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
*, umode_t
, struct nameidata
*);
51 static int nfs_mkdir(struct inode
*, struct dentry
*, umode_t
);
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
*, umode_t
, dev_t
);
57 static int nfs_rename(struct inode
*, struct dentry
*,
58 struct inode
*, struct dentry
*);
59 static int nfs_fsync_dir(struct file
*, loff_t
, loff_t
, 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 int nfs_atomic_open(struct inode
*, struct dentry
*,
115 struct file
*, unsigned, umode_t
,
117 const struct inode_operations nfs4_dir_inode_operations
= {
118 .create
= nfs_create
,
119 .lookup
= nfs_lookup
,
120 .atomic_open
= nfs_atomic_open
,
122 .unlink
= nfs_unlink
,
123 .symlink
= nfs_symlink
,
127 .rename
= nfs_rename
,
128 .permission
= nfs_permission
,
129 .getattr
= nfs_getattr
,
130 .setattr
= nfs_setattr
,
131 .getxattr
= generic_getxattr
,
132 .setxattr
= generic_setxattr
,
133 .listxattr
= generic_listxattr
,
134 .removexattr
= generic_removexattr
,
137 #endif /* CONFIG_NFS_V4 */
139 static struct nfs_open_dir_context
*alloc_nfs_open_dir_context(struct inode
*dir
, struct rpc_cred
*cred
)
141 struct nfs_open_dir_context
*ctx
;
142 ctx
= kmalloc(sizeof(*ctx
), GFP_KERNEL
);
145 ctx
->attr_gencount
= NFS_I(dir
)->attr_gencount
;
148 ctx
->cred
= get_rpccred(cred
);
151 return ERR_PTR(-ENOMEM
);
154 static void put_nfs_open_dir_context(struct nfs_open_dir_context
*ctx
)
156 put_rpccred(ctx
->cred
);
164 nfs_opendir(struct inode
*inode
, struct file
*filp
)
167 struct nfs_open_dir_context
*ctx
;
168 struct rpc_cred
*cred
;
170 dfprintk(FILE, "NFS: open dir(%s/%s)\n",
171 filp
->f_path
.dentry
->d_parent
->d_name
.name
,
172 filp
->f_path
.dentry
->d_name
.name
);
174 nfs_inc_stats(inode
, NFSIOS_VFSOPEN
);
176 cred
= rpc_lookup_cred();
178 return PTR_ERR(cred
);
179 ctx
= alloc_nfs_open_dir_context(inode
, cred
);
184 filp
->private_data
= ctx
;
185 if (filp
->f_path
.dentry
== filp
->f_path
.mnt
->mnt_root
) {
186 /* This is a mountpoint, so d_revalidate will never
187 * have been called, so we need to refresh the
188 * inode (for close-open consistency) ourselves.
190 __nfs_revalidate_inode(NFS_SERVER(inode
), inode
);
198 nfs_closedir(struct inode
*inode
, struct file
*filp
)
200 put_nfs_open_dir_context(filp
->private_data
);
204 struct nfs_cache_array_entry
{
208 unsigned char d_type
;
211 struct nfs_cache_array
{
215 struct nfs_cache_array_entry array
[0];
218 typedef int (*decode_dirent_t
)(struct xdr_stream
*, struct nfs_entry
*, int);
222 unsigned long page_index
;
225 loff_t current_index
;
226 decode_dirent_t decode
;
228 unsigned long timestamp
;
229 unsigned long gencount
;
230 unsigned int cache_entry_index
;
233 } nfs_readdir_descriptor_t
;
236 * The caller is responsible for calling nfs_readdir_release_array(page)
239 struct nfs_cache_array
*nfs_readdir_get_array(struct page
*page
)
243 return ERR_PTR(-EIO
);
246 return ERR_PTR(-ENOMEM
);
251 void nfs_readdir_release_array(struct page
*page
)
257 * we are freeing strings created by nfs_add_to_readdir_array()
260 void nfs_readdir_clear_array(struct page
*page
)
262 struct nfs_cache_array
*array
;
265 array
= kmap_atomic(page
);
266 for (i
= 0; i
< array
->size
; i
++)
267 kfree(array
->array
[i
].string
.name
);
268 kunmap_atomic(array
);
272 * the caller is responsible for freeing qstr.name
273 * when called by nfs_readdir_add_to_array, the strings will be freed in
274 * nfs_clear_readdir_array()
277 int nfs_readdir_make_qstr(struct qstr
*string
, const char *name
, unsigned int len
)
280 string
->name
= kmemdup(name
, len
, GFP_KERNEL
);
281 if (string
->name
== NULL
)
284 * Avoid a kmemleak false positive. The pointer to the name is stored
285 * in a page cache page which kmemleak does not scan.
287 kmemleak_not_leak(string
->name
);
288 string
->hash
= full_name_hash(name
, len
);
293 int nfs_readdir_add_to_array(struct nfs_entry
*entry
, struct page
*page
)
295 struct nfs_cache_array
*array
= nfs_readdir_get_array(page
);
296 struct nfs_cache_array_entry
*cache_entry
;
300 return PTR_ERR(array
);
302 cache_entry
= &array
->array
[array
->size
];
304 /* Check that this entry lies within the page bounds */
306 if ((char *)&cache_entry
[1] - (char *)page_address(page
) > PAGE_SIZE
)
309 cache_entry
->cookie
= entry
->prev_cookie
;
310 cache_entry
->ino
= entry
->ino
;
311 cache_entry
->d_type
= entry
->d_type
;
312 ret
= nfs_readdir_make_qstr(&cache_entry
->string
, entry
->name
, entry
->len
);
315 array
->last_cookie
= entry
->cookie
;
318 array
->eof_index
= array
->size
;
320 nfs_readdir_release_array(page
);
325 int nfs_readdir_search_for_pos(struct nfs_cache_array
*array
, nfs_readdir_descriptor_t
*desc
)
327 loff_t diff
= desc
->file
->f_pos
- desc
->current_index
;
332 if (diff
>= array
->size
) {
333 if (array
->eof_index
>= 0)
338 index
= (unsigned int)diff
;
339 *desc
->dir_cookie
= array
->array
[index
].cookie
;
340 desc
->cache_entry_index
= index
;
348 int nfs_readdir_search_for_cookie(struct nfs_cache_array
*array
, nfs_readdir_descriptor_t
*desc
)
352 int status
= -EAGAIN
;
354 for (i
= 0; i
< array
->size
; i
++) {
355 if (array
->array
[i
].cookie
== *desc
->dir_cookie
) {
356 struct nfs_inode
*nfsi
= NFS_I(desc
->file
->f_path
.dentry
->d_inode
);
357 struct nfs_open_dir_context
*ctx
= desc
->file
->private_data
;
359 new_pos
= desc
->current_index
+ i
;
360 if (ctx
->attr_gencount
!= nfsi
->attr_gencount
361 || (nfsi
->cache_validity
& (NFS_INO_INVALID_ATTR
|NFS_INO_INVALID_DATA
))) {
363 ctx
->attr_gencount
= nfsi
->attr_gencount
;
364 } else if (new_pos
< desc
->file
->f_pos
) {
366 && ctx
->dup_cookie
== *desc
->dir_cookie
) {
367 if (printk_ratelimit()) {
368 pr_notice("NFS: directory %s/%s contains a readdir loop."
