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/pagevec.h>
33 #include <linux/namei.h>
34 #include <linux/mount.h>
35 #include <linux/sched.h>
37 #include "delegation.h"
42 /* #define NFS_DEBUG_VERBOSE 1 */
44 static int nfs_opendir(struct inode
*, struct file
*);
45 static int nfs_readdir(struct file
*, void *, filldir_t
);
46 static struct dentry
*nfs_lookup(struct inode
*, struct dentry
*, struct nameidata
*);
47 static int nfs_create(struct inode
*, struct dentry
*, int, struct nameidata
*);
48 static int nfs_mkdir(struct inode
*, struct dentry
*, int);
49 static int nfs_rmdir(struct inode
*, struct dentry
*);
50 static int nfs_unlink(struct inode
*, struct dentry
*);
51 static int nfs_symlink(struct inode
*, struct dentry
*, const char *);
52 static int nfs_link(struct dentry
*, struct inode
*, struct dentry
*);
53 static int nfs_mknod(struct inode
*, struct dentry
*, int, dev_t
);
54 static int nfs_rename(struct inode
*, struct dentry
*,
55 struct inode
*, struct dentry
*);
56 static int nfs_fsync_dir(struct file
*, int);
57 static loff_t
nfs_llseek_dir(struct file
*, loff_t
, int);
58 static int nfs_readdir_clear_array(struct page
*, gfp_t
);
60 const struct file_operations nfs_dir_operations
= {
61 .llseek
= nfs_llseek_dir
,
62 .read
= generic_read_dir
,
63 .readdir
= nfs_readdir
,
65 .release
= nfs_release
,
66 .fsync
= nfs_fsync_dir
,
69 const struct inode_operations nfs_dir_inode_operations
= {
74 .symlink
= nfs_symlink
,
79 .permission
= nfs_permission
,
80 .getattr
= nfs_getattr
,
81 .setattr
= nfs_setattr
,
84 const struct address_space_operations nfs_dir_addr_space_ops
= {
85 .releasepage
= nfs_readdir_clear_array
,
89 const struct inode_operations nfs3_dir_inode_operations
= {
94 .symlink
= nfs_symlink
,
99 .permission
= nfs_permission
,
100 .getattr
= nfs_getattr
,
101 .setattr
= nfs_setattr
,
102 .listxattr
= nfs3_listxattr
,
103 .getxattr
= nfs3_getxattr
,
104 .setxattr
= nfs3_setxattr
,
105 .removexattr
= nfs3_removexattr
,
107 #endif /* CONFIG_NFS_V3 */
111 static struct dentry
*nfs_atomic_lookup(struct inode
*, struct dentry
*, struct nameidata
*);
112 static int nfs_open_create(struct inode
*dir
, struct dentry
*dentry
, int mode
, struct nameidata
*nd
);
113 const struct inode_operations nfs4_dir_inode_operations
= {
114 .create
= nfs_open_create
,
115 .lookup
= nfs_atomic_lookup
,
117 .unlink
= nfs_unlink
,
118 .symlink
= nfs_symlink
,
122 .rename
= nfs_rename
,
123 .permission
= nfs_permission
,
124 .getattr
= nfs_getattr
,
125 .setattr
= nfs_setattr
,
126 .getxattr
= nfs4_getxattr
,
127 .setxattr
= nfs4_setxattr
,
128 .listxattr
= nfs4_listxattr
,
131 #endif /* CONFIG_NFS_V4 */
137 nfs_opendir(struct inode
*inode
, struct file
*filp
)
141 dfprintk(FILE, "NFS: open dir(%s/%s)\n",
142 filp
->f_path
.dentry
->d_parent
->d_name
.name
,
143 filp
->f_path
.dentry
->d_name
.name
);
145 nfs_inc_stats(inode
, NFSIOS_VFSOPEN
);
147 /* Call generic open code in order to cache credentials */
148 res
= nfs_open(inode
, filp
);
149 if (filp
->f_path
.dentry
== filp
->f_path
.mnt
->mnt_root
) {
150 /* This is a mountpoint, so d_revalidate will never
151 * have been called, so we need to refresh the
152 * inode (for close-open consistency) ourselves.
154 __nfs_revalidate_inode(NFS_SERVER(inode
), inode
);
159 struct nfs_cache_array_entry
{
165 struct nfs_cache_array
{
169 struct nfs_cache_array_entry array
[0];
172 #define MAX_READDIR_ARRAY ((PAGE_SIZE - sizeof(struct nfs_cache_array)) / sizeof(struct nfs_cache_array_entry))
174 typedef __be32
* (*decode_dirent_t
)(__be32
*, struct nfs_entry
*, int);
178 unsigned long page_index
;
180 loff_t current_index
;
181 decode_dirent_t decode
;
183 unsigned long timestamp
;
184 unsigned long gencount
;
185 unsigned int cache_entry_index
;
188 } nfs_readdir_descriptor_t
;
191 * The caller is responsible for calling nfs_readdir_release_array(page)
194 struct nfs_cache_array
*nfs_readdir_get_array(struct page
*page
)
197 return ERR_PTR(-EIO
);
198 return (struct nfs_cache_array
*)kmap(page
);
202 void nfs_readdir_release_array(struct page
*page
)
208 * we are freeing strings created by nfs_add_to_readdir_array()
211 int nfs_readdir_clear_array(struct page
*page
, gfp_t mask
)
213 struct nfs_cache_array
*array
= nfs_readdir_get_array(page
);
215 for (i
= 0; i
< array
->size
; i
++)
216 kfree(array
->array
[i
].string
.name
);
217 nfs_readdir_release_array(page
);
222 * the caller is responsible for freeing qstr.name
223 * when called by nfs_readdir_add_to_array, the strings will be freed in
224 * nfs_clear_readdir_array()
227 void nfs_readdir_make_qstr(struct qstr
*string
, const char *name
, unsigned int len
)
230 string
->name
= kmemdup(name
, len
, GFP_KERNEL
);
231 string
->hash
= full_name_hash(string
->name
, string
->len
);
235 int nfs_readdir_add_to_array(struct nfs_entry
*entry
, struct page
*page
)
237 struct nfs_cache_array
*array
= nfs_readdir_get_array(page
);
239 return PTR_ERR(array
);
240 if (array
->size
>= MAX_READDIR_ARRAY
) {
241 nfs_readdir_release_array(page
);
245 array
->array
[array
->size
].cookie
= entry
->prev_cookie
;
246 array
->last_cookie
= entry
->cookie
;
247 array
->array
[array
->size
].ino
= entry
->ino
;
248 nfs_readdir_make_qstr(&array
->array
[array
->size
].string
, entry
->name
, entry
->len
);
250 array
->eof_index
= array
->size
;
252 nfs_readdir_release_array(page
);
257 int nfs_readdir_search_for_pos(struct nfs_cache_array
*array
, nfs_readdir_descriptor_t
*desc
)
259 loff_t diff
= desc
->file
->f_pos
- desc
->current_index
;
264 if (diff
>= array
->size
) {
265 if (array
->eof_index
> 0)
267 desc
->current_index
+= array
->size
;
271 index
= (unsigned int)diff
;
272 *desc
->dir_cookie
= array
->array
[index
].cookie
;
273 desc
->cache_entry_index
= index
;
274 if (index
== array
->eof_index
)
283 int nfs_readdir_search_for_cookie(struct nfs_cache_array
*array
, nfs_readdir_descriptor_t
*desc
)
286 int status
= -EAGAIN
;
288 for (i
= 0; i
< array
->size
; i
++) {
289 if (i
== array
->eof_index
) {
291 status
= -EBADCOOKIE
;
293 if (array
->array
[i
].cookie
== *desc
->dir_cookie
) {
294 desc
->cache_entry_index
= i
;
