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 list_add(&ctx
->list
, &nfsi
->open_files
);
84 spin_unlock(&dir
->i_lock
);
87 return ERR_PTR(-ENOMEM
);
90 static void put_nfs_open_dir_context(struct inode
*dir
, struct nfs_open_dir_context
*ctx
)
92 spin_lock(&dir
->i_lock
);
94 spin_unlock(&dir
->i_lock
);
103 nfs_opendir(struct inode
*inode
, struct file
*filp
)
106 struct nfs_open_dir_context
*ctx
;
108 dfprintk(FILE, "NFS: open dir(%pD2)\n", filp
);
110 nfs_inc_stats(inode
, NFSIOS_VFSOPEN
);
112 ctx
= alloc_nfs_open_dir_context(inode
, current_cred());
117 filp
->private_data
= ctx
;
123 nfs_closedir(struct inode
*inode
, struct file
*filp
)
125 put_nfs_open_dir_context(file_inode(filp
), filp
->private_data
);
129 struct nfs_cache_array_entry
{
133 unsigned char d_type
;
136 struct nfs_cache_array
{
140 struct nfs_cache_array_entry array
[0];
146 struct page
*pages
[NFS_MAX_READDIR_RAPAGES
];
149 typedef int (*decode_dirent_t
)(struct xdr_stream
*, struct nfs_entry
*, bool);
153 struct dir_context
*ctx
;
154 unsigned long page_index
;
155 struct readdirvec pvec
;
158 loff_t current_index
;
159 decode_dirent_t decode
;
161 unsigned long timestamp
;
162 unsigned long gencount
;
163 unsigned int cache_entry_index
;
166 } nfs_readdir_descriptor_t
;
169 * we are freeing strings created by nfs_add_to_readdir_array()
172 void nfs_readdir_clear_array(struct page
*page
)
174 struct nfs_cache_array
*array
;
177 array
= kmap_atomic(page
);
178 for (i
= 0; i
< array
->size
; i
++)
179 kfree(array
->array
[i
].string
.name
);
180 kunmap_atomic(array
);
184 * the caller is responsible for freeing qstr.name
185 * when called by nfs_readdir_add_to_array, the strings will be freed in
186 * nfs_clear_readdir_array()
189 int nfs_readdir_make_qstr(struct qstr
*string
, const char *name
, unsigned int len
)
192 string
->name
= kmemdup(name
, len
, GFP_KERNEL
);
193 if (string
->name
== NULL
)
196 * Avoid a kmemleak false positive. The pointer to the name is stored
197 * in a page cache page which kmemleak does not scan.
199 kmemleak_not_leak(string
->name
);
200 string
->hash
= full_name_hash(NULL
, name
, len
);
205 int nfs_readdir_add_to_array(struct nfs_entry
*entry
, struct page
*page
)
207 struct nfs_cache_array
*array
= kmap(page
);
208 struct nfs_cache_array_entry
*cache_entry
;
211 cache_entry
= &array
->array
[array
->size
];
213 /* Check that this entry lies within the page bounds */
215 if ((char *)&cache_entry
[1] - (char *)page_address(page
) > PAGE_SIZE
)
218 cache_entry
->cookie
= entry
->prev_cookie
;
219 cache_entry
->ino
= entry
->ino
;
220 cache_entry
->d_type
= entry
->d_type
;
221 ret
= nfs_readdir_make_qstr(&cache_entry
->string
, entry
->name
, entry
->len
);
224 array
->last_cookie
= entry
->cookie
;
227 array
->eof_index
= array
->size
;
234 int nfs_readdir_search_for_pos(struct nfs_cache_array
*array
, nfs_readdir_descriptor_t
*desc
)
236 loff_t diff
= desc
->ctx
->pos
- desc
->current_index
;
241 if (diff
>= array
->size
) {
242 if (array
->eof_index
>= 0)
247 index
= (unsigned int)diff
;
248 *desc
->dir_cookie
= array
->array
[index
].cookie
;
249 desc
->cache_entry_index
= index
;
257 nfs_readdir_inode_mapping_valid(struct nfs_inode
*nfsi
)
259 if (nfsi
->cache_validity
& (NFS_INO_INVALID_ATTR
|NFS_INO_INVALID_DATA
))
262 return !test_bit(NFS_INO_INVALIDATING
, &nfsi
->flags
);
266 int nfs_readdir_search_for_cookie(struct nfs_cache_array
*array
, nfs_readdir_descriptor_t
*desc
)
270 int status
= -EAGAIN
;
272 for (i
= 0; i
< array
->size
; i
++) {
273 if (array
->array
[i
].cookie
== *desc
->dir_cookie
) {
274 struct nfs_inode
*nfsi
= NFS_I(file_inode(desc
->file
));
275 struct nfs_open_dir_context
*ctx
= desc
->file
->private_data
;
277 new_pos
= desc
->current_index
+ i
;
278 if (ctx
->attr_gencount
!= nfsi
->attr_gencount
||
279 !nfs_readdir_inode_mapping_valid(nfsi
)) {
281 ctx
->attr_gencount
= nfsi
->attr_gencount
;
282 } else if (new_pos
< desc
->ctx
->pos
) {
284 && ctx
->dup_cookie
== *desc
->dir_cookie
) {
285 if (printk_ratelimit()) {
286 pr_notice("NFS: directory %pD2 contains a readdir loop."
287 "Please contact your server vendor. "
288 "The file: %.*s has duplicate cookie %llu\n",
289 desc
->file
, array
->array
[i
].string
.len
,
290 array
->array
[i
].string
.name
, *desc
->dir_cookie
);
295 ctx
->dup_cookie
= *desc
->dir_cookie
;
298 desc
->ctx
->pos
= new_pos
;
299 desc
->cache_entry_index
= i
;
303 if (array
->eof_index
>= 0) {
304 status
= -EBADCOOKIE
;
305 if (*desc
->dir_cookie
== array
->last_cookie
)
313 int nfs_readdir_search_array(nfs_readdir_descriptor_t
*desc
)
315 struct nfs_cache_array
*array
;
318 array
= kmap(desc
->page
);
320 if (*desc
->dir_cookie
== 0)
321 status
= nfs_readdir_search_for_pos(array
, desc
);
323 status
= nfs_readdir_search_for_cookie(array
, desc
);
325 if (status
== -EAGAIN
) {
326 desc
->last_cookie
= array
->last_cookie
;
327 desc
->current_index
+= array
->size
;
334 /* Fill a page with xdr information before transferring to the cache page */
336 int nfs_readdir_xdr_filler(struct page
**pages
, nfs_readdir_descriptor_t
*desc
,
337 struct nfs_entry
*entry
, struct file
*file
, struct inode
*inode
)
339 struct nfs_open_dir_context
*ctx
= file
->private_data
;
340 const struct cred
*cred
= ctx
->cred
;
341 unsigned long timestamp
, gencount
;
346 gencount
= nfs_inc_attr_generation_counter();
347 error
= NFS_PROTO(inode
)->readdir(file_dentry(file
), cred
, entry
->cookie
, pages
,
348 NFS_SERVER(inode
)->dtsize
, desc
->plus
);
350 /* We requested READDIRPLUS, but the server doesn't grok it */
351 if (error
== -ENOTSUPP
&& desc
->plus
) {
352 NFS_SERVER(inode
)->caps
&= ~NFS_CAP_READDIRPLUS
;
353 clear_bit(NFS_INO_ADVISE_RDPLUS
, &NFS_I(inode
)->flags
);
359 desc
->timestamp
= timestamp
;
360 desc
->gencount
= gencount
;
365 static int xdr_decode(nfs_readdir_descriptor_t
*desc
,
366 struct nfs_entry
*entry
, struct xdr_stream
*xdr
)
370 error
= desc
->decode(xdr
, entry
, desc
->plus
);
373 entry
->fattr
->time_start
= desc
->timestamp
;
374 entry
->fattr
->gencount
= desc
->gencount
;
378 /* Match file and dirent using either filehandle or fileid
379 * Note: caller is responsible for checking the fsid
382 int nfs_same_file(struct dentry
*dentry
, struct nfs_entry
*entry
)
385 struct nfs_inode
*nfsi
;
387 if (d_really_is_negative(dentry
))
390 inode
= d_inode(dentry
);
391 if (is_bad_inode(inode
) || NFS_STALE(inode
))
395 if (entry
->fattr
->fileid
!= nfsi
->fileid
)
397 if (entry
->fh
->size
&& nfs_compare_fh(entry
->fh
, &nfsi
->fh
) != 0)
403 bool nfs_use_readdirplus(struct inode
*dir
, struct dir_context
*ctx
)
405 if (!nfs_server_capable(dir
, NFS_CAP_READDIRPLUS
))
407 if (test_and_clear_bit(NFS_INO_ADVISE_RDPLUS
, &NFS_I(dir
)->flags
))
415 * This function is called by the lookup and getattr code to request the
416 * use of readdirplus to accelerate any future lookups in the same
419 void nfs_advise_use_readdirplus(struct inode
*dir
)
421 struct nfs_inode
*nfsi
= NFS_I(dir
);
423 if (nfs_server_capable(dir
, NFS_CAP_READDIRPLUS
) &&
424 !list_empty(&nfsi
->open_files
))
425 set_bit(NFS_INO_ADVISE_RDPLUS
, &nfsi
->flags
);
429 * This function is mainly for use by nfs_getattr().
431 * If this is an 'ls -l', we want to force use of readdirplus.
