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/module.h>
21 #include <linux/time.h>
22 #include <linux/errno.h>
23 #include <linux/stat.h>
24 #include <linux/fcntl.h>
25 #include <linux/string.h>
26 #include <linux/kernel.h>
27 #include <linux/slab.h>
29 #include <linux/sunrpc/clnt.h>
30 #include <linux/nfs_fs.h>
31 #include <linux/nfs_mount.h>
32 #include <linux/pagemap.h>
33 #include <linux/pagevec.h>
34 #include <linux/namei.h>
35 #include <linux/mount.h>
36 #include <linux/sched.h>
37 #include <linux/kmemleak.h>
38 #include <linux/xattr.h>
40 #include "delegation.h"
45 /* #define NFS_DEBUG_VERBOSE 1 */
47 static int nfs_opendir(struct inode
*, struct file
*);
48 static int nfs_closedir(struct inode
*, struct file
*);
49 static int nfs_readdir(struct file
*, void *, filldir_t
);
50 static int nfs_fsync_dir(struct file
*, loff_t
, loff_t
, int);
51 static loff_t
nfs_llseek_dir(struct file
*, loff_t
, int);
52 static void nfs_readdir_clear_array(struct page
*);
54 const struct file_operations nfs_dir_operations
= {
55 .llseek
= nfs_llseek_dir
,
56 .read
= generic_read_dir
,
57 .readdir
= nfs_readdir
,
59 .release
= nfs_closedir
,
60 .fsync
= nfs_fsync_dir
,
63 const struct address_space_operations nfs_dir_aops
= {
64 .freepage
= nfs_readdir_clear_array
,
67 static struct nfs_open_dir_context
*alloc_nfs_open_dir_context(struct inode
*dir
, struct rpc_cred
*cred
)
69 struct nfs_open_dir_context
*ctx
;
70 ctx
= kmalloc(sizeof(*ctx
), GFP_KERNEL
);
73 ctx
->attr_gencount
= NFS_I(dir
)->attr_gencount
;
76 ctx
->cred
= get_rpccred(cred
);
79 return ERR_PTR(-ENOMEM
);
82 static void put_nfs_open_dir_context(struct nfs_open_dir_context
*ctx
)
84 put_rpccred(ctx
->cred
);
92 nfs_opendir(struct inode
*inode
, struct file
*filp
)
95 struct nfs_open_dir_context
*ctx
;
96 struct rpc_cred
*cred
;
98 dfprintk(FILE, "NFS: open dir(%s/%s)\n",
99 filp
->f_path
.dentry
->d_parent
->d_name
.name
,
100 filp
->f_path
.dentry
->d_name
.name
);
102 nfs_inc_stats(inode
, NFSIOS_VFSOPEN
);
104 cred
= rpc_lookup_cred();
106 return PTR_ERR(cred
);
107 ctx
= alloc_nfs_open_dir_context(inode
, cred
);
112 filp
->private_data
= ctx
;
113 if (filp
->f_path
.dentry
== filp
->f_path
.mnt
->mnt_root
) {
114 /* This is a mountpoint, so d_revalidate will never
115 * have been called, so we need to refresh the
116 * inode (for close-open consistency) ourselves.
118 __nfs_revalidate_inode(NFS_SERVER(inode
), inode
);
126 nfs_closedir(struct inode
*inode
, struct file
*filp
)
128 put_nfs_open_dir_context(filp
->private_data
);
132 struct nfs_cache_array_entry
{
136 unsigned char d_type
;
139 struct nfs_cache_array
{
143 struct nfs_cache_array_entry array
[0];
146 typedef int (*decode_dirent_t
)(struct xdr_stream
*, struct nfs_entry
*, int);
150 unsigned long page_index
;
153 loff_t current_index
;
154 decode_dirent_t decode
;
156 unsigned long timestamp
;
157 unsigned long gencount
;
158 unsigned int cache_entry_index
;
161 } nfs_readdir_descriptor_t
;
164 * The caller is responsible for calling nfs_readdir_release_array(page)
167 struct nfs_cache_array
*nfs_readdir_get_array(struct page
*page
)
171 return ERR_PTR(-EIO
);
174 return ERR_PTR(-ENOMEM
);
179 void nfs_readdir_release_array(struct page
*page
)
185 * we are freeing strings created by nfs_add_to_readdir_array()
188 void nfs_readdir_clear_array(struct page
*page
)
190 struct nfs_cache_array
*array
;
193 array
= kmap_atomic(page
);
194 for (i
= 0; i
< array
->size
; i
++)
195 kfree(array
->array
[i
].string
.name
);
196 kunmap_atomic(array
);
200 * the caller is responsible for freeing qstr.name
201 * when called by nfs_readdir_add_to_array, the strings will be freed in
202 * nfs_clear_readdir_array()
205 int nfs_readdir_make_qstr(struct qstr
*string
, const char *name
, unsigned int len
)
208 string
->name
= kmemdup(name
, len
, GFP_KERNEL
);
209 if (string
->name
== NULL
)
212 * Avoid a kmemleak false positive. The pointer to the name is stored
213 * in a page cache page which kmemleak does not scan.
215 kmemleak_not_leak(string
->name
);
216 string
->hash
= full_name_hash(name
, len
);
221 int nfs_readdir_add_to_array(struct nfs_entry
*entry
, struct page
*page
)
223 struct nfs_cache_array
*array
= nfs_readdir_get_array(page
);
224 struct nfs_cache_array_entry
*cache_entry
;
228 return PTR_ERR(array
);
230 cache_entry
= &array
->array
[array
->size
];
232 /* Check that this entry lies within the page bounds */
234 if ((char *)&cache_entry
[1] - (char *)page_address(page
) > PAGE_SIZE
)
237 cache_entry
->cookie
= entry
->prev_cookie
;
238 cache_entry
->ino
= entry
->ino
;
239 cache_entry
->d_type
= entry
->d_type
;
240 ret
= nfs_readdir_make_qstr(&cache_entry
->string
, entry
->name
, entry
->len
);
243 array
->last_cookie
= entry
->cookie
;
246 array
->eof_index
= array
->size
;
248 nfs_readdir_release_array(page
);
253 int nfs_readdir_search_for_pos(struct nfs_cache_array
*array
, nfs_readdir_descriptor_t
*desc
)
255 loff_t diff
= desc
->file
->f_pos
- desc
->current_index
;
260 if (diff
>= array
->size
) {
261 if (array
->eof_index
>= 0)
266 index
= (unsigned int)diff
;
267 *desc
->dir_cookie
= array
->array
[index
].cookie
;
268 desc
->cache_entry_index
= index
;
276 int nfs_readdir_search_for_cookie(struct nfs_cache_array
*array
, nfs_readdir_descriptor_t
*desc
)
280 int status
= -EAGAIN
;
282 for (i
= 0; i
< array
->size
; i
++) {
283 if (array
->array
[i
].cookie
== *desc
->dir_cookie
) {
284 struct nfs_inode
*nfsi
= NFS_I(file_inode(desc
->file
));
285 struct nfs_open_dir_context
*ctx
= desc
->file
->private_data
;
287 new_pos
= desc
->current_index
+ i
;
288 if (ctx
->attr_gencount
!= nfsi
->attr_gencount
289 || (nfsi
->cache_validity
& (NFS_INO_INVALID_ATTR
|NFS_INO_INVALID_DATA
))) {
291 ctx
->attr_gencount
= nfsi
->attr_gencount
;
292 } else if (new_pos
< desc
->file
->f_pos
) {
294 && ctx
->dup_cookie
== *desc
->dir_cookie
) {
295 if (printk_ratelimit()) {
296 pr_notice("NFS: directory %s/%s contains a readdir loop."
