1 // SPDX-License-Identifier: GPL-2.0-only
5 * Copyright (C) 1992 Rick Sladkey
7 * nfs directory handling functions
9 * 10 Apr 1996 Added silly rename for unlink --okir
10 * 28 Sep 1996 Improved directory cache --okir
11 * 23 Aug 1997 Claus Heine claus@momo.math.rwth-aachen.de
12 * Re-implemented silly rename for unlink, newly implemented
13 * silly rename for nfs_rename() following the suggestions
14 * of Olaf Kirch (okir) found in this file.
15 * Following Linus comments on my original hack, this version
16 * depends only on the dcache stuff and doesn't touch the inode
17 * layer (iput() and friends).
18 * 6 Jun 1999 Cache readdir lookups in the page cache. -DaveM
21 #include <linux/module.h>
22 #include <linux/time.h>
23 #include <linux/errno.h>
24 #include <linux/stat.h>
25 #include <linux/fcntl.h>
26 #include <linux/string.h>
27 #include <linux/kernel.h>
28 #include <linux/slab.h>
30 #include <linux/sunrpc/clnt.h>
31 #include <linux/nfs_fs.h>
32 #include <linux/nfs_mount.h>
33 #include <linux/pagemap.h>
34 #include <linux/pagevec.h>
35 #include <linux/namei.h>
36 #include <linux/mount.h>
37 #include <linux/swap.h>
38 #include <linux/sched.h>
39 #include <linux/kmemleak.h>
40 #include <linux/xattr.h>
42 #include "delegation.h"
49 /* #define NFS_DEBUG_VERBOSE 1 */
51 static int nfs_opendir(struct inode
*, struct file
*);
52 static int nfs_closedir(struct inode
*, struct file
*);
53 static int nfs_readdir(struct file
*, struct dir_context
*);
54 static int nfs_fsync_dir(struct file
*, loff_t
, loff_t
, int);
55 static loff_t
nfs_llseek_dir(struct file
*, loff_t
, int);
56 static void nfs_readdir_clear_array(struct page
*);
58 const struct file_operations nfs_dir_operations
= {
59 .llseek
= nfs_llseek_dir
,
60 .read
= generic_read_dir
,
61 .iterate
= nfs_readdir
,
63 .release
= nfs_closedir
,
64 .fsync
= nfs_fsync_dir
,
67 const struct address_space_operations nfs_dir_aops
= {
68 .freepage
= nfs_readdir_clear_array
,
71 static struct nfs_open_dir_context
*alloc_nfs_open_dir_context(struct inode
*dir
, const struct cred
*cred
)
73 struct nfs_inode
*nfsi
= NFS_I(dir
);
74 struct nfs_open_dir_context
*ctx
;
75 ctx
= kmalloc(sizeof(*ctx
), GFP_KERNEL
);
78 ctx
->attr_gencount
= nfsi
->attr_gencount
;
81 ctx
->cred
= get_cred(cred
);
82 spin_lock(&dir
->i_lock
);
83 if (list_empty(&nfsi
->open_files
) &&
84 (nfsi
->cache_validity
& NFS_INO_DATA_INVAL_DEFER
))
85 nfsi
->cache_validity
|= NFS_INO_INVALID_DATA
|
87 list_add(&ctx
->list
, &nfsi
->open_files
);
88 spin_unlock(&dir
->i_lock
);
91 return ERR_PTR(-ENOMEM
);
94 static void put_nfs_open_dir_context(struct inode
*dir
, struct nfs_open_dir_context
*ctx
)
96 spin_lock(&dir
->i_lock
);
98 spin_unlock(&dir
->i_lock
);
107 nfs_opendir(struct inode
*inode
, struct file
*filp
)
110 struct nfs_open_dir_context
*ctx
;
112 dfprintk(FILE, "NFS: open dir(%pD2)\n", filp
);
114 nfs_inc_stats(inode
, NFSIOS_VFSOPEN
);
116 ctx
= alloc_nfs_open_dir_context(inode
, current_cred());
121 filp
->private_data
= ctx
;
127 nfs_closedir(struct inode
*inode
, struct file
*filp
)
129 put_nfs_open_dir_context(file_inode(filp
), 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
*, bool);
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 * we are freeing strings created by nfs_add_to_readdir_array()
169 void nfs_readdir_clear_array(struct page
*page
)
171 struct nfs_cache_array
*array
;
174 array
= kmap_atomic(page
);
175 for (i
= 0; i
< array
->size
; i
++)
176 kfree(array
->array
[i
].string
.name
);
177 kunmap_atomic(array
);
181 * the caller is responsible for freeing qstr.name
182 * when called by nfs_readdir_add_to_array, the strings will be freed in
183 * nfs_clear_readdir_array()
186 int nfs_readdir_make_qstr(struct qstr
*string
, const char *name
, unsigned int len
)
189 string
->name
= kmemdup(name
, len
, GFP_KERNEL
);
190 if (string
->name
== NULL
)
193 * Avoid a kmemleak false positive. The pointer to the name is stored
194 * in a page cache page which kmemleak does not scan.
196 kmemleak_not_leak(string
->name
);
197 string
->hash
= full_name_hash(NULL
, name
, len
);
202 int nfs_readdir_add_to_array(struct nfs_entry
*entry
, struct page
*page
)
204 struct nfs_cache_array
*array
= kmap(page
);
205 struct nfs_cache_array_entry
*cache_entry
;
208 cache_entry
= &array
->array
[array
->size
];
210 /* Check that this entry lies within the page bounds */
212 if ((char *)&cache_entry
[1] - (char *)page_address(page
) > PAGE_SIZE
)
215 cache_entry
->cookie
= entry
->prev_cookie
;
216 cache_entry
->ino
= entry
->ino
;
217 cache_entry
->d_type
= entry
->d_type
;
218 ret
= nfs_readdir_make_qstr(&cache_entry
->string
, entry
->name
, entry
->len
);
221 array
->last_cookie
= entry
->cookie
;
224 array
->eof_index
= array
->size
;
231 int nfs_readdir_search_for_pos(struct nfs_cache_array
*array
, nfs_readdir_descriptor_t
*desc
)
233 loff_t diff
= desc
->ctx
->pos
- desc
->current_index
;
238 if (diff
>= array
->size
) {
239 if (array
->eof_index
>= 0)
244 index
= (unsigned int)diff
;
245 *desc
->dir_cookie
= array
->array
[index
].cookie
;
246 desc
->cache_entry_index
= index
;
254 nfs_readdir_inode_mapping_valid(struct nfs_inode
*nfsi
)
256 if (nfsi
->cache_validity
& (NFS_INO_INVALID_ATTR
|NFS_INO_INVALID_DATA
))
259 return !test_bit(NFS_INO_INVALIDATING
, &nfsi
->flags
);
263 int nfs_readdir_search_for_cookie(struct nfs_cache_array
*array
, nfs_readdir_descriptor_t
*desc
)
267 int status
= -EAGAIN
;
269 for (i
= 0; i
< array
->size
; i
++) {
270 if (array
->array
[i
].cookie
== *desc
->dir_cookie
) {
271 struct nfs_inode
*nfsi
= NFS_I(file_inode(desc
->file
));
272 struct nfs_open_dir_context
*ctx
= desc
->file
->private_data
;
274 new_pos
= desc
->current_index
+ i
;
275 if (ctx
->attr_gencount
!= nfsi
->attr_gencount
||
276 !nfs_readdir_inode_mapping_valid(nfsi
)) {
278 ctx
->attr_gencount
= nfsi
->attr_gencount
;
279 } else if (new_pos
< desc
->ctx
->pos
) {
281 && ctx
->dup_cookie
== *desc
->dir_cookie
) {
282 if (printk_ratelimit()) {
283 pr_notice("NFS: directory %pD2 contains a readdir loop."
284 "Please contact your server vendor. "
285 "The file: %.*s has duplicate cookie %llu\n",
286 desc
->file
, array
->array
[i
].string
.len
,
287 array
->array
[i
].string
.name
, *desc
->dir_cookie
);
292 ctx
->dup_cookie
= *desc
->dir_cookie
;
295 desc
->ctx
->pos
= new_pos
;
296 desc
->cache_entry_index
= i
;
300 if (array
->eof_index
>= 0) {
301 status
= -EBADCOOKIE
;
302 if (*desc
->dir_cookie
== array
->last_cookie
)
310 int nfs_readdir_search_array(nfs_readdir_descriptor_t
*desc
)
312 struct nfs_cache_array
*array
;
315 array
= kmap(desc
->page
);
317 if (*desc
->dir_cookie
== 0)
318 status
= nfs_readdir_search_for_pos(array
, desc
);
320 status
= nfs_readdir_search_for_cookie(array
, desc
);
322 if (status
== -EAGAIN
) {
323 desc
->last_cookie
= array
->last_cookie
;
324 desc
->current_index
+= array
->size
;
331 /* Fill a page with xdr information before transferring to the cache page */
333 int nfs_readdir_xdr_filler(struct page
**pages
, nfs_readdir_descriptor_t
*desc
,
334 struct nfs_entry
*entry
, struct file
*file
, struct inode
*inode
)
336 struct nfs_open_dir_context
*ctx
= file
->private_data
;
337 const struct cred
*cred
= ctx
->cred
;
338 unsigned long timestamp
, gencount
;
343 gencount
= nfs_inc_attr_generation_counter();
344 error
= NFS_PROTO(inode
)->readdir(file_dentry(file
), cred
, entry
->cookie
, pages
,
345 NFS_SERVER(inode
)->dtsize
, desc
->plus
);
347 /* We requested READDIRPLUS, but the server doesn't grok it */
348 if (error
== -ENOTSUPP
&& desc
->plus
) {
349 NFS_SERVER(inode
)->caps
&= ~NFS_CAP_READDIRPLUS
;
350 clear_bit(NFS_INO_ADVISE_RDPLUS
, &NFS_I(inode
)->flags
);
356 desc
->timestamp
= timestamp
;
357 desc
->gencount
= gencount
;
362 static int xdr_decode(nfs_readdir_descriptor_t
*desc
,
363 struct nfs_entry
*entry
, struct xdr_stream
*xdr
)
367 error
= desc
->decode(xdr
, entry
, desc
->plus
);
370 entry
->fattr
->time_start
= desc
->timestamp
;
371 entry
->fattr
->gencount
= desc
->gencount
;
375 /* Match file and dirent using either filehandle or fileid
376 * Note: caller is responsible for checking the fsid
379 int nfs_same_file(struct dentry
*dentry
, struct nfs_entry
*entry
)
382 struct nfs_inode
*nfsi
;
384 if (d_really_is_negative(dentry
))
387 inode
= d_inode(dentry
);
388 if (is_bad_inode(inode
) || NFS_STALE(inode
))
392 if (entry
->fattr
->fileid
!= nfsi
->fileid
)
394 if (entry
->fh
->size
&& nfs_compare_fh(entry
->fh
, &nfsi
->fh
) != 0)
400 bool nfs_use_readdirplus(struct inode
*dir
, struct dir_context
*ctx
)
402 if (!nfs_server_capable(dir
, NFS_CAP_READDIRPLUS
))
404 if (test_and_clear_bit(NFS_INO_ADVISE_RDPLUS
, &NFS_I(dir
)->flags
))
412 * This function is called by the lookup and getattr code to request the
413 * use of readdirplus to accelerate any future lookups in the same
416 void nfs_advise_use_readdirplus(struct inode
*dir
)
418 struct nfs_inode
*nfsi
= NFS_I(dir
);
420 if (nfs_server_capable(dir
, NFS_CAP_READDIRPLUS
) &&
421 !list_empty(&nfsi
->open_files
))
422 set_bit(NFS_INO_ADVISE_RDPLUS
, &nfsi
->flags
);
426 * This function is mainly for use by nfs_getattr().
