4 * Copyright (C) 1992 Rick Sladkey
6 * nfs directory handling functions
8 * 10 Apr 1996 Added silly rename for unlink --okir
9 * 28 Sep 1996 Improved directory cache --okir
10 * 23 Aug 1997 Claus Heine claus@momo.math.rwth-aachen.de
11 * Re-implemented silly rename for unlink, newly implemented
12 * silly rename for nfs_rename() following the suggestions
13 * of Olaf Kirch (okir) found in this file.
14 * Following Linus comments on my original hack, this version
15 * depends only on the dcache stuff and doesn't touch the inode
16 * layer (iput() and friends).
17 * 6 Jun 1999 Cache readdir lookups in the page cache. -DaveM
20 #include <linux/time.h>
21 #include <linux/errno.h>
22 #include <linux/stat.h>
23 #include <linux/fcntl.h>
24 #include <linux/string.h>
25 #include <linux/kernel.h>
26 #include <linux/slab.h>
28 #include <linux/sunrpc/clnt.h>
29 #include <linux/nfs_fs.h>
30 #include <linux/nfs_mount.h>
31 #include <linux/pagemap.h>
32 #include <linux/pagevec.h>
33 #include <linux/namei.h>
34 #include <linux/mount.h>
35 #include <linux/sched.h>
36 #include <linux/vmalloc.h>
37 #include <linux/kmemleak.h>
39 #include "delegation.h"
44 /* #define NFS_DEBUG_VERBOSE 1 */
46 static int nfs_opendir(struct inode
*, struct file
*);
47 static int nfs_readdir(struct file
*, void *, filldir_t
);
48 static struct dentry
*nfs_lookup(struct inode
*, struct dentry
*, struct nameidata
*);
49 static int nfs_create(struct inode
*, struct dentry
*, int, struct nameidata
*);
50 static int nfs_mkdir(struct inode
*, struct dentry
*, int);
51 static int nfs_rmdir(struct inode
*, struct dentry
*);
52 static int nfs_unlink(struct inode
*, struct dentry
*);
53 static int nfs_symlink(struct inode
*, struct dentry
*, const char *);
54 static int nfs_link(struct dentry
*, struct inode
*, struct dentry
*);
55 static int nfs_mknod(struct inode
*, struct dentry
*, int, dev_t
);
56 static int nfs_rename(struct inode
*, struct dentry
*,
57 struct inode
*, struct dentry
*);
58 static int nfs_fsync_dir(struct file
*, int);
59 static loff_t
nfs_llseek_dir(struct file
*, loff_t
, int);
60 static int nfs_readdir_clear_array(struct page
*, gfp_t
);
62 const struct file_operations nfs_dir_operations
= {
63 .llseek
= nfs_llseek_dir
,
64 .read
= generic_read_dir
,
65 .readdir
= nfs_readdir
,
67 .release
= nfs_release
,
68 .fsync
= nfs_fsync_dir
,
71 const struct inode_operations nfs_dir_inode_operations
= {
76 .symlink
= nfs_symlink
,
81 .permission
= nfs_permission
,
82 .getattr
= nfs_getattr
,
83 .setattr
= nfs_setattr
,
86 const struct address_space_operations nfs_dir_addr_space_ops
= {
87 .releasepage
= nfs_readdir_clear_array
,
91 const struct inode_operations nfs3_dir_inode_operations
= {
96 .symlink
= nfs_symlink
,
100 .rename
= nfs_rename
,
101 .permission
= nfs_permission
,
102 .getattr
= nfs_getattr
,
103 .setattr
= nfs_setattr
,
104 .listxattr
= nfs3_listxattr
,
105 .getxattr
= nfs3_getxattr
,
106 .setxattr
= nfs3_setxattr
,
107 .removexattr
= nfs3_removexattr
,
109 #endif /* CONFIG_NFS_V3 */
113 static struct dentry
*nfs_atomic_lookup(struct inode
*, struct dentry
*, struct nameidata
*);
114 static int nfs_open_create(struct inode
*dir
, struct dentry
*dentry
, int mode
, struct nameidata
*nd
);
115 const struct inode_operations nfs4_dir_inode_operations
= {
116 .create
= nfs_open_create
,
117 .lookup
= nfs_atomic_lookup
,
119 .unlink
= nfs_unlink
,
120 .symlink
= nfs_symlink
,
124 .rename
= nfs_rename
,
125 .permission
= nfs_permission
,
126 .getattr
= nfs_getattr
,
127 .setattr
= nfs_setattr
,
128 .getxattr
= nfs4_getxattr
,
129 .setxattr
= nfs4_setxattr
,
130 .listxattr
= nfs4_listxattr
,
133 #endif /* CONFIG_NFS_V4 */
139 nfs_opendir(struct inode
*inode
, struct file
*filp
)
143 dfprintk(FILE, "NFS: open dir(%s/%s)\n",
144 filp
->f_path
.dentry
->d_parent
->d_name
.name
,
145 filp
->f_path
.dentry
->d_name
.name
);
147 nfs_inc_stats(inode
, NFSIOS_VFSOPEN
);
149 /* Call generic open code in order to cache credentials */
150 res
= nfs_open(inode
, filp
);
151 if (filp
->f_path
.dentry
== filp
->f_path
.mnt
->mnt_root
) {
152 /* This is a mountpoint, so d_revalidate will never
153 * have been called, so we need to refresh the
154 * inode (for close-open consistency) ourselves.
156 __nfs_revalidate_inode(NFS_SERVER(inode
), inode
);
161 struct nfs_cache_array_entry
{
167 struct nfs_cache_array
{
171 struct nfs_cache_array_entry array
[0];
174 #define MAX_READDIR_ARRAY ((PAGE_SIZE - sizeof(struct nfs_cache_array)) / sizeof(struct nfs_cache_array_entry))
176 typedef __be32
* (*decode_dirent_t
)(struct xdr_stream
*, struct nfs_entry
*, struct nfs_server
*, int);
180 unsigned long page_index
;
182 loff_t current_index
;
183 decode_dirent_t decode
;
185 unsigned long timestamp
;
186 unsigned long gencount
;
187 unsigned int cache_entry_index
;
190 } nfs_readdir_descriptor_t
;
193 * The caller is responsible for calling nfs_readdir_release_array(page)
196 struct nfs_cache_array
*nfs_readdir_get_array(struct page
*page
)
200 return ERR_PTR(-EIO
);
203 return ERR_PTR(-ENOMEM
);
208 void nfs_readdir_release_array(struct page
*page
)
214 * we are freeing strings created by nfs_add_to_readdir_array()
217 int nfs_readdir_clear_array(struct page
*page
, gfp_t mask
)
219 struct nfs_cache_array
*array
= nfs_readdir_get_array(page
);
223 return PTR_ERR(array
);
224 for (i
= 0; i
< array
->size
; i
++)
225 kfree(array
->array
[i
].string
.name
);
226 nfs_readdir_release_array(page
);
231 * the caller is responsible for freeing qstr.name
232 * when called by nfs_readdir_add_to_array, the strings will be freed in
233 * nfs_clear_readdir_array()
236 int nfs_readdir_make_qstr(struct qstr
*string
, const char *name
, unsigned int len
)
239 string
->name
= kmemdup(name
, len
, GFP_KERNEL
);
240 if (string
->name
== NULL
)
243 * Avoid a kmemleak false positive. The pointer to the name is stored
244 * in a page cache page which kmemleak does not scan.
