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>
34 #include <linux/vserver/xid.h>
36 #include "delegation.h"
38 #define NFS_PARANOIA 1
39 /* #define NFS_DEBUG_VERBOSE 1 */
41 static int nfs_opendir(struct inode *, struct file *);
42 static int nfs_readdir(struct file *, void *, filldir_t);
43 static struct dentry *nfs_lookup(struct inode *, struct dentry *, struct nameidata *);
44 static int nfs_create(struct inode *, struct dentry *, int, struct nameidata *);
45 static int nfs_mkdir(struct inode *, struct dentry *, int);
46 static int nfs_rmdir(struct inode *, struct dentry *);
47 static int nfs_unlink(struct inode *, struct dentry *);
48 static int nfs_symlink(struct inode *, struct dentry *, const char *);
49 static int nfs_link(struct dentry *, struct inode *, struct dentry *);
50 static int nfs_mknod(struct inode *, struct dentry *, int, dev_t);
51 static int nfs_rename(struct inode *, struct dentry *,
52 struct inode *, struct dentry *);
53 static int nfs_fsync_dir(struct file *, struct dentry *, int);
55 struct file_operations nfs_dir_operations = {
56 .read = generic_read_dir,
57 .readdir = nfs_readdir,
59 .release = nfs_release,
60 .fsync = nfs_fsync_dir,
63 struct inode_operations nfs_dir_inode_operations = {
68 .symlink = nfs_symlink,
73 .permission = nfs_permission,
74 .getattr = nfs_getattr,
75 .setattr = nfs_setattr,
80 static struct dentry *nfs_atomic_lookup(struct inode *, struct dentry *, struct nameidata *);
81 struct inode_operations nfs4_dir_inode_operations = {
83 .lookup = nfs_atomic_lookup,
86 .symlink = nfs_symlink,
91 .permission = nfs_permission,
92 .getattr = nfs_getattr,
93 .setattr = nfs_setattr,
96 #endif /* CONFIG_NFS_V4 */
102 nfs_opendir(struct inode *inode, struct file *filp)
107 /* Call generic open code in order to cache credentials */
109 res = nfs_open(inode, filp);
114 typedef u32 * (*decode_dirent_t)(u32 *, struct nfs_entry *, int);
118 unsigned long page_index;
121 struct nfs_entry *entry;
122 decode_dirent_t decode;
125 } nfs_readdir_descriptor_t;
127 /* Now we cache directories properly, by stuffing the dirent
128 * data directly in the page cache.
130 * Inode invalidation due to refresh etc. takes care of
131 * _everything_, no sloppy entry flushing logic, no extraneous
132 * copying, network direct to page cache, the way it was meant
135 * NOTE: Dirent information verification is done always by the
136 * page-in of the RPC reply, nowhere else, this simplies
137 * things substantially.
140 int nfs_readdir_filler(nfs_readdir_descriptor_t *desc, struct page *page)
142 struct file *file = desc->file;
143 struct inode *inode = file->f_dentry->d_inode;
144 struct rpc_cred *cred = nfs_file_cred(file);
145 unsigned long timestamp;
148 dfprintk(VFS, "NFS: nfs_readdir_filler() reading cookie %Lu into page %lu.\n", (long long)desc->entry->cookie, page->index);
152 error = NFS_PROTO(inode)->readdir(file->f_dentry, cred, desc->entry->cookie, page,
153 NFS_SERVER(inode)->dtsize, desc->plus);
155 /* We requested READDIRPLUS, but the server doesn't grok it */
156 if (error == -ENOTSUPP && desc->plus) {
157 NFS_SERVER(inode)->caps &= ~NFS_CAP_READDIRPLUS;
158 NFS_FLAGS(inode) &= ~NFS_INO_ADVISE_RDPLUS;
164 SetPageUptodate(page);
165 NFS_FLAGS(inode) |= NFS_INO_INVALID_ATIME;
166 /* Ensure consistent page alignment of the data.
167 * Note: assumes we have exclusive access to this mapping either
168 * throught inode->i_sem or some other mechanism.
170 if (page->index == 0) {
171 invalidate_inode_pages(inode->i_mapping);
172 NFS_I(inode)->readdir_timestamp = timestamp;
179 nfs_zap_caches(inode);
185 int dir_decode(nfs_readdir_descriptor_t *desc)
188 p = desc->decode(p, desc->entry, desc->plus);
196 void dir_page_release(nfs_readdir_descriptor_t *desc)
199 page_cache_release(desc->page);
205 * Given a pointer to a buffer that has already been filled by a call
206 * to readdir, find the next entry.
208 * If the end of the buffer has been reached, return -EAGAIN, if not,
209 * return the offset within the buffer of the next entry to be
213 int find_dirent(nfs_readdir_descriptor_t *desc, struct page *page)
215 struct nfs_entry *entry = desc->entry;
219 while((status = dir_decode(desc)) == 0) {
220 dfprintk(VFS, "NFS: found cookie %Lu\n", (long long)entry->cookie);
221 if (entry->prev_cookie == desc->target)
223 if (loop_count++ > 200) {
228 dfprintk(VFS, "NFS: find_dirent() returns %d\n", status);
233 * Find the given page, and call find_dirent() in order to try to
234 * return the next entry.
