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>
36 #include "delegation.h"
38 #define NFS_PARANOIA 1
39 /* #define NFS_DEBUG_VERBOSE 1 */
41 static int nfs_opendir(struct inode
*, struct file
*);
42 static int nfs_readdir(struct file
*, void *, filldir_t
);
43 static struct dentry
*nfs_lookup(struct inode
*, struct dentry
*, struct nameidata
*);
44 static int nfs_create(struct inode
*, struct dentry
*, int, struct nameidata
*);
45 static int nfs_mkdir(struct inode
*, struct dentry
*, int);
46 static int nfs_rmdir(struct inode
*, struct dentry
*);
47 static int nfs_unlink(struct inode
*, struct dentry
*);
48 static int nfs_symlink(struct inode
*, struct dentry
*, const char *);
49 static int nfs_link(struct dentry
*, struct inode
*, struct dentry
*);
50 static int nfs_mknod(struct inode
*, struct dentry
*, int, dev_t
);
51 static int nfs_rename(struct inode
*, struct dentry
*,
52 struct inode
*, struct dentry
*);
53 static int nfs_fsync_dir(struct file
*, struct dentry
*, int);
55 struct file_operations nfs_dir_operations
= {
56 .read
= generic_read_dir
,
57 .readdir
= nfs_readdir
,
59 .release
= nfs_release
,
60 .fsync
= nfs_fsync_dir
,
63 struct inode_operations nfs_dir_inode_operations
= {
68 .symlink
= nfs_symlink
,
73 .permission
= nfs_permission
,
74 .getattr
= nfs_getattr
,
75 .setattr
= nfs_setattr
,
80 static struct dentry
*nfs_atomic_lookup(struct inode
*, struct dentry
*, struct nameidata
*);
81 struct inode_operations nfs4_dir_inode_operations
= {
83 .lookup
= nfs_atomic_lookup
,
86 .symlink
= nfs_symlink
,
91 .permission
= nfs_permission
,
92 .getattr
= nfs_getattr
,
93 .setattr
= nfs_setattr
,
96 #endif /* CONFIG_NFS_V4 */
102 nfs_opendir(struct inode
*inode
, struct file
*filp
)
107 /* Call generic open code in order to cache credentials */
109 res
= nfs_open(inode
, filp
);
114 typedef u32
* (*decode_dirent_t
)(u32
*, struct nfs_entry
*, int);
118 unsigned long page_index
;
121 struct nfs_entry
*entry
;
122 decode_dirent_t decode
;
125 } nfs_readdir_descriptor_t
;
127 /* Now we cache directories properly, by stuffing the dirent
128 * data directly in the page cache.
130 * Inode invalidation due to refresh etc. takes care of
131 * _everything_, no sloppy entry flushing logic, no extraneous
132 * copying, network direct to page cache, the way it was meant
135 * NOTE: Dirent information verification is done always by the
136 * page-in of the RPC reply, nowhere else, this simplies
137 * things substantially.
140 int nfs_readdir_filler(nfs_readdir_descriptor_t
*desc
, struct page
*page
)
142 struct file
*file
= desc
->file
;
143 struct inode
*inode
= file
->f_dentry
->d_inode
;
144 struct rpc_cred
*cred
= nfs_file_cred(file
);
145 unsigned long timestamp
;
148 dfprintk(VFS
, "NFS: nfs_readdir_filler() reading cookie %Lu into page %lu.\n", (long long)desc
->entry
->cookie
, page
->index
);
152 error
= NFS_PROTO(inode
)->readdir(file
->f_dentry
, cred
, desc
->entry
->cookie
, page
,
153 NFS_SERVER(inode
)->dtsize
, desc
->plus
);
155 /* We requested READDIRPLUS, but the server doesn't grok it */
156 if (error
== -ENOTSUPP
&& desc
->plus
) {
157 NFS_SERVER(inode
)->caps
&= ~NFS_CAP_READDIRPLUS
;
158 NFS_FLAGS(inode
) &= ~NFS_INO_ADVISE_RDPLUS
;
164 SetPageUptodate(page
);
165 NFS_FLAGS(inode
) |= NFS_INO_INVALID_ATIME
;
166 /* Ensure consistent page alignment of the data.
167 * Note: assumes we have exclusive access to this mapping either
168 * throught inode->i_sem or some other mechanism.
170 if (page
->index
== 0) {
171 invalidate_inode_pages(inode
->i_mapping
);
172 NFS_I(inode
)->readdir_timestamp
= timestamp
;
179 nfs_zap_caches(inode
);
185 int dir_decode(nfs_readdir_descriptor_t
*desc
)
188 p
= desc
->decode(p
, desc
->entry
, desc
->plus
);
196 void dir_page_release(nfs_readdir_descriptor_t
*desc
)
199 page_cache_release(desc
->page
);
205 * Given a pointer to a buffer that has already been filled by a call
206 * to readdir, find the next entry.
208 * If the end of the buffer has been reached, return -EAGAIN, if not,
209 * return the offset within the buffer of the next entry to be
213 int find_dirent(nfs_readdir_descriptor_t
*desc
, struct page
*page
)
215 struct nfs_entry
*entry
= desc
->entry
;
219 while((status
= dir_decode(desc
)) == 0) {
220 dfprintk(VFS
, "NFS: found cookie %Lu\n", (long long)entry
->cookie
);
221 if (entry
->prev_cookie
== desc
->target
)
223 if (loop_count
++ > 200) {
228 dfprintk(VFS
, "NFS: find_dirent() returns %d\n", status
);
233 * Find the given page, and call find_dirent() in order to try to
234 * return the next entry.
