4 * Copyright (C) 1992 Rick Sladkey
6 * nfs directory handling functions
8 * 10 Apr 1996 Added silly rename for unlink --okir
9 * 28 Sep 1996 Improved directory cache --okir
10 * 23 Aug 1997 Claus Heine claus@momo.math.rwth-aachen.de
11 * Re-implemented silly rename for unlink, newly implemented
12 * silly rename for nfs_rename() following the suggestions
13 * of Olaf Kirch (okir) found in this file.
14 * Following Linus comments on my original hack, this version
15 * depends only on the dcache stuff and doesn't touch the inode
16 * layer (iput() and friends).
17 * 6 Jun 1999 Cache readdir lookups in the page cache. -DaveM
20 #include <linux/time.h>
21 #include <linux/errno.h>
22 #include <linux/stat.h>
23 #include <linux/fcntl.h>
24 #include <linux/string.h>
25 #include <linux/kernel.h>
26 #include <linux/slab.h>
28 #include <linux/sunrpc/clnt.h>
29 #include <linux/nfs_fs.h>
30 #include <linux/nfs_mount.h>
31 #include <linux/pagemap.h>
32 #include <linux/smp_lock.h>
33 #include <linux/pagevec.h>
34 #include <linux/namei.h>
35 #include <linux/mount.h>
38 #include "delegation.h"
41 #define NFS_PARANOIA 1
42 /* #define NFS_DEBUG_VERBOSE 1 */
44 static int nfs_opendir(struct inode *, struct file *);
45 static int nfs_readdir(struct file *, void *, filldir_t);
46 static struct dentry *nfs_lookup(struct inode *, struct dentry *, struct nameidata *);
47 static int nfs_create(struct inode *, struct dentry *, int, struct nameidata *);
48 static int nfs_mkdir(struct inode *, struct dentry *, int);
49 static int nfs_rmdir(struct inode *, struct dentry *);
50 static int nfs_unlink(struct inode *, struct dentry *);
51 static int nfs_symlink(struct inode *, struct dentry *, const char *);
52 static int nfs_link(struct dentry *, struct inode *, struct dentry *);
53 static int nfs_mknod(struct inode *, struct dentry *, int, dev_t);
54 static int nfs_rename(struct inode *, struct dentry *,
55 struct inode *, struct dentry *);
56 static int nfs_fsync_dir(struct file *, struct dentry *, int);
57 static loff_t nfs_llseek_dir(struct file *, loff_t, int);
59 const struct file_operations nfs_dir_operations = {
60 .llseek = nfs_llseek_dir,
61 .read = generic_read_dir,
62 .readdir = nfs_readdir,
64 .release = nfs_release,
65 .fsync = nfs_fsync_dir,
68 struct inode_operations nfs_dir_inode_operations = {
73 .symlink = nfs_symlink,
78 .permission = nfs_permission,
79 .getattr = nfs_getattr,
80 .setattr = nfs_setattr,
84 struct inode_operations nfs3_dir_inode_operations = {
89 .symlink = nfs_symlink,
94 .permission = nfs_permission,
95 .getattr = nfs_getattr,
96 .setattr = nfs_setattr,
97 .listxattr = nfs3_listxattr,
98 .getxattr = nfs3_getxattr,
99 .setxattr = nfs3_setxattr,
100 .removexattr = nfs3_removexattr,
102 #endif /* CONFIG_NFS_V3 */
106 static struct dentry *nfs_atomic_lookup(struct inode *, struct dentry *, struct nameidata *);
107 struct inode_operations nfs4_dir_inode_operations = {
108 .create = nfs_create,
109 .lookup = nfs_atomic_lookup,
111 .unlink = nfs_unlink,
112 .symlink = nfs_symlink,
116 .rename = nfs_rename,
117 .permission = nfs_permission,
118 .getattr = nfs_getattr,
119 .setattr = nfs_setattr,
120 .getxattr = nfs4_getxattr,
121 .setxattr = nfs4_setxattr,
122 .listxattr = nfs4_listxattr,
125 #endif /* CONFIG_NFS_V4 */
131 nfs_opendir(struct inode *inode, struct file *filp)
135 dfprintk(VFS, "NFS: opendir(%s/%ld)\n",
136 inode->i_sb->s_id, inode->i_ino);
139 /* Call generic open code in order to cache credentials */
140 res = nfs_open(inode, filp);
145 typedef __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 } nfs_readdir_descriptor_t;
159 /* Now we cache directories properly, by stuffing the dirent
160 * data directly in the page cache.
162 * Inode invalidation due to refresh etc. takes care of
163 * _everything_, no sloppy entry flushing logic, no extraneous
164 * copying, network direct to page cache, the way it was meant
167 * NOTE: Dirent information verification is done always by the
168 * page-in of the RPC reply, nowhere else, this simplies
169 * things substantially.
172 int nfs_readdir_filler(nfs_readdir_descriptor_t *desc, struct page *page)
174 struct file *file = desc->file;
175 struct inode *inode = file->f_path.dentry->d_inode;
176 struct rpc_cred *cred = nfs_file_cred(file);
177 unsigned long timestamp;
180 dfprintk(DIRCACHE, "NFS: %s: reading cookie %Lu into page %lu\n",
181 __FUNCTION__, (long long)desc->entry->cookie,
186 error = NFS_PROTO(inode)->readdir(file->f_path.dentry, cred, desc->entry->cookie, page,
187 NFS_SERVER(inode)->dtsize, desc->plus);
189 /* We requested READDIRPLUS, but the server doesn't grok it */
190 if (error == -ENOTSUPP && desc->plus) {
191 NFS_SERVER(inode)->caps &= ~NFS_CAP_READDIRPLUS;
192 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_FLAGS(inode));
198 SetPageUptodate(page);
199 spin_lock(&inode->i_lock);
200 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ATIME;
201 spin_unlock(&inode->i_lock);
202 /* Ensure consistent page alignment of the data.
203 * Note: assumes we have exclusive access to this mapping either
204 * through inode->i_mutex or some other mechanism.
206 if (page->index == 0 && invalidate_inode_pages2_range(inode->i_mapping, PAGE_CACHE_SIZE, -1) < 0) {
207 /* Should never happen */
208 nfs_zap_mapping(inode, inode->i_mapping);
215 nfs_zap_caches(inode);
221 int dir_decode(nfs_readdir_descriptor_t *desc)
223 __be32 *p = desc->ptr;
224 p = desc->decode(p, desc->entry, desc->plus);
232 void dir_page_release(nfs_readdir_descriptor_t *desc)
235 page_cache_release(desc->page);
241 * Given a pointer to a buffer that has already been filled by a call
242 * to readdir, find the next entry with cookie '*desc->dir_cookie'.