237 int find_dirent_page(nfs_readdir_descriptor_t *desc)
239 struct inode *inode = desc->file->f_dentry->d_inode;
243 dfprintk(VFS, "NFS: find_dirent_page() searching directory page %ld\n", desc->page_index);
245 page = read_cache_page(inode->i_mapping, desc->page_index,
246 (filler_t *)nfs_readdir_filler, desc);
248 status = PTR_ERR(page);
251 if (!PageUptodate(page))
254 /* NOTE: Someone else may have changed the READDIRPLUS flag */
256 desc->ptr = kmap(page); /* matching kunmap in nfs_do_filldir */
257 status = find_dirent(desc, page);
259 dir_page_release(desc);
261 dfprintk(VFS, "NFS: find_dirent_page() returns %d\n", status);
264 page_cache_release(page);
269 * Recurse through the page cache pages, and return a
270 * filled nfs_entry structure of the next directory entry if possible.
272 * The target for the search is 'desc->target'.
275 int readdir_search_pagecache(nfs_readdir_descriptor_t *desc)
280 dfprintk(VFS, "NFS: readdir_search_pagecache() searching for cookie %Lu\n", (long long)desc->target);
282 res = find_dirent_page(desc);
285 /* Align to beginning of next page */
287 if (loop_count++ > 200) {
292 dfprintk(VFS, "NFS: readdir_search_pagecache() returned %d\n", res);
296 static inline unsigned int dt_type(struct inode *inode)
298 return (inode->i_mode >> 12) & 15;
301 static struct dentry *nfs_readdir_lookup(nfs_readdir_descriptor_t *desc);
304 * Once we've found the start of the dirent within a page: fill 'er up...
307 int nfs_do_filldir(nfs_readdir_descriptor_t *desc, void *dirent,
310 struct file *file = desc->file;
311 struct nfs_entry *entry = desc->entry;
312 struct dentry *dentry = NULL;
313 unsigned long fileid;
317 dfprintk(VFS, "NFS: nfs_do_filldir() filling starting @ cookie %Lu\n", (long long)desc->target);
320 unsigned d_type = DT_UNKNOWN;
321 /* Note: entry->prev_cookie contains the cookie for
322 * retrieving the current dirent on the server */
323 fileid = nfs_fileid_to_ino_t(entry->ino);
325 /* Get a dentry if we have one */
328 dentry = nfs_readdir_lookup(desc);
330 /* Use readdirplus info */
331 if (dentry != NULL && dentry->d_inode != NULL) {
332 d_type = dt_type(dentry->d_inode);
333 fileid = dentry->d_inode->i_ino;
336 res = filldir(dirent, entry->name, entry->len,
337 entry->prev_cookie, fileid, d_type);
340 file->f_pos = desc->target = entry->cookie;
341 if (dir_decode(desc) != 0) {
345 if (loop_count++ > 200) {
350 dir_page_release(desc);
353 dfprintk(VFS, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n", (long long)desc->target, res);
358 * If we cannot find a cookie in our cache, we suspect that this is
359 * because it points to a deleted file, so we ask the server to return
360 * whatever it thinks is the next entry. We then feed this to filldir.
361 * If all goes well, we should then be able to find our way round the
362 * cache on the next call to readdir_search_pagecache();
364 * NOTE: we cannot add the anonymous page to the pagecache because
365 * the data it contains might not be page aligned. Besides,
366 * we should already have a complete representation of the
367 * directory in the page cache by the time we get here.
370 int uncached_readdir(nfs_readdir_descriptor_t *desc, void *dirent,
373 struct file *file = desc->file;
374 struct inode *inode = file->f_dentry->d_inode;
375 struct rpc_cred *cred = nfs_file_cred(file);
376 struct page *page = NULL;
379 dfprintk(VFS, "NFS: uncached_readdir() searching for cookie %Lu\n", (long long)desc->target);
381 page = alloc_page(GFP_HIGHUSER);
386 desc->error = NFS_PROTO(inode)->readdir(file->f_dentry, cred, desc->target,
388 NFS_SERVER(inode)->dtsize,
390 NFS_FLAGS(inode) |= NFS_INO_INVALID_ATIME;
392 desc->ptr = kmap(page); /* matching kunmap in nfs_do_filldir */
393 if (desc->error >= 0) {
394 if ((status = dir_decode(desc)) == 0)
395 desc->entry->prev_cookie = desc->target;
401 status = nfs_do_filldir(desc, dirent, filldir);
403 /* Reset read descriptor so it searches the page cache from
404 * the start upon the next call to readdir_search_pagecache() */
405 desc->page_index = 0;
406 desc->entry->cookie = desc->entry->prev_cookie = 0;
407 desc->entry->eof = 0;
409 dfprintk(VFS, "NFS: uncached_readdir() returns %d\n", status);
412 dir_page_release(desc);
416 /* The file offset position is now represented as a true offset into the
417 * page cache as is the case in most of the other filesystems.
