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);
54 static loff_t
nfs_llseek_dir(struct file
*, loff_t
, int);
56 struct file_operations nfs_dir_operations
= {
57 .llseek
= nfs_llseek_dir
,
58 .read
= generic_read_dir
,
59 .readdir
= nfs_readdir
,
61 .release
= nfs_release
,
62 .fsync
= nfs_fsync_dir
,
65 struct inode_operations nfs_dir_inode_operations
= {
70 .symlink
= nfs_symlink
,
75 .permission
= nfs_permission
,
76 .getattr
= nfs_getattr
,
77 .setattr
= nfs_setattr
,
81 struct inode_operations nfs3_dir_inode_operations
= {
86 .symlink
= nfs_symlink
,
91 .permission
= nfs_permission
,
92 .getattr
= nfs_getattr
,
93 .setattr
= nfs_setattr
,
94 .listxattr
= nfs3_listxattr
,
95 .getxattr
= nfs3_getxattr
,
96 .setxattr
= nfs3_setxattr
,
97 .removexattr
= nfs3_removexattr
,
99 #endif /* CONFIG_NFS_V3 */
103 static struct dentry
*nfs_atomic_lookup(struct inode
*, struct dentry
*, struct nameidata
*);
104 struct inode_operations nfs4_dir_inode_operations
= {
105 .create
= nfs_create
,
106 .lookup
= nfs_atomic_lookup
,
108 .unlink
= nfs_unlink
,
109 .symlink
= nfs_symlink
,
113 .rename
= nfs_rename
,
114 .permission
= nfs_permission
,
115 .getattr
= nfs_getattr
,
116 .setattr
= nfs_setattr
,
117 .getxattr
= nfs4_getxattr
,
118 .setxattr
= nfs4_setxattr
,
119 .listxattr
= nfs4_listxattr
,
122 #endif /* CONFIG_NFS_V4 */
128 nfs_opendir(struct inode
*inode
, struct file
*filp
)
133 /* Call generic open code in order to cache credentials */
135 res
= nfs_open(inode
, filp
);
140 typedef u32
* (*decode_dirent_t
)(u32
*, struct nfs_entry
*, int);
144 unsigned long page_index
;
147 loff_t current_index
;
148 struct nfs_entry
*entry
;
149 decode_dirent_t decode
;
152 } nfs_readdir_descriptor_t
;
154 /* Now we cache directories properly, by stuffing the dirent
155 * data directly in the page cache.
157 * Inode invalidation due to refresh etc. takes care of
158 * _everything_, no sloppy entry flushing logic, no extraneous
159 * copying, network direct to page cache, the way it was meant
162 * NOTE: Dirent information verification is done always by the
163 * page-in of the RPC reply, nowhere else, this simplies
164 * things substantially.
167 int nfs_readdir_filler(nfs_readdir_descriptor_t
*desc
, struct page
*page
)
169 struct file
*file
= desc
->file
;
170 struct inode
*inode
= file
->f_dentry
->d_inode
;
171 struct rpc_cred
*cred
= nfs_file_cred(file
);
172 unsigned long timestamp
;
175 dfprintk(VFS
, "NFS: nfs_readdir_filler() reading cookie %Lu into page %lu.\n", (long long)desc
->entry
->cookie
, page
->index
);
179 error
= NFS_PROTO(inode
)->readdir(file
->f_dentry
, cred
, desc
->entry
->cookie
, page
,
180 NFS_SERVER(inode
)->dtsize
, desc
->plus
);
182 /* We requested READDIRPLUS, but the server doesn't grok it */
183 if (error
== -ENOTSUPP
&& desc
->plus
) {
184 NFS_SERVER(inode
)->caps
&= ~NFS_CAP_READDIRPLUS
;
185 NFS_FLAGS(inode
) &= ~NFS_INO_ADVISE_RDPLUS
;
191 SetPageUptodate(page
);
192 NFS_FLAGS(inode
) |= NFS_INO_INVALID_ATIME
;
193 /* Ensure consistent page alignment of the data.
194 * Note: assumes we have exclusive access to this mapping either
195 * through inode->i_sem or some other mechanism.
197 if (page
->index
== 0)
198 invalidate_inode_pages2_range(inode
->i_mapping
, PAGE_CACHE_SIZE
, -1);
204 nfs_zap_caches(inode
);
210 int dir_decode(nfs_readdir_descriptor_t
*desc
)
213 p
= desc
->decode(p
, desc
->entry
, desc
->plus
);
221 void dir_page_release(nfs_readdir_descriptor_t
*desc
)
224 page_cache_release(desc
->page
);
230 * Given a pointer to a buffer that has already been filled by a call
231 * to readdir, find the next entry with cookie '*desc->dir_cookie'.
233 * If the end of the buffer has been reached, return -EAGAIN, if not,
234 * return the offset within the buffer of the next entry to be
238 int find_dirent(nfs_readdir_descriptor_t
*desc
)
240 struct nfs_entry
*entry
= desc
->entry
;
244 while((status
= dir_decode(desc
)) == 0) {
245 dfprintk(VFS
, "NFS: found cookie %Lu\n", (unsigned long long)entry
->cookie
);
246 if (entry
->prev_cookie
== *desc
->dir_cookie
)
248 if (loop_count
++ > 200) {
253 dfprintk(VFS
, "NFS: find_dirent() returns %d\n", status
);
258 * Given a pointer to a buffer that has already been filled by a call
259 * to readdir, find the entry at offset 'desc->file->f_pos'.