304 int nfs_readdir_search_array(nfs_readdir_descriptor_t
*desc
)
306 struct nfs_cache_array
*array
;
307 int status
= -EBADCOOKIE
;
309 if (desc
->dir_cookie
== NULL
)
312 array
= nfs_readdir_get_array(desc
->page
);
314 status
= PTR_ERR(array
);
318 if (*desc
->dir_cookie
== 0)
319 status
= nfs_readdir_search_for_pos(array
, desc
);
321 status
= nfs_readdir_search_for_cookie(array
, desc
);
323 nfs_readdir_release_array(desc
->page
);
328 /* Fill a page with xdr information before transferring to the cache page */
330 int nfs_readdir_xdr_filler(struct page
*xdr_page
, nfs_readdir_descriptor_t
*desc
,
331 struct nfs_entry
*entry
, struct file
*file
, struct inode
*inode
)
333 struct rpc_cred
*cred
= nfs_file_cred(file
);
334 unsigned long timestamp
, gencount
;
339 gencount
= nfs_inc_attr_generation_counter();
340 error
= NFS_PROTO(inode
)->readdir(file
->f_path
.dentry
, cred
, entry
->cookie
, xdr_page
,
341 NFS_SERVER(inode
)->dtsize
, desc
->plus
);
343 /* We requested READDIRPLUS, but the server doesn't grok it */
344 if (error
== -ENOTSUPP
&& desc
->plus
) {
345 NFS_SERVER(inode
)->caps
&= ~NFS_CAP_READDIRPLUS
;
346 clear_bit(NFS_INO_ADVISE_RDPLUS
, &NFS_I(inode
)->flags
);
352 desc
->timestamp
= timestamp
;
353 desc
->gencount
= gencount
;
358 /* Fill in an entry based on the xdr code stored in desc->page */
360 int xdr_decode(nfs_readdir_descriptor_t
*desc
, struct nfs_entry
*entry
, __be32
**ptr
)
363 p
= desc
->decode(p
, entry
, desc
->plus
);
368 entry
->fattr
->time_start
= desc
->timestamp
;
369 entry
->fattr
->gencount
= desc
->gencount
;
373 /* Perform conversion from xdr to cache array */
375 void nfs_readdir_page_filler(nfs_readdir_descriptor_t
*desc
, struct nfs_entry
*entry
,
376 struct page
*xdr_page
, struct page
*page
)
378 __be32
*ptr
= kmap(xdr_page
);
379 while (xdr_decode(desc
, entry
, &ptr
) == 0) {
380 if (nfs_readdir_add_to_array(entry
, page
) == -1)
387 int nfs_readdir_xdr_to_array(nfs_readdir_descriptor_t
*desc
, struct page
*page
, struct inode
*inode
)
389 struct page
*xdr_page
;
390 struct nfs_entry entry
;
391 struct file
*file
= desc
->file
;
392 struct nfs_cache_array
*array
;
395 entry
.prev_cookie
= 0;
396 entry
.cookie
= *desc
->dir_cookie
;
398 entry
.fh
= nfs_alloc_fhandle();
399 entry
.fattr
= nfs_alloc_fattr();
400 if (entry
.fh
== NULL
|| entry
.fattr
== NULL
)
403 array
= nfs_readdir_get_array(page
);
404 memset(array
, 0, sizeof(struct nfs_cache_array
));
405 array
->eof_index
= -1;
407 xdr_page
= alloc_page(GFP_KERNEL
);
409 goto out_release_array
;
411 status
= nfs_readdir_xdr_filler(xdr_page
, desc
, &entry
, file
, inode
);
414 nfs_readdir_page_filler(desc
, &entry
, xdr_page
, page
);
415 } while (array
->eof_index
< 0 && array
->size
< MAX_READDIR_ARRAY
);
419 nfs_readdir_release_array(page
);
421 nfs_free_fattr(entry
.fattr
);
422 nfs_free_fhandle(entry
.fh
);
427 * Now we cache directories properly, by converting xdr information
428 * to an array that can be used for lookups later. This results in
429 * fewer cache pages, since we can store more information on each page.
430 * We only need to convert from xdr once so future lookups are much simpler
433 int nfs_readdir_filler(nfs_readdir_descriptor_t
*desc
, struct page
* page
)
435 struct inode
*inode
= desc
->file
->f_path
.dentry
->d_inode
;
437 if (nfs_readdir_xdr_to_array(desc
, page
, inode
) == -1)
439 SetPageUptodate(page
);
441 if (invalidate_inode_pages2_range(inode
->i_mapping
, page
->index
+ 1, -1) < 0) {
442 /* Should never happen */
443 nfs_zap_mapping(inode
, inode
->i_mapping
);
453 void cache_page_release(nfs_readdir_descriptor_t
*desc
)
455 page_cache_release(desc
->page
);
460 struct page
*get_cache_page(nfs_readdir_descriptor_t
*desc
)
463 page
= read_cache_page(desc
->file
->f_path
.dentry
->d_inode
->i_mapping
,
464 desc
->page_index
, (filler_t
*)nfs_readdir_filler
, desc
);
471 * Returns 0 if desc->dir_cookie was found on page desc->page_index
474 int find_cache_page(nfs_readdir_descriptor_t
*desc
)
478 desc
->page
= get_cache_page(desc
);
479 if (IS_ERR(desc
->page
))
480 return PTR_ERR(desc
->page
);
482 res
= nfs_readdir_search_array(desc
);
485 cache_page_release(desc
);
489 /* Search for desc->dir_cookie from the beginning of the page cache */
491 int readdir_search_pagecache(nfs_readdir_descriptor_t
*desc
)
494 desc
->page_index
= 0;
496 if (*desc
->dir_cookie
== 0)
497 desc
->cache_entry_index
= 0;
500 res
= find_cache_page(desc
);
508 static inline unsigned int dt_type(struct inode
*inode
)
510 return (inode
->i_mode
>> 12) & 15;
514 * Once we've found the start of the dirent within a page: fill 'er up...
517 int nfs_do_filldir(nfs_readdir_descriptor_t
*desc
, void *dirent
,
520 struct file
*file
= desc
->file
;
523 struct nfs_cache_array
*array
= NULL
;
524 unsigned int d_type
= DT_UNKNOWN
;
525 struct dentry
*dentry
= NULL
;
527 array
= nfs_readdir_get_array(desc
->page
);
529 for (i
= desc
->cache_entry_index
; i
< array
->size
; i
++) {
532 res
= filldir(dirent
, array
->array
[i
].string
.name
,
533 array
->array
[i
].string
.len
, file
->f_pos
,
534 nfs_compat_user_ino64(array
->array
[i
].ino
), d_type
);
538 desc
->cache_entry_index
= i
;
539 if (i
< (array
->size
-1))
540 *desc
->dir_cookie
= array
->array
[i
+1].cookie
;
542 *desc
->dir_cookie
= array
->last_cookie
;
543 if (i
== array
->eof_index
) {
549 nfs_readdir_release_array(desc
->page
);
550 cache_page_release(desc
);
553 dfprintk(DIRCACHE
, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n",
554 (unsigned long long)*desc
->dir_cookie
, res
);
559 * If we cannot find a cookie in our cache, we suspect that this is
560 * because it points to a deleted file, so we ask the server to return
561 * whatever it thinks is the next entry. We then feed this to filldir.
562 * If all goes well, we should then be able to find our way round the
563 * cache on the next call to readdir_search_pagecache();
565 * NOTE: we cannot add the anonymous page to the pagecache because
566 * the data it contains might not be page aligned. Besides,
567 * we should already have a complete representation of the
568 * directory in the page cache by the time we get here.
