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);
55 struct file_operations nfs_dir_operations = {
56 .read = generic_read_dir,
57 .readdir = nfs_readdir,
59 .release = nfs_release,
60 .fsync = nfs_fsync_dir,
63 struct inode_operations nfs_dir_inode_operations = {
68 .symlink = nfs_symlink,
73 .permission = nfs_permission,
74 .getattr = nfs_getattr,
75 .setattr = nfs_setattr,
80 static struct dentry *nfs_atomic_lookup(struct inode *, struct dentry *, struct nameidata *);
81 struct inode_operations nfs4_dir_inode_operations = {
83 .lookup = nfs_atomic_lookup,
86 .symlink = nfs_symlink,
91 .permission = nfs_permission,
92 .getattr = nfs_getattr,
93 .setattr = nfs_setattr,
96 #endif /* CONFIG_NFS_V4 */
102 nfs_opendir(struct inode *inode, struct file *filp)
107 /* Call generic open code in order to cache credentials */
109 res = nfs_open(inode, filp);
114 typedef u32 * (*decode_dirent_t)(u32 *, struct nfs_entry *, int);
118 unsigned long page_index;
121 struct nfs_entry *entry;
122 decode_dirent_t decode;
125 } nfs_readdir_descriptor_t;
127 /* Now we cache directories properly, by stuffing the dirent
128 * data directly in the page cache.
130 * Inode invalidation due to refresh etc. takes care of
131 * _everything_, no sloppy entry flushing logic, no extraneous
132 * copying, network direct to page cache, the way it was meant
135 * NOTE: Dirent information verification is done always by the
136 * page-in of the RPC reply, nowhere else, this simplies
137 * things substantially.
140 int nfs_readdir_filler(nfs_readdir_descriptor_t *desc, struct page *page)
142 struct file *file = desc->file;
143 struct inode *inode = file->f_dentry->d_inode;
144 struct rpc_cred *cred = nfs_file_cred(file);
145 unsigned long timestamp;
148 dfprintk(VFS, "NFS: nfs_readdir_filler() reading cookie %Lu into page %lu.\n", (long long)desc->entry->cookie, page->index);
152 error = NFS_PROTO(inode)->readdir(file->f_dentry, cred, desc->entry->cookie, page,
153 NFS_SERVER(inode)->dtsize, desc->plus);
155 /* We requested READDIRPLUS, but the server doesn't grok it */
156 if (error == -ENOTSUPP && desc->plus) {
157 NFS_SERVER(inode)->caps &= ~NFS_CAP_READDIRPLUS;
158 NFS_FLAGS(inode) &= ~NFS_INO_ADVISE_RDPLUS;
164 SetPageUptodate(page);
165 NFS_FLAGS(inode) |= NFS_INO_INVALID_ATIME;
166 /* Ensure consistent page alignment of the data.
167 * Note: assumes we have exclusive access to this mapping either
168 * through inode->i_sem or some other mechanism.
170 if (page->index == 0)
171 invalidate_inode_pages2_range(inode->i_mapping, PAGE_CACHE_SIZE, -1);
177 nfs_zap_caches(inode);
183 int dir_decode(nfs_readdir_descriptor_t *desc)
186 p = desc->decode(p, desc->entry, desc->plus);
194 void dir_page_release(nfs_readdir_descriptor_t *desc)
197 page_cache_release(desc->page);
203 * Given a pointer to a buffer that has already been filled by a call
204 * to readdir, find the next entry.
206 * If the end of the buffer has been reached, return -EAGAIN, if not,
207 * return the offset within the buffer of the next entry to be
211 int find_dirent(nfs_readdir_descriptor_t *desc, struct page *page)
213 struct nfs_entry *entry = desc->entry;
217 while((status = dir_decode(desc)) == 0) {
218 dfprintk(VFS, "NFS: found cookie %Lu\n", (long long)entry->cookie);
219 if (entry->prev_cookie == desc->target)
221 if (loop_count++ > 200) {
226 dfprintk(VFS, "NFS: find_dirent() returns %d\n", status);
231 * Find the given page, and call find_dirent() in order to try to
232 * return the next entry.
235 int find_dirent_page(nfs_readdir_descriptor_t *desc)
237 struct inode *inode = desc->file->f_dentry->d_inode;
241 dfprintk(VFS, "NFS: find_dirent_page() searching directory page %ld\n", desc->page_index);
243 page = read_cache_page(inode->i_mapping, desc->page_index,
244 (filler_t *)nfs_readdir_filler, desc);
246 status = PTR_ERR(page);
249 if (!PageUptodate(page))
252 /* NOTE: Someone else may have changed the READDIRPLUS flag */
254 desc->ptr = kmap(page); /* matching kunmap in nfs_do_filldir */
255 status = find_dirent(desc, page);
257 dir_page_release(desc);
259 dfprintk(VFS, "NFS: find_dirent_page() returns %d\n", status);
262 page_cache_release(page);
267 * Recurse through the page cache pages, and return a
268 * filled nfs_entry structure of the next directory entry if possible.