261 * If the end of the buffer has been reached, return -EAGAIN, if not,
262 * return the offset within the buffer of the next entry to be
266 int find_dirent_index(nfs_readdir_descriptor_t
*desc
)
268 struct nfs_entry
*entry
= desc
->entry
;
273 status
= dir_decode(desc
);
277 dfprintk(VFS
, "NFS: found cookie %Lu at index %Ld\n", (unsigned long long)entry
->cookie
, desc
->current_index
);
279 if (desc
->file
->f_pos
== desc
->current_index
) {
280 *desc
->dir_cookie
= entry
->cookie
;
283 desc
->current_index
++;
284 if (loop_count
++ > 200) {
289 dfprintk(VFS
, "NFS: find_dirent_index() returns %d\n", status
);
294 * Find the given page, and call find_dirent() or find_dirent_index in
295 * order to try to return the next entry.
298 int find_dirent_page(nfs_readdir_descriptor_t
*desc
)
300 struct inode
*inode
= desc
->file
->f_dentry
->d_inode
;
304 dfprintk(VFS
, "NFS: find_dirent_page() searching directory page %ld\n", desc
->page_index
);
306 page
= read_cache_page(inode
->i_mapping
, desc
->page_index
,
307 (filler_t
*)nfs_readdir_filler
, desc
);
309 status
= PTR_ERR(page
);
312 if (!PageUptodate(page
))
315 /* NOTE: Someone else may have changed the READDIRPLUS flag */
317 desc
->ptr
= kmap(page
); /* matching kunmap in nfs_do_filldir */
318 if (*desc
->dir_cookie
!= 0)
319 status
= find_dirent(desc
);
321 status
= find_dirent_index(desc
);
323 dir_page_release(desc
);
325 dfprintk(VFS
, "NFS: find_dirent_page() returns %d\n", status
);
328 page_cache_release(page
);
333 * Recurse through the page cache pages, and return a
334 * filled nfs_entry structure of the next directory entry if possible.
336 * The target for the search is '*desc->dir_cookie' if non-0,
337 * 'desc->file->f_pos' otherwise
340 int readdir_search_pagecache(nfs_readdir_descriptor_t
*desc
)
345 /* Always search-by-index from the beginning of the cache */
346 if (*desc
->dir_cookie
== 0) {
347 dfprintk(VFS
, "NFS: readdir_search_pagecache() searching for offset %Ld\n", (long long)desc
->file
->f_pos
);
348 desc
->page_index
= 0;
349 desc
->entry
->cookie
= desc
->entry
->prev_cookie
= 0;
350 desc
->entry
->eof
= 0;
351 desc
->current_index
= 0;
353 dfprintk(VFS
, "NFS: readdir_search_pagecache() searching for cookie %Lu\n", (unsigned long long)*desc
->dir_cookie
);
356 res
= find_dirent_page(desc
);
359 /* Align to beginning of next page */
361 if (loop_count
++ > 200) {
366 dfprintk(VFS
, "NFS: readdir_search_pagecache() returned %d\n", res
);
370 static inline unsigned int dt_type(struct inode
*inode
)
372 return (inode
->i_mode
>> 12) & 15;
375 static struct dentry
*nfs_readdir_lookup(nfs_readdir_descriptor_t
*desc
);
378 * Once we've found the start of the dirent within a page: fill 'er up...
381 int nfs_do_filldir(nfs_readdir_descriptor_t
*desc
, void *dirent
,
384 struct file
*file
= desc
->file
;
385 struct nfs_entry
*entry
= desc
->entry
;
386 struct dentry
*dentry
= NULL
;
387 unsigned long fileid
;
391 dfprintk(VFS
, "NFS: nfs_do_filldir() filling starting @ cookie %Lu\n", (long long)entry
->cookie
);
394 unsigned d_type
= DT_UNKNOWN
;
395 /* Note: entry->prev_cookie contains the cookie for
396 * retrieving the current dirent on the server */
397 fileid
= nfs_fileid_to_ino_t(entry
->ino
);
399 /* Get a dentry if we have one */
402 dentry
= nfs_readdir_lookup(desc
);
404 /* Use readdirplus info */
405 if (dentry
!= NULL
&& dentry
->d_inode
!= NULL
) {
406 d_type
= dt_type(dentry
->d_inode
);
407 fileid
= dentry
->d_inode
->i_ino
;
410 res
= filldir(dirent
, entry
->name
, entry
->len
,
411 file
->f_pos
, fileid
, d_type
);
415 *desc
->dir_cookie
= entry
->cookie
;
416 if (dir_decode(desc
) != 0) {
420 if (loop_count
++ > 200) {
425 dir_page_release(desc
);
428 dfprintk(VFS
, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n", (unsigned long long)*desc
->dir_cookie
, res
);
433 * If we cannot find a cookie in our cache, we suspect that this is
434 * because it points to a deleted file, so we ask the server to return
435 * whatever it thinks is the next entry. We then feed this to filldir.
436 * If all goes well, we should then be able to find our way round the
437 * cache on the next call to readdir_search_pagecache();
439 * NOTE: we cannot add the anonymous page to the pagecache because
440 * the data it contains might not be page aligned. Besides,
441 * we should already have a complete representation of the
442 * directory in the page cache by the time we get here.
