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 const struct inode_operations nfs_dir_inode_operations = {
73 .symlink = nfs_symlink,
78 .permission = nfs_permission,
79 .getattr = nfs_getattr,
80 .setattr = nfs_setattr,
84 const struct inode_operations nfs3_dir_inode_operations = {
89 .symlink = nfs_symlink,
94 .permission = nfs_permission,
95 .getattr = nfs_getattr,
96 .setattr = nfs_setattr,
97 .listxattr = nfs3_listxattr,
98 .getxattr = nfs3_getxattr,
99 .setxattr = nfs3_setxattr,
100 .removexattr = nfs3_removexattr,
102 #endif /* CONFIG_NFS_V3 */
106 static struct dentry *nfs_atomic_lookup(struct inode *, struct dentry *, struct nameidata *);
107 const struct inode_operations nfs4_dir_inode_operations = {
108 .create = nfs_create,
109 .lookup = nfs_atomic_lookup,
111 .unlink = nfs_unlink,
112 .symlink = nfs_symlink,
116 .rename = nfs_rename,
117 .permission = nfs_permission,
118 .getattr = nfs_getattr,
119 .setattr = nfs_setattr,
120 .getxattr = nfs4_getxattr,
121 .setxattr = nfs4_setxattr,
122 .listxattr = nfs4_listxattr,
125 #endif /* CONFIG_NFS_V4 */
131 nfs_opendir(struct inode *inode, struct file *filp)
135 dfprintk(VFS, "NFS: opendir(%s/%ld)\n",
136 inode->i_sb->s_id, inode->i_ino);
139 /* Call generic open code in order to cache credentials */
140 res = nfs_open(inode, filp);
145 typedef __be32 * (*decode_dirent_t)(__be32 *, struct nfs_entry *, int);
149 unsigned long page_index;
152 loff_t current_index;
153 struct nfs_entry *entry;
154 decode_dirent_t decode;
157 unsigned long timestamp;
159 } nfs_readdir_descriptor_t;
161 /* Now we cache directories properly, by stuffing the dirent
162 * data directly in the page cache.
164 * Inode invalidation due to refresh etc. takes care of
165 * _everything_, no sloppy entry flushing logic, no extraneous
166 * copying, network direct to page cache, the way it was meant
169 * NOTE: Dirent information verification is done always by the
170 * page-in of the RPC reply, nowhere else, this simplies
171 * things substantially.
174 int nfs_readdir_filler(nfs_readdir_descriptor_t *desc, struct page *page)
176 struct file *file = desc->file;
177 struct inode *inode = file->f_path.dentry->d_inode;
178 struct rpc_cred *cred = nfs_file_cred(file);
179 unsigned long timestamp;
182 dfprintk(DIRCACHE, "NFS: %s: reading cookie %Lu into page %lu\n",
183 __FUNCTION__, (long long)desc->entry->cookie,
188 error = NFS_PROTO(inode)->readdir(file->f_path.dentry, cred, desc->entry->cookie, page,
189 NFS_SERVER(inode)->dtsize, desc->plus);
191 /* We requested READDIRPLUS, but the server doesn't grok it */
192 if (error == -ENOTSUPP && desc->plus) {
193 NFS_SERVER(inode)->caps &= ~NFS_CAP_READDIRPLUS;
194 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_FLAGS(inode));
200 desc->timestamp = timestamp;
201 desc->timestamp_valid = 1;
202 SetPageUptodate(page);
203 spin_lock(&inode->i_lock);
204 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ATIME;
205 spin_unlock(&inode->i_lock);
206 /* Ensure consistent page alignment of the data.
207 * Note: assumes we have exclusive access to this mapping either
208 * through inode->i_mutex or some other mechanism.
210 if (page->index == 0 && invalidate_inode_pages2_range(inode->i_mapping, PAGE_CACHE_SIZE, -1) < 0) {
211 /* Should never happen */
212 nfs_zap_mapping(inode, inode->i_mapping);
219 nfs_zap_caches(inode);
225 int dir_decode(nfs_readdir_descriptor_t *desc)
227 __be32 *p = desc->ptr;
228 p = desc->decode(p, desc->entry, desc->plus);
232 if (desc->timestamp_valid)
233 desc->entry->fattr->time_start = desc->timestamp;
235 desc->entry->fattr->valid &= ~NFS_ATTR_FATTR;
240 void dir_page_release(nfs_readdir_descriptor_t *desc)
243 page_cache_release(desc->page);
249 * Given a pointer to a buffer that has already been filled by a call
250 * to readdir, find the next entry with cookie '*desc->dir_cookie'.
252 * If the end of the buffer has been reached, return -EAGAIN, if not,
253 * return the offset within the buffer of the next entry to be
257 int find_dirent(nfs_readdir_descriptor_t *desc)
259 struct nfs_entry *entry = desc->entry;
263 while((status = dir_decode(desc)) == 0) {
264 dfprintk(DIRCACHE, "NFS: %s: examining cookie %Lu\n",
265 __FUNCTION__, (unsigned long long)entry->cookie);
266 if (entry->prev_cookie == *desc->dir_cookie)
268 if (loop_count++ > 200) {
277 * Given a pointer to a buffer that has already been filled by a call
278 * to readdir, find the entry at offset 'desc->file->f_pos'.
280 * If the end of the buffer has been reached, return -EAGAIN, if not,
281 * return the offset within the buffer of the next entry to be
285 int find_dirent_index(nfs_readdir_descriptor_t *desc)
287 struct nfs_entry *entry = desc->entry;
292 status = dir_decode(desc);
296 dfprintk(DIRCACHE, "NFS: found cookie %Lu at index %Ld\n",
297 (unsigned long long)entry->cookie, desc->current_index);
299 if (desc->file->f_pos == desc->current_index) {
300 *desc->dir_cookie = entry->cookie;
303 desc->current_index++;
304 if (loop_count++ > 200) {
313 * Find the given page, and call find_dirent() or find_dirent_index in
314 * order to try to return the next entry.
317 int find_dirent_page(nfs_readdir_descriptor_t *desc)
319 struct inode *inode = desc->file->f_path.dentry->d_inode;
323 dfprintk(DIRCACHE, "NFS: %s: searching page %ld for target %Lu\n",
324 __FUNCTION__, desc->page_index,
325 (long long) *desc->dir_cookie);
327 /* If we find the page in the page_cache, we cannot be sure
328 * how fresh the data is, so we will ignore readdir_plus attributes.
