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>
38 #include "delegation.h"
41 #define NFS_PARANOIA 1
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
*, struct dentry
*, 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(VFS
, "NFS: opendir(%s/%ld)\n",
136 inode
->i_sb
->s_id
, inode
->i_ino
);
139 /* Call generic open code in order to cache credentials */
140 res
= nfs_open(inode
, filp
);
145 typedef __be32
* (*decode_dirent_t
)(__be32
*, struct nfs_entry
*, int);
149 unsigned long page_index
;
152 loff_t current_index
;
153 struct nfs_entry
*entry
;
154 decode_dirent_t decode
;
157 unsigned long timestamp
;
159 } nfs_readdir_descriptor_t
;
161 /* Now we cache directories properly, by stuffing the dirent
162 * data directly in the page cache.
164 * Inode invalidation due to refresh etc. takes care of
165 * _everything_, no sloppy entry flushing logic, no extraneous
166 * copying, network direct to page cache, the way it was meant
169 * NOTE: Dirent information verification is done always by the
170 * page-in of the RPC reply, nowhere else, this simplies
171 * things substantially.
174 int nfs_readdir_filler(nfs_readdir_descriptor_t
*desc
, struct page
*page
)
176 struct file
*file
= desc
->file
;
177 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
178 struct rpc_cred
*cred
= nfs_file_cred(file
);
179 unsigned long timestamp
;
182 dfprintk(DIRCACHE
, "NFS: %s: reading cookie %Lu into page %lu\n",
183 __FUNCTION__
, (long long)desc
->entry
->cookie
,
188 error
= NFS_PROTO(inode
)->readdir(file
->f_path
.dentry
, cred
, desc
->entry
->cookie
, page
,
189 NFS_SERVER(inode
)->dtsize
, desc
->plus
);
191 /* We requested READDIRPLUS, but the server doesn't grok it */
192 if (error
== -ENOTSUPP
&& desc
->plus
) {
193 NFS_SERVER(inode
)->caps
&= ~NFS_CAP_READDIRPLUS
;
194 clear_bit(NFS_INO_ADVISE_RDPLUS
, &NFS_FLAGS(inode
));
200 desc
->timestamp
= timestamp
;
201 desc
->timestamp_valid
= 1;
202 SetPageUptodate(page
);
203 spin_lock(&inode
->i_lock
);
204 NFS_I(inode
)->cache_validity
|= NFS_INO_INVALID_ATIME
;
205 spin_unlock(&inode
->i_lock
);
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 (page
->index
== 0 && invalidate_inode_pages2_range(inode
->i_mapping
, PAGE_CACHE_SIZE
, -1) < 0) {
211 /* Should never happen */
212 nfs_zap_mapping(inode
, inode
->i_mapping
);
219 nfs_zap_caches(inode
);
225 int dir_decode(nfs_readdir_descriptor_t
*desc
)
227 __be32
*p
= desc
->ptr
;
228 p
= desc
->decode(p
, desc
->entry
, desc
->plus
);
232 if (desc
->timestamp_valid
)
233 desc
->entry
->fattr
->time_start
= desc
->timestamp
;
235 desc
->entry
->fattr
->valid
&= ~NFS_ATTR_FATTR
;
240 void dir_page_release(nfs_readdir_descriptor_t
*desc
)
243 page_cache_release(desc
->page
);
249 * Given a pointer to a buffer that has already been filled by a call
250 * to readdir, find the next entry with cookie '*desc->dir_cookie'.
252 * If the end of the buffer has been reached, return -EAGAIN, if not,
253 * return the offset within the buffer of the next entry to be
257 int find_dirent(nfs_readdir_descriptor_t
*desc
)
259 struct nfs_entry
*entry
= desc
->entry
;
263 while((status
= dir_decode(desc
)) == 0) {
264 dfprintk(DIRCACHE
, "NFS: %s: examining cookie %Lu\n",
265 __FUNCTION__
, (unsigned long long)entry
->cookie
);
266 if (entry
->prev_cookie
== *desc
->dir_cookie
)
268 if (loop_count
++ > 200) {
277 * Given a pointer to a buffer that has already been filled by a call
278 * to readdir, find the entry at offset 'desc->file->f_pos'.
280 * If the end of the buffer has been reached, return -EAGAIN, if not,
281 * return the offset within the buffer of the next entry to be
285 int find_dirent_index(nfs_readdir_descriptor_t
*desc
)
287 struct nfs_entry
*entry
= desc
->entry
;
292 status
= dir_decode(desc
);
296 dfprintk(DIRCACHE
, "NFS: found cookie %Lu at index %Ld\n",
297 (unsigned long long)entry
->cookie
, desc
->current_index
);
299 if (desc
->file
->f_pos
== desc
->current_index
) {
300 *desc
->dir_cookie
= entry
->cookie
;
303 desc
->current_index
++;
304 if (loop_count
++ > 200) {
313 * Find the given page, and call find_dirent() or find_dirent_index in
314 * order to try to return the next entry.
317 int find_dirent_page(nfs_readdir_descriptor_t
*desc
)
319 struct inode
*inode
= desc
->file
->f_path
.dentry
->d_inode
;
323 dfprintk(DIRCACHE
, "NFS: %s: searching page %ld for target %Lu\n",
324 __FUNCTION__
, desc
->page_index
,
325 (long long) *desc
->dir_cookie
);
327 /* If we find the page in the page_cache, we cannot be sure
328 * how fresh the data is, so we will ignore readdir_plus attributes.
330 desc
->timestamp_valid
= 0;
331 page
= read_cache_page(inode
->i_mapping
, desc
->page_index
,
332 (filler_t
*)nfs_readdir_filler
, desc
);
334 status
= PTR_ERR(page
);
338 /* NOTE: Someone else may have changed the READDIRPLUS flag */
340 desc
->ptr
= kmap(page
); /* matching kunmap in nfs_do_filldir */
341 if (*desc
->dir_cookie
!= 0)
342 status
= find_dirent(desc
);
344 status
= find_dirent_index(desc
);
346 dir_page_release(desc
);
348 dfprintk(DIRCACHE
, "NFS: %s: returns %d\n", __FUNCTION__
, status
);
353 * Recurse through the page cache pages, and return a
354 * filled nfs_entry structure of the next directory entry if possible.
356 * The target for the search is '*desc->dir_cookie' if non-0,
357 * 'desc->file->f_pos' otherwise
360 int readdir_search_pagecache(nfs_readdir_descriptor_t
*desc
)
365 /* Always search-by-index from the beginning of the cache */
366 if (*desc
->dir_cookie
== 0) {
367 dfprintk(DIRCACHE
, "NFS: readdir_search_pagecache() searching for offset %Ld\n",
368 (long long)desc
->file
->f_pos
);
369 desc
->page_index
= 0;
370 desc
->entry
->cookie
= desc
->entry
->prev_cookie
= 0;
371 desc
->entry
->eof
= 0;
372 desc
->current_index
= 0;
374 dfprintk(DIRCACHE
, "NFS: readdir_search_pagecache() searching for cookie %Lu\n",
375 (unsigned long long)*desc
->dir_cookie
);
378 res
= find_dirent_page(desc
);
381 /* Align to beginning of next page */
383 if (loop_count
++ > 200) {
389 dfprintk(DIRCACHE
, "NFS: %s: returns %d\n", __FUNCTION__
, res
);
393 static inline unsigned int dt_type(struct inode
*inode
)
395 return (inode
->i_mode
>> 12) & 15;
398 static struct dentry
*nfs_readdir_lookup(nfs_readdir_descriptor_t
*desc
);
401 * Once we've found the start of the dirent within a page: fill 'er up...