246 kmemleak_not_leak(string
->name
);
247 string
->hash
= full_name_hash(name
, len
);
252 int nfs_readdir_add_to_array(struct nfs_entry
*entry
, struct page
*page
)
254 struct nfs_cache_array
*array
= nfs_readdir_get_array(page
);
255 struct nfs_cache_array_entry
*cache_entry
;
259 return PTR_ERR(array
);
261 if (array
->size
>= MAX_READDIR_ARRAY
)
264 cache_entry
= &array
->array
[array
->size
];
265 cache_entry
->cookie
= entry
->prev_cookie
;
266 cache_entry
->ino
= entry
->ino
;
267 ret
= nfs_readdir_make_qstr(&cache_entry
->string
, entry
->name
, entry
->len
);
270 array
->last_cookie
= entry
->cookie
;
273 array
->eof_index
= array
->size
;
275 nfs_readdir_release_array(page
);
280 int nfs_readdir_search_for_pos(struct nfs_cache_array
*array
, nfs_readdir_descriptor_t
*desc
)
282 loff_t diff
= desc
->file
->f_pos
- desc
->current_index
;
287 if (diff
>= array
->size
) {
288 if (array
->eof_index
>= 0)
290 desc
->current_index
+= array
->size
;
294 index
= (unsigned int)diff
;
295 *desc
->dir_cookie
= array
->array
[index
].cookie
;
296 desc
->cache_entry_index
= index
;
304 int nfs_readdir_search_for_cookie(struct nfs_cache_array
*array
, nfs_readdir_descriptor_t
*desc
)
307 int status
= -EAGAIN
;
309 for (i
= 0; i
< array
->size
; i
++) {
310 if (array
->array
[i
].cookie
== *desc
->dir_cookie
) {
311 desc
->cache_entry_index
= i
;
316 if (i
== array
->eof_index
) {
318 status
= -EBADCOOKIE
;
325 int nfs_readdir_search_array(nfs_readdir_descriptor_t
*desc
)
327 struct nfs_cache_array
*array
;
328 int status
= -EBADCOOKIE
;
330 if (desc
->dir_cookie
== NULL
)
333 array
= nfs_readdir_get_array(desc
->page
);
335 status
= PTR_ERR(array
);
339 if (*desc
->dir_cookie
== 0)
340 status
= nfs_readdir_search_for_pos(array
, desc
);
342 status
= nfs_readdir_search_for_cookie(array
, desc
);
344 nfs_readdir_release_array(desc
->page
);
349 /* Fill a page with xdr information before transferring to the cache page */
351 int nfs_readdir_xdr_filler(struct page
**pages
, nfs_readdir_descriptor_t
*desc
,
352 struct nfs_entry
*entry
, struct file
*file
, struct inode
*inode
)
354 struct rpc_cred
*cred
= nfs_file_cred(file
);
355 unsigned long timestamp
, gencount
;
360 gencount
= nfs_inc_attr_generation_counter();
361 error
= NFS_PROTO(inode
)->readdir(file
->f_path
.dentry
, cred
, entry
->cookie
, pages
,
362 NFS_SERVER(inode
)->dtsize
, desc
->plus
);
364 /* We requested READDIRPLUS, but the server doesn't grok it */
365 if (error
== -ENOTSUPP
&& desc
->plus
) {
366 NFS_SERVER(inode
)->caps
&= ~NFS_CAP_READDIRPLUS
;
367 clear_bit(NFS_INO_ADVISE_RDPLUS
, &NFS_I(inode
)->flags
);
373 desc
->timestamp
= timestamp
;
374 desc
->gencount
= gencount
;
379 /* Fill in an entry based on the xdr code stored in desc->page */
381 int xdr_decode(nfs_readdir_descriptor_t
*desc
, struct nfs_entry
*entry
, struct xdr_stream
*stream
)
383 __be32
*p
= desc
->decode(stream
, entry
, NFS_SERVER(desc
->file
->f_path
.dentry
->d_inode
), desc
->plus
);
387 entry
->fattr
->time_start
= desc
->timestamp
;
388 entry
->fattr
->gencount
= desc
->gencount
;
393 int nfs_same_file(struct dentry
*dentry
, struct nfs_entry
*entry
)
395 struct nfs_inode
*node
;
396 if (dentry
->d_inode
== NULL
)
398 node
= NFS_I(dentry
->d_inode
);
399 if (node
->fh
.size
!= entry
->fh
->size
)
401 if (strncmp(node
->fh
.data
, entry
->fh
->data
, node
->fh
.size
) != 0)
409 void nfs_prime_dcache(struct dentry
*parent
, struct nfs_entry
*entry
)
411 struct qstr filename
= {
415 struct dentry
*dentry
;
416 struct dentry
*alias
;
417 struct inode
*dir
= parent
->d_inode
;
420 if (filename
.name
[0] == '.') {
421 if (filename
.len
== 1)
423 if (filename
.len
== 2 && filename
.name
[1] == '.')
426 filename
.hash
= full_name_hash(filename
.name
, filename
.len
);
428 dentry
= d_lookup(parent
, &filename
);
429 if (dentry
!= NULL
) {
430 if (nfs_same_file(dentry
, entry
)) {
431 nfs_refresh_inode(dentry
->d_inode
, entry
->fattr
);
439 dentry
= d_alloc(parent
, &filename
);
443 dentry
->d_op
= NFS_PROTO(dir
)->dentry_ops
;
444 inode
= nfs_fhget(dentry
->d_sb
, entry
->fh
, entry
->fattr
);
448 alias
= d_materialise_unique(dentry
, inode
);
452 nfs_set_verifier(alias
, nfs_save_change_attribute(dir
));
455 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
461 /* Perform conversion from xdr to cache array */
463 int nfs_readdir_page_filler(nfs_readdir_descriptor_t
*desc
, struct nfs_entry
*entry
,
464 void *xdr_page
, struct page
*page
, unsigned int buflen
)
466 struct xdr_stream stream
;
468 __be32
*ptr
= xdr_page
;
469 struct nfs_cache_array
*array
;
470 unsigned int count
= 0;
473 buf
.head
->iov_base
= xdr_page
;
474 buf
.head
->iov_len
= buflen
;
475 buf
.tail
->iov_len
= 0;
478 buf
.buflen
= buf
.head
->iov_len
;
479 buf
.len
= buf
.head
->iov_len
;
481 xdr_init_decode(&stream
, &buf
, ptr
);
485 status
= xdr_decode(desc
, entry
, &stream
);
487 if (status
== -EAGAIN
)
495 nfs_prime_dcache(desc
->file
->f_path
.dentry
, entry
);
497 status
= nfs_readdir_add_to_array(entry
, page
);
500 } while (!entry
->eof
);
502 if (count
== 0 || (status
== -EBADCOOKIE
&& entry
->eof
== 1)) {
503 array
= nfs_readdir_get_array(page
);
504 if (!IS_ERR(array
)) {
505 array
->eof_index
= array
->size
;
507 nfs_readdir_release_array(page
);
509 status
= PTR_ERR(array
);
515 void nfs_readdir_free_pagearray(struct page
**pages
, unsigned int npages
)
518 for (i
= 0; i
< npages
; i
++)
523 void nfs_readdir_free_large_page(void *ptr
, struct page
**pages
,
526 vm_unmap_ram(ptr
, npages
);
527 nfs_readdir_free_pagearray(pages
, npages
);
531 * nfs_readdir_large_page will allocate pages that must be freed with a call
532 * to nfs_readdir_free_large_page
535 void *nfs_readdir_large_page(struct page
**pages
, unsigned int npages
)
540 for (i
= 0; i
< npages
; i
++) {
541 struct page
*page
= alloc_page(GFP_KERNEL
);
547 ptr
= vm_map_ram(pages
, npages
, 0, PAGE_KERNEL
);
548 if (!IS_ERR_OR_NULL(ptr
))
551 nfs_readdir_free_pagearray(pages
, i
);
556 int nfs_readdir_xdr_to_array(nfs_readdir_descriptor_t
*desc
, struct page
*page
, struct inode
*inode
)
558 struct page
*pages
[NFS_MAX_READDIR_PAGES
];
559 void *pages_ptr
= NULL
;
560 struct nfs_entry entry
;
561 struct file
*file
= desc
->file
;
562 struct nfs_cache_array
*array
;
563 int status
= -ENOMEM
;
564 unsigned int array_size
= ARRAY_SIZE(pages
);
566 entry
.prev_cookie
= 0;
567 entry
.cookie
= *desc
->dir_cookie
;
569 entry
.fh
= nfs_alloc_fhandle();
570 entry
.fattr
= nfs_alloc_fattr();
571 if (entry
.fh
== NULL
|| entry
.fattr
== NULL
)
574 array
= nfs_readdir_get_array(page
);
576 status
= PTR_ERR(array
);
579 memset(array
, 0, sizeof(struct nfs_cache_array
));
580 array
->eof_index
= -1;
582 pages_ptr
= nfs_readdir_large_page(pages
, array_size
);
584 goto out_release_array
;
587 status
= nfs_readdir_xdr_filler(pages
, desc
, &entry
, file
, inode
);
592 status
= nfs_readdir_page_filler(desc
, &entry
, pages_ptr
, page
, pglen
);
594 if (status
== -ENOSPC
)
598 } while (array
->eof_index
< 0);
600 nfs_readdir_free_large_page(pages_ptr
, pages
, array_size
);
602 nfs_readdir_release_array(page
);
604 nfs_free_fattr(entry
.fattr
);
605 nfs_free_fhandle(entry
.fh
);
610 * Now we cache directories properly, by converting xdr information
611 * to an array that can be used for lookups later. This results in
612 * fewer cache pages, since we can store more information on each page.