369 "Please contact your server vendor. "
370 "The file: %s has duplicate cookie %llu\n",
371 desc
->file
->f_dentry
->d_parent
->d_name
.name
,
372 desc
->file
->f_dentry
->d_name
.name
,
373 array
->array
[i
].string
.name
,
379 ctx
->dup_cookie
= *desc
->dir_cookie
;
382 desc
->file
->f_pos
= new_pos
;
383 desc
->cache_entry_index
= i
;
387 if (array
->eof_index
>= 0) {
388 status
= -EBADCOOKIE
;
389 if (*desc
->dir_cookie
== array
->last_cookie
)
397 int nfs_readdir_search_array(nfs_readdir_descriptor_t
*desc
)
399 struct nfs_cache_array
*array
;
402 array
= nfs_readdir_get_array(desc
->page
);
404 status
= PTR_ERR(array
);
408 if (*desc
->dir_cookie
== 0)
409 status
= nfs_readdir_search_for_pos(array
, desc
);
411 status
= nfs_readdir_search_for_cookie(array
, desc
);
413 if (status
== -EAGAIN
) {
414 desc
->last_cookie
= array
->last_cookie
;
415 desc
->current_index
+= array
->size
;
418 nfs_readdir_release_array(desc
->page
);
423 /* Fill a page with xdr information before transferring to the cache page */
425 int nfs_readdir_xdr_filler(struct page
**pages
, nfs_readdir_descriptor_t
*desc
,
426 struct nfs_entry
*entry
, struct file
*file
, struct inode
*inode
)
428 struct nfs_open_dir_context
*ctx
= file
->private_data
;
429 struct rpc_cred
*cred
= ctx
->cred
;
430 unsigned long timestamp
, gencount
;
435 gencount
= nfs_inc_attr_generation_counter();
436 error
= NFS_PROTO(inode
)->readdir(file
->f_path
.dentry
, cred
, entry
->cookie
, pages
,
437 NFS_SERVER(inode
)->dtsize
, desc
->plus
);
439 /* We requested READDIRPLUS, but the server doesn't grok it */
440 if (error
== -ENOTSUPP
&& desc
->plus
) {
441 NFS_SERVER(inode
)->caps
&= ~NFS_CAP_READDIRPLUS
;
442 clear_bit(NFS_INO_ADVISE_RDPLUS
, &NFS_I(inode
)->flags
);
448 desc
->timestamp
= timestamp
;
449 desc
->gencount
= gencount
;
454 static int xdr_decode(nfs_readdir_descriptor_t
*desc
,
455 struct nfs_entry
*entry
, struct xdr_stream
*xdr
)
459 error
= desc
->decode(xdr
, entry
, desc
->plus
);
462 entry
->fattr
->time_start
= desc
->timestamp
;
463 entry
->fattr
->gencount
= desc
->gencount
;
468 int nfs_same_file(struct dentry
*dentry
, struct nfs_entry
*entry
)
470 if (dentry
->d_inode
== NULL
)
472 if (nfs_compare_fh(entry
->fh
, NFS_FH(dentry
->d_inode
)) != 0)
480 bool nfs_use_readdirplus(struct inode
*dir
, struct file
*filp
)
482 if (!nfs_server_capable(dir
, NFS_CAP_READDIRPLUS
))
484 if (test_and_clear_bit(NFS_INO_ADVISE_RDPLUS
, &NFS_I(dir
)->flags
))
486 if (filp
->f_pos
== 0)
492 * This function is called by the lookup code to request the use of
493 * readdirplus to accelerate any future lookups in the same
497 void nfs_advise_use_readdirplus(struct inode
*dir
)
499 set_bit(NFS_INO_ADVISE_RDPLUS
, &NFS_I(dir
)->flags
);
503 void nfs_prime_dcache(struct dentry
*parent
, struct nfs_entry
*entry
)
505 struct qstr filename
= QSTR_INIT(entry
->name
, entry
->len
);
506 struct dentry
*dentry
;
507 struct dentry
*alias
;
508 struct inode
*dir
= parent
->d_inode
;
511 if (filename
.name
[0] == '.') {
512 if (filename
.len
== 1)
514 if (filename
.len
== 2 && filename
.name
[1] == '.')
517 filename
.hash
= full_name_hash(filename
.name
, filename
.len
);
519 dentry
= d_lookup(parent
, &filename
);
520 if (dentry
!= NULL
) {
521 if (nfs_same_file(dentry
, entry
)) {
522 nfs_refresh_inode(dentry
->d_inode
, entry
->fattr
);
530 dentry
= d_alloc(parent
, &filename
);
534 inode
= nfs_fhget(dentry
->d_sb
, entry
->fh
, entry
->fattr
);
538 alias
= d_materialise_unique(dentry
, inode
);
542 nfs_set_verifier(alias
, nfs_save_change_attribute(dir
));
545 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
551 /* Perform conversion from xdr to cache array */
553 int nfs_readdir_page_filler(nfs_readdir_descriptor_t
*desc
, struct nfs_entry
*entry
,
554 struct page
**xdr_pages
, struct page
*page
, unsigned int buflen
)
556 struct xdr_stream stream
;
558 struct page
*scratch
;
559 struct nfs_cache_array
*array
;
560 unsigned int count
= 0;
563 scratch
= alloc_page(GFP_KERNEL
);
567 xdr_init_decode_pages(&stream
, &buf
, xdr_pages
, buflen
);
568 xdr_set_scratch_buffer(&stream
, page_address(scratch
), PAGE_SIZE
);
571 status
= xdr_decode(desc
, entry
, &stream
);
573 if (status
== -EAGAIN
)
581 nfs_prime_dcache(desc
->file
->f_path
.dentry
, entry
);
583 status
= nfs_readdir_add_to_array(entry
, page
);
586 } while (!entry
->eof
);
588 if (count
== 0 || (status
== -EBADCOOKIE
&& entry
->eof
!= 0)) {
589 array
= nfs_readdir_get_array(page
);
590 if (!IS_ERR(array
)) {
591 array
->eof_index
= array
->size
;
593 nfs_readdir_release_array(page
);
595 status
= PTR_ERR(array
);
603 void nfs_readdir_free_pagearray(struct page
**pages
, unsigned int npages
)
606 for (i
= 0; i
< npages
; i
++)
611 void nfs_readdir_free_large_page(void *ptr
, struct page
**pages
,
614 nfs_readdir_free_pagearray(pages
, npages
);
618 * nfs_readdir_large_page will allocate pages that must be freed with a call
619 * to nfs_readdir_free_large_page
622 int nfs_readdir_large_page(struct page
**pages
, unsigned int npages
)
626 for (i
= 0; i
< npages
; i
++) {
627 struct page
*page
= alloc_page(GFP_KERNEL
);
635 nfs_readdir_free_pagearray(pages
, i
);
640 int nfs_readdir_xdr_to_array(nfs_readdir_descriptor_t
*desc
, struct page
*page
, struct inode
*inode
)
642 struct page
*pages
[NFS_MAX_READDIR_PAGES
];
643 void *pages_ptr
= NULL
;
644 struct nfs_entry entry
;
645 struct file
*file
= desc
->file
;
646 struct nfs_cache_array
*array
;
647 int status
= -ENOMEM
;
648 unsigned int array_size
= ARRAY_SIZE(pages
);
650 entry
.prev_cookie
= 0;
651 entry
.cookie
= desc
->last_cookie
;
653 entry
.fh
= nfs_alloc_fhandle();
654 entry
.fattr
= nfs_alloc_fattr();
655 entry
.server
= NFS_SERVER(inode
);
656 if (entry
.fh
== NULL
|| entry
.fattr
== NULL
)
659 array
= nfs_readdir_get_array(page
);
661 status
= PTR_ERR(array
);
664 memset(array
, 0, sizeof(struct nfs_cache_array
));
665 array
->eof_index
= -1;
667 status
= nfs_readdir_large_page(pages
, array_size
);
669 goto out_release_array
;
672 status
= nfs_readdir_xdr_filler(pages
, desc
, &entry
, file
, inode
);
677 status
= nfs_readdir_page_filler(desc
, &entry
, pages
, page
, pglen
);
679 if (status
== -ENOSPC
)
683 } while (array
->eof_index
< 0);
685 nfs_readdir_free_large_page(pages_ptr
, pages
, array_size
);
687 nfs_readdir_release_array(page
);
689 nfs_free_fattr(entry
.fattr
);
690 nfs_free_fhandle(entry
.fh
);
695 * Now we cache directories properly, by converting xdr information
696 * to an array that can be used for lookups later. This results in
697 * fewer cache pages, since we can store more information on each page.
