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/kmemleak.h>
37 #include <linux/xattr.h>
39 #include "delegation.h"
44 /* #define NFS_DEBUG_VERBOSE 1 */
46 static int nfs_opendir(struct inode
*, struct file
*);
47 static int nfs_closedir(struct inode
*, struct file
*);
48 static int nfs_readdir(struct file
*, void *, filldir_t
);
49 static struct dentry
*nfs_lookup(struct inode
*, struct dentry
*, struct nameidata
*);
50 static int nfs_create(struct inode
*, struct dentry
*, int, struct nameidata
*);
51 static int nfs_mkdir(struct inode
*, struct dentry
*, int);
52 static int nfs_rmdir(struct inode
*, struct dentry
*);
53 static int nfs_unlink(struct inode
*, struct dentry
*);
54 static int nfs_symlink(struct inode
*, struct dentry
*, const char *);
55 static int nfs_link(struct dentry
*, struct inode
*, struct dentry
*);
56 static int nfs_mknod(struct inode
*, struct dentry
*, int, dev_t
);
57 static int nfs_rename(struct inode
*, struct dentry
*,
58 struct inode
*, struct dentry
*);
59 static int nfs_fsync_dir(struct file
*, int);
60 static loff_t
nfs_llseek_dir(struct file
*, loff_t
, int);
61 static void nfs_readdir_clear_array(struct page
*);
63 const struct file_operations nfs_dir_operations
= {
64 .llseek
= nfs_llseek_dir
,
65 .read
= generic_read_dir
,
66 .readdir
= nfs_readdir
,
68 .release
= nfs_closedir
,
69 .fsync
= nfs_fsync_dir
,
72 const struct inode_operations nfs_dir_inode_operations
= {
77 .symlink
= nfs_symlink
,
82 .permission
= nfs_permission
,
83 .getattr
= nfs_getattr
,
84 .setattr
= nfs_setattr
,
87 const struct address_space_operations nfs_dir_aops
= {
88 .freepage
= nfs_readdir_clear_array
,
92 const struct inode_operations nfs3_dir_inode_operations
= {
97 .symlink
= nfs_symlink
,
101 .rename
= nfs_rename
,
102 .permission
= nfs_permission
,
103 .getattr
= nfs_getattr
,
104 .setattr
= nfs_setattr
,
105 .listxattr
= nfs3_listxattr
,
106 .getxattr
= nfs3_getxattr
,
107 .setxattr
= nfs3_setxattr
,
108 .removexattr
= nfs3_removexattr
,
110 #endif /* CONFIG_NFS_V3 */
114 static struct dentry
*nfs_atomic_lookup(struct inode
*, struct dentry
*, struct nameidata
*);
115 static int nfs_open_create(struct inode
*dir
, struct dentry
*dentry
, int mode
, struct nameidata
*nd
);
116 const struct inode_operations nfs4_dir_inode_operations
= {
117 .create
= nfs_open_create
,
118 .lookup
= nfs_atomic_lookup
,
120 .unlink
= nfs_unlink
,
121 .symlink
= nfs_symlink
,
125 .rename
= nfs_rename
,
126 .permission
= nfs_permission
,
127 .getattr
= nfs_getattr
,
128 .setattr
= nfs_setattr
,
129 .getxattr
= generic_getxattr
,
130 .setxattr
= generic_setxattr
,
131 .listxattr
= generic_listxattr
,
132 .removexattr
= generic_removexattr
,
135 #endif /* CONFIG_NFS_V4 */
137 static struct nfs_open_dir_context
*alloc_nfs_open_dir_context(struct rpc_cred
*cred
)
139 struct nfs_open_dir_context
*ctx
;
140 ctx
= kmalloc(sizeof(*ctx
), GFP_KERNEL
);
145 ctx
->cred
= get_rpccred(cred
);
147 ctx
= ERR_PTR(-ENOMEM
);
151 static void put_nfs_open_dir_context(struct nfs_open_dir_context
*ctx
)
153 put_rpccred(ctx
->cred
);
161 nfs_opendir(struct inode
*inode
, struct file
*filp
)
164 struct nfs_open_dir_context
*ctx
;
165 struct rpc_cred
*cred
;
167 dfprintk(FILE, "NFS: open dir(%s/%s)\n",
168 filp
->f_path
.dentry
->d_parent
->d_name
.name
,
169 filp
->f_path
.dentry
->d_name
.name
);
171 nfs_inc_stats(inode
, NFSIOS_VFSOPEN
);
173 cred
= rpc_lookup_cred();
175 return PTR_ERR(cred
);
176 ctx
= alloc_nfs_open_dir_context(cred
);
181 filp
->private_data
= ctx
;
182 if (filp
->f_path
.dentry
== filp
->f_path
.mnt
->mnt_root
) {
183 /* This is a mountpoint, so d_revalidate will never
184 * have been called, so we need to refresh the
185 * inode (for close-open consistency) ourselves.
187 __nfs_revalidate_inode(NFS_SERVER(inode
), inode
);
195 nfs_closedir(struct inode
*inode
, struct file
*filp
)
197 put_nfs_open_dir_context(filp
->private_data
);
201 struct nfs_cache_array_entry
{
205 unsigned char d_type
;
208 struct nfs_cache_array
{
212 struct nfs_cache_array_entry array
[0];
215 typedef int (*decode_dirent_t
)(struct xdr_stream
*, struct nfs_entry
*, int);
219 unsigned long page_index
;
222 loff_t current_index
;
223 decode_dirent_t decode
;
225 unsigned long timestamp
;
226 unsigned long gencount
;
227 unsigned int cache_entry_index
;
230 } nfs_readdir_descriptor_t
;
233 * The caller is responsible for calling nfs_readdir_release_array(page)
236 struct nfs_cache_array
*nfs_readdir_get_array(struct page
*page
)
240 return ERR_PTR(-EIO
);
243 return ERR_PTR(-ENOMEM
);
248 void nfs_readdir_release_array(struct page
*page
)
254 * we are freeing strings created by nfs_add_to_readdir_array()
257 void nfs_readdir_clear_array(struct page
*page
)
259 struct nfs_cache_array
*array
;
262 array
= kmap_atomic(page
, KM_USER0
);
263 for (i
= 0; i
< array
->size
; i
++)
264 kfree(array
->array
[i
].string
.name
);
265 kunmap_atomic(array
, KM_USER0
);
269 * the caller is responsible for freeing qstr.name
270 * when called by nfs_readdir_add_to_array, the strings will be freed in
271 * nfs_clear_readdir_array()
274 int nfs_readdir_make_qstr(struct qstr
*string
, const char *name
, unsigned int len
)
277 string
->name
= kmemdup(name
, len
, GFP_KERNEL
);
278 if (string
->name
== NULL
)
281 * Avoid a kmemleak false positive. The pointer to the name is stored
282 * in a page cache page which kmemleak does not scan.
