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"
48 /* #define NFS_DEBUG_VERBOSE 1 */
50 static int nfs_opendir(struct inode
*, struct file
*);
51 static int nfs_closedir(struct inode
*, struct file
*);
52 static int nfs_readdir(struct file
*, struct dir_context
*);
53 static int nfs_fsync_dir(struct file
*, loff_t
, loff_t
, int);
54 static loff_t
nfs_llseek_dir(struct file
*, loff_t
, int);
55 static void nfs_readdir_clear_array(struct page
*);
57 const struct file_operations nfs_dir_operations
= {
58 .llseek
= nfs_llseek_dir
,
59 .read
= generic_read_dir
,
60 .iterate
= nfs_readdir
,
62 .release
= nfs_closedir
,
63 .fsync
= nfs_fsync_dir
,
66 const struct address_space_operations nfs_dir_aops
= {
67 .freepage
= nfs_readdir_clear_array
,
70 static struct nfs_open_dir_context
*alloc_nfs_open_dir_context(struct inode
*dir
, struct rpc_cred
*cred
)
72 struct nfs_open_dir_context
*ctx
;
73 ctx
= kmalloc(sizeof(*ctx
), GFP_KERNEL
);
76 ctx
->attr_gencount
= NFS_I(dir
)->attr_gencount
;
79 ctx
->cred
= get_rpccred(cred
);
82 return ERR_PTR(-ENOMEM
);
85 static void put_nfs_open_dir_context(struct nfs_open_dir_context
*ctx
)
87 put_rpccred(ctx
->cred
);
95 nfs_opendir(struct inode
*inode
, struct file
*filp
)
98 struct nfs_open_dir_context
*ctx
;
99 struct rpc_cred
*cred
;
101 dfprintk(FILE, "NFS: open dir(%s/%s)\n",
102 filp
->f_path
.dentry
->d_parent
->d_name
.name
,
103 filp
->f_path
.dentry
->d_name
.name
);
105 nfs_inc_stats(inode
, NFSIOS_VFSOPEN
);
107 cred
= rpc_lookup_cred();
109 return PTR_ERR(cred
);
110 ctx
= alloc_nfs_open_dir_context(inode
, cred
);
115 filp
->private_data
= ctx
;
116 if (filp
->f_path
.dentry
== filp
->f_path
.mnt
->mnt_root
) {
117 /* This is a mountpoint, so d_revalidate will never
118 * have been called, so we need to refresh the
119 * inode (for close-open consistency) ourselves.
121 __nfs_revalidate_inode(NFS_SERVER(inode
), inode
);
129 nfs_closedir(struct inode
*inode
, struct file
*filp
)
131 put_nfs_open_dir_context(filp
->private_data
);
135 struct nfs_cache_array_entry
{
139 unsigned char d_type
;
142 struct nfs_cache_array
{
146 struct nfs_cache_array_entry array
[0];
149 typedef int (*decode_dirent_t
)(struct xdr_stream
*, struct nfs_entry
*, int);
153 struct dir_context
*ctx
;
154 unsigned long page_index
;
157 loff_t current_index
;
158 decode_dirent_t decode
;
160 unsigned long timestamp
;
161 unsigned long gencount
;
162 unsigned int cache_entry_index
;
165 } nfs_readdir_descriptor_t
;
168 * The caller is responsible for calling nfs_readdir_release_array(page)
171 struct nfs_cache_array
*nfs_readdir_get_array(struct page
*page
)
175 return ERR_PTR(-EIO
);
178 return ERR_PTR(-ENOMEM
);
183 void nfs_readdir_release_array(struct page
*page
)
189 * we are freeing strings created by nfs_add_to_readdir_array()
192 void nfs_readdir_clear_array(struct page
*page
)
194 struct nfs_cache_array
*array
;
197 array
= kmap_atomic(page
);
198 for (i
= 0; i
< array
->size
; i
++)
199 kfree(array
->array
[i
].string
.name
);
200 kunmap_atomic(array
);
204 * the caller is responsible for freeing qstr.name
205 * when called by nfs_readdir_add_to_array, the strings will be freed in
206 * nfs_clear_readdir_array()
209 int nfs_readdir_make_qstr(struct qstr
*string
, const char *name
, unsigned int len
)
212 string
->name
= kmemdup(name
, len
, GFP_KERNEL
);
213 if (string
->name
== NULL
)
216 * Avoid a kmemleak false positive. The pointer to the name is stored
217 * in a page cache page which kmemleak does not scan.
219 kmemleak_not_leak(string
->name
);
220 string
->hash
= full_name_hash(name
, len
);
225 int nfs_readdir_add_to_array(struct nfs_entry
*entry
, struct page
*page
)
227 struct nfs_cache_array
*array
= nfs_readdir_get_array(page
);
228 struct nfs_cache_array_entry
*cache_entry
;
232 return PTR_ERR(array
);
234 cache_entry
= &array
->array
[array
->size
];
236 /* Check that this entry lies within the page bounds */
238 if ((char *)&cache_entry
[1] - (char *)page_address(page
) > PAGE_SIZE
)
241 cache_entry
->cookie
= entry
->prev_cookie
;
242 cache_entry
->ino
= entry
->ino
;
243 cache_entry
->d_type
= entry
->d_type
;
244 ret
= nfs_readdir_make_qstr(&cache_entry
->string
, entry
->name
, entry
->len
);
247 array
->last_cookie
= entry
->cookie
;
250 array
->eof_index
= array
->size
;
252 nfs_readdir_release_array(page
);
257 int nfs_readdir_search_for_pos(struct nfs_cache_array
*array
, nfs_readdir_descriptor_t
*desc
)
259 loff_t diff
= desc
->ctx
->pos
- desc
->current_index
;
264 if (diff
>= array
->size
) {
265 if (array
->eof_index
>= 0)
270 index
= (unsigned int)diff
;
271 *desc
->dir_cookie
= array
->array
[index
].cookie
;
272 desc
->cache_entry_index
= index
;
280 int nfs_readdir_search_for_cookie(struct nfs_cache_array
*array
, nfs_readdir_descriptor_t
*desc
)
284 int status
= -EAGAIN
;
286 for (i
= 0; i
< array
->size
; i
++) {
287 if (array
->array
[i
].cookie
== *desc
->dir_cookie
) {
288 struct nfs_inode
*nfsi
= NFS_I(file_inode(desc
->file
));
289 struct nfs_open_dir_context
*ctx
= desc
->file
->private_data
;
291 new_pos
= desc
->current_index
+ i
;
292 if (ctx
->attr_gencount
!= nfsi
->attr_gencount
293 || (nfsi
->cache_validity
& (NFS_INO_INVALID_ATTR
|NFS_INO_INVALID_DATA
))) {
295 ctx
->attr_gencount
= nfsi
->attr_gencount
;
296 } else if (new_pos
< desc
->ctx
->pos
) {
298 && ctx
->dup_cookie
== *desc
->dir_cookie
) {
299 if (printk_ratelimit()) {
300 pr_notice("NFS: directory %s/%s contains a readdir loop."
