4 * Copyright (C) 1992 Rick Sladkey
6 * nfs directory handling functions
8 * 10 Apr 1996 Added silly rename for unlink --okir
9 * 28 Sep 1996 Improved directory cache --okir
10 * 23 Aug 1997 Claus Heine claus@momo.math.rwth-aachen.de
11 * Re-implemented silly rename for unlink, newly implemented
12 * silly rename for nfs_rename() following the suggestions
13 * of Olaf Kirch (okir) found in this file.
14 * Following Linus comments on my original hack, this version
15 * depends only on the dcache stuff and doesn't touch the inode
16 * layer (iput() and friends).
17 * 6 Jun 1999 Cache readdir lookups in the page cache. -DaveM
20 #include <linux/time.h>
21 #include <linux/errno.h>
22 #include <linux/stat.h>
23 #include <linux/fcntl.h>
24 #include <linux/string.h>
25 #include <linux/kernel.h>
26 #include <linux/slab.h>
28 #include <linux/sunrpc/clnt.h>
29 #include <linux/nfs_fs.h>
30 #include <linux/nfs_mount.h>
31 #include <linux/pagemap.h>
32 #include <linux/smp_lock.h>
33 #include <linux/namei.h>
36 #include "delegation.h"
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);
54 static loff_t nfs_llseek_dir(struct file *, loff_t, int);
56 struct file_operations nfs_dir_operations = {
57 .llseek = nfs_llseek_dir,
58 .read = generic_read_dir,
59 .readdir = nfs_readdir,
61 .release = nfs_release,
62 .fsync = nfs_fsync_dir,
65 struct inode_operations nfs_dir_inode_operations = {
70 .symlink = nfs_symlink,
75 .permission = nfs_permission,
76 .getattr = nfs_getattr,
77 .setattr = nfs_setattr,
81 struct inode_operations nfs3_dir_inode_operations = {
86 .symlink = nfs_symlink,
91 .permission = nfs_permission,
92 .getattr = nfs_getattr,
93 .setattr = nfs_setattr,
94 .listxattr = nfs3_listxattr,
95 .getxattr = nfs3_getxattr,
96 .setxattr = nfs3_setxattr,
97 .removexattr = nfs3_removexattr,
99 #endif /* CONFIG_NFS_V3 */
103 static struct dentry *nfs_atomic_lookup(struct inode *, struct dentry *, struct nameidata *);
104 struct inode_operations nfs4_dir_inode_operations = {
105 .create = nfs_create,
106 .lookup = nfs_atomic_lookup,
108 .unlink = nfs_unlink,
109 .symlink = nfs_symlink,
113 .rename = nfs_rename,
114 .permission = nfs_permission,
115 .getattr = nfs_getattr,
116 .setattr = nfs_setattr,
117 .getxattr = nfs4_getxattr,
118 .setxattr = nfs4_setxattr,
119 .listxattr = nfs4_listxattr,
122 #endif /* CONFIG_NFS_V4 */
128 nfs_opendir(struct inode *inode, struct file *filp)
133 /* Call generic open code in order to cache credentials */
135 res = nfs_open(inode, filp);
140 typedef u32 * (*decode_dirent_t)(u32 *, struct nfs_entry *, int);
144 unsigned long page_index;
147 loff_t current_index;
148 struct nfs_entry *entry;
149 decode_dirent_t decode;
152 } nfs_readdir_descriptor_t;
154 /* Now we cache directories properly, by stuffing the dirent
155 * data directly in the page cache.
157 * Inode invalidation due to refresh etc. takes care of
158 * _everything_, no sloppy entry flushing logic, no extraneous
159 * copying, network direct to page cache, the way it was meant
162 * NOTE: Dirent information verification is done always by the
163 * page-in of the RPC reply, nowhere else, this simplies
164 * things substantially.
167 int nfs_readdir_filler(nfs_readdir_descriptor_t *desc, struct page *page)
169 struct file *file = desc->file;
170 struct inode *inode = file->f_dentry->d_inode;
171 struct rpc_cred *cred = nfs_file_cred(file);
172 unsigned long timestamp;
175 dfprintk(VFS, "NFS: nfs_readdir_filler() reading cookie %Lu into page %lu.\n", (long long)desc->entry->cookie, page->index);
179 error = NFS_PROTO(inode)->readdir(file->f_dentry, cred, desc->entry->cookie, page,
180 NFS_SERVER(inode)->dtsize, desc->plus);
182 /* We requested READDIRPLUS, but the server doesn't grok it */
183 if (error == -ENOTSUPP && desc->plus) {
184 NFS_SERVER(inode)->caps &= ~NFS_CAP_READDIRPLUS;
185 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_FLAGS(inode));
191 SetPageUptodate(page);
192 spin_lock(&inode->i_lock);
193 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ATIME;
194 spin_unlock(&inode->i_lock);
195 /* Ensure consistent page alignment of the data.
196 * Note: assumes we have exclusive access to this mapping either
197 * through inode->i_sem or some other mechanism.
199 if (page->index == 0)
200 invalidate_inode_pages2_range(inode->i_mapping, PAGE_CACHE_SIZE, -1);
206 nfs_zap_caches(inode);
212 int dir_decode(nfs_readdir_descriptor_t *desc)
215 p = desc->decode(p, desc->entry, desc->plus);
223 void dir_page_release(nfs_readdir_descriptor_t *desc)
226 page_cache_release(desc->page);
232 * Given a pointer to a buffer that has already been filled by a call
233 * to readdir, find the next entry with cookie '*desc->dir_cookie'.