284 kmemleak_not_leak(string
->name
);
285 string
->hash
= full_name_hash(name
, len
);
290 int nfs_readdir_add_to_array(struct nfs_entry
*entry
, struct page
*page
)
292 struct nfs_cache_array
*array
= nfs_readdir_get_array(page
);
293 struct nfs_cache_array_entry
*cache_entry
;
297 return PTR_ERR(array
);
299 cache_entry
= &array
->array
[array
->size
];
301 /* Check that this entry lies within the page bounds */
303 if ((char *)&cache_entry
[1] - (char *)page_address(page
) > PAGE_SIZE
)
306 cache_entry
->cookie
= entry
->prev_cookie
;
307 cache_entry
->ino
= entry
->ino
;
308 cache_entry
->d_type
= entry
->d_type
;
309 ret
= nfs_readdir_make_qstr(&cache_entry
->string
, entry
->name
, entry
->len
);
312 array
->last_cookie
= entry
->cookie
;
315 array
->eof_index
= array
->size
;
317 nfs_readdir_release_array(page
);
322 int nfs_readdir_search_for_pos(struct nfs_cache_array
*array
, nfs_readdir_descriptor_t
*desc
)
324 loff_t diff
= desc
->file
->f_pos
- desc
->current_index
;
326 struct nfs_open_dir_context
*ctx
= desc
->file
->private_data
;
330 if (diff
>= array
->size
) {
331 if (array
->eof_index
>= 0)
336 index
= (unsigned int)diff
;
337 *desc
->dir_cookie
= array
->array
[index
].cookie
;
338 desc
->cache_entry_index
= index
;
347 int nfs_readdir_search_for_cookie(struct nfs_cache_array
*array
, nfs_readdir_descriptor_t
*desc
)
351 int status
= -EAGAIN
;
352 struct nfs_open_dir_context
*ctx
= desc
->file
->private_data
;
354 for (i
= 0; i
< array
->size
; i
++) {
355 if (array
->array
[i
].cookie
== *desc
->dir_cookie
) {
356 new_pos
= desc
->current_index
+ i
;
357 if (new_pos
< desc
->file
->f_pos
) {
358 ctx
->dup_cookie
= *desc
->dir_cookie
;
361 desc
->file
->f_pos
= new_pos
;
362 desc
->cache_entry_index
= i
;
366 if (array
->eof_index
>= 0) {
367 status
= -EBADCOOKIE
;
368 if (*desc
->dir_cookie
== array
->last_cookie
)
375 int nfs_readdir_search_array(nfs_readdir_descriptor_t
*desc
)
377 struct nfs_cache_array
*array
;
380 array
= nfs_readdir_get_array(desc
->page
);
382 status
= PTR_ERR(array
);
386 if (*desc
->dir_cookie
== 0)
387 status
= nfs_readdir_search_for_pos(array
, desc
);
389 status
= nfs_readdir_search_for_cookie(array
, desc
);
391 if (status
== -EAGAIN
) {
392 desc
->last_cookie
= array
->last_cookie
;
393 desc
->current_index
+= array
->size
;
396 nfs_readdir_release_array(desc
->page
);
401 /* Fill a page with xdr information before transferring to the cache page */
403 int nfs_readdir_xdr_filler(struct page
**pages
, nfs_readdir_descriptor_t
*desc
,
404 struct nfs_entry
*entry
, struct file
*file
, struct inode
*inode
)
406 struct nfs_open_dir_context
*ctx
= file
->private_data
;
407 struct rpc_cred
*cred
= ctx
->cred
;
408 unsigned long timestamp
, gencount
;
413 gencount
= nfs_inc_attr_generation_counter();
414 error
= NFS_PROTO(inode
)->readdir(file
->f_path
.dentry
, cred
, entry
->cookie
, pages
,
415 NFS_SERVER(inode
)->dtsize
, desc
->plus
);
417 /* We requested READDIRPLUS, but the server doesn't grok it */
418 if (error
== -ENOTSUPP
&& desc
->plus
) {
419 NFS_SERVER(inode
)->caps
&= ~NFS_CAP_READDIRPLUS
;
420 clear_bit(NFS_INO_ADVISE_RDPLUS
, &NFS_I(inode
)->flags
);
426 desc
->timestamp
= timestamp
;
427 desc
->gencount
= gencount
;
432 static int xdr_decode(nfs_readdir_descriptor_t
*desc
,
433 struct nfs_entry
*entry
, struct xdr_stream
*xdr
)
437 error
= desc
->decode(xdr
, entry
, desc
->plus
);
440 entry
->fattr
->time_start
= desc
->timestamp
;
441 entry
->fattr
->gencount
= desc
->gencount
;
446 int nfs_same_file(struct dentry
*dentry
, struct nfs_entry
*entry
)
448 if (dentry
->d_inode
== NULL
)
450 if (nfs_compare_fh(entry
->fh
, NFS_FH(dentry
->d_inode
)) != 0)
458 void nfs_prime_dcache(struct dentry
*parent
, struct nfs_entry
*entry
)
460 struct qstr filename
= {
464 struct dentry
*dentry
;
465 struct dentry
*alias
;
466 struct inode
*dir
= parent
->d_inode
;
469 if (filename
.name
[0] == '.') {
470 if (filename
.len
== 1)
472 if (filename
.len
== 2 && filename
.name
[1] == '.')
475 filename
.hash
= full_name_hash(filename
.name
, filename
.len
);
477 dentry
= d_lookup(parent
, &filename
);
478 if (dentry
!= NULL
) {
479 if (nfs_same_file(dentry
, entry
)) {
480 nfs_refresh_inode(dentry
->d_inode
, entry
->fattr
);
488 dentry
= d_alloc(parent
, &filename
);
492 inode
= nfs_fhget(dentry
->d_sb
, entry
->fh
, entry
->fattr
);
496 alias
= d_materialise_unique(dentry
, inode
);
500 nfs_set_verifier(alias
, nfs_save_change_attribute(dir
));
503 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
509 /* Perform conversion from xdr to cache array */
511 int nfs_readdir_page_filler(nfs_readdir_descriptor_t
*desc
, struct nfs_entry
*entry
,
512 struct page
**xdr_pages
, struct page
*page
, unsigned int buflen
)
514 struct xdr_stream stream
;
516 struct page
*scratch
;
517 struct nfs_cache_array
*array
;
518 unsigned int count
= 0;
521 scratch
= alloc_page(GFP_KERNEL
);
525 xdr_init_decode_pages(&stream
, &buf
, xdr_pages
, buflen
);
526 xdr_set_scratch_buffer(&stream
, page_address(scratch
), PAGE_SIZE
);
529 status
= xdr_decode(desc
, entry
, &stream
);
531 if (status
== -EAGAIN
)
539 nfs_prime_dcache(desc
->file
->f_path
.dentry
, entry
);
541 status
= nfs_readdir_add_to_array(entry
, page
);
544 } while (!entry
->eof
);
546 if (count
== 0 || (status
== -EBADCOOKIE
&& entry
->eof
!= 0)) {
547 array
= nfs_readdir_get_array(page
);
548 if (!IS_ERR(array
)) {
549 array
->eof_index
= array
->size
;
551 nfs_readdir_release_array(page
);
553 status
= PTR_ERR(array
);
561 void nfs_readdir_free_pagearray(struct page
**pages
, unsigned int npages
)
564 for (i
= 0; i
< npages
; i
++)
569 void nfs_readdir_free_large_page(void *ptr
, struct page
**pages
,
572 nfs_readdir_free_pagearray(pages
, npages
);
576 * nfs_readdir_large_page will allocate pages that must be freed with a call
577 * to nfs_readdir_free_large_page
580 int nfs_readdir_large_page(struct page
**pages
, unsigned int npages
)
584 for (i
= 0; i
< npages
; i
++) {
585 struct page
*page
= alloc_page(GFP_KERNEL
);
593 nfs_readdir_free_pagearray(pages
, i
);
598 int nfs_readdir_xdr_to_array(nfs_readdir_descriptor_t
*desc
, struct page
*page
, struct inode
*inode
)
600 struct page
*pages
[NFS_MAX_READDIR_PAGES
];
601 void *pages_ptr
= NULL
;
602 struct nfs_entry entry
;
603 struct file
*file
= desc
->file
;
604 struct nfs_cache_array
*array
;
605 int status
= -ENOMEM
;
606 unsigned int array_size
= ARRAY_SIZE(pages
);
608 entry
.prev_cookie
= 0;
609 entry
.cookie
= desc
->last_cookie
;
611 entry
.fh
= nfs_alloc_fhandle();
612 entry
.fattr
= nfs_alloc_fattr();
613 entry
.server
= NFS_SERVER(inode
);
614 if (entry
.fh
== NULL
|| entry
.fattr
== NULL
)
617 array
= nfs_readdir_get_array(page
);
619 status
= PTR_ERR(array
);
622 memset(array
, 0, sizeof(struct nfs_cache_array
));
623 array
->eof_index
= -1;
625 status
= nfs_readdir_large_page(pages
, array_size
);
627 goto out_release_array
;
630 status
= nfs_readdir_xdr_filler(pages
, desc
, &entry
, file
, inode
);
635 status
= nfs_readdir_page_filler(desc
, &entry
, pages
, page
, pglen
);
637 if (status
== -ENOSPC
)
641 } while (array
->eof_index
< 0);
643 nfs_readdir_free_large_page(pages_ptr
, pages
, array_size
);
645 nfs_readdir_release_array(page
);
647 nfs_free_fattr(entry
.fattr
);
648 nfs_free_fhandle(entry
.fh
);
653 * Now we cache directories properly, by converting xdr information
654 * to an array that can be used for lookups later. This results in
655 * fewer cache pages, since we can store more information on each page.