301 "Please contact your server vendor. "
302 "The file: %s has duplicate cookie %llu\n",
303 desc
->file
->f_dentry
->d_parent
->d_name
.name
,
304 desc
->file
->f_dentry
->d_name
.name
,
305 array
->array
[i
].string
.name
,
311 ctx
->dup_cookie
= *desc
->dir_cookie
;
314 desc
->ctx
->pos
= new_pos
;
315 desc
->cache_entry_index
= i
;
319 if (array
->eof_index
>= 0) {
320 status
= -EBADCOOKIE
;
321 if (*desc
->dir_cookie
== array
->last_cookie
)
329 int nfs_readdir_search_array(nfs_readdir_descriptor_t
*desc
)
331 struct nfs_cache_array
*array
;
334 array
= nfs_readdir_get_array(desc
->page
);
336 status
= PTR_ERR(array
);
340 if (*desc
->dir_cookie
== 0)
341 status
= nfs_readdir_search_for_pos(array
, desc
);
343 status
= nfs_readdir_search_for_cookie(array
, desc
);
345 if (status
== -EAGAIN
) {
346 desc
->last_cookie
= array
->last_cookie
;
347 desc
->current_index
+= array
->size
;
350 nfs_readdir_release_array(desc
->page
);
355 /* Fill a page with xdr information before transferring to the cache page */
357 int nfs_readdir_xdr_filler(struct page
**pages
, nfs_readdir_descriptor_t
*desc
,
358 struct nfs_entry
*entry
, struct file
*file
, struct inode
*inode
)
360 struct nfs_open_dir_context
*ctx
= file
->private_data
;
361 struct rpc_cred
*cred
= ctx
->cred
;
362 unsigned long timestamp
, gencount
;
367 gencount
= nfs_inc_attr_generation_counter();
368 error
= NFS_PROTO(inode
)->readdir(file
->f_path
.dentry
, cred
, entry
->cookie
, pages
,
369 NFS_SERVER(inode
)->dtsize
, desc
->plus
);
371 /* We requested READDIRPLUS, but the server doesn't grok it */
372 if (error
== -ENOTSUPP
&& desc
->plus
) {
373 NFS_SERVER(inode
)->caps
&= ~NFS_CAP_READDIRPLUS
;
374 clear_bit(NFS_INO_ADVISE_RDPLUS
, &NFS_I(inode
)->flags
);
380 desc
->timestamp
= timestamp
;
381 desc
->gencount
= gencount
;
386 static int xdr_decode(nfs_readdir_descriptor_t
*desc
,
387 struct nfs_entry
*entry
, struct xdr_stream
*xdr
)
391 error
= desc
->decode(xdr
, entry
, desc
->plus
);
394 entry
->fattr
->time_start
= desc
->timestamp
;
395 entry
->fattr
->gencount
= desc
->gencount
;
400 int nfs_same_file(struct dentry
*dentry
, struct nfs_entry
*entry
)
402 if (dentry
->d_inode
== NULL
)
404 if (nfs_compare_fh(entry
->fh
, NFS_FH(dentry
->d_inode
)) != 0)
412 bool nfs_use_readdirplus(struct inode
*dir
, struct dir_context
*ctx
)
414 if (!nfs_server_capable(dir
, NFS_CAP_READDIRPLUS
))
416 if (test_and_clear_bit(NFS_INO_ADVISE_RDPLUS
, &NFS_I(dir
)->flags
))
424 * This function is called by the lookup code to request the use of
425 * readdirplus to accelerate any future lookups in the same
429 void nfs_advise_use_readdirplus(struct inode
*dir
)
431 set_bit(NFS_INO_ADVISE_RDPLUS
, &NFS_I(dir
)->flags
);
435 void nfs_prime_dcache(struct dentry
*parent
, struct nfs_entry
*entry
)
437 struct qstr filename
= QSTR_INIT(entry
->name
, entry
->len
);
438 struct dentry
*dentry
;
439 struct dentry
*alias
;
440 struct inode
*dir
= parent
->d_inode
;
444 if (filename
.name
[0] == '.') {
445 if (filename
.len
== 1)
447 if (filename
.len
== 2 && filename
.name
[1] == '.')
450 filename
.hash
= full_name_hash(filename
.name
, filename
.len
);
452 dentry
= d_lookup(parent
, &filename
);
453 if (dentry
!= NULL
) {
454 if (nfs_same_file(dentry
, entry
)) {
455 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
456 status
= nfs_refresh_inode(dentry
->d_inode
, entry
->fattr
);
458 nfs_setsecurity(dentry
->d_inode
, entry
->fattr
, entry
->label
);
461 if (d_invalidate(dentry
) != 0)
467 dentry
= d_alloc(parent
, &filename
);
471 inode
= nfs_fhget(dentry
->d_sb
, entry
->fh
, entry
->fattr
, entry
->label
);
475 alias
= d_materialise_unique(dentry
, inode
);
479 nfs_set_verifier(alias
, nfs_save_change_attribute(dir
));
482 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
488 /* Perform conversion from xdr to cache array */
490 int nfs_readdir_page_filler(nfs_readdir_descriptor_t
*desc
, struct nfs_entry
*entry
,
491 struct page
**xdr_pages
, struct page
*page
, unsigned int buflen
)
493 struct xdr_stream stream
;
495 struct page
*scratch
;
496 struct nfs_cache_array
*array
;
497 unsigned int count
= 0;
500 scratch
= alloc_page(GFP_KERNEL
);
504 xdr_init_decode_pages(&stream
, &buf
, xdr_pages
, buflen
);
505 xdr_set_scratch_buffer(&stream
, page_address(scratch
), PAGE_SIZE
);
508 status
= xdr_decode(desc
, entry
, &stream
);
510 if (status
== -EAGAIN
)
518 nfs_prime_dcache(desc
->file
->f_path
.dentry
, entry
);
520 status
= nfs_readdir_add_to_array(entry
, page
);
523 } while (!entry
->eof
);
525 if (count
== 0 || (status
== -EBADCOOKIE
&& entry
->eof
!= 0)) {
526 array
= nfs_readdir_get_array(page
);
527 if (!IS_ERR(array
)) {
528 array
->eof_index
= array
->size
;
530 nfs_readdir_release_array(page
);
532 status
= PTR_ERR(array
);
540 void nfs_readdir_free_pagearray(struct page
**pages
, unsigned int npages
)
543 for (i
= 0; i
< npages
; i
++)
548 void nfs_readdir_free_large_page(void *ptr
, struct page
**pages
,
551 nfs_readdir_free_pagearray(pages
, npages
);
555 * nfs_readdir_large_page will allocate pages that must be freed with a call
556 * to nfs_readdir_free_large_page
559 int nfs_readdir_large_page(struct page
**pages
, unsigned int npages
)
563 for (i
= 0; i
< npages
; i
++) {
564 struct page
*page
= alloc_page(GFP_KERNEL
);
572 nfs_readdir_free_pagearray(pages
, i
);
577 int nfs_readdir_xdr_to_array(nfs_readdir_descriptor_t
*desc
, struct page
*page
, struct inode
*inode
)
579 struct page
*pages
[NFS_MAX_READDIR_PAGES
];
580 void *pages_ptr
= NULL
;
581 struct nfs_entry entry
;
582 struct file
*file
= desc
->file
;
583 struct nfs_cache_array
*array
;
584 int status
= -ENOMEM
;
585 unsigned int array_size
= ARRAY_SIZE(pages
);
587 entry
.prev_cookie
= 0;
588 entry
.cookie
= desc
->last_cookie
;
590 entry
.fh
= nfs_alloc_fhandle();
591 entry
.fattr
= nfs_alloc_fattr();
592 entry
.server
= NFS_SERVER(inode
);
593 if (entry
.fh
== NULL
|| entry
.fattr
== NULL
)
596 entry
.label
= nfs4_label_alloc(NFS_SERVER(inode
), GFP_NOWAIT
);
597 if (IS_ERR(entry
.label
)) {
598 status
= PTR_ERR(entry
.label
);
602 array
= nfs_readdir_get_array(page
);
604 status
= PTR_ERR(array
);
607 memset(array
, 0, sizeof(struct nfs_cache_array
));
608 array
->eof_index
= -1;
610 status
= nfs_readdir_large_page(pages
, array_size
);
612 goto out_release_array
;
615 status
= nfs_readdir_xdr_filler(pages
, desc
, &entry
, file
, inode
);
620 status
= nfs_readdir_page_filler(desc
, &entry
, pages
, page
, pglen
);
622 if (status
== -ENOSPC
)
626 } while (array
->eof_index
< 0);
628 nfs_readdir_free_large_page(pages_ptr
, pages
, array_size
);
630 nfs_readdir_release_array(page
);
632 nfs4_label_free(entry
.label
);
634 nfs_free_fattr(entry
.fattr
);
635 nfs_free_fhandle(entry
.fh
);
640 * Now we cache directories properly, by converting xdr information
641 * to an array that can be used for lookups later. This results in
642 * fewer cache pages, since we can store more information on each page.
643 * We only need to convert from xdr once so future lookups are much simpler
646 int nfs_readdir_filler(nfs_readdir_descriptor_t
*desc
, struct page
* page
)
648 struct inode
*inode
= file_inode(desc
->file
);
651 ret
= nfs_readdir_xdr_to_array(desc
, page
, inode
);
654 SetPageUptodate(page
);
656 if (invalidate_inode_pages2_range(inode
->i_mapping
, page
->index
+ 1, -1) < 0) {
657 /* Should never happen */
658 nfs_zap_mapping(inode
, inode
->i_mapping
);
668 void cache_page_release(nfs_readdir_descriptor_t
*desc
)
670 if (!desc
->page
->mapping
)
671 nfs_readdir_clear_array(desc
->page
);
672 page_cache_release(desc
->page
);
677 struct page
*get_cache_page(nfs_readdir_descriptor_t
*desc
)
679 return read_cache_page(file_inode(desc
->file
)->i_mapping
,
680 desc
->page_index
, (filler_t
*)nfs_readdir_filler
, desc
);
684 * Returns 0 if desc->dir_cookie was found on page desc->page_index
687 int find_cache_page(nfs_readdir_descriptor_t
*desc
)
691 desc
->page
= get_cache_page(desc
);
692 if (IS_ERR(desc
->page
))
693 return PTR_ERR(desc
->page
);
695 res
= nfs_readdir_search_array(desc
);
697 cache_page_release(desc
);
701 /* Search for desc->dir_cookie from the beginning of the page cache */
703 int readdir_search_pagecache(nfs_readdir_descriptor_t
*desc
)
707 if (desc
->page_index
== 0) {
708 desc
->current_index
= 0;
709 desc
->last_cookie
= 0;
712 res
= find_cache_page(desc
);
713 } while (res
== -EAGAIN
);
718 * Once we've found the start of the dirent within a page: fill 'er up...
