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/smp_lock.h>
33 #include <linux/pagevec.h>
34 #include <linux/namei.h>
35 #include <linux/mount.h>
36 #include <linux/sched.h>
39 #include "delegation.h"
43 /* #define NFS_DEBUG_VERBOSE 1 */
45 static int nfs_opendir(struct inode
*, struct file
*);
46 static int nfs_readdir(struct file
*, void *, filldir_t
);
47 static struct dentry
*nfs_lookup(struct inode
*, struct dentry
*, struct nameidata
*);
48 static int nfs_create(struct inode
*, struct dentry
*, int, struct nameidata
*);
49 static int nfs_mkdir(struct inode
*, struct dentry
*, int);
50 static int nfs_rmdir(struct inode
*, struct dentry
*);
51 static int nfs_unlink(struct inode
*, struct dentry
*);
52 static int nfs_symlink(struct inode
*, struct dentry
*, const char *);
53 static int nfs_link(struct dentry
*, struct inode
*, struct dentry
*);
54 static int nfs_mknod(struct inode
*, struct dentry
*, int, dev_t
);
55 static int nfs_rename(struct inode
*, struct dentry
*,
56 struct inode
*, struct dentry
*);
57 static int nfs_fsync_dir(struct file
*, struct dentry
*, int);
58 static loff_t
nfs_llseek_dir(struct file
*, loff_t
, int);
60 const struct file_operations nfs_dir_operations
= {
61 .llseek
= nfs_llseek_dir
,
62 .read
= generic_read_dir
,
63 .readdir
= nfs_readdir
,
65 .release
= nfs_release
,
66 .fsync
= nfs_fsync_dir
,
69 const struct inode_operations nfs_dir_inode_operations
= {
74 .symlink
= nfs_symlink
,
79 .permission
= nfs_permission
,
80 .getattr
= nfs_getattr
,
81 .setattr
= nfs_setattr
,
85 const struct inode_operations nfs3_dir_inode_operations
= {
90 .symlink
= nfs_symlink
,
95 .permission
= nfs_permission
,
96 .getattr
= nfs_getattr
,
97 .setattr
= nfs_setattr
,
98 .listxattr
= nfs3_listxattr
,
99 .getxattr
= nfs3_getxattr
,
100 .setxattr
= nfs3_setxattr
,
101 .removexattr
= nfs3_removexattr
,
103 #endif /* CONFIG_NFS_V3 */
107 static struct dentry
*nfs_atomic_lookup(struct inode
*, struct dentry
*, struct nameidata
*);
108 const struct inode_operations nfs4_dir_inode_operations
= {
109 .create
= nfs_create
,
110 .lookup
= nfs_atomic_lookup
,
112 .unlink
= nfs_unlink
,
113 .symlink
= nfs_symlink
,
117 .rename
= nfs_rename
,
118 .permission
= nfs_permission
,
119 .getattr
= nfs_getattr
,
120 .setattr
= nfs_setattr
,
121 .getxattr
= nfs4_getxattr
,
122 .setxattr
= nfs4_setxattr
,
123 .listxattr
= nfs4_listxattr
,
126 #endif /* CONFIG_NFS_V4 */
132 nfs_opendir(struct inode
*inode
, struct file
*filp
)
136 dfprintk(FILE, "NFS: open dir(%s/%s)\n",
137 filp
->f_path
.dentry
->d_parent
->d_name
.name
,
138 filp
->f_path
.dentry
->d_name
.name
);
140 nfs_inc_stats(inode
, NFSIOS_VFSOPEN
);
142 /* Call generic open code in order to cache credentials */
143 res
= nfs_open(inode
, filp
);
147 typedef __be32
* (*decode_dirent_t
)(__be32
*, struct nfs_entry
*, int);
151 unsigned long page_index
;
154 loff_t current_index
;
155 struct nfs_entry
*entry
;
156 decode_dirent_t decode
;
158 unsigned long timestamp
;
159 unsigned long gencount
;
161 } nfs_readdir_descriptor_t
;
163 /* Now we cache directories properly, by stuffing the dirent
164 * data directly in the page cache.
166 * Inode invalidation due to refresh etc. takes care of
167 * _everything_, no sloppy entry flushing logic, no extraneous
168 * copying, network direct to page cache, the way it was meant
171 * NOTE: Dirent information verification is done always by the
172 * page-in of the RPC reply, nowhere else, this simplies
173 * things substantially.
176 int nfs_readdir_filler(nfs_readdir_descriptor_t
*desc
, struct page
*page
)
178 struct file
*file
= desc
->file
;
179 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
180 struct rpc_cred
*cred
= nfs_file_cred(file
);
181 unsigned long timestamp
, gencount
;
184 dfprintk(DIRCACHE
, "NFS: %s: reading cookie %Lu into page %lu\n",
185 __func__
, (long long)desc
->entry
->cookie
,
190 gencount
= nfs_inc_attr_generation_counter();
191 error
= NFS_PROTO(inode
)->readdir(file
->f_path
.dentry
, cred
, desc
->entry
->cookie
, page
,
192 NFS_SERVER(inode
)->dtsize
, desc
->plus
);
194 /* We requested READDIRPLUS, but the server doesn't grok it */
195 if (error
== -ENOTSUPP
&& desc
->plus
) {
196 NFS_SERVER(inode
)->caps
&= ~NFS_CAP_READDIRPLUS
;
197 clear_bit(NFS_INO_ADVISE_RDPLUS
, &NFS_I(inode
)->flags
);
203 desc
->timestamp
= timestamp
;
204 desc
->gencount
= gencount
;
205 desc
->timestamp_valid
= 1;
206 SetPageUptodate(page
);
207 /* Ensure consistent page alignment of the data.
208 * Note: assumes we have exclusive access to this mapping either
209 * through inode->i_mutex or some other mechanism.
211 if (invalidate_inode_pages2_range(inode
->i_mapping
, page
->index
+ 1, -1) < 0) {
212 /* Should never happen */
213 nfs_zap_mapping(inode
, inode
->i_mapping
);
223 int dir_decode(nfs_readdir_descriptor_t
*desc
)
225 __be32
*p
= desc
->ptr
;
226 p
= desc
->decode(p
, desc
->entry
, desc
->plus
);
230 if (desc
->timestamp_valid
) {
231 desc
->entry
->fattr
->time_start
= desc
->timestamp
;
232 desc
->entry
->fattr
->gencount
= desc
->gencount
;
234 desc
->entry
->fattr
->valid
&= ~NFS_ATTR_FATTR
;
239 void dir_page_release(nfs_readdir_descriptor_t
*desc
)
242 page_cache_release(desc
->page
);
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'.
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
256 int find_dirent(nfs_readdir_descriptor_t
*desc
)
258 struct nfs_entry
*entry
= desc
->entry
;
262 while((status
= dir_decode(desc
)) == 0) {
263 dfprintk(DIRCACHE
, "NFS: %s: examining cookie %Lu\n",
264 __func__
, (unsigned long long)entry
->cookie
);
265 if (entry
->prev_cookie
== *desc
->dir_cookie
)
267 if (loop_count
++ > 200) {
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'.
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
284 int find_dirent_index(nfs_readdir_descriptor_t
*desc
)
286 struct nfs_entry
*entry
= desc
->entry
;
291 status
= dir_decode(desc
);
295 dfprintk(DIRCACHE
, "NFS: found cookie %Lu at index %Ld\n",
296 (unsigned long long)entry
->cookie
, desc
->current_index
);
298 if (desc
->file
->f_pos
== desc
->current_index
) {
299 *desc
->dir_cookie
= entry
->cookie
;
302 desc
->current_index
++;
303 if (loop_count
++ > 200) {
312 * Find the given page, and call find_dirent() or find_dirent_index in
313 * order to try to return the next entry.