270 * The target for the search is 'desc->target'.
273 int readdir_search_pagecache(nfs_readdir_descriptor_t *desc)
278 dfprintk(VFS, "NFS: readdir_search_pagecache() searching for cookie %Lu\n", (long long)desc->target);
280 res = find_dirent_page(desc);
283 /* Align to beginning of next page */
285 if (loop_count++ > 200) {
290 dfprintk(VFS, "NFS: readdir_search_pagecache() returned %d\n", res);
294 static inline unsigned int dt_type(struct inode *inode)
296 return (inode->i_mode >> 12) & 15;
299 static struct dentry *nfs_readdir_lookup(nfs_readdir_descriptor_t *desc);
302 * Once we've found the start of the dirent within a page: fill 'er up...
305 int nfs_do_filldir(nfs_readdir_descriptor_t *desc, void *dirent,
308 struct file *file = desc->file;
309 struct nfs_entry *entry = desc->entry;
310 struct dentry *dentry = NULL;
311 unsigned long fileid;
315 dfprintk(VFS, "NFS: nfs_do_filldir() filling starting @ cookie %Lu\n", (long long)desc->target);
318 unsigned d_type = DT_UNKNOWN;
319 /* Note: entry->prev_cookie contains the cookie for
320 * retrieving the current dirent on the server */
321 fileid = nfs_fileid_to_ino_t(entry->ino);
323 /* Get a dentry if we have one */
326 dentry = nfs_readdir_lookup(desc);
328 /* Use readdirplus info */
329 if (dentry != NULL && dentry->d_inode != NULL) {
330 d_type = dt_type(dentry->d_inode);
331 fileid = dentry->d_inode->i_ino;
334 res = filldir(dirent, entry->name, entry->len,
335 entry->prev_cookie, fileid, d_type);
338 file->f_pos = desc->target = entry->cookie;
339 if (dir_decode(desc) != 0) {
343 if (loop_count++ > 200) {
348 dir_page_release(desc);
351 dfprintk(VFS, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n", (long long)desc->target, res);
356 * If we cannot find a cookie in our cache, we suspect that this is
357 * because it points to a deleted file, so we ask the server to return
358 * whatever it thinks is the next entry. We then feed this to filldir.
359 * If all goes well, we should then be able to find our way round the
360 * cache on the next call to readdir_search_pagecache();
362 * NOTE: we cannot add the anonymous page to the pagecache because
363 * the data it contains might not be page aligned. Besides,
364 * we should already have a complete representation of the
365 * directory in the page cache by the time we get here.
368 int uncached_readdir(nfs_readdir_descriptor_t *desc, void *dirent,
371 struct file *file = desc->file;
372 struct inode *inode = file->f_dentry->d_inode;
373 struct rpc_cred *cred = nfs_file_cred(file);
374 struct page *page = NULL;
377 dfprintk(VFS, "NFS: uncached_readdir() searching for cookie %Lu\n", (long long)desc->target);
379 page = alloc_page(GFP_HIGHUSER);
384 desc->error = NFS_PROTO(inode)->readdir(file->f_dentry, cred, desc->target,
386 NFS_SERVER(inode)->dtsize,
388 NFS_FLAGS(inode) |= NFS_INO_INVALID_ATIME;
390 desc->ptr = kmap(page); /* matching kunmap in nfs_do_filldir */
391 if (desc->error >= 0) {
392 if ((status = dir_decode(desc)) == 0)
393 desc->entry->prev_cookie = desc->target;
399 status = nfs_do_filldir(desc, dirent, filldir);
401 /* Reset read descriptor so it searches the page cache from
402 * the start upon the next call to readdir_search_pagecache() */
403 desc->page_index = 0;
404 desc->entry->cookie = desc->entry->prev_cookie = 0;
405 desc->entry->eof = 0;
407 dfprintk(VFS, "NFS: uncached_readdir() returns %d\n", status);
410 dir_page_release(desc);
414 /* The file offset position is now represented as a true offset into the
415 * page cache as is the case in most of the other filesystems.
417 static int nfs_readdir(struct file *filp, void *dirent, filldir_t filldir)
419 struct dentry *dentry = filp->f_dentry;
420 struct inode *inode = dentry->d_inode;
421 nfs_readdir_descriptor_t my_desc,
423 struct nfs_entry my_entry;
425 struct nfs_fattr fattr;
430 res = nfs_revalidate_inode(NFS_SERVER(inode), inode);
437 * filp->f_pos points to the file offset in the page cache.
