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"
43 /* #define NFS_DEBUG_VERBOSE 1 */
45 static int nfs_opendir(struct inode
*, struct file
*);
46 static int nfs_readdir(struct file
*, void *, filldir_t
);
47 static struct dentry
*nfs_lookup(struct inode
*, struct dentry
*, struct nameidata
*);
48 static int nfs_create(struct inode
*, struct dentry
*, int, struct nameidata
*);
49 static int nfs_mkdir(struct inode
*, struct dentry
*, int);
50 static int nfs_rmdir(struct inode
*, struct dentry
*);
51 static int nfs_unlink(struct inode
*, struct dentry
*);
52 static int nfs_symlink(struct inode
*, struct dentry
*, const char *);
53 static int nfs_link(struct dentry
*, struct inode
*, struct dentry
*);
54 static int nfs_mknod(struct inode
*, struct dentry
*, int, dev_t
);
55 static int nfs_rename(struct inode
*, struct dentry
*,
56 struct inode
*, struct dentry
*);
57 static int nfs_fsync_dir(struct file
*, struct dentry
*, int);
58 static loff_t
nfs_llseek_dir(struct file
*, loff_t
, int);
60 const struct file_operations nfs_dir_operations
= {
61 .llseek
= nfs_llseek_dir
,
62 .read
= generic_read_dir
,
63 .readdir
= nfs_readdir
,
65 .release
= nfs_release
,
66 .fsync
= nfs_fsync_dir
,
69 const struct inode_operations nfs_dir_inode_operations
= {
74 .symlink
= nfs_symlink
,
79 .permission
= nfs_permission
,
80 .getattr
= nfs_getattr
,
81 .setattr
= nfs_setattr
,
85 const struct inode_operations nfs3_dir_inode_operations
= {
90 .symlink
= nfs_symlink
,
95 .permission
= nfs_permission
,
96 .getattr
= nfs_getattr
,
97 .setattr
= nfs_setattr
,
98 .listxattr
= nfs3_listxattr
,
99 .getxattr
= nfs3_getxattr
,
100 .setxattr
= nfs3_setxattr
,
101 .removexattr
= nfs3_removexattr
,
103 #endif /* CONFIG_NFS_V3 */
107 static struct dentry
*nfs_atomic_lookup(struct inode
*, struct dentry
*, struct nameidata
*);
108 const struct inode_operations nfs4_dir_inode_operations
= {
109 .create
= nfs_create
,
110 .lookup
= nfs_atomic_lookup
,
112 .unlink
= nfs_unlink
,
113 .symlink
= nfs_symlink
,
117 .rename
= nfs_rename
,
118 .permission
= nfs_permission
,
119 .getattr
= nfs_getattr
,
120 .setattr
= nfs_setattr
,
121 .getxattr
= nfs4_getxattr
,
122 .setxattr
= nfs4_setxattr
,
123 .listxattr
= nfs4_listxattr
,
126 #endif /* CONFIG_NFS_V4 */
132 nfs_opendir(struct inode
*inode
, struct file
*filp
)
136 dfprintk(FILE, "NFS: open dir(%s/%s)\n",
137 filp
->f_path
.dentry
->d_parent
->d_name
.name
,
138 filp
->f_path
.dentry
->d_name
.name
);
140 nfs_inc_stats(inode
, NFSIOS_VFSOPEN
);
142 /* Call generic open code in order to cache credentials */
143 res
= nfs_open(inode
, filp
);
147 typedef __be32
* (*decode_dirent_t
)(__be32
*, struct nfs_entry
*, int);
151 unsigned long page_index
;
154 loff_t current_index
;
155 struct nfs_entry
*entry
;
156 decode_dirent_t decode
;
158 unsigned long timestamp
;
160 } nfs_readdir_descriptor_t
;
162 /* Now we cache directories properly, by stuffing the dirent
163 * data directly in the page cache.
165 * Inode invalidation due to refresh etc. takes care of
166 * _everything_, no sloppy entry flushing logic, no extraneous
167 * copying, network direct to page cache, the way it was meant
170 * NOTE: Dirent information verification is done always by the
171 * page-in of the RPC reply, nowhere else, this simplies
172 * things substantially.
175 int nfs_readdir_filler(nfs_readdir_descriptor_t
*desc
, struct page
*page
)
177 struct file
*file
= desc
->file
;
178 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
179 struct rpc_cred
*cred
= nfs_file_cred(file
);
180 unsigned long timestamp
;
183 dfprintk(DIRCACHE
, "NFS: %s: reading cookie %Lu into page %lu\n",
184 __func__
, (long long)desc
->entry
->cookie
,
189 error
= NFS_PROTO(inode
)->readdir(file
->f_path
.dentry
, cred
, desc
->entry
->cookie
, page
,
190 NFS_SERVER(inode
)->dtsize
, desc
->plus
);
192 /* We requested READDIRPLUS, but the server doesn't grok it */
193 if (error
== -ENOTSUPP
&& desc
->plus
) {
194 NFS_SERVER(inode
)->caps
&= ~NFS_CAP_READDIRPLUS
;
195 clear_bit(NFS_INO_ADVISE_RDPLUS
, &NFS_I(inode
)->flags
);
201 desc
->timestamp
= timestamp
;
202 desc
->timestamp_valid
= 1;
203 SetPageUptodate(page
);
204 /* Ensure consistent page alignment of the data.
205 * Note: assumes we have exclusive access to this mapping either
206 * through inode->i_mutex or some other mechanism.
208 if (invalidate_inode_pages2_range(inode
->i_mapping
, page
->index
+ 1, -1) < 0) {
209 /* Should never happen */
210 nfs_zap_mapping(inode
, inode
->i_mapping
);
220 int dir_decode(nfs_readdir_descriptor_t
*desc
)
222 __be32
*p
= desc
->ptr
;
223 p
= desc
->decode(p
, desc
->entry
, desc
->plus
);
227 if (desc
->timestamp_valid
)
228 desc
->entry
->fattr
->time_start
= desc
->timestamp
;
230 desc
->entry
->fattr
->valid
&= ~NFS_ATTR_FATTR
;
235 void dir_page_release(nfs_readdir_descriptor_t
*desc
)
238 page_cache_release(desc
->page
);
244 * Given a pointer to a buffer that has already been filled by a call
245 * to readdir, find the next entry with cookie '*desc->dir_cookie'.
247 * If the end of the buffer has been reached, return -EAGAIN, if not,
248 * return the offset within the buffer of the next entry to be
252 int find_dirent(nfs_readdir_descriptor_t
*desc
)
254 struct nfs_entry
*entry
= desc
->entry
;
258 while((status
= dir_decode(desc
)) == 0) {
259 dfprintk(DIRCACHE
, "NFS: %s: examining cookie %Lu\n",
260 __func__
, (unsigned long long)entry
->cookie
);
261 if (entry
->prev_cookie
== *desc
->dir_cookie
)
263 if (loop_count
++ > 200) {
272 * Given a pointer to a buffer that has already been filled by a call
273 * to readdir, find the entry at offset 'desc->file->f_pos'.
