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>
38 #include "delegation.h"
41 #define NFS_PARANOIA 1
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 struct inode_operations nfs_dir_inode_operations
= {
73 .symlink
= nfs_symlink
,
78 .permission
= nfs_permission
,
79 .getattr
= nfs_getattr
,
80 .setattr
= nfs_setattr
,
84 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 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 u32
* (*decode_dirent_t
)(u32
*, 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 } nfs_readdir_descriptor_t
;
159 /* Now we cache directories properly, by stuffing the dirent
160 * data directly in the page cache.
162 * Inode invalidation due to refresh etc. takes care of
163 * _everything_, no sloppy entry flushing logic, no extraneous
164 * copying, network direct to page cache, the way it was meant
167 * NOTE: Dirent information verification is done always by the
168 * page-in of the RPC reply, nowhere else, this simplies
169 * things substantially.
172 int nfs_readdir_filler(nfs_readdir_descriptor_t
*desc
, struct page
*page
)
174 struct file
*file
= desc
->file
;
175 struct inode
*inode
= file
->f_dentry
->d_inode
;
176 struct rpc_cred
*cred
= nfs_file_cred(file
);
177 unsigned long timestamp
;
180 dfprintk(DIRCACHE
, "NFS: %s: reading cookie %Lu into page %lu\n",
181 __FUNCTION__
, (long long)desc
->entry
->cookie
,
186 error
= NFS_PROTO(inode
)->readdir(file
->f_dentry
, cred
, desc
->entry
->cookie
, page
,
187 NFS_SERVER(inode
)->dtsize
, desc
->plus
);
189 /* We requested READDIRPLUS, but the server doesn't grok it */
190 if (error
== -ENOTSUPP
&& desc
->plus
) {
191 NFS_SERVER(inode
)->caps
&= ~NFS_CAP_READDIRPLUS
;
192 clear_bit(NFS_INO_ADVISE_RDPLUS
, &NFS_FLAGS(inode
));
198 SetPageUptodate(page
);
199 spin_lock(&inode
->i_lock
);
200 NFS_I(inode
)->cache_validity
|= NFS_INO_INVALID_ATIME
;
201 spin_unlock(&inode
->i_lock
);
202 /* Ensure consistent page alignment of the data.
203 * Note: assumes we have exclusive access to this mapping either
204 * through inode->i_mutex or some other mechanism.
206 if (page
->index
== 0)
207 invalidate_inode_pages2_range(inode
->i_mapping
, PAGE_CACHE_SIZE
, -1);
213 nfs_zap_caches(inode
);
219 int dir_decode(nfs_readdir_descriptor_t
*desc
)
222 p
= desc
->decode(p
, desc
->entry
, desc
->plus
);
230 void dir_page_release(nfs_readdir_descriptor_t
*desc
)
233 page_cache_release(desc
->page
);
239 * Given a pointer to a buffer that has already been filled by a call
240 * to readdir, find the next entry with cookie '*desc->dir_cookie'.
242 * If the end of the buffer has been reached, return -EAGAIN, if not,
243 * return the offset within the buffer of the next entry to be
247 int find_dirent(nfs_readdir_descriptor_t
*desc
)
249 struct nfs_entry
*entry
= desc
->entry
;
253 while((status
= dir_decode(desc
)) == 0) {
254 dfprintk(DIRCACHE
, "NFS: %s: examining cookie %Lu\n",
255 __FUNCTION__
, (unsigned long long)entry
->cookie
);
256 if (entry
->prev_cookie
== *desc
->dir_cookie
)
258 if (loop_count
++ > 200) {
267 * Given a pointer to a buffer that has already been filled by a call
268 * to readdir, find the entry at offset 'desc->file->f_pos'.
270 * If the end of the buffer has been reached, return -EAGAIN, if not,
271 * return the offset within the buffer of the next entry to be
275 int find_dirent_index(nfs_readdir_descriptor_t
*desc
)
277 struct nfs_entry
*entry
= desc
->entry
;
282 status
= dir_decode(desc
);
286 dfprintk(DIRCACHE
, "NFS: found cookie %Lu at index %Ld\n",
287 (unsigned long long)entry
->cookie
, desc
->current_index
);
289 if (desc
->file
->f_pos
== desc
->current_index
) {
290 *desc
->dir_cookie
= entry
->cookie
;
293 desc
->current_index
++;
294 if (loop_count
++ > 200) {
303 * Find the given page, and call find_dirent() or find_dirent_index in
304 * order to try to return the next entry.
307 int find_dirent_page(nfs_readdir_descriptor_t
*desc
)
309 struct inode
*inode
= desc
->file
->f_dentry
->d_inode
;
313 dfprintk(DIRCACHE
, "NFS: %s: searching page %ld for target %Lu\n",
314 __FUNCTION__
, desc
->page_index
,
315 (long long) *desc
->dir_cookie
);
317 page
= read_cache_page(inode
->i_mapping
, desc
->page_index
,
318 (filler_t
*)nfs_readdir_filler
, desc
);
320 status
= PTR_ERR(page
);
323 if (!PageUptodate(page
))
326 /* NOTE: Someone else may have changed the READDIRPLUS flag */
328 desc
->ptr
= kmap(page
); /* matching kunmap in nfs_do_filldir */
329 if (*desc
->dir_cookie
!= 0)
330 status
= find_dirent(desc
);
332 status
= find_dirent_index(desc
);
334 dir_page_release(desc
);
336 dfprintk(DIRCACHE
, "NFS: %s: returns %d\n", __FUNCTION__
, status
);
339 page_cache_release(page
);
344 * Recurse through the page cache pages, and return a
345 * filled nfs_entry structure of the next directory entry if possible.
347 * The target for the search is '*desc->dir_cookie' if non-0,
348 * 'desc->file->f_pos' otherwise
351 int readdir_search_pagecache(nfs_readdir_descriptor_t
*desc
)
356 /* Always search-by-index from the beginning of the cache */
357 if (*desc
->dir_cookie
== 0) {
358 dfprintk(DIRCACHE
, "NFS: readdir_search_pagecache() searching for offset %Ld\n",
359 (long long)desc
->file
->f_pos
);
360 desc
->page_index
= 0;
361 desc
->entry
->cookie
= desc
->entry
->prev_cookie
= 0;
362 desc
->entry
->eof
= 0;
363 desc
->current_index
= 0;
365 dfprintk(DIRCACHE
, "NFS: readdir_search_pagecache() searching for cookie %Lu\n",
366 (unsigned long long)*desc
->dir_cookie
);
369 res
= find_dirent_page(desc
);
372 /* Align to beginning of next page */
374 if (loop_count
++ > 200) {
380 dfprintk(DIRCACHE
, "NFS: %s: returns %d\n", __FUNCTION__
, res
);
384 static inline unsigned int dt_type(struct inode
*inode
)
386 return (inode
->i_mode
>> 12) & 15;
389 static struct dentry
*nfs_readdir_lookup(nfs_readdir_descriptor_t
*desc
);
392 * Once we've found the start of the dirent within a page: fill 'er up...
