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>
36 #include <linux/vmalloc.h>
38 #include "delegation.h"
43 /* #define NFS_DEBUG_VERBOSE 1 */
45 static int nfs_opendir(struct inode
*, struct file
*);
46 static int nfs_readdir(struct file
*, void *, filldir_t
);
47 static struct dentry
*nfs_lookup(struct inode
*, struct dentry
*, struct nameidata
*);
48 static int nfs_create(struct inode
*, struct dentry
*, int, struct nameidata
*);
49 static int nfs_mkdir(struct inode
*, struct dentry
*, int);
50 static int nfs_rmdir(struct inode
*, struct dentry
*);
51 static int nfs_unlink(struct inode
*, struct dentry
*);
52 static int nfs_symlink(struct inode
*, struct dentry
*, const char *);
53 static int nfs_link(struct dentry
*, struct inode
*, struct dentry
*);
54 static int nfs_mknod(struct inode
*, struct dentry
*, int, dev_t
);
55 static int nfs_rename(struct inode
*, struct dentry
*,
56 struct inode
*, struct dentry
*);
57 static int nfs_fsync_dir(struct file
*, int);
58 static loff_t
nfs_llseek_dir(struct file
*, loff_t
, int);
59 static int nfs_readdir_clear_array(struct page
*, gfp_t
);
61 const struct file_operations nfs_dir_operations
= {
62 .llseek
= nfs_llseek_dir
,
63 .read
= generic_read_dir
,
64 .readdir
= nfs_readdir
,
66 .release
= nfs_release
,
67 .fsync
= nfs_fsync_dir
,
70 const struct inode_operations nfs_dir_inode_operations
= {
75 .symlink
= nfs_symlink
,
80 .permission
= nfs_permission
,
81 .getattr
= nfs_getattr
,
82 .setattr
= nfs_setattr
,
85 const struct address_space_operations nfs_dir_addr_space_ops
= {
86 .releasepage
= nfs_readdir_clear_array
,
90 const struct inode_operations nfs3_dir_inode_operations
= {
95 .symlink
= nfs_symlink
,
100 .permission
= nfs_permission
,
101 .getattr
= nfs_getattr
,
102 .setattr
= nfs_setattr
,
103 .listxattr
= nfs3_listxattr
,
104 .getxattr
= nfs3_getxattr
,
105 .setxattr
= nfs3_setxattr
,
106 .removexattr
= nfs3_removexattr
,
108 #endif /* CONFIG_NFS_V3 */
112 static struct dentry
*nfs_atomic_lookup(struct inode
*, struct dentry
*, struct nameidata
*);
113 static int nfs_open_create(struct inode
*dir
, struct dentry
*dentry
, int mode
, struct nameidata
*nd
);
114 const struct inode_operations nfs4_dir_inode_operations
= {
115 .create
= nfs_open_create
,
116 .lookup
= nfs_atomic_lookup
,
118 .unlink
= nfs_unlink
,
119 .symlink
= nfs_symlink
,
123 .rename
= nfs_rename
,
124 .permission
= nfs_permission
,
125 .getattr
= nfs_getattr
,
126 .setattr
= nfs_setattr
,
127 .getxattr
= nfs4_getxattr
,
128 .setxattr
= nfs4_setxattr
,
129 .listxattr
= nfs4_listxattr
,
132 #endif /* CONFIG_NFS_V4 */
138 nfs_opendir(struct inode
*inode
, struct file
*filp
)
142 dfprintk(FILE, "NFS: open dir(%s/%s)\n",
143 filp
->f_path
.dentry
->d_parent
->d_name
.name
,
144 filp
->f_path
.dentry
->d_name
.name
);
146 nfs_inc_stats(inode
, NFSIOS_VFSOPEN
);
148 /* Call generic open code in order to cache credentials */
149 res
= nfs_open(inode
, filp
);
150 if (filp
->f_path
.dentry
== filp
->f_path
.mnt
->mnt_root
) {
151 /* This is a mountpoint, so d_revalidate will never
152 * have been called, so we need to refresh the
153 * inode (for close-open consistency) ourselves.
155 __nfs_revalidate_inode(NFS_SERVER(inode
), inode
);
160 struct nfs_cache_array_entry
{
166 struct nfs_cache_array
{
170 struct nfs_cache_array_entry array
[0];
173 #define MAX_READDIR_ARRAY ((PAGE_SIZE - sizeof(struct nfs_cache_array)) / sizeof(struct nfs_cache_array_entry))
175 typedef __be32
* (*decode_dirent_t
)(struct xdr_stream
*, struct nfs_entry
*, struct nfs_server
*, int);
179 unsigned long page_index
;
181 loff_t current_index
;
182 decode_dirent_t decode
;
184 unsigned long timestamp
;
185 unsigned long gencount
;
186 unsigned int cache_entry_index
;
189 } nfs_readdir_descriptor_t
;
192 * The caller is responsible for calling nfs_readdir_release_array(page)
195 struct nfs_cache_array
*nfs_readdir_get_array(struct page
*page
)
198 return ERR_PTR(-EIO
);
199 return (struct nfs_cache_array
*)kmap(page
);
203 void nfs_readdir_release_array(struct page
*page
)
209 * we are freeing strings created by nfs_add_to_readdir_array()
212 int nfs_readdir_clear_array(struct page
*page
, gfp_t mask
)
214 struct nfs_cache_array
*array
= nfs_readdir_get_array(page
);
216 for (i
= 0; i
< array
->size
; i
++)
217 kfree(array
->array
[i
].string
.name
);
218 nfs_readdir_release_array(page
);
223 * the caller is responsible for freeing qstr.name
224 * when called by nfs_readdir_add_to_array, the strings will be freed in
225 * nfs_clear_readdir_array()
228 void nfs_readdir_make_qstr(struct qstr
*string
, const char *name
, unsigned int len
)
231 string
->name
= kmemdup(name
, len
, GFP_KERNEL
);
232 string
->hash
= full_name_hash(string
->name
, string
->len
);
236 int nfs_readdir_add_to_array(struct nfs_entry
*entry
, struct page
*page
)
238 struct nfs_cache_array
*array
= nfs_readdir_get_array(page
);
240 return PTR_ERR(array
);
241 if (array
->size
>= MAX_READDIR_ARRAY
) {
242 nfs_readdir_release_array(page
);
246 array
->array
[array
->size
].cookie
= entry
->prev_cookie
;
247 array
->last_cookie
= entry
->cookie
;
248 array
->array
[array
->size
].ino
= entry
->ino
;
249 nfs_readdir_make_qstr(&array
->array
[array
->size
].string
, entry
->name
, entry
->len
);
251 array
->eof_index
= array
->size
;
253 nfs_readdir_release_array(page
);
258 int nfs_readdir_search_for_pos(struct nfs_cache_array
*array
, nfs_readdir_descriptor_t
*desc
)
260 loff_t diff
= desc
->file
->f_pos
- desc
->current_index
;
265 if (diff
>= array
->size
) {
266 if (array
->eof_index
> 0)
268 desc
->current_index
+= array
->size
;
272 index
= (unsigned int)diff
;
273 *desc
->dir_cookie
= array
->array
[index
].cookie
;
274 desc
->cache_entry_index
= index
;
275 if (index
== array
->eof_index
)
284 int nfs_readdir_search_for_cookie(struct nfs_cache_array
*array
, nfs_readdir_descriptor_t
*desc
)
287 int status
= -EAGAIN
;
289 for (i
= 0; i
< array
->size
; i
++) {
290 if (i
== array
->eof_index
) {
292 status
= -EBADCOOKIE
;
294 if (array
->array
[i
].cookie
== *desc
->dir_cookie
) {
295 desc
->cache_entry_index
= i
;
305 int nfs_readdir_search_array(nfs_readdir_descriptor_t
*desc
)
307 struct nfs_cache_array
*array
;
308 int status
= -EBADCOOKIE
;
310 if (desc
->dir_cookie
== NULL
)
313 array
= nfs_readdir_get_array(desc
->page
);
315 status
= PTR_ERR(array
);
319 if (*desc
->dir_cookie
== 0)
320 status
= nfs_readdir_search_for_pos(array
, desc
);
322 status
= nfs_readdir_search_for_cookie(array
, desc
);
324 nfs_readdir_release_array(desc
->page
);
329 /* Fill a page with xdr information before transferring to the cache page */
331 int nfs_readdir_xdr_filler(struct page
**pages
, nfs_readdir_descriptor_t
*desc
,
332 struct nfs_entry
*entry
, struct file
*file
, struct inode
*inode
)
334 struct rpc_cred
*cred
= nfs_file_cred(file
);
335 unsigned long timestamp
, gencount
;
340 gencount
= nfs_inc_attr_generation_counter();
341 error
= NFS_PROTO(inode
)->readdir(file
->f_path
.dentry
, cred
, entry
->cookie
, pages
,
342 NFS_SERVER(inode
)->dtsize
, desc
->plus
);
344 /* We requested READDIRPLUS, but the server doesn't grok it */
345 if (error
== -ENOTSUPP
&& desc
->plus
) {
346 NFS_SERVER(inode
)->caps
&= ~NFS_CAP_READDIRPLUS
;
347 clear_bit(NFS_INO_ADVISE_RDPLUS
, &NFS_I(inode
)->flags
);
353 desc
->timestamp
= timestamp
;
354 desc
->gencount
= gencount
;
359 /* Fill in an entry based on the xdr code stored in desc->page */
361 int xdr_decode(nfs_readdir_descriptor_t
*desc
, struct nfs_entry
*entry
, struct xdr_stream
*stream
)
363 __be32
*p
= desc
->decode(stream
, entry
, NFS_SERVER(desc
->file
->f_path
.dentry
->d_inode
), desc
->plus
);
367 entry
->fattr
->time_start
= desc
->timestamp
;
368 entry
->fattr
->gencount
= desc
->gencount
;
373 int nfs_same_file(struct dentry
*dentry
, struct nfs_entry
*entry
)
375 struct nfs_inode
*node
;
376 if (dentry
->d_inode
== NULL
)
378 node
= NFS_I(dentry
->d_inode
);
379 if (node
->fh
.size
!= entry
->fh
->size
)
381 if (strncmp(node
->fh
.data
, entry
->fh
->data
, node
->fh
.size
) != 0)
389 void nfs_prime_dcache(struct dentry
*parent
, struct nfs_entry
*entry
)
391 struct qstr filename
;
392 struct dentry
*dentry
= NULL
;
393 struct dentry
*alias
= NULL
;
394 struct inode
*dir
= parent
->d_inode
;
397 nfs_readdir_make_qstr(&filename
, entry
->name
, entry
->len
);
398 if (filename
.len
== 1 && filename
.name
[0] == '.')
