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/namei.h>
35 #include "delegation.h"
37 #define NFS_PARANOIA 1
38 /* #define NFS_DEBUG_VERBOSE 1 */
40 static int nfs_opendir(struct inode
*, struct file
*);
41 static int nfs_readdir(struct file
*, void *, filldir_t
);
42 static struct dentry
*nfs_lookup(struct inode
*, struct dentry
*, struct nameidata
*);
43 static int nfs_create(struct inode
*, struct dentry
*, int, struct nameidata
*);
44 static int nfs_mkdir(struct inode
*, struct dentry
*, int);
45 static int nfs_rmdir(struct inode
*, struct dentry
*);
46 static int nfs_unlink(struct inode
*, struct dentry
*);
47 static int nfs_symlink(struct inode
*, struct dentry
*, const char *);
48 static int nfs_link(struct dentry
*, struct inode
*, struct dentry
*);
49 static int nfs_mknod(struct inode
*, struct dentry
*, int, dev_t
);
50 static int nfs_rename(struct inode
*, struct dentry
*,
51 struct inode
*, struct dentry
*);
52 static int nfs_fsync_dir(struct file
*, struct dentry
*, int);
54 struct file_operations nfs_dir_operations
= {
55 .read
= generic_read_dir
,
56 .readdir
= nfs_readdir
,
58 .release
= nfs_release
,
59 .fsync
= nfs_fsync_dir
,
62 struct inode_operations nfs_dir_inode_operations
= {
67 .symlink
= nfs_symlink
,
72 .permission
= nfs_permission
,
73 .getattr
= nfs_getattr
,
74 .setattr
= nfs_setattr
,
79 static struct dentry
*nfs_atomic_lookup(struct inode
*, struct dentry
*, struct nameidata
*);
80 struct inode_operations nfs4_dir_inode_operations
= {
82 .lookup
= nfs_atomic_lookup
,
85 .symlink
= nfs_symlink
,
90 .permission
= nfs_permission
,
91 .getattr
= nfs_getattr
,
92 .setattr
= nfs_setattr
,
95 #endif /* CONFIG_NFS_V4 */
101 nfs_opendir(struct inode
*inode
, struct file
*filp
)
106 /* Call generic open code in order to cache credentials */
108 res
= nfs_open(inode
, filp
);
113 typedef u32
* (*decode_dirent_t
)(u32
*, struct nfs_entry
*, int);
117 unsigned long page_index
;
120 struct nfs_entry
*entry
;
121 decode_dirent_t decode
;
124 } nfs_readdir_descriptor_t
;
126 /* Now we cache directories properly, by stuffing the dirent
127 * data directly in the page cache.
129 * Inode invalidation due to refresh etc. takes care of
130 * _everything_, no sloppy entry flushing logic, no extraneous
131 * copying, network direct to page cache, the way it was meant
134 * NOTE: Dirent information verification is done always by the
135 * page-in of the RPC reply, nowhere else, this simplies
136 * things substantially.
139 int nfs_readdir_filler(nfs_readdir_descriptor_t
*desc
, struct page
*page
)
141 struct file
*file
= desc
->file
;
142 struct inode
*inode
= file
->f_dentry
->d_inode
;
143 struct rpc_cred
*cred
= nfs_file_cred(file
);
144 unsigned long timestamp
;
147 dfprintk(VFS
, "NFS: nfs_readdir_filler() reading cookie %Lu into page %lu.\n", (long long)desc
->entry
->cookie
, page
->index
);
151 error
= NFS_PROTO(inode
)->readdir(file
->f_dentry
, cred
, desc
->entry
->cookie
, page
,
152 NFS_SERVER(inode
)->dtsize
, desc
->plus
);
154 /* We requested READDIRPLUS, but the server doesn't grok it */
155 if (error
== -ENOTSUPP
&& desc
->plus
) {
156 NFS_SERVER(inode
)->caps
&= ~NFS_CAP_READDIRPLUS
;
157 NFS_FLAGS(inode
) &= ~NFS_INO_ADVISE_RDPLUS
;
163 SetPageUptodate(page
);
164 NFS_FLAGS(inode
) |= NFS_INO_INVALID_ATIME
;
165 /* Ensure consistent page alignment of the data.
166 * Note: assumes we have exclusive access to this mapping either
167 * throught inode->i_sem or some other mechanism.
169 if (page
->index
== 0) {
170 invalidate_inode_pages(inode
->i_mapping
);
171 NFS_I(inode
)->readdir_timestamp
= timestamp
;
178 nfs_zap_caches(inode
);
184 int dir_decode(nfs_readdir_descriptor_t
*desc
)
187 p
= desc
->decode(p
, desc
->entry
, desc
->plus
);
195 void dir_page_release(nfs_readdir_descriptor_t
*desc
)
198 page_cache_release(desc
->page
);
204 * Given a pointer to a buffer that has already been filled by a call
205 * to readdir, find the next entry.
207 * If the end of the buffer has been reached, return -EAGAIN, if not,
208 * return the offset within the buffer of the next entry to be
212 int find_dirent(nfs_readdir_descriptor_t
*desc
, struct page
*page
)
214 struct nfs_entry
*entry
= desc
->entry
;
218 while((status
= dir_decode(desc
)) == 0) {
219 dfprintk(VFS
, "NFS: found cookie %Lu\n", (long long)entry
->cookie
);
220 if (entry
->prev_cookie
== desc
->target
)
222 if (loop_count
++ > 200) {
227 dfprintk(VFS
, "NFS: find_dirent() returns %d\n", status
);
232 * Find the given page, and call find_dirent() in order to try to
233 * return the next entry.
