
1298 
1299 /*
1300  * See comments for nfs_proc_create regarding failed operations.
1301  */
1302 static int nfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1303 {
1304 	struct iattr attr;
1305 	int error;
1306 
1307 	dfprintk(VFS, "NFS: mkdir(%s/%ld), %s\n",
1308 			dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1309 
1310 	attr.ia_valid = ATTR_MODE;
1311 	attr.ia_mode = mode | S_IFDIR;
1312 
1313 	lock_kernel();
1314 	nfs_begin_data_update(dir);
1315 	error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr);
1316 	nfs_end_data_update(dir);
1317 	if (error != 0)
1318 		goto out_err;
1319 	nfs_renew_times(dentry);
1320 	nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1321 	unlock_kernel();
1322 	return 0;
1323 out_err:
1324 	d_drop(dentry);
1325 	unlock_kernel();
1326 	return error;
1327 }
1328 
1329 static int nfs_rmdir(struct inode *dir, struct dentry *dentry)
1330 {
1331 	int error;
1332 
1333 	dfprintk(VFS, "NFS: rmdir(%s/%ld), %s\n",
1334 			dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1335 
1336 	lock_kernel();
1337 	nfs_begin_data_update(dir);
1338 	error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
1339 	/* Ensure the VFS deletes this inode */
1340 	if (error == 0 && dentry->d_inode != NULL)
1341 		clear_nlink(dentry->d_inode);
1342 	nfs_end_data_update(dir);
1343 	unlock_kernel();
1344 
1345 	return error;
1346 }
1347 
1348 static int nfs_sillyrename(struct inode *dir, struct dentry *dentry)
1349 {
1350 	static unsigned int sillycounter;
1351 	const int      i_inosize  = sizeof(dir->i_ino)*2;
1352 	const int      countersize = sizeof(sillycounter)*2;
1353 	const int      slen       = sizeof(".nfs") + i_inosize + countersize - 1;
1354 	char           silly[slen+1];
1355 	struct qstr    qsilly;
1356 	struct dentry *sdentry;
1357 	int            error = -EIO;
1358 
1359 	dfprintk(VFS, "NFS: silly-rename(%s/%s, ct=%d)\n",
1360 		dentry->d_parent->d_name.name, dentry->d_name.name,
1361 		atomic_read(&dentry->d_count));
1362 	nfs_inc_stats(dir, NFSIOS_SILLYRENAME);
1363 
1364 	/*
1365 	 * We don't allow a dentry to be silly-renamed twice.
1366 	 */
1367 	error = -EBUSY;
1368 	if (dentry->d_flags & DCACHE_NFSFS_RENAMED)
1369 		goto out;
1370 
1371 	sprintf(silly, ".nfs%*.*lx",
1372 		i_inosize, i_inosize, dentry->d_inode->i_ino);
1373 
1374 	/* Return delegation in anticipation of the rename */
1375 	nfs_inode_return_delegation(dentry->d_inode);
1376 
1377 	sdentry = NULL;
1378 	do {
1379 		char *suffix = silly + slen - countersize;
1380 
1381 		dput(sdentry);
1382 		sillycounter++;
1383 		sprintf(suffix, "%*.*x", countersize, countersize, sillycounter);
1384 
1385 		dfprintk(VFS, "NFS: trying to rename %s to %s\n",
1386 				dentry->d_name.name, silly);
1387 
1388 		sdentry = lookup_one_len(silly, dentry->d_parent, slen);
1389 		/*
1390 		 * N.B. Better to return EBUSY here ... it could be
1391 		 * dangerous to delete the file while it's in use.