316 int find_dirent_page(nfs_readdir_descriptor_t
*desc
)
318 struct inode
*inode
= desc
->file
->f_path
.dentry
->d_inode
;
322 dfprintk(DIRCACHE
, "NFS: %s: searching page %ld for target %Lu\n",
323 __func__
, desc
->page_index
,
324 (long long) *desc
->dir_cookie
);
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.
329 desc
->timestamp_valid
= 0;
330 page
= read_cache_page(inode
->i_mapping
, desc
->page_index
,
331 (filler_t
*)nfs_readdir_filler
, desc
);
333 status
= PTR_ERR(page
);
337 /* NOTE: Someone else may have changed the READDIRPLUS flag */
339 desc
->ptr
= kmap(page
); /* matching kunmap in nfs_do_filldir */
340 if (*desc
->dir_cookie
!= 0)
341 status
= find_dirent(desc
);
343 status
= find_dirent_index(desc
);
345 dir_page_release(desc
);
347 dfprintk(DIRCACHE
, "NFS: %s: returns %d\n", __func__
, status
);
352 * Recurse through the page cache pages, and return a
353 * filled nfs_entry structure of the next directory entry if possible.
355 * The target for the search is '*desc->dir_cookie' if non-0,
356 * 'desc->file->f_pos' otherwise
359 int readdir_search_pagecache(nfs_readdir_descriptor_t
*desc
)
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;
373 dfprintk(DIRCACHE
, "NFS: readdir_search_pagecache() searching for cookie %Lu\n",
374 (unsigned long long)*desc
->dir_cookie
);
377 res
= find_dirent_page(desc
);
380 /* Align to beginning of next page */
382 if (loop_count
++ > 200) {
388 dfprintk(DIRCACHE
, "NFS: %s: returns %d\n", __func__
, res
);
392 static inline unsigned int dt_type(struct inode
*inode
)
394 return (inode
->i_mode
>> 12) & 15;
397 static struct dentry
*nfs_readdir_lookup(nfs_readdir_descriptor_t
*desc
);
400 * Once we've found the start of the dirent within a page: fill 'er up...
403 int nfs_do_filldir(nfs_readdir_descriptor_t
*desc
, void *dirent
,
406 struct file
*file
= desc
->file
;
407 struct nfs_entry
*entry
= desc
->entry
;
408 struct dentry
*dentry
= NULL
;
413 dfprintk(DIRCACHE
, "NFS: nfs_do_filldir() filling starting @ cookie %Lu\n",
414 (unsigned long long)entry
->cookie
);
417 unsigned d_type
= DT_UNKNOWN
;
418 /* Note: entry->prev_cookie contains the cookie for
419 * retrieving the current dirent on the server */
422 /* Get a dentry if we have one */
425 dentry
= nfs_readdir_lookup(desc
);
427 /* Use readdirplus info */
428 if (dentry
!= NULL
&& dentry
->d_inode
!= NULL
) {
429 d_type
= dt_type(dentry
->d_inode
);
430 fileid
= NFS_FILEID(dentry
->d_inode
);
433 res
= filldir(dirent
, entry
->name
, entry
->len
,
434 file
->f_pos
, nfs_compat_user_ino64(fileid
),
439 *desc
->dir_cookie
= entry
->cookie
;
440 if (dir_decode(desc
) != 0) {
444 if (loop_count
++ > 200) {
449 dir_page_release(desc
);
452 dfprintk(DIRCACHE
, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n",
453 (unsigned long long)*desc
->dir_cookie
, res
);
458 * If we cannot find a cookie in our cache, we suspect that this is
459 * because it points to a deleted file, so we ask the server to return
460 * whatever it thinks is the next entry. We then feed this to filldir.
461 * If all goes well, we should then be able to find our way round the
462 * cache on the next call to readdir_search_pagecache();
464 * NOTE: we cannot add the anonymous page to the pagecache because
465 * the data it contains might not be page aligned. Besides,
466 * we should already have a complete representation of the
467 * directory in the page cache by the time we get here.
470 int uncached_readdir(nfs_readdir_descriptor_t
*desc
, void *dirent
,
473 struct file
*file
= desc
->file
;
474 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
475 struct rpc_cred
*cred
= nfs_file_cred(file
);
476 struct page
*page
= NULL
;
478 unsigned long timestamp
, gencount
;
480 dfprintk(DIRCACHE
, "NFS: uncached_readdir() searching for cookie %Lu\n",
481 (unsigned long long)*desc
->dir_cookie
);
483 page
= alloc_page(GFP_HIGHUSER
);
489 gencount
= nfs_inc_attr_generation_counter();
490 status
= NFS_PROTO(inode
)->readdir(file
->f_path
.dentry
, cred
,
491 *desc
->dir_cookie
, page
,
492 NFS_SERVER(inode
)->dtsize
,
495 desc
->ptr
= kmap(page
); /* matching kunmap in nfs_do_filldir */
497 desc
->timestamp
= timestamp
;
498 desc
->gencount
= gencount
;
499 desc
->timestamp_valid
= 1;
500 if ((status
= dir_decode(desc
)) == 0)
501 desc
->entry
->prev_cookie
= *desc
->dir_cookie
;
507 status
= nfs_do_filldir(desc
, dirent
, filldir
);
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;
515 dfprintk(DIRCACHE
, "NFS: %s: returns %d\n",
519 dir_page_release(desc
);
523 /* The file offset position represents the dirent entry number. A
524 last cookie cache takes care of the common case of reading the
527 static int nfs_readdir(struct file
*filp
, void *dirent
, filldir_t filldir
)
529 struct dentry
*dentry
= filp
->f_path
.dentry
;
530 struct inode
*inode
= dentry
->d_inode
;
531 nfs_readdir_descriptor_t my_desc
,
533 struct nfs_entry my_entry
;
535 struct nfs_fattr fattr
;
538 dfprintk(FILE, "NFS: readdir(%s/%s) starting at cookie %llu\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
);
544 * filp->f_pos points to the dirent entry number.
545 * *desc->dir_cookie has the cookie for the next entry. We have
546 * to either find the entry with the appropriate number or
547 * revalidate the cookie.
