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 * Return the intent data that applies to this particular path component
534 * Note that the current set of intents only apply to the very last
535 * component of the path.
536 * We check for this using LOOKUP_CONTINUE and LOOKUP_PARENT.
538 static inline unsigned int nfs_lookup_check_intent(struct nameidata
*nd
, unsigned int mask
)
540 if (nd
->flags
& (LOOKUP_CONTINUE
|LOOKUP_PARENT
))
542 return nd
->flags
& mask
;
546 * Inode and filehandle revalidation for lookups.
548 * We force revalidation in the cases where the VFS sets LOOKUP_REVAL,
549 * or if the intent information indicates that we're about to open this
550 * particular file and the "nocto" mount flag is not set.
554 int nfs_lookup_verify_inode(struct inode
*inode
, struct nameidata
*nd
)
556 struct nfs_server
*server
= NFS_SERVER(inode
);
559 /* VFS wants an on-the-wire revalidation */
560 if (nd
->flags
& LOOKUP_REVAL
)
562 /* This is an open(2) */
563 if (nfs_lookup_check_intent(nd
, LOOKUP_OPEN
) != 0 &&
564 !(server
->flags
& NFS_MOUNT_NOCTO
))
567 return nfs_revalidate_inode(server
, inode
);
569 return __nfs_revalidate_inode(server
, inode
);
573 * We judge how long we want to trust negative
574 * dentries by looking at the parent inode mtime.
576 * If parent mtime has changed, we revalidate, else we wait for a
577 * period corresponding to the parent's attribute cache timeout value.
580 int nfs_neg_need_reval(struct inode
*dir
, struct dentry
*dentry
,
581 struct nameidata
*nd
)
583 /* Don't revalidate a negative dentry if we're creating a new file */
584 if (nd
!= NULL
&& nfs_lookup_check_intent(nd
, LOOKUP_CREATE
) != 0)
586 return !nfs_check_verifier(dir
, dentry
);
590 * This is called every time the dcache has a lookup hit,
591 * and we should check whether we can really trust that
594 * NOTE! The hit can be a negative hit too, don't assume
597 * If the parent directory is seen to have changed, we throw out the
598 * cached dentry and do a new lookup.
600 static int nfs_lookup_revalidate(struct dentry
* dentry
, struct nameidata
*nd
)
604 struct dentry
*parent
;
606 struct nfs_fh fhandle
;
607 struct nfs_fattr fattr
;
608 unsigned long verifier
;
610 parent
= dget_parent(dentry
);
612 dir
= parent
->d_inode
;
613 inode
= dentry
->d_inode
;
616 if (nfs_neg_need_reval(dir
, dentry
, nd
))
621 if (is_bad_inode(inode
)) {
622 dfprintk(VFS
, "nfs_lookup_validate: %s/%s has dud inode\n",
623 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
);
627 /* Revalidate parent directory attribute cache */
628 if (nfs_revalidate_inode(NFS_SERVER(dir
), dir
) < 0)
631 /* Force a full look up iff the parent directory has changed */
632 if (nfs_check_verifier(dir
, dentry
)) {
633 if (nfs_lookup_verify_inode(inode
, nd
))
638 if (NFS_STALE(inode
))
641 verifier
= nfs_save_change_attribute(dir
);
642 error
= NFS_PROTO(dir
)->lookup(dir
, &dentry
->d_name
, &fhandle
, &fattr
);
645 if (nfs_compare_fh(NFS_FH(inode
), &fhandle
))
647 if ((error
= nfs_refresh_inode(inode
, &fattr
)) != 0)
650 nfs_renew_times(dentry
);
651 nfs_set_verifier(dentry
, verifier
);
660 if (inode
&& S_ISDIR(inode
->i_mode
)) {
661 /* Purge readdir caches. */
662 nfs_zap_caches(inode
);
663 /* If we have submounts, don't unhash ! */
664 if (have_submounts(dentry
))
666 shrink_dcache_parent(dentry
);
675 * This is called from dput() when d_count is going to 0.