297 "Please contact your server vendor. "
298 "The file: %s has duplicate cookie %llu\n",
299 desc
->file
->f_dentry
->d_parent
->d_name
.name
,
300 desc
->file
->f_dentry
->d_name
.name
,
301 array
->array
[i
].string
.name
,
307 ctx
->dup_cookie
= *desc
->dir_cookie
;
310 desc
->file
->f_pos
= new_pos
;
311 desc
->cache_entry_index
= i
;
315 if (array
->eof_index
>= 0) {
316 status
= -EBADCOOKIE
;
317 if (*desc
->dir_cookie
== array
->last_cookie
)
325 int nfs_readdir_search_array(nfs_readdir_descriptor_t
*desc
)
327 struct nfs_cache_array
*array
;
330 array
= nfs_readdir_get_array(desc
->page
);
332 status
= PTR_ERR(array
);
336 if (*desc
->dir_cookie
== 0)
337 status
= nfs_readdir_search_for_pos(array
, desc
);
339 status
= nfs_readdir_search_for_cookie(array
, desc
);
341 if (status
== -EAGAIN
) {
342 desc
->last_cookie
= array
->last_cookie
;
343 desc
->current_index
+= array
->size
;
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 nfs_open_dir_context
*ctx
= file
->private_data
;
357 struct rpc_cred
*cred
= ctx
->cred
;
358 unsigned long timestamp
, gencount
;
363 gencount
= nfs_inc_attr_generation_counter();
364 error
= NFS_PROTO(inode
)->readdir(file
->f_path
.dentry
, cred
, entry
->cookie
, pages
,
365 NFS_SERVER(inode
)->dtsize
, desc
->plus
);
367 /* We requested READDIRPLUS, but the server doesn't grok it */
368 if (error
== -ENOTSUPP
&& desc
->plus
) {
369 NFS_SERVER(inode
)->caps
&= ~NFS_CAP_READDIRPLUS
;
370 clear_bit(NFS_INO_ADVISE_RDPLUS
, &NFS_I(inode
)->flags
);
376 desc
->timestamp
= timestamp
;
377 desc
->gencount
= gencount
;
382 static int xdr_decode(nfs_readdir_descriptor_t
*desc
,
383 struct nfs_entry
*entry
, struct xdr_stream
*xdr
)
387 error
= desc
->decode(xdr
, entry
, desc
->plus
);
390 entry
->fattr
->time_start
= desc
->timestamp
;
391 entry
->fattr
->gencount
= desc
->gencount
;
396 int nfs_same_file(struct dentry
*dentry
, struct nfs_entry
*entry
)
398 if (dentry
->d_inode
== NULL
)
400 if (nfs_compare_fh(entry
->fh
, NFS_FH(dentry
->d_inode
)) != 0)
408 bool nfs_use_readdirplus(struct inode
*dir
, struct file
*filp
)
410 if (!nfs_server_capable(dir
, NFS_CAP_READDIRPLUS
))
412 if (test_and_clear_bit(NFS_INO_ADVISE_RDPLUS
, &NFS_I(dir
)->flags
))
414 if (filp
->f_pos
== 0)
420 * This function is called by the lookup code to request the use of
421 * readdirplus to accelerate any future lookups in the same
425 void nfs_advise_use_readdirplus(struct inode
*dir
)
427 set_bit(NFS_INO_ADVISE_RDPLUS
, &NFS_I(dir
)->flags
);
431 void nfs_prime_dcache(struct dentry
*parent
, struct nfs_entry
*entry
)
433 struct qstr filename
= QSTR_INIT(entry
->name
, entry
->len
);
434 struct dentry
*dentry
;
435 struct dentry
*alias
;
436 struct inode
*dir
= parent
->d_inode
;
439 if (filename
.name
[0] == '.') {
440 if (filename
.len
== 1)
442 if (filename
.len
== 2 && filename
.name
[1] == '.')
445 filename
.hash
= full_name_hash(filename
.name
, filename
.len
);
447 dentry
= d_lookup(parent
, &filename
);
448 if (dentry
!= NULL
) {
449 if (nfs_same_file(dentry
, entry
)) {
450 nfs_refresh_inode(dentry
->d_inode
, entry
->fattr
);
453 if (d_invalidate(dentry
) != 0)
459 dentry
= d_alloc(parent
, &filename
);
463 inode
= nfs_fhget(dentry
->d_sb
, entry
->fh
, entry
->fattr
, entry
->label
);
467 alias
= d_materialise_unique(dentry
, inode
);
471 nfs_set_verifier(alias
, nfs_save_change_attribute(dir
));
474 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
480 /* Perform conversion from xdr to cache array */
482 int nfs_readdir_page_filler(nfs_readdir_descriptor_t
*desc
, struct nfs_entry
*entry
,
483 struct page
**xdr_pages
, struct page
*page
, unsigned int buflen
)
485 struct xdr_stream stream
;
487 struct page
*scratch
;
488 struct nfs_cache_array
*array
;
489 unsigned int count
= 0;
492 scratch
= alloc_page(GFP_KERNEL
);
496 xdr_init_decode_pages(&stream
, &buf
, xdr_pages
, buflen
);
497 xdr_set_scratch_buffer(&stream
, page_address(scratch
), PAGE_SIZE
);
500 status
= xdr_decode(desc
, entry
, &stream
);
502 if (status
== -EAGAIN
)
510 nfs_prime_dcache(desc
->file
->f_path
.dentry
, entry
);
512 status
= nfs_readdir_add_to_array(entry
, page
);
515 } while (!entry
->eof
);
517 if (count
== 0 || (status
== -EBADCOOKIE
&& entry
->eof
!= 0)) {
518 array
= nfs_readdir_get_array(page
);
519 if (!IS_ERR(array
)) {
520 array
->eof_index
= array
->size
;
522 nfs_readdir_release_array(page
);
524 status
= PTR_ERR(array
);
532 void nfs_readdir_free_pagearray(struct page
**pages
, unsigned int npages
)
535 for (i
= 0; i
< npages
; i
++)
540 void nfs_readdir_free_large_page(void *ptr
, struct page
**pages
,
543 nfs_readdir_free_pagearray(pages
, npages
);
547 * nfs_readdir_large_page will allocate pages that must be freed with a call
548 * to nfs_readdir_free_large_page
551 int nfs_readdir_large_page(struct page
**pages
, unsigned int npages
)
555 for (i
= 0; i
< npages
; i
++) {
556 struct page
*page
= alloc_page(GFP_KERNEL
);
564 nfs_readdir_free_pagearray(pages
, i
);
569 int nfs_readdir_xdr_to_array(nfs_readdir_descriptor_t
*desc
, struct page
*page
, struct inode
*inode
)
571 struct page
*pages
[NFS_MAX_READDIR_PAGES
];
572 void *pages_ptr
= NULL
;
573 struct nfs_entry entry
;
574 struct file
*file
= desc
->file
;
575 struct nfs_cache_array
*array
;
576 int status
= -ENOMEM
;
577 unsigned int array_size
= ARRAY_SIZE(pages
);
579 entry
.prev_cookie
= 0;
580 entry
.cookie
= desc
->last_cookie
;
582 entry
.fh
= nfs_alloc_fhandle();
583 entry
.fattr
= nfs_alloc_fattr();
584 entry
.server
= NFS_SERVER(inode
);
585 if (entry
.fh
== NULL
|| entry
.fattr
== NULL
)
588 array
= nfs_readdir_get_array(page
);
590 status
= PTR_ERR(array
);
593 memset(array
, 0, sizeof(struct nfs_cache_array
));
594 array
->eof_index
= -1;
596 status
= nfs_readdir_large_page(pages
, array_size
);
598 goto out_release_array
;
601 status
= nfs_readdir_xdr_filler(pages
, desc
, &entry
, file
, inode
);
606 status
= nfs_readdir_page_filler(desc
, &entry
, pages
, page
, pglen
);
608 if (status
== -ENOSPC
)
612 } while (array
->eof_index
< 0);
614 nfs_readdir_free_large_page(pages_ptr
, pages
, array_size
);
616 nfs_readdir_release_array(page
);
618 nfs_free_fattr(entry
.fattr
);
619 nfs_free_fhandle(entry
.fh
);
624 * Now we cache directories properly, by converting xdr information
625 * to an array that can be used for lookups later. This results in
626 * fewer cache pages, since we can store more information on each page.
627 * We only need to convert from xdr once so future lookups are much simpler
630 int nfs_readdir_filler(nfs_readdir_descriptor_t
*desc
, struct page
* page
)
632 struct inode
*inode
= file_inode(desc
->file
);
635 ret
= nfs_readdir_xdr_to_array(desc
, page
, inode
);
638 SetPageUptodate(page
);
640 if (invalidate_inode_pages2_range(inode
->i_mapping
, page
->index
+ 1, -1) < 0) {
641 /* Should never happen */
642 nfs_zap_mapping(inode
, inode
->i_mapping
);
652 void cache_page_release(nfs_readdir_descriptor_t
*desc
)
654 if (!desc
->page
->mapping
)
655 nfs_readdir_clear_array(desc
->page
);
656 page_cache_release(desc
->page
);
661 struct page
*get_cache_page(nfs_readdir_descriptor_t
*desc
)
663 return read_cache_page(file_inode(desc
->file
)->i_mapping
,
664 desc
->page_index
, (filler_t
*)nfs_readdir_filler
, desc
);
668 * Returns 0 if desc->dir_cookie was found on page desc->page_index
671 int find_cache_page(nfs_readdir_descriptor_t
*desc
)
675 desc
->page
= get_cache_page(desc
);
676 if (IS_ERR(desc
->page
))
677 return PTR_ERR(desc
->page
);
679 res
= nfs_readdir_search_array(desc
);
681 cache_page_release(desc
);
685 /* Search for desc->dir_cookie from the beginning of the page cache */
687 int readdir_search_pagecache(nfs_readdir_descriptor_t
*desc
)
691 if (desc
->page_index
== 0) {
692 desc
->current_index
= 0;
693 desc
->last_cookie
= 0;
696 res
= find_cache_page(desc
);
697 } while (res
== -EAGAIN
);
702 * Once we've found the start of the dirent within a page: fill 'er up...