330 desc->timestamp_valid = 0;
331 page = read_cache_page(inode->i_mapping, desc->page_index,
332 (filler_t *)nfs_readdir_filler, desc);
334 status = PTR_ERR(page);
338 /* NOTE: Someone else may have changed the READDIRPLUS flag */
340 desc->ptr = kmap(page); /* matching kunmap in nfs_do_filldir */
341 if (*desc->dir_cookie != 0)
342 status = find_dirent(desc);
344 status = find_dirent_index(desc);
346 dir_page_release(desc);
348 dfprintk(DIRCACHE, "NFS: %s: returns %d\n", __FUNCTION__, status);
353 * Recurse through the page cache pages, and return a
354 * filled nfs_entry structure of the next directory entry if possible.
356 * The target for the search is '*desc->dir_cookie' if non-0,
357 * 'desc->file->f_pos' otherwise
360 int readdir_search_pagecache(nfs_readdir_descriptor_t *desc)
365 /* Always search-by-index from the beginning of the cache */
366 if (*desc->dir_cookie == 0) {
367 dfprintk(DIRCACHE, "NFS: readdir_search_pagecache() searching for offset %Ld\n",
368 (long long)desc->file->f_pos);
369 desc->page_index = 0;
370 desc->entry->cookie = desc->entry->prev_cookie = 0;
371 desc->entry->eof = 0;
372 desc->current_index = 0;
374 dfprintk(DIRCACHE, "NFS: readdir_search_pagecache() searching for cookie %Lu\n",
375 (unsigned long long)*desc->dir_cookie);
378 res = find_dirent_page(desc);
381 /* Align to beginning of next page */
383 if (loop_count++ > 200) {
389 dfprintk(DIRCACHE, "NFS: %s: returns %d\n", __FUNCTION__, res);
393 static inline unsigned int dt_type(struct inode *inode)
395 return (inode->i_mode >> 12) & 15;
398 static struct dentry *nfs_readdir_lookup(nfs_readdir_descriptor_t *desc);
401 * Once we've found the start of the dirent within a page: fill 'er up...
404 int nfs_do_filldir(nfs_readdir_descriptor_t *desc, void *dirent,
407 struct file *file = desc->file;
408 struct nfs_entry *entry = desc->entry;
409 struct dentry *dentry = NULL;
410 unsigned long fileid;
414 dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling starting @ cookie %Lu\n",
415 (unsigned long long)entry->cookie);
418 unsigned d_type = DT_UNKNOWN;
419 /* Note: entry->prev_cookie contains the cookie for
420 * retrieving the current dirent on the server */
421 fileid = nfs_fileid_to_ino_t(entry->ino);
423 /* Get a dentry if we have one */
426 dentry = nfs_readdir_lookup(desc);
428 /* Use readdirplus info */
429 if (dentry != NULL && dentry->d_inode != NULL) {
430 d_type = dt_type(dentry->d_inode);
431 fileid = dentry->d_inode->i_ino;
434 res = filldir(dirent, entry->name, entry->len,
435 file->f_pos, fileid, d_type);
439 *desc->dir_cookie = entry->cookie;
440 if (dir_decode(desc) != 0) {
444 if (loop_count++ > 200) {
449 dir_page_release(desc);
452 dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n",
453 (unsigned long long)*desc->dir_cookie, res);
458 * If we cannot find a cookie in our cache, we suspect that this is
459 * because it points to a deleted file, so we ask the server to return
460 * whatever it thinks is the next entry. We then feed this to filldir.
461 * If all goes well, we should then be able to find our way round the
462 * cache on the next call to readdir_search_pagecache();
464 * NOTE: we cannot add the anonymous page to the pagecache because
465 * the data it contains might not be page aligned. Besides,
466 * we should already have a complete representation of the
467 * directory in the page cache by the time we get here.
470 int uncached_readdir(nfs_readdir_descriptor_t *desc, void *dirent,
473 struct file *file = desc->file;
474 struct inode *inode = file->f_path.dentry->d_inode;
475 struct rpc_cred *cred = nfs_file_cred(file);
476 struct page *page = NULL;
478 unsigned long timestamp;
480 dfprintk(DIRCACHE, "NFS: uncached_readdir() searching for cookie %Lu\n",
481 (unsigned long long)*desc->dir_cookie);
483 page = alloc_page(GFP_HIGHUSER);
489 desc->error = NFS_PROTO(inode)->readdir(file->f_path.dentry, cred, *desc->dir_cookie,
491 NFS_SERVER(inode)->dtsize,
493 spin_lock(&inode->i_lock);
494 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ATIME;
495 spin_unlock(&inode->i_lock);
497 desc->ptr = kmap(page); /* matching kunmap in nfs_do_filldir */
498 if (desc->error >= 0) {
499 desc->timestamp = timestamp;
500 desc->timestamp_valid = 1;
501 if ((status = dir_decode(desc)) == 0)
502 desc->entry->prev_cookie = *desc->dir_cookie;
508 status = nfs_do_filldir(desc, dirent, filldir);
510 /* Reset read descriptor so it searches the page cache from
511 * the start upon the next call to readdir_search_pagecache() */
512 desc->page_index = 0;
513 desc->entry->cookie = desc->entry->prev_cookie = 0;
514 desc->entry->eof = 0;
516 dfprintk(DIRCACHE, "NFS: %s: returns %d\n",
517 __FUNCTION__, status);
520 dir_page_release(desc);
524 /* The file offset position represents the dirent entry number. A
525 last cookie cache takes care of the common case of reading the
528 static int nfs_readdir(struct file *filp, void *dirent, filldir_t filldir)
530 struct dentry *dentry = filp->f_path.dentry;
531 struct inode *inode = dentry->d_inode;
532 nfs_readdir_descriptor_t my_desc,
534 struct nfs_entry my_entry;
536 struct nfs_fattr fattr;
539 dfprintk(VFS, "NFS: readdir(%s/%s) starting at cookie %Lu\n",
540 dentry->d_parent->d_name.name, dentry->d_name.name,
541 (long long)filp->f_pos);
542 nfs_inc_stats(inode, NFSIOS_VFSGETDENTS);
546 res = nfs_revalidate_mapping_nolock(inode, filp->f_mapping);
553 * filp->f_pos points to the dirent entry number.
554 * *desc->dir_cookie has the cookie for the next entry. We have
555 * to either find the entry with the appropriate number or
556 * revalidate the cookie.