395 int nfs_do_filldir(nfs_readdir_descriptor_t
*desc
, void *dirent
,
398 struct file
*file
= desc
->file
;
399 struct nfs_entry
*entry
= desc
->entry
;
400 struct dentry
*dentry
= NULL
;
401 unsigned long fileid
;
405 dfprintk(DIRCACHE
, "NFS: nfs_do_filldir() filling starting @ cookie %Lu\n",
406 (unsigned long long)entry
->cookie
);
409 unsigned d_type
= DT_UNKNOWN
;
410 /* Note: entry->prev_cookie contains the cookie for
411 * retrieving the current dirent on the server */
412 fileid
= nfs_fileid_to_ino_t(entry
->ino
);
414 /* Get a dentry if we have one */
417 dentry
= nfs_readdir_lookup(desc
);
419 /* Use readdirplus info */
420 if (dentry
!= NULL
&& dentry
->d_inode
!= NULL
) {
421 d_type
= dt_type(dentry
->d_inode
);
422 fileid
= dentry
->d_inode
->i_ino
;
425 res
= filldir(dirent
, entry
->name
, entry
->len
,
426 file
->f_pos
, fileid
, d_type
);
430 *desc
->dir_cookie
= entry
->cookie
;
431 if (dir_decode(desc
) != 0) {
435 if (loop_count
++ > 200) {
440 dir_page_release(desc
);
443 dfprintk(DIRCACHE
, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n",
444 (unsigned long long)*desc
->dir_cookie
, res
);
449 * If we cannot find a cookie in our cache, we suspect that this is
450 * because it points to a deleted file, so we ask the server to return
451 * whatever it thinks is the next entry. We then feed this to filldir.
452 * If all goes well, we should then be able to find our way round the
453 * cache on the next call to readdir_search_pagecache();
455 * NOTE: we cannot add the anonymous page to the pagecache because
456 * the data it contains might not be page aligned. Besides,
457 * we should already have a complete representation of the
458 * directory in the page cache by the time we get here.
461 int uncached_readdir(nfs_readdir_descriptor_t
*desc
, void *dirent
,
464 struct file
*file
= desc
->file
;
465 struct inode
*inode
= file
->f_dentry
->d_inode
;
466 struct rpc_cred
*cred
= nfs_file_cred(file
);
467 struct page
*page
= NULL
;
470 dfprintk(DIRCACHE
, "NFS: uncached_readdir() searching for cookie %Lu\n",
471 (unsigned long long)*desc
->dir_cookie
);
473 page
= alloc_page(GFP_HIGHUSER
);
478 desc
->error
= NFS_PROTO(inode
)->readdir(file
->f_dentry
, cred
, *desc
->dir_cookie
,
480 NFS_SERVER(inode
)->dtsize
,
482 spin_lock(&inode
->i_lock
);
483 NFS_I(inode
)->cache_validity
|= NFS_INO_INVALID_ATIME
;
484 spin_unlock(&inode
->i_lock
);
486 desc
->ptr
= kmap(page
); /* matching kunmap in nfs_do_filldir */
487 if (desc
->error
>= 0) {
488 if ((status
= dir_decode(desc
)) == 0)
489 desc
->entry
->prev_cookie
= *desc
->dir_cookie
;
495 status
= nfs_do_filldir(desc
, dirent
, filldir
);
497 /* Reset read descriptor so it searches the page cache from
498 * the start upon the next call to readdir_search_pagecache() */
499 desc
->page_index
= 0;
500 desc
->entry
->cookie
= desc
->entry
->prev_cookie
= 0;
501 desc
->entry
->eof
= 0;
503 dfprintk(DIRCACHE
, "NFS: %s: returns %d\n",
504 __FUNCTION__
, status
);
507 dir_page_release(desc
);
511 /* The file offset position represents the dirent entry number. A
512 last cookie cache takes care of the common case of reading the
515 static int nfs_readdir(struct file
*filp
, void *dirent
, filldir_t filldir
)
517 struct dentry
*dentry
= filp
->f_dentry
;
518 struct inode
*inode
= dentry
->d_inode
;
519 nfs_readdir_descriptor_t my_desc
,
521 struct nfs_entry my_entry
;
523 struct nfs_fattr fattr
;
526 dfprintk(VFS
, "NFS: readdir(%s/%s) starting at cookie %Lu\n",
527 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
,
528 (long long)filp
->f_pos
);
529 nfs_inc_stats(inode
, NFSIOS_VFSGETDENTS
);
533 res
= nfs_revalidate_mapping(inode
, filp
->f_mapping
);
540 * filp->f_pos points to the dirent entry number.
541 * *desc->dir_cookie has the cookie for the next entry. We have
542 * to either find the entry with the appropriate number or
543 * revalidate the cookie.
545 memset(desc
, 0, sizeof(*desc
));
548 desc
->dir_cookie
= &((struct nfs_open_context
*)filp
->private_data
)->dir_cookie
;
549 desc
->decode
= NFS_PROTO(inode
)->decode_dirent
;
550 desc
->plus
= NFS_USE_READDIRPLUS(inode
);
552 my_entry
.cookie
= my_entry
.prev_cookie
= 0;
555 my_entry
.fattr
= &fattr
;
556 nfs_fattr_init(&fattr
);
557 desc
->entry
= &my_entry
;
559 while(!desc
->entry
->eof
) {
560 res
= readdir_search_pagecache(desc
);
562 if (res
== -EBADCOOKIE
) {
563 /* This means either end of directory */
564 if (*desc
->dir_cookie
&& desc
->entry
->cookie
!= *desc
->dir_cookie
) {
565 /* Or that the server has 'lost' a cookie */
566 res
= uncached_readdir(desc
, dirent
, filldir
);
573 if (res
== -ETOOSMALL
&& desc
->plus
) {
574 clear_bit(NFS_INO_ADVISE_RDPLUS
, &NFS_FLAGS(inode
));
575 nfs_zap_caches(inode
);
577 desc
->entry
->eof
= 0;
583 res
= nfs_do_filldir(desc
, dirent
, filldir
);
592 dfprintk(VFS
, "NFS: readdir(%s/%s) returns %ld\n",
593 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
,
598 loff_t
nfs_llseek_dir(struct file
*filp
, loff_t offset
, int origin
)
600 mutex_lock(&filp
->f_dentry
->d_inode
->i_mutex
);
603 offset
+= filp
->f_pos
;
611 if (offset
!= filp
->f_pos
) {
612 filp
->f_pos
= offset
;
613 ((struct nfs_open_context
*)filp
->private_data
)->dir_cookie
= 0;
616 mutex_unlock(&filp
->f_dentry
->d_inode
->i_mutex
);
621 * All directory operations under NFS are synchronous, so fsync()
622 * is a dummy operation.