613 * We only need to convert from xdr once so future lookups are much simpler
616 int nfs_readdir_filler(nfs_readdir_descriptor_t
*desc
, struct page
* page
)
618 struct inode
*inode
= desc
->file
->f_path
.dentry
->d_inode
;
621 ret
= nfs_readdir_xdr_to_array(desc
, page
, inode
);
624 SetPageUptodate(page
);
626 if (invalidate_inode_pages2_range(inode
->i_mapping
, page
->index
+ 1, -1) < 0) {
627 /* Should never happen */
628 nfs_zap_mapping(inode
, inode
->i_mapping
);
638 void cache_page_release(nfs_readdir_descriptor_t
*desc
)
640 page_cache_release(desc
->page
);
645 struct page
*get_cache_page(nfs_readdir_descriptor_t
*desc
)
647 return read_cache_page(desc
->file
->f_path
.dentry
->d_inode
->i_mapping
,
648 desc
->page_index
, (filler_t
*)nfs_readdir_filler
, desc
);
652 * Returns 0 if desc->dir_cookie was found on page desc->page_index
655 int find_cache_page(nfs_readdir_descriptor_t
*desc
)
659 desc
->page
= get_cache_page(desc
);
660 if (IS_ERR(desc
->page
))
661 return PTR_ERR(desc
->page
);
663 res
= nfs_readdir_search_array(desc
);
666 cache_page_release(desc
);
670 /* Search for desc->dir_cookie from the beginning of the page cache */
672 int readdir_search_pagecache(nfs_readdir_descriptor_t
*desc
)
676 if (desc
->page_index
== 0)
677 desc
->current_index
= 0;
679 res
= find_cache_page(desc
);
687 static inline unsigned int dt_type(struct inode
*inode
)
689 return (inode
->i_mode
>> 12) & 15;
693 * Once we've found the start of the dirent within a page: fill 'er up...
696 int nfs_do_filldir(nfs_readdir_descriptor_t
*desc
, void *dirent
,
699 struct file
*file
= desc
->file
;
702 struct nfs_cache_array
*array
= NULL
;
703 unsigned int d_type
= DT_UNKNOWN
;
705 array
= nfs_readdir_get_array(desc
->page
);
707 res
= PTR_ERR(array
);
711 for (i
= desc
->cache_entry_index
; i
< array
->size
; i
++) {
712 struct nfs_cache_array_entry
*ent
;
715 ent
= &array
->array
[i
];
716 if (filldir(dirent
, ent
->string
.name
, ent
->string
.len
,
717 file
->f_pos
, nfs_compat_user_ino64(ent
->ino
), d_type
) < 0) {
722 desc
->cache_entry_index
= i
;
723 if (i
< (array
->size
-1))
724 *desc
->dir_cookie
= array
->array
[i
+1].cookie
;
726 *desc
->dir_cookie
= array
->last_cookie
;
728 if (i
== array
->eof_index
)
731 nfs_readdir_release_array(desc
->page
);
733 cache_page_release(desc
);
734 dfprintk(DIRCACHE
, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n",
735 (unsigned long long)*desc
->dir_cookie
, res
);
740 * If we cannot find a cookie in our cache, we suspect that this is
741 * because it points to a deleted file, so we ask the server to return
742 * whatever it thinks is the next entry. We then feed this to filldir.
743 * If all goes well, we should then be able to find our way round the
744 * cache on the next call to readdir_search_pagecache();
746 * NOTE: we cannot add the anonymous page to the pagecache because
747 * the data it contains might not be page aligned. Besides,
748 * we should already have a complete representation of the
749 * directory in the page cache by the time we get here.
752 int uncached_readdir(nfs_readdir_descriptor_t
*desc
, void *dirent
,
755 struct page
*page
= NULL
;
757 struct inode
*inode
= desc
->file
->f_path
.dentry
->d_inode
;
759 dfprintk(DIRCACHE
, "NFS: uncached_readdir() searching for cookie %Lu\n",
760 (unsigned long long)*desc
->dir_cookie
);
762 page
= alloc_page(GFP_HIGHUSER
);
768 desc
->page_index
= 0;
771 status
= nfs_readdir_xdr_to_array(desc
, page
, inode
);
775 status
= nfs_do_filldir(desc
, dirent
, filldir
);
778 dfprintk(DIRCACHE
, "NFS: %s: returns %d\n",
782 cache_page_release(desc
);
786 /* The file offset position represents the dirent entry number. A
787 last cookie cache takes care of the common case of reading the
790 static int nfs_readdir(struct file
*filp
, void *dirent
, filldir_t filldir
)
792 struct dentry
*dentry
= filp
->f_path
.dentry
;
793 struct inode
*inode
= dentry
->d_inode
;
794 nfs_readdir_descriptor_t my_desc
,
798 dfprintk(FILE, "NFS: readdir(%s/%s) starting at cookie %llu\n",
799 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
,
800 (long long)filp
->f_pos
);
801 nfs_inc_stats(inode
, NFSIOS_VFSGETDENTS
);
804 * filp->f_pos points to the dirent entry number.
805 * *desc->dir_cookie has the cookie for the next entry. We have
806 * to either find the entry with the appropriate number or
807 * revalidate the cookie.
809 memset(desc
, 0, sizeof(*desc
));
812 desc
->dir_cookie
= &nfs_file_open_context(filp
)->dir_cookie
;
813 desc
->decode
= NFS_PROTO(inode
)->decode_dirent
;
814 desc
->plus
= NFS_USE_READDIRPLUS(inode
);
816 nfs_block_sillyrename(dentry
);
817 res
= nfs_revalidate_mapping(inode
, filp
->f_mapping
);
821 while (desc
->eof
!= 1) {
822 res
= readdir_search_pagecache(desc
);
824 if (res
== -EBADCOOKIE
) {
826 /* This means either end of directory */
827 if (*desc
->dir_cookie
&& desc
->eof
== 0) {
828 /* Or that the server has 'lost' a cookie */
829 res
= uncached_readdir(desc
, dirent
, filldir
);
835 if (res
== -ETOOSMALL
&& desc
->plus
) {
836 clear_bit(NFS_INO_ADVISE_RDPLUS
, &NFS_I(inode
)->flags
);
837 nfs_zap_caches(inode
);
838 desc
->page_index
= 0;
846 res
= nfs_do_filldir(desc
, dirent
, filldir
);
851 nfs_unblock_sillyrename(dentry
);
854 dfprintk(FILE, "NFS: readdir(%s/%s) returns %d\n",
855 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
,
860 static loff_t
nfs_llseek_dir(struct file
*filp
, loff_t offset
, int origin
)
862 struct dentry
*dentry
= filp
->f_path
.dentry
;
863 struct inode
*inode
= dentry
->d_inode
;
865 dfprintk(FILE, "NFS: llseek dir(%s/%s, %lld, %d)\n",
866 dentry
->d_parent
->d_name
.name
,
870 mutex_lock(&inode
->i_mutex
);
873 offset
+= filp
->f_pos
;
881 if (offset
!= filp
->f_pos
) {
882 filp
->f_pos
= offset
;
883 nfs_file_open_context(filp
)->dir_cookie
= 0;
886 mutex_unlock(&inode
->i_mutex
);
891 * All directory operations under NFS are synchronous, so fsync()
892 * is a dummy operation.
894 static int nfs_fsync_dir(struct file
*filp
, int datasync
)
896 struct dentry
*dentry
= filp
->f_path
.dentry
;
898 dfprintk(FILE, "NFS: fsync dir(%s/%s) datasync %d\n",
899 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
,
902 nfs_inc_stats(dentry
->d_inode
, NFSIOS_VFSFSYNC
);
907 * nfs_force_lookup_revalidate - Mark the directory as having changed
908 * @dir - pointer to directory inode
910 * This forces the revalidation code in nfs_lookup_revalidate() to do a
911 * full lookup on all child dentries of 'dir' whenever a change occurs
912 * on the server that might have invalidated our dcache.
914 * The caller should be holding dir->i_lock
916 void nfs_force_lookup_revalidate(struct inode
*dir
)
918 NFS_I(dir
)->cache_change_attribute
++;
922 * A check for whether or not the parent directory has changed.
923 * In the case it has, we assume that the dentries are untrustworthy
924 * and may need to be looked up again.
926 static int nfs_check_verifier(struct inode
*dir
, struct dentry
*dentry
)
930 if (NFS_SERVER(dir
)->flags
& NFS_MOUNT_LOOKUP_CACHE_NONE
)
932 if (!nfs_verify_change_attribute(dir
, dentry
->d_time
))
934 /* Revalidate nfsi->cache_change_attribute before we declare a match */
935 if (nfs_revalidate_inode(NFS_SERVER(dir
), dir
) < 0)
937 if (!nfs_verify_change_attribute(dir
, dentry
->d_time
))
943 * Return the intent data that applies to this particular path component
945 * Note that the current set of intents only apply to the very last
946 * component of the path.