656 * We only need to convert from xdr once so future lookups are much simpler
659 int nfs_readdir_filler(nfs_readdir_descriptor_t
*desc
, struct page
* page
)
661 struct inode
*inode
= desc
->file
->f_path
.dentry
->d_inode
;
664 ret
= nfs_readdir_xdr_to_array(desc
, page
, inode
);
667 SetPageUptodate(page
);
669 if (invalidate_inode_pages2_range(inode
->i_mapping
, page
->index
+ 1, -1) < 0) {
670 /* Should never happen */
671 nfs_zap_mapping(inode
, inode
->i_mapping
);
681 void cache_page_release(nfs_readdir_descriptor_t
*desc
)
683 if (!desc
->page
->mapping
)
684 nfs_readdir_clear_array(desc
->page
);
685 page_cache_release(desc
->page
);
690 struct page
*get_cache_page(nfs_readdir_descriptor_t
*desc
)
692 return read_cache_page(desc
->file
->f_path
.dentry
->d_inode
->i_mapping
,
693 desc
->page_index
, (filler_t
*)nfs_readdir_filler
, desc
);
697 * Returns 0 if desc->dir_cookie was found on page desc->page_index
700 int find_cache_page(nfs_readdir_descriptor_t
*desc
)
704 desc
->page
= get_cache_page(desc
);
705 if (IS_ERR(desc
->page
))
706 return PTR_ERR(desc
->page
);
708 res
= nfs_readdir_search_array(desc
);
710 cache_page_release(desc
);
714 /* Search for desc->dir_cookie from the beginning of the page cache */
716 int readdir_search_pagecache(nfs_readdir_descriptor_t
*desc
)
720 if (desc
->page_index
== 0) {
721 desc
->current_index
= 0;
722 desc
->last_cookie
= 0;
725 res
= find_cache_page(desc
);
726 } while (res
== -EAGAIN
);
731 * Once we've found the start of the dirent within a page: fill 'er up...
734 int nfs_do_filldir(nfs_readdir_descriptor_t
*desc
, void *dirent
,
737 struct file
*file
= desc
->file
;
740 struct nfs_cache_array
*array
= NULL
;
741 struct nfs_open_dir_context
*ctx
= file
->private_data
;
743 if (ctx
->duped
!= 0 && ctx
->dup_cookie
== *desc
->dir_cookie
) {
744 if (printk_ratelimit()) {
745 pr_notice("NFS: directory %s/%s contains a readdir loop. "
746 "Please contact your server vendor. "
747 "Offending cookie: %llu\n",
748 file
->f_dentry
->d_parent
->d_name
.name
,
749 file
->f_dentry
->d_name
.name
,
756 array
= nfs_readdir_get_array(desc
->page
);
758 res
= PTR_ERR(array
);
762 for (i
= desc
->cache_entry_index
; i
< array
->size
; i
++) {
763 struct nfs_cache_array_entry
*ent
;
765 ent
= &array
->array
[i
];
766 if (filldir(dirent
, ent
->string
.name
, ent
->string
.len
,
767 file
->f_pos
, nfs_compat_user_ino64(ent
->ino
),
773 if (i
< (array
->size
-1))
774 *desc
->dir_cookie
= array
->array
[i
+1].cookie
;
776 *desc
->dir_cookie
= array
->last_cookie
;
778 if (array
->eof_index
>= 0)
781 nfs_readdir_release_array(desc
->page
);
783 cache_page_release(desc
);
784 dfprintk(DIRCACHE
, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n",
785 (unsigned long long)*desc
->dir_cookie
, res
);
790 * If we cannot find a cookie in our cache, we suspect that this is
791 * because it points to a deleted file, so we ask the server to return
792 * whatever it thinks is the next entry. We then feed this to filldir.
793 * If all goes well, we should then be able to find our way round the
794 * cache on the next call to readdir_search_pagecache();
796 * NOTE: we cannot add the anonymous page to the pagecache because
797 * the data it contains might not be page aligned. Besides,
798 * we should already have a complete representation of the
799 * directory in the page cache by the time we get here.
802 int uncached_readdir(nfs_readdir_descriptor_t
*desc
, void *dirent
,
805 struct page
*page
= NULL
;
807 struct inode
*inode
= desc
->file
->f_path
.dentry
->d_inode
;
809 dfprintk(DIRCACHE
, "NFS: uncached_readdir() searching for cookie %Lu\n",
810 (unsigned long long)*desc
->dir_cookie
);
812 page
= alloc_page(GFP_HIGHUSER
);
818 desc
->page_index
= 0;
819 desc
->last_cookie
= *desc
->dir_cookie
;
822 status
= nfs_readdir_xdr_to_array(desc
, page
, inode
);
826 status
= nfs_do_filldir(desc
, dirent
, filldir
);
829 dfprintk(DIRCACHE
, "NFS: %s: returns %d\n",
833 cache_page_release(desc
);
837 /* The file offset position represents the dirent entry number. A
838 last cookie cache takes care of the common case of reading the
841 static int nfs_readdir(struct file
*filp
, void *dirent
, filldir_t filldir
)
843 struct dentry
*dentry
= filp
->f_path
.dentry
;
844 struct inode
*inode
= dentry
->d_inode
;
845 nfs_readdir_descriptor_t my_desc
,
847 struct nfs_open_dir_context
*dir_ctx
= filp
->private_data
;
850 dfprintk(FILE, "NFS: readdir(%s/%s) starting at cookie %llu\n",
851 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
,
852 (long long)filp
->f_pos
);
853 nfs_inc_stats(inode
, NFSIOS_VFSGETDENTS
);
856 * filp->f_pos points to the dirent entry number.
857 * *desc->dir_cookie has the cookie for the next entry. We have
858 * to either find the entry with the appropriate number or
859 * revalidate the cookie.
861 memset(desc
, 0, sizeof(*desc
));
864 desc
->dir_cookie
= &dir_ctx
->dir_cookie
;
865 desc
->decode
= NFS_PROTO(inode
)->decode_dirent
;
866 desc
->plus
= NFS_USE_READDIRPLUS(inode
);
868 nfs_block_sillyrename(dentry
);
869 res
= nfs_revalidate_mapping(inode
, filp
->f_mapping
);
874 res
= readdir_search_pagecache(desc
);
876 if (res
== -EBADCOOKIE
) {
878 /* This means either end of directory */
879 if (*desc
->dir_cookie
&& desc
->eof
== 0) {
880 /* Or that the server has 'lost' a cookie */
881 res
= uncached_readdir(desc
, dirent
, filldir
);
887 if (res
== -ETOOSMALL
&& desc
->plus
) {
888 clear_bit(NFS_INO_ADVISE_RDPLUS
, &NFS_I(inode
)->flags
);
889 nfs_zap_caches(inode
);
890 desc
->page_index
= 0;
898 res
= nfs_do_filldir(desc
, dirent
, filldir
);
901 } while (!desc
->eof
);
903 nfs_unblock_sillyrename(dentry
);
906 dfprintk(FILE, "NFS: readdir(%s/%s) returns %d\n",
907 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
,
912 static loff_t
nfs_llseek_dir(struct file
*filp
, loff_t offset
, int origin
)
914 struct dentry
*dentry
= filp
->f_path
.dentry
;
915 struct inode
*inode
= dentry
->d_inode
;
916 struct nfs_open_dir_context
*dir_ctx
= filp
->private_data
;
918 dfprintk(FILE, "NFS: llseek dir(%s/%s, %lld, %d)\n",
919 dentry
->d_parent
->d_name
.name
,
923 mutex_lock(&inode
->i_mutex
);
926 offset
+= filp
->f_pos
;
934 if (offset
!= filp
->f_pos
) {
935 filp
->f_pos
= offset
;
936 dir_ctx
->dir_cookie
= 0;
940 mutex_unlock(&inode
->i_mutex
);
945 * All directory operations under NFS are synchronous, so fsync()
946 * is a dummy operation.
948 static int nfs_fsync_dir(struct file
*filp
, int datasync
)
950 struct dentry
*dentry
= filp
->f_path
.dentry
;
952 dfprintk(FILE, "NFS: fsync dir(%s/%s) datasync %d\n",
953 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
,
956 nfs_inc_stats(dentry
->d_inode
, NFSIOS_VFSFSYNC
);
961 * nfs_force_lookup_revalidate - Mark the directory as having changed
962 * @dir - pointer to directory inode
964 * This forces the revalidation code in nfs_lookup_revalidate() to do a
965 * full lookup on all child dentries of 'dir' whenever a change occurs
966 * on the server that might have invalidated our dcache.
968 * The caller should be holding dir->i_lock
970 void nfs_force_lookup_revalidate(struct inode
*dir
)
972 NFS_I(dir
)->cache_change_attribute
++;
976 * A check for whether or not the parent directory has changed.
977 * In the case it has, we assume that the dentries are untrustworthy
978 * and may need to be looked up again.
980 static int nfs_check_verifier(struct inode
*dir
, struct dentry
*dentry
)
984 if (NFS_SERVER(dir
)->flags
& NFS_MOUNT_LOOKUP_CACHE_NONE
)
986 if (!nfs_verify_change_attribute(dir
, dentry
->d_time
))
988 /* Revalidate nfsi->cache_change_attribute before we declare a match */
989 if (nfs_revalidate_inode(NFS_SERVER(dir
), dir
) < 0)
991 if (!nfs_verify_change_attribute(dir
, dentry
->d_time
))
997 * Return the intent data that applies to this particular path component
999 * Note that the current set of intents only apply to the very last
1000 * component of the path.
1001 * We check for this using LOOKUP_CONTINUE and LOOKUP_PARENT.