244 * If the end of the buffer has been reached, return -EAGAIN, if not,
245 * return the offset within the buffer of the next entry to be
249 int find_dirent(nfs_readdir_descriptor_t *desc)
251 struct nfs_entry *entry = desc->entry;
255 while((status = dir_decode(desc)) == 0) {
256 dfprintk(DIRCACHE, "NFS: %s: examining cookie %Lu\n",
257 __FUNCTION__, (unsigned long long)entry->cookie);
258 if (entry->prev_cookie == *desc->dir_cookie)
260 if (loop_count++ > 200) {
269 * Given a pointer to a buffer that has already been filled by a call
270 * to readdir, find the entry at offset 'desc->file->f_pos'.
272 * If the end of the buffer has been reached, return -EAGAIN, if not,
273 * return the offset within the buffer of the next entry to be
277 int find_dirent_index(nfs_readdir_descriptor_t *desc)
279 struct nfs_entry *entry = desc->entry;
284 status = dir_decode(desc);
288 dfprintk(DIRCACHE, "NFS: found cookie %Lu at index %Ld\n",
289 (unsigned long long)entry->cookie, desc->current_index);
291 if (desc->file->f_pos == desc->current_index) {
292 *desc->dir_cookie = entry->cookie;
295 desc->current_index++;
296 if (loop_count++ > 200) {
305 * Find the given page, and call find_dirent() or find_dirent_index in
306 * order to try to return the next entry.
309 int find_dirent_page(nfs_readdir_descriptor_t *desc)
311 struct inode *inode = desc->file->f_path.dentry->d_inode;
315 dfprintk(DIRCACHE, "NFS: %s: searching page %ld for target %Lu\n",
316 __FUNCTION__, desc->page_index,
317 (long long) *desc->dir_cookie);
319 page = read_cache_page(inode->i_mapping, desc->page_index,
320 (filler_t *)nfs_readdir_filler, desc);
322 status = PTR_ERR(page);
325 if (!PageUptodate(page))
328 /* NOTE: Someone else may have changed the READDIRPLUS flag */
330 desc->ptr = kmap(page); /* matching kunmap in nfs_do_filldir */
331 if (*desc->dir_cookie != 0)
332 status = find_dirent(desc);
334 status = find_dirent_index(desc);
336 dir_page_release(desc);
338 dfprintk(DIRCACHE, "NFS: %s: returns %d\n", __FUNCTION__, status);
341 page_cache_release(page);
346 * Recurse through the page cache pages, and return a
347 * filled nfs_entry structure of the next directory entry if possible.
349 * The target for the search is '*desc->dir_cookie' if non-0,
350 * 'desc->file->f_pos' otherwise
353 int readdir_search_pagecache(nfs_readdir_descriptor_t *desc)
358 /* Always search-by-index from the beginning of the cache */
359 if (*desc->dir_cookie == 0) {
360 dfprintk(DIRCACHE, "NFS: readdir_search_pagecache() searching for offset %Ld\n",
361 (long long)desc->file->f_pos);
362 desc->page_index = 0;
363 desc->entry->cookie = desc->entry->prev_cookie = 0;
364 desc->entry->eof = 0;
365 desc->current_index = 0;
367 dfprintk(DIRCACHE, "NFS: readdir_search_pagecache() searching for cookie %Lu\n",
368 (unsigned long long)*desc->dir_cookie);
371 res = find_dirent_page(desc);
374 /* Align to beginning of next page */
376 if (loop_count++ > 200) {
382 dfprintk(DIRCACHE, "NFS: %s: returns %d\n", __FUNCTION__, res);
386 static inline unsigned int dt_type(struct inode *inode)
388 return (inode->i_mode >> 12) & 15;
391 static struct dentry *nfs_readdir_lookup(nfs_readdir_descriptor_t *desc);
394 * Once we've found the start of the dirent within a page: fill 'er up...
397 int nfs_do_filldir(nfs_readdir_descriptor_t *desc, void *dirent,
400 struct file *file = desc->file;
401 struct nfs_entry *entry = desc->entry;
402 struct dentry *dentry = NULL;
403 unsigned long fileid;
407 dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling starting @ cookie %Lu\n",
408 (unsigned long long)entry->cookie);
411 unsigned d_type = DT_UNKNOWN;
412 /* Note: entry->prev_cookie contains the cookie for
413 * retrieving the current dirent on the server */
414 fileid = nfs_fileid_to_ino_t(entry->ino);
416 /* Get a dentry if we have one */
419 dentry = nfs_readdir_lookup(desc);
421 /* Use readdirplus info */
422 if (dentry != NULL && dentry->d_inode != NULL) {
423 d_type = dt_type(dentry->d_inode);
424 fileid = dentry->d_inode->i_ino;
427 res = filldir(dirent, entry->name, entry->len,
428 file->f_pos, fileid, d_type);
432 *desc->dir_cookie = entry->cookie;
433 if (dir_decode(desc) != 0) {
437 if (loop_count++ > 200) {
442 dir_page_release(desc);
445 dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n",
446 (unsigned long long)*desc->dir_cookie, res);
451 * If we cannot find a cookie in our cache, we suspect that this is
452 * because it points to a deleted file, so we ask the server to return
453 * whatever it thinks is the next entry. We then feed this to filldir.
454 * If all goes well, we should then be able to find our way round the
455 * cache on the next call to readdir_search_pagecache();
457 * NOTE: we cannot add the anonymous page to the pagecache because
458 * the data it contains might not be page aligned. Besides,
459 * we should already have a complete representation of the
460 * directory in the page cache by the time we get here.
463 int uncached_readdir(nfs_readdir_descriptor_t *desc, void *dirent,
466 struct file *file = desc->file;
467 struct inode *inode = file->f_path.dentry->d_inode;
468 struct rpc_cred *cred = nfs_file_cred(file);
469 struct page *page = NULL;
472 dfprintk(DIRCACHE, "NFS: uncached_readdir() searching for cookie %Lu\n",
473 (unsigned long long)*desc->dir_cookie);
475 page = alloc_page(GFP_HIGHUSER);
480 desc->error = NFS_PROTO(inode)->readdir(file->f_path.dentry, cred, *desc->dir_cookie,
482 NFS_SERVER(inode)->dtsize,
484 spin_lock(&inode->i_lock);
485 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ATIME;
486 spin_unlock(&inode->i_lock);
488 desc->ptr = kmap(page); /* matching kunmap in nfs_do_filldir */
489 if (desc->error >= 0) {
490 if ((status = dir_decode(desc)) == 0)
491 desc->entry->prev_cookie = *desc->dir_cookie;
497 status = nfs_do_filldir(desc, dirent, filldir);
499 /* Reset read descriptor so it searches the page cache from
500 * the start upon the next call to readdir_search_pagecache() */
501 desc->page_index = 0;
502 desc->entry->cookie = desc->entry->prev_cookie = 0;
503 desc->entry->eof = 0;
505 dfprintk(DIRCACHE, "NFS: %s: returns %d\n",
506 __FUNCTION__, status);
509 dir_page_release(desc);
513 /* The file offset position represents the dirent entry number. A
514 last cookie cache takes care of the common case of reading the
517 static int nfs_readdir(struct file *filp, void *dirent, filldir_t filldir)
519 struct dentry *dentry = filp->f_path.dentry;
520 struct inode *inode = dentry->d_inode;
521 nfs_readdir_descriptor_t my_desc,
523 struct nfs_entry my_entry;
525 struct nfs_fattr fattr;
528 dfprintk(VFS, "NFS: readdir(%s/%s) starting at cookie %Lu\n",
529 dentry->d_parent->d_name.name, dentry->d_name.name,
530 (long long)filp->f_pos);
531 nfs_inc_stats(inode, NFSIOS_VFSGETDENTS);
535 res = nfs_revalidate_mapping_nolock(inode, filp->f_mapping);
542 * filp->f_pos points to the dirent entry number.