698 * We only need to convert from xdr once so future lookups are much simpler
701 int nfs_readdir_filler(nfs_readdir_descriptor_t
*desc
, struct page
* page
)
703 struct inode
*inode
= desc
->file
->f_path
.dentry
->d_inode
;
706 ret
= nfs_readdir_xdr_to_array(desc
, page
, inode
);
709 SetPageUptodate(page
);
711 if (invalidate_inode_pages2_range(inode
->i_mapping
, page
->index
+ 1, -1) < 0) {
712 /* Should never happen */
713 nfs_zap_mapping(inode
, inode
->i_mapping
);
723 void cache_page_release(nfs_readdir_descriptor_t
*desc
)
725 if (!desc
->page
->mapping
)
726 nfs_readdir_clear_array(desc
->page
);
727 page_cache_release(desc
->page
);
732 struct page
*get_cache_page(nfs_readdir_descriptor_t
*desc
)
734 return read_cache_page(desc
->file
->f_path
.dentry
->d_inode
->i_mapping
,
735 desc
->page_index
, (filler_t
*)nfs_readdir_filler
, desc
);
739 * Returns 0 if desc->dir_cookie was found on page desc->page_index
742 int find_cache_page(nfs_readdir_descriptor_t
*desc
)
746 desc
->page
= get_cache_page(desc
);
747 if (IS_ERR(desc
->page
))
748 return PTR_ERR(desc
->page
);
750 res
= nfs_readdir_search_array(desc
);
752 cache_page_release(desc
);
756 /* Search for desc->dir_cookie from the beginning of the page cache */
758 int readdir_search_pagecache(nfs_readdir_descriptor_t
*desc
)
762 if (desc
->page_index
== 0) {
763 desc
->current_index
= 0;
764 desc
->last_cookie
= 0;
767 res
= find_cache_page(desc
);
768 } while (res
== -EAGAIN
);
773 * Once we've found the start of the dirent within a page: fill 'er up...
776 int nfs_do_filldir(nfs_readdir_descriptor_t
*desc
, void *dirent
,
779 struct file
*file
= desc
->file
;
782 struct nfs_cache_array
*array
= NULL
;
783 struct nfs_open_dir_context
*ctx
= file
->private_data
;
785 array
= nfs_readdir_get_array(desc
->page
);
787 res
= PTR_ERR(array
);
791 for (i
= desc
->cache_entry_index
; i
< array
->size
; i
++) {
792 struct nfs_cache_array_entry
*ent
;
794 ent
= &array
->array
[i
];
795 if (filldir(dirent
, ent
->string
.name
, ent
->string
.len
,
796 file
->f_pos
, nfs_compat_user_ino64(ent
->ino
),
802 if (i
< (array
->size
-1))
803 *desc
->dir_cookie
= array
->array
[i
+1].cookie
;
805 *desc
->dir_cookie
= array
->last_cookie
;
809 if (array
->eof_index
>= 0)
812 nfs_readdir_release_array(desc
->page
);
814 cache_page_release(desc
);
815 dfprintk(DIRCACHE
, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n",
816 (unsigned long long)*desc
->dir_cookie
, res
);
821 * If we cannot find a cookie in our cache, we suspect that this is
822 * because it points to a deleted file, so we ask the server to return
823 * whatever it thinks is the next entry. We then feed this to filldir.
824 * If all goes well, we should then be able to find our way round the
825 * cache on the next call to readdir_search_pagecache();
827 * NOTE: we cannot add the anonymous page to the pagecache because
828 * the data it contains might not be page aligned. Besides,
829 * we should already have a complete representation of the
830 * directory in the page cache by the time we get here.
833 int uncached_readdir(nfs_readdir_descriptor_t
*desc
, void *dirent
,
836 struct page
*page
= NULL
;
838 struct inode
*inode
= desc
->file
->f_path
.dentry
->d_inode
;
839 struct nfs_open_dir_context
*ctx
= desc
->file
->private_data
;
841 dfprintk(DIRCACHE
, "NFS: uncached_readdir() searching for cookie %Lu\n",
842 (unsigned long long)*desc
->dir_cookie
);
844 page
= alloc_page(GFP_HIGHUSER
);
850 desc
->page_index
= 0;
851 desc
->last_cookie
= *desc
->dir_cookie
;
855 status
= nfs_readdir_xdr_to_array(desc
, page
, inode
);
859 status
= nfs_do_filldir(desc
, dirent
, filldir
);
862 dfprintk(DIRCACHE
, "NFS: %s: returns %d\n",
866 cache_page_release(desc
);
870 /* The file offset position represents the dirent entry number. A
871 last cookie cache takes care of the common case of reading the
874 static int nfs_readdir(struct file
*filp
, void *dirent
, filldir_t filldir
)
876 struct dentry
*dentry
= filp
->f_path
.dentry
;
877 struct inode
*inode
= dentry
->d_inode
;
878 nfs_readdir_descriptor_t my_desc
,
880 struct nfs_open_dir_context
*dir_ctx
= filp
->private_data
;
883 dfprintk(FILE, "NFS: readdir(%s/%s) starting at cookie %llu\n",
884 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
,
885 (long long)filp
->f_pos
);
886 nfs_inc_stats(inode
, NFSIOS_VFSGETDENTS
);
889 * filp->f_pos points to the dirent entry number.
890 * *desc->dir_cookie has the cookie for the next entry. We have
891 * to either find the entry with the appropriate number or
892 * revalidate the cookie.
894 memset(desc
, 0, sizeof(*desc
));
897 desc
->dir_cookie
= &dir_ctx
->dir_cookie
;
898 desc
->decode
= NFS_PROTO(inode
)->decode_dirent
;
899 desc
->plus
= nfs_use_readdirplus(inode
, filp
) ? 1 : 0;
901 nfs_block_sillyrename(dentry
);
902 res
= nfs_revalidate_mapping(inode
, filp
->f_mapping
);
907 res
= readdir_search_pagecache(desc
);
909 if (res
== -EBADCOOKIE
) {
911 /* This means either end of directory */
912 if (*desc
->dir_cookie
&& desc
->eof
== 0) {
913 /* Or that the server has 'lost' a cookie */
914 res
= uncached_readdir(desc
, dirent
, filldir
);
920 if (res
== -ETOOSMALL
&& desc
->plus
) {
921 clear_bit(NFS_INO_ADVISE_RDPLUS
, &NFS_I(inode
)->flags
);
922 nfs_zap_caches(inode
);
923 desc
->page_index
= 0;
931 res
= nfs_do_filldir(desc
, dirent
, filldir
);
934 } while (!desc
->eof
);
936 nfs_unblock_sillyrename(dentry
);
939 dfprintk(FILE, "NFS: readdir(%s/%s) returns %d\n",
940 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
,
945 static loff_t
nfs_llseek_dir(struct file
*filp
, loff_t offset
, int origin
)
947 struct dentry
*dentry
= filp
->f_path
.dentry
;
948 struct inode
*inode
= dentry
->d_inode
;
949 struct nfs_open_dir_context
*dir_ctx
= filp
->private_data
;
951 dfprintk(FILE, "NFS: llseek dir(%s/%s, %lld, %d)\n",
952 dentry
->d_parent
->d_name
.name
,
956 mutex_lock(&inode
->i_mutex
);
959 offset
+= filp
->f_pos
;
967 if (offset
!= filp
->f_pos
) {
968 filp
->f_pos
= offset
;
969 dir_ctx
->dir_cookie
= 0;
973 mutex_unlock(&inode
->i_mutex
);
978 * All directory operations under NFS are synchronous, so fsync()
979 * is a dummy operation.
981 static int nfs_fsync_dir(struct file
*filp
, loff_t start
, loff_t end
,
984 struct dentry
*dentry
= filp
->f_path
.dentry
;
985 struct inode
*inode
= dentry
->d_inode
;
987 dfprintk(FILE, "NFS: fsync dir(%s/%s) datasync %d\n",
988 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
,
991 mutex_lock(&inode
->i_mutex
);
992 nfs_inc_stats(dentry
->d_inode
, NFSIOS_VFSFSYNC
);
993 mutex_unlock(&inode
->i_mutex
);
998 * nfs_force_lookup_revalidate - Mark the directory as having changed
999 * @dir - pointer to directory inode
1001 * This forces the revalidation code in nfs_lookup_revalidate() to do a
1002 * full lookup on all child dentries of 'dir' whenever a change occurs
1003 * on the server that might have invalidated our dcache.
1005 * The caller should be holding dir->i_lock
1007 void nfs_force_lookup_revalidate(struct inode
*dir
)
1009 NFS_I(dir
)->cache_change_attribute
++;
1013 * A check for whether or not the parent directory has changed.
1014 * In the case it has, we assume that the dentries are untrustworthy
1015 * and may need to be looked up again.