445 int uncached_readdir(nfs_readdir_descriptor_t
*desc
, void *dirent
,
448 struct file
*file
= desc
->file
;
449 struct inode
*inode
= file
->f_dentry
->d_inode
;
450 struct rpc_cred
*cred
= nfs_file_cred(file
);
451 struct page
*page
= NULL
;
454 dfprintk(VFS
, "NFS: uncached_readdir() searching for cookie %Lu\n", (unsigned long long)*desc
->dir_cookie
);
456 page
= alloc_page(GFP_HIGHUSER
);
461 desc
->error
= NFS_PROTO(inode
)->readdir(file
->f_dentry
, cred
, *desc
->dir_cookie
,
463 NFS_SERVER(inode
)->dtsize
,
465 NFS_FLAGS(inode
) |= NFS_INO_INVALID_ATIME
;
467 desc
->ptr
= kmap(page
); /* matching kunmap in nfs_do_filldir */
468 if (desc
->error
>= 0) {
469 if ((status
= dir_decode(desc
)) == 0)
470 desc
->entry
->prev_cookie
= *desc
->dir_cookie
;
476 status
= nfs_do_filldir(desc
, dirent
, filldir
);
478 /* Reset read descriptor so it searches the page cache from
479 * the start upon the next call to readdir_search_pagecache() */
480 desc
->page_index
= 0;
481 desc
->entry
->cookie
= desc
->entry
->prev_cookie
= 0;
482 desc
->entry
->eof
= 0;
484 dfprintk(VFS
, "NFS: uncached_readdir() returns %d\n", status
);
487 dir_page_release(desc
);
491 /* The file offset position represents the dirent entry number. A
492 last cookie cache takes care of the common case of reading the
495 static int nfs_readdir(struct file
*filp
, void *dirent
, filldir_t filldir
)
497 struct dentry
*dentry
= filp
->f_dentry
;
498 struct inode
*inode
= dentry
->d_inode
;
499 nfs_readdir_descriptor_t my_desc
,
501 struct nfs_entry my_entry
;
503 struct nfs_fattr fattr
;
508 res
= nfs_revalidate_inode(NFS_SERVER(inode
), inode
);
515 * filp->f_pos points to the dirent entry number.
516 * *desc->dir_cookie has the cookie for the next entry. We have
517 * to either find the entry with the appropriate number or
518 * revalidate the cookie.
520 memset(desc
, 0, sizeof(*desc
));
523 desc
->dir_cookie
= &((struct nfs_open_context
*)filp
->private_data
)->dir_cookie
;
524 desc
->decode
= NFS_PROTO(inode
)->decode_dirent
;
525 desc
->plus
= NFS_USE_READDIRPLUS(inode
);
527 my_entry
.cookie
= my_entry
.prev_cookie
= 0;
530 my_entry
.fattr
= &fattr
;
531 desc
->entry
= &my_entry
;
533 while(!desc
->entry
->eof
) {
534 res
= readdir_search_pagecache(desc
);
536 if (res
== -EBADCOOKIE
) {
537 /* This means either end of directory */
538 if (*desc
->dir_cookie
&& desc
->entry
->cookie
!= *desc
->dir_cookie
) {
539 /* Or that the server has 'lost' a cookie */
540 res
= uncached_readdir(desc
, dirent
, filldir
);
547 if (res
== -ETOOSMALL
&& desc
->plus
) {
548 NFS_FLAGS(inode
) &= ~NFS_INO_ADVISE_RDPLUS
;
549 nfs_zap_caches(inode
);
551 desc
->entry
->eof
= 0;
557 res
= nfs_do_filldir(desc
, dirent
, filldir
);
571 loff_t
nfs_llseek_dir(struct file
*filp
, loff_t offset
, int origin
)
573 down(&filp
->f_dentry
->d_inode
->i_sem
);
576 offset
+= filp
->f_pos
;
584 if (offset
!= filp
->f_pos
) {
585 filp
->f_pos
= offset
;
586 ((struct nfs_open_context
*)filp
->private_data
)->dir_cookie
= 0;
589 up(&filp
->f_dentry
->d_inode
->i_sem
);
594 * All directory operations under NFS are synchronous, so fsync()
595 * is a dummy operation.
597 int nfs_fsync_dir(struct file
*filp
, struct dentry
*dentry
, int datasync
)
603 * A check for whether or not the parent directory has changed.
604 * In the case it has, we assume that the dentries are untrustworthy
605 * and may need to be looked up again.
607 static inline int nfs_check_verifier(struct inode
*dir
, struct dentry
*dentry
)
611 if ((NFS_FLAGS(dir
) & NFS_INO_INVALID_ATTR
) != 0
612 || nfs_attribute_timeout(dir
))
614 return nfs_verify_change_attribute(dir
, (unsigned long)dentry
->d_fsdata
);
617 static inline void nfs_set_verifier(struct dentry
* dentry
, unsigned long verf
)
619 dentry
->d_fsdata
= (void *)verf
;
623 * Whenever an NFS operation succeeds, we know that the dentry
624 * is valid, so we update the revalidation timestamp.
626 static inline void nfs_renew_times(struct dentry
* dentry
)
628 dentry
->d_time
= jiffies
;
632 * Return the intent data that applies to this particular path component
634 * Note that the current set of intents only apply to the very last
635 * component of the path.
636 * We check for this using LOOKUP_CONTINUE and LOOKUP_PARENT.
638 static inline unsigned int nfs_lookup_check_intent(struct nameidata
*nd
, unsigned int mask
)
640 if (nd
->flags
& (LOOKUP_CONTINUE
|LOOKUP_PARENT
))
642 return nd
->flags
& mask
;
646 * Inode and filehandle revalidation for lookups.
648 * We force revalidation in the cases where the VFS sets LOOKUP_REVAL,
649 * or if the intent information indicates that we're about to open this
650 * particular file and the "nocto" mount flag is not set.
654 int nfs_lookup_verify_inode(struct inode
*inode
, struct nameidata
*nd
)
656 struct nfs_server
*server
= NFS_SERVER(inode
);
659 /* VFS wants an on-the-wire revalidation */
660 if (nd
->flags
& LOOKUP_REVAL
)
662 /* This is an open(2) */
663 if (nfs_lookup_check_intent(nd
, LOOKUP_OPEN
) != 0 &&
664 !(server
->flags
& NFS_MOUNT_NOCTO
))
667 return nfs_revalidate_inode(server
, inode
);
669 return __nfs_revalidate_inode(server
, inode
);
673 * We judge how long we want to trust negative
674 * dentries by looking at the parent inode mtime.
676 * If parent mtime has changed, we revalidate, else we wait for a
677 * period corresponding to the parent's attribute cache timeout value.