432 * Do this by checking if there is an active file descriptor
433 * and calling nfs_advise_use_readdirplus, then forcing a
436 void nfs_force_use_readdirplus(struct inode
*dir
)
438 struct nfs_inode
*nfsi
= NFS_I(dir
);
440 if (nfs_server_capable(dir
, NFS_CAP_READDIRPLUS
) &&
441 !list_empty(&nfsi
->open_files
)) {
442 set_bit(NFS_INO_ADVISE_RDPLUS
, &nfsi
->flags
);
443 invalidate_mapping_pages(dir
->i_mapping
, 0, -1);
448 void nfs_prime_dcache(struct dentry
*parent
, struct nfs_entry
*entry
)
450 struct qstr filename
= QSTR_INIT(entry
->name
, entry
->len
);
451 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq
);
452 struct dentry
*dentry
;
453 struct dentry
*alias
;
454 struct inode
*dir
= d_inode(parent
);
458 if (!(entry
->fattr
->valid
& NFS_ATTR_FATTR_FILEID
))
460 if (!(entry
->fattr
->valid
& NFS_ATTR_FATTR_FSID
))
462 if (filename
.len
== 0)
464 /* Validate that the name doesn't contain any illegal '\0' */
465 if (strnlen(filename
.name
, filename
.len
) != filename
.len
)
468 if (strnchr(filename
.name
, filename
.len
, '/'))
470 if (filename
.name
[0] == '.') {
471 if (filename
.len
== 1)
473 if (filename
.len
== 2 && filename
.name
[1] == '.')
476 filename
.hash
= full_name_hash(parent
, filename
.name
, filename
.len
);
478 dentry
= d_lookup(parent
, &filename
);
481 dentry
= d_alloc_parallel(parent
, &filename
, &wq
);
485 if (!d_in_lookup(dentry
)) {
486 /* Is there a mountpoint here? If so, just exit */
487 if (!nfs_fsid_equal(&NFS_SB(dentry
->d_sb
)->fsid
,
488 &entry
->fattr
->fsid
))
490 if (nfs_same_file(dentry
, entry
)) {
491 if (!entry
->fh
->size
)
493 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
494 status
= nfs_refresh_inode(d_inode(dentry
), entry
->fattr
);
496 nfs_setsecurity(d_inode(dentry
), entry
->fattr
, entry
->label
);
499 d_invalidate(dentry
);
505 if (!entry
->fh
->size
) {
506 d_lookup_done(dentry
);
510 inode
= nfs_fhget(dentry
->d_sb
, entry
->fh
, entry
->fattr
, entry
->label
);
511 alias
= d_splice_alias(inode
, dentry
);
512 d_lookup_done(dentry
);
519 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
524 /* Perform conversion from xdr to cache array */
526 int nfs_readdir_page_filler(nfs_readdir_descriptor_t
*desc
, struct nfs_entry
*entry
,
527 struct page
**xdr_pages
, struct page
*page
, unsigned int buflen
)
529 struct xdr_stream stream
;
531 struct page
*scratch
;
532 struct nfs_cache_array
*array
;
533 unsigned int count
= 0;
535 int max_rapages
= NFS_MAX_READDIR_RAPAGES
;
537 desc
->pvec
.index
= desc
->page_index
;
540 scratch
= alloc_page(GFP_KERNEL
);
547 xdr_init_decode_pages(&stream
, &buf
, xdr_pages
, buflen
);
548 xdr_set_scratch_buffer(&stream
, page_address(scratch
), PAGE_SIZE
);
551 status
= xdr_decode(desc
, entry
, &stream
);
553 if (status
== -EAGAIN
)
561 nfs_prime_dcache(file_dentry(desc
->file
), entry
);
563 status
= nfs_readdir_add_to_array(entry
, desc
->pvec
.pages
[desc
->pvec
.nr
]);
564 if (status
== -ENOSPC
) {
566 if (desc
->pvec
.nr
== max_rapages
)
568 status
= nfs_readdir_add_to_array(entry
, desc
->pvec
.pages
[desc
->pvec
.nr
]);
572 } while (!entry
->eof
);
575 * page and desc->pvec.pages[0] are valid, don't need to check
576 * whether or not to be NULL.
578 copy_highpage(page
, desc
->pvec
.pages
[0]);
581 if (count
== 0 || (status
== -EBADCOOKIE
&& entry
->eof
!= 0)) {
582 array
= kmap_atomic(desc
->pvec
.pages
[desc
->pvec
.nr
]);
583 array
->eof_index
= array
->size
;
585 kunmap_atomic(array
);
591 * desc->pvec.nr > 0 means at least one page was completely filled,
592 * we should return -ENOSPC. Otherwise function
593 * nfs_readdir_xdr_to_array will enter infinite loop.
595 if (desc
->pvec
.nr
> 0)
601 void nfs_readdir_free_pages(struct page
**pages
, unsigned int npages
)
604 for (i
= 0; i
< npages
; i
++)
609 * nfs_readdir_alloc_pages() will allocate pages that must be freed with a call
610 * to nfs_readdir_free_pages()
613 int nfs_readdir_alloc_pages(struct page
**pages
, unsigned int npages
)
617 for (i
= 0; i
< npages
; i
++) {
618 struct page
*page
= alloc_page(GFP_KERNEL
);
626 nfs_readdir_free_pages(pages
, i
);
631 * nfs_readdir_rapages_init initialize rapages by nfs_cache_array structure.
634 void nfs_readdir_rapages_init(nfs_readdir_descriptor_t
*desc
)
636 struct nfs_cache_array
*array
;
637 int max_rapages
= NFS_MAX_READDIR_RAPAGES
;
640 for (index
= 0; index
< max_rapages
; index
++) {
641 array
= kmap_atomic(desc
->pvec
.pages
[index
]);
642 memset(array
, 0, sizeof(struct nfs_cache_array
));
643 array
->eof_index
= -1;
644 kunmap_atomic(array
);
649 int nfs_readdir_xdr_to_array(nfs_readdir_descriptor_t
*desc
, struct page
*page
, struct inode
*inode
)
651 struct page
*pages
[NFS_MAX_READDIR_PAGES
];
652 struct nfs_entry entry
;
653 struct file
*file
= desc
->file
;
654 struct nfs_cache_array
*array
;
655 int status
= -ENOMEM
;
656 unsigned int array_size
= ARRAY_SIZE(pages
);
659 * This means we hit readdir rdpages miss, the preallocated rdpages
660 * are useless, the preallocate rdpages should be reinitialized.
662 nfs_readdir_rapages_init(desc
);
664 entry
.prev_cookie
= 0;
665 entry
.cookie
= desc
->last_cookie
;
667 entry
.fh
= nfs_alloc_fhandle();
668 entry
.fattr
= nfs_alloc_fattr();
669 entry
.server
= NFS_SERVER(inode
);
670 if (entry
.fh
== NULL
|| entry
.fattr
== NULL
)
673 entry
.label
= nfs4_label_alloc(NFS_SERVER(inode
), GFP_NOWAIT
);
674 if (IS_ERR(entry
.label
)) {
675 status
= PTR_ERR(entry
.label
);
680 memset(array
, 0, sizeof(struct nfs_cache_array
));
681 array
->eof_index
= -1;
683 status
= nfs_readdir_alloc_pages(pages
, array_size
);
685 goto out_release_array
;
688 status
= nfs_readdir_xdr_filler(pages
, desc
, &entry
, file
, inode
);
693 status
= nfs_readdir_page_filler(desc
, &entry
, pages
, page
, pglen
);
695 if (status
== -ENOSPC
)
699 } while (array
->eof_index
< 0);
701 nfs_readdir_free_pages(pages
, array_size
);
704 nfs4_label_free(entry
.label
);
706 nfs_free_fattr(entry
.fattr
);
707 nfs_free_fhandle(entry
.fh
);
712 * Now we cache directories properly, by converting xdr information
713 * to an array that can be used for lookups later. This results in
714 * fewer cache pages, since we can store more information on each page.
715 * We only need to convert from xdr once so future lookups are much simpler
718 int nfs_readdir_filler(void *data
, struct page
* page
)
720 nfs_readdir_descriptor_t
*desc
= data
;
721 struct inode
*inode
= file_inode(desc
->file
);
725 * If desc->page_index in range desc->pvec.index and
726 * desc->pvec.index + desc->pvec.nr, we get readdir cache hit.
728 if (desc
->page_index
>= desc
->pvec
.index
&&
729 desc
->page_index
< (desc
->pvec
.index
+ desc
->pvec
.nr
)) {
731 * page and desc->pvec.pages[x] are valid, don't need to check
732 * whether or not to be NULL.
734 copy_highpage(page
, desc
->pvec
.pages
[desc
->page_index
- desc
->pvec
.index
]);
737 ret
= nfs_readdir_xdr_to_array(desc
, page
, inode
);
742 SetPageUptodate(page
);
744 if (invalidate_inode_pages2_range(inode
->i_mapping
, page
->index
+ 1, -1) < 0) {
745 /* Should never happen */
746 nfs_zap_mapping(inode
, inode
->i_mapping
);
756 void cache_page_release(nfs_readdir_descriptor_t
*desc
)
758 if (!desc
->page
->mapping
)
759 nfs_readdir_clear_array(desc
->page
);
760 put_page(desc
->page
);
765 struct page
*get_cache_page(nfs_readdir_descriptor_t
*desc
)
767 return read_cache_page(desc
->file
->f_mapping
, desc
->page_index
,
768 nfs_readdir_filler
, desc
);
772 * Returns 0 if desc->dir_cookie was found on page desc->page_index
775 int find_cache_page(nfs_readdir_descriptor_t
*desc
)
779 desc
->page
= get_cache_page(desc
);
780 if (IS_ERR(desc
->page
))
781 return PTR_ERR(desc
->page
);
783 res
= nfs_readdir_search_array(desc
);
785 cache_page_release(desc
);
789 /* Search for desc->dir_cookie from the beginning of the page cache */
791 int readdir_search_pagecache(nfs_readdir_descriptor_t
*desc
)
795 if (desc
->page_index
== 0) {
796 desc
->current_index
= 0;
797 desc
->last_cookie
= 0;
800 res
= find_cache_page(desc
);
801 } while (res
== -EAGAIN
);
806 * Once we've found the start of the dirent within a page: fill 'er up...
