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 void nfs_readdir_clear_array(struct page
*);
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_aops
= {
87 .freepage
= 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
;
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 void nfs_readdir_clear_array(struct page
*page
)
219 struct nfs_cache_array
*array
;
222 array
= kmap_atomic(page
, KM_USER0
);
223 for (i
= 0; i
< array
->size
; i
++)
224 kfree(array
->array
[i
].string
.name
);
225 kunmap_atomic(array
, KM_USER0
);
229 * the caller is responsible for freeing qstr.name
230 * when called by nfs_readdir_add_to_array, the strings will be freed in
231 * nfs_clear_readdir_array()
234 int nfs_readdir_make_qstr(struct qstr
*string
, const char *name
, unsigned int len
)
237 string
->name
= kmemdup(name
, len
, GFP_KERNEL
);
238 if (string
->name
== NULL
)
241 * Avoid a kmemleak false positive. The pointer to the name is stored
242 * in a page cache page which kmemleak does not scan.
244 kmemleak_not_leak(string
->name
);
245 string
->hash
= full_name_hash(name
, len
);
250 int nfs_readdir_add_to_array(struct nfs_entry
*entry
, struct page
*page
)
252 struct nfs_cache_array
*array
= nfs_readdir_get_array(page
);
253 struct nfs_cache_array_entry
*cache_entry
;
257 return PTR_ERR(array
);
259 cache_entry
= &array
->array
[array
->size
];
261 /* Check that this entry lies within the page bounds */
263 if ((char *)&cache_entry
[1] - (char *)page_address(page
) > PAGE_SIZE
)
266 cache_entry
->cookie
= entry
->prev_cookie
;
267 cache_entry
->ino
= entry
->ino
;
268 cache_entry
->d_type
= entry
->d_type
;
269 ret
= nfs_readdir_make_qstr(&cache_entry
->string
, entry
->name
, entry
->len
);
272 array
->last_cookie
= entry
->cookie
;
275 array
->eof_index
= array
->size
;
277 nfs_readdir_release_array(page
);
282 int nfs_readdir_search_for_pos(struct nfs_cache_array
*array
, nfs_readdir_descriptor_t
*desc
)
284 loff_t diff
= desc
->file
->f_pos
- desc
->current_index
;
289 if (diff
>= array
->size
) {
290 if (array
->eof_index
>= 0)
292 desc
->current_index
+= array
->size
;
296 index
= (unsigned int)diff
;
297 *desc
->dir_cookie
= array
->array
[index
].cookie
;
298 desc
->cache_entry_index
= index
;
306 int nfs_readdir_search_for_cookie(struct nfs_cache_array
*array
, nfs_readdir_descriptor_t
*desc
)
309 int status
= -EAGAIN
;
311 for (i
= 0; i
< array
->size
; i
++) {
312 if (array
->array
[i
].cookie
== *desc
->dir_cookie
) {
313 desc
->cache_entry_index
= i
;
317 if (array
->eof_index
>= 0) {
318 status
= -EBADCOOKIE
;
319 if (*desc
->dir_cookie
== array
->last_cookie
)
326 int nfs_readdir_search_array(nfs_readdir_descriptor_t
*desc
)
328 struct nfs_cache_array
*array
;
331 array
= nfs_readdir_get_array(desc
->page
);
333 status
= PTR_ERR(array
);
337 if (*desc
->dir_cookie
== 0)
338 status
= nfs_readdir_search_for_pos(array
, desc
);
340 status
= nfs_readdir_search_for_cookie(array
, desc
);
342 if (status
== -EAGAIN
) {
343 desc
->last_cookie
= array
->last_cookie
;
346 nfs_readdir_release_array(desc
->page
);
351 /* Fill a page with xdr information before transferring to the cache page */
353 int nfs_readdir_xdr_filler(struct page
**pages
, nfs_readdir_descriptor_t
*desc
,
354 struct nfs_entry
*entry
, struct file
*file
, struct inode
*inode
)
356 struct rpc_cred
*cred
= nfs_file_cred(file
);
357 unsigned long timestamp
, gencount
;
362 gencount
= nfs_inc_attr_generation_counter();
363 error
= NFS_PROTO(inode
)->readdir(file
->f_path
.dentry
, cred
, entry
->cookie
, pages
,
364 NFS_SERVER(inode
)->dtsize
, desc
->plus
);
366 /* We requested READDIRPLUS, but the server doesn't grok it */
367 if (error
== -ENOTSUPP
&& desc
->plus
) {
368 NFS_SERVER(inode
)->caps
&= ~NFS_CAP_READDIRPLUS
;
369 clear_bit(NFS_INO_ADVISE_RDPLUS
, &NFS_I(inode
)->flags
);
375 desc
->timestamp
= timestamp
;
376 desc
->gencount
= gencount
;
381 /* Fill in an entry based on the xdr code stored in desc->page */
383 int xdr_decode(nfs_readdir_descriptor_t
*desc
, struct nfs_entry
*entry
, struct xdr_stream
*stream
)
385 __be32
*p
= desc
->decode(stream
, entry
, NFS_SERVER(desc
->file
->f_path
.dentry
->d_inode
), desc
->plus
);
389 entry
->fattr
->time_start
= desc
->timestamp
;
390 entry
->fattr
->gencount
= desc
->gencount
;
395 int nfs_same_file(struct dentry
*dentry
, struct nfs_entry
*entry
)
397 if (dentry
->d_inode
== NULL
)
399 if (nfs_compare_fh(entry
->fh
, NFS_FH(dentry
->d_inode
)) != 0)
407 void nfs_prime_dcache(struct dentry
*parent
, struct nfs_entry
*entry
)
409 struct qstr filename
= {
413 struct dentry
*dentry
;
414 struct dentry
*alias
;
415 struct inode
*dir
= parent
->d_inode
;
418 if (filename
.name
[0] == '.') {
419 if (filename
.len
== 1)
421 if (filename
.len
== 2 && filename
.name
[1] == '.')