235 * If the end of the buffer has been reached, return -EAGAIN, if not,
236 * return the offset within the buffer of the next entry to be
240 int find_dirent(nfs_readdir_descriptor_t *desc)
242 struct nfs_entry *entry = desc->entry;
246 while((status = dir_decode(desc)) == 0) {
247 dfprintk(VFS, "NFS: found cookie %Lu\n", (unsigned long long)entry->cookie);
248 if (entry->prev_cookie == *desc->dir_cookie)
250 if (loop_count++ > 200) {
255 dfprintk(VFS, "NFS: find_dirent() returns %d\n", status);
260 * Given a pointer to a buffer that has already been filled by a call
261 * to readdir, find the entry at offset 'desc->file->f_pos'.
263 * If the end of the buffer has been reached, return -EAGAIN, if not,
264 * return the offset within the buffer of the next entry to be
268 int find_dirent_index(nfs_readdir_descriptor_t *desc)
270 struct nfs_entry *entry = desc->entry;
275 status = dir_decode(desc);
279 dfprintk(VFS, "NFS: found cookie %Lu at index %Ld\n", (unsigned long long)entry->cookie, desc->current_index);
281 if (desc->file->f_pos == desc->current_index) {
282 *desc->dir_cookie = entry->cookie;
285 desc->current_index++;
286 if (loop_count++ > 200) {
291 dfprintk(VFS, "NFS: find_dirent_index() returns %d\n", status);
296 * Find the given page, and call find_dirent() or find_dirent_index in
297 * order to try to return the next entry.
300 int find_dirent_page(nfs_readdir_descriptor_t *desc)
302 struct inode *inode = desc->file->f_dentry->d_inode;
306 dfprintk(VFS, "NFS: find_dirent_page() searching directory page %ld\n", desc->page_index);
308 page = read_cache_page(inode->i_mapping, desc->page_index,
309 (filler_t *)nfs_readdir_filler, desc);
311 status = PTR_ERR(page);
314 if (!PageUptodate(page))
317 /* NOTE: Someone else may have changed the READDIRPLUS flag */
319 desc->ptr = kmap(page); /* matching kunmap in nfs_do_filldir */
320 if (*desc->dir_cookie != 0)
321 status = find_dirent(desc);
323 status = find_dirent_index(desc);
325 dir_page_release(desc);
327 dfprintk(VFS, "NFS: find_dirent_page() returns %d\n", status);
330 page_cache_release(page);
335 * Recurse through the page cache pages, and return a
336 * filled nfs_entry structure of the next directory entry if possible.
338 * The target for the search is '*desc->dir_cookie' if non-0,
339 * 'desc->file->f_pos' otherwise
342 int readdir_search_pagecache(nfs_readdir_descriptor_t *desc)
347 /* Always search-by-index from the beginning of the cache */
348 if (*desc->dir_cookie == 0) {
349 dfprintk(VFS, "NFS: readdir_search_pagecache() searching for offset %Ld\n", (long long)desc->file->f_pos);
350 desc->page_index = 0;
351 desc->entry->cookie = desc->entry->prev_cookie = 0;
352 desc->entry->eof = 0;
353 desc->current_index = 0;
355 dfprintk(VFS, "NFS: readdir_search_pagecache() searching for cookie %Lu\n", (unsigned long long)*desc->dir_cookie);
358 res = find_dirent_page(desc);
361 /* Align to beginning of next page */
363 if (loop_count++ > 200) {
368 dfprintk(VFS, "NFS: readdir_search_pagecache() returned %d\n", res);
372 static inline unsigned int dt_type(struct inode *inode)
374 return (inode->i_mode >> 12) & 15;
377 static struct dentry *nfs_readdir_lookup(nfs_readdir_descriptor_t *desc);
380 * Once we've found the start of the dirent within a page: fill 'er up...
383 int nfs_do_filldir(nfs_readdir_descriptor_t *desc, void *dirent,
386 struct file *file = desc->file;
387 struct nfs_entry *entry = desc->entry;
388 struct dentry *dentry = NULL;
389 unsigned long fileid;
393 dfprintk(VFS, "NFS: nfs_do_filldir() filling starting @ cookie %Lu\n", (long long)entry->cookie);
396 unsigned d_type = DT_UNKNOWN;
397 /* Note: entry->prev_cookie contains the cookie for
398 * retrieving the current dirent on the server */
399 fileid = nfs_fileid_to_ino_t(entry->ino);
401 /* Get a dentry if we have one */
404 dentry = nfs_readdir_lookup(desc);
406 /* Use readdirplus info */
407 if (dentry != NULL && dentry->d_inode != NULL) {
408 d_type = dt_type(dentry->d_inode);
409 fileid = dentry->d_inode->i_ino;
412 res = filldir(dirent, entry->name, entry->len,
413 file->f_pos, fileid, d_type);
417 *desc->dir_cookie = entry->cookie;
418 if (dir_decode(desc) != 0) {
422 if (loop_count++ > 200) {
427 dir_page_release(desc);
430 dfprintk(VFS, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n", (unsigned long long)*desc->dir_cookie, res);
435 * If we cannot find a cookie in our cache, we suspect that this is
436 * because it points to a deleted file, so we ask the server to return
437 * whatever it thinks is the next entry. We then feed this to filldir.