571 int uncached_readdir(nfs_readdir_descriptor_t
*desc
, void *dirent
,
574 struct page
*page
= NULL
;
576 struct inode
*inode
= desc
->file
->f_path
.dentry
->d_inode
;
578 dfprintk(DIRCACHE
, "NFS: uncached_readdir() searching for cookie %Lu\n",
579 (unsigned long long)*desc
->dir_cookie
);
581 page
= alloc_page(GFP_HIGHUSER
);
587 if (nfs_readdir_xdr_to_array(desc
, page
, inode
) == -1) {
593 status
= nfs_do_filldir(desc
, dirent
, filldir
);
596 dfprintk(DIRCACHE
, "NFS: %s: returns %d\n",
600 cache_page_release(desc
);
604 /* The file offset position represents the dirent entry number. A
605 last cookie cache takes care of the common case of reading the
608 static int nfs_readdir(struct file
*filp
, void *dirent
, filldir_t filldir
)
610 struct dentry
*dentry
= filp
->f_path
.dentry
;
611 struct inode
*inode
= dentry
->d_inode
;
612 nfs_readdir_descriptor_t my_desc
,
616 dfprintk(FILE, "NFS: readdir(%s/%s) starting at cookie %llu\n",
617 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
,
618 (long long)filp
->f_pos
);
619 nfs_inc_stats(inode
, NFSIOS_VFSGETDENTS
);
622 * filp->f_pos points to the dirent entry number.
623 * *desc->dir_cookie has the cookie for the next entry. We have
624 * to either find the entry with the appropriate number or
625 * revalidate the cookie.
627 memset(desc
, 0, sizeof(*desc
));
630 desc
->dir_cookie
= &nfs_file_open_context(filp
)->dir_cookie
;
631 desc
->decode
= NFS_PROTO(inode
)->decode_dirent
;
632 desc
->plus
= NFS_USE_READDIRPLUS(inode
);
634 nfs_block_sillyrename(dentry
);
635 res
= nfs_revalidate_mapping(inode
, filp
->f_mapping
);
639 while (desc
->eof
!= 1) {
640 res
= readdir_search_pagecache(desc
);
642 if (res
== -EBADCOOKIE
) {
643 /* This means either end of directory */
644 if (*desc
->dir_cookie
&& desc
->eof
== 0) {
645 /* Or that the server has 'lost' a cookie */
646 res
= uncached_readdir(desc
, dirent
, filldir
);
653 if (res
== -ETOOSMALL
&& desc
->plus
) {
654 clear_bit(NFS_INO_ADVISE_RDPLUS
, &NFS_I(inode
)->flags
);
655 nfs_zap_caches(inode
);
663 res
= nfs_do_filldir(desc
, dirent
, filldir
);
670 nfs_unblock_sillyrename(dentry
);
673 dfprintk(FILE, "NFS: readdir(%s/%s) returns %d\n",
674 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
,
679 static loff_t
nfs_llseek_dir(struct file
*filp
, loff_t offset
, int origin
)
681 struct dentry
*dentry
= filp
->f_path
.dentry
;
682 struct inode
*inode
= dentry
->d_inode
;
684 dfprintk(FILE, "NFS: llseek dir(%s/%s, %lld, %d)\n",
685 dentry
->d_parent
->d_name
.name
,
689 mutex_lock(&inode
->i_mutex
);
692 offset
+= filp
->f_pos
;
700 if (offset
!= filp
->f_pos
) {
701 filp
->f_pos
= offset
;
702 nfs_file_open_context(filp
)->dir_cookie
= 0;
705 mutex_unlock(&inode
->i_mutex
);
710 * All directory operations under NFS are synchronous, so fsync()
711 * is a dummy operation.
713 static int nfs_fsync_dir(struct file
*filp
, int datasync
)
715 struct dentry
*dentry
= filp
->f_path
.dentry
;
717 dfprintk(FILE, "NFS: fsync dir(%s/%s) datasync %d\n",
718 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
,
721 nfs_inc_stats(dentry
->d_inode
, NFSIOS_VFSFSYNC
);
726 * nfs_force_lookup_revalidate - Mark the directory as having changed
727 * @dir - pointer to directory inode
729 * This forces the revalidation code in nfs_lookup_revalidate() to do a
730 * full lookup on all child dentries of 'dir' whenever a change occurs
731 * on the server that might have invalidated our dcache.
733 * The caller should be holding dir->i_lock
735 void nfs_force_lookup_revalidate(struct inode
*dir
)
737 NFS_I(dir
)->cache_change_attribute
++;
741 * A check for whether or not the parent directory has changed.
742 * In the case it has, we assume that the dentries are untrustworthy
743 * and may need to be looked up again.
745 static int nfs_check_verifier(struct inode
*dir
, struct dentry
*dentry
)
749 if (NFS_SERVER(dir
)->flags
& NFS_MOUNT_LOOKUP_CACHE_NONE
)
751 if (!nfs_verify_change_attribute(dir
, dentry
->d_time
))
753 /* Revalidate nfsi->cache_change_attribute before we declare a match */
754 if (nfs_revalidate_inode(NFS_SERVER(dir
), dir
) < 0)
756 if (!nfs_verify_change_attribute(dir
, dentry
->d_time
))
762 * Return the intent data that applies to this particular path component
764 * Note that the current set of intents only apply to the very last
765 * component of the path.
766 * We check for this using LOOKUP_CONTINUE and LOOKUP_PARENT.
768 static inline unsigned int nfs_lookup_check_intent(struct nameidata
*nd
, unsigned int mask
)
770 if (nd
->flags
& (LOOKUP_CONTINUE
|LOOKUP_PARENT
))
772 return nd
->flags
& mask
;
776 * Use intent information to check whether or not we're going to do
777 * an O_EXCL create using this path component.
779 static int nfs_is_exclusive_create(struct inode
*dir
, struct nameidata
*nd
)
781 if (NFS_PROTO(dir
)->version
== 2)
783 return nd
&& nfs_lookup_check_intent(nd
, LOOKUP_EXCL
);
787 * Inode and filehandle revalidation for lookups.
789 * We force revalidation in the cases where the VFS sets LOOKUP_REVAL,
790 * or if the intent information indicates that we're about to open this
791 * particular file and the "nocto" mount flag is not set.
795 int nfs_lookup_verify_inode(struct inode
*inode
, struct nameidata
*nd
)
797 struct nfs_server
*server
= NFS_SERVER(inode
);
799 if (test_bit(NFS_INO_MOUNTPOINT
, &NFS_I(inode
)->flags
))
802 /* VFS wants an on-the-wire revalidation */
803 if (nd
->flags
& LOOKUP_REVAL
)
805 /* This is an open(2) */
806 if (nfs_lookup_check_intent(nd
, LOOKUP_OPEN
) != 0 &&
807 !(server
->flags
& NFS_MOUNT_NOCTO
) &&
808 (S_ISREG(inode
->i_mode
) ||
809 S_ISDIR(inode
->i_mode
)))
813 return nfs_revalidate_inode(server
, inode
);
815 return __nfs_revalidate_inode(server
, inode
);
819 * We judge how long we want to trust negative
820 * dentries by looking at the parent inode mtime.
822 * If parent mtime has changed, we revalidate, else we wait for a
823 * period corresponding to the parent's attribute cache timeout value.
826 int nfs_neg_need_reval(struct inode
*dir
, struct dentry
*dentry
,
827 struct nameidata
*nd
)
829 /* Don't revalidate a negative dentry if we're creating a new file */
830 if (nd
!= NULL
&& nfs_lookup_check_intent(nd
, LOOKUP_CREATE
) != 0)
832 if (NFS_SERVER(dir
)->flags
& NFS_MOUNT_LOOKUP_CACHE_NONEG
)
834 return !nfs_check_verifier(dir
, dentry
);
838 * This is called every time the dcache has a lookup hit,
839 * and we should check whether we can really trust that
842 * NOTE! The hit can be a negative hit too, don't assume
845 * If the parent directory is seen to have changed, we throw out the
846 * cached dentry and do a new lookup.