424 filename
.hash
= full_name_hash(filename
.name
, filename
.len
);
426 dentry
= d_lookup(parent
, &filename
);
427 if (dentry
!= NULL
) {
428 if (nfs_same_file(dentry
, entry
)) {
429 nfs_refresh_inode(dentry
->d_inode
, entry
->fattr
);
437 dentry
= d_alloc(parent
, &filename
);
441 dentry
->d_op
= NFS_PROTO(dir
)->dentry_ops
;
442 inode
= nfs_fhget(dentry
->d_sb
, entry
->fh
, entry
->fattr
);
446 alias
= d_materialise_unique(dentry
, inode
);
450 nfs_set_verifier(alias
, nfs_save_change_attribute(dir
));
453 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
459 /* Perform conversion from xdr to cache array */
461 int nfs_readdir_page_filler(nfs_readdir_descriptor_t
*desc
, struct nfs_entry
*entry
,
462 void *xdr_page
, struct page
*page
, unsigned int buflen
)
464 struct xdr_stream stream
;
466 __be32
*ptr
= xdr_page
;
467 struct nfs_cache_array
*array
;
468 unsigned int count
= 0;
471 buf
.head
->iov_base
= xdr_page
;
472 buf
.head
->iov_len
= buflen
;
473 buf
.tail
->iov_len
= 0;
476 buf
.buflen
= buf
.head
->iov_len
;
477 buf
.len
= buf
.head
->iov_len
;
479 xdr_init_decode(&stream
, &buf
, ptr
);
483 status
= xdr_decode(desc
, entry
, &stream
);
485 if (status
== -EAGAIN
)
493 nfs_prime_dcache(desc
->file
->f_path
.dentry
, entry
);
495 status
= nfs_readdir_add_to_array(entry
, page
);
498 } while (!entry
->eof
);
500 if (count
== 0 || (status
== -EBADCOOKIE
&& entry
->eof
!= 0)) {
501 array
= nfs_readdir_get_array(page
);
502 if (!IS_ERR(array
)) {
503 array
->eof_index
= array
->size
;
505 nfs_readdir_release_array(page
);
507 status
= PTR_ERR(array
);
513 void nfs_readdir_free_pagearray(struct page
**pages
, unsigned int npages
)
516 for (i
= 0; i
< npages
; i
++)
521 void nfs_readdir_free_large_page(void *ptr
, struct page
**pages
,
524 vm_unmap_ram(ptr
, npages
);
525 nfs_readdir_free_pagearray(pages
, npages
);
529 * nfs_readdir_large_page will allocate pages that must be freed with a call
530 * to nfs_readdir_free_large_page
533 void *nfs_readdir_large_page(struct page
**pages
, unsigned int npages
)
538 for (i
= 0; i
< npages
; i
++) {
539 struct page
*page
= alloc_page(GFP_KERNEL
);
545 ptr
= vm_map_ram(pages
, npages
, 0, PAGE_KERNEL
);
546 if (!IS_ERR_OR_NULL(ptr
))
549 nfs_readdir_free_pagearray(pages
, i
);
554 int nfs_readdir_xdr_to_array(nfs_readdir_descriptor_t
*desc
, struct page
*page
, struct inode
*inode
)
556 struct page
*pages
[NFS_MAX_READDIR_PAGES
];
557 void *pages_ptr
= NULL
;
558 struct nfs_entry entry
;
559 struct file
*file
= desc
->file
;
560 struct nfs_cache_array
*array
;
561 int status
= -ENOMEM
;
562 unsigned int array_size
= ARRAY_SIZE(pages
);
564 entry
.prev_cookie
= 0;
565 entry
.cookie
= desc
->last_cookie
;
567 entry
.fh
= nfs_alloc_fhandle();
568 entry
.fattr
= nfs_alloc_fattr();
569 if (entry
.fh
== NULL
|| entry
.fattr
== NULL
)
572 array
= nfs_readdir_get_array(page
);
574 status
= PTR_ERR(array
);
577 memset(array
, 0, sizeof(struct nfs_cache_array
));
578 array
->eof_index
= -1;
580 pages_ptr
= nfs_readdir_large_page(pages
, array_size
);
582 goto out_release_array
;
585 status
= nfs_readdir_xdr_filler(pages
, desc
, &entry
, file
, inode
);
590 status
= nfs_readdir_page_filler(desc
, &entry
, pages_ptr
, page
, pglen
);
592 if (status
== -ENOSPC
)
596 } while (array
->eof_index
< 0);
598 nfs_readdir_free_large_page(pages_ptr
, pages
, array_size
);
600 nfs_readdir_release_array(page
);
602 nfs_free_fattr(entry
.fattr
);
603 nfs_free_fhandle(entry
.fh
);
608 * Now we cache directories properly, by converting xdr information
609 * to an array that can be used for lookups later. This results in
610 * fewer cache pages, since we can store more information on each page.
611 * We only need to convert from xdr once so future lookups are much simpler
614 int nfs_readdir_filler(nfs_readdir_descriptor_t
*desc
, struct page
* page
)
616 struct inode
*inode
= desc
->file
->f_path
.dentry
->d_inode
;
619 ret
= nfs_readdir_xdr_to_array(desc
, page
, inode
);
622 SetPageUptodate(page
);
624 if (invalidate_inode_pages2_range(inode
->i_mapping
, page
->index
+ 1, -1) < 0) {
625 /* Should never happen */
626 nfs_zap_mapping(inode
, inode
->i_mapping
);
636 void cache_page_release(nfs_readdir_descriptor_t
*desc
)
638 if (!desc
->page
->mapping
)
639 nfs_readdir_clear_array(desc
->page
);
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
);
665 cache_page_release(desc
);
669 /* Search for desc->dir_cookie from the beginning of the page cache */
671 int readdir_search_pagecache(nfs_readdir_descriptor_t
*desc
)
675 if (desc
->page_index
== 0) {
676 desc
->current_index
= 0;
677 desc
->last_cookie
= 0;
680 res
= find_cache_page(desc
);
681 } while (res
== -EAGAIN
);
686 * Once we've found the start of the dirent within a page: fill 'er up...
689 int nfs_do_filldir(nfs_readdir_descriptor_t
*desc
, void *dirent
,
692 struct file
*file
= desc
->file
;
695 struct nfs_cache_array
*array
= NULL
;
697 array
= nfs_readdir_get_array(desc
->page
);
699 res
= PTR_ERR(array
);
703 for (i
= desc
->cache_entry_index
; i
< array
->size
; i
++) {
704 struct nfs_cache_array_entry
*ent
;
706 ent
= &array
->array
[i
];
707 if (filldir(dirent
, ent
->string
.name
, ent
->string
.len
,
708 file
->f_pos
, nfs_compat_user_ino64(ent
->ino
),
714 if (i
< (array
->size
-1))
715 *desc
->dir_cookie
= array
->array
[i
+1].cookie
;
717 *desc
->dir_cookie
= array
->last_cookie
;
719 if (array
->eof_index
>= 0)
722 nfs_readdir_release_array(desc
->page
);
724 cache_page_release(desc
);
725 dfprintk(DIRCACHE
, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n",
726 (unsigned long long)*desc
->dir_cookie
, res
);
731 * If we cannot find a cookie in our cache, we suspect that this is
732 * because it points to a deleted file, so we ask the server to return
733 * whatever it thinks is the next entry. We then feed this to filldir.
734 * If all goes well, we should then be able to find our way round the
735 * cache on the next call to readdir_search_pagecache();
737 * NOTE: we cannot add the anonymous page to the pagecache because
738 * the data it contains might not be page aligned. Besides,
739 * we should already have a complete representation of the
740 * directory in the page cache by the time we get here.
743 int uncached_readdir(nfs_readdir_descriptor_t
*desc
, void *dirent
,
746 struct page
*page
= NULL
;
748 struct inode
*inode
= desc
->file
->f_path
.dentry
->d_inode
;
750 dfprintk(DIRCACHE
, "NFS: uncached_readdir() searching for cookie %Lu\n",
751 (unsigned long long)*desc
->dir_cookie
);
753 page
= alloc_page(GFP_HIGHUSER
);
759 desc
->page_index
= 0;
760 desc
->last_cookie
= *desc
->dir_cookie
;
763 status
= nfs_readdir_xdr_to_array(desc
, page
, inode
);
767 status
= nfs_do_filldir(desc
, dirent
, filldir
);
770 dfprintk(DIRCACHE
, "NFS: %s: returns %d\n",
774 cache_page_release(desc
);
778 /* The file offset position represents the dirent entry number. A
779 last cookie cache takes care of the common case of reading the
782 static int nfs_readdir(struct file
*filp
, void *dirent
, filldir_t filldir
)
784 struct dentry
*dentry
= filp
->f_path
.dentry
;
785 struct inode
*inode
= dentry
->d_inode
;
786 nfs_readdir_descriptor_t my_desc
,
790 dfprintk(FILE, "NFS: readdir(%s/%s) starting at cookie %llu\n",
791 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
,
792 (long long)filp
->f_pos
);
793 nfs_inc_stats(inode
, NFSIOS_VFSGETDENTS
);
796 * filp->f_pos points to the dirent entry number.
797 * *desc->dir_cookie has the cookie for the next entry. We have
798 * to either find the entry with the appropriate number or
799 * revalidate the cookie.
