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
{
165 unsigned char d_type
;
168 struct nfs_cache_array
{
172 struct nfs_cache_array_entry array
[0];
175 typedef __be32
* (*decode_dirent_t
)(struct xdr_stream
*, struct nfs_entry
*, struct nfs_server
*, int);
179 unsigned long page_index
;
181 loff_t current_index
;
182 decode_dirent_t decode
;
184 unsigned long timestamp
;
185 unsigned long gencount
;
186 unsigned int cache_entry_index
;
189 } nfs_readdir_descriptor_t
;
192 * The caller is responsible for calling nfs_readdir_release_array(page)
195 struct nfs_cache_array
*nfs_readdir_get_array(struct page
*page
)
199 return ERR_PTR(-EIO
);
202 return ERR_PTR(-ENOMEM
);
207 void nfs_readdir_release_array(struct page
*page
)
213 * we are freeing strings created by nfs_add_to_readdir_array()
216 int nfs_readdir_clear_array(struct page
*page
, gfp_t mask
)
218 struct nfs_cache_array
*array
= nfs_readdir_get_array(page
);
222 return PTR_ERR(array
);
223 for (i
= 0; i
< array
->size
; i
++)
224 kfree(array
->array
[i
].string
.name
);
225 nfs_readdir_release_array(page
);
230 * the caller is responsible for freeing qstr.name
231 * when called by nfs_readdir_add_to_array, the strings will be freed in
232 * nfs_clear_readdir_array()
235 int nfs_readdir_make_qstr(struct qstr
*string
, const char *name
, unsigned int len
)
238 string
->name
= kmemdup(name
, len
, GFP_KERNEL
);
239 if (string
->name
== NULL
)
242 * Avoid a kmemleak false positive. The pointer to the name is stored
243 * in a page cache page which kmemleak does not scan.
245 kmemleak_not_leak(string
->name
);
246 string
->hash
= full_name_hash(name
, len
);
251 int nfs_readdir_add_to_array(struct nfs_entry
*entry
, struct page
*page
)
253 struct nfs_cache_array
*array
= nfs_readdir_get_array(page
);
254 struct nfs_cache_array_entry
*cache_entry
;
258 return PTR_ERR(array
);
260 cache_entry
= &array
->array
[array
->size
];
262 /* Check that this entry lies within the page bounds */
264 if ((char *)&cache_entry
[1] - (char *)page_address(page
) > PAGE_SIZE
)
267 cache_entry
->cookie
= entry
->prev_cookie
;
268 cache_entry
->ino
= entry
->ino
;
269 cache_entry
->d_type
= entry
->d_type
;
270 ret
= nfs_readdir_make_qstr(&cache_entry
->string
, entry
->name
, entry
->len
);
273 array
->last_cookie
= entry
->cookie
;
276 array
->eof_index
= array
->size
;
278 nfs_readdir_release_array(page
);
283 int nfs_readdir_search_for_pos(struct nfs_cache_array
*array
, nfs_readdir_descriptor_t
*desc
)
285 loff_t diff
= desc
->file
->f_pos
- desc
->current_index
;
290 if (diff
>= array
->size
) {
291 if (array
->eof_index
>= 0)
293 desc
->current_index
+= array
->size
;
297 index
= (unsigned int)diff
;
298 *desc
->dir_cookie
= array
->array
[index
].cookie
;
299 desc
->cache_entry_index
= index
;
307 int nfs_readdir_search_for_cookie(struct nfs_cache_array
*array
, nfs_readdir_descriptor_t
*desc
)
310 int status
= -EAGAIN
;
312 for (i
= 0; i
< array
->size
; i
++) {
313 if (array
->array
[i
].cookie
== *desc
->dir_cookie
) {
314 desc
->cache_entry_index
= i
;
319 if (i
== array
->eof_index
) {
321 status
= -EBADCOOKIE
;
328 int nfs_readdir_search_array(nfs_readdir_descriptor_t
*desc
)
330 struct nfs_cache_array
*array
;
331 int status
= -EBADCOOKIE
;
333 if (desc
->dir_cookie
== NULL
)
336 array
= nfs_readdir_get_array(desc
->page
);
338 status
= PTR_ERR(array
);
342 if (*desc
->dir_cookie
== 0)
343 status
= nfs_readdir_search_for_pos(array
, desc
);
345 status
= nfs_readdir_search_for_cookie(array
, desc
);
347 nfs_readdir_release_array(desc
->page
);
352 /* Fill a page with xdr information before transferring to the cache page */
354 int nfs_readdir_xdr_filler(struct page
**pages
, nfs_readdir_descriptor_t
*desc
,
355 struct nfs_entry
*entry
, struct file
*file
, struct inode
*inode
)
357 struct rpc_cred
*cred
= nfs_file_cred(file
);
358 unsigned long timestamp
, gencount
;
363 gencount
= nfs_inc_attr_generation_counter();
364 error
= NFS_PROTO(inode
)->readdir(file
->f_path
.dentry
, cred
, entry
->cookie
, pages
,
365 NFS_SERVER(inode
)->dtsize
, desc
->plus
);
367 /* We requested READDIRPLUS, but the server doesn't grok it */
368 if (error
== -ENOTSUPP
&& desc
->plus
) {
369 NFS_SERVER(inode
)->caps
&= ~NFS_CAP_READDIRPLUS
;
370 clear_bit(NFS_INO_ADVISE_RDPLUS
, &NFS_I(inode
)->flags
);
376 desc
->timestamp
= timestamp
;
377 desc
->gencount
= gencount
;
382 /* Fill in an entry based on the xdr code stored in desc->page */
384 int xdr_decode(nfs_readdir_descriptor_t
*desc
, struct nfs_entry
*entry
, struct xdr_stream
*stream
)
386 __be32
*p
= desc
->decode(stream
, entry
, NFS_SERVER(desc
->file
->f_path
.dentry
->d_inode
), desc
->plus
);
390 entry
->fattr
->time_start
= desc
->timestamp
;
391 entry
->fattr
->gencount
= desc
->gencount
;
396 int nfs_same_file(struct dentry
*dentry
, struct nfs_entry
*entry
)
398 struct nfs_inode
*node
;
399 if (dentry
->d_inode
== NULL
)
401 node
= NFS_I(dentry
->d_inode
);
402 if (node
->fh
.size
!= entry
->fh
->size
)
404 if (strncmp(node
->fh
.data
, entry
->fh
->data
, node
->fh
.size
) != 0)
412 void nfs_prime_dcache(struct dentry
*parent
, struct nfs_entry
*entry
)
414 struct qstr filename
= {
418 struct dentry
*dentry
;
419 struct dentry
*alias
;
420 struct inode
*dir
= parent
->d_inode
;
423 if (filename
.name
[0] == '.') {
424 if (filename
.len
== 1)
426 if (filename
.len
== 2 && filename
.name
[1] == '.')
