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>
35 #define NFS_PARANOIA 1
36 /* #define NFS_DEBUG_VERBOSE 1 */
38 static int nfs_opendir(struct inode *, struct file *);
39 static int nfs_readdir(struct file *, void *, filldir_t);
40 static struct dentry *nfs_lookup(struct inode *, struct dentry *, struct nameidata *);
41 static int nfs_cached_lookup(struct inode *, struct dentry *,
42 struct nfs_fh *, struct nfs_fattr *);
43 static int nfs_create(struct inode *, struct dentry *, int, struct nameidata *);
44 static int nfs_mkdir(struct inode *, struct dentry *, int);
45 static int nfs_rmdir(struct inode *, struct dentry *);
46 static int nfs_unlink(struct inode *, struct dentry *);
47 static int nfs_symlink(struct inode *, struct dentry *, const char *);
48 static int nfs_link(struct dentry *, struct inode *, struct dentry *);
49 static int nfs_mknod(struct inode *, struct dentry *, int, dev_t);
50 static int nfs_rename(struct inode *, struct dentry *,
51 struct inode *, struct dentry *);
53 struct file_operations nfs_dir_operations = {
54 .read = generic_read_dir,
55 .readdir = nfs_readdir,
57 .release = nfs_release,
60 struct inode_operations nfs_dir_inode_operations = {
65 .symlink = nfs_symlink,
70 .permission = nfs_permission,
71 .getattr = nfs_getattr,
72 .setattr = nfs_setattr,
77 static struct dentry *nfs_atomic_lookup(struct inode *, struct dentry *, struct nameidata *);
78 struct inode_operations nfs4_dir_inode_operations = {
80 .lookup = nfs_atomic_lookup,
83 .symlink = nfs_symlink,
88 .permission = nfs_permission,
89 .getattr = nfs_getattr,
90 .setattr = nfs_setattr,
93 #endif /* CONFIG_NFS_V4 */
99 nfs_opendir(struct inode *inode, struct file *filp)
104 /* Call generic open code in order to cache credentials */
106 res = nfs_open(inode, filp);
111 typedef u32 * (*decode_dirent_t)(u32 *, struct nfs_entry *, int);
115 unsigned long page_index;
118 struct nfs_entry *entry;
119 decode_dirent_t decode;
122 } nfs_readdir_descriptor_t;
124 /* Now we cache directories properly, by stuffing the dirent
125 * data directly in the page cache.
127 * Inode invalidation due to refresh etc. takes care of
128 * _everything_, no sloppy entry flushing logic, no extraneous
129 * copying, network direct to page cache, the way it was meant
132 * NOTE: Dirent information verification is done always by the
133 * page-in of the RPC reply, nowhere else, this simplies
134 * things substantially.
137 int nfs_readdir_filler(nfs_readdir_descriptor_t *desc, struct page *page)
139 struct file *file = desc->file;
140 struct inode *inode = file->f_dentry->d_inode;
141 struct rpc_cred *cred = nfs_file_cred(file);
142 unsigned long timestamp;
145 dfprintk(VFS, "NFS: nfs_readdir_filler() reading cookie %Lu into page %lu.\n", (long long)desc->entry->cookie, page->index);
149 error = NFS_PROTO(inode)->readdir(file->f_dentry, cred, desc->entry->cookie, page,
150 NFS_SERVER(inode)->dtsize, desc->plus);
152 /* We requested READDIRPLUS, but the server doesn't grok it */
153 if (error == -ENOTSUPP && desc->plus) {
154 NFS_SERVER(inode)->caps &= ~NFS_CAP_READDIRPLUS;
155 NFS_FLAGS(inode) &= ~NFS_INO_ADVISE_RDPLUS;
161 SetPageUptodate(page);
162 NFS_FLAGS(inode) |= NFS_INO_INVALID_ATIME;
163 /* Ensure consistent page alignment of the data.
164 * Note: assumes we have exclusive access to this mapping either
165 * throught inode->i_sem or some other mechanism.
167 if (page->index == 0) {
168 invalidate_inode_pages(inode->i_mapping);
169 NFS_I(inode)->readdir_timestamp = timestamp;
176 nfs_zap_caches(inode);
182 int dir_decode(nfs_readdir_descriptor_t *desc)
185 p = desc->decode(p, desc->entry, desc->plus);
193 void dir_page_release(nfs_readdir_descriptor_t *desc)
196 page_cache_release(desc->page);
202 * Given a pointer to a buffer that has already been filled by a call
203 * to readdir, find the next entry.
205 * If the end of the buffer has been reached, return -EAGAIN, if not,
206 * return the offset within the buffer of the next entry to be
210 int find_dirent(nfs_readdir_descriptor_t *desc, struct page *page)
212 struct nfs_entry *entry = desc->entry;
216 while((status = dir_decode(desc)) == 0) {
217 dfprintk(VFS, "NFS: found cookie %Lu\n", (long long)entry->cookie);
218 if (entry->prev_cookie == desc->target)
220 if (loop_count++ > 200) {
225 dfprintk(VFS, "NFS: find_dirent() returns %d\n", status);
230 * Find the given page, and call find_dirent() in order to try to
231 * return the next entry.