438 * but if the cache has meanwhile been zapped, we need to
439 * read from the last dirent to revalidate f_pos
442 memset(desc, 0, sizeof(*desc));
445 desc->target = filp->f_pos;
446 desc->decode = NFS_PROTO(inode)->decode_dirent;
447 desc->plus = NFS_USE_READDIRPLUS(inode);
449 my_entry.cookie = my_entry.prev_cookie = 0;
452 my_entry.fattr = &fattr;
453 desc->entry = &my_entry;
455 while(!desc->entry->eof) {
456 res = readdir_search_pagecache(desc);
457 if (res == -EBADCOOKIE) {
458 /* This means either end of directory */
459 if (desc->entry->cookie != desc->target) {
460 /* Or that the server has 'lost' a cookie */
461 res = uncached_readdir(desc, dirent, filldir);
468 if (res == -ETOOSMALL && desc->plus) {
469 NFS_FLAGS(inode) &= ~NFS_INO_ADVISE_RDPLUS;
470 nfs_zap_caches(inode);
472 desc->entry->eof = 0;
478 res = nfs_do_filldir(desc, dirent, filldir);
493 * All directory operations under NFS are synchronous, so fsync()
494 * is a dummy operation.
496 int nfs_fsync_dir(struct file *filp, struct dentry *dentry, int datasync)
502 * A check for whether or not the parent directory has changed.
503 * In the case it has, we assume that the dentries are untrustworthy
504 * and may need to be looked up again.
506 static inline int nfs_check_verifier(struct inode *dir, struct dentry *dentry)
510 if ((NFS_FLAGS(dir) & NFS_INO_INVALID_ATTR) != 0
511 || nfs_attribute_timeout(dir))
513 return nfs_verify_change_attribute(dir, (unsigned long)dentry->d_fsdata);
516 static inline void nfs_set_verifier(struct dentry * dentry, unsigned long verf)
518 dentry->d_fsdata = (void *)verf;
522 * Whenever an NFS operation succeeds, we know that the dentry
523 * is valid, so we update the revalidation timestamp.
525 static inline void nfs_renew_times(struct dentry * dentry)
527 dentry->d_time = jiffies;
531 * Return the intent data that applies to this particular path component
533 * Note that the current set of intents only apply to the very last
534 * component of the path.
535 * We check for this using LOOKUP_CONTINUE and LOOKUP_PARENT.
537 static inline unsigned int nfs_lookup_check_intent(struct nameidata *nd, unsigned int mask)
539 if (nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))
541 return nd->flags & mask;
545 * Inode and filehandle revalidation for lookups.
547 * We force revalidation in the cases where the VFS sets LOOKUP_REVAL,
548 * or if the intent information indicates that we're about to open this
549 * particular file and the "nocto" mount flag is not set.
553 int nfs_lookup_verify_inode(struct inode *inode, struct nameidata *nd)
555 struct nfs_server *server = NFS_SERVER(inode);
558 /* VFS wants an on-the-wire revalidation */
559 if (nd->flags & LOOKUP_REVAL)
561 /* This is an open(2) */
562 if (nfs_lookup_check_intent(nd, LOOKUP_OPEN) != 0 &&
563 !(server->flags & NFS_MOUNT_NOCTO))
566 return nfs_revalidate_inode(server, inode);
568 return __nfs_revalidate_inode(server, inode);
572 * We judge how long we want to trust negative
573 * dentries by looking at the parent inode mtime.
575 * If parent mtime has changed, we revalidate, else we wait for a
576 * period corresponding to the parent's attribute cache timeout value.
579 int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry,
580 struct nameidata *nd)
582 /* Don't revalidate a negative dentry if we're creating a new file */
583 if (nd != NULL && nfs_lookup_check_intent(nd, LOOKUP_CREATE) != 0)
585 return !nfs_check_verifier(dir, dentry);
589 * This is called every time the dcache has a lookup hit,
590 * and we should check whether we can really trust that
593 * NOTE! The hit can be a negative hit too, don't assume
596 * If the parent directory is seen to have changed, we throw out the
597 * cached dentry and do a new lookup.
599 static int nfs_lookup_revalidate(struct dentry * dentry, struct nameidata *nd)
603 struct dentry *parent;
605 struct nfs_fh fhandle;
606 struct nfs_fattr fattr;
607 unsigned long verifier;
609 parent = dget_parent(dentry);
611 dir = parent->d_inode;
612 inode = dentry->d_inode;
615 if (nfs_neg_need_reval(dir, dentry, nd))
620 if (is_bad_inode(inode)) {
621 dfprintk(VFS, "nfs_lookup_validate: %s/%s has dud inode\n",
622 dentry->d_parent->d_name.name, dentry->d_name.name);
626 /* Revalidate parent directory attribute cache */
627 if (nfs_revalidate_inode(NFS_SERVER(dir), dir) < 0)
630 /* Force a full look up iff the parent directory has changed */
631 if (nfs_check_verifier(dir, dentry)) {
632 if (nfs_lookup_verify_inode(inode, nd))
637 if (NFS_STALE(inode))
640 verifier = nfs_save_change_attribute(dir);
641 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, &fhandle, &fattr);
644 if (nfs_compare_fh(NFS_FH(inode), &fhandle))
646 if ((error = nfs_refresh_inode(inode, &fattr)) != 0)
649 nfs_renew_times(dentry);
650 nfs_set_verifier(dentry, verifier);
659 if (inode && S_ISDIR(inode->i_mode)) {
660 /* Purge readdir caches. */
661 nfs_zap_caches(inode);
662 /* If we have submounts, don't unhash ! */
663 if (have_submounts(dentry))
665 shrink_dcache_parent(dentry);
674 * This is called from dput() when d_count is going to 0.
