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/mount.h>
32 #include <linux/pagemap.h>
33 #include <linux/smp_lock.h>
34 #include <linux/namei.h>
35 #include <linux/vserver/xid.h>
38 #include "delegation.h"
40 #define NFS_PARANOIA 1
41 /* #define NFS_DEBUG_VERBOSE 1 */
43 static int nfs_opendir(struct inode *, struct file *);
44 static int nfs_readdir(struct file *, void *, filldir_t);
45 static struct dentry *nfs_lookup(struct inode *, struct dentry *, struct nameidata *);
46 static int nfs_create(struct inode *, struct dentry *, int, struct nameidata *);
47 static int nfs_mkdir(struct inode *, struct dentry *, int);
48 static int nfs_rmdir(struct inode *, struct dentry *);
49 static int nfs_unlink(struct inode *, struct dentry *);
50 static int nfs_symlink(struct inode *, struct dentry *, const char *);
51 static int nfs_link(struct dentry *, struct inode *, struct dentry *);
52 static int nfs_mknod(struct inode *, struct dentry *, int, dev_t);
53 static int nfs_rename(struct inode *, struct dentry *,
54 struct inode *, struct dentry *);
55 static int nfs_fsync_dir(struct file *, struct dentry *, int);
56 static loff_t nfs_llseek_dir(struct file *, loff_t, int);
58 struct file_operations nfs_dir_operations = {
59 .llseek = nfs_llseek_dir,
60 .read = generic_read_dir,
61 .readdir = nfs_readdir,
63 .release = nfs_release,
64 .fsync = nfs_fsync_dir,
67 struct inode_operations nfs_dir_inode_operations = {
72 .symlink = nfs_symlink,
77 .permission = nfs_permission,
78 .getattr = nfs_getattr,
79 .setattr = nfs_setattr,
83 struct inode_operations nfs3_dir_inode_operations = {
88 .symlink = nfs_symlink,
93 .permission = nfs_permission,
94 .getattr = nfs_getattr,
95 .setattr = nfs_setattr,
96 .listxattr = nfs3_listxattr,
97 .getxattr = nfs3_getxattr,
98 .setxattr = nfs3_setxattr,
99 .removexattr = nfs3_removexattr,
101 #endif /* CONFIG_NFS_V3 */
105 static struct dentry *nfs_atomic_lookup(struct inode *, struct dentry *, struct nameidata *);
106 struct inode_operations nfs4_dir_inode_operations = {
107 .create = nfs_create,
108 .lookup = nfs_atomic_lookup,
110 .unlink = nfs_unlink,
111 .symlink = nfs_symlink,
115 .rename = nfs_rename,
116 .permission = nfs_permission,
117 .getattr = nfs_getattr,
118 .setattr = nfs_setattr,
119 .getxattr = nfs4_getxattr,
120 .setxattr = nfs4_setxattr,
121 .listxattr = nfs4_listxattr,
124 #endif /* CONFIG_NFS_V4 */
130 nfs_opendir(struct inode *inode, struct file *filp)
135 /* Call generic open code in order to cache credentials */
137 res = nfs_open(inode, filp);
142 typedef u32 * (*decode_dirent_t)(u32 *, struct nfs_entry *, int);
146 unsigned long page_index;
149 loff_t current_index;
150 struct nfs_entry *entry;
151 decode_dirent_t decode;
154 } nfs_readdir_descriptor_t;
156 /* Now we cache directories properly, by stuffing the dirent
157 * data directly in the page cache.
159 * Inode invalidation due to refresh etc. takes care of
160 * _everything_, no sloppy entry flushing logic, no extraneous
161 * copying, network direct to page cache, the way it was meant
164 * NOTE: Dirent information verification is done always by the
165 * page-in of the RPC reply, nowhere else, this simplies
166 * things substantially.
169 int nfs_readdir_filler(nfs_readdir_descriptor_t *desc, struct page *page)
171 struct file *file = desc->file;
172 struct inode *inode = file->f_dentry->d_inode;
173 struct rpc_cred *cred = nfs_file_cred(file);
174 unsigned long timestamp;
177 dfprintk(VFS, "NFS: nfs_readdir_filler() reading cookie %Lu into page %lu.\n", (long long)desc->entry->cookie, page->index);
181 error = NFS_PROTO(inode)->readdir(file->f_dentry, cred, desc->entry->cookie, page,
182 NFS_SERVER(inode)->dtsize, desc->plus);
184 /* We requested READDIRPLUS, but the server doesn't grok it */
185 if (error == -ENOTSUPP && desc->plus) {
186 NFS_SERVER(inode)->caps &= ~NFS_CAP_READDIRPLUS;
187 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_FLAGS(inode));
193 SetPageUptodate(page);
194 spin_lock(&inode->i_lock);
195 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ATIME;
196 spin_unlock(&inode->i_lock);
197 /* Ensure consistent page alignment of the data.
198 * Note: assumes we have exclusive access to this mapping either
199 * through inode->i_mutex or some other mechanism.
201 if (page->index == 0)
202 invalidate_inode_pages2_range(inode->i_mapping, PAGE_CACHE_SIZE, -1);
208 nfs_zap_caches(inode);
214 int dir_decode(nfs_readdir_descriptor_t *desc)
217 p = desc->decode(p, desc->entry, desc->plus);
225 void dir_page_release(nfs_readdir_descriptor_t *desc)
228 page_cache_release(desc->page);
234 * Given a pointer to a buffer that has already been filled by a call
235 * to readdir, find the next entry with cookie '*desc->dir_cookie'.