1392 		 */
1393 		if (IS_ERR(sdentry))
1394 			goto out;
1395 	} while(sdentry->d_inode != NULL); /* need negative lookup */
1396 
1397 	qsilly.name = silly;
1398 	qsilly.len  = strlen(silly);
1399 	nfs_begin_data_update(dir);
1400 	if (dentry->d_inode) {
1401 		nfs_begin_data_update(dentry->d_inode);
1402 		error = NFS_PROTO(dir)->rename(dir, &dentry->d_name,
1403 				dir, &qsilly);
1404 		nfs_mark_for_revalidate(dentry->d_inode);
1405 		nfs_end_data_update(dentry->d_inode);
1406 	} else
1407 		error = NFS_PROTO(dir)->rename(dir, &dentry->d_name,
1408 				dir, &qsilly);
1409 	nfs_end_data_update(dir);
1410 	if (!error) {
1411 		nfs_renew_times(dentry);
1412 		nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1413 		d_move(dentry, sdentry);
1414 		error = nfs_async_unlink(dentry);
1415  		/* If we return 0 we don't unlink */
1416 	}
1417 	dput(sdentry);
1418 out:
1419 	return error;
1420 }
1421 
1422 /*
1423  * Remove a file after making sure there are no pending writes,
1424  * and after checking that the file has only one user.
1425  *
1426  * We invalidate the attribute cache and free the inode prior to the operation
1427  * to avoid possible races if the server reuses the inode.
1428  */
1429 static int nfs_safe_remove(struct dentry *dentry)
1430 {
1431 	struct inode *dir = dentry->d_parent->d_inode;
1432 	struct inode *inode = dentry->d_inode;
1433 	int error = -EBUSY;
1434 
1435 	dfprintk(VFS, "NFS: safe_remove(%s/%s)\n",
1436 		dentry->d_parent->d_name.name, dentry->d_name.name);
1437 
1438 	/* If the dentry was sillyrenamed, we simply call d_delete() */
1439 	if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1440 		error = 0;
1441 		goto out;
1442 	}
1443 
1444 	nfs_begin_data_update(dir);
1445 	if (inode != NULL) {
1446 		nfs_inode_return_delegation(inode);
1447 		nfs_begin_data_update(inode);
1448 		error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1449 		/* The VFS may want to delete this inode */
1450 		if (error == 0)
1451 			drop_nlink(inode);
1452 		nfs_mark_for_revalidate(inode);
1453 		nfs_end_data_update(inode);
1454 	} else
1455 		error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1456 	nfs_end_data_update(dir);
1457 out:
1458 	return error;
1459 }
1460 
1461 /*  We do silly rename. In case sillyrename() returns -EBUSY, the inode
1462  *  belongs to an active ".nfs..." file and we return -EBUSY.
1463  *
1464  *  If sillyrename() returns 0, we do nothing, otherwise we unlink.
1465  */
1466 static int nfs_unlink(struct inode *dir, struct dentry *dentry)
1467 {
1468 	int error;
1469 	int need_rehash = 0;
1470 
1471 	dfprintk(VFS, "NFS: unlink(%s/%ld, %s)\n", dir->i_sb->s_id,
1472 		dir->i_ino, dentry->d_name.name);
1473 
1474 	lock_kernel();
1475 	spin_lock(&dcache_lock);
1476 	spin_lock(&dentry->d_lock);
1477 	if (atomic_read(&dentry->d_count) > 1) {
1478 		spin_unlock(&dentry->d_lock);
1479 		spin_unlock(&dcache_lock);
1480 		/* Start asynchronous writeout of the inode */
1481 		write_inode_now(dentry->d_inode, 0);
1482 		error = nfs_sillyrename(dir, dentry);
1483 		unlock_kernel();
1484 		return error;
1485 	}
1486 	if (!d_unhashed(dentry)) {
1487 		__d_drop(dentry);
1488 		need_rehash = 1;
1489 	}
1490 	spin_unlock(&dentry->d_lock);
1491 	spin_unlock(&dcache_lock);
1492 	error = nfs_safe_remove(dentry);
1493 	if (!error) {
1494 		nfs_renew_times(dentry);
1495 		nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1496 	} else if (need_rehash)
1497 		d_rehash(dentry);
1498 	unlock_kernel();
1499 	return error;
1500 }
1501 
1502 /*
1503  * To create a symbolic link, most file systems instantiate a new inode,
1504  * add a page to it containing the path, then write it out to the disk
1505  * using prepare_write/commit_write.
1506  *
1507  * Unfortunately the NFS client can't create the in-core inode first
1508  * because it needs a file handle to create an in-core inode (see
1509  * fs/nfs/inode.c:nfs_fhget).  We only have a file handle *after* the
1510  * symlink request has completed on the server.
