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/swap.h>
37 #include <linux/sched.h>
38 #include <linux/kmemleak.h>
39 #include <linux/xattr.h>
41 #include "delegation.h"
46 /* #define NFS_DEBUG_VERBOSE 1 */
48 static int nfs_opendir(struct inode
*, struct file
*);
49 static int nfs_closedir(struct inode
*, struct file
*);
50 static int nfs_readdir(struct file
*, struct dir_context
*);
51 static int nfs_fsync_dir(struct file
*, loff_t
, loff_t
, int);
52 static loff_t
nfs_llseek_dir(struct file
*, loff_t
, int);
53 static void nfs_readdir_clear_array(struct page
*);
55 const struct file_operations nfs_dir_operations
= {
56 .llseek
= nfs_llseek_dir
,
57 .read
= generic_read_dir
,
58 .iterate
= nfs_readdir
,
60 .release
= nfs_closedir
,
61 .fsync
= nfs_fsync_dir
,
64 const struct address_space_operations nfs_dir_aops
= {
65 .freepage
= nfs_readdir_clear_array
,
68 static struct nfs_open_dir_context
*alloc_nfs_open_dir_context(struct inode
*dir
, struct rpc_cred
*cred
)
70 struct nfs_open_dir_context
*ctx
;
71 ctx
= kmalloc(sizeof(*ctx
), GFP_KERNEL
);
74 ctx
->attr_gencount
= NFS_I(dir
)->attr_gencount
;
77 ctx
->cred
= get_rpccred(cred
);
80 return ERR_PTR(-ENOMEM
);
83 static void put_nfs_open_dir_context(struct nfs_open_dir_context
*ctx
)
85 put_rpccred(ctx
->cred
);
93 nfs_opendir(struct inode
*inode
, struct file
*filp
)
96 struct nfs_open_dir_context
*ctx
;
97 struct rpc_cred
*cred
;
99 dfprintk(FILE, "NFS: open dir(%s/%s)\n",
100 filp
->f_path
.dentry
->d_parent
->d_name
.name
,
101 filp
->f_path
.dentry
->d_name
.name
);
103 nfs_inc_stats(inode
, NFSIOS_VFSOPEN
);
105 cred
= rpc_lookup_cred();
107 return PTR_ERR(cred
);
108 ctx
= alloc_nfs_open_dir_context(inode
, cred
);
113 filp
->private_data
= ctx
;
114 if (filp
->f_path
.dentry
== filp
->f_path
.mnt
->mnt_root
) {
115 /* This is a mountpoint, so d_revalidate will never
116 * have been called, so we need to refresh the
117 * inode (for close-open consistency) ourselves.
119 __nfs_revalidate_inode(NFS_SERVER(inode
), inode
);
127 nfs_closedir(struct inode
*inode
, struct file
*filp
)
129 put_nfs_open_dir_context(filp
->private_data
);
133 struct nfs_cache_array_entry
{
137 unsigned char d_type
;
140 struct nfs_cache_array
{
144 struct nfs_cache_array_entry array
[0];
147 typedef int (*decode_dirent_t
)(struct xdr_stream
*, struct nfs_entry
*, int);
151 struct dir_context
*ctx
;
152 unsigned long page_index
;
155 loff_t current_index
;
156 decode_dirent_t decode
;
158 unsigned long timestamp
;
159 unsigned long gencount
;
160 unsigned int cache_entry_index
;
163 } nfs_readdir_descriptor_t
;
166 * The caller is responsible for calling nfs_readdir_release_array(page)
169 struct nfs_cache_array
*nfs_readdir_get_array(struct page
*page
)
173 return ERR_PTR(-EIO
);
176 return ERR_PTR(-ENOMEM
);
181 void nfs_readdir_release_array(struct page
*page
)
187 * we are freeing strings created by nfs_add_to_readdir_array()
190 void nfs_readdir_clear_array(struct page
*page
)
192 struct nfs_cache_array
*array
;
195 array
= kmap_atomic(page
);
196 for (i
= 0; i
< array
->size
; i
++)
197 kfree(array
->array
[i
].string
.name
);
198 kunmap_atomic(array
);
202 * the caller is responsible for freeing qstr.name
203 * when called by nfs_readdir_add_to_array, the strings will be freed in
204 * nfs_clear_readdir_array()
207 int nfs_readdir_make_qstr(struct qstr
*string
, const char *name
, unsigned int len
)
210 string
->name
= kmemdup(name
, len
, GFP_KERNEL
);
211 if (string
->name
== NULL
)
214 * Avoid a kmemleak false positive. The pointer to the name is stored
215 * in a page cache page which kmemleak does not scan.
217 kmemleak_not_leak(string
->name
);
218 string
->hash
= full_name_hash(name
, len
);
223 int nfs_readdir_add_to_array(struct nfs_entry
*entry
, struct page
*page
)
225 struct nfs_cache_array
*array
= nfs_readdir_get_array(page
);
226 struct nfs_cache_array_entry
*cache_entry
;
230 return PTR_ERR(array
);
232 cache_entry
= &array
->array
[array
->size
];
234 /* Check that this entry lies within the page bounds */
236 if ((char *)&cache_entry
[1] - (char *)page_address(page
) > PAGE_SIZE
)
239 cache_entry
->cookie
= entry
->prev_cookie
;
240 cache_entry
->ino
= entry
->ino
;
241 cache_entry
->d_type
= entry
->d_type
;
242 ret
= nfs_readdir_make_qstr(&cache_entry
->string
, entry
->name
, entry
->len
);
245 array
->last_cookie
= entry
->cookie
;
248 array
->eof_index
= array
->size
;
250 nfs_readdir_release_array(page
);
255 int nfs_readdir_search_for_pos(struct nfs_cache_array
*array
, nfs_readdir_descriptor_t
*desc
)
257 loff_t diff
= desc
->ctx
->pos
- desc
->current_index
;
262 if (diff
>= array
->size
) {
263 if (array
->eof_index
>= 0)
268 index
= (unsigned int)diff
;
269 *desc
->dir_cookie
= array
->array
[index
].cookie
;
270 desc
->cache_entry_index
= index
;
278 int nfs_readdir_search_for_cookie(struct nfs_cache_array
*array
, nfs_readdir_descriptor_t
*desc
)
282 int status
= -EAGAIN
;
284 for (i
= 0; i
< array
->size
; i
++) {
285 if (array
->array
[i
].cookie
== *desc
->dir_cookie
) {
286 struct nfs_inode
*nfsi
= NFS_I(file_inode(desc
->file
));
287 struct nfs_open_dir_context
*ctx
= desc
->file
->private_data
;
289 new_pos
= desc
->current_index
+ i
;
290 if (ctx
->attr_gencount
!= nfsi
->attr_gencount
291 || (nfsi
->cache_validity
& (NFS_INO_INVALID_ATTR
|NFS_INO_INVALID_DATA
))) {
293 ctx
->attr_gencount
= nfsi
->attr_gencount
;
294 } else if (new_pos
< desc
->ctx
->pos
) {
296 && ctx
->dup_cookie
== *desc
->dir_cookie
) {
297 if (printk_ratelimit()) {
298 pr_notice("NFS: directory %s/%s contains a readdir loop."