399 dentry
= dget(parent
);
400 else if (filename
.len
== 2 && filename
.name
[0] == '.'
401 && filename
.name
[1] == '.')
402 dentry
= dget_parent(parent
);
404 dentry
= d_lookup(parent
, &filename
);
406 if (dentry
!= NULL
) {
407 if (nfs_same_file(dentry
, entry
)) {
408 nfs_refresh_inode(dentry
->d_inode
, entry
->fattr
);
416 dentry
= d_alloc(parent
, &filename
);
417 dentry
->d_op
= NFS_PROTO(dir
)->dentry_ops
;
418 inode
= nfs_fhget(dentry
->d_sb
, entry
->fh
, entry
->fattr
);
422 alias
= d_materialise_unique(dentry
, inode
);
426 nfs_set_verifier(alias
, nfs_save_change_attribute(dir
));
429 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
433 kfree(filename
.name
);
437 /* Perform conversion from xdr to cache array */
439 void nfs_readdir_page_filler(nfs_readdir_descriptor_t
*desc
, struct nfs_entry
*entry
,
440 void *xdr_page
, struct page
*page
, unsigned int buflen
)
442 struct xdr_stream stream
;
444 __be32
*ptr
= xdr_page
;
446 struct nfs_cache_array
*array
;
448 buf
.head
->iov_base
= xdr_page
;
449 buf
.head
->iov_len
= buflen
;
450 buf
.tail
->iov_len
= 0;
453 buf
.buflen
= buf
.head
->iov_len
;
454 buf
.len
= buf
.head
->iov_len
;
456 xdr_init_decode(&stream
, &buf
, ptr
);
460 status
= xdr_decode(desc
, entry
, &stream
);
464 if (nfs_readdir_add_to_array(entry
, page
) == -1)
467 nfs_prime_dcache(desc
->file
->f_path
.dentry
, entry
);
468 } while (!entry
->eof
);
470 if (status
== -EBADCOOKIE
&& entry
->eof
) {
471 array
= nfs_readdir_get_array(page
);
472 array
->eof_index
= array
->size
- 1;
474 nfs_readdir_release_array(page
);
479 void nfs_readdir_free_pagearray(struct page
**pages
, unsigned int npages
)
482 for (i
= 0; i
< npages
; i
++)
487 void nfs_readdir_free_large_page(void *ptr
, struct page
**pages
,
490 vm_unmap_ram(ptr
, npages
);
491 nfs_readdir_free_pagearray(pages
, npages
);
495 * nfs_readdir_large_page will allocate pages that must be freed with a call
496 * to nfs_readdir_free_large_page
499 void *nfs_readdir_large_page(struct page
**pages
, unsigned int npages
)
504 for (i
= 0; i
< npages
; i
++) {
505 struct page
*page
= alloc_page(GFP_KERNEL
);
511 ptr
= vm_map_ram(pages
, NFS_MAX_READDIR_PAGES
, 0, PAGE_KERNEL
);
512 if (!IS_ERR_OR_NULL(ptr
))
515 nfs_readdir_free_pagearray(pages
, i
);
520 int nfs_readdir_xdr_to_array(nfs_readdir_descriptor_t
*desc
, struct page
*page
, struct inode
*inode
)
522 struct page
*pages
[NFS_MAX_READDIR_PAGES
];
523 void *pages_ptr
= NULL
;
524 struct nfs_entry entry
;
525 struct file
*file
= desc
->file
;
526 struct nfs_cache_array
*array
;
528 unsigned int array_size
= ARRAY_SIZE(pages
);
530 entry
.prev_cookie
= 0;
531 entry
.cookie
= *desc
->dir_cookie
;
533 entry
.fh
= nfs_alloc_fhandle();
534 entry
.fattr
= nfs_alloc_fattr();
535 if (entry
.fh
== NULL
|| entry
.fattr
== NULL
)
538 array
= nfs_readdir_get_array(page
);
539 memset(array
, 0, sizeof(struct nfs_cache_array
));
540 array
->eof_index
= -1;
542 pages_ptr
= nfs_readdir_large_page(pages
, array_size
);
544 goto out_release_array
;
546 status
= nfs_readdir_xdr_filler(pages
, desc
, &entry
, file
, inode
);
550 nfs_readdir_page_filler(desc
, &entry
, pages_ptr
, page
, array_size
* PAGE_SIZE
);
551 } while (array
->eof_index
< 0 && array
->size
< MAX_READDIR_ARRAY
);
553 nfs_readdir_free_large_page(pages_ptr
, pages
, array_size
);
555 nfs_readdir_release_array(page
);
557 nfs_free_fattr(entry
.fattr
);
558 nfs_free_fhandle(entry
.fh
);
563 * Now we cache directories properly, by converting xdr information
564 * to an array that can be used for lookups later. This results in
565 * fewer cache pages, since we can store more information on each page.
566 * We only need to convert from xdr once so future lookups are much simpler
569 int nfs_readdir_filler(nfs_readdir_descriptor_t
*desc
, struct page
* page
)
571 struct inode
*inode
= desc
->file
->f_path
.dentry
->d_inode
;
573 if (nfs_readdir_xdr_to_array(desc
, page
, inode
) < 0)
575 SetPageUptodate(page
);
577 if (invalidate_inode_pages2_range(inode
->i_mapping
, page
->index
+ 1, -1) < 0) {
578 /* Should never happen */
579 nfs_zap_mapping(inode
, inode
->i_mapping
);
589 void cache_page_release(nfs_readdir_descriptor_t
*desc
)
591 page_cache_release(desc
->page
);
596 struct page
*get_cache_page(nfs_readdir_descriptor_t
*desc
)
599 page
= read_cache_page(desc
->file
->f_path
.dentry
->d_inode
->i_mapping
,
600 desc
->page_index
, (filler_t
*)nfs_readdir_filler
, desc
);
607 * Returns 0 if desc->dir_cookie was found on page desc->page_index
610 int find_cache_page(nfs_readdir_descriptor_t
*desc
)
614 desc
->page
= get_cache_page(desc
);
615 if (IS_ERR(desc
->page
))
616 return PTR_ERR(desc
->page
);
618 res
= nfs_readdir_search_array(desc
);
621 cache_page_release(desc
);
625 /* Search for desc->dir_cookie from the beginning of the page cache */
627 int readdir_search_pagecache(nfs_readdir_descriptor_t
*desc
)
632 res
= find_cache_page(desc
);
640 static inline unsigned int dt_type(struct inode
*inode
)
642 return (inode
->i_mode
>> 12) & 15;
646 * Once we've found the start of the dirent within a page: fill 'er up...
