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/vs_tag.h>
39 #include "delegation.h"
42 #define NFS_PARANOIA 1
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 struct inode_operations nfs_dir_inode_operations = {
74 .symlink = nfs_symlink,
79 .permission = nfs_permission,
80 .getattr = nfs_getattr,
81 .setattr = nfs_setattr,
85 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 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(VFS, "NFS: opendir(%s/%ld)\n",
137 inode->i_sb->s_id, inode->i_ino);
140 /* Call generic open code in order to cache credentials */
141 res = nfs_open(inode, filp);
146 typedef __be32 * (*decode_dirent_t)(__be32 *, struct nfs_entry *, int);
150 unsigned long page_index;
153 loff_t current_index;
154 struct nfs_entry *entry;
155 decode_dirent_t decode;
158 } nfs_readdir_descriptor_t;
160 /* Now we cache directories properly, by stuffing the dirent
161 * data directly in the page cache.
163 * Inode invalidation due to refresh etc. takes care of
164 * _everything_, no sloppy entry flushing logic, no extraneous
165 * copying, network direct to page cache, the way it was meant
168 * NOTE: Dirent information verification is done always by the
169 * page-in of the RPC reply, nowhere else, this simplies
170 * things substantially.
173 int nfs_readdir_filler(nfs_readdir_descriptor_t *desc, struct page *page)
175 struct file *file = desc->file;
176 struct inode *inode = file->f_path.dentry->d_inode;
177 struct rpc_cred *cred = nfs_file_cred(file);
178 unsigned long timestamp;
181 dfprintk(DIRCACHE, "NFS: %s: reading cookie %Lu into page %lu\n",
182 __FUNCTION__, (long long)desc->entry->cookie,
187 error = NFS_PROTO(inode)->readdir(file->f_path.dentry, cred, desc->entry->cookie, page,
188 NFS_SERVER(inode)->dtsize, desc->plus);
190 /* We requested READDIRPLUS, but the server doesn't grok it */
191 if (error == -ENOTSUPP && desc->plus) {
192 NFS_SERVER(inode)->caps &= ~NFS_CAP_READDIRPLUS;
193 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_FLAGS(inode));
199 SetPageUptodate(page);
200 spin_lock(&inode->i_lock);
201 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ATIME;
202 spin_unlock(&inode->i_lock);
203 /* Ensure consistent page alignment of the data.
204 * Note: assumes we have exclusive access to this mapping either
205 * through inode->i_mutex or some other mechanism.
207 if (page->index == 0 && invalidate_inode_pages2_range(inode->i_mapping, PAGE_CACHE_SIZE, -1) < 0) {
208 /* Should never happen */
209 nfs_zap_mapping(inode, inode->i_mapping);
216 nfs_zap_caches(inode);
222 int dir_decode(nfs_readdir_descriptor_t *desc)
224 __be32 *p = desc->ptr;
225 p = desc->decode(p, desc->entry, desc->plus);
233 void dir_page_release(nfs_readdir_descriptor_t *desc)
236 page_cache_release(desc->page);
242 * Given a pointer to a buffer that has already been filled by a call
243 * to readdir, find the next entry with cookie '*desc->dir_cookie'.
245 * If the end of the buffer has been reached, return -EAGAIN, if not,
246 * return the offset within the buffer of the next entry to be
250 int find_dirent(nfs_readdir_descriptor_t *desc)
252 struct nfs_entry *entry = desc->entry;
256 while((status = dir_decode(desc)) == 0) {
257 dfprintk(DIRCACHE, "NFS: %s: examining cookie %Lu\n",
258 __FUNCTION__, (unsigned long long)entry->cookie);
259 if (entry->prev_cookie == *desc->dir_cookie)
261 if (loop_count++ > 200) {
270 * Given a pointer to a buffer that has already been filled by a call
271 * to readdir, find the entry at offset 'desc->file->f_pos'.
273 * If the end of the buffer has been reached, return -EAGAIN, if not,
274 * return the offset within the buffer of the next entry to be
278 int find_dirent_index(nfs_readdir_descriptor_t *desc)
280 struct nfs_entry *entry = desc->entry;
285 status = dir_decode(desc);
289 dfprintk(DIRCACHE, "NFS: found cookie %Lu at index %Ld\n",
290 (unsigned long long)entry->cookie, desc->current_index);
292 if (desc->file->f_pos == desc->current_index) {
293 *desc->dir_cookie = entry->cookie;
296 desc->current_index++;
297 if (loop_count++ > 200) {
306 * Find the given page, and call find_dirent() or find_dirent_index in
307 * order to try to return the next entry.
310 int find_dirent_page(nfs_readdir_descriptor_t *desc)
312 struct inode *inode = desc->file->f_path.dentry->d_inode;
316 dfprintk(DIRCACHE, "NFS: %s: searching page %ld for target %Lu\n",
317 __FUNCTION__, desc->page_index,
318 (long long) *desc->dir_cookie);
320 page = read_cache_page(inode->i_mapping, desc->page_index,
321 (filler_t *)nfs_readdir_filler, desc);
323 status = PTR_ERR(page);
326 if (!PageUptodate(page))
329 /* NOTE: Someone else may have changed the READDIRPLUS flag */
331 desc->ptr = kmap(page); /* matching kunmap in nfs_do_filldir */
332 if (*desc->dir_cookie != 0)
333 status = find_dirent(desc);
335 status = find_dirent_index(desc);
337 dir_page_release(desc);
339 dfprintk(DIRCACHE, "NFS: %s: returns %d\n", __FUNCTION__, status);
342 page_cache_release(page);
347 * Recurse through the page cache pages, and return a
348 * filled nfs_entry structure of the next directory entry if possible.