237 int find_dirent_page(nfs_readdir_descriptor_t
*desc
)
239 struct inode
*inode
= desc
->file
->f_dentry
->d_inode
;
243 dfprintk(VFS
, "NFS: find_dirent_page() searching directory page %ld\n", desc
->page_index
);
245 page
= read_cache_page(inode
->i_mapping
, desc
->page_index
,
246 (filler_t
*)nfs_readdir_filler
, desc
);
248 status
= PTR_ERR(page
);
251 if (!PageUptodate(page
))
254 /* NOTE: Someone else may have changed the READDIRPLUS flag */
256 desc
->ptr
= kmap(page
); /* matching kunmap in nfs_do_filldir */
257 status
= find_dirent(desc
, page
);
259 dir_page_release(desc
);
261 dfprintk(VFS
, "NFS: find_dirent_page() returns %d\n", status
);
264 page_cache_release(page
);
269 * Recurse through the page cache pages, and return a
270 * filled nfs_entry structure of the next directory entry if possible.
272 * The target for the search is 'desc->target'.
275 int readdir_search_pagecache(nfs_readdir_descriptor_t
*desc
)
280 dfprintk(VFS
, "NFS: readdir_search_pagecache() searching for cookie %Lu\n", (long long)desc
->target
);
282 res
= find_dirent_page(desc
);
285 /* Align to beginning of next page */
287 if (loop_count
++ > 200) {
292 dfprintk(VFS
, "NFS: readdir_search_pagecache() returned %d\n", res
);
296 static inline unsigned int dt_type(struct inode
*inode
)
298 return (inode
->i_mode
>> 12) & 15;
301 static struct dentry
*nfs_readdir_lookup(nfs_readdir_descriptor_t
*desc
);
304 * Once we've found the start of the dirent within a page: fill 'er up...
307 int nfs_do_filldir(nfs_readdir_descriptor_t
*desc
, void *dirent
,
310 struct file
*file
= desc
->file
;
311 struct nfs_entry
*entry
= desc
->entry
;
312 struct dentry
*dentry
= NULL
;
313 unsigned long fileid
;
317 dfprintk(VFS
, "NFS: nfs_do_filldir() filling starting @ cookie %Lu\n", (long long)desc
->target
);
320 unsigned d_type
= DT_UNKNOWN
;
321 /* Note: entry->prev_cookie contains the cookie for
322 * retrieving the current dirent on the server */
323 fileid
= nfs_fileid_to_ino_t(entry
->ino
);
325 /* Get a dentry if we have one */
328 dentry
= nfs_readdir_lookup(desc
);
330 /* Use readdirplus info */
331 if (dentry
!= NULL
&& dentry
->d_inode
!= NULL
) {
332 d_type
= dt_type(dentry
->d_inode
);
333 fileid
= dentry
->d_inode
->i_ino
;
336 res
= filldir(dirent
, entry
->name
, entry
->len
,
337 entry
->prev_cookie
, fileid
, d_type
);
340 file
->f_pos
= desc
->target
= entry
->cookie
;
341 if (dir_decode(desc
) != 0) {
345 if (loop_count
++ > 200) {
350 dir_page_release(desc
);
353 dfprintk(VFS
, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n", (long long)desc
->target
, res
);
358 * If we cannot find a cookie in our cache, we suspect that this is
359 * because it points to a deleted file, so we ask the server to return
360 * whatever it thinks is the next entry. We then feed this to filldir.
361 * If all goes well, we should then be able to find our way round the
362 * cache on the next call to readdir_search_pagecache();
364 * NOTE: we cannot add the anonymous page to the pagecache because
365 * the data it contains might not be page aligned. Besides,
366 * we should already have a complete representation of the
367 * directory in the page cache by the time we get here.
370 int uncached_readdir(nfs_readdir_descriptor_t
*desc
, void *dirent
,
373 struct file
*file
= desc
->file
;
374 struct inode
*inode
= file
->f_dentry
->d_inode
;
375 struct rpc_cred
*cred
= nfs_file_cred(file
);
376 struct page
*page
= NULL
;
379 dfprintk(VFS
, "NFS: uncached_readdir() searching for cookie %Lu\n", (long long)desc
->target
);
381 page
= alloc_page(GFP_HIGHUSER
);
386 desc
->error
= NFS_PROTO(inode
)->readdir(file
->f_dentry
, cred
, desc
->target
,
388 NFS_SERVER(inode
)->dtsize
,
390 NFS_FLAGS(inode
) |= NFS_INO_INVALID_ATIME
;
392 desc
->ptr
= kmap(page
); /* matching kunmap in nfs_do_filldir */
393 if (desc
->error
>= 0) {
394 if ((status
= dir_decode(desc
)) == 0)
395 desc
->entry
->prev_cookie
= desc
->target
;
401 status
= nfs_do_filldir(desc
, dirent
, filldir
);
403 /* Reset read descriptor so it searches the page cache from
404 * the start upon the next call to readdir_search_pagecache() */
405 desc
->page_index
= 0;
406 desc
->entry
->cookie
= desc
->entry
->prev_cookie
= 0;
407 desc
->entry
->eof
= 0;
409 dfprintk(VFS
, "NFS: uncached_readdir() returns %d\n", status
);
412 dir_page_release(desc
);
416 /* The file offset position is now represented as a true offset into the
417 * page cache as is the case in most of the other filesystems.