809 int nfs_do_filldir(nfs_readdir_descriptor_t
*desc
)
811 struct file
*file
= desc
->file
;
814 struct nfs_cache_array
*array
= NULL
;
815 struct nfs_open_dir_context
*ctx
= file
->private_data
;
817 array
= kmap(desc
->page
);
818 for (i
= desc
->cache_entry_index
; i
< array
->size
; i
++) {
819 struct nfs_cache_array_entry
*ent
;
821 ent
= &array
->array
[i
];
822 if (!dir_emit(desc
->ctx
, ent
->string
.name
, ent
->string
.len
,
823 nfs_compat_user_ino64(ent
->ino
), ent
->d_type
)) {
828 if (i
< (array
->size
-1))
829 *desc
->dir_cookie
= array
->array
[i
+1].cookie
;
831 *desc
->dir_cookie
= array
->last_cookie
;
835 if (array
->eof_index
>= 0)
839 cache_page_release(desc
);
840 dfprintk(DIRCACHE
, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n",
841 (unsigned long long)*desc
->dir_cookie
, res
);
846 * If we cannot find a cookie in our cache, we suspect that this is
847 * because it points to a deleted file, so we ask the server to return
848 * whatever it thinks is the next entry. We then feed this to filldir.
849 * If all goes well, we should then be able to find our way round the
850 * cache on the next call to readdir_search_pagecache();
852 * NOTE: we cannot add the anonymous page to the pagecache because
853 * the data it contains might not be page aligned. Besides,
854 * we should already have a complete representation of the
855 * directory in the page cache by the time we get here.
858 int uncached_readdir(nfs_readdir_descriptor_t
*desc
)
860 struct page
*page
= NULL
;
862 struct inode
*inode
= file_inode(desc
->file
);
863 struct nfs_open_dir_context
*ctx
= desc
->file
->private_data
;
865 dfprintk(DIRCACHE
, "NFS: uncached_readdir() searching for cookie %Lu\n",
866 (unsigned long long)*desc
->dir_cookie
);
868 page
= alloc_page(GFP_HIGHUSER
);
874 desc
->page_index
= 0;
875 desc
->last_cookie
= *desc
->dir_cookie
;
879 status
= nfs_readdir_xdr_to_array(desc
, page
, inode
);
883 status
= nfs_do_filldir(desc
);
886 dfprintk(DIRCACHE
, "NFS: %s: returns %d\n",
890 cache_page_release(desc
);
894 /* The file offset position represents the dirent entry number. A
895 last cookie cache takes care of the common case of reading the
898 static int nfs_readdir(struct file
*file
, struct dir_context
*ctx
)
900 struct dentry
*dentry
= file_dentry(file
);
901 struct inode
*inode
= d_inode(dentry
);
902 nfs_readdir_descriptor_t my_desc
,
904 struct nfs_open_dir_context
*dir_ctx
= file
->private_data
;
906 int max_rapages
= NFS_MAX_READDIR_RAPAGES
;
908 dfprintk(FILE, "NFS: readdir(%pD2) starting at cookie %llu\n",
909 file
, (long long)ctx
->pos
);
910 nfs_inc_stats(inode
, NFSIOS_VFSGETDENTS
);
913 * ctx->pos points to the dirent entry number.
914 * *desc->dir_cookie has the cookie for the next entry. We have
915 * to either find the entry with the appropriate number or
916 * revalidate the cookie.
918 memset(desc
, 0, sizeof(*desc
));
922 desc
->dir_cookie
= &dir_ctx
->dir_cookie
;
923 desc
->decode
= NFS_PROTO(inode
)->decode_dirent
;
924 desc
->plus
= nfs_use_readdirplus(inode
, ctx
);
926 res
= nfs_readdir_alloc_pages(desc
->pvec
.pages
, max_rapages
);
930 nfs_readdir_rapages_init(desc
);
932 if (ctx
->pos
== 0 || nfs_attribute_cache_expired(inode
))
933 res
= nfs_revalidate_mapping(inode
, file
->f_mapping
);
938 res
= readdir_search_pagecache(desc
);
940 if (res
== -EBADCOOKIE
) {
942 /* This means either end of directory */
943 if (*desc
->dir_cookie
&& !desc
->eof
) {
944 /* Or that the server has 'lost' a cookie */
945 res
= uncached_readdir(desc
);
951 if (res
== -ETOOSMALL
&& desc
->plus
) {
952 clear_bit(NFS_INO_ADVISE_RDPLUS
, &NFS_I(inode
)->flags
);
953 nfs_zap_caches(inode
);
954 desc
->page_index
= 0;
962 res
= nfs_do_filldir(desc
);
965 } while (!desc
->eof
);
967 nfs_readdir_free_pages(desc
->pvec
.pages
, max_rapages
);
970 dfprintk(FILE, "NFS: readdir(%pD2) returns %d\n", file
, res
);
974 static loff_t
nfs_llseek_dir(struct file
*filp
, loff_t offset
, int whence
)
976 struct inode
*inode
= file_inode(filp
);
977 struct nfs_open_dir_context
*dir_ctx
= filp
->private_data
;
979 dfprintk(FILE, "NFS: llseek dir(%pD2, %lld, %d)\n",
980 filp
, offset
, whence
);
994 offset
+= filp
->f_pos
;
1000 if (offset
!= filp
->f_pos
) {
1001 filp
->f_pos
= offset
;
1002 dir_ctx
->dir_cookie
= 0;
1005 inode_unlock(inode
);
1010 * All directory operations under NFS are synchronous, so fsync()
1011 * is a dummy operation.
1013 static int nfs_fsync_dir(struct file
*filp
, loff_t start
, loff_t end
,
1016 struct inode
*inode
= file_inode(filp
);
1018 dfprintk(FILE, "NFS: fsync dir(%pD2) datasync %d\n", filp
, datasync
);
1021 nfs_inc_stats(inode
, NFSIOS_VFSFSYNC
);
1022 inode_unlock(inode
);
1027 * nfs_force_lookup_revalidate - Mark the directory as having changed
1028 * @dir: pointer to directory inode
1030 * This forces the revalidation code in nfs_lookup_revalidate() to do a
1031 * full lookup on all child dentries of 'dir' whenever a change occurs
1032 * on the server that might have invalidated our dcache.
1034 * The caller should be holding dir->i_lock
1036 void nfs_force_lookup_revalidate(struct inode
*dir
)
1038 NFS_I(dir
)->cache_change_attribute
++;
1040 EXPORT_SYMBOL_GPL(nfs_force_lookup_revalidate
);
1043 * A check for whether or not the parent directory has changed.
1044 * In the case it has, we assume that the dentries are untrustworthy
1045 * and may need to be looked up again.
1046 * If rcu_walk prevents us from performing a full check, return 0.
1048 static int nfs_check_verifier(struct inode
*dir
, struct dentry
*dentry
,
1051 if (IS_ROOT(dentry
))
1053 if (NFS_SERVER(dir
)->flags
& NFS_MOUNT_LOOKUP_CACHE_NONE
)
1055 if (!nfs_verify_change_attribute(dir
, dentry
->d_time
))
1057 /* Revalidate nfsi->cache_change_attribute before we declare a match */
1058 if (nfs_mapping_need_revalidate_inode(dir
)) {
1061 if (__nfs_revalidate_inode(NFS_SERVER(dir
), dir
) < 0)
1064 if (!nfs_verify_change_attribute(dir
, dentry
->d_time
))
1070 * Use intent information to check whether or not we're going to do
1071 * an O_EXCL create using this path component.
1073 static int nfs_is_exclusive_create(struct inode
*dir
, unsigned int flags
)
1075 if (NFS_PROTO(dir
)->version
== 2)
1077 return flags
& LOOKUP_EXCL
;
1081 * Inode and filehandle revalidation for lookups.
1083 * We force revalidation in the cases where the VFS sets LOOKUP_REVAL,
1084 * or if the intent information indicates that we're about to open this
1085 * particular file and the "nocto" mount flag is not set.
1089 int nfs_lookup_verify_inode(struct inode
*inode
, unsigned int flags
)
1091 struct nfs_server
*server
= NFS_SERVER(inode
);
1094 if (IS_AUTOMOUNT(inode
))
1097 if (flags
& LOOKUP_OPEN
) {
1098 switch (inode
->i_mode
& S_IFMT
) {
1100 /* A NFSv4 OPEN will revalidate later */
1101 if (server
->caps
& NFS_CAP_ATOMIC_OPEN
)
1105 if (server
->flags
& NFS_MOUNT_NOCTO
)
1107 /* NFS close-to-open cache consistency validation */
1112 /* VFS wants an on-the-wire revalidation */
1113 if (flags
& LOOKUP_REVAL
)
1116 return (inode
->i_nlink
== 0) ? -ESTALE
: 0;
1118 if (flags
& LOOKUP_RCU
)
1120 ret
= __nfs_revalidate_inode(server
, inode
);
1127 * We judge how long we want to trust negative
1128 * dentries by looking at the parent inode mtime.
1130 * If parent mtime has changed, we revalidate, else we wait for a
1131 * period corresponding to the parent's attribute cache timeout value.
1133 * If LOOKUP_RCU prevents us from performing a full check, return 1
1134 * suggesting a reval is needed.
1136 * Note that when creating a new file, or looking up a rename target,
1137 * then it shouldn't be necessary to revalidate a negative dentry.
1140 int nfs_neg_need_reval(struct inode
*dir
, struct dentry
*dentry
,
1143 if (flags
& (LOOKUP_CREATE
| LOOKUP_RENAME_TARGET
))
1145 if (NFS_SERVER(dir
)->flags
& NFS_MOUNT_LOOKUP_CACHE_NONEG
)
1147 return !nfs_check_verifier(dir
, dentry
, flags
& LOOKUP_RCU
);
1151 nfs_lookup_revalidate_done(struct inode
*dir
, struct dentry
*dentry
,
1152 struct inode
*inode
, int error
)
1156 dfprintk(LOOKUPCACHE
, "NFS: %s(%pd2) is valid\n",
1160 nfs_mark_for_revalidate(dir
);
1161 if (inode
&& S_ISDIR(inode
->i_mode
)) {
1162 /* Purge readdir caches. */
1163 nfs_zap_caches(inode
);
1165 * We can't d_drop the root of a disconnected tree:
1166 * its d_hash is on the s_anon list and d_drop() would hide
1167 * it from shrink_dcache_for_unmount(), leading to busy
1168 * inodes on unmount and further oopses.