624 int nfs_fsync_dir(struct file
*filp
, struct dentry
*dentry
, int datasync
)
626 dfprintk(VFS
, "NFS: fsync_dir(%s/%s) datasync %d\n",
627 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
,
634 * A check for whether or not the parent directory has changed.
635 * In the case it has, we assume that the dentries are untrustworthy
636 * and may need to be looked up again.
638 static inline int nfs_check_verifier(struct inode
*dir
, struct dentry
*dentry
)
642 if ((NFS_I(dir
)->cache_validity
& NFS_INO_INVALID_ATTR
) != 0
643 || nfs_attribute_timeout(dir
))
645 return nfs_verify_change_attribute(dir
, (unsigned long)dentry
->d_fsdata
);
648 static inline void nfs_set_verifier(struct dentry
* dentry
, unsigned long verf
)
650 dentry
->d_fsdata
= (void *)verf
;
654 * Whenever an NFS operation succeeds, we know that the dentry
655 * is valid, so we update the revalidation timestamp.
657 static inline void nfs_renew_times(struct dentry
* dentry
)
659 dentry
->d_time
= jiffies
;
663 * Return the intent data that applies to this particular path component
665 * Note that the current set of intents only apply to the very last
666 * component of the path.
667 * We check for this using LOOKUP_CONTINUE and LOOKUP_PARENT.
669 static inline unsigned int nfs_lookup_check_intent(struct nameidata
*nd
, unsigned int mask
)
671 if (nd
->flags
& (LOOKUP_CONTINUE
|LOOKUP_PARENT
))
673 return nd
->flags
& mask
;
677 * Inode and filehandle revalidation for lookups.
679 * We force revalidation in the cases where the VFS sets LOOKUP_REVAL,
680 * or if the intent information indicates that we're about to open this
681 * particular file and the "nocto" mount flag is not set.
685 int nfs_lookup_verify_inode(struct inode
*inode
, struct nameidata
*nd
)
687 struct nfs_server
*server
= NFS_SERVER(inode
);
690 /* VFS wants an on-the-wire revalidation */
691 if (nd
->flags
& LOOKUP_REVAL
)
693 /* This is an open(2) */
694 if (nfs_lookup_check_intent(nd
, LOOKUP_OPEN
) != 0 &&
695 !(server
->flags
& NFS_MOUNT_NOCTO
) &&
696 (S_ISREG(inode
->i_mode
) ||
697 S_ISDIR(inode
->i_mode
)))
700 return nfs_revalidate_inode(server
, inode
);
702 return __nfs_revalidate_inode(server
, inode
);
706 * We judge how long we want to trust negative
707 * dentries by looking at the parent inode mtime.
709 * If parent mtime has changed, we revalidate, else we wait for a
710 * period corresponding to the parent's attribute cache timeout value.
713 int nfs_neg_need_reval(struct inode
*dir
, struct dentry
*dentry
,
714 struct nameidata
*nd
)
716 /* Don't revalidate a negative dentry if we're creating a new file */
717 if (nd
!= NULL
&& nfs_lookup_check_intent(nd
, LOOKUP_CREATE
) != 0)
719 return !nfs_check_verifier(dir
, dentry
);
723 * This is called every time the dcache has a lookup hit,
724 * and we should check whether we can really trust that
727 * NOTE! The hit can be a negative hit too, don't assume
730 * If the parent directory is seen to have changed, we throw out the
731 * cached dentry and do a new lookup.
733 static int nfs_lookup_revalidate(struct dentry
* dentry
, struct nameidata
*nd
)
737 struct dentry
*parent
;
739 struct nfs_fh fhandle
;
740 struct nfs_fattr fattr
;
741 unsigned long verifier
;
743 parent
= dget_parent(dentry
);
745 dir
= parent
->d_inode
;
746 nfs_inc_stats(dir
, NFSIOS_DENTRYREVALIDATE
);
747 inode
= dentry
->d_inode
;
750 if (nfs_neg_need_reval(dir
, dentry
, nd
))
755 if (is_bad_inode(inode
)) {
756 dfprintk(LOOKUPCACHE
, "%s: %s/%s has dud inode\n",
757 __FUNCTION__
, dentry
->d_parent
->d_name
.name
,
758 dentry
->d_name
.name
);
762 /* Revalidate parent directory attribute cache */
763 if (nfs_revalidate_inode(NFS_SERVER(dir
), dir
) < 0)
766 /* Force a full look up iff the parent directory has changed */
767 if (nfs_check_verifier(dir
, dentry
)) {
768 if (nfs_lookup_verify_inode(inode
, nd
))
773 if (NFS_STALE(inode
))
776 verifier
= nfs_save_change_attribute(dir
);
777 error
= NFS_PROTO(dir
)->lookup(dir
, &dentry
->d_name
, &fhandle
, &fattr
);
780 if (nfs_compare_fh(NFS_FH(inode
), &fhandle
))
782 if ((error
= nfs_refresh_inode(inode
, &fattr
)) != 0)
785 nfs_renew_times(dentry
);
786 nfs_set_verifier(dentry
, verifier
);
790 dfprintk(LOOKUPCACHE
, "NFS: %s(%s/%s) is valid\n",
791 __FUNCTION__
, dentry
->d_parent
->d_name
.name
,
792 dentry
->d_name
.name
);
798 if (inode
&& S_ISDIR(inode
->i_mode
)) {
799 /* Purge readdir caches. */
800 nfs_zap_caches(inode
);
801 /* If we have submounts, don't unhash ! */
802 if (have_submounts(dentry
))
804 shrink_dcache_parent(dentry
);
809 dfprintk(LOOKUPCACHE
, "NFS: %s(%s/%s) is invalid\n",
810 __FUNCTION__
, dentry
->d_parent
->d_name
.name
,
811 dentry
->d_name
.name
);
816 * This is called from dput() when d_count is going to 0.
818 static int nfs_dentry_delete(struct dentry
*dentry
)
820 dfprintk(VFS
, "NFS: dentry_delete(%s/%s, %x)\n",
821 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
,
824 if (dentry
->d_flags
& DCACHE_NFSFS_RENAMED
) {
825 /* Unhash it, so that ->d_iput() would be called */
828 if (!(dentry
->d_sb
->s_flags
& MS_ACTIVE
)) {
829 /* Unhash it, so that ancestors of killed async unlink
830 * files will be cleaned up during umount */
838 * Called when the dentry loses inode.
839 * We use it to clean up silly-renamed files.