237 * If the end of the buffer has been reached, return -EAGAIN, if not,
238 * return the offset within the buffer of the next entry to be
242 int find_dirent(nfs_readdir_descriptor_t *desc)
244 struct nfs_entry *entry = desc->entry;
248 while((status = dir_decode(desc)) == 0) {
249 dfprintk(VFS, "NFS: found cookie %Lu\n", (unsigned long long)entry->cookie);
250 if (entry->prev_cookie == *desc->dir_cookie)
252 if (loop_count++ > 200) {
257 dfprintk(VFS, "NFS: find_dirent() returns %d\n", status);
262 * Given a pointer to a buffer that has already been filled by a call
263 * to readdir, find the entry at offset 'desc->file->f_pos'.
265 * If the end of the buffer has been reached, return -EAGAIN, if not,
266 * return the offset within the buffer of the next entry to be
270 int find_dirent_index(nfs_readdir_descriptor_t *desc)
272 struct nfs_entry *entry = desc->entry;
277 status = dir_decode(desc);
281 dfprintk(VFS, "NFS: found cookie %Lu at index %Ld\n", (unsigned long long)entry->cookie, desc->current_index);
283 if (desc->file->f_pos == desc->current_index) {
284 *desc->dir_cookie = entry->cookie;
287 desc->current_index++;
288 if (loop_count++ > 200) {
293 dfprintk(VFS, "NFS: find_dirent_index() returns %d\n", status);
298 * Find the given page, and call find_dirent() or find_dirent_index in
299 * order to try to return the next entry.
302 int find_dirent_page(nfs_readdir_descriptor_t *desc)
304 struct inode *inode = desc->file->f_dentry->d_inode;
308 dfprintk(VFS, "NFS: find_dirent_page() searching directory page %ld\n", desc->page_index);
310 page = read_cache_page(inode->i_mapping, desc->page_index,
311 (filler_t *)nfs_readdir_filler, desc);
313 status = PTR_ERR(page);
316 if (!PageUptodate(page))
319 /* NOTE: Someone else may have changed the READDIRPLUS flag */
321 desc->ptr = kmap(page); /* matching kunmap in nfs_do_filldir */
322 if (*desc->dir_cookie != 0)
323 status = find_dirent(desc);
325 status = find_dirent_index(desc);
327 dir_page_release(desc);
329 dfprintk(VFS, "NFS: find_dirent_page() returns %d\n", status);
332 page_cache_release(page);
337 * Recurse through the page cache pages, and return a
338 * filled nfs_entry structure of the next directory entry if possible.
340 * The target for the search is '*desc->dir_cookie' if non-0,
341 * 'desc->file->f_pos' otherwise
344 int readdir_search_pagecache(nfs_readdir_descriptor_t *desc)
349 /* Always search-by-index from the beginning of the cache */
350 if (*desc->dir_cookie == 0) {
351 dfprintk(VFS, "NFS: readdir_search_pagecache() searching for offset %Ld\n", (long long)desc->file->f_pos);
352 desc->page_index = 0;
353 desc->entry->cookie = desc->entry->prev_cookie = 0;
354 desc->entry->eof = 0;
355 desc->current_index = 0;
357 dfprintk(VFS, "NFS: readdir_search_pagecache() searching for cookie %Lu\n", (unsigned long long)*desc->dir_cookie);
360 res = find_dirent_page(desc);
363 /* Align to beginning of next page */
365 if (loop_count++ > 200) {
370 dfprintk(VFS, "NFS: readdir_search_pagecache() returned %d\n", res);
374 static inline unsigned int dt_type(struct inode *inode)
376 return (inode->i_mode >> 12) & 15;
379 static struct dentry *nfs_readdir_lookup(nfs_readdir_descriptor_t *desc);
382 * Once we've found the start of the dirent within a page: fill 'er up...
385 int nfs_do_filldir(nfs_readdir_descriptor_t *desc, void *dirent,
388 struct file *file = desc->file;
389 struct nfs_entry *entry = desc->entry;
390 struct dentry *dentry = NULL;
391 unsigned long fileid;
395 dfprintk(VFS, "NFS: nfs_do_filldir() filling starting @ cookie %Lu\n", (long long)entry->cookie);
398 unsigned d_type = DT_UNKNOWN;
399 /* Note: entry->prev_cookie contains the cookie for
400 * retrieving the current dirent on the server */
401 fileid = nfs_fileid_to_ino_t(entry->ino);
403 /* Get a dentry if we have one */
406 dentry = nfs_readdir_lookup(desc);
408 /* Use readdirplus info */
409 if (dentry != NULL && dentry->d_inode != NULL) {
410 d_type = dt_type(dentry->d_inode);
411 fileid = dentry->d_inode->i_ino;
414 res = filldir(dirent, entry->name, entry->len,
415 file->f_pos, fileid, d_type);
419 *desc->dir_cookie = entry->cookie;
420 if (dir_decode(desc) != 0) {
424 if (loop_count++ > 200) {
429 dir_page_release(desc);
432 dfprintk(VFS, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n", (unsigned long long)*desc->dir_cookie, res);
437 * If we cannot find a cookie in our cache, we suspect that this is
438 * because it points to a deleted file, so we ask the server to return
439 * whatever it thinks is the next entry. We then feed this to filldir.