236 int find_dirent_page(nfs_readdir_descriptor_t
*desc
)
238 struct inode
*inode
= desc
->file
->f_dentry
->d_inode
;
242 dfprintk(VFS
, "NFS: find_dirent_page() searching directory page %ld\n", desc
->page_index
);
244 page
= read_cache_page(inode
->i_mapping
, desc
->page_index
,
245 (filler_t
*)nfs_readdir_filler
, desc
);
247 status
= PTR_ERR(page
);
250 if (!PageUptodate(page
))
253 /* NOTE: Someone else may have changed the READDIRPLUS flag */
255 desc
->ptr
= kmap(page
); /* matching kunmap in nfs_do_filldir */
256 status
= find_dirent(desc
, page
);
258 dir_page_release(desc
);
260 dfprintk(VFS
, "NFS: find_dirent_page() returns %d\n", status
);
263 page_cache_release(page
);
268 * Recurse through the page cache pages, and return a
269 * filled nfs_entry structure of the next directory entry if possible.
271 * The target for the search is 'desc->target'.
274 int readdir_search_pagecache(nfs_readdir_descriptor_t
*desc
)
279 dfprintk(VFS
, "NFS: readdir_search_pagecache() searching for cookie %Lu\n", (long long)desc
->target
);
281 res
= find_dirent_page(desc
);
284 /* Align to beginning of next page */
286 if (loop_count
++ > 200) {
291 dfprintk(VFS
, "NFS: readdir_search_pagecache() returned %d\n", res
);
295 static inline unsigned int dt_type(struct inode
*inode
)
297 return (inode
->i_mode
>> 12) & 15;
300 static struct dentry
*nfs_readdir_lookup(nfs_readdir_descriptor_t
*desc
);
303 * Once we've found the start of the dirent within a page: fill 'er up...
306 int nfs_do_filldir(nfs_readdir_descriptor_t
*desc
, void *dirent
,
309 struct file
*file
= desc
->file
;
310 struct nfs_entry
*entry
= desc
->entry
;
311 struct dentry
*dentry
= NULL
;
312 unsigned long fileid
;
316 dfprintk(VFS
, "NFS: nfs_do_filldir() filling starting @ cookie %Lu\n", (long long)desc
->target
);
319 unsigned d_type
= DT_UNKNOWN
;
320 /* Note: entry->prev_cookie contains the cookie for
321 * retrieving the current dirent on the server */
322 fileid
= nfs_fileid_to_ino_t(entry
->ino
);
324 /* Get a dentry if we have one */
327 dentry
= nfs_readdir_lookup(desc
);
329 /* Use readdirplus info */
330 if (dentry
!= NULL
&& dentry
->d_inode
!= NULL
) {
331 d_type
= dt_type(dentry
->d_inode
);
332 fileid
= dentry
->d_inode
->i_ino
;
335 res
= filldir(dirent
, entry
->name
, entry
->len
,
336 entry
->prev_cookie
, fileid
, d_type
);
339 file
->f_pos
= desc
->target
= entry
->cookie
;
340 if (dir_decode(desc
) != 0) {
344 if (loop_count
++ > 200) {
349 dir_page_release(desc
);
352 dfprintk(VFS
, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n", (long long)desc
->target
, res
);
357 * If we cannot find a cookie in our cache, we suspect that this is
358 * because it points to a deleted file, so we ask the server to return
359 * whatever it thinks is the next entry. We then feed this to filldir.
360 * If all goes well, we should then be able to find our way round the
361 * cache on the next call to readdir_search_pagecache();
363 * NOTE: we cannot add the anonymous page to the pagecache because
364 * the data it contains might not be page aligned. Besides,
365 * we should already have a complete representation of the
366 * directory in the page cache by the time we get here.
369 int uncached_readdir(nfs_readdir_descriptor_t
*desc
, void *dirent
,
372 struct file
*file
= desc
->file
;
373 struct inode
*inode
= file
->f_dentry
->d_inode
;
374 struct rpc_cred
*cred
= nfs_file_cred(file
);
375 struct page
*page
= NULL
;
378 dfprintk(VFS
, "NFS: uncached_readdir() searching for cookie %Lu\n", (long long)desc
->target
);
380 page
= alloc_page(GFP_HIGHUSER
);
385 desc
->error
= NFS_PROTO(inode
)->readdir(file
->f_dentry
, cred
, desc
->target
,
387 NFS_SERVER(inode
)->dtsize
,
389 NFS_FLAGS(inode
) |= NFS_INO_INVALID_ATIME
;
391 desc
->ptr
= kmap(page
); /* matching kunmap in nfs_do_filldir */
392 if (desc
->error
>= 0) {
393 if ((status
= dir_decode(desc
)) == 0)
394 desc
->entry
->prev_cookie
= desc
->target
;
400 status
= nfs_do_filldir(desc
, dirent
, filldir
);
402 /* Reset read descriptor so it searches the page cache from
403 * the start upon the next call to readdir_search_pagecache() */
404 desc
->page_index
= 0;
405 desc
->entry
->cookie
= desc
->entry
->prev_cookie
= 0;
406 desc
->entry
->eof
= 0;
408 dfprintk(VFS
, "NFS: uncached_readdir() returns %d\n", status
);
411 dir_page_release(desc
);
415 /* The file offset position is now represented as a true offset into the
416 * page cache as is the case in most of the other filesystems.