801 memset(desc
, 0, sizeof(*desc
));
804 desc
->dir_cookie
= &nfs_file_open_context(filp
)->dir_cookie
;
805 desc
->decode
= NFS_PROTO(inode
)->decode_dirent
;
806 desc
->plus
= NFS_USE_READDIRPLUS(inode
);
808 nfs_block_sillyrename(dentry
);
809 res
= nfs_revalidate_mapping(inode
, filp
->f_mapping
);
814 res
= readdir_search_pagecache(desc
);
816 if (res
== -EBADCOOKIE
) {
818 /* This means either end of directory */
819 if (*desc
->dir_cookie
&& desc
->eof
== 0) {
820 /* Or that the server has 'lost' a cookie */
821 res
= uncached_readdir(desc
, dirent
, filldir
);
827 if (res
== -ETOOSMALL
&& desc
->plus
) {
828 clear_bit(NFS_INO_ADVISE_RDPLUS
, &NFS_I(inode
)->flags
);
829 nfs_zap_caches(inode
);
830 desc
->page_index
= 0;
838 res
= nfs_do_filldir(desc
, dirent
, filldir
);
841 } while (!desc
->eof
);
843 nfs_unblock_sillyrename(dentry
);
846 dfprintk(FILE, "NFS: readdir(%s/%s) returns %d\n",
847 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
,
852 static loff_t
nfs_llseek_dir(struct file
*filp
, loff_t offset
, int origin
)
854 struct dentry
*dentry
= filp
->f_path
.dentry
;
855 struct inode
*inode
= dentry
->d_inode
;
857 dfprintk(FILE, "NFS: llseek dir(%s/%s, %lld, %d)\n",
858 dentry
->d_parent
->d_name
.name
,
862 mutex_lock(&inode
->i_mutex
);
865 offset
+= filp
->f_pos
;
873 if (offset
!= filp
->f_pos
) {
874 filp
->f_pos
= offset
;
875 nfs_file_open_context(filp
)->dir_cookie
= 0;
878 mutex_unlock(&inode
->i_mutex
);
883 * All directory operations under NFS are synchronous, so fsync()
884 * is a dummy operation.
886 static int nfs_fsync_dir(struct file
*filp
, int datasync
)
888 struct dentry
*dentry
= filp
->f_path
.dentry
;
890 dfprintk(FILE, "NFS: fsync dir(%s/%s) datasync %d\n",
891 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
,
894 nfs_inc_stats(dentry
->d_inode
, NFSIOS_VFSFSYNC
);
899 * nfs_force_lookup_revalidate - Mark the directory as having changed
900 * @dir - pointer to directory inode
902 * This forces the revalidation code in nfs_lookup_revalidate() to do a
903 * full lookup on all child dentries of 'dir' whenever a change occurs
904 * on the server that might have invalidated our dcache.
906 * The caller should be holding dir->i_lock
908 void nfs_force_lookup_revalidate(struct inode
*dir
)
910 NFS_I(dir
)->cache_change_attribute
++;
914 * A check for whether or not the parent directory has changed.
915 * In the case it has, we assume that the dentries are untrustworthy
916 * and may need to be looked up again.
918 static int nfs_check_verifier(struct inode
*dir
, struct dentry
*dentry
)
922 if (NFS_SERVER(dir
)->flags
& NFS_MOUNT_LOOKUP_CACHE_NONE
)
924 if (!nfs_verify_change_attribute(dir
, dentry
->d_time
))
926 /* Revalidate nfsi->cache_change_attribute before we declare a match */
927 if (nfs_revalidate_inode(NFS_SERVER(dir
), dir
) < 0)
929 if (!nfs_verify_change_attribute(dir
, dentry
->d_time
))
935 * Return the intent data that applies to this particular path component
937 * Note that the current set of intents only apply to the very last
938 * component of the path.
939 * We check for this using LOOKUP_CONTINUE and LOOKUP_PARENT.
941 static inline unsigned int nfs_lookup_check_intent(struct nameidata
*nd
, unsigned int mask
)
943 if (nd
->flags
& (LOOKUP_CONTINUE
|LOOKUP_PARENT
))
945 return nd
->flags
& mask
;
949 * Use intent information to check whether or not we're going to do
950 * an O_EXCL create using this path component.
952 static int nfs_is_exclusive_create(struct inode
*dir
, struct nameidata
*nd
)
954 if (NFS_PROTO(dir
)->version
== 2)
956 return nd
&& nfs_lookup_check_intent(nd
, LOOKUP_EXCL
);
960 * Inode and filehandle revalidation for lookups.
962 * We force revalidation in the cases where the VFS sets LOOKUP_REVAL,
963 * or if the intent information indicates that we're about to open this
964 * particular file and the "nocto" mount flag is not set.
968 int nfs_lookup_verify_inode(struct inode
*inode
, struct nameidata
*nd
)
970 struct nfs_server
*server
= NFS_SERVER(inode
);
972 if (test_bit(NFS_INO_MOUNTPOINT
, &NFS_I(inode
)->flags
))
975 /* VFS wants an on-the-wire revalidation */
976 if (nd
->flags
& LOOKUP_REVAL
)
978 /* This is an open(2) */
979 if (nfs_lookup_check_intent(nd
, LOOKUP_OPEN
) != 0 &&
980 !(server
->flags
& NFS_MOUNT_NOCTO
) &&
981 (S_ISREG(inode
->i_mode
) ||
982 S_ISDIR(inode
->i_mode
)))
986 return nfs_revalidate_inode(server
, inode
);
988 return __nfs_revalidate_inode(server
, inode
);
992 * We judge how long we want to trust negative
993 * dentries by looking at the parent inode mtime.
995 * If parent mtime has changed, we revalidate, else we wait for a
996 * period corresponding to the parent's attribute cache timeout value.
999 int nfs_neg_need_reval(struct inode
*dir
, struct dentry
*dentry
,
1000 struct nameidata
*nd
)
1002 /* Don't revalidate a negative dentry if we're creating a new file */
1003 if (nd
!= NULL
&& nfs_lookup_check_intent(nd
, LOOKUP_CREATE
) != 0)
1005 if (NFS_SERVER(dir
)->flags
& NFS_MOUNT_LOOKUP_CACHE_NONEG
)
1007 return !nfs_check_verifier(dir
, dentry
);
1011 * This is called every time the dcache has a lookup hit,
1012 * and we should check whether we can really trust that
1015 * NOTE! The hit can be a negative hit too, don't assume
1018 * If the parent directory is seen to have changed, we throw out the
1019 * cached dentry and do a new lookup.