1511  *
1512  * So instead we allocate a raw page, copy the symname into it, then do
1513  * the SYMLINK request with the page as the buffer.  If it succeeds, we
1514  * now have a new file handle and can instantiate an in-core NFS inode
1515  * and move the raw page into its mapping.
1516  */
1517 static int nfs_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1518 {
1519 	struct pagevec lru_pvec;
1520 	struct page *page;
1521 	char *kaddr;
1522 	struct iattr attr;
1523 	unsigned int pathlen = strlen(symname);
1524 	int error;
1525 
1526 	dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s)\n", dir->i_sb->s_id,
1527 		dir->i_ino, dentry->d_name.name, symname);
1528 
1529 	if (pathlen > PAGE_SIZE)
1530 		return -ENAMETOOLONG;
1531 
1532 	attr.ia_mode = S_IFLNK | S_IRWXUGO;
1533 	attr.ia_valid = ATTR_MODE;
1534 
1535 	lock_kernel();
1536 
1537 	page = alloc_page(GFP_KERNEL);
1538 	if (!page) {
1539 		unlock_kernel();
1540 		return -ENOMEM;
1541 	}
1542 
1543 	kaddr = kmap_atomic(page, KM_USER0);
1544 	memcpy(kaddr, symname, pathlen);
1545 	if (pathlen < PAGE_SIZE)
1546 		memset(kaddr + pathlen, 0, PAGE_SIZE - pathlen);
1547 	kunmap_atomic(kaddr, KM_USER0);
1548 
1549 	nfs_begin_data_update(dir);
1550 	error = NFS_PROTO(dir)->symlink(dir, dentry, page, pathlen, &attr);
1551 	nfs_end_data_update(dir);
1552 	if (error != 0) {
1553 		dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s) error %d\n",
1554 			dir->i_sb->s_id, dir->i_ino,
1555 			dentry->d_name.name, symname, error);
1556 		d_drop(dentry);
1557 		__free_page(page);
1558 		unlock_kernel();
1559 		return error;
1560 	}
1561 
1562 	/*
1563 	 * No big deal if we can't add this page to the page cache here.
1564 	 * READLINK will get the missing page from the server if needed.
1565 	 */
1566 	pagevec_init(&lru_pvec, 0);
1567 	if (!add_to_page_cache(page, dentry->d_inode->i_mapping, 0,
1568 							GFP_KERNEL)) {
1569 		pagevec_add(&lru_pvec, page);
1570 		pagevec_lru_add(&lru_pvec);
1571 		SetPageUptodate(page);
1572 		unlock_page(page);
1573 	} else
1574 		__free_page(page);
1575 
1576 	unlock_kernel();
1577 	return 0;
1578 }
1579 
1580 static int
1581 nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1582 {
1583 	struct inode *inode = old_dentry->d_inode;
1584 	int error;
1585 
1586 	dfprintk(VFS, "NFS: link(%s/%s -> %s/%s)\n",
1587 		old_dentry->d_parent->d_name.name, old_dentry->d_name.name,
1588 		dentry->d_parent->d_name.name, dentry->d_name.name);
1589 
1590 	lock_kernel();
1591 	nfs_begin_data_update(dir);
1592 	nfs_begin_data_update(inode);
1593 	error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name);
1594 	if (error == 0) {
1595 		atomic_inc(&inode->i_count);
1596 		d_instantiate(dentry, inode);
1597 	}
1598 	nfs_end_data_update(inode);
1599 	nfs_end_data_update(dir);
1600 	unlock_kernel();
1601 	return error;
1602 }
1603 
1604 /*
1605  * RENAME
1606  * FIXME: Some nfsds, like the Linux user space nfsd, may generate a
1607  * different file handle for the same inode after a rename (e.g. when
1608  * moving to a different directory). A fail-safe method to do so would
1609  * be to look up old_dir/old_name, create a link to new_dir/new_name and
1610  * rename the old file using the sillyrename stuff. This way, the original
1611  * file in old_dir will go away when the last process iput()s the inode.
1612  *
1613  * FIXED.