275 * If the end of the buffer has been reached, return -EAGAIN, if not,
276 * return the offset within the buffer of the next entry to be
280 int find_dirent_index(nfs_readdir_descriptor_t
*desc
)
282 struct nfs_entry
*entry
= desc
->entry
;
287 status
= dir_decode(desc
);
291 dfprintk(DIRCACHE
, "NFS: found cookie %Lu at index %Ld\n",
292 (unsigned long long)entry
->cookie
, desc
->current_index
);
294 if (desc
->file
->f_pos
== desc
->current_index
) {
295 *desc
->dir_cookie
= entry
->cookie
;
298 desc
->current_index
++;
299 if (loop_count
++ > 200) {
308 * Find the given page, and call find_dirent() or find_dirent_index in
309 * order to try to return the next entry.
312 int find_dirent_page(nfs_readdir_descriptor_t
*desc
)
314 struct inode
*inode
= desc
->file
->f_path
.dentry
->d_inode
;
318 dfprintk(DIRCACHE
, "NFS: %s: searching page %ld for target %Lu\n",
319 __func__
, desc
->page_index
,
320 (long long) *desc
->dir_cookie
);
322 /* If we find the page in the page_cache, we cannot be sure
323 * how fresh the data is, so we will ignore readdir_plus attributes.
325 desc
->timestamp_valid
= 0;
326 page
= read_cache_page(inode
->i_mapping
, desc
->page_index
,
327 (filler_t
*)nfs_readdir_filler
, desc
);
329 status
= PTR_ERR(page
);
333 /* NOTE: Someone else may have changed the READDIRPLUS flag */
335 desc
->ptr
= kmap(page
); /* matching kunmap in nfs_do_filldir */
336 if (*desc
->dir_cookie
!= 0)
337 status
= find_dirent(desc
);
339 status
= find_dirent_index(desc
);
341 dir_page_release(desc
);
343 dfprintk(DIRCACHE
, "NFS: %s: returns %d\n", __func__
, status
);
348 * Recurse through the page cache pages, and return a
349 * filled nfs_entry structure of the next directory entry if possible.
351 * The target for the search is '*desc->dir_cookie' if non-0,
352 * 'desc->file->f_pos' otherwise
355 int readdir_search_pagecache(nfs_readdir_descriptor_t
*desc
)
360 /* Always search-by-index from the beginning of the cache */
361 if (*desc
->dir_cookie
== 0) {
362 dfprintk(DIRCACHE
, "NFS: readdir_search_pagecache() searching for offset %Ld\n",
363 (long long)desc
->file
->f_pos
);
364 desc
->page_index
= 0;
365 desc
->entry
->cookie
= desc
->entry
->prev_cookie
= 0;
366 desc
->entry
->eof
= 0;
367 desc
->current_index
= 0;
369 dfprintk(DIRCACHE
, "NFS: readdir_search_pagecache() searching for cookie %Lu\n",
370 (unsigned long long)*desc
->dir_cookie
);
373 res
= find_dirent_page(desc
);
376 /* Align to beginning of next page */
378 if (loop_count
++ > 200) {
384 dfprintk(DIRCACHE
, "NFS: %s: returns %d\n", __func__
, res
);
388 static inline unsigned int dt_type(struct inode
*inode
)
390 return (inode
->i_mode
>> 12) & 15;
393 static struct dentry
*nfs_readdir_lookup(nfs_readdir_descriptor_t
*desc
);
396 * Once we've found the start of the dirent within a page: fill 'er up...
399 int nfs_do_filldir(nfs_readdir_descriptor_t
*desc
, void *dirent
,
402 struct file
*file
= desc
->file
;
403 struct nfs_entry
*entry
= desc
->entry
;
404 struct dentry
*dentry
= NULL
;
409 dfprintk(DIRCACHE
, "NFS: nfs_do_filldir() filling starting @ cookie %Lu\n",
410 (unsigned long long)entry
->cookie
);
413 unsigned d_type
= DT_UNKNOWN
;
414 /* Note: entry->prev_cookie contains the cookie for
415 * retrieving the current dirent on the server */
418 /* Get a dentry if we have one */
421 dentry
= nfs_readdir_lookup(desc
);
423 /* Use readdirplus info */
424 if (dentry
!= NULL
&& dentry
->d_inode
!= NULL
) {
425 d_type
= dt_type(dentry
->d_inode
);
426 fileid
= NFS_FILEID(dentry
->d_inode
);
429 res
= filldir(dirent
, entry
->name
, entry
->len
,
430 file
->f_pos
, nfs_compat_user_ino64(fileid
),
435 *desc
->dir_cookie
= entry
->cookie
;
436 if (dir_decode(desc
) != 0) {
440 if (loop_count
++ > 200) {
445 dir_page_release(desc
);
448 dfprintk(DIRCACHE
, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n",
449 (unsigned long long)*desc
->dir_cookie
, res
);
454 * If we cannot find a cookie in our cache, we suspect that this is
455 * because it points to a deleted file, so we ask the server to return
456 * whatever it thinks is the next entry. We then feed this to filldir.
457 * If all goes well, we should then be able to find our way round the
458 * cache on the next call to readdir_search_pagecache();
460 * NOTE: we cannot add the anonymous page to the pagecache because
461 * the data it contains might not be page aligned. Besides,
462 * we should already have a complete representation of the
463 * directory in the page cache by the time we get here.
466 int uncached_readdir(nfs_readdir_descriptor_t
*desc
, void *dirent
,
469 struct file
*file
= desc
->file
;
470 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
471 struct rpc_cred
*cred
= nfs_file_cred(file
);
472 struct page
*page
= NULL
;
474 unsigned long timestamp
;
476 dfprintk(DIRCACHE
, "NFS: uncached_readdir() searching for cookie %Lu\n",
477 (unsigned long long)*desc
->dir_cookie
);
479 page
= alloc_page(GFP_HIGHUSER
);
485 status
= NFS_PROTO(inode
)->readdir(file
->f_path
.dentry
, cred
,
486 *desc
->dir_cookie
, page
,
487 NFS_SERVER(inode
)->dtsize
,
490 desc
->ptr
= kmap(page
); /* matching kunmap in nfs_do_filldir */
492 desc
->timestamp
= timestamp
;
493 desc
->timestamp_valid
= 1;
494 if ((status
= dir_decode(desc
)) == 0)
495 desc
->entry
->prev_cookie
= *desc
->dir_cookie
;
501 status
= nfs_do_filldir(desc
, dirent
, filldir
);
503 /* Reset read descriptor so it searches the page cache from
504 * the start upon the next call to readdir_search_pagecache() */
505 desc
->page_index
= 0;
506 desc
->entry
->cookie
= desc
->entry
->prev_cookie
= 0;
507 desc
->entry
->eof
= 0;
509 dfprintk(DIRCACHE
, "NFS: %s: returns %d\n",
513 dir_page_release(desc
);
517 /* The file offset position represents the dirent entry number. A
518 last cookie cache takes care of the common case of reading the
521 static int nfs_readdir(struct file
*filp
, void *dirent
, filldir_t filldir
)
523 struct dentry
*dentry
= filp
->f_path
.dentry
;
524 struct inode
*inode
= dentry
->d_inode
;
525 nfs_readdir_descriptor_t my_desc
,
527 struct nfs_entry my_entry
;
529 struct nfs_fattr fattr
;
532 dfprintk(FILE, "NFS: readdir(%s/%s) starting at cookie %llu\n",
533 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
,
534 (long long)filp
->f_pos
);
535 nfs_inc_stats(inode
, NFSIOS_VFSGETDENTS
);
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
= &nfs_file_open_context(filp
)->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 nfs_block_sillyrename(dentry
);
560 res
= nfs_revalidate_mapping_nolock(inode
, filp
->f_mapping
);
564 while(!desc
->entry
->eof
) {
565 res
= readdir_search_pagecache(desc
);
567 if (res
== -EBADCOOKIE
) {
568 /* This means either end of directory */
569 if (*desc
->dir_cookie
&& desc
->entry
->cookie
!= *desc
->dir_cookie
) {
570 /* Or that the server has 'lost' a cookie */
571 res
= uncached_readdir(desc
, dirent
, filldir
);
578 if (res
== -ETOOSMALL
&& desc
->plus
) {
579 clear_bit(NFS_INO_ADVISE_RDPLUS
, &NFS_I(inode
)->flags
);
580 nfs_zap_caches(inode
);
582 desc
->entry
->eof
= 0;
588 res
= nfs_do_filldir(desc
, dirent
, filldir
);
595 nfs_unblock_sillyrename(dentry
);
599 dfprintk(FILE, "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 struct dentry
*dentry
= filp
->f_path
.dentry
;
608 struct inode
*inode
= dentry
->d_inode
;
610 dfprintk(FILE, "NFS: llseek dir(%s/%s, %lld, %d)\n",
611 dentry
->d_parent
->d_name
.name
,
615 mutex_lock(&inode
->i_mutex
);
618 offset
+= filp
->f_pos
;
626 if (offset
!= filp
->f_pos
) {
627 filp
->f_pos
= offset
;
628 nfs_file_open_context(filp
)->dir_cookie
= 0;
631 mutex_unlock(&inode
->i_mutex
);
636 * All directory operations under NFS are synchronous, so fsync()
637 * is a dummy operation.
