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
);
216 nfs_zap_caches(inode
);
222 int dir_decode(nfs_readdir_descriptor_t
*desc
)
224 __be32
*p
= desc
->ptr
;
225 p
= desc
->decode(p
, desc
->entry
, desc
->plus
);
229 if (desc
->timestamp_valid
)
230 desc
->entry
->fattr
->time_start
= desc
->timestamp
;
232 desc
->entry
->fattr
->valid
&= ~NFS_ATTR_FATTR
;
237 void dir_page_release(nfs_readdir_descriptor_t
*desc
)
240 page_cache_release(desc
->page
);
246 * Given a pointer to a buffer that has already been filled by a call
247 * to readdir, find the next entry with cookie '*desc->dir_cookie'.
249 * If the end of the buffer has been reached, return -EAGAIN, if not,
250 * return the offset within the buffer of the next entry to be
254 int find_dirent(nfs_readdir_descriptor_t
*desc
)
256 struct nfs_entry
*entry
= desc
->entry
;
260 while((status
= dir_decode(desc
)) == 0) {
261 dfprintk(DIRCACHE
, "NFS: %s: examining cookie %Lu\n",
262 __FUNCTION__
, (unsigned long long)entry
->cookie
);
263 if (entry
->prev_cookie
== *desc
->dir_cookie
)
265 if (loop_count
++ > 200) {
274 * Given a pointer to a buffer that has already been filled by a call
275 * to readdir, find the entry at offset 'desc->file->f_pos'.
277 * If the end of the buffer has been reached, return -EAGAIN, if not,
278 * return the offset within the buffer of the next entry to be
282 int find_dirent_index(nfs_readdir_descriptor_t
*desc
)
284 struct nfs_entry
*entry
= desc
->entry
;
289 status
= dir_decode(desc
);
293 dfprintk(DIRCACHE
, "NFS: found cookie %Lu at index %Ld\n",
294 (unsigned long long)entry
->cookie
, desc
->current_index
);
296 if (desc
->file
->f_pos
== desc
->current_index
) {
297 *desc
->dir_cookie
= entry
->cookie
;
300 desc
->current_index
++;
301 if (loop_count
++ > 200) {
310 * Find the given page, and call find_dirent() or find_dirent_index in
311 * order to try to return the next entry.
314 int find_dirent_page(nfs_readdir_descriptor_t
*desc
)
316 struct inode
*inode
= desc
->file
->f_path
.dentry
->d_inode
;
320 dfprintk(DIRCACHE
, "NFS: %s: searching page %ld for target %Lu\n",
321 __FUNCTION__
, desc
->page_index
,
322 (long long) *desc
->dir_cookie
);
324 /* If we find the page in the page_cache, we cannot be sure
325 * how fresh the data is, so we will ignore readdir_plus attributes.
327 desc
->timestamp_valid
= 0;
328 page
= read_cache_page(inode
->i_mapping
, desc
->page_index
,
329 (filler_t
*)nfs_readdir_filler
, desc
);
331 status
= PTR_ERR(page
);
335 /* NOTE: Someone else may have changed the READDIRPLUS flag */
337 desc
->ptr
= kmap(page
); /* matching kunmap in nfs_do_filldir */
338 if (*desc
->dir_cookie
!= 0)
339 status
= find_dirent(desc
);
341 status
= find_dirent_index(desc
);
343 dir_page_release(desc
);
345 dfprintk(DIRCACHE
, "NFS: %s: returns %d\n", __FUNCTION__
, status
);
350 * Recurse through the page cache pages, and return a
351 * filled nfs_entry structure of the next directory entry if possible.
353 * The target for the search is '*desc->dir_cookie' if non-0,
354 * 'desc->file->f_pos' otherwise
357 int readdir_search_pagecache(nfs_readdir_descriptor_t
*desc
)
362 /* Always search-by-index from the beginning of the cache */
363 if (*desc
->dir_cookie
== 0) {
364 dfprintk(DIRCACHE
, "NFS: readdir_search_pagecache() searching for offset %Ld\n",
365 (long long)desc
->file
->f_pos
);
366 desc
->page_index
= 0;
367 desc
->entry
->cookie
= desc
->entry
->prev_cookie
= 0;
368 desc
->entry
->eof
= 0;
369 desc
->current_index
= 0;
371 dfprintk(DIRCACHE
, "NFS: readdir_search_pagecache() searching for cookie %Lu\n",
372 (unsigned long long)*desc
->dir_cookie
);
375 res
= find_dirent_page(desc
);
378 /* Align to beginning of next page */
380 if (loop_count
++ > 200) {
386 dfprintk(DIRCACHE
, "NFS: %s: returns %d\n", __FUNCTION__
, res
);
390 static inline unsigned int dt_type(struct inode
*inode
)
392 return (inode
->i_mode
>> 12) & 15;
395 static struct dentry
*nfs_readdir_lookup(nfs_readdir_descriptor_t
*desc
);
398 * Once we've found the start of the dirent within a page: fill 'er up...
401 int nfs_do_filldir(nfs_readdir_descriptor_t
*desc
, void *dirent
,
404 struct file
*file
= desc
->file
;
405 struct nfs_entry
*entry
= desc
->entry
;
406 struct dentry
*dentry
= NULL
;
411 dfprintk(DIRCACHE
, "NFS: nfs_do_filldir() filling starting @ cookie %Lu\n",
412 (unsigned long long)entry
->cookie
);
415 unsigned d_type
= DT_UNKNOWN
;
416 /* Note: entry->prev_cookie contains the cookie for
417 * retrieving the current dirent on the server */
420 /* Get a dentry if we have one */
423 dentry
= nfs_readdir_lookup(desc
);
425 /* Use readdirplus info */
426 if (dentry
!= NULL
&& dentry
->d_inode
!= NULL
) {
427 d_type
= dt_type(dentry
->d_inode
);
428 fileid
= NFS_FILEID(dentry
->d_inode
);
431 res
= filldir(dirent
, entry
->name
, entry
->len
,
432 file
->f_pos
, fileid
, d_type
);
436 *desc
->dir_cookie
= entry
->cookie
;
437 if (dir_decode(desc
) != 0) {
441 if (loop_count
++ > 200) {
446 dir_page_release(desc
);
449 dfprintk(DIRCACHE
, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n",
450 (unsigned long long)*desc
->dir_cookie
, res
);
455 * If we cannot find a cookie in our cache, we suspect that this is
456 * because it points to a deleted file, so we ask the server to return
457 * whatever it thinks is the next entry. We then feed this to filldir.
