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"
41 #define NFS_PARANOIA 1
42 /* #define NFS_DEBUG_VERBOSE 1 */
44 static int nfs_opendir(struct inode *, struct file *);
45 static int nfs_readdir(struct file *, void *, filldir_t);
46 static struct dentry *nfs_lookup(struct inode *, struct dentry *, struct nameidata *);
47 static int nfs_create(struct inode *, struct dentry *, int, struct nameidata *);
48 static int nfs_mkdir(struct inode *, struct dentry *, int);
49 static int nfs_rmdir(struct inode *, struct dentry *);
50 static int nfs_unlink(struct inode *, struct dentry *);
51 static int nfs_symlink(struct inode *, struct dentry *, const char *);
52 static int nfs_link(struct dentry *, struct inode *, struct dentry *);
53 static int nfs_mknod(struct inode *, struct dentry *, int, dev_t);
54 static int nfs_rename(struct inode *, struct dentry *,
55 struct inode *, struct dentry *);
56 static int nfs_fsync_dir(struct file *, struct dentry *, int);
57 static loff_t nfs_llseek_dir(struct file *, loff_t, int);
59 const struct file_operations nfs_dir_operations = {
60 .llseek = nfs_llseek_dir,
61 .read = generic_read_dir,
62 .readdir = nfs_readdir,
64 .release = nfs_release,
65 .fsync = nfs_fsync_dir,
68 struct inode_operations nfs_dir_inode_operations = {
73 .symlink = nfs_symlink,
78 .permission = nfs_permission,
79 .getattr = nfs_getattr,
80 .setattr = nfs_setattr,
84 struct inode_operations nfs3_dir_inode_operations = {
89 .symlink = nfs_symlink,
94 .permission = nfs_permission,
95 .getattr = nfs_getattr,
96 .setattr = nfs_setattr,
97 .listxattr = nfs3_listxattr,
98 .getxattr = nfs3_getxattr,
99 .setxattr = nfs3_setxattr,
100 .removexattr = nfs3_removexattr,
102 #endif /* CONFIG_NFS_V3 */
106 static struct dentry *nfs_atomic_lookup(struct inode *, struct dentry *, struct nameidata *);
107 struct inode_operations nfs4_dir_inode_operations = {
108 .create = nfs_create,
109 .lookup = nfs_atomic_lookup,
111 .unlink = nfs_unlink,
112 .symlink = nfs_symlink,
116 .rename = nfs_rename,
117 .permission = nfs_permission,
118 .getattr = nfs_getattr,
119 .setattr = nfs_setattr,
120 .getxattr = nfs4_getxattr,
121 .setxattr = nfs4_setxattr,
122 .listxattr = nfs4_listxattr,
125 #endif /* CONFIG_NFS_V4 */
131 nfs_opendir(struct inode *inode, struct file *filp)
135 dfprintk(VFS, "NFS: opendir(%s/%ld)\n",
136 inode->i_sb->s_id, inode->i_ino);
139 /* Call generic open code in order to cache credentials */
140 res = nfs_open(inode, filp);
145 typedef u32 * (*decode_dirent_t)(u32 *, struct nfs_entry *, int);
149 unsigned long page_index;
152 loff_t current_index;
153 struct nfs_entry *entry;
154 decode_dirent_t decode;
157 } nfs_readdir_descriptor_t;
159 /* Now we cache directories properly, by stuffing the dirent
160 * data directly in the page cache.
162 * Inode invalidation due to refresh etc. takes care of
163 * _everything_, no sloppy entry flushing logic, no extraneous
164 * copying, network direct to page cache, the way it was meant
167 * NOTE: Dirent information verification is done always by the
168 * page-in of the RPC reply, nowhere else, this simplies
169 * things substantially.
172 int nfs_readdir_filler(nfs_readdir_descriptor_t *desc, struct page *page)
174 struct file *file = desc->file;
175 struct inode *inode = file->f_dentry->d_inode;
176 struct rpc_cred *cred = nfs_file_cred(file);
177 unsigned long timestamp;
180 dfprintk(DIRCACHE, "NFS: %s: reading cookie %Lu into page %lu\n",
181 __FUNCTION__, (long long)desc->entry->cookie,
186 error = NFS_PROTO(inode)->readdir(file->f_dentry, cred, desc->entry->cookie, page,
187 NFS_SERVER(inode)->dtsize, desc->plus);
189 /* We requested READDIRPLUS, but the server doesn't grok it */
190 if (error == -ENOTSUPP && desc->plus) {
191 NFS_SERVER(inode)->caps &= ~NFS_CAP_READDIRPLUS;
192 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_FLAGS(inode));
198 SetPageUptodate(page);
199 spin_lock(&inode->i_lock);
200 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ATIME;
201 spin_unlock(&inode->i_lock);
202 /* Ensure consistent page alignment of the data.
203 * Note: assumes we have exclusive access to this mapping either
204 * through inode->i_mutex or some other mechanism.
206 if (page->index == 0)
207 invalidate_inode_pages2_range(inode->i_mapping, PAGE_CACHE_SIZE, -1);
213 nfs_zap_caches(inode);
219 int dir_decode(nfs_readdir_descriptor_t *desc)
222 p = desc->decode(p, desc->entry, desc->plus);
230 void dir_page_release(nfs_readdir_descriptor_t *desc)
233 page_cache_release(desc->page);
239 * Given a pointer to a buffer that has already been filled by a call
240 * to readdir, find the next entry with cookie '*desc->dir_cookie'.