419 static int nfs_readdir(struct file
*filp
, void *dirent
, filldir_t filldir
)
421 struct dentry
*dentry
= filp
->f_dentry
;
422 struct inode
*inode
= dentry
->d_inode
;
423 nfs_readdir_descriptor_t my_desc
,
425 struct nfs_entry my_entry
;
427 struct nfs_fattr fattr
;
432 res
= nfs_revalidate_inode(NFS_SERVER(inode
), inode
);
439 * filp->f_pos points to the file offset in the page cache.
440 * but if the cache has meanwhile been zapped, we need to
441 * read from the last dirent to revalidate f_pos
444 memset(desc
, 0, sizeof(*desc
));
447 desc
->target
= filp
->f_pos
;
448 desc
->decode
= NFS_PROTO(inode
)->decode_dirent
;
449 desc
->plus
= NFS_USE_READDIRPLUS(inode
);
451 my_entry
.cookie
= my_entry
.prev_cookie
= 0;
454 my_entry
.fattr
= &fattr
;
455 desc
->entry
= &my_entry
;
457 while(!desc
->entry
->eof
) {
458 res
= readdir_search_pagecache(desc
);
459 if (res
== -EBADCOOKIE
) {
460 /* This means either end of directory */
461 if (desc
->entry
->cookie
!= desc
->target
) {
462 /* Or that the server has 'lost' a cookie */
463 res
= uncached_readdir(desc
, dirent
, filldir
);
470 if (res
== -ETOOSMALL
&& desc
->plus
) {
471 NFS_FLAGS(inode
) &= ~NFS_INO_ADVISE_RDPLUS
;
472 nfs_zap_caches(inode
);
474 desc
->entry
->eof
= 0;
480 res
= nfs_do_filldir(desc
, dirent
, filldir
);
495 * All directory operations under NFS are synchronous, so fsync()
496 * is a dummy operation.
498 int nfs_fsync_dir(struct file
*filp
, struct dentry
*dentry
, int datasync
)
504 * A check for whether or not the parent directory has changed.
505 * In the case it has, we assume that the dentries are untrustworthy
506 * and may need to be looked up again.
508 static inline int nfs_check_verifier(struct inode
*dir
, struct dentry
*dentry
)
512 if ((NFS_FLAGS(dir
) & NFS_INO_INVALID_ATTR
) != 0
513 || nfs_attribute_timeout(dir
))
515 return nfs_verify_change_attribute(dir
, (unsigned long)dentry
->d_fsdata
);
518 static inline void nfs_set_verifier(struct dentry
* dentry
, unsigned long verf
)
520 dentry
->d_fsdata
= (void *)verf
;
524 * Whenever an NFS operation succeeds, we know that the dentry
525 * is valid, so we update the revalidation timestamp.
527 static inline void nfs_renew_times(struct dentry
* dentry
)
529 dentry
->d_time
= jiffies
;
533 * Return the intent data that applies to this particular path component
535 * Note that the current set of intents only apply to the very last
536 * component of the path.
537 * We check for this using LOOKUP_CONTINUE and LOOKUP_PARENT.
539 static inline unsigned int nfs_lookup_check_intent(struct nameidata
*nd
, unsigned int mask
)
541 if (nd
->flags
& (LOOKUP_CONTINUE
|LOOKUP_PARENT
))
543 return nd
->flags
& mask
;
547 * Inode and filehandle revalidation for lookups.
549 * We force revalidation in the cases where the VFS sets LOOKUP_REVAL,
550 * or if the intent information indicates that we're about to open this
551 * particular file and the "nocto" mount flag is not set.
555 int nfs_lookup_verify_inode(struct inode
*inode
, struct nameidata
*nd
)
557 struct nfs_server
*server
= NFS_SERVER(inode
);
560 /* VFS wants an on-the-wire revalidation */
561 if (nd
->flags
& LOOKUP_REVAL
)
563 /* This is an open(2) */
564 if (nfs_lookup_check_intent(nd
, LOOKUP_OPEN
) != 0 &&
565 !(server
->flags
& NFS_MOUNT_NOCTO
))
568 return nfs_revalidate_inode(server
, inode
);
570 return __nfs_revalidate_inode(server
, inode
);
574 * We judge how long we want to trust negative
575 * dentries by looking at the parent inode mtime.
577 * If parent mtime has changed, we revalidate, else we wait for a
578 * period corresponding to the parent's attribute cache timeout value.
581 int nfs_neg_need_reval(struct inode
*dir
, struct dentry
*dentry
,
582 struct nameidata
*nd
)
584 /* Don't revalidate a negative dentry if we're creating a new file */
585 if (nd
!= NULL
&& nfs_lookup_check_intent(nd
, LOOKUP_CREATE
) != 0)
587 return !nfs_check_verifier(dir
, dentry
);
591 * This is called every time the dcache has a lookup hit,
592 * and we should check whether we can really trust that
595 * NOTE! The hit can be a negative hit too, don't assume
598 * If the parent directory is seen to have changed, we throw out the
599 * cached dentry and do a new lookup.