404 int nfs_do_filldir(nfs_readdir_descriptor_t
*desc
, void *dirent
,
407 struct file
*file
= desc
->file
;
408 struct nfs_entry
*entry
= desc
->entry
;
409 struct dentry
*dentry
= NULL
;
410 unsigned long fileid
;
414 dfprintk(DIRCACHE
, "NFS: nfs_do_filldir() filling starting @ cookie %Lu\n",
415 (unsigned long long)entry
->cookie
);
418 unsigned d_type
= DT_UNKNOWN
;
419 /* Note: entry->prev_cookie contains the cookie for
420 * retrieving the current dirent on the server */
421 fileid
= nfs_fileid_to_ino_t(entry
->ino
);
423 /* Get a dentry if we have one */
426 dentry
= nfs_readdir_lookup(desc
);
428 /* Use readdirplus info */
429 if (dentry
!= NULL
&& dentry
->d_inode
!= NULL
) {
430 d_type
= dt_type(dentry
->d_inode
);
431 fileid
= dentry
->d_inode
->i_ino
;
434 res
= filldir(dirent
, entry
->name
, entry
->len
,
435 file
->f_pos
, fileid
, d_type
);
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
;
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 desc
->error
= NFS_PROTO(inode
)->readdir(file
->f_path
.dentry
, cred
, *desc
->dir_cookie
,
491 NFS_SERVER(inode
)->dtsize
,
493 spin_lock(&inode
->i_lock
);
494 NFS_I(inode
)->cache_validity
|= NFS_INO_INVALID_ATIME
;
495 spin_unlock(&inode
->i_lock
);
497 desc
->ptr
= kmap(page
); /* matching kunmap in nfs_do_filldir */
498 if (desc
->error
>= 0) {
499 desc
->timestamp
= timestamp
;
500 desc
->timestamp_valid
= 1;
501 if ((status
= dir_decode(desc
)) == 0)
502 desc
->entry
->prev_cookie
= *desc
->dir_cookie
;
508 status
= nfs_do_filldir(desc
, dirent
, filldir
);
510 /* Reset read descriptor so it searches the page cache from
511 * the start upon the next call to readdir_search_pagecache() */
512 desc
->page_index
= 0;
513 desc
->entry
->cookie
= desc
->entry
->prev_cookie
= 0;
514 desc
->entry
->eof
= 0;
516 dfprintk(DIRCACHE
, "NFS: %s: returns %d\n",
517 __FUNCTION__
, status
);
520 dir_page_release(desc
);
524 /* The file offset position represents the dirent entry number. A
525 last cookie cache takes care of the common case of reading the
528 static int nfs_readdir(struct file
*filp
, void *dirent
, filldir_t filldir
)
530 struct dentry
*dentry
= filp
->f_path
.dentry
;
531 struct inode
*inode
= dentry
->d_inode
;
532 nfs_readdir_descriptor_t my_desc
,
534 struct nfs_entry my_entry
;
536 struct nfs_fattr fattr
;
539 dfprintk(VFS
, "NFS: readdir(%s/%s) starting at cookie %Lu\n",
540 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
,
541 (long long)filp
->f_pos
);
542 nfs_inc_stats(inode
, NFSIOS_VFSGETDENTS
);
546 res
= nfs_revalidate_mapping_nolock(inode
, filp
->f_mapping
);
553 * filp->f_pos points to the dirent entry number.
554 * *desc->dir_cookie has the cookie for the next entry. We have
555 * to either find the entry with the appropriate number or
556 * revalidate the cookie.
558 memset(desc
, 0, sizeof(*desc
));
561 desc
->dir_cookie
= &((struct nfs_open_context
*)filp
->private_data
)->dir_cookie
;
562 desc
->decode
= NFS_PROTO(inode
)->decode_dirent
;
563 desc
->plus
= NFS_USE_READDIRPLUS(inode
);
565 my_entry
.cookie
= my_entry
.prev_cookie
= 0;
568 my_entry
.fattr
= &fattr
;
569 nfs_fattr_init(&fattr
);
570 desc
->entry
= &my_entry
;
572 while(!desc
->entry
->eof
) {
573 res
= readdir_search_pagecache(desc
);
575 if (res
== -EBADCOOKIE
) {
576 /* This means either end of directory */
577 if (*desc
->dir_cookie
&& desc
->entry
->cookie
!= *desc
->dir_cookie
) {
578 /* Or that the server has 'lost' a cookie */
579 res
= uncached_readdir(desc
, dirent
, filldir
);
586 if (res
== -ETOOSMALL
&& desc
->plus
) {
587 clear_bit(NFS_INO_ADVISE_RDPLUS
, &NFS_FLAGS(inode
));
588 nfs_zap_caches(inode
);
590 desc
->entry
->eof
= 0;
596 res
= nfs_do_filldir(desc
, dirent
, filldir
);
605 dfprintk(VFS
, "NFS: readdir(%s/%s) returns %ld\n",
606 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
,
611 loff_t
nfs_llseek_dir(struct file
*filp
, loff_t offset
, int origin
)
613 mutex_lock(&filp
->f_path
.dentry
->d_inode
->i_mutex
);
616 offset
+= filp
->f_pos
;
624 if (offset
!= filp
->f_pos
) {
625 filp
->f_pos
= offset
;
626 ((struct nfs_open_context
*)filp
->private_data
)->dir_cookie
= 0;
629 mutex_unlock(&filp
->f_path
.dentry
->d_inode
->i_mutex
);
634 * All directory operations under NFS are synchronous, so fsync()
635 * is a dummy operation.
637 int nfs_fsync_dir(struct file
*filp
, struct dentry
*dentry
, int datasync
)
639 dfprintk(VFS
, "NFS: fsync_dir(%s/%s) datasync %d\n",
640 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
,
647 * A check for whether or not the parent directory has changed.
648 * In the case it has, we assume that the dentries are untrustworthy
649 * and may need to be looked up again.
651 static int nfs_check_verifier(struct inode
*dir
, struct dentry
*dentry
)
655 if ((NFS_I(dir
)->cache_validity
& NFS_INO_INVALID_ATTR
) != 0
656 || nfs_attribute_timeout(dir
))
658 return nfs_verify_change_attribute(dir
, (unsigned long)dentry
->d_fsdata
);
661 static inline void nfs_set_verifier(struct dentry
* dentry
, unsigned long verf
)
663 dentry
->d_fsdata
= (void *)verf
;
666 static void nfs_refresh_verifier(struct dentry
* dentry
, unsigned long verf
)
668 if (time_after(verf
, (unsigned long)dentry
->d_fsdata
))
669 nfs_set_verifier(dentry
, verf
);
673 * Whenever an NFS operation succeeds, we know that the dentry
674 * is valid, so we update the revalidation timestamp.
