4 * Copyright (C) 1992 Rick Sladkey
6 * nfs directory handling functions
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
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>
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/pagevec.h>
33 #include <linux/namei.h>
34 #include <linux/mount.h>
35 #include <linux/sched.h>
38 #include "delegation.h"
42 /* #define NFS_DEBUG_VERBOSE 1 */
44 static int nfs_opendir(struct inode
*, struct file
*);
45 static int nfs_readdir(struct file
*, void *, filldir_t
);
46 static struct dentry
*nfs_lookup(struct inode
*, struct dentry
*, struct nameidata
*);
47 static int nfs_create(struct inode
*, struct dentry
*, int, struct nameidata
*);
48 static int nfs_mkdir(struct inode
*, struct dentry
*, int);
49 static int nfs_rmdir(struct inode
*, struct dentry
*);
50 static int nfs_unlink(struct inode
*, struct dentry
*);
51 static int nfs_symlink(struct inode
*, struct dentry
*, const char *);
52 static int nfs_link(struct dentry
*, struct inode
*, struct dentry
*);
53 static int nfs_mknod(struct inode
*, struct dentry
*, int, dev_t
);
54 static int nfs_rename(struct inode
*, struct dentry
*,
55 struct inode
*, struct dentry
*);
56 static int nfs_fsync_dir(struct file
*, int);
57 static loff_t
nfs_llseek_dir(struct file
*, loff_t
, int);
59 const struct file_operations nfs_dir_operations
= {
60 .llseek
= nfs_llseek_dir
,
61 .read
= generic_read_dir
,
62 .readdir
= nfs_readdir
,
64 .release
= nfs_release
,
65 .fsync
= nfs_fsync_dir
,
68 const struct inode_operations nfs_dir_inode_operations
= {
73 .symlink
= nfs_symlink
,
78 .permission
= nfs_permission
,
79 .getattr
= nfs_getattr
,
80 .setattr
= nfs_setattr
,
84 const struct inode_operations nfs3_dir_inode_operations
= {
89 .symlink
= nfs_symlink
,
94 .permission
= nfs_permission
,
95 .getattr
= nfs_getattr
,
96 .setattr
= nfs_setattr
,
97 .listxattr
= nfs3_listxattr
,
98 .getxattr
= nfs3_getxattr
,
99 .setxattr
= nfs3_setxattr
,
100 .removexattr
= nfs3_removexattr
,
102 #endif /* CONFIG_NFS_V3 */
106 static struct dentry
*nfs_atomic_lookup(struct inode
*, struct dentry
*, struct nameidata
*);
107 const struct inode_operations nfs4_dir_inode_operations
= {
108 .create
= nfs_create
,
109 .lookup
= nfs_atomic_lookup
,
111 .unlink
= nfs_unlink
,
112 .symlink
= nfs_symlink
,
116 .rename
= nfs_rename
,
117 .permission
= nfs_permission
,
118 .getattr
= nfs_getattr
,
119 .setattr
= nfs_setattr
,
120 .getxattr
= nfs4_getxattr
,
121 .setxattr
= nfs4_setxattr
,
122 .listxattr
= nfs4_listxattr
,
125 #endif /* CONFIG_NFS_V4 */
131 nfs_opendir(struct inode
*inode
, struct file
*filp
)
135 dfprintk(FILE, "NFS: open dir(%s/%s)\n",
136 filp
->f_path
.dentry
->d_parent
->d_name
.name
,
137 filp
->f_path
.dentry
->d_name
.name
);
139 nfs_inc_stats(inode
, NFSIOS_VFSOPEN
);
141 /* Call generic open code in order to cache credentials */
142 res
= nfs_open(inode
, filp
);
146 typedef __be32
* (*decode_dirent_t
)(__be32
*, struct nfs_entry
*, int);
150 unsigned long page_index
;
153 loff_t current_index
;
154 struct nfs_entry
*entry
;
155 decode_dirent_t decode
;
157 unsigned long timestamp
;
158 unsigned long gencount
;
160 } nfs_readdir_descriptor_t
;
162 /* Now we cache directories properly, by stuffing the dirent
163 * data directly in the page cache.
165 * Inode invalidation due to refresh etc. takes care of
166 * _everything_, no sloppy entry flushing logic, no extraneous
167 * copying, network direct to page cache, the way it was meant
170 * NOTE: Dirent information verification is done always by the
171 * page-in of the RPC reply, nowhere else, this simplies
172 * things substantially.
175 int nfs_readdir_filler(nfs_readdir_descriptor_t
*desc
, struct page
*page
)
177 struct file
*file
= desc
->file
;
178 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
179 struct rpc_cred
*cred
= nfs_file_cred(file
);
180 unsigned long timestamp
, gencount
;
183 dfprintk(DIRCACHE
, "NFS: %s: reading cookie %Lu into page %lu\n",
184 __func__
, (long long)desc
->entry
->cookie
,
189 gencount
= nfs_inc_attr_generation_counter();
190 error
= NFS_PROTO(inode
)->readdir(file
->f_path
.dentry
, cred
, desc
->entry
->cookie
, page
,
191 NFS_SERVER(inode
)->dtsize
, desc
->plus
);
193 /* We requested READDIRPLUS, but the server doesn't grok it */
194 if (error
== -ENOTSUPP
&& desc
->plus
) {
195 NFS_SERVER(inode
)->caps
&= ~NFS_CAP_READDIRPLUS
;
196 clear_bit(NFS_INO_ADVISE_RDPLUS
, &NFS_I(inode
)->flags
);
202 desc
->timestamp
= timestamp
;
203 desc
->gencount
= gencount
;
204 desc
->timestamp_valid
= 1;
205 SetPageUptodate(page
);
206 /* Ensure consistent page alignment of the data.
207 * Note: assumes we have exclusive access to this mapping either
208 * through inode->i_mutex or some other mechanism.
210 if (invalidate_inode_pages2_range(inode
->i_mapping
, page
->index
+ 1, -1) < 0) {
211 /* Should never happen */
212 nfs_zap_mapping(inode
, inode
->i_mapping
);
222 int dir_decode(nfs_readdir_descriptor_t
*desc
)
224 __be32
*p
= desc
->ptr
;
225 p
= desc
->decode(p
, desc
->entry
, desc
->plus
);
229 if (desc
->timestamp_valid
) {
230 desc
->entry
->fattr
->time_start
= desc
->timestamp
;
231 desc
->entry
->fattr
->gencount
= desc
->gencount
;
233 desc
->entry
->fattr
->valid
&= ~NFS_ATTR_FATTR
;
238 void dir_page_release(nfs_readdir_descriptor_t
*desc
)
241 page_cache_release(desc
->page
);
247 * Given a pointer to a buffer that has already been filled by a call
248 * to readdir, find the next entry with cookie '*desc->dir_cookie'.
250 * If the end of the buffer has been reached, return -EAGAIN, if not,
251 * return the offset within the buffer of the next entry to be
255 int find_dirent(nfs_readdir_descriptor_t
*desc
)
257 struct nfs_entry
*entry
= desc
->entry
;
261 while((status
= dir_decode(desc
)) == 0) {
262 dfprintk(DIRCACHE
, "NFS: %s: examining cookie %Lu\n",
263 __func__
, (unsigned long long)entry
->cookie
);
264 if (entry
->prev_cookie
== *desc
->dir_cookie
)
266 if (loop_count
++ > 200) {
275 * Given a pointer to a buffer that has already been filled by a call
276 * to readdir, find the entry at offset 'desc->file->f_pos'.
