4 * Copyright (C) 1992 Rick Sladkey
6 * nfs directory handling functions
8 * 10 Apr 1996 Added silly rename for unlink --okir
9 * 28 Sep 1996 Improved directory cache --okir
10 * 23 Aug 1997 Claus Heine claus@momo.math.rwth-aachen.de
11 * Re-implemented silly rename for unlink, newly implemented
12 * silly rename for nfs_rename() following the suggestions
13 * of Olaf Kirch (okir) found in this file.
14 * Following Linus comments on my original hack, this version
15 * depends only on the dcache stuff and doesn't touch the inode
16 * layer (iput() and friends).
17 * 6 Jun 1999 Cache readdir lookups in the page cache. -DaveM
20 #include <linux/time.h>
21 #include <linux/errno.h>
22 #include <linux/stat.h>
23 #include <linux/fcntl.h>
24 #include <linux/string.h>
25 #include <linux/kernel.h>
26 #include <linux/slab.h>
28 #include <linux/sunrpc/clnt.h>
29 #include <linux/nfs_fs.h>
30 #include <linux/nfs_mount.h>
31 #include <linux/pagemap.h>
32 #include <linux/pagevec.h>
33 #include <linux/namei.h>
34 #include <linux/mount.h>
35 #include <linux/sched.h>
36 #include <linux/kmemleak.h>
37 #include <linux/xattr.h>
39 #include "delegation.h"
44 /* #define NFS_DEBUG_VERBOSE 1 */
46 static int nfs_opendir(struct inode
*, struct file
*);
47 static int nfs_closedir(struct inode
*, struct file
*);
48 static int nfs_readdir(struct file
*, void *, filldir_t
);
49 static struct dentry
*nfs_lookup(struct inode
*, struct dentry
*, struct nameidata
*);
50 static int nfs_create(struct inode
*, struct dentry
*, umode_t
, struct nameidata
*);
51 static int nfs_mkdir(struct inode
*, struct dentry
*, umode_t
);
52 static int nfs_rmdir(struct inode
*, struct dentry
*);
53 static int nfs_unlink(struct inode
*, struct dentry
*);
54 static int nfs_symlink(struct inode
*, struct dentry
*, const char *);
55 static int nfs_link(struct dentry
*, struct inode
*, struct dentry
*);
56 static int nfs_mknod(struct inode
*, struct dentry
*, umode_t
, dev_t
);
57 static int nfs_rename(struct inode
*, struct dentry
*,
58 struct inode
*, struct dentry
*);
59 static int nfs_fsync_dir(struct file
*, loff_t
, loff_t
, int);
60 static loff_t
nfs_llseek_dir(struct file
*, loff_t
, int);
61 static void nfs_readdir_clear_array(struct page
*);
63 const struct file_operations nfs_dir_operations
= {
64 .llseek
= nfs_llseek_dir
,
65 .read
= generic_read_dir
,
66 .readdir
= nfs_readdir
,
68 .release
= nfs_closedir
,
69 .fsync
= nfs_fsync_dir
,
72 const struct inode_operations nfs_dir_inode_operations
= {
77 .symlink
= nfs_symlink
,
82 .permission
= nfs_permission
,
83 .getattr
= nfs_getattr
,
84 .setattr
= nfs_setattr
,
87 const struct address_space_operations nfs_dir_aops
= {
88 .freepage
= nfs_readdir_clear_array
,
92 const struct inode_operations nfs3_dir_inode_operations
= {
97 .symlink
= nfs_symlink
,
101 .rename
= nfs_rename
,
102 .permission
= nfs_permission
,
103 .getattr
= nfs_getattr
,
104 .setattr
= nfs_setattr
,
105 .listxattr
= nfs3_listxattr
,
106 .getxattr
= nfs3_getxattr
,
107 .setxattr
= nfs3_setxattr
,
108 .removexattr
= nfs3_removexattr
,
110 #endif /* CONFIG_NFS_V3 */
114 static struct dentry
*nfs_atomic_lookup(struct inode
*, struct dentry
*, struct nameidata
*);
115 static int nfs_open_create(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
, struct nameidata
*nd
);
116 const struct inode_operations nfs4_dir_inode_operations
= {
117 .create
= nfs_open_create
,
118 .lookup
= nfs_atomic_lookup
,
120 .unlink
= nfs_unlink
,
121 .symlink
= nfs_symlink
,
125 .rename
= nfs_rename
,
126 .permission
= nfs_permission
,
127 .getattr
= nfs_getattr
,
128 .setattr
= nfs_setattr
,
129 .getxattr
= generic_getxattr
,
130 .setxattr
= generic_setxattr
,
131 .listxattr
= generic_listxattr
,
132 .removexattr
= generic_removexattr
,
135 #endif /* CONFIG_NFS_V4 */
137 static struct nfs_open_dir_context
*alloc_nfs_open_dir_context(struct inode
*dir
, struct rpc_cred
*cred
)
139 struct nfs_open_dir_context
*ctx
;
140 ctx
= kmalloc(sizeof(*ctx
), GFP_KERNEL
);
143 ctx
->attr_gencount
= NFS_I(dir
)->attr_gencount
;
146 ctx
->cred
= get_rpccred(cred
);
149 return ERR_PTR(-ENOMEM
);
152 static void put_nfs_open_dir_context(struct nfs_open_dir_context
*ctx
)
154 put_rpccred(ctx
->cred
);
162 nfs_opendir(struct inode
*inode
, struct file
*filp
)
165 struct nfs_open_dir_context
*ctx
;
166 struct rpc_cred
*cred
;
168 dfprintk(FILE, "NFS: open dir(%s/%s)\n",
169 filp
->f_path
.dentry
->d_parent
->d_name
.name
,
170 filp
->f_path
.dentry
->d_name
.name
);
172 nfs_inc_stats(inode
, NFSIOS_VFSOPEN
);
174 cred
= rpc_lookup_cred();
176 return PTR_ERR(cred
);
177 ctx
= alloc_nfs_open_dir_context(inode
, cred
);
182 filp
->private_data
= ctx
;
183 if (filp
->f_path
.dentry
== filp
->f_path
.mnt
->mnt_root
) {
184 /* This is a mountpoint, so d_revalidate will never
185 * have been called, so we need to refresh the
186 * inode (for close-open consistency) ourselves.
188 __nfs_revalidate_inode(NFS_SERVER(inode
), inode
);
196 nfs_closedir(struct inode
*inode
, struct file
*filp
)
198 put_nfs_open_dir_context(filp
->private_data
);
202 struct nfs_cache_array_entry
{
206 unsigned char d_type
;
209 struct nfs_cache_array
{
213 struct nfs_cache_array_entry array
[0];
216 typedef int (*decode_dirent_t
)(struct xdr_stream
*, struct nfs_entry
*, int);
220 unsigned long page_index
;
223 loff_t current_index
;
224 decode_dirent_t decode
;
226 unsigned long timestamp
;
227 unsigned long gencount
;
228 unsigned int cache_entry_index
;
231 } nfs_readdir_descriptor_t
;
234 * The caller is responsible for calling nfs_readdir_release_array(page)
237 struct nfs_cache_array
*nfs_readdir_get_array(struct page
*page
)
241 return ERR_PTR(-EIO
);
244 return ERR_PTR(-ENOMEM
);
249 void nfs_readdir_release_array(struct page
*page
)
255 * we are freeing strings created by nfs_add_to_readdir_array()
258 void nfs_readdir_clear_array(struct page
*page
)
260 struct nfs_cache_array
*array
;
263 array
= kmap_atomic(page
);
264 for (i
= 0; i
< array
->size
; i
++)
265 kfree(array
->array
[i
].string
.name
);
266 kunmap_atomic(array
);
270 * the caller is responsible for freeing qstr.name
271 * when called by nfs_readdir_add_to_array, the strings will be freed in
272 * nfs_clear_readdir_array()
275 int nfs_readdir_make_qstr(struct qstr
*string
, const char *name
, unsigned int len
)
278 string
->name
= kmemdup(name
, len
, GFP_KERNEL
);
279 if (string
->name
== NULL
)
282 * Avoid a kmemleak false positive. The pointer to the name is stored
283 * in a page cache page which kmemleak does not scan.