947 * We check for this using LOOKUP_CONTINUE and LOOKUP_PARENT.
949 static inline unsigned int nfs_lookup_check_intent(struct nameidata
*nd
, unsigned int mask
)
951 if (nd
->flags
& (LOOKUP_CONTINUE
|LOOKUP_PARENT
))
953 return nd
->flags
& mask
;
957 * Use intent information to check whether or not we're going to do
958 * an O_EXCL create using this path component.
960 static int nfs_is_exclusive_create(struct inode
*dir
, struct nameidata
*nd
)
962 if (NFS_PROTO(dir
)->version
== 2)
964 return nd
&& nfs_lookup_check_intent(nd
, LOOKUP_EXCL
);
968 * Inode and filehandle revalidation for lookups.
970 * We force revalidation in the cases where the VFS sets LOOKUP_REVAL,
971 * or if the intent information indicates that we're about to open this
972 * particular file and the "nocto" mount flag is not set.
976 int nfs_lookup_verify_inode(struct inode
*inode
, struct nameidata
*nd
)
978 struct nfs_server
*server
= NFS_SERVER(inode
);
980 if (test_bit(NFS_INO_MOUNTPOINT
, &NFS_I(inode
)->flags
))
983 /* VFS wants an on-the-wire revalidation */
984 if (nd
->flags
& LOOKUP_REVAL
)
986 /* This is an open(2) */
987 if (nfs_lookup_check_intent(nd
, LOOKUP_OPEN
) != 0 &&
988 !(server
->flags
& NFS_MOUNT_NOCTO
) &&
989 (S_ISREG(inode
->i_mode
) ||
990 S_ISDIR(inode
->i_mode
)))
994 return nfs_revalidate_inode(server
, inode
);
996 return __nfs_revalidate_inode(server
, inode
);
1000 * We judge how long we want to trust negative
1001 * dentries by looking at the parent inode mtime.
1003 * If parent mtime has changed, we revalidate, else we wait for a
1004 * period corresponding to the parent's attribute cache timeout value.
1007 int nfs_neg_need_reval(struct inode
*dir
, struct dentry
*dentry
,
1008 struct nameidata
*nd
)
1010 /* Don't revalidate a negative dentry if we're creating a new file */
1011 if (nd
!= NULL
&& nfs_lookup_check_intent(nd
, LOOKUP_CREATE
) != 0)
1013 if (NFS_SERVER(dir
)->flags
& NFS_MOUNT_LOOKUP_CACHE_NONEG
)
1015 return !nfs_check_verifier(dir
, dentry
);
1019 * This is called every time the dcache has a lookup hit,
1020 * and we should check whether we can really trust that
1023 * NOTE! The hit can be a negative hit too, don't assume
1026 * If the parent directory is seen to have changed, we throw out the
1027 * cached dentry and do a new lookup.
1029 static int nfs_lookup_revalidate(struct dentry
* dentry
, struct nameidata
*nd
)
1032 struct inode
*inode
;
1033 struct dentry
*parent
;
1034 struct nfs_fh
*fhandle
= NULL
;
1035 struct nfs_fattr
*fattr
= NULL
;
1038 parent
= dget_parent(dentry
);
1039 dir
= parent
->d_inode
;
1040 nfs_inc_stats(dir
, NFSIOS_DENTRYREVALIDATE
);
1041 inode
= dentry
->d_inode
;
1044 if (nfs_neg_need_reval(dir
, dentry
, nd
))
1049 if (is_bad_inode(inode
)) {
1050 dfprintk(LOOKUPCACHE
, "%s: %s/%s has dud inode\n",
1051 __func__
, dentry
->d_parent
->d_name
.name
,
1052 dentry
->d_name
.name
);
1056 if (nfs_have_delegation(inode
, FMODE_READ
))
1057 goto out_set_verifier
;
1059 /* Force a full look up iff the parent directory has changed */
1060 if (!nfs_is_exclusive_create(dir
, nd
) && nfs_check_verifier(dir
, dentry
)) {
1061 if (nfs_lookup_verify_inode(inode
, nd
))
1062 goto out_zap_parent
;
1066 if (NFS_STALE(inode
))
1070 fhandle
= nfs_alloc_fhandle();
1071 fattr
= nfs_alloc_fattr();
1072 if (fhandle
== NULL
|| fattr
== NULL
)
1075 error
= NFS_PROTO(dir
)->lookup(dir
, &dentry
->d_name
, fhandle
, fattr
);
1078 if (nfs_compare_fh(NFS_FH(inode
), fhandle
))
1080 if ((error
= nfs_refresh_inode(inode
, fattr
)) != 0)
1083 nfs_free_fattr(fattr
);
1084 nfs_free_fhandle(fhandle
);
1086 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1089 dfprintk(LOOKUPCACHE
, "NFS: %s(%s/%s) is valid\n",
1090 __func__
, dentry
->d_parent
->d_name
.name
,
1091 dentry
->d_name
.name
);
1094 nfs_zap_caches(dir
);
1096 nfs_mark_for_revalidate(dir
);
1097 if (inode
&& S_ISDIR(inode
->i_mode
)) {
1098 /* Purge readdir caches. */
1099 nfs_zap_caches(inode
);
1100 /* If we have submounts, don't unhash ! */
1101 if (have_submounts(dentry
))
1103 if (dentry
->d_flags
& DCACHE_DISCONNECTED
)
1105 shrink_dcache_parent(dentry
);
1108 nfs_free_fattr(fattr
);
1109 nfs_free_fhandle(fhandle
);
1111 dfprintk(LOOKUPCACHE
, "NFS: %s(%s/%s) is invalid\n",
1112 __func__
, dentry
->d_parent
->d_name
.name
,
1113 dentry
->d_name
.name
);
1116 nfs_free_fattr(fattr
);
1117 nfs_free_fhandle(fhandle
);
1119 dfprintk(LOOKUPCACHE
, "NFS: %s(%s/%s) lookup returned error %d\n",
1120 __func__
, dentry
->d_parent
->d_name
.name
,
1121 dentry
->d_name
.name
, error
);
1126 * This is called from dput() when d_count is going to 0.
1128 static int nfs_dentry_delete(struct dentry
*dentry
)
1130 dfprintk(VFS
, "NFS: dentry_delete(%s/%s, %x)\n",
1131 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
,
1134 /* Unhash any dentry with a stale inode */
1135 if (dentry
->d_inode
!= NULL
&& NFS_STALE(dentry
->d_inode
))
1138 if (dentry
->d_flags
& DCACHE_NFSFS_RENAMED
) {
1139 /* Unhash it, so that ->d_iput() would be called */
1142 if (!(dentry
->d_sb
->s_flags
& MS_ACTIVE
)) {
1143 /* Unhash it, so that ancestors of killed async unlink
1144 * files will be cleaned up during umount */
1151 static void nfs_drop_nlink(struct inode
*inode
)
1153 spin_lock(&inode
->i_lock
);
1154 if (inode
->i_nlink
> 0)
1156 spin_unlock(&inode
->i_lock
);
1160 * Called when the dentry loses inode.
1161 * We use it to clean up silly-renamed files.
1163 static void nfs_dentry_iput(struct dentry
*dentry
, struct inode
*inode
)
1165 if (S_ISDIR(inode
->i_mode
))
1166 /* drop any readdir cache as it could easily be old */
1167 NFS_I(inode
)->cache_validity
|= NFS_INO_INVALID_DATA
;
1169 if (dentry
->d_flags
& DCACHE_NFSFS_RENAMED
) {
1171 nfs_complete_unlink(dentry
, inode
);
1176 const struct dentry_operations nfs_dentry_operations
= {
1177 .d_revalidate
= nfs_lookup_revalidate
,
1178 .d_delete
= nfs_dentry_delete
,
1179 .d_iput
= nfs_dentry_iput
,
1182 static struct dentry
*nfs_lookup(struct inode
*dir
, struct dentry
* dentry
, struct nameidata
*nd
)
1185 struct dentry
*parent
;
1186 struct inode
*inode
= NULL
;
1187 struct nfs_fh
*fhandle
= NULL
;
1188 struct nfs_fattr
*fattr
= NULL
;
1191 dfprintk(VFS
, "NFS: lookup(%s/%s)\n",
1192 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
);
1193 nfs_inc_stats(dir
, NFSIOS_VFSLOOKUP
);
1195 res
= ERR_PTR(-ENAMETOOLONG
);
1196 if (dentry
->d_name
.len
> NFS_SERVER(dir
)->namelen
)
1199 dentry
->d_op
= NFS_PROTO(dir
)->dentry_ops
;
1202 * If we're doing an exclusive create, optimize away the lookup
1203 * but don't hash the dentry.
