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/time.h>
21 #include <linux/errno.h>
22 #include <linux/stat.h>
23 #include <linux/fcntl.h>
24 #include <linux/string.h>
25 #include <linux/kernel.h>
26 #include <linux/slab.h>
28 #include <linux/sunrpc/clnt.h>
29 #include <linux/nfs_fs.h>
30 #include <linux/nfs_mount.h>
31 #include <linux/pagemap.h>
32 #include <linux/smp_lock.h>
33 #include <linux/pagevec.h>
34 #include <linux/namei.h>
35 #include <linux/mount.h>
36 #include <linux/sched.h>
39 #include "delegation.h"
42 /* #define NFS_DEBUG_VERBOSE 1 */
44 static int nfs_opendir(struct inode
*, struct file
*);
45 static int nfs_readdir(struct file
*, void *, filldir_t
);
46 static struct dentry
*nfs_lookup(struct inode
*, struct dentry
*, struct nameidata
*);
47 static int nfs_create(struct inode
*, struct dentry
*, int, struct nameidata
*);
48 static int nfs_mkdir(struct inode
*, struct dentry
*, int);
49 static int nfs_rmdir(struct inode
*, struct dentry
*);
50 static int nfs_unlink(struct inode
*, struct dentry
*);
51 static int nfs_symlink(struct inode
*, struct dentry
*, const char *);
52 static int nfs_link(struct dentry
*, struct inode
*, struct dentry
*);
53 static int nfs_mknod(struct inode
*, struct dentry
*, int, dev_t
);
54 static int nfs_rename(struct inode
*, struct dentry
*,
55 struct inode
*, struct dentry
*);
56 static int nfs_fsync_dir(struct file
*, struct dentry
*, int);
57 static loff_t
nfs_llseek_dir(struct file
*, loff_t
, int);
59 const struct file_operations nfs_dir_operations
= {
60 .llseek
= nfs_llseek_dir
,
61 .read
= generic_read_dir
,
62 .readdir
= nfs_readdir
,
64 .release
= nfs_release
,
65 .fsync
= nfs_fsync_dir
,
68 const struct inode_operations nfs_dir_inode_operations
= {
73 .symlink
= nfs_symlink
,
78 .permission
= nfs_permission
,
79 .getattr
= nfs_getattr
,
80 .setattr
= nfs_setattr
,
84 const struct inode_operations nfs3_dir_inode_operations
= {
89 .symlink
= nfs_symlink
,
94 .permission
= nfs_permission
,
95 .getattr
= nfs_getattr
,
96 .setattr
= nfs_setattr
,
97 .listxattr
= nfs3_listxattr
,
98 .getxattr
= nfs3_getxattr
,
99 .setxattr
= nfs3_setxattr
,
100 .removexattr
= nfs3_removexattr
,
102 #endif /* CONFIG_NFS_V3 */
106 static struct dentry
*nfs_atomic_lookup(struct inode
*, struct dentry
*, struct nameidata
*);
107 const struct inode_operations nfs4_dir_inode_operations
= {
108 .create
= nfs_create
,
109 .lookup
= nfs_atomic_lookup
,
111 .unlink
= nfs_unlink
,
112 .symlink
= nfs_symlink
,
116 .rename
= nfs_rename
,
117 .permission
= nfs_permission
,
118 .getattr
= nfs_getattr
,
119 .setattr
= nfs_setattr
,
120 .getxattr
= nfs4_getxattr
,
121 .setxattr
= nfs4_setxattr
,
122 .listxattr
= nfs4_listxattr
,
125 #endif /* CONFIG_NFS_V4 */
131 nfs_opendir(struct inode
*inode
, struct file
*filp
)
135 dfprintk(VFS
, "NFS: opendir(%s/%ld)\n",
136 inode
->i_sb
->s_id
, inode
->i_ino
);
139 /* Call generic open code in order to cache credentials */
140 res
= nfs_open(inode
, filp
);
145 typedef __be32
* (*decode_dirent_t
)(__be32
*, struct nfs_entry
*, int);
149 unsigned long page_index
;
152 loff_t current_index
;
153 struct nfs_entry
*entry
;
154 decode_dirent_t decode
;
157 unsigned long timestamp
;
159 } nfs_readdir_descriptor_t
;
161 /* Now we cache directories properly, by stuffing the dirent
162 * data directly in the page cache.
164 * Inode invalidation due to refresh etc. takes care of
165 * _everything_, no sloppy entry flushing logic, no extraneous
166 * copying, network direct to page cache, the way it was meant
169 * NOTE: Dirent information verification is done always by the
170 * page-in of the RPC reply, nowhere else, this simplies
171 * things substantially.
174 int nfs_readdir_filler(nfs_readdir_descriptor_t
*desc
, struct page
*page
)
176 struct file
*file
= desc
->file
;
177 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
178 struct rpc_cred
*cred
= nfs_file_cred(file
);
179 unsigned long timestamp
;
182 dfprintk(DIRCACHE
, "NFS: %s: reading cookie %Lu into page %lu\n",
183 __FUNCTION__
, (long long)desc
->entry
->cookie
,
188 error
= NFS_PROTO(inode
)->readdir(file
->f_path
.dentry
, cred
, desc
->entry
->cookie
, page
,
189 NFS_SERVER(inode
)->dtsize
, desc
->plus
);
191 /* We requested READDIRPLUS, but the server doesn't grok it */
192 if (error
== -ENOTSUPP
&& desc
->plus
) {
193 NFS_SERVER(inode
)->caps
&= ~NFS_CAP_READDIRPLUS
;
194 clear_bit(NFS_INO_ADVISE_RDPLUS
, &NFS_FLAGS(inode
));
200 desc
->timestamp
= timestamp
;
201 desc
->timestamp_valid
= 1;
202 SetPageUptodate(page
);
203 /* Ensure consistent page alignment of the data.
204 * Note: assumes we have exclusive access to this mapping either
205 * through inode->i_mutex or some other mechanism.
207 if (page
->index
== 0 && invalidate_inode_pages2_range(inode
->i_mapping
, PAGE_CACHE_SIZE
, -1) < 0) {
208 /* Should never happen */
209 nfs_zap_mapping(inode
, inode
->i_mapping
);
220 int dir_decode(nfs_readdir_descriptor_t
*desc
)
222 __be32
*p
= desc
->ptr
;
223 p
= desc
->decode(p
, desc
->entry
, desc
->plus
);
227 if (desc
->timestamp_valid
)
228 desc
->entry
->fattr
->time_start
= desc
->timestamp
;
230 desc
->entry
->fattr
->valid
&= ~NFS_ATTR_FATTR
;
235 void dir_page_release(nfs_readdir_descriptor_t
*desc
)
238 page_cache_release(desc
->page
);
244 * Given a pointer to a buffer that has already been filled by a call
245 * to readdir, find the next entry with cookie '*desc->dir_cookie'.
247 * If the end of the buffer has been reached, return -EAGAIN, if not,
248 * return the offset within the buffer of the next entry to be
252 int find_dirent(nfs_readdir_descriptor_t
*desc
)
254 struct nfs_entry
*entry
= desc
->entry
;
258 while((status
= dir_decode(desc
)) == 0) {
259 dfprintk(DIRCACHE
, "NFS: %s: examining cookie %Lu\n",
260 __FUNCTION__
, (unsigned long long)entry
->cookie
);
261 if (entry
->prev_cookie
== *desc
->dir_cookie
)
263 if (loop_count
++ > 200) {
272 * Given a pointer to a buffer that has already been filled by a call
273 * to readdir, find the entry at offset 'desc->file->f_pos'.
