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 } nfs_readdir_descriptor_t
;
159 /* Now we cache directories properly, by stuffing the dirent
160 * data directly in the page cache.
162 * Inode invalidation due to refresh etc. takes care of
163 * _everything_, no sloppy entry flushing logic, no extraneous
164 * copying, network direct to page cache, the way it was meant
167 * NOTE: Dirent information verification is done always by the
168 * page-in of the RPC reply, nowhere else, this simplies
169 * things substantially.
172 int nfs_readdir_filler(nfs_readdir_descriptor_t
*desc
, struct page
*page
)
174 struct file
*file
= desc
->file
;
175 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
176 struct rpc_cred
*cred
= nfs_file_cred(file
);
177 unsigned long timestamp
;
180 dfprintk(DIRCACHE
, "NFS: %s: reading cookie %Lu into page %lu\n",
181 __FUNCTION__
, (long long)desc
->entry
->cookie
,
186 error
= NFS_PROTO(inode
)->readdir(file
->f_path
.dentry
, cred
, desc
->entry
->cookie
, page
,
187 NFS_SERVER(inode
)->dtsize
, desc
->plus
);
189 /* We requested READDIRPLUS, but the server doesn't grok it */
190 if (error
== -ENOTSUPP
&& desc
->plus
) {
191 NFS_SERVER(inode
)->caps
&= ~NFS_CAP_READDIRPLUS
;
192 clear_bit(NFS_INO_ADVISE_RDPLUS
, &NFS_FLAGS(inode
));
198 SetPageUptodate(page
);
199 spin_lock(&inode
->i_lock
);
200 NFS_I(inode
)->cache_validity
|= NFS_INO_INVALID_ATIME
;
201 spin_unlock(&inode
->i_lock
);
202 /* Ensure consistent page alignment of the data.
203 * Note: assumes we have exclusive access to this mapping either
204 * through inode->i_mutex or some other mechanism.
206 if (page
->index
== 0 && invalidate_inode_pages2_range(inode
->i_mapping
, PAGE_CACHE_SIZE
, -1) < 0) {
207 /* Should never happen */
208 nfs_zap_mapping(inode
, inode
->i_mapping
);
215 nfs_zap_caches(inode
);
221 int dir_decode(nfs_readdir_descriptor_t
*desc
)
223 __be32
*p
= desc
->ptr
;
224 p
= desc
->decode(p
, desc
->entry
, desc
->plus
);
232 void dir_page_release(nfs_readdir_descriptor_t
*desc
)
235 page_cache_release(desc
->page
);
241 * Given a pointer to a buffer that has already been filled by a call
242 * to readdir, find the next entry with cookie '*desc->dir_cookie'.
244 * If the end of the buffer has been reached, return -EAGAIN, if not,
245 * return the offset within the buffer of the next entry to be
249 int find_dirent(nfs_readdir_descriptor_t
*desc
)
251 struct nfs_entry
*entry
= desc
->entry
;
255 while((status
= dir_decode(desc
)) == 0) {
256 dfprintk(DIRCACHE
, "NFS: %s: examining cookie %Lu\n",
257 __FUNCTION__
, (unsigned long long)entry
->cookie
);
258 if (entry
->prev_cookie
== *desc
->dir_cookie
)
260 if (loop_count
++ > 200) {
269 * Given a pointer to a buffer that has already been filled by a call
270 * to readdir, find the entry at offset 'desc->file->f_pos'.
272 * If the end of the buffer has been reached, return -EAGAIN, if not,
273 * return the offset within the buffer of the next entry to be
277 int find_dirent_index(nfs_readdir_descriptor_t
*desc
)
279 struct nfs_entry
*entry
= desc
->entry
;
284 status
= dir_decode(desc
);
288 dfprintk(DIRCACHE
, "NFS: found cookie %Lu at index %Ld\n",
289 (unsigned long long)entry
->cookie
, desc
->current_index
);
291 if (desc
->file
->f_pos
== desc
->current_index
) {
292 *desc
->dir_cookie
= entry
->cookie
;
295 desc
->current_index
++;
296 if (loop_count
++ > 200) {
305 * Find the given page, and call find_dirent() or find_dirent_index in
306 * order to try to return the next entry.
309 int find_dirent_page(nfs_readdir_descriptor_t
*desc
)
311 struct inode
*inode
= desc
->file
->f_path
.dentry
->d_inode
;
315 dfprintk(DIRCACHE
, "NFS: %s: searching page %ld for target %Lu\n",
316 __FUNCTION__
, desc
->page_index
,
317 (long long) *desc
->dir_cookie
);
319 page
= read_cache_page(inode
->i_mapping
, desc
->page_index
,
320 (filler_t
*)nfs_readdir_filler
, desc
);
322 status
= PTR_ERR(page
);
325 if (!PageUptodate(page
))
328 /* NOTE: Someone else may have changed the READDIRPLUS flag */
330 desc
->ptr
= kmap(page
); /* matching kunmap in nfs_do_filldir */
331 if (*desc
->dir_cookie
!= 0)
332 status
= find_dirent(desc
);
334 status
= find_dirent_index(desc
);
336 dir_page_release(desc
);
338 dfprintk(DIRCACHE
, "NFS: %s: returns %d\n", __FUNCTION__
, status
);
341 page_cache_release(page
);
346 * Recurse through the page cache pages, and return a
347 * filled nfs_entry structure of the next directory entry if possible.
349 * The target for the search is '*desc->dir_cookie' if non-0,
350 * 'desc->file->f_pos' otherwise
353 int readdir_search_pagecache(nfs_readdir_descriptor_t
*desc
)
358 /* Always search-by-index from the beginning of the cache */
359 if (*desc
->dir_cookie
== 0) {
360 dfprintk(DIRCACHE
, "NFS: readdir_search_pagecache() searching for offset %Ld\n",
361 (long long)desc
->file
->f_pos
);
362 desc
->page_index
= 0;
363 desc
->entry
->cookie
= desc
->entry
->prev_cookie
= 0;
364 desc
->entry
->eof
= 0;
365 desc
->current_index
= 0;
367 dfprintk(DIRCACHE
, "NFS: readdir_search_pagecache() searching for cookie %Lu\n",
368 (unsigned long long)*desc
->dir_cookie
);
371 res
= find_dirent_page(desc
);
374 /* Align to beginning of next page */
376 if (loop_count
++ > 200) {
382 dfprintk(DIRCACHE
, "NFS: %s: returns %d\n", __FUNCTION__
, res
);
386 static inline unsigned int dt_type(struct inode
*inode
)
388 return (inode
->i_mode
>> 12) & 15;
391 static struct dentry
*nfs_readdir_lookup(nfs_readdir_descriptor_t
*desc
);
394 * Once we've found the start of the dirent within a page: fill 'er up...
