4 * Copyright (C) 1992 Rick Sladkey
6 * nfs directory handling functions
8 * 10 Apr 1996 Added silly rename for unlink --okir
9 * 28 Sep 1996 Improved directory cache --okir
10 * 23 Aug 1997 Claus Heine claus@momo.math.rwth-aachen.de
11 * Re-implemented silly rename for unlink, newly implemented
12 * silly rename for nfs_rename() following the suggestions
13 * of Olaf Kirch (okir) found in this file.
14 * Following Linus comments on my original hack, this version
15 * depends only on the dcache stuff and doesn't touch the inode
16 * layer (iput() and friends).
17 * 6 Jun 1999 Cache readdir lookups in the page cache. -DaveM
20 #include <linux/module.h>
21 #include <linux/time.h>
22 #include <linux/errno.h>
23 #include <linux/stat.h>
24 #include <linux/fcntl.h>
25 #include <linux/string.h>
26 #include <linux/kernel.h>
27 #include <linux/slab.h>
29 #include <linux/sunrpc/clnt.h>
30 #include <linux/nfs_fs.h>
31 #include <linux/nfs_mount.h>
32 #include <linux/pagemap.h>
33 #include <linux/pagevec.h>
34 #include <linux/namei.h>
35 #include <linux/mount.h>
36 #include <linux/swap.h>
37 #include <linux/sched.h>
38 #include <linux/kmemleak.h>
39 #include <linux/xattr.h>
41 #include "delegation.h"
48 /* #define NFS_DEBUG_VERBOSE 1 */
50 static int nfs_opendir(struct inode
*, struct file
*);
51 static int nfs_closedir(struct inode
*, struct file
*);
52 static int nfs_readdir(struct file
*, struct dir_context
*);
53 static int nfs_fsync_dir(struct file
*, loff_t
, loff_t
, int);
54 static loff_t
nfs_llseek_dir(struct file
*, loff_t
, int);
55 static void nfs_readdir_clear_array(struct page
*);
57 const struct file_operations nfs_dir_operations
= {
58 .llseek
= nfs_llseek_dir
,
59 .read
= generic_read_dir
,
60 .iterate
= nfs_readdir
,
62 .release
= nfs_closedir
,
63 .fsync
= nfs_fsync_dir
,
66 const struct address_space_operations nfs_dir_aops
= {
67 .freepage
= nfs_readdir_clear_array
,
70 static struct nfs_open_dir_context
*alloc_nfs_open_dir_context(struct inode
*dir
, const struct cred
*cred
)
72 struct nfs_inode
*nfsi
= NFS_I(dir
);
73 struct nfs_open_dir_context
*ctx
;
74 ctx
= kmalloc(sizeof(*ctx
), GFP_KERNEL
);
77 ctx
->attr_gencount
= nfsi
->attr_gencount
;
80 ctx
->cred
= get_cred(cred
);
81 spin_lock(&dir
->i_lock
);
82 list_add(&ctx
->list
, &nfsi
->open_files
);
83 spin_unlock(&dir
->i_lock
);
86 return ERR_PTR(-ENOMEM
);
89 static void put_nfs_open_dir_context(struct inode
*dir
, struct nfs_open_dir_context
*ctx
)
91 spin_lock(&dir
->i_lock
);
93 spin_unlock(&dir
->i_lock
);
102 nfs_opendir(struct inode
*inode
, struct file
*filp
)
105 struct nfs_open_dir_context
*ctx
;
107 dfprintk(FILE, "NFS: open dir(%pD2)\n", filp
);
109 nfs_inc_stats(inode
, NFSIOS_VFSOPEN
);
111 ctx
= alloc_nfs_open_dir_context(inode
, current_cred());
116 filp
->private_data
= ctx
;
122 nfs_closedir(struct inode
*inode
, struct file
*filp
)
124 put_nfs_open_dir_context(file_inode(filp
), filp
->private_data
);
128 struct nfs_cache_array_entry
{
132 unsigned char d_type
;
135 struct nfs_cache_array
{
139 struct nfs_cache_array_entry array
[0];
142 typedef int (*decode_dirent_t
)(struct xdr_stream
*, struct nfs_entry
*, bool);
146 struct dir_context
*ctx
;
147 unsigned long page_index
;
150 loff_t current_index
;
151 decode_dirent_t decode
;
153 unsigned long timestamp
;
154 unsigned long gencount
;
155 unsigned int cache_entry_index
;
158 } nfs_readdir_descriptor_t
;
161 * we are freeing strings created by nfs_add_to_readdir_array()
164 void nfs_readdir_clear_array(struct page
*page
)
166 struct nfs_cache_array
*array
;
169 array
= kmap_atomic(page
);
170 for (i
= 0; i
< array
->size
; i
++)
171 kfree(array
->array
[i
].string
.name
);
172 kunmap_atomic(array
);
176 * the caller is responsible for freeing qstr.name
177 * when called by nfs_readdir_add_to_array, the strings will be freed in
178 * nfs_clear_readdir_array()
181 int nfs_readdir_make_qstr(struct qstr
*string
, const char *name
, unsigned int len
)
184 string
->name
= kmemdup(name
, len
, GFP_KERNEL
);
185 if (string
->name
== NULL
)
188 * Avoid a kmemleak false positive. The pointer to the name is stored
189 * in a page cache page which kmemleak does not scan.
191 kmemleak_not_leak(string
->name
);
192 string
->hash
= full_name_hash(NULL
, name
, len
);
197 int nfs_readdir_add_to_array(struct nfs_entry
*entry
, struct page
*page
)
199 struct nfs_cache_array
*array
= kmap(page
);
200 struct nfs_cache_array_entry
*cache_entry
;
203 cache_entry
= &array
->array
[array
->size
];
205 /* Check that this entry lies within the page bounds */
207 if ((char *)&cache_entry
[1] - (char *)page_address(page
) > PAGE_SIZE
)
210 cache_entry
->cookie
= entry
->prev_cookie
;
211 cache_entry
->ino
= entry
->ino
;
212 cache_entry
->d_type
= entry
->d_type
;
213 ret
= nfs_readdir_make_qstr(&cache_entry
->string
, entry
->name
, entry
->len
);
216 array
->last_cookie
= entry
->cookie
;
219 array
->eof_index
= array
->size
;
226 int nfs_readdir_search_for_pos(struct nfs_cache_array
*array
, nfs_readdir_descriptor_t
*desc
)
228 loff_t diff
= desc
->ctx
->pos
- desc
->current_index
;
233 if (diff
>= array
->size
) {
234 if (array
->eof_index
>= 0)
239 index
= (unsigned int)diff
;
240 *desc
->dir_cookie
= array
->array
[index
].cookie
;
241 desc
->cache_entry_index
= index
;
249 nfs_readdir_inode_mapping_valid(struct nfs_inode
*nfsi
)
251 if (nfsi
->cache_validity
& (NFS_INO_INVALID_ATTR
|NFS_INO_INVALID_DATA
))
254 return !test_bit(NFS_INO_INVALIDATING
, &nfsi
->flags
);
258 int nfs_readdir_search_for_cookie(struct nfs_cache_array
*array
, nfs_readdir_descriptor_t
*desc
)
262 int status
= -EAGAIN
;
264 for (i
= 0; i
< array
->size
; i
++) {
265 if (array
->array
[i
].cookie
== *desc
->dir_cookie
) {
266 struct nfs_inode
*nfsi
= NFS_I(file_inode(desc
->file
));
267 struct nfs_open_dir_context
*ctx
= desc
->file
->private_data
;
269 new_pos
= desc
->current_index
+ i
;
270 if (ctx
->attr_gencount
!= nfsi
->attr_gencount
||
271 !nfs_readdir_inode_mapping_valid(nfsi
)) {
273 ctx
->attr_gencount
= nfsi
->attr_gencount
;
274 } else if (new_pos
< desc
->ctx
->pos
) {
276 && ctx
->dup_cookie
== *desc
->dir_cookie
) {
277 if (printk_ratelimit()) {
278 pr_notice("NFS: directory %pD2 contains a readdir loop."