1614  *
1615  * It actually works quite well. One needs to have the possibility for
1616  * at least one ".nfs..." file in each directory the file ever gets
1617  * moved or linked to which happens automagically with the new
1618  * implementation that only depends on the dcache stuff instead of
1619  * using the inode layer
1620  *
1621  * Unfortunately, things are a little more complicated than indicated
1622  * above. For a cross-directory move, we want to make sure we can get
1623  * rid of the old inode after the operation.  This means there must be
1624  * no pending writes (if it's a file), and the use count must be 1.
1625  * If these conditions are met, we can drop the dentries before doing
1626  * the rename.
1627  */
1628 static int nfs_rename(struct inode *old_dir, struct dentry *old_dentry,
1629 		      struct inode *new_dir, struct dentry *new_dentry)
1630 {
1631 	struct inode *old_inode = old_dentry->d_inode;
1632 	struct inode *new_inode = new_dentry->d_inode;
1633 	struct dentry *dentry = NULL, *rehash = NULL;
1634 	int error = -EBUSY;
1635 
1636 	/*
1637 	 * To prevent any new references to the target during the rename,
1638 	 * we unhash the dentry and free the inode in advance.
1639 	 */
1640 	lock_kernel();
1641 	if (!d_unhashed(new_dentry)) {
1642 		d_drop(new_dentry);
1643 		rehash = new_dentry;
1644 	}
1645 
1646 	dfprintk(VFS, "NFS: rename(%s/%s -> %s/%s, ct=%d)\n",
1647 		 old_dentry->d_parent->d_name.name, old_dentry->d_name.name,
1648 		 new_dentry->d_parent->d_name.name, new_dentry->d_name.name,
1649 		 atomic_read(&new_dentry->d_count));
1650 
1651 	/*
1652 	 * First check whether the target is busy ... we can't
1653 	 * safely do _any_ rename if the target is in use.
1654 	 *
1655 	 * For files, make a copy of the dentry and then do a
1656 	 * silly-rename. If the silly-rename succeeds, the
1657 	 * copied dentry is hashed and becomes the new target.
1658 	 */
1659 	if (!new_inode)
1660 		goto go_ahead;
1661 	if (S_ISDIR(new_inode->i_mode)) {
1662 		error = -EISDIR;
1663 		if (!S_ISDIR(old_inode->i_mode))
1664 			goto out;
1665 	} else if (atomic_read(&new_dentry->d_count) > 2) {
1666 		int err;
1667 		/* copy the target dentry's name */
1668 		dentry = d_alloc(new_dentry->d_parent,
1669 				 &new_dentry->d_name);
1670 		if (!dentry)
1671 			goto out;
1672 
1673 		/* silly-rename the existing target ... */
1674 		err = nfs_sillyrename(new_dir, new_dentry);
1675 		if (!err) {
1676 			new_dentry = rehash = dentry;
1677 			new_inode = NULL;
1678 			/* instantiate the replacement target */
1679 			d_instantiate(new_dentry, NULL);
1680 		} else if (atomic_read(&new_dentry->d_count) > 1)
1681 			/* dentry still busy? */
1682 			goto out;
1683 	} else
1684 		drop_nlink(new_inode);
1685 
1686 go_ahead:
1687 	/*
1688 	 * ... prune child dentries and writebacks if needed.