677 static int nfs_dentry_delete(struct dentry
*dentry
)
679 dfprintk(VFS
, "NFS: dentry_delete(%s/%s, %x)\n",
680 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
,
683 if (dentry
->d_flags
& DCACHE_NFSFS_RENAMED
) {
684 /* Unhash it, so that ->d_iput() would be called */
687 if (!(dentry
->d_sb
->s_flags
& MS_ACTIVE
)) {
688 /* Unhash it, so that ancestors of killed async unlink
689 * files will be cleaned up during umount */
697 * Called when the dentry loses inode.
698 * We use it to clean up silly-renamed files.
700 static void nfs_dentry_iput(struct dentry
*dentry
, struct inode
*inode
)
702 if (dentry
->d_flags
& DCACHE_NFSFS_RENAMED
) {
705 nfs_complete_unlink(dentry
);
708 /* When creating a negative dentry, we want to renew d_time */
709 nfs_renew_times(dentry
);
713 struct dentry_operations nfs_dentry_operations
= {
714 .d_revalidate
= nfs_lookup_revalidate
,
715 .d_delete
= nfs_dentry_delete
,
716 .d_iput
= nfs_dentry_iput
,
720 * Use intent information to check whether or not we're going to do
721 * an O_EXCL create using this path component.
724 int nfs_is_exclusive_create(struct inode
*dir
, struct nameidata
*nd
)
726 if (NFS_PROTO(dir
)->version
== 2)
728 if (nd
== NULL
|| nfs_lookup_check_intent(nd
, LOOKUP_CREATE
) == 0)
730 return (nd
->intent
.open
.flags
& O_EXCL
) != 0;
733 static struct dentry
*nfs_lookup(struct inode
*dir
, struct dentry
* dentry
, struct nameidata
*nd
)
736 struct inode
*inode
= NULL
;
738 struct nfs_fh fhandle
;
739 struct nfs_fattr fattr
;
741 dfprintk(VFS
, "NFS: lookup(%s/%s)\n",
742 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
);
744 res
= ERR_PTR(-ENAMETOOLONG
);
745 if (dentry
->d_name
.len
> NFS_SERVER(dir
)->namelen
)
748 res
= ERR_PTR(-ENOMEM
);
749 dentry
->d_op
= NFS_PROTO(dir
)->dentry_ops
;
752 /* Revalidate parent directory attribute cache */
753 error
= nfs_revalidate_inode(NFS_SERVER(dir
), dir
);
755 res
= ERR_PTR(error
);
759 /* If we're doing an exclusive create, optimize away the lookup */
760 if (nfs_is_exclusive_create(dir
, nd
))
763 error
= NFS_PROTO(dir
)->lookup(dir
, &dentry
->d_name
, &fhandle
, &fattr
);
764 if (error
== -ENOENT
)
767 res
= ERR_PTR(error
);
770 res
= ERR_PTR(-EACCES
);
771 inode
= nfs_fhget(dentry
->d_sb
, &fhandle
, &fattr
);
775 res
= d_add_unique(dentry
, inode
);
778 nfs_renew_times(dentry
);
779 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
787 static int nfs_open_revalidate(struct dentry
*, struct nameidata
*);
789 struct dentry_operations nfs4_dentry_operations
= {
790 .d_revalidate
= nfs_open_revalidate
,
791 .d_delete
= nfs_dentry_delete
,
792 .d_iput
= nfs_dentry_iput
,
796 * Use intent information to determine whether we need to substitute
797 * the NFSv4-style stateful OPEN for the LOOKUP call
799 static int is_atomic_open(struct inode
*dir
, struct nameidata
*nd
)
801 if (nd
== NULL
|| nfs_lookup_check_intent(nd
, LOOKUP_OPEN
) == 0)
803 /* NFS does not (yet) have a stateful open for directories */
804 if (nd
->flags
& LOOKUP_DIRECTORY