458 * If all goes well, we should then be able to find our way round the
459 * cache on the next call to readdir_search_pagecache();
461 * NOTE: we cannot add the anonymous page to the pagecache because
462 * the data it contains might not be page aligned. Besides,
463 * we should already have a complete representation of the
464 * directory in the page cache by the time we get here.
467 int uncached_readdir(nfs_readdir_descriptor_t
*desc
, void *dirent
,
470 struct file
*file
= desc
->file
;
471 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
472 struct rpc_cred
*cred
= nfs_file_cred(file
);
473 struct page
*page
= NULL
;
475 unsigned long timestamp
;
477 dfprintk(DIRCACHE
, "NFS: uncached_readdir() searching for cookie %Lu\n",
478 (unsigned long long)*desc
->dir_cookie
);
480 page
= alloc_page(GFP_HIGHUSER
);
486 desc
->error
= NFS_PROTO(inode
)->readdir(file
->f_path
.dentry
, cred
, *desc
->dir_cookie
,
488 NFS_SERVER(inode
)->dtsize
,
491 desc
->ptr
= kmap(page
); /* matching kunmap in nfs_do_filldir */
492 if (desc
->error
>= 0) {
493 desc
->timestamp
= timestamp
;
494 desc
->timestamp_valid
= 1;
495 if ((status
= dir_decode(desc
)) == 0)
496 desc
->entry
->prev_cookie
= *desc
->dir_cookie
;
502 status
= nfs_do_filldir(desc
, dirent
, filldir
);
504 /* Reset read descriptor so it searches the page cache from
505 * the start upon the next call to readdir_search_pagecache() */
506 desc
->page_index
= 0;
507 desc
->entry
->cookie
= desc
->entry
->prev_cookie
= 0;
508 desc
->entry
->eof
= 0;
510 dfprintk(DIRCACHE
, "NFS: %s: returns %d\n",
511 __FUNCTION__
, status
);
514 dir_page_release(desc
);
518 /* The file offset position represents the dirent entry number. A
519 last cookie cache takes care of the common case of reading the
522 static int nfs_readdir(struct file
*filp
, void *dirent
, filldir_t filldir
)
524 struct dentry
*dentry
= filp
->f_path
.dentry
;
525 struct inode
*inode
= dentry
->d_inode
;
526 nfs_readdir_descriptor_t my_desc
,
528 struct nfs_entry my_entry
;
530 struct nfs_fattr fattr
;
533 dfprintk(VFS
, "NFS: readdir(%s/%s) starting at cookie %Lu\n",
534 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
,
535 (long long)filp
->f_pos
);
536 nfs_inc_stats(inode
, NFSIOS_VFSGETDENTS
);
540 res
= nfs_revalidate_mapping_nolock(inode
, filp
->f_mapping
);
547 * filp->f_pos points to the dirent entry number.
548 * *desc->dir_cookie has the cookie for the next entry. We have
549 * to either find the entry with the appropriate number or
550 * revalidate the cookie.
552 memset(desc
, 0, sizeof(*desc
));
555 desc
->dir_cookie
= &nfs_file_open_context(filp
)->dir_cookie
;
556 desc
->decode
= NFS_PROTO(inode
)->decode_dirent
;
557 desc
->plus
= NFS_USE_READDIRPLUS(inode
);
559 my_entry
.cookie
= my_entry
.prev_cookie
= 0;
562 my_entry
.fattr
= &fattr
;
563 nfs_fattr_init(&fattr
);
564 desc
->entry
= &my_entry
;
566 while(!desc
->entry
->eof
) {
567 res
= readdir_search_pagecache(desc
);
569 if (res
== -EBADCOOKIE
) {
570 /* This means either end of directory */
571 if (*desc
->dir_cookie
&& desc
->entry
->cookie
!= *desc
->dir_cookie
) {
572 /* Or that the server has 'lost' a cookie */
573 res
= uncached_readdir(desc
, dirent
, filldir
);
580 if (res
== -ETOOSMALL
&& desc
->plus
) {
581 clear_bit(NFS_INO_ADVISE_RDPLUS
, &NFS_FLAGS(inode
));
582 nfs_zap_caches(inode
);
584 desc
->entry
->eof
= 0;
590 res
= nfs_do_filldir(desc
, dirent
, filldir
);
599 dfprintk(VFS
, "NFS: readdir(%s/%s) returns %ld\n",
600 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
,
605 static loff_t
nfs_llseek_dir(struct file
*filp
, loff_t offset
, int origin
)
607 mutex_lock(&filp
->f_path
.dentry
->d_inode
->i_mutex
);
610 offset
+= filp
->f_pos
;
618 if (offset
!= filp
->f_pos
) {
619 filp
->f_pos
= offset
;
620 nfs_file_open_context(filp
)->dir_cookie
= 0;
623 mutex_unlock(&filp
->f_path
.dentry
->d_inode
->i_mutex
);
628 * All directory operations under NFS are synchronous, so fsync()
629 * is a dummy operation.
631 static int nfs_fsync_dir(struct file
*filp
, struct dentry
*dentry
, int datasync
)
633 dfprintk(VFS
, "NFS: fsync_dir(%s/%s) datasync %d\n",
634 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
,
641 * A check for whether or not the parent directory has changed.
642 * In the case it has, we assume that the dentries are untrustworthy
643 * and may need to be looked up again.
645 static int nfs_check_verifier(struct inode
*dir
, struct dentry
*dentry
)
649 if (!nfs_verify_change_attribute(dir
, dentry
->d_time
))
651 /* Revalidate nfsi->cache_change_attribute before we declare a match */
652 if (nfs_revalidate_inode(NFS_SERVER(dir
), dir
) < 0)
654 if (!nfs_verify_change_attribute(dir
, dentry
->d_time
))
659 static inline void nfs_set_verifier(struct dentry
* dentry
, unsigned long verf
)
661 dentry
->d_time
= verf
;
665 * Return the intent data that applies to this particular path component
667 * Note that the current set of intents only apply to the very last
668 * component of the path.
669 * We check for this using LOOKUP_CONTINUE and LOOKUP_PARENT.
671 static inline unsigned int nfs_lookup_check_intent(struct nameidata
*nd
, unsigned int mask
)
673 if (nd
->flags
& (LOOKUP_CONTINUE
|LOOKUP_PARENT
))
675 return nd
->flags
& mask
;
679 * Inode and filehandle revalidation for lookups.
681 * We force revalidation in the cases where the VFS sets LOOKUP_REVAL,
682 * or if the intent information indicates that we're about to open this
683 * particular file and the "nocto" mount flag is not set.