1017 static int nfs_check_verifier(struct inode
*dir
, struct dentry
*dentry
)
1019 if (IS_ROOT(dentry
))
1021 if (NFS_SERVER(dir
)->flags
& NFS_MOUNT_LOOKUP_CACHE_NONE
)
1023 if (!nfs_verify_change_attribute(dir
, dentry
->d_time
))
1025 /* Revalidate nfsi->cache_change_attribute before we declare a match */
1026 if (nfs_revalidate_inode(NFS_SERVER(dir
), dir
) < 0)
1028 if (!nfs_verify_change_attribute(dir
, dentry
->d_time
))
1034 * Return the intent data that applies to this particular path component
1036 * Note that the current set of intents only apply to the very last
1037 * component of the path and none of them is set before that last
1040 static inline unsigned int nfs_lookup_check_intent(unsigned int flags
,
1043 return flags
& mask
;
1047 * Use intent information to check whether or not we're going to do
1048 * an O_EXCL create using this path component.
1050 static int nfs_is_exclusive_create(struct inode
*dir
, struct nameidata
*nd
)
1052 if (NFS_PROTO(dir
)->version
== 2)
1054 return nd
&& nfs_lookup_check_intent(nd
->flags
, LOOKUP_EXCL
);
1058 * Inode and filehandle revalidation for lookups.
1060 * We force revalidation in the cases where the VFS sets LOOKUP_REVAL,
1061 * or if the intent information indicates that we're about to open this
1062 * particular file and the "nocto" mount flag is not set.
1066 int nfs_lookup_verify_inode(struct inode
*inode
, struct nameidata
*nd
)
1068 struct nfs_server
*server
= NFS_SERVER(inode
);
1070 if (IS_AUTOMOUNT(inode
))
1073 /* VFS wants an on-the-wire revalidation */
1074 if (nd
->flags
& LOOKUP_REVAL
)
1076 /* This is an open(2) */
1077 if (nfs_lookup_check_intent(nd
->flags
, LOOKUP_OPEN
) != 0 &&
1078 !(server
->flags
& NFS_MOUNT_NOCTO
) &&
1079 (S_ISREG(inode
->i_mode
) ||
1080 S_ISDIR(inode
->i_mode
)))
1084 return nfs_revalidate_inode(server
, inode
);
1086 return __nfs_revalidate_inode(server
, inode
);
1090 * We judge how long we want to trust negative
1091 * dentries by looking at the parent inode mtime.
1093 * If parent mtime has changed, we revalidate, else we wait for a
1094 * period corresponding to the parent's attribute cache timeout value.
1097 int nfs_neg_need_reval(struct inode
*dir
, struct dentry
*dentry
,
1098 struct nameidata
*nd
)
1100 /* Don't revalidate a negative dentry if we're creating a new file */
1101 if (nd
!= NULL
&& nfs_lookup_check_intent(nd
->flags
, LOOKUP_CREATE
) != 0)
1103 if (NFS_SERVER(dir
)->flags
& NFS_MOUNT_LOOKUP_CACHE_NONEG
)
1105 return !nfs_check_verifier(dir
, dentry
);
1109 * This is called every time the dcache has a lookup hit,
1110 * and we should check whether we can really trust that
1113 * NOTE! The hit can be a negative hit too, don't assume
1116 * If the parent directory is seen to have changed, we throw out the
1117 * cached dentry and do a new lookup.
1119 static int nfs_lookup_revalidate(struct dentry
*dentry
, struct nameidata
*nd
)
1122 struct inode
*inode
;
1123 struct dentry
*parent
;
1124 struct nfs_fh
*fhandle
= NULL
;
1125 struct nfs_fattr
*fattr
= NULL
;
1128 if (nd
->flags
& LOOKUP_RCU
)
1131 parent
= dget_parent(dentry
);
1132 dir
= parent
->d_inode
;
1133 nfs_inc_stats(dir
, NFSIOS_DENTRYREVALIDATE
);
1134 inode
= dentry
->d_inode
;
1137 if (nfs_neg_need_reval(dir
, dentry
, nd
))
1139 goto out_valid_noent
;
1142 if (is_bad_inode(inode
)) {
1143 dfprintk(LOOKUPCACHE
, "%s: %s/%s has dud inode\n",
1144 __func__
, dentry
->d_parent
->d_name
.name
,
1145 dentry
->d_name
.name
);
1149 if (nfs_have_delegation(inode
, FMODE_READ
))
1150 goto out_set_verifier
;
1152 /* Force a full look up iff the parent directory has changed */
1153 if (!nfs_is_exclusive_create(dir
, nd
) && nfs_check_verifier(dir
, dentry
)) {
1154 if (nfs_lookup_verify_inode(inode
, nd
))
1155 goto out_zap_parent
;
1159 if (NFS_STALE(inode
))
1163 fhandle
= nfs_alloc_fhandle();
1164 fattr
= nfs_alloc_fattr();
1165 if (fhandle
== NULL
|| fattr
== NULL
)
1168 error
= NFS_PROTO(dir
)->lookup(dir
, &dentry
->d_name
, fhandle
, fattr
);
1171 if (nfs_compare_fh(NFS_FH(inode
), fhandle
))
1173 if ((error
= nfs_refresh_inode(inode
, fattr
)) != 0)
1176 nfs_free_fattr(fattr
);
1177 nfs_free_fhandle(fhandle
);
1179 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1181 /* Success: notify readdir to use READDIRPLUS */
1182 nfs_advise_use_readdirplus(dir
);
1185 dfprintk(LOOKUPCACHE
, "NFS: %s(%s/%s) is valid\n",
1186 __func__
, dentry
->d_parent
->d_name
.name
,
1187 dentry
->d_name
.name
);
1190 nfs_zap_caches(dir
);
1192 nfs_mark_for_revalidate(dir
);
1193 if (inode
&& S_ISDIR(inode
->i_mode
)) {
1194 /* Purge readdir caches. */
1195 nfs_zap_caches(inode
);
1196 /* If we have submounts, don't unhash ! */
1197 if (have_submounts(dentry
))
1199 if (dentry
->d_flags
& DCACHE_DISCONNECTED
)
1201 shrink_dcache_parent(dentry
);
1204 nfs_free_fattr(fattr
);
1205 nfs_free_fhandle(fhandle
);
1207 dfprintk(LOOKUPCACHE
, "NFS: %s(%s/%s) is invalid\n",
1208 __func__
, dentry
->d_parent
->d_name
.name
,
1209 dentry
->d_name
.name
);
1212 nfs_free_fattr(fattr
);
1213 nfs_free_fhandle(fhandle
);
1215 dfprintk(LOOKUPCACHE
, "NFS: %s(%s/%s) lookup returned error %d\n",
1216 __func__
, dentry
->d_parent
->d_name
.name
,
1217 dentry
->d_name
.name
, error
);
1222 * This is called from dput() when d_count is going to 0.
1224 static int nfs_dentry_delete(const struct dentry
*dentry
)
1226 dfprintk(VFS
, "NFS: dentry_delete(%s/%s, %x)\n",
1227 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
,
1230 /* Unhash any dentry with a stale inode */
1231 if (dentry
->d_inode
!= NULL
&& NFS_STALE(dentry
->d_inode
))
1234 if (dentry
->d_flags
& DCACHE_NFSFS_RENAMED
) {
1235 /* Unhash it, so that ->d_iput() would be called */
1238 if (!(dentry
->d_sb
->s_flags
& MS_ACTIVE
)) {
1239 /* Unhash it, so that ancestors of killed async unlink
1240 * files will be cleaned up during umount */
1247 static void nfs_drop_nlink(struct inode
*inode
)
1249 spin_lock(&inode
->i_lock
);
1250 if (inode
->i_nlink
> 0)
1252 spin_unlock(&inode
->i_lock
);
1256 * Called when the dentry loses inode.
1257 * We use it to clean up silly-renamed files.