285 kmemleak_not_leak(string
->name
);
286 string
->hash
= full_name_hash(name
, len
);
291 int nfs_readdir_add_to_array(struct nfs_entry
*entry
, struct page
*page
)
293 struct nfs_cache_array
*array
= nfs_readdir_get_array(page
);
294 struct nfs_cache_array_entry
*cache_entry
;
298 return PTR_ERR(array
);
300 cache_entry
= &array
->array
[array
->size
];
302 /* Check that this entry lies within the page bounds */
304 if ((char *)&cache_entry
[1] - (char *)page_address(page
) > PAGE_SIZE
)
307 cache_entry
->cookie
= entry
->prev_cookie
;
308 cache_entry
->ino
= entry
->ino
;
309 cache_entry
->d_type
= entry
->d_type
;
310 ret
= nfs_readdir_make_qstr(&cache_entry
->string
, entry
->name
, entry
->len
);
313 array
->last_cookie
= entry
->cookie
;
316 array
->eof_index
= array
->size
;
318 nfs_readdir_release_array(page
);
323 int nfs_readdir_search_for_pos(struct nfs_cache_array
*array
, nfs_readdir_descriptor_t
*desc
)
325 loff_t diff
= desc
->file
->f_pos
- desc
->current_index
;
330 if (diff
>= array
->size
) {
331 if (array
->eof_index
>= 0)
336 index
= (unsigned int)diff
;
337 *desc
->dir_cookie
= array
->array
[index
].cookie
;
338 desc
->cache_entry_index
= index
;
346 int nfs_readdir_search_for_cookie(struct nfs_cache_array
*array
, nfs_readdir_descriptor_t
*desc
)
350 int status
= -EAGAIN
;
352 for (i
= 0; i
< array
->size
; i
++) {
353 if (array
->array
[i
].cookie
== *desc
->dir_cookie
) {
354 struct nfs_inode
*nfsi
= NFS_I(desc
->file
->f_path
.dentry
->d_inode
);
355 struct nfs_open_dir_context
*ctx
= desc
->file
->private_data
;
357 new_pos
= desc
->current_index
+ i
;
358 if (ctx
->attr_gencount
!= nfsi
->attr_gencount
359 || (nfsi
->cache_validity
& (NFS_INO_INVALID_ATTR
|NFS_INO_INVALID_DATA
))) {
361 ctx
->attr_gencount
= nfsi
->attr_gencount
;
362 } else if (new_pos
< desc
->file
->f_pos
) {
364 && ctx
->dup_cookie
== *desc
->dir_cookie
) {
365 if (printk_ratelimit()) {
366 pr_notice("NFS: directory %s/%s contains a readdir loop."
367 "Please contact your server vendor. "
368 "The file: %s has duplicate cookie %llu\n",
369 desc
->file
->f_dentry
->d_parent
->d_name
.name
,
370 desc
->file
->f_dentry
->d_name
.name
,
371 array
->array
[i
].string
.name
,
377 ctx
->dup_cookie
= *desc
->dir_cookie
;
380 desc
->file
->f_pos
= new_pos
;
381 desc
->cache_entry_index
= i
;
385 if (array
->eof_index
>= 0) {
386 status
= -EBADCOOKIE
;
387 if (*desc
->dir_cookie
== array
->last_cookie
)
395 int nfs_readdir_search_array(nfs_readdir_descriptor_t
*desc
)
397 struct nfs_cache_array
*array
;
400 array
= nfs_readdir_get_array(desc
->page
);
402 status
= PTR_ERR(array
);
406 if (*desc
->dir_cookie
== 0)
407 status
= nfs_readdir_search_for_pos(array
, desc
);
409 status
= nfs_readdir_search_for_cookie(array
, desc
);
411 if (status
== -EAGAIN
) {
412 desc
->last_cookie
= array
->last_cookie
;
413 desc
->current_index
+= array
->size
;
416 nfs_readdir_release_array(desc
->page
);
421 /* Fill a page with xdr information before transferring to the cache page */
423 int nfs_readdir_xdr_filler(struct page
**pages
, nfs_readdir_descriptor_t
*desc
,
424 struct nfs_entry
*entry
, struct file
*file
, struct inode
*inode
)
426 struct nfs_open_dir_context
*ctx
= file
->private_data
;
427 struct rpc_cred
*cred
= ctx
->cred
;
428 unsigned long timestamp
, gencount
;
433 gencount
= nfs_inc_attr_generation_counter();
434 error
= NFS_PROTO(inode
)->readdir(file
->f_path
.dentry
, cred
, entry
->cookie
, pages
,
435 NFS_SERVER(inode
)->dtsize
, desc
->plus
);
437 /* We requested READDIRPLUS, but the server doesn't grok it */
438 if (error
== -ENOTSUPP
&& desc
->plus
) {
439 NFS_SERVER(inode
)->caps
&= ~NFS_CAP_READDIRPLUS
;
440 clear_bit(NFS_INO_ADVISE_RDPLUS
, &NFS_I(inode
)->flags
);
446 desc
->timestamp
= timestamp
;
447 desc
->gencount
= gencount
;
452 static int xdr_decode(nfs_readdir_descriptor_t
*desc
,
453 struct nfs_entry
*entry
, struct xdr_stream
*xdr
)
457 error
= desc
->decode(xdr
, entry
, desc
->plus
);
460 entry
->fattr
->time_start
= desc
->timestamp
;
461 entry
->fattr
->gencount
= desc
->gencount
;
466 int nfs_same_file(struct dentry
*dentry
, struct nfs_entry
*entry
)
468 if (dentry
->d_inode
== NULL
)
470 if (nfs_compare_fh(entry
->fh
, NFS_FH(dentry
->d_inode
)) != 0)
478 bool nfs_use_readdirplus(struct inode
*dir
, struct file
*filp
)
480 if (!nfs_server_capable(dir
, NFS_CAP_READDIRPLUS
))
482 if (test_and_clear_bit(NFS_INO_ADVISE_RDPLUS
, &NFS_I(dir
)->flags
))
484 if (filp
->f_pos
== 0)
490 * This function is called by the lookup code to request the use of
491 * readdirplus to accelerate any future lookups in the same
495 void nfs_advise_use_readdirplus(struct inode
*dir
)
497 set_bit(NFS_INO_ADVISE_RDPLUS
, &NFS_I(dir
)->flags
);
501 void nfs_prime_dcache(struct dentry
*parent
, struct nfs_entry
*entry
)
503 struct qstr filename
= QSTR_INIT(entry
->name
, entry
->len
);
504 struct dentry
*dentry
;
505 struct dentry
*alias
;
506 struct inode
*dir
= parent
->d_inode
;
509 if (filename
.name
[0] == '.') {
510 if (filename
.len
== 1)
512 if (filename
.len
== 2 && filename
.name
[1] == '.')
515 filename
.hash
= full_name_hash(filename
.name
, filename
.len
);
517 dentry
= d_lookup(parent
, &filename
);
518 if (dentry
!= NULL
) {
519 if (nfs_same_file(dentry
, entry
)) {
520 nfs_refresh_inode(dentry
->d_inode
, entry
->fattr
);
528 dentry
= d_alloc(parent
, &filename
);
532 inode
= nfs_fhget(dentry
->d_sb
, entry
->fh
, entry
->fattr
);
536 alias
= d_materialise_unique(dentry
, inode
);
540 nfs_set_verifier(alias
, nfs_save_change_attribute(dir
));
543 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
549 /* Perform conversion from xdr to cache array */
551 int nfs_readdir_page_filler(nfs_readdir_descriptor_t
*desc
, struct nfs_entry
*entry
,
552 struct page
**xdr_pages
, struct page
*page
, unsigned int buflen
)
554 struct xdr_stream stream
;
556 struct page
*scratch
;
557 struct nfs_cache_array
*array
;
558 unsigned int count
= 0;
561 scratch
= alloc_page(GFP_KERNEL
);
565 xdr_init_decode_pages(&stream
, &buf
, xdr_pages
, buflen
);
566 xdr_set_scratch_buffer(&stream
, page_address(scratch
), PAGE_SIZE
);
569 status
= xdr_decode(desc
, entry
, &stream
);
571 if (status
== -EAGAIN
)
579 nfs_prime_dcache(desc
->file
->f_path
.dentry
, entry
);
581 status
= nfs_readdir_add_to_array(entry
, page
);
584 } while (!entry
->eof
);
586 if (count
== 0 || (status
== -EBADCOOKIE
&& entry
->eof
!= 0)) {
587 array
= nfs_readdir_get_array(page
);
588 if (!IS_ERR(array
)) {
589 array
->eof_index
= array
->size
;
591 nfs_readdir_release_array(page
);
593 status
= PTR_ERR(array
);
601 void nfs_readdir_free_pagearray(struct page
**pages
, unsigned int npages
)
604 for (i
= 0; i
< npages
; i
++)
609 void nfs_readdir_free_large_page(void *ptr
, struct page
**pages
,
612 nfs_readdir_free_pagearray(pages
, npages
);
616 * nfs_readdir_large_page will allocate pages that must be freed with a call
617 * to nfs_readdir_free_large_page
620 int nfs_readdir_large_page(struct page
**pages
, unsigned int npages
)
624 for (i
= 0; i
< npages
; i
++) {
625 struct page
*page
= alloc_page(GFP_KERNEL
);
633 nfs_readdir_free_pagearray(pages
, i
);
638 int nfs_readdir_xdr_to_array(nfs_readdir_descriptor_t
*desc
, struct page
*page
, struct inode
*inode
)
640 struct page
*pages
[NFS_MAX_READDIR_PAGES
];
641 void *pages_ptr
= NULL
;
642 struct nfs_entry entry
;
643 struct file
*file
= desc
->file
;
644 struct nfs_cache_array
*array
;
645 int status
= -ENOMEM
;
646 unsigned int array_size
= ARRAY_SIZE(pages
);
648 entry
.prev_cookie
= 0;
649 entry
.cookie
= desc
->last_cookie
;
651 entry
.fh
= nfs_alloc_fhandle();
652 entry
.fattr
= nfs_alloc_fattr();
653 entry
.server
= NFS_SERVER(inode
);
654 if (entry
.fh
== NULL
|| entry
.fattr
== NULL
)
657 array
= nfs_readdir_get_array(page
);
659 status
= PTR_ERR(array
);
662 memset(array
, 0, sizeof(struct nfs_cache_array
));
663 array
->eof_index
= -1;
665 status
= nfs_readdir_large_page(pages
, array_size
);
667 goto out_release_array
;
670 status
= nfs_readdir_xdr_filler(pages
, desc
, &entry
, file
, inode
);
675 status
= nfs_readdir_page_filler(desc
, &entry
, pages
, page
, pglen
);
677 if (status
== -ENOSPC
)
681 } while (array
->eof_index
< 0);
683 nfs_readdir_free_large_page(pages_ptr
, pages
, array_size
);
685 nfs_readdir_release_array(page
);
687 nfs_free_fattr(entry
.fattr
);
688 nfs_free_fhandle(entry
.fh
);
693 * Now we cache directories properly, by converting xdr information
694 * to an array that can be used for lookups later. This results in
695 * fewer cache pages, since we can store more information on each page.