429 filename
.hash
= full_name_hash(filename
.name
, filename
.len
);
431 dentry
= d_lookup(parent
, &filename
);
432 if (dentry
!= NULL
) {
433 if (nfs_same_file(dentry
, entry
)) {
434 nfs_refresh_inode(dentry
->d_inode
, entry
->fattr
);
442 dentry
= d_alloc(parent
, &filename
);
446 dentry
->d_op
= NFS_PROTO(dir
)->dentry_ops
;
447 inode
= nfs_fhget(dentry
->d_sb
, entry
->fh
, entry
->fattr
);
451 alias
= d_materialise_unique(dentry
, inode
);
455 nfs_set_verifier(alias
, nfs_save_change_attribute(dir
));
458 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
464 /* Perform conversion from xdr to cache array */
466 int nfs_readdir_page_filler(nfs_readdir_descriptor_t
*desc
, struct nfs_entry
*entry
,
467 void *xdr_page
, struct page
*page
, unsigned int buflen
)
469 struct xdr_stream stream
;
471 __be32
*ptr
= xdr_page
;
472 struct nfs_cache_array
*array
;
473 unsigned int count
= 0;
476 buf
.head
->iov_base
= xdr_page
;
477 buf
.head
->iov_len
= buflen
;
478 buf
.tail
->iov_len
= 0;
481 buf
.buflen
= buf
.head
->iov_len
;
482 buf
.len
= buf
.head
->iov_len
;
484 xdr_init_decode(&stream
, &buf
, ptr
);
488 status
= xdr_decode(desc
, entry
, &stream
);
490 if (status
== -EAGAIN
)
498 nfs_prime_dcache(desc
->file
->f_path
.dentry
, entry
);
500 status
= nfs_readdir_add_to_array(entry
, page
);
503 } while (!entry
->eof
);
505 if (count
== 0 || (status
== -EBADCOOKIE
&& entry
->eof
== 1)) {
506 array
= nfs_readdir_get_array(page
);
507 if (!IS_ERR(array
)) {
508 array
->eof_index
= array
->size
;
510 nfs_readdir_release_array(page
);
512 status
= PTR_ERR(array
);
518 void nfs_readdir_free_pagearray(struct page
**pages
, unsigned int npages
)
521 for (i
= 0; i
< npages
; i
++)
526 void nfs_readdir_free_large_page(void *ptr
, struct page
**pages
,
529 vm_unmap_ram(ptr
, npages
);
530 nfs_readdir_free_pagearray(pages
, npages
);
534 * nfs_readdir_large_page will allocate pages that must be freed with a call
535 * to nfs_readdir_free_large_page
538 void *nfs_readdir_large_page(struct page
**pages
, unsigned int npages
)
543 for (i
= 0; i
< npages
; i
++) {
544 struct page
*page
= alloc_page(GFP_KERNEL
);
550 ptr
= vm_map_ram(pages
, npages
, 0, PAGE_KERNEL
);
551 if (!IS_ERR_OR_NULL(ptr
))
554 nfs_readdir_free_pagearray(pages
, i
);
559 int nfs_readdir_xdr_to_array(nfs_readdir_descriptor_t
*desc
, struct page
*page
, struct inode
*inode
)
561 struct page
*pages
[NFS_MAX_READDIR_PAGES
];
562 void *pages_ptr
= NULL
;
563 struct nfs_entry entry
;
564 struct file
*file
= desc
->file
;
565 struct nfs_cache_array
*array
;
566 int status
= -ENOMEM
;
567 unsigned int array_size
= ARRAY_SIZE(pages
);
569 entry
.prev_cookie
= 0;
570 entry
.cookie
= *desc
->dir_cookie
;
572 entry
.fh
= nfs_alloc_fhandle();
573 entry
.fattr
= nfs_alloc_fattr();
574 if (entry
.fh
== NULL
|| entry
.fattr
== NULL
)
577 array
= nfs_readdir_get_array(page
);
579 status
= PTR_ERR(array
);
582 memset(array
, 0, sizeof(struct nfs_cache_array
));
583 array
->eof_index
= -1;
585 pages_ptr
= nfs_readdir_large_page(pages
, array_size
);
587 goto out_release_array
;
590 status
= nfs_readdir_xdr_filler(pages
, desc
, &entry
, file
, inode
);
595 status
= nfs_readdir_page_filler(desc
, &entry
, pages_ptr
, page
, pglen
);
597 if (status
== -ENOSPC
)
601 } while (array
->eof_index
< 0);
603 nfs_readdir_free_large_page(pages_ptr
, pages
, array_size
);
605 nfs_readdir_release_array(page
);
607 nfs_free_fattr(entry
.fattr
);
608 nfs_free_fhandle(entry
.fh
);
613 * Now we cache directories properly, by converting xdr information
614 * to an array that can be used for lookups later. This results in
615 * fewer cache pages, since we can store more information on each page.
616 * We only need to convert from xdr once so future lookups are much simpler
619 int nfs_readdir_filler(nfs_readdir_descriptor_t
*desc
, struct page
* page
)
621 struct inode
*inode
= desc
->file
->f_path
.dentry
->d_inode
;
624 ret
= nfs_readdir_xdr_to_array(desc
, page
, inode
);
627 SetPageUptodate(page
);
629 if (invalidate_inode_pages2_range(inode
->i_mapping
, page
->index
+ 1, -1) < 0) {
630 /* Should never happen */
631 nfs_zap_mapping(inode
, inode
->i_mapping
);
641 void cache_page_release(nfs_readdir_descriptor_t
*desc
)
643 page_cache_release(desc
->page
);
648 struct page
*get_cache_page(nfs_readdir_descriptor_t
*desc
)
650 return read_cache_page(desc
->file
->f_path
.dentry
->d_inode
->i_mapping
,
651 desc
->page_index
, (filler_t
*)nfs_readdir_filler
, desc
);
655 * Returns 0 if desc->dir_cookie was found on page desc->page_index
658 int find_cache_page(nfs_readdir_descriptor_t
*desc
)
662 desc
->page
= get_cache_page(desc
);
663 if (IS_ERR(desc
->page
))
664 return PTR_ERR(desc
->page
);
666 res
= nfs_readdir_search_array(desc
);
669 cache_page_release(desc
);
673 /* Search for desc->dir_cookie from the beginning of the page cache */
675 int readdir_search_pagecache(nfs_readdir_descriptor_t
*desc
)
679 if (desc
->page_index
== 0)
680 desc
->current_index
= 0;
682 res
= find_cache_page(desc
);
690 static inline unsigned int dt_type(struct inode
*inode
)
692 return (inode
->i_mode
>> 12) & 15;
696 * Once we've found the start of the dirent within a page: fill 'er up...
699 int nfs_do_filldir(nfs_readdir_descriptor_t
*desc
, void *dirent
,
702 struct file
*file
= desc
->file
;
705 struct nfs_cache_array
*array
= NULL
;
707 array
= nfs_readdir_get_array(desc
->page
);
709 res
= PTR_ERR(array
);
713 for (i
= desc
->cache_entry_index
; i
< array
->size
; i
++) {
714 struct nfs_cache_array_entry
*ent
;
716 ent
= &array
->array
[i
];
717 if (filldir(dirent
, ent
->string
.name
, ent
->string
.len
,
718 file
->f_pos
, nfs_compat_user_ino64(ent
->ino
),
724 desc
->cache_entry_index
= i
;
725 if (i
< (array
->size
-1))
726 *desc
->dir_cookie
= array
->array
[i
+1].cookie
;
728 *desc
->dir_cookie
= array
->last_cookie
;
730 if (i
== array
->eof_index
)
733 nfs_readdir_release_array(desc
->page
);
735 cache_page_release(desc
);
736 dfprintk(DIRCACHE
, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n",
737 (unsigned long long)*desc
->dir_cookie
, res
);
742 * If we cannot find a cookie in our cache, we suspect that this is
743 * because it points to a deleted file, so we ask the server to return
744 * whatever it thinks is the next entry. We then feed this to filldir.
745 * If all goes well, we should then be able to find our way round the
746 * cache on the next call to readdir_search_pagecache();
748 * NOTE: we cannot add the anonymous page to the pagecache because
749 * the data it contains might not be page aligned. Besides,
750 * we should already have a complete representation of the
751 * directory in the page cache by the time we get here.