275 * If the end of the buffer has been reached, return -EAGAIN, if not,
276 * return the offset within the buffer of the next entry to be
280 int find_dirent_index(nfs_readdir_descriptor_t
*desc
)
282 struct nfs_entry
*entry
= desc
->entry
;
287 status
= dir_decode(desc
);
291 dfprintk(DIRCACHE
, "NFS: found cookie %Lu at index %Ld\n",
292 (unsigned long long)entry
->cookie
, desc
->current_index
);
294 if (desc
->file
->f_pos
== desc
->current_index
) {
295 *desc
->dir_cookie
= entry
->cookie
;
298 desc
->current_index
++;
299 if (loop_count
++ > 200) {
308 * Find the given page, and call find_dirent() or find_dirent_index in
309 * order to try to return the next entry.
312 int find_dirent_page(nfs_readdir_descriptor_t
*desc
)
314 struct inode
*inode
= desc
->file
->f_path
.dentry
->d_inode
;
318 dfprintk(DIRCACHE
, "NFS: %s: searching page %ld for target %Lu\n",
319 __FUNCTION__
, desc
->page_index
,
320 (long long) *desc
->dir_cookie
);
322 /* If we find the page in the page_cache, we cannot be sure
323 * how fresh the data is, so we will ignore readdir_plus attributes.
325 desc
->timestamp_valid
= 0;
326 page
= read_cache_page(inode
->i_mapping
, desc
->page_index
,
327 (filler_t
*)nfs_readdir_filler
, desc
);
329 status
= PTR_ERR(page
);
333 /* NOTE: Someone else may have changed the READDIRPLUS flag */
335 desc
->ptr
= kmap(page
); /* matching kunmap in nfs_do_filldir */
336 if (*desc
->dir_cookie
!= 0)
337 status
= find_dirent(desc
);
339 status
= find_dirent_index(desc
);
341 dir_page_release(desc
);
343 dfprintk(DIRCACHE
, "NFS: %s: returns %d\n", __FUNCTION__
, status
);
348 * Recurse through the page cache pages, and return a
349 * filled nfs_entry structure of the next directory entry if possible.
351 * The target for the search is '*desc->dir_cookie' if non-0,
352 * 'desc->file->f_pos' otherwise
355 int readdir_search_pagecache(nfs_readdir_descriptor_t
*desc
)
360 /* Always search-by-index from the beginning of the cache */
361 if (*desc
->dir_cookie
== 0) {
362 dfprintk(DIRCACHE
, "NFS: readdir_search_pagecache() searching for offset %Ld\n",
363 (long long)desc
->file
->f_pos
);
364 desc
->page_index
= 0;
365 desc
->entry
->cookie
= desc
->entry
->prev_cookie
= 0;
366 desc
->entry
->eof
= 0;
367 desc
->current_index
= 0;
369 dfprintk(DIRCACHE
, "NFS: readdir_search_pagecache() searching for cookie %Lu\n",
370 (unsigned long long)*desc
->dir_cookie
);
373 res
= find_dirent_page(desc
);
376 /* Align to beginning of next page */
378 if (loop_count
++ > 200) {
384 dfprintk(DIRCACHE
, "NFS: %s: returns %d\n", __FUNCTION__
, res
);
388 static inline unsigned int dt_type(struct inode
*inode
)
390 return (inode
->i_mode
>> 12) & 15;
393 static struct dentry
*nfs_readdir_lookup(nfs_readdir_descriptor_t
*desc
);
396 * Once we've found the start of the dirent within a page: fill 'er up...
399 int nfs_do_filldir(nfs_readdir_descriptor_t
*desc
, void *dirent
,
402 struct file
*file
= desc
->file
;
403 struct nfs_entry
*entry
= desc
->entry
;
404 struct dentry
*dentry
= NULL
;
409 dfprintk(DIRCACHE
, "NFS: nfs_do_filldir() filling starting @ cookie %Lu\n",
410 (unsigned long long)entry
->cookie
);
413 unsigned d_type
= DT_UNKNOWN
;
414 /* Note: entry->prev_cookie contains the cookie for
415 * retrieving the current dirent on the server */
418 /* Get a dentry if we have one */
421 dentry
= nfs_readdir_lookup(desc
);
423 /* Use readdirplus info */
424 if (dentry
!= NULL
&& dentry
->d_inode
!= NULL
) {
425 d_type
= dt_type(dentry
->d_inode
);
426 fileid
= NFS_FILEID(dentry
->d_inode
);
429 res
= filldir(dirent
, entry
->name
, entry
->len
,
430 file
->f_pos
, fileid
, d_type
);
434 *desc
->dir_cookie
= entry
->cookie
;
435 if (dir_decode(desc
) != 0) {
439 if (loop_count
++ > 200) {
444 dir_page_release(desc
);
447 dfprintk(DIRCACHE
, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n",
448 (unsigned long long)*desc
->dir_cookie
, res
);
453 * If we cannot find a cookie in our cache, we suspect that this is
454 * because it points to a deleted file, so we ask the server to return
455 * whatever it thinks is the next entry. We then feed this to filldir.
456 * If all goes well, we should then be able to find our way round the
457 * cache on the next call to readdir_search_pagecache();
459 * NOTE: we cannot add the anonymous page to the pagecache because
460 * the data it contains might not be page aligned. Besides,
461 * we should already have a complete representation of the
462 * directory in the page cache by the time we get here.
465 int uncached_readdir(nfs_readdir_descriptor_t
*desc
, void *dirent
,
468 struct file
*file
= desc
->file
;
469 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
470 struct rpc_cred
*cred
= nfs_file_cred(file
);
471 struct page
*page
= NULL
;
473 unsigned long timestamp
;
475 dfprintk(DIRCACHE
, "NFS: uncached_readdir() searching for cookie %Lu\n",
476 (unsigned long long)*desc
->dir_cookie
);
478 page
= alloc_page(GFP_HIGHUSER
);
484 desc
->error
= NFS_PROTO(inode
)->readdir(file
->f_path
.dentry
, cred
, *desc
->dir_cookie
,
486 NFS_SERVER(inode
)->dtsize
,
489 desc
->ptr
= kmap(page
); /* matching kunmap in nfs_do_filldir */
490 if (desc
->error
>= 0) {
491 desc
->timestamp
= timestamp
;
492 desc
->timestamp_valid
= 1;
493 if ((status
= dir_decode(desc
)) == 0)
494 desc
->entry
->prev_cookie
= *desc
->dir_cookie
;
500 status
= nfs_do_filldir(desc
, dirent
, filldir
);
502 /* Reset read descriptor so it searches the page cache from
503 * the start upon the next call to readdir_search_pagecache() */
504 desc
->page_index
= 0;
505 desc
->entry
->cookie
= desc
->entry
->prev_cookie
= 0;
506 desc
->entry
->eof
= 0;
508 dfprintk(DIRCACHE
, "NFS: %s: returns %d\n",
509 __FUNCTION__
, status
);
512 dir_page_release(desc
);
516 /* The file offset position represents the dirent entry number. A
517 last cookie cache takes care of the common case of reading the
520 static int nfs_readdir(struct file
*filp
, void *dirent
, filldir_t filldir
)
522 struct dentry
*dentry
= filp
->f_path
.dentry
;
523 struct inode
*inode
= dentry
->d_inode
;
524 nfs_readdir_descriptor_t my_desc
,
526 struct nfs_entry my_entry
;
528 struct nfs_fattr fattr
;
531 dfprintk(VFS
, "NFS: readdir(%s/%s) starting at cookie %Lu\n",
532 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
,
533 (long long)filp
->f_pos
);
534 nfs_inc_stats(inode
, NFSIOS_VFSGETDENTS
);
538 res
= nfs_revalidate_mapping_nolock(inode
, filp
->f_mapping
);
545 * filp->f_pos points to the dirent entry number.