696 * We only need to convert from xdr once so future lookups are much simpler
699 int nfs_readdir_filler(nfs_readdir_descriptor_t
*desc
, struct page
* page
)
701 struct inode
*inode
= desc
->file
->f_path
.dentry
->d_inode
;
704 ret
= nfs_readdir_xdr_to_array(desc
, page
, inode
);
707 SetPageUptodate(page
);
709 if (invalidate_inode_pages2_range(inode
->i_mapping
, page
->index
+ 1, -1) < 0) {
710 /* Should never happen */
711 nfs_zap_mapping(inode
, inode
->i_mapping
);
721 void cache_page_release(nfs_readdir_descriptor_t
*desc
)
723 if (!desc
->page
->mapping
)
724 nfs_readdir_clear_array(desc
->page
);
725 page_cache_release(desc
->page
);
730 struct page
*get_cache_page(nfs_readdir_descriptor_t
*desc
)
732 return read_cache_page(desc
->file
->f_path
.dentry
->d_inode
->i_mapping
,
733 desc
->page_index
, (filler_t
*)nfs_readdir_filler
, desc
);
737 * Returns 0 if desc->dir_cookie was found on page desc->page_index
740 int find_cache_page(nfs_readdir_descriptor_t
*desc
)
744 desc
->page
= get_cache_page(desc
);
745 if (IS_ERR(desc
->page
))
746 return PTR_ERR(desc
->page
);
748 res
= nfs_readdir_search_array(desc
);
750 cache_page_release(desc
);
754 /* Search for desc->dir_cookie from the beginning of the page cache */
756 int readdir_search_pagecache(nfs_readdir_descriptor_t
*desc
)
760 if (desc
->page_index
== 0) {
761 desc
->current_index
= 0;
762 desc
->last_cookie
= 0;
765 res
= find_cache_page(desc
);
766 } while (res
== -EAGAIN
);
771 * Once we've found the start of the dirent within a page: fill 'er up...
774 int nfs_do_filldir(nfs_readdir_descriptor_t
*desc
, void *dirent
,
777 struct file
*file
= desc
->file
;
780 struct nfs_cache_array
*array
= NULL
;
781 struct nfs_open_dir_context
*ctx
= file
->private_data
;
783 array
= nfs_readdir_get_array(desc
->page
);
785 res
= PTR_ERR(array
);
789 for (i
= desc
->cache_entry_index
; i
< array
->size
; i
++) {
790 struct nfs_cache_array_entry
*ent
;
792 ent
= &array
->array
[i
];
793 if (filldir(dirent
, ent
->string
.name
, ent
->string
.len
,
794 file
->f_pos
, nfs_compat_user_ino64(ent
->ino
),
800 if (i
< (array
->size
-1))
801 *desc
->dir_cookie
= array
->array
[i
+1].cookie
;
803 *desc
->dir_cookie
= array
->last_cookie
;
807 if (array
->eof_index
>= 0)
810 nfs_readdir_release_array(desc
->page
);
812 cache_page_release(desc
);
813 dfprintk(DIRCACHE
, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n",
814 (unsigned long long)*desc
->dir_cookie
, res
);
819 * If we cannot find a cookie in our cache, we suspect that this is
820 * because it points to a deleted file, so we ask the server to return
821 * whatever it thinks is the next entry. We then feed this to filldir.
822 * If all goes well, we should then be able to find our way round the
823 * cache on the next call to readdir_search_pagecache();
825 * NOTE: we cannot add the anonymous page to the pagecache because
826 * the data it contains might not be page aligned. Besides,
827 * we should already have a complete representation of the
828 * directory in the page cache by the time we get here.
831 int uncached_readdir(nfs_readdir_descriptor_t
*desc
, void *dirent
,
834 struct page
*page
= NULL
;
836 struct inode
*inode
= desc
->file
->f_path
.dentry
->d_inode
;
837 struct nfs_open_dir_context
*ctx
= desc
->file
->private_data
;
839 dfprintk(DIRCACHE
, "NFS: uncached_readdir() searching for cookie %Lu\n",
840 (unsigned long long)*desc
->dir_cookie
);
842 page
= alloc_page(GFP_HIGHUSER
);
848 desc
->page_index
= 0;
849 desc
->last_cookie
= *desc
->dir_cookie
;
853 status
= nfs_readdir_xdr_to_array(desc
, page
, inode
);
857 status
= nfs_do_filldir(desc
, dirent
, filldir
);
860 dfprintk(DIRCACHE
, "NFS: %s: returns %d\n",
864 cache_page_release(desc
);
868 /* The file offset position represents the dirent entry number. A
869 last cookie cache takes care of the common case of reading the
872 static int nfs_readdir(struct file
*filp
, void *dirent
, filldir_t filldir
)
874 struct dentry
*dentry
= filp
->f_path
.dentry
;
875 struct inode
*inode
= dentry
->d_inode
;
876 nfs_readdir_descriptor_t my_desc
,
878 struct nfs_open_dir_context
*dir_ctx
= filp
->private_data
;
881 dfprintk(FILE, "NFS: readdir(%s/%s) starting at cookie %llu\n",
882 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
,
883 (long long)filp
->f_pos
);
884 nfs_inc_stats(inode
, NFSIOS_VFSGETDENTS
);
887 * filp->f_pos points to the dirent entry number.
888 * *desc->dir_cookie has the cookie for the next entry. We have
889 * to either find the entry with the appropriate number or
890 * revalidate the cookie.
892 memset(desc
, 0, sizeof(*desc
));
895 desc
->dir_cookie
= &dir_ctx
->dir_cookie
;
896 desc
->decode
= NFS_PROTO(inode
)->decode_dirent
;
897 desc
->plus
= nfs_use_readdirplus(inode
, filp
) ? 1 : 0;
899 nfs_block_sillyrename(dentry
);
900 res
= nfs_revalidate_mapping(inode
, filp
->f_mapping
);
905 res
= readdir_search_pagecache(desc
);
907 if (res
== -EBADCOOKIE
) {
909 /* This means either end of directory */
910 if (*desc
->dir_cookie
&& desc
->eof
== 0) {
911 /* Or that the server has 'lost' a cookie */
912 res
= uncached_readdir(desc
, dirent
, filldir
);
918 if (res
== -ETOOSMALL
&& desc
->plus
) {
919 clear_bit(NFS_INO_ADVISE_RDPLUS
, &NFS_I(inode
)->flags
);
920 nfs_zap_caches(inode
);
921 desc
->page_index
= 0;
929 res
= nfs_do_filldir(desc
, dirent
, filldir
);
932 } while (!desc
->eof
);
934 nfs_unblock_sillyrename(dentry
);
937 dfprintk(FILE, "NFS: readdir(%s/%s) returns %d\n",
938 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
,
943 static loff_t
nfs_llseek_dir(struct file
*filp
, loff_t offset
, int origin
)
945 struct dentry
*dentry
= filp
->f_path
.dentry
;
946 struct inode
*inode
= dentry
->d_inode
;
947 struct nfs_open_dir_context
*dir_ctx
= filp
->private_data
;
949 dfprintk(FILE, "NFS: llseek dir(%s/%s, %lld, %d)\n",
950 dentry
->d_parent
->d_name
.name
,
954 mutex_lock(&inode
->i_mutex
);
957 offset
+= filp
->f_pos
;
965 if (offset
!= filp
->f_pos
) {
966 filp
->f_pos
= offset
;
967 dir_ctx
->dir_cookie
= 0;
971 mutex_unlock(&inode
->i_mutex
);
976 * All directory operations under NFS are synchronous, so fsync()
977 * is a dummy operation.