242 * If the end of the buffer has been reached, return -EAGAIN, if not,
243 * return the offset within the buffer of the next entry to be
247 int find_dirent(nfs_readdir_descriptor_t *desc)
249 struct nfs_entry *entry = desc->entry;
253 while((status = dir_decode(desc)) == 0) {
254 dfprintk(DIRCACHE, "NFS: %s: examining cookie %Lu\n",
255 __FUNCTION__, (unsigned long long)entry->cookie);
256 if (entry->prev_cookie == *desc->dir_cookie)
258 if (loop_count++ > 200) {
267 * Given a pointer to a buffer that has already been filled by a call
268 * to readdir, find the entry at offset 'desc->file->f_pos'.
270 * If the end of the buffer has been reached, return -EAGAIN, if not,
271 * return the offset within the buffer of the next entry to be
275 int find_dirent_index(nfs_readdir_descriptor_t *desc)
277 struct nfs_entry *entry = desc->entry;
282 status = dir_decode(desc);
286 dfprintk(DIRCACHE, "NFS: found cookie %Lu at index %Ld\n",
287 (unsigned long long)entry->cookie, desc->current_index);
289 if (desc->file->f_pos == desc->current_index) {
290 *desc->dir_cookie = entry->cookie;
293 desc->current_index++;
294 if (loop_count++ > 200) {
303 * Find the given page, and call find_dirent() or find_dirent_index in
304 * order to try to return the next entry.
307 int find_dirent_page(nfs_readdir_descriptor_t *desc)
309 struct inode *inode = desc->file->f_dentry->d_inode;
313 dfprintk(DIRCACHE, "NFS: %s: searching page %ld for target %Lu\n",
314 __FUNCTION__, desc->page_index,
315 (long long) *desc->dir_cookie);
317 page = read_cache_page(inode->i_mapping, desc->page_index,
318 (filler_t *)nfs_readdir_filler, desc);
320 status = PTR_ERR(page);
323 if (!PageUptodate(page))
326 /* NOTE: Someone else may have changed the READDIRPLUS flag */
328 desc->ptr = kmap(page); /* matching kunmap in nfs_do_filldir */
329 if (*desc->dir_cookie != 0)
330 status = find_dirent(desc);
332 status = find_dirent_index(desc);
334 dir_page_release(desc);
336 dfprintk(DIRCACHE, "NFS: %s: returns %d\n", __FUNCTION__, status);
339 page_cache_release(page);
344 * Recurse through the page cache pages, and return a
345 * filled nfs_entry structure of the next directory entry if possible.
347 * The target for the search is '*desc->dir_cookie' if non-0,
348 * 'desc->file->f_pos' otherwise
351 int readdir_search_pagecache(nfs_readdir_descriptor_t *desc)
356 /* Always search-by-index from the beginning of the cache */
357 if (*desc->dir_cookie == 0) {
358 dfprintk(DIRCACHE, "NFS: readdir_search_pagecache() searching for offset %Ld\n",
359 (long long)desc->file->f_pos);
360 desc->page_index = 0;
361 desc->entry->cookie = desc->entry->prev_cookie = 0;
362 desc->entry->eof = 0;
363 desc->current_index = 0;
365 dfprintk(DIRCACHE, "NFS: readdir_search_pagecache() searching for cookie %Lu\n",
366 (unsigned long long)*desc->dir_cookie);
369 res = find_dirent_page(desc);
372 /* Align to beginning of next page */
374 if (loop_count++ > 200) {
380 dfprintk(DIRCACHE, "NFS: %s: returns %d\n", __FUNCTION__, res);
384 static inline unsigned int dt_type(struct inode *inode)
386 return (inode->i_mode >> 12) & 15;
389 static struct dentry *nfs_readdir_lookup(nfs_readdir_descriptor_t *desc);
392 * Once we've found the start of the dirent within a page: fill 'er up...
395 int nfs_do_filldir(nfs_readdir_descriptor_t *desc, void *dirent,
398 struct file *file = desc->file;
399 struct nfs_entry *entry = desc->entry;
400 struct dentry *dentry = NULL;
401 unsigned long fileid;
405 dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling starting @ cookie %Lu\n",
406 (unsigned long long)entry->cookie);
409 unsigned d_type = DT_UNKNOWN;
410 /* Note: entry->prev_cookie contains the cookie for
411 * retrieving the current dirent on the server */
412 fileid = nfs_fileid_to_ino_t(entry->ino);
414 /* Get a dentry if we have one */
417 dentry = nfs_readdir_lookup(desc);
419 /* Use readdirplus info */
420 if (dentry != NULL && dentry->d_inode != NULL) {
421 d_type = dt_type(dentry->d_inode);
422 fileid = dentry->d_inode->i_ino;
425 res = filldir(dirent, entry->name, entry->len,
426 file->f_pos, fileid, d_type);
430 *desc->dir_cookie = entry->cookie;
431 if (dir_decode(desc) != 0) {
435 if (loop_count++ > 200) {
440 dir_page_release(desc);
443 dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n",
444 (unsigned long long)*desc->dir_cookie, res);
449 * If we cannot find a cookie in our cache, we suspect that this is
450 * because it points to a deleted file, so we ask the server to return
451 * whatever it thinks is the next entry. We then feed this to filldir.
452 * If all goes well, we should then be able to find our way round the
453 * cache on the next call to readdir_search_pagecache();
455 * NOTE: we cannot add the anonymous page to the pagecache because
456 * the data it contains might not be page aligned. Besides,
457 * we should already have a complete representation of the
458 * directory in the page cache by the time we get here.