299 "Please contact your server vendor. "
300 "The file: %s has duplicate cookie %llu\n",
301 desc
->file
->f_dentry
->d_parent
->d_name
.name
,
302 desc
->file
->f_dentry
->d_name
.name
,
303 array
->array
[i
].string
.name
,
309 ctx
->dup_cookie
= *desc
->dir_cookie
;
312 desc
->ctx
->pos
= new_pos
;
313 desc
->cache_entry_index
= i
;
317 if (array
->eof_index
>= 0) {
318 status
= -EBADCOOKIE
;
319 if (*desc
->dir_cookie
== array
->last_cookie
)
327 int nfs_readdir_search_array(nfs_readdir_descriptor_t
*desc
)
329 struct nfs_cache_array
*array
;
332 array
= nfs_readdir_get_array(desc
->page
);
334 status
= PTR_ERR(array
);
338 if (*desc
->dir_cookie
== 0)
339 status
= nfs_readdir_search_for_pos(array
, desc
);
341 status
= nfs_readdir_search_for_cookie(array
, desc
);
343 if (status
== -EAGAIN
) {
344 desc
->last_cookie
= array
->last_cookie
;
345 desc
->current_index
+= array
->size
;
348 nfs_readdir_release_array(desc
->page
);
353 /* Fill a page with xdr information before transferring to the cache page */
355 int nfs_readdir_xdr_filler(struct page
**pages
, nfs_readdir_descriptor_t
*desc
,
356 struct nfs_entry
*entry
, struct file
*file
, struct inode
*inode
)
358 struct nfs_open_dir_context
*ctx
= file
->private_data
;
359 struct rpc_cred
*cred
= ctx
->cred
;
360 unsigned long timestamp
, gencount
;
365 gencount
= nfs_inc_attr_generation_counter();
366 error
= NFS_PROTO(inode
)->readdir(file
->f_path
.dentry
, cred
, entry
->cookie
, pages
,
367 NFS_SERVER(inode
)->dtsize
, desc
->plus
);
369 /* We requested READDIRPLUS, but the server doesn't grok it */
370 if (error
== -ENOTSUPP
&& desc
->plus
) {
371 NFS_SERVER(inode
)->caps
&= ~NFS_CAP_READDIRPLUS
;
372 clear_bit(NFS_INO_ADVISE_RDPLUS
, &NFS_I(inode
)->flags
);
378 desc
->timestamp
= timestamp
;
379 desc
->gencount
= gencount
;
384 static int xdr_decode(nfs_readdir_descriptor_t
*desc
,
385 struct nfs_entry
*entry
, struct xdr_stream
*xdr
)
389 error
= desc
->decode(xdr
, entry
, desc
->plus
);
392 entry
->fattr
->time_start
= desc
->timestamp
;
393 entry
->fattr
->gencount
= desc
->gencount
;
398 int nfs_same_file(struct dentry
*dentry
, struct nfs_entry
*entry
)
400 if (dentry
->d_inode
== NULL
)
402 if (nfs_compare_fh(entry
->fh
, NFS_FH(dentry
->d_inode
)) != 0)
410 bool nfs_use_readdirplus(struct inode
*dir
, struct dir_context
*ctx
)
412 if (!nfs_server_capable(dir
, NFS_CAP_READDIRPLUS
))
414 if (test_and_clear_bit(NFS_INO_ADVISE_RDPLUS
, &NFS_I(dir
)->flags
))
422 * This function is called by the lookup code to request the use of
423 * readdirplus to accelerate any future lookups in the same
427 void nfs_advise_use_readdirplus(struct inode
*dir
)
429 set_bit(NFS_INO_ADVISE_RDPLUS
, &NFS_I(dir
)->flags
);
433 void nfs_prime_dcache(struct dentry
*parent
, struct nfs_entry
*entry
)
435 struct qstr filename
= QSTR_INIT(entry
->name
, entry
->len
);
436 struct dentry
*dentry
;
437 struct dentry
*alias
;
438 struct inode
*dir
= parent
->d_inode
;
442 if (filename
.name
[0] == '.') {
443 if (filename
.len
== 1)
445 if (filename
.len
== 2 && filename
.name
[1] == '.')
448 filename
.hash
= full_name_hash(filename
.name
, filename
.len
);
450 dentry
= d_lookup(parent
, &filename
);
451 if (dentry
!= NULL
) {
452 if (nfs_same_file(dentry
, entry
)) {
453 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
454 status
= nfs_refresh_inode(dentry
->d_inode
, entry
->fattr
);
456 nfs_setsecurity(dentry
->d_inode
, entry
->fattr
, entry
->label
);
459 if (d_invalidate(dentry
) != 0)
465 dentry
= d_alloc(parent
, &filename
);
469 inode
= nfs_fhget(dentry
->d_sb
, entry
->fh
, entry
->fattr
, entry
->label
);
473 alias
= d_materialise_unique(dentry
, inode
);
477 nfs_set_verifier(alias
, nfs_save_change_attribute(dir
));
480 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
486 /* Perform conversion from xdr to cache array */
488 int nfs_readdir_page_filler(nfs_readdir_descriptor_t
*desc
, struct nfs_entry
*entry
,
489 struct page
**xdr_pages
, struct page
*page
, unsigned int buflen
)
491 struct xdr_stream stream
;
493 struct page
*scratch
;
494 struct nfs_cache_array
*array
;
495 unsigned int count
= 0;
498 scratch
= alloc_page(GFP_KERNEL
);
502 xdr_init_decode_pages(&stream
, &buf
, xdr_pages
, buflen
);
503 xdr_set_scratch_buffer(&stream
, page_address(scratch
), PAGE_SIZE
);
506 status
= xdr_decode(desc
, entry
, &stream
);
508 if (status
== -EAGAIN
)
516 nfs_prime_dcache(desc
->file
->f_path
.dentry
, entry
);
518 status
= nfs_readdir_add_to_array(entry
, page
);
521 } while (!entry
->eof
);
523 if (count
== 0 || (status
== -EBADCOOKIE
&& entry
->eof
!= 0)) {
524 array
= nfs_readdir_get_array(page
);
525 if (!IS_ERR(array
)) {
526 array
->eof_index
= array
->size
;
528 nfs_readdir_release_array(page
);
530 status
= PTR_ERR(array
);
538 void nfs_readdir_free_pagearray(struct page
**pages
, unsigned int npages
)
541 for (i
= 0; i
< npages
; i
++)
546 void nfs_readdir_free_large_page(void *ptr
, struct page
**pages
,
549 nfs_readdir_free_pagearray(pages
, npages
);
553 * nfs_readdir_large_page will allocate pages that must be freed with a call
554 * to nfs_readdir_free_large_page
557 int nfs_readdir_large_page(struct page
**pages
, unsigned int npages
)
561 for (i
= 0; i
< npages
; i
++) {
562 struct page
*page
= alloc_page(GFP_KERNEL
);
570 nfs_readdir_free_pagearray(pages
, i
);
575 int nfs_readdir_xdr_to_array(nfs_readdir_descriptor_t
*desc
, struct page
*page
, struct inode
*inode
)
577 struct page
*pages
[NFS_MAX_READDIR_PAGES
];
578 void *pages_ptr
= NULL
;
579 struct nfs_entry entry
;
580 struct file
*file
= desc
->file
;
581 struct nfs_cache_array
*array
;
582 int status
= -ENOMEM
;
583 unsigned int array_size
= ARRAY_SIZE(pages
);
585 entry
.prev_cookie
= 0;
586 entry
.cookie
= desc
->last_cookie
;
588 entry
.fh
= nfs_alloc_fhandle();
589 entry
.fattr
= nfs_alloc_fattr();
590 entry
.server
= NFS_SERVER(inode
);
591 if (entry
.fh
== NULL
|| entry
.fattr
== NULL
)
594 entry
.label
= nfs4_label_alloc(NFS_SERVER(inode
), GFP_NOWAIT
);
595 if (IS_ERR(entry
.label
)) {
596 status
= PTR_ERR(entry
.label
);
600 array
= nfs_readdir_get_array(page
);
602 status
= PTR_ERR(array
);
605 memset(array
, 0, sizeof(struct nfs_cache_array
));
606 array
->eof_index
= -1;
608 status
= nfs_readdir_large_page(pages
, array_size
);
610 goto out_release_array
;
613 status
= nfs_readdir_xdr_filler(pages
, desc
, &entry
, file
, inode
);
618 status
= nfs_readdir_page_filler(desc
, &entry
, pages
, page
, pglen
);
620 if (status
== -ENOSPC
)
624 } while (array
->eof_index
< 0);
626 nfs_readdir_free_large_page(pages_ptr
, pages
, array_size
);
628 nfs_readdir_release_array(page
);
630 nfs4_label_free(entry
.label
);
632 nfs_free_fattr(entry
.fattr
);
633 nfs_free_fhandle(entry
.fh
);
638 * Now we cache directories properly, by converting xdr information
639 * to an array that can be used for lookups later. This results in
640 * fewer cache pages, since we can store more information on each page.
641 * We only need to convert from xdr once so future lookups are much simpler
644 int nfs_readdir_filler(nfs_readdir_descriptor_t
*desc
, struct page
* page
)
646 struct inode
*inode
= file_inode(desc
->file
);
649 ret
= nfs_readdir_xdr_to_array(desc
, page
, inode
);
652 SetPageUptodate(page
);
654 if (invalidate_inode_pages2_range(inode
->i_mapping
, page
->index
+ 1, -1) < 0) {
655 /* Should never happen */
656 nfs_zap_mapping(inode
, inode
->i_mapping
);
666 void cache_page_release(nfs_readdir_descriptor_t
*desc
)
668 if (!desc
->page
->mapping
)
669 nfs_readdir_clear_array(desc
->page
);
670 page_cache_release(desc
->page
);
675 struct page
*get_cache_page(nfs_readdir_descriptor_t
*desc
)
677 return read_cache_page(file_inode(desc
->file
)->i_mapping
,
678 desc
->page_index
, (filler_t
*)nfs_readdir_filler
, desc
);
682 * Returns 0 if desc->dir_cookie was found on page desc->page_index
685 int find_cache_page(nfs_readdir_descriptor_t
*desc
)
689 desc
->page
= get_cache_page(desc
);
690 if (IS_ERR(desc
->page
))
691 return PTR_ERR(desc
->page
);
693 res
= nfs_readdir_search_array(desc
);
695 cache_page_release(desc
);
699 /* Search for desc->dir_cookie from the beginning of the page cache */
701 int readdir_search_pagecache(nfs_readdir_descriptor_t
*desc
)
705 if (desc
->page_index
== 0) {
706 desc
->current_index
= 0;
707 desc
->last_cookie
= 0;
710 res
= find_cache_page(desc
);
711 } while (res
== -EAGAIN
);
716 * Once we've found the start of the dirent within a page: fill 'er up...