705 int nfs_do_filldir(nfs_readdir_descriptor_t
*desc
, void *dirent
,
708 struct file
*file
= desc
->file
;
711 struct nfs_cache_array
*array
= NULL
;
712 struct nfs_open_dir_context
*ctx
= file
->private_data
;
714 array
= nfs_readdir_get_array(desc
->page
);
716 res
= PTR_ERR(array
);
720 for (i
= desc
->cache_entry_index
; i
< array
->size
; i
++) {
721 struct nfs_cache_array_entry
*ent
;
723 ent
= &array
->array
[i
];
724 if (filldir(dirent
, ent
->string
.name
, ent
->string
.len
,
725 file
->f_pos
, nfs_compat_user_ino64(ent
->ino
),
731 if (i
< (array
->size
-1))
732 *desc
->dir_cookie
= array
->array
[i
+1].cookie
;
734 *desc
->dir_cookie
= array
->last_cookie
;
738 if (array
->eof_index
>= 0)
741 nfs_readdir_release_array(desc
->page
);
743 cache_page_release(desc
);
744 dfprintk(DIRCACHE
, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n",
745 (unsigned long long)*desc
->dir_cookie
, res
);
750 * If we cannot find a cookie in our cache, we suspect that this is
751 * because it points to a deleted file, so we ask the server to return
752 * whatever it thinks is the next entry. We then feed this to filldir.
753 * If all goes well, we should then be able to find our way round the
754 * cache on the next call to readdir_search_pagecache();
756 * NOTE: we cannot add the anonymous page to the pagecache because
757 * the data it contains might not be page aligned. Besides,
758 * we should already have a complete representation of the
759 * directory in the page cache by the time we get here.
762 int uncached_readdir(nfs_readdir_descriptor_t
*desc
, void *dirent
,
765 struct page
*page
= NULL
;
767 struct inode
*inode
= file_inode(desc
->file
);
768 struct nfs_open_dir_context
*ctx
= desc
->file
->private_data
;
770 dfprintk(DIRCACHE
, "NFS: uncached_readdir() searching for cookie %Lu\n",
771 (unsigned long long)*desc
->dir_cookie
);
773 page
= alloc_page(GFP_HIGHUSER
);
779 desc
->page_index
= 0;
780 desc
->last_cookie
= *desc
->dir_cookie
;
784 status
= nfs_readdir_xdr_to_array(desc
, page
, inode
);
788 status
= nfs_do_filldir(desc
, dirent
, filldir
);
791 dfprintk(DIRCACHE
, "NFS: %s: returns %d\n",
795 cache_page_release(desc
);
799 /* The file offset position represents the dirent entry number. A
800 last cookie cache takes care of the common case of reading the
803 static int nfs_readdir(struct file
*filp
, void *dirent
, filldir_t filldir
)
805 struct dentry
*dentry
= filp
->f_path
.dentry
;
806 struct inode
*inode
= dentry
->d_inode
;
807 nfs_readdir_descriptor_t my_desc
,
809 struct nfs_open_dir_context
*dir_ctx
= filp
->private_data
;
812 dfprintk(FILE, "NFS: readdir(%s/%s) starting at cookie %llu\n",
813 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
,
814 (long long)filp
->f_pos
);
815 nfs_inc_stats(inode
, NFSIOS_VFSGETDENTS
);
818 * filp->f_pos points to the dirent entry number.
819 * *desc->dir_cookie has the cookie for the next entry. We have
820 * to either find the entry with the appropriate number or
821 * revalidate the cookie.
823 memset(desc
, 0, sizeof(*desc
));
826 desc
->dir_cookie
= &dir_ctx
->dir_cookie
;
827 desc
->decode
= NFS_PROTO(inode
)->decode_dirent
;
828 desc
->plus
= nfs_use_readdirplus(inode
, filp
) ? 1 : 0;
830 nfs_block_sillyrename(dentry
);
831 res
= nfs_revalidate_mapping(inode
, filp
->f_mapping
);
836 res
= readdir_search_pagecache(desc
);
838 if (res
== -EBADCOOKIE
) {
840 /* This means either end of directory */
841 if (*desc
->dir_cookie
&& desc
->eof
== 0) {
842 /* Or that the server has 'lost' a cookie */
843 res
= uncached_readdir(desc
, dirent
, filldir
);
849 if (res
== -ETOOSMALL
&& desc
->plus
) {
850 clear_bit(NFS_INO_ADVISE_RDPLUS
, &NFS_I(inode
)->flags
);
851 nfs_zap_caches(inode
);
852 desc
->page_index
= 0;
860 res
= nfs_do_filldir(desc
, dirent
, filldir
);
863 } while (!desc
->eof
);
865 nfs_unblock_sillyrename(dentry
);
868 dfprintk(FILE, "NFS: readdir(%s/%s) returns %d\n",
869 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
,
874 static loff_t
nfs_llseek_dir(struct file
*filp
, loff_t offset
, int whence
)
876 struct dentry
*dentry
= filp
->f_path
.dentry
;
877 struct inode
*inode
= dentry
->d_inode
;
878 struct nfs_open_dir_context
*dir_ctx
= filp
->private_data
;
880 dfprintk(FILE, "NFS: llseek dir(%s/%s, %lld, %d)\n",
881 dentry
->d_parent
->d_name
.name
,
885 mutex_lock(&inode
->i_mutex
);
888 offset
+= filp
->f_pos
;
896 if (offset
!= filp
->f_pos
) {
897 filp
->f_pos
= offset
;
898 dir_ctx
->dir_cookie
= 0;
902 mutex_unlock(&inode
->i_mutex
);
907 * All directory operations under NFS are synchronous, so fsync()
908 * is a dummy operation.
910 static int nfs_fsync_dir(struct file
*filp
, loff_t start
, loff_t end
,
913 struct dentry
*dentry
= filp
->f_path
.dentry
;
914 struct inode
*inode
= dentry
->d_inode
;
916 dfprintk(FILE, "NFS: fsync dir(%s/%s) datasync %d\n",
917 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
,
920 mutex_lock(&inode
->i_mutex
);
921 nfs_inc_stats(dentry
->d_inode
, NFSIOS_VFSFSYNC
);
922 mutex_unlock(&inode
->i_mutex
);
927 * nfs_force_lookup_revalidate - Mark the directory as having changed
928 * @dir - pointer to directory inode
930 * This forces the revalidation code in nfs_lookup_revalidate() to do a
931 * full lookup on all child dentries of 'dir' whenever a change occurs
932 * on the server that might have invalidated our dcache.
934 * The caller should be holding dir->i_lock
936 void nfs_force_lookup_revalidate(struct inode
*dir
)
938 NFS_I(dir
)->cache_change_attribute
++;
940 EXPORT_SYMBOL_GPL(nfs_force_lookup_revalidate
);
943 * A check for whether or not the parent directory has changed.
944 * In the case it has, we assume that the dentries are untrustworthy
945 * and may need to be looked up again.
947 static int nfs_check_verifier(struct inode
*dir
, struct dentry
*dentry
)
951 if (NFS_SERVER(dir
)->flags
& NFS_MOUNT_LOOKUP_CACHE_NONE
)
953 if (!nfs_verify_change_attribute(dir
, dentry
->d_time
))
955 /* Revalidate nfsi->cache_change_attribute before we declare a match */
956 if (nfs_revalidate_inode(NFS_SERVER(dir
), dir
) < 0)
958 if (!nfs_verify_change_attribute(dir
, dentry
->d_time
))
964 * Use intent information to check whether or not we're going to do
965 * an O_EXCL create using this path component.
967 static int nfs_is_exclusive_create(struct inode
*dir
, unsigned int flags
)
969 if (NFS_PROTO(dir
)->version
== 2)
971 return flags
& LOOKUP_EXCL
;
975 * Inode and filehandle revalidation for lookups.
977 * We force revalidation in the cases where the VFS sets LOOKUP_REVAL,
978 * or if the intent information indicates that we're about to open this
979 * particular file and the "nocto" mount flag is not set.