1689 	 */
1690 	if (atomic_read(&old_dentry->d_count) > 1) {
1691 		if (S_ISREG(old_inode->i_mode))
1692 			nfs_wb_all(old_inode);
1693 		shrink_dcache_parent(old_dentry);
1694 	}
1695 	nfs_inode_return_delegation(old_inode);
1696 
1697 	if (new_inode != NULL) {
1698 		nfs_inode_return_delegation(new_inode);
1699 		d_delete(new_dentry);
1700 	}
1701 
1702 	nfs_begin_data_update(old_dir);
1703 	nfs_begin_data_update(new_dir);
1704 	nfs_begin_data_update(old_inode);
1705 	error = NFS_PROTO(old_dir)->rename(old_dir, &old_dentry->d_name,
1706 					   new_dir, &new_dentry->d_name);
1707 	nfs_mark_for_revalidate(old_inode);
1708 	nfs_end_data_update(old_inode);
1709 	nfs_end_data_update(new_dir);
1710 	nfs_end_data_update(old_dir);
1711 out:
1712 	if (rehash)
1713 		d_rehash(rehash);
1714 	if (!error) {
1715 		d_move(old_dentry, new_dentry);
1716 		nfs_renew_times(new_dentry);
1717 		nfs_refresh_verifier(new_dentry, nfs_save_change_attribute(new_dir));
1718 	}
1719 
1720 	/* new dentry created? */
1721 	if (dentry)
1722 		dput(dentry);
1723 	unlock_kernel();
1724 	return error;
1725 }
1726 
1727 static DEFINE_SPINLOCK(nfs_access_lru_lock);
1728 static LIST_HEAD(nfs_access_lru_list);
1729 static atomic_long_t nfs_access_nr_entries;
1730 
1731 static void nfs_access_free_entry(struct nfs_access_entry *entry)
1732 {
1733 	put_rpccred(entry->cred);
1734 	kfree(entry);
1735 	smp_mb__before_atomic_dec();
1736 	atomic_long_dec(&nfs_access_nr_entries);
1737 	smp_mb__after_atomic_dec();
1738 }
1739 
1740 int nfs_access_cache_shrinker(int nr_to_scan, gfp_t gfp_mask)
1741 {
1742 	LIST_HEAD(head);
1743 	struct nfs_inode *nfsi;
1744 	struct nfs_access_entry *cache;
1745 
1746 	spin_lock(&nfs_access_lru_lock);
1747 restart:
1748 	list_for_each_entry(nfsi, &nfs_access_lru_list, access_cache_inode_lru) {
1749 		struct inode *inode;
1750 
1751 		if (nr_to_scan-- == 0)
1752 			break;
1753 		inode = igrab(&nfsi->vfs_inode);
1754 		if (inode == NULL)
1755 			continue;
1756 		spin_lock(&inode->i_lock);
1757 		if (list_empty(&nfsi->access_cache_entry_lru))
1758 			goto remove_lru_entry;
1759 		cache = list_entry(nfsi->access_cache_entry_lru.next,
1760 				struct nfs_access_entry, lru);
1761 		list_move(&cache->lru, &head);
1762 		rb_erase(&cache->rb_node, &nfsi->access_cache);
1763 		if (!list_empty(&nfsi->access_cache_entry_lru))
1764 			list_move_tail(&nfsi->access_cache_inode_lru,
1765 					&nfs_access_lru_list);
1766 		else {
1767 remove_lru_entry:
1768 			list_del_init(&nfsi->access_cache_inode_lru);
1769 			clear_bit(NFS_INO_ACL_LRU_SET, &nfsi->flags);
1770 		}
1771 		spin_unlock(&inode->i_lock);
1772 		iput(inode);
1773 		goto restart;
1774 	}
1775 	spin_unlock(&nfs_access_lru_lock);
1776 	while (!list_empty(&head)) {
1777 		cache = list_entry(head.next, struct nfs_access_entry, lru);
1778 		list_del(&cache->lru);
1779 		nfs_access_free_entry(cache);
1780 	}
1781 	return (atomic_long_read(&nfs_access_nr_entries) / 100) * sysctl_vfs_cache_pressure;
1782 }
1783 
1784 static void __nfs_access_zap_cache(struct inode *inode)
1785 {
1786 	struct nfs_inode *nfsi = NFS_I(inode);
1787 	struct rb_root *root_node = &nfsi->access_cache;
1788 	struct rb_node *n, *dispose = NULL;
1789 	struct nfs_access_entry *entry;
1790 
1791 	/* Unhook entries from the cache */
1792 	while ((n = rb_first(root_node)) != NULL) {
1793 		entry = rb_entry(n, struct nfs_access_entry, rb_node);