558 memset(desc, 0, sizeof(*desc));
561 desc->dir_cookie = &((struct nfs_open_context *)filp->private_data)->dir_cookie;
562 desc->decode = NFS_PROTO(inode)->decode_dirent;
563 desc->plus = NFS_USE_READDIRPLUS(inode);
565 my_entry.cookie = my_entry.prev_cookie = 0;
568 my_entry.fattr = &fattr;
569 nfs_fattr_init(&fattr);
570 desc->entry = &my_entry;
572 while(!desc->entry->eof) {
573 res = readdir_search_pagecache(desc);
575 if (res == -EBADCOOKIE) {
576 /* This means either end of directory */
577 if (*desc->dir_cookie && desc->entry->cookie != *desc->dir_cookie) {
578 /* Or that the server has 'lost' a cookie */
579 res = uncached_readdir(desc, dirent, filldir);
586 if (res == -ETOOSMALL && desc->plus) {
587 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_FLAGS(inode));
588 nfs_zap_caches(inode);
590 desc->entry->eof = 0;
596 res = nfs_do_filldir(desc, dirent, filldir);
605 dfprintk(VFS, "NFS: readdir(%s/%s) returns %ld\n",
606 dentry->d_parent->d_name.name, dentry->d_name.name,
611 loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int origin)
613 mutex_lock(&filp->f_path.dentry->d_inode->i_mutex);
616 offset += filp->f_pos;
624 if (offset != filp->f_pos) {
625 filp->f_pos = offset;
626 ((struct nfs_open_context *)filp->private_data)->dir_cookie = 0;
629 mutex_unlock(&filp->f_path.dentry->d_inode->i_mutex);
634 * All directory operations under NFS are synchronous, so fsync()
635 * is a dummy operation.
637 int nfs_fsync_dir(struct file *filp, struct dentry *dentry, int datasync)
639 dfprintk(VFS, "NFS: fsync_dir(%s/%s) datasync %d\n",
640 dentry->d_parent->d_name.name, dentry->d_name.name,
647 * A check for whether or not the parent directory has changed.
648 * In the case it has, we assume that the dentries are untrustworthy
649 * and may need to be looked up again.
651 static int nfs_check_verifier(struct inode *dir, struct dentry *dentry)
657 verf = (unsigned long)dentry->d_fsdata;
658 if ((NFS_I(dir)->cache_validity & NFS_INO_INVALID_ATTR) != 0
659 || nfs_attribute_timeout(dir)
660 || nfs_caches_unstable(dir)
661 || verf != NFS_I(dir)->cache_change_attribute)
666 static inline void nfs_set_verifier(struct dentry * dentry, unsigned long verf)
668 dentry->d_fsdata = (void *)verf;
671 static void nfs_refresh_verifier(struct dentry * dentry, unsigned long verf)
673 nfs_set_verifier(dentry, verf);
677 * Whenever an NFS operation succeeds, we know that the dentry
678 * is valid, so we update the revalidation timestamp.
680 static inline void nfs_renew_times(struct dentry * dentry)
682 dentry->d_time = jiffies;
686 * Return the intent data that applies to this particular path component
688 * Note that the current set of intents only apply to the very last
689 * component of the path.
690 * We check for this using LOOKUP_CONTINUE and LOOKUP_PARENT.
692 static inline unsigned int nfs_lookup_check_intent(struct nameidata *nd, unsigned int mask)
694 if (nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))
696 return nd->flags & mask;
700 * Inode and filehandle revalidation for lookups.
702 * We force revalidation in the cases where the VFS sets LOOKUP_REVAL,
703 * or if the intent information indicates that we're about to open this
704 * particular file and the "nocto" mount flag is not set.
708 int nfs_lookup_verify_inode(struct inode *inode, struct nameidata *nd)
710 struct nfs_server *server = NFS_SERVER(inode);
713 /* VFS wants an on-the-wire revalidation */
714 if (nd->flags & LOOKUP_REVAL)
716 /* This is an open(2) */
717 if (nfs_lookup_check_intent(nd, LOOKUP_OPEN) != 0 &&
718 !(server->flags & NFS_MOUNT_NOCTO) &&
719 (S_ISREG(inode->i_mode) ||
720 S_ISDIR(inode->i_mode)))
723 return nfs_revalidate_inode(server, inode);
725 return __nfs_revalidate_inode(server, inode);
729 * We judge how long we want to trust negative
730 * dentries by looking at the parent inode mtime.
732 * If parent mtime has changed, we revalidate, else we wait for a
733 * period corresponding to the parent's attribute cache timeout value.
736 int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry,
737 struct nameidata *nd)
739 /* Don't revalidate a negative dentry if we're creating a new file */
740 if (nd != NULL && nfs_lookup_check_intent(nd, LOOKUP_CREATE) != 0)
742 return !nfs_check_verifier(dir, dentry);
746 * This is called every time the dcache has a lookup hit,
747 * and we should check whether we can really trust that
750 * NOTE! The hit can be a negative hit too, don't assume
753 * If the parent directory is seen to have changed, we throw out the
754 * cached dentry and do a new lookup.
756 static int nfs_lookup_revalidate(struct dentry * dentry, struct nameidata *nd)
760 struct dentry *parent;
762 struct nfs_fh fhandle;
763 struct nfs_fattr fattr;
764 unsigned long verifier;
766 parent = dget_parent(dentry);
768 dir = parent->d_inode;
769 nfs_inc_stats(dir, NFSIOS_DENTRYREVALIDATE);
770 inode = dentry->d_inode;
773 if (nfs_neg_need_reval(dir, dentry, nd))
778 if (is_bad_inode(inode)) {
779 dfprintk(LOOKUPCACHE, "%s: %s/%s has dud inode\n",
780 __FUNCTION__, dentry->d_parent->d_name.name,
781 dentry->d_name.name);
785 /* Revalidate parent directory attribute cache */
786 if (nfs_revalidate_inode(NFS_SERVER(dir), dir) < 0)
789 /* Force a full look up iff the parent directory has changed */
790 if (nfs_check_verifier(dir, dentry)) {
791 if (nfs_lookup_verify_inode(inode, nd))
796 if (NFS_STALE(inode))
799 verifier = nfs_save_change_attribute(dir);
800 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, &fhandle, &fattr);
803 if (nfs_compare_fh(NFS_FH(inode), &fhandle))
805 if ((error = nfs_refresh_inode(inode, &fattr)) != 0)
808 nfs_renew_times(dentry);
809 nfs_refresh_verifier(dentry, verifier);
813 dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is valid\n",
814 __FUNCTION__, dentry->d_parent->d_name.name,
815 dentry->d_name.name);
821 if (inode && S_ISDIR(inode->i_mode)) {
822 /* Purge readdir caches. */
823 nfs_zap_caches(inode);
824 /* If we have submounts, don't unhash ! */
825 if (have_submounts(dentry))
827 shrink_dcache_parent(dentry);
832 dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is invalid\n",
833 __FUNCTION__, dentry->d_parent->d_name.name,
834 dentry->d_name.name);
839 * This is called from dput() when d_count is going to 0.