841 static void nfs_dentry_iput(struct dentry
*dentry
, struct inode
*inode
)
843 nfs_inode_return_delegation(inode
);
844 if (dentry
->d_flags
& DCACHE_NFSFS_RENAMED
) {
847 nfs_complete_unlink(dentry
);
850 /* When creating a negative dentry, we want to renew d_time */
851 nfs_renew_times(dentry
);
855 struct dentry_operations nfs_dentry_operations
= {
856 .d_revalidate
= nfs_lookup_revalidate
,
857 .d_delete
= nfs_dentry_delete
,
858 .d_iput
= nfs_dentry_iput
,
862 * Use intent information to check whether or not we're going to do
863 * an O_EXCL create using this path component.
866 int nfs_is_exclusive_create(struct inode
*dir
, struct nameidata
*nd
)
868 if (NFS_PROTO(dir
)->version
== 2)
870 if (nd
== NULL
|| nfs_lookup_check_intent(nd
, LOOKUP_CREATE
) == 0)
872 return (nd
->intent
.open
.flags
& O_EXCL
) != 0;
875 static inline int nfs_reval_fsid(struct vfsmount
*mnt
, struct inode
*dir
,
876 struct nfs_fh
*fh
, struct nfs_fattr
*fattr
)
878 struct nfs_server
*server
= NFS_SERVER(dir
);
880 if (!nfs_fsid_equal(&server
->fsid
, &fattr
->fsid
))
881 /* Revalidate fsid on root dir */
882 return __nfs_revalidate_inode(server
, mnt
->mnt_root
->d_inode
);
886 static struct dentry
*nfs_lookup(struct inode
*dir
, struct dentry
* dentry
, struct nameidata
*nd
)
889 struct inode
*inode
= NULL
;
891 struct nfs_fh fhandle
;
892 struct nfs_fattr fattr
;
894 dfprintk(VFS
, "NFS: lookup(%s/%s)\n",
895 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
);
896 nfs_inc_stats(dir
, NFSIOS_VFSLOOKUP
);
898 res
= ERR_PTR(-ENAMETOOLONG
);
899 if (dentry
->d_name
.len
> NFS_SERVER(dir
)->namelen
)
902 res
= ERR_PTR(-ENOMEM
);
903 dentry
->d_op
= NFS_PROTO(dir
)->dentry_ops
;
908 * If we're doing an exclusive create, optimize away the lookup
909 * but don't hash the dentry.
911 if (nfs_is_exclusive_create(dir
, nd
)) {
912 d_instantiate(dentry
, NULL
);
917 error
= NFS_PROTO(dir
)->lookup(dir
, &dentry
->d_name
, &fhandle
, &fattr
);
918 if (error
== -ENOENT
)
921 res
= ERR_PTR(error
);
924 error
= nfs_reval_fsid(nd
->mnt
, dir
, &fhandle
, &fattr
);
926 res
= ERR_PTR(error
);
929 inode
= nfs_fhget(dentry
->d_sb
, &fhandle
, &fattr
);
930 res
= (struct dentry
*)inode
;
935 res
= d_materialise_unique(dentry
, inode
);
938 nfs_renew_times(dentry
);
939 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
947 static int nfs_open_revalidate(struct dentry
*, struct nameidata
*);
949 struct dentry_operations nfs4_dentry_operations
= {
950 .d_revalidate
= nfs_open_revalidate
,
951 .d_delete
= nfs_dentry_delete
,
952 .d_iput
= nfs_dentry_iput
,
956 * Use intent information to determine whether we need to substitute
957 * the NFSv4-style stateful OPEN for the LOOKUP call
959 static int is_atomic_open(struct inode
*dir
, struct nameidata
*nd
)
961 if (nd
== NULL
|| nfs_lookup_check_intent(nd
, LOOKUP_OPEN
) == 0)
963 /* NFS does not (yet) have a stateful open for directories */
964 if (nd
->flags
& LOOKUP_DIRECTORY
)
966 /* Are we trying to write to a read only partition? */
967 if (IS_RDONLY(dir
) && (nd
->intent
.open
.flags
& (O_CREAT
|O_TRUNC
|FMODE_WRITE
)))
972 static struct dentry
*nfs_atomic_lookup(struct inode
*dir
, struct dentry
*dentry
, struct nameidata
*nd
)
974 struct dentry
*res
= NULL
;
977 dfprintk(VFS
, "NFS: atomic_lookup(%s/%ld), %s\n",
978 dir
->i_sb
->s_id
, dir
->i_ino
, dentry
->d_name
.name
);
980 /* Check that we are indeed trying to open this file */
981 if (!is_atomic_open(dir
, nd
))
984 if (dentry
->d_name
.len
> NFS_SERVER(dir
)->namelen
) {
985 res
= ERR_PTR(-ENAMETOOLONG
);
988 dentry
->d_op
= NFS_PROTO(dir
)->dentry_ops
;
990 /* Let vfs_create() deal with O_EXCL */
991 if (nd
->intent
.open
.flags
& O_EXCL
) {
996 /* Open the file on the server */
998 /* Revalidate parent directory attribute cache */
999 error
= nfs_revalidate_inode(NFS_SERVER(dir
), dir
);
1001 res
= ERR_PTR(error
);
1006 if (nd
->intent
.open
.flags
& O_CREAT
) {
1007 nfs_begin_data_update(dir
);
1008 res
= nfs4_atomic_open(dir
, dentry
, nd
);
1009 nfs_end_data_update(dir
);
1011 res
= nfs4_atomic_open(dir
, dentry
, nd
);
1014 error
= PTR_ERR(res
);
1016 /* Make a negative dentry */
1020 /* This turned out not to be a regular file */
1025 if (!(nd
->intent
.open
.flags
& O_NOFOLLOW
))
1031 } else if (res
!= NULL
)
1033 nfs_renew_times(dentry
);
1034 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1038 return nfs_lookup(dir
, dentry
, nd
);
1041 static int nfs_open_revalidate(struct dentry
*dentry
, struct nameidata
*nd
)
1043 struct dentry
*parent
= NULL
;
1044 struct inode
*inode
= dentry
->d_inode
;
1046 unsigned long verifier
;
1047 int openflags
, ret
= 0;
1049 parent
= dget_parent(dentry
);
1050 dir
= parent
->d_inode
;
1051 if (!is_atomic_open(dir
, nd
))
1053 /* We can't create new files in nfs_open_revalidate(), so we
1054 * optimize away revalidation of negative dentries.
1058 /* NFS only supports OPEN on regular files */
1059 if (!S_ISREG(inode
->i_mode
))
1061 openflags
= nd
->intent
.open
.flags
;
1062 /* We cannot do exclusive creation on a positive dentry */
1063 if ((openflags
& (O_CREAT
|O_EXCL
)) == (O_CREAT
|O_EXCL
))
1065 /* We can't create new files, or truncate existing ones here */
1066 openflags
&= ~(O_CREAT
|O_TRUNC
);
1069 * Note: we're not holding inode->i_mutex and so may be racing with
1070 * operations that change the directory. We therefore save the
1071 * change attribute *before* we do the RPC call.