)
806 /* Are we trying to write to a read only partition? */
807 if (IS_RDONLY(dir
) && (nd
->intent
.open
.flags
& (O_CREAT
|O_TRUNC
|FMODE_WRITE
)))
812 static struct dentry
*nfs_atomic_lookup(struct inode
*dir
, struct dentry
*dentry
, struct nameidata
*nd
)
814 struct dentry
*res
= NULL
;
815 struct inode
*inode
= NULL
;
818 /* Check that we are indeed trying to open this file */
819 if (!is_atomic_open(dir
, nd
))
822 if (dentry
->d_name
.len
> NFS_SERVER(dir
)->namelen
) {
823 res
= ERR_PTR(-ENAMETOOLONG
);
826 dentry
->d_op
= NFS_PROTO(dir
)->dentry_ops
;
828 /* Let vfs_create() deal with O_EXCL */
829 if (nd
->intent
.open
.flags
& O_EXCL
)
832 /* Open the file on the server */
834 /* Revalidate parent directory attribute cache */
835 error
= nfs_revalidate_inode(NFS_SERVER(dir
), dir
);
837 res
= ERR_PTR(error
);
841 if (nd
->intent
.open
.flags
& O_CREAT
) {
842 nfs_begin_data_update(dir
);
843 inode
= nfs4_atomic_open(dir
, dentry
, nd
);
844 nfs_end_data_update(dir
);
846 inode
= nfs4_atomic_open(dir
, dentry
, nd
);
849 error
= PTR_ERR(inode
);
851 /* Make a negative dentry */
855 /* This turned out not to be a regular file */
857 if (!(nd
->intent
.open
.flags
& O_NOFOLLOW
))
862 res
= ERR_PTR(error
);
867 res
= d_add_unique(dentry
, inode
);
870 nfs_renew_times(dentry
);
871 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
875 return nfs_lookup(dir
, dentry
, nd
);
878 static int nfs_open_revalidate(struct dentry
*dentry
, struct nameidata
*nd
)
880 struct dentry
*parent
= NULL
;
881 struct inode
*inode
= dentry
->d_inode
;
883 unsigned long verifier
;
884 int openflags
, ret
= 0;
886 parent
= dget_parent(dentry
);
887 dir
= parent
->d_inode
;
888 if (!is_atomic_open(dir
, nd
))
890 /* We can't create new files in nfs_open_revalidate(), so we
891 * optimize away revalidation of negative dentries.
895 /* NFS only supports OPEN on regular files */
896 if (!S_ISREG(inode
->i_mode
))
898 openflags
= nd
->intent
.open
.flags
;
899 /* We cannot do exclusive creation on a positive dentry */
900 if ((openflags
& (O_CREAT
|O_EXCL
)) == (O_CREAT
|O_EXCL
))
902 /* We can't create new files, or truncate existing ones here */
903 openflags
&= ~(O_CREAT
|O_TRUNC
);
906 * Note: we're not holding inode->i_sem and so may be racing with
907 * operations that change the directory. We therefore save the
908 * change attribute *before* we do the RPC call.
911 verifier
= nfs_save_change_attribute(dir
);
912 ret
= nfs4_open_revalidate(dir
, dentry
, openflags
);
914 nfs_set_verifier(dentry
, verifier
);
923 if (inode
!= NULL
&& nfs_have_delegation(inode
, FMODE_READ
))
925 return nfs_lookup_revalidate(dentry
, nd
);
927 #endif /* CONFIG_NFSV4 */
929 static struct dentry
*nfs_readdir_lookup(nfs_readdir_descriptor_t
*desc
)
931 struct dentry
*parent
= desc
->file
->f_dentry
;
932 struct inode
*dir
= parent
->d_inode
;
933 struct nfs_entry
*entry
= desc
->entry
;
934 struct dentry
*dentry
, *alias
;
943 if (name
.name
[0] == '.' && name
.name
[1] == '.')