438 * If all goes well, we should then be able to find our way round the
439 * cache on the next call to readdir_search_pagecache();
441 * NOTE: we cannot add the anonymous page to the pagecache because
442 * the data it contains might not be page aligned. Besides,
443 * we should already have a complete representation of the
444 * directory in the page cache by the time we get here.
447 int uncached_readdir(nfs_readdir_descriptor_t *desc, void *dirent,
450 struct file *file = desc->file;
451 struct inode *inode = file->f_dentry->d_inode;
452 struct rpc_cred *cred = nfs_file_cred(file);
453 struct page *page = NULL;
456 dfprintk(VFS, "NFS: uncached_readdir() searching for cookie %Lu\n", (unsigned long long)*desc->dir_cookie);
458 page = alloc_page(GFP_HIGHUSER);
463 desc->error = NFS_PROTO(inode)->readdir(file->f_dentry, cred, *desc->dir_cookie,
465 NFS_SERVER(inode)->dtsize,
467 spin_lock(&inode->i_lock);
468 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ATIME;
469 spin_unlock(&inode->i_lock);
471 desc->ptr = kmap(page); /* matching kunmap in nfs_do_filldir */
472 if (desc->error >= 0) {
473 if ((status = dir_decode(desc)) == 0)
474 desc->entry->prev_cookie = *desc->dir_cookie;
480 status = nfs_do_filldir(desc, dirent, filldir);
482 /* Reset read descriptor so it searches the page cache from
483 * the start upon the next call to readdir_search_pagecache() */
484 desc->page_index = 0;
485 desc->entry->cookie = desc->entry->prev_cookie = 0;
486 desc->entry->eof = 0;
488 dfprintk(VFS, "NFS: uncached_readdir() returns %d\n", status);
491 dir_page_release(desc);
495 /* The file offset position represents the dirent entry number. A
496 last cookie cache takes care of the common case of reading the
499 static int nfs_readdir(struct file *filp, void *dirent, filldir_t filldir)
501 struct dentry *dentry = filp->f_dentry;
502 struct inode *inode = dentry->d_inode;
503 nfs_readdir_descriptor_t my_desc,
505 struct nfs_entry my_entry;
507 struct nfs_fattr fattr;
512 res = nfs_revalidate_inode(NFS_SERVER(inode), inode);
519 * filp->f_pos points to the dirent entry number.
520 * *desc->dir_cookie has the cookie for the next entry. We have
521 * to either find the entry with the appropriate number or
522 * revalidate the cookie.
524 memset(desc, 0, sizeof(*desc));
527 desc->dir_cookie = &((struct nfs_open_context *)filp->private_data)->dir_cookie;
528 desc->decode = NFS_PROTO(inode)->decode_dirent;
529 desc->plus = NFS_USE_READDIRPLUS(inode);
531 my_entry.cookie = my_entry.prev_cookie = 0;
534 my_entry.fattr = &fattr;
535 desc->entry = &my_entry;
537 while(!desc->entry->eof) {
538 res = readdir_search_pagecache(desc);
540 if (res == -EBADCOOKIE) {
541 /* This means either end of directory */
542 if (*desc->dir_cookie && desc->entry->cookie != *desc->dir_cookie) {
543 /* Or that the server has 'lost' a cookie */
544 res = uncached_readdir(desc, dirent, filldir);
551 if (res == -ETOOSMALL && desc->plus) {
552 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_FLAGS(inode));
553 nfs_zap_caches(inode);
555 desc->entry->eof = 0;
561 res = nfs_do_filldir(desc, dirent, filldir);
575 loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int origin)
577 down(&filp->f_dentry->d_inode->i_sem);
580 offset += filp->f_pos;
588 if (offset != filp->f_pos) {
589 filp->f_pos = offset;
590 ((struct nfs_open_context *)filp->private_data)->dir_cookie = 0;
593 up(&filp->f_dentry->d_inode->i_sem);
598 * All directory operations under NFS are synchronous, so fsync()
599 * is a dummy operation.
601 int nfs_fsync_dir(struct file *filp, struct dentry *dentry, int datasync)
607 * A check for whether or not the parent directory has changed.
608 * In the case it has, we assume that the dentries are untrustworthy
609 * and may need to be looked up again.
611 static inline int nfs_check_verifier(struct inode *dir, struct dentry *dentry)
615 if ((NFS_I(dir)->cache_validity & NFS_INO_INVALID_ATTR) != 0
616 || nfs_attribute_timeout(dir))
618 return nfs_verify_change_attribute(dir, (unsigned long)dentry->d_fsdata);
621 static inline void nfs_set_verifier(struct dentry * dentry, unsigned long verf)
623 dentry->d_fsdata = (void *)verf;
627 * Whenever an NFS operation succeeds, we know that the dentry
628 * is valid, so we update the revalidation timestamp.
630 static inline void nfs_renew_times(struct dentry * dentry)
632 dentry->d_time = jiffies;
636 * Return the intent data that applies to this particular path component
638 * Note that the current set of intents only apply to the very last
639 * component of the path.
640 * We check for this using LOOKUP_CONTINUE and LOOKUP_PARENT.
642 static inline unsigned int nfs_lookup_check_intent(struct nameidata *nd, unsigned int mask)
644 if (nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))
646 return nd->flags & mask;
650 * Inode and filehandle revalidation for lookups.