418 static int nfs_readdir(struct file
*filp
, void *dirent
, filldir_t filldir
)
420 struct dentry
*dentry
= filp
->f_dentry
;
421 struct inode
*inode
= dentry
->d_inode
;
422 nfs_readdir_descriptor_t my_desc
,
424 struct nfs_entry my_entry
;
426 struct nfs_fattr fattr
;
431 res
= nfs_revalidate_inode(NFS_SERVER(inode
), inode
);
438 * filp->f_pos points to the file offset in the page cache.
439 * but if the cache has meanwhile been zapped, we need to
440 * read from the last dirent to revalidate f_pos
443 memset(desc
, 0, sizeof(*desc
));
446 desc
->target
= filp
->f_pos
;
447 desc
->decode
= NFS_PROTO(inode
)->decode_dirent
;
448 desc
->plus
= NFS_USE_READDIRPLUS(inode
);
450 my_entry
.cookie
= my_entry
.prev_cookie
= 0;
453 my_entry
.fattr
= &fattr
;
454 desc
->entry
= &my_entry
;
456 while(!desc
->entry
->eof
) {
457 res
= readdir_search_pagecache(desc
);
458 if (res
== -EBADCOOKIE
) {
459 /* This means either end of directory */
460 if (desc
->entry
->cookie
!= desc
->target
) {
461 /* Or that the server has 'lost' a cookie */
462 res
= uncached_readdir(desc
, dirent
, filldir
);
469 if (res
== -ETOOSMALL
&& desc
->plus
) {
470 NFS_FLAGS(inode
) &= ~NFS_INO_ADVISE_RDPLUS
;
471 nfs_zap_caches(inode
);
473 desc
->entry
->eof
= 0;
479 res
= nfs_do_filldir(desc
, dirent
, filldir
);
494 * All directory operations under NFS are synchronous, so fsync()
495 * is a dummy operation.
497 int nfs_fsync_dir(struct file
*filp
, struct dentry
*dentry
, int datasync
)
503 * A check for whether or not the parent directory has changed.
504 * In the case it has, we assume that the dentries are untrustworthy
505 * and may need to be looked up again.
507 static inline int nfs_check_verifier(struct inode
*dir
, struct dentry
*dentry
)
511 if ((NFS_FLAGS(dir
) & NFS_INO_INVALID_ATTR
) != 0
512 || nfs_attribute_timeout(dir
))
514 return nfs_verify_change_attribute(dir
, (unsigned long)dentry
->d_fsdata
);
517 static inline void nfs_set_verifier(struct dentry
* dentry
, unsigned long verf
)
519 dentry
->d_fsdata
= (void *)verf
;
523 * Whenever an NFS operation succeeds, we know that the dentry
524 * is valid, so we update the revalidation timestamp.
526 static inline void nfs_renew_times(struct dentry
* dentry
)
528 dentry
->d_time
= jiffies
;
532 int nfs_lookup_verify_inode(struct inode
*inode
, struct nameidata
*nd
)
534 struct nfs_server
*server
= NFS_SERVER(inode
);
537 int ndflags
= nd
->flags
;
538 /* VFS wants an on-the-wire revalidation */
539 if (ndflags
& LOOKUP_REVAL
)
541 /* This is an open(2) */
542 if ((ndflags
& LOOKUP_OPEN
) &&
543 !(ndflags
& LOOKUP_CONTINUE
) &&
544 !(server
->flags
& NFS_MOUNT_NOCTO
))
547 return nfs_revalidate_inode(server
, inode
);
549 return __nfs_revalidate_inode(server
, inode
);
553 * We judge how long we want to trust negative
554 * dentries by looking at the parent inode mtime.
556 * If parent mtime has changed, we revalidate, else we wait for a
557 * period corresponding to the parent's attribute cache timeout value.
560 int nfs_neg_need_reval(struct inode
*dir
, struct dentry
*dentry
,
561 struct nameidata
*nd
)
567 /* Don't revalidate a negative dentry if we're creating a new file */
568 if ((ndflags
& LOOKUP_CREATE
) && !(ndflags
& LOOKUP_CONTINUE
))
570 return !nfs_check_verifier(dir
, dentry
);
574 * This is called every time the dcache has a lookup hit,
575 * and we should check whether we can really trust that
578 * NOTE! The hit can be a negative hit too, don't assume
581 * If the parent directory is seen to have changed, we throw out the
582 * cached dentry and do a new lookup.
584 static int nfs_lookup_revalidate(struct dentry
* dentry
, struct nameidata
*nd
)
588 struct dentry
*parent
;
590 struct nfs_fh fhandle
;
591 struct nfs_fattr fattr
;
592 unsigned long verifier
;
594 parent
= dget_parent(dentry
);
596 dir
= parent
->d_inode
;
597 inode
= dentry
->d_inode
;
600 if (nfs_neg_need_reval(dir
, dentry
, nd
))
605 if (is_bad_inode(inode
)) {
606 dfprintk(VFS
, "nfs_lookup_validate: %s/%s has dud inode\n",
607 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
);
611 /* Revalidate parent directory attribute cache */
612 if (nfs_revalidate_inode(NFS_SERVER(dir
), dir
) < 0)
615 /* Force a full look up iff the parent directory has changed */
616 if (nfs_check_verifier(dir
, dentry
)) {
617 if (nfs_lookup_verify_inode(inode
, nd
))
622 if (NFS_STALE(inode
))
625 verifier
= nfs_save_change_attribute(dir
);
626 error
= NFS_PROTO(dir
)->lookup(dir
, &dentry
->d_name
, &fhandle
, &fattr
);
629 if (nfs_compare_fh(NFS_FH(inode
), &fhandle
))
631 if ((error
= nfs_refresh_inode(inode
, &fattr
)) != 0)
634 nfs_renew_times(dentry
);
635 nfs_set_verifier(dentry
, verifier
);
644 if (inode
&& S_ISDIR(inode
->i_mode
)) {
645 /* Purge readdir caches. */
646 nfs_zap_caches(inode
);
647 /* If we have submounts, don't unhash ! */
648 if (have_submounts(dentry
))
650 shrink_dcache_parent(dentry
);
659 * This is called from dput() when d_count is going to 0.