461 int uncached_readdir(nfs_readdir_descriptor_t *desc, void *dirent,
464 struct file *file = desc->file;
465 struct inode *inode = file->f_dentry->d_inode;
466 struct rpc_cred *cred = nfs_file_cred(file);
467 struct page *page = NULL;
470 dfprintk(DIRCACHE, "NFS: uncached_readdir() searching for cookie %Lu\n",
471 (unsigned long long)*desc->dir_cookie);
473 page = alloc_page(GFP_HIGHUSER);
478 desc->error = NFS_PROTO(inode)->readdir(file->f_dentry, cred, *desc->dir_cookie,
480 NFS_SERVER(inode)->dtsize,
482 spin_lock(&inode->i_lock);
483 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ATIME;
484 spin_unlock(&inode->i_lock);
486 desc->ptr = kmap(page); /* matching kunmap in nfs_do_filldir */
487 if (desc->error >= 0) {
488 if ((status = dir_decode(desc)) == 0)
489 desc->entry->prev_cookie = *desc->dir_cookie;
495 status = nfs_do_filldir(desc, dirent, filldir);
497 /* Reset read descriptor so it searches the page cache from
498 * the start upon the next call to readdir_search_pagecache() */
499 desc->page_index = 0;
500 desc->entry->cookie = desc->entry->prev_cookie = 0;
501 desc->entry->eof = 0;
503 dfprintk(DIRCACHE, "NFS: %s: returns %d\n",
504 __FUNCTION__, status);
507 dir_page_release(desc);
511 /* The file offset position represents the dirent entry number. A
512 last cookie cache takes care of the common case of reading the
515 static int nfs_readdir(struct file *filp, void *dirent, filldir_t filldir)
517 struct dentry *dentry = filp->f_dentry;
518 struct inode *inode = dentry->d_inode;
519 nfs_readdir_descriptor_t my_desc,
521 struct nfs_entry my_entry;
523 struct nfs_fattr fattr;
526 dfprintk(VFS, "NFS: readdir(%s/%s) starting at cookie %Lu\n",
527 dentry->d_parent->d_name.name, dentry->d_name.name,
528 (long long)filp->f_pos);
529 nfs_inc_stats(inode, NFSIOS_VFSGETDENTS);
533 res = nfs_revalidate_inode(NFS_SERVER(inode), inode);
540 * filp->f_pos points to the dirent entry number.
541 * *desc->dir_cookie has the cookie for the next entry. We have
542 * to either find the entry with the appropriate number or
543 * revalidate the cookie.
545 memset(desc, 0, sizeof(*desc));
548 desc->dir_cookie = &((struct nfs_open_context *)filp->private_data)->dir_cookie;
549 desc->decode = NFS_PROTO(inode)->decode_dirent;
550 desc->plus = NFS_USE_READDIRPLUS(inode);
552 my_entry.cookie = my_entry.prev_cookie = 0;
555 my_entry.fattr = &fattr;
556 nfs_fattr_init(&fattr);
557 desc->entry = &my_entry;
559 while(!desc->entry->eof) {
560 res = readdir_search_pagecache(desc);
562 if (res == -EBADCOOKIE) {
563 /* This means either end of directory */
564 if (*desc->dir_cookie && desc->entry->cookie != *desc->dir_cookie) {
565 /* Or that the server has 'lost' a cookie */
566 res = uncached_readdir(desc, dirent, filldir);
573 if (res == -ETOOSMALL && desc->plus) {
574 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_FLAGS(inode));
575 nfs_zap_caches(inode);
577 desc->entry->eof = 0;
583 res = nfs_do_filldir(desc, dirent, filldir);
592 dfprintk(VFS, "NFS: readdir(%s/%s) returns %ld\n",
593 dentry->d_parent->d_name.name, dentry->d_name.name,
598 loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int origin)
600 mutex_lock(&filp->f_dentry->d_inode->i_mutex);
603 offset += filp->f_pos;
611 if (offset != filp->f_pos) {
612 filp->f_pos = offset;
613 ((struct nfs_open_context *)filp->private_data)->dir_cookie = 0;
616 mutex_unlock(&filp->f_dentry->d_inode->i_mutex);
621 * All directory operations under NFS are synchronous, so fsync()
622 * is a dummy operation.
624 int nfs_fsync_dir(struct file *filp, struct dentry *dentry, int datasync)
626 dfprintk(VFS, "NFS: fsync_dir(%s/%s) datasync %d\n",
627 dentry->d_parent->d_name.name, dentry->d_name.name,
634 * A check for whether or not the parent directory has changed.
635 * In the case it has, we assume that the dentries are untrustworthy
636 * and may need to be looked up again.
638 static inline int nfs_check_verifier(struct inode *dir, struct dentry *dentry)
642 if ((NFS_I(dir)->cache_validity & NFS_INO_INVALID_ATTR) != 0
643 || nfs_attribute_timeout(dir))
645 return nfs_verify_change_attribute(dir, (unsigned long)dentry->d_fsdata);
648 static inline void nfs_set_verifier(struct dentry * dentry, unsigned long verf)
650 dentry->d_fsdata = (void *)verf;
654 * Whenever an NFS operation succeeds, we know that the dentry
655 * is valid, so we update the revalidation timestamp.
657 static inline void nfs_renew_times(struct dentry * dentry)
659 dentry->d_time = jiffies;
663 * Return the intent data that applies to this particular path component
665 * Note that the current set of intents only apply to the very last
666 * component of the path.
667 * We check for this using LOOKUP_CONTINUE and LOOKUP_PARENT.
669 static inline unsigned int nfs_lookup_check_intent(struct nameidata *nd, unsigned int mask)
671 if (nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))
673 return nd->flags & mask;
677 * Inode and filehandle revalidation for lookups.