1021 static int nfs_lookup_revalidate(struct dentry
* dentry
, struct nameidata
*nd
)
1024 struct inode
*inode
;
1025 struct dentry
*parent
;
1026 struct nfs_fh
*fhandle
= NULL
;
1027 struct nfs_fattr
*fattr
= NULL
;
1030 parent
= dget_parent(dentry
);
1031 dir
= parent
->d_inode
;
1032 nfs_inc_stats(dir
, NFSIOS_DENTRYREVALIDATE
);
1033 inode
= dentry
->d_inode
;
1036 if (nfs_neg_need_reval(dir
, dentry
, nd
))
1041 if (is_bad_inode(inode
)) {
1042 dfprintk(LOOKUPCACHE
, "%s: %s/%s has dud inode\n",
1043 __func__
, dentry
->d_parent
->d_name
.name
,
1044 dentry
->d_name
.name
);
1048 if (nfs_have_delegation(inode
, FMODE_READ
))
1049 goto out_set_verifier
;
1051 /* Force a full look up iff the parent directory has changed */
1052 if (!nfs_is_exclusive_create(dir
, nd
) && nfs_check_verifier(dir
, dentry
)) {
1053 if (nfs_lookup_verify_inode(inode
, nd
))
1054 goto out_zap_parent
;
1058 if (NFS_STALE(inode
))
1062 fhandle
= nfs_alloc_fhandle();
1063 fattr
= nfs_alloc_fattr();
1064 if (fhandle
== NULL
|| fattr
== NULL
)
1067 error
= NFS_PROTO(dir
)->lookup(dir
, &dentry
->d_name
, fhandle
, fattr
);
1070 if (nfs_compare_fh(NFS_FH(inode
), fhandle
))
1072 if ((error
= nfs_refresh_inode(inode
, fattr
)) != 0)
1075 nfs_free_fattr(fattr
);
1076 nfs_free_fhandle(fhandle
);
1078 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1081 dfprintk(LOOKUPCACHE
, "NFS: %s(%s/%s) is valid\n",
1082 __func__
, dentry
->d_parent
->d_name
.name
,
1083 dentry
->d_name
.name
);
1086 nfs_zap_caches(dir
);
1088 nfs_mark_for_revalidate(dir
);
1089 if (inode
&& S_ISDIR(inode
->i_mode
)) {
1090 /* Purge readdir caches. */
1091 nfs_zap_caches(inode
);
1092 /* If we have submounts, don't unhash ! */
1093 if (have_submounts(dentry
))
1095 if (dentry
->d_flags
& DCACHE_DISCONNECTED
)
1097 shrink_dcache_parent(dentry
);
1100 nfs_free_fattr(fattr
);
1101 nfs_free_fhandle(fhandle
);
1103 dfprintk(LOOKUPCACHE
, "NFS: %s(%s/%s) is invalid\n",
1104 __func__
, dentry
->d_parent
->d_name
.name
,
1105 dentry
->d_name
.name
);
1108 nfs_free_fattr(fattr
);
1109 nfs_free_fhandle(fhandle
);
1111 dfprintk(LOOKUPCACHE
, "NFS: %s(%s/%s) lookup returned error %d\n",
1112 __func__
, dentry
->d_parent
->d_name
.name
,
1113 dentry
->d_name
.name
, error
);
1118 * This is called from dput() when d_count is going to 0.
1120 static int nfs_dentry_delete(const struct dentry
*dentry
)
1122 dfprintk(VFS
, "NFS: dentry_delete(%s/%s, %x)\n",
1123 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
,
1126 /* Unhash any dentry with a stale inode */
1127 if (dentry
->d_inode
!= NULL
&& NFS_STALE(dentry
->d_inode
))
1130 if (dentry
->d_flags
& DCACHE_NFSFS_RENAMED
) {
1131 /* Unhash it, so that ->d_iput() would be called */
1134 if (!(dentry
->d_sb
->s_flags
& MS_ACTIVE
)) {
1135 /* Unhash it, so that ancestors of killed async unlink
1136 * files will be cleaned up during umount */
1143 static void nfs_drop_nlink(struct inode
*inode
)
1145 spin_lock(&inode
->i_lock
);
1146 if (inode
->i_nlink
> 0)
1148 spin_unlock(&inode
->i_lock
);
1152 * Called when the dentry loses inode.
1153 * We use it to clean up silly-renamed files.
1155 static void nfs_dentry_iput(struct dentry
*dentry
, struct inode
*inode
)
1157 if (S_ISDIR(inode
->i_mode
))
1158 /* drop any readdir cache as it could easily be old */
1159 NFS_I(inode
)->cache_validity
|= NFS_INO_INVALID_DATA
;
1161 if (dentry
->d_flags
& DCACHE_NFSFS_RENAMED
) {
1163 nfs_complete_unlink(dentry
, inode
);
1168 const struct dentry_operations nfs_dentry_operations
= {
1169 .d_revalidate
= nfs_lookup_revalidate
,
1170 .d_delete
= nfs_dentry_delete
,
1171 .d_iput
= nfs_dentry_iput
,
1174 static struct dentry
*nfs_lookup(struct inode
*dir
, struct dentry
* dentry
, struct nameidata
*nd
)
1177 struct dentry
*parent
;
1178 struct inode
*inode
= NULL
;
1179 struct nfs_fh
*fhandle
= NULL
;
1180 struct nfs_fattr
*fattr
= NULL
;
1183 dfprintk(VFS
, "NFS: lookup(%s/%s)\n",
1184 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
);
1185 nfs_inc_stats(dir
, NFSIOS_VFSLOOKUP
);
1187 res
= ERR_PTR(-ENAMETOOLONG
);
1188 if (dentry
->d_name
.len
> NFS_SERVER(dir
)->namelen
)
1191 dentry
->d_op
= NFS_PROTO(dir
)->dentry_ops
;
1194 * If we're doing an exclusive create, optimize away the lookup
1195 * but don't hash the dentry.
1197 if (nfs_is_exclusive_create(dir
, nd
)) {
1198 d_instantiate(dentry
, NULL
);
1203 res
= ERR_PTR(-ENOMEM
);
1204 fhandle
= nfs_alloc_fhandle();
1205 fattr
= nfs_alloc_fattr();
1206 if (fhandle
== NULL
|| fattr
== NULL
)
1209 parent
= dentry
->d_parent
;
1210 /* Protect against concurrent sillydeletes */
1211 nfs_block_sillyrename(parent
);
1212 error
= NFS_PROTO(dir
)->lookup(dir
, &dentry
->d_name
, fhandle
, fattr
);
1213 if (error
== -ENOENT
)
1216 res
= ERR_PTR(error
);
1217 goto out_unblock_sillyrename
;
1219 inode
= nfs_fhget(dentry
->d_sb
, fhandle
, fattr
);
1220 res
= (struct dentry
*)inode
;
1222 goto out_unblock_sillyrename
;
1225 res
= d_materialise_unique(dentry
, inode
);
1228 goto out_unblock_sillyrename
;
1231 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1232 out_unblock_sillyrename
:
1233 nfs_unblock_sillyrename(parent
);
1235 nfs_free_fattr(fattr
);
1236 nfs_free_fhandle(fhandle
);
1240 #ifdef CONFIG_NFS_V4
1241 static int nfs_open_revalidate(struct dentry
*, struct nameidata
*);
1243 const struct dentry_operations nfs4_dentry_operations
= {
1244 .d_revalidate
= nfs_open_revalidate
,
1245 .d_delete
= nfs_dentry_delete
,
1246 .d_iput
= nfs_dentry_iput
,
1250 * Use intent information to determine whether we need to substitute
1251 * the NFSv4-style stateful OPEN for the LOOKUP call
1253 static int is_atomic_open(struct nameidata
*nd
)
1255 if (nd
== NULL
|| nfs_lookup_check_intent(nd
, LOOKUP_OPEN
) == 0)
1257 /* NFS does not (yet) have a stateful open for directories */
1258 if (nd
->flags
& LOOKUP_DIRECTORY
)
1260 /* Are we trying to write to a read only partition? */
1261 if (__mnt_is_readonly(nd
->path
.mnt
) &&
1262 (nd
->intent
.open
.flags
& (O_CREAT
|O_TRUNC
|FMODE_WRITE
)))
1267 static struct nfs_open_context
*nameidata_to_nfs_open_context(struct dentry
*dentry
, struct nameidata
*nd
)
1269 struct path path
= {
1270 .mnt
= nd
->path
.mnt
,