676 static inline void nfs_renew_times(struct dentry
* dentry
)
678 dentry
->d_time
= jiffies
;
682 * Return the intent data that applies to this particular path component
684 * Note that the current set of intents only apply to the very last
685 * component of the path.
686 * We check for this using LOOKUP_CONTINUE and LOOKUP_PARENT.
688 static inline unsigned int nfs_lookup_check_intent(struct nameidata
*nd
, unsigned int mask
)
690 if (nd
->flags
& (LOOKUP_CONTINUE
|LOOKUP_PARENT
))
692 return nd
->flags
& mask
;
696 * Inode and filehandle revalidation for lookups.
698 * We force revalidation in the cases where the VFS sets LOOKUP_REVAL,
699 * or if the intent information indicates that we're about to open this
700 * particular file and the "nocto" mount flag is not set.
704 int nfs_lookup_verify_inode(struct inode
*inode
, struct nameidata
*nd
)
706 struct nfs_server
*server
= NFS_SERVER(inode
);
709 /* VFS wants an on-the-wire revalidation */
710 if (nd
->flags
& LOOKUP_REVAL
)
712 /* This is an open(2) */
713 if (nfs_lookup_check_intent(nd
, LOOKUP_OPEN
) != 0 &&
714 !(server
->flags
& NFS_MOUNT_NOCTO
) &&
715 (S_ISREG(inode
->i_mode
) ||
716 S_ISDIR(inode
->i_mode
)))
719 return nfs_revalidate_inode(server
, inode
);
721 return __nfs_revalidate_inode(server
, inode
);
725 * We judge how long we want to trust negative
726 * dentries by looking at the parent inode mtime.
728 * If parent mtime has changed, we revalidate, else we wait for a
729 * period corresponding to the parent's attribute cache timeout value.
732 int nfs_neg_need_reval(struct inode
*dir
, struct dentry
*dentry
,
733 struct nameidata
*nd
)
735 /* Don't revalidate a negative dentry if we're creating a new file */
736 if (nd
!= NULL
&& nfs_lookup_check_intent(nd
, LOOKUP_CREATE
) != 0)
738 return !nfs_check_verifier(dir
, dentry
);
742 * This is called every time the dcache has a lookup hit,
743 * and we should check whether we can really trust that
746 * NOTE! The hit can be a negative hit too, don't assume
749 * If the parent directory is seen to have changed, we throw out the
750 * cached dentry and do a new lookup.
752 static int nfs_lookup_revalidate(struct dentry
* dentry
, struct nameidata
*nd
)
756 struct dentry
*parent
;
758 struct nfs_fh fhandle
;
759 struct nfs_fattr fattr
;
760 unsigned long verifier
;
762 parent
= dget_parent(dentry
);
764 dir
= parent
->d_inode
;
765 nfs_inc_stats(dir
, NFSIOS_DENTRYREVALIDATE
);
766 inode
= dentry
->d_inode
;
769 if (nfs_neg_need_reval(dir
, dentry
, nd
))
774 if (is_bad_inode(inode
)) {
775 dfprintk(LOOKUPCACHE
, "%s: %s/%s has dud inode\n",
776 __FUNCTION__
, dentry
->d_parent
->d_name
.name
,
777 dentry
->d_name
.name
);
781 /* Revalidate parent directory attribute cache */
782 if (nfs_revalidate_inode(NFS_SERVER(dir
), dir
) < 0)
785 /* Force a full look up iff the parent directory has changed */
786 if (nfs_check_verifier(dir
, dentry
)) {
787 if (nfs_lookup_verify_inode(inode
, nd
))
792 if (NFS_STALE(inode
))
795 verifier
= nfs_save_change_attribute(dir
);
796 error
= NFS_PROTO(dir
)->lookup(dir
, &dentry
->d_name
, &fhandle
, &fattr
);
799 if (nfs_compare_fh(NFS_FH(inode
), &fhandle
))
801 if ((error
= nfs_refresh_inode(inode
, &fattr
)) != 0)
804 nfs_renew_times(dentry
);
805 nfs_refresh_verifier(dentry
, verifier
);
809 dfprintk(LOOKUPCACHE
, "NFS: %s(%s/%s) is valid\n",
810 __FUNCTION__
, dentry
->d_parent
->d_name
.name
,
811 dentry
->d_name
.name
);
817 if (inode
&& S_ISDIR(inode
->i_mode
)) {
818 /* Purge readdir caches. */
819 nfs_zap_caches(inode
);
820 /* If we have submounts, don't unhash ! */
821 if (have_submounts(dentry
))
823 shrink_dcache_parent(dentry
);
828 dfprintk(LOOKUPCACHE
, "NFS: %s(%s/%s) is invalid\n",
829 __FUNCTION__
, dentry
->d_parent
->d_name
.name
,
830 dentry
->d_name
.name
);
835 * This is called from dput() when d_count is going to 0.
837 static int nfs_dentry_delete(struct dentry
*dentry
)
839 dfprintk(VFS
, "NFS: dentry_delete(%s/%s, %x)\n",
840 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
,
843 if (dentry
->d_flags
& DCACHE_NFSFS_RENAMED
) {
844 /* Unhash it, so that ->d_iput() would be called */
847 if (!(dentry
->d_sb
->s_flags
& MS_ACTIVE
)) {
848 /* Unhash it, so that ancestors of killed async unlink
849 * files will be cleaned up during umount */
857 * Called when the dentry loses inode.
858 * We use it to clean up silly-renamed files.
860 static void nfs_dentry_iput(struct dentry
*dentry
, struct inode
*inode
)
862 nfs_inode_return_delegation(inode
);
863 if (S_ISDIR(inode
->i_mode
))
864 /* drop any readdir cache as it could easily be old */
865 NFS_I(inode
)->cache_validity
|= NFS_INO_INVALID_DATA
;
867 if (dentry
->d_flags
& DCACHE_NFSFS_RENAMED
) {
870 nfs_complete_unlink(dentry
);
873 /* When creating a negative dentry, we want to renew d_time */
874 nfs_renew_times(dentry
);
878 struct dentry_operations nfs_dentry_operations
= {
879 .d_revalidate
= nfs_lookup_revalidate
,
880 .d_delete
= nfs_dentry_delete
,
881 .d_iput
= nfs_dentry_iput
,
885 * Use intent information to check whether or not we're going to do
886 * an O_EXCL create using this path component.