1003 static inline unsigned int nfs_lookup_check_intent(struct nameidata
*nd
,
1006 if (nd
->flags
& (LOOKUP_CONTINUE
|LOOKUP_PARENT
))
1008 return nd
->flags
& mask
;
1012 * Use intent information to check whether or not we're going to do
1013 * an O_EXCL create using this path component.
1015 static int nfs_is_exclusive_create(struct inode
*dir
, struct nameidata
*nd
)
1017 if (NFS_PROTO(dir
)->version
== 2)
1019 return nd
&& nfs_lookup_check_intent(nd
, LOOKUP_EXCL
);
1023 * Inode and filehandle revalidation for lookups.
1025 * We force revalidation in the cases where the VFS sets LOOKUP_REVAL,
1026 * or if the intent information indicates that we're about to open this
1027 * particular file and the "nocto" mount flag is not set.
1031 int nfs_lookup_verify_inode(struct inode
*inode
, struct nameidata
*nd
)
1033 struct nfs_server
*server
= NFS_SERVER(inode
);
1035 if (IS_AUTOMOUNT(inode
))
1038 /* VFS wants an on-the-wire revalidation */
1039 if (nd
->flags
& LOOKUP_REVAL
)
1041 /* This is an open(2) */
1042 if (nfs_lookup_check_intent(nd
, LOOKUP_OPEN
) != 0 &&
1043 !(server
->flags
& NFS_MOUNT_NOCTO
) &&
1044 (S_ISREG(inode
->i_mode
) ||
1045 S_ISDIR(inode
->i_mode
)))
1049 return nfs_revalidate_inode(server
, inode
);
1051 return __nfs_revalidate_inode(server
, inode
);
1055 * We judge how long we want to trust negative
1056 * dentries by looking at the parent inode mtime.
1058 * If parent mtime has changed, we revalidate, else we wait for a
1059 * period corresponding to the parent's attribute cache timeout value.
1062 int nfs_neg_need_reval(struct inode
*dir
, struct dentry
*dentry
,
1063 struct nameidata
*nd
)
1065 /* Don't revalidate a negative dentry if we're creating a new file */
1066 if (nd
!= NULL
&& nfs_lookup_check_intent(nd
, LOOKUP_CREATE
) != 0)
1068 if (NFS_SERVER(dir
)->flags
& NFS_MOUNT_LOOKUP_CACHE_NONEG
)
1070 return !nfs_check_verifier(dir
, dentry
);
1074 * This is called every time the dcache has a lookup hit,
1075 * and we should check whether we can really trust that
1078 * NOTE! The hit can be a negative hit too, don't assume
1081 * If the parent directory is seen to have changed, we throw out the
1082 * cached dentry and do a new lookup.
1084 static int nfs_lookup_revalidate(struct dentry
*dentry
, struct nameidata
*nd
)
1087 struct inode
*inode
;
1088 struct dentry
*parent
;
1089 struct nfs_fh
*fhandle
= NULL
;
1090 struct nfs_fattr
*fattr
= NULL
;
1093 if (nd
->flags
& LOOKUP_RCU
)
1096 parent
= dget_parent(dentry
);
1097 dir
= parent
->d_inode
;
1098 nfs_inc_stats(dir
, NFSIOS_DENTRYREVALIDATE
);
1099 inode
= dentry
->d_inode
;
1102 if (nfs_neg_need_reval(dir
, dentry
, nd
))
1107 if (is_bad_inode(inode
)) {
1108 dfprintk(LOOKUPCACHE
, "%s: %s/%s has dud inode\n",
1109 __func__
, dentry
->d_parent
->d_name
.name
,
1110 dentry
->d_name
.name
);
1114 if (nfs_have_delegation(inode
, FMODE_READ
))
1115 goto out_set_verifier
;
1117 /* Force a full look up iff the parent directory has changed */
1118 if (!nfs_is_exclusive_create(dir
, nd
) && nfs_check_verifier(dir
, dentry
)) {
1119 if (nfs_lookup_verify_inode(inode
, nd
))
1120 goto out_zap_parent
;
1124 if (NFS_STALE(inode
))
1128 fhandle
= nfs_alloc_fhandle();
1129 fattr
= nfs_alloc_fattr();
1130 if (fhandle
== NULL
|| fattr
== NULL
)
1133 error
= NFS_PROTO(dir
)->lookup(NFS_SERVER(dir
)->client
, dir
, &dentry
->d_name
, fhandle
, fattr
);
1136 if (nfs_compare_fh(NFS_FH(inode
), fhandle
))
1138 if ((error
= nfs_refresh_inode(inode
, fattr
)) != 0)
1141 nfs_free_fattr(fattr
);
1142 nfs_free_fhandle(fhandle
);
1144 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1147 dfprintk(LOOKUPCACHE
, "NFS: %s(%s/%s) is valid\n",
1148 __func__
, dentry
->d_parent
->d_name
.name
,
1149 dentry
->d_name
.name
);
1152 nfs_zap_caches(dir
);
1154 nfs_mark_for_revalidate(dir
);
1155 if (inode
&& S_ISDIR(inode
->i_mode
)) {
1156 /* Purge readdir caches. */
1157 nfs_zap_caches(inode
);
1158 /* If we have submounts, don't unhash ! */
1159 if (have_submounts(dentry
))
1161 if (dentry
->d_flags
& DCACHE_DISCONNECTED
)
1163 shrink_dcache_parent(dentry
);
1166 nfs_free_fattr(fattr
);
1167 nfs_free_fhandle(fhandle
);
1169 dfprintk(LOOKUPCACHE
, "NFS: %s(%s/%s) is invalid\n",
1170 __func__
, dentry
->d_parent
->d_name
.name
,
1171 dentry
->d_name
.name
);
1174 nfs_free_fattr(fattr
);
1175 nfs_free_fhandle(fhandle
);
1177 dfprintk(LOOKUPCACHE
, "NFS: %s(%s/%s) lookup returned error %d\n",
1178 __func__
, dentry
->d_parent
->d_name
.name
,
1179 dentry
->d_name
.name
, error
);
1184 * This is called from dput() when d_count is going to 0.
1186 static int nfs_dentry_delete(const struct dentry
*dentry
)
1188 dfprintk(VFS
, "NFS: dentry_delete(%s/%s, %x)\n",
1189 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
,
1192 /* Unhash any dentry with a stale inode */
1193 if (dentry
->d_inode
!= NULL
&& NFS_STALE(dentry
->d_inode
))
1196 if (dentry
->d_flags
& DCACHE_NFSFS_RENAMED
) {
1197 /* Unhash it, so that ->d_iput() would be called */
1200 if (!(dentry
->d_sb
->s_flags
& MS_ACTIVE
)) {
1201 /* Unhash it, so that ancestors of killed async unlink
1202 * files will be cleaned up during umount */
1209 static void nfs_drop_nlink(struct inode
*inode
)
1211 spin_lock(&inode
->i_lock
);
1212 if (inode
->i_nlink
> 0)
1214 spin_unlock(&inode
->i_lock
);
1218 * Called when the dentry loses inode.
1219 * We use it to clean up silly-renamed files.
1221 static void nfs_dentry_iput(struct dentry
*dentry
, struct inode
*inode
)
1223 if (S_ISDIR(inode
->i_mode
))
1224 /* drop any readdir cache as it could easily be old */
1225 NFS_I(inode
)->cache_validity
|= NFS_INO_INVALID_DATA
;
1227 if (dentry
->d_flags
& DCACHE_NFSFS_RENAMED
) {
1229 nfs_complete_unlink(dentry
, inode
);
1234 static void nfs_d_release(struct dentry
*dentry
)
1236 /* free cached devname value, if it survived that far */
1237 if (unlikely(dentry
->d_fsdata
)) {
1238 if (dentry
->d_flags
& DCACHE_NFSFS_RENAMED
)
1241 kfree(dentry
->d_fsdata
);
1245 const struct dentry_operations nfs_dentry_operations
= {
1246 .d_revalidate
= nfs_lookup_revalidate
,
1247 .d_delete
= nfs_dentry_delete
,
1248 .d_iput
= nfs_dentry_iput
,
1249 .d_automount
= nfs_d_automount
,
1250 .d_release
= nfs_d_release
,
1253 static struct dentry
*nfs_lookup(struct inode
*dir
, struct dentry
* dentry
, struct nameidata
*nd
)
1256 struct dentry
*parent
;
1257 struct inode
*inode
= NULL
;
1258 struct nfs_fh
*fhandle
= NULL
;
1259 struct nfs_fattr
*fattr
= NULL
;
1262 dfprintk(VFS
, "NFS: lookup(%s/%s)\n",
1263 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
);
1264 nfs_inc_stats(dir
, NFSIOS_VFSLOOKUP
);
1266 res
= ERR_PTR(-ENAMETOOLONG
);
1267 if (dentry
->d_name
.len
> NFS_SERVER(dir
)->namelen
)
1271 * If we're doing an exclusive create, optimize away the lookup
1272 * but don't hash the dentry.