350 * The target for the search is '*desc->dir_cookie' if non-0,
351 * 'desc->file->f_pos' otherwise
354 int readdir_search_pagecache(nfs_readdir_descriptor_t *desc)
359 /* Always search-by-index from the beginning of the cache */
360 if (*desc->dir_cookie == 0) {
361 dfprintk(DIRCACHE, "NFS: readdir_search_pagecache() searching for offset %Ld\n",
362 (long long)desc->file->f_pos);
363 desc->page_index = 0;
364 desc->entry->cookie = desc->entry->prev_cookie = 0;
365 desc->entry->eof = 0;
366 desc->current_index = 0;
368 dfprintk(DIRCACHE, "NFS: readdir_search_pagecache() searching for cookie %Lu\n",
369 (unsigned long long)*desc->dir_cookie);
372 res = find_dirent_page(desc);
375 /* Align to beginning of next page */
377 if (loop_count++ > 200) {
383 dfprintk(DIRCACHE, "NFS: %s: returns %d\n", __FUNCTION__, res);
387 static inline unsigned int dt_type(struct inode *inode)
389 return (inode->i_mode >> 12) & 15;
392 static struct dentry *nfs_readdir_lookup(nfs_readdir_descriptor_t *desc);
395 * Once we've found the start of the dirent within a page: fill 'er up...
398 int nfs_do_filldir(nfs_readdir_descriptor_t *desc, void *dirent,
401 struct file *file = desc->file;
402 struct nfs_entry *entry = desc->entry;
403 struct dentry *dentry = NULL;
404 unsigned long fileid;
408 dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling starting @ cookie %Lu\n",
409 (unsigned long long)entry->cookie);
412 unsigned d_type = DT_UNKNOWN;
413 /* Note: entry->prev_cookie contains the cookie for
414 * retrieving the current dirent on the server */
415 fileid = nfs_fileid_to_ino_t(entry->ino);
417 /* Get a dentry if we have one */
420 dentry = nfs_readdir_lookup(desc);
422 /* Use readdirplus info */
423 if (dentry != NULL && dentry->d_inode != NULL) {
424 d_type = dt_type(dentry->d_inode);
425 fileid = dentry->d_inode->i_ino;
428 res = filldir(dirent, entry->name, entry->len,
429 file->f_pos, fileid, d_type);
433 *desc->dir_cookie = entry->cookie;
434 if (dir_decode(desc) != 0) {
438 if (loop_count++ > 200) {
443 dir_page_release(desc);
446 dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n",
447 (unsigned long long)*desc->dir_cookie, res);
452 * If we cannot find a cookie in our cache, we suspect that this is
453 * because it points to a deleted file, so we ask the server to return
454 * whatever it thinks is the next entry. We then feed this to filldir.
455 * If all goes well, we should then be able to find our way round the
456 * cache on the next call to readdir_search_pagecache();
458 * NOTE: we cannot add the anonymous page to the pagecache because
459 * the data it contains might not be page aligned. Besides,
460 * we should already have a complete representation of the
461 * directory in the page cache by the time we get here.
464 int uncached_readdir(nfs_readdir_descriptor_t *desc, void *dirent,
467 struct file *file = desc->file;
468 struct inode *inode = file->f_path.dentry->d_inode;
469 struct rpc_cred *cred = nfs_file_cred(file);
470 struct page *page = NULL;
473 dfprintk(DIRCACHE, "NFS: uncached_readdir() searching for cookie %Lu\n",
474 (unsigned long long)*desc->dir_cookie);
476 page = alloc_page(GFP_HIGHUSER);
481 desc->error = NFS_PROTO(inode)->readdir(file->f_path.dentry, cred, *desc->dir_cookie,
483 NFS_SERVER(inode)->dtsize,
485 spin_lock(&inode->i_lock);
486 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ATIME;
487 spin_unlock(&inode->i_lock);
489 desc->ptr = kmap(page); /* matching kunmap in nfs_do_filldir */
490 if (desc->error >= 0) {
491 if ((status = dir_decode(desc)) == 0)
492 desc->entry->prev_cookie = *desc->dir_cookie;
498 status = nfs_do_filldir(desc, dirent, filldir);
500 /* Reset read descriptor so it searches the page cache from
501 * the start upon the next call to readdir_search_pagecache() */
502 desc->page_index = 0;
503 desc->entry->cookie = desc->entry->prev_cookie = 0;
504 desc->entry->eof = 0;
506 dfprintk(DIRCACHE, "NFS: %s: returns %d\n",
507 __FUNCTION__, status);
510 dir_page_release(desc);
514 /* The file offset position represents the dirent entry number. A
515 last cookie cache takes care of the common case of reading the
518 static int nfs_readdir(struct file *filp, void *dirent, filldir_t filldir)
520 struct dentry *dentry = filp->f_path.dentry;
521 struct inode *inode = dentry->d_inode;
522 nfs_readdir_descriptor_t my_desc,
524 struct nfs_entry my_entry;
526 struct nfs_fattr fattr;
529 dfprintk(VFS, "NFS: readdir(%s/%s) starting at cookie %Lu\n",
530 dentry->d_parent->d_name.name, dentry->d_name.name,
531 (long long)filp->f_pos);
532 nfs_inc_stats(inode, NFSIOS_VFSGETDENTS);
536 res = nfs_revalidate_mapping_nolock(inode, filp->f_mapping);
543 * filp->f_pos points to the dirent entry number.
544 * *desc->dir_cookie has the cookie for the next entry. We have
545 * to either find the entry with the appropriate number or
546 * revalidate the cookie.
548 memset(desc, 0, sizeof(*desc));
551 desc->dir_cookie = &((struct nfs_open_context *)filp->private_data)->dir_cookie;
552 desc->decode = NFS_PROTO(inode)->decode_dirent;
553 desc->plus = NFS_USE_READDIRPLUS(inode);
555 my_entry.cookie = my_entry.prev_cookie = 0;
558 my_entry.fattr = &fattr;
559 nfs_fattr_init(&fattr);
560 desc->entry = &my_entry;
562 while(!desc->entry->eof) {
563 res = readdir_search_pagecache(desc);
565 if (res == -EBADCOOKIE) {
566 /* This means either end of directory */
567 if (*desc->dir_cookie && desc->entry->cookie != *desc->dir_cookie) {
568 /* Or that the server has 'lost' a cookie */
569 res = uncached_readdir(desc, dirent, filldir);
576 if (res == -ETOOSMALL && desc->plus) {
577 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_FLAGS(inode));
578 nfs_zap_caches(inode);
580 desc->entry->eof = 0;
586 res = nfs_do_filldir(desc, dirent, filldir);
595 dfprintk(VFS, "NFS: readdir(%s/%s) returns %ld\n",
596 dentry->d_parent->d_name.name, dentry->d_name.name,
601 loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int origin)
603 mutex_lock(&filp->f_path.dentry->d_inode->i_mutex);
606 offset += filp->f_pos;
614 if (offset != filp->f_pos) {
615 filp->f_pos = offset;
616 ((struct nfs_open_context *)filp->private_data)->dir_cookie = 0;
619 mutex_unlock(&filp->f_path.dentry->d_inode->i_mutex);
624 * All directory operations under NFS are synchronous, so fsync()
625 * is a dummy operation.