428 * If this is an 'ls -l', we want to force use of readdirplus.
429 * Do this by checking if there is an active file descriptor
430 * and calling nfs_advise_use_readdirplus, then forcing a
433 void nfs_force_use_readdirplus(struct inode
*dir
)
435 struct nfs_inode
*nfsi
= NFS_I(dir
);
437 if (nfs_server_capable(dir
, NFS_CAP_READDIRPLUS
) &&
438 !list_empty(&nfsi
->open_files
)) {
439 set_bit(NFS_INO_ADVISE_RDPLUS
, &nfsi
->flags
);
440 invalidate_mapping_pages(dir
->i_mapping
, 0, -1);
445 void nfs_prime_dcache(struct dentry
*parent
, struct nfs_entry
*entry
)
447 struct qstr filename
= QSTR_INIT(entry
->name
, entry
->len
);
448 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq
);
449 struct dentry
*dentry
;
450 struct dentry
*alias
;
451 struct inode
*dir
= d_inode(parent
);
455 if (!(entry
->fattr
->valid
& NFS_ATTR_FATTR_FILEID
))
457 if (!(entry
->fattr
->valid
& NFS_ATTR_FATTR_FSID
))
459 if (filename
.len
== 0)
461 /* Validate that the name doesn't contain any illegal '\0' */
462 if (strnlen(filename
.name
, filename
.len
) != filename
.len
)
465 if (strnchr(filename
.name
, filename
.len
, '/'))
467 if (filename
.name
[0] == '.') {
468 if (filename
.len
== 1)
470 if (filename
.len
== 2 && filename
.name
[1] == '.')
473 filename
.hash
= full_name_hash(parent
, filename
.name
, filename
.len
);
475 dentry
= d_lookup(parent
, &filename
);
478 dentry
= d_alloc_parallel(parent
, &filename
, &wq
);
482 if (!d_in_lookup(dentry
)) {
483 /* Is there a mountpoint here? If so, just exit */
484 if (!nfs_fsid_equal(&NFS_SB(dentry
->d_sb
)->fsid
,
485 &entry
->fattr
->fsid
))
487 if (nfs_same_file(dentry
, entry
)) {
488 if (!entry
->fh
->size
)
490 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
491 status
= nfs_refresh_inode(d_inode(dentry
), entry
->fattr
);
493 nfs_setsecurity(d_inode(dentry
), entry
->fattr
, entry
->label
);
496 d_invalidate(dentry
);
502 if (!entry
->fh
->size
) {
503 d_lookup_done(dentry
);
507 inode
= nfs_fhget(dentry
->d_sb
, entry
->fh
, entry
->fattr
, entry
->label
);
508 alias
= d_splice_alias(inode
, dentry
);
509 d_lookup_done(dentry
);
516 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
521 /* Perform conversion from xdr to cache array */
523 int nfs_readdir_page_filler(nfs_readdir_descriptor_t
*desc
, struct nfs_entry
*entry
,
524 struct page
**xdr_pages
, struct page
*page
, unsigned int buflen
)
526 struct xdr_stream stream
;
528 struct page
*scratch
;
529 struct nfs_cache_array
*array
;
530 unsigned int count
= 0;
533 scratch
= alloc_page(GFP_KERNEL
);
540 xdr_init_decode_pages(&stream
, &buf
, xdr_pages
, buflen
);
541 xdr_set_scratch_buffer(&stream
, page_address(scratch
), PAGE_SIZE
);
544 status
= xdr_decode(desc
, entry
, &stream
);
546 if (status
== -EAGAIN
)
554 nfs_prime_dcache(file_dentry(desc
->file
), entry
);
556 status
= nfs_readdir_add_to_array(entry
, page
);
559 } while (!entry
->eof
);
562 if (count
== 0 || (status
== -EBADCOOKIE
&& entry
->eof
!= 0)) {
564 array
->eof_index
= array
->size
;
574 void nfs_readdir_free_pages(struct page
**pages
, unsigned int npages
)
577 for (i
= 0; i
< npages
; i
++)
582 * nfs_readdir_alloc_pages() will allocate pages that must be freed with a call
583 * to nfs_readdir_free_pages()
586 int nfs_readdir_alloc_pages(struct page
**pages
, unsigned int npages
)
590 for (i
= 0; i
< npages
; i
++) {
591 struct page
*page
= alloc_page(GFP_KERNEL
);
599 nfs_readdir_free_pages(pages
, i
);
604 int nfs_readdir_xdr_to_array(nfs_readdir_descriptor_t
*desc
, struct page
*page
, struct inode
*inode
)
606 struct page
*pages
[NFS_MAX_READDIR_PAGES
];
607 struct nfs_entry entry
;
608 struct file
*file
= desc
->file
;
609 struct nfs_cache_array
*array
;
610 int status
= -ENOMEM
;
611 unsigned int array_size
= ARRAY_SIZE(pages
);
613 entry
.prev_cookie
= 0;
614 entry
.cookie
= desc
->last_cookie
;
616 entry
.fh
= nfs_alloc_fhandle();
617 entry
.fattr
= nfs_alloc_fattr();
618 entry
.server
= NFS_SERVER(inode
);
619 if (entry
.fh
== NULL
|| entry
.fattr
== NULL
)
622 entry
.label
= nfs4_label_alloc(NFS_SERVER(inode
), GFP_NOWAIT
);
623 if (IS_ERR(entry
.label
)) {
624 status
= PTR_ERR(entry
.label
);
629 memset(array
, 0, sizeof(struct nfs_cache_array
));
630 array
->eof_index
= -1;
632 status
= nfs_readdir_alloc_pages(pages
, array_size
);
634 goto out_release_array
;
637 status
= nfs_readdir_xdr_filler(pages
, desc
, &entry
, file
, inode
);
642 status
= nfs_readdir_page_filler(desc
, &entry
, pages
, page
, pglen
);
644 if (status
== -ENOSPC
)
648 } while (array
->eof_index
< 0);
650 nfs_readdir_free_pages(pages
, array_size
);
653 nfs4_label_free(entry
.label
);
655 nfs_free_fattr(entry
.fattr
);
656 nfs_free_fhandle(entry
.fh
);
661 * Now we cache directories properly, by converting xdr information
662 * to an array that can be used for lookups later. This results in
663 * fewer cache pages, since we can store more information on each page.
664 * We only need to convert from xdr once so future lookups are much simpler
667 int nfs_readdir_filler(void *data
, struct page
* page
)
669 nfs_readdir_descriptor_t
*desc
= data
;
670 struct inode
*inode
= file_inode(desc
->file
);
673 ret
= nfs_readdir_xdr_to_array(desc
, page
, inode
);
676 SetPageUptodate(page
);
678 if (invalidate_inode_pages2_range(inode
->i_mapping
, page
->index
+ 1, -1) < 0) {
679 /* Should never happen */
680 nfs_zap_mapping(inode
, inode
->i_mapping
);
690 void cache_page_release(nfs_readdir_descriptor_t
*desc
)
692 if (!desc
->page
->mapping
)
693 nfs_readdir_clear_array(desc
->page
);
694 put_page(desc
->page
);
699 struct page
*get_cache_page(nfs_readdir_descriptor_t
*desc
)
701 return read_cache_page(desc
->file
->f_mapping
, desc
->page_index
,
702 nfs_readdir_filler
, desc
);
706 * Returns 0 if desc->dir_cookie was found on page desc->page_index
709 int find_cache_page(nfs_readdir_descriptor_t
*desc
)
713 desc
->page
= get_cache_page(desc
);
714 if (IS_ERR(desc
->page
))
715 return PTR_ERR(desc
->page
);
717 res
= nfs_readdir_search_array(desc
);
719 cache_page_release(desc
);
723 /* Search for desc->dir_cookie from the beginning of the page cache */
725 int readdir_search_pagecache(nfs_readdir_descriptor_t
*desc
)
729 if (desc
->page_index
== 0) {
730 desc
->current_index
= 0;
731 desc
->last_cookie
= 0;
734 res
= find_cache_page(desc
);
735 } while (res
== -EAGAIN
);
740 * Once we've found the start of the dirent within a page: fill 'er up...
