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
,
94 .getxattr
= nfs4_getxattr
,
95 .setxattr
= nfs4_setxattr
,
96 .listxattr
= nfs4_listxattr
,
99 #endif /* CONFIG_NFS_V4 */
105 nfs_opendir(struct inode
*inode
, struct file
*filp
)
110 /* Call generic open code in order to cache credentials */
112 res
= nfs_open(inode
, filp
);
117 typedef u32
* (*decode_dirent_t
)(u32
*, struct nfs_entry
*, int);
121 unsigned long page_index
;
124 struct nfs_entry
*entry
;
125 decode_dirent_t decode
;
128 } nfs_readdir_descriptor_t
;
130 /* Now we cache directories properly, by stuffing the dirent
131 * data directly in the page cache.
133 * Inode invalidation due to refresh etc. takes care of
134 * _everything_, no sloppy entry flushing logic, no extraneous
135 * copying, network direct to page cache, the way it was meant
138 * NOTE: Dirent information verification is done always by the
139 * page-in of the RPC reply, nowhere else, this simplies
140 * things substantially.
143 int nfs_readdir_filler(nfs_readdir_descriptor_t
*desc
, struct page
*page
)
145 struct file
*file
= desc
->file
;
146 struct inode
*inode
= file
->f_dentry
->d_inode
;
147 struct rpc_cred
*cred
= nfs_file_cred(file
);
148 unsigned long timestamp
;
151 dfprintk(VFS
, "NFS: nfs_readdir_filler() reading cookie %Lu into page %lu.\n", (long long)desc
->entry
->cookie
, page
->index
);
155 error
= NFS_PROTO(inode
)->readdir(file
->f_dentry
, cred
, desc
->entry
->cookie
, page
,
156 NFS_SERVER(inode
)->dtsize
, desc
->plus
);
158 /* We requested READDIRPLUS, but the server doesn't grok it */
159 if (error
== -ENOTSUPP
&& desc
->plus
) {
160 NFS_SERVER(inode
)->caps
&= ~NFS_CAP_READDIRPLUS
;
161 NFS_FLAGS(inode
) &= ~NFS_INO_ADVISE_RDPLUS
;
167 SetPageUptodate(page
);
168 NFS_FLAGS(inode
) |= NFS_INO_INVALID_ATIME
;
169 /* Ensure consistent page alignment of the data.
170 * Note: assumes we have exclusive access to this mapping either
171 * through inode->i_sem or some other mechanism.
173 if (page
->index
== 0)
174 invalidate_inode_pages2_range(inode
->i_mapping
, PAGE_CACHE_SIZE
, -1);
180 nfs_zap_caches(inode
);
186 int dir_decode(nfs_readdir_descriptor_t
*desc
)
189 p
= desc
->decode(p
, desc
->entry
, desc
->plus
);
197 void dir_page_release(nfs_readdir_descriptor_t
*desc
)
200 page_cache_release(desc
->page
);
206 * Given a pointer to a buffer that has already been filled by a call
207 * to readdir, find the next entry.
209 * If the end of the buffer has been reached, return -EAGAIN, if not,
210 * return the offset within the buffer of the next entry to be
214 int find_dirent(nfs_readdir_descriptor_t
*desc
, struct page
*page
)
216 struct nfs_entry
*entry
= desc
->entry
;
220 while((status
= dir_decode(desc
)) == 0) {
221 dfprintk(VFS
, "NFS: found cookie %Lu\n", (long long)entry
->cookie
);
222 if (entry
->prev_cookie
== desc
->target
)
224 if (loop_count
++ > 200) {
229 dfprintk(VFS
, "NFS: find_dirent() returns %d\n", status
);
234 * Find the given page, and call find_dirent() in order to try to
235 * return the next entry.
238 int find_dirent_page(nfs_readdir_descriptor_t
*desc
)
240 struct inode
*inode
= desc
->file
->f_dentry
->d_inode
;
244 dfprintk(VFS
, "NFS: find_dirent_page() searching directory page %ld\n", desc
->page_index
);
246 page
= read_cache_page(inode
->i_mapping
, desc
->page_index
,
247 (filler_t
*)nfs_readdir_filler
, desc
);
249 status
= PTR_ERR(page
);
252 if (!PageUptodate(page
))
255 /* NOTE: Someone else may have changed the READDIRPLUS flag */
257 desc
->ptr
= kmap(page
); /* matching kunmap in nfs_do_filldir */
258 status
= find_dirent(desc
, page
);
260 dir_page_release(desc
);
262 dfprintk(VFS
, "NFS: find_dirent_page() returns %d\n", status
);
265 page_cache_release(page
);
270 * Recurse through the page cache pages, and return a
271 * filled nfs_entry structure of the next directory entry if possible.
273 * The target for the search is 'desc->target'.
276 int readdir_search_pagecache(nfs_readdir_descriptor_t
*desc
)
281 dfprintk(VFS
, "NFS: readdir_search_pagecache() searching for cookie %Lu\n", (long long)desc
->target
);
283 res
= find_dirent_page(desc
);
286 /* Align to beginning of next page */
288 if (loop_count
++ > 200) {
293 dfprintk(VFS
, "NFS: readdir_search_pagecache() returned %d\n", res
);
297 static inline unsigned int dt_type(struct inode
*inode
)
299 return (inode
->i_mode
>> 12) & 15;
302 static struct dentry
*nfs_readdir_lookup(nfs_readdir_descriptor_t
*desc
);
305 * Once we've found the start of the dirent within a page: fill 'er up...