1170 if (IS_ROOT(dentry
))
1173 dfprintk(LOOKUPCACHE
, "NFS: %s(%pd2) is invalid\n",
1177 dfprintk(LOOKUPCACHE
, "NFS: %s(%pd2) lookup returned error %d\n",
1178 __func__
, dentry
, error
);
1183 nfs_lookup_revalidate_negative(struct inode
*dir
, struct dentry
*dentry
,
1187 if (nfs_neg_need_reval(dir
, dentry
, flags
)) {
1188 if (flags
& LOOKUP_RCU
)
1192 return nfs_lookup_revalidate_done(dir
, dentry
, NULL
, ret
);
1196 nfs_lookup_revalidate_delegated(struct inode
*dir
, struct dentry
*dentry
,
1197 struct inode
*inode
)
1199 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1200 return nfs_lookup_revalidate_done(dir
, dentry
, inode
, 1);
1204 nfs_lookup_revalidate_dentry(struct inode
*dir
, struct dentry
*dentry
,
1205 struct inode
*inode
)
1207 struct nfs_fh
*fhandle
;
1208 struct nfs_fattr
*fattr
;
1209 struct nfs4_label
*label
;
1213 fhandle
= nfs_alloc_fhandle();
1214 fattr
= nfs_alloc_fattr();
1215 label
= nfs4_label_alloc(NFS_SERVER(inode
), GFP_KERNEL
);
1216 if (fhandle
== NULL
|| fattr
== NULL
|| IS_ERR(label
))
1219 ret
= NFS_PROTO(dir
)->lookup(dir
, &dentry
->d_name
, fhandle
, fattr
, label
);
1221 if (ret
== -ESTALE
|| ret
== -ENOENT
)
1226 if (nfs_compare_fh(NFS_FH(inode
), fhandle
))
1228 if (nfs_refresh_inode(inode
, fattr
) < 0)
1231 nfs_setsecurity(inode
, fattr
, label
);
1232 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1234 /* set a readdirplus hint that we had a cache miss */
1235 nfs_force_use_readdirplus(dir
);
1238 nfs_free_fattr(fattr
);
1239 nfs_free_fhandle(fhandle
);
1240 nfs4_label_free(label
);
1241 return nfs_lookup_revalidate_done(dir
, dentry
, inode
, ret
);
1245 * This is called every time the dcache has a lookup hit,
1246 * and we should check whether we can really trust that
1249 * NOTE! The hit can be a negative hit too, don't assume
1252 * If the parent directory is seen to have changed, we throw out the
1253 * cached dentry and do a new lookup.
1256 nfs_do_lookup_revalidate(struct inode
*dir
, struct dentry
*dentry
,
1259 struct inode
*inode
;
1262 nfs_inc_stats(dir
, NFSIOS_DENTRYREVALIDATE
);
1263 inode
= d_inode(dentry
);
1266 return nfs_lookup_revalidate_negative(dir
, dentry
, flags
);
1268 if (is_bad_inode(inode
)) {
1269 dfprintk(LOOKUPCACHE
, "%s: %pd2 has dud inode\n",
1274 if (NFS_PROTO(dir
)->have_delegation(inode
, FMODE_READ
))
1275 return nfs_lookup_revalidate_delegated(dir
, dentry
, inode
);
1277 /* Force a full look up iff the parent directory has changed */
1278 if (!(flags
& (LOOKUP_EXCL
| LOOKUP_REVAL
)) &&
1279 nfs_check_verifier(dir
, dentry
, flags
& LOOKUP_RCU
)) {
1280 error
= nfs_lookup_verify_inode(inode
, flags
);
1282 if (error
== -ESTALE
)
1283 nfs_zap_caches(dir
);
1286 nfs_advise_use_readdirplus(dir
);
1290 if (flags
& LOOKUP_RCU
)
1293 if (NFS_STALE(inode
))
1296 trace_nfs_lookup_revalidate_enter(dir
, dentry
, flags
);
1297 error
= nfs_lookup_revalidate_dentry(dir
, dentry
, inode
);
1298 trace_nfs_lookup_revalidate_exit(dir
, dentry
, flags
, error
);
1301 return nfs_lookup_revalidate_done(dir
, dentry
, inode
, 1);
1303 if (flags
& LOOKUP_RCU
)
1305 return nfs_lookup_revalidate_done(dir
, dentry
, inode
, 0);
1309 __nfs_lookup_revalidate(struct dentry
*dentry
, unsigned int flags
,
1310 int (*reval
)(struct inode
*, struct dentry
*, unsigned int))
1312 struct dentry
*parent
;
1316 if (flags
& LOOKUP_RCU
) {
1317 parent
= READ_ONCE(dentry
->d_parent
);
1318 dir
= d_inode_rcu(parent
);
1321 ret
= reval(dir
, dentry
, flags
);
1322 if (parent
!= READ_ONCE(dentry
->d_parent
))
1325 parent
= dget_parent(dentry
);
1326 ret
= reval(d_inode(parent
), dentry
, flags
);
1332 static int nfs_lookup_revalidate(struct dentry
*dentry
, unsigned int flags
)
1334 return __nfs_lookup_revalidate(dentry
, flags
, nfs_do_lookup_revalidate
);
1338 * A weaker form of d_revalidate for revalidating just the d_inode(dentry)
1339 * when we don't really care about the dentry name. This is called when a
1340 * pathwalk ends on a dentry that was not found via a normal lookup in the
1341 * parent dir (e.g.: ".", "..", procfs symlinks or mountpoint traversals).
1343 * In this situation, we just want to verify that the inode itself is OK
1344 * since the dentry might have changed on the server.
1346 static int nfs_weak_revalidate(struct dentry
*dentry
, unsigned int flags
)
1348 struct inode
*inode
= d_inode(dentry
);
1352 * I believe we can only get a negative dentry here in the case of a
1353 * procfs-style symlink. Just assume it's correct for now, but we may
1354 * eventually need to do something more here.
1357 dfprintk(LOOKUPCACHE
, "%s: %pd2 has negative inode\n",
1362 if (is_bad_inode(inode
)) {
1363 dfprintk(LOOKUPCACHE
, "%s: %pd2 has dud inode\n",
1368 error
= nfs_lookup_verify_inode(inode
, flags
);
1369 dfprintk(LOOKUPCACHE
, "NFS: %s: inode %lu is %s\n",
1370 __func__
, inode
->i_ino
, error
? "invalid" : "valid");
1375 * This is called from dput() when d_count is going to 0.
1377 static int nfs_dentry_delete(const struct dentry
*dentry
)
1379 dfprintk(VFS
, "NFS: dentry_delete(%pd2, %x)\n",
1380 dentry
, dentry
->d_flags
);
1382 /* Unhash any dentry with a stale inode */
1383 if (d_really_is_positive(dentry
) && NFS_STALE(d_inode(dentry
)))
1386 if (dentry
->d_flags
& DCACHE_NFSFS_RENAMED
) {
1387 /* Unhash it, so that ->d_iput() would be called */
1390 if (!(dentry
->d_sb
->s_flags
& SB_ACTIVE
)) {
1391 /* Unhash it, so that ancestors of killed async unlink
1392 * files will be cleaned up during umount */
1399 /* Ensure that we revalidate inode->i_nlink */
1400 static void nfs_drop_nlink(struct inode
*inode
)
1402 spin_lock(&inode
->i_lock
);
1403 /* drop the inode if we're reasonably sure this is the last link */
1404 if (inode
->i_nlink
> 0)
1406 NFS_I(inode
)->attr_gencount
= nfs_inc_attr_generation_counter();
1407 NFS_I(inode
)->cache_validity
|= NFS_INO_INVALID_CHANGE
1408 | NFS_INO_INVALID_CTIME
1409 | NFS_INO_INVALID_OTHER
1410 | NFS_INO_REVAL_FORCED
;
1411 spin_unlock(&inode
->i_lock
);
1415 * Called when the dentry loses inode.
1416 * We use it to clean up silly-renamed files.
1418 static void nfs_dentry_iput(struct dentry
*dentry
, struct inode
*inode
)
1420 if (S_ISDIR(inode
->i_mode
))
1421 /* drop any readdir cache as it could easily be old */
1422 NFS_I(inode
)->cache_validity
|= NFS_INO_INVALID_DATA
;
1424 if (dentry
->d_flags
& DCACHE_NFSFS_RENAMED
) {
1425 nfs_complete_unlink(dentry
, inode
);
1426 nfs_drop_nlink(inode
);
1431 static void nfs_d_release(struct dentry
*dentry
)
1433 /* free cached devname value, if it survived that far */
1434 if (unlikely(dentry
->d_fsdata
)) {
1435 if (dentry
->d_flags
& DCACHE_NFSFS_RENAMED
)
1438 kfree(dentry
->d_fsdata
);
1442 const struct dentry_operations nfs_dentry_operations
= {
1443 .d_revalidate
= nfs_lookup_revalidate
,
1444 .d_weak_revalidate
= nfs_weak_revalidate
,
1445 .d_delete
= nfs_dentry_delete
,
1446 .d_iput
= nfs_dentry_iput
,
1447 .d_automount
= nfs_d_automount
,
1448 .d_release
= nfs_d_release
,
1450 EXPORT_SYMBOL_GPL(nfs_dentry_operations
);
1452 struct dentry
*nfs_lookup(struct inode
*dir
, struct dentry
* dentry
, unsigned int flags
)
1455 struct inode
*inode
= NULL
;
1456 struct nfs_fh
*fhandle
= NULL
;
1457 struct nfs_fattr
*fattr
= NULL
;
1458 struct nfs4_label
*label
= NULL
;
1461 dfprintk(VFS
, "NFS: lookup(%pd2)\n", dentry
);
1462 nfs_inc_stats(dir
, NFSIOS_VFSLOOKUP
);
1464 if (unlikely(dentry
->d_name
.len
> NFS_SERVER(dir
)->namelen
))
1465 return ERR_PTR(-ENAMETOOLONG
);
1468 * If we're doing an exclusive create, optimize away the lookup
1469 * but don't hash the dentry.