680 int nfs_neg_need_reval(struct inode
*dir
, struct dentry
*dentry
,
681 struct nameidata
*nd
)
683 /* Don't revalidate a negative dentry if we're creating a new file */
684 if (nd
!= NULL
&& nfs_lookup_check_intent(nd
, LOOKUP_CREATE
) != 0)
686 return !nfs_check_verifier(dir
, dentry
);
690 * This is called every time the dcache has a lookup hit,
691 * and we should check whether we can really trust that
694 * NOTE! The hit can be a negative hit too, don't assume
697 * If the parent directory is seen to have changed, we throw out the
698 * cached dentry and do a new lookup.
700 static int nfs_lookup_revalidate(struct dentry
* dentry
, struct nameidata
*nd
)
704 struct dentry
*parent
;
706 struct nfs_fh fhandle
;
707 struct nfs_fattr fattr
;
708 unsigned long verifier
;
710 parent
= dget_parent(dentry
);
712 dir
= parent
->d_inode
;
713 inode
= dentry
->d_inode
;
716 if (nfs_neg_need_reval(dir
, dentry
, nd
))
721 if (is_bad_inode(inode
)) {
722 dfprintk(VFS
, "nfs_lookup_validate: %s/%s has dud inode\n",
723 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
);
727 /* Revalidate parent directory attribute cache */
728 if (nfs_revalidate_inode(NFS_SERVER(dir
), dir
) < 0)
731 /* Force a full look up iff the parent directory has changed */
732 if (nfs_check_verifier(dir
, dentry
)) {
733 if (nfs_lookup_verify_inode(inode
, nd
))
738 if (NFS_STALE(inode
))
741 verifier
= nfs_save_change_attribute(dir
);
742 error
= NFS_PROTO(dir
)->lookup(dir
, &dentry
->d_name
, &fhandle
, &fattr
);
745 if (nfs_compare_fh(NFS_FH(inode
), &fhandle
))
747 if ((error
= nfs_refresh_inode(inode
, &fattr
)) != 0)
750 nfs_renew_times(dentry
);
751 nfs_set_verifier(dentry
, verifier
);
760 if (inode
&& S_ISDIR(inode
->i_mode
)) {
761 /* Purge readdir caches. */
762 nfs_zap_caches(inode
);
763 /* If we have submounts, don't unhash ! */
764 if (have_submounts(dentry
))
766 shrink_dcache_parent(dentry
);
775 * This is called from dput() when d_count is going to 0.
777 static int nfs_dentry_delete(struct dentry
*dentry
)
779 dfprintk(VFS
, "NFS: dentry_delete(%s/%s, %x)\n",
780 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
,
783 if (dentry
->d_flags
& DCACHE_NFSFS_RENAMED
) {
784 /* Unhash it, so that ->d_iput() would be called */
787 if (!(dentry
->d_sb
->s_flags
& MS_ACTIVE
)) {
788 /* Unhash it, so that ancestors of killed async unlink
789 * files will be cleaned up during umount */
797 * Called when the dentry loses inode.
798 * We use it to clean up silly-renamed files.
800 static void nfs_dentry_iput(struct dentry
*dentry
, struct inode
*inode
)
802 if (dentry
->d_flags
& DCACHE_NFSFS_RENAMED
) {
805 nfs_complete_unlink(dentry
);
808 /* When creating a negative dentry, we want to renew d_time */
809 nfs_renew_times(dentry
);
813 struct dentry_operations nfs_dentry_operations
= {
814 .d_revalidate
= nfs_lookup_revalidate
,
815 .d_delete
= nfs_dentry_delete
,
816 .d_iput
= nfs_dentry_iput
,
820 * Use intent information to check whether or not we're going to do
821 * an O_EXCL create using this path component.
824 int nfs_is_exclusive_create(struct inode
*dir
, struct nameidata
*nd
)
826 if (NFS_PROTO(dir
)->version
== 2)
828 if (nd
== NULL
|| nfs_lookup_check_intent(nd
, LOOKUP_CREATE
) == 0)
830 return (nd
->intent
.open
.flags
& O_EXCL
) != 0;
833 static struct dentry
*nfs_lookup(struct inode
*dir
, struct dentry
* dentry
, struct nameidata
*nd
)
836 struct inode
*inode
= NULL
;
838 struct nfs_fh fhandle
;
839 struct nfs_fattr fattr
;
841 dfprintk(VFS
, "NFS: lookup(%s/%s)\n",
842 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
);
844 res
= ERR_PTR(-ENAMETOOLONG
);
845 if (dentry
->d_name
.len
> NFS_SERVER(dir
)->namelen
)
848 res
= ERR_PTR(-ENOMEM
);
849 dentry
->d_op
= NFS_PROTO(dir
)->dentry_ops
;
852 /* Revalidate parent directory attribute cache */
853 error
= nfs_revalidate_inode(NFS_SERVER(dir
), dir
);
855 res
= ERR_PTR(error
);
859 /* If we're doing an exclusive create, optimize away the lookup */
860 if (nfs_is_exclusive_create(dir
, nd
))
863 error
= NFS_PROTO(dir
)->lookup(dir
, &dentry
->d_name
, &fhandle
, &fattr
);
864 if (error
== -ENOENT
)
867 res
= ERR_PTR(error
);
870 res
= ERR_PTR(-EACCES
);
871 inode
= nfs_fhget(dentry
->d_sb
, &fhandle
, &fattr
);
875 res
= d_add_unique(dentry
, inode
);
878 nfs_renew_times(dentry
);
879 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
887 static int nfs_open_revalidate(struct dentry
*, struct nameidata
*);
889 struct dentry_operations nfs4_dentry_operations
= {
890 .d_revalidate
= nfs_open_revalidate
,
891 .d_delete
= nfs_dentry_delete
,
892 .d_iput
= nfs_dentry_iput
,
896 * Use intent information to determine whether we need to substitute
897 * the NFSv4-style stateful OPEN for the LOOKUP call
899 static int is_atomic_open(struct inode
*dir
, struct nameidata
*nd
)
901 if (nd
== NULL
|| nfs_lookup_check_intent(nd