649 int nfs_do_filldir(nfs_readdir_descriptor_t
*desc
, void *dirent
,
652 struct file
*file
= desc
->file
;
655 struct nfs_cache_array
*array
= NULL
;
656 unsigned int d_type
= DT_UNKNOWN
;
657 struct dentry
*dentry
= NULL
;
659 array
= nfs_readdir_get_array(desc
->page
);
661 for (i
= desc
->cache_entry_index
; i
< array
->size
; i
++) {
664 res
= filldir(dirent
, array
->array
[i
].string
.name
,
665 array
->array
[i
].string
.len
, file
->f_pos
,
666 nfs_compat_user_ino64(array
->array
[i
].ino
), d_type
);
670 desc
->cache_entry_index
= i
;
671 if (i
< (array
->size
-1))
672 *desc
->dir_cookie
= array
->array
[i
+1].cookie
;
674 *desc
->dir_cookie
= array
->last_cookie
;
675 if (i
== array
->eof_index
) {
681 nfs_readdir_release_array(desc
->page
);
682 cache_page_release(desc
);
685 dfprintk(DIRCACHE
, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n",
686 (unsigned long long)*desc
->dir_cookie
, res
);
691 * If we cannot find a cookie in our cache, we suspect that this is
692 * because it points to a deleted file, so we ask the server to return
693 * whatever it thinks is the next entry. We then feed this to filldir.
694 * If all goes well, we should then be able to find our way round the
695 * cache on the next call to readdir_search_pagecache();
697 * NOTE: we cannot add the anonymous page to the pagecache because
698 * the data it contains might not be page aligned. Besides,
699 * we should already have a complete representation of the
700 * directory in the page cache by the time we get here.
703 int uncached_readdir(nfs_readdir_descriptor_t
*desc
, void *dirent
,
706 struct page
*page
= NULL
;
708 struct inode
*inode
= desc
->file
->f_path
.dentry
->d_inode
;
710 dfprintk(DIRCACHE
, "NFS: uncached_readdir() searching for cookie %Lu\n",
711 (unsigned long long)*desc
->dir_cookie
);
713 page
= alloc_page(GFP_HIGHUSER
);
719 if (nfs_readdir_xdr_to_array(desc
, page
, inode
) == -1) {
724 desc
->page_index
= 0;
726 status
= nfs_do_filldir(desc
, dirent
, filldir
);
729 dfprintk(DIRCACHE
, "NFS: %s: returns %d\n",
733 cache_page_release(desc
);
737 /* The file offset position represents the dirent entry number. A
738 last cookie cache takes care of the common case of reading the
741 static int nfs_readdir(struct file
*filp
, void *dirent
, filldir_t filldir
)
743 struct dentry
*dentry
= filp
->f_path
.dentry
;
744 struct inode
*inode
= dentry
->d_inode
;
745 nfs_readdir_descriptor_t my_desc
,
749 dfprintk(FILE, "NFS: readdir(%s/%s) starting at cookie %llu\n",
750 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
,
751 (long long)filp
->f_pos
);
752 nfs_inc_stats(inode
, NFSIOS_VFSGETDENTS
);
755 * filp->f_pos points to the dirent entry number.
756 * *desc->dir_cookie has the cookie for the next entry. We have
757 * to either find the entry with the appropriate number or
758 * revalidate the cookie.
760 memset(desc
, 0, sizeof(*desc
));
763 desc
->dir_cookie
= &nfs_file_open_context(filp
)->dir_cookie
;
764 desc
->decode
= NFS_PROTO(inode
)->decode_dirent
;
765 desc
->plus
= NFS_USE_READDIRPLUS(inode
);
767 nfs_block_sillyrename(dentry
);
768 res
= nfs_revalidate_mapping(inode
, filp
->f_mapping
);
772 while (desc
->eof
!= 1) {
773 res
= readdir_search_pagecache(desc
);
775 if (res
== -EBADCOOKIE
) {
776 /* This means either end of directory */
777 if (*desc
->dir_cookie
&& desc
->eof
== 0) {
778 /* Or that the server has 'lost' a cookie */
779 res
= uncached_readdir(desc
, dirent
, filldir
);
786 if (res
== -ETOOSMALL
&& desc
->plus
) {
787 clear_bit(NFS_INO_ADVISE_RDPLUS
, &NFS_I(inode
)->flags
);
788 nfs_zap_caches(inode
);
789 desc
->page_index
= 0;
797 res
= nfs_do_filldir(desc
, dirent
, filldir
);
804 nfs_unblock_sillyrename(dentry
);
807 dfprintk(FILE, "NFS: readdir(%s/%s) returns %d\n",
808 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
,
813 static loff_t
nfs_llseek_dir(struct file
*filp
, loff_t offset
, int origin
)
815 struct dentry
*dentry
= filp
->f_path
.dentry
;
816 struct inode
*inode
= dentry
->d_inode
;
818 dfprintk(FILE, "NFS: llseek dir(%s/%s, %lld, %d)\n",
819 dentry
->d_parent
->d_name
.name
,
823 mutex_lock(&inode
->i_mutex
);
826 offset
+= filp
->f_pos
;
834 if (offset
!= filp
->f_pos
) {
835 filp
->f_pos
= offset
;
836 nfs_file_open_context(filp
)->dir_cookie
= 0;
839 mutex_unlock(&inode
->i_mutex
);
844 * All directory operations under NFS are synchronous, so fsync()
845 * is a dummy operation.
847 static int nfs_fsync_dir(struct file
*filp
, int datasync
)
849 struct dentry
*dentry
= filp
->f_path
.dentry
;
851 dfprintk(FILE, "NFS: fsync dir(%s/%s) datasync %d\n",
852 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
,
855 nfs_inc_stats(dentry
->d_inode
, NFSIOS_VFSFSYNC
);
860 * nfs_force_lookup_revalidate - Mark the directory as having changed
861 * @dir - pointer to directory inode
863 * This forces the revalidation code in nfs_lookup_revalidate() to do a
864 * full lookup on all child dentries of 'dir' whenever a change occurs
865 * on the server that might have invalidated our dcache.
867 * The caller should be holding dir->i_lock
869 void nfs_force_lookup_revalidate(struct inode
*dir
)
871 NFS_I(dir
)->cache_change_attribute
++;
875 * A check for whether or not the parent directory has changed.
876 * In the case it has, we assume that the dentries are untrustworthy
877 * and may need to be looked up again.
879 static int nfs_check_verifier(struct inode
*dir
, struct dentry
*dentry
)
883 if (NFS_SERVER(dir
)->flags
& NFS_MOUNT_LOOKUP_CACHE_NONE
)
885 if (!nfs_verify_change_attribute(dir
, dentry
->d_time
))
887 /* Revalidate nfsi->cache_change_attribute before we declare a match */
888 if (nfs_revalidate_inode(NFS_SERVER(dir
), dir
) < 0)
890 if (!nfs_verify_change_attribute(dir
, dentry
->d_time
))
896 * Return the intent data that applies to this particular path component
898 * Note that the current set of intents only apply to the very last
899 * component of the path.
900 * We check for this using LOOKUP_CONTINUE and LOOKUP_PARENT.
902 static inline unsigned int nfs_lookup_check_intent(struct nameidata
*nd
, unsigned int mask
)
904 if (nd
->flags
& (LOOKUP_CONTINUE
|LOOKUP_PARENT
))
906 return nd
->flags
& mask
;
910 * Use intent information to check whether or not we're going to do
911 * an O_EXCL create using this path component.
913 static int nfs_is_exclusive_create(struct inode
*dir
, struct nameidata
*nd
)
915 if (NFS_PROTO(dir
)->version
== 2)
917 return nd
&& nfs_lookup_check_intent(nd
, LOOKUP_EXCL
);
921 * Inode and filehandle revalidation for lookups.