278 * If the end of the buffer has been reached, return -EAGAIN, if not,
279 * return the offset within the buffer of the next entry to be
283 int find_dirent_index(nfs_readdir_descriptor_t
*desc
)
285 struct nfs_entry
*entry
= desc
->entry
;
290 status
= dir_decode(desc
);
294 dfprintk(DIRCACHE
, "NFS: found cookie %Lu at index %Ld\n",
295 (unsigned long long)entry
->cookie
, desc
->current_index
);
297 if (desc
->file
->f_pos
== desc
->current_index
) {
298 *desc
->dir_cookie
= entry
->cookie
;
301 desc
->current_index
++;
302 if (loop_count
++ > 200) {
311 * Find the given page, and call find_dirent() or find_dirent_index in
312 * order to try to return the next entry.
315 int find_dirent_page(nfs_readdir_descriptor_t
*desc
)
317 struct inode
*inode
= desc
->file
->f_path
.dentry
->d_inode
;
321 dfprintk(DIRCACHE
, "NFS: %s: searching page %ld for target %Lu\n",
322 __func__
, desc
->page_index
,
323 (long long) *desc
->dir_cookie
);
325 /* If we find the page in the page_cache, we cannot be sure
326 * how fresh the data is, so we will ignore readdir_plus attributes.
328 desc
->timestamp_valid
= 0;
329 page
= read_cache_page(inode
->i_mapping
, desc
->page_index
,
330 (filler_t
*)nfs_readdir_filler
, desc
);
332 status
= PTR_ERR(page
);
336 /* NOTE: Someone else may have changed the READDIRPLUS flag */
338 desc
->ptr
= kmap(page
); /* matching kunmap in nfs_do_filldir */
339 if (*desc
->dir_cookie
!= 0)
340 status
= find_dirent(desc
);
342 status
= find_dirent_index(desc
);
344 dir_page_release(desc
);
346 dfprintk(DIRCACHE
, "NFS: %s: returns %d\n", __func__
, status
);
351 * Recurse through the page cache pages, and return a
352 * filled nfs_entry structure of the next directory entry if possible.
354 * The target for the search is '*desc->dir_cookie' if non-0,
355 * 'desc->file->f_pos' otherwise
358 int readdir_search_pagecache(nfs_readdir_descriptor_t
*desc
)
363 /* Always search-by-index from the beginning of the cache */
364 if (*desc
->dir_cookie
== 0) {
365 dfprintk(DIRCACHE
, "NFS: readdir_search_pagecache() searching for offset %Ld\n",
366 (long long)desc
->file
->f_pos
);
367 desc
->page_index
= 0;
368 desc
->entry
->cookie
= desc
->entry
->prev_cookie
= 0;
369 desc
->entry
->eof
= 0;
370 desc
->current_index
= 0;
372 dfprintk(DIRCACHE
, "NFS: readdir_search_pagecache() searching for cookie %Lu\n",
373 (unsigned long long)*desc
->dir_cookie
);
376 res
= find_dirent_page(desc
);
379 /* Align to beginning of next page */
381 if (loop_count
++ > 200) {
387 dfprintk(DIRCACHE
, "NFS: %s: returns %d\n", __func__
, res
);
391 static inline unsigned int dt_type(struct inode
*inode
)
393 return (inode
->i_mode
>> 12) & 15;
396 static struct dentry
*nfs_readdir_lookup(nfs_readdir_descriptor_t
*desc
);
399 * Once we've found the start of the dirent within a page: fill 'er up...
402 int nfs_do_filldir(nfs_readdir_descriptor_t
*desc
, void *dirent
,
405 struct file
*file
= desc
->file
;
406 struct nfs_entry
*entry
= desc
->entry
;
407 struct dentry
*dentry
= NULL
;
412 dfprintk(DIRCACHE
, "NFS: nfs_do_filldir() filling starting @ cookie %Lu\n",
413 (unsigned long long)entry
->cookie
);
416 unsigned d_type
= DT_UNKNOWN
;
417 /* Note: entry->prev_cookie contains the cookie for
418 * retrieving the current dirent on the server */
421 /* Get a dentry if we have one */
424 dentry
= nfs_readdir_lookup(desc
);
426 /* Use readdirplus info */
427 if (dentry
!= NULL
&& dentry
->d_inode
!= NULL
) {
428 d_type
= dt_type(dentry
->d_inode
);
429 fileid
= NFS_FILEID(dentry
->d_inode
);
432 res
= filldir(dirent
, entry
->name
, entry
->len
,
433 file
->f_pos
, nfs_compat_user_ino64(fileid
),
438 *desc
->dir_cookie
= entry
->cookie
;
439 if (dir_decode(desc
) != 0) {
443 if (loop_count
++ > 200) {
448 dir_page_release(desc
);
451 dfprintk(DIRCACHE
, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n",
452 (unsigned long long)*desc
->dir_cookie
, res
);
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();
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.
469 int uncached_readdir(nfs_readdir_descriptor_t
*desc
, void *dirent
,
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
;
477 unsigned long timestamp
, gencount
;
479 dfprintk(DIRCACHE
, "NFS: uncached_readdir() searching for cookie %Lu\n",
480 (unsigned long long)*desc
->dir_cookie
);
482 page
= alloc_page(GFP_HIGHUSER
);
488 gencount
= nfs_inc_attr_generation_counter();
489 status
= NFS_PROTO(inode
)->readdir(file
->f_path
.dentry
, cred
,
490 *desc
->dir_cookie
, page
,
491 NFS_SERVER(inode
)->dtsize
,
494 desc
->ptr
= kmap(page
); /* matching kunmap in nfs_do_filldir */
496 desc
->timestamp
= timestamp
;
497 desc
->gencount
= gencount
;
498 desc
->timestamp_valid
= 1;
499 if ((status
= dir_decode(desc
)) == 0)
500 desc
->entry
->prev_cookie
= *desc
->dir_cookie
;
506 status
= nfs_do_filldir(desc
, dirent
, filldir
);
508 /* Reset read descriptor so it searches the page cache from
509 * the start upon the next call to readdir_search_pagecache() */
510 desc
->page_index
= 0;
511 desc
->entry
->cookie
= desc
->entry
->prev_cookie
= 0;
512 desc
->entry
->eof
= 0;
514 dfprintk(DIRCACHE
, "NFS: %s: returns %d\n",
518 dir_page_release(desc
);
522 /* The file offset position represents the dirent entry number. A
523 last cookie cache takes care of the common case of reading the
526 static int nfs_readdir(struct file
*filp
, void *dirent
, filldir_t filldir
)
528 struct dentry
*dentry
= filp
->f_path
.dentry
;
529 struct inode
*inode
= dentry
->d_inode
;
530 nfs_readdir_descriptor_t my_desc
,
532 struct nfs_entry my_entry
;
535 dfprintk(FILE, "NFS: readdir(%s/%s) starting at cookie %llu\n",
536 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
,
537 (long long)filp
->f_pos
);
538 nfs_inc_stats(inode
, NFSIOS_VFSGETDENTS
);
541 * filp->f_pos points to the dirent entry number.
542 * *desc->dir_cookie has the cookie for the next entry. We have
543 * to either find the entry with the appropriate number or
544 * revalidate the cookie.