719 int nfs_do_filldir(nfs_readdir_descriptor_t
*desc
)
721 struct file
*file
= desc
->file
;
724 struct nfs_cache_array
*array
= NULL
;
725 struct nfs_open_dir_context
*ctx
= file
->private_data
;
727 array
= nfs_readdir_get_array(desc
->page
);
729 res
= PTR_ERR(array
);
733 for (i
= desc
->cache_entry_index
; i
< array
->size
; i
++) {
734 struct nfs_cache_array_entry
*ent
;
736 ent
= &array
->array
[i
];
737 if (!dir_emit(desc
->ctx
, ent
->string
.name
, ent
->string
.len
,
738 nfs_compat_user_ino64(ent
->ino
), ent
->d_type
)) {
743 if (i
< (array
->size
-1))
744 *desc
->dir_cookie
= array
->array
[i
+1].cookie
;
746 *desc
->dir_cookie
= array
->last_cookie
;
750 if (array
->eof_index
>= 0)
753 nfs_readdir_release_array(desc
->page
);
755 cache_page_release(desc
);
756 dfprintk(DIRCACHE
, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n",
757 (unsigned long long)*desc
->dir_cookie
, res
);
762 * If we cannot find a cookie in our cache, we suspect that this is
763 * because it points to a deleted file, so we ask the server to return
764 * whatever it thinks is the next entry. We then feed this to filldir.
765 * If all goes well, we should then be able to find our way round the
766 * cache on the next call to readdir_search_pagecache();
768 * NOTE: we cannot add the anonymous page to the pagecache because
769 * the data it contains might not be page aligned. Besides,
770 * we should already have a complete representation of the
771 * directory in the page cache by the time we get here.
774 int uncached_readdir(nfs_readdir_descriptor_t
*desc
)
776 struct page
*page
= NULL
;
778 struct inode
*inode
= file_inode(desc
->file
);
779 struct nfs_open_dir_context
*ctx
= desc
->file
->private_data
;
781 dfprintk(DIRCACHE
, "NFS: uncached_readdir() searching for cookie %Lu\n",
782 (unsigned long long)*desc
->dir_cookie
);
784 page
= alloc_page(GFP_HIGHUSER
);
790 desc
->page_index
= 0;
791 desc
->last_cookie
= *desc
->dir_cookie
;
795 status
= nfs_readdir_xdr_to_array(desc
, page
, inode
);
799 status
= nfs_do_filldir(desc
);
802 dfprintk(DIRCACHE
, "NFS: %s: returns %d\n",
806 cache_page_release(desc
);
810 /* The file offset position represents the dirent entry number. A
811 last cookie cache takes care of the common case of reading the
814 static int nfs_readdir(struct file
*file
, struct dir_context
*ctx
)
816 struct dentry
*dentry
= file
->f_path
.dentry
;
817 struct inode
*inode
= dentry
->d_inode
;
818 nfs_readdir_descriptor_t my_desc
,
820 struct nfs_open_dir_context
*dir_ctx
= file
->private_data
;
823 dfprintk(FILE, "NFS: readdir(%s/%s) starting at cookie %llu\n",
824 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
,
825 (long long)ctx
->pos
);
826 nfs_inc_stats(inode
, NFSIOS_VFSGETDENTS
);
829 * ctx->pos points to the dirent entry number.
830 * *desc->dir_cookie has the cookie for the next entry. We have
831 * to either find the entry with the appropriate number or
832 * revalidate the cookie.
834 memset(desc
, 0, sizeof(*desc
));
838 desc
->dir_cookie
= &dir_ctx
->dir_cookie
;
839 desc
->decode
= NFS_PROTO(inode
)->decode_dirent
;
840 desc
->plus
= nfs_use_readdirplus(inode
, ctx
) ? 1 : 0;
842 nfs_block_sillyrename(dentry
);
843 if (ctx
->pos
== 0 || nfs_attribute_cache_expired(inode
))
844 res
= nfs_revalidate_mapping(inode
, file
->f_mapping
);
849 res
= readdir_search_pagecache(desc
);
851 if (res
== -EBADCOOKIE
) {
853 /* This means either end of directory */
854 if (*desc
->dir_cookie
&& desc
->eof
== 0) {
855 /* Or that the server has 'lost' a cookie */
856 res
= uncached_readdir(desc
);
862 if (res
== -ETOOSMALL
&& desc
->plus
) {
863 clear_bit(NFS_INO_ADVISE_RDPLUS
, &NFS_I(inode
)->flags
);
864 nfs_zap_caches(inode
);
865 desc
->page_index
= 0;
873 res
= nfs_do_filldir(desc
);
876 } while (!desc
->eof
);
878 nfs_unblock_sillyrename(dentry
);
881 dfprintk(FILE, "NFS: readdir(%s/%s) returns %d\n",
882 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
,
887 static loff_t
nfs_llseek_dir(struct file
*filp
, loff_t offset
, int whence
)
889 struct dentry
*dentry
= filp
->f_path
.dentry
;
890 struct inode
*inode
= dentry
->d_inode
;
891 struct nfs_open_dir_context
*dir_ctx
= filp
->private_data
;
893 dfprintk(FILE, "NFS: llseek dir(%s/%s, %lld, %d)\n",
894 dentry
->d_parent
->d_name
.name
,
898 mutex_lock(&inode
->i_mutex
);
901 offset
+= filp
->f_pos
;
909 if (offset
!= filp
->f_pos
) {
910 filp
->f_pos
= offset
;
911 dir_ctx
->dir_cookie
= 0;
915 mutex_unlock(&inode
->i_mutex
);
920 * All directory operations under NFS are synchronous, so fsync()
921 * is a dummy operation.
923 static int nfs_fsync_dir(struct file
*filp
, loff_t start
, loff_t end
,
926 struct dentry
*dentry
= filp
->f_path
.dentry
;
927 struct inode
*inode
= dentry
->d_inode
;
929 dfprintk(FILE, "NFS: fsync dir(%s/%s) datasync %d\n",
930 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
,
933 mutex_lock(&inode
->i_mutex
);
934 nfs_inc_stats(dentry
->d_inode
, NFSIOS_VFSFSYNC
);
935 mutex_unlock(&inode
->i_mutex
);
940 * nfs_force_lookup_revalidate - Mark the directory as having changed
941 * @dir - pointer to directory inode
943 * This forces the revalidation code in nfs_lookup_revalidate() to do a
944 * full lookup on all child dentries of 'dir' whenever a change occurs
945 * on the server that might have invalidated our dcache.
947 * The caller should be holding dir->i_lock
949 void nfs_force_lookup_revalidate(struct inode
*dir
)
951 NFS_I(dir
)->cache_change_attribute
++;
953 EXPORT_SYMBOL_GPL(nfs_force_lookup_revalidate
);
956 * A check for whether or not the parent directory has changed.
957 * In the case it has, we assume that the dentries are untrustworthy
958 * and may need to be looked up again.
960 static int nfs_check_verifier(struct inode
*dir
, struct dentry
*dentry
)
964 if (NFS_SERVER(dir
)->flags
& NFS_MOUNT_LOOKUP_CACHE_NONE
)
966 if (!nfs_verify_change_attribute(dir
, dentry
->d_time
))
968 /* Revalidate nfsi->cache_change_attribute before we declare a match */
969 if (nfs_revalidate_inode(NFS_SERVER(dir
), dir
) < 0)
971 if (!nfs_verify_change_attribute(dir
, dentry
->d_time
))
977 * Use intent information to check whether or not we're going to do
978 * an O_EXCL create using this path component.
980 static int nfs_is_exclusive_create(struct inode
*dir
, unsigned int flags
)
982 if (NFS_PROTO(dir
)->version
== 2)
984 return flags
& LOOKUP_EXCL
;
988 * Inode and filehandle revalidation for lookups.
990 * We force revalidation in the cases where the VFS sets LOOKUP_REVAL,
991 * or if the intent information indicates that we're about to open this
992 * particular file and the "nocto" mount flag is not set.