652 * We force revalidation in the cases where the VFS sets LOOKUP_REVAL,
653 * or if the intent information indicates that we're about to open this
654 * particular file and the "nocto" mount flag is not set.
658 int nfs_lookup_verify_inode(struct inode *inode, struct nameidata *nd)
660 struct nfs_server *server = NFS_SERVER(inode);
663 /* VFS wants an on-the-wire revalidation */
664 if (nd->flags & LOOKUP_REVAL)
666 /* This is an open(2) */
667 if (nfs_lookup_check_intent(nd, LOOKUP_OPEN) != 0 &&
668 !(server->flags & NFS_MOUNT_NOCTO))
671 return nfs_revalidate_inode(server, inode);
673 return __nfs_revalidate_inode(server, inode);
677 * We judge how long we want to trust negative
678 * dentries by looking at the parent inode mtime.
680 * If parent mtime has changed, we revalidate, else we wait for a
681 * period corresponding to the parent's attribute cache timeout value.
684 int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry,
685 struct nameidata *nd)
687 /* Don't revalidate a negative dentry if we're creating a new file */
688 if (nd != NULL && nfs_lookup_check_intent(nd, LOOKUP_CREATE) != 0)
690 return !nfs_check_verifier(dir, dentry);
694 * This is called every time the dcache has a lookup hit,
695 * and we should check whether we can really trust that
698 * NOTE! The hit can be a negative hit too, don't assume
701 * If the parent directory is seen to have changed, we throw out the
702 * cached dentry and do a new lookup.
704 static int nfs_lookup_revalidate(struct dentry * dentry, struct nameidata *nd)
708 struct dentry *parent;
710 struct nfs_fh fhandle;
711 struct nfs_fattr fattr;
712 unsigned long verifier;
714 parent = dget_parent(dentry);
716 dir = parent->d_inode;
717 inode = dentry->d_inode;
720 if (nfs_neg_need_reval(dir, dentry, nd))
725 if (is_bad_inode(inode)) {
726 dfprintk(VFS, "nfs_lookup_validate: %s/%s has dud inode\n",
727 dentry->d_parent->d_name.name, dentry->d_name.name);
731 /* Revalidate parent directory attribute cache */
732 if (nfs_revalidate_inode(NFS_SERVER(dir), dir) < 0)
735 /* Force a full look up iff the parent directory has changed */
736 if (nfs_check_verifier(dir, dentry)) {
737 if (nfs_lookup_verify_inode(inode, nd))
742 if (NFS_STALE(inode))
745 verifier = nfs_save_change_attribute(dir);
746 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, &fhandle, &fattr);
749 if (nfs_compare_fh(NFS_FH(inode), &fhandle))
751 if ((error = nfs_refresh_inode(inode, &fattr)) != 0)
754 nfs_renew_times(dentry);
755 nfs_set_verifier(dentry, verifier);
764 if (inode && S_ISDIR(inode->i_mode)) {
765 /* Purge readdir caches. */
766 nfs_zap_caches(inode);
767 /* If we have submounts, don't unhash ! */
768 if (have_submounts(dentry))
770 shrink_dcache_parent(dentry);
779 * This is called from dput() when d_count is going to 0.
781 static int nfs_dentry_delete(struct dentry *dentry)
783 dfprintk(VFS, "NFS: dentry_delete(%s/%s, %x)\n",
784 dentry->d_parent->d_name.name, dentry->d_name.name,
787 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
788 /* Unhash it, so that ->d_iput() would be called */
791 if (!(dentry->d_sb->s_flags & MS_ACTIVE)) {
792 /* Unhash it, so that ancestors of killed async unlink
793 * files will be cleaned up during umount */
801 * Called when the dentry loses inode.
802 * We use it to clean up silly-renamed files.
804 static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode)
806 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
809 nfs_complete_unlink(dentry);
812 /* When creating a negative dentry, we want to renew d_time */
813 nfs_renew_times(dentry);
817 struct dentry_operations nfs_dentry_operations = {
818 .d_revalidate = nfs_lookup_revalidate,
819 .d_delete = nfs_dentry_delete,
820 .d_iput = nfs_dentry_iput,
824 * Use intent information to check whether or not we're going to do
825 * an O_EXCL create using this path component.