419 static int nfs_readdir(struct file *filp, void *dirent, filldir_t filldir)
421 struct dentry *dentry = filp->f_dentry;
422 struct inode *inode = dentry->d_inode;
423 nfs_readdir_descriptor_t my_desc,
425 struct nfs_entry my_entry;
427 struct nfs_fattr fattr;
432 res = nfs_revalidate_inode(NFS_SERVER(inode), inode);
439 * filp->f_pos points to the file offset in the page cache.
440 * but if the cache has meanwhile been zapped, we need to
441 * read from the last dirent to revalidate f_pos
444 memset(desc, 0, sizeof(*desc));
447 desc->target = filp->f_pos;
448 desc->decode = NFS_PROTO(inode)->decode_dirent;
449 desc->plus = NFS_USE_READDIRPLUS(inode);
451 my_entry.cookie = my_entry.prev_cookie = 0;
454 my_entry.fattr = &fattr;
455 desc->entry = &my_entry;
457 while(!desc->entry->eof) {
458 res = readdir_search_pagecache(desc);
459 if (res == -EBADCOOKIE) {
460 /* This means either end of directory */
461 if (desc->entry->cookie != desc->target) {
462 /* Or that the server has 'lost' a cookie */
463 res = uncached_readdir(desc, dirent, filldir);
470 if (res == -ETOOSMALL && desc->plus) {
471 NFS_FLAGS(inode) &= ~NFS_INO_ADVISE_RDPLUS;
472 nfs_zap_caches(inode);
474 desc->entry->eof = 0;
480 res = nfs_do_filldir(desc, dirent, filldir);
495 * All directory operations under NFS are synchronous, so fsync()
496 * is a dummy operation.
498 int nfs_fsync_dir(struct file *filp, struct dentry *dentry, int datasync)
504 * A check for whether or not the parent directory has changed.
505 * In the case it has, we assume that the dentries are untrustworthy
506 * and may need to be looked up again.
508 static inline int nfs_check_verifier(struct inode *dir, struct dentry *dentry)
512 if ((NFS_FLAGS(dir) & NFS_INO_INVALID_ATTR) != 0
513 || nfs_attribute_timeout(dir))
515 return nfs_verify_change_attribute(dir, (unsigned long)dentry->d_fsdata);
518 static inline void nfs_set_verifier(struct dentry * dentry, unsigned long verf)
520 dentry->d_fsdata = (void *)verf;
524 * Whenever an NFS operation succeeds, we know that the dentry
525 * is valid, so we update the revalidation timestamp.
527 static inline void nfs_renew_times(struct dentry * dentry)
529 dentry->d_time = jiffies;
533 int nfs_lookup_verify_inode(struct inode *inode, int isopen)
535 struct nfs_server *server = NFS_SERVER(inode);
537 if (isopen && !(server->flags & NFS_MOUNT_NOCTO))
538 return __nfs_revalidate_inode(server, inode);
539 return nfs_revalidate_inode(server, inode);
543 * We judge how long we want to trust negative
544 * dentries by looking at the parent inode mtime.
546 * If parent mtime has changed, we revalidate, else we wait for a
547 * period corresponding to the parent's attribute cache timeout value.
550 int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry,
551 struct nameidata *nd)
557 /* Don't revalidate a negative dentry if we're creating a new file */
558 if ((ndflags & LOOKUP_CREATE) && !(ndflags & LOOKUP_CONTINUE))
560 return !nfs_check_verifier(dir, dentry);
564 * This is called every time the dcache has a lookup hit,
565 * and we should check whether we can really trust that
568 * NOTE! The hit can be a negative hit too, don't assume
571 * If the parent directory is seen to have changed, we throw out the
572 * cached dentry and do a new lookup.
574 static int nfs_lookup_revalidate(struct dentry * dentry, struct nameidata *nd)
578 struct dentry *parent;
580 struct nfs_fh fhandle;
581 struct nfs_fattr fattr;
582 unsigned long verifier;
585 parent = dget_parent(dentry);
587 dir = parent->d_inode;
588 inode = dentry->d_inode;
590 if (nd && !(nd->flags & LOOKUP_CONTINUE) && (nd->flags & LOOKUP_OPEN))
594 if (nfs_neg_need_reval(dir, dentry, nd))
599 if (is_bad_inode(inode)) {
600 dfprintk(VFS, "nfs_lookup_validate: %s/%s has dud inode\n",
601 dentry->d_parent->d_name.name, dentry->d_name.name);
605 /* Revalidate parent directory attribute cache */
606 nfs_revalidate_inode(NFS_SERVER(dir), dir);
608 /* Force a full look up iff the parent directory has changed */
609 if (nfs_check_verifier(dir, dentry)) {
610 if (nfs_lookup_verify_inode(inode, isopen))
615 if (NFS_STALE(inode))
618 verifier = nfs_save_change_attribute(dir);
619 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, &fhandle, &fattr);
622 if (nfs_compare_fh(NFS_FH(inode), &fhandle))
624 if ((error = nfs_refresh_inode(inode, &fattr)) != 0)
627 nfs_renew_times(dentry);
628 nfs_set_verifier(dentry, verifier);
637 if (inode && S_ISDIR(inode->i_mode)) {
638 /* Purge readdir caches. */
639 nfs_zap_caches(inode);
640 /* If we have submounts, don't unhash ! */
641 if (have_submounts(dentry))
643 shrink_dcache_parent(dentry);
652 * This is called from dput() when d_count is going to 0.