721 int nfs_do_filldir(nfs_readdir_descriptor_t
*desc
)
723 struct file
*file
= desc
->file
;
726 struct nfs_cache_array
*array
= NULL
;
727 struct nfs_open_dir_context
*ctx
= file
->private_data
;
729 array
= nfs_readdir_get_array(desc
->page
);
731 res
= PTR_ERR(array
);
735 for (i
= desc
->cache_entry_index
; i
< array
->size
; i
++) {
736 struct nfs_cache_array_entry
*ent
;
738 ent
= &array
->array
[i
];
739 if (!dir_emit(desc
->ctx
, ent
->string
.name
, ent
->string
.len
,
740 nfs_compat_user_ino64(ent
->ino
), ent
->d_type
)) {
745 if (i
< (array
->size
-1))
746 *desc
->dir_cookie
= array
->array
[i
+1].cookie
;
748 *desc
->dir_cookie
= array
->last_cookie
;
752 if (array
->eof_index
>= 0)
755 nfs_readdir_release_array(desc
->page
);
757 cache_page_release(desc
);
758 dfprintk(DIRCACHE
, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n",
759 (unsigned long long)*desc
->dir_cookie
, res
);
764 * If we cannot find a cookie in our cache, we suspect that this is
765 * because it points to a deleted file, so we ask the server to return
766 * whatever it thinks is the next entry. We then feed this to filldir.
767 * If all goes well, we should then be able to find our way round the
768 * cache on the next call to readdir_search_pagecache();
770 * NOTE: we cannot add the anonymous page to the pagecache because
771 * the data it contains might not be page aligned. Besides,
772 * we should already have a complete representation of the
773 * directory in the page cache by the time we get here.
776 int uncached_readdir(nfs_readdir_descriptor_t
*desc
)
778 struct page
*page
= NULL
;
780 struct inode
*inode
= file_inode(desc
->file
);
781 struct nfs_open_dir_context
*ctx
= desc
->file
->private_data
;
783 dfprintk(DIRCACHE
, "NFS: uncached_readdir() searching for cookie %Lu\n",
784 (unsigned long long)*desc
->dir_cookie
);
786 page
= alloc_page(GFP_HIGHUSER
);
792 desc
->page_index
= 0;
793 desc
->last_cookie
= *desc
->dir_cookie
;
797 status
= nfs_readdir_xdr_to_array(desc
, page
, inode
);
801 status
= nfs_do_filldir(desc
);
804 dfprintk(DIRCACHE
, "NFS: %s: returns %d\n",
808 cache_page_release(desc
);
812 /* The file offset position represents the dirent entry number. A
813 last cookie cache takes care of the common case of reading the
816 static int nfs_readdir(struct file
*file
, struct dir_context
*ctx
)
818 struct dentry
*dentry
= file
->f_path
.dentry
;
819 struct inode
*inode
= dentry
->d_inode
;
820 nfs_readdir_descriptor_t my_desc
,
822 struct nfs_open_dir_context
*dir_ctx
= file
->private_data
;
825 dfprintk(FILE, "NFS: readdir(%s/%s) starting at cookie %llu\n",
826 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
,
827 (long long)ctx
->pos
);
828 nfs_inc_stats(inode
, NFSIOS_VFSGETDENTS
);
831 * ctx->pos points to the dirent entry number.
832 * *desc->dir_cookie has the cookie for the next entry. We have
833 * to either find the entry with the appropriate number or
834 * revalidate the cookie.
836 memset(desc
, 0, sizeof(*desc
));
840 desc
->dir_cookie
= &dir_ctx
->dir_cookie
;
841 desc
->decode
= NFS_PROTO(inode
)->decode_dirent
;
842 desc
->plus
= nfs_use_readdirplus(inode
, ctx
) ? 1 : 0;
844 nfs_block_sillyrename(dentry
);
845 if (ctx
->pos
== 0 || nfs_attribute_cache_expired(inode
))
846 res
= nfs_revalidate_mapping(inode
, file
->f_mapping
);
851 res
= readdir_search_pagecache(desc
);
853 if (res
== -EBADCOOKIE
) {
855 /* This means either end of directory */
856 if (*desc
->dir_cookie
&& desc
->eof
== 0) {
857 /* Or that the server has 'lost' a cookie */
858 res
= uncached_readdir(desc
);
864 if (res
== -ETOOSMALL
&& desc
->plus
) {
865 clear_bit(NFS_INO_ADVISE_RDPLUS
, &NFS_I(inode
)->flags
);
866 nfs_zap_caches(inode
);
867 desc
->page_index
= 0;
875 res
= nfs_do_filldir(desc
);
878 } while (!desc
->eof
);
880 nfs_unblock_sillyrename(dentry
);
883 dfprintk(FILE, "NFS: readdir(%s/%s) returns %d\n",
884 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
,
889 static loff_t
nfs_llseek_dir(struct file
*filp
, loff_t offset
, int whence
)
891 struct dentry
*dentry
= filp
->f_path
.dentry
;
892 struct inode
*inode
= dentry
->d_inode
;
893 struct nfs_open_dir_context
*dir_ctx
= filp
->private_data
;
895 dfprintk(FILE, "NFS: llseek dir(%s/%s, %lld, %d)\n",
896 dentry
->d_parent
->d_name
.name
,
900 mutex_lock(&inode
->i_mutex
);
903 offset
+= filp
->f_pos
;
911 if (offset
!= filp
->f_pos
) {
912 filp
->f_pos
= offset
;
913 dir_ctx
->dir_cookie
= 0;
917 mutex_unlock(&inode
->i_mutex
);
922 * All directory operations under NFS are synchronous, so fsync()
923 * is a dummy operation.
925 static int nfs_fsync_dir(struct file
*filp
, loff_t start
, loff_t end
,
928 struct dentry
*dentry
= filp
->f_path
.dentry
;
929 struct inode
*inode
= dentry
->d_inode
;
931 dfprintk(FILE, "NFS: fsync dir(%s/%s) datasync %d\n",
932 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
,
935 mutex_lock(&inode
->i_mutex
);
936 nfs_inc_stats(dentry
->d_inode
, NFSIOS_VFSFSYNC
);
937 mutex_unlock(&inode
->i_mutex
);
942 * nfs_force_lookup_revalidate - Mark the directory as having changed
943 * @dir - pointer to directory inode
945 * This forces the revalidation code in nfs_lookup_revalidate() to do a
946 * full lookup on all child dentries of 'dir' whenever a change occurs
947 * on the server that might have invalidated our dcache.
949 * The caller should be holding dir->i_lock
951 void nfs_force_lookup_revalidate(struct inode
*dir
)
953 NFS_I(dir
)->cache_change_attribute
++;
955 EXPORT_SYMBOL_GPL(nfs_force_lookup_revalidate
);
958 * A check for whether or not the parent directory has changed.
959 * In the case it has, we assume that the dentries are untrustworthy
960 * and may need to be looked up again.
962 static int nfs_check_verifier(struct inode
*dir
, struct dentry
*dentry
)
966 if (NFS_SERVER(dir
)->flags
& NFS_MOUNT_LOOKUP_CACHE_NONE
)
968 if (!nfs_verify_change_attribute(dir
, dentry
->d_time
))
970 /* Revalidate nfsi->cache_change_attribute before we declare a match */
971 if (nfs_revalidate_inode(NFS_SERVER(dir
), dir
) < 0)
973 if (!nfs_verify_change_attribute(dir
, dentry
->d_time
))
979 * Use intent information to check whether or not we're going to do
980 * an O_EXCL create using this path component.
982 static int nfs_is_exclusive_create(struct inode
*dir
, unsigned int flags
)
984 if (NFS_PROTO(dir
)->version
== 2)
986 return flags
& LOOKUP_EXCL
;
990 * Inode and filehandle revalidation for lookups.
992 * We force revalidation in the cases where the VFS sets LOOKUP_REVAL,
993 * or if the intent information indicates that we're about to open this
994 * particular file and the "nocto" mount flag is not set.
998 int nfs_lookup_verify_inode(struct inode
*inode
, unsigned int flags
)
1000 struct nfs_server
*server
= NFS_SERVER(inode
);
1003 if (IS_AUTOMOUNT(inode
))
1005 /* VFS wants an on-the-wire revalidation */
1006 if (flags
& LOOKUP_REVAL
)
1008 /* This is an open(2) */
1009 if ((flags
& LOOKUP_OPEN
) && !(server
->flags
& NFS_MOUNT_NOCTO
) &&
1010 (S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
)))
1013 return (inode
->i_nlink
== 0) ? -ENOENT
: 0;
1015 ret
= __nfs_revalidate_inode(server
, inode
);
1022 * We judge how long we want to trust negative
1023 * dentries by looking at the parent inode mtime.
1025 * If parent mtime has changed, we revalidate, else we wait for a
1026 * period corresponding to the parent's attribute cache timeout value.