979 static int nfs_fsync_dir(struct file
*filp
, loff_t start
, loff_t end
,
982 struct dentry
*dentry
= filp
->f_path
.dentry
;
983 struct inode
*inode
= dentry
->d_inode
;
985 dfprintk(FILE, "NFS: fsync dir(%s/%s) datasync %d\n",
986 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
,
989 mutex_lock(&inode
->i_mutex
);
990 nfs_inc_stats(dentry
->d_inode
, NFSIOS_VFSFSYNC
);
991 mutex_unlock(&inode
->i_mutex
);
996 * nfs_force_lookup_revalidate - Mark the directory as having changed
997 * @dir - pointer to directory inode
999 * This forces the revalidation code in nfs_lookup_revalidate() to do a
1000 * full lookup on all child dentries of 'dir' whenever a change occurs
1001 * on the server that might have invalidated our dcache.
1003 * The caller should be holding dir->i_lock
1005 void nfs_force_lookup_revalidate(struct inode
*dir
)
1007 NFS_I(dir
)->cache_change_attribute
++;
1011 * A check for whether or not the parent directory has changed.
1012 * In the case it has, we assume that the dentries are untrustworthy
1013 * and may need to be looked up again.
1015 static int nfs_check_verifier(struct inode
*dir
, struct dentry
*dentry
)
1017 if (IS_ROOT(dentry
))
1019 if (NFS_SERVER(dir
)->flags
& NFS_MOUNT_LOOKUP_CACHE_NONE
)
1021 if (!nfs_verify_change_attribute(dir
, dentry
->d_time
))
1023 /* Revalidate nfsi->cache_change_attribute before we declare a match */
1024 if (nfs_revalidate_inode(NFS_SERVER(dir
), dir
) < 0)
1026 if (!nfs_verify_change_attribute(dir
, dentry
->d_time
))
1032 * Return the intent data that applies to this particular path component
1034 * Note that the current set of intents only apply to the very last
1035 * component of the path and none of them is set before that last
1038 static inline unsigned int nfs_lookup_check_intent(struct nameidata
*nd
,
1041 return nd
->flags
& mask
;
1045 * Use intent information to check whether or not we're going to do
1046 * an O_EXCL create using this path component.
1048 static int nfs_is_exclusive_create(struct inode
*dir
, struct nameidata
*nd
)
1050 if (NFS_PROTO(dir
)->version
== 2)
1052 return nd
&& nfs_lookup_check_intent(nd
, LOOKUP_EXCL
);
1056 * Inode and filehandle revalidation for lookups.
1058 * We force revalidation in the cases where the VFS sets LOOKUP_REVAL,
1059 * or if the intent information indicates that we're about to open this
1060 * particular file and the "nocto" mount flag is not set.
1064 int nfs_lookup_verify_inode(struct inode
*inode
, struct nameidata
*nd
)
1066 struct nfs_server
*server
= NFS_SERVER(inode
);
1068 if (IS_AUTOMOUNT(inode
))
1071 /* VFS wants an on-the-wire revalidation */
1072 if (nd
->flags
& LOOKUP_REVAL
)
1074 /* This is an open(2) */
1075 if (nfs_lookup_check_intent(nd
, LOOKUP_OPEN
) != 0 &&
1076 !(server
->flags
& NFS_MOUNT_NOCTO
) &&
1077 (S_ISREG(inode
->i_mode
) ||
1078 S_ISDIR(inode
->i_mode
)))
1082 return nfs_revalidate_inode(server
, inode
);
1084 return __nfs_revalidate_inode(server
, inode
);
1088 * We judge how long we want to trust negative
1089 * dentries by looking at the parent inode mtime.
1091 * If parent mtime has changed, we revalidate, else we wait for a
1092 * period corresponding to the parent's attribute cache timeout value.
1095 int nfs_neg_need_reval(struct inode
*dir
, struct dentry
*dentry
,
1096 struct nameidata
*nd
)
1098 /* Don't revalidate a negative dentry if we're creating a new file */
1099 if (nd
!= NULL
&& nfs_lookup_check_intent(nd
, LOOKUP_CREATE
) != 0)
1101 if (NFS_SERVER(dir
)->flags
& NFS_MOUNT_LOOKUP_CACHE_NONEG
)
1103 return !nfs_check_verifier(dir
, dentry
);
1107 * This is called every time the dcache has a lookup hit,
1108 * and we should check whether we can really trust that
1111 * NOTE! The hit can be a negative hit too, don't assume
1114 * If the parent directory is seen to have changed, we throw out the
1115 * cached dentry and do a new lookup.
1117 static int nfs_lookup_revalidate(struct dentry
*dentry
, struct nameidata
*nd
)
1120 struct inode
*inode
;
1121 struct dentry
*parent
;
1122 struct nfs_fh
*fhandle
= NULL
;
1123 struct nfs_fattr
*fattr
= NULL
;
1126 if (nd
->flags
& LOOKUP_RCU
)
1129 parent
= dget_parent(dentry
);
1130 dir
= parent
->d_inode
;
1131 nfs_inc_stats(dir
, NFSIOS_DENTRYREVALIDATE
);
1132 inode
= dentry
->d_inode
;
1135 if (nfs_neg_need_reval(dir
, dentry
, nd
))
1137 goto out_valid_noent
;
1140 if (is_bad_inode(inode
)) {
1141 dfprintk(LOOKUPCACHE
, "%s: %s/%s has dud inode\n",
1142 __func__
, dentry
->d_parent
->d_name
.name
,
1143 dentry
->d_name
.name
);
1147 if (nfs_have_delegation(inode
, FMODE_READ
))
1148 goto out_set_verifier
;
1150 /* Force a full look up iff the parent directory has changed */
1151 if (!nfs_is_exclusive_create(dir
, nd
) && nfs_check_verifier(dir
, dentry
)) {
1152 if (nfs_lookup_verify_inode(inode
, nd
))
1153 goto out_zap_parent
;
1157 if (NFS_STALE(inode
))
1161 fhandle
= nfs_alloc_fhandle();
1162 fattr
= nfs_alloc_fattr();
1163 if (fhandle
== NULL
|| fattr
== NULL
)
1166 error
= NFS_PROTO(dir
)->lookup(dir
, &dentry
->d_name
, fhandle
, fattr
);
1169 if (nfs_compare_fh(NFS_FH(inode
), fhandle
))
1171 if ((error
= nfs_refresh_inode(inode
, fattr
)) != 0)
1174 nfs_free_fattr(fattr
);
1175 nfs_free_fhandle(fhandle
);
1177 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1179 /* Success: notify readdir to use READDIRPLUS */
1180 nfs_advise_use_readdirplus(dir
);
1183 dfprintk(LOOKUPCACHE
, "NFS: %s(%s/%s) is valid\n",
1184 __func__
, dentry
->d_parent
->d_name
.name
,
1185 dentry
->d_name
.name
);
1188 nfs_zap_caches(dir
);
1190 nfs_mark_for_revalidate(dir
);
1191 if (inode
&& S_ISDIR(inode
->i_mode
)) {
1192 /* Purge readdir caches. */
1193 nfs_zap_caches(inode
);
1194 /* If we have submounts, don't unhash ! */
1195 if (have_submounts(dentry
))
1197 if (dentry
->d_flags
& DCACHE_DISCONNECTED
)
1199 shrink_dcache_parent(dentry
);
1202 nfs_free_fattr(fattr
);
1203 nfs_free_fhandle(fhandle
);
1205 dfprintk(LOOKUPCACHE
, "NFS: %s(%s/%s) is invalid\n",
1206 __func__
, dentry
->d_parent
->d_name
.name
,
1207 dentry
->d_name
.name
);
1210 nfs_free_fattr(fattr
);
1211 nfs_free_fhandle(fhandle
);
1213 dfprintk(LOOKUPCACHE
, "NFS: %s(%s/%s) lookup returned error %d\n",
1214 __func__
, dentry
->d_parent
->d_name
.name
,
1215 dentry
->d_name
.name
, error
);
1220 * This is called from dput() when d_count is going to 0.
1222 static int nfs_dentry_delete(const struct dentry
*dentry
)
1224 dfprintk(VFS
, "NFS: dentry_delete(%s/%s, %x)\n",
1225 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
,
1228 /* Unhash any dentry with a stale inode */
1229 if (dentry
->d_inode
!= NULL
&& NFS_STALE(dentry
->d_inode
))
1232 if (dentry
->d_flags
& DCACHE_NFSFS_RENAMED
) {
1233 /* Unhash it, so that ->d_iput() would be called */
1236 if (!(dentry
->d_sb
->s_flags
& MS_ACTIVE
)) {
1237 /* Unhash it, so that ancestors of killed async unlink
1238 * files will be cleaned up during umount */
1245 static void nfs_drop_nlink(struct inode
*inode
)
1247 spin_lock(&inode
->i_lock
);
1248 if (inode
->i_nlink
> 0)
1250 spin_unlock(&inode
->i_lock
);
1254 * Called when the dentry loses inode.
1255 * We use it to clean up silly-renamed files.