308 int nfs_do_filldir(nfs_readdir_descriptor_t
*desc
, void *dirent
,
311 struct file
*file
= desc
->file
;
312 struct nfs_entry
*entry
= desc
->entry
;
313 struct dentry
*dentry
= NULL
;
314 unsigned long fileid
;
318 dfprintk(VFS
, "NFS: nfs_do_filldir() filling starting @ cookie %Lu\n", (long long)desc
->target
);
321 unsigned d_type
= DT_UNKNOWN
;
322 /* Note: entry->prev_cookie contains the cookie for
323 * retrieving the current dirent on the server */
324 fileid
= nfs_fileid_to_ino_t(entry
->ino
);
326 /* Get a dentry if we have one */
329 dentry
= nfs_readdir_lookup(desc
);
331 /* Use readdirplus info */
332 if (dentry
!= NULL
&& dentry
->d_inode
!= NULL
) {
333 d_type
= dt_type(dentry
->d_inode
);
334 fileid
= dentry
->d_inode
->i_ino
;
337 res
= filldir(dirent
, entry
->name
, entry
->len
,
338 entry
->prev_cookie
, fileid
, d_type
);
341 file
->f_pos
= desc
->target
= entry
->cookie
;
342 if (dir_decode(desc
) != 0) {
346 if (loop_count
++ > 200) {
351 dir_page_release(desc
);
354 dfprintk(VFS
, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n", (long long)desc
->target
, res
);
359 * If we cannot find a cookie in our cache, we suspect that this is
360 * because it points to a deleted file, so we ask the server to return
361 * whatever it thinks is the next entry. We then feed this to filldir.
362 * If all goes well, we should then be able to find our way round the
363 * cache on the next call to readdir_search_pagecache();
365 * NOTE: we cannot add the anonymous page to the pagecache because
366 * the data it contains might not be page aligned. Besides,
367 * we should already have a complete representation of the
368 * directory in the page cache by the time we get here.
371 int uncached_readdir(nfs_readdir_descriptor_t
*desc
, void *dirent
,
374 struct file
*file
= desc
->file
;
375 struct inode
*inode
= file
->f_dentry
->d_inode
;
376 struct rpc_cred
*cred
= nfs_file_cred(file
);
377 struct page
*page
= NULL
;
380 dfprintk(VFS
, "NFS: uncached_readdir() searching for cookie %Lu\n", (long long)desc
->target
);
382 page
= alloc_page(GFP_HIGHUSER
);
387 desc
->error
= NFS_PROTO(inode
)->readdir(file
->f_dentry
, cred
, desc
->target
,
389 NFS_SERVER(inode
)->dtsize
,
391 NFS_FLAGS(inode
) |= NFS_INO_INVALID_ATIME
;
393 desc
->ptr
= kmap(page
); /* matching kunmap in nfs_do_filldir */
394 if (desc
->error
>= 0) {
395 if ((status
= dir_decode(desc
)) == 0)
396 desc
->entry
->prev_cookie
= desc
->target
;
402 status
= nfs_do_filldir(desc
, dirent
, filldir
);
404 /* Reset read descriptor so it searches the page cache from
405 * the start upon the next call to readdir_search_pagecache() */
406 desc
->page_index
= 0;
407 desc
->entry
->cookie
= desc
->entry
->prev_cookie
= 0;
408 desc
->entry
->eof
= 0;
410 dfprintk(VFS
, "NFS: uncached_readdir() returns %d\n", status
);
413 dir_page_release(desc
);
417 /* The file offset position is now represented as a true offset into the
418 * page cache as is the case in most of the other filesystems.
420 static int nfs_readdir(struct file
*filp
, void *dirent
, filldir_t filldir
)
422 struct dentry
*dentry
= filp
->f_dentry
;
423 struct inode
*inode
= dentry
->d_inode
;
424 nfs_readdir_descriptor_t my_desc
,
426 struct nfs_entry my_entry
;
428 struct nfs_fattr fattr
;
433 res
= nfs_revalidate_inode(NFS_SERVER(inode
), inode
);
440 * filp->f_pos points to the file offset in the page cache.
441 * but if the cache has meanwhile been zapped, we need to
442 * read from the last dirent to revalidate f_pos
445 memset(desc
, 0, sizeof(*desc
));
448 desc
->target
= filp
->f_pos
;
449 desc
->decode
= NFS_PROTO(inode
)->decode_dirent
;
450 desc
->plus
= NFS_USE_READDIRPLUS(inode
);
452 my_entry
.cookie
= my_entry
.prev_cookie
= 0;
455 my_entry
.fattr
= &fattr
;
456 desc
->entry
= &my_entry
;
458 while(!desc
->entry
->eof
) {
459 res
= readdir_search_pagecache(desc
);
460 if (res
== -EBADCOOKIE
) {
461 /* This means either end of directory */
462 if (desc
->entry
->cookie
!= desc
->target
) {
463 /* Or that the server has 'lost' a cookie */
464 res
= uncached_readdir(desc
, dirent
, filldir
);
471 if (res
== -ETOOSMALL
&& desc
->plus
) {
472 NFS_FLAGS(inode
) &= ~NFS_INO_ADVISE_RDPLUS
;
473 nfs_zap_caches(inode
);
475 desc
->entry
->eof
= 0;
481 res
= nfs_do_filldir(desc
, dirent
, filldir
);
496 * All directory operations under NFS are synchronous, so fsync()
497 * is a dummy operation.