279 "Please contact your server vendor. "
280 "The file: %.*s has duplicate cookie %llu\n",
281 desc
->file
, array
->array
[i
].string
.len
,
282 array
->array
[i
].string
.name
, *desc
->dir_cookie
);
287 ctx
->dup_cookie
= *desc
->dir_cookie
;
290 desc
->ctx
->pos
= new_pos
;
291 desc
->cache_entry_index
= i
;
295 if (array
->eof_index
>= 0) {
296 status
= -EBADCOOKIE
;
297 if (*desc
->dir_cookie
== array
->last_cookie
)
305 int nfs_readdir_search_array(nfs_readdir_descriptor_t
*desc
)
307 struct nfs_cache_array
*array
;
310 array
= kmap(desc
->page
);
312 if (*desc
->dir_cookie
== 0)
313 status
= nfs_readdir_search_for_pos(array
, desc
);
315 status
= nfs_readdir_search_for_cookie(array
, desc
);
317 if (status
== -EAGAIN
) {
318 desc
->last_cookie
= array
->last_cookie
;
319 desc
->current_index
+= array
->size
;
326 /* Fill a page with xdr information before transferring to the cache page */
328 int nfs_readdir_xdr_filler(struct page
**pages
, nfs_readdir_descriptor_t
*desc
,
329 struct nfs_entry
*entry
, struct file
*file
, struct inode
*inode
)
331 struct nfs_open_dir_context
*ctx
= file
->private_data
;
332 const struct cred
*cred
= ctx
->cred
;
333 unsigned long timestamp
, gencount
;
338 gencount
= nfs_inc_attr_generation_counter();
339 error
= NFS_PROTO(inode
)->readdir(file_dentry(file
), cred
, entry
->cookie
, pages
,
340 NFS_SERVER(inode
)->dtsize
, desc
->plus
);
342 /* We requested READDIRPLUS, but the server doesn't grok it */
343 if (error
== -ENOTSUPP
&& desc
->plus
) {
344 NFS_SERVER(inode
)->caps
&= ~NFS_CAP_READDIRPLUS
;
345 clear_bit(NFS_INO_ADVISE_RDPLUS
, &NFS_I(inode
)->flags
);
351 desc
->timestamp
= timestamp
;
352 desc
->gencount
= gencount
;
357 static int xdr_decode(nfs_readdir_descriptor_t
*desc
,
358 struct nfs_entry
*entry
, struct xdr_stream
*xdr
)
362 error
= desc
->decode(xdr
, entry
, desc
->plus
);
365 entry
->fattr
->time_start
= desc
->timestamp
;
366 entry
->fattr
->gencount
= desc
->gencount
;
370 /* Match file and dirent using either filehandle or fileid
371 * Note: caller is responsible for checking the fsid
374 int nfs_same_file(struct dentry
*dentry
, struct nfs_entry
*entry
)
377 struct nfs_inode
*nfsi
;
379 if (d_really_is_negative(dentry
))
382 inode
= d_inode(dentry
);
383 if (is_bad_inode(inode
) || NFS_STALE(inode
))
387 if (entry
->fattr
->fileid
!= nfsi
->fileid
)
389 if (entry
->fh
->size
&& nfs_compare_fh(entry
->fh
, &nfsi
->fh
) != 0)
395 bool nfs_use_readdirplus(struct inode
*dir
, struct dir_context
*ctx
)
397 if (!nfs_server_capable(dir
, NFS_CAP_READDIRPLUS
))
399 if (test_and_clear_bit(NFS_INO_ADVISE_RDPLUS
, &NFS_I(dir
)->flags
))
407 * This function is called by the lookup and getattr code to request the
408 * use of readdirplus to accelerate any future lookups in the same
411 void nfs_advise_use_readdirplus(struct inode
*dir
)
413 struct nfs_inode
*nfsi
= NFS_I(dir
);
415 if (nfs_server_capable(dir
, NFS_CAP_READDIRPLUS
) &&
416 !list_empty(&nfsi
->open_files
))
417 set_bit(NFS_INO_ADVISE_RDPLUS
, &nfsi
->flags
);
421 * This function is mainly for use by nfs_getattr().
423 * If this is an 'ls -l', we want to force use of readdirplus.
424 * Do this by checking if there is an active file descriptor
425 * and calling nfs_advise_use_readdirplus, then forcing a
428 void nfs_force_use_readdirplus(struct inode
*dir
)
430 struct nfs_inode
*nfsi
= NFS_I(dir
);
432 if (nfs_server_capable(dir
, NFS_CAP_READDIRPLUS
) &&
433 !list_empty(&nfsi
->open_files
)) {
434 set_bit(NFS_INO_ADVISE_RDPLUS
, &nfsi
->flags
);
435 invalidate_mapping_pages(dir
->i_mapping
, 0, -1);
440 void nfs_prime_dcache(struct dentry
*parent
, struct nfs_entry
*entry
)
442 struct qstr filename
= QSTR_INIT(entry
->name
, entry
->len
);
443 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq
);
444 struct dentry
*dentry
;
445 struct dentry
*alias
;
446 struct inode
*dir
= d_inode(parent
);
450 if (!(entry
->fattr
->valid
& NFS_ATTR_FATTR_FILEID
))
452 if (!(entry
->fattr
->valid
& NFS_ATTR_FATTR_FSID
))
454 if (filename
.len
== 0)
456 /* Validate that the name doesn't contain any illegal '\0' */
457 if (strnlen(filename
.name
, filename
.len
) != filename
.len
)
460 if (strnchr(filename
.name
, filename
.len
, '/'))
462 if (filename
.name
[0] == '.') {
463 if (filename
.len
== 1)
465 if (filename
.len
== 2 && filename
.name
[1] == '.')
468 filename
.hash
= full_name_hash(parent
, filename
.name
, filename
.len
);
470 dentry
= d_lookup(parent
, &filename
);
473 dentry
= d_alloc_parallel(parent
, &filename
, &wq
);
477 if (!d_in_lookup(dentry
)) {
478 /* Is there a mountpoint here? If so, just exit */
479 if (!nfs_fsid_equal(&NFS_SB(dentry
->d_sb
)->fsid
,
480 &entry
->fattr
->fsid
))
482 if (nfs_same_file(dentry
, entry
)) {
483 if (!entry
->fh
->size
)
485 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
486 status
= nfs_refresh_inode(d_inode(dentry
), entry
->fattr
);
488 nfs_setsecurity(d_inode(dentry
), entry
->fattr
, entry
->label
);
491 d_invalidate(dentry
);
497 if (!entry
->fh
->size
) {
498 d_lookup_done(dentry
);
502 inode
= nfs_fhget(dentry
->d_sb
, entry
->fh
, entry
->fattr
, entry
->label
);
503 alias
= d_splice_alias(inode
, dentry
);
504 d_lookup_done(dentry
);
511 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
516 /* Perform conversion from xdr to cache array */
518 int nfs_readdir_page_filler(nfs_readdir_descriptor_t
*desc
, struct nfs_entry
*entry
,
519 struct page
**xdr_pages
, struct page
*page
, unsigned int buflen
)
521 struct xdr_stream stream
;
523 struct page
*scratch
;
524 struct nfs_cache_array
*array
;
525 unsigned int count
= 0;
528 scratch
= alloc_page(GFP_KERNEL
);
535 xdr_init_decode_pages(&stream
, &buf
, xdr_pages
, buflen
);
536 xdr_set_scratch_buffer(&stream
, page_address(scratch
), PAGE_SIZE
);
539 status
= xdr_decode(desc
, entry
, &stream
);
541 if (status
== -EAGAIN
)
549 nfs_prime_dcache(file_dentry(desc
->file
), entry
);
551 status
= nfs_readdir_add_to_array(entry
, page
);
554 } while (!entry
->eof
);
557 if (count
== 0 || (status
== -EBADCOOKIE
&& entry
->eof
!= 0)) {
559 array
->eof_index
= array
->size
;
569 void nfs_readdir_free_pages(struct page
**pages
, unsigned int npages
)
572 for (i
= 0; i
< npages
; i
++)
577 * nfs_readdir_large_page will allocate pages that must be freed with a call
578 * to nfs_readdir_free_pagearray
581 int nfs_readdir_alloc_pages(struct page
**pages
, unsigned int npages
)
585 for (i
= 0; i
< npages
; i
++) {
586 struct page
*page
= alloc_page(GFP_KERNEL
);
594 nfs_readdir_free_pages(pages
, i
);
599 int nfs_readdir_xdr_to_array(nfs_readdir_descriptor_t
*desc
, struct page
*page
, struct inode
*inode
)
601 struct page
*pages
[NFS_MAX_READDIR_PAGES
];
602 struct nfs_entry entry
;
603 struct file
*file
= desc
->file
;
604 struct nfs_cache_array
*array
;
605 int status
= -ENOMEM
;
606 unsigned int array_size
= ARRAY_SIZE(pages
);
608 entry
.prev_cookie
= 0;
609 entry
.cookie
= desc
->last_cookie
;
611 entry
.fh
= nfs_alloc_fhandle();
612 entry
.fattr
= nfs_alloc_fattr();
613 entry
.server
= NFS_SERVER(inode
);
614 if (entry
.fh
== NULL
|| entry
.fattr
== NULL
)
617 entry
.label
= nfs4_label_alloc(NFS_SERVER(inode
), GFP_NOWAIT
);
618 if (IS_ERR(entry
.label
)) {
619 status
= PTR_ERR(entry
.label
);
624 memset(array
, 0, sizeof(struct nfs_cache_array
));
625 array
->eof_index
= -1;
627 status
= nfs_readdir_alloc_pages(pages
, array_size
);
629 goto out_release_array
;
632 status
= nfs_readdir_xdr_filler(pages
, desc
, &entry
, file
, inode
);
637 status
= nfs_readdir_page_filler(desc
, &entry
, pages
, page
, pglen
);
639 if (status
== -ENOSPC
)
643 } while (array
->eof_index
< 0);
645 nfs_readdir_free_pages(pages
, array_size
);
648 nfs4_label_free(entry
.label
);
650 nfs_free_fattr(entry
.fattr
);
651 nfs_free_fhandle(entry
.fh
);
656 * Now we cache directories properly, by converting xdr information
657 * to an array that can be used for lookups later. This results in
658 * fewer cache pages, since we can store more information on each page.
659 * We only need to convert from xdr once so future lookups are much simpler
662 int nfs_readdir_filler(nfs_readdir_descriptor_t
*desc
, struct page
* page
)
664 struct inode
*inode
= file_inode(desc
->file
);
667 ret
= nfs_readdir_xdr_to_array(desc
, page
, inode
);
670 SetPageUptodate(page
);
672 if (invalidate_inode_pages2_range(inode
->i_mapping
, page
->index
+ 1, -1) < 0) {
673 /* Should never happen */
674 nfs_zap_mapping(inode
, inode
->i_mapping
);
684 void cache_page_release(nfs_readdir_descriptor_t
*desc
)
686 if (!desc
->page
->mapping
)
687 nfs_readdir_clear_array(desc
->page
);
688 put_page(desc
->page
);
693 struct page
*get_cache_page(nfs_readdir_descriptor_t
*desc
)
695 return read_cache_page(desc
->file
->f_mapping
,
696 desc
->page_index
, (filler_t
*)nfs_readdir_filler
, desc
);
700 * Returns 0 if desc->dir_cookie was found on page desc->page_index
703 int find_cache_page(nfs_readdir_descriptor_t
*desc
)
707 desc
->page
= get_cache_page(desc
);
708 if (IS_ERR(desc
->page
))
709 return PTR_ERR(desc
->page
);
711 res
= nfs_readdir_search_array(desc
);
713 cache_page_release(desc
);
717 /* Search for desc->dir_cookie from the beginning of the page cache */
719 int readdir_search_pagecache(nfs_readdir_descriptor_t
*desc
)
723 if (desc
->page_index
== 0) {
724 desc
->current_index
= 0;
725 desc
->last_cookie
= 0;
728 res
= find_cache_page(desc
);
729 } while (res
== -EAGAIN
);
734 * Once we've found the start of the dirent within a page: fill 'er up...