1205 if (nfs_is_exclusive_create(dir
, nd
)) {
1206 d_instantiate(dentry
, NULL
);
1211 res
= ERR_PTR(-ENOMEM
);
1212 fhandle
= nfs_alloc_fhandle();
1213 fattr
= nfs_alloc_fattr();
1214 if (fhandle
== NULL
|| fattr
== NULL
)
1217 parent
= dentry
->d_parent
;
1218 /* Protect against concurrent sillydeletes */
1219 nfs_block_sillyrename(parent
);
1220 error
= NFS_PROTO(dir
)->lookup(dir
, &dentry
->d_name
, fhandle
, fattr
);
1221 if (error
== -ENOENT
)
1224 res
= ERR_PTR(error
);
1225 goto out_unblock_sillyrename
;
1227 inode
= nfs_fhget(dentry
->d_sb
, fhandle
, fattr
);
1228 res
= (struct dentry
*)inode
;
1230 goto out_unblock_sillyrename
;
1233 res
= d_materialise_unique(dentry
, inode
);
1236 goto out_unblock_sillyrename
;
1239 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1240 out_unblock_sillyrename
:
1241 nfs_unblock_sillyrename(parent
);
1243 nfs_free_fattr(fattr
);
1244 nfs_free_fhandle(fhandle
);
1248 #ifdef CONFIG_NFS_V4
1249 static int nfs_open_revalidate(struct dentry
*, struct nameidata
*);
1251 const struct dentry_operations nfs4_dentry_operations
= {
1252 .d_revalidate
= nfs_open_revalidate
,
1253 .d_delete
= nfs_dentry_delete
,
1254 .d_iput
= nfs_dentry_iput
,
1258 * Use intent information to determine whether we need to substitute
1259 * the NFSv4-style stateful OPEN for the LOOKUP call
1261 static int is_atomic_open(struct nameidata
*nd
)
1263 if (nd
== NULL
|| nfs_lookup_check_intent(nd
, LOOKUP_OPEN
) == 0)
1265 /* NFS does not (yet) have a stateful open for directories */
1266 if (nd
->flags
& LOOKUP_DIRECTORY
)
1268 /* Are we trying to write to a read only partition? */
1269 if (__mnt_is_readonly(nd
->path
.mnt
) &&
1270 (nd
->intent
.open
.flags
& (O_CREAT
|O_TRUNC
|FMODE_WRITE
)))
1275 static struct nfs_open_context
*nameidata_to_nfs_open_context(struct dentry
*dentry
, struct nameidata
*nd
)
1277 struct path path
= {
1278 .mnt
= nd
->path
.mnt
,
1281 struct nfs_open_context
*ctx
;
1282 struct rpc_cred
*cred
;
1283 fmode_t fmode
= nd
->intent
.open
.flags
& (FMODE_READ
| FMODE_WRITE
| FMODE_EXEC
);
1285 cred
= rpc_lookup_cred();
1287 return ERR_CAST(cred
);
1288 ctx
= alloc_nfs_open_context(&path
, cred
, fmode
);
1291 return ERR_PTR(-ENOMEM
);
1295 static int do_open(struct inode
*inode
, struct file
*filp
)
1297 nfs_fscache_set_inode_cookie(inode
, filp
);
1301 static int nfs_intent_set_file(struct nameidata
*nd
, struct nfs_open_context
*ctx
)
1306 /* If the open_intent is for execute, we have an extra check to make */
1307 if (ctx
->mode
& FMODE_EXEC
) {
1308 ret
= nfs_may_open(ctx
->path
.dentry
->d_inode
,
1310 nd
->intent
.open
.flags
);
1314 filp
= lookup_instantiate_filp(nd
, ctx
->path
.dentry
, do_open
);
1316 ret
= PTR_ERR(filp
);
1318 nfs_file_set_open_context(filp
, ctx
);
1320 put_nfs_open_context(ctx
);
1324 static struct dentry
*nfs_atomic_lookup(struct inode
*dir
, struct dentry
*dentry
, struct nameidata
*nd
)
1326 struct nfs_open_context
*ctx
;
1328 struct dentry
*res
= NULL
;
1329 struct inode
*inode
;
1333 dfprintk(VFS
, "NFS: atomic_lookup(%s/%ld), %s\n",
1334 dir
->i_sb
->s_id
, dir
->i_ino
, dentry
->d_name
.name
);
1336 /* Check that we are indeed trying to open this file */
1337 if (!is_atomic_open(nd
))
1340 if (dentry
->d_name
.len
> NFS_SERVER(dir
)->namelen
) {
1341 res
= ERR_PTR(-ENAMETOOLONG
);
1344 dentry
->d_op
= NFS_PROTO(dir
)->dentry_ops
;
1346 /* Let vfs_create() deal with O_EXCL. Instantiate, but don't hash
1348 if (nd
->flags
& LOOKUP_EXCL
) {
1349 d_instantiate(dentry
, NULL
);
1353 ctx
= nameidata_to_nfs_open_context(dentry
, nd
);
1354 res
= ERR_CAST(ctx
);
1358 open_flags
= nd
->intent
.open
.flags
;
1359 if (nd
->flags
& LOOKUP_CREATE
) {
1360 attr
.ia_mode
= nd
->intent
.open
.create_mode
;
1361 attr
.ia_valid
= ATTR_MODE
;
1362 if (!IS_POSIXACL(dir
))
1363 attr
.ia_mode
&= ~current_umask();
1365 open_flags
&= ~(O_EXCL
| O_CREAT
);
1369 /* Open the file on the server */
1370 nfs_block_sillyrename(dentry
->d_parent
);
1371 inode
= NFS_PROTO(dir
)->open_context(dir
, ctx
, open_flags
, &attr
);
1372 if (IS_ERR(inode
)) {
1373 nfs_unblock_sillyrename(dentry
->d_parent
);
1374 put_nfs_open_context(ctx
);
1375 switch (PTR_ERR(inode
)) {
1376 /* Make a negative dentry */
1378 d_add(dentry
, NULL
);
1381 /* This turned out not to be a regular file */
1385 if (!(nd
->intent
.open
.flags
& O_NOFOLLOW
))
1390 res
= ERR_CAST(inode
);
1394 res
= d_add_unique(dentry
, inode
);
1395 nfs_unblock_sillyrename(dentry
->d_parent
);
1397 dput(ctx
->path
.dentry
);
1398 ctx
->path
.dentry
= dget(res
);
1401 err
= nfs_intent_set_file(nd
, ctx
);
1405 return ERR_PTR(err
);
1408 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1411 return nfs_lookup(dir
, dentry
, nd
);
1414 static int nfs_open_revalidate(struct dentry
*dentry
, struct nameidata
*nd
)
1416 struct dentry
*parent
= NULL
;
1417 struct inode
*inode
= dentry
->d_inode
;
1419 struct nfs_open_context
*ctx
;
1420 int openflags
, ret
= 0;
1422 if (!is_atomic_open(nd
) || d_mountpoint(dentry
))
1425 parent
= dget_parent(dentry
);
1426 dir
= parent
->d_inode
;
1428 /* We can't create new files in nfs_open_revalidate(), so we
1429 * optimize away revalidation of negative dentries.
1431 if (inode
== NULL
) {
1432 if (!nfs_neg_need_reval(dir
, dentry
, nd
))
1437 /* NFS only supports OPEN on regular files */
1438 if (!S_ISREG(inode
->i_mode
))
1440 openflags
= nd
->intent
.open
.flags
;
1441 /* We cannot do exclusive creation on a positive dentry */
1442 if ((openflags
& (O_CREAT
|O_EXCL
)) == (O_CREAT
|O_EXCL
))
1444 /* We can't create new files, or truncate existing ones here */
1445 openflags
&= ~(O_CREAT
|O_EXCL
|O_TRUNC
);
1447 ctx
= nameidata_to_nfs_open_context(dentry
, nd
);
1452 * Note: we're not holding inode->i_mutex and so may be racing with
1453 * operations that change the directory. We therefore save the
1454 * change attribute *before* we do the RPC call.