1259 static void nfs_dentry_iput(struct dentry
*dentry
, struct inode
*inode
)
1261 if (S_ISDIR(inode
->i_mode
))
1262 /* drop any readdir cache as it could easily be old */
1263 NFS_I(inode
)->cache_validity
|= NFS_INO_INVALID_DATA
;
1265 if (dentry
->d_flags
& DCACHE_NFSFS_RENAMED
) {
1267 nfs_complete_unlink(dentry
, inode
);
1272 static void nfs_d_release(struct dentry
*dentry
)
1274 /* free cached devname value, if it survived that far */
1275 if (unlikely(dentry
->d_fsdata
)) {
1276 if (dentry
->d_flags
& DCACHE_NFSFS_RENAMED
)
1279 kfree(dentry
->d_fsdata
);
1283 const struct dentry_operations nfs_dentry_operations
= {
1284 .d_revalidate
= nfs_lookup_revalidate
,
1285 .d_delete
= nfs_dentry_delete
,
1286 .d_iput
= nfs_dentry_iput
,
1287 .d_automount
= nfs_d_automount
,
1288 .d_release
= nfs_d_release
,
1291 static struct dentry
*nfs_lookup(struct inode
*dir
, struct dentry
* dentry
, struct nameidata
*nd
)
1294 struct dentry
*parent
;
1295 struct inode
*inode
= NULL
;
1296 struct nfs_fh
*fhandle
= NULL
;
1297 struct nfs_fattr
*fattr
= NULL
;
1300 dfprintk(VFS
, "NFS: lookup(%s/%s)\n",
1301 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
);
1302 nfs_inc_stats(dir
, NFSIOS_VFSLOOKUP
);
1304 res
= ERR_PTR(-ENAMETOOLONG
);
1305 if (dentry
->d_name
.len
> NFS_SERVER(dir
)->namelen
)
1309 * If we're doing an exclusive create, optimize away the lookup
1310 * but don't hash the dentry.
1312 if (nfs_is_exclusive_create(dir
, nd
)) {
1313 d_instantiate(dentry
, NULL
);
1318 res
= ERR_PTR(-ENOMEM
);
1319 fhandle
= nfs_alloc_fhandle();
1320 fattr
= nfs_alloc_fattr();
1321 if (fhandle
== NULL
|| fattr
== NULL
)
1324 parent
= dentry
->d_parent
;
1325 /* Protect against concurrent sillydeletes */
1326 nfs_block_sillyrename(parent
);
1327 error
= NFS_PROTO(dir
)->lookup(dir
, &dentry
->d_name
, fhandle
, fattr
);
1328 if (error
== -ENOENT
)
1331 res
= ERR_PTR(error
);
1332 goto out_unblock_sillyrename
;
1334 inode
= nfs_fhget(dentry
->d_sb
, fhandle
, fattr
);
1335 res
= ERR_CAST(inode
);
1337 goto out_unblock_sillyrename
;
1339 /* Success: notify readdir to use READDIRPLUS */
1340 nfs_advise_use_readdirplus(dir
);
1343 res
= d_materialise_unique(dentry
, inode
);
1346 goto out_unblock_sillyrename
;
1349 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1350 out_unblock_sillyrename
:
1351 nfs_unblock_sillyrename(parent
);
1353 nfs_free_fattr(fattr
);
1354 nfs_free_fhandle(fhandle
);
1358 #ifdef CONFIG_NFS_V4
1359 static int nfs4_lookup_revalidate(struct dentry
*, struct nameidata
*);
1361 const struct dentry_operations nfs4_dentry_operations
= {
1362 .d_revalidate
= nfs4_lookup_revalidate
,
1363 .d_delete
= nfs_dentry_delete
,
1364 .d_iput
= nfs_dentry_iput
,
1365 .d_automount
= nfs_d_automount
,
1366 .d_release
= nfs_d_release
,
1369 static fmode_t
flags_to_mode(int flags
)
1371 fmode_t res
= (__force fmode_t
)flags
& FMODE_EXEC
;
1372 if ((flags
& O_ACCMODE
) != O_WRONLY
)
1374 if ((flags
& O_ACCMODE
) != O_RDONLY
)
1379 static struct nfs_open_context
*create_nfs_open_context(struct dentry
*dentry
, int open_flags
)
1381 return alloc_nfs_open_context(dentry
, flags_to_mode(open_flags
));
1384 static int do_open(struct inode
*inode
, struct file
*filp
)
1386 nfs_fscache_set_inode_cookie(inode
, filp
);
1390 static int nfs_finish_open(struct nfs_open_context
*ctx
,
1391 struct dentry
*dentry
,
1392 struct file
*file
, unsigned open_flags
,
1397 if (ctx
->dentry
!= dentry
) {
1399 ctx
->dentry
= dget(dentry
);
1402 /* If the open_intent is for execute, we have an extra check to make */
1403 if (ctx
->mode
& FMODE_EXEC
) {
1404 err
= nfs_may_open(dentry
->d_inode
, ctx
->cred
, open_flags
);
1409 err
= finish_open(file
, dentry
, do_open
, opened
);
1412 nfs_file_set_open_context(file
, ctx
);
1415 put_nfs_open_context(ctx
);
1419 static int nfs_atomic_open(struct inode
*dir
, struct dentry
*dentry
,
1420 struct file
*file
, unsigned open_flags
,
1421 umode_t mode
, int *opened
)
1423 struct nfs_open_context
*ctx
;
1425 struct iattr attr
= { .ia_valid
= ATTR_OPEN
};
1426 struct inode
*inode
;
1429 /* Expect a negative dentry */
1430 BUG_ON(dentry
->d_inode
);
1432 dfprintk(VFS
, "NFS: atomic_open(%s/%ld), %s\n",
1433 dir
->i_sb
->s_id
, dir
->i_ino
, dentry
->d_name
.name
);
1435 /* NFS only supports OPEN on regular files */
1436 if ((open_flags
& O_DIRECTORY
)) {
1437 if (!d_unhashed(dentry
)) {
1439 * Hashed negative dentry with O_DIRECTORY: dentry was
1440 * revalidated and is fine, no need to perform lookup
1448 if (dentry
->d_name
.len
> NFS_SERVER(dir
)->namelen
)
1449 return -ENAMETOOLONG
;
1451 if (open_flags
& O_CREAT
) {
1452 attr
.ia_valid
|= ATTR_MODE
;
1453 attr
.ia_mode
= mode
& ~current_umask();
1455 if (open_flags
& O_TRUNC
) {
1456 attr
.ia_valid
|= ATTR_SIZE
;
1460 ctx
= create_nfs_open_context(dentry
, open_flags
);
1465 nfs_block_sillyrename(dentry
->d_parent
);
1466 inode
= NFS_PROTO(dir
)->open_context(dir
, ctx
, open_flags
, &attr
);
1468 if (IS_ERR(inode
)) {
1469 nfs_unblock_sillyrename(dentry
->d_parent
);
1470 put_nfs_open_context(ctx
);
1471 err
= PTR_ERR(inode
);
1474 d_add(dentry
, NULL
);
1480 if (!(open_flags
& O_NOFOLLOW
))
1489 res
= d_add_unique(dentry
, inode
);
1493 nfs_unblock_sillyrename(dentry
->d_parent
);
1494 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1496 err
= nfs_finish_open(ctx
, dentry
, file
, open_flags
, opened
);
1503 res
= nfs_lookup(dir
, dentry
, NULL
);
1508 return finish_no_open(file
, res
);
1511 static int nfs4_lookup_revalidate(struct dentry
*dentry
, struct nameidata
*nd
)
1513 struct dentry
*parent
= NULL
;
1514 struct inode
*inode
;
1518 if (nd
->flags
& LOOKUP_RCU
)
1521 if (!(nd
->flags
& LOOKUP_OPEN
) || (nd
->flags
& LOOKUP_DIRECTORY
))
1523 if (d_mountpoint(dentry
))
1526 inode
= dentry
->d_inode
;
1527 parent
= dget_parent(dentry
);
1528 dir
= parent
->d_inode
;
1530 /* We can't create new files in nfs_open_revalidate(), so we
1531 * optimize away revalidation of negative dentries.
1533 if (inode
== NULL
) {
1534 if (!nfs_neg_need_reval(dir
, dentry
, nd
))
1539 /* NFS only supports OPEN on regular files */
1540 if (!S_ISREG(inode
->i_mode
))
1542 /* We cannot do exclusive creation on a positive dentry */
1543 if (nd
&& nd
->flags
& LOOKUP_EXCL
)
1546 /* Let f_op->open() actually open (and revalidate) the file */
1556 return nfs_lookup_revalidate(dentry
, nd
);
1559 #endif /* CONFIG_NFSV4 */
1562 * Code common to create, mkdir, and mknod.