543 * *desc->dir_cookie has the cookie for the next entry. We have
544 * to either find the entry with the appropriate number or
545 * revalidate the cookie.
547 memset(desc, 0, sizeof(*desc));
550 desc->dir_cookie = &((struct nfs_open_context *)filp->private_data)->dir_cookie;
551 desc->decode = NFS_PROTO(inode)->decode_dirent;
552 desc->plus = NFS_USE_READDIRPLUS(inode);
554 my_entry.cookie = my_entry.prev_cookie = 0;
557 my_entry.fattr = &fattr;
558 nfs_fattr_init(&fattr);
559 desc->entry = &my_entry;
561 while(!desc->entry->eof) {
562 res = readdir_search_pagecache(desc);
564 if (res == -EBADCOOKIE) {
565 /* This means either end of directory */
566 if (*desc->dir_cookie && desc->entry->cookie != *desc->dir_cookie) {
567 /* Or that the server has 'lost' a cookie */
568 res = uncached_readdir(desc, dirent, filldir);
575 if (res == -ETOOSMALL && desc->plus) {
576 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_FLAGS(inode));
577 nfs_zap_caches(inode);
579 desc->entry->eof = 0;
585 res = nfs_do_filldir(desc, dirent, filldir);
594 dfprintk(VFS, "NFS: readdir(%s/%s) returns %ld\n",
595 dentry->d_parent->d_name.name, dentry->d_name.name,
600 loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int origin)
602 mutex_lock(&filp->f_path.dentry->d_inode->i_mutex);
605 offset += filp->f_pos;
613 if (offset != filp->f_pos) {
614 filp->f_pos = offset;
615 ((struct nfs_open_context *)filp->private_data)->dir_cookie = 0;
618 mutex_unlock(&filp->f_path.dentry->d_inode->i_mutex);
623 * All directory operations under NFS are synchronous, so fsync()
624 * is a dummy operation.
626 int nfs_fsync_dir(struct file *filp, struct dentry *dentry, int datasync)
628 dfprintk(VFS, "NFS: fsync_dir(%s/%s) datasync %d\n",
629 dentry->d_parent->d_name.name, dentry->d_name.name,
636 * A check for whether or not the parent directory has changed.
637 * In the case it has, we assume that the dentries are untrustworthy
638 * and may need to be looked up again.
640 static inline int nfs_check_verifier(struct inode *dir, struct dentry *dentry)
644 if ((NFS_I(dir)->cache_validity & NFS_INO_INVALID_ATTR) != 0
645 || nfs_attribute_timeout(dir))
647 return nfs_verify_change_attribute(dir, (unsigned long)dentry->d_fsdata);
650 static inline void nfs_set_verifier(struct dentry * dentry, unsigned long verf)
652 dentry->d_fsdata = (void *)verf;
656 * Whenever an NFS operation succeeds, we know that the dentry
657 * is valid, so we update the revalidation timestamp.
659 static inline void nfs_renew_times(struct dentry * dentry)
661 dentry->d_time = jiffies;
665 * Return the intent data that applies to this particular path component
667 * Note that the current set of intents only apply to the very last
668 * component of the path.
669 * We check for this using LOOKUP_CONTINUE and LOOKUP_PARENT.
671 static inline unsigned int nfs_lookup_check_intent(struct nameidata *nd, unsigned int mask)
673 if (nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))
675 return nd->flags & mask;
679 * Inode and filehandle revalidation for lookups.
681 * We force revalidation in the cases where the VFS sets LOOKUP_REVAL,
682 * or if the intent information indicates that we're about to open this
683 * particular file and the "nocto" mount flag is not set.
687 int nfs_lookup_verify_inode(struct inode *inode, struct nameidata *nd)
689 struct nfs_server *server = NFS_SERVER(inode);
692 /* VFS wants an on-the-wire revalidation */
693 if (nd->flags & LOOKUP_REVAL)
695 /* This is an open(2) */
696 if (nfs_lookup_check_intent(nd, LOOKUP_OPEN) != 0 &&
697 !(server->flags & NFS_MOUNT_NOCTO) &&
698 (S_ISREG(inode->i_mode) ||
699 S_ISDIR(inode->i_mode)))
702 return nfs_revalidate_inode(server, inode);
704 return __nfs_revalidate_inode(server, inode);
708 * We judge how long we want to trust negative
709 * dentries by looking at the parent inode mtime.
711 * If parent mtime has changed, we revalidate, else we wait for a
712 * period corresponding to the parent's attribute cache timeout value.
715 int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry,
716 struct nameidata *nd)
718 /* Don't revalidate a negative dentry if we're creating a new file */
719 if (nd != NULL && nfs_lookup_check_intent(nd, LOOKUP_CREATE) != 0)
721 return !nfs_check_verifier(dir, dentry);
725 * This is called every time the dcache has a lookup hit,
726 * and we should check whether we can really trust that
729 * NOTE! The hit can be a negative hit too, don't assume
732 * If the parent directory is seen to have changed, we throw out the
733 * cached dentry and do a new lookup.