679 * We force revalidation in the cases where the VFS sets LOOKUP_REVAL,
680 * or if the intent information indicates that we're about to open this
681 * particular file and the "nocto" mount flag is not set.
685 int nfs_lookup_verify_inode(struct inode *inode, struct nameidata *nd)
687 struct nfs_server *server = NFS_SERVER(inode);
690 /* VFS wants an on-the-wire revalidation */
691 if (nd->flags & LOOKUP_REVAL)
693 /* This is an open(2) */
694 if (nfs_lookup_check_intent(nd, LOOKUP_OPEN) != 0 &&
695 !(server->flags & NFS_MOUNT_NOCTO))
698 return nfs_revalidate_inode(server, inode);
700 return __nfs_revalidate_inode(server, inode);
704 * We judge how long we want to trust negative
705 * dentries by looking at the parent inode mtime.
707 * If parent mtime has changed, we revalidate, else we wait for a
708 * period corresponding to the parent's attribute cache timeout value.
711 int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry,
712 struct nameidata *nd)
714 /* Don't revalidate a negative dentry if we're creating a new file */
715 if (nd != NULL && nfs_lookup_check_intent(nd, LOOKUP_CREATE) != 0)
717 return !nfs_check_verifier(dir, dentry);
721 * This is called every time the dcache has a lookup hit,
722 * and we should check whether we can really trust that
725 * NOTE! The hit can be a negative hit too, don't assume
728 * If the parent directory is seen to have changed, we throw out the
729 * cached dentry and do a new lookup.
731 static int nfs_lookup_revalidate(struct dentry * dentry, struct nameidata *nd)
735 struct dentry *parent;
737 struct nfs_fh fhandle;
738 struct nfs_fattr fattr;
739 unsigned long verifier;
741 parent = dget_parent(dentry);
743 dir = parent->d_inode;
744 nfs_inc_stats(dir, NFSIOS_DENTRYREVALIDATE);
745 inode = dentry->d_inode;
748 if (nfs_neg_need_reval(dir, dentry, nd))
753 if (is_bad_inode(inode)) {
754 dfprintk(LOOKUPCACHE, "%s: %s/%s has dud inode\n",
755 __FUNCTION__, dentry->d_parent->d_name.name,
756 dentry->d_name.name);
760 /* Revalidate parent directory attribute cache */
761 if (nfs_revalidate_inode(NFS_SERVER(dir), dir) < 0)
764 /* Force a full look up iff the parent directory has changed */
765 if (nfs_check_verifier(dir, dentry)) {
766 if (nfs_lookup_verify_inode(inode, nd))
771 if (NFS_STALE(inode))
774 verifier = nfs_save_change_attribute(dir);
775 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, &fhandle, &fattr);
778 if (nfs_compare_fh(NFS_FH(inode), &fhandle))
780 if ((error = nfs_refresh_inode(inode, &fattr)) != 0)
783 nfs_renew_times(dentry);
784 nfs_set_verifier(dentry, verifier);
788 dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is valid\n",
789 __FUNCTION__, dentry->d_parent->d_name.name,
790 dentry->d_name.name);
796 if (inode && S_ISDIR(inode->i_mode)) {
797 /* Purge readdir caches. */
798 nfs_zap_caches(inode);
799 /* If we have submounts, don't unhash ! */
800 if (have_submounts(dentry))
802 shrink_dcache_parent(dentry);
807 dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is invalid\n",
808 __FUNCTION__, dentry->d_parent->d_name.name,
809 dentry->d_name.name);
814 * This is called from dput() when d_count is going to 0.
816 static int nfs_dentry_delete(struct dentry *dentry)
818 dfprintk(VFS, "NFS: dentry_delete(%s/%s, %x)\n",
819 dentry->d_parent->d_name.name, dentry->d_name.name,
822 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
823 /* Unhash it, so that ->d_iput() would be called */
826 if (!(dentry->d_sb->s_flags & MS_ACTIVE)) {
827 /* Unhash it, so that ancestors of killed async unlink
828 * files will be cleaned up during umount */
836 * Called when the dentry loses inode.
837 * We use it to clean up silly-renamed files.
839 static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode)
841 nfs_inode_return_delegation(inode);
842 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
845 nfs_complete_unlink(dentry);
848 /* When creating a negative dentry, we want to renew d_time */
849 nfs_renew_times(dentry);
853 struct dentry_operations nfs_dentry_operations = {
854 .d_revalidate = nfs_lookup_revalidate,
855 .d_delete = nfs_dentry_delete,
856 .d_iput = nfs_dentry_iput,
860 * Use intent information to check whether or not we're going to do
861 * an O_EXCL create using this path component.