737 int nfs_do_filldir(nfs_readdir_descriptor_t
*desc
)
739 struct file
*file
= desc
->file
;
742 struct nfs_cache_array
*array
= NULL
;
743 struct nfs_open_dir_context
*ctx
= file
->private_data
;
745 array
= kmap(desc
->page
);
746 for (i
= desc
->cache_entry_index
; i
< array
->size
; i
++) {
747 struct nfs_cache_array_entry
*ent
;
749 ent
= &array
->array
[i
];
750 if (!dir_emit(desc
->ctx
, ent
->string
.name
, ent
->string
.len
,
751 nfs_compat_user_ino64(ent
->ino
), ent
->d_type
)) {
756 if (i
< (array
->size
-1))
757 *desc
->dir_cookie
= array
->array
[i
+1].cookie
;
759 *desc
->dir_cookie
= array
->last_cookie
;
763 if (array
->eof_index
>= 0)
767 cache_page_release(desc
);
768 dfprintk(DIRCACHE
, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n",
769 (unsigned long long)*desc
->dir_cookie
, res
);
774 * If we cannot find a cookie in our cache, we suspect that this is
775 * because it points to a deleted file, so we ask the server to return
776 * whatever it thinks is the next entry. We then feed this to filldir.
777 * If all goes well, we should then be able to find our way round the
778 * cache on the next call to readdir_search_pagecache();
780 * NOTE: we cannot add the anonymous page to the pagecache because
781 * the data it contains might not be page aligned. Besides,
782 * we should already have a complete representation of the
783 * directory in the page cache by the time we get here.
786 int uncached_readdir(nfs_readdir_descriptor_t
*desc
)
788 struct page
*page
= NULL
;
790 struct inode
*inode
= file_inode(desc
->file
);
791 struct nfs_open_dir_context
*ctx
= desc
->file
->private_data
;
793 dfprintk(DIRCACHE
, "NFS: uncached_readdir() searching for cookie %Lu\n",
794 (unsigned long long)*desc
->dir_cookie
);
796 page
= alloc_page(GFP_HIGHUSER
);
802 desc
->page_index
= 0;
803 desc
->last_cookie
= *desc
->dir_cookie
;
807 status
= nfs_readdir_xdr_to_array(desc
, page
, inode
);
811 status
= nfs_do_filldir(desc
);
814 dfprintk(DIRCACHE
, "NFS: %s: returns %d\n",
818 cache_page_release(desc
);
822 /* The file offset position represents the dirent entry number. A
823 last cookie cache takes care of the common case of reading the
826 static int nfs_readdir(struct file
*file
, struct dir_context
*ctx
)
828 struct dentry
*dentry
= file_dentry(file
);
829 struct inode
*inode
= d_inode(dentry
);
830 nfs_readdir_descriptor_t my_desc
,
832 struct nfs_open_dir_context
*dir_ctx
= file
->private_data
;
835 dfprintk(FILE, "NFS: readdir(%pD2) starting at cookie %llu\n",
836 file
, (long long)ctx
->pos
);
837 nfs_inc_stats(inode
, NFSIOS_VFSGETDENTS
);
840 * ctx->pos points to the dirent entry number.
841 * *desc->dir_cookie has the cookie for the next entry. We have
842 * to either find the entry with the appropriate number or
843 * revalidate the cookie.
845 memset(desc
, 0, sizeof(*desc
));
849 desc
->dir_cookie
= &dir_ctx
->dir_cookie
;
850 desc
->decode
= NFS_PROTO(inode
)->decode_dirent
;
851 desc
->plus
= nfs_use_readdirplus(inode
, ctx
);
853 if (ctx
->pos
== 0 || nfs_attribute_cache_expired(inode
))
854 res
= nfs_revalidate_mapping(inode
, file
->f_mapping
);
859 res
= readdir_search_pagecache(desc
);
861 if (res
== -EBADCOOKIE
) {
863 /* This means either end of directory */
864 if (*desc
->dir_cookie
&& !desc
->eof
) {
865 /* Or that the server has 'lost' a cookie */
866 res
= uncached_readdir(desc
);
872 if (res
== -ETOOSMALL
&& desc
->plus
) {
873 clear_bit(NFS_INO_ADVISE_RDPLUS
, &NFS_I(inode
)->flags
);
874 nfs_zap_caches(inode
);
875 desc
->page_index
= 0;
883 res
= nfs_do_filldir(desc
);
886 } while (!desc
->eof
);
890 dfprintk(FILE, "NFS: readdir(%pD2) returns %d\n", file
, res
);
894 static loff_t
nfs_llseek_dir(struct file
*filp
, loff_t offset
, int whence
)
896 struct inode
*inode
= file_inode(filp
);
897 struct nfs_open_dir_context
*dir_ctx
= filp
->private_data
;
899 dfprintk(FILE, "NFS: llseek dir(%pD2, %lld, %d)\n",
900 filp
, offset
, whence
);
914 offset
+= filp
->f_pos
;
920 if (offset
!= filp
->f_pos
) {
921 filp
->f_pos
= offset
;
922 dir_ctx
->dir_cookie
= 0;
930 * All directory operations under NFS are synchronous, so fsync()
931 * is a dummy operation.
933 static int nfs_fsync_dir(struct file
*filp
, loff_t start
, loff_t end
,
936 struct inode
*inode
= file_inode(filp
);
938 dfprintk(FILE, "NFS: fsync dir(%pD2) datasync %d\n", filp
, datasync
);
941 nfs_inc_stats(inode
, NFSIOS_VFSFSYNC
);
947 * nfs_force_lookup_revalidate - Mark the directory as having changed
948 * @dir: pointer to directory inode
950 * This forces the revalidation code in nfs_lookup_revalidate() to do a
951 * full lookup on all child dentries of 'dir' whenever a change occurs
952 * on the server that might have invalidated our dcache.
954 * The caller should be holding dir->i_lock
956 void nfs_force_lookup_revalidate(struct inode
*dir
)
958 NFS_I(dir
)->cache_change_attribute
++;
960 EXPORT_SYMBOL_GPL(nfs_force_lookup_revalidate
);
963 * A check for whether or not the parent directory has changed.
964 * In the case it has, we assume that the dentries are untrustworthy
965 * and may need to be looked up again.
966 * If rcu_walk prevents us from performing a full check, return 0.
968 static int nfs_check_verifier(struct inode
*dir
, struct dentry
*dentry
,
973 if (NFS_SERVER(dir
)->flags
& NFS_MOUNT_LOOKUP_CACHE_NONE
)
975 if (!nfs_verify_change_attribute(dir
, dentry
->d_time
))
977 /* Revalidate nfsi->cache_change_attribute before we declare a match */
978 if (nfs_mapping_need_revalidate_inode(dir
)) {
981 if (__nfs_revalidate_inode(NFS_SERVER(dir
), dir
) < 0)
984 if (!nfs_verify_change_attribute(dir
, dentry
->d_time
))
990 * Use intent information to check whether or not we're going to do
991 * an O_EXCL create using this path component.
993 static int nfs_is_exclusive_create(struct inode
*dir
, unsigned int flags
)
995 if (NFS_PROTO(dir
)->version
== 2)
997 return flags
& LOOKUP_EXCL
;
1001 * Inode and filehandle revalidation for lookups.
1003 * We force revalidation in the cases where the VFS sets LOOKUP_REVAL,
1004 * or if the intent information indicates that we're about to open this
1005 * particular file and the "nocto" mount flag is not set.
1009 int nfs_lookup_verify_inode(struct inode
*inode
, unsigned int flags
)
1011 struct nfs_server
*server
= NFS_SERVER(inode
);
1014 if (IS_AUTOMOUNT(inode
))
1017 if (flags
& LOOKUP_OPEN
) {
1018 switch (inode
->i_mode
& S_IFMT
) {
1020 /* A NFSv4 OPEN will revalidate later */
1021 if (server
->caps
& NFS_CAP_ATOMIC_OPEN
)
1025 if (server
->flags
& NFS_MOUNT_NOCTO
)
1027 /* NFS close-to-open cache consistency validation */
1032 /* VFS wants an on-the-wire revalidation */
1033 if (flags
& LOOKUP_REVAL
)
1036 return (inode
->i_nlink
== 0) ? -ESTALE
: 0;
1038 if (flags
& LOOKUP_RCU
)
1040 ret
= __nfs_revalidate_inode(server
, inode
);
1047 * We judge how long we want to trust negative
1048 * dentries by looking at the parent inode mtime.
1050 * If parent mtime has changed, we revalidate, else we wait for a
1051 * period corresponding to the parent's attribute cache timeout value.
1053 * If LOOKUP_RCU prevents us from performing a full check, return 1
1054 * suggesting a reval is needed.
1056 * Note that when creating a new file, or looking up a rename target,
1057 * then it shouldn't be necessary to revalidate a negative dentry.
1060 int nfs_neg_need_reval(struct inode
*dir
, struct dentry
*dentry
,
1063 if (flags
& (LOOKUP_CREATE
| LOOKUP_RENAME_TARGET
))
1065 if (NFS_SERVER(dir
)->flags
& NFS_MOUNT_LOOKUP_CACHE_NONEG
)
1067 return !nfs_check_verifier(dir
, dentry
, flags
& LOOKUP_RCU
);
1071 nfs_lookup_revalidate_done(struct inode
*dir
, struct dentry
*dentry
,
1072 struct inode
*inode
, int error
)
1076 dfprintk(LOOKUPCACHE
, "NFS: %s(%pd2) is valid\n",
1080 nfs_mark_for_revalidate(dir
);
1081 if (inode
&& S_ISDIR(inode
->i_mode
)) {
1082 /* Purge readdir caches. */
1083 nfs_zap_caches(inode
);
1085 * We can't d_drop the root of a disconnected tree:
1086 * its d_hash is on the s_anon list and d_drop() would hide
1087 * it from shrink_dcache_for_unmount(), leading to busy
1088 * inodes on unmount and further oopses.