546 * *desc->dir_cookie has the cookie for the next entry. We have
547 * to either find the entry with the appropriate number or
548 * revalidate the cookie.
550 memset(desc
, 0, sizeof(*desc
));
553 desc
->dir_cookie
= &nfs_file_open_context(filp
)->dir_cookie
;
554 desc
->decode
= NFS_PROTO(inode
)->decode_dirent
;
555 desc
->plus
= NFS_USE_READDIRPLUS(inode
);
557 my_entry
.cookie
= my_entry
.prev_cookie
= 0;
560 my_entry
.fattr
= &fattr
;
561 nfs_fattr_init(&fattr
);
562 desc
->entry
= &my_entry
;
564 while(!desc
->entry
->eof
) {
565 res
= readdir_search_pagecache(desc
);
567 if (res
== -EBADCOOKIE
) {
568 /* This means either end of directory */
569 if (*desc
->dir_cookie
&& desc
->entry
->cookie
!= *desc
->dir_cookie
) {
570 /* Or that the server has 'lost' a cookie */
571 res
= uncached_readdir(desc
, dirent
, filldir
);
578 if (res
== -ETOOSMALL
&& desc
->plus
) {
579 clear_bit(NFS_INO_ADVISE_RDPLUS
, &NFS_FLAGS(inode
));
580 nfs_zap_caches(inode
);
582 desc
->entry
->eof
= 0;
588 res
= nfs_do_filldir(desc
, dirent
, filldir
);
597 dfprintk(VFS
, "NFS: readdir(%s/%s) returns %ld\n",
598 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
,
603 static loff_t
nfs_llseek_dir(struct file
*filp
, loff_t offset
, int origin
)
605 mutex_lock(&filp
->f_path
.dentry
->d_inode
->i_mutex
);
608 offset
+= filp
->f_pos
;
616 if (offset
!= filp
->f_pos
) {
617 filp
->f_pos
= offset
;
618 nfs_file_open_context(filp
)->dir_cookie
= 0;
621 mutex_unlock(&filp
->f_path
.dentry
->d_inode
->i_mutex
);
626 * All directory operations under NFS are synchronous, so fsync()
627 * is a dummy operation.
629 static int nfs_fsync_dir(struct file
*filp
, struct dentry
*dentry
, int datasync
)
631 dfprintk(VFS
, "NFS: fsync_dir(%s/%s) datasync %d\n",
632 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
,
639 * A check for whether or not the parent directory has changed.
640 * In the case it has, we assume that the dentries are untrustworthy
641 * and may need to be looked up again.
643 static int nfs_check_verifier(struct inode
*dir
, struct dentry
*dentry
)
647 if (!nfs_verify_change_attribute(dir
, dentry
->d_time
))
649 /* Revalidate nfsi->cache_change_attribute before we declare a match */
650 if (nfs_revalidate_inode(NFS_SERVER(dir
), dir
) < 0)
652 if (!nfs_verify_change_attribute(dir
, dentry
->d_time
))
658 * Return the intent data that applies to this particular path component
660 * Note that the current set of intents only apply to the very last
661 * component of the path.
662 * We check for this using LOOKUP_CONTINUE and LOOKUP_PARENT.
664 static inline unsigned int nfs_lookup_check_intent(struct nameidata
*nd
, unsigned int mask
)
666 if (nd
->flags
& (LOOKUP_CONTINUE
|LOOKUP_PARENT
))
668 return nd
->flags
& mask
;
672 * Use intent information to check whether or not we're going to do
673 * an O_EXCL create using this path component.
675 static int nfs_is_exclusive_create(struct inode
*dir
, struct nameidata
*nd
)
677 if (NFS_PROTO(dir
)->version
== 2)
679 if (nd
== NULL
|| nfs_lookup_check_intent(nd
, LOOKUP_CREATE
) == 0)
681 return (nd
->intent
.open
.flags
& O_EXCL
) != 0;
685 * Inode and filehandle revalidation for lookups.
687 * We force revalidation in the cases where the VFS sets LOOKUP_REVAL,
688 * or if the intent information indicates that we're about to open this
689 * particular file and the "nocto" mount flag is not set.
693 int nfs_lookup_verify_inode(struct inode
*inode
, struct nameidata
*nd
)
695 struct nfs_server
*server
= NFS_SERVER(inode
);
698 /* VFS wants an on-the-wire revalidation */
699 if (nd
->flags
& LOOKUP_REVAL
)
701 /* This is an open(2) */
702 if (nfs_lookup_check_intent(nd
, LOOKUP_OPEN
) != 0 &&
703 !(server
->flags
& NFS_MOUNT_NOCTO
) &&
704 (S_ISREG(inode
->i_mode
) ||
705 S_ISDIR(inode
->i_mode
)))
708 return nfs_revalidate_inode(server
, inode
);
710 return __nfs_revalidate_inode(server
, inode
);
714 * We judge how long we want to trust negative
715 * dentries by looking at the parent inode mtime.
717 * If parent mtime has changed, we revalidate, else we wait for a
718 * period corresponding to the parent's attribute cache timeout value.
721 int nfs_neg_need_reval(struct inode
*dir
, struct dentry
*dentry
,
722 struct nameidata
*nd
)
724 /* Don't revalidate a negative dentry if we're creating a new file */
725 if (nd
!= NULL
&& nfs_lookup_check_intent(nd
, LOOKUP_CREATE
) != 0)
727 return !nfs_check_verifier(dir
, dentry
);
731 * This is called every time the dcache has a lookup hit,
732 * and we should check whether we can really trust that
735 * NOTE! The hit can be a negative hit too, don't assume
738 * If the parent directory is seen to have changed, we throw out the
739 * cached dentry and do a new lookup.