234 int find_dirent_page(nfs_readdir_descriptor_t *desc)
236 struct inode *inode = desc->file->f_dentry->d_inode;
240 dfprintk(VFS, "NFS: find_dirent_page() searching directory page %ld\n", desc->page_index);
242 page = read_cache_page(inode->i_mapping, desc->page_index,
243 (filler_t *)nfs_readdir_filler, desc);
245 status = PTR_ERR(page);
248 if (!PageUptodate(page))
251 /* NOTE: Someone else may have changed the READDIRPLUS flag */
253 desc->ptr = kmap(page); /* matching kunmap in nfs_do_filldir */
254 status = find_dirent(desc, page);
256 dir_page_release(desc);
258 dfprintk(VFS, "NFS: find_dirent_page() returns %d\n", status);
261 page_cache_release(page);
266 * Recurse through the page cache pages, and return a
267 * filled nfs_entry structure of the next directory entry if possible.
269 * The target for the search is 'desc->target'.
272 int readdir_search_pagecache(nfs_readdir_descriptor_t *desc)
277 dfprintk(VFS, "NFS: readdir_search_pagecache() searching for cookie %Lu\n", (long long)desc->target);
279 res = find_dirent_page(desc);
282 /* Align to beginning of next page */
284 if (loop_count++ > 200) {
289 dfprintk(VFS, "NFS: readdir_search_pagecache() returned %d\n", res);
293 static unsigned int nfs_type2dtype[] = {
306 unsigned int nfs_type_to_d_type(enum nfs_ftype type)
308 return nfs_type2dtype[type];
312 * Once we've found the start of the dirent within a page: fill 'er up...
315 int nfs_do_filldir(nfs_readdir_descriptor_t *desc, void *dirent,
318 struct file *file = desc->file;
319 struct nfs_entry *entry = desc->entry;
320 unsigned long fileid;
324 dfprintk(VFS, "NFS: nfs_do_filldir() filling starting @ cookie %Lu\n", (long long)desc->target);
327 unsigned d_type = DT_UNKNOWN;
328 /* Note: entry->prev_cookie contains the cookie for
329 * retrieving the current dirent on the server */
330 fileid = nfs_fileid_to_ino_t(entry->ino);
332 /* Use readdirplus info */
333 if (desc->plus && (entry->fattr->valid & NFS_ATTR_FATTR))
334 d_type = nfs_type_to_d_type(entry->fattr->type);
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);
352 dfprintk(VFS, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n", (long long)desc->target, res);
357 * If we cannot find a cookie in our cache, we suspect that this is
358 * because it points to a deleted file, so we ask the server to return
359 * whatever it thinks is the next entry. We then feed this to filldir.
360 * If all goes well, we should then be able to find our way round the
361 * cache on the next call to readdir_search_pagecache();
363 * NOTE: we cannot add the anonymous page to the pagecache because
364 * the data it contains might not be page aligned. Besides,
365 * we should already have a complete representation of the
366 * directory in the page cache by the time we get here.
369 int uncached_readdir(nfs_readdir_descriptor_t *desc, void *dirent,
372 struct file *file = desc->file;
373 struct inode *inode = file->f_dentry->d_inode;
374 struct rpc_cred *cred = nfs_file_cred(file);
375 struct page *page = NULL;
378 dfprintk(VFS, "NFS: uncached_readdir() searching for cookie %Lu\n", (long long)desc->target);
380 page = alloc_page(GFP_HIGHUSER);
385 desc->error = NFS_PROTO(inode)->readdir(file->f_dentry, cred, desc->target,
387 NFS_SERVER(inode)->dtsize,
389 NFS_FLAGS(inode) |= NFS_INO_INVALID_ATIME;
391 desc->ptr = kmap(page); /* matching kunmap in nfs_do_filldir */
392 if (desc->error >= 0) {
393 if ((status = dir_decode(desc)) == 0)
394 desc->entry->prev_cookie = desc->target;
400 status = nfs_do_filldir(desc, dirent, filldir);
402 /* Reset read descriptor so it searches the page cache from
403 * the start upon the next call to readdir_search_pagecache() */
404 desc->page_index = 0;
405 desc->entry->cookie = desc->entry->prev_cookie = 0;
406 desc->entry->eof = 0;
408 dfprintk(VFS, "NFS: uncached_readdir() returns %d\n", status);
411 dir_page_release(desc);
415 /* The file offset position is now represented as a true offset into the
416 * page cache as is the case in most of the other filesystems.
418 static int nfs_readdir(struct file *filp, void *dirent, filldir_t filldir)
420 struct dentry *dentry = filp->f_dentry;
421 struct inode *inode = dentry->d_inode;
422 nfs_readdir_descriptor_t my_desc,
424 struct nfs_entry my_entry;
426 struct nfs_fattr fattr;
431 res = nfs_revalidate_inode(NFS_SERVER(inode), inode);
438 * filp->f_pos points to the file offset in the page cache.