889 int nfs_is_exclusive_create(struct inode
*dir
, struct nameidata
*nd
)
891 if (NFS_PROTO(dir
)->version
== 2)
893 if (nd
== NULL
|| nfs_lookup_check_intent(nd
, LOOKUP_CREATE
) == 0)
895 return (nd
->intent
.open
.flags
& O_EXCL
) != 0;
898 static inline int nfs_reval_fsid(struct vfsmount
*mnt
, struct inode
*dir
,
899 struct nfs_fh
*fh
, struct nfs_fattr
*fattr
)
901 struct nfs_server
*server
= NFS_SERVER(dir
);
903 if (!nfs_fsid_equal(&server
->fsid
, &fattr
->fsid
))
904 /* Revalidate fsid on root dir */
905 return __nfs_revalidate_inode(server
, mnt
->mnt_root
->d_inode
);
909 static struct dentry
*nfs_lookup(struct inode
*dir
, struct dentry
* dentry
, struct nameidata
*nd
)
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
;
931 * If we're doing an exclusive create, optimize away the lookup
932 * but don't hash the dentry.
934 if (nfs_is_exclusive_create(dir
, nd
)) {
935 d_instantiate(dentry
, NULL
);
940 error
= NFS_PROTO(dir
)->lookup(dir
, &dentry
->d_name
, &fhandle
, &fattr
);
941 if (error
== -ENOENT
)
944 res
= ERR_PTR(error
);
947 error
= nfs_reval_fsid(nd
->mnt
, dir
, &fhandle
, &fattr
);
949 res
= ERR_PTR(error
);
952 inode
= nfs_fhget(dentry
->d_sb
, &fhandle
, &fattr
);
953 res
= (struct dentry
*)inode
;
958 res
= d_materialise_unique(dentry
, inode
);
960 struct dentry
*parent
;
963 /* Was a directory renamed! */
964 parent
= dget_parent(res
);
965 if (!IS_ROOT(parent
))
966 nfs_mark_for_revalidate(parent
->d_inode
);
970 nfs_renew_times(dentry
);
971 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
979 static int nfs_open_revalidate(struct dentry
*, struct nameidata
*);
981 struct dentry_operations nfs4_dentry_operations
= {
982 .d_revalidate
= nfs_open_revalidate
,
983 .d_delete
= nfs_dentry_delete
,
984 .d_iput
= nfs_dentry_iput
,
988 * Use intent information to determine whether we need to substitute
989 * the NFSv4-style stateful OPEN for the LOOKUP call
991 static int is_atomic_open(struct inode
*dir
, struct nameidata
*nd
)
993 if (nd
== NULL
|| nfs_lookup_check_intent(nd
, LOOKUP_OPEN
) == 0)
995 /* NFS does not (yet) have a stateful open for directories */
996 if (nd
->flags
& LOOKUP_DIRECTORY
)
998 /* Are we trying to write to a read only partition? */
999 if (IS_RDONLY(dir
) && (nd
->intent
.open
.flags
& (O_CREAT
|O_TRUNC
|FMODE_WRITE
)))
1004 static struct dentry
*nfs_atomic_lookup(struct inode
*dir
, struct dentry
*dentry
, struct nameidata
*nd
)
1006 struct dentry
*res
= NULL
;
1009 dfprintk(VFS
, "NFS: atomic_lookup(%s/%ld), %s\n",
1010 dir
->i_sb
->s_id
, dir
->i_ino
, dentry
->d_name
.name
);
1012 /* Check that we are indeed trying to open this file */
1013 if (!is_atomic_open(dir
, nd
))
1016 if (dentry
->d_name
.len
> NFS_SERVER(dir
)->namelen
) {
1017 res
= ERR_PTR(-ENAMETOOLONG
);
1020 dentry
->d_op
= NFS_PROTO(dir
)->dentry_ops
;
1022 /* Let vfs_create() deal with O_EXCL */
1023 if (nd
->intent
.open
.flags
& O_EXCL
) {
1024 d_add(dentry
, NULL
);
1028 /* Open the file on the server */
1030 /* Revalidate parent directory attribute cache */
1031 error
= nfs_revalidate_inode(NFS_SERVER(dir
), dir
);
1033 res
= ERR_PTR(error
);
1038 if (nd
->intent
.open
.flags
& O_CREAT
) {
1039 nfs_begin_data_update(dir
);
1040 res
= nfs4_atomic_open(dir
, dentry
, nd
);
1041 nfs_end_data_update(dir
);
1043 res
= nfs4_atomic_open(dir
, dentry
, nd
);
1046 error
= PTR_ERR(res
);
1048 /* Make a negative dentry */
1052 /* This turned out not to be a regular file */
1057 if (!(nd
->intent
.open
.flags
& O_NOFOLLOW
))
1063 } else if (res
!= NULL
)
1065 nfs_renew_times(dentry
);
1066 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1070 return nfs_lookup(dir
, dentry
, nd
);
1073 static int nfs_open_revalidate(struct dentry
*dentry
, struct nameidata
*nd
)
1075 struct dentry
*parent
= NULL
;
1076 struct inode
*inode
= dentry
->d_inode
;
1078 unsigned long verifier
;
1079 int openflags
, ret
= 0;
1081 parent
= dget_parent(dentry
);
1082 dir
= parent
->d_inode
;
1083 if (!is_atomic_open(dir
, nd
))
1085 /* We can't create new files in nfs_open_revalidate(), so we
1086 * optimize away revalidation of negative dentries.
1090 /* NFS only supports OPEN on regular files */
1091 if (!S_ISREG(inode
->i_mode
))
1093 openflags
= nd
->intent
.open
.flags
;
1094 /* We cannot do exclusive creation on a positive dentry */
1095 if ((openflags
& (O_CREAT
|O_EXCL
)) == (O_CREAT
|O_EXCL
))
1097 /* We can't create new files, or truncate existing ones here */
1098 openflags
&= ~(O_CREAT
|O_TRUNC
);
1101 * Note: we're not holding inode->i_mutex and so may be racing with
1102 * operations that change the directory. We therefore save the
1103 * change attribute *before* we do the RPC call.
1106 verifier
= nfs_save_change_attribute(dir
);
1107 ret
= nfs4_open_revalidate(dir
, dentry
, openflags
, nd
);
1109 nfs_refresh_verifier(dentry
, verifier
);
1118 if (inode
!= NULL
&& nfs_have_delegation(inode
, FMODE_READ
))
1120 return nfs_lookup_revalidate(dentry
, nd
);
1122 #endif /* CONFIG_NFSV4 */
1124 static struct dentry
*nfs_readdir_lookup(nfs_readdir_descriptor_t
*desc
)
1126 struct dentry
*parent
= desc
->file
->f_path
.dentry
;
1127 struct inode
*dir
= parent
->d_inode
;
1128 struct nfs_entry
*entry
= desc
->entry
;
1129 struct dentry
*dentry
, *alias
;
1130 struct qstr name
= {
1131 .name
= entry
->name
,
1134 struct inode
*inode
;
1138 if (name
.name
[0] == '.' && name
.name
[1] == '.')
1139 return dget_parent(parent
);
1142 if (name
.name
[0] == '.')