741 static int nfs_lookup_revalidate(struct dentry
* dentry
, struct nameidata
*nd
)
745 struct dentry
*parent
;
747 struct nfs_fh fhandle
;
748 struct nfs_fattr fattr
;
750 parent
= dget_parent(dentry
);
752 dir
= parent
->d_inode
;
753 nfs_inc_stats(dir
, NFSIOS_DENTRYREVALIDATE
);
754 inode
= dentry
->d_inode
;
757 if (nfs_neg_need_reval(dir
, dentry
, nd
))
762 if (is_bad_inode(inode
)) {
763 dfprintk(LOOKUPCACHE
, "%s: %s/%s has dud inode\n",
764 __FUNCTION__
, dentry
->d_parent
->d_name
.name
,
765 dentry
->d_name
.name
);
769 /* Force a full look up iff the parent directory has changed */
770 if (!nfs_is_exclusive_create(dir
, nd
) && nfs_check_verifier(dir
, dentry
)) {
771 if (nfs_lookup_verify_inode(inode
, nd
))
776 if (NFS_STALE(inode
))
779 error
= NFS_PROTO(dir
)->lookup(dir
, &dentry
->d_name
, &fhandle
, &fattr
);
782 if (nfs_compare_fh(NFS_FH(inode
), &fhandle
))
784 if ((error
= nfs_refresh_inode(inode
, &fattr
)) != 0)
787 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
791 dfprintk(LOOKUPCACHE
, "NFS: %s(%s/%s) is valid\n",
792 __FUNCTION__
, dentry
->d_parent
->d_name
.name
,
793 dentry
->d_name
.name
);
798 nfs_mark_for_revalidate(dir
);
799 if (inode
&& S_ISDIR(inode
->i_mode
)) {
800 /* Purge readdir caches. */
801 nfs_zap_caches(inode
);
802 /* If we have submounts, don't unhash ! */
803 if (have_submounts(dentry
))
805 shrink_dcache_parent(dentry
);
810 dfprintk(LOOKUPCACHE
, "NFS: %s(%s/%s) is invalid\n",
811 __FUNCTION__
, dentry
->d_parent
->d_name
.name
,
812 dentry
->d_name
.name
);
817 * This is called from dput() when d_count is going to 0.
819 static int nfs_dentry_delete(struct dentry
*dentry
)
821 dfprintk(VFS
, "NFS: dentry_delete(%s/%s, %x)\n",
822 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
,
825 if (dentry
->d_flags
& DCACHE_NFSFS_RENAMED
) {
826 /* Unhash it, so that ->d_iput() would be called */
829 if (!(dentry
->d_sb
->s_flags
& MS_ACTIVE
)) {
830 /* Unhash it, so that ancestors of killed async unlink
831 * files will be cleaned up during umount */
839 * Called when the dentry loses inode.
840 * We use it to clean up silly-renamed files.
842 static void nfs_dentry_iput(struct dentry
*dentry
, struct inode
*inode
)
844 nfs_inode_return_delegation(inode
);
845 if (S_ISDIR(inode
->i_mode
))
846 /* drop any readdir cache as it could easily be old */
847 NFS_I(inode
)->cache_validity
|= NFS_INO_INVALID_DATA
;
849 if (dentry
->d_flags
& DCACHE_NFSFS_RENAMED
) {
852 nfs_complete_unlink(dentry
, inode
);
858 struct dentry_operations nfs_dentry_operations
= {
859 .d_revalidate
= nfs_lookup_revalidate
,
860 .d_delete
= nfs_dentry_delete
,
861 .d_iput
= nfs_dentry_iput
,
864 static struct dentry
*nfs_lookup(struct inode
*dir
, struct dentry
* dentry
, struct nameidata
*nd
)
867 struct inode
*inode
= NULL
;
869 struct nfs_fh fhandle
;
870 struct nfs_fattr fattr
;
872 dfprintk(VFS
, "NFS: lookup(%s/%s)\n",
873 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
);
874 nfs_inc_stats(dir
, NFSIOS_VFSLOOKUP
);
876 res
= ERR_PTR(-ENAMETOOLONG
);
877 if (dentry
->d_name
.len
> NFS_SERVER(dir
)->namelen
)
880 res
= ERR_PTR(-ENOMEM
);
881 dentry
->d_op
= NFS_PROTO(dir
)->dentry_ops
;
886 * If we're doing an exclusive create, optimize away the lookup
887 * but don't hash the dentry.
889 if (nfs_is_exclusive_create(dir
, nd
)) {
890 d_instantiate(dentry
, NULL
);
895 error
= NFS_PROTO(dir
)->lookup(dir
, &dentry
->d_name
, &fhandle
, &fattr
);
896 if (error
== -ENOENT
)
899 res
= ERR_PTR(error
);
902 inode
= nfs_fhget(dentry
->d_sb
, &fhandle
, &fattr
);
903 res
= (struct dentry
*)inode
;
908 res
= d_materialise_unique(dentry
, inode
);
914 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
922 static int nfs_open_revalidate(struct dentry
*, struct nameidata
*);
924 struct dentry_operations nfs4_dentry_operations
= {
925 .d_revalidate
= nfs_open_revalidate
,
926 .d_delete
= nfs_dentry_delete
,
927 .d_iput
= nfs_dentry_iput
,
931 * Use intent information to determine whether we need to substitute
932 * the NFSv4-style stateful OPEN for the LOOKUP call
934 static int is_atomic_open(struct inode
*dir
, struct nameidata
*nd
)
936 if (nd
== NULL
|| nfs_lookup_check_intent(nd
, LOOKUP_OPEN
) == 0)
938 /* NFS does not (yet) have a stateful open for directories */
939 if (nd
->flags
& LOOKUP_DIRECTORY
)
941 /* Are we trying to write to a read only partition? */
942 if (IS_RDONLY(dir
) && (nd
->intent
.open
.flags
& (O_CREAT
|O_TRUNC
|FMODE_WRITE
)))
947 static struct dentry
*nfs_atomic_lookup(struct inode
*dir
, struct dentry
*dentry
, struct nameidata
*nd
)
949 struct dentry
*res
= NULL
;
952 dfprintk(VFS
, "NFS: atomic_lookup(%s/%ld), %s\n",
953 dir
->i_sb
->s_id
, dir
->i_ino
, dentry
->d_name
.name
);
955 /* Check that we are indeed trying to open this file */
956 if (!is_atomic_open(dir
, nd
))
959 if (dentry
->d_name
.len
> NFS_SERVER(dir
)->namelen
) {
960 res
= ERR_PTR(-ENAMETOOLONG
);
963 dentry
->d_op
= NFS_PROTO(dir
)->dentry_ops
;
965 /* Let vfs_create() deal with O_EXCL. Instantiate, but don't hash
967 if (nd
->intent
.open
.flags
& O_EXCL
) {
968 d_instantiate(dentry
, NULL
);
972 /* Open the file on the server */
974 res
= nfs4_atomic_open(dir
, dentry
, nd
);
977 error
= PTR_ERR(res
);
979 /* Make a negative dentry */
983 /* This turned out not to be a regular file */
988 if (!(nd
->intent
.open
.flags
& O_NOFOLLOW
))
994 } else if (res
!= NULL
)
999 return nfs_lookup(dir
, dentry
, nd
);
1002 static int nfs_open_revalidate(struct dentry
*dentry
, struct nameidata
*nd
)
1004 struct dentry
*parent
= NULL
;
1005 struct inode
*inode
= dentry
->d_inode
;
1007 int openflags
, ret
= 0;
1009 parent
= dget_parent(dentry
);
1010 dir
= parent
->d_inode
;
1011 if (!is_atomic_open(dir
, nd
))
1013 /* We can't create new files in nfs_open_revalidate(), so we
1014 * optimize away revalidation of negative dentries.
1016 if (inode
== NULL
) {
1017 if (!nfs_neg_need_reval(dir
, dentry
, nd
))
1022 /* NFS only supports OPEN on regular files */
1023 if (!S_ISREG(inode
->i_mode
))
1025 openflags
= nd
->intent
.open
.flags
;
1026 /* We cannot do exclusive creation on a positive dentry */
1027 if ((openflags
& (O_CREAT
|O_EXCL
)) == (O_CREAT
|O_EXCL
))
1029 /* We can't create new files, or truncate existing ones here */
1030 openflags
&= ~(O_CREAT
|O_TRUNC
);
1033 * Note: we're not holding inode->i_mutex and so may be racing with
1034 * operations that change the directory. We therefore save the
1035 * change attribute *before* we do the RPC call.