439 * but if the cache has meanwhile been zapped, we need to
440 * read from the last dirent to revalidate f_pos
443 memset(desc, 0, sizeof(*desc));
446 desc->target = filp->f_pos;
447 desc->decode = NFS_PROTO(inode)->decode_dirent;
448 desc->plus = NFS_USE_READDIRPLUS(inode);
450 my_entry.cookie = my_entry.prev_cookie = 0;
453 my_entry.fattr = &fattr;
454 desc->entry = &my_entry;
456 while(!desc->entry->eof) {
457 res = readdir_search_pagecache(desc);
458 if (res == -EBADCOOKIE) {
459 /* This means either end of directory */
460 if (desc->entry->cookie != desc->target) {
461 /* Or that the server has 'lost' a cookie */
462 res = uncached_readdir(desc, dirent, filldir);
469 if (res == -ETOOSMALL && desc->plus) {
470 NFS_FLAGS(inode) &= ~NFS_INO_ADVISE_RDPLUS;
471 nfs_zap_caches(inode);
473 desc->entry->eof = 0;
479 res = nfs_do_filldir(desc, dirent, filldir);
494 * A check for whether or not the parent directory has changed.
495 * In the case it has, we assume that the dentries are untrustworthy
496 * and may need to be looked up again.
498 static inline int nfs_check_verifier(struct inode *dir, struct dentry *dentry)
502 if ((NFS_FLAGS(dir) & NFS_INO_INVALID_ATTR) != 0
503 || nfs_attribute_timeout(dir))
505 return nfs_verify_change_attribute(dir, (unsigned long)dentry->d_fsdata);
508 static inline void nfs_set_verifier(struct dentry * dentry, unsigned long verf)
510 dentry->d_fsdata = (void *)verf;
514 * Whenever an NFS operation succeeds, we know that the dentry
515 * is valid, so we update the revalidation timestamp.
517 static inline void nfs_renew_times(struct dentry * dentry)
519 dentry->d_time = jiffies;
523 int nfs_lookup_verify_inode(struct inode *inode, int isopen)
525 struct nfs_server *server = NFS_SERVER(inode);
527 if (isopen && !(server->flags & NFS_MOUNT_NOCTO))
528 return __nfs_revalidate_inode(server, inode);
529 return nfs_revalidate_inode(server, inode);
533 * We judge how long we want to trust negative
534 * dentries by looking at the parent inode mtime.
536 * If parent mtime has changed, we revalidate, else we wait for a
537 * period corresponding to the parent's attribute cache timeout value.
540 int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry,
541 struct nameidata *nd)
547 /* Don't revalidate a negative dentry if we're creating a new file */
548 if ((ndflags & LOOKUP_CREATE) && !(ndflags & LOOKUP_CONTINUE))
550 return !nfs_check_verifier(dir, dentry);
554 * This is called every time the dcache has a lookup hit,
555 * and we should check whether we can really trust that
558 * NOTE! The hit can be a negative hit too, don't assume
561 * If the parent directory is seen to have changed, we throw out the
562 * cached dentry and do a new lookup.
564 static int nfs_lookup_revalidate(struct dentry * dentry, struct nameidata *nd)
568 struct dentry *parent;
570 struct nfs_fh fhandle;
571 struct nfs_fattr fattr;
572 unsigned long verifier;
575 parent = dget_parent(dentry);
577 dir = parent->d_inode;
578 inode = dentry->d_inode;
580 if (nd && !(nd->flags & LOOKUP_CONTINUE) && (nd->flags & LOOKUP_OPEN))
584 if (nfs_neg_need_reval(dir, dentry, nd))
589 if (is_bad_inode(inode)) {
590 dfprintk(VFS, "nfs_lookup_validate: %s/%s has dud inode\n",
591 dentry->d_parent->d_name.name, dentry->d_name.name);
595 /* Revalidate parent directory attribute cache */
596 nfs_revalidate_inode(NFS_SERVER(dir), dir);
598 /* Force a full look up iff the parent directory has changed */
599 if (nfs_check_verifier(dir, dentry)) {
600 if (nfs_lookup_verify_inode(inode, isopen))
606 * Note: we're not holding inode->i_sem and so may be racing with
607 * operations that change the directory. We therefore save the
608 * change attribute *before* we do the RPC call.
610 verifier = nfs_save_change_attribute(dir);
611 error = nfs_cached_lookup(dir, dentry, &fhandle, &fattr);
613 if (memcmp(NFS_FH(inode), &fhandle, sizeof(struct nfs_fh))!= 0)
615 if (nfs_lookup_verify_inode(inode, isopen))
617 goto out_valid_renew;
620 if (NFS_STALE(inode))
623 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, &fhandle, &fattr);
626 if (memcmp(NFS_FH(inode), &fhandle, sizeof(struct nfs_fh))!= 0)
628 if ((error = nfs_refresh_inode(inode, &fattr)) != 0)
632 nfs_renew_times(dentry);
633 nfs_set_verifier(dentry, verifier);
642 if (inode && S_ISDIR(inode->i_mode)) {
643 /* Purge readdir caches. */
644 nfs_zap_caches(inode);
645 /* If we have submounts, don't unhash ! */
646 if (have_submounts(dentry))
648 shrink_dcache_parent(dentry);
657 * This is called from dput() when d_count is going to 0.