627 int nfs_fsync_dir(struct file *filp, struct dentry *dentry, int datasync)
629 dfprintk(VFS, "NFS: fsync_dir(%s/%s) datasync %d\n",
630 dentry->d_parent->d_name.name, dentry->d_name.name,
637 * A check for whether or not the parent directory has changed.
638 * In the case it has, we assume that the dentries are untrustworthy
639 * and may need to be looked up again.
641 static inline int nfs_check_verifier(struct inode *dir, struct dentry *dentry)
645 if ((NFS_I(dir)->cache_validity & NFS_INO_INVALID_ATTR) != 0
646 || nfs_attribute_timeout(dir))
648 return nfs_verify_change_attribute(dir, (unsigned long)dentry->d_fsdata);
651 static inline void nfs_set_verifier(struct dentry * dentry, unsigned long verf)
653 dentry->d_fsdata = (void *)verf;
657 * Whenever an NFS operation succeeds, we know that the dentry
658 * is valid, so we update the revalidation timestamp.
660 static inline void nfs_renew_times(struct dentry * dentry)
662 dentry->d_time = jiffies;
666 * Return the intent data that applies to this particular path component
668 * Note that the current set of intents only apply to the very last
669 * component of the path.
670 * We check for this using LOOKUP_CONTINUE and LOOKUP_PARENT.
672 static inline unsigned int nfs_lookup_check_intent(struct nameidata *nd, unsigned int mask)
674 if (nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))
676 return nd->flags & mask;
680 * Inode and filehandle revalidation for lookups.
682 * We force revalidation in the cases where the VFS sets LOOKUP_REVAL,
683 * or if the intent information indicates that we're about to open this
684 * particular file and the "nocto" mount flag is not set.
688 int nfs_lookup_verify_inode(struct inode *inode, struct nameidata *nd)
690 struct nfs_server *server = NFS_SERVER(inode);
693 /* VFS wants an on-the-wire revalidation */
694 if (nd->flags & LOOKUP_REVAL)
696 /* This is an open(2) */
697 if (nfs_lookup_check_intent(nd, LOOKUP_OPEN) != 0 &&
698 !(server->flags & NFS_MOUNT_NOCTO) &&
699 (S_ISREG(inode->i_mode) ||
700 S_ISDIR(inode->i_mode)))
703 return nfs_revalidate_inode(server, inode);
705 return __nfs_revalidate_inode(server, inode);
709 * We judge how long we want to trust negative
710 * dentries by looking at the parent inode mtime.
712 * If parent mtime has changed, we revalidate, else we wait for a
713 * period corresponding to the parent's attribute cache timeout value.
716 int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry,
717 struct nameidata *nd)
719 /* Don't revalidate a negative dentry if we're creating a new file */
720 if (nd != NULL && nfs_lookup_check_intent(nd, LOOKUP_CREATE) != 0)
722 return !nfs_check_verifier(dir, dentry);
726 * This is called every time the dcache has a lookup hit,
727 * and we should check whether we can really trust that
730 * NOTE! The hit can be a negative hit too, don't assume
733 * If the parent directory is seen to have changed, we throw out the
734 * cached dentry and do a new lookup.
736 static int nfs_lookup_revalidate(struct dentry * dentry, struct nameidata *nd)
740 struct dentry *parent;
742 struct nfs_fh fhandle;
743 struct nfs_fattr fattr;
744 unsigned long verifier;
746 parent = dget_parent(dentry);
748 dir = parent->d_inode;
749 nfs_inc_stats(dir, NFSIOS_DENTRYREVALIDATE);
750 inode = dentry->d_inode;
753 if (nfs_neg_need_reval(dir, dentry, nd))
758 if (is_bad_inode(inode)) {
759 dfprintk(LOOKUPCACHE, "%s: %s/%s has dud inode\n",
760 __FUNCTION__, dentry->d_parent->d_name.name,
761 dentry->d_name.name);
765 /* Revalidate parent directory attribute cache */
766 if (nfs_revalidate_inode(NFS_SERVER(dir), dir) < 0)
769 /* Force a full look up iff the parent directory has changed */
770 if (nfs_check_verifier(dir, dentry)) {
771 if (nfs_lookup_verify_inode(inode, nd))
776 if (NFS_STALE(inode))
779 verifier = nfs_save_change_attribute(dir);
780 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, &fhandle, &fattr);
783 if (nfs_compare_fh(NFS_FH(inode), &fhandle))
785 if ((error = nfs_refresh_inode(inode, &fattr)) != 0)
788 nfs_renew_times(dentry);
789 nfs_set_verifier(dentry, verifier);
793 dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is valid\n",
794 __FUNCTION__, dentry->d_parent->d_name.name,
795 dentry->d_name.name);
801 if (inode && S_ISDIR(inode->i_mode)) {
802 /* Purge readdir caches. */
803 nfs_zap_caches(inode);
804 /* If we have submounts, don't unhash ! */
805 if (have_submounts(dentry))
807 shrink_dcache_parent(dentry);
812 dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is invalid\n",
813 __FUNCTION__, dentry->d_parent->d_name.name,
814 dentry->d_name.name);
819 * This is called from dput() when d_count is going to 0.
821 static int nfs_dentry_delete(struct dentry *dentry)
823 dfprintk(VFS, "NFS: dentry_delete(%s/%s, %x)\n",
824 dentry->d_parent->d_name.name, dentry->d_name.name,
827 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
828 /* Unhash it, so that ->d_iput() would be called */
831 if (!(dentry->d_sb->s_flags & MS_ACTIVE)) {
832 /* Unhash it, so that ancestors of killed async unlink
833 * files will be cleaned up during umount */
841 * Called when the dentry loses inode.
842 * We use it to clean up silly-renamed files.