654 static int nfs_dentry_delete(struct dentry *dentry)
656 dfprintk(VFS, "NFS: dentry_delete(%s/%s, %x)\n",
657 dentry->d_parent->d_name.name, dentry->d_name.name,
660 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
661 /* Unhash it, so that ->d_iput() would be called */
664 if (!(dentry->d_sb->s_flags & MS_ACTIVE)) {
665 /* Unhash it, so that ancestors of killed async unlink
666 * files will be cleaned up during umount */
674 * Called when the dentry loses inode.
675 * We use it to clean up silly-renamed files.
677 static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode)
679 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
682 nfs_complete_unlink(dentry);
685 /* When creating a negative dentry, we want to renew d_time */
686 nfs_renew_times(dentry);
690 struct dentry_operations nfs_dentry_operations = {
691 .d_revalidate = nfs_lookup_revalidate,
692 .d_delete = nfs_dentry_delete,
693 .d_iput = nfs_dentry_iput,
697 int nfs_is_exclusive_create(struct inode *dir, struct nameidata *nd)
699 if (NFS_PROTO(dir)->version == 2)
701 if (!nd || (nd->flags & LOOKUP_CONTINUE) || !(nd->flags & LOOKUP_CREATE))
703 return (nd->intent.open.flags & O_EXCL) != 0;
706 static struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
709 struct inode *inode = NULL;
711 struct nfs_fh fhandle;
712 struct nfs_fattr fattr;
714 dfprintk(VFS, "NFS: lookup(%s/%s)\n",
715 dentry->d_parent->d_name.name, dentry->d_name.name);
717 res = ERR_PTR(-ENAMETOOLONG);
718 if (dentry->d_name.len > NFS_SERVER(dir)->namelen)
721 res = ERR_PTR(-ENOMEM);
722 dentry->d_op = NFS_PROTO(dir)->dentry_ops;
725 /* Revalidate parent directory attribute cache */
726 nfs_revalidate_inode(NFS_SERVER(dir), dir);
728 /* If we're doing an exclusive create, optimize away the lookup */
729 if (nfs_is_exclusive_create(dir, nd))
732 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, &fhandle, &fattr);
733 if (error == -ENOENT)
736 res = ERR_PTR(error);
739 res = ERR_PTR(-EACCES);
740 inode = nfs_fhget(dentry->d_sb, &fhandle, &fattr);
743 vx_propagate_xid(nd, inode);
745 res = d_add_unique(dentry, inode);
748 nfs_renew_times(dentry);
749 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
757 static int nfs_open_revalidate(struct dentry *, struct nameidata *);
759 struct dentry_operations nfs4_dentry_operations = {
760 .d_revalidate = nfs_open_revalidate,
761 .d_delete = nfs_dentry_delete,
762 .d_iput = nfs_dentry_iput,
765 static int is_atomic_open(struct inode *dir, struct nameidata *nd)
769 /* Check that we are indeed trying to open this file */
770 if ((nd->flags & LOOKUP_CONTINUE) || !(nd->flags & LOOKUP_OPEN))
772 /* NFS does not (yet) have a stateful open for directories */
773 if (nd->flags & LOOKUP_DIRECTORY)
775 /* Are we trying to write to a read only partition? */
776 if (IS_RDONLY(dir) && (nd->intent.open.flags & (O_CREAT|O_TRUNC|FMODE_WRITE)))
781 static struct dentry *nfs_atomic_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd)
783 struct dentry *res = NULL;
784 struct inode *inode = NULL;
786 /* Check that we are indeed trying to open this file */
787 if (!is_atomic_open(dir, nd))
790 if (dentry->d_name.len > NFS_SERVER(dir)->namelen) {
791 res = ERR_PTR(-ENAMETOOLONG);
794 dentry->d_op = NFS_PROTO(dir)->dentry_ops;
796 /* Let vfs_create() deal with O_EXCL */
797 if (nd->intent.open.flags & O_EXCL)
800 /* Open the file on the server */
802 /* Revalidate parent directory attribute cache */
803 nfs_revalidate_inode(NFS_SERVER(dir), dir);
805 if (nd->intent.open.flags & O_CREAT) {
806 nfs_begin_data_update(dir);
807 inode = nfs4_atomic_open(dir, dentry, nd);
808 nfs_end_data_update(dir);
810 inode = nfs4_atomic_open(dir, dentry, nd);
813 int error = PTR_ERR(inode);
815 /* Make a negative dentry */
819 /* This turned out not to be a regular file */
821 if (!(nd->intent.open.flags & O_NOFOLLOW))
826 res = ERR_PTR(error);
831 res = d_add_unique(dentry, inode);
834 nfs_renew_times(dentry);
835 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
839 return nfs_lookup(dir, dentry, nd);
842 static int nfs_open_revalidate(struct dentry *dentry, struct nameidata *nd)
844 struct dentry *parent = NULL;
845 struct inode *inode = dentry->d_inode;
847 unsigned long verifier;
848 int openflags, ret = 0;
850 parent = dget_parent(dentry);
851 dir = parent->d_inode;
852 if (!is_atomic_open(dir, nd))
854 /* We can't create new files in nfs_open_revalidate(), so we
855 * optimize away revalidation of negative dentries.