848 static int nfs_lookup_revalidate(struct dentry
* dentry
, struct nameidata
*nd
)
852 struct dentry
*parent
;
853 struct nfs_fh
*fhandle
= NULL
;
854 struct nfs_fattr
*fattr
= NULL
;
857 parent
= dget_parent(dentry
);
858 dir
= parent
->d_inode
;
859 nfs_inc_stats(dir
, NFSIOS_DENTRYREVALIDATE
);
860 inode
= dentry
->d_inode
;
863 if (nfs_neg_need_reval(dir
, dentry
, nd
))
868 if (is_bad_inode(inode
)) {
869 dfprintk(LOOKUPCACHE
, "%s: %s/%s has dud inode\n",
870 __func__
, dentry
->d_parent
->d_name
.name
,
871 dentry
->d_name
.name
);
875 if (nfs_have_delegation(inode
, FMODE_READ
))
876 goto out_set_verifier
;
878 /* Force a full look up iff the parent directory has changed */
879 if (!nfs_is_exclusive_create(dir
, nd
) && nfs_check_verifier(dir
, dentry
)) {
880 if (nfs_lookup_verify_inode(inode
, nd
))
885 if (NFS_STALE(inode
))
889 fhandle
= nfs_alloc_fhandle();
890 fattr
= nfs_alloc_fattr();
891 if (fhandle
== NULL
|| fattr
== NULL
)
894 error
= NFS_PROTO(dir
)->lookup(dir
, &dentry
->d_name
, fhandle
, fattr
);
897 if (nfs_compare_fh(NFS_FH(inode
), fhandle
))
899 if ((error
= nfs_refresh_inode(inode
, fattr
)) != 0)
902 nfs_free_fattr(fattr
);
903 nfs_free_fhandle(fhandle
);
905 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
908 dfprintk(LOOKUPCACHE
, "NFS: %s(%s/%s) is valid\n",
909 __func__
, dentry
->d_parent
->d_name
.name
,
910 dentry
->d_name
.name
);
915 nfs_mark_for_revalidate(dir
);
916 if (inode
&& S_ISDIR(inode
->i_mode
)) {
917 /* Purge readdir caches. */
918 nfs_zap_caches(inode
);
919 /* If we have submounts, don't unhash ! */
920 if (have_submounts(dentry
))
922 if (dentry
->d_flags
& DCACHE_DISCONNECTED
)
924 shrink_dcache_parent(dentry
);
927 nfs_free_fattr(fattr
);
928 nfs_free_fhandle(fhandle
);
930 dfprintk(LOOKUPCACHE
, "NFS: %s(%s/%s) is invalid\n",
931 __func__
, dentry
->d_parent
->d_name
.name
,
932 dentry
->d_name
.name
);
935 nfs_free_fattr(fattr
);
936 nfs_free_fhandle(fhandle
);
938 dfprintk(LOOKUPCACHE
, "NFS: %s(%s/%s) lookup returned error %d\n",
939 __func__
, dentry
->d_parent
->d_name
.name
,
940 dentry
->d_name
.name
, error
);
945 * This is called from dput() when d_count is going to 0.
947 static int nfs_dentry_delete(struct dentry
*dentry
)
949 dfprintk(VFS
, "NFS: dentry_delete(%s/%s, %x)\n",
950 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
,
953 /* Unhash any dentry with a stale inode */
954 if (dentry
->d_inode
!= NULL
&& NFS_STALE(dentry
->d_inode
))
957 if (dentry
->d_flags
& DCACHE_NFSFS_RENAMED
) {
958 /* Unhash it, so that ->d_iput() would be called */
961 if (!(dentry
->d_sb
->s_flags
& MS_ACTIVE
)) {
962 /* Unhash it, so that ancestors of killed async unlink
963 * files will be cleaned up during umount */
970 static void nfs_drop_nlink(struct inode
*inode
)
972 spin_lock(&inode
->i_lock
);
973 if (inode
->i_nlink
> 0)
975 spin_unlock(&inode
->i_lock
);
979 * Called when the dentry loses inode.
980 * We use it to clean up silly-renamed files.
982 static void nfs_dentry_iput(struct dentry
*dentry
, struct inode
*inode
)
984 if (S_ISDIR(inode
->i_mode
))
985 /* drop any readdir cache as it could easily be old */
986 NFS_I(inode
)->cache_validity
|= NFS_INO_INVALID_DATA
;
988 if (dentry
->d_flags
& DCACHE_NFSFS_RENAMED
) {
990 nfs_complete_unlink(dentry
, inode
);
995 const struct dentry_operations nfs_dentry_operations
= {
996 .d_revalidate
= nfs_lookup_revalidate
,
997 .d_delete
= nfs_dentry_delete
,
998 .d_iput
= nfs_dentry_iput
,
1001 static struct dentry
*nfs_lookup(struct inode
*dir
, struct dentry
* dentry
, struct nameidata
*nd
)
1004 struct dentry
*parent
;
1005 struct inode
*inode
= NULL
;
1006 struct nfs_fh
*fhandle
= NULL
;
1007 struct nfs_fattr
*fattr
= NULL
;
1010 dfprintk(VFS
, "NFS: lookup(%s/%s)\n",
1011 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
);
1012 nfs_inc_stats(dir
, NFSIOS_VFSLOOKUP
);
1014 res
= ERR_PTR(-ENAMETOOLONG
);
1015 if (dentry
->d_name
.len
> NFS_SERVER(dir
)->namelen
)
1018 dentry
->d_op
= NFS_PROTO(dir
)->dentry_ops
;
1021 * If we're doing an exclusive create, optimize away the lookup
1022 * but don't hash the dentry.
1024 if (nfs_is_exclusive_create(dir
, nd
)) {
1025 d_instantiate(dentry
, NULL
);
1030 res
= ERR_PTR(-ENOMEM
);
1031 fhandle
= nfs_alloc_fhandle();
1032 fattr
= nfs_alloc_fattr();
1033 if (fhandle
== NULL
|| fattr
== NULL
)
1036 parent
= dentry
->d_parent
;
1037 /* Protect against concurrent sillydeletes */
1038 nfs_block_sillyrename(parent
);
1039 error
= NFS_PROTO(dir
)->lookup(dir
, &dentry
->d_name
, fhandle
, fattr
);
1040 if (error
== -ENOENT
)
1043 res
= ERR_PTR(error
);
1044 goto out_unblock_sillyrename
;
1046 inode
= nfs_fhget(dentry
->d_sb
, fhandle
, fattr
);
1047 res
= (struct dentry
*)inode
;
1049 goto out_unblock_sillyrename
;
1052 res
= d_materialise_unique(dentry
, inode
);
1055 goto out_unblock_sillyrename
;
1058 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1059 out_unblock_sillyrename
:
1060 nfs_unblock_sillyrename(parent
);
1062 nfs_free_fattr(fattr
);
1063 nfs_free_fhandle(fhandle
);
1067 #ifdef CONFIG_NFS_V4
1068 static int nfs_open_revalidate(struct dentry
*, struct nameidata
*);
1070 const struct dentry_operations nfs4_dentry_operations
= {
1071 .d_revalidate
= nfs_open_revalidate
,
1072 .d_delete
= nfs_dentry_delete
,
1073 .d_iput
= nfs_dentry_iput
,
1077 * Use intent information to determine whether we need to substitute
1078 * the NFSv4-style stateful OPEN for the LOOKUP call
1080 static int is_atomic_open(struct nameidata
*nd
)
1082 if (nd
== NULL
|| nfs_lookup_check_intent(nd
, LOOKUP_OPEN
) == 0)
1084 /* NFS does not (yet) have a stateful open for directories */
1085 if (nd
->flags
& LOOKUP_DIRECTORY
)
1087 /* Are we trying to write to a read only partition? */
1088 if (__mnt_is_readonly(nd
->path
.mnt
) &&
1089 (nd
->intent
.open
.flags
& (O_CREAT
|O_TRUNC
|FMODE_WRITE
)))
1094 static struct nfs_open_context
*nameidata_to_nfs_open_context(struct dentry
*dentry
, struct nameidata
*nd
)
1096 struct path path
= {
1097 .mnt
= nd
->path
.mnt
,
1100 struct nfs_open_context
*ctx
;
1101 struct rpc_cred
*cred
;
1102 fmode_t fmode