, LOOKUP_OPEN
) == 0)
903 /* NFS does not (yet) have a stateful open for directories */
904 if (nd
->flags
& LOOKUP_DIRECTORY
)
906 /* Are we trying to write to a read only partition? */
907 if (IS_RDONLY(dir
) && (nd
->intent
.open
.flags
& (O_CREAT
|O_TRUNC
|FMODE_WRITE
)))
912 static struct dentry
*nfs_atomic_lookup(struct inode
*dir
, struct dentry
*dentry
, struct nameidata
*nd
)
914 struct dentry
*res
= NULL
;
915 struct inode
*inode
= NULL
;
918 /* Check that we are indeed trying to open this file */
919 if (!is_atomic_open(dir
, nd
))
922 if (dentry
->d_name
.len
> NFS_SERVER(dir
)->namelen
) {
923 res
= ERR_PTR(-ENAMETOOLONG
);
926 dentry
->d_op
= NFS_PROTO(dir
)->dentry_ops
;
928 /* Let vfs_create() deal with O_EXCL */
929 if (nd
->intent
.open
.flags
& O_EXCL
)
932 /* Open the file on the server */
934 /* Revalidate parent directory attribute cache */
935 error
= nfs_revalidate_inode(NFS_SERVER(dir
), dir
);
937 res
= ERR_PTR(error
);
941 if (nd
->intent
.open
.flags
& O_CREAT
) {
942 nfs_begin_data_update(dir
);
943 inode
= nfs4_atomic_open(dir
, dentry
, nd
);
944 nfs_end_data_update(dir
);
946 inode
= nfs4_atomic_open(dir
, dentry
, nd
);
949 error
= PTR_ERR(inode
);
951 /* Make a negative dentry */
955 /* This turned out not to be a regular file */
957 if (!(nd
->intent
.open
.flags
& O_NOFOLLOW
))
962 res
= ERR_PTR(error
);
967 res
= d_add_unique(dentry
, inode
);
970 nfs_renew_times(dentry
);
971 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
975 return nfs_lookup(dir
, dentry
, nd
);
978 static int nfs_open_revalidate(struct dentry
*dentry
, struct nameidata
*nd
)
980 struct dentry
*parent
= NULL
;
981 struct inode
*inode
= dentry
->d_inode
;
983 unsigned long verifier
;
984 int openflags
, ret
= 0;
986 parent
= dget_parent(dentry
);
987 dir
= parent
->d_inode
;
988 if (!is_atomic_open(dir
, nd
))
990 /* We can't create new files in nfs_open_revalidate(), so we
991 * optimize away revalidation of negative dentries.
995 /* NFS only supports OPEN on regular files */
996 if (!S_ISREG(inode
->i_mode
))
998 openflags
= nd
->intent
.open
.flags
;
999 /* We cannot do exclusive creation on a positive dentry */
1000 if ((openflags
& (O_CREAT
|O_EXCL
)) == (O_CREAT
|O_EXCL
))
1002 /* We can't create new files, or truncate existing ones here */
1003 openflags
&= ~(O_CREAT
|O_TRUNC
);
1006 * Note: we're not holding inode->i_sem and so may be racing with
1007 * operations that change the directory. We therefore save the
1008 * change attribute *before* we do the RPC call.
1011 verifier
= nfs_save_change_attribute(dir
);
1012 ret
= nfs4_open_revalidate(dir
, dentry
, openflags
);
1014 nfs_set_verifier(dentry
, verifier
);
1023 if (inode
!= NULL
&& nfs_have_delegation(inode
, FMODE_READ
))
1025 return nfs_lookup_revalidate(dentry
, nd
);
1027 #endif /* CONFIG_NFSV4 */
1029 static struct dentry
*nfs_readdir_lookup(nfs_readdir_descriptor_t
*desc
)
1031 struct dentry
*parent
= desc
->file
->f_dentry
;
1032 struct inode
*dir
= parent
->d_inode
;
1033 struct nfs_entry
*entry
= desc
->entry
;
1034 struct dentry
*dentry
, *alias
;
1035 struct qstr name
= {
1036 .name
= entry
->name
,
1039 struct inode
*inode
;
1043 if (name
.name
[0] == '.' && name
.name
[1] == '.')
1044 return dget_parent(parent
);
1047 if (name
.name
[0] == '.')
1048 return dget(parent
);
1050 name
.hash
= full_name_hash(name
.name
, name
.len
);
1051 dentry
= d_lookup(parent
, &name
);
1054 if (!desc
->plus
|| !(entry
->fattr
->valid
& NFS_ATTR_FATTR
))
1056 /* Note: caller is already holding the dir->i_sem! */
1057 dentry
= d_alloc(parent
, &name
);
1060 dentry
->d_op
= NFS_PROTO(dir
)->dentry_ops
;
1061 inode
= nfs_fhget(dentry
->d_sb
, entry
->fh
, entry
->fattr
);
1066 alias
= d_add_unique(dentry
, inode
);
1067 if (alias
!= NULL
) {
1071 nfs_renew_times(dentry
);
1072 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1077 * Code common to create, mkdir, and mknod.
1079 int nfs_instantiate(struct dentry
*dentry
, struct nfs_fh
*fhandle
,
1080 struct nfs_fattr
*fattr
)
1082 struct inode
*inode
;
1083 int error
= -EACCES
;
1085 /* We may have been initialized further down */
1086 if (dentry
->d_inode
)
1088 if (fhandle
->size
== 0) {
1089 struct inode
*dir
= dentry
->d_parent
->d_inode
;
1090 error
= NFS_PROTO(dir
)->lookup(dir
, &dentry
->d_name
, fhandle
, fattr
);
1094 if (!(fattr
->valid
& NFS_ATTR_FATTR
)) {
1095 struct nfs_server
*server
= NFS_SB(dentry
->d_sb
);
1096 error
= server
->rpc_ops
->getattr(server
, fhandle
, fattr
);
1101 inode
= nfs_fhget(dentry
->d_sb
, fhandle
, fattr
);
1104 d_instantiate(dentry
, inode
);
1112 * Following a failed create operation, we drop the dentry rather
1113 * than retain a negative dentry. This avoids a problem in the event
1114 * that the operation succeeded on the server, but an error in the
1115 * reply path made it appear to have failed.