1090 if (IS_ROOT(dentry
))
1093 dfprintk(LOOKUPCACHE
, "NFS: %s(%pd2) is invalid\n",
1097 dfprintk(LOOKUPCACHE
, "NFS: %s(%pd2) lookup returned error %d\n",
1098 __func__
, dentry
, error
);
1103 nfs_lookup_revalidate_negative(struct inode
*dir
, struct dentry
*dentry
,
1107 if (nfs_neg_need_reval(dir
, dentry
, flags
)) {
1108 if (flags
& LOOKUP_RCU
)
1112 return nfs_lookup_revalidate_done(dir
, dentry
, NULL
, ret
);
1116 nfs_lookup_revalidate_delegated(struct inode
*dir
, struct dentry
*dentry
,
1117 struct inode
*inode
)
1119 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1120 return nfs_lookup_revalidate_done(dir
, dentry
, inode
, 1);
1124 nfs_lookup_revalidate_dentry(struct inode
*dir
, struct dentry
*dentry
,
1125 struct inode
*inode
)
1127 struct nfs_fh
*fhandle
;
1128 struct nfs_fattr
*fattr
;
1129 struct nfs4_label
*label
;
1133 fhandle
= nfs_alloc_fhandle();
1134 fattr
= nfs_alloc_fattr();
1135 label
= nfs4_label_alloc(NFS_SERVER(inode
), GFP_KERNEL
);
1136 if (fhandle
== NULL
|| fattr
== NULL
|| IS_ERR(label
))
1139 ret
= NFS_PROTO(dir
)->lookup(dir
, &dentry
->d_name
, fhandle
, fattr
, label
);
1141 if (ret
== -ESTALE
|| ret
== -ENOENT
)
1146 if (nfs_compare_fh(NFS_FH(inode
), fhandle
))
1148 if (nfs_refresh_inode(inode
, fattr
) < 0)
1151 nfs_setsecurity(inode
, fattr
, label
);
1152 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1154 /* set a readdirplus hint that we had a cache miss */
1155 nfs_force_use_readdirplus(dir
);
1158 nfs_free_fattr(fattr
);
1159 nfs_free_fhandle(fhandle
);
1160 nfs4_label_free(label
);
1161 return nfs_lookup_revalidate_done(dir
, dentry
, inode
, ret
);
1165 * This is called every time the dcache has a lookup hit,
1166 * and we should check whether we can really trust that
1169 * NOTE! The hit can be a negative hit too, don't assume
1172 * If the parent directory is seen to have changed, we throw out the
1173 * cached dentry and do a new lookup.
1176 nfs_do_lookup_revalidate(struct inode
*dir
, struct dentry
*dentry
,
1179 struct inode
*inode
;
1182 nfs_inc_stats(dir
, NFSIOS_DENTRYREVALIDATE
);
1183 inode
= d_inode(dentry
);
1186 return nfs_lookup_revalidate_negative(dir
, dentry
, flags
);
1188 if (is_bad_inode(inode
)) {
1189 dfprintk(LOOKUPCACHE
, "%s: %pd2 has dud inode\n",
1194 if (NFS_PROTO(dir
)->have_delegation(inode
, FMODE_READ
))
1195 return nfs_lookup_revalidate_delegated(dir
, dentry
, inode
);
1197 /* Force a full look up iff the parent directory has changed */
1198 if (!(flags
& (LOOKUP_EXCL
| LOOKUP_REVAL
)) &&
1199 nfs_check_verifier(dir
, dentry
, flags
& LOOKUP_RCU
)) {
1200 error
= nfs_lookup_verify_inode(inode
, flags
);
1202 if (error
== -ESTALE
)
1203 nfs_zap_caches(dir
);
1206 nfs_advise_use_readdirplus(dir
);
1210 if (flags
& LOOKUP_RCU
)
1213 if (NFS_STALE(inode
))
1216 trace_nfs_lookup_revalidate_enter(dir
, dentry
, flags
);
1217 error
= nfs_lookup_revalidate_dentry(dir
, dentry
, inode
);
1218 trace_nfs_lookup_revalidate_exit(dir
, dentry
, flags
, error
);
1221 return nfs_lookup_revalidate_done(dir
, dentry
, inode
, 1);
1223 if (flags
& LOOKUP_RCU
)
1225 return nfs_lookup_revalidate_done(dir
, dentry
, inode
, 0);
1229 __nfs_lookup_revalidate(struct dentry
*dentry
, unsigned int flags
,
1230 int (*reval
)(struct inode
*, struct dentry
*, unsigned int))
1232 struct dentry
*parent
;
1236 if (flags
& LOOKUP_RCU
) {
1237 parent
= READ_ONCE(dentry
->d_parent
);
1238 dir
= d_inode_rcu(parent
);
1241 ret
= reval(dir
, dentry
, flags
);
1242 if (parent
!= READ_ONCE(dentry
->d_parent
))
1245 parent
= dget_parent(dentry
);
1246 ret
= reval(d_inode(parent
), dentry
, flags
);
1252 static int nfs_lookup_revalidate(struct dentry
*dentry
, unsigned int flags
)
1254 return __nfs_lookup_revalidate(dentry
, flags
, nfs_do_lookup_revalidate
);
1258 * A weaker form of d_revalidate for revalidating just the d_inode(dentry)
1259 * when we don't really care about the dentry name. This is called when a
1260 * pathwalk ends on a dentry that was not found via a normal lookup in the
1261 * parent dir (e.g.: ".", "..", procfs symlinks or mountpoint traversals).
1263 * In this situation, we just want to verify that the inode itself is OK
1264 * since the dentry might have changed on the server.
1266 static int nfs_weak_revalidate(struct dentry
*dentry
, unsigned int flags
)
1268 struct inode
*inode
= d_inode(dentry
);
1272 * I believe we can only get a negative dentry here in the case of a
1273 * procfs-style symlink. Just assume it's correct for now, but we may
1274 * eventually need to do something more here.
1277 dfprintk(LOOKUPCACHE
, "%s: %pd2 has negative inode\n",
1282 if (is_bad_inode(inode
)) {
1283 dfprintk(LOOKUPCACHE
, "%s: %pd2 has dud inode\n",
1288 error
= nfs_lookup_verify_inode(inode
, flags
);
1289 dfprintk(LOOKUPCACHE
, "NFS: %s: inode %lu is %s\n",
1290 __func__
, inode
->i_ino
, error
? "invalid" : "valid");
1295 * This is called from dput() when d_count is going to 0.
1297 static int nfs_dentry_delete(const struct dentry
*dentry
)
1299 dfprintk(VFS
, "NFS: dentry_delete(%pd2, %x)\n",
1300 dentry
, dentry
->d_flags
);
1302 /* Unhash any dentry with a stale inode */
1303 if (d_really_is_positive(dentry
) && NFS_STALE(d_inode(dentry
)))
1306 if (dentry
->d_flags
& DCACHE_NFSFS_RENAMED
) {
1307 /* Unhash it, so that ->d_iput() would be called */
1310 if (!(dentry
->d_sb
->s_flags
& SB_ACTIVE
)) {
1311 /* Unhash it, so that ancestors of killed async unlink
1312 * files will be cleaned up during umount */
1319 /* Ensure that we revalidate inode->i_nlink */
1320 static void nfs_drop_nlink(struct inode
*inode
)
1322 spin_lock(&inode
->i_lock
);
1323 /* drop the inode if we're reasonably sure this is the last link */
1324 if (inode
->i_nlink
> 0)
1326 NFS_I(inode
)->attr_gencount
= nfs_inc_attr_generation_counter();
1327 NFS_I(inode
)->cache_validity
|= NFS_INO_INVALID_CHANGE
1328 | NFS_INO_INVALID_CTIME
1329 | NFS_INO_INVALID_OTHER
1330 | NFS_INO_REVAL_FORCED
;
1331 spin_unlock(&inode
->i_lock
);
1335 * Called when the dentry loses inode.
1336 * We use it to clean up silly-renamed files.
1338 static void nfs_dentry_iput(struct dentry
*dentry
, struct inode
*inode
)
1340 if (S_ISDIR(inode
->i_mode
))
1341 /* drop any readdir cache as it could easily be old */
1342 NFS_I(inode
)->cache_validity
|= NFS_INO_INVALID_DATA
;
1344 if (dentry
->d_flags
& DCACHE_NFSFS_RENAMED
) {
1345 nfs_complete_unlink(dentry
, inode
);
1346 nfs_drop_nlink(inode
);
1351 static void nfs_d_release(struct dentry
*dentry
)
1353 /* free cached devname value, if it survived that far */
1354 if (unlikely(dentry
->d_fsdata
)) {
1355 if (dentry
->d_flags
& DCACHE_NFSFS_RENAMED
)
1358 kfree(dentry
->d_fsdata
);
1362 const struct dentry_operations nfs_dentry_operations
= {
1363 .d_revalidate
= nfs_lookup_revalidate
,
1364 .d_weak_revalidate
= nfs_weak_revalidate
,
1365 .d_delete
= nfs_dentry_delete
,
1366 .d_iput
= nfs_dentry_iput
,
1367 .d_automount
= nfs_d_automount
,
1368 .d_release
= nfs_d_release
,
1370 EXPORT_SYMBOL_GPL(nfs_dentry_operations
);
1372 struct dentry
*nfs_lookup(struct inode
*dir
, struct dentry
* dentry
, unsigned int flags
)
1375 struct inode
*inode
= NULL
;
1376 struct nfs_fh
*fhandle
= NULL
;
1377 struct nfs_fattr
*fattr
= NULL
;
1378 struct nfs4_label
*label
= NULL
;
1381 dfprintk(VFS
, "NFS: lookup(%pd2)\n", dentry
);
1382 nfs_inc_stats(dir
, NFSIOS_VFSLOOKUP
);
1384 if (unlikely(dentry
->d_name
.len
> NFS_SERVER(dir
)->namelen
))
1385 return ERR_PTR(-ENAMETOOLONG
);
1388 * If we're doing an exclusive create, optimize away the lookup
1389 * but don't hash the dentry.