661 static int nfs_dentry_delete(struct dentry
*dentry
)
663 dfprintk(VFS
, "NFS: dentry_delete(%s/%s, %x)\n",
664 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
,
667 if (dentry
->d_flags
& DCACHE_NFSFS_RENAMED
) {
668 /* Unhash it, so that ->d_iput() would be called */
671 if (!(dentry
->d_sb
->s_flags
& MS_ACTIVE
)) {
672 /* Unhash it, so that ancestors of killed async unlink
673 * files will be cleaned up during umount */
681 * Called when the dentry loses inode.
682 * We use it to clean up silly-renamed files.
684 static void nfs_dentry_iput(struct dentry
*dentry
, struct inode
*inode
)
686 if (dentry
->d_flags
& DCACHE_NFSFS_RENAMED
) {
689 nfs_complete_unlink(dentry
);
692 /* When creating a negative dentry, we want to renew d_time */
693 nfs_renew_times(dentry
);
697 struct dentry_operations nfs_dentry_operations
= {
698 .d_revalidate
= nfs_lookup_revalidate
,
699 .d_delete
= nfs_dentry_delete
,
700 .d_iput
= nfs_dentry_iput
,
704 int nfs_is_exclusive_create(struct inode
*dir
, struct nameidata
*nd
)
706 if (NFS_PROTO(dir
)->version
== 2)
708 if (!nd
|| (nd
->flags
& LOOKUP_CONTINUE
) || !(nd
->flags
& LOOKUP_CREATE
))
710 return (nd
->intent
.open
.flags
& O_EXCL
) != 0;
713 static struct dentry
*nfs_lookup(struct inode
*dir
, struct dentry
* dentry
, struct nameidata
*nd
)
716 struct inode
*inode
= NULL
;
718 struct nfs_fh fhandle
;
719 struct nfs_fattr fattr
;
721 dfprintk(VFS
, "NFS: lookup(%s/%s)\n",
722 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
);
724 res
= ERR_PTR(-ENAMETOOLONG
);
725 if (dentry
->d_name
.len
> NFS_SERVER(dir
)->namelen
)
728 res
= ERR_PTR(-ENOMEM
);
729 dentry
->d_op
= NFS_PROTO(dir
)->dentry_ops
;
732 /* Revalidate parent directory attribute cache */
733 error
= nfs_revalidate_inode(NFS_SERVER(dir
), dir
);
735 res
= ERR_PTR(error
);
739 /* If we're doing an exclusive create, optimize away the lookup */
740 if (nfs_is_exclusive_create(dir
, nd
))
743 error
= NFS_PROTO(dir
)->lookup(dir
, &dentry
->d_name
, &fhandle
, &fattr
);
744 if (error
== -ENOENT
)
747 res
= ERR_PTR(error
);
750 res
= ERR_PTR(-EACCES
);
751 inode
= nfs_fhget(dentry
->d_sb
, &fhandle
, &fattr
);
755 res
= d_add_unique(dentry
, inode
);
758 nfs_renew_times(dentry
);
759 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
767 static int nfs_open_revalidate(struct dentry
*, struct nameidata
*);
769 struct dentry_operations nfs4_dentry_operations
= {
770 .d_revalidate
= nfs_open_revalidate
,
771 .d_delete
= nfs_dentry_delete
,
772 .d_iput
= nfs_dentry_iput
,
775 static int is_atomic_open(struct inode
*dir
, struct nameidata
*nd
)
779 /* Check that we are indeed trying to open this file */
780 if ((nd
->flags
& LOOKUP_CONTINUE
) || !(nd
->flags
& LOOKUP_OPEN
))
782 /* NFS does not (yet) have a stateful open for directories */
783 if (nd
->flags
& LOOKUP_DIRECTORY
)
785 /* Are we trying to write to a read only partition? */
786 if (IS_RDONLY(dir
) && (nd
->intent
.open
.flags
& (O_CREAT
|O_TRUNC
|FMODE_WRITE
)))
791 static struct dentry
*nfs_atomic_lookup(struct inode
*dir
, struct dentry
*dentry
, struct nameidata
*nd
)
793 struct dentry
*res
= NULL
;
794 struct inode
*inode
= NULL
;
797 /* Check that we are indeed trying to open this file */
798 if (!is_atomic_open(dir
, nd
))
801 if (dentry
->d_name
.len
> NFS_SERVER(dir
)->namelen
) {
802 res
= ERR_PTR(-ENAMETOOLONG
);
805 dentry
->d_op
= NFS_PROTO(dir
)->dentry_ops
;
807 /* Let vfs_create() deal with O_EXCL */
808 if (nd
->intent
.open
.flags
& O_EXCL
)
811 /* Open the file on the server */
813 /* Revalidate parent directory attribute cache */
814 error
= nfs_revalidate_inode(NFS_SERVER(dir
), dir
);
816 res
= ERR_PTR(error
);
820 if (nd
->intent
.open
.flags
& O_CREAT
) {
821 nfs_begin_data_update(dir
);
822 inode
= nfs4_atomic_open(dir
, dentry
, nd
);
823 nfs_end_data_update(dir
);
825 inode
= nfs4_atomic_open(dir
, dentry
, nd
);
828 error
= PTR_ERR(inode
);
830 /* Make a negative dentry */
834 /* This turned out not to be a regular file */
836 if (!(nd
->intent
.open
.flags
& O_NOFOLLOW
))
841 res
= ERR_PTR(error
);
846 res
= d_add_unique(dentry
, inode
);
849 nfs_renew_times(dentry
);
850 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
854 return nfs_lookup(dir
, dentry
, nd
);
857 static int nfs_open_revalidate(struct dentry
*dentry
, struct nameidata
*nd
)
859 struct dentry
*parent
= NULL
;
860 struct inode
*inode
= dentry
->d_inode
;
862 unsigned long verifier
;
863 int openflags
, ret
= 0;
865 parent
= dget_parent(dentry
);
866 dir
= parent
->d_inode
;
867 if (!is_atomic_open(dir
, nd
))
869 /* We can't create new files in nfs_open_revalidate(), so we
870 * optimize away revalidation of negative dentries.