440 * If all goes well, we should then be able to find our way round the
441 * cache on the next call to readdir_search_pagecache();
443 * NOTE: we cannot add the anonymous page to the pagecache because
444 * the data it contains might not be page aligned. Besides,
445 * we should already have a complete representation of the
446 * directory in the page cache by the time we get here.
449 int uncached_readdir(nfs_readdir_descriptor_t *desc, void *dirent,
452 struct file *file = desc->file;
453 struct inode *inode = file->f_dentry->d_inode;
454 struct rpc_cred *cred = nfs_file_cred(file);
455 struct page *page = NULL;
458 dfprintk(VFS, "NFS: uncached_readdir() searching for cookie %Lu\n", (unsigned long long)*desc->dir_cookie);
460 page = alloc_page(GFP_HIGHUSER);
465 desc->error = NFS_PROTO(inode)->readdir(file->f_dentry, cred, *desc->dir_cookie,
467 NFS_SERVER(inode)->dtsize,
469 spin_lock(&inode->i_lock);
470 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ATIME;
471 spin_unlock(&inode->i_lock);
473 desc->ptr = kmap(page); /* matching kunmap in nfs_do_filldir */
474 if (desc->error >= 0) {
475 if ((status = dir_decode(desc)) == 0)
476 desc->entry->prev_cookie = *desc->dir_cookie;
482 status = nfs_do_filldir(desc, dirent, filldir);
484 /* Reset read descriptor so it searches the page cache from
485 * the start upon the next call to readdir_search_pagecache() */
486 desc->page_index = 0;
487 desc->entry->cookie = desc->entry->prev_cookie = 0;
488 desc->entry->eof = 0;
490 dfprintk(VFS, "NFS: uncached_readdir() returns %d\n", status);
493 dir_page_release(desc);
497 /* The file offset position represents the dirent entry number. A
498 last cookie cache takes care of the common case of reading the
501 static int nfs_readdir(struct file *filp, void *dirent, filldir_t filldir)
503 struct dentry *dentry = filp->f_dentry;
504 struct inode *inode = dentry->d_inode;
505 nfs_readdir_descriptor_t my_desc,
507 struct nfs_entry my_entry;
509 struct nfs_fattr fattr;
514 res = nfs_revalidate_inode(NFS_SERVER(inode), inode);
521 * filp->f_pos points to the dirent entry number.
522 * *desc->dir_cookie has the cookie for the next entry. We have
523 * to either find the entry with the appropriate number or
524 * revalidate the cookie.
526 memset(desc, 0, sizeof(*desc));
529 desc->dir_cookie = &((struct nfs_open_context *)filp->private_data)->dir_cookie;
530 desc->decode = NFS_PROTO(inode)->decode_dirent;
531 desc->plus = NFS_USE_READDIRPLUS(inode);
533 my_entry.cookie = my_entry.prev_cookie = 0;
536 my_entry.fattr = &fattr;
537 nfs_fattr_init(&fattr);
538 desc->entry = &my_entry;
540 while(!desc->entry->eof) {
541 res = readdir_search_pagecache(desc);
543 if (res == -EBADCOOKIE) {
544 /* This means either end of directory */
545 if (*desc->dir_cookie && desc->entry->cookie != *desc->dir_cookie) {
546 /* Or that the server has 'lost' a cookie */
547 res = uncached_readdir(desc, dirent, filldir);
554 if (res == -ETOOSMALL && desc->plus) {
555 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_FLAGS(inode));
556 nfs_zap_caches(inode);
558 desc->entry->eof = 0;
564 res = nfs_do_filldir(desc, dirent, filldir);
576 loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int origin)
578 mutex_lock(&filp->f_dentry->d_inode->i_mutex);
581 offset += filp->f_pos;
589 if (offset != filp->f_pos) {
590 filp->f_pos = offset;
591 ((struct nfs_open_context *)filp->private_data)->dir_cookie = 0;
594 mutex_unlock(&filp->f_dentry->d_inode->i_mutex);
599 * All directory operations under NFS are synchronous, so fsync()
600 * is a dummy operation.
602 int nfs_fsync_dir(struct file *filp, struct dentry *dentry, int datasync)
608 * A check for whether or not the parent directory has changed.
609 * In the case it has, we assume that the dentries are untrustworthy
610 * and may need to be looked up again.
612 static inline int nfs_check_verifier(struct inode *dir, struct dentry *dentry)
616 if ((NFS_I(dir)->cache_validity & NFS_INO_INVALID_ATTR) != 0
617 || nfs_attribute_timeout(dir))
619 return nfs_verify_change_attribute(dir, (unsigned long)dentry->d_fsdata);
622 static inline void nfs_set_verifier(struct dentry * dentry, unsigned long verf)
624 dentry->d_fsdata = (void *)verf;
628 * Whenever an NFS operation succeeds, we know that the dentry
629 * is valid, so we update the revalidation timestamp.
631 static inline void nfs_renew_times(struct dentry * dentry)
633 dentry->d_time = jiffies;
637 * Return the intent data that applies to this particular path component
639 * Note that the current set of intents only apply to the very last
640 * component of the path.
641 * We check for this using LOOKUP_CONTINUE and LOOKUP_PARENT.
643 static inline unsigned int nfs_lookup_check_intent(struct nameidata *nd, unsigned int mask)
645 if (nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))
647 return nd->flags & mask;
651 * Inode and filehandle revalidation for lookups.
653 * We force revalidation in the cases where the VFS sets LOOKUP_REVAL,
654 * or if the intent information indicates that we're about to open this
655 * particular file and the "nocto" mount flag is not set.