1564 int nfs_instantiate(struct dentry
*dentry
, struct nfs_fh
*fhandle
,
1565 struct nfs_fattr
*fattr
)
1567 struct dentry
*parent
= dget_parent(dentry
);
1568 struct inode
*dir
= parent
->d_inode
;
1569 struct inode
*inode
;
1570 int error
= -EACCES
;
1574 /* We may have been initialized further down */
1575 if (dentry
->d_inode
)
1577 if (fhandle
->size
== 0) {
1578 error
= NFS_PROTO(dir
)->lookup(dir
, &dentry
->d_name
, fhandle
, fattr
);
1582 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1583 if (!(fattr
->valid
& NFS_ATTR_FATTR
)) {
1584 struct nfs_server
*server
= NFS_SB(dentry
->d_sb
);
1585 error
= server
->nfs_client
->rpc_ops
->getattr(server
, fhandle
, fattr
);
1589 inode
= nfs_fhget(dentry
->d_sb
, fhandle
, fattr
);
1590 error
= PTR_ERR(inode
);
1593 d_add(dentry
, inode
);
1598 nfs_mark_for_revalidate(dir
);
1604 * Following a failed create operation, we drop the dentry rather
1605 * than retain a negative dentry. This avoids a problem in the event
1606 * that the operation succeeded on the server, but an error in the
1607 * reply path made it appear to have failed.
1609 static int nfs_create(struct inode
*dir
, struct dentry
*dentry
,
1610 umode_t mode
, struct nameidata
*nd
)
1614 int open_flags
= O_CREAT
|O_EXCL
;
1616 dfprintk(VFS
, "NFS: create(%s/%ld), %s\n",
1617 dir
->i_sb
->s_id
, dir
->i_ino
, dentry
->d_name
.name
);
1619 attr
.ia_mode
= mode
;
1620 attr
.ia_valid
= ATTR_MODE
;
1622 if (nd
&& !(nd
->flags
& LOOKUP_EXCL
))
1623 open_flags
= O_CREAT
;
1625 error
= NFS_PROTO(dir
)->create(dir
, dentry
, &attr
, open_flags
);
1635 * See comments for nfs_proc_create regarding failed operations.
1638 nfs_mknod(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
, dev_t rdev
)
1643 dfprintk(VFS
, "NFS: mknod(%s/%ld), %s\n",
1644 dir
->i_sb
->s_id
, dir
->i_ino
, dentry
->d_name
.name
);
1646 if (!new_valid_dev(rdev
))
1649 attr
.ia_mode
= mode
;
1650 attr
.ia_valid
= ATTR_MODE
;
1652 status
= NFS_PROTO(dir
)->mknod(dir
, dentry
, &attr
, rdev
);
1662 * See comments for nfs_proc_create regarding failed operations.
1664 static int nfs_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
1669 dfprintk(VFS
, "NFS: mkdir(%s/%ld), %s\n",
1670 dir
->i_sb
->s_id
, dir
->i_ino
, dentry
->d_name
.name
);
1672 attr
.ia_valid
= ATTR_MODE
;
1673 attr
.ia_mode
= mode
| S_IFDIR
;
1675 error
= NFS_PROTO(dir
)->mkdir(dir
, dentry
, &attr
);
1684 static void nfs_dentry_handle_enoent(struct dentry
*dentry
)
1686 if (dentry
->d_inode
!= NULL
&& !d_unhashed(dentry
))
1690 static int nfs_rmdir(struct inode
*dir
, struct dentry
*dentry
)
1694 dfprintk(VFS
, "NFS: rmdir(%s/%ld), %s\n",
1695 dir
->i_sb
->s_id
, dir
->i_ino
, dentry
->d_name
.name
);
1697 error
= NFS_PROTO(dir
)->rmdir(dir
, &dentry
->d_name
);
1698 /* Ensure the VFS deletes this inode */
1699 if (error
== 0 && dentry
->d_inode
!= NULL
)
1700 clear_nlink(dentry
->d_inode
);
1701 else if (error
== -ENOENT
)
1702 nfs_dentry_handle_enoent(dentry
);
1708 * Remove a file after making sure there are no pending writes,
1709 * and after checking that the file has only one user.
1711 * We invalidate the attribute cache and free the inode prior to the operation
1712 * to avoid possible races if the server reuses the inode.
1714 static int nfs_safe_remove(struct dentry
*dentry
)
1716 struct inode
*dir
= dentry
->d_parent
->d_inode
;
1717 struct inode
*inode
= dentry
->d_inode
;
1720 dfprintk(VFS
, "NFS: safe_remove(%s/%s)\n",
1721 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
);
1723 /* If the dentry was sillyrenamed, we simply call d_delete() */
1724 if (dentry
->d_flags
& DCACHE_NFSFS_RENAMED
) {
1729 if (inode
!= NULL
) {
1730 nfs_inode_return_delegation(inode
);
1731 error
= NFS_PROTO(dir
)->remove(dir
, &dentry
->d_name
);
1732 /* The VFS may want to delete this inode */
1734 nfs_drop_nlink(inode
);
1735 nfs_mark_for_revalidate(inode
);
1737 error
= NFS_PROTO(dir
)->remove(dir
, &dentry
->d_name
);
1738 if (error
== -ENOENT
)
1739 nfs_dentry_handle_enoent(dentry
);
1744 /* We do silly rename. In case sillyrename() returns -EBUSY, the inode
1745 * belongs to an active ".nfs..." file and we return -EBUSY.
1747 * If sillyrename() returns 0, we do nothing, otherwise we unlink.
1749 static int nfs_unlink(struct inode
*dir
, struct dentry
*dentry
)
1752 int need_rehash
= 0;
1754 dfprintk(VFS
, "NFS: unlink(%s/%ld, %s)\n", dir
->i_sb
->s_id
,
1755 dir
->i_ino
, dentry
->d_name
.name
);
1757 spin_lock(&dentry
->d_lock
);
1758 if (dentry
->d_count
> 1) {
1759 spin_unlock(&dentry
->d_lock
);
1760 /* Start asynchronous writeout of the inode */
1761 write_inode_now(dentry
->d_inode
, 0);
1762 error
= nfs_sillyrename(dir
, dentry
);
1765 if (!d_unhashed(dentry
)) {
1769 spin_unlock(&dentry
->d_lock
);
1770 error
= nfs_safe_remove(dentry
);
1771 if (!error
|| error
== -ENOENT
) {
1772 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1773 } else if (need_rehash
)
1779 * To create a symbolic link, most file systems instantiate a new inode,
1780 * add a page to it containing the path, then write it out to the disk
1781 * using prepare_write/commit_write.
1783 * Unfortunately the NFS client can't create the in-core inode first
1784 * because it needs a file handle to create an in-core inode (see
1785 * fs/nfs/inode.c:nfs_fhget). We only have a file handle *after* the
1786 * symlink request has completed on the server.
1788 * So instead we allocate a raw page, copy the symname into it, then do
1789 * the SYMLINK request with the page as the buffer. If it succeeds, we
1790 * now have a new file handle and can instantiate an in-core NFS inode
1791 * and move the raw page into its mapping.
1793 static int nfs_symlink(struct inode
*dir
, struct dentry
*dentry
, const char *symname
)
1795 struct pagevec lru_pvec
;
1799 unsigned int pathlen
= strlen(symname
);
1802 dfprintk(VFS
, "NFS: symlink(%s/%ld, %s, %s)\n", dir
->i_sb
->s_id
,
1803 dir
->i_ino
, dentry
->d_name
.name
, symname
);
1805 if (pathlen
> PAGE_SIZE
)
1806 return -ENAMETOOLONG
;
1808 attr
.ia_mode
= S_IFLNK
| S_IRWXUGO
;
1809 attr
.ia_valid
= ATTR_MODE
;
1811 page
= alloc_page(GFP_HIGHUSER
);
1815 kaddr
= kmap_atomic(page
);
1816 memcpy(kaddr
, symname
, pathlen
);
1817 if (pathlen
< PAGE_SIZE
)
1818 memset(kaddr
+ pathlen
, 0, PAGE_SIZE
- pathlen
);
1819 kunmap_atomic(kaddr
);
1821 error
= NFS_PROTO(dir
)->symlink(dir
, dentry
, page
, pathlen
, &attr
);
1823 dfprintk(VFS
, "NFS: symlink(%s/%ld, %s, %s) error %d\n",
1824 dir
->i_sb
->s_id
, dir
->i_ino
,
1825 dentry
->d_name
.name
, symname
, error
);
1832 * No big deal if we can't add this page to the page cache here.