639 static int nfs_fsync_dir(struct file
*filp
, struct dentry
*dentry
, int datasync
)
641 dfprintk(FILE, "NFS: fsync dir(%s/%s) datasync %d\n",
642 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
,
645 nfs_inc_stats(dentry
->d_inode
, NFSIOS_VFSFSYNC
);
650 * nfs_force_lookup_revalidate - Mark the directory as having changed
651 * @dir - pointer to directory inode
653 * This forces the revalidation code in nfs_lookup_revalidate() to do a
654 * full lookup on all child dentries of 'dir' whenever a change occurs
655 * on the server that might have invalidated our dcache.
657 * The caller should be holding dir->i_lock
659 void nfs_force_lookup_revalidate(struct inode
*dir
)
661 NFS_I(dir
)->cache_change_attribute
= jiffies
;
665 * A check for whether or not the parent directory has changed.
666 * In the case it has, we assume that the dentries are untrustworthy
667 * and may need to be looked up again.
669 static int nfs_check_verifier(struct inode
*dir
, struct dentry
*dentry
)
673 if (!nfs_verify_change_attribute(dir
, dentry
->d_time
))
675 /* Revalidate nfsi->cache_change_attribute before we declare a match */
676 if (nfs_revalidate_inode(NFS_SERVER(dir
), dir
) < 0)
678 if (!nfs_verify_change_attribute(dir
, dentry
->d_time
))
684 * Return the intent data that applies to this particular path component
686 * Note that the current set of intents only apply to the very last
687 * component of the path.
688 * We check for this using LOOKUP_CONTINUE and LOOKUP_PARENT.
690 static inline unsigned int nfs_lookup_check_intent(struct nameidata
*nd
, unsigned int mask
)
692 if (nd
->flags
& (LOOKUP_CONTINUE
|LOOKUP_PARENT
))
694 return nd
->flags
& mask
;
698 * Use intent information to check whether or not we're going to do
699 * an O_EXCL create using this path component.
701 static int nfs_is_exclusive_create(struct inode
*dir
, struct nameidata
*nd
)
703 if (NFS_PROTO(dir
)->version
== 2)
705 if (nd
== NULL
|| nfs_lookup_check_intent(nd
, LOOKUP_CREATE
) == 0)
707 return (nd
->intent
.open
.flags
& O_EXCL
) != 0;
711 * Inode and filehandle revalidation for lookups.
713 * We force revalidation in the cases where the VFS sets LOOKUP_REVAL,
714 * or if the intent information indicates that we're about to open this
715 * particular file and the "nocto" mount flag is not set.
719 int nfs_lookup_verify_inode(struct inode
*inode
, struct nameidata
*nd
)
721 struct nfs_server
*server
= NFS_SERVER(inode
);
723 if (test_bit(NFS_INO_MOUNTPOINT
, &NFS_I(inode
)->flags
))
726 /* VFS wants an on-the-wire revalidation */
727 if (nd
->flags
& LOOKUP_REVAL
)
729 /* This is an open(2) */
730 if (nfs_lookup_check_intent(nd
, LOOKUP_OPEN
) != 0 &&
731 !(server
->flags
& NFS_MOUNT_NOCTO
) &&
732 (S_ISREG(inode
->i_mode
) ||
733 S_ISDIR(inode
->i_mode
)))
737 return nfs_revalidate_inode(server
, inode
);
739 return __nfs_revalidate_inode(server
, inode
);
743 * We judge how long we want to trust negative
744 * dentries by looking at the parent inode mtime.
746 * If parent mtime has changed, we revalidate, else we wait for a
747 * period corresponding to the parent's attribute cache timeout value.
750 int nfs_neg_need_reval(struct inode
*dir
, struct dentry
*dentry
,
751 struct nameidata
*nd
)
753 /* Don't revalidate a negative dentry if we're creating a new file */
754 if (nd
!= NULL
&& nfs_lookup_check_intent(nd
, LOOKUP_CREATE
) != 0)
756 return !nfs_check_verifier(dir
, dentry
);
760 * This is called every time the dcache has a lookup hit,
761 * and we should check whether we can really trust that
764 * NOTE! The hit can be a negative hit too, don't assume
767 * If the parent directory is seen to have changed, we throw out the
768 * cached dentry and do a new lookup.
770 static int nfs_lookup_revalidate(struct dentry
* dentry
, struct nameidata
*nd
)
774 struct dentry
*parent
;
776 struct nfs_fh fhandle
;
777 struct nfs_fattr fattr
;
779 parent
= dget_parent(dentry
);
780 dir
= parent
->d_inode
;
781 nfs_inc_stats(dir
, NFSIOS_DENTRYREVALIDATE
);
782 inode
= dentry
->d_inode
;
785 if (nfs_neg_need_reval(dir
, dentry
, nd
))
790 if (is_bad_inode(inode
)) {
791 dfprintk(LOOKUPCACHE
, "%s: %s/%s has dud inode\n",
792 __func__
, dentry
->d_parent
->d_name
.name
,
793 dentry
->d_name
.name
);
797 /* Force a full look up iff the parent directory has changed */
798 if (!nfs_is_exclusive_create(dir
, nd
) && nfs_check_verifier(dir
, dentry
)) {
799 if (nfs_lookup_verify_inode(inode
, nd
))
804 if (NFS_STALE(inode
))
807 error
= NFS_PROTO(dir
)->lookup(dir
, &dentry
->d_name
, &fhandle
, &fattr
);
810 if (nfs_compare_fh(NFS_FH(inode
), &fhandle
))
812 if ((error
= nfs_refresh_inode(inode
, &fattr
)) != 0)
815 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
818 dfprintk(LOOKUPCACHE
, "NFS: %s(%s/%s) is valid\n",
819 __func__
, dentry
->d_parent
->d_name
.name
,
820 dentry
->d_name
.name
);
825 nfs_mark_for_revalidate(dir
);
826 if (inode
&& S_ISDIR(inode
->i_mode
)) {
827 /* Purge readdir caches. */
828 nfs_zap_caches(inode
);
829 /* If we have submounts, don't unhash ! */
830 if (have_submounts(dentry
))
832 shrink_dcache_parent(dentry
);
836 dfprintk(LOOKUPCACHE
, "NFS: %s(%s/%s) is invalid\n",
837 __func__
, dentry
->d_parent
->d_name
.name
,
838 dentry
->d_name
.name
);
843 * This is called from dput() when d_count is going to 0.