1274 if (nfs_is_exclusive_create(dir
, nd
)) {
1275 d_instantiate(dentry
, NULL
);
1280 res
= ERR_PTR(-ENOMEM
);
1281 fhandle
= nfs_alloc_fhandle();
1282 fattr
= nfs_alloc_fattr();
1283 if (fhandle
== NULL
|| fattr
== NULL
)
1286 parent
= dentry
->d_parent
;
1287 /* Protect against concurrent sillydeletes */
1288 nfs_block_sillyrename(parent
);
1289 error
= NFS_PROTO(dir
)->lookup(NFS_SERVER(dir
)->client
, dir
, &dentry
->d_name
, fhandle
, fattr
);
1290 if (error
== -ENOENT
)
1293 res
= ERR_PTR(error
);
1294 goto out_unblock_sillyrename
;
1296 inode
= nfs_fhget(dentry
->d_sb
, fhandle
, fattr
);
1297 res
= ERR_CAST(inode
);
1299 goto out_unblock_sillyrename
;
1302 res
= d_materialise_unique(dentry
, inode
);
1305 goto out_unblock_sillyrename
;
1308 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1309 out_unblock_sillyrename
:
1310 nfs_unblock_sillyrename(parent
);
1312 nfs_free_fattr(fattr
);
1313 nfs_free_fhandle(fhandle
);
1317 #ifdef CONFIG_NFS_V4
1318 static int nfs_open_revalidate(struct dentry
*, struct nameidata
*);
1320 const struct dentry_operations nfs4_dentry_operations
= {
1321 .d_revalidate
= nfs_open_revalidate
,
1322 .d_delete
= nfs_dentry_delete
,
1323 .d_iput
= nfs_dentry_iput
,
1324 .d_automount
= nfs_d_automount
,
1325 .d_release
= nfs_d_release
,
1329 * Use intent information to determine whether we need to substitute
1330 * the NFSv4-style stateful OPEN for the LOOKUP call
1332 static int is_atomic_open(struct nameidata
*nd
)
1334 if (nd
== NULL
|| nfs_lookup_check_intent(nd
, LOOKUP_OPEN
) == 0)
1336 /* NFS does not (yet) have a stateful open for directories */
1337 if (nd
->flags
& LOOKUP_DIRECTORY
)
1339 /* Are we trying to write to a read only partition? */
1340 if (__mnt_is_readonly(nd
->path
.mnt
) &&
1341 (nd
->intent
.open
.flags
& (O_CREAT
|O_TRUNC
|O_ACCMODE
)))
1346 static fmode_t
flags_to_mode(int flags
)
1348 fmode_t res
= (__force fmode_t
)flags
& FMODE_EXEC
;
1349 if ((flags
& O_ACCMODE
) != O_WRONLY
)
1351 if ((flags
& O_ACCMODE
) != O_RDONLY
)
1356 static struct nfs_open_context
*create_nfs_open_context(struct dentry
*dentry
, int open_flags
)
1358 struct nfs_open_context
*ctx
;
1359 struct rpc_cred
*cred
;
1360 fmode_t fmode
= flags_to_mode(open_flags
);
1362 cred
= rpc_lookup_cred();
1364 return ERR_CAST(cred
);
1365 ctx
= alloc_nfs_open_context(dentry
, cred
, fmode
);
1368 return ERR_PTR(-ENOMEM
);
1372 static int do_open(struct inode
*inode
, struct file
*filp
)
1374 nfs_fscache_set_inode_cookie(inode
, filp
);
1378 static int nfs_intent_set_file(struct nameidata
*nd
, struct nfs_open_context
*ctx
)
1383 /* If the open_intent is for execute, we have an extra check to make */
1384 if (ctx
->mode
& FMODE_EXEC
) {
1385 ret
= nfs_may_open(ctx
->dentry
->d_inode
,
1387 nd
->intent
.open
.flags
);
1391 filp
= lookup_instantiate_filp(nd
, ctx
->dentry
, do_open
);
1393 ret
= PTR_ERR(filp
);
1395 nfs_file_set_open_context(filp
, ctx
);
1397 put_nfs_open_context(ctx
);
1401 static struct dentry
*nfs_atomic_lookup(struct inode
*dir
, struct dentry
*dentry
, struct nameidata
*nd
)
1403 struct nfs_open_context
*ctx
;
1405 struct dentry
*res
= NULL
;
1406 struct inode
*inode
;
1410 dfprintk(VFS
, "NFS: atomic_lookup(%s/%ld), %s\n",
1411 dir
->i_sb
->s_id
, dir
->i_ino
, dentry
->d_name
.name
);
1413 /* Check that we are indeed trying to open this file */
1414 if (!is_atomic_open(nd
))
1417 if (dentry
->d_name
.len
> NFS_SERVER(dir
)->namelen
) {
1418 res
= ERR_PTR(-ENAMETOOLONG
);
1422 /* Let vfs_create() deal with O_EXCL. Instantiate, but don't hash
1424 if (nd
->flags
& LOOKUP_EXCL
) {
1425 d_instantiate(dentry
, NULL
);
1429 open_flags
= nd
->intent
.open
.flags
;
1431 ctx
= create_nfs_open_context(dentry
, open_flags
);
1432 res
= ERR_CAST(ctx
);
1436 if (nd
->flags
& LOOKUP_CREATE
) {
1437 attr
.ia_mode
= nd
->intent
.open
.create_mode
;
1438 attr
.ia_valid
= ATTR_MODE
;
1439 attr
.ia_mode
&= ~current_umask();
1441 open_flags
&= ~(O_EXCL
| O_CREAT
);
1445 /* Open the file on the server */
1446 nfs_block_sillyrename(dentry
->d_parent
);
1447 inode
= NFS_PROTO(dir
)->open_context(dir
, ctx
, open_flags
, &attr
);
1448 if (IS_ERR(inode
)) {
1449 nfs_unblock_sillyrename(dentry
->d_parent
);
1450 put_nfs_open_context(ctx
);
1451 switch (PTR_ERR(inode
)) {
1452 /* Make a negative dentry */
1454 d_add(dentry
, NULL
);
1457 /* This turned out not to be a regular file */
1461 if (!(nd
->intent
.open
.flags
& O_NOFOLLOW
))
1466 res
= ERR_CAST(inode
);
1470 res
= d_add_unique(dentry
, inode
);
1471 nfs_unblock_sillyrename(dentry
->d_parent
);
1474 ctx
->dentry
= dget(res
);
1477 err
= nfs_intent_set_file(nd
, ctx
);
1481 return ERR_PTR(err
);
1484 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1487 return nfs_lookup(dir
, dentry
, nd
);
1490 static int nfs_open_revalidate(struct dentry
*dentry
, struct nameidata
*nd
)
1492 struct dentry
*parent
= NULL
;
1493 struct inode
*inode
;
1495 struct nfs_open_context
*ctx
;
1496 int openflags
, ret
= 0;
1498 if (nd
->flags
& LOOKUP_RCU
)
1501 inode
= dentry
->d_inode
;
1502 if (!is_atomic_open(nd
) || d_mountpoint(dentry
))
1505 parent
= dget_parent(dentry
);
1506 dir
= parent
->d_inode
;
1508 /* We can't create new files in nfs_open_revalidate(), so we
1509 * optimize away revalidation of negative dentries.
1511 if (inode
== NULL
) {
1512 if (!nfs_neg_need_reval(dir
, dentry
, nd
))
1517 /* NFS only supports OPEN on regular files */
1518 if (!S_ISREG(inode
->i_mode
))
1520 openflags
= nd
->intent
.open
.flags
;
1521 /* We cannot do exclusive creation on a positive dentry */
1522 if ((openflags
& (O_CREAT
|O_EXCL
)) == (O_CREAT
|O_EXCL
))
1524 /* We can't create new files, or truncate existing ones here */
1525 openflags
&= ~(O_CREAT
|O_EXCL
|O_TRUNC
);
1527 ctx
= create_nfs_open_context(dentry
, openflags
);
1532 * Note: we're not holding inode->i_mutex and so may be racing with
1533 * operations that change the directory. We therefore save the
1534 * change attribute *before* we do the RPC call.