1833 * READLINK will get the missing page from the server if needed.
1835 pagevec_init(&lru_pvec
, 0);
1836 if (!add_to_page_cache(page
, dentry
->d_inode
->i_mapping
, 0,
1838 pagevec_add(&lru_pvec
, page
);
1839 pagevec_lru_add_file(&lru_pvec
);
1840 SetPageUptodate(page
);
1849 nfs_link(struct dentry
*old_dentry
, struct inode
*dir
, struct dentry
*dentry
)
1851 struct inode
*inode
= old_dentry
->d_inode
;
1854 dfprintk(VFS
, "NFS: link(%s/%s -> %s/%s)\n",
1855 old_dentry
->d_parent
->d_name
.name
, old_dentry
->d_name
.name
,
1856 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
);
1858 nfs_inode_return_delegation(inode
);
1861 error
= NFS_PROTO(dir
)->link(inode
, dir
, &dentry
->d_name
);
1864 d_add(dentry
, inode
);
1871 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a
1872 * different file handle for the same inode after a rename (e.g. when
1873 * moving to a different directory). A fail-safe method to do so would
1874 * be to look up old_dir/old_name, create a link to new_dir/new_name and
1875 * rename the old file using the sillyrename stuff. This way, the original
1876 * file in old_dir will go away when the last process iput()s the inode.
1880 * It actually works quite well. One needs to have the possibility for
1881 * at least one ".nfs..." file in each directory the file ever gets
1882 * moved or linked to which happens automagically with the new
1883 * implementation that only depends on the dcache stuff instead of
1884 * using the inode layer
1886 * Unfortunately, things are a little more complicated than indicated
1887 * above. For a cross-directory move, we want to make sure we can get
1888 * rid of the old inode after the operation. This means there must be
1889 * no pending writes (if it's a file), and the use count must be 1.
1890 * If these conditions are met, we can drop the dentries before doing
1893 static int nfs_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
1894 struct inode
*new_dir
, struct dentry
*new_dentry
)
1896 struct inode
*old_inode
= old_dentry
->d_inode
;
1897 struct inode
*new_inode
= new_dentry
->d_inode
;
1898 struct dentry
*dentry
= NULL
, *rehash
= NULL
;
1901 dfprintk(VFS
, "NFS: rename(%s/%s -> %s/%s, ct=%d)\n",
1902 old_dentry
->d_parent
->d_name
.name
, old_dentry
->d_name
.name
,
1903 new_dentry
->d_parent
->d_name
.name
, new_dentry
->d_name
.name
,
1904 new_dentry
->d_count
);
1907 * For non-directories, check whether the target is busy and if so,
1908 * make a copy of the dentry and then do a silly-rename. If the
1909 * silly-rename succeeds, the copied dentry is hashed and becomes
1912 if (new_inode
&& !S_ISDIR(new_inode
->i_mode
)) {
1914 * To prevent any new references to the target during the
1915 * rename, we unhash the dentry in advance.
1917 if (!d_unhashed(new_dentry
)) {
1919 rehash
= new_dentry
;
1922 if (new_dentry
->d_count
> 2) {
1925 /* copy the target dentry's name */
1926 dentry
= d_alloc(new_dentry
->d_parent
,
1927 &new_dentry
->d_name
);
1931 /* silly-rename the existing target ... */
1932 err
= nfs_sillyrename(new_dir
, new_dentry
);
1936 new_dentry
= dentry
;
1942 nfs_inode_return_delegation(old_inode
);
1943 if (new_inode
!= NULL
)
1944 nfs_inode_return_delegation(new_inode
);
1946 error
= NFS_PROTO(old_dir
)->rename(old_dir
, &old_dentry
->d_name
,
1947 new_dir
, &new_dentry
->d_name
);
1948 nfs_mark_for_revalidate(old_inode
);
1953 if (new_inode
!= NULL
)
1954 nfs_drop_nlink(new_inode
);
1955 d_move(old_dentry
, new_dentry
);
1956 nfs_set_verifier(new_dentry
,
1957 nfs_save_change_attribute(new_dir
));
1958 } else if (error
== -ENOENT
)
1959 nfs_dentry_handle_enoent(old_dentry
);
1961 /* new dentry created? */
1967 static DEFINE_SPINLOCK(nfs_access_lru_lock
);
1968 static LIST_HEAD(nfs_access_lru_list
);
1969 static atomic_long_t nfs_access_nr_entries
;
1971 static void nfs_access_free_entry(struct nfs_access_entry
*entry
)
1973 put_rpccred(entry
->cred
);
1975 smp_mb__before_atomic_dec();
1976 atomic_long_dec(&nfs_access_nr_entries
);
1977 smp_mb__after_atomic_dec();
1980 static void nfs_access_free_list(struct list_head
*head
)
1982 struct nfs_access_entry
*cache
;
1984 while (!list_empty(head
)) {
1985 cache
= list_entry(head
->next
, struct nfs_access_entry
, lru
);
1986 list_del(&cache
->lru
);
1987 nfs_access_free_entry(cache
);
1991 int nfs_access_cache_shrinker(struct shrinker
*shrink
,
1992 struct shrink_control
*sc
)
1995 struct nfs_inode
*nfsi
, *next
;
1996 struct nfs_access_entry
*cache
;
1997 int nr_to_scan
= sc
->nr_to_scan
;
1998 gfp_t gfp_mask
= sc
->gfp_mask
;
2000 if ((gfp_mask
& GFP_KERNEL
) != GFP_KERNEL
)
2001 return (nr_to_scan
== 0) ? 0 : -1;
2003 spin_lock(&nfs_access_lru_lock
);
2004 list_for_each_entry_safe(nfsi
, next
, &nfs_access_lru_list
, access_cache_inode_lru
) {
2005 struct inode
*inode
;
2007 if (nr_to_scan
-- == 0)
2009 inode
= &nfsi
->vfs_inode
;
2010 spin_lock(&inode
->i_lock
);
2011 if (list_empty(&nfsi
->access_cache_entry_lru
))
2012 goto remove_lru_entry
;
2013 cache
= list_entry(nfsi
->access_cache_entry_lru
.next
,
2014 struct nfs_access_entry
, lru
);
2015 list_move(&cache
->lru
, &head
);
2016 rb_erase(&cache
->rb_node
, &nfsi
->access_cache
);
2017 if (!list_empty(&nfsi
->access_cache_entry_lru
))
2018 list_move_tail(&nfsi
->access_cache_inode_lru
,
2019 &nfs_access_lru_list
);
2022 list_del_init(&nfsi
->access_cache_inode_lru
);
2023 smp_mb__before_clear_bit();
2024 clear_bit(NFS_INO_ACL_LRU_SET
, &nfsi
->flags
);
2025 smp_mb__after_clear_bit();
2027 spin_unlock(&inode
->i_lock
);
2029 spin_unlock(&nfs_access_lru_lock
);
2030 nfs_access_free_list(&head
);
2031 return (atomic_long_read(&nfs_access_nr_entries
) / 100) * sysctl_vfs_cache_pressure
;
2034 static void __nfs_access_zap_cache(struct nfs_inode
*nfsi
, struct list_head
*head
)
2036 struct rb_root
*root_node
= &nfsi
->access_cache
;
2038 struct nfs_access_entry
*entry
;
2040 /* Unhook entries from the cache */
2041 while ((n
= rb_first(root_node
)) != NULL
) {
2042 entry
= rb_entry(n
, struct nfs_access_entry