944 return dget_parent(parent
);
947 if (name
.name
[0] == '.')
950 name
.hash
= full_name_hash(name
.name
, name
.len
);
951 dentry
= d_lookup(parent
, &name
);
954 if (!desc
->plus
|| !(entry
->fattr
->valid
& NFS_ATTR_FATTR
))
956 /* Note: caller is already holding the dir->i_sem! */
957 dentry
= d_alloc(parent
, &name
);
960 dentry
->d_op
= NFS_PROTO(dir
)->dentry_ops
;
961 inode
= nfs_fhget(dentry
->d_sb
, entry
->fh
, entry
->fattr
);
966 alias
= d_add_unique(dentry
, inode
);
971 nfs_renew_times(dentry
);
972 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
977 * Code common to create, mkdir, and mknod.
979 int nfs_instantiate(struct dentry
*dentry
, struct nfs_fh
*fhandle
,
980 struct nfs_fattr
*fattr
)
985 /* We may have been initialized further down */
988 if (fhandle
->size
== 0) {
989 struct inode
*dir
= dentry
->d_parent
->d_inode
;
990 error
= NFS_PROTO(dir
)->lookup(dir
, &dentry
->d_name
, fhandle
, fattr
);
994 if (!(fattr
->valid
& NFS_ATTR_FATTR
)) {
995 struct nfs_server
*server
= NFS_SB(dentry
->d_sb
);
996 error
= server
->rpc_ops
->getattr(server
, fhandle
, fattr
);
1001 inode
= nfs_fhget(dentry
->d_sb
, fhandle
, fattr
);
1004 d_instantiate(dentry
, inode
);
1012 * Following a failed create operation, we drop the dentry rather
1013 * than retain a negative dentry. This avoids a problem in the event
1014 * that the operation succeeded on the server, but an error in the
1015 * reply path made it appear to have failed.
1017 static int nfs_create(struct inode
*dir
, struct dentry
*dentry
, int mode
,
1018 struct nameidata
*nd
)
1024 dfprintk(VFS
, "NFS: create(%s/%ld, %s\n", dir
->i_sb
->s_id
,
1025 dir
->i_ino
, dentry
->d_name
.name
);
1027 attr
.ia_mode
= mode
;
1028 attr
.ia_valid
= ATTR_MODE
;
1030 if (nd
&& (nd
->flags
& LOOKUP_CREATE
))
1031 open_flags
= nd
->intent
.open
.flags
;
1034 nfs_begin_data_update(dir
);
1035 error
= NFS_PROTO(dir
)->create(dir
, dentry
, &attr
, open_flags
);
1036 nfs_end_data_update(dir
);
1039 nfs_renew_times(dentry
);
1040 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1050 * See comments for nfs_proc_create regarding failed operations.
1053 nfs_mknod(struct inode
*dir
, struct dentry
*dentry
, int mode
, dev_t rdev
)
1058 dfprintk(VFS
, "NFS: mknod(%s/%ld, %s\n", dir
->i_sb
->s_id
,
1059 dir
->i_ino
, dentry
->d_name
.name
);
1061 if (!new_valid_dev(rdev
))
1064 attr
.ia_mode
= mode
;
1065 attr
.ia_valid
= ATTR_MODE
;
1068 nfs_begin_data_update(dir
);
1069 status
= NFS_PROTO(dir
)->mknod(dir
, dentry
, &attr
, rdev
);
1070 nfs_end_data_update(dir
);
1073 nfs_renew_times(dentry
);
1074 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1084 * See comments for nfs_proc_create regarding failed operations.