754 int uncached_readdir(nfs_readdir_descriptor_t
*desc
, void *dirent
,
757 struct page
*page
= NULL
;
759 struct inode
*inode
= desc
->file
->f_path
.dentry
->d_inode
;
761 dfprintk(DIRCACHE
, "NFS: uncached_readdir() searching for cookie %Lu\n",
762 (unsigned long long)*desc
->dir_cookie
);
764 page
= alloc_page(GFP_HIGHUSER
);
770 desc
->page_index
= 0;
773 status
= nfs_readdir_xdr_to_array(desc
, page
, inode
);
777 status
= nfs_do_filldir(desc
, dirent
, filldir
);
780 dfprintk(DIRCACHE
, "NFS: %s: returns %d\n",
784 cache_page_release(desc
);
788 /* The file offset position represents the dirent entry number. A
789 last cookie cache takes care of the common case of reading the
792 static int nfs_readdir(struct file
*filp
, void *dirent
, filldir_t filldir
)
794 struct dentry
*dentry
= filp
->f_path
.dentry
;
795 struct inode
*inode
= dentry
->d_inode
;
796 nfs_readdir_descriptor_t my_desc
,
800 dfprintk(FILE, "NFS: readdir(%s/%s) starting at cookie %llu\n",
801 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
,
802 (long long)filp
->f_pos
);
803 nfs_inc_stats(inode
, NFSIOS_VFSGETDENTS
);
806 * filp->f_pos points to the dirent entry number.
807 * *desc->dir_cookie has the cookie for the next entry. We have
808 * to either find the entry with the appropriate number or
809 * revalidate the cookie.
811 memset(desc
, 0, sizeof(*desc
));
814 desc
->dir_cookie
= &nfs_file_open_context(filp
)->dir_cookie
;
815 desc
->decode
= NFS_PROTO(inode
)->decode_dirent
;
816 desc
->plus
= NFS_USE_READDIRPLUS(inode
);
818 nfs_block_sillyrename(dentry
);
819 res
= nfs_revalidate_mapping(inode
, filp
->f_mapping
);
823 while (desc
->eof
!= 1) {
824 res
= readdir_search_pagecache(desc
);
826 if (res
== -EBADCOOKIE
) {
828 /* This means either end of directory */
829 if (*desc
->dir_cookie
&& desc
->eof
== 0) {
830 /* Or that the server has 'lost' a cookie */
831 res
= uncached_readdir(desc
, dirent
, filldir
);
837 if (res
== -ETOOSMALL
&& desc
->plus
) {
838 clear_bit(NFS_INO_ADVISE_RDPLUS
, &NFS_I(inode
)->flags
);
839 nfs_zap_caches(inode
);
840 desc
->page_index
= 0;
848 res
= nfs_do_filldir(desc
, dirent
, filldir
);
853 nfs_unblock_sillyrename(dentry
);
856 dfprintk(FILE, "NFS: readdir(%s/%s) returns %d\n",
857 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
,
862 static loff_t
nfs_llseek_dir(struct file
*filp
, loff_t offset
, int origin
)
864 struct dentry
*dentry
= filp
->f_path
.dentry
;
865 struct inode
*inode
= dentry
->d_inode
;
867 dfprintk(FILE, "NFS: llseek dir(%s/%s, %lld, %d)\n",
868 dentry
->d_parent
->d_name
.name
,
872 mutex_lock(&inode
->i_mutex
);
875 offset
+= filp
->f_pos
;
883 if (offset
!= filp
->f_pos
) {
884 filp
->f_pos
= offset
;
885 nfs_file_open_context(filp
)->dir_cookie
= 0;
888 mutex_unlock(&inode
->i_mutex
);
893 * All directory operations under NFS are synchronous, so fsync()
894 * is a dummy operation.
896 static int nfs_fsync_dir(struct file
*filp
, int datasync
)
898 struct dentry
*dentry
= filp
->f_path
.dentry
;
900 dfprintk(FILE, "NFS: fsync dir(%s/%s) datasync %d\n",
901 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
,
904 nfs_inc_stats(dentry
->d_inode
, NFSIOS_VFSFSYNC
);
909 * nfs_force_lookup_revalidate - Mark the directory as having changed
910 * @dir - pointer to directory inode
912 * This forces the revalidation code in nfs_lookup_revalidate() to do a
913 * full lookup on all child dentries of 'dir' whenever a change occurs
914 * on the server that might have invalidated our dcache.
916 * The caller should be holding dir->i_lock
918 void nfs_force_lookup_revalidate(struct inode
*dir
)
920 NFS_I(dir
)->cache_change_attribute
++;
924 * A check for whether or not the parent directory has changed.
925 * In the case it has, we assume that the dentries are untrustworthy
926 * and may need to be looked up again.
928 static int nfs_check_verifier(struct inode
*dir
, struct dentry
*dentry
)
932 if (NFS_SERVER(dir
)->flags
& NFS_MOUNT_LOOKUP_CACHE_NONE
)
934 if (!nfs_verify_change_attribute(dir
, dentry
->d_time
))
936 /* Revalidate nfsi->cache_change_attribute before we declare a match */
937 if (nfs_revalidate_inode(NFS_SERVER(dir
), dir
) < 0)
939 if (!nfs_verify_change_attribute(dir
, dentry
->d_time
))
945 * Return the intent data that applies to this particular path component
947 * Note that the current set of intents only apply to the very last
948 * component of the path.
949 * We check for this using LOOKUP_CONTINUE and LOOKUP_PARENT.
951 static inline unsigned int nfs_lookup_check_intent(struct nameidata
*nd
, unsigned int mask
)
953 if (nd
->flags
& (LOOKUP_CONTINUE
|LOOKUP_PARENT
))
955 return nd
->flags
& mask
;
959 * Use intent information to check whether or not we're going to do
960 * an O_EXCL create using this path component.
962 static int nfs_is_exclusive_create(struct inode
*dir
, struct nameidata
*nd
)
964 if (NFS_PROTO(dir
)->version
== 2)
966 return nd
&& nfs_lookup_check_intent(nd
, LOOKUP_EXCL
);
970 * Inode and filehandle revalidation for lookups.
972 * We force revalidation in the cases where the VFS sets LOOKUP_REVAL,
973 * or if the intent information indicates that we're about to open this
974 * particular file and the "nocto" mount flag is not set.
978 int nfs_lookup_verify_inode(struct inode
*inode
, struct nameidata
*nd
)
980 struct nfs_server
*server
= NFS_SERVER(inode
);
982 if (test_bit(NFS_INO_MOUNTPOINT
, &NFS_I(inode
)->flags
))
985 /* VFS wants an on-the-wire revalidation */
986 if (nd
->flags
& LOOKUP_REVAL
)
988 /* This is an open(2) */
989 if (nfs_lookup_check_intent(nd
, LOOKUP_OPEN
) != 0 &&
990 !(server
->flags
& NFS_MOUNT_NOCTO
) &&
991 (S_ISREG(inode
->i_mode
) ||
992 S_ISDIR(inode
->i_mode
)))
996 return nfs_revalidate_inode(server
, inode
);
998 return __nfs_revalidate_inode(server
, inode
);
1002 * We judge how long we want to trust negative
1003 * dentries by looking at the parent inode mtime.
1005 * If parent mtime has changed, we revalidate, else we wait for a
1006 * period corresponding to the parent's attribute cache timeout value.
1009 int nfs_neg_need_reval(struct inode
*dir
, struct dentry
*dentry
,
1010 struct nameidata
*nd
)
1012 /* Don't revalidate a negative dentry if we're creating a new file */
1013 if (nd
!= NULL
&& nfs_lookup_check_intent(nd
, LOOKUP_CREATE
) != 0)
1015 if (NFS_SERVER(dir
)->flags
& NFS_MOUNT_LOOKUP_CACHE_NONEG
)
1017 return !nfs_check_verifier(dir
, dentry
);
1021 * This is called every time the dcache has a lookup hit,
1022 * and we should check whether we can really trust that
1025 * NOTE! The hit can be a negative hit too, don't assume
1028 * If the parent directory is seen to have changed, we throw out the
1029 * cached dentry and do a new lookup.