546 memset(desc
, 0, sizeof(*desc
));
549 desc
->dir_cookie
= &nfs_file_open_context(filp
)->dir_cookie
;
550 desc
->decode
= NFS_PROTO(inode
)->decode_dirent
;
551 desc
->plus
= NFS_USE_READDIRPLUS(inode
);
553 my_entry
.cookie
= my_entry
.prev_cookie
= 0;
555 my_entry
.fh
= nfs_alloc_fhandle();
556 my_entry
.fattr
= nfs_alloc_fattr();
557 if (my_entry
.fh
== NULL
|| my_entry
.fattr
== NULL
)
558 goto out_alloc_failed
;
560 desc
->entry
= &my_entry
;
562 nfs_block_sillyrename(dentry
);
563 res
= nfs_revalidate_mapping(inode
, filp
->f_mapping
);
567 while(!desc
->entry
->eof
) {
568 res
= readdir_search_pagecache(desc
);
570 if (res
== -EBADCOOKIE
) {
571 /* This means either end of directory */
572 if (*desc
->dir_cookie
&& desc
->entry
->cookie
!= *desc
->dir_cookie
) {
573 /* Or that the server has 'lost' a cookie */
574 res
= uncached_readdir(desc
, dirent
, filldir
);
581 if (res
== -ETOOSMALL
&& desc
->plus
) {
582 clear_bit(NFS_INO_ADVISE_RDPLUS
, &NFS_I(inode
)->flags
);
583 nfs_zap_caches(inode
);
585 desc
->entry
->eof
= 0;
591 res
= nfs_do_filldir(desc
, dirent
, filldir
);
598 nfs_unblock_sillyrename(dentry
);
602 nfs_free_fattr(my_entry
.fattr
);
603 nfs_free_fhandle(my_entry
.fh
);
604 dfprintk(FILE, "NFS: readdir(%s/%s) returns %d\n",
605 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
,
610 static loff_t
nfs_llseek_dir(struct file
*filp
, loff_t offset
, int origin
)
612 struct dentry
*dentry
= filp
->f_path
.dentry
;
613 struct inode
*inode
= dentry
->d_inode
;
615 dfprintk(FILE, "NFS: llseek dir(%s/%s, %lld, %d)\n",
616 dentry
->d_parent
->d_name
.name
,
620 mutex_lock(&inode
->i_mutex
);
623 offset
+= filp
->f_pos
;
631 if (offset
!= filp
->f_pos
) {
632 filp
->f_pos
= offset
;
633 nfs_file_open_context(filp
)->dir_cookie
= 0;
636 mutex_unlock(&inode
->i_mutex
);
641 * All directory operations under NFS are synchronous, so fsync()
642 * is a dummy operation.
644 static int nfs_fsync_dir(struct file
*filp
, int datasync
)
646 struct dentry
*dentry
= filp
->f_path
.dentry
;
648 dfprintk(FILE, "NFS: fsync dir(%s/%s) datasync %d\n",
649 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
,
652 nfs_inc_stats(dentry
->d_inode
, NFSIOS_VFSFSYNC
);
657 * nfs_force_lookup_revalidate - Mark the directory as having changed
658 * @dir - pointer to directory inode
660 * This forces the revalidation code in nfs_lookup_revalidate() to do a
661 * full lookup on all child dentries of 'dir' whenever a change occurs
662 * on the server that might have invalidated our dcache.
664 * The caller should be holding dir->i_lock
666 void nfs_force_lookup_revalidate(struct inode
*dir
)
668 NFS_I(dir
)->cache_change_attribute
++;
672 * A check for whether or not the parent directory has changed.
673 * In the case it has, we assume that the dentries are untrustworthy
674 * and may need to be looked up again.
676 static int nfs_check_verifier(struct inode
*dir
, struct dentry
*dentry
)
680 if (NFS_SERVER(dir
)->flags
& NFS_MOUNT_LOOKUP_CACHE_NONE
)
682 if (!nfs_verify_change_attribute(dir
, dentry
->d_time
))
684 /* Revalidate nfsi->cache_change_attribute before we declare a match */
685 if (nfs_revalidate_inode(NFS_SERVER(dir
), dir
) < 0)
687 if (!nfs_verify_change_attribute(dir
, dentry
->d_time
))
693 * Return the intent data that applies to this particular path component
695 * Note that the current set of intents only apply to the very last
696 * component of the path.
697 * We check for this using LOOKUP_CONTINUE and LOOKUP_PARENT.
699 static inline unsigned int nfs_lookup_check_intent(struct nameidata
*nd
, unsigned int mask
)
701 if (nd
->flags
& (LOOKUP_CONTINUE
|LOOKUP_PARENT
))
703 return nd
->flags
& mask
;
707 * Use intent information to check whether or not we're going to do
708 * an O_EXCL create using this path component.
710 static int nfs_is_exclusive_create(struct inode
*dir
, struct nameidata
*nd
)
712 if (NFS_PROTO(dir
)->version
== 2)
714 return nd
&& nfs_lookup_check_intent(nd
, LOOKUP_EXCL
);
718 * Inode and filehandle revalidation for lookups.
720 * We force revalidation in the cases where the VFS sets LOOKUP_REVAL,
721 * or if the intent information indicates that we're about to open this
722 * particular file and the "nocto" mount flag is not set.
726 int nfs_lookup_verify_inode(struct inode
*inode
, struct nameidata
*nd
)
728 struct nfs_server
*server
= NFS_SERVER(inode
);
730 if (test_bit(NFS_INO_MOUNTPOINT
, &NFS_I(inode
)->flags
))
733 /* VFS wants an on-the-wire revalidation */
734 if (nd
->flags
& LOOKUP_REVAL
)
736 /* This is an open(2) */
737 if (nfs_lookup_check_intent(nd
, LOOKUP_OPEN
) != 0 &&
738 !(server
->flags
& NFS_MOUNT_NOCTO
) &&
739 (S_ISREG(inode
->i_mode
) ||
740 S_ISDIR(inode
->i_mode
)))
744 return nfs_revalidate_inode(server
, inode
);
746 return __nfs_revalidate_inode(server
, inode
);
750 * We judge how long we want to trust negative
751 * dentries by looking at the parent inode mtime.
753 * If parent mtime has changed, we revalidate, else we wait for a
754 * period corresponding to the parent's attribute cache timeout value.
757 int nfs_neg_need_reval(struct inode
*dir
, struct dentry
*dentry
,
758 struct nameidata
*nd
)
760 /* Don't revalidate a negative dentry if we're creating a new file */
761 if (nd
!= NULL
&& nfs_lookup_check_intent(nd
, LOOKUP_CREATE
) != 0)
763 if (NFS_SERVER(dir
)->flags
& NFS_MOUNT_LOOKUP_CACHE_NONEG
)
765 return !nfs_check_verifier(dir
, dentry
);
769 * This is called every time the dcache has a lookup hit,
770 * and we should check whether we can really trust that
773 * NOTE! The hit can be a negative hit too, don't assume
776 * If the parent directory is seen to have changed, we throw out the
777 * cached dentry and do a new lookup.