1273 struct nfs_open_context
*ctx
;
1274 struct rpc_cred
*cred
;
1275 fmode_t fmode
= nd
->intent
.open
.flags
& (FMODE_READ
| FMODE_WRITE
| FMODE_EXEC
);
1277 cred
= rpc_lookup_cred();
1279 return ERR_CAST(cred
);
1280 ctx
= alloc_nfs_open_context(&path
, cred
, fmode
);
1283 return ERR_PTR(-ENOMEM
);
1287 static int do_open(struct inode
*inode
, struct file
*filp
)
1289 nfs_fscache_set_inode_cookie(inode
, filp
);
1293 static int nfs_intent_set_file(struct nameidata
*nd
, struct nfs_open_context
*ctx
)
1298 /* If the open_intent is for execute, we have an extra check to make */
1299 if (ctx
->mode
& FMODE_EXEC
) {
1300 ret
= nfs_may_open(ctx
->path
.dentry
->d_inode
,
1302 nd
->intent
.open
.flags
);
1306 filp
= lookup_instantiate_filp(nd
, ctx
->path
.dentry
, do_open
);
1308 ret
= PTR_ERR(filp
);
1310 nfs_file_set_open_context(filp
, ctx
);
1312 put_nfs_open_context(ctx
);
1316 static struct dentry
*nfs_atomic_lookup(struct inode
*dir
, struct dentry
*dentry
, struct nameidata
*nd
)
1318 struct nfs_open_context
*ctx
;
1320 struct dentry
*res
= NULL
;
1321 struct inode
*inode
;
1325 dfprintk(VFS
, "NFS: atomic_lookup(%s/%ld), %s\n",
1326 dir
->i_sb
->s_id
, dir
->i_ino
, dentry
->d_name
.name
);
1328 /* Check that we are indeed trying to open this file */
1329 if (!is_atomic_open(nd
))
1332 if (dentry
->d_name
.len
> NFS_SERVER(dir
)->namelen
) {
1333 res
= ERR_PTR(-ENAMETOOLONG
);
1336 dentry
->d_op
= NFS_PROTO(dir
)->dentry_ops
;
1338 /* Let vfs_create() deal with O_EXCL. Instantiate, but don't hash
1340 if (nd
->flags
& LOOKUP_EXCL
) {
1341 d_instantiate(dentry
, NULL
);
1345 ctx
= nameidata_to_nfs_open_context(dentry
, nd
);
1346 res
= ERR_CAST(ctx
);
1350 open_flags
= nd
->intent
.open
.flags
;
1351 if (nd
->flags
& LOOKUP_CREATE
) {
1352 attr
.ia_mode
= nd
->intent
.open
.create_mode
;
1353 attr
.ia_valid
= ATTR_MODE
;
1354 if (!IS_POSIXACL(dir
))
1355 attr
.ia_mode
&= ~current_umask();
1357 open_flags
&= ~(O_EXCL
| O_CREAT
);
1361 /* Open the file on the server */
1362 nfs_block_sillyrename(dentry
->d_parent
);
1363 inode
= NFS_PROTO(dir
)->open_context(dir
, ctx
, open_flags
, &attr
);
1364 if (IS_ERR(inode
)) {
1365 nfs_unblock_sillyrename(dentry
->d_parent
);
1366 put_nfs_open_context(ctx
);
1367 switch (PTR_ERR(inode
)) {
1368 /* Make a negative dentry */
1370 d_add(dentry
, NULL
);
1373 /* This turned out not to be a regular file */
1377 if (!(nd
->intent
.open
.flags
& O_NOFOLLOW
))
1382 res
= ERR_CAST(inode
);
1386 res
= d_add_unique(dentry
, inode
);
1387 nfs_unblock_sillyrename(dentry
->d_parent
);
1389 dput(ctx
->path
.dentry
);
1390 ctx
->path
.dentry
= dget(res
);
1393 err
= nfs_intent_set_file(nd
, ctx
);
1397 return ERR_PTR(err
);
1400 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1403 return nfs_lookup(dir
, dentry
, nd
);
1406 static int nfs_open_revalidate(struct dentry
*dentry
, struct nameidata
*nd
)
1408 struct dentry
*parent
= NULL
;
1409 struct inode
*inode
= dentry
->d_inode
;
1411 struct nfs_open_context
*ctx
;
1412 int openflags
, ret
= 0;
1414 if (!is_atomic_open(nd
) || d_mountpoint(dentry
))
1417 parent
= dget_parent(dentry
);
1418 dir
= parent
->d_inode
;
1420 /* We can't create new files in nfs_open_revalidate(), so we
1421 * optimize away revalidation of negative dentries.
1423 if (inode
== NULL
) {
1424 if (!nfs_neg_need_reval(dir
, dentry
, nd
))
1429 /* NFS only supports OPEN on regular files */
1430 if (!S_ISREG(inode
->i_mode
))
1432 openflags
= nd
->intent
.open
.flags
;
1433 /* We cannot do exclusive creation on a positive dentry */
1434 if ((openflags
& (O_CREAT
|O_EXCL
)) == (O_CREAT
|O_EXCL
))
1436 /* We can't create new files, or truncate existing ones here */
1437 openflags
&= ~(O_CREAT
|O_EXCL
|O_TRUNC
);
1439 ctx
= nameidata_to_nfs_open_context(dentry
, nd
);
1444 * Note: we're not holding inode->i_mutex and so may be racing with
1445 * operations that change the directory. We therefore save the
1446 * change attribute *before* we do the RPC call.
1448 inode
= NFS_PROTO(dir
)->open_context(dir
, ctx
, openflags
, NULL
);
1449 if (IS_ERR(inode
)) {
1450 ret
= PTR_ERR(inode
);
1463 if (inode
!= dentry
->d_inode
)
1466 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1467 ret
= nfs_intent_set_file(nd
, ctx
);
1477 put_nfs_open_context(ctx
);
1483 return nfs_lookup_revalidate(dentry
, nd
);
1486 static int nfs_open_create(struct inode
*dir
, struct dentry
*dentry
, int mode
,
1487 struct nameidata
*nd
)
1489 struct nfs_open_context
*ctx
= NULL
;
1494 dfprintk(VFS
, "NFS: create(%s/%ld), %s\n",
1495 dir
->i_sb
->s_id
, dir
->i_ino
, dentry
->d_name
.name
);
1497 attr
.ia_mode
= mode
;
1498 attr
.ia_valid
= ATTR_MODE
;
1500 if ((nd
->flags
& LOOKUP_CREATE
) != 0) {
1501 open_flags
= nd
->intent
.open
.flags
;
1503 ctx
= nameidata_to_nfs_open_context(dentry
, nd
);
1504 error
= PTR_ERR(ctx
);
1509 error
= NFS_PROTO(dir
)->create(dir
, dentry
, &attr
, open_flags
, ctx
);
1513 error
= nfs_intent_set_file(nd
, ctx
);
1520 put_nfs_open_context(ctx
);
1527 #endif /* CONFIG_NFSV4 */
1530 * Code common to create, mkdir, and mknod.
1532 int nfs_instantiate(struct dentry
*dentry
, struct nfs_fh
*fhandle
,
1533 struct nfs_fattr
*fattr
)
1535 struct dentry
*parent
= dget_parent(dentry
);
1536 struct inode
*dir
= parent
->d_inode
;
1537 struct inode
*inode
;
1538 int error
= -EACCES
;
1542 /* We may have been initialized further down */
1543 if (dentry
->d_inode
)
1545 if (fhandle
->size
== 0) {
1546 error
= NFS_PROTO(dir
)->lookup(dir
, &dentry
->d_name
, fhandle
, fattr
);
1550 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1551 if (!(fattr
->valid
& NFS_ATTR_FATTR
)) {
1552 struct nfs_server
*server
= NFS_SB(dentry
->d_sb
);
1553 error
= server
->nfs_client
->rpc_ops
->getattr(server
, fhandle
, fattr
);
1557 inode
= nfs_fhget(dentry
->d_sb
, fhandle
, fattr
);
1558 error
= PTR_ERR(inode
);
1561 d_add(dentry
, inode
);
1566 nfs_mark_for_revalidate(dir
);
1572 * Following a failed create operation, we drop the dentry rather
1573 * than retain a negative dentry. This avoids a problem in the event
1574 * that the operation succeeded on the server, but an error in the
1575 * reply path made it appear to have failed.