841 static int nfs_dentry_delete(struct dentry *dentry)
843 dfprintk(VFS, "NFS: dentry_delete(%s/%s, %x)\n",
844 dentry->d_parent->d_name.name, dentry->d_name.name,
847 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
848 /* Unhash it, so that ->d_iput() would be called */
851 if (!(dentry->d_sb->s_flags & MS_ACTIVE)) {
852 /* Unhash it, so that ancestors of killed async unlink
853 * files will be cleaned up during umount */
861 * Called when the dentry loses inode.
862 * We use it to clean up silly-renamed files.
864 static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode)
866 nfs_inode_return_delegation(inode);
867 if (S_ISDIR(inode->i_mode))
868 /* drop any readdir cache as it could easily be old */
869 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA;
871 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
874 nfs_complete_unlink(dentry);
877 /* When creating a negative dentry, we want to renew d_time */
878 nfs_renew_times(dentry);
882 struct dentry_operations nfs_dentry_operations = {
883 .d_revalidate = nfs_lookup_revalidate,
884 .d_delete = nfs_dentry_delete,
885 .d_iput = nfs_dentry_iput,
889 * Use intent information to check whether or not we're going to do
890 * an O_EXCL create using this path component.
893 int nfs_is_exclusive_create(struct inode *dir, struct nameidata *nd)
895 if (NFS_PROTO(dir)->version == 2)
897 if (nd == NULL || nfs_lookup_check_intent(nd, LOOKUP_CREATE) == 0)
899 return (nd->intent.open.flags & O_EXCL) != 0;
902 static inline int nfs_reval_fsid(struct vfsmount *mnt, struct inode *dir,
903 struct nfs_fh *fh, struct nfs_fattr *fattr)
905 struct nfs_server *server = NFS_SERVER(dir);
907 if (!nfs_fsid_equal(&server->fsid, &fattr->fsid))
908 /* Revalidate fsid on root dir */
909 return __nfs_revalidate_inode(server, mnt->mnt_root->d_inode);
913 static struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
916 struct inode *inode = NULL;
918 struct nfs_fh fhandle;
919 struct nfs_fattr fattr;
921 dfprintk(VFS, "NFS: lookup(%s/%s)\n",
922 dentry->d_parent->d_name.name, dentry->d_name.name);
923 nfs_inc_stats(dir, NFSIOS_VFSLOOKUP);
925 res = ERR_PTR(-ENAMETOOLONG);
926 if (dentry->d_name.len > NFS_SERVER(dir)->namelen)
929 res = ERR_PTR(-ENOMEM);
930 dentry->d_op = NFS_PROTO(dir)->dentry_ops;
935 * If we're doing an exclusive create, optimize away the lookup
936 * but don't hash the dentry.
938 if (nfs_is_exclusive_create(dir, nd)) {
939 d_instantiate(dentry, NULL);
944 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, &fhandle, &fattr);
945 if (error == -ENOENT)
948 res = ERR_PTR(error);
951 error = nfs_reval_fsid(nd->mnt, dir, &fhandle, &fattr);
953 res = ERR_PTR(error);
956 inode = nfs_fhget(dentry->d_sb, &fhandle, &fattr);
957 res = (struct dentry *)inode;
962 res = d_materialise_unique(dentry, inode);
964 struct dentry *parent;
967 /* Was a directory renamed! */
968 parent = dget_parent(res);
969 if (!IS_ROOT(parent))
970 nfs_mark_for_revalidate(parent->d_inode);
974 nfs_renew_times(dentry);
975 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
983 static int nfs_open_revalidate(struct dentry *, struct nameidata *);
985 struct dentry_operations nfs4_dentry_operations = {
986 .d_revalidate = nfs_open_revalidate,
987 .d_delete = nfs_dentry_delete,
988 .d_iput = nfs_dentry_iput,
992 * Use intent information to determine whether we need to substitute
993 * the NFSv4-style stateful OPEN for the LOOKUP call
995 static int is_atomic_open(struct inode *dir, struct nameidata *nd)
997 if (nd == NULL || nfs_lookup_check_intent(nd, LOOKUP_OPEN) == 0)
999 /* NFS does not (yet) have a stateful open for directories */
1000 if (nd->flags & LOOKUP_DIRECTORY)
1002 /* Are we trying to write to a read only partition? */
1003 if (IS_RDONLY(dir) && (nd->intent.open.flags & (O_CREAT|O_TRUNC|FMODE_WRITE)))
1008 static struct dentry *nfs_atomic_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd)
1010 struct dentry *res = NULL;
1013 dfprintk(VFS, "NFS: atomic_lookup(%s/%ld), %s\n",
1014 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1016 /* Check that we are indeed trying to open this file */
1017 if (!is_atomic_open(dir, nd))
1020 if (dentry->d_name.len > NFS_SERVER(dir)->namelen) {
1021 res = ERR_PTR(-ENAMETOOLONG);
1024 dentry->d_op = NFS_PROTO(dir)->dentry_ops;
1026 /* Let vfs_create() deal with O_EXCL */
1027 if (nd->intent.open.flags & O_EXCL) {
1028 d_add(dentry, NULL);
1032 /* Open the file on the server */
1034 /* Revalidate parent directory attribute cache */
1035 error = nfs_revalidate_inode(NFS_SERVER(dir), dir);
1037 res = ERR_PTR(error);
1042 if (nd->intent.open.flags & O_CREAT) {
1043 nfs_begin_data_update(dir);
1044 res = nfs4_atomic_open(dir, dentry, nd);
1045 nfs_end_data_update(dir);
1047 res = nfs4_atomic_open(dir, dentry, nd);
1050 error = PTR_ERR(res);
1052 /* Make a negative dentry */
1056 /* This turned out not to be a regular file */
1061 if (!(nd->intent.open.flags & O_NOFOLLOW))
1067 } else if (res != NULL)
1069 nfs_renew_times(dentry);
1070 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1074 return nfs_lookup(dir, dentry, nd);
1077 static int nfs_open_revalidate(struct dentry *dentry, struct nameidata *nd)
1079 struct dentry *parent = NULL;
1080 struct inode *inode = dentry->d_inode;
1082 unsigned long verifier;
1083 int openflags, ret = 0;
1085 parent = dget_parent(dentry);
1086 dir = parent->d_inode;
1087 if (!is_atomic_open(dir, nd))
1089 /* We can't create new files in nfs_open_revalidate(), so we
1090 * optimize away revalidation of negative dentries.