779 static int nfs_lookup_revalidate(struct dentry
* dentry
, struct nameidata
*nd
)
783 struct dentry
*parent
;
784 struct nfs_fh
*fhandle
= NULL
;
785 struct nfs_fattr
*fattr
= NULL
;
788 parent
= dget_parent(dentry
);
789 dir
= parent
->d_inode
;
790 nfs_inc_stats(dir
, NFSIOS_DENTRYREVALIDATE
);
791 inode
= dentry
->d_inode
;
794 if (nfs_neg_need_reval(dir
, dentry
, nd
))
799 if (is_bad_inode(inode
)) {
800 dfprintk(LOOKUPCACHE
, "%s: %s/%s has dud inode\n",
801 __func__
, dentry
->d_parent
->d_name
.name
,
802 dentry
->d_name
.name
);
806 if (nfs_have_delegation(inode
, FMODE_READ
))
807 goto out_set_verifier
;
809 /* Force a full look up iff the parent directory has changed */
810 if (!nfs_is_exclusive_create(dir
, nd
) && nfs_check_verifier(dir
, dentry
)) {
811 if (nfs_lookup_verify_inode(inode
, nd
))
816 if (NFS_STALE(inode
))
820 fhandle
= nfs_alloc_fhandle();
821 fattr
= nfs_alloc_fattr();
822 if (fhandle
== NULL
|| fattr
== NULL
)
825 error
= NFS_PROTO(dir
)->lookup(dir
, &dentry
->d_name
, fhandle
, fattr
);
828 if (nfs_compare_fh(NFS_FH(inode
), fhandle
))
830 if ((error
= nfs_refresh_inode(inode
, fattr
)) != 0)
833 nfs_free_fattr(fattr
);
834 nfs_free_fhandle(fhandle
);
836 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
839 dfprintk(LOOKUPCACHE
, "NFS: %s(%s/%s) is valid\n",
840 __func__
, dentry
->d_parent
->d_name
.name
,
841 dentry
->d_name
.name
);
846 nfs_mark_for_revalidate(dir
);
847 if (inode
&& S_ISDIR(inode
->i_mode
)) {
848 /* Purge readdir caches. */
849 nfs_zap_caches(inode
);
850 /* If we have submounts, don't unhash ! */
851 if (have_submounts(dentry
))
853 if (dentry
->d_flags
& DCACHE_DISCONNECTED
)
855 shrink_dcache_parent(dentry
);
858 nfs_free_fattr(fattr
);
859 nfs_free_fhandle(fhandle
);
861 dfprintk(LOOKUPCACHE
, "NFS: %s(%s/%s) is invalid\n",
862 __func__
, dentry
->d_parent
->d_name
.name
,
863 dentry
->d_name
.name
);
866 nfs_free_fattr(fattr
);
867 nfs_free_fhandle(fhandle
);
869 dfprintk(LOOKUPCACHE
, "NFS: %s(%s/%s) lookup returned error %d\n",
870 __func__
, dentry
->d_parent
->d_name
.name
,
871 dentry
->d_name
.name
, error
);
876 * This is called from dput() when d_count is going to 0.
878 static int nfs_dentry_delete(struct dentry
*dentry
)
880 dfprintk(VFS
, "NFS: dentry_delete(%s/%s, %x)\n",
881 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
,
884 /* Unhash any dentry with a stale inode */
885 if (dentry
->d_inode
!= NULL
&& NFS_STALE(dentry
->d_inode
))
888 if (dentry
->d_flags
& DCACHE_NFSFS_RENAMED
) {
889 /* Unhash it, so that ->d_iput() would be called */
892 if (!(dentry
->d_sb
->s_flags
& MS_ACTIVE
)) {
893 /* Unhash it, so that ancestors of killed async unlink
894 * files will be cleaned up during umount */
901 static void nfs_drop_nlink(struct inode
*inode
)
903 spin_lock(&inode
->i_lock
);
904 if (inode
->i_nlink
> 0)
906 spin_unlock(&inode
->i_lock
);
910 * Called when the dentry loses inode.
911 * We use it to clean up silly-renamed files.
913 static void nfs_dentry_iput(struct dentry
*dentry
, struct inode
*inode
)
915 if (S_ISDIR(inode
->i_mode
))
916 /* drop any readdir cache as it could easily be old */
917 NFS_I(inode
)->cache_validity
|= NFS_INO_INVALID_DATA
;
919 if (dentry
->d_flags
& DCACHE_NFSFS_RENAMED
) {
921 nfs_complete_unlink(dentry
, inode
);
926 const struct dentry_operations nfs_dentry_operations
= {
927 .d_revalidate
= nfs_lookup_revalidate
,
928 .d_delete
= nfs_dentry_delete
,
929 .d_iput
= nfs_dentry_iput
,
932 static struct dentry
*nfs_lookup(struct inode
*dir
, struct dentry
* dentry
, struct nameidata
*nd
)
935 struct dentry
*parent
;
936 struct inode
*inode
= NULL
;
937 struct nfs_fh
*fhandle
= NULL
;
938 struct nfs_fattr
*fattr
= NULL
;
941 dfprintk(VFS
, "NFS: lookup(%s/%s)\n",
942 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
);
943 nfs_inc_stats(dir
, NFSIOS_VFSLOOKUP
);
945 res
= ERR_PTR(-ENAMETOOLONG
);
946 if (dentry
->d_name
.len
> NFS_SERVER(dir
)->namelen
)
949 dentry
->d_op
= NFS_PROTO(dir
)->dentry_ops
;
952 * If we're doing an exclusive create, optimize away the lookup
953 * but don't hash the dentry.
955 if (nfs_is_exclusive_create(dir
, nd
)) {
956 d_instantiate(dentry
, NULL
);
961 res
= ERR_PTR(-ENOMEM
);
962 fhandle
= nfs_alloc_fhandle();
963 fattr
= nfs_alloc_fattr();
964 if (fhandle
== NULL
|| fattr
== NULL
)
967 parent
= dentry
->d_parent
;
968 /* Protect against concurrent sillydeletes */
969 nfs_block_sillyrename(parent
);
970 error
= NFS_PROTO(dir
)->lookup(dir
, &dentry
->d_name
, fhandle
, fattr
);
971 if (error
== -ENOENT
)
974 res
= ERR_PTR(error
);
975 goto out_unblock_sillyrename
;
977 inode
= nfs_fhget(dentry
->d_sb
, fhandle
, fattr
);
978 res
= (struct dentry
*)inode
;
980 goto out_unblock_sillyrename
;
983 res
= d_materialise_unique(dentry
, inode
);
986 goto out_unblock_sillyrename
;
989 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
990 out_unblock_sillyrename
:
991 nfs_unblock_sillyrename(parent
);
993 nfs_free_fattr(fattr
);
994 nfs_free_fhandle(fhandle
);
999 static int nfs_open_revalidate(struct dentry
*, struct nameidata
*);
1001 const struct dentry_operations nfs4_dentry_operations
= {
1002 .d_revalidate
= nfs_open_revalidate
,
1003 .d_delete
= nfs_dentry_delete
,
1004 .d_iput
= nfs_dentry_iput
,
1008 * Use intent information to determine whether we need to substitute
1009 * the NFSv4-style stateful OPEN for the LOOKUP call
1011 static int is_atomic_open(struct nameidata
*nd
)
1013 if (nd
== NULL
|| nfs_lookup_check_intent(nd
, LOOKUP_OPEN
) == 0)
1015 /* NFS does not (yet) have a stateful open for directories */
1016 if (nd
->flags
& LOOKUP_DIRECTORY
)
1018 /* Are we trying to write to a read only partition? */
1019 if (__mnt_is_readonly(nd
->path
.mnt
) &&
1020 (nd
->intent
.open
.flags
& (O_CREAT
|O_TRUNC
|FMODE_WRITE
)))
1025 static struct dentry
*nfs_atomic_lookup(struct inode
*dir
, struct dentry
*dentry
, struct nameidata
*nd
)
1027 struct dentry
*res
= NULL
;
1030 dfprintk(VFS
, "NFS: atomic_lookup(%s/%ld), %s\n",
1031 dir
->i_sb
->s_id
, dir
->i_ino
, dentry
->d_name
.name
);
1033 /* Check that we are indeed trying to open this file */
1034 if (!is_atomic_open(nd
))
1037 if (dentry
->d_name
.len
> NFS_SERVER(dir
)->namelen
) {
1038 res
= ERR_PTR(-ENAMETOOLONG
);
1041 dentry
->d_op
= NFS_PROTO(dir
)->dentry_ops
;
1043 /* Let vfs_create() deal with O_EXCL. Instantiate, but don't hash
1045 if (nd
->flags
& LOOKUP_EXCL
) {
1046 d_instantiate(dentry
, NULL
);
1050 /* Open the file on the server */
1051 res
= nfs4_atomic_open(dir
, dentry
, nd
);
1053 error
= PTR_ERR(res
);
1055 /* Make a negative dentry */
1059 /* This turned out not to be a regular file */
1064 if (!(nd
->intent
.open
.flags
& O_NOFOLLOW
))
1070 } else if (res
!= NULL
)
1075 return nfs_lookup(dir
, dentry
, nd
);
1078 static int nfs_open_revalidate(struct dentry
*dentry
, struct nameidata
*nd
)
1080 struct dentry
*parent
= NULL
;
1081 struct inode
*inode
= dentry
->d_inode
;
1083 int openflags
, ret
= 0;
1085 if (!is_atomic_open(nd
) || d_mountpoint(dentry
))
1087 parent
= dget_parent(dentry
);
1088 dir
= parent
->d_inode
;
1089 /* We can't create new files in nfs_open_revalidate(), so we
1090 * optimize away revalidation of negative dentries.