1456 inode
= NFS_PROTO(dir
)->open_context(dir
, ctx
, openflags
, NULL
);
1457 if (IS_ERR(inode
)) {
1458 ret
= PTR_ERR(inode
);
1471 if (inode
!= dentry
->d_inode
)
1474 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1475 ret
= nfs_intent_set_file(nd
, ctx
);
1485 put_nfs_open_context(ctx
);
1491 return nfs_lookup_revalidate(dentry
, nd
);
1494 static int nfs_open_create(struct inode
*dir
, struct dentry
*dentry
, int mode
,
1495 struct nameidata
*nd
)
1497 struct nfs_open_context
*ctx
= NULL
;
1502 dfprintk(VFS
, "NFS: create(%s/%ld), %s\n",
1503 dir
->i_sb
->s_id
, dir
->i_ino
, dentry
->d_name
.name
);
1505 attr
.ia_mode
= mode
;
1506 attr
.ia_valid
= ATTR_MODE
;
1508 if ((nd
->flags
& LOOKUP_CREATE
) != 0) {
1509 open_flags
= nd
->intent
.open
.flags
;
1511 ctx
= nameidata_to_nfs_open_context(dentry
, nd
);
1512 error
= PTR_ERR(ctx
);
1517 error
= NFS_PROTO(dir
)->create(dir
, dentry
, &attr
, open_flags
, ctx
);
1521 error
= nfs_intent_set_file(nd
, ctx
);
1528 put_nfs_open_context(ctx
);
1535 #endif /* CONFIG_NFSV4 */
1538 * Code common to create, mkdir, and mknod.
1540 int nfs_instantiate(struct dentry
*dentry
, struct nfs_fh
*fhandle
,
1541 struct nfs_fattr
*fattr
)
1543 struct dentry
*parent
= dget_parent(dentry
);
1544 struct inode
*dir
= parent
->d_inode
;
1545 struct inode
*inode
;
1546 int error
= -EACCES
;
1550 /* We may have been initialized further down */
1551 if (dentry
->d_inode
)
1553 if (fhandle
->size
== 0) {
1554 error
= NFS_PROTO(dir
)->lookup(dir
, &dentry
->d_name
, fhandle
, fattr
);
1558 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1559 if (!(fattr
->valid
& NFS_ATTR_FATTR
)) {
1560 struct nfs_server
*server
= NFS_SB(dentry
->d_sb
);
1561 error
= server
->nfs_client
->rpc_ops
->getattr(server
, fhandle
, fattr
);
1565 inode
= nfs_fhget(dentry
->d_sb
, fhandle
, fattr
);
1566 error
= PTR_ERR(inode
);
1569 d_add(dentry
, inode
);
1574 nfs_mark_for_revalidate(dir
);
1580 * Following a failed create operation, we drop the dentry rather
1581 * than retain a negative dentry. This avoids a problem in the event
1582 * that the operation succeeded on the server, but an error in the
1583 * reply path made it appear to have failed.
1585 static int nfs_create(struct inode
*dir
, struct dentry
*dentry
, int mode
,
1586 struct nameidata
*nd
)
1591 dfprintk(VFS
, "NFS: create(%s/%ld), %s\n",
1592 dir
->i_sb
->s_id
, dir
->i_ino
, dentry
->d_name
.name
);
1594 attr
.ia_mode
= mode
;
1595 attr
.ia_valid
= ATTR_MODE
;
1597 error
= NFS_PROTO(dir
)->create(dir
, dentry
, &attr
, 0, NULL
);
1607 * See comments for nfs_proc_create regarding failed operations.
1610 nfs_mknod(struct inode
*dir
, struct dentry
*dentry
, int mode
, dev_t rdev
)
1615 dfprintk(VFS
, "NFS: mknod(%s/%ld), %s\n",
1616 dir
->i_sb
->s_id
, dir
->i_ino
, dentry
->d_name
.name
);
1618 if (!new_valid_dev(rdev
))
1621 attr
.ia_mode
= mode
;
1622 attr
.ia_valid
= ATTR_MODE
;
1624 status
= NFS_PROTO(dir
)->mknod(dir
, dentry
, &attr
, rdev
);
1634 * See comments for nfs_proc_create regarding failed operations.
1636 static int nfs_mkdir(struct inode
*dir
, struct dentry
*dentry
, int mode
)
1641 dfprintk(VFS
, "NFS: mkdir(%s/%ld), %s\n",
1642 dir
->i_sb
->s_id
, dir
->i_ino
, dentry
->d_name
.name
);
1644 attr
.ia_valid
= ATTR_MODE
;
1645 attr
.ia_mode
= mode
| S_IFDIR
;
1647 error
= NFS_PROTO(dir
)->mkdir(dir
, dentry
, &attr
);
1656 static void nfs_dentry_handle_enoent(struct dentry
*dentry
)
1658 if (dentry
->d_inode
!= NULL
&& !d_unhashed(dentry
))
1662 static int nfs_rmdir(struct inode
*dir
, struct dentry
*dentry
)
1666 dfprintk(VFS
, "NFS: rmdir(%s/%ld), %s\n",
1667 dir
->i_sb
->s_id
, dir
->i_ino
, dentry
->d_name
.name
);
1669 error
= NFS_PROTO(dir
)->rmdir(dir
, &dentry
->d_name
);
1670 /* Ensure the VFS deletes this inode */
1671 if (error
== 0 && dentry
->d_inode
!= NULL
)
1672 clear_nlink(dentry
->d_inode
);
1673 else if (error
== -ENOENT
)
1674 nfs_dentry_handle_enoent(dentry
);
1680 * Remove a file after making sure there are no pending writes,
1681 * and after checking that the file has only one user.
1683 * We invalidate the attribute cache and free the inode prior to the operation
1684 * to avoid possible races if the server reuses the inode.
1686 static int nfs_safe_remove(struct dentry
*dentry
)
1688 struct inode
*dir
= dentry
->d_parent
->d_inode
;
1689 struct inode
*inode
= dentry
->d_inode
;
1692 dfprintk(VFS
, "NFS: safe_remove(%s/%s)\n",
1693 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
);
1695 /* If the dentry was sillyrenamed, we simply call d_delete() */
1696 if (dentry
->d_flags
& DCACHE_NFSFS_RENAMED
) {
1701 if (inode
!= NULL
) {
1702 nfs_inode_return_delegation(inode
);
1703 error
= NFS_PROTO(dir
)->remove(dir
, &dentry
->d_name
);
1704 /* The VFS may want to delete this inode */
1706 nfs_drop_nlink(inode
);
1707 nfs_mark_for_revalidate(inode
);
1709 error
= NFS_PROTO(dir
)->remove(dir
, &dentry
->d_name
);
1710 if (error
== -ENOENT
)
1711 nfs_dentry_handle_enoent(dentry
);
1716 /* We do silly rename. In case sillyrename() returns -EBUSY, the inode
1717 * belongs to an active ".nfs..." file and we return -EBUSY.
1719 * If sillyrename() returns 0, we do nothing, otherwise we unlink.
1721 static int nfs_unlink(struct inode
*dir
, struct dentry
*dentry
)
1724 int need_rehash
= 0;
1726 dfprintk(VFS
, "NFS: unlink(%s/%ld, %s)\n", dir
->i_sb
->s_id
,
1727 dir
->i_ino
, dentry
->d_name
.name
);
1729 spin_lock(&dcache_lock
);
1730 spin_lock(&dentry
->d_lock
);
1731 if (atomic_read(&dentry
->d_count
) > 1) {
1732 spin_unlock(&dentry
->d_lock
);
1733 spin_unlock(&dcache_lock
);
1734 /* Start asynchronous writeout of the inode */
1735 write_inode_now(dentry
->d_inode
, 0);
1736 error
= nfs_sillyrename(dir
, dentry
);
1739 if (!d_unhashed(dentry
)) {
1743 spin_unlock(&dentry
->d_lock
);
1744 spin_unlock(&dcache_lock
);
1745 error
= nfs_safe_remove(dentry
);
1746 if (!error
|| error
== -ENOENT
) {
1747 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1748 } else if (need_rehash
)
1754 * To create a symbolic link, most file systems instantiate a new inode,
1755 * add a page to it containing the path, then write it out to the disk
1756 * using prepare_write/commit_write.
1758 * Unfortunately the NFS client can't create the in-core inode first
1759 * because it needs a file handle to create an in-core inode (see
1760 * fs/nfs/inode.c:nfs_fhget). We only have a file handle *after* the
1761 * symlink request has completed on the server.
1763 * So instead we allocate a raw page, copy the symname into it, then do
1764 * the SYMLINK request with the page as the buffer. If it succeeds, we
1765 * now have a new file handle and can instantiate an in-core NFS inode
1766 * and move the raw page into its mapping.