1536 inode
= NFS_PROTO(dir
)->open_context(dir
, ctx
, openflags
, NULL
);
1537 if (IS_ERR(inode
)) {
1538 ret
= PTR_ERR(inode
);
1551 if (inode
!= dentry
->d_inode
)
1554 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1555 ret
= nfs_intent_set_file(nd
, ctx
);
1565 put_nfs_open_context(ctx
);
1571 return nfs_lookup_revalidate(dentry
, nd
);
1574 static int nfs_open_create(struct inode
*dir
, struct dentry
*dentry
, int mode
,
1575 struct nameidata
*nd
)
1577 struct nfs_open_context
*ctx
= NULL
;
1580 int open_flags
= O_CREAT
|O_EXCL
;
1582 dfprintk(VFS
, "NFS: create(%s/%ld), %s\n",
1583 dir
->i_sb
->s_id
, dir
->i_ino
, dentry
->d_name
.name
);
1585 attr
.ia_mode
= mode
;
1586 attr
.ia_valid
= ATTR_MODE
;
1588 if (nd
&& (nd
->flags
& LOOKUP_OPEN
) != 0)
1589 open_flags
= nd
->intent
.open
.flags
;
1591 ctx
= create_nfs_open_context(dentry
, open_flags
);
1592 error
= PTR_ERR(ctx
);
1596 error
= NFS_PROTO(dir
)->create(dir
, dentry
, &attr
, open_flags
, ctx
);
1599 if (nd
&& (nd
->flags
& LOOKUP_OPEN
) != 0) {
1600 error
= nfs_intent_set_file(nd
, ctx
);
1604 put_nfs_open_context(ctx
);
1608 put_nfs_open_context(ctx
);
1615 #endif /* CONFIG_NFSV4 */
1618 * Code common to create, mkdir, and mknod.
1620 int nfs_instantiate(struct dentry
*dentry
, struct nfs_fh
*fhandle
,
1621 struct nfs_fattr
*fattr
)
1623 struct dentry
*parent
= dget_parent(dentry
);
1624 struct inode
*dir
= parent
->d_inode
;
1625 struct inode
*inode
;
1626 int error
= -EACCES
;
1630 /* We may have been initialized further down */
1631 if (dentry
->d_inode
)
1633 if (fhandle
->size
== 0) {
1634 error
= NFS_PROTO(dir
)->lookup(NFS_SERVER(dir
)->client
, dir
, &dentry
->d_name
, fhandle
, fattr
);
1638 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1639 if (!(fattr
->valid
& NFS_ATTR_FATTR
)) {
1640 struct nfs_server
*server
= NFS_SB(dentry
->d_sb
);
1641 error
= server
->nfs_client
->rpc_ops
->getattr(server
, fhandle
, fattr
);
1645 inode
= nfs_fhget(dentry
->d_sb
, fhandle
, fattr
);
1646 error
= PTR_ERR(inode
);
1649 d_add(dentry
, inode
);
1654 nfs_mark_for_revalidate(dir
);
1660 * Following a failed create operation, we drop the dentry rather
1661 * than retain a negative dentry. This avoids a problem in the event
1662 * that the operation succeeded on the server, but an error in the
1663 * reply path made it appear to have failed.
1665 static int nfs_create(struct inode
*dir
, struct dentry
*dentry
, int mode
,
1666 struct nameidata
*nd
)
1670 int open_flags
= O_CREAT
|O_EXCL
;
1672 dfprintk(VFS
, "NFS: create(%s/%ld), %s\n",
1673 dir
->i_sb
->s_id
, dir
->i_ino
, dentry
->d_name
.name
);
1675 attr
.ia_mode
= mode
;
1676 attr
.ia_valid
= ATTR_MODE
;
1678 if (nd
&& (nd
->flags
& LOOKUP_OPEN
) != 0)
1679 open_flags
= nd
->intent
.open
.flags
;
1681 error
= NFS_PROTO(dir
)->create(dir
, dentry
, &attr
, open_flags
, NULL
);
1691 * See comments for nfs_proc_create regarding failed operations.
1694 nfs_mknod(struct inode
*dir
, struct dentry
*dentry
, int mode
, dev_t rdev
)
1699 dfprintk(VFS
, "NFS: mknod(%s/%ld), %s\n",
1700 dir
->i_sb
->s_id
, dir
->i_ino
, dentry
->d_name
.name
);
1702 if (!new_valid_dev(rdev
))
1705 attr
.ia_mode
= mode
;
1706 attr
.ia_valid
= ATTR_MODE
;
1708 status
= NFS_PROTO(dir
)->mknod(dir
, dentry
, &attr
, rdev
);
1718 * See comments for nfs_proc_create regarding failed operations.
1720 static int nfs_mkdir(struct inode
*dir
, struct dentry
*dentry
, int mode
)
1725 dfprintk(VFS
, "NFS: mkdir(%s/%ld), %s\n",
1726 dir
->i_sb
->s_id
, dir
->i_ino
, dentry
->d_name
.name
);
1728 attr
.ia_valid
= ATTR_MODE
;
1729 attr
.ia_mode
= mode
| S_IFDIR
;
1731 error
= NFS_PROTO(dir
)->mkdir(dir
, dentry
, &attr
);
1740 static void nfs_dentry_handle_enoent(struct dentry
*dentry
)
1742 if (dentry
->d_inode
!= NULL
&& !d_unhashed(dentry
))
1746 static int nfs_rmdir(struct inode
*dir
, struct dentry
*dentry
)
1750 dfprintk(VFS
, "NFS: rmdir(%s/%ld), %s\n",
1751 dir
->i_sb
->s_id
, dir
->i_ino
, dentry
->d_name
.name
);
1753 error
= NFS_PROTO(dir
)->rmdir(dir
, &dentry
->d_name
);
1754 /* Ensure the VFS deletes this inode */
1755 if (error
== 0 && dentry
->d_inode
!= NULL
)
1756 clear_nlink(dentry
->d_inode
);
1757 else if (error
== -ENOENT
)
1758 nfs_dentry_handle_enoent(dentry
);
1764 * Remove a file after making sure there are no pending writes,
1765 * and after checking that the file has only one user.
1767 * We invalidate the attribute cache and free the inode prior to the operation
1768 * to avoid possible races if the server reuses the inode.
1770 static int nfs_safe_remove(struct dentry
*dentry
)
1772 struct inode
*dir
= dentry
->d_parent
->d_inode
;
1773 struct inode
*inode
= dentry
->d_inode
;
1776 dfprintk(VFS
, "NFS: safe_remove(%s/%s)\n",
1777 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
);
1779 /* If the dentry was sillyrenamed, we simply call d_delete() */
1780 if (dentry
->d_flags
& DCACHE_NFSFS_RENAMED
) {
1785 if (inode
!= NULL
) {
1786 nfs_inode_return_delegation(inode
);
1787 error
= NFS_PROTO(dir
)->remove(dir
, &dentry
->d_name
);
1788 /* The VFS may want to delete this inode */
1790 nfs_drop_nlink(inode
);
1791 nfs_mark_for_revalidate(inode
);
1793 error
= NFS_PROTO(dir
)->remove(dir
, &dentry
->d_name
);
1794 if (error
== -ENOENT
)
1795 nfs_dentry_handle_enoent(dentry
);
1800 /* We do silly rename. In case sillyrename() returns -EBUSY, the inode
1801 * belongs to an active ".nfs..." file and we return -EBUSY.
1803 * If sillyrename() returns 0, we do nothing, otherwise we unlink.
1805 static int nfs_unlink(struct inode
*dir
, struct dentry
*dentry
)
1808 int need_rehash
= 0;
1810 dfprintk(VFS
, "NFS: unlink(%s/%ld, %s)\n", dir
->i_sb
->s_id
,
1811 dir
->i_ino
, dentry
->d_name
.name
);
1813 spin_lock(&dentry
->d_lock
);
1814 if (dentry
->d_count
> 1) {
1815 spin_unlock(&dentry
->d_lock
);
1816 /* Start asynchronous writeout of the inode */
1817 write_inode_now(dentry
->d_inode
, 0);
1818 error
= nfs_sillyrename(dir
, dentry
);
1821 if (!d_unhashed(dentry
)) {
1825 spin_unlock(&dentry
->d_lock
);
1826 error
= nfs_safe_remove(dentry
);
1827 if (!error
|| error
== -ENOENT
) {
1828 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1829 } else if (need_rehash
)
1835 * To create a symbolic link, most file systems instantiate a new inode,
1836 * add a page to it containing the path, then write it out to the disk
1837 * using prepare_write/commit_write.
1839 * Unfortunately the NFS client can't create the in-core inode first
1840 * because it needs a file handle to create an in-core inode (see
1841 * fs/nfs/inode.c:nfs_fhget). We only have a file handle *after* the
1842 * symlink request has completed on the server.
1844 * So instead we allocate a raw page, copy the symname into it, then do
1845 * the SYMLINK request with the page as the buffer. If it succeeds, we
1846 * now have a new file handle and can instantiate an in-core NFS inode
1847 * and move the raw page into its mapping.