1577 static int nfs_create(struct inode
*dir
, struct dentry
*dentry
, int mode
,
1578 struct nameidata
*nd
)
1583 dfprintk(VFS
, "NFS: create(%s/%ld), %s\n",
1584 dir
->i_sb
->s_id
, dir
->i_ino
, dentry
->d_name
.name
);
1586 attr
.ia_mode
= mode
;
1587 attr
.ia_valid
= ATTR_MODE
;
1589 error
= NFS_PROTO(dir
)->create(dir
, dentry
, &attr
, 0, NULL
);
1599 * See comments for nfs_proc_create regarding failed operations.
1602 nfs_mknod(struct inode
*dir
, struct dentry
*dentry
, int mode
, dev_t rdev
)
1607 dfprintk(VFS
, "NFS: mknod(%s/%ld), %s\n",
1608 dir
->i_sb
->s_id
, dir
->i_ino
, dentry
->d_name
.name
);
1610 if (!new_valid_dev(rdev
))
1613 attr
.ia_mode
= mode
;
1614 attr
.ia_valid
= ATTR_MODE
;
1616 status
= NFS_PROTO(dir
)->mknod(dir
, dentry
, &attr
, rdev
);
1626 * See comments for nfs_proc_create regarding failed operations.
1628 static int nfs_mkdir(struct inode
*dir
, struct dentry
*dentry
, int mode
)
1633 dfprintk(VFS
, "NFS: mkdir(%s/%ld), %s\n",
1634 dir
->i_sb
->s_id
, dir
->i_ino
, dentry
->d_name
.name
);
1636 attr
.ia_valid
= ATTR_MODE
;
1637 attr
.ia_mode
= mode
| S_IFDIR
;
1639 error
= NFS_PROTO(dir
)->mkdir(dir
, dentry
, &attr
);
1648 static void nfs_dentry_handle_enoent(struct dentry
*dentry
)
1650 if (dentry
->d_inode
!= NULL
&& !d_unhashed(dentry
))
1654 static int nfs_rmdir(struct inode
*dir
, struct dentry
*dentry
)
1658 dfprintk(VFS
, "NFS: rmdir(%s/%ld), %s\n",
1659 dir
->i_sb
->s_id
, dir
->i_ino
, dentry
->d_name
.name
);
1661 error
= NFS_PROTO(dir
)->rmdir(dir
, &dentry
->d_name
);
1662 /* Ensure the VFS deletes this inode */
1663 if (error
== 0 && dentry
->d_inode
!= NULL
)
1664 clear_nlink(dentry
->d_inode
);
1665 else if (error
== -ENOENT
)
1666 nfs_dentry_handle_enoent(dentry
);
1672 * Remove a file after making sure there are no pending writes,
1673 * and after checking that the file has only one user.
1675 * We invalidate the attribute cache and free the inode prior to the operation
1676 * to avoid possible races if the server reuses the inode.
1678 static int nfs_safe_remove(struct dentry
*dentry
)
1680 struct inode
*dir
= dentry
->d_parent
->d_inode
;
1681 struct inode
*inode
= dentry
->d_inode
;
1684 dfprintk(VFS
, "NFS: safe_remove(%s/%s)\n",
1685 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
);
1687 /* If the dentry was sillyrenamed, we simply call d_delete() */
1688 if (dentry
->d_flags
& DCACHE_NFSFS_RENAMED
) {
1693 if (inode
!= NULL
) {
1694 nfs_inode_return_delegation(inode
);
1695 error
= NFS_PROTO(dir
)->remove(dir
, &dentry
->d_name
);
1696 /* The VFS may want to delete this inode */
1698 nfs_drop_nlink(inode
);
1699 nfs_mark_for_revalidate(inode
);
1701 error
= NFS_PROTO(dir
)->remove(dir
, &dentry
->d_name
);
1702 if (error
== -ENOENT
)
1703 nfs_dentry_handle_enoent(dentry
);
1708 /* We do silly rename. In case sillyrename() returns -EBUSY, the inode
1709 * belongs to an active ".nfs..." file and we return -EBUSY.
1711 * If sillyrename() returns 0, we do nothing, otherwise we unlink.
1713 static int nfs_unlink(struct inode
*dir
, struct dentry
*dentry
)
1716 int need_rehash
= 0;
1718 dfprintk(VFS
, "NFS: unlink(%s/%ld, %s)\n", dir
->i_sb
->s_id
,
1719 dir
->i_ino
, dentry
->d_name
.name
);
1721 spin_lock(&dcache_lock
);
1722 spin_lock(&dentry
->d_lock
);
1723 if (atomic_read(&dentry
->d_count
) > 1) {
1724 spin_unlock(&dentry
->d_lock
);
1725 spin_unlock(&dcache_lock
);
1726 /* Start asynchronous writeout of the inode */
1727 write_inode_now(dentry
->d_inode
, 0);
1728 error
= nfs_sillyrename(dir
, dentry
);
1731 if (!d_unhashed(dentry
)) {
1735 spin_unlock(&dentry
->d_lock
);
1736 spin_unlock(&dcache_lock
);
1737 error
= nfs_safe_remove(dentry
);
1738 if (!error
|| error
== -ENOENT
) {
1739 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1740 } else if (need_rehash
)
1746 * To create a symbolic link, most file systems instantiate a new inode,
1747 * add a page to it containing the path, then write it out to the disk
1748 * using prepare_write/commit_write.
1750 * Unfortunately the NFS client can't create the in-core inode first
1751 * because it needs a file handle to create an in-core inode (see
1752 * fs/nfs/inode.c:nfs_fhget). We only have a file handle *after* the
1753 * symlink request has completed on the server.
1755 * So instead we allocate a raw page, copy the symname into it, then do
1756 * the SYMLINK request with the page as the buffer. If it succeeds, we
1757 * now have a new file handle and can instantiate an in-core NFS inode
1758 * and move the raw page into its mapping.
1760 static int nfs_symlink(struct inode
*dir
, struct dentry
*dentry
, const char *symname
)
1762 struct pagevec lru_pvec
;
1766 unsigned int pathlen
= strlen(symname
);
1769 dfprintk(VFS
, "NFS: symlink(%s/%ld, %s, %s)\n", dir
->i_sb
->s_id
,
1770 dir
->i_ino
, dentry
->d_name
.name
, symname
);
1772 if (pathlen
> PAGE_SIZE
)
1773 return -ENAMETOOLONG
;
1775 attr
.ia_mode
= S_IFLNK
| S_IRWXUGO
;
1776 attr
.ia_valid
= ATTR_MODE
;
1778 page
= alloc_page(GFP_HIGHUSER
);
1782 kaddr
= kmap_atomic(page
, KM_USER0
);
1783 memcpy(kaddr
, symname
, pathlen
);
1784 if (pathlen
< PAGE_SIZE
)
1785 memset(kaddr
+ pathlen
, 0, PAGE_SIZE
- pathlen
);
1786 kunmap_atomic(kaddr
, KM_USER0
);
1788 error
= NFS_PROTO(dir
)->symlink(dir
, dentry
, page
, pathlen
, &attr
);
1790 dfprintk(VFS
, "NFS: symlink(%s/%ld, %s, %s) error %d\n",
1791 dir
->i_sb
->s_id
, dir
->i_ino
,
1792 dentry
->d_name
.name
, symname
, error
);
1799 * No big deal if we can't add this page to the page cache here.
1800 * READLINK will get the missing page from the server if needed.