659 int nfs_lookup_verify_inode(struct inode *inode, struct nameidata *nd)
661 struct nfs_server *server = NFS_SERVER(inode);
664 /* VFS wants an on-the-wire revalidation */
665 if (nd->flags & LOOKUP_REVAL)
667 /* This is an open(2) */
668 if (nfs_lookup_check_intent(nd, LOOKUP_OPEN) != 0 &&
669 !(server->flags & NFS_MOUNT_NOCTO))
672 return nfs_revalidate_inode(server, inode);
674 return __nfs_revalidate_inode(server, inode);
678 * We judge how long we want to trust negative
679 * dentries by looking at the parent inode mtime.
681 * If parent mtime has changed, we revalidate, else we wait for a
682 * period corresponding to the parent's attribute cache timeout value.
685 int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry,
686 struct nameidata *nd)
688 /* Don't revalidate a negative dentry if we're creating a new file */
689 if (nd != NULL && nfs_lookup_check_intent(nd, LOOKUP_CREATE) != 0)
691 return !nfs_check_verifier(dir, dentry);
695 * This is called every time the dcache has a lookup hit,
696 * and we should check whether we can really trust that
699 * NOTE! The hit can be a negative hit too, don't assume
702 * If the parent directory is seen to have changed, we throw out the
703 * cached dentry and do a new lookup.
705 static int nfs_lookup_revalidate(struct dentry * dentry, struct nameidata *nd)
709 struct dentry *parent;
711 struct nfs_fh fhandle;
712 struct nfs_fattr fattr;
713 unsigned long verifier;
715 parent = dget_parent(dentry);
717 dir = parent->d_inode;
718 inode = dentry->d_inode;
721 if (nfs_neg_need_reval(dir, dentry, nd))
726 if (is_bad_inode(inode)) {
727 dfprintk(VFS, "nfs_lookup_validate: %s/%s has dud inode\n",
728 dentry->d_parent->d_name.name, dentry->d_name.name);
732 /* Revalidate parent directory attribute cache */
733 if (nfs_revalidate_inode(NFS_SERVER(dir), dir) < 0)
736 /* Force a full look up iff the parent directory has changed */
737 if (nfs_check_verifier(dir, dentry)) {
738 if (nfs_lookup_verify_inode(inode, nd))
743 if (NFS_STALE(inode))
746 verifier = nfs_save_change_attribute(dir);
747 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, &fhandle, &fattr);
750 if (nfs_compare_fh(NFS_FH(inode), &fhandle))
752 if ((error = nfs_refresh_inode(inode, &fattr)) != 0)
755 nfs_renew_times(dentry);
756 nfs_set_verifier(dentry, verifier);
765 if (inode && S_ISDIR(inode->i_mode)) {
766 /* Purge readdir caches. */
767 nfs_zap_caches(inode);
768 /* If we have submounts, don't unhash ! */
769 if (have_submounts(dentry))
771 shrink_dcache_parent(dentry);
780 * This is called from dput() when d_count is going to 0.
782 static int nfs_dentry_delete(struct dentry *dentry)
784 dfprintk(VFS, "NFS: dentry_delete(%s/%s, %x)\n",
785 dentry->d_parent->d_name.name, dentry->d_name.name,
788 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
789 /* Unhash it, so that ->d_iput() would be called */
792 if (!(dentry->d_sb->s_flags & MS_ACTIVE)) {
793 /* Unhash it, so that ancestors of killed async unlink
794 * files will be cleaned up during umount */
802 * Called when the dentry loses inode.
803 * We use it to clean up silly-renamed files.
805 static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode)
807 nfs_inode_return_delegation(inode);
808 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
811 nfs_complete_unlink(dentry);
814 /* When creating a negative dentry, we want to renew d_time */
815 nfs_renew_times(dentry);
819 struct dentry_operations nfs_dentry_operations = {
820 .d_revalidate = nfs_lookup_revalidate,
821 .d_delete = nfs_dentry_delete,
822 .d_iput = nfs_dentry_iput,
826 * Use intent information to check whether or not we're going to do
827 * an O_EXCL create using this path component.
830 int nfs_is_exclusive_create(struct inode *dir, struct nameidata *nd)
832 if (NFS_PROTO(dir)->version == 2)
834 if (nd == NULL || nfs_lookup_check_intent(nd, LOOKUP_CREATE) == 0)
836 return (nd->intent.open.flags & O_EXCL) != 0;
839 static struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
842 struct inode *inode = NULL;
844 struct nfs_fh fhandle;
845 struct nfs_fattr fattr;
847 dfprintk(VFS, "NFS: lookup(%s/%s)\n",
848 dentry->d_parent->d_name.name, dentry->d_name.name);
850 res = ERR_PTR(-ENAMETOOLONG);
851 if (dentry->d_name.len > NFS_SERVER(dir)->namelen)
854 res = ERR_PTR(-ENOMEM);
855 dentry->d_op = NFS_PROTO(dir)->dentry_ops;
860 * If we're doing an exclusive create, optimize away the lookup
861 * but don't hash the dentry.