1029 int nfs_neg_need_reval(struct inode
*dir
, struct dentry
*dentry
,
1032 /* Don't revalidate a negative dentry if we're creating a new file */
1033 if (flags
& LOOKUP_CREATE
)
1035 if (NFS_SERVER(dir
)->flags
& NFS_MOUNT_LOOKUP_CACHE_NONEG
)
1037 return !nfs_check_verifier(dir
, dentry
);
1041 * This is called every time the dcache has a lookup hit,
1042 * and we should check whether we can really trust that
1045 * NOTE! The hit can be a negative hit too, don't assume
1048 * If the parent directory is seen to have changed, we throw out the
1049 * cached dentry and do a new lookup.
1051 static int nfs_lookup_revalidate(struct dentry
*dentry
, unsigned int flags
)
1054 struct inode
*inode
;
1055 struct dentry
*parent
;
1056 struct nfs_fh
*fhandle
= NULL
;
1057 struct nfs_fattr
*fattr
= NULL
;
1058 struct nfs4_label
*label
= NULL
;
1061 if (flags
& LOOKUP_RCU
)
1064 parent
= dget_parent(dentry
);
1065 dir
= parent
->d_inode
;
1066 nfs_inc_stats(dir
, NFSIOS_DENTRYREVALIDATE
);
1067 inode
= dentry
->d_inode
;
1070 if (nfs_neg_need_reval(dir
, dentry
, flags
))
1072 goto out_valid_noent
;
1075 if (is_bad_inode(inode
)) {
1076 dfprintk(LOOKUPCACHE
, "%s: %s/%s has dud inode\n",
1077 __func__
, dentry
->d_parent
->d_name
.name
,
1078 dentry
->d_name
.name
);
1082 if (NFS_PROTO(dir
)->have_delegation(inode
, FMODE_READ
))
1083 goto out_set_verifier
;
1085 /* Force a full look up iff the parent directory has changed */
1086 if (!nfs_is_exclusive_create(dir
, flags
) && nfs_check_verifier(dir
, dentry
)) {
1087 if (nfs_lookup_verify_inode(inode
, flags
))
1088 goto out_zap_parent
;
1092 if (NFS_STALE(inode
))
1096 fhandle
= nfs_alloc_fhandle();
1097 fattr
= nfs_alloc_fattr();
1098 if (fhandle
== NULL
|| fattr
== NULL
)
1101 label
= nfs4_label_alloc(NFS_SERVER(inode
), GFP_NOWAIT
);
1105 trace_nfs_lookup_revalidate_enter(dir
, dentry
, flags
);
1106 error
= NFS_PROTO(dir
)->lookup(dir
, &dentry
->d_name
, fhandle
, fattr
, label
);
1107 trace_nfs_lookup_revalidate_exit(dir
, dentry
, flags
, error
);
1110 if (nfs_compare_fh(NFS_FH(inode
), fhandle
))
1112 if ((error
= nfs_refresh_inode(inode
, fattr
)) != 0)
1115 nfs_setsecurity(inode
, fattr
, label
);
1117 nfs_free_fattr(fattr
);
1118 nfs_free_fhandle(fhandle
);
1119 nfs4_label_free(label
);
1122 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1124 /* Success: notify readdir to use READDIRPLUS */
1125 nfs_advise_use_readdirplus(dir
);
1128 dfprintk(LOOKUPCACHE
, "NFS: %s(%s/%s) is valid\n",
1129 __func__
, dentry
->d_parent
->d_name
.name
,
1130 dentry
->d_name
.name
);
1133 nfs_zap_caches(dir
);
1135 nfs_free_fattr(fattr
);
1136 nfs_free_fhandle(fhandle
);
1137 nfs4_label_free(label
);
1138 nfs_mark_for_revalidate(dir
);
1139 if (inode
&& S_ISDIR(inode
->i_mode
)) {
1140 /* Purge readdir caches. */
1141 nfs_zap_caches(inode
);
1142 if (dentry
->d_flags
& DCACHE_DISCONNECTED
)
1145 /* If we have submounts, don't unhash ! */
1146 if (check_submounts_and_drop(dentry
) != 0)
1150 dfprintk(LOOKUPCACHE
, "NFS: %s(%s/%s) is invalid\n",
1151 __func__
, dentry
->d_parent
->d_name
.name
,
1152 dentry
->d_name
.name
);
1155 nfs_free_fattr(fattr
);
1156 nfs_free_fhandle(fhandle
);
1157 nfs4_label_free(label
);
1159 dfprintk(LOOKUPCACHE
, "NFS: %s(%s/%s) lookup returned error %d\n",
1160 __func__
, dentry
->d_parent
->d_name
.name
,
1161 dentry
->d_name
.name
, error
);
1166 * A weaker form of d_revalidate for revalidating just the dentry->d_inode
1167 * when we don't really care about the dentry name. This is called when a
1168 * pathwalk ends on a dentry that was not found via a normal lookup in the
1169 * parent dir (e.g.: ".", "..", procfs symlinks or mountpoint traversals).
1171 * In this situation, we just want to verify that the inode itself is OK
1172 * since the dentry might have changed on the server.
1174 static int nfs_weak_revalidate(struct dentry
*dentry
, unsigned int flags
)
1177 struct inode
*inode
= dentry
->d_inode
;
1180 * I believe we can only get a negative dentry here in the case of a
1181 * procfs-style symlink. Just assume it's correct for now, but we may
1182 * eventually need to do something more here.
1185 dfprintk(LOOKUPCACHE
, "%s: %s/%s has negative inode\n",
1186 __func__
, dentry
->d_parent
->d_name
.name
,
1187 dentry
->d_name
.name
);
1191 if (is_bad_inode(inode
)) {
1192 dfprintk(LOOKUPCACHE
, "%s: %s/%s has dud inode\n",
1193 __func__
, dentry
->d_parent
->d_name
.name
,
1194 dentry
->d_name
.name
);
1198 error
= nfs_revalidate_inode(NFS_SERVER(inode
), inode
);
1199 dfprintk(LOOKUPCACHE
, "NFS: %s: inode %lu is %s\n",
1200 __func__
, inode
->i_ino
, error
? "invalid" : "valid");
1205 * This is called from dput() when d_count is going to 0.
1207 static int nfs_dentry_delete(const struct dentry
*dentry
)
1209 dfprintk(VFS
, "NFS: dentry_delete(%s/%s, %x)\n",
1210 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
,
1213 /* Unhash any dentry with a stale inode */
1214 if (dentry
->d_inode
!= NULL
&& NFS_STALE(dentry
->d_inode
))
1217 if (dentry
->d_flags
& DCACHE_NFSFS_RENAMED
) {
1218 /* Unhash it, so that ->d_iput() would be called */
1221 if (!(dentry
->d_sb
->s_flags
& MS_ACTIVE
)) {
1222 /* Unhash it, so that ancestors of killed async unlink
1223 * files will be cleaned up during umount */
1230 /* Ensure that we revalidate inode->i_nlink */
1231 static void nfs_drop_nlink(struct inode
*inode
)
1233 spin_lock(&inode
->i_lock
);
1234 /* drop the inode if we're reasonably sure this is the last link */
1235 if (inode
->i_nlink
== 1)
1237 NFS_I(inode
)->cache_validity
|= NFS_INO_INVALID_ATTR
;
1238 spin_unlock(&inode
->i_lock
);
1242 * Called when the dentry loses inode.
1243 * We use it to clean up silly-renamed files.
1245 static void nfs_dentry_iput(struct dentry
*dentry
, struct inode
*inode
)
1247 if (S_ISDIR(inode
->i_mode
))
1248 /* drop any readdir cache as it could easily be old */
1249 NFS_I(inode
)->cache_validity
|= NFS_INO_INVALID_DATA
;
1251 if (dentry
->d_flags
& DCACHE_NFSFS_RENAMED
) {
1252 nfs_complete_unlink(dentry
, inode
);
1253 nfs_drop_nlink(inode
);
1258 static void nfs_d_release(struct dentry
*dentry
)
1260 /* free cached devname value, if it survived that far */
1261 if (unlikely(dentry
->d_fsdata
)) {
1262 if (dentry
->d_flags
& DCACHE_NFSFS_RENAMED
)
1265 kfree(dentry
->d_fsdata
);
1269 const struct dentry_operations nfs_dentry_operations
= {
1270 .d_revalidate
= nfs_lookup_revalidate
,
1271 .d_weak_revalidate
= nfs_weak_revalidate
,
1272 .d_delete
= nfs_dentry_delete
,
1273 .d_iput
= nfs_dentry_iput
,
1274 .d_automount
= nfs_d_automount
,
1275 .d_release
= nfs_d_release
,
1277 EXPORT_SYMBOL_GPL(nfs_dentry_operations
);
1279 struct dentry
*nfs_lookup(struct inode
*dir
, struct dentry
* dentry
, unsigned int flags
)
1282 struct dentry
*parent
;
1283 struct inode
*inode
= NULL
;
1284 struct nfs_fh
*fhandle
= NULL
;
1285 struct nfs_fattr
*fattr
= NULL
;
1286 struct nfs4_label
*label
= NULL
;
1289 dfprintk(VFS
, "NFS: lookup(%s/%s)\n",
1290 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
);
1291 nfs_inc_stats(dir
, NFSIOS_VFSLOOKUP
);
1293 res
= ERR_PTR(-ENAMETOOLONG
);
1294 if (dentry
->d_name
.len
> NFS_SERVER(dir
)->namelen
)
1298 * If we're doing an exclusive create, optimize away the lookup
1299 * but don't hash the dentry.