1802 pagevec_init(&lru_pvec
, 0);
1803 if (!add_to_page_cache(page
, dentry
->d_inode
->i_mapping
, 0,
1805 pagevec_add(&lru_pvec
, page
);
1806 pagevec_lru_add_file(&lru_pvec
);
1807 SetPageUptodate(page
);
1816 nfs_link(struct dentry
*old_dentry
, struct inode
*dir
, struct dentry
*dentry
)
1818 struct inode
*inode
= old_dentry
->d_inode
;
1821 dfprintk(VFS
, "NFS: link(%s/%s -> %s/%s)\n",
1822 old_dentry
->d_parent
->d_name
.name
, old_dentry
->d_name
.name
,
1823 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
);
1825 nfs_inode_return_delegation(inode
);
1828 error
= NFS_PROTO(dir
)->link(inode
, dir
, &dentry
->d_name
);
1831 d_add(dentry
, inode
);
1838 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a
1839 * different file handle for the same inode after a rename (e.g. when
1840 * moving to a different directory). A fail-safe method to do so would
1841 * be to look up old_dir/old_name, create a link to new_dir/new_name and
1842 * rename the old file using the sillyrename stuff. This way, the original
1843 * file in old_dir will go away when the last process iput()s the inode.
1847 * It actually works quite well. One needs to have the possibility for
1848 * at least one ".nfs..." file in each directory the file ever gets
1849 * moved or linked to which happens automagically with the new
1850 * implementation that only depends on the dcache stuff instead of
1851 * using the inode layer
1853 * Unfortunately, things are a little more complicated than indicated
1854 * above. For a cross-directory move, we want to make sure we can get
1855 * rid of the old inode after the operation. This means there must be
1856 * no pending writes (if it's a file), and the use count must be 1.
1857 * If these conditions are met, we can drop the dentries before doing
1860 static int nfs_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
1861 struct inode
*new_dir
, struct dentry
*new_dentry
)
1863 struct inode
*old_inode
= old_dentry
->d_inode
;
1864 struct inode
*new_inode
= new_dentry
->d_inode
;
1865 struct dentry
*dentry
= NULL
, *rehash
= NULL
;
1868 dfprintk(VFS
, "NFS: rename(%s/%s -> %s/%s, ct=%d)\n",
1869 old_dentry
->d_parent
->d_name
.name
, old_dentry
->d_name
.name
,
1870 new_dentry
->d_parent
->d_name
.name
, new_dentry
->d_name
.name
,
1871 atomic_read(&new_dentry
->d_count
));
1874 * For non-directories, check whether the target is busy and if so,
1875 * make a copy of the dentry and then do a silly-rename. If the
1876 * silly-rename succeeds, the copied dentry is hashed and becomes
1879 if (new_inode
&& !S_ISDIR(new_inode
->i_mode
)) {
1881 * To prevent any new references to the target during the
1882 * rename, we unhash the dentry in advance.
1884 if (!d_unhashed(new_dentry
)) {
1886 rehash
= new_dentry
;
1889 if (atomic_read(&new_dentry
->d_count
) > 2) {
1892 /* copy the target dentry's name */
1893 dentry
= d_alloc(new_dentry
->d_parent
,
1894 &new_dentry
->d_name
);
1898 /* silly-rename the existing target ... */
1899 err
= nfs_sillyrename(new_dir
, new_dentry
);
1903 new_dentry
= dentry
;
1909 nfs_inode_return_delegation(old_inode
);
1910 if (new_inode
!= NULL
)
1911 nfs_inode_return_delegation(new_inode
);
1913 error
= NFS_PROTO(old_dir
)->rename(old_dir
, &old_dentry
->d_name
,
1914 new_dir
, &new_dentry
->d_name
);
1915 nfs_mark_for_revalidate(old_inode
);
1920 if (new_inode
!= NULL
)
1921 nfs_drop_nlink(new_inode
);
1922 d_move(old_dentry
, new_dentry
);
1923 nfs_set_verifier(new_dentry
,
1924 nfs_save_change_attribute(new_dir
));
1925 } else if (error
== -ENOENT
)
1926 nfs_dentry_handle_enoent(old_dentry
);
1928 /* new dentry created? */
1934 static DEFINE_SPINLOCK(nfs_access_lru_lock
);
1935 static LIST_HEAD(nfs_access_lru_list
);
1936 static atomic_long_t nfs_access_nr_entries
;
1938 static void nfs_access_free_entry(struct nfs_access_entry
*entry
)
1940 put_rpccred(entry
->cred
);
1942 smp_mb__before_atomic_dec();
1943 atomic_long_dec(&nfs_access_nr_entries
);
1944 smp_mb__after_atomic_dec();
1947 static void nfs_access_free_list(struct list_head
*head
)
1949 struct nfs_access_entry
*cache
;
1951 while (!list_empty(head
)) {
1952 cache
= list_entry(head
->next
, struct nfs_access_entry
, lru
);
1953 list_del(&cache
->lru
);
1954 nfs_access_free_entry(cache
);
1958 int nfs_access_cache_shrinker(struct shrinker
*shrink
, int nr_to_scan
, gfp_t gfp_mask
)
1961 struct nfs_inode
*nfsi
, *next
;
1962 struct nfs_access_entry
*cache
;
1964 if ((gfp_mask
& GFP_KERNEL
) != GFP_KERNEL
)
1965 return (nr_to_scan
== 0) ? 0 : -1;
1967 spin_lock(&nfs_access_lru_lock
);
1968 list_for_each_entry_safe(nfsi
, next
, &nfs_access_lru_list
, access_cache_inode_lru
) {
1969 struct inode
*inode
;
1971 if (nr_to_scan
-- == 0)
1973 inode
= &nfsi
->vfs_inode
;
1974 spin_lock(&inode
->i_lock
);
1975 if (list_empty(&nfsi
->access_cache_entry_lru
))
1976 goto remove_lru_entry
;
1977 cache
= list_entry(nfsi
->access_cache_entry_lru
.next
,
1978 struct nfs_access_entry
, lru
);
1979 list_move(&cache
->lru
, &head
);
1980 rb_erase(&cache
->rb_node
, &nfsi
->access_cache
);
1981 if (!list_empty(&nfsi
->access_cache_entry_lru
))
1982 list_move_tail(&nfsi
->access_cache_inode_lru
,
1983 &nfs_access_lru_list
);
1986 list_del_init(&nfsi
->access_cache_inode_lru
);
1987 smp_mb__before_clear_bit();
1988 clear_bit(NFS_INO_ACL_LRU_SET
, &nfsi
->flags
);
1989 smp_mb__after_clear_bit();
1991 spin_unlock(&inode
->i_lock
);
1993 spin_unlock(&nfs_access_lru_lock
);
1994 nfs_access_free_list(&head
);
1995 return (atomic_long_read(&nfs_access_nr_entries
) / 100) * sysctl_vfs_cache_pressure
;
1998 static void __nfs_access_zap_cache(struct nfs_inode
*nfsi
, struct list_head
*head
)
2000 struct rb_root
*root_node
= &nfsi
->access_cache
;
2002 struct nfs_access_entry
*entry
;
2004 /* Unhook entries from the cache */
2005 while ((n
= rb_first(root_node
)) != NULL
) {
2006 entry
= rb_entry(n
, struct nfs_access_entry
, rb_node
);
2007 rb_erase(n