844 static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode)
846 nfs_inode_return_delegation(inode);
847 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
850 nfs_complete_unlink(dentry);
853 /* When creating a negative dentry, we want to renew d_time */
854 nfs_renew_times(dentry);
858 struct dentry_operations nfs_dentry_operations = {
859 .d_revalidate = nfs_lookup_revalidate,
860 .d_delete = nfs_dentry_delete,
861 .d_iput = nfs_dentry_iput,
865 * Use intent information to check whether or not we're going to do
866 * an O_EXCL create using this path component.
869 int nfs_is_exclusive_create(struct inode *dir, struct nameidata *nd)
871 if (NFS_PROTO(dir)->version == 2)
873 if (nd == NULL || nfs_lookup_check_intent(nd, LOOKUP_CREATE) == 0)
875 return (nd->intent.open.flags & O_EXCL) != 0;
878 static inline int nfs_reval_fsid(struct vfsmount *mnt, struct inode *dir,
879 struct nfs_fh *fh, struct nfs_fattr *fattr)
881 struct nfs_server *server = NFS_SERVER(dir);
883 if (!nfs_fsid_equal(&server->fsid, &fattr->fsid))
884 /* Revalidate fsid on root dir */
885 return __nfs_revalidate_inode(server, mnt->mnt_root->d_inode);
889 static struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
892 struct inode *inode = NULL;
894 struct nfs_fh fhandle;
895 struct nfs_fattr fattr;
897 dfprintk(VFS, "NFS: lookup(%s/%s)\n",
898 dentry->d_parent->d_name.name, dentry->d_name.name);
899 nfs_inc_stats(dir, NFSIOS_VFSLOOKUP);
901 res = ERR_PTR(-ENAMETOOLONG);
902 if (dentry->d_name.len > NFS_SERVER(dir)->namelen)
905 res = ERR_PTR(-ENOMEM);
906 dentry->d_op = NFS_PROTO(dir)->dentry_ops;
911 * If we're doing an exclusive create, optimize away the lookup
912 * but don't hash the dentry.
914 if (nfs_is_exclusive_create(dir, nd)) {
915 d_instantiate(dentry, NULL);
920 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, &fhandle, &fattr);
921 if (error == -ENOENT)
924 res = ERR_PTR(error);
927 error = nfs_reval_fsid(nd->mnt, dir, &fhandle, &fattr);
929 res = ERR_PTR(error);
932 inode = nfs_fhget(dentry->d_sb, &fhandle, &fattr);
933 res = (struct dentry *)inode;
937 dx_propagate_tag(nd, inode);
939 res = d_materialise_unique(dentry, inode);
941 struct dentry *parent;
944 /* Was a directory renamed! */
945 parent = dget_parent(res);
946 if (!IS_ROOT(parent))
947 nfs_mark_for_revalidate(parent->d_inode);
951 nfs_renew_times(dentry);
952 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
960 static int nfs_open_revalidate(struct dentry *, struct nameidata *);
962 struct dentry_operations nfs4_dentry_operations = {
963 .d_revalidate = nfs_open_revalidate,
964 .d_delete = nfs_dentry_delete,
965 .d_iput = nfs_dentry_iput,
969 * Use intent information to determine whether we need to substitute
970 * the NFSv4-style stateful OPEN for the LOOKUP call
972 static int is_atomic_open(struct inode *dir, struct nameidata *nd)
974 if (nd == NULL || nfs_lookup_check_intent(nd, LOOKUP_OPEN) == 0)
976 /* NFS does not (yet) have a stateful open for directories */
977 if (nd->flags & LOOKUP_DIRECTORY)
979 /* Are we trying to write to a read only partition? */
980 if ((IS_RDONLY(dir) || MNT_IS_RDONLY(nd->mnt)) &&
981 (nd->intent.open.flags & (O_CREAT|O_TRUNC|FMODE_WRITE)))
986 static struct dentry *nfs_atomic_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd)
988 struct dentry *res = NULL;
991 dfprintk(VFS, "NFS: atomic_lookup(%s/%ld), %s\n",
992 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
994 /* Check that we are indeed trying to open this file */
995 if (!is_atomic_open(dir, nd))
998 if (dentry->d_name.len > NFS_SERVER(dir)->namelen) {
999 res = ERR_PTR(-ENAMETOOLONG);
1002 dentry->d_op = NFS_PROTO(dir)->dentry_ops;
1004 /* Let vfs_create() deal with O_EXCL */
1005 if (nd->intent.open.flags & O_EXCL) {
1006 d_add(dentry, NULL);
1010 /* Open the file on the server */
1012 /* Revalidate parent directory attribute cache */
1013 error = nfs_revalidate_inode(NFS_SERVER(dir), dir);
1015 res = ERR_PTR(error);
1020 if (nd->intent.open.flags & O_CREAT) {
1021 nfs_begin_data_update(dir);
1022 res = nfs4_atomic_open(dir, dentry, nd);
1023 nfs_end_data_update(dir);
1025 res = nfs4_atomic_open(dir, dentry, nd);
1028 error = PTR_ERR(res);
1030 /* Make a negative dentry */
1034 /* This turned out not to be a regular file */
1039 if (!(nd->intent.open.flags & O_NOFOLLOW))
1045 } else if (res != NULL)
1047 nfs_renew_times(dentry);
1048 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1052 return nfs_lookup(dir, dentry, nd);
1055 static int nfs_open_revalidate(struct dentry *dentry, struct nameidata *nd)
1057 struct dentry *parent = NULL;
1058 struct inode *inode = dentry->d_inode;
1060 unsigned long verifier;
1061 int openflags, ret = 0;
1063 parent = dget_parent(dentry);
1064 dir = parent->d_inode;
1065 if (!is_atomic_open(dir, nd))
1067 /* We can't create new files in nfs_open_revalidate(), so we
1068 * optimize away revalidation of negative dentries.
1072 /* NFS only supports OPEN on regular files */
1073 if (!S_ISREG(inode->i_mode))
1075 openflags = nd->intent.open.flags;
1076 /* We cannot do exclusive creation on a positive dentry */
1077 if ((openflags & (O_CREAT|O_EXCL)) == (O_CREAT|O_EXCL))
1079 /* We can't create new files, or truncate existing ones here */
1080 openflags &= ~(O_CREAT|O_TRUNC);
1083 * Note: we're not holding inode->i_mutex and so may be racing with
1084 * operations that change the directory. We therefore save the
1085 * change attribute *before* we do the RPC call.