659 static int nfs_dentry_delete(struct dentry *dentry)
661 dfprintk(VFS, "NFS: dentry_delete(%s/%s, %x)\n",
662 dentry->d_parent->d_name.name, dentry->d_name.name,
665 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
666 /* Unhash it, so that ->d_iput() would be called */
669 if (!(dentry->d_sb->s_flags & MS_ACTIVE)) {
670 /* Unhash it, so that ancestors of killed async unlink
671 * files will be cleaned up during umount */
679 * Called when the dentry loses inode.
680 * We use it to clean up silly-renamed files.
682 static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode)
684 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
687 nfs_complete_unlink(dentry);
690 /* When creating a negative dentry, we want to renew d_time */
691 nfs_renew_times(dentry);
695 struct dentry_operations nfs_dentry_operations = {
696 .d_revalidate = nfs_lookup_revalidate,
697 .d_delete = nfs_dentry_delete,
698 .d_iput = nfs_dentry_iput,
702 int nfs_is_exclusive_create(struct inode *dir, struct nameidata *nd)
704 if (NFS_PROTO(dir)->version == 2)
706 if (!nd || (nd->flags & LOOKUP_CONTINUE) || !(nd->flags & LOOKUP_CREATE))
708 return (nd->intent.open.flags & O_EXCL) != 0;
711 static struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
713 struct inode *inode = NULL;
715 struct nfs_fh fhandle;
716 struct nfs_fattr fattr;
718 dfprintk(VFS, "NFS: lookup(%s/%s)\n",
719 dentry->d_parent->d_name.name, dentry->d_name.name);
721 error = -ENAMETOOLONG;
722 if (dentry->d_name.len > NFS_SERVER(dir)->namelen)
726 dentry->d_op = NFS_PROTO(dir)->dentry_ops;
729 /* Revalidate parent directory attribute cache */
730 nfs_revalidate_inode(NFS_SERVER(dir), dir);
732 /* If we're doing an exclusive create, optimize away the lookup */
733 if (nfs_is_exclusive_create(dir, nd))
736 error = nfs_cached_lookup(dir, dentry, &fhandle, &fattr);
738 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name,
740 if (error == -ENOENT)
746 inode = nfs_fhget(dentry->d_sb, &fhandle, &fattr);
751 d_add(dentry, inode);
752 nfs_renew_times(dentry);
753 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
758 return ERR_PTR(error);
762 static int nfs_open_revalidate(struct dentry *, struct nameidata *);
764 struct dentry_operations nfs4_dentry_operations = {
765 .d_revalidate = nfs_open_revalidate,
766 .d_delete = nfs_dentry_delete,
767 .d_iput = nfs_dentry_iput,
770 static int is_atomic_open(struct inode *dir, struct nameidata *nd)
774 /* Check that we are indeed trying to open this file */
775 if ((nd->flags & LOOKUP_CONTINUE) || !(nd->flags & LOOKUP_OPEN))
777 /* NFS does not (yet) have a stateful open for directories */
778 if (nd->flags & LOOKUP_DIRECTORY)
780 /* Are we trying to write to a read only partition? */
781 if (IS_RDONLY(dir) && (nd->intent.open.flags & (O_CREAT|O_TRUNC|FMODE_WRITE)))
786 static struct dentry *nfs_atomic_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd)
788 struct inode *inode = NULL;
791 /* Check that we are indeed trying to open this file */
792 if (!is_atomic_open(dir, nd))
795 if (dentry->d_name.len > NFS_SERVER(dir)->namelen) {
796 error = -ENAMETOOLONG;
799 dentry->d_op = NFS_PROTO(dir)->dentry_ops;
801 /* Let vfs_create() deal with O_EXCL */
802 if (nd->intent.open.flags & O_EXCL)
805 /* Open the file on the server */
807 /* Revalidate parent directory attribute cache */
808 nfs_revalidate_inode(NFS_SERVER(dir), dir);
810 if (nd->intent.open.flags & O_CREAT) {
811 nfs_begin_data_update(dir);
812 inode = nfs4_atomic_open(dir, dentry, nd);
813 nfs_end_data_update(dir);
815 inode = nfs4_atomic_open(dir, dentry, nd);
818 error = PTR_ERR(inode);
820 /* Make a negative dentry */
824 /* This turned out not to be a regular file */
826 if (!(nd->intent.open.flags & O_NOFOLLOW))
835 d_add(dentry, inode);
836 nfs_renew_times(dentry);
837 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
840 return ERR_PTR(error);
842 return nfs_lookup(dir, dentry, nd);
845 static int nfs_open_revalidate(struct dentry *dentry, struct nameidata *nd)
847 struct dentry *parent = NULL;
848 struct inode *inode = dentry->d_inode;
850 unsigned long verifier;
851 int openflags, ret = 0;
853 /* NFS only supports OPEN for regular files */
854 if (inode && !S_ISREG(inode->i_mode))
856 parent = dget_parent(dentry);
857 dir = parent->d_inode;
858 if (!is_atomic_open(dir, nd))
860 openflags = nd->intent.open.flags;
861 if (openflags & O_CREAT) {
862 /* If this is a negative dentry, just drop it */
865 /* If this is exclusive open, just revalidate */
866 if (openflags & O_EXCL)
869 /* We can't create new files, or truncate existing ones here */
870 openflags &= ~(O_CREAT|O_TRUNC);
873 * Note: we're not holding inode->i_sem and so may be racing with
874 * operations that change the directory. We therefore save the
875 * change attribute *before* we do the RPC call.