845 static int nfs_dentry_delete(struct dentry
*dentry
)
847 dfprintk(VFS
, "NFS: dentry_delete(%s/%s, %x)\n",
848 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
,
851 /* Unhash any dentry with a stale inode */
852 if (dentry
->d_inode
!= NULL
&& NFS_STALE(dentry
->d_inode
))
855 if (dentry
->d_flags
& DCACHE_NFSFS_RENAMED
) {
856 /* Unhash it, so that ->d_iput() would be called */
859 if (!(dentry
->d_sb
->s_flags
& MS_ACTIVE
)) {
860 /* Unhash it, so that ancestors of killed async unlink
861 * files will be cleaned up during umount */
868 static void nfs_drop_nlink(struct inode
*inode
)
870 spin_lock(&inode
->i_lock
);
871 if (inode
->i_nlink
> 0)
873 spin_unlock(&inode
->i_lock
);
877 * Called when the dentry loses inode.
878 * We use it to clean up silly-renamed files.
880 static void nfs_dentry_iput(struct dentry
*dentry
, struct inode
*inode
)
882 if (S_ISDIR(inode
->i_mode
))
883 /* drop any readdir cache as it could easily be old */
884 NFS_I(inode
)->cache_validity
|= NFS_INO_INVALID_DATA
;
886 if (dentry
->d_flags
& DCACHE_NFSFS_RENAMED
) {
889 nfs_complete_unlink(dentry
, inode
);
895 struct dentry_operations nfs_dentry_operations
= {
896 .d_revalidate
= nfs_lookup_revalidate
,
897 .d_delete
= nfs_dentry_delete
,
898 .d_iput
= nfs_dentry_iput
,
901 static struct dentry
*nfs_lookup(struct inode
*dir
, struct dentry
* dentry
, struct nameidata
*nd
)
904 struct dentry
*parent
;
905 struct inode
*inode
= NULL
;
907 struct nfs_fh fhandle
;
908 struct nfs_fattr fattr
;
910 dfprintk(VFS
, "NFS: lookup(%s/%s)\n",
911 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
);
912 nfs_inc_stats(dir
, NFSIOS_VFSLOOKUP
);
914 res
= ERR_PTR(-ENAMETOOLONG
);
915 if (dentry
->d_name
.len
> NFS_SERVER(dir
)->namelen
)
918 res
= ERR_PTR(-ENOMEM
);
919 dentry
->d_op
= NFS_PROTO(dir
)->dentry_ops
;
922 * If we're doing an exclusive create, optimize away the lookup
923 * but don't hash the dentry.
925 if (nfs_is_exclusive_create(dir
, nd
)) {
926 d_instantiate(dentry
, NULL
);
931 parent
= dentry
->d_parent
;
932 /* Protect against concurrent sillydeletes */
933 nfs_block_sillyrename(parent
);
934 error
= NFS_PROTO(dir
)->lookup(dir
, &dentry
->d_name
, &fhandle
, &fattr
);
935 if (error
== -ENOENT
)
938 res
= ERR_PTR(error
);
939 goto out_unblock_sillyrename
;
941 inode
= nfs_fhget(dentry
->d_sb
, &fhandle
, &fattr
);
942 res
= (struct dentry
*)inode
;
944 goto out_unblock_sillyrename
;
947 res
= d_materialise_unique(dentry
, inode
);
950 goto out_unblock_sillyrename
;
953 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
954 out_unblock_sillyrename
:
955 nfs_unblock_sillyrename(parent
);
961 static int nfs_open_revalidate(struct dentry
*, struct nameidata
*);
963 struct dentry_operations nfs4_dentry_operations
= {
964 .d_revalidate
= nfs_open_revalidate
,
965 .d_delete
= nfs_dentry_delete
,
966 .d_iput
= nfs_dentry_iput
,
970 * Use intent information to determine whether we need to substitute
971 * the NFSv4-style stateful OPEN for the LOOKUP call
973 static int is_atomic_open(struct inode
*dir
, struct nameidata
*nd
)
975 if (nd
== NULL
|| nfs_lookup_check_intent(nd
, LOOKUP_OPEN
) == 0)
977 /* NFS does not (yet) have a stateful open for directories */
978 if (nd
->flags
& LOOKUP_DIRECTORY
)
980 /* Are we trying to write to a read only partition? */
981 if (__mnt_is_readonly(nd
->path
.mnt
) &&
982 (nd
->intent
.open
.flags
& (O_CREAT
|O_TRUNC
|FMODE_WRITE
)))
987 static struct dentry
*nfs_atomic_lookup(struct inode
*dir
, struct dentry
*dentry
, struct nameidata
*nd
)
989 struct dentry
*res
= NULL
;
992 dfprintk(VFS
, "NFS: atomic_lookup(%s/%ld), %s\n",
993 dir
->i_sb
->s_id
, dir
->i_ino
, dentry
->d_name
.name
);
995 /* Check that we are indeed trying to open this file */
996 if (!is_atomic_open(dir
, nd
))
999 if (dentry
->d_name
.len
> NFS_SERVER(dir
)->namelen
) {
1000 res
= ERR_PTR(-ENAMETOOLONG
);
1003 dentry
->d_op
= NFS_PROTO(dir
)->dentry_ops
;
1005 /* Let vfs_create() deal with O_EXCL. Instantiate, but don't hash
1007 if (nd
->intent
.open
.flags
& O_EXCL
) {
1008 d_instantiate(dentry
, NULL
);
1012 /* Open the file on the server */
1013 res
= nfs4_atomic_open(dir
, dentry
, nd
);
1015 error
= PTR_ERR(res
);
1017 /* Make a negative dentry */
1021 /* This turned out not to be a regular file */
1026 if (!(nd
->intent
.open
.flags
& O_NOFOLLOW
))
1032 } else if (res
!= NULL
)
1037 return nfs_lookup(dir
, dentry
, nd
);
1040 static int nfs_open_revalidate(struct dentry
*dentry
, struct nameidata
*nd
)
1042 struct dentry
*parent
= NULL
;
1043 struct inode
*inode
= dentry
->d_inode
;
1045 int openflags
, ret
= 0;
1047 parent
= dget_parent(dentry
);
1048 dir
= parent
->d_inode
;
1049 if (!is_atomic_open(dir
, nd
))
1051 /* We can't create new files in nfs_open_revalidate(), so we
1052 * optimize away revalidation of negative dentries.
1054 if (inode
== NULL
) {
1055 if (!nfs_neg_need_reval(dir
, dentry
, nd
))
1060 /* NFS only supports OPEN on regular files */
1061 if (!S_ISREG(inode
->i_mode
))
1063 openflags
= nd
->intent
.open
.flags
;
1064 /* We cannot do exclusive creation on a positive dentry */
1065 if ((openflags
& (O_CREAT
|O_EXCL
)) == (O_CREAT
|O_EXCL
))
1067 /* We can't create new files, or truncate existing ones here */
1068 openflags
&= ~(O_CREAT
|O_TRUNC
);
1071 * Note: we're not holding inode->i_mutex and so may be racing with
1072 * operations that change the directory. We therefore save the
1073 * change attribute *before* we do the RPC call.