601 static int nfs_lookup_revalidate(struct dentry
* dentry
, struct nameidata
*nd
)
605 struct dentry
*parent
;
607 struct nfs_fh fhandle
;
608 struct nfs_fattr fattr
;
609 unsigned long verifier
;
611 parent
= dget_parent(dentry
);
613 dir
= parent
->d_inode
;
614 inode
= dentry
->d_inode
;
617 if (nfs_neg_need_reval(dir
, dentry
, nd
))
622 if (is_bad_inode(inode
)) {
623 dfprintk(VFS
, "nfs_lookup_validate: %s/%s has dud inode\n",
624 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
);
628 /* Revalidate parent directory attribute cache */
629 if (nfs_revalidate_inode(NFS_SERVER(dir
), dir
) < 0)
632 /* Force a full look up iff the parent directory has changed */
633 if (nfs_check_verifier(dir
, dentry
)) {
634 if (nfs_lookup_verify_inode(inode
, nd
))
639 if (NFS_STALE(inode
))
642 verifier
= nfs_save_change_attribute(dir
);
643 error
= NFS_PROTO(dir
)->lookup(dir
, &dentry
->d_name
, &fhandle
, &fattr
);
646 if (nfs_compare_fh(NFS_FH(inode
), &fhandle
))
648 if ((error
= nfs_refresh_inode(inode
, &fattr
)) != 0)
651 nfs_renew_times(dentry
);
652 nfs_set_verifier(dentry
, verifier
);
661 if (inode
&& S_ISDIR(inode
->i_mode
)) {
662 /* Purge readdir caches. */
663 nfs_zap_caches(inode
);
664 /* If we have submounts, don't unhash ! */
665 if (have_submounts(dentry
))
667 shrink_dcache_parent(dentry
);
676 * This is called from dput() when d_count is going to 0.
678 static int nfs_dentry_delete(struct dentry
*dentry
)
680 dfprintk(VFS
, "NFS: dentry_delete(%s/%s, %x)\n",
681 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
,
684 if (dentry
->d_flags
& DCACHE_NFSFS_RENAMED
) {
685 /* Unhash it, so that ->d_iput() would be called */
688 if (!(dentry
->d_sb
->s_flags
& MS_ACTIVE
)) {
689 /* Unhash it, so that ancestors of killed async unlink
690 * files will be cleaned up during umount */
698 * Called when the dentry loses inode.
699 * We use it to clean up silly-renamed files.
701 static void nfs_dentry_iput(struct dentry
*dentry
, struct inode
*inode
)
703 if (dentry
->d_flags
& DCACHE_NFSFS_RENAMED
) {
706 nfs_complete_unlink(dentry
);
709 /* When creating a negative dentry, we want to renew d_time */
710 nfs_renew_times(dentry
);
714 struct dentry_operations nfs_dentry_operations
= {
715 .d_revalidate
= nfs_lookup_revalidate
,
716 .d_delete
= nfs_dentry_delete
,
717 .d_iput
= nfs_dentry_iput
,
721 * Use intent information to check whether or not we're going to do
722 * an O_EXCL create using this path component.
725 int nfs_is_exclusive_create(struct inode
*dir
, struct nameidata
*nd
)
727 if (NFS_PROTO(dir
)->version
== 2)
729 if (nd
== NULL
|| nfs_lookup_check_intent(nd
, LOOKUP_CREATE
) == 0)
731 return (nd
->intent
.open
.flags
& O_EXCL
) != 0;
734 static struct dentry
*nfs_lookup(struct inode
*dir
, struct dentry
* dentry
, struct nameidata
*nd
)
737 struct inode
*inode
= NULL
;
739 struct nfs_fh fhandle
;
740 struct nfs_fattr fattr
;
742 dfprintk(VFS
, "NFS: lookup(%s/%s)\n",
743 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
);
745 res
= ERR_PTR(-ENAMETOOLONG
);
746 if (dentry
->d_name
.len
> NFS_SERVER(dir
)->namelen
)
749 res
= ERR_PTR(-ENOMEM
);
750 dentry
->d_op
= NFS_PROTO(dir
)->dentry_ops
;
753 /* Revalidate parent directory attribute cache */
754 error
= nfs_revalidate_inode(NFS_SERVER(dir
), dir
);
756 res
= ERR_PTR(error
);
760 /* If we're doing an exclusive create, optimize away the lookup */
761 if (nfs_is_exclusive_create(dir
, nd
))
764 error
= NFS_PROTO(dir
)->lookup(dir
, &dentry
->d_name
, &fhandle
, &fattr
);
765 if (error
== -ENOENT
)
768 res
= ERR_PTR(error
);
771 res
= ERR_PTR(-EACCES
);
772 inode
= nfs_fhget(dentry
->d_sb
, &fhandle
, &fattr
);
776 res
= d_add_unique(dentry
, inode
);
779 nfs_renew_times(dentry
);
780 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
788 static int nfs_open_revalidate(struct dentry
*, struct nameidata
*);
790 struct dentry_operations nfs4_dentry_operations
= {
791 .d_revalidate
= nfs_open_revalidate
,
792 .d_delete
= nfs_dentry_delete
,
793 .d_iput
= nfs_dentry_iput
,
797 * Use intent information to determine whether we need to substitute
798 * the NFSv4-style stateful OPEN for the LOOKUP call
800 static int is_atomic_open(struct inode
*dir
, struct nameidata
*nd
)
802 if (nd
== NULL
|| nfs_lookup_check_intent(nd
, LOOKUP_OPEN
) == 0)
804 /* NFS does not (yet) have a stateful open for directories */
805 if (nd
->flags
& LOOKUP_DIRECTORY
)
807 /* Are we trying to write to a read only partition? */
808 if (IS_RDONLY(dir
) && (nd
->intent
.open
.flags
& (O_CREAT
|O_TRUNC
|FMODE_WRITE
)))
813 static struct dentry
*nfs_atomic_lookup(struct inode
*dir
, struct dentry
*dentry
, struct nameidata
*nd
)
815 struct dentry
*res
= NULL
;
816 struct inode
*inode
= NULL
;
819 /* Check that we are indeed trying to open this file */