828 int nfs_is_exclusive_create(struct inode *dir, struct nameidata *nd)
830 if (NFS_PROTO(dir)->version == 2)
832 if (nd == NULL || nfs_lookup_check_intent(nd, LOOKUP_CREATE) == 0)
834 return (nd->intent.open.flags & O_EXCL) != 0;
837 static struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
840 struct inode *inode = NULL;
842 struct nfs_fh fhandle;
843 struct nfs_fattr fattr;
845 dfprintk(VFS, "NFS: lookup(%s/%s)\n",
846 dentry->d_parent->d_name.name, dentry->d_name.name);
848 res = ERR_PTR(-ENAMETOOLONG);
849 if (dentry->d_name.len > NFS_SERVER(dir)->namelen)
852 res = ERR_PTR(-ENOMEM);
853 dentry->d_op = NFS_PROTO(dir)->dentry_ops;
856 /* Revalidate parent directory attribute cache */
857 error = nfs_revalidate_inode(NFS_SERVER(dir), dir);
859 res = ERR_PTR(error);
863 /* If we're doing an exclusive create, optimize away the lookup */
864 if (nfs_is_exclusive_create(dir, nd))
867 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, &fhandle, &fattr);
868 if (error == -ENOENT)
871 res = ERR_PTR(error);
874 res = ERR_PTR(-EACCES);
875 inode = nfs_fhget(dentry->d_sb, &fhandle, &fattr);
879 res = d_add_unique(dentry, inode);
882 nfs_renew_times(dentry);
883 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
891 static int nfs_open_revalidate(struct dentry *, struct nameidata *);
893 struct dentry_operations nfs4_dentry_operations = {
894 .d_revalidate = nfs_open_revalidate,
895 .d_delete = nfs_dentry_delete,
896 .d_iput = nfs_dentry_iput,
900 * Use intent information to determine whether we need to substitute
901 * the NFSv4-style stateful OPEN for the LOOKUP call
903 static int is_atomic_open(struct inode *dir, struct nameidata *nd)
905 if (nd == NULL || nfs_lookup_check_intent(nd, LOOKUP_OPEN) == 0)
907 /* NFS does not (yet) have a stateful open for directories */
908 if (nd->flags & LOOKUP_DIRECTORY)
910 /* Are we trying to write to a read only partition? */
911 if (IS_RDONLY(dir) && (nd->intent.open.flags & (O_CREAT|O_TRUNC|FMODE_WRITE)))
916 static struct dentry *nfs_atomic_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd)
918 struct dentry *res = NULL;
919 struct inode *inode = NULL;
922 /* Check that we are indeed trying to open this file */
923 if (!is_atomic_open(dir, nd))
926 if (dentry->d_name.len > NFS_SERVER(dir)->namelen) {
927 res = ERR_PTR(-ENAMETOOLONG);
930 dentry->d_op = NFS_PROTO(dir)->dentry_ops;
932 /* Let vfs_create() deal with O_EXCL */
933 if (nd->intent.open.flags & O_EXCL)
936 /* Open the file on the server */
938 /* Revalidate parent directory attribute cache */
939 error = nfs_revalidate_inode(NFS_SERVER(dir), dir);
941 res = ERR_PTR(error);
945 if (nd->intent.open.flags & O_CREAT) {
946 nfs_begin_data_update(dir);
947 inode = nfs4_atomic_open(dir, dentry, nd);
948 nfs_end_data_update(dir);
950 inode = nfs4_atomic_open(dir, dentry, nd);
953 error = PTR_ERR(inode);
955 /* Make a negative dentry */
959 /* This turned out not to be a regular file */
961 if (!(nd->intent.open.flags & O_NOFOLLOW))
966 res = ERR_PTR(error);
971 res = d_add_unique(dentry, inode);
974 nfs_renew_times(dentry);
975 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
979 return nfs_lookup(dir, dentry, nd);
982 static int nfs_open_revalidate(struct dentry *dentry, struct nameidata *nd)
984 struct dentry *parent = NULL;
985 struct inode *inode = dentry->d_inode;
987 unsigned long verifier;
988 int openflags, ret = 0;
990 parent = dget_parent(dentry);
991 dir = parent->d_inode;
992 if (!is_atomic_open(dir, nd))
994 /* We can't create new files in nfs_open_revalidate(), so we
995 * optimize away revalidation of negative dentries.
999 /* NFS only supports OPEN on regular files */
1000 if (!S_ISREG(inode->i_mode))
1002 openflags = nd->intent.open.flags;
1003 /* We cannot do exclusive creation on a positive dentry */
1004 if ((openflags & (O_CREAT|O_EXCL)) == (O_CREAT|O_EXCL))
1006 /* We can't create new files, or truncate existing ones here */
1007 openflags &= ~(O_CREAT|O_TRUNC);
1010 * Note: we're not holding inode->i_sem and so may be racing with
1011 * operations that change the directory. We therefore save the
1012 * change attribute *before* we do the RPC call.
1015 verifier = nfs_save_change_attribute(dir);
1016 ret = nfs4_open_revalidate(dir, dentry, openflags);
1018 nfs_set_verifier(dentry, verifier);
1027 if (inode != NULL && nfs_have_delegation(inode, FMODE_READ))
1029 return nfs_lookup_revalidate(dentry, nd);
1031 #endif /* CONFIG_NFSV4 */
1033 static struct dentry *nfs_readdir_lookup(nfs_readdir_descriptor_t *desc)
1035 struct dentry *parent = desc->file->f_dentry;
1036 struct inode *dir = parent->d_inode;
1037 struct nfs_entry *entry = desc->entry;
1038 struct dentry *dentry, *alias;
1039 struct qstr name = {
1040 .name = entry->name,
1043 struct inode *inode;
1047 if (name.name[0] == '.' && name.name[1] == '.')
1048 return dget_parent(parent);
1051 if (name.name[0] == '.')
1052 return dget(parent);
1054 name.hash = full_name_hash(name.name, name.len);
1055 dentry = d_lookup(parent, &name);
1058 if (!desc->plus || !(entry->fattr->valid & NFS_ATTR_FATTR))
1060 /* Note: caller is already holding the dir->i_sem! */
1061 dentry = d_alloc(parent, &name);
1064 dentry->d_op = NFS_PROTO(dir)->dentry_ops;
1065 inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr);
1070 alias = d_add_unique(dentry, inode);
1071 if (alias != NULL) {
1075 nfs_renew_times(dentry);
1076 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1081 * Code common to create, mkdir, and mknod.