983 int nfs_lookup_verify_inode(struct inode
*inode
, unsigned int flags
)
985 struct nfs_server
*server
= NFS_SERVER(inode
);
988 if (IS_AUTOMOUNT(inode
))
990 /* VFS wants an on-the-wire revalidation */
991 if (flags
& LOOKUP_REVAL
)
993 /* This is an open(2) */
994 if ((flags
& LOOKUP_OPEN
) && !(server
->flags
& NFS_MOUNT_NOCTO
) &&
995 (S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
)))
998 return (inode
->i_nlink
== 0) ? -ENOENT
: 0;
1000 ret
= __nfs_revalidate_inode(server
, inode
);
1007 * We judge how long we want to trust negative
1008 * dentries by looking at the parent inode mtime.
1010 * If parent mtime has changed, we revalidate, else we wait for a
1011 * period corresponding to the parent's attribute cache timeout value.
1014 int nfs_neg_need_reval(struct inode
*dir
, struct dentry
*dentry
,
1017 /* Don't revalidate a negative dentry if we're creating a new file */
1018 if (flags
& LOOKUP_CREATE
)
1020 if (NFS_SERVER(dir
)->flags
& NFS_MOUNT_LOOKUP_CACHE_NONEG
)
1022 return !nfs_check_verifier(dir
, dentry
);
1026 * This is called every time the dcache has a lookup hit,
1027 * and we should check whether we can really trust that
1030 * NOTE! The hit can be a negative hit too, don't assume
1033 * If the parent directory is seen to have changed, we throw out the
1034 * cached dentry and do a new lookup.
1036 static int nfs_lookup_revalidate(struct dentry
*dentry
, unsigned int flags
)
1039 struct inode
*inode
;
1040 struct dentry
*parent
;
1041 struct nfs_fh
*fhandle
= NULL
;
1042 struct nfs_fattr
*fattr
= NULL
;
1043 struct nfs4_label
*label
= NULL
;
1046 if (flags
& LOOKUP_RCU
)
1049 parent
= dget_parent(dentry
);
1050 dir
= parent
->d_inode
;
1051 nfs_inc_stats(dir
, NFSIOS_DENTRYREVALIDATE
);
1052 inode
= dentry
->d_inode
;
1055 if (nfs_neg_need_reval(dir
, dentry
, flags
))
1057 goto out_valid_noent
;
1060 if (is_bad_inode(inode
)) {
1061 dfprintk(LOOKUPCACHE
, "%s: %s/%s has dud inode\n",
1062 __func__
, dentry
->d_parent
->d_name
.name
,
1063 dentry
->d_name
.name
);
1067 if (NFS_PROTO(dir
)->have_delegation(inode
, FMODE_READ
))
1068 goto out_set_verifier
;
1070 /* Force a full look up iff the parent directory has changed */
1071 if (!nfs_is_exclusive_create(dir
, flags
) && nfs_check_verifier(dir
, dentry
)) {
1072 if (nfs_lookup_verify_inode(inode
, flags
))
1073 goto out_zap_parent
;
1077 if (NFS_STALE(inode
))
1081 fhandle
= nfs_alloc_fhandle();
1082 fattr
= nfs_alloc_fattr();
1083 if (fhandle
== NULL
|| fattr
== NULL
)
1086 error
= NFS_PROTO(dir
)->lookup(dir
, &dentry
->d_name
, fhandle
, fattr
, label
);
1089 if (nfs_compare_fh(NFS_FH(inode
), fhandle
))
1091 if ((error
= nfs_refresh_inode(inode
, fattr
)) != 0)
1094 nfs_free_fattr(fattr
);
1095 nfs_free_fhandle(fhandle
);
1097 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1099 /* Success: notify readdir to use READDIRPLUS */
1100 nfs_advise_use_readdirplus(dir
);
1103 dfprintk(LOOKUPCACHE
, "NFS: %s(%s/%s) is valid\n",
1104 __func__
, dentry
->d_parent
->d_name
.name
,
1105 dentry
->d_name
.name
);
1108 nfs_zap_caches(dir
);
1110 nfs_free_fattr(fattr
);
1111 nfs_free_fhandle(fhandle
);
1112 nfs_mark_for_revalidate(dir
);
1113 if (inode
&& S_ISDIR(inode
->i_mode
)) {
1114 /* Purge readdir caches. */
1115 nfs_zap_caches(inode
);
1116 /* If we have submounts, don't unhash ! */
1117 if (have_submounts(dentry
))
1119 if (dentry
->d_flags
& DCACHE_DISCONNECTED
)
1121 shrink_dcache_parent(dentry
);
1125 dfprintk(LOOKUPCACHE
, "NFS: %s(%s/%s) is invalid\n",
1126 __func__
, dentry
->d_parent
->d_name
.name
,
1127 dentry
->d_name
.name
);
1130 nfs_free_fattr(fattr
);
1131 nfs_free_fhandle(fhandle
);
1133 dfprintk(LOOKUPCACHE
, "NFS: %s(%s/%s) lookup returned error %d\n",
1134 __func__
, dentry
->d_parent
->d_name
.name
,
1135 dentry
->d_name
.name
, error
);
1140 * A weaker form of d_revalidate for revalidating just the dentry->d_inode
1141 * when we don't really care about the dentry name. This is called when a
1142 * pathwalk ends on a dentry that was not found via a normal lookup in the
1143 * parent dir (e.g.: ".", "..", procfs symlinks or mountpoint traversals).
1145 * In this situation, we just want to verify that the inode itself is OK
1146 * since the dentry might have changed on the server.
1148 static int nfs_weak_revalidate(struct dentry
*dentry
, unsigned int flags
)
1151 struct inode
*inode
= dentry
->d_inode
;
1154 * I believe we can only get a negative dentry here in the case of a
1155 * procfs-style symlink. Just assume it's correct for now, but we may
1156 * eventually need to do something more here.
1159 dfprintk(LOOKUPCACHE
, "%s: %s/%s has negative inode\n",
1160 __func__
, dentry
->d_parent
->d_name
.name
,
1161 dentry
->d_name
.name
);
1165 if (is_bad_inode(inode
)) {
1166 dfprintk(LOOKUPCACHE
, "%s: %s/%s has dud inode\n",
1167 __func__
, dentry
->d_parent
->d_name
.name
,
1168 dentry
->d_name
.name
);
1172 error
= nfs_revalidate_inode(NFS_SERVER(inode
), inode
);
1173 dfprintk(LOOKUPCACHE
, "NFS: %s: inode %lu is %s\n",
1174 __func__
, inode
->i_ino
, error
? "invalid" : "valid");
1179 * This is called from dput() when d_count is going to 0.
1181 static int nfs_dentry_delete(const struct dentry
*dentry
)
1183 dfprintk(VFS
, "NFS: dentry_delete(%s/%s, %x)\n",
1184 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
,
1187 /* Unhash any dentry with a stale inode */
1188 if (dentry
->d_inode
!= NULL
&& NFS_STALE(dentry
->d_inode
))
1191 if (dentry
->d_flags
& DCACHE_NFSFS_RENAMED
) {
1192 /* Unhash it, so that ->d_iput() would be called */
1195 if (!(dentry
->d_sb
->s_flags
& MS_ACTIVE
)) {
1196 /* Unhash it, so that ancestors of killed async unlink
1197 * files will be cleaned up during umount */
1204 /* Ensure that we revalidate inode->i_nlink */
1205 static void nfs_drop_nlink(struct inode
*inode
)
1207 spin_lock(&inode
->i_lock
);
1208 /* drop the inode if we're reasonably sure this is the last link */
1209 if (inode
->i_nlink
== 1)
1211 NFS_I(inode
)->cache_validity
|= NFS_INO_INVALID_ATTR
;
1212 spin_unlock(&inode
->i_lock
);
1216 * Called when the dentry loses inode.
1217 * We use it to clean up silly-renamed files.