996 int nfs_lookup_verify_inode(struct inode
*inode
, unsigned int flags
)
998 struct nfs_server
*server
= NFS_SERVER(inode
);
1001 if (IS_AUTOMOUNT(inode
))
1003 /* VFS wants an on-the-wire revalidation */
1004 if (flags
& LOOKUP_REVAL
)
1006 /* This is an open(2) */
1007 if ((flags
& LOOKUP_OPEN
) && !(server
->flags
& NFS_MOUNT_NOCTO
) &&
1008 (S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
)))
1011 return (inode
->i_nlink
== 0) ? -ENOENT
: 0;
1013 ret
= __nfs_revalidate_inode(server
, inode
);
1020 * We judge how long we want to trust negative
1021 * dentries by looking at the parent inode mtime.
1023 * If parent mtime has changed, we revalidate, else we wait for a
1024 * period corresponding to the parent's attribute cache timeout value.
1027 int nfs_neg_need_reval(struct inode
*dir
, struct dentry
*dentry
,
1030 /* Don't revalidate a negative dentry if we're creating a new file */
1031 if (flags
& LOOKUP_CREATE
)
1033 if (NFS_SERVER(dir
)->flags
& NFS_MOUNT_LOOKUP_CACHE_NONEG
)
1035 return !nfs_check_verifier(dir
, dentry
);
1039 * This is called every time the dcache has a lookup hit,
1040 * and we should check whether we can really trust that
1043 * NOTE! The hit can be a negative hit too, don't assume
1046 * If the parent directory is seen to have changed, we throw out the
1047 * cached dentry and do a new lookup.
1049 static int nfs_lookup_revalidate(struct dentry
*dentry
, unsigned int flags
)
1052 struct inode
*inode
;
1053 struct dentry
*parent
;
1054 struct nfs_fh
*fhandle
= NULL
;
1055 struct nfs_fattr
*fattr
= NULL
;
1056 struct nfs4_label
*label
= NULL
;
1059 if (flags
& LOOKUP_RCU
)
1062 parent
= dget_parent(dentry
);
1063 dir
= parent
->d_inode
;
1064 nfs_inc_stats(dir
, NFSIOS_DENTRYREVALIDATE
);
1065 inode
= dentry
->d_inode
;
1068 if (nfs_neg_need_reval(dir
, dentry
, flags
))
1070 goto out_valid_noent
;
1073 if (is_bad_inode(inode
)) {
1074 dfprintk(LOOKUPCACHE
, "%s: %s/%s has dud inode\n",
1075 __func__
, dentry
->d_parent
->d_name
.name
,
1076 dentry
->d_name
.name
);
1080 if (NFS_PROTO(dir
)->have_delegation(inode
, FMODE_READ
))
1081 goto out_set_verifier
;
1083 /* Force a full look up iff the parent directory has changed */
1084 if (!nfs_is_exclusive_create(dir
, flags
) && nfs_check_verifier(dir
, dentry
)) {
1085 if (nfs_lookup_verify_inode(inode
, flags
))
1086 goto out_zap_parent
;
1090 if (NFS_STALE(inode
))
1094 fhandle
= nfs_alloc_fhandle();
1095 fattr
= nfs_alloc_fattr();
1096 if (fhandle
== NULL
|| fattr
== NULL
)
1099 label
= nfs4_label_alloc(NFS_SERVER(inode
), GFP_NOWAIT
);
1103 error
= NFS_PROTO(dir
)->lookup(dir
, &dentry
->d_name
, fhandle
, fattr
, label
);
1106 if (nfs_compare_fh(NFS_FH(inode
), fhandle
))
1108 if ((error
= nfs_refresh_inode(inode
, fattr
)) != 0)
1111 nfs_setsecurity(inode
, fattr
, label
);
1113 nfs_free_fattr(fattr
);
1114 nfs_free_fhandle(fhandle
);
1115 nfs4_label_free(label
);
1118 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1120 /* Success: notify readdir to use READDIRPLUS */
1121 nfs_advise_use_readdirplus(dir
);
1124 dfprintk(LOOKUPCACHE
, "NFS: %s(%s/%s) is valid\n",
1125 __func__
, dentry
->d_parent
->d_name
.name
,
1126 dentry
->d_name
.name
);
1129 nfs_zap_caches(dir
);
1131 nfs_free_fattr(fattr
);
1132 nfs_free_fhandle(fhandle
);
1133 nfs4_label_free(label
);
1134 nfs_mark_for_revalidate(dir
);
1135 if (inode
&& S_ISDIR(inode
->i_mode
)) {
1136 /* Purge readdir caches. */
1137 nfs_zap_caches(inode
);
1138 /* If we have submounts, don't unhash ! */
1139 if (have_submounts(dentry
))
1141 if (dentry
->d_flags
& DCACHE_DISCONNECTED
)
1143 shrink_dcache_parent(dentry
);
1147 dfprintk(LOOKUPCACHE
, "NFS: %s(%s/%s) is invalid\n",
1148 __func__
, dentry
->d_parent
->d_name
.name
,
1149 dentry
->d_name
.name
);
1152 nfs_free_fattr(fattr
);
1153 nfs_free_fhandle(fhandle
);
1154 nfs4_label_free(label
);
1156 dfprintk(LOOKUPCACHE
, "NFS: %s(%s/%s) lookup returned error %d\n",
1157 __func__
, dentry
->d_parent
->d_name
.name
,
1158 dentry
->d_name
.name
, error
);
1163 * A weaker form of d_revalidate for revalidating just the dentry->d_inode
1164 * when we don't really care about the dentry name. This is called when a
1165 * pathwalk ends on a dentry that was not found via a normal lookup in the
1166 * parent dir (e.g.: ".", "..", procfs symlinks or mountpoint traversals).
1168 * In this situation, we just want to verify that the inode itself is OK
1169 * since the dentry might have changed on the server.
1171 static int nfs_weak_revalidate(struct dentry
*dentry
, unsigned int flags
)
1174 struct inode
*inode
= dentry
->d_inode
;
1177 * I believe we can only get a negative dentry here in the case of a
1178 * procfs-style symlink. Just assume it's correct for now, but we may
1179 * eventually need to do something more here.
1182 dfprintk(LOOKUPCACHE
, "%s: %s/%s has negative inode\n",
1183 __func__
, dentry
->d_parent
->d_name
.name
,
1184 dentry
->d_name
.name
);
1188 if (is_bad_inode(inode
)) {
1189 dfprintk(LOOKUPCACHE
, "%s: %s/%s has dud inode\n",
1190 __func__
, dentry
->d_parent
->d_name
.name
,
1191 dentry
->d_name
.name
);
1195 error
= nfs_revalidate_inode(NFS_SERVER(inode
), inode
);
1196 dfprintk(LOOKUPCACHE
, "NFS: %s: inode %lu is %s\n",
1197 __func__
, inode
->i_ino
, error
? "invalid" : "valid");
1202 * This is called from dput() when d_count is going to 0.
1204 static int nfs_dentry_delete(const struct dentry
*dentry
)
1206 dfprintk(VFS
, "NFS: dentry_delete(%s/%s, %x)\n",
1207 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
,
1210 /* Unhash any dentry with a stale inode */
1211 if (dentry
->d_inode
!= NULL
&& NFS_STALE(dentry
->d_inode
))
1214 if (dentry
->d_flags
& DCACHE_NFSFS_RENAMED
) {
1215 /* Unhash it, so that ->d_iput() would be called */
1218 if (!(dentry
->d_sb
->s_flags
& MS_ACTIVE
)) {
1219 /* Unhash it, so that ancestors of killed async unlink
1220 * files will be cleaned up during umount */
1227 /* Ensure that we revalidate inode->i_nlink */
1228 static void nfs_drop_nlink(struct inode
*inode
)
1230 spin_lock(&inode
->i_lock
);
1231 /* drop the inode if we're reasonably sure this is the last link */
1232 if (inode
->i_nlink
== 1)
1234 NFS_I(inode
)->cache_validity
|= NFS_INO_INVALID_ATTR
;
1235 spin_unlock(&inode
->i_lock
);
1239 * Called when the dentry loses inode.
1240 * We use it to clean up silly-renamed files.