820 if (!is_atomic_open(dir
, nd
))
823 if (dentry
->d_name
.len
> NFS_SERVER(dir
)->namelen
) {
824 res
= ERR_PTR(-ENAMETOOLONG
);
827 dentry
->d_op
= NFS_PROTO(dir
)->dentry_ops
;
829 /* Let vfs_create() deal with O_EXCL */
830 if (nd
->intent
.open
.flags
& O_EXCL
)
833 /* Open the file on the server */
835 /* Revalidate parent directory attribute cache */
836 error
= nfs_revalidate_inode(NFS_SERVER(dir
), dir
);
838 res
= ERR_PTR(error
);
842 if (nd
->intent
.open
.flags
& O_CREAT
) {
843 nfs_begin_data_update(dir
);
844 inode
= nfs4_atomic_open(dir
, dentry
, nd
);
845 nfs_end_data_update(dir
);
847 inode
= nfs4_atomic_open(dir
, dentry
, nd
);
850 error
= PTR_ERR(inode
);
852 /* Make a negative dentry */
856 /* This turned out not to be a regular file */
858 if (!(nd
->intent
.open
.flags
& O_NOFOLLOW
))
863 res
= ERR_PTR(error
);
868 res
= d_add_unique(dentry
, inode
);
871 nfs_renew_times(dentry
);
872 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
876 return nfs_lookup(dir
, dentry
, nd
);
879 static int nfs_open_revalidate(struct dentry
*dentry
, struct nameidata
*nd
)
881 struct dentry
*parent
= NULL
;
882 struct inode
*inode
= dentry
->d_inode
;
884 unsigned long verifier
;
885 int openflags
, ret
= 0;
887 parent
= dget_parent(dentry
);
888 dir
= parent
->d_inode
;
889 if (!is_atomic_open(dir
, nd
))
891 /* We can't create new files in nfs_open_revalidate(), so we
892 * optimize away revalidation of negative dentries.
896 /* NFS only supports OPEN on regular files */
897 if (!S_ISREG(inode
->i_mode
))
899 openflags
= nd
->intent
.open
.flags
;
900 /* We cannot do exclusive creation on a positive dentry */
901 if ((openflags
& (O_CREAT
|O_EXCL
)) == (O_CREAT
|O_EXCL
))
903 /* We can't create new files, or truncate existing ones here */
904 openflags
&= ~(O_CREAT
|O_TRUNC
);
907 * Note: we're not holding inode->i_sem and so may be racing with
908 * operations that change the directory. We therefore save the
909 * change attribute *before* we do the RPC call.
912 verifier
= nfs_save_change_attribute(dir
);
913 ret
= nfs4_open_revalidate(dir
, dentry
, openflags
);
915 nfs_set_verifier(dentry
, verifier
);
924 if (inode
!= NULL
&& nfs_have_delegation(inode
, FMODE_READ
))
926 return nfs_lookup_revalidate(dentry
, nd
);
928 #endif /* CONFIG_NFSV4 */
930 static struct dentry
*nfs_readdir_lookup(nfs_readdir_descriptor_t
*desc
)
932 struct dentry
*parent
= desc
->file
->f_dentry
;
933 struct inode
*dir
= parent
->d_inode
;
934 struct nfs_entry
*entry
= desc
->entry
;
935 struct dentry
*dentry
, *alias
;
944 if (name
.name
[0] == '.' && name
.name
[1] == '.')
945 return dget_parent(parent
);
948 if (name
.name
[0] == '.')
951 name
.hash
= full_name_hash(name
.name
, name
.len
);
952 dentry
= d_lookup(parent
, &name
);
955 if (!desc
->plus
|| !(entry
->fattr
->valid
& NFS_ATTR_FATTR
))
957 /* Note: caller is already holding the dir->i_sem! */
958 dentry
= d_alloc(parent
, &name
);
961 dentry
->d_op
= NFS_PROTO(dir
)->dentry_ops
;
962 inode
= nfs_fhget(dentry
->d_sb
, entry
->fh
, entry
->fattr
);
967 alias
= d_add_unique(dentry
, inode
);
972 nfs_renew_times(dentry
);
973 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
978 * Code common to create, mkdir, and mknod.
980 int nfs_instantiate(struct dentry
*dentry
, struct nfs_fh
*fhandle
,
981 struct nfs_fattr
*fattr
)
986 /* We may have been initialized further down */
989 if (fhandle
->size
== 0) {
990 struct inode
*dir
= dentry
->d_parent
->d_inode
;
991 error
= NFS_PROTO(dir
)->lookup(dir
, &dentry
->d_name
, fhandle
, fattr
);
995 if (!(fattr
->valid
& NFS_ATTR_FATTR
)) {
996 struct nfs_server
*server
= NFS_SB(dentry
->d_sb
);
997 error
= server
->rpc_ops
->getattr(server
, fhandle
, fattr
);
1002 inode
= nfs_fhget(dentry
->d_sb
, fhandle
, fattr
);
1005 d_instantiate(dentry
, inode
);
1013 * Following a failed create operation, we drop the dentry rather
1014 * than retain a negative dentry. This avoids a problem in the event
1015 * that the operation succeeded on the server, but an error in the
1016 * reply path made it appear to have failed.
1018 static int nfs_create(struct inode
*dir
, struct dentry
*dentry
, int mode
,
1019 struct nameidata
*nd
)
1025 dfprintk(VFS
, "NFS: create(%s/%ld, %s\n", dir
->i_sb
->s_id
,
1026 dir
->i_ino
, dentry
->d_name
.name
);
1028 attr
.ia_mode
= mode
;
1029 attr
.ia_valid
= ATTR_MODE
;
1031 if (nd
&& (nd
->flags
& LOOKUP_CREATE
))
1032 open_flags
= nd
->intent
.open
.flags
;
1035 nfs_begin_data_update(dir
);
1036 error
= NFS_PROTO(dir
)->create(dir
, dentry
, &attr
, open_flags
);
1037 nfs_end_data_update(dir
);
1040 nfs_renew_times(dentry
);
1041 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1051 * See comments for nfs_proc_create regarding failed operations.