397 int nfs_do_filldir(nfs_readdir_descriptor_t
*desc
, void *dirent
,
400 struct file
*file
= desc
->file
;
401 struct nfs_entry
*entry
= desc
->entry
;
402 struct dentry
*dentry
= NULL
;
403 unsigned long fileid
;
407 dfprintk(DIRCACHE
, "NFS: nfs_do_filldir() filling starting @ cookie %Lu\n",
408 (unsigned long long)entry
->cookie
);
411 unsigned d_type
= DT_UNKNOWN
;
412 /* Note: entry->prev_cookie contains the cookie for
413 * retrieving the current dirent on the server */
414 fileid
= nfs_fileid_to_ino_t(entry
->ino
);
416 /* Get a dentry if we have one */
419 dentry
= nfs_readdir_lookup(desc
);
421 /* Use readdirplus info */
422 if (dentry
!= NULL
&& dentry
->d_inode
!= NULL
) {
423 d_type
= dt_type(dentry
->d_inode
);
424 fileid
= dentry
->d_inode
->i_ino
;
427 res
= filldir(dirent
, entry
->name
, entry
->len
,
428 file
->f_pos
, fileid
, d_type
);
432 *desc
->dir_cookie
= entry
->cookie
;
433 if (dir_decode(desc
) != 0) {
437 if (loop_count
++ > 200) {
442 dir_page_release(desc
);
445 dfprintk(DIRCACHE
, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n",
446 (unsigned long long)*desc
->dir_cookie
, res
);
451 * If we cannot find a cookie in our cache, we suspect that this is
452 * because it points to a deleted file, so we ask the server to return
453 * whatever it thinks is the next entry. We then feed this to filldir.
454 * If all goes well, we should then be able to find our way round the
455 * cache on the next call to readdir_search_pagecache();
457 * NOTE: we cannot add the anonymous page to the pagecache because
458 * the data it contains might not be page aligned. Besides,
459 * we should already have a complete representation of the
460 * directory in the page cache by the time we get here.
463 int uncached_readdir(nfs_readdir_descriptor_t
*desc
, void *dirent
,
466 struct file
*file
= desc
->file
;
467 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
468 struct rpc_cred
*cred
= nfs_file_cred(file
);
469 struct page
*page
= NULL
;
472 dfprintk(DIRCACHE
, "NFS: uncached_readdir() searching for cookie %Lu\n",
473 (unsigned long long)*desc
->dir_cookie
);
475 page
= alloc_page(GFP_HIGHUSER
);
480 desc
->error
= NFS_PROTO(inode
)->readdir(file
->f_path
.dentry
, cred
, *desc
->dir_cookie
,
482 NFS_SERVER(inode
)->dtsize
,
484 spin_lock(&inode
->i_lock
);
485 NFS_I(inode
)->cache_validity
|= NFS_INO_INVALID_ATIME
;
486 spin_unlock(&inode
->i_lock
);
488 desc
->ptr
= kmap(page
); /* matching kunmap in nfs_do_filldir */
489 if (desc
->error
>= 0) {
490 if ((status
= dir_decode(desc
)) == 0)
491 desc
->entry
->prev_cookie
= *desc
->dir_cookie
;
497 status
= nfs_do_filldir(desc
, dirent
, filldir
);
499 /* Reset read descriptor so it searches the page cache from
500 * the start upon the next call to readdir_search_pagecache() */
501 desc
->page_index
= 0;
502 desc
->entry
->cookie
= desc
->entry
->prev_cookie
= 0;
503 desc
->entry
->eof
= 0;
505 dfprintk(DIRCACHE
, "NFS: %s: returns %d\n",
506 __FUNCTION__
, status
);
509 dir_page_release(desc
);
513 /* The file offset position represents the dirent entry number. A
514 last cookie cache takes care of the common case of reading the
517 static int nfs_readdir(struct file
*filp
, void *dirent
, filldir_t filldir
)
519 struct dentry
*dentry
= filp
->f_path
.dentry
;
520 struct inode
*inode
= dentry
->d_inode
;
521 nfs_readdir_descriptor_t my_desc
,
523 struct nfs_entry my_entry
;
525 struct nfs_fattr fattr
;
528 dfprintk(VFS
, "NFS: readdir(%s/%s) starting at cookie %Lu\n",
529 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
,
530 (long long)filp
->f_pos
);
531 nfs_inc_stats(inode
, NFSIOS_VFSGETDENTS
);
535 res
= nfs_revalidate_mapping_nolock(inode
, filp
->f_mapping
);
542 * filp->f_pos points to the dirent entry number.
543 * *desc->dir_cookie has the cookie for the next entry. We have
544 * to either find the entry with the appropriate number or
545 * revalidate the cookie.
547 memset(desc
, 0, sizeof(*desc
));
550 desc
->dir_cookie
= &((struct nfs_open_context
*)filp
->private_data
)->dir_cookie
;
551 desc
->decode
= NFS_PROTO(inode
)->decode_dirent
;
552 desc
->plus
= NFS_USE_READDIRPLUS(inode
);
554 my_entry
.cookie
= my_entry
.prev_cookie
= 0;
557 my_entry
.fattr
= &fattr
;
558 nfs_fattr_init(&fattr
);
559 desc
->entry
= &my_entry
;
561 while(!desc
->entry
->eof
) {
562 res
= readdir_search_pagecache(desc
);
564 if (res
== -EBADCOOKIE
) {
565 /* This means either end of directory */
566 if (*desc
->dir_cookie
&& desc
->entry
->cookie
!= *desc
->dir_cookie
) {
567 /* Or that the server has 'lost' a cookie */
568 res
= uncached_readdir(desc
, dirent
, filldir
);
575 if (res
== -ETOOSMALL
&& desc
->plus
) {
576 clear_bit(NFS_INO_ADVISE_RDPLUS
, &NFS_FLAGS(inode
));
577 nfs_zap_caches(inode
);
579 desc
->entry
->eof
= 0;
585 res
= nfs_do_filldir(desc
, dirent
, filldir
);
594 dfprintk(VFS
, "NFS: readdir(%s/%s) returns %ld\n",
595 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
,
600 loff_t
nfs_llseek_dir(struct file
*filp
, loff_t offset
, int origin
)
602 mutex_lock(&filp
->f_path
.dentry
->d_inode
->i_mutex
);
605 offset
+= filp
->f_pos
;
613 if (offset
!= filp
->f_pos
) {
614 filp
->f_pos
= offset
;
615 ((struct nfs_open_context
*)filp
->private_data
)->dir_cookie
= 0;
618 mutex_unlock(&filp
->f_path
.dentry
->d_inode
->i_mutex
);
623 * All directory operations under NFS are synchronous, so fsync()
624 * is a dummy operation.
626 int nfs_fsync_dir(struct file
*filp
, struct dentry
*dentry
, int datasync
)
628 dfprintk(VFS
, "NFS: fsync_dir(%s/%s) datasync %d\n",
629 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
,
636 * A check for whether or not the parent directory has changed.
637 * In the case it has, we assume that the dentries are untrustworthy
638 * and may need to be looked up again.
640 static int nfs_check_verifier(struct inode
*dir
, struct dentry
*dentry
)
644 if ((NFS_I(dir
)->cache_validity
& NFS_INO_INVALID_ATTR
) != 0
645 || nfs_attribute_timeout(dir
))
647 return nfs_verify_change_attribute(dir
, (unsigned long)dentry
->d_fsdata
);
650 static inline void nfs_set_verifier(struct dentry
* dentry
, unsigned long verf
)
652 dentry
->d_fsdata
= (void *)verf
;
655 static void nfs_refresh_verifier(struct dentry
* dentry
, unsigned long verf
)
657 if (time_after(verf
, (unsigned long)dentry
->d_fsdata
))
658 nfs_set_verifier(dentry
, verf
);
662 * Whenever an NFS operation succeeds, we know that the dentry
663 * is valid, so we update the revalidation timestamp.