1794 		rb_erase(n, root_node);
1795 		list_del(&entry->lru);
1796 		n->rb_left = dispose;
1797 		dispose = n;
1798 	}
1799 	nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS;
1800 	spin_unlock(&inode->i_lock);
1801 
1802 	/* Now kill them all! */
1803 	while (dispose != NULL) {
1804 		n = dispose;
1805 		dispose = n->rb_left;
1806 		nfs_access_free_entry(rb_entry(n, struct nfs_access_entry, rb_node));
1807 	}
1808 }
1809 
1810 void nfs_access_zap_cache(struct inode *inode)
1811 {
1812 	/* Remove from global LRU init */
1813 	if (test_and_clear_bit(NFS_INO_ACL_LRU_SET, &NFS_FLAGS(inode))) {
1814 		spin_lock(&nfs_access_lru_lock);
1815 		list_del_init(&NFS_I(inode)->access_cache_inode_lru);
1816 		spin_unlock(&nfs_access_lru_lock);
1817 	}
1818 
1819 	spin_lock(&inode->i_lock);
1820 	/* This will release the spinlock */
1821 	__nfs_access_zap_cache(inode);
1822 }
1823 
1824 static struct nfs_access_entry *nfs_access_search_rbtree(struct inode *inode, struct rpc_cred *cred)
1825 {
1826 	struct rb_node *n = NFS_I(inode)->access_cache.rb_node;
1827 	struct nfs_access_entry *entry;
1828 
1829 	while (n != NULL) {
1830 		entry = rb_entry(n, struct nfs_access_entry, rb_node);
1831 
1832 		if (cred < entry->cred)
1833 			n = n->rb_left;
1834 		else if (cred > entry->cred)
1835 			n = n->rb_right;
1836 		else
1837 			return entry;
1838 	}
1839 	return NULL;
1840 }
1841 
1842 int nfs_access_get_cached(struct inode *inode, struct rpc_cred *cred, struct nfs_access_entry *res)
1843 {
1844 	struct nfs_inode *nfsi = NFS_I(inode);
1845 	struct nfs_access_entry *cache;
1846 	int err = -ENOENT;
1847 
1848 	spin_lock(&inode->i_lock);
1849 	if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS)
1850 		goto out_zap;
1851 	cache = nfs_access_search_rbtree(inode, cred);
1852 	if (cache == NULL)
1853 		goto out;
1854 	if (time_after(jiffies, cache->jiffies + NFS_ATTRTIMEO(inode)))
1855 		goto out_stale;
1856 	res->jiffies = cache->jiffies;
1857 	res->cred = cache->cred;
1858 	res->mask = cache->mask;
1859 	list_move_tail(&cache->lru, &nfsi->access_cache_entry_lru);
1860 	err = 0;
1861 out:
1862 	spin_unlock(&inode->i_lock);
1863 	return err;
1864 out_stale:
1865 	rb_erase(&cache->rb_node, &nfsi->access_cache);
1866 	list_del(&cache->lru);
1867 	spin_unlock(&inode->i_lock);
1868 	nfs_access_free_entry(cache);
1869 	return -ENOENT;
1870 out_zap:
1871 	/* This will release the spinlock */
1872 	__nfs_access_zap_cache(inode);
1873 	return -ENOENT;
1874 }
1875 
1876 static void nfs_access_add_rbtree(struct inode *inode, struct nfs_access_entry *set)
1877 {
1878 	struct nfs_inode *nfsi = NFS_I(inode);
1879 	struct rb_root *root_node = &nfsi->access_cache;
1880 	struct rb_node **p = &root_node->rb_node;
1881 	struct rb_node *parent = NULL;
1882 	struct nfs_access_entry *entry;
1883 
1884 	spin_lock(&inode->i_lock);
1885 	while (*p != NULL) {
1886 		parent = *p;
1887 		entry = rb_entry(parent, struct nfs_access_entry, rb_node);
1888 
1889 		if (set->cred < entry->cred)
1890 			p = &parent->rb_left;
1891 		else if (set->cred > entry->cred)
1892 			p = &parent->rb_right;
1893 		else
1894 			goto found;
1895 	}
1896 	rb_link_node(&set->rb_node, parent, p);
1897 	rb_insert_color(&set->rb_node, root_node);
1898 	list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
1899 	spin_unlock(&inode->i_lock);
1900 	return;
1901 found:
1902 	rb_replace_node(parent, &set->rb_node, root_node);