1074 verifier
= nfs_save_change_attribute(dir
);
1075 ret
= nfs4_open_revalidate(dir
, dentry
, openflags
, nd
);
1077 nfs_set_verifier(dentry
, verifier
);
1086 if (inode
!= NULL
&& nfs_have_delegation(inode
, FMODE_READ
))
1088 return nfs_lookup_revalidate(dentry
, nd
);
1090 #endif /* CONFIG_NFSV4 */
1092 static struct dentry
*nfs_readdir_lookup(nfs_readdir_descriptor_t
*desc
)
1094 struct dentry
*parent
= desc
->file
->f_dentry
;
1095 struct inode
*dir
= parent
->d_inode
;
1096 struct nfs_entry
*entry
= desc
->entry
;
1097 struct dentry
*dentry
, *alias
;
1098 struct qstr name
= {
1099 .name
= entry
->name
,
1102 struct inode
*inode
;
1106 if (name
.name
[0] == '.' && name
.name
[1] == '.')
1107 return dget_parent(parent
);
1110 if (name
.name
[0] == '.')
1111 return dget(parent
);
1113 name
.hash
= full_name_hash(name
.name
, name
.len
);
1114 dentry
= d_lookup(parent
, &name
);
1117 if (!desc
->plus
|| !(entry
->fattr
->valid
& NFS_ATTR_FATTR
))
1119 /* Note: caller is already holding the dir->i_mutex! */
1120 dentry
= d_alloc(parent
, &name
);
1123 dentry
->d_op
= NFS_PROTO(dir
)->dentry_ops
;
1124 inode
= nfs_fhget(dentry
->d_sb
, entry
->fh
, entry
->fattr
);
1125 if (IS_ERR(inode
)) {
1130 alias
= d_materialise_unique(dentry
, inode
);
1131 if (alias
!= NULL
) {
1136 nfs_renew_times(dentry
);
1137 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1142 * Code common to create, mkdir, and mknod.
1144 int nfs_instantiate(struct dentry
*dentry
, struct nfs_fh
*fhandle
,
1145 struct nfs_fattr
*fattr
)
1147 struct inode
*inode
;
1148 int error
= -EACCES
;
1150 /* We may have been initialized further down */
1151 if (dentry
->d_inode
)
1153 if (fhandle
->size
== 0) {
1154 struct inode
*dir
= dentry
->d_parent
->d_inode
;
1155 error
= NFS_PROTO(dir
)->lookup(dir
, &dentry
->d_name
, fhandle
, fattr
);
1159 if (!(fattr
->valid
& NFS_ATTR_FATTR
)) {
1160 struct nfs_server
*server
= NFS_SB(dentry
->d_sb
);
1161 error
= server
->nfs_client
->rpc_ops
->getattr(server
, fhandle
, fattr
);
1165 inode
= nfs_fhget(dentry
->d_sb
, fhandle
, fattr
);
1166 error
= PTR_ERR(inode
);
1169 d_instantiate(dentry
, inode
);
1170 if (d_unhashed(dentry
))
1176 * Following a failed create operation, we drop the dentry rather
1177 * than retain a negative dentry. This avoids a problem in the event
1178 * that the operation succeeded on the server, but an error in the
1179 * reply path made it appear to have failed.
1181 static int nfs_create(struct inode
*dir
, struct dentry
*dentry
, int mode
,
1182 struct nameidata
*nd
)
1188 dfprintk(VFS
, "NFS: create(%s/%ld), %s\n",
1189 dir
->i_sb
->s_id
, dir
->i_ino
, dentry
->d_name
.name
);
1191 attr
.ia_mode
= mode
;
1192 attr
.ia_valid
= ATTR_MODE
;
1194 if (nd
&& (nd
->flags
& LOOKUP_CREATE
))
1195 open_flags
= nd
->intent
.open
.flags
;
1198 nfs_begin_data_update(dir
);
1199 error
= NFS_PROTO(dir
)->create(dir
, dentry
, &attr
, open_flags
, nd
);
1200 nfs_end_data_update(dir
);
1203 nfs_renew_times(dentry
);
1204 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1214 * See comments for nfs_proc_create regarding failed operations.
1217 nfs_mknod(struct inode
*dir
, struct dentry
*dentry
, int mode
, dev_t rdev
)
1222 dfprintk(VFS
, "NFS: mknod(%s/%ld), %s\n",
1223 dir
->i_sb
->s_id
, dir
->i_ino
, dentry
->d_name
.name
);
1225 if (!new_valid_dev(rdev
))
1228 attr
.ia_mode
= mode
;
1229 attr
.ia_valid
= ATTR_MODE
;
1232 nfs_begin_data_update(dir
);
1233 status
= NFS_PROTO(dir
)->mknod(dir
, dentry
, &attr
, rdev
);
1234 nfs_end_data_update(dir
);
1237 nfs_renew_times(dentry
);
1238 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1248 * See comments for nfs_proc_create regarding failed operations.
1250 static int nfs_mkdir(struct inode
*dir
, struct dentry
*dentry
, int mode
)
1255 dfprintk(VFS
, "NFS: mkdir(%s/%ld), %s\n",
1256 dir
->i_sb
->s_id
, dir
->i_ino
, dentry
->d_name
.name
);
1258 attr
.ia_valid
= ATTR_MODE
;
1259 attr
.ia_mode
= mode
| S_IFDIR
;
1262 nfs_begin_data_update(dir
);
1263 error
= NFS_PROTO(dir
)->mkdir(dir
, dentry
, &attr
);
1264 nfs_end_data_update(dir
);
1267 nfs_renew_times(dentry
);
1268 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1277 static int nfs_rmdir(struct inode
*dir
, struct dentry
*dentry
)
1281 dfprintk(VFS
, "NFS: rmdir(%s/%ld), %s\n",
1282 dir
->i_sb
->s_id
, dir
->i_ino
, dentry
->d_name
.name
);
1285 nfs_begin_data_update(dir
);
1286 error
= NFS_PROTO(dir
)->rmdir(dir
, &dentry
->d_name
);
1287 /* Ensure the VFS deletes this inode */
1288 if (error
== 0 && dentry
->d_inode
!= NULL
)
1289 clear_nlink(dentry
->d_inode
);
1290 nfs_end_data_update(dir
);
1296 static int nfs_sillyrename(struct inode
*dir
, struct dentry
*dentry
)
1298 static unsigned int sillycounter
;
1299 const int i_inosize
= sizeof(dir
->i_ino
)*2;
1300 const int countersize
= sizeof(sillycounter
)*2;
1301 const int slen
= sizeof(".nfs") + i_inosize
+ countersize
- 1;
1304 struct dentry
*sdentry
;
1307 dfprintk(VFS
, "NFS: silly-rename(%s/%s, ct=%d)\n",
1308 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
,
1309 atomic_read(&dentry
->d_count
));
1310 nfs_inc_stats(dir
, NFSIOS_SILLYRENAME
);
1313 if (!dentry
->d_inode
)
1314 printk("NFS: silly-renaming %s/%s, negative dentry??\n",
1315 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
);
1318 * We don't allow a dentry to be silly-renamed twice.