1219 static void nfs_dentry_iput(struct dentry
*dentry
, struct inode
*inode
)
1221 if (S_ISDIR(inode
->i_mode
))
1222 /* drop any readdir cache as it could easily be old */
1223 NFS_I(inode
)->cache_validity
|= NFS_INO_INVALID_DATA
;
1225 if (dentry
->d_flags
& DCACHE_NFSFS_RENAMED
) {
1226 nfs_complete_unlink(dentry
, inode
);
1227 nfs_drop_nlink(inode
);
1232 static void nfs_d_release(struct dentry
*dentry
)
1234 /* free cached devname value, if it survived that far */
1235 if (unlikely(dentry
->d_fsdata
)) {
1236 if (dentry
->d_flags
& DCACHE_NFSFS_RENAMED
)
1239 kfree(dentry
->d_fsdata
);
1243 const struct dentry_operations nfs_dentry_operations
= {
1244 .d_revalidate
= nfs_lookup_revalidate
,
1245 .d_weak_revalidate
= nfs_weak_revalidate
,
1246 .d_delete
= nfs_dentry_delete
,
1247 .d_iput
= nfs_dentry_iput
,
1248 .d_automount
= nfs_d_automount
,
1249 .d_release
= nfs_d_release
,
1251 EXPORT_SYMBOL_GPL(nfs_dentry_operations
);
1253 struct dentry
*nfs_lookup(struct inode
*dir
, struct dentry
* dentry
, unsigned int flags
)
1256 struct dentry
*parent
;
1257 struct inode
*inode
= NULL
;
1258 struct nfs_fh
*fhandle
= NULL
;
1259 struct nfs_fattr
*fattr
= NULL
;
1260 struct nfs4_label
*label
= NULL
;
1263 dfprintk(VFS
, "NFS: lookup(%s/%s)\n",
1264 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
);
1265 nfs_inc_stats(dir
, NFSIOS_VFSLOOKUP
);
1267 res
= ERR_PTR(-ENAMETOOLONG
);
1268 if (dentry
->d_name
.len
> NFS_SERVER(dir
)->namelen
)
1272 * If we're doing an exclusive create, optimize away the lookup
1273 * but don't hash the dentry.
1275 if (nfs_is_exclusive_create(dir
, flags
)) {
1276 d_instantiate(dentry
, NULL
);
1281 res
= ERR_PTR(-ENOMEM
);
1282 fhandle
= nfs_alloc_fhandle();
1283 fattr
= nfs_alloc_fattr();
1284 if (fhandle
== NULL
|| fattr
== NULL
)
1287 parent
= dentry
->d_parent
;
1288 /* Protect against concurrent sillydeletes */
1289 nfs_block_sillyrename(parent
);
1290 error
= NFS_PROTO(dir
)->lookup(dir
, &dentry
->d_name
, fhandle
, fattr
, label
);
1291 if (error
== -ENOENT
)
1294 res
= ERR_PTR(error
);
1295 goto out_unblock_sillyrename
;
1297 inode
= nfs_fhget(dentry
->d_sb
, fhandle
, fattr
, label
);
1298 res
= ERR_CAST(inode
);
1300 goto out_unblock_sillyrename
;
1302 /* Success: notify readdir to use READDIRPLUS */
1303 nfs_advise_use_readdirplus(dir
);
1306 res
= d_materialise_unique(dentry
, inode
);
1309 goto out_unblock_sillyrename
;
1312 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1313 out_unblock_sillyrename
:
1314 nfs_unblock_sillyrename(parent
);
1316 nfs_free_fattr(fattr
);
1317 nfs_free_fhandle(fhandle
);
1320 EXPORT_SYMBOL_GPL(nfs_lookup
);
1322 #if IS_ENABLED(CONFIG_NFS_V4)
1323 static int nfs4_lookup_revalidate(struct dentry
*, unsigned int);
1325 const struct dentry_operations nfs4_dentry_operations
= {
1326 .d_revalidate
= nfs4_lookup_revalidate
,
1327 .d_delete
= nfs_dentry_delete
,
1328 .d_iput
= nfs_dentry_iput
,
1329 .d_automount
= nfs_d_automount
,
1330 .d_release
= nfs_d_release
,
1332 EXPORT_SYMBOL_GPL(nfs4_dentry_operations
);
1334 static fmode_t
flags_to_mode(int flags
)
1336 fmode_t res
= (__force fmode_t
)flags
& FMODE_EXEC
;
1337 if ((flags
& O_ACCMODE
) != O_WRONLY
)
1339 if ((flags
& O_ACCMODE
) != O_RDONLY
)
1344 static struct nfs_open_context
*create_nfs_open_context(struct dentry
*dentry
, int open_flags
)
1346 return alloc_nfs_open_context(dentry
, flags_to_mode(open_flags
));
1349 static int do_open(struct inode
*inode
, struct file
*filp
)
1351 nfs_fscache_set_inode_cookie(inode
, filp
);
1355 static int nfs_finish_open(struct nfs_open_context
*ctx
,
1356 struct dentry
*dentry
,
1357 struct file
*file
, unsigned open_flags
,
1362 if (ctx
->dentry
!= dentry
) {
1364 ctx
->dentry
= dget(dentry
);
1367 /* If the open_intent is for execute, we have an extra check to make */
1368 if (ctx
->mode
& FMODE_EXEC
) {
1369 err
= nfs_may_open(dentry
->d_inode
, ctx
->cred
, open_flags
);
1374 err
= finish_open(file
, dentry
, do_open
, opened
);
1377 nfs_file_set_open_context(file
, ctx
);
1380 put_nfs_open_context(ctx
);
1384 int nfs_atomic_open(struct inode
*dir
, struct dentry
*dentry
,
1385 struct file
*file
, unsigned open_flags
,
1386 umode_t mode
, int *opened
)
1388 struct nfs_open_context
*ctx
;
1390 struct iattr attr
= { .ia_valid
= ATTR_OPEN
};
1391 struct inode
*inode
;
1394 /* Expect a negative dentry */
1395 BUG_ON(dentry
->d_inode
);
1397 dfprintk(VFS
, "NFS: atomic_open(%s/%ld), %s\n",
1398 dir
->i_sb
->s_id
, dir
->i_ino
, dentry
->d_name
.name
);
1400 /* NFS only supports OPEN on regular files */
1401 if ((open_flags
& O_DIRECTORY
)) {
1402 if (!d_unhashed(dentry
)) {
1404 * Hashed negative dentry with O_DIRECTORY: dentry was
1405 * revalidated and is fine, no need to perform lookup
1413 if (dentry
->d_name
.len
> NFS_SERVER(dir
)->namelen
)
1414 return -ENAMETOOLONG
;
1416 if (open_flags
& O_CREAT
) {
1417 attr
.ia_valid
|= ATTR_MODE
;
1418 attr
.ia_mode
= mode
& ~current_umask();
1420 if (open_flags
& O_TRUNC
) {
1421 attr
.ia_valid
|= ATTR_SIZE
;
1425 ctx
= create_nfs_open_context(dentry
, open_flags
);
1430 nfs_block_sillyrename(dentry
->d_parent
);
1431 inode
= NFS_PROTO(dir
)->open_context(dir
, ctx
, open_flags
, &attr
);
1433 if (IS_ERR(inode
)) {
1434 nfs_unblock_sillyrename(dentry
->d_parent
);
1435 put_nfs_open_context(ctx
);
1436 err
= PTR_ERR(inode
);
1439 d_add(dentry
, NULL
);
1445 if (!(open_flags
& O_NOFOLLOW
))
1454 res
= d_add_unique(dentry
, inode
);
1458 nfs_unblock_sillyrename(dentry
->d_parent
);
1459 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1461 err
= nfs_finish_open(ctx
, dentry
, file
, open_flags
, opened
);
1468 res
= nfs_lookup(dir
, dentry
, 0);
1473 return finish_no_open(file
, res
);
1475 EXPORT_SYMBOL_GPL(nfs_atomic_open
);
1477 static int nfs4_lookup_revalidate(struct dentry
*dentry
, unsigned int flags
)
1479 struct dentry
*parent
= NULL
;
1480 struct inode
*inode
;
1484 if (flags
& LOOKUP_RCU
)
1487 if (!(flags
& LOOKUP_OPEN
) || (flags
& LOOKUP_DIRECTORY
))
1489 if (d_mountpoint(dentry
))
1491 if (NFS_SB(dentry
->d_sb
)->caps
& NFS_CAP_ATOMIC_OPEN_V1
)
1494 inode
= dentry
->d_inode
;
1495 parent
= dget_parent(dentry
);
1496 dir
= parent
->d_inode
;
1498 /* We can't create new files in nfs_open_revalidate(), so we
1499 * optimize away revalidation of negative dentries.