499 int nfs_fsync_dir(struct file
*filp
, struct dentry
*dentry
, int datasync
)
505 * A check for whether or not the parent directory has changed.
506 * In the case it has, we assume that the dentries are untrustworthy
507 * and may need to be looked up again.
509 static inline int nfs_check_verifier(struct inode
*dir
, struct dentry
*dentry
)
513 if ((NFS_FLAGS(dir
) & NFS_INO_INVALID_ATTR
) != 0
514 || nfs_attribute_timeout(dir
))
516 return nfs_verify_change_attribute(dir
, (unsigned long)dentry
->d_fsdata
);
519 static inline void nfs_set_verifier(struct dentry
* dentry
, unsigned long verf
)
521 dentry
->d_fsdata
= (void *)verf
;
525 * Whenever an NFS operation succeeds, we know that the dentry
526 * is valid, so we update the revalidation timestamp.
528 static inline void nfs_renew_times(struct dentry
* dentry
)
530 dentry
->d_time
= jiffies
;
534 * Return the intent data that applies to this particular path component
536 * Note that the current set of intents only apply to the very last
537 * component of the path.
538 * We check for this using LOOKUP_CONTINUE and LOOKUP_PARENT.
540 static inline unsigned int nfs_lookup_check_intent(struct nameidata
*nd
, unsigned int mask
)
542 if (nd
->flags
& (LOOKUP_CONTINUE
|LOOKUP_PARENT
))
544 return nd
->flags
& mask
;
548 * Inode and filehandle revalidation for lookups.
550 * We force revalidation in the cases where the VFS sets LOOKUP_REVAL,
551 * or if the intent information indicates that we're about to open this
552 * particular file and the "nocto" mount flag is not set.
556 int nfs_lookup_verify_inode(struct inode
*inode
, struct nameidata
*nd
)
558 struct nfs_server
*server
= NFS_SERVER(inode
);
561 /* VFS wants an on-the-wire revalidation */
562 if (nd
->flags
& LOOKUP_REVAL
)
564 /* This is an open(2) */
565 if (nfs_lookup_check_intent(nd
, LOOKUP_OPEN
) != 0 &&
566 !(server
->flags
& NFS_MOUNT_NOCTO
))
569 return nfs_revalidate_inode(server
, inode
);
571 return __nfs_revalidate_inode(server
, inode
);
575 * We judge how long we want to trust negative
576 * dentries by looking at the parent inode mtime.
578 * If parent mtime has changed, we revalidate, else we wait for a
579 * period corresponding to the parent's attribute cache timeout value.
582 int nfs_neg_need_reval(struct inode
*dir
, struct dentry
*dentry
,
583 struct nameidata
*nd
)
585 /* Don't revalidate a negative dentry if we're creating a new file */
586 if (nd
!= NULL
&& nfs_lookup_check_intent(nd
, LOOKUP_CREATE
) != 0)
588 return !nfs_check_verifier(dir
, dentry
);
592 * This is called every time the dcache has a lookup hit,
593 * and we should check whether we can really trust that
596 * NOTE! The hit can be a negative hit too, don't assume
599 * If the parent directory is seen to have changed, we throw out the
600 * cached dentry and do a new lookup.
602 static int nfs_lookup_revalidate(struct dentry
* dentry
, struct nameidata
*nd
)
606 struct dentry
*parent
;
608 struct nfs_fh fhandle
;
609 struct nfs_fattr fattr
;
610 unsigned long verifier
;
612 parent
= dget_parent(dentry
);
614 dir
= parent
->d_inode
;
615 inode
= dentry
->d_inode
;
618 if (nfs_neg_need_reval(dir
, dentry
, nd
))
623 if (is_bad_inode(inode
)) {
624 dfprintk(VFS
, "nfs_lookup_validate: %s/%s has dud inode\n",
625 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
);
629 /* Revalidate parent directory attribute cache */
630 if (nfs_revalidate_inode(NFS_SERVER(dir
), dir
) < 0)
633 /* Force a full look up iff the parent directory has changed */
634 if (nfs_check_verifier(dir
, dentry
)) {
635 if (nfs_lookup_verify_inode(inode
, nd
))
640 if (NFS_STALE(inode
))
643 verifier
= nfs_save_change_attribute(dir
);
644 error
= NFS_PROTO(dir
)->lookup(dir
, &dentry
->d_name
, &fhandle
, &fattr
);
647 if (nfs_compare_fh(NFS_FH(inode
), &fhandle
))
649 if ((error
= nfs_refresh_inode(inode
, &fattr
)) != 0)
652 nfs_renew_times(dentry
);
653 nfs_set_verifier(dentry
, verifier
);
662 if (inode
&& S_ISDIR(inode
->i_mode
)) {
663 /* Purge readdir caches. */
664 nfs_zap_caches(inode
);
665 /* If we have submounts, don't unhash ! */
666 if (have_submounts(dentry
))
668 shrink_dcache_parent(dentry
);
677 * This is called from dput() when d_count is going to 0.