1242 static void nfs_dentry_iput(struct dentry
*dentry
, struct inode
*inode
)
1244 if (S_ISDIR(inode
->i_mode
))
1245 /* drop any readdir cache as it could easily be old */
1246 NFS_I(inode
)->cache_validity
|= NFS_INO_INVALID_DATA
;
1248 if (dentry
->d_flags
& DCACHE_NFSFS_RENAMED
) {
1249 nfs_complete_unlink(dentry
, inode
);
1250 nfs_drop_nlink(inode
);
1255 static void nfs_d_release(struct dentry
*dentry
)
1257 /* free cached devname value, if it survived that far */
1258 if (unlikely(dentry
->d_fsdata
)) {
1259 if (dentry
->d_flags
& DCACHE_NFSFS_RENAMED
)
1262 kfree(dentry
->d_fsdata
);
1266 const struct dentry_operations nfs_dentry_operations
= {
1267 .d_revalidate
= nfs_lookup_revalidate
,
1268 .d_weak_revalidate
= nfs_weak_revalidate
,
1269 .d_delete
= nfs_dentry_delete
,
1270 .d_iput
= nfs_dentry_iput
,
1271 .d_automount
= nfs_d_automount
,
1272 .d_release
= nfs_d_release
,
1274 EXPORT_SYMBOL_GPL(nfs_dentry_operations
);
1276 struct dentry
*nfs_lookup(struct inode
*dir
, struct dentry
* dentry
, unsigned int flags
)
1279 struct dentry
*parent
;
1280 struct inode
*inode
= NULL
;
1281 struct nfs_fh
*fhandle
= NULL
;
1282 struct nfs_fattr
*fattr
= NULL
;
1283 struct nfs4_label
*label
= NULL
;
1286 dfprintk(VFS
, "NFS: lookup(%s/%s)\n",
1287 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
);
1288 nfs_inc_stats(dir
, NFSIOS_VFSLOOKUP
);
1290 res
= ERR_PTR(-ENAMETOOLONG
);
1291 if (dentry
->d_name
.len
> NFS_SERVER(dir
)->namelen
)
1295 * If we're doing an exclusive create, optimize away the lookup
1296 * but don't hash the dentry.
1298 if (nfs_is_exclusive_create(dir
, flags
)) {
1299 d_instantiate(dentry
, NULL
);
1304 res
= ERR_PTR(-ENOMEM
);
1305 fhandle
= nfs_alloc_fhandle();
1306 fattr
= nfs_alloc_fattr();
1307 if (fhandle
== NULL
|| fattr
== NULL
)
1310 label
= nfs4_label_alloc(NFS_SERVER(dir
), GFP_NOWAIT
);
1314 parent
= dentry
->d_parent
;
1315 /* Protect against concurrent sillydeletes */
1316 nfs_block_sillyrename(parent
);
1317 error
= NFS_PROTO(dir
)->lookup(dir
, &dentry
->d_name
, fhandle
, fattr
, label
);
1318 if (error
== -ENOENT
)
1321 res
= ERR_PTR(error
);
1322 goto out_unblock_sillyrename
;
1324 inode
= nfs_fhget(dentry
->d_sb
, fhandle
, fattr
, label
);
1325 res
= ERR_CAST(inode
);
1327 goto out_unblock_sillyrename
;
1329 /* Success: notify readdir to use READDIRPLUS */
1330 nfs_advise_use_readdirplus(dir
);
1333 res
= d_materialise_unique(dentry
, inode
);
1336 goto out_unblock_sillyrename
;
1339 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1340 out_unblock_sillyrename
:
1341 nfs_unblock_sillyrename(parent
);
1342 nfs4_label_free(label
);
1344 nfs_free_fattr(fattr
);
1345 nfs_free_fhandle(fhandle
);
1348 EXPORT_SYMBOL_GPL(nfs_lookup
);
1350 #if IS_ENABLED(CONFIG_NFS_V4)
1351 static int nfs4_lookup_revalidate(struct dentry
*, unsigned int);
1353 const struct dentry_operations nfs4_dentry_operations
= {
1354 .d_revalidate
= nfs4_lookup_revalidate
,
1355 .d_delete
= nfs_dentry_delete
,
1356 .d_iput
= nfs_dentry_iput
,
1357 .d_automount
= nfs_d_automount
,
1358 .d_release
= nfs_d_release
,
1360 EXPORT_SYMBOL_GPL(nfs4_dentry_operations
);
1362 static fmode_t
flags_to_mode(int flags
)
1364 fmode_t res
= (__force fmode_t
)flags
& FMODE_EXEC
;
1365 if ((flags
& O_ACCMODE
) != O_WRONLY
)
1367 if ((flags
& O_ACCMODE
) != O_RDONLY
)
1372 static struct nfs_open_context
*create_nfs_open_context(struct dentry
*dentry
, int open_flags
)
1374 return alloc_nfs_open_context(dentry
, flags_to_mode(open_flags
));
1377 static int do_open(struct inode
*inode
, struct file
*filp
)
1379 nfs_fscache_set_inode_cookie(inode
, filp
);
1383 static int nfs_finish_open(struct nfs_open_context
*ctx
,
1384 struct dentry
*dentry
,
1385 struct file
*file
, unsigned open_flags
,
1390 err
= finish_open(file
, dentry
, do_open
, opened
);
1393 nfs_file_set_open_context(file
, ctx
);
1396 put_nfs_open_context(ctx
);
1400 int nfs_atomic_open(struct inode
*dir
, struct dentry
*dentry
,
1401 struct file
*file
, unsigned open_flags
,
1402 umode_t mode
, int *opened
)
1404 struct nfs_open_context
*ctx
;
1406 struct iattr attr
= { .ia_valid
= ATTR_OPEN
};
1407 struct inode
*inode
;
1410 /* Expect a negative dentry */
1411 BUG_ON(dentry
->d_inode
);
1413 dfprintk(VFS
, "NFS: atomic_open(%s/%ld), %s\n",
1414 dir
->i_sb
->s_id
, dir
->i_ino
, dentry
->d_name
.name
);
1416 /* NFS only supports OPEN on regular files */
1417 if ((open_flags
& O_DIRECTORY
)) {
1418 if (!d_unhashed(dentry
)) {
1420 * Hashed negative dentry with O_DIRECTORY: dentry was
1421 * revalidated and is fine, no need to perform lookup
1429 if (dentry
->d_name
.len
> NFS_SERVER(dir
)->namelen
)
1430 return -ENAMETOOLONG
;
1432 if (open_flags
& O_CREAT
) {
1433 attr
.ia_valid
|= ATTR_MODE
;
1434 attr
.ia_mode
= mode
& ~current_umask();
1436 if (open_flags
& O_TRUNC
) {
1437 attr
.ia_valid
|= ATTR_SIZE
;
1441 ctx
= create_nfs_open_context(dentry
, open_flags
);
1446 nfs_block_sillyrename(dentry
->d_parent
);
1447 inode
= NFS_PROTO(dir
)->open_context(dir
, ctx
, open_flags
, &attr
);
1448 nfs_unblock_sillyrename(dentry
->d_parent
);
1449 if (IS_ERR(inode
)) {
1450 put_nfs_open_context(ctx
);
1451 err
= PTR_ERR(inode
);
1455 d_add(dentry
, NULL
);
1461 if (!(open_flags
& O_NOFOLLOW
))
1471 err
= nfs_finish_open(ctx
, ctx
->dentry
, file
, open_flags
, opened
);
1476 res
= nfs_lookup(dir
, dentry
, 0);
1481 return finish_no_open(file
, res
);
1483 EXPORT_SYMBOL_GPL(nfs_atomic_open
);
1485 static int nfs4_lookup_revalidate(struct dentry
*dentry
, unsigned int flags
)
1487 struct dentry
*parent
= NULL
;
1488 struct inode
*inode
;
1492 if (flags
& LOOKUP_RCU
)
1495 if (!(flags
& LOOKUP_OPEN
) || (flags
& LOOKUP_DIRECTORY
))
1497 if (d_mountpoint(dentry
))
1499 if (NFS_SB(dentry
->d_sb
)->caps
& NFS_CAP_ATOMIC_OPEN_V1
)
1502 inode
= dentry
->d_inode
;
1503 parent
= dget_parent(dentry
);
1504 dir
= parent
->d_inode
;
1506 /* We can't create new files in nfs_open_revalidate(), so we
1507 * optimize away revalidation of negative dentries.
1509 if (inode
== NULL
) {
1510 if (!nfs_neg_need_reval(dir
, dentry
, flags
))
1515 /* NFS only supports OPEN on regular files */
1516 if (!S_ISREG(inode
->i_mode
))
1518 /* We cannot do exclusive creation on a positive dentry */
1519 if (flags
& LOOKUP_EXCL
)
1522 /* Let f_op->open() actually open (and revalidate) the file */
1532 return nfs_lookup_revalidate(dentry
, flags
);
1535 #endif /* CONFIG_NFSV4 */
1538 * Code common to create, mkdir, and mknod.