1088 verifier = nfs_save_change_attribute(dir);
1089 ret = nfs4_open_revalidate(dir, dentry, openflags, nd);
1091 nfs_set_verifier(dentry, verifier);
1100 if (inode != NULL && nfs_have_delegation(inode, FMODE_READ))
1102 return nfs_lookup_revalidate(dentry, nd);
1104 #endif /* CONFIG_NFSV4 */
1106 static struct dentry *nfs_readdir_lookup(nfs_readdir_descriptor_t *desc)
1108 struct dentry *parent = desc->file->f_path.dentry;
1109 struct inode *dir = parent->d_inode;
1110 struct nfs_entry *entry = desc->entry;
1111 struct dentry *dentry, *alias;
1112 struct qstr name = {
1113 .name = entry->name,
1116 struct inode *inode;
1120 if (name.name[0] == '.' && name.name[1] == '.')
1121 return dget_parent(parent);
1124 if (name.name[0] == '.')
1125 return dget(parent);
1127 name.hash = full_name_hash(name.name, name.len);
1128 dentry = d_lookup(parent, &name);
1131 if (!desc->plus || !(entry->fattr->valid & NFS_ATTR_FATTR))
1133 /* Note: caller is already holding the dir->i_mutex! */
1134 dentry = d_alloc(parent, &name);
1137 dentry->d_op = NFS_PROTO(dir)->dentry_ops;
1138 inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr);
1139 if (IS_ERR(inode)) {
1144 alias = d_materialise_unique(dentry, inode);
1145 if (alias != NULL) {
1152 nfs_renew_times(dentry);
1153 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1158 * Code common to create, mkdir, and mknod.
1160 int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle,
1161 struct nfs_fattr *fattr)
1163 struct inode *inode;
1164 int error = -EACCES;
1166 /* We may have been initialized further down */
1167 if (dentry->d_inode)
1169 if (fhandle->size == 0) {
1170 struct inode *dir = dentry->d_parent->d_inode;
1171 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr);
1175 if (!(fattr->valid & NFS_ATTR_FATTR)) {
1176 struct nfs_server *server = NFS_SB(dentry->d_sb);
1177 error = server->nfs_client->rpc_ops->getattr(server, fhandle, fattr);
1181 inode = nfs_fhget(dentry->d_sb, fhandle, fattr);
1182 error = PTR_ERR(inode);
1185 d_instantiate(dentry, inode);
1186 if (d_unhashed(dentry))
1192 * Following a failed create operation, we drop the dentry rather
1193 * than retain a negative dentry. This avoids a problem in the event
1194 * that the operation succeeded on the server, but an error in the
1195 * reply path made it appear to have failed.
1197 static int nfs_create(struct inode *dir, struct dentry *dentry, int mode,
1198 struct nameidata *nd)
1204 dfprintk(VFS, "NFS: create(%s/%ld), %s\n",
1205 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1207 attr.ia_mode = mode;
1208 attr.ia_valid = ATTR_MODE;
1210 if (nd && (nd->flags & LOOKUP_CREATE))
1211 open_flags = nd->intent.open.flags;
1214 nfs_begin_data_update(dir);
1215 error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags, nd);
1216 nfs_end_data_update(dir);
1219 nfs_renew_times(dentry);
1220 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1230 * See comments for nfs_proc_create regarding failed operations.
1233 nfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t rdev)
1238 dfprintk(VFS, "NFS: mknod(%s/%ld), %s\n",
1239 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1241 if (!new_valid_dev(rdev))
1244 attr.ia_mode = mode;
1245 attr.ia_valid = ATTR_MODE;
1248 nfs_begin_data_update(dir);
1249 status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev);
1250 nfs_end_data_update(dir);
1253 nfs_renew_times(dentry);
1254 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1264 * See comments for nfs_proc_create regarding failed operations.
1266 static int nfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1271 dfprintk(VFS, "NFS: mkdir(%s/%ld), %s\n",
1272 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1274 attr.ia_valid = ATTR_MODE;
1275 attr.ia_mode = mode | S_IFDIR;
1278 nfs_begin_data_update(dir);
1279 error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr);
1280 nfs_end_data_update(dir);
1283 nfs_renew_times(dentry);
1284 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
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);
1301 nfs_begin_data_update(dir);
1302 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
1303 /* Ensure the VFS deletes this inode */
1304 if (error == 0 && dentry->d_inode != NULL)
1305 clear_nlink(dentry->d_inode);
1306 nfs_end_data_update(dir);
1312 static int nfs_sillyrename(struct inode *dir, struct dentry *dentry)
1314 static unsigned int sillycounter;
1315 const int i_inosize = sizeof(dir->i_ino)*2;
1316 const int countersize = sizeof(sillycounter)*2;
1317 const int slen = sizeof(".nfs") + i_inosize + countersize - 1;
1320 struct dentry *sdentry;
1323 dfprintk(VFS, "NFS: silly-rename(%s/%s, ct=%d)\n",
1324 dentry->d_parent->d_name.name, dentry->d_name.name,
1325 atomic_read(&dentry->d_count));
1326 nfs_inc_stats(dir, NFSIOS_SILLYRENAME);
1329 if (!dentry->d_inode)
1330 printk("NFS: silly-renaming %s/%s, negative dentry??\n",
1331 dentry->d_parent->d_name.name, dentry->d_name.name);
1334 * We don't allow a dentry to be silly-renamed twice.
1337 if (dentry->d_flags & DCACHE_NFSFS_RENAMED)
1340 sprintf(silly, ".nfs%*.*lx",
1341 i_inosize, i_inosize, dentry->d_inode->i_ino);
1343 /* Return delegation in anticipation of the rename */
1344 nfs_inode_return_delegation(dentry->d_inode);
1348 char *suffix = silly + slen - countersize;
1352 sprintf(suffix, "%*.*x", countersize, countersize, sillycounter);
1354 dfprintk(VFS, "NFS: trying to rename %s to %s\n",
1355 dentry->d_name.name, silly);
1357 sdentry = lookup_one_len(silly, dentry->d_parent, slen);
1359 * N.B. Better to return EBUSY here ... it could be
1360 * dangerous to delete the file while it's in use.