687 int nfs_lookup_verify_inode(struct inode
*inode
, struct nameidata
*nd
)
689 struct nfs_server
*server
= NFS_SERVER(inode
);
692 /* VFS wants an on-the-wire revalidation */
693 if (nd
->flags
& LOOKUP_REVAL
)
695 /* This is an open(2) */
696 if (nfs_lookup_check_intent(nd
, LOOKUP_OPEN
) != 0 &&
697 !(server
->flags
& NFS_MOUNT_NOCTO
) &&
698 (S_ISREG(inode
->i_mode
) ||
699 S_ISDIR(inode
->i_mode
)))
702 return nfs_revalidate_inode(server
, inode
);
704 return __nfs_revalidate_inode(server
, inode
);
708 * We judge how long we want to trust negative
709 * dentries by looking at the parent inode mtime.
711 * If parent mtime has changed, we revalidate, else we wait for a
712 * period corresponding to the parent's attribute cache timeout value.
715 int nfs_neg_need_reval(struct inode
*dir
, struct dentry
*dentry
,
716 struct nameidata
*nd
)
718 /* Don't revalidate a negative dentry if we're creating a new file */
719 if (nd
!= NULL
&& nfs_lookup_check_intent(nd
, LOOKUP_CREATE
) != 0)
721 return !nfs_check_verifier(dir
, dentry
);
725 * This is called every time the dcache has a lookup hit,
726 * and we should check whether we can really trust that
729 * NOTE! The hit can be a negative hit too, don't assume
732 * If the parent directory is seen to have changed, we throw out the
733 * cached dentry and do a new lookup.
735 static int nfs_lookup_revalidate(struct dentry
* dentry
, struct nameidata
*nd
)
739 struct dentry
*parent
;
741 struct nfs_fh fhandle
;
742 struct nfs_fattr fattr
;
744 parent
= dget_parent(dentry
);
746 dir
= parent
->d_inode
;
747 nfs_inc_stats(dir
, NFSIOS_DENTRYREVALIDATE
);
748 inode
= dentry
->d_inode
;
751 if (nfs_neg_need_reval(dir
, dentry
, nd
))
756 if (is_bad_inode(inode
)) {
757 dfprintk(LOOKUPCACHE
, "%s: %s/%s has dud inode\n",
758 __FUNCTION__
, dentry
->d_parent
->d_name
.name
,
759 dentry
->d_name
.name
);
763 /* Force a full look up iff the parent directory has changed */
764 if (nfs_check_verifier(dir
, dentry
)) {
765 if (nfs_lookup_verify_inode(inode
, nd
))
770 if (NFS_STALE(inode
))
773 error
= NFS_PROTO(dir
)->lookup(dir
, &dentry
->d_name
, &fhandle
, &fattr
);
776 if (nfs_compare_fh(NFS_FH(inode
), &fhandle
))
778 if ((error
= nfs_refresh_inode(inode
, &fattr
)) != 0)
781 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
785 dfprintk(LOOKUPCACHE
, "NFS: %s(%s/%s) is valid\n",
786 __FUNCTION__
, dentry
->d_parent
->d_name
.name
,
787 dentry
->d_name
.name
);
792 nfs_mark_for_revalidate(dir
);
793 if (inode
&& S_ISDIR(inode
->i_mode
)) {
794 /* Purge readdir caches. */
795 nfs_zap_caches(inode
);
796 /* If we have submounts, don't unhash ! */
797 if (have_submounts(dentry
))
799 shrink_dcache_parent(dentry
);
804 dfprintk(LOOKUPCACHE
, "NFS: %s(%s/%s) is invalid\n",
805 __FUNCTION__
, dentry
->d_parent
->d_name
.name
,
806 dentry
->d_name
.name
);
811 * This is called from dput() when d_count is going to 0.
813 static int nfs_dentry_delete(struct dentry
*dentry
)
815 dfprintk(VFS
, "NFS: dentry_delete(%s/%s, %x)\n",
816 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
,
819 if (dentry
->d_flags
& DCACHE_NFSFS_RENAMED
) {
820 /* Unhash it, so that ->d_iput() would be called */
823 if (!(dentry
->d_sb
->s_flags
& MS_ACTIVE
)) {
824 /* Unhash it, so that ancestors of killed async unlink
825 * files will be cleaned up during umount */
833 * Called when the dentry loses inode.
834 * We use it to clean up silly-renamed files.
836 static void nfs_dentry_iput(struct dentry
*dentry
, struct inode
*inode
)
838 nfs_inode_return_delegation(inode
);
839 if (S_ISDIR(inode
->i_mode
))
840 /* drop any readdir cache as it could easily be old */
841 NFS_I(inode
)->cache_validity
|= NFS_INO_INVALID_DATA
;
843 if (dentry
->d_flags
& DCACHE_NFSFS_RENAMED
) {
846 nfs_complete_unlink(dentry
, inode
);
852 struct dentry_operations nfs_dentry_operations
= {
853 .d_revalidate
= nfs_lookup_revalidate
,
854 .d_delete
= nfs_dentry_delete
,
855 .d_iput
= nfs_dentry_iput
,
859 * Use intent information to check whether or not we're going to do
860 * an O_EXCL create using this path component.
863 int nfs_is_exclusive_create(struct inode
*dir
, struct nameidata
*nd
)
865 if (NFS_PROTO(dir
)->version
== 2)
867 if (nd
== NULL
|| nfs_lookup_check_intent(nd
, LOOKUP_CREATE
) == 0)
869 return (nd
->intent
.open
.flags
& O_EXCL
) != 0;
872 static inline int nfs_reval_fsid(struct inode
*dir
, const struct nfs_fattr
*fattr
)
874 struct nfs_server
*server
= NFS_SERVER(dir
);
876 if (!nfs_fsid_equal(&server
->fsid
, &fattr
->fsid
))
877 /* Revalidate fsid using the parent directory */
878 return __nfs_revalidate_inode(server
, dir
);
882 static struct dentry
*nfs_lookup(struct inode
*dir
, struct dentry
* dentry
, struct nameidata
*nd
)
885 struct inode
*inode
= NULL
;
887 struct nfs_fh fhandle
;
888 struct nfs_fattr fattr
;
890 dfprintk(VFS
, "NFS: lookup(%s/%s)\n",
891 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
);
892 nfs_inc_stats(dir
, NFSIOS_VFSLOOKUP
);
894 res
= ERR_PTR(-ENAMETOOLONG
);
895 if (dentry
->d_name
.len
> NFS_SERVER(dir
)->namelen
)
898 res
= ERR_PTR(-ENOMEM
);
899 dentry
->d_op
= NFS_PROTO(dir
)->dentry_ops
;
904 * If we're doing an exclusive create, optimize away the lookup
905 * but don't hash the dentry.