1301 if (nfs_is_exclusive_create(dir
, flags
)) {
1302 d_instantiate(dentry
, NULL
);
1307 res
= ERR_PTR(-ENOMEM
);
1308 fhandle
= nfs_alloc_fhandle();
1309 fattr
= nfs_alloc_fattr();
1310 if (fhandle
== NULL
|| fattr
== NULL
)
1313 label
= nfs4_label_alloc(NFS_SERVER(dir
), GFP_NOWAIT
);
1317 parent
= dentry
->d_parent
;
1318 /* Protect against concurrent sillydeletes */
1319 trace_nfs_lookup_enter(dir
, dentry
, flags
);
1320 nfs_block_sillyrename(parent
);
1321 error
= NFS_PROTO(dir
)->lookup(dir
, &dentry
->d_name
, fhandle
, fattr
, label
);
1322 if (error
== -ENOENT
)
1325 res
= ERR_PTR(error
);
1326 goto out_unblock_sillyrename
;
1328 inode
= nfs_fhget(dentry
->d_sb
, fhandle
, fattr
, label
);
1329 res
= ERR_CAST(inode
);
1331 goto out_unblock_sillyrename
;
1333 /* Success: notify readdir to use READDIRPLUS */
1334 nfs_advise_use_readdirplus(dir
);
1337 res
= d_materialise_unique(dentry
, inode
);
1340 goto out_unblock_sillyrename
;
1343 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1344 out_unblock_sillyrename
:
1345 nfs_unblock_sillyrename(parent
);
1346 trace_nfs_lookup_exit(dir
, dentry
, flags
, error
);
1347 nfs4_label_free(label
);
1349 nfs_free_fattr(fattr
);
1350 nfs_free_fhandle(fhandle
);
1353 EXPORT_SYMBOL_GPL(nfs_lookup
);
1355 #if IS_ENABLED(CONFIG_NFS_V4)
1356 static int nfs4_lookup_revalidate(struct dentry
*, unsigned int);
1358 const struct dentry_operations nfs4_dentry_operations
= {
1359 .d_revalidate
= nfs4_lookup_revalidate
,
1360 .d_delete
= nfs_dentry_delete
,
1361 .d_iput
= nfs_dentry_iput
,
1362 .d_automount
= nfs_d_automount
,
1363 .d_release
= nfs_d_release
,
1365 EXPORT_SYMBOL_GPL(nfs4_dentry_operations
);
1367 static fmode_t
flags_to_mode(int flags
)
1369 fmode_t res
= (__force fmode_t
)flags
& FMODE_EXEC
;
1370 if ((flags
& O_ACCMODE
) != O_WRONLY
)
1372 if ((flags
& O_ACCMODE
) != O_RDONLY
)
1377 static struct nfs_open_context
*create_nfs_open_context(struct dentry
*dentry
, int open_flags
)
1379 return alloc_nfs_open_context(dentry
, flags_to_mode(open_flags
));
1382 static int do_open(struct inode
*inode
, struct file
*filp
)
1384 nfs_fscache_set_inode_cookie(inode
, filp
);
1388 static int nfs_finish_open(struct nfs_open_context
*ctx
,
1389 struct dentry
*dentry
,
1390 struct file
*file
, unsigned open_flags
,
1395 if ((open_flags
& (O_CREAT
| O_EXCL
)) == (O_CREAT
| O_EXCL
))
1396 *opened
|= FILE_CREATED
;
1398 err
= finish_open(file
, dentry
, do_open
, opened
);
1401 nfs_file_set_open_context(file
, ctx
);
1407 int nfs_atomic_open(struct inode
*dir
, struct dentry
*dentry
,
1408 struct file
*file
, unsigned open_flags
,
1409 umode_t mode
, int *opened
)
1411 struct nfs_open_context
*ctx
;
1413 struct iattr attr
= { .ia_valid
= ATTR_OPEN
};
1414 struct inode
*inode
;
1415 unsigned int lookup_flags
= 0;
1418 /* Expect a negative dentry */
1419 BUG_ON(dentry
->d_inode
);
1421 dfprintk(VFS
, "NFS: atomic_open(%s/%ld), %s\n",
1422 dir
->i_sb
->s_id
, dir
->i_ino
, dentry
->d_name
.name
);
1424 err
= nfs_check_flags(open_flags
);
1428 /* NFS only supports OPEN on regular files */
1429 if ((open_flags
& O_DIRECTORY
)) {
1430 if (!d_unhashed(dentry
)) {
1432 * Hashed negative dentry with O_DIRECTORY: dentry was
1433 * revalidated and is fine, no need to perform lookup
1438 lookup_flags
= LOOKUP_OPEN
|LOOKUP_DIRECTORY
;
1442 if (dentry
->d_name
.len
> NFS_SERVER(dir
)->namelen
)
1443 return -ENAMETOOLONG
;
1445 if (open_flags
& O_CREAT
) {
1446 attr
.ia_valid
|= ATTR_MODE
;
1447 attr
.ia_mode
= mode
& ~current_umask();
1449 if (open_flags
& O_TRUNC
) {
1450 attr
.ia_valid
|= ATTR_SIZE
;
1454 ctx
= create_nfs_open_context(dentry
, open_flags
);
1459 trace_nfs_atomic_open_enter(dir
, ctx
, open_flags
);
1460 nfs_block_sillyrename(dentry
->d_parent
);
1461 inode
= NFS_PROTO(dir
)->open_context(dir
, ctx
, open_flags
, &attr
, opened
);
1462 nfs_unblock_sillyrename(dentry
->d_parent
);
1463 if (IS_ERR(inode
)) {
1464 err
= PTR_ERR(inode
);
1465 trace_nfs_atomic_open_exit(dir
, ctx
, open_flags
, err
);
1466 put_nfs_open_context(ctx
);
1470 d_add(dentry
, NULL
);
1476 if (!(open_flags
& O_NOFOLLOW
))
1486 err
= nfs_finish_open(ctx
, ctx
->dentry
, file
, open_flags
, opened
);
1487 trace_nfs_atomic_open_exit(dir
, ctx
, open_flags
, err
);
1488 put_nfs_open_context(ctx
);
1493 res
= nfs_lookup(dir
, dentry
, lookup_flags
);
1498 return finish_no_open(file
, res
);
1500 EXPORT_SYMBOL_GPL(nfs_atomic_open
);
1502 static int nfs4_lookup_revalidate(struct dentry
*dentry
, unsigned int flags
)
1504 struct dentry
*parent
= NULL
;
1505 struct inode
*inode
;
1509 if (flags
& LOOKUP_RCU
)
1512 if (!(flags
& LOOKUP_OPEN
) || (flags
& LOOKUP_DIRECTORY
))
1514 if (d_mountpoint(dentry
))
1516 if (NFS_SB(dentry
->d_sb
)->caps
& NFS_CAP_ATOMIC_OPEN_V1
)
1519 inode
= dentry
->d_inode
;
1520 parent
= dget_parent(dentry
);
1521 dir
= parent
->d_inode
;
1523 /* We can't create new files in nfs_open_revalidate(), so we
1524 * optimize away revalidation of negative dentries.
1526 if (inode
== NULL
) {
1527 if (!nfs_neg_need_reval(dir
, dentry
, flags
))
1532 /* NFS only supports OPEN on regular files */
1533 if (!S_ISREG(inode
->i_mode
))
1535 /* We cannot do exclusive creation on a positive dentry */
1536 if (flags
& LOOKUP_EXCL
)
1539 /* Let f_op->open() actually open (and revalidate) the file */
1549 return nfs_lookup_revalidate(dentry
, flags
);
1552 #endif /* CONFIG_NFSV4 */
1555 * Code common to create, mkdir, and mknod.