1092 if (inode
== NULL
) {
1093 if (!nfs_neg_need_reval(dir
, dentry
, nd
))
1098 /* NFS only supports OPEN on regular files */
1099 if (!S_ISREG(inode
->i_mode
))
1101 openflags
= nd
->intent
.open
.flags
;
1102 /* We cannot do exclusive creation on a positive dentry */
1103 if ((openflags
& (O_CREAT
|O_EXCL
)) == (O_CREAT
|O_EXCL
))
1105 /* We can't create new files, or truncate existing ones here */
1106 openflags
&= ~(O_CREAT
|O_TRUNC
);
1109 * Note: we're not holding inode->i_mutex and so may be racing with
1110 * operations that change the directory. We therefore save the
1111 * change attribute *before* we do the RPC call.
1113 ret
= nfs4_open_revalidate(dir
, dentry
, openflags
, nd
);
1122 return nfs_lookup_revalidate(dentry
, nd
);
1124 #endif /* CONFIG_NFSV4 */
1126 static struct dentry
*nfs_readdir_lookup(nfs_readdir_descriptor_t
*desc
)
1128 struct dentry
*parent
= desc
->file
->f_path
.dentry
;
1129 struct inode
*dir
= parent
->d_inode
;
1130 struct nfs_entry
*entry
= desc
->entry
;
1131 struct dentry
*dentry
, *alias
;
1132 struct qstr name
= {
1133 .name
= entry
->name
,
1136 struct inode
*inode
;
1137 unsigned long verf
= nfs_save_change_attribute(dir
);
1141 if (name
.name
[0] == '.' && name
.name
[1] == '.')
1142 return dget_parent(parent
);
1145 if (name
.name
[0] == '.')
1146 return dget(parent
);
1149 spin_lock(&dir
->i_lock
);
1150 if (NFS_I(dir
)->cache_validity
& NFS_INO_INVALID_DATA
) {
1151 spin_unlock(&dir
->i_lock
);
1154 spin_unlock(&dir
->i_lock
);
1156 name
.hash
= full_name_hash(name
.name
, name
.len
);
1157 dentry
= d_lookup(parent
, &name
);
1158 if (dentry
!= NULL
) {
1159 /* Is this a positive dentry that matches the readdir info? */
1160 if (dentry
->d_inode
!= NULL
&&
1161 (NFS_FILEID(dentry
->d_inode
) == entry
->ino
||
1162 d_mountpoint(dentry
))) {
1163 if (!desc
->plus
|| entry
->fh
->size
== 0)
1165 if (nfs_compare_fh(NFS_FH(dentry
->d_inode
),
1169 /* No, so d_drop to allow one to be created */
1173 if (!desc
->plus
|| !(entry
->fattr
->valid
& NFS_ATTR_FATTR
))
1175 if (name
.len
> NFS_SERVER(dir
)->namelen
)
1177 /* Note: caller is already holding the dir->i_mutex! */
1178 dentry
= d_alloc(parent
, &name
);
1181 dentry
->d_op
= NFS_PROTO(dir
)->dentry_ops
;
1182 inode
= nfs_fhget(dentry
->d_sb
, entry
->fh
, entry
->fattr
);
1183 if (IS_ERR(inode
)) {
1188 alias
= d_materialise_unique(dentry
, inode
);
1189 if (alias
!= NULL
) {
1197 nfs_set_verifier(dentry
, verf
);
1202 * Code common to create, mkdir, and mknod.
1204 int nfs_instantiate(struct dentry
*dentry
, struct nfs_fh
*fhandle
,
1205 struct nfs_fattr
*fattr
)
1207 struct dentry
*parent
= dget_parent(dentry
);
1208 struct inode
*dir
= parent
->d_inode
;
1209 struct inode
*inode
;
1210 int error
= -EACCES
;
1214 /* We may have been initialized further down */
1215 if (dentry
->d_inode
)
1217 if (fhandle
->size
== 0) {
1218 error
= NFS_PROTO(dir
)->lookup(dir
, &dentry
->d_name
, fhandle
, fattr
);
1222 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1223 if (!(fattr
->valid
& NFS_ATTR_FATTR
)) {
1224 struct nfs_server
*server
= NFS_SB(dentry
->d_sb
);
1225 error
= server
->nfs_client
->rpc_ops
->getattr(server
, fhandle
, fattr
);
1229 inode
= nfs_fhget(dentry
->d_sb
, fhandle
, fattr
);
1230 error
= PTR_ERR(inode
);
1233 d_add(dentry
, inode
);
1238 nfs_mark_for_revalidate(dir
);
1244 * Following a failed create operation, we drop the dentry rather
1245 * than retain a negative dentry. This avoids a problem in the event
1246 * that the operation succeeded on the server, but an error in the
1247 * reply path made it appear to have failed.
1249 static int nfs_create(struct inode
*dir
, struct dentry
*dentry
, int mode
,
1250 struct nameidata
*nd
)
1256 dfprintk(VFS
, "NFS: create(%s/%ld), %s\n",
1257 dir
->i_sb
->s_id
, dir
->i_ino
, dentry
->d_name
.name
);
1259 attr
.ia_mode
= mode
;
1260 attr
.ia_valid
= ATTR_MODE
;
1262 if ((nd
->flags
& LOOKUP_CREATE
) != 0)
1263 open_flags
= nd
->intent
.open
.flags
;
1265 error
= NFS_PROTO(dir
)->create(dir
, dentry
, &attr
, open_flags
, nd
);
1275 * See comments for nfs_proc_create regarding failed operations.
1278 nfs_mknod(struct inode
*dir
, struct dentry
*dentry
, int mode
, dev_t rdev
)
1283 dfprintk(VFS
, "NFS: mknod(%s/%ld), %s\n",
1284 dir
->i_sb
->s_id
, dir
->i_ino
, dentry
->d_name
.name
);
1286 if (!new_valid_dev(rdev
))
1289 attr
.ia_mode
= mode
;
1290 attr
.ia_valid
= ATTR_MODE
;
1292 status
= NFS_PROTO(dir
)->mknod(dir
, dentry
, &attr
, rdev
);
1302 * See comments for nfs_proc_create regarding failed operations.
1304 static int nfs_mkdir(struct inode
*dir
, struct dentry
*dentry
, int mode
)
1309 dfprintk(VFS
, "NFS: mkdir(%s/%ld), %s\n",
1310 dir
->i_sb
->s_id
, dir
->i_ino
, dentry
->d_name
.name
);
1312 attr
.ia_valid
= ATTR_MODE
;
1313 attr
.ia_mode
= mode
| S_IFDIR
;
1315 error
= NFS_PROTO(dir
)->mkdir(dir
, dentry
, &attr
);
1324 static void nfs_dentry_handle_enoent(struct dentry
*dentry
)
1326 if (dentry
->d_inode
!= NULL
&& !d_unhashed(dentry
))
1330 static int nfs_rmdir(struct inode
*dir
, struct dentry
*dentry
)
1334 dfprintk(VFS
, "NFS: rmdir(%s/%ld), %s\n",
1335 dir
->i_sb
->s_id
, dir
->i_ino
, dentry
->d_name
.name
);
1337 error
= NFS_PROTO(dir
)->rmdir(dir
, &dentry
->d_name
);
1338 /* Ensure the VFS deletes this inode */
1339 if (error
== 0 && dentry
->d_inode
!= NULL
)
1340 clear_nlink(dentry
->d_inode
);
1341 else if (error
== -ENOENT
)
1342 nfs_dentry_handle_enoent(dentry
);
1347 static int nfs_sillyrename(struct inode
*dir
, struct dentry
*dentry
)
1349 static unsigned int sillycounter
;
1350 const int fileidsize
= sizeof(NFS_FILEID(dentry
->d_inode
))*2;
1351 const int countersize
= sizeof(sillycounter
)*2;
1352 const int slen
= sizeof(".nfs")+fileidsize
+countersize
-1;
1355 struct dentry
*sdentry
;
1358 dfprintk(VFS
, "NFS: silly-rename(%s/%s, ct=%d)\n",
1359 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
,
1360 atomic_read(&dentry
->d_count
));
1361 nfs_inc_stats(dir
, NFSIOS_SILLYRENAME
);
1364 * We don't allow a dentry to be silly-renamed twice.