676 static int nfs_dentry_delete(struct dentry *dentry)
678 dfprintk(VFS, "NFS: dentry_delete(%s/%s, %x)\n",
679 dentry->d_parent->d_name.name, dentry->d_name.name,
682 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
683 /* Unhash it, so that ->d_iput() would be called */
686 if (!(dentry->d_sb->s_flags & MS_ACTIVE)) {
687 /* Unhash it, so that ancestors of killed async unlink
688 * files will be cleaned up during umount */
696 * Called when the dentry loses inode.
697 * We use it to clean up silly-renamed files.
699 static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode)
701 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
704 nfs_complete_unlink(dentry);
707 /* When creating a negative dentry, we want to renew d_time */
708 nfs_renew_times(dentry);
712 struct dentry_operations nfs_dentry_operations = {
713 .d_revalidate = nfs_lookup_revalidate,
714 .d_delete = nfs_dentry_delete,
715 .d_iput = nfs_dentry_iput,
719 * Use intent information to check whether or not we're going to do
720 * an O_EXCL create using this path component.
723 int nfs_is_exclusive_create(struct inode *dir, struct nameidata *nd)
725 if (NFS_PROTO(dir)->version == 2)
727 if (nd == NULL || nfs_lookup_check_intent(nd, LOOKUP_CREATE) == 0)
729 return (nd->intent.open.flags & O_EXCL) != 0;
732 static struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
735 struct inode *inode = NULL;
737 struct nfs_fh fhandle;
738 struct nfs_fattr fattr;
740 dfprintk(VFS, "NFS: lookup(%s/%s)\n",
741 dentry->d_parent->d_name.name, dentry->d_name.name);
743 res = ERR_PTR(-ENAMETOOLONG);
744 if (dentry->d_name.len > NFS_SERVER(dir)->namelen)
747 res = ERR_PTR(-ENOMEM);
748 dentry->d_op = NFS_PROTO(dir)->dentry_ops;
751 /* Revalidate parent directory attribute cache */
752 error = nfs_revalidate_inode(NFS_SERVER(dir), dir);
754 res = ERR_PTR(error);
758 /* If we're doing an exclusive create, optimize away the lookup */
759 if (nfs_is_exclusive_create(dir, nd))
762 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, &fhandle, &fattr);
763 if (error == -ENOENT)
766 res = ERR_PTR(error);
769 res = ERR_PTR(-EACCES);
770 inode = nfs_fhget(dentry->d_sb, &fhandle, &fattr);
774 res = d_add_unique(dentry, inode);
777 nfs_renew_times(dentry);
778 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
786 static int nfs_open_revalidate(struct dentry *, struct nameidata *);
788 struct dentry_operations nfs4_dentry_operations = {
789 .d_revalidate = nfs_open_revalidate,
790 .d_delete = nfs_dentry_delete,
791 .d_iput = nfs_dentry_iput,
795 * Use intent information to determine whether we need to substitute
796 * the NFSv4-style stateful OPEN for the LOOKUP call
798 static int is_atomic_open(struct inode *dir, struct nameidata *nd)
800 if (nd == NULL || nfs_lookup_check_intent(nd, LOOKUP_OPEN) == 0)
802 /* NFS does not (yet) have a stateful open for directories */
803 if (nd->flags & LOOKUP_DIRECTORY)
805 /* Are we trying to write to a read only partition? */
806 if (IS_RDONLY(dir) && (nd->intent.open.flags & (O_CREAT|O_TRUNC|FMODE_WRITE)))
811 static struct dentry *nfs_atomic_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd)
813 struct dentry *res = NULL;
814 struct inode *inode = NULL;
817 /* Check that we are indeed trying to open this file */
818 if (!is_atomic_open(dir, nd))
821 if (dentry->d_name.len > NFS_SERVER(dir)->namelen) {
822 res = ERR_PTR(-ENAMETOOLONG);
825 dentry->d_op = NFS_PROTO(dir)->dentry_ops;
827 /* Let vfs_create() deal with O_EXCL */
828 if (nd->intent.open.flags & O_EXCL)
831 /* Open the file on the server */
833 /* Revalidate parent directory attribute cache */
834 error = nfs_revalidate_inode(NFS_SERVER(dir), dir);
836 res = ERR_PTR(error);
840 if (nd->intent.open.flags & O_CREAT) {
841 nfs_begin_data_update(dir);
842 inode = nfs4_atomic_open(dir, dentry, nd);
843 nfs_end_data_update(dir);
845 inode = nfs4_atomic_open(dir, dentry, nd);
848 error = PTR_ERR(inode);
850 /* Make a negative dentry */
854 /* This turned out not to be a regular file */
856 if (!(nd->intent.open.flags & O_NOFOLLOW))
861 res = ERR_PTR(error);
866 res = d_add_unique(dentry, inode);
869 nfs_renew_times(dentry);
870 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
874 return nfs_lookup(dir, dentry, nd);
877 static int nfs_open_revalidate(struct dentry *dentry, struct nameidata *nd)
879 struct dentry *parent = NULL;
880 struct inode *inode = dentry->d_inode;
882 unsigned long verifier;
883 int openflags, ret = 0;
885 parent = dget_parent(dentry);
886 dir = parent->d_inode;
887 if (!is_atomic_open(dir, nd))
889 /* We can't create new files in nfs_open_revalidate(), so we
890 * optimize away revalidation of negative dentries.