735 static int nfs_lookup_revalidate(struct dentry * dentry, struct nameidata *nd)
739 struct dentry *parent;
741 struct nfs_fh fhandle;
742 struct nfs_fattr fattr;
743 unsigned long verifier;
745 parent = dget_parent(dentry);
747 dir = parent->d_inode;
748 nfs_inc_stats(dir, NFSIOS_DENTRYREVALIDATE);
749 inode = dentry->d_inode;
752 if (nfs_neg_need_reval(dir, dentry, nd))
757 if (is_bad_inode(inode)) {
758 dfprintk(LOOKUPCACHE, "%s: %s/%s has dud inode\n",
759 __FUNCTION__, dentry->d_parent->d_name.name,
760 dentry->d_name.name);
764 /* Revalidate parent directory attribute cache */
765 if (nfs_revalidate_inode(NFS_SERVER(dir), dir) < 0)
768 /* Force a full look up iff the parent directory has changed */
769 if (nfs_check_verifier(dir, dentry)) {
770 if (nfs_lookup_verify_inode(inode, nd))
775 if (NFS_STALE(inode))
778 verifier = nfs_save_change_attribute(dir);
779 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, &fhandle, &fattr);
782 if (nfs_compare_fh(NFS_FH(inode), &fhandle))
784 if ((error = nfs_refresh_inode(inode, &fattr)) != 0)
787 nfs_renew_times(dentry);
788 nfs_set_verifier(dentry, verifier);
792 dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is valid\n",
793 __FUNCTION__, dentry->d_parent->d_name.name,
794 dentry->d_name.name);
800 if (inode && S_ISDIR(inode->i_mode)) {
801 /* Purge readdir caches. */
802 nfs_zap_caches(inode);
803 /* If we have submounts, don't unhash ! */
804 if (have_submounts(dentry))
806 shrink_dcache_parent(dentry);
811 dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is invalid\n",
812 __FUNCTION__, dentry->d_parent->d_name.name,
813 dentry->d_name.name);
818 * This is called from dput() when d_count is going to 0.
820 static int nfs_dentry_delete(struct dentry *dentry)
822 dfprintk(VFS, "NFS: dentry_delete(%s/%s, %x)\n",
823 dentry->d_parent->d_name.name, dentry->d_name.name,
826 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
827 /* Unhash it, so that ->d_iput() would be called */
830 if (!(dentry->d_sb->s_flags & MS_ACTIVE)) {
831 /* Unhash it, so that ancestors of killed async unlink
832 * files will be cleaned up during umount */
840 * Called when the dentry loses inode.
841 * We use it to clean up silly-renamed files.
843 static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode)
845 nfs_inode_return_delegation(inode);
846 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
849 nfs_complete_unlink(dentry);
852 /* When creating a negative dentry, we want to renew d_time */
853 nfs_renew_times(dentry);
857 struct dentry_operations nfs_dentry_operations = {
858 .d_revalidate = nfs_lookup_revalidate,
859 .d_delete = nfs_dentry_delete,
860 .d_iput = nfs_dentry_iput,
864 * Use intent information to check whether or not we're going to do
865 * an O_EXCL create using this path component.
868 int nfs_is_exclusive_create(struct inode *dir, struct nameidata *nd)
870 if (NFS_PROTO(dir)->version == 2)
872 if (nd == NULL || nfs_lookup_check_intent(nd, LOOKUP_CREATE) == 0)
874 return (nd->intent.open.flags & O_EXCL) != 0;
877 static inline int nfs_reval_fsid(struct vfsmount *mnt, struct inode *dir,
878 struct nfs_fh *fh, struct nfs_fattr *fattr)
880 struct nfs_server *server = NFS_SERVER(dir);
882 if (!nfs_fsid_equal(&server->fsid, &fattr->fsid))
883 /* Revalidate fsid on root dir */
884 return __nfs_revalidate_inode(server, mnt->mnt_root->d_inode);
888 static struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
891 struct inode *inode = NULL;
893 struct nfs_fh fhandle;
894 struct nfs_fattr fattr;
896 dfprintk(VFS, "NFS: lookup(%s/%s)\n",
897 dentry->d_parent->d_name.name, dentry->d_name.name);
898 nfs_inc_stats(dir, NFSIOS_VFSLOOKUP);
900 res = ERR_PTR(-ENAMETOOLONG);
901 if (dentry->d_name.len > NFS_SERVER(dir)->namelen)
904 res = ERR_PTR(-ENOMEM);
905 dentry->d_op = NFS_PROTO(dir)->dentry_ops;
910 * If we're doing an exclusive create, optimize away the lookup
911 * but don't hash the dentry.
913 if (nfs_is_exclusive_create(dir, nd)) {
914 d_instantiate(dentry, NULL);
919 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, &fhandle, &fattr);
920 if (error == -ENOENT)
923 res = ERR_PTR(error);
926 error = nfs_reval_fsid(nd->mnt, dir, &fhandle, &fattr);
928 res = ERR_PTR(error);
931 inode = nfs_fhget(dentry->d_sb, &fhandle, &fattr);
932 res = (struct dentry *)inode;
937 res = d_materialise_unique(dentry, inode);
939 struct dentry *parent;
942 /* Was a directory renamed! */
943 parent = dget_parent(res);
944 if (!IS_ROOT(parent))
945 nfs_mark_for_revalidate(parent->d_inode);
949 nfs_renew_times(dentry);
950 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
958 static int nfs_open_revalidate(struct dentry *, struct nameidata *);
960 struct dentry_operations nfs4_dentry_operations = {
961 .d_revalidate = nfs_open_revalidate,
962 .d_delete = nfs_dentry_delete,
963 .d_iput = nfs_dentry_iput,
967 * Use intent information to determine whether we need to substitute
968 * the NFSv4-style stateful OPEN for the LOOKUP call
970 static int is_atomic_open(struct inode *dir, struct nameidata *nd)
972 if (nd == NULL || nfs_lookup_check_intent(nd, LOOKUP_OPEN) == 0)
974 /* NFS does not (yet) have a stateful open for directories */
975 if (nd->flags & LOOKUP_DIRECTORY)
977 /* Are we trying to write to a read only partition? */
978 if (IS_RDONLY(dir) && (nd->intent.open.flags & (O_CREAT|O_TRUNC|FMODE_WRITE)))
983 static struct dentry *nfs_atomic_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd)
985 struct dentry *res = NULL;
988 dfprintk(VFS, "NFS: atomic_lookup(%s/%ld), %s\n",
989 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
991 /* Check that we are indeed trying to open this file */
992 if (!is_atomic_open(dir, nd))
995 if (dentry->d_name.len > NFS_SERVER(dir)->namelen) {
996 res = ERR_PTR(-ENAMETOOLONG);
999 dentry->d_op = NFS_PROTO(dir)->dentry_ops;
1001 /* Let vfs_create() deal with O_EXCL */
1002 if (nd->intent.open.flags & O_EXCL) {
1003 d_add(dentry, NULL);
1007 /* Open the file on the server */
1009 /* Revalidate parent directory attribute cache */
1010 error = nfs_revalidate_inode(NFS_SERVER(dir), dir);
1012 res = ERR_PTR(error);
1017 if (nd->intent.open.flags & O_CREAT) {
1018 nfs_begin_data_update(dir);
1019 res = nfs4_atomic_open(dir, dentry, nd);
1020 nfs_end_data_update(dir);
1022 res = nfs4_atomic_open(dir, dentry, nd);
1025 error = PTR_ERR(res);
1027 /* Make a negative dentry */
1031 /* This turned out not to be a regular file */
1036 if (!(nd->intent.open.flags & O_NOFOLLOW))
1042 } else if (res != NULL)
1044 nfs_renew_times(dentry);
1045 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1049 return nfs_lookup(dir, dentry, nd);
1052 static int nfs_open_revalidate(struct dentry *dentry, struct nameidata *nd)
1054 struct dentry *parent = NULL;
1055 struct inode *inode = dentry->d_inode;
1057 unsigned long verifier;
1058 int openflags, ret = 0;
1060 parent = dget_parent(dentry);
1061 dir = parent->d_inode;
1062 if (!is_atomic_open(dir, nd))
1064 /* We can't create new files in nfs_open_revalidate(), so we
1065 * optimize away revalidation of negative dentries.