923 * We force revalidation in the cases where the VFS sets LOOKUP_REVAL,
924 * or if the intent information indicates that we're about to open this
925 * particular file and the "nocto" mount flag is not set.
929 int nfs_lookup_verify_inode(struct inode
*inode
, struct nameidata
*nd
)
931 struct nfs_server
*server
= NFS_SERVER(inode
);
933 if (test_bit(NFS_INO_MOUNTPOINT
, &NFS_I(inode
)->flags
))
936 /* VFS wants an on-the-wire revalidation */
937 if (nd
->flags
& LOOKUP_REVAL
)
939 /* This is an open(2) */
940 if (nfs_lookup_check_intent(nd
, LOOKUP_OPEN
) != 0 &&
941 !(server
->flags
& NFS_MOUNT_NOCTO
) &&
942 (S_ISREG(inode
->i_mode
) ||
943 S_ISDIR(inode
->i_mode
)))
947 return nfs_revalidate_inode(server
, inode
);
949 return __nfs_revalidate_inode(server
, inode
);
953 * We judge how long we want to trust negative
954 * dentries by looking at the parent inode mtime.
956 * If parent mtime has changed, we revalidate, else we wait for a
957 * period corresponding to the parent's attribute cache timeout value.
960 int nfs_neg_need_reval(struct inode
*dir
, struct dentry
*dentry
,
961 struct nameidata
*nd
)
963 /* Don't revalidate a negative dentry if we're creating a new file */
964 if (nd
!= NULL
&& nfs_lookup_check_intent(nd
, LOOKUP_CREATE
) != 0)
966 if (NFS_SERVER(dir
)->flags
& NFS_MOUNT_LOOKUP_CACHE_NONEG
)
968 return !nfs_check_verifier(dir
, dentry
);
972 * This is called every time the dcache has a lookup hit,
973 * and we should check whether we can really trust that
976 * NOTE! The hit can be a negative hit too, don't assume
979 * If the parent directory is seen to have changed, we throw out the
980 * cached dentry and do a new lookup.
982 static int nfs_lookup_revalidate(struct dentry
* dentry
, struct nameidata
*nd
)
986 struct dentry
*parent
;
987 struct nfs_fh
*fhandle
= NULL
;
988 struct nfs_fattr
*fattr
= NULL
;
991 parent
= dget_parent(dentry
);
992 dir
= parent
->d_inode
;
993 nfs_inc_stats(dir
, NFSIOS_DENTRYREVALIDATE
);
994 inode
= dentry
->d_inode
;
997 if (nfs_neg_need_reval(dir
, dentry
, nd
))
1002 if (is_bad_inode(inode
)) {
1003 dfprintk(LOOKUPCACHE
, "%s: %s/%s has dud inode\n",
1004 __func__
, dentry
->d_parent
->d_name
.name
,
1005 dentry
->d_name
.name
);
1009 if (nfs_have_delegation(inode
, FMODE_READ
))
1010 goto out_set_verifier
;
1012 /* Force a full look up iff the parent directory has changed */
1013 if (!nfs_is_exclusive_create(dir
, nd
) && nfs_check_verifier(dir
, dentry
)) {
1014 if (nfs_lookup_verify_inode(inode
, nd
))
1015 goto out_zap_parent
;
1019 if (NFS_STALE(inode
))
1023 fhandle
= nfs_alloc_fhandle();
1024 fattr
= nfs_alloc_fattr();
1025 if (fhandle
== NULL
|| fattr
== NULL
)
1028 error
= NFS_PROTO(dir
)->lookup(dir
, &dentry
->d_name
, fhandle
, fattr
);
1031 if (nfs_compare_fh(NFS_FH(inode
), fhandle
))
1033 if ((error
= nfs_refresh_inode(inode
, fattr
)) != 0)
1036 nfs_free_fattr(fattr
);
1037 nfs_free_fhandle(fhandle
);
1039 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1042 dfprintk(LOOKUPCACHE
, "NFS: %s(%s/%s) is valid\n",
1043 __func__
, dentry
->d_parent
->d_name
.name
,
1044 dentry
->d_name
.name
);
1047 nfs_zap_caches(dir
);
1049 nfs_mark_for_revalidate(dir
);
1050 if (inode
&& S_ISDIR(inode
->i_mode
)) {
1051 /* Purge readdir caches. */
1052 nfs_zap_caches(inode
);
1053 /* If we have submounts, don't unhash ! */
1054 if (have_submounts(dentry
))
1056 if (dentry
->d_flags
& DCACHE_DISCONNECTED
)
1058 shrink_dcache_parent(dentry
);
1061 nfs_free_fattr(fattr
);
1062 nfs_free_fhandle(fhandle
);
1064 dfprintk(LOOKUPCACHE
, "NFS: %s(%s/%s) is invalid\n",
1065 __func__
, dentry
->d_parent
->d_name
.name
,
1066 dentry
->d_name
.name
);
1069 nfs_free_fattr(fattr
);
1070 nfs_free_fhandle(fhandle
);
1072 dfprintk(LOOKUPCACHE
, "NFS: %s(%s/%s) lookup returned error %d\n",
1073 __func__
, dentry
->d_parent
->d_name
.name
,
1074 dentry
->d_name
.name
, error
);
1079 * This is called from dput() when d_count is going to 0.
1081 static int nfs_dentry_delete(struct dentry
*dentry
)
1083 dfprintk(VFS
, "NFS: dentry_delete(%s/%s, %x)\n",
1084 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
,
1087 /* Unhash any dentry with a stale inode */
1088 if (dentry
->d_inode
!= NULL
&& NFS_STALE(dentry
->d_inode
))
1091 if (dentry
->d_flags
& DCACHE_NFSFS_RENAMED
) {
1092 /* Unhash it, so that ->d_iput() would be called */
1095 if (!(dentry
->d_sb
->s_flags
& MS_ACTIVE
)) {
1096 /* Unhash it, so that ancestors of killed async unlink
1097 * files will be cleaned up during umount */
1104 static void nfs_drop_nlink(struct inode
*inode
)
1106 spin_lock(&inode
->i_lock
);
1107 if (inode
->i_nlink
> 0)
1109 spin_unlock(&inode
->i_lock
);
1113 * Called when the dentry loses inode.
1114 * We use it to clean up silly-renamed files.
1116 static void nfs_dentry_iput(struct dentry
*dentry
, struct inode
*inode
)
1118 if (S_ISDIR(inode
->i_mode
))
1119 /* drop any readdir cache as it could easily be old */
1120 NFS_I(inode
)->cache_validity
|= NFS_INO_INVALID_DATA
;
1122 if (dentry
->d_flags
& DCACHE_NFSFS_RENAMED
) {
1124 nfs_complete_unlink(dentry
, inode
);
1129 const struct dentry_operations nfs_dentry_operations
= {
1130 .d_revalidate
= nfs_lookup_revalidate
,
1131 .d_delete
= nfs_dentry_delete
,
1132 .d_iput
= nfs_dentry_iput
,
1135 static struct dentry
*nfs_lookup(struct inode
*dir
, struct dentry
* dentry
, struct nameidata
*nd
)
1138 struct dentry
*parent
;
1139 struct inode
*inode
= NULL
;
1140 struct nfs_fh
*fhandle
= NULL
;
1141 struct nfs_fattr
*fattr
= NULL
;
1144 dfprintk(VFS
, "NFS: lookup(%s/%s)\n",
1145 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
);
1146 nfs_inc_stats(dir
, NFSIOS_VFSLOOKUP
);
1148 res
= ERR_PTR(-ENAMETOOLONG
);
1149 if (dentry
->d_name
.len
> NFS_SERVER(dir
)->namelen
)
1152 dentry
->d_op
= NFS_PROTO(dir
)->dentry_ops
;
1155 * If we're doing an exclusive create, optimize away the lookup
1156 * but don't hash the dentry.