1083 int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle,
1084 struct nfs_fattr *fattr)
1086 struct inode *inode;
1087 int error = -EACCES;
1089 /* We may have been initialized further down */
1090 if (dentry->d_inode)
1092 if (fhandle->size == 0) {
1093 struct inode *dir = dentry->d_parent->d_inode;
1094 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr);
1098 if (!(fattr->valid & NFS_ATTR_FATTR)) {
1099 struct nfs_server *server = NFS_SB(dentry->d_sb);
1100 error = server->rpc_ops->getattr(server, fhandle, fattr);
1105 inode = nfs_fhget(dentry->d_sb, fhandle, fattr);
1108 d_instantiate(dentry, inode);
1116 * Following a failed create operation, we drop the dentry rather
1117 * than retain a negative dentry. This avoids a problem in the event
1118 * that the operation succeeded on the server, but an error in the
1119 * reply path made it appear to have failed.
1121 static int nfs_create(struct inode *dir, struct dentry *dentry, int mode,
1122 struct nameidata *nd)
1128 dfprintk(VFS, "NFS: create(%s/%ld, %s\n", dir->i_sb->s_id,
1129 dir->i_ino, dentry->d_name.name);
1131 attr.ia_mode = mode;
1132 attr.ia_valid = ATTR_MODE;
1134 if (nd && (nd->flags & LOOKUP_CREATE))
1135 open_flags = nd->intent.open.flags;
1138 nfs_begin_data_update(dir);
1139 error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags);
1140 nfs_end_data_update(dir);
1143 nfs_renew_times(dentry);
1144 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1154 * See comments for nfs_proc_create regarding failed operations.
1157 nfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t rdev)
1162 dfprintk(VFS, "NFS: mknod(%s/%ld, %s\n", dir->i_sb->s_id,
1163 dir->i_ino, dentry->d_name.name);
1165 if (!new_valid_dev(rdev))
1168 attr.ia_mode = mode;
1169 attr.ia_valid = ATTR_MODE;
1172 nfs_begin_data_update(dir);
1173 status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev);
1174 nfs_end_data_update(dir);
1177 nfs_renew_times(dentry);
1178 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1188 * See comments for nfs_proc_create regarding failed operations.
1190 static int nfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1195 dfprintk(VFS, "NFS: mkdir(%s/%ld, %s\n", dir->i_sb->s_id,
1196 dir->i_ino, dentry->d_name.name);
1198 attr.ia_valid = ATTR_MODE;
1199 attr.ia_mode = mode | S_IFDIR;
1202 nfs_begin_data_update(dir);
1203 error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr);
1204 nfs_end_data_update(dir);
1207 nfs_renew_times(dentry);
1208 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1217 static int nfs_rmdir(struct inode *dir, struct dentry *dentry)
1221 dfprintk(VFS, "NFS: rmdir(%s/%ld, %s\n", dir->i_sb->s_id,
1222 dir->i_ino, dentry->d_name.name);
1225 nfs_begin_data_update(dir);
1226 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
1227 /* Ensure the VFS deletes this inode */
1228 if (error == 0 && dentry->d_inode != NULL)
1229 dentry->d_inode->i_nlink = 0;
1230 nfs_end_data_update(dir);
1236 static int nfs_sillyrename(struct inode *dir, struct dentry *dentry)
1238 static unsigned int sillycounter;
1239 const int i_inosize = sizeof(dir->i_ino)*2;
1240 const int countersize = sizeof(sillycounter)*2;
1241 const int slen = sizeof(".nfs") + i_inosize + countersize - 1;
1244 struct dentry *sdentry;
1247 dfprintk(VFS, "NFS: silly-rename(%s/%s, ct=%d)\n",
1248 dentry->d_parent->d_name.name, dentry->d_name.name,
1249 atomic_read(&dentry->d_count));
1252 if (!dentry->d_inode)
1253 printk("NFS: silly-renaming %s/%s, negative dentry??\n",
1254 dentry->d_parent->d_name.name, dentry->d_name.name);
1257 * We don't allow a dentry to be silly-renamed twice.
1260 if (dentry->d_flags & DCACHE_NFSFS_RENAMED)
1263 sprintf(silly, ".nfs%*.*lx",
1264 i_inosize, i_inosize, dentry->d_inode->i_ino);
1268 char *suffix = silly + slen - countersize;
1272 sprintf(suffix, "%*.*x", countersize, countersize, sillycounter);
1274 dfprintk(VFS, "trying to rename %s to %s\n",
1275 dentry->d_name.name, silly);
1277 sdentry = lookup_one_len(silly, dentry->d_parent, slen);
1279 * N.B. Better to return EBUSY here ... it could be
1280 * dangerous to delete the file while it's in use.
1282 if (IS_ERR(sdentry))
1284 } while(sdentry->d_inode != NULL); /* need negative lookup */
1286 qsilly.name = silly;
1287 qsilly.len = strlen(silly);
1288 nfs_begin_data_update(dir);
1289 if (dentry->d_inode) {
1290 nfs_begin_data_update(dentry->d_inode);
1291 error = NFS_PROTO(dir)->rename(dir, &dentry->d_name,
1293 nfs_end_data_update(dentry->d_inode);
1295 error = NFS_PROTO(dir)->rename(dir, &dentry->d_name,
1297 nfs_end_data_update(dir);
1299 nfs_renew_times(dentry);
1300 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1301 d_move(dentry, sdentry);
1302 error = nfs_async_unlink(dentry);
1303 /* If we return 0 we don't unlink */
1311 * Remove a file after making sure there are no pending writes,
1312 * and after checking that the file has only one user.