1069 /* NFS only supports OPEN on regular files */
1070 if (!S_ISREG(inode->i_mode))
1072 openflags = nd->intent.open.flags;
1073 /* We cannot do exclusive creation on a positive dentry */
1074 if ((openflags & (O_CREAT|O_EXCL)) == (O_CREAT|O_EXCL))
1076 /* We can't create new files, or truncate existing ones here */
1077 openflags &= ~(O_CREAT|O_TRUNC);
1080 * Note: we're not holding inode->i_mutex and so may be racing with
1081 * operations that change the directory. We therefore save the
1082 * change attribute *before* we do the RPC call.
1085 verifier = nfs_save_change_attribute(dir);
1086 ret = nfs4_open_revalidate(dir, dentry, openflags, nd);
1088 nfs_set_verifier(dentry, verifier);
1097 if (inode != NULL && nfs_have_delegation(inode, FMODE_READ))
1099 return nfs_lookup_revalidate(dentry, nd);
1101 #endif /* CONFIG_NFSV4 */
1103 static struct dentry *nfs_readdir_lookup(nfs_readdir_descriptor_t *desc)
1105 struct dentry *parent = desc->file->f_path.dentry;
1106 struct inode *dir = parent->d_inode;
1107 struct nfs_entry *entry = desc->entry;
1108 struct dentry *dentry, *alias;
1109 struct qstr name = {
1110 .name = entry->name,
1113 struct inode *inode;
1117 if (name.name[0] == '.' && name.name[1] == '.')
1118 return dget_parent(parent);
1121 if (name.name[0] == '.')
1122 return dget(parent);
1124 name.hash = full_name_hash(name.name, name.len);
1125 dentry = d_lookup(parent, &name);
1126 if (dentry != NULL) {
1127 /* Is this a positive dentry that matches the readdir info? */
1128 if (dentry->d_inode != NULL &&
1129 (NFS_FILEID(dentry->d_inode) == entry->ino ||
1130 d_mountpoint(dentry))) {
1131 if (!desc->plus || entry->fh->size == 0)
1133 if (nfs_compare_fh(NFS_FH(dentry->d_inode),
1137 /* No, so d_drop to allow one to be created */
1141 if (!desc->plus || !(entry->fattr->valid & NFS_ATTR_FATTR))
1143 /* Note: caller is already holding the dir->i_mutex! */
1144 dentry = d_alloc(parent, &name);
1147 dentry->d_op = NFS_PROTO(dir)->dentry_ops;
1148 inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr);
1149 if (IS_ERR(inode)) {
1154 alias = d_materialise_unique(dentry, inode);
1155 if (alias != NULL) {
1163 nfs_renew_times(dentry);
1164 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1169 * Code common to create, mkdir, and mknod.
1171 int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle,
1172 struct nfs_fattr *fattr)
1174 struct inode *inode;
1175 int error = -EACCES;
1177 /* We may have been initialized further down */
1178 if (dentry->d_inode)
1180 if (fhandle->size == 0) {
1181 struct inode *dir = dentry->d_parent->d_inode;
1182 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr);
1186 if (!(fattr->valid & NFS_ATTR_FATTR)) {
1187 struct nfs_server *server = NFS_SB(dentry->d_sb);
1188 error = server->nfs_client->rpc_ops->getattr(server, fhandle, fattr);
1192 inode = nfs_fhget(dentry->d_sb, fhandle, fattr);
1193 error = PTR_ERR(inode);
1196 d_instantiate(dentry, inode);
1197 if (d_unhashed(dentry))
1203 * Following a failed create operation, we drop the dentry rather
1204 * than retain a negative dentry. This avoids a problem in the event
1205 * that the operation succeeded on the server, but an error in the
1206 * reply path made it appear to have failed.
1208 static int nfs_create(struct inode *dir, struct dentry *dentry, int mode,
1209 struct nameidata *nd)
1215 dfprintk(VFS, "NFS: create(%s/%ld), %s\n",
1216 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1218 attr.ia_mode = mode;
1219 attr.ia_valid = ATTR_MODE;
1221 if (nd && (nd->flags & LOOKUP_CREATE))
1222 open_flags = nd->intent.open.flags;
1225 nfs_begin_data_update(dir);
1226 error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags, nd);
1227 nfs_end_data_update(dir);
1230 nfs_renew_times(dentry);
1231 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1241 * See comments for nfs_proc_create regarding failed operations.
1244 nfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t rdev)
1249 dfprintk(VFS, "NFS: mknod(%s/%ld), %s\n",
1250 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1252 if (!new_valid_dev(rdev))
1255 attr.ia_mode = mode;
1256 attr.ia_valid = ATTR_MODE;
1259 nfs_begin_data_update(dir);
1260 status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev);
1261 nfs_end_data_update(dir);
1264 nfs_renew_times(dentry);
1265 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1275 * See comments for nfs_proc_create regarding failed operations.
1277 static int nfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1282 dfprintk(VFS, "NFS: mkdir(%s/%ld), %s\n",
1283 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1285 attr.ia_valid = ATTR_MODE;
1286 attr.ia_mode = mode | S_IFDIR;
1289 nfs_begin_data_update(dir);
1290 error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr);
1291 nfs_end_data_update(dir);
1294 nfs_renew_times(dentry);
1295 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1304 static int nfs_rmdir(struct inode *dir, struct dentry *dentry)
1308 dfprintk(VFS, "NFS: rmdir(%s/%ld), %s\n",
1309 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1312 nfs_begin_data_update(dir);
1313 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
1314 /* Ensure the VFS deletes this inode */
1315 if (error == 0 && dentry->d_inode != NULL)
1316 clear_nlink(dentry->d_inode);
1317 nfs_end_data_update(dir);
1323 static int nfs_sillyrename(struct inode *dir, struct dentry *dentry)
1325 static unsigned int sillycounter;
1326 const int i_inosize = sizeof(dir->i_ino)*2;
1327 const int countersize = sizeof(sillycounter)*2;
1328 const int slen = sizeof(".nfs") + i_inosize + countersize - 1;
1331 struct dentry *sdentry;
1334 dfprintk(VFS, "NFS: silly-rename(%s/%s, ct=%d)\n",
1335 dentry->d_parent->d_name.name, dentry->d_name.name,
1336 atomic_read(&dentry->d_count));
1337 nfs_inc_stats(dir, NFSIOS_SILLYRENAME);
1340 if (!dentry->d_inode)
1341 printk("NFS: silly-renaming %s/%s, negative dentry??\n",
1342 dentry->d_parent->d_name.name, dentry->d_name.name);
1345 * We don't allow a dentry to be silly-renamed twice.