859 /* NFS only supports OPEN on regular files */
860 if (!S_ISREG(inode->i_mode))
862 openflags = nd->intent.open.flags;
863 /* We cannot do exclusive creation on a positive dentry */
864 if ((openflags & (O_CREAT|O_EXCL)) == (O_CREAT|O_EXCL))
866 /* We can't create new files, or truncate existing ones here */
867 openflags &= ~(O_CREAT|O_TRUNC);
870 * Note: we're not holding inode->i_sem and so may be racing with
871 * operations that change the directory. We therefore save the
872 * change attribute *before* we do the RPC call.
875 verifier = nfs_save_change_attribute(dir);
876 ret = nfs4_open_revalidate(dir, dentry, openflags);
878 nfs_set_verifier(dentry, verifier);
887 if (inode != NULL && nfs_have_delegation(inode, FMODE_READ))
889 return nfs_lookup_revalidate(dentry, nd);
891 #endif /* CONFIG_NFSV4 */
893 static struct dentry *nfs_readdir_lookup(nfs_readdir_descriptor_t *desc)
895 struct dentry *parent = desc->file->f_dentry;
896 struct inode *dir = parent->d_inode;
897 struct nfs_entry *entry = desc->entry;
898 struct dentry *dentry, *alias;
907 if (name.name[0] == '.' && name.name[1] == '.')
908 return dget_parent(parent);
911 if (name.name[0] == '.')
914 name.hash = full_name_hash(name.name, name.len);
915 dentry = d_lookup(parent, &name);
918 if (!desc->plus || !(entry->fattr->valid & NFS_ATTR_FATTR))
920 /* Note: caller is already holding the dir->i_sem! */
921 dentry = d_alloc(parent, &name);
924 dentry->d_op = NFS_PROTO(dir)->dentry_ops;
925 inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr);
930 alias = d_add_unique(dentry, inode);
935 nfs_renew_times(dentry);
936 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
941 * Code common to create, mkdir, and mknod.
943 static int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle,
944 struct nfs_fattr *fattr)
949 /* We may have been initialized further down */
952 if (fhandle->size == 0) {
953 struct inode *dir = dentry->d_parent->d_inode;
954 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr);
958 if (!(fattr->valid & NFS_ATTR_FATTR)) {
959 struct nfs_server *server = NFS_SB(dentry->d_sb);
960 error = server->rpc_ops->getattr(server, fhandle, fattr);
964 inode = nfs_fhget(dentry->d_sb, fhandle, fattr);
966 d_instantiate(dentry, inode);
967 nfs_renew_times(dentry);
968 nfs_set_verifier(dentry, nfs_save_change_attribute(dentry->d_parent->d_inode));
978 * Following a failed create operation, we drop the dentry rather
979 * than retain a negative dentry. This avoids a problem in the event
980 * that the operation succeeded on the server, but an error in the
981 * reply path made it appear to have failed.
983 static int nfs_create(struct inode *dir, struct dentry *dentry, int mode,
984 struct nameidata *nd)
991 dfprintk(VFS, "NFS: create(%s/%ld, %s\n", dir->i_sb->s_id,
992 dir->i_ino, dentry->d_name.name);
995 attr.ia_valid = ATTR_MODE;
997 if (nd && (nd->flags & LOOKUP_CREATE))
998 open_flags = nd->intent.open.flags;
1001 nfs_begin_data_update(dir);
1002 inode = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags);
1003 nfs_end_data_update(dir);
1004 if (!IS_ERR(inode)) {
1005 d_instantiate(dentry, inode);
1006 nfs_renew_times(dentry);
1007 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1010 error = PTR_ERR(inode);
1018 * See comments for nfs_proc_create regarding failed operations.