1257 static void nfs_dentry_iput(struct dentry
*dentry
, struct inode
*inode
)
1259 if (S_ISDIR(inode
->i_mode
))
1260 /* drop any readdir cache as it could easily be old */
1261 NFS_I(inode
)->cache_validity
|= NFS_INO_INVALID_DATA
;
1263 if (dentry
->d_flags
& DCACHE_NFSFS_RENAMED
) {
1265 nfs_complete_unlink(dentry
, inode
);
1270 static void nfs_d_release(struct dentry
*dentry
)
1272 /* free cached devname value, if it survived that far */
1273 if (unlikely(dentry
->d_fsdata
)) {
1274 if (dentry
->d_flags
& DCACHE_NFSFS_RENAMED
)
1277 kfree(dentry
->d_fsdata
);
1281 const struct dentry_operations nfs_dentry_operations
= {
1282 .d_revalidate
= nfs_lookup_revalidate
,
1283 .d_delete
= nfs_dentry_delete
,
1284 .d_iput
= nfs_dentry_iput
,
1285 .d_automount
= nfs_d_automount
,
1286 .d_release
= nfs_d_release
,
1289 static struct dentry
*nfs_lookup(struct inode
*dir
, struct dentry
* dentry
, struct nameidata
*nd
)
1292 struct dentry
*parent
;
1293 struct inode
*inode
= NULL
;
1294 struct nfs_fh
*fhandle
= NULL
;
1295 struct nfs_fattr
*fattr
= NULL
;
1298 dfprintk(VFS
, "NFS: lookup(%s/%s)\n",
1299 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
);
1300 nfs_inc_stats(dir
, NFSIOS_VFSLOOKUP
);
1302 res
= ERR_PTR(-ENAMETOOLONG
);
1303 if (dentry
->d_name
.len
> NFS_SERVER(dir
)->namelen
)
1307 * If we're doing an exclusive create, optimize away the lookup
1308 * but don't hash the dentry.
1310 if (nfs_is_exclusive_create(dir
, nd
)) {
1311 d_instantiate(dentry
, NULL
);
1316 res
= ERR_PTR(-ENOMEM
);
1317 fhandle
= nfs_alloc_fhandle();
1318 fattr
= nfs_alloc_fattr();
1319 if (fhandle
== NULL
|| fattr
== NULL
)
1322 parent
= dentry
->d_parent
;
1323 /* Protect against concurrent sillydeletes */
1324 nfs_block_sillyrename(parent
);
1325 error
= NFS_PROTO(dir
)->lookup(dir
, &dentry
->d_name
, fhandle
, fattr
);
1326 if (error
== -ENOENT
)
1329 res
= ERR_PTR(error
);
1330 goto out_unblock_sillyrename
;
1332 inode
= nfs_fhget(dentry
->d_sb
, fhandle
, fattr
);
1333 res
= ERR_CAST(inode
);
1335 goto out_unblock_sillyrename
;
1337 /* Success: notify readdir to use READDIRPLUS */
1338 nfs_advise_use_readdirplus(dir
);
1341 res
= d_materialise_unique(dentry
, inode
);
1344 goto out_unblock_sillyrename
;
1347 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1348 out_unblock_sillyrename
:
1349 nfs_unblock_sillyrename(parent
);
1351 nfs_free_fattr(fattr
);
1352 nfs_free_fhandle(fhandle
);
1356 #ifdef CONFIG_NFS_V4
1357 static int nfs_open_revalidate(struct dentry
*, struct nameidata
*);
1359 const struct dentry_operations nfs4_dentry_operations
= {
1360 .d_revalidate
= nfs_open_revalidate
,
1361 .d_delete
= nfs_dentry_delete
,
1362 .d_iput
= nfs_dentry_iput
,
1363 .d_automount
= nfs_d_automount
,
1364 .d_release
= nfs_d_release
,
1368 * Use intent information to determine whether we need to substitute
1369 * the NFSv4-style stateful OPEN for the LOOKUP call
1371 static int is_atomic_open(struct nameidata
*nd
)
1373 if (nd
== NULL
|| nfs_lookup_check_intent(nd
, LOOKUP_OPEN
) == 0)
1375 /* NFS does not (yet) have a stateful open for directories */
1376 if (nd
->flags
& LOOKUP_DIRECTORY
)
1378 /* Are we trying to write to a read only partition? */
1379 if (__mnt_is_readonly(nd
->path
.mnt
) &&
1380 (nd
->intent
.open
.flags
& (O_CREAT
|O_TRUNC
|O_ACCMODE
)))
1385 static fmode_t
flags_to_mode(int flags
)
1387 fmode_t res
= (__force fmode_t
)flags
& FMODE_EXEC
;
1388 if ((flags
& O_ACCMODE
) != O_WRONLY
)
1390 if ((flags
& O_ACCMODE
) != O_RDONLY
)
1395 static struct nfs_open_context
*create_nfs_open_context(struct dentry
*dentry
, int open_flags
)
1397 return alloc_nfs_open_context(dentry
, flags_to_mode(open_flags
));
1400 static int do_open(struct inode
*inode
, struct file
*filp
)
1402 nfs_fscache_set_inode_cookie(inode
, filp
);
1406 static int nfs_intent_set_file(struct nameidata
*nd
, struct nfs_open_context
*ctx
)
1411 /* If the open_intent is for execute, we have an extra check to make */
1412 if (ctx
->mode
& FMODE_EXEC
) {
1413 ret
= nfs_may_open(ctx
->dentry
->d_inode
,
1415 nd
->intent
.open
.flags
);
1419 filp
= lookup_instantiate_filp(nd
, ctx
->dentry
, do_open
);
1421 ret
= PTR_ERR(filp
);
1423 nfs_file_set_open_context(filp
, ctx
);
1425 put_nfs_open_context(ctx
);
1429 static struct dentry
*nfs_atomic_lookup(struct inode
*dir
, struct dentry
*dentry
, struct nameidata
*nd
)
1431 struct nfs_open_context
*ctx
;
1433 struct dentry
*res
= NULL
;
1434 struct inode
*inode
;
1438 dfprintk(VFS
, "NFS: atomic_lookup(%s/%ld), %s\n",
1439 dir
->i_sb
->s_id
, dir
->i_ino
, dentry
->d_name
.name
);
1441 /* Check that we are indeed trying to open this file */
1442 if (!is_atomic_open(nd
))
1445 if (dentry
->d_name
.len
> NFS_SERVER(dir
)->namelen
) {
1446 res
= ERR_PTR(-ENAMETOOLONG
);
1450 /* Let vfs_create() deal with O_EXCL. Instantiate, but don't hash
1452 if (nd
->flags
& LOOKUP_EXCL
) {
1453 d_instantiate(dentry
, NULL
);
1457 open_flags
= nd
->intent
.open
.flags
;
1458 attr
.ia_valid
= ATTR_OPEN
;
1460 ctx
= create_nfs_open_context(dentry
, open_flags
);
1461 res
= ERR_CAST(ctx
);
1465 if (nd
->flags
& LOOKUP_CREATE
) {
1466 attr
.ia_mode
= nd
->intent
.open
.create_mode
;
1467 attr
.ia_valid
|= ATTR_MODE
;
1468 attr
.ia_mode
&= ~current_umask();
1470 open_flags
&= ~(O_EXCL
| O_CREAT
);
1472 if (open_flags
& O_TRUNC
) {
1473 attr
.ia_valid
|= ATTR_SIZE
;
1477 /* Open the file on the server */
1478 nfs_block_sillyrename(dentry
->d_parent
);
1479 inode
= NFS_PROTO(dir
)->open_context(dir
, ctx
, open_flags
, &attr
);
1480 if (IS_ERR(inode
)) {
1481 nfs_unblock_sillyrename(dentry
->d_parent
);
1482 put_nfs_open_context(ctx
);
1483 switch (PTR_ERR(inode
)) {
1484 /* Make a negative dentry */
1486 d_add(dentry
, NULL
);
1489 /* This turned out not to be a regular file */
1494 if (!(nd
->intent
.open
.flags
& O_NOFOLLOW
))
1498 res
= ERR_CAST(inode
);
1502 res
= d_add_unique(dentry
, inode
);
1503 nfs_unblock_sillyrename(dentry
->d_parent
);
1506 ctx
->dentry
= dget(res
);
1509 err
= nfs_intent_set_file(nd
, ctx
);
1513 return ERR_PTR(err
);
1516 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1519 return nfs_lookup(dir
, dentry
, nd
);
1522 static int nfs_open_revalidate(struct dentry
*dentry
, struct nameidata
*nd
)
1524 struct dentry
*parent
= NULL
;
1525 struct inode
*inode
;
1527 struct nfs_open_context
*ctx
;
1529 int openflags
, ret
= 0;
1531 if (nd
->flags
& LOOKUP_RCU
)
1534 inode
= dentry
->d_inode
;
1535 if (!is_atomic_open(nd
) || d_mountpoint(dentry
))
1538 parent
= dget_parent(dentry
);
1539 dir
= parent
->d_inode
;
1541 /* We can't create new files in nfs_open_revalidate(), so we
1542 * optimize away revalidation of negative dentries.