1348 if (dentry->d_flags & DCACHE_NFSFS_RENAMED)
1351 sprintf(silly, ".nfs%*.*lx",
1352 i_inosize, i_inosize, dentry->d_inode->i_ino);
1354 /* Return delegation in anticipation of the rename */
1355 nfs_inode_return_delegation(dentry->d_inode);
1359 char *suffix = silly + slen - countersize;
1363 sprintf(suffix, "%*.*x", countersize, countersize, sillycounter);
1365 dfprintk(VFS, "NFS: trying to rename %s to %s\n",
1366 dentry->d_name.name, silly);
1368 sdentry = lookup_one_len(silly, dentry->d_parent, slen);
1370 * N.B. Better to return EBUSY here ... it could be
1371 * dangerous to delete the file while it's in use.
1373 if (IS_ERR(sdentry))
1375 } while(sdentry->d_inode != NULL); /* need negative lookup */
1377 qsilly.name = silly;
1378 qsilly.len = strlen(silly);
1379 nfs_begin_data_update(dir);
1380 if (dentry->d_inode) {
1381 nfs_begin_data_update(dentry->d_inode);
1382 error = NFS_PROTO(dir)->rename(dir, &dentry->d_name,
1384 nfs_mark_for_revalidate(dentry->d_inode);
1385 nfs_end_data_update(dentry->d_inode);
1387 error = NFS_PROTO(dir)->rename(dir, &dentry->d_name,
1389 nfs_end_data_update(dir);
1391 nfs_renew_times(dentry);
1392 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1393 d_move(dentry, sdentry);
1394 error = nfs_async_unlink(dentry);
1395 /* If we return 0 we don't unlink */
1403 * Remove a file after making sure there are no pending writes,
1404 * and after checking that the file has only one user.
1406 * We invalidate the attribute cache and free the inode prior to the operation
1407 * to avoid possible races if the server reuses the inode.
1409 static int nfs_safe_remove(struct dentry *dentry)
1411 struct inode *dir = dentry->d_parent->d_inode;
1412 struct inode *inode = dentry->d_inode;
1415 dfprintk(VFS, "NFS: safe_remove(%s/%s)\n",
1416 dentry->d_parent->d_name.name, dentry->d_name.name);
1418 /* If the dentry was sillyrenamed, we simply call d_delete() */
1419 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1424 nfs_begin_data_update(dir);
1425 if (inode != NULL) {
1426 nfs_inode_return_delegation(inode);
1427 nfs_begin_data_update(inode);
1428 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1429 /* The VFS may want to delete this inode */
1432 nfs_mark_for_revalidate(inode);
1433 nfs_end_data_update(inode);
1435 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1436 nfs_end_data_update(dir);
1441 /* We do silly rename. In case sillyrename() returns -EBUSY, the inode
1442 * belongs to an active ".nfs..." file and we return -EBUSY.
1444 * If sillyrename() returns 0, we do nothing, otherwise we unlink.
1446 static int nfs_unlink(struct inode *dir, struct dentry *dentry)
1449 int need_rehash = 0;
1451 dfprintk(VFS, "NFS: unlink(%s/%ld, %s)\n", dir->i_sb->s_id,
1452 dir->i_ino, dentry->d_name.name);
1455 spin_lock(&dcache_lock);
1456 spin_lock(&dentry->d_lock);
1457 if (atomic_read(&dentry->d_count) > 1) {
1458 spin_unlock(&dentry->d_lock);
1459 spin_unlock(&dcache_lock);
1460 /* Start asynchronous writeout of the inode */
1461 write_inode_now(dentry->d_inode, 0);
1462 error = nfs_sillyrename(dir, dentry);
1466 if (!d_unhashed(dentry)) {
1470 spin_unlock(&dentry->d_lock);
1471 spin_unlock(&dcache_lock);
1472 error = nfs_safe_remove(dentry);
1474 nfs_renew_times(dentry);
1475 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1476 } else if (need_rehash)
1483 * To create a symbolic link, most file systems instantiate a new inode,
1484 * add a page to it containing the path, then write it out to the disk
1485 * using prepare_write/commit_write.
1487 * Unfortunately the NFS client can't create the in-core inode first
1488 * because it needs a file handle to create an in-core inode (see
1489 * fs/nfs/inode.c:nfs_fhget). We only have a file handle *after* the
1490 * symlink request has completed on the server.
1492 * So instead we allocate a raw page, copy the symname into it, then do
1493 * the SYMLINK request with the page as the buffer. If it succeeds, we
1494 * now have a new file handle and can instantiate an in-core NFS inode
1495 * and move the raw page into its mapping.
1497 static int nfs_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1499 struct pagevec lru_pvec;
1503 unsigned int pathlen = strlen(symname);
1506 dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s)\n", dir->i_sb->s_id,
1507 dir->i_ino, dentry->d_name.name, symname);
1509 if (pathlen > PAGE_SIZE)
1510 return -ENAMETOOLONG;
1512 attr.ia_mode = S_IFLNK | S_IRWXUGO;
1513 attr.ia_valid = ATTR_MODE;
1517 page = alloc_page(GFP_KERNEL);
1523 kaddr = kmap_atomic(page, KM_USER0);
1524 memcpy(kaddr, symname, pathlen);
1525 if (pathlen < PAGE_SIZE)
1526 memset(kaddr + pathlen, 0, PAGE_SIZE - pathlen);
1527 kunmap_atomic(kaddr, KM_USER0);
1529 nfs_begin_data_update(dir);
1530 error = NFS_PROTO(dir)->symlink(dir, dentry, page, pathlen, &attr);
1531 nfs_end_data_update(dir);
1533 dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s) error %d\n",
1534 dir->i_sb->s_id, dir->i_ino,
1535 dentry->d_name.name, symname, error);
1543 * No big deal if we can't add this page to the page cache here.
1544 * READLINK will get the missing page from the server if needed.