864 int nfs_is_exclusive_create(struct inode *dir, struct nameidata *nd)
866 if (NFS_PROTO(dir)->version == 2)
868 if (nd == NULL || nfs_lookup_check_intent(nd, LOOKUP_CREATE) == 0)
870 return (nd->intent.open.flags & O_EXCL) != 0;
873 static struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
876 struct inode *inode = NULL;
878 struct nfs_fh fhandle;
879 struct nfs_fattr fattr;
881 dfprintk(VFS, "NFS: lookup(%s/%s)\n",
882 dentry->d_parent->d_name.name, dentry->d_name.name);
883 nfs_inc_stats(dir, NFSIOS_VFSLOOKUP);
885 res = ERR_PTR(-ENAMETOOLONG);
886 if (dentry->d_name.len > NFS_SERVER(dir)->namelen)
889 res = ERR_PTR(-ENOMEM);
890 dentry->d_op = NFS_PROTO(dir)->dentry_ops;
894 /* If we're doing an exclusive create, optimize away the lookup */
895 if (nfs_is_exclusive_create(dir, nd))
898 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, &fhandle, &fattr);
899 if (error == -ENOENT)
902 res = ERR_PTR(error);
905 inode = nfs_fhget(dentry->d_sb, &fhandle, &fattr);
906 res = (struct dentry *)inode;
909 vx_propagate_xid(nd, inode);
911 res = d_add_unique(dentry, inode);
914 nfs_renew_times(dentry);
915 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
923 static int nfs_open_revalidate(struct dentry *, struct nameidata *);
925 struct dentry_operations nfs4_dentry_operations = {
926 .d_revalidate = nfs_open_revalidate,
927 .d_delete = nfs_dentry_delete,
928 .d_iput = nfs_dentry_iput,
932 * Use intent information to determine whether we need to substitute
933 * the NFSv4-style stateful OPEN for the LOOKUP call
935 static int is_atomic_open(struct inode *dir, struct nameidata *nd)
937 if (nd == NULL || nfs_lookup_check_intent(nd, LOOKUP_OPEN) == 0)
939 /* NFS does not (yet) have a stateful open for directories */
940 if (nd->flags & LOOKUP_DIRECTORY)
942 /* Are we trying to write to a read only partition? */
943 if ((IS_RDONLY(dir) || MNT_IS_RDONLY(nd->mnt)) &&
944 (nd->intent.open.flags & (O_CREAT|O_TRUNC|FMODE_WRITE)))
949 static struct dentry *nfs_atomic_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd)
951 struct dentry *res = NULL;
954 dfprintk(VFS, "NFS: atomic_lookup(%s/%ld), %s\n",
955 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
957 /* Check that we are indeed trying to open this file */
958 if (!is_atomic_open(dir, nd))
961 if (dentry->d_name.len > NFS_SERVER(dir)->namelen) {
962 res = ERR_PTR(-ENAMETOOLONG);
965 dentry->d_op = NFS_PROTO(dir)->dentry_ops;
967 /* Let vfs_create() deal with O_EXCL */
968 if (nd->intent.open.flags & O_EXCL) {
973 /* Open the file on the server */
975 /* Revalidate parent directory attribute cache */
976 error = nfs_revalidate_inode(NFS_SERVER(dir), dir);
978 res = ERR_PTR(error);
983 if (nd->intent.open.flags & O_CREAT) {
984 nfs_begin_data_update(dir);
985 res = nfs4_atomic_open(dir, dentry, nd);
986 nfs_end_data_update(dir);
988 res = nfs4_atomic_open(dir, dentry, nd);
991 error = PTR_ERR(res);
993 /* Make a negative dentry */
997 /* This turned out not to be a regular file */
1002 if (!(nd->intent.open.flags & O_NOFOLLOW))
1008 } else if (res != NULL)
1010 nfs_renew_times(dentry);
1011 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1015 return nfs_lookup(dir, dentry, nd);
1018 static int nfs_open_revalidate(struct dentry *dentry, struct nameidata *nd)
1020 struct dentry *parent = NULL;
1021 struct inode *inode = dentry->d_inode;
1023 unsigned long verifier;
1024 int openflags, ret = 0;
1026 parent = dget_parent(dentry);
1027 dir = parent->d_inode;
1028 if (!is_atomic_open(dir, nd))
1030 /* We can't create new files in nfs_open_revalidate(), so we
1031 * optimize away revalidation of negative dentries.
1035 /* NFS only supports OPEN on regular files */
1036 if (!S_ISREG(inode->i_mode))
1038 openflags = nd->intent.open.flags;
1039 /* We cannot do exclusive creation on a positive dentry */
1040 if ((openflags & (O_CREAT|O_EXCL)) == (O_CREAT|O_EXCL))
1042 /* We can't create new files, or truncate existing ones here */
1043 openflags &= ~(O_CREAT|O_TRUNC);
1046 * Note: we're not holding inode->i_mutex and so may be racing with
1047 * operations that change the directory. We therefore save the
1048 * change attribute *before* we do the RPC call.
1051 verifier = nfs_save_change_attribute(dir);
1052 ret = nfs4_open_revalidate(dir, dentry, openflags, nd);
1054 nfs_set_verifier(dentry, verifier);
1063 if (inode != NULL && nfs_have_delegation(inode, FMODE_READ))
1065 return nfs_lookup_revalidate(dentry, nd);
1067 #endif /* CONFIG_NFSV4 */
1069 static struct dentry *nfs_readdir_lookup(nfs_readdir_descriptor_t *desc)
1071 struct dentry *parent = desc->file->f_dentry;
1072 struct inode *dir = parent->d_inode;
1073 struct nfs_entry *entry = desc->entry;
1074 struct dentry *dentry, *alias;
1075 struct qstr name = {
1076 .name = entry->name,
1079 struct inode *inode;
1083 if (name.name[0] == '.' && name.name[1] == '.')
1084 return dget_parent(parent);
1087 if (name.name[0] == '.')
1088 return dget(parent);
1090 name.hash = full_name_hash(name.name, name.len);
1091 dentry = d_lookup(parent, &name);
1094 if (!desc->plus || !(entry->fattr->valid & NFS_ATTR_FATTR))
1096 /* Note: caller is already holding the dir->i_mutex! */
1097 dentry = d_alloc(parent, &name);
1100 dentry->d_op = NFS_PROTO(dir)->dentry_ops;
1101 inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr);
1102 if (IS_ERR(inode)) {
1106 alias = d_add_unique(dentry, inode);
1107 if (alias != NULL) {
1111 nfs_renew_times(dentry);
1112 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1117 * Code common to create, mkdir, and mknod.