665 static inline void nfs_renew_times(struct dentry
* dentry
)
667 dentry
->d_time
= jiffies
;
671 * Return the intent data that applies to this particular path component
673 * Note that the current set of intents only apply to the very last
674 * component of the path.
675 * We check for this using LOOKUP_CONTINUE and LOOKUP_PARENT.
677 static inline unsigned int nfs_lookup_check_intent(struct nameidata
*nd
, unsigned int mask
)
679 if (nd
->flags
& (LOOKUP_CONTINUE
|LOOKUP_PARENT
))
681 return nd
->flags
& mask
;
685 * Inode and filehandle revalidation for lookups.
687 * We force revalidation in the cases where the VFS sets LOOKUP_REVAL,
688 * or if the intent information indicates that we're about to open this
689 * particular file and the "nocto" mount flag is not set.
693 int nfs_lookup_verify_inode(struct inode
*inode
, struct nameidata
*nd
)
695 struct nfs_server
*server
= NFS_SERVER(inode
);
698 /* VFS wants an on-the-wire revalidation */
699 if (nd
->flags
& LOOKUP_REVAL
)
701 /* This is an open(2) */
702 if (nfs_lookup_check_intent(nd
, LOOKUP_OPEN
) != 0 &&
703 !(server
->flags
& NFS_MOUNT_NOCTO
) &&
704 (S_ISREG(inode
->i_mode
) ||
705 S_ISDIR(inode
->i_mode
)))
708 return nfs_revalidate_inode(server
, inode
);
710 return __nfs_revalidate_inode(server
, inode
);
714 * We judge how long we want to trust negative
715 * dentries by looking at the parent inode mtime.
717 * If parent mtime has changed, we revalidate, else we wait for a
718 * period corresponding to the parent's attribute cache timeout value.
721 int nfs_neg_need_reval(struct inode
*dir
, struct dentry
*dentry
,
722 struct nameidata
*nd
)
724 /* Don't revalidate a negative dentry if we're creating a new file */
725 if (nd
!= NULL
&& nfs_lookup_check_intent(nd
, LOOKUP_CREATE
) != 0)
727 return !nfs_check_verifier(dir
, dentry
);
731 * This is called every time the dcache has a lookup hit,
732 * and we should check whether we can really trust that
735 * NOTE! The hit can be a negative hit too, don't assume
738 * If the parent directory is seen to have changed, we throw out the
739 * cached dentry and do a new lookup.
741 static int nfs_lookup_revalidate(struct dentry
* dentry
, struct nameidata
*nd
)
745 struct dentry
*parent
;
747 struct nfs_fh fhandle
;
748 struct nfs_fattr fattr
;
749 unsigned long verifier
;
751 parent
= dget_parent(dentry
);
753 dir
= parent
->d_inode
;
754 nfs_inc_stats(dir
, NFSIOS_DENTRYREVALIDATE
);
755 inode
= dentry
->d_inode
;
758 if (nfs_neg_need_reval(dir
, dentry
, nd
))
763 if (is_bad_inode(inode
)) {
764 dfprintk(LOOKUPCACHE
, "%s: %s/%s has dud inode\n",
765 __FUNCTION__
, dentry
->d_parent
->d_name
.name
,
766 dentry
->d_name
.name
);
770 /* Revalidate parent directory attribute cache */
771 if (nfs_revalidate_inode(NFS_SERVER(dir
), dir
) < 0)
774 /* Force a full look up iff the parent directory has changed */
775 if (nfs_check_verifier(dir
, dentry
)) {
776 if (nfs_lookup_verify_inode(inode
, nd
))
781 if (NFS_STALE(inode
))
784 verifier
= nfs_save_change_attribute(dir
);
785 error
= NFS_PROTO(dir
)->lookup(dir
, &dentry
->d_name
, &fhandle
, &fattr
);
788 if (nfs_compare_fh(NFS_FH(inode
), &fhandle
))
790 if ((error
= nfs_refresh_inode(inode
, &fattr
)) != 0)
793 nfs_renew_times(dentry
);
794 nfs_refresh_verifier(dentry
, verifier
);
798 dfprintk(LOOKUPCACHE
, "NFS: %s(%s/%s) is valid\n",
799 __FUNCTION__
, dentry
->d_parent
->d_name
.name
,
800 dentry
->d_name
.name
);
806 if (inode
&& S_ISDIR(inode
->i_mode
)) {
807 /* Purge readdir caches. */
808 nfs_zap_caches(inode
);
809 /* If we have submounts, don't unhash ! */
810 if (have_submounts(dentry
))
812 shrink_dcache_parent(dentry
);
817 dfprintk(LOOKUPCACHE
, "NFS: %s(%s/%s) is invalid\n",
818 __FUNCTION__
, dentry
->d_parent
->d_name
.name
,
819 dentry
->d_name
.name
);
824 * This is called from dput() when d_count is going to 0.
826 static int nfs_dentry_delete(struct dentry
*dentry
)
828 dfprintk(VFS
, "NFS: dentry_delete(%s/%s, %x)\n",
829 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
,
832 if (dentry
->d_flags
& DCACHE_NFSFS_RENAMED
) {
833 /* Unhash it, so that ->d_iput() would be called */
836 if (!(dentry
->d_sb
->s_flags
& MS_ACTIVE
)) {
837 /* Unhash it, so that ancestors of killed async unlink
838 * files will be cleaned up during umount */
846 * Called when the dentry loses inode.
847 * We use it to clean up silly-renamed files.
849 static void nfs_dentry_iput(struct dentry
*dentry
, struct inode
*inode
)
851 nfs_inode_return_delegation(inode
);
852 if (dentry
->d_flags
& DCACHE_NFSFS_RENAMED
) {
855 nfs_complete_unlink(dentry
);
858 /* When creating a negative dentry, we want to renew d_time */
859 nfs_renew_times(dentry
);
863 struct dentry_operations nfs_dentry_operations
= {
864 .d_revalidate
= nfs_lookup_revalidate
,
865 .d_delete
= nfs_dentry_delete
,
866 .d_iput
= nfs_dentry_iput
,
870 * Use intent information to check whether or not we're going to do
871 * an O_EXCL create using this path component.