679 static int nfs_dentry_delete(struct dentry
*dentry
)
681 dfprintk(VFS
, "NFS: dentry_delete(%s/%s, %x)\n",
682 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
,
685 if (dentry
->d_flags
& DCACHE_NFSFS_RENAMED
) {
686 /* Unhash it, so that ->d_iput() would be called */
689 if (!(dentry
->d_sb
->s_flags
& MS_ACTIVE
)) {
690 /* Unhash it, so that ancestors of killed async unlink
691 * files will be cleaned up during umount */
699 * Called when the dentry loses inode.
700 * We use it to clean up silly-renamed files.
702 static void nfs_dentry_iput(struct dentry
*dentry
, struct inode
*inode
)
704 if (dentry
->d_flags
& DCACHE_NFSFS_RENAMED
) {
707 nfs_complete_unlink(dentry
);
710 /* When creating a negative dentry, we want to renew d_time */
711 nfs_renew_times(dentry
);
715 struct dentry_operations nfs_dentry_operations
= {
716 .d_revalidate
= nfs_lookup_revalidate
,
717 .d_delete
= nfs_dentry_delete
,
718 .d_iput
= nfs_dentry_iput
,
722 * Use intent information to check whether or not we're going to do
723 * an O_EXCL create using this path component.
726 int nfs_is_exclusive_create(struct inode
*dir
, struct nameidata
*nd
)
728 if (NFS_PROTO(dir
)->version
== 2)
730 if (nd
== NULL
|| nfs_lookup_check_intent(nd
, LOOKUP_CREATE
) == 0)
732 return (nd
->intent
.open
.flags
& O_EXCL
) != 0;
735 static struct dentry
*nfs_lookup(struct inode
*dir
, struct dentry
* dentry
, struct nameidata
*nd
)
738 struct inode
*inode
= NULL
;
740 struct nfs_fh fhandle
;
741 struct nfs_fattr fattr
;
743 dfprintk(VFS
, "NFS: lookup(%s/%s)\n",
744 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
);
746 res
= ERR_PTR(-ENAMETOOLONG
);
747 if (dentry
->d_name
.len
> NFS_SERVER(dir
)->namelen
)
750 res
= ERR_PTR(-ENOMEM
);
751 dentry
->d_op
= NFS_PROTO(dir
)->dentry_ops
;
754 /* Revalidate parent directory attribute cache */
755 error
= nfs_revalidate_inode(NFS_SERVER(dir
), dir
);
757 res
= ERR_PTR(error
);
761 /* If we're doing an exclusive create, optimize away the lookup */
762 if (nfs_is_exclusive_create(dir
, nd
))
765 error
= NFS_PROTO(dir
)->lookup(dir
, &dentry
->d_name
, &fhandle
, &fattr
);
766 if (error
== -ENOENT
)
769 res
= ERR_PTR(error
);
772 res
= ERR_PTR(-EACCES
);
773 inode
= nfs_fhget(dentry
->d_sb
, &fhandle
, &fattr
);
777 res
= d_add_unique(dentry
, inode
);
780 nfs_renew_times(dentry
);
781 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
789 static int nfs_open_revalidate(struct dentry
*, struct nameidata
*);
791 struct dentry_operations nfs4_dentry_operations
= {
792 .d_revalidate
= nfs_open_revalidate
,
793 .d_delete
= nfs_dentry_delete
,
794 .d_iput
= nfs_dentry_iput
,
798 * Use intent information to determine whether we need to substitute
799 * the NFSv4-style stateful OPEN for the LOOKUP call
801 static int is_atomic_open(struct inode
*dir
, struct nameidata
*nd
)
803 if (nd
== NULL
|| nfs_lookup_check_intent(nd
, LOOKUP_OPEN
) == 0)
805 /* NFS does not (yet) have a stateful open for directories */
806 if (nd
->flags
& LOOKUP_DIRECTORY
)
808 /* Are we trying to write to a read only partition? */
809 if (IS_RDONLY(dir
) && (nd
->intent
.open
.flags
& (O_CREAT
|O_TRUNC
|FMODE_WRITE
)))
814 static struct dentry
*nfs_atomic_lookup(struct inode
*dir
, struct dentry
*dentry
, struct nameidata
*nd
)
816 struct dentry
*res
= NULL
;
817 struct inode
*inode
= NULL
;
820 /* Check that we are indeed trying to open this file */
821 if (!is_atomic_open(dir
, nd
))
824 if (dentry
->d_name
.len
> NFS_SERVER(dir
)->namelen
) {
825 res
= ERR_PTR(-ENAMETOOLONG
);
828 dentry
->d_op
= NFS_PROTO(dir
)->dentry_ops
;
830 /* Let vfs_create() deal with O_EXCL */
831 if (nd
->intent
.open
.flags
& O_EXCL
)
834 /* Open the file on the server */
836 /* Revalidate parent directory attribute cache */
837 error
= nfs_revalidate_inode(NFS_SERVER(dir
), dir
);
839 res
= ERR_PTR(error
);
843 if (nd
->intent
.open
.flags
& O_CREAT
) {
844 nfs_begin_data_update(dir
);
845 inode
= nfs4_atomic_open(dir
, dentry
, nd
);
846 nfs_end_data_update(dir
);
848 inode
= nfs4_atomic_open(dir
, dentry
, nd
);
851 error
= PTR_ERR(inode
);
853 /* Make a negative dentry */
857 /* This turned out not to be a regular file */
859 if (!(nd
->intent
.open
.flags
& O_NOFOLLOW
))
864 res
= ERR_PTR(error
);
869 res
= d_add_unique(dentry
, inode
);
872 nfs_renew_times(dentry
);
873 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
877 return nfs_lookup(dir
, dentry
, nd
);
880 static int nfs_open_revalidate(struct dentry
*dentry
, struct nameidata
*nd
)
882 struct dentry
*parent
= NULL
;
883 struct inode
*inode
= dentry
->d_inode
;
885 unsigned long verifier
;
886 int openflags
, ret
= 0;
888 parent
= dget_parent(dentry
);
889 dir
= parent
->d_inode
;
890 if (!is_atomic_open(dir
, nd
))
892 /* We can't create new files in nfs_open_revalidate(), so we
893 * optimize away revalidation of negative dentries.