1557 int nfs_instantiate(struct dentry
*dentry
, struct nfs_fh
*fhandle
,
1558 struct nfs_fattr
*fattr
,
1559 struct nfs4_label
*label
)
1561 struct dentry
*parent
= dget_parent(dentry
);
1562 struct inode
*dir
= parent
->d_inode
;
1563 struct inode
*inode
;
1564 int error
= -EACCES
;
1568 /* We may have been initialized further down */
1569 if (dentry
->d_inode
)
1571 if (fhandle
->size
== 0) {
1572 error
= NFS_PROTO(dir
)->lookup(dir
, &dentry
->d_name
, fhandle
, fattr
, NULL
);
1576 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1577 if (!(fattr
->valid
& NFS_ATTR_FATTR
)) {
1578 struct nfs_server
*server
= NFS_SB(dentry
->d_sb
);
1579 error
= server
->nfs_client
->rpc_ops
->getattr(server
, fhandle
, fattr
, NULL
);
1583 inode
= nfs_fhget(dentry
->d_sb
, fhandle
, fattr
, label
);
1584 error
= PTR_ERR(inode
);
1587 d_add(dentry
, inode
);
1592 nfs_mark_for_revalidate(dir
);
1596 EXPORT_SYMBOL_GPL(nfs_instantiate
);
1599 * Following a failed create operation, we drop the dentry rather
1600 * than retain a negative dentry. This avoids a problem in the event
1601 * that the operation succeeded on the server, but an error in the
1602 * reply path made it appear to have failed.
1604 int nfs_create(struct inode
*dir
, struct dentry
*dentry
,
1605 umode_t mode
, bool excl
)
1608 int open_flags
= excl
? O_CREAT
| O_EXCL
: O_CREAT
;
1611 dfprintk(VFS
, "NFS: create(%s/%ld), %s\n",
1612 dir
->i_sb
->s_id
, dir
->i_ino
, dentry
->d_name
.name
);
1614 attr
.ia_mode
= mode
;
1615 attr
.ia_valid
= ATTR_MODE
;
1617 trace_nfs_create_enter(dir
, dentry
, open_flags
);
1618 error
= NFS_PROTO(dir
)->create(dir
, dentry
, &attr
, open_flags
);
1619 trace_nfs_create_exit(dir
, dentry
, open_flags
, error
);
1627 EXPORT_SYMBOL_GPL(nfs_create
);
1630 * See comments for nfs_proc_create regarding failed operations.
1633 nfs_mknod(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
, dev_t rdev
)
1638 dfprintk(VFS
, "NFS: mknod(%s/%ld), %s\n",
1639 dir
->i_sb
->s_id
, dir
->i_ino
, dentry
->d_name
.name
);
1641 if (!new_valid_dev(rdev
))
1644 attr
.ia_mode
= mode
;
1645 attr
.ia_valid
= ATTR_MODE
;
1647 trace_nfs_mknod_enter(dir
, dentry
);
1648 status
= NFS_PROTO(dir
)->mknod(dir
, dentry
, &attr
, rdev
);
1649 trace_nfs_mknod_exit(dir
, dentry
, status
);
1657 EXPORT_SYMBOL_GPL(nfs_mknod
);
1660 * See comments for nfs_proc_create regarding failed operations.
1662 int nfs_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
1667 dfprintk(VFS
, "NFS: mkdir(%s/%ld), %s\n",
1668 dir
->i_sb
->s_id
, dir
->i_ino
, dentry
->d_name
.name
);
1670 attr
.ia_valid
= ATTR_MODE
;
1671 attr
.ia_mode
= mode
| S_IFDIR
;
1673 trace_nfs_mkdir_enter(dir
, dentry
);
1674 error
= NFS_PROTO(dir
)->mkdir(dir
, dentry
, &attr
);
1675 trace_nfs_mkdir_exit(dir
, dentry
, error
);
1683 EXPORT_SYMBOL_GPL(nfs_mkdir
);
1685 static void nfs_dentry_handle_enoent(struct dentry
*dentry
)
1687 if (dentry
->d_inode
!= NULL
&& !d_unhashed(dentry
))
1691 int nfs_rmdir(struct inode
*dir
, struct dentry
*dentry
)
1695 dfprintk(VFS
, "NFS: rmdir(%s/%ld), %s\n",
1696 dir
->i_sb
->s_id
, dir
->i_ino
, dentry
->d_name
.name
);
1698 trace_nfs_rmdir_enter(dir
, dentry
);
1699 if (dentry
->d_inode
) {
1700 nfs_wait_on_sillyrename(dentry
);
1701 error
= NFS_PROTO(dir
)->rmdir(dir
, &dentry
->d_name
);
1702 /* Ensure the VFS deletes this inode */
1705 clear_nlink(dentry
->d_inode
);
1708 nfs_dentry_handle_enoent(dentry
);
1711 error
= NFS_PROTO(dir
)->rmdir(dir
, &dentry
->d_name
);
1712 trace_nfs_rmdir_exit(dir
, dentry
, error
);
1716 EXPORT_SYMBOL_GPL(nfs_rmdir
);
1719 * Remove a file after making sure there are no pending writes,
1720 * and after checking that the file has only one user.
1722 * We invalidate the attribute cache and free the inode prior to the operation
1723 * to avoid possible races if the server reuses the inode.
1725 static int nfs_safe_remove(struct dentry
*dentry
)
1727 struct inode
*dir
= dentry
->d_parent
->d_inode
;
1728 struct inode
*inode
= dentry
->d_inode
;
1731 dfprintk(VFS
, "NFS: safe_remove(%s/%s)\n",
1732 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
);
1734 /* If the dentry was sillyrenamed, we simply call d_delete() */
1735 if (dentry
->d_flags
& DCACHE_NFSFS_RENAMED
) {
1740 trace_nfs_remove_enter(dir
, dentry
);
1741 if (inode
!= NULL
) {
1742 NFS_PROTO(inode
)->return_delegation(inode
);
1743 error
= NFS_PROTO(dir
)->remove(dir
, &dentry
->d_name
);
1745 nfs_drop_nlink(inode
);
1747 error
= NFS_PROTO(dir
)->remove(dir
, &dentry
->d_name
);
1748 if (error
== -ENOENT
)
1749 nfs_dentry_handle_enoent(dentry
);
1750 trace_nfs_remove_exit(dir
, dentry
, error
);
1755 /* We do silly rename. In case sillyrename() returns -EBUSY, the inode
1756 * belongs to an active ".nfs..." file and we return -EBUSY.
1758 * If sillyrename() returns 0, we do nothing, otherwise we unlink.
1760 int nfs_unlink(struct inode
*dir
, struct dentry
*dentry
)
1763 int need_rehash
= 0;
1765 dfprintk(VFS
, "NFS: unlink(%s/%ld, %s)\n", dir
->i_sb
->s_id
,
1766 dir
->i_ino
, dentry
->d_name
.name
);
1768 trace_nfs_unlink_enter(dir
, dentry
);
1769 spin_lock(&dentry
->d_lock
);
1770 if (d_count(dentry
) > 1) {
1771 spin_unlock(&dentry
->d_lock
);
1772 /* Start asynchronous writeout of the inode */
1773 write_inode_now(dentry
->d_inode
, 0);
1774 error
= nfs_sillyrename(dir
, dentry
);
1777 if (!d_unhashed(dentry
)) {
1781 spin_unlock(&dentry
->d_lock
);
1782 error
= nfs_safe_remove(dentry
);
1783 if (!error
|| error
== -ENOENT
) {
1784 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1785 } else if (need_rehash
)
1788 trace_nfs_unlink_exit(dir
, dentry
, error
);
1791 EXPORT_SYMBOL_GPL(nfs_unlink
);
1794 * To create a symbolic link, most file systems instantiate a new inode,
1795 * add a page to it containing the path, then write it out to the disk
1796 * using prepare_write/commit_write.
1798 * Unfortunately the NFS client can't create the in-core inode first
1799 * because it needs a file handle to create an in-core inode (see
1800 * fs/nfs/inode.c:nfs_fhget). We only have a file handle *after* the
1801 * symlink request has completed on the server.
1803 * So instead we allocate a raw page, copy the symname into it, then do
1804 * the SYMLINK request with the page as the buffer. If it succeeds, we
1805 * now have a new file handle and can instantiate an in-core NFS inode
1806 * and move the raw page into its mapping.
1808 int nfs_symlink(struct inode
*dir
, struct dentry
*dentry
, const char *symname
)
1813 unsigned int pathlen
= strlen(symname
);
1816 dfprintk(VFS
, "NFS: symlink(%s/%ld, %s, %s)\n", dir
->i_sb
->s_id
,
1817 dir
->i_ino
, dentry
->d_name
.name
, symname
);
1819 if (pathlen
> PAGE_SIZE
)
1820 return -ENAMETOOLONG
;
1822 attr
.ia_mode
= S_IFLNK
| S_IRWXUGO
;
1823 attr
.ia_valid
= ATTR_MODE
;
1825 page
= alloc_page(GFP_HIGHUSER
);
1829 kaddr
= kmap_atomic(page
);
1830 memcpy(kaddr
, symname
, pathlen
);
1831 if (pathlen
< PAGE_SIZE
)
1832 memset(kaddr
+ pathlen
, 0, PAGE_SIZE
- pathlen
);
1833 kunmap_atomic(kaddr
);
1835 trace_nfs_symlink_enter(dir
, dentry
);
1836 error
= NFS_PROTO(dir
)->symlink(dir
, dentry
, page
, pathlen
, &attr
);
1837 trace_nfs_symlink_exit(dir
, dentry
, error
);
1839 dfprintk(VFS
, "NFS: symlink(%s/%ld, %s, %s) error %d\n",
1840 dir
->i_sb
->s_id
, dir
->i_ino
,
1841 dentry
->d_name
.name
, symname
, error
);
1848 * No big deal if we can't add this page to the page cache here.