549 memset(desc
, 0, sizeof(*desc
));
552 desc
->dir_cookie
= &nfs_file_open_context(filp
)->dir_cookie
;
553 desc
->decode
= NFS_PROTO(inode
)->decode_dirent
;
554 desc
->plus
= NFS_USE_READDIRPLUS(inode
);
556 my_entry
.cookie
= my_entry
.prev_cookie
= 0;
559 my_entry
.fattr
= &fattr
;
560 nfs_fattr_init(&fattr
);
561 desc
->entry
= &my_entry
;
563 nfs_block_sillyrename(dentry
);
564 res
= nfs_revalidate_mapping_nolock(inode
, filp
->f_mapping
);
568 while(!desc
->entry
->eof
) {
569 res
= readdir_search_pagecache(desc
);
571 if (res
== -EBADCOOKIE
) {
572 /* This means either end of directory */
573 if (*desc
->dir_cookie
&& desc
->entry
->cookie
!= *desc
->dir_cookie
) {
574 /* Or that the server has 'lost' a cookie */
575 res
= uncached_readdir(desc
, dirent
, filldir
);
582 if (res
== -ETOOSMALL
&& desc
->plus
) {
583 clear_bit(NFS_INO_ADVISE_RDPLUS
, &NFS_I(inode
)->flags
);
584 nfs_zap_caches(inode
);
586 desc
->entry
->eof
= 0;
592 res
= nfs_do_filldir(desc
, dirent
, filldir
);
599 nfs_unblock_sillyrename(dentry
);
602 dfprintk(FILE, "NFS: readdir(%s/%s) returns %ld\n",
603 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
,
608 static loff_t
nfs_llseek_dir(struct file
*filp
, loff_t offset
, int origin
)
610 struct dentry
*dentry
= filp
->f_path
.dentry
;
611 struct inode
*inode
= dentry
->d_inode
;
613 dfprintk(FILE, "NFS: llseek dir(%s/%s, %lld, %d)\n",
614 dentry
->d_parent
->d_name
.name
,
618 mutex_lock(&inode
->i_mutex
);
621 offset
+= filp
->f_pos
;
629 if (offset
!= filp
->f_pos
) {
630 filp
->f_pos
= offset
;
631 nfs_file_open_context(filp
)->dir_cookie
= 0;
634 mutex_unlock(&inode
->i_mutex
);
639 * All directory operations under NFS are synchronous, so fsync()
640 * is a dummy operation.
642 static int nfs_fsync_dir(struct file
*filp
, struct dentry
*dentry
, int datasync
)
644 dfprintk(FILE, "NFS: fsync dir(%s/%s) datasync %d\n",
645 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
,
648 nfs_inc_stats(dentry
->d_inode
, NFSIOS_VFSFSYNC
);
653 * nfs_force_lookup_revalidate - Mark the directory as having changed
654 * @dir - pointer to directory inode
656 * This forces the revalidation code in nfs_lookup_revalidate() to do a
657 * full lookup on all child dentries of 'dir' whenever a change occurs
658 * on the server that might have invalidated our dcache.
660 * The caller should be holding dir->i_lock
662 void nfs_force_lookup_revalidate(struct inode
*dir
)
664 NFS_I(dir
)->cache_change_attribute
++;
668 * A check for whether or not the parent directory has changed.
669 * In the case it has, we assume that the dentries are untrustworthy
670 * and may need to be looked up again.
672 static int nfs_check_verifier(struct inode
*dir
, struct dentry
*dentry
)
676 if (NFS_SERVER(dir
)->flags
& NFS_MOUNT_LOOKUP_CACHE_NONE
)
678 if (!nfs_verify_change_attribute(dir
, dentry
->d_time
))
680 /* Revalidate nfsi->cache_change_attribute before we declare a match */
681 if (nfs_revalidate_inode(NFS_SERVER(dir
), dir
) < 0)
683 if (!nfs_verify_change_attribute(dir
, dentry
->d_time
))
689 * Return the intent data that applies to this particular path component
691 * Note that the current set of intents only apply to the very last
692 * component of the path.
693 * We check for this using LOOKUP_CONTINUE and LOOKUP_PARENT.
695 static inline unsigned int nfs_lookup_check_intent(struct nameidata
*nd
, unsigned int mask
)
697 if (nd
->flags
& (LOOKUP_CONTINUE
|LOOKUP_PARENT
))
699 return nd
->flags
& mask
;
703 * Use intent information to check whether or not we're going to do
704 * an O_EXCL create using this path component.
706 static int nfs_is_exclusive_create(struct inode
*dir
, struct nameidata
*nd
)
708 if (NFS_PROTO(dir
)->version
== 2)
710 return nd
&& nfs_lookup_check_intent(nd
, LOOKUP_EXCL
);
714 * Inode and filehandle revalidation for lookups.
716 * We force revalidation in the cases where the VFS sets LOOKUP_REVAL,
717 * or if the intent information indicates that we're about to open this
718 * particular file and the "nocto" mount flag is not set.
722 int nfs_lookup_verify_inode(struct inode
*inode
, struct nameidata
*nd
)
724 struct nfs_server
*server
= NFS_SERVER(inode
);
726 if (test_bit(NFS_INO_MOUNTPOINT
, &NFS_I(inode
)->flags
))
729 /* VFS wants an on-the-wire revalidation */
730 if (nd
->flags
& LOOKUP_REVAL
)
732 /* This is an open(2) */
733 if (nfs_lookup_check_intent(nd
, LOOKUP_OPEN
) != 0 &&
734 !(server
->flags
& NFS_MOUNT_NOCTO
) &&
735 (S_ISREG(inode
->i_mode
) ||
736 S_ISDIR(inode
->i_mode
)))
740 return nfs_revalidate_inode(server
, inode
);
742 return __nfs_revalidate_inode(server
, inode
);
746 * We judge how long we want to trust negative
747 * dentries by looking at the parent inode mtime.
749 * If parent mtime has changed, we revalidate, else we wait for a
750 * period corresponding to the parent's attribute cache timeout value.
753 int nfs_neg_need_reval(struct inode
*dir
, struct dentry
*dentry
,
754 struct nameidata
*nd
)
756 /* Don't revalidate a negative dentry if we're creating a new file */
757 if (nd
!= NULL
&& nfs_lookup_check_intent(nd
, LOOKUP_CREATE
) != 0)
759 if (NFS_SERVER(dir
)->flags
& NFS_MOUNT_LOOKUP_CACHE_NONEG
)
761 return !nfs_check_verifier(dir
, dentry
);
765 * This is called every time the dcache has a lookup hit,
766 * and we should check whether we can really trust that
769 * NOTE! The hit can be a negative hit too, don't assume
772 * If the parent directory is seen to have changed, we throw out the
773 * cached dentry and do a new lookup.
775 static int nfs_lookup_revalidate(struct dentry
* dentry
, struct nameidata
*nd
)
779 struct dentry
*parent
;
781 struct nfs_fh fhandle
;
782 struct nfs_fattr fattr
;
784 parent
= dget_parent(dentry
);
785 dir
= parent
->d_inode
;
786 nfs_inc_stats(dir
, NFSIOS_DENTRYREVALIDATE
);
787 inode
= dentry
->d_inode
;
790 if (nfs_neg_need_reval(dir
, dentry
, nd
))
795 if (is_bad_inode(inode
)) {
796 dfprintk(LOOKUPCACHE
, "%s: %s/%s has dud inode\n",
797 __func__
, dentry
->d_parent
->d_name
.name
,
798 dentry
->d_name
.name
);
802 if (nfs_have_delegation(inode
, FMODE_READ
))
803 goto out_set_verifier
;
805 /* Force a full look up iff the parent directory has changed */
806 if (!nfs_is_exclusive_create(dir
, nd
) && nfs_check_verifier(dir
, dentry
)) {
807 if (nfs_lookup_verify_inode(inode
, nd
))
812 if (NFS_STALE(inode
))
815 error
= NFS_PROTO(dir
)->lookup(dir
, &dentry
->d_name
, &fhandle
, &fattr
);
818 if (nfs_compare_fh(NFS_FH(inode
), &fhandle
))
820 if ((error
= nfs_refresh_inode(inode
, &fattr
)) != 0)
824 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
827 dfprintk(LOOKUPCACHE
, "NFS: %s(%s/%s) is valid\n",
828 __func__
, dentry
->d_parent
->d_name
.name
,
829 dentry
->d_name
.name
);
834 nfs_mark_for_revalidate(dir
);
835 if (inode
&& S_ISDIR(inode
->i_mode
)) {
836 /* Purge readdir caches. */
837 nfs_zap_caches(inode
);
838 /* If we have submounts, don't unhash ! */
839 if (have_submounts(dentry
))
841 shrink_dcache_parent(dentry
);
845 dfprintk(LOOKUPCACHE
, "NFS: %s(%s/%s) is invalid\n",
846 __func__
, dentry
->d_parent
->d_name
.name
,
847 dentry
->d_name
.name
);
852 * This is called from dput() when d_count is going to 0.