= nd
->intent
.open
.flags
& (FMODE_READ
| FMODE_WRITE
| FMODE_EXEC
);
1104 cred
= rpc_lookup_cred();
1106 return ERR_CAST(cred
);
1107 ctx
= alloc_nfs_open_context(&path
, cred
, fmode
);
1110 return ERR_PTR(-ENOMEM
);
1114 static int do_open(struct inode
*inode
, struct file
*filp
)
1116 nfs_fscache_set_inode_cookie(inode
, filp
);
1120 static int nfs_intent_set_file(struct nameidata
*nd
, struct nfs_open_context
*ctx
)
1125 /* If the open_intent is for execute, we have an extra check to make */
1126 if (ctx
->mode
& FMODE_EXEC
) {
1127 ret
= nfs_may_open(ctx
->path
.dentry
->d_inode
,
1129 nd
->intent
.open
.flags
);
1133 filp
= lookup_instantiate_filp(nd
, ctx
->path
.dentry
, do_open
);
1135 ret
= PTR_ERR(filp
);
1137 nfs_file_set_open_context(filp
, ctx
);
1139 put_nfs_open_context(ctx
);
1143 static struct dentry
*nfs_atomic_lookup(struct inode
*dir
, struct dentry
*dentry
, struct nameidata
*nd
)
1145 struct nfs_open_context
*ctx
;
1147 struct dentry
*res
= NULL
;
1148 struct inode
*inode
;
1152 dfprintk(VFS
, "NFS: atomic_lookup(%s/%ld), %s\n",
1153 dir
->i_sb
->s_id
, dir
->i_ino
, dentry
->d_name
.name
);
1155 /* Check that we are indeed trying to open this file */
1156 if (!is_atomic_open(nd
))
1159 if (dentry
->d_name
.len
> NFS_SERVER(dir
)->namelen
) {
1160 res
= ERR_PTR(-ENAMETOOLONG
);
1163 dentry
->d_op
= NFS_PROTO(dir
)->dentry_ops
;
1165 /* Let vfs_create() deal with O_EXCL. Instantiate, but don't hash
1167 if (nd
->flags
& LOOKUP_EXCL
) {
1168 d_instantiate(dentry
, NULL
);
1172 ctx
= nameidata_to_nfs_open_context(dentry
, nd
);
1173 res
= ERR_CAST(ctx
);
1177 open_flags
= nd
->intent
.open
.flags
;
1178 if (nd
->flags
& LOOKUP_CREATE
) {
1179 attr
.ia_mode
= nd
->intent
.open
.create_mode
;
1180 attr
.ia_valid
= ATTR_MODE
;
1181 if (!IS_POSIXACL(dir
))
1182 attr
.ia_mode
&= ~current_umask();
1184 open_flags
&= ~(O_EXCL
| O_CREAT
);
1188 /* Open the file on the server */
1189 nfs_block_sillyrename(dentry
->d_parent
);
1190 inode
= NFS_PROTO(dir
)->open_context(dir
, ctx
, open_flags
, &attr
);
1191 if (IS_ERR(inode
)) {
1192 nfs_unblock_sillyrename(dentry
->d_parent
);
1193 put_nfs_open_context(ctx
);
1194 switch (PTR_ERR(inode
)) {
1195 /* Make a negative dentry */
1197 d_add(dentry
, NULL
);
1200 /* This turned out not to be a regular file */
1205 if (!(nd
->intent
.open
.flags
& O_NOFOLLOW
))
1209 res
= ERR_CAST(inode
);
1213 res
= d_add_unique(dentry
, inode
);
1214 nfs_unblock_sillyrename(dentry
->d_parent
);
1216 dput(ctx
->path
.dentry
);
1217 ctx
->path
.dentry
= dget(res
);
1220 err
= nfs_intent_set_file(nd
, ctx
);
1224 return ERR_PTR(err
);
1227 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1230 return nfs_lookup(dir
, dentry
, nd
);
1233 static int nfs_open_revalidate(struct dentry
*dentry
, struct nameidata
*nd
)
1235 struct dentry
*parent
= NULL
;
1236 struct inode
*inode
= dentry
->d_inode
;
1238 struct nfs_open_context
*ctx
;
1239 int openflags
, ret
= 0;
1241 if (!is_atomic_open(nd
) || d_mountpoint(dentry
))
1244 parent
= dget_parent(dentry
);
1245 dir
= parent
->d_inode
;
1247 /* We can't create new files in nfs_open_revalidate(), so we
1248 * optimize away revalidation of negative dentries.
1250 if (inode
== NULL
) {
1251 if (!nfs_neg_need_reval(dir
, dentry
, nd
))
1256 /* NFS only supports OPEN on regular files */
1257 if (!S_ISREG(inode
->i_mode
))
1259 openflags
= nd
->intent
.open
.flags
;
1260 /* We cannot do exclusive creation on a positive dentry */
1261 if ((openflags
& (O_CREAT
|O_EXCL
)) == (O_CREAT
|O_EXCL
))
1263 /* We can't create new files, or truncate existing ones here */
1264 openflags
&= ~(O_CREAT
|O_EXCL
|O_TRUNC
);
1266 ctx
= nameidata_to_nfs_open_context(dentry
, nd
);
1271 * Note: we're not holding inode->i_mutex and so may be racing with
1272 * operations that change the directory. We therefore save the
1273 * change attribute *before* we do the RPC call.
1275 inode
= NFS_PROTO(dir
)->open_context(dir
, ctx
, openflags
, NULL
);
1276 if (IS_ERR(inode
)) {
1277 ret
= PTR_ERR(inode
);
1290 if (inode
!= dentry
->d_inode
)
1293 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1294 ret
= nfs_intent_set_file(nd
, ctx
);
1304 put_nfs_open_context(ctx
);
1310 return nfs_lookup_revalidate(dentry
, nd
);
1313 static int nfs_open_create(struct inode
*dir
, struct dentry
*dentry
, int mode
,
1314 struct nameidata
*nd
)
1316 struct nfs_open_context
*ctx
= NULL
;
1321 dfprintk(VFS
, "NFS: create(%s/%ld), %s\n",
1322 dir
->i_sb
->s_id
, dir
->i_ino
, dentry
->d_name
.name
);
1324 attr
.ia_mode
= mode
;
1325 attr
.ia_valid
= ATTR_MODE
;
1327 if ((nd
->flags
& LOOKUP_CREATE
) != 0) {
1328 open_flags
= nd
->intent
.open
.flags
;
1330 ctx
= nameidata_to_nfs_open_context(dentry
, nd
);
1331 error
= PTR_ERR(ctx
);
1336 error
= NFS_PROTO(dir
)->create(dir
, dentry
, &attr
, open_flags
, ctx
);
1340 error
= nfs_intent_set_file(nd
, ctx
);
1347 put_nfs_open_context(ctx
);
1354 #endif /* CONFIG_NFSV4 */
1357 * Code common to create, mkdir, and mknod.
1359 int nfs_instantiate(struct dentry
*dentry
, struct nfs_fh
*fhandle
,
1360 struct nfs_fattr
*fattr
)
1362 struct dentry
*parent
= dget_parent(dentry
);
1363 struct inode
*dir
= parent
->d_inode
;
1364 struct inode
*inode
;
1365 int error
= -EACCES
;
1369 /* We may have been initialized further down */
1370 if (dentry
->d_inode
)
1372 if (fhandle
->size
== 0) {
1373 error
= NFS_PROTO(dir
)->lookup(dir
, &dentry
->d_name
, fhandle
, fattr
);
1377 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1378 if (!(fattr
->valid
& NFS_ATTR_FATTR
)) {
1379 struct nfs_server
*server
= NFS_SB(dentry
->d_sb
);
1380 error
= server
->nfs_client
->rpc_ops
->getattr(server
, fhandle
, fattr
);
1384 inode
= nfs_fhget(dentry
->d_sb
, fhandle
, fattr
);
1385 error
= PTR_ERR(inode
);
1388 d_add(dentry
, inode
);
1393 nfs_mark_for_revalidate(dir
);
1399 * Following a failed create operation, we drop the dentry rather
1400 * than retain a negative dentry. This avoids a problem in the event
1401 * that the operation succeeded on the server, but an error in the
1402 * reply path made it appear to have failed.