1075 ret
= nfs4_open_revalidate(dir
, dentry
, openflags
, nd
);
1083 if (inode
!= NULL
&& nfs_have_delegation(inode
, FMODE_READ
))
1085 return nfs_lookup_revalidate(dentry
, nd
);
1087 #endif /* CONFIG_NFSV4 */
1089 static struct dentry
*nfs_readdir_lookup(nfs_readdir_descriptor_t
*desc
)
1091 struct dentry
*parent
= desc
->file
->f_path
.dentry
;
1092 struct inode
*dir
= parent
->d_inode
;
1093 struct nfs_entry
*entry
= desc
->entry
;
1094 struct dentry
*dentry
, *alias
;
1095 struct qstr name
= {
1096 .name
= entry
->name
,
1099 struct inode
*inode
;
1100 unsigned long verf
= nfs_save_change_attribute(dir
);
1104 if (name
.name
[0] == '.' && name
.name
[1] == '.')
1105 return dget_parent(parent
);
1108 if (name
.name
[0] == '.')
1109 return dget(parent
);
1112 spin_lock(&dir
->i_lock
);
1113 if (NFS_I(dir
)->cache_validity
& NFS_INO_INVALID_DATA
) {
1114 spin_unlock(&dir
->i_lock
);
1117 spin_unlock(&dir
->i_lock
);
1119 name
.hash
= full_name_hash(name
.name
, name
.len
);
1120 dentry
= d_lookup(parent
, &name
);
1121 if (dentry
!= NULL
) {
1122 /* Is this a positive dentry that matches the readdir info? */
1123 if (dentry
->d_inode
!= NULL
&&
1124 (NFS_FILEID(dentry
->d_inode
) == entry
->ino
||
1125 d_mountpoint(dentry
))) {
1126 if (!desc
->plus
|| entry
->fh
->size
== 0)
1128 if (nfs_compare_fh(NFS_FH(dentry
->d_inode
),
1132 /* No, so d_drop to allow one to be created */
1136 if (!desc
->plus
|| !(entry
->fattr
->valid
& NFS_ATTR_FATTR
))
1138 if (name
.len
> NFS_SERVER(dir
)->namelen
)
1140 /* Note: caller is already holding the dir->i_mutex! */
1141 dentry
= d_alloc(parent
, &name
);
1144 dentry
->d_op
= NFS_PROTO(dir
)->dentry_ops
;
1145 inode
= nfs_fhget(dentry
->d_sb
, entry
->fh
, entry
->fattr
);
1146 if (IS_ERR(inode
)) {
1151 alias
= d_materialise_unique(dentry
, inode
);
1152 if (alias
!= NULL
) {
1160 nfs_set_verifier(dentry
, verf
);
1165 * Code common to create, mkdir, and mknod.
1167 int nfs_instantiate(struct dentry
*dentry
, struct nfs_fh
*fhandle
,
1168 struct nfs_fattr
*fattr
)
1170 struct dentry
*parent
= dget_parent(dentry
);
1171 struct inode
*dir
= parent
->d_inode
;
1172 struct inode
*inode
;
1173 int error
= -EACCES
;
1177 /* We may have been initialized further down */
1178 if (dentry
->d_inode
)
1180 if (fhandle
->size
== 0) {
1181 error
= NFS_PROTO(dir
)->lookup(dir
, &dentry
->d_name
, fhandle
, fattr
);
1185 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1186 if (!(fattr
->valid
& NFS_ATTR_FATTR
)) {
1187 struct nfs_server
*server
= NFS_SB(dentry
->d_sb
);
1188 error
= server
->nfs_client
->rpc_ops
->getattr(server
, fhandle
, fattr
);
1192 inode
= nfs_fhget(dentry
->d_sb
, fhandle
, fattr
);
1193 error
= PTR_ERR(inode
);
1196 d_add(dentry
, inode
);
1201 nfs_mark_for_revalidate(dir
);
1207 * Following a failed create operation, we drop the dentry rather
1208 * than retain a negative dentry. This avoids a problem in the event
1209 * that the operation succeeded on the server, but an error in the
1210 * reply path made it appear to have failed.
1212 static int nfs_create(struct inode
*dir
, struct dentry
*dentry
, int mode
,
1213 struct nameidata
*nd
)
1219 dfprintk(VFS
, "NFS: create(%s/%ld), %s\n",
1220 dir
->i_sb
->s_id
, dir
->i_ino
, dentry
->d_name
.name
);
1222 attr
.ia_mode
= mode
;
1223 attr
.ia_valid
= ATTR_MODE
;
1225 if ((nd
->flags
& LOOKUP_CREATE
) != 0)
1226 open_flags
= nd
->intent
.open
.flags
;
1228 error
= NFS_PROTO(dir
)->create(dir
, dentry
, &attr
, open_flags
, nd
);
1238 * See comments for nfs_proc_create regarding failed operations.
1241 nfs_mknod(struct inode
*dir
, struct dentry
*dentry
, int mode
, dev_t rdev
)
1246 dfprintk(VFS
, "NFS: mknod(%s/%ld), %s\n",
1247 dir
->i_sb
->s_id
, dir
->i_ino
, dentry
->d_name
.name
);
1249 if (!new_valid_dev(rdev
))
1252 attr
.ia_mode
= mode
;
1253 attr
.ia_valid
= ATTR_MODE
;
1255 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
;
1278 error
= NFS_PROTO(dir
)->mkdir(dir
, dentry
, &attr
);
1287 static void nfs_dentry_handle_enoent(struct dentry
*dentry
)
1289 if (dentry
->d_inode
!= NULL
&& !d_unhashed(dentry
))
1293 static int nfs_rmdir(struct inode
*dir
, struct dentry
*dentry
)
1297 dfprintk(VFS
, "NFS: rmdir(%s/%ld), %s\n",
1298 dir
->i_sb
->s_id
, dir
->i_ino
, dentry
->d_name
.name
);
1300 error
= NFS_PROTO(dir
)->rmdir(dir
, &dentry
->d_name
);
1301 /* Ensure the VFS deletes this inode */
1302 if (error
== 0 && dentry
->d_inode
!= NULL
)
1303 clear_nlink(dentry
->d_inode
);
1304 else if (error
== -ENOENT
)
1305 nfs_dentry_handle_enoent(dentry
);
1310 static int nfs_sillyrename(struct inode
*dir
, struct dentry
*dentry
)
1312 static unsigned int sillycounter
;
1313 const int fileidsize
= sizeof(NFS_FILEID(dentry
->d_inode
))*2;
1314 const int countersize
= sizeof(sillycounter
)*2;
1315 const int slen
= sizeof(".nfs")+fileidsize
+countersize
-1;
1318 struct dentry
*sdentry
;
1321 dfprintk(VFS
, "NFS: silly-rename(%s/%s, ct=%d)\n",
1322 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
,
1323 atomic_read(&dentry
->d_count
));
1324 nfs_inc_stats(dir
, NFSIOS_SILLYRENAME
);
1327 * We don't allow a dentry to be silly-renamed twice.