1471 if (nfs_is_exclusive_create(dir
, flags
) || flags
& LOOKUP_RENAME_TARGET
)
1474 res
= ERR_PTR(-ENOMEM
);
1475 fhandle
= nfs_alloc_fhandle();
1476 fattr
= nfs_alloc_fattr();
1477 if (fhandle
== NULL
|| fattr
== NULL
)
1480 label
= nfs4_label_alloc(NFS_SERVER(dir
), GFP_NOWAIT
);
1484 trace_nfs_lookup_enter(dir
, dentry
, flags
);
1485 error
= NFS_PROTO(dir
)->lookup(dir
, &dentry
->d_name
, fhandle
, fattr
, label
);
1486 if (error
== -ENOENT
)
1489 res
= ERR_PTR(error
);
1492 inode
= nfs_fhget(dentry
->d_sb
, fhandle
, fattr
, label
);
1493 res
= ERR_CAST(inode
);
1497 /* Notify readdir to use READDIRPLUS */
1498 nfs_force_use_readdirplus(dir
);
1501 res
= d_splice_alias(inode
, dentry
);
1507 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1509 trace_nfs_lookup_exit(dir
, dentry
, flags
, error
);
1510 nfs4_label_free(label
);
1512 nfs_free_fattr(fattr
);
1513 nfs_free_fhandle(fhandle
);
1516 EXPORT_SYMBOL_GPL(nfs_lookup
);
1518 #if IS_ENABLED(CONFIG_NFS_V4)
1519 static int nfs4_lookup_revalidate(struct dentry
*, unsigned int);
1521 const struct dentry_operations nfs4_dentry_operations
= {
1522 .d_revalidate
= nfs4_lookup_revalidate
,
1523 .d_weak_revalidate
= nfs_weak_revalidate
,
1524 .d_delete
= nfs_dentry_delete
,
1525 .d_iput
= nfs_dentry_iput
,
1526 .d_automount
= nfs_d_automount
,
1527 .d_release
= nfs_d_release
,
1529 EXPORT_SYMBOL_GPL(nfs4_dentry_operations
);
1531 static fmode_t
flags_to_mode(int flags
)
1533 fmode_t res
= (__force fmode_t
)flags
& FMODE_EXEC
;
1534 if ((flags
& O_ACCMODE
) != O_WRONLY
)
1536 if ((flags
& O_ACCMODE
) != O_RDONLY
)
1541 static struct nfs_open_context
*create_nfs_open_context(struct dentry
*dentry
, int open_flags
, struct file
*filp
)
1543 return alloc_nfs_open_context(dentry
, flags_to_mode(open_flags
), filp
);
1546 static int do_open(struct inode
*inode
, struct file
*filp
)
1548 nfs_fscache_open_file(inode
, filp
);
1552 static int nfs_finish_open(struct nfs_open_context
*ctx
,
1553 struct dentry
*dentry
,
1554 struct file
*file
, unsigned open_flags
)
1558 err
= finish_open(file
, dentry
, do_open
);
1561 if (S_ISREG(file
->f_path
.dentry
->d_inode
->i_mode
))
1562 nfs_file_set_open_context(file
, ctx
);
1569 int nfs_atomic_open(struct inode
*dir
, struct dentry
*dentry
,
1570 struct file
*file
, unsigned open_flags
,
1573 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq
);
1574 struct nfs_open_context
*ctx
;
1576 struct iattr attr
= { .ia_valid
= ATTR_OPEN
};
1577 struct inode
*inode
;
1578 unsigned int lookup_flags
= 0;
1579 bool switched
= false;
1583 /* Expect a negative dentry */
1584 BUG_ON(d_inode(dentry
));
1586 dfprintk(VFS
, "NFS: atomic_open(%s/%lu), %pd\n",
1587 dir
->i_sb
->s_id
, dir
->i_ino
, dentry
);
1589 err
= nfs_check_flags(open_flags
);
1593 /* NFS only supports OPEN on regular files */
1594 if ((open_flags
& O_DIRECTORY
)) {
1595 if (!d_in_lookup(dentry
)) {
1597 * Hashed negative dentry with O_DIRECTORY: dentry was
1598 * revalidated and is fine, no need to perform lookup
1603 lookup_flags
= LOOKUP_OPEN
|LOOKUP_DIRECTORY
;
1607 if (dentry
->d_name
.len
> NFS_SERVER(dir
)->namelen
)
1608 return -ENAMETOOLONG
;
1610 if (open_flags
& O_CREAT
) {
1611 struct nfs_server
*server
= NFS_SERVER(dir
);
1613 if (!(server
->attr_bitmask
[2] & FATTR4_WORD2_MODE_UMASK
))
1614 mode
&= ~current_umask();
1616 attr
.ia_valid
|= ATTR_MODE
;
1617 attr
.ia_mode
= mode
;
1619 if (open_flags
& O_TRUNC
) {
1620 attr
.ia_valid
|= ATTR_SIZE
;
1624 if (!(open_flags
& O_CREAT
) && !d_in_lookup(dentry
)) {
1627 dentry
= d_alloc_parallel(dentry
->d_parent
,
1628 &dentry
->d_name
, &wq
);
1630 return PTR_ERR(dentry
);
1631 if (unlikely(!d_in_lookup(dentry
)))
1632 return finish_no_open(file
, dentry
);
1635 ctx
= create_nfs_open_context(dentry
, open_flags
, file
);
1640 trace_nfs_atomic_open_enter(dir
, ctx
, open_flags
);
1641 inode
= NFS_PROTO(dir
)->open_context(dir
, ctx
, open_flags
, &attr
, &created
);
1643 file
->f_mode
|= FMODE_CREATED
;
1644 if (IS_ERR(inode
)) {
1645 err
= PTR_ERR(inode
);
1646 trace_nfs_atomic_open_exit(dir
, ctx
, open_flags
, err
);
1647 put_nfs_open_context(ctx
);
1651 d_splice_alias(NULL
, dentry
);
1652 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1658 if (!(open_flags
& O_NOFOLLOW
))
1668 err
= nfs_finish_open(ctx
, ctx
->dentry
, file
, open_flags
);
1669 trace_nfs_atomic_open_exit(dir
, ctx
, open_flags
, err
);
1670 put_nfs_open_context(ctx
);
1672 if (unlikely(switched
)) {
1673 d_lookup_done(dentry
);
1679 res
= nfs_lookup(dir
, dentry
, lookup_flags
);
1681 d_lookup_done(dentry
);
1688 return PTR_ERR(res
);
1689 return finish_no_open(file
, res
);
1691 EXPORT_SYMBOL_GPL(nfs_atomic_open
);
1694 nfs4_do_lookup_revalidate(struct inode
*dir
, struct dentry
*dentry
,
1697 struct inode
*inode
;
1699 if (!(flags
& LOOKUP_OPEN
) || (flags
& LOOKUP_DIRECTORY
))
1701 if (d_mountpoint(dentry
))
1704 inode
= d_inode(dentry
);
1706 /* We can't create new files in nfs_open_revalidate(), so we
1707 * optimize away revalidation of negative dentries.
1712 if (NFS_PROTO(dir
)->have_delegation(inode
, FMODE_READ
))
1713 return nfs_lookup_revalidate_delegated(dir
, dentry
, inode
);
1715 /* NFS only supports OPEN on regular files */
1716 if (!S_ISREG(inode
->i_mode
))
1719 /* We cannot do exclusive creation on a positive dentry */
1720 if (flags
& (LOOKUP_EXCL
| LOOKUP_REVAL
))
1723 /* Check if the directory changed */
1724 if (!nfs_check_verifier(dir
, dentry
, flags
& LOOKUP_RCU
))
1727 /* Let f_op->open() actually open (and revalidate) the file */
1730 if (flags
& LOOKUP_RCU
)
1732 return nfs_lookup_revalidate_dentry(dir
, dentry
, inode
);
1735 return nfs_do_lookup_revalidate(dir
, dentry
, flags
);
1738 static int nfs4_lookup_revalidate(struct dentry
*dentry
, unsigned int flags
)
1740 return __nfs_lookup_revalidate(dentry
, flags
,
1741 nfs4_do_lookup_revalidate
);
1744 #endif /* CONFIG_NFSV4 */
1747 * Code common to create, mkdir, and mknod.
1749 int nfs_instantiate(struct dentry
*dentry
, struct nfs_fh
*fhandle
,
1750 struct nfs_fattr
*fattr
,
1751 struct nfs4_label
*label
)
1753 struct dentry
*parent
= dget_parent(dentry
);
1754 struct inode
*dir
= d_inode(parent
);
1755 struct inode
*inode
;
1757 int error
= -EACCES
;
1761 /* We may have been initialized further down */
1762 if (d_really_is_positive(dentry
))
1764 if (fhandle
->size
== 0) {
1765 error
= NFS_PROTO(dir
)->lookup(dir
, &dentry
->d_name
, fhandle
, fattr
, NULL
);
1769 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1770 if (!(fattr
->valid
& NFS_ATTR_FATTR
)) {
1771 struct nfs_server
*server
= NFS_SB(dentry
->d_sb
);
1772 error
= server
->nfs_client
->rpc_ops
->getattr(server
, fhandle
,
1777 inode
= nfs_fhget(dentry
->d_sb
, fhandle
, fattr
, label
);
1778 d
= d_splice_alias(inode
, dentry
);
1788 nfs_mark_for_revalidate(dir
);
1792 EXPORT_SYMBOL_GPL(nfs_instantiate
);
1795 * Following a failed create operation, we drop the dentry rather
1796 * than retain a negative dentry. This avoids a problem in the event
1797 * that the operation succeeded on the server, but an error in the
1798 * reply path made it appear to have failed.