1117 static int nfs_create(struct inode
*dir
, struct dentry
*dentry
, int mode
,
1118 struct nameidata
*nd
)
1124 dfprintk(VFS
, "NFS: create(%s/%ld, %s\n", dir
->i_sb
->s_id
,
1125 dir
->i_ino
, dentry
->d_name
.name
);
1127 attr
.ia_mode
= mode
;
1128 attr
.ia_valid
= ATTR_MODE
;
1130 if (nd
&& (nd
->flags
& LOOKUP_CREATE
))
1131 open_flags
= nd
->intent
.open
.flags
;
1134 nfs_begin_data_update(dir
);
1135 error
= NFS_PROTO(dir
)->create(dir
, dentry
, &attr
, open_flags
);
1136 nfs_end_data_update(dir
);
1139 nfs_renew_times(dentry
);
1140 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1150 * See comments for nfs_proc_create regarding failed operations.
1153 nfs_mknod(struct inode
*dir
, struct dentry
*dentry
, int mode
, dev_t rdev
)
1158 dfprintk(VFS
, "NFS: mknod(%s/%ld, %s\n", dir
->i_sb
->s_id
,
1159 dir
->i_ino
, dentry
->d_name
.name
);
1161 if (!new_valid_dev(rdev
))
1164 attr
.ia_mode
= mode
;
1165 attr
.ia_valid
= ATTR_MODE
;
1168 nfs_begin_data_update(dir
);
1169 status
= NFS_PROTO(dir
)->mknod(dir
, dentry
, &attr
, rdev
);
1170 nfs_end_data_update(dir
);
1173 nfs_renew_times(dentry
);
1174 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1184 * See comments for nfs_proc_create regarding failed operations.
1186 static int nfs_mkdir(struct inode
*dir
, struct dentry
*dentry
, int mode
)
1191 dfprintk(VFS
, "NFS: mkdir(%s/%ld, %s\n", dir
->i_sb
->s_id
,
1192 dir
->i_ino
, dentry
->d_name
.name
);
1194 attr
.ia_valid
= ATTR_MODE
;
1195 attr
.ia_mode
= mode
| S_IFDIR
;
1198 nfs_begin_data_update(dir
);
1199 error
= NFS_PROTO(dir
)->mkdir(dir
, dentry
, &attr
);
1200 nfs_end_data_update(dir
);
1203 nfs_renew_times(dentry
);
1204 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1213 static int nfs_rmdir(struct inode
*dir
, struct dentry
*dentry
)
1217 dfprintk(VFS
, "NFS: rmdir(%s/%ld, %s\n", dir
->i_sb
->s_id
,
1218 dir
->i_ino
, dentry
->d_name
.name
);
1221 nfs_begin_data_update(dir
);
1222 error
= NFS_PROTO(dir
)->rmdir(dir
, &dentry
->d_name
);
1223 /* Ensure the VFS deletes this inode */
1224 if (error
== 0 && dentry
->d_inode
!= NULL
)
1225 dentry
->d_inode
->i_nlink
= 0;
1226 nfs_end_data_update(dir
);
1232 static int nfs_sillyrename(struct inode
*dir
, struct dentry
*dentry
)
1234 static unsigned int sillycounter
;
1235 const int i_inosize
= sizeof(dir
->i_ino
)*2;
1236 const int countersize
= sizeof(sillycounter
)*2;
1237 const int slen
= sizeof(".nfs") + i_inosize
+ countersize
- 1;
1240 struct dentry
*sdentry
;
1243 dfprintk(VFS
, "NFS: silly-rename(%s/%s, ct=%d)\n",
1244 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
,
1245 atomic_read(&dentry
->d_count
));
1248 if (!dentry
->d_inode
)
1249 printk("NFS: silly-renaming %s/%s, negative dentry??\n",
1250 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
);
1253 * We don't allow a dentry to be silly-renamed twice.
1256 if (dentry
->d_flags
& DCACHE_NFSFS_RENAMED
)
1259 sprintf(silly
, ".nfs%*.*lx",
1260 i_inosize
, i_inosize
, dentry
->d_inode
->i_ino
);
1264 char *suffix
= silly
+ slen
- countersize
;
1268 sprintf(suffix
, "%*.*x", countersize
, countersize
, sillycounter
);
1270 dfprintk(VFS
, "trying to rename %s to %s\n",
1271 dentry
->d_name
.name
, silly
);
1273 sdentry
= lookup_one_len(silly
, dentry
->d_parent
, slen
);
1275 * N.B. Better to return EBUSY here ... it could be
1276 * dangerous to delete the file while it's in use.
1278 if (IS_ERR(sdentry
))
1280 } while(sdentry
->d_inode
!= NULL
); /* need negative lookup */
1282 qsilly
.name
= silly
;
1283 qsilly
.len
= strlen(silly
);
1284 nfs_begin_data_update(dir
);
1285 if (dentry
->d_inode
) {
1286 nfs_begin_data_update(dentry
->d_inode
);
1287 error
= NFS_PROTO(dir
)->rename(dir
, &dentry
->d_name
,
1289 nfs_end_data_update(dentry
->d_inode
);
1291 error
= NFS_PROTO(dir
)->rename(dir
, &dentry
->d_name
,
1293 nfs_end_data_update(dir
);
1295 nfs_renew_times(dentry
);
1296 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1297 d_move(dentry
, sdentry
);
1298 error
= nfs_async_unlink(dentry
);
1299 /* If we return 0 we don't unlink */
1307 * Remove a file after making sure there are no pending writes,
1308 * and after checking that the file has only one user.