874 int nfs_is_exclusive_create(struct inode
*dir
, struct nameidata
*nd
)
876 if (NFS_PROTO(dir
)->version
== 2)
878 if (nd
== NULL
|| nfs_lookup_check_intent(nd
, LOOKUP_CREATE
) == 0)
880 return (nd
->intent
.open
.flags
& O_EXCL
) != 0;
883 static inline int nfs_reval_fsid(struct vfsmount
*mnt
, struct inode
*dir
,
884 struct nfs_fh
*fh
, struct nfs_fattr
*fattr
)
886 struct nfs_server
*server
= NFS_SERVER(dir
);
888 if (!nfs_fsid_equal(&server
->fsid
, &fattr
->fsid
))
889 /* Revalidate fsid on root dir */
890 return __nfs_revalidate_inode(server
, mnt
->mnt_root
->d_inode
);
894 static struct dentry
*nfs_lookup(struct inode
*dir
, struct dentry
* dentry
, struct nameidata
*nd
)
897 struct inode
*inode
= NULL
;
899 struct nfs_fh fhandle
;
900 struct nfs_fattr fattr
;
902 dfprintk(VFS
, "NFS: lookup(%s/%s)\n",
903 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
);
904 nfs_inc_stats(dir
, NFSIOS_VFSLOOKUP
);
906 res
= ERR_PTR(-ENAMETOOLONG
);
907 if (dentry
->d_name
.len
> NFS_SERVER(dir
)->namelen
)
910 res
= ERR_PTR(-ENOMEM
);
911 dentry
->d_op
= NFS_PROTO(dir
)->dentry_ops
;
916 * If we're doing an exclusive create, optimize away the lookup
917 * but don't hash the dentry.
919 if (nfs_is_exclusive_create(dir
, nd
)) {
920 d_instantiate(dentry
, NULL
);
925 error
= NFS_PROTO(dir
)->lookup(dir
, &dentry
->d_name
, &fhandle
, &fattr
);
926 if (error
== -ENOENT
)
929 res
= ERR_PTR(error
);
932 error
= nfs_reval_fsid(nd
->mnt
, dir
, &fhandle
, &fattr
);
934 res
= ERR_PTR(error
);
937 inode
= nfs_fhget(dentry
->d_sb
, &fhandle
, &fattr
);
938 res
= (struct dentry
*)inode
;
943 res
= d_materialise_unique(dentry
, inode
);
945 struct dentry
*parent
;
948 /* Was a directory renamed! */
949 parent
= dget_parent(res
);
950 if (!IS_ROOT(parent
))
951 nfs_mark_for_revalidate(parent
->d_inode
);
955 nfs_renew_times(dentry
);
956 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
964 static int nfs_open_revalidate(struct dentry
*, struct nameidata
*);
966 struct dentry_operations nfs4_dentry_operations
= {
967 .d_revalidate
= nfs_open_revalidate
,
968 .d_delete
= nfs_dentry_delete
,
969 .d_iput
= nfs_dentry_iput
,
973 * Use intent information to determine whether we need to substitute
974 * the NFSv4-style stateful OPEN for the LOOKUP call
976 static int is_atomic_open(struct inode
*dir
, struct nameidata
*nd
)
978 if (nd
== NULL
|| nfs_lookup_check_intent(nd
, LOOKUP_OPEN
) == 0)
980 /* NFS does not (yet) have a stateful open for directories */
981 if (nd
->flags
& LOOKUP_DIRECTORY
)
983 /* Are we trying to write to a read only partition? */
984 if (IS_RDONLY(dir
) && (nd
->intent
.open
.flags
& (O_CREAT
|O_TRUNC
|FMODE_WRITE
)))
989 static struct dentry
*nfs_atomic_lookup(struct inode
*dir
, struct dentry
*dentry
, struct nameidata
*nd
)
991 struct dentry
*res
= NULL
;
994 dfprintk(VFS
, "NFS: atomic_lookup(%s/%ld), %s\n",
995 dir
->i_sb
->s_id
, dir
->i_ino
, dentry
->d_name
.name
);
997 /* Check that we are indeed trying to open this file */
998 if (!is_atomic_open(dir
, nd
))
1001 if (dentry
->d_name
.len
> NFS_SERVER(dir
)->namelen
) {
1002 res
= ERR_PTR(-ENAMETOOLONG
);
1005 dentry
->d_op
= NFS_PROTO(dir
)->dentry_ops
;
1007 /* Let vfs_create() deal with O_EXCL */
1008 if (nd
->intent
.open
.flags
& O_EXCL
) {
1009 d_add(dentry
, NULL
);
1013 /* Open the file on the server */
1015 /* Revalidate parent directory attribute cache */
1016 error
= nfs_revalidate_inode(NFS_SERVER(dir
), dir
);
1018 res
= ERR_PTR(error
);
1023 if (nd
->intent
.open
.flags
& O_CREAT
) {
1024 nfs_begin_data_update(dir
);
1025 res
= nfs4_atomic_open(dir
, dentry
, nd
);
1026 nfs_end_data_update(dir
);
1028 res
= nfs4_atomic_open(dir
, dentry
, nd
);
1031 error
= PTR_ERR(res
);
1033 /* Make a negative dentry */
1037 /* This turned out not to be a regular file */
1042 if (!(nd
->intent
.open
.flags
& O_NOFOLLOW
))
1048 } else if (res
!= NULL
)
1050 nfs_renew_times(dentry
);
1051 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1055 return nfs_lookup(dir
, dentry
, nd
);
1058 static int nfs_open_revalidate(struct dentry
*dentry
, struct nameidata
*nd
)
1060 struct dentry
*parent
= NULL
;
1061 struct inode
*inode
= dentry
->d_inode
;
1063 unsigned long verifier
;
1064 int openflags
, ret
= 0;
1066 parent
= dget_parent(dentry
);
1067 dir
= parent
->d_inode
;
1068 if (!is_atomic_open(dir
, nd
))
1070 /* We can't create new files in nfs_open_revalidate(), so we
1071 * optimize away revalidation of negative dentries.
1075 /* NFS only supports OPEN on regular files */
1076 if (!S_ISREG(inode
->i_mode
))
1078 openflags
= nd
->intent
.open
.flags
;
1079 /* We cannot do exclusive creation on a positive dentry */
1080 if ((openflags
& (O_CREAT
|O_EXCL
)) == (O_CREAT
|O_EXCL
))
1082 /* We can't create new files, or truncate existing ones here */
1083 openflags
&= ~(O_CREAT
|O_TRUNC
);
1086 * Note: we're not holding inode->i_mutex and so may be racing with
1087 * operations that change the directory. We therefore save the
1088 * change attribute *before* we do the RPC call.
1091 verifier
= nfs_save_change_attribute(dir
);
1092 ret
= nfs4_open_revalidate(dir
, dentry
, openflags
, nd
);
1094 nfs_refresh_verifier(dentry
, verifier
);
1103 if (inode
!= NULL
&& nfs_have_delegation(inode
, FMODE_READ
))
1105 return nfs_lookup_revalidate(dentry
, nd
);
1107 #endif /* CONFIG_NFSV4 */
1109 static struct dentry
*nfs_readdir_lookup(nfs_readdir_descriptor_t
*desc
)
1111 struct dentry
*parent
= desc
->file
->f_path
.dentry
;
1112 struct inode
*dir
= parent
->d_inode
;
1113 struct nfs_entry
*entry
= desc
->entry
;
1114 struct dentry
*dentry
, *alias
;
1115 struct qstr name
= {
1116 .name
= entry
->name
,
1119 struct inode
*inode
;
1123 if (name
.name
[0] == '.' && name
.name
[1] == '.')
1124 return dget_parent(parent
);
1127 if (name
.name
[0] == '.')