878 verifier = nfs_save_change_attribute(dir);
879 ret = nfs4_open_revalidate(dir, dentry, openflags);
881 nfs_set_verifier(dentry, verifier);
890 return nfs_lookup_revalidate(dentry, nd);
892 #endif /* CONFIG_NFSV4 */
895 int find_dirent_name(nfs_readdir_descriptor_t *desc, struct page *page, struct dentry *dentry)
897 struct nfs_entry *entry = desc->entry;
900 while((status = dir_decode(desc)) == 0) {
901 if (entry->len != dentry->d_name.len)
903 if (memcmp(entry->name, dentry->d_name.name, entry->len))
905 if (!(entry->fattr->valid & NFS_ATTR_FATTR))
913 * Use the cached Readdirplus results in order to avoid a LOOKUP call
914 * whenever we believe that the parent directory has not changed.
916 * We assume that any file creation/rename changes the directory mtime.
917 * As this results in a page cache invalidation whenever it occurs,
918 * we don't require any other tests for cache coherency.
921 int nfs_cached_lookup(struct inode *dir, struct dentry *dentry,
922 struct nfs_fh *fh, struct nfs_fattr *fattr)
924 nfs_readdir_descriptor_t desc;
925 struct nfs_server *server;
926 struct nfs_entry entry;
928 unsigned long timestamp;
931 if (!NFS_USE_READDIRPLUS(dir))
933 server = NFS_SERVER(dir);
934 /* Don't use readdirplus unless the cache is stable */
935 if ((server->flags & NFS_MOUNT_NOAC) != 0
936 || nfs_caches_unstable(dir)
937 || nfs_attribute_timeout(dir))
939 if ((NFS_FLAGS(dir) & (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA)) != 0)
941 timestamp = NFS_I(dir)->readdir_timestamp;
946 desc.decode = NFS_PROTO(dir)->decode_dirent;
951 for(;(page = find_get_page(dir->i_mapping, desc.page_index)); desc.page_index++) {
954 if (PageUptodate(page)) {
955 void * kaddr = kmap_atomic(page, KM_USER0);
957 res = find_dirent_name(&desc, page, dentry);
958 kunmap_atomic(kaddr, KM_USER0);
960 page_cache_release(page);
969 fattr->timestamp = timestamp;
974 * Code common to create, mkdir, and mknod.
976 static int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle,
977 struct nfs_fattr *fattr)
982 /* We may have been initialized further down */
985 if (fhandle->size == 0 || !(fattr->valid & NFS_ATTR_FATTR)) {
986 struct inode *dir = dentry->d_parent->d_inode;
987 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr);
991 inode = nfs_fhget(dentry->d_sb, fhandle, fattr);
993 d_instantiate(dentry, inode);
994 nfs_renew_times(dentry);
995 nfs_set_verifier(dentry, nfs_save_change_attribute(dentry->d_parent->d_inode));
1005 * Following a failed create operation, we drop the dentry rather
1006 * than retain a negative dentry. This avoids a problem in the event
1007 * that the operation succeeded on the server, but an error in the
1008 * reply path made it appear to have failed.
1010 static int nfs_create(struct inode *dir, struct dentry *dentry, int mode,
1011 struct nameidata *nd)
1014 struct inode *inode;
1018 dfprintk(VFS, "NFS: create(%s/%ld, %s\n", dir->i_sb->s_id,
1019 dir->i_ino, dentry->d_name.name);
1021 attr.ia_mode = mode;
1022 attr.ia_valid = ATTR_MODE;
1024 if (nd && (nd->flags & LOOKUP_CREATE))
1025 open_flags = nd->intent.open.flags;
1028 * The 0 argument passed into the create function should one day
1029 * contain the O_EXCL flag if requested. This allows NFSv3 to
1030 * select the appropriate create strategy. Currently open_namei
1031 * does not pass the create flags.