1094 /* NFS only supports OPEN on regular files */
1095 if (!S_ISREG(inode->i_mode))
1097 openflags = nd->intent.open.flags;
1098 /* We cannot do exclusive creation on a positive dentry */
1099 if ((openflags & (O_CREAT|O_EXCL)) == (O_CREAT|O_EXCL))
1101 /* We can't create new files, or truncate existing ones here */
1102 openflags &= ~(O_CREAT|O_TRUNC);
1105 * Note: we're not holding inode->i_mutex and so may be racing with
1106 * operations that change the directory. We therefore save the
1107 * change attribute *before* we do the RPC call.
1110 verifier = nfs_save_change_attribute(dir);
1111 ret = nfs4_open_revalidate(dir, dentry, openflags, nd);
1113 nfs_refresh_verifier(dentry, verifier);
1122 if (inode != NULL && nfs_have_delegation(inode, FMODE_READ))
1124 return nfs_lookup_revalidate(dentry, nd);
1126 #endif /* CONFIG_NFSV4 */
1128 static struct dentry *nfs_readdir_lookup(nfs_readdir_descriptor_t *desc)
1130 struct dentry *parent = desc->file->f_path.dentry;
1131 struct inode *dir = parent->d_inode;
1132 struct nfs_entry *entry = desc->entry;
1133 struct dentry *dentry, *alias;
1134 struct qstr name = {
1135 .name = entry->name,
1138 struct inode *inode;
1142 if (name.name[0] == '.' && name.name[1] == '.')
1143 return dget_parent(parent);
1146 if (name.name[0] == '.')
1147 return dget(parent);
1149 name.hash = full_name_hash(name.name, name.len);
1150 dentry = d_lookup(parent, &name);
1151 if (dentry != NULL) {
1152 /* Is this a positive dentry that matches the readdir info? */
1153 if (dentry->d_inode != NULL &&
1154 (NFS_FILEID(dentry->d_inode) == entry->ino ||
1155 d_mountpoint(dentry))) {
1156 if (!desc->plus || entry->fh->size == 0)
1158 if (nfs_compare_fh(NFS_FH(dentry->d_inode),
1162 /* No, so d_drop to allow one to be created */
1166 if (!desc->plus || !(entry->fattr->valid & NFS_ATTR_FATTR))
1168 /* Note: caller is already holding the dir->i_mutex! */
1169 dentry = d_alloc(parent, &name);
1172 dentry->d_op = NFS_PROTO(dir)->dentry_ops;
1173 inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr);
1174 if (IS_ERR(inode)) {
1179 alias = d_materialise_unique(dentry, inode);
1180 if (alias != NULL) {
1187 nfs_renew_times(dentry);
1188 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1191 nfs_renew_times(dentry);
1192 nfs_refresh_verifier(dentry, nfs_save_change_attribute(dir));
1197 * Code common to create, mkdir, and mknod.
1199 int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle,
1200 struct nfs_fattr *fattr)
1202 struct inode *inode;
1203 int error = -EACCES;
1205 /* We may have been initialized further down */
1206 if (dentry->d_inode)
1208 if (fhandle->size == 0) {
1209 struct inode *dir = dentry->d_parent->d_inode;
1210 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr);
1214 if (!(fattr->valid & NFS_ATTR_FATTR)) {
1215 struct nfs_server *server = NFS_SB(dentry->d_sb);
1216 error = server->nfs_client->rpc_ops->getattr(server, fhandle, fattr);
1220 inode = nfs_fhget(dentry->d_sb, fhandle, fattr);
1221 error = PTR_ERR(inode);
1224 d_instantiate(dentry, inode);
1225 if (d_unhashed(dentry))
1231 * Following a failed create operation, we drop the dentry rather
1232 * than retain a negative dentry. This avoids a problem in the event
1233 * that the operation succeeded on the server, but an error in the
1234 * reply path made it appear to have failed.
1236 static int nfs_create(struct inode *dir, struct dentry *dentry, int mode,
1237 struct nameidata *nd)
1243 dfprintk(VFS, "NFS: create(%s/%ld), %s\n",
1244 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1246 attr.ia_mode = mode;
1247 attr.ia_valid = ATTR_MODE;
1249 if (nd && (nd->flags & LOOKUP_CREATE))
1250 open_flags = nd->intent.open.flags;
1253 nfs_begin_data_update(dir);
1254 error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags, nd);
1255 nfs_end_data_update(dir);
1258 nfs_renew_times(dentry);
1259 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1269 * See comments for nfs_proc_create regarding failed operations.
1272 nfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t rdev)
1277 dfprintk(VFS, "NFS: mknod(%s/%ld), %s\n",
1278 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1280 if (!new_valid_dev(rdev))
1283 attr.ia_mode = mode;
1284 attr.ia_valid = ATTR_MODE;
1287 nfs_begin_data_update(dir);
1288 status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev);
1289 nfs_end_data_update(dir);
1292 nfs_renew_times(dentry);
1293 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1303 * See comments for nfs_proc_create regarding failed operations.
1305 static int nfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1310 dfprintk(VFS, "NFS: mkdir(%s/%ld), %s\n",
1311 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1313 attr.ia_valid = ATTR_MODE;
1314 attr.ia_mode = mode | S_IFDIR;
1317 nfs_begin_data_update(dir);
1318 error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr);
1319 nfs_end_data_update(dir);
1322 nfs_renew_times(dentry);
1323 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1332 static int nfs_rmdir(struct inode *dir, struct dentry *dentry)
1336 dfprintk(VFS, "NFS: rmdir(%s/%ld), %s\n",
1337 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1340 nfs_begin_data_update(dir);
1341 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
1342 /* Ensure the VFS deletes this inode */
1343 if (error == 0 && dentry->d_inode != NULL)
1344 clear_nlink(dentry->d_inode);
1345 nfs_end_data_update(dir);
1351 static int nfs_sillyrename(struct inode *dir, struct dentry *dentry)
1353 static unsigned int sillycounter;
1354 const int i_inosize = sizeof(dir->i_ino)*2;
1355 const int countersize = sizeof(sillycounter)*2;
1356 const int slen = sizeof(".nfs") + i_inosize + countersize - 1;
1359 struct dentry *sdentry;
1362 dfprintk(VFS, "NFS: silly-rename(%s/%s, ct=%d)\n",
1363 dentry->d_parent->d_name.name, dentry->d_name.name,
1364 atomic_read(&dentry->d_count));
1365 nfs_inc_stats(dir, NFSIOS_SILLYRENAME);
1368 if (!dentry->d_inode)
1369 printk("NFS: silly-renaming %s/%s, negative dentry??\n",
1370 dentry->d_parent->d_name.name, dentry->d_name.name);
1373 * We don't allow a dentry to be silly-renamed twice.