863 if (nfs_is_exclusive_create(dir, nd)) {
864 d_instantiate(dentry, NULL);
869 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, &fhandle, &fattr);
870 if (error == -ENOENT)
873 res = ERR_PTR(error);
876 res = ERR_PTR(-EACCES);
877 inode = nfs_fhget(dentry->d_sb, &fhandle, &fattr);
880 vx_propagate_xid(nd, inode);
882 res = d_add_unique(dentry, inode);
885 nfs_renew_times(dentry);
886 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
894 static int nfs_open_revalidate(struct dentry *, struct nameidata *);
896 struct dentry_operations nfs4_dentry_operations = {
897 .d_revalidate = nfs_open_revalidate,
898 .d_delete = nfs_dentry_delete,
899 .d_iput = nfs_dentry_iput,
903 * Use intent information to determine whether we need to substitute
904 * the NFSv4-style stateful OPEN for the LOOKUP call
906 static int is_atomic_open(struct inode *dir, struct nameidata *nd)
908 if (nd == NULL || nfs_lookup_check_intent(nd, LOOKUP_OPEN) == 0)
910 /* NFS does not (yet) have a stateful open for directories */
911 if (nd->flags & LOOKUP_DIRECTORY)
913 /* Are we trying to write to a read only partition? */
914 if ((IS_RDONLY(dir) || MNT_IS_RDONLY(nd->mnt)) &&
915 (nd->intent.open.flags & (O_CREAT|O_TRUNC|FMODE_WRITE)))
920 static struct dentry *nfs_atomic_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd)
922 struct dentry *res = NULL;
925 /* Check that we are indeed trying to open this file */
926 if (!is_atomic_open(dir, nd))
929 if (dentry->d_name.len > NFS_SERVER(dir)->namelen) {
930 res = ERR_PTR(-ENAMETOOLONG);
933 dentry->d_op = NFS_PROTO(dir)->dentry_ops;
935 /* Let vfs_create() deal with O_EXCL */
936 if (nd->intent.open.flags & O_EXCL) {
941 /* Open the file on the server */
943 /* Revalidate parent directory attribute cache */
944 error = nfs_revalidate_inode(NFS_SERVER(dir), dir);
946 res = ERR_PTR(error);
951 if (nd->intent.open.flags & O_CREAT) {
952 nfs_begin_data_update(dir);
953 res = nfs4_atomic_open(dir, dentry, nd);
954 nfs_end_data_update(dir);
956 res = nfs4_atomic_open(dir, dentry, nd);
959 error = PTR_ERR(res);
961 /* Make a negative dentry */
965 /* This turned out not to be a regular file */
970 if (!(nd->intent.open.flags & O_NOFOLLOW))
976 } else if (res != NULL)
978 nfs_renew_times(dentry);
979 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
983 return nfs_lookup(dir, dentry, nd);
986 static int nfs_open_revalidate(struct dentry *dentry, struct nameidata *nd)
988 struct dentry *parent = NULL;
989 struct inode *inode = dentry->d_inode;
991 unsigned long verifier;
992 int openflags, ret = 0;
994 parent = dget_parent(dentry);
995 dir = parent->d_inode;
996 if (!is_atomic_open(dir, nd))
998 /* We can't create new files in nfs_open_revalidate(), so we
999 * optimize away revalidation of negative dentries.
1003 /* NFS only supports OPEN on regular files */
1004 if (!S_ISREG(inode->i_mode))
1006 openflags = nd->intent.open.flags;
1007 /* We cannot do exclusive creation on a positive dentry */
1008 if ((openflags & (O_CREAT|O_EXCL)) == (O_CREAT|O_EXCL))
1010 /* We can't create new files, or truncate existing ones here */
1011 openflags &= ~(O_CREAT|O_TRUNC);
1014 * Note: we're not holding inode->i_mutex and so may be racing with
1015 * operations that change the directory. We therefore save the
1016 * change attribute *before* we do the RPC call.
1019 verifier = nfs_save_change_attribute(dir);
1020 ret = nfs4_open_revalidate(dir, dentry, openflags, nd);
1022 nfs_set_verifier(dentry, verifier);
1031 if (inode != NULL && nfs_have_delegation(inode, FMODE_READ))
1033 return nfs_lookup_revalidate(dentry, nd);
1035 #endif /* CONFIG_NFSV4 */
1037 static struct dentry *nfs_readdir_lookup(nfs_readdir_descriptor_t *desc)
1039 struct dentry *parent = desc->file->f_dentry;
1040 struct inode *dir = parent->d_inode;
1041 struct nfs_entry *entry = desc->entry;
1042 struct dentry *dentry, *alias;
1043 struct qstr name = {
1044 .name = entry->name,
1047 struct inode *inode;
1051 if (name.name[0] == '.' && name.name[1] == '.')
1052 return dget_parent(parent);
1055 if (name.name[0] == '.')
1056 return dget(parent);
1058 name.hash = full_name_hash(name.name, name.len);
1059 dentry = d_lookup(parent, &name);
1062 if (!desc->plus || !(entry->fattr->valid & NFS_ATTR_FATTR))
1064 /* Note: caller is already holding the dir->i_mutex! */
1065 dentry = d_alloc(parent, &name);
1068 dentry->d_op = NFS_PROTO(dir)->dentry_ops;
1069 inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr);
1074 alias = d_add_unique(dentry, inode);
1075 if (alias != NULL) {
1079 nfs_renew_times(dentry);
1080 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1085 * Code common to create, mkdir, and mknod.