1544 if (inode
== NULL
) {
1545 if (!nfs_neg_need_reval(dir
, dentry
, nd
))
1550 /* NFS only supports OPEN on regular files */
1551 if (!S_ISREG(inode
->i_mode
))
1553 openflags
= nd
->intent
.open
.flags
;
1554 /* We cannot do exclusive creation on a positive dentry */
1555 if ((openflags
& (O_CREAT
|O_EXCL
)) == (O_CREAT
|O_EXCL
))
1557 /* We can't create new files here */
1558 openflags
&= ~(O_CREAT
|O_EXCL
);
1560 ctx
= create_nfs_open_context(dentry
, openflags
);
1565 attr
.ia_valid
= ATTR_OPEN
;
1566 if (openflags
& O_TRUNC
) {
1567 attr
.ia_valid
|= ATTR_SIZE
;
1573 * Note: we're not holding inode->i_mutex and so may be racing with
1574 * operations that change the directory. We therefore save the
1575 * change attribute *before* we do the RPC call.
1577 inode
= NFS_PROTO(dir
)->open_context(dir
, ctx
, openflags
, &attr
);
1578 if (IS_ERR(inode
)) {
1579 ret
= PTR_ERR(inode
);
1592 if (inode
!= dentry
->d_inode
)
1595 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1596 ret
= nfs_intent_set_file(nd
, ctx
);
1606 put_nfs_open_context(ctx
);
1612 return nfs_lookup_revalidate(dentry
, nd
);
1615 static int nfs_open_create(struct inode
*dir
, struct dentry
*dentry
,
1616 umode_t mode
, struct nameidata
*nd
)
1618 struct nfs_open_context
*ctx
= NULL
;
1621 int open_flags
= O_CREAT
|O_EXCL
;
1623 dfprintk(VFS
, "NFS: create(%s/%ld), %s\n",
1624 dir
->i_sb
->s_id
, dir
->i_ino
, dentry
->d_name
.name
);
1626 attr
.ia_mode
= mode
;
1627 attr
.ia_valid
= ATTR_MODE
;
1630 open_flags
= nd
->intent
.open
.flags
;
1632 ctx
= create_nfs_open_context(dentry
, open_flags
);
1633 error
= PTR_ERR(ctx
);
1637 error
= NFS_PROTO(dir
)->create(dir
, dentry
, &attr
, open_flags
, ctx
);
1641 error
= nfs_intent_set_file(nd
, ctx
);
1645 put_nfs_open_context(ctx
);
1649 put_nfs_open_context(ctx
);
1656 #endif /* CONFIG_NFSV4 */
1659 * Code common to create, mkdir, and mknod.
1661 int nfs_instantiate(struct dentry
*dentry
, struct nfs_fh
*fhandle
,
1662 struct nfs_fattr
*fattr
)
1664 struct dentry
*parent
= dget_parent(dentry
);
1665 struct inode
*dir
= parent
->d_inode
;
1666 struct inode
*inode
;
1667 int error
= -EACCES
;
1671 /* We may have been initialized further down */
1672 if (dentry
->d_inode
)
1674 if (fhandle
->size
== 0) {
1675 error
= NFS_PROTO(dir
)->lookup(dir
, &dentry
->d_name
, fhandle
, fattr
);
1679 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1680 if (!(fattr
->valid
& NFS_ATTR_FATTR
)) {
1681 struct nfs_server
*server
= NFS_SB(dentry
->d_sb
);
1682 error
= server
->nfs_client
->rpc_ops
->getattr(server
, fhandle
, fattr
);
1686 inode
= nfs_fhget(dentry
->d_sb
, fhandle
, fattr
);
1687 error
= PTR_ERR(inode
);
1690 d_add(dentry
, inode
);
1695 nfs_mark_for_revalidate(dir
);
1701 * Following a failed create operation, we drop the dentry rather
1702 * than retain a negative dentry. This avoids a problem in the event
1703 * that the operation succeeded on the server, but an error in the
1704 * reply path made it appear to have failed.
1706 static int nfs_create(struct inode
*dir
, struct dentry
*dentry
,
1707 umode_t mode
, struct nameidata
*nd
)
1711 int open_flags
= O_CREAT
|O_EXCL
;
1713 dfprintk(VFS
, "NFS: create(%s/%ld), %s\n",
1714 dir
->i_sb
->s_id
, dir
->i_ino
, dentry
->d_name
.name
);
1716 attr
.ia_mode
= mode
;
1717 attr
.ia_valid
= ATTR_MODE
;
1720 open_flags
= nd
->intent
.open
.flags
;
1722 error
= NFS_PROTO(dir
)->create(dir
, dentry
, &attr
, open_flags
, NULL
);
1732 * See comments for nfs_proc_create regarding failed operations.
1735 nfs_mknod(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
, dev_t rdev
)
1740 dfprintk(VFS
, "NFS: mknod(%s/%ld), %s\n",
1741 dir
->i_sb
->s_id
, dir
->i_ino
, dentry
->d_name
.name
);
1743 if (!new_valid_dev(rdev
))
1746 attr
.ia_mode
= mode
;
1747 attr
.ia_valid
= ATTR_MODE
;
1749 status
= NFS_PROTO(dir
)->mknod(dir
, dentry
, &attr
, rdev
);
1759 * See comments for nfs_proc_create regarding failed operations.
1761 static int nfs_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
1766 dfprintk(VFS
, "NFS: mkdir(%s/%ld), %s\n",
1767 dir
->i_sb
->s_id
, dir
->i_ino
, dentry
->d_name
.name
);
1769 attr
.ia_valid
= ATTR_MODE
;
1770 attr
.ia_mode
= mode
| S_IFDIR
;
1772 error
= NFS_PROTO(dir
)->mkdir(dir
, dentry
, &attr
);
1781 static void nfs_dentry_handle_enoent(struct dentry
*dentry
)
1783 if (dentry
->d_inode
!= NULL
&& !d_unhashed(dentry
))
1787 static int nfs_rmdir(struct inode
*dir
, struct dentry
*dentry
)
1791 dfprintk(VFS
, "NFS: rmdir(%s/%ld), %s\n",
1792 dir
->i_sb
->s_id
, dir
->i_ino
, dentry
->d_name
.name
);
1794 error
= NFS_PROTO(dir
)->rmdir(dir
, &dentry
->d_name
);
1795 /* Ensure the VFS deletes this inode */
1796 if (error
== 0 && dentry
->d_inode
!= NULL
)
1797 clear_nlink(dentry
->d_inode
);
1798 else if (error
== -ENOENT
)
1799 nfs_dentry_handle_enoent(dentry
);
1805 * Remove a file after making sure there are no pending writes,
1806 * and after checking that the file has only one user.
1808 * We invalidate the attribute cache and free the inode prior to the operation
1809 * to avoid possible races if the server reuses the inode.
1811 static int nfs_safe_remove(struct dentry
*dentry
)
1813 struct inode
*dir
= dentry
->d_parent
->d_inode
;
1814 struct inode
*inode
= dentry
->d_inode
;
1817 dfprintk(VFS
, "NFS: safe_remove(%s/%s)\n",
1818 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
);
1820 /* If the dentry was sillyrenamed, we simply call d_delete() */
1821 if (dentry
->d_flags
& DCACHE_NFSFS_RENAMED
) {
1826 if (inode
!= NULL
) {
1827 nfs_inode_return_delegation(inode
);
1828 error
= NFS_PROTO(dir
)->remove(dir
, &dentry
->d_name
);
1829 /* The VFS may want to delete this inode */
1831 nfs_drop_nlink(inode
);
1832 nfs_mark_for_revalidate(inode
);
1834 error
= NFS_PROTO(dir
)->remove(dir
, &dentry
->d_name
);
1835 if (error
== -ENOENT
)
1836 nfs_dentry_handle_enoent(dentry
);
1841 /* We do silly rename. In case sillyrename() returns -EBUSY, the inode
1842 * belongs to an active ".nfs..." file and we return -EBUSY.
1844 * If sillyrename() returns 0, we do nothing, otherwise we unlink.
1846 static int nfs_unlink(struct inode
*dir
, struct dentry
*dentry
)
1849 int need_rehash
= 0;
1851 dfprintk(VFS
, "NFS: unlink(%s/%ld, %s)\n", dir
->i_sb
->s_id
,
1852 dir
->i_ino
, dentry
->d_name
.name
);
1854 spin_lock(&dentry
->d_lock
);
1855 if (dentry
->d_count
> 1) {
1856 spin_unlock(&dentry
->d_lock
);
1857 /* Start asynchronous writeout of the inode */
1858 write_inode_now(dentry
->d_inode
, 0);
1859 error
= nfs_sillyrename(dir
, dentry
);
1862 if (!d_unhashed(dentry
)) {
1866 spin_unlock(&dentry
->d_lock
);
1867 error
= nfs_safe_remove(dentry
);
1868 if (!error
|| error
== -ENOENT
) {
1869 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1870 } else if (need_rehash
)
1876 * To create a symbolic link, most file systems instantiate a new inode,
1877 * add a page to it containing the path, then write it out to the disk
1878 * using prepare_write/commit_write.