1376 if (dentry->d_flags & DCACHE_NFSFS_RENAMED)
1379 sprintf(silly, ".nfs%*.*lx",
1380 i_inosize, i_inosize, dentry->d_inode->i_ino);
1382 /* Return delegation in anticipation of the rename */
1383 nfs_inode_return_delegation(dentry->d_inode);
1387 char *suffix = silly + slen - countersize;
1391 sprintf(suffix, "%*.*x", countersize, countersize, sillycounter);
1393 dfprintk(VFS, "NFS: trying to rename %s to %s\n",
1394 dentry->d_name.name, silly);
1396 sdentry = lookup_one_len(silly, dentry->d_parent, slen);
1398 * N.B. Better to return EBUSY here ... it could be
1399 * dangerous to delete the file while it's in use.
1401 if (IS_ERR(sdentry))
1403 } while(sdentry->d_inode != NULL); /* need negative lookup */
1405 qsilly.name = silly;
1406 qsilly.len = strlen(silly);
1407 nfs_begin_data_update(dir);
1408 if (dentry->d_inode) {
1409 nfs_begin_data_update(dentry->d_inode);
1410 error = NFS_PROTO(dir)->rename(dir, &dentry->d_name,
1412 nfs_mark_for_revalidate(dentry->d_inode);
1413 nfs_end_data_update(dentry->d_inode);
1415 error = NFS_PROTO(dir)->rename(dir, &dentry->d_name,
1417 nfs_end_data_update(dir);
1419 nfs_renew_times(dentry);
1420 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1421 d_move(dentry, sdentry);
1422 error = nfs_async_unlink(dentry);
1423 /* If we return 0 we don't unlink */
1431 * Remove a file after making sure there are no pending writes,
1432 * and after checking that the file has only one user.
1434 * We invalidate the attribute cache and free the inode prior to the operation
1435 * to avoid possible races if the server reuses the inode.
1437 static int nfs_safe_remove(struct dentry *dentry)
1439 struct inode *dir = dentry->d_parent->d_inode;
1440 struct inode *inode = dentry->d_inode;
1443 dfprintk(VFS, "NFS: safe_remove(%s/%s)\n",
1444 dentry->d_parent->d_name.name, dentry->d_name.name);
1446 /* If the dentry was sillyrenamed, we simply call d_delete() */
1447 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1452 nfs_begin_data_update(dir);
1453 if (inode != NULL) {
1454 nfs_inode_return_delegation(inode);
1455 nfs_begin_data_update(inode);
1456 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1457 /* The VFS may want to delete this inode */
1460 nfs_mark_for_revalidate(inode);
1461 nfs_end_data_update(inode);
1463 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1464 nfs_end_data_update(dir);
1469 /* We do silly rename. In case sillyrename() returns -EBUSY, the inode
1470 * belongs to an active ".nfs..." file and we return -EBUSY.
1472 * If sillyrename() returns 0, we do nothing, otherwise we unlink.
1474 static int nfs_unlink(struct inode *dir, struct dentry *dentry)
1477 int need_rehash = 0;
1479 dfprintk(VFS, "NFS: unlink(%s/%ld, %s)\n", dir->i_sb->s_id,
1480 dir->i_ino, dentry->d_name.name);
1483 spin_lock(&dcache_lock);
1484 spin_lock(&dentry->d_lock);
1485 if (atomic_read(&dentry->d_count) > 1) {
1486 spin_unlock(&dentry->d_lock);
1487 spin_unlock(&dcache_lock);
1488 /* Start asynchronous writeout of the inode */
1489 write_inode_now(dentry->d_inode, 0);
1490 error = nfs_sillyrename(dir, dentry);
1494 if (!d_unhashed(dentry)) {
1498 spin_unlock(&dentry->d_lock);
1499 spin_unlock(&dcache_lock);
1500 error = nfs_safe_remove(dentry);
1502 nfs_renew_times(dentry);
1503 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1504 } else if (need_rehash)
1511 * To create a symbolic link, most file systems instantiate a new inode,
1512 * add a page to it containing the path, then write it out to the disk
1513 * using prepare_write/commit_write.
1515 * Unfortunately the NFS client can't create the in-core inode first
1516 * because it needs a file handle to create an in-core inode (see
1517 * fs/nfs/inode.c:nfs_fhget). We only have a file handle *after* the
1518 * symlink request has completed on the server.
1520 * So instead we allocate a raw page, copy the symname into it, then do
1521 * the SYMLINK request with the page as the buffer. If it succeeds, we
1522 * now have a new file handle and can instantiate an in-core NFS inode
1523 * and move the raw page into its mapping.
1525 static int nfs_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1527 struct pagevec lru_pvec;
1531 unsigned int pathlen = strlen(symname);
1534 dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s)\n", dir->i_sb->s_id,
1535 dir->i_ino, dentry->d_name.name, symname);
1537 if (pathlen > PAGE_SIZE)
1538 return -ENAMETOOLONG;
1540 attr.ia_mode = S_IFLNK | S_IRWXUGO;
1541 attr.ia_valid = ATTR_MODE;
1545 page = alloc_page(GFP_KERNEL);
1551 kaddr = kmap_atomic(page, KM_USER0);
1552 memcpy(kaddr, symname, pathlen);
1553 if (pathlen < PAGE_SIZE)
1554 memset(kaddr + pathlen, 0, PAGE_SIZE - pathlen);
1555 kunmap_atomic(kaddr, KM_USER0);
1557 nfs_begin_data_update(dir);
1558 error = NFS_PROTO(dir)->symlink(dir, dentry, page, pathlen, &attr);
1559 nfs_end_data_update(dir);
1561 dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s) error %d\n",
1562 dir->i_sb->s_id, dir->i_ino,
1563 dentry->d_name.name, symname, error);
1571 * No big deal if we can't add this page to the page cache here.
1572 * READLINK will get the missing page from the server if needed.
1574 pagevec_init(&lru_pvec, 0);
1575 if (!add_to_page_cache(page, dentry->d_inode->i_mapping, 0,
1577 pagevec_add(&lru_pvec, page);
1578 pagevec_lru_add(&lru_pvec);
1579 SetPageUptodate(page);
1589 nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1591 struct inode *inode = old_dentry->d_inode;
1594 dfprintk(VFS, "NFS: link(%s/%s -> %s/%s)\n",
1595 old_dentry->d_parent->d_name.name, old_dentry->d_name.name,
1596 dentry->d_parent->d_name.name, dentry->d_name.name);
1599 nfs_begin_data_update(dir);
1600 nfs_begin_data_update(inode);
1601 error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name);
1603 atomic_inc(&inode->i_count);
1604 d_instantiate(dentry, inode);
1606 nfs_end_data_update(inode);
1607 nfs_end_data_update(dir);
1614 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a
1615 * different file handle for the same inode after a rename (e.g. when
1616 * moving to a different directory). A fail-safe method to do so would
1617 * be to look up old_dir/old_name, create a link to new_dir/new_name and
1618 * rename the old file using the sillyrename stuff. This way, the original
1619 * file in old_dir will go away when the last process iput()s the inode.