1158 if (nfs_is_exclusive_create(dir
, nd
)) {
1159 d_instantiate(dentry
, NULL
);
1164 res
= ERR_PTR(-ENOMEM
);
1165 fhandle
= nfs_alloc_fhandle();
1166 fattr
= nfs_alloc_fattr();
1167 if (fhandle
== NULL
|| fattr
== NULL
)
1170 parent
= dentry
->d_parent
;
1171 /* Protect against concurrent sillydeletes */
1172 nfs_block_sillyrename(parent
);
1173 error
= NFS_PROTO(dir
)->lookup(dir
, &dentry
->d_name
, fhandle
, fattr
);
1174 if (error
== -ENOENT
)
1177 res
= ERR_PTR(error
);
1178 goto out_unblock_sillyrename
;
1180 inode
= nfs_fhget(dentry
->d_sb
, fhandle
, fattr
);
1181 res
= (struct dentry
*)inode
;
1183 goto out_unblock_sillyrename
;
1186 res
= d_materialise_unique(dentry
, inode
);
1189 goto out_unblock_sillyrename
;
1192 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1193 out_unblock_sillyrename
:
1194 nfs_unblock_sillyrename(parent
);
1196 nfs_free_fattr(fattr
);
1197 nfs_free_fhandle(fhandle
);
1201 #ifdef CONFIG_NFS_V4
1202 static int nfs_open_revalidate(struct dentry
*, struct nameidata
*);
1204 const struct dentry_operations nfs4_dentry_operations
= {
1205 .d_revalidate
= nfs_open_revalidate
,
1206 .d_delete
= nfs_dentry_delete
,
1207 .d_iput
= nfs_dentry_iput
,
1211 * Use intent information to determine whether we need to substitute
1212 * the NFSv4-style stateful OPEN for the LOOKUP call
1214 static int is_atomic_open(struct nameidata
*nd
)
1216 if (nd
== NULL
|| nfs_lookup_check_intent(nd
, LOOKUP_OPEN
) == 0)
1218 /* NFS does not (yet) have a stateful open for directories */
1219 if (nd
->flags
& LOOKUP_DIRECTORY
)
1221 /* Are we trying to write to a read only partition? */
1222 if (__mnt_is_readonly(nd
->path
.mnt
) &&
1223 (nd
->intent
.open
.flags
& (O_CREAT
|O_TRUNC
|FMODE_WRITE
)))
1228 static struct nfs_open_context
*nameidata_to_nfs_open_context(struct dentry
*dentry
, struct nameidata
*nd
)
1230 struct path path
= {
1231 .mnt
= nd
->path
.mnt
,
1234 struct nfs_open_context
*ctx
;
1235 struct rpc_cred
*cred
;
1236 fmode_t fmode
= nd
->intent
.open
.flags
& (FMODE_READ
| FMODE_WRITE
| FMODE_EXEC
);
1238 cred
= rpc_lookup_cred();
1240 return ERR_CAST(cred
);
1241 ctx
= alloc_nfs_open_context(&path
, cred
, fmode
);
1244 return ERR_PTR(-ENOMEM
);
1248 static int do_open(struct inode
*inode
, struct file
*filp
)
1250 nfs_fscache_set_inode_cookie(inode
, filp
);
1254 static int nfs_intent_set_file(struct nameidata
*nd
, struct nfs_open_context
*ctx
)
1259 /* If the open_intent is for execute, we have an extra check to make */
1260 if (ctx
->mode
& FMODE_EXEC
) {
1261 ret
= nfs_may_open(ctx
->path
.dentry
->d_inode
,
1263 nd
->intent
.open
.flags
);
1267 filp
= lookup_instantiate_filp(nd
, ctx
->path
.dentry
, do_open
);
1269 ret
= PTR_ERR(filp
);
1271 nfs_file_set_open_context(filp
, ctx
);
1273 put_nfs_open_context(ctx
);
1277 static struct dentry
*nfs_atomic_lookup(struct inode
*dir
, struct dentry
*dentry
, struct nameidata
*nd
)
1279 struct nfs_open_context
*ctx
;
1281 struct dentry
*res
= NULL
;
1282 struct inode
*inode
;
1286 dfprintk(VFS
, "NFS: atomic_lookup(%s/%ld), %s\n",
1287 dir
->i_sb
->s_id
, dir
->i_ino
, dentry
->d_name
.name
);
1289 /* Check that we are indeed trying to open this file */
1290 if (!is_atomic_open(nd
))
1293 if (dentry
->d_name
.len
> NFS_SERVER(dir
)->namelen
) {
1294 res
= ERR_PTR(-ENAMETOOLONG
);
1297 dentry
->d_op
= NFS_PROTO(dir
)->dentry_ops
;
1299 /* Let vfs_create() deal with O_EXCL. Instantiate, but don't hash
1301 if (nd
->flags
& LOOKUP_EXCL
) {
1302 d_instantiate(dentry
, NULL
);
1306 ctx
= nameidata_to_nfs_open_context(dentry
, nd
);
1307 res
= ERR_CAST(ctx
);
1311 open_flags
= nd
->intent
.open
.flags
;
1312 if (nd
->flags
& LOOKUP_CREATE
) {
1313 attr
.ia_mode
= nd
->intent
.open
.create_mode
;
1314 attr
.ia_valid
= ATTR_MODE
;
1315 if (!IS_POSIXACL(dir
))
1316 attr
.ia_mode
&= ~current_umask();
1318 open_flags
&= ~(O_EXCL
| O_CREAT
);
1322 /* Open the file on the server */
1323 nfs_block_sillyrename(dentry
->d_parent
);
1324 inode
= NFS_PROTO(dir
)->open_context(dir
, ctx
, open_flags
, &attr
);
1325 if (IS_ERR(inode
)) {
1326 nfs_unblock_sillyrename(dentry
->d_parent
);
1327 put_nfs_open_context(ctx
);
1328 switch (PTR_ERR(inode
)) {
1329 /* Make a negative dentry */
1331 d_add(dentry
, NULL
);
1334 /* This turned out not to be a regular file */
1339 if (!(nd
->intent
.open
.flags
& O_NOFOLLOW
))
1343 res
= ERR_CAST(inode
);
1347 res
= d_add_unique(dentry
, inode
);
1348 nfs_unblock_sillyrename(dentry
->d_parent
);
1350 dput(ctx
->path
.dentry
);
1351 ctx
->path
.dentry
= dget(res
);
1354 err
= nfs_intent_set_file(nd
, ctx
);
1358 return ERR_PTR(err
);
1361 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1364 return nfs_lookup(dir
, dentry
, nd
);
1367 static int nfs_open_revalidate(struct dentry
*dentry
, struct nameidata
*nd
)
1369 struct dentry
*parent
= NULL
;
1370 struct inode
*inode
= dentry
->d_inode
;
1372 struct nfs_open_context
*ctx
;
1373 int openflags
, ret
= 0;
1375 if (!is_atomic_open(nd
) || d_mountpoint(dentry
))
1378 parent
= dget_parent(dentry
);
1379 dir
= parent
->d_inode
;
1381 /* We can't create new files in nfs_open_revalidate(), so we
1382 * optimize away revalidation of negative dentries.
1384 if (inode
== NULL
) {
1385 if (!nfs_neg_need_reval(dir
, dentry
, nd
))
1390 /* NFS only supports OPEN on regular files */
1391 if (!S_ISREG(inode
->i_mode
))
1393 openflags
= nd
->intent
.open
.flags
;
1394 /* We cannot do exclusive creation on a positive dentry */
1395 if ((openflags
& (O_CREAT
|O_EXCL
)) == (O_CREAT
|O_EXCL
))
1397 /* We can't create new files, or truncate existing ones here */
1398 openflags
&= ~(O_CREAT
|O_EXCL
|O_TRUNC
);
1400 ctx
= nameidata_to_nfs_open_context(dentry
, nd
);
1405 * Note: we're not holding inode->i_mutex and so may be racing with
1406 * operations that change the directory. We therefore save the
1407 * change attribute *before* we do the RPC call.
1409 inode
= NFS_PROTO(dir
)->open_context(dir
, ctx
, openflags
, NULL
);
1410 if (IS_ERR(inode
)) {
1411 ret
= PTR_ERR(inode
);
1424 if (inode
!= dentry
->d_inode
)
1427 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1428 ret
= nfs_intent_set_file(nd
, ctx
);
1438 put_nfs_open_context(ctx
);
1444 return nfs_lookup_revalidate(dentry
, nd
);
1447 static int nfs_open_create(struct inode
*dir
, struct dentry
*dentry
, int mode
,
1448 struct nameidata
*nd
)
1450 struct nfs_open_context
*ctx
= NULL
;
1455 dfprintk(VFS
, "NFS: create(%s/%ld), %s\n",
1456 dir
->i_sb
->s_id
, dir
->i_ino
, dentry
->d_name
.name
);
1458 attr
.ia_mode
= mode
;
1459 attr
.ia_valid
= ATTR_MODE
;
1461 if ((nd
->flags
& LOOKUP_CREATE
) != 0) {
1462 open_flags
= nd
->intent
.open
.flags
;
1464 ctx
= nameidata_to_nfs_open_context(dentry
, nd
);
1465 error
= PTR_ERR(ctx
);
1470 error
= NFS_PROTO(dir
)->create(dir
, dentry
, &attr
, open_flags
, ctx
);
1474 error
= nfs_intent_set_file(nd
, ctx
);
1481 put_nfs_open_context(ctx
);
1488 #endif /* CONFIG_NFSV4 */
1491 * Code common to create, mkdir, and mknod.