1054 nfs_mknod(struct inode
*dir
, struct dentry
*dentry
, int mode
, dev_t rdev
)
1059 dfprintk(VFS
, "NFS: mknod(%s/%ld, %s\n", dir
->i_sb
->s_id
,
1060 dir
->i_ino
, dentry
->d_name
.name
);
1062 if (!new_valid_dev(rdev
))
1065 attr
.ia_mode
= mode
;
1066 attr
.ia_valid
= ATTR_MODE
;
1069 nfs_begin_data_update(dir
);
1070 status
= NFS_PROTO(dir
)->mknod(dir
, dentry
, &attr
, rdev
);
1071 nfs_end_data_update(dir
);
1074 nfs_renew_times(dentry
);
1075 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1085 * See comments for nfs_proc_create regarding failed operations.
1087 static int nfs_mkdir(struct inode
*dir
, struct dentry
*dentry
, int mode
)
1092 dfprintk(VFS
, "NFS: mkdir(%s/%ld, %s\n", dir
->i_sb
->s_id
,
1093 dir
->i_ino
, dentry
->d_name
.name
);
1095 attr
.ia_valid
= ATTR_MODE
;
1096 attr
.ia_mode
= mode
| S_IFDIR
;
1099 nfs_begin_data_update(dir
);
1100 error
= NFS_PROTO(dir
)->mkdir(dir
, dentry
, &attr
);
1101 nfs_end_data_update(dir
);
1104 nfs_renew_times(dentry
);
1105 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1114 static int nfs_rmdir(struct inode
*dir
, struct dentry
*dentry
)
1118 dfprintk(VFS
, "NFS: rmdir(%s/%ld, %s\n", dir
->i_sb
->s_id
,
1119 dir
->i_ino
, dentry
->d_name
.name
);
1122 nfs_begin_data_update(dir
);
1123 error
= NFS_PROTO(dir
)->rmdir(dir
, &dentry
->d_name
);
1124 /* Ensure the VFS deletes this inode */
1125 if (error
== 0 && dentry
->d_inode
!= NULL
)
1126 dentry
->d_inode
->i_nlink
= 0;
1127 nfs_end_data_update(dir
);
1133 static int nfs_sillyrename(struct inode
*dir
, struct dentry
*dentry
)
1135 static unsigned int sillycounter
;
1136 const int i_inosize
= sizeof(dir
->i_ino
)*2;
1137 const int countersize
= sizeof(sillycounter
)*2;
1138 const int slen
= sizeof(".nfs") + i_inosize
+ countersize
- 1;
1141 struct dentry
*sdentry
;
1144 dfprintk(VFS
, "NFS: silly-rename(%s/%s, ct=%d)\n",
1145 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
,
1146 atomic_read(&dentry
->d_count
));
1149 if (!dentry
->d_inode
)
1150 printk("NFS: silly-renaming %s/%s, negative dentry??\n",
1151 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
);
1154 * We don't allow a dentry to be silly-renamed twice.
1157 if (dentry
->d_flags
& DCACHE_NFSFS_RENAMED
)
1160 sprintf(silly
, ".nfs%*.*lx",
1161 i_inosize
, i_inosize
, dentry
->d_inode
->i_ino
);
1165 char *suffix
= silly
+ slen
- countersize
;
1169 sprintf(suffix
, "%*.*x", countersize
, countersize
, sillycounter
);
1171 dfprintk(VFS
, "trying to rename %s to %s\n",
1172 dentry
->d_name
.name
, silly
);
1174 sdentry
= lookup_one_len(silly
, dentry
->d_parent
, slen
);
1176 * N.B. Better to return EBUSY here ... it could be
1177 * dangerous to delete the file while it's in use.
1179 if (IS_ERR(sdentry
))
1181 } while(sdentry
->d_inode
!= NULL
); /* need negative lookup */
1183 qsilly
.name
= silly
;
1184 qsilly
.len
= strlen(silly
);
1185 nfs_begin_data_update(dir
);
1186 if (dentry
->d_inode
) {
1187 nfs_begin_data_update(dentry
->d_inode
);
1188 error
= NFS_PROTO(dir
)->rename(dir
, &dentry
->d_name
,
1190 nfs_end_data_update(dentry
->d_inode
);
1192 error
= NFS_PROTO(dir
)->rename(dir
, &dentry
->d_name
,
1194 nfs_end_data_update(dir
);
1196 nfs_renew_times(dentry
);
1197 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1198 d_move(dentry
, sdentry
);
1199 error
= nfs_async_unlink(dentry
);
1200 /* If we return 0 we don't unlink */
1208 * Remove a file after making sure there are no pending writes,
1209 * and after checking that the file has only one user.
1211 * We invalidate the attribute cache and free the inode prior to the operation
1212 * to avoid possible races if the server reuses the inode.