1031 static int nfs_lookup_revalidate(struct dentry
* dentry
, struct nameidata
*nd
)
1034 struct inode
*inode
;
1035 struct dentry
*parent
;
1036 struct nfs_fh
*fhandle
= NULL
;
1037 struct nfs_fattr
*fattr
= NULL
;
1040 parent
= dget_parent(dentry
);
1041 dir
= parent
->d_inode
;
1042 nfs_inc_stats(dir
, NFSIOS_DENTRYREVALIDATE
);
1043 inode
= dentry
->d_inode
;
1046 if (nfs_neg_need_reval(dir
, dentry
, nd
))
1051 if (is_bad_inode(inode
)) {
1052 dfprintk(LOOKUPCACHE
, "%s: %s/%s has dud inode\n",
1053 __func__
, dentry
->d_parent
->d_name
.name
,
1054 dentry
->d_name
.name
);
1058 if (nfs_have_delegation(inode
, FMODE_READ
))
1059 goto out_set_verifier
;
1061 /* Force a full look up iff the parent directory has changed */
1062 if (!nfs_is_exclusive_create(dir
, nd
) && nfs_check_verifier(dir
, dentry
)) {
1063 if (nfs_lookup_verify_inode(inode
, nd
))
1064 goto out_zap_parent
;
1068 if (NFS_STALE(inode
))
1072 fhandle
= nfs_alloc_fhandle();
1073 fattr
= nfs_alloc_fattr();
1074 if (fhandle
== NULL
|| fattr
== NULL
)
1077 error
= NFS_PROTO(dir
)->lookup(dir
, &dentry
->d_name
, fhandle
, fattr
);
1080 if (nfs_compare_fh(NFS_FH(inode
), fhandle
))
1082 if ((error
= nfs_refresh_inode(inode
, fattr
)) != 0)
1085 nfs_free_fattr(fattr
);
1086 nfs_free_fhandle(fhandle
);
1088 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1091 dfprintk(LOOKUPCACHE
, "NFS: %s(%s/%s) is valid\n",
1092 __func__
, dentry
->d_parent
->d_name
.name
,
1093 dentry
->d_name
.name
);
1096 nfs_zap_caches(dir
);
1098 nfs_mark_for_revalidate(dir
);
1099 if (inode
&& S_ISDIR(inode
->i_mode
)) {
1100 /* Purge readdir caches. */
1101 nfs_zap_caches(inode
);
1102 /* If we have submounts, don't unhash ! */
1103 if (have_submounts(dentry
))
1105 if (dentry
->d_flags
& DCACHE_DISCONNECTED
)
1107 shrink_dcache_parent(dentry
);
1110 nfs_free_fattr(fattr
);
1111 nfs_free_fhandle(fhandle
);
1113 dfprintk(LOOKUPCACHE
, "NFS: %s(%s/%s) is invalid\n",
1114 __func__
, dentry
->d_parent
->d_name
.name
,
1115 dentry
->d_name
.name
);
1118 nfs_free_fattr(fattr
);
1119 nfs_free_fhandle(fhandle
);
1121 dfprintk(LOOKUPCACHE
, "NFS: %s(%s/%s) lookup returned error %d\n",
1122 __func__
, dentry
->d_parent
->d_name
.name
,
1123 dentry
->d_name
.name
, error
);
1128 * This is called from dput() when d_count is going to 0.
1130 static int nfs_dentry_delete(struct dentry
*dentry
)
1132 dfprintk(VFS
, "NFS: dentry_delete(%s/%s, %x)\n",
1133 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
,
1136 /* Unhash any dentry with a stale inode */
1137 if (dentry
->d_inode
!= NULL
&& NFS_STALE(dentry
->d_inode
))
1140 if (dentry
->d_flags
& DCACHE_NFSFS_RENAMED
) {
1141 /* Unhash it, so that ->d_iput() would be called */
1144 if (!(dentry
->d_sb
->s_flags
& MS_ACTIVE
)) {
1145 /* Unhash it, so that ancestors of killed async unlink
1146 * files will be cleaned up during umount */
1153 static void nfs_drop_nlink(struct inode
*inode
)
1155 spin_lock(&inode
->i_lock
);
1156 if (inode
->i_nlink
> 0)
1158 spin_unlock(&inode
->i_lock
);
1162 * Called when the dentry loses inode.
1163 * We use it to clean up silly-renamed files.
1165 static void nfs_dentry_iput(struct dentry
*dentry
, struct inode
*inode
)
1167 if (S_ISDIR(inode
->i_mode
))
1168 /* drop any readdir cache as it could easily be old */
1169 NFS_I(inode
)->cache_validity
|= NFS_INO_INVALID_DATA
;
1171 if (dentry
->d_flags
& DCACHE_NFSFS_RENAMED
) {
1173 nfs_complete_unlink(dentry
, inode
);
1178 const struct dentry_operations nfs_dentry_operations
= {
1179 .d_revalidate
= nfs_lookup_revalidate
,
1180 .d_delete
= nfs_dentry_delete
,
1181 .d_iput
= nfs_dentry_iput
,
1184 static struct dentry
*nfs_lookup(struct inode
*dir
, struct dentry
* dentry
, struct nameidata
*nd
)
1187 struct dentry
*parent
;
1188 struct inode
*inode
= NULL
;
1189 struct nfs_fh
*fhandle
= NULL
;
1190 struct nfs_fattr
*fattr
= NULL
;
1193 dfprintk(VFS
, "NFS: lookup(%s/%s)\n",
1194 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
);
1195 nfs_inc_stats(dir
, NFSIOS_VFSLOOKUP
);
1197 res
= ERR_PTR(-ENAMETOOLONG
);
1198 if (dentry
->d_name
.len
> NFS_SERVER(dir
)->namelen
)
1201 dentry
->d_op
= NFS_PROTO(dir
)->dentry_ops
;
1204 * If we're doing an exclusive create, optimize away the lookup
1205 * but don't hash the dentry.
1207 if (nfs_is_exclusive_create(dir
, nd
)) {
1208 d_instantiate(dentry
, NULL
);
1213 res
= ERR_PTR(-ENOMEM
);
1214 fhandle
= nfs_alloc_fhandle();
1215 fattr
= nfs_alloc_fattr();
1216 if (fhandle
== NULL
|| fattr
== NULL
)
1219 parent
= dentry
->d_parent
;
1220 /* Protect against concurrent sillydeletes */
1221 nfs_block_sillyrename(parent
);
1222 error
= NFS_PROTO(dir
)->lookup(dir
, &dentry
->d_name
, fhandle
, fattr
);
1223 if (error
== -ENOENT
)
1226 res
= ERR_PTR(error
);
1227 goto out_unblock_sillyrename
;
1229 inode
= nfs_fhget(dentry
->d_sb
, fhandle
, fattr
);
1230 res
= (struct dentry
*)inode
;
1232 goto out_unblock_sillyrename
;
1235 res
= d_materialise_unique(dentry
, inode
);
1238 goto out_unblock_sillyrename
;
1241 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1242 out_unblock_sillyrename
:
1243 nfs_unblock_sillyrename(parent
);
1245 nfs_free_fattr(fattr
);
1246 nfs_free_fhandle(fhandle
);
1250 #ifdef CONFIG_NFS_V4
1251 static int nfs_open_revalidate(struct dentry
*, struct nameidata
*);
1253 const struct dentry_operations nfs4_dentry_operations
= {
1254 .d_revalidate
= nfs_open_revalidate
,
1255 .d_delete
= nfs_dentry_delete
,
1256 .d_iput
= nfs_dentry_iput
,
1260 * Use intent information to determine whether we need to substitute
1261 * the NFSv4-style stateful OPEN for the LOOKUP call
1263 static int is_atomic_open(struct nameidata
*nd
)
1265 if (nd
== NULL
|| nfs_lookup_check_intent(nd
, LOOKUP_OPEN
) == 0)
1267 /* NFS does not (yet) have a stateful open for directories */
1268 if (nd
->flags
& LOOKUP_DIRECTORY
)
1270 /* Are we trying to write to a read only partition? */
1271 if (__mnt_is_readonly(nd
->path
.mnt
) &&
1272 (nd
->intent
.open
.flags
& (O_CREAT
|O_TRUNC
|FMODE_WRITE
)))
1277 static struct nfs_open_context
*nameidata_to_nfs_open_context(struct dentry
*dentry
, struct nameidata
*nd
)
1279 struct path path
= {
1280 .mnt
= nd
->path
.mnt
,