1493 int nfs_instantiate(struct dentry
*dentry
, struct nfs_fh
*fhandle
,
1494 struct nfs_fattr
*fattr
)
1496 struct dentry
*parent
= dget_parent(dentry
);
1497 struct inode
*dir
= parent
->d_inode
;
1498 struct inode
*inode
;
1499 int error
= -EACCES
;
1503 /* We may have been initialized further down */
1504 if (dentry
->d_inode
)
1506 if (fhandle
->size
== 0) {
1507 error
= NFS_PROTO(dir
)->lookup(dir
, &dentry
->d_name
, fhandle
, fattr
);
1511 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1512 if (!(fattr
->valid
& NFS_ATTR_FATTR
)) {
1513 struct nfs_server
*server
= NFS_SB(dentry
->d_sb
);
1514 error
= server
->nfs_client
->rpc_ops
->getattr(server
, fhandle
, fattr
);
1518 inode
= nfs_fhget(dentry
->d_sb
, fhandle
, fattr
);
1519 error
= PTR_ERR(inode
);
1522 d_add(dentry
, inode
);
1527 nfs_mark_for_revalidate(dir
);
1533 * Following a failed create operation, we drop the dentry rather
1534 * than retain a negative dentry. This avoids a problem in the event
1535 * that the operation succeeded on the server, but an error in the
1536 * reply path made it appear to have failed.
1538 static int nfs_create(struct inode
*dir
, struct dentry
*dentry
, int mode
,
1539 struct nameidata
*nd
)
1544 dfprintk(VFS
, "NFS: create(%s/%ld), %s\n",
1545 dir
->i_sb
->s_id
, dir
->i_ino
, dentry
->d_name
.name
);
1547 attr
.ia_mode
= mode
;
1548 attr
.ia_valid
= ATTR_MODE
;
1550 error
= NFS_PROTO(dir
)->create(dir
, dentry
, &attr
, 0, NULL
);
1560 * See comments for nfs_proc_create regarding failed operations.
1563 nfs_mknod(struct inode
*dir
, struct dentry
*dentry
, int mode
, dev_t rdev
)
1568 dfprintk(VFS
, "NFS: mknod(%s/%ld), %s\n",
1569 dir
->i_sb
->s_id
, dir
->i_ino
, dentry
->d_name
.name
);
1571 if (!new_valid_dev(rdev
))
1574 attr
.ia_mode
= mode
;
1575 attr
.ia_valid
= ATTR_MODE
;
1577 status
= NFS_PROTO(dir
)->mknod(dir
, dentry
, &attr
, rdev
);
1587 * See comments for nfs_proc_create regarding failed operations.
1589 static int nfs_mkdir(struct inode
*dir
, struct dentry
*dentry
, int mode
)
1594 dfprintk(VFS
, "NFS: mkdir(%s/%ld), %s\n",
1595 dir
->i_sb
->s_id
, dir
->i_ino
, dentry
->d_name
.name
);
1597 attr
.ia_valid
= ATTR_MODE
;
1598 attr
.ia_mode
= mode
| S_IFDIR
;
1600 error
= NFS_PROTO(dir
)->mkdir(dir
, dentry
, &attr
);
1609 static void nfs_dentry_handle_enoent(struct dentry
*dentry
)
1611 if (dentry
->d_inode
!= NULL
&& !d_unhashed(dentry
))
1615 static int nfs_rmdir(struct inode
*dir
, struct dentry
*dentry
)
1619 dfprintk(VFS
, "NFS: rmdir(%s/%ld), %s\n",
1620 dir
->i_sb
->s_id
, dir
->i_ino
, dentry
->d_name
.name
);
1622 error
= NFS_PROTO(dir
)->rmdir(dir
, &dentry
->d_name
);
1623 /* Ensure the VFS deletes this inode */
1624 if (error
== 0 && dentry
->d_inode
!= NULL
)
1625 clear_nlink(dentry
->d_inode
);
1626 else if (error
== -ENOENT
)
1627 nfs_dentry_handle_enoent(dentry
);
1633 * Remove a file after making sure there are no pending writes,
1634 * and after checking that the file has only one user.
1636 * We invalidate the attribute cache and free the inode prior to the operation
1637 * to avoid possible races if the server reuses the inode.
1639 static int nfs_safe_remove(struct dentry
*dentry
)
1641 struct inode
*dir
= dentry
->d_parent
->d_inode
;
1642 struct inode
*inode
= dentry
->d_inode
;
1645 dfprintk(VFS
, "NFS: safe_remove(%s/%s)\n",
1646 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
);
1648 /* If the dentry was sillyrenamed, we simply call d_delete() */
1649 if (dentry
->d_flags
& DCACHE_NFSFS_RENAMED
) {
1654 if (inode
!= NULL
) {
1655 nfs_inode_return_delegation(inode
);
1656 error
= NFS_PROTO(dir
)->remove(dir
, &dentry
->d_name
);
1657 /* The VFS may want to delete this inode */
1659 nfs_drop_nlink(inode
);
1660 nfs_mark_for_revalidate(inode
);
1662 error
= NFS_PROTO(dir
)->remove(dir
, &dentry
->d_name
);
1663 if (error
== -ENOENT
)
1664 nfs_dentry_handle_enoent(dentry
);
1669 /* We do silly rename. In case sillyrename() returns -EBUSY, the inode
1670 * belongs to an active ".nfs..." file and we return -EBUSY.
1672 * If sillyrename() returns 0, we do nothing, otherwise we unlink.
1674 static int nfs_unlink(struct inode
*dir
, struct dentry
*dentry
)
1677 int need_rehash
= 0;
1679 dfprintk(VFS
, "NFS: unlink(%s/%ld, %s)\n", dir
->i_sb
->s_id
,
1680 dir
->i_ino
, dentry
->d_name
.name
);
1682 spin_lock(&dcache_lock
);
1683 spin_lock(&dentry
->d_lock
);
1684 if (atomic_read(&dentry
->d_count
) > 1) {
1685 spin_unlock(&dentry
->d_lock
);
1686 spin_unlock(&dcache_lock
);
1687 /* Start asynchronous writeout of the inode */
1688 write_inode_now(dentry
->d_inode
, 0);
1689 error
= nfs_sillyrename(dir
, dentry
);
1692 if (!d_unhashed(dentry
)) {
1696 spin_unlock(&dentry
->d_lock
);
1697 spin_unlock(&dcache_lock
);
1698 error
= nfs_safe_remove(dentry
);
1699 if (!error
|| error
== -ENOENT
) {
1700 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1701 } else if (need_rehash
)
1707 * To create a symbolic link, most file systems instantiate a new inode,
1708 * add a page to it containing the path, then write it out to the disk
1709 * using prepare_write/commit_write.
1711 * Unfortunately the NFS client can't create the in-core inode first
1712 * because it needs a file handle to create an in-core inode (see
1713 * fs/nfs/inode.c:nfs_fhget). We only have a file handle *after* the
1714 * symlink request has completed on the server.
1716 * So instead we allocate a raw page, copy the symname into it, then do
1717 * the SYMLINK request with the page as the buffer. If it succeeds, we
1718 * now have a new file handle and can instantiate an in-core NFS inode
1719 * and move the raw page into its mapping.