743 int nfs_do_filldir(nfs_readdir_descriptor_t
*desc
)
745 struct file
*file
= desc
->file
;
748 struct nfs_cache_array
*array
= NULL
;
749 struct nfs_open_dir_context
*ctx
= file
->private_data
;
751 array
= kmap(desc
->page
);
752 for (i
= desc
->cache_entry_index
; i
< array
->size
; i
++) {
753 struct nfs_cache_array_entry
*ent
;
755 ent
= &array
->array
[i
];
756 if (!dir_emit(desc
->ctx
, ent
->string
.name
, ent
->string
.len
,
757 nfs_compat_user_ino64(ent
->ino
), ent
->d_type
)) {
762 if (i
< (array
->size
-1))
763 *desc
->dir_cookie
= array
->array
[i
+1].cookie
;
765 *desc
->dir_cookie
= array
->last_cookie
;
769 if (array
->eof_index
>= 0)
773 cache_page_release(desc
);
774 dfprintk(DIRCACHE
, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n",
775 (unsigned long long)*desc
->dir_cookie
, res
);
780 * If we cannot find a cookie in our cache, we suspect that this is
781 * because it points to a deleted file, so we ask the server to return
782 * whatever it thinks is the next entry. We then feed this to filldir.
783 * If all goes well, we should then be able to find our way round the
784 * cache on the next call to readdir_search_pagecache();
786 * NOTE: we cannot add the anonymous page to the pagecache because
787 * the data it contains might not be page aligned. Besides,
788 * we should already have a complete representation of the
789 * directory in the page cache by the time we get here.
792 int uncached_readdir(nfs_readdir_descriptor_t
*desc
)
794 struct page
*page
= NULL
;
796 struct inode
*inode
= file_inode(desc
->file
);
797 struct nfs_open_dir_context
*ctx
= desc
->file
->private_data
;
799 dfprintk(DIRCACHE
, "NFS: uncached_readdir() searching for cookie %Lu\n",
800 (unsigned long long)*desc
->dir_cookie
);
802 page
= alloc_page(GFP_HIGHUSER
);
808 desc
->page_index
= 0;
809 desc
->last_cookie
= *desc
->dir_cookie
;
813 status
= nfs_readdir_xdr_to_array(desc
, page
, inode
);
817 status
= nfs_do_filldir(desc
);
820 dfprintk(DIRCACHE
, "NFS: %s: returns %d\n",
824 cache_page_release(desc
);
828 /* The file offset position represents the dirent entry number. A
829 last cookie cache takes care of the common case of reading the
832 static int nfs_readdir(struct file
*file
, struct dir_context
*ctx
)
834 struct dentry
*dentry
= file_dentry(file
);
835 struct inode
*inode
= d_inode(dentry
);
836 nfs_readdir_descriptor_t my_desc
,
838 struct nfs_open_dir_context
*dir_ctx
= file
->private_data
;
841 dfprintk(FILE, "NFS: readdir(%pD2) starting at cookie %llu\n",
842 file
, (long long)ctx
->pos
);
843 nfs_inc_stats(inode
, NFSIOS_VFSGETDENTS
);
846 * ctx->pos points to the dirent entry number.
847 * *desc->dir_cookie has the cookie for the next entry. We have
848 * to either find the entry with the appropriate number or
849 * revalidate the cookie.
851 memset(desc
, 0, sizeof(*desc
));
855 desc
->dir_cookie
= &dir_ctx
->dir_cookie
;
856 desc
->decode
= NFS_PROTO(inode
)->decode_dirent
;
857 desc
->plus
= nfs_use_readdirplus(inode
, ctx
);
859 if (ctx
->pos
== 0 || nfs_attribute_cache_expired(inode
))
860 res
= nfs_revalidate_mapping(inode
, file
->f_mapping
);
865 res
= readdir_search_pagecache(desc
);
867 if (res
== -EBADCOOKIE
) {
869 /* This means either end of directory */
870 if (*desc
->dir_cookie
&& !desc
->eof
) {
871 /* Or that the server has 'lost' a cookie */
872 res
= uncached_readdir(desc
);
878 if (res
== -ETOOSMALL
&& desc
->plus
) {
879 clear_bit(NFS_INO_ADVISE_RDPLUS
, &NFS_I(inode
)->flags
);
880 nfs_zap_caches(inode
);
881 desc
->page_index
= 0;
889 res
= nfs_do_filldir(desc
);
892 } while (!desc
->eof
);
896 dfprintk(FILE, "NFS: readdir(%pD2) returns %d\n", file
, res
);
900 static loff_t
nfs_llseek_dir(struct file
*filp
, loff_t offset
, int whence
)
902 struct inode
*inode
= file_inode(filp
);
903 struct nfs_open_dir_context
*dir_ctx
= filp
->private_data
;
905 dfprintk(FILE, "NFS: llseek dir(%pD2, %lld, %d)\n",
906 filp
, offset
, whence
);
920 offset
+= filp
->f_pos
;
926 if (offset
!= filp
->f_pos
) {
927 filp
->f_pos
= offset
;
928 dir_ctx
->dir_cookie
= 0;
936 * All directory operations under NFS are synchronous, so fsync()
937 * is a dummy operation.
939 static int nfs_fsync_dir(struct file
*filp
, loff_t start
, loff_t end
,
942 struct inode
*inode
= file_inode(filp
);
944 dfprintk(FILE, "NFS: fsync dir(%pD2) datasync %d\n", filp
, datasync
);
947 nfs_inc_stats(inode
, NFSIOS_VFSFSYNC
);
953 * nfs_force_lookup_revalidate - Mark the directory as having changed
954 * @dir: pointer to directory inode
956 * This forces the revalidation code in nfs_lookup_revalidate() to do a
957 * full lookup on all child dentries of 'dir' whenever a change occurs
958 * on the server that might have invalidated our dcache.
960 * The caller should be holding dir->i_lock
962 void nfs_force_lookup_revalidate(struct inode
*dir
)
964 NFS_I(dir
)->cache_change_attribute
++;
966 EXPORT_SYMBOL_GPL(nfs_force_lookup_revalidate
);
969 * A check for whether or not the parent directory has changed.
970 * In the case it has, we assume that the dentries are untrustworthy
971 * and may need to be looked up again.
972 * If rcu_walk prevents us from performing a full check, return 0.
974 static int nfs_check_verifier(struct inode
*dir
, struct dentry
*dentry
,
979 if (NFS_SERVER(dir
)->flags
& NFS_MOUNT_LOOKUP_CACHE_NONE
)
981 if (!nfs_verify_change_attribute(dir
, dentry
->d_time
))
983 /* Revalidate nfsi->cache_change_attribute before we declare a match */
984 if (nfs_mapping_need_revalidate_inode(dir
)) {
987 if (__nfs_revalidate_inode(NFS_SERVER(dir
), dir
) < 0)
990 if (!nfs_verify_change_attribute(dir
, dentry
->d_time
))
996 * Use intent information to check whether or not we're going to do
997 * an O_EXCL create using this path component.
999 static int nfs_is_exclusive_create(struct inode
*dir
, unsigned int flags
)
1001 if (NFS_PROTO(dir
)->version
== 2)
1003 return flags
& LOOKUP_EXCL
;
1007 * Inode and filehandle revalidation for lookups.
1009 * We force revalidation in the cases where the VFS sets LOOKUP_REVAL,
1010 * or if the intent information indicates that we're about to open this
1011 * particular file and the "nocto" mount flag is not set.
1015 int nfs_lookup_verify_inode(struct inode
*inode
, unsigned int flags
)
1017 struct nfs_server
*server
= NFS_SERVER(inode
);
1020 if (IS_AUTOMOUNT(inode
))
1023 if (flags
& LOOKUP_OPEN
) {
1024 switch (inode
->i_mode
& S_IFMT
) {
1026 /* A NFSv4 OPEN will revalidate later */
1027 if (server
->caps
& NFS_CAP_ATOMIC_OPEN
)
1031 if (server
->flags
& NFS_MOUNT_NOCTO
)
1033 /* NFS close-to-open cache consistency validation */
1038 /* VFS wants an on-the-wire revalidation */
1039 if (flags
& LOOKUP_REVAL
)
1042 return (inode
->i_nlink
== 0) ? -ESTALE
: 0;
1044 if (flags
& LOOKUP_RCU
)
1046 ret
= __nfs_revalidate_inode(server
, inode
);
1053 * We judge how long we want to trust negative
1054 * dentries by looking at the parent inode mtime.
1056 * If parent mtime has changed, we revalidate, else we wait for a
1057 * period corresponding to the parent's attribute cache timeout value.
1059 * If LOOKUP_RCU prevents us from performing a full check, return 1
1060 * suggesting a reval is needed.
1062 * Note that when creating a new file, or looking up a rename target,
1063 * then it shouldn't be necessary to revalidate a negative dentry.
1066 int nfs_neg_need_reval(struct inode
*dir
, struct dentry
*dentry
,
1069 if (flags
& (LOOKUP_CREATE
| LOOKUP_RENAME_TARGET
))
1071 if (NFS_SERVER(dir
)->flags
& NFS_MOUNT_LOOKUP_CACHE_NONEG
)
1073 return !nfs_check_verifier(dir
, dentry
, flags
& LOOKUP_RCU
);
1077 nfs_lookup_revalidate_done(struct inode
*dir
, struct dentry
*dentry
,
1078 struct inode
*inode
, int error
)
1082 dfprintk(LOOKUPCACHE
, "NFS: %s(%pd2) is valid\n",
1086 nfs_mark_for_revalidate(dir
);
1087 if (inode
&& S_ISDIR(inode
->i_mode
)) {
1088 /* Purge readdir caches. */
1089 nfs_zap_caches(inode
);
1091 * We can't d_drop the root of a disconnected tree:
1092 * its d_hash is on the s_anon list and d_drop() would hide
1093 * it from shrink_dcache_for_unmount(), leading to busy
1094 * inodes on unmount and further oopses.