1283 struct nfs_open_context
*ctx
;
1284 struct rpc_cred
*cred
;
1285 fmode_t fmode
= nd
->intent
.open
.flags
& (FMODE_READ
| FMODE_WRITE
| FMODE_EXEC
);
1287 cred
= rpc_lookup_cred();
1289 return ERR_CAST(cred
);
1290 ctx
= alloc_nfs_open_context(&path
, cred
, fmode
);
1293 return ERR_PTR(-ENOMEM
);
1297 static int do_open(struct inode
*inode
, struct file
*filp
)
1299 nfs_fscache_set_inode_cookie(inode
, filp
);
1303 static int nfs_intent_set_file(struct nameidata
*nd
, struct nfs_open_context
*ctx
)
1308 /* If the open_intent is for execute, we have an extra check to make */
1309 if (ctx
->mode
& FMODE_EXEC
) {
1310 ret
= nfs_may_open(ctx
->path
.dentry
->d_inode
,
1312 nd
->intent
.open
.flags
);
1316 filp
= lookup_instantiate_filp(nd
, ctx
->path
.dentry
, do_open
);
1318 ret
= PTR_ERR(filp
);
1320 nfs_file_set_open_context(filp
, ctx
);
1322 put_nfs_open_context(ctx
);
1326 static struct dentry
*nfs_atomic_lookup(struct inode
*dir
, struct dentry
*dentry
, struct nameidata
*nd
)
1328 struct nfs_open_context
*ctx
;
1330 struct dentry
*res
= NULL
;
1331 struct inode
*inode
;
1335 dfprintk(VFS
, "NFS: atomic_lookup(%s/%ld), %s\n",
1336 dir
->i_sb
->s_id
, dir
->i_ino
, dentry
->d_name
.name
);
1338 /* Check that we are indeed trying to open this file */
1339 if (!is_atomic_open(nd
))
1342 if (dentry
->d_name
.len
> NFS_SERVER(dir
)->namelen
) {
1343 res
= ERR_PTR(-ENAMETOOLONG
);
1346 dentry
->d_op
= NFS_PROTO(dir
)->dentry_ops
;
1348 /* Let vfs_create() deal with O_EXCL. Instantiate, but don't hash
1350 if (nd
->flags
& LOOKUP_EXCL
) {
1351 d_instantiate(dentry
, NULL
);
1355 ctx
= nameidata_to_nfs_open_context(dentry
, nd
);
1356 res
= ERR_CAST(ctx
);
1360 open_flags
= nd
->intent
.open
.flags
;
1361 if (nd
->flags
& LOOKUP_CREATE
) {
1362 attr
.ia_mode
= nd
->intent
.open
.create_mode
;
1363 attr
.ia_valid
= ATTR_MODE
;
1364 if (!IS_POSIXACL(dir
))
1365 attr
.ia_mode
&= ~current_umask();
1367 open_flags
&= ~(O_EXCL
| O_CREAT
);
1371 /* Open the file on the server */
1372 nfs_block_sillyrename(dentry
->d_parent
);
1373 inode
= NFS_PROTO(dir
)->open_context(dir
, ctx
, open_flags
, &attr
);
1374 if (IS_ERR(inode
)) {
1375 nfs_unblock_sillyrename(dentry
->d_parent
);
1376 put_nfs_open_context(ctx
);
1377 switch (PTR_ERR(inode
)) {
1378 /* Make a negative dentry */
1380 d_add(dentry
, NULL
);
1383 /* This turned out not to be a regular file */
1387 if (!(nd
->intent
.open
.flags
& O_NOFOLLOW
))
1392 res
= ERR_CAST(inode
);
1396 res
= d_add_unique(dentry
, inode
);
1397 nfs_unblock_sillyrename(dentry
->d_parent
);
1399 dput(ctx
->path
.dentry
);
1400 ctx
->path
.dentry
= dget(res
);
1403 err
= nfs_intent_set_file(nd
, ctx
);
1407 return ERR_PTR(err
);
1410 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1413 return nfs_lookup(dir
, dentry
, nd
);
1416 static int nfs_open_revalidate(struct dentry
*dentry
, struct nameidata
*nd
)
1418 struct dentry
*parent
= NULL
;
1419 struct inode
*inode
= dentry
->d_inode
;
1421 struct nfs_open_context
*ctx
;
1422 int openflags
, ret
= 0;
1424 if (!is_atomic_open(nd
) || d_mountpoint(dentry
))
1427 parent
= dget_parent(dentry
);
1428 dir
= parent
->d_inode
;
1430 /* We can't create new files in nfs_open_revalidate(), so we
1431 * optimize away revalidation of negative dentries.
1433 if (inode
== NULL
) {
1434 if (!nfs_neg_need_reval(dir
, dentry
, nd
))
1439 /* NFS only supports OPEN on regular files */
1440 if (!S_ISREG(inode
->i_mode
))
1442 openflags
= nd
->intent
.open
.flags
;
1443 /* We cannot do exclusive creation on a positive dentry */
1444 if ((openflags
& (O_CREAT
|O_EXCL
)) == (O_CREAT
|O_EXCL
))
1446 /* We can't create new files, or truncate existing ones here */
1447 openflags
&= ~(O_CREAT
|O_EXCL
|O_TRUNC
);
1449 ctx
= nameidata_to_nfs_open_context(dentry
, nd
);
1454 * Note: we're not holding inode->i_mutex and so may be racing with
1455 * operations that change the directory. We therefore save the
1456 * change attribute *before* we do the RPC call.
1458 inode
= NFS_PROTO(dir
)->open_context(dir
, ctx
, openflags
, NULL
);
1459 if (IS_ERR(inode
)) {
1460 ret
= PTR_ERR(inode
);
1473 if (inode
!= dentry
->d_inode
)
1476 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1477 ret
= nfs_intent_set_file(nd
, ctx
);
1487 put_nfs_open_context(ctx
);
1493 return nfs_lookup_revalidate(dentry
, nd
);
1496 static int nfs_open_create(struct inode
*dir
, struct dentry
*dentry
, int mode
,
1497 struct nameidata
*nd
)
1499 struct nfs_open_context
*ctx
= NULL
;
1504 dfprintk(VFS
, "NFS: create(%s/%ld), %s\n",
1505 dir
->i_sb
->s_id
, dir
->i_ino
, dentry
->d_name
.name
);
1507 attr
.ia_mode
= mode
;
1508 attr
.ia_valid
= ATTR_MODE
;
1510 if ((nd
->flags
& LOOKUP_CREATE
) != 0) {
1511 open_flags
= nd
->intent
.open
.flags
;
1513 ctx
= nameidata_to_nfs_open_context(dentry
, nd
);
1514 error
= PTR_ERR(ctx
);
1519 error
= NFS_PROTO(dir
)->create(dir
, dentry
, &attr
, open_flags
, ctx
);
1523 error
= nfs_intent_set_file(nd
, ctx
);
1530 put_nfs_open_context(ctx
);
1537 #endif /* CONFIG_NFSV4 */
1540 * Code common to create, mkdir, and mknod.
1542 int nfs_instantiate(struct dentry
*dentry
, struct nfs_fh
*fhandle
,
1543 struct nfs_fattr
*fattr
)
1545 struct dentry
*parent
= dget_parent(dentry
);
1546 struct inode
*dir
= parent
->d_inode
;
1547 struct inode
*inode
;
1548 int error
= -EACCES
;
1552 /* We may have been initialized further down */
1553 if (dentry
->d_inode
)
1555 if (fhandle
->size
== 0) {
1556 error
= NFS_PROTO(dir
)->lookup(dir
, &dentry
->d_name
, fhandle
, fattr
);
1560 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1561 if (!(fattr
->valid
& NFS_ATTR_FATTR
)) {
1562 struct nfs_server
*server
= NFS_SB(dentry
->d_sb
);
1563 error
= server
->nfs_client
->rpc_ops
->getattr(server
, fhandle
, fattr
);
1567 inode
= nfs_fhget(dentry
->d_sb
, fhandle
, fattr
);
1568 error
= PTR_ERR(inode
);
1571 d_add(dentry
, inode
);
1576 nfs_mark_for_revalidate(dir
);
1582 * Following a failed create operation, we drop the dentry rather
1583 * than retain a negative dentry. This avoids a problem in the event
1584 * that the operation succeeded on the server, but an error in the
1585 * reply path made it appear to have failed.