907 if (nfs_is_exclusive_create(dir
, nd
)) {
908 d_instantiate(dentry
, NULL
);
913 error
= NFS_PROTO(dir
)->lookup(dir
, &dentry
->d_name
, &fhandle
, &fattr
);
914 if (error
== -ENOENT
)
917 res
= ERR_PTR(error
);
920 error
= nfs_reval_fsid(dir
, &fattr
);
922 res
= ERR_PTR(error
);
925 inode
= nfs_fhget(dentry
->d_sb
, &fhandle
, &fattr
);
926 res
= (struct dentry
*)inode
;
931 res
= d_materialise_unique(dentry
, inode
);
937 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
945 static int nfs_open_revalidate(struct dentry
*, struct nameidata
*);
947 struct dentry_operations nfs4_dentry_operations
= {
948 .d_revalidate
= nfs_open_revalidate
,
949 .d_delete
= nfs_dentry_delete
,
950 .d_iput
= nfs_dentry_iput
,
954 * Use intent information to determine whether we need to substitute
955 * the NFSv4-style stateful OPEN for the LOOKUP call
957 static int is_atomic_open(struct inode
*dir
, struct nameidata
*nd
)
959 if (nd
== NULL
|| nfs_lookup_check_intent(nd
, LOOKUP_OPEN
) == 0)
961 /* NFS does not (yet) have a stateful open for directories */
962 if (nd
->flags
& LOOKUP_DIRECTORY
)
964 /* Are we trying to write to a read only partition? */
965 if (IS_RDONLY(dir
) && (nd
->intent
.open
.flags
& (O_CREAT
|O_TRUNC
|FMODE_WRITE
)))
970 static struct dentry
*nfs_atomic_lookup(struct inode
*dir
, struct dentry
*dentry
, struct nameidata
*nd
)
972 struct dentry
*res
= NULL
;
975 dfprintk(VFS
, "NFS: atomic_lookup(%s/%ld), %s\n",
976 dir
->i_sb
->s_id
, dir
->i_ino
, dentry
->d_name
.name
);
978 /* Check that we are indeed trying to open this file */
979 if (!is_atomic_open(dir
, nd
))
982 if (dentry
->d_name
.len
> NFS_SERVER(dir
)->namelen
) {
983 res
= ERR_PTR(-ENAMETOOLONG
);
986 dentry
->d_op
= NFS_PROTO(dir
)->dentry_ops
;
988 /* Let vfs_create() deal with O_EXCL. Instantiate, but don't hash
990 if (nd
->intent
.open
.flags
& O_EXCL
) {
991 d_instantiate(dentry
, NULL
);
995 /* Open the file on the server */
997 /* Revalidate parent directory attribute cache */
998 error
= nfs_revalidate_inode(NFS_SERVER(dir
), dir
);
1000 res
= ERR_PTR(error
);
1005 res
= nfs4_atomic_open(dir
, dentry
, nd
);
1008 error
= PTR_ERR(res
);
1010 /* Make a negative dentry */
1014 /* This turned out not to be a regular file */
1019 if (!(nd
->intent
.open
.flags
& O_NOFOLLOW
))
1025 } else if (res
!= NULL
)
1027 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1031 return nfs_lookup(dir
, dentry
, nd
);
1034 static int nfs_open_revalidate(struct dentry
*dentry
, struct nameidata
*nd
)
1036 struct dentry
*parent
= NULL
;
1037 struct inode
*inode
= dentry
->d_inode
;
1039 int openflags
, ret
= 0;
1041 parent
= dget_parent(dentry
);
1042 dir
= parent
->d_inode
;
1043 if (!is_atomic_open(dir
, nd
))
1045 /* We can't create new files in nfs_open_revalidate(), so we
1046 * optimize away revalidation of negative dentries.
1050 /* NFS only supports OPEN on regular files */
1051 if (!S_ISREG(inode
->i_mode
))
1053 openflags
= nd
->intent
.open
.flags
;
1054 /* We cannot do exclusive creation on a positive dentry */
1055 if ((openflags
& (O_CREAT
|O_EXCL
)) == (O_CREAT
|O_EXCL
))
1057 /* We can't create new files, or truncate existing ones here */
1058 openflags
&= ~(O_CREAT
|O_TRUNC
);
1061 * Note: we're not holding inode->i_mutex and so may be racing with
1062 * operations that change the directory. We therefore save the
1063 * change attribute *before* we do the RPC call.
1066 ret
= nfs4_open_revalidate(dir
, dentry
, openflags
, nd
);
1068 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1077 if (inode
!= NULL
&& nfs_have_delegation(inode
, FMODE_READ
))
1079 return nfs_lookup_revalidate(dentry
, nd
);
1081 #endif /* CONFIG_NFSV4 */
1083 static struct dentry
*nfs_readdir_lookup(nfs_readdir_descriptor_t
*desc
)
1085 struct dentry
*parent
= desc
->file
->f_path
.dentry
;
1086 struct inode
*dir
= parent
->d_inode
;
1087 struct nfs_entry
*entry
= desc
->entry
;
1088 struct dentry
*dentry
, *alias
;
1089 struct qstr name
= {
1090 .name
= entry
->name
,
1093 struct inode
*inode
;
1094 unsigned long verf
= nfs_save_change_attribute(dir
);
1098 if (name
.name
[0] == '.' && name
.name
[1] == '.')
1099 return dget_parent(parent
);
1102 if (name
.name
[0] == '.')
1103 return dget(parent
);
1106 spin_lock(&dir
->i_lock
);
1107 if (NFS_I(dir
)->cache_validity
& NFS_INO_INVALID_DATA
) {
1108 spin_unlock(&dir
->i_lock
);
1111 spin_unlock(&dir
->i_lock
);
1113 name
.hash
= full_name_hash(name
.name
, name
.len
);
1114 dentry
= d_lookup(parent
, &name
);
1115 if (dentry
!= NULL
) {
1116 /* Is this a positive dentry that matches the readdir info? */
1117 if (dentry
->d_inode
!= NULL
&&
1118 (NFS_FILEID(dentry
->d_inode
) == entry
->ino
||
1119 d_mountpoint(dentry
))) {
1120 if (!desc
->plus
|| entry
->fh
->size
== 0)
1122 if (nfs_compare_fh(NFS_FH(dentry
->d_inode
),
1126 /* No, so d_drop to allow one to be created */
1130 if (!desc
->plus
|| !(entry
->fattr
->valid
& NFS_ATTR_FATTR
))
1132 if (name
.len
> NFS_SERVER(dir
)->namelen
)
1134 /* Note: caller is already holding the dir->i_mutex! */
1135 dentry
= d_alloc(parent
, &name
);
1138 dentry
->d_op
= NFS_PROTO(dir
)->dentry_ops
;
1139 inode
= nfs_fhget(dentry
->d_sb
, entry
->fh
, entry
->fattr
);
1140 if (IS_ERR(inode
)) {
1145 alias
= d_materialise_unique(dentry
, inode
);
1146 if (alias
!= NULL
) {
1154 nfs_set_verifier(dentry
, verf
);
1159 * Code common to create, mkdir, and mknod.