897 /* NFS only supports OPEN on regular files */
898 if (!S_ISREG(inode
->i_mode
))
900 openflags
= nd
->intent
.open
.flags
;
901 /* We cannot do exclusive creation on a positive dentry */
902 if ((openflags
& (O_CREAT
|O_EXCL
)) == (O_CREAT
|O_EXCL
))
904 /* We can't create new files, or truncate existing ones here */
905 openflags
&= ~(O_CREAT
|O_TRUNC
);
908 * Note: we're not holding inode->i_sem and so may be racing with
909 * operations that change the directory. We therefore save the
910 * change attribute *before* we do the RPC call.
913 verifier
= nfs_save_change_attribute(dir
);
914 ret
= nfs4_open_revalidate(dir
, dentry
, openflags
);
916 nfs_set_verifier(dentry
, verifier
);
925 if (inode
!= NULL
&& nfs_have_delegation(inode
, FMODE_READ
))
927 return nfs_lookup_revalidate(dentry
, nd
);
929 #endif /* CONFIG_NFSV4 */
931 static struct dentry
*nfs_readdir_lookup(nfs_readdir_descriptor_t
*desc
)
933 struct dentry
*parent
= desc
->file
->f_dentry
;
934 struct inode
*dir
= parent
->d_inode
;
935 struct nfs_entry
*entry
= desc
->entry
;
936 struct dentry
*dentry
, *alias
;
945 if (name
.name
[0] == '.' && name
.name
[1] == '.')
946 return dget_parent(parent
);
949 if (name
.name
[0] == '.')
952 name
.hash
= full_name_hash(name
.name
, name
.len
);
953 dentry
= d_lookup(parent
, &name
);
956 if (!desc
->plus
|| !(entry
->fattr
->valid
& NFS_ATTR_FATTR
))
958 /* Note: caller is already holding the dir->i_sem! */
959 dentry
= d_alloc(parent
, &name
);
962 dentry
->d_op
= NFS_PROTO(dir
)->dentry_ops
;
963 inode
= nfs_fhget(dentry
->d_sb
, entry
->fh
, entry
->fattr
);
968 alias
= d_add_unique(dentry
, inode
);
973 nfs_renew_times(dentry
);
974 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
979 * Code common to create, mkdir, and mknod.
981 int nfs_instantiate(struct dentry
*dentry
, struct nfs_fh
*fhandle
,
982 struct nfs_fattr
*fattr
)
987 /* We may have been initialized further down */
990 if (fhandle
->size
== 0) {
991 struct inode
*dir
= dentry
->d_parent
->d_inode
;
992 error
= NFS_PROTO(dir
)->lookup(dir
, &dentry
->d_name
, fhandle
, fattr
);
996 if (!(fattr
->valid
& NFS_ATTR_FATTR
)) {
997 struct nfs_server
*server
= NFS_SB(dentry
->d_sb
);
998 error
= server
->rpc_ops
->getattr(server
, fhandle
, fattr
);
1003 inode
= nfs_fhget(dentry
->d_sb
, fhandle
, fattr
);
1006 d_instantiate(dentry
, inode
);
1014 * Following a failed create operation, we drop the dentry rather
1015 * than retain a negative dentry. This avoids a problem in the event
1016 * that the operation succeeded on the server, but an error in the
1017 * reply path made it appear to have failed.
1019 static int nfs_create(struct inode
*dir
, struct dentry
*dentry
, int mode
,
1020 struct nameidata
*nd
)
1026 dfprintk(VFS
, "NFS: create(%s/%ld, %s\n", dir
->i_sb
->s_id
,
1027 dir
->i_ino
, dentry
->d_name
.name
);
1029 attr
.ia_mode
= mode
;
1030 attr
.ia_valid
= ATTR_MODE
;
1032 if (nd
&& (nd
->flags
& LOOKUP_CREATE
))
1033 open_flags
= nd
->intent
.open
.flags
;
1036 nfs_begin_data_update(dir
);
1037 error
= NFS_PROTO(dir
)->create(dir
, dentry
, &attr
, open_flags
);
1038 nfs_end_data_update(dir
);
1041 nfs_renew_times(dentry
);
1042 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1052 * See comments for nfs_proc_create regarding failed operations.