1321 if (dentry
->d_flags
& DCACHE_NFSFS_RENAMED
)
1324 sprintf(silly
, ".nfs%*.*lx",
1325 i_inosize
, i_inosize
, dentry
->d_inode
->i_ino
);
1327 /* Return delegation in anticipation of the rename */
1328 nfs_inode_return_delegation(dentry
->d_inode
);
1332 char *suffix
= silly
+ slen
- countersize
;
1336 sprintf(suffix
, "%*.*x", countersize
, countersize
, sillycounter
);
1338 dfprintk(VFS
, "NFS: trying to rename %s to %s\n",
1339 dentry
->d_name
.name
, silly
);
1341 sdentry
= lookup_one_len(silly
, dentry
->d_parent
, slen
);
1343 * N.B. Better to return EBUSY here ... it could be
1344 * dangerous to delete the file while it's in use.
1346 if (IS_ERR(sdentry
))
1348 } while(sdentry
->d_inode
!= NULL
); /* need negative lookup */
1350 qsilly
.name
= silly
;
1351 qsilly
.len
= strlen(silly
);
1352 nfs_begin_data_update(dir
);
1353 if (dentry
->d_inode
) {
1354 nfs_begin_data_update(dentry
->d_inode
);
1355 error
= NFS_PROTO(dir
)->rename(dir
, &dentry
->d_name
,
1357 nfs_mark_for_revalidate(dentry
->d_inode
);
1358 nfs_end_data_update(dentry
->d_inode
);
1360 error
= NFS_PROTO(dir
)->rename(dir
, &dentry
->d_name
,
1362 nfs_end_data_update(dir
);
1364 nfs_renew_times(dentry
);
1365 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1366 d_move(dentry
, sdentry
);
1367 error
= nfs_async_unlink(dentry
);
1368 /* If we return 0 we don't unlink */
1376 * Remove a file after making sure there are no pending writes,
1377 * and after checking that the file has only one user.
1379 * We invalidate the attribute cache and free the inode prior to the operation
1380 * to avoid possible races if the server reuses the inode.
1382 static int nfs_safe_remove(struct dentry
*dentry
)
1384 struct inode
*dir
= dentry
->d_parent
->d_inode
;
1385 struct inode
*inode
= dentry
->d_inode
;
1388 dfprintk(VFS
, "NFS: safe_remove(%s/%s)\n",
1389 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
);
1391 /* If the dentry was sillyrenamed, we simply call d_delete() */
1392 if (dentry
->d_flags
& DCACHE_NFSFS_RENAMED
) {
1397 nfs_begin_data_update(dir
);
1398 if (inode
!= NULL
) {
1399 nfs_inode_return_delegation(inode
);
1400 nfs_begin_data_update(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
);
1406 nfs_end_data_update(inode
);
1408 error
= NFS_PROTO(dir
)->remove(dir
, &dentry
->d_name
);
1409 nfs_end_data_update(dir
);
1414 /* We do silly rename. In case sillyrename() returns -EBUSY, the inode
1415 * belongs to an active ".nfs..." file and we return -EBUSY.
1417 * If sillyrename() returns 0, we do nothing, otherwise we unlink.
1419 static int nfs_unlink(struct inode
*dir
, struct dentry
*dentry
)
1422 int need_rehash
= 0;
1424 dfprintk(VFS
, "NFS: unlink(%s/%ld, %s)\n", dir
->i_sb
->s_id
,
1425 dir
->i_ino
, dentry
->d_name
.name
);
1428 spin_lock(&dcache_lock
);
1429 spin_lock(&dentry
->d_lock
);
1430 if (atomic_read(&dentry
->d_count
) > 1) {
1431 spin_unlock(&dentry
->d_lock
);
1432 spin_unlock(&dcache_lock
);
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_renew_times(dentry
);
1446 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1447 } else if (need_rehash
)
1454 * To create a symbolic link, most file systems instantiate a new inode,
1455 * add a page to it containing the path, then write it out to the disk
1456 * using prepare_write/commit_write.
1458 * Unfortunately the NFS client can't create the in-core inode first
1459 * because it needs a file handle to create an in-core inode (see
1460 * fs/nfs/inode.c:nfs_fhget). We only have a file handle *after* the
1461 * symlink request has completed on the server.
1463 * So instead we allocate a raw page, copy the symname into it, then do
1464 * the SYMLINK request with the page as the buffer. If it succeeds, we
1465 * now have a new file handle and can instantiate an in-core NFS inode
1466 * and move the raw page into its mapping.
1468 static int nfs_symlink(struct inode
*dir
, struct dentry
*dentry
, const char *symname
)
1470 struct pagevec lru_pvec
;
1474 unsigned int pathlen
= strlen(symname
);
1477 dfprintk(VFS
, "NFS: symlink(%s/%ld, %s, %s)\n", dir
->i_sb
->s_id
,
1478 dir
->i_ino
, dentry
->d_name
.name
, symname
);
1480 if (pathlen
> PAGE_SIZE
)
1481 return -ENAMETOOLONG
;
1483 attr
.ia_mode
= S_IFLNK
| S_IRWXUGO
;
1484 attr
.ia_valid
= ATTR_MODE
;
1488 page
= alloc_page(GFP_KERNEL
);
1494 kaddr
= kmap_atomic(page
, KM_USER0
);
1495 memcpy(kaddr
, symname
, pathlen
);
1496 if (pathlen
< PAGE_SIZE
)
1497 memset(kaddr
+ pathlen
, 0, PAGE_SIZE
- pathlen
);
1498 kunmap_atomic(kaddr
, KM_USER0
);
1500 nfs_begin_data_update(dir
);
1501 error
= NFS_PROTO(dir
)->symlink(dir
, dentry
, page
, pathlen
, &attr
);
1502 nfs_end_data_update(dir
);
1504 dfprintk(VFS
, "NFS: symlink(%s/%ld, %s, %s) error %d\n",
1505 dir
->i_sb
->s_id
, dir
->i_ino
,
1506 dentry
->d_name
.name
, symname
, error
);
1514 * No big deal if we can't add this page to the page cache here.
1515 * READLINK will get the missing page from the server if needed.