1038 ret
= nfs4_open_revalidate(dir
, dentry
, openflags
, nd
);
1047 if (inode
!= NULL
&& nfs_have_delegation(inode
, FMODE_READ
))
1049 return nfs_lookup_revalidate(dentry
, nd
);
1051 #endif /* CONFIG_NFSV4 */
1053 static struct dentry
*nfs_readdir_lookup(nfs_readdir_descriptor_t
*desc
)
1055 struct dentry
*parent
= desc
->file
->f_path
.dentry
;
1056 struct inode
*dir
= parent
->d_inode
;
1057 struct nfs_entry
*entry
= desc
->entry
;
1058 struct dentry
*dentry
, *alias
;
1059 struct qstr name
= {
1060 .name
= entry
->name
,
1063 struct inode
*inode
;
1064 unsigned long verf
= nfs_save_change_attribute(dir
);
1068 if (name
.name
[0] == '.' && name
.name
[1] == '.')
1069 return dget_parent(parent
);
1072 if (name
.name
[0] == '.')
1073 return dget(parent
);
1076 spin_lock(&dir
->i_lock
);
1077 if (NFS_I(dir
)->cache_validity
& NFS_INO_INVALID_DATA
) {
1078 spin_unlock(&dir
->i_lock
);
1081 spin_unlock(&dir
->i_lock
);
1083 name
.hash
= full_name_hash(name
.name
, name
.len
);
1084 dentry
= d_lookup(parent
, &name
);
1085 if (dentry
!= NULL
) {
1086 /* Is this a positive dentry that matches the readdir info? */
1087 if (dentry
->d_inode
!= NULL
&&
1088 (NFS_FILEID(dentry
->d_inode
) == entry
->ino
||
1089 d_mountpoint(dentry
))) {
1090 if (!desc
->plus
|| entry
->fh
->size
== 0)
1092 if (nfs_compare_fh(NFS_FH(dentry
->d_inode
),
1096 /* No, so d_drop to allow one to be created */
1100 if (!desc
->plus
|| !(entry
->fattr
->valid
& NFS_ATTR_FATTR
))
1102 if (name
.len
> NFS_SERVER(dir
)->namelen
)
1104 /* Note: caller is already holding the dir->i_mutex! */
1105 dentry
= d_alloc(parent
, &name
);
1108 dentry
->d_op
= NFS_PROTO(dir
)->dentry_ops
;
1109 inode
= nfs_fhget(dentry
->d_sb
, entry
->fh
, entry
->fattr
);
1110 if (IS_ERR(inode
)) {
1115 alias
= d_materialise_unique(dentry
, inode
);
1116 if (alias
!= NULL
) {
1124 nfs_set_verifier(dentry
, verf
);
1129 * Code common to create, mkdir, and mknod.
1131 int nfs_instantiate(struct dentry
*dentry
, struct nfs_fh
*fhandle
,
1132 struct nfs_fattr
*fattr
)
1134 struct dentry
*parent
= dget_parent(dentry
);
1135 struct inode
*dir
= parent
->d_inode
;
1136 struct inode
*inode
;
1137 int error
= -EACCES
;
1141 /* We may have been initialized further down */
1142 if (dentry
->d_inode
)
1144 if (fhandle
->size
== 0) {
1145 error
= NFS_PROTO(dir
)->lookup(dir
, &dentry
->d_name
, fhandle
, fattr
);
1149 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1150 if (!(fattr
->valid
& NFS_ATTR_FATTR
)) {
1151 struct nfs_server
*server
= NFS_SB(dentry
->d_sb
);
1152 error
= server
->nfs_client
->rpc_ops
->getattr(server
, fhandle
, fattr
);
1156 inode
= nfs_fhget(dentry
->d_sb
, fhandle
, fattr
);
1157 error
= PTR_ERR(inode
);
1160 d_add(dentry
, inode
);
1165 nfs_mark_for_revalidate(dir
);
1171 * Following a failed create operation, we drop the dentry rather
1172 * than retain a negative dentry. This avoids a problem in the event
1173 * that the operation succeeded on the server, but an error in the
1174 * reply path made it appear to have failed.
1176 static int nfs_create(struct inode
*dir
, struct dentry
*dentry
, int mode
,
1177 struct nameidata
*nd
)
1183 dfprintk(VFS
, "NFS: create(%s/%ld), %s\n",
1184 dir
->i_sb
->s_id
, dir
->i_ino
, dentry
->d_name
.name
);
1186 attr
.ia_mode
= mode
;
1187 attr
.ia_valid
= ATTR_MODE
;
1189 if ((nd
->flags
& LOOKUP_CREATE
) != 0)
1190 open_flags
= nd
->intent
.open
.flags
;
1193 error
= NFS_PROTO(dir
)->create(dir
, dentry
, &attr
, open_flags
, nd
);
1205 * See comments for nfs_proc_create regarding failed operations.
1208 nfs_mknod(struct inode
*dir
, struct dentry
*dentry
, int mode
, dev_t rdev
)
1213 dfprintk(VFS
, "NFS: mknod(%s/%ld), %s\n",
1214 dir
->i_sb
->s_id
, dir
->i_ino
, dentry
->d_name
.name
);
1216 if (!new_valid_dev(rdev
))
1219 attr
.ia_mode
= mode
;
1220 attr
.ia_valid
= ATTR_MODE
;
1223 status
= NFS_PROTO(dir
)->mknod(dir
, dentry
, &attr
, rdev
);
1235 * See comments for nfs_proc_create regarding failed operations.
1237 static int nfs_mkdir(struct inode
*dir
, struct dentry
*dentry
, int mode
)
1242 dfprintk(VFS
, "NFS: mkdir(%s/%ld), %s\n",
1243 dir
->i_sb
->s_id
, dir
->i_ino
, dentry
->d_name
.name
);
1245 attr
.ia_valid
= ATTR_MODE
;
1246 attr
.ia_mode
= mode
| S_IFDIR
;
1249 error
= NFS_PROTO(dir
)->mkdir(dir
, dentry
, &attr
);
1260 static int nfs_rmdir(struct inode
*dir
, struct dentry
*dentry
)
1264 dfprintk(VFS
, "NFS: rmdir(%s/%ld), %s\n",
1265 dir
->i_sb
->s_id
, dir
->i_ino
, dentry
->d_name
.name
);
1268 error
= NFS_PROTO(dir
)->rmdir(dir
, &dentry
->d_name
);
1269 /* Ensure the VFS deletes this inode */
1270 if (error
== 0 && dentry
->d_inode
!= NULL
)
1271 clear_nlink(dentry
->d_inode
);
1277 static int nfs_sillyrename(struct inode
*dir
, struct dentry
*dentry
)
1279 static unsigned int sillycounter
;
1280 const int fileidsize
= sizeof(NFS_FILEID(dentry
->d_inode
))*2;
1281 const int countersize
= sizeof(sillycounter
)*2;
1282 const int slen
= sizeof(".nfs")+fileidsize
+countersize
-1;
1285 struct dentry
*sdentry
;
1288 dfprintk(VFS
, "NFS: silly-rename(%s/%s, ct=%d)\n",
1289 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
,
1290 atomic_read(&dentry
->d_count
));
1291 nfs_inc_stats(dir
, NFSIOS_SILLYRENAME
);
1294 * We don't allow a dentry to be silly-renamed twice.