1161 int nfs_instantiate(struct dentry
*dentry
, struct nfs_fh
*fhandle
,
1162 struct nfs_fattr
*fattr
)
1164 struct dentry
*parent
= dget_parent(dentry
);
1165 struct inode
*dir
= parent
->d_inode
;
1166 struct inode
*inode
;
1167 int error
= -EACCES
;
1171 /* We may have been initialized further down */
1172 if (dentry
->d_inode
)
1174 if (fhandle
->size
== 0) {
1175 error
= NFS_PROTO(dir
)->lookup(dir
, &dentry
->d_name
, fhandle
, fattr
);
1179 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1180 if (!(fattr
->valid
& NFS_ATTR_FATTR
)) {
1181 struct nfs_server
*server
= NFS_SB(dentry
->d_sb
);
1182 error
= server
->nfs_client
->rpc_ops
->getattr(server
, fhandle
, fattr
);
1186 inode
= nfs_fhget(dentry
->d_sb
, fhandle
, fattr
);
1187 error
= PTR_ERR(inode
);
1190 d_add(dentry
, inode
);
1195 nfs_mark_for_revalidate(dir
);
1201 * Following a failed create operation, we drop the dentry rather
1202 * than retain a negative dentry. This avoids a problem in the event
1203 * that the operation succeeded on the server, but an error in the
1204 * reply path made it appear to have failed.
1206 static int nfs_create(struct inode
*dir
, struct dentry
*dentry
, int mode
,
1207 struct nameidata
*nd
)
1213 dfprintk(VFS
, "NFS: create(%s/%ld), %s\n",
1214 dir
->i_sb
->s_id
, dir
->i_ino
, dentry
->d_name
.name
);
1216 attr
.ia_mode
= mode
;
1217 attr
.ia_valid
= ATTR_MODE
;
1219 if ((nd
->flags
& LOOKUP_CREATE
) != 0)
1220 open_flags
= nd
->intent
.open
.flags
;
1223 error
= NFS_PROTO(dir
)->create(dir
, dentry
, &attr
, open_flags
, nd
);
1235 * See comments for nfs_proc_create regarding failed operations.
1238 nfs_mknod(struct inode
*dir
, struct dentry
*dentry
, int mode
, dev_t rdev
)
1243 dfprintk(VFS
, "NFS: mknod(%s/%ld), %s\n",
1244 dir
->i_sb
->s_id
, dir
->i_ino
, dentry
->d_name
.name
);
1246 if (!new_valid_dev(rdev
))
1249 attr
.ia_mode
= mode
;
1250 attr
.ia_valid
= ATTR_MODE
;
1253 status
= NFS_PROTO(dir
)->mknod(dir
, dentry
, &attr
, rdev
);
1265 * See comments for nfs_proc_create regarding failed operations.
1267 static int nfs_mkdir(struct inode
*dir
, struct dentry
*dentry
, int mode
)
1272 dfprintk(VFS
, "NFS: mkdir(%s/%ld), %s\n",
1273 dir
->i_sb
->s_id
, dir
->i_ino
, dentry
->d_name
.name
);
1275 attr
.ia_valid
= ATTR_MODE
;
1276 attr
.ia_mode
= mode
| S_IFDIR
;
1279 error
= NFS_PROTO(dir
)->mkdir(dir
, dentry
, &attr
);
1290 static int nfs_rmdir(struct inode
*dir
, struct dentry
*dentry
)
1294 dfprintk(VFS
, "NFS: rmdir(%s/%ld), %s\n",
1295 dir
->i_sb
->s_id
, dir
->i_ino
, dentry
->d_name
.name
);
1298 error
= NFS_PROTO(dir
)->rmdir(dir
, &dentry
->d_name
);
1299 /* Ensure the VFS deletes this inode */
1300 if (error
== 0 && dentry
->d_inode
!= NULL
)
1301 clear_nlink(dentry
->d_inode
);
1307 static int nfs_sillyrename(struct inode
*dir
, struct dentry
*dentry
)
1309 static unsigned int sillycounter
;
1310 const int fileidsize
= sizeof(NFS_FILEID(dentry
->d_inode
))*2;
1311 const int countersize
= sizeof(sillycounter
)*2;
1312 const int slen
= sizeof(".nfs")+fileidsize
+countersize
-1;
1315 struct dentry
*sdentry
;
1318 dfprintk(VFS
, "NFS: silly-rename(%s/%s, ct=%d)\n",
1319 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
,
1320 atomic_read(&dentry
->d_count
));
1321 nfs_inc_stats(dir
, NFSIOS_SILLYRENAME
);
1324 * We don't allow a dentry to be silly-renamed twice.
1327 if (dentry
->d_flags
& DCACHE_NFSFS_RENAMED
)
1330 sprintf(silly
, ".nfs%*.*Lx",
1331 fileidsize
, fileidsize
,
1332 (unsigned long long)NFS_FILEID(dentry
->d_inode
));
1334 /* Return delegation in anticipation of the rename */
1335 nfs_inode_return_delegation(dentry
->d_inode
);
1339 char *suffix
= silly
+ slen
- countersize
;
1343 sprintf(suffix
, "%*.*x", countersize
, countersize
, sillycounter
);
1345 dfprintk(VFS
, "NFS: trying to rename %s to %s\n",
1346 dentry
->d_name
.name
, silly
);
1348 sdentry
= lookup_one_len(silly
, dentry
->d_parent
, slen
);
1350 * N.B. Better to return EBUSY here ... it could be
1351 * dangerous to delete the file while it's in use.
1353 if (IS_ERR(sdentry
))
1355 } while(sdentry
->d_inode
!= NULL
); /* need negative lookup */
1357 qsilly
.name
= silly
;
1358 qsilly
.len
= strlen(silly
);
1359 if (dentry
->d_inode
) {
1360 error
= NFS_PROTO(dir
)->rename(dir
, &dentry
->d_name
,
1362 nfs_mark_for_revalidate(dentry
->d_inode
);
1364 error
= NFS_PROTO(dir
)->rename(dir
, &dentry
->d_name
,
1367 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1368 d_move(dentry
, sdentry
);
1369 error
= nfs_async_unlink(dir
, dentry
);
1370 /* If we return 0 we don't unlink */
1378 * Remove a file after making sure there are no pending writes,
1379 * and after checking that the file has only one user.