1096 if (IS_ROOT(dentry
))
1099 dfprintk(LOOKUPCACHE
, "NFS: %s(%pd2) is invalid\n",
1103 dfprintk(LOOKUPCACHE
, "NFS: %s(%pd2) lookup returned error %d\n",
1104 __func__
, dentry
, error
);
1109 nfs_lookup_revalidate_negative(struct inode
*dir
, struct dentry
*dentry
,
1113 if (nfs_neg_need_reval(dir
, dentry
, flags
)) {
1114 if (flags
& LOOKUP_RCU
)
1118 return nfs_lookup_revalidate_done(dir
, dentry
, NULL
, ret
);
1122 nfs_lookup_revalidate_delegated(struct inode
*dir
, struct dentry
*dentry
,
1123 struct inode
*inode
)
1125 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1126 return nfs_lookup_revalidate_done(dir
, dentry
, inode
, 1);
1130 nfs_lookup_revalidate_dentry(struct inode
*dir
, struct dentry
*dentry
,
1131 struct inode
*inode
)
1133 struct nfs_fh
*fhandle
;
1134 struct nfs_fattr
*fattr
;
1135 struct nfs4_label
*label
;
1139 fhandle
= nfs_alloc_fhandle();
1140 fattr
= nfs_alloc_fattr();
1141 label
= nfs4_label_alloc(NFS_SERVER(inode
), GFP_KERNEL
);
1142 if (fhandle
== NULL
|| fattr
== NULL
|| IS_ERR(label
))
1145 ret
= NFS_PROTO(dir
)->lookup(dir
, &dentry
->d_name
, fhandle
, fattr
, label
);
1147 if (ret
== -ESTALE
|| ret
== -ENOENT
)
1152 if (nfs_compare_fh(NFS_FH(inode
), fhandle
))
1154 if (nfs_refresh_inode(inode
, fattr
) < 0)
1157 nfs_setsecurity(inode
, fattr
, label
);
1158 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1160 /* set a readdirplus hint that we had a cache miss */
1161 nfs_force_use_readdirplus(dir
);
1164 nfs_free_fattr(fattr
);
1165 nfs_free_fhandle(fhandle
);
1166 nfs4_label_free(label
);
1167 return nfs_lookup_revalidate_done(dir
, dentry
, inode
, ret
);
1171 * This is called every time the dcache has a lookup hit,
1172 * and we should check whether we can really trust that
1175 * NOTE! The hit can be a negative hit too, don't assume
1178 * If the parent directory is seen to have changed, we throw out the
1179 * cached dentry and do a new lookup.
1182 nfs_do_lookup_revalidate(struct inode
*dir
, struct dentry
*dentry
,
1185 struct inode
*inode
;
1188 nfs_inc_stats(dir
, NFSIOS_DENTRYREVALIDATE
);
1189 inode
= d_inode(dentry
);
1192 return nfs_lookup_revalidate_negative(dir
, dentry
, flags
);
1194 if (is_bad_inode(inode
)) {
1195 dfprintk(LOOKUPCACHE
, "%s: %pd2 has dud inode\n",
1200 if (NFS_PROTO(dir
)->have_delegation(inode
, FMODE_READ
))
1201 return nfs_lookup_revalidate_delegated(dir
, dentry
, inode
);
1203 /* Force a full look up iff the parent directory has changed */
1204 if (!(flags
& (LOOKUP_EXCL
| LOOKUP_REVAL
)) &&
1205 nfs_check_verifier(dir
, dentry
, flags
& LOOKUP_RCU
)) {
1206 error
= nfs_lookup_verify_inode(inode
, flags
);
1208 if (error
== -ESTALE
)
1209 nfs_zap_caches(dir
);
1212 nfs_advise_use_readdirplus(dir
);
1216 if (flags
& LOOKUP_RCU
)
1219 if (NFS_STALE(inode
))
1222 trace_nfs_lookup_revalidate_enter(dir
, dentry
, flags
);
1223 error
= nfs_lookup_revalidate_dentry(dir
, dentry
, inode
);
1224 trace_nfs_lookup_revalidate_exit(dir
, dentry
, flags
, error
);
1227 return nfs_lookup_revalidate_done(dir
, dentry
, inode
, 1);
1229 if (flags
& LOOKUP_RCU
)
1231 return nfs_lookup_revalidate_done(dir
, dentry
, inode
, 0);
1235 __nfs_lookup_revalidate(struct dentry
*dentry
, unsigned int flags
,
1236 int (*reval
)(struct inode
*, struct dentry
*, unsigned int))
1238 struct dentry
*parent
;
1242 if (flags
& LOOKUP_RCU
) {
1243 parent
= READ_ONCE(dentry
->d_parent
);
1244 dir
= d_inode_rcu(parent
);
1247 ret
= reval(dir
, dentry
, flags
);
1248 if (parent
!= READ_ONCE(dentry
->d_parent
))
1251 parent
= dget_parent(dentry
);
1252 ret
= reval(d_inode(parent
), dentry
, flags
);
1258 static int nfs_lookup_revalidate(struct dentry
*dentry
, unsigned int flags
)
1260 return __nfs_lookup_revalidate(dentry
, flags
, nfs_do_lookup_revalidate
);
1264 * A weaker form of d_revalidate for revalidating just the d_inode(dentry)
1265 * when we don't really care about the dentry name. This is called when a
1266 * pathwalk ends on a dentry that was not found via a normal lookup in the
1267 * parent dir (e.g.: ".", "..", procfs symlinks or mountpoint traversals).
1269 * In this situation, we just want to verify that the inode itself is OK
1270 * since the dentry might have changed on the server.
1272 static int nfs_weak_revalidate(struct dentry
*dentry
, unsigned int flags
)
1274 struct inode
*inode
= d_inode(dentry
);
1278 * I believe we can only get a negative dentry here in the case of a
1279 * procfs-style symlink. Just assume it's correct for now, but we may
1280 * eventually need to do something more here.
1283 dfprintk(LOOKUPCACHE
, "%s: %pd2 has negative inode\n",
1288 if (is_bad_inode(inode
)) {
1289 dfprintk(LOOKUPCACHE
, "%s: %pd2 has dud inode\n",
1294 error
= nfs_lookup_verify_inode(inode
, flags
);
1295 dfprintk(LOOKUPCACHE
, "NFS: %s: inode %lu is %s\n",
1296 __func__
, inode
->i_ino
, error
? "invalid" : "valid");
1301 * This is called from dput() when d_count is going to 0.
1303 static int nfs_dentry_delete(const struct dentry
*dentry
)
1305 dfprintk(VFS
, "NFS: dentry_delete(%pd2, %x)\n",
1306 dentry
, dentry
->d_flags
);
1308 /* Unhash any dentry with a stale inode */
1309 if (d_really_is_positive(dentry
) && NFS_STALE(d_inode(dentry
)))
1312 if (dentry
->d_flags
& DCACHE_NFSFS_RENAMED
) {
1313 /* Unhash it, so that ->d_iput() would be called */
1316 if (!(dentry
->d_sb
->s_flags
& SB_ACTIVE
)) {
1317 /* Unhash it, so that ancestors of killed async unlink
1318 * files will be cleaned up during umount */
1325 /* Ensure that we revalidate inode->i_nlink */
1326 static void nfs_drop_nlink(struct inode
*inode
)
1328 spin_lock(&inode
->i_lock
);
1329 /* drop the inode if we're reasonably sure this is the last link */
1330 if (inode
->i_nlink
> 0)
1332 NFS_I(inode
)->attr_gencount
= nfs_inc_attr_generation_counter();
1333 NFS_I(inode
)->cache_validity
|= NFS_INO_INVALID_CHANGE
1334 | NFS_INO_INVALID_CTIME
1335 | NFS_INO_INVALID_OTHER
1336 | NFS_INO_REVAL_FORCED
;
1337 spin_unlock(&inode
->i_lock
);
1341 * Called when the dentry loses inode.
1342 * We use it to clean up silly-renamed files.
1344 static void nfs_dentry_iput(struct dentry
*dentry
, struct inode
*inode
)
1346 if (S_ISDIR(inode
->i_mode
))
1347 /* drop any readdir cache as it could easily be old */
1348 NFS_I(inode
)->cache_validity
|= NFS_INO_INVALID_DATA
;
1350 if (dentry
->d_flags
& DCACHE_NFSFS_RENAMED
) {
1351 nfs_complete_unlink(dentry
, inode
);
1352 nfs_drop_nlink(inode
);
1357 static void nfs_d_release(struct dentry
*dentry
)
1359 /* free cached devname value, if it survived that far */
1360 if (unlikely(dentry
->d_fsdata
)) {
1361 if (dentry
->d_flags
& DCACHE_NFSFS_RENAMED
)
1364 kfree(dentry
->d_fsdata
);
1368 const struct dentry_operations nfs_dentry_operations
= {
1369 .d_revalidate
= nfs_lookup_revalidate
,
1370 .d_weak_revalidate
= nfs_weak_revalidate
,
1371 .d_delete
= nfs_dentry_delete
,
1372 .d_iput
= nfs_dentry_iput
,
1373 .d_automount
= nfs_d_automount
,
1374 .d_release
= nfs_d_release
,
1376 EXPORT_SYMBOL_GPL(nfs_dentry_operations
);
1378 struct dentry
*nfs_lookup(struct inode
*dir
, struct dentry
* dentry
, unsigned int flags
)
1381 struct inode
*inode
= NULL
;
1382 struct nfs_fh
*fhandle
= NULL
;
1383 struct nfs_fattr
*fattr
= NULL
;
1384 struct nfs4_label
*label
= NULL
;
1387 dfprintk(VFS
, "NFS: lookup(%pd2)\n", dentry
);
1388 nfs_inc_stats(dir
, NFSIOS_VFSLOOKUP
);
1390 if (unlikely(dentry
->d_name
.len
> NFS_SERVER(dir
)->namelen
))
1391 return ERR_PTR(-ENAMETOOLONG
);
1394 * If we're doing an exclusive create, optimize away the lookup
1395 * but don't hash the dentry.