1055 nfs_mknod(struct inode
*dir
, struct dentry
*dentry
, int mode
, dev_t rdev
)
1060 dfprintk(VFS
, "NFS: mknod(%s/%ld, %s\n", dir
->i_sb
->s_id
,
1061 dir
->i_ino
, dentry
->d_name
.name
);
1063 if (!new_valid_dev(rdev
))
1066 attr
.ia_mode
= mode
;
1067 attr
.ia_valid
= ATTR_MODE
;
1070 nfs_begin_data_update(dir
);
1071 status
= NFS_PROTO(dir
)->mknod(dir
, dentry
, &attr
, rdev
);
1072 nfs_end_data_update(dir
);
1075 nfs_renew_times(dentry
);
1076 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1086 * See comments for nfs_proc_create regarding failed operations.
1088 static int nfs_mkdir(struct inode
*dir
, struct dentry
*dentry
, int mode
)
1093 dfprintk(VFS
, "NFS: mkdir(%s/%ld, %s\n", dir
->i_sb
->s_id
,
1094 dir
->i_ino
, dentry
->d_name
.name
);
1096 attr
.ia_valid
= ATTR_MODE
;
1097 attr
.ia_mode
= mode
| S_IFDIR
;
1100 nfs_begin_data_update(dir
);
1101 error
= NFS_PROTO(dir
)->mkdir(dir
, dentry
, &attr
);
1102 nfs_end_data_update(dir
);
1105 nfs_renew_times(dentry
);
1106 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1115 static int nfs_rmdir(struct inode
*dir
, struct dentry
*dentry
)
1119 dfprintk(VFS
, "NFS: rmdir(%s/%ld, %s\n", dir
->i_sb
->s_id
,
1120 dir
->i_ino
, dentry
->d_name
.name
);
1123 nfs_begin_data_update(dir
);
1124 error
= NFS_PROTO(dir
)->rmdir(dir
, &dentry
->d_name
);
1125 /* Ensure the VFS deletes this inode */
1126 if (error
== 0 && dentry
->d_inode
!= NULL
)
1127 dentry
->d_inode
->i_nlink
= 0;
1128 nfs_end_data_update(dir
);
1134 static int nfs_sillyrename(struct inode
*dir
, struct dentry
*dentry
)
1136 static unsigned int sillycounter
;
1137 const int i_inosize
= sizeof(dir
->i_ino
)*2;
1138 const int countersize
= sizeof(sillycounter
)*2;
1139 const int slen
= sizeof(".nfs") + i_inosize
+ countersize
- 1;
1142 struct dentry
*sdentry
;
1145 dfprintk(VFS
, "NFS: silly-rename(%s/%s, ct=%d)\n",
1146 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
,
1147 atomic_read(&dentry
->d_count
));
1150 if (!dentry
->d_inode
)
1151 printk("NFS: silly-renaming %s/%s, negative dentry??\n",
1152 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
);
1155 * We don't allow a dentry to be silly-renamed twice.
1158 if (dentry
->d_flags
& DCACHE_NFSFS_RENAMED
)
1161 sprintf(silly
, ".nfs%*.*lx",
1162 i_inosize
, i_inosize
, dentry
->d_inode
->i_ino
);
1166 char *suffix
= silly
+ slen
- countersize
;
1170 sprintf(suffix
, "%*.*x", countersize
, countersize
, sillycounter
);
1172 dfprintk(VFS
, "trying to rename %s to %s\n",
1173 dentry
->d_name
.name
, silly
);
1175 sdentry
= lookup_one_len(silly
, dentry
->d_parent
, slen
);
1177 * N.B. Better to return EBUSY here ... it could be
1178 * dangerous to delete the file while it's in use.
1180 if (IS_ERR(sdentry
))
1182 } while(sdentry
->d_inode
!= NULL
); /* need negative lookup */
1184 qsilly
.name
= silly
;
1185 qsilly
.len
= strlen(silly
);
1186 nfs_begin_data_update(dir
);
1187 if (dentry
->d_inode
) {
1188 nfs_begin_data_update(dentry
->d_inode
);
1189 error
= NFS_PROTO(dir
)->rename(dir
, &dentry
->d_name
,
1191 nfs_end_data_update(dentry
->d_inode
);
1193 error
= NFS_PROTO(dir
)->rename(dir
, &dentry
->d_name
,
1195 nfs_end_data_update(dir
);
1197 nfs_renew_times(dentry
);
1198 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1199 d_move(dentry
, sdentry
);
1200 error
= nfs_async_unlink(dentry
);
1201 /* If we return 0 we don't unlink */
1209 * Remove a file after making sure there are no pending writes,
1210 * and after checking that the file has only one user.
1212 * We invalidate the attribute cache and free the inode prior to the operation
1213 * to avoid possible races if the server reuses the inode.
1215 static int nfs_safe_remove(struct dentry
*dentry
)
1217 struct inode
*dir
= dentry
->d_parent
->d_inode
;
1218 struct inode
*inode
= dentry
->d_inode
;
1221 dfprintk(VFS
, "NFS: safe_remove(%s/%s)\n",
1222 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
);
1224 /* If the dentry was sillyrenamed, we simply call d_delete() */
1225 if (dentry
->d_flags
& DCACHE_NFSFS_RENAMED
) {
1230 nfs_begin_data_update(dir
);
1231 if (inode
!= NULL
) {
1232 nfs_begin_data_update(inode
);
1233 error
= NFS_PROTO(dir
)->remove(dir
, &dentry
->d_name
);
1234 /* The VFS may want to delete this inode */
1237 nfs_end_data_update(inode
);
1239 error
= NFS_PROTO(dir
)->remove(dir
, &dentry
->d_name
);
1240 nfs_end_data_update(dir
);
1245 /* We do silly rename. In case sillyrename() returns -EBUSY, the inode
1246 * belongs to an active ".nfs..." file and we return -EBUSY.