1034 nfs_begin_data_update(dir);
1035 inode = NFS_PROTO(dir)->create(dir, &dentry->d_name, &attr, open_flags);
1036 nfs_end_data_update(dir);
1037 if (!IS_ERR(inode)) {
1038 d_instantiate(dentry, inode);
1039 nfs_renew_times(dentry);
1040 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1043 error = PTR_ERR(inode);
1051 * See comments for nfs_proc_create regarding failed operations.
1054 nfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t rdev)
1057 struct nfs_fattr fattr;
1058 struct nfs_fh fhandle;
1061 dfprintk(VFS, "NFS: mknod(%s/%ld, %s\n", dir->i_sb->s_id,
1062 dir->i_ino, dentry->d_name.name);
1064 if (!new_valid_dev(rdev))
1067 attr.ia_mode = mode;
1068 attr.ia_valid = ATTR_MODE;
1071 nfs_begin_data_update(dir);
1072 error = NFS_PROTO(dir)->mknod(dir, &dentry->d_name, &attr, rdev,
1074 nfs_end_data_update(dir);
1076 error = nfs_instantiate(dentry, &fhandle, &fattr);
1084 * See comments for nfs_proc_create regarding failed operations.
1086 static int nfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1089 struct nfs_fattr fattr;
1090 struct nfs_fh fhandle;
1093 dfprintk(VFS, "NFS: mkdir(%s/%ld, %s\n", dir->i_sb->s_id,
1094 dir->i_ino, dentry->d_name.name);
1096 attr.ia_valid = ATTR_MODE;
1097 attr.ia_mode = mode | S_IFDIR;
1102 * Always drop the dentry, we can't always depend on
1103 * the fattr returned by the server (AIX seems to be
1104 * broken). We're better off doing another lookup than
1105 * depending on potentially bogus information.
1109 nfs_begin_data_update(dir);
1110 error = NFS_PROTO(dir)->mkdir(dir, &dentry->d_name, &attr, &fhandle,
1112 nfs_end_data_update(dir);
1114 error = nfs_instantiate(dentry, &fhandle, &fattr);
1121 static int nfs_rmdir(struct inode *dir, struct dentry *dentry)
1125 dfprintk(VFS, "NFS: rmdir(%s/%ld, %s\n", dir->i_sb->s_id,
1126 dir->i_ino, dentry->d_name.name);
1129 nfs_begin_data_update(dir);
1130 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
1131 /* Ensure the VFS deletes this inode */
1132 if (error == 0 && dentry->d_inode != NULL)
1133 dentry->d_inode->i_nlink = 0;
1134 nfs_end_data_update(dir);
1140 static int nfs_sillyrename(struct inode *dir, struct dentry *dentry)
1142 static unsigned int sillycounter;
1143 const int i_inosize = sizeof(dir->i_ino)*2;
1144 const int countersize = sizeof(sillycounter)*2;
1145 const int slen = strlen(".nfs") + i_inosize + countersize;
1148 struct dentry *sdentry;
1151 dfprintk(VFS, "NFS: silly-rename(%s/%s, ct=%d)\n",
1152 dentry->d_parent->d_name.name, dentry->d_name.name,
1153 atomic_read(&dentry->d_count));
1156 if (!dentry->d_inode)
1157 printk("NFS: silly-renaming %s/%s, negative dentry??\n",
1158 dentry->d_parent->d_name.name, dentry->d_name.name);
1161 * We don't allow a dentry to be silly-renamed twice.
1164 if (dentry->d_flags & DCACHE_NFSFS_RENAMED)
1167 sprintf(silly, ".nfs%*.*lx",
1168 i_inosize, i_inosize, dentry->d_inode->i_ino);
1172 char *suffix = silly + slen - countersize;
1176 sprintf(suffix, "%*.*x", countersize, countersize, sillycounter);
1178 dfprintk(VFS, "trying to rename %s to %s\n",
1179 dentry->d_name.name, silly);
1181 sdentry = lookup_one_len(silly, dentry->d_parent, slen);
1183 * N.B. Better to return EBUSY here ... it could be
1184 * dangerous to delete the file while it's in use.
1186 if (IS_ERR(sdentry))
1188 } while(sdentry->d_inode != NULL); /* need negative lookup */
1190 qsilly.name = silly;
1191 qsilly.len = strlen(silly);
1192 nfs_begin_data_update(dir);
1193 if (dentry->d_inode) {
1194 nfs_begin_data_update(dentry->d_inode);
1195 error = NFS_PROTO(dir)->rename(dir, &dentry->d_name,
1197 nfs_end_data_update(dentry->d_inode);
1199 error = NFS_PROTO(dir)->rename(dir, &dentry->d_name,
1201 nfs_end_data_update(dir);
1203 nfs_renew_times(dentry);
1204 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1205 d_move(dentry, sdentry);
1206 error = nfs_async_unlink(dentry);
1207 /* If we return 0 we don't unlink */
1215 * Remove a file after making sure there are no pending writes,
1216 * and after checking that the file has only one user.
1218 * We invalidate the attribute cache and free the inode prior to the operation
1219 * to avoid possible races if the server reuses the inode.