1362 if (IS_ERR(sdentry))
1364 } while(sdentry->d_inode != NULL); /* need negative lookup */
1366 qsilly.name = silly;
1367 qsilly.len = strlen(silly);
1368 nfs_begin_data_update(dir);
1369 if (dentry->d_inode) {
1370 nfs_begin_data_update(dentry->d_inode);
1371 error = NFS_PROTO(dir)->rename(dir, &dentry->d_name,
1373 nfs_mark_for_revalidate(dentry->d_inode);
1374 nfs_end_data_update(dentry->d_inode);
1376 error = NFS_PROTO(dir)->rename(dir, &dentry->d_name,
1378 nfs_end_data_update(dir);
1380 nfs_renew_times(dentry);
1381 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1382 d_move(dentry, sdentry);
1383 error = nfs_async_unlink(dentry);
1384 /* If we return 0 we don't unlink */
1392 * Remove a file after making sure there are no pending writes,
1393 * and after checking that the file has only one user.
1395 * We invalidate the attribute cache and free the inode prior to the operation
1396 * to avoid possible races if the server reuses the inode.
1398 static int nfs_safe_remove(struct dentry *dentry)
1400 struct inode *dir = dentry->d_parent->d_inode;
1401 struct inode *inode = dentry->d_inode;
1404 dfprintk(VFS, "NFS: safe_remove(%s/%s)\n",
1405 dentry->d_parent->d_name.name, dentry->d_name.name);
1407 /* If the dentry was sillyrenamed, we simply call d_delete() */
1408 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1413 nfs_begin_data_update(dir);
1414 if (inode != NULL) {
1415 nfs_inode_return_delegation(inode);
1416 nfs_begin_data_update(inode);
1417 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1418 /* The VFS may want to delete this inode */
1421 nfs_mark_for_revalidate(inode);
1422 nfs_end_data_update(inode);
1424 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1425 nfs_end_data_update(dir);
1430 /* We do silly rename. In case sillyrename() returns -EBUSY, the inode
1431 * belongs to an active ".nfs..." file and we return -EBUSY.
1433 * If sillyrename() returns 0, we do nothing, otherwise we unlink.
1435 static int nfs_unlink(struct inode *dir, struct dentry *dentry)
1438 int need_rehash = 0;
1440 dfprintk(VFS, "NFS: unlink(%s/%ld, %s)\n", dir->i_sb->s_id,
1441 dir->i_ino, dentry->d_name.name);
1444 spin_lock(&dcache_lock);
1445 spin_lock(&dentry->d_lock);
1446 if (atomic_read(&dentry->d_count) > 1) {
1447 spin_unlock(&dentry->d_lock);
1448 spin_unlock(&dcache_lock);
1449 error = nfs_sillyrename(dir, dentry);
1453 if (!d_unhashed(dentry)) {
1457 spin_unlock(&dentry->d_lock);
1458 spin_unlock(&dcache_lock);
1459 error = nfs_safe_remove(dentry);
1461 nfs_renew_times(dentry);
1462 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1463 } else if (need_rehash)
1470 * To create a symbolic link, most file systems instantiate a new inode,
1471 * add a page to it containing the path, then write it out to the disk
1472 * using prepare_write/commit_write.
1474 * Unfortunately the NFS client can't create the in-core inode first
1475 * because it needs a file handle to create an in-core inode (see
1476 * fs/nfs/inode.c:nfs_fhget). We only have a file handle *after* the
1477 * symlink request has completed on the server.
1479 * So instead we allocate a raw page, copy the symname into it, then do
1480 * the SYMLINK request with the page as the buffer. If it succeeds, we
1481 * now have a new file handle and can instantiate an in-core NFS inode
1482 * and move the raw page into its mapping.
1484 static int nfs_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1486 struct pagevec lru_pvec;
1490 unsigned int pathlen = strlen(symname);
1493 dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s)\n", dir->i_sb->s_id,
1494 dir->i_ino, dentry->d_name.name, symname);
1496 if (pathlen > PAGE_SIZE)
1497 return -ENAMETOOLONG;
1499 attr.ia_mode = S_IFLNK | S_IRWXUGO;
1500 attr.ia_valid = ATTR_MODE;
1504 page = alloc_page(GFP_KERNEL);
1510 kaddr = kmap_atomic(page, KM_USER0);
1511 memcpy(kaddr, symname, pathlen);
1512 if (pathlen < PAGE_SIZE)
1513 memset(kaddr + pathlen, 0, PAGE_SIZE - pathlen);
1514 kunmap_atomic(kaddr, KM_USER0);
1516 nfs_begin_data_update(dir);
1517 error = NFS_PROTO(dir)->symlink(dir, dentry, page, pathlen, &attr);
1518 nfs_end_data_update(dir);
1520 dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s) error %d\n",
1521 dir->i_sb->s_id, dir->i_ino,
1522 dentry->d_name.name, symname, error);
1530 * No big deal if we can't add this page to the page cache here.
1531 * READLINK will get the missing page from the server if needed.
1533 pagevec_init(&lru_pvec, 0);
1534 if (!add_to_page_cache(page, dentry->d_inode->i_mapping, 0,
1536 pagevec_add(&lru_pvec, page);
1537 pagevec_lru_add(&lru_pvec);
1538 SetPageUptodate(page);
1548 nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1550 struct inode *inode = old_dentry->d_inode;
1553 dfprintk(VFS, "NFS: link(%s/%s -> %s/%s)\n",
1554 old_dentry->d_parent->d_name.name, old_dentry->d_name.name,
1555 dentry->d_parent->d_name.name, dentry->d_name.name);
1558 nfs_begin_data_update(dir);
1559 nfs_begin_data_update(inode);
1560 error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name);
1562 atomic_inc(&inode->i_count);
1563 d_instantiate(dentry, inode);
1565 nfs_end_data_update(inode);
1566 nfs_end_data_update(dir);
1573 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a
1574 * different file handle for the same inode after a rename (e.g. when
1575 * moving to a different directory). A fail-safe method to do so would
1576 * be to look up old_dir/old_name, create a link to new_dir/new_name and
1577 * rename the old file using the sillyrename stuff. This way, the original
1578 * file in old_dir will go away when the last process iput()s the inode.
1582 * It actually works quite well. One needs to have the possibility for
1583 * at least one ".nfs..." file in each directory the file ever gets
1584 * moved or linked to which happens automagically with the new
1585 * implementation that only depends on the dcache stuff instead of
1586 * using the inode layer
1588 * Unfortunately, things are a little more complicated than indicated
1589 * above. For a cross-directory move, we want to make sure we can get
1590 * rid of the old inode after the operation. This means there must be
1591 * no pending writes (if it's a file), and the use count must be 1.