1021 nfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t rdev)
1024 struct nfs_fattr fattr;
1025 struct nfs_fh fhandle;
1028 dfprintk(VFS, "NFS: mknod(%s/%ld, %s\n", dir->i_sb->s_id,
1029 dir->i_ino, dentry->d_name.name);
1031 if (!new_valid_dev(rdev))
1034 attr.ia_mode = mode;
1035 attr.ia_valid = ATTR_MODE;
1038 nfs_begin_data_update(dir);
1039 error = NFS_PROTO(dir)->mknod(dir, &dentry->d_name, &attr, rdev,
1041 nfs_end_data_update(dir);
1043 error = nfs_instantiate(dentry, &fhandle, &fattr);
1051 * See comments for nfs_proc_create regarding failed operations.
1053 static int nfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1056 struct nfs_fattr fattr;
1057 struct nfs_fh fhandle;
1060 dfprintk(VFS, "NFS: mkdir(%s/%ld, %s\n", dir->i_sb->s_id,
1061 dir->i_ino, dentry->d_name.name);
1063 attr.ia_valid = ATTR_MODE;
1064 attr.ia_mode = mode | S_IFDIR;
1069 * Always drop the dentry, we can't always depend on
1070 * the fattr returned by the server (AIX seems to be
1071 * broken). We're better off doing another lookup than
1072 * depending on potentially bogus information.
1076 nfs_begin_data_update(dir);
1077 error = NFS_PROTO(dir)->mkdir(dir, &dentry->d_name, &attr, &fhandle,
1079 nfs_end_data_update(dir);
1081 error = nfs_instantiate(dentry, &fhandle, &fattr);
1088 static int nfs_rmdir(struct inode *dir, struct dentry *dentry)
1092 dfprintk(VFS, "NFS: rmdir(%s/%ld, %s\n", dir->i_sb->s_id,
1093 dir->i_ino, dentry->d_name.name);
1096 nfs_begin_data_update(dir);
1097 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
1098 /* Ensure the VFS deletes this inode */
1099 if (error == 0 && dentry->d_inode != NULL)
1100 dentry->d_inode->i_nlink = 0;
1101 nfs_end_data_update(dir);
1107 static int nfs_sillyrename(struct inode *dir, struct dentry *dentry)
1109 static unsigned int sillycounter;
1110 const int i_inosize = sizeof(dir->i_ino)*2;
1111 const int countersize = sizeof(sillycounter)*2;
1112 const int slen = strlen(".nfs") + i_inosize + countersize;
1115 struct dentry *sdentry;
1118 dfprintk(VFS, "NFS: silly-rename(%s/%s, ct=%d)\n",
1119 dentry->d_parent->d_name.name, dentry->d_name.name,
1120 atomic_read(&dentry->d_count));
1123 if (!dentry->d_inode)
1124 printk("NFS: silly-renaming %s/%s, negative dentry??\n",
1125 dentry->d_parent->d_name.name, dentry->d_name.name);
1128 * We don't allow a dentry to be silly-renamed twice.
1131 if (dentry->d_flags & DCACHE_NFSFS_RENAMED)
1134 sprintf(silly, ".nfs%*.*lx",
1135 i_inosize, i_inosize, dentry->d_inode->i_ino);
1139 char *suffix = silly + slen - countersize;
1143 sprintf(suffix, "%*.*x", countersize, countersize, sillycounter);
1145 dfprintk(VFS, "trying to rename %s to %s\n",
1146 dentry->d_name.name, silly);
1148 sdentry = lookup_one_len(silly, dentry->d_parent, slen);
1150 * N.B. Better to return EBUSY here ... it could be
1151 * dangerous to delete the file while it's in use.
1153 if (IS_ERR(sdentry))
1155 } while(sdentry->d_inode != NULL); /* need negative lookup */
1157 qsilly.name = silly;
1158 qsilly.len = strlen(silly);
1159 nfs_begin_data_update(dir);
1160 if (dentry->d_inode) {
1161 nfs_begin_data_update(dentry->d_inode);
1162 error = NFS_PROTO(dir)->rename(dir, &dentry->d_name,
1164 nfs_end_data_update(dentry->d_inode);
1166 error = NFS_PROTO(dir)->rename(dir, &dentry->d_name,
1168 nfs_end_data_update(dir);
1170 nfs_renew_times(dentry);
1171 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1172 d_move(dentry, sdentry);
1173 error = nfs_async_unlink(dentry);
1174 /* If we return 0 we don't unlink */
1182 * Remove a file after making sure there are no pending writes,
1183 * and after checking that the file has only one user.
1185 * We invalidate the attribute cache and free the inode prior to the operation
1186 * to avoid possible races if the server reuses the inode.