1367 if (dentry
->d_flags
& DCACHE_NFSFS_RENAMED
)
1370 sprintf(silly
, ".nfs%*.*Lx",
1371 fileidsize
, fileidsize
,
1372 (unsigned long long)NFS_FILEID(dentry
->d_inode
));
1374 /* Return delegation in anticipation of the rename */
1375 nfs_inode_return_delegation(dentry
->d_inode
);
1379 char *suffix
= silly
+ slen
- countersize
;
1383 sprintf(suffix
, "%*.*x", countersize
, countersize
, sillycounter
);
1385 dfprintk(VFS
, "NFS: trying to rename %s to %s\n",
1386 dentry
->d_name
.name
, silly
);
1388 sdentry
= lookup_one_len(silly
, dentry
->d_parent
, slen
);
1390 * N.B. Better to return EBUSY here ... it could be
1391 * dangerous to delete the file while it's in use.
1393 if (IS_ERR(sdentry
))
1395 } while(sdentry
->d_inode
!= NULL
); /* need negative lookup */
1397 qsilly
.name
= silly
;
1398 qsilly
.len
= strlen(silly
);
1399 if (dentry
->d_inode
) {
1400 error
= NFS_PROTO(dir
)->rename(dir
, &dentry
->d_name
,
1402 nfs_mark_for_revalidate(dentry
->d_inode
);
1404 error
= NFS_PROTO(dir
)->rename(dir
, &dentry
->d_name
,
1407 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1408 d_move(dentry
, sdentry
);
1409 error
= nfs_async_unlink(dir
, dentry
);
1410 /* If we return 0 we don't unlink */
1418 * Remove a file after making sure there are no pending writes,
1419 * and after checking that the file has only one user.
1421 * We invalidate the attribute cache and free the inode prior to the operation
1422 * to avoid possible races if the server reuses the inode.
1424 static int nfs_safe_remove(struct dentry
*dentry
)
1426 struct inode
*dir
= dentry
->d_parent
->d_inode
;
1427 struct inode
*inode
= dentry
->d_inode
;
1430 dfprintk(VFS
, "NFS: safe_remove(%s/%s)\n",
1431 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
);
1433 /* If the dentry was sillyrenamed, we simply call d_delete() */
1434 if (dentry
->d_flags
& DCACHE_NFSFS_RENAMED
) {
1439 if (inode
!= NULL
) {
1440 nfs_inode_return_delegation(inode
);
1441 error
= NFS_PROTO(dir
)->remove(dir
, &dentry
->d_name
);
1442 /* The VFS may want to delete this inode */
1444 nfs_drop_nlink(inode
);
1445 nfs_mark_for_revalidate(inode
);
1447 error
= NFS_PROTO(dir
)->remove(dir
, &dentry
->d_name
);
1448 if (error
== -ENOENT
)
1449 nfs_dentry_handle_enoent(dentry
);
1454 /* We do silly rename. In case sillyrename() returns -EBUSY, the inode
1455 * belongs to an active ".nfs..." file and we return -EBUSY.
1457 * If sillyrename() returns 0, we do nothing, otherwise we unlink.
1459 static int nfs_unlink(struct inode
*dir
, struct dentry
*dentry
)
1462 int need_rehash
= 0;
1464 dfprintk(VFS
, "NFS: unlink(%s/%ld, %s)\n", dir
->i_sb
->s_id
,
1465 dir
->i_ino
, dentry
->d_name
.name
);
1467 spin_lock(&dcache_lock
);
1468 spin_lock(&dentry
->d_lock
);
1469 if (atomic_read(&dentry
->d_count
) > 1) {
1470 spin_unlock(&dentry
->d_lock
);
1471 spin_unlock(&dcache_lock
);
1472 /* Start asynchronous writeout of the inode */
1473 write_inode_now(dentry
->d_inode
, 0);
1474 error
= nfs_sillyrename(dir
, dentry
);
1477 if (!d_unhashed(dentry
)) {
1481 spin_unlock(&dentry
->d_lock
);
1482 spin_unlock(&dcache_lock
);
1483 error
= nfs_safe_remove(dentry
);
1484 if (!error
|| error
== -ENOENT
) {
1485 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1486 } else if (need_rehash
)
1492 * To create a symbolic link, most file systems instantiate a new inode,
1493 * add a page to it containing the path, then write it out to the disk
1494 * using prepare_write/commit_write.
1496 * Unfortunately the NFS client can't create the in-core inode first
1497 * because it needs a file handle to create an in-core inode (see
1498 * fs/nfs/inode.c:nfs_fhget). We only have a file handle *after* the
1499 * symlink request has completed on the server.
1501 * So instead we allocate a raw page, copy the symname into it, then do
1502 * the SYMLINK request with the page as the buffer. If it succeeds, we
1503 * now have a new file handle and can instantiate an in-core NFS inode
1504 * and move the raw page into its mapping.
1506 static int nfs_symlink(struct inode
*dir
, struct dentry
*dentry
, const char *symname
)
1508 struct pagevec lru_pvec
;
1512 unsigned int pathlen
= strlen(symname
);
1515 dfprintk(VFS
, "NFS: symlink(%s/%ld, %s, %s)\n", dir
->i_sb
->s_id
,
1516 dir
->i_ino
, dentry
->d_name
.name
, symname
);
1518 if (pathlen
> PAGE_SIZE
)
1519 return -ENAMETOOLONG
;
1521 attr
.ia_mode
= S_IFLNK
| S_IRWXUGO
;
1522 attr
.ia_valid
= ATTR_MODE
;
1524 page
= alloc_page(GFP_HIGHUSER
);
1528 kaddr
= kmap_atomic(page
, KM_USER0
);
1529 memcpy(kaddr
, symname
, pathlen
);
1530 if (pathlen
< PAGE_SIZE
)
1531 memset(kaddr
+ pathlen
, 0, PAGE_SIZE
- pathlen
);
1532 kunmap_atomic(kaddr
, KM_USER0
);
1534 error
= NFS_PROTO(dir
)->symlink(dir
, dentry
, page
, pathlen
, &attr
);
1536 dfprintk(VFS
, "NFS: symlink(%s/%ld, %s, %s) error %d\n",
1537 dir
->i_sb
->s_id
, dir
->i_ino
,
1538 dentry
->d_name
.name
, symname
, error
);
1545 * No big deal if we can't add this page to the page cache here.
1546 * READLINK will get the missing page from the server if needed.