1248 * If sillyrename() returns 0, we do nothing, otherwise we unlink.
1250 static int nfs_unlink(struct inode
*dir
, struct dentry
*dentry
)
1253 int need_rehash
= 0;
1255 dfprintk(VFS
, "NFS: unlink(%s/%ld, %s)\n", dir
->i_sb
->s_id
,
1256 dir
->i_ino
, dentry
->d_name
.name
);
1259 spin_lock(&dcache_lock
);
1260 spin_lock(&dentry
->d_lock
);
1261 if (atomic_read(&dentry
->d_count
) > 1) {
1262 spin_unlock(&dentry
->d_lock
);
1263 spin_unlock(&dcache_lock
);
1264 error
= nfs_sillyrename(dir
, dentry
);
1268 if (!d_unhashed(dentry
)) {
1272 spin_unlock(&dentry
->d_lock
);
1273 spin_unlock(&dcache_lock
);
1274 error
= nfs_safe_remove(dentry
);
1276 nfs_renew_times(dentry
);
1277 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1278 } else if (need_rehash
)
1285 nfs_symlink(struct inode
*dir
, struct dentry
*dentry
, const char *symname
)
1288 struct nfs_fattr sym_attr
;
1289 struct nfs_fh sym_fh
;
1290 struct qstr qsymname
;
1293 dfprintk(VFS
, "NFS: symlink(%s/%ld, %s, %s)\n", dir
->i_sb
->s_id
,
1294 dir
->i_ino
, dentry
->d_name
.name
, symname
);
1297 if (dentry
->d_inode
)
1298 printk("nfs_proc_symlink: %s/%s not negative!\n",
1299 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
);
1302 * Fill in the sattr for the call.
1303 * Note: SunOS 4.1.2 crashes if the mode isn't initialized!
1305 attr
.ia_valid
= ATTR_MODE
;
1306 attr
.ia_mode
= S_IFLNK
| S_IRWXUGO
;
1308 qsymname
.name
= symname
;
1309 qsymname
.len
= strlen(symname
);
1312 nfs_begin_data_update(dir
);
1313 error
= NFS_PROTO(dir
)->symlink(dir
, &dentry
->d_name
, &qsymname
,
1314 &attr
, &sym_fh
, &sym_attr
);
1315 nfs_end_data_update(dir
);
1317 error
= nfs_instantiate(dentry
, &sym_fh
, &sym_attr
);
1319 if (error
== -EEXIST
)
1320 printk("nfs_proc_symlink: %s/%s already exists??\n",
1321 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
);
1329 nfs_link(struct dentry
*old_dentry
, struct inode
*dir
, struct dentry
*dentry
)
1331 struct inode
*inode
= old_dentry
->d_inode
;
1334 dfprintk(VFS
, "NFS: link(%s/%s -> %s/%s)\n",
1335 old_dentry
->d_parent
->d_name
.name
, old_dentry
->d_name
.name
,
1336 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
);
1339 * Drop the dentry in advance to force a new lookup.
1340 * Since nfs_proc_link doesn't return a file handle,
1341 * we can't use the existing dentry.
1346 nfs_begin_data_update(dir
);
1347 nfs_begin_data_update(inode
);
1348 error
= NFS_PROTO(dir
)->link(inode
, dir
, &dentry
->d_name
);
1349 nfs_end_data_update(inode
);
1350 nfs_end_data_update(dir
);
1357 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a
1358 * different file handle for the same inode after a rename (e.g. when
1359 * moving to a different directory). A fail-safe method to do so would
1360 * be to look up old_dir/old_name, create a link to new_dir/new_name and
1361 * rename the old file using the sillyrename stuff. This way, the original
1362 * file in old_dir will go away when the last process iput()s the inode.
1366 * It actually works quite well. One needs to have the possibility for
1367 * at least one ".nfs..." file in each directory the file ever gets
1368 * moved or linked to which happens automagically with the new
1369 * implementation that only depends on the dcache stuff instead of
1370 * using the inode layer
1372 * Unfortunately, things are a little more complicated than indicated
1373 * above. For a cross-directory move, we want to make sure we can get
1374 * rid of the old inode after the operation. This means there must be
1375 * no pending writes (if it's a file), and the use count must be 1.
1376 * If these conditions are met, we can drop the dentries before doing
1379 static int nfs_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
1380 struct inode
*new_dir
, struct dentry
*new_dentry
)
1382 struct inode
*old_inode
= old_dentry
->d_inode
;
1383 struct inode
*new_inode
= new_dentry
->d_inode
;
1384 struct dentry
*dentry
= NULL
, *rehash
= NULL
;
1388 * To prevent any new references to the target during the rename,
1389 * we unhash the dentry and free the inode in advance.
1392 if (!d_unhashed(new_dentry
)) {
1394 rehash
= new_dentry
;
1397 dfprintk(VFS
, "NFS: rename(%s/%s -> %s/%s, ct=%d)\n",
1398 old_dentry
->d_parent
->d_name
.name
, old_dentry
->d_name
.name
,
1399 new_dentry
->d_parent
->d_name
.name
, new_dentry
->d_name
.name
,
1400 atomic_read(&new_dentry
->d_count
));
1403 * First check whether the target is busy ... we can't
1404 * safely do _any_ rename if the target is in use.