1501 if (inode
== NULL
) {
1502 if (!nfs_neg_need_reval(dir
, dentry
, flags
))
1507 /* NFS only supports OPEN on regular files */
1508 if (!S_ISREG(inode
->i_mode
))
1510 /* We cannot do exclusive creation on a positive dentry */
1511 if (flags
& LOOKUP_EXCL
)
1514 /* Let f_op->open() actually open (and revalidate) the file */
1524 return nfs_lookup_revalidate(dentry
, flags
);
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
,
1534 struct nfs4_label
*label
)
1536 struct dentry
*parent
= dget_parent(dentry
);
1537 struct inode
*dir
= parent
->d_inode
;
1538 struct inode
*inode
;
1539 int error
= -EACCES
;
1543 /* We may have been initialized further down */
1544 if (dentry
->d_inode
)
1546 if (fhandle
->size
== 0) {
1547 error
= NFS_PROTO(dir
)->lookup(dir
, &dentry
->d_name
, fhandle
, fattr
, NULL
);
1551 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1552 if (!(fattr
->valid
& NFS_ATTR_FATTR
)) {
1553 struct nfs_server
*server
= NFS_SB(dentry
->d_sb
);
1554 error
= server
->nfs_client
->rpc_ops
->getattr(server
, fhandle
, fattr
, NULL
);
1558 inode
= nfs_fhget(dentry
->d_sb
, fhandle
, fattr
, label
);
1559 error
= PTR_ERR(inode
);
1562 d_add(dentry
, inode
);
1567 nfs_mark_for_revalidate(dir
);
1571 EXPORT_SYMBOL_GPL(nfs_instantiate
);
1574 * Following a failed create operation, we drop the dentry rather
1575 * than retain a negative dentry. This avoids a problem in the event
1576 * that the operation succeeded on the server, but an error in the
1577 * reply path made it appear to have failed.
1579 int nfs_create(struct inode
*dir
, struct dentry
*dentry
,
1580 umode_t mode
, bool excl
)
1583 int open_flags
= excl
? O_CREAT
| O_EXCL
: O_CREAT
;
1586 dfprintk(VFS
, "NFS: create(%s/%ld), %s\n",
1587 dir
->i_sb
->s_id
, dir
->i_ino
, dentry
->d_name
.name
);
1589 attr
.ia_mode
= mode
;
1590 attr
.ia_valid
= ATTR_MODE
;
1592 error
= NFS_PROTO(dir
)->create(dir
, dentry
, &attr
, open_flags
);
1600 EXPORT_SYMBOL_GPL(nfs_create
);
1603 * See comments for nfs_proc_create regarding failed operations.
1606 nfs_mknod(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
, dev_t rdev
)
1611 dfprintk(VFS
, "NFS: mknod(%s/%ld), %s\n",
1612 dir
->i_sb
->s_id
, dir
->i_ino
, dentry
->d_name
.name
);
1614 if (!new_valid_dev(rdev
))
1617 attr
.ia_mode
= mode
;
1618 attr
.ia_valid
= ATTR_MODE
;
1620 status
= NFS_PROTO(dir
)->mknod(dir
, dentry
, &attr
, rdev
);
1628 EXPORT_SYMBOL_GPL(nfs_mknod
);
1631 * See comments for nfs_proc_create regarding failed operations.
1633 int nfs_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
1638 dfprintk(VFS
, "NFS: mkdir(%s/%ld), %s\n",
1639 dir
->i_sb
->s_id
, dir
->i_ino
, dentry
->d_name
.name
);
1641 attr
.ia_valid
= ATTR_MODE
;
1642 attr
.ia_mode
= mode
| S_IFDIR
;
1644 error
= NFS_PROTO(dir
)->mkdir(dir
, dentry
, &attr
);
1652 EXPORT_SYMBOL_GPL(nfs_mkdir
);
1654 static void nfs_dentry_handle_enoent(struct dentry
*dentry
)
1656 if (dentry
->d_inode
!= NULL
&& !d_unhashed(dentry
))
1660 int nfs_rmdir(struct inode
*dir
, struct dentry
*dentry
)
1664 dfprintk(VFS
, "NFS: rmdir(%s/%ld), %s\n",
1665 dir
->i_sb
->s_id
, dir
->i_ino
, dentry
->d_name
.name
);
1667 error
= NFS_PROTO(dir
)->rmdir(dir
, &dentry
->d_name
);
1668 /* Ensure the VFS deletes this inode */
1669 if (error
== 0 && dentry
->d_inode
!= NULL
)
1670 clear_nlink(dentry
->d_inode
);
1671 else if (error
== -ENOENT
)
1672 nfs_dentry_handle_enoent(dentry
);
1676 EXPORT_SYMBOL_GPL(nfs_rmdir
);
1679 * Remove a file after making sure there are no pending writes,
1680 * and after checking that the file has only one user.
1682 * We invalidate the attribute cache and free the inode prior to the operation
1683 * to avoid possible races if the server reuses the inode.
1685 static int nfs_safe_remove(struct dentry
*dentry
)
1687 struct inode
*dir
= dentry
->d_parent
->d_inode
;
1688 struct inode
*inode
= dentry
->d_inode
;
1691 dfprintk(VFS
, "NFS: safe_remove(%s/%s)\n",
1692 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
);
1694 /* If the dentry was sillyrenamed, we simply call d_delete() */
1695 if (dentry
->d_flags
& DCACHE_NFSFS_RENAMED
) {
1700 if (inode
!= NULL
) {
1701 NFS_PROTO(inode
)->return_delegation(inode
);
1702 error
= NFS_PROTO(dir
)->remove(dir
, &dentry
->d_name
);
1704 nfs_drop_nlink(inode
);
1706 error
= NFS_PROTO(dir
)->remove(dir
, &dentry
->d_name
);
1707 if (error
== -ENOENT
)
1708 nfs_dentry_handle_enoent(dentry
);
1713 /* We do silly rename. In case sillyrename() returns -EBUSY, the inode
1714 * belongs to an active ".nfs..." file and we return -EBUSY.
1716 * If sillyrename() returns 0, we do nothing, otherwise we unlink.
1718 int nfs_unlink(struct inode
*dir
, struct dentry
*dentry
)
1721 int need_rehash
= 0;
1723 dfprintk(VFS
, "NFS: unlink(%s/%ld, %s)\n", dir
->i_sb
->s_id
,
1724 dir
->i_ino
, dentry
->d_name
.name
);
1726 spin_lock(&dentry
->d_lock
);
1727 if (dentry
->d_count
> 1) {
1728 spin_unlock(&dentry
->d_lock
);
1729 /* Start asynchronous writeout of the inode */
1730 write_inode_now(dentry
->d_inode
, 0);
1731 error
= nfs_sillyrename(dir
, dentry
);
1734 if (!d_unhashed(dentry
)) {
1738 spin_unlock(&dentry
->d_lock
);
1739 error
= nfs_safe_remove(dentry
);
1740 if (!error
|| error
== -ENOENT
) {
1741 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1742 } else if (need_rehash
)
1746 EXPORT_SYMBOL_GPL(nfs_unlink
);
1749 * To create a symbolic link, most file systems instantiate a new inode,
1750 * add a page to it containing the path, then write it out to the disk
1751 * using prepare_write/commit_write.
1753 * Unfortunately the NFS client can't create the in-core inode first
1754 * because it needs a file handle to create an in-core inode (see
1755 * fs/nfs/inode.c:nfs_fhget). We only have a file handle *after* the
1756 * symlink request has completed on the server.
1758 * So instead we allocate a raw page, copy the symname into it, then do
1759 * the SYMLINK request with the page as the buffer. If it succeeds, we
1760 * now have a new file handle and can instantiate an in-core NFS inode
1761 * and move the raw page into its mapping.
1763 int nfs_symlink(struct inode
*dir
, struct dentry
*dentry
, const char *symname
)
1765 struct pagevec lru_pvec
;
1769 unsigned int pathlen
= strlen(symname
);
1772 dfprintk(VFS
, "NFS: symlink(%s/%ld, %s, %s)\n", dir
->i_sb
->s_id
,
1773 dir
->i_ino
, dentry
->d_name
.name
, symname
);
1775 if (pathlen
> PAGE_SIZE
)
1776 return -ENAMETOOLONG
;
1778 attr
.ia_mode
= S_IFLNK
| S_IRWXUGO
;
1779 attr
.ia_valid
= ATTR_MODE
;
1781 page
= alloc_page(GFP_HIGHUSER
);
1785 kaddr
= kmap_atomic(page
);
1786 memcpy(kaddr
, symname
, pathlen
);
1787 if (pathlen
< PAGE_SIZE
)
1788 memset(kaddr
+ pathlen
, 0, PAGE_SIZE
- pathlen
);
1789 kunmap_atomic(kaddr
);
1791 error
= NFS_PROTO(dir
)->symlink(dir
, dentry
, page
, pathlen
, &attr
);
1793 dfprintk(VFS
, "NFS: symlink(%s/%ld, %s, %s) error %d\n",
1794 dir
->i_sb
->s_id
, dir
->i_ino
,
1795 dentry
->d_name
.name
, symname
, error
);
1802 * No big deal if we can't add this page to the page cache here.
1803 * READLINK will get the missing page from the server if needed.