1397 if (nfs_is_exclusive_create(dir
, flags
) || flags
& LOOKUP_RENAME_TARGET
)
1400 res
= ERR_PTR(-ENOMEM
);
1401 fhandle
= nfs_alloc_fhandle();
1402 fattr
= nfs_alloc_fattr();
1403 if (fhandle
== NULL
|| fattr
== NULL
)
1406 label
= nfs4_label_alloc(NFS_SERVER(dir
), GFP_NOWAIT
);
1410 trace_nfs_lookup_enter(dir
, dentry
, flags
);
1411 error
= NFS_PROTO(dir
)->lookup(dir
, &dentry
->d_name
, fhandle
, fattr
, label
);
1412 if (error
== -ENOENT
)
1415 res
= ERR_PTR(error
);
1418 inode
= nfs_fhget(dentry
->d_sb
, fhandle
, fattr
, label
);
1419 res
= ERR_CAST(inode
);
1423 /* Notify readdir to use READDIRPLUS */
1424 nfs_force_use_readdirplus(dir
);
1427 res
= d_splice_alias(inode
, dentry
);
1433 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1435 trace_nfs_lookup_exit(dir
, dentry
, flags
, error
);
1436 nfs4_label_free(label
);
1438 nfs_free_fattr(fattr
);
1439 nfs_free_fhandle(fhandle
);
1442 EXPORT_SYMBOL_GPL(nfs_lookup
);
1444 #if IS_ENABLED(CONFIG_NFS_V4)
1445 static int nfs4_lookup_revalidate(struct dentry
*, unsigned int);
1447 const struct dentry_operations nfs4_dentry_operations
= {
1448 .d_revalidate
= nfs4_lookup_revalidate
,
1449 .d_weak_revalidate
= nfs_weak_revalidate
,
1450 .d_delete
= nfs_dentry_delete
,
1451 .d_iput
= nfs_dentry_iput
,
1452 .d_automount
= nfs_d_automount
,
1453 .d_release
= nfs_d_release
,
1455 EXPORT_SYMBOL_GPL(nfs4_dentry_operations
);
1457 static fmode_t
flags_to_mode(int flags
)
1459 fmode_t res
= (__force fmode_t
)flags
& FMODE_EXEC
;
1460 if ((flags
& O_ACCMODE
) != O_WRONLY
)
1462 if ((flags
& O_ACCMODE
) != O_RDONLY
)
1467 static struct nfs_open_context
*create_nfs_open_context(struct dentry
*dentry
, int open_flags
, struct file
*filp
)
1469 return alloc_nfs_open_context(dentry
, flags_to_mode(open_flags
), filp
);
1472 static int do_open(struct inode
*inode
, struct file
*filp
)
1474 nfs_fscache_open_file(inode
, filp
);
1478 static int nfs_finish_open(struct nfs_open_context
*ctx
,
1479 struct dentry
*dentry
,
1480 struct file
*file
, unsigned open_flags
)
1484 err
= finish_open(file
, dentry
, do_open
);
1487 if (S_ISREG(file
->f_path
.dentry
->d_inode
->i_mode
))
1488 nfs_file_set_open_context(file
, ctx
);
1495 int nfs_atomic_open(struct inode
*dir
, struct dentry
*dentry
,
1496 struct file
*file
, unsigned open_flags
,
1499 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq
);
1500 struct nfs_open_context
*ctx
;
1502 struct iattr attr
= { .ia_valid
= ATTR_OPEN
};
1503 struct inode
*inode
;
1504 unsigned int lookup_flags
= 0;
1505 bool switched
= false;
1509 /* Expect a negative dentry */
1510 BUG_ON(d_inode(dentry
));
1512 dfprintk(VFS
, "NFS: atomic_open(%s/%lu), %pd\n",
1513 dir
->i_sb
->s_id
, dir
->i_ino
, dentry
);
1515 err
= nfs_check_flags(open_flags
);
1519 /* NFS only supports OPEN on regular files */
1520 if ((open_flags
& O_DIRECTORY
)) {
1521 if (!d_in_lookup(dentry
)) {
1523 * Hashed negative dentry with O_DIRECTORY: dentry was
1524 * revalidated and is fine, no need to perform lookup
1529 lookup_flags
= LOOKUP_OPEN
|LOOKUP_DIRECTORY
;
1533 if (dentry
->d_name
.len
> NFS_SERVER(dir
)->namelen
)
1534 return -ENAMETOOLONG
;
1536 if (open_flags
& O_CREAT
) {
1537 struct nfs_server
*server
= NFS_SERVER(dir
);
1539 if (!(server
->attr_bitmask
[2] & FATTR4_WORD2_MODE_UMASK
))
1540 mode
&= ~current_umask();
1542 attr
.ia_valid
|= ATTR_MODE
;
1543 attr
.ia_mode
= mode
;
1545 if (open_flags
& O_TRUNC
) {
1546 attr
.ia_valid
|= ATTR_SIZE
;
1550 if (!(open_flags
& O_CREAT
) && !d_in_lookup(dentry
)) {
1553 dentry
= d_alloc_parallel(dentry
->d_parent
,
1554 &dentry
->d_name
, &wq
);
1556 return PTR_ERR(dentry
);
1557 if (unlikely(!d_in_lookup(dentry
)))
1558 return finish_no_open(file
, dentry
);
1561 ctx
= create_nfs_open_context(dentry
, open_flags
, file
);
1566 trace_nfs_atomic_open_enter(dir
, ctx
, open_flags
);
1567 inode
= NFS_PROTO(dir
)->open_context(dir
, ctx
, open_flags
, &attr
, &created
);
1569 file
->f_mode
|= FMODE_CREATED
;
1570 if (IS_ERR(inode
)) {
1571 err
= PTR_ERR(inode
);
1572 trace_nfs_atomic_open_exit(dir
, ctx
, open_flags
, err
);
1573 put_nfs_open_context(ctx
);
1577 d_splice_alias(NULL
, dentry
);
1578 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1584 if (!(open_flags
& O_NOFOLLOW
))
1594 err
= nfs_finish_open(ctx
, ctx
->dentry
, file
, open_flags
);
1595 trace_nfs_atomic_open_exit(dir
, ctx
, open_flags
, err
);
1596 put_nfs_open_context(ctx
);
1598 if (unlikely(switched
)) {
1599 d_lookup_done(dentry
);
1605 res
= nfs_lookup(dir
, dentry
, lookup_flags
);
1607 d_lookup_done(dentry
);
1614 return PTR_ERR(res
);
1615 return finish_no_open(file
, res
);
1617 EXPORT_SYMBOL_GPL(nfs_atomic_open
);
1620 nfs4_do_lookup_revalidate(struct inode
*dir
, struct dentry
*dentry
,
1623 struct inode
*inode
;
1625 if (!(flags
& LOOKUP_OPEN
) || (flags
& LOOKUP_DIRECTORY
))
1627 if (d_mountpoint(dentry
))
1630 inode
= d_inode(dentry
);
1632 /* We can't create new files in nfs_open_revalidate(), so we
1633 * optimize away revalidation of negative dentries.
1638 if (NFS_PROTO(dir
)->have_delegation(inode
, FMODE_READ
))
1639 return nfs_lookup_revalidate_delegated(dir
, dentry
, inode
);
1641 /* NFS only supports OPEN on regular files */
1642 if (!S_ISREG(inode
->i_mode
))
1645 /* We cannot do exclusive creation on a positive dentry */
1646 if (flags
& (LOOKUP_EXCL
| LOOKUP_REVAL
))
1649 /* Check if the directory changed */
1650 if (!nfs_check_verifier(dir
, dentry
, flags
& LOOKUP_RCU
))
1653 /* Let f_op->open() actually open (and revalidate) the file */
1656 if (flags
& LOOKUP_RCU
)
1658 return nfs_lookup_revalidate_dentry(dir
, dentry
, inode
);
1661 return nfs_do_lookup_revalidate(dir
, dentry
, flags
);
1664 static int nfs4_lookup_revalidate(struct dentry
*dentry
, unsigned int flags
)
1666 return __nfs_lookup_revalidate(dentry
, flags
,
1667 nfs4_do_lookup_revalidate
);
1670 #endif /* CONFIG_NFSV4 */
1673 * Code common to create, mkdir, and mknod.
1675 int nfs_instantiate(struct dentry
*dentry
, struct nfs_fh
*fhandle
,
1676 struct nfs_fattr
*fattr
,
1677 struct nfs4_label
*label
)
1679 struct dentry
*parent
= dget_parent(dentry
);
1680 struct inode
*dir
= d_inode(parent
);
1681 struct inode
*inode
;
1683 int error
= -EACCES
;
1687 /* We may have been initialized further down */
1688 if (d_really_is_positive(dentry
))
1690 if (fhandle
->size
== 0) {
1691 error
= NFS_PROTO(dir
)->lookup(dir
, &dentry
->d_name
, fhandle
, fattr
, NULL
);
1695 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1696 if (!(fattr
->valid
& NFS_ATTR_FATTR
)) {
1697 struct nfs_server
*server
= NFS_SB(dentry
->d_sb
);
1698 error
= server
->nfs_client
->rpc_ops
->getattr(server
, fhandle
,
1703 inode
= nfs_fhget(dentry
->d_sb
, fhandle
, fattr
, label
);
1704 d
= d_splice_alias(inode
, dentry
);
1714 nfs_mark_for_revalidate(dir
);
1718 EXPORT_SYMBOL_GPL(nfs_instantiate
);
1721 * Following a failed create operation, we drop the dentry rather
1722 * than retain a negative dentry. This avoids a problem in the event
1723 * that the operation succeeded on the server, but an error in the
1724 * reply path made it appear to have failed.