1721 static int nfs_symlink(struct inode
*dir
, struct dentry
*dentry
, const char *symname
)
1723 struct pagevec lru_pvec
;
1727 unsigned int pathlen
= strlen(symname
);
1730 dfprintk(VFS
, "NFS: symlink(%s/%ld, %s, %s)\n", dir
->i_sb
->s_id
,
1731 dir
->i_ino
, dentry
->d_name
.name
, symname
);
1733 if (pathlen
> PAGE_SIZE
)
1734 return -ENAMETOOLONG
;
1736 attr
.ia_mode
= S_IFLNK
| S_IRWXUGO
;
1737 attr
.ia_valid
= ATTR_MODE
;
1739 page
= alloc_page(GFP_HIGHUSER
);
1743 kaddr
= kmap_atomic(page
, KM_USER0
);
1744 memcpy(kaddr
, symname
, pathlen
);
1745 if (pathlen
< PAGE_SIZE
)
1746 memset(kaddr
+ pathlen
, 0, PAGE_SIZE
- pathlen
);
1747 kunmap_atomic(kaddr
, KM_USER0
);
1749 error
= NFS_PROTO(dir
)->symlink(dir
, dentry
, page
, pathlen
, &attr
);
1751 dfprintk(VFS
, "NFS: symlink(%s/%ld, %s, %s) error %d\n",
1752 dir
->i_sb
->s_id
, dir
->i_ino
,
1753 dentry
->d_name
.name
, symname
, error
);
1760 * No big deal if we can't add this page to the page cache here.
1761 * READLINK will get the missing page from the server if needed.
1763 pagevec_init(&lru_pvec
, 0);
1764 if (!add_to_page_cache(page
, dentry
->d_inode
->i_mapping
, 0,
1766 pagevec_add(&lru_pvec
, page
);
1767 pagevec_lru_add_file(&lru_pvec
);
1768 SetPageUptodate(page
);
1777 nfs_link(struct dentry
*old_dentry
, struct inode
*dir
, struct dentry
*dentry
)
1779 struct inode
*inode
= old_dentry
->d_inode
;
1782 dfprintk(VFS
, "NFS: link(%s/%s -> %s/%s)\n",
1783 old_dentry
->d_parent
->d_name
.name
, old_dentry
->d_name
.name
,
1784 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
);
1786 nfs_inode_return_delegation(inode
);
1789 error
= NFS_PROTO(dir
)->link(inode
, dir
, &dentry
->d_name
);
1791 atomic_inc(&inode
->i_count
);
1792 d_add(dentry
, inode
);
1799 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a
1800 * different file handle for the same inode after a rename (e.g. when
1801 * moving to a different directory). A fail-safe method to do so would
1802 * be to look up old_dir/old_name, create a link to new_dir/new_name and
1803 * rename the old file using the sillyrename stuff. This way, the original
1804 * file in old_dir will go away when the last process iput()s the inode.
1808 * It actually works quite well. One needs to have the possibility for
1809 * at least one ".nfs..." file in each directory the file ever gets
1810 * moved or linked to which happens automagically with the new
1811 * implementation that only depends on the dcache stuff instead of
1812 * using the inode layer
1814 * Unfortunately, things are a little more complicated than indicated
1815 * above. For a cross-directory move, we want to make sure we can get
1816 * rid of the old inode after the operation. This means there must be
1817 * no pending writes (if it's a file), and the use count must be 1.
1818 * If these conditions are met, we can drop the dentries before doing
1821 static int nfs_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
1822 struct inode
*new_dir
, struct dentry
*new_dentry
)
1824 struct inode
*old_inode
= old_dentry
->d_inode
;
1825 struct inode
*new_inode
= new_dentry
->d_inode
;
1826 struct dentry
*dentry
= NULL
, *rehash
= NULL
;
1829 dfprintk(VFS
, "NFS: rename(%s/%s -> %s/%s, ct=%d)\n",
1830 old_dentry
->d_parent
->d_name
.name
, old_dentry
->d_name
.name
,
1831 new_dentry
->d_parent
->d_name
.name
, new_dentry
->d_name
.name
,
1832 atomic_read(&new_dentry
->d_count
));
1835 * For non-directories, check whether the target is busy and if so,
1836 * make a copy of the dentry and then do a silly-rename. If the
1837 * silly-rename succeeds, the copied dentry is hashed and becomes
1840 if (new_inode
&& !S_ISDIR(new_inode
->i_mode
)) {
1842 * To prevent any new references to the target during the
1843 * rename, we unhash the dentry in advance.
1845 if (!d_unhashed(new_dentry
)) {
1847 rehash
= new_dentry
;
1850 if (atomic_read(&new_dentry
->d_count
) > 2) {
1853 /* copy the target dentry's name */
1854 dentry
= d_alloc(new_dentry
->d_parent
,
1855 &new_dentry
->d_name
);
1859 /* silly-rename the existing target ... */
1860 err
= nfs_sillyrename(new_dir
, new_dentry
);
1864 new_dentry
= dentry
;
1870 nfs_inode_return_delegation(old_inode
);
1871 if (new_inode
!= NULL
)
1872 nfs_inode_return_delegation(new_inode
);
1874 error
= NFS_PROTO(old_dir
)->rename(old_dir
, &old_dentry
->d_name
,
1875 new_dir
, &new_dentry
->d_name
);
1876 nfs_mark_for_revalidate(old_inode
);
1881 if (new_inode
!= NULL
)
1882 nfs_drop_nlink(new_inode
);
1883 d_move(old_dentry
, new_dentry
);
1884 nfs_set_verifier(new_dentry
,
1885 nfs_save_change_attribute(new_dir
));
1886 } else if (error
== -ENOENT
)
1887 nfs_dentry_handle_enoent(old_dentry
);
1889 /* new dentry created? */
1895 static DEFINE_SPINLOCK(nfs_access_lru_lock
);
1896 static LIST_HEAD(nfs_access_lru_list
);
1897 static atomic_long_t nfs_access_nr_entries
;
1899 static void nfs_access_free_entry(struct nfs_access_entry
*entry
)
1901 put_rpccred(entry
->cred
);
1903 smp_mb__before_atomic_dec();
1904 atomic_long_dec(&nfs_access_nr_entries
);
1905 smp_mb__after_atomic_dec();
1908 static void nfs_access_free_list(struct list_head
*head
)
1910 struct nfs_access_entry
*cache
;
1912 while (!list_empty(head
)) {
1913 cache
= list_entry(head
->next
, struct nfs_access_entry
, lru
);
1914 list_del(&cache
->lru
);
1915 nfs_access_free_entry(cache
);
1919 int nfs_access_cache_shrinker(struct shrinker
*shrink
, int nr_to_scan
, gfp_t gfp_mask
)
1922 struct nfs_inode
*nfsi
, *next
;
1923 struct nfs_access_entry
*cache
;
1925 if ((gfp_mask
& GFP_KERNEL
) != GFP_KERNEL
)
1926 return (nr_to_scan
== 0) ? 0 : -1;
1928 spin_lock(&nfs_access_lru_lock
);
1929 list_for_each_entry_safe(nfsi
, next
, &nfs_access_lru_list
, access_cache_inode_lru
) {
1930 struct inode
*inode
;
1932 if (nr_to_scan
-- == 0)
1934 inode
= &nfsi
->vfs_inode
;
1935 spin_lock(&inode
->i_lock
);
1936 if (list_empty(&nfsi
->access_cache_entry_lru
))
1937 goto remove_lru_entry
;
1938 cache
= list_entry(nfsi
->access_cache_entry_lru
.next
,
1939 struct nfs_access_entry
, lru
);
1940 list_move(&cache
->lru
, &head
);
1941 rb_erase(&cache
->rb_node
, &nfsi
->access_cache
);
1942 if (!list_empty(&nfsi
->access_cache_entry_lru
))
1943 list_move_tail(&nfsi
->access_cache_inode_lru
,
1944 &nfs_access_lru_list
);
1947 list_del_init(&nfsi
->access_cache_inode_lru
);
1948 smp_mb__before_clear_bit();
1949 clear_bit(NFS_INO_ACL_LRU_SET
, &nfsi
->flags
);
1950 smp_mb__after_clear_bit();