1330 if (dentry
->d_flags
& DCACHE_NFSFS_RENAMED
)
1333 sprintf(silly
, ".nfs%*.*Lx",
1334 fileidsize
, fileidsize
,
1335 (unsigned long long)NFS_FILEID(dentry
->d_inode
));
1337 /* Return delegation in anticipation of the rename */
1338 nfs_inode_return_delegation(dentry
->d_inode
);
1342 char *suffix
= silly
+ slen
- countersize
;
1346 sprintf(suffix
, "%*.*x", countersize
, countersize
, sillycounter
);
1348 dfprintk(VFS
, "NFS: trying to rename %s to %s\n",
1349 dentry
->d_name
.name
, silly
);
1351 sdentry
= lookup_one_len(silly
, dentry
->d_parent
, slen
);
1353 * N.B. Better to return EBUSY here ... it could be
1354 * dangerous to delete the file while it's in use.
1356 if (IS_ERR(sdentry
))
1358 } while(sdentry
->d_inode
!= NULL
); /* need negative lookup */
1360 qsilly
.name
= silly
;
1361 qsilly
.len
= strlen(silly
);
1362 if (dentry
->d_inode
) {
1363 error
= NFS_PROTO(dir
)->rename(dir
, &dentry
->d_name
,
1365 nfs_mark_for_revalidate(dentry
->d_inode
);
1367 error
= NFS_PROTO(dir
)->rename(dir
, &dentry
->d_name
,
1370 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1371 d_move(dentry
, sdentry
);
1372 error
= nfs_async_unlink(dir
, dentry
);
1373 /* If we return 0 we don't unlink */
1381 * Remove a file after making sure there are no pending writes,
1382 * and after checking that the file has only one user.
1384 * We invalidate the attribute cache and free the inode prior to the operation
1385 * to avoid possible races if the server reuses the inode.
1387 static int nfs_safe_remove(struct dentry
*dentry
)
1389 struct inode
*dir
= dentry
->d_parent
->d_inode
;
1390 struct inode
*inode
= dentry
->d_inode
;
1393 dfprintk(VFS
, "NFS: safe_remove(%s/%s)\n",
1394 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
);
1396 /* If the dentry was sillyrenamed, we simply call d_delete() */
1397 if (dentry
->d_flags
& DCACHE_NFSFS_RENAMED
) {
1402 if (inode
!= NULL
) {
1403 nfs_inode_return_delegation(inode
);
1404 error
= NFS_PROTO(dir
)->remove(dir
, &dentry
->d_name
);
1405 /* The VFS may want to delete this inode */
1407 nfs_drop_nlink(inode
);
1408 nfs_mark_for_revalidate(inode
);
1410 error
= NFS_PROTO(dir
)->remove(dir
, &dentry
->d_name
);
1411 if (error
== -ENOENT
)
1412 nfs_dentry_handle_enoent(dentry
);
1417 /* We do silly rename. In case sillyrename() returns -EBUSY, the inode
1418 * belongs to an active ".nfs..." file and we return -EBUSY.
1420 * If sillyrename() returns 0, we do nothing, otherwise we unlink.
1422 static int nfs_unlink(struct inode
*dir
, struct dentry
*dentry
)
1425 int need_rehash
= 0;
1427 dfprintk(VFS
, "NFS: unlink(%s/%ld, %s)\n", dir
->i_sb
->s_id
,
1428 dir
->i_ino
, dentry
->d_name
.name
);
1430 spin_lock(&dcache_lock
);
1431 spin_lock(&dentry
->d_lock
);
1432 if (atomic_read(&dentry
->d_count
) > 1) {
1433 spin_unlock(&dentry
->d_lock
);
1434 spin_unlock(&dcache_lock
);
1435 /* Start asynchronous writeout of the inode */
1436 write_inode_now(dentry
->d_inode
, 0);
1438 error
= nfs_sillyrename(dir
, dentry
);
1442 if (!d_unhashed(dentry
)) {
1446 spin_unlock(&dentry
->d_lock
);
1447 spin_unlock(&dcache_lock
);
1448 error
= nfs_safe_remove(dentry
);
1449 if (!error
|| error
== -ENOENT
) {
1450 nfs_set_verifier(dentry
, nfs_save_change_attribute(dir
));
1451 } else if (need_rehash
)
1457 * To create a symbolic link, most file systems instantiate a new inode,
1458 * add a page to it containing the path, then write it out to the disk
1459 * using prepare_write/commit_write.
1461 * Unfortunately the NFS client can't create the in-core inode first
1462 * because it needs a file handle to create an in-core inode (see
1463 * fs/nfs/inode.c:nfs_fhget). We only have a file handle *after* the
1464 * symlink request has completed on the server.
1466 * So instead we allocate a raw page, copy the symname into it, then do
1467 * the SYMLINK request with the page as the buffer. If it succeeds, we
1468 * now have a new file handle and can instantiate an in-core NFS inode
1469 * and move the raw page into its mapping.
1471 static int nfs_symlink(struct inode
*dir
, struct dentry
*dentry
, const char *symname
)
1473 struct pagevec lru_pvec
;
1477 unsigned int pathlen
= strlen(symname
);
1480 dfprintk(VFS
, "NFS: symlink(%s/%ld, %s, %s)\n", dir
->i_sb
->s_id
,
1481 dir
->i_ino
, dentry
->d_name
.name
, symname
);
1483 if (pathlen
> PAGE_SIZE
)
1484 return -ENAMETOOLONG
;
1486 attr
.ia_mode
= S_IFLNK
| S_IRWXUGO
;
1487 attr
.ia_valid
= ATTR_MODE
;
1491 page
= alloc_page(GFP_HIGHUSER
);
1497 kaddr
= kmap_atomic(page
, KM_USER0
);
1498 memcpy(kaddr
, symname
, pathlen
);
1499 if (pathlen
< PAGE_SIZE
)
1500 memset(kaddr
+ pathlen
, 0, PAGE_SIZE
- pathlen
);
1501 kunmap_atomic(kaddr
, KM_USER0
);
1503 error
= NFS_PROTO(dir
)->symlink(dir
, dentry
, page
, pathlen
, &attr
);
1505 dfprintk(VFS
, "NFS: symlink(%s/%ld, %s, %s) error %d\n",
1506 dir
->i_sb
->s_id
, dir
->i_ino
,
1507 dentry
->d_name
.name
, symname
, error
);
1515 * No big deal if we can't add this page to the page cache here.
1516 * READLINK will get the missing page from the server if needed.
1518 pagevec_init(&lru_pvec
, 0);
1519 if (!add_to_page_cache(page
, dentry
->d_inode
->i_mapping
, 0,
1521 pagevec_add(&lru_pvec
, page
);
1522 pagevec_lru_add(&lru_pvec
);
1523 SetPageUptodate(page
);
1533 nfs_link(struct dentry
*old_dentry
, struct inode
*dir
, struct dentry
*dentry
)
1535 struct inode
*inode
= old_dentry
->d_inode
;
1538 dfprintk(VFS
, "NFS: link(%s/%s -> %s/%s)\n",
1539 old_dentry
->d_parent
->d_name
.name
, old_dentry
->d_name
.name
,
1540 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
);
1543 error
= NFS_PROTO(dir
)->link(inode
, dir
, &dentry
->d_name
);
1545 atomic_inc(&inode
->i_count
);
1546 d_add(dentry
, inode
);
1553 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a
1554 * different file handle for the same inode after a rename (e.g. when
1555 * moving to a different directory). A fail-safe method to do so would
1556 * be to look up old_dir/old_name, create a link to new_dir/new_name and
1557 * rename the old file using the sillyrename stuff. This way, the original
1558 * file in old_dir will go away when the last process iput()s the inode.