894 /* NFS only supports OPEN on regular files */
895 if (!S_ISREG(inode->i_mode))
897 openflags = nd->intent.open.flags;
898 /* We cannot do exclusive creation on a positive dentry */
899 if ((openflags & (O_CREAT|O_EXCL)) == (O_CREAT|O_EXCL))
901 /* We can't create new files, or truncate existing ones here */
902 openflags &= ~(O_CREAT|O_TRUNC);
905 * Note: we're not holding inode->i_sem and so may be racing with
906 * operations that change the directory. We therefore save the
907 * change attribute *before* we do the RPC call.
910 verifier = nfs_save_change_attribute(dir);
911 ret = nfs4_open_revalidate(dir, dentry, openflags);
913 nfs_set_verifier(dentry, verifier);
922 if (inode != NULL && nfs_have_delegation(inode, FMODE_READ))
924 return nfs_lookup_revalidate(dentry, nd);
926 #endif /* CONFIG_NFSV4 */
928 static struct dentry *nfs_readdir_lookup(nfs_readdir_descriptor_t *desc)
930 struct dentry *parent = desc->file->f_dentry;
931 struct inode *dir = parent->d_inode;
932 struct nfs_entry *entry = desc->entry;
933 struct dentry *dentry, *alias;
942 if (name.name[0] == '.' && name.name[1] == '.')
943 return dget_parent(parent);
946 if (name.name[0] == '.')
949 name.hash = full_name_hash(name.name, name.len);
950 dentry = d_lookup(parent, &name);
953 if (!desc->plus || !(entry->fattr->valid & NFS_ATTR_FATTR))
955 /* Note: caller is already holding the dir->i_sem! */
956 dentry = d_alloc(parent, &name);
959 dentry->d_op = NFS_PROTO(dir)->dentry_ops;
960 inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr);
965 alias = d_add_unique(dentry, inode);
970 nfs_renew_times(dentry);
971 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
976 * Code common to create, mkdir, and mknod.
978 int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle,
979 struct nfs_fattr *fattr)
984 /* We may have been initialized further down */
987 if (fhandle->size == 0) {
988 struct inode *dir = dentry->d_parent->d_inode;
989 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr);
993 if (!(fattr->valid & NFS_ATTR_FATTR)) {
994 struct nfs_server *server = NFS_SB(dentry->d_sb);
995 error = server->rpc_ops->getattr(server, fhandle, fattr);
1000 inode = nfs_fhget(dentry->d_sb, fhandle, fattr);
1003 d_instantiate(dentry, inode);
1011 * Following a failed create operation, we drop the dentry rather
1012 * than retain a negative dentry. This avoids a problem in the event
1013 * that the operation succeeded on the server, but an error in the
1014 * reply path made it appear to have failed.
1016 static int nfs_create(struct inode *dir, struct dentry *dentry, int mode,
1017 struct nameidata *nd)
1023 dfprintk(VFS, "NFS: create(%s/%ld, %s\n", dir->i_sb->s_id,
1024 dir->i_ino, dentry->d_name.name);
1026 attr.ia_mode = mode;
1027 attr.ia_valid = ATTR_MODE;
1029 if (nd && (nd->flags & LOOKUP_CREATE))
1030 open_flags = nd->intent.open.flags;
1033 nfs_begin_data_update(dir);
1034 error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags);
1035 nfs_end_data_update(dir);
1038 nfs_renew_times(dentry);
1039 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1049 * See comments for nfs_proc_create regarding failed operations.