874 /* NFS only supports OPEN on regular files */
875 if (!S_ISREG(inode
->i_mode
))
877 openflags
= nd
->intent
.open
.flags
;
878 /* We cannot do exclusive creation on a positive dentry */
879 if ((openflags
& (O_CREAT
|O_EXCL
)) == (O_CREAT
|O_EXCL
))
881 /* We can't create new files, or truncate existing ones here */
882 openflags
&= ~(O_CREAT
|O_TRUNC
);
885 * Note: we're not holding inode->i_sem and so may be racing with
886 * operations that change the directory. We therefore save the
887 * change attribute *before* we do the RPC call.
890 verifier
= nfs_save_change_attribute(dir
);
891 ret
= nfs4_open_revalidate(dir
, dentry
, openflags
);
893 nfs_set_verifier(dentry
, verifier
);
902 if (inode
!= NULL
&& nfs_have_delegation(inode
, FMODE_READ
))
904 return nfs_lookup_revalidate(dentry
, nd
);
906 #endif /* CONFIG_NFSV4 */
908 static struct dentry
*nfs_readdir_lookup(nfs_readdir_descriptor_t
*desc
)
910 struct dentry
*parent
= desc
->file
->f_dentry
;
911 struct inode
*dir
= parent
->d_inode
;
912 struct nfs_entry
*entry
= desc
->entry
;
913 struct dentry
*dentry
, *alias
;
922 if (name
.name
[0] == '.' && name
.name
[1] == '.')
923 return dget_parent(parent
);
926 if (name
.name
[0] == '.')
929 name
.hash
= full_name_hash(name
.name
, name
.len
);
930 dentry
= d_lookup(parent
, &name
);
933 if (!desc
->plus
|| !(entry
->fattr
->valid
& NFS_ATTR_FATTR
))
935 /* Note: caller is already holding the dir->i_sem! */
936 dentry
= d_alloc(parent
, &name
);
939 dentry
->d_op
= NFS_PROTO(dir
)->dentry_ops
;
940 inode
= nfs_fhget(dentry
->d_sb
, entry
->fh
, entry
->fattr
);
945 alias
= d_add_unique(dentry
, inode
);
950 nfs_renew_times(dentry
);
951 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
956 * Code common to create, mkdir, and mknod.
958 int nfs_instantiate(struct dentry
*dentry
, struct nfs_fh
*fhandle
,
959 struct nfs_fattr
*fattr
)
964 /* We may have been initialized further down */
967 if (fhandle
->size
== 0) {
968 struct inode
*dir
= dentry
->d_parent
->d_inode
;
969 error
= NFS_PROTO(dir
)->lookup(dir
, &dentry
->d_name
, fhandle
, fattr
);
973 if (!(fattr
->valid
& NFS_ATTR_FATTR
)) {
974 struct nfs_server
*server
= NFS_SB(dentry
->d_sb
);
975 error
= server
->rpc_ops
->getattr(server
, fhandle
, fattr
);
980 inode
= nfs_fhget(dentry
->d_sb
, fhandle
, fattr
);
983 d_instantiate(dentry
, inode
);
991 * Following a failed create operation, we drop the dentry rather
992 * than retain a negative dentry. This avoids a problem in the event
993 * that the operation succeeded on the server, but an error in the
994 * reply path made it appear to have failed.
996 static int nfs_create(struct inode
*dir
, struct dentry
*dentry
, int mode
,
997 struct nameidata
*nd
)
1003 dfprintk(VFS
, "NFS: create(%s/%ld, %s\n", dir
->i_sb
->s_id
,
1004 dir
->i_ino
, dentry
->d_name
.name
);
1006 attr
.ia_mode
= mode
;
1007 attr
.ia_valid
= ATTR_MODE
;
1009 if (nd
&& (nd
->flags
& LOOKUP_CREATE
))
1010 open_flags
= nd
->intent
.open
.flags
;
1013 nfs_begin_data_update(dir
);
1014 error
= NFS_PROTO(dir
)->create(dir
, dentry
, &attr
, open_flags
);
1015 nfs_end_data_update(dir
);
1018 nfs_renew_times(dentry
);
1019 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1029 * See comments for nfs_proc_create regarding failed operations.