1314 * We invalidate the attribute cache and free the inode prior to the operation
1315 * to avoid possible races if the server reuses the inode.
1317 static int nfs_safe_remove(struct dentry *dentry)
1319 struct inode *dir = dentry->d_parent->d_inode;
1320 struct inode *inode = dentry->d_inode;
1323 dfprintk(VFS, "NFS: safe_remove(%s/%s)\n",
1324 dentry->d_parent->d_name.name, dentry->d_name.name);
1326 /* If the dentry was sillyrenamed, we simply call d_delete() */
1327 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1332 nfs_begin_data_update(dir);
1333 if (inode != NULL) {
1334 nfs_begin_data_update(inode);
1335 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1336 /* The VFS may want to delete this inode */
1339 nfs_end_data_update(inode);
1341 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1342 nfs_end_data_update(dir);
1347 /* We do silly rename. In case sillyrename() returns -EBUSY, the inode
1348 * belongs to an active ".nfs..." file and we return -EBUSY.
1350 * If sillyrename() returns 0, we do nothing, otherwise we unlink.
1352 static int nfs_unlink(struct inode *dir, struct dentry *dentry)
1355 int need_rehash = 0;
1357 dfprintk(VFS, "NFS: unlink(%s/%ld, %s)\n", dir->i_sb->s_id,
1358 dir->i_ino, dentry->d_name.name);
1361 spin_lock(&dcache_lock);
1362 spin_lock(&dentry->d_lock);
1363 if (atomic_read(&dentry->d_count) > 1) {
1364 spin_unlock(&dentry->d_lock);
1365 spin_unlock(&dcache_lock);
1366 error = nfs_sillyrename(dir, dentry);
1370 if (!d_unhashed(dentry)) {
1374 spin_unlock(&dentry->d_lock);
1375 spin_unlock(&dcache_lock);
1376 error = nfs_safe_remove(dentry);
1378 nfs_renew_times(dentry);
1379 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1380 } else if (need_rehash)
1387 nfs_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1390 struct nfs_fattr sym_attr;
1391 struct nfs_fh sym_fh;
1392 struct qstr qsymname;
1395 dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s)\n", dir->i_sb->s_id,
1396 dir->i_ino, dentry->d_name.name, symname);
1399 if (dentry->d_inode)
1400 printk("nfs_proc_symlink: %s/%s not negative!\n",
1401 dentry->d_parent->d_name.name, dentry->d_name.name);
1404 * Fill in the sattr for the call.
1405 * Note: SunOS 4.1.2 crashes if the mode isn't initialized!
1407 attr.ia_valid = ATTR_MODE;
1408 attr.ia_mode = S_IFLNK | S_IRWXUGO;
1410 qsymname.name = symname;
1411 qsymname.len = strlen(symname);
1414 nfs_begin_data_update(dir);
1415 error = NFS_PROTO(dir)->symlink(dir, &dentry->d_name, &qsymname,
1416 &attr, &sym_fh, &sym_attr);
1417 nfs_end_data_update(dir);
1419 error = nfs_instantiate(dentry, &sym_fh, &sym_attr);
1421 if (error == -EEXIST)
1422 printk("nfs_proc_symlink: %s/%s already exists??\n",
1423 dentry->d_parent->d_name.name, dentry->d_name.name);
1431 nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1433 struct inode *inode = old_dentry->d_inode;
1436 dfprintk(VFS, "NFS: link(%s/%s -> %s/%s)\n",
1437 old_dentry->d_parent->d_name.name, old_dentry->d_name.name,
1438 dentry->d_parent->d_name.name, dentry->d_name.name);
1441 * Drop the dentry in advance to force a new lookup.
1442 * Since nfs_proc_link doesn't return a file handle,
1443 * we can't use the existing dentry.
1448 nfs_begin_data_update(dir);
1449 nfs_begin_data_update(inode);
1450 error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name);
1451 nfs_end_data_update(inode);
1452 nfs_end_data_update(dir);
1459 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a
1460 * different file handle for the same inode after a rename (e.g. when
1461 * moving to a different directory). A fail-safe method to do so would
1462 * be to look up old_dir/old_name, create a link to new_dir/new_name and
1463 * rename the old file using the sillyrename stuff. This way, the original
1464 * file in old_dir will go away when the last process iput()s the inode.
1468 * It actually works quite well. One needs to have the possibility for
1469 * at least one ".nfs..." file in each directory the file ever gets
1470 * moved or linked to which happens automagically with the new
1471 * implementation that only depends on the dcache stuff instead of
1472 * using the inode layer
1474 * Unfortunately, things are a little more complicated than indicated
1475 * above. For a cross-directory move, we want to make sure we can get
1476 * rid of the old inode after the operation. This means there must be
1477 * no pending writes (if it's a file), and the use count must be 1.
1478 * If these conditions are met, we can drop the dentries before doing
1481 static int nfs_rename(struct inode *old_dir, struct dentry *old_dentry,
1482 struct inode *new_dir, struct dentry *new_dentry)
1484 struct inode *old_inode = old_dentry->d_inode;
1485 struct inode *new_inode = new_dentry->d_inode;
1486 struct dentry *dentry = NULL, *rehash = NULL;
1490 * To prevent any new references to the target during the rename,
1491 * we unhash the dentry and free the inode in advance.