1548 pagevec_init(&lru_pvec
, 0);
1549 if (!add_to_page_cache(page
, dentry
->d_inode
->i_mapping
, 0,
1551 pagevec_add(&lru_pvec
, page
);
1552 pagevec_lru_add_file(&lru_pvec
);
1553 SetPageUptodate(page
);
1562 nfs_link(struct dentry
*old_dentry
, struct inode
*dir
, struct dentry
*dentry
)
1564 struct inode
*inode
= old_dentry
->d_inode
;
1567 dfprintk(VFS
, "NFS: link(%s/%s -> %s/%s)\n",
1568 old_dentry
->d_parent
->d_name
.name
, old_dentry
->d_name
.name
,
1569 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
);
1571 nfs_inode_return_delegation(inode
);
1574 error
= NFS_PROTO(dir
)->link(inode
, dir
, &dentry
->d_name
);
1576 atomic_inc(&inode
->i_count
);
1577 d_add(dentry
, inode
);
1584 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a
1585 * different file handle for the same inode after a rename (e.g. when
1586 * moving to a different directory). A fail-safe method to do so would
1587 * be to look up old_dir/old_name, create a link to new_dir/new_name and
1588 * rename the old file using the sillyrename stuff. This way, the original
1589 * file in old_dir will go away when the last process iput()s the inode.
1593 * It actually works quite well. One needs to have the possibility for
1594 * at least one ".nfs..." file in each directory the file ever gets
1595 * moved or linked to which happens automagically with the new
1596 * implementation that only depends on the dcache stuff instead of
1597 * using the inode layer
1599 * Unfortunately, things are a little more complicated than indicated
1600 * above. For a cross-directory move, we want to make sure we can get
1601 * rid of the old inode after the operation. This means there must be
1602 * no pending writes (if it's a file), and the use count must be 1.
1603 * If these conditions are met, we can drop the dentries before doing
1606 static int nfs_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
1607 struct inode
*new_dir
, struct dentry
*new_dentry
)
1609 struct inode
*old_inode
= old_dentry
->d_inode
;
1610 struct inode
*new_inode
= new_dentry
->d_inode
;
1611 struct dentry
*dentry
= NULL
, *rehash
= NULL
;
1614 dfprintk(VFS
, "NFS: rename(%s/%s -> %s/%s, ct=%d)\n",
1615 old_dentry
->d_parent
->d_name
.name
, old_dentry
->d_name
.name
,
1616 new_dentry
->d_parent
->d_name
.name
, new_dentry
->d_name
.name
,
1617 atomic_read(&new_dentry
->d_count
));
1620 * For non-directories, check whether the target is busy and if so,
1621 * make a copy of the dentry and then do a silly-rename. If the
1622 * silly-rename succeeds, the copied dentry is hashed and becomes
1625 if (new_inode
&& !S_ISDIR(new_inode
->i_mode
)) {
1627 * To prevent any new references to the target during the
1628 * rename, we unhash the dentry in advance.
1630 if (!d_unhashed(new_dentry
)) {
1632 rehash
= new_dentry
;
1635 if (atomic_read(&new_dentry
->d_count
) > 2) {
1638 /* copy the target dentry's name */
1639 dentry
= d_alloc(new_dentry
->d_parent
,
1640 &new_dentry
->d_name
);
1644 /* silly-rename the existing target ... */
1645 err
= nfs_sillyrename(new_dir
, new_dentry
);
1649 new_dentry
= dentry
;
1655 nfs_inode_return_delegation(old_inode
);
1656 if (new_inode
!= NULL
)
1657 nfs_inode_return_delegation(new_inode
);
1659 error
= NFS_PROTO(old_dir
)->rename(old_dir
, &old_dentry
->d_name
,
1660 new_dir
, &new_dentry
->d_name
);
1661 nfs_mark_for_revalidate(old_inode
);
1666 if (new_inode
!= NULL
)
1667 nfs_drop_nlink(new_inode
);
1668 d_move(old_dentry
, new_dentry
);
1669 nfs_set_verifier(new_dentry
,
1670 nfs_save_change_attribute(new_dir
));
1671 } else if (error
== -ENOENT
)
1672 nfs_dentry_handle_enoent(old_dentry
);
1674 /* new dentry created? */
1680 static DEFINE_SPINLOCK(nfs_access_lru_lock
);
1681 static LIST_HEAD(nfs_access_lru_list
);
1682 static atomic_long_t nfs_access_nr_entries
;
1684 static void nfs_access_free_entry(struct nfs_access_entry
*entry
)
1686 put_rpccred(entry
->cred
);
1688 smp_mb__before_atomic_dec();
1689 atomic_long_dec(&nfs_access_nr_entries
);
1690 smp_mb__after_atomic_dec();
1693 static void nfs_access_free_list(struct list_head
*head
)
1695 struct nfs_access_entry
*cache
;
1697 while (!list_empty(head
)) {
1698 cache
= list_entry(head
->next
, struct nfs_access_entry
, lru
);
1699 list_del(&cache
->lru
);
1700 nfs_access_free_entry(cache
);
1704 int nfs_access_cache_shrinker(struct shrinker
*shrink
, int nr_to_scan
, gfp_t gfp_mask
)
1707 struct nfs_inode
*nfsi
;
1708 struct nfs_access_entry
*cache
;
1710 if ((gfp_mask
& GFP_KERNEL
) != GFP_KERNEL
)
1711 return (nr_to_scan
== 0) ? 0 : -1;
1713 spin_lock(&nfs_access_lru_lock
);
1714 list_for_each_entry(nfsi
, &nfs_access_lru_list
, access_cache_inode_lru
) {
1715 struct inode
*inode
;
1717 if (nr_to_scan
-- == 0)
1719 inode
= &nfsi
->vfs_inode
;
1720 spin_lock(&inode
->i_lock
);
1721 if (list_empty(&nfsi
->access_cache_entry_lru
))
1722 goto remove_lru_entry
;
1723 cache
= list_entry(nfsi
->access_cache_entry_lru
.next
,
1724 struct nfs_access_entry
, lru
);
1725 list_move(&cache
->lru
, &head
);
1726 rb_erase(&cache
->rb_node
, &nfsi
->access_cache
);
1727 if (!list_empty(&nfsi
->access_cache_entry_lru
))
1728 list_move_tail(&nfsi
->access_cache_inode_lru
,
1729 &nfs_access_lru_list
);
1732 list_del_init(&nfsi
->access_cache_inode_lru
);
1733 smp_mb__before_clear_bit();
1734 clear_bit(NFS_INO_ACL_LRU_SET
, &nfsi
->flags
);
1735 smp_mb__after_clear_bit();
1737 spin_unlock(&inode
->i_lock
);
1739 spin_unlock(&nfs_access_lru_lock
);
1740 nfs_access_free_list(&head
);
1741 return (atomic_long_read(&nfs_access_nr_entries
) / 100) * sysctl_vfs_cache_pressure
;
1744 static void __nfs_access_zap_cache(struct nfs_inode
*nfsi
, struct list_head
*head
)
1746 struct rb_root
*root_node
= &nfsi
->access_cache
;
1748 struct nfs_access_entry
*entry
;
1750 /* Unhook entries from the cache */
1751 while ((n
= rb_first(root_node
)) != NULL
) {
1752 entry
= rb_entry(n
, struct nfs_access_entry
, rb_node
);