1032 nfs_mknod(struct inode
*dir
, struct dentry
*dentry
, int mode
, dev_t rdev
)
1037 dfprintk(VFS
, "NFS: mknod(%s/%ld, %s\n", dir
->i_sb
->s_id
,
1038 dir
->i_ino
, dentry
->d_name
.name
);
1040 if (!new_valid_dev(rdev
))
1043 attr
.ia_mode
= mode
;
1044 attr
.ia_valid
= ATTR_MODE
;
1047 nfs_begin_data_update(dir
);
1048 status
= NFS_PROTO(dir
)->mknod(dir
, dentry
, &attr
, rdev
);
1049 nfs_end_data_update(dir
);
1052 nfs_renew_times(dentry
);
1053 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1063 * See comments for nfs_proc_create regarding failed operations.
1065 static int nfs_mkdir(struct inode
*dir
, struct dentry
*dentry
, int mode
)
1070 dfprintk(VFS
, "NFS: mkdir(%s/%ld, %s\n", dir
->i_sb
->s_id
,
1071 dir
->i_ino
, dentry
->d_name
.name
);
1073 attr
.ia_valid
= ATTR_MODE
;
1074 attr
.ia_mode
= mode
| S_IFDIR
;
1077 nfs_begin_data_update(dir
);
1078 error
= NFS_PROTO(dir
)->mkdir(dir
, dentry
, &attr
);
1079 nfs_end_data_update(dir
);
1082 nfs_renew_times(dentry
);
1083 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1092 static int nfs_rmdir(struct inode
*dir
, struct dentry
*dentry
)
1096 dfprintk(VFS
, "NFS: rmdir(%s/%ld, %s\n", dir
->i_sb
->s_id
,
1097 dir
->i_ino
, dentry
->d_name
.name
);
1100 nfs_begin_data_update(dir
);
1101 error
= NFS_PROTO(dir
)->rmdir(dir
, &dentry
->d_name
);
1102 /* Ensure the VFS deletes this inode */
1103 if (error
== 0 && dentry
->d_inode
!= NULL
)
1104 dentry
->d_inode
->i_nlink
= 0;
1105 nfs_end_data_update(dir
);
1111 static int nfs_sillyrename(struct inode
*dir
, struct dentry
*dentry
)
1113 static unsigned int sillycounter
;
1114 const int i_inosize
= sizeof(dir
->i_ino
)*2;
1115 const int countersize
= sizeof(sillycounter
)*2;
1116 const int slen
= sizeof(".nfs") + i_inosize
+ countersize
- 1;
1119 struct dentry
*sdentry
;
1122 dfprintk(VFS
, "NFS: silly-rename(%s/%s, ct=%d)\n",
1123 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
,
1124 atomic_read(&dentry
->d_count
));
1127 if (!dentry
->d_inode
)
1128 printk("NFS: silly-renaming %s/%s, negative dentry??\n",
1129 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
);
1132 * We don't allow a dentry to be silly-renamed twice.
1135 if (dentry
->d_flags
& DCACHE_NFSFS_RENAMED
)
1138 sprintf(silly
, ".nfs%*.*lx",
1139 i_inosize
, i_inosize
, dentry
->d_inode
->i_ino
);
1143 char *suffix
= silly
+ slen
- countersize
;
1147 sprintf(suffix
, "%*.*x", countersize
, countersize
, sillycounter
);
1149 dfprintk(VFS
, "trying to rename %s to %s\n",
1150 dentry
->d_name
.name
, silly
);
1152 sdentry
= lookup_one_len(silly
, dentry
->d_parent
, slen
);
1154 * N.B. Better to return EBUSY here ... it could be
1155 * dangerous to delete the file while it's in use.
1157 if (IS_ERR(sdentry
))
1159 } while(sdentry
->d_inode
!= NULL
); /* need negative lookup */
1161 qsilly
.name
= silly
;
1162 qsilly
.len
= strlen(silly
);
1163 nfs_begin_data_update(dir
);
1164 if (dentry
->d_inode
) {
1165 nfs_begin_data_update(dentry
->d_inode
);
1166 error
= NFS_PROTO(dir
)->rename(dir
, &dentry
->d_name
,
1168 nfs_end_data_update(dentry
->d_inode
);
1170 error
= NFS_PROTO(dir
)->rename(dir
, &dentry
->d_name
,
1172 nfs_end_data_update(dir
);
1174 nfs_renew_times(dentry
);
1175 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1176 d_move(dentry
, sdentry
);
1177 error
= nfs_async_unlink(dentry
);
1178 /* If we return 0 we don't unlink */
1186 * Remove a file after making sure there are no pending writes,
1187 * and after checking that the file has only one user.
1189 * We invalidate the attribute cache and free the inode prior to the operation
1190 * to avoid possible races if the server reuses the inode.
1192 static int nfs_safe_remove(struct dentry
*dentry
)
1194 struct inode
*dir
= dentry
->d_parent
->d_inode
;
1195 struct inode
*inode
= dentry
->d_inode
;
1198 dfprintk(VFS
, "NFS: safe_remove(%s/%s)\n",
1199 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
);
1201 /* If the dentry was sillyrenamed, we simply call d_delete() */
1202 if (dentry
->d_flags
& DCACHE_NFSFS_RENAMED
) {
1207 nfs_begin_data_update(dir
);
1208 if (inode
!= NULL
) {
1209 nfs_begin_data_update(inode
);
1210 error
= NFS_PROTO(dir
)->remove(dir
, &dentry
->d_name
);
1211 /* The VFS may want to delete this inode */
1214 nfs_end_data_update(inode
);
1216 error
= NFS_PROTO(dir
)->remove(dir
, &dentry
->d_name
);
1217 nfs_end_data_update(dir
);
1222 /* We do silly rename. In case sillyrename() returns -EBUSY, the inode
1223 * belongs to an active ".nfs..." file and we return -EBUSY.