1494 if (!d_unhashed(new_dentry)) {
1496 rehash = new_dentry;
1499 dfprintk(VFS, "NFS: rename(%s/%s -> %s/%s, ct=%d)\n",
1500 old_dentry->d_parent->d_name.name, old_dentry->d_name.name,
1501 new_dentry->d_parent->d_name.name, new_dentry->d_name.name,
1502 atomic_read(&new_dentry->d_count));
1505 * First check whether the target is busy ... we can't
1506 * safely do _any_ rename if the target is in use.
1508 * For files, make a copy of the dentry and then do a
1509 * silly-rename. If the silly-rename succeeds, the
1510 * copied dentry is hashed and becomes the new target.
1514 if (S_ISDIR(new_inode->i_mode))
1516 else if (atomic_read(&new_dentry->d_count) > 2) {
1518 /* copy the target dentry's name */
1519 dentry = d_alloc(new_dentry->d_parent,
1520 &new_dentry->d_name);
1524 /* silly-rename the existing target ... */
1525 err = nfs_sillyrename(new_dir, new_dentry);
1527 new_dentry = rehash = dentry;
1529 /* instantiate the replacement target */
1530 d_instantiate(new_dentry, NULL);
1531 } else if (atomic_read(&new_dentry->d_count) > 1) {
1532 /* dentry still busy? */
1534 printk("nfs_rename: target %s/%s busy, d_count=%d\n",
1535 new_dentry->d_parent->d_name.name,
1536 new_dentry->d_name.name,
1537 atomic_read(&new_dentry->d_count));
1545 * ... prune child dentries and writebacks if needed.
1547 if (atomic_read(&old_dentry->d_count) > 1) {
1548 nfs_wb_all(old_inode);
1549 shrink_dcache_parent(old_dentry);
1553 d_delete(new_dentry);
1555 nfs_begin_data_update(old_dir);
1556 nfs_begin_data_update(new_dir);
1557 nfs_begin_data_update(old_inode);
1558 error = NFS_PROTO(old_dir)->rename(old_dir, &old_dentry->d_name,
1559 new_dir, &new_dentry->d_name);
1560 nfs_end_data_update(old_inode);
1561 nfs_end_data_update(new_dir);
1562 nfs_end_data_update(old_dir);
1567 if (!S_ISDIR(old_inode->i_mode))
1568 d_move(old_dentry, new_dentry);
1569 nfs_renew_times(new_dentry);
1570 nfs_set_verifier(new_dentry, nfs_save_change_attribute(new_dir));
1573 /* new dentry created? */
1580 int nfs_access_get_cached(struct inode *inode, struct rpc_cred *cred, struct nfs_access_entry *res)
1582 struct nfs_inode *nfsi = NFS_I(inode);
1583 struct nfs_access_entry *cache = &nfsi->cache_access;
1585 if (cache->cred != cred
1586 || time_after(jiffies, cache->jiffies + NFS_ATTRTIMEO(inode))
1587 || (nfsi->cache_validity & NFS_INO_INVALID_ACCESS))
1589 memcpy(res, cache, sizeof(*res));
1593 void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set)
1595 struct nfs_inode *nfsi = NFS_I(inode);
1596 struct nfs_access_entry *cache = &nfsi->cache_access;
1598 if (cache->cred != set->cred) {
1600 put_rpccred(cache->cred);
1601 cache->cred = get_rpccred(set->cred);
1603 /* FIXME: replace current access_cache BKL reliance with inode->i_lock */
1604 spin_lock(&inode->i_lock);
1605 nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS;
1606 spin_unlock(&inode->i_lock);
1607 cache->jiffies = set->jiffies;
1608 cache->mask = set->mask;
1611 static int nfs_do_access(struct inode *inode, struct rpc_cred *cred, int mask)
1613 struct nfs_access_entry cache;
1616 status = nfs_access_get_cached(inode, cred, &cache);
1620 /* Be clever: ask server to check for all possible rights */
1621 cache.mask = MAY_EXEC | MAY_WRITE | MAY_READ;
1623 cache.jiffies = jiffies;
1624 status = NFS_PROTO(inode)->access(inode, &cache);
1627 nfs_access_add_cache(inode, &cache);
1629 if ((cache.mask & mask) == mask)
1634 int nfs_permission(struct inode *inode, int mask, struct nameidata *nd)
1636 struct rpc_cred *cred;
1641 /* Is this sys_access() ? */
1642 if (nd != NULL && (nd->flags & LOOKUP_ACCESS))
1645 switch (inode->i_mode & S_IFMT) {
1649 /* NFSv4 has atomic_open... */
1650 if (nfs_server_capable(inode, NFS_CAP_ATOMIC_OPEN)
1652 && (nd->flags & LOOKUP_OPEN))
1657 * Optimize away all write operations, since the server
1658 * will check permissions when we perform the op.
1660 if ((mask & MAY_WRITE) && !(mask & MAY_READ))
1667 if (!NFS_PROTO(inode)->access)
1670 cred = rpcauth_lookupcred(NFS_CLIENT(inode)->cl_auth, 0);
1671 if (!IS_ERR(cred)) {
1672 res = nfs_do_access(inode, cred, mask);
1675 res = PTR_ERR(cred);
1680 res = nfs_revalidate_inode(NFS_SERVER(inode), inode);
1682 res = generic_permission(inode, mask, NULL);
1689 * version-control: t
1690 * kept-new-versions: 5