1540 int nfs_instantiate(struct dentry
*dentry
, struct nfs_fh
*fhandle
,
1541 struct nfs_fattr
*fattr
,
1542 struct nfs4_label
*label
)
1544 struct dentry
*parent
= dget_parent(dentry
);
1545 struct inode
*dir
= parent
->d_inode
;
1546 struct inode
*inode
;
1547 int error
= -EACCES
;
1551 /* We may have been initialized further down */
1552 if (dentry
->d_inode
)
1554 if (fhandle
->size
== 0) {
1555 error
= NFS_PROTO(dir
)->lookup(dir
, &dentry
->d_name
, fhandle
, fattr
, NULL
);
1559 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1560 if (!(fattr
->valid
& NFS_ATTR_FATTR
)) {
1561 struct nfs_server
*server
= NFS_SB(dentry
->d_sb
);
1562 error
= server
->nfs_client
->rpc_ops
->getattr(server
, fhandle
, fattr
, NULL
);
1566 inode
= nfs_fhget(dentry
->d_sb
, fhandle
, fattr
, label
);
1567 error
= PTR_ERR(inode
);
1570 d_add(dentry
, inode
);
1575 nfs_mark_for_revalidate(dir
);
1579 EXPORT_SYMBOL_GPL(nfs_instantiate
);
1582 * Following a failed create operation, we drop the dentry rather
1583 * than retain a negative dentry. This avoids a problem in the event
1584 * that the operation succeeded on the server, but an error in the
1585 * reply path made it appear to have failed.
1587 int nfs_create(struct inode
*dir
, struct dentry
*dentry
,
1588 umode_t mode
, bool excl
)
1591 int open_flags
= excl
? O_CREAT
| O_EXCL
: O_CREAT
;
1594 dfprintk(VFS
, "NFS: create(%s/%ld), %s\n",
1595 dir
->i_sb
->s_id
, dir
->i_ino
, dentry
->d_name
.name
);
1597 attr
.ia_mode
= mode
;
1598 attr
.ia_valid
= ATTR_MODE
;
1600 error
= NFS_PROTO(dir
)->create(dir
, dentry
, &attr
, open_flags
);
1608 EXPORT_SYMBOL_GPL(nfs_create
);
1611 * See comments for nfs_proc_create regarding failed operations.
1614 nfs_mknod(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
, dev_t rdev
)
1619 dfprintk(VFS
, "NFS: mknod(%s/%ld), %s\n",
1620 dir
->i_sb
->s_id
, dir
->i_ino
, dentry
->d_name
.name
);
1622 if (!new_valid_dev(rdev
))
1625 attr
.ia_mode
= mode
;
1626 attr
.ia_valid
= ATTR_MODE
;
1628 status
= NFS_PROTO(dir
)->mknod(dir
, dentry
, &attr
, rdev
);
1636 EXPORT_SYMBOL_GPL(nfs_mknod
);
1639 * See comments for nfs_proc_create regarding failed operations.
1641 int nfs_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
1646 dfprintk(VFS
, "NFS: mkdir(%s/%ld), %s\n",
1647 dir
->i_sb
->s_id
, dir
->i_ino
, dentry
->d_name
.name
);
1649 attr
.ia_valid
= ATTR_MODE
;
1650 attr
.ia_mode
= mode
| S_IFDIR
;
1652 error
= NFS_PROTO(dir
)->mkdir(dir
, dentry
, &attr
);
1660 EXPORT_SYMBOL_GPL(nfs_mkdir
);
1662 static void nfs_dentry_handle_enoent(struct dentry
*dentry
)
1664 if (dentry
->d_inode
!= NULL
&& !d_unhashed(dentry
))
1668 int nfs_rmdir(struct inode
*dir
, struct dentry
*dentry
)
1672 dfprintk(VFS
, "NFS: rmdir(%s/%ld), %s\n",
1673 dir
->i_sb
->s_id
, dir
->i_ino
, dentry
->d_name
.name
);
1675 error
= NFS_PROTO(dir
)->rmdir(dir
, &dentry
->d_name
);
1676 /* Ensure the VFS deletes this inode */
1677 if (error
== 0 && dentry
->d_inode
!= NULL
)
1678 clear_nlink(dentry
->d_inode
);
1679 else if (error
== -ENOENT
)
1680 nfs_dentry_handle_enoent(dentry
);
1684 EXPORT_SYMBOL_GPL(nfs_rmdir
);
1687 * Remove a file after making sure there are no pending writes,
1688 * and after checking that the file has only one user.
1690 * We invalidate the attribute cache and free the inode prior to the operation
1691 * to avoid possible races if the server reuses the inode.
1693 static int nfs_safe_remove(struct dentry
*dentry
)
1695 struct inode
*dir
= dentry
->d_parent
->d_inode
;
1696 struct inode
*inode
= dentry
->d_inode
;
1699 dfprintk(VFS
, "NFS: safe_remove(%s/%s)\n",
1700 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
);
1702 /* If the dentry was sillyrenamed, we simply call d_delete() */
1703 if (dentry
->d_flags
& DCACHE_NFSFS_RENAMED
) {
1708 if (inode
!= NULL
) {
1709 NFS_PROTO(inode
)->return_delegation(inode
);
1710 error
= NFS_PROTO(dir
)->remove(dir
, &dentry
->d_name
);
1712 nfs_drop_nlink(inode
);
1714 error
= NFS_PROTO(dir
)->remove(dir
, &dentry
->d_name
);
1715 if (error
== -ENOENT
)
1716 nfs_dentry_handle_enoent(dentry
);
1721 /* We do silly rename. In case sillyrename() returns -EBUSY, the inode
1722 * belongs to an active ".nfs..." file and we return -EBUSY.
1724 * If sillyrename() returns 0, we do nothing, otherwise we unlink.
1726 int nfs_unlink(struct inode
*dir
, struct dentry
*dentry
)
1729 int need_rehash
= 0;
1731 dfprintk(VFS
, "NFS: unlink(%s/%ld, %s)\n", dir
->i_sb
->s_id
,
1732 dir
->i_ino
, dentry
->d_name
.name
);
1734 spin_lock(&dentry
->d_lock
);
1735 if (d_count(dentry
) > 1) {
1736 spin_unlock(&dentry
->d_lock
);
1737 /* Start asynchronous writeout of the inode */
1738 write_inode_now(dentry
->d_inode
, 0);
1739 error
= nfs_sillyrename(dir
, dentry
);
1742 if (!d_unhashed(dentry
)) {
1746 spin_unlock(&dentry
->d_lock
);
1747 error
= nfs_safe_remove(dentry
);
1748 if (!error
|| error
== -ENOENT
) {
1749 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1750 } else if (need_rehash
)
1754 EXPORT_SYMBOL_GPL(nfs_unlink
);
1757 * To create a symbolic link, most file systems instantiate a new inode,
1758 * add a page to it containing the path, then write it out to the disk
1759 * using prepare_write/commit_write.
1761 * Unfortunately the NFS client can't create the in-core inode first
1762 * because it needs a file handle to create an in-core inode (see
1763 * fs/nfs/inode.c:nfs_fhget). We only have a file handle *after* the
1764 * symlink request has completed on the server.
1766 * So instead we allocate a raw page, copy the symname into it, then do
1767 * the SYMLINK request with the page as the buffer. If it succeeds, we
1768 * now have a new file handle and can instantiate an in-core NFS inode
1769 * and move the raw page into its mapping.
1771 int nfs_symlink(struct inode
*dir
, struct dentry
*dentry
, const char *symname
)
1776 unsigned int pathlen
= strlen(symname
);
1779 dfprintk(VFS
, "NFS: symlink(%s/%ld, %s, %s)\n", dir
->i_sb
->s_id
,
1780 dir
->i_ino
, dentry
->d_name
.name
, symname
);
1782 if (pathlen
> PAGE_SIZE
)
1783 return -ENAMETOOLONG
;
1785 attr
.ia_mode
= S_IFLNK
| S_IRWXUGO
;
1786 attr
.ia_valid
= ATTR_MODE
;
1788 page
= alloc_page(GFP_HIGHUSER
);
1792 kaddr
= kmap_atomic(page
);
1793 memcpy(kaddr
, symname
, pathlen
);
1794 if (pathlen
< PAGE_SIZE
)
1795 memset(kaddr
+ pathlen
, 0, PAGE_SIZE
- pathlen
);
1796 kunmap_atomic(kaddr
);
1798 error
= NFS_PROTO(dir
)->symlink(dir
, dentry
, page
, pathlen
, &attr
);
1800 dfprintk(VFS
, "NFS: symlink(%s/%ld, %s, %s) error %d\n",
1801 dir
->i_sb
->s_id
, dir
->i_ino
,
1802 dentry
->d_name
.name
, symname
, error
);
1809 * No big deal if we can't add this page to the page cache here.