1381 * We invalidate the attribute cache and free the inode prior to the operation
1382 * to avoid possible races if the server reuses the inode.
1384 static int nfs_safe_remove(struct dentry
*dentry
)
1386 struct inode
*dir
= dentry
->d_parent
->d_inode
;
1387 struct inode
*inode
= dentry
->d_inode
;
1390 dfprintk(VFS
, "NFS: safe_remove(%s/%s)\n",
1391 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
);
1393 /* If the dentry was sillyrenamed, we simply call d_delete() */
1394 if (dentry
->d_flags
& DCACHE_NFSFS_RENAMED
) {
1399 if (inode
!= NULL
) {
1400 nfs_inode_return_delegation(inode
);
1401 error
= NFS_PROTO(dir
)->remove(dir
, &dentry
->d_name
);
1402 /* The VFS may want to delete this inode */
1405 nfs_mark_for_revalidate(inode
);
1407 error
= NFS_PROTO(dir
)->remove(dir
, &dentry
->d_name
);
1412 /* We do silly rename. In case sillyrename() returns -EBUSY, the inode
1413 * belongs to an active ".nfs..." file and we return -EBUSY.
1415 * If sillyrename() returns 0, we do nothing, otherwise we unlink.
1417 static int nfs_unlink(struct inode
*dir
, struct dentry
*dentry
)
1420 int need_rehash
= 0;
1422 dfprintk(VFS
, "NFS: unlink(%s/%ld, %s)\n", dir
->i_sb
->s_id
,
1423 dir
->i_ino
, dentry
->d_name
.name
);
1426 spin_lock(&dcache_lock
);
1427 spin_lock(&dentry
->d_lock
);
1428 if (atomic_read(&dentry
->d_count
) > 1) {
1429 spin_unlock(&dentry
->d_lock
);
1430 spin_unlock(&dcache_lock
);
1431 /* Start asynchronous writeout of the inode */
1432 write_inode_now(dentry
->d_inode
, 0);
1433 error
= nfs_sillyrename(dir
, dentry
);
1437 if (!d_unhashed(dentry
)) {
1441 spin_unlock(&dentry
->d_lock
);
1442 spin_unlock(&dcache_lock
);
1443 error
= nfs_safe_remove(dentry
);
1445 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1446 } else if (need_rehash
)
1453 * To create a symbolic link, most file systems instantiate a new inode,
1454 * add a page to it containing the path, then write it out to the disk
1455 * using prepare_write/commit_write.
1457 * Unfortunately the NFS client can't create the in-core inode first
1458 * because it needs a file handle to create an in-core inode (see
1459 * fs/nfs/inode.c:nfs_fhget). We only have a file handle *after* the
1460 * symlink request has completed on the server.
1462 * So instead we allocate a raw page, copy the symname into it, then do
1463 * the SYMLINK request with the page as the buffer. If it succeeds, we
1464 * now have a new file handle and can instantiate an in-core NFS inode
1465 * and move the raw page into its mapping.
1467 static int nfs_symlink(struct inode
*dir
, struct dentry
*dentry
, const char *symname
)
1469 struct pagevec lru_pvec
;
1473 unsigned int pathlen
= strlen(symname
);
1476 dfprintk(VFS
, "NFS: symlink(%s/%ld, %s, %s)\n", dir
->i_sb
->s_id
,
1477 dir
->i_ino
, dentry
->d_name
.name
, symname
);
1479 if (pathlen
> PAGE_SIZE
)
1480 return -ENAMETOOLONG
;
1482 attr
.ia_mode
= S_IFLNK
| S_IRWXUGO
;
1483 attr
.ia_valid
= ATTR_MODE
;
1487 page
= alloc_page(GFP_HIGHUSER
);
1493 kaddr
= kmap_atomic(page
, KM_USER0
);
1494 memcpy(kaddr
, symname
, pathlen
);
1495 if (pathlen
< PAGE_SIZE
)
1496 memset(kaddr
+ pathlen
, 0, PAGE_SIZE
- pathlen
);
1497 kunmap_atomic(kaddr
, KM_USER0
);
1499 error
= NFS_PROTO(dir
)->symlink(dir
, dentry
, page
, pathlen
, &attr
);
1501 dfprintk(VFS
, "NFS: symlink(%s/%ld, %s, %s) error %d\n",
1502 dir
->i_sb
->s_id
, dir
->i_ino
,
1503 dentry
->d_name
.name
, symname
, error
);
1511 * No big deal if we can't add this page to the page cache here.
1512 * READLINK will get the missing page from the server if needed.
1514 pagevec_init(&lru_pvec
, 0);
1515 if (!add_to_page_cache(page
, dentry
->d_inode
->i_mapping
, 0,
1517 pagevec_add(&lru_pvec
, page
);
1518 pagevec_lru_add(&lru_pvec
);
1519 SetPageUptodate(page
);
1529 nfs_link(struct dentry
*old_dentry
, struct inode
*dir
, struct dentry
*dentry
)
1531 struct inode
*inode
= old_dentry
->d_inode
;
1534 dfprintk(VFS
, "NFS: link(%s/%s -> %s/%s)\n",
1535 old_dentry
->d_parent
->d_name
.name
, old_dentry
->d_name
.name
,
1536 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
);
1539 error
= NFS_PROTO(dir
)->link(inode
, dir
, &dentry
->d_name
);
1541 atomic_inc(&inode
->i_count
);
1542 d_instantiate(dentry
, inode
);
1550 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a
1551 * different file handle for the same inode after a rename (e.g. when
1552 * moving to a different directory). A fail-safe method to do so would
1553 * be to look up old_dir/old_name, create a link to new_dir/new_name and
1554 * rename the old file using the sillyrename stuff. This way, the original
1555 * file in old_dir will go away when the last process iput()s the inode.
1559 * It actually works quite well. One needs to have the possibility for
1560 * at least one ".nfs..." file in each directory the file ever gets
1561 * moved or linked to which happens automagically with the new
1562 * implementation that only depends on the dcache stuff instead of
1563 * using the inode layer
1565 * Unfortunately, things are a little more complicated than indicated
1566 * above. For a cross-directory move, we want to make sure we can get
1567 * rid of the old inode after the operation. This means there must be
1568 * no pending writes (if it's a file), and the use count must be 1.
1569 * If these conditions are met, we can drop the dentries before doing
1572 static int nfs_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
1573 struct inode
*new_dir
, struct dentry
*new_dentry
)
1575 struct inode
*old_inode
= old_dentry
->d_inode
;
1576 struct inode
*new_inode
= new_dentry
->d_inode
;
1577 struct dentry
*dentry
= NULL
, *rehash
= NULL
;
1581 * To prevent any new references to the target during the rename,
1582 * we unhash the dentry and free the inode in advance.