1086 static int nfs_mkdir(struct inode
*dir
, struct dentry
*dentry
, int mode
)
1091 dfprintk(VFS
, "NFS: mkdir(%s/%ld, %s\n", dir
->i_sb
->s_id
,
1092 dir
->i_ino
, dentry
->d_name
.name
);
1094 attr
.ia_valid
= ATTR_MODE
;
1095 attr
.ia_mode
= mode
| S_IFDIR
;
1098 nfs_begin_data_update(dir
);
1099 error
= NFS_PROTO(dir
)->mkdir(dir
, dentry
, &attr
);
1100 nfs_end_data_update(dir
);
1103 nfs_renew_times(dentry
);
1104 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1113 static int nfs_rmdir(struct inode
*dir
, struct dentry
*dentry
)
1117 dfprintk(VFS
, "NFS: rmdir(%s/%ld, %s\n", dir
->i_sb
->s_id
,
1118 dir
->i_ino
, dentry
->d_name
.name
);
1121 nfs_begin_data_update(dir
);
1122 error
= NFS_PROTO(dir
)->rmdir(dir
, &dentry
->d_name
);
1123 /* Ensure the VFS deletes this inode */
1124 if (error
== 0 && dentry
->d_inode
!= NULL
)
1125 dentry
->d_inode
->i_nlink
= 0;
1126 nfs_end_data_update(dir
);
1132 static int nfs_sillyrename(struct inode
*dir
, struct dentry
*dentry
)
1134 static unsigned int sillycounter
;
1135 const int i_inosize
= sizeof(dir
->i_ino
)*2;
1136 const int countersize
= sizeof(sillycounter
)*2;
1137 const int slen
= sizeof(".nfs") + i_inosize
+ countersize
- 1;
1140 struct dentry
*sdentry
;
1143 dfprintk(VFS
, "NFS: silly-rename(%s/%s, ct=%d)\n",
1144 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
,
1145 atomic_read(&dentry
->d_count
));
1148 if (!dentry
->d_inode
)
1149 printk("NFS: silly-renaming %s/%s, negative dentry??\n",
1150 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
);
1153 * We don't allow a dentry to be silly-renamed twice.
1156 if (dentry
->d_flags
& DCACHE_NFSFS_RENAMED
)
1159 sprintf(silly
, ".nfs%*.*lx",
1160 i_inosize
, i_inosize
, dentry
->d_inode
->i_ino
);
1164 char *suffix
= silly
+ slen
- countersize
;
1168 sprintf(suffix
, "%*.*x", countersize
, countersize
, sillycounter
);
1170 dfprintk(VFS
, "trying to rename %s to %s\n",
1171 dentry
->d_name
.name
, silly
);
1173 sdentry
= lookup_one_len(silly
, dentry
->d_parent
, slen
);
1175 * N.B. Better to return EBUSY here ... it could be
1176 * dangerous to delete the file while it's in use.
1178 if (IS_ERR(sdentry
))
1180 } while(sdentry
->d_inode
!= NULL
); /* need negative lookup */
1182 qsilly
.name
= silly
;
1183 qsilly
.len
= strlen(silly
);
1184 nfs_begin_data_update(dir
);
1185 if (dentry
->d_inode
) {
1186 nfs_begin_data_update(dentry
->d_inode
);
1187 error
= NFS_PROTO(dir
)->rename(dir
, &dentry
->d_name
,
1189 nfs_end_data_update(dentry
->d_inode
);
1191 error
= NFS_PROTO(dir
)->rename(dir
, &dentry
->d_name
,
1193 nfs_end_data_update(dir
);
1195 nfs_renew_times(dentry
);
1196 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1197 d_move(dentry
, sdentry
);
1198 error
= nfs_async_unlink(dentry
);
1199 /* If we return 0 we don't unlink */
1207 * Remove a file after making sure there are no pending writes,
1208 * and after checking that the file has only one user.
1210 * We invalidate the attribute cache and free the inode prior to the operation
1211 * to avoid possible races if the server reuses the inode.