1592 * If these conditions are met, we can drop the dentries before doing
1595 static int nfs_rename(struct inode *old_dir, struct dentry *old_dentry,
1596 struct inode *new_dir, struct dentry *new_dentry)
1598 struct inode *old_inode = old_dentry->d_inode;
1599 struct inode *new_inode = new_dentry->d_inode;
1600 struct dentry *dentry = NULL, *rehash = NULL;
1604 * To prevent any new references to the target during the rename,
1605 * we unhash the dentry and free the inode in advance.
1608 if (!d_unhashed(new_dentry)) {
1610 rehash = new_dentry;
1613 dfprintk(VFS, "NFS: rename(%s/%s -> %s/%s, ct=%d)\n",
1614 old_dentry->d_parent->d_name.name, old_dentry->d_name.name,
1615 new_dentry->d_parent->d_name.name, new_dentry->d_name.name,
1616 atomic_read(&new_dentry->d_count));
1619 * First check whether the target is busy ... we can't
1620 * safely do _any_ rename if the target is in use.
1622 * For files, make a copy of the dentry and then do a
1623 * silly-rename. If the silly-rename succeeds, the
1624 * copied dentry is hashed and becomes the new target.
1628 if (S_ISDIR(new_inode->i_mode)) {
1630 if (!S_ISDIR(old_inode->i_mode))
1632 } else if (atomic_read(&new_dentry->d_count) > 2) {
1634 /* copy the target dentry's name */
1635 dentry = d_alloc(new_dentry->d_parent,
1636 &new_dentry->d_name);
1640 /* silly-rename the existing target ... */
1641 err = nfs_sillyrename(new_dir, new_dentry);
1643 new_dentry = rehash = dentry;
1645 /* instantiate the replacement target */
1646 d_instantiate(new_dentry, NULL);
1647 } else if (atomic_read(&new_dentry->d_count) > 1) {
1648 /* dentry still busy? */
1650 printk("nfs_rename: target %s/%s busy, d_count=%d\n",
1651 new_dentry->d_parent->d_name.name,
1652 new_dentry->d_name.name,
1653 atomic_read(&new_dentry->d_count));
1658 drop_nlink(new_inode);
1662 * ... prune child dentries and writebacks if needed.
1664 if (atomic_read(&old_dentry->d_count) > 1) {
1665 if (S_ISREG(old_inode->i_mode))
1666 nfs_wb_all(old_inode);
1667 shrink_dcache_parent(old_dentry);
1669 nfs_inode_return_delegation(old_inode);
1671 if (new_inode != NULL) {
1672 nfs_inode_return_delegation(new_inode);
1673 d_delete(new_dentry);
1676 nfs_begin_data_update(old_dir);
1677 nfs_begin_data_update(new_dir);
1678 nfs_begin_data_update(old_inode);
1679 error = NFS_PROTO(old_dir)->rename(old_dir, &old_dentry->d_name,
1680 new_dir, &new_dentry->d_name);
1681 nfs_mark_for_revalidate(old_inode);
1682 nfs_end_data_update(old_inode);
1683 nfs_end_data_update(new_dir);
1684 nfs_end_data_update(old_dir);
1689 d_move(old_dentry, new_dentry);
1690 nfs_renew_times(new_dentry);
1691 nfs_set_verifier(new_dentry, nfs_save_change_attribute(new_dir));
1694 /* new dentry created? */
1701 static DEFINE_SPINLOCK(nfs_access_lru_lock);
1702 static LIST_HEAD(nfs_access_lru_list);
1703 static atomic_long_t nfs_access_nr_entries;
1705 static void nfs_access_free_entry(struct nfs_access_entry *entry)
1707 put_rpccred(entry->cred);
1709 smp_mb__before_atomic_dec();
1710 atomic_long_dec(&nfs_access_nr_entries);
1711 smp_mb__after_atomic_dec();
1714 int nfs_access_cache_shrinker(int nr_to_scan, gfp_t gfp_mask)
1717 struct nfs_inode *nfsi;
1718 struct nfs_access_entry *cache;
1720 spin_lock(&nfs_access_lru_lock);
1722 list_for_each_entry(nfsi, &nfs_access_lru_list, access_cache_inode_lru) {
1723 struct inode *inode;
1725 if (nr_to_scan-- == 0)
1727 inode = igrab(&nfsi->vfs_inode);
1730 spin_lock(&inode->i_lock);
1731 if (list_empty(&nfsi->access_cache_entry_lru))
1732 goto remove_lru_entry;
1733 cache = list_entry(nfsi->access_cache_entry_lru.next,
1734 struct nfs_access_entry, lru);
1735 list_move(&cache->lru, &head);
1736 rb_erase(&cache->rb_node, &nfsi->access_cache);
1737 if (!list_empty(&nfsi->access_cache_entry_lru))
1738 list_move_tail(&nfsi->access_cache_inode_lru,
1739 &nfs_access_lru_list);
1742 list_del_init(&nfsi->access_cache_inode_lru);
1743 clear_bit(NFS_INO_ACL_LRU_SET, &nfsi->flags);
1745 spin_unlock(&inode->i_lock);
1749 spin_unlock(&nfs_access_lru_lock);
1750 while (!list_empty(&head)) {
1751 cache = list_entry(head.next, struct nfs_access_entry, lru);
1752 list_del(&cache->lru);
1753 nfs_access_free_entry(cache);
1755 return (atomic_long_read(&nfs_access_nr_entries) / 100) * sysctl_vfs_cache_pressure;
1758 static void __nfs_access_zap_cache(struct inode *inode)
1760 struct nfs_inode *nfsi = NFS_I(inode);
1761 struct rb_root *root_node = &nfsi->access_cache;
1762 struct rb_node *n, *dispose = NULL;
1763 struct nfs_access_entry *entry;
1765 /* Unhook entries from the cache */
1766 while ((n = rb_first(root_node)) != NULL) {
1767 entry = rb_entry(n, struct nfs_access_entry, rb_node);
1768 rb_erase(n, root_node);
1769 list_del(&entry->lru);
1770 n->rb_left = dispose;
1773 nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS;
1774 spin_unlock(&inode->i_lock);
1776 /* Now kill them all! */