1585 if (!d_unhashed(new_dentry
)) {
1587 rehash
= new_dentry
;
1590 dfprintk(VFS
, "NFS: rename(%s/%s -> %s/%s, ct=%d)\n",
1591 old_dentry
->d_parent
->d_name
.name
, old_dentry
->d_name
.name
,
1592 new_dentry
->d_parent
->d_name
.name
, new_dentry
->d_name
.name
,
1593 atomic_read(&new_dentry
->d_count
));
1596 * First check whether the target is busy ... we can't
1597 * safely do _any_ rename if the target is in use.
1599 * For files, make a copy of the dentry and then do a
1600 * silly-rename. If the silly-rename succeeds, the
1601 * copied dentry is hashed and becomes the new target.
1605 if (S_ISDIR(new_inode
->i_mode
)) {
1607 if (!S_ISDIR(old_inode
->i_mode
))
1609 } else if (atomic_read(&new_dentry
->d_count
) > 2) {
1611 /* copy the target dentry's name */
1612 dentry
= d_alloc(new_dentry
->d_parent
,
1613 &new_dentry
->d_name
);
1617 /* silly-rename the existing target ... */
1618 err
= nfs_sillyrename(new_dir
, new_dentry
);
1620 new_dentry
= rehash
= dentry
;
1622 /* instantiate the replacement target */
1623 d_instantiate(new_dentry
, NULL
);
1624 } else if (atomic_read(&new_dentry
->d_count
) > 1)
1625 /* dentry still busy? */
1628 drop_nlink(new_inode
);
1632 * ... prune child dentries and writebacks if needed.
1634 if (atomic_read(&old_dentry
->d_count
) > 1) {
1635 if (S_ISREG(old_inode
->i_mode
))
1636 nfs_wb_all(old_inode
);
1637 shrink_dcache_parent(old_dentry
);
1639 nfs_inode_return_delegation(old_inode
);
1641 if (new_inode
!= NULL
) {
1642 nfs_inode_return_delegation(new_inode
);
1643 d_delete(new_dentry
);
1646 error
= NFS_PROTO(old_dir
)->rename(old_dir
, &old_dentry
->d_name
,
1647 new_dir
, &new_dentry
->d_name
);
1648 nfs_mark_for_revalidate(old_inode
);
1653 d_move(old_dentry
, new_dentry
);
1654 nfs_set_verifier(new_dentry
,
1655 nfs_save_change_attribute(new_dir
));
1658 /* new dentry created? */
1665 static DEFINE_SPINLOCK(nfs_access_lru_lock
);
1666 static LIST_HEAD(nfs_access_lru_list
);
1667 static atomic_long_t nfs_access_nr_entries
;
1669 static void nfs_access_free_entry(struct nfs_access_entry
*entry
)
1671 put_rpccred(entry
->cred
);
1673 smp_mb__before_atomic_dec();
1674 atomic_long_dec(&nfs_access_nr_entries
);
1675 smp_mb__after_atomic_dec();
1678 int nfs_access_cache_shrinker(int nr_to_scan
, gfp_t gfp_mask
)
1681 struct nfs_inode
*nfsi
;
1682 struct nfs_access_entry
*cache
;
1685 spin_lock(&nfs_access_lru_lock
);
1686 list_for_each_entry(nfsi
, &nfs_access_lru_list
, access_cache_inode_lru
) {
1687 struct inode
*inode
;
1689 if (nr_to_scan
-- == 0)
1691 inode
= igrab(&nfsi
->vfs_inode
);
1694 spin_lock(&inode
->i_lock
);
1695 if (list_empty(&nfsi
->access_cache_entry_lru
))
1696 goto remove_lru_entry
;
1697 cache
= list_entry(nfsi
->access_cache_entry_lru
.next
,
1698 struct nfs_access_entry
, lru
);
1699 list_move(&cache
->lru
, &head
);
1700 rb_erase(&cache
->rb_node
, &nfsi
->access_cache
);
1701 if (!list_empty(&nfsi
->access_cache_entry_lru
))
1702 list_move_tail(&nfsi
->access_cache_inode_lru
,
1703 &nfs_access_lru_list
);
1706 list_del_init(&nfsi
->access_cache_inode_lru
);
1707 clear_bit(NFS_INO_ACL_LRU_SET
, &nfsi
->flags
);
1709 spin_unlock(&inode
->i_lock
);
1710 spin_unlock(&nfs_access_lru_lock
);
1714 spin_unlock(&nfs_access_lru_lock
);
1715 while (!list_empty(&head
)) {
1716 cache
= list_entry(head
.next
, struct nfs_access_entry
, lru
);
1717 list_del(&cache
->lru
);
1718 nfs_access_free_entry(cache
);
1720 return (atomic_long_read(&nfs_access_nr_entries
) / 100) * sysctl_vfs_cache_pressure
;
1723 static void __nfs_access_zap_cache(struct inode
*inode
)
1725 struct nfs_inode
*nfsi
= NFS_I(inode
);
1726 struct rb_root
*root_node
= &nfsi
->access_cache
;
1727 struct rb_node
*n
, *dispose
= NULL
;
1728 struct nfs_access_entry
*entry
;
1730 /* Unhook entries from the cache */
1731 while ((n
= rb_first(root_node
)) != NULL
) {
1732 entry
= rb_entry(n
, struct nfs_access_entry
, rb_node
);
1733 rb_erase(n
, root_node
);
1734 list_del(&entry
->lru
);
1735 n
->rb_left
= dispose
;
1738 nfsi
->cache_validity
&= ~NFS_INO_INVALID_ACCESS
;
1739 spin_unlock(&inode
->i_lock
);
1741 /* Now kill them all! */
1742 while (dispose
!= NULL
) {
1744 dispose
= n
->rb_left
;
1745 nfs_access_free_entry(rb_entry(n
, struct nfs_access_entry
, rb_node
));
1749 void nfs_access_zap_cache(struct inode
*inode
)
1751 /* Remove from global LRU init */
1752 if (test_and_clear_bit(NFS_INO_ACL_LRU_SET
, &NFS_FLAGS(inode
))) {
1753 spin_lock(&nfs_access_lru_lock
);
1754 list_del_init(&NFS_I(inode
)->access_cache_inode_lru
);
1755 spin_unlock(&nfs_access_lru_lock
);
1758 spin_lock(&inode
->i_lock
);
1759 /* This will release the spinlock */
1760 __nfs_access_zap_cache(inode
);
1763 static struct nfs_access_entry
*nfs_access_search_rbtree(struct inode
*inode
, struct rpc_cred
*cred
)