1800 int nfs_create(struct inode
*dir
, struct dentry
*dentry
,
1801 umode_t mode
, bool excl
)
1804 int open_flags
= excl
? O_CREAT
| O_EXCL
: O_CREAT
;
1807 dfprintk(VFS
, "NFS: create(%s/%lu), %pd\n",
1808 dir
->i_sb
->s_id
, dir
->i_ino
, dentry
);
1810 attr
.ia_mode
= mode
;
1811 attr
.ia_valid
= ATTR_MODE
;
1813 trace_nfs_create_enter(dir
, dentry
, open_flags
);
1814 error
= NFS_PROTO(dir
)->create(dir
, dentry
, &attr
, open_flags
);
1815 trace_nfs_create_exit(dir
, dentry
, open_flags
, error
);
1823 EXPORT_SYMBOL_GPL(nfs_create
);
1826 * See comments for nfs_proc_create regarding failed operations.
1829 nfs_mknod(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
, dev_t rdev
)
1834 dfprintk(VFS
, "NFS: mknod(%s/%lu), %pd\n",
1835 dir
->i_sb
->s_id
, dir
->i_ino
, dentry
);
1837 attr
.ia_mode
= mode
;
1838 attr
.ia_valid
= ATTR_MODE
;
1840 trace_nfs_mknod_enter(dir
, dentry
);
1841 status
= NFS_PROTO(dir
)->mknod(dir
, dentry
, &attr
, rdev
);
1842 trace_nfs_mknod_exit(dir
, dentry
, status
);
1850 EXPORT_SYMBOL_GPL(nfs_mknod
);
1853 * See comments for nfs_proc_create regarding failed operations.
1855 int nfs_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
1860 dfprintk(VFS
, "NFS: mkdir(%s/%lu), %pd\n",
1861 dir
->i_sb
->s_id
, dir
->i_ino
, dentry
);
1863 attr
.ia_valid
= ATTR_MODE
;
1864 attr
.ia_mode
= mode
| S_IFDIR
;
1866 trace_nfs_mkdir_enter(dir
, dentry
);
1867 error
= NFS_PROTO(dir
)->mkdir(dir
, dentry
, &attr
);
1868 trace_nfs_mkdir_exit(dir
, dentry
, error
);
1876 EXPORT_SYMBOL_GPL(nfs_mkdir
);
1878 static void nfs_dentry_handle_enoent(struct dentry
*dentry
)
1880 if (simple_positive(dentry
))
1884 int nfs_rmdir(struct inode
*dir
, struct dentry
*dentry
)
1888 dfprintk(VFS
, "NFS: rmdir(%s/%lu), %pd\n",
1889 dir
->i_sb
->s_id
, dir
->i_ino
, dentry
);
1891 trace_nfs_rmdir_enter(dir
, dentry
);
1892 if (d_really_is_positive(dentry
)) {
1893 down_write(&NFS_I(d_inode(dentry
))->rmdir_sem
);
1894 error
= NFS_PROTO(dir
)->rmdir(dir
, &dentry
->d_name
);
1895 /* Ensure the VFS deletes this inode */
1898 clear_nlink(d_inode(dentry
));
1901 nfs_dentry_handle_enoent(dentry
);
1903 up_write(&NFS_I(d_inode(dentry
))->rmdir_sem
);
1905 error
= NFS_PROTO(dir
)->rmdir(dir
, &dentry
->d_name
);
1906 trace_nfs_rmdir_exit(dir
, dentry
, error
);
1910 EXPORT_SYMBOL_GPL(nfs_rmdir
);
1913 * Remove a file after making sure there are no pending writes,
1914 * and after checking that the file has only one user.
1916 * We invalidate the attribute cache and free the inode prior to the operation
1917 * to avoid possible races if the server reuses the inode.
1919 static int nfs_safe_remove(struct dentry
*dentry
)
1921 struct inode
*dir
= d_inode(dentry
->d_parent
);
1922 struct inode
*inode
= d_inode(dentry
);
1925 dfprintk(VFS
, "NFS: safe_remove(%pd2)\n", dentry
);
1927 /* If the dentry was sillyrenamed, we simply call d_delete() */
1928 if (dentry
->d_flags
& DCACHE_NFSFS_RENAMED
) {
1933 trace_nfs_remove_enter(dir
, dentry
);
1934 if (inode
!= NULL
) {
1935 error
= NFS_PROTO(dir
)->remove(dir
, dentry
);
1937 nfs_drop_nlink(inode
);
1939 error
= NFS_PROTO(dir
)->remove(dir
, dentry
);
1940 if (error
== -ENOENT
)
1941 nfs_dentry_handle_enoent(dentry
);
1942 trace_nfs_remove_exit(dir
, dentry
, error
);
1947 /* We do silly rename. In case sillyrename() returns -EBUSY, the inode
1948 * belongs to an active ".nfs..." file and we return -EBUSY.
1950 * If sillyrename() returns 0, we do nothing, otherwise we unlink.
1952 int nfs_unlink(struct inode
*dir
, struct dentry
*dentry
)
1955 int need_rehash
= 0;
1957 dfprintk(VFS
, "NFS: unlink(%s/%lu, %pd)\n", dir
->i_sb
->s_id
,
1958 dir
->i_ino
, dentry
);
1960 trace_nfs_unlink_enter(dir
, dentry
);
1961 spin_lock(&dentry
->d_lock
);
1962 if (d_count(dentry
) > 1) {
1963 spin_unlock(&dentry
->d_lock
);
1964 /* Start asynchronous writeout of the inode */
1965 write_inode_now(d_inode(dentry
), 0);
1966 error
= nfs_sillyrename(dir
, dentry
);
1969 if (!d_unhashed(dentry
)) {
1973 spin_unlock(&dentry
->d_lock
);
1974 error
= nfs_safe_remove(dentry
);
1975 if (!error
|| error
== -ENOENT
) {
1976 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1977 } else if (need_rehash
)
1980 trace_nfs_unlink_exit(dir
, dentry
, error
);
1983 EXPORT_SYMBOL_GPL(nfs_unlink
);
1986 * To create a symbolic link, most file systems instantiate a new inode,
1987 * add a page to it containing the path, then write it out to the disk
1988 * using prepare_write/commit_write.
1990 * Unfortunately the NFS client can't create the in-core inode first
1991 * because it needs a file handle to create an in-core inode (see
1992 * fs/nfs/inode.c:nfs_fhget). We only have a file handle *after* the
1993 * symlink request has completed on the server.
1995 * So instead we allocate a raw page, copy the symname into it, then do
1996 * the SYMLINK request with the page as the buffer. If it succeeds, we
1997 * now have a new file handle and can instantiate an in-core NFS inode
1998 * and move the raw page into its mapping.
2000 int nfs_symlink(struct inode
*dir
, struct dentry
*dentry
, const char *symname
)
2005 unsigned int pathlen
= strlen(symname
);
2008 dfprintk(VFS
, "NFS: symlink(%s/%lu, %pd, %s)\n", dir
->i_sb
->s_id
,
2009 dir
->i_ino
, dentry
, symname
);
2011 if (pathlen
> PAGE_SIZE
)
2012 return -ENAMETOOLONG
;
2014 attr
.ia_mode
= S_IFLNK
| S_IRWXUGO
;
2015 attr
.ia_valid
= ATTR_MODE
;
2017 page
= alloc_page(GFP_USER
);
2021 kaddr
= page_address(page
);
2022 memcpy(kaddr
, symname
, pathlen
);
2023 if (pathlen
< PAGE_SIZE
)
2024 memset(kaddr
+ pathlen
, 0, PAGE_SIZE
- pathlen
);
2026 trace_nfs_symlink_enter(dir
, dentry
);
2027 error
= NFS_PROTO(dir
)->symlink(dir
, dentry
, page
, pathlen
, &attr
);
2028 trace_nfs_symlink_exit(dir
, dentry
, error
);
2030 dfprintk(VFS
, "NFS: symlink(%s/%lu, %pd, %s) error %d\n",
2031 dir
->i_sb
->s_id
, dir
->i_ino
,
2032 dentry
, symname
, error
);
2039 * No big deal if we can't add this page to the page cache here.
2040 * READLINK will get the missing page from the server if needed.
2042 if (!add_to_page_cache_lru(page
, d_inode(dentry
)->i_mapping
, 0,
2044 SetPageUptodate(page
);
2047 * add_to_page_cache_lru() grabs an extra page refcount.
2048 * Drop it here to avoid leaking this page later.
2056 EXPORT_SYMBOL_GPL(nfs_symlink
);
2059 nfs_link(struct dentry
*old_dentry
, struct inode
*dir
, struct dentry
*dentry
)
2061 struct inode
*inode
= d_inode(old_dentry
);
2064 dfprintk(VFS
, "NFS: link(%pd2 -> %pd2)\n",
2065 old_dentry
, dentry
);
2067 trace_nfs_link_enter(inode
, dir
, dentry
);
2069 error
= NFS_PROTO(dir
)->link(inode
, dir
, &dentry
->d_name
);
2072 d_add(dentry
, inode
);
2074 trace_nfs_link_exit(inode
, dir
, dentry
, error
);
2077 EXPORT_SYMBOL_GPL(nfs_link
);
2081 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a
2082 * different file handle for the same inode after a rename (e.g. when
2083 * moving to a different directory). A fail-safe method to do so would
2084 * be to look up old_dir/old_name, create a link to new_dir/new_name and
2085 * rename the old file using the sillyrename stuff. This way, the original
2086 * file in old_dir will go away when the last process iput()s the inode.
2090 * It actually works quite well. One needs to have the possibility for
2091 * at least one ".nfs..." file in each directory the file ever gets
2092 * moved or linked to which happens automagically with the new
2093 * implementation that only depends on the dcache stuff instead of
2094 * using the inode layer
2096 * Unfortunately, things are a little more complicated than indicated
2097 * above. For a cross-directory move, we want to make sure we can get
2098 * rid of the old inode after the operation. This means there must be
2099 * no pending writes (if it's a file), and the use count must be 1.