1087 int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle,
1088 struct nfs_fattr *fattr)
1090 struct inode *inode;
1091 int error = -EACCES;
1093 /* We may have been initialized further down */
1094 if (dentry->d_inode)
1096 if (fhandle->size == 0) {
1097 struct inode *dir = dentry->d_parent->d_inode;
1098 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr);
1102 if (!(fattr->valid & NFS_ATTR_FATTR)) {
1103 struct nfs_server *server = NFS_SB(dentry->d_sb);
1104 error = server->rpc_ops->getattr(server, fhandle, fattr);
1109 inode = nfs_fhget(dentry->d_sb, fhandle, fattr);
1112 d_instantiate(dentry, inode);
1113 if (d_unhashed(dentry))
1122 * Following a failed create operation, we drop the dentry rather
1123 * than retain a negative dentry. This avoids a problem in the event
1124 * that the operation succeeded on the server, but an error in the
1125 * reply path made it appear to have failed.
1127 static int nfs_create(struct inode *dir, struct dentry *dentry, int mode,
1128 struct nameidata *nd)
1134 dfprintk(VFS, "NFS: create(%s/%ld, %s\n", dir->i_sb->s_id,
1135 dir->i_ino, dentry->d_name.name);
1137 attr.ia_mode = mode;
1138 attr.ia_valid = ATTR_MODE;
1140 if (nd && (nd->flags & LOOKUP_CREATE))
1141 open_flags = nd->intent.open.flags;
1144 nfs_begin_data_update(dir);
1145 error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags, nd);
1146 nfs_end_data_update(dir);
1149 nfs_renew_times(dentry);
1150 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1160 * See comments for nfs_proc_create regarding failed operations.
1163 nfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t rdev)
1168 dfprintk(VFS, "NFS: mknod(%s/%ld, %s\n", dir->i_sb->s_id,
1169 dir->i_ino, dentry->d_name.name);
1171 if (!new_valid_dev(rdev))
1174 attr.ia_mode = mode;
1175 attr.ia_valid = ATTR_MODE;
1178 nfs_begin_data_update(dir);
1179 status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev);
1180 nfs_end_data_update(dir);
1183 nfs_renew_times(dentry);
1184 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1194 * See comments for nfs_proc_create regarding failed operations.
1196 static int nfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1201 dfprintk(VFS, "NFS: mkdir(%s/%ld, %s\n", dir->i_sb->s_id,
1202 dir->i_ino, dentry->d_name.name);
1204 attr.ia_valid = ATTR_MODE;
1205 attr.ia_mode = mode | S_IFDIR;
1208 nfs_begin_data_update(dir);
1209 error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr);
1210 nfs_end_data_update(dir);
1213 nfs_renew_times(dentry);
1214 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1223 static int nfs_rmdir(struct inode *dir, struct dentry *dentry)
1227 dfprintk(VFS, "NFS: rmdir(%s/%ld, %s\n", dir->i_sb->s_id,
1228 dir->i_ino, dentry->d_name.name);
1231 nfs_begin_data_update(dir);
1232 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
1233 /* Ensure the VFS deletes this inode */
1234 if (error == 0 && dentry->d_inode != NULL)
1235 dentry->d_inode->i_nlink = 0;
1236 nfs_end_data_update(dir);
1242 static int nfs_sillyrename(struct inode *dir, struct dentry *dentry)
1244 static unsigned int sillycounter;
1245 const int i_inosize = sizeof(dir->i_ino)*2;
1246 const int countersize = sizeof(sillycounter)*2;
1247 const int slen = sizeof(".nfs") + i_inosize + countersize - 1;
1250 struct dentry *sdentry;
1253 dfprintk(VFS, "NFS: silly-rename(%s/%s, ct=%d)\n",
1254 dentry->d_parent->d_name.name, dentry->d_name.name,
1255 atomic_read(&dentry->d_count));
1258 if (!dentry->d_inode)
1259 printk("NFS: silly-renaming %s/%s, negative dentry??\n",
1260 dentry->d_parent->d_name.name, dentry->d_name.name);
1263 * We don't allow a dentry to be silly-renamed twice.
1266 if (dentry->d_flags & DCACHE_NFSFS_RENAMED)
1269 sprintf(silly, ".nfs%*.*lx",
1270 i_inosize, i_inosize, dentry->d_inode->i_ino);
1272 /* Return delegation in anticipation of the rename */
1273 nfs_inode_return_delegation(dentry->d_inode);
1277 char *suffix = silly + slen - countersize;
1281 sprintf(suffix, "%*.*x", countersize, countersize, sillycounter);
1283 dfprintk(VFS, "trying to rename %s to %s\n",
1284 dentry->d_name.name, silly);
1286 sdentry = lookup_one_len(silly, dentry->d_parent, slen);
1288 * N.B. Better to return EBUSY here ... it could be
1289 * dangerous to delete the file while it's in use.
1291 if (IS_ERR(sdentry))
1293 } while(sdentry->d_inode != NULL); /* need negative lookup */
1295 qsilly.name = silly;
1296 qsilly.len = strlen(silly);
1297 nfs_begin_data_update(dir);
1298 if (dentry->d_inode) {
1299 nfs_begin_data_update(dentry->d_inode);
1300 error = NFS_PROTO(dir)->rename(dir, &dentry->d_name,
1302 nfs_mark_for_revalidate(dentry->d_inode);
1303 nfs_end_data_update(dentry->d_inode);
1305 error = NFS_PROTO(dir)->rename(dir, &dentry->d_name,
1307 nfs_end_data_update(dir);
1309 nfs_renew_times(dentry);
1310 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1311 d_move(dentry, sdentry);
1312 error = nfs_async_unlink(dentry);
1313 /* If we return 0 we don't unlink */
1321 * Remove a file after making sure there are no pending writes,
1322 * and after checking that the file has only one user.
1324 * We invalidate the attribute cache and free the inode prior to the operation
1325 * to avoid possible races if the server reuses the inode.