1765 struct rb_node
*n
= NFS_I(inode
)->access_cache
.rb_node
;
1766 struct nfs_access_entry
*entry
;
1769 entry
= rb_entry(n
, struct nfs_access_entry
, rb_node
);
1771 if (cred
< entry
->cred
)
1773 else if (cred
> entry
->cred
)
1781 static int nfs_access_get_cached(struct inode
*inode
, struct rpc_cred
*cred
, struct nfs_access_entry
*res
)
1783 struct nfs_inode
*nfsi
= NFS_I(inode
);
1784 struct nfs_access_entry
*cache
;
1787 spin_lock(&inode
->i_lock
);
1788 if (nfsi
->cache_validity
& NFS_INO_INVALID_ACCESS
)
1790 cache
= nfs_access_search_rbtree(inode
, cred
);
1793 if (!time_in_range(jiffies
, cache
->jiffies
, cache
->jiffies
+ NFS_ATTRTIMEO(inode
)))
1795 res
->jiffies
= cache
->jiffies
;
1796 res
->cred
= cache
->cred
;
1797 res
->mask
= cache
->mask
;
1798 list_move_tail(&cache
->lru
, &nfsi
->access_cache_entry_lru
);
1801 spin_unlock(&inode
->i_lock
);
1804 rb_erase(&cache
->rb_node
, &nfsi
->access_cache
);
1805 list_del(&cache
->lru
);
1806 spin_unlock(&inode
->i_lock
);
1807 nfs_access_free_entry(cache
);
1810 /* This will release the spinlock */
1811 __nfs_access_zap_cache(inode
);
1815 static void nfs_access_add_rbtree(struct inode
*inode
, struct nfs_access_entry
*set
)
1817 struct nfs_inode
*nfsi
= NFS_I(inode
);
1818 struct rb_root
*root_node
= &nfsi
->access_cache
;
1819 struct rb_node
**p
= &root_node
->rb_node
;
1820 struct rb_node
*parent
= NULL
;
1821 struct nfs_access_entry
*entry
;
1823 spin_lock(&inode
->i_lock
);
1824 while (*p
!= NULL
) {
1826 entry
= rb_entry(parent
, struct nfs_access_entry
, rb_node
);
1828 if (set
->cred
< entry
->cred
)
1829 p
= &parent
->rb_left
;
1830 else if (set
->cred
> entry
->cred
)
1831 p
= &parent
->rb_right
;
1835 rb_link_node(&set
->rb_node
, parent
, p
);
1836 rb_insert_color(&set
->rb_node
, root_node
);
1837 list_add_tail(&set
->lru
, &nfsi
->access_cache_entry_lru
);
1838 spin_unlock(&inode
->i_lock
);
1841 rb_replace_node(parent
, &set
->rb_node
, root_node
);
1842 list_add_tail(&set
->lru
, &nfsi
->access_cache_entry_lru
);
1843 list_del(&entry
->lru
);
1844 spin_unlock(&inode
->i_lock
);
1845 nfs_access_free_entry(entry
);
1848 static void nfs_access_add_cache(struct inode
*inode
, struct nfs_access_entry
*set
)
1850 struct nfs_access_entry
*cache
= kmalloc(sizeof(*cache
), GFP_KERNEL
);
1853 RB_CLEAR_NODE(&cache
->rb_node
);
1854 cache
->jiffies
= set
->jiffies
;
1855 cache
->cred
= get_rpccred(set
->cred
);
1856 cache
->mask
= set
->mask
;
1858 nfs_access_add_rbtree(inode
, cache
);
1860 /* Update accounting */
1861 smp_mb__before_atomic_inc();
1862 atomic_long_inc(&nfs_access_nr_entries
);
1863 smp_mb__after_atomic_inc();
1865 /* Add inode to global LRU list */
1866 if (!test_and_set_bit(NFS_INO_ACL_LRU_SET
, &NFS_FLAGS(inode
))) {
1867 spin_lock(&nfs_access_lru_lock
);
1868 list_add_tail(&NFS_I(inode
)->access_cache_inode_lru
, &nfs_access_lru_list
);
1869 spin_unlock(&nfs_access_lru_lock
);
1873 static int nfs_do_access(struct inode
*inode
, struct rpc_cred
*cred
, int mask
)
1875 struct nfs_access_entry cache
;
1878 status
= nfs_access_get_cached(inode
, cred
, &cache
);
1882 /* Be clever: ask server to check for all possible rights */
1883 cache
.mask
= MAY_EXEC
| MAY_WRITE
| MAY_READ
;
1885 cache
.jiffies
= jiffies
;
1886 status
= NFS_PROTO(inode
)->access(inode
, &cache
);
1889 nfs_access_add_cache(inode
, &cache
);
1891 if ((cache
.mask
& mask
) == mask
)
1896 static int nfs_open_permission_mask(int openflags
)
1900 if (openflags
& FMODE_READ
)
1902 if (openflags
& FMODE_WRITE
)
1904 if (openflags
& FMODE_EXEC
)
1909 int nfs_may_open(struct inode
*inode
, struct rpc_cred
*cred
, int openflags
)
1911 return nfs_do_access(inode
, cred
, nfs_open_permission_mask(openflags
));
1914 int nfs_permission(struct inode
*inode
, int mask
, struct nameidata
*nd
)
1916 struct rpc_cred
*cred
;
1919 nfs_inc_stats(inode
, NFSIOS_VFSACCESS
);
1923 /* Is this sys_access() ? */
1924 if (nd
!= NULL
&& (nd
->flags
& LOOKUP_ACCESS
))
1927 switch (inode
->i_mode
& S_IFMT
) {
1931 /* NFSv4 has atomic_open... */
1932 if (nfs_server_capable(inode
, NFS_CAP_ATOMIC_OPEN
)
1934 && (nd
->flags
& LOOKUP_OPEN
))
1939 * Optimize away all write operations, since the server
1940 * will check permissions when we perform the op.
1942 if ((mask
& MAY_WRITE
) && !(mask
& MAY_READ
))
1949 if (!NFS_PROTO(inode
)->access
)
1952 cred
= rpcauth_lookupcred(NFS_CLIENT(inode
)->cl_auth
, 0);
1953 if (!IS_ERR(cred
)) {
1954 res
= nfs_do_access(inode
, cred
, mask
);
1957 res
= PTR_ERR(cred
);
1960 dfprintk(VFS
, "NFS: permission(%s/%ld), mask=0x%x, res=%d\n",
1961 inode
->i_sb
->s_id
, inode
->i_ino
, mask
, res
);
1964 res
= nfs_revalidate_inode(NFS_SERVER(inode
), inode
);
1966 res
= generic_permission(inode
, mask
, NULL
);
1973 * version-control: t
1974 * kept-new-versions: 5