1391 if (nfs_is_exclusive_create(dir
, flags
) || flags
& LOOKUP_RENAME_TARGET
)
1394 res
= ERR_PTR(-ENOMEM
);
1395 fhandle
= nfs_alloc_fhandle();
1396 fattr
= nfs_alloc_fattr();
1397 if (fhandle
== NULL
|| fattr
== NULL
)
1400 label
= nfs4_label_alloc(NFS_SERVER(dir
), GFP_NOWAIT
);
1404 trace_nfs_lookup_enter(dir
, dentry
, flags
);
1405 error
= NFS_PROTO(dir
)->lookup(dir
, &dentry
->d_name
, fhandle
, fattr
, label
);
1406 if (error
== -ENOENT
)
1409 res
= ERR_PTR(error
);
1412 inode
= nfs_fhget(dentry
->d_sb
, fhandle
, fattr
, label
);
1413 res
= ERR_CAST(inode
);
1417 /* Notify readdir to use READDIRPLUS */
1418 nfs_force_use_readdirplus(dir
);
1421 res
= d_splice_alias(inode
, dentry
);
1427 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1429 trace_nfs_lookup_exit(dir
, dentry
, flags
, error
);
1430 nfs4_label_free(label
);
1432 nfs_free_fattr(fattr
);
1433 nfs_free_fhandle(fhandle
);
1436 EXPORT_SYMBOL_GPL(nfs_lookup
);
1438 #if IS_ENABLED(CONFIG_NFS_V4)
1439 static int nfs4_lookup_revalidate(struct dentry
*, unsigned int);
1441 const struct dentry_operations nfs4_dentry_operations
= {
1442 .d_revalidate
= nfs4_lookup_revalidate
,
1443 .d_weak_revalidate
= nfs_weak_revalidate
,
1444 .d_delete
= nfs_dentry_delete
,
1445 .d_iput
= nfs_dentry_iput
,
1446 .d_automount
= nfs_d_automount
,
1447 .d_release
= nfs_d_release
,
1449 EXPORT_SYMBOL_GPL(nfs4_dentry_operations
);
1451 static fmode_t
flags_to_mode(int flags
)
1453 fmode_t res
= (__force fmode_t
)flags
& FMODE_EXEC
;
1454 if ((flags
& O_ACCMODE
) != O_WRONLY
)
1456 if ((flags
& O_ACCMODE
) != O_RDONLY
)
1461 static struct nfs_open_context
*create_nfs_open_context(struct dentry
*dentry
, int open_flags
, struct file
*filp
)
1463 return alloc_nfs_open_context(dentry
, flags_to_mode(open_flags
), filp
);
1466 static int do_open(struct inode
*inode
, struct file
*filp
)
1468 nfs_fscache_open_file(inode
, filp
);
1472 static int nfs_finish_open(struct nfs_open_context
*ctx
,
1473 struct dentry
*dentry
,
1474 struct file
*file
, unsigned open_flags
)
1478 err
= finish_open(file
, dentry
, do_open
);
1481 if (S_ISREG(file
->f_path
.dentry
->d_inode
->i_mode
))
1482 nfs_file_set_open_context(file
, ctx
);
1489 int nfs_atomic_open(struct inode
*dir
, struct dentry
*dentry
,
1490 struct file
*file
, unsigned open_flags
,
1493 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq
);
1494 struct nfs_open_context
*ctx
;
1496 struct iattr attr
= { .ia_valid
= ATTR_OPEN
};
1497 struct inode
*inode
;
1498 unsigned int lookup_flags
= 0;
1499 bool switched
= false;
1503 /* Expect a negative dentry */
1504 BUG_ON(d_inode(dentry
));
1506 dfprintk(VFS
, "NFS: atomic_open(%s/%lu), %pd\n",
1507 dir
->i_sb
->s_id
, dir
->i_ino
, dentry
);
1509 err
= nfs_check_flags(open_flags
);
1513 /* NFS only supports OPEN on regular files */
1514 if ((open_flags
& O_DIRECTORY
)) {
1515 if (!d_in_lookup(dentry
)) {
1517 * Hashed negative dentry with O_DIRECTORY: dentry was
1518 * revalidated and is fine, no need to perform lookup
1523 lookup_flags
= LOOKUP_OPEN
|LOOKUP_DIRECTORY
;
1527 if (dentry
->d_name
.len
> NFS_SERVER(dir
)->namelen
)
1528 return -ENAMETOOLONG
;
1530 if (open_flags
& O_CREAT
) {
1531 struct nfs_server
*server
= NFS_SERVER(dir
);
1533 if (!(server
->attr_bitmask
[2] & FATTR4_WORD2_MODE_UMASK
))
1534 mode
&= ~current_umask();
1536 attr
.ia_valid
|= ATTR_MODE
;
1537 attr
.ia_mode
= mode
;
1539 if (open_flags
& O_TRUNC
) {
1540 attr
.ia_valid
|= ATTR_SIZE
;
1544 if (!(open_flags
& O_CREAT
) && !d_in_lookup(dentry
)) {
1547 dentry
= d_alloc_parallel(dentry
->d_parent
,
1548 &dentry
->d_name
, &wq
);
1550 return PTR_ERR(dentry
);
1551 if (unlikely(!d_in_lookup(dentry
)))
1552 return finish_no_open(file
, dentry
);
1555 ctx
= create_nfs_open_context(dentry
, open_flags
, file
);
1560 trace_nfs_atomic_open_enter(dir
, ctx
, open_flags
);
1561 inode
= NFS_PROTO(dir
)->open_context(dir
, ctx
, open_flags
, &attr
, &created
);
1563 file
->f_mode
|= FMODE_CREATED
;
1564 if (IS_ERR(inode
)) {
1565 err
= PTR_ERR(inode
);
1566 trace_nfs_atomic_open_exit(dir
, ctx
, open_flags
, err
);
1567 put_nfs_open_context(ctx
);
1571 d_splice_alias(NULL
, dentry
);
1572 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1578 if (!(open_flags
& O_NOFOLLOW
))
1588 err
= nfs_finish_open(ctx
, ctx
->dentry
, file
, open_flags
);
1589 trace_nfs_atomic_open_exit(dir
, ctx
, open_flags
, err
);
1590 put_nfs_open_context(ctx
);
1592 if (unlikely(switched
)) {
1593 d_lookup_done(dentry
);
1599 res
= nfs_lookup(dir
, dentry
, lookup_flags
);
1601 d_lookup_done(dentry
);
1608 return PTR_ERR(res
);
1609 return finish_no_open(file
, res
);
1611 EXPORT_SYMBOL_GPL(nfs_atomic_open
);
1614 nfs4_do_lookup_revalidate(struct inode
*dir
, struct dentry
*dentry
,
1617 struct inode
*inode
;
1619 if (!(flags
& LOOKUP_OPEN
) || (flags
& LOOKUP_DIRECTORY
))
1621 if (d_mountpoint(dentry
))
1624 inode
= d_inode(dentry
);
1626 /* We can't create new files in nfs_open_revalidate(), so we
1627 * optimize away revalidation of negative dentries.
1632 if (NFS_PROTO(dir
)->have_delegation(inode
, FMODE_READ
))
1633 return nfs_lookup_revalidate_delegated(dir
, dentry
, inode
);
1635 /* NFS only supports OPEN on regular files */
1636 if (!S_ISREG(inode
->i_mode
))
1639 /* We cannot do exclusive creation on a positive dentry */
1640 if (flags
& (LOOKUP_EXCL
| LOOKUP_REVAL
))
1643 /* Check if the directory changed */
1644 if (!nfs_check_verifier(dir
, dentry
, flags
& LOOKUP_RCU
))
1647 /* Let f_op->open() actually open (and revalidate) the file */
1650 if (flags
& LOOKUP_RCU
)
1652 return nfs_lookup_revalidate_dentry(dir
, dentry
, inode
);;
1655 return nfs_do_lookup_revalidate(dir
, dentry
, flags
);
1658 static int nfs4_lookup_revalidate(struct dentry
*dentry
, unsigned int flags
)
1660 return __nfs_lookup_revalidate(dentry
, flags
,
1661 nfs4_do_lookup_revalidate
);
1664 #endif /* CONFIG_NFSV4 */
1667 * Code common to create, mkdir, and mknod.
1669 int nfs_instantiate(struct dentry
*dentry
, struct nfs_fh
*fhandle
,
1670 struct nfs_fattr
*fattr
,
1671 struct nfs4_label
*label
)
1673 struct dentry
*parent
= dget_parent(dentry
);
1674 struct inode
*dir
= d_inode(parent
);
1675 struct inode
*inode
;
1677 int error
= -EACCES
;
1681 /* We may have been initialized further down */
1682 if (d_really_is_positive(dentry
))
1684 if (fhandle
->size
== 0) {
1685 error
= NFS_PROTO(dir
)->lookup(dir
, &dentry
->d_name
, fhandle
, fattr
, NULL
);
1689 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1690 if (!(fattr
->valid
& NFS_ATTR_FATTR
)) {
1691 struct nfs_server
*server
= NFS_SB(dentry
->d_sb
);
1692 error
= server
->nfs_client
->rpc_ops
->getattr(server
, fhandle
,
1697 inode
= nfs_fhget(dentry
->d_sb
, fhandle
, fattr
, label
);
1698 d
= d_splice_alias(inode
, dentry
);
1708 nfs_mark_for_revalidate(dir
);
1712 EXPORT_SYMBOL_GPL(nfs_instantiate
);
1715 * Following a failed create operation, we drop the dentry rather
1716 * than retain a negative dentry. This avoids a problem in the event
1717 * that the operation succeeded on the server, but an error in the
1718 * reply path made it appear to have failed.