854 static int nfs_dentry_delete(struct dentry
*dentry
)
856 dfprintk(VFS
, "NFS: dentry_delete(%s/%s, %x)\n",
857 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
,
860 /* Unhash any dentry with a stale inode */
861 if (dentry
->d_inode
!= NULL
&& NFS_STALE(dentry
->d_inode
))
864 if (dentry
->d_flags
& DCACHE_NFSFS_RENAMED
) {
865 /* Unhash it, so that ->d_iput() would be called */
868 if (!(dentry
->d_sb
->s_flags
& MS_ACTIVE
)) {
869 /* Unhash it, so that ancestors of killed async unlink
870 * files will be cleaned up during umount */
877 static void nfs_drop_nlink(struct inode
*inode
)
879 spin_lock(&inode
->i_lock
);
880 if (inode
->i_nlink
> 0)
882 spin_unlock(&inode
->i_lock
);
886 * Called when the dentry loses inode.
887 * We use it to clean up silly-renamed files.
889 static void nfs_dentry_iput(struct dentry
*dentry
, struct inode
*inode
)
891 if (S_ISDIR(inode
->i_mode
))
892 /* drop any readdir cache as it could easily be old */
893 NFS_I(inode
)->cache_validity
|= NFS_INO_INVALID_DATA
;
895 if (dentry
->d_flags
& DCACHE_NFSFS_RENAMED
) {
897 nfs_complete_unlink(dentry
, inode
);
902 struct dentry_operations nfs_dentry_operations
= {
903 .d_revalidate
= nfs_lookup_revalidate
,
904 .d_delete
= nfs_dentry_delete
,
905 .d_iput
= nfs_dentry_iput
,
908 static struct dentry
*nfs_lookup(struct inode
*dir
, struct dentry
* dentry
, struct nameidata
*nd
)
911 struct dentry
*parent
;
912 struct inode
*inode
= NULL
;
914 struct nfs_fh fhandle
;
915 struct nfs_fattr fattr
;
917 dfprintk(VFS
, "NFS: lookup(%s/%s)\n",
918 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
);
919 nfs_inc_stats(dir
, NFSIOS_VFSLOOKUP
);
921 res
= ERR_PTR(-ENAMETOOLONG
);
922 if (dentry
->d_name
.len
> NFS_SERVER(dir
)->namelen
)
925 res
= ERR_PTR(-ENOMEM
);
926 dentry
->d_op
= NFS_PROTO(dir
)->dentry_ops
;
929 * If we're doing an exclusive create, optimize away the lookup
930 * but don't hash the dentry.
932 if (nfs_is_exclusive_create(dir
, nd
)) {
933 d_instantiate(dentry
, NULL
);
938 parent
= dentry
->d_parent
;
939 /* Protect against concurrent sillydeletes */
940 nfs_block_sillyrename(parent
);
941 error
= NFS_PROTO(dir
)->lookup(dir
, &dentry
->d_name
, &fhandle
, &fattr
);
942 if (error
== -ENOENT
)
945 res
= ERR_PTR(error
);
946 goto out_unblock_sillyrename
;
948 inode
= nfs_fhget(dentry
->d_sb
, &fhandle
, &fattr
);
949 res
= (struct dentry
*)inode
;
951 goto out_unblock_sillyrename
;
954 res
= d_materialise_unique(dentry
, inode
);
957 goto out_unblock_sillyrename
;
960 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
961 out_unblock_sillyrename
:
962 nfs_unblock_sillyrename(parent
);
968 static int nfs_open_revalidate(struct dentry
*, struct nameidata
*);
970 struct dentry_operations nfs4_dentry_operations
= {
971 .d_revalidate
= nfs_open_revalidate
,
972 .d_delete
= nfs_dentry_delete
,
973 .d_iput
= nfs_dentry_iput
,
977 * Use intent information to determine whether we need to substitute
978 * the NFSv4-style stateful OPEN for the LOOKUP call
980 static int is_atomic_open(struct nameidata
*nd
)
982 if (nd
== NULL
|| nfs_lookup_check_intent(nd
, LOOKUP_OPEN
) == 0)
984 /* NFS does not (yet) have a stateful open for directories */
985 if (nd
->flags
& LOOKUP_DIRECTORY
)
987 /* Are we trying to write to a read only partition? */
988 if (__mnt_is_readonly(nd
->path
.mnt
) &&
989 (nd
->intent
.open
.flags
& (O_CREAT
|O_TRUNC
|FMODE_WRITE
)))
994 static struct dentry
*nfs_atomic_lookup(struct inode
*dir
, struct dentry
*dentry
, struct nameidata
*nd
)
996 struct dentry
*res
= NULL
;
999 dfprintk(VFS
, "NFS: atomic_lookup(%s/%ld), %s\n",
1000 dir
->i_sb
->s_id
, dir
->i_ino
, dentry
->d_name
.name
);
1002 /* Check that we are indeed trying to open this file */
1003 if (!is_atomic_open(nd
))
1006 if (dentry
->d_name
.len
> NFS_SERVER(dir
)->namelen
) {
1007 res
= ERR_PTR(-ENAMETOOLONG
);
1010 dentry
->d_op
= NFS_PROTO(dir
)->dentry_ops
;
1012 /* Let vfs_create() deal with O_EXCL. Instantiate, but don't hash
1014 if (nd
->flags
& LOOKUP_EXCL
) {
1015 d_instantiate(dentry
, NULL
);
1019 /* Open the file on the server */
1020 res
= nfs4_atomic_open(dir
, dentry
, nd
);
1022 error
= PTR_ERR(res
);
1024 /* Make a negative dentry */
1028 /* This turned out not to be a regular file */
1033 if (!(nd
->intent
.open
.flags
& O_NOFOLLOW
))
1039 } else if (res
!= NULL
)
1044 return nfs_lookup(dir
, dentry
, nd
);
1047 static int nfs_open_revalidate(struct dentry
*dentry
, struct nameidata
*nd
)
1049 struct dentry
*parent
= NULL
;
1050 struct inode
*inode
= dentry
->d_inode
;
1052 int openflags
, ret
= 0;
1054 if (!is_atomic_open(nd
))
1056 parent
= dget_parent(dentry
);
1057 dir
= parent
->d_inode
;
1058 /* We can't create new files in nfs_open_revalidate(), so we
1059 * optimize away revalidation of negative dentries.
1061 if (inode
== NULL
) {
1062 if (!nfs_neg_need_reval(dir
, dentry
, nd
))
1067 /* NFS only supports OPEN on regular files */
1068 if (!S_ISREG(inode
->i_mode
))
1070 openflags
= nd
->intent
.open
.flags
;
1071 /* We cannot do exclusive creation on a positive dentry */
1072 if ((openflags
& (O_CREAT
|O_EXCL
)) == (O_CREAT
|O_EXCL
))
1074 /* We can't create new files, or truncate existing ones here */
1075 openflags
&= ~(O_CREAT
|O_TRUNC
);
1078 * Note: we're not holding inode->i_mutex and so may be racing with
1079 * operations that change the directory. We therefore save the
1080 * change attribute *before* we do the RPC call.