1805 pagevec_init(&lru_pvec
, 0);
1806 if (!add_to_page_cache(page
, dentry
->d_inode
->i_mapping
, 0,
1808 pagevec_add(&lru_pvec
, page
);
1809 pagevec_lru_add_file(&lru_pvec
);
1810 SetPageUptodate(page
);
1817 EXPORT_SYMBOL_GPL(nfs_symlink
);
1820 nfs_link(struct dentry
*old_dentry
, struct inode
*dir
, struct dentry
*dentry
)
1822 struct inode
*inode
= old_dentry
->d_inode
;
1825 dfprintk(VFS
, "NFS: link(%s/%s -> %s/%s)\n",
1826 old_dentry
->d_parent
->d_name
.name
, old_dentry
->d_name
.name
,
1827 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
);
1829 NFS_PROTO(inode
)->return_delegation(inode
);
1832 error
= NFS_PROTO(dir
)->link(inode
, dir
, &dentry
->d_name
);
1835 d_add(dentry
, inode
);
1839 EXPORT_SYMBOL_GPL(nfs_link
);
1843 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a
1844 * different file handle for the same inode after a rename (e.g. when
1845 * moving to a different directory). A fail-safe method to do so would
1846 * be to look up old_dir/old_name, create a link to new_dir/new_name and
1847 * rename the old file using the sillyrename stuff. This way, the original
1848 * file in old_dir will go away when the last process iput()s the inode.
1852 * It actually works quite well. One needs to have the possibility for
1853 * at least one ".nfs..." file in each directory the file ever gets
1854 * moved or linked to which happens automagically with the new
1855 * implementation that only depends on the dcache stuff instead of
1856 * using the inode layer
1858 * Unfortunately, things are a little more complicated than indicated
1859 * above. For a cross-directory move, we want to make sure we can get
1860 * rid of the old inode after the operation. This means there must be
1861 * no pending writes (if it's a file), and the use count must be 1.
1862 * If these conditions are met, we can drop the dentries before doing
1865 int nfs_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
1866 struct inode
*new_dir
, struct dentry
*new_dentry
)
1868 struct inode
*old_inode
= old_dentry
->d_inode
;
1869 struct inode
*new_inode
= new_dentry
->d_inode
;
1870 struct dentry
*dentry
= NULL
, *rehash
= NULL
;
1873 dfprintk(VFS
, "NFS: rename(%s/%s -> %s/%s, ct=%d)\n",
1874 old_dentry
->d_parent
->d_name
.name
, old_dentry
->d_name
.name
,
1875 new_dentry
->d_parent
->d_name
.name
, new_dentry
->d_name
.name
,
1876 new_dentry
->d_count
);
1879 * For non-directories, check whether the target is busy and if so,
1880 * make a copy of the dentry and then do a silly-rename. If the
1881 * silly-rename succeeds, the copied dentry is hashed and becomes
1884 if (new_inode
&& !S_ISDIR(new_inode
->i_mode
)) {
1886 * To prevent any new references to the target during the
1887 * rename, we unhash the dentry in advance.
1889 if (!d_unhashed(new_dentry
)) {
1891 rehash
= new_dentry
;
1894 if (new_dentry
->d_count
> 2) {
1897 /* copy the target dentry's name */
1898 dentry
= d_alloc(new_dentry
->d_parent
,
1899 &new_dentry
->d_name
);
1903 /* silly-rename the existing target ... */
1904 err
= nfs_sillyrename(new_dir
, new_dentry
);
1908 new_dentry
= dentry
;
1914 NFS_PROTO(old_inode
)->return_delegation(old_inode
);
1915 if (new_inode
!= NULL
)
1916 NFS_PROTO(new_inode
)->return_delegation(new_inode
);
1918 error
= NFS_PROTO(old_dir
)->rename(old_dir
, &old_dentry
->d_name
,
1919 new_dir
, &new_dentry
->d_name
);
1920 nfs_mark_for_revalidate(old_inode
);
1925 if (new_inode
!= NULL
)
1926 nfs_drop_nlink(new_inode
);
1927 d_move(old_dentry
, new_dentry
);
1928 nfs_set_verifier(new_dentry
,
1929 nfs_save_change_attribute(new_dir
));
1930 } else if (error
== -ENOENT
)
1931 nfs_dentry_handle_enoent(old_dentry
);
1933 /* new dentry created? */
1938 EXPORT_SYMBOL_GPL(nfs_rename
);
1940 static DEFINE_SPINLOCK(nfs_access_lru_lock
);
1941 static LIST_HEAD(nfs_access_lru_list
);
1942 static atomic_long_t nfs_access_nr_entries
;
1944 static void nfs_access_free_entry(struct nfs_access_entry
*entry
)
1946 put_rpccred(entry
->cred
);
1948 smp_mb__before_atomic_dec();
1949 atomic_long_dec(&nfs_access_nr_entries
);
1950 smp_mb__after_atomic_dec();
1953 static void nfs_access_free_list(struct list_head
*head
)
1955 struct nfs_access_entry
*cache
;
1957 while (!list_empty(head
)) {
1958 cache
= list_entry(head
->next
, struct nfs_access_entry
, lru
);
1959 list_del(&cache
->lru
);
1960 nfs_access_free_entry(cache
);
1964 int nfs_access_cache_shrinker(struct shrinker
*shrink
,
1965 struct shrink_control
*sc
)
1968 struct nfs_inode
*nfsi
, *next
;
1969 struct nfs_access_entry
*cache
;
1970 int nr_to_scan
= sc
->nr_to_scan
;
1971 gfp_t gfp_mask
= sc
->gfp_mask
;
1973 if ((gfp_mask
& GFP_KERNEL
) != GFP_KERNEL
)
1974 return (nr_to_scan
== 0) ? 0 : -1;
1976 spin_lock(&nfs_access_lru_lock
);
1977 list_for_each_entry_safe(nfsi
, next
, &nfs_access_lru_list
, access_cache_inode_lru
) {
1978 struct inode
*inode
;
1980 if (nr_to_scan
-- == 0)
1982 inode
= &nfsi
->vfs_inode
;
1983 spin_lock(&inode
->i_lock
);
1984 if (list_empty(&nfsi
->access_cache_entry_lru
))
1985 goto remove_lru_entry
;
1986 cache
= list_entry(nfsi
->access_cache_entry_lru
.next
,
1987 struct nfs_access_entry
, lru
);
1988 list_move(&cache
->lru
, &head
);
1989 rb_erase(&cache
->rb_node
, &nfsi
->access_cache
);
1990 if (!list_empty(&nfsi
->access_cache_entry_lru
))
1991 list_move_tail(&nfsi
->access_cache_inode_lru
,
1992 &nfs_access_lru_list
);
1995 list_del_init(&nfsi
->access_cache_inode_lru
);
1996 smp_mb__before_clear_bit();
1997 clear_bit(NFS_INO_ACL_LRU_SET
, &nfsi
->flags
);
1998 smp_mb__after_clear_bit();
2000 spin_unlock(&inode
->i_lock
);
2002 spin_unlock(&nfs_access_lru_lock
);
2003 nfs_access_free_list(&head
);
2004 return (atomic_long_read(&nfs_access_nr_entries
) / 100) * sysctl_vfs_cache_pressure
;
2007 static void __nfs_access_zap_cache(struct nfs_inode
*nfsi
, struct list_head
*head
)
2009 struct rb_root
*root_node
= &nfsi
->access_cache
;
2011 struct nfs_access_entry
*entry
;
2013 /* Unhook entries from the cache */
2014 while ((n
= rb_first(root_node
)) != NULL
) {
2015 entry
= rb_entry(n
, struct nfs_access_entry
, rb_node
);
2016 rb_erase(n
, root_node
);
2017 list_move(&entry
->lru
, head
);
2019 nfsi
->cache_validity
&= ~NFS_INO_INVALID_ACCESS
;
2022 void nfs_access_zap_cache(struct inode
*inode
)
2026 if (test_bit(NFS_INO_ACL_LRU_SET
, &NFS_I(inode
)->flags
) == 0)