1849 * READLINK will get the missing page from the server if needed.
1851 if (!add_to_page_cache_lru(page
, dentry
->d_inode
->i_mapping
, 0,
1853 SetPageUptodate(page
);
1860 EXPORT_SYMBOL_GPL(nfs_symlink
);
1863 nfs_link(struct dentry
*old_dentry
, struct inode
*dir
, struct dentry
*dentry
)
1865 struct inode
*inode
= old_dentry
->d_inode
;
1868 dfprintk(VFS
, "NFS: link(%s/%s -> %s/%s)\n",
1869 old_dentry
->d_parent
->d_name
.name
, old_dentry
->d_name
.name
,
1870 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
);
1872 trace_nfs_link_enter(inode
, dir
, dentry
);
1873 NFS_PROTO(inode
)->return_delegation(inode
);
1876 error
= NFS_PROTO(dir
)->link(inode
, dir
, &dentry
->d_name
);
1879 d_add(dentry
, inode
);
1881 trace_nfs_link_exit(inode
, dir
, dentry
, error
);
1884 EXPORT_SYMBOL_GPL(nfs_link
);
1888 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a
1889 * different file handle for the same inode after a rename (e.g. when
1890 * moving to a different directory). A fail-safe method to do so would
1891 * be to look up old_dir/old_name, create a link to new_dir/new_name and
1892 * rename the old file using the sillyrename stuff. This way, the original
1893 * file in old_dir will go away when the last process iput()s the inode.
1897 * It actually works quite well. One needs to have the possibility for
1898 * at least one ".nfs..." file in each directory the file ever gets
1899 * moved or linked to which happens automagically with the new
1900 * implementation that only depends on the dcache stuff instead of
1901 * using the inode layer
1903 * Unfortunately, things are a little more complicated than indicated
1904 * above. For a cross-directory move, we want to make sure we can get
1905 * rid of the old inode after the operation. This means there must be
1906 * no pending writes (if it's a file), and the use count must be 1.
1907 * If these conditions are met, we can drop the dentries before doing
1910 int nfs_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
1911 struct inode
*new_dir
, struct dentry
*new_dentry
)
1913 struct inode
*old_inode
= old_dentry
->d_inode
;
1914 struct inode
*new_inode
= new_dentry
->d_inode
;
1915 struct dentry
*dentry
= NULL
, *rehash
= NULL
;
1918 dfprintk(VFS
, "NFS: rename(%s/%s -> %s/%s, ct=%d)\n",
1919 old_dentry
->d_parent
->d_name
.name
, old_dentry
->d_name
.name
,
1920 new_dentry
->d_parent
->d_name
.name
, new_dentry
->d_name
.name
,
1921 d_count(new_dentry
));
1923 trace_nfs_rename_enter(old_dir
, old_dentry
, new_dir
, new_dentry
);
1925 * For non-directories, check whether the target is busy and if so,
1926 * make a copy of the dentry and then do a silly-rename. If the
1927 * silly-rename succeeds, the copied dentry is hashed and becomes
1930 if (new_inode
&& !S_ISDIR(new_inode
->i_mode
)) {
1932 * To prevent any new references to the target during the
1933 * rename, we unhash the dentry in advance.
1935 if (!d_unhashed(new_dentry
)) {
1937 rehash
= new_dentry
;
1940 if (d_count(new_dentry
) > 2) {
1943 /* copy the target dentry's name */
1944 dentry
= d_alloc(new_dentry
->d_parent
,
1945 &new_dentry
->d_name
);
1949 /* silly-rename the existing target ... */
1950 err
= nfs_sillyrename(new_dir
, new_dentry
);
1954 new_dentry
= dentry
;
1960 NFS_PROTO(old_inode
)->return_delegation(old_inode
);
1961 if (new_inode
!= NULL
)
1962 NFS_PROTO(new_inode
)->return_delegation(new_inode
);
1964 error
= NFS_PROTO(old_dir
)->rename(old_dir
, &old_dentry
->d_name
,
1965 new_dir
, &new_dentry
->d_name
);
1966 nfs_mark_for_revalidate(old_inode
);
1970 trace_nfs_rename_exit(old_dir
, old_dentry
,
1971 new_dir
, new_dentry
, error
);
1973 if (new_inode
!= NULL
)
1974 nfs_drop_nlink(new_inode
);
1975 d_move(old_dentry
, new_dentry
);
1976 nfs_set_verifier(new_dentry
,
1977 nfs_save_change_attribute(new_dir
));
1978 } else if (error
== -ENOENT
)
1979 nfs_dentry_handle_enoent(old_dentry
);
1981 /* new dentry created? */
1986 EXPORT_SYMBOL_GPL(nfs_rename
);
1988 static DEFINE_SPINLOCK(nfs_access_lru_lock
);
1989 static LIST_HEAD(nfs_access_lru_list
);
1990 static atomic_long_t nfs_access_nr_entries
;
1992 static void nfs_access_free_entry(struct nfs_access_entry
*entry
)
1994 put_rpccred(entry
->cred
);
1996 smp_mb__before_atomic_dec();
1997 atomic_long_dec(&nfs_access_nr_entries
);
1998 smp_mb__after_atomic_dec();
2001 static void nfs_access_free_list(struct list_head
*head
)
2003 struct nfs_access_entry
*cache
;
2005 while (!list_empty(head
)) {
2006 cache
= list_entry(head
->next
, struct nfs_access_entry
, lru
);
2007 list_del(&cache
->lru
);
2008 nfs_access_free_entry(cache
);
2013 nfs_access_cache_scan(struct shrinker
*shrink
, struct shrink_control
*sc
)
2016 struct nfs_inode
*nfsi
, *next
;
2017 struct nfs_access_entry
*cache
;
2018 int nr_to_scan
= sc
->nr_to_scan
;
2019 gfp_t gfp_mask
= sc
->gfp_mask
;
2022 if ((gfp_mask
& GFP_KERNEL
) != GFP_KERNEL
)
2025 spin_lock(&nfs_access_lru_lock
);
2026 list_for_each_entry_safe(nfsi
, next
, &nfs_access_lru_list
, access_cache_inode_lru
) {
2027 struct inode
*inode
;
2029 if (nr_to_scan
-- == 0)
2031 inode
= &nfsi
->vfs_inode
;
2032 spin_lock(&inode
->i_lock
);
2033 if (list_empty(&nfsi
->access_cache_entry_lru
))
2034 goto remove_lru_entry
;
2035 cache
= list_entry(nfsi
->access_cache_entry_lru
.next
,
2036 struct nfs_access_entry
, lru
);
2037 list_move(&cache
->lru
, &head
);
2038 rb_erase(&cache
->rb_node
, &nfsi
->access_cache
);
2040 if (!list_empty(&nfsi
->access_cache_entry_lru
))
2041 list_move_tail(&nfsi
->access_cache_inode_lru
,
2042 &nfs_access_lru_list
);
2045 list_del_init(&nfsi
->access_cache_inode_lru
);
2046 smp_mb__before_clear_bit();
2047 clear_bit(NFS_INO_ACL_LRU_SET
, &nfsi
->flags
);
2048 smp_mb__after_clear_bit();
2050 spin_unlock(&inode
->i_lock
);
2052 spin_unlock(&nfs_access_lru_lock
);
2053 nfs_access_free_list(&head
);
2058 nfs_access_cache_count(struct shrinker
*shrink
, struct shrink_control
*sc
)
2060 return vfs_pressure_ratio(atomic_long_read(&nfs_access_nr_entries
));
2063 static void __nfs_access_zap_cache(struct nfs_inode
*nfsi
, struct list_head
*head
)
2065 struct rb_root
*root_node
= &nfsi
->access_cache
;
2067 struct nfs_access_entry
*entry
;
2069 /* Unhook entries from the cache */
2070 while ((n
= rb_first(root_node
)) != NULL
) {
2071 entry
= rb_entry(n
, struct nfs_access_entry
, rb_node
);
2072 rb_erase(n
, root_node
);
2073 list_move(&entry
->lru
, head
);
2075 nfsi
->cache_validity