1082 ret
= nfs4_open_revalidate(dir
, dentry
, openflags
, nd
);
1091 return nfs_lookup_revalidate(dentry
, nd
);
1093 #endif /* CONFIG_NFSV4 */
1095 static struct dentry
*nfs_readdir_lookup(nfs_readdir_descriptor_t
*desc
)
1097 struct dentry
*parent
= desc
->file
->f_path
.dentry
;
1098 struct inode
*dir
= parent
->d_inode
;
1099 struct nfs_entry
*entry
= desc
->entry
;
1100 struct dentry
*dentry
, *alias
;
1101 struct qstr name
= {
1102 .name
= entry
->name
,
1105 struct inode
*inode
;
1106 unsigned long verf
= nfs_save_change_attribute(dir
);
1110 if (name
.name
[0] == '.' && name
.name
[1] == '.')
1111 return dget_parent(parent
);
1114 if (name
.name
[0] == '.')
1115 return dget(parent
);
1118 spin_lock(&dir
->i_lock
);
1119 if (NFS_I(dir
)->cache_validity
& NFS_INO_INVALID_DATA
) {
1120 spin_unlock(&dir
->i_lock
);
1123 spin_unlock(&dir
->i_lock
);
1125 name
.hash
= full_name_hash(name
.name
, name
.len
);
1126 dentry
= d_lookup(parent
, &name
);
1127 if (dentry
!= NULL
) {
1128 /* Is this a positive dentry that matches the readdir info? */
1129 if (dentry
->d_inode
!= NULL
&&
1130 (NFS_FILEID(dentry
->d_inode
) == entry
->ino
||
1131 d_mountpoint(dentry
))) {
1132 if (!desc
->plus
|| entry
->fh
->size
== 0)
1134 if (nfs_compare_fh(NFS_FH(dentry
->d_inode
),
1138 /* No, so d_drop to allow one to be created */
1142 if (!desc
->plus
|| !(entry
->fattr
->valid
& NFS_ATTR_FATTR
))
1144 if (name
.len
> NFS_SERVER(dir
)->namelen
)
1146 /* Note: caller is already holding the dir->i_mutex! */
1147 dentry
= d_alloc(parent
, &name
);
1150 dentry
->d_op
= NFS_PROTO(dir
)->dentry_ops
;
1151 inode
= nfs_fhget(dentry
->d_sb
, entry
->fh
, entry
->fattr
);
1152 if (IS_ERR(inode
)) {
1157 alias
= d_materialise_unique(dentry
, inode
);
1158 if (alias
!= NULL
) {
1166 nfs_set_verifier(dentry
, verf
);
1171 * Code common to create, mkdir, and mknod.
1173 int nfs_instantiate(struct dentry
*dentry
, struct nfs_fh
*fhandle
,
1174 struct nfs_fattr
*fattr
)
1176 struct dentry
*parent
= dget_parent(dentry
);
1177 struct inode
*dir
= parent
->d_inode
;
1178 struct inode
*inode
;
1179 int error
= -EACCES
;
1183 /* We may have been initialized further down */
1184 if (dentry
->d_inode
)
1186 if (fhandle
->size
== 0) {
1187 error
= NFS_PROTO(dir
)->lookup(dir
, &dentry
->d_name
, fhandle
, fattr
);
1191 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1192 if (!(fattr
->valid
& NFS_ATTR_FATTR
)) {
1193 struct nfs_server
*server
= NFS_SB(dentry
->d_sb
);
1194 error
= server
->nfs_client
->rpc_ops
->getattr(server
, fhandle
, fattr
);
1198 inode
= nfs_fhget(dentry
->d_sb
, fhandle
, fattr
);
1199 error
= PTR_ERR(inode
);
1202 d_add(dentry
, inode
);
1207 nfs_mark_for_revalidate(dir
);
1213 * Following a failed create operation, we drop the dentry rather
1214 * than retain a negative dentry. This avoids a problem in the event
1215 * that the operation succeeded on the server, but an error in the
1216 * reply path made it appear to have failed.
1218 static int nfs_create(struct inode
*dir
, struct dentry
*dentry
, int mode
,
1219 struct nameidata
*nd
)
1225 dfprintk(VFS
, "NFS: create(%s/%ld), %s\n",
1226 dir
->i_sb
->s_id
, dir
->i_ino
, dentry
->d_name
.name
);
1228 attr
.ia_mode
= mode
;
1229 attr
.ia_valid
= ATTR_MODE
;
1231 if ((nd
->flags
& LOOKUP_CREATE
) != 0)
1232 open_flags
= nd
->intent
.open
.flags
;
1234 error
= NFS_PROTO(dir
)->create(dir
, dentry
, &attr
, open_flags
, nd
);
1244 * See comments for nfs_proc_create regarding failed operations.
1247 nfs_mknod(struct inode
*dir
, struct dentry
*dentry
, int mode
, dev_t rdev
)
1252 dfprintk(VFS
, "NFS: mknod(%s/%ld), %s\n",
1253 dir
->i_sb
->s_id
, dir
->i_ino
, dentry
->d_name
.name
);
1255 if (!new_valid_dev(rdev
))
1258 attr
.ia_mode
= mode
;
1259 attr
.ia_valid
= ATTR_MODE
;
1261 status
= NFS_PROTO(dir
)->mknod(dir
, dentry
, &attr
, rdev
);
1271 * See comments for nfs_proc_create regarding failed operations.
1273 static int nfs_mkdir(struct inode
*dir
, struct dentry
*dentry
, int mode
)
1278 dfprintk(VFS
, "NFS: mkdir(%s/%ld), %s\n",
1279 dir
->i_sb
->s_id
, dir
->i_ino
, dentry
->d_name
.name
);
1281 attr
.ia_valid
= ATTR_MODE
;
1282 attr
.ia_mode
= mode
| S_IFDIR
;
1284 error
= NFS_PROTO(dir
)->mkdir(dir
, dentry
, &attr
);
1293 static void nfs_dentry_handle_enoent(struct dentry
*dentry
)
1295 if (dentry
->d_inode
!= NULL
&& !d_unhashed(dentry
))
1299 static int nfs_rmdir(struct inode
*dir
, struct dentry
*dentry
)
1303 dfprintk(VFS
, "NFS: rmdir(%s/%ld), %s\n",
1304 dir
->i_sb
->s_id
, dir
->i_ino
, dentry
->d_name
.name
);
1306 error
= NFS_PROTO(dir
)->rmdir(dir
, &dentry
->d_name
);
1307 /* Ensure the VFS deletes this inode */
1308 if (error
== 0 && dentry
->d_inode
!= NULL
)
1309 clear_nlink(dentry
->d_inode
);
1310 else if (error
== -ENOENT
)
1311 nfs_dentry_handle_enoent(dentry
);
1316 static int nfs_sillyrename(struct inode
*dir
, struct dentry
*dentry
)
1318 static unsigned int sillycounter
;
1319 const int fileidsize
= sizeof(NFS_FILEID(dentry
->d_inode
))*2;
1320 const int countersize
= sizeof(sillycounter
)*2;
1321 const int slen
= sizeof(".nfs")+fileidsize
+countersize
-1;
1324 struct dentry
*sdentry
;
1327 dfprintk(VFS
, "NFS: silly-rename(%s/%s, ct=%d)\n",
1328 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
,
1329 atomic_read(&dentry
->d_count
));
1330 nfs_inc_stats(dir
, NFSIOS_SILLYRENAME
);
1333 * We don't allow a dentry to be silly-renamed twice.