1587 static int nfs_create(struct inode
*dir
, struct dentry
*dentry
, int mode
,
1588 struct nameidata
*nd
)
1593 dfprintk(VFS
, "NFS: create(%s/%ld), %s\n",
1594 dir
->i_sb
->s_id
, dir
->i_ino
, dentry
->d_name
.name
);
1596 attr
.ia_mode
= mode
;
1597 attr
.ia_valid
= ATTR_MODE
;
1599 error
= NFS_PROTO(dir
)->create(dir
, dentry
, &attr
, 0, NULL
);
1609 * See comments for nfs_proc_create regarding failed operations.
1612 nfs_mknod(struct inode
*dir
, struct dentry
*dentry
, int mode
, dev_t rdev
)
1617 dfprintk(VFS
, "NFS: mknod(%s/%ld), %s\n",
1618 dir
->i_sb
->s_id
, dir
->i_ino
, dentry
->d_name
.name
);
1620 if (!new_valid_dev(rdev
))
1623 attr
.ia_mode
= mode
;
1624 attr
.ia_valid
= ATTR_MODE
;
1626 status
= NFS_PROTO(dir
)->mknod(dir
, dentry
, &attr
, rdev
);
1636 * See comments for nfs_proc_create regarding failed operations.
1638 static int nfs_mkdir(struct inode
*dir
, struct dentry
*dentry
, int mode
)
1643 dfprintk(VFS
, "NFS: mkdir(%s/%ld), %s\n",
1644 dir
->i_sb
->s_id
, dir
->i_ino
, dentry
->d_name
.name
);
1646 attr
.ia_valid
= ATTR_MODE
;
1647 attr
.ia_mode
= mode
| S_IFDIR
;
1649 error
= NFS_PROTO(dir
)->mkdir(dir
, dentry
, &attr
);
1658 static void nfs_dentry_handle_enoent(struct dentry
*dentry
)
1660 if (dentry
->d_inode
!= NULL
&& !d_unhashed(dentry
))
1664 static int nfs_rmdir(struct inode
*dir
, struct dentry
*dentry
)
1668 dfprintk(VFS
, "NFS: rmdir(%s/%ld), %s\n",
1669 dir
->i_sb
->s_id
, dir
->i_ino
, dentry
->d_name
.name
);
1671 error
= NFS_PROTO(dir
)->rmdir(dir
, &dentry
->d_name
);
1672 /* Ensure the VFS deletes this inode */
1673 if (error
== 0 && dentry
->d_inode
!= NULL
)
1674 clear_nlink(dentry
->d_inode
);
1675 else if (error
== -ENOENT
)
1676 nfs_dentry_handle_enoent(dentry
);
1682 * Remove a file after making sure there are no pending writes,
1683 * and after checking that the file has only one user.
1685 * We invalidate the attribute cache and free the inode prior to the operation
1686 * to avoid possible races if the server reuses the inode.
1688 static int nfs_safe_remove(struct dentry
*dentry
)
1690 struct inode
*dir
= dentry
->d_parent
->d_inode
;
1691 struct inode
*inode
= dentry
->d_inode
;
1694 dfprintk(VFS
, "NFS: safe_remove(%s/%s)\n",
1695 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
);
1697 /* If the dentry was sillyrenamed, we simply call d_delete() */
1698 if (dentry
->d_flags
& DCACHE_NFSFS_RENAMED
) {
1703 if (inode
!= NULL
) {
1704 nfs_inode_return_delegation(inode
);
1705 error
= NFS_PROTO(dir
)->remove(dir
, &dentry
->d_name
);
1706 /* The VFS may want to delete this inode */
1708 nfs_drop_nlink(inode
);
1709 nfs_mark_for_revalidate(inode
);
1711 error
= NFS_PROTO(dir
)->remove(dir
, &dentry
->d_name
);
1712 if (error
== -ENOENT
)
1713 nfs_dentry_handle_enoent(dentry
);
1718 /* We do silly rename. In case sillyrename() returns -EBUSY, the inode
1719 * belongs to an active ".nfs..." file and we return -EBUSY.
1721 * If sillyrename() returns 0, we do nothing, otherwise we unlink.
1723 static int nfs_unlink(struct inode
*dir
, struct dentry
*dentry
)
1726 int need_rehash
= 0;
1728 dfprintk(VFS
, "NFS: unlink(%s/%ld, %s)\n", dir
->i_sb
->s_id
,
1729 dir
->i_ino
, dentry
->d_name
.name
);
1731 spin_lock(&dcache_lock
);
1732 spin_lock(&dentry
->d_lock
);
1733 if (atomic_read(&dentry
->d_count
) > 1) {
1734 spin_unlock(&dentry
->d_lock
);
1735 spin_unlock(&dcache_lock
);
1736 /* Start asynchronous writeout of the inode */
1737 write_inode_now(dentry
->d_inode
, 0);
1738 error
= nfs_sillyrename(dir
, dentry
);
1741 if (!d_unhashed(dentry
)) {
1745 spin_unlock(&dentry
->d_lock
);
1746 spin_unlock(&dcache_lock
);
1747 error
= nfs_safe_remove(dentry
);
1748 if (!error
|| error
== -ENOENT
) {
1749 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1750 } else if (need_rehash
)
1756 * To create a symbolic link, most file systems instantiate a new inode,
1757 * add a page to it containing the path, then write it out to the disk
1758 * using prepare_write/commit_write.
1760 * Unfortunately the NFS client can't create the in-core inode first
1761 * because it needs a file handle to create an in-core inode (see
1762 * fs/nfs/inode.c:nfs_fhget). We only have a file handle *after* the
1763 * symlink request has completed on the server.
1765 * So instead we allocate a raw page, copy the symname into it, then do
1766 * the SYMLINK request with the page as the buffer. If it succeeds, we
1767 * now have a new file handle and can instantiate an in-core NFS inode
1768 * and move the raw page into its mapping.
1770 static int nfs_symlink(struct inode
*dir
, struct dentry
*dentry
, const char *symname
)
1772 struct pagevec lru_pvec
;
1776 unsigned int pathlen
= strlen(symname
);
1779 dfprintk(VFS
, "NFS: symlink(%s/%ld, %s, %s)\n", dir
->i_sb
->s_id
,
1780 dir
->i_ino
, dentry
->d_name
.name
, symname
);
1782 if (pathlen
> PAGE_SIZE
)
1783 return -ENAMETOOLONG
;
1785 attr
.ia_mode
= S_IFLNK
| S_IRWXUGO
;
1786 attr
.ia_valid
= ATTR_MODE
;
1788 page
= alloc_page(GFP_HIGHUSER
);
1792 kaddr
= kmap_atomic(page
, KM_USER0
);
1793 memcpy(kaddr
, symname
, pathlen
);
1794 if (pathlen
< PAGE_SIZE
)
1795 memset(kaddr
+ pathlen
, 0, PAGE_SIZE
- pathlen
);
1796 kunmap_atomic(kaddr
, KM_USER0
);
1798 error
= NFS_PROTO(dir
)->symlink(dir
, dentry
, page
, pathlen
, &attr
);
1800 dfprintk(VFS
, "NFS: symlink(%s/%ld, %s, %s) error %d\n",
1801 dir
->i_sb
->s_id
, dir
->i_ino
,
1802 dentry
->d_name
.name
, symname
, error
);
1809 * No big deal if we can't add this page to the page cache here.