1623 * It actually works quite well. One needs to have the possibility for
1624 * at least one ".nfs..." file in each directory the file ever gets
1625 * moved or linked to which happens automagically with the new
1626 * implementation that only depends on the dcache stuff instead of
1627 * using the inode layer
1629 * Unfortunately, things are a little more complicated than indicated
1630 * above. For a cross-directory move, we want to make sure we can get
1631 * rid of the old inode after the operation. This means there must be
1632 * no pending writes (if it's a file), and the use count must be 1.
1633 * If these conditions are met, we can drop the dentries before doing
1636 static int nfs_rename(struct inode *old_dir, struct dentry *old_dentry,
1637 struct inode *new_dir, struct dentry *new_dentry)
1639 struct inode *old_inode = old_dentry->d_inode;
1640 struct inode *new_inode = new_dentry->d_inode;
1641 struct dentry *dentry = NULL, *rehash = NULL;
1645 * To prevent any new references to the target during the rename,
1646 * we unhash the dentry and free the inode in advance.
1649 if (!d_unhashed(new_dentry)) {
1651 rehash = new_dentry;
1654 dfprintk(VFS, "NFS: rename(%s/%s -> %s/%s, ct=%d)\n",
1655 old_dentry->d_parent->d_name.name, old_dentry->d_name.name,
1656 new_dentry->d_parent->d_name.name, new_dentry->d_name.name,
1657 atomic_read(&new_dentry->d_count));
1660 * First check whether the target is busy ... we can't
1661 * safely do _any_ rename if the target is in use.
1663 * For files, make a copy of the dentry and then do a
1664 * silly-rename. If the silly-rename succeeds, the
1665 * copied dentry is hashed and becomes the new target.
1669 if (S_ISDIR(new_inode->i_mode)) {
1671 if (!S_ISDIR(old_inode->i_mode))
1673 } else if (atomic_read(&new_dentry->d_count) > 2) {
1675 /* copy the target dentry's name */
1676 dentry = d_alloc(new_dentry->d_parent,
1677 &new_dentry->d_name);
1681 /* silly-rename the existing target ... */
1682 err = nfs_sillyrename(new_dir, new_dentry);
1684 new_dentry = rehash = dentry;
1686 /* instantiate the replacement target */
1687 d_instantiate(new_dentry, NULL);
1688 } else if (atomic_read(&new_dentry->d_count) > 1) {
1689 /* dentry still busy? */
1691 printk("nfs_rename: target %s/%s busy, d_count=%d\n",
1692 new_dentry->d_parent->d_name.name,
1693 new_dentry->d_name.name,
1694 atomic_read(&new_dentry->d_count));
1699 drop_nlink(new_inode);
1703 * ... prune child dentries and writebacks if needed.
1705 if (atomic_read(&old_dentry->d_count) > 1) {
1706 if (S_ISREG(old_inode->i_mode))
1707 nfs_wb_all(old_inode);
1708 shrink_dcache_parent(old_dentry);
1710 nfs_inode_return_delegation(old_inode);
1712 if (new_inode != NULL) {
1713 nfs_inode_return_delegation(new_inode);
1714 d_delete(new_dentry);
1717 nfs_begin_data_update(old_dir);
1718 nfs_begin_data_update(new_dir);
1719 nfs_begin_data_update(old_inode);
1720 error = NFS_PROTO(old_dir)->rename(old_dir, &old_dentry->d_name,
1721 new_dir, &new_dentry->d_name);
1722 nfs_mark_for_revalidate(old_inode);
1723 nfs_end_data_update(old_inode);
1724 nfs_end_data_update(new_dir);
1725 nfs_end_data_update(old_dir);
1730 d_move(old_dentry, new_dentry);
1731 nfs_renew_times(new_dentry);
1732 nfs_refresh_verifier(new_dentry, nfs_save_change_attribute(new_dir));
1735 /* new dentry created? */
1742 static DEFINE_SPINLOCK(nfs_access_lru_lock);
1743 static LIST_HEAD(nfs_access_lru_list);
1744 static atomic_long_t nfs_access_nr_entries;
1746 static void nfs_access_free_entry(struct nfs_access_entry *entry)
1748 put_rpccred(entry->cred);
1750 smp_mb__before_atomic_dec();
1751 atomic_long_dec(&nfs_access_nr_entries);
1752 smp_mb__after_atomic_dec();
1755 int nfs_access_cache_shrinker(int nr_to_scan, gfp_t gfp_mask)
1758 struct nfs_inode *nfsi;
1759 struct nfs_access_entry *cache;
1761 spin_lock(&nfs_access_lru_lock);
1763 list_for_each_entry(nfsi, &nfs_access_lru_list, access_cache_inode_lru) {
1764 struct inode *inode;
1766 if (nr_to_scan-- == 0)
1768 inode = igrab(&nfsi->vfs_inode);
1771 spin_lock(&inode->i_lock);
1772 if (list_empty(&nfsi->access_cache_entry_lru))
1773 goto remove_lru_entry;
1774 cache = list_entry(nfsi->access_cache_entry_lru.next,
1775 struct nfs_access_entry, lru);
1776 list_move(&cache->lru, &head);
1777 rb_erase(&cache->rb_node, &nfsi->access_cache);
1778 if (!list_empty(&nfsi->access_cache_entry_lru))
1779 list_move_tail(&nfsi->access_cache_inode_lru,
1780 &nfs_access_lru_list);
1783 list_del_init(&nfsi->access_cache_inode_lru);
1784 clear_bit(NFS_INO_ACL_LRU_SET, &nfsi->flags);
1786 spin_unlock(&inode->i_lock);
1790 spin_unlock(&nfs_access_lru_lock);
1791 while (!list_empty(&head)) {
1792 cache = list_entry(head.next, struct nfs_access_entry, lru);
1793 list_del(&cache->lru);
1794 nfs_access_free_entry(cache);
1796 return (atomic_long_read(&nfs_access_nr_entries) / 100) * sysctl_vfs_cache_pressure;
1799 static void __nfs_access_zap_cache(struct inode *inode)
1801 struct nfs_inode *nfsi = NFS_I(inode);
1802 struct rb_root *root_node = &nfsi->access_cache;
1803 struct rb_node *n, *dispose = NULL;
1804 struct nfs_access_entry *entry;
1806 /* Unhook entries from the cache */
1807 while ((n = rb_first(root_node)) != NULL) {