1952 spin_unlock(&inode
->i_lock
);
1954 spin_unlock(&nfs_access_lru_lock
);
1955 nfs_access_free_list(&head
);
1956 return (atomic_long_read(&nfs_access_nr_entries
) / 100) * sysctl_vfs_cache_pressure
;
1959 static void __nfs_access_zap_cache(struct nfs_inode
*nfsi
, struct list_head
*head
)
1961 struct rb_root
*root_node
= &nfsi
->access_cache
;
1963 struct nfs_access_entry
*entry
;
1965 /* Unhook entries from the cache */
1966 while ((n
= rb_first(root_node
)) != NULL
) {
1967 entry
= rb_entry(n
, struct nfs_access_entry
, rb_node
);
1968 rb_erase(n
, root_node
);
1969 list_move(&entry
->lru
, head
);
1971 nfsi
->cache_validity
&= ~NFS_INO_INVALID_ACCESS
;
1974 void nfs_access_zap_cache(struct inode
*inode
)
1978 if (test_bit(NFS_INO_ACL_LRU_SET
, &NFS_I(inode
)->flags
) == 0)
1980 /* Remove from global LRU init */
1981 spin_lock(&nfs_access_lru_lock
);
1982 if (test_and_clear_bit(NFS_INO_ACL_LRU_SET
, &NFS_I(inode
)->flags
))
1983 list_del_init(&NFS_I(inode
)->access_cache_inode_lru
);
1985 spin_lock(&inode
->i_lock
);
1986 __nfs_access_zap_cache(NFS_I(inode
), &head
);
1987 spin_unlock(&inode
->i_lock
);
1988 spin_unlock(&nfs_access_lru_lock
);
1989 nfs_access_free_list(&head
);
1992 static struct nfs_access_entry
*nfs_access_search_rbtree(struct inode
*inode
, struct rpc_cred
*cred
)
1994 struct rb_node
*n
= NFS_I(inode
)->access_cache
.rb_node
;
1995 struct nfs_access_entry
*entry
;
1998 entry
= rb_entry(n
, struct nfs_access_entry
, rb_node
);
2000 if (cred
< entry
->cred
)
2002 else if (cred
> entry
->cred
)
2010 static int nfs_access_get_cached(struct inode
*inode
, struct rpc_cred
*cred
, struct nfs_access_entry
*res
)
2012 struct nfs_inode
*nfsi
= NFS_I(inode
);
2013 struct nfs_access_entry
*cache
;
2016 spin_lock(&inode
->i_lock
);
2017 if (nfsi
->cache_validity
& NFS_INO_INVALID_ACCESS
)
2019 cache
= nfs_access_search_rbtree(inode
, cred
);
2022 if (!nfs_have_delegated_attributes(inode
) &&
2023 !time_in_range_open(jiffies
, cache
->jiffies
, cache
->jiffies
+ nfsi
->attrtimeo
))
2025 res
->jiffies
= cache
->jiffies
;
2026 res
->cred
= cache
->cred
;
2027 res
->mask
= cache
->mask
;
2028 list_move_tail(&cache
->lru
, &nfsi
->access_cache_entry_lru
);
2031 spin_unlock(&inode
->i_lock
);
2034 rb_erase(&cache
->rb_node
, &nfsi
->access_cache
);
2035 list_del(&cache
->lru
);
2036 spin_unlock(&inode
->i_lock
);
2037 nfs_access_free_entry(cache
);
2040 spin_unlock(&inode
->i_lock
);
2041 nfs_access_zap_cache(inode
);
2045 static void nfs_access_add_rbtree(struct inode
*inode
, struct nfs_access_entry
*set
)
2047 struct nfs_inode
*nfsi
= NFS_I(inode
);
2048 struct rb_root
*root_node
= &nfsi
->access_cache
;
2049 struct rb_node
**p
= &root_node
->rb_node
;
2050 struct rb_node
*parent
= NULL
;
2051 struct nfs_access_entry
*entry
;
2053 spin_lock(&inode
->i_lock
);
2054 while (*p
!= NULL
) {
2056 entry
= rb_entry(parent
, struct nfs_access_entry
, rb_node
);
2058 if (set
->cred
< entry
->cred
)
2059 p
= &parent
->rb_left
;
2060 else if (set
->cred
> entry
->cred
)
2061 p
= &parent
->rb_right
;
2065 rb_link_node(&set
->rb_node
, parent
, p
);
2066 rb_insert_color(&set
->rb_node
, root_node
);
2067 list_add_tail(&set
->lru
, &nfsi
->access_cache_entry_lru
);
2068 spin_unlock(&inode
->i_lock
);
2071 rb_replace_node(parent
, &set
->rb_node
, root_node
);
2072 list_add_tail(&set
->lru
, &nfsi
->access_cache_entry_lru
);
2073 list_del(&entry
->lru
);
2074 spin_unlock(&inode
->i_lock
);
2075 nfs_access_free_entry(entry
);
2078 static void nfs_access_add_cache(struct inode
*inode
, struct nfs_access_entry
*set
)
2080 struct nfs_access_entry
*cache
= kmalloc(sizeof(*cache
), GFP_KERNEL
);
2083 RB_CLEAR_NODE(&cache
->rb_node
);
2084 cache
->jiffies
= set
->jiffies
;
2085 cache
->cred
= get_rpccred(set
->cred
);
2086 cache
->mask
= set
->mask
;
2088 nfs_access_add_rbtree(inode
, cache
);
2090 /* Update accounting */
2091 smp_mb__before_atomic_inc();
2092 atomic_long_inc(&nfs_access_nr_entries
);
2093 smp_mb__after_atomic_inc();
2095 /* Add inode to global LRU list */
2096 if (!test_bit(NFS_INO_ACL_LRU_SET
, &NFS_I(inode
)->flags
)) {
2097 spin_lock(&nfs_access_lru_lock
);
2098 if (!test_and_set_bit(NFS_INO_ACL_LRU_SET
, &NFS_I(inode
)->flags
))
2099 list_add_tail(&NFS_I(inode
)->access_cache_inode_lru
,
2100 &nfs_access_lru_list
);
2101 spin_unlock(&nfs_access_lru_lock
);
2105 static int nfs_do_access(struct inode
*inode
, struct rpc_cred
*cred
, int mask
)
2107 struct nfs_access_entry cache
;
2110 status
= nfs_access_get_cached(inode
, cred
, &cache
);
2114 /* Be clever: ask server to check for all possible rights */
2115 cache
.mask
= MAY_EXEC
| MAY_WRITE
| MAY_READ
;
2117 cache
.jiffies
= jiffies
;
2118 status
= NFS_PROTO(inode
)->access(inode
, &cache
);
2120 if (status
== -ESTALE
) {
2121 nfs_zap_caches(inode
);
2122 if (!S_ISDIR(inode
->i_mode
))
2123 set_bit(NFS_INO_STALE
, &NFS_I(inode
)->flags
);
2127 nfs_access_add_cache(inode
, &cache
);
2129 if ((mask
& ~cache
.mask
& (MAY_READ
| MAY_WRITE
| MAY_EXEC
)) == 0)
2134 static int nfs_open_permission_mask(int openflags
)
2138 if (openflags
& FMODE_READ
)
2140 if (openflags
& FMODE_WRITE
)
2142 if (openflags
& FMODE_EXEC
)
2147 int nfs_may_open(struct inode
*inode
, struct rpc_cred
*cred
, int openflags
)
2149 return nfs_do_access(inode
, cred
, nfs_open_permission_mask(openflags
));
2152 int nfs_permission(struct inode
*inode
, int mask
)
2154 struct rpc_cred
*cred
;
2157 nfs_inc_stats(inode
, NFSIOS_VFSACCESS
);
2159 if ((mask
& (MAY_READ
| MAY_WRITE
| MAY_EXEC
)) == 0)
2161 /* Is this sys_access() ? */
2162 if (mask
& (MAY_ACCESS
| MAY_CHDIR
))
2165 switch (inode
->i_mode
& S_IFMT
) {
2169 /* NFSv4 has atomic_open... */
2170 if (nfs_server_capable(inode
, NFS_CAP_ATOMIC_OPEN
)
2171 && (mask
& MAY_OPEN
)
2172 && !(mask
& MAY_EXEC
))
2177 * Optimize away all write operations, since the server
2178 * will check permissions when we perform the op.
2180 if ((mask
& MAY_WRITE
) && !(mask
& MAY_READ
))
2185 if (!NFS_PROTO(inode
)->access
)
2188 cred
= rpc_lookup_cred();
2189 if (!IS_ERR(cred
)) {
2190 res
= nfs_do_access(inode
, cred
, mask
);
2193 res
= PTR_ERR(cred
);
2195 if (!res
&& (mask
& MAY_EXEC
) && !execute_ok(inode
))
2198 dfprintk(VFS
, "NFS: permission(%s/%ld), mask=0x%x, res=%d\n",
2199 inode
->i_sb
->s_id
, inode
->i_ino
, mask
, res
);
2202 res
= nfs_revalidate_inode(NFS_SERVER(inode
), inode
);
2204 res
= generic_permission(inode
, mask
, NULL
);
2210 * version-control: t
2211 * kept-new-versions: 5