1810 * READLINK will get the missing page from the server if needed.
1812 if (!add_to_page_cache_lru(page
, dentry
->d_inode
->i_mapping
, 0,
1814 SetPageUptodate(page
);
1821 EXPORT_SYMBOL_GPL(nfs_symlink
);
1824 nfs_link(struct dentry
*old_dentry
, struct inode
*dir
, struct dentry
*dentry
)
1826 struct inode
*inode
= old_dentry
->d_inode
;
1829 dfprintk(VFS
, "NFS: link(%s/%s -> %s/%s)\n",
1830 old_dentry
->d_parent
->d_name
.name
, old_dentry
->d_name
.name
,
1831 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
);
1833 NFS_PROTO(inode
)->return_delegation(inode
);
1836 error
= NFS_PROTO(dir
)->link(inode
, dir
, &dentry
->d_name
);
1839 d_add(dentry
, inode
);
1843 EXPORT_SYMBOL_GPL(nfs_link
);
1847 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a
1848 * different file handle for the same inode after a rename (e.g. when
1849 * moving to a different directory). A fail-safe method to do so would
1850 * be to look up old_dir/old_name, create a link to new_dir/new_name and
1851 * rename the old file using the sillyrename stuff. This way, the original
1852 * file in old_dir will go away when the last process iput()s the inode.
1856 * It actually works quite well. One needs to have the possibility for
1857 * at least one ".nfs..." file in each directory the file ever gets
1858 * moved or linked to which happens automagically with the new
1859 * implementation that only depends on the dcache stuff instead of
1860 * using the inode layer
1862 * Unfortunately, things are a little more complicated than indicated
1863 * above. For a cross-directory move, we want to make sure we can get
1864 * rid of the old inode after the operation. This means there must be
1865 * no pending writes (if it's a file), and the use count must be 1.
1866 * If these conditions are met, we can drop the dentries before doing
1869 int nfs_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
1870 struct inode
*new_dir
, struct dentry
*new_dentry
)
1872 struct inode
*old_inode
= old_dentry
->d_inode
;
1873 struct inode
*new_inode
= new_dentry
->d_inode
;
1874 struct dentry
*dentry
= NULL
, *rehash
= NULL
;
1877 dfprintk(VFS
, "NFS: rename(%s/%s -> %s/%s, ct=%d)\n",
1878 old_dentry
->d_parent
->d_name
.name
, old_dentry
->d_name
.name
,
1879 new_dentry
->d_parent
->d_name
.name
, new_dentry
->d_name
.name
,
1880 d_count(new_dentry
));
1883 * For non-directories, check whether the target is busy and if so,
1884 * make a copy of the dentry and then do a silly-rename. If the
1885 * silly-rename succeeds, the copied dentry is hashed and becomes
1888 if (new_inode
&& !S_ISDIR(new_inode
->i_mode
)) {
1890 * To prevent any new references to the target during the
1891 * rename, we unhash the dentry in advance.
1893 if (!d_unhashed(new_dentry
)) {
1895 rehash
= new_dentry
;
1898 if (d_count(new_dentry
) > 2) {
1901 /* copy the target dentry's name */
1902 dentry
= d_alloc(new_dentry
->d_parent
,
1903 &new_dentry
->d_name
);
1907 /* silly-rename the existing target ... */
1908 err
= nfs_sillyrename(new_dir
, new_dentry
);
1912 new_dentry
= dentry
;
1918 NFS_PROTO(old_inode
)->return_delegation(old_inode
);
1919 if (new_inode
!= NULL
)
1920 NFS_PROTO(new_inode
)->return_delegation(new_inode
);
1922 error
= NFS_PROTO(old_dir
)->rename(old_dir
, &old_dentry
->d_name
,
1923 new_dir
, &new_dentry
->d_name
);
1924 nfs_mark_for_revalidate(old_inode
);
1929 if (new_inode
!= NULL
)
1930 nfs_drop_nlink(new_inode
);
1931 d_move(old_dentry
, new_dentry
);
1932 nfs_set_verifier(new_dentry
,
1933 nfs_save_change_attribute(new_dir
));
1934 } else if (error
== -ENOENT
)
1935 nfs_dentry_handle_enoent(old_dentry
);
1937 /* new dentry created? */
1942 EXPORT_SYMBOL_GPL(nfs_rename
);
1944 static DEFINE_SPINLOCK(nfs_access_lru_lock
);
1945 static LIST_HEAD(nfs_access_lru_list
);
1946 static atomic_long_t nfs_access_nr_entries
;
1948 static void nfs_access_free_entry(struct nfs_access_entry
*entry
)
1950 put_rpccred(entry
->cred
);
1952 smp_mb__before_atomic_dec();
1953 atomic_long_dec(&nfs_access_nr_entries
);
1954 smp_mb__after_atomic_dec();
1957 static void nfs_access_free_list(struct list_head
*head
)
1959 struct nfs_access_entry
*cache
;
1961 while (!list_empty(head
)) {
1962 cache
= list_entry(head
->next
, struct nfs_access_entry
, lru
);
1963 list_del(&cache
->lru
);
1964 nfs_access_free_entry(cache
);
1968 int nfs_access_cache_shrinker(struct shrinker
*shrink
,
1969 struct shrink_control
*sc
)
1972 struct nfs_inode
*nfsi
, *next
;
1973 struct nfs_access_entry
*cache
;
1974 int nr_to_scan
= sc
->nr_to_scan
;
1975 gfp_t gfp_mask
= sc
->gfp_mask
;
1977 if ((gfp_mask
& GFP_KERNEL
) != GFP_KERNEL
)
1978 return (nr_to_scan
== 0) ? 0 : -1;
1980 spin_lock(&nfs_access_lru_lock
);
1981 list_for_each_entry_safe(nfsi
, next
, &nfs_access_lru_list
, access_cache_inode_lru
) {
1982 struct inode
*inode
;
1984 if (nr_to_scan
-- == 0)
1986 inode
= &nfsi
->vfs_inode
;
1987 spin_lock(&inode
->i_lock
);
1988 if (list_empty(&nfsi
->access_cache_entry_lru
))
1989 goto remove_lru_entry
;
1990 cache
= list_entry(nfsi
->access_cache_entry_lru
.next
,
1991 struct nfs_access_entry
, lru
);
1992 list_move(&cache
->lru
, &head
);
1993 rb_erase(&cache
->rb_node
, &nfsi
->access_cache
);
1994 if (!list_empty(&nfsi
->access_cache_entry_lru
))
1995 list_move_tail(&nfsi
->access_cache_inode_lru
,
1996 &nfs_access_lru_list
);
1999 list_del_init(&nfsi
->access_cache_inode_lru
);
2000 smp_mb__before_clear_bit();
2001 clear_bit(NFS_INO_ACL_LRU_SET
, &nfsi
->flags
);
2002 smp_mb__after_clear_bit();
2004 spin_unlock(&inode
->i_lock
);
2006 spin_unlock(&nfs_access_lru_lock
);
2007 nfs_access_free_list(&head
);
2008 return (atomic_long_read(&nfs_access_nr_entries
) / 100) * sysctl_vfs_cache_pressure
;
2011 static void __nfs_access_zap_cache(struct nfs_inode
*nfsi
, struct list_head
*head
)
2013 struct rb_root
*root_node
= &nfsi
->access_cache
;
2015 struct nfs_access_entry
*entry
;
2017 /* Unhook entries from the cache */
2018 while ((n
= rb_first(root_node
)) != NULL
) {
2019 entry
= rb_entry(n
, struct nfs_access_entry
, rb_node
);
2020 rb_erase(n
, root_node
);
2021 list_move(&entry
->lru
, head
);
2023 nfsi
->cache_validity
&= ~NFS_INO_INVALID_ACCESS
;
2026 void nfs_access_zap_cache(struct inode
*inode
)
2030 if (test_bit(NFS_INO_ACL_LRU_SET
, &NFS_I(inode
)->flags
) == 0)
2032 /* Remove from global LRU init */