2100 * If these conditions are met, we can drop the dentries before doing
2103 int nfs_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
2104 struct inode
*new_dir
, struct dentry
*new_dentry
,
2107 struct inode
*old_inode
= d_inode(old_dentry
);
2108 struct inode
*new_inode
= d_inode(new_dentry
);
2109 struct dentry
*dentry
= NULL
, *rehash
= NULL
;
2110 struct rpc_task
*task
;
2116 dfprintk(VFS
, "NFS: rename(%pd2 -> %pd2, ct=%d)\n",
2117 old_dentry
, new_dentry
,
2118 d_count(new_dentry
));
2120 trace_nfs_rename_enter(old_dir
, old_dentry
, new_dir
, new_dentry
);
2122 * For non-directories, check whether the target is busy and if so,
2123 * make a copy of the dentry and then do a silly-rename. If the
2124 * silly-rename succeeds, the copied dentry is hashed and becomes
2127 if (new_inode
&& !S_ISDIR(new_inode
->i_mode
)) {
2129 * To prevent any new references to the target during the
2130 * rename, we unhash the dentry in advance.
2132 if (!d_unhashed(new_dentry
)) {
2134 rehash
= new_dentry
;
2137 if (d_count(new_dentry
) > 2) {
2140 /* copy the target dentry's name */
2141 dentry
= d_alloc(new_dentry
->d_parent
,
2142 &new_dentry
->d_name
);
2146 /* silly-rename the existing target ... */
2147 err
= nfs_sillyrename(new_dir
, new_dentry
);
2151 new_dentry
= dentry
;
2157 task
= nfs_async_rename(old_dir
, new_dir
, old_dentry
, new_dentry
, NULL
);
2159 error
= PTR_ERR(task
);
2163 error
= rpc_wait_for_completion_task(task
);
2165 ((struct nfs_renamedata
*)task
->tk_calldata
)->cancelled
= 1;
2166 /* Paired with the atomic_dec_and_test() barrier in rpc_do_put_task() */
2169 error
= task
->tk_status
;
2171 /* Ensure the inode attributes are revalidated */
2173 spin_lock(&old_inode
->i_lock
);
2174 NFS_I(old_inode
)->attr_gencount
= nfs_inc_attr_generation_counter();
2175 NFS_I(old_inode
)->cache_validity
|= NFS_INO_INVALID_CHANGE
2176 | NFS_INO_INVALID_CTIME
2177 | NFS_INO_REVAL_FORCED
;
2178 spin_unlock(&old_inode
->i_lock
);
2183 trace_nfs_rename_exit(old_dir
, old_dentry
,
2184 new_dir
, new_dentry
, error
);
2186 if (new_inode
!= NULL
)
2187 nfs_drop_nlink(new_inode
);
2189 * The d_move() should be here instead of in an async RPC completion
2190 * handler because we need the proper locks to move the dentry. If
2191 * we're interrupted by a signal, the async RPC completion handler
2192 * should mark the directories for revalidation.
2194 d_move(old_dentry
, new_dentry
);
2195 nfs_set_verifier(old_dentry
,
2196 nfs_save_change_attribute(new_dir
));
2197 } else if (error
== -ENOENT
)
2198 nfs_dentry_handle_enoent(old_dentry
);
2200 /* new dentry created? */
2205 EXPORT_SYMBOL_GPL(nfs_rename
);
2207 static DEFINE_SPINLOCK(nfs_access_lru_lock
);
2208 static LIST_HEAD(nfs_access_lru_list
);
2209 static atomic_long_t nfs_access_nr_entries
;
2211 static unsigned long nfs_access_max_cachesize
= ULONG_MAX
;
2212 module_param(nfs_access_max_cachesize
, ulong
, 0644);
2213 MODULE_PARM_DESC(nfs_access_max_cachesize
, "NFS access maximum total cache length");
2215 static void nfs_access_free_entry(struct nfs_access_entry
*entry
)
2217 put_cred(entry
->cred
);
2218 kfree_rcu(entry
, rcu_head
);
2219 smp_mb__before_atomic();
2220 atomic_long_dec(&nfs_access_nr_entries
);
2221 smp_mb__after_atomic();
2224 static void nfs_access_free_list(struct list_head
*head
)
2226 struct nfs_access_entry
*cache
;
2228 while (!list_empty(head
)) {
2229 cache
= list_entry(head
->next
, struct nfs_access_entry
, lru
);
2230 list_del(&cache
->lru
);
2231 nfs_access_free_entry(cache
);
2235 static unsigned long
2236 nfs_do_access_cache_scan(unsigned int nr_to_scan
)
2239 struct nfs_inode
*nfsi
, *next
;
2240 struct nfs_access_entry
*cache
;
2243 spin_lock(&nfs_access_lru_lock
);
2244 list_for_each_entry_safe(nfsi
, next
, &nfs_access_lru_list
, access_cache_inode_lru
) {
2245 struct inode
*inode
;
2247 if (nr_to_scan
-- == 0)
2249 inode
= &nfsi
->vfs_inode
;
2250 spin_lock(&inode
->i_lock
);
2251 if (list_empty(&nfsi
->access_cache_entry_lru
))
2252 goto remove_lru_entry
;
2253 cache
= list_entry(nfsi
->access_cache_entry_lru
.next
,
2254 struct nfs_access_entry
, lru
);
2255 list_move(&cache
->lru
, &head
);
2256 rb_erase(&cache
->rb_node
, &nfsi
->access_cache
);
2258 if (!list_empty(&nfsi
->access_cache_entry_lru
))
2259 list_move_tail(&nfsi
->access_cache_inode_lru
,
2260 &nfs_access_lru_list
);
2263 list_del_init(&nfsi
->access_cache_inode_lru
);
2264 smp_mb__before_atomic();
2265 clear_bit(NFS_INO_ACL_LRU_SET
, &nfsi
->flags
);
2266 smp_mb__after_atomic();
2268 spin_unlock(&inode
->i_lock
);
2270 spin_unlock(&nfs_access_lru_lock
);
2271 nfs_access_free_list(&head
);
2276 nfs_access_cache_scan(struct shrinker
*shrink
, struct shrink_control
*sc
)
2278 int nr_to_scan
= sc
->nr_to_scan
;
2279 gfp_t gfp_mask
= sc
->gfp_mask
;
2281 if ((gfp_mask
& GFP_KERNEL
) != GFP_KERNEL
)
2283 return nfs_do_access_cache_scan(nr_to_scan
);
2288 nfs_access_cache_count(struct shrinker
*shrink
, struct shrink_control
*sc
)
2290 return vfs_pressure_ratio(atomic_long_read(&nfs_access_nr_entries
));
2294 nfs_access_cache_enforce_limit(void)
2296 long nr_entries
= atomic_long_read(&nfs_access_nr_entries
);
2298 unsigned int nr_to_scan
;
2300 if (nr_entries
< 0 || nr_entries
<= nfs_access_max_cachesize
)
2303 diff
= nr_entries
- nfs_access_max_cachesize
;
2304 if (diff
< nr_to_scan
)
2306 nfs_do_access_cache_scan(nr_to_scan
);
2309 static void __nfs_access_zap_cache(struct nfs_inode
*nfsi
, struct list_head
*head
)
2311 struct rb_root
*root_node
= &nfsi
->access_cache
;
2313 struct nfs_access_entry
*entry
;
2315 /* Unhook entries from the cache */
2316 while ((n
= rb_first(root_node
)) != NULL
) {
2317 entry
= rb_entry(n
, struct nfs_access_entry
, rb_node
);
2318 rb_erase(n
, root_node
);
2319 list_move(&entry
->lru
, head
);
2321 nfsi
->cache_validity
&= ~NFS_INO_INVALID_ACCESS
;
2324 void nfs_access_zap_cache(struct inode
*inode
)
2328 if (test_bit(NFS_INO_ACL_LRU_SET
, &NFS_I(inode
)->flags
) == 0)
2330 /* Remove from global LRU init */
2331 spin_lock(&nfs_access_lru_lock
);
2332 if (test_and_clear_bit(NFS_INO_ACL_LRU_SET
, &NFS_I(inode
)->flags
))
2333 list_del_init(&NFS_I(inode
)->access_cache_inode_lru
);
2335 spin_lock(&inode
->i_lock
);
2336 __nfs_access_zap_cache(NFS_I(inode
), &head
);
2337 spin_unlock(&inode
->i_lock
);
2338 spin_unlock(&nfs_access_lru_lock
);
2339 nfs_access_free_list(&head
);
2341 EXPORT_SYMBOL_GPL(nfs_access_zap_cache
);
2343 static struct nfs_access_entry
*nfs_access_search_rbtree(struct inode
*inode
, const struct cred
*cred
)
2345 struct rb_node
*n
= NFS_I(inode
)->access_cache
.rb_node
;
2348 struct nfs_access_entry
*entry
=
2349 rb_entry(n
, struct nfs_access_entry
, rb_node
);
2350 int cmp
= cred_fscmp(cred
, entry
->cred
);
2362 static int nfs_access_get_cached(struct inode
*inode
, const struct cred
*cred
, struct nfs_access_entry
*res
, bool may_block
)
2364 struct nfs_inode
*nfsi
= NFS_I(inode
);
2365 struct nfs_access_entry
*cache
;
2369 spin_lock(&inode
->i_lock
);
2371 if (nfsi
->cache_validity
& NFS_INO_INVALID_ACCESS
)
2373 cache
= nfs_access_search_rbtree(inode
, cred
);
2377 /* Found an entry, is our attribute cache valid? */
2378 if (!nfs_check_cache_invalid(inode
, NFS_INO_INVALID_ACCESS
))
2385 spin_unlock(&inode
->i_lock
);
2386 err
= __nfs_revalidate_inode(NFS_SERVER(inode
), inode
);
2389 spin_lock(&inode
->i_lock
);
2392 res
->cred
= cache
->cred
;
2393 res
->mask
= cache
->mask
;
2394 list_move_tail(&cache
->lru
, &nfsi
->access_cache_entry_lru
);
2397 spin_unlock(&inode
->i_lock
);
2400 spin_unlock(&inode
->i_lock
);
2401 nfs_access_zap_cache(inode
);
2405 static int nfs_access_get_cached_rcu(struct inode
*inode
, const struct cred
*cred
, struct nfs_access_entry
*res
)
2407 /* Only check the most recently returned cache entry,
2408 * but do it without locking.