1221 static int nfs_safe_remove(struct dentry *dentry)
1223 struct inode *dir = dentry->d_parent->d_inode;
1224 struct inode *inode = dentry->d_inode;
1227 dfprintk(VFS, "NFS: safe_remove(%s/%s)\n",
1228 dentry->d_parent->d_name.name, dentry->d_name.name);
1230 /* If the dentry was sillyrenamed, we simply call d_delete() */
1231 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1236 nfs_begin_data_update(dir);
1237 if (inode != NULL) {
1238 nfs_begin_data_update(inode);
1239 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1240 /* The VFS may want to delete this inode */
1243 nfs_end_data_update(inode);
1245 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1246 nfs_end_data_update(dir);
1251 /* We do silly rename. In case sillyrename() returns -EBUSY, the inode
1252 * belongs to an active ".nfs..." file and we return -EBUSY.
1254 * If sillyrename() returns 0, we do nothing, otherwise we unlink.
1256 static int nfs_unlink(struct inode *dir, struct dentry *dentry)
1259 int need_rehash = 0;
1261 dfprintk(VFS, "NFS: unlink(%s/%ld, %s)\n", dir->i_sb->s_id,
1262 dir->i_ino, dentry->d_name.name);
1265 spin_lock(&dcache_lock);
1266 spin_lock(&dentry->d_lock);
1267 if (atomic_read(&dentry->d_count) > 1) {
1268 spin_unlock(&dentry->d_lock);
1269 spin_unlock(&dcache_lock);
1270 error = nfs_sillyrename(dir, dentry);
1274 if (!d_unhashed(dentry)) {
1278 spin_unlock(&dentry->d_lock);
1279 spin_unlock(&dcache_lock);
1280 error = nfs_safe_remove(dentry);
1282 nfs_renew_times(dentry);
1283 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1284 } else if (need_rehash)
1291 nfs_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1294 struct nfs_fattr sym_attr;
1295 struct nfs_fh sym_fh;
1296 struct qstr qsymname;
1299 dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s)\n", dir->i_sb->s_id,
1300 dir->i_ino, dentry->d_name.name, symname);
1302 error = -ENAMETOOLONG;
1303 switch (NFS_PROTO(dir)->version) {
1305 if (strlen(symname) > NFS2_MAXPATHLEN)
1309 if (strlen(symname) > NFS3_MAXPATHLEN)
1316 if (dentry->d_inode)
1317 printk("nfs_proc_symlink: %s/%s not negative!\n",
1318 dentry->d_parent->d_name.name, dentry->d_name.name);
1321 * Fill in the sattr for the call.
1322 * Note: SunOS 4.1.2 crashes if the mode isn't initialized!
1324 attr.ia_valid = ATTR_MODE;
1325 attr.ia_mode = S_IFLNK | S_IRWXUGO;
1327 qsymname.name = symname;
1328 qsymname.len = strlen(symname);
1331 nfs_begin_data_update(dir);
1332 error = NFS_PROTO(dir)->symlink(dir, &dentry->d_name, &qsymname,
1333 &attr, &sym_fh, &sym_attr);
1334 nfs_end_data_update(dir);
1336 error = nfs_instantiate(dentry, &sym_fh, &sym_attr);
1338 if (error == -EEXIST)
1339 printk("nfs_proc_symlink: %s/%s already exists??\n",
1340 dentry->d_parent->d_name.name, dentry->d_name.name);
1350 nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1352 struct inode *inode = old_dentry->d_inode;
1355 dfprintk(VFS, "NFS: link(%s/%s -> %s/%s)\n",
1356 old_dentry->d_parent->d_name.name, old_dentry->d_name.name,
1357 dentry->d_parent->d_name.name, dentry->d_name.name);
1360 * Drop the dentry in advance to force a new lookup.
1361 * Since nfs_proc_link doesn't return a file handle,
1362 * we can't use the existing dentry.
1367 nfs_begin_data_update(dir);
1368 nfs_begin_data_update(inode);
1369 error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name);
1370 nfs_end_data_update(inode);
1371 nfs_end_data_update(dir);
1378 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a
1379 * different file handle for the same inode after a rename (e.g. when
1380 * moving to a different directory). A fail-safe method to do so would
1381 * be to look up old_dir/old_name, create a link to new_dir/new_name and
1382 * rename the old file using the sillyrename stuff. This way, the original
1383 * file in old_dir will go away when the last process iput()s the inode.
1387 * It actually works quite well. One needs to have the possibility for
1388 * at least one ".nfs..." file in each directory the file ever gets
1389 * moved or linked to which happens automagically with the new
1390 * implementation that only depends on the dcache stuff instead of
1391 * using the inode layer
1393 * Unfortunately, things are a little more complicated than indicated
1394 * above. For a cross-directory move, we want to make sure we can get
1395 * rid of the old inode after the operation. This means there must be
1396 * no pending writes (if it's a file), and the use count must be 1.