1849 static int nfs_symlink(struct inode
*dir
, struct dentry
*dentry
, const char *symname
)
1851 struct pagevec lru_pvec
;
1855 unsigned int pathlen
= strlen(symname
);
1858 dfprintk(VFS
, "NFS: symlink(%s/%ld, %s, %s)\n", dir
->i_sb
->s_id
,
1859 dir
->i_ino
, dentry
->d_name
.name
, symname
);
1861 if (pathlen
> PAGE_SIZE
)
1862 return -ENAMETOOLONG
;
1864 attr
.ia_mode
= S_IFLNK
| S_IRWXUGO
;
1865 attr
.ia_valid
= ATTR_MODE
;
1867 page
= alloc_page(GFP_HIGHUSER
);
1871 kaddr
= kmap_atomic(page
, KM_USER0
);
1872 memcpy(kaddr
, symname
, pathlen
);
1873 if (pathlen
< PAGE_SIZE
)
1874 memset(kaddr
+ pathlen
, 0, PAGE_SIZE
- pathlen
);
1875 kunmap_atomic(kaddr
, KM_USER0
);
1877 error
= NFS_PROTO(dir
)->symlink(dir
, dentry
, page
, pathlen
, &attr
);
1879 dfprintk(VFS
, "NFS: symlink(%s/%ld, %s, %s) error %d\n",
1880 dir
->i_sb
->s_id
, dir
->i_ino
,
1881 dentry
->d_name
.name
, symname
, error
);
1888 * No big deal if we can't add this page to the page cache here.
1889 * READLINK will get the missing page from the server if needed.
1891 pagevec_init(&lru_pvec
, 0);
1892 if (!add_to_page_cache(page
, dentry
->d_inode
->i_mapping
, 0,
1894 pagevec_add(&lru_pvec
, page
);
1895 pagevec_lru_add_file(&lru_pvec
);
1896 SetPageUptodate(page
);
1905 nfs_link(struct dentry
*old_dentry
, struct inode
*dir
, struct dentry
*dentry
)
1907 struct inode
*inode
= old_dentry
->d_inode
;
1910 dfprintk(VFS
, "NFS: link(%s/%s -> %s/%s)\n",
1911 old_dentry
->d_parent
->d_name
.name
, old_dentry
->d_name
.name
,
1912 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
);
1914 nfs_inode_return_delegation(inode
);
1917 error
= NFS_PROTO(dir
)->link(inode
, dir
, &dentry
->d_name
);
1920 d_add(dentry
, inode
);
1927 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a
1928 * different file handle for the same inode after a rename (e.g. when
1929 * moving to a different directory). A fail-safe method to do so would
1930 * be to look up old_dir/old_name, create a link to new_dir/new_name and
1931 * rename the old file using the sillyrename stuff. This way, the original
1932 * file in old_dir will go away when the last process iput()s the inode.
1936 * It actually works quite well. One needs to have the possibility for
1937 * at least one ".nfs..." file in each directory the file ever gets
1938 * moved or linked to which happens automagically with the new
1939 * implementation that only depends on the dcache stuff instead of
1940 * using the inode layer
1942 * Unfortunately, things are a little more complicated than indicated
1943 * above. For a cross-directory move, we want to make sure we can get
1944 * rid of the old inode after the operation. This means there must be
1945 * no pending writes (if it's a file), and the use count must be 1.
1946 * If these conditions are met, we can drop the dentries before doing
1949 static int nfs_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
1950 struct inode
*new_dir
, struct dentry
*new_dentry
)
1952 struct inode
*old_inode
= old_dentry
->d_inode
;
1953 struct inode
*new_inode
= new_dentry
->d_inode
;
1954 struct dentry
*dentry
= NULL
, *rehash
= NULL
;
1957 dfprintk(VFS
, "NFS: rename(%s/%s -> %s/%s, ct=%d)\n",
1958 old_dentry
->d_parent
->d_name
.name
, old_dentry
->d_name
.name
,
1959 new_dentry
->d_parent
->d_name
.name
, new_dentry
->d_name
.name
,
1960 new_dentry
->d_count
);
1963 * For non-directories, check whether the target is busy and if so,
1964 * make a copy of the dentry and then do a silly-rename. If the
1965 * silly-rename succeeds, the copied dentry is hashed and becomes
1968 if (new_inode
&& !S_ISDIR(new_inode
->i_mode
)) {
1970 * To prevent any new references to the target during the
1971 * rename, we unhash the dentry in advance.
1973 if (!d_unhashed(new_dentry
)) {
1975 rehash
= new_dentry
;
1978 if (new_dentry
->d_count
> 2) {
1981 /* copy the target dentry's name */
1982 dentry
= d_alloc(new_dentry
->d_parent
,
1983 &new_dentry
->d_name
);
1987 /* silly-rename the existing target ... */
1988 err
= nfs_sillyrename(new_dir
, new_dentry
);
1992 new_dentry
= dentry
;
1998 nfs_inode_return_delegation(old_inode
);
1999 if (new_inode
!= NULL
)
2000 nfs_inode_return_delegation(new_inode
);
2002 error
= NFS_PROTO(old_dir
)->rename(old_dir
, &old_dentry
->d_name
,
2003 new_dir
, &new_dentry
->d_name
);
2004 nfs_mark_for_revalidate(old_inode
);
2009 if (new_inode
!= NULL
)
2010 nfs_drop_nlink(new_inode
);
2011 d_move(old_dentry
, new_dentry
);
2012 nfs_set_verifier(new_dentry
,
2013 nfs_save_change_attribute(new_dir
));
2014 } else if (error
== -ENOENT
)
2015 nfs_dentry_handle_enoent(old_dentry
);
2017 /* new dentry created? */
2023 static DEFINE_SPINLOCK(nfs_access_lru_lock
);
2024 static LIST_HEAD(nfs_access_lru_list
);
2025 static atomic_long_t nfs_access_nr_entries
;
2027 static void nfs_access_free_entry(struct nfs_access_entry
*entry
)
2029 put_rpccred(entry
->cred
);
2031 smp_mb__before_atomic_dec();
2032 atomic_long_dec(&nfs_access_nr_entries
);
2033 smp_mb__after_atomic_dec();
2036 static void nfs_access_free_list(struct list_head
*head
)
2038 struct nfs_access_entry
*cache
;
2040 while (!list_empty(head
)) {
2041 cache
= list_entry(head
->next
, struct nfs_access_entry
, lru
);
2042 list_del(&cache
->lru
);
2043 nfs_access_free_entry(cache
);
2047 int nfs_access_cache_shrinker(struct shrinker
*shrink
,
2048 struct shrink_control
*sc
)
2051 struct nfs_inode
*nfsi
, *next
;
2052 struct nfs_access_entry
*cache
;
2053 int nr_to_scan
= sc
->nr_to_scan
;
2054 gfp_t gfp_mask
= sc
->gfp_mask
;
2056 if ((gfp_mask
& GFP_KERNEL
) != GFP_KERNEL
)
2057 return (nr_to_scan
== 0) ? 0 : -1;
2059 spin_lock(&nfs_access_lru_lock
);
2060 list_for_each_entry_safe(nfsi
, next
, &nfs_access_lru_list
, access_cache_inode_lru
) {
2061 struct inode
*inode
;
2063 if (nr_to_scan
-- == 0)
2065 inode
= &nfsi
->vfs_inode
;
2066 spin_lock(&inode
->i_lock
);
2067 if (list_empty(&nfsi
->access_cache_entry_lru
))
2068 goto remove_lru_entry
;
2069 cache
= list_entry(nfsi
->access_cache_entry_lru
.next
,
2070 struct nfs_access_entry
, lru
);
2071 list_move(&cache
->lru
, &head
);
2072 rb_erase(&cache
->rb_node
, &nfsi
->access_cache
);
2073 if (!list_empty(&nfsi
->access_cache_entry_lru
))
2074 list_move_tail(&nfsi
->access_cache_inode_lru
,
2075 &nfs_access_lru_list
);
2078 list_del_init(&nfsi
->access_cache_inode_lru
);
2079 smp_mb__before_clear_bit();
2080 clear_bit(NFS_INO_ACL_LRU_SET
, &nfsi
->flags
);
2081 smp_mb__after_clear_bit();
2083 spin_unlock(&inode
->i_lock