1310 * We invalidate the attribute cache and free the inode prior to the operation
1311 * to avoid possible races if the server reuses the inode.
1313 static int nfs_safe_remove(struct dentry
*dentry
)
1315 struct inode
*dir
= dentry
->d_parent
->d_inode
;
1316 struct inode
*inode
= dentry
->d_inode
;
1319 dfprintk(VFS
, "NFS: safe_remove(%s/%s)\n",
1320 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
);
1322 /* If the dentry was sillyrenamed, we simply call d_delete() */
1323 if (dentry
->d_flags
& DCACHE_NFSFS_RENAMED
) {
1328 nfs_begin_data_update(dir
);
1329 if (inode
!= NULL
) {
1330 nfs_begin_data_update(inode
);
1331 error
= NFS_PROTO(dir
)->remove(dir
, &dentry
->d_name
);
1332 /* The VFS may want to delete this inode */
1335 nfs_end_data_update(inode
);
1337 error
= NFS_PROTO(dir
)->remove(dir
, &dentry
->d_name
);
1338 nfs_end_data_update(dir
);
1343 /* We do silly rename. In case sillyrename() returns -EBUSY, the inode
1344 * belongs to an active ".nfs..." file and we return -EBUSY.
1346 * If sillyrename() returns 0, we do nothing, otherwise we unlink.
1348 static int nfs_unlink(struct inode
*dir
, struct dentry
*dentry
)
1351 int need_rehash
= 0;
1353 dfprintk(VFS
, "NFS: unlink(%s/%ld, %s)\n", dir
->i_sb
->s_id
,
1354 dir
->i_ino
, dentry
->d_name
.name
);
1357 spin_lock(&dcache_lock
);
1358 spin_lock(&dentry
->d_lock
);
1359 if (atomic_read(&dentry
->d_count
) > 1) {
1360 spin_unlock(&dentry
->d_lock
);
1361 spin_unlock(&dcache_lock
);
1362 error
= nfs_sillyrename(dir
, dentry
);
1366 if (!d_unhashed(dentry
)) {
1370 spin_unlock(&dentry
->d_lock
);
1371 spin_unlock(&dcache_lock
);
1372 error
= nfs_safe_remove(dentry
);
1374 nfs_renew_times(dentry
);
1375 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1376 } else if (need_rehash
)
1383 nfs_symlink(struct inode
*dir
, struct dentry
*dentry
, const char *symname
)
1386 struct nfs_fattr sym_attr
;
1387 struct nfs_fh sym_fh
;
1388 struct qstr qsymname
;
1391 dfprintk(VFS
, "NFS: symlink(%s/%ld, %s, %s)\n", dir
->i_sb
->s_id
,
1392 dir
->i_ino
, dentry
->d_name
.name
, symname
);
1395 if (dentry
->d_inode
)
1396 printk("nfs_proc_symlink: %s/%s not negative!\n",
1397 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
);
1400 * Fill in the sattr for the call.
1401 * Note: SunOS 4.1.2 crashes if the mode isn't initialized!
1403 attr
.ia_valid
= ATTR_MODE
;
1404 attr
.ia_mode
= S_IFLNK
| S_IRWXUGO
;
1406 qsymname
.name
= symname
;
1407 qsymname
.len
= strlen(symname
);
1410 nfs_begin_data_update(dir
);
1411 error
= NFS_PROTO(dir
)->symlink(dir
, &dentry
->d_name
, &qsymname
,
1412 &attr
, &sym_fh
, &sym_attr
);
1413 nfs_end_data_update(dir
);
1415 error
= nfs_instantiate(dentry
, &sym_fh
, &sym_attr
);
1417 if (error
== -EEXIST
)
1418 printk("nfs_proc_symlink: %s/%s already exists??\n",
1419 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
);
1427 nfs_link(struct dentry
*old_dentry
, struct inode
*dir
, struct dentry
*dentry
)
1429 struct inode
*inode
= old_dentry
->d_inode
;
1432 dfprintk(VFS
, "NFS: link(%s/%s -> %s/%s)\n",
1433 old_dentry
->d_parent
->d_name
.name
, old_dentry
->d_name
.name
,
1434 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
);
1437 * Drop the dentry in advance to force a new lookup.
1438 * Since nfs_proc_link doesn't return a file handle,
1439 * we can't use the existing dentry.
1444 nfs_begin_data_update(dir
);
1445 nfs_begin_data_update(inode
);
1446 error
= NFS_PROTO(dir
)->link(inode
, dir
, &dentry
->d_name
);
1447 nfs_end_data_update(inode
);
1448 nfs_end_data_update(dir
);
1455 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a
1456 * different file handle for the same inode after a rename (e.g. when
1457 * moving to a different directory). A fail-safe method to do so would
1458 * be to look up old_dir/old_name, create a link to new_dir/new_name and
1459 * rename the old file using the sillyrename stuff. This way, the original
1460 * file in old_dir will go away when the last process iput()s the inode.
1464 * It actually works quite well. One needs to have the possibility for
1465 * at least one ".nfs..." file in each directory the file ever gets
1466 * moved or linked to which happens automagically with the new
1467 * implementation that only depends on the dcache stuff instead of
1468 * using the inode layer
1470 * Unfortunately, things are a little more complicated than indicated
1471 * above. For a cross-directory move, we want to make sure we can get
1472 * rid of the old inode after the operation. This means there must be
1473 * no pending writes (if it's a file), and the use count must be 1.
1474 * If these conditions are met, we can drop the dentries before doing
1477 static int nfs_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
1478 struct inode
*new_dir
, struct dentry
*new_dentry
)
1480 struct inode
*old_inode
= old_dentry
->d_inode
;
1481 struct inode
*new_inode
= new_dentry
->d_inode
;
1482 struct dentry
*dentry
= NULL
, *rehash
= NULL
;
1486 * To prevent any new references to the target during the rename,
1487 * we unhash the dentry and free the inode in advance.