1225 * If sillyrename() returns 0, we do nothing, otherwise we unlink.
1227 static int nfs_unlink(struct inode
*dir
, struct dentry
*dentry
)
1230 int need_rehash
= 0;
1232 dfprintk(VFS
, "NFS: unlink(%s/%ld, %s)\n", dir
->i_sb
->s_id
,
1233 dir
->i_ino
, dentry
->d_name
.name
);
1236 spin_lock(&dcache_lock
);
1237 spin_lock(&dentry
->d_lock
);
1238 if (atomic_read(&dentry
->d_count
) > 1) {
1239 spin_unlock(&dentry
->d_lock
);
1240 spin_unlock(&dcache_lock
);
1241 error
= nfs_sillyrename(dir
, dentry
);
1245 if (!d_unhashed(dentry
)) {
1249 spin_unlock(&dentry
->d_lock
);
1250 spin_unlock(&dcache_lock
);
1251 error
= nfs_safe_remove(dentry
);
1253 nfs_renew_times(dentry
);
1254 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1255 } else if (need_rehash
)
1262 nfs_symlink(struct inode
*dir
, struct dentry
*dentry
, const char *symname
)
1265 struct nfs_fattr sym_attr
;
1266 struct nfs_fh sym_fh
;
1267 struct qstr qsymname
;
1270 dfprintk(VFS
, "NFS: symlink(%s/%ld, %s, %s)\n", dir
->i_sb
->s_id
,
1271 dir
->i_ino
, dentry
->d_name
.name
, symname
);
1274 if (dentry
->d_inode
)
1275 printk("nfs_proc_symlink: %s/%s not negative!\n",
1276 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
);
1279 * Fill in the sattr for the call.
1280 * Note: SunOS 4.1.2 crashes if the mode isn't initialized!
1282 attr
.ia_valid
= ATTR_MODE
;
1283 attr
.ia_mode
= S_IFLNK
| S_IRWXUGO
;
1285 qsymname
.name
= symname
;
1286 qsymname
.len
= strlen(symname
);
1289 nfs_begin_data_update(dir
);
1290 error
= NFS_PROTO(dir
)->symlink(dir
, &dentry
->d_name
, &qsymname
,
1291 &attr
, &sym_fh
, &sym_attr
);
1292 nfs_end_data_update(dir
);
1294 error
= nfs_instantiate(dentry
, &sym_fh
, &sym_attr
);
1296 if (error
== -EEXIST
)
1297 printk("nfs_proc_symlink: %s/%s already exists??\n",
1298 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
);
1306 nfs_link(struct dentry
*old_dentry
, struct inode
*dir
, struct dentry
*dentry
)
1308 struct inode
*inode
= old_dentry
->d_inode
;
1311 dfprintk(VFS
, "NFS: link(%s/%s -> %s/%s)\n",
1312 old_dentry
->d_parent
->d_name
.name
, old_dentry
->d_name
.name
,
1313 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
);
1316 * Drop the dentry in advance to force a new lookup.
1317 * Since nfs_proc_link doesn't return a file handle,
1318 * we can't use the existing dentry.
1323 nfs_begin_data_update(dir
);
1324 nfs_begin_data_update(inode
);
1325 error
= NFS_PROTO(dir
)->link(inode
, dir
, &dentry
->d_name
);
1326 nfs_end_data_update(inode
);
1327 nfs_end_data_update(dir
);
1334 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a
1335 * different file handle for the same inode after a rename (e.g. when
1336 * moving to a different directory). A fail-safe method to do so would
1337 * be to look up old_dir/old_name, create a link to new_dir/new_name and
1338 * rename the old file using the sillyrename stuff. This way, the original
1339 * file in old_dir will go away when the last process iput()s the inode.
1343 * It actually works quite well. One needs to have the possibility for
1344 * at least one ".nfs..." file in each directory the file ever gets
1345 * moved or linked to which happens automagically with the new
1346 * implementation that only depends on the dcache stuff instead of
1347 * using the inode layer
1349 * Unfortunately, things are a little more complicated than indicated
1350 * above. For a cross-directory move, we want to make sure we can get
1351 * rid of the old inode after the operation. This means there must be
1352 * no pending writes (if it's a file), and the use count must be 1.
1353 * If these conditions are met, we can drop the dentries before doing
1356 static int nfs_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
1357 struct inode
*new_dir
, struct dentry
*new_dentry
)
1359 struct inode
*old_inode
= old_dentry
->d_inode
;
1360 struct inode
*new_inode
= new_dentry
->d_inode
;
1361 struct dentry
*dentry
= NULL
, *rehash
= NULL
;
1365 * To prevent any new references to the target during the rename,
1366 * we unhash the dentry and free the inode in advance.
1369 if (!d_unhashed(new_dentry
)) {
1371 rehash
= new_dentry
;
1374 dfprintk(VFS
, "NFS: rename(%s/%s -> %s/%s, ct=%d)\n",
1375 old_dentry
->d_parent
->d_name
.name
, old_dentry
->d_name
.name
,
1376 new_dentry
->d_parent
->d_name
.name
, new_dentry
->d_name
.name
,
1377 atomic_read(&new_dentry
->d_count
));
1380 * First check whether the target is busy ... we can't
1381 * safely do _any_ rename if the target is in use.