1214 static int nfs_safe_remove(struct dentry
*dentry
)
1216 struct inode
*dir
= dentry
->d_parent
->d_inode
;
1217 struct inode
*inode
= dentry
->d_inode
;
1220 dfprintk(VFS
, "NFS: safe_remove(%s/%s)\n",
1221 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
);
1223 /* If the dentry was sillyrenamed, we simply call d_delete() */
1224 if (dentry
->d_flags
& DCACHE_NFSFS_RENAMED
) {
1229 nfs_begin_data_update(dir
);
1230 if (inode
!= NULL
) {
1231 nfs_begin_data_update(inode
);
1232 error
= NFS_PROTO(dir
)->remove(dir
, &dentry
->d_name
);
1233 /* The VFS may want to delete this inode */
1236 nfs_end_data_update(inode
);
1238 error
= NFS_PROTO(dir
)->remove(dir
, &dentry
->d_name
);
1239 nfs_end_data_update(dir
);
1244 /* We do silly rename. In case sillyrename() returns -EBUSY, the inode
1245 * belongs to an active ".nfs..." file and we return -EBUSY.
1247 * If sillyrename() returns 0, we do nothing, otherwise we unlink.
1249 static int nfs_unlink(struct inode
*dir
, struct dentry
*dentry
)
1252 int need_rehash
= 0;
1254 dfprintk(VFS
, "NFS: unlink(%s/%ld, %s)\n", dir
->i_sb
->s_id
,
1255 dir
->i_ino
, dentry
->d_name
.name
);
1258 spin_lock(&dcache_lock
);
1259 spin_lock(&dentry
->d_lock
);
1260 if (atomic_read(&dentry
->d_count
) > 1) {
1261 spin_unlock(&dentry
->d_lock
);
1262 spin_unlock(&dcache_lock
);
1263 error
= nfs_sillyrename(dir
, dentry
);
1267 if (!d_unhashed(dentry
)) {
1271 spin_unlock(&dentry
->d_lock
);
1272 spin_unlock(&dcache_lock
);
1273 error
= nfs_safe_remove(dentry
);
1275 nfs_renew_times(dentry
);
1276 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1277 } else if (need_rehash
)
1284 nfs_symlink(struct inode
*dir
, struct dentry
*dentry
, const char *symname
)
1287 struct nfs_fattr sym_attr
;
1288 struct nfs_fh sym_fh
;
1289 struct qstr qsymname
;
1292 dfprintk(VFS
, "NFS: symlink(%s/%ld, %s, %s)\n", dir
->i_sb
->s_id
,
1293 dir
->i_ino
, dentry
->d_name
.name
, symname
);
1296 if (dentry
->d_inode
)
1297 printk("nfs_proc_symlink: %s/%s not negative!\n",
1298 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
);
1301 * Fill in the sattr for the call.
1302 * Note: SunOS 4.1.2 crashes if the mode isn't initialized!
1304 attr
.ia_valid
= ATTR_MODE
;
1305 attr
.ia_mode
= S_IFLNK
| S_IRWXUGO
;
1307 qsymname
.name
= symname
;
1308 qsymname
.len
= strlen(symname
);
1311 nfs_begin_data_update(dir
);
1312 error
= NFS_PROTO(dir
)->symlink(dir
, &dentry
->d_name
, &qsymname
,
1313 &attr
, &sym_fh
, &sym_attr
);
1314 nfs_end_data_update(dir
);
1316 error
= nfs_instantiate(dentry
, &sym_fh
, &sym_attr
);
1318 if (error
== -EEXIST
)
1319 printk("nfs_proc_symlink: %s/%s already exists??\n",
1320 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
);
1328 nfs_link(struct dentry
*old_dentry
, struct inode
*dir
, struct dentry
*dentry
)
1330 struct inode
*inode
= old_dentry
->d_inode
;
1333 dfprintk(VFS
, "NFS: link(%s/%s -> %s/%s)\n",
1334 old_dentry
->d_parent
->d_name
.name
, old_dentry
->d_name
.name
,
1335 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
);
1338 * Drop the dentry in advance to force a new lookup.
1339 * Since nfs_proc_link doesn't return a file handle,
1340 * we can't use the existing dentry.
1345 nfs_begin_data_update(dir
);
1346 nfs_begin_data_update(inode
);
1347 error
= NFS_PROTO(dir
)->link(inode
, dir
, &dentry
->d_name
);
1348 nfs_end_data_update(inode
);
1349 nfs_end_data_update(dir
);
1356 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a
1357 * different file handle for the same inode after a rename (e.g. when
1358 * moving to a different directory). A fail-safe method to do so would
1359 * be to look up old_dir/old_name, create a link to new_dir/new_name and
1360 * rename the old file using the sillyrename stuff. This way, the original
1361 * file in old_dir will go away when the last process iput()s the inode.
1365 * It actually works quite well. One needs to have the possibility for
1366 * at least one ".nfs..." file in each directory the file ever gets
1367 * moved or linked to which happens automagically with the new
1368 * implementation that only depends on the dcache stuff instead of
1369 * using the inode layer
1371 * Unfortunately, things are a little more complicated than indicated
1372 * above. For a cross-directory move, we want to make sure we can get
1373 * rid of the old inode after the operation. This means there must be
1374 * no pending writes (if it's a file), and the use count must be 1.
1375 * If these conditions are met, we can drop the dentries before doing
1378 static int nfs_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
1379 struct inode
*new_dir
, struct dentry
*new_dentry
)
1381 struct inode
*old_inode
= old_dentry
->d_inode
;
1382 struct inode
*new_inode
= new_dentry
->d_inode
;
1383 struct dentry
*dentry
= NULL
, *rehash
= NULL
;
1387 * To prevent any new references to the target during the rename,
1388 * we unhash the dentry and free the inode in advance.