1128 return dget(parent
);
1130 name
.hash
= full_name_hash(name
.name
, name
.len
);
1131 dentry
= d_lookup(parent
, &name
);
1132 if (dentry
!= NULL
) {
1133 /* Is this a positive dentry that matches the readdir info? */
1134 if (dentry
->d_inode
!= NULL
&&
1135 (NFS_FILEID(dentry
->d_inode
) == entry
->ino
||
1136 d_mountpoint(dentry
))) {
1137 if (!desc
->plus
|| entry
->fh
->size
== 0)
1139 if (nfs_compare_fh(NFS_FH(dentry
->d_inode
),
1143 /* No, so d_drop to allow one to be created */
1147 if (!desc
->plus
|| !(entry
->fattr
->valid
& NFS_ATTR_FATTR
))
1149 /* Note: caller is already holding the dir->i_mutex! */
1150 dentry
= d_alloc(parent
, &name
);
1153 dentry
->d_op
= NFS_PROTO(dir
)->dentry_ops
;
1154 inode
= nfs_fhget(dentry
->d_sb
, entry
->fh
, entry
->fattr
);
1155 if (IS_ERR(inode
)) {
1160 alias
= d_materialise_unique(dentry
, inode
);
1161 if (alias
!= NULL
) {
1168 nfs_renew_times(dentry
);
1169 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1172 nfs_renew_times(dentry
);
1173 nfs_refresh_verifier(dentry
, nfs_save_change_attribute(dir
));
1178 * Code common to create, mkdir, and mknod.
1180 int nfs_instantiate(struct dentry
*dentry
, struct nfs_fh
*fhandle
,
1181 struct nfs_fattr
*fattr
)
1183 struct inode
*inode
;
1184 int error
= -EACCES
;
1186 /* We may have been initialized further down */
1187 if (dentry
->d_inode
)
1189 if (fhandle
->size
== 0) {
1190 struct inode
*dir
= dentry
->d_parent
->d_inode
;
1191 error
= NFS_PROTO(dir
)->lookup(dir
, &dentry
->d_name
, fhandle
, fattr
);
1195 if (!(fattr
->valid
& NFS_ATTR_FATTR
)) {
1196 struct nfs_server
*server
= NFS_SB(dentry
->d_sb
);
1197 error
= server
->nfs_client
->rpc_ops
->getattr(server
, fhandle
, fattr
);
1201 inode
= nfs_fhget(dentry
->d_sb
, fhandle
, fattr
);
1202 error
= PTR_ERR(inode
);
1205 d_instantiate(dentry
, inode
);
1206 if (d_unhashed(dentry
))
1212 * Following a failed create operation, we drop the dentry rather
1213 * than retain a negative dentry. This avoids a problem in the event
1214 * that the operation succeeded on the server, but an error in the
1215 * reply path made it appear to have failed.
1217 static int nfs_create(struct inode
*dir
, struct dentry
*dentry
, int mode
,
1218 struct nameidata
*nd
)
1224 dfprintk(VFS
, "NFS: create(%s/%ld), %s\n",
1225 dir
->i_sb
->s_id
, dir
->i_ino
, dentry
->d_name
.name
);
1227 attr
.ia_mode
= mode
;
1228 attr
.ia_valid
= ATTR_MODE
;
1230 if (nd
&& (nd
->flags
& LOOKUP_CREATE
))
1231 open_flags
= nd
->intent
.open
.flags
;
1234 nfs_begin_data_update(dir
);
1235 error
= NFS_PROTO(dir
)->create(dir
, dentry
, &attr
, open_flags
, nd
);
1236 nfs_end_data_update(dir
);
1239 nfs_renew_times(dentry
);
1240 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1250 * See comments for nfs_proc_create regarding failed operations.
1253 nfs_mknod(struct inode
*dir
, struct dentry
*dentry
, int mode
, dev_t rdev
)
1258 dfprintk(VFS
, "NFS: mknod(%s/%ld), %s\n",
1259 dir
->i_sb
->s_id
, dir
->i_ino
, dentry
->d_name
.name
);
1261 if (!new_valid_dev(rdev
))
1264 attr
.ia_mode
= mode
;
1265 attr
.ia_valid
= ATTR_MODE
;
1268 nfs_begin_data_update(dir
);
1269 status
= NFS_PROTO(dir
)->mknod(dir
, dentry
, &attr
, rdev
);
1270 nfs_end_data_update(dir
);
1273 nfs_renew_times(dentry
);
1274 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1284 * See comments for nfs_proc_create regarding failed operations.
1286 static int nfs_mkdir(struct inode
*dir
, struct dentry
*dentry
, int mode
)
1291 dfprintk(VFS
, "NFS: mkdir(%s/%ld), %s\n",
1292 dir
->i_sb
->s_id
, dir
->i_ino
, dentry
->d_name
.name
);
1294 attr
.ia_valid
= ATTR_MODE
;
1295 attr
.ia_mode
= mode
| S_IFDIR
;
1298 nfs_begin_data_update(dir
);
1299 error
= NFS_PROTO(dir
)->mkdir(dir
, dentry
, &attr
);
1300 nfs_end_data_update(dir
);
1303 nfs_renew_times(dentry
);
1304 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1313 static int nfs_rmdir(struct inode
*dir
, struct dentry
*dentry
)
1317 dfprintk(VFS
, "NFS: rmdir(%s/%ld), %s\n",
1318 dir
->i_sb
->s_id
, dir
->i_ino
, dentry
->d_name
.name
);
1321 nfs_begin_data_update(dir
);
1322 error
= NFS_PROTO(dir
)->rmdir(dir
, &dentry
->d_name
);
1323 /* Ensure the VFS deletes this inode */
1324 if (error
== 0 && dentry
->d_inode
!= NULL
)
1325 clear_nlink(dentry
->d_inode
);
1326 nfs_end_data_update(dir
);
1332 static int nfs_sillyrename(struct inode
*dir
, struct dentry
*dentry
)
1334 static unsigned int sillycounter
;
1335 const int i_inosize
= sizeof(dir
->i_ino
)*2;
1336 const int countersize
= sizeof(sillycounter
)*2;
1337 const int slen
= sizeof(".nfs") + i_inosize
+ countersize
- 1;
1340 struct dentry
*sdentry
;
1343 dfprintk(VFS
, "NFS: silly-rename(%s/%s, ct=%d)\n",
1344 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
,
1345 atomic_read(&dentry
->d_count
));
1346 nfs_inc_stats(dir
, NFSIOS_SILLYRENAME
);
1349 if (!dentry
->d_inode
)
1350 printk("NFS: silly-renaming %s/%s, negative dentry??\n",
1351 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
);
1354 * We don't allow a dentry to be silly-renamed twice.
1357 if (dentry
->d_flags
& DCACHE_NFSFS_RENAMED
)
1360 sprintf(silly
, ".nfs%*.*lx",
1361 i_inosize
, i_inosize
, dentry
->d_inode
->i_ino
);
1363 /* Return delegation in anticipation of the rename */
1364 nfs_inode_return_delegation(dentry
->d_inode
);
1368 char *suffix
= silly
+ slen
- countersize
;
1372 sprintf(suffix
, "%*.*x", countersize
, countersize
, sillycounter
);
1374 dfprintk(VFS
, "NFS: trying to rename %s to %s\n",
1375 dentry
->d_name
.name
, silly
);
1377 sdentry
= lookup_one_len(silly
, dentry
->d_parent
, slen
);
1379 * N.B. Better to return EBUSY here ... it could be
1380 * dangerous to delete the file while it's in use.