1143 return dget(parent
);
1145 name
.hash
= full_name_hash(name
.name
, name
.len
);
1146 dentry
= d_lookup(parent
, &name
);
1147 if (dentry
!= NULL
) {
1148 /* Is this a positive dentry that matches the readdir info? */
1149 if (dentry
->d_inode
!= NULL
&&
1150 (NFS_FILEID(dentry
->d_inode
) == entry
->ino
||
1151 d_mountpoint(dentry
))) {
1152 if (!desc
->plus
|| entry
->fh
->size
== 0)
1154 if (nfs_compare_fh(NFS_FH(dentry
->d_inode
),
1158 /* No, so d_drop to allow one to be created */
1162 if (!desc
->plus
|| !(entry
->fattr
->valid
& NFS_ATTR_FATTR
))
1164 /* Note: caller is already holding the dir->i_mutex! */
1165 dentry
= d_alloc(parent
, &name
);
1168 dentry
->d_op
= NFS_PROTO(dir
)->dentry_ops
;
1169 inode
= nfs_fhget(dentry
->d_sb
, entry
->fh
, entry
->fattr
);
1170 if (IS_ERR(inode
)) {
1175 alias
= d_materialise_unique(dentry
, inode
);
1176 if (alias
!= NULL
) {
1183 nfs_renew_times(dentry
);
1184 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1187 nfs_renew_times(dentry
);
1188 nfs_refresh_verifier(dentry
, nfs_save_change_attribute(dir
));
1193 * Code common to create, mkdir, and mknod.
1195 int nfs_instantiate(struct dentry
*dentry
, struct nfs_fh
*fhandle
,
1196 struct nfs_fattr
*fattr
)
1198 struct inode
*inode
;
1199 int error
= -EACCES
;
1201 /* We may have been initialized further down */
1202 if (dentry
->d_inode
)
1204 if (fhandle
->size
== 0) {
1205 struct inode
*dir
= dentry
->d_parent
->d_inode
;
1206 error
= NFS_PROTO(dir
)->lookup(dir
, &dentry
->d_name
, fhandle
, fattr
);
1210 if (!(fattr
->valid
& NFS_ATTR_FATTR
)) {
1211 struct nfs_server
*server
= NFS_SB(dentry
->d_sb
);
1212 error
= server
->nfs_client
->rpc_ops
->getattr(server
, fhandle
, fattr
);
1216 inode
= nfs_fhget(dentry
->d_sb
, fhandle
, fattr
);
1217 error
= PTR_ERR(inode
);
1220 d_instantiate(dentry
, inode
);
1221 if (d_unhashed(dentry
))
1227 * Following a failed create operation, we drop the dentry rather
1228 * than retain a negative dentry. This avoids a problem in the event
1229 * that the operation succeeded on the server, but an error in the
1230 * reply path made it appear to have failed.
1232 static int nfs_create(struct inode
*dir
, struct dentry
*dentry
, int mode
,
1233 struct nameidata
*nd
)
1239 dfprintk(VFS
, "NFS: create(%s/%ld), %s\n",
1240 dir
->i_sb
->s_id
, dir
->i_ino
, dentry
->d_name
.name
);
1242 attr
.ia_mode
= mode
;
1243 attr
.ia_valid
= ATTR_MODE
;
1245 if (nd
&& (nd
->flags
& LOOKUP_CREATE
))
1246 open_flags
= nd
->intent
.open
.flags
;
1249 nfs_begin_data_update(dir
);
1250 error
= NFS_PROTO(dir
)->create(dir
, dentry
, &attr
, open_flags
, nd
);
1251 nfs_end_data_update(dir
);
1254 nfs_renew_times(dentry
);
1255 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1265 * See comments for nfs_proc_create regarding failed operations.
1268 nfs_mknod(struct inode
*dir
, struct dentry
*dentry
, int mode
, dev_t rdev
)
1273 dfprintk(VFS
, "NFS: mknod(%s/%ld), %s\n",
1274 dir
->i_sb
->s_id
, dir
->i_ino
, dentry
->d_name
.name
);
1276 if (!new_valid_dev(rdev
))
1279 attr
.ia_mode
= mode
;
1280 attr
.ia_valid
= ATTR_MODE
;
1283 nfs_begin_data_update(dir
);
1284 status
= NFS_PROTO(dir
)->mknod(dir
, dentry
, &attr
, rdev
);
1285 nfs_end_data_update(dir
);
1288 nfs_renew_times(dentry
);
1289 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1299 * See comments for nfs_proc_create regarding failed operations.
1301 static int nfs_mkdir(struct inode
*dir
, struct dentry
*dentry
, int mode
)
1306 dfprintk(VFS
, "NFS: mkdir(%s/%ld), %s\n",
1307 dir
->i_sb
->s_id
, dir
->i_ino
, dentry
->d_name
.name
);
1309 attr
.ia_valid
= ATTR_MODE
;
1310 attr
.ia_mode
= mode
| S_IFDIR
;
1313 nfs_begin_data_update(dir
);
1314 error
= NFS_PROTO(dir
)->mkdir(dir
, dentry
, &attr
);
1315 nfs_end_data_update(dir
);
1318 nfs_renew_times(dentry
);
1319 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1328 static int nfs_rmdir(struct inode
*dir
, struct dentry
*dentry
)
1332 dfprintk(VFS
, "NFS: rmdir(%s/%ld), %s\n",
1333 dir
->i_sb
->s_id
, dir
->i_ino
, dentry
->d_name
.name
);
1336 nfs_begin_data_update(dir
);
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 nfs_end_data_update(dir
);
1347 static int nfs_sillyrename(struct inode
*dir
, struct dentry
*dentry
)
1349 static unsigned int sillycounter
;
1350 const int i_inosize
= sizeof(dir
->i_ino
)*2;
1351 const int countersize
= sizeof(sillycounter
)*2;
1352 const int slen
= sizeof(".nfs") + i_inosize
+ 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 if (!dentry
->d_inode
)
1365 printk("NFS: silly-renaming %s/%s, negative dentry??\n",
1366 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
);
1369 * We don't allow a dentry to be silly-renamed twice.
1372 if (dentry
->d_flags
& DCACHE_NFSFS_RENAMED
)
1375 sprintf(silly
, ".nfs%*.*lx",
1376 i_inosize
, i_inosize
, dentry
->d_inode
->i_ino
);
1378 /* Return delegation in anticipation of the rename */
1379 nfs_inode_return_delegation(dentry
->d_inode
);
1383 char *suffix
= silly
+ slen
- countersize
;
1387 sprintf(suffix
, "%*.*x", countersize
, countersize
, sillycounter
);
1389 dfprintk(VFS
, "NFS: trying to rename %s to %s\n",
1390 dentry
->d_name
.name
, silly
);
1392 sdentry
= lookup_one_len(silly
, dentry
->d_parent
, slen
);
1394 * N.B. Better to return EBUSY here ... it could be
1395 * dangerous to delete the file while it's in use.