1297 if (dentry
->d_flags
& DCACHE_NFSFS_RENAMED
)
1300 sprintf(silly
, ".nfs%*.*Lx",
1301 fileidsize
, fileidsize
,
1302 (unsigned long long)NFS_FILEID(dentry
->d_inode
));
1304 /* Return delegation in anticipation of the rename */
1305 nfs_inode_return_delegation(dentry
->d_inode
);
1309 char *suffix
= silly
+ slen
- countersize
;
1313 sprintf(suffix
, "%*.*x", countersize
, countersize
, sillycounter
);
1315 dfprintk(VFS
, "NFS: trying to rename %s to %s\n",
1316 dentry
->d_name
.name
, silly
);
1318 sdentry
= lookup_one_len(silly
, dentry
->d_parent
, slen
);
1320 * N.B. Better to return EBUSY here ... it could be
1321 * dangerous to delete the file while it's in use.
1323 if (IS_ERR(sdentry
))
1325 } while(sdentry
->d_inode
!= NULL
); /* need negative lookup */
1327 qsilly
.name
= silly
;
1328 qsilly
.len
= strlen(silly
);
1329 if (dentry
->d_inode
) {
1330 error
= NFS_PROTO(dir
)->rename(dir
, &dentry
->d_name
,
1332 nfs_mark_for_revalidate(dentry
->d_inode
);
1334 error
= NFS_PROTO(dir
)->rename(dir
, &dentry
->d_name
,
1337 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1338 d_move(dentry
, sdentry
);
1339 error
= nfs_async_unlink(dir
, dentry
);
1340 /* If we return 0 we don't unlink */
1348 * Remove a file after making sure there are no pending writes,
1349 * and after checking that the file has only one user.
1351 * We invalidate the attribute cache and free the inode prior to the operation
1352 * to avoid possible races if the server reuses the inode.
1354 static int nfs_safe_remove(struct dentry
*dentry
)
1356 struct inode
*dir
= dentry
->d_parent
->d_inode
;
1357 struct inode
*inode
= dentry
->d_inode
;
1360 dfprintk(VFS
, "NFS: safe_remove(%s/%s)\n",
1361 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
);
1363 /* If the dentry was sillyrenamed, we simply call d_delete() */
1364 if (dentry
->d_flags
& DCACHE_NFSFS_RENAMED
) {
1369 if (inode
!= NULL
) {
1370 nfs_inode_return_delegation(inode
);
1371 error
= NFS_PROTO(dir
)->remove(dir
, &dentry
->d_name
);
1372 /* The VFS may want to delete this inode */
1375 nfs_mark_for_revalidate(inode
);
1377 error
= NFS_PROTO(dir
)->remove(dir
, &dentry
->d_name
);
1382 /* We do silly rename. In case sillyrename() returns -EBUSY, the inode
1383 * belongs to an active ".nfs..." file and we return -EBUSY.
1385 * If sillyrename() returns 0, we do nothing, otherwise we unlink.
1387 static int nfs_unlink(struct inode
*dir
, struct dentry
*dentry
)
1390 int need_rehash
= 0;
1392 dfprintk(VFS
, "NFS: unlink(%s/%ld, %s)\n", dir
->i_sb
->s_id
,
1393 dir
->i_ino
, dentry
->d_name
.name
);
1396 spin_lock(&dcache_lock
);
1397 spin_lock(&dentry
->d_lock
);
1398 if (atomic_read(&dentry
->d_count
) > 1) {
1399 spin_unlock(&dentry
->d_lock
);
1400 spin_unlock(&dcache_lock
);
1401 /* Start asynchronous writeout of the inode */
1402 write_inode_now(dentry
->d_inode
, 0);
1403 error
= nfs_sillyrename(dir
, dentry
);
1407 if (!d_unhashed(dentry
)) {
1411 spin_unlock(&dentry
->d_lock
);
1412 spin_unlock(&dcache_lock
);
1413 error
= nfs_safe_remove(dentry
);
1415 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1416 } else if (need_rehash
)
1423 * To create a symbolic link, most file systems instantiate a new inode,
1424 * add a page to it containing the path, then write it out to the disk
1425 * using prepare_write/commit_write.
1427 * Unfortunately the NFS client can't create the in-core inode first
1428 * because it needs a file handle to create an in-core inode (see
1429 * fs/nfs/inode.c:nfs_fhget). We only have a file handle *after* the
1430 * symlink request has completed on the server.
1432 * So instead we allocate a raw page, copy the symname into it, then do
1433 * the SYMLINK request with the page as the buffer. If it succeeds, we
1434 * now have a new file handle and can instantiate an in-core NFS inode
1435 * and move the raw page into its mapping.
1437 static int nfs_symlink(struct inode
*dir
, struct dentry
*dentry
, const char *symname
)
1439 struct pagevec lru_pvec
;
1443 unsigned int pathlen
= strlen(symname
);
1446 dfprintk(VFS
, "NFS: symlink(%s/%ld, %s, %s)\n", dir
->i_sb
->s_id
,
1447 dir
->i_ino
, dentry
->d_name
.name
, symname
);
1449 if (pathlen
> PAGE_SIZE
)
1450 return -ENAMETOOLONG
;
1452 attr
.ia_mode
= S_IFLNK
| S_IRWXUGO
;
1453 attr
.ia_valid
= ATTR_MODE
;
1457 page
= alloc_page(GFP_HIGHUSER
);
1463 kaddr
= kmap_atomic(page
, KM_USER0
);
1464 memcpy(kaddr
, symname
, pathlen
);
1465 if (pathlen
< PAGE_SIZE
)
1466 memset(kaddr
+ pathlen
, 0, PAGE_SIZE
- pathlen
);
1467 kunmap_atomic(kaddr
, KM_USER0
);
1469 error
= NFS_PROTO(dir
)->symlink(dir
, dentry
, page
, pathlen
, &attr
);
1471 dfprintk(VFS
, "NFS: symlink(%s/%ld, %s, %s) error %d\n",
1472 dir
->i_sb
->s_id
, dir
->i_ino
,
1473 dentry
->d_name
.name
, symname
, error
);
1481 * No big deal if we can't add this page to the page cache here.
1482 * READLINK will get the missing page from the server if needed.
1484 pagevec_init(&lru_pvec
, 0);
1485 if (!add_to_page_cache(page
, dentry
->d_inode
->i_mapping
, 0,
1487 pagevec_add(&lru_pvec
, page
);
1488 pagevec_lru_add(&lru_pvec
);
1489 SetPageUptodate(page
);
1499 nfs_link(struct dentry
*old_dentry
, struct inode
*dir
, struct dentry
*dentry
)
1501 struct inode
*inode
= old_dentry
->d_inode
;
1504 dfprintk(VFS
, "NFS: link(%s/%s -> %s/%s)\n",
1505 old_dentry
->d_parent
->d_name
.name
, old_dentry
->d_name
.name
,
1506 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
);
1509 error
= NFS_PROTO(dir
)->link(inode
, dir
, &dentry
->d_name
);
1511 atomic_inc(&inode
->i_count
);
1512 d_instantiate(dentry
, inode
);
1520 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a
1521 * different file handle for the same inode after a rename (e.g. when
1522 * moving to a different directory). A fail-safe method to do so would
1523 * be to look up old_dir/old_name, create a link to new_dir/new_name and
1524 * rename the old file using the sillyrename stuff. This way, the original
1525 * file in old_dir will go away when the last process iput()s the inode.