1336 if (dentry
->d_flags
& DCACHE_NFSFS_RENAMED
)
1339 sprintf(silly
, ".nfs%*.*Lx",
1340 fileidsize
, fileidsize
,
1341 (unsigned long long)NFS_FILEID(dentry
->d_inode
));
1343 /* Return delegation in anticipation of the rename */
1344 nfs_inode_return_delegation(dentry
->d_inode
);
1348 char *suffix
= silly
+ slen
- countersize
;
1352 sprintf(suffix
, "%*.*x", countersize
, countersize
, sillycounter
);
1354 dfprintk(VFS
, "NFS: trying to rename %s to %s\n",
1355 dentry
->d_name
.name
, silly
);
1357 sdentry
= lookup_one_len(silly
, dentry
->d_parent
, slen
);
1359 * N.B. Better to return EBUSY here ... it could be
1360 * dangerous to delete the file while it's in use.
1362 if (IS_ERR(sdentry
))
1364 } while(sdentry
->d_inode
!= NULL
); /* need negative lookup */
1366 qsilly
.name
= silly
;
1367 qsilly
.len
= strlen(silly
);
1368 if (dentry
->d_inode
) {
1369 error
= NFS_PROTO(dir
)->rename(dir
, &dentry
->d_name
,
1371 nfs_mark_for_revalidate(dentry
->d_inode
);
1373 error
= NFS_PROTO(dir
)->rename(dir
, &dentry
->d_name
,
1376 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1377 d_move(dentry
, sdentry
);
1378 error
= nfs_async_unlink(dir
, dentry
);
1379 /* If we return 0 we don't unlink */
1387 * Remove a file after making sure there are no pending writes,
1388 * and after checking that the file has only one user.
1390 * We invalidate the attribute cache and free the inode prior to the operation
1391 * to avoid possible races if the server reuses the inode.
1393 static int nfs_safe_remove(struct dentry
*dentry
)
1395 struct inode
*dir
= dentry
->d_parent
->d_inode
;
1396 struct inode
*inode
= dentry
->d_inode
;
1399 dfprintk(VFS
, "NFS: safe_remove(%s/%s)\n",
1400 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
);
1402 /* If the dentry was sillyrenamed, we simply call d_delete() */
1403 if (dentry
->d_flags
& DCACHE_NFSFS_RENAMED
) {
1408 if (inode
!= NULL
) {
1409 nfs_inode_return_delegation(inode
);
1410 error
= NFS_PROTO(dir
)->remove(dir
, &dentry
->d_name
);
1411 /* The VFS may want to delete this inode */
1413 nfs_drop_nlink(inode
);
1414 nfs_mark_for_revalidate(inode
);
1416 error
= NFS_PROTO(dir
)->remove(dir
, &dentry
->d_name
);
1417 if (error
== -ENOENT
)
1418 nfs_dentry_handle_enoent(dentry
);
1423 /* We do silly rename. In case sillyrename() returns -EBUSY, the inode
1424 * belongs to an active ".nfs..." file and we return -EBUSY.
1426 * If sillyrename() returns 0, we do nothing, otherwise we unlink.
1428 static int nfs_unlink(struct inode
*dir
, struct dentry
*dentry
)
1431 int need_rehash
= 0;
1433 dfprintk(VFS
, "NFS: unlink(%s/%ld, %s)\n", dir
->i_sb
->s_id
,
1434 dir
->i_ino
, dentry
->d_name
.name
);
1436 spin_lock(&dcache_lock
);
1437 spin_lock(&dentry
->d_lock
);
1438 if (atomic_read(&dentry
->d_count
) > 1) {
1439 spin_unlock(&dentry
->d_lock
);
1440 spin_unlock(&dcache_lock
);
1441 /* Start asynchronous writeout of the inode */
1442 write_inode_now(dentry
->d_inode
, 0);
1443 error
= nfs_sillyrename(dir
, dentry
);
1446 if (!d_unhashed(dentry
)) {
1450 spin_unlock(&dentry
->d_lock
);
1451 spin_unlock(&dcache_lock
);
1452 error
= nfs_safe_remove(dentry
);
1453 if (!error
|| error
== -ENOENT
) {
1454 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1455 } else if (need_rehash
)
1461 * To create a symbolic link, most file systems instantiate a new inode,
1462 * add a page to it containing the path, then write it out to the disk
1463 * using prepare_write/commit_write.
1465 * Unfortunately the NFS client can't create the in-core inode first
1466 * because it needs a file handle to create an in-core inode (see
1467 * fs/nfs/inode.c:nfs_fhget). We only have a file handle *after* the
1468 * symlink request has completed on the server.
1470 * So instead we allocate a raw page, copy the symname into it, then do
1471 * the SYMLINK request with the page as the buffer. If it succeeds, we
1472 * now have a new file handle and can instantiate an in-core NFS inode
1473 * and move the raw page into its mapping.
1475 static int nfs_symlink(struct inode
*dir
, struct dentry
*dentry
, const char *symname
)
1477 struct pagevec lru_pvec
;
1481 unsigned int pathlen
= strlen(symname
);
1484 dfprintk(VFS
, "NFS: symlink(%s/%ld, %s, %s)\n", dir
->i_sb
->s_id
,
1485 dir
->i_ino
, dentry
->d_name
.name
, symname
);
1487 if (pathlen
> PAGE_SIZE
)
1488 return -ENAMETOOLONG
;
1490 attr
.ia_mode
= S_IFLNK
| S_IRWXUGO
;
1491 attr
.ia_valid
= ATTR_MODE
;
1493 page
= alloc_page(GFP_HIGHUSER
);
1497 kaddr
= kmap_atomic(page
, KM_USER0
);
1498 memcpy(kaddr
, symname
, pathlen
);
1499 if (pathlen
< PAGE_SIZE
)
1500 memset(kaddr
+ pathlen
, 0, PAGE_SIZE
- pathlen
);
1501 kunmap_atomic(kaddr
, KM_USER0
);
1503 error
= NFS_PROTO(dir
)->symlink(dir
, dentry
, page
, pathlen
, &attr
);
1505 dfprintk(VFS
, "NFS: symlink(%s/%ld, %s, %s) error %d\n",
1506 dir
->i_sb
->s_id
, dir
->i_ino
,
1507 dentry
->d_name
.name
, symname
, error
);
1514 * No big deal if we can't add this page to the page cache here.
1515 * READLINK will get the missing page from the server if needed.
1517 pagevec_init(&lru_pvec
, 0);
1518 if (!add_to_page_cache(page
, dentry
->d_inode
->i_mapping
, 0,
1520 pagevec_add(&lru_pvec
, page
);
1521 pagevec_lru_add_file(&lru_pvec
);
1522 SetPageUptodate(page
);
1531 nfs_link(struct dentry
*old_dentry
, struct inode
*dir
, struct dentry
*dentry
)
1533 struct inode
*inode
= old_dentry
->d_inode
;
1536 dfprintk(VFS
, "NFS: link(%s/%s -> %s/%s)\n",
1537 old_dentry
->d_parent
->d_name
.name
, old_dentry
->d_name
.name
,
1538 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
);
1541 error
= NFS_PROTO(dir
)->link(inode
, dir
, &dentry
->d_name
);
1543 atomic_inc(&inode
->i_count
);
1544 d_add(dentry
, inode
);
1551 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a
1552 * different file handle for the same inode after a rename (e.g. when
1553 * moving to a different directory). A fail-safe method to do so would
1554 * be to look up old_dir/old_name, create a link to new_dir/new_name and
1555 * rename the old file using the sillyrename stuff. This way, the original
1556 * file in old_dir will go away when the last process iput()s the inode.