1720 int nfs_create(struct inode
*dir
, struct dentry
*dentry
,
1721 umode_t mode
, bool excl
)
1724 int open_flags
= excl
? O_CREAT
| O_EXCL
: O_CREAT
;
1727 dfprintk(VFS
, "NFS: create(%s/%lu), %pd\n",
1728 dir
->i_sb
->s_id
, dir
->i_ino
, dentry
);
1730 attr
.ia_mode
= mode
;
1731 attr
.ia_valid
= ATTR_MODE
;
1733 trace_nfs_create_enter(dir
, dentry
, open_flags
);
1734 error
= NFS_PROTO(dir
)->create(dir
, dentry
, &attr
, open_flags
);
1735 trace_nfs_create_exit(dir
, dentry
, open_flags
, error
);
1743 EXPORT_SYMBOL_GPL(nfs_create
);
1746 * See comments for nfs_proc_create regarding failed operations.
1749 nfs_mknod(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
, dev_t rdev
)
1754 dfprintk(VFS
, "NFS: mknod(%s/%lu), %pd\n",
1755 dir
->i_sb
->s_id
, dir
->i_ino
, dentry
);
1757 attr
.ia_mode
= mode
;
1758 attr
.ia_valid
= ATTR_MODE
;
1760 trace_nfs_mknod_enter(dir
, dentry
);
1761 status
= NFS_PROTO(dir
)->mknod(dir
, dentry
, &attr
, rdev
);
1762 trace_nfs_mknod_exit(dir
, dentry
, status
);
1770 EXPORT_SYMBOL_GPL(nfs_mknod
);
1773 * See comments for nfs_proc_create regarding failed operations.
1775 int nfs_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
1780 dfprintk(VFS
, "NFS: mkdir(%s/%lu), %pd\n",
1781 dir
->i_sb
->s_id
, dir
->i_ino
, dentry
);
1783 attr
.ia_valid
= ATTR_MODE
;
1784 attr
.ia_mode
= mode
| S_IFDIR
;
1786 trace_nfs_mkdir_enter(dir
, dentry
);
1787 error
= NFS_PROTO(dir
)->mkdir(dir
, dentry
, &attr
);
1788 trace_nfs_mkdir_exit(dir
, dentry
, error
);
1796 EXPORT_SYMBOL_GPL(nfs_mkdir
);
1798 static void nfs_dentry_handle_enoent(struct dentry
*dentry
)
1800 if (simple_positive(dentry
))
1804 int nfs_rmdir(struct inode
*dir
, struct dentry
*dentry
)
1808 dfprintk(VFS
, "NFS: rmdir(%s/%lu), %pd\n",
1809 dir
->i_sb
->s_id
, dir
->i_ino
, dentry
);
1811 trace_nfs_rmdir_enter(dir
, dentry
);
1812 if (d_really_is_positive(dentry
)) {
1813 down_write(&NFS_I(d_inode(dentry
))->rmdir_sem
);
1814 error
= NFS_PROTO(dir
)->rmdir(dir
, &dentry
->d_name
);
1815 /* Ensure the VFS deletes this inode */
1818 clear_nlink(d_inode(dentry
));
1821 nfs_dentry_handle_enoent(dentry
);
1823 up_write(&NFS_I(d_inode(dentry
))->rmdir_sem
);
1825 error
= NFS_PROTO(dir
)->rmdir(dir
, &dentry
->d_name
);
1826 trace_nfs_rmdir_exit(dir
, dentry
, error
);
1830 EXPORT_SYMBOL_GPL(nfs_rmdir
);
1833 * Remove a file after making sure there are no pending writes,
1834 * and after checking that the file has only one user.
1836 * We invalidate the attribute cache and free the inode prior to the operation
1837 * to avoid possible races if the server reuses the inode.
1839 static int nfs_safe_remove(struct dentry
*dentry
)
1841 struct inode
*dir
= d_inode(dentry
->d_parent
);
1842 struct inode
*inode
= d_inode(dentry
);
1845 dfprintk(VFS
, "NFS: safe_remove(%pd2)\n", dentry
);
1847 /* If the dentry was sillyrenamed, we simply call d_delete() */
1848 if (dentry
->d_flags
& DCACHE_NFSFS_RENAMED
) {
1853 trace_nfs_remove_enter(dir
, dentry
);
1854 if (inode
!= NULL
) {
1855 error
= NFS_PROTO(dir
)->remove(dir
, dentry
);
1857 nfs_drop_nlink(inode
);
1859 error
= NFS_PROTO(dir
)->remove(dir
, dentry
);
1860 if (error
== -ENOENT
)
1861 nfs_dentry_handle_enoent(dentry
);
1862 trace_nfs_remove_exit(dir
, dentry
, error
);
1867 /* We do silly rename. In case sillyrename() returns -EBUSY, the inode
1868 * belongs to an active ".nfs..." file and we return -EBUSY.
1870 * If sillyrename() returns 0, we do nothing, otherwise we unlink.
1872 int nfs_unlink(struct inode
*dir
, struct dentry
*dentry
)
1875 int need_rehash
= 0;
1877 dfprintk(VFS
, "NFS: unlink(%s/%lu, %pd)\n", dir
->i_sb
->s_id
,
1878 dir
->i_ino
, dentry
);
1880 trace_nfs_unlink_enter(dir
, dentry
);
1881 spin_lock(&dentry
->d_lock
);
1882 if (d_count(dentry
) > 1) {
1883 spin_unlock(&dentry
->d_lock
);
1884 /* Start asynchronous writeout of the inode */
1885 write_inode_now(d_inode(dentry
), 0);
1886 error
= nfs_sillyrename(dir
, dentry
);
1889 if (!d_unhashed(dentry
)) {
1893 spin_unlock(&dentry
->d_lock
);
1894 error
= nfs_safe_remove(dentry
);
1895 if (!error
|| error
== -ENOENT
) {
1896 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1897 } else if (need_rehash
)
1900 trace_nfs_unlink_exit(dir
, dentry
, error
);
1903 EXPORT_SYMBOL_GPL(nfs_unlink
);
1906 * To create a symbolic link, most file systems instantiate a new inode,
1907 * add a page to it containing the path, then write it out to the disk
1908 * using prepare_write/commit_write.
1910 * Unfortunately the NFS client can't create the in-core inode first
1911 * because it needs a file handle to create an in-core inode (see
1912 * fs/nfs/inode.c:nfs_fhget). We only have a file handle *after* the
1913 * symlink request has completed on the server.
1915 * So instead we allocate a raw page, copy the symname into it, then do
1916 * the SYMLINK request with the page as the buffer. If it succeeds, we
1917 * now have a new file handle and can instantiate an in-core NFS inode
1918 * and move the raw page into its mapping.
1920 int nfs_symlink(struct inode
*dir
, struct dentry
*dentry
, const char *symname
)
1925 unsigned int pathlen
= strlen(symname
);
1928 dfprintk(VFS
, "NFS: symlink(%s/%lu, %pd, %s)\n", dir
->i_sb
->s_id
,
1929 dir
->i_ino
, dentry
, symname
);
1931 if (pathlen
> PAGE_SIZE
)
1932 return -ENAMETOOLONG
;
1934 attr
.ia_mode
= S_IFLNK
| S_IRWXUGO
;
1935 attr
.ia_valid
= ATTR_MODE
;
1937 page
= alloc_page(GFP_USER
);
1941 kaddr
= page_address(page
);
1942 memcpy(kaddr
, symname
, pathlen
);
1943 if (pathlen
< PAGE_SIZE
)
1944 memset(kaddr
+ pathlen
, 0, PAGE_SIZE
- pathlen
);
1946 trace_nfs_symlink_enter(dir
, dentry
);
1947 error
= NFS_PROTO(dir
)->symlink(dir
, dentry
, page
, pathlen
, &attr
);
1948 trace_nfs_symlink_exit(dir
, dentry
, error
);
1950 dfprintk(VFS
, "NFS: symlink(%s/%lu, %pd, %s) error %d\n",
1951 dir
->i_sb
->s_id
, dir
->i_ino
,
1952 dentry
, symname
, error
);
1959 * No big deal if we can't add this page to the page cache here.
1960 * READLINK will get the missing page from the server if needed.
1962 if (!add_to_page_cache_lru(page
, d_inode(dentry
)->i_mapping
, 0,
1964 SetPageUptodate(page
);
1967 * add_to_page_cache_lru() grabs an extra page refcount.
1968 * Drop it here to avoid leaking this page later.
1976 EXPORT_SYMBOL_GPL(nfs_symlink
);
1979 nfs_link(struct dentry
*old_dentry
, struct inode
*dir
, struct dentry
*dentry
)
1981 struct inode
*inode
= d_inode(old_dentry
);
1984 dfprintk(VFS
, "NFS: link(%pd2 -> %pd2)\n",
1985 old_dentry
, dentry
);
1987 trace_nfs_link_enter(inode
, dir
, dentry
);
1989 error
= NFS_PROTO(dir
)->link(inode
, dir
, &dentry
->d_name
);
1992 d_add(dentry
, inode
);
1994 trace_nfs_link_exit(inode
, dir
, dentry
, error
);
1997 EXPORT_SYMBOL_GPL(nfs_link
);
2001 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a
2002 * different file handle for the same inode after a rename (e.g. when
2003 * moving to a different directory). A fail-safe method to do so would
2004 * be to look up old_dir/old_name, create a link to new_dir/new_name and
2005 * rename the old file using the sillyrename stuff. This way, the original
2006 * file in old_dir will go away when the last process iput()s the inode.
2010 * It actually works quite well. One needs to have the possibility for
2011 * at least one ".nfs..." file in each directory the file ever gets
2012 * moved or linked to which happens automagically with the new
2013 * implementation that only depends on the dcache stuff instead of
2014 * using the inode layer
2016 * Unfortunately, things are a little more complicated than indicated
2017 * above. For a cross-directory move, we want to make sure we can get
2018 * rid of the old inode after the operation. This means there must be
2019 * no pending writes (if it's a file), and the use count must be 1.