1213 static int nfs_safe_remove(struct dentry
*dentry
)
1215 struct inode
*dir
= dentry
->d_parent
->d_inode
;
1216 struct inode
*inode
= dentry
->d_inode
;
1219 dfprintk(VFS
, "NFS: safe_remove(%s/%s)\n",
1220 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
);
1222 /* If the dentry was sillyrenamed, we simply call d_delete() */
1223 if (dentry
->d_flags
& DCACHE_NFSFS_RENAMED
) {
1228 nfs_begin_data_update(dir
);
1229 if (inode
!= NULL
) {
1230 nfs_begin_data_update(inode
);
1231 error
= NFS_PROTO(dir
)->remove(dir
, &dentry
->d_name
);
1232 /* The VFS may want to delete this inode */
1235 nfs_end_data_update(inode
);
1237 error
= NFS_PROTO(dir
)->remove(dir
, &dentry
->d_name
);
1238 nfs_end_data_update(dir
);
1243 /* We do silly rename. In case sillyrename() returns -EBUSY, the inode
1244 * belongs to an active ".nfs..." file and we return -EBUSY.
1246 * If sillyrename() returns 0, we do nothing, otherwise we unlink.
1248 static int nfs_unlink(struct inode
*dir
, struct dentry
*dentry
)
1251 int need_rehash
= 0;
1253 dfprintk(VFS
, "NFS: unlink(%s/%ld, %s)\n", dir
->i_sb
->s_id
,
1254 dir
->i_ino
, dentry
->d_name
.name
);
1257 spin_lock(&dcache_lock
);
1258 spin_lock(&dentry
->d_lock
);
1259 if (atomic_read(&dentry
->d_count
) > 1) {
1260 spin_unlock(&dentry
->d_lock
);
1261 spin_unlock(&dcache_lock
);
1262 error
= nfs_sillyrename(dir
, dentry
);
1266 if (!d_unhashed(dentry
)) {
1270 spin_unlock(&dentry
->d_lock
);
1271 spin_unlock(&dcache_lock
);
1272 error
= nfs_safe_remove(dentry
);
1274 nfs_renew_times(dentry
);
1275 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1276 } else if (need_rehash
)
1283 nfs_symlink(struct inode
*dir
, struct dentry
*dentry
, const char *symname
)
1286 struct nfs_fattr sym_attr
;
1287 struct nfs_fh sym_fh
;
1288 struct qstr qsymname
;
1291 dfprintk(VFS
, "NFS: symlink(%s/%ld, %s, %s)\n", dir
->i_sb
->s_id
,
1292 dir
->i_ino
, dentry
->d_name
.name
, symname
);
1295 if (dentry
->d_inode
)
1296 printk("nfs_proc_symlink: %s/%s not negative!\n",
1297 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
);
1300 * Fill in the sattr for the call.
1301 * Note: SunOS 4.1.2 crashes if the mode isn't initialized!
1303 attr
.ia_valid
= ATTR_MODE
;
1304 attr
.ia_mode
= S_IFLNK
| S_IRWXUGO
;
1306 qsymname
.name
= symname
;
1307 qsymname
.len
= strlen(symname
);
1310 nfs_begin_data_update(dir
);
1311 error
= NFS_PROTO(dir
)->symlink(dir
, &dentry
->d_name
, &qsymname
,
1312 &attr
, &sym_fh
, &sym_attr
);
1313 nfs_end_data_update(dir
);
1315 error
= nfs_instantiate(dentry
, &sym_fh
, &sym_attr
);
1317 if (error
== -EEXIST
)
1318 printk("nfs_proc_symlink: %s/%s already exists??\n",
1319 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
);
1327 nfs_link(struct dentry
*old_dentry
, struct inode
*dir
, struct dentry
*dentry
)
1329 struct inode
*inode
= old_dentry
->d_inode
;
1332 dfprintk(VFS
, "NFS: link(%s/%s -> %s/%s)\n",
1333 old_dentry
->d_parent
->d_name
.name
, old_dentry
->d_name
.name
,
1334 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
);
1337 * Drop the dentry in advance to force a new lookup.