1753 rb_erase(n
, root_node
);
1754 list_move(&entry
->lru
, head
);
1756 nfsi
->cache_validity
&= ~NFS_INO_INVALID_ACCESS
;
1759 void nfs_access_zap_cache(struct inode
*inode
)
1763 if (test_bit(NFS_INO_ACL_LRU_SET
, &NFS_I(inode
)->flags
) == 0)
1765 /* Remove from global LRU init */
1766 spin_lock(&nfs_access_lru_lock
);
1767 if (test_and_clear_bit(NFS_INO_ACL_LRU_SET
, &NFS_I(inode
)->flags
))
1768 list_del_init(&NFS_I(inode
)->access_cache_inode_lru
);
1770 spin_lock(&inode
->i_lock
);
1771 __nfs_access_zap_cache(NFS_I(inode
), &head
);
1772 spin_unlock(&inode
->i_lock
);
1773 spin_unlock(&nfs_access_lru_lock
);
1774 nfs_access_free_list(&head
);
1777 static struct nfs_access_entry
*nfs_access_search_rbtree(struct inode
*inode
, struct rpc_cred
*cred
)
1779 struct rb_node
*n
= NFS_I(inode
)->access_cache
.rb_node
;
1780 struct nfs_access_entry
*entry
;
1783 entry
= rb_entry(n
, struct nfs_access_entry
, rb_node
);
1785 if (cred
< entry
->cred
)
1787 else if (cred
> entry
->cred
)
1795 static int nfs_access_get_cached(struct inode
*inode
, struct rpc_cred
*cred
, struct nfs_access_entry
*res
)
1797 struct nfs_inode
*nfsi
= NFS_I(inode
);
1798 struct nfs_access_entry
*cache
;
1801 spin_lock(&inode
->i_lock
);
1802 if (nfsi
->cache_validity
& NFS_INO_INVALID_ACCESS
)
1804 cache
= nfs_access_search_rbtree(inode
, cred
);
1807 if (!nfs_have_delegated_attributes(inode
) &&
1808 !time_in_range_open(jiffies
, cache
->jiffies
, cache
->jiffies
+ nfsi
->attrtimeo
))
1810 res
->jiffies
= cache
->jiffies
;
1811 res
->cred
= cache
->cred
;
1812 res
->mask
= cache
->mask
;
1813 list_move_tail(&cache
->lru
, &nfsi
->access_cache_entry_lru
);
1816 spin_unlock(&inode
->i_lock
);
1819 rb_erase(&cache
->rb_node
, &nfsi
->access_cache
);
1820 list_del(&cache
->lru
);
1821 spin_unlock(&inode
->i_lock
);
1822 nfs_access_free_entry(cache
);
1825 spin_unlock(&inode
->i_lock
);
1826 nfs_access_zap_cache(inode
);
1830 static void nfs_access_add_rbtree(struct inode
*inode
, struct nfs_access_entry
*set
)
1832 struct nfs_inode
*nfsi
= NFS_I(inode
);
1833 struct rb_root
*root_node
= &nfsi
->access_cache
;
1834 struct rb_node
**p
= &root_node
->rb_node
;
1835 struct rb_node
*parent
= NULL
;
1836 struct nfs_access_entry
*entry
;
1838 spin_lock(&inode
->i_lock
);
1839 while (*p
!= NULL
) {
1841 entry
= rb_entry(parent
, struct nfs_access_entry
, rb_node
);
1843 if (set
->cred
< entry
->cred
)
1844 p
= &parent
->rb_left
;
1845 else if (set
->cred
> entry
->cred
)
1846 p
= &parent
->rb_right
;
1850 rb_link_node(&set
->rb_node
, parent
, p
);
1851 rb_insert_color(&set
->rb_node
, root_node
);
1852 list_add_tail(&set
->lru
, &nfsi
->access_cache_entry_lru
);
1853 spin_unlock(&inode
->i_lock
);
1856 rb_replace_node(parent
, &set
->rb_node
, root_node
);
1857 list_add_tail(&set
->lru
, &nfsi
->access_cache_entry_lru
);
1858 list_del(&entry
->lru
);
1859 spin_unlock(&inode
->i_lock
);
1860 nfs_access_free_entry(entry
);
1863 static void nfs_access_add_cache(struct inode
*inode
, struct nfs_access_entry
*set
)
1865 struct nfs_access_entry
*cache
= kmalloc(sizeof(*cache
), GFP_KERNEL
);
1868 RB_CLEAR_NODE(&cache
->rb_node
);
1869 cache
->jiffies
= set
->jiffies
;
1870 cache
->cred
= get_rpccred(set
->cred
);
1871 cache
->mask
= set
->mask
;
1873 nfs_access_add_rbtree(inode
, cache
);
1875 /* Update accounting */
1876 smp_mb__before_atomic_inc();
1877 atomic_long_inc(&nfs_access_nr_entries
);
1878 smp_mb__after_atomic_inc();
1880 /* Add inode to global LRU list */
1881 if (!test_bit(NFS_INO_ACL_LRU_SET
, &NFS_I(inode
)->flags
)) {
1882 spin_lock(&nfs_access_lru_lock
);
1883 if (!test_and_set_bit(NFS_INO_ACL_LRU_SET
, &NFS_I(inode
)->flags
))
1884 list_add_tail(&NFS_I(inode
)->access_cache_inode_lru
,
1885 &nfs_access_lru_list
);
1886 spin_unlock(&nfs_access_lru_lock
);
1890 static int nfs_do_access(struct inode
*inode
, struct rpc_cred
*cred
, int mask
)
1892 struct nfs_access_entry cache
;
1895 status
= nfs_access_get_cached(inode
, cred
, &cache
);
1899 /* Be clever: ask server to check for all possible rights */
1900 cache
.mask
= MAY_EXEC
| MAY_WRITE
| MAY_READ
;
1902 cache
.jiffies
= jiffies
;
1903 status
= NFS_PROTO(inode
)->access(inode
, &cache
);
1905 if (status
== -ESTALE
) {
1906 nfs_zap_caches(inode
);
1907 if (!S_ISDIR(inode
->i_mode
))
1908 set_bit(NFS_INO_STALE
, &NFS_I(inode
)->flags
);
1912 nfs_access_add_cache(inode
, &cache
);
1914 if ((mask
& ~cache
.mask
& (MAY_READ
| MAY_WRITE
| MAY_EXEC
)) == 0)
1919 static int nfs_open_permission_mask(int openflags
)
1923 if (openflags
& FMODE_READ
)
1925 if (openflags
& FMODE_WRITE
)
1927 if (openflags
& FMODE_EXEC
)
1932 int nfs_may_open(struct inode
*inode
, struct rpc_cred
*cred
, int openflags
)
1934 return nfs_do_access(inode
, cred
, nfs_open_permission_mask(openflags
));
1937 int nfs_permission(struct inode
*inode
, int mask
)
1939 struct rpc_cred
*cred
;
1942 nfs_inc_stats(inode
, NFSIOS_VFSACCESS
);
1944 if ((mask
& (MAY_READ
| MAY_WRITE
| MAY_EXEC
)) == 0)
1946 /* Is this sys_access() ? */
1947 if (mask
& (MAY_ACCESS
| MAY_CHDIR
))
1950 switch (inode
->i_mode
& S_IFMT
) {
1954 /* NFSv4 has atomic_open... */
1955 if (nfs_server_capable(inode
, NFS_CAP_ATOMIC_OPEN
)
1956 && (mask
& MAY_OPEN
)
1957 && !(mask
& MAY_EXEC
))
1962 * Optimize away all write operations, since the server
1963 * will check permissions when we perform the op.
1965 if ((mask
& MAY_WRITE
) && !(mask
& MAY_READ
))
1970 if (!NFS_PROTO(inode
)->access
)
1973 cred
= rpc_lookup_cred();
1974 if (!IS_ERR(cred
)) {
1975 res
= nfs_do_access(inode
, cred
, mask
);
1978 res
= PTR_ERR(cred
);
1980 if (!res
&& (mask
& MAY_EXEC
) && !execute_ok(inode
))
1983 dfprintk(VFS
, "NFS: permission(%s/%ld), mask=0x%x, res=%d\n",
1984 inode
->i_sb
->s_id
, inode
->i_ino
, mask
, res
);
1987 res
= nfs_revalidate_inode(NFS_SERVER(inode
), inode
);
1989 res
= generic_permission(inode
, mask
, NULL
);
1995 * version-control: t
1996 * kept-new-versions: 5