1327 static int nfs_safe_remove(struct dentry *dentry)
1329 struct inode *dir = dentry->d_parent->d_inode;
1330 struct inode *inode = dentry->d_inode;
1333 dfprintk(VFS, "NFS: safe_remove(%s/%s)\n",
1334 dentry->d_parent->d_name.name, dentry->d_name.name);
1336 /* If the dentry was sillyrenamed, we simply call d_delete() */
1337 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1342 nfs_begin_data_update(dir);
1343 if (inode != NULL) {
1344 nfs_inode_return_delegation(inode);
1345 nfs_begin_data_update(inode);
1346 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1347 /* The VFS may want to delete this inode */
1350 nfs_mark_for_revalidate(inode);
1351 nfs_end_data_update(inode);
1353 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1354 nfs_end_data_update(dir);
1359 /* We do silly rename. In case sillyrename() returns -EBUSY, the inode
1360 * belongs to an active ".nfs..." file and we return -EBUSY.
1362 * If sillyrename() returns 0, we do nothing, otherwise we unlink.
1364 static int nfs_unlink(struct inode *dir, struct dentry *dentry)
1367 int need_rehash = 0;
1369 dfprintk(VFS, "NFS: unlink(%s/%ld, %s)\n", dir->i_sb->s_id,
1370 dir->i_ino, dentry->d_name.name);
1373 spin_lock(&dcache_lock);
1374 spin_lock(&dentry->d_lock);
1375 if (atomic_read(&dentry->d_count) > 1) {
1376 spin_unlock(&dentry->d_lock);
1377 spin_unlock(&dcache_lock);
1378 error = nfs_sillyrename(dir, dentry);
1382 if (!d_unhashed(dentry)) {
1386 spin_unlock(&dentry->d_lock);
1387 spin_unlock(&dcache_lock);
1388 error = nfs_safe_remove(dentry);
1390 nfs_renew_times(dentry);
1391 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1392 } else if (need_rehash)
1399 nfs_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1402 struct nfs_fattr sym_attr;
1403 struct nfs_fh sym_fh;
1404 struct qstr qsymname;
1407 dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s)\n", dir->i_sb->s_id,
1408 dir->i_ino, dentry->d_name.name, symname);
1411 if (dentry->d_inode)
1412 printk("nfs_proc_symlink: %s/%s not negative!\n",
1413 dentry->d_parent->d_name.name, dentry->d_name.name);
1416 * Fill in the sattr for the call.
1417 * Note: SunOS 4.1.2 crashes if the mode isn't initialized!
1419 attr.ia_valid = ATTR_MODE;
1420 attr.ia_mode = S_IFLNK | S_IRWXUGO;
1422 qsymname.name = symname;
1423 qsymname.len = strlen(symname);
1426 nfs_begin_data_update(dir);
1427 error = NFS_PROTO(dir)->symlink(dir, &dentry->d_name, &qsymname,
1428 &attr, &sym_fh, &sym_attr);
1429 nfs_end_data_update(dir);
1431 error = nfs_instantiate(dentry, &sym_fh, &sym_attr);
1433 if (error == -EEXIST)
1434 printk("nfs_proc_symlink: %s/%s already exists??\n",
1435 dentry->d_parent->d_name.name, dentry->d_name.name);
1443 nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1445 struct inode *inode = old_dentry->d_inode;
1448 dfprintk(VFS, "NFS: link(%s/%s -> %s/%s)\n",
1449 old_dentry->d_parent->d_name.name, old_dentry->d_name.name,
1450 dentry->d_parent->d_name.name, dentry->d_name.name);
1453 nfs_begin_data_update(dir);
1454 nfs_begin_data_update(inode);
1455 error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name);
1457 atomic_inc(&inode->i_count);
1458 d_instantiate(dentry, inode);
1460 nfs_end_data_update(inode);
1461 nfs_end_data_update(dir);
1468 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a
1469 * different file handle for the same inode after a rename (e.g. when
1470 * moving to a different directory). A fail-safe method to do so would
1471 * be to look up old_dir/old_name, create a link to new_dir/new_name and
1472 * rename the old file using the sillyrename stuff. This way, the original
1473 * file in old_dir will go away when the last process iput()s the inode.
1477 * It actually works quite well. One needs to have the possibility for
1478 * at least one ".nfs..." file in each directory the file ever gets
1479 * moved or linked to which happens automagically with the new
1480 * implementation that only depends on the dcache stuff instead of
1481 * using the inode layer
1483 * Unfortunately, things are a little more complicated than indicated
1484 * above. For a cross-directory move, we want to make sure we can get
1485 * rid of the old inode after the operation. This means there must be
1486 * no pending writes (if it's a file), and the use count must be 1.
1487 * If these conditions are met, we can drop the dentries before doing
1490 static int nfs_rename(struct inode *old_dir, struct dentry *old_dentry,
1491 struct inode *new_dir, struct dentry *new_dentry)
1493 struct inode *old_inode = old_dentry->d_inode;
1494 struct inode *new_inode = new_dentry->d_inode;
1495 struct dentry *dentry = NULL, *rehash = NULL;
1499 * To prevent any new references to the target during the rename,
1500 * we unhash the dentry and free the inode in advance.