1052 nfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t rdev)
1057 dfprintk(VFS, "NFS: mknod(%s/%ld, %s\n", dir->i_sb->s_id,
1058 dir->i_ino, dentry->d_name.name);
1060 if (!new_valid_dev(rdev))
1063 attr.ia_mode = mode;
1064 attr.ia_valid = ATTR_MODE;
1067 nfs_begin_data_update(dir);
1068 status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev);
1069 nfs_end_data_update(dir);
1072 nfs_renew_times(dentry);
1073 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1083 * See comments for nfs_proc_create regarding failed operations.
1085 static int nfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1090 dfprintk(VFS, "NFS: mkdir(%s/%ld, %s\n", dir->i_sb->s_id,
1091 dir->i_ino, dentry->d_name.name);
1093 attr.ia_valid = ATTR_MODE;
1094 attr.ia_mode = mode | S_IFDIR;
1097 nfs_begin_data_update(dir);
1098 error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr);
1099 nfs_end_data_update(dir);
1102 nfs_renew_times(dentry);
1103 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1112 static int nfs_rmdir(struct inode *dir, struct dentry *dentry)
1116 dfprintk(VFS, "NFS: rmdir(%s/%ld, %s\n", dir->i_sb->s_id,
1117 dir->i_ino, dentry->d_name.name);
1120 nfs_begin_data_update(dir);
1121 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
1122 /* Ensure the VFS deletes this inode */
1123 if (error == 0 && dentry->d_inode != NULL)
1124 dentry->d_inode->i_nlink = 0;
1125 nfs_end_data_update(dir);
1131 static int nfs_sillyrename(struct inode *dir, struct dentry *dentry)
1133 static unsigned int sillycounter;
1134 const int i_inosize = sizeof(dir->i_ino)*2;
1135 const int countersize = sizeof(sillycounter)*2;
1136 const int slen = sizeof(".nfs") + i_inosize + countersize - 1;
1139 struct dentry *sdentry;
1142 dfprintk(VFS, "NFS: silly-rename(%s/%s, ct=%d)\n",
1143 dentry->d_parent->d_name.name, dentry->d_name.name,
1144 atomic_read(&dentry->d_count));
1147 if (!dentry->d_inode)
1148 printk("NFS: silly-renaming %s/%s, negative dentry??\n",
1149 dentry->d_parent->d_name.name, dentry->d_name.name);
1152 * We don't allow a dentry to be silly-renamed twice.
1155 if (dentry->d_flags & DCACHE_NFSFS_RENAMED)
1158 sprintf(silly, ".nfs%*.*lx",
1159 i_inosize, i_inosize, dentry->d_inode->i_ino);
1163 char *suffix = silly + slen - countersize;
1167 sprintf(suffix, "%*.*x", countersize, countersize, sillycounter);
1169 dfprintk(VFS, "trying to rename %s to %s\n",
1170 dentry->d_name.name, silly);
1172 sdentry = lookup_one_len(silly, dentry->d_parent, slen);
1174 * N.B. Better to return EBUSY here ... it could be
1175 * dangerous to delete the file while it's in use.
1177 if (IS_ERR(sdentry))
1179 } while(sdentry->d_inode != NULL); /* need negative lookup */
1181 qsilly.name = silly;
1182 qsilly.len = strlen(silly);
1183 nfs_begin_data_update(dir);
1184 if (dentry->d_inode) {
1185 nfs_begin_data_update(dentry->d_inode);
1186 error = NFS_PROTO(dir)->rename(dir, &dentry->d_name,
1188 nfs_end_data_update(dentry->d_inode);
1190 error = NFS_PROTO(dir)->rename(dir, &dentry->d_name,
1192 nfs_end_data_update(dir);
1194 nfs_renew_times(dentry);
1195 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1196 d_move(dentry, sdentry);
1197 error = nfs_async_unlink(dentry);
1198 /* If we return 0 we don't unlink */
1206 * Remove a file after making sure there are no pending writes,
1207 * and after checking that the file has only one user.
1209 * We invalidate the attribute cache and free the inode prior to the operation
1210 * to avoid possible races if the server reuses the inode.
1212 static int nfs_safe_remove(struct dentry *dentry)
1214 struct inode *dir = dentry->d_parent->d_inode;
1215 struct inode *inode = dentry->d_inode;
1218 dfprintk(VFS, "NFS: safe_remove(%s/%s)\n",
1219 dentry->d_parent->d_name.name, dentry->d_name.name);
1221 /* If the dentry was sillyrenamed, we simply call d_delete() */
1222 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1227 nfs_begin_data_update(dir);
1228 if (inode != NULL) {
1229 nfs_begin_data_update(inode);
1230 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1231 /* The VFS may want to delete this inode */
1234 nfs_end_data_update(inode);
1236 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1237 nfs_end_data_update(dir);
1242 /* We do silly rename. In case sillyrename() returns -EBUSY, the inode
1243 * belongs to an active ".nfs..." file and we return -EBUSY.