1404 static int nfs_create(struct inode
*dir
, struct dentry
*dentry
, int mode
,
1405 struct nameidata
*nd
)
1410 dfprintk(VFS
, "NFS: create(%s/%ld), %s\n",
1411 dir
->i_sb
->s_id
, dir
->i_ino
, dentry
->d_name
.name
);
1413 attr
.ia_mode
= mode
;
1414 attr
.ia_valid
= ATTR_MODE
;
1416 error
= NFS_PROTO(dir
)->create(dir
, dentry
, &attr
, 0, NULL
);
1426 * See comments for nfs_proc_create regarding failed operations.
1429 nfs_mknod(struct inode
*dir
, struct dentry
*dentry
, int mode
, dev_t rdev
)
1434 dfprintk(VFS
, "NFS: mknod(%s/%ld), %s\n",
1435 dir
->i_sb
->s_id
, dir
->i_ino
, dentry
->d_name
.name
);
1437 if (!new_valid_dev(rdev
))
1440 attr
.ia_mode
= mode
;
1441 attr
.ia_valid
= ATTR_MODE
;
1443 status
= NFS_PROTO(dir
)->mknod(dir
, dentry
, &attr
, rdev
);
1453 * See comments for nfs_proc_create regarding failed operations.
1455 static int nfs_mkdir(struct inode
*dir
, struct dentry
*dentry
, int mode
)
1460 dfprintk(VFS
, "NFS: mkdir(%s/%ld), %s\n",
1461 dir
->i_sb
->s_id
, dir
->i_ino
, dentry
->d_name
.name
);
1463 attr
.ia_valid
= ATTR_MODE
;
1464 attr
.ia_mode
= mode
| S_IFDIR
;
1466 error
= NFS_PROTO(dir
)->mkdir(dir
, dentry
, &attr
);
1475 static void nfs_dentry_handle_enoent(struct dentry
*dentry
)
1477 if (dentry
->d_inode
!= NULL
&& !d_unhashed(dentry
))
1481 static int nfs_rmdir(struct inode
*dir
, struct dentry
*dentry
)
1485 dfprintk(VFS
, "NFS: rmdir(%s/%ld), %s\n",
1486 dir
->i_sb
->s_id
, dir
->i_ino
, dentry
->d_name
.name
);
1488 error
= NFS_PROTO(dir
)->rmdir(dir
, &dentry
->d_name
);
1489 /* Ensure the VFS deletes this inode */
1490 if (error
== 0 && dentry
->d_inode
!= NULL
)
1491 clear_nlink(dentry
->d_inode
);
1492 else if (error
== -ENOENT
)
1493 nfs_dentry_handle_enoent(dentry
);
1499 * Remove a file after making sure there are no pending writes,
1500 * and after checking that the file has only one user.
1502 * We invalidate the attribute cache and free the inode prior to the operation
1503 * to avoid possible races if the server reuses the inode.
1505 static int nfs_safe_remove(struct dentry
*dentry
)
1507 struct inode
*dir
= dentry
->d_parent
->d_inode
;
1508 struct inode
*inode
= dentry
->d_inode
;
1511 dfprintk(VFS
, "NFS: safe_remove(%s/%s)\n",
1512 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
);
1514 /* If the dentry was sillyrenamed, we simply call d_delete() */
1515 if (dentry
->d_flags
& DCACHE_NFSFS_RENAMED
) {
1520 if (inode
!= NULL
) {
1521 nfs_inode_return_delegation(inode
);
1522 error
= NFS_PROTO(dir
)->remove(dir
, &dentry
->d_name
);
1523 /* The VFS may want to delete this inode */
1525 nfs_drop_nlink(inode
);
1526 nfs_mark_for_revalidate(inode
);
1528 error
= NFS_PROTO(dir
)->remove(dir
, &dentry
->d_name
);
1529 if (error
== -ENOENT
)
1530 nfs_dentry_handle_enoent(dentry
);
1535 /* We do silly rename. In case sillyrename() returns -EBUSY, the inode
1536 * belongs to an active ".nfs..." file and we return -EBUSY.
1538 * If sillyrename() returns 0, we do nothing, otherwise we unlink.
1540 static int nfs_unlink(struct inode
*dir
, struct dentry
*dentry
)
1543 int need_rehash
= 0;
1545 dfprintk(VFS
, "NFS: unlink(%s/%ld, %s)\n", dir
->i_sb
->s_id
,
1546 dir
->i_ino
, dentry
->d_name
.name
);
1548 spin_lock(&dcache_lock
);
1549 spin_lock(&dentry
->d_lock
);
1550 if (atomic_read(&dentry
->d_count
) > 1) {
1551 spin_unlock(&dentry
->d_lock
);
1552 spin_unlock(&dcache_lock
);
1553 /* Start asynchronous writeout of the inode */
1554 write_inode_now(dentry
->d_inode
, 0);
1555 error
= nfs_sillyrename(dir
, dentry
);
1558 if (!d_unhashed(dentry
)) {
1562 spin_unlock(&dentry
->d_lock
);
1563 spin_unlock(&dcache_lock
);
1564 error
= nfs_safe_remove(dentry
);
1565 if (!error
|| error
== -ENOENT
) {
1566 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1567 } else if (need_rehash
)
1573 * To create a symbolic link, most file systems instantiate a new inode,
1574 * add a page to it containing the path, then write it out to the disk
1575 * using prepare_write/commit_write.
1577 * Unfortunately the NFS client can't create the in-core inode first
1578 * because it needs a file handle to create an in-core inode (see
1579 * fs/nfs/inode.c:nfs_fhget). We only have a file handle *after* the
1580 * symlink request has completed on the server.
1582 * So instead we allocate a raw page, copy the symname into it, then do
1583 * the SYMLINK request with the page as the buffer. If it succeeds, we
1584 * now have a new file handle and can instantiate an in-core NFS inode
1585 * and move the raw page into its mapping.
1587 static int nfs_symlink(struct inode
*dir
, struct dentry
*dentry
, const char *symname
)
1589 struct pagevec lru_pvec
;
1593 unsigned int pathlen
= strlen(symname
);
1596 dfprintk(VFS
, "NFS: symlink(%s/%ld, %s, %s)\n", dir
->i_sb
->s_id
,
1597 dir
->i_ino
, dentry
->d_name
.name
, symname
);
1599 if (pathlen
> PAGE_SIZE
)
1600 return -ENAMETOOLONG
;
1602 attr
.ia_mode
= S_IFLNK
| S_IRWXUGO
;
1603 attr
.ia_valid
= ATTR_MODE
;
1605 page
= alloc_page(GFP_HIGHUSER
);
1609 kaddr
= kmap_atomic(page
, KM_USER0
);
1610 memcpy(kaddr
, symname
, pathlen
);
1611 if (pathlen
< PAGE_SIZE
)
1612 memset(kaddr
+ pathlen
, 0, PAGE_SIZE
- pathlen
);
1613 kunmap_atomic(kaddr
, KM_USER0
);
1615 error
= NFS_PROTO(dir
)->symlink(dir
, dentry
, page
, pathlen
, &attr
);
1617 dfprintk(VFS
, "NFS: symlink(%s/%ld, %s, %s) error %d\n",
1618 dir
->i_sb
->s_id
, dir
->i_ino
,
1619 dentry
->d_name
.name
, symname
, error
);
1626 * No big deal if we can't add this page to the page cache here.
1627 * READLINK will get the missing page from the server if needed.