1562 * It actually works quite well. One needs to have the possibility for
1563 * at least one ".nfs..." file in each directory the file ever gets
1564 * moved or linked to which happens automagically with the new
1565 * implementation that only depends on the dcache stuff instead of
1566 * using the inode layer
1568 * Unfortunately, things are a little more complicated than indicated
1569 * above. For a cross-directory move, we want to make sure we can get
1570 * rid of the old inode after the operation. This means there must be
1571 * no pending writes (if it's a file), and the use count must be 1.
1572 * If these conditions are met, we can drop the dentries before doing
1575 static int nfs_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
1576 struct inode
*new_dir
, struct dentry
*new_dentry
)
1578 struct inode
*old_inode
= old_dentry
->d_inode
;
1579 struct inode
*new_inode
= new_dentry
->d_inode
;
1580 struct dentry
*dentry
= NULL
, *rehash
= NULL
;
1584 * To prevent any new references to the target during the rename,
1585 * we unhash the dentry and free the inode in advance.
1587 if (!d_unhashed(new_dentry
)) {
1589 rehash
= new_dentry
;
1592 dfprintk(VFS
, "NFS: rename(%s/%s -> %s/%s, ct=%d)\n",
1593 old_dentry
->d_parent
->d_name
.name
, old_dentry
->d_name
.name
,
1594 new_dentry
->d_parent
->d_name
.name
, new_dentry
->d_name
.name
,
1595 atomic_read(&new_dentry
->d_count
));
1598 * First check whether the target is busy ... we can't
1599 * safely do _any_ rename if the target is in use.
1601 * For files, make a copy of the dentry and then do a
1602 * silly-rename. If the silly-rename succeeds, the
1603 * copied dentry is hashed and becomes the new target.
1607 if (S_ISDIR(new_inode
->i_mode
)) {
1609 if (!S_ISDIR(old_inode
->i_mode
))
1611 } else if (atomic_read(&new_dentry
->d_count
) > 2) {
1613 /* copy the target dentry's name */
1614 dentry
= d_alloc(new_dentry
->d_parent
,
1615 &new_dentry
->d_name
);
1619 /* silly-rename the existing target ... */
1621 err
= nfs_sillyrename(new_dir
, new_dentry
);
1624 new_dentry
= rehash
= dentry
;
1626 /* instantiate the replacement target */
1627 d_instantiate(new_dentry
, NULL
);
1628 } else if (atomic_read(&new_dentry
->d_count
) > 1)
1629 /* dentry still busy? */
1632 nfs_drop_nlink(new_inode
);
1636 * ... prune child dentries and writebacks if needed.
1638 if (atomic_read(&old_dentry
->d_count
) > 1) {
1639 if (S_ISREG(old_inode
->i_mode
))
1640 nfs_wb_all(old_inode
);
1641 shrink_dcache_parent(old_dentry
);
1643 nfs_inode_return_delegation(old_inode
);
1645 if (new_inode
!= NULL
) {
1646 nfs_inode_return_delegation(new_inode
);
1647 d_delete(new_dentry
);
1650 error
= NFS_PROTO(old_dir
)->rename(old_dir
, &old_dentry
->d_name
,
1651 new_dir
, &new_dentry
->d_name
);
1652 nfs_mark_for_revalidate(old_inode
);
1657 d_move(old_dentry
, new_dentry
);
1658 nfs_set_verifier(new_dentry
,
1659 nfs_save_change_attribute(new_dir
));
1660 } else if (error
== -ENOENT
)
1661 nfs_dentry_handle_enoent(old_dentry
);
1663 /* new dentry created? */
1669 static DEFINE_SPINLOCK(nfs_access_lru_lock
);
1670 static LIST_HEAD(nfs_access_lru_list
);
1671 static atomic_long_t nfs_access_nr_entries
;
1673 static void nfs_access_free_entry(struct nfs_access_entry
*entry
)
1675 put_rpccred(entry
->cred
);
1677 smp_mb__before_atomic_dec();
1678 atomic_long_dec(&nfs_access_nr_entries
);
1679 smp_mb__after_atomic_dec();
1682 int nfs_access_cache_shrinker(int nr_to_scan
, gfp_t gfp_mask
)
1685 struct nfs_inode
*nfsi
;
1686 struct nfs_access_entry
*cache
;
1689 spin_lock(&nfs_access_lru_lock
);
1690 list_for_each_entry(nfsi
, &nfs_access_lru_list
, access_cache_inode_lru
) {
1691 struct rw_semaphore
*s_umount
;
1692 struct inode
*inode
;
1694 if (nr_to_scan
-- == 0)
1696 s_umount
= &nfsi
->vfs_inode
.i_sb
->s_umount
;
1697 if (!down_read_trylock(s_umount
))
1699 inode
= igrab(&nfsi
->vfs_inode
);
1700 if (inode
== NULL
) {
1704 spin_lock(&inode
->i_lock
);
1705 if (list_empty(&nfsi
->access_cache_entry_lru
))
1706 goto remove_lru_entry
;
1707 cache
= list_entry(nfsi
->access_cache_entry_lru
.next
,
1708 struct nfs_access_entry
, lru
);
1709 list_move(&cache
->lru
, &head
);
1710 rb_erase(&cache
->rb_node
, &nfsi
->access_cache
);
1711 if (!list_empty(&nfsi
->access_cache_entry_lru
))
1712 list_move_tail(&nfsi
->access_cache_inode_lru
,
1713 &nfs_access_lru_list
);
1716 list_del_init(&nfsi
->access_cache_inode_lru
);
1717 clear_bit(NFS_INO_ACL_LRU_SET
, &nfsi
->flags
);
1719 spin_unlock(&inode
->i_lock
);
1720 spin_unlock(&nfs_access_lru_lock
);
1725 spin_unlock(&nfs_access_lru_lock
);
1726 while (!list_empty(&head
)) {
1727 cache
= list_entry(head
.next
, struct nfs_access_entry
, lru
);
1728 list_del(&cache
->lru
);
1729 nfs_access_free_entry(cache
);
1731 return (atomic_long_read(&nfs_access_nr_entries
) / 100) * sysctl_vfs_cache_pressure
;
1734 static void __nfs_access_zap_cache(struct inode
*inode
)
1736 struct nfs_inode
*nfsi
= NFS_I(inode
);
1737 struct rb_root
*root_node
= &nfsi
->access_cache
;
1738 struct rb_node
*n
, *dispose
= NULL
;
1739 struct nfs_access_entry
*entry
;
1741 /* Unhook entries from the cache */
1742 while ((n
= rb_first(root_node
)) != NULL
) {
1743 entry
= rb_entry(n
, struct nfs_access_entry
, rb_node
);
1744 rb_erase(n
, root_node
);
1745 list_del(&entry
->lru
);
1746 n
->rb_left
= dispose
;
1749 nfsi
->cache_validity
&= ~NFS_INO_INVALID_ACCESS
;
1750 spin_unlock(&inode
->i_lock
);