1490 if (!d_unhashed(new_dentry
)) {
1492 rehash
= new_dentry
;
1495 dfprintk(VFS
, "NFS: rename(%s/%s -> %s/%s, ct=%d)\n",
1496 old_dentry
->d_parent
->d_name
.name
, old_dentry
->d_name
.name
,
1497 new_dentry
->d_parent
->d_name
.name
, new_dentry
->d_name
.name
,
1498 atomic_read(&new_dentry
->d_count
));
1501 * First check whether the target is busy ... we can't
1502 * safely do _any_ rename if the target is in use.
1504 * For files, make a copy of the dentry and then do a
1505 * silly-rename. If the silly-rename succeeds, the
1506 * copied dentry is hashed and becomes the new target.
1510 if (S_ISDIR(new_inode
->i_mode
))
1512 else if (atomic_read(&new_dentry
->d_count
) > 2) {
1514 /* copy the target dentry's name */
1515 dentry
= d_alloc(new_dentry
->d_parent
,
1516 &new_dentry
->d_name
);
1520 /* silly-rename the existing target ... */
1521 err
= nfs_sillyrename(new_dir
, new_dentry
);
1523 new_dentry
= rehash
= dentry
;
1525 /* instantiate the replacement target */
1526 d_instantiate(new_dentry
, NULL
);
1527 } else if (atomic_read(&new_dentry
->d_count
) > 1) {
1528 /* dentry still busy? */
1530 printk("nfs_rename: target %s/%s busy, d_count=%d\n",
1531 new_dentry
->d_parent
->d_name
.name
,
1532 new_dentry
->d_name
.name
,
1533 atomic_read(&new_dentry
->d_count
));
1541 * ... prune child dentries and writebacks if needed.
1543 if (atomic_read(&old_dentry
->d_count
) > 1) {
1544 nfs_wb_all(old_inode
);
1545 shrink_dcache_parent(old_dentry
);
1549 d_delete(new_dentry
);
1551 nfs_begin_data_update(old_dir
);
1552 nfs_begin_data_update(new_dir
);
1553 nfs_begin_data_update(old_inode
);
1554 error
= NFS_PROTO(old_dir
)->rename(old_dir
, &old_dentry
->d_name
,
1555 new_dir
, &new_dentry
->d_name
);
1556 nfs_end_data_update(old_inode
);
1557 nfs_end_data_update(new_dir
);
1558 nfs_end_data_update(old_dir
);
1563 if (!S_ISDIR(old_inode
->i_mode
))
1564 d_move(old_dentry
, new_dentry
);
1565 nfs_renew_times(new_dentry
);
1566 nfs_set_verifier(new_dentry
, nfs_save_change_attribute(new_dir
));
1569 /* new dentry created? */
1576 int nfs_access_get_cached(struct inode
*inode
, struct rpc_cred
*cred
, struct nfs_access_entry
*res
)
1578 struct nfs_access_entry
*cache
= &NFS_I(inode
)->cache_access
;
1580 if (cache
->cred
!= cred
1581 || time_after(jiffies
, cache
->jiffies
+ NFS_ATTRTIMEO(inode
))
1582 || (NFS_FLAGS(inode
) & NFS_INO_INVALID_ACCESS
))
1584 memcpy(res
, cache
, sizeof(*res
));
1588 void nfs_access_add_cache(struct inode
*inode
, struct nfs_access_entry
*set
)
1590 struct nfs_access_entry
*cache
= &NFS_I(inode
)->cache_access
;
1592 if (cache
->cred
!= set
->cred
) {
1594 put_rpccred(cache
->cred
);
1595 cache
->cred
= get_rpccred(set
->cred
);
1597 NFS_FLAGS(inode
) &= ~NFS_INO_INVALID_ACCESS
;
1598 cache
->jiffies
= set
->jiffies
;
1599 cache
->mask
= set
->mask
;
1602 static int nfs_do_access(struct inode
*inode
, struct rpc_cred
*cred
, int mask
)
1604 struct nfs_access_entry cache
;
1607 status
= nfs_access_get_cached(inode
, cred
, &cache
);
1611 /* Be clever: ask server to check for all possible rights */
1612 cache
.mask
= MAY_EXEC
| MAY_WRITE
| MAY_READ
;
1614 cache
.jiffies
= jiffies
;
1615 status
= NFS_PROTO(inode
)->access(inode
, &cache
);
1618 nfs_access_add_cache(inode
, &cache
);
1620 if ((cache
.mask
& mask
) == mask
)
1625 int nfs_permission(struct inode
*inode
, int mask
, struct nameidata
*nd
)
1627 struct rpc_cred
*cred
;
1632 /* Is this sys_access() ? */
1633 if (nd
!= NULL
&& (nd
->flags
& LOOKUP_ACCESS
))
1636 switch (inode
->i_mode
& S_IFMT
) {
1640 /* NFSv4 has atomic_open... */
1641 if (nfs_server_capable(inode
, NFS_CAP_ATOMIC_OPEN
)
1643 && (nd
->flags
& LOOKUP_OPEN
))
1648 * Optimize away all write operations, since the server
1649 * will check permissions when we perform the op.
1651 if ((mask
& MAY_WRITE
) && !(mask
& MAY_READ
))
1658 if (!NFS_PROTO(inode
)->access
)
1661 cred
= rpcauth_lookupcred(NFS_CLIENT(inode
)->cl_auth
, 0);
1662 if (!IS_ERR(cred
)) {
1663 res
= nfs_do_access(inode
, cred
, mask
);
1666 res
= PTR_ERR(cred
);
1671 res
= nfs_revalidate_inode(NFS_SERVER(inode
), inode
);
1673 res
= generic_permission(inode
, mask
, NULL
);
1680 * version-control: t
1681 * kept-new-versions: 5