1517 pagevec_init(&lru_pvec
, 0);
1518 if (!add_to_page_cache(page
, dentry
->d_inode
->i_mapping
, 0,
1520 if (!pagevec_add(&lru_pvec
, page
))
1521 __pagevec_lru_add(&lru_pvec
);
1522 SetPageUptodate(page
);
1532 nfs_link(struct dentry
*old_dentry
, struct inode
*dir
, struct dentry
*dentry
)
1534 struct inode
*inode
= old_dentry
->d_inode
;
1537 dfprintk(VFS
, "NFS: link(%s/%s -> %s/%s)\n",
1538 old_dentry
->d_parent
->d_name
.name
, old_dentry
->d_name
.name
,
1539 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
);
1542 nfs_begin_data_update(dir
);
1543 nfs_begin_data_update(inode
);
1544 error
= NFS_PROTO(dir
)->link(inode
, dir
, &dentry
->d_name
);
1546 atomic_inc(&inode
->i_count
);
1547 d_instantiate(dentry
, inode
);
1549 nfs_end_data_update(inode
);
1550 nfs_end_data_update(dir
);
1557 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a
1558 * different file handle for the same inode after a rename (e.g. when
1559 * moving to a different directory). A fail-safe method to do so would
1560 * be to look up old_dir/old_name, create a link to new_dir/new_name and
1561 * rename the old file using the sillyrename stuff. This way, the original
1562 * file in old_dir will go away when the last process iput()s the inode.
1566 * It actually works quite well. One needs to have the possibility for
1567 * at least one ".nfs..." file in each directory the file ever gets
1568 * moved or linked to which happens automagically with the new
1569 * implementation that only depends on the dcache stuff instead of
1570 * using the inode layer
1572 * Unfortunately, things are a little more complicated than indicated
1573 * above. For a cross-directory move, we want to make sure we can get
1574 * rid of the old inode after the operation. This means there must be
1575 * no pending writes (if it's a file), and the use count must be 1.
1576 * If these conditions are met, we can drop the dentries before doing
1579 static int nfs_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
1580 struct inode
*new_dir
, struct dentry
*new_dentry
)
1582 struct inode
*old_inode
= old_dentry
->d_inode
;
1583 struct inode
*new_inode
= new_dentry
->d_inode
;
1584 struct dentry
*dentry
= NULL
, *rehash
= NULL
;
1588 * To prevent any new references to the target during the rename,
1589 * we unhash the dentry and free the inode in advance.
1592 if (!d_unhashed(new_dentry
)) {
1594 rehash
= new_dentry
;
1597 dfprintk(VFS
, "NFS: rename(%s/%s -> %s/%s, ct=%d)\n",
1598 old_dentry
->d_parent
->d_name
.name
, old_dentry
->d_name
.name
,
1599 new_dentry
->d_parent
->d_name
.name
, new_dentry
->d_name
.name
,
1600 atomic_read(&new_dentry
->d_count
));
1603 * First check whether the target is busy ... we can't
1604 * safely do _any_ rename if the target is in use.
1606 * For files, make a copy of the dentry and then do a
1607 * silly-rename. If the silly-rename succeeds, the
1608 * copied dentry is hashed and becomes the new target.
1612 if (S_ISDIR(new_inode
->i_mode
)) {
1614 if (!S_ISDIR(old_inode
->i_mode
))
1616 } else if (atomic_read(&new_dentry
->d_count
) > 2) {
1618 /* copy the target dentry's name */
1619 dentry
= d_alloc(new_dentry
->d_parent
,
1620 &new_dentry
->d_name
);
1624 /* silly-rename the existing target ... */
1625 err
= nfs_sillyrename(new_dir
, new_dentry
);
1627 new_dentry
= rehash
= dentry
;
1629 /* instantiate the replacement target */
1630 d_instantiate(new_dentry
, NULL
);
1631 } else if (atomic_read(&new_dentry
->d_count
) > 1) {
1632 /* dentry still busy? */
1634 printk("nfs_rename: target %s/%s busy, d_count=%d\n",
1635 new_dentry
->d_parent
->d_name
.name
,
1636 new_dentry
->d_name
.name
,
1637 atomic_read(&new_dentry
->d_count
));
1642 drop_nlink(new_inode
);
1646 * ... prune child dentries and writebacks if needed.
1648 if (atomic_read(&old_dentry
->d_count
) > 1) {
1649 nfs_wb_all(old_inode
);
1650 shrink_dcache_parent(old_dentry
);
1652 nfs_inode_return_delegation(old_inode
);
1654 if (new_inode
!= NULL
) {
1655 nfs_inode_return_delegation(new_inode
);
1656 d_delete(new_dentry
);
1659 nfs_begin_data_update(old_dir
);
1660 nfs_begin_data_update(new_dir
);
1661 nfs_begin_data_update(old_inode
);
1662 error
= NFS_PROTO(old_dir
)->rename(old_dir
, &old_dentry
->d_name
,
1663 new_dir
, &new_dentry
->d_name
);
1664 nfs_mark_for_revalidate(old_inode
);
1665 nfs_end_data_update(old_inode
);
1666 nfs_end_data_update(new_dir
);
1667 nfs_end_data_update(old_dir
);
1672 d_move(old_dentry
, new_dentry
);
1673 nfs_renew_times(new_dentry
);
1674 nfs_set_verifier(new_dentry
, nfs_save_change_attribute(new_dir
));
1677 /* new dentry created? */
1684 static DEFINE_SPINLOCK(nfs_access_lru_lock
);
1685 static LIST_HEAD(nfs_access_lru_list
);
1686 static atomic_long_t nfs_access_nr_entries
;
1688 static void nfs_access_free_entry(struct nfs_access_entry
*entry
)
1690 put_rpccred(entry
->cred
);
1692 smp_mb__before_atomic_dec();
1693 atomic_long_dec(&nfs_access_nr_entries
);
1694 smp_mb__after_atomic_dec();
1697 int nfs_access_cache_shrinker(int nr_to_scan
, gfp_t gfp_mask
)
1700 struct nfs_inode
*nfsi
;
1701 struct nfs_access_entry
*cache
;
1703 spin_lock(&nfs_access_lru_lock
);
1705 list_for_each_entry(nfsi
, &nfs_access_lru_list
, access_cache_inode_lru
) {
1706 struct inode
*inode
;
1708 if (nr_to_scan
-- == 0)
1710 inode
= igrab(&nfsi
->vfs_inode
);
1713 spin_lock(&inode
->i_lock
);
1714 if (list_empty(&nfsi
->access_cache_entry_lru
))
1715 goto remove_lru_entry
;
1716 cache