1880 * Unfortunately the NFS client can't create the in-core inode first
1881 * because it needs a file handle to create an in-core inode (see
1882 * fs/nfs/inode.c:nfs_fhget). We only have a file handle *after* the
1883 * symlink request has completed on the server.
1885 * So instead we allocate a raw page, copy the symname into it, then do
1886 * the SYMLINK request with the page as the buffer. If it succeeds, we
1887 * now have a new file handle and can instantiate an in-core NFS inode
1888 * and move the raw page into its mapping.
1890 static int nfs_symlink(struct inode
*dir
, struct dentry
*dentry
, const char *symname
)
1892 struct pagevec lru_pvec
;
1896 unsigned int pathlen
= strlen(symname
);
1899 dfprintk(VFS
, "NFS: symlink(%s/%ld, %s, %s)\n", dir
->i_sb
->s_id
,
1900 dir
->i_ino
, dentry
->d_name
.name
, symname
);
1902 if (pathlen
> PAGE_SIZE
)
1903 return -ENAMETOOLONG
;
1905 attr
.ia_mode
= S_IFLNK
| S_IRWXUGO
;
1906 attr
.ia_valid
= ATTR_MODE
;
1908 page
= alloc_page(GFP_HIGHUSER
);
1912 kaddr
= kmap_atomic(page
);
1913 memcpy(kaddr
, symname
, pathlen
);
1914 if (pathlen
< PAGE_SIZE
)
1915 memset(kaddr
+ pathlen
, 0, PAGE_SIZE
- pathlen
);
1916 kunmap_atomic(kaddr
);
1918 error
= NFS_PROTO(dir
)->symlink(dir
, dentry
, page
, pathlen
, &attr
);
1920 dfprintk(VFS
, "NFS: symlink(%s/%ld, %s, %s) error %d\n",
1921 dir
->i_sb
->s_id
, dir
->i_ino
,
1922 dentry
->d_name
.name
, symname
, error
);
1929 * No big deal if we can't add this page to the page cache here.
1930 * READLINK will get the missing page from the server if needed.
1932 pagevec_init(&lru_pvec
, 0);
1933 if (!add_to_page_cache(page
, dentry
->d_inode
->i_mapping
, 0,
1935 pagevec_add(&lru_pvec
, page
);
1936 pagevec_lru_add_file(&lru_pvec
);
1937 SetPageUptodate(page
);
1946 nfs_link(struct dentry
*old_dentry
, struct inode
*dir
, struct dentry
*dentry
)
1948 struct inode
*inode
= old_dentry
->d_inode
;
1951 dfprintk(VFS
, "NFS: link(%s/%s -> %s/%s)\n",
1952 old_dentry
->d_parent
->d_name
.name
, old_dentry
->d_name
.name
,
1953 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
);
1955 nfs_inode_return_delegation(inode
);
1958 error
= NFS_PROTO(dir
)->link(inode
, dir
, &dentry
->d_name
);
1961 d_add(dentry
, inode
);
1968 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a
1969 * different file handle for the same inode after a rename (e.g. when
1970 * moving to a different directory). A fail-safe method to do so would
1971 * be to look up old_dir/old_name, create a link to new_dir/new_name and
1972 * rename the old file using the sillyrename stuff. This way, the original
1973 * file in old_dir will go away when the last process iput()s the inode.
1977 * It actually works quite well. One needs to have the possibility for
1978 * at least one ".nfs..." file in each directory the file ever gets
1979 * moved or linked to which happens automagically with the new
1980 * implementation that only depends on the dcache stuff instead of
1981 * using the inode layer
1983 * Unfortunately, things are a little more complicated than indicated
1984 * above. For a cross-directory move, we want to make sure we can get
1985 * rid of the old inode after the operation. This means there must be
1986 * no pending writes (if it's a file), and the use count must be 1.
1987 * If these conditions are met, we can drop the dentries before doing
1990 static int nfs_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
1991 struct inode
*new_dir
, struct dentry
*new_dentry
)
1993 struct inode
*old_inode
= old_dentry
->d_inode
;
1994 struct inode
*new_inode
= new_dentry
->d_inode
;
1995 struct dentry
*dentry
= NULL
, *rehash
= NULL
;
1998 dfprintk(VFS
, "NFS: rename(%s/%s -> %s/%s, ct=%d)\n",
1999 old_dentry
->d_parent
->d_name
.name
, old_dentry
->d_name
.name
,
2000 new_dentry
->d_parent
->d_name
.name
, new_dentry
->d_name
.name
,
2001 new_dentry
->d_count
);
2004 * For non-directories, check whether the target is busy and if so,
2005 * make a copy of the dentry and then do a silly-rename. If the
2006 * silly-rename succeeds, the copied dentry is hashed and becomes
2009 if (new_inode
&& !S_ISDIR(new_inode
->i_mode
)) {
2011 * To prevent any new references to the target during the
2012 * rename, we unhash the dentry in advance.
2014 if (!d_unhashed(new_dentry
)) {
2016 rehash
= new_dentry
;
2019 if (new_dentry
->d_count
> 2) {
2022 /* copy the target dentry's name */
2023 dentry
= d_alloc(new_dentry
->d_parent
,
2024 &new_dentry
->d_name
);
2028 /* silly-rename the existing target ... */
2029 err
= nfs_sillyrename(new_dir
, new_dentry
);
2033 new_dentry
= dentry
;
2039 nfs_inode_return_delegation(old_inode
);
2040 if (new_inode
!= NULL
)
2041 nfs_inode_return_delegation(new_inode
);
2043 error
= NFS_PROTO(old_dir
)->rename(old_dir
, &old_dentry
->d_name
,
2044 new_dir
, &new_dentry
->d_name
);
2045 nfs_mark_for_revalidate(old_inode
);
2050 if (new_inode
!= NULL
)
2051 nfs_drop_nlink(new_inode
);
2052 d_move(old_dentry
, new_dentry
);
2053 nfs_set_verifier(new_dentry
,
2054 nfs_save_change_attribute(new_dir
));
2055 } else if (error
== -ENOENT
)
2056 nfs_dentry_handle_enoent(old_dentry
);
2058 /* new dentry created? */
2064 static DEFINE_SPINLOCK(nfs_access_lru_lock
);
2065 static LIST_HEAD(nfs_access_lru_list
);
2066 static atomic_long_t nfs_access_nr_entries
;
2068 static void nfs_access_free_entry(struct nfs_access_entry
*entry
)
2070 put_rpccred(entry
->cred
);
2072 smp_mb__before_atomic_dec();
2073 atomic_long_dec(&nfs_access_nr_entries
);
2074 smp_mb__after_atomic_dec();
2077 static void nfs_access_free_list(struct list_head
*head
)
2079 struct nfs_access_entry
*cache
;
2081 while (!list_empty(head
)) {
2082 cache
= list_entry(head
->next
, struct nfs_access_entry
, lru
);
2083 list_del(&cache
->lru
);
2084 nfs_access_free_entry(cache
);
2088 int nfs_access_cache_shrinker(struct shrinker
*shrink
,
2089 struct shrink_control
*sc
)
2092 struct nfs_inode
*nfsi
, *next
;
2093 struct nfs_access_entry
*cache
;
2094 int nr_to_scan
= sc
->nr_to_scan
;
2095 gfp_t gfp_mask
= sc
->gfp_mask
;
2097 if ((gfp_mask
& GFP_KERNEL
) != GFP_KERNEL
)
2098 return (nr_to_scan
== 0) ? 0 : -1;
2100 spin_lock(&nfs_access_lru_lock
);
2101 list_for_each_entry_safe(nfsi
, next
, &nfs_access_lru_list
, access_cache_inode_lru
) {
2102 struct inode
*inode
;
2104 if (nr_to_scan
-- == 0)
2106 inode
= &nfsi
->vfs_inode
;
2107 spin_lock(&inode
->i_lock
);
2108 if (list_empty(&nfsi
->access_cache_entry_lru
))
2109 goto remove_lru_entry
;
2110 cache
= list_entry(nfsi
->access_cache_entry_lru
.next
,
2111 struct nfs_access_entry
, lru
);
2112 list_move(&cache
->lru
, &head
);
2113 rb_erase(&cache
->rb_node
, &nfsi
->access_cache
);
2114 if (!list_empty(&nfsi
->access_cache_entry_lru
))
2115 list_move_tail(&nfsi