1808 entry = rb_entry(n, struct nfs_access_entry, rb_node);
1809 rb_erase(n, root_node);
1810 list_del(&entry->lru);
1811 n->rb_left = dispose;
1814 nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS;
1815 spin_unlock(&inode->i_lock);
1817 /* Now kill them all! */
1818 while (dispose != NULL) {
1820 dispose = n->rb_left;
1821 nfs_access_free_entry(rb_entry(n, struct nfs_access_entry, rb_node));
1825 void nfs_access_zap_cache(struct inode *inode)
1827 /* Remove from global LRU init */
1828 if (test_and_clear_bit(NFS_INO_ACL_LRU_SET, &NFS_FLAGS(inode))) {
1829 spin_lock(&nfs_access_lru_lock);
1830 list_del_init(&NFS_I(inode)->access_cache_inode_lru);
1831 spin_unlock(&nfs_access_lru_lock);
1834 spin_lock(&inode->i_lock);
1835 /* This will release the spinlock */
1836 __nfs_access_zap_cache(inode);
1839 static struct nfs_access_entry *nfs_access_search_rbtree(struct inode *inode, struct rpc_cred *cred)
1841 struct rb_node *n = NFS_I(inode)->access_cache.rb_node;
1842 struct nfs_access_entry *entry;
1845 entry = rb_entry(n, struct nfs_access_entry, rb_node);
1847 if (cred < entry->cred)
1849 else if (cred > entry->cred)
1857 int nfs_access_get_cached(struct inode *inode, struct rpc_cred *cred, struct nfs_access_entry *res)
1859 struct nfs_inode *nfsi = NFS_I(inode);
1860 struct nfs_access_entry *cache;
1863 spin_lock(&inode->i_lock);
1864 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS)
1866 cache = nfs_access_search_rbtree(inode, cred);
1869 if (time_after(jiffies, cache->jiffies + NFS_ATTRTIMEO(inode)))
1871 res->jiffies = cache->jiffies;
1872 res->cred = cache->cred;
1873 res->mask = cache->mask;
1874 list_move_tail(&cache->lru, &nfsi->access_cache_entry_lru);
1877 spin_unlock(&inode->i_lock);
1880 rb_erase(&cache->rb_node, &nfsi->access_cache);
1881 list_del(&cache->lru);
1882 spin_unlock(&inode->i_lock);
1883 nfs_access_free_entry(cache);
1886 /* This will release the spinlock */
1887 __nfs_access_zap_cache(inode);
1891 static void nfs_access_add_rbtree(struct inode *inode, struct nfs_access_entry *set)
1893 struct nfs_inode *nfsi = NFS_I(inode);
1894 struct rb_root *root_node = &nfsi->access_cache;
1895 struct rb_node **p = &root_node->rb_node;
1896 struct rb_node *parent = NULL;
1897 struct nfs_access_entry *entry;
1899 spin_lock(&inode->i_lock);
1900 while (*p != NULL) {
1902 entry = rb_entry(parent, struct nfs_access_entry, rb_node);
1904 if (set->cred < entry->cred)
1905 p = &parent->rb_left;
1906 else if (set->cred > entry->cred)
1907 p = &parent->rb_right;
1911 rb_link_node(&set->rb_node, parent, p);
1912 rb_insert_color(&set->rb_node, root_node);
1913 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
1914 spin_unlock(&inode->i_lock);
1917 rb_replace_node(parent, &set->rb_node, root_node);
1918 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
1919 list_del(&entry->lru);
1920 spin_unlock(&inode->i_lock);
1921 nfs_access_free_entry(entry);
1924 void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set)
1926 struct nfs_access_entry *cache = kmalloc(sizeof(*cache), GFP_KERNEL);
1929 RB_CLEAR_NODE(&cache->rb_node);
1930 cache->jiffies = set->jiffies;
1931 cache->cred = get_rpccred(set->cred);
1932 cache->mask = set->mask;
1934 nfs_access_add_rbtree(inode, cache);
1936 /* Update accounting */
1937 smp_mb__before_atomic_inc();
1938 atomic_long_inc(&nfs_access_nr_entries);
1939 smp_mb__after_atomic_inc();
1941 /* Add inode to global LRU list */
1942 if (!test_and_set_bit(NFS_INO_ACL_LRU_SET, &NFS_FLAGS(inode))) {
1943 spin_lock(&nfs_access_lru_lock);
1944 list_add_tail(&NFS_I(inode)->access_cache_inode_lru, &nfs_access_lru_list);
1945 spin_unlock(&nfs_access_lru_lock);
1949 static int nfs_do_access(struct inode *inode, struct rpc_cred *cred, int mask)
1951 struct nfs_access_entry cache;
1954 status = nfs_access_get_cached(inode, cred, &cache);
1958 /* Be clever: ask server to check for all possible rights */
1959 cache.mask = MAY_EXEC | MAY_WRITE | MAY_READ;
1961 cache.jiffies = jiffies;
1962 status = NFS_PROTO(inode)->access(inode, &cache);
1965 nfs_access_add_cache(inode, &cache);
1967 if ((cache.mask & mask) == mask)
1972 int nfs_permission(struct inode *inode, int mask, struct nameidata *nd)
1974 struct rpc_cred *cred;
1977 nfs_inc_stats(inode, NFSIOS_VFSACCESS);
1981 /* Is this sys_access() ? */
1982 if (nd != NULL && (nd->flags & LOOKUP_ACCESS))
1985 switch (inode->i_mode & S_IFMT) {
1989 /* NFSv4 has atomic_open... */
1990 if (nfs_server_capable(inode, NFS_CAP_ATOMIC_OPEN)
1992 && (nd->flags & LOOKUP_OPEN))
1997 * Optimize away all write operations, since the server
1998 * will check permissions when we perform the op.
2000 if ((mask & MAY_WRITE) && !(mask & MAY_READ))
2007 if (!NFS_PROTO(inode)->access)
2010 cred = rpcauth_lookupcred(NFS_CLIENT(inode)->cl_auth, 0);
2011 if (!IS_ERR(cred)) {
2012 res = nfs_do_access(inode, cred, mask);
2015 res = PTR_ERR(cred);
2018 dfprintk(VFS, "NFS: permission(%s/%ld), mask=0x%x, res=%d\n",
2019 inode->i_sb->s_id, inode->i_ino, mask, res);
2022 res = nfs_revalidate_inode(NFS_SERVER(inode), inode);
2024 res = generic_permission(inode, mask, NULL);
2031 * version-control: t
2032 * kept-new-versions: 5