1752 /* Now kill them all! */
1753 while (dispose
!= NULL
) {
1755 dispose
= n
->rb_left
;
1756 nfs_access_free_entry(rb_entry(n
, struct nfs_access_entry
, rb_node
));
1760 void nfs_access_zap_cache(struct inode
*inode
)
1762 /* Remove from global LRU init */
1763 if (test_and_clear_bit(NFS_INO_ACL_LRU_SET
, &NFS_I(inode
)->flags
)) {
1764 spin_lock(&nfs_access_lru_lock
);
1765 list_del_init(&NFS_I(inode
)->access_cache_inode_lru
);
1766 spin_unlock(&nfs_access_lru_lock
);
1769 spin_lock(&inode
->i_lock
);
1770 /* This will release the spinlock */
1771 __nfs_access_zap_cache(inode
);
1774 static struct nfs_access_entry
*nfs_access_search_rbtree(struct inode
*inode
, struct rpc_cred
*cred
)
1776 struct rb_node
*n
= NFS_I(inode
)->access_cache
.rb_node
;
1777 struct nfs_access_entry
*entry
;
1780 entry
= rb_entry(n
, struct nfs_access_entry
, rb_node
);
1782 if (cred
< entry
->cred
)
1784 else if (cred
> entry
->cred
)
1792 static int nfs_access_get_cached(struct inode
*inode
, struct rpc_cred
*cred
, struct nfs_access_entry
*res
)
1794 struct nfs_inode
*nfsi
= NFS_I(inode
);
1795 struct nfs_access_entry
*cache
;
1798 spin_lock(&inode
->i_lock
);
1799 if (nfsi
->cache_validity
& NFS_INO_INVALID_ACCESS
)
1801 cache
= nfs_access_search_rbtree(inode
, cred
);
1804 if (!time_in_range(jiffies
, cache
->jiffies
, cache
->jiffies
+ nfsi
->attrtimeo
))
1806 res
->jiffies
= cache
->jiffies
;
1807 res
->cred
= cache
->cred
;
1808 res
->mask
= cache
->mask
;
1809 list_move_tail(&cache
->lru
, &nfsi
->access_cache_entry_lru
);
1812 spin_unlock(&inode
->i_lock
);
1815 rb_erase(&cache
->rb_node
, &nfsi
->access_cache
);
1816 list_del(&cache
->lru
);
1817 spin_unlock(&inode
->i_lock
);
1818 nfs_access_free_entry(cache
);
1821 /* This will release the spinlock */
1822 __nfs_access_zap_cache(inode
);
1826 static void nfs_access_add_rbtree(struct inode
*inode
, struct nfs_access_entry
*set
)
1828 struct nfs_inode
*nfsi
= NFS_I(inode
);
1829 struct rb_root
*root_node
= &nfsi
->access_cache
;
1830 struct rb_node
**p
= &root_node
->rb_node
;
1831 struct rb_node
*parent
= NULL
;
1832 struct nfs_access_entry
*entry
;
1834 spin_lock(&inode
->i_lock
);
1835 while (*p
!= NULL
) {
1837 entry
= rb_entry(parent
, struct nfs_access_entry
, rb_node
);
1839 if (set
->cred
< entry
->cred
)
1840 p
= &parent
->rb_left
;
1841 else if (set
->cred
> entry
->cred
)
1842 p
= &parent
->rb_right
;
1846 rb_link_node(&set
->rb_node
, parent
, p
);
1847 rb_insert_color(&set
->rb_node
, root_node
);
1848 list_add_tail(&set
->lru
, &nfsi
->access_cache_entry_lru
);
1849 spin_unlock(&inode
->i_lock
);
1852 rb_replace_node(parent
, &set
->rb_node
, root_node
);
1853 list_add_tail(&set
->lru
, &nfsi
->access_cache_entry_lru
);
1854 list_del(&entry
->lru
);
1855 spin_unlock(&inode
->i_lock
);
1856 nfs_access_free_entry(entry
);
1859 static void nfs_access_add_cache(struct inode
*inode
, struct nfs_access_entry
*set
)
1861 struct nfs_access_entry
*cache
= kmalloc(sizeof(*cache
), GFP_KERNEL
);
1864 RB_CLEAR_NODE(&cache
->rb_node
);
1865 cache
->jiffies
= set
->jiffies
;
1866 cache
->cred
= get_rpccred(set
->cred
);
1867 cache
->mask
= set
->mask
;
1869 nfs_access_add_rbtree(inode
, cache
);
1871 /* Update accounting */
1872 smp_mb__before_atomic_inc();
1873 atomic_long_inc(&nfs_access_nr_entries
);
1874 smp_mb__after_atomic_inc();
1876 /* Add inode to global LRU list */
1877 if (!test_and_set_bit(NFS_INO_ACL_LRU_SET
, &NFS_I(inode
)->flags
)) {
1878 spin_lock(&nfs_access_lru_lock
);
1879 list_add_tail(&NFS_I(inode
)->access_cache_inode_lru
, &nfs_access_lru_list
);
1880 spin_unlock(&nfs_access_lru_lock
);
1884 static int nfs_do_access(struct inode
*inode
, struct rpc_cred
*cred
, int mask
)
1886 struct nfs_access_entry cache
;
1889 status
= nfs_access_get_cached(inode
, cred
, &cache
);
1893 /* Be clever: ask server to check for all possible rights */
1894 cache
.mask
= MAY_EXEC
| MAY_WRITE
| MAY_READ
;
1896 cache
.jiffies
= jiffies
;
1897 status
= NFS_PROTO(inode
)->access(inode
, &cache
);
1900 nfs_access_add_cache(inode
, &cache
);
1902 if ((cache
.mask
& mask
) == mask
)
1907 static int nfs_open_permission_mask(int openflags
)
1911 if (openflags
& FMODE_READ
)
1913 if (openflags
& FMODE_WRITE
)
1915 if (openflags
& FMODE_EXEC
)
1920 int nfs_may_open(struct inode
*inode
, struct rpc_cred
*cred
, int openflags
)
1922 return nfs_do_access(inode
, cred
, nfs_open_permission_mask(openflags
));
1925 int nfs_permission(struct inode
*inode
, int mask
, struct nameidata
*nd
)
1927 struct rpc_cred
*cred
;
1930 nfs_inc_stats(inode
, NFSIOS_VFSACCESS
);
1934 /* Is this sys_access() ? */
1935 if (nd
!= NULL
&& (nd
->flags
& LOOKUP_ACCESS
))
1938 switch (inode
->i_mode
& S_IFMT
) {
1942 /* NFSv4 has atomic_open... */
1943 if (nfs_server_capable(inode
, NFS_CAP_ATOMIC_OPEN
)
1945 && (nd
->flags
& LOOKUP_OPEN
))
1950 * Optimize away all write operations, since the server
1951 * will check permissions when we perform the op.
1953 if ((mask
& MAY_WRITE
) && !(mask
& MAY_READ
))
1958 if (!NFS_PROTO(inode
)->access
)
1961 cred
= rpc_lookup_cred();
1962 if (!IS_ERR(cred
)) {
1963 res
= nfs_do_access(inode
, cred
, mask
);
1966 res
= PTR_ERR(cred
);
1968 dfprintk(VFS
, "NFS: permission(%s/%ld), mask=0x%x, res=%d\n",
1969 inode
->i_sb
->s_id
, inode
->i_ino
, mask
, res
);
1972 res
= nfs_revalidate_inode(NFS_SERVER(inode
), inode
);
1974 res
= generic_permission(inode
, mask
, NULL
);
1980 * version-control: t
1981 * kept-new-versions: 5