1810 * READLINK will get the missing page from the server if needed.
1812 pagevec_init(&lru_pvec
, 0);
1813 if (!add_to_page_cache(page
, dentry
->d_inode
->i_mapping
, 0,
1815 pagevec_add(&lru_pvec
, page
);
1816 pagevec_lru_add_file(&lru_pvec
);
1817 SetPageUptodate(page
);
1826 nfs_link(struct dentry
*old_dentry
, struct inode
*dir
, struct dentry
*dentry
)
1828 struct inode
*inode
= old_dentry
->d_inode
;
1831 dfprintk(VFS
, "NFS: link(%s/%s -> %s/%s)\n",
1832 old_dentry
->d_parent
->d_name
.name
, old_dentry
->d_name
.name
,
1833 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
);
1835 nfs_inode_return_delegation(inode
);
1838 error
= NFS_PROTO(dir
)->link(inode
, dir
, &dentry
->d_name
);
1841 d_add(dentry
, inode
);
1848 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a
1849 * different file handle for the same inode after a rename (e.g. when
1850 * moving to a different directory). A fail-safe method to do so would
1851 * be to look up old_dir/old_name, create a link to new_dir/new_name and
1852 * rename the old file using the sillyrename stuff. This way, the original
1853 * file in old_dir will go away when the last process iput()s the inode.
1857 * It actually works quite well. One needs to have the possibility for
1858 * at least one ".nfs..." file in each directory the file ever gets
1859 * moved or linked to which happens automagically with the new
1860 * implementation that only depends on the dcache stuff instead of
1861 * using the inode layer
1863 * Unfortunately, things are a little more complicated than indicated
1864 * above. For a cross-directory move, we want to make sure we can get
1865 * rid of the old inode after the operation. This means there must be
1866 * no pending writes (if it's a file), and the use count must be 1.
1867 * If these conditions are met, we can drop the dentries before doing
1870 static int nfs_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
1871 struct inode
*new_dir
, struct dentry
*new_dentry
)
1873 struct inode
*old_inode
= old_dentry
->d_inode
;
1874 struct inode
*new_inode
= new_dentry
->d_inode
;
1875 struct dentry
*dentry
= NULL
, *rehash
= NULL
;
1878 dfprintk(VFS
, "NFS: rename(%s/%s -> %s/%s, ct=%d)\n",
1879 old_dentry
->d_parent
->d_name
.name
, old_dentry
->d_name
.name
,
1880 new_dentry
->d_parent
->d_name
.name
, new_dentry
->d_name
.name
,
1881 atomic_read(&new_dentry
->d_count
));
1884 * For non-directories, check whether the target is busy and if so,
1885 * make a copy of the dentry and then do a silly-rename. If the
1886 * silly-rename succeeds, the copied dentry is hashed and becomes
1889 if (new_inode
&& !S_ISDIR(new_inode
->i_mode
)) {
1891 * To prevent any new references to the target during the
1892 * rename, we unhash the dentry in advance.
1894 if (!d_unhashed(new_dentry
)) {
1896 rehash
= new_dentry
;
1899 if (atomic_read(&new_dentry
->d_count
) > 2) {
1902 /* copy the target dentry's name */
1903 dentry
= d_alloc(new_dentry
->d_parent
,
1904 &new_dentry
->d_name
);
1908 /* silly-rename the existing target ... */
1909 err
= nfs_sillyrename(new_dir
, new_dentry
);
1913 new_dentry
= dentry
;
1919 nfs_inode_return_delegation(old_inode
);
1920 if (new_inode
!= NULL
)
1921 nfs_inode_return_delegation(new_inode
);
1923 error
= NFS_PROTO(old_dir
)->rename(old_dir
, &old_dentry
->d_name
,
1924 new_dir
, &new_dentry
->d_name
);
1925 nfs_mark_for_revalidate(old_inode
);
1930 if (new_inode
!= NULL
)
1931 nfs_drop_nlink(new_inode
);
1932 d_move(old_dentry
, new_dentry
);
1933 nfs_set_verifier(new_dentry
,
1934 nfs_save_change_attribute(new_dir
));
1935 } else if (error
== -ENOENT
)
1936 nfs_dentry_handle_enoent(old_dentry
);
1938 /* new dentry created? */
1944 static DEFINE_SPINLOCK(nfs_access_lru_lock
);
1945 static LIST_HEAD(nfs_access_lru_list
);
1946 static atomic_long_t nfs_access_nr_entries
;
1948 static void nfs_access_free_entry(struct nfs_access_entry
*entry
)
1950 put_rpccred(entry
->cred
);
1952 smp_mb__before_atomic_dec();
1953 atomic_long_dec(&nfs_access_nr_entries
);
1954 smp_mb__after_atomic_dec();
1957 static void nfs_access_free_list(struct list_head
*head
)
1959 struct nfs_access_entry
*cache
;
1961 while (!list_empty(head
)) {
1962 cache
= list_entry(head
->next
, struct nfs_access_entry
, lru
);
1963 list_del(&cache
->lru
);
1964 nfs_access_free_entry(cache
);
1968 int nfs_access_cache_shrinker(struct shrinker
*shrink
, int nr_to_scan
, gfp_t gfp_mask
)
1971 struct nfs_inode
*nfsi
, *next
;
1972 struct nfs_access_entry
*cache
;
1974 if ((gfp_mask
& GFP_KERNEL
) != GFP_KERNEL
)
1975 return (nr_to_scan
== 0) ? 0 : -1;
1977 spin_lock(&nfs_access_lru_lock
);
1978 list_for_each_entry_safe(nfsi
, next
, &nfs_access_lru_list
, access_cache_inode_lru
) {
1979 struct inode
*inode
;
1981 if (nr_to_scan
-- == 0)
1983 inode
= &nfsi
->vfs_inode
;
1984 spin_lock(&inode
->i_lock
);
1985 if (list_empty(&nfsi
->access_cache_entry_lru
))
1986 goto remove_lru_entry
;
1987 cache
= list_entry(nfsi
->access_cache_entry_lru
.next
,
1988 struct nfs_access_entry
, lru
);
1989 list_move(&cache
->lru
, &head
);
1990 rb_erase(&cache
->rb_node
, &nfsi
->access_cache
);
1991 if (!list_empty(&nfsi
->access_cache_entry_lru
))
1992 list_move_tail(&nfsi
->access_cache_inode_lru
,
1993 &nfs_access_lru_list
);
1996 list_del_init(&nfsi
->access_cache_inode_lru
);
1997 smp_mb__before_clear_bit();
1998 clear_bit(NFS_INO_ACL_LRU_SET
, &nfsi
->flags
);
1999 smp_mb__after_clear_bit();
2001 spin_unlock(&inode
->i_lock
);
2003 spin_unlock(&nfs_access_lru_lock
);
2004 nfs_access_free_list(&head
);
2005 return (atomic_long_read(&nfs_access_nr_entries
) / 100) * sysctl_vfs_cache_pressure
;
2008 static void __nfs_access_zap_cache(struct nfs_inode
*nfsi
, struct list_head
*head
)
2010 struct rb_root
*root_node
= &nfsi
->access_cache
;
2012 struct nfs_access_entry
*entry
;
2014 /* Unhook entries from the cache */
2015 while ((n
= rb_first(root_node
)) != NULL
) {
2016 entry
= rb_entry(n
, struct nfs_access_entry