1726 int nfs_create(struct inode
*dir
, struct dentry
*dentry
,
1727 umode_t mode
, bool excl
)
1730 int open_flags
= excl
? O_CREAT
| O_EXCL
: O_CREAT
;
1733 dfprintk(VFS
, "NFS: create(%s/%lu), %pd\n",
1734 dir
->i_sb
->s_id
, dir
->i_ino
, dentry
);
1736 attr
.ia_mode
= mode
;
1737 attr
.ia_valid
= ATTR_MODE
;
1739 trace_nfs_create_enter(dir
, dentry
, open_flags
);
1740 error
= NFS_PROTO(dir
)->create(dir
, dentry
, &attr
, open_flags
);
1741 trace_nfs_create_exit(dir
, dentry
, open_flags
, error
);
1749 EXPORT_SYMBOL_GPL(nfs_create
);
1752 * See comments for nfs_proc_create regarding failed operations.
1755 nfs_mknod(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
, dev_t rdev
)
1760 dfprintk(VFS
, "NFS: mknod(%s/%lu), %pd\n",
1761 dir
->i_sb
->s_id
, dir
->i_ino
, dentry
);
1763 attr
.ia_mode
= mode
;
1764 attr
.ia_valid
= ATTR_MODE
;
1766 trace_nfs_mknod_enter(dir
, dentry
);
1767 status
= NFS_PROTO(dir
)->mknod(dir
, dentry
, &attr
, rdev
);
1768 trace_nfs_mknod_exit(dir
, dentry
, status
);
1776 EXPORT_SYMBOL_GPL(nfs_mknod
);
1779 * See comments for nfs_proc_create regarding failed operations.
1781 int nfs_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
1786 dfprintk(VFS
, "NFS: mkdir(%s/%lu), %pd\n",
1787 dir
->i_sb
->s_id
, dir
->i_ino
, dentry
);
1789 attr
.ia_valid
= ATTR_MODE
;
1790 attr
.ia_mode
= mode
| S_IFDIR
;
1792 trace_nfs_mkdir_enter(dir
, dentry
);
1793 error
= NFS_PROTO(dir
)->mkdir(dir
, dentry
, &attr
);
1794 trace_nfs_mkdir_exit(dir
, dentry
, error
);
1802 EXPORT_SYMBOL_GPL(nfs_mkdir
);
1804 static void nfs_dentry_handle_enoent(struct dentry
*dentry
)
1806 if (simple_positive(dentry
))
1810 int nfs_rmdir(struct inode
*dir
, struct dentry
*dentry
)
1814 dfprintk(VFS
, "NFS: rmdir(%s/%lu), %pd\n",
1815 dir
->i_sb
->s_id
, dir
->i_ino
, dentry
);
1817 trace_nfs_rmdir_enter(dir
, dentry
);
1818 if (d_really_is_positive(dentry
)) {
1819 down_write(&NFS_I(d_inode(dentry
))->rmdir_sem
);
1820 error
= NFS_PROTO(dir
)->rmdir(dir
, &dentry
->d_name
);
1821 /* Ensure the VFS deletes this inode */
1824 clear_nlink(d_inode(dentry
));
1827 nfs_dentry_handle_enoent(dentry
);
1829 up_write(&NFS_I(d_inode(dentry
))->rmdir_sem
);
1831 error
= NFS_PROTO(dir
)->rmdir(dir
, &dentry
->d_name
);
1832 trace_nfs_rmdir_exit(dir
, dentry
, error
);
1836 EXPORT_SYMBOL_GPL(nfs_rmdir
);
1839 * Remove a file after making sure there are no pending writes,
1840 * and after checking that the file has only one user.
1842 * We invalidate the attribute cache and free the inode prior to the operation
1843 * to avoid possible races if the server reuses the inode.
1845 static int nfs_safe_remove(struct dentry
*dentry
)
1847 struct inode
*dir
= d_inode(dentry
->d_parent
);
1848 struct inode
*inode
= d_inode(dentry
);
1851 dfprintk(VFS
, "NFS: safe_remove(%pd2)\n", dentry
);
1853 /* If the dentry was sillyrenamed, we simply call d_delete() */
1854 if (dentry
->d_flags
& DCACHE_NFSFS_RENAMED
) {
1859 trace_nfs_remove_enter(dir
, dentry
);
1860 if (inode
!= NULL
) {
1861 error
= NFS_PROTO(dir
)->remove(dir
, dentry
);
1863 nfs_drop_nlink(inode
);
1865 error
= NFS_PROTO(dir
)->remove(dir
, dentry
);
1866 if (error
== -ENOENT
)
1867 nfs_dentry_handle_enoent(dentry
);
1868 trace_nfs_remove_exit(dir
, dentry
, error
);
1873 /* We do silly rename. In case sillyrename() returns -EBUSY, the inode
1874 * belongs to an active ".nfs..." file and we return -EBUSY.
1876 * If sillyrename() returns 0, we do nothing, otherwise we unlink.
1878 int nfs_unlink(struct inode
*dir
, struct dentry
*dentry
)
1881 int need_rehash
= 0;
1883 dfprintk(VFS
, "NFS: unlink(%s/%lu, %pd)\n", dir
->i_sb
->s_id
,
1884 dir
->i_ino
, dentry
);
1886 trace_nfs_unlink_enter(dir
, dentry
);
1887 spin_lock(&dentry
->d_lock
);
1888 if (d_count(dentry
) > 1) {
1889 spin_unlock(&dentry
->d_lock
);
1890 /* Start asynchronous writeout of the inode */
1891 write_inode_now(d_inode(dentry
), 0);
1892 error
= nfs_sillyrename(dir
, dentry
);
1895 if (!d_unhashed(dentry
)) {
1899 spin_unlock(&dentry
->d_lock
);
1900 error
= nfs_safe_remove(dentry
);
1901 if (!error
|| error
== -ENOENT
) {
1902 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1903 } else if (need_rehash
)
1906 trace_nfs_unlink_exit(dir
, dentry
, error
);
1909 EXPORT_SYMBOL_GPL(nfs_unlink
);
1912 * To create a symbolic link, most file systems instantiate a new inode,
1913 * add a page to it containing the path, then write it out to the disk
1914 * using prepare_write/commit_write.
1916 * Unfortunately the NFS client can't create the in-core inode first
1917 * because it needs a file handle to create an in-core inode (see
1918 * fs/nfs/inode.c:nfs_fhget). We only have a file handle *after* the
1919 * symlink request has completed on the server.
1921 * So instead we allocate a raw page, copy the symname into it, then do
1922 * the SYMLINK request with the page as the buffer. If it succeeds, we
1923 * now have a new file handle and can instantiate an in-core NFS inode
1924 * and move the raw page into its mapping.
1926 int nfs_symlink(struct inode
*dir
, struct dentry
*dentry
, const char *symname
)
1931 unsigned int pathlen
= strlen(symname
);
1934 dfprintk(VFS
, "NFS: symlink(%s/%lu, %pd, %s)\n", dir
->i_sb
->s_id
,
1935 dir
->i_ino
, dentry
, symname
);
1937 if (pathlen
> PAGE_SIZE
)
1938 return -ENAMETOOLONG
;
1940 attr
.ia_mode
= S_IFLNK
| S_IRWXUGO
;
1941 attr
.ia_valid
= ATTR_MODE
;
1943 page
= alloc_page(GFP_USER
);
1947 kaddr
= page_address(page
);
1948 memcpy(kaddr
, symname
, pathlen
);
1949 if (pathlen
< PAGE_SIZE
)
1950 memset(kaddr
+ pathlen
, 0, PAGE_SIZE
- pathlen
);
1952 trace_nfs_symlink_enter(dir
, dentry
);
1953 error
= NFS_PROTO(dir
)->symlink(dir
, dentry
, page
, pathlen
, &attr
);
1954 trace_nfs_symlink_exit(dir
, dentry
, error
);
1956 dfprintk(VFS
, "NFS: symlink(%s/%lu, %pd, %s) error %d\n",
1957 dir
->i_sb
->s_id
, dir
->i_ino
,
1958 dentry
, symname
, error
);
1965 * No big deal if we can't add this page to the page cache here.
1966 * READLINK will get the missing page from the server if needed.
1968 if (!add_to_page_cache_lru(page
, d_inode(dentry
)->i_mapping
, 0,
1970 SetPageUptodate(page
);
1973 * add_to_page_cache_lru() grabs an extra page refcount.
1974 * Drop it here to avoid leaking this page later.
1982 EXPORT_SYMBOL_GPL(nfs_symlink
);
1985 nfs_link(struct dentry
*old_dentry
, struct inode
*dir
, struct dentry
*dentry
)
1987 struct inode
*inode
= d_inode(old_dentry
);
1990 dfprintk(VFS
, "NFS: link(%pd2 -> %pd2)\n",
1991 old_dentry
, dentry
);
1993 trace_nfs_link_enter(inode
, dir
, dentry
);
1995 error
= NFS_PROTO(dir
)->link(inode
, dir
, &dentry
->d_name
);
1998 d_add(dentry
, inode
);
2000 trace_nfs_link_exit(inode
, dir
, dentry
, error
);
2003 EXPORT_SYMBOL_GPL(nfs_link
);
2007 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a
2008 * different file handle for the same inode after a rename (e.g. when
2009 * moving to a different directory). A fail-safe method to do so would
2010 * be to look up old_dir/old_name, create a link to new_dir/new_name and
2011 * rename the old file using the sillyrename stuff. This way, the original
2012 * file in old_dir will go away when the last process iput()s the inode.
2016 * It actually works quite well. One needs to have the possibility for
2017 * at least one ".nfs..." file in each directory the file ever gets
2018 * moved or linked to which happens automagically with the new
2019 * implementation that only depends on the dcache stuff instead of
2020 * using the inode layer
2022 * Unfortunately, things are a little more complicated than indicated
2023 * above. For a cross-directory move, we want to make sure we can get
2024 * rid of the old inode after the operation. This means there must be
2025 * no pending writes (if it's a file), and the use count must be 1.