1119 int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle,
1120 struct nfs_fattr *fattr)
1122 struct inode *inode;
1123 int error = -EACCES;
1125 /* We may have been initialized further down */
1126 if (dentry->d_inode)
1128 if (fhandle->size == 0) {
1129 struct inode *dir = dentry->d_parent->d_inode;
1130 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr);
1134 if (!(fattr->valid & NFS_ATTR_FATTR)) {
1135 struct nfs_server *server = NFS_SB(dentry->d_sb);
1136 error = server->rpc_ops->getattr(server, fhandle, fattr);
1140 inode = nfs_fhget(dentry->d_sb, fhandle, fattr);
1141 error = PTR_ERR(inode);
1144 d_instantiate(dentry, inode);
1152 * Following a failed create operation, we drop the dentry rather
1153 * than retain a negative dentry. This avoids a problem in the event
1154 * that the operation succeeded on the server, but an error in the
1155 * reply path made it appear to have failed.
1157 static int nfs_create(struct inode *dir, struct dentry *dentry, int mode,
1158 struct nameidata *nd)
1164 dfprintk(VFS, "NFS: create(%s/%ld), %s\n",
1165 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1167 attr.ia_mode = mode;
1168 attr.ia_valid = ATTR_MODE;
1170 if (nd && (nd->flags & LOOKUP_CREATE))
1171 open_flags = nd->intent.open.flags;
1174 nfs_begin_data_update(dir);
1175 error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags, nd);
1176 nfs_end_data_update(dir);
1179 nfs_renew_times(dentry);
1180 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1190 * See comments for nfs_proc_create regarding failed operations.
1193 nfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t rdev)
1198 dfprintk(VFS, "NFS: mknod(%s/%ld), %s\n",
1199 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1201 if (!new_valid_dev(rdev))
1204 attr.ia_mode = mode;
1205 attr.ia_valid = ATTR_MODE;
1208 nfs_begin_data_update(dir);
1209 status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev);
1210 nfs_end_data_update(dir);
1213 nfs_renew_times(dentry);
1214 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1224 * See comments for nfs_proc_create regarding failed operations.
1226 static int nfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1231 dfprintk(VFS, "NFS: mkdir(%s/%ld), %s\n",
1232 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1234 attr.ia_valid = ATTR_MODE;
1235 attr.ia_mode = mode | S_IFDIR;
1238 nfs_begin_data_update(dir);
1239 error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr);
1240 nfs_end_data_update(dir);
1243 nfs_renew_times(dentry);
1244 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1253 static int nfs_rmdir(struct inode *dir, struct dentry *dentry)
1257 dfprintk(VFS, "NFS: rmdir(%s/%ld), %s\n",
1258 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1261 nfs_begin_data_update(dir);
1262 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
1263 /* Ensure the VFS deletes this inode */
1264 if (error == 0 && dentry->d_inode != NULL)
1265 dentry->d_inode->i_nlink = 0;
1266 nfs_end_data_update(dir);
1272 static int nfs_sillyrename(struct inode *dir, struct dentry *dentry)
1274 static unsigned int sillycounter;
1275 const int i_inosize = sizeof(dir->i_ino)*2;
1276 const int countersize = sizeof(sillycounter)*2;
1277 const int slen = sizeof(".nfs") + i_inosize + countersize - 1;
1280 struct dentry *sdentry;
1283 dfprintk(VFS, "NFS: silly-rename(%s/%s, ct=%d)\n",
1284 dentry->d_parent->d_name.name, dentry->d_name.name,
1285 atomic_read(&dentry->d_count));
1286 nfs_inc_stats(dir, NFSIOS_SILLYRENAME);
1289 if (!dentry->d_inode)
1290 printk("NFS: silly-renaming %s/%s, negative dentry??\n",
1291 dentry->d_parent->d_name.name, dentry->d_name.name);
1294 * We don't allow a dentry to be silly-renamed twice.
1297 if (dentry->d_flags & DCACHE_NFSFS_RENAMED)
1300 sprintf(silly, ".nfs%*.*lx",
1301 i_inosize, i_inosize, dentry->d_inode->i_ino);
1303 /* Return delegation in anticipation of the rename */
1304 nfs_inode_return_delegation(dentry->d_inode);
1308 char *suffix = silly + slen - countersize;
1312 sprintf(suffix, "%*.*x", countersize, countersize, sillycounter);
1314 dfprintk(VFS, "NFS: trying to rename %s to %s\n",
1315 dentry->d_name.name, silly);
1317 sdentry = lookup_one_len(silly, dentry->d_parent, slen);
1319 * N.B. Better to return EBUSY here ... it could be
1320 * dangerous to delete the file while it's in use.
1322 if (IS_ERR(sdentry))
1324 } while(sdentry->d_inode != NULL); /* need negative lookup */
1326 qsilly.name = silly;
1327 qsilly.len = strlen(silly);
1328 nfs_begin_data_update(dir);
1329 if (dentry->d_inode) {
1330 nfs_begin_data_update(dentry->d_inode);
1331 error = NFS_PROTO(dir)->rename(dir, &dentry->d_name,
1333 nfs_mark_for_revalidate(dentry->d_inode);
1334 nfs_end_data_update(dentry->d_inode);
1336 error = NFS_PROTO(dir)->rename(dir, &dentry->d_name,
1338 nfs_end_data_update(dir);
1340 nfs_renew_times(dentry);
1341 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1342 d_move(dentry, sdentry);
1343 error = nfs_async_unlink(dentry);
1344 /* If we return 0 we don't unlink */
1352 * Remove a file after making sure there are no pending writes,
1353 * and after checking that the file has only one user.