1529 * It actually works quite well. One needs to have the possibility for
1530 * at least one ".nfs..." file in each directory the file ever gets
1531 * moved or linked to which happens automagically with the new
1532 * implementation that only depends on the dcache stuff instead of
1533 * using the inode layer
1535 * Unfortunately, things are a little more complicated than indicated
1536 * above. For a cross-directory move, we want to make sure we can get
1537 * rid of the old inode after the operation. This means there must be
1538 * no pending writes (if it's a file), and the use count must be 1.
1539 * If these conditions are met, we can drop the dentries before doing
1542 static int nfs_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
1543 struct inode
*new_dir
, struct dentry
*new_dentry
)
1545 struct inode
*old_inode
= old_dentry
->d_inode
;
1546 struct inode
*new_inode
= new_dentry
->d_inode
;
1547 struct dentry
*dentry
= NULL
, *rehash
= NULL
;
1551 * To prevent any new references to the target during the rename,
1552 * we unhash the dentry and free the inode in advance.
1555 if (!d_unhashed(new_dentry
)) {
1557 rehash
= new_dentry
;
1560 dfprintk(VFS
, "NFS: rename(%s/%s -> %s/%s, ct=%d)\n",
1561 old_dentry
->d_parent
->d_name
.name
, old_dentry
->d_name
.name
,
1562 new_dentry
->d_parent
->d_name
.name
, new_dentry
->d_name
.name
,
1563 atomic_read(&new_dentry
->d_count
));
1566 * First check whether the target is busy ... we can't
1567 * safely do _any_ rename if the target is in use.
1569 * For files, make a copy of the dentry and then do a
1570 * silly-rename. If the silly-rename succeeds, the
1571 * copied dentry is hashed and becomes the new target.
1575 if (S_ISDIR(new_inode
->i_mode
)) {
1577 if (!S_ISDIR(old_inode
->i_mode
))
1579 } else if (atomic_read(&new_dentry
->d_count
) > 2) {
1581 /* copy the target dentry's name */
1582 dentry
= d_alloc(new_dentry
->d_parent
,
1583 &new_dentry
->d_name
);
1587 /* silly-rename the existing target ... */
1588 err
= nfs_sillyrename(new_dir
, new_dentry
);
1590 new_dentry
= rehash
= dentry
;
1592 /* instantiate the replacement target */
1593 d_instantiate(new_dentry
, NULL
);
1594 } else if (atomic_read(&new_dentry
->d_count
) > 1)
1595 /* dentry still busy? */
1598 drop_nlink(new_inode
);
1602 * ... prune child dentries and writebacks if needed.
1604 if (atomic_read(&old_dentry
->d_count
) > 1) {
1605 if (S_ISREG(old_inode
->i_mode
))
1606 nfs_wb_all(old_inode
);
1607 shrink_dcache_parent(old_dentry
);
1609 nfs_inode_return_delegation(old_inode
);
1611 if (new_inode
!= NULL
) {
1612 nfs_inode_return_delegation(new_inode
);
1613 d_delete(new_dentry
);
1616 error
= NFS_PROTO(old_dir
)->rename(old_dir
, &old_dentry
->d_name
,
1617 new_dir
, &new_dentry
->d_name
);
1618 nfs_mark_for_revalidate(old_inode
);
1623 d_move(old_dentry
, new_dentry
);
1624 nfs_set_verifier(new_dentry
,
1625 nfs_save_change_attribute(new_dir
));
1628 /* new dentry created? */
1635 static DEFINE_SPINLOCK(nfs_access_lru_lock
);
1636 static LIST_HEAD(nfs_access_lru_list
);
1637 static atomic_long_t nfs_access_nr_entries
;
1639 static void nfs_access_free_entry(struct nfs_access_entry
*entry
)
1641 put_rpccred(entry
->cred
);
1643 smp_mb__before_atomic_dec();
1644 atomic_long_dec(&nfs_access_nr_entries
);
1645 smp_mb__after_atomic_dec();
1648 int nfs_access_cache_shrinker(int nr_to_scan
, gfp_t gfp_mask
)
1651 struct nfs_inode
*nfsi
;
1652 struct nfs_access_entry
*cache
;
1655 spin_lock(&nfs_access_lru_lock
);
1656 list_for_each_entry(nfsi
, &nfs_access_lru_list
, access_cache_inode_lru
) {
1657 struct inode
*inode
;
1659 if (nr_to_scan
-- == 0)
1661 inode
= igrab(&nfsi
->vfs_inode
);
1664 spin_lock(&inode
->i_lock
);
1665 if (list_empty(&nfsi
->access_cache_entry_lru
))
1666 goto remove_lru_entry
;
1667 cache
= list_entry(nfsi
->access_cache_entry_lru
.next
,
1668 struct nfs_access_entry
, lru
);
1669 list_move(&cache
->lru
, &head
);
1670 rb_erase(&cache
->rb_node
, &nfsi
->access_cache
);
1671 if (!list_empty(&nfsi
->access_cache_entry_lru
))
1672 list_move_tail(&nfsi
->access_cache_inode_lru
,
1673 &nfs_access_lru_list
);
1676 list_del_init(&nfsi
->access_cache_inode_lru
);
1677 clear_bit(NFS_INO_ACL_LRU_SET
, &nfsi
->flags
);
1679 spin_unlock(&inode
->i_lock
);
1680 spin_unlock(&nfs_access_lru_lock
);
1684 spin_unlock(&nfs_access_lru_lock
);
1685 while (!list_empty(&head
)) {
1686 cache
= list_entry(head
.next
, struct nfs_access_entry
, lru
);
1687 list_del(&cache
->lru
);
1688 nfs_access_free_entry(cache
);
1690 return (atomic_long_read(&nfs_access_nr_entries
) / 100) * sysctl_vfs_cache_pressure
;
1693 static void __nfs_access_zap_cache(struct inode
*inode
)
1695 struct nfs_inode
*nfsi
= NFS_I(inode
);
1696 struct rb_root
*root_node
= &nfsi
->access_cache
;
1697 struct rb_node
*n
, *dispose
= NULL
;
1698 struct nfs_access_entry
*entry
;
1700 /* Unhook entries from the cache */
1701 while ((n
= rb_first(root_node
)) != NULL
) {
1702 entry
= rb_entry(n
, struct nfs_access_entry
, rb_node
);
1703 rb_erase(n
, root_node
);
1704 list_del(&entry
->lru
);
1705 n
->rb_left
= dispose
;
1708 nfsi
->cache_validity
&= ~NFS_INO_INVALID_ACCESS
;
1709 spin_unlock(&inode
->i_lock
);
1711 /* Now kill them all! */
1712 while (dispose
!= NULL
) {
1714 dispose
= n
->rb_left
;
1715 nfs_access_free_entry(rb_entry(n
, struct nfs_access_entry