1382 if (IS_ERR(sdentry
))
1384 } while(sdentry
->d_inode
!= NULL
); /* need negative lookup */
1386 qsilly
.name
= silly
;
1387 qsilly
.len
= strlen(silly
);
1388 nfs_begin_data_update(dir
);
1389 if (dentry
->d_inode
) {
1390 nfs_begin_data_update(dentry
->d_inode
);
1391 error
= NFS_PROTO(dir
)->rename(dir
, &dentry
->d_name
,
1393 nfs_mark_for_revalidate(dentry
->d_inode
);
1394 nfs_end_data_update(dentry
->d_inode
);
1396 error
= NFS_PROTO(dir
)->rename(dir
, &dentry
->d_name
,
1398 nfs_end_data_update(dir
);
1400 nfs_renew_times(dentry
);
1401 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1402 d_move(dentry
, sdentry
);
1403 error
= nfs_async_unlink(dentry
);
1404 /* If we return 0 we don't unlink */
1412 * Remove a file after making sure there are no pending writes,
1413 * and after checking that the file has only one user.
1415 * We invalidate the attribute cache and free the inode prior to the operation
1416 * to avoid possible races if the server reuses the inode.
1418 static int nfs_safe_remove(struct dentry
*dentry
)
1420 struct inode
*dir
= dentry
->d_parent
->d_inode
;
1421 struct inode
*inode
= dentry
->d_inode
;
1424 dfprintk(VFS
, "NFS: safe_remove(%s/%s)\n",
1425 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
);
1427 /* If the dentry was sillyrenamed, we simply call d_delete() */
1428 if (dentry
->d_flags
& DCACHE_NFSFS_RENAMED
) {
1433 nfs_begin_data_update(dir
);
1434 if (inode
!= NULL
) {
1435 nfs_inode_return_delegation(inode
);
1436 nfs_begin_data_update(inode
);
1437 error
= NFS_PROTO(dir
)->remove(dir
, &dentry
->d_name
);
1438 /* The VFS may want to delete this inode */
1441 nfs_mark_for_revalidate(inode
);
1442 nfs_end_data_update(inode
);
1444 error
= NFS_PROTO(dir
)->remove(dir
, &dentry
->d_name
);
1445 nfs_end_data_update(dir
);
1450 /* We do silly rename. In case sillyrename() returns -EBUSY, the inode
1451 * belongs to an active ".nfs..." file and we return -EBUSY.
1453 * If sillyrename() returns 0, we do nothing, otherwise we unlink.
1455 static int nfs_unlink(struct inode
*dir
, struct dentry
*dentry
)
1458 int need_rehash
= 0;
1460 dfprintk(VFS
, "NFS: unlink(%s/%ld, %s)\n", dir
->i_sb
->s_id
,
1461 dir
->i_ino
, dentry
->d_name
.name
);
1464 spin_lock(&dcache_lock
);
1465 spin_lock(&dentry
->d_lock
);
1466 if (atomic_read(&dentry
->d_count
) > 1) {
1467 spin_unlock(&dentry
->d_lock
);
1468 spin_unlock(&dcache_lock
);
1469 /* Start asynchronous writeout of the inode */
1470 write_inode_now(dentry
->d_inode
, 0);
1471 error
= nfs_sillyrename(dir
, dentry
);
1475 if (!d_unhashed(dentry
)) {
1479 spin_unlock(&dentry
->d_lock
);
1480 spin_unlock(&dcache_lock
);
1481 error
= nfs_safe_remove(dentry
);
1483 nfs_renew_times(dentry
);
1484 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1485 } else if (need_rehash
)
1492 * To create a symbolic link, most file systems instantiate a new inode,
1493 * add a page to it containing the path, then write it out to the disk
1494 * using prepare_write/commit_write.
1496 * Unfortunately the NFS client can't create the in-core inode first
1497 * because it needs a file handle to create an in-core inode (see
1498 * fs/nfs/inode.c:nfs_fhget). We only have a file handle *after* the
1499 * symlink request has completed on the server.
1501 * So instead we allocate a raw page, copy the symname into it, then do
1502 * the SYMLINK request with the page as the buffer. If it succeeds, we
1503 * now have a new file handle and can instantiate an in-core NFS inode
1504 * and move the raw page into its mapping.
1506 static int nfs_symlink(struct inode
*dir
, struct dentry
*dentry
, const char *symname
)
1508 struct pagevec lru_pvec
;
1512 unsigned int pathlen
= strlen(symname
);
1515 dfprintk(VFS
, "NFS: symlink(%s/%ld, %s, %s)\n", dir
->i_sb
->s_id
,
1516 dir
->i_ino
, dentry
->d_name
.name
, symname
);
1518 if (pathlen
> PAGE_SIZE
)
1519 return -ENAMETOOLONG
;
1521 attr
.ia_mode
= S_IFLNK
| S_IRWXUGO
;
1522 attr
.ia_valid
= ATTR_MODE
;
1526 page
= alloc_page(GFP_KERNEL
);
1532 kaddr
= kmap_atomic(page
, KM_USER0
);
1533 memcpy(kaddr
, symname
, pathlen
);
1534 if (pathlen
< PAGE_SIZE
)
1535 memset(kaddr
+ pathlen
, 0, PAGE_SIZE
- pathlen
);
1536 kunmap_atomic(kaddr
, KM_USER0
);
1538 nfs_begin_data_update(dir
);
1539 error
= NFS_PROTO(dir
)->symlink(dir
, dentry
, page
, pathlen
, &attr
);
1540 nfs_end_data_update(dir
);
1542 dfprintk(VFS
, "NFS: symlink(%s/%ld, %s, %s) error %d\n",
1543 dir
->i_sb
->s_id
, dir
->i_ino
,
1544 dentry
->d_name
.name
, symname
, error
);
1552 * No big deal if we can't add this page to the page cache here.
1553 * READLINK will get the missing page from the server if needed.
1555 pagevec_init(&lru_pvec
, 0);
1556 if (!add_to_page_cache(page
, dentry
->d_inode
->i_mapping
, 0,
1558 pagevec_add(&lru_pvec
, page
);
1559 pagevec_lru_add(&lru_pvec
);
1560 SetPageUptodate(page
);
1570 nfs_link(struct dentry
*old_dentry
, struct inode
*dir
, struct dentry
*dentry
)
1572 struct inode
*inode
= old_dentry
->d_inode
;
1575 dfprintk(VFS
, "NFS: link(%s/%s -> %s/%s)\n",
1576 old_dentry
->d_parent
->d_name
.name
, old_dentry
->d_name
.name
,
1577 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
);
1580 nfs_begin_data_update(dir
);
1581 nfs_begin_data_update(inode
);
1582 error
= NFS_PROTO(dir
)->link(inode
, dir
, &dentry
->d_name
);
1584 atomic_inc(&inode
->i_count
);
1585 d_instantiate(dentry
, inode
);
1587 nfs_end_data_update(inode
);
1588 nfs_end_data_update(dir
);
1595 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a
1596 * different file handle for the same inode after a rename (e.g. when
1597 * moving to a different directory). A fail-safe method to do so would
1598 * be to look up old_dir/old_name, create a link to new_dir/new_name and
1599 * rename the old file using the sillyrename stuff. This way, the original
1600 * file in old_dir will go away when the last process iput()s the inode.