1397 if (IS_ERR(sdentry
))
1399 } while(sdentry
->d_inode
!= NULL
); /* need negative lookup */
1401 qsilly
.name
= silly
;
1402 qsilly
.len
= strlen(silly
);
1403 nfs_begin_data_update(dir
);
1404 if (dentry
->d_inode
) {
1405 nfs_begin_data_update(dentry
->d_inode
);
1406 error
= NFS_PROTO(dir
)->rename(dir
, &dentry
->d_name
,
1408 nfs_mark_for_revalidate(dentry
->d_inode
);
1409 nfs_end_data_update(dentry
->d_inode
);
1411 error
= NFS_PROTO(dir
)->rename(dir
, &dentry
->d_name
,
1413 nfs_end_data_update(dir
);
1415 nfs_renew_times(dentry
);
1416 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1417 d_move(dentry
, sdentry
);
1418 error
= nfs_async_unlink(dentry
);
1419 /* If we return 0 we don't unlink */
1427 * Remove a file after making sure there are no pending writes,
1428 * and after checking that the file has only one user.
1430 * We invalidate the attribute cache and free the inode prior to the operation
1431 * to avoid possible races if the server reuses the inode.
1433 static int nfs_safe_remove(struct dentry
*dentry
)
1435 struct inode
*dir
= dentry
->d_parent
->d_inode
;
1436 struct inode
*inode
= dentry
->d_inode
;
1439 dfprintk(VFS
, "NFS: safe_remove(%s/%s)\n",
1440 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
);
1442 /* If the dentry was sillyrenamed, we simply call d_delete() */
1443 if (dentry
->d_flags
& DCACHE_NFSFS_RENAMED
) {
1448 nfs_begin_data_update(dir
);
1449 if (inode
!= NULL
) {
1450 nfs_inode_return_delegation(inode
);
1451 nfs_begin_data_update(inode
);
1452 error
= NFS_PROTO(dir
)->remove(dir
, &dentry
->d_name
);
1453 /* The VFS may want to delete this inode */
1456 nfs_mark_for_revalidate(inode
);
1457 nfs_end_data_update(inode
);
1459 error
= NFS_PROTO(dir
)->remove(dir
, &dentry
->d_name
);
1460 nfs_end_data_update(dir
);
1465 /* We do silly rename. In case sillyrename() returns -EBUSY, the inode
1466 * belongs to an active ".nfs..." file and we return -EBUSY.
1468 * If sillyrename() returns 0, we do nothing, otherwise we unlink.
1470 static int nfs_unlink(struct inode
*dir
, struct dentry
*dentry
)
1473 int need_rehash
= 0;
1475 dfprintk(VFS
, "NFS: unlink(%s/%ld, %s)\n", dir
->i_sb
->s_id
,
1476 dir
->i_ino
, dentry
->d_name
.name
);
1479 spin_lock(&dcache_lock
);
1480 spin_lock(&dentry
->d_lock
);
1481 if (atomic_read(&dentry
->d_count
) > 1) {
1482 spin_unlock(&dentry
->d_lock
);
1483 spin_unlock(&dcache_lock
);
1484 /* Start asynchronous writeout of the inode */
1485 write_inode_now(dentry
->d_inode
, 0);
1486 error
= nfs_sillyrename(dir
, dentry
);
1490 if (!d_unhashed(dentry
)) {
1494 spin_unlock(&dentry
->d_lock
);
1495 spin_unlock(&dcache_lock
);
1496 error
= nfs_safe_remove(dentry
);
1498 nfs_renew_times(dentry
);
1499 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1500 } else if (need_rehash
)
1507 * To create a symbolic link, most file systems instantiate a new inode,
1508 * add a page to it containing the path, then write it out to the disk
1509 * using prepare_write/commit_write.
1511 * Unfortunately the NFS client can't create the in-core inode first
1512 * because it needs a file handle to create an in-core inode (see
1513 * fs/nfs/inode.c:nfs_fhget). We only have a file handle *after* the
1514 * symlink request has completed on the server.
1516 * So instead we allocate a raw page, copy the symname into it, then do
1517 * the SYMLINK request with the page as the buffer. If it succeeds, we
1518 * now have a new file handle and can instantiate an in-core NFS inode
1519 * and move the raw page into its mapping.
1521 static int nfs_symlink(struct inode
*dir
, struct dentry
*dentry
, const char *symname
)
1523 struct pagevec lru_pvec
;
1527 unsigned int pathlen
= strlen(symname
);
1530 dfprintk(VFS
, "NFS: symlink(%s/%ld, %s, %s)\n", dir
->i_sb
->s_id
,
1531 dir
->i_ino
, dentry
->d_name
.name
, symname
);
1533 if (pathlen
> PAGE_SIZE
)
1534 return -ENAMETOOLONG
;
1536 attr
.ia_mode
= S_IFLNK
| S_IRWXUGO
;
1537 attr
.ia_valid
= ATTR_MODE
;
1541 page
= alloc_page(GFP_KERNEL
);
1547 kaddr
= kmap_atomic(page
, KM_USER0
);
1548 memcpy(kaddr
, symname
, pathlen
);
1549 if (pathlen
< PAGE_SIZE
)
1550 memset(kaddr
+ pathlen
, 0, PAGE_SIZE
- pathlen
);
1551 kunmap_atomic(kaddr
, KM_USER0
);
1553 nfs_begin_data_update(dir
);
1554 error
= NFS_PROTO(dir
)->symlink(dir
, dentry
, page
, pathlen
, &attr
);
1555 nfs_end_data_update(dir
);
1557 dfprintk(VFS
, "NFS: symlink(%s/%ld, %s, %s) error %d\n",
1558 dir
->i_sb
->s_id
, dir
->i_ino
,
1559 dentry
->d_name
.name
, symname
, error
);
1567 * No big deal if we can't add this page to the page cache here.
1568 * READLINK will get the missing page from the server if needed.
1570 pagevec_init(&lru_pvec
, 0);
1571 if (!add_to_page_cache(page
, dentry
->d_inode
->i_mapping
, 0,
1573 pagevec_add(&lru_pvec
, page
);
1574 pagevec_lru_add(&lru_pvec
);
1575 SetPageUptodate(page
);
1585 nfs_link(struct dentry
*old_dentry
, struct inode
*dir
, struct dentry
*dentry
)
1587 struct inode
*inode
= old_dentry
->d_inode
;
1590 dfprintk(VFS
, "NFS: link(%s/%s -> %s/%s)\n",
1591 old_dentry
->d_parent
->d_name
.name
, old_dentry
->d_name
.name
,
1592 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
);
1595 nfs_begin_data_update(dir
);
1596 nfs_begin_data_update(inode
);
1597 error
= NFS_PROTO(dir
)->link(inode
, dir
, &dentry
->d_name
);
1599 atomic_inc(&inode
->i_count
);
1600 d_instantiate(dentry
, inode
);
1602 nfs_end_data_update(inode
);
1603 nfs_end_data_update(dir
);
1610 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a
1611 * different file handle for the same inode after a rename (e.g. when
1612 * moving to a different directory). A fail-safe method to do so would
1613 * be to look up old_dir/old_name, create a link to new_dir/new_name and
1614 * rename the old file using the sillyrename stuff. This way, the original
1615 * file in old_dir will go away when the last process iput()s the inode.