1768 static int nfs_symlink(struct inode
*dir
, struct dentry
*dentry
, const char *symname
)
1770 struct pagevec lru_pvec
;
1774 unsigned int pathlen
= strlen(symname
);
1777 dfprintk(VFS
, "NFS: symlink(%s/%ld, %s, %s)\n", dir
->i_sb
->s_id
,
1778 dir
->i_ino
, dentry
->d_name
.name
, symname
);
1780 if (pathlen
> PAGE_SIZE
)
1781 return -ENAMETOOLONG
;
1783 attr
.ia_mode
= S_IFLNK
| S_IRWXUGO
;
1784 attr
.ia_valid
= ATTR_MODE
;
1786 page
= alloc_page(GFP_HIGHUSER
);
1790 kaddr
= kmap_atomic(page
, KM_USER0
);
1791 memcpy(kaddr
, symname
, pathlen
);
1792 if (pathlen
< PAGE_SIZE
)
1793 memset(kaddr
+ pathlen
, 0, PAGE_SIZE
- pathlen
);
1794 kunmap_atomic(kaddr
, KM_USER0
);
1796 error
= NFS_PROTO(dir
)->symlink(dir
, dentry
, page
, pathlen
, &attr
);
1798 dfprintk(VFS
, "NFS: symlink(%s/%ld, %s, %s) error %d\n",
1799 dir
->i_sb
->s_id
, dir
->i_ino
,
1800 dentry
->d_name
.name
, symname
, error
);
1807 * No big deal if we can't add this page to the page cache here.
1808 * READLINK will get the missing page from the server if needed.
1810 pagevec_init(&lru_pvec
, 0);
1811 if (!add_to_page_cache(page
, dentry
->d_inode
->i_mapping
, 0,
1813 pagevec_add(&lru_pvec
, page
);
1814 pagevec_lru_add_file(&lru_pvec
);
1815 SetPageUptodate(page
);
1824 nfs_link(struct dentry
*old_dentry
, struct inode
*dir
, struct dentry
*dentry
)
1826 struct inode
*inode
= old_dentry
->d_inode
;
1829 dfprintk(VFS
, "NFS: link(%s/%s -> %s/%s)\n",
1830 old_dentry
->d_parent
->d_name
.name
, old_dentry
->d_name
.name
,
1831 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
);
1833 nfs_inode_return_delegation(inode
);
1836 error
= NFS_PROTO(dir
)->link(inode
, dir
, &dentry
->d_name
);
1839 d_add(dentry
, inode
);
1846 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a
1847 * different file handle for the same inode after a rename (e.g. when
1848 * moving to a different directory). A fail-safe method to do so would
1849 * be to look up old_dir/old_name, create a link to new_dir/new_name and
1850 * rename the old file using the sillyrename stuff. This way, the original
1851 * file in old_dir will go away when the last process iput()s the inode.
1855 * It actually works quite well. One needs to have the possibility for
1856 * at least one ".nfs..." file in each directory the file ever gets
1857 * moved or linked to which happens automagically with the new
1858 * implementation that only depends on the dcache stuff instead of
1859 * using the inode layer
1861 * Unfortunately, things are a little more complicated than indicated
1862 * above. For a cross-directory move, we want to make sure we can get
1863 * rid of the old inode after the operation. This means there must be
1864 * no pending writes (if it's a file), and the use count must be 1.
1865 * If these conditions are met, we can drop the dentries before doing
1868 static int nfs_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
1869 struct inode
*new_dir
, struct dentry
*new_dentry
)
1871 struct inode
*old_inode
= old_dentry
->d_inode
;
1872 struct inode
*new_inode
= new_dentry
->d_inode
;
1873 struct dentry
*dentry
= NULL
, *rehash
= NULL
;
1876 dfprintk(VFS
, "NFS: rename(%s/%s -> %s/%s, ct=%d)\n",
1877 old_dentry
->d_parent
->d_name
.name
, old_dentry
->d_name
.name
,
1878 new_dentry
->d_parent
->d_name
.name
, new_dentry
->d_name
.name
,
1879 atomic_read(&new_dentry
->d_count
));
1882 * For non-directories, check whether the target is busy and if so,
1883 * make a copy of the dentry and then do a silly-rename. If the
1884 * silly-rename succeeds, the copied dentry is hashed and becomes
1887 if (new_inode
&& !S_ISDIR(new_inode
->i_mode
)) {
1889 * To prevent any new references to the target during the
1890 * rename, we unhash the dentry in advance.
1892 if (!d_unhashed(new_dentry
)) {
1894 rehash
= new_dentry
;
1897 if (atomic_read(&new_dentry
->d_count
) > 2) {
1900 /* copy the target dentry's name */
1901 dentry
= d_alloc(new_dentry
->d_parent
,
1902 &new_dentry
->d_name
);
1906 /* silly-rename the existing target ... */
1907 err
= nfs_sillyrename(new_dir
, new_dentry
);
1911 new_dentry
= dentry
;
1917 nfs_inode_return_delegation(old_inode
);
1918 if (new_inode
!= NULL
)
1919 nfs_inode_return_delegation(new_inode
);
1921 error
= NFS_PROTO(old_dir
)->rename(old_dir
, &old_dentry
->d_name
,
1922 new_dir
, &new_dentry
->d_name
);
1923 nfs_mark_for_revalidate(old_inode
);
1928 if (new_inode
!= NULL
)
1929 nfs_drop_nlink(new_inode
);
1930 d_move(old_dentry
, new_dentry
);
1931 nfs_set_verifier(new_dentry
,
1932 nfs_save_change_attribute(new_dir
));
1933 } else if (error
== -ENOENT
)
1934 nfs_dentry_handle_enoent(old_dentry
);
1936 /* new dentry created? */
1942 static DEFINE_SPINLOCK(nfs_access_lru_lock
);
1943 static LIST_HEAD(nfs_access_lru_list
);
1944 static atomic_long_t nfs_access_nr_entries
;
1946 static void nfs_access_free_entry(struct nfs_access_entry
*entry
)
1948 put_rpccred(entry
->cred
);
1950 smp_mb__before_atomic_dec();
1951 atomic_long_dec(&nfs_access_nr_entries
);
1952 smp_mb__after_atomic_dec();
1955 static void nfs_access_free_list(struct list_head
*head
)
1957 struct nfs_access_entry
*cache
;
1959 while (!list_empty(head
)) {
1960 cache
= list_entry(head
->next
, struct nfs_access_entry
, lru
);
1961 list_del(&cache
->lru
);
1962 nfs_access_free_entry(cache
);
1966 int nfs_access_cache_shrinker(struct shrinker
*shrink
, int nr_to_scan
, gfp_t gfp_mask
)
1969 struct nfs_inode
*nfsi
, *next
;
1970 struct nfs_access_entry
*cache
;
1972 if ((gfp_mask
& GFP_KERNEL
) != GFP_KERNEL
)
1973 return (nr_to_scan
== 0) ? 0 : -1;
1975 spin_lock(&nfs_access_lru_lock
);
1976 list_for_each_entry_safe(nfsi
, next
, &nfs_access_lru_list
, access_cache_inode_lru
) {
1977 struct inode
*inode
;
1979 if (nr_to_scan
-- == 0)
1981 inode
= &nfsi
->vfs_inode
;
1982 spin_lock(&inode
->i_lock
);
1983 if (list_empty(&nfsi
->access_cache_entry_lru
))
1984 goto remove_lru_entry
;
1985 cache
= list_entry(nfsi
->access_cache_entry_lru
.next
,
1986 struct nfs_access_entry
, lru
);
1987 list_move(&cache
->lru
, &head
);
1988 rb_erase(&cache
->rb_node
, &nfsi
->access_cache
);
1989 if (!list_empty(&nfsi
->access_cache_entry_lru
))
1990 list_move_tail(&nfsi
->access_cache_inode_lru
,
1991 &nfs_access_lru_list
);
1994 list_del_init(&nfsi
->access_cache_inode_lru
);
1995 smp_mb__before_clear_bit();
1996 clear_bit(NFS_INO_ACL_LRU_SET
, &nfsi
->flags
);
1997 smp_mb__after_clear_bit();
1999 spin_unlock(&inode
->i_lock