1604 * It actually works quite well. One needs to have the possibility for
1605 * at least one ".nfs..." file in each directory the file ever gets
1606 * moved or linked to which happens automagically with the new
1607 * implementation that only depends on the dcache stuff instead of
1608 * using the inode layer
1610 * Unfortunately, things are a little more complicated than indicated
1611 * above. For a cross-directory move, we want to make sure we can get
1612 * rid of the old inode after the operation. This means there must be
1613 * no pending writes (if it's a file), and the use count must be 1.
1614 * If these conditions are met, we can drop the dentries before doing
1617 static int nfs_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
1618 struct inode
*new_dir
, struct dentry
*new_dentry
)
1620 struct inode
*old_inode
= old_dentry
->d_inode
;
1621 struct inode
*new_inode
= new_dentry
->d_inode
;
1622 struct dentry
*dentry
= NULL
, *rehash
= NULL
;
1626 * To prevent any new references to the target during the rename,
1627 * we unhash the dentry and free the inode in advance.
1630 if (!d_unhashed(new_dentry
)) {
1632 rehash
= new_dentry
;
1635 dfprintk(VFS
, "NFS: rename(%s/%s -> %s/%s, ct=%d)\n",
1636 old_dentry
->d_parent
->d_name
.name
, old_dentry
->d_name
.name
,
1637 new_dentry
->d_parent
->d_name
.name
, new_dentry
->d_name
.name
,
1638 atomic_read(&new_dentry
->d_count
));
1641 * First check whether the target is busy ... we can't
1642 * safely do _any_ rename if the target is in use.
1644 * For files, make a copy of the dentry and then do a
1645 * silly-rename. If the silly-rename succeeds, the
1646 * copied dentry is hashed and becomes the new target.
1650 if (S_ISDIR(new_inode
->i_mode
)) {
1652 if (!S_ISDIR(old_inode
->i_mode
))
1654 } else if (atomic_read(&new_dentry
->d_count
) > 2) {
1656 /* copy the target dentry's name */
1657 dentry
= d_alloc(new_dentry
->d_parent
,
1658 &new_dentry
->d_name
);
1662 /* silly-rename the existing target ... */
1663 err
= nfs_sillyrename(new_dir
, new_dentry
);
1665 new_dentry
= rehash
= dentry
;
1667 /* instantiate the replacement target */
1668 d_instantiate(new_dentry
, NULL
);
1669 } else if (atomic_read(&new_dentry
->d_count
) > 1) {
1670 /* dentry still busy? */
1672 printk("nfs_rename: target %s/%s busy, d_count=%d\n",
1673 new_dentry
->d_parent
->d_name
.name
,
1674 new_dentry
->d_name
.name
,
1675 atomic_read(&new_dentry
->d_count
));
1680 drop_nlink(new_inode
);
1684 * ... prune child dentries and writebacks if needed.
1686 if (atomic_read(&old_dentry
->d_count
) > 1) {
1687 if (S_ISREG(old_inode
->i_mode
))
1688 nfs_wb_all(old_inode
);
1689 shrink_dcache_parent(old_dentry
);
1691 nfs_inode_return_delegation(old_inode
);
1693 if (new_inode
!= NULL
) {
1694 nfs_inode_return_delegation(new_inode
);
1695 d_delete(new_dentry
);
1698 nfs_begin_data_update(old_dir
);
1699 nfs_begin_data_update(new_dir
);
1700 nfs_begin_data_update(old_inode
);
1701 error
= NFS_PROTO(old_dir
)->rename(old_dir
, &old_dentry
->d_name
,
1702 new_dir
, &new_dentry
->d_name
);
1703 nfs_mark_for_revalidate(old_inode
);
1704 nfs_end_data_update(old_inode
);
1705 nfs_end_data_update(new_dir
);
1706 nfs_end_data_update(old_dir
);
1711 d_move(old_dentry
, new_dentry
);
1712 nfs_renew_times(new_dentry
);
1713 nfs_refresh_verifier(new_dentry
, nfs_save_change_attribute(new_dir
));
1716 /* new dentry created? */
1723 static DEFINE_SPINLOCK(nfs_access_lru_lock
);
1724 static LIST_HEAD(nfs_access_lru_list
);
1725 static atomic_long_t nfs_access_nr_entries
;
1727 static void nfs_access_free_entry(struct nfs_access_entry
*entry
)
1729 put_rpccred(entry
->cred
);
1731 smp_mb__before_atomic_dec();
1732 atomic_long_dec(&nfs_access_nr_entries
);
1733 smp_mb__after_atomic_dec();
1736 int nfs_access_cache_shrinker(int nr_to_scan
, gfp_t gfp_mask
)
1739 struct nfs_inode
*nfsi
;
1740 struct nfs_access_entry
*cache
;
1742 spin_lock(&nfs_access_lru_lock
);
1744 list_for_each_entry(nfsi
, &nfs_access_lru_list
, access_cache_inode_lru
) {
1745 struct inode
*inode
;
1747 if (nr_to_scan
-- == 0)
1749 inode
= igrab(&nfsi
->vfs_inode
);
1752 spin_lock(&inode
->i_lock
);
1753 if (list_empty(&nfsi
->access_cache_entry_lru
))
1754 goto remove_lru_entry
;
1755 cache
= list_entry(nfsi
->access_cache_entry_lru
.next
,
1756 struct nfs_access_entry
, lru
);
1757 list_move(&cache
->lru
, &head
);
1758 rb_erase(&cache
->rb_node
, &nfsi
->access_cache
);
1759 if (!list_empty(&nfsi
->access_cache_entry_lru
))
1760 list_move_tail(&nfsi
->access_cache_inode_lru
,
1761 &nfs_access_lru_list
);
1764 list_del_init(&nfsi
->access_cache_inode_lru
);
1765 clear_bit(NFS_INO_ACL_LRU_SET
, &nfsi
->flags
);
1767 spin_unlock(&inode
->i_lock
);
1771 spin_unlock(&nfs_access_lru_lock
);
1772 while (!list_empty(&head
)) {
1773 cache
= list_entry(head
.next
, struct nfs_access_entry
, lru
);
1774 list_del(&cache
->lru
);
1775 nfs_access_free_entry(cache
);
1777 return (atomic_long_read(&nfs_access_nr_entries
) / 100) * sysctl_vfs_cache_pressure
;
1780 static void __nfs_access_zap_cache(struct inode
*inode
)
1782 struct nfs_inode
*nfsi
= NFS_I(inode
);
1783 struct rb_root
*root_node
= &nfsi
->access_cache
;
1784 struct rb_node
*n
, *dispose
= NULL
;
1785 struct nfs_access_entry
*entry
;
1787 /* Unhook entries from the cache */
1788 while ((n
= rb_first(root_node
)) != NULL
) {
1789 entry