2020 * If these conditions are met, we can drop the dentries before doing
2023 int nfs_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
2024 struct inode
*new_dir
, struct dentry
*new_dentry
,
2027 struct inode
*old_inode
= d_inode(old_dentry
);
2028 struct inode
*new_inode
= d_inode(new_dentry
);
2029 struct dentry
*dentry
= NULL
, *rehash
= NULL
;
2030 struct rpc_task
*task
;
2036 dfprintk(VFS
, "NFS: rename(%pd2 -> %pd2, ct=%d)\n",
2037 old_dentry
, new_dentry
,
2038 d_count(new_dentry
));
2040 trace_nfs_rename_enter(old_dir
, old_dentry
, new_dir
, new_dentry
);
2042 * For non-directories, check whether the target is busy and if so,
2043 * make a copy of the dentry and then do a silly-rename. If the
2044 * silly-rename succeeds, the copied dentry is hashed and becomes
2047 if (new_inode
&& !S_ISDIR(new_inode
->i_mode
)) {
2049 * To prevent any new references to the target during the
2050 * rename, we unhash the dentry in advance.
2052 if (!d_unhashed(new_dentry
)) {
2054 rehash
= new_dentry
;
2057 if (d_count(new_dentry
) > 2) {
2060 /* copy the target dentry's name */
2061 dentry
= d_alloc(new_dentry
->d_parent
,
2062 &new_dentry
->d_name
);
2066 /* silly-rename the existing target ... */
2067 err
= nfs_sillyrename(new_dir
, new_dentry
);
2071 new_dentry
= dentry
;
2077 task
= nfs_async_rename(old_dir
, new_dir
, old_dentry
, new_dentry
, NULL
);
2079 error
= PTR_ERR(task
);
2083 error
= rpc_wait_for_completion_task(task
);
2085 ((struct nfs_renamedata
*)task
->tk_calldata
)->cancelled
= 1;
2086 /* Paired with the atomic_dec_and_test() barrier in rpc_do_put_task() */
2089 error
= task
->tk_status
;
2091 /* Ensure the inode attributes are revalidated */
2093 spin_lock(&old_inode
->i_lock
);
2094 NFS_I(old_inode
)->attr_gencount
= nfs_inc_attr_generation_counter();
2095 NFS_I(old_inode
)->cache_validity
|= NFS_INO_INVALID_CHANGE
2096 | NFS_INO_INVALID_CTIME
2097 | NFS_INO_REVAL_FORCED
;
2098 spin_unlock(&old_inode
->i_lock
);
2103 trace_nfs_rename_exit(old_dir
, old_dentry
,
2104 new_dir
, new_dentry
, error
);
2106 if (new_inode
!= NULL
)
2107 nfs_drop_nlink(new_inode
);
2109 * The d_move() should be here instead of in an async RPC completion
2110 * handler because we need the proper locks to move the dentry. If
2111 * we're interrupted by a signal, the async RPC completion handler
2112 * should mark the directories for revalidation.
2114 d_move(old_dentry
, new_dentry
);
2115 nfs_set_verifier(old_dentry
,
2116 nfs_save_change_attribute(new_dir
));
2117 } else if (error
== -ENOENT
)
2118 nfs_dentry_handle_enoent(old_dentry
);
2120 /* new dentry created? */
2125 EXPORT_SYMBOL_GPL(nfs_rename
);
2127 static DEFINE_SPINLOCK(nfs_access_lru_lock
);
2128 static LIST_HEAD(nfs_access_lru_list
);
2129 static atomic_long_t nfs_access_nr_entries
;
2131 static unsigned long nfs_access_max_cachesize
= ULONG_MAX
;
2132 module_param(nfs_access_max_cachesize
, ulong
, 0644);
2133 MODULE_PARM_DESC(nfs_access_max_cachesize
, "NFS access maximum total cache length");
2135 static void nfs_access_free_entry(struct nfs_access_entry
*entry
)
2137 put_cred(entry
->cred
);
2138 kfree_rcu(entry
, rcu_head
);
2139 smp_mb__before_atomic();
2140 atomic_long_dec(&nfs_access_nr_entries
);
2141 smp_mb__after_atomic();
2144 static void nfs_access_free_list(struct list_head
*head
)
2146 struct nfs_access_entry
*cache
;
2148 while (!list_empty(head
)) {
2149 cache
= list_entry(head
->next
, struct nfs_access_entry
, lru
);
2150 list_del(&cache
->lru
);
2151 nfs_access_free_entry(cache
);
2155 static unsigned long
2156 nfs_do_access_cache_scan(unsigned int nr_to_scan
)
2159 struct nfs_inode
*nfsi
, *next
;
2160 struct nfs_access_entry
*cache
;
2163 spin_lock(&nfs_access_lru_lock
);
2164 list_for_each_entry_safe(nfsi
, next
, &nfs_access_lru_list
, access_cache_inode_lru
) {
2165 struct inode
*inode
;
2167 if (nr_to_scan
-- == 0)
2169 inode
= &nfsi
->vfs_inode
;
2170 spin_lock(&inode
->i_lock
);
2171 if (list_empty(&nfsi
->access_cache_entry_lru
))
2172 goto remove_lru_entry
;
2173 cache
= list_entry(nfsi
->access_cache_entry_lru
.next
,
2174 struct nfs_access_entry
, lru
);
2175 list_move(&cache
->lru
, &head
);
2176 rb_erase(&cache
->rb_node
, &nfsi
->access_cache
);
2178 if (!list_empty(&nfsi
->access_cache_entry_lru
))
2179 list_move_tail(&nfsi
->access_cache_inode_lru
,
2180 &nfs_access_lru_list
);
2183 list_del_init(&nfsi
->access_cache_inode_lru
);
2184 smp_mb__before_atomic();
2185 clear_bit(NFS_INO_ACL_LRU_SET
, &nfsi
->flags
);
2186 smp_mb__after_atomic();
2188 spin_unlock(&inode
->i_lock
);
2190 spin_unlock(&nfs_access_lru_lock
);
2191 nfs_access_free_list(&head
);
2196 nfs_access_cache_scan(struct shrinker
*shrink
, struct shrink_control
*sc
)
2198 int nr_to_scan
= sc
->nr_to_scan
;
2199 gfp_t gfp_mask
= sc
->gfp_mask
;
2201 if ((gfp_mask
& GFP_KERNEL
) != GFP_KERNEL
)
2203 return nfs_do_access_cache_scan(nr_to_scan
);
2208 nfs_access_cache_count(struct shrinker
*shrink
, struct shrink_control
*sc
)
2210 return vfs_pressure_ratio(atomic_long_read(&nfs_access_nr_entries
));
2214 nfs_access_cache_enforce_limit(void)
2216 long nr_entries
= atomic_long_read(&nfs_access_nr_entries
);
2218 unsigned int nr_to_scan
;
2220 if (nr_entries
< 0 || nr_entries
<= nfs_access_max_cachesize
)
2223 diff
= nr_entries
- nfs_access_max_cachesize
;
2224 if (diff
< nr_to_scan
)
2226 nfs_do_access_cache_scan(nr_to_scan
);
2229 static void __nfs_access_zap_cache(struct nfs_inode
*nfsi
, struct list_head
*head
)
2231 struct rb_root
*root_node
= &nfsi
->access_cache
;
2233 struct nfs_access_entry
*entry
;
2235 /* Unhook entries from the cache */
2236 while ((n
= rb_first(root_node
)) != NULL
) {
2237 entry
= rb_entry(n
, struct nfs_access_entry
, rb_node
);
2238 rb_erase(n
, root_node
);
2239 list_move(&entry
->lru
, head
);
2241 nfsi
->cache_validity
&= ~NFS_INO_INVALID_ACCESS
;
2244 void nfs_access_zap_cache(struct inode
*inode
)
2248 if (test_bit(NFS_INO_ACL_LRU_SET
, &NFS_I(inode
)->flags
) == 0)
2250 /* Remove from global LRU init */
2251 spin_lock(&nfs_access_lru_lock
);
2252 if (test_and_clear_bit(NFS_INO_ACL_LRU_SET
, &NFS_I(inode
)->flags
))
2253 list_del_init(&NFS_I(inode
)->access_cache_inode_lru
);
2255 spin_lock(&inode
->i_lock
);
2256 __nfs_access_zap_cache(NFS_I(inode
), &head
);
2257 spin_unlock(&inode
->i_lock
);
2258 spin_unlock(&nfs_access_lru_lock
);
2259 nfs_access_free_list(&head
);
2261 EXPORT_SYMBOL_GPL(nfs_access_zap_cache
);
2263 static struct nfs_access_entry
*nfs_access_search_rbtree(struct inode
*inode
, const struct cred
*cred
)
2265 struct rb_node
*n
= NFS_I(inode
)->access_cache
.rb_node
;
2268 struct nfs_access_entry
*entry
=
2269 rb_entry(n
, struct nfs_access_entry
, rb_node
);
2270 int cmp
= cred_fscmp(cred
, entry
->cred
);
2282 static int nfs_access_get_cached(struct inode
*inode
, const struct cred
*cred
, struct nfs_access_entry
*res
, bool may_block
)
2284 struct nfs_inode
*nfsi
= NFS_I(inode
);
2285 struct nfs_access_entry
*cache
;
2289 spin_lock(&inode
->i_lock
);
2291 if (nfsi
->cache_validity
& NFS_INO_INVALID_ACCESS
)
2293 cache
= nfs_access_search_rbtree(inode
, cred
);
2297 /* Found an entry, is our attribute cache valid? */
2298 if (!nfs_check_cache_invalid(inode
, NFS_INO_INVALID_ACCESS
))
2305 spin_unlock(&inode
->i_lock
);
2306 err
= __nfs_revalidate_inode(NFS_SERVER(inode
), inode
);
2309 spin_lock(&inode
->i_lock
);
2312 res
->cred
= cache
->cred
;
2313 res
->mask
= cache
->mask
;
2314 list_move_tail(&cache
->lru
, &nfsi
->access_cache_entry_lru
);
2317 spin_unlock(&inode
->i_lock
);
2320 spin_unlock(&inode
->i_lock
);
2321 nfs_access_zap_cache(inode
);
2325 static int nfs_access_get_cached_rcu(struct inode
*inode
, const struct cred
*cred
, struct nfs_access_entry
*res
)
2327 /* Only check the most recently returned cache entry,
2328 * but do it without locking.