1560 * It actually works quite well. One needs to have the possibility for
1561 * at least one ".nfs..." file in each directory the file ever gets
1562 * moved or linked to which happens automagically with the new
1563 * implementation that only depends on the dcache stuff instead of
1564 * using the inode layer
1566 * Unfortunately, things are a little more complicated than indicated
1567 * above. For a cross-directory move, we want to make sure we can get
1568 * rid of the old inode after the operation. This means there must be
1569 * no pending writes (if it's a file), and the use count must be 1.
1570 * If these conditions are met, we can drop the dentries before doing
1573 static int nfs_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
1574 struct inode
*new_dir
, struct dentry
*new_dentry
)
1576 struct inode
*old_inode
= old_dentry
->d_inode
;
1577 struct inode
*new_inode
= new_dentry
->d_inode
;
1578 struct dentry
*dentry
= NULL
, *rehash
= NULL
;
1582 * To prevent any new references to the target during the rename,
1583 * we unhash the dentry and free the inode in advance.
1585 if (!d_unhashed(new_dentry
)) {
1587 rehash
= new_dentry
;
1590 dfprintk(VFS
, "NFS: rename(%s/%s -> %s/%s, ct=%d)\n",
1591 old_dentry
->d_parent
->d_name
.name
, old_dentry
->d_name
.name
,
1592 new_dentry
->d_parent
->d_name
.name
, new_dentry
->d_name
.name
,
1593 atomic_read(&new_dentry
->d_count
));
1596 * First check whether the target is busy ... we can't
1597 * safely do _any_ rename if the target is in use.
1599 * For files, make a copy of the dentry and then do a
1600 * silly-rename. If the silly-rename succeeds, the
1601 * copied dentry is hashed and becomes the new target.
1605 if (S_ISDIR(new_inode
->i_mode
)) {
1607 if (!S_ISDIR(old_inode
->i_mode
))
1609 } else if (atomic_read(&new_dentry
->d_count
) > 2) {
1611 /* copy the target dentry's name */
1612 dentry
= d_alloc(new_dentry
->d_parent
,
1613 &new_dentry
->d_name
);
1617 /* silly-rename the existing target ... */
1618 err
= nfs_sillyrename(new_dir
, new_dentry
);
1620 new_dentry
= rehash
= dentry
;
1622 /* instantiate the replacement target */
1623 d_instantiate(new_dentry
, NULL
);
1624 } else if (atomic_read(&new_dentry
->d_count
) > 1)
1625 /* dentry still busy? */
1631 * ... prune child dentries and writebacks if needed.
1633 if (atomic_read(&old_dentry
->d_count
) > 1) {
1634 if (S_ISREG(old_inode
->i_mode
))
1635 nfs_wb_all(old_inode
);
1636 shrink_dcache_parent(old_dentry
);
1638 nfs_inode_return_delegation(old_inode
);
1640 if (new_inode
!= NULL
)
1641 nfs_inode_return_delegation(new_inode
);
1643 error
= NFS_PROTO(old_dir
)->rename(old_dir
, &old_dentry
->d_name
,
1644 new_dir
, &new_dentry
->d_name
);
1645 nfs_mark_for_revalidate(old_inode
);
1650 if (new_inode
!= NULL
)
1651 nfs_drop_nlink(new_inode
);
1652 d_move(old_dentry
, new_dentry
);
1653 nfs_set_verifier(new_dentry
,
1654 nfs_save_change_attribute(new_dir
));
1655 } else if (error
== -ENOENT
)
1656 nfs_dentry_handle_enoent(old_dentry
);
1658 /* new dentry created? */
1664 static DEFINE_SPINLOCK(nfs_access_lru_lock
);
1665 static LIST_HEAD(nfs_access_lru_list
);
1666 static atomic_long_t nfs_access_nr_entries
;
1668 static void nfs_access_free_entry(struct nfs_access_entry
*entry
)
1670 put_rpccred(entry
->cred
);
1672 smp_mb__before_atomic_dec();
1673 atomic_long_dec(&nfs_access_nr_entries
);
1674 smp_mb__after_atomic_dec();
1677 int nfs_access_cache_shrinker(int nr_to_scan
, gfp_t gfp_mask
)
1680 struct nfs_inode
*nfsi
;
1681 struct nfs_access_entry
*cache
;
1684 spin_lock(&nfs_access_lru_lock
);
1685 list_for_each_entry(nfsi
, &nfs_access_lru_list
, access_cache_inode_lru
) {
1686 struct rw_semaphore
*s_umount
;
1687 struct inode
*inode
;
1689 if (nr_to_scan
-- == 0)
1691 s_umount
= &nfsi
->vfs_inode
.i_sb
->s_umount
;
1692 if (!down_read_trylock(s_umount
))
1694 inode
= igrab(&nfsi
->vfs_inode
);
1695 if (inode
== NULL
) {
1699 spin_lock(&inode
->i_lock
);
1700 if (list_empty(&nfsi
->access_cache_entry_lru
))
1701 goto remove_lru_entry
;
1702 cache
= list_entry(nfsi
->access_cache_entry_lru
.next
,
1703 struct nfs_access_entry
, lru
);
1704 list_move(&cache
->lru
, &head
);
1705 rb_erase(&cache
->rb_node
, &nfsi
->access_cache
);
1706 if (!list_empty(&nfsi
->access_cache_entry_lru
))
1707 list_move_tail(&nfsi
->access_cache_inode_lru
,
1708 &nfs_access_lru_list
);
1711 list_del_init(&nfsi
->access_cache_inode_lru
);
1712 clear_bit(NFS_INO_ACL_LRU_SET
, &nfsi
->flags
);
1714 spin_unlock(&inode
->i_lock
);
1715 spin_unlock(&nfs_access_lru_lock
);
1720 spin_unlock(&nfs_access_lru_lock
);
1721 while (!list_empty(&head
)) {
1722 cache
= list_entry(head
.next
, struct nfs_access_entry
, lru
);
1723 list_del(&cache
->lru
);
1724 nfs_access_free_entry(cache
);
1726 return (atomic_long_read(&nfs_access_nr_entries
) / 100) * sysctl_vfs_cache_pressure
;
1729 static void __nfs_access_zap_cache(struct inode
*inode
)
1731 struct nfs_inode
*nfsi
= NFS_I(inode
);
1732 struct rb_root
*root_node
= &nfsi
->access_cache
;
1733 struct rb_node
*n
, *dispose
= NULL
;
1734 struct nfs_access_entry
*entry
;
1736 /* Unhook entries from the cache */
1737 while ((n
= rb_first(root_node
)) != NULL
) {
1738 entry
= rb_entry(n
, struct nfs_access_entry
, rb_node
);
1739 rb_erase(n
, root_node
);
1740 list_del(&entry
->lru
);
1741 n
->rb_left
= dispose
;
1744 nfsi
->cache_validity
&= ~NFS_INO_INVALID_ACCESS
;
1745 spin_unlock(&inode
->i_lock
);
1747 /* Now kill them all! */
1748 while (dispose
!= NULL
) {
1750 dispose
= n
->rb_left
;
1751 nfs_access_free_entry(rb_entry(n
, struct nfs_access_entry
, rb_node
));