1355 * We invalidate the attribute cache and free the inode prior to the operation
1356 * to avoid possible races if the server reuses the inode.
1358 static int nfs_safe_remove(struct dentry *dentry)
1360 struct inode *dir = dentry->d_parent->d_inode;
1361 struct inode *inode = dentry->d_inode;
1364 dfprintk(VFS, "NFS: safe_remove(%s/%s)\n",
1365 dentry->d_parent->d_name.name, dentry->d_name.name);
1367 /* If the dentry was sillyrenamed, we simply call d_delete() */
1368 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1373 nfs_begin_data_update(dir);
1374 if (inode != NULL) {
1375 nfs_inode_return_delegation(inode);
1376 nfs_begin_data_update(inode);
1377 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1378 /* The VFS may want to delete this inode */
1381 nfs_mark_for_revalidate(inode);
1382 nfs_end_data_update(inode);
1384 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1385 nfs_end_data_update(dir);
1390 /* We do silly rename. In case sillyrename() returns -EBUSY, the inode
1391 * belongs to an active ".nfs..." file and we return -EBUSY.
1393 * If sillyrename() returns 0, we do nothing, otherwise we unlink.
1395 static int nfs_unlink(struct inode *dir, struct dentry *dentry)
1398 int need_rehash = 0;
1400 dfprintk(VFS, "NFS: unlink(%s/%ld, %s)\n", dir->i_sb->s_id,
1401 dir->i_ino, dentry->d_name.name);
1404 spin_lock(&dcache_lock);
1405 spin_lock(&dentry->d_lock);
1406 if (atomic_read(&dentry->d_count) > 1) {
1407 spin_unlock(&dentry->d_lock);
1408 spin_unlock(&dcache_lock);
1409 error = nfs_sillyrename(dir, dentry);
1413 if (!d_unhashed(dentry)) {
1417 spin_unlock(&dentry->d_lock);
1418 spin_unlock(&dcache_lock);
1419 error = nfs_safe_remove(dentry);
1421 nfs_renew_times(dentry);
1422 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1423 } else if (need_rehash)
1430 nfs_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1433 struct nfs_fattr sym_attr;
1434 struct nfs_fh sym_fh;
1435 struct qstr qsymname;
1438 dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s)\n", dir->i_sb->s_id,
1439 dir->i_ino, dentry->d_name.name, symname);
1442 if (dentry->d_inode)
1443 printk("nfs_proc_symlink: %s/%s not negative!\n",
1444 dentry->d_parent->d_name.name, dentry->d_name.name);
1447 * Fill in the sattr for the call.
1448 * Note: SunOS 4.1.2 crashes if the mode isn't initialized!
1450 attr.ia_valid = ATTR_MODE;
1451 attr.ia_mode = S_IFLNK | S_IRWXUGO;
1453 qsymname.name = symname;
1454 qsymname.len = strlen(symname);
1457 nfs_begin_data_update(dir);
1458 error = NFS_PROTO(dir)->symlink(dir, &dentry->d_name, &qsymname,
1459 &attr, &sym_fh, &sym_attr);
1460 nfs_end_data_update(dir);
1462 error = nfs_instantiate(dentry, &sym_fh, &sym_attr);
1464 if (error == -EEXIST)
1465 printk("nfs_proc_symlink: %s/%s already exists??\n",
1466 dentry->d_parent->d_name.name, dentry->d_name.name);
1474 nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1476 struct inode *inode = old_dentry->d_inode;
1479 dfprintk(VFS, "NFS: link(%s/%s -> %s/%s)\n",
1480 old_dentry->d_parent->d_name.name, old_dentry->d_name.name,
1481 dentry->d_parent->d_name.name, dentry->d_name.name);
1484 nfs_begin_data_update(dir);
1485 nfs_begin_data_update(inode);
1486 error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name);
1488 atomic_inc(&inode->i_count);
1489 d_instantiate(dentry, inode);
1491 nfs_end_data_update(inode);
1492 nfs_end_data_update(dir);
1499 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a
1500 * different file handle for the same inode after a rename (e.g. when
1501 * moving to a different directory). A fail-safe method to do so would
1502 * be to look up old_dir/old_name, create a link to new_dir/new_name and
1503 * rename the old file using the sillyrename stuff. This way, the original
1504 * file in old_dir will go away when the last process iput()s the inode.
1508 * It actually works quite well. One needs to have the possibility for
1509 * at least one ".nfs..." file in each directory the file ever gets
1510 * moved or linked to which happens automagically with the new
1511 * implementation that only depends on the dcache stuff instead of
1512 * using the inode layer
1514 * Unfortunately, things are a little more complicated than indicated
1515 * above. For a cross-directory move, we want to make sure we can get
1516 * rid of the old inode after the operation. This means there must be
1517 * no pending writes (if it's a file), and the use count must be 1.
1518 * If these conditions are met, we can drop the dentries before doing
1521 static int nfs_rename(struct inode *old_dir, struct dentry *old_dentry,
1522 struct inode *new_dir, struct dentry *new_dentry)
1524 struct inode *old_inode = old_dentry->d_inode;
1525 struct inode *new_inode = new_dentry->d_inode;
1526 struct dentry *dentry = NULL, *rehash = NULL;
1530 * To prevent any new references to the target during the rename,
1531 * we unhash the dentry and free the inode in advance.
1534 if (!d_unhashed(new_dentry)) {
1536 rehash = new_dentry;
1539 dfprintk(VFS, "NFS: rename(%s/%s -> %s/%s, ct=%d)\n",
1540 old_dentry->d_parent->d_name.name, old_dentry->d_name.name,
1541 new_dentry->d_parent->d_name.name, new_dentry->d_name.name,
1542 atomic_read(&new_dentry->d_count));
1545 * First check whether the target is busy ... we can't
1546 * safely do _any_ rename if the target is in use.