1503 if (!d_unhashed(new_dentry)) {
1505 rehash = new_dentry;
1508 dfprintk(VFS, "NFS: rename(%s/%s -> %s/%s, ct=%d)\n",
1509 old_dentry->d_parent->d_name.name, old_dentry->d_name.name,
1510 new_dentry->d_parent->d_name.name, new_dentry->d_name.name,
1511 atomic_read(&new_dentry->d_count));
1514 * First check whether the target is busy ... we can't
1515 * safely do _any_ rename if the target is in use.
1517 * For files, make a copy of the dentry and then do a
1518 * silly-rename. If the silly-rename succeeds, the
1519 * copied dentry is hashed and becomes the new target.
1523 if (S_ISDIR(new_inode->i_mode)) {
1525 if (!S_ISDIR(old_inode->i_mode))
1527 } else if (atomic_read(&new_dentry->d_count) > 2) {
1529 /* copy the target dentry's name */
1530 dentry = d_alloc(new_dentry->d_parent,
1531 &new_dentry->d_name);
1535 /* silly-rename the existing target ... */
1536 err = nfs_sillyrename(new_dir, new_dentry);
1538 new_dentry = rehash = dentry;
1540 /* instantiate the replacement target */
1541 d_instantiate(new_dentry, NULL);
1542 } else if (atomic_read(&new_dentry->d_count) > 1) {
1543 /* dentry still busy? */
1545 printk("nfs_rename: target %s/%s busy, d_count=%d\n",
1546 new_dentry->d_parent->d_name.name,
1547 new_dentry->d_name.name,
1548 atomic_read(&new_dentry->d_count));
1553 new_inode->i_nlink--;
1557 * ... prune child dentries and writebacks if needed.
1559 if (atomic_read(&old_dentry->d_count) > 1) {
1560 nfs_wb_all(old_inode);
1561 shrink_dcache_parent(old_dentry);
1563 nfs_inode_return_delegation(old_inode);
1565 if (new_inode != NULL) {
1566 nfs_inode_return_delegation(new_inode);
1567 d_delete(new_dentry);
1570 nfs_begin_data_update(old_dir);
1571 nfs_begin_data_update(new_dir);
1572 nfs_begin_data_update(old_inode);
1573 error = NFS_PROTO(old_dir)->rename(old_dir, &old_dentry->d_name,
1574 new_dir, &new_dentry->d_name);
1575 nfs_mark_for_revalidate(old_inode);
1576 nfs_end_data_update(old_inode);
1577 nfs_end_data_update(new_dir);
1578 nfs_end_data_update(old_dir);
1583 if (!S_ISDIR(old_inode->i_mode))
1584 d_move(old_dentry, new_dentry);
1585 nfs_renew_times(new_dentry);
1586 nfs_set_verifier(new_dentry, nfs_save_change_attribute(new_dir));
1589 /* new dentry created? */
1596 int nfs_access_get_cached(struct inode *inode, struct rpc_cred *cred, struct nfs_access_entry *res)
1598 struct nfs_inode *nfsi = NFS_I(inode);
1599 struct nfs_access_entry *cache = &nfsi->cache_access;
1601 if (cache->cred != cred
1602 || time_after(jiffies, cache->jiffies + NFS_ATTRTIMEO(inode))
1603 || (nfsi->cache_validity & NFS_INO_INVALID_ACCESS))
1605 memcpy(res, cache, sizeof(*res));
1609 void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set)
1611 struct nfs_inode *nfsi = NFS_I(inode);
1612 struct nfs_access_entry *cache = &nfsi->cache_access;
1614 if (cache->cred != set->cred) {
1616 put_rpccred(cache->cred);
1617 cache->cred = get_rpccred(set->cred);
1619 /* FIXME: replace current access_cache BKL reliance with inode->i_lock */
1620 spin_lock(&inode->i_lock);
1621 nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS;
1622 spin_unlock(&inode->i_lock);
1623 cache->jiffies = set->jiffies;
1624 cache->mask = set->mask;
1627 static int nfs_do_access(struct inode *inode, struct rpc_cred *cred, int mask)
1629 struct nfs_access_entry cache;
1632 status = nfs_access_get_cached(inode, cred, &cache);
1636 /* Be clever: ask server to check for all possible rights */
1637 cache.mask = MAY_EXEC | MAY_WRITE | MAY_READ;
1639 cache.jiffies = jiffies;
1640 status = NFS_PROTO(inode)->access(inode, &cache);
1643 nfs_access_add_cache(inode, &cache);
1645 if ((cache.mask & mask) == mask)
1650 int nfs_permission(struct inode *inode, int mask, struct nameidata *nd)
1652 struct rpc_cred *cred;
1657 /* Is this sys_access() ? */
1658 if (nd != NULL && (nd->flags & LOOKUP_ACCESS))
1661 switch (inode->i_mode & S_IFMT) {
1665 /* NFSv4 has atomic_open... */
1666 if (nfs_server_capable(inode, NFS_CAP_ATOMIC_OPEN)
1668 && (nd->flags & LOOKUP_OPEN))
1673 * Optimize away all write operations, since the server
1674 * will check permissions when we perform the op.
1676 if ((mask & MAY_WRITE) && !(mask & MAY_READ))
1683 if (!NFS_PROTO(inode)->access)
1686 cred = rpcauth_lookupcred(NFS_CLIENT(inode)->cl_auth, 0);
1687 if (!IS_ERR(cred)) {
1688 res = nfs_do_access(inode, cred, mask);
1691 res = PTR_ERR(cred);
1696 res = nfs_revalidate_inode(NFS_SERVER(inode), inode);
1698 res = generic_permission(inode, mask, NULL);
1705 * version-control: t
1706 * kept-new-versions: 5