1245 * If sillyrename() returns 0, we do nothing, otherwise we unlink.
1247 static int nfs_unlink(struct inode *dir, struct dentry *dentry)
1250 int need_rehash = 0;
1252 dfprintk(VFS, "NFS: unlink(%s/%ld, %s)\n", dir->i_sb->s_id,
1253 dir->i_ino, dentry->d_name.name);
1256 spin_lock(&dcache_lock);
1257 spin_lock(&dentry->d_lock);
1258 if (atomic_read(&dentry->d_count) > 1) {
1259 spin_unlock(&dentry->d_lock);
1260 spin_unlock(&dcache_lock);
1261 error = nfs_sillyrename(dir, dentry);
1265 if (!d_unhashed(dentry)) {
1269 spin_unlock(&dentry->d_lock);
1270 spin_unlock(&dcache_lock);
1271 error = nfs_safe_remove(dentry);
1273 nfs_renew_times(dentry);
1274 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1275 } else if (need_rehash)
1282 nfs_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1285 struct nfs_fattr sym_attr;
1286 struct nfs_fh sym_fh;
1287 struct qstr qsymname;
1290 dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s)\n", dir->i_sb->s_id,
1291 dir->i_ino, dentry->d_name.name, symname);
1294 if (dentry->d_inode)
1295 printk("nfs_proc_symlink: %s/%s not negative!\n",
1296 dentry->d_parent->d_name.name, dentry->d_name.name);
1299 * Fill in the sattr for the call.
1300 * Note: SunOS 4.1.2 crashes if the mode isn't initialized!
1302 attr.ia_valid = ATTR_MODE;
1303 attr.ia_mode = S_IFLNK | S_IRWXUGO;
1305 qsymname.name = symname;
1306 qsymname.len = strlen(symname);
1309 nfs_begin_data_update(dir);
1310 error = NFS_PROTO(dir)->symlink(dir, &dentry->d_name, &qsymname,
1311 &attr, &sym_fh, &sym_attr);
1312 nfs_end_data_update(dir);
1314 error = nfs_instantiate(dentry, &sym_fh, &sym_attr);
1316 if (error == -EEXIST)
1317 printk("nfs_proc_symlink: %s/%s already exists??\n",
1318 dentry->d_parent->d_name.name, dentry->d_name.name);
1326 nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1328 struct inode *inode = old_dentry->d_inode;
1331 dfprintk(VFS, "NFS: link(%s/%s -> %s/%s)\n",
1332 old_dentry->d_parent->d_name.name, old_dentry->d_name.name,
1333 dentry->d_parent->d_name.name, dentry->d_name.name);
1336 * Drop the dentry in advance to force a new lookup.
1337 * Since nfs_proc_link doesn't return a file handle,
1338 * we can't use the existing dentry.
1343 nfs_begin_data_update(dir);
1344 nfs_begin_data_update(inode);
1345 error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name);
1346 nfs_end_data_update(inode);
1347 nfs_end_data_update(dir);
1354 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a
1355 * different file handle for the same inode after a rename (e.g. when
1356 * moving to a different directory). A fail-safe method to do so would
1357 * be to look up old_dir/old_name, create a link to new_dir/new_name and
1358 * rename the old file using the sillyrename stuff. This way, the original
1359 * file in old_dir will go away when the last process iput()s the inode.
1363 * It actually works quite well. One needs to have the possibility for
1364 * at least one ".nfs..." file in each directory the file ever gets
1365 * moved or linked to which happens automagically with the new
1366 * implementation that only depends on the dcache stuff instead of
1367 * using the inode layer
1369 * Unfortunately, things are a little more complicated than indicated
1370 * above. For a cross-directory move, we want to make sure we can get
1371 * rid of the old inode after the operation. This means there must be
1372 * no pending writes (if it's a file), and the use count must be 1.
1373 * If these conditions are met, we can drop the dentries before doing
1376 static int nfs_rename(struct inode *old_dir, struct dentry *old_dentry,
1377 struct inode *new_dir, struct dentry *new_dentry)
1379 struct inode *old_inode = old_dentry->d_inode;
1380 struct inode *new_inode = new_dentry->d_inode;
1381 struct dentry *dentry = NULL, *rehash = NULL;
1385 * To prevent any new references to the target during the rename,
1386 * we unhash the dentry and free the inode in advance.
1389 if (!d_unhashed(new_dentry)) {
1391 rehash = new_dentry;
1394 dfprintk(VFS, "NFS: rename(%s/%s -> %s/%s, ct=%d)\n",
1395 old_dentry->d_parent->d_name.name, old_dentry->d_name.name,
1396 new_dentry->d_parent->d_name.name, new_dentry->d_name.name,
1397 atomic_read(&new_dentry->d_count));
1400 * First check whether the target is busy ... we can't
1401 * safely do _any_ rename if the target is in use.