1619 * It actually works quite well. One needs to have the possibility for
1620 * at least one ".nfs..." file in each directory the file ever gets
1621 * moved or linked to which happens automagically with the new
1622 * implementation that only depends on the dcache stuff instead of
1623 * using the inode layer
1625 * Unfortunately, things are a little more complicated than indicated
1626 * above. For a cross-directory move, we want to make sure we can get
1627 * rid of the old inode after the operation. This means there must be
1628 * no pending writes (if it's a file), and the use count must be 1.
1629 * If these conditions are met, we can drop the dentries before doing
1632 static int nfs_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
1633 struct inode
*new_dir
, struct dentry
*new_dentry
)
1635 struct inode
*old_inode
= old_dentry
->d_inode
;
1636 struct inode
*new_inode
= new_dentry
->d_inode
;
1637 struct dentry
*dentry
= NULL
, *rehash
= NULL
;
1641 * To prevent any new references to the target during the rename,
1642 * we unhash the dentry and free the inode in advance.
1645 if (!d_unhashed(new_dentry
)) {
1647 rehash
= new_dentry
;
1650 dfprintk(VFS
, "NFS: rename(%s/%s -> %s/%s, ct=%d)\n",
1651 old_dentry
->d_parent
->d_name
.name
, old_dentry
->d_name
.name
,
1652 new_dentry
->d_parent
->d_name
.name
, new_dentry
->d_name
.name
,
1653 atomic_read(&new_dentry
->d_count
));
1656 * First check whether the target is busy ... we can't
1657 * safely do _any_ rename if the target is in use.
1659 * For files, make a copy of the dentry and then do a
1660 * silly-rename. If the silly-rename succeeds, the
1661 * copied dentry is hashed and becomes the new target.
1665 if (S_ISDIR(new_inode
->i_mode
)) {
1667 if (!S_ISDIR(old_inode
->i_mode
))
1669 } else if (atomic_read(&new_dentry
->d_count
) > 2) {
1671 /* copy the target dentry's name */
1672 dentry
= d_alloc(new_dentry
->d_parent
,
1673 &new_dentry
->d_name
);
1677 /* silly-rename the existing target ... */
1678 err
= nfs_sillyrename(new_dir
, new_dentry
);
1680 new_dentry
= rehash
= dentry
;
1682 /* instantiate the replacement target */
1683 d_instantiate(new_dentry
, NULL
);
1684 } else if (atomic_read(&new_dentry
->d_count
) > 1) {
1685 /* dentry still busy? */
1687 printk("nfs_rename: target %s/%s busy, d_count=%d\n",
1688 new_dentry
->d_parent
->d_name
.name
,
1689 new_dentry
->d_name
.name
,
1690 atomic_read(&new_dentry
->d_count
));
1695 drop_nlink(new_inode
);
1699 * ... prune child dentries and writebacks if needed.
1701 if (atomic_read(&old_dentry
->d_count
) > 1) {
1702 if (S_ISREG(old_inode
->i_mode
))
1703 nfs_wb_all(old_inode
);
1704 shrink_dcache_parent(old_dentry
);
1706 nfs_inode_return_delegation(old_inode
);
1708 if (new_inode
!= NULL
) {
1709 nfs_inode_return_delegation(new_inode
);
1710 d_delete(new_dentry
);
1713 nfs_begin_data_update(old_dir
);
1714 nfs_begin_data_update(new_dir
);
1715 nfs_begin_data_update(old_inode
);
1716 error
= NFS_PROTO(old_dir
)->rename(old_dir
, &old_dentry
->d_name
,
1717 new_dir
, &new_dentry
->d_name
);
1718 nfs_mark_for_revalidate(old_inode
);
1719 nfs_end_data_update(old_inode
);
1720 nfs_end_data_update(new_dir
);
1721 nfs_end_data_update(old_dir
);
1726 d_move(old_dentry
, new_dentry
);
1727 nfs_renew_times(new_dentry
);
1728 nfs_refresh_verifier(new_dentry
, nfs_save_change_attribute(new_dir
));
1731 /* new dentry created? */
1738 static DEFINE_SPINLOCK(nfs_access_lru_lock
);
1739 static LIST_HEAD(nfs_access_lru_list
);
1740 static atomic_long_t nfs_access_nr_entries
;
1742 static void nfs_access_free_entry(struct nfs_access_entry
*entry
)
1744 put_rpccred(entry
->cred
);
1746 smp_mb__before_atomic_dec();
1747 atomic_long_dec(&nfs_access_nr_entries
);
1748 smp_mb__after_atomic_dec();
1751 int nfs_access_cache_shrinker(int nr_to_scan
, gfp_t gfp_mask
)
1754 struct nfs_inode
*nfsi
;
1755 struct nfs_access_entry
*cache
;
1757 spin_lock(&nfs_access_lru_lock
);
1759 list_for_each_entry(nfsi
, &nfs_access_lru_list
, access_cache_inode_lru
) {
1760 struct inode
*inode
;
1762 if (nr_to_scan
-- == 0)
1764 inode
= igrab(&nfsi
->vfs_inode
);
1767 spin_lock(&inode
->i_lock
);
1768 if (list_empty(&nfsi
->access_cache_entry_lru
))
1769 goto remove_lru_entry
;
1770 cache
= list_entry(nfsi
->access_cache_entry_lru
.next
,
1771 struct nfs_access_entry
, lru
);
1772 list_move(&cache
->lru
, &head
);
1773 rb_erase(&cache
->rb_node
, &nfsi
->access_cache
);
1774 if (!list_empty(&nfsi
->access_cache_entry_lru
))
1775 list_move_tail(&nfsi
->access_cache_inode_lru
,
1776 &nfs_access_lru_list
);
1779 list_del_init(&nfsi
->access_cache_inode_lru
);
1780 clear_bit(NFS_INO_ACL_LRU_SET
, &nfsi
->flags
);
1782 spin_unlock(&inode
->i_lock
);
1786 spin_unlock(&nfs_access_lru_lock
);
1787 while (!list_empty(&head
)) {
1788 cache
= list_entry(head
.next
, struct nfs_access_entry
, lru
);
1789 list_del(&cache
->lru
);
1790 nfs_access_free_entry(cache
);
1792 return (atomic_long_read(&nfs_access_nr_entries
) / 100) * sysctl_vfs_cache_pressure
;
1795 static void __nfs_access_zap_cache(struct inode
*inode
)
1797 struct nfs_inode
*nfsi
= NFS_I(inode
);
1798 struct rb_root
*root_node
= &nfsi
->access_cache
;
1799 struct rb_node
*n
, *dispose
= NULL
;
1800 struct nfs_access_entry
*entry
;
1802 /* Unhook entries from the cache */
1803 while ((n
= rb_first(root_node
)) != NULL
) {
1804 entry