2028 /* Remove from global LRU init */
2029 spin_lock(&nfs_access_lru_lock
);
2030 if (test_and_clear_bit(NFS_INO_ACL_LRU_SET
, &NFS_I(inode
)->flags
))
2031 list_del_init(&NFS_I(inode
)->access_cache_inode_lru
);
2033 spin_lock(&inode
->i_lock
);
2034 __nfs_access_zap_cache(NFS_I(inode
), &head
);
2035 spin_unlock(&inode
->i_lock
);
2036 spin_unlock(&nfs_access_lru_lock
);
2037 nfs_access_free_list(&head
);
2039 EXPORT_SYMBOL_GPL(nfs_access_zap_cache
);
2041 static struct nfs_access_entry
*nfs_access_search_rbtree(struct inode
*inode
, struct rpc_cred
*cred
)
2043 struct rb_node
*n
= NFS_I(inode
)->access_cache
.rb_node
;
2044 struct nfs_access_entry
*entry
;
2047 entry
= rb_entry(n
, struct nfs_access_entry
, rb_node
);
2049 if (cred
< entry
->cred
)
2051 else if (cred
> entry
->cred
)
2059 static int nfs_access_get_cached(struct inode
*inode
, struct rpc_cred
*cred
, struct nfs_access_entry
*res
)
2061 struct nfs_inode
*nfsi
= NFS_I(inode
);
2062 struct nfs_access_entry
*cache
;
2065 spin_lock(&inode
->i_lock
);
2066 if (nfsi
->cache_validity
& NFS_INO_INVALID_ACCESS
)
2068 cache
= nfs_access_search_rbtree(inode
, cred
);
2071 if (!nfs_have_delegated_attributes(inode
) &&
2072 !time_in_range_open(jiffies
, cache
->jiffies
, cache
->jiffies
+ nfsi
->attrtimeo
))
2074 res
->jiffies
= cache
->jiffies
;
2075 res
->cred
= cache
->cred
;
2076 res
->mask
= cache
->mask
;
2077 list_move_tail(&cache
->lru
, &nfsi
->access_cache_entry_lru
);
2080 spin_unlock(&inode
->i_lock
);
2083 rb_erase(&cache
->rb_node
, &nfsi
->access_cache
);
2084 list_del(&cache
->lru
);
2085 spin_unlock(&inode
->i_lock
);
2086 nfs_access_free_entry(cache
);
2089 spin_unlock(&inode
->i_lock
);
2090 nfs_access_zap_cache(inode
);
2094 static void nfs_access_add_rbtree(struct inode
*inode
, struct nfs_access_entry
*set
)
2096 struct nfs_inode
*nfsi
= NFS_I(inode
);
2097 struct rb_root
*root_node
= &nfsi
->access_cache
;
2098 struct rb_node
**p
= &root_node
->rb_node
;
2099 struct rb_node
*parent
= NULL
;
2100 struct nfs_access_entry
*entry
;
2102 spin_lock(&inode
->i_lock
);
2103 while (*p
!= NULL
) {
2105 entry
= rb_entry(parent
, struct nfs_access_entry
, rb_node
);
2107 if (set
->cred
< entry
->cred
)
2108 p
= &parent
->rb_left
;
2109 else if (set
->cred
> entry
->cred
)
2110 p
= &parent
->rb_right
;
2114 rb_link_node(&set
->rb_node
, parent
, p
);
2115 rb_insert_color(&set
->rb_node
, root_node
);
2116 list_add_tail(&set
->lru
, &nfsi
->access_cache_entry_lru
);
2117 spin_unlock(&inode
->i_lock
);
2120 rb_replace_node(parent
, &set
->rb_node
, root_node
);
2121 list_add_tail(&set
->lru
, &nfsi
->access_cache_entry_lru
);
2122 list_del(&entry
->lru
);
2123 spin_unlock(&inode
->i_lock
);
2124 nfs_access_free_entry(entry
);
2127 void nfs_access_add_cache(struct inode
*inode
, struct nfs_access_entry
*set
)
2129 struct nfs_access_entry
*cache
= kmalloc(sizeof(*cache
), GFP_KERNEL
);
2132 RB_CLEAR_NODE(&cache
->rb_node
);
2133 cache
->jiffies
= set
->jiffies
;
2134 cache
->cred
= get_rpccred(set
->cred
);
2135 cache
->mask
= set
->mask
;
2137 nfs_access_add_rbtree(inode
, cache
);
2139 /* Update accounting */
2140 smp_mb__before_atomic_inc();
2141 atomic_long_inc(&nfs_access_nr_entries
);
2142 smp_mb__after_atomic_inc();
2144 /* Add inode to global LRU list */
2145 if (!test_bit(NFS_INO_ACL_LRU_SET
, &NFS_I(inode
)->flags
)) {
2146 spin_lock(&nfs_access_lru_lock
);
2147 if (!test_and_set_bit(NFS_INO_ACL_LRU_SET
, &NFS_I(inode
)->flags
))
2148 list_add_tail(&NFS_I(inode
)->access_cache_inode_lru
,
2149 &nfs_access_lru_list
);
2150 spin_unlock(&nfs_access_lru_lock
);
2153 EXPORT_SYMBOL_GPL(nfs_access_add_cache
);
2155 void nfs_access_set_mask(struct nfs_access_entry
*entry
, u32 access_result
)
2158 if (access_result
& NFS4_ACCESS_READ
)
2159 entry
->mask
|= MAY_READ
;
2161 (NFS4_ACCESS_MODIFY
| NFS4_ACCESS_EXTEND
| NFS4_ACCESS_DELETE
))
2162 entry
->mask
|= MAY_WRITE
;
2163 if (access_result
& (NFS4_ACCESS_LOOKUP
|NFS4_ACCESS_EXECUTE
))
2164 entry
->mask
|= MAY_EXEC
;
2166 EXPORT_SYMBOL_GPL(nfs_access_set_mask
);
2168 static int nfs_do_access(struct inode
*inode
, struct rpc_cred
*cred
, int mask
)
2170 struct nfs_access_entry cache
;
2173 status
= nfs_access_get_cached(inode
, cred
, &cache
);
2177 /* Be clever: ask server to check for all possible rights */
2178 cache
.mask
= MAY_EXEC
| MAY_WRITE
| MAY_READ
;
2180 cache
.jiffies
= jiffies
;
2181 status
= NFS_PROTO(inode
)->access(inode
, &cache
);
2183 if (status
== -ESTALE
) {
2184 nfs_zap_caches(inode
);
2185 if (!S_ISDIR(inode
->i_mode
))
2186 set_bit(NFS_INO_STALE
, &NFS_I(inode
)->flags
);
2190 nfs_access_add_cache(inode
, &cache
);
2192 if ((mask
& ~cache
.mask
& (MAY_READ
| MAY_WRITE
| MAY_EXEC
)) == 0)
2197 static int nfs_open_permission_mask(int openflags
)
2201 if (openflags
& __FMODE_EXEC
) {
2202 /* ONLY check exec rights */
2205 if ((openflags
& O_ACCMODE
) != O_WRONLY
)
2207 if ((openflags
& O_ACCMODE
) != O_RDONLY
)
2214 int nfs_may_open(struct inode
*inode
, struct rpc_cred
*cred
, int openflags
)
2216 return nfs_do_access(inode
, cred
, nfs_open_permission_mask(openflags
));
2218 EXPORT_SYMBOL_GPL(nfs_may_open
);
2220 int nfs_permission(struct inode
*inode
, int mask
)
2222 struct rpc_cred
*cred
;
2225 if (mask
& MAY_NOT_BLOCK
)
2228 nfs_inc_stats(inode
, NFSIOS_VFSACCESS
);
2230 if ((mask
& (MAY_READ
| MAY_WRITE
| MAY_EXEC
)) == 0)
2232 /* Is this sys_access() ? */
2233 if (mask
& (MAY_ACCESS
| MAY_CHDIR
))
2236 switch (inode
->i_mode
& S_IFMT
) {
2240 /* NFSv4 has atomic_open... */
2241 if (nfs_server_capable(inode
, NFS_CAP_ATOMIC_OPEN
)
2242 && (mask
& MAY_OPEN
)
2243 && !(mask
& MAY_EXEC
))
2248 * Optimize away all write operations, since the server
2249 * will check permissions when we perform the op.
2251 if ((mask
& MAY_WRITE
) && !(mask
& MAY_READ
))
2256 if (!NFS_PROTO(inode
)->access
)
2259 cred
= rpc_lookup_cred();
2260 if (!IS_ERR(cred
)) {
2261 res
= nfs_do_access(inode
, cred
, mask
);
2264 res
= PTR_ERR(cred
);
2266 if (!res
&& (mask
& MAY_EXEC
) && !execute_ok(inode
))
2269 dfprintk(VFS
, "NFS: permission(%s/%ld), mask=0x%x, res=%d\n",
2270 inode
->i_sb
->s_id
, inode
->i_ino
, mask
, res
);
2273 res
= nfs_revalidate_inode(NFS_SERVER(inode
), inode
);
2275 res
= generic_permission(inode
, mask
);
2278 EXPORT_SYMBOL_GPL(nfs_permission
);
2282 * version-control: t
2283 * kept-new-versions: 5