2033 spin_lock(&nfs_access_lru_lock
);
2034 if (test_and_clear_bit(NFS_INO_ACL_LRU_SET
, &NFS_I(inode
)->flags
))
2035 list_del_init(&NFS_I(inode
)->access_cache_inode_lru
);
2037 spin_lock(&inode
->i_lock
);
2038 __nfs_access_zap_cache(NFS_I(inode
), &head
);
2039 spin_unlock(&inode
->i_lock
);
2040 spin_unlock(&nfs_access_lru_lock
);
2041 nfs_access_free_list(&head
);
2043 EXPORT_SYMBOL_GPL(nfs_access_zap_cache
);
2045 static struct nfs_access_entry
*nfs_access_search_rbtree(struct inode
*inode
, struct rpc_cred
*cred
)
2047 struct rb_node
*n
= NFS_I(inode
)->access_cache
.rb_node
;
2048 struct nfs_access_entry
*entry
;
2051 entry
= rb_entry(n
, struct nfs_access_entry
, rb_node
);
2053 if (cred
< entry
->cred
)
2055 else if (cred
> entry
->cred
)
2063 static int nfs_access_get_cached(struct inode
*inode
, struct rpc_cred
*cred
, struct nfs_access_entry
*res
)
2065 struct nfs_inode
*nfsi
= NFS_I(inode
);
2066 struct nfs_access_entry
*cache
;
2069 spin_lock(&inode
->i_lock
);
2070 if (nfsi
->cache_validity
& NFS_INO_INVALID_ACCESS
)
2072 cache
= nfs_access_search_rbtree(inode
, cred
);
2075 if (!nfs_have_delegated_attributes(inode
) &&
2076 !time_in_range_open(jiffies
, cache
->jiffies
, cache
->jiffies
+ nfsi
->attrtimeo
))
2078 res
->jiffies
= cache
->jiffies
;
2079 res
->cred
= cache
->cred
;
2080 res
->mask
= cache
->mask
;
2081 list_move_tail(&cache
->lru
, &nfsi
->access_cache_entry_lru
);
2084 spin_unlock(&inode
->i_lock
);
2087 rb_erase(&cache
->rb_node
, &nfsi
->access_cache
);
2088 list_del(&cache
->lru
);
2089 spin_unlock(&inode
->i_lock
);
2090 nfs_access_free_entry(cache
);
2093 spin_unlock(&inode
->i_lock
);
2094 nfs_access_zap_cache(inode
);
2098 static void nfs_access_add_rbtree(struct inode
*inode
, struct nfs_access_entry
*set
)
2100 struct nfs_inode
*nfsi
= NFS_I(inode
);
2101 struct rb_root
*root_node
= &nfsi
->access_cache
;
2102 struct rb_node
**p
= &root_node
->rb_node
;
2103 struct rb_node
*parent
= NULL
;
2104 struct nfs_access_entry
*entry
;
2106 spin_lock(&inode
->i_lock
);
2107 while (*p
!= NULL
) {
2109 entry
= rb_entry(parent
, struct nfs_access_entry
, rb_node
);
2111 if (set
->cred
< entry
->cred
)
2112 p
= &parent
->rb_left
;
2113 else if (set
->cred
> entry
->cred
)
2114 p
= &parent
->rb_right
;
2118 rb_link_node(&set
->rb_node
, parent
, p
);
2119 rb_insert_color(&set
->rb_node
, root_node
);
2120 list_add_tail(&set
->lru
, &nfsi
->access_cache_entry_lru
);
2121 spin_unlock(&inode
->i_lock
);
2124 rb_replace_node(parent
, &set
->rb_node
, root_node
);
2125 list_add_tail(&set
->lru
, &nfsi
->access_cache_entry_lru
);
2126 list_del(&entry
->lru
);
2127 spin_unlock(&inode
->i_lock
);
2128 nfs_access_free_entry(entry
);
2131 void nfs_access_add_cache(struct inode
*inode
, struct nfs_access_entry
*set
)
2133 struct nfs_access_entry
*cache
= kmalloc(sizeof(*cache
), GFP_KERNEL
);
2136 RB_CLEAR_NODE(&cache
->rb_node
);
2137 cache
->jiffies
= set
->jiffies
;
2138 cache
->cred
= get_rpccred(set
->cred
);
2139 cache
->mask
= set
->mask
;
2141 nfs_access_add_rbtree(inode
, cache
);
2143 /* Update accounting */
2144 smp_mb__before_atomic_inc();
2145 atomic_long_inc(&nfs_access_nr_entries
);
2146 smp_mb__after_atomic_inc();
2148 /* Add inode to global LRU list */
2149 if (!test_bit(NFS_INO_ACL_LRU_SET
, &NFS_I(inode
)->flags
)) {
2150 spin_lock(&nfs_access_lru_lock
);
2151 if (!test_and_set_bit(NFS_INO_ACL_LRU_SET
, &NFS_I(inode
)->flags
))
2152 list_add_tail(&NFS_I(inode
)->access_cache_inode_lru
,
2153 &nfs_access_lru_list
);
2154 spin_unlock(&nfs_access_lru_lock
);
2157 EXPORT_SYMBOL_GPL(nfs_access_add_cache
);
2159 void nfs_access_set_mask(struct nfs_access_entry
*entry
, u32 access_result
)
2162 if (access_result
& NFS4_ACCESS_READ
)
2163 entry
->mask
|= MAY_READ
;
2165 (NFS4_ACCESS_MODIFY
| NFS4_ACCESS_EXTEND
| NFS4_ACCESS_DELETE
))
2166 entry
->mask
|= MAY_WRITE
;
2167 if (access_result
& (NFS4_ACCESS_LOOKUP
|NFS4_ACCESS_EXECUTE
))
2168 entry
->mask
|= MAY_EXEC
;
2170 EXPORT_SYMBOL_GPL(nfs_access_set_mask
);
2172 static int nfs_do_access(struct inode
*inode
, struct rpc_cred
*cred
, int mask
)
2174 struct nfs_access_entry cache
;
2177 status
= nfs_access_get_cached(inode
, cred
, &cache
);
2181 /* Be clever: ask server to check for all possible rights */
2182 cache
.mask
= MAY_EXEC
| MAY_WRITE
| MAY_READ
;
2184 cache
.jiffies
= jiffies
;
2185 status
= NFS_PROTO(inode
)->access(inode
, &cache
);
2187 if (status
== -ESTALE
) {
2188 nfs_zap_caches(inode
);
2189 if (!S_ISDIR(inode
->i_mode
))
2190 set_bit(NFS_INO_STALE
, &NFS_I(inode
)->flags
);
2194 nfs_access_add_cache(inode
, &cache
);
2196 if ((mask
& ~cache
.mask
& (MAY_READ
| MAY_WRITE
| MAY_EXEC
)) == 0)
2201 static int nfs_open_permission_mask(int openflags
)
2205 if (openflags
& __FMODE_EXEC
) {
2206 /* ONLY check exec rights */
2209 if ((openflags
& O_ACCMODE
) != O_WRONLY
)
2211 if ((openflags
& O_ACCMODE
) != O_RDONLY
)
2218 int nfs_may_open(struct inode
*inode
, struct rpc_cred
*cred
, int openflags
)
2220 return nfs_do_access(inode
, cred
, nfs_open_permission_mask(openflags
));
2222 EXPORT_SYMBOL_GPL(nfs_may_open
);
2224 int nfs_permission(struct inode
*inode
, int mask
)
2226 struct rpc_cred
*cred
;
2229 if (mask
& MAY_NOT_BLOCK
)
2232 nfs_inc_stats(inode
, NFSIOS_VFSACCESS
);
2234 if ((mask
& (MAY_READ
| MAY_WRITE
| MAY_EXEC
)) == 0)
2236 /* Is this sys_access() ? */
2237 if (mask
& (MAY_ACCESS
| MAY_CHDIR
))
2240 switch (inode
->i_mode
& S_IFMT
) {
2244 /* NFSv4 has atomic_open... */
2245 if (nfs_server_capable(inode
, NFS_CAP_ATOMIC_OPEN
)
2246 && (mask
& MAY_OPEN
)
2247 && !(mask
& MAY_EXEC
))
2252 * Optimize away all write operations, since the server
2253 * will check permissions when we perform the op.
2255 if ((mask
& MAY_WRITE
) && !(mask
& MAY_READ
))
2260 if (!NFS_PROTO(inode
)->access
)
2263 cred
= rpc_lookup_cred();
2264 if (!IS_ERR(cred
)) {
2265 res
= nfs_do_access(inode
, cred
, mask
);
2268 res
= PTR_ERR(cred
);
2270 if (!res
&& (mask
& MAY_EXEC
) && !execute_ok(inode
))
2273 dfprintk(VFS
, "NFS: permission(%s/%ld), mask=0x%x, res=%d\n",
2274 inode
->i_sb
->s_id
, inode
->i_ino
, mask
, res
);
2277 res
= nfs_revalidate_inode(NFS_SERVER(inode
), inode
);
2279 res
= generic_permission(inode
, mask
);
2282 EXPORT_SYMBOL_GPL(nfs_permission
);
2286 * version-control: t
2287 * kept-new-versions: 5