= rb_entry(n
, struct nfs_access_entry
, rb_node
);
1790 rb_erase(n
, root_node
);
1791 list_del(&entry
->lru
);
1792 n
->rb_left
= dispose
;
1795 nfsi
->cache_validity
&= ~NFS_INO_INVALID_ACCESS
;
1796 spin_unlock(&inode
->i_lock
);
1798 /* Now kill them all! */
1799 while (dispose
!= NULL
) {
1801 dispose
= n
->rb_left
;
1802 nfs_access_free_entry(rb_entry(n
, struct nfs_access_entry
, rb_node
));
1806 void nfs_access_zap_cache(struct inode
*inode
)
1808 /* Remove from global LRU init */
1809 if (test_and_clear_bit(NFS_INO_ACL_LRU_SET
, &NFS_FLAGS(inode
))) {
1810 spin_lock(&nfs_access_lru_lock
);
1811 list_del_init(&NFS_I(inode
)->access_cache_inode_lru
);
1812 spin_unlock(&nfs_access_lru_lock
);
1815 spin_lock(&inode
->i_lock
);
1816 /* This will release the spinlock */
1817 __nfs_access_zap_cache(inode
);
1820 static struct nfs_access_entry
*nfs_access_search_rbtree(struct inode
*inode
, struct rpc_cred
*cred
)
1822 struct rb_node
*n
= NFS_I(inode
)->access_cache
.rb_node
;
1823 struct nfs_access_entry
*entry
;
1826 entry
= rb_entry(n
, struct nfs_access_entry
, rb_node
);
1828 if (cred
< entry
->cred
)
1830 else if (cred
> entry
->cred
)
1838 int nfs_access_get_cached(struct inode
*inode
, struct rpc_cred
*cred
, struct nfs_access_entry
*res
)
1840 struct nfs_inode
*nfsi
= NFS_I(inode
);
1841 struct nfs_access_entry
*cache
;
1844 spin_lock(&inode
->i_lock
);
1845 if (nfsi
->cache_validity
& NFS_INO_INVALID_ACCESS
)
1847 cache
= nfs_access_search_rbtree(inode
, cred
);
1850 if (time_after(jiffies
, cache
->jiffies
+ NFS_ATTRTIMEO(inode
)))
1852 res
->jiffies
= cache
->jiffies
;
1853 res
->cred
= cache
->cred
;
1854 res
->mask
= cache
->mask
;
1855 list_move_tail(&cache
->lru
, &nfsi
->access_cache_entry_lru
);
1858 spin_unlock(&inode
->i_lock
);
1861 rb_erase(&cache
->rb_node
, &nfsi
->access_cache
);
1862 list_del(&cache
->lru
);
1863 spin_unlock(&inode
->i_lock
);
1864 nfs_access_free_entry(cache
);
1867 /* This will release the spinlock */
1868 __nfs_access_zap_cache(inode
);
1872 static void nfs_access_add_rbtree(struct inode
*inode
, struct nfs_access_entry
*set
)
1874 struct nfs_inode
*nfsi
= NFS_I(inode
);
1875 struct rb_root
*root_node
= &nfsi
->access_cache
;
1876 struct rb_node
**p
= &root_node
->rb_node
;
1877 struct rb_node
*parent
= NULL
;
1878 struct nfs_access_entry
*entry
;
1880 spin_lock(&inode
->i_lock
);
1881 while (*p
!= NULL
) {
1883 entry
= rb_entry(parent
, struct nfs_access_entry
, rb_node
);
1885 if (set
->cred
< entry
->cred
)
1886 p
= &parent
->rb_left
;
1887 else if (set
->cred
> entry
->cred
)
1888 p
= &parent
->rb_right
;
1892 rb_link_node(&set
->rb_node
, parent
, p
);
1893 rb_insert_color(&set
->rb_node
, root_node
);
1894 list_add_tail(&set
->lru
, &nfsi
->access_cache_entry_lru
);
1895 spin_unlock(&inode
->i_lock
);
1898 rb_replace_node(parent
, &set
->rb_node
, root_node
);
1899 list_add_tail(&set
->lru
, &nfsi
->access_cache_entry_lru
);
1900 list_del(&entry
->lru
);
1901 spin_unlock(&inode
->i_lock
);
1902 nfs_access_free_entry(entry
);
1905 void nfs_access_add_cache(struct inode
*inode
, struct nfs_access_entry
*set
)
1907 struct nfs_access_entry
*cache
= kmalloc(sizeof(*cache
), GFP_KERNEL
);
1910 RB_CLEAR_NODE(&cache
->rb_node
);
1911 cache
->jiffies
= set
->jiffies
;
1912 cache
->cred
= get_rpccred(set
->cred
);
1913 cache
->mask
= set
->mask
;
1915 nfs_access_add_rbtree(inode
, cache
);
1917 /* Update accounting */
1918 smp_mb__before_atomic_inc();
1919 atomic_long_inc(&nfs_access_nr_entries
);
1920 smp_mb__after_atomic_inc();
1922 /* Add inode to global LRU list */
1923 if (!test_and_set_bit(NFS_INO_ACL_LRU_SET
, &NFS_FLAGS(inode
))) {
1924 spin_lock(&nfs_access_lru_lock
);
1925 list_add_tail(&NFS_I(inode
)->access_cache_inode_lru
, &nfs_access_lru_list
);
1926 spin_unlock(&nfs_access_lru_lock
);
1930 static int nfs_do_access(struct inode
*inode
, struct rpc_cred
*cred
, int mask
)
1932 struct nfs_access_entry cache
;
1935 status
= nfs_access_get_cached(inode
, cred
, &cache
);
1939 /* Be clever: ask server to check for all possible rights */
1940 cache
.mask
= MAY_EXEC
| MAY_WRITE
| MAY_READ
;
1942 cache
.jiffies
= jiffies
;
1943 status
= NFS_PROTO(inode
)->access(inode
, &cache
);
1946 nfs_access_add_cache(inode
, &cache
);
1948 if ((cache
.mask
& mask
) == mask
)
1953 int nfs_permission(struct inode
*inode
, int mask
, struct nameidata
*nd
)
1955 struct rpc_cred
*cred
;
1958 nfs_inc_stats(inode
, NFSIOS_VFSACCESS
);
1962 /* Is this sys_access() ? */
1963 if (nd
!= NULL
&& (nd
->flags
& LOOKUP_ACCESS
))
1966 switch (inode
->i_mode
& S_IFMT
) {
1970 /* NFSv4 has atomic_open... */
1971 if (nfs_server_capable(inode
, NFS_CAP_ATOMIC_OPEN
)
1973 && (nd
->flags
& LOOKUP_OPEN
))
1978 * Optimize away all write operations, since the server
1979 * will check permissions when we perform the op.
1981 if ((mask
& MAY_WRITE
) && !(mask
& MAY_READ
))
1988 if (!NFS_PROTO(inode
)->access
)
1991 cred
= rpcauth_lookupcred(NFS_CLIENT(inode
)->cl_auth
, 0);
1992 if (!IS_ERR(cred
)) {
1993 res
= nfs_do_access(inode
, cred
, mask
);
1996 res
= PTR_ERR(cred
);
1999 dfprintk(VFS
, "NFS: permission(%s/%ld), mask=0x%x, res=%d\n",
2000 inode
->i_sb
->s_id
, inode
->i_ino
, mask
, res
);
2003 res
= nfs_revalidate_inode(NFS_SERVER(inode
), inode
);
2005 res
= generic_permission(inode
, mask
, NULL
);
2012 * version-control: t
2013 * kept-new-versions: 5