1406 * For files, make a copy of the dentry and then do a
1407 * silly-rename. If the silly-rename succeeds, the
1408 * copied dentry is hashed and becomes the new target.
1412 if (S_ISDIR(new_inode
->i_mode
))
1414 else if (atomic_read(&new_dentry
->d_count
) > 2) {
1416 /* copy the target dentry's name */
1417 dentry
= d_alloc(new_dentry
->d_parent
,
1418 &new_dentry
->d_name
);
1422 /* silly-rename the existing target ... */
1423 err
= nfs_sillyrename(new_dir
, new_dentry
);
1425 new_dentry
= rehash
= dentry
;
1427 /* instantiate the replacement target */
1428 d_instantiate(new_dentry
, NULL
);
1429 } else if (atomic_read(&new_dentry
->d_count
) > 1) {
1430 /* dentry still busy? */
1432 printk("nfs_rename: target %s/%s busy, d_count=%d\n",
1433 new_dentry
->d_parent
->d_name
.name
,
1434 new_dentry
->d_name
.name
,
1435 atomic_read(&new_dentry
->d_count
));
1443 * ... prune child dentries and writebacks if needed.
1445 if (atomic_read(&old_dentry
->d_count
) > 1) {
1446 nfs_wb_all(old_inode
);
1447 shrink_dcache_parent(old_dentry
);
1451 d_delete(new_dentry
);
1453 nfs_begin_data_update(old_dir
);
1454 nfs_begin_data_update(new_dir
);
1455 nfs_begin_data_update(old_inode
);
1456 error
= NFS_PROTO(old_dir
)->rename(old_dir
, &old_dentry
->d_name
,
1457 new_dir
, &new_dentry
->d_name
);
1458 nfs_end_data_update(old_inode
);
1459 nfs_end_data_update(new_dir
);
1460 nfs_end_data_update(old_dir
);
1465 if (!S_ISDIR(old_inode
->i_mode
))
1466 d_move(old_dentry
, new_dentry
);
1467 nfs_renew_times(new_dentry
);
1468 nfs_set_verifier(new_dentry
, nfs_save_change_attribute(new_dir
));
1471 /* new dentry created? */
1478 int nfs_access_get_cached(struct inode
*inode
, struct rpc_cred
*cred
, struct nfs_access_entry
*res
)
1480 struct nfs_access_entry
*cache
= &NFS_I(inode
)->cache_access
;
1482 if (cache
->cred
!= cred
1483 || time_after(jiffies
, cache
->jiffies
+ NFS_ATTRTIMEO(inode
))
1484 || (NFS_FLAGS(inode
) & NFS_INO_INVALID_ACCESS
))
1486 memcpy(res
, cache
, sizeof(*res
));
1490 void nfs_access_add_cache(struct inode
*inode
, struct nfs_access_entry
*set
)
1492 struct nfs_access_entry
*cache
= &NFS_I(inode
)->cache_access
;
1494 if (cache
->cred
!= set
->cred
) {
1496 put_rpccred(cache
->cred
);
1497 cache
->cred
= get_rpccred(set
->cred
);
1499 NFS_FLAGS(inode
) &= ~NFS_INO_INVALID_ACCESS
;
1500 cache
->jiffies
= set
->jiffies
;
1501 cache
->mask
= set
->mask
;
1504 static int nfs_do_access(struct inode
*inode
, struct rpc_cred
*cred
, int mask
)
1506 struct nfs_access_entry cache
;
1509 status
= nfs_access_get_cached(inode
, cred
, &cache
);
1513 /* Be clever: ask server to check for all possible rights */
1514 cache
.mask
= MAY_EXEC
| MAY_WRITE
| MAY_READ
;
1516 cache
.jiffies
= jiffies
;
1517 status
= NFS_PROTO(inode
)->access(inode
, &cache
);
1520 nfs_access_add_cache(inode
, &cache
);
1522 if ((cache
.mask
& mask
) == mask
)
1527 int nfs_permission(struct inode
*inode
, int mask
, struct nameidata
*nd
)
1529 struct rpc_cred
*cred
;
1534 /* Is this sys_access() ? */
1535 if (nd
!= NULL
&& (nd
->flags
& LOOKUP_ACCESS
))
1538 switch (inode
->i_mode
& S_IFMT
) {
1542 /* NFSv4 has atomic_open... */
1543 if (nfs_server_capable(inode
, NFS_CAP_ATOMIC_OPEN
)
1545 && (nd
->flags
& LOOKUP_OPEN
))
1550 * Optimize away all write operations, since the server
1551 * will check permissions when we perform the op.
1553 if ((mask
& MAY_WRITE
) && !(mask
& MAY_READ
))
1560 if (!NFS_PROTO(inode
)->access
)
1563 cred
= rpcauth_lookupcred(NFS_CLIENT(inode
)->cl_auth
, 0);
1564 if (!IS_ERR(cred
)) {
1565 res
= nfs_do_access(inode
, cred
, mask
);
1568 res
= PTR_ERR(cred
);
1573 res
= nfs_revalidate_inode(NFS_SERVER(inode
), inode
);
1575 res
= generic_permission(inode
, mask
, NULL
);
1582 * version-control: t
1583 * kept-new-versions: 5