);
2001 spin_unlock(&nfs_access_lru_lock
);
2002 nfs_access_free_list(&head
);
2003 return (atomic_long_read(&nfs_access_nr_entries
) / 100) * sysctl_vfs_cache_pressure
;
2006 static void __nfs_access_zap_cache(struct nfs_inode
*nfsi
, struct list_head
*head
)
2008 struct rb_root
*root_node
= &nfsi
->access_cache
;
2010 struct nfs_access_entry
*entry
;
2012 /* Unhook entries from the cache */
2013 while ((n
= rb_first(root_node
)) != NULL
) {
2014 entry
= rb_entry(n
, struct nfs_access_entry
, rb_node
);
2015 rb_erase(n
, root_node
);
2016 list_move(&entry
->lru
, head
);
2018 nfsi
->cache_validity
&= ~NFS_INO_INVALID_ACCESS
;
2021 void nfs_access_zap_cache(struct inode
*inode
)
2025 if (test_bit(NFS_INO_ACL_LRU_SET
, &NFS_I(inode
)->flags
) == 0)
2027 /* Remove from global LRU init */
2028 spin_lock(&nfs_access_lru_lock
);
2029 if (test_and_clear_bit(NFS_INO_ACL_LRU_SET
, &NFS_I(inode
)->flags
))
2030 list_del_init(&NFS_I(inode
)->access_cache_inode_lru
);
2032 spin_lock(&inode
->i_lock
);
2033 __nfs_access_zap_cache(NFS_I(inode
), &head
);
2034 spin_unlock(&inode
->i_lock
);
2035 spin_unlock(&nfs_access_lru_lock
);
2036 nfs_access_free_list(&head
);
2039 static struct nfs_access_entry
*nfs_access_search_rbtree(struct inode
*inode
, struct rpc_cred
*cred
)
2041 struct rb_node
*n
= NFS_I(inode
)->access_cache
.rb_node
;
2042 struct nfs_access_entry
*entry
;
2045 entry
= rb_entry(n
, struct nfs_access_entry
, rb_node
);
2047 if (cred
< entry
->cred
)
2049 else if (cred
> entry
->cred
)
2057 static int nfs_access_get_cached(struct inode
*inode
, struct rpc_cred
*cred
, struct nfs_access_entry
*res
)
2059 struct nfs_inode
*nfsi
= NFS_I(inode
);
2060 struct nfs_access_entry
*cache
;
2063 spin_lock(&inode
->i_lock
);
2064 if (nfsi
->cache_validity
& NFS_INO_INVALID_ACCESS
)
2066 cache
= nfs_access_search_rbtree(inode
, cred
);
2069 if (!nfs_have_delegated_attributes(inode
) &&
2070 !time_in_range_open(jiffies
, cache
->jiffies
, cache
->jiffies
+ nfsi
->attrtimeo
))
2072 res
->jiffies
= cache
->jiffies
;
2073 res
->cred
= cache
->cred
;
2074 res
->mask
= cache
->mask
;
2075 list_move_tail(&cache
->lru
, &nfsi
->access_cache_entry_lru
);
2078 spin_unlock(&inode
->i_lock
);
2081 rb_erase(&cache
->rb_node
, &nfsi
->access_cache
);
2082 list_del(&cache
->lru
);
2083 spin_unlock(&inode
->i_lock
);
2084 nfs_access_free_entry(cache
);
2087 spin_unlock(&inode
->i_lock
);
2088 nfs_access_zap_cache(inode
);
2092 static void nfs_access_add_rbtree(struct inode
*inode
, struct nfs_access_entry
*set
)
2094 struct nfs_inode
*nfsi
= NFS_I(inode
);
2095 struct rb_root
*root_node
= &nfsi
->access_cache
;
2096 struct rb_node
**p
= &root_node
->rb_node
;
2097 struct rb_node
*parent
= NULL
;
2098 struct nfs_access_entry
*entry
;
2100 spin_lock(&inode
->i_lock
);
2101 while (*p
!= NULL
) {
2103 entry
= rb_entry(parent
, struct nfs_access_entry
, rb_node
);
2105 if (set
->cred
< entry
->cred
)
2106 p
= &parent
->rb_left
;
2107 else if (set
->cred
> entry
->cred
)
2108 p
= &parent
->rb_right
;
2112 rb_link_node(&set
->rb_node
, parent
, p
);
2113 rb_insert_color(&set
->rb_node
, root_node
);
2114 list_add_tail(&set
->lru
, &nfsi
->access_cache_entry_lru
);
2115 spin_unlock(&inode
->i_lock
);
2118 rb_replace_node(parent
, &set
->rb_node
, root_node
);
2119 list_add_tail(&set
->lru
, &nfsi
->access_cache_entry_lru
);
2120 list_del(&entry
->lru
);
2121 spin_unlock(&inode
->i_lock
);
2122 nfs_access_free_entry(entry
);
2125 static void nfs_access_add_cache(struct inode
*inode
, struct nfs_access_entry
*set
)
2127 struct nfs_access_entry
*cache
= kmalloc(sizeof(*cache
), GFP_KERNEL
);
2130 RB_CLEAR_NODE(&cache
->rb_node
);
2131 cache
->jiffies
= set
->jiffies
;
2132 cache
->cred
= get_rpccred(set
->cred
);
2133 cache
->mask
= set
->mask
;
2135 nfs_access_add_rbtree(inode
, cache
);
2137 /* Update accounting */
2138 smp_mb__before_atomic_inc();
2139 atomic_long_inc(&nfs_access_nr_entries
);
2140 smp_mb__after_atomic_inc();
2142 /* Add inode to global LRU list */
2143 if (!test_bit(NFS_INO_ACL_LRU_SET
, &NFS_I(inode
)->flags
)) {
2144 spin_lock(&nfs_access_lru_lock
);
2145 if (!test_and_set_bit(NFS_INO_ACL_LRU_SET
, &NFS_I(inode
)->flags
))
2146 list_add_tail(&NFS_I(inode
)->access_cache_inode_lru
,
2147 &nfs_access_lru_list
);
2148 spin_unlock(&nfs_access_lru_lock
);
2152 static int nfs_do_access(struct inode
*inode
, struct rpc_cred
*cred
, int mask
)
2154 struct nfs_access_entry cache
;
2157 status
= nfs_access_get_cached(inode
, cred
, &cache
);
2161 /* Be clever: ask server to check for all possible rights */
2162 cache
.mask
= MAY_EXEC
| MAY_WRITE
| MAY_READ
;
2164 cache
.jiffies
= jiffies
;
2165 status
= NFS_PROTO(inode
)->access(inode
, &cache
);
2167 if (status
== -ESTALE
) {
2168 nfs_zap_caches(inode
);
2169 if (!S_ISDIR(inode
->i_mode
))
2170 set_bit(NFS_INO_STALE
, &NFS_I(inode
)->flags
);
2174 nfs_access_add_cache(inode
, &cache
);
2176 if ((mask
& ~cache
.mask
& (MAY_READ
| MAY_WRITE
| MAY_EXEC
)) == 0)
2181 static int nfs_open_permission_mask(int openflags
)
2185 if (openflags
& FMODE_READ
)
2187 if (openflags
& FMODE_WRITE
)
2189 if (openflags
& FMODE_EXEC
)
2194 int nfs_may_open(struct inode
*inode
, struct rpc_cred
*cred
, int openflags
)
2196 return nfs_do_access(inode
, cred
, nfs_open_permission_mask(openflags
));
2199 int nfs_permission(struct inode
*inode
, int mask
)
2201 struct rpc_cred
*cred
;
2204 nfs_inc_stats(inode
, NFSIOS_VFSACCESS
);
2206 if ((mask
& (MAY_READ
| MAY_WRITE
| MAY_EXEC
)) == 0)
2208 /* Is this sys_access() ? */
2209 if (mask
& (MAY_ACCESS
| MAY_CHDIR
))
2212 switch (inode
->i_mode
& S_IFMT
) {
2216 /* NFSv4 has atomic_open... */
2217 if (nfs_server_capable(inode
, NFS_CAP_ATOMIC_OPEN
)
2218 && (mask
& MAY_OPEN
)
2219 && !(mask
& MAY_EXEC
))
2224 * Optimize away all write operations, since the server
2225 * will check permissions when we perform the op.
2227 if ((mask
& MAY_WRITE
) && !(mask
& MAY_READ
))
2232 if (!NFS_PROTO(inode
)->access
)
2235 cred
= rpc_lookup_cred();
2236 if (!IS_ERR(cred
)) {
2237 res
= nfs_do_access(inode
, cred
, mask
);
2240 res
= PTR_ERR(cred
);
2242 if (!res
&& (mask
& MAY_EXEC
) && !execute_ok(inode
))
2245 dfprintk(VFS
, "NFS: permission(%s/%ld), mask=0x%x, res=%d\n",
2246 inode
->i_sb
->s_id
, inode
->i_ino
, mask
, res
);
2249 res
= nfs_revalidate_inode(NFS_SERVER(inode
), inode
);
2251 res
= generic_permission(inode
, mask
, NULL
);
2257 * version-control: t
2258 * kept-new-versions: 5