, root_node
);
2008 list_move(&entry
->lru
, head
);
2010 nfsi
->cache_validity
&= ~NFS_INO_INVALID_ACCESS
;
2013 void nfs_access_zap_cache(struct inode
*inode
)
2017 if (test_bit(NFS_INO_ACL_LRU_SET
, &NFS_I(inode
)->flags
) == 0)
2019 /* Remove from global LRU init */
2020 spin_lock(&nfs_access_lru_lock
);
2021 if (test_and_clear_bit(NFS_INO_ACL_LRU_SET
, &NFS_I(inode
)->flags
))
2022 list_del_init(&NFS_I(inode
)->access_cache_inode_lru
);
2024 spin_lock(&inode
->i_lock
);
2025 __nfs_access_zap_cache(NFS_I(inode
), &head
);
2026 spin_unlock(&inode
->i_lock
);
2027 spin_unlock(&nfs_access_lru_lock
);
2028 nfs_access_free_list(&head
);
2031 static struct nfs_access_entry
*nfs_access_search_rbtree(struct inode
*inode
, struct rpc_cred
*cred
)
2033 struct rb_node
*n
= NFS_I(inode
)->access_cache
.rb_node
;
2034 struct nfs_access_entry
*entry
;
2037 entry
= rb_entry(n
, struct nfs_access_entry
, rb_node
);
2039 if (cred
< entry
->cred
)
2041 else if (cred
> entry
->cred
)
2049 static int nfs_access_get_cached(struct inode
*inode
, struct rpc_cred
*cred
, struct nfs_access_entry
*res
)
2051 struct nfs_inode
*nfsi
= NFS_I(inode
);
2052 struct nfs_access_entry
*cache
;
2055 spin_lock(&inode
->i_lock
);
2056 if (nfsi
->cache_validity
& NFS_INO_INVALID_ACCESS
)
2058 cache
= nfs_access_search_rbtree(inode
, cred
);
2061 if (!nfs_have_delegated_attributes(inode
) &&
2062 !time_in_range_open(jiffies
, cache
->jiffies
, cache
->jiffies
+ nfsi
->attrtimeo
))
2064 res
->jiffies
= cache
->jiffies
;
2065 res
->cred
= cache
->cred
;
2066 res
->mask
= cache
->mask
;
2067 list_move_tail(&cache
->lru
, &nfsi
->access_cache_entry_lru
);
2070 spin_unlock(&inode
->i_lock
);
2073 rb_erase(&cache
->rb_node
, &nfsi
->access_cache
);
2074 list_del(&cache
->lru
);
2075 spin_unlock(&inode
->i_lock
);
2076 nfs_access_free_entry(cache
);
2079 spin_unlock(&inode
->i_lock
);
2080 nfs_access_zap_cache(inode
);
2084 static void nfs_access_add_rbtree(struct inode
*inode
, struct nfs_access_entry
*set
)
2086 struct nfs_inode
*nfsi
= NFS_I(inode
);
2087 struct rb_root
*root_node
= &nfsi
->access_cache
;
2088 struct rb_node
**p
= &root_node
->rb_node
;
2089 struct rb_node
*parent
= NULL
;
2090 struct nfs_access_entry
*entry
;
2092 spin_lock(&inode
->i_lock
);
2093 while (*p
!= NULL
) {
2095 entry
= rb_entry(parent
, struct nfs_access_entry
, rb_node
);
2097 if (set
->cred
< entry
->cred
)
2098 p
= &parent
->rb_left
;
2099 else if (set
->cred
> entry
->cred
)
2100 p
= &parent
->rb_right
;
2104 rb_link_node(&set
->rb_node
, parent
, p
);
2105 rb_insert_color(&set
->rb_node
, root_node
);
2106 list_add_tail(&set
->lru
, &nfsi
->access_cache_entry_lru
);
2107 spin_unlock(&inode
->i_lock
);
2110 rb_replace_node(parent
, &set
->rb_node
, root_node
);
2111 list_add_tail(&set
->lru
, &nfsi
->access_cache_entry_lru
);
2112 list_del(&entry
->lru
);
2113 spin_unlock(&inode
->i_lock
);
2114 nfs_access_free_entry(entry
);
2117 static void nfs_access_add_cache(struct inode
*inode
, struct nfs_access_entry
*set
)
2119 struct nfs_access_entry
*cache
= kmalloc(sizeof(*cache
), GFP_KERNEL
);
2122 RB_CLEAR_NODE(&cache
->rb_node
);
2123 cache
->jiffies
= set
->jiffies
;
2124 cache
->cred
= get_rpccred(set
->cred
);
2125 cache
->mask
= set
->mask
;
2127 nfs_access_add_rbtree(inode
, cache
);
2129 /* Update accounting */
2130 smp_mb__before_atomic_inc();
2131 atomic_long_inc(&nfs_access_nr_entries
);
2132 smp_mb__after_atomic_inc();
2134 /* Add inode to global LRU list */
2135 if (!test_bit(NFS_INO_ACL_LRU_SET
, &NFS_I(inode
)->flags
)) {
2136 spin_lock(&nfs_access_lru_lock
);
2137 if (!test_and_set_bit(NFS_INO_ACL_LRU_SET
, &NFS_I(inode
)->flags
))
2138 list_add_tail(&NFS_I(inode
)->access_cache_inode_lru
,
2139 &nfs_access_lru_list
);
2140 spin_unlock(&nfs_access_lru_lock
);
2144 static int nfs_do_access(struct inode
*inode
, struct rpc_cred
*cred
, int mask
)
2146 struct nfs_access_entry cache
;
2149 status
= nfs_access_get_cached(inode
, cred
, &cache
);
2153 /* Be clever: ask server to check for all possible rights */
2154 cache
.mask
= MAY_EXEC
| MAY_WRITE
| MAY_READ
;
2156 cache
.jiffies
= jiffies
;
2157 status
= NFS_PROTO(inode
)->access(inode
, &cache
);
2159 if (status
== -ESTALE
) {
2160 nfs_zap_caches(inode
);
2161 if (!S_ISDIR(inode
->i_mode
))
2162 set_bit(NFS_INO_STALE
, &NFS_I(inode
)->flags
);
2166 nfs_access_add_cache(inode
, &cache
);
2168 if ((mask
& ~cache
.mask
& (MAY_READ
| MAY_WRITE
| MAY_EXEC
)) == 0)
2173 static int nfs_open_permission_mask(int openflags
)
2177 if (openflags
& FMODE_READ
)
2179 if (openflags
& FMODE_WRITE
)
2181 if (openflags
& FMODE_EXEC
)
2186 int nfs_may_open(struct inode
*inode
, struct rpc_cred
*cred
, int openflags
)
2188 return nfs_do_access(inode
, cred
, nfs_open_permission_mask(openflags
));
2191 int nfs_permission(struct inode
*inode
, int mask
)
2193 struct rpc_cred
*cred
;
2196 nfs_inc_stats(inode
, NFSIOS_VFSACCESS
);
2198 if ((mask
& (MAY_READ
| MAY_WRITE
| MAY_EXEC
)) == 0)
2200 /* Is this sys_access() ? */
2201 if (mask
& (MAY_ACCESS
| MAY_CHDIR
))
2204 switch (inode
->i_mode
& S_IFMT
) {
2208 /* NFSv4 has atomic_open... */
2209 if (nfs_server_capable(inode
, NFS_CAP_ATOMIC_OPEN
)
2210 && (mask
& MAY_OPEN
)
2211 && !(mask
& MAY_EXEC
))
2216 * Optimize away all write operations, since the server
2217 * will check permissions when we perform the op.
2219 if ((mask
& MAY_WRITE
) && !(mask
& MAY_READ
))
2224 if (!NFS_PROTO(inode
)->access
)
2227 cred
= rpc_lookup_cred();
2228 if (!IS_ERR(cred
)) {
2229 res
= nfs_do_access(inode
, cred
, mask
);
2232 res
= PTR_ERR(cred
);
2234 if (!res
&& (mask
& MAY_EXEC
) && !execute_ok(inode
))
2237 dfprintk(VFS
, "NFS: permission(%s/%ld), mask=0x%x, res=%d\n",
2238 inode
->i_sb
->s_id
, inode
->i_ino
, mask
, res
);
2241 res
= nfs_revalidate_inode(NFS_SERVER(inode
), inode
);
2243 res
= generic_permission(inode
, mask
, NULL
);
2249 * version-control: t
2250 * kept-new-versions: 5