1338 * Since nfs_proc_link doesn't return a file handle,
1339 * we can't use the existing dentry.
1344 nfs_begin_data_update(dir
);
1345 nfs_begin_data_update(inode
);
1346 error
= NFS_PROTO(dir
)->link(inode
, dir
, &dentry
->d_name
);
1347 nfs_end_data_update(inode
);
1348 nfs_end_data_update(dir
);
1355 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a
1356 * different file handle for the same inode after a rename (e.g. when
1357 * moving to a different directory). A fail-safe method to do so would
1358 * be to look up old_dir/old_name, create a link to new_dir/new_name and
1359 * rename the old file using the sillyrename stuff. This way, the original
1360 * file in old_dir will go away when the last process iput()s the inode.
1364 * It actually works quite well. One needs to have the possibility for
1365 * at least one ".nfs..." file in each directory the file ever gets
1366 * moved or linked to which happens automagically with the new
1367 * implementation that only depends on the dcache stuff instead of
1368 * using the inode layer
1370 * Unfortunately, things are a little more complicated than indicated
1371 * above. For a cross-directory move, we want to make sure we can get
1372 * rid of the old inode after the operation. This means there must be
1373 * no pending writes (if it's a file), and the use count must be 1.
1374 * If these conditions are met, we can drop the dentries before doing
1377 static int nfs_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
1378 struct inode
*new_dir
, struct dentry
*new_dentry
)
1380 struct inode
*old_inode
= old_dentry
->d_inode
;
1381 struct inode
*new_inode
= new_dentry
->d_inode
;
1382 struct dentry
*dentry
= NULL
, *rehash
= NULL
;
1386 * To prevent any new references to the target during the rename,
1387 * we unhash the dentry and free the inode in advance.
1390 if (!d_unhashed(new_dentry
)) {
1392 rehash
= new_dentry
;
1395 dfprintk(VFS
, "NFS: rename(%s/%s -> %s/%s, ct=%d)\n",
1396 old_dentry
->d_parent
->d_name
.name
, old_dentry
->d_name
.name
,
1397 new_dentry
->d_parent
->d_name
.name
, new_dentry
->d_name
.name
,
1398 atomic_read(&new_dentry
->d_count
));
1401 * First check whether the target is busy ... we can't
1402 * safely do _any_ rename if the target is in use.
1404 * For files, make a copy of the dentry and then do a
1405 * silly-rename. If the silly-rename succeeds, the
1406 * copied dentry is hashed and becomes the new target.
1410 if (S_ISDIR(new_inode
->i_mode
))
1412 else if (atomic_read(&new_dentry
->d_count
) > 2) {
1414 /* copy the target dentry's name */
1415 dentry
= d_alloc(new_dentry
->d_parent
,
1416 &new_dentry
->d_name
);
1420 /* silly-rename the existing target ... */
1421 err
= nfs_sillyrename(new_dir
, new_dentry
);
1423 new_dentry
= rehash
= dentry
;
1425 /* instantiate the replacement target */
1426 d_instantiate(new_dentry
, NULL
);
1427 } else if (atomic_read(&new_dentry
->d_count
) > 1) {
1428 /* dentry still busy? */
1430 printk("nfs_rename: target %s/%s busy, d_count=%d\n",
1431 new_dentry
->d_parent
->d_name
.name
,
1432 new_dentry
->d_name
.name
,
1433 atomic_read(&new_dentry
->d_count
));
1441 * ... prune child dentries and writebacks if needed.