&= ~NFS_INO_INVALID_ACCESS
;
2078 void nfs_access_zap_cache(struct inode
*inode
)
2082 if (test_bit(NFS_INO_ACL_LRU_SET
, &NFS_I(inode
)->flags
) == 0)
2084 /* Remove from global LRU init */
2085 spin_lock(&nfs_access_lru_lock
);
2086 if (test_and_clear_bit(NFS_INO_ACL_LRU_SET
, &NFS_I(inode
)->flags
))
2087 list_del_init(&NFS_I(inode
)->access_cache_inode_lru
);
2089 spin_lock(&inode
->i_lock
);
2090 __nfs_access_zap_cache(NFS_I(inode
), &head
);
2091 spin_unlock(&inode
->i_lock
);
2092 spin_unlock(&nfs_access_lru_lock
);
2093 nfs_access_free_list(&head
);
2095 EXPORT_SYMBOL_GPL(nfs_access_zap_cache
);
2097 static struct nfs_access_entry
*nfs_access_search_rbtree(struct inode
*inode
, struct rpc_cred
*cred
)
2099 struct rb_node
*n
= NFS_I(inode
)->access_cache
.rb_node
;
2100 struct nfs_access_entry
*entry
;
2103 entry
= rb_entry(n
, struct nfs_access_entry
, rb_node
);
2105 if (cred
< entry
->cred
)
2107 else if (cred
> entry
->cred
)
2115 static int nfs_access_get_cached(struct inode
*inode
, struct rpc_cred
*cred
, struct nfs_access_entry
*res
)
2117 struct nfs_inode
*nfsi
= NFS_I(inode
);
2118 struct nfs_access_entry
*cache
;
2121 spin_lock(&inode
->i_lock
);
2122 if (nfsi
->cache_validity
& NFS_INO_INVALID_ACCESS
)
2124 cache
= nfs_access_search_rbtree(inode
, cred
);
2127 if (!nfs_have_delegated_attributes(inode
) &&
2128 !time_in_range_open(jiffies
, cache
->jiffies
, cache
->jiffies
+ nfsi
->attrtimeo
))
2130 res
->jiffies
= cache
->jiffies
;
2131 res
->cred
= cache
->cred
;
2132 res
->mask
= cache
->mask
;
2133 list_move_tail(&cache
->lru
, &nfsi
->access_cache_entry_lru
);
2136 spin_unlock(&inode
->i_lock
);
2139 rb_erase(&cache
->rb_node
, &nfsi
->access_cache
);
2140 list_del(&cache
->lru
);
2141 spin_unlock(&inode
->i_lock
);
2142 nfs_access_free_entry(cache
);
2145 spin_unlock(&inode
->i_lock
);
2146 nfs_access_zap_cache(inode
);
2150 static void nfs_access_add_rbtree(struct inode
*inode
, struct nfs_access_entry
*set
)
2152 struct nfs_inode
*nfsi
= NFS_I(inode
);
2153 struct rb_root
*root_node
= &nfsi
->access_cache
;
2154 struct rb_node
**p
= &root_node
->rb_node
;
2155 struct rb_node
*parent
= NULL
;
2156 struct nfs_access_entry
*entry
;
2158 spin_lock(&inode
->i_lock
);
2159 while (*p
!= NULL
) {
2161 entry
= rb_entry(parent
, struct nfs_access_entry
, rb_node
);
2163 if (set
->cred
< entry
->cred
)
2164 p
= &parent
->rb_left
;
2165 else if (set
->cred
> entry
->cred
)
2166 p
= &parent
->rb_right
;
2170 rb_link_node(&set
->rb_node
, parent
, p
);
2171 rb_insert_color(&set
->rb_node
, root_node
);
2172 list_add_tail(&set
->lru
, &nfsi
->access_cache_entry_lru
);
2173 spin_unlock(&inode
->i_lock
);
2176 rb_replace_node(parent
, &set
->rb_node
, root_node
);
2177 list_add_tail(&set
->lru
, &nfsi
->access_cache_entry_lru
);
2178 list_del(&entry
->lru
);
2179 spin_unlock(&inode
->i_lock
);
2180 nfs_access_free_entry(entry
);
2183 void nfs_access_add_cache(struct inode
*inode
, struct nfs_access_entry
*set
)
2185 struct nfs_access_entry
*cache
= kmalloc(sizeof(*cache
), GFP_KERNEL
);
2188 RB_CLEAR_NODE(&cache
->rb_node
);
2189 cache
->jiffies
= set
->jiffies
;
2190 cache
->cred
= get_rpccred(set
->cred
);
2191 cache
->mask
= set
->mask
;
2193 nfs_access_add_rbtree(inode
, cache
);
2195 /* Update accounting */
2196 smp_mb__before_atomic_inc();
2197 atomic_long_inc(&nfs_access_nr_entries
);
2198 smp_mb__after_atomic_inc();
2200 /* Add inode to global LRU list */
2201 if (!test_bit(NFS_INO_ACL_LRU_SET
, &NFS_I(inode
)->flags
)) {
2202 spin_lock(&nfs_access_lru_lock
);
2203 if (!test_and_set_bit(NFS_INO_ACL_LRU_SET
, &NFS_I(inode
)->flags
))
2204 list_add_tail(&NFS_I(inode
)->access_cache_inode_lru
,
2205 &nfs_access_lru_list
);
2206 spin_unlock(&nfs_access_lru_lock
);
2209 EXPORT_SYMBOL_GPL(nfs_access_add_cache
);
2211 void nfs_access_set_mask(struct nfs_access_entry
*entry
, u32 access_result
)
2214 if (access_result
& NFS4_ACCESS_READ
)
2215 entry
->mask
|= MAY_READ
;
2217 (NFS4_ACCESS_MODIFY
| NFS4_ACCESS_EXTEND
| NFS4_ACCESS_DELETE
))
2218 entry
->mask
|= MAY_WRITE
;
2219 if (access_result
& (NFS4_ACCESS_LOOKUP
|NFS4_ACCESS_EXECUTE
))
2220 entry
->mask
|= MAY_EXEC
;
2222 EXPORT_SYMBOL_GPL(nfs_access_set_mask
);
2224 static int nfs_do_access(struct inode
*inode
, struct rpc_cred
*cred
, int mask
)
2226 struct nfs_access_entry cache
;
2229 trace_nfs_access_enter(inode
);
2231 status
= nfs_access_get_cached(inode
, cred
, &cache
);
2235 /* Be clever: ask server to check for all possible rights */
2236 cache
.mask
= MAY_EXEC
| MAY_WRITE
| MAY_READ
;
2238 cache
.jiffies
= jiffies
;
2239 status
= NFS_PROTO(inode
)->access(inode
, &cache
);
2241 if (status
== -ESTALE
) {
2242 nfs_zap_caches(inode
);
2243 if (!S_ISDIR(inode
->i_mode
))
2244 set_bit(NFS_INO_STALE
, &NFS_I(inode
)->flags
);
2248 nfs_access_add_cache(inode
, &cache
);
2250 if ((mask
& ~cache
.mask
& (MAY_READ
| MAY_WRITE
| MAY_EXEC
)) != 0)
2253 trace_nfs_access_exit(inode
, status
);
2257 static int nfs_open_permission_mask(int openflags
)
2261 if (openflags
& __FMODE_EXEC
) {
2262 /* ONLY check exec rights */
2265 if ((openflags
& O_ACCMODE
) != O_WRONLY
)
2267 if ((openflags
& O_ACCMODE
) != O_RDONLY
)
2274 int nfs_may_open(struct inode
*inode
, struct rpc_cred
*cred
, int openflags
)
2276 return nfs_do_access(inode
, cred
, nfs_open_permission_mask(openflags
));
2278 EXPORT_SYMBOL_GPL(nfs_may_open
);
2280 int nfs_permission(struct inode
*inode
, int mask
)
2282 struct rpc_cred
*cred
;
2285 if (mask
& MAY_NOT_BLOCK
)
2288 nfs_inc_stats(inode
, NFSIOS_VFSACCESS
);
2290 if ((mask
& (MAY_READ
| MAY_WRITE
| MAY_EXEC
)) == 0)
2292 /* Is this sys_access() ? */
2293 if (mask
& (MAY_ACCESS
| MAY_CHDIR
))
2296 switch (inode
->i_mode
& S_IFMT
) {
2303 * Optimize away all write operations, since the server
2304 * will check permissions when we perform the op.
2306 if ((mask
& MAY_WRITE
) && !(mask
& MAY_READ
))
2311 if (!NFS_PROTO(inode
)->access
)
2314 cred
= rpc_lookup_cred();
2315 if (!IS_ERR(cred
)) {
2316 res
= nfs_do_access(inode
, cred
, mask
);
2319 res
= PTR_ERR(cred
);
2321 if (!res
&& (mask
& MAY_EXEC
) && !execute_ok(inode
))
2324 dfprintk(VFS
, "NFS: permission(%s/%ld), mask=0x%x, res=%d\n",
2325 inode
->i_sb
->s_id
, inode
->i_ino
, mask
, res
);
2328 res
= nfs_revalidate_inode(NFS_SERVER(inode
), inode
);
2330 res
= generic_permission(inode
, mask
);
2333 EXPORT_SYMBOL_GPL(nfs_permission
);
2337 * version-control: t
2338 * kept-new-versions: 5