2026 * If these conditions are met, we can drop the dentries before doing
2029 int nfs_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
2030 struct inode
*new_dir
, struct dentry
*new_dentry
,
2033 struct inode
*old_inode
= d_inode(old_dentry
);
2034 struct inode
*new_inode
= d_inode(new_dentry
);
2035 struct dentry
*dentry
= NULL
, *rehash
= NULL
;
2036 struct rpc_task
*task
;
2042 dfprintk(VFS
, "NFS: rename(%pd2 -> %pd2, ct=%d)\n",
2043 old_dentry
, new_dentry
,
2044 d_count(new_dentry
));
2046 trace_nfs_rename_enter(old_dir
, old_dentry
, new_dir
, new_dentry
);
2048 * For non-directories, check whether the target is busy and if so,
2049 * make a copy of the dentry and then do a silly-rename. If the
2050 * silly-rename succeeds, the copied dentry is hashed and becomes
2053 if (new_inode
&& !S_ISDIR(new_inode
->i_mode
)) {
2055 * To prevent any new references to the target during the
2056 * rename, we unhash the dentry in advance.
2058 if (!d_unhashed(new_dentry
)) {
2060 rehash
= new_dentry
;
2063 if (d_count(new_dentry
) > 2) {
2066 /* copy the target dentry's name */
2067 dentry
= d_alloc(new_dentry
->d_parent
,
2068 &new_dentry
->d_name
);
2072 /* silly-rename the existing target ... */
2073 err
= nfs_sillyrename(new_dir
, new_dentry
);
2077 new_dentry
= dentry
;
2083 task
= nfs_async_rename(old_dir
, new_dir
, old_dentry
, new_dentry
, NULL
);
2085 error
= PTR_ERR(task
);
2089 error
= rpc_wait_for_completion_task(task
);
2091 ((struct nfs_renamedata
*)task
->tk_calldata
)->cancelled
= 1;
2092 /* Paired with the atomic_dec_and_test() barrier in rpc_do_put_task() */
2095 error
= task
->tk_status
;
2097 /* Ensure the inode attributes are revalidated */
2099 spin_lock(&old_inode
->i_lock
);
2100 NFS_I(old_inode
)->attr_gencount
= nfs_inc_attr_generation_counter();
2101 NFS_I(old_inode
)->cache_validity
|= NFS_INO_INVALID_CHANGE
2102 | NFS_INO_INVALID_CTIME
2103 | NFS_INO_REVAL_FORCED
;
2104 spin_unlock(&old_inode
->i_lock
);
2109 trace_nfs_rename_exit(old_dir
, old_dentry
,
2110 new_dir
, new_dentry
, error
);
2112 if (new_inode
!= NULL
)
2113 nfs_drop_nlink(new_inode
);
2115 * The d_move() should be here instead of in an async RPC completion
2116 * handler because we need the proper locks to move the dentry. If
2117 * we're interrupted by a signal, the async RPC completion handler
2118 * should mark the directories for revalidation.
2120 d_move(old_dentry
, new_dentry
);
2121 nfs_set_verifier(old_dentry
,
2122 nfs_save_change_attribute(new_dir
));
2123 } else if (error
== -ENOENT
)
2124 nfs_dentry_handle_enoent(old_dentry
);
2126 /* new dentry created? */
2131 EXPORT_SYMBOL_GPL(nfs_rename
);
2133 static DEFINE_SPINLOCK(nfs_access_lru_lock
);
2134 static LIST_HEAD(nfs_access_lru_list
);
2135 static atomic_long_t nfs_access_nr_entries
;
2137 static unsigned long nfs_access_max_cachesize
= ULONG_MAX
;
2138 module_param(nfs_access_max_cachesize
, ulong
, 0644);
2139 MODULE_PARM_DESC(nfs_access_max_cachesize
, "NFS access maximum total cache length");
2141 static void nfs_access_free_entry(struct nfs_access_entry
*entry
)
2143 put_cred(entry
->cred
);
2144 kfree_rcu(entry
, rcu_head
);
2145 smp_mb__before_atomic();
2146 atomic_long_dec(&nfs_access_nr_entries
);
2147 smp_mb__after_atomic();
2150 static void nfs_access_free_list(struct list_head
*head
)
2152 struct nfs_access_entry
*cache
;
2154 while (!list_empty(head
)) {
2155 cache
= list_entry(head
->next
, struct nfs_access_entry
, lru
);
2156 list_del(&cache
->lru
);
2157 nfs_access_free_entry(cache
);
2161 static unsigned long
2162 nfs_do_access_cache_scan(unsigned int nr_to_scan
)
2165 struct nfs_inode
*nfsi
, *next
;
2166 struct nfs_access_entry
*cache
;
2169 spin_lock(&nfs_access_lru_lock
);
2170 list_for_each_entry_safe(nfsi
, next
, &nfs_access_lru_list
, access_cache_inode_lru
) {
2171 struct inode
*inode
;
2173 if (nr_to_scan
-- == 0)
2175 inode
= &nfsi
->vfs_inode
;
2176 spin_lock(&inode
->i_lock
);
2177 if (list_empty(&nfsi
->access_cache_entry_lru
))
2178 goto remove_lru_entry
;
2179 cache
= list_entry(nfsi
->access_cache_entry_lru
.next
,
2180 struct nfs_access_entry
, lru
);
2181 list_move(&cache
->lru
, &head
);
2182 rb_erase(&cache
->rb_node
, &nfsi
->access_cache
);
2184 if (!list_empty(&nfsi
->access_cache_entry_lru
))
2185 list_move_tail(&nfsi
->access_cache_inode_lru
,
2186 &nfs_access_lru_list
);
2189 list_del_init(&nfsi
->access_cache_inode_lru
);
2190 smp_mb__before_atomic();
2191 clear_bit(NFS_INO_ACL_LRU_SET
, &nfsi
->flags
);
2192 smp_mb__after_atomic();
2194 spin_unlock(&inode
->i_lock
);
2196 spin_unlock(&nfs_access_lru_lock
);
2197 nfs_access_free_list(&head
);
2202 nfs_access_cache_scan(struct shrinker
*shrink
, struct shrink_control
*sc
)
2204 int nr_to_scan
= sc
->nr_to_scan
;
2205 gfp_t gfp_mask
= sc
->gfp_mask
;
2207 if ((gfp_mask
& GFP_KERNEL
) != GFP_KERNEL
)
2209 return nfs_do_access_cache_scan(nr_to_scan
);
2214 nfs_access_cache_count(struct shrinker
*shrink
, struct shrink_control
*sc
)
2216 return vfs_pressure_ratio(atomic_long_read(&nfs_access_nr_entries
));
2220 nfs_access_cache_enforce_limit(void)
2222 long nr_entries
= atomic_long_read(&nfs_access_nr_entries
);
2224 unsigned int nr_to_scan
;
2226 if (nr_entries
< 0 || nr_entries
<= nfs_access_max_cachesize
)
2229 diff
= nr_entries
- nfs_access_max_cachesize
;
2230 if (diff
< nr_to_scan
)
2232 nfs_do_access_cache_scan(nr_to_scan
);
2235 static void __nfs_access_zap_cache(struct nfs_inode
*nfsi
, struct list_head
*head
)
2237 struct rb_root
*root_node
= &nfsi
->access_cache
;
2239 struct nfs_access_entry
*entry
;
2241 /* Unhook entries from the cache */
2242 while ((n
= rb_first(root_node
)) != NULL
) {
2243 entry
= rb_entry(n
, struct nfs_access_entry
, rb_node
);
2244 rb_erase(n
, root_node
);
2245 list_move(&entry
->lru
, head
);
2247 nfsi
->cache_validity
&= ~NFS_INO_INVALID_ACCESS
;
2250 void nfs_access_zap_cache(struct inode
*inode
)
2254 if (test_bit(NFS_INO_ACL_LRU_SET
, &NFS_I(inode
)->flags
) == 0)
2256 /* Remove from global LRU init */
2257 spin_lock(&nfs_access_lru_lock
);
2258 if (test_and_clear_bit(NFS_INO_ACL_LRU_SET
, &NFS_I(inode
)->flags
))
2259 list_del_init(&NFS_I(inode
)->access_cache_inode_lru
);
2261 spin_lock(&inode
->i_lock
);
2262 __nfs_access_zap_cache(NFS_I(inode
), &head
);
2263 spin_unlock(&inode
->i_lock
);
2264 spin_unlock(&nfs_access_lru_lock
);
2265 nfs_access_free_list(&head
);
2267 EXPORT_SYMBOL_GPL(nfs_access_zap_cache
);
2269 static struct nfs_access_entry
*nfs_access_search_rbtree(struct inode
*inode
, const struct cred
*cred
)
2271 struct rb_node
*n
= NFS_I(inode
)->access_cache
.rb_node
;
2274 struct nfs_access_entry
*entry
=
2275 rb_entry(n
, struct nfs_access_entry
, rb_node
);
2276 int cmp
= cred_fscmp(cred
, entry
->cred
);
2288 static int nfs_access_get_cached(struct inode
*inode
, const struct cred
*cred
, struct nfs_access_entry
*res
, bool may_block
)
2290 struct nfs_inode
*nfsi
= NFS_I(inode
);
2291 struct nfs_access_entry
*cache
;
2295 spin_lock(&inode
->i_lock
);
2297 if (nfsi
->cache_validity
& NFS_INO_INVALID_ACCESS
)
2299 cache
= nfs_access_search_rbtree(inode
, cred
);
2303 /* Found an entry, is our attribute cache valid? */
2304 if (!nfs_check_cache_invalid(inode
, NFS_INO_INVALID_ACCESS
))
2311 spin_unlock(&inode
->i_lock
);
2312 err
= __nfs_revalidate_inode(NFS_SERVER(inode
), inode
);
2315 spin_lock(&inode
->i_lock
);
2318 res
->cred
= cache
->cred
;
2319 res
->mask
= cache
->mask
;
2320 list_move_tail(&cache
->lru
, &nfsi
->access_cache_entry_lru
);
2323 spin_unlock(&inode
->i_lock
);
2326 spin_unlock(&inode
->i_lock
);
2327 nfs_access_zap_cache(inode
);
2331 static int nfs_access_get_cached_rcu(struct inode
*inode
, const struct cred
*cred
, struct nfs_access_entry
*res
)
2333 /* Only check the most recently returned cache entry,
2334 * but do it without locking.