1443 if (atomic_read(&old_dentry
->d_count
) > 1) {
1444 nfs_wb_all(old_inode
);
1445 shrink_dcache_parent(old_dentry
);
1449 d_delete(new_dentry
);
1451 nfs_begin_data_update(old_dir
);
1452 nfs_begin_data_update(new_dir
);
1453 nfs_begin_data_update(old_inode
);
1454 error
= NFS_PROTO(old_dir
)->rename(old_dir
, &old_dentry
->d_name
,
1455 new_dir
, &new_dentry
->d_name
);
1456 nfs_end_data_update(old_inode
);
1457 nfs_end_data_update(new_dir
);
1458 nfs_end_data_update(old_dir
);
1463 if (!S_ISDIR(old_inode
->i_mode
))
1464 d_move(old_dentry
, new_dentry
);
1465 nfs_renew_times(new_dentry
);
1466 nfs_set_verifier(new_dentry
, nfs_save_change_attribute(new_dir
));
1469 /* new dentry created? */
1476 int nfs_access_get_cached(struct inode
*inode
, struct rpc_cred
*cred
, struct nfs_access_entry
*res
)
1478 struct nfs_access_entry
*cache
= &NFS_I(inode
)->cache_access
;
1480 if (cache
->cred
!= cred
1481 || time_after(jiffies
, cache
->jiffies
+ NFS_ATTRTIMEO(inode
))
1482 || (NFS_FLAGS(inode
) & NFS_INO_INVALID_ACCESS
))
1484 memcpy(res
, cache
, sizeof(*res
));
1488 void nfs_access_add_cache(struct inode
*inode
, struct nfs_access_entry
*set
)
1490 struct nfs_access_entry
*cache
= &NFS_I(inode
)->cache_access
;
1492 if (cache
->cred
!= set
->cred
) {
1494 put_rpccred(cache
->cred
);
1495 cache
->cred
= get_rpccred(set
->cred
);
1497 NFS_FLAGS(inode
) &= ~NFS_INO_INVALID_ACCESS
;
1498 cache
->jiffies
= set
->jiffies
;
1499 cache
->mask
= set
->mask
;
1502 static int nfs_do_access(struct inode
*inode
, struct rpc_cred
*cred
, int mask
)
1504 struct nfs_access_entry cache
;
1507 status
= nfs_access_get_cached(inode
, cred
, &cache
);
1511 /* Be clever: ask server to check for all possible rights */
1512 cache
.mask
= MAY_EXEC
| MAY_WRITE
| MAY_READ
;
1514 cache
.jiffies
= jiffies
;
1515 status
= NFS_PROTO(inode
)->access(inode
, &cache
);
1518 nfs_access_add_cache(inode
, &cache
);
1520 if ((cache
.mask
& mask
) == mask
)
1525 int nfs_permission(struct inode
*inode
, int mask
, struct nameidata
*nd
)
1527 struct rpc_cred
*cred
;
1532 /* Is this sys_access() ? */
1533 if (nd
!= NULL
&& (nd
->flags
& LOOKUP_ACCESS
))
1536 switch (inode
->i_mode
& S_IFMT
) {
1540 /* NFSv4 has atomic_open... */
1541 if (nfs_server_capable(inode
, NFS_CAP_ATOMIC_OPEN
)
1543 && (nd
->flags
& LOOKUP_OPEN
))
1548 * Optimize away all write operations, since the server
1549 * will check permissions when we perform the op.
1551 if ((mask
& MAY_WRITE
) && !(mask
& MAY_READ
))
1558 if (!NFS_PROTO(inode
)->access
)
1561 cred
= rpcauth_lookupcred(NFS_CLIENT(inode
)->cl_auth
, 0);
1562 if (!IS_ERR(cred
)) {
1563 res
= nfs_do_access(inode
, cred
, mask
);
1566 res
= PTR_ERR(cred
);
1571 res
= nfs_revalidate_inode(NFS_SERVER(inode
), inode
);
1573 res
= generic_permission(inode
, mask
, NULL
);
1580 * version-control: t
1581 * kept-new-versions: 5