2330 struct nfs_inode
*nfsi
= NFS_I(inode
);
2331 struct nfs_access_entry
*cache
;
2333 struct list_head
*lh
;
2336 if (nfsi
->cache_validity
& NFS_INO_INVALID_ACCESS
)
2338 lh
= rcu_dereference(nfsi
->access_cache_entry_lru
.prev
);
2339 cache
= list_entry(lh
, struct nfs_access_entry
, lru
);
2340 if (lh
== &nfsi
->access_cache_entry_lru
||
2341 cred
!= cache
->cred
)
2345 if (nfs_check_cache_invalid(inode
, NFS_INO_INVALID_ACCESS
))
2347 res
->cred
= cache
->cred
;
2348 res
->mask
= cache
->mask
;
2355 static void nfs_access_add_rbtree(struct inode
*inode
, struct nfs_access_entry
*set
)
2357 struct nfs_inode
*nfsi
= NFS_I(inode
);
2358 struct rb_root
*root_node
= &nfsi
->access_cache
;
2359 struct rb_node
**p
= &root_node
->rb_node
;
2360 struct rb_node
*parent
= NULL
;
2361 struct nfs_access_entry
*entry
;
2364 spin_lock(&inode
->i_lock
);
2365 while (*p
!= NULL
) {
2367 entry
= rb_entry(parent
, struct nfs_access_entry
, rb_node
);
2368 cmp
= cred_fscmp(set
->cred
, entry
->cred
);
2371 p
= &parent
->rb_left
;
2373 p
= &parent
->rb_right
;
2377 rb_link_node(&set
->rb_node
, parent
, p
);
2378 rb_insert_color(&set
->rb_node
, root_node
);
2379 list_add_tail(&set
->lru
, &nfsi
->access_cache_entry_lru
);
2380 spin_unlock(&inode
->i_lock
);
2383 rb_replace_node(parent
, &set
->rb_node
, root_node
);
2384 list_add_tail(&set
->lru
, &nfsi
->access_cache_entry_lru
);
2385 list_del(&entry
->lru
);
2386 spin_unlock(&inode
->i_lock
);
2387 nfs_access_free_entry(entry
);
2390 void nfs_access_add_cache(struct inode
*inode
, struct nfs_access_entry
*set
)
2392 struct nfs_access_entry
*cache
= kmalloc(sizeof(*cache
), GFP_KERNEL
);
2395 RB_CLEAR_NODE(&cache
->rb_node
);
2396 cache
->cred
= get_cred(set
->cred
);
2397 cache
->mask
= set
->mask
;
2399 /* The above field assignments must be visible
2400 * before this item appears on the lru. We cannot easily
2401 * use rcu_assign_pointer, so just force the memory barrier.
2404 nfs_access_add_rbtree(inode
, cache
);
2406 /* Update accounting */
2407 smp_mb__before_atomic();
2408 atomic_long_inc(&nfs_access_nr_entries
);
2409 smp_mb__after_atomic();
2411 /* Add inode to global LRU list */
2412 if (!test_bit(NFS_INO_ACL_LRU_SET
, &NFS_I(inode
)->flags
)) {
2413 spin_lock(&nfs_access_lru_lock
);
2414 if (!test_and_set_bit(NFS_INO_ACL_LRU_SET
, &NFS_I(inode
)->flags
))
2415 list_add_tail(&NFS_I(inode
)->access_cache_inode_lru
,
2416 &nfs_access_lru_list
);
2417 spin_unlock(&nfs_access_lru_lock
);
2419 nfs_access_cache_enforce_limit();
2421 EXPORT_SYMBOL_GPL(nfs_access_add_cache
);
2423 #define NFS_MAY_READ (NFS_ACCESS_READ)
2424 #define NFS_MAY_WRITE (NFS_ACCESS_MODIFY | \
2425 NFS_ACCESS_EXTEND | \
2427 #define NFS_FILE_MAY_WRITE (NFS_ACCESS_MODIFY | \
2429 #define NFS_DIR_MAY_WRITE NFS_MAY_WRITE
2430 #define NFS_MAY_LOOKUP (NFS_ACCESS_LOOKUP)
2431 #define NFS_MAY_EXECUTE (NFS_ACCESS_EXECUTE)
2433 nfs_access_calc_mask(u32 access_result
, umode_t umode
)
2437 if (access_result
& NFS_MAY_READ
)
2439 if (S_ISDIR(umode
)) {
2440 if ((access_result
& NFS_DIR_MAY_WRITE
) == NFS_DIR_MAY_WRITE
)
2442 if ((access_result
& NFS_MAY_LOOKUP
) == NFS_MAY_LOOKUP
)
2444 } else if (S_ISREG(umode
)) {
2445 if ((access_result
& NFS_FILE_MAY_WRITE
) == NFS_FILE_MAY_WRITE
)
2447 if ((access_result
& NFS_MAY_EXECUTE
) == NFS_MAY_EXECUTE
)
2449 } else if (access_result
& NFS_MAY_WRITE
)
2454 void nfs_access_set_mask(struct nfs_access_entry
*entry
, u32 access_result
)
2456 entry
->mask
= access_result
;
2458 EXPORT_SYMBOL_GPL(nfs_access_set_mask
);
2460 static int nfs_do_access(struct inode
*inode
, const struct cred
*cred
, int mask
)
2462 struct nfs_access_entry cache
;
2463 bool may_block
= (mask
& MAY_NOT_BLOCK
) == 0;
2467 trace_nfs_access_enter(inode
);
2469 status
= nfs_access_get_cached_rcu(inode
, cred
, &cache
);
2471 status
= nfs_access_get_cached(inode
, cred
, &cache
, may_block
);
2480 * Determine which access bits we want to ask for...
2482 cache
.mask
= NFS_ACCESS_READ
| NFS_ACCESS_MODIFY
| NFS_ACCESS_EXTEND
;
2483 if (S_ISDIR(inode
->i_mode
))
2484 cache
.mask
|= NFS_ACCESS_DELETE
| NFS_ACCESS_LOOKUP
;
2486 cache
.mask
|= NFS_ACCESS_EXECUTE
;
2488 status
= NFS_PROTO(inode
)->access(inode
, &cache
);
2490 if (status
== -ESTALE
) {
2491 nfs_zap_caches(inode
);
2492 if (!S_ISDIR(inode
->i_mode
))
2493 set_bit(NFS_INO_STALE
, &NFS_I(inode
)->flags
);
2497 nfs_access_add_cache(inode
, &cache
);
2499 cache_mask
= nfs_access_calc_mask(cache
.mask
, inode
->i_mode
);
2500 if ((mask
& ~cache_mask
& (MAY_READ
| MAY_WRITE
| MAY_EXEC
)) != 0)
2503 trace_nfs_access_exit(inode
, status
);
2507 static int nfs_open_permission_mask(int openflags
)
2511 if (openflags
& __FMODE_EXEC
) {
2512 /* ONLY check exec rights */
2515 if ((openflags
& O_ACCMODE
) != O_WRONLY
)
2517 if ((openflags
& O_ACCMODE
) != O_RDONLY
)
2524 int nfs_may_open(struct inode
*inode
, const struct cred
*cred
, int openflags
)
2526 return nfs_do_access(inode
, cred
, nfs_open_permission_mask(openflags
));
2528 EXPORT_SYMBOL_GPL(nfs_may_open
);
2530 static int nfs_execute_ok(struct inode
*inode
, int mask
)
2532 struct nfs_server
*server
= NFS_SERVER(inode
);
2535 if (S_ISDIR(inode
->i_mode
))
2537 if (nfs_check_cache_invalid(inode
, NFS_INO_INVALID_OTHER
)) {
2538 if (mask
& MAY_NOT_BLOCK
)
2540 ret
= __nfs_revalidate_inode(server
, inode
);
2542 if (ret
== 0 && !execute_ok(inode
))
2547 int nfs_permission(struct inode
*inode
, int mask
)
2549 const struct cred
*cred
= current_cred();
2552 nfs_inc_stats(inode
, NFSIOS_VFSACCESS
);
2554 if ((mask
& (MAY_READ
| MAY_WRITE
| MAY_EXEC
)) == 0)
2556 /* Is this sys_access() ? */
2557 if (mask
& (MAY_ACCESS
| MAY_CHDIR
))
2560 switch (inode
->i_mode
& S_IFMT
) {
2564 if ((mask
& MAY_OPEN
) &&
2565 nfs_server_capable(inode
, NFS_CAP_ATOMIC_OPEN
))
2570 * Optimize away all write operations, since the server
2571 * will check permissions when we perform the op.
2573 if ((mask
& MAY_WRITE
) && !(mask
& MAY_READ
))
2578 if (!NFS_PROTO(inode
)->access
)
2581 /* Always try fast lookups first */
2583 res
= nfs_do_access(inode
, cred
, mask
|MAY_NOT_BLOCK
);
2585 if (res
== -ECHILD
&& !(mask
& MAY_NOT_BLOCK
)) {
2586 /* Fast lookup failed, try the slow way */
2587 res
= nfs_do_access(inode
, cred
, mask
);
2590 if (!res
&& (mask
& MAY_EXEC
))
2591 res
= nfs_execute_ok(inode
, mask
);
2593 dfprintk(VFS
, "NFS: permission(%s/%lu), mask=0x%x, res=%d\n",
2594 inode
->i_sb
->s_id
, inode
->i_ino
, mask
, res
);
2597 if (mask
& MAY_NOT_BLOCK
)
2600 res
= nfs_revalidate_inode(NFS_SERVER(inode
), inode
);
2602 res
= generic_permission(inode
, mask
);
2605 EXPORT_SYMBOL_GPL(nfs_permission
);
2609 * version-control: t
2610 * kept-new-versions: 5