1188 static int nfs_safe_remove(struct dentry *dentry)
1190 struct inode *dir = dentry->d_parent->d_inode;
1191 struct inode *inode = dentry->d_inode;
1194 dfprintk(VFS, "NFS: safe_remove(%s/%s)\n",
1195 dentry->d_parent->d_name.name, dentry->d_name.name);
1197 /* If the dentry was sillyrenamed, we simply call d_delete() */
1198 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1203 nfs_begin_data_update(dir);
1204 if (inode != NULL) {
1205 nfs_begin_data_update(inode);
1206 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1207 /* The VFS may want to delete this inode */
1210 nfs_end_data_update(inode);
1212 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1213 nfs_end_data_update(dir);
1218 /* We do silly rename. In case sillyrename() returns -EBUSY, the inode
1219 * belongs to an active ".nfs..." file and we return -EBUSY.
1221 * If sillyrename() returns 0, we do nothing, otherwise we unlink.
1223 static int nfs_unlink(struct inode *dir, struct dentry *dentry)
1226 int need_rehash = 0;
1228 dfprintk(VFS, "NFS: unlink(%s/%ld, %s)\n", dir->i_sb->s_id,
1229 dir->i_ino, dentry->d_name.name);
1232 spin_lock(&dcache_lock);
1233 spin_lock(&dentry->d_lock);
1234 if (atomic_read(&dentry->d_count) > 1) {
1235 spin_unlock(&dentry->d_lock);
1236 spin_unlock(&dcache_lock);
1237 error = nfs_sillyrename(dir, dentry);
1241 if (!d_unhashed(dentry)) {
1245 spin_unlock(&dentry->d_lock);
1246 spin_unlock(&dcache_lock);
1247 error = nfs_safe_remove(dentry);
1249 nfs_renew_times(dentry);
1250 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1251 } else if (need_rehash)
1258 nfs_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1261 struct nfs_fattr sym_attr;
1262 struct nfs_fh sym_fh;
1263 struct qstr qsymname;
1266 dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s)\n", dir->i_sb->s_id,
1267 dir->i_ino, dentry->d_name.name, symname);
1270 if (dentry->d_inode)
1271 printk("nfs_proc_symlink: %s/%s not negative!\n",
1272 dentry->d_parent->d_name.name, dentry->d_name.name);
1275 * Fill in the sattr for the call.
1276 * Note: SunOS 4.1.2 crashes if the mode isn't initialized!
1278 attr.ia_valid = ATTR_MODE;
1279 attr.ia_mode = S_IFLNK | S_IRWXUGO;
1281 qsymname.name = symname;
1282 qsymname.len = strlen(symname);
1285 nfs_begin_data_update(dir);
1286 error = NFS_PROTO(dir)->symlink(dir, &dentry->d_name, &qsymname,
1287 &attr, &sym_fh, &sym_attr);
1288 nfs_end_data_update(dir);
1290 error = nfs_instantiate(dentry, &sym_fh, &sym_attr);
1292 if (error == -EEXIST)
1293 printk("nfs_proc_symlink: %s/%s already exists??\n",
1294 dentry->d_parent->d_name.name, dentry->d_name.name);
1302 nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1304 struct inode *inode = old_dentry->d_inode;
1307 dfprintk(VFS, "NFS: link(%s/%s -> %s/%s)\n",
1308 old_dentry->d_parent->d_name.name, old_dentry->d_name.name,
1309 dentry->d_parent->d_name.name, dentry->d_name.name);
1312 * Drop the dentry in advance to force a new lookup.
1313 * Since nfs_proc_link doesn't return a file handle,
1314 * we can't use the existing dentry.
1319 nfs_begin_data_update(dir);
1320 nfs_begin_data_update(inode);
1321 error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name);
1322 nfs_end_data_update(inode);
1323 nfs_end_data_update(dir);
1330 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a
1331 * different file handle for the same inode after a rename (e.g. when
1332 * moving to a different directory). A fail-safe method to do so would
1333 * be to look up old_dir/old_name, create a link to new_dir/new_name and
1334 * rename the old file using the sillyrename stuff. This way, the original
1335 * file in old_dir will go away when the last process iput()s the inode.
1339 * It actually works quite well. One needs to have the possibility for
1340 * at least one ".nfs..." file in each directory the file ever gets
1341 * moved or linked to which happens automagically with the new
1342 * implementation that only depends on the dcache stuff instead of
1343 * using the inode layer
1345 * Unfortunately, things are a little more complicated than indicated
1346 * above. For a cross-directory move, we want to make sure we can get
1347 * rid of the old inode after the operation. This means there must be
1348 * no pending writes (if it's a file), and the use count must be 1.
1349 * If these conditions are met, we can drop the dentries before doing
1352 static int nfs_rename(struct inode *old_dir, struct dentry *old_dentry,
1353 struct inode *new_dir, struct dentry *new_dentry)
1355 struct inode *old_inode = old_dentry->d_inode;
1356 struct inode *new_inode = new_dentry->d_inode;
1357 struct dentry *dentry = NULL, *rehash = NULL;
1361 * To prevent any new references to the target during the rename,
1362 * we unhash the dentry and free the inode in advance.