1903 	list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
1904 	list_del(&entry->lru);
1905 	spin_unlock(&inode->i_lock);
1906 	nfs_access_free_entry(entry);
1907 }
1908 
1909 void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set)
1910 {
1911 	struct nfs_access_entry *cache = kmalloc(sizeof(*cache), GFP_KERNEL);
1912 	if (cache == NULL)
1913 		return;
1914 	RB_CLEAR_NODE(&cache->rb_node);
1915 	cache->jiffies = set->jiffies;
1916 	cache->cred = get_rpccred(set->cred);
1917 	cache->mask = set->mask;
1918 
1919 	nfs_access_add_rbtree(inode, cache);
1920 
1921 	/* Update accounting */
1922 	smp_mb__before_atomic_inc();
1923 	atomic_long_inc(&nfs_access_nr_entries);
1924 	smp_mb__after_atomic_inc();
1925 
1926 	/* Add inode to global LRU list */
1927 	if (!test_and_set_bit(NFS_INO_ACL_LRU_SET, &NFS_FLAGS(inode))) {
1928 		spin_lock(&nfs_access_lru_lock);
1929 		list_add_tail(&NFS_I(inode)->access_cache_inode_lru, &nfs_access_lru_list);
1930 		spin_unlock(&nfs_access_lru_lock);
1931 	}
1932 }
1933 
1934 static int nfs_do_access(struct inode *inode, struct rpc_cred *cred, int mask)
1935 {
1936 	struct nfs_access_entry cache;
1937 	int status;
1938 
1939 	status = nfs_access_get_cached(inode, cred, &cache);
1940 	if (status == 0)
1941 		goto out;
1942 
1943 	/* Be clever: ask server to check for all possible rights */
1944 	cache.mask = MAY_EXEC | MAY_WRITE | MAY_READ;
1945 	cache.cred = cred;
1946 	cache.jiffies = jiffies;
1947 	status = NFS_PROTO(inode)->access(inode, &cache);
1948 	if (status != 0)
1949 		return status;
1950 	nfs_access_add_cache(inode, &cache);
1951 out:
1952 	if ((cache.mask & mask) == mask)
1953 		return 0;
1954 	return -EACCES;
1955 }
1956 
1957 int nfs_permission(struct inode *inode, int mask, struct nameidata *nd)
1958 {
1959 	struct rpc_cred *cred;
1960 	int res = 0;
1961 
1962 	nfs_inc_stats(inode, NFSIOS_VFSACCESS);
1963 
1964 	if (mask == 0)
1965 		goto out;
1966 	/* Is this sys_access() ? */
1967 	if (nd != NULL && (nd->flags & LOOKUP_ACCESS))
1968 		goto force_lookup;
1969 
1970 	switch (inode->i_mode & S_IFMT) {
1971 		case S_IFLNK:
1972 			goto out;
1973 		case S_IFREG:
1974 			/* NFSv4 has atomic_open... */
1975 			if (nfs_server_capable(inode, NFS_CAP_ATOMIC_OPEN)
1976 					&& nd != NULL
1977 					&& (nd->flags & LOOKUP_OPEN))
1978 				goto out;
1979 			break;
1980 		case S_IFDIR:
1981 			/*
1982 			 * Optimize away all write operations, since the server
1983 			 * will check permissions when we perform the op.
1984 			 */
1985 			if ((mask & MAY_WRITE) && !(mask & MAY_READ))
1986 				goto out;
1987 	}
1988 
1989 force_lookup:
1990 	lock_kernel();
1991 
1992 	if (!NFS_PROTO(inode)->access)
1993 		goto out_notsup;
1994 
1995 	cred = rpcauth_lookupcred(NFS_CLIENT(inode)->cl_auth, 0);
1996 	if (!IS_ERR(cred)) {
1997 		res = nfs_do_access(inode, cred, mask);
1998 		put_rpccred(cred);
1999 	} else
2000 		res = PTR_ERR(cred);
2001 	unlock_kernel();
2002 out:
2003 	dfprintk(VFS, "NFS: permission(%s/%ld), mask=0x%x, res=%d\n",
2004 		inode->i_sb->s_id, inode->i_ino, mask, res);
2005 	return res;
2006 out_notsup:
2007 	res = nfs_revalidate_inode(NFS_SERVER(inode), inode);
2008 	if (res == 0)
2009 		res = generic_permission(inode, mask, NULL);
2010 	unlock_kernel();
2011 	goto out;
2012 }
2013 
2014 /*
2015  * Local variables:
2016  *  version-control: t
2017  *  kept-new-versions: 5
2018  * End:
2019  */
2020