1397 * If these conditions are met, we can drop the dentries before doing
1400 static int nfs_rename(struct inode *old_dir, struct dentry *old_dentry,
1401 struct inode *new_dir, struct dentry *new_dentry)
1403 struct inode *old_inode = old_dentry->d_inode;
1404 struct inode *new_inode = new_dentry->d_inode;
1405 struct dentry *dentry = NULL, *rehash = NULL;
1409 * To prevent any new references to the target during the rename,
1410 * we unhash the dentry and free the inode in advance.
1413 if (!d_unhashed(new_dentry)) {
1415 rehash = new_dentry;
1418 dfprintk(VFS, "NFS: rename(%s/%s -> %s/%s, ct=%d)\n",
1419 old_dentry->d_parent->d_name.name, old_dentry->d_name.name,
1420 new_dentry->d_parent->d_name.name, new_dentry->d_name.name,
1421 atomic_read(&new_dentry->d_count));
1424 * First check whether the target is busy ... we can't
1425 * safely do _any_ rename if the target is in use.
1427 * For files, make a copy of the dentry and then do a
1428 * silly-rename. If the silly-rename succeeds, the
1429 * copied dentry is hashed and becomes the new target.
1433 if (S_ISDIR(new_inode->i_mode))
1435 else if (atomic_read(&new_dentry->d_count) > 1) {
1437 /* copy the target dentry's name */
1438 dentry = d_alloc(new_dentry->d_parent,
1439 &new_dentry->d_name);
1443 /* silly-rename the existing target ... */
1444 err = nfs_sillyrename(new_dir, new_dentry);
1446 new_dentry = rehash = dentry;
1448 /* instantiate the replacement target */
1449 d_instantiate(new_dentry, NULL);
1452 /* dentry still busy? */
1453 if (atomic_read(&new_dentry->d_count) > 1) {
1455 printk("nfs_rename: target %s/%s busy, d_count=%d\n",
1456 new_dentry->d_parent->d_name.name,
1457 new_dentry->d_name.name,
1458 atomic_read(&new_dentry->d_count));
1466 * ... prune child dentries and writebacks if needed.
1468 if (atomic_read(&old_dentry->d_count) > 1) {
1469 nfs_wb_all(old_inode);
1470 shrink_dcache_parent(old_dentry);
1474 d_delete(new_dentry);
1476 nfs_begin_data_update(old_dir);
1477 nfs_begin_data_update(new_dir);
1478 nfs_begin_data_update(old_inode);
1479 error = NFS_PROTO(old_dir)->rename(old_dir, &old_dentry->d_name,
1480 new_dir, &new_dentry->d_name);
1481 nfs_end_data_update(old_inode);
1482 nfs_end_data_update(new_dir);
1483 nfs_end_data_update(old_dir);
1488 if (!S_ISDIR(old_inode->i_mode))
1489 d_move(old_dentry, new_dentry);
1490 nfs_renew_times(new_dentry);
1491 nfs_set_verifier(new_dentry, nfs_save_change_attribute(new_dir));
1494 /* new dentry created? */
1502 nfs_permission(struct inode *inode, int mask, struct nameidata *nd)
1504 struct nfs_access_cache *cache = &NFS_I(inode)->cache_access;
1505 struct rpc_cred *cred;
1506 int mode = inode->i_mode;
1511 if (mask & MAY_WRITE) {
1514 * Nobody gets write access to a read-only fs.
1517 if (IS_RDONLY(inode) &&
1518 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
1523 * Nobody gets write access to an immutable file.
1526 if (IS_IMMUTABLE(inode))
1529 /* Are we checking permissions on anything other than lookup/execute? */
1530 if ((mask & MAY_EXEC) == 0) {
1531 /* We only need to check permissions on file open() and access() */
1532 if (!nd || !(nd->flags & (LOOKUP_OPEN|LOOKUP_ACCESS)))
1534 /* NFSv4 has atomic_open... */
1535 if (NFS_PROTO(inode)->version > 3 && (nd->flags & LOOKUP_OPEN))
1541 if (!NFS_PROTO(inode)->access)
1544 cred = rpcauth_lookupcred(NFS_CLIENT(inode)->cl_auth, 0);
1545 if (cache->cred == cred
1546 && time_before(jiffies, cache->jiffies + NFS_ATTRTIMEO(inode))
1547 && !(NFS_FLAGS(inode) & NFS_INO_INVALID_ATTR)) {
1548 if (!(res = cache->err)) {
1549 /* Is the mask a subset of an accepted mask? */
1550 if ((cache->mask & mask) == mask)
1553 /* ...or is it a superset of a rejected mask? */
1554 if ((cache->mask & mask) == cache->mask)
1559 res = NFS_PROTO(inode)->access(inode, cred, mask);
1560 if (!res || res == -EACCES)
1567 nfs_revalidate_inode(NFS_SERVER(inode), inode);
1568 res = vfs_permission(inode, mask);
1572 cache->jiffies = jiffies;
1574 put_rpccred(cache->cred);
1584 * version-control: t
1585 * kept-new-versions: 5