1546 pagevec_init(&lru_pvec, 0);
1547 if (!add_to_page_cache(page, dentry->d_inode->i_mapping, 0,
1549 pagevec_add(&lru_pvec, page);
1550 pagevec_lru_add(&lru_pvec);
1551 SetPageUptodate(page);
1561 nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1563 struct inode *inode = old_dentry->d_inode;
1566 dfprintk(VFS, "NFS: link(%s/%s -> %s/%s)\n",
1567 old_dentry->d_parent->d_name.name, old_dentry->d_name.name,
1568 dentry->d_parent->d_name.name, dentry->d_name.name);
1571 nfs_begin_data_update(dir);
1572 nfs_begin_data_update(inode);
1573 error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name);
1575 atomic_inc(&inode->i_count);
1576 d_instantiate(dentry, inode);
1578 nfs_end_data_update(inode);
1579 nfs_end_data_update(dir);
1586 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a
1587 * different file handle for the same inode after a rename (e.g. when
1588 * moving to a different directory). A fail-safe method to do so would
1589 * be to look up old_dir/old_name, create a link to new_dir/new_name and
1590 * rename the old file using the sillyrename stuff. This way, the original
1591 * file in old_dir will go away when the last process iput()s the inode.
1595 * It actually works quite well. One needs to have the possibility for
1596 * at least one ".nfs..." file in each directory the file ever gets
1597 * moved or linked to which happens automagically with the new
1598 * implementation that only depends on the dcache stuff instead of
1599 * using the inode layer
1601 * Unfortunately, things are a little more complicated than indicated
1602 * above. For a cross-directory move, we want to make sure we can get
1603 * rid of the old inode after the operation. This means there must be
1604 * no pending writes (if it's a file), and the use count must be 1.
1605 * If these conditions are met, we can drop the dentries before doing
1608 static int nfs_rename(struct inode *old_dir, struct dentry *old_dentry,
1609 struct inode *new_dir, struct dentry *new_dentry)
1611 struct inode *old_inode = old_dentry->d_inode;
1612 struct inode *new_inode = new_dentry->d_inode;
1613 struct dentry *dentry = NULL, *rehash = NULL;
1617 * To prevent any new references to the target during the rename,
1618 * we unhash the dentry and free the inode in advance.
1621 if (!d_unhashed(new_dentry)) {
1623 rehash = new_dentry;
1626 dfprintk(VFS, "NFS: rename(%s/%s -> %s/%s, ct=%d)\n",
1627 old_dentry->d_parent->d_name.name, old_dentry->d_name.name,
1628 new_dentry->d_parent->d_name.name, new_dentry->d_name.name,
1629 atomic_read(&new_dentry->d_count));
1632 * First check whether the target is busy ... we can't
1633 * safely do _any_ rename if the target is in use.
1635 * For files, make a copy of the dentry and then do a
1636 * silly-rename. If the silly-rename succeeds, the
1637 * copied dentry is hashed and becomes the new target.
1641 if (S_ISDIR(new_inode->i_mode)) {
1643 if (!S_ISDIR(old_inode->i_mode))
1645 } else if (atomic_read(&new_dentry->d_count) > 2) {
1647 /* copy the target dentry's name */
1648 dentry = d_alloc(new_dentry->d_parent,
1649 &new_dentry->d_name);
1653 /* silly-rename the existing target ... */
1654 err = nfs_sillyrename(new_dir, new_dentry);
1656 new_dentry = rehash = dentry;
1658 /* instantiate the replacement target */
1659 d_instantiate(new_dentry, NULL);
1660 } else if (atomic_read(&new_dentry->d_count) > 1) {
1661 /* dentry still busy? */
1663 printk("nfs_rename: target %s/%s busy, d_count=%d\n",
1664 new_dentry->d_parent->d_name.name,
1665 new_dentry->d_name.name,
1666 atomic_read(&new_dentry->d_count));
1671 drop_nlink(new_inode);
1675 * ... prune child dentries and writebacks if needed.
1677 if (atomic_read(&old_dentry->d_count) > 1) {
1678 nfs_wb_all(old_inode);
1679 shrink_dcache_parent(old_dentry);
1681 nfs_inode_return_delegation(old_inode);
1683 if (new_inode != NULL) {
1684 nfs_inode_return_delegation(new_inode);
1685 d_delete(new_dentry);
1688 nfs_begin_data_update(old_dir);
1689 nfs_begin_data_update(new_dir);
1690 nfs_begin_data_update(old_inode);
1691 error = NFS_PROTO(old_dir)->rename(old_dir, &old_dentry->d_name,
1692 new_dir, &new_dentry->d_name);
1693 nfs_mark_for_revalidate(old_inode);
1694 nfs_end_data_update(old_inode);
1695 nfs_end_data_update(new_dir);
1696 nfs_end_data_update(old_dir);
1701 d_move(old_dentry, new_dentry);
1702 nfs_renew_times(new_dentry);
1703 nfs_set_verifier(new_dentry, nfs_save_change_attribute(new_dir));
1706 /* new dentry created? */
1713 static DEFINE_SPINLOCK(nfs_access_lru_lock);
1714 static LIST_HEAD(nfs_access_lru_list);
1715 static atomic_long_t nfs_access_nr_entries;
1717 static void nfs_access_free_entry(struct nfs_access_entry *entry)
1719 put_rpccred(entry->cred);
1721 smp_mb__before_atomic_dec();
1722 atomic_long_dec(&nfs_access_nr_entries);
1723 smp_mb__after_atomic_dec();
1726 int nfs_access_cache_shrinker(int nr_to_scan, gfp_t gfp_mask)
1729 struct nfs_inode *nfsi;
1730 struct nfs_access_entry *cache;
1732 spin_lock(&nfs_access_lru_lock);
1734 list_for_each_entry(nfsi, &nfs_access_lru_list, access_cache_inode_lru) {
1735 struct inode *inode;
1737 if (nr_to_scan-- == 0)
1739 inode = igrab(&nfsi->vfs_inode);
1742 spin_lock(&inode->i_lock);
1743 if (list_empty(&nfsi->access_cache_entry_lru))
1744 goto remove_lru_entry;
1745 cache = list_entry(nfsi->access_cache_entry_lru.next,
1746 struct nfs_access_entry, lru);
1747 list_move(&cache->lru, &head);
1748 rb_erase(&cache->rb_node, &nfsi->access_cache);
1749 if (!list_empty(&nfsi->access_cache_entry_lru))
1750 list_move_tail(&nfsi->access_cache_inode_lru,
1751 &nfs_access_lru_list);
1754 list_del_init(&nfsi->access_cache_inode_lru);
1755 clear_bit(NFS_INO_ACL_LRU_SET, &nfsi->flags);
1757 spin_unlock(&inode->i_lock);
1761 spin_unlock(&nfs_access_lru_lock);