1365 if (!d_unhashed(new_dentry)) {
1367 rehash = new_dentry;
1370 dfprintk(VFS, "NFS: rename(%s/%s -> %s/%s, ct=%d)\n",
1371 old_dentry->d_parent->d_name.name, old_dentry->d_name.name,
1372 new_dentry->d_parent->d_name.name, new_dentry->d_name.name,
1373 atomic_read(&new_dentry->d_count));
1376 * First check whether the target is busy ... we can't
1377 * safely do _any_ rename if the target is in use.
1379 * For files, make a copy of the dentry and then do a
1380 * silly-rename. If the silly-rename succeeds, the
1381 * copied dentry is hashed and becomes the new target.
1385 if (S_ISDIR(new_inode->i_mode))
1387 else if (atomic_read(&new_dentry->d_count) > 2) {
1389 /* copy the target dentry's name */
1390 dentry = d_alloc(new_dentry->d_parent,
1391 &new_dentry->d_name);
1395 /* silly-rename the existing target ... */
1396 err = nfs_sillyrename(new_dir, new_dentry);
1398 new_dentry = rehash = dentry;
1400 /* instantiate the replacement target */
1401 d_instantiate(new_dentry, NULL);
1402 } else if (atomic_read(&new_dentry->d_count) > 1) {
1403 /* dentry still busy? */
1405 printk("nfs_rename: target %s/%s busy, d_count=%d\n",
1406 new_dentry->d_parent->d_name.name,
1407 new_dentry->d_name.name,
1408 atomic_read(&new_dentry->d_count));
1416 * ... prune child dentries and writebacks if needed.
1418 if (atomic_read(&old_dentry->d_count) > 1) {
1419 nfs_wb_all(old_inode);
1420 shrink_dcache_parent(old_dentry);
1424 d_delete(new_dentry);
1426 nfs_begin_data_update(old_dir);
1427 nfs_begin_data_update(new_dir);
1428 nfs_begin_data_update(old_inode);
1429 error = NFS_PROTO(old_dir)->rename(old_dir, &old_dentry->d_name,
1430 new_dir, &new_dentry->d_name);
1431 nfs_end_data_update(old_inode);
1432 nfs_end_data_update(new_dir);
1433 nfs_end_data_update(old_dir);
1438 if (!S_ISDIR(old_inode->i_mode))
1439 d_move(old_dentry, new_dentry);
1440 nfs_renew_times(new_dentry);
1441 nfs_set_verifier(new_dentry, nfs_save_change_attribute(new_dir));
1444 /* new dentry created? */
1451 int nfs_access_get_cached(struct inode *inode, struct rpc_cred *cred, struct nfs_access_entry *res)
1453 struct nfs_access_entry *cache = &NFS_I(inode)->cache_access;
1455 if (cache->cred != cred
1456 || time_after(jiffies, cache->jiffies + NFS_ATTRTIMEO(inode))
1457 || (NFS_FLAGS(inode) & NFS_INO_INVALID_ACCESS))
1459 memcpy(res, cache, sizeof(*res));
1463 void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set)
1465 struct nfs_access_entry *cache = &NFS_I(inode)->cache_access;
1467 if (cache->cred != set->cred) {
1469 put_rpccred(cache->cred);
1470 cache->cred = get_rpccred(set->cred);
1471 NFS_FLAGS(inode) &= ~NFS_INO_INVALID_ACCESS;
1473 cache->jiffies = set->jiffies;
1474 cache->mask = set->mask;
1477 static int nfs_do_access(struct inode *inode, struct rpc_cred *cred, int mask)
1479 struct nfs_access_entry cache;
1482 status = nfs_access_get_cached(inode, cred, &cache);
1486 /* Be clever: ask server to check for all possible rights */
1487 cache.mask = MAY_EXEC | MAY_WRITE | MAY_READ;
1489 cache.jiffies = jiffies;
1490 status = NFS_PROTO(inode)->access(inode, &cache);
1493 nfs_access_add_cache(inode, &cache);
1495 if ((cache.mask & mask) == mask)
1500 int nfs_permission(struct inode *inode, int mask, struct nameidata *nd)
1502 struct rpc_cred *cred;
1508 /* Are we checking permissions on anything other than lookup/execute? */
1509 if ((mask & MAY_EXEC) == 0) {
1510 /* We only need to check permissions on file open() and access() */
1511 if (!nd || !(nd->flags & (LOOKUP_OPEN|LOOKUP_ACCESS)))
1513 /* NFSv4 has atomic_open... */
1514 if (NFS_PROTO(inode)->version > 3 && (nd->flags & LOOKUP_OPEN))
1520 if (!NFS_PROTO(inode)->access)
1523 cred = rpcauth_lookupcred(NFS_CLIENT(inode)->cl_auth, 0);
1524 res = nfs_do_access(inode, cred, mask);
1529 nfs_revalidate_inode(NFS_SERVER(inode), inode);
1530 res = generic_permission(inode, mask, NULL);
1537 * version-control: t
1538 * kept-new-versions: 5