);
2085 spin_unlock(&nfs_access_lru_lock
);
2086 nfs_access_free_list(&head
);
2087 return (atomic_long_read(&nfs_access_nr_entries
) / 100) * sysctl_vfs_cache_pressure
;
2090 static void __nfs_access_zap_cache(struct nfs_inode
*nfsi
, struct list_head
*head
)
2092 struct rb_root
*root_node
= &nfsi
->access_cache
;
2094 struct nfs_access_entry
*entry
;
2096 /* Unhook entries from the cache */
2097 while ((n
= rb_first(root_node
)) != NULL
) {
2098 entry
= rb_entry(n
, struct nfs_access_entry
, rb_node
);
2099 rb_erase(n
, root_node
);
2100 list_move(&entry
->lru
, head
);
2102 nfsi
->cache_validity
&= ~NFS_INO_INVALID_ACCESS
;
2105 void nfs_access_zap_cache(struct inode
*inode
)
2109 if (test_bit(NFS_INO_ACL_LRU_SET
, &NFS_I(inode
)->flags
) == 0)
2111 /* Remove from global LRU init */
2112 spin_lock(&nfs_access_lru_lock
);
2113 if (test_and_clear_bit(NFS_INO_ACL_LRU_SET
, &NFS_I(inode
)->flags
))
2114 list_del_init(&NFS_I(inode
)->access_cache_inode_lru
);
2116 spin_lock(&inode
->i_lock
);
2117 __nfs_access_zap_cache(NFS_I(inode
), &head
);
2118 spin_unlock(&inode
->i_lock
);
2119 spin_unlock(&nfs_access_lru_lock
);
2120 nfs_access_free_list(&head
);
2123 static struct nfs_access_entry
*nfs_access_search_rbtree(struct inode
*inode
, struct rpc_cred
*cred
)
2125 struct rb_node
*n
= NFS_I(inode
)->access_cache
.rb_node
;
2126 struct nfs_access_entry
*entry
;
2129 entry
= rb_entry(n
, struct nfs_access_entry
, rb_node
);
2131 if (cred
< entry
->cred
)
2133 else if (cred
> entry
->cred
)
2141 static int nfs_access_get_cached(struct inode
*inode
, struct rpc_cred
*cred
, struct nfs_access_entry
*res
)
2143 struct nfs_inode
*nfsi
= NFS_I(inode
);
2144 struct nfs_access_entry
*cache
;
2147 spin_lock(&inode
->i_lock
);
2148 if (nfsi
->cache_validity
& NFS_INO_INVALID_ACCESS
)
2150 cache
= nfs_access_search_rbtree(inode
, cred
);
2153 if (!nfs_have_delegated_attributes(inode
) &&
2154 !time_in_range_open(jiffies
, cache
->jiffies
, cache
->jiffies
+ nfsi
->attrtimeo
))
2156 res
->jiffies
= cache
->jiffies
;
2157 res
->cred
= cache
->cred
;
2158 res
->mask
= cache
->mask
;
2159 list_move_tail(&cache
->lru
, &nfsi
->access_cache_entry_lru
);
2162 spin_unlock(&inode
->i_lock
);
2165 rb_erase(&cache
->rb_node
, &nfsi
->access_cache
);
2166 list_del(&cache
->lru
);
2167 spin_unlock(&inode
->i_lock
);
2168 nfs_access_free_entry(cache
);
2171 spin_unlock(&inode
->i_lock
);
2172 nfs_access_zap_cache(inode
);
2176 static void nfs_access_add_rbtree(struct inode
*inode
, struct nfs_access_entry
*set
)
2178 struct nfs_inode
*nfsi
= NFS_I(inode
);
2179 struct rb_root
*root_node
= &nfsi
->access_cache
;
2180 struct rb_node
**p
= &root_node
->rb_node
;
2181 struct rb_node
*parent
= NULL
;
2182 struct nfs_access_entry
*entry
;
2184 spin_lock(&inode
->i_lock
);
2185 while (*p
!= NULL
) {
2187 entry
= rb_entry(parent
, struct nfs_access_entry
, rb_node
);
2189 if (set
->cred
< entry
->cred
)
2190 p
= &parent
->rb_left
;
2191 else if (set
->cred
> entry
->cred
)
2192 p
= &parent
->rb_right
;
2196 rb_link_node(&set
->rb_node
, parent
, p
);
2197 rb_insert_color(&set
->rb_node
, root_node
);
2198 list_add_tail(&set
->lru
, &nfsi
->access_cache_entry_lru
);
2199 spin_unlock(&inode
->i_lock
);
2202 rb_replace_node(parent
, &set
->rb_node
, root_node
);
2203 list_add_tail(&set
->lru
, &nfsi
->access_cache_entry_lru
);
2204 list_del(&entry
->lru
);
2205 spin_unlock(&inode
->i_lock
);
2206 nfs_access_free_entry(entry
);
2209 static void nfs_access_add_cache(struct inode
*inode
, struct nfs_access_entry
*set
)
2211 struct nfs_access_entry
*cache
= kmalloc(sizeof(*cache
), GFP_KERNEL
);
2214 RB_CLEAR_NODE(&cache
->rb_node
);
2215 cache
->jiffies
= set
->jiffies
;
2216 cache
->cred
= get_rpccred(set
->cred
);
2217 cache
->mask
= set
->mask
;
2219 nfs_access_add_rbtree(inode
, cache
);
2221 /* Update accounting */
2222 smp_mb__before_atomic_inc();
2223 atomic_long_inc(&nfs_access_nr_entries
);
2224 smp_mb__after_atomic_inc();
2226 /* Add inode to global LRU list */
2227 if (!test_bit(NFS_INO_ACL_LRU_SET
, &NFS_I(inode
)->flags
)) {
2228 spin_lock(&nfs_access_lru_lock
);
2229 if (!test_and_set_bit(NFS_INO_ACL_LRU_SET
, &NFS_I(inode
)->flags
))
2230 list_add_tail(&NFS_I(inode
)->access_cache_inode_lru
,
2231 &nfs_access_lru_list
);
2232 spin_unlock(&nfs_access_lru_lock
);
2236 static int nfs_do_access(struct inode
*inode
, struct rpc_cred
*cred
, int mask
)
2238 struct nfs_access_entry cache
;
2241 status
= nfs_access_get_cached(inode
, cred
, &cache
);
2245 /* Be clever: ask server to check for all possible rights */
2246 cache
.mask
= MAY_EXEC
| MAY_WRITE
| MAY_READ
;
2248 cache
.jiffies
= jiffies
;
2249 status
= NFS_PROTO(inode
)->access(inode
, &cache
);
2251 if (status
== -ESTALE
) {
2252 nfs_zap_caches(inode
);
2253 if (!S_ISDIR(inode
->i_mode
))
2254 set_bit(NFS_INO_STALE
, &NFS_I(inode
)->flags
);
2258 nfs_access_add_cache(inode
, &cache
);
2260 if ((mask
& ~cache
.mask
& (MAY_READ
| MAY_WRITE
| MAY_EXEC
)) == 0)
2265 static int nfs_open_permission_mask(int openflags
)
2269 if ((openflags
& O_ACCMODE
) != O_WRONLY
)
2271 if ((openflags
& O_ACCMODE
) != O_RDONLY
)
2273 if (openflags
& __FMODE_EXEC
)
2278 int nfs_may_open(struct inode
*inode
, struct rpc_cred
*cred
, int openflags
)
2280 return nfs_do_access(inode
, cred
, nfs_open_permission_mask(openflags
));
2283 int nfs_permission(struct inode
*inode
, int mask
)
2285 struct rpc_cred
*cred
;
2288 if (mask
& MAY_NOT_BLOCK
)
2291 nfs_inc_stats(inode
, NFSIOS_VFSACCESS
);
2293 if ((mask
& (MAY_READ
| MAY_WRITE
| MAY_EXEC
)) == 0)
2295 /* Is this sys_access() ? */
2296 if (mask
& (MAY_ACCESS
| MAY_CHDIR
))
2299 switch (inode
->i_mode
& S_IFMT
) {
2303 /* NFSv4 has atomic_open... */
2304 if (nfs_server_capable(inode
, NFS_CAP_ATOMIC_OPEN
)
2305 && (mask
& MAY_OPEN
)
2306 && !(mask
& MAY_EXEC
))
2311 * Optimize away all write operations, since the server
2312 * will check permissions when we perform the op.
2314 if ((mask
& MAY_WRITE
) && !(mask
& MAY_READ
))
2319 if (!NFS_PROTO(inode
)->access
)
2322 cred
= rpc_lookup_cred();
2323 if (!IS_ERR(cred
)) {
2324 res
= nfs_do_access(inode
, cred
, mask
);
2327 res
= PTR_ERR(cred
);
2329 if (!res
&& (mask
& MAY_EXEC
) && !execute_ok(inode
))
2332 dfprintk(VFS
, "NFS: permission(%s/%ld), mask=0x%x, res=%d\n",
2333 inode
->i_sb
->s_id
, inode
->i_ino
, mask
, res
);
2336 res
= nfs_revalidate_inode(NFS_SERVER(inode
), inode
);
2338 res
= generic_permission(inode
, mask
);
2344 * version-control: t
2345 * kept-new-versions: 5