1762 while (!list_empty(&head)) {
1763 cache = list_entry(head.next, struct nfs_access_entry, lru);
1764 list_del(&cache->lru);
1765 nfs_access_free_entry(cache);
1767 return (atomic_long_read(&nfs_access_nr_entries) / 100) * sysctl_vfs_cache_pressure;
1770 static void __nfs_access_zap_cache(struct inode *inode)
1772 struct nfs_inode *nfsi = NFS_I(inode);
1773 struct rb_root *root_node = &nfsi->access_cache;
1774 struct rb_node *n, *dispose = NULL;
1775 struct nfs_access_entry *entry;
1777 /* Unhook entries from the cache */
1778 while ((n = rb_first(root_node)) != NULL) {
1779 entry = rb_entry(n, struct nfs_access_entry, rb_node);
1780 rb_erase(n, root_node);
1781 list_del(&entry->lru);
1782 n->rb_left = dispose;
1785 nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS;
1786 spin_unlock(&inode->i_lock);
1788 /* Now kill them all! */
1789 while (dispose != NULL) {
1791 dispose = n->rb_left;
1792 nfs_access_free_entry(rb_entry(n, struct nfs_access_entry, rb_node));
1796 void nfs_access_zap_cache(struct inode *inode)
1798 /* Remove from global LRU init */
1799 if (test_and_clear_bit(NFS_INO_ACL_LRU_SET, &NFS_FLAGS(inode))) {
1800 spin_lock(&nfs_access_lru_lock);
1801 list_del_init(&NFS_I(inode)->access_cache_inode_lru);
1802 spin_unlock(&nfs_access_lru_lock);
1805 spin_lock(&inode->i_lock);
1806 /* This will release the spinlock */
1807 __nfs_access_zap_cache(inode);
1810 static struct nfs_access_entry *nfs_access_search_rbtree(struct inode *inode, struct rpc_cred *cred)
1812 struct rb_node *n = NFS_I(inode)->access_cache.rb_node;
1813 struct nfs_access_entry *entry;
1816 entry = rb_entry(n, struct nfs_access_entry, rb_node);
1818 if (cred < entry->cred)
1820 else if (cred > entry->cred)
1828 int nfs_access_get_cached(struct inode *inode, struct rpc_cred *cred, struct nfs_access_entry *res)
1830 struct nfs_inode *nfsi = NFS_I(inode);
1831 struct nfs_access_entry *cache;
1834 spin_lock(&inode->i_lock);
1835 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS)
1837 cache = nfs_access_search_rbtree(inode, cred);
1840 if (time_after(jiffies, cache->jiffies + NFS_ATTRTIMEO(inode)))
1842 res->jiffies = cache->jiffies;
1843 res->cred = cache->cred;
1844 res->mask = cache->mask;
1845 list_move_tail(&cache->lru, &nfsi->access_cache_entry_lru);
1848 spin_unlock(&inode->i_lock);
1851 rb_erase(&cache->rb_node, &nfsi->access_cache);
1852 list_del(&cache->lru);
1853 spin_unlock(&inode->i_lock);
1854 nfs_access_free_entry(cache);
1857 /* This will release the spinlock */
1858 __nfs_access_zap_cache(inode);
1862 static void nfs_access_add_rbtree(struct inode *inode, struct nfs_access_entry *set)
1864 struct nfs_inode *nfsi = NFS_I(inode);
1865 struct rb_root *root_node = &nfsi->access_cache;
1866 struct rb_node **p = &root_node->rb_node;
1867 struct rb_node *parent = NULL;
1868 struct nfs_access_entry *entry;
1870 spin_lock(&inode->i_lock);
1871 while (*p != NULL) {
1873 entry = rb_entry(parent, struct nfs_access_entry, rb_node);
1875 if (set->cred < entry->cred)
1876 p = &parent->rb_left;
1877 else if (set->cred > entry->cred)
1878 p = &parent->rb_right;
1882 rb_link_node(&set->rb_node, parent, p);
1883 rb_insert_color(&set->rb_node, root_node);
1884 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
1885 spin_unlock(&inode->i_lock);
1888 rb_replace_node(parent, &set->rb_node, root_node);
1889 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
1890 list_del(&entry->lru);
1891 spin_unlock(&inode->i_lock);
1892 nfs_access_free_entry(entry);
1895 void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set)
1897 struct nfs_access_entry *cache = kmalloc(sizeof(*cache), GFP_KERNEL);
1900 RB_CLEAR_NODE(&cache->rb_node);
1901 cache->jiffies = set->jiffies;
1902 cache->cred = get_rpccred(set->cred);
1903 cache->mask = set->mask;
1905 nfs_access_add_rbtree(inode, cache);
1907 /* Update accounting */
1908 smp_mb__before_atomic_inc();
1909 atomic_long_inc(&nfs_access_nr_entries);
1910 smp_mb__after_atomic_inc();
1912 /* Add inode to global LRU list */
1913 if (!test_and_set_bit(NFS_INO_ACL_LRU_SET, &NFS_FLAGS(inode))) {
1914 spin_lock(&nfs_access_lru_lock);
1915 list_add_tail(&NFS_I(inode)->access_cache_inode_lru, &nfs_access_lru_list);
1916 spin_unlock(&nfs_access_lru_lock);
1920 static int nfs_do_access(struct inode *inode, struct rpc_cred *cred, int mask)
1922 struct nfs_access_entry cache;
1925 status = nfs_access_get_cached(inode, cred, &cache);
1929 /* Be clever: ask server to check for all possible rights */
1930 cache.mask = MAY_EXEC | MAY_WRITE | MAY_READ;
1932 cache.jiffies = jiffies;
1933 status = NFS_PROTO(inode)->access(inode, &cache);
1936 nfs_access_add_cache(inode, &cache);
1938 if ((cache.mask & mask) == mask)
1943 int nfs_permission(struct inode *inode, int mask, struct nameidata *nd)
1945 struct rpc_cred *cred;
1948 nfs_inc_stats(inode, NFSIOS_VFSACCESS);
1952 /* Is this sys_access() ? */
1953 if (nd != NULL && (nd->flags & LOOKUP_ACCESS))
1956 switch (inode->i_mode & S_IFMT) {
1960 /* NFSv4 has atomic_open... */
1961 if (nfs_server_capable(inode, NFS_CAP_ATOMIC_OPEN)
1963 && (nd->flags & LOOKUP_OPEN))
1968 * Optimize away all write operations, since the server
1969 * will check permissions when we perform the op.
1971 if ((mask & MAY_WRITE) && !(mask & MAY_READ))
1978 if (!NFS_PROTO(inode)->access)
1981 cred = rpcauth_lookupcred(NFS_CLIENT(inode)->cl_auth, 0);
1982 if (!IS_ERR(cred)) {
1983 res = nfs_do_access(inode, cred, mask);
1986 res = PTR_ERR(cred);
1989 dfprintk(VFS, "NFS: permission(%s/%ld), mask=0x%x, res=%d\n",
1990 inode->i_sb->s_id, inode->i_ino, mask, res);
1993 res = nfs_revalidate_inode(NFS_SERVER(inode), inode);
1995 res = generic_permission(inode, mask, NULL);
2002 * version-control: t
2003 * kept-new-versions: 5