1548 * For files, make a copy of the dentry and then do a
1549 * silly-rename. If the silly-rename succeeds, the
1550 * copied dentry is hashed and becomes the new target.
1554 if (S_ISDIR(new_inode->i_mode)) {
1556 if (!S_ISDIR(old_inode->i_mode))
1558 } else if (atomic_read(&new_dentry->d_count) > 2) {
1560 /* copy the target dentry's name */
1561 dentry = d_alloc(new_dentry->d_parent,
1562 &new_dentry->d_name);
1566 /* silly-rename the existing target ... */
1567 err = nfs_sillyrename(new_dir, new_dentry);
1569 new_dentry = rehash = dentry;
1571 /* instantiate the replacement target */
1572 d_instantiate(new_dentry, NULL);
1573 } else if (atomic_read(&new_dentry->d_count) > 1) {
1574 /* dentry still busy? */
1576 printk("nfs_rename: target %s/%s busy, d_count=%d\n",
1577 new_dentry->d_parent->d_name.name,
1578 new_dentry->d_name.name,
1579 atomic_read(&new_dentry->d_count));
1584 new_inode->i_nlink--;
1588 * ... prune child dentries and writebacks if needed.
1590 if (atomic_read(&old_dentry->d_count) > 1) {
1591 nfs_wb_all(old_inode);
1592 shrink_dcache_parent(old_dentry);
1594 nfs_inode_return_delegation(old_inode);
1596 if (new_inode != NULL) {
1597 nfs_inode_return_delegation(new_inode);
1598 d_delete(new_dentry);
1601 nfs_begin_data_update(old_dir);
1602 nfs_begin_data_update(new_dir);
1603 nfs_begin_data_update(old_inode);
1604 error = NFS_PROTO(old_dir)->rename(old_dir, &old_dentry->d_name,
1605 new_dir, &new_dentry->d_name);
1606 nfs_mark_for_revalidate(old_inode);
1607 nfs_end_data_update(old_inode);
1608 nfs_end_data_update(new_dir);
1609 nfs_end_data_update(old_dir);
1614 if (!S_ISDIR(old_inode->i_mode))
1615 d_move(old_dentry, new_dentry);
1616 nfs_renew_times(new_dentry);
1617 nfs_set_verifier(new_dentry, nfs_save_change_attribute(new_dir));
1620 /* new dentry created? */
1627 int nfs_access_get_cached(struct inode *inode, struct rpc_cred *cred, struct nfs_access_entry *res)
1629 struct nfs_inode *nfsi = NFS_I(inode);
1630 struct nfs_access_entry *cache = &nfsi->cache_access;
1632 if (cache->cred != cred
1633 || time_after(jiffies, cache->jiffies + NFS_ATTRTIMEO(inode))
1634 || (nfsi->cache_validity & NFS_INO_INVALID_ACCESS))
1636 memcpy(res, cache, sizeof(*res));
1640 void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set)
1642 struct nfs_inode *nfsi = NFS_I(inode);
1643 struct nfs_access_entry *cache = &nfsi->cache_access;
1645 if (cache->cred != set->cred) {
1647 put_rpccred(cache->cred);
1648 cache->cred = get_rpccred(set->cred);
1650 /* FIXME: replace current access_cache BKL reliance with inode->i_lock */
1651 spin_lock(&inode->i_lock);
1652 nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS;
1653 spin_unlock(&inode->i_lock);
1654 cache->jiffies = set->jiffies;
1655 cache->mask = set->mask;
1658 static int nfs_do_access(struct inode *inode, struct rpc_cred *cred, int mask)
1660 struct nfs_access_entry cache;
1663 status = nfs_access_get_cached(inode, cred, &cache);
1667 /* Be clever: ask server to check for all possible rights */
1668 cache.mask = MAY_EXEC | MAY_WRITE | MAY_READ;
1670 cache.jiffies = jiffies;
1671 status = NFS_PROTO(inode)->access(inode, &cache);
1674 nfs_access_add_cache(inode, &cache);
1676 if ((cache.mask & mask) == mask)
1681 int nfs_permission(struct inode *inode, int mask, struct nameidata *nd)
1683 struct rpc_cred *cred;
1686 nfs_inc_stats(inode, NFSIOS_VFSACCESS);
1690 /* Is this sys_access() ? */
1691 if (nd != NULL && (nd->flags & LOOKUP_ACCESS))
1694 switch (inode->i_mode & S_IFMT) {
1698 /* NFSv4 has atomic_open... */
1699 if (nfs_server_capable(inode, NFS_CAP_ATOMIC_OPEN)
1701 && (nd->flags & LOOKUP_OPEN))
1706 * Optimize away all write operations, since the server
1707 * will check permissions when we perform the op.
1709 if ((mask & MAY_WRITE) && !(mask & MAY_READ))
1716 if (!NFS_PROTO(inode)->access)
1719 cred = rpcauth_lookupcred(NFS_CLIENT(inode)->cl_auth, 0);
1720 if (!IS_ERR(cred)) {
1721 res = nfs_do_access(inode, cred, mask);
1724 res = PTR_ERR(cred);
1727 dfprintk(VFS, "NFS: permission(%s/%ld), mask=0x%x, res=%d\n",
1728 inode->i_sb->s_id, inode->i_ino, mask, res);
1731 res = nfs_revalidate_inode(NFS_SERVER(inode), inode);
1733 res = generic_permission(inode, mask, NULL);
1740 * version-control: t
1741 * kept-new-versions: 5