1777 while (dispose != NULL) {
1779 dispose = n->rb_left;
1780 nfs_access_free_entry(rb_entry(n, struct nfs_access_entry, rb_node));
1784 void nfs_access_zap_cache(struct inode *inode)
1786 /* Remove from global LRU init */
1787 if (test_and_clear_bit(NFS_INO_ACL_LRU_SET, &NFS_FLAGS(inode))) {
1788 spin_lock(&nfs_access_lru_lock);
1789 list_del_init(&NFS_I(inode)->access_cache_inode_lru);
1790 spin_unlock(&nfs_access_lru_lock);
1793 spin_lock(&inode->i_lock);
1794 /* This will release the spinlock */
1795 __nfs_access_zap_cache(inode);
1798 static struct nfs_access_entry *nfs_access_search_rbtree(struct inode *inode, struct rpc_cred *cred)
1800 struct rb_node *n = NFS_I(inode)->access_cache.rb_node;
1801 struct nfs_access_entry *entry;
1804 entry = rb_entry(n, struct nfs_access_entry, rb_node);
1806 if (cred < entry->cred)
1808 else if (cred > entry->cred)
1816 int nfs_access_get_cached(struct inode *inode, struct rpc_cred *cred, struct nfs_access_entry *res)
1818 struct nfs_inode *nfsi = NFS_I(inode);
1819 struct nfs_access_entry *cache;
1822 spin_lock(&inode->i_lock);
1823 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS)
1825 cache = nfs_access_search_rbtree(inode, cred);
1828 if (time_after(jiffies, cache->jiffies + NFS_ATTRTIMEO(inode)))
1830 res->jiffies = cache->jiffies;
1831 res->cred = cache->cred;
1832 res->mask = cache->mask;
1833 list_move_tail(&cache->lru, &nfsi->access_cache_entry_lru);
1836 spin_unlock(&inode->i_lock);
1839 rb_erase(&cache->rb_node, &nfsi->access_cache);
1840 list_del(&cache->lru);
1841 spin_unlock(&inode->i_lock);
1842 nfs_access_free_entry(cache);
1845 /* This will release the spinlock */
1846 __nfs_access_zap_cache(inode);
1850 static void nfs_access_add_rbtree(struct inode *inode, struct nfs_access_entry *set)
1852 struct nfs_inode *nfsi = NFS_I(inode);
1853 struct rb_root *root_node = &nfsi->access_cache;
1854 struct rb_node **p = &root_node->rb_node;
1855 struct rb_node *parent = NULL;
1856 struct nfs_access_entry *entry;
1858 spin_lock(&inode->i_lock);
1859 while (*p != NULL) {
1861 entry = rb_entry(parent, struct nfs_access_entry, rb_node);
1863 if (set->cred < entry->cred)
1864 p = &parent->rb_left;
1865 else if (set->cred > entry->cred)
1866 p = &parent->rb_right;
1870 rb_link_node(&set->rb_node, parent, p);
1871 rb_insert_color(&set->rb_node, root_node);
1872 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
1873 spin_unlock(&inode->i_lock);
1876 rb_replace_node(parent, &set->rb_node, root_node);
1877 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
1878 list_del(&entry->lru);
1879 spin_unlock(&inode->i_lock);
1880 nfs_access_free_entry(entry);
1883 void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set)
1885 struct nfs_access_entry *cache = kmalloc(sizeof(*cache), GFP_KERNEL);
1888 RB_CLEAR_NODE(&cache->rb_node);
1889 cache->jiffies = set->jiffies;
1890 cache->cred = get_rpccred(set->cred);
1891 cache->mask = set->mask;
1893 nfs_access_add_rbtree(inode, cache);
1895 /* Update accounting */
1896 smp_mb__before_atomic_inc();
1897 atomic_long_inc(&nfs_access_nr_entries);
1898 smp_mb__after_atomic_inc();
1900 /* Add inode to global LRU list */
1901 if (!test_and_set_bit(NFS_INO_ACL_LRU_SET, &NFS_FLAGS(inode))) {
1902 spin_lock(&nfs_access_lru_lock);
1903 list_add_tail(&NFS_I(inode)->access_cache_inode_lru, &nfs_access_lru_list);
1904 spin_unlock(&nfs_access_lru_lock);
1908 static int nfs_do_access(struct inode *inode, struct rpc_cred *cred, int mask)
1910 struct nfs_access_entry cache;
1913 status = nfs_access_get_cached(inode, cred, &cache);
1917 /* Be clever: ask server to check for all possible rights */
1918 cache.mask = MAY_EXEC | MAY_WRITE | MAY_READ;
1920 cache.jiffies = jiffies;
1921 status = NFS_PROTO(inode)->access(inode, &cache);
1924 nfs_access_add_cache(inode, &cache);
1926 if ((cache.mask & mask) == mask)
1931 int nfs_permission(struct inode *inode, int mask, struct nameidata *nd)
1933 struct rpc_cred *cred;
1936 nfs_inc_stats(inode, NFSIOS_VFSACCESS);
1940 /* Is this sys_access() ? */
1941 if (nd != NULL && (nd->flags & LOOKUP_ACCESS))
1944 switch (inode->i_mode & S_IFMT) {
1948 /* NFSv4 has atomic_open... */
1949 if (nfs_server_capable(inode, NFS_CAP_ATOMIC_OPEN)
1951 && (nd->flags & LOOKUP_OPEN))
1956 * Optimize away all write operations, since the server
1957 * will check permissions when we perform the op.
1959 if ((mask & MAY_WRITE) && !(mask & MAY_READ))
1966 if (!NFS_PROTO(inode)->access)
1969 cred = rpcauth_lookupcred(NFS_CLIENT(inode)->cl_auth, 0);
1970 if (!IS_ERR(cred)) {
1971 res = nfs_do_access(inode, cred, mask);
1974 res = PTR_ERR(cred);
1977 dfprintk(VFS, "NFS: permission(%s/%ld), mask=0x%x, res=%d\n",
1978 inode->i_sb->s_id, inode->i_ino, mask, res);
1981 res = nfs_revalidate_inode(NFS_SERVER(inode), inode);
1983 res = generic_permission(inode, mask, NULL);
1990 * version-control: t
1991 * kept-new-versions: 5