, rb_node
));
1719 void nfs_access_zap_cache(struct inode
*inode
)
1721 /* Remove from global LRU init */
1722 if (test_and_clear_bit(NFS_INO_ACL_LRU_SET
, &NFS_FLAGS(inode
))) {
1723 spin_lock(&nfs_access_lru_lock
);
1724 list_del_init(&NFS_I(inode
)->access_cache_inode_lru
);
1725 spin_unlock(&nfs_access_lru_lock
);
1728 spin_lock(&inode
->i_lock
);
1729 /* This will release the spinlock */
1730 __nfs_access_zap_cache(inode
);
1733 static struct nfs_access_entry
*nfs_access_search_rbtree(struct inode
*inode
, struct rpc_cred
*cred
)
1735 struct rb_node
*n
= NFS_I(inode
)->access_cache
.rb_node
;
1736 struct nfs_access_entry
*entry
;
1739 entry
= rb_entry(n
, struct nfs_access_entry
, rb_node
);
1741 if (cred
< entry
->cred
)
1743 else if (cred
> entry
->cred
)
1751 static int nfs_access_get_cached(struct inode
*inode
, struct rpc_cred
*cred
, struct nfs_access_entry
*res
)
1753 struct nfs_inode
*nfsi
= NFS_I(inode
);
1754 struct nfs_access_entry
*cache
;
1757 spin_lock(&inode
->i_lock
);
1758 if (nfsi
->cache_validity
& NFS_INO_INVALID_ACCESS
)
1760 cache
= nfs_access_search_rbtree(inode
, cred
);
1763 if (!time_in_range(jiffies
, cache
->jiffies
, cache
->jiffies
+ NFS_ATTRTIMEO(inode
)))
1765 res
->jiffies
= cache
->jiffies
;
1766 res
->cred
= cache
->cred
;
1767 res
->mask
= cache
->mask
;
1768 list_move_tail(&cache
->lru
, &nfsi
->access_cache_entry_lru
);
1771 spin_unlock(&inode
->i_lock
);
1774 rb_erase(&cache
->rb_node
, &nfsi
->access_cache
);
1775 list_del(&cache
->lru
);
1776 spin_unlock(&inode
->i_lock
);
1777 nfs_access_free_entry(cache
);
1780 /* This will release the spinlock */
1781 __nfs_access_zap_cache(inode
);
1785 static void nfs_access_add_rbtree(struct inode
*inode
, struct nfs_access_entry
*set
)
1787 struct nfs_inode
*nfsi
= NFS_I(inode
);
1788 struct rb_root
*root_node
= &nfsi
->access_cache
;
1789 struct rb_node
**p
= &root_node
->rb_node
;
1790 struct rb_node
*parent
= NULL
;
1791 struct nfs_access_entry
*entry
;
1793 spin_lock(&inode
->i_lock
);
1794 while (*p
!= NULL
) {
1796 entry
= rb_entry(parent
, struct nfs_access_entry
, rb_node
);
1798 if (set
->cred
< entry
->cred
)
1799 p
= &parent
->rb_left
;
1800 else if (set
->cred
> entry
->cred
)
1801 p
= &parent
->rb_right
;
1805 rb_link_node(&set
->rb_node
, parent
, p
);
1806 rb_insert_color(&set
->rb_node
, root_node
);
1807 list_add_tail(&set
->lru
, &nfsi
->access_cache_entry_lru
);
1808 spin_unlock(&inode
->i_lock
);
1811 rb_replace_node(parent
, &set
->rb_node
, root_node
);
1812 list_add_tail(&set
->lru
, &nfsi
->access_cache_entry_lru
);
1813 list_del(&entry
->lru
);
1814 spin_unlock(&inode
->i_lock
);
1815 nfs_access_free_entry(entry
);
1818 static void nfs_access_add_cache(struct inode
*inode
, struct nfs_access_entry
*set
)
1820 struct nfs_access_entry
*cache
= kmalloc(sizeof(*cache
), GFP_KERNEL
);
1823 RB_CLEAR_NODE(&cache
->rb_node
);
1824 cache
->jiffies
= set
->jiffies
;
1825 cache
->cred
= get_rpccred(set
->cred
);
1826 cache
->mask
= set
->mask
;
1828 nfs_access_add_rbtree(inode
, cache
);
1830 /* Update accounting */
1831 smp_mb__before_atomic_inc();
1832 atomic_long_inc(&nfs_access_nr_entries
);
1833 smp_mb__after_atomic_inc();
1835 /* Add inode to global LRU list */
1836 if (!test_and_set_bit(NFS_INO_ACL_LRU_SET
, &NFS_FLAGS(inode
))) {
1837 spin_lock(&nfs_access_lru_lock
);
1838 list_add_tail(&NFS_I(inode
)->access_cache_inode_lru
, &nfs_access_lru_list
);
1839 spin_unlock(&nfs_access_lru_lock
);
1843 static int nfs_do_access(struct inode
*inode
, struct rpc_cred
*cred
, int mask
)
1845 struct nfs_access_entry cache
;
1848 status
= nfs_access_get_cached(inode
, cred
, &cache
);
1852 /* Be clever: ask server to check for all possible rights */
1853 cache
.mask
= MAY_EXEC
| MAY_WRITE
| MAY_READ
;
1855 cache
.jiffies
= jiffies
;
1856 status
= NFS_PROTO(inode
)->access(inode
, &cache
);
1859 nfs_access_add_cache(inode
, &cache
);
1861 if ((cache
.mask
& mask
) == mask
)
1866 static int nfs_open_permission_mask(int openflags
)
1870 if (openflags
& FMODE_READ
)
1872 if (openflags
& FMODE_WRITE
)
1874 if (openflags
& FMODE_EXEC
)
1879 int nfs_may_open(struct inode
*inode
, struct rpc_cred
*cred
, int openflags
)
1881 return nfs_do_access(inode
, cred
, nfs_open_permission_mask(openflags
));
1884 int nfs_permission(struct inode
*inode
, int mask
, struct nameidata
*nd
)
1886 struct rpc_cred
*cred
;
1889 nfs_inc_stats(inode
, NFSIOS_VFSACCESS
);
1893 /* Is this sys_access() ? */
1894 if (nd
!= NULL
&& (nd
->flags
& LOOKUP_ACCESS
))
1897 switch (inode
->i_mode
& S_IFMT
) {
1901 /* NFSv4 has atomic_open... */
1902 if (nfs_server_capable(inode
, NFS_CAP_ATOMIC_OPEN
)
1904 && (nd
->flags
& LOOKUP_OPEN
))
1909 * Optimize away all write operations, since the server
1910 * will check permissions when we perform the op.
1912 if ((mask
& MAY_WRITE
) && !(mask
& MAY_READ
))
1919 if (!NFS_PROTO(inode
)->access
)
1922 cred
= rpcauth_lookupcred(NFS_CLIENT(inode
)->cl_auth
, 0);
1923 if (!IS_ERR(cred
)) {
1924 res
= nfs_do_access(inode
, cred
, mask
);
1927 res
= PTR_ERR(cred
);
1930 dfprintk(VFS
, "NFS: permission(%s/%ld), mask=0x%x, res=%d\n",
1931 inode
->i_sb
->s_id
, inode
->i_ino
, mask
, res
);
1934 res
= nfs_revalidate_inode(NFS_SERVER(inode
), inode
);
1936 res
= generic_permission(inode
, mask
, NULL
);
1943 * version-control: t
1944 * kept-new-versions: 5