1391 if (!d_unhashed(new_dentry
)) {
1393 rehash
= new_dentry
;
1396 dfprintk(VFS
, "NFS: rename(%s/%s -> %s/%s, ct=%d)\n",
1397 old_dentry
->d_parent
->d_name
.name
, old_dentry
->d_name
.name
,
1398 new_dentry
->d_parent
->d_name
.name
, new_dentry
->d_name
.name
,
1399 atomic_read(&new_dentry
->d_count
));
1402 * First check whether the target is busy ... we can't
1403 * safely do _any_ rename if the target is in use.
1405 * For files, make a copy of the dentry and then do a
1406 * silly-rename. If the silly-rename succeeds, the
1407 * copied dentry is hashed and becomes the new target.
1411 if (S_ISDIR(new_inode
->i_mode
))
1413 else if (atomic_read(&new_dentry
->d_count
) > 2) {
1415 /* copy the target dentry's name */
1416 dentry
= d_alloc(new_dentry
->d_parent
,
1417 &new_dentry
->d_name
);
1421 /* silly-rename the existing target ... */
1422 err
= nfs_sillyrename(new_dir
, new_dentry
);
1424 new_dentry
= rehash
= dentry
;
1426 /* instantiate the replacement target */
1427 d_instantiate(new_dentry
, NULL
);
1428 } else if (atomic_read(&new_dentry
->d_count
) > 1) {
1429 /* dentry still busy? */
1431 printk("nfs_rename: target %s/%s busy, d_count=%d\n",
1432 new_dentry
->d_parent
->d_name
.name
,
1433 new_dentry
->d_name
.name
,
1434 atomic_read(&new_dentry
->d_count
));
1442 * ... prune child dentries and writebacks if needed.
1444 if (atomic_read(&old_dentry
->d_count
) > 1) {
1445 nfs_wb_all(old_inode
);
1446 shrink_dcache_parent(old_dentry
);
1450 d_delete(new_dentry
);
1452 nfs_begin_data_update(old_dir
);
1453 nfs_begin_data_update(new_dir
);
1454 nfs_begin_data_update(old_inode
);
1455 error
= NFS_PROTO(old_dir
)->rename(old_dir
, &old_dentry
->d_name
,
1456 new_dir
, &new_dentry
->d_name
);
1457 nfs_end_data_update(old_inode
);
1458 nfs_end_data_update(new_dir
);
1459 nfs_end_data_update(old_dir
);
1464 if (!S_ISDIR(old_inode
->i_mode
))
1465 d_move(old_dentry
, new_dentry
);
1466 nfs_renew_times(new_dentry
);
1467 nfs_set_verifier(new_dentry
, nfs_save_change_attribute(new_dir
));
1470 /* new dentry created? */
1477 int nfs_access_get_cached(struct inode
*inode
, struct rpc_cred
*cred
, struct nfs_access_entry
*res
)
1479 struct nfs_access_entry
*cache
= &NFS_I(inode
)->cache_access
;
1481 if (cache
->cred
!= cred
1482 || time_after(jiffies
, cache
->jiffies
+ NFS_ATTRTIMEO(inode
))
1483 || (NFS_FLAGS(inode
) & NFS_INO_INVALID_ACCESS
))
1485 memcpy(res
, cache
, sizeof(*res
));
1489 void nfs_access_add_cache(struct inode
*inode
, struct nfs_access_entry
*set
)
1491 struct nfs_access_entry
*cache
= &NFS_I(inode
)->cache_access
;
1493 if (cache
->cred
!= set
->cred
) {
1495 put_rpccred(cache
->cred
);
1496 cache
->cred
= get_rpccred(set
->cred
);
1498 NFS_FLAGS(inode
) &= ~NFS_INO_INVALID_ACCESS
;
1499 cache
->jiffies
= set
->jiffies
;
1500 cache
->mask
= set
->mask
;
1503 static int nfs_do_access(struct inode
*inode
, struct rpc_cred
*cred
, int mask
)
1505 struct nfs_access_entry cache
;
1508 status
= nfs_access_get_cached(inode
, cred
, &cache
);
1512 /* Be clever: ask server to check for all possible rights */
1513 cache
.mask
= MAY_EXEC
| MAY_WRITE
| MAY_READ
;
1515 cache
.jiffies
= jiffies
;
1516 status
= NFS_PROTO(inode
)->access(inode
, &cache
);
1519 nfs_access_add_cache(inode
, &cache
);
1521 if ((cache
.mask
& mask
) == mask
)
1526 int nfs_permission(struct inode
*inode
, int mask
, struct nameidata
*nd
)
1528 struct rpc_cred
*cred
;
1533 /* Is this sys_access() ? */
1534 if (nd
!= NULL
&& (nd
->flags
& LOOKUP_ACCESS
))
1537 switch (inode
->i_mode
& S_IFMT
) {
1541 /* NFSv4 has atomic_open... */
1542 if (nfs_server_capable(inode
, NFS_CAP_ATOMIC_OPEN
)
1544 && (nd
->flags
& LOOKUP_OPEN
))
1549 * Optimize away all write operations, since the server
1550 * will check permissions when we perform the op.
1552 if ((mask
& MAY_WRITE
) && !(mask
& MAY_READ
))
1559 if (!NFS_PROTO(inode
)->access
)
1562 cred
= rpcauth_lookupcred(NFS_CLIENT(inode
)->cl_auth
, 0);
1563 if (!IS_ERR(cred
)) {
1564 res
= nfs_do_access(inode
, cred
, mask
);
1567 res
= PTR_ERR(cred
);
1572 res
= nfs_revalidate_inode(NFS_SERVER(inode
), inode
);
1574 res
= generic_permission(inode
, mask
, NULL
);
1581 * version-control: t
1582 * kept-new-versions: 5