, rb_node
);
2017 rb_erase(n
, root_node
);
2018 list_move(&entry
->lru
, head
);
2020 nfsi
->cache_validity
&= ~NFS_INO_INVALID_ACCESS
;
2023 void nfs_access_zap_cache(struct inode
*inode
)
2027 if (test_bit(NFS_INO_ACL_LRU_SET
, &NFS_I(inode
)->flags
) == 0)
2029 /* Remove from global LRU init */
2030 spin_lock(&nfs_access_lru_lock
);
2031 if (test_and_clear_bit(NFS_INO_ACL_LRU_SET
, &NFS_I(inode
)->flags
))
2032 list_del_init(&NFS_I(inode
)->access_cache_inode_lru
);
2034 spin_lock(&inode
->i_lock
);
2035 __nfs_access_zap_cache(NFS_I(inode
), &head
);
2036 spin_unlock(&inode
->i_lock
);
2037 spin_unlock(&nfs_access_lru_lock
);
2038 nfs_access_free_list(&head
);
2041 static struct nfs_access_entry
*nfs_access_search_rbtree(struct inode
*inode
, struct rpc_cred
*cred
)
2043 struct rb_node
*n
= NFS_I(inode
)->access_cache
.rb_node
;
2044 struct nfs_access_entry
*entry
;
2047 entry
= rb_entry(n
, struct nfs_access_entry
, rb_node
);
2049 if (cred
< entry
->cred
)
2051 else if (cred
> entry
->cred
)
2059 static int nfs_access_get_cached(struct inode
*inode
, struct rpc_cred
*cred
, struct nfs_access_entry
*res
)
2061 struct nfs_inode
*nfsi
= NFS_I(inode
);
2062 struct nfs_access_entry
*cache
;
2065 spin_lock(&inode
->i_lock
);
2066 if (nfsi
->cache_validity
& NFS_INO_INVALID_ACCESS
)
2068 cache
= nfs_access_search_rbtree(inode
, cred
);
2071 if (!nfs_have_delegated_attributes(inode
) &&
2072 !time_in_range_open(jiffies
, cache
->jiffies
, cache
->jiffies
+ nfsi
->attrtimeo
))
2074 res
->jiffies
= cache
->jiffies
;
2075 res
->cred
= cache
->cred
;
2076 res
->mask
= cache
->mask
;
2077 list_move_tail(&cache
->lru
, &nfsi
->access_cache_entry_lru
);
2080 spin_unlock(&inode
->i_lock
);
2083 rb_erase(&cache
->rb_node
, &nfsi
->access_cache
);
2084 list_del(&cache
->lru
);
2085 spin_unlock(&inode
->i_lock
);
2086 nfs_access_free_entry(cache
);
2089 spin_unlock(&inode
->i_lock
);
2090 nfs_access_zap_cache(inode
);
2094 static void nfs_access_add_rbtree(struct inode
*inode
, struct nfs_access_entry
*set
)
2096 struct nfs_inode
*nfsi
= NFS_I(inode
);
2097 struct rb_root
*root_node
= &nfsi
->access_cache
;
2098 struct rb_node
**p
= &root_node
->rb_node
;
2099 struct rb_node
*parent
= NULL
;
2100 struct nfs_access_entry
*entry
;
2102 spin_lock(&inode
->i_lock
);
2103 while (*p
!= NULL
) {
2105 entry
= rb_entry(parent
, struct nfs_access_entry
, rb_node
);
2107 if (set
->cred
< entry
->cred
)
2108 p
= &parent
->rb_left
;
2109 else if (set
->cred
> entry
->cred
)
2110 p
= &parent
->rb_right
;
2114 rb_link_node(&set
->rb_node
, parent
, p
);
2115 rb_insert_color(&set
->rb_node
, root_node
);
2116 list_add_tail(&set
->lru
, &nfsi
->access_cache_entry_lru
);
2117 spin_unlock(&inode
->i_lock
);
2120 rb_replace_node(parent
, &set
->rb_node
, root_node
);
2121 list_add_tail(&set
->lru
, &nfsi
->access_cache_entry_lru
);
2122 list_del(&entry
->lru
);
2123 spin_unlock(&inode
->i_lock
);
2124 nfs_access_free_entry(entry
);
2127 static void nfs_access_add_cache(struct inode
*inode
, struct nfs_access_entry
*set
)
2129 struct nfs_access_entry
*cache
= kmalloc(sizeof(*cache
), GFP_KERNEL
);
2132 RB_CLEAR_NODE(&cache
->rb_node
);
2133 cache
->jiffies
= set
->jiffies
;
2134 cache
->cred
= get_rpccred(set
->cred
);
2135 cache
->mask
= set
->mask
;
2137 nfs_access_add_rbtree(inode
, cache
);
2139 /* Update accounting */
2140 smp_mb__before_atomic_inc();
2141 atomic_long_inc(&nfs_access_nr_entries
);
2142 smp_mb__after_atomic_inc();
2144 /* Add inode to global LRU list */
2145 if (!test_bit(NFS_INO_ACL_LRU_SET
, &NFS_I(inode
)->flags
)) {
2146 spin_lock(&nfs_access_lru_lock
);
2147 if (!test_and_set_bit(NFS_INO_ACL_LRU_SET
, &NFS_I(inode
)->flags
))
2148 list_add_tail(&NFS_I(inode
)->access_cache_inode_lru
,
2149 &nfs_access_lru_list
);
2150 spin_unlock(&nfs_access_lru_lock
);
2154 static int nfs_do_access(struct inode
*inode
, struct rpc_cred
*cred
, int mask
)
2156 struct nfs_access_entry cache
;
2159 status
= nfs_access_get_cached(inode
, cred
, &cache
);
2163 /* Be clever: ask server to check for all possible rights */
2164 cache
.mask
= MAY_EXEC
| MAY_WRITE
| MAY_READ
;
2166 cache
.jiffies
= jiffies
;
2167 status
= NFS_PROTO(inode
)->access(inode
, &cache
);
2169 if (status
== -ESTALE
) {
2170 nfs_zap_caches(inode
);
2171 if (!S_ISDIR(inode
->i_mode
))
2172 set_bit(NFS_INO_STALE
, &NFS_I(inode
)->flags
);
2176 nfs_access_add_cache(inode
, &cache
);
2178 if ((mask
& ~cache
.mask
& (MAY_READ
| MAY_WRITE
| MAY_EXEC
)) == 0)
2183 static int nfs_open_permission_mask(int openflags
)
2187 if (openflags
& FMODE_READ
)
2189 if (openflags
& FMODE_WRITE
)
2191 if (openflags
& FMODE_EXEC
)
2196 int nfs_may_open(struct inode
*inode
, struct rpc_cred
*cred
, int openflags
)
2198 return nfs_do_access(inode
, cred
, nfs_open_permission_mask(openflags
));
2201 int nfs_permission(struct inode
*inode
, int mask
)
2203 struct rpc_cred
*cred
;
2206 nfs_inc_stats(inode
, NFSIOS_VFSACCESS
);
2208 if ((mask
& (MAY_READ
| MAY_WRITE
| MAY_EXEC
)) == 0)
2210 /* Is this sys_access() ? */
2211 if (mask
& (MAY_ACCESS
| MAY_CHDIR
))
2214 switch (inode
->i_mode
& S_IFMT
) {
2218 /* NFSv4 has atomic_open... */
2219 if (nfs_server_capable(inode
, NFS_CAP_ATOMIC_OPEN
)
2220 && (mask
& MAY_OPEN
)
2221 && !(mask
& MAY_EXEC
))
2226 * Optimize away all write operations, since the server
2227 * will check permissions when we perform the op.
2229 if ((mask
& MAY_WRITE
) && !(mask
& MAY_READ
))
2234 if (!NFS_PROTO(inode
)->access
)
2237 cred
= rpc_lookup_cred();
2238 if (!IS_ERR(cred
)) {
2239 res
= nfs_do_access(inode
, cred
, mask
);
2242 res
= PTR_ERR(cred
);
2244 if (!res
&& (mask
& MAY_EXEC
) && !execute_ok(inode
))
2247 dfprintk(VFS
, "NFS: permission(%s/%ld), mask=0x%x, res=%d\n",
2248 inode
->i_sb
->s_id
, inode
->i_ino
, mask
, res
);
2251 res
= nfs_revalidate_inode(NFS_SERVER(inode
), inode
);
2253 res
= generic_permission(inode
, mask
, NULL
);
2259 * version-control: t
2260 * kept-new-versions: 5