1629 pagevec_init(&lru_pvec
, 0);
1630 if (!add_to_page_cache(page
, dentry
->d_inode
->i_mapping
, 0,
1632 pagevec_add(&lru_pvec
, page
);
1633 pagevec_lru_add_file(&lru_pvec
);
1634 SetPageUptodate(page
);
1643 nfs_link(struct dentry
*old_dentry
, struct inode
*dir
, struct dentry
*dentry
)
1645 struct inode
*inode
= old_dentry
->d_inode
;
1648 dfprintk(VFS
, "NFS: link(%s/%s -> %s/%s)\n",
1649 old_dentry
->d_parent
->d_name
.name
, old_dentry
->d_name
.name
,
1650 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
);
1652 nfs_inode_return_delegation(inode
);
1655 error
= NFS_PROTO(dir
)->link(inode
, dir
, &dentry
->d_name
);
1657 atomic_inc(&inode
->i_count
);
1658 d_add(dentry
, inode
);
1665 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a
1666 * different file handle for the same inode after a rename (e.g. when
1667 * moving to a different directory). A fail-safe method to do so would
1668 * be to look up old_dir/old_name, create a link to new_dir/new_name and
1669 * rename the old file using the sillyrename stuff. This way, the original
1670 * file in old_dir will go away when the last process iput()s the inode.
1674 * It actually works quite well. One needs to have the possibility for
1675 * at least one ".nfs..." file in each directory the file ever gets
1676 * moved or linked to which happens automagically with the new
1677 * implementation that only depends on the dcache stuff instead of
1678 * using the inode layer
1680 * Unfortunately, things are a little more complicated than indicated
1681 * above. For a cross-directory move, we want to make sure we can get
1682 * rid of the old inode after the operation. This means there must be
1683 * no pending writes (if it's a file), and the use count must be 1.
1684 * If these conditions are met, we can drop the dentries before doing
1687 static int nfs_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
1688 struct inode
*new_dir
, struct dentry
*new_dentry
)
1690 struct inode
*old_inode
= old_dentry
->d_inode
;
1691 struct inode
*new_inode
= new_dentry
->d_inode
;
1692 struct dentry
*dentry
= NULL
, *rehash
= NULL
;
1695 dfprintk(VFS
, "NFS: rename(%s/%s -> %s/%s, ct=%d)\n",
1696 old_dentry
->d_parent
->d_name
.name
, old_dentry
->d_name
.name
,
1697 new_dentry
->d_parent
->d_name
.name
, new_dentry
->d_name
.name
,
1698 atomic_read(&new_dentry
->d_count
));
1701 * For non-directories, check whether the target is busy and if so,
1702 * make a copy of the dentry and then do a silly-rename. If the
1703 * silly-rename succeeds, the copied dentry is hashed and becomes
1706 if (new_inode
&& !S_ISDIR(new_inode
->i_mode
)) {
1708 * To prevent any new references to the target during the
1709 * rename, we unhash the dentry in advance.
1711 if (!d_unhashed(new_dentry
)) {
1713 rehash
= new_dentry
;
1716 if (atomic_read(&new_dentry
->d_count
) > 2) {
1719 /* copy the target dentry's name */
1720 dentry
= d_alloc(new_dentry
->d_parent
,
1721 &new_dentry
->d_name
);
1725 /* silly-rename the existing target ... */
1726 err
= nfs_sillyrename(new_dir
, new_dentry
);
1730 new_dentry
= dentry
;
1736 nfs_inode_return_delegation(old_inode
);
1737 if (new_inode
!= NULL
)
1738 nfs_inode_return_delegation(new_inode
);
1740 error
= NFS_PROTO(old_dir
)->rename(old_dir
, &old_dentry
->d_name
,
1741 new_dir
, &new_dentry
->d_name
);
1742 nfs_mark_for_revalidate(old_inode
);
1747 if (new_inode
!= NULL
)
1748 nfs_drop_nlink(new_inode
);
1749 d_move(old_dentry
, new_dentry
);
1750 nfs_set_verifier(new_dentry
,
1751 nfs_save_change_attribute(new_dir
));
1752 } else if (error
== -ENOENT
)
1753 nfs_dentry_handle_enoent(old_dentry
);
1755 /* new dentry created? */
1761 static DEFINE_SPINLOCK(nfs_access_lru_lock
);
1762 static LIST_HEAD(nfs_access_lru_list
);
1763 static atomic_long_t nfs_access_nr_entries
;
1765 static void nfs_access_free_entry(struct nfs_access_entry
*entry
)
1767 put_rpccred(entry
->cred
);
1769 smp_mb__before_atomic_dec();
1770 atomic_long_dec(&nfs_access_nr_entries
);
1771 smp_mb__after_atomic_dec();
1774 static void nfs_access_free_list(struct list_head
*head
)
1776 struct nfs_access_entry
*cache
;
1778 while (!list_empty(head
)) {
1779 cache
= list_entry(head
->next
, struct nfs_access_entry
, lru
);
1780 list_del(&cache
->lru
);
1781 nfs_access_free_entry(cache
);
1785 int nfs_access_cache_shrinker(struct shrinker
*shrink
, int nr_to_scan
, gfp_t gfp_mask
)
1788 struct nfs_inode
*nfsi
, *next
;
1789 struct nfs_access_entry
*cache
;
1791 if ((gfp_mask
& GFP_KERNEL
) != GFP_KERNEL
)
1792 return (nr_to_scan
== 0) ? 0 : -1;
1794 spin_lock(&nfs_access_lru_lock
);
1795 list_for_each_entry_safe(nfsi
, next
, &nfs_access_lru_list
, access_cache_inode_lru
) {
1796 struct inode
*inode
;
1798 if (nr_to_scan
-- == 0)
1800 inode
= &nfsi
->vfs_inode
;
1801 spin_lock(&inode
->i_lock
);
1802 if (list_empty(&nfsi
->access_cache_entry_lru
))
1803 goto remove_lru_entry
;
1804 cache
= list_entry(nfsi
->access_cache_entry_lru
.next
,
1805 struct nfs_access_entry
, lru
);
1806 list_move(&cache
->lru
, &head
);
1807 rb_erase(&cache
->rb_node
, &nfsi
->access_cache
);
1808 if (!list_empty(&nfsi
->access_cache_entry_lru
))
1809 list_move_tail(&nfsi
->access_cache_inode_lru
,
1810 &nfs_access_lru_list
);
1813 list_del_init(&nfsi
->access_cache_inode_lru
);
1814 smp_mb__before_clear_bit();
1815 clear_bit(NFS_INO_ACL_LRU_SET
, &nfsi
->flags
);
1816 smp_mb__after_clear_bit();
1818 spin_unlock(&inode
->i_lock
);
1820 spin_unlock(&nfs_access_lru_lock
);
1821 nfs_access_free_list(&head
);
1822 return (atomic_long_read(&nfs_access_nr_entries
) / 100) * sysctl_vfs_cache_pressure
;
1825 static void __nfs_access_zap_cache(struct nfs_inode
*nfsi
, struct list_head
*head
)
1827 struct rb_root
*root_node
= &nfsi
->access_cache
;
1829 struct nfs_access_entry
*entry
;
1831 /* Unhook entries from the cache */
1832 while ((n
= rb_first(root_node
)) != NULL
) {
1833 entry
= rb_entry(n
, struct nfs_access_entry
, rb_node