= rb_entry(n
, struct nfs_access_entry
, rb_node
);
1805 rb_erase(n
, root_node
);
1806 list_del(&entry
->lru
);
1807 n
->rb_left
= dispose
;
1810 nfsi
->cache_validity
&= ~NFS_INO_INVALID_ACCESS
;
1811 spin_unlock(&inode
->i_lock
);
1813 /* Now kill them all! */
1814 while (dispose
!= NULL
) {
1816 dispose
= n
->rb_left
;
1817 nfs_access_free_entry(rb_entry(n
, struct nfs_access_entry
, rb_node
));
1821 void nfs_access_zap_cache(struct inode
*inode
)
1823 /* Remove from global LRU init */
1824 if (test_and_clear_bit(NFS_INO_ACL_LRU_SET
, &NFS_FLAGS(inode
))) {
1825 spin_lock(&nfs_access_lru_lock
);
1826 list_del_init(&NFS_I(inode
)->access_cache_inode_lru
);
1827 spin_unlock(&nfs_access_lru_lock
);
1830 spin_lock(&inode
->i_lock
);
1831 /* This will release the spinlock */
1832 __nfs_access_zap_cache(inode
);
1835 static struct nfs_access_entry
*nfs_access_search_rbtree(struct inode
*inode
, struct rpc_cred
*cred
)
1837 struct rb_node
*n
= NFS_I(inode
)->access_cache
.rb_node
;
1838 struct nfs_access_entry
*entry
;
1841 entry
= rb_entry(n
, struct nfs_access_entry
, rb_node
);
1843 if (cred
< entry
->cred
)
1845 else if (cred
> entry
->cred
)
1853 int nfs_access_get_cached(struct inode
*inode
, struct rpc_cred
*cred
, struct nfs_access_entry
*res
)
1855 struct nfs_inode
*nfsi
= NFS_I(inode
);
1856 struct nfs_access_entry
*cache
;
1859 spin_lock(&inode
->i_lock
);
1860 if (nfsi
->cache_validity
& NFS_INO_INVALID_ACCESS
)
1862 cache
= nfs_access_search_rbtree(inode
, cred
);
1865 if (time_after(jiffies
, cache
->jiffies
+ NFS_ATTRTIMEO(inode
)))
1867 res
->jiffies
= cache
->jiffies
;
1868 res
->cred
= cache
->cred
;
1869 res
->mask
= cache
->mask
;
1870 list_move_tail(&cache
->lru
, &nfsi
->access_cache_entry_lru
);
1873 spin_unlock(&inode
->i_lock
);
1876 rb_erase(&cache
->rb_node
, &nfsi
->access_cache
);
1877 list_del(&cache
->lru
);
1878 spin_unlock(&inode
->i_lock
);
1879 nfs_access_free_entry(cache
);
1882 /* This will release the spinlock */
1883 __nfs_access_zap_cache(inode
);
1887 static void nfs_access_add_rbtree(struct inode
*inode
, struct nfs_access_entry
*set
)
1889 struct nfs_inode
*nfsi
= NFS_I(inode
);
1890 struct rb_root
*root_node
= &nfsi
->access_cache
;
1891 struct rb_node
**p
= &root_node
->rb_node
;
1892 struct rb_node
*parent
= NULL
;
1893 struct nfs_access_entry
*entry
;
1895 spin_lock(&inode
->i_lock
);
1896 while (*p
!= NULL
) {
1898 entry
= rb_entry(parent
, struct nfs_access_entry
, rb_node
);
1900 if (set
->cred
< entry
->cred
)
1901 p
= &parent
->rb_left
;
1902 else if (set
->cred
> entry
->cred
)
1903 p
= &parent
->rb_right
;
1907 rb_link_node(&set
->rb_node
, parent
, p
);
1908 rb_insert_color(&set
->rb_node
, root_node
);
1909 list_add_tail(&set
->lru
, &nfsi
->access_cache_entry_lru
);
1910 spin_unlock(&inode
->i_lock
);
1913 rb_replace_node(parent
, &set
->rb_node
, root_node
);
1914 list_add_tail(&set
->lru
, &nfsi
->access_cache_entry_lru
);
1915 list_del(&entry
->lru
);
1916 spin_unlock(&inode
->i_lock
);
1917 nfs_access_free_entry(entry
);
1920 void nfs_access_add_cache(struct inode
*inode
, struct nfs_access_entry
*set
)
1922 struct nfs_access_entry
*cache
= kmalloc(sizeof(*cache
), GFP_KERNEL
);
1925 RB_CLEAR_NODE(&cache
->rb_node
);
1926 cache
->jiffies
= set
->jiffies
;
1927 cache
->cred
= get_rpccred(set
->cred
);
1928 cache
->mask
= set
->mask
;
1930 nfs_access_add_rbtree(inode
, cache
);
1932 /* Update accounting */
1933 smp_mb__before_atomic_inc();
1934 atomic_long_inc(&nfs_access_nr_entries
);
1935 smp_mb__after_atomic_inc();
1937 /* Add inode to global LRU list */
1938 if (!test_and_set_bit(NFS_INO_ACL_LRU_SET
, &NFS_FLAGS(inode
))) {
1939 spin_lock(&nfs_access_lru_lock
);
1940 list_add_tail(&NFS_I(inode
)->access_cache_inode_lru
, &nfs_access_lru_list
);
1941 spin_unlock(&nfs_access_lru_lock
);
1945 static int nfs_do_access(struct inode
*inode
, struct rpc_cred
*cred
, int mask
)
1947 struct nfs_access_entry cache
;
1950 status
= nfs_access_get_cached(inode
, cred
, &cache
);
1954 /* Be clever: ask server to check for all possible rights */
1955 cache
.mask
= MAY_EXEC
| MAY_WRITE
| MAY_READ
;
1957 cache
.jiffies
= jiffies
;
1958 status
= NFS_PROTO(inode
)->access(inode
, &cache
);
1961 nfs_access_add_cache(inode
, &cache
);
1963 if ((cache
.mask
& mask
) == mask
)
1968 int nfs_permission(struct inode
*inode
, int mask
, struct nameidata
*nd
)
1970 struct rpc_cred
*cred
;
1973 nfs_inc_stats(inode
, NFSIOS_VFSACCESS
);
1977 /* Is this sys_access() ? */
1978 if (nd
!= NULL
&& (nd
->flags
& LOOKUP_ACCESS
))
1981 switch (inode
->i_mode
& S_IFMT
) {
1985 /* NFSv4 has atomic_open... */
1986 if (nfs_server_capable(inode
, NFS_CAP_ATOMIC_OPEN
)
1988 && (nd
->flags
& LOOKUP_OPEN
))
1993 * Optimize away all write operations, since the server
1994 * will check permissions when we perform the op.
1996 if ((mask
& MAY_WRITE
) && !(mask
& MAY_READ
))
2003 if (!NFS_PROTO(inode
)->access
)
2006 cred
= rpcauth_lookupcred(NFS_CLIENT(inode
)->cl_auth
, 0);
2007 if (!IS_ERR(cred
)) {
2008 res
= nfs_do_access(inode
, cred
, mask
);
2011 res
= PTR_ERR(cred
);
2014 dfprintk(VFS
, "NFS: permission(%s/%ld), mask=0x%x, res=%d\n",
2015 inode
->i_sb
->s_id
, inode
->i_ino
, mask
, res
);
2018 res
= nfs_revalidate_inode(NFS_SERVER(inode
), inode
);
2020 res
= generic_permission(inode
, mask
, NULL
);
2027 * version-control: t
2028 * kept-new-versions: 5