= list_entry(nfsi
->access_cache_entry_lru
.next
,
1717 struct nfs_access_entry
, lru
);
1718 list_move(&cache
->lru
, &head
);
1719 rb_erase(&cache
->rb_node
, &nfsi
->access_cache
);
1720 if (!list_empty(&nfsi
->access_cache_entry_lru
))
1721 list_move_tail(&nfsi
->access_cache_inode_lru
,
1722 &nfs_access_lru_list
);
1725 list_del_init(&nfsi
->access_cache_inode_lru
);
1726 clear_bit(NFS_INO_ACL_LRU_SET
, &nfsi
->flags
);
1728 spin_unlock(&inode
->i_lock
);
1732 spin_unlock(&nfs_access_lru_lock
);
1733 while (!list_empty(&head
)) {
1734 cache
= list_entry(head
.next
, struct nfs_access_entry
, lru
);
1735 list_del(&cache
->lru
);
1736 nfs_access_free_entry(cache
);
1738 return (atomic_long_read(&nfs_access_nr_entries
) / 100) * sysctl_vfs_cache_pressure
;
1741 static void __nfs_access_zap_cache(struct inode
*inode
)
1743 struct nfs_inode
*nfsi
= NFS_I(inode
);
1744 struct rb_root
*root_node
= &nfsi
->access_cache
;
1745 struct rb_node
*n
, *dispose
= NULL
;
1746 struct nfs_access_entry
*entry
;
1748 /* Unhook entries from the cache */
1749 while ((n
= rb_first(root_node
)) != NULL
) {
1750 entry
= rb_entry(n
, struct nfs_access_entry
, rb_node
);
1751 rb_erase(n
, root_node
);
1752 list_del(&entry
->lru
);
1753 n
->rb_left
= dispose
;
1756 nfsi
->cache_validity
&= ~NFS_INO_INVALID_ACCESS
;
1757 spin_unlock(&inode
->i_lock
);
1759 /* Now kill them all! */
1760 while (dispose
!= NULL
) {
1762 dispose
= n
->rb_left
;
1763 nfs_access_free_entry(rb_entry(n
, struct nfs_access_entry
, rb_node
));
1767 void nfs_access_zap_cache(struct inode
*inode
)
1769 /* Remove from global LRU init */
1770 if (test_and_clear_bit(NFS_INO_ACL_LRU_SET
, &NFS_FLAGS(inode
))) {
1771 spin_lock(&nfs_access_lru_lock
);
1772 list_del_init(&NFS_I(inode
)->access_cache_inode_lru
);
1773 spin_unlock(&nfs_access_lru_lock
);
1776 spin_lock(&inode
->i_lock
);
1777 /* This will release the spinlock */
1778 __nfs_access_zap_cache(inode
);
1781 static struct nfs_access_entry
*nfs_access_search_rbtree(struct inode
*inode
, struct rpc_cred
*cred
)
1783 struct rb_node
*n
= NFS_I(inode
)->access_cache
.rb_node
;
1784 struct nfs_access_entry
*entry
;
1787 entry
= rb_entry(n
, struct nfs_access_entry
, rb_node
);
1789 if (cred
< entry
->cred
)
1791 else if (cred
> entry
->cred
)
1799 int nfs_access_get_cached(struct inode
*inode
, struct rpc_cred
*cred
, struct nfs_access_entry
*res
)
1801 struct nfs_inode
*nfsi
= NFS_I(inode
);
1802 struct nfs_access_entry
*cache
;
1805 spin_lock(&inode
->i_lock
);
1806 if (nfsi
->cache_validity
& NFS_INO_INVALID_ACCESS
)
1808 cache
= nfs_access_search_rbtree(inode
, cred
);
1811 if (time_after(jiffies
, cache
->jiffies
+ NFS_ATTRTIMEO(inode
)))
1813 res
->jiffies
= cache
->jiffies
;
1814 res
->cred
= cache
->cred
;
1815 res
->mask
= cache
->mask
;
1816 list_move_tail(&cache
->lru
, &nfsi
->access_cache_entry_lru
);
1819 spin_unlock(&inode
->i_lock
);
1822 rb_erase(&cache
->rb_node
, &nfsi
->access_cache
);
1823 list_del(&cache
->lru
);
1824 spin_unlock(&inode
->i_lock
);
1825 nfs_access_free_entry(cache
);
1828 /* This will release the spinlock */
1829 __nfs_access_zap_cache(inode
);
1833 static void nfs_access_add_rbtree(struct inode
*inode
, struct nfs_access_entry
*set
)
1835 struct nfs_inode
*nfsi
= NFS_I(inode
);
1836 struct rb_root
*root_node
= &nfsi
->access_cache
;
1837 struct rb_node
**p
= &root_node
->rb_node
;
1838 struct rb_node
*parent
= NULL
;
1839 struct nfs_access_entry
*entry
;
1841 spin_lock(&inode
->i_lock
);
1842 while (*p
!= NULL
) {
1844 entry
= rb_entry(parent
, struct nfs_access_entry
, rb_node
);
1846 if (set
->cred
< entry
->cred
)
1847 p
= &parent
->rb_left
;
1848 else if (set
->cred
> entry
->cred
)
1849 p
= &parent
->rb_right
;
1853 rb_link_node(&set
->rb_node
, parent
, p
);
1854 rb_insert_color(&set
->rb_node
, root_node
);
1855 list_add_tail(&set
->lru
, &nfsi
->access_cache_entry_lru
);
1856 spin_unlock(&inode
->i_lock
);
1859 rb_replace_node(parent
, &set
->rb_node
, root_node
);
1860 list_add_tail(&set
->lru
, &nfsi
->access_cache_entry_lru
);
1861 list_del(&entry
->lru
);
1862 spin_unlock(&inode
->i_lock
);
1863 nfs_access_free_entry(entry
);
1866 void nfs_access_add_cache(struct inode
*inode
, struct nfs_access_entry
*set
)
1868 struct nfs_access_entry
*cache
= kmalloc(sizeof(*cache
), GFP_KERNEL
);
1871 RB_CLEAR_NODE(&cache
->rb_node
);
1872 cache
->jiffies
= set
->jiffies
;
1873 cache
->cred
= get_rpccred(set
->cred
);
1874 cache
->mask
= set
->mask
;
1876 nfs_access_add_rbtree(inode
, cache
);
1878 /* Update accounting */
1879 smp_mb__before_atomic_inc();
1880 atomic_long_inc(&nfs_access_nr_entries
);
1881 smp_mb__after_atomic_inc();
1883 /* Add inode to global LRU list */
1884 if (!test_and_set_bit(NFS_INO_ACL_LRU_SET
, &NFS_FLAGS(inode
))) {
1885 spin_lock(&nfs_access_lru_lock
);
1886 list_add_tail(&NFS_I(inode
)->access_cache_inode_lru
, &nfs_access_lru_list
);
1887 spin_unlock(&nfs_access_lru_lock
);
1891 static int nfs_do_access(struct inode
*inode
, struct rpc_cred
*cred
, int mask
)
1893 struct nfs_access_entry cache
;
1896 status
= nfs_access_get_cached(inode
, cred
, &cache
);
1900 /* Be clever: ask server to check for all possible rights */
1901 cache
.mask
= MAY_EXEC
| MAY_WRITE
| MAY_READ
;
1903 cache
.jiffies
= jiffies
;
1904 status
= NFS_PROTO(inode
)->access(inode
, &cache
);
1907 nfs_access_add_cache(inode
, &cache
);
1909 if ((cache
.mask
& mask
) == mask
)
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