);
1834 rb_erase(n
, root_node
);
1835 list_move(&entry
->lru
, head
);
1837 nfsi
->cache_validity
&= ~NFS_INO_INVALID_ACCESS
;
1840 void nfs_access_zap_cache(struct inode
*inode
)
1844 if (test_bit(NFS_INO_ACL_LRU_SET
, &NFS_I(inode
)->flags
) == 0)
1846 /* Remove from global LRU init */
1847 spin_lock(&nfs_access_lru_lock
);
1848 if (test_and_clear_bit(NFS_INO_ACL_LRU_SET
, &NFS_I(inode
)->flags
))
1849 list_del_init(&NFS_I(inode
)->access_cache_inode_lru
);
1851 spin_lock(&inode
->i_lock
);
1852 __nfs_access_zap_cache(NFS_I(inode
), &head
);
1853 spin_unlock(&inode
->i_lock
);
1854 spin_unlock(&nfs_access_lru_lock
);
1855 nfs_access_free_list(&head
);
1858 static struct nfs_access_entry
*nfs_access_search_rbtree(struct inode
*inode
, struct rpc_cred
*cred
)
1860 struct rb_node
*n
= NFS_I(inode
)->access_cache
.rb_node
;
1861 struct nfs_access_entry
*entry
;
1864 entry
= rb_entry(n
, struct nfs_access_entry
, rb_node
);
1866 if (cred
< entry
->cred
)
1868 else if (cred
> entry
->cred
)
1876 static int nfs_access_get_cached(struct inode
*inode
, struct rpc_cred
*cred
, struct nfs_access_entry
*res
)
1878 struct nfs_inode
*nfsi
= NFS_I(inode
);
1879 struct nfs_access_entry
*cache
;
1882 spin_lock(&inode
->i_lock
);
1883 if (nfsi
->cache_validity
& NFS_INO_INVALID_ACCESS
)
1885 cache
= nfs_access_search_rbtree(inode
, cred
);
1888 if (!nfs_have_delegated_attributes(inode
) &&
1889 !time_in_range_open(jiffies
, cache
->jiffies
, cache
->jiffies
+ nfsi
->attrtimeo
))
1891 res
->jiffies
= cache
->jiffies
;
1892 res
->cred
= cache
->cred
;
1893 res
->mask
= cache
->mask
;
1894 list_move_tail(&cache
->lru
, &nfsi
->access_cache_entry_lru
);
1897 spin_unlock(&inode
->i_lock
);
1900 rb_erase(&cache
->rb_node
, &nfsi
->access_cache
);
1901 list_del(&cache
->lru
);
1902 spin_unlock(&inode
->i_lock
);
1903 nfs_access_free_entry(cache
);
1906 spin_unlock(&inode
->i_lock
);
1907 nfs_access_zap_cache(inode
);
1911 static void nfs_access_add_rbtree(struct inode
*inode
, struct nfs_access_entry
*set
)
1913 struct nfs_inode
*nfsi
= NFS_I(inode
);
1914 struct rb_root
*root_node
= &nfsi
->access_cache
;
1915 struct rb_node
**p
= &root_node
->rb_node
;
1916 struct rb_node
*parent
= NULL
;
1917 struct nfs_access_entry
*entry
;
1919 spin_lock(&inode
->i_lock
);
1920 while (*p
!= NULL
) {
1922 entry
= rb_entry(parent
, struct nfs_access_entry
, rb_node
);
1924 if (set
->cred
< entry
->cred
)
1925 p
= &parent
->rb_left
;
1926 else if (set
->cred
> entry
->cred
)
1927 p
= &parent
->rb_right
;
1931 rb_link_node(&set
->rb_node
, parent
, p
);
1932 rb_insert_color(&set
->rb_node
, root_node
);
1933 list_add_tail(&set
->lru
, &nfsi
->access_cache_entry_lru
);
1934 spin_unlock(&inode
->i_lock
);
1937 rb_replace_node(parent
, &set
->rb_node
, root_node
);
1938 list_add_tail(&set
->lru
, &nfsi
->access_cache_entry_lru
);
1939 list_del(&entry
->lru
);
1940 spin_unlock(&inode
->i_lock
);
1941 nfs_access_free_entry(entry
);
1944 static void nfs_access_add_cache(struct inode
*inode
, struct nfs_access_entry
*set
)
1946 struct nfs_access_entry
*cache
= kmalloc(sizeof(*cache
), GFP_KERNEL
);
1949 RB_CLEAR_NODE(&cache
->rb_node
);
1950 cache
->jiffies
= set
->jiffies
;
1951 cache
->cred
= get_rpccred(set
->cred
);
1952 cache
->mask
= set
->mask
;
1954 nfs_access_add_rbtree(inode
, cache
);
1956 /* Update accounting */
1957 smp_mb__before_atomic_inc();
1958 atomic_long_inc(&nfs_access_nr_entries
);
1959 smp_mb__after_atomic_inc();
1961 /* Add inode to global LRU list */
1962 if (!test_bit(NFS_INO_ACL_LRU_SET
, &NFS_I(inode
)->flags
)) {
1963 spin_lock(&nfs_access_lru_lock
);
1964 if (!test_and_set_bit(NFS_INO_ACL_LRU_SET
, &NFS_I(inode
)->flags
))
1965 list_add_tail(&NFS_I(inode
)->access_cache_inode_lru
,
1966 &nfs_access_lru_list
);
1967 spin_unlock(&nfs_access_lru_lock
);
1971 static int nfs_do_access(struct inode
*inode
, struct rpc_cred
*cred
, int mask
)
1973 struct nfs_access_entry cache
;
1976 status
= nfs_access_get_cached(inode
, cred
, &cache
);
1980 /* Be clever: ask server to check for all possible rights */
1981 cache
.mask
= MAY_EXEC
| MAY_WRITE
| MAY_READ
;
1983 cache
.jiffies
= jiffies
;
1984 status
= NFS_PROTO(inode
)->access(inode
, &cache
);
1986 if (status
== -ESTALE
) {
1987 nfs_zap_caches(inode
);
1988 if (!S_ISDIR(inode
->i_mode
))
1989 set_bit(NFS_INO_STALE
, &NFS_I(inode
)->flags
);
1993 nfs_access_add_cache(inode
, &cache
);
1995 if ((mask
& ~cache
.mask
& (MAY_READ
| MAY_WRITE
| MAY_EXEC
)) == 0)
2000 static int nfs_open_permission_mask(int openflags
)
2004 if (openflags
& FMODE_READ
)
2006 if (openflags
& FMODE_WRITE
)
2008 if (openflags
& FMODE_EXEC
)
2013 int nfs_may_open(struct inode
*inode
, struct rpc_cred
*cred
, int openflags
)
2015 return nfs_do_access(inode
, cred
, nfs_open_permission_mask(openflags
));
2018 int nfs_permission(struct inode
*inode
, int mask
)
2020 struct rpc_cred
*cred
;
2023 nfs_inc_stats(inode
, NFSIOS_VFSACCESS
);
2025 if ((mask
& (MAY_READ
| MAY_WRITE
| MAY_EXEC
)) == 0)
2027 /* Is this sys_access() ? */
2028 if (mask
& (MAY_ACCESS
| MAY_CHDIR
))
2031 switch (inode
->i_mode
& S_IFMT
) {
2035 /* NFSv4 has atomic_open... */
2036 if (nfs_server_capable(inode
, NFS_CAP_ATOMIC_OPEN
)
2037 && (mask
& MAY_OPEN
)
2038 && !(mask
& MAY_EXEC
))
2043 * Optimize away all write operations, since the server
2044 * will check permissions when we perform the op.
2046 if ((mask
& MAY_WRITE
) && !(mask
& MAY_READ
))
2051 if (!NFS_PROTO(inode
)->access
)
2054 cred
= rpc_lookup_cred();
2055 if (!IS_ERR(cred
)) {
2056 res
= nfs_do_access(inode
, cred
, mask
);
2059 res
= PTR_ERR(cred
);
2061 if (!res
&& (mask
& MAY_EXEC
) && !execute_ok(inode
))
2064 dfprintk(VFS
, "NFS: permission(%s/%ld), mask=0x%x, res=%d\n",
2065 inode
->i_sb
->s_id
, inode
->i_ino
, mask
, res
);
2068 res
= nfs_revalidate_inode(NFS_SERVER(inode
), inode
);
2070 res
= generic_permission(inode
, mask
, NULL
);
2076 * version-control: t
2077 * kept-new-versions: 5