1403 * For files, make a copy of the dentry and then do a
1404 * silly-rename. If the silly-rename succeeds, the
1405 * copied dentry is hashed and becomes the new target.
1409 if (S_ISDIR(new_inode->i_mode))
1411 else if (atomic_read(&new_dentry->d_count) > 2) {
1413 /* copy the target dentry's name */
1414 dentry = d_alloc(new_dentry->d_parent,
1415 &new_dentry->d_name);
1419 /* silly-rename the existing target ... */
1420 err = nfs_sillyrename(new_dir, new_dentry);
1422 new_dentry = rehash = dentry;
1424 /* instantiate the replacement target */
1425 d_instantiate(new_dentry, NULL);
1426 } else if (atomic_read(&new_dentry->d_count) > 1) {
1427 /* dentry still busy? */
1429 printk("nfs_rename: target %s/%s busy, d_count=%d\n",
1430 new_dentry->d_parent->d_name.name,
1431 new_dentry->d_name.name,
1432 atomic_read(&new_dentry->d_count));
1440 * ... prune child dentries and writebacks if needed.
1442 if (atomic_read(&old_dentry->d_count) > 1) {
1443 nfs_wb_all(old_inode);
1444 shrink_dcache_parent(old_dentry);
1448 d_delete(new_dentry);
1450 nfs_begin_data_update(old_dir);
1451 nfs_begin_data_update(new_dir);
1452 nfs_begin_data_update(old_inode);
1453 error = NFS_PROTO(old_dir)->rename(old_dir, &old_dentry->d_name,
1454 new_dir, &new_dentry->d_name);
1455 nfs_end_data_update(old_inode);
1456 nfs_end_data_update(new_dir);
1457 nfs_end_data_update(old_dir);
1462 if (!S_ISDIR(old_inode->i_mode))
1463 d_move(old_dentry, new_dentry);
1464 nfs_renew_times(new_dentry);
1465 nfs_set_verifier(new_dentry, nfs_save_change_attribute(new_dir));
1468 /* new dentry created? */
1475 int nfs_access_get_cached(struct inode *inode, struct rpc_cred *cred, struct nfs_access_entry *res)
1477 struct nfs_access_entry *cache = &NFS_I(inode)->cache_access;
1479 if (cache->cred != cred
1480 || time_after(jiffies, cache->jiffies + NFS_ATTRTIMEO(inode))
1481 || (NFS_FLAGS(inode) & NFS_INO_INVALID_ACCESS))
1483 memcpy(res, cache, sizeof(*res));
1487 void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set)
1489 struct nfs_access_entry *cache = &NFS_I(inode)->cache_access;
1491 if (cache->cred != set->cred) {
1493 put_rpccred(cache->cred);
1494 cache->cred = get_rpccred(set->cred);
1496 NFS_FLAGS(inode) &= ~NFS_INO_INVALID_ACCESS;
1497 cache->jiffies = set->jiffies;
1498 cache->mask = set->mask;
1501 static int nfs_do_access(struct inode *inode, struct rpc_cred *cred, int mask)
1503 struct nfs_access_entry cache;
1506 status = nfs_access_get_cached(inode, cred, &cache);
1510 /* Be clever: ask server to check for all possible rights */
1511 cache.mask = MAY_EXEC | MAY_WRITE | MAY_READ;
1513 cache.jiffies = jiffies;
1514 status = NFS_PROTO(inode)->access(inode, &cache);
1517 nfs_access_add_cache(inode, &cache);
1519 if ((cache.mask & mask) == mask)
1524 int nfs_permission(struct inode *inode, int mask, struct nameidata *nd)
1526 struct rpc_cred *cred;
1531 /* Is this sys_access() ? */
1532 if (nd != NULL && (nd->flags & LOOKUP_ACCESS))
1535 switch (inode->i_mode & S_IFMT) {
1539 /* NFSv4 has atomic_open... */
1540 if (nfs_server_capable(inode, NFS_CAP_ATOMIC_OPEN)
1542 && (nd->flags & LOOKUP_OPEN))
1547 * Optimize away all write operations, since the server
1548 * will check permissions when we perform the op.
1550 if ((mask & MAY_WRITE) && !(mask & MAY_READ))
1557 if (!NFS_PROTO(inode)->access)
1560 cred = rpcauth_lookupcred(NFS_CLIENT(inode)->cl_auth, 0);
1561 if (!IS_ERR(cred)) {
1562 res = nfs_do_access(inode, cred, mask);
1565 res = PTR_ERR(cred);
1570 res = nfs_revalidate_inode(NFS_SERVER(inode), inode);
1572 res = generic_permission(inode, mask, NULL);
1579 * version-control: t
1580 * kept-new-versions: 5