1383 * For files, make a copy of the dentry and then do a
1384 * silly-rename. If the silly-rename succeeds, the
1385 * copied dentry is hashed and becomes the new target.
1389 if (S_ISDIR(new_inode
->i_mode
))
1391 else if (atomic_read(&new_dentry
->d_count
) > 2) {
1393 /* copy the target dentry's name */
1394 dentry
= d_alloc(new_dentry
->d_parent
,
1395 &new_dentry
->d_name
);
1399 /* silly-rename the existing target ... */
1400 err
= nfs_sillyrename(new_dir
, new_dentry
);
1402 new_dentry
= rehash
= dentry
;
1404 /* instantiate the replacement target */
1405 d_instantiate(new_dentry
, NULL
);
1406 } else if (atomic_read(&new_dentry
->d_count
) > 1) {
1407 /* dentry still busy? */
1409 printk("nfs_rename: target %s/%s busy, d_count=%d\n",
1410 new_dentry
->d_parent
->d_name
.name
,
1411 new_dentry
->d_name
.name
,
1412 atomic_read(&new_dentry
->d_count
));
1420 * ... prune child dentries and writebacks if needed.
1422 if (atomic_read(&old_dentry
->d_count
) > 1) {
1423 nfs_wb_all(old_inode
);
1424 shrink_dcache_parent(old_dentry
);
1428 d_delete(new_dentry
);
1430 nfs_begin_data_update(old_dir
);
1431 nfs_begin_data_update(new_dir
);
1432 nfs_begin_data_update(old_inode
);
1433 error
= NFS_PROTO(old_dir
)->rename(old_dir
, &old_dentry
->d_name
,
1434 new_dir
, &new_dentry
->d_name
);
1435 nfs_end_data_update(old_inode
);
1436 nfs_end_data_update(new_dir
);
1437 nfs_end_data_update(old_dir
);
1442 if (!S_ISDIR(old_inode
->i_mode
))
1443 d_move(old_dentry
, new_dentry
);
1444 nfs_renew_times(new_dentry
);
1445 nfs_set_verifier(new_dentry
, nfs_save_change_attribute(new_dir
));
1448 /* new dentry created? */
1455 int nfs_access_get_cached(struct inode
*inode
, struct rpc_cred
*cred
, struct nfs_access_entry
*res
)
1457 struct nfs_access_entry
*cache
= &NFS_I(inode
)->cache_access
;
1459 if (cache
->cred
!= cred
1460 || time_after(jiffies
, cache
->jiffies
+ NFS_ATTRTIMEO(inode
))
1461 || (NFS_FLAGS(inode
) & NFS_INO_INVALID_ACCESS
))
1463 memcpy(res
, cache
, sizeof(*res
));
1467 void nfs_access_add_cache(struct inode
*inode
, struct nfs_access_entry
*set
)
1469 struct nfs_access_entry
*cache
= &NFS_I(inode
)->cache_access
;
1471 if (cache
->cred
!= set
->cred
) {
1473 put_rpccred(cache
->cred
);
1474 cache
->cred
= get_rpccred(set
->cred
);
1476 NFS_FLAGS(inode
) &= ~NFS_INO_INVALID_ACCESS
;
1477 cache
->jiffies
= set
->jiffies
;
1478 cache
->mask
= set
->mask
;
1481 static int nfs_do_access(struct inode
*inode
, struct rpc_cred
*cred
, int mask
)
1483 struct nfs_access_entry cache
;
1486 status
= nfs_access_get_cached(inode
, cred
, &cache
);
1490 /* Be clever: ask server to check for all possible rights */
1491 cache
.mask
= MAY_EXEC
| MAY_WRITE
| MAY_READ
;
1493 cache
.jiffies
= jiffies
;
1494 status
= NFS_PROTO(inode
)->access(inode
, &cache
);
1497 nfs_access_add_cache(inode
, &cache
);
1499 if ((cache
.mask
& mask
) == mask
)
1504 int nfs_permission(struct inode
*inode
, int mask
, struct nameidata
*nd
)
1506 struct rpc_cred
*cred
;
1511 /* Is this sys_access() ? */
1512 if (nd
!= NULL
&& (nd
->flags
& LOOKUP_ACCESS
))
1515 switch (inode
->i_mode
& S_IFMT
) {
1519 /* NFSv4 has atomic_open... */
1520 if (nfs_server_capable(inode
, NFS_CAP_ATOMIC_OPEN
)
1522 && (nd
->flags
& LOOKUP_OPEN
))
1527 * Optimize away all write operations, since the server
1528 * will check permissions when we perform the op.
1530 if ((mask
& MAY_WRITE
) && !(mask
& MAY_READ
))
1537 if (!NFS_PROTO(inode
)->access
)
1540 cred
= rpcauth_lookupcred(NFS_CLIENT(inode
)->cl_auth
, 0);
1541 if (!IS_ERR(cred
)) {
1542 res
= nfs_do_access(inode
, cred
, mask
);
1545 res
= PTR_ERR(cred
);
1550 res
= nfs_revalidate_inode(NFS_SERVER(inode
), inode
);
1552 res
= generic_permission(inode
, mask
, NULL
);
1559 * version-control: t
1560 * kept-new-versions: 5