2336 struct nfs_inode
*nfsi
= NFS_I(inode
);
2337 struct nfs_access_entry
*cache
;
2339 struct list_head
*lh
;
2342 if (nfsi
->cache_validity
& NFS_INO_INVALID_ACCESS
)
2344 lh
= rcu_dereference(nfsi
->access_cache_entry_lru
.prev
);
2345 cache
= list_entry(lh
, struct nfs_access_entry
, lru
);
2346 if (lh
== &nfsi
->access_cache_entry_lru
||
2347 cred
!= cache
->cred
)
2351 if (nfs_check_cache_invalid(inode
, NFS_INO_INVALID_ACCESS
))
2353 res
->cred
= cache
->cred
;
2354 res
->mask
= cache
->mask
;
2361 static void nfs_access_add_rbtree(struct inode
*inode
, struct nfs_access_entry
*set
)
2363 struct nfs_inode
*nfsi
= NFS_I(inode
);
2364 struct rb_root
*root_node
= &nfsi
->access_cache
;
2365 struct rb_node
**p
= &root_node
->rb_node
;
2366 struct rb_node
*parent
= NULL
;
2367 struct nfs_access_entry
*entry
;
2370 spin_lock(&inode
->i_lock
);
2371 while (*p
!= NULL
) {
2373 entry
= rb_entry(parent
, struct nfs_access_entry
, rb_node
);
2374 cmp
= cred_fscmp(set
->cred
, entry
->cred
);
2377 p
= &parent
->rb_left
;
2379 p
= &parent
->rb_right
;
2383 rb_link_node(&set
->rb_node
, parent
, p
);
2384 rb_insert_color(&set
->rb_node
, root_node
);
2385 list_add_tail(&set
->lru
, &nfsi
->access_cache_entry_lru
);
2386 spin_unlock(&inode
->i_lock
);
2389 rb_replace_node(parent
, &set
->rb_node
, root_node
);
2390 list_add_tail(&set
->lru
, &nfsi
->access_cache_entry_lru
);
2391 list_del(&entry
->lru
);
2392 spin_unlock(&inode
->i_lock
);
2393 nfs_access_free_entry(entry
);
2396 void nfs_access_add_cache(struct inode
*inode
, struct nfs_access_entry
*set
)
2398 struct nfs_access_entry
*cache
= kmalloc(sizeof(*cache
), GFP_KERNEL
);
2401 RB_CLEAR_NODE(&cache
->rb_node
);
2402 cache
->cred
= get_cred(set
->cred
);
2403 cache
->mask
= set
->mask
;
2405 /* The above field assignments must be visible
2406 * before this item appears on the lru. We cannot easily
2407 * use rcu_assign_pointer, so just force the memory barrier.
2410 nfs_access_add_rbtree(inode
, cache
);
2412 /* Update accounting */
2413 smp_mb__before_atomic();
2414 atomic_long_inc(&nfs_access_nr_entries
);
2415 smp_mb__after_atomic();
2417 /* Add inode to global LRU list */
2418 if (!test_bit(NFS_INO_ACL_LRU_SET
, &NFS_I(inode
)->flags
)) {
2419 spin_lock(&nfs_access_lru_lock
);
2420 if (!test_and_set_bit(NFS_INO_ACL_LRU_SET
, &NFS_I(inode
)->flags
))
2421 list_add_tail(&NFS_I(inode
)->access_cache_inode_lru
,
2422 &nfs_access_lru_list
);
2423 spin_unlock(&nfs_access_lru_lock
);
2425 nfs_access_cache_enforce_limit();
2427 EXPORT_SYMBOL_GPL(nfs_access_add_cache
);
2429 #define NFS_MAY_READ (NFS_ACCESS_READ)
2430 #define NFS_MAY_WRITE (NFS_ACCESS_MODIFY | \
2431 NFS_ACCESS_EXTEND | \
2433 #define NFS_FILE_MAY_WRITE (NFS_ACCESS_MODIFY | \
2435 #define NFS_DIR_MAY_WRITE NFS_MAY_WRITE
2436 #define NFS_MAY_LOOKUP (NFS_ACCESS_LOOKUP)
2437 #define NFS_MAY_EXECUTE (NFS_ACCESS_EXECUTE)
2439 nfs_access_calc_mask(u32 access_result
, umode_t umode
)
2443 if (access_result
& NFS_MAY_READ
)
2445 if (S_ISDIR(umode
)) {
2446 if ((access_result
& NFS_DIR_MAY_WRITE
) == NFS_DIR_MAY_WRITE
)
2448 if ((access_result
& NFS_MAY_LOOKUP
) == NFS_MAY_LOOKUP
)
2450 } else if (S_ISREG(umode
)) {
2451 if ((access_result
& NFS_FILE_MAY_WRITE
) == NFS_FILE_MAY_WRITE
)
2453 if ((access_result
& NFS_MAY_EXECUTE
) == NFS_MAY_EXECUTE
)
2455 } else if (access_result
& NFS_MAY_WRITE
)
2460 void nfs_access_set_mask(struct nfs_access_entry
*entry
, u32 access_result
)
2462 entry
->mask
= access_result
;
2464 EXPORT_SYMBOL_GPL(nfs_access_set_mask
);
2466 static int nfs_do_access(struct inode
*inode
, const struct cred
*cred
, int mask
)
2468 struct nfs_access_entry cache
;
2469 bool may_block
= (mask
& MAY_NOT_BLOCK
) == 0;
2473 trace_nfs_access_enter(inode
);
2475 status
= nfs_access_get_cached_rcu(inode
, cred
, &cache
);
2477 status
= nfs_access_get_cached(inode
, cred
, &cache
, may_block
);
2486 * Determine which access bits we want to ask for...
2488 cache
.mask
= NFS_ACCESS_READ
| NFS_ACCESS_MODIFY
| NFS_ACCESS_EXTEND
;
2489 if (S_ISDIR(inode
->i_mode
))
2490 cache
.mask
|= NFS_ACCESS_DELETE
| NFS_ACCESS_LOOKUP
;
2492 cache
.mask
|= NFS_ACCESS_EXECUTE
;
2494 status
= NFS_PROTO(inode
)->access(inode
, &cache
);
2496 if (status
== -ESTALE
) {
2497 nfs_zap_caches(inode
);
2498 if (!S_ISDIR(inode
->i_mode
))
2499 set_bit(NFS_INO_STALE
, &NFS_I(inode
)->flags
);
2503 nfs_access_add_cache(inode
, &cache
);
2505 cache_mask
= nfs_access_calc_mask(cache
.mask
, inode
->i_mode
);
2506 if ((mask
& ~cache_mask
& (MAY_READ
| MAY_WRITE
| MAY_EXEC
)) != 0)
2509 trace_nfs_access_exit(inode
, status
);
2513 static int nfs_open_permission_mask(int openflags
)
2517 if (openflags
& __FMODE_EXEC
) {
2518 /* ONLY check exec rights */
2521 if ((openflags
& O_ACCMODE
) != O_WRONLY
)
2523 if ((openflags
& O_ACCMODE
) != O_RDONLY
)
2530 int nfs_may_open(struct inode
*inode
, const struct cred
*cred
, int openflags
)
2532 return nfs_do_access(inode
, cred
, nfs_open_permission_mask(openflags
));
2534 EXPORT_SYMBOL_GPL(nfs_may_open
);
2536 static int nfs_execute_ok(struct inode
*inode
, int mask
)
2538 struct nfs_server
*server
= NFS_SERVER(inode
);
2541 if (S_ISDIR(inode
->i_mode
))
2543 if (nfs_check_cache_invalid(inode
, NFS_INO_INVALID_OTHER
)) {
2544 if (mask
& MAY_NOT_BLOCK
)
2546 ret
= __nfs_revalidate_inode(server
, inode
);
2548 if (ret
== 0 && !execute_ok(inode
))
2553 int nfs_permission(struct inode
*inode
, int mask
)
2555 const struct cred
*cred
= current_cred();
2558 nfs_inc_stats(inode
, NFSIOS_VFSACCESS
);
2560 if ((mask
& (MAY_READ
| MAY_WRITE
| MAY_EXEC
)) == 0)
2562 /* Is this sys_access() ? */
2563 if (mask
& (MAY_ACCESS
| MAY_CHDIR
))
2566 switch (inode
->i_mode
& S_IFMT
) {
2570 if ((mask
& MAY_OPEN
) &&
2571 nfs_server_capable(inode
, NFS_CAP_ATOMIC_OPEN
))
2576 * Optimize away all write operations, since the server
2577 * will check permissions when we perform the op.
2579 if ((mask
& MAY_WRITE
) && !(mask
& MAY_READ
))
2584 if (!NFS_PROTO(inode
)->access
)
2587 /* Always try fast lookups first */
2589 res
= nfs_do_access(inode
, cred
, mask
|MAY_NOT_BLOCK
);
2591 if (res
== -ECHILD
&& !(mask
& MAY_NOT_BLOCK
)) {
2592 /* Fast lookup failed, try the slow way */
2593 res
= nfs_do_access(inode
, cred
, mask
);
2596 if (!res
&& (mask
& MAY_EXEC
))
2597 res
= nfs_execute_ok(inode
, mask
);
2599 dfprintk(VFS
, "NFS: permission(%s/%lu), mask=0x%x, res=%d\n",
2600 inode
->i_sb
->s_id
, inode
->i_ino
, mask
, res
);
2603 if (mask
& MAY_NOT_BLOCK
)
2606 res
= nfs_revalidate_inode(NFS_SERVER(inode
), inode
);
2608 res
= generic_permission(inode
, mask
);
2611 EXPORT_SYMBOL_GPL(nfs_permission
);
2615 * version-control: t
2616 * kept-new-versions: 5