, rb_node
);
2043 rb_erase(n
, root_node
);
2044 list_move(&entry
->lru
, head
);
2046 nfsi
->cache_validity
&= ~NFS_INO_INVALID_ACCESS
;
2049 void nfs_access_zap_cache(struct inode
*inode
)
2053 if (test_bit(NFS_INO_ACL_LRU_SET
, &NFS_I(inode
)->flags
) == 0)
2055 /* Remove from global LRU init */
2056 spin_lock(&nfs_access_lru_lock
);
2057 if (test_and_clear_bit(NFS_INO_ACL_LRU_SET
, &NFS_I(inode
)->flags
))
2058 list_del_init(&NFS_I(inode
)->access_cache_inode_lru
);
2060 spin_lock(&inode
->i_lock
);
2061 __nfs_access_zap_cache(NFS_I(inode
), &head
);
2062 spin_unlock(&inode
->i_lock
);
2063 spin_unlock(&nfs_access_lru_lock
);
2064 nfs_access_free_list(&head
);
2067 static struct nfs_access_entry
*nfs_access_search_rbtree(struct inode
*inode
, struct rpc_cred
*cred
)
2069 struct rb_node
*n
= NFS_I(inode
)->access_cache
.rb_node
;
2070 struct nfs_access_entry
*entry
;
2073 entry
= rb_entry(n
, struct nfs_access_entry
, rb_node
);
2075 if (cred
< entry
->cred
)
2077 else if (cred
> entry
->cred
)
2085 static int nfs_access_get_cached(struct inode
*inode
, struct rpc_cred
*cred
, struct nfs_access_entry
*res
)
2087 struct nfs_inode
*nfsi
= NFS_I(inode
);
2088 struct nfs_access_entry
*cache
;
2091 spin_lock(&inode
->i_lock
);
2092 if (nfsi
->cache_validity
& NFS_INO_INVALID_ACCESS
)
2094 cache
= nfs_access_search_rbtree(inode
, cred
);
2097 if (!nfs_have_delegated_attributes(inode
) &&
2098 !time_in_range_open(jiffies
, cache
->jiffies
, cache
->jiffies
+ nfsi
->attrtimeo
))
2100 res
->jiffies
= cache
->jiffies
;
2101 res
->cred
= cache
->cred
;
2102 res
->mask
= cache
->mask
;
2103 list_move_tail(&cache
->lru
, &nfsi
->access_cache_entry_lru
);
2106 spin_unlock(&inode
->i_lock
);
2109 rb_erase(&cache
->rb_node
, &nfsi
->access_cache
);
2110 list_del(&cache
->lru
);
2111 spin_unlock(&inode
->i_lock
);
2112 nfs_access_free_entry(cache
);
2115 spin_unlock(&inode
->i_lock
);
2116 nfs_access_zap_cache(inode
);
2120 static void nfs_access_add_rbtree(struct inode
*inode
, struct nfs_access_entry
*set
)
2122 struct nfs_inode
*nfsi
= NFS_I(inode
);
2123 struct rb_root
*root_node
= &nfsi
->access_cache
;
2124 struct rb_node
**p
= &root_node
->rb_node
;
2125 struct rb_node
*parent
= NULL
;
2126 struct nfs_access_entry
*entry
;
2128 spin_lock(&inode
->i_lock
);
2129 while (*p
!= NULL
) {
2131 entry
= rb_entry(parent
, struct nfs_access_entry
, rb_node
);
2133 if (set
->cred
< entry
->cred
)
2134 p
= &parent
->rb_left
;
2135 else if (set
->cred
> entry
->cred
)
2136 p
= &parent
->rb_right
;
2140 rb_link_node(&set
->rb_node
, parent
, p
);
2141 rb_insert_color(&set
->rb_node
, root_node
);
2142 list_add_tail(&set
->lru
, &nfsi
->access_cache_entry_lru
);
2143 spin_unlock(&inode
->i_lock
);
2146 rb_replace_node(parent
, &set
->rb_node
, root_node
);
2147 list_add_tail(&set
->lru
, &nfsi
->access_cache_entry_lru
);
2148 list_del(&entry
->lru
);
2149 spin_unlock(&inode
->i_lock
);
2150 nfs_access_free_entry(entry
);
2153 static void nfs_access_add_cache(struct inode
*inode
, struct nfs_access_entry
*set
)
2155 struct nfs_access_entry
*cache
= kmalloc(sizeof(*cache
), GFP_KERNEL
);
2158 RB_CLEAR_NODE(&cache
->rb_node
);
2159 cache
->jiffies
= set
->jiffies
;
2160 cache
->cred
= get_rpccred(set
->cred
);
2161 cache
->mask
= set
->mask
;
2163 nfs_access_add_rbtree(inode
, cache
);
2165 /* Update accounting */
2166 smp_mb__before_atomic_inc();
2167 atomic_long_inc(&nfs_access_nr_entries
);
2168 smp_mb__after_atomic_inc();
2170 /* Add inode to global LRU list */
2171 if (!test_bit(NFS_INO_ACL_LRU_SET
, &NFS_I(inode
)->flags
)) {
2172 spin_lock(&nfs_access_lru_lock
);
2173 if (!test_and_set_bit(NFS_INO_ACL_LRU_SET
, &NFS_I(inode
)->flags
))
2174 list_add_tail(&NFS_I(inode
)->access_cache_inode_lru
,
2175 &nfs_access_lru_list
);
2176 spin_unlock(&nfs_access_lru_lock
);
2180 static int nfs_do_access(struct inode
*inode
, struct rpc_cred
*cred
, int mask
)
2182 struct nfs_access_entry cache
;
2185 status
= nfs_access_get_cached(inode
, cred
, &cache
);
2189 /* Be clever: ask server to check for all possible rights */
2190 cache
.mask
= MAY_EXEC
| MAY_WRITE
| MAY_READ
;
2192 cache
.jiffies
= jiffies
;
2193 status
= NFS_PROTO(inode
)->access(inode
, &cache
);
2195 if (status
== -ESTALE
) {
2196 nfs_zap_caches(inode
);
2197 if (!S_ISDIR(inode
->i_mode
))
2198 set_bit(NFS_INO_STALE
, &NFS_I(inode
)->flags
);
2202 nfs_access_add_cache(inode
, &cache
);
2204 if ((mask
& ~cache
.mask
& (MAY_READ
| MAY_WRITE
| MAY_EXEC
)) == 0)
2209 static int nfs_open_permission_mask(int openflags
)
2213 if ((openflags
& O_ACCMODE
) != O_WRONLY
)
2215 if ((openflags
& O_ACCMODE
) != O_RDONLY
)
2217 if (openflags
& __FMODE_EXEC
)
2222 int nfs_may_open(struct inode
*inode
, struct rpc_cred
*cred
, int openflags
)
2224 return nfs_do_access(inode
, cred
, nfs_open_permission_mask(openflags
));
2227 int nfs_permission(struct inode
*inode
, int mask
)
2229 struct rpc_cred
*cred
;
2232 if (mask
& MAY_NOT_BLOCK
)
2235 nfs_inc_stats(inode
, NFSIOS_VFSACCESS
);
2237 if ((mask
& (MAY_READ
| MAY_WRITE
| MAY_EXEC
)) == 0)
2239 /* Is this sys_access() ? */
2240 if (mask
& (MAY_ACCESS
| MAY_CHDIR
))
2243 switch (inode
->i_mode
& S_IFMT
) {
2247 /* NFSv4 has atomic_open... */
2248 if (nfs_server_capable(inode
, NFS_CAP_ATOMIC_OPEN
)
2249 && (mask
& MAY_OPEN
)
2250 && !(mask
& MAY_EXEC
))
2255 * Optimize away all write operations, since the server
2256 * will check permissions when we perform the op.
2258 if ((mask
& MAY_WRITE
) && !(mask
& MAY_READ
))
2263 if (!NFS_PROTO(inode
)->access
)
2266 cred
= rpc_lookup_cred();
2267 if (!IS_ERR(cred
)) {
2268 res
= nfs_do_access(inode
, cred
, mask
);
2271 res
= PTR_ERR(cred
);
2273 if (!res
&& (mask
& MAY_EXEC
) && !execute_ok(inode
))
2276 dfprintk(VFS
, "NFS: permission(%s/%ld), mask=0x%x, res=%d\n",
2277 inode
->i_sb
->s_id
, inode
->i_ino
, mask
, res
);
2280 res
= nfs_revalidate_inode(NFS_SERVER(inode
), inode
);
2282 res
= generic_permission(inode
, mask
);
2288 * version-control: t
2289 * kept-new-versions: 5