->access_cache_inode_lru
,
2116 &nfs_access_lru_list
);
2119 list_del_init(&nfsi
->access_cache_inode_lru
);
2120 smp_mb__before_clear_bit();
2121 clear_bit(NFS_INO_ACL_LRU_SET
, &nfsi
->flags
);
2122 smp_mb__after_clear_bit();
2124 spin_unlock(&inode
->i_lock
);
2126 spin_unlock(&nfs_access_lru_lock
);
2127 nfs_access_free_list(&head
);
2128 return (atomic_long_read(&nfs_access_nr_entries
) / 100) * sysctl_vfs_cache_pressure
;
2131 static void __nfs_access_zap_cache(struct nfs_inode
*nfsi
, struct list_head
*head
)
2133 struct rb_root
*root_node
= &nfsi
->access_cache
;
2135 struct nfs_access_entry
*entry
;
2137 /* Unhook entries from the cache */
2138 while ((n
= rb_first(root_node
)) != NULL
) {
2139 entry
= rb_entry(n
, struct nfs_access_entry
, rb_node
);
2140 rb_erase(n
, root_node
);
2141 list_move(&entry
->lru
, head
);
2143 nfsi
->cache_validity
&= ~NFS_INO_INVALID_ACCESS
;
2146 void nfs_access_zap_cache(struct inode
*inode
)
2150 if (test_bit(NFS_INO_ACL_LRU_SET
, &NFS_I(inode
)->flags
) == 0)
2152 /* Remove from global LRU init */
2153 spin_lock(&nfs_access_lru_lock
);
2154 if (test_and_clear_bit(NFS_INO_ACL_LRU_SET
, &NFS_I(inode
)->flags
))
2155 list_del_init(&NFS_I(inode
)->access_cache_inode_lru
);
2157 spin_lock(&inode
->i_lock
);
2158 __nfs_access_zap_cache(NFS_I(inode
), &head
);
2159 spin_unlock(&inode
->i_lock
);
2160 spin_unlock(&nfs_access_lru_lock
);
2161 nfs_access_free_list(&head
);
2164 static struct nfs_access_entry
*nfs_access_search_rbtree(struct inode
*inode
, struct rpc_cred
*cred
)
2166 struct rb_node
*n
= NFS_I(inode
)->access_cache
.rb_node
;
2167 struct nfs_access_entry
*entry
;
2170 entry
= rb_entry(n
, struct nfs_access_entry
, rb_node
);
2172 if (cred
< entry
->cred
)
2174 else if (cred
> entry
->cred
)
2182 static int nfs_access_get_cached(struct inode
*inode
, struct rpc_cred
*cred
, struct nfs_access_entry
*res
)
2184 struct nfs_inode
*nfsi
= NFS_I(inode
);
2185 struct nfs_access_entry
*cache
;
2188 spin_lock(&inode
->i_lock
);
2189 if (nfsi
->cache_validity
& NFS_INO_INVALID_ACCESS
)
2191 cache
= nfs_access_search_rbtree(inode
, cred
);
2194 if (!nfs_have_delegated_attributes(inode
) &&
2195 !time_in_range_open(jiffies
, cache
->jiffies
, cache
->jiffies
+ nfsi
->attrtimeo
))
2197 res
->jiffies
= cache
->jiffies
;
2198 res
->cred
= cache
->cred
;
2199 res
->mask
= cache
->mask
;
2200 list_move_tail(&cache
->lru
, &nfsi
->access_cache_entry_lru
);
2203 spin_unlock(&inode
->i_lock
);
2206 rb_erase(&cache
->rb_node
, &nfsi
->access_cache
);
2207 list_del(&cache
->lru
);
2208 spin_unlock(&inode
->i_lock
);
2209 nfs_access_free_entry(cache
);
2212 spin_unlock(&inode
->i_lock
);
2213 nfs_access_zap_cache(inode
);
2217 static void nfs_access_add_rbtree(struct inode
*inode
, struct nfs_access_entry
*set
)
2219 struct nfs_inode
*nfsi
= NFS_I(inode
);
2220 struct rb_root
*root_node
= &nfsi
->access_cache
;
2221 struct rb_node
**p
= &root_node
->rb_node
;
2222 struct rb_node
*parent
= NULL
;
2223 struct nfs_access_entry
*entry
;
2225 spin_lock(&inode
->i_lock
);
2226 while (*p
!= NULL
) {
2228 entry
= rb_entry(parent
, struct nfs_access_entry
, rb_node
);
2230 if (set
->cred
< entry
->cred
)
2231 p
= &parent
->rb_left
;
2232 else if (set
->cred
> entry
->cred
)
2233 p
= &parent
->rb_right
;
2237 rb_link_node(&set
->rb_node
, parent
, p
);
2238 rb_insert_color(&set
->rb_node
, root_node
);
2239 list_add_tail(&set
->lru
, &nfsi
->access_cache_entry_lru
);
2240 spin_unlock(&inode
->i_lock
);
2243 rb_replace_node(parent
, &set
->rb_node
, root_node
);
2244 list_add_tail(&set
->lru
, &nfsi
->access_cache_entry_lru
);
2245 list_del(&entry
->lru
);
2246 spin_unlock(&inode
->i_lock
);
2247 nfs_access_free_entry(entry
);
2250 static void nfs_access_add_cache(struct inode
*inode
, struct nfs_access_entry
*set
)
2252 struct nfs_access_entry
*cache
= kmalloc(sizeof(*cache
), GFP_KERNEL
);
2255 RB_CLEAR_NODE(&cache
->rb_node
);
2256 cache
->jiffies
= set
->jiffies
;
2257 cache
->cred
= get_rpccred(set
->cred
);
2258 cache
->mask
= set
->mask
;
2260 nfs_access_add_rbtree(inode
, cache
);
2262 /* Update accounting */
2263 smp_mb__before_atomic_inc();
2264 atomic_long_inc(&nfs_access_nr_entries
);
2265 smp_mb__after_atomic_inc();
2267 /* Add inode to global LRU list */
2268 if (!test_bit(NFS_INO_ACL_LRU_SET
, &NFS_I(inode
)->flags
)) {
2269 spin_lock(&nfs_access_lru_lock
);
2270 if (!test_and_set_bit(NFS_INO_ACL_LRU_SET
, &NFS_I(inode
)->flags
))
2271 list_add_tail(&NFS_I(inode
)->access_cache_inode_lru
,
2272 &nfs_access_lru_list
);
2273 spin_unlock(&nfs_access_lru_lock
);
2277 static int nfs_do_access(struct inode
*inode
, struct rpc_cred
*cred
, int mask
)
2279 struct nfs_access_entry cache
;
2282 status
= nfs_access_get_cached(inode
, cred
, &cache
);
2286 /* Be clever: ask server to check for all possible rights */
2287 cache
.mask
= MAY_EXEC
| MAY_WRITE
| MAY_READ
;
2289 cache
.jiffies
= jiffies
;
2290 status
= NFS_PROTO(inode
)->access(inode
, &cache
);
2292 if (status
== -ESTALE
) {
2293 nfs_zap_caches(inode
);
2294 if (!S_ISDIR(inode
->i_mode
))
2295 set_bit(NFS_INO_STALE
, &NFS_I(inode
)->flags
);
2299 nfs_access_add_cache(inode
, &cache
);
2301 if ((mask
& ~cache
.mask
& (MAY_READ
| MAY_WRITE
| MAY_EXEC
)) == 0)
2306 static int nfs_open_permission_mask(int openflags
)
2310 if ((openflags
& O_ACCMODE
) != O_WRONLY
)
2312 if ((openflags
& O_ACCMODE
) != O_RDONLY
)
2314 if (openflags
& __FMODE_EXEC
)
2319 int nfs_may_open(struct inode
*inode
, struct rpc_cred
*cred
, int openflags
)
2321 return nfs_do_access(inode
, cred
, nfs_open_permission_mask(openflags
));
2324 int nfs_permission(struct inode
*inode
, int mask
)
2326 struct rpc_cred
*cred
;
2329 if (mask
& MAY_NOT_BLOCK
)
2332 nfs_inc_stats(inode
, NFSIOS_VFSACCESS
);
2334 if ((mask
& (MAY_READ
| MAY_WRITE
| MAY_EXEC
)) == 0)
2336 /* Is this sys_access() ? */
2337 if (mask
& (MAY_ACCESS
| MAY_CHDIR
))
2340 switch (inode
->i_mode
& S_IFMT
) {
2344 /* NFSv4 has atomic_open... */
2345 if (nfs_server_capable(inode
, NFS_CAP_ATOMIC_OPEN
)
2346 && (mask
& MAY_OPEN
)
2347 && !(mask
& MAY_EXEC
))
2352 * Optimize away all write operations, since the server
2353 * will check permissions when we perform the op.
2355 if ((mask
& MAY_WRITE
) && !(mask
& MAY_READ
))
2360 if (!NFS_PROTO(inode
)->access
)
2363 cred
= rpc_lookup_cred();
2364 if (!IS_ERR(cred
)) {
2365 res
= nfs_do_access(inode
, cred
, mask
);
2368 res
= PTR_ERR(cred
);
2370 if (!res
&& (mask
& MAY_EXEC
) && !execute_ok(inode
))
2373 dfprintk(VFS
, "NFS: permission(%s/%ld), mask=0x%x, res=%d\n",
2374 inode
->i_sb
->s_id
, inode
->i_ino
, mask
, res
);
2377 res
= nfs_revalidate_inode(NFS_SERVER(inode
), inode
);
2379 res
= generic_permission(inode
, mask
);
2385 * version-control: t
2386 * kept-new-versions: 5