2410 struct nfs_inode
*nfsi
= NFS_I(inode
);
2411 struct nfs_access_entry
*cache
;
2413 struct list_head
*lh
;
2416 if (nfsi
->cache_validity
& NFS_INO_INVALID_ACCESS
)
2418 lh
= rcu_dereference(nfsi
->access_cache_entry_lru
.prev
);
2419 cache
= list_entry(lh
, struct nfs_access_entry
, lru
);
2420 if (lh
== &nfsi
->access_cache_entry_lru
||
2421 cred
!= cache
->cred
)
2425 if (nfs_check_cache_invalid(inode
, NFS_INO_INVALID_ACCESS
))
2427 res
->cred
= cache
->cred
;
2428 res
->mask
= cache
->mask
;
2435 static void nfs_access_add_rbtree(struct inode
*inode
, struct nfs_access_entry
*set
)
2437 struct nfs_inode
*nfsi
= NFS_I(inode
);
2438 struct rb_root
*root_node
= &nfsi
->access_cache
;
2439 struct rb_node
**p
= &root_node
->rb_node
;
2440 struct rb_node
*parent
= NULL
;
2441 struct nfs_access_entry
*entry
;
2444 spin_lock(&inode
->i_lock
);
2445 while (*p
!= NULL
) {
2447 entry
= rb_entry(parent
, struct nfs_access_entry
, rb_node
);
2448 cmp
= cred_fscmp(set
->cred
, entry
->cred
);
2451 p
= &parent
->rb_left
;
2453 p
= &parent
->rb_right
;
2457 rb_link_node(&set
->rb_node
, parent
, p
);
2458 rb_insert_color(&set
->rb_node
, root_node
);
2459 list_add_tail(&set
->lru
, &nfsi
->access_cache_entry_lru
);
2460 spin_unlock(&inode
->i_lock
);
2463 rb_replace_node(parent
, &set
->rb_node
, root_node
);
2464 list_add_tail(&set
->lru
, &nfsi
->access_cache_entry_lru
);
2465 list_del(&entry
->lru
);
2466 spin_unlock(&inode
->i_lock
);
2467 nfs_access_free_entry(entry
);
2470 void nfs_access_add_cache(struct inode
*inode
, struct nfs_access_entry
*set
)
2472 struct nfs_access_entry
*cache
= kmalloc(sizeof(*cache
), GFP_KERNEL
);
2475 RB_CLEAR_NODE(&cache
->rb_node
);
2476 cache
->cred
= get_cred(set
->cred
);
2477 cache
->mask
= set
->mask
;
2479 /* The above field assignments must be visible
2480 * before this item appears on the lru. We cannot easily
2481 * use rcu_assign_pointer, so just force the memory barrier.
2484 nfs_access_add_rbtree(inode
, cache
);
2486 /* Update accounting */
2487 smp_mb__before_atomic();
2488 atomic_long_inc(&nfs_access_nr_entries
);
2489 smp_mb__after_atomic();
2491 /* Add inode to global LRU list */
2492 if (!test_bit(NFS_INO_ACL_LRU_SET
, &NFS_I(inode
)->flags
)) {
2493 spin_lock(&nfs_access_lru_lock
);
2494 if (!test_and_set_bit(NFS_INO_ACL_LRU_SET
, &NFS_I(inode
)->flags
))
2495 list_add_tail(&NFS_I(inode
)->access_cache_inode_lru
,
2496 &nfs_access_lru_list
);
2497 spin_unlock(&nfs_access_lru_lock
);
2499 nfs_access_cache_enforce_limit();
2501 EXPORT_SYMBOL_GPL(nfs_access_add_cache
);
2503 #define NFS_MAY_READ (NFS_ACCESS_READ)
2504 #define NFS_MAY_WRITE (NFS_ACCESS_MODIFY | \
2505 NFS_ACCESS_EXTEND | \
2507 #define NFS_FILE_MAY_WRITE (NFS_ACCESS_MODIFY | \
2509 #define NFS_DIR_MAY_WRITE NFS_MAY_WRITE
2510 #define NFS_MAY_LOOKUP (NFS_ACCESS_LOOKUP)
2511 #define NFS_MAY_EXECUTE (NFS_ACCESS_EXECUTE)
2513 nfs_access_calc_mask(u32 access_result
, umode_t umode
)
2517 if (access_result
& NFS_MAY_READ
)
2519 if (S_ISDIR(umode
)) {
2520 if ((access_result
& NFS_DIR_MAY_WRITE
) == NFS_DIR_MAY_WRITE
)
2522 if ((access_result
& NFS_MAY_LOOKUP
) == NFS_MAY_LOOKUP
)
2524 } else if (S_ISREG(umode
)) {
2525 if ((access_result
& NFS_FILE_MAY_WRITE
) == NFS_FILE_MAY_WRITE
)
2527 if ((access_result
& NFS_MAY_EXECUTE
) == NFS_MAY_EXECUTE
)
2529 } else if (access_result
& NFS_MAY_WRITE
)
2534 void nfs_access_set_mask(struct nfs_access_entry
*entry
, u32 access_result
)
2536 entry
->mask
= access_result
;
2538 EXPORT_SYMBOL_GPL(nfs_access_set_mask
);
2540 static int nfs_do_access(struct inode
*inode
, const struct cred
*cred
, int mask
)
2542 struct nfs_access_entry cache
;
2543 bool may_block
= (mask
& MAY_NOT_BLOCK
) == 0;
2547 trace_nfs_access_enter(inode
);
2549 status
= nfs_access_get_cached_rcu(inode
, cred
, &cache
);
2551 status
= nfs_access_get_cached(inode
, cred
, &cache
, may_block
);
2560 * Determine which access bits we want to ask for...
2562 cache
.mask
= NFS_ACCESS_READ
| NFS_ACCESS_MODIFY
| NFS_ACCESS_EXTEND
;
2563 if (S_ISDIR(inode
->i_mode
))
2564 cache
.mask
|= NFS_ACCESS_DELETE
| NFS_ACCESS_LOOKUP
;
2566 cache
.mask
|= NFS_ACCESS_EXECUTE
;
2568 status
= NFS_PROTO(inode
)->access(inode
, &cache
);
2570 if (status
== -ESTALE
) {
2571 nfs_zap_caches(inode
);
2572 if (!S_ISDIR(inode
->i_mode
))
2573 set_bit(NFS_INO_STALE
, &NFS_I(inode
)->flags
);
2577 nfs_access_add_cache(inode
, &cache
);
2579 cache_mask
= nfs_access_calc_mask(cache
.mask
, inode
->i_mode
);
2580 if ((mask
& ~cache_mask
& (MAY_READ
| MAY_WRITE
| MAY_EXEC
)) != 0)
2583 trace_nfs_access_exit(inode
, status
);
2587 static int nfs_open_permission_mask(int openflags
)
2591 if (openflags
& __FMODE_EXEC
) {
2592 /* ONLY check exec rights */
2595 if ((openflags
& O_ACCMODE
) != O_WRONLY
)
2597 if ((openflags
& O_ACCMODE
) != O_RDONLY
)
2604 int nfs_may_open(struct inode
*inode
, const struct cred
*cred
, int openflags
)
2606 return nfs_do_access(inode
, cred
, nfs_open_permission_mask(openflags
));
2608 EXPORT_SYMBOL_GPL(nfs_may_open
);
2610 static int nfs_execute_ok(struct inode
*inode
, int mask
)
2612 struct nfs_server
*server
= NFS_SERVER(inode
);
2615 if (S_ISDIR(inode
->i_mode
))
2617 if (nfs_check_cache_invalid(inode
, NFS_INO_INVALID_OTHER
)) {
2618 if (mask
& MAY_NOT_BLOCK
)
2620 ret
= __nfs_revalidate_inode(server
, inode
);
2622 if (ret
== 0 && !execute_ok(inode
))
2627 int nfs_permission(struct inode
*inode
, int mask
)
2629 const struct cred
*cred
= current_cred();
2632 nfs_inc_stats(inode
, NFSIOS_VFSACCESS
);
2634 if ((mask
& (MAY_READ
| MAY_WRITE
| MAY_EXEC
)) == 0)
2636 /* Is this sys_access() ? */
2637 if (mask
& (MAY_ACCESS
| MAY_CHDIR
))
2640 switch (inode
->i_mode
& S_IFMT
) {
2644 if ((mask
& MAY_OPEN
) &&
2645 nfs_server_capable(inode
, NFS_CAP_ATOMIC_OPEN
))
2650 * Optimize away all write operations, since the server
2651 * will check permissions when we perform the op.
2653 if ((mask
& MAY_WRITE
) && !(mask
& MAY_READ
))
2658 if (!NFS_PROTO(inode
)->access
)
2661 /* Always try fast lookups first */
2663 res
= nfs_do_access(inode
, cred
, mask
|MAY_NOT_BLOCK
);
2665 if (res
== -ECHILD
&& !(mask
& MAY_NOT_BLOCK
)) {
2666 /* Fast lookup failed, try the slow way */
2667 res
= nfs_do_access(inode
, cred
, mask
);
2670 if (!res
&& (mask
& MAY_EXEC
))
2671 res
= nfs_execute_ok(inode
, mask
);
2673 dfprintk(VFS
, "NFS: permission(%s/%lu), mask=0x%x, res=%d\n",
2674 inode
->i_sb
->s_id
, inode
->i_ino
, mask
, res
);
2677 if (mask
& MAY_NOT_BLOCK
)
2680 res
= nfs_revalidate_inode(NFS_SERVER(inode
), inode
);
2682 res
= generic_permission(inode
, mask
);
2685 EXPORT_SYMBOL_GPL(nfs_permission
);
2689 * version-control: t
2690 * kept-new-versions: 5