1755 void nfs_access_zap_cache(struct inode
*inode
)
1757 /* Remove from global LRU init */
1758 if (test_and_clear_bit(NFS_INO_ACL_LRU_SET
, &NFS_I(inode
)->flags
)) {
1759 spin_lock(&nfs_access_lru_lock
);
1760 list_del_init(&NFS_I(inode
)->access_cache_inode_lru
);
1761 spin_unlock(&nfs_access_lru_lock
);
1764 spin_lock(&inode
->i_lock
);
1765 /* This will release the spinlock */
1766 __nfs_access_zap_cache(inode
);
1769 static struct nfs_access_entry
*nfs_access_search_rbtree(struct inode
*inode
, struct rpc_cred
*cred
)
1771 struct rb_node
*n
= NFS_I(inode
)->access_cache
.rb_node
;
1772 struct nfs_access_entry
*entry
;
1775 entry
= rb_entry(n
, struct nfs_access_entry
, rb_node
);
1777 if (cred
< entry
->cred
)
1779 else if (cred
> entry
->cred
)
1787 static int nfs_access_get_cached(struct inode
*inode
, struct rpc_cred
*cred
, struct nfs_access_entry
*res
)
1789 struct nfs_inode
*nfsi
= NFS_I(inode
);
1790 struct nfs_access_entry
*cache
;
1793 spin_lock(&inode
->i_lock
);
1794 if (nfsi
->cache_validity
& NFS_INO_INVALID_ACCESS
)
1796 cache
= nfs_access_search_rbtree(inode
, cred
);
1799 if (!nfs_have_delegation(inode
, FMODE_READ
) &&
1800 !time_in_range_open(jiffies
, cache
->jiffies
, cache
->jiffies
+ nfsi
->attrtimeo
))
1802 res
->jiffies
= cache
->jiffies
;
1803 res
->cred
= cache
->cred
;
1804 res
->mask
= cache
->mask
;
1805 list_move_tail(&cache
->lru
, &nfsi
->access_cache_entry_lru
);
1808 spin_unlock(&inode
->i_lock
);
1811 rb_erase(&cache
->rb_node
, &nfsi
->access_cache
);
1812 list_del(&cache
->lru
);
1813 spin_unlock(&inode
->i_lock
);
1814 nfs_access_free_entry(cache
);
1817 /* This will release the spinlock */
1818 __nfs_access_zap_cache(inode
);
1822 static void nfs_access_add_rbtree(struct inode
*inode
, struct nfs_access_entry
*set
)
1824 struct nfs_inode
*nfsi
= NFS_I(inode
);
1825 struct rb_root
*root_node
= &nfsi
->access_cache
;
1826 struct rb_node
**p
= &root_node
->rb_node
;
1827 struct rb_node
*parent
= NULL
;
1828 struct nfs_access_entry
*entry
;
1830 spin_lock(&inode
->i_lock
);
1831 while (*p
!= NULL
) {
1833 entry
= rb_entry(parent
, struct nfs_access_entry
, rb_node
);
1835 if (set
->cred
< entry
->cred
)
1836 p
= &parent
->rb_left
;
1837 else if (set
->cred
> entry
->cred
)
1838 p
= &parent
->rb_right
;
1842 rb_link_node(&set
->rb_node
, parent
, p
);
1843 rb_insert_color(&set
->rb_node
, root_node
);
1844 list_add_tail(&set
->lru
, &nfsi
->access_cache_entry_lru
);
1845 spin_unlock(&inode
->i_lock
);
1848 rb_replace_node(parent
, &set
->rb_node
, root_node
);
1849 list_add_tail(&set
->lru
, &nfsi
->access_cache_entry_lru
);
1850 list_del(&entry
->lru
);
1851 spin_unlock(&inode
->i_lock
);
1852 nfs_access_free_entry(entry
);
1855 static void nfs_access_add_cache(struct inode
*inode
, struct nfs_access_entry
*set
)
1857 struct nfs_access_entry
*cache
= kmalloc(sizeof(*cache
), GFP_KERNEL
);
1860 RB_CLEAR_NODE(&cache
->rb_node
);
1861 cache
->jiffies
= set
->jiffies
;
1862 cache
->cred
= get_rpccred(set
->cred
);
1863 cache
->mask
= set
->mask
;
1865 nfs_access_add_rbtree(inode
, cache
);
1867 /* Update accounting */
1868 smp_mb__before_atomic_inc();
1869 atomic_long_inc(&nfs_access_nr_entries
);
1870 smp_mb__after_atomic_inc();
1872 /* Add inode to global LRU list */
1873 if (!test_and_set_bit(NFS_INO_ACL_LRU_SET
, &NFS_I(inode
)->flags
)) {
1874 spin_lock(&nfs_access_lru_lock
);
1875 list_add_tail(&NFS_I(inode
)->access_cache_inode_lru
, &nfs_access_lru_list
);
1876 spin_unlock(&nfs_access_lru_lock
);
1880 static int nfs_do_access(struct inode
*inode
, struct rpc_cred
*cred
, int mask
)
1882 struct nfs_access_entry cache
;
1885 status
= nfs_access_get_cached(inode
, cred
, &cache
);
1889 /* Be clever: ask server to check for all possible rights */
1890 cache
.mask
= MAY_EXEC
| MAY_WRITE
| MAY_READ
;
1892 cache
.jiffies
= jiffies
;
1893 status
= NFS_PROTO(inode
)->access(inode
, &cache
);
1895 if (status
== -ESTALE
) {
1896 nfs_zap_caches(inode
);
1897 if (!S_ISDIR(inode
->i_mode
))
1898 set_bit(NFS_INO_STALE
, &NFS_I(inode
)->flags
);
1902 nfs_access_add_cache(inode
, &cache
);
1904 if ((mask
& ~cache
.mask
& (MAY_READ
| MAY_WRITE
| MAY_EXEC
)) == 0)
1909 static int nfs_open_permission_mask(int openflags
)
1913 if (openflags
& FMODE_READ
)
1915 if (openflags
& FMODE_WRITE
)
1917 if (openflags
& FMODE_EXEC
)
1922 int nfs_may_open(struct inode
*inode
, struct rpc_cred
*cred
, int openflags
)
1924 return nfs_do_access(inode
, cred
, nfs_open_permission_mask(openflags
));
1927 int nfs_permission(struct inode
*inode
, int mask
)
1929 struct rpc_cred
*cred
;
1932 nfs_inc_stats(inode
, NFSIOS_VFSACCESS
);
1934 if ((mask
& (MAY_READ
| MAY_WRITE
| MAY_EXEC
)) == 0)
1936 /* Is this sys_access() ? */
1937 if (mask
& MAY_ACCESS
)
1940 switch (inode
->i_mode
& S_IFMT
) {
1944 /* NFSv4 has atomic_open... */
1945 if (nfs_server_capable(inode
, NFS_CAP_ATOMIC_OPEN
)
1946 && (mask
& MAY_OPEN
)
1947 && !(mask
& MAY_EXEC
))
1952 * Optimize away all write operations, since the server
1953 * will check permissions when we perform the op.
1955 if ((mask
& MAY_WRITE
) && !(mask
& MAY_READ
))
1960 if (!NFS_PROTO(inode
)->access
)
1963 cred
= rpc_lookup_cred();
1964 if (!IS_ERR(cred
)) {
1965 res
= nfs_do_access(inode
, cred
, mask
);
1968 res
= PTR_ERR(cred
);
1970 if (!res
&& (mask
& MAY_EXEC
) && !execute_ok(inode
))
1973 dfprintk(VFS
, "NFS: permission(%s/%ld), mask=0x%x, res=%d\n",
1974 inode
->i_sb
->s_id
, inode
->i_ino
, mask
, res
);
1977 res
= nfs_revalidate_inode(NFS_SERVER(inode
), inode
);
1979 res
= generic_permission(inode
, mask
, NULL
);
1985 * version-control: t
1986 * kept-new-versions: 5