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/namei.h>
36 #include "delegation.h"
39 #define NFS_PARANOIA 1
40 /* #define NFS_DEBUG_VERBOSE 1 */
42 static int nfs_opendir(struct inode *, struct file *);
43 static int nfs_readdir(struct file *, void *, filldir_t);
44 static struct dentry *nfs_lookup(struct inode *, struct dentry *, struct nameidata *);
45 static int nfs_create(struct inode *, struct dentry *, int, struct nameidata *);
46 static int nfs_mkdir(struct inode *, struct dentry *, int);
47 static int nfs_rmdir(struct inode *, struct dentry *);
48 static int nfs_unlink(struct inode *, struct dentry *);
49 static int nfs_symlink(struct inode *, struct dentry *, const char *);
50 static int nfs_link(struct dentry *, struct inode *, struct dentry *);
51 static int nfs_mknod(struct inode *, struct dentry *, int, dev_t);
52 static int nfs_rename(struct inode *, struct dentry *,
53 struct inode *, struct dentry *);
54 static int nfs_fsync_dir(struct file *, struct dentry *, int);
55 static loff_t nfs_llseek_dir(struct file *, loff_t, int);
57 const struct file_operations nfs_dir_operations = {
58 .llseek = nfs_llseek_dir,
59 .read = generic_read_dir,
60 .readdir = nfs_readdir,
62 .release = nfs_release,
63 .fsync = nfs_fsync_dir,
66 struct inode_operations nfs_dir_inode_operations = {
71 .symlink = nfs_symlink,
76 .permission = nfs_permission,
77 .getattr = nfs_getattr,
78 .setattr = nfs_setattr,
82 struct inode_operations nfs3_dir_inode_operations = {
87 .symlink = nfs_symlink,
92 .permission = nfs_permission,
93 .getattr = nfs_getattr,
94 .setattr = nfs_setattr,
95 .listxattr = nfs3_listxattr,
96 .getxattr = nfs3_getxattr,
97 .setxattr = nfs3_setxattr,
98 .removexattr = nfs3_removexattr,
100 #endif /* CONFIG_NFS_V3 */
104 static struct dentry *nfs_atomic_lookup(struct inode *, struct dentry *, struct nameidata *);
105 struct inode_operations nfs4_dir_inode_operations = {
106 .create = nfs_create,
107 .lookup = nfs_atomic_lookup,
109 .unlink = nfs_unlink,
110 .symlink = nfs_symlink,
114 .rename = nfs_rename,
115 .permission = nfs_permission,
116 .getattr = nfs_getattr,
117 .setattr = nfs_setattr,
118 .getxattr = nfs4_getxattr,
119 .setxattr = nfs4_setxattr,
120 .listxattr = nfs4_listxattr,
123 #endif /* CONFIG_NFS_V4 */
129 nfs_opendir(struct inode *inode, struct file *filp)
133 dfprintk(VFS, "NFS: opendir(%s/%ld)\n",
134 inode->i_sb->s_id, inode->i_ino);
137 /* Call generic open code in order to cache credentials */
138 res = nfs_open(inode, filp);
143 typedef u32 * (*decode_dirent_t)(u32 *, struct nfs_entry *, int);
147 unsigned long page_index;
150 loff_t current_index;
151 struct nfs_entry *entry;
152 decode_dirent_t decode;
155 } nfs_readdir_descriptor_t;
157 /* Now we cache directories properly, by stuffing the dirent
158 * data directly in the page cache.
160 * Inode invalidation due to refresh etc. takes care of
161 * _everything_, no sloppy entry flushing logic, no extraneous
162 * copying, network direct to page cache, the way it was meant
165 * NOTE: Dirent information verification is done always by the
166 * page-in of the RPC reply, nowhere else, this simplies
167 * things substantially.
170 int nfs_readdir_filler(nfs_readdir_descriptor_t *desc, struct page *page)
172 struct file *file = desc->file;
173 struct inode *inode = file->f_dentry->d_inode;
174 struct rpc_cred *cred = nfs_file_cred(file);
175 unsigned long timestamp;
178 dfprintk(DIRCACHE, "NFS: %s: reading cookie %Lu into page %lu\n",
179 __FUNCTION__, (long long)desc->entry->cookie,
184 error = NFS_PROTO(inode)->readdir(file->f_dentry, cred, desc->entry->cookie, page,
185 NFS_SERVER(inode)->dtsize, desc->plus);
187 /* We requested READDIRPLUS, but the server doesn't grok it */
188 if (error == -ENOTSUPP && desc->plus) {
189 NFS_SERVER(inode)->caps &= ~NFS_CAP_READDIRPLUS;
190 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_FLAGS(inode));
196 SetPageUptodate(page);
197 spin_lock(&inode->i_lock);
198 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ATIME;
199 spin_unlock(&inode->i_lock);
200 /* Ensure consistent page alignment of the data.
201 * Note: assumes we have exclusive access to this mapping either
202 * through inode->i_mutex or some other mechanism.
204 if (page->index == 0)
205 invalidate_inode_pages2_range(inode->i_mapping, PAGE_CACHE_SIZE, -1);
211 nfs_zap_caches(inode);
217 int dir_decode(nfs_readdir_descriptor_t *desc)
220 p = desc->decode(p, desc->entry, desc->plus);
228 void dir_page_release(nfs_readdir_descriptor_t *desc)
231 page_cache_release(desc->page);
237 * Given a pointer to a buffer that has already been filled by a call
238 * to readdir, find the next entry with cookie '*desc->dir_cookie'.
240 * If the end of the buffer has been reached, return -EAGAIN, if not,
241 * return the offset within the buffer of the next entry to be
245 int find_dirent(nfs_readdir_descriptor_t *desc)
247 struct nfs_entry *entry = desc->entry;
251 while((status = dir_decode(desc)) == 0) {
252 dfprintk(DIRCACHE, "NFS: %s: examining cookie %Lu\n",
253 __FUNCTION__, (unsigned long long)entry->cookie);
254 if (entry->prev_cookie == *desc->dir_cookie)
256 if (loop_count++ > 200) {
265 * Given a pointer to a buffer that has already been filled by a call
266 * to readdir, find the entry at offset 'desc->file->f_pos'.
268 * If the end of the buffer has been reached, return -EAGAIN, if not,
269 * return the offset within the buffer of the next entry to be
273 int find_dirent_index(nfs_readdir_descriptor_t *desc)
275 struct nfs_entry *entry = desc->entry;
280 status = dir_decode(desc);
284 dfprintk(DIRCACHE, "NFS: found cookie %Lu at index %Ld\n",
285 (unsigned long long)entry->cookie, desc->current_index);
287 if (desc->file->f_pos == desc->current_index) {
288 *desc->dir_cookie = entry->cookie;
291 desc->current_index++;
292 if (loop_count++ > 200) {
301 * Find the given page, and call find_dirent() or find_dirent_index in
302 * order to try to return the next entry.
305 int find_dirent_page(nfs_readdir_descriptor_t *desc)
307 struct inode *inode = desc->file->f_dentry->d_inode;
311 dfprintk(DIRCACHE, "NFS: %s: searching page %ld for target %Lu\n",
312 __FUNCTION__, desc->page_index,
313 (long long) *desc->dir_cookie);
315 page = read_cache_page(inode->i_mapping, desc->page_index,
316 (filler_t *)nfs_readdir_filler, desc);
318 status = PTR_ERR(page);
321 if (!PageUptodate(page))
324 /* NOTE: Someone else may have changed the READDIRPLUS flag */
326 desc->ptr = kmap(page); /* matching kunmap in nfs_do_filldir */
327 if (*desc->dir_cookie != 0)
328 status = find_dirent(desc);
330 status = find_dirent_index(desc);
332 dir_page_release(desc);
334 dfprintk(DIRCACHE, "NFS: %s: returns %d\n", __FUNCTION__, status);
337 page_cache_release(page);
342 * Recurse through the page cache pages, and return a
343 * filled nfs_entry structure of the next directory entry if possible.
345 * The target for the search is '*desc->dir_cookie' if non-0,
346 * 'desc->file->f_pos' otherwise
349 int readdir_search_pagecache(nfs_readdir_descriptor_t *desc)
354 /* Always search-by-index from the beginning of the cache */
355 if (*desc->dir_cookie == 0) {
356 dfprintk(DIRCACHE, "NFS: readdir_search_pagecache() searching for offset %Ld\n",
357 (long long)desc->file->f_pos);
358 desc->page_index = 0;
359 desc->entry->cookie = desc->entry->prev_cookie = 0;
360 desc->entry->eof = 0;
361 desc->current_index = 0;
363 dfprintk(DIRCACHE, "NFS: readdir_search_pagecache() searching for cookie %Lu\n",
364 (unsigned long long)*desc->dir_cookie);
367 res = find_dirent_page(desc);
370 /* Align to beginning of next page */
372 if (loop_count++ > 200) {
378 dfprintk(DIRCACHE, "NFS: %s: returns %d\n", __FUNCTION__, res);
382 static inline unsigned int dt_type(struct inode *inode)
384 return (inode->i_mode >> 12) & 15;
387 static struct dentry *nfs_readdir_lookup(nfs_readdir_descriptor_t *desc);
390 * Once we've found the start of the dirent within a page: fill 'er up...
393 int nfs_do_filldir(nfs_readdir_descriptor_t *desc, void *dirent,
396 struct file *file = desc->file;
397 struct nfs_entry *entry = desc->entry;
398 struct dentry *dentry = NULL;
399 unsigned long fileid;
403 dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling starting @ cookie %Lu\n",
404 (unsigned long long)entry->cookie);
407 unsigned d_type = DT_UNKNOWN;
408 /* Note: entry->prev_cookie contains the cookie for
409 * retrieving the current dirent on the server */
410 fileid = nfs_fileid_to_ino_t(entry->ino);
412 /* Get a dentry if we have one */
415 dentry = nfs_readdir_lookup(desc);
417 /* Use readdirplus info */
418 if (dentry != NULL && dentry->d_inode != NULL) {
419 d_type = dt_type(dentry->d_inode);
420 fileid = dentry->d_inode->i_ino;
423 res = filldir(dirent, entry->name, entry->len,
424 file->f_pos, fileid, d_type);
428 *desc->dir_cookie = entry->cookie;
429 if (dir_decode(desc) != 0) {
433 if (loop_count++ > 200) {
438 dir_page_release(desc);
441 dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n",
442 (unsigned long long)*desc->dir_cookie, res);
447 * If we cannot find a cookie in our cache, we suspect that this is
448 * because it points to a deleted file, so we ask the server to return
449 * whatever it thinks is the next entry. We then feed this to filldir.
450 * If all goes well, we should then be able to find our way round the
451 * cache on the next call to readdir_search_pagecache();
453 * NOTE: we cannot add the anonymous page to the pagecache because
454 * the data it contains might not be page aligned. Besides,
455 * we should already have a complete representation of the
456 * directory in the page cache by the time we get here.
459 int uncached_readdir(nfs_readdir_descriptor_t *desc, void *dirent,
462 struct file *file = desc->file;
463 struct inode *inode = file->f_dentry->d_inode;
464 struct rpc_cred *cred = nfs_file_cred(file);
465 struct page *page = NULL;
468 dfprintk(DIRCACHE, "NFS: uncached_readdir() searching for cookie %Lu\n",
469 (unsigned long long)*desc->dir_cookie);
471 page = alloc_page(GFP_HIGHUSER);
476 desc->error = NFS_PROTO(inode)->readdir(file->f_dentry, cred, *desc->dir_cookie,
478 NFS_SERVER(inode)->dtsize,
480 spin_lock(&inode->i_lock);
481 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ATIME;
482 spin_unlock(&inode->i_lock);
484 desc->ptr = kmap(page); /* matching kunmap in nfs_do_filldir */
485 if (desc->error >= 0) {
486 if ((status = dir_decode(desc)) == 0)
487 desc->entry->prev_cookie = *desc->dir_cookie;
493 status = nfs_do_filldir(desc, dirent, filldir);
495 /* Reset read descriptor so it searches the page cache from
496 * the start upon the next call to readdir_search_pagecache() */
497 desc->page_index = 0;
498 desc->entry->cookie = desc->entry->prev_cookie = 0;
499 desc->entry->eof = 0;
501 dfprintk(DIRCACHE, "NFS: %s: returns %d\n",
502 __FUNCTION__, status);
505 dir_page_release(desc);
509 /* The file offset position represents the dirent entry number. A
510 last cookie cache takes care of the common case of reading the
513 static int nfs_readdir(struct file *filp, void *dirent, filldir_t filldir)
515 struct dentry *dentry = filp->f_dentry;
516 struct inode *inode = dentry->d_inode;
517 nfs_readdir_descriptor_t my_desc,
519 struct nfs_entry my_entry;
521 struct nfs_fattr fattr;
524 dfprintk(VFS, "NFS: readdir(%s/%s) starting at cookie %Lu\n",
525 dentry->d_parent->d_name.name, dentry->d_name.name,
526 (long long)filp->f_pos);
527 nfs_inc_stats(inode, NFSIOS_VFSGETDENTS);
531 res = nfs_revalidate_mapping(inode, filp->f_mapping);
538 * filp->f_pos points to the dirent entry number.
539 * *desc->dir_cookie has the cookie for the next entry. We have
540 * to either find the entry with the appropriate number or
541 * revalidate the cookie.
543 memset(desc, 0, sizeof(*desc));
546 desc->dir_cookie = &((struct nfs_open_context *)filp->private_data)->dir_cookie;
547 desc->decode = NFS_PROTO(inode)->decode_dirent;
548 desc->plus = NFS_USE_READDIRPLUS(inode);
550 my_entry.cookie = my_entry.prev_cookie = 0;
553 my_entry.fattr = &fattr;
554 nfs_fattr_init(&fattr);
555 desc->entry = &my_entry;
557 while(!desc->entry->eof) {
558 res = readdir_search_pagecache(desc);
560 if (res == -EBADCOOKIE) {
561 /* This means either end of directory */
562 if (*desc->dir_cookie && desc->entry->cookie != *desc->dir_cookie) {
563 /* Or that the server has 'lost' a cookie */
564 res = uncached_readdir(desc, dirent, filldir);
571 if (res == -ETOOSMALL && desc->plus) {
572 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_FLAGS(inode));
573 nfs_zap_caches(inode);
575 desc->entry->eof = 0;
581 res = nfs_do_filldir(desc, dirent, filldir);
590 dfprintk(VFS, "NFS: readdir(%s/%s) returns %ld\n",
591 dentry->d_parent->d_name.name, dentry->d_name.name,
596 loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int origin)
598 mutex_lock(&filp->f_dentry->d_inode->i_mutex);
601 offset += filp->f_pos;
609 if (offset != filp->f_pos) {
610 filp->f_pos = offset;
611 ((struct nfs_open_context *)filp->private_data)->dir_cookie = 0;
614 mutex_unlock(&filp->f_dentry->d_inode->i_mutex);
619 * All directory operations under NFS are synchronous, so fsync()
620 * is a dummy operation.
622 int nfs_fsync_dir(struct file *filp, struct dentry *dentry, int datasync)
624 dfprintk(VFS, "NFS: fsync_dir(%s/%s) datasync %d\n",
625 dentry->d_parent->d_name.name, dentry->d_name.name,
632 * A check for whether or not the parent directory has changed.
633 * In the case it has, we assume that the dentries are untrustworthy
634 * and may need to be looked up again.
636 static inline int nfs_check_verifier(struct inode *dir, struct dentry *dentry)
640 if ((NFS_I(dir)->cache_validity & NFS_INO_INVALID_ATTR) != 0
641 || nfs_attribute_timeout(dir))
643 return nfs_verify_change_attribute(dir, (unsigned long)dentry->d_fsdata);
646 static inline void nfs_set_verifier(struct dentry * dentry, unsigned long verf)
648 dentry->d_fsdata = (void *)verf;
652 * Whenever an NFS operation succeeds, we know that the dentry
653 * is valid, so we update the revalidation timestamp.
655 static inline void nfs_renew_times(struct dentry * dentry)
657 dentry->d_time = jiffies;
661 * Return the intent data that applies to this particular path component
663 * Note that the current set of intents only apply to the very last
664 * component of the path.
665 * We check for this using LOOKUP_CONTINUE and LOOKUP_PARENT.
667 static inline unsigned int nfs_lookup_check_intent(struct nameidata *nd, unsigned int mask)
669 if (nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))
671 return nd->flags & mask;
675 * Inode and filehandle revalidation for lookups.
677 * We force revalidation in the cases where the VFS sets LOOKUP_REVAL,
678 * or if the intent information indicates that we're about to open this
679 * particular file and the "nocto" mount flag is not set.
683 int nfs_lookup_verify_inode(struct inode *inode, struct nameidata *nd)
685 struct nfs_server *server = NFS_SERVER(inode);
688 /* VFS wants an on-the-wire revalidation */
689 if (nd->flags & LOOKUP_REVAL)
691 /* This is an open(2) */
692 if (nfs_lookup_check_intent(nd, LOOKUP_OPEN) != 0 &&
693 !(server->flags & NFS_MOUNT_NOCTO) &&
694 (S_ISREG(inode->i_mode) ||
695 S_ISDIR(inode->i_mode)))
698 return nfs_revalidate_inode(server, inode);
700 return __nfs_revalidate_inode(server, inode);
704 * We judge how long we want to trust negative
705 * dentries by looking at the parent inode mtime.
707 * If parent mtime has changed, we revalidate, else we wait for a
708 * period corresponding to the parent's attribute cache timeout value.
711 int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry,
712 struct nameidata *nd)
714 /* Don't revalidate a negative dentry if we're creating a new file */
715 if (nd != NULL && nfs_lookup_check_intent(nd, LOOKUP_CREATE) != 0)
717 return !nfs_check_verifier(dir, dentry);
721 * This is called every time the dcache has a lookup hit,
722 * and we should check whether we can really trust that
725 * NOTE! The hit can be a negative hit too, don't assume
728 * If the parent directory is seen to have changed, we throw out the
729 * cached dentry and do a new lookup.
731 static int nfs_lookup_revalidate(struct dentry * dentry, struct nameidata *nd)
735 struct dentry *parent;
737 struct nfs_fh fhandle;
738 struct nfs_fattr fattr;
739 unsigned long verifier;
741 parent = dget_parent(dentry);
743 dir = parent->d_inode;
744 nfs_inc_stats(dir, NFSIOS_DENTRYREVALIDATE);
745 inode = dentry->d_inode;
748 if (nfs_neg_need_reval(dir, dentry, nd))
753 if (is_bad_inode(inode)) {
754 dfprintk(LOOKUPCACHE, "%s: %s/%s has dud inode\n",
755 __FUNCTION__, dentry->d_parent->d_name.name,
756 dentry->d_name.name);
760 /* Revalidate parent directory attribute cache */
761 if (nfs_revalidate_inode(NFS_SERVER(dir), dir) < 0)
764 /* Force a full look up iff the parent directory has changed */
765 if (nfs_check_verifier(dir, dentry)) {
766 if (nfs_lookup_verify_inode(inode, nd))
771 if (NFS_STALE(inode))
774 verifier = nfs_save_change_attribute(dir);
775 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, &fhandle, &fattr);
778 if (nfs_compare_fh(NFS_FH(inode), &fhandle))
780 if ((error = nfs_refresh_inode(inode, &fattr)) != 0)
783 nfs_renew_times(dentry);
784 nfs_set_verifier(dentry, verifier);
788 dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is valid\n",
789 __FUNCTION__, dentry->d_parent->d_name.name,
790 dentry->d_name.name);
796 if (inode && S_ISDIR(inode->i_mode)) {
797 /* Purge readdir caches. */
798 nfs_zap_caches(inode);
799 /* If we have submounts, don't unhash ! */
800 if (have_submounts(dentry))
802 shrink_dcache_parent(dentry);
807 dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is invalid\n",
808 __FUNCTION__, dentry->d_parent->d_name.name,
809 dentry->d_name.name);
814 * This is called from dput() when d_count is going to 0.
816 static int nfs_dentry_delete(struct dentry *dentry)
818 dfprintk(VFS, "NFS: dentry_delete(%s/%s, %x)\n",
819 dentry->d_parent->d_name.name, dentry->d_name.name,
822 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
823 /* Unhash it, so that ->d_iput() would be called */
826 if (!(dentry->d_sb->s_flags & MS_ACTIVE)) {
827 /* Unhash it, so that ancestors of killed async unlink
828 * files will be cleaned up during umount */
836 * Called when the dentry loses inode.
837 * We use it to clean up silly-renamed files.
839 static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode)
841 nfs_inode_return_delegation(inode);
842 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
845 nfs_complete_unlink(dentry);
848 /* When creating a negative dentry, we want to renew d_time */
849 nfs_renew_times(dentry);
853 struct dentry_operations nfs_dentry_operations = {
854 .d_revalidate = nfs_lookup_revalidate,
855 .d_delete = nfs_dentry_delete,
856 .d_iput = nfs_dentry_iput,
860 * Use intent information to check whether or not we're going to do
861 * an O_EXCL create using this path component.
864 int nfs_is_exclusive_create(struct inode *dir, struct nameidata *nd)
866 if (NFS_PROTO(dir)->version == 2)
868 if (nd == NULL || nfs_lookup_check_intent(nd, LOOKUP_CREATE) == 0)
870 return (nd->intent.open.flags & O_EXCL) != 0;
873 static inline int nfs_reval_fsid(struct inode *dir,
874 struct nfs_fh *fh, struct nfs_fattr *fattr)
876 struct nfs_server *server = NFS_SERVER(dir);
878 if (!nfs_fsid_equal(&server->fsid, &fattr->fsid))
879 /* Revalidate fsid on root dir */
880 return __nfs_revalidate_inode(server, dir->i_sb->s_root->d_inode);
884 static struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
887 struct inode *inode = NULL;
889 struct nfs_fh fhandle;
890 struct nfs_fattr fattr;
892 dfprintk(VFS, "NFS: lookup(%s/%s)\n",
893 dentry->d_parent->d_name.name, dentry->d_name.name);
894 nfs_inc_stats(dir, NFSIOS_VFSLOOKUP);
896 res = ERR_PTR(-ENAMETOOLONG);
897 if (dentry->d_name.len > NFS_SERVER(dir)->namelen)
900 res = ERR_PTR(-ENOMEM);
901 dentry->d_op = NFS_PROTO(dir)->dentry_ops;
905 /* If we're doing an exclusive create, optimize away the lookup */
906 if (nfs_is_exclusive_create(dir, nd))
909 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, &fhandle, &fattr);
910 if (error == -ENOENT)
913 res = ERR_PTR(error);
916 error = nfs_reval_fsid(dir, &fhandle, &fattr);
918 res = ERR_PTR(error);
921 inode = nfs_fhget(dentry->d_sb, &fhandle, &fattr);
922 res = (struct dentry *)inode;
926 res = d_add_unique(dentry, inode);
929 nfs_renew_times(dentry);
930 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
938 static int nfs_open_revalidate(struct dentry *, struct nameidata *);
940 struct dentry_operations nfs4_dentry_operations = {
941 .d_revalidate = nfs_open_revalidate,
942 .d_delete = nfs_dentry_delete,
943 .d_iput = nfs_dentry_iput,
947 * Use intent information to determine whether we need to substitute
948 * the NFSv4-style stateful OPEN for the LOOKUP call
950 static int is_atomic_open(struct inode *dir, struct nameidata *nd)
952 if (nd == NULL || nfs_lookup_check_intent(nd, LOOKUP_OPEN) == 0)
954 /* NFS does not (yet) have a stateful open for directories */
955 if (nd->flags & LOOKUP_DIRECTORY)
957 /* Are we trying to write to a read only partition? */
958 if (IS_RDONLY(dir) && (nd->intent.open.flags & (O_CREAT|O_TRUNC|FMODE_WRITE)))
963 static struct dentry *nfs_atomic_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd)
965 struct dentry *res = NULL;
968 dfprintk(VFS, "NFS: atomic_lookup(%s/%ld), %s\n",
969 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
971 /* Check that we are indeed trying to open this file */
972 if (!is_atomic_open(dir, nd))
975 if (dentry->d_name.len > NFS_SERVER(dir)->namelen) {
976 res = ERR_PTR(-ENAMETOOLONG);
979 dentry->d_op = NFS_PROTO(dir)->dentry_ops;
981 /* Let vfs_create() deal with O_EXCL */
982 if (nd->intent.open.flags & O_EXCL) {
987 /* Open the file on the server */
989 /* Revalidate parent directory attribute cache */
990 error = nfs_revalidate_inode(NFS_SERVER(dir), dir);
992 res = ERR_PTR(error);
997 if (nd->intent.open.flags & O_CREAT) {
998 nfs_begin_data_update(dir);
999 res = nfs4_atomic_open(dir, dentry, nd);
1000 nfs_end_data_update(dir);
1002 res = nfs4_atomic_open(dir, dentry, nd);
1005 error = PTR_ERR(res);
1007 /* Make a negative dentry */
1011 /* This turned out not to be a regular file */
1016 if (!(nd->intent.open.flags & O_NOFOLLOW))
1022 } else if (res != NULL)
1024 nfs_renew_times(dentry);
1025 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1029 return nfs_lookup(dir, dentry, nd);
1032 static int nfs_open_revalidate(struct dentry *dentry, struct nameidata *nd)
1034 struct dentry *parent = NULL;
1035 struct inode *inode = dentry->d_inode;
1037 unsigned long verifier;
1038 int openflags, ret = 0;
1040 parent = dget_parent(dentry);
1041 dir = parent->d_inode;
1042 if (!is_atomic_open(dir, nd))
1044 /* We can't create new files in nfs_open_revalidate(), so we
1045 * optimize away revalidation of negative dentries.
1049 /* NFS only supports OPEN on regular files */
1050 if (!S_ISREG(inode->i_mode))
1052 openflags = nd->intent.open.flags;
1053 /* We cannot do exclusive creation on a positive dentry */
1054 if ((openflags & (O_CREAT|O_EXCL)) == (O_CREAT|O_EXCL))
1056 /* We can't create new files, or truncate existing ones here */
1057 openflags &= ~(O_CREAT|O_TRUNC);
1060 * Note: we're not holding inode->i_mutex and so may be racing with
1061 * operations that change the directory. We therefore save the
1062 * change attribute *before* we do the RPC call.
1065 verifier = nfs_save_change_attribute(dir);
1066 ret = nfs4_open_revalidate(dir, dentry, openflags, nd);
1068 nfs_set_verifier(dentry, verifier);
1077 if (inode != NULL && nfs_have_delegation(inode, FMODE_READ))
1079 return nfs_lookup_revalidate(dentry, nd);
1081 #endif /* CONFIG_NFSV4 */
1083 static struct dentry *nfs_readdir_lookup(nfs_readdir_descriptor_t *desc)
1085 struct dentry *parent = desc->file->f_dentry;
1086 struct inode *dir = parent->d_inode;
1087 struct nfs_entry *entry = desc->entry;
1088 struct dentry *dentry, *alias;
1089 struct qstr name = {
1090 .name = entry->name,
1093 struct inode *inode;
1097 if (name.name[0] == '.' && name.name[1] == '.')
1098 return dget_parent(parent);
1101 if (name.name[0] == '.')
1102 return dget(parent);
1104 name.hash = full_name_hash(name.name, name.len);
1105 dentry = d_lookup(parent, &name);
1108 if (!desc->plus || !(entry->fattr->valid & NFS_ATTR_FATTR))
1110 /* Note: caller is already holding the dir->i_mutex! */
1111 dentry = d_alloc(parent, &name);
1114 dentry->d_op = NFS_PROTO(dir)->dentry_ops;
1115 inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr);
1116 if (IS_ERR(inode)) {
1120 alias = d_add_unique(dentry, inode);
1121 if (alias != NULL) {
1125 nfs_renew_times(dentry);
1126 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1131 * Code common to create, mkdir, and mknod.
1133 int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle,
1134 struct nfs_fattr *fattr)
1136 struct inode *inode;
1137 int error = -EACCES;
1139 /* We may have been initialized further down */
1140 if (dentry->d_inode)
1142 if (fhandle->size == 0) {
1143 struct inode *dir = dentry->d_parent->d_inode;
1144 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr);
1148 if (!(fattr->valid & NFS_ATTR_FATTR)) {
1149 struct nfs_server *server = NFS_SB(dentry->d_sb);
1150 error = server->rpc_ops->getattr(server, fhandle, fattr);
1154 inode = nfs_fhget(dentry->d_sb, fhandle, fattr);
1155 error = PTR_ERR(inode);
1158 d_instantiate(dentry, inode);
1166 * Following a failed create operation, we drop the dentry rather
1167 * than retain a negative dentry. This avoids a problem in the event
1168 * that the operation succeeded on the server, but an error in the
1169 * reply path made it appear to have failed.
1171 static int nfs_create(struct inode *dir, struct dentry *dentry, int mode,
1172 struct nameidata *nd)
1178 dfprintk(VFS, "NFS: create(%s/%ld), %s\n",
1179 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1181 attr.ia_mode = mode;
1182 attr.ia_valid = ATTR_MODE;
1184 if (nd && (nd->flags & LOOKUP_CREATE))
1185 open_flags = nd->intent.open.flags;
1188 nfs_begin_data_update(dir);
1189 error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags, nd);
1190 nfs_end_data_update(dir);
1193 nfs_renew_times(dentry);
1194 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1204 * See comments for nfs_proc_create regarding failed operations.
1207 nfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t rdev)
1212 dfprintk(VFS, "NFS: mknod(%s/%ld), %s\n",
1213 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1215 if (!new_valid_dev(rdev))
1218 attr.ia_mode = mode;
1219 attr.ia_valid = ATTR_MODE;
1222 nfs_begin_data_update(dir);
1223 status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev);
1224 nfs_end_data_update(dir);
1227 nfs_renew_times(dentry);
1228 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1238 * See comments for nfs_proc_create regarding failed operations.
1240 static int nfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1245 dfprintk(VFS, "NFS: mkdir(%s/%ld), %s\n",
1246 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1248 attr.ia_valid = ATTR_MODE;
1249 attr.ia_mode = mode | S_IFDIR;
1252 nfs_begin_data_update(dir);
1253 error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr);
1254 nfs_end_data_update(dir);
1257 nfs_renew_times(dentry);
1258 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1267 static int nfs_rmdir(struct inode *dir, struct dentry *dentry)
1271 dfprintk(VFS, "NFS: rmdir(%s/%ld), %s\n",
1272 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1275 nfs_begin_data_update(dir);
1276 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
1277 /* Ensure the VFS deletes this inode */
1278 if (error == 0 && dentry->d_inode != NULL)
1279 dentry->d_inode->i_nlink = 0;
1280 nfs_end_data_update(dir);
1286 static int nfs_sillyrename(struct inode *dir, struct dentry *dentry)
1288 static unsigned int sillycounter;
1289 const int i_inosize = sizeof(dir->i_ino)*2;
1290 const int countersize = sizeof(sillycounter)*2;
1291 const int slen = sizeof(".nfs") + i_inosize + countersize - 1;
1294 struct dentry *sdentry;
1297 dfprintk(VFS, "NFS: silly-rename(%s/%s, ct=%d)\n",
1298 dentry->d_parent->d_name.name, dentry->d_name.name,
1299 atomic_read(&dentry->d_count));
1300 nfs_inc_stats(dir, NFSIOS_SILLYRENAME);
1303 if (!dentry->d_inode)
1304 printk("NFS: silly-renaming %s/%s, negative dentry??\n",
1305 dentry->d_parent->d_name.name, dentry->d_name.name);
1308 * We don't allow a dentry to be silly-renamed twice.
1311 if (dentry->d_flags & DCACHE_NFSFS_RENAMED)
1314 sprintf(silly, ".nfs%*.*lx",
1315 i_inosize, i_inosize, dentry->d_inode->i_ino);
1317 /* Return delegation in anticipation of the rename */
1318 nfs_inode_return_delegation(dentry->d_inode);
1322 char *suffix = silly + slen - countersize;
1326 sprintf(suffix, "%*.*x", countersize, countersize, sillycounter);
1328 dfprintk(VFS, "NFS: trying to rename %s to %s\n",
1329 dentry->d_name.name, silly);
1331 sdentry = lookup_one_len(silly, dentry->d_parent, slen);
1333 * N.B. Better to return EBUSY here ... it could be
1334 * dangerous to delete the file while it's in use.
1336 if (IS_ERR(sdentry))
1338 } while(sdentry->d_inode != NULL); /* need negative lookup */
1340 qsilly.name = silly;
1341 qsilly.len = strlen(silly);
1342 nfs_begin_data_update(dir);
1343 if (dentry->d_inode) {
1344 nfs_begin_data_update(dentry->d_inode);
1345 error = NFS_PROTO(dir)->rename(dir, &dentry->d_name,
1347 nfs_mark_for_revalidate(dentry->d_inode);
1348 nfs_end_data_update(dentry->d_inode);
1350 error = NFS_PROTO(dir)->rename(dir, &dentry->d_name,
1352 nfs_end_data_update(dir);
1354 nfs_renew_times(dentry);
1355 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1356 d_move(dentry, sdentry);
1357 error = nfs_async_unlink(dentry);
1358 /* If we return 0 we don't unlink */
1366 * Remove a file after making sure there are no pending writes,
1367 * and after checking that the file has only one user.
1369 * We invalidate the attribute cache and free the inode prior to the operation
1370 * to avoid possible races if the server reuses the inode.
1372 static int nfs_safe_remove(struct dentry *dentry)
1374 struct inode *dir = dentry->d_parent->d_inode;
1375 struct inode *inode = dentry->d_inode;
1378 dfprintk(VFS, "NFS: safe_remove(%s/%s)\n",
1379 dentry->d_parent->d_name.name, dentry->d_name.name);
1381 /* If the dentry was sillyrenamed, we simply call d_delete() */
1382 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1387 nfs_begin_data_update(dir);
1388 if (inode != NULL) {
1389 nfs_inode_return_delegation(inode);
1390 nfs_begin_data_update(inode);
1391 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1392 /* The VFS may want to delete this inode */
1395 nfs_mark_for_revalidate(inode);
1396 nfs_end_data_update(inode);
1398 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1399 nfs_end_data_update(dir);
1404 /* We do silly rename. In case sillyrename() returns -EBUSY, the inode
1405 * belongs to an active ".nfs..." file and we return -EBUSY.
1407 * If sillyrename() returns 0, we do nothing, otherwise we unlink.
1409 static int nfs_unlink(struct inode *dir, struct dentry *dentry)
1412 int need_rehash = 0;
1414 dfprintk(VFS, "NFS: unlink(%s/%ld, %s)\n", dir->i_sb->s_id,
1415 dir->i_ino, dentry->d_name.name);
1418 spin_lock(&dcache_lock);
1419 spin_lock(&dentry->d_lock);
1420 if (atomic_read(&dentry->d_count) > 1) {
1421 spin_unlock(&dentry->d_lock);
1422 spin_unlock(&dcache_lock);
1423 error = nfs_sillyrename(dir, dentry);
1427 if (!d_unhashed(dentry)) {
1431 spin_unlock(&dentry->d_lock);
1432 spin_unlock(&dcache_lock);
1433 error = nfs_safe_remove(dentry);
1435 nfs_renew_times(dentry);
1436 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1437 } else if (need_rehash)
1444 nfs_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1447 struct nfs_fattr sym_attr;
1448 struct nfs_fh sym_fh;
1449 struct qstr qsymname;
1452 dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s)\n", dir->i_sb->s_id,
1453 dir->i_ino, dentry->d_name.name, symname);
1456 if (dentry->d_inode)
1457 printk("nfs_proc_symlink: %s/%s not negative!\n",
1458 dentry->d_parent->d_name.name, dentry->d_name.name);
1461 * Fill in the sattr for the call.
1462 * Note: SunOS 4.1.2 crashes if the mode isn't initialized!
1464 attr.ia_valid = ATTR_MODE;
1465 attr.ia_mode = S_IFLNK | S_IRWXUGO;
1467 qsymname.name = symname;
1468 qsymname.len = strlen(symname);
1471 nfs_begin_data_update(dir);
1472 error = NFS_PROTO(dir)->symlink(dir, &dentry->d_name, &qsymname,
1473 &attr, &sym_fh, &sym_attr);
1474 nfs_end_data_update(dir);
1476 error = nfs_instantiate(dentry, &sym_fh, &sym_attr);
1478 if (error == -EEXIST)
1479 printk("nfs_proc_symlink: %s/%s already exists??\n",
1480 dentry->d_parent->d_name.name, dentry->d_name.name);
1488 nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1490 struct inode *inode = old_dentry->d_inode;
1493 dfprintk(VFS, "NFS: link(%s/%s -> %s/%s)\n",
1494 old_dentry->d_parent->d_name.name, old_dentry->d_name.name,
1495 dentry->d_parent->d_name.name, dentry->d_name.name);
1498 nfs_begin_data_update(dir);
1499 nfs_begin_data_update(inode);
1500 error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name);
1502 atomic_inc(&inode->i_count);
1503 d_instantiate(dentry, inode);
1505 nfs_end_data_update(inode);
1506 nfs_end_data_update(dir);
1513 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a
1514 * different file handle for the same inode after a rename (e.g. when
1515 * moving to a different directory). A fail-safe method to do so would
1516 * be to look up old_dir/old_name, create a link to new_dir/new_name and
1517 * rename the old file using the sillyrename stuff. This way, the original
1518 * file in old_dir will go away when the last process iput()s the inode.
1522 * It actually works quite well. One needs to have the possibility for
1523 * at least one ".nfs..." file in each directory the file ever gets
1524 * moved or linked to which happens automagically with the new
1525 * implementation that only depends on the dcache stuff instead of
1526 * using the inode layer
1528 * Unfortunately, things are a little more complicated than indicated
1529 * above. For a cross-directory move, we want to make sure we can get
1530 * rid of the old inode after the operation. This means there must be
1531 * no pending writes (if it's a file), and the use count must be 1.
1532 * If these conditions are met, we can drop the dentries before doing
1535 static int nfs_rename(struct inode *old_dir, struct dentry *old_dentry,
1536 struct inode *new_dir, struct dentry *new_dentry)
1538 struct inode *old_inode = old_dentry->d_inode;
1539 struct inode *new_inode = new_dentry->d_inode;
1540 struct dentry *dentry = NULL, *rehash = NULL;
1544 * To prevent any new references to the target during the rename,
1545 * we unhash the dentry and free the inode in advance.
1548 if (!d_unhashed(new_dentry)) {
1550 rehash = new_dentry;
1553 dfprintk(VFS, "NFS: rename(%s/%s -> %s/%s, ct=%d)\n",
1554 old_dentry->d_parent->d_name.name, old_dentry->d_name.name,
1555 new_dentry->d_parent->d_name.name, new_dentry->d_name.name,
1556 atomic_read(&new_dentry->d_count));
1559 * First check whether the target is busy ... we can't
1560 * safely do _any_ rename if the target is in use.
1562 * For files, make a copy of the dentry and then do a
1563 * silly-rename. If the silly-rename succeeds, the
1564 * copied dentry is hashed and becomes the new target.
1568 if (S_ISDIR(new_inode->i_mode)) {
1570 if (!S_ISDIR(old_inode->i_mode))
1572 } else if (atomic_read(&new_dentry->d_count) > 2) {
1574 /* copy the target dentry's name */
1575 dentry = d_alloc(new_dentry->d_parent,
1576 &new_dentry->d_name);
1580 /* silly-rename the existing target ... */
1581 err = nfs_sillyrename(new_dir, new_dentry);
1583 new_dentry = rehash = dentry;
1585 /* instantiate the replacement target */
1586 d_instantiate(new_dentry, NULL);
1587 } else if (atomic_read(&new_dentry->d_count) > 1) {
1588 /* dentry still busy? */
1590 printk("nfs_rename: target %s/%s busy, d_count=%d\n",
1591 new_dentry->d_parent->d_name.name,
1592 new_dentry->d_name.name,
1593 atomic_read(&new_dentry->d_count));
1598 new_inode->i_nlink--;
1602 * ... prune child dentries and writebacks if needed.
1604 if (atomic_read(&old_dentry->d_count) > 1) {
1605 nfs_wb_all(old_inode);
1606 shrink_dcache_parent(old_dentry);
1608 nfs_inode_return_delegation(old_inode);
1610 if (new_inode != NULL) {
1611 nfs_inode_return_delegation(new_inode);
1612 d_delete(new_dentry);
1615 nfs_begin_data_update(old_dir);
1616 nfs_begin_data_update(new_dir);
1617 nfs_begin_data_update(old_inode);
1618 error = NFS_PROTO(old_dir)->rename(old_dir, &old_dentry->d_name,
1619 new_dir, &new_dentry->d_name);
1620 nfs_mark_for_revalidate(old_inode);
1621 nfs_end_data_update(old_inode);
1622 nfs_end_data_update(new_dir);
1623 nfs_end_data_update(old_dir);
1628 if (!S_ISDIR(old_inode->i_mode))
1629 d_move(old_dentry, new_dentry);
1630 nfs_renew_times(new_dentry);
1631 nfs_set_verifier(new_dentry, nfs_save_change_attribute(new_dir));
1634 /* new dentry created? */
1641 int nfs_access_get_cached(struct inode *inode, struct rpc_cred *cred, struct nfs_access_entry *res)
1643 struct nfs_inode *nfsi = NFS_I(inode);
1644 struct nfs_access_entry *cache = &nfsi->cache_access;
1646 if (cache->cred != cred
1647 || time_after(jiffies, cache->jiffies + NFS_ATTRTIMEO(inode))
1648 || (nfsi->cache_validity & NFS_INO_INVALID_ACCESS))
1650 memcpy(res, cache, sizeof(*res));
1654 void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set)
1656 struct nfs_inode *nfsi = NFS_I(inode);
1657 struct nfs_access_entry *cache = &nfsi->cache_access;
1659 if (cache->cred != set->cred) {
1661 put_rpccred(cache->cred);
1662 cache->cred = get_rpccred(set->cred);
1664 /* FIXME: replace current access_cache BKL reliance with inode->i_lock */
1665 spin_lock(&inode->i_lock);
1666 nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS;
1667 spin_unlock(&inode->i_lock);
1668 cache->jiffies = set->jiffies;
1669 cache->mask = set->mask;
1672 static int nfs_do_access(struct inode *inode, struct rpc_cred *cred, int mask)
1674 struct nfs_access_entry cache;
1677 status = nfs_access_get_cached(inode, cred, &cache);
1681 /* Be clever: ask server to check for all possible rights */
1682 cache.mask = MAY_EXEC | MAY_WRITE | MAY_READ;
1684 cache.jiffies = jiffies;
1685 status = NFS_PROTO(inode)->access(inode, &cache);
1688 nfs_access_add_cache(inode, &cache);
1690 if ((cache.mask & mask) == mask)
1695 int nfs_permission(struct inode *inode, int mask, struct nameidata *nd)
1697 struct rpc_cred *cred;
1700 nfs_inc_stats(inode, NFSIOS_VFSACCESS);
1704 /* Is this sys_access() ? */
1705 if (nd != NULL && (nd->flags & LOOKUP_ACCESS))
1708 switch (inode->i_mode & S_IFMT) {
1712 /* NFSv4 has atomic_open... */
1713 if (nfs_server_capable(inode, NFS_CAP_ATOMIC_OPEN)
1715 && (nd->flags & LOOKUP_OPEN))
1720 * Optimize away all write operations, since the server
1721 * will check permissions when we perform the op.
1723 if ((mask & MAY_WRITE) && !(mask & MAY_READ))
1730 if (!NFS_PROTO(inode)->access)
1733 cred = rpcauth_lookupcred(NFS_CLIENT(inode)->cl_auth, 0);
1734 if (!IS_ERR(cred)) {
1735 res = nfs_do_access(inode, cred, mask);
1738 res = PTR_ERR(cred);
1741 dfprintk(VFS, "NFS: permission(%s/%ld), mask=0x%x, res=%d\n",
1742 inode->i_sb->s_id, inode->i_ino, mask, res);
1745 res = nfs_revalidate_inode(NFS_SERVER(inode), inode);
1747 res = generic_permission